diff options
Diffstat (limited to 'sys')
175 files changed, 56028 insertions, 619 deletions
diff --git a/sys/amd64/amd64/exception.S b/sys/amd64/amd64/exception.S index b0c4ba134964..c473362099e3 100644 --- a/sys/amd64/amd64/exception.S +++ b/sys/amd64/amd64/exception.S @@ -1,8 +1,12 @@ /*- * Copyright (c) 1989, 1990 William F. Jolitz. * Copyright (c) 1990 The Regents of the University of California. + * Copyright (c) 2007 The FreeBSD Foundation * All rights reserved. * + * Portions of this software were developed by A. Joseph Koshy under + * sponsorship from the FreeBSD Foundation and Google, Inc. + * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: @@ -32,6 +36,8 @@ #include "opt_atpic.h" #include "opt_compat.h" +#include "opt_hwpmc_hooks.h" +#include "opt_kdtrace.h" #include <machine/asmacros.h> #include <machine/psl.h> @@ -39,7 +45,25 @@ #include "assym.s" +#ifdef KDTRACE_HOOKS + .bss + .globl dtrace_invop_jump_addr + .align 8 + .type dtrace_invop_jump_addr, @object + .size dtrace_invop_jump_addr, 8 +dtrace_invop_jump_addr: + .zero 8 + .globl dtrace_invop_calltrap_addr + .align 8 + .type dtrace_invop_calltrap_addr, @object + .size dtrace_invop_calltrap_addr, 8 +dtrace_invop_calltrap_addr: + .zero 8 +#endif .text +#ifdef HWPMC_HOOKS + ENTRY(start_exceptions) +#endif /*****************************************************************************/ /* Trap handling */ @@ -162,6 +186,30 @@ alltraps_pushregs_no_rdi: movq %r14,TF_R14(%rsp) movq %r15,TF_R15(%rsp) FAKE_MCOUNT(TF_RIP(%rsp)) +#ifdef KDTRACE_HOOKS + /* + * DTrace Function Boundary Trace (fbt) and Statically Defined + * Trace (sdt) probes are triggered by int3 (0xcc) which causes + * the #BP (T_BPTFLT) breakpoint interrupt. For all other trap + * types, just handle them in the usual way. + */ + cmpq $T_BPTFLT,TF_TRAPNO(%rsp) + jne calltrap + + /* Check if there is no DTrace hook registered. */ + cmpq $0,dtrace_invop_jump_addr + je calltrap + + /* + * Set our jump address for the jump back in the event that + * the breakpoint wasn't caused by DTrace at all. + */ + movq $calltrap, dtrace_invop_calltrap_addr(%rip) + + /* Jump to the code hooked in by DTrace. */ + movq dtrace_invop_jump_addr, %rax + jmpq *dtrace_invop_jump_addr +#endif .globl calltrap .type calltrap,@function calltrap: @@ -348,6 +396,9 @@ IDTVEC(fast_syscall32) * execute the NMI handler with interrupts disabled to prevent a * nested interrupt from executing an 'iretq' instruction and * inadvertently taking the processor out of NMI mode. + * + * Third, the NMI handler runs on its own stack (tss_ist1), shared + * with the double fault handler. */ IDTVEC(nmi) @@ -386,6 +437,61 @@ nmi_calltrap: movq %rsp, %rdi call trap MEXITCOUNT +#ifdef HWPMC_HOOKS + /* + * Check if the current trap was from user mode and if so + * whether the current thread needs a user call chain to be + * captured. We are still in NMI mode at this point. + */ + testb $SEL_RPL_MASK,TF_CS(%rsp) + jz nocallchain + movq PCPU(CURTHREAD),%rax /* curthread present? */ + orq %rax,%rax + jz nocallchain + testl $TDP_CALLCHAIN,TD_PFLAGS(%rax) /* flagged for capture? */ + jz nocallchain + /* + * A user callchain is to be captured, so: + * - Move execution to the regular kernel stack, to allow for + * nested NMI interrupts. + * - Take the processor out of "NMI" mode by faking an "iret". + * - Enable interrupts, so that copyin() can work. + */ + movq %rsp,%rsi /* source stack pointer */ + movq $TF_SIZE,%rcx + movq PCPU(RSP0),%rbx + subq %rcx,%rbx + movq %rbx,%rdi /* destination stack pointer */ + + shrq $3,%rcx /* trap frame size in long words */ + cld + rep + movsq /* copy trapframe */ + + movl %ss,%eax + pushq %rax /* tf_ss */ + pushq %rbx /* tf_rsp (on kernel stack) */ + pushfq /* tf_rflags */ + movl %cs,%eax + pushq %rax /* tf_cs */ + pushq $outofnmi /* tf_rip */ + iretq +outofnmi: + /* + * At this point the processor has exited NMI mode and is running + * with interrupts turned off on the normal kernel stack. + * We turn interrupts back on, and take the usual 'doreti' exit + * path. + * + * If a pending NMI gets recognized at or after this point, it + * will cause a kernel callchain to be traced. Since this path + * is only taken for NMI interrupts from user space, our `swapgs' + * state is correct for taking the doreti path. + */ + sti + jmp doreti +nocallchain: +#endif testl %ebx,%ebx jz nmi_restoreregs swapgs @@ -553,3 +659,6 @@ doreti_iret_fault: movq $0,TF_ADDR(%rsp) FAKE_MCOUNT(TF_RIP(%rsp)) jmp calltrap +#ifdef HWPMC_HOOKS + ENTRY(end_exceptions) +#endif diff --git a/sys/amd64/amd64/local_apic.c b/sys/amd64/amd64/local_apic.c index 9a3ec2a834ee..47bb78dadf56 100644 --- a/sys/amd64/amd64/local_apic.c +++ b/sys/amd64/amd64/local_apic.c @@ -35,6 +35,7 @@ __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" +#include "opt_kdtrace.h" #include "opt_ddb.h" @@ -65,6 +66,11 @@ __FBSDID("$FreeBSD$"); #include <ddb/ddb.h> #endif +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> +cyclic_clock_func_t lapic_cyclic_clock_func[MAXCPU]; +#endif + /* Sanity checks on IDT vectors. */ CTASSERT(APIC_IO_INTS + APIC_NUM_IOINTS == APIC_TIMER_INT); CTASSERT(APIC_TIMER_INT < APIC_LOCAL_INTS); @@ -668,6 +674,17 @@ lapic_handle_timer(struct trapframe *frame) (*la->la_timer_count)++; critical_enter(); +#ifdef KDTRACE_HOOKS + /* + * If the DTrace hooks are configured and a callback function + * has been registered, then call it to process the high speed + * timers. + */ + int cpu = PCPU_GET(cpuid); + if (lapic_cyclic_clock_func[cpu] != NULL) + (*lapic_cyclic_clock_func[cpu])(frame); +#endif + /* Fire hardclock at hz. */ la->la_hard_ticks += hz; if (la->la_hard_ticks >= lapic_timer_hz) { diff --git a/sys/amd64/amd64/trap.c b/sys/amd64/amd64/trap.c index f7a2d1ac63e4..6b083415d50c 100644 --- a/sys/amd64/amd64/trap.c +++ b/sys/amd64/amd64/trap.c @@ -49,6 +49,7 @@ __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" #include "opt_isa.h" #include "opt_kdb.h" +#include "opt_kdtrace.h" #include "opt_ktrace.h" #include <sys/param.h> @@ -94,6 +95,24 @@ __FBSDID("$FreeBSD$"); #endif #include <machine/tss.h> +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> + +/* + * This is a hook which is initialised by the dtrace module + * to handle traps which might occur during DTrace probe + * execution. + */ +dtrace_trap_func_t dtrace_trap_func; + +/* + * This is a hook which is initialised by the systrace module + * when it is loaded. This keeps the DTrace syscall provider + * implementation opaque. + */ +systrace_probe_func_t systrace_probe_func; +#endif + extern void trap(struct trapframe *frame); extern void syscall(struct trapframe *frame); void dblfault_handler(struct trapframe *frame); @@ -195,11 +214,28 @@ trap(struct trapframe *frame) * the NMI was handled by it and we can return immediately. */ if (type == T_NMI && pmc_intr && - (*pmc_intr)(PCPU_GET(cpuid), (uintptr_t) frame->tf_rip, - TRAPF_USERMODE(frame))) + (*pmc_intr)(PCPU_GET(cpuid), frame)) goto out; #endif +#ifdef KDTRACE_HOOKS + /* + * A trap can occur while DTrace executes a probe. Before + * executing the probe, DTrace blocks re-scheduling and sets + * a flag in it's per-cpu flags to indicate that it doesn't + * want to fault. On returning from the the probe, the no-fault + * flag is cleared and finally re-scheduling is enabled. + * + * If the DTrace kernel module has registered a trap handler, + * call it and if it returns non-zero, assume that it has + * handled the trap and modified the trap frame so that this + * function can return normally. + */ + if (dtrace_trap_func != NULL) + if ((*dtrace_trap_func)(frame, type)) + goto out; +#endif + if ((frame->tf_rflags & PSL_I) == 0) { /* * Buggy application or kernel code has disabled @@ -211,7 +247,7 @@ trap(struct trapframe *frame) if (ISPL(frame->tf_cs) == SEL_UPL) printf( "pid %ld (%s): trap %d with interrupts disabled\n", - (long)curproc->p_pid, curproc->p_comm, type); + (long)curproc->p_pid, curthread->td_name, type); else if (type != T_NMI && type != T_BPTFLT && type != T_TRCTRAP) { /* @@ -708,8 +744,8 @@ trap_fatal(frame, eva) printf("current process = "); if (curproc) { printf("%lu (%s)\n", - (u_long)curproc->p_pid, curproc->p_comm ? - curproc->p_comm : ""); + (u_long)curproc->p_pid, curthread->td_name ? + curthread->td_name : ""); } else { printf("Idle\n"); } @@ -836,7 +872,7 @@ syscall(struct trapframe *frame) #endif CTR4(KTR_SYSC, "syscall enter thread %p pid %d proc %s code %d", td, - td->td_proc->p_pid, td->td_proc->p_comm, code); + td->td_proc->p_pid, td->td_name, code); td->td_syscalls++; @@ -848,9 +884,34 @@ syscall(struct trapframe *frame) PTRACESTOP_SC(p, td, S_PT_SCE); +#ifdef KDTRACE_HOOKS + /* + * If the systrace module has registered it's probe + * callback and if there is a probe active for the + * syscall 'entry', process the probe. + */ + if (systrace_probe_func != NULL && callp->sy_entry != 0) + (*systrace_probe_func)(callp->sy_entry, code, callp, + args); +#endif + AUDIT_SYSCALL_ENTER(code, td); error = (*callp->sy_call)(td, argp); AUDIT_SYSCALL_EXIT(error, td); + + /* Save the latest error return value. */ + td->td_errno = error; + +#ifdef KDTRACE_HOOKS + /* + * If the systrace module has registered it's probe + * callback and if there is a probe active for the + * syscall 'return', process the probe. + */ + if (systrace_probe_func != NULL && callp->sy_return != 0) + (*systrace_probe_func)(callp->sy_return, code, callp, + args); +#endif } switch (error) { @@ -918,7 +979,7 @@ syscall(struct trapframe *frame) userret(td, frame); CTR4(KTR_SYSC, "syscall exit thread %p pid %d proc %s code %d", td, - td->td_proc->p_pid, td->td_proc->p_comm, code); + td->td_proc->p_pid, td->td_name, code); #ifdef KTRACE if (KTRPOINT(td, KTR_SYSRET)) diff --git a/sys/amd64/amd64/tsc.c b/sys/amd64/amd64/tsc.c index 3c66a3355da4..416a4a795dfe 100644 --- a/sys/amd64/amd64/tsc.c +++ b/sys/amd64/amd64/tsc.c @@ -28,6 +28,7 @@ __FBSDID("$FreeBSD$"); #include "opt_clock.h" +#include "opt_kdtrace.h" #include <sys/param.h> #include <sys/bus.h> @@ -44,6 +45,10 @@ __FBSDID("$FreeBSD$"); #include <machine/md_var.h> #include <machine/specialreg.h> +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> +#endif + #include "cpufreq_if.h" uint64_t tsc_freq; @@ -220,3 +225,18 @@ tsc_get_timecount(struct timecounter *tc) { return (rdtsc()); } + +#ifdef KDTRACE_HOOKS +/* + * DTrace needs a high resolution time function which can + * be called from a probe context and guaranteed not to have + * instrumented with probes itself. + * + * Returns nanoseconds since boot. + */ +uint64_t +dtrace_gethrtime() +{ + return (rdtsc() * (uint64_t) 1000000000 / tsc_freq); +} +#endif diff --git a/sys/amd64/conf/GENERIC b/sys/amd64/conf/GENERIC index 57c7eb4f0644..d69ed146997e 100644 --- a/sys/amd64/conf/GENERIC +++ b/sys/amd64/conf/GENERIC @@ -64,6 +64,9 @@ options KBD_INSTALL_CDEV # install a CDEV entry in /dev options ADAPTIVE_GIANT # Giant mutex is adaptive. options STOP_NMI # Stop CPUS using NMI instead of IPI options AUDIT # Security event auditing +options PRINTF_BUFR_SIZE=128 # Prevent printf output being interspersed. +options KDTRACE_FRAME # Ensure frames are compiled in +options KDTRACE_HOOKS # Kernel DTrace hooks # Make an SMP-capable kernel by default options SMP # Symmetric MultiProcessor Kernel diff --git a/sys/amd64/include/param.h b/sys/amd64/include/param.h index b1b90f467658..3c1f974b9fe3 100644 --- a/sys/amd64/include/param.h +++ b/sys/amd64/include/param.h @@ -79,7 +79,7 @@ #define MACHINE_ARCH "amd64" #endif -#ifdef SMP +#if defined(SMP) || defined(KLD_MODULE) #define MAXCPU 16 #else #define MAXCPU 1 diff --git a/sys/arm/include/param.h b/sys/arm/include/param.h index 4c59eee53c6d..92affb6c9be0 100644 --- a/sys/arm/include/param.h +++ b/sys/arm/include/param.h @@ -72,8 +72,8 @@ #endif #define MID_MACHINE MID_ARM6 -#ifdef SMP -#define MAXCPU 2 +#if defined(SMP) || defined(KLD_MODULE) +#define MAXCPU 2 #else #define MAXCPU 1 #endif /* SMP */ diff --git a/sys/cddl/compat/opensolaris/kern/opensolaris.c b/sys/cddl/compat/opensolaris/kern/opensolaris.c new file mode 100644 index 000000000000..f1f84fb570a4 --- /dev/null +++ b/sys/cddl/compat/opensolaris/kern/opensolaris.c @@ -0,0 +1,101 @@ +/*- + * Copyright 2007 John Birrell <jb@FreeBSD.org> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + */ + +#include <sys/cdefs.h> +#include <sys/types.h> +#include <sys/conf.h> +#include <sys/cpuvar.h> +#include <sys/errno.h> +#include <sys/kernel.h> +#include <sys/module.h> +#include <sys/mutex.h> + +cpu_core_t cpu_core[MAXCPU]; +kmutex_t cpu_lock; +solaris_cpu_t solaris_cpu[MAXCPU]; + +/* + * OpenSolaris subsystem initialisation. + */ +static void +opensolaris_load(void *dummy) +{ + int i; + + printf("This module (opensolaris) contains code covered by the\n"); + printf("Common Development and Distribution License (CDDL)\n"); + printf("see http://opensolaris.org/os/licensing/opensolaris_license/\n"); + + /* + * "Enable" all CPUs even though they may not exist just so + * that the asserts work. On FreeBSD, if a CPU exists, it is + * enabled. + */ + for (i = 0; i < MAXCPU; i++) { + solaris_cpu[i].cpuid = i; + solaris_cpu[i].cpu_flags &= CPU_ENABLE; + } + + mutex_init(&cpu_lock, "OpenSolaris CPU lock", MUTEX_DEFAULT, NULL); +} + +SYSINIT(opensolaris_register, SI_SUB_OPENSOLARIS, SI_ORDER_FIRST, opensolaris_load, NULL); + +static void +opensolaris_unload(void) +{ + mutex_destroy(&cpu_lock); +} + +SYSUNINIT(opensolaris_unregister, SI_SUB_OPENSOLARIS, SI_ORDER_FIRST, opensolaris_unload, NULL); + +static int +opensolaris_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + return (error); +} + +DEV_MODULE(opensolaris, opensolaris_modevent, NULL); +MODULE_VERSION(opensolaris, 1); diff --git a/sys/cddl/compat/opensolaris/kern/opensolaris_cmn_err.c b/sys/cddl/compat/opensolaris/kern/opensolaris_cmn_err.c new file mode 100644 index 000000000000..665ef53bf3a3 --- /dev/null +++ b/sys/cddl/compat/opensolaris/kern/opensolaris_cmn_err.c @@ -0,0 +1,64 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + */ + +#include <sys/cmn_err.h> + +void +vcmn_err(int ce, const char *fmt, va_list adx) +{ + char buf[256]; + + switch (ce) { + case CE_CONT: + snprintf(buf, sizeof(buf), "Solaris(cont): %s\n", fmt); + break; + case CE_NOTE: + snprintf(buf, sizeof(buf), "Solaris: NOTICE: %s\n", fmt); + break; + case CE_WARN: + snprintf(buf, sizeof(buf), "Solaris: WARNING: %s\n", fmt); + break; + case CE_PANIC: + snprintf(buf, sizeof(buf), "Solaris(panic): %s\n", fmt); + break; + case CE_IGNORE: + break; + default: + panic("Solaris: unknown severity level"); + } + if (ce == CE_PANIC) + panic(buf); + if (ce != CE_IGNORE) + vprintf(buf, adx); +} + +void +cmn_err(int type, const char *fmt, ...) +{ + va_list ap; + + va_start(ap, fmt); + vcmn_err(type, fmt, ap); + va_end(ap); +} diff --git a/sys/cddl/compat/opensolaris/kern/opensolaris_kmem.c b/sys/cddl/compat/opensolaris/kern/opensolaris_kmem.c index a73349f583e7..cfbad3d75739 100644 --- a/sys/cddl/compat/opensolaris/kern/opensolaris_kmem.c +++ b/sys/cddl/compat/opensolaris/kern/opensolaris_kmem.c @@ -40,6 +40,8 @@ __FBSDID("$FreeBSD$"); #include <vm/vm_kern.h> #include <vm/vm_map.h> +#define KMEM_DEBUG + #ifdef KMEM_DEBUG #include <sys/queue.h> #include <sys/stack.h> @@ -93,6 +95,10 @@ void zfs_kmem_free(void *buf, size_t size __unused) { #ifdef KMEM_DEBUG + if (buf == NULL) { + printf("%s: attempt to free NULL\n",__func__); + return; + } struct kmem_item *i; buf = (u_char *)buf - sizeof(struct kmem_item); @@ -108,18 +114,18 @@ zfs_kmem_free(void *buf, size_t size __unused) free(buf, M_SOLARIS); } -u_long +uint64_t kmem_size(void) { - return ((u_long)vm_kmem_size); + return ((uint64_t)vm_kmem_size); } -u_long +uint64_t kmem_used(void) { - return ((u_long)kmem_map->size); + return ((uint64_t)kmem_map->size); } static int @@ -200,7 +206,6 @@ kmem_cache_free(kmem_cache_t *cache, void *buf) } #ifdef _KERNEL -extern void zone_drain(uma_zone_t zone); void kmem_cache_reap_now(kmem_cache_t *cache) { @@ -237,7 +242,8 @@ calloc(size_t n, size_t s) } #ifdef KMEM_DEBUG -static void +void kmem_show(void *); +void kmem_show(void *dummy __unused) { struct kmem_item *i; @@ -249,12 +255,10 @@ kmem_show(void *dummy __unused) printf("KMEM_DEBUG: Leaked elements:\n\n"); LIST_FOREACH(i, &kmem_items, next) { printf("address=%p\n", i); - stack_print(&i->stack); - printf("\n"); } } mtx_unlock(&kmem_items_mtx); } -SYSUNINIT(sol_kmem, SI_SUB_DRIVERS, SI_ORDER_FIRST, kmem_show, NULL); +SYSUNINIT(sol_kmem, SI_SUB_CPU, SI_ORDER_FIRST, kmem_show, NULL); #endif /* KMEM_DEBUG */ diff --git a/sys/cddl/compat/opensolaris/sys/atomic.h b/sys/cddl/compat/opensolaris/sys/atomic.h index 2559415da3fa..17aa0971caa7 100644 --- a/sys/cddl/compat/opensolaris/sys/atomic.h +++ b/sys/cddl/compat/opensolaris/sys/atomic.h @@ -32,6 +32,10 @@ #include <sys/types.h> #include <machine/atomic.h> +#define casptr(_a, _b, _c) \ + atomic_cmpset_ptr((volatile uintptr_t *)(_a), (uintptr_t)(_b), (uintptr_t) (_c)) +#define cas32 atomic_cmpset_32 + #ifndef __LP64__ extern void atomic_add_64(volatile uint64_t *target, int64_t delta); extern void *atomic_cas_ptr(volatile void *target, void *cmp, void *newval); @@ -106,7 +110,7 @@ atomic_inc_64_nv(volatile uint64_t *target) static __inline void * atomic_cas_ptr(volatile void *target, void *cmp, void *newval) { - return ((void *)atomic_cas_64((uint64_t *)target, (uint64_t)cmp, + return ((void *)atomic_cas_64((volatile uint64_t *)target, (uint64_t)cmp, (uint64_t)newval)); } #endif diff --git a/sys/cddl/compat/opensolaris/sys/bitmap.h b/sys/cddl/compat/opensolaris/sys/bitmap.h new file mode 100644 index 000000000000..f3130dfb822f --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/bitmap.h @@ -0,0 +1,118 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ + +/* + * Copyright 2005 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */ +/* All Rights Reserved */ + + +#ifndef _COMPAT_OPENSOLARIS_SYS_BITMAP_H +#define _COMPAT_OPENSOLARIS_SYS_BITMAP_H + +#include <sys/atomic.h> + +/* + * Operations on bitmaps of arbitrary size + * A bitmap is a vector of 1 or more ulong_t's. + * The user of the package is responsible for range checks and keeping + * track of sizes. + */ + +#ifdef _LP64 +#define BT_ULSHIFT 6 /* log base 2 of BT_NBIPUL, to extract word index */ +#define BT_ULSHIFT32 5 /* log base 2 of BT_NBIPUL, to extract word index */ +#else +#define BT_ULSHIFT 5 /* log base 2 of BT_NBIPUL, to extract word index */ +#endif + +#define BT_NBIPUL (1 << BT_ULSHIFT) /* n bits per ulong_t */ +#define BT_ULMASK (BT_NBIPUL - 1) /* to extract bit index */ + +#ifdef _LP64 +#define BT_NBIPUL32 (1 << BT_ULSHIFT32) /* n bits per ulong_t */ +#define BT_ULMASK32 (BT_NBIPUL32 - 1) /* to extract bit index */ +#define BT_ULMAXMASK 0xffffffffffffffff /* used by bt_getlowbit */ +#else +#define BT_ULMAXMASK 0xffffffff +#endif + +/* + * bitmap is a ulong_t *, bitindex an index_t + * + * The macros BT_WIM and BT_BIW internal; there is no need + * for users of this package to use them. + */ + +/* + * word in map + */ +#define BT_WIM(bitmap, bitindex) \ + ((bitmap)[(bitindex) >> BT_ULSHIFT]) +/* + * bit in word + */ +#define BT_BIW(bitindex) \ + (1UL << ((bitindex) & BT_ULMASK)) + +#ifdef _LP64 +#define BT_WIM32(bitmap, bitindex) \ + ((bitmap)[(bitindex) >> BT_ULSHIFT32]) + +#define BT_BIW32(bitindex) \ + (1UL << ((bitindex) & BT_ULMASK32)) +#endif + +/* + * These are public macros + * + * BT_BITOUL == n bits to n ulong_t's + */ +#define BT_BITOUL(nbits) \ + (((nbits) + BT_NBIPUL - 1l) / BT_NBIPUL) +#define BT_SIZEOFMAP(nbits) \ + (BT_BITOUL(nbits) * sizeof (ulong_t)) +#define BT_TEST(bitmap, bitindex) \ + ((BT_WIM((bitmap), (bitindex)) & BT_BIW(bitindex)) ? 1 : 0) +#define BT_SET(bitmap, bitindex) \ + { BT_WIM((bitmap), (bitindex)) |= BT_BIW(bitindex); } +#define BT_CLEAR(bitmap, bitindex) \ + { BT_WIM((bitmap), (bitindex)) &= ~BT_BIW(bitindex); } + +#ifdef _LP64 +#define BT_BITOUL32(nbits) \ + (((nbits) + BT_NBIPUL32 - 1l) / BT_NBIPUL32) +#define BT_SIZEOFMAP32(nbits) \ + (BT_BITOUL32(nbits) * sizeof (uint_t)) +#define BT_TEST32(bitmap, bitindex) \ + ((BT_WIM32((bitmap), (bitindex)) & BT_BIW32(bitindex)) ? 1 : 0) +#define BT_SET32(bitmap, bitindex) \ + { BT_WIM32((bitmap), (bitindex)) |= BT_BIW32(bitindex); } +#define BT_CLEAR32(bitmap, bitindex) \ + { BT_WIM32((bitmap), (bitindex)) &= ~BT_BIW32(bitindex); } +#endif /* _LP64 */ + +#endif /* _COMPAT_OPENSOLARIS_SYS_BITMAP_H */ diff --git a/sys/cddl/compat/opensolaris/sys/cmn_err.h b/sys/cddl/compat/opensolaris/sys/cmn_err.h index b9987e8c75a9..9979a3f042c4 100644 --- a/sys/cddl/compat/opensolaris/sys/cmn_err.h +++ b/sys/cddl/compat/opensolaris/sys/cmn_err.h @@ -44,44 +44,8 @@ extern "C" { #define CE_PANIC 3 /* panic */ #define CE_IGNORE 4 /* print nothing */ -static __inline void -vcmn_err(int ce, const char *fmt, va_list adx) -{ - char buf[256]; - - switch (ce) { - case CE_CONT: - snprintf(buf, sizeof(buf), "ZFS(cont): %s\n", fmt); - break; - case CE_NOTE: - snprintf(buf, sizeof(buf), "ZFS: NOTICE: %s\n", fmt); - break; - case CE_WARN: - snprintf(buf, sizeof(buf), "ZFS: WARNING: %s\n", fmt); - break; - case CE_PANIC: - snprintf(buf, sizeof(buf), "ZFS(panic): %s\n", fmt); - break; - case CE_IGNORE: - break; - default: - panic("unknown severity level"); - } - if (ce != CE_IGNORE) - vprintf(buf, adx); - if (ce == CE_PANIC) - panic("ZFS"); -} - -static __inline void -cmn_err(int ce, const char *fmt, ...) -{ - va_list adx; - - va_start(adx, fmt); - vcmn_err(ce, fmt, adx); - va_end(adx); -} +void cmn_err(int, const char *, ...); +void vcmn_err(int, const char *, va_list); #ifdef __cplusplus } diff --git a/sys/cddl/compat/opensolaris/sys/cpupart.h b/sys/cddl/compat/opensolaris/sys/cpupart.h new file mode 100644 index 000000000000..560339b0b35e --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/cpupart.h @@ -0,0 +1,38 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ + +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_CPUPART_H +#define _COMPAT_OPENSOLARIS_SYS_CPUPART_H + +typedef int cpupartid_t; + +typedef struct cpupart { + cpupartid_t cp_id; /* partition ID */ +} cpupart_t; + +#endif /* _COMPAT_OPENSOLARIS_SYS_CPUPART_H */ diff --git a/sys/cddl/compat/opensolaris/sys/cpuvar.h b/sys/cddl/compat/opensolaris/sys/cpuvar.h new file mode 100644 index 000000000000..df0af5c41f14 --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/cpuvar.h @@ -0,0 +1,136 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ + +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_CPUVAR_H +#define _COMPAT_OPENSOLARIS_SYS_CPUVAR_H + +#include <sys/mutex.h> +#include <sys/cpuvar_defs.h> + +#ifdef _KERNEL + +struct cyc_cpu; + +typedef struct { + int cpuid; + struct cyc_cpu *cpu_cyclic; + uint32_t cpu_flags; + uint_t cpu_intr_actv; + uintptr_t cpu_profile_pc; + uintptr_t cpu_profile_upc; + uintptr_t cpu_dtrace_caller; /* DTrace: caller, if any */ + hrtime_t cpu_dtrace_chillmark; /* DTrace: chill mark time */ + hrtime_t cpu_dtrace_chilled; /* DTrace: total chill time */ +} solaris_cpu_t; + +/* Some code may choose to redefine this if pcpu_t would be more useful. */ +#define cpu_t solaris_cpu_t +#define cpu_id cpuid + +extern solaris_cpu_t solaris_cpu[]; + +#define CPU_CACHE_COHERENCE_SIZE 64 + +/* + * The cpu_core structure consists of per-CPU state available in any context. + * On some architectures, this may mean that the page(s) containing the + * NCPU-sized array of cpu_core structures must be locked in the TLB -- it + * is up to the platform to assure that this is performed properly. Note that + * the structure is sized to avoid false sharing. + */ +#define CPUC_SIZE (sizeof (uint16_t) + sizeof (uintptr_t) + \ + sizeof (kmutex_t)) +#define CPUC_PADSIZE CPU_CACHE_COHERENCE_SIZE - CPUC_SIZE + +typedef struct cpu_core { + uint16_t cpuc_dtrace_flags; /* DTrace flags */ + uint8_t cpuc_pad[CPUC_PADSIZE]; /* padding */ + uintptr_t cpuc_dtrace_illval; /* DTrace illegal value */ + kmutex_t cpuc_pid_lock; /* DTrace pid provider lock */ +} cpu_core_t; + +extern cpu_core_t cpu_core[]; + +extern kmutex_t cpu_lock; +#endif /* _KERNEL */ + +/* + * Flags in the CPU structure. + * + * These are protected by cpu_lock (except during creation). + * + * Offlined-CPUs have three stages of being offline: + * + * CPU_ENABLE indicates that the CPU is participating in I/O interrupts + * that can be directed at a number of different CPUs. If CPU_ENABLE + * is off, the CPU will not be given interrupts that can be sent elsewhere, + * but will still get interrupts from devices associated with that CPU only, + * and from other CPUs. + * + * CPU_OFFLINE indicates that the dispatcher should not allow any threads + * other than interrupt threads to run on that CPU. A CPU will not have + * CPU_OFFLINE set if there are any bound threads (besides interrupts). + * + * CPU_QUIESCED is set if p_offline was able to completely turn idle the + * CPU and it will not have to run interrupt threads. In this case it'll + * stay in the idle loop until CPU_QUIESCED is turned off. + * + * CPU_FROZEN is used only by CPR to mark CPUs that have been successfully + * suspended (in the suspend path), or have yet to be resumed (in the resume + * case). + * + * On some platforms CPUs can be individually powered off. + * The following flags are set for powered off CPUs: CPU_QUIESCED, + * CPU_OFFLINE, and CPU_POWEROFF. The following flags are cleared: + * CPU_RUNNING, CPU_READY, CPU_EXISTS, and CPU_ENABLE. + */ +#define CPU_RUNNING 0x001 /* CPU running */ +#define CPU_READY 0x002 /* CPU ready for cross-calls */ +#define CPU_QUIESCED 0x004 /* CPU will stay in idle */ +#define CPU_EXISTS 0x008 /* CPU is configured */ +#define CPU_ENABLE 0x010 /* CPU enabled for interrupts */ +#define CPU_OFFLINE 0x020 /* CPU offline via p_online */ +#define CPU_POWEROFF 0x040 /* CPU is powered off */ +#define CPU_FROZEN 0x080 /* CPU is frozen via CPR suspend */ +#define CPU_SPARE 0x100 /* CPU offline available for use */ +#define CPU_FAULTED 0x200 /* CPU offline diagnosed faulty */ + +typedef enum { + CPU_INIT, + CPU_CONFIG, + CPU_UNCONFIG, + CPU_ON, + CPU_OFF, + CPU_CPUPART_IN, + CPU_CPUPART_OUT +} cpu_setup_t; + +typedef int cpu_setup_func_t(cpu_setup_t, int, void *); + + +#endif /* _COMPAT_OPENSOLARIS_SYS_CPUVAR_H */ diff --git a/sys/cddl/compat/opensolaris/sys/cpuvar_defs.h b/sys/cddl/compat/opensolaris/sys/cpuvar_defs.h new file mode 100644 index 000000000000..af0947ba8ad6 --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/cpuvar_defs.h @@ -0,0 +1,61 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ + +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_CPUVAR_DEFS_H +#define _COMPAT_OPENSOLARIS_SYS_CPUVAR_DEFS_H + +/* + * DTrace flags. + */ +#define CPU_DTRACE_NOFAULT 0x0001 /* Don't fault */ +#define CPU_DTRACE_DROP 0x0002 /* Drop this ECB */ +#define CPU_DTRACE_BADADDR 0x0004 /* DTrace fault: bad address */ +#define CPU_DTRACE_BADALIGN 0x0008 /* DTrace fault: bad alignment */ +#define CPU_DTRACE_DIVZERO 0x0010 /* DTrace fault: divide by zero */ +#define CPU_DTRACE_ILLOP 0x0020 /* DTrace fault: illegal operation */ +#define CPU_DTRACE_NOSCRATCH 0x0040 /* DTrace fault: out of scratch */ +#define CPU_DTRACE_KPRIV 0x0080 /* DTrace fault: bad kernel access */ +#define CPU_DTRACE_UPRIV 0x0100 /* DTrace fault: bad user access */ +#define CPU_DTRACE_TUPOFLOW 0x0200 /* DTrace fault: tuple stack overflow */ +#if defined(__sparc) +#define CPU_DTRACE_FAKERESTORE 0x0400 /* pid provider hint to getreg */ +#endif +#define CPU_DTRACE_ENTRY 0x0800 /* pid provider hint to ustack() */ +#define CPU_DTRACE_BADSTACK 0x1000 /* DTrace fault: bad stack */ + +#define CPU_DTRACE_FAULT (CPU_DTRACE_BADADDR | CPU_DTRACE_BADALIGN | \ + CPU_DTRACE_DIVZERO | CPU_DTRACE_ILLOP | \ + CPU_DTRACE_NOSCRATCH | CPU_DTRACE_KPRIV | \ + CPU_DTRACE_UPRIV | CPU_DTRACE_TUPOFLOW | \ + CPU_DTRACE_BADSTACK) +#define CPU_DTRACE_ERROR (CPU_DTRACE_FAULT | CPU_DTRACE_DROP) + +#define PANICSTKSIZE 8192 +#define REGSIZE 256 + +#endif /* _COMPAT_OPENSOLARIS_SYS_CPUVAR_DEFS_H */ diff --git a/sys/cddl/compat/opensolaris/sys/cyclic.h b/sys/cddl/compat/opensolaris/sys/cyclic.h new file mode 100644 index 000000000000..fced5dfb43b5 --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/cyclic.h @@ -0,0 +1,79 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + */ +/* + * Copyright (c) 1999-2001 by Sun Microsystems, Inc. + * All rights reserved. + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_CYCLIC_H_ +#define _COMPAT_OPENSOLARIS_SYS_CYCLIC_H_ + +#ifndef _KERNEL +typedef void cpu_t; +#endif + + +#ifndef _ASM +#include <sys/time.h> +#include <sys/cpuvar.h> +#endif /* !_ASM */ + +#ifndef _ASM + +typedef uintptr_t cyclic_id_t; +typedef int cyc_index_t; +typedef uint16_t cyc_level_t; +typedef void (*cyc_func_t)(void *); +typedef void *cyb_arg_t; + +#define CYCLIC_NONE ((cyclic_id_t)0) + +typedef struct cyc_handler { + cyc_func_t cyh_func; + void *cyh_arg; +} cyc_handler_t; + +typedef struct cyc_time { + hrtime_t cyt_when; + hrtime_t cyt_interval; +} cyc_time_t; + +typedef struct cyc_omni_handler { + void (*cyo_online)(void *, cpu_t *, cyc_handler_t *, cyc_time_t *); + void (*cyo_offline)(void *, cpu_t *, void *); + void *cyo_arg; +} cyc_omni_handler_t; + +#ifdef _KERNEL + +cyclic_id_t cyclic_add(cyc_handler_t *, cyc_time_t *); +cyclic_id_t cyclic_add_omni(cyc_omni_handler_t *); +void cyclic_remove(cyclic_id_t); + +#endif /* _KERNEL */ + +#endif /* !_ASM */ + +#endif diff --git a/sys/cddl/compat/opensolaris/sys/cyclic_impl.h b/sys/cddl/compat/opensolaris/sys/cyclic_impl.h new file mode 100644 index 000000000000..a19525115cf5 --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/cyclic_impl.h @@ -0,0 +1,304 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + */ +/* + * Copyright 2004 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_CYCLIC_IMPL_H_ +#define _COMPAT_OPENSOLARIS_SYS_CYCLIC_IMPL_H_ + +#include <sys/cyclic.h> + +/* + * Cyclic Subsystem Backend-supplied Interfaces + * -------------------------------------------- + * + * 0 Background + * + * The design, implementation and interfaces of the cyclic subsystem are + * covered in detail in block comments in the implementation. This + * comment covers the interface from the cyclic subsystem into the cyclic + * backend. The backend is specified by a structure of function pointers + * defined below. + * + * 1 Overview + * + * cyb_configure() <-- Configures the backend on the specified CPU + * cyb_unconfigure() <-- Unconfigures the backend + * cyb_enable() <-- Enables the CY_HIGH_LEVEL interrupt source + * cyb_disable() <-- Disables the CY_HIGH_LEVEL interrupt source + * cyb_reprogram() <-- Reprograms the CY_HIGH_LEVEL interrupt source + * cyb_xcall() <-- Cross calls to the specified CPU + * + * 2 cyb_arg_t cyb_configure(cpu_t *) + * + * 2.1 Overview + * + * cyb_configure() should configure the specified CPU for cyclic operation. + * + * 2.2 Arguments and notes + * + * cyb_configure() should initialize any backend-specific per-CPU + * structures for the specified CPU. cyb_configure() will be called for + * each CPU (including the boot CPU) during boot. If the platform + * supports dynamic reconfiguration, cyb_configure() will be called for + * new CPUs as they are configured into the system. + * + * 2.3 Return value + * + * cyb_configure() is expected to return a cookie (a cyb_arg_t, which is + * of type void *) which will be used as the first argument for all future + * cyclic calls into the backend on the specified CPU. + * + * 2.4 Caller's context + * + * cpu_lock will be held. The caller's CPU is unspecified, and may or + * may not be the CPU specified to cyb_configure(). + * + * 3 void cyb_unconfigure(cyb_arg_t arg) + * + * 3.1 Overview + * + * cyb_unconfigure() should unconfigure the specified backend. + * + * 3.2 Arguments and notes + * + * The only argument to cyb_unconfigure() is a cookie as returned from + * cyb_configure(). + * + * cyb_unconfigure() should free any backend-specific per-CPU structures + * for the specified backend. cyb_unconfigure() will _only_ be called on + * platforms which support dynamic reconfiguration. If the platform does + * not support dynamic reconfiguration, cyb_unconfigure() may panic. + * + * After cyb_unconfigure() returns, the backend must not call cyclic_fire() + * on the corresponding CPU; doing so will result in a bad trap. + * + * 3.3 Return value + * + * None. + * + * 3.4 Caller's context + * + * cpu_lock will be held. The caller's CPU is unspecified, and may or + * may not be the CPU specified to cyb_unconfigure(). The specified + * CPU is guaranteed to exist at the time cyb_unconfigure() is called. + * The cyclic subsystem is guaranteed to be suspended when cyb_unconfigure() + * is called, and interrupts are guaranteed to be disabled. + * + * 4 void cyb_enable(cyb_arg_t arg) + * + * 4.1 Overview + * + * cyb_enable() should enable the CY_HIGH_LEVEL interrupt source on + * the specified backend. + * + * 4.2 Arguments and notes + * + * The only argument to cyb_enable() is a backend cookie as returned from + * cyb_configure(). + * + * cyb_enable() will only be called if a) the specified backend has never + * been enabled or b) the specified backend has been explicitly disabled with + * cyb_disable(). In either case, cyb_enable() will only be called if + * the cyclic subsystem wishes to add a cyclic to the CPU corresponding + * to the specified backend. cyb_enable() will be called before + * cyb_reprogram() for a given backend. + * + * cyclic_fire() should not be called on a CPU which has not had its backend + * explicitly cyb_enable()'d, but to do so does not constitute fatal error. + * + * 4.3 Return value + * + * None. + * + * 4.4 Caller's context + * + * cyb_enable() will only be called from CY_HIGH_LEVEL context on the CPU + * corresponding to the specified backend. + * + * 5 void cyb_disable(cyb_arg_t arg) + * + * 5.1 Overview + * + * cyb_disable() should disable the CY_HIGH_LEVEL interrupt source on + * the specified backend. + * + * 5.2 Arguments and notes + * + * The only argument to cyb_disable() is a backend cookie as returned from + * cyb_configure(). + * + * cyb_disable() will only be called on backends which have been previously + * been cyb_enable()'d. cyb_disable() will be called when all cyclics have + * been juggled away or removed from a cyb_enable()'d CPU. + * + * cyclic_fire() should not be called on a CPU which has had its backend + * explicitly cyb_disable()'d, but to do so does not constitute fatal + * error. cyb_disable() is thus not required to check for a pending + * CY_HIGH_LEVEL interrupt. + * + * 5.3 Return value + * + * None. + * + * 5.4 Caller's context + * + * cyb_disable() will only be called from CY_HIGH_LEVEL context on the CPU + * corresponding to the specified backend. + * + * 6 void cyb_reprogram(cyb_arg_t arg, hrtime_t time) + * + * 6.1 Overview + * + * cyb_reprogram() should reprogram the CY_HIGH_LEVEL interrupt source + * to fire at the absolute time specified. + * + * 6.2 Arguments and notes + * + * The first argument to cyb_reprogram() is a backend cookie as returned from + * cyb_configure(). + * + * The second argument is an absolute time at which the CY_HIGH_LEVEL + * interrupt should fire. The specified time _may_ be in the past (albeit + * the very recent past). If this is the case, the backend should generate + * a CY_HIGH_LEVEL interrupt as soon as possible. + * + * The platform should not assume that cyb_reprogram() will be called with + * monotonically increasing values. + * + * If the platform does not allow for interrupts at arbitrary times in the + * future, cyb_reprogram() may do nothing -- as long as cyclic_fire() is + * called periodically at CY_HIGH_LEVEL. While this is clearly suboptimal + * (cyclic granularity will be bounded by the length of the period between + * cyclic_fire()'s), it allows the cyclic subsystem to be implemented on + * inferior hardware. + * + * 6.3 Return value + * + * None. + * + * 6.4 Caller's context + * + * cyb_reprogram() will only be called from CY_HIGH_LEVEL context on the CPU + * corresponding to the specified backend. + * + * 10 cyb_xcall(cyb_arg_t arg, cpu_t *, void(*func)(void *), void *farg) + * + * 10.1 Overview + * + * cyb_xcall() should execute the specified function on the specified CPU. + * + * 10.2 Arguments and notes + * + * The first argument to cyb_restore_level() is a backend cookie as returned + * from cyb_configure(). The second argument is a CPU on which the third + * argument, a function pointer, should be executed. The fourth argument, + * a void *, should be passed as the argument to the specified function. + * + * cyb_xcall() must provide exactly-once semantics. If the specified + * function is called more than once, or not at all, the cyclic subsystem + * will become internally inconsistent. The specified function must be + * be executed on the specified CPU, but may be executed in any context + * (any interrupt context or kernel context). + * + * cyb_xcall() cannot block. Any resources which cyb_xcall() needs to + * acquire must thus be protected by synchronization primitives which + * never require the caller to block. + * + * 10.3 Return value + * + * None. + * + * 10.4 Caller's context + * + * cpu_lock will be held and kernel preemption may be disabled. The caller + * may be unable to block, giving rise to the constraint outlined in + * 10.2, above. + * + */ +typedef struct cyc_backend { + cyb_arg_t (*cyb_configure)(cpu_t *); + void (*cyb_unconfigure)(cyb_arg_t); + void (*cyb_enable)(cyb_arg_t); + void (*cyb_disable)(cyb_arg_t); + void (*cyb_reprogram)(cyb_arg_t, hrtime_t); + void (*cyb_xcall)(cyb_arg_t, cpu_t *, cyc_func_t, void *); + cyb_arg_t cyb_arg; +} cyc_backend_t; + +#define CYF_FREE 0x0001 + +typedef struct cyclic { + hrtime_t cy_expire; + hrtime_t cy_interval; + void (*cy_handler)(void *); + void *cy_arg; + uint16_t cy_flags; +} cyclic_t; + +typedef struct cyc_cpu { + cpu_t *cyp_cpu; + cyc_index_t *cyp_heap; + cyclic_t *cyp_cyclics; + cyc_index_t cyp_nelems; + cyc_index_t cyp_size; + cyc_backend_t *cyp_backend; + struct mtx cyp_mtx; +} cyc_cpu_t; + +typedef struct cyc_omni_cpu { + cyc_cpu_t *cyo_cpu; + cyc_index_t cyo_ndx; + void *cyo_arg; + struct cyc_omni_cpu *cyo_next; +} cyc_omni_cpu_t; + +typedef struct cyc_id { + cyc_cpu_t *cyi_cpu; + cyc_index_t cyi_ndx; + struct cyc_id *cyi_prev; + struct cyc_id *cyi_next; + cyc_omni_handler_t cyi_omni_hdlr; + cyc_omni_cpu_t *cyi_omni_list; +} cyc_id_t; + +typedef struct cyc_xcallarg { + cyc_cpu_t *cyx_cpu; + hrtime_t cyx_exp; +} cyc_xcallarg_t; + +#define CY_DEFAULT_PERCPU 1 +#define CY_PASSIVE_LEVEL -1 + +#define CY_WAIT 0 +#define CY_NOWAIT 1 + +#define CYC_HEAP_PARENT(ndx) (((ndx) - 1) >> 1) +#define CYC_HEAP_RIGHT(ndx) (((ndx) + 1) << 1) +#define CYC_HEAP_LEFT(ndx) ((((ndx) + 1) << 1) - 1) + +#endif diff --git a/sys/cddl/compat/opensolaris/sys/elf.h b/sys/cddl/compat/opensolaris/sys/elf.h new file mode 100644 index 000000000000..a630f28fbc21 --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/elf.h @@ -0,0 +1,116 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + * ELF compatibility definitions for OpenSolaris source. + * + */ + +#ifndef _SYS__ELF_SOLARIS_H_ +#define _SYS__ELF_SOLARIS_H_ + +#include_next <sys/elf.h> + +#define __sElfN(x) typedef __CONCAT(__CONCAT(__CONCAT(Elf,__ELF_WORD_SIZE),_),x) x + +__sElfN(Addr); +__sElfN(Cap); +__sElfN(Dyn); +__sElfN(Ehdr); +__sElfN(Move); +__sElfN(Off); +__sElfN(Phdr); +__sElfN(Rel); +__sElfN(Rela); +__sElfN(Shdr); +__sElfN(Sym); +__sElfN(Syminfo); +__sElfN(Verdaux); +__sElfN(Verdef); +__sElfN(Vernaux); +__sElfN(Verneed); +__sElfN(Versym); + +__sElfN(Half); +__sElfN(Sword); +__sElfN(Word); + +#if __ELF_WORD_SIZE == 32 +typedef Elf32_Word Xword; /* Xword/Sxword are 32-bits in Elf32 */ +typedef Elf32_Sword Sxword; +#else +typedef Elf64_Xword Xword; +typedef Elf64_Sxword Sxword; +#endif + +#define ELF_M_INFO __ELFN(M_INFO) +#define ELF_M_SIZE __ELFN(M_SIZE) +#define ELF_M_SYM __ELFN(M_SYM) + +/* + * Elf `printf' type-cast macros. These force arguments to be a fixed size + * so that Elf32 and Elf64 can share common format strings. + */ +#define EC_ADDR(a) ((Elf64_Addr)(a)) /* "ull" */ +#define EC_OFF(a) ((Elf64_Off)(a)) /* "ull" */ +#define EC_HALF(a) ((Elf64_Half)(a)) /* "d" */ +#define EC_WORD(a) ((Elf64_Word)(a)) /* "u" */ +#define EC_SWORD(a) ((Elf64_Sword)(a)) /* "d" */ +#define EC_XWORD(a) ((Elf64_Xword)(a)) /* "ull" */ +#define EC_SXWORD(a) ((Elf64_Sxword)(a)) /* "ll" */ +#define EC_LWORD(a) ((Elf64_Lword)(a)) /* "ull" */ + +#define elf_checksum __elfN(checksum) +#define elf_fsize __elfN(fsize) +#define elf_getehdr __elfN(getehdr) +#define elf_getphdr __elfN(getphdr) +#define elf_newehdr __elfN(newehdr) +#define elf_newphdr __elfN(newphdr) +#define elf_getshdr __elfN(getshdr) +#define elf_xlatetof __elfN(xlatetof) +#define elf_xlatetom __elfN(xlatetom) + +#define Elf_cap_entry __ElfN(cap_entry) +#define Elf_cap_title __ElfN(cap_title) +#define Elf_demangle_name __ElfN(demangle_name) +#define Elf_dyn_entry __ElfN(dyn_entry) +#define Elf_dyn_title __ElfN(dyn_title) +#define Elf_ehdr __ElfN(ehdr) +#define Elf_got_entry __ElfN(got_entry) +#define Elf_got_title __ElfN(got_title) +#define Elf_reloc_apply_reg __ElfN(reloc_apply_reg) +#define Elf_reloc_apply_val __ElfN(reloc_apply_val) +#define Elf_reloc_entry_1 __ElfN(reloc_entry_1) +#define Elf_reloc_entry_2 __ElfN(reloc_entry_2) +#define Elf_reloc_title __ElfN(reloc_title) +#define Elf_phdr __ElfN(phdr) +#define Elf_shdr __ElfN(shdr) +#define Elf_syms_table_entry __ElfN(syms_table_entry) +#define Elf_syms_table_title __ElfN(syms_table_title) +#define Elf_ver_def_title __ElfN(ver_def_title) +#define Elf_ver_line_1 __ElfN(ver_line_1) +#define Elf_ver_line_2 __ElfN(ver_line_2) +#define Elf_ver_line_3 __ElfN(ver_line_3) +#define Elf_ver_line_4 __ElfN(ver_line_4) +#define Elf_ver_line_5 __ElfN(ver_line_5) +#define Elf_ver_need_title __ElfN(ver_need_title) + +#endif /* !_SYS__ELF_SOLARIS_H_ */ diff --git a/sys/cddl/compat/opensolaris/sys/feature_tests.h b/sys/cddl/compat/opensolaris/sys/feature_tests.h new file mode 100644 index 000000000000..7e2c74ff3fdf --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/feature_tests.h @@ -0,0 +1,35 @@ +/* + * Copyright (C) 2007 John Birrell <jb@freebsd.org> + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_FEATURE_TESTS_H_ +#define _COMPAT_OPENSOLARIS_SYS_FEATURE_TESTS_H_ + +/* Empty. */ + +#endif diff --git a/sys/cddl/compat/opensolaris/sys/kcondvar.h b/sys/cddl/compat/opensolaris/sys/kcondvar.h index 547d622b8dfd..0422ba0791ca 100644 --- a/sys/cddl/compat/opensolaris/sys/kcondvar.h +++ b/sys/cddl/compat/opensolaris/sys/kcondvar.h @@ -45,8 +45,15 @@ typedef enum { } kcv_type_t; #define zfs_cv_init(cv, name, type, arg) do { \ + const char *_name; \ ASSERT((type) == CV_DEFAULT); \ - cv_init((cv), "zfs:" #cv); \ + for (_name = #cv; *_name != '\0'; _name++) { \ + if (*_name >= 'a' && *_name <= 'z') \ + break; \ + } \ + if (*_name == '\0') \ + _name = #cv; \ + cv_init((cv), _name); \ } while (0) #define cv_init(cv, name, type, arg) zfs_cv_init((cv), (name), (type), (arg)) diff --git a/sys/cddl/compat/opensolaris/sys/kmem.h b/sys/cddl/compat/opensolaris/sys/kmem.h index 89dfac9bf634..5258cffe4edf 100644 --- a/sys/cddl/compat/opensolaris/sys/kmem.h +++ b/sys/cddl/compat/opensolaris/sys/kmem.h @@ -41,8 +41,6 @@ #define KM_NOSLEEP M_NOWAIT #define KMC_NODEBUG 0 -typedef void vmem_t; - typedef struct kmem_cache { char kc_name[32]; #ifdef _KERNEL @@ -55,10 +53,12 @@ typedef struct kmem_cache { void *kc_private; } kmem_cache_t; +#define vmem_t void + void *zfs_kmem_alloc(size_t size, int kmflags); void zfs_kmem_free(void *buf, size_t size); -u_long kmem_size(void); -u_long kmem_used(void); +uint64_t kmem_size(void); +uint64_t kmem_used(void); kmem_cache_t *kmem_cache_create(char *name, size_t bufsize, size_t align, int (*constructor)(void *, void *, int), void (*destructor)(void *, void *), void (*reclaim)(void *) __unused, void *private, vmem_t *vmp, int cflags); diff --git a/sys/cddl/compat/opensolaris/sys/lock.h b/sys/cddl/compat/opensolaris/sys/lock.h index 51fcd6789684..27663f46e446 100644 --- a/sys/cddl/compat/opensolaris/sys/lock.h +++ b/sys/cddl/compat/opensolaris/sys/lock.h @@ -35,8 +35,7 @@ #define LO_ALLMASK (LO_INITIALIZED | LO_WITNESS | LO_QUIET | \ LO_RECURSABLE | LO_SLEEPABLE | LO_UPGRADABLE | \ - LO_DUPOK | LO_ENROLLPEND | LO_CLASSMASK | \ - LO_NOPROFILE) + LO_DUPOK | LO_CLASSMASK | LO_NOPROFILE) #define LO_EXPECTED (LO_INITIALIZED | LO_WITNESS | LO_RECURSABLE | \ LO_SLEEPABLE | LO_UPGRADABLE | LO_DUPOK | \ /* sx lock class */(2 << LO_CLASSSHIFT)) diff --git a/sys/cddl/compat/opensolaris/machine/endian.h b/sys/cddl/compat/opensolaris/sys/mman.h index 855189f64c96..ca746898f65d 100644 --- a/sys/cddl/compat/opensolaris/machine/endian.h +++ b/sys/cddl/compat/opensolaris/sys/mman.h @@ -1,7 +1,7 @@ -/*- - * Copyright (c) 2007 Pawel Jakub Dawidek <pjd@FreeBSD.org> +/* + * Copyright (C) 2007 John Birrell <jb@freebsd.org> * All rights reserved. - * + * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: @@ -10,11 +10,11 @@ * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. - * - * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE - * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) @@ -24,21 +24,14 @@ * SUCH DAMAGE. * * $FreeBSD$ + * */ -#ifndef _OPENSOLARIS_MACHINE_ENDIAN_H_ -#define _OPENSOLARIS_MACHINE_ENDIAN_H_ +#ifndef _COMPAT_OPENSOLARIS_SYS_MMAN_H_ +#define _COMPAT_OPENSOLARIS_SYS_MMAN_H_ -#include_next <machine/endian.h> +#include_next <sys/mman.h> -/* - * Solaris defines _LITTLE_ENDIAN or _BIG_ENDIAN, but never both and decides - * which architecture it is based on which thing is defined. - */ -#if _BYTE_ORDER == _LITTLE_ENDIAN -#undef _BIG_ENDIAN -#else -#undef _LITTLE_ENDIAN -#endif +#define mmap64(_a,_b,_c,_d,_e,_f) mmap(_a,_b,_c,_d,_e,_f) -#endif /* !_OPENSOLARIS_MACHINE_ENDIAN_H_ */ +#endif diff --git a/sys/cddl/compat/opensolaris/sys/modctl.h b/sys/cddl/compat/opensolaris/sys/modctl.h new file mode 100644 index 000000000000..7af39b090f3b --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/modctl.h @@ -0,0 +1,38 @@ +/* + * Copyright (C) 2007 John Birrell <jb@freebsd.org> + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_MODCTL_H +#define _COMPAT_OPENSOLARIS_SYS_MODCTL_H + +#include <sys/param.h> +#include <sys/linker.h> + +typedef struct linker_file modctl_t; + +#endif /* _COMPAT_OPENSOLARIS_SYS_MODCTL_H */ diff --git a/sys/cddl/compat/opensolaris/sys/mutex.h b/sys/cddl/compat/opensolaris/sys/mutex.h index d7543af4363e..8756cd0534f4 100644 --- a/sys/cddl/compat/opensolaris/sys/mutex.h +++ b/sys/cddl/compat/opensolaris/sys/mutex.h @@ -53,11 +53,18 @@ typedef struct sx kmutex_t; #endif #define mutex_init(lock, desc, type, arg) do { \ + const char *_name; \ ASSERT((type) == MUTEX_DEFAULT); \ KASSERT(((lock)->lock_object.lo_flags & LO_ALLMASK) != \ LO_EXPECTED, ("lock %s already initialized", #lock)); \ bzero((lock), sizeof(struct sx)); \ - sx_init_flags((lock), "zfs:" #lock, MUTEX_FLAGS); \ + for (_name = #lock; *_name != '\0'; _name++) { \ + if (*_name >= 'a' && *_name <= 'z') \ + break; \ + } \ + if (*_name == '\0') \ + _name = #lock; \ + sx_init_flags((lock), _name, MUTEX_FLAGS); \ } while (0) #define mutex_destroy(lock) sx_destroy(lock) #define mutex_enter(lock) sx_xlock(lock) diff --git a/sys/cddl/compat/opensolaris/sys/objfs.h b/sys/cddl/compat/opensolaris/sys/objfs.h new file mode 100644 index 000000000000..b656e7803edc --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/objfs.h @@ -0,0 +1,35 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_OBJFS_H +#define _COMPAT_OPENSOLARIS_SYS_OBJFS_H + +/* + * Private data structure found in '.info' section + */ +typedef struct objfs_info { + int objfs_info_primary; +} objfs_info_t; + + +#endif /* _COMPAT_OPENSOLARIS_SYS_OBJFS_H */ diff --git a/sys/cddl/compat/opensolaris/sys/param.h b/sys/cddl/compat/opensolaris/sys/param.h new file mode 100644 index 000000000000..8d36a9d07700 --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/param.h @@ -0,0 +1,37 @@ +/* + * Copyright (C) 2007 John Birrell <jb@freebsd.org> + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_PARAM_H_ +#define _COMPAT_OPENSOLARIS_SYS_PARAM_H_ + +#include_next <sys/param.h> + +#define PAGESIZE PAGE_SIZE + +#endif diff --git a/sys/cddl/compat/opensolaris/sys/pcpu.h b/sys/cddl/compat/opensolaris/sys/pcpu.h new file mode 100644 index 000000000000..eb9e6a68384b --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/pcpu.h @@ -0,0 +1,37 @@ +/* + * Copyright (C) 2007 John Birrell <jb@freebsd.org> + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_PCPU_H_ +#define _COMPAT_OPENSOLARIS_SYS_PCPU_H_ + +#include_next <sys/pcpu.h> + +typedef struct pcpu pcpu_t; + +#endif diff --git a/sys/cddl/compat/opensolaris/sys/proc.h b/sys/cddl/compat/opensolaris/sys/proc.h index 2532c5760704..2e9200cb670c 100644 --- a/sys/cddl/compat/opensolaris/sys/proc.h +++ b/sys/cddl/compat/opensolaris/sys/proc.h @@ -44,8 +44,6 @@ #define max_ncpus mp_ncpus #define boot_max_ncpus mp_ncpus -extern int hz; /* system clock's frequency */ - #define TS_RUN 0 #define p0 proc0 diff --git a/sys/cddl/compat/opensolaris/sys/rwlock.h b/sys/cddl/compat/opensolaris/sys/rwlock.h index e87717566c15..a3e55153bbcf 100644 --- a/sys/cddl/compat/opensolaris/sys/rwlock.h +++ b/sys/cddl/compat/opensolaris/sys/rwlock.h @@ -60,10 +60,17 @@ typedef struct sx krwlock_t; #define RW_ISWRITER(x) (rw_iswriter(x)) #define rw_init(lock, desc, type, arg) do { \ + const char *_name; \ KASSERT(((lock)->lock_object.lo_flags & LO_ALLMASK) != \ LO_EXPECTED, ("lock %s already initialized", #lock)); \ bzero((lock), sizeof(struct sx)); \ - sx_init_flags((lock), "zfs:" #lock, RW_FLAGS); \ + for (_name = #lock; *_name != '\0'; _name++) { \ + if (*_name >= 'a' && *_name <= 'z') \ + break; \ + } \ + if (*_name == '\0') \ + _name = #lock; \ + sx_init_flags((lock), _name, RW_FLAGS); \ } while (0) #define rw_destroy(lock) sx_destroy(lock) #define rw_enter(lock, how) do { \ diff --git a/sys/cddl/compat/opensolaris/sys/sema.h b/sys/cddl/compat/opensolaris/sys/sema.h new file mode 100644 index 000000000000..08e4721c8664 --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/sema.h @@ -0,0 +1,40 @@ +/* + * Copyright (C) 2007 John Birrell <jb@freebsd.org> + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_SEMA_H_ +#define _COMPAT_OPENSOLARIS_SYS_SEMA_H_ + +#include_next <sys/sema.h> + +typedef struct sema ksema_t; + +#define sema_p sema_post +#define sema_v sema_value + +#endif diff --git a/sys/cddl/compat/opensolaris/sys/stat.h b/sys/cddl/compat/opensolaris/sys/stat.h new file mode 100644 index 000000000000..5f45ebe08e90 --- /dev/null +++ b/sys/cddl/compat/opensolaris/sys/stat.h @@ -0,0 +1,38 @@ +/* + * Copyright (C) 2007 John Birrell <jb@freebsd.org> + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + */ + +#ifndef _COMPAT_OPENSOLARIS_SYS_STAT_H_ +#define _COMPAT_OPENSOLARIS_SYS_STAT_H_ + +#include_next <sys/stat.h> + +#define stat64 stat +#define fstat64 fstat + +#endif diff --git a/sys/cddl/compat/opensolaris/sys/time.h b/sys/cddl/compat/opensolaris/sys/time.h index 94d24a52f6a9..770b25163191 100644 --- a/sys/cddl/compat/opensolaris/sys/time.h +++ b/sys/cddl/compat/opensolaris/sys/time.h @@ -31,9 +31,17 @@ #include_next <sys/time.h> +#define SEC 1 +#define MILLISEC 1000 +#define MICROSEC 1000000 +#define NANOSEC 1000000000 + +typedef longlong_t hrtime_t; + +#define LBOLT ((gethrtime() * hz) / NANOSEC) + #ifdef _KERNEL -#define lbolt ((gethrtime() * hz) / NANOSEC) -#define lbolt64 (int64_t)(lbolt) +#define lbolt64 (int64_t)(LBOLT) static __inline hrtime_t gethrtime(void) { @@ -53,6 +61,15 @@ gethrtime(void) { #define gethrestime_sec() (time_second) #define gethrestime(ts) getnanotime(ts) +#else + +static __inline hrtime_t gethrtime(void) { + struct timespec ts; + clock_gettime(CLOCK_UPTIME,&ts); + return (((u_int64_t) ts.tv_sec) * NANOSEC + ts.tv_nsec); +} + + #endif /* _KERNEL */ #endif /* !_OPENSOLARIS_SYS_TIME_H_ */ diff --git a/sys/cddl/compat/opensolaris/sys/types.h b/sys/cddl/compat/opensolaris/sys/types.h index 4006aec84f8a..7d5d9e41ed14 100644 --- a/sys/cddl/compat/opensolaris/sys/types.h +++ b/sys/cddl/compat/opensolaris/sys/types.h @@ -39,6 +39,18 @@ #define MAXNAMELEN 256 typedef struct timespec timestruc_t; +typedef u_int uint_t; +typedef u_char uchar_t; +typedef u_short ushort_t; +typedef u_long ulong_t; +typedef long long longlong_t; +typedef unsigned long long u_longlong_t; +typedef off_t off64_t; +typedef id_t taskid_t; +typedef id_t projid_t; +typedef id_t poolid_t; +typedef id_t zoneid_t; +typedef id_t ctid_t; #ifdef _KERNEL @@ -52,6 +64,13 @@ typedef void pathname_t; typedef int64_t rlim64_t; #else +#ifdef NEED_SOLARIS_BOOLEAN +#if defined(__XOPEN_OR_POSIX) +typedef enum { _B_FALSE, _B_TRUE } boolean_t; +#else +typedef enum { B_FALSE, B_TRUE } boolean_t; +#endif /* defined(__XOPEN_OR_POSIX) */ +#endif typedef longlong_t offset_t; typedef u_longlong_t u_offset_t; diff --git a/sys/cddl/compat/opensolaris/sys/uio.h b/sys/cddl/compat/opensolaris/sys/uio.h index 02ee1d831d91..d219ff0e960c 100644 --- a/sys/cddl/compat/opensolaris/sys/uio.h +++ b/sys/cddl/compat/opensolaris/sys/uio.h @@ -51,6 +51,7 @@ typedef struct iovec iovec_t; #define uio_loffset uio_offset +#ifdef BUILDING_ZFS static __inline int zfs_uiomove(void *cp, size_t n, enum uio_rw dir, uio_t *uio) { @@ -59,5 +60,6 @@ zfs_uiomove(void *cp, size_t n, enum uio_rw dir, uio_t *uio) return (uiomove(cp, (int)n, uio)); } #define uiomove(cp, n, dir, uio) zfs_uiomove((cp), (n), (dir), (uio)) +#endif #endif /* !_OPENSOLARIS_SYS_UIO_H_ */ diff --git a/sys/cddl/contrib/opensolaris/OPENSOLARIS.LICENSE b/sys/cddl/contrib/opensolaris/OPENSOLARIS.LICENSE new file mode 100644 index 000000000000..da23621dc843 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/OPENSOLARIS.LICENSE @@ -0,0 +1,384 @@ +Unless otherwise noted, all files in this distribution are released +under the Common Development and Distribution License (CDDL). +Exceptions are noted within the associated source files. + +-------------------------------------------------------------------- + + +COMMON DEVELOPMENT AND DISTRIBUTION LICENSE Version 1.0 + +1. Definitions. + + 1.1. "Contributor" means each individual or entity that creates + or contributes to the creation of Modifications. + + 1.2. "Contributor Version" means the combination of the Original + Software, prior Modifications used by a Contributor (if any), + and the Modifications made by that particular Contributor. + + 1.3. "Covered Software" means (a) the Original Software, or (b) + Modifications, or (c) the combination of files containing + Original Software with files containing Modifications, in + each case including portions thereof. + + 1.4. "Executable" means the Covered Software in any form other + than Source Code. + + 1.5. "Initial Developer" means the individual or entity that first + makes Original Software available under this License. + + 1.6. "Larger Work" means a work which combines Covered Software or + portions thereof with code not governed by the terms of this + License. + + 1.7. "License" means this document. + + 1.8. "Licensable" means having the right to grant, to the maximum + extent possible, whether at the time of the initial grant or + subsequently acquired, any and all of the rights conveyed + herein. + + 1.9. "Modifications" means the Source Code and Executable form of + any of the following: + + A. Any file that results from an addition to, deletion from or + modification of the contents of a file containing Original + Software or previous Modifications; + + B. Any new file that contains any part of the Original + Software or previous Modifications; or + + C. Any new file that is contributed or otherwise made + available under the terms of this License. + + 1.10. "Original Software" means the Source Code and Executable + form of computer software code that is originally released + under this License. + + 1.11. "Patent Claims" means any patent claim(s), now owned or + hereafter acquired, including without limitation, method, + process, and apparatus claims, in any patent Licensable by + grantor. + + 1.12. "Source Code" means (a) the common form of computer software + code in which modifications are made and (b) associated + documentation included in or with such code. + + 1.13. "You" (or "Your") means an individual or a legal entity + exercising rights under, and complying with all of the terms + of, this License. For legal entities, "You" includes any + entity which controls, is controlled by, or is under common + control with You. For purposes of this definition, + "control" means (a) the power, direct or indirect, to cause + the direction or management of such entity, whether by + contract or otherwise, or (b) ownership of more than fifty + percent (50%) of the outstanding shares or beneficial + ownership of such entity. + +2. License Grants. + + 2.1. The Initial Developer Grant. + + Conditioned upon Your compliance with Section 3.1 below and + subject to third party intellectual property claims, the Initial + Developer hereby grants You a world-wide, royalty-free, + non-exclusive license: + + (a) under intellectual property rights (other than patent or + trademark) Licensable by Initial Developer, to use, + reproduce, modify, display, perform, sublicense and + distribute the Original Software (or portions thereof), + with or without Modifications, and/or as part of a Larger + Work; and + + (b) under Patent Claims infringed by the making, using or + selling of Original Software, to make, have made, use, + practice, sell, and offer for sale, and/or otherwise + dispose of the Original Software (or portions thereof). + + (c) The licenses granted in Sections 2.1(a) and (b) are + effective on the date Initial Developer first distributes + or otherwise makes the Original Software available to a + third party under the terms of this License. + + (d) Notwithstanding Section 2.1(b) above, no patent license is + granted: (1) for code that You delete from the Original + Software, or (2) for infringements caused by: (i) the + modification of the Original Software, or (ii) the + combination of the Original Software with other software + or devices. + + 2.2. Contributor Grant. + + Conditioned upon Your compliance with Section 3.1 below and + subject to third party intellectual property claims, each + Contributor hereby grants You a world-wide, royalty-free, + non-exclusive license: + + (a) under intellectual property rights (other than patent or + trademark) Licensable by Contributor to use, reproduce, + modify, display, perform, sublicense and distribute the + Modifications created by such Contributor (or portions + thereof), either on an unmodified basis, with other + Modifications, as Covered Software and/or as part of a + Larger Work; and + + (b) under Patent Claims infringed by the making, using, or + selling of Modifications made by that Contributor either + alone and/or in combination with its Contributor Version + (or portions of such combination), to make, use, sell, + offer for sale, have made, and/or otherwise dispose of: + (1) Modifications made by that Contributor (or portions + thereof); and (2) the combination of Modifications made by + that Contributor with its Contributor Version (or portions + of such combination). + + (c) The licenses granted in Sections 2.2(a) and 2.2(b) are + effective on the date Contributor first distributes or + otherwise makes the Modifications available to a third + party. + + (d) Notwithstanding Section 2.2(b) above, no patent license is + granted: (1) for any code that Contributor has deleted + from the Contributor Version; (2) for infringements caused + by: (i) third party modifications of Contributor Version, + or (ii) the combination of Modifications made by that + Contributor with other software (except as part of the + Contributor Version) or other devices; or (3) under Patent + Claims infringed by Covered Software in the absence of + Modifications made by that Contributor. + +3. Distribution Obligations. + + 3.1. Availability of Source Code. + + Any Covered Software that You distribute or otherwise make + available in Executable form must also be made available in Source + Code form and that Source Code form must be distributed only under + the terms of this License. You must include a copy of this + License with every copy of the Source Code form of the Covered + Software You distribute or otherwise make available. You must + inform recipients of any such Covered Software in Executable form + as to how they can obtain such Covered Software in Source Code + form in a reasonable manner on or through a medium customarily + used for software exchange. + + 3.2. Modifications. + + The Modifications that You create or to which You contribute are + governed by the terms of this License. You represent that You + believe Your Modifications are Your original creation(s) and/or + You have sufficient rights to grant the rights conveyed by this + License. + + 3.3. Required Notices. + + You must include a notice in each of Your Modifications that + identifies You as the Contributor of the Modification. You may + not remove or alter any copyright, patent or trademark notices + contained within the Covered Software, or any notices of licensing + or any descriptive text giving attribution to any Contributor or + the Initial Developer. + + 3.4. Application of Additional Terms. + + You may not offer or impose any terms on any Covered Software in + Source Code form that alters or restricts the applicable version + of this License or the recipients' rights hereunder. You may + choose to offer, and to charge a fee for, warranty, support, + indemnity or liability obligations to one or more recipients of + Covered Software. However, you may do so only on Your own behalf, + and not on behalf of the Initial Developer or any Contributor. + You must make it absolutely clear that any such warranty, support, + indemnity or liability obligation is offered by You alone, and You + hereby agree to indemnify the Initial Developer and every + Contributor for any liability incurred by the Initial Developer or + such Contributor as a result of warranty, support, indemnity or + liability terms You offer. + + 3.5. Distribution of Executable Versions. + + You may distribute the Executable form of the Covered Software + under the terms of this License or under the terms of a license of + Your choice, which may contain terms different from this License, + provided that You are in compliance with the terms of this License + and that the license for the Executable form does not attempt to + limit or alter the recipient's rights in the Source Code form from + the rights set forth in this License. If You distribute the + Covered Software in Executable form under a different license, You + must make it absolutely clear that any terms which differ from + this License are offered by You alone, not by the Initial + Developer or Contributor. You hereby agree to indemnify the + Initial Developer and every Contributor for any liability incurred + by the Initial Developer or such Contributor as a result of any + such terms You offer. + + 3.6. Larger Works. + + You may create a Larger Work by combining Covered Software with + other code not governed by the terms of this License and + distribute the Larger Work as a single product. In such a case, + You must make sure the requirements of this License are fulfilled + for the Covered Software. + +4. Versions of the License. + + 4.1. New Versions. + + Sun Microsystems, Inc. is the initial license steward and may + publish revised and/or new versions of this License from time to + time. Each version will be given a distinguishing version number. + Except as provided in Section 4.3, no one other than the license + steward has the right to modify this License. + + 4.2. Effect of New Versions. + + You may always continue to use, distribute or otherwise make the + Covered Software available under the terms of the version of the + License under which You originally received the Covered Software. + If the Initial Developer includes a notice in the Original + Software prohibiting it from being distributed or otherwise made + available under any subsequent version of the License, You must + distribute and make the Covered Software available under the terms + of the version of the License under which You originally received + the Covered Software. Otherwise, You may also choose to use, + distribute or otherwise make the Covered Software available under + the terms of any subsequent version of the License published by + the license steward. + + 4.3. Modified Versions. + + When You are an Initial Developer and You want to create a new + license for Your Original Software, You may create and use a + modified version of this License if You: (a) rename the license + and remove any references to the name of the license steward + (except to note that the license differs from this License); and + (b) otherwise make it clear that the license contains terms which + differ from this License. + +5. DISCLAIMER OF WARRANTY. + + COVERED SOFTWARE IS PROVIDED UNDER THIS LICENSE ON AN "AS IS" + BASIS, WITHOUT WARRANTY OF ANY KIND, EITHER EXPRESSED OR IMPLIED, + INCLUDING, WITHOUT LIMITATION, WARRANTIES THAT THE COVERED + SOFTWARE IS FREE OF DEFECTS, MERCHANTABLE, FIT FOR A PARTICULAR + PURPOSE OR NON-INFRINGING. THE ENTIRE RISK AS TO THE QUALITY AND + PERFORMANCE OF THE COVERED SOFTWARE IS WITH YOU. SHOULD ANY + COVERED SOFTWARE PROVE DEFECTIVE IN ANY RESPECT, YOU (NOT THE + INITIAL DEVELOPER OR ANY OTHER CONTRIBUTOR) ASSUME THE COST OF ANY + NECESSARY SERVICING, REPAIR OR CORRECTION. THIS DISCLAIMER OF + WARRANTY CONSTITUTES AN ESSENTIAL PART OF THIS LICENSE. NO USE OF + ANY COVERED SOFTWARE IS AUTHORIZED HEREUNDER EXCEPT UNDER THIS + DISCLAIMER. + +6. TERMINATION. + + 6.1. This License and the rights granted hereunder will terminate + automatically if You fail to comply with terms herein and fail to + cure such breach within 30 days of becoming aware of the breach. + Provisions which, by their nature, must remain in effect beyond + the termination of this License shall survive. + + 6.2. If You assert a patent infringement claim (excluding + declaratory judgment actions) against Initial Developer or a + Contributor (the Initial Developer or Contributor against whom You + assert such claim is referred to as "Participant") alleging that + the Participant Software (meaning the Contributor Version where + the Participant is a Contributor or the Original Software where + the Participant is the Initial Developer) directly or indirectly + infringes any patent, then any and all rights granted directly or + indirectly to You by such Participant, the Initial Developer (if + the Initial Developer is not the Participant) and all Contributors + under Sections 2.1 and/or 2.2 of this License shall, upon 60 days + notice from Participant terminate prospectively and automatically + at the expiration of such 60 day notice period, unless if within + such 60 day period You withdraw Your claim with respect to the + Participant Software against such Participant either unilaterally + or pursuant to a written agreement with Participant. + + 6.3. In the event of termination under Sections 6.1 or 6.2 above, + all end user licenses that have been validly granted by You or any + distributor hereunder prior to termination (excluding licenses + granted to You by any distributor) shall survive termination. + +7. LIMITATION OF LIABILITY. + + UNDER NO CIRCUMSTANCES AND UNDER NO LEGAL THEORY, WHETHER TORT + (INCLUDING NEGLIGENCE), CONTRACT, OR OTHERWISE, SHALL YOU, THE + INITIAL DEVELOPER, ANY OTHER CONTRIBUTOR, OR ANY DISTRIBUTOR OF + COVERED SOFTWARE, OR ANY SUPPLIER OF ANY OF SUCH PARTIES, BE + LIABLE TO ANY PERSON FOR ANY INDIRECT, SPECIAL, INCIDENTAL, OR + CONSEQUENTIAL DAMAGES OF ANY CHARACTER INCLUDING, WITHOUT + LIMITATION, DAMAGES FOR LOST PROFITS, LOSS OF GOODWILL, WORK + STOPPAGE, COMPUTER FAILURE OR MALFUNCTION, OR ANY AND ALL OTHER + COMMERCIAL DAMAGES OR LOSSES, EVEN IF SUCH PARTY SHALL HAVE BEEN + INFORMED OF THE POSSIBILITY OF SUCH DAMAGES. THIS LIMITATION OF + LIABILITY SHALL NOT APPLY TO LIABILITY FOR DEATH OR PERSONAL + INJURY RESULTING FROM SUCH PARTY'S NEGLIGENCE TO THE EXTENT + APPLICABLE LAW PROHIBITS SUCH LIMITATION. SOME JURISDICTIONS DO + NOT ALLOW THE EXCLUSION OR LIMITATION OF INCIDENTAL OR + CONSEQUENTIAL DAMAGES, SO THIS EXCLUSION AND LIMITATION MAY NOT + APPLY TO YOU. + +8. U.S. GOVERNMENT END USERS. + + The Covered Software is a "commercial item," as that term is + defined in 48 C.F.R. 2.101 (Oct. 1995), consisting of "commercial + computer software" (as that term is defined at 48 + C.F.R. 252.227-7014(a)(1)) and "commercial computer software + documentation" as such terms are used in 48 C.F.R. 12.212 + (Sept. 1995). Consistent with 48 C.F.R. 12.212 and 48 + C.F.R. 227.7202-1 through 227.7202-4 (June 1995), all + U.S. Government End Users acquire Covered Software with only those + rights set forth herein. This U.S. Government Rights clause is in + lieu of, and supersedes, any other FAR, DFAR, or other clause or + provision that addresses Government rights in computer software + under this License. + +9. MISCELLANEOUS. + + This License represents the complete agreement concerning subject + matter hereof. If any provision of this License is held to be + unenforceable, such provision shall be reformed only to the extent + necessary to make it enforceable. This License shall be governed + by the law of the jurisdiction specified in a notice contained + within the Original Software (except to the extent applicable law, + if any, provides otherwise), excluding such jurisdiction's + conflict-of-law provisions. Any litigation relating to this + License shall be subject to the jurisdiction of the courts located + in the jurisdiction and venue specified in a notice contained + within the Original Software, with the losing party responsible + for costs, including, without limitation, court costs and + reasonable attorneys' fees and expenses. The application of the + United Nations Convention on Contracts for the International Sale + of Goods is expressly excluded. Any law or regulation which + provides that the language of a contract shall be construed + against the drafter shall not apply to this License. You agree + that You alone are responsible for compliance with the United + States export administration regulations (and the export control + laws and regulation of any other countries) when You use, + distribute or otherwise make available any Covered Software. + +10. RESPONSIBILITY FOR CLAIMS. + + As between Initial Developer and the Contributors, each party is + responsible for claims and damages arising, directly or + indirectly, out of its utilization of rights under this License + and You agree to work with Initial Developer and Contributors to + distribute such responsibility on an equitable basis. Nothing + herein is intended or shall be deemed to constitute any admission + of liability. + +-------------------------------------------------------------------- + +NOTICE PURSUANT TO SECTION 9 OF THE COMMON DEVELOPMENT AND +DISTRIBUTION LICENSE (CDDL) + +For Covered Software in this distribution, this License shall +be governed by the laws of the State of California (excluding +conflict-of-law provisions). + +Any litigation relating to this License shall be subject to the +jurisdiction of the Federal Courts of the Northern District of +California and the state courts of the State of California, with +venue lying in Santa Clara County, California. diff --git a/sys/cddl/contrib/opensolaris/uts/common/ctf/ctf_mod.c b/sys/cddl/contrib/opensolaris/uts/common/ctf/ctf_mod.c new file mode 100644 index 000000000000..b34cf400cd2c --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/ctf/ctf_mod.c @@ -0,0 +1,177 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2003 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/sysmacros.h> +#include <sys/modctl.h> +#include <sys/debug.h> +#include <sys/mman.h> +#include <sys/modctl.h> +#include <sys/kobj.h> +#include <ctf_impl.h> + +int ctf_leave_compressed = 0; + +static struct modlmisc modlmisc = { + &mod_miscops, "Compact C Type Format routines" +}; + +static struct modlinkage modlinkage = { + MODREV_1, &modlmisc, NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *mip) +{ + return (mod_info(&modlinkage, mip)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} + +/*ARGSUSED*/ +void * +ctf_zopen(int *errp) +{ + return ((void *)1); /* zmod is always loaded because we depend on it */ +} + +/*ARGSUSED*/ +const void * +ctf_sect_mmap(ctf_sect_t *sp, int fd) +{ + return (MAP_FAILED); /* we don't support this in the kernel */ +} + +/*ARGSUSED*/ +void +ctf_sect_munmap(const ctf_sect_t *sp) +{ + /* we don't support this in the kernel */ +} + +/*ARGSUSED*/ +ctf_file_t * +ctf_fdopen(int fd, int *errp) +{ + return (ctf_set_open_errno(errp, ENOTSUP)); +} + +/*ARGSUSED*/ +ctf_file_t * +ctf_open(const char *filename, int *errp) +{ + return (ctf_set_open_errno(errp, ENOTSUP)); +} + +/*ARGSUSED*/ +int +ctf_write(ctf_file_t *fp, int fd) +{ + return (ctf_set_errno(fp, ENOTSUP)); +} + +int +ctf_version(int version) +{ + ASSERT(version > 0 && version <= CTF_VERSION); + + if (version > 0) + _libctf_version = MIN(CTF_VERSION, version); + + return (_libctf_version); +} + +/*ARGSUSED*/ +ctf_file_t * +ctf_modopen(struct module *mp, int *error) +{ + ctf_sect_t ctfsect, symsect, strsect; + ctf_file_t *fp = NULL; + int err; + + if (error == NULL) + error = &err; + + ctfsect.cts_name = ".SUNW_ctf"; + ctfsect.cts_type = SHT_PROGBITS; + ctfsect.cts_flags = SHF_ALLOC; + ctfsect.cts_data = mp->ctfdata; + ctfsect.cts_size = mp->ctfsize; + ctfsect.cts_entsize = 1; + ctfsect.cts_offset = 0; + + symsect.cts_name = ".symtab"; + symsect.cts_type = SHT_SYMTAB; + symsect.cts_flags = 0; + symsect.cts_data = mp->symtbl; + symsect.cts_size = mp->symhdr->sh_size; +#ifdef _LP64 + symsect.cts_entsize = sizeof (Elf64_Sym); +#else + symsect.cts_entsize = sizeof (Elf32_Sym); +#endif + symsect.cts_offset = 0; + + strsect.cts_name = ".strtab"; + strsect.cts_type = SHT_STRTAB; + strsect.cts_flags = 0; + strsect.cts_data = mp->strings; + strsect.cts_size = mp->strhdr->sh_size; + strsect.cts_entsize = 1; + strsect.cts_offset = 0; + + ASSERT(MUTEX_HELD(&mod_lock)); + + if ((fp = ctf_bufopen(&ctfsect, &symsect, &strsect, error)) == NULL) + return (NULL); + + if (!ctf_leave_compressed && (caddr_t)fp->ctf_base != mp->ctfdata) { + /* + * We must have just uncompressed the CTF data. To avoid + * others having to pay the (substantial) cost of decompressing + * the data, we're going to substitute the uncompressed version + * for the compressed version. Note that this implies that the + * first CTF consumer will induce memory impact on the system + * (but in the name of performance of future CTF consumers). + */ + kobj_set_ctf(mp, (caddr_t)fp->ctf_base, fp->ctf_size); + fp->ctf_data.cts_data = fp->ctf_base; + fp->ctf_data.cts_size = fp->ctf_size; + } + + return (fp); +} diff --git a/sys/cddl/contrib/opensolaris/uts/common/ctf/ctf_subr.c b/sys/cddl/contrib/opensolaris/uts/common/ctf/ctf_subr.c new file mode 100644 index 000000000000..cd0a828628d4 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/ctf/ctf_subr.c @@ -0,0 +1,96 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2004 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <ctf_impl.h> +#include <sys/kobj.h> +#include <sys/kobj_impl.h> + +/* + * This module is used both during the normal operation of the kernel (i.e. + * after kmem has been initialized) and during boot (before unix`_start has + * been called). kobj_alloc is able to tell the difference between the two + * cases, and as such must be used instead of kmem_alloc. + */ + +void * +ctf_data_alloc(size_t size) +{ + void *buf = kobj_alloc(size, KM_NOWAIT|KM_SCRATCH); + + if (buf == NULL) + return (MAP_FAILED); + + return (buf); +} + +void +ctf_data_free(void *buf, size_t size) +{ + kobj_free(buf, size); +} + +/*ARGSUSED*/ +void +ctf_data_protect(void *buf, size_t size) +{ + /* we don't support this operation in the kernel */ +} + +void * +ctf_alloc(size_t size) +{ + return (kobj_alloc(size, KM_NOWAIT|KM_TMP)); +} + +/*ARGSUSED*/ +void +ctf_free(void *buf, size_t size) +{ + kobj_free(buf, size); +} + +/*ARGSUSED*/ +const char * +ctf_strerror(int err) +{ + return (NULL); /* we don't support this operation in the kernel */ +} + +/*PRINTFLIKE1*/ +void +ctf_dprintf(const char *format, ...) +{ + if (_libctf_debug) { + va_list alist; + + va_start(alist, format); + (void) printf("ctf DEBUG: "); + (void) vprintf(format, alist); + va_end(alist); + } +} diff --git a/sys/cddl/contrib/opensolaris/uts/common/dtrace/dtrace.c b/sys/cddl/contrib/opensolaris/uts/common/dtrace/dtrace.c new file mode 100644 index 000000000000..7600708dba75 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/dtrace/dtrace.c @@ -0,0 +1,16449 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD: src/sys/cddl/contrib/opensolaris/uts/common/dtrace/dtrace.c,v 1.6 2008/08/19 21:28:58 jb Exp $ + */ + +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#pragma ident "%Z%%M% %I% %E% SMI" + +/* + * DTrace - Dynamic Tracing for Solaris + * + * This is the implementation of the Solaris Dynamic Tracing framework + * (DTrace). The user-visible interface to DTrace is described at length in + * the "Solaris Dynamic Tracing Guide". The interfaces between the libdtrace + * library, the in-kernel DTrace framework, and the DTrace providers are + * described in the block comments in the <sys/dtrace.h> header file. The + * internal architecture of DTrace is described in the block comments in the + * <sys/dtrace_impl.h> header file. The comments contained within the DTrace + * implementation very much assume mastery of all of these sources; if one has + * an unanswered question about the implementation, one should consult them + * first. + * + * The functions here are ordered roughly as follows: + * + * - Probe context functions + * - Probe hashing functions + * - Non-probe context utility functions + * - Matching functions + * - Provider-to-Framework API functions + * - Probe management functions + * - DIF object functions + * - Format functions + * - Predicate functions + * - ECB functions + * - Buffer functions + * - Enabling functions + * - DOF functions + * - Anonymous enabling functions + * - Consumer state functions + * - Helper functions + * - Hook functions + * - Driver cookbook functions + * + * Each group of functions begins with a block comment labelled the "DTrace + * [Group] Functions", allowing one to find each block by searching forward + * on capital-f functions. + */ +#include <sys/errno.h> +#if !defined(sun) +#include <sys/time.h> +#endif +#include <sys/stat.h> +#include <sys/modctl.h> +#include <sys/conf.h> +#include <sys/systm.h> +#if defined(sun) +#include <sys/ddi.h> +#include <sys/sunddi.h> +#endif +#include <sys/cpuvar.h> +#include <sys/kmem.h> +#if defined(sun) +#include <sys/strsubr.h> +#endif +#include <sys/sysmacros.h> +#include <sys/dtrace_impl.h> +#include <sys/atomic.h> +#include <sys/cmn_err.h> +#if defined(sun) +#include <sys/mutex_impl.h> +#include <sys/rwlock_impl.h> +#endif +#include <sys/ctf_api.h> +#if defined(sun) +#include <sys/panic.h> +#include <sys/priv_impl.h> +#endif +#include <sys/policy.h> +#if defined(sun) +#include <sys/cred_impl.h> +#include <sys/procfs_isa.h> +#endif +#include <sys/taskq.h> +#if defined(sun) +#include <sys/mkdev.h> +#include <sys/kdi.h> +#endif +#include <sys/zone.h> +#include <sys/socket.h> +#include <netinet/in.h> + +/* FreeBSD includes: */ +#if !defined(sun) +#include <sys/callout.h> +#include <sys/ctype.h> +#include <sys/limits.h> +#include <sys/kdb.h> +#include <sys/kernel.h> +#include <sys/malloc.h> +#include <sys/sysctl.h> +#include <sys/lock.h> +#include <sys/mutex.h> +#include <sys/sx.h> +#include <sys/dtrace_bsd.h> +#include <netinet/in.h> +#include "dtrace_cddl.h" +#include "dtrace_debug.c" +#endif + +/* + * DTrace Tunable Variables + * + * The following variables may be tuned by adding a line to /etc/system that + * includes both the name of the DTrace module ("dtrace") and the name of the + * variable. For example: + * + * set dtrace:dtrace_destructive_disallow = 1 + * + * In general, the only variables that one should be tuning this way are those + * that affect system-wide DTrace behavior, and for which the default behavior + * is undesirable. Most of these variables are tunable on a per-consumer + * basis using DTrace options, and need not be tuned on a system-wide basis. + * When tuning these variables, avoid pathological values; while some attempt + * is made to verify the integrity of these variables, they are not considered + * part of the supported interface to DTrace, and they are therefore not + * checked comprehensively. Further, these variables should not be tuned + * dynamically via "mdb -kw" or other means; they should only be tuned via + * /etc/system. + */ +int dtrace_destructive_disallow = 0; +dtrace_optval_t dtrace_nonroot_maxsize = (16 * 1024 * 1024); +size_t dtrace_difo_maxsize = (256 * 1024); +dtrace_optval_t dtrace_dof_maxsize = (256 * 1024); +size_t dtrace_global_maxsize = (16 * 1024); +size_t dtrace_actions_max = (16 * 1024); +size_t dtrace_retain_max = 1024; +dtrace_optval_t dtrace_helper_actions_max = 32; +dtrace_optval_t dtrace_helper_providers_max = 32; +dtrace_optval_t dtrace_dstate_defsize = (1 * 1024 * 1024); +size_t dtrace_strsize_default = 256; +dtrace_optval_t dtrace_cleanrate_default = 9900990; /* 101 hz */ +dtrace_optval_t dtrace_cleanrate_min = 200000; /* 5000 hz */ +dtrace_optval_t dtrace_cleanrate_max = (uint64_t)60 * NANOSEC; /* 1/minute */ +dtrace_optval_t dtrace_aggrate_default = NANOSEC; /* 1 hz */ +dtrace_optval_t dtrace_statusrate_default = NANOSEC; /* 1 hz */ +dtrace_optval_t dtrace_statusrate_max = (hrtime_t)10 * NANOSEC; /* 6/minute */ +dtrace_optval_t dtrace_switchrate_default = NANOSEC; /* 1 hz */ +dtrace_optval_t dtrace_nspec_default = 1; +dtrace_optval_t dtrace_specsize_default = 32 * 1024; +dtrace_optval_t dtrace_stackframes_default = 20; +dtrace_optval_t dtrace_ustackframes_default = 20; +dtrace_optval_t dtrace_jstackframes_default = 50; +dtrace_optval_t dtrace_jstackstrsize_default = 512; +int dtrace_msgdsize_max = 128; +hrtime_t dtrace_chill_max = 500 * (NANOSEC / MILLISEC); /* 500 ms */ +hrtime_t dtrace_chill_interval = NANOSEC; /* 1000 ms */ +int dtrace_devdepth_max = 32; +int dtrace_err_verbose; +hrtime_t dtrace_deadman_interval = NANOSEC; +hrtime_t dtrace_deadman_timeout = (hrtime_t)10 * NANOSEC; +hrtime_t dtrace_deadman_user = (hrtime_t)30 * NANOSEC; + +/* + * DTrace External Variables + * + * As dtrace(7D) is a kernel module, any DTrace variables are obviously + * available to DTrace consumers via the backtick (`) syntax. One of these, + * dtrace_zero, is made deliberately so: it is provided as a source of + * well-known, zero-filled memory. While this variable is not documented, + * it is used by some translators as an implementation detail. + */ +const char dtrace_zero[256] = { 0 }; /* zero-filled memory */ + +/* + * DTrace Internal Variables + */ +#if defined(sun) +static dev_info_t *dtrace_devi; /* device info */ +#endif +#if defined(sun) +static vmem_t *dtrace_arena; /* probe ID arena */ +static vmem_t *dtrace_minor; /* minor number arena */ +static taskq_t *dtrace_taskq; /* task queue */ +#else +static struct unrhdr *dtrace_arena; /* Probe ID number. */ +#endif +static dtrace_probe_t **dtrace_probes; /* array of all probes */ +static int dtrace_nprobes; /* number of probes */ +static dtrace_provider_t *dtrace_provider; /* provider list */ +static dtrace_meta_t *dtrace_meta_pid; /* user-land meta provider */ +static int dtrace_opens; /* number of opens */ +static int dtrace_helpers; /* number of helpers */ +#if defined(sun) +static void *dtrace_softstate; /* softstate pointer */ +#endif +static dtrace_hash_t *dtrace_bymod; /* probes hashed by module */ +static dtrace_hash_t *dtrace_byfunc; /* probes hashed by function */ +static dtrace_hash_t *dtrace_byname; /* probes hashed by name */ +static dtrace_toxrange_t *dtrace_toxrange; /* toxic range array */ +static int dtrace_toxranges; /* number of toxic ranges */ +static int dtrace_toxranges_max; /* size of toxic range array */ +static dtrace_anon_t dtrace_anon; /* anonymous enabling */ +static kmem_cache_t *dtrace_state_cache; /* cache for dynamic state */ +static uint64_t dtrace_vtime_references; /* number of vtimestamp refs */ +static kthread_t *dtrace_panicked; /* panicking thread */ +static dtrace_ecb_t *dtrace_ecb_create_cache; /* cached created ECB */ +static dtrace_genid_t dtrace_probegen; /* current probe generation */ +static dtrace_helpers_t *dtrace_deferred_pid; /* deferred helper list */ +static dtrace_enabling_t *dtrace_retained; /* list of retained enablings */ +static dtrace_dynvar_t dtrace_dynhash_sink; /* end of dynamic hash chains */ +#if !defined(sun) +static struct mtx dtrace_unr_mtx; +MTX_SYSINIT(dtrace_unr_mtx, &dtrace_unr_mtx, "Unique resource identifier", MTX_DEF); +int dtrace_in_probe; /* non-zero if executing a probe */ +#if defined(__i386__) || defined(__amd64__) +uintptr_t dtrace_in_probe_addr; /* Address of invop when already in probe */ +#endif +#endif + +/* + * DTrace Locking + * DTrace is protected by three (relatively coarse-grained) locks: + * + * (1) dtrace_lock is required to manipulate essentially any DTrace state, + * including enabling state, probes, ECBs, consumer state, helper state, + * etc. Importantly, dtrace_lock is _not_ required when in probe context; + * probe context is lock-free -- synchronization is handled via the + * dtrace_sync() cross call mechanism. + * + * (2) dtrace_provider_lock is required when manipulating provider state, or + * when provider state must be held constant. + * + * (3) dtrace_meta_lock is required when manipulating meta provider state, or + * when meta provider state must be held constant. + * + * The lock ordering between these three locks is dtrace_meta_lock before + * dtrace_provider_lock before dtrace_lock. (In particular, there are + * several places where dtrace_provider_lock is held by the framework as it + * calls into the providers -- which then call back into the framework, + * grabbing dtrace_lock.) + * + * There are two other locks in the mix: mod_lock and cpu_lock. With respect + * to dtrace_provider_lock and dtrace_lock, cpu_lock continues its historical + * role as a coarse-grained lock; it is acquired before both of these locks. + * With respect to dtrace_meta_lock, its behavior is stranger: cpu_lock must + * be acquired _between_ dtrace_meta_lock and any other DTrace locks. + * mod_lock is similar with respect to dtrace_provider_lock in that it must be + * acquired _between_ dtrace_provider_lock and dtrace_lock. + */ +static kmutex_t dtrace_lock; /* probe state lock */ +static kmutex_t dtrace_provider_lock; /* provider state lock */ +static kmutex_t dtrace_meta_lock; /* meta-provider state lock */ + +#if !defined(sun) +/* XXX FreeBSD hacks. */ +static kmutex_t mod_lock; + +#define cr_suid cr_svuid +#define cr_sgid cr_svgid +#define ipaddr_t in_addr_t +#define mod_modname pathname +#define vuprintf vprintf +#define ttoproc(_a) ((_a)->td_proc) +#define crgetzoneid(_a) 0 +#define NCPU MAXCPU +#define SNOCD 0 +#define CPU_ON_INTR(_a) 0 + +#define PRIV_EFFECTIVE (1 << 0) +#define PRIV_DTRACE_KERNEL (1 << 1) +#define PRIV_DTRACE_PROC (1 << 2) +#define PRIV_DTRACE_USER (1 << 3) +#define PRIV_PROC_OWNER (1 << 4) +#define PRIV_PROC_ZONE (1 << 5) +#define PRIV_ALL ~0 + +SYSCTL_NODE(_debug, OID_AUTO, dtrace, CTLFLAG_RD, 0, "DTrace Information"); +#endif + +#if defined(sun) +#define curcpu CPU->cpu_id +#endif + + +/* + * DTrace Provider Variables + * + * These are the variables relating to DTrace as a provider (that is, the + * provider of the BEGIN, END, and ERROR probes). + */ +static dtrace_pattr_t dtrace_provider_attr = { +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +}; + +static void +dtrace_nullop(void) +{} + +static dtrace_pops_t dtrace_provider_ops = { + (void (*)(void *, dtrace_probedesc_t *))dtrace_nullop, + (void (*)(void *, modctl_t *))dtrace_nullop, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, + NULL, + NULL, + NULL, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop +}; + +static dtrace_id_t dtrace_probeid_begin; /* special BEGIN probe */ +static dtrace_id_t dtrace_probeid_end; /* special END probe */ +dtrace_id_t dtrace_probeid_error; /* special ERROR probe */ + +/* + * DTrace Helper Tracing Variables + */ +uint32_t dtrace_helptrace_next = 0; +uint32_t dtrace_helptrace_nlocals; +char *dtrace_helptrace_buffer; +int dtrace_helptrace_bufsize = 512 * 1024; + +#ifdef DEBUG +int dtrace_helptrace_enabled = 1; +#else +int dtrace_helptrace_enabled = 0; +#endif + +/* + * DTrace Error Hashing + * + * On DEBUG kernels, DTrace will track the errors that has seen in a hash + * table. This is very useful for checking coverage of tests that are + * expected to induce DIF or DOF processing errors, and may be useful for + * debugging problems in the DIF code generator or in DOF generation . The + * error hash may be examined with the ::dtrace_errhash MDB dcmd. + */ +#ifdef DEBUG +static dtrace_errhash_t dtrace_errhash[DTRACE_ERRHASHSZ]; +static const char *dtrace_errlast; +static kthread_t *dtrace_errthread; +static kmutex_t dtrace_errlock; +#endif + +/* + * DTrace Macros and Constants + * + * These are various macros that are useful in various spots in the + * implementation, along with a few random constants that have no meaning + * outside of the implementation. There is no real structure to this cpp + * mishmash -- but is there ever? + */ +#define DTRACE_HASHSTR(hash, probe) \ + dtrace_hash_str(*((char **)((uintptr_t)(probe) + (hash)->dth_stroffs))) + +#define DTRACE_HASHNEXT(hash, probe) \ + (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_nextoffs) + +#define DTRACE_HASHPREV(hash, probe) \ + (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_prevoffs) + +#define DTRACE_HASHEQ(hash, lhs, rhs) \ + (strcmp(*((char **)((uintptr_t)(lhs) + (hash)->dth_stroffs)), \ + *((char **)((uintptr_t)(rhs) + (hash)->dth_stroffs))) == 0) + +#define DTRACE_AGGHASHSIZE_SLEW 17 + +#define DTRACE_V4MAPPED_OFFSET (sizeof (uint32_t) * 3) + +/* + * The key for a thread-local variable consists of the lower 61 bits of the + * t_did, plus the 3 bits of the highest active interrupt above LOCK_LEVEL. + * We add DIF_VARIABLE_MAX to t_did to assure that the thread key is never + * equal to a variable identifier. This is necessary (but not sufficient) to + * assure that global associative arrays never collide with thread-local + * variables. To guarantee that they cannot collide, we must also define the + * order for keying dynamic variables. That order is: + * + * [ key0 ] ... [ keyn ] [ variable-key ] [ tls-key ] + * + * Because the variable-key and the tls-key are in orthogonal spaces, there is + * no way for a global variable key signature to match a thread-local key + * signature. + */ +#if defined(sun) +#define DTRACE_TLS_THRKEY(where) { \ + uint_t intr = 0; \ + uint_t actv = CPU->cpu_intr_actv >> (LOCK_LEVEL + 1); \ + for (; actv; actv >>= 1) \ + intr++; \ + ASSERT(intr < (1 << 3)); \ + (where) = ((curthread->t_did + DIF_VARIABLE_MAX) & \ + (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \ +} +#else +#define DTRACE_TLS_THRKEY(where) { \ + solaris_cpu_t *_c = &solaris_cpu[curcpu]; \ + uint_t intr = 0; \ + uint_t actv = _c->cpu_intr_actv; \ + for (; actv; actv >>= 1) \ + intr++; \ + ASSERT(intr < (1 << 3)); \ + (where) = ((curthread->td_tid + DIF_VARIABLE_MAX) & \ + (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \ +} +#endif + +#define DT_BSWAP_8(x) ((x) & 0xff) +#define DT_BSWAP_16(x) ((DT_BSWAP_8(x) << 8) | DT_BSWAP_8((x) >> 8)) +#define DT_BSWAP_32(x) ((DT_BSWAP_16(x) << 16) | DT_BSWAP_16((x) >> 16)) +#define DT_BSWAP_64(x) ((DT_BSWAP_32(x) << 32) | DT_BSWAP_32((x) >> 32)) + +#define DT_MASK_LO 0x00000000FFFFFFFFULL + +#define DTRACE_STORE(type, tomax, offset, what) \ + *((type *)((uintptr_t)(tomax) + (uintptr_t)offset)) = (type)(what); + +#ifndef __i386 +#define DTRACE_ALIGNCHECK(addr, size, flags) \ + if (addr & (size - 1)) { \ + *flags |= CPU_DTRACE_BADALIGN; \ + cpu_core[curcpu].cpuc_dtrace_illval = addr; \ + return (0); \ + } +#else +#define DTRACE_ALIGNCHECK(addr, size, flags) +#endif + +/* + * Test whether a range of memory starting at testaddr of size testsz falls + * within the range of memory described by addr, sz. We take care to avoid + * problems with overflow and underflow of the unsigned quantities, and + * disallow all negative sizes. Ranges of size 0 are allowed. + */ +#define DTRACE_INRANGE(testaddr, testsz, baseaddr, basesz) \ + ((testaddr) - (baseaddr) < (basesz) && \ + (testaddr) + (testsz) - (baseaddr) <= (basesz) && \ + (testaddr) + (testsz) >= (testaddr)) + +/* + * Test whether alloc_sz bytes will fit in the scratch region. We isolate + * alloc_sz on the righthand side of the comparison in order to avoid overflow + * or underflow in the comparison with it. This is simpler than the INRANGE + * check above, because we know that the dtms_scratch_ptr is valid in the + * range. Allocations of size zero are allowed. + */ +#define DTRACE_INSCRATCH(mstate, alloc_sz) \ + ((mstate)->dtms_scratch_base + (mstate)->dtms_scratch_size - \ + (mstate)->dtms_scratch_ptr >= (alloc_sz)) + +#define DTRACE_LOADFUNC(bits) \ +/*CSTYLED*/ \ +uint##bits##_t \ +dtrace_load##bits(uintptr_t addr) \ +{ \ + size_t size = bits / NBBY; \ + /*CSTYLED*/ \ + uint##bits##_t rval; \ + int i; \ + volatile uint16_t *flags = (volatile uint16_t *) \ + &cpu_core[curcpu].cpuc_dtrace_flags; \ + \ + DTRACE_ALIGNCHECK(addr, size, flags); \ + \ + for (i = 0; i < dtrace_toxranges; i++) { \ + if (addr >= dtrace_toxrange[i].dtt_limit) \ + continue; \ + \ + if (addr + size <= dtrace_toxrange[i].dtt_base) \ + continue; \ + \ + /* \ + * This address falls within a toxic region; return 0. \ + */ \ + *flags |= CPU_DTRACE_BADADDR; \ + cpu_core[curcpu].cpuc_dtrace_illval = addr; \ + return (0); \ + } \ + \ + *flags |= CPU_DTRACE_NOFAULT; \ + /*CSTYLED*/ \ + rval = *((volatile uint##bits##_t *)addr); \ + *flags &= ~CPU_DTRACE_NOFAULT; \ + \ + return (!(*flags & CPU_DTRACE_FAULT) ? rval : 0); \ +} + +#ifdef _LP64 +#define dtrace_loadptr dtrace_load64 +#else +#define dtrace_loadptr dtrace_load32 +#endif + +#define DTRACE_DYNHASH_FREE 0 +#define DTRACE_DYNHASH_SINK 1 +#define DTRACE_DYNHASH_VALID 2 + +#define DTRACE_MATCH_NEXT 0 +#define DTRACE_MATCH_DONE 1 +#define DTRACE_ANCHORED(probe) ((probe)->dtpr_func[0] != '\0') +#define DTRACE_STATE_ALIGN 64 + +#define DTRACE_FLAGS2FLT(flags) \ + (((flags) & CPU_DTRACE_BADADDR) ? DTRACEFLT_BADADDR : \ + ((flags) & CPU_DTRACE_ILLOP) ? DTRACEFLT_ILLOP : \ + ((flags) & CPU_DTRACE_DIVZERO) ? DTRACEFLT_DIVZERO : \ + ((flags) & CPU_DTRACE_KPRIV) ? DTRACEFLT_KPRIV : \ + ((flags) & CPU_DTRACE_UPRIV) ? DTRACEFLT_UPRIV : \ + ((flags) & CPU_DTRACE_TUPOFLOW) ? DTRACEFLT_TUPOFLOW : \ + ((flags) & CPU_DTRACE_BADALIGN) ? DTRACEFLT_BADALIGN : \ + ((flags) & CPU_DTRACE_NOSCRATCH) ? DTRACEFLT_NOSCRATCH : \ + ((flags) & CPU_DTRACE_BADSTACK) ? DTRACEFLT_BADSTACK : \ + DTRACEFLT_UNKNOWN) + +#define DTRACEACT_ISSTRING(act) \ + ((act)->dta_kind == DTRACEACT_DIFEXPR && \ + (act)->dta_difo->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) + +/* Function prototype definitions: */ +static size_t dtrace_strlen(const char *, size_t); +static dtrace_probe_t *dtrace_probe_lookup_id(dtrace_id_t id); +static void dtrace_enabling_provide(dtrace_provider_t *); +static int dtrace_enabling_match(dtrace_enabling_t *, int *); +static void dtrace_enabling_matchall(void); +static dtrace_state_t *dtrace_anon_grab(void); +#if defined(sun) +static uint64_t dtrace_helper(int, dtrace_mstate_t *, + dtrace_state_t *, uint64_t, uint64_t); +static dtrace_helpers_t *dtrace_helpers_create(proc_t *); +#endif +static void dtrace_buffer_drop(dtrace_buffer_t *); +static intptr_t dtrace_buffer_reserve(dtrace_buffer_t *, size_t, size_t, + dtrace_state_t *, dtrace_mstate_t *); +static int dtrace_state_option(dtrace_state_t *, dtrace_optid_t, + dtrace_optval_t); +static int dtrace_ecb_create_enable(dtrace_probe_t *, void *); +#if defined(sun) +static void dtrace_helper_provider_destroy(dtrace_helper_provider_t *); +#endif +uint16_t dtrace_load16(uintptr_t); +uint32_t dtrace_load32(uintptr_t); +uint64_t dtrace_load64(uintptr_t); +uint8_t dtrace_load8(uintptr_t); +void dtrace_dynvar_clean(dtrace_dstate_t *); +dtrace_dynvar_t *dtrace_dynvar(dtrace_dstate_t *, uint_t, dtrace_key_t *, + size_t, dtrace_dynvar_op_t, dtrace_mstate_t *, dtrace_vstate_t *); +uintptr_t dtrace_dif_varstr(uintptr_t, dtrace_state_t *, dtrace_mstate_t *); + +/* + * DTrace Probe Context Functions + * + * These functions are called from probe context. Because probe context is + * any context in which C may be called, arbitrarily locks may be held, + * interrupts may be disabled, we may be in arbitrary dispatched state, etc. + * As a result, functions called from probe context may only call other DTrace + * support functions -- they may not interact at all with the system at large. + * (Note that the ASSERT macro is made probe-context safe by redefining it in + * terms of dtrace_assfail(), a probe-context safe function.) If arbitrary + * loads are to be performed from probe context, they _must_ be in terms of + * the safe dtrace_load*() variants. + * + * Some functions in this block are not actually called from probe context; + * for these functions, there will be a comment above the function reading + * "Note: not called from probe context." + */ +void +dtrace_panic(const char *format, ...) +{ + va_list alist; + + va_start(alist, format); + dtrace_vpanic(format, alist); + va_end(alist); +} + +int +dtrace_assfail(const char *a, const char *f, int l) +{ + dtrace_panic("assertion failed: %s, file: %s, line: %d", a, f, l); + + /* + * We just need something here that even the most clever compiler + * cannot optimize away. + */ + return (a[(uintptr_t)f]); +} + +/* + * Atomically increment a specified error counter from probe context. + */ +static void +dtrace_error(uint32_t *counter) +{ + /* + * Most counters stored to in probe context are per-CPU counters. + * However, there are some error conditions that are sufficiently + * arcane that they don't merit per-CPU storage. If these counters + * are incremented concurrently on different CPUs, scalability will be + * adversely affected -- but we don't expect them to be white-hot in a + * correctly constructed enabling... + */ + uint32_t oval, nval; + + do { + oval = *counter; + + if ((nval = oval + 1) == 0) { + /* + * If the counter would wrap, set it to 1 -- assuring + * that the counter is never zero when we have seen + * errors. (The counter must be 32-bits because we + * aren't guaranteed a 64-bit compare&swap operation.) + * To save this code both the infamy of being fingered + * by a priggish news story and the indignity of being + * the target of a neo-puritan witch trial, we're + * carefully avoiding any colorful description of the + * likelihood of this condition -- but suffice it to + * say that it is only slightly more likely than the + * overflow of predicate cache IDs, as discussed in + * dtrace_predicate_create(). + */ + nval = 1; + } + } while (dtrace_cas32(counter, oval, nval) != oval); +} + +/* + * Use the DTRACE_LOADFUNC macro to define functions for each of loading a + * uint8_t, a uint16_t, a uint32_t and a uint64_t. + */ +DTRACE_LOADFUNC(8) +DTRACE_LOADFUNC(16) +DTRACE_LOADFUNC(32) +DTRACE_LOADFUNC(64) + +static int +dtrace_inscratch(uintptr_t dest, size_t size, dtrace_mstate_t *mstate) +{ + if (dest < mstate->dtms_scratch_base) + return (0); + + if (dest + size < dest) + return (0); + + if (dest + size > mstate->dtms_scratch_ptr) + return (0); + + return (1); +} + +static int +dtrace_canstore_statvar(uint64_t addr, size_t sz, + dtrace_statvar_t **svars, int nsvars) +{ + int i; + + for (i = 0; i < nsvars; i++) { + dtrace_statvar_t *svar = svars[i]; + + if (svar == NULL || svar->dtsv_size == 0) + continue; + + if (DTRACE_INRANGE(addr, sz, svar->dtsv_data, svar->dtsv_size)) + return (1); + } + + return (0); +} + +/* + * Check to see if the address is within a memory region to which a store may + * be issued. This includes the DTrace scratch areas, and any DTrace variable + * region. The caller of dtrace_canstore() is responsible for performing any + * alignment checks that are needed before stores are actually executed. + */ +static int +dtrace_canstore(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + /* + * First, check to see if the address is in scratch space... + */ + if (DTRACE_INRANGE(addr, sz, mstate->dtms_scratch_base, + mstate->dtms_scratch_size)) + return (1); + + /* + * Now check to see if it's a dynamic variable. This check will pick + * up both thread-local variables and any global dynamically-allocated + * variables. + */ + if (DTRACE_INRANGE(addr, sz, (uintptr_t)vstate->dtvs_dynvars.dtds_base, + vstate->dtvs_dynvars.dtds_size)) { + dtrace_dstate_t *dstate = &vstate->dtvs_dynvars; + uintptr_t base = (uintptr_t)dstate->dtds_base + + (dstate->dtds_hashsize * sizeof (dtrace_dynhash_t)); + uintptr_t chunkoffs; + + /* + * Before we assume that we can store here, we need to make + * sure that it isn't in our metadata -- storing to our + * dynamic variable metadata would corrupt our state. For + * the range to not include any dynamic variable metadata, + * it must: + * + * (1) Start above the hash table that is at the base of + * the dynamic variable space + * + * (2) Have a starting chunk offset that is beyond the + * dtrace_dynvar_t that is at the base of every chunk + * + * (3) Not span a chunk boundary + * + */ + if (addr < base) + return (0); + + chunkoffs = (addr - base) % dstate->dtds_chunksize; + + if (chunkoffs < sizeof (dtrace_dynvar_t)) + return (0); + + if (chunkoffs + sz > dstate->dtds_chunksize) + return (0); + + return (1); + } + + /* + * Finally, check the static local and global variables. These checks + * take the longest, so we perform them last. + */ + if (dtrace_canstore_statvar(addr, sz, + vstate->dtvs_locals, vstate->dtvs_nlocals)) + return (1); + + if (dtrace_canstore_statvar(addr, sz, + vstate->dtvs_globals, vstate->dtvs_nglobals)) + return (1); + + return (0); +} + + +/* + * Convenience routine to check to see if the address is within a memory + * region in which a load may be issued given the user's privilege level; + * if not, it sets the appropriate error flags and loads 'addr' into the + * illegal value slot. + * + * DTrace subroutines (DIF_SUBR_*) should use this helper to implement + * appropriate memory access protection. + */ +static int +dtrace_canload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval; + + /* + * If we hold the privilege to read from kernel memory, then + * everything is readable. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return (1); + + /* + * You can obviously read that which you can store. + */ + if (dtrace_canstore(addr, sz, mstate, vstate)) + return (1); + + /* + * We're allowed to read from our own string table. + */ + if (DTRACE_INRANGE(addr, sz, (uintptr_t)mstate->dtms_difo->dtdo_strtab, + mstate->dtms_difo->dtdo_strlen)) + return (1); + + DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV); + *illval = addr; + return (0); +} + +/* + * Convenience routine to check to see if a given string is within a memory + * region in which a load may be issued given the user's privilege level; + * this exists so that we don't need to issue unnecessary dtrace_strlen() + * calls in the event that the user has all privileges. + */ +static int +dtrace_strcanload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + size_t strsz; + + /* + * If we hold the privilege to read from kernel memory, then + * everything is readable. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return (1); + + strsz = 1 + dtrace_strlen((char *)(uintptr_t)addr, sz); + if (dtrace_canload(addr, strsz, mstate, vstate)) + return (1); + + return (0); +} + +/* + * Convenience routine to check to see if a given variable is within a memory + * region in which a load may be issued given the user's privilege level. + */ +static int +dtrace_vcanload(void *src, dtrace_diftype_t *type, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + size_t sz; + ASSERT(type->dtdt_flags & DIF_TF_BYREF); + + /* + * If we hold the privilege to read from kernel memory, then + * everything is readable. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return (1); + + if (type->dtdt_kind == DIF_TYPE_STRING) + sz = dtrace_strlen(src, + vstate->dtvs_state->dts_options[DTRACEOPT_STRSIZE]) + 1; + else + sz = type->dtdt_size; + + return (dtrace_canload((uintptr_t)src, sz, mstate, vstate)); +} + +/* + * Compare two strings using safe loads. + */ +static int +dtrace_strncmp(char *s1, char *s2, size_t limit) +{ + uint8_t c1, c2; + volatile uint16_t *flags; + + if (s1 == s2 || limit == 0) + return (0); + + flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; + + do { + if (s1 == NULL) { + c1 = '\0'; + } else { + c1 = dtrace_load8((uintptr_t)s1++); + } + + if (s2 == NULL) { + c2 = '\0'; + } else { + c2 = dtrace_load8((uintptr_t)s2++); + } + + if (c1 != c2) + return (c1 - c2); + } while (--limit && c1 != '\0' && !(*flags & CPU_DTRACE_FAULT)); + + return (0); +} + +/* + * Compute strlen(s) for a string using safe memory accesses. The additional + * len parameter is used to specify a maximum length to ensure completion. + */ +static size_t +dtrace_strlen(const char *s, size_t lim) +{ + uint_t len; + + for (len = 0; len != lim; len++) { + if (dtrace_load8((uintptr_t)s++) == '\0') + break; + } + + return (len); +} + +/* + * Check if an address falls within a toxic region. + */ +static int +dtrace_istoxic(uintptr_t kaddr, size_t size) +{ + uintptr_t taddr, tsize; + int i; + + for (i = 0; i < dtrace_toxranges; i++) { + taddr = dtrace_toxrange[i].dtt_base; + tsize = dtrace_toxrange[i].dtt_limit - taddr; + + if (kaddr - taddr < tsize) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = kaddr; + return (1); + } + + if (taddr - kaddr < size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = taddr; + return (1); + } + } + + return (0); +} + +/* + * Copy src to dst using safe memory accesses. The src is assumed to be unsafe + * memory specified by the DIF program. The dst is assumed to be safe memory + * that we can store to directly because it is managed by DTrace. As with + * standard bcopy, overlapping copies are handled properly. + */ +static void +dtrace_bcopy(const void *src, void *dst, size_t len) +{ + if (len != 0) { + uint8_t *s1 = dst; + const uint8_t *s2 = src; + + if (s1 <= s2) { + do { + *s1++ = dtrace_load8((uintptr_t)s2++); + } while (--len != 0); + } else { + s2 += len; + s1 += len; + + do { + *--s1 = dtrace_load8((uintptr_t)--s2); + } while (--len != 0); + } + } +} + +/* + * Copy src to dst using safe memory accesses, up to either the specified + * length, or the point that a nul byte is encountered. The src is assumed to + * be unsafe memory specified by the DIF program. The dst is assumed to be + * safe memory that we can store to directly because it is managed by DTrace. + * Unlike dtrace_bcopy(), overlapping regions are not handled. + */ +static void +dtrace_strcpy(const void *src, void *dst, size_t len) +{ + if (len != 0) { + uint8_t *s1 = dst, c; + const uint8_t *s2 = src; + + do { + *s1++ = c = dtrace_load8((uintptr_t)s2++); + } while (--len != 0 && c != '\0'); + } +} + +/* + * Copy src to dst, deriving the size and type from the specified (BYREF) + * variable type. The src is assumed to be unsafe memory specified by the DIF + * program. The dst is assumed to be DTrace variable memory that is of the + * specified type; we assume that we can store to directly. + */ +static void +dtrace_vcopy(void *src, void *dst, dtrace_diftype_t *type) +{ + ASSERT(type->dtdt_flags & DIF_TF_BYREF); + + if (type->dtdt_kind == DIF_TYPE_STRING) { + dtrace_strcpy(src, dst, type->dtdt_size); + } else { + dtrace_bcopy(src, dst, type->dtdt_size); + } +} + +/* + * Compare s1 to s2 using safe memory accesses. The s1 data is assumed to be + * unsafe memory specified by the DIF program. The s2 data is assumed to be + * safe memory that we can access directly because it is managed by DTrace. + */ +static int +dtrace_bcmp(const void *s1, const void *s2, size_t len) +{ + volatile uint16_t *flags; + + flags = (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; + + if (s1 == s2) + return (0); + + if (s1 == NULL || s2 == NULL) + return (1); + + if (s1 != s2 && len != 0) { + const uint8_t *ps1 = s1; + const uint8_t *ps2 = s2; + + do { + if (dtrace_load8((uintptr_t)ps1++) != *ps2++) + return (1); + } while (--len != 0 && !(*flags & CPU_DTRACE_FAULT)); + } + return (0); +} + +/* + * Zero the specified region using a simple byte-by-byte loop. Note that this + * is for safe DTrace-managed memory only. + */ +static void +dtrace_bzero(void *dst, size_t len) +{ + uchar_t *cp; + + for (cp = dst; len != 0; len--) + *cp++ = 0; +} + +static void +dtrace_add_128(uint64_t *addend1, uint64_t *addend2, uint64_t *sum) +{ + uint64_t result[2]; + + result[0] = addend1[0] + addend2[0]; + result[1] = addend1[1] + addend2[1] + + (result[0] < addend1[0] || result[0] < addend2[0] ? 1 : 0); + + sum[0] = result[0]; + sum[1] = result[1]; +} + +/* + * Shift the 128-bit value in a by b. If b is positive, shift left. + * If b is negative, shift right. + */ +static void +dtrace_shift_128(uint64_t *a, int b) +{ + uint64_t mask; + + if (b == 0) + return; + + if (b < 0) { + b = -b; + if (b >= 64) { + a[0] = a[1] >> (b - 64); + a[1] = 0; + } else { + a[0] >>= b; + mask = 1LL << (64 - b); + mask -= 1; + a[0] |= ((a[1] & mask) << (64 - b)); + a[1] >>= b; + } + } else { + if (b >= 64) { + a[1] = a[0] << (b - 64); + a[0] = 0; + } else { + a[1] <<= b; + mask = a[0] >> (64 - b); + a[1] |= mask; + a[0] <<= b; + } + } +} + +/* + * The basic idea is to break the 2 64-bit values into 4 32-bit values, + * use native multiplication on those, and then re-combine into the + * resulting 128-bit value. + * + * (hi1 << 32 + lo1) * (hi2 << 32 + lo2) = + * hi1 * hi2 << 64 + + * hi1 * lo2 << 32 + + * hi2 * lo1 << 32 + + * lo1 * lo2 + */ +static void +dtrace_multiply_128(uint64_t factor1, uint64_t factor2, uint64_t *product) +{ + uint64_t hi1, hi2, lo1, lo2; + uint64_t tmp[2]; + + hi1 = factor1 >> 32; + hi2 = factor2 >> 32; + + lo1 = factor1 & DT_MASK_LO; + lo2 = factor2 & DT_MASK_LO; + + product[0] = lo1 * lo2; + product[1] = hi1 * hi2; + + tmp[0] = hi1 * lo2; + tmp[1] = 0; + dtrace_shift_128(tmp, 32); + dtrace_add_128(product, tmp, product); + + tmp[0] = hi2 * lo1; + tmp[1] = 0; + dtrace_shift_128(tmp, 32); + dtrace_add_128(product, tmp, product); +} + +/* + * This privilege check should be used by actions and subroutines to + * verify that the user credentials of the process that enabled the + * invoking ECB match the target credentials + */ +static int +dtrace_priv_proc_common_user(dtrace_state_t *state) +{ + cred_t *cr, *s_cr = state->dts_cred.dcr_cred; + + /* + * We should always have a non-NULL state cred here, since if cred + * is null (anonymous tracing), we fast-path bypass this routine. + */ + ASSERT(s_cr != NULL); + + if ((cr = CRED()) != NULL && + s_cr->cr_uid == cr->cr_uid && + s_cr->cr_uid == cr->cr_ruid && + s_cr->cr_uid == cr->cr_suid && + s_cr->cr_gid == cr->cr_gid && + s_cr->cr_gid == cr->cr_rgid && + s_cr->cr_gid == cr->cr_sgid) + return (1); + + return (0); +} + +/* + * This privilege check should be used by actions and subroutines to + * verify that the zone of the process that enabled the invoking ECB + * matches the target credentials + */ +static int +dtrace_priv_proc_common_zone(dtrace_state_t *state) +{ +#if defined(sun) + cred_t *cr, *s_cr = state->dts_cred.dcr_cred; + + /* + * We should always have a non-NULL state cred here, since if cred + * is null (anonymous tracing), we fast-path bypass this routine. + */ + ASSERT(s_cr != NULL); + + if ((cr = CRED()) != NULL && + s_cr->cr_zone == cr->cr_zone) + return (1); + + return (0); +#else + return (1); +#endif +} + +/* + * This privilege check should be used by actions and subroutines to + * verify that the process has not setuid or changed credentials. + */ +static int +dtrace_priv_proc_common_nocd(void) +{ + proc_t *proc; + + if ((proc = ttoproc(curthread)) != NULL && + !(proc->p_flag & SNOCD)) + return (1); + + return (0); +} + +static int +dtrace_priv_proc_destructive(dtrace_state_t *state) +{ + int action = state->dts_cred.dcr_action; + + if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE) == 0) && + dtrace_priv_proc_common_zone(state) == 0) + goto bad; + + if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER) == 0) && + dtrace_priv_proc_common_user(state) == 0) + goto bad; + + if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG) == 0) && + dtrace_priv_proc_common_nocd() == 0) + goto bad; + + return (1); + +bad: + cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; + + return (0); +} + +static int +dtrace_priv_proc_control(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_PROC_CONTROL) + return (1); + + if (dtrace_priv_proc_common_zone(state) && + dtrace_priv_proc_common_user(state) && + dtrace_priv_proc_common_nocd()) + return (1); + + cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; + + return (0); +} + +static int +dtrace_priv_proc(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_PROC) + return (1); + + cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; + + return (0); +} + +static int +dtrace_priv_kernel(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL) + return (1); + + cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV; + + return (0); +} + +static int +dtrace_priv_kernel_destructive(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL_DESTRUCTIVE) + return (1); + + cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV; + + return (0); +} + +/* + * Note: not called from probe context. This function is called + * asynchronously (and at a regular interval) from outside of probe context to + * clean the dirty dynamic variable lists on all CPUs. Dynamic variable + * cleaning is explained in detail in <sys/dtrace_impl.h>. + */ +void +dtrace_dynvar_clean(dtrace_dstate_t *dstate) +{ + dtrace_dynvar_t *dirty; + dtrace_dstate_percpu_t *dcpu; + int i, work = 0; + + for (i = 0; i < NCPU; i++) { + dcpu = &dstate->dtds_percpu[i]; + + ASSERT(dcpu->dtdsc_rinsing == NULL); + + /* + * If the dirty list is NULL, there is no dirty work to do. + */ + if (dcpu->dtdsc_dirty == NULL) + continue; + + /* + * If the clean list is non-NULL, then we're not going to do + * any work for this CPU -- it means that there has not been + * a dtrace_dynvar() allocation on this CPU (or from this CPU) + * since the last time we cleaned house. + */ + if (dcpu->dtdsc_clean != NULL) + continue; + + work = 1; + + /* + * Atomically move the dirty list aside. + */ + do { + dirty = dcpu->dtdsc_dirty; + + /* + * Before we zap the dirty list, set the rinsing list. + * (This allows for a potential assertion in + * dtrace_dynvar(): if a free dynamic variable appears + * on a hash chain, either the dirty list or the + * rinsing list for some CPU must be non-NULL.) + */ + dcpu->dtdsc_rinsing = dirty; + dtrace_membar_producer(); + } while (dtrace_casptr(&dcpu->dtdsc_dirty, + dirty, NULL) != dirty); + } + + if (!work) { + /* + * We have no work to do; we can simply return. + */ + return; + } + + dtrace_sync(); + + for (i = 0; i < NCPU; i++) { + dcpu = &dstate->dtds_percpu[i]; + + if (dcpu->dtdsc_rinsing == NULL) + continue; + + /* + * We are now guaranteed that no hash chain contains a pointer + * into this dirty list; we can make it clean. + */ + ASSERT(dcpu->dtdsc_clean == NULL); + dcpu->dtdsc_clean = dcpu->dtdsc_rinsing; + dcpu->dtdsc_rinsing = NULL; + } + + /* + * Before we actually set the state to be DTRACE_DSTATE_CLEAN, make + * sure that all CPUs have seen all of the dtdsc_clean pointers. + * This prevents a race whereby a CPU incorrectly decides that + * the state should be something other than DTRACE_DSTATE_CLEAN + * after dtrace_dynvar_clean() has completed. + */ + dtrace_sync(); + + dstate->dtds_state = DTRACE_DSTATE_CLEAN; +} + +/* + * Depending on the value of the op parameter, this function looks-up, + * allocates or deallocates an arbitrarily-keyed dynamic variable. If an + * allocation is requested, this function will return a pointer to a + * dtrace_dynvar_t corresponding to the allocated variable -- or NULL if no + * variable can be allocated. If NULL is returned, the appropriate counter + * will be incremented. + */ +dtrace_dynvar_t * +dtrace_dynvar(dtrace_dstate_t *dstate, uint_t nkeys, + dtrace_key_t *key, size_t dsize, dtrace_dynvar_op_t op, + dtrace_mstate_t *mstate, dtrace_vstate_t *vstate) +{ + uint64_t hashval = DTRACE_DYNHASH_VALID; + dtrace_dynhash_t *hash = dstate->dtds_hash; + dtrace_dynvar_t *free, *new_free, *next, *dvar, *start, *prev = NULL; + processorid_t me = curcpu, cpu = me; + dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[me]; + size_t bucket, ksize; + size_t chunksize = dstate->dtds_chunksize; + uintptr_t kdata, lock, nstate; + uint_t i; + + ASSERT(nkeys != 0); + + /* + * Hash the key. As with aggregations, we use Jenkins' "One-at-a-time" + * algorithm. For the by-value portions, we perform the algorithm in + * 16-bit chunks (as opposed to 8-bit chunks). This speeds things up a + * bit, and seems to have only a minute effect on distribution. For + * the by-reference data, we perform "One-at-a-time" iterating (safely) + * over each referenced byte. It's painful to do this, but it's much + * better than pathological hash distribution. The efficacy of the + * hashing algorithm (and a comparison with other algorithms) may be + * found by running the ::dtrace_dynstat MDB dcmd. + */ + for (i = 0; i < nkeys; i++) { + if (key[i].dttk_size == 0) { + uint64_t val = key[i].dttk_value; + + hashval += (val >> 48) & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + hashval += (val >> 32) & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + hashval += (val >> 16) & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + hashval += val & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + } else { + /* + * This is incredibly painful, but it beats the hell + * out of the alternative. + */ + uint64_t j, size = key[i].dttk_size; + uintptr_t base = (uintptr_t)key[i].dttk_value; + + if (!dtrace_canload(base, size, mstate, vstate)) + break; + + for (j = 0; j < size; j++) { + hashval += dtrace_load8(base + j); + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + } + } + } + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT)) + return (NULL); + + hashval += (hashval << 3); + hashval ^= (hashval >> 11); + hashval += (hashval << 15); + + /* + * There is a remote chance (ideally, 1 in 2^31) that our hashval + * comes out to be one of our two sentinel hash values. If this + * actually happens, we set the hashval to be a value known to be a + * non-sentinel value. + */ + if (hashval == DTRACE_DYNHASH_FREE || hashval == DTRACE_DYNHASH_SINK) + hashval = DTRACE_DYNHASH_VALID; + + /* + * Yes, it's painful to do a divide here. If the cycle count becomes + * important here, tricks can be pulled to reduce it. (However, it's + * critical that hash collisions be kept to an absolute minimum; + * they're much more painful than a divide.) It's better to have a + * solution that generates few collisions and still keeps things + * relatively simple. + */ + bucket = hashval % dstate->dtds_hashsize; + + if (op == DTRACE_DYNVAR_DEALLOC) { + volatile uintptr_t *lockp = &hash[bucket].dtdh_lock; + + for (;;) { + while ((lock = *lockp) & 1) + continue; + + if (dtrace_casptr((volatile void *)lockp, + (volatile void *)lock, (volatile void *)(lock + 1)) == (void *)lock) + break; + } + + dtrace_membar_producer(); + } + +top: + prev = NULL; + lock = hash[bucket].dtdh_lock; + + dtrace_membar_consumer(); + + start = hash[bucket].dtdh_chain; + ASSERT(start != NULL && (start->dtdv_hashval == DTRACE_DYNHASH_SINK || + start->dtdv_hashval != DTRACE_DYNHASH_FREE || + op != DTRACE_DYNVAR_DEALLOC)); + + for (dvar = start; dvar != NULL; dvar = dvar->dtdv_next) { + dtrace_tuple_t *dtuple = &dvar->dtdv_tuple; + dtrace_key_t *dkey = &dtuple->dtt_key[0]; + + if (dvar->dtdv_hashval != hashval) { + if (dvar->dtdv_hashval == DTRACE_DYNHASH_SINK) { + /* + * We've reached the sink, and therefore the + * end of the hash chain; we can kick out of + * the loop knowing that we have seen a valid + * snapshot of state. + */ + ASSERT(dvar->dtdv_next == NULL); + ASSERT(dvar == &dtrace_dynhash_sink); + break; + } + + if (dvar->dtdv_hashval == DTRACE_DYNHASH_FREE) { + /* + * We've gone off the rails: somewhere along + * the line, one of the members of this hash + * chain was deleted. Note that we could also + * detect this by simply letting this loop run + * to completion, as we would eventually hit + * the end of the dirty list. However, we + * want to avoid running the length of the + * dirty list unnecessarily (it might be quite + * long), so we catch this as early as + * possible by detecting the hash marker. In + * this case, we simply set dvar to NULL and + * break; the conditional after the loop will + * send us back to top. + */ + dvar = NULL; + break; + } + + goto next; + } + + if (dtuple->dtt_nkeys != nkeys) + goto next; + + for (i = 0; i < nkeys; i++, dkey++) { + if (dkey->dttk_size != key[i].dttk_size) + goto next; /* size or type mismatch */ + + if (dkey->dttk_size != 0) { + if (dtrace_bcmp( + (void *)(uintptr_t)key[i].dttk_value, + (void *)(uintptr_t)dkey->dttk_value, + dkey->dttk_size)) + goto next; + } else { + if (dkey->dttk_value != key[i].dttk_value) + goto next; + } + } + + if (op != DTRACE_DYNVAR_DEALLOC) + return (dvar); + + ASSERT(dvar->dtdv_next == NULL || + dvar->dtdv_next->dtdv_hashval != DTRACE_DYNHASH_FREE); + + if (prev != NULL) { + ASSERT(hash[bucket].dtdh_chain != dvar); + ASSERT(start != dvar); + ASSERT(prev->dtdv_next == dvar); + prev->dtdv_next = dvar->dtdv_next; + } else { + if (dtrace_casptr(&hash[bucket].dtdh_chain, + start, dvar->dtdv_next) != start) { + /* + * We have failed to atomically swing the + * hash table head pointer, presumably because + * of a conflicting allocation on another CPU. + * We need to reread the hash chain and try + * again. + */ + goto top; + } + } + + dtrace_membar_producer(); + + /* + * Now set the hash value to indicate that it's free. + */ + ASSERT(hash[bucket].dtdh_chain != dvar); + dvar->dtdv_hashval = DTRACE_DYNHASH_FREE; + + dtrace_membar_producer(); + + /* + * Set the next pointer to point at the dirty list, and + * atomically swing the dirty pointer to the newly freed dvar. + */ + do { + next = dcpu->dtdsc_dirty; + dvar->dtdv_next = next; + } while (dtrace_casptr(&dcpu->dtdsc_dirty, next, dvar) != next); + + /* + * Finally, unlock this hash bucket. + */ + ASSERT(hash[bucket].dtdh_lock == lock); + ASSERT(lock & 1); + hash[bucket].dtdh_lock++; + + return (NULL); +next: + prev = dvar; + continue; + } + + if (dvar == NULL) { + /* + * If dvar is NULL, it is because we went off the rails: + * one of the elements that we traversed in the hash chain + * was deleted while we were traversing it. In this case, + * we assert that we aren't doing a dealloc (deallocs lock + * the hash bucket to prevent themselves from racing with + * one another), and retry the hash chain traversal. + */ + ASSERT(op != DTRACE_DYNVAR_DEALLOC); + goto top; + } + + if (op != DTRACE_DYNVAR_ALLOC) { + /* + * If we are not to allocate a new variable, we want to + * return NULL now. Before we return, check that the value + * of the lock word hasn't changed. If it has, we may have + * seen an inconsistent snapshot. + */ + if (op == DTRACE_DYNVAR_NOALLOC) { + if (hash[bucket].dtdh_lock != lock) + goto top; + } else { + ASSERT(op == DTRACE_DYNVAR_DEALLOC); + ASSERT(hash[bucket].dtdh_lock == lock); + ASSERT(lock & 1); + hash[bucket].dtdh_lock++; + } + + return (NULL); + } + + /* + * We need to allocate a new dynamic variable. The size we need is the + * size of dtrace_dynvar plus the size of nkeys dtrace_key_t's plus the + * size of any auxiliary key data (rounded up to 8-byte alignment) plus + * the size of any referred-to data (dsize). We then round the final + * size up to the chunksize for allocation. + */ + for (ksize = 0, i = 0; i < nkeys; i++) + ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t)); + + /* + * This should be pretty much impossible, but could happen if, say, + * strange DIF specified the tuple. Ideally, this should be an + * assertion and not an error condition -- but that requires that the + * chunksize calculation in dtrace_difo_chunksize() be absolutely + * bullet-proof. (That is, it must not be able to be fooled by + * malicious DIF.) Given the lack of backwards branches in DIF, + * solving this would presumably not amount to solving the Halting + * Problem -- but it still seems awfully hard. + */ + if (sizeof (dtrace_dynvar_t) + sizeof (dtrace_key_t) * (nkeys - 1) + + ksize + dsize > chunksize) { + dcpu->dtdsc_drops++; + return (NULL); + } + + nstate = DTRACE_DSTATE_EMPTY; + + do { +retry: + free = dcpu->dtdsc_free; + + if (free == NULL) { + dtrace_dynvar_t *clean = dcpu->dtdsc_clean; + void *rval; + + if (clean == NULL) { + /* + * We're out of dynamic variable space on + * this CPU. Unless we have tried all CPUs, + * we'll try to allocate from a different + * CPU. + */ + switch (dstate->dtds_state) { + case DTRACE_DSTATE_CLEAN: { + void *sp = &dstate->dtds_state; + + if (++cpu >= NCPU) + cpu = 0; + + if (dcpu->dtdsc_dirty != NULL && + nstate == DTRACE_DSTATE_EMPTY) + nstate = DTRACE_DSTATE_DIRTY; + + if (dcpu->dtdsc_rinsing != NULL) + nstate = DTRACE_DSTATE_RINSING; + + dcpu = &dstate->dtds_percpu[cpu]; + + if (cpu != me) + goto retry; + + (void) dtrace_cas32(sp, + DTRACE_DSTATE_CLEAN, nstate); + + /* + * To increment the correct bean + * counter, take another lap. + */ + goto retry; + } + + case DTRACE_DSTATE_DIRTY: + dcpu->dtdsc_dirty_drops++; + break; + + case DTRACE_DSTATE_RINSING: + dcpu->dtdsc_rinsing_drops++; + break; + + case DTRACE_DSTATE_EMPTY: + dcpu->dtdsc_drops++; + break; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_DROP); + return (NULL); + } + + /* + * The clean list appears to be non-empty. We want to + * move the clean list to the free list; we start by + * moving the clean pointer aside. + */ + if (dtrace_casptr(&dcpu->dtdsc_clean, + clean, NULL) != clean) { + /* + * We are in one of two situations: + * + * (a) The clean list was switched to the + * free list by another CPU. + * + * (b) The clean list was added to by the + * cleansing cyclic. + * + * In either of these situations, we can + * just reattempt the free list allocation. + */ + goto retry; + } + + ASSERT(clean->dtdv_hashval == DTRACE_DYNHASH_FREE); + + /* + * Now we'll move the clean list to the free list. + * It's impossible for this to fail: the only way + * the free list can be updated is through this + * code path, and only one CPU can own the clean list. + * Thus, it would only be possible for this to fail if + * this code were racing with dtrace_dynvar_clean(). + * (That is, if dtrace_dynvar_clean() updated the clean + * list, and we ended up racing to update the free + * list.) This race is prevented by the dtrace_sync() + * in dtrace_dynvar_clean() -- which flushes the + * owners of the clean lists out before resetting + * the clean lists. + */ + rval = dtrace_casptr(&dcpu->dtdsc_free, NULL, clean); + ASSERT(rval == NULL); + goto retry; + } + + dvar = free; + new_free = dvar->dtdv_next; + } while (dtrace_casptr(&dcpu->dtdsc_free, free, new_free) != free); + + /* + * We have now allocated a new chunk. We copy the tuple keys into the + * tuple array and copy any referenced key data into the data space + * following the tuple array. As we do this, we relocate dttk_value + * in the final tuple to point to the key data address in the chunk. + */ + kdata = (uintptr_t)&dvar->dtdv_tuple.dtt_key[nkeys]; + dvar->dtdv_data = (void *)(kdata + ksize); + dvar->dtdv_tuple.dtt_nkeys = nkeys; + + for (i = 0; i < nkeys; i++) { + dtrace_key_t *dkey = &dvar->dtdv_tuple.dtt_key[i]; + size_t kesize = key[i].dttk_size; + + if (kesize != 0) { + dtrace_bcopy( + (const void *)(uintptr_t)key[i].dttk_value, + (void *)kdata, kesize); + dkey->dttk_value = kdata; + kdata += P2ROUNDUP(kesize, sizeof (uint64_t)); + } else { + dkey->dttk_value = key[i].dttk_value; + } + + dkey->dttk_size = kesize; + } + + ASSERT(dvar->dtdv_hashval == DTRACE_DYNHASH_FREE); + dvar->dtdv_hashval = hashval; + dvar->dtdv_next = start; + + if (dtrace_casptr(&hash[bucket].dtdh_chain, start, dvar) == start) + return (dvar); + + /* + * The cas has failed. Either another CPU is adding an element to + * this hash chain, or another CPU is deleting an element from this + * hash chain. The simplest way to deal with both of these cases + * (though not necessarily the most efficient) is to free our + * allocated block and tail-call ourselves. Note that the free is + * to the dirty list and _not_ to the free list. This is to prevent + * races with allocators, above. + */ + dvar->dtdv_hashval = DTRACE_DYNHASH_FREE; + + dtrace_membar_producer(); + + do { + free = dcpu->dtdsc_dirty; + dvar->dtdv_next = free; + } while (dtrace_casptr(&dcpu->dtdsc_dirty, free, dvar) != free); + + return (dtrace_dynvar(dstate, nkeys, key, dsize, op, mstate, vstate)); +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_min(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + if ((int64_t)nval < (int64_t)*oval) + *oval = nval; +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_max(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + if ((int64_t)nval > (int64_t)*oval) + *oval = nval; +} + +static void +dtrace_aggregate_quantize(uint64_t *quanta, uint64_t nval, uint64_t incr) +{ + int i, zero = DTRACE_QUANTIZE_ZEROBUCKET; + int64_t val = (int64_t)nval; + + if (val < 0) { + for (i = 0; i < zero; i++) { + if (val <= DTRACE_QUANTIZE_BUCKETVAL(i)) { + quanta[i] += incr; + return; + } + } + } else { + for (i = zero + 1; i < DTRACE_QUANTIZE_NBUCKETS; i++) { + if (val < DTRACE_QUANTIZE_BUCKETVAL(i)) { + quanta[i - 1] += incr; + return; + } + } + + quanta[DTRACE_QUANTIZE_NBUCKETS - 1] += incr; + return; + } + + ASSERT(0); +} + +static void +dtrace_aggregate_lquantize(uint64_t *lquanta, uint64_t nval, uint64_t incr) +{ + uint64_t arg = *lquanta++; + int32_t base = DTRACE_LQUANTIZE_BASE(arg); + uint16_t step = DTRACE_LQUANTIZE_STEP(arg); + uint16_t levels = DTRACE_LQUANTIZE_LEVELS(arg); + int32_t val = (int32_t)nval, level; + + ASSERT(step != 0); + ASSERT(levels != 0); + + if (val < base) { + /* + * This is an underflow. + */ + lquanta[0] += incr; + return; + } + + level = (val - base) / step; + + if (level < levels) { + lquanta[level + 1] += incr; + return; + } + + /* + * This is an overflow. + */ + lquanta[levels + 1] += incr; +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_avg(uint64_t *data, uint64_t nval, uint64_t arg) +{ + data[0]++; + data[1] += nval; +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_stddev(uint64_t *data, uint64_t nval, uint64_t arg) +{ + int64_t snval = (int64_t)nval; + uint64_t tmp[2]; + + data[0]++; + data[1] += nval; + + /* + * What we want to say here is: + * + * data[2] += nval * nval; + * + * But given that nval is 64-bit, we could easily overflow, so + * we do this as 128-bit arithmetic. + */ + if (snval < 0) + snval = -snval; + + dtrace_multiply_128((uint64_t)snval, (uint64_t)snval, tmp); + dtrace_add_128(data + 2, tmp, data + 2); +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_count(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + *oval = *oval + 1; +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_sum(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + *oval += nval; +} + +/* + * Aggregate given the tuple in the principal data buffer, and the aggregating + * action denoted by the specified dtrace_aggregation_t. The aggregation + * buffer is specified as the buf parameter. This routine does not return + * failure; if there is no space in the aggregation buffer, the data will be + * dropped, and a corresponding counter incremented. + */ +static void +dtrace_aggregate(dtrace_aggregation_t *agg, dtrace_buffer_t *dbuf, + intptr_t offset, dtrace_buffer_t *buf, uint64_t expr, uint64_t arg) +{ + dtrace_recdesc_t *rec = &agg->dtag_action.dta_rec; + uint32_t i, ndx, size, fsize; + uint32_t align = sizeof (uint64_t) - 1; + dtrace_aggbuffer_t *agb; + dtrace_aggkey_t *key; + uint32_t hashval = 0, limit, isstr; + caddr_t tomax, data, kdata; + dtrace_actkind_t action; + dtrace_action_t *act; + uintptr_t offs; + + if (buf == NULL) + return; + + if (!agg->dtag_hasarg) { + /* + * Currently, only quantize() and lquantize() take additional + * arguments, and they have the same semantics: an increment + * value that defaults to 1 when not present. If additional + * aggregating actions take arguments, the setting of the + * default argument value will presumably have to become more + * sophisticated... + */ + arg = 1; + } + + action = agg->dtag_action.dta_kind - DTRACEACT_AGGREGATION; + size = rec->dtrd_offset - agg->dtag_base; + fsize = size + rec->dtrd_size; + + ASSERT(dbuf->dtb_tomax != NULL); + data = dbuf->dtb_tomax + offset + agg->dtag_base; + + if ((tomax = buf->dtb_tomax) == NULL) { + dtrace_buffer_drop(buf); + return; + } + + /* + * The metastructure is always at the bottom of the buffer. + */ + agb = (dtrace_aggbuffer_t *)(tomax + buf->dtb_size - + sizeof (dtrace_aggbuffer_t)); + + if (buf->dtb_offset == 0) { + /* + * We just kludge up approximately 1/8th of the size to be + * buckets. If this guess ends up being routinely + * off-the-mark, we may need to dynamically readjust this + * based on past performance. + */ + uintptr_t hashsize = (buf->dtb_size >> 3) / sizeof (uintptr_t); + + if ((uintptr_t)agb - hashsize * sizeof (dtrace_aggkey_t *) < + (uintptr_t)tomax || hashsize == 0) { + /* + * We've been given a ludicrously small buffer; + * increment our drop count and leave. + */ + dtrace_buffer_drop(buf); + return; + } + + /* + * And now, a pathetic attempt to try to get a an odd (or + * perchance, a prime) hash size for better hash distribution. + */ + if (hashsize > (DTRACE_AGGHASHSIZE_SLEW << 3)) + hashsize -= DTRACE_AGGHASHSIZE_SLEW; + + agb->dtagb_hashsize = hashsize; + agb->dtagb_hash = (dtrace_aggkey_t **)((uintptr_t)agb - + agb->dtagb_hashsize * sizeof (dtrace_aggkey_t *)); + agb->dtagb_free = (uintptr_t)agb->dtagb_hash; + + for (i = 0; i < agb->dtagb_hashsize; i++) + agb->dtagb_hash[i] = NULL; + } + + ASSERT(agg->dtag_first != NULL); + ASSERT(agg->dtag_first->dta_intuple); + + /* + * Calculate the hash value based on the key. Note that we _don't_ + * include the aggid in the hashing (but we will store it as part of + * the key). The hashing algorithm is Bob Jenkins' "One-at-a-time" + * algorithm: a simple, quick algorithm that has no known funnels, and + * gets good distribution in practice. The efficacy of the hashing + * algorithm (and a comparison with other algorithms) may be found by + * running the ::dtrace_aggstat MDB dcmd. + */ + for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) { + i = act->dta_rec.dtrd_offset - agg->dtag_base; + limit = i + act->dta_rec.dtrd_size; + ASSERT(limit <= size); + isstr = DTRACEACT_ISSTRING(act); + + for (; i < limit; i++) { + hashval += data[i]; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + if (isstr && data[i] == '\0') + break; + } + } + + hashval += (hashval << 3); + hashval ^= (hashval >> 11); + hashval += (hashval << 15); + + /* + * Yes, the divide here is expensive -- but it's generally the least + * of the performance issues given the amount of data that we iterate + * over to compute hash values, compare data, etc. + */ + ndx = hashval % agb->dtagb_hashsize; + + for (key = agb->dtagb_hash[ndx]; key != NULL; key = key->dtak_next) { + ASSERT((caddr_t)key >= tomax); + ASSERT((caddr_t)key < tomax + buf->dtb_size); + + if (hashval != key->dtak_hashval || key->dtak_size != size) + continue; + + kdata = key->dtak_data; + ASSERT(kdata >= tomax && kdata < tomax + buf->dtb_size); + + for (act = agg->dtag_first; act->dta_intuple; + act = act->dta_next) { + i = act->dta_rec.dtrd_offset - agg->dtag_base; + limit = i + act->dta_rec.dtrd_size; + ASSERT(limit <= size); + isstr = DTRACEACT_ISSTRING(act); + + for (; i < limit; i++) { + if (kdata[i] != data[i]) + goto next; + + if (isstr && data[i] == '\0') + break; + } + } + + if (action != key->dtak_action) { + /* + * We are aggregating on the same value in the same + * aggregation with two different aggregating actions. + * (This should have been picked up in the compiler, + * so we may be dealing with errant or devious DIF.) + * This is an error condition; we indicate as much, + * and return. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return; + } + + /* + * This is a hit: we need to apply the aggregator to + * the value at this key. + */ + agg->dtag_aggregate((uint64_t *)(kdata + size), expr, arg); + return; +next: + continue; + } + + /* + * We didn't find it. We need to allocate some zero-filled space, + * link it into the hash table appropriately, and apply the aggregator + * to the (zero-filled) value. + */ + offs = buf->dtb_offset; + while (offs & (align - 1)) + offs += sizeof (uint32_t); + + /* + * If we don't have enough room to both allocate a new key _and_ + * its associated data, increment the drop count and return. + */ + if ((uintptr_t)tomax + offs + fsize > + agb->dtagb_free - sizeof (dtrace_aggkey_t)) { + dtrace_buffer_drop(buf); + return; + } + + /*CONSTCOND*/ + ASSERT(!(sizeof (dtrace_aggkey_t) & (sizeof (uintptr_t) - 1))); + key = (dtrace_aggkey_t *)(agb->dtagb_free - sizeof (dtrace_aggkey_t)); + agb->dtagb_free -= sizeof (dtrace_aggkey_t); + + key->dtak_data = kdata = tomax + offs; + buf->dtb_offset = offs + fsize; + + /* + * Now copy the data across. + */ + *((dtrace_aggid_t *)kdata) = agg->dtag_id; + + for (i = sizeof (dtrace_aggid_t); i < size; i++) + kdata[i] = data[i]; + + /* + * Because strings are not zeroed out by default, we need to iterate + * looking for actions that store strings, and we need to explicitly + * pad these strings out with zeroes. + */ + for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) { + int nul; + + if (!DTRACEACT_ISSTRING(act)) + continue; + + i = act->dta_rec.dtrd_offset - agg->dtag_base; + limit = i + act->dta_rec.dtrd_size; + ASSERT(limit <= size); + + for (nul = 0; i < limit; i++) { + if (nul) { + kdata[i] = '\0'; + continue; + } + + if (data[i] != '\0') + continue; + + nul = 1; + } + } + + for (i = size; i < fsize; i++) + kdata[i] = 0; + + key->dtak_hashval = hashval; + key->dtak_size = size; + key->dtak_action = action; + key->dtak_next = agb->dtagb_hash[ndx]; + agb->dtagb_hash[ndx] = key; + + /* + * Finally, apply the aggregator. + */ + *((uint64_t *)(key->dtak_data + size)) = agg->dtag_initial; + agg->dtag_aggregate((uint64_t *)(key->dtak_data + size), expr, arg); +} + +/* + * Given consumer state, this routine finds a speculation in the INACTIVE + * state and transitions it into the ACTIVE state. If there is no speculation + * in the INACTIVE state, 0 is returned. In this case, no error counter is + * incremented -- it is up to the caller to take appropriate action. + */ +static int +dtrace_speculation(dtrace_state_t *state) +{ + int i = 0; + dtrace_speculation_state_t current; + uint32_t *stat = &state->dts_speculations_unavail, count; + + while (i < state->dts_nspeculations) { + dtrace_speculation_t *spec = &state->dts_speculations[i]; + + current = spec->dtsp_state; + + if (current != DTRACESPEC_INACTIVE) { + if (current == DTRACESPEC_COMMITTINGMANY || + current == DTRACESPEC_COMMITTING || + current == DTRACESPEC_DISCARDING) + stat = &state->dts_speculations_busy; + i++; + continue; + } + + if (dtrace_cas32((uint32_t *)&spec->dtsp_state, + current, DTRACESPEC_ACTIVE) == current) + return (i + 1); + } + + /* + * We couldn't find a speculation. If we found as much as a single + * busy speculation buffer, we'll attribute this failure as "busy" + * instead of "unavail". + */ + do { + count = *stat; + } while (dtrace_cas32(stat, count, count + 1) != count); + + return (0); +} + +/* + * This routine commits an active speculation. If the specified speculation + * is not in a valid state to perform a commit(), this routine will silently do + * nothing. The state of the specified speculation is transitioned according + * to the state transition diagram outlined in <sys/dtrace_impl.h> + */ +static void +dtrace_speculation_commit(dtrace_state_t *state, processorid_t cpu, + dtrace_specid_t which) +{ + dtrace_speculation_t *spec; + dtrace_buffer_t *src, *dest; + uintptr_t daddr, saddr, dlimit; + dtrace_speculation_state_t current, new = 0; + intptr_t offs; + + if (which == 0) + return; + + if (which > state->dts_nspeculations) { + cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return; + } + + spec = &state->dts_speculations[which - 1]; + src = &spec->dtsp_buffer[cpu]; + dest = &state->dts_buffer[cpu]; + + do { + current = spec->dtsp_state; + + if (current == DTRACESPEC_COMMITTINGMANY) + break; + + switch (current) { + case DTRACESPEC_INACTIVE: + case DTRACESPEC_DISCARDING: + return; + + case DTRACESPEC_COMMITTING: + /* + * This is only possible if we are (a) commit()'ing + * without having done a prior speculate() on this CPU + * and (b) racing with another commit() on a different + * CPU. There's nothing to do -- we just assert that + * our offset is 0. + */ + ASSERT(src->dtb_offset == 0); + return; + + case DTRACESPEC_ACTIVE: + new = DTRACESPEC_COMMITTING; + break; + + case DTRACESPEC_ACTIVEONE: + /* + * This speculation is active on one CPU. If our + * buffer offset is non-zero, we know that the one CPU + * must be us. Otherwise, we are committing on a + * different CPU from the speculate(), and we must + * rely on being asynchronously cleaned. + */ + if (src->dtb_offset != 0) { + new = DTRACESPEC_COMMITTING; + break; + } + /*FALLTHROUGH*/ + + case DTRACESPEC_ACTIVEMANY: + new = DTRACESPEC_COMMITTINGMANY; + break; + + default: + ASSERT(0); + } + } while (dtrace_cas32((uint32_t *)&spec->dtsp_state, + current, new) != current); + + /* + * We have set the state to indicate that we are committing this + * speculation. Now reserve the necessary space in the destination + * buffer. + */ + if ((offs = dtrace_buffer_reserve(dest, src->dtb_offset, + sizeof (uint64_t), state, NULL)) < 0) { + dtrace_buffer_drop(dest); + goto out; + } + + /* + * We have the space; copy the buffer across. (Note that this is a + * highly subobtimal bcopy(); in the unlikely event that this becomes + * a serious performance issue, a high-performance DTrace-specific + * bcopy() should obviously be invented.) + */ + daddr = (uintptr_t)dest->dtb_tomax + offs; + dlimit = daddr + src->dtb_offset; + saddr = (uintptr_t)src->dtb_tomax; + + /* + * First, the aligned portion. + */ + while (dlimit - daddr >= sizeof (uint64_t)) { + *((uint64_t *)daddr) = *((uint64_t *)saddr); + + daddr += sizeof (uint64_t); + saddr += sizeof (uint64_t); + } + + /* + * Now any left-over bit... + */ + while (dlimit - daddr) + *((uint8_t *)daddr++) = *((uint8_t *)saddr++); + + /* + * Finally, commit the reserved space in the destination buffer. + */ + dest->dtb_offset = offs + src->dtb_offset; + +out: + /* + * If we're lucky enough to be the only active CPU on this speculation + * buffer, we can just set the state back to DTRACESPEC_INACTIVE. + */ + if (current == DTRACESPEC_ACTIVE || + (current == DTRACESPEC_ACTIVEONE && new == DTRACESPEC_COMMITTING)) { + uint32_t rval = dtrace_cas32((uint32_t *)&spec->dtsp_state, + DTRACESPEC_COMMITTING, DTRACESPEC_INACTIVE); + + ASSERT(rval == DTRACESPEC_COMMITTING); + } + + src->dtb_offset = 0; + src->dtb_xamot_drops += src->dtb_drops; + src->dtb_drops = 0; +} + +/* + * This routine discards an active speculation. If the specified speculation + * is not in a valid state to perform a discard(), this routine will silently + * do nothing. The state of the specified speculation is transitioned + * according to the state transition diagram outlined in <sys/dtrace_impl.h> + */ +static void +dtrace_speculation_discard(dtrace_state_t *state, processorid_t cpu, + dtrace_specid_t which) +{ + dtrace_speculation_t *spec; + dtrace_speculation_state_t current, new = 0; + dtrace_buffer_t *buf; + + if (which == 0) + return; + + if (which > state->dts_nspeculations) { + cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return; + } + + spec = &state->dts_speculations[which - 1]; + buf = &spec->dtsp_buffer[cpu]; + + do { + current = spec->dtsp_state; + + switch (current) { + case DTRACESPEC_INACTIVE: + case DTRACESPEC_COMMITTINGMANY: + case DTRACESPEC_COMMITTING: + case DTRACESPEC_DISCARDING: + return; + + case DTRACESPEC_ACTIVE: + case DTRACESPEC_ACTIVEMANY: + new = DTRACESPEC_DISCARDING; + break; + + case DTRACESPEC_ACTIVEONE: + if (buf->dtb_offset != 0) { + new = DTRACESPEC_INACTIVE; + } else { + new = DTRACESPEC_DISCARDING; + } + break; + + default: + ASSERT(0); + } + } while (dtrace_cas32((uint32_t *)&spec->dtsp_state, + current, new) != current); + + buf->dtb_offset = 0; + buf->dtb_drops = 0; +} + +/* + * Note: not called from probe context. This function is called + * asynchronously from cross call context to clean any speculations that are + * in the COMMITTINGMANY or DISCARDING states. These speculations may not be + * transitioned back to the INACTIVE state until all CPUs have cleaned the + * speculation. + */ +static void +dtrace_speculation_clean_here(dtrace_state_t *state) +{ + dtrace_icookie_t cookie; + processorid_t cpu = curcpu; + dtrace_buffer_t *dest = &state->dts_buffer[cpu]; + dtrace_specid_t i; + + cookie = dtrace_interrupt_disable(); + + if (dest->dtb_tomax == NULL) { + dtrace_interrupt_enable(cookie); + return; + } + + for (i = 0; i < state->dts_nspeculations; i++) { + dtrace_speculation_t *spec = &state->dts_speculations[i]; + dtrace_buffer_t *src = &spec->dtsp_buffer[cpu]; + + if (src->dtb_tomax == NULL) + continue; + + if (spec->dtsp_state == DTRACESPEC_DISCARDING) { + src->dtb_offset = 0; + continue; + } + + if (spec->dtsp_state != DTRACESPEC_COMMITTINGMANY) + continue; + + if (src->dtb_offset == 0) + continue; + + dtrace_speculation_commit(state, cpu, i + 1); + } + + dtrace_interrupt_enable(cookie); +} + +/* + * Note: not called from probe context. This function is called + * asynchronously (and at a regular interval) to clean any speculations that + * are in the COMMITTINGMANY or DISCARDING states. If it discovers that there + * is work to be done, it cross calls all CPUs to perform that work; + * COMMITMANY and DISCARDING speculations may not be transitioned back to the + * INACTIVE state until they have been cleaned by all CPUs. + */ +static void +dtrace_speculation_clean(dtrace_state_t *state) +{ + int work = 0, rv; + dtrace_specid_t i; + + for (i = 0; i < state->dts_nspeculations; i++) { + dtrace_speculation_t *spec = &state->dts_speculations[i]; + + ASSERT(!spec->dtsp_cleaning); + + if (spec->dtsp_state != DTRACESPEC_DISCARDING && + spec->dtsp_state != DTRACESPEC_COMMITTINGMANY) + continue; + + work++; + spec->dtsp_cleaning = 1; + } + + if (!work) + return; + + dtrace_xcall(DTRACE_CPUALL, + (dtrace_xcall_t)dtrace_speculation_clean_here, state); + + /* + * We now know that all CPUs have committed or discarded their + * speculation buffers, as appropriate. We can now set the state + * to inactive. + */ + for (i = 0; i < state->dts_nspeculations; i++) { + dtrace_speculation_t *spec = &state->dts_speculations[i]; + dtrace_speculation_state_t current, new; + + if (!spec->dtsp_cleaning) + continue; + + current = spec->dtsp_state; + ASSERT(current == DTRACESPEC_DISCARDING || + current == DTRACESPEC_COMMITTINGMANY); + + new = DTRACESPEC_INACTIVE; + + rv = dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new); + ASSERT(rv == current); + spec->dtsp_cleaning = 0; + } +} + +/* + * Called as part of a speculate() to get the speculative buffer associated + * with a given speculation. Returns NULL if the specified speculation is not + * in an ACTIVE state. If the speculation is in the ACTIVEONE state -- and + * the active CPU is not the specified CPU -- the speculation will be + * atomically transitioned into the ACTIVEMANY state. + */ +static dtrace_buffer_t * +dtrace_speculation_buffer(dtrace_state_t *state, processorid_t cpuid, + dtrace_specid_t which) +{ + dtrace_speculation_t *spec; + dtrace_speculation_state_t current, new = 0; + dtrace_buffer_t *buf; + + if (which == 0) + return (NULL); + + if (which > state->dts_nspeculations) { + cpu_core[cpuid].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return (NULL); + } + + spec = &state->dts_speculations[which - 1]; + buf = &spec->dtsp_buffer[cpuid]; + + do { + current = spec->dtsp_state; + + switch (current) { + case DTRACESPEC_INACTIVE: + case DTRACESPEC_COMMITTINGMANY: + case DTRACESPEC_DISCARDING: + return (NULL); + + case DTRACESPEC_COMMITTING: + ASSERT(buf->dtb_offset == 0); + return (NULL); + + case DTRACESPEC_ACTIVEONE: + /* + * This speculation is currently active on one CPU. + * Check the offset in the buffer; if it's non-zero, + * that CPU must be us (and we leave the state alone). + * If it's zero, assume that we're starting on a new + * CPU -- and change the state to indicate that the + * speculation is active on more than one CPU. + */ + if (buf->dtb_offset != 0) + return (buf); + + new = DTRACESPEC_ACTIVEMANY; + break; + + case DTRACESPEC_ACTIVEMANY: + return (buf); + + case DTRACESPEC_ACTIVE: + new = DTRACESPEC_ACTIVEONE; + break; + + default: + ASSERT(0); + } + } while (dtrace_cas32((uint32_t *)&spec->dtsp_state, + current, new) != current); + + ASSERT(new == DTRACESPEC_ACTIVEONE || new == DTRACESPEC_ACTIVEMANY); + return (buf); +} + +/* + * Return a string. In the event that the user lacks the privilege to access + * arbitrary kernel memory, we copy the string out to scratch memory so that we + * don't fail access checking. + * + * dtrace_dif_variable() uses this routine as a helper for various + * builtin values such as 'execname' and 'probefunc.' + */ +uintptr_t +dtrace_dif_varstr(uintptr_t addr, dtrace_state_t *state, + dtrace_mstate_t *mstate) +{ + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t ret; + size_t strsz; + + /* + * The easy case: this probe is allowed to read all of memory, so + * we can just return this as a vanilla pointer. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return (addr); + + /* + * This is the tougher case: we copy the string in question from + * kernel memory into scratch memory and return it that way: this + * ensures that we won't trip up when access checking tests the + * BYREF return value. + */ + strsz = dtrace_strlen((char *)addr, size) + 1; + + if (mstate->dtms_scratch_ptr + strsz > + mstate->dtms_scratch_base + mstate->dtms_scratch_size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + return (0); + } + + dtrace_strcpy((const void *)addr, (void *)mstate->dtms_scratch_ptr, + strsz); + ret = mstate->dtms_scratch_ptr; + mstate->dtms_scratch_ptr += strsz; + return (ret); +} + +/* + * Return a string from a memoy address which is known to have one or + * more concatenated, individually zero terminated, sub-strings. + * In the event that the user lacks the privilege to access + * arbitrary kernel memory, we copy the string out to scratch memory so that we + * don't fail access checking. + * + * dtrace_dif_variable() uses this routine as a helper for various + * builtin values such as 'execargs'. + */ +static uintptr_t +dtrace_dif_varstrz(uintptr_t addr, size_t strsz, dtrace_state_t *state, + dtrace_mstate_t *mstate) +{ + char *p; + size_t i; + uintptr_t ret; + + if (mstate->dtms_scratch_ptr + strsz > + mstate->dtms_scratch_base + mstate->dtms_scratch_size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + return (0); + } + + dtrace_bcopy((const void *)addr, (void *)mstate->dtms_scratch_ptr, + strsz); + + /* Replace sub-string termination characters with a space. */ + for (p = (char *) mstate->dtms_scratch_ptr, i = 0; i < strsz - 1; + p++, i++) + if (*p == '\0') + *p = ' '; + + ret = mstate->dtms_scratch_ptr; + mstate->dtms_scratch_ptr += strsz; + return (ret); +} + +/* + * This function implements the DIF emulator's variable lookups. The emulator + * passes a reserved variable identifier and optional built-in array index. + */ +static uint64_t +dtrace_dif_variable(dtrace_mstate_t *mstate, dtrace_state_t *state, uint64_t v, + uint64_t ndx) +{ + /* + * If we're accessing one of the uncached arguments, we'll turn this + * into a reference in the args array. + */ + if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) { + ndx = v - DIF_VAR_ARG0; + v = DIF_VAR_ARGS; + } + + switch (v) { + case DIF_VAR_ARGS: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_ARGS); + if (ndx >= sizeof (mstate->dtms_arg) / + sizeof (mstate->dtms_arg[0])) { + int aframes = mstate->dtms_probe->dtpr_aframes + 2; + dtrace_provider_t *pv; + uint64_t val; + + pv = mstate->dtms_probe->dtpr_provider; + if (pv->dtpv_pops.dtps_getargval != NULL) + val = pv->dtpv_pops.dtps_getargval(pv->dtpv_arg, + mstate->dtms_probe->dtpr_id, + mstate->dtms_probe->dtpr_arg, ndx, aframes); + else + val = dtrace_getarg(ndx, aframes); + + /* + * This is regrettably required to keep the compiler + * from tail-optimizing the call to dtrace_getarg(). + * The condition always evaluates to true, but the + * compiler has no way of figuring that out a priori. + * (None of this would be necessary if the compiler + * could be relied upon to _always_ tail-optimize + * the call to dtrace_getarg() -- but it can't.) + */ + if (mstate->dtms_probe != NULL) + return (val); + + ASSERT(0); + } + + return (mstate->dtms_arg[ndx]); + +#if defined(sun) + case DIF_VAR_UREGS: { + klwp_t *lwp; + + if (!dtrace_priv_proc(state)) + return (0); + + if ((lwp = curthread->t_lwp) == NULL) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = NULL; + return (0); + } + + return (dtrace_getreg(lwp->lwp_regs, ndx)); + return (0); + } +#endif + + case DIF_VAR_CURTHREAD: + if (!dtrace_priv_kernel(state)) + return (0); + return ((uint64_t)(uintptr_t)curthread); + + case DIF_VAR_TIMESTAMP: + if (!(mstate->dtms_present & DTRACE_MSTATE_TIMESTAMP)) { + mstate->dtms_timestamp = dtrace_gethrtime(); + mstate->dtms_present |= DTRACE_MSTATE_TIMESTAMP; + } + return (mstate->dtms_timestamp); + + case DIF_VAR_VTIMESTAMP: + ASSERT(dtrace_vtime_references != 0); + return (curthread->t_dtrace_vtime); + + case DIF_VAR_WALLTIMESTAMP: + if (!(mstate->dtms_present & DTRACE_MSTATE_WALLTIMESTAMP)) { + mstate->dtms_walltimestamp = dtrace_gethrestime(); + mstate->dtms_present |= DTRACE_MSTATE_WALLTIMESTAMP; + } + return (mstate->dtms_walltimestamp); + +#if defined(sun) + case DIF_VAR_IPL: + if (!dtrace_priv_kernel(state)) + return (0); + if (!(mstate->dtms_present & DTRACE_MSTATE_IPL)) { + mstate->dtms_ipl = dtrace_getipl(); + mstate->dtms_present |= DTRACE_MSTATE_IPL; + } + return (mstate->dtms_ipl); +#endif + + case DIF_VAR_EPID: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_EPID); + return (mstate->dtms_epid); + + case DIF_VAR_ID: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return (mstate->dtms_probe->dtpr_id); + + case DIF_VAR_STACKDEPTH: + if (!dtrace_priv_kernel(state)) + return (0); + if (!(mstate->dtms_present & DTRACE_MSTATE_STACKDEPTH)) { + int aframes = mstate->dtms_probe->dtpr_aframes + 2; + + mstate->dtms_stackdepth = dtrace_getstackdepth(aframes); + mstate->dtms_present |= DTRACE_MSTATE_STACKDEPTH; + } + return (mstate->dtms_stackdepth); + +#if defined(sun) + case DIF_VAR_USTACKDEPTH: + if (!dtrace_priv_proc(state)) + return (0); + if (!(mstate->dtms_present & DTRACE_MSTATE_USTACKDEPTH)) { + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && + CPU_ON_INTR(CPU)) { + mstate->dtms_ustackdepth = 0; + } else { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + mstate->dtms_ustackdepth = + dtrace_getustackdepth(); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + mstate->dtms_present |= DTRACE_MSTATE_USTACKDEPTH; + } + return (mstate->dtms_ustackdepth); +#endif + + case DIF_VAR_CALLER: + if (!dtrace_priv_kernel(state)) + return (0); + if (!(mstate->dtms_present & DTRACE_MSTATE_CALLER)) { + int aframes = mstate->dtms_probe->dtpr_aframes + 2; + + if (!DTRACE_ANCHORED(mstate->dtms_probe)) { + /* + * If this is an unanchored probe, we are + * required to go through the slow path: + * dtrace_caller() only guarantees correct + * results for anchored probes. + */ + pc_t caller[2] = {0, 0}; + + dtrace_getpcstack(caller, 2, aframes, + (uint32_t *)(uintptr_t)mstate->dtms_arg[0]); + mstate->dtms_caller = caller[1]; + } else if ((mstate->dtms_caller = + dtrace_caller(aframes)) == -1) { + /* + * We have failed to do this the quick way; + * we must resort to the slower approach of + * calling dtrace_getpcstack(). + */ + pc_t caller = 0; + + dtrace_getpcstack(&caller, 1, aframes, NULL); + mstate->dtms_caller = caller; + } + + mstate->dtms_present |= DTRACE_MSTATE_CALLER; + } + return (mstate->dtms_caller); + +#if defined(sun) + case DIF_VAR_UCALLER: + if (!dtrace_priv_proc(state)) + return (0); + + if (!(mstate->dtms_present & DTRACE_MSTATE_UCALLER)) { + uint64_t ustack[3]; + + /* + * dtrace_getupcstack() fills in the first uint64_t + * with the current PID. The second uint64_t will + * be the program counter at user-level. The third + * uint64_t will contain the caller, which is what + * we're after. + */ + ustack[2] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_getupcstack(ustack, 3); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + mstate->dtms_ucaller = ustack[2]; + mstate->dtms_present |= DTRACE_MSTATE_UCALLER; + } + + return (mstate->dtms_ucaller); +#endif + + case DIF_VAR_PROBEPROV: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return (dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_provider->dtpv_name, + state, mstate)); + + case DIF_VAR_PROBEMOD: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return (dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_mod, + state, mstate)); + + case DIF_VAR_PROBEFUNC: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return (dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_func, + state, mstate)); + + case DIF_VAR_PROBENAME: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return (dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_name, + state, mstate)); + + case DIF_VAR_PID: + if (!dtrace_priv_proc(state)) + return (0); + +#if defined(sun) + /* + * Note that we are assuming that an unanchored probe is + * always due to a high-level interrupt. (And we're assuming + * that there is only a single high level interrupt.) + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return (pid0.pid_id); + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * Further, it is always safe to dereference the p_pidp member + * of one's own proc structure. (These are truisms becuase + * threads and processes don't clean up their own state -- + * they leave that task to whomever reaps them.) + */ + return ((uint64_t)curthread->t_procp->p_pidp->pid_id); +#else + return ((uint64_t)curproc->p_pid); +#endif + + case DIF_VAR_PPID: + if (!dtrace_priv_proc(state)) + return (0); + +#if defined(sun) + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return (pid0.pid_id); + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * (This is true because threads don't clean up their own + * state -- they leave that task to whomever reaps them.) + */ + return ((uint64_t)curthread->t_procp->p_ppid); +#else + return ((uint64_t)curproc->p_pptr->p_pid); +#endif + + case DIF_VAR_TID: +#if defined(sun) + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return (0); +#endif + + return ((uint64_t)curthread->t_tid); + + case DIF_VAR_EXECARGS: { + struct pargs *p_args = curthread->td_proc->p_args; + + if (p_args == NULL) + return(0); + + return (dtrace_dif_varstrz( + (uintptr_t) p_args->ar_args, p_args->ar_length, state, mstate)); + } + + case DIF_VAR_EXECNAME: +#if defined(sun) + if (!dtrace_priv_proc(state)) + return (0); + + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return ((uint64_t)(uintptr_t)p0.p_user.u_comm); + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * (This is true because threads don't clean up their own + * state -- they leave that task to whomever reaps them.) + */ + return (dtrace_dif_varstr( + (uintptr_t)curthread->t_procp->p_user.u_comm, + state, mstate)); +#else + return (dtrace_dif_varstr( + (uintptr_t) curthread->td_proc->p_comm, state, mstate)); +#endif + + case DIF_VAR_ZONENAME: +#if defined(sun) + if (!dtrace_priv_proc(state)) + return (0); + + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return ((uint64_t)(uintptr_t)p0.p_zone->zone_name); + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * (This is true because threads don't clean up their own + * state -- they leave that task to whomever reaps them.) + */ + return (dtrace_dif_varstr( + (uintptr_t)curthread->t_procp->p_zone->zone_name, + state, mstate)); +#else + return (0); +#endif + + case DIF_VAR_UID: + if (!dtrace_priv_proc(state)) + return (0); + +#if defined(sun) + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return ((uint64_t)p0.p_cred->cr_uid); +#endif + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * (This is true because threads don't clean up their own + * state -- they leave that task to whomever reaps them.) + * + * Additionally, it is safe to dereference one's own process + * credential, since this is never NULL after process birth. + */ + return ((uint64_t)curthread->t_procp->p_cred->cr_uid); + + case DIF_VAR_GID: + if (!dtrace_priv_proc(state)) + return (0); + +#if defined(sun) + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return ((uint64_t)p0.p_cred->cr_gid); +#endif + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * (This is true because threads don't clean up their own + * state -- they leave that task to whomever reaps them.) + * + * Additionally, it is safe to dereference one's own process + * credential, since this is never NULL after process birth. + */ + return ((uint64_t)curthread->t_procp->p_cred->cr_gid); + + case DIF_VAR_ERRNO: { +#if defined(sun) + klwp_t *lwp; + if (!dtrace_priv_proc(state)) + return (0); + + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return (0); + + /* + * It is always safe to dereference one's own t_lwp pointer in + * the event that this pointer is non-NULL. (This is true + * because threads and lwps don't clean up their own state -- + * they leave that task to whomever reaps them.) + */ + if ((lwp = curthread->t_lwp) == NULL) + return (0); + + return ((uint64_t)lwp->lwp_errno); +#else + return (curthread->td_errno); +#endif + } + default: + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } +} + +/* + * Emulate the execution of DTrace ID subroutines invoked by the call opcode. + * Notice that we don't bother validating the proper number of arguments or + * their types in the tuple stack. This isn't needed because all argument + * interpretation is safe because of our load safety -- the worst that can + * happen is that a bogus program can obtain bogus results. + */ +static void +dtrace_dif_subr(uint_t subr, uint_t rd, uint64_t *regs, + dtrace_key_t *tupregs, int nargs, + dtrace_mstate_t *mstate, dtrace_state_t *state) +{ + volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; + volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval; + dtrace_vstate_t *vstate = &state->dts_vstate; + +#if defined(sun) + union { + mutex_impl_t mi; + uint64_t mx; + } m; + + union { + krwlock_t ri; + uintptr_t rw; + } r; +#else + union { + struct mtx *mi; + uintptr_t mx; + } m; + union { + struct sx *si; + uintptr_t sx; + } s; +#endif + + switch (subr) { + case DIF_SUBR_RAND: + regs[rd] = (dtrace_gethrtime() * 2416 + 374441) % 1771875; + break; + +#if defined(sun) + case DIF_SUBR_MUTEX_OWNED: + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), + mstate, vstate)) { + regs[rd] = 0; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); + if (MUTEX_TYPE_ADAPTIVE(&m.mi)) + regs[rd] = MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER; + else + regs[rd] = LOCK_HELD(&m.mi.m_spin.m_spinlock); + break; + + case DIF_SUBR_MUTEX_OWNER: + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), + mstate, vstate)) { + regs[rd] = 0; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); + if (MUTEX_TYPE_ADAPTIVE(&m.mi) && + MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER) + regs[rd] = (uintptr_t)MUTEX_OWNER(&m.mi); + else + regs[rd] = 0; + break; + + case DIF_SUBR_MUTEX_TYPE_ADAPTIVE: + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), + mstate, vstate)) { + regs[rd] = 0; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); + regs[rd] = MUTEX_TYPE_ADAPTIVE(&m.mi); + break; + + case DIF_SUBR_MUTEX_TYPE_SPIN: + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), + mstate, vstate)) { + regs[rd] = 0; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); + regs[rd] = MUTEX_TYPE_SPIN(&m.mi); + break; + + case DIF_SUBR_RW_READ_HELD: { + uintptr_t tmp; + + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t), + mstate, vstate)) { + regs[rd] = 0; + break; + } + + r.rw = dtrace_loadptr(tupregs[0].dttk_value); + regs[rd] = _RW_READ_HELD(&r.ri, tmp); + break; + } + + case DIF_SUBR_RW_WRITE_HELD: + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t), + mstate, vstate)) { + regs[rd] = 0; + break; + } + + r.rw = dtrace_loadptr(tupregs[0].dttk_value); + regs[rd] = _RW_WRITE_HELD(&r.ri); + break; + + case DIF_SUBR_RW_ISWRITER: + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t), + mstate, vstate)) { + regs[rd] = 0; + break; + } + + r.rw = dtrace_loadptr(tupregs[0].dttk_value); + regs[rd] = _RW_ISWRITER(&r.ri); + break; + +#else + /* + * XXX - The following code works because mutex, rwlocks, & sxlocks + * all have similar data structures in FreeBSD. This may not be + * good if someone changes one of the lock data structures. + * Ideally, it would be nice if all these shared a common lock + * object. + */ + case DIF_SUBR_MUTEX_OWNED: + /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ + m.mx = tupregs[0].dttk_value; + +#ifdef DOODAD + if (LO_CLASSINDEX(&(m.mi->lock_object)) < 2) { + regs[rd] = !(m.mi->mtx_lock & MTX_UNOWNED); + } else { + regs[rd] = !(m.mi->mtx_lock & SX_UNLOCKED); + } +#endif + break; + + case DIF_SUBR_MUTEX_OWNER: + /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ + m.mx = tupregs[0].dttk_value; + + if (LO_CLASSINDEX(&(m.mi->lock_object)) < 2) { + regs[rd] = m.mi->mtx_lock & ~MTX_FLAGMASK; + } else { + if (!(m.mi->mtx_lock & SX_LOCK_SHARED)) + regs[rd] = SX_OWNER(m.mi->mtx_lock); + else + regs[rd] = 0; + } + break; + + case DIF_SUBR_MUTEX_TYPE_ADAPTIVE: + /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ + m.mx = tupregs[0].dttk_value; + + regs[rd] = (LO_CLASSINDEX(&(m.mi->lock_object)) != 0); + break; + + case DIF_SUBR_MUTEX_TYPE_SPIN: + /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ + m.mx = tupregs[0].dttk_value; + + regs[rd] = (LO_CLASSINDEX(&(m.mi->lock_object)) == 0); + break; + + case DIF_SUBR_RW_READ_HELD: + case DIF_SUBR_SX_SHARED_HELD: + /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ + s.sx = tupregs[0].dttk_value; + regs[rd] = ((s.si->sx_lock & SX_LOCK_SHARED) && + (SX_OWNER(s.si->sx_lock) >> SX_SHARERS_SHIFT) != 0); + break; + + case DIF_SUBR_RW_WRITE_HELD: + case DIF_SUBR_SX_EXCLUSIVE_HELD: + /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ + s.sx = tupregs[0].dttk_value; + regs[rd] = (SX_OWNER(s.si->sx_lock) == (uintptr_t) curthread); + break; + + case DIF_SUBR_RW_ISWRITER: + case DIF_SUBR_SX_ISEXCLUSIVE: + /* XXX - need to use dtrace_canload() and dtrace_loadptr() */ + s.sx = tupregs[0].dttk_value; + regs[rd] = ((s.si->sx_lock & SX_LOCK_EXCLUSIVE_WAITERS) || + !(s.si->sx_lock & SX_LOCK_SHARED)); + break; +#endif /* ! defined(sun) */ + + case DIF_SUBR_BCOPY: { + /* + * We need to be sure that the destination is in the scratch + * region -- no other region is allowed. + */ + uintptr_t src = tupregs[0].dttk_value; + uintptr_t dest = tupregs[1].dttk_value; + size_t size = tupregs[2].dttk_value; + + if (!dtrace_inscratch(dest, size, mstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + if (!dtrace_canload(src, size, mstate, vstate)) { + regs[rd] = 0; + break; + } + + dtrace_bcopy((void *)src, (void *)dest, size); + break; + } + + case DIF_SUBR_ALLOCA: + case DIF_SUBR_COPYIN: { + uintptr_t dest = P2ROUNDUP(mstate->dtms_scratch_ptr, 8); + uint64_t size = + tupregs[subr == DIF_SUBR_ALLOCA ? 0 : 1].dttk_value; + size_t scratch_size = (dest - mstate->dtms_scratch_ptr) + size; + + /* + * This action doesn't require any credential checks since + * probes will not activate in user contexts to which the + * enabling user does not have permissions. + */ + + /* + * Rounding up the user allocation size could have overflowed + * a large, bogus allocation (like -1ULL) to 0. + */ + if (scratch_size < size || + !DTRACE_INSCRATCH(mstate, scratch_size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + if (subr == DIF_SUBR_COPYIN) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyin(tupregs[0].dttk_value, dest, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + + mstate->dtms_scratch_ptr += scratch_size; + regs[rd] = dest; + break; + } + + case DIF_SUBR_COPYINTO: { + uint64_t size = tupregs[1].dttk_value; + uintptr_t dest = tupregs[2].dttk_value; + + /* + * This action doesn't require any credential checks since + * probes will not activate in user contexts to which the + * enabling user does not have permissions. + */ + if (!dtrace_inscratch(dest, size, mstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyin(tupregs[0].dttk_value, dest, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + break; + } + + case DIF_SUBR_COPYINSTR: { + uintptr_t dest = mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + + if (nargs > 1 && tupregs[1].dttk_value < size) + size = tupregs[1].dttk_value + 1; + + /* + * This action doesn't require any credential checks since + * probes will not activate in user contexts to which the + * enabling user does not have permissions. + */ + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyinstr(tupregs[0].dttk_value, dest, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + ((char *)dest)[size - 1] = '\0'; + mstate->dtms_scratch_ptr += size; + regs[rd] = dest; + break; + } + +#if defined(sun) + case DIF_SUBR_MSGSIZE: + case DIF_SUBR_MSGDSIZE: { + uintptr_t baddr = tupregs[0].dttk_value, daddr; + uintptr_t wptr, rptr; + size_t count = 0; + int cont = 0; + + while (baddr != 0 && !(*flags & CPU_DTRACE_FAULT)) { + + if (!dtrace_canload(baddr, sizeof (mblk_t), mstate, + vstate)) { + regs[rd] = 0; + break; + } + + wptr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_wptr)); + + rptr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_rptr)); + + if (wptr < rptr) { + *flags |= CPU_DTRACE_BADADDR; + *illval = tupregs[0].dttk_value; + break; + } + + daddr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_datap)); + + baddr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_cont)); + + /* + * We want to prevent against denial-of-service here, + * so we're only going to search the list for + * dtrace_msgdsize_max mblks. + */ + if (cont++ > dtrace_msgdsize_max) { + *flags |= CPU_DTRACE_ILLOP; + break; + } + + if (subr == DIF_SUBR_MSGDSIZE) { + if (dtrace_load8(daddr + + offsetof(dblk_t, db_type)) != M_DATA) + continue; + } + + count += wptr - rptr; + } + + if (!(*flags & CPU_DTRACE_FAULT)) + regs[rd] = count; + + break; + } +#endif + + case DIF_SUBR_PROGENYOF: { + pid_t pid = tupregs[0].dttk_value; + proc_t *p; + int rval = 0; + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + + for (p = curthread->t_procp; p != NULL; p = p->p_parent) { +#if defined(sun) + if (p->p_pidp->pid_id == pid) { +#else + if (p->p_pid == pid) { +#endif + rval = 1; + break; + } + } + + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + regs[rd] = rval; + break; + } + + case DIF_SUBR_SPECULATION: + regs[rd] = dtrace_speculation(state); + break; + + case DIF_SUBR_COPYOUT: { + uintptr_t kaddr = tupregs[0].dttk_value; + uintptr_t uaddr = tupregs[1].dttk_value; + uint64_t size = tupregs[2].dttk_value; + + if (!dtrace_destructive_disallow && + dtrace_priv_proc_control(state) && + !dtrace_istoxic(kaddr, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyout(kaddr, uaddr, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + break; + } + + case DIF_SUBR_COPYOUTSTR: { + uintptr_t kaddr = tupregs[0].dttk_value; + uintptr_t uaddr = tupregs[1].dttk_value; + uint64_t size = tupregs[2].dttk_value; + + if (!dtrace_destructive_disallow && + dtrace_priv_proc_control(state) && + !dtrace_istoxic(kaddr, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyoutstr(kaddr, uaddr, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + break; + } + + case DIF_SUBR_STRLEN: { + size_t sz; + uintptr_t addr = (uintptr_t)tupregs[0].dttk_value; + sz = dtrace_strlen((char *)addr, + state->dts_options[DTRACEOPT_STRSIZE]); + + if (!dtrace_canload(addr, sz + 1, mstate, vstate)) { + regs[rd] = 0; + break; + } + + regs[rd] = sz; + + break; + } + + case DIF_SUBR_STRCHR: + case DIF_SUBR_STRRCHR: { + /* + * We're going to iterate over the string looking for the + * specified character. We will iterate until we have reached + * the string length or we have found the character. If this + * is DIF_SUBR_STRRCHR, we will look for the last occurrence + * of the specified character instead of the first. + */ + uintptr_t saddr = tupregs[0].dttk_value; + uintptr_t addr = tupregs[0].dttk_value; + uintptr_t limit = addr + state->dts_options[DTRACEOPT_STRSIZE]; + char c, target = (char)tupregs[1].dttk_value; + + for (regs[rd] = 0; addr < limit; addr++) { + if ((c = dtrace_load8(addr)) == target) { + regs[rd] = addr; + + if (subr == DIF_SUBR_STRCHR) + break; + } + + if (c == '\0') + break; + } + + if (!dtrace_canload(saddr, addr - saddr, mstate, vstate)) { + regs[rd] = 0; + break; + } + + break; + } + + case DIF_SUBR_STRSTR: + case DIF_SUBR_INDEX: + case DIF_SUBR_RINDEX: { + /* + * We're going to iterate over the string looking for the + * specified string. We will iterate until we have reached + * the string length or we have found the string. (Yes, this + * is done in the most naive way possible -- but considering + * that the string we're searching for is likely to be + * relatively short, the complexity of Rabin-Karp or similar + * hardly seems merited.) + */ + char *addr = (char *)(uintptr_t)tupregs[0].dttk_value; + char *substr = (char *)(uintptr_t)tupregs[1].dttk_value; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + size_t len = dtrace_strlen(addr, size); + size_t sublen = dtrace_strlen(substr, size); + char *limit = addr + len, *orig = addr; + int notfound = subr == DIF_SUBR_STRSTR ? 0 : -1; + int inc = 1; + + regs[rd] = notfound; + + if (!dtrace_canload((uintptr_t)addr, len + 1, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!dtrace_canload((uintptr_t)substr, sublen + 1, mstate, + vstate)) { + regs[rd] = 0; + break; + } + + /* + * strstr() and index()/rindex() have similar semantics if + * both strings are the empty string: strstr() returns a + * pointer to the (empty) string, and index() and rindex() + * both return index 0 (regardless of any position argument). + */ + if (sublen == 0 && len == 0) { + if (subr == DIF_SUBR_STRSTR) + regs[rd] = (uintptr_t)addr; + else + regs[rd] = 0; + break; + } + + if (subr != DIF_SUBR_STRSTR) { + if (subr == DIF_SUBR_RINDEX) { + limit = orig - 1; + addr += len; + inc = -1; + } + + /* + * Both index() and rindex() take an optional position + * argument that denotes the starting position. + */ + if (nargs == 3) { + int64_t pos = (int64_t)tupregs[2].dttk_value; + + /* + * If the position argument to index() is + * negative, Perl implicitly clamps it at + * zero. This semantic is a little surprising + * given the special meaning of negative + * positions to similar Perl functions like + * substr(), but it appears to reflect a + * notion that index() can start from a + * negative index and increment its way up to + * the string. Given this notion, Perl's + * rindex() is at least self-consistent in + * that it implicitly clamps positions greater + * than the string length to be the string + * length. Where Perl completely loses + * coherence, however, is when the specified + * substring is the empty string (""). In + * this case, even if the position is + * negative, rindex() returns 0 -- and even if + * the position is greater than the length, + * index() returns the string length. These + * semantics violate the notion that index() + * should never return a value less than the + * specified position and that rindex() should + * never return a value greater than the + * specified position. (One assumes that + * these semantics are artifacts of Perl's + * implementation and not the results of + * deliberate design -- it beggars belief that + * even Larry Wall could desire such oddness.) + * While in the abstract one would wish for + * consistent position semantics across + * substr(), index() and rindex() -- or at the + * very least self-consistent position + * semantics for index() and rindex() -- we + * instead opt to keep with the extant Perl + * semantics, in all their broken glory. (Do + * we have more desire to maintain Perl's + * semantics than Perl does? Probably.) + */ + if (subr == DIF_SUBR_RINDEX) { + if (pos < 0) { + if (sublen == 0) + regs[rd] = 0; + break; + } + + if (pos > len) + pos = len; + } else { + if (pos < 0) + pos = 0; + + if (pos >= len) { + if (sublen == 0) + regs[rd] = len; + break; + } + } + + addr = orig + pos; + } + } + + for (regs[rd] = notfound; addr != limit; addr += inc) { + if (dtrace_strncmp(addr, substr, sublen) == 0) { + if (subr != DIF_SUBR_STRSTR) { + /* + * As D index() and rindex() are + * modeled on Perl (and not on awk), + * we return a zero-based (and not a + * one-based) index. (For you Perl + * weenies: no, we're not going to add + * $[ -- and shouldn't you be at a con + * or something?) + */ + regs[rd] = (uintptr_t)(addr - orig); + break; + } + + ASSERT(subr == DIF_SUBR_STRSTR); + regs[rd] = (uintptr_t)addr; + break; + } + } + + break; + } + + case DIF_SUBR_STRTOK: { + uintptr_t addr = tupregs[0].dttk_value; + uintptr_t tokaddr = tupregs[1].dttk_value; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t limit, toklimit = tokaddr + size; + uint8_t c = 0, tokmap[32]; /* 256 / 8 */ + char *dest = (char *)mstate->dtms_scratch_ptr; + int i; + + /* + * Check both the token buffer and (later) the input buffer, + * since both could be non-scratch addresses. + */ + if (!dtrace_strcanload(tokaddr, size, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + if (addr == 0) { + /* + * If the address specified is NULL, we use our saved + * strtok pointer from the mstate. Note that this + * means that the saved strtok pointer is _only_ + * valid within multiple enablings of the same probe -- + * it behaves like an implicit clause-local variable. + */ + addr = mstate->dtms_strtok; + } else { + /* + * If the user-specified address is non-NULL we must + * access check it. This is the only time we have + * a chance to do so, since this address may reside + * in the string table of this clause-- future calls + * (when we fetch addr from mstate->dtms_strtok) + * would fail this access check. + */ + if (!dtrace_strcanload(addr, size, mstate, vstate)) { + regs[rd] = 0; + break; + } + } + + /* + * First, zero the token map, and then process the token + * string -- setting a bit in the map for every character + * found in the token string. + */ + for (i = 0; i < sizeof (tokmap); i++) + tokmap[i] = 0; + + for (; tokaddr < toklimit; tokaddr++) { + if ((c = dtrace_load8(tokaddr)) == '\0') + break; + + ASSERT((c >> 3) < sizeof (tokmap)); + tokmap[c >> 3] |= (1 << (c & 0x7)); + } + + for (limit = addr + size; addr < limit; addr++) { + /* + * We're looking for a character that is _not_ contained + * in the token string. + */ + if ((c = dtrace_load8(addr)) == '\0') + break; + + if (!(tokmap[c >> 3] & (1 << (c & 0x7)))) + break; + } + + if (c == '\0') { + /* + * We reached the end of the string without finding + * any character that was not in the token string. + * We return NULL in this case, and we set the saved + * address to NULL as well. + */ + regs[rd] = 0; + mstate->dtms_strtok = 0; + break; + } + + /* + * From here on, we're copying into the destination string. + */ + for (i = 0; addr < limit && i < size - 1; addr++) { + if ((c = dtrace_load8(addr)) == '\0') + break; + + if (tokmap[c >> 3] & (1 << (c & 0x7))) + break; + + ASSERT(i < size); + dest[i++] = c; + } + + ASSERT(i < size); + dest[i] = '\0'; + regs[rd] = (uintptr_t)dest; + mstate->dtms_scratch_ptr += size; + mstate->dtms_strtok = addr; + break; + } + + case DIF_SUBR_SUBSTR: { + uintptr_t s = tupregs[0].dttk_value; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + char *d = (char *)mstate->dtms_scratch_ptr; + int64_t index = (int64_t)tupregs[1].dttk_value; + int64_t remaining = (int64_t)tupregs[2].dttk_value; + size_t len = dtrace_strlen((char *)s, size); + int64_t i = 0; + + if (!dtrace_canload(s, len + 1, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + if (nargs <= 2) + remaining = (int64_t)size; + + if (index < 0) { + index += len; + + if (index < 0 && index + remaining > 0) { + remaining += index; + index = 0; + } + } + + if (index >= len || index < 0) { + remaining = 0; + } else if (remaining < 0) { + remaining += len - index; + } else if (index + remaining > size) { + remaining = size - index; + } + + for (i = 0; i < remaining; i++) { + if ((d[i] = dtrace_load8(s + index + i)) == '\0') + break; + } + + d[i] = '\0'; + + mstate->dtms_scratch_ptr += size; + regs[rd] = (uintptr_t)d; + break; + } + +#if defined(sun) + case DIF_SUBR_GETMAJOR: +#ifdef _LP64 + regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR64) & MAXMAJ64; +#else + regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR) & MAXMAJ; +#endif + break; + + case DIF_SUBR_GETMINOR: +#ifdef _LP64 + regs[rd] = tupregs[0].dttk_value & MAXMIN64; +#else + regs[rd] = tupregs[0].dttk_value & MAXMIN; +#endif + break; + + case DIF_SUBR_DDI_PATHNAME: { + /* + * This one is a galactic mess. We are going to roughly + * emulate ddi_pathname(), but it's made more complicated + * by the fact that we (a) want to include the minor name and + * (b) must proceed iteratively instead of recursively. + */ + uintptr_t dest = mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + char *start = (char *)dest, *end = start + size - 1; + uintptr_t daddr = tupregs[0].dttk_value; + int64_t minor = (int64_t)tupregs[1].dttk_value; + char *s; + int i, len, depth = 0; + + /* + * Due to all the pointer jumping we do and context we must + * rely upon, we just mandate that the user must have kernel + * read privileges to use this routine. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) == 0) { + *flags |= CPU_DTRACE_KPRIV; + *illval = daddr; + regs[rd] = 0; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + *end = '\0'; + + /* + * We want to have a name for the minor. In order to do this, + * we need to walk the minor list from the devinfo. We want + * to be sure that we don't infinitely walk a circular list, + * so we check for circularity by sending a scout pointer + * ahead two elements for every element that we iterate over; + * if the list is circular, these will ultimately point to the + * same element. You may recognize this little trick as the + * answer to a stupid interview question -- one that always + * seems to be asked by those who had to have it laboriously + * explained to them, and who can't even concisely describe + * the conditions under which one would be forced to resort to + * this technique. Needless to say, those conditions are + * found here -- and probably only here. Is this the only use + * of this infamous trick in shipping, production code? If it + * isn't, it probably should be... + */ + if (minor != -1) { + uintptr_t maddr = dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_minor)); + + uintptr_t next = offsetof(struct ddi_minor_data, next); + uintptr_t name = offsetof(struct ddi_minor_data, + d_minor) + offsetof(struct ddi_minor, name); + uintptr_t dev = offsetof(struct ddi_minor_data, + d_minor) + offsetof(struct ddi_minor, dev); + uintptr_t scout; + + if (maddr != NULL) + scout = dtrace_loadptr(maddr + next); + + while (maddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { + uint64_t m; +#ifdef _LP64 + m = dtrace_load64(maddr + dev) & MAXMIN64; +#else + m = dtrace_load32(maddr + dev) & MAXMIN; +#endif + if (m != minor) { + maddr = dtrace_loadptr(maddr + next); + + if (scout == NULL) + continue; + + scout = dtrace_loadptr(scout + next); + + if (scout == NULL) + continue; + + scout = dtrace_loadptr(scout + next); + + if (scout == NULL) + continue; + + if (scout == maddr) { + *flags |= CPU_DTRACE_ILLOP; + break; + } + + continue; + } + + /* + * We have the minor data. Now we need to + * copy the minor's name into the end of the + * pathname. + */ + s = (char *)dtrace_loadptr(maddr + name); + len = dtrace_strlen(s, size); + + if (*flags & CPU_DTRACE_FAULT) + break; + + if (len != 0) { + if ((end -= (len + 1)) < start) + break; + + *end = ':'; + } + + for (i = 1; i <= len; i++) + end[i] = dtrace_load8((uintptr_t)s++); + break; + } + } + + while (daddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { + ddi_node_state_t devi_state; + + devi_state = dtrace_load32(daddr + + offsetof(struct dev_info, devi_node_state)); + + if (*flags & CPU_DTRACE_FAULT) + break; + + if (devi_state >= DS_INITIALIZED) { + s = (char *)dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_addr)); + len = dtrace_strlen(s, size); + + if (*flags & CPU_DTRACE_FAULT) + break; + + if (len != 0) { + if ((end -= (len + 1)) < start) + break; + + *end = '@'; + } + + for (i = 1; i <= len; i++) + end[i] = dtrace_load8((uintptr_t)s++); + } + + /* + * Now for the node name... + */ + s = (char *)dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_node_name)); + + daddr = dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_parent)); + + /* + * If our parent is NULL (that is, if we're the root + * node), we're going to use the special path + * "devices". + */ + if (daddr == 0) + s = "devices"; + + len = dtrace_strlen(s, size); + if (*flags & CPU_DTRACE_FAULT) + break; + + if ((end -= (len + 1)) < start) + break; + + for (i = 1; i <= len; i++) + end[i] = dtrace_load8((uintptr_t)s++); + *end = '/'; + + if (depth++ > dtrace_devdepth_max) { + *flags |= CPU_DTRACE_ILLOP; + break; + } + } + + if (end < start) + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + + if (daddr == 0) { + regs[rd] = (uintptr_t)end; + mstate->dtms_scratch_ptr += size; + } + + break; + } +#endif + + case DIF_SUBR_STRJOIN: { + char *d = (char *)mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t s1 = tupregs[0].dttk_value; + uintptr_t s2 = tupregs[1].dttk_value; + int i = 0; + + if (!dtrace_strcanload(s1, size, mstate, vstate) || + !dtrace_strcanload(s2, size, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + for (;;) { + if (i >= size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + if ((d[i++] = dtrace_load8(s1++)) == '\0') { + i--; + break; + } + } + + for (;;) { + if (i >= size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + if ((d[i++] = dtrace_load8(s2++)) == '\0') + break; + } + + if (i < size) { + mstate->dtms_scratch_ptr += i; + regs[rd] = (uintptr_t)d; + } + + break; + } + + case DIF_SUBR_LLTOSTR: { + int64_t i = (int64_t)tupregs[0].dttk_value; + int64_t val = i < 0 ? i * -1 : i; + uint64_t size = 22; /* enough room for 2^64 in decimal */ + char *end = (char *)mstate->dtms_scratch_ptr + size - 1; + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + for (*end-- = '\0'; val; val /= 10) + *end-- = '0' + (val % 10); + + if (i == 0) + *end-- = '0'; + + if (i < 0) + *end-- = '-'; + + regs[rd] = (uintptr_t)end + 1; + mstate->dtms_scratch_ptr += size; + break; + } + + case DIF_SUBR_HTONS: + case DIF_SUBR_NTOHS: +#if BYTE_ORDER == BIG_ENDIAN + regs[rd] = (uint16_t)tupregs[0].dttk_value; +#else + regs[rd] = DT_BSWAP_16((uint16_t)tupregs[0].dttk_value); +#endif + break; + + + case DIF_SUBR_HTONL: + case DIF_SUBR_NTOHL: +#if BYTE_ORDER == BIG_ENDIAN + regs[rd] = (uint32_t)tupregs[0].dttk_value; +#else + regs[rd] = DT_BSWAP_32((uint32_t)tupregs[0].dttk_value); +#endif + break; + + + case DIF_SUBR_HTONLL: + case DIF_SUBR_NTOHLL: +#if BYTE_ORDER == BIG_ENDIAN + regs[rd] = (uint64_t)tupregs[0].dttk_value; +#else + regs[rd] = DT_BSWAP_64((uint64_t)tupregs[0].dttk_value); +#endif + break; + + + case DIF_SUBR_DIRNAME: + case DIF_SUBR_BASENAME: { + char *dest = (char *)mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t src = tupregs[0].dttk_value; + int i, j, len = dtrace_strlen((char *)src, size); + int lastbase = -1, firstbase = -1, lastdir = -1; + int start, end; + + if (!dtrace_canload(src, len + 1, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + /* + * The basename and dirname for a zero-length string is + * defined to be "." + */ + if (len == 0) { + len = 1; + src = (uintptr_t)"."; + } + + /* + * Start from the back of the string, moving back toward the + * front until we see a character that isn't a slash. That + * character is the last character in the basename. + */ + for (i = len - 1; i >= 0; i--) { + if (dtrace_load8(src + i) != '/') + break; + } + + if (i >= 0) + lastbase = i; + + /* + * Starting from the last character in the basename, move + * towards the front until we find a slash. The character + * that we processed immediately before that is the first + * character in the basename. + */ + for (; i >= 0; i--) { + if (dtrace_load8(src + i) == '/') + break; + } + + if (i >= 0) + firstbase = i + 1; + + /* + * Now keep going until we find a non-slash character. That + * character is the last character in the dirname. + */ + for (; i >= 0; i--) { + if (dtrace_load8(src + i) != '/') + break; + } + + if (i >= 0) + lastdir = i; + + ASSERT(!(lastbase == -1 && firstbase != -1)); + ASSERT(!(firstbase == -1 && lastdir != -1)); + + if (lastbase == -1) { + /* + * We didn't find a non-slash character. We know that + * the length is non-zero, so the whole string must be + * slashes. In either the dirname or the basename + * case, we return '/'. + */ + ASSERT(firstbase == -1); + firstbase = lastbase = lastdir = 0; + } + + if (firstbase == -1) { + /* + * The entire string consists only of a basename + * component. If we're looking for dirname, we need + * to change our string to be just "."; if we're + * looking for a basename, we'll just set the first + * character of the basename to be 0. + */ + if (subr == DIF_SUBR_DIRNAME) { + ASSERT(lastdir == -1); + src = (uintptr_t)"."; + lastdir = 0; + } else { + firstbase = 0; + } + } + + if (subr == DIF_SUBR_DIRNAME) { + if (lastdir == -1) { + /* + * We know that we have a slash in the name -- + * or lastdir would be set to 0, above. And + * because lastdir is -1, we know that this + * slash must be the first character. (That + * is, the full string must be of the form + * "/basename".) In this case, the last + * character of the directory name is 0. + */ + lastdir = 0; + } + + start = 0; + end = lastdir; + } else { + ASSERT(subr == DIF_SUBR_BASENAME); + ASSERT(firstbase != -1 && lastbase != -1); + start = firstbase; + end = lastbase; + } + + for (i = start, j = 0; i <= end && j < size - 1; i++, j++) + dest[j] = dtrace_load8(src + i); + + dest[j] = '\0'; + regs[rd] = (uintptr_t)dest; + mstate->dtms_scratch_ptr += size; + break; + } + + case DIF_SUBR_CLEANPATH: { + char *dest = (char *)mstate->dtms_scratch_ptr, c; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t src = tupregs[0].dttk_value; + int i = 0, j = 0; + + if (!dtrace_strcanload(src, size, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + /* + * Move forward, loading each character. + */ + do { + c = dtrace_load8(src + i++); +next: + if (j + 5 >= size) /* 5 = strlen("/..c\0") */ + break; + + if (c != '/') { + dest[j++] = c; + continue; + } + + c = dtrace_load8(src + i++); + + if (c == '/') { + /* + * We have two slashes -- we can just advance + * to the next character. + */ + goto next; + } + + if (c != '.') { + /* + * This is not "." and it's not ".." -- we can + * just store the "/" and this character and + * drive on. + */ + dest[j++] = '/'; + dest[j++] = c; + continue; + } + + c = dtrace_load8(src + i++); + + if (c == '/') { + /* + * This is a "/./" component. We're not going + * to store anything in the destination buffer; + * we're just going to go to the next component. + */ + goto next; + } + + if (c != '.') { + /* + * This is not ".." -- we can just store the + * "/." and this character and continue + * processing. + */ + dest[j++] = '/'; + dest[j++] = '.'; + dest[j++] = c; + continue; + } + + c = dtrace_load8(src + i++); + + if (c != '/' && c != '\0') { + /* + * This is not ".." -- it's "..[mumble]". + * We'll store the "/.." and this character + * and continue processing. + */ + dest[j++] = '/'; + dest[j++] = '.'; + dest[j++] = '.'; + dest[j++] = c; + continue; + } + + /* + * This is "/../" or "/..\0". We need to back up + * our destination pointer until we find a "/". + */ + i--; + while (j != 0 && dest[--j] != '/') + continue; + + if (c == '\0') + dest[++j] = '/'; + } while (c != '\0'); + + dest[j] = '\0'; + regs[rd] = (uintptr_t)dest; + mstate->dtms_scratch_ptr += size; + break; + } + + case DIF_SUBR_INET_NTOA: + case DIF_SUBR_INET_NTOA6: + case DIF_SUBR_INET_NTOP: { + size_t size; + int af, argi, i; + char *base, *end; + + if (subr == DIF_SUBR_INET_NTOP) { + af = (int)tupregs[0].dttk_value; + argi = 1; + } else { + af = subr == DIF_SUBR_INET_NTOA ? AF_INET: AF_INET6; + argi = 0; + } + + if (af == AF_INET) { + ipaddr_t ip4; + uint8_t *ptr8, val; + + /* + * Safely load the IPv4 address. + */ + ip4 = dtrace_load32(tupregs[argi].dttk_value); + + /* + * Check an IPv4 string will fit in scratch. + */ + size = INET_ADDRSTRLEN; + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + base = (char *)mstate->dtms_scratch_ptr; + end = (char *)mstate->dtms_scratch_ptr + size - 1; + + /* + * Stringify as a dotted decimal quad. + */ + *end-- = '\0'; + ptr8 = (uint8_t *)&ip4; + for (i = 3; i >= 0; i--) { + val = ptr8[i]; + + if (val == 0) { + *end-- = '0'; + } else { + for (; val; val /= 10) { + *end-- = '0' + (val % 10); + } + } + + if (i > 0) + *end-- = '.'; + } + ASSERT(end + 1 >= base); + + } else if (af == AF_INET6) { + struct in6_addr ip6; + int firstzero, tryzero, numzero, v6end; + uint16_t val; + const char digits[] = "0123456789abcdef"; + + /* + * Stringify using RFC 1884 convention 2 - 16 bit + * hexadecimal values with a zero-run compression. + * Lower case hexadecimal digits are used. + * eg, fe80::214:4fff:fe0b:76c8. + * The IPv4 embedded form is returned for inet_ntop, + * just the IPv4 string is returned for inet_ntoa6. + */ + + /* + * Safely load the IPv6 address. + */ + dtrace_bcopy( + (void *)(uintptr_t)tupregs[argi].dttk_value, + (void *)(uintptr_t)&ip6, sizeof (struct in6_addr)); + + /* + * Check an IPv6 string will fit in scratch. + */ + size = INET6_ADDRSTRLEN; + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + base = (char *)mstate->dtms_scratch_ptr; + end = (char *)mstate->dtms_scratch_ptr + size - 1; + *end-- = '\0'; + + /* + * Find the longest run of 16 bit zero values + * for the single allowed zero compression - "::". + */ + firstzero = -1; + tryzero = -1; + numzero = 1; + for (i = 0; i < sizeof (struct in6_addr); i++) { +#if defined(sun) + if (ip6._S6_un._S6_u8[i] == 0 && +#else + if (ip6.__u6_addr.__u6_addr8[i] == 0 && +#endif + tryzero == -1 && i % 2 == 0) { + tryzero = i; + continue; + } + + if (tryzero != -1 && +#if defined(sun) + (ip6._S6_un._S6_u8[i] != 0 || +#else + (ip6.__u6_addr.__u6_addr8[i] != 0 || +#endif + i == sizeof (struct in6_addr) - 1)) { + + if (i - tryzero <= numzero) { + tryzero = -1; + continue; + } + + firstzero = tryzero; + numzero = i - i % 2 - tryzero; + tryzero = -1; + +#if defined(sun) + if (ip6._S6_un._S6_u8[i] == 0 && +#else + if (ip6.__u6_addr.__u6_addr8[i] == 0 && +#endif + i == sizeof (struct in6_addr) - 1) + numzero += 2; + } + } + ASSERT(firstzero + numzero <= sizeof (struct in6_addr)); + + /* + * Check for an IPv4 embedded address. + */ + v6end = sizeof (struct in6_addr) - 2; + if (IN6_IS_ADDR_V4MAPPED(&ip6) || + IN6_IS_ADDR_V4COMPAT(&ip6)) { + for (i = sizeof (struct in6_addr) - 1; + i >= DTRACE_V4MAPPED_OFFSET; i--) { + ASSERT(end >= base); + +#if defined(sun) + val = ip6._S6_un._S6_u8[i]; +#else + val = ip6.__u6_addr.__u6_addr8[i]; +#endif + + if (val == 0) { + *end-- = '0'; + } else { + for (; val; val /= 10) { + *end-- = '0' + val % 10; + } + } + + if (i > DTRACE_V4MAPPED_OFFSET) + *end-- = '.'; + } + + if (subr == DIF_SUBR_INET_NTOA6) + goto inetout; + + /* + * Set v6end to skip the IPv4 address that + * we have already stringified. + */ + v6end = 10; + } + + /* + * Build the IPv6 string by working through the + * address in reverse. + */ + for (i = v6end; i >= 0; i -= 2) { + ASSERT(end >= base); + + if (i == firstzero + numzero - 2) { + *end-- = ':'; + *end-- = ':'; + i -= numzero - 2; + continue; + } + + if (i < 14 && i != firstzero - 2) + *end-- = ':'; + +#if defined(sun) + val = (ip6._S6_un._S6_u8[i] << 8) + + ip6._S6_un._S6_u8[i + 1]; +#else + val = (ip6.__u6_addr.__u6_addr8[i] << 8) + + ip6.__u6_addr.__u6_addr8[i + 1]; +#endif + + if (val == 0) { + *end-- = '0'; + } else { + for (; val; val /= 16) { + *end-- = digits[val % 16]; + } + } + } + ASSERT(end + 1 >= base); + + } else { + /* + * The user didn't use AH_INET or AH_INET6. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + regs[rd] = 0; + break; + } + +inetout: regs[rd] = (uintptr_t)end + 1; + mstate->dtms_scratch_ptr += size; + break; + } + + case DIF_SUBR_MEMREF: { + uintptr_t size = 2 * sizeof(uintptr_t); + uintptr_t *memref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t)); + size_t scratch_size = ((uintptr_t) memref - mstate->dtms_scratch_ptr) + size; + + /* address and length */ + memref[0] = tupregs[0].dttk_value; + memref[1] = tupregs[1].dttk_value; + + regs[rd] = (uintptr_t) memref; + mstate->dtms_scratch_ptr += scratch_size; + break; + } + + case DIF_SUBR_TYPEREF: { + uintptr_t size = 4 * sizeof(uintptr_t); + uintptr_t *typeref = (uintptr_t *) P2ROUNDUP(mstate->dtms_scratch_ptr, sizeof(uintptr_t)); + size_t scratch_size = ((uintptr_t) typeref - mstate->dtms_scratch_ptr) + size; + + /* address, num_elements, type_str, type_len */ + typeref[0] = tupregs[0].dttk_value; + typeref[1] = tupregs[1].dttk_value; + typeref[2] = tupregs[2].dttk_value; + typeref[3] = tupregs[3].dttk_value; + + regs[rd] = (uintptr_t) typeref; + mstate->dtms_scratch_ptr += scratch_size; + break; + } + } +} + +/* + * Emulate the execution of DTrace IR instructions specified by the given + * DIF object. This function is deliberately void of assertions as all of + * the necessary checks are handled by a call to dtrace_difo_validate(). + */ +static uint64_t +dtrace_dif_emulate(dtrace_difo_t *difo, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate, dtrace_state_t *state) +{ + const dif_instr_t *text = difo->dtdo_buf; + const uint_t textlen = difo->dtdo_len; + const char *strtab = difo->dtdo_strtab; + const uint64_t *inttab = difo->dtdo_inttab; + + uint64_t rval = 0; + dtrace_statvar_t *svar; + dtrace_dstate_t *dstate = &vstate->dtvs_dynvars; + dtrace_difv_t *v; + volatile uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; + volatile uintptr_t *illval = &cpu_core[curcpu].cpuc_dtrace_illval; + + dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */ + uint64_t regs[DIF_DIR_NREGS]; + uint64_t *tmp; + + uint8_t cc_n = 0, cc_z = 0, cc_v = 0, cc_c = 0; + int64_t cc_r; + uint_t pc = 0, id, opc = 0; + uint8_t ttop = 0; + dif_instr_t instr; + uint_t r1, r2, rd; + + /* + * We stash the current DIF object into the machine state: we need it + * for subsequent access checking. + */ + mstate->dtms_difo = difo; + + regs[DIF_REG_R0] = 0; /* %r0 is fixed at zero */ + + while (pc < textlen && !(*flags & CPU_DTRACE_FAULT)) { + opc = pc; + + instr = text[pc++]; + r1 = DIF_INSTR_R1(instr); + r2 = DIF_INSTR_R2(instr); + rd = DIF_INSTR_RD(instr); + + switch (DIF_INSTR_OP(instr)) { + case DIF_OP_OR: + regs[rd] = regs[r1] | regs[r2]; + break; + case DIF_OP_XOR: + regs[rd] = regs[r1] ^ regs[r2]; + break; + case DIF_OP_AND: + regs[rd] = regs[r1] & regs[r2]; + break; + case DIF_OP_SLL: + regs[rd] = regs[r1] << regs[r2]; + break; + case DIF_OP_SRL: + regs[rd] = regs[r1] >> regs[r2]; + break; + case DIF_OP_SUB: + regs[rd] = regs[r1] - regs[r2]; + break; + case DIF_OP_ADD: + regs[rd] = regs[r1] + regs[r2]; + break; + case DIF_OP_MUL: + regs[rd] = regs[r1] * regs[r2]; + break; + case DIF_OP_SDIV: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + regs[rd] = (int64_t)regs[r1] / + (int64_t)regs[r2]; + } + break; + + case DIF_OP_UDIV: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + regs[rd] = regs[r1] / regs[r2]; + } + break; + + case DIF_OP_SREM: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + regs[rd] = (int64_t)regs[r1] % + (int64_t)regs[r2]; + } + break; + + case DIF_OP_UREM: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + regs[rd] = regs[r1] % regs[r2]; + } + break; + + case DIF_OP_NOT: + regs[rd] = ~regs[r1]; + break; + case DIF_OP_MOV: + regs[rd] = regs[r1]; + break; + case DIF_OP_CMP: + cc_r = regs[r1] - regs[r2]; + cc_n = cc_r < 0; + cc_z = cc_r == 0; + cc_v = 0; + cc_c = regs[r1] < regs[r2]; + break; + case DIF_OP_TST: + cc_n = cc_v = cc_c = 0; + cc_z = regs[r1] == 0; + break; + case DIF_OP_BA: + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BE: + if (cc_z) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BNE: + if (cc_z == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BG: + if ((cc_z | (cc_n ^ cc_v)) == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BGU: + if ((cc_c | cc_z) == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BGE: + if ((cc_n ^ cc_v) == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BGEU: + if (cc_c == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BL: + if (cc_n ^ cc_v) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BLU: + if (cc_c) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BLE: + if (cc_z | (cc_n ^ cc_v)) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BLEU: + if (cc_c | cc_z) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_RLDSB: + if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDSB: + regs[rd] = (int8_t)dtrace_load8(regs[r1]); + break; + case DIF_OP_RLDSH: + if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDSH: + regs[rd] = (int16_t)dtrace_load16(regs[r1]); + break; + case DIF_OP_RLDSW: + if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDSW: + regs[rd] = (int32_t)dtrace_load32(regs[r1]); + break; + case DIF_OP_RLDUB: + if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDUB: + regs[rd] = dtrace_load8(regs[r1]); + break; + case DIF_OP_RLDUH: + if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDUH: + regs[rd] = dtrace_load16(regs[r1]); + break; + case DIF_OP_RLDUW: + if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDUW: + regs[rd] = dtrace_load32(regs[r1]); + break; + case DIF_OP_RLDX: + if (!dtrace_canstore(regs[r1], 8, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDX: + regs[rd] = dtrace_load64(regs[r1]); + break; + case DIF_OP_ULDSB: + regs[rd] = (int8_t) + dtrace_fuword8((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDSH: + regs[rd] = (int16_t) + dtrace_fuword16((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDSW: + regs[rd] = (int32_t) + dtrace_fuword32((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDUB: + regs[rd] = + dtrace_fuword8((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDUH: + regs[rd] = + dtrace_fuword16((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDUW: + regs[rd] = + dtrace_fuword32((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDX: + regs[rd] = + dtrace_fuword64((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_RET: + rval = regs[rd]; + pc = textlen; + break; + case DIF_OP_NOP: + break; + case DIF_OP_SETX: + regs[rd] = inttab[DIF_INSTR_INTEGER(instr)]; + break; + case DIF_OP_SETS: + regs[rd] = (uint64_t)(uintptr_t) + (strtab + DIF_INSTR_STRING(instr)); + break; + case DIF_OP_SCMP: { + size_t sz = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t s1 = regs[r1]; + uintptr_t s2 = regs[r2]; + + if (s1 != 0 && + !dtrace_strcanload(s1, sz, mstate, vstate)) + break; + if (s2 != 0 && + !dtrace_strcanload(s2, sz, mstate, vstate)) + break; + + cc_r = dtrace_strncmp((char *)s1, (char *)s2, sz); + + cc_n = cc_r < 0; + cc_z = cc_r == 0; + cc_v = cc_c = 0; + break; + } + case DIF_OP_LDGA: + regs[rd] = dtrace_dif_variable(mstate, state, + r1, regs[r2]); + break; + case DIF_OP_LDGS: + id = DIF_INSTR_VAR(instr); + + if (id >= DIF_VAR_OTHER_UBASE) { + uintptr_t a; + + id -= DIF_VAR_OTHER_UBASE; + svar = vstate->dtvs_globals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (!(v->dtdv_type.dtdt_flags & DIF_TF_BYREF)) { + regs[rd] = svar->dtsv_data; + break; + } + + a = (uintptr_t)svar->dtsv_data; + + if (*(uint8_t *)a == UINT8_MAX) { + /* + * If the 0th byte is set to UINT8_MAX + * then this is to be treated as a + * reference to a NULL variable. + */ + regs[rd] = 0; + } else { + regs[rd] = a + sizeof (uint64_t); + } + + break; + } + + regs[rd] = dtrace_dif_variable(mstate, state, id, 0); + break; + + case DIF_OP_STGS: + id = DIF_INSTR_VAR(instr); + + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + svar = vstate->dtvs_globals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + uintptr_t a = (uintptr_t)svar->dtsv_data; + + ASSERT(a != 0); + ASSERT(svar->dtsv_size != 0); + + if (regs[rd] == 0) { + *(uint8_t *)a = UINT8_MAX; + break; + } else { + *(uint8_t *)a = 0; + a += sizeof (uint64_t); + } + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], &v->dtdv_type, + mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + (void *)a, &v->dtdv_type); + break; + } + + svar->dtsv_data = regs[rd]; + break; + + case DIF_OP_LDTA: + /* + * There are no DTrace built-in thread-local arrays at + * present. This opcode is saved for future work. + */ + *flags |= CPU_DTRACE_ILLOP; + regs[rd] = 0; + break; + + case DIF_OP_LDLS: + id = DIF_INSTR_VAR(instr); + + if (id < DIF_VAR_OTHER_UBASE) { + /* + * For now, this has no meaning. + */ + regs[rd] = 0; + break; + } + + id -= DIF_VAR_OTHER_UBASE; + + ASSERT(id < vstate->dtvs_nlocals); + ASSERT(vstate->dtvs_locals != NULL); + + svar = vstate->dtvs_locals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + uintptr_t a = (uintptr_t)svar->dtsv_data; + size_t sz = v->dtdv_type.dtdt_size; + + sz += sizeof (uint64_t); + ASSERT(svar->dtsv_size == NCPU * sz); + a += curcpu * sz; + + if (*(uint8_t *)a == UINT8_MAX) { + /* + * If the 0th byte is set to UINT8_MAX + * then this is to be treated as a + * reference to a NULL variable. + */ + regs[rd] = 0; + } else { + regs[rd] = a + sizeof (uint64_t); + } + + break; + } + + ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t)); + tmp = (uint64_t *)(uintptr_t)svar->dtsv_data; + regs[rd] = tmp[curcpu]; + break; + + case DIF_OP_STLS: + id = DIF_INSTR_VAR(instr); + + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + ASSERT(id < vstate->dtvs_nlocals); + + ASSERT(vstate->dtvs_locals != NULL); + svar = vstate->dtvs_locals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + uintptr_t a = (uintptr_t)svar->dtsv_data; + size_t sz = v->dtdv_type.dtdt_size; + + sz += sizeof (uint64_t); + ASSERT(svar->dtsv_size == NCPU * sz); + a += curcpu * sz; + + if (regs[rd] == 0) { + *(uint8_t *)a = UINT8_MAX; + break; + } else { + *(uint8_t *)a = 0; + a += sizeof (uint64_t); + } + + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], &v->dtdv_type, + mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + (void *)a, &v->dtdv_type); + break; + } + + ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t)); + tmp = (uint64_t *)(uintptr_t)svar->dtsv_data; + tmp[curcpu] = regs[rd]; + break; + + case DIF_OP_LDTS: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + v = &vstate->dtvs_tlocals[id]; + + key = &tupregs[DIF_DTR_NREGS]; + key[0].dttk_value = (uint64_t)id; + key[0].dttk_size = 0; + DTRACE_TLS_THRKEY(key[1].dttk_value); + key[1].dttk_size = 0; + + dvar = dtrace_dynvar(dstate, 2, key, + sizeof (uint64_t), DTRACE_DYNVAR_NOALLOC, + mstate, vstate); + + if (dvar == NULL) { + regs[rd] = 0; + break; + } + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data; + } else { + regs[rd] = *((uint64_t *)dvar->dtdv_data); + } + + break; + } + + case DIF_OP_STTS: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + key = &tupregs[DIF_DTR_NREGS]; + key[0].dttk_value = (uint64_t)id; + key[0].dttk_size = 0; + DTRACE_TLS_THRKEY(key[1].dttk_value); + key[1].dttk_size = 0; + v = &vstate->dtvs_tlocals[id]; + + dvar = dtrace_dynvar(dstate, 2, key, + v->dtdv_type.dtdt_size > sizeof (uint64_t) ? + v->dtdv_type.dtdt_size : sizeof (uint64_t), + regs[rd] ? DTRACE_DYNVAR_ALLOC : + DTRACE_DYNVAR_DEALLOC, mstate, vstate); + + /* + * Given that we're storing to thread-local data, + * we need to flush our predicate cache. + */ + curthread->t_predcache = 0; + + if (dvar == NULL) + break; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], + &v->dtdv_type, mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + dvar->dtdv_data, &v->dtdv_type); + } else { + *((uint64_t *)dvar->dtdv_data) = regs[rd]; + } + + break; + } + + case DIF_OP_SRA: + regs[rd] = (int64_t)regs[r1] >> regs[r2]; + break; + + case DIF_OP_CALL: + dtrace_dif_subr(DIF_INSTR_SUBR(instr), rd, + regs, tupregs, ttop, mstate, state); + break; + + case DIF_OP_PUSHTR: + if (ttop == DIF_DTR_NREGS) { + *flags |= CPU_DTRACE_TUPOFLOW; + break; + } + + if (r1 == DIF_TYPE_STRING) { + /* + * If this is a string type and the size is 0, + * we'll use the system-wide default string + * size. Note that we are _not_ looking at + * the value of the DTRACEOPT_STRSIZE option; + * had this been set, we would expect to have + * a non-zero size value in the "pushtr". + */ + tupregs[ttop].dttk_size = + dtrace_strlen((char *)(uintptr_t)regs[rd], + regs[r2] ? regs[r2] : + dtrace_strsize_default) + 1; + } else { + tupregs[ttop].dttk_size = regs[r2]; + } + + tupregs[ttop++].dttk_value = regs[rd]; + break; + + case DIF_OP_PUSHTV: + if (ttop == DIF_DTR_NREGS) { + *flags |= CPU_DTRACE_TUPOFLOW; + break; + } + + tupregs[ttop].dttk_value = regs[rd]; + tupregs[ttop++].dttk_size = 0; + break; + + case DIF_OP_POPTS: + if (ttop != 0) + ttop--; + break; + + case DIF_OP_FLUSHTS: + ttop = 0; + break; + + case DIF_OP_LDGAA: + case DIF_OP_LDTAA: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key = tupregs; + uint_t nkeys = ttop; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + key[nkeys].dttk_value = (uint64_t)id; + key[nkeys++].dttk_size = 0; + + if (DIF_INSTR_OP(instr) == DIF_OP_LDTAA) { + DTRACE_TLS_THRKEY(key[nkeys].dttk_value); + key[nkeys++].dttk_size = 0; + v = &vstate->dtvs_tlocals[id]; + } else { + v = &vstate->dtvs_globals[id]->dtsv_var; + } + + dvar = dtrace_dynvar(dstate, nkeys, key, + v->dtdv_type.dtdt_size > sizeof (uint64_t) ? + v->dtdv_type.dtdt_size : sizeof (uint64_t), + DTRACE_DYNVAR_NOALLOC, mstate, vstate); + + if (dvar == NULL) { + regs[rd] = 0; + break; + } + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data; + } else { + regs[rd] = *((uint64_t *)dvar->dtdv_data); + } + + break; + } + + case DIF_OP_STGAA: + case DIF_OP_STTAA: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key = tupregs; + uint_t nkeys = ttop; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + key[nkeys].dttk_value = (uint64_t)id; + key[nkeys++].dttk_size = 0; + + if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) { + DTRACE_TLS_THRKEY(key[nkeys].dttk_value); + key[nkeys++].dttk_size = 0; + v = &vstate->dtvs_tlocals[id]; + } else { + v = &vstate->dtvs_globals[id]->dtsv_var; + } + + dvar = dtrace_dynvar(dstate, nkeys, key, + v->dtdv_type.dtdt_size > sizeof (uint64_t) ? + v->dtdv_type.dtdt_size : sizeof (uint64_t), + regs[rd] ? DTRACE_DYNVAR_ALLOC : + DTRACE_DYNVAR_DEALLOC, mstate, vstate); + + if (dvar == NULL) + break; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], &v->dtdv_type, + mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + dvar->dtdv_data, &v->dtdv_type); + } else { + *((uint64_t *)dvar->dtdv_data) = regs[rd]; + } + + break; + } + + case DIF_OP_ALLOCS: { + uintptr_t ptr = P2ROUNDUP(mstate->dtms_scratch_ptr, 8); + size_t size = ptr - mstate->dtms_scratch_ptr + regs[r1]; + + /* + * Rounding up the user allocation size could have + * overflowed large, bogus allocations (like -1ULL) to + * 0. + */ + if (size < regs[r1] || + !DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + dtrace_bzero((void *) mstate->dtms_scratch_ptr, size); + mstate->dtms_scratch_ptr += size; + regs[rd] = ptr; + break; + } + + case DIF_OP_COPYS: + if (!dtrace_canstore(regs[rd], regs[r2], + mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + if (!dtrace_canload(regs[r1], regs[r2], mstate, vstate)) + break; + + dtrace_bcopy((void *)(uintptr_t)regs[r1], + (void *)(uintptr_t)regs[rd], (size_t)regs[r2]); + break; + + case DIF_OP_STB: + if (!dtrace_canstore(regs[rd], 1, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + *((uint8_t *)(uintptr_t)regs[rd]) = (uint8_t)regs[r1]; + break; + + case DIF_OP_STH: + if (!dtrace_canstore(regs[rd], 2, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + if (regs[rd] & 1) { + *flags |= CPU_DTRACE_BADALIGN; + *illval = regs[rd]; + break; + } + *((uint16_t *)(uintptr_t)regs[rd]) = (uint16_t)regs[r1]; + break; + + case DIF_OP_STW: + if (!dtrace_canstore(regs[rd], 4, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + if (regs[rd] & 3) { + *flags |= CPU_DTRACE_BADALIGN; + *illval = regs[rd]; + break; + } + *((uint32_t *)(uintptr_t)regs[rd]) = (uint32_t)regs[r1]; + break; + + case DIF_OP_STX: + if (!dtrace_canstore(regs[rd], 8, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + if (regs[rd] & 7) { + *flags |= CPU_DTRACE_BADALIGN; + *illval = regs[rd]; + break; + } + *((uint64_t *)(uintptr_t)regs[rd]) = regs[r1]; + break; + } + } + + if (!(*flags & CPU_DTRACE_FAULT)) + return (rval); + + mstate->dtms_fltoffs = opc * sizeof (dif_instr_t); + mstate->dtms_present |= DTRACE_MSTATE_FLTOFFS; + + return (0); +} + +static void +dtrace_action_breakpoint(dtrace_ecb_t *ecb) +{ + dtrace_probe_t *probe = ecb->dte_probe; + dtrace_provider_t *prov = probe->dtpr_provider; + char c[DTRACE_FULLNAMELEN + 80], *str; + char *msg = "dtrace: breakpoint action at probe "; + char *ecbmsg = " (ecb "; + uintptr_t mask = (0xf << (sizeof (uintptr_t) * NBBY / 4)); + uintptr_t val = (uintptr_t)ecb; + int shift = (sizeof (uintptr_t) * NBBY) - 4, i = 0; + + if (dtrace_destructive_disallow) + return; + + /* + * It's impossible to be taking action on the NULL probe. + */ + ASSERT(probe != NULL); + + /* + * This is a poor man's (destitute man's?) sprintf(): we want to + * print the provider name, module name, function name and name of + * the probe, along with the hex address of the ECB with the breakpoint + * action -- all of which we must place in the character buffer by + * hand. + */ + while (*msg != '\0') + c[i++] = *msg++; + + for (str = prov->dtpv_name; *str != '\0'; str++) + c[i++] = *str; + c[i++] = ':'; + + for (str = probe->dtpr_mod; *str != '\0'; str++) + c[i++] = *str; + c[i++] = ':'; + + for (str = probe->dtpr_func; *str != '\0'; str++) + c[i++] = *str; + c[i++] = ':'; + + for (str = probe->dtpr_name; *str != '\0'; str++) + c[i++] = *str; + + while (*ecbmsg != '\0') + c[i++] = *ecbmsg++; + + while (shift >= 0) { + mask = (uintptr_t)0xf << shift; + + if (val >= ((uintptr_t)1 << shift)) + c[i++] = "0123456789abcdef"[(val & mask) >> shift]; + shift -= 4; + } + + c[i++] = ')'; + c[i] = '\0'; + +#if defined(sun) + debug_enter(c); +#else + kdb_enter_why(KDB_WHY_DTRACE, "breakpoint action"); +#endif +} + +static void +dtrace_action_panic(dtrace_ecb_t *ecb) +{ + dtrace_probe_t *probe = ecb->dte_probe; + + /* + * It's impossible to be taking action on the NULL probe. + */ + ASSERT(probe != NULL); + + if (dtrace_destructive_disallow) + return; + + if (dtrace_panicked != NULL) + return; + + if (dtrace_casptr(&dtrace_panicked, NULL, curthread) != NULL) + return; + + /* + * We won the right to panic. (We want to be sure that only one + * thread calls panic() from dtrace_probe(), and that panic() is + * called exactly once.) + */ + dtrace_panic("dtrace: panic action at probe %s:%s:%s:%s (ecb %p)", + probe->dtpr_provider->dtpv_name, probe->dtpr_mod, + probe->dtpr_func, probe->dtpr_name, (void *)ecb); +} + +static void +dtrace_action_raise(uint64_t sig) +{ + if (dtrace_destructive_disallow) + return; + + if (sig >= NSIG) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return; + } + +#if defined(sun) + /* + * raise() has a queue depth of 1 -- we ignore all subsequent + * invocations of the raise() action. + */ + if (curthread->t_dtrace_sig == 0) + curthread->t_dtrace_sig = (uint8_t)sig; + + curthread->t_sig_check = 1; + aston(curthread); +#else + struct proc *p = curproc; + PROC_LOCK(p); + psignal(p, sig); + PROC_UNLOCK(p); +#endif +} + +static void +dtrace_action_stop(void) +{ + if (dtrace_destructive_disallow) + return; + +#if defined(sun) + if (!curthread->t_dtrace_stop) { + curthread->t_dtrace_stop = 1; + curthread->t_sig_check = 1; + aston(curthread); + } +#else + struct proc *p = curproc; + PROC_LOCK(p); + psignal(p, SIGSTOP); + PROC_UNLOCK(p); +#endif +} + +static void +dtrace_action_chill(dtrace_mstate_t *mstate, hrtime_t val) +{ + hrtime_t now; + volatile uint16_t *flags; +#if defined(sun) + cpu_t *cpu = CPU; +#else + cpu_t *cpu = &solaris_cpu[curcpu]; +#endif + + if (dtrace_destructive_disallow) + return; + + flags = (volatile uint16_t *)&cpu_core[cpu->cpu_id].cpuc_dtrace_flags; + + now = dtrace_gethrtime(); + + if (now - cpu->cpu_dtrace_chillmark > dtrace_chill_interval) { + /* + * We need to advance the mark to the current time. + */ + cpu->cpu_dtrace_chillmark = now; + cpu->cpu_dtrace_chilled = 0; + } + + /* + * Now check to see if the requested chill time would take us over + * the maximum amount of time allowed in the chill interval. (Or + * worse, if the calculation itself induces overflow.) + */ + if (cpu->cpu_dtrace_chilled + val > dtrace_chill_max || + cpu->cpu_dtrace_chilled + val < cpu->cpu_dtrace_chilled) { + *flags |= CPU_DTRACE_ILLOP; + return; + } + + while (dtrace_gethrtime() - now < val) + continue; + + /* + * Normally, we assure that the value of the variable "timestamp" does + * not change within an ECB. The presence of chill() represents an + * exception to this rule, however. + */ + mstate->dtms_present &= ~DTRACE_MSTATE_TIMESTAMP; + cpu->cpu_dtrace_chilled += val; +} + +#if defined(sun) +static void +dtrace_action_ustack(dtrace_mstate_t *mstate, dtrace_state_t *state, + uint64_t *buf, uint64_t arg) +{ + int nframes = DTRACE_USTACK_NFRAMES(arg); + int strsize = DTRACE_USTACK_STRSIZE(arg); + uint64_t *pcs = &buf[1], *fps; + char *str = (char *)&pcs[nframes]; + int size, offs = 0, i, j; + uintptr_t old = mstate->dtms_scratch_ptr, saved; + uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; + char *sym; + + /* + * Should be taking a faster path if string space has not been + * allocated. + */ + ASSERT(strsize != 0); + + /* + * We will first allocate some temporary space for the frame pointers. + */ + fps = (uint64_t *)P2ROUNDUP(mstate->dtms_scratch_ptr, 8); + size = (uintptr_t)fps - mstate->dtms_scratch_ptr + + (nframes * sizeof (uint64_t)); + + if (!DTRACE_INSCRATCH(mstate, size)) { + /* + * Not enough room for our frame pointers -- need to indicate + * that we ran out of scratch space. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + return; + } + + mstate->dtms_scratch_ptr += size; + saved = mstate->dtms_scratch_ptr; + + /* + * Now get a stack with both program counters and frame pointers. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_getufpstack(buf, fps, nframes + 1); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + /* + * If that faulted, we're cooked. + */ + if (*flags & CPU_DTRACE_FAULT) + goto out; + + /* + * Now we want to walk up the stack, calling the USTACK helper. For + * each iteration, we restore the scratch pointer. + */ + for (i = 0; i < nframes; i++) { + mstate->dtms_scratch_ptr = saved; + + if (offs >= strsize) + break; + + sym = (char *)(uintptr_t)dtrace_helper( + DTRACE_HELPER_ACTION_USTACK, + mstate, state, pcs[i], fps[i]); + + /* + * If we faulted while running the helper, we're going to + * clear the fault and null out the corresponding string. + */ + if (*flags & CPU_DTRACE_FAULT) { + *flags &= ~CPU_DTRACE_FAULT; + str[offs++] = '\0'; + continue; + } + + if (sym == NULL) { + str[offs++] = '\0'; + continue; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + + /* + * Now copy in the string that the helper returned to us. + */ + for (j = 0; offs + j < strsize; j++) { + if ((str[offs + j] = sym[j]) == '\0') + break; + } + + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + offs += j + 1; + } + + if (offs >= strsize) { + /* + * If we didn't have room for all of the strings, we don't + * abort processing -- this needn't be a fatal error -- but we + * still want to increment a counter (dts_stkstroverflows) to + * allow this condition to be warned about. (If this is from + * a jstack() action, it is easily tuned via jstackstrsize.) + */ + dtrace_error(&state->dts_stkstroverflows); + } + + while (offs < strsize) + str[offs++] = '\0'; + +out: + mstate->dtms_scratch_ptr = old; +} +#endif + +/* + * If you're looking for the epicenter of DTrace, you just found it. This + * is the function called by the provider to fire a probe -- from which all + * subsequent probe-context DTrace activity emanates. + */ +void +dtrace_probe(dtrace_id_t id, uintptr_t arg0, uintptr_t arg1, + uintptr_t arg2, uintptr_t arg3, uintptr_t arg4) +{ + processorid_t cpuid; + dtrace_icookie_t cookie; + dtrace_probe_t *probe; + dtrace_mstate_t mstate; + dtrace_ecb_t *ecb; + dtrace_action_t *act; + intptr_t offs; + size_t size; + int vtime, onintr; + volatile uint16_t *flags; + hrtime_t now; + +#if defined(sun) + /* + * Kick out immediately if this CPU is still being born (in which case + * curthread will be set to -1) or the current thread can't allow + * probes in its current context. + */ + if (((uintptr_t)curthread & 1) || (curthread->t_flag & T_DONTDTRACE)) + return; +#endif + + cookie = dtrace_interrupt_disable(); + probe = dtrace_probes[id - 1]; + cpuid = curcpu; + onintr = CPU_ON_INTR(CPU); + + if (!onintr && probe->dtpr_predcache != DTRACE_CACHEIDNONE && + probe->dtpr_predcache == curthread->t_predcache) { + /* + * We have hit in the predicate cache; we know that + * this predicate would evaluate to be false. + */ + dtrace_interrupt_enable(cookie); + return; + } + +#if defined(sun) + if (panic_quiesce) { +#else + if (panicstr != NULL) { +#endif + /* + * We don't trace anything if we're panicking. + */ + dtrace_interrupt_enable(cookie); + return; + } + + now = dtrace_gethrtime(); + vtime = dtrace_vtime_references != 0; + + if (vtime && curthread->t_dtrace_start) + curthread->t_dtrace_vtime += now - curthread->t_dtrace_start; + + mstate.dtms_difo = NULL; + mstate.dtms_probe = probe; + mstate.dtms_strtok = 0; + mstate.dtms_arg[0] = arg0; + mstate.dtms_arg[1] = arg1; + mstate.dtms_arg[2] = arg2; + mstate.dtms_arg[3] = arg3; + mstate.dtms_arg[4] = arg4; + + flags = (volatile uint16_t *)&cpu_core[cpuid].cpuc_dtrace_flags; + + for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) { + dtrace_predicate_t *pred = ecb->dte_predicate; + dtrace_state_t *state = ecb->dte_state; + dtrace_buffer_t *buf = &state->dts_buffer[cpuid]; + dtrace_buffer_t *aggbuf = &state->dts_aggbuffer[cpuid]; + dtrace_vstate_t *vstate = &state->dts_vstate; + dtrace_provider_t *prov = probe->dtpr_provider; + int committed = 0; + caddr_t tomax; + + /* + * A little subtlety with the following (seemingly innocuous) + * declaration of the automatic 'val': by looking at the + * code, you might think that it could be declared in the + * action processing loop, below. (That is, it's only used in + * the action processing loop.) However, it must be declared + * out of that scope because in the case of DIF expression + * arguments to aggregating actions, one iteration of the + * action loop will use the last iteration's value. + */ + uint64_t val = 0; + + mstate.dtms_present = DTRACE_MSTATE_ARGS | DTRACE_MSTATE_PROBE; + *flags &= ~CPU_DTRACE_ERROR; + + if (prov == dtrace_provider) { + /* + * If dtrace itself is the provider of this probe, + * we're only going to continue processing the ECB if + * arg0 (the dtrace_state_t) is equal to the ECB's + * creating state. (This prevents disjoint consumers + * from seeing one another's metaprobes.) + */ + if (arg0 != (uint64_t)(uintptr_t)state) + continue; + } + + if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) { + /* + * We're not currently active. If our provider isn't + * the dtrace pseudo provider, we're not interested. + */ + if (prov != dtrace_provider) + continue; + + /* + * Now we must further check if we are in the BEGIN + * probe. If we are, we will only continue processing + * if we're still in WARMUP -- if one BEGIN enabling + * has invoked the exit() action, we don't want to + * evaluate subsequent BEGIN enablings. + */ + if (probe->dtpr_id == dtrace_probeid_begin && + state->dts_activity != DTRACE_ACTIVITY_WARMUP) { + ASSERT(state->dts_activity == + DTRACE_ACTIVITY_DRAINING); + continue; + } + } + + if (ecb->dte_cond) { + /* + * If the dte_cond bits indicate that this + * consumer is only allowed to see user-mode firings + * of this probe, call the provider's dtps_usermode() + * entry point to check that the probe was fired + * while in a user context. Skip this ECB if that's + * not the case. + */ + if ((ecb->dte_cond & DTRACE_COND_USERMODE) && + prov->dtpv_pops.dtps_usermode(prov->dtpv_arg, + probe->dtpr_id, probe->dtpr_arg) == 0) + continue; + +#if defined(sun) + /* + * This is more subtle than it looks. We have to be + * absolutely certain that CRED() isn't going to + * change out from under us so it's only legit to + * examine that structure if we're in constrained + * situations. Currently, the only times we'll this + * check is if a non-super-user has enabled the + * profile or syscall providers -- providers that + * allow visibility of all processes. For the + * profile case, the check above will ensure that + * we're examining a user context. + */ + if (ecb->dte_cond & DTRACE_COND_OWNER) { + cred_t *cr; + cred_t *s_cr = + ecb->dte_state->dts_cred.dcr_cred; + proc_t *proc; + + ASSERT(s_cr != NULL); + + if ((cr = CRED()) == NULL || + s_cr->cr_uid != cr->cr_uid || + s_cr->cr_uid != cr->cr_ruid || + s_cr->cr_uid != cr->cr_suid || + s_cr->cr_gid != cr->cr_gid || + s_cr->cr_gid != cr->cr_rgid || + s_cr->cr_gid != cr->cr_sgid || + (proc = ttoproc(curthread)) == NULL || + (proc->p_flag & SNOCD)) + continue; + } + + if (ecb->dte_cond & DTRACE_COND_ZONEOWNER) { + cred_t *cr; + cred_t *s_cr = + ecb->dte_state->dts_cred.dcr_cred; + + ASSERT(s_cr != NULL); + + if ((cr = CRED()) == NULL || + s_cr->cr_zone->zone_id != + cr->cr_zone->zone_id) + continue; + } +#endif + } + + if (now - state->dts_alive > dtrace_deadman_timeout) { + /* + * We seem to be dead. Unless we (a) have kernel + * destructive permissions (b) have expicitly enabled + * destructive actions and (c) destructive actions have + * not been disabled, we're going to transition into + * the KILLED state, from which no further processing + * on this state will be performed. + */ + if (!dtrace_priv_kernel_destructive(state) || + !state->dts_cred.dcr_destructive || + dtrace_destructive_disallow) { + void *activity = &state->dts_activity; + dtrace_activity_t current; + + do { + current = state->dts_activity; + } while (dtrace_cas32(activity, current, + DTRACE_ACTIVITY_KILLED) != current); + + continue; + } + } + + if ((offs = dtrace_buffer_reserve(buf, ecb->dte_needed, + ecb->dte_alignment, state, &mstate)) < 0) + continue; + + tomax = buf->dtb_tomax; + ASSERT(tomax != NULL); + + if (ecb->dte_size != 0) + DTRACE_STORE(uint32_t, tomax, offs, ecb->dte_epid); + + mstate.dtms_epid = ecb->dte_epid; + mstate.dtms_present |= DTRACE_MSTATE_EPID; + + if (state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) + mstate.dtms_access = DTRACE_ACCESS_KERNEL; + else + mstate.dtms_access = 0; + + if (pred != NULL) { + dtrace_difo_t *dp = pred->dtp_difo; + int rval; + + rval = dtrace_dif_emulate(dp, &mstate, vstate, state); + + if (!(*flags & CPU_DTRACE_ERROR) && !rval) { + dtrace_cacheid_t cid = probe->dtpr_predcache; + + if (cid != DTRACE_CACHEIDNONE && !onintr) { + /* + * Update the predicate cache... + */ + ASSERT(cid == pred->dtp_cacheid); + curthread->t_predcache = cid; + } + + continue; + } + } + + for (act = ecb->dte_action; !(*flags & CPU_DTRACE_ERROR) && + act != NULL; act = act->dta_next) { + size_t valoffs; + dtrace_difo_t *dp; + dtrace_recdesc_t *rec = &act->dta_rec; + + size = rec->dtrd_size; + valoffs = offs + rec->dtrd_offset; + + if (DTRACEACT_ISAGG(act->dta_kind)) { + uint64_t v = 0xbad; + dtrace_aggregation_t *agg; + + agg = (dtrace_aggregation_t *)act; + + if ((dp = act->dta_difo) != NULL) + v = dtrace_dif_emulate(dp, + &mstate, vstate, state); + + if (*flags & CPU_DTRACE_ERROR) + continue; + + /* + * Note that we always pass the expression + * value from the previous iteration of the + * action loop. This value will only be used + * if there is an expression argument to the + * aggregating action, denoted by the + * dtag_hasarg field. + */ + dtrace_aggregate(agg, buf, + offs, aggbuf, v, val); + continue; + } + + switch (act->dta_kind) { + case DTRACEACT_STOP: + if (dtrace_priv_proc_destructive(state)) + dtrace_action_stop(); + continue; + + case DTRACEACT_BREAKPOINT: + if (dtrace_priv_kernel_destructive(state)) + dtrace_action_breakpoint(ecb); + continue; + + case DTRACEACT_PANIC: + if (dtrace_priv_kernel_destructive(state)) + dtrace_action_panic(ecb); + continue; + + case DTRACEACT_STACK: + if (!dtrace_priv_kernel(state)) + continue; + + dtrace_getpcstack((pc_t *)(tomax + valoffs), + size / sizeof (pc_t), probe->dtpr_aframes, + DTRACE_ANCHORED(probe) ? NULL : + (uint32_t *)arg0); + continue; + +#if defined(sun) + case DTRACEACT_JSTACK: + case DTRACEACT_USTACK: + if (!dtrace_priv_proc(state)) + continue; + + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate.dtms_probe) && + CPU_ON_INTR(CPU)) { + int depth = DTRACE_USTACK_NFRAMES( + rec->dtrd_arg) + 1; + + dtrace_bzero((void *)(tomax + valoffs), + DTRACE_USTACK_STRSIZE(rec->dtrd_arg) + + depth * sizeof (uint64_t)); + + continue; + } + + if (DTRACE_USTACK_STRSIZE(rec->dtrd_arg) != 0 && + curproc->p_dtrace_helpers != NULL) { + /* + * This is the slow path -- we have + * allocated string space, and we're + * getting the stack of a process that + * has helpers. Call into a separate + * routine to perform this processing. + */ + dtrace_action_ustack(&mstate, state, + (uint64_t *)(tomax + valoffs), + rec->dtrd_arg); + continue; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_getupcstack((uint64_t *) + (tomax + valoffs), + DTRACE_USTACK_NFRAMES(rec->dtrd_arg) + 1); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + continue; +#endif + + default: + break; + } + + dp = act->dta_difo; + ASSERT(dp != NULL); + + val = dtrace_dif_emulate(dp, &mstate, vstate, state); + + if (*flags & CPU_DTRACE_ERROR) + continue; + + switch (act->dta_kind) { + case DTRACEACT_SPECULATE: + ASSERT(buf == &state->dts_buffer[cpuid]); + buf = dtrace_speculation_buffer(state, + cpuid, val); + + if (buf == NULL) { + *flags |= CPU_DTRACE_DROP; + continue; + } + + offs = dtrace_buffer_reserve(buf, + ecb->dte_needed, ecb->dte_alignment, + state, NULL); + + if (offs < 0) { + *flags |= CPU_DTRACE_DROP; + continue; + } + + tomax = buf->dtb_tomax; + ASSERT(tomax != NULL); + + if (ecb->dte_size != 0) + DTRACE_STORE(uint32_t, tomax, offs, + ecb->dte_epid); + continue; + + case DTRACEACT_PRINTM: { + /* The DIF returns a 'memref'. */ + uintptr_t *memref = (uintptr_t *)(uintptr_t) val; + + /* Get the size from the memref. */ + size = memref[1]; + + /* + * Check if the size exceeds the allocated + * buffer size. + */ + if (size + sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) { + /* Flag a drop! */ + *flags |= CPU_DTRACE_DROP; + continue; + } + + /* Store the size in the buffer first. */ + DTRACE_STORE(uintptr_t, tomax, + valoffs, size); + + /* + * Offset the buffer address to the start + * of the data. + */ + valoffs += sizeof(uintptr_t); + + /* + * Reset to the memory address rather than + * the memref array, then let the BYREF + * code below do the work to store the + * memory data in the buffer. + */ + val = memref[0]; + break; + } + + case DTRACEACT_PRINTT: { + /* The DIF returns a 'typeref'. */ + uintptr_t *typeref = (uintptr_t *)(uintptr_t) val; + char c = '\0' + 1; + size_t s; + + /* + * Get the type string length and round it + * up so that the data that follows is + * aligned for easy access. + */ + size_t typs = strlen((char *) typeref[2]) + 1; + typs = roundup(typs, sizeof(uintptr_t)); + + /* + *Get the size from the typeref using the + * number of elements and the type size. + */ + size = typeref[1] * typeref[3]; + + /* + * Check if the size exceeds the allocated + * buffer size. + */ + if (size + typs + 2 * sizeof(uintptr_t) > dp->dtdo_rtype.dtdt_size) { + /* Flag a drop! */ + *flags |= CPU_DTRACE_DROP; + + } + + /* Store the size in the buffer first. */ + DTRACE_STORE(uintptr_t, tomax, + valoffs, size); + valoffs += sizeof(uintptr_t); + + /* Store the type size in the buffer. */ + DTRACE_STORE(uintptr_t, tomax, + valoffs, typeref[3]); + valoffs += sizeof(uintptr_t); + + val = typeref[2]; + + for (s = 0; s < typs; s++) { + if (c != '\0') + c = dtrace_load8(val++); + + DTRACE_STORE(uint8_t, tomax, + valoffs++, c); + } + + /* + * Reset to the memory address rather than + * the typeref array, then let the BYREF + * code below do the work to store the + * memory data in the buffer. + */ + val = typeref[0]; + break; + } + + case DTRACEACT_CHILL: + if (dtrace_priv_kernel_destructive(state)) + dtrace_action_chill(&mstate, val); + continue; + + case DTRACEACT_RAISE: + if (dtrace_priv_proc_destructive(state)) + dtrace_action_raise(val); + continue; + + case DTRACEACT_COMMIT: + ASSERT(!committed); + + /* + * We need to commit our buffer state. + */ + if (ecb->dte_size) + buf->dtb_offset = offs + ecb->dte_size; + buf = &state->dts_buffer[cpuid]; + dtrace_speculation_commit(state, cpuid, val); + committed = 1; + continue; + + case DTRACEACT_DISCARD: + dtrace_speculation_discard(state, cpuid, val); + continue; + + case DTRACEACT_DIFEXPR: + case DTRACEACT_LIBACT: + case DTRACEACT_PRINTF: + case DTRACEACT_PRINTA: + case DTRACEACT_SYSTEM: + case DTRACEACT_FREOPEN: + break; + + case DTRACEACT_SYM: + case DTRACEACT_MOD: + if (!dtrace_priv_kernel(state)) + continue; + break; + + case DTRACEACT_USYM: + case DTRACEACT_UMOD: + case DTRACEACT_UADDR: { +#if defined(sun) + struct pid *pid = curthread->t_procp->p_pidp; +#endif + + if (!dtrace_priv_proc(state)) + continue; + + DTRACE_STORE(uint64_t, tomax, +#if defined(sun) + valoffs, (uint64_t)pid->pid_id); +#else + valoffs, (uint64_t) curproc->p_pid); +#endif + DTRACE_STORE(uint64_t, tomax, + valoffs + sizeof (uint64_t), val); + + continue; + } + + case DTRACEACT_EXIT: { + /* + * For the exit action, we are going to attempt + * to atomically set our activity to be + * draining. If this fails (either because + * another CPU has beat us to the exit action, + * or because our current activity is something + * other than ACTIVE or WARMUP), we will + * continue. This assures that the exit action + * can be successfully recorded at most once + * when we're in the ACTIVE state. If we're + * encountering the exit() action while in + * COOLDOWN, however, we want to honor the new + * status code. (We know that we're the only + * thread in COOLDOWN, so there is no race.) + */ + void *activity = &state->dts_activity; + dtrace_activity_t current = state->dts_activity; + + if (current == DTRACE_ACTIVITY_COOLDOWN) + break; + + if (current != DTRACE_ACTIVITY_WARMUP) + current = DTRACE_ACTIVITY_ACTIVE; + + if (dtrace_cas32(activity, current, + DTRACE_ACTIVITY_DRAINING) != current) { + *flags |= CPU_DTRACE_DROP; + continue; + } + + break; + } + + default: + ASSERT(0); + } + + if (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF) { + uintptr_t end = valoffs + size; + + if (!dtrace_vcanload((void *)(uintptr_t)val, + &dp->dtdo_rtype, &mstate, vstate)) + continue; + + /* + * If this is a string, we're going to only + * load until we find the zero byte -- after + * which we'll store zero bytes. + */ + if (dp->dtdo_rtype.dtdt_kind == + DIF_TYPE_STRING) { + char c = '\0' + 1; + int intuple = act->dta_intuple; + size_t s; + + for (s = 0; s < size; s++) { + if (c != '\0') + c = dtrace_load8(val++); + + DTRACE_STORE(uint8_t, tomax, + valoffs++, c); + + if (c == '\0' && intuple) + break; + } + + continue; + } + + while (valoffs < end) { + DTRACE_STORE(uint8_t, tomax, valoffs++, + dtrace_load8(val++)); + } + + continue; + } + + switch (size) { + case 0: + break; + + case sizeof (uint8_t): + DTRACE_STORE(uint8_t, tomax, valoffs, val); + break; + case sizeof (uint16_t): + DTRACE_STORE(uint16_t, tomax, valoffs, val); + break; + case sizeof (uint32_t): + DTRACE_STORE(uint32_t, tomax, valoffs, val); + break; + case sizeof (uint64_t): + DTRACE_STORE(uint64_t, tomax, valoffs, val); + break; + default: + /* + * Any other size should have been returned by + * reference, not by value. + */ + ASSERT(0); + break; + } + } + + if (*flags & CPU_DTRACE_DROP) + continue; + + if (*flags & CPU_DTRACE_FAULT) { + int ndx; + dtrace_action_t *err; + + buf->dtb_errors++; + + if (probe->dtpr_id == dtrace_probeid_error) { + /* + * There's nothing we can do -- we had an + * error on the error probe. We bump an + * error counter to at least indicate that + * this condition happened. + */ + dtrace_error(&state->dts_dblerrors); + continue; + } + + if (vtime) { + /* + * Before recursing on dtrace_probe(), we + * need to explicitly clear out our start + * time to prevent it from being accumulated + * into t_dtrace_vtime. + */ + curthread->t_dtrace_start = 0; + } + + /* + * Iterate over the actions to figure out which action + * we were processing when we experienced the error. + * Note that act points _past_ the faulting action; if + * act is ecb->dte_action, the fault was in the + * predicate, if it's ecb->dte_action->dta_next it's + * in action #1, and so on. + */ + for (err = ecb->dte_action, ndx = 0; + err != act; err = err->dta_next, ndx++) + continue; + + dtrace_probe_error(state, ecb->dte_epid, ndx, + (mstate.dtms_present & DTRACE_MSTATE_FLTOFFS) ? + mstate.dtms_fltoffs : -1, DTRACE_FLAGS2FLT(*flags), + cpu_core[cpuid].cpuc_dtrace_illval); + + continue; + } + + if (!committed) + buf->dtb_offset = offs + ecb->dte_size; + } + + if (vtime) + curthread->t_dtrace_start = dtrace_gethrtime(); + + dtrace_interrupt_enable(cookie); +} + +/* + * DTrace Probe Hashing Functions + * + * The functions in this section (and indeed, the functions in remaining + * sections) are not _called_ from probe context. (Any exceptions to this are + * marked with a "Note:".) Rather, they are called from elsewhere in the + * DTrace framework to look-up probes in, add probes to and remove probes from + * the DTrace probe hashes. (Each probe is hashed by each element of the + * probe tuple -- allowing for fast lookups, regardless of what was + * specified.) + */ +static uint_t +dtrace_hash_str(const char *p) +{ + unsigned int g; + uint_t hval = 0; + + while (*p) { + hval = (hval << 4) + *p++; + if ((g = (hval & 0xf0000000)) != 0) + hval ^= g >> 24; + hval &= ~g; + } + return (hval); +} + +static dtrace_hash_t * +dtrace_hash_create(uintptr_t stroffs, uintptr_t nextoffs, uintptr_t prevoffs) +{ + dtrace_hash_t *hash = kmem_zalloc(sizeof (dtrace_hash_t), KM_SLEEP); + + hash->dth_stroffs = stroffs; + hash->dth_nextoffs = nextoffs; + hash->dth_prevoffs = prevoffs; + + hash->dth_size = 1; + hash->dth_mask = hash->dth_size - 1; + + hash->dth_tab = kmem_zalloc(hash->dth_size * + sizeof (dtrace_hashbucket_t *), KM_SLEEP); + + return (hash); +} + +static void +dtrace_hash_destroy(dtrace_hash_t *hash) +{ +#ifdef DEBUG + int i; + + for (i = 0; i < hash->dth_size; i++) + ASSERT(hash->dth_tab[i] == NULL); +#endif + + kmem_free(hash->dth_tab, + hash->dth_size * sizeof (dtrace_hashbucket_t *)); + kmem_free(hash, sizeof (dtrace_hash_t)); +} + +static void +dtrace_hash_resize(dtrace_hash_t *hash) +{ + int size = hash->dth_size, i, ndx; + int new_size = hash->dth_size << 1; + int new_mask = new_size - 1; + dtrace_hashbucket_t **new_tab, *bucket, *next; + + ASSERT((new_size & new_mask) == 0); + + new_tab = kmem_zalloc(new_size * sizeof (void *), KM_SLEEP); + + for (i = 0; i < size; i++) { + for (bucket = hash->dth_tab[i]; bucket != NULL; bucket = next) { + dtrace_probe_t *probe = bucket->dthb_chain; + + ASSERT(probe != NULL); + ndx = DTRACE_HASHSTR(hash, probe) & new_mask; + + next = bucket->dthb_next; + bucket->dthb_next = new_tab[ndx]; + new_tab[ndx] = bucket; + } + } + + kmem_free(hash->dth_tab, hash->dth_size * sizeof (void *)); + hash->dth_tab = new_tab; + hash->dth_size = new_size; + hash->dth_mask = new_mask; +} + +static void +dtrace_hash_add(dtrace_hash_t *hash, dtrace_probe_t *new) +{ + int hashval = DTRACE_HASHSTR(hash, new); + int ndx = hashval & hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + dtrace_probe_t **nextp, **prevp; + + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, new)) + goto add; + } + + if ((hash->dth_nbuckets >> 1) > hash->dth_size) { + dtrace_hash_resize(hash); + dtrace_hash_add(hash, new); + return; + } + + bucket = kmem_zalloc(sizeof (dtrace_hashbucket_t), KM_SLEEP); + bucket->dthb_next = hash->dth_tab[ndx]; + hash->dth_tab[ndx] = bucket; + hash->dth_nbuckets++; + +add: + nextp = DTRACE_HASHNEXT(hash, new); + ASSERT(*nextp == NULL && *(DTRACE_HASHPREV(hash, new)) == NULL); + *nextp = bucket->dthb_chain; + + if (bucket->dthb_chain != NULL) { + prevp = DTRACE_HASHPREV(hash, bucket->dthb_chain); + ASSERT(*prevp == NULL); + *prevp = new; + } + + bucket->dthb_chain = new; + bucket->dthb_len++; +} + +static dtrace_probe_t * +dtrace_hash_lookup(dtrace_hash_t *hash, dtrace_probe_t *template) +{ + int hashval = DTRACE_HASHSTR(hash, template); + int ndx = hashval & hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template)) + return (bucket->dthb_chain); + } + + return (NULL); +} + +static int +dtrace_hash_collisions(dtrace_hash_t *hash, dtrace_probe_t *template) +{ + int hashval = DTRACE_HASHSTR(hash, template); + int ndx = hashval & hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template)) + return (bucket->dthb_len); + } + + return (0); +} + +static void +dtrace_hash_remove(dtrace_hash_t *hash, dtrace_probe_t *probe) +{ + int ndx = DTRACE_HASHSTR(hash, probe) & hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + + dtrace_probe_t **prevp = DTRACE_HASHPREV(hash, probe); + dtrace_probe_t **nextp = DTRACE_HASHNEXT(hash, probe); + + /* + * Find the bucket that we're removing this probe from. + */ + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, probe)) + break; + } + + ASSERT(bucket != NULL); + + if (*prevp == NULL) { + if (*nextp == NULL) { + /* + * The removed probe was the only probe on this + * bucket; we need to remove the bucket. + */ + dtrace_hashbucket_t *b = hash->dth_tab[ndx]; + + ASSERT(bucket->dthb_chain == probe); + ASSERT(b != NULL); + + if (b == bucket) { + hash->dth_tab[ndx] = bucket->dthb_next; + } else { + while (b->dthb_next != bucket) + b = b->dthb_next; + b->dthb_next = bucket->dthb_next; + } + + ASSERT(hash->dth_nbuckets > 0); + hash->dth_nbuckets--; + kmem_free(bucket, sizeof (dtrace_hashbucket_t)); + return; + } + + bucket->dthb_chain = *nextp; + } else { + *(DTRACE_HASHNEXT(hash, *prevp)) = *nextp; + } + + if (*nextp != NULL) + *(DTRACE_HASHPREV(hash, *nextp)) = *prevp; +} + +/* + * DTrace Utility Functions + * + * These are random utility functions that are _not_ called from probe context. + */ +static int +dtrace_badattr(const dtrace_attribute_t *a) +{ + return (a->dtat_name > DTRACE_STABILITY_MAX || + a->dtat_data > DTRACE_STABILITY_MAX || + a->dtat_class > DTRACE_CLASS_MAX); +} + +/* + * Return a duplicate copy of a string. If the specified string is NULL, + * this function returns a zero-length string. + */ +static char * +dtrace_strdup(const char *str) +{ + char *new = kmem_zalloc((str != NULL ? strlen(str) : 0) + 1, KM_SLEEP); + + if (str != NULL) + (void) strcpy(new, str); + + return (new); +} + +#define DTRACE_ISALPHA(c) \ + (((c) >= 'a' && (c) <= 'z') || ((c) >= 'A' && (c) <= 'Z')) + +static int +dtrace_badname(const char *s) +{ + char c; + + if (s == NULL || (c = *s++) == '\0') + return (0); + + if (!DTRACE_ISALPHA(c) && c != '-' && c != '_' && c != '.') + return (1); + + while ((c = *s++) != '\0') { + if (!DTRACE_ISALPHA(c) && (c < '0' || c > '9') && + c != '-' && c != '_' && c != '.' && c != '`') + return (1); + } + + return (0); +} + +static void +dtrace_cred2priv(cred_t *cr, uint32_t *privp, uid_t *uidp, zoneid_t *zoneidp) +{ + uint32_t priv; + +#if defined(sun) + if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) { + /* + * For DTRACE_PRIV_ALL, the uid and zoneid don't matter. + */ + priv = DTRACE_PRIV_ALL; + } else { + *uidp = crgetuid(cr); + *zoneidp = crgetzoneid(cr); + + priv = 0; + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) + priv |= DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER; + else if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) + priv |= DTRACE_PRIV_USER; + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) + priv |= DTRACE_PRIV_PROC; + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) + priv |= DTRACE_PRIV_OWNER; + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) + priv |= DTRACE_PRIV_ZONEOWNER; + } +#else + priv = DTRACE_PRIV_ALL; +#endif + + *privp = priv; +} + +#ifdef DTRACE_ERRDEBUG +static void +dtrace_errdebug(const char *str) +{ + int hval = dtrace_hash_str(str) % DTRACE_ERRHASHSZ; + int occupied = 0; + + mutex_enter(&dtrace_errlock); + dtrace_errlast = str; + dtrace_errthread = curthread; + + while (occupied++ < DTRACE_ERRHASHSZ) { + if (dtrace_errhash[hval].dter_msg == str) { + dtrace_errhash[hval].dter_count++; + goto out; + } + + if (dtrace_errhash[hval].dter_msg != NULL) { + hval = (hval + 1) % DTRACE_ERRHASHSZ; + continue; + } + + dtrace_errhash[hval].dter_msg = str; + dtrace_errhash[hval].dter_count = 1; + goto out; + } + + panic("dtrace: undersized error hash"); +out: + mutex_exit(&dtrace_errlock); +} +#endif + +/* + * DTrace Matching Functions + * + * These functions are used to match groups of probes, given some elements of + * a probe tuple, or some globbed expressions for elements of a probe tuple. + */ +static int +dtrace_match_priv(const dtrace_probe_t *prp, uint32_t priv, uid_t uid, + zoneid_t zoneid) +{ + if (priv != DTRACE_PRIV_ALL) { + uint32_t ppriv = prp->dtpr_provider->dtpv_priv.dtpp_flags; + uint32_t match = priv & ppriv; + + /* + * No PRIV_DTRACE_* privileges... + */ + if ((priv & (DTRACE_PRIV_PROC | DTRACE_PRIV_USER | + DTRACE_PRIV_KERNEL)) == 0) + return (0); + + /* + * No matching bits, but there were bits to match... + */ + if (match == 0 && ppriv != 0) + return (0); + + /* + * Need to have permissions to the process, but don't... + */ + if (((ppriv & ~match) & DTRACE_PRIV_OWNER) != 0 && + uid != prp->dtpr_provider->dtpv_priv.dtpp_uid) { + return (0); + } + + /* + * Need to be in the same zone unless we possess the + * privilege to examine all zones. + */ + if (((ppriv & ~match) & DTRACE_PRIV_ZONEOWNER) != 0 && + zoneid != prp->dtpr_provider->dtpv_priv.dtpp_zoneid) { + return (0); + } + } + + return (1); +} + +/* + * dtrace_match_probe compares a dtrace_probe_t to a pre-compiled key, which + * consists of input pattern strings and an ops-vector to evaluate them. + * This function returns >0 for match, 0 for no match, and <0 for error. + */ +static int +dtrace_match_probe(const dtrace_probe_t *prp, const dtrace_probekey_t *pkp, + uint32_t priv, uid_t uid, zoneid_t zoneid) +{ + dtrace_provider_t *pvp = prp->dtpr_provider; + int rv; + + if (pvp->dtpv_defunct) + return (0); + + if ((rv = pkp->dtpk_pmatch(pvp->dtpv_name, pkp->dtpk_prov, 0)) <= 0) + return (rv); + + if ((rv = pkp->dtpk_mmatch(prp->dtpr_mod, pkp->dtpk_mod, 0)) <= 0) + return (rv); + + if ((rv = pkp->dtpk_fmatch(prp->dtpr_func, pkp->dtpk_func, 0)) <= 0) + return (rv); + + if ((rv = pkp->dtpk_nmatch(prp->dtpr_name, pkp->dtpk_name, 0)) <= 0) + return (rv); + + if (dtrace_match_priv(prp, priv, uid, zoneid) == 0) + return (0); + + return (rv); +} + +/* + * dtrace_match_glob() is a safe kernel implementation of the gmatch(3GEN) + * interface for matching a glob pattern 'p' to an input string 's'. Unlike + * libc's version, the kernel version only applies to 8-bit ASCII strings. + * In addition, all of the recursion cases except for '*' matching have been + * unwound. For '*', we still implement recursive evaluation, but a depth + * counter is maintained and matching is aborted if we recurse too deep. + * The function returns 0 if no match, >0 if match, and <0 if recursion error. + */ +static int +dtrace_match_glob(const char *s, const char *p, int depth) +{ + const char *olds; + char s1, c; + int gs; + + if (depth > DTRACE_PROBEKEY_MAXDEPTH) + return (-1); + + if (s == NULL) + s = ""; /* treat NULL as empty string */ + +top: + olds = s; + s1 = *s++; + + if (p == NULL) + return (0); + + if ((c = *p++) == '\0') + return (s1 == '\0'); + + switch (c) { + case '[': { + int ok = 0, notflag = 0; + char lc = '\0'; + + if (s1 == '\0') + return (0); + + if (*p == '!') { + notflag = 1; + p++; + } + + if ((c = *p++) == '\0') + return (0); + + do { + if (c == '-' && lc != '\0' && *p != ']') { + if ((c = *p++) == '\0') + return (0); + if (c == '\\' && (c = *p++) == '\0') + return (0); + + if (notflag) { + if (s1 < lc || s1 > c) + ok++; + else + return (0); + } else if (lc <= s1 && s1 <= c) + ok++; + + } else if (c == '\\' && (c = *p++) == '\0') + return (0); + + lc = c; /* save left-hand 'c' for next iteration */ + + if (notflag) { + if (s1 != c) + ok++; + else + return (0); + } else if (s1 == c) + ok++; + + if ((c = *p++) == '\0') + return (0); + + } while (c != ']'); + + if (ok) + goto top; + + return (0); + } + + case '\\': + if ((c = *p++) == '\0') + return (0); + /*FALLTHRU*/ + + default: + if (c != s1) + return (0); + /*FALLTHRU*/ + + case '?': + if (s1 != '\0') + goto top; + return (0); + + case '*': + while (*p == '*') + p++; /* consecutive *'s are identical to a single one */ + + if (*p == '\0') + return (1); + + for (s = olds; *s != '\0'; s++) { + if ((gs = dtrace_match_glob(s, p, depth + 1)) != 0) + return (gs); + } + + return (0); + } +} + +/*ARGSUSED*/ +static int +dtrace_match_string(const char *s, const char *p, int depth) +{ + return (s != NULL && strcmp(s, p) == 0); +} + +/*ARGSUSED*/ +static int +dtrace_match_nul(const char *s, const char *p, int depth) +{ + return (1); /* always match the empty pattern */ +} + +/*ARGSUSED*/ +static int +dtrace_match_nonzero(const char *s, const char *p, int depth) +{ + return (s != NULL && s[0] != '\0'); +} + +static int +dtrace_match(const dtrace_probekey_t *pkp, uint32_t priv, uid_t uid, + zoneid_t zoneid, int (*matched)(dtrace_probe_t *, void *), void *arg) +{ + dtrace_probe_t template, *probe; + dtrace_hash_t *hash = NULL; + int len, best = INT_MAX, nmatched = 0; + dtrace_id_t i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + /* + * If the probe ID is specified in the key, just lookup by ID and + * invoke the match callback once if a matching probe is found. + */ + if (pkp->dtpk_id != DTRACE_IDNONE) { + if ((probe = dtrace_probe_lookup_id(pkp->dtpk_id)) != NULL && + dtrace_match_probe(probe, pkp, priv, uid, zoneid) > 0) { + (void) (*matched)(probe, arg); + nmatched++; + } + return (nmatched); + } + + template.dtpr_mod = (char *)pkp->dtpk_mod; + template.dtpr_func = (char *)pkp->dtpk_func; + template.dtpr_name = (char *)pkp->dtpk_name; + + /* + * We want to find the most distinct of the module name, function + * name, and name. So for each one that is not a glob pattern or + * empty string, we perform a lookup in the corresponding hash and + * use the hash table with the fewest collisions to do our search. + */ + if (pkp->dtpk_mmatch == &dtrace_match_string && + (len = dtrace_hash_collisions(dtrace_bymod, &template)) < best) { + best = len; + hash = dtrace_bymod; + } + + if (pkp->dtpk_fmatch == &dtrace_match_string && + (len = dtrace_hash_collisions(dtrace_byfunc, &template)) < best) { + best = len; + hash = dtrace_byfunc; + } + + if (pkp->dtpk_nmatch == &dtrace_match_string && + (len = dtrace_hash_collisions(dtrace_byname, &template)) < best) { + best = len; + hash = dtrace_byname; + } + + /* + * If we did not select a hash table, iterate over every probe and + * invoke our callback for each one that matches our input probe key. + */ + if (hash == NULL) { + for (i = 0; i < dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i]) == NULL || + dtrace_match_probe(probe, pkp, priv, uid, + zoneid) <= 0) + continue; + + nmatched++; + + if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT) + break; + } + + return (nmatched); + } + + /* + * If we selected a hash table, iterate over each probe of the same key + * name and invoke the callback for every probe that matches the other + * attributes of our input probe key. + */ + for (probe = dtrace_hash_lookup(hash, &template); probe != NULL; + probe = *(DTRACE_HASHNEXT(hash, probe))) { + + if (dtrace_match_probe(probe, pkp, priv, uid, zoneid) <= 0) + continue; + + nmatched++; + + if ((*matched)(probe, arg) != DTRACE_MATCH_NEXT) + break; + } + + return (nmatched); +} + +/* + * Return the function pointer dtrace_probecmp() should use to compare the + * specified pattern with a string. For NULL or empty patterns, we select + * dtrace_match_nul(). For glob pattern strings, we use dtrace_match_glob(). + * For non-empty non-glob strings, we use dtrace_match_string(). + */ +static dtrace_probekey_f * +dtrace_probekey_func(const char *p) +{ + char c; + + if (p == NULL || *p == '\0') + return (&dtrace_match_nul); + + while ((c = *p++) != '\0') { + if (c == '[' || c == '?' || c == '*' || c == '\\') + return (&dtrace_match_glob); + } + + return (&dtrace_match_string); +} + +/* + * Build a probe comparison key for use with dtrace_match_probe() from the + * given probe description. By convention, a null key only matches anchored + * probes: if each field is the empty string, reset dtpk_fmatch to + * dtrace_match_nonzero(). + */ +static void +dtrace_probekey(dtrace_probedesc_t *pdp, dtrace_probekey_t *pkp) +{ + pkp->dtpk_prov = pdp->dtpd_provider; + pkp->dtpk_pmatch = dtrace_probekey_func(pdp->dtpd_provider); + + pkp->dtpk_mod = pdp->dtpd_mod; + pkp->dtpk_mmatch = dtrace_probekey_func(pdp->dtpd_mod); + + pkp->dtpk_func = pdp->dtpd_func; + pkp->dtpk_fmatch = dtrace_probekey_func(pdp->dtpd_func); + + pkp->dtpk_name = pdp->dtpd_name; + pkp->dtpk_nmatch = dtrace_probekey_func(pdp->dtpd_name); + + pkp->dtpk_id = pdp->dtpd_id; + + if (pkp->dtpk_id == DTRACE_IDNONE && + pkp->dtpk_pmatch == &dtrace_match_nul && + pkp->dtpk_mmatch == &dtrace_match_nul && + pkp->dtpk_fmatch == &dtrace_match_nul && + pkp->dtpk_nmatch == &dtrace_match_nul) + pkp->dtpk_fmatch = &dtrace_match_nonzero; +} + +/* + * DTrace Provider-to-Framework API Functions + * + * These functions implement much of the Provider-to-Framework API, as + * described in <sys/dtrace.h>. The parts of the API not in this section are + * the functions in the API for probe management (found below), and + * dtrace_probe() itself (found above). + */ + +/* + * Register the calling provider with the DTrace framework. This should + * generally be called by DTrace providers in their attach(9E) entry point. + */ +int +dtrace_register(const char *name, const dtrace_pattr_t *pap, uint32_t priv, + cred_t *cr, const dtrace_pops_t *pops, void *arg, dtrace_provider_id_t *idp) +{ + dtrace_provider_t *provider; + + if (name == NULL || pap == NULL || pops == NULL || idp == NULL) { + cmn_err(CE_WARN, "failed to register provider '%s': invalid " + "arguments", name ? name : "<NULL>"); + return (EINVAL); + } + + if (name[0] == '\0' || dtrace_badname(name)) { + cmn_err(CE_WARN, "failed to register provider '%s': invalid " + "provider name", name); + return (EINVAL); + } + + if ((pops->dtps_provide == NULL && pops->dtps_provide_module == NULL) || + pops->dtps_enable == NULL || pops->dtps_disable == NULL || + pops->dtps_destroy == NULL || + ((pops->dtps_resume == NULL) != (pops->dtps_suspend == NULL))) { + cmn_err(CE_WARN, "failed to register provider '%s': invalid " + "provider ops", name); + return (EINVAL); + } + + if (dtrace_badattr(&pap->dtpa_provider) || + dtrace_badattr(&pap->dtpa_mod) || + dtrace_badattr(&pap->dtpa_func) || + dtrace_badattr(&pap->dtpa_name) || + dtrace_badattr(&pap->dtpa_args)) { + cmn_err(CE_WARN, "failed to register provider '%s': invalid " + "provider attributes", name); + return (EINVAL); + } + + if (priv & ~DTRACE_PRIV_ALL) { + cmn_err(CE_WARN, "failed to register provider '%s': invalid " + "privilege attributes", name); + return (EINVAL); + } + + if ((priv & DTRACE_PRIV_KERNEL) && + (priv & (DTRACE_PRIV_USER | DTRACE_PRIV_OWNER)) && + pops->dtps_usermode == NULL) { + cmn_err(CE_WARN, "failed to register provider '%s': need " + "dtps_usermode() op for given privilege attributes", name); + return (EINVAL); + } + + provider = kmem_zalloc(sizeof (dtrace_provider_t), KM_SLEEP); + provider->dtpv_name = kmem_alloc(strlen(name) + 1, KM_SLEEP); + (void) strcpy(provider->dtpv_name, name); + + provider->dtpv_attr = *pap; + provider->dtpv_priv.dtpp_flags = priv; + if (cr != NULL) { + provider->dtpv_priv.dtpp_uid = crgetuid(cr); + provider->dtpv_priv.dtpp_zoneid = crgetzoneid(cr); + } + provider->dtpv_pops = *pops; + + if (pops->dtps_provide == NULL) { + ASSERT(pops->dtps_provide_module != NULL); + provider->dtpv_pops.dtps_provide = + (void (*)(void *, dtrace_probedesc_t *))dtrace_nullop; + } + + if (pops->dtps_provide_module == NULL) { + ASSERT(pops->dtps_provide != NULL); + provider->dtpv_pops.dtps_provide_module = + (void (*)(void *, modctl_t *))dtrace_nullop; + } + + if (pops->dtps_suspend == NULL) { + ASSERT(pops->dtps_resume == NULL); + provider->dtpv_pops.dtps_suspend = + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop; + provider->dtpv_pops.dtps_resume = + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop; + } + + provider->dtpv_arg = arg; + *idp = (dtrace_provider_id_t)provider; + + if (pops == &dtrace_provider_ops) { + ASSERT(MUTEX_HELD(&dtrace_provider_lock)); + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dtrace_anon.dta_enabling == NULL); + + /* + * We make sure that the DTrace provider is at the head of + * the provider chain. + */ + provider->dtpv_next = dtrace_provider; + dtrace_provider = provider; + return (0); + } + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + /* + * If there is at least one provider registered, we'll add this + * provider after the first provider. + */ + if (dtrace_provider != NULL) { + provider->dtpv_next = dtrace_provider->dtpv_next; + dtrace_provider->dtpv_next = provider; + } else { + dtrace_provider = provider; + } + + if (dtrace_retained != NULL) { + dtrace_enabling_provide(provider); + + /* + * Now we need to call dtrace_enabling_matchall() -- which + * will acquire cpu_lock and dtrace_lock. We therefore need + * to drop all of our locks before calling into it... + */ + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + dtrace_enabling_matchall(); + + return (0); + } + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + return (0); +} + +/* + * Unregister the specified provider from the DTrace framework. This should + * generally be called by DTrace providers in their detach(9E) entry point. + */ +int +dtrace_unregister(dtrace_provider_id_t id) +{ + dtrace_provider_t *old = (dtrace_provider_t *)id; + dtrace_provider_t *prev = NULL; + int i, self = 0; + dtrace_probe_t *probe, *first = NULL; + + if (old->dtpv_pops.dtps_enable == + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop) { + /* + * If DTrace itself is the provider, we're called with locks + * already held. + */ + ASSERT(old == dtrace_provider); +#if defined(sun) + ASSERT(dtrace_devi != NULL); +#endif + ASSERT(MUTEX_HELD(&dtrace_provider_lock)); + ASSERT(MUTEX_HELD(&dtrace_lock)); + self = 1; + + if (dtrace_provider->dtpv_next != NULL) { + /* + * There's another provider here; return failure. + */ + return (EBUSY); + } + } else { + mutex_enter(&dtrace_provider_lock); + mutex_enter(&mod_lock); + mutex_enter(&dtrace_lock); + } + + /* + * If anyone has /dev/dtrace open, or if there are anonymous enabled + * probes, we refuse to let providers slither away, unless this + * provider has already been explicitly invalidated. + */ + if (!old->dtpv_defunct && + (dtrace_opens || (dtrace_anon.dta_state != NULL && + dtrace_anon.dta_state->dts_necbs > 0))) { + if (!self) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + } + return (EBUSY); + } + + /* + * Attempt to destroy the probes associated with this provider. + */ + for (i = 0; i < dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i]) == NULL) + continue; + + if (probe->dtpr_provider != old) + continue; + + if (probe->dtpr_ecb == NULL) + continue; + + /* + * We have at least one ECB; we can't remove this provider. + */ + if (!self) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + } + return (EBUSY); + } + + /* + * All of the probes for this provider are disabled; we can safely + * remove all of them from their hash chains and from the probe array. + */ + for (i = 0; i < dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i]) == NULL) + continue; + + if (probe->dtpr_provider != old) + continue; + + dtrace_probes[i] = NULL; + + dtrace_hash_remove(dtrace_bymod, probe); + dtrace_hash_remove(dtrace_byfunc, probe); + dtrace_hash_remove(dtrace_byname, probe); + + if (first == NULL) { + first = probe; + probe->dtpr_nextmod = NULL; + } else { + probe->dtpr_nextmod = first; + first = probe; + } + } + + /* + * The provider's probes have been removed from the hash chains and + * from the probe array. Now issue a dtrace_sync() to be sure that + * everyone has cleared out from any probe array processing. + */ + dtrace_sync(); + + for (probe = first; probe != NULL; probe = first) { + first = probe->dtpr_nextmod; + + old->dtpv_pops.dtps_destroy(old->dtpv_arg, probe->dtpr_id, + probe->dtpr_arg); + kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); + kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); + kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); +#if defined(sun) + vmem_free(dtrace_arena, (void *)(uintptr_t)(probe->dtpr_id), 1); +#else + free_unr(dtrace_arena, probe->dtpr_id); +#endif + kmem_free(probe, sizeof (dtrace_probe_t)); + } + + if ((prev = dtrace_provider) == old) { +#if defined(sun) + ASSERT(self || dtrace_devi == NULL); + ASSERT(old->dtpv_next == NULL || dtrace_devi == NULL); +#endif + dtrace_provider = old->dtpv_next; + } else { + while (prev != NULL && prev->dtpv_next != old) + prev = prev->dtpv_next; + + if (prev == NULL) { + panic("attempt to unregister non-existent " + "dtrace provider %p\n", (void *)id); + } + + prev->dtpv_next = old->dtpv_next; + } + + if (!self) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + } + + kmem_free(old->dtpv_name, strlen(old->dtpv_name) + 1); + kmem_free(old, sizeof (dtrace_provider_t)); + + return (0); +} + +/* + * Invalidate the specified provider. All subsequent probe lookups for the + * specified provider will fail, but its probes will not be removed. + */ +void +dtrace_invalidate(dtrace_provider_id_t id) +{ + dtrace_provider_t *pvp = (dtrace_provider_t *)id; + + ASSERT(pvp->dtpv_pops.dtps_enable != + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop); + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + pvp->dtpv_defunct = 1; + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); +} + +/* + * Indicate whether or not DTrace has attached. + */ +int +dtrace_attached(void) +{ + /* + * dtrace_provider will be non-NULL iff the DTrace driver has + * attached. (It's non-NULL because DTrace is always itself a + * provider.) + */ + return (dtrace_provider != NULL); +} + +/* + * Remove all the unenabled probes for the given provider. This function is + * not unlike dtrace_unregister(), except that it doesn't remove the provider + * -- just as many of its associated probes as it can. + */ +int +dtrace_condense(dtrace_provider_id_t id) +{ + dtrace_provider_t *prov = (dtrace_provider_t *)id; + int i; + dtrace_probe_t *probe; + + /* + * Make sure this isn't the dtrace provider itself. + */ + ASSERT(prov->dtpv_pops.dtps_enable != + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop); + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + /* + * Attempt to destroy the probes associated with this provider. + */ + for (i = 0; i < dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i]) == NULL) + continue; + + if (probe->dtpr_provider != prov) + continue; + + if (probe->dtpr_ecb != NULL) + continue; + + dtrace_probes[i] = NULL; + + dtrace_hash_remove(dtrace_bymod, probe); + dtrace_hash_remove(dtrace_byfunc, probe); + dtrace_hash_remove(dtrace_byname, probe); + + prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, i + 1, + probe->dtpr_arg); + kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); + kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); + kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); + kmem_free(probe, sizeof (dtrace_probe_t)); +#if defined(sun) + vmem_free(dtrace_arena, (void *)((uintptr_t)i + 1), 1); +#else + free_unr(dtrace_arena, i + 1); +#endif + } + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + return (0); +} + +/* + * DTrace Probe Management Functions + * + * The functions in this section perform the DTrace probe management, + * including functions to create probes, look-up probes, and call into the + * providers to request that probes be provided. Some of these functions are + * in the Provider-to-Framework API; these functions can be identified by the + * fact that they are not declared "static". + */ + +/* + * Create a probe with the specified module name, function name, and name. + */ +dtrace_id_t +dtrace_probe_create(dtrace_provider_id_t prov, const char *mod, + const char *func, const char *name, int aframes, void *arg) +{ + dtrace_probe_t *probe, **probes; + dtrace_provider_t *provider = (dtrace_provider_t *)prov; + dtrace_id_t id; + + if (provider == dtrace_provider) { + ASSERT(MUTEX_HELD(&dtrace_lock)); + } else { + mutex_enter(&dtrace_lock); + } + +#if defined(sun) + id = (dtrace_id_t)(uintptr_t)vmem_alloc(dtrace_arena, 1, + VM_BESTFIT | VM_SLEEP); +#else + id = alloc_unr(dtrace_arena); +#endif + probe = kmem_zalloc(sizeof (dtrace_probe_t), KM_SLEEP); + + probe->dtpr_id = id; + probe->dtpr_gen = dtrace_probegen++; + probe->dtpr_mod = dtrace_strdup(mod); + probe->dtpr_func = dtrace_strdup(func); + probe->dtpr_name = dtrace_strdup(name); + probe->dtpr_arg = arg; + probe->dtpr_aframes = aframes; + probe->dtpr_provider = provider; + + dtrace_hash_add(dtrace_bymod, probe); + dtrace_hash_add(dtrace_byfunc, probe); + dtrace_hash_add(dtrace_byname, probe); + + if (id - 1 >= dtrace_nprobes) { + size_t osize = dtrace_nprobes * sizeof (dtrace_probe_t *); + size_t nsize = osize << 1; + + if (nsize == 0) { + ASSERT(osize == 0); + ASSERT(dtrace_probes == NULL); + nsize = sizeof (dtrace_probe_t *); + } + + probes = kmem_zalloc(nsize, KM_SLEEP); + + if (dtrace_probes == NULL) { + ASSERT(osize == 0); + dtrace_probes = probes; + dtrace_nprobes = 1; + } else { + dtrace_probe_t **oprobes = dtrace_probes; + + bcopy(oprobes, probes, osize); + dtrace_membar_producer(); + dtrace_probes = probes; + + dtrace_sync(); + + /* + * All CPUs are now seeing the new probes array; we can + * safely free the old array. + */ + kmem_free(oprobes, osize); + dtrace_nprobes <<= 1; + } + + ASSERT(id - 1 < dtrace_nprobes); + } + + ASSERT(dtrace_probes[id - 1] == NULL); + dtrace_probes[id - 1] = probe; + + if (provider != dtrace_provider) + mutex_exit(&dtrace_lock); + + return (id); +} + +static dtrace_probe_t * +dtrace_probe_lookup_id(dtrace_id_t id) +{ + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (id == 0 || id > dtrace_nprobes) + return (NULL); + + return (dtrace_probes[id - 1]); +} + +static int +dtrace_probe_lookup_match(dtrace_probe_t *probe, void *arg) +{ + *((dtrace_id_t *)arg) = probe->dtpr_id; + + return (DTRACE_MATCH_DONE); +} + +/* + * Look up a probe based on provider and one or more of module name, function + * name and probe name. + */ +dtrace_id_t +dtrace_probe_lookup(dtrace_provider_id_t prid, char *mod, + char *func, char *name) +{ + dtrace_probekey_t pkey; + dtrace_id_t id; + int match; + + pkey.dtpk_prov = ((dtrace_provider_t *)prid)->dtpv_name; + pkey.dtpk_pmatch = &dtrace_match_string; + pkey.dtpk_mod = mod; + pkey.dtpk_mmatch = mod ? &dtrace_match_string : &dtrace_match_nul; + pkey.dtpk_func = func; + pkey.dtpk_fmatch = func ? &dtrace_match_string : &dtrace_match_nul; + pkey.dtpk_name = name; + pkey.dtpk_nmatch = name ? &dtrace_match_string : &dtrace_match_nul; + pkey.dtpk_id = DTRACE_IDNONE; + + mutex_enter(&dtrace_lock); + match = dtrace_match(&pkey, DTRACE_PRIV_ALL, 0, 0, + dtrace_probe_lookup_match, &id); + mutex_exit(&dtrace_lock); + + ASSERT(match == 1 || match == 0); + return (match ? id : 0); +} + +/* + * Returns the probe argument associated with the specified probe. + */ +void * +dtrace_probe_arg(dtrace_provider_id_t id, dtrace_id_t pid) +{ + dtrace_probe_t *probe; + void *rval = NULL; + + mutex_enter(&dtrace_lock); + + if ((probe = dtrace_probe_lookup_id(pid)) != NULL && + probe->dtpr_provider == (dtrace_provider_t *)id) + rval = probe->dtpr_arg; + + mutex_exit(&dtrace_lock); + + return (rval); +} + +/* + * Copy a probe into a probe description. + */ +static void +dtrace_probe_description(const dtrace_probe_t *prp, dtrace_probedesc_t *pdp) +{ + bzero(pdp, sizeof (dtrace_probedesc_t)); + pdp->dtpd_id = prp->dtpr_id; + + (void) strncpy(pdp->dtpd_provider, + prp->dtpr_provider->dtpv_name, DTRACE_PROVNAMELEN - 1); + + (void) strncpy(pdp->dtpd_mod, prp->dtpr_mod, DTRACE_MODNAMELEN - 1); + (void) strncpy(pdp->dtpd_func, prp->dtpr_func, DTRACE_FUNCNAMELEN - 1); + (void) strncpy(pdp->dtpd_name, prp->dtpr_name, DTRACE_NAMELEN - 1); +} + +#if !defined(sun) +static int +dtrace_probe_provide_cb(linker_file_t lf, void *arg) +{ + dtrace_provider_t *prv = (dtrace_provider_t *) arg; + + prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, lf); + + return(0); +} +#endif + + +/* + * Called to indicate that a probe -- or probes -- should be provided by a + * specfied provider. If the specified description is NULL, the provider will + * be told to provide all of its probes. (This is done whenever a new + * consumer comes along, or whenever a retained enabling is to be matched.) If + * the specified description is non-NULL, the provider is given the + * opportunity to dynamically provide the specified probe, allowing providers + * to support the creation of probes on-the-fly. (So-called _autocreated_ + * probes.) If the provider is NULL, the operations will be applied to all + * providers; if the provider is non-NULL the operations will only be applied + * to the specified provider. The dtrace_provider_lock must be held, and the + * dtrace_lock must _not_ be held -- the provider's dtps_provide() operation + * will need to grab the dtrace_lock when it reenters the framework through + * dtrace_probe_lookup(), dtrace_probe_create(), etc. + */ +static void +dtrace_probe_provide(dtrace_probedesc_t *desc, dtrace_provider_t *prv) +{ +#if defined(sun) + modctl_t *ctl; +#endif + int all = 0; + + ASSERT(MUTEX_HELD(&dtrace_provider_lock)); + + if (prv == NULL) { + all = 1; + prv = dtrace_provider; + } + + do { + /* + * First, call the blanket provide operation. + */ + prv->dtpv_pops.dtps_provide(prv->dtpv_arg, desc); + + /* + * Now call the per-module provide operation. We will grab + * mod_lock to prevent the list from being modified. Note + * that this also prevents the mod_busy bits from changing. + * (mod_busy can only be changed with mod_lock held.) + */ + mutex_enter(&mod_lock); + +#if defined(sun) + ctl = &modules; + do { + if (ctl->mod_busy || ctl->mod_mp == NULL) + continue; + + prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl); + + } while ((ctl = ctl->mod_next) != &modules); +#else + (void) linker_file_foreach(dtrace_probe_provide_cb, prv); +#endif + + mutex_exit(&mod_lock); + } while (all && (prv = prv->dtpv_next) != NULL); +} + +#if defined(sun) +/* + * Iterate over each probe, and call the Framework-to-Provider API function + * denoted by offs. + */ +static void +dtrace_probe_foreach(uintptr_t offs) +{ + dtrace_provider_t *prov; + void (*func)(void *, dtrace_id_t, void *); + dtrace_probe_t *probe; + dtrace_icookie_t cookie; + int i; + + /* + * We disable interrupts to walk through the probe array. This is + * safe -- the dtrace_sync() in dtrace_unregister() assures that we + * won't see stale data. + */ + cookie = dtrace_interrupt_disable(); + + for (i = 0; i < dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i]) == NULL) + continue; + + if (probe->dtpr_ecb == NULL) { + /* + * This probe isn't enabled -- don't call the function. + */ + continue; + } + + prov = probe->dtpr_provider; + func = *((void(**)(void *, dtrace_id_t, void *)) + ((uintptr_t)&prov->dtpv_pops + offs)); + + func(prov->dtpv_arg, i + 1, probe->dtpr_arg); + } + + dtrace_interrupt_enable(cookie); +} +#endif + +static int +dtrace_probe_enable(dtrace_probedesc_t *desc, dtrace_enabling_t *enab) +{ + dtrace_probekey_t pkey; + uint32_t priv; + uid_t uid; + zoneid_t zoneid; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + dtrace_ecb_create_cache = NULL; + + if (desc == NULL) { + /* + * If we're passed a NULL description, we're being asked to + * create an ECB with a NULL probe. + */ + (void) dtrace_ecb_create_enable(NULL, enab); + return (0); + } + + dtrace_probekey(desc, &pkey); + dtrace_cred2priv(enab->dten_vstate->dtvs_state->dts_cred.dcr_cred, + &priv, &uid, &zoneid); + + return (dtrace_match(&pkey, priv, uid, zoneid, dtrace_ecb_create_enable, + enab)); +} + +/* + * DTrace Helper Provider Functions + */ +static void +dtrace_dofattr2attr(dtrace_attribute_t *attr, const dof_attr_t dofattr) +{ + attr->dtat_name = DOF_ATTR_NAME(dofattr); + attr->dtat_data = DOF_ATTR_DATA(dofattr); + attr->dtat_class = DOF_ATTR_CLASS(dofattr); +} + +static void +dtrace_dofprov2hprov(dtrace_helper_provdesc_t *hprov, + const dof_provider_t *dofprov, char *strtab) +{ + hprov->dthpv_provname = strtab + dofprov->dofpv_name; + dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_provider, + dofprov->dofpv_provattr); + dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_mod, + dofprov->dofpv_modattr); + dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_func, + dofprov->dofpv_funcattr); + dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_name, + dofprov->dofpv_nameattr); + dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_args, + dofprov->dofpv_argsattr); +} + +static void +dtrace_helper_provide_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid) +{ + uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; + dof_hdr_t *dof = (dof_hdr_t *)daddr; + dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec; + dof_provider_t *provider; + dof_probe_t *probe; + uint32_t *off, *enoff; + uint8_t *arg; + char *strtab; + uint_t i, nprobes; + dtrace_helper_provdesc_t dhpv; + dtrace_helper_probedesc_t dhpb; + dtrace_meta_t *meta = dtrace_meta_pid; + dtrace_mops_t *mops = &meta->dtm_mops; + void *parg; + + provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); + str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_strtab * dof->dofh_secsize); + prb_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_probes * dof->dofh_secsize); + arg_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_prargs * dof->dofh_secsize); + off_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_proffs * dof->dofh_secsize); + + strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); + off = (uint32_t *)(uintptr_t)(daddr + off_sec->dofs_offset); + arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset); + enoff = NULL; + + /* + * See dtrace_helper_provider_validate(). + */ + if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && + provider->dofpv_prenoffs != DOF_SECT_NONE) { + enoff_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_prenoffs * dof->dofh_secsize); + enoff = (uint32_t *)(uintptr_t)(daddr + enoff_sec->dofs_offset); + } + + nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize; + + /* + * Create the provider. + */ + dtrace_dofprov2hprov(&dhpv, provider, strtab); + + if ((parg = mops->dtms_provide_pid(meta->dtm_arg, &dhpv, pid)) == NULL) + return; + + meta->dtm_count++; + + /* + * Create the probes. + */ + for (i = 0; i < nprobes; i++) { + probe = (dof_probe_t *)(uintptr_t)(daddr + + prb_sec->dofs_offset + i * prb_sec->dofs_entsize); + + dhpb.dthpb_mod = dhp->dofhp_mod; + dhpb.dthpb_func = strtab + probe->dofpr_func; + dhpb.dthpb_name = strtab + probe->dofpr_name; + dhpb.dthpb_base = probe->dofpr_addr; + dhpb.dthpb_offs = off + probe->dofpr_offidx; + dhpb.dthpb_noffs = probe->dofpr_noffs; + if (enoff != NULL) { + dhpb.dthpb_enoffs = enoff + probe->dofpr_enoffidx; + dhpb.dthpb_nenoffs = probe->dofpr_nenoffs; + } else { + dhpb.dthpb_enoffs = NULL; + dhpb.dthpb_nenoffs = 0; + } + dhpb.dthpb_args = arg + probe->dofpr_argidx; + dhpb.dthpb_nargc = probe->dofpr_nargc; + dhpb.dthpb_xargc = probe->dofpr_xargc; + dhpb.dthpb_ntypes = strtab + probe->dofpr_nargv; + dhpb.dthpb_xtypes = strtab + probe->dofpr_xargv; + + mops->dtms_create_probe(meta->dtm_arg, parg, &dhpb); + } +} + +static void +dtrace_helper_provide(dof_helper_t *dhp, pid_t pid) +{ + uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; + dof_hdr_t *dof = (dof_hdr_t *)daddr; + int i; + + ASSERT(MUTEX_HELD(&dtrace_meta_lock)); + + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + + dof->dofh_secoff + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_PROVIDER) + continue; + + dtrace_helper_provide_one(dhp, sec, pid); + } + + /* + * We may have just created probes, so we must now rematch against + * any retained enablings. Note that this call will acquire both + * cpu_lock and dtrace_lock; the fact that we are holding + * dtrace_meta_lock now is what defines the ordering with respect to + * these three locks. + */ + dtrace_enabling_matchall(); +} + +#if defined(sun) +static void +dtrace_helper_provider_remove_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid) +{ + uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; + dof_hdr_t *dof = (dof_hdr_t *)daddr; + dof_sec_t *str_sec; + dof_provider_t *provider; + char *strtab; + dtrace_helper_provdesc_t dhpv; + dtrace_meta_t *meta = dtrace_meta_pid; + dtrace_mops_t *mops = &meta->dtm_mops; + + provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); + str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_strtab * dof->dofh_secsize); + + strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); + + /* + * Create the provider. + */ + dtrace_dofprov2hprov(&dhpv, provider, strtab); + + mops->dtms_remove_pid(meta->dtm_arg, &dhpv, pid); + + meta->dtm_count--; +} + +static void +dtrace_helper_provider_remove(dof_helper_t *dhp, pid_t pid) +{ + uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; + dof_hdr_t *dof = (dof_hdr_t *)daddr; + int i; + + ASSERT(MUTEX_HELD(&dtrace_meta_lock)); + + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + + dof->dofh_secoff + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_PROVIDER) + continue; + + dtrace_helper_provider_remove_one(dhp, sec, pid); + } +} +#endif + +/* + * DTrace Meta Provider-to-Framework API Functions + * + * These functions implement the Meta Provider-to-Framework API, as described + * in <sys/dtrace.h>. + */ +int +dtrace_meta_register(const char *name, const dtrace_mops_t *mops, void *arg, + dtrace_meta_provider_id_t *idp) +{ + dtrace_meta_t *meta; + dtrace_helpers_t *help, *next; + int i; + + *idp = DTRACE_METAPROVNONE; + + /* + * We strictly don't need the name, but we hold onto it for + * debuggability. All hail error queues! + */ + if (name == NULL) { + cmn_err(CE_WARN, "failed to register meta-provider: " + "invalid name"); + return (EINVAL); + } + + if (mops == NULL || + mops->dtms_create_probe == NULL || + mops->dtms_provide_pid == NULL || + mops->dtms_remove_pid == NULL) { + cmn_err(CE_WARN, "failed to register meta-register %s: " + "invalid ops", name); + return (EINVAL); + } + + meta = kmem_zalloc(sizeof (dtrace_meta_t), KM_SLEEP); + meta->dtm_mops = *mops; + meta->dtm_name = kmem_alloc(strlen(name) + 1, KM_SLEEP); + (void) strcpy(meta->dtm_name, name); + meta->dtm_arg = arg; + + mutex_enter(&dtrace_meta_lock); + mutex_enter(&dtrace_lock); + + if (dtrace_meta_pid != NULL) { + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_meta_lock); + cmn_err(CE_WARN, "failed to register meta-register %s: " + "user-land meta-provider exists", name); + kmem_free(meta->dtm_name, strlen(meta->dtm_name) + 1); + kmem_free(meta, sizeof (dtrace_meta_t)); + return (EINVAL); + } + + dtrace_meta_pid = meta; + *idp = (dtrace_meta_provider_id_t)meta; + + /* + * If there are providers and probes ready to go, pass them + * off to the new meta provider now. + */ + + help = dtrace_deferred_pid; + dtrace_deferred_pid = NULL; + + mutex_exit(&dtrace_lock); + + while (help != NULL) { + for (i = 0; i < help->dthps_nprovs; i++) { + dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov, + help->dthps_pid); + } + + next = help->dthps_next; + help->dthps_next = NULL; + help->dthps_prev = NULL; + help->dthps_deferred = 0; + help = next; + } + + mutex_exit(&dtrace_meta_lock); + + return (0); +} + +int +dtrace_meta_unregister(dtrace_meta_provider_id_t id) +{ + dtrace_meta_t **pp, *old = (dtrace_meta_t *)id; + + mutex_enter(&dtrace_meta_lock); + mutex_enter(&dtrace_lock); + + if (old == dtrace_meta_pid) { + pp = &dtrace_meta_pid; + } else { + panic("attempt to unregister non-existent " + "dtrace meta-provider %p\n", (void *)old); + } + + if (old->dtm_count != 0) { + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_meta_lock); + return (EBUSY); + } + + *pp = NULL; + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_meta_lock); + + kmem_free(old->dtm_name, strlen(old->dtm_name) + 1); + kmem_free(old, sizeof (dtrace_meta_t)); + + return (0); +} + + +/* + * DTrace DIF Object Functions + */ +static int +dtrace_difo_err(uint_t pc, const char *format, ...) +{ + if (dtrace_err_verbose) { + va_list alist; + + (void) uprintf("dtrace DIF object error: [%u]: ", pc); + va_start(alist, format); + (void) vuprintf(format, alist); + va_end(alist); + } + +#ifdef DTRACE_ERRDEBUG + dtrace_errdebug(format); +#endif + return (1); +} + +/* + * Validate a DTrace DIF object by checking the IR instructions. The following + * rules are currently enforced by dtrace_difo_validate(): + * + * 1. Each instruction must have a valid opcode + * 2. Each register, string, variable, or subroutine reference must be valid + * 3. No instruction can modify register %r0 (must be zero) + * 4. All instruction reserved bits must be set to zero + * 5. The last instruction must be a "ret" instruction + * 6. All branch targets must reference a valid instruction _after_ the branch + */ +static int +dtrace_difo_validate(dtrace_difo_t *dp, dtrace_vstate_t *vstate, uint_t nregs, + cred_t *cr) +{ + int err = 0, i; + int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err; + int kcheckload; + uint_t pc; + + kcheckload = cr == NULL || + (vstate->dtvs_state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) == 0; + + dp->dtdo_destructive = 0; + + for (pc = 0; pc < dp->dtdo_len && err == 0; pc++) { + dif_instr_t instr = dp->dtdo_buf[pc]; + + uint_t r1 = DIF_INSTR_R1(instr); + uint_t r2 = DIF_INSTR_R2(instr); + uint_t rd = DIF_INSTR_RD(instr); + uint_t rs = DIF_INSTR_RS(instr); + uint_t label = DIF_INSTR_LABEL(instr); + uint_t v = DIF_INSTR_VAR(instr); + uint_t subr = DIF_INSTR_SUBR(instr); + uint_t type = DIF_INSTR_TYPE(instr); + uint_t op = DIF_INSTR_OP(instr); + + switch (op) { + case DIF_OP_OR: + case DIF_OP_XOR: + case DIF_OP_AND: + case DIF_OP_SLL: + case DIF_OP_SRL: + case DIF_OP_SRA: + case DIF_OP_SUB: + case DIF_OP_ADD: + case DIF_OP_MUL: + case DIF_OP_SDIV: + case DIF_OP_UDIV: + case DIF_OP_SREM: + case DIF_OP_UREM: + case DIF_OP_COPYS: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_NOT: + case DIF_OP_MOV: + case DIF_OP_ALLOCS: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_LDSB: + case DIF_OP_LDSH: + case DIF_OP_LDSW: + case DIF_OP_LDUB: + case DIF_OP_LDUH: + case DIF_OP_LDUW: + case DIF_OP_LDX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + if (kcheckload) + dp->dtdo_buf[pc] = DIF_INSTR_LOAD(op + + DIF_OP_RLDSB - DIF_OP_LDSB, r1, rd); + break; + case DIF_OP_RLDSB: + case DIF_OP_RLDSH: + case DIF_OP_RLDSW: + case DIF_OP_RLDUB: + case DIF_OP_RLDUH: + case DIF_OP_RLDUW: + case DIF_OP_RLDX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_ULDSB: + case DIF_OP_ULDSH: + case DIF_OP_ULDSW: + case DIF_OP_ULDUB: + case DIF_OP_ULDUH: + case DIF_OP_ULDUW: + case DIF_OP_ULDX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_STB: + case DIF_OP_STH: + case DIF_OP_STW: + case DIF_OP_STX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to 0 address\n"); + break; + case DIF_OP_CMP: + case DIF_OP_SCMP: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rd != 0) + err += efunc(pc, "non-zero reserved bits\n"); + break; + case DIF_OP_TST: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0 || rd != 0) + err += efunc(pc, "non-zero reserved bits\n"); + break; + case DIF_OP_BA: + case DIF_OP_BE: + case DIF_OP_BNE: + case DIF_OP_BG: + case DIF_OP_BGU: + case DIF_OP_BGE: + case DIF_OP_BGEU: + case DIF_OP_BL: + case DIF_OP_BLU: + case DIF_OP_BLE: + case DIF_OP_BLEU: + if (label >= dp->dtdo_len) { + err += efunc(pc, "invalid branch target %u\n", + label); + } + if (label <= pc) { + err += efunc(pc, "backward branch to %u\n", + label); + } + break; + case DIF_OP_RET: + if (r1 != 0 || r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + break; + case DIF_OP_NOP: + case DIF_OP_POPTS: + case DIF_OP_FLUSHTS: + if (r1 != 0 || r2 != 0 || rd != 0) + err += efunc(pc, "non-zero reserved bits\n"); + break; + case DIF_OP_SETX: + if (DIF_INSTR_INTEGER(instr) >= dp->dtdo_intlen) { + err += efunc(pc, "invalid integer ref %u\n", + DIF_INSTR_INTEGER(instr)); + } + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_SETS: + if (DIF_INSTR_STRING(instr) >= dp->dtdo_strlen) { + err += efunc(pc, "invalid string ref %u\n", + DIF_INSTR_STRING(instr)); + } + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_LDGA: + case DIF_OP_LDTA: + if (r1 > DIF_VAR_ARRAY_MAX) + err += efunc(pc, "invalid array %u\n", r1); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_LDGS: + case DIF_OP_LDTS: + case DIF_OP_LDLS: + case DIF_OP_LDGAA: + case DIF_OP_LDTAA: + if (v < DIF_VAR_OTHER_MIN || v > DIF_VAR_OTHER_MAX) + err += efunc(pc, "invalid variable %u\n", v); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_STGS: + case DIF_OP_STTS: + case DIF_OP_STLS: + case DIF_OP_STGAA: + case DIF_OP_STTAA: + if (v < DIF_VAR_OTHER_UBASE || v > DIF_VAR_OTHER_MAX) + err += efunc(pc, "invalid variable %u\n", v); + if (rs >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + break; + case DIF_OP_CALL: + if (subr > DIF_SUBR_MAX) + err += efunc(pc, "invalid subr %u\n", subr); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + + if (subr == DIF_SUBR_COPYOUT || + subr == DIF_SUBR_COPYOUTSTR) { + dp->dtdo_destructive = 1; + } + break; + case DIF_OP_PUSHTR: + if (type != DIF_TYPE_STRING && type != DIF_TYPE_CTF) + err += efunc(pc, "invalid ref type %u\n", type); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rs >= nregs) + err += efunc(pc, "invalid register %u\n", rs); + break; + case DIF_OP_PUSHTV: + if (type != DIF_TYPE_CTF) + err += efunc(pc, "invalid val type %u\n", type); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rs >= nregs) + err += efunc(pc, "invalid register %u\n", rs); + break; + default: + err += efunc(pc, "invalid opcode %u\n", + DIF_INSTR_OP(instr)); + } + } + + if (dp->dtdo_len != 0 && + DIF_INSTR_OP(dp->dtdo_buf[dp->dtdo_len - 1]) != DIF_OP_RET) { + err += efunc(dp->dtdo_len - 1, + "expected 'ret' as last DIF instruction\n"); + } + + if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) { + /* + * If we're not returning by reference, the size must be either + * 0 or the size of one of the base types. + */ + switch (dp->dtdo_rtype.dtdt_size) { + case 0: + case sizeof (uint8_t): + case sizeof (uint16_t): + case sizeof (uint32_t): + case sizeof (uint64_t): + break; + + default: + err += efunc(dp->dtdo_len - 1, "bad return size"); + } + } + + for (i = 0; i < dp->dtdo_varlen && err == 0; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i], *existing = NULL; + dtrace_diftype_t *vt, *et; + uint_t id, ndx; + + if (v->dtdv_scope != DIFV_SCOPE_GLOBAL && + v->dtdv_scope != DIFV_SCOPE_THREAD && + v->dtdv_scope != DIFV_SCOPE_LOCAL) { + err += efunc(i, "unrecognized variable scope %d\n", + v->dtdv_scope); + break; + } + + if (v->dtdv_kind != DIFV_KIND_ARRAY && + v->dtdv_kind != DIFV_KIND_SCALAR) { + err += efunc(i, "unrecognized variable type %d\n", + v->dtdv_kind); + break; + } + + if ((id = v->dtdv_id) > DIF_VARIABLE_MAX) { + err += efunc(i, "%d exceeds variable id limit\n", id); + break; + } + + if (id < DIF_VAR_OTHER_UBASE) + continue; + + /* + * For user-defined variables, we need to check that this + * definition is identical to any previous definition that we + * encountered. + */ + ndx = id - DIF_VAR_OTHER_UBASE; + + switch (v->dtdv_scope) { + case DIFV_SCOPE_GLOBAL: + if (ndx < vstate->dtvs_nglobals) { + dtrace_statvar_t *svar; + + if ((svar = vstate->dtvs_globals[ndx]) != NULL) + existing = &svar->dtsv_var; + } + + break; + + case DIFV_SCOPE_THREAD: + if (ndx < vstate->dtvs_ntlocals) + existing = &vstate->dtvs_tlocals[ndx]; + break; + + case DIFV_SCOPE_LOCAL: + if (ndx < vstate->dtvs_nlocals) { + dtrace_statvar_t *svar; + + if ((svar = vstate->dtvs_locals[ndx]) != NULL) + existing = &svar->dtsv_var; + } + + break; + } + + vt = &v->dtdv_type; + + if (vt->dtdt_flags & DIF_TF_BYREF) { + if (vt->dtdt_size == 0) { + err += efunc(i, "zero-sized variable\n"); + break; + } + + if (v->dtdv_scope == DIFV_SCOPE_GLOBAL && + vt->dtdt_size > dtrace_global_maxsize) { + err += efunc(i, "oversized by-ref global\n"); + break; + } + } + + if (existing == NULL || existing->dtdv_id == 0) + continue; + + ASSERT(existing->dtdv_id == v->dtdv_id); + ASSERT(existing->dtdv_scope == v->dtdv_scope); + + if (existing->dtdv_kind != v->dtdv_kind) + err += efunc(i, "%d changed variable kind\n", id); + + et = &existing->dtdv_type; + + if (vt->dtdt_flags != et->dtdt_flags) { + err += efunc(i, "%d changed variable type flags\n", id); + break; + } + + if (vt->dtdt_size != 0 && vt->dtdt_size != et->dtdt_size) { + err += efunc(i, "%d changed variable type size\n", id); + break; + } + } + + return (err); +} + +#if defined(sun) +/* + * Validate a DTrace DIF object that it is to be used as a helper. Helpers + * are much more constrained than normal DIFOs. Specifically, they may + * not: + * + * 1. Make calls to subroutines other than copyin(), copyinstr() or + * miscellaneous string routines + * 2. Access DTrace variables other than the args[] array, and the + * curthread, pid, ppid, tid, execname, zonename, uid and gid variables. + * 3. Have thread-local variables. + * 4. Have dynamic variables. + */ +static int +dtrace_difo_validate_helper(dtrace_difo_t *dp) +{ + int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err; + int err = 0; + uint_t pc; + + for (pc = 0; pc < dp->dtdo_len; pc++) { + dif_instr_t instr = dp->dtdo_buf[pc]; + + uint_t v = DIF_INSTR_VAR(instr); + uint_t subr = DIF_INSTR_SUBR(instr); + uint_t op = DIF_INSTR_OP(instr); + + switch (op) { + case DIF_OP_OR: + case DIF_OP_XOR: + case DIF_OP_AND: + case DIF_OP_SLL: + case DIF_OP_SRL: + case DIF_OP_SRA: + case DIF_OP_SUB: + case DIF_OP_ADD: + case DIF_OP_MUL: + case DIF_OP_SDIV: + case DIF_OP_UDIV: + case DIF_OP_SREM: + case DIF_OP_UREM: + case DIF_OP_COPYS: + case DIF_OP_NOT: + case DIF_OP_MOV: + case DIF_OP_RLDSB: + case DIF_OP_RLDSH: + case DIF_OP_RLDSW: + case DIF_OP_RLDUB: + case DIF_OP_RLDUH: + case DIF_OP_RLDUW: + case DIF_OP_RLDX: + case DIF_OP_ULDSB: + case DIF_OP_ULDSH: + case DIF_OP_ULDSW: + case DIF_OP_ULDUB: + case DIF_OP_ULDUH: + case DIF_OP_ULDUW: + case DIF_OP_ULDX: + case DIF_OP_STB: + case DIF_OP_STH: + case DIF_OP_STW: + case DIF_OP_STX: + case DIF_OP_ALLOCS: + case DIF_OP_CMP: + case DIF_OP_SCMP: + case DIF_OP_TST: + case DIF_OP_BA: + case DIF_OP_BE: + case DIF_OP_BNE: + case DIF_OP_BG: + case DIF_OP_BGU: + case DIF_OP_BGE: + case DIF_OP_BGEU: + case DIF_OP_BL: + case DIF_OP_BLU: + case DIF_OP_BLE: + case DIF_OP_BLEU: + case DIF_OP_RET: + case DIF_OP_NOP: + case DIF_OP_POPTS: + case DIF_OP_FLUSHTS: + case DIF_OP_SETX: + case DIF_OP_SETS: + case DIF_OP_LDGA: + case DIF_OP_LDLS: + case DIF_OP_STGS: + case DIF_OP_STLS: + case DIF_OP_PUSHTR: + case DIF_OP_PUSHTV: + break; + + case DIF_OP_LDGS: + if (v >= DIF_VAR_OTHER_UBASE) + break; + + if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) + break; + + if (v == DIF_VAR_CURTHREAD || v == DIF_VAR_PID || + v == DIF_VAR_PPID || v == DIF_VAR_TID || + v == DIF_VAR_EXECARGS || + v == DIF_VAR_EXECNAME || v == DIF_VAR_ZONENAME || + v == DIF_VAR_UID || v == DIF_VAR_GID) + break; + + err += efunc(pc, "illegal variable %u\n", v); + break; + + case DIF_OP_LDTA: + case DIF_OP_LDTS: + case DIF_OP_LDGAA: + case DIF_OP_LDTAA: + err += efunc(pc, "illegal dynamic variable load\n"); + break; + + case DIF_OP_STTS: + case DIF_OP_STGAA: + case DIF_OP_STTAA: + err += efunc(pc, "illegal dynamic variable store\n"); + break; + + case DIF_OP_CALL: + if (subr == DIF_SUBR_ALLOCA || + subr == DIF_SUBR_BCOPY || + subr == DIF_SUBR_COPYIN || + subr == DIF_SUBR_COPYINTO || + subr == DIF_SUBR_COPYINSTR || + subr == DIF_SUBR_INDEX || + subr == DIF_SUBR_INET_NTOA || + subr == DIF_SUBR_INET_NTOA6 || + subr == DIF_SUBR_INET_NTOP || + subr == DIF_SUBR_LLTOSTR || + subr == DIF_SUBR_RINDEX || + subr == DIF_SUBR_STRCHR || + subr == DIF_SUBR_STRJOIN || + subr == DIF_SUBR_STRRCHR || + subr == DIF_SUBR_STRSTR || + subr == DIF_SUBR_HTONS || + subr == DIF_SUBR_HTONL || + subr == DIF_SUBR_HTONLL || + subr == DIF_SUBR_NTOHS || + subr == DIF_SUBR_NTOHL || + subr == DIF_SUBR_NTOHLL || + subr == DIF_SUBR_MEMREF || + subr == DIF_SUBR_TYPEREF) + break; + + err += efunc(pc, "invalid subr %u\n", subr); + break; + + default: + err += efunc(pc, "invalid opcode %u\n", + DIF_INSTR_OP(instr)); + } + } + + return (err); +} +#endif + +/* + * Returns 1 if the expression in the DIF object can be cached on a per-thread + * basis; 0 if not. + */ +static int +dtrace_difo_cacheable(dtrace_difo_t *dp) +{ + int i; + + if (dp == NULL) + return (0); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_scope != DIFV_SCOPE_GLOBAL) + continue; + + switch (v->dtdv_id) { + case DIF_VAR_CURTHREAD: + case DIF_VAR_PID: + case DIF_VAR_TID: + case DIF_VAR_EXECARGS: + case DIF_VAR_EXECNAME: + case DIF_VAR_ZONENAME: + break; + + default: + return (0); + } + } + + /* + * This DIF object may be cacheable. Now we need to look for any + * array loading instructions, any memory loading instructions, or + * any stores to thread-local variables. + */ + for (i = 0; i < dp->dtdo_len; i++) { + uint_t op = DIF_INSTR_OP(dp->dtdo_buf[i]); + + if ((op >= DIF_OP_LDSB && op <= DIF_OP_LDX) || + (op >= DIF_OP_ULDSB && op <= DIF_OP_ULDX) || + (op >= DIF_OP_RLDSB && op <= DIF_OP_RLDX) || + op == DIF_OP_LDGA || op == DIF_OP_STTS) + return (0); + } + + return (1); +} + +static void +dtrace_difo_hold(dtrace_difo_t *dp) +{ + int i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + dp->dtdo_refcnt++; + ASSERT(dp->dtdo_refcnt != 0); + + /* + * We need to check this DIF object for references to the variable + * DIF_VAR_VTIMESTAMP. + */ + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_id != DIF_VAR_VTIMESTAMP) + continue; + + if (dtrace_vtime_references++ == 0) + dtrace_vtime_enable(); + } +} + +/* + * This routine calculates the dynamic variable chunksize for a given DIF + * object. The calculation is not fool-proof, and can probably be tricked by + * malicious DIF -- but it works for all compiler-generated DIF. Because this + * calculation is likely imperfect, dtrace_dynvar() is able to gracefully fail + * if a dynamic variable size exceeds the chunksize. + */ +static void +dtrace_difo_chunksize(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + uint64_t sval = 0; + dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */ + const dif_instr_t *text = dp->dtdo_buf; + uint_t pc, srd = 0; + uint_t ttop = 0; + size_t size, ksize; + uint_t id, i; + + for (pc = 0; pc < dp->dtdo_len; pc++) { + dif_instr_t instr = text[pc]; + uint_t op = DIF_INSTR_OP(instr); + uint_t rd = DIF_INSTR_RD(instr); + uint_t r1 = DIF_INSTR_R1(instr); + uint_t nkeys = 0; + uchar_t scope = 0; + + dtrace_key_t *key = tupregs; + + switch (op) { + case DIF_OP_SETX: + sval = dp->dtdo_inttab[DIF_INSTR_INTEGER(instr)]; + srd = rd; + continue; + + case DIF_OP_STTS: + key = &tupregs[DIF_DTR_NREGS]; + key[0].dttk_size = 0; + key[1].dttk_size = 0; + nkeys = 2; + scope = DIFV_SCOPE_THREAD; + break; + + case DIF_OP_STGAA: + case DIF_OP_STTAA: + nkeys = ttop; + + if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) + key[nkeys++].dttk_size = 0; + + key[nkeys++].dttk_size = 0; + + if (op == DIF_OP_STTAA) { + scope = DIFV_SCOPE_THREAD; + } else { + scope = DIFV_SCOPE_GLOBAL; + } + + break; + + case DIF_OP_PUSHTR: + if (ttop == DIF_DTR_NREGS) + return; + + if ((srd == 0 || sval == 0) && r1 == DIF_TYPE_STRING) { + /* + * If the register for the size of the "pushtr" + * is %r0 (or the value is 0) and the type is + * a string, we'll use the system-wide default + * string size. + */ + tupregs[ttop++].dttk_size = + dtrace_strsize_default; + } else { + if (srd == 0) + return; + + tupregs[ttop++].dttk_size = sval; + } + + break; + + case DIF_OP_PUSHTV: + if (ttop == DIF_DTR_NREGS) + return; + + tupregs[ttop++].dttk_size = 0; + break; + + case DIF_OP_FLUSHTS: + ttop = 0; + break; + + case DIF_OP_POPTS: + if (ttop != 0) + ttop--; + break; + } + + sval = 0; + srd = 0; + + if (nkeys == 0) + continue; + + /* + * We have a dynamic variable allocation; calculate its size. + */ + for (ksize = 0, i = 0; i < nkeys; i++) + ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t)); + + size = sizeof (dtrace_dynvar_t); + size += sizeof (dtrace_key_t) * (nkeys - 1); + size += ksize; + + /* + * Now we need to determine the size of the stored data. + */ + id = DIF_INSTR_VAR(instr); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_id == id && v->dtdv_scope == scope) { + size += v->dtdv_type.dtdt_size; + break; + } + } + + if (i == dp->dtdo_varlen) + return; + + /* + * We have the size. If this is larger than the chunk size + * for our dynamic variable state, reset the chunk size. + */ + size = P2ROUNDUP(size, sizeof (uint64_t)); + + if (size > vstate->dtvs_dynvars.dtds_chunksize) + vstate->dtvs_dynvars.dtds_chunksize = size; + } +} + +static void +dtrace_difo_init(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + int i, oldsvars, osz, nsz, otlocals, ntlocals; + uint_t id; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dp->dtdo_buf != NULL && dp->dtdo_len != 0); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + dtrace_statvar_t *svar, ***svarp = NULL; + size_t dsize = 0; + uint8_t scope = v->dtdv_scope; + int *np = NULL; + + if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE) + continue; + + id -= DIF_VAR_OTHER_UBASE; + + switch (scope) { + case DIFV_SCOPE_THREAD: + while (id >= (otlocals = vstate->dtvs_ntlocals)) { + dtrace_difv_t *tlocals; + + if ((ntlocals = (otlocals << 1)) == 0) + ntlocals = 1; + + osz = otlocals * sizeof (dtrace_difv_t); + nsz = ntlocals * sizeof (dtrace_difv_t); + + tlocals = kmem_zalloc(nsz, KM_SLEEP); + + if (osz != 0) { + bcopy(vstate->dtvs_tlocals, + tlocals, osz); + kmem_free(vstate->dtvs_tlocals, osz); + } + + vstate->dtvs_tlocals = tlocals; + vstate->dtvs_ntlocals = ntlocals; + } + + vstate->dtvs_tlocals[id] = *v; + continue; + + case DIFV_SCOPE_LOCAL: + np = &vstate->dtvs_nlocals; + svarp = &vstate->dtvs_locals; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) + dsize = NCPU * (v->dtdv_type.dtdt_size + + sizeof (uint64_t)); + else + dsize = NCPU * sizeof (uint64_t); + + break; + + case DIFV_SCOPE_GLOBAL: + np = &vstate->dtvs_nglobals; + svarp = &vstate->dtvs_globals; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) + dsize = v->dtdv_type.dtdt_size + + sizeof (uint64_t); + + break; + + default: + ASSERT(0); + } + + while (id >= (oldsvars = *np)) { + dtrace_statvar_t **statics; + int newsvars, oldsize, newsize; + + if ((newsvars = (oldsvars << 1)) == 0) + newsvars = 1; + + oldsize = oldsvars * sizeof (dtrace_statvar_t *); + newsize = newsvars * sizeof (dtrace_statvar_t *); + + statics = kmem_zalloc(newsize, KM_SLEEP); + + if (oldsize != 0) { + bcopy(*svarp, statics, oldsize); + kmem_free(*svarp, oldsize); + } + + *svarp = statics; + *np = newsvars; + } + + if ((svar = (*svarp)[id]) == NULL) { + svar = kmem_zalloc(sizeof (dtrace_statvar_t), KM_SLEEP); + svar->dtsv_var = *v; + + if ((svar->dtsv_size = dsize) != 0) { + svar->dtsv_data = (uint64_t)(uintptr_t) + kmem_zalloc(dsize, KM_SLEEP); + } + + (*svarp)[id] = svar; + } + + svar->dtsv_refcnt++; + } + + dtrace_difo_chunksize(dp, vstate); + dtrace_difo_hold(dp); +} + +#if defined(sun) +static dtrace_difo_t * +dtrace_difo_duplicate(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + dtrace_difo_t *new; + size_t sz; + + ASSERT(dp->dtdo_buf != NULL); + ASSERT(dp->dtdo_refcnt != 0); + + new = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP); + + ASSERT(dp->dtdo_buf != NULL); + sz = dp->dtdo_len * sizeof (dif_instr_t); + new->dtdo_buf = kmem_alloc(sz, KM_SLEEP); + bcopy(dp->dtdo_buf, new->dtdo_buf, sz); + new->dtdo_len = dp->dtdo_len; + + if (dp->dtdo_strtab != NULL) { + ASSERT(dp->dtdo_strlen != 0); + new->dtdo_strtab = kmem_alloc(dp->dtdo_strlen, KM_SLEEP); + bcopy(dp->dtdo_strtab, new->dtdo_strtab, dp->dtdo_strlen); + new->dtdo_strlen = dp->dtdo_strlen; + } + + if (dp->dtdo_inttab != NULL) { + ASSERT(dp->dtdo_intlen != 0); + sz = dp->dtdo_intlen * sizeof (uint64_t); + new->dtdo_inttab = kmem_alloc(sz, KM_SLEEP); + bcopy(dp->dtdo_inttab, new->dtdo_inttab, sz); + new->dtdo_intlen = dp->dtdo_intlen; + } + + if (dp->dtdo_vartab != NULL) { + ASSERT(dp->dtdo_varlen != 0); + sz = dp->dtdo_varlen * sizeof (dtrace_difv_t); + new->dtdo_vartab = kmem_alloc(sz, KM_SLEEP); + bcopy(dp->dtdo_vartab, new->dtdo_vartab, sz); + new->dtdo_varlen = dp->dtdo_varlen; + } + + dtrace_difo_init(new, vstate); + return (new); +} +#endif + +static void +dtrace_difo_destroy(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + int i; + + ASSERT(dp->dtdo_refcnt == 0); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + dtrace_statvar_t *svar, **svarp = NULL; + uint_t id; + uint8_t scope = v->dtdv_scope; + int *np = NULL; + + switch (scope) { + case DIFV_SCOPE_THREAD: + continue; + + case DIFV_SCOPE_LOCAL: + np = &vstate->dtvs_nlocals; + svarp = vstate->dtvs_locals; + break; + + case DIFV_SCOPE_GLOBAL: + np = &vstate->dtvs_nglobals; + svarp = vstate->dtvs_globals; + break; + + default: + ASSERT(0); + } + + if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE) + continue; + + id -= DIF_VAR_OTHER_UBASE; + ASSERT(id < *np); + + svar = svarp[id]; + ASSERT(svar != NULL); + ASSERT(svar->dtsv_refcnt > 0); + + if (--svar->dtsv_refcnt > 0) + continue; + + if (svar->dtsv_size != 0) { + ASSERT(svar->dtsv_data != 0); + kmem_free((void *)(uintptr_t)svar->dtsv_data, + svar->dtsv_size); + } + + kmem_free(svar, sizeof (dtrace_statvar_t)); + svarp[id] = NULL; + } + + if (dp->dtdo_buf != NULL) + kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t)); + if (dp->dtdo_inttab != NULL) + kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t)); + if (dp->dtdo_strtab != NULL) + kmem_free(dp->dtdo_strtab, dp->dtdo_strlen); + if (dp->dtdo_vartab != NULL) + kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t)); + + kmem_free(dp, sizeof (dtrace_difo_t)); +} + +static void +dtrace_difo_release(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + int i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dp->dtdo_refcnt != 0); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_id != DIF_VAR_VTIMESTAMP) + continue; + + ASSERT(dtrace_vtime_references > 0); + if (--dtrace_vtime_references == 0) + dtrace_vtime_disable(); + } + + if (--dp->dtdo_refcnt == 0) + dtrace_difo_destroy(dp, vstate); +} + +/* + * DTrace Format Functions + */ +static uint16_t +dtrace_format_add(dtrace_state_t *state, char *str) +{ + char *fmt, **new; + uint16_t ndx, len = strlen(str) + 1; + + fmt = kmem_zalloc(len, KM_SLEEP); + bcopy(str, fmt, len); + + for (ndx = 0; ndx < state->dts_nformats; ndx++) { + if (state->dts_formats[ndx] == NULL) { + state->dts_formats[ndx] = fmt; + return (ndx + 1); + } + } + + if (state->dts_nformats == USHRT_MAX) { + /* + * This is only likely if a denial-of-service attack is being + * attempted. As such, it's okay to fail silently here. + */ + kmem_free(fmt, len); + return (0); + } + + /* + * For simplicity, we always resize the formats array to be exactly the + * number of formats. + */ + ndx = state->dts_nformats++; + new = kmem_alloc((ndx + 1) * sizeof (char *), KM_SLEEP); + + if (state->dts_formats != NULL) { + ASSERT(ndx != 0); + bcopy(state->dts_formats, new, ndx * sizeof (char *)); + kmem_free(state->dts_formats, ndx * sizeof (char *)); + } + + state->dts_formats = new; + state->dts_formats[ndx] = fmt; + + return (ndx + 1); +} + +static void +dtrace_format_remove(dtrace_state_t *state, uint16_t format) +{ + char *fmt; + + ASSERT(state->dts_formats != NULL); + ASSERT(format <= state->dts_nformats); + ASSERT(state->dts_formats[format - 1] != NULL); + + fmt = state->dts_formats[format - 1]; + kmem_free(fmt, strlen(fmt) + 1); + state->dts_formats[format - 1] = NULL; +} + +static void +dtrace_format_destroy(dtrace_state_t *state) +{ + int i; + + if (state->dts_nformats == 0) { + ASSERT(state->dts_formats == NULL); + return; + } + + ASSERT(state->dts_formats != NULL); + + for (i = 0; i < state->dts_nformats; i++) { + char *fmt = state->dts_formats[i]; + + if (fmt == NULL) + continue; + + kmem_free(fmt, strlen(fmt) + 1); + } + + kmem_free(state->dts_formats, state->dts_nformats * sizeof (char *)); + state->dts_nformats = 0; + state->dts_formats = NULL; +} + +/* + * DTrace Predicate Functions + */ +static dtrace_predicate_t * +dtrace_predicate_create(dtrace_difo_t *dp) +{ + dtrace_predicate_t *pred; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dp->dtdo_refcnt != 0); + + pred = kmem_zalloc(sizeof (dtrace_predicate_t), KM_SLEEP); + pred->dtp_difo = dp; + pred->dtp_refcnt = 1; + + if (!dtrace_difo_cacheable(dp)) + return (pred); + + if (dtrace_predcache_id == DTRACE_CACHEIDNONE) { + /* + * This is only theoretically possible -- we have had 2^32 + * cacheable predicates on this machine. We cannot allow any + * more predicates to become cacheable: as unlikely as it is, + * there may be a thread caching a (now stale) predicate cache + * ID. (N.B.: the temptation is being successfully resisted to + * have this cmn_err() "Holy shit -- we executed this code!") + */ + return (pred); + } + + pred->dtp_cacheid = dtrace_predcache_id++; + + return (pred); +} + +static void +dtrace_predicate_hold(dtrace_predicate_t *pred) +{ + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(pred->dtp_difo != NULL && pred->dtp_difo->dtdo_refcnt != 0); + ASSERT(pred->dtp_refcnt > 0); + + pred->dtp_refcnt++; +} + +static void +dtrace_predicate_release(dtrace_predicate_t *pred, dtrace_vstate_t *vstate) +{ + dtrace_difo_t *dp = pred->dtp_difo; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dp != NULL && dp->dtdo_refcnt != 0); + ASSERT(pred->dtp_refcnt > 0); + + if (--pred->dtp_refcnt == 0) { + dtrace_difo_release(pred->dtp_difo, vstate); + kmem_free(pred, sizeof (dtrace_predicate_t)); + } +} + +/* + * DTrace Action Description Functions + */ +static dtrace_actdesc_t * +dtrace_actdesc_create(dtrace_actkind_t kind, uint32_t ntuple, + uint64_t uarg, uint64_t arg) +{ + dtrace_actdesc_t *act; + +#if defined(sun) + ASSERT(!DTRACEACT_ISPRINTFLIKE(kind) || (arg != NULL && + arg >= KERNELBASE) || (arg == NULL && kind == DTRACEACT_PRINTA)); +#endif + + act = kmem_zalloc(sizeof (dtrace_actdesc_t), KM_SLEEP); + act->dtad_kind = kind; + act->dtad_ntuple = ntuple; + act->dtad_uarg = uarg; + act->dtad_arg = arg; + act->dtad_refcnt = 1; + + return (act); +} + +static void +dtrace_actdesc_hold(dtrace_actdesc_t *act) +{ + ASSERT(act->dtad_refcnt >= 1); + act->dtad_refcnt++; +} + +static void +dtrace_actdesc_release(dtrace_actdesc_t *act, dtrace_vstate_t *vstate) +{ + dtrace_actkind_t kind = act->dtad_kind; + dtrace_difo_t *dp; + + ASSERT(act->dtad_refcnt >= 1); + + if (--act->dtad_refcnt != 0) + return; + + if ((dp = act->dtad_difo) != NULL) + dtrace_difo_release(dp, vstate); + + if (DTRACEACT_ISPRINTFLIKE(kind)) { + char *str = (char *)(uintptr_t)act->dtad_arg; + +#if defined(sun) + ASSERT((str != NULL && (uintptr_t)str >= KERNELBASE) || + (str == NULL && act->dtad_kind == DTRACEACT_PRINTA)); +#endif + + if (str != NULL) + kmem_free(str, strlen(str) + 1); + } + + kmem_free(act, sizeof (dtrace_actdesc_t)); +} + +/* + * DTrace ECB Functions + */ +static dtrace_ecb_t * +dtrace_ecb_add(dtrace_state_t *state, dtrace_probe_t *probe) +{ + dtrace_ecb_t *ecb; + dtrace_epid_t epid; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + ecb = kmem_zalloc(sizeof (dtrace_ecb_t), KM_SLEEP); + ecb->dte_predicate = NULL; + ecb->dte_probe = probe; + + /* + * The default size is the size of the default action: recording + * the epid. + */ + ecb->dte_size = ecb->dte_needed = sizeof (dtrace_epid_t); + ecb->dte_alignment = sizeof (dtrace_epid_t); + + epid = state->dts_epid++; + + if (epid - 1 >= state->dts_necbs) { + dtrace_ecb_t **oecbs = state->dts_ecbs, **ecbs; + int necbs = state->dts_necbs << 1; + + ASSERT(epid == state->dts_necbs + 1); + + if (necbs == 0) { + ASSERT(oecbs == NULL); + necbs = 1; + } + + ecbs = kmem_zalloc(necbs * sizeof (*ecbs), KM_SLEEP); + + if (oecbs != NULL) + bcopy(oecbs, ecbs, state->dts_necbs * sizeof (*ecbs)); + + dtrace_membar_producer(); + state->dts_ecbs = ecbs; + + if (oecbs != NULL) { + /* + * If this state is active, we must dtrace_sync() + * before we can free the old dts_ecbs array: we're + * coming in hot, and there may be active ring + * buffer processing (which indexes into the dts_ecbs + * array) on another CPU. + */ + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) + dtrace_sync(); + + kmem_free(oecbs, state->dts_necbs * sizeof (*ecbs)); + } + + dtrace_membar_producer(); + state->dts_necbs = necbs; + } + + ecb->dte_state = state; + + ASSERT(state->dts_ecbs[epid - 1] == NULL); + dtrace_membar_producer(); + state->dts_ecbs[(ecb->dte_epid = epid) - 1] = ecb; + + return (ecb); +} + +static void +dtrace_ecb_enable(dtrace_ecb_t *ecb) +{ + dtrace_probe_t *probe = ecb->dte_probe; + + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(ecb->dte_next == NULL); + + if (probe == NULL) { + /* + * This is the NULL probe -- there's nothing to do. + */ + return; + } + + if (probe->dtpr_ecb == NULL) { + dtrace_provider_t *prov = probe->dtpr_provider; + + /* + * We're the first ECB on this probe. + */ + probe->dtpr_ecb = probe->dtpr_ecb_last = ecb; + + if (ecb->dte_predicate != NULL) + probe->dtpr_predcache = ecb->dte_predicate->dtp_cacheid; + + prov->dtpv_pops.dtps_enable(prov->dtpv_arg, + probe->dtpr_id, probe->dtpr_arg); + } else { + /* + * This probe is already active. Swing the last pointer to + * point to the new ECB, and issue a dtrace_sync() to assure + * that all CPUs have seen the change. + */ + ASSERT(probe->dtpr_ecb_last != NULL); + probe->dtpr_ecb_last->dte_next = ecb; + probe->dtpr_ecb_last = ecb; + probe->dtpr_predcache = 0; + + dtrace_sync(); + } +} + +static void +dtrace_ecb_resize(dtrace_ecb_t *ecb) +{ + uint32_t maxalign = sizeof (dtrace_epid_t); + uint32_t align = sizeof (uint8_t), offs, diff; + dtrace_action_t *act; + int wastuple = 0; + uint32_t aggbase = UINT32_MAX; + dtrace_state_t *state = ecb->dte_state; + + /* + * If we record anything, we always record the epid. (And we always + * record it first.) + */ + offs = sizeof (dtrace_epid_t); + ecb->dte_size = ecb->dte_needed = sizeof (dtrace_epid_t); + + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + dtrace_recdesc_t *rec = &act->dta_rec; + + if ((align = rec->dtrd_alignment) > maxalign) + maxalign = align; + + if (!wastuple && act->dta_intuple) { + /* + * This is the first record in a tuple. Align the + * offset to be at offset 4 in an 8-byte aligned + * block. + */ + diff = offs + sizeof (dtrace_aggid_t); + + if ((diff = (diff & (sizeof (uint64_t) - 1)))) + offs += sizeof (uint64_t) - diff; + + aggbase = offs - sizeof (dtrace_aggid_t); + ASSERT(!(aggbase & (sizeof (uint64_t) - 1))); + } + + /*LINTED*/ + if (rec->dtrd_size != 0 && (diff = (offs & (align - 1)))) { + /* + * The current offset is not properly aligned; align it. + */ + offs += align - diff; + } + + rec->dtrd_offset = offs; + + if (offs + rec->dtrd_size > ecb->dte_needed) { + ecb->dte_needed = offs + rec->dtrd_size; + + if (ecb->dte_needed > state->dts_needed) + state->dts_needed = ecb->dte_needed; + } + + if (DTRACEACT_ISAGG(act->dta_kind)) { + dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act; + dtrace_action_t *first = agg->dtag_first, *prev; + + ASSERT(rec->dtrd_size != 0 && first != NULL); + ASSERT(wastuple); + ASSERT(aggbase != UINT32_MAX); + + agg->dtag_base = aggbase; + + while ((prev = first->dta_prev) != NULL && + DTRACEACT_ISAGG(prev->dta_kind)) { + agg = (dtrace_aggregation_t *)prev; + first = agg->dtag_first; + } + + if (prev != NULL) { + offs = prev->dta_rec.dtrd_offset + + prev->dta_rec.dtrd_size; + } else { + offs = sizeof (dtrace_epid_t); + } + wastuple = 0; + } else { + if (!act->dta_intuple) + ecb->dte_size = offs + rec->dtrd_size; + + offs += rec->dtrd_size; + } + + wastuple = act->dta_intuple; + } + + if ((act = ecb->dte_action) != NULL && + !(act->dta_kind == DTRACEACT_SPECULATE && act->dta_next == NULL) && + ecb->dte_size == sizeof (dtrace_epid_t)) { + /* + * If the size is still sizeof (dtrace_epid_t), then all + * actions store no data; set the size to 0. + */ + ecb->dte_alignment = maxalign; + ecb->dte_size = 0; + + /* + * If the needed space is still sizeof (dtrace_epid_t), then + * all actions need no additional space; set the needed + * size to 0. + */ + if (ecb->dte_needed == sizeof (dtrace_epid_t)) + ecb->dte_needed = 0; + + return; + } + + /* + * Set our alignment, and make sure that the dte_size and dte_needed + * are aligned to the size of an EPID. + */ + ecb->dte_alignment = maxalign; + ecb->dte_size = (ecb->dte_size + (sizeof (dtrace_epid_t) - 1)) & + ~(sizeof (dtrace_epid_t) - 1); + ecb->dte_needed = (ecb->dte_needed + (sizeof (dtrace_epid_t) - 1)) & + ~(sizeof (dtrace_epid_t) - 1); + ASSERT(ecb->dte_size <= ecb->dte_needed); +} + +static dtrace_action_t * +dtrace_ecb_aggregation_create(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc) +{ + dtrace_aggregation_t *agg; + size_t size = sizeof (uint64_t); + int ntuple = desc->dtad_ntuple; + dtrace_action_t *act; + dtrace_recdesc_t *frec; + dtrace_aggid_t aggid; + dtrace_state_t *state = ecb->dte_state; + + agg = kmem_zalloc(sizeof (dtrace_aggregation_t), KM_SLEEP); + agg->dtag_ecb = ecb; + + ASSERT(DTRACEACT_ISAGG(desc->dtad_kind)); + + switch (desc->dtad_kind) { + case DTRACEAGG_MIN: + agg->dtag_initial = INT64_MAX; + agg->dtag_aggregate = dtrace_aggregate_min; + break; + + case DTRACEAGG_MAX: + agg->dtag_initial = INT64_MIN; + agg->dtag_aggregate = dtrace_aggregate_max; + break; + + case DTRACEAGG_COUNT: + agg->dtag_aggregate = dtrace_aggregate_count; + break; + + case DTRACEAGG_QUANTIZE: + agg->dtag_aggregate = dtrace_aggregate_quantize; + size = (((sizeof (uint64_t) * NBBY) - 1) * 2 + 1) * + sizeof (uint64_t); + break; + + case DTRACEAGG_LQUANTIZE: { + uint16_t step = DTRACE_LQUANTIZE_STEP(desc->dtad_arg); + uint16_t levels = DTRACE_LQUANTIZE_LEVELS(desc->dtad_arg); + + agg->dtag_initial = desc->dtad_arg; + agg->dtag_aggregate = dtrace_aggregate_lquantize; + + if (step == 0 || levels == 0) + goto err; + + size = levels * sizeof (uint64_t) + 3 * sizeof (uint64_t); + break; + } + + case DTRACEAGG_AVG: + agg->dtag_aggregate = dtrace_aggregate_avg; + size = sizeof (uint64_t) * 2; + break; + + case DTRACEAGG_STDDEV: + agg->dtag_aggregate = dtrace_aggregate_stddev; + size = sizeof (uint64_t) * 4; + break; + + case DTRACEAGG_SUM: + agg->dtag_aggregate = dtrace_aggregate_sum; + break; + + default: + goto err; + } + + agg->dtag_action.dta_rec.dtrd_size = size; + + if (ntuple == 0) + goto err; + + /* + * We must make sure that we have enough actions for the n-tuple. + */ + for (act = ecb->dte_action_last; act != NULL; act = act->dta_prev) { + if (DTRACEACT_ISAGG(act->dta_kind)) + break; + + if (--ntuple == 0) { + /* + * This is the action with which our n-tuple begins. + */ + agg->dtag_first = act; + goto success; + } + } + + /* + * This n-tuple is short by ntuple elements. Return failure. + */ + ASSERT(ntuple != 0); +err: + kmem_free(agg, sizeof (dtrace_aggregation_t)); + return (NULL); + +success: + /* + * If the last action in the tuple has a size of zero, it's actually + * an expression argument for the aggregating action. + */ + ASSERT(ecb->dte_action_last != NULL); + act = ecb->dte_action_last; + + if (act->dta_kind == DTRACEACT_DIFEXPR) { + ASSERT(act->dta_difo != NULL); + + if (act->dta_difo->dtdo_rtype.dtdt_size == 0) + agg->dtag_hasarg = 1; + } + + /* + * We need to allocate an id for this aggregation. + */ +#if defined(sun) + aggid = (dtrace_aggid_t)(uintptr_t)vmem_alloc(state->dts_aggid_arena, 1, + VM_BESTFIT | VM_SLEEP); +#else + aggid = alloc_unr(state->dts_aggid_arena); +#endif + + if (aggid - 1 >= state->dts_naggregations) { + dtrace_aggregation_t **oaggs = state->dts_aggregations; + dtrace_aggregation_t **aggs; + int naggs = state->dts_naggregations << 1; + int onaggs = state->dts_naggregations; + + ASSERT(aggid == state->dts_naggregations + 1); + + if (naggs == 0) { + ASSERT(oaggs == NULL); + naggs = 1; + } + + aggs = kmem_zalloc(naggs * sizeof (*aggs), KM_SLEEP); + + if (oaggs != NULL) { + bcopy(oaggs, aggs, onaggs * sizeof (*aggs)); + kmem_free(oaggs, onaggs * sizeof (*aggs)); + } + + state->dts_aggregations = aggs; + state->dts_naggregations = naggs; + } + + ASSERT(state->dts_aggregations[aggid - 1] == NULL); + state->dts_aggregations[(agg->dtag_id = aggid) - 1] = agg; + + frec = &agg->dtag_first->dta_rec; + if (frec->dtrd_alignment < sizeof (dtrace_aggid_t)) + frec->dtrd_alignment = sizeof (dtrace_aggid_t); + + for (act = agg->dtag_first; act != NULL; act = act->dta_next) { + ASSERT(!act->dta_intuple); + act->dta_intuple = 1; + } + + return (&agg->dtag_action); +} + +static void +dtrace_ecb_aggregation_destroy(dtrace_ecb_t *ecb, dtrace_action_t *act) +{ + dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act; + dtrace_state_t *state = ecb->dte_state; + dtrace_aggid_t aggid = agg->dtag_id; + + ASSERT(DTRACEACT_ISAGG(act->dta_kind)); +#if defined(sun) + vmem_free(state->dts_aggid_arena, (void *)(uintptr_t)aggid, 1); +#else + free_unr(state->dts_aggid_arena, aggid); +#endif + + ASSERT(state->dts_aggregations[aggid - 1] == agg); + state->dts_aggregations[aggid - 1] = NULL; + + kmem_free(agg, sizeof (dtrace_aggregation_t)); +} + +static int +dtrace_ecb_action_add(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc) +{ + dtrace_action_t *action, *last; + dtrace_difo_t *dp = desc->dtad_difo; + uint32_t size = 0, align = sizeof (uint8_t), mask; + uint16_t format = 0; + dtrace_recdesc_t *rec; + dtrace_state_t *state = ecb->dte_state; + dtrace_optval_t *opt = state->dts_options, nframes = 0, strsize; + uint64_t arg = desc->dtad_arg; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(ecb->dte_action == NULL || ecb->dte_action->dta_refcnt == 1); + + if (DTRACEACT_ISAGG(desc->dtad_kind)) { + /* + * If this is an aggregating action, there must be neither + * a speculate nor a commit on the action chain. + */ + dtrace_action_t *act; + + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + if (act->dta_kind == DTRACEACT_COMMIT) + return (EINVAL); + + if (act->dta_kind == DTRACEACT_SPECULATE) + return (EINVAL); + } + + action = dtrace_ecb_aggregation_create(ecb, desc); + + if (action == NULL) + return (EINVAL); + } else { + if (DTRACEACT_ISDESTRUCTIVE(desc->dtad_kind) || + (desc->dtad_kind == DTRACEACT_DIFEXPR && + dp != NULL && dp->dtdo_destructive)) { + state->dts_destructive = 1; + } + + switch (desc->dtad_kind) { + case DTRACEACT_PRINTF: + case DTRACEACT_PRINTA: + case DTRACEACT_SYSTEM: + case DTRACEACT_FREOPEN: + /* + * We know that our arg is a string -- turn it into a + * format. + */ + if (arg == 0) { + ASSERT(desc->dtad_kind == DTRACEACT_PRINTA); + format = 0; + } else { + ASSERT(arg != 0); +#if defined(sun) + ASSERT(arg > KERNELBASE); +#endif + format = dtrace_format_add(state, + (char *)(uintptr_t)arg); + } + + /*FALLTHROUGH*/ + case DTRACEACT_LIBACT: + case DTRACEACT_DIFEXPR: + if (dp == NULL) + return (EINVAL); + + if ((size = dp->dtdo_rtype.dtdt_size) != 0) + break; + + if (dp->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) { + if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return (EINVAL); + + size = opt[DTRACEOPT_STRSIZE]; + } + + break; + + case DTRACEACT_STACK: + if ((nframes = arg) == 0) { + nframes = opt[DTRACEOPT_STACKFRAMES]; + ASSERT(nframes > 0); + arg = nframes; + } + + size = nframes * sizeof (pc_t); + break; + + case DTRACEACT_JSTACK: + if ((strsize = DTRACE_USTACK_STRSIZE(arg)) == 0) + strsize = opt[DTRACEOPT_JSTACKSTRSIZE]; + + if ((nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) + nframes = opt[DTRACEOPT_JSTACKFRAMES]; + + arg = DTRACE_USTACK_ARG(nframes, strsize); + + /*FALLTHROUGH*/ + case DTRACEACT_USTACK: + if (desc->dtad_kind != DTRACEACT_JSTACK && + (nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) { + strsize = DTRACE_USTACK_STRSIZE(arg); + nframes = opt[DTRACEOPT_USTACKFRAMES]; + ASSERT(nframes > 0); + arg = DTRACE_USTACK_ARG(nframes, strsize); + } + + /* + * Save a slot for the pid. + */ + size = (nframes + 1) * sizeof (uint64_t); + size += DTRACE_USTACK_STRSIZE(arg); + size = P2ROUNDUP(size, (uint32_t)(sizeof (uintptr_t))); + + break; + + case DTRACEACT_SYM: + case DTRACEACT_MOD: + if (dp == NULL || ((size = dp->dtdo_rtype.dtdt_size) != + sizeof (uint64_t)) || + (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return (EINVAL); + break; + + case DTRACEACT_USYM: + case DTRACEACT_UMOD: + case DTRACEACT_UADDR: + if (dp == NULL || + (dp->dtdo_rtype.dtdt_size != sizeof (uint64_t)) || + (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return (EINVAL); + + /* + * We have a slot for the pid, plus a slot for the + * argument. To keep things simple (aligned with + * bitness-neutral sizing), we store each as a 64-bit + * quantity. + */ + size = 2 * sizeof (uint64_t); + break; + + case DTRACEACT_STOP: + case DTRACEACT_BREAKPOINT: + case DTRACEACT_PANIC: + break; + + case DTRACEACT_CHILL: + case DTRACEACT_DISCARD: + case DTRACEACT_RAISE: + if (dp == NULL) + return (EINVAL); + break; + + case DTRACEACT_EXIT: + if (dp == NULL || + (size = dp->dtdo_rtype.dtdt_size) != sizeof (int) || + (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return (EINVAL); + break; + + case DTRACEACT_SPECULATE: + if (ecb->dte_size > sizeof (dtrace_epid_t)) + return (EINVAL); + + if (dp == NULL) + return (EINVAL); + + state->dts_speculates = 1; + break; + + case DTRACEACT_PRINTM: + size = dp->dtdo_rtype.dtdt_size; + break; + + case DTRACEACT_PRINTT: + size = dp->dtdo_rtype.dtdt_size; + break; + + case DTRACEACT_COMMIT: { + dtrace_action_t *act = ecb->dte_action; + + for (; act != NULL; act = act->dta_next) { + if (act->dta_kind == DTRACEACT_COMMIT) + return (EINVAL); + } + + if (dp == NULL) + return (EINVAL); + break; + } + + default: + return (EINVAL); + } + + if (size != 0 || desc->dtad_kind == DTRACEACT_SPECULATE) { + /* + * If this is a data-storing action or a speculate, + * we must be sure that there isn't a commit on the + * action chain. + */ + dtrace_action_t *act = ecb->dte_action; + + for (; act != NULL; act = act->dta_next) { + if (act->dta_kind == DTRACEACT_COMMIT) + return (EINVAL); + } + } + + action = kmem_zalloc(sizeof (dtrace_action_t), KM_SLEEP); + action->dta_rec.dtrd_size = size; + } + + action->dta_refcnt = 1; + rec = &action->dta_rec; + size = rec->dtrd_size; + + for (mask = sizeof (uint64_t) - 1; size != 0 && mask > 0; mask >>= 1) { + if (!(size & mask)) { + align = mask + 1; + break; + } + } + + action->dta_kind = desc->dtad_kind; + + if ((action->dta_difo = dp) != NULL) + dtrace_difo_hold(dp); + + rec->dtrd_action = action->dta_kind; + rec->dtrd_arg = arg; + rec->dtrd_uarg = desc->dtad_uarg; + rec->dtrd_alignment = (uint16_t)align; + rec->dtrd_format = format; + + if ((last = ecb->dte_action_last) != NULL) { + ASSERT(ecb->dte_action != NULL); + action->dta_prev = last; + last->dta_next = action; + } else { + ASSERT(ecb->dte_action == NULL); + ecb->dte_action = action; + } + + ecb->dte_action_last = action; + + return (0); +} + +static void +dtrace_ecb_action_remove(dtrace_ecb_t *ecb) +{ + dtrace_action_t *act = ecb->dte_action, *next; + dtrace_vstate_t *vstate = &ecb->dte_state->dts_vstate; + dtrace_difo_t *dp; + uint16_t format; + + if (act != NULL && act->dta_refcnt > 1) { + ASSERT(act->dta_next == NULL || act->dta_next->dta_refcnt == 1); + act->dta_refcnt--; + } else { + for (; act != NULL; act = next) { + next = act->dta_next; + ASSERT(next != NULL || act == ecb->dte_action_last); + ASSERT(act->dta_refcnt == 1); + + if ((format = act->dta_rec.dtrd_format) != 0) + dtrace_format_remove(ecb->dte_state, format); + + if ((dp = act->dta_difo) != NULL) + dtrace_difo_release(dp, vstate); + + if (DTRACEACT_ISAGG(act->dta_kind)) { + dtrace_ecb_aggregation_destroy(ecb, act); + } else { + kmem_free(act, sizeof (dtrace_action_t)); + } + } + } + + ecb->dte_action = NULL; + ecb->dte_action_last = NULL; + ecb->dte_size = sizeof (dtrace_epid_t); +} + +static void +dtrace_ecb_disable(dtrace_ecb_t *ecb) +{ + /* + * We disable the ECB by removing it from its probe. + */ + dtrace_ecb_t *pecb, *prev = NULL; + dtrace_probe_t *probe = ecb->dte_probe; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (probe == NULL) { + /* + * This is the NULL probe; there is nothing to disable. + */ + return; + } + + for (pecb = probe->dtpr_ecb; pecb != NULL; pecb = pecb->dte_next) { + if (pecb == ecb) + break; + prev = pecb; + } + + ASSERT(pecb != NULL); + + if (prev == NULL) { + probe->dtpr_ecb = ecb->dte_next; + } else { + prev->dte_next = ecb->dte_next; + } + + if (ecb == probe->dtpr_ecb_last) { + ASSERT(ecb->dte_next == NULL); + probe->dtpr_ecb_last = prev; + } + + /* + * The ECB has been disconnected from the probe; now sync to assure + * that all CPUs have seen the change before returning. + */ + dtrace_sync(); + + if (probe->dtpr_ecb == NULL) { + /* + * That was the last ECB on the probe; clear the predicate + * cache ID for the probe, disable it and sync one more time + * to assure that we'll never hit it again. + */ + dtrace_provider_t *prov = probe->dtpr_provider; + + ASSERT(ecb->dte_next == NULL); + ASSERT(probe->dtpr_ecb_last == NULL); + probe->dtpr_predcache = DTRACE_CACHEIDNONE; + prov->dtpv_pops.dtps_disable(prov->dtpv_arg, + probe->dtpr_id, probe->dtpr_arg); + dtrace_sync(); + } else { + /* + * There is at least one ECB remaining on the probe. If there + * is _exactly_ one, set the probe's predicate cache ID to be + * the predicate cache ID of the remaining ECB. + */ + ASSERT(probe->dtpr_ecb_last != NULL); + ASSERT(probe->dtpr_predcache == DTRACE_CACHEIDNONE); + + if (probe->dtpr_ecb == probe->dtpr_ecb_last) { + dtrace_predicate_t *p = probe->dtpr_ecb->dte_predicate; + + ASSERT(probe->dtpr_ecb->dte_next == NULL); + + if (p != NULL) + probe->dtpr_predcache = p->dtp_cacheid; + } + + ecb->dte_next = NULL; + } +} + +static void +dtrace_ecb_destroy(dtrace_ecb_t *ecb) +{ + dtrace_state_t *state = ecb->dte_state; + dtrace_vstate_t *vstate = &state->dts_vstate; + dtrace_predicate_t *pred; + dtrace_epid_t epid = ecb->dte_epid; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(ecb->dte_next == NULL); + ASSERT(ecb->dte_probe == NULL || ecb->dte_probe->dtpr_ecb != ecb); + + if ((pred = ecb->dte_predicate) != NULL) + dtrace_predicate_release(pred, vstate); + + dtrace_ecb_action_remove(ecb); + + ASSERT(state->dts_ecbs[epid - 1] == ecb); + state->dts_ecbs[epid - 1] = NULL; + + kmem_free(ecb, sizeof (dtrace_ecb_t)); +} + +static dtrace_ecb_t * +dtrace_ecb_create(dtrace_state_t *state, dtrace_probe_t *probe, + dtrace_enabling_t *enab) +{ + dtrace_ecb_t *ecb; + dtrace_predicate_t *pred; + dtrace_actdesc_t *act; + dtrace_provider_t *prov; + dtrace_ecbdesc_t *desc = enab->dten_current; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(state != NULL); + + ecb = dtrace_ecb_add(state, probe); + ecb->dte_uarg = desc->dted_uarg; + + if ((pred = desc->dted_pred.dtpdd_predicate) != NULL) { + dtrace_predicate_hold(pred); + ecb->dte_predicate = pred; + } + + if (probe != NULL) { + /* + * If the provider shows more leg than the consumer is old + * enough to see, we need to enable the appropriate implicit + * predicate bits to prevent the ecb from activating at + * revealing times. + * + * Providers specifying DTRACE_PRIV_USER at register time + * are stating that they need the /proc-style privilege + * model to be enforced, and this is what DTRACE_COND_OWNER + * and DTRACE_COND_ZONEOWNER will then do at probe time. + */ + prov = probe->dtpr_provider; + if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLPROC) && + (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER)) + ecb->dte_cond |= DTRACE_COND_OWNER; + + if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLZONE) && + (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER)) + ecb->dte_cond |= DTRACE_COND_ZONEOWNER; + + /* + * If the provider shows us kernel innards and the user + * is lacking sufficient privilege, enable the + * DTRACE_COND_USERMODE implicit predicate. + */ + if (!(state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) && + (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_KERNEL)) + ecb->dte_cond |= DTRACE_COND_USERMODE; + } + + if (dtrace_ecb_create_cache != NULL) { + /* + * If we have a cached ecb, we'll use its action list instead + * of creating our own (saving both time and space). + */ + dtrace_ecb_t *cached = dtrace_ecb_create_cache; + dtrace_action_t *act = cached->dte_action; + + if (act != NULL) { + ASSERT(act->dta_refcnt > 0); + act->dta_refcnt++; + ecb->dte_action = act; + ecb->dte_action_last = cached->dte_action_last; + ecb->dte_needed = cached->dte_needed; + ecb->dte_size = cached->dte_size; + ecb->dte_alignment = cached->dte_alignment; + } + + return (ecb); + } + + for (act = desc->dted_action; act != NULL; act = act->dtad_next) { + if ((enab->dten_error = dtrace_ecb_action_add(ecb, act)) != 0) { + dtrace_ecb_destroy(ecb); + return (NULL); + } + } + + dtrace_ecb_resize(ecb); + + return (dtrace_ecb_create_cache = ecb); +} + +static int +dtrace_ecb_create_enable(dtrace_probe_t *probe, void *arg) +{ + dtrace_ecb_t *ecb; + dtrace_enabling_t *enab = arg; + dtrace_state_t *state = enab->dten_vstate->dtvs_state; + + ASSERT(state != NULL); + + if (probe != NULL && probe->dtpr_gen < enab->dten_probegen) { + /* + * This probe was created in a generation for which this + * enabling has previously created ECBs; we don't want to + * enable it again, so just kick out. + */ + return (DTRACE_MATCH_NEXT); + } + + if ((ecb = dtrace_ecb_create(state, probe, enab)) == NULL) + return (DTRACE_MATCH_DONE); + + dtrace_ecb_enable(ecb); + return (DTRACE_MATCH_NEXT); +} + +static dtrace_ecb_t * +dtrace_epid2ecb(dtrace_state_t *state, dtrace_epid_t id) +{ + dtrace_ecb_t *ecb; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (id == 0 || id > state->dts_necbs) + return (NULL); + + ASSERT(state->dts_necbs > 0 && state->dts_ecbs != NULL); + ASSERT((ecb = state->dts_ecbs[id - 1]) == NULL || ecb->dte_epid == id); + + return (state->dts_ecbs[id - 1]); +} + +static dtrace_aggregation_t * +dtrace_aggid2agg(dtrace_state_t *state, dtrace_aggid_t id) +{ + dtrace_aggregation_t *agg; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (id == 0 || id > state->dts_naggregations) + return (NULL); + + ASSERT(state->dts_naggregations > 0 && state->dts_aggregations != NULL); + ASSERT((agg = state->dts_aggregations[id - 1]) == NULL || + agg->dtag_id == id); + + return (state->dts_aggregations[id - 1]); +} + +/* + * DTrace Buffer Functions + * + * The following functions manipulate DTrace buffers. Most of these functions + * are called in the context of establishing or processing consumer state; + * exceptions are explicitly noted. + */ + +/* + * Note: called from cross call context. This function switches the two + * buffers on a given CPU. The atomicity of this operation is assured by + * disabling interrupts while the actual switch takes place; the disabling of + * interrupts serializes the execution with any execution of dtrace_probe() on + * the same CPU. + */ +static void +dtrace_buffer_switch(dtrace_buffer_t *buf) +{ + caddr_t tomax = buf->dtb_tomax; + caddr_t xamot = buf->dtb_xamot; + dtrace_icookie_t cookie; + + ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); + ASSERT(!(buf->dtb_flags & DTRACEBUF_RING)); + + cookie = dtrace_interrupt_disable(); + buf->dtb_tomax = xamot; + buf->dtb_xamot = tomax; + buf->dtb_xamot_drops = buf->dtb_drops; + buf->dtb_xamot_offset = buf->dtb_offset; + buf->dtb_xamot_errors = buf->dtb_errors; + buf->dtb_xamot_flags = buf->dtb_flags; + buf->dtb_offset = 0; + buf->dtb_drops = 0; + buf->dtb_errors = 0; + buf->dtb_flags &= ~(DTRACEBUF_ERROR | DTRACEBUF_DROPPED); + dtrace_interrupt_enable(cookie); +} + +/* + * Note: called from cross call context. This function activates a buffer + * on a CPU. As with dtrace_buffer_switch(), the atomicity of the operation + * is guaranteed by the disabling of interrupts. + */ +static void +dtrace_buffer_activate(dtrace_state_t *state) +{ + dtrace_buffer_t *buf; + dtrace_icookie_t cookie = dtrace_interrupt_disable(); + + buf = &state->dts_buffer[curcpu]; + + if (buf->dtb_tomax != NULL) { + /* + * We might like to assert that the buffer is marked inactive, + * but this isn't necessarily true: the buffer for the CPU + * that processes the BEGIN probe has its buffer activated + * manually. In this case, we take the (harmless) action + * re-clearing the bit INACTIVE bit. + */ + buf->dtb_flags &= ~DTRACEBUF_INACTIVE; + } + + dtrace_interrupt_enable(cookie); +} + +static int +dtrace_buffer_alloc(dtrace_buffer_t *bufs, size_t size, int flags, + processorid_t cpu) +{ +#if defined(sun) + cpu_t *cp; +#else + struct pcpu *cp; +#endif + dtrace_buffer_t *buf; + +#if defined(sun) + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (size > dtrace_nonroot_maxsize && + !PRIV_POLICY_CHOICE(CRED(), PRIV_ALL, B_FALSE)) + return (EFBIG); + + cp = cpu_list; + + do { + if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id) + continue; + + buf = &bufs[cp->cpu_id]; + + /* + * If there is already a buffer allocated for this CPU, it + * is only possible that this is a DR event. In this case, + */ + if (buf->dtb_tomax != NULL) { + ASSERT(buf->dtb_size == size); + continue; + } + + ASSERT(buf->dtb_xamot == NULL); + + if ((buf->dtb_tomax = kmem_zalloc(size, KM_NOSLEEP)) == NULL) + goto err; + + buf->dtb_size = size; + buf->dtb_flags = flags; + buf->dtb_offset = 0; + buf->dtb_drops = 0; + + if (flags & DTRACEBUF_NOSWITCH) + continue; + + if ((buf->dtb_xamot = kmem_zalloc(size, KM_NOSLEEP)) == NULL) + goto err; + } while ((cp = cp->cpu_next) != cpu_list); + + return (0); + +err: + cp = cpu_list; + + do { + if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id) + continue; + + buf = &bufs[cp->cpu_id]; + + if (buf->dtb_xamot != NULL) { + ASSERT(buf->dtb_tomax != NULL); + ASSERT(buf->dtb_size == size); + kmem_free(buf->dtb_xamot, size); + } + + if (buf->dtb_tomax != NULL) { + ASSERT(buf->dtb_size == size); + kmem_free(buf->dtb_tomax, size); + } + + buf->dtb_tomax = NULL; + buf->dtb_xamot = NULL; + buf->dtb_size = 0; + } while ((cp = cp->cpu_next) != cpu_list); + + return (ENOMEM); +#else + int i; + +#if defined(__amd64__) + /* + * FreeBSD isn't good at limiting the amount of memory we + * ask to malloc, so let's place a limit here before trying + * to do something that might well end in tears at bedtime. + */ + if (size > physmem * PAGE_SIZE / (128 * (mp_maxid + 1))) + return(ENOMEM); +#endif + + ASSERT(MUTEX_HELD(&dtrace_lock)); + for (i = 0; i <= mp_maxid; i++) { + if ((cp = pcpu_find(i)) == NULL) + continue; + + if (cpu != DTRACE_CPUALL && cpu != i) + continue; + + buf = &bufs[i]; + + /* + * If there is already a buffer allocated for this CPU, it + * is only possible that this is a DR event. In this case, + * the buffer size must match our specified size. + */ + if (buf->dtb_tomax != NULL) { + ASSERT(buf->dtb_size == size); + continue; + } + + ASSERT(buf->dtb_xamot == NULL); + + if ((buf->dtb_tomax = kmem_zalloc(size, KM_NOSLEEP)) == NULL) + goto err; + + buf->dtb_size = size; + buf->dtb_flags = flags; + buf->dtb_offset = 0; + buf->dtb_drops = 0; + + if (flags & DTRACEBUF_NOSWITCH) + continue; + + if ((buf->dtb_xamot = kmem_zalloc(size, KM_NOSLEEP)) == NULL) + goto err; + } + + return (0); + +err: + /* + * Error allocating memory, so free the buffers that were + * allocated before the failed allocation. + */ + for (i = 0; i <= mp_maxid; i++) { + if ((cp = pcpu_find(i)) == NULL) + continue; + + if (cpu != DTRACE_CPUALL && cpu != i) + continue; + + buf = &bufs[i]; + + if (buf->dtb_xamot != NULL) { + ASSERT(buf->dtb_tomax != NULL); + ASSERT(buf->dtb_size == size); + kmem_free(buf->dtb_xamot, size); + } + + if (buf->dtb_tomax != NULL) { + ASSERT(buf->dtb_size == size); + kmem_free(buf->dtb_tomax, size); + } + + buf->dtb_tomax = NULL; + buf->dtb_xamot = NULL; + buf->dtb_size = 0; + + } + + return (ENOMEM); +#endif +} + +/* + * Note: called from probe context. This function just increments the drop + * count on a buffer. It has been made a function to allow for the + * possibility of understanding the source of mysterious drop counts. (A + * problem for which one may be particularly disappointed that DTrace cannot + * be used to understand DTrace.) + */ +static void +dtrace_buffer_drop(dtrace_buffer_t *buf) +{ + buf->dtb_drops++; +} + +/* + * Note: called from probe context. This function is called to reserve space + * in a buffer. If mstate is non-NULL, sets the scratch base and size in the + * mstate. Returns the new offset in the buffer, or a negative value if an + * error has occurred. + */ +static intptr_t +dtrace_buffer_reserve(dtrace_buffer_t *buf, size_t needed, size_t align, + dtrace_state_t *state, dtrace_mstate_t *mstate) +{ + intptr_t offs = buf->dtb_offset, soffs; + intptr_t woffs; + caddr_t tomax; + size_t total; + + if (buf->dtb_flags & DTRACEBUF_INACTIVE) + return (-1); + + if ((tomax = buf->dtb_tomax) == NULL) { + dtrace_buffer_drop(buf); + return (-1); + } + + if (!(buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL))) { + while (offs & (align - 1)) { + /* + * Assert that our alignment is off by a number which + * is itself sizeof (uint32_t) aligned. + */ + ASSERT(!((align - (offs & (align - 1))) & + (sizeof (uint32_t) - 1))); + DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE); + offs += sizeof (uint32_t); + } + + if ((soffs = offs + needed) > buf->dtb_size) { + dtrace_buffer_drop(buf); + return (-1); + } + + if (mstate == NULL) + return (offs); + + mstate->dtms_scratch_base = (uintptr_t)tomax + soffs; + mstate->dtms_scratch_size = buf->dtb_size - soffs; + mstate->dtms_scratch_ptr = mstate->dtms_scratch_base; + + return (offs); + } + + if (buf->dtb_flags & DTRACEBUF_FILL) { + if (state->dts_activity != DTRACE_ACTIVITY_COOLDOWN && + (buf->dtb_flags & DTRACEBUF_FULL)) + return (-1); + goto out; + } + + total = needed + (offs & (align - 1)); + + /* + * For a ring buffer, life is quite a bit more complicated. Before + * we can store any padding, we need to adjust our wrapping offset. + * (If we've never before wrapped or we're not about to, no adjustment + * is required.) + */ + if ((buf->dtb_flags & DTRACEBUF_WRAPPED) || + offs + total > buf->dtb_size) { + woffs = buf->dtb_xamot_offset; + + if (offs + total > buf->dtb_size) { + /* + * We can't fit in the end of the buffer. First, a + * sanity check that we can fit in the buffer at all. + */ + if (total > buf->dtb_size) { + dtrace_buffer_drop(buf); + return (-1); + } + + /* + * We're going to be storing at the top of the buffer, + * so now we need to deal with the wrapped offset. We + * only reset our wrapped offset to 0 if it is + * currently greater than the current offset. If it + * is less than the current offset, it is because a + * previous allocation induced a wrap -- but the + * allocation didn't subsequently take the space due + * to an error or false predicate evaluation. In this + * case, we'll just leave the wrapped offset alone: if + * the wrapped offset hasn't been advanced far enough + * for this allocation, it will be adjusted in the + * lower loop. + */ + if (buf->dtb_flags & DTRACEBUF_WRAPPED) { + if (woffs >= offs) + woffs = 0; + } else { + woffs = 0; + } + + /* + * Now we know that we're going to be storing to the + * top of the buffer and that there is room for us + * there. We need to clear the buffer from the current + * offset to the end (there may be old gunk there). + */ + while (offs < buf->dtb_size) + tomax[offs++] = 0; + + /* + * We need to set our offset to zero. And because we + * are wrapping, we need to set the bit indicating as + * much. We can also adjust our needed space back + * down to the space required by the ECB -- we know + * that the top of the buffer is aligned. + */ + offs = 0; + total = needed; + buf->dtb_flags |= DTRACEBUF_WRAPPED; + } else { + /* + * There is room for us in the buffer, so we simply + * need to check the wrapped offset. + */ + if (woffs < offs) { + /* + * The wrapped offset is less than the offset. + * This can happen if we allocated buffer space + * that induced a wrap, but then we didn't + * subsequently take the space due to an error + * or false predicate evaluation. This is + * okay; we know that _this_ allocation isn't + * going to induce a wrap. We still can't + * reset the wrapped offset to be zero, + * however: the space may have been trashed in + * the previous failed probe attempt. But at + * least the wrapped offset doesn't need to + * be adjusted at all... + */ + goto out; + } + } + + while (offs + total > woffs) { + dtrace_epid_t epid = *(uint32_t *)(tomax + woffs); + size_t size; + + if (epid == DTRACE_EPIDNONE) { + size = sizeof (uint32_t); + } else { + ASSERT(epid <= state->dts_necbs); + ASSERT(state->dts_ecbs[epid - 1] != NULL); + + size = state->dts_ecbs[epid - 1]->dte_size; + } + + ASSERT(woffs + size <= buf->dtb_size); + ASSERT(size != 0); + + if (woffs + size == buf->dtb_size) { + /* + * We've reached the end of the buffer; we want + * to set the wrapped offset to 0 and break + * out. However, if the offs is 0, then we're + * in a strange edge-condition: the amount of + * space that we want to reserve plus the size + * of the record that we're overwriting is + * greater than the size of the buffer. This + * is problematic because if we reserve the + * space but subsequently don't consume it (due + * to a failed predicate or error) the wrapped + * offset will be 0 -- yet the EPID at offset 0 + * will not be committed. This situation is + * relatively easy to deal with: if we're in + * this case, the buffer is indistinguishable + * from one that hasn't wrapped; we need only + * finish the job by clearing the wrapped bit, + * explicitly setting the offset to be 0, and + * zero'ing out the old data in the buffer. + */ + if (offs == 0) { + buf->dtb_flags &= ~DTRACEBUF_WRAPPED; + buf->dtb_offset = 0; + woffs = total; + + while (woffs < buf->dtb_size) + tomax[woffs++] = 0; + } + + woffs = 0; + break; + } + + woffs += size; + } + + /* + * We have a wrapped offset. It may be that the wrapped offset + * has become zero -- that's okay. + */ + buf->dtb_xamot_offset = woffs; + } + +out: + /* + * Now we can plow the buffer with any necessary padding. + */ + while (offs & (align - 1)) { + /* + * Assert that our alignment is off by a number which + * is itself sizeof (uint32_t) aligned. + */ + ASSERT(!((align - (offs & (align - 1))) & + (sizeof (uint32_t) - 1))); + DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE); + offs += sizeof (uint32_t); + } + + if (buf->dtb_flags & DTRACEBUF_FILL) { + if (offs + needed > buf->dtb_size - state->dts_reserve) { + buf->dtb_flags |= DTRACEBUF_FULL; + return (-1); + } + } + + if (mstate == NULL) + return (offs); + + /* + * For ring buffers and fill buffers, the scratch space is always + * the inactive buffer. + */ + mstate->dtms_scratch_base = (uintptr_t)buf->dtb_xamot; + mstate->dtms_scratch_size = buf->dtb_size; + mstate->dtms_scratch_ptr = mstate->dtms_scratch_base; + + return (offs); +} + +static void +dtrace_buffer_polish(dtrace_buffer_t *buf) +{ + ASSERT(buf->dtb_flags & DTRACEBUF_RING); + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (!(buf->dtb_flags & DTRACEBUF_WRAPPED)) + return; + + /* + * We need to polish the ring buffer. There are three cases: + * + * - The first (and presumably most common) is that there is no gap + * between the buffer offset and the wrapped offset. In this case, + * there is nothing in the buffer that isn't valid data; we can + * mark the buffer as polished and return. + * + * - The second (less common than the first but still more common + * than the third) is that there is a gap between the buffer offset + * and the wrapped offset, and the wrapped offset is larger than the + * buffer offset. This can happen because of an alignment issue, or + * can happen because of a call to dtrace_buffer_reserve() that + * didn't subsequently consume the buffer space. In this case, + * we need to zero the data from the buffer offset to the wrapped + * offset. + * + * - The third (and least common) is that there is a gap between the + * buffer offset and the wrapped offset, but the wrapped offset is + * _less_ than the buffer offset. This can only happen because a + * call to dtrace_buffer_reserve() induced a wrap, but the space + * was not subsequently consumed. In this case, we need to zero the + * space from the offset to the end of the buffer _and_ from the + * top of the buffer to the wrapped offset. + */ + if (buf->dtb_offset < buf->dtb_xamot_offset) { + bzero(buf->dtb_tomax + buf->dtb_offset, + buf->dtb_xamot_offset - buf->dtb_offset); + } + + if (buf->dtb_offset > buf->dtb_xamot_offset) { + bzero(buf->dtb_tomax + buf->dtb_offset, + buf->dtb_size - buf->dtb_offset); + bzero(buf->dtb_tomax, buf->dtb_xamot_offset); + } +} + +static void +dtrace_buffer_free(dtrace_buffer_t *bufs) +{ + int i; + + for (i = 0; i < NCPU; i++) { + dtrace_buffer_t *buf = &bufs[i]; + + if (buf->dtb_tomax == NULL) { + ASSERT(buf->dtb_xamot == NULL); + ASSERT(buf->dtb_size == 0); + continue; + } + + if (buf->dtb_xamot != NULL) { + ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); + kmem_free(buf->dtb_xamot, buf->dtb_size); + } + + kmem_free(buf->dtb_tomax, buf->dtb_size); + buf->dtb_size = 0; + buf->dtb_tomax = NULL; + buf->dtb_xamot = NULL; + } +} + +/* + * DTrace Enabling Functions + */ +static dtrace_enabling_t * +dtrace_enabling_create(dtrace_vstate_t *vstate) +{ + dtrace_enabling_t *enab; + + enab = kmem_zalloc(sizeof (dtrace_enabling_t), KM_SLEEP); + enab->dten_vstate = vstate; + + return (enab); +} + +static void +dtrace_enabling_add(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb) +{ + dtrace_ecbdesc_t **ndesc; + size_t osize, nsize; + + /* + * We can't add to enablings after we've enabled them, or after we've + * retained them. + */ + ASSERT(enab->dten_probegen == 0); + ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL); + + if (enab->dten_ndesc < enab->dten_maxdesc) { + enab->dten_desc[enab->dten_ndesc++] = ecb; + return; + } + + osize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *); + + if (enab->dten_maxdesc == 0) { + enab->dten_maxdesc = 1; + } else { + enab->dten_maxdesc <<= 1; + } + + ASSERT(enab->dten_ndesc < enab->dten_maxdesc); + + nsize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *); + ndesc = kmem_zalloc(nsize, KM_SLEEP); + bcopy(enab->dten_desc, ndesc, osize); + if (enab->dten_desc != NULL) + kmem_free(enab->dten_desc, osize); + + enab->dten_desc = ndesc; + enab->dten_desc[enab->dten_ndesc++] = ecb; +} + +static void +dtrace_enabling_addlike(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb, + dtrace_probedesc_t *pd) +{ + dtrace_ecbdesc_t *new; + dtrace_predicate_t *pred; + dtrace_actdesc_t *act; + + /* + * We're going to create a new ECB description that matches the + * specified ECB in every way, but has the specified probe description. + */ + new = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP); + + if ((pred = ecb->dted_pred.dtpdd_predicate) != NULL) + dtrace_predicate_hold(pred); + + for (act = ecb->dted_action; act != NULL; act = act->dtad_next) + dtrace_actdesc_hold(act); + + new->dted_action = ecb->dted_action; + new->dted_pred = ecb->dted_pred; + new->dted_probe = *pd; + new->dted_uarg = ecb->dted_uarg; + + dtrace_enabling_add(enab, new); +} + +static void +dtrace_enabling_dump(dtrace_enabling_t *enab) +{ + int i; + + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_probedesc_t *desc = &enab->dten_desc[i]->dted_probe; + + cmn_err(CE_NOTE, "enabling probe %d (%s:%s:%s:%s)", i, + desc->dtpd_provider, desc->dtpd_mod, + desc->dtpd_func, desc->dtpd_name); + } +} + +static void +dtrace_enabling_destroy(dtrace_enabling_t *enab) +{ + int i; + dtrace_ecbdesc_t *ep; + dtrace_vstate_t *vstate = enab->dten_vstate; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_actdesc_t *act, *next; + dtrace_predicate_t *pred; + + ep = enab->dten_desc[i]; + + if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) + dtrace_predicate_release(pred, vstate); + + for (act = ep->dted_action; act != NULL; act = next) { + next = act->dtad_next; + dtrace_actdesc_release(act, vstate); + } + + kmem_free(ep, sizeof (dtrace_ecbdesc_t)); + } + + if (enab->dten_desc != NULL) + kmem_free(enab->dten_desc, + enab->dten_maxdesc * sizeof (dtrace_enabling_t *)); + + /* + * If this was a retained enabling, decrement the dts_nretained count + * and take it off of the dtrace_retained list. + */ + if (enab->dten_prev != NULL || enab->dten_next != NULL || + dtrace_retained == enab) { + ASSERT(enab->dten_vstate->dtvs_state != NULL); + ASSERT(enab->dten_vstate->dtvs_state->dts_nretained > 0); + enab->dten_vstate->dtvs_state->dts_nretained--; + } + + if (enab->dten_prev == NULL) { + if (dtrace_retained == enab) { + dtrace_retained = enab->dten_next; + + if (dtrace_retained != NULL) + dtrace_retained->dten_prev = NULL; + } + } else { + ASSERT(enab != dtrace_retained); + ASSERT(dtrace_retained != NULL); + enab->dten_prev->dten_next = enab->dten_next; + } + + if (enab->dten_next != NULL) { + ASSERT(dtrace_retained != NULL); + enab->dten_next->dten_prev = enab->dten_prev; + } + + kmem_free(enab, sizeof (dtrace_enabling_t)); +} + +static int +dtrace_enabling_retain(dtrace_enabling_t *enab) +{ + dtrace_state_t *state; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL); + ASSERT(enab->dten_vstate != NULL); + + state = enab->dten_vstate->dtvs_state; + ASSERT(state != NULL); + + /* + * We only allow each state to retain dtrace_retain_max enablings. + */ + if (state->dts_nretained >= dtrace_retain_max) + return (ENOSPC); + + state->dts_nretained++; + + if (dtrace_retained == NULL) { + dtrace_retained = enab; + return (0); + } + + enab->dten_next = dtrace_retained; + dtrace_retained->dten_prev = enab; + dtrace_retained = enab; + + return (0); +} + +static int +dtrace_enabling_replicate(dtrace_state_t *state, dtrace_probedesc_t *match, + dtrace_probedesc_t *create) +{ + dtrace_enabling_t *new, *enab; + int found = 0, err = ENOENT; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(strlen(match->dtpd_provider) < DTRACE_PROVNAMELEN); + ASSERT(strlen(match->dtpd_mod) < DTRACE_MODNAMELEN); + ASSERT(strlen(match->dtpd_func) < DTRACE_FUNCNAMELEN); + ASSERT(strlen(match->dtpd_name) < DTRACE_NAMELEN); + + new = dtrace_enabling_create(&state->dts_vstate); + + /* + * Iterate over all retained enablings, looking for enablings that + * match the specified state. + */ + for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { + int i; + + /* + * dtvs_state can only be NULL for helper enablings -- and + * helper enablings can't be retained. + */ + ASSERT(enab->dten_vstate->dtvs_state != NULL); + + if (enab->dten_vstate->dtvs_state != state) + continue; + + /* + * Now iterate over each probe description; we're looking for + * an exact match to the specified probe description. + */ + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_ecbdesc_t *ep = enab->dten_desc[i]; + dtrace_probedesc_t *pd = &ep->dted_probe; + + if (strcmp(pd->dtpd_provider, match->dtpd_provider)) + continue; + + if (strcmp(pd->dtpd_mod, match->dtpd_mod)) + continue; + + if (strcmp(pd->dtpd_func, match->dtpd_func)) + continue; + + if (strcmp(pd->dtpd_name, match->dtpd_name)) + continue; + + /* + * We have a winning probe! Add it to our growing + * enabling. + */ + found = 1; + dtrace_enabling_addlike(new, ep, create); + } + } + + if (!found || (err = dtrace_enabling_retain(new)) != 0) { + dtrace_enabling_destroy(new); + return (err); + } + + return (0); +} + +static void +dtrace_enabling_retract(dtrace_state_t *state) +{ + dtrace_enabling_t *enab, *next; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + /* + * Iterate over all retained enablings, destroy the enablings retained + * for the specified state. + */ + for (enab = dtrace_retained; enab != NULL; enab = next) { + next = enab->dten_next; + + /* + * dtvs_state can only be NULL for helper enablings -- and + * helper enablings can't be retained. + */ + ASSERT(enab->dten_vstate->dtvs_state != NULL); + + if (enab->dten_vstate->dtvs_state == state) { + ASSERT(state->dts_nretained > 0); + dtrace_enabling_destroy(enab); + } + } + + ASSERT(state->dts_nretained == 0); +} + +static int +dtrace_enabling_match(dtrace_enabling_t *enab, int *nmatched) +{ + int i = 0; + int matched = 0; + + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(MUTEX_HELD(&dtrace_lock)); + + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_ecbdesc_t *ep = enab->dten_desc[i]; + + enab->dten_current = ep; + enab->dten_error = 0; + + matched += dtrace_probe_enable(&ep->dted_probe, enab); + + if (enab->dten_error != 0) { + /* + * If we get an error half-way through enabling the + * probes, we kick out -- perhaps with some number of + * them enabled. Leaving enabled probes enabled may + * be slightly confusing for user-level, but we expect + * that no one will attempt to actually drive on in + * the face of such errors. If this is an anonymous + * enabling (indicated with a NULL nmatched pointer), + * we cmn_err() a message. We aren't expecting to + * get such an error -- such as it can exist at all, + * it would be a result of corrupted DOF in the driver + * properties. + */ + if (nmatched == NULL) { + cmn_err(CE_WARN, "dtrace_enabling_match() " + "error on %p: %d", (void *)ep, + enab->dten_error); + } + + return (enab->dten_error); + } + } + + enab->dten_probegen = dtrace_probegen; + if (nmatched != NULL) + *nmatched = matched; + + return (0); +} + +static void +dtrace_enabling_matchall(void) +{ + dtrace_enabling_t *enab; + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_lock); + + /* + * Iterate over all retained enablings to see if any probes match + * against them. We only perform this operation on enablings for which + * we have sufficient permissions by virtue of being in the global zone + * or in the same zone as the DTrace client. Because we can be called + * after dtrace_detach() has been called, we cannot assert that there + * are retained enablings. We can safely load from dtrace_retained, + * however: the taskq_destroy() at the end of dtrace_detach() will + * block pending our completion. + */ + for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { +#if defined(sun) + cred_t *cr = enab->dten_vstate->dtvs_state->dts_cred.dcr_cred; + + if (INGLOBALZONE(curproc) || getzoneid() == crgetzoneid(cr)) +#endif + (void) dtrace_enabling_match(enab, NULL); + } + + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); +} + +/* + * If an enabling is to be enabled without having matched probes (that is, if + * dtrace_state_go() is to be called on the underlying dtrace_state_t), the + * enabling must be _primed_ by creating an ECB for every ECB description. + * This must be done to assure that we know the number of speculations, the + * number of aggregations, the minimum buffer size needed, etc. before we + * transition out of DTRACE_ACTIVITY_INACTIVE. To do this without actually + * enabling any probes, we create ECBs for every ECB decription, but with a + * NULL probe -- which is exactly what this function does. + */ +static void +dtrace_enabling_prime(dtrace_state_t *state) +{ + dtrace_enabling_t *enab; + int i; + + for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { + ASSERT(enab->dten_vstate->dtvs_state != NULL); + + if (enab->dten_vstate->dtvs_state != state) + continue; + + /* + * We don't want to prime an enabling more than once, lest + * we allow a malicious user to induce resource exhaustion. + * (The ECBs that result from priming an enabling aren't + * leaked -- but they also aren't deallocated until the + * consumer state is destroyed.) + */ + if (enab->dten_primed) + continue; + + for (i = 0; i < enab->dten_ndesc; i++) { + enab->dten_current = enab->dten_desc[i]; + (void) dtrace_probe_enable(NULL, enab); + } + + enab->dten_primed = 1; + } +} + +/* + * Called to indicate that probes should be provided due to retained + * enablings. This is implemented in terms of dtrace_probe_provide(), but it + * must take an initial lap through the enabling calling the dtps_provide() + * entry point explicitly to allow for autocreated probes. + */ +static void +dtrace_enabling_provide(dtrace_provider_t *prv) +{ + int i, all = 0; + dtrace_probedesc_t desc; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(MUTEX_HELD(&dtrace_provider_lock)); + + if (prv == NULL) { + all = 1; + prv = dtrace_provider; + } + + do { + dtrace_enabling_t *enab = dtrace_retained; + void *parg = prv->dtpv_arg; + + for (; enab != NULL; enab = enab->dten_next) { + for (i = 0; i < enab->dten_ndesc; i++) { + desc = enab->dten_desc[i]->dted_probe; + mutex_exit(&dtrace_lock); + prv->dtpv_pops.dtps_provide(parg, &desc); + mutex_enter(&dtrace_lock); + } + } + } while (all && (prv = prv->dtpv_next) != NULL); + + mutex_exit(&dtrace_lock); + dtrace_probe_provide(NULL, all ? NULL : prv); + mutex_enter(&dtrace_lock); +} + +/* + * DTrace DOF Functions + */ +/*ARGSUSED*/ +static void +dtrace_dof_error(dof_hdr_t *dof, const char *str) +{ + if (dtrace_err_verbose) + cmn_err(CE_WARN, "failed to process DOF: %s", str); + +#ifdef DTRACE_ERRDEBUG + dtrace_errdebug(str); +#endif +} + +/* + * Create DOF out of a currently enabled state. Right now, we only create + * DOF containing the run-time options -- but this could be expanded to create + * complete DOF representing the enabled state. + */ +static dof_hdr_t * +dtrace_dof_create(dtrace_state_t *state) +{ + dof_hdr_t *dof; + dof_sec_t *sec; + dof_optdesc_t *opt; + int i, len = sizeof (dof_hdr_t) + + roundup(sizeof (dof_sec_t), sizeof (uint64_t)) + + sizeof (dof_optdesc_t) * DTRACEOPT_MAX; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + dof = kmem_zalloc(len, KM_SLEEP); + dof->dofh_ident[DOF_ID_MAG0] = DOF_MAG_MAG0; + dof->dofh_ident[DOF_ID_MAG1] = DOF_MAG_MAG1; + dof->dofh_ident[DOF_ID_MAG2] = DOF_MAG_MAG2; + dof->dofh_ident[DOF_ID_MAG3] = DOF_MAG_MAG3; + + dof->dofh_ident[DOF_ID_MODEL] = DOF_MODEL_NATIVE; + dof->dofh_ident[DOF_ID_ENCODING] = DOF_ENCODE_NATIVE; + dof->dofh_ident[DOF_ID_VERSION] = DOF_VERSION; + dof->dofh_ident[DOF_ID_DIFVERS] = DIF_VERSION; + dof->dofh_ident[DOF_ID_DIFIREG] = DIF_DIR_NREGS; + dof->dofh_ident[DOF_ID_DIFTREG] = DIF_DTR_NREGS; + + dof->dofh_flags = 0; + dof->dofh_hdrsize = sizeof (dof_hdr_t); + dof->dofh_secsize = sizeof (dof_sec_t); + dof->dofh_secnum = 1; /* only DOF_SECT_OPTDESC */ + dof->dofh_secoff = sizeof (dof_hdr_t); + dof->dofh_loadsz = len; + dof->dofh_filesz = len; + dof->dofh_pad = 0; + + /* + * Fill in the option section header... + */ + sec = (dof_sec_t *)((uintptr_t)dof + sizeof (dof_hdr_t)); + sec->dofs_type = DOF_SECT_OPTDESC; + sec->dofs_align = sizeof (uint64_t); + sec->dofs_flags = DOF_SECF_LOAD; + sec->dofs_entsize = sizeof (dof_optdesc_t); + + opt = (dof_optdesc_t *)((uintptr_t)sec + + roundup(sizeof (dof_sec_t), sizeof (uint64_t))); + + sec->dofs_offset = (uintptr_t)opt - (uintptr_t)dof; + sec->dofs_size = sizeof (dof_optdesc_t) * DTRACEOPT_MAX; + + for (i = 0; i < DTRACEOPT_MAX; i++) { + opt[i].dofo_option = i; + opt[i].dofo_strtab = DOF_SECIDX_NONE; + opt[i].dofo_value = state->dts_options[i]; + } + + return (dof); +} + +static dof_hdr_t * +dtrace_dof_copyin(uintptr_t uarg, int *errp) +{ + dof_hdr_t hdr, *dof; + + ASSERT(!MUTEX_HELD(&dtrace_lock)); + + /* + * First, we're going to copyin() the sizeof (dof_hdr_t). + */ + if (copyin((void *)uarg, &hdr, sizeof (hdr)) != 0) { + dtrace_dof_error(NULL, "failed to copyin DOF header"); + *errp = EFAULT; + return (NULL); + } + + /* + * Now we'll allocate the entire DOF and copy it in -- provided + * that the length isn't outrageous. + */ + if (hdr.dofh_loadsz >= dtrace_dof_maxsize) { + dtrace_dof_error(&hdr, "load size exceeds maximum"); + *errp = E2BIG; + return (NULL); + } + + if (hdr.dofh_loadsz < sizeof (hdr)) { + dtrace_dof_error(&hdr, "invalid load size"); + *errp = EINVAL; + return (NULL); + } + + dof = kmem_alloc(hdr.dofh_loadsz, KM_SLEEP); + + if (copyin((void *)uarg, dof, hdr.dofh_loadsz) != 0) { + kmem_free(dof, hdr.dofh_loadsz); + *errp = EFAULT; + return (NULL); + } + + return (dof); +} + +#if !defined(sun) +static __inline uchar_t +dtrace_dof_char(char c) { + switch (c) { + case '0': + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + return (c - '0'); + case 'A': + case 'B': + case 'C': + case 'D': + case 'E': + case 'F': + return (c - 'A' + 10); + case 'a': + case 'b': + case 'c': + case 'd': + case 'e': + case 'f': + return (c - 'a' + 10); + } + /* Should not reach here. */ + return (0); +} +#endif + +static dof_hdr_t * +dtrace_dof_property(const char *name) +{ + uchar_t *buf; + uint64_t loadsz; + unsigned int len, i; + dof_hdr_t *dof; + +#if defined(sun) + /* + * Unfortunately, array of values in .conf files are always (and + * only) interpreted to be integer arrays. We must read our DOF + * as an integer array, and then squeeze it into a byte array. + */ + if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dtrace_devi, 0, + (char *)name, (int **)&buf, &len) != DDI_PROP_SUCCESS) + return (NULL); + + for (i = 0; i < len; i++) + buf[i] = (uchar_t)(((int *)buf)[i]); + + if (len < sizeof (dof_hdr_t)) { + ddi_prop_free(buf); + dtrace_dof_error(NULL, "truncated header"); + return (NULL); + } + + if (len < (loadsz = ((dof_hdr_t *)buf)->dofh_loadsz)) { + ddi_prop_free(buf); + dtrace_dof_error(NULL, "truncated DOF"); + return (NULL); + } + + if (loadsz >= dtrace_dof_maxsize) { + ddi_prop_free(buf); + dtrace_dof_error(NULL, "oversized DOF"); + return (NULL); + } + + dof = kmem_alloc(loadsz, KM_SLEEP); + bcopy(buf, dof, loadsz); + ddi_prop_free(buf); +#else + char *p; + char *p_env; + + if ((p_env = getenv(name)) == NULL) + return (NULL); + + len = strlen(p_env) / 2; + + buf = kmem_alloc(len, KM_SLEEP); + + dof = (dof_hdr_t *) buf; + + p = p_env; + + for (i = 0; i < len; i++) { + buf[i] = (dtrace_dof_char(p[0]) << 4) | + dtrace_dof_char(p[1]); + p += 2; + } + + freeenv(p_env); + + if (len < sizeof (dof_hdr_t)) { + kmem_free(buf, 0); + dtrace_dof_error(NULL, "truncated header"); + return (NULL); + } + + if (len < (loadsz = dof->dofh_loadsz)) { + kmem_free(buf, 0); + dtrace_dof_error(NULL, "truncated DOF"); + return (NULL); + } + + if (loadsz >= dtrace_dof_maxsize) { + kmem_free(buf, 0); + dtrace_dof_error(NULL, "oversized DOF"); + return (NULL); + } +#endif + + return (dof); +} + +static void +dtrace_dof_destroy(dof_hdr_t *dof) +{ + kmem_free(dof, dof->dofh_loadsz); +} + +/* + * Return the dof_sec_t pointer corresponding to a given section index. If the + * index is not valid, dtrace_dof_error() is called and NULL is returned. If + * a type other than DOF_SECT_NONE is specified, the header is checked against + * this type and NULL is returned if the types do not match. + */ +static dof_sec_t * +dtrace_dof_sect(dof_hdr_t *dof, uint32_t type, dof_secidx_t i) +{ + dof_sec_t *sec = (dof_sec_t *)(uintptr_t) + ((uintptr_t)dof + dof->dofh_secoff + i * dof->dofh_secsize); + + if (i >= dof->dofh_secnum) { + dtrace_dof_error(dof, "referenced section index is invalid"); + return (NULL); + } + + if (!(sec->dofs_flags & DOF_SECF_LOAD)) { + dtrace_dof_error(dof, "referenced section is not loadable"); + return (NULL); + } + + if (type != DOF_SECT_NONE && type != sec->dofs_type) { + dtrace_dof_error(dof, "referenced section is the wrong type"); + return (NULL); + } + + return (sec); +} + +static dtrace_probedesc_t * +dtrace_dof_probedesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_probedesc_t *desc) +{ + dof_probedesc_t *probe; + dof_sec_t *strtab; + uintptr_t daddr = (uintptr_t)dof; + uintptr_t str; + size_t size; + + if (sec->dofs_type != DOF_SECT_PROBEDESC) { + dtrace_dof_error(dof, "invalid probe section"); + return (NULL); + } + + if (sec->dofs_align != sizeof (dof_secidx_t)) { + dtrace_dof_error(dof, "bad alignment in probe description"); + return (NULL); + } + + if (sec->dofs_offset + sizeof (dof_probedesc_t) > dof->dofh_loadsz) { + dtrace_dof_error(dof, "truncated probe description"); + return (NULL); + } + + probe = (dof_probedesc_t *)(uintptr_t)(daddr + sec->dofs_offset); + strtab = dtrace_dof_sect(dof, DOF_SECT_STRTAB, probe->dofp_strtab); + + if (strtab == NULL) + return (NULL); + + str = daddr + strtab->dofs_offset; + size = strtab->dofs_size; + + if (probe->dofp_provider >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe provider"); + return (NULL); + } + + (void) strncpy(desc->dtpd_provider, + (char *)(str + probe->dofp_provider), + MIN(DTRACE_PROVNAMELEN - 1, size - probe->dofp_provider)); + + if (probe->dofp_mod >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe module"); + return (NULL); + } + + (void) strncpy(desc->dtpd_mod, (char *)(str + probe->dofp_mod), + MIN(DTRACE_MODNAMELEN - 1, size - probe->dofp_mod)); + + if (probe->dofp_func >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe function"); + return (NULL); + } + + (void) strncpy(desc->dtpd_func, (char *)(str + probe->dofp_func), + MIN(DTRACE_FUNCNAMELEN - 1, size - probe->dofp_func)); + + if (probe->dofp_name >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe name"); + return (NULL); + } + + (void) strncpy(desc->dtpd_name, (char *)(str + probe->dofp_name), + MIN(DTRACE_NAMELEN - 1, size - probe->dofp_name)); + + return (desc); +} + +static dtrace_difo_t * +dtrace_dof_difo(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, + cred_t *cr) +{ + dtrace_difo_t *dp; + size_t ttl = 0; + dof_difohdr_t *dofd; + uintptr_t daddr = (uintptr_t)dof; + size_t max = dtrace_difo_maxsize; + int i, l, n; + + static const struct { + int section; + int bufoffs; + int lenoffs; + int entsize; + int align; + const char *msg; + } difo[] = { + { DOF_SECT_DIF, offsetof(dtrace_difo_t, dtdo_buf), + offsetof(dtrace_difo_t, dtdo_len), sizeof (dif_instr_t), + sizeof (dif_instr_t), "multiple DIF sections" }, + + { DOF_SECT_INTTAB, offsetof(dtrace_difo_t, dtdo_inttab), + offsetof(dtrace_difo_t, dtdo_intlen), sizeof (uint64_t), + sizeof (uint64_t), "multiple integer tables" }, + + { DOF_SECT_STRTAB, offsetof(dtrace_difo_t, dtdo_strtab), + offsetof(dtrace_difo_t, dtdo_strlen), 0, + sizeof (char), "multiple string tables" }, + + { DOF_SECT_VARTAB, offsetof(dtrace_difo_t, dtdo_vartab), + offsetof(dtrace_difo_t, dtdo_varlen), sizeof (dtrace_difv_t), + sizeof (uint_t), "multiple variable tables" }, + + { DOF_SECT_NONE, 0, 0, 0, 0, NULL } + }; + + if (sec->dofs_type != DOF_SECT_DIFOHDR) { + dtrace_dof_error(dof, "invalid DIFO header section"); + return (NULL); + } + + if (sec->dofs_align != sizeof (dof_secidx_t)) { + dtrace_dof_error(dof, "bad alignment in DIFO header"); + return (NULL); + } + + if (sec->dofs_size < sizeof (dof_difohdr_t) || + sec->dofs_size % sizeof (dof_secidx_t)) { + dtrace_dof_error(dof, "bad size in DIFO header"); + return (NULL); + } + + dofd = (dof_difohdr_t *)(uintptr_t)(daddr + sec->dofs_offset); + n = (sec->dofs_size - sizeof (*dofd)) / sizeof (dof_secidx_t) + 1; + + dp = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP); + dp->dtdo_rtype = dofd->dofd_rtype; + + for (l = 0; l < n; l++) { + dof_sec_t *subsec; + void **bufp; + uint32_t *lenp; + + if ((subsec = dtrace_dof_sect(dof, DOF_SECT_NONE, + dofd->dofd_links[l])) == NULL) + goto err; /* invalid section link */ + + if (ttl + subsec->dofs_size > max) { + dtrace_dof_error(dof, "exceeds maximum size"); + goto err; + } + + ttl += subsec->dofs_size; + + for (i = 0; difo[i].section != DOF_SECT_NONE; i++) { + if (subsec->dofs_type != difo[i].section) + continue; + + if (!(subsec->dofs_flags & DOF_SECF_LOAD)) { + dtrace_dof_error(dof, "section not loaded"); + goto err; + } + + if (subsec->dofs_align != difo[i].align) { + dtrace_dof_error(dof, "bad alignment"); + goto err; + } + + bufp = (void **)((uintptr_t)dp + difo[i].bufoffs); + lenp = (uint32_t *)((uintptr_t)dp + difo[i].lenoffs); + + if (*bufp != NULL) { + dtrace_dof_error(dof, difo[i].msg); + goto err; + } + + if (difo[i].entsize != subsec->dofs_entsize) { + dtrace_dof_error(dof, "entry size mismatch"); + goto err; + } + + if (subsec->dofs_entsize != 0 && + (subsec->dofs_size % subsec->dofs_entsize) != 0) { + dtrace_dof_error(dof, "corrupt entry size"); + goto err; + } + + *lenp = subsec->dofs_size; + *bufp = kmem_alloc(subsec->dofs_size, KM_SLEEP); + bcopy((char *)(uintptr_t)(daddr + subsec->dofs_offset), + *bufp, subsec->dofs_size); + + if (subsec->dofs_entsize != 0) + *lenp /= subsec->dofs_entsize; + + break; + } + + /* + * If we encounter a loadable DIFO sub-section that is not + * known to us, assume this is a broken program and fail. + */ + if (difo[i].section == DOF_SECT_NONE && + (subsec->dofs_flags & DOF_SECF_LOAD)) { + dtrace_dof_error(dof, "unrecognized DIFO subsection"); + goto err; + } + } + + if (dp->dtdo_buf == NULL) { + /* + * We can't have a DIF object without DIF text. + */ + dtrace_dof_error(dof, "missing DIF text"); + goto err; + } + + /* + * Before we validate the DIF object, run through the variable table + * looking for the strings -- if any of their size are under, we'll set + * their size to be the system-wide default string size. Note that + * this should _not_ happen if the "strsize" option has been set -- + * in this case, the compiler should have set the size to reflect the + * setting of the option. + */ + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + dtrace_diftype_t *t = &v->dtdv_type; + + if (v->dtdv_id < DIF_VAR_OTHER_UBASE) + continue; + + if (t->dtdt_kind == DIF_TYPE_STRING && t->dtdt_size == 0) + t->dtdt_size = dtrace_strsize_default; + } + + if (dtrace_difo_validate(dp, vstate, DIF_DIR_NREGS, cr) != 0) + goto err; + + dtrace_difo_init(dp, vstate); + return (dp); + +err: + kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t)); + kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t)); + kmem_free(dp->dtdo_strtab, dp->dtdo_strlen); + kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t)); + + kmem_free(dp, sizeof (dtrace_difo_t)); + return (NULL); +} + +static dtrace_predicate_t * +dtrace_dof_predicate(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, + cred_t *cr) +{ + dtrace_difo_t *dp; + + if ((dp = dtrace_dof_difo(dof, sec, vstate, cr)) == NULL) + return (NULL); + + return (dtrace_predicate_create(dp)); +} + +static dtrace_actdesc_t * +dtrace_dof_actdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, + cred_t *cr) +{ + dtrace_actdesc_t *act, *first = NULL, *last = NULL, *next; + dof_actdesc_t *desc; + dof_sec_t *difosec; + size_t offs; + uintptr_t daddr = (uintptr_t)dof; + uint64_t arg; + dtrace_actkind_t kind; + + if (sec->dofs_type != DOF_SECT_ACTDESC) { + dtrace_dof_error(dof, "invalid action section"); + return (NULL); + } + + if (sec->dofs_offset + sizeof (dof_actdesc_t) > dof->dofh_loadsz) { + dtrace_dof_error(dof, "truncated action description"); + return (NULL); + } + + if (sec->dofs_align != sizeof (uint64_t)) { + dtrace_dof_error(dof, "bad alignment in action description"); + return (NULL); + } + + if (sec->dofs_size < sec->dofs_entsize) { + dtrace_dof_error(dof, "section entry size exceeds total size"); + return (NULL); + } + + if (sec->dofs_entsize != sizeof (dof_actdesc_t)) { + dtrace_dof_error(dof, "bad entry size in action description"); + return (NULL); + } + + if (sec->dofs_size / sec->dofs_entsize > dtrace_actions_max) { + dtrace_dof_error(dof, "actions exceed dtrace_actions_max"); + return (NULL); + } + + for (offs = 0; offs < sec->dofs_size; offs += sec->dofs_entsize) { + desc = (dof_actdesc_t *)(daddr + + (uintptr_t)sec->dofs_offset + offs); + kind = (dtrace_actkind_t)desc->dofa_kind; + + if (DTRACEACT_ISPRINTFLIKE(kind) && + (kind != DTRACEACT_PRINTA || + desc->dofa_strtab != DOF_SECIDX_NONE)) { + dof_sec_t *strtab; + char *str, *fmt; + uint64_t i; + + /* + * printf()-like actions must have a format string. + */ + if ((strtab = dtrace_dof_sect(dof, + DOF_SECT_STRTAB, desc->dofa_strtab)) == NULL) + goto err; + + str = (char *)((uintptr_t)dof + + (uintptr_t)strtab->dofs_offset); + + for (i = desc->dofa_arg; i < strtab->dofs_size; i++) { + if (str[i] == '\0') + break; + } + + if (i >= strtab->dofs_size) { + dtrace_dof_error(dof, "bogus format string"); + goto err; + } + + if (i == desc->dofa_arg) { + dtrace_dof_error(dof, "empty format string"); + goto err; + } + + i -= desc->dofa_arg; + fmt = kmem_alloc(i + 1, KM_SLEEP); + bcopy(&str[desc->dofa_arg], fmt, i + 1); + arg = (uint64_t)(uintptr_t)fmt; + } else { + if (kind == DTRACEACT_PRINTA) { + ASSERT(desc->dofa_strtab == DOF_SECIDX_NONE); + arg = 0; + } else { + arg = desc->dofa_arg; + } + } + + act = dtrace_actdesc_create(kind, desc->dofa_ntuple, + desc->dofa_uarg, arg); + + if (last != NULL) { + last->dtad_next = act; + } else { + first = act; + } + + last = act; + + if (desc->dofa_difo == DOF_SECIDX_NONE) + continue; + + if ((difosec = dtrace_dof_sect(dof, + DOF_SECT_DIFOHDR, desc->dofa_difo)) == NULL) + goto err; + + act->dtad_difo = dtrace_dof_difo(dof, difosec, vstate, cr); + + if (act->dtad_difo == NULL) + goto err; + } + + ASSERT(first != NULL); + return (first); + +err: + for (act = first; act != NULL; act = next) { + next = act->dtad_next; + dtrace_actdesc_release(act, vstate); + } + + return (NULL); +} + +static dtrace_ecbdesc_t * +dtrace_dof_ecbdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, + cred_t *cr) +{ + dtrace_ecbdesc_t *ep; + dof_ecbdesc_t *ecb; + dtrace_probedesc_t *desc; + dtrace_predicate_t *pred = NULL; + + if (sec->dofs_size < sizeof (dof_ecbdesc_t)) { + dtrace_dof_error(dof, "truncated ECB description"); + return (NULL); + } + + if (sec->dofs_align != sizeof (uint64_t)) { + dtrace_dof_error(dof, "bad alignment in ECB description"); + return (NULL); + } + + ecb = (dof_ecbdesc_t *)((uintptr_t)dof + (uintptr_t)sec->dofs_offset); + sec = dtrace_dof_sect(dof, DOF_SECT_PROBEDESC, ecb->dofe_probes); + + if (sec == NULL) + return (NULL); + + ep = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP); + ep->dted_uarg = ecb->dofe_uarg; + desc = &ep->dted_probe; + + if (dtrace_dof_probedesc(dof, sec, desc) == NULL) + goto err; + + if (ecb->dofe_pred != DOF_SECIDX_NONE) { + if ((sec = dtrace_dof_sect(dof, + DOF_SECT_DIFOHDR, ecb->dofe_pred)) == NULL) + goto err; + + if ((pred = dtrace_dof_predicate(dof, sec, vstate, cr)) == NULL) + goto err; + + ep->dted_pred.dtpdd_predicate = pred; + } + + if (ecb->dofe_actions != DOF_SECIDX_NONE) { + if ((sec = dtrace_dof_sect(dof, + DOF_SECT_ACTDESC, ecb->dofe_actions)) == NULL) + goto err; + + ep->dted_action = dtrace_dof_actdesc(dof, sec, vstate, cr); + + if (ep->dted_action == NULL) + goto err; + } + + return (ep); + +err: + if (pred != NULL) + dtrace_predicate_release(pred, vstate); + kmem_free(ep, sizeof (dtrace_ecbdesc_t)); + return (NULL); +} + +/* + * Apply the relocations from the specified 'sec' (a DOF_SECT_URELHDR) to the + * specified DOF. At present, this amounts to simply adding 'ubase' to the + * site of any user SETX relocations to account for load object base address. + * In the future, if we need other relocations, this function can be extended. + */ +static int +dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase) +{ + uintptr_t daddr = (uintptr_t)dof; + dof_relohdr_t *dofr = + (dof_relohdr_t *)(uintptr_t)(daddr + sec->dofs_offset); + dof_sec_t *ss, *rs, *ts; + dof_relodesc_t *r; + uint_t i, n; + + if (sec->dofs_size < sizeof (dof_relohdr_t) || + sec->dofs_align != sizeof (dof_secidx_t)) { + dtrace_dof_error(dof, "invalid relocation header"); + return (-1); + } + + ss = dtrace_dof_sect(dof, DOF_SECT_STRTAB, dofr->dofr_strtab); + rs = dtrace_dof_sect(dof, DOF_SECT_RELTAB, dofr->dofr_relsec); + ts = dtrace_dof_sect(dof, DOF_SECT_NONE, dofr->dofr_tgtsec); + + if (ss == NULL || rs == NULL || ts == NULL) + return (-1); /* dtrace_dof_error() has been called already */ + + if (rs->dofs_entsize < sizeof (dof_relodesc_t) || + rs->dofs_align != sizeof (uint64_t)) { + dtrace_dof_error(dof, "invalid relocation section"); + return (-1); + } + + r = (dof_relodesc_t *)(uintptr_t)(daddr + rs->dofs_offset); + n = rs->dofs_size / rs->dofs_entsize; + + for (i = 0; i < n; i++) { + uintptr_t taddr = daddr + ts->dofs_offset + r->dofr_offset; + + switch (r->dofr_type) { + case DOF_RELO_NONE: + break; + case DOF_RELO_SETX: + if (r->dofr_offset >= ts->dofs_size || r->dofr_offset + + sizeof (uint64_t) > ts->dofs_size) { + dtrace_dof_error(dof, "bad relocation offset"); + return (-1); + } + + if (!IS_P2ALIGNED(taddr, sizeof (uint64_t))) { + dtrace_dof_error(dof, "misaligned setx relo"); + return (-1); + } + + *(uint64_t *)taddr += ubase; + break; + default: + dtrace_dof_error(dof, "invalid relocation type"); + return (-1); + } + + r = (dof_relodesc_t *)((uintptr_t)r + rs->dofs_entsize); + } + + return (0); +} + +/* + * The dof_hdr_t passed to dtrace_dof_slurp() should be a partially validated + * header: it should be at the front of a memory region that is at least + * sizeof (dof_hdr_t) in size -- and then at least dof_hdr.dofh_loadsz in + * size. It need not be validated in any other way. + */ +static int +dtrace_dof_slurp(dof_hdr_t *dof, dtrace_vstate_t *vstate, cred_t *cr, + dtrace_enabling_t **enabp, uint64_t ubase, int noprobes) +{ + uint64_t len = dof->dofh_loadsz, seclen; + uintptr_t daddr = (uintptr_t)dof; + dtrace_ecbdesc_t *ep; + dtrace_enabling_t *enab; + uint_t i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dof->dofh_loadsz >= sizeof (dof_hdr_t)); + + /* + * Check the DOF header identification bytes. In addition to checking + * valid settings, we also verify that unused bits/bytes are zeroed so + * we can use them later without fear of regressing existing binaries. + */ + if (bcmp(&dof->dofh_ident[DOF_ID_MAG0], + DOF_MAG_STRING, DOF_MAG_STRLEN) != 0) { + dtrace_dof_error(dof, "DOF magic string mismatch"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_ILP32 && + dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_LP64) { + dtrace_dof_error(dof, "DOF has invalid data model"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_ENCODING] != DOF_ENCODE_NATIVE) { + dtrace_dof_error(dof, "DOF encoding mismatch"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && + dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_2) { + dtrace_dof_error(dof, "DOF version mismatch"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_DIFVERS] != DIF_VERSION_2) { + dtrace_dof_error(dof, "DOF uses unsupported instruction set"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_DIFIREG] > DIF_DIR_NREGS) { + dtrace_dof_error(dof, "DOF uses too many integer registers"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_DIFTREG] > DIF_DTR_NREGS) { + dtrace_dof_error(dof, "DOF uses too many tuple registers"); + return (-1); + } + + for (i = DOF_ID_PAD; i < DOF_ID_SIZE; i++) { + if (dof->dofh_ident[i] != 0) { + dtrace_dof_error(dof, "DOF has invalid ident byte set"); + return (-1); + } + } + + if (dof->dofh_flags & ~DOF_FL_VALID) { + dtrace_dof_error(dof, "DOF has invalid flag bits set"); + return (-1); + } + + if (dof->dofh_secsize == 0) { + dtrace_dof_error(dof, "zero section header size"); + return (-1); + } + + /* + * Check that the section headers don't exceed the amount of DOF + * data. Note that we cast the section size and number of sections + * to uint64_t's to prevent possible overflow in the multiplication. + */ + seclen = (uint64_t)dof->dofh_secnum * (uint64_t)dof->dofh_secsize; + + if (dof->dofh_secoff > len || seclen > len || + dof->dofh_secoff + seclen > len) { + dtrace_dof_error(dof, "truncated section headers"); + return (-1); + } + + if (!IS_P2ALIGNED(dof->dofh_secoff, sizeof (uint64_t))) { + dtrace_dof_error(dof, "misaligned section headers"); + return (-1); + } + + if (!IS_P2ALIGNED(dof->dofh_secsize, sizeof (uint64_t))) { + dtrace_dof_error(dof, "misaligned section size"); + return (-1); + } + + /* + * Take an initial pass through the section headers to be sure that + * the headers don't have stray offsets. If the 'noprobes' flag is + * set, do not permit sections relating to providers, probes, or args. + */ + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(daddr + + (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); + + if (noprobes) { + switch (sec->dofs_type) { + case DOF_SECT_PROVIDER: + case DOF_SECT_PROBES: + case DOF_SECT_PRARGS: + case DOF_SECT_PROFFS: + dtrace_dof_error(dof, "illegal sections " + "for enabling"); + return (-1); + } + } + + if (!(sec->dofs_flags & DOF_SECF_LOAD)) + continue; /* just ignore non-loadable sections */ + + if (sec->dofs_align & (sec->dofs_align - 1)) { + dtrace_dof_error(dof, "bad section alignment"); + return (-1); + } + + if (sec->dofs_offset & (sec->dofs_align - 1)) { + dtrace_dof_error(dof, "misaligned section"); + return (-1); + } + + if (sec->dofs_offset > len || sec->dofs_size > len || + sec->dofs_offset + sec->dofs_size > len) { + dtrace_dof_error(dof, "corrupt section header"); + return (-1); + } + + if (sec->dofs_type == DOF_SECT_STRTAB && *((char *)daddr + + sec->dofs_offset + sec->dofs_size - 1) != '\0') { + dtrace_dof_error(dof, "non-terminating string table"); + return (-1); + } + } + + /* + * Take a second pass through the sections and locate and perform any + * relocations that are present. We do this after the first pass to + * be sure that all sections have had their headers validated. + */ + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(daddr + + (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); + + if (!(sec->dofs_flags & DOF_SECF_LOAD)) + continue; /* skip sections that are not loadable */ + + switch (sec->dofs_type) { + case DOF_SECT_URELHDR: + if (dtrace_dof_relocate(dof, sec, ubase) != 0) + return (-1); + break; + } + } + + if ((enab = *enabp) == NULL) + enab = *enabp = dtrace_enabling_create(vstate); + + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(daddr + + (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_ECBDESC) + continue; + + if ((ep = dtrace_dof_ecbdesc(dof, sec, vstate, cr)) == NULL) { + dtrace_enabling_destroy(enab); + *enabp = NULL; + return (-1); + } + + dtrace_enabling_add(enab, ep); + } + + return (0); +} + +/* + * Process DOF for any options. This routine assumes that the DOF has been + * at least processed by dtrace_dof_slurp(). + */ +static int +dtrace_dof_options(dof_hdr_t *dof, dtrace_state_t *state) +{ + int i, rval; + uint32_t entsize; + size_t offs; + dof_optdesc_t *desc; + + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)((uintptr_t)dof + + (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_OPTDESC) + continue; + + if (sec->dofs_align != sizeof (uint64_t)) { + dtrace_dof_error(dof, "bad alignment in " + "option description"); + return (EINVAL); + } + + if ((entsize = sec->dofs_entsize) == 0) { + dtrace_dof_error(dof, "zeroed option entry size"); + return (EINVAL); + } + + if (entsize < sizeof (dof_optdesc_t)) { + dtrace_dof_error(dof, "bad option entry size"); + return (EINVAL); + } + + for (offs = 0; offs < sec->dofs_size; offs += entsize) { + desc = (dof_optdesc_t *)((uintptr_t)dof + + (uintptr_t)sec->dofs_offset + offs); + + if (desc->dofo_strtab != DOF_SECIDX_NONE) { + dtrace_dof_error(dof, "non-zero option string"); + return (EINVAL); + } + + if (desc->dofo_value == DTRACEOPT_UNSET) { + dtrace_dof_error(dof, "unset option"); + return (EINVAL); + } + + if ((rval = dtrace_state_option(state, + desc->dofo_option, desc->dofo_value)) != 0) { + dtrace_dof_error(dof, "rejected option"); + return (rval); + } + } + } + + return (0); +} + +/* + * DTrace Consumer State Functions + */ +static int +dtrace_dstate_init(dtrace_dstate_t *dstate, size_t size) +{ + size_t hashsize, maxper, min, chunksize = dstate->dtds_chunksize; + void *base; + uintptr_t limit; + dtrace_dynvar_t *dvar, *next, *start; + int i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dstate->dtds_base == NULL && dstate->dtds_percpu == NULL); + + bzero(dstate, sizeof (dtrace_dstate_t)); + + if ((dstate->dtds_chunksize = chunksize) == 0) + dstate->dtds_chunksize = DTRACE_DYNVAR_CHUNKSIZE; + + if (size < (min = dstate->dtds_chunksize + sizeof (dtrace_dynhash_t))) + size = min; + + if ((base = kmem_zalloc(size, KM_NOSLEEP)) == NULL) + return (ENOMEM); + + dstate->dtds_size = size; + dstate->dtds_base = base; + dstate->dtds_percpu = kmem_cache_alloc(dtrace_state_cache, KM_SLEEP); + bzero(dstate->dtds_percpu, NCPU * sizeof (dtrace_dstate_percpu_t)); + + hashsize = size / (dstate->dtds_chunksize + sizeof (dtrace_dynhash_t)); + + if (hashsize != 1 && (hashsize & 1)) + hashsize--; + + dstate->dtds_hashsize = hashsize; + dstate->dtds_hash = dstate->dtds_base; + + /* + * Set all of our hash buckets to point to the single sink, and (if + * it hasn't already been set), set the sink's hash value to be the + * sink sentinel value. The sink is needed for dynamic variable + * lookups to know that they have iterated over an entire, valid hash + * chain. + */ + for (i = 0; i < hashsize; i++) + dstate->dtds_hash[i].dtdh_chain = &dtrace_dynhash_sink; + + if (dtrace_dynhash_sink.dtdv_hashval != DTRACE_DYNHASH_SINK) + dtrace_dynhash_sink.dtdv_hashval = DTRACE_DYNHASH_SINK; + + /* + * Determine number of active CPUs. Divide free list evenly among + * active CPUs. + */ + start = (dtrace_dynvar_t *) + ((uintptr_t)base + hashsize * sizeof (dtrace_dynhash_t)); + limit = (uintptr_t)base + size; + + maxper = (limit - (uintptr_t)start) / NCPU; + maxper = (maxper / dstate->dtds_chunksize) * dstate->dtds_chunksize; + + for (i = 0; i < NCPU; i++) { +#if !defined(sun) + if (CPU_ABSENT(i)) + continue; +#endif + dstate->dtds_percpu[i].dtdsc_free = dvar = start; + + /* + * If we don't even have enough chunks to make it once through + * NCPUs, we're just going to allocate everything to the first + * CPU. And if we're on the last CPU, we're going to allocate + * whatever is left over. In either case, we set the limit to + * be the limit of the dynamic variable space. + */ + if (maxper == 0 || i == NCPU - 1) { + limit = (uintptr_t)base + size; + start = NULL; + } else { + limit = (uintptr_t)start + maxper; + start = (dtrace_dynvar_t *)limit; + } + + ASSERT(limit <= (uintptr_t)base + size); + + for (;;) { + next = (dtrace_dynvar_t *)((uintptr_t)dvar + + dstate->dtds_chunksize); + + if ((uintptr_t)next + dstate->dtds_chunksize >= limit) + break; + + dvar->dtdv_next = next; + dvar = next; + } + + if (maxper == 0) + break; + } + + return (0); +} + +static void +dtrace_dstate_fini(dtrace_dstate_t *dstate) +{ + ASSERT(MUTEX_HELD(&cpu_lock)); + + if (dstate->dtds_base == NULL) + return; + + kmem_free(dstate->dtds_base, dstate->dtds_size); + kmem_cache_free(dtrace_state_cache, dstate->dtds_percpu); +} + +static void +dtrace_vstate_fini(dtrace_vstate_t *vstate) +{ + /* + * Logical XOR, where are you? + */ + ASSERT((vstate->dtvs_nglobals == 0) ^ (vstate->dtvs_globals != NULL)); + + if (vstate->dtvs_nglobals > 0) { + kmem_free(vstate->dtvs_globals, vstate->dtvs_nglobals * + sizeof (dtrace_statvar_t *)); + } + + if (vstate->dtvs_ntlocals > 0) { + kmem_free(vstate->dtvs_tlocals, vstate->dtvs_ntlocals * + sizeof (dtrace_difv_t)); + } + + ASSERT((vstate->dtvs_nlocals == 0) ^ (vstate->dtvs_locals != NULL)); + + if (vstate->dtvs_nlocals > 0) { + kmem_free(vstate->dtvs_locals, vstate->dtvs_nlocals * + sizeof (dtrace_statvar_t *)); + } +} + +#if defined(sun) +static void +dtrace_state_clean(dtrace_state_t *state) +{ + if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) + return; + + dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars); + dtrace_speculation_clean(state); +} + +static void +dtrace_state_deadman(dtrace_state_t *state) +{ + hrtime_t now; + + dtrace_sync(); + + now = dtrace_gethrtime(); + + if (state != dtrace_anon.dta_state && + now - state->dts_laststatus >= dtrace_deadman_user) + return; + + /* + * We must be sure that dts_alive never appears to be less than the + * value upon entry to dtrace_state_deadman(), and because we lack a + * dtrace_cas64(), we cannot store to it atomically. We thus instead + * store INT64_MAX to it, followed by a memory barrier, followed by + * the new value. This assures that dts_alive never appears to be + * less than its true value, regardless of the order in which the + * stores to the underlying storage are issued. + */ + state->dts_alive = INT64_MAX; + dtrace_membar_producer(); + state->dts_alive = now; +} +#else +static void +dtrace_state_clean(void *arg) +{ + dtrace_state_t *state = arg; + dtrace_optval_t *opt = state->dts_options; + + if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) + return; + + dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars); + dtrace_speculation_clean(state); + + callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC, + dtrace_state_clean, state); +} + +static void +dtrace_state_deadman(void *arg) +{ + dtrace_state_t *state = arg; + hrtime_t now; + + dtrace_sync(); + + dtrace_debug_output(); + + now = dtrace_gethrtime(); + + if (state != dtrace_anon.dta_state && + now - state->dts_laststatus >= dtrace_deadman_user) + return; + + /* + * We must be sure that dts_alive never appears to be less than the + * value upon entry to dtrace_state_deadman(), and because we lack a + * dtrace_cas64(), we cannot store to it atomically. We thus instead + * store INT64_MAX to it, followed by a memory barrier, followed by + * the new value. This assures that dts_alive never appears to be + * less than its true value, regardless of the order in which the + * stores to the underlying storage are issued. + */ + state->dts_alive = INT64_MAX; + dtrace_membar_producer(); + state->dts_alive = now; + + callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC, + dtrace_state_deadman, state); +} +#endif + +static dtrace_state_t * +#if defined(sun) +dtrace_state_create(dev_t *devp, cred_t *cr) +#else +dtrace_state_create(struct cdev *dev) +#endif +{ +#if defined(sun) + minor_t minor; + major_t major; +#else + cred_t *cr = NULL; + int m = 0; +#endif + char c[30]; + dtrace_state_t *state; + dtrace_optval_t *opt; + int bufsize = NCPU * sizeof (dtrace_buffer_t), i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(MUTEX_HELD(&cpu_lock)); + +#if defined(sun) + minor = (minor_t)(uintptr_t)vmem_alloc(dtrace_minor, 1, + VM_BESTFIT | VM_SLEEP); + + if (ddi_soft_state_zalloc(dtrace_softstate, minor) != DDI_SUCCESS) { + vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1); + return (NULL); + } + + state = ddi_get_soft_state(dtrace_softstate, minor); +#else + if (dev != NULL) { + cr = dev->si_cred; + m = minor(dev); + } + + /* Allocate memory for the state. */ + state = kmem_zalloc(sizeof(dtrace_state_t), KM_SLEEP); +#endif + + state->dts_epid = DTRACE_EPIDNONE + 1; + + (void) snprintf(c, sizeof (c), "dtrace_aggid_%d", m); +#if defined(sun) + state->dts_aggid_arena = vmem_create(c, (void *)1, UINT32_MAX, 1, + NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER); + + if (devp != NULL) { + major = getemajor(*devp); + } else { + major = ddi_driver_major(dtrace_devi); + } + + state->dts_dev = makedevice(major, minor); + + if (devp != NULL) + *devp = state->dts_dev; +#else + state->dts_aggid_arena = new_unrhdr(1, INT_MAX, &dtrace_unr_mtx); + state->dts_dev = dev; +#endif + + /* + * We allocate NCPU buffers. On the one hand, this can be quite + * a bit of memory per instance (nearly 36K on a Starcat). On the + * other hand, it saves an additional memory reference in the probe + * path. + */ + state->dts_buffer = kmem_zalloc(bufsize, KM_SLEEP); + state->dts_aggbuffer = kmem_zalloc(bufsize, KM_SLEEP); + +#if defined(sun) + state->dts_cleaner = CYCLIC_NONE; + state->dts_deadman = CYCLIC_NONE; +#else + callout_init(&state->dts_cleaner, CALLOUT_MPSAFE); + callout_init(&state->dts_deadman, CALLOUT_MPSAFE); +#endif + state->dts_vstate.dtvs_state = state; + + for (i = 0; i < DTRACEOPT_MAX; i++) + state->dts_options[i] = DTRACEOPT_UNSET; + + /* + * Set the default options. + */ + opt = state->dts_options; + opt[DTRACEOPT_BUFPOLICY] = DTRACEOPT_BUFPOLICY_SWITCH; + opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_AUTO; + opt[DTRACEOPT_NSPEC] = dtrace_nspec_default; + opt[DTRACEOPT_SPECSIZE] = dtrace_specsize_default; + opt[DTRACEOPT_CPU] = (dtrace_optval_t)DTRACE_CPUALL; + opt[DTRACEOPT_STRSIZE] = dtrace_strsize_default; + opt[DTRACEOPT_STACKFRAMES] = dtrace_stackframes_default; + opt[DTRACEOPT_USTACKFRAMES] = dtrace_ustackframes_default; + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_default; + opt[DTRACEOPT_AGGRATE] = dtrace_aggrate_default; + opt[DTRACEOPT_SWITCHRATE] = dtrace_switchrate_default; + opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_default; + opt[DTRACEOPT_JSTACKFRAMES] = dtrace_jstackframes_default; + opt[DTRACEOPT_JSTACKSTRSIZE] = dtrace_jstackstrsize_default; + + state->dts_activity = DTRACE_ACTIVITY_INACTIVE; + + /* + * Depending on the user credentials, we set flag bits which alter probe + * visibility or the amount of destructiveness allowed. In the case of + * actual anonymous tracing, or the possession of all privileges, all of + * the normal checks are bypassed. + */ + if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) { + state->dts_cred.dcr_visible = DTRACE_CRV_ALL; + state->dts_cred.dcr_action = DTRACE_CRA_ALL; + } else { + /* + * Set up the credentials for this instantiation. We take a + * hold on the credential to prevent it from disappearing on + * us; this in turn prevents the zone_t referenced by this + * credential from disappearing. This means that we can + * examine the credential and the zone from probe context. + */ + crhold(cr); + state->dts_cred.dcr_cred = cr; + + /* + * CRA_PROC means "we have *some* privilege for dtrace" and + * unlocks the use of variables like pid, zonename, etc. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE) || + PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) { + state->dts_cred.dcr_action |= DTRACE_CRA_PROC; + } + + /* + * dtrace_user allows use of syscall and profile providers. + * If the user also has proc_owner and/or proc_zone, we + * extend the scope to include additional visibility and + * destructive power. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) { + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) { + state->dts_cred.dcr_visible |= + DTRACE_CRV_ALLPROC; + + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; + } + + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) { + state->dts_cred.dcr_visible |= + DTRACE_CRV_ALLZONE; + + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; + } + + /* + * If we have all privs in whatever zone this is, + * we can do destructive things to processes which + * have altered credentials. + */ +#if defined(sun) + if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE), + cr->cr_zone->zone_privset)) { + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG; + } +#endif + } + + /* + * Holding the dtrace_kernel privilege also implies that + * the user has the dtrace_user privilege from a visibility + * perspective. But without further privileges, some + * destructive actions are not available. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) { + /* + * Make all probes in all zones visible. However, + * this doesn't mean that all actions become available + * to all zones. + */ + state->dts_cred.dcr_visible |= DTRACE_CRV_KERNEL | + DTRACE_CRV_ALLPROC | DTRACE_CRV_ALLZONE; + + state->dts_cred.dcr_action |= DTRACE_CRA_KERNEL | + DTRACE_CRA_PROC; + /* + * Holding proc_owner means that destructive actions + * for *this* zone are allowed. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; + + /* + * Holding proc_zone means that destructive actions + * for this user/group ID in all zones is allowed. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; + +#if defined(sun) + /* + * If we have all privs in whatever zone this is, + * we can do destructive things to processes which + * have altered credentials. + */ + if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE), + cr->cr_zone->zone_privset)) { + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG; + } +#endif + } + + /* + * Holding the dtrace_proc privilege gives control over fasttrap + * and pid providers. We need to grant wider destructive + * privileges in the event that the user has proc_owner and/or + * proc_zone. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) { + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; + + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; + } + } + + return (state); +} + +static int +dtrace_state_buffer(dtrace_state_t *state, dtrace_buffer_t *buf, int which) +{ + dtrace_optval_t *opt = state->dts_options, size; + processorid_t cpu = 0;; + int flags = 0, rval; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(which < DTRACEOPT_MAX); + ASSERT(state->dts_activity == DTRACE_ACTIVITY_INACTIVE || + (state == dtrace_anon.dta_state && + state->dts_activity == DTRACE_ACTIVITY_ACTIVE)); + + if (opt[which] == DTRACEOPT_UNSET || opt[which] == 0) + return (0); + + if (opt[DTRACEOPT_CPU] != DTRACEOPT_UNSET) + cpu = opt[DTRACEOPT_CPU]; + + if (which == DTRACEOPT_SPECSIZE) + flags |= DTRACEBUF_NOSWITCH; + + if (which == DTRACEOPT_BUFSIZE) { + if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_RING) + flags |= DTRACEBUF_RING; + + if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_FILL) + flags |= DTRACEBUF_FILL; + + if (state != dtrace_anon.dta_state || + state->dts_activity != DTRACE_ACTIVITY_ACTIVE) + flags |= DTRACEBUF_INACTIVE; + } + + for (size = opt[which]; size >= sizeof (uint64_t); size >>= 1) { + /* + * The size must be 8-byte aligned. If the size is not 8-byte + * aligned, drop it down by the difference. + */ + if (size & (sizeof (uint64_t) - 1)) + size -= size & (sizeof (uint64_t) - 1); + + if (size < state->dts_reserve) { + /* + * Buffers always must be large enough to accommodate + * their prereserved space. We return E2BIG instead + * of ENOMEM in this case to allow for user-level + * software to differentiate the cases. + */ + return (E2BIG); + } + + rval = dtrace_buffer_alloc(buf, size, flags, cpu); + + if (rval != ENOMEM) { + opt[which] = size; + return (rval); + } + + if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL) + return (rval); + } + + return (ENOMEM); +} + +static int +dtrace_state_buffers(dtrace_state_t *state) +{ + dtrace_speculation_t *spec = state->dts_speculations; + int rval, i; + + if ((rval = dtrace_state_buffer(state, state->dts_buffer, + DTRACEOPT_BUFSIZE)) != 0) + return (rval); + + if ((rval = dtrace_state_buffer(state, state->dts_aggbuffer, + DTRACEOPT_AGGSIZE)) != 0) + return (rval); + + for (i = 0; i < state->dts_nspeculations; i++) { + if ((rval = dtrace_state_buffer(state, + spec[i].dtsp_buffer, DTRACEOPT_SPECSIZE)) != 0) + return (rval); + } + + return (0); +} + +static void +dtrace_state_prereserve(dtrace_state_t *state) +{ + dtrace_ecb_t *ecb; + dtrace_probe_t *probe; + + state->dts_reserve = 0; + + if (state->dts_options[DTRACEOPT_BUFPOLICY] != DTRACEOPT_BUFPOLICY_FILL) + return; + + /* + * If our buffer policy is a "fill" buffer policy, we need to set the + * prereserved space to be the space required by the END probes. + */ + probe = dtrace_probes[dtrace_probeid_end - 1]; + ASSERT(probe != NULL); + + for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) { + if (ecb->dte_state != state) + continue; + + state->dts_reserve += ecb->dte_needed + ecb->dte_alignment; + } +} + +static int +dtrace_state_go(dtrace_state_t *state, processorid_t *cpu) +{ + dtrace_optval_t *opt = state->dts_options, sz, nspec; + dtrace_speculation_t *spec; + dtrace_buffer_t *buf; +#if defined(sun) + cyc_handler_t hdlr; + cyc_time_t when; +#endif + int rval = 0, i, bufsize = NCPU * sizeof (dtrace_buffer_t); + dtrace_icookie_t cookie; + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_lock); + + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) { + rval = EBUSY; + goto out; + } + + /* + * Before we can perform any checks, we must prime all of the + * retained enablings that correspond to this state. + */ + dtrace_enabling_prime(state); + + if (state->dts_destructive && !state->dts_cred.dcr_destructive) { + rval = EACCES; + goto out; + } + + dtrace_state_prereserve(state); + + /* + * Now we want to do is try to allocate our speculations. + * We do not automatically resize the number of speculations; if + * this fails, we will fail the operation. + */ + nspec = opt[DTRACEOPT_NSPEC]; + ASSERT(nspec != DTRACEOPT_UNSET); + + if (nspec > INT_MAX) { + rval = ENOMEM; + goto out; + } + + spec = kmem_zalloc(nspec * sizeof (dtrace_speculation_t), KM_NOSLEEP); + + if (spec == NULL) { + rval = ENOMEM; + goto out; + } + + state->dts_speculations = spec; + state->dts_nspeculations = (int)nspec; + + for (i = 0; i < nspec; i++) { + if ((buf = kmem_zalloc(bufsize, KM_NOSLEEP)) == NULL) { + rval = ENOMEM; + goto err; + } + + spec[i].dtsp_buffer = buf; + } + + if (opt[DTRACEOPT_GRABANON] != DTRACEOPT_UNSET) { + if (dtrace_anon.dta_state == NULL) { + rval = ENOENT; + goto out; + } + + if (state->dts_necbs != 0) { + rval = EALREADY; + goto out; + } + + state->dts_anon = dtrace_anon_grab(); + ASSERT(state->dts_anon != NULL); + state = state->dts_anon; + + /* + * We want "grabanon" to be set in the grabbed state, so we'll + * copy that option value from the grabbing state into the + * grabbed state. + */ + state->dts_options[DTRACEOPT_GRABANON] = + opt[DTRACEOPT_GRABANON]; + + *cpu = dtrace_anon.dta_beganon; + + /* + * If the anonymous state is active (as it almost certainly + * is if the anonymous enabling ultimately matched anything), + * we don't allow any further option processing -- but we + * don't return failure. + */ + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) + goto out; + } + + if (opt[DTRACEOPT_AGGSIZE] != DTRACEOPT_UNSET && + opt[DTRACEOPT_AGGSIZE] != 0) { + if (state->dts_aggregations == NULL) { + /* + * We're not going to create an aggregation buffer + * because we don't have any ECBs that contain + * aggregations -- set this option to 0. + */ + opt[DTRACEOPT_AGGSIZE] = 0; + } else { + /* + * If we have an aggregation buffer, we must also have + * a buffer to use as scratch. + */ + if (opt[DTRACEOPT_BUFSIZE] == DTRACEOPT_UNSET || + opt[DTRACEOPT_BUFSIZE] < state->dts_needed) { + opt[DTRACEOPT_BUFSIZE] = state->dts_needed; + } + } + } + + if (opt[DTRACEOPT_SPECSIZE] != DTRACEOPT_UNSET && + opt[DTRACEOPT_SPECSIZE] != 0) { + if (!state->dts_speculates) { + /* + * We're not going to create speculation buffers + * because we don't have any ECBs that actually + * speculate -- set the speculation size to 0. + */ + opt[DTRACEOPT_SPECSIZE] = 0; + } + } + + /* + * The bare minimum size for any buffer that we're actually going to + * do anything to is sizeof (uint64_t). + */ + sz = sizeof (uint64_t); + + if ((state->dts_needed != 0 && opt[DTRACEOPT_BUFSIZE] < sz) || + (state->dts_speculates && opt[DTRACEOPT_SPECSIZE] < sz) || + (state->dts_aggregations != NULL && opt[DTRACEOPT_AGGSIZE] < sz)) { + /* + * A buffer size has been explicitly set to 0 (or to a size + * that will be adjusted to 0) and we need the space -- we + * need to return failure. We return ENOSPC to differentiate + * it from failing to allocate a buffer due to failure to meet + * the reserve (for which we return E2BIG). + */ + rval = ENOSPC; + goto out; + } + + if ((rval = dtrace_state_buffers(state)) != 0) + goto err; + + if ((sz = opt[DTRACEOPT_DYNVARSIZE]) == DTRACEOPT_UNSET) + sz = dtrace_dstate_defsize; + + do { + rval = dtrace_dstate_init(&state->dts_vstate.dtvs_dynvars, sz); + + if (rval == 0) + break; + + if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL) + goto err; + } while (sz >>= 1); + + opt[DTRACEOPT_DYNVARSIZE] = sz; + + if (rval != 0) + goto err; + + if (opt[DTRACEOPT_STATUSRATE] > dtrace_statusrate_max) + opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_max; + + if (opt[DTRACEOPT_CLEANRATE] == 0) + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max; + + if (opt[DTRACEOPT_CLEANRATE] < dtrace_cleanrate_min) + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_min; + + if (opt[DTRACEOPT_CLEANRATE] > dtrace_cleanrate_max) + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max; + + state->dts_alive = state->dts_laststatus = dtrace_gethrtime(); +#if defined(sun) + hdlr.cyh_func = (cyc_func_t)dtrace_state_clean; + hdlr.cyh_arg = state; + hdlr.cyh_level = CY_LOW_LEVEL; + + when.cyt_when = 0; + when.cyt_interval = opt[DTRACEOPT_CLEANRATE]; + + state->dts_cleaner = cyclic_add(&hdlr, &when); + + hdlr.cyh_func = (cyc_func_t)dtrace_state_deadman; + hdlr.cyh_arg = state; + hdlr.cyh_level = CY_LOW_LEVEL; + + when.cyt_when = 0; + when.cyt_interval = dtrace_deadman_interval; + + state->dts_deadman = cyclic_add(&hdlr, &when); +#else + callout_reset(&state->dts_cleaner, hz * opt[DTRACEOPT_CLEANRATE] / NANOSEC, + dtrace_state_clean, state); + callout_reset(&state->dts_deadman, hz * dtrace_deadman_interval / NANOSEC, + dtrace_state_deadman, state); +#endif + + state->dts_activity = DTRACE_ACTIVITY_WARMUP; + + /* + * Now it's time to actually fire the BEGIN probe. We need to disable + * interrupts here both to record the CPU on which we fired the BEGIN + * probe (the data from this CPU will be processed first at user + * level) and to manually activate the buffer for this CPU. + */ + cookie = dtrace_interrupt_disable(); + *cpu = curcpu; + ASSERT(state->dts_buffer[*cpu].dtb_flags & DTRACEBUF_INACTIVE); + state->dts_buffer[*cpu].dtb_flags &= ~DTRACEBUF_INACTIVE; + + dtrace_probe(dtrace_probeid_begin, + (uint64_t)(uintptr_t)state, 0, 0, 0, 0); + dtrace_interrupt_enable(cookie); + /* + * We may have had an exit action from a BEGIN probe; only change our + * state to ACTIVE if we're still in WARMUP. + */ + ASSERT(state->dts_activity == DTRACE_ACTIVITY_WARMUP || + state->dts_activity == DTRACE_ACTIVITY_DRAINING); + + if (state->dts_activity == DTRACE_ACTIVITY_WARMUP) + state->dts_activity = DTRACE_ACTIVITY_ACTIVE; + + /* + * Regardless of whether or not now we're in ACTIVE or DRAINING, we + * want each CPU to transition its principal buffer out of the + * INACTIVE state. Doing this assures that no CPU will suddenly begin + * processing an ECB halfway down a probe's ECB chain; all CPUs will + * atomically transition from processing none of a state's ECBs to + * processing all of them. + */ + dtrace_xcall(DTRACE_CPUALL, + (dtrace_xcall_t)dtrace_buffer_activate, state); + goto out; + +err: + dtrace_buffer_free(state->dts_buffer); + dtrace_buffer_free(state->dts_aggbuffer); + + if ((nspec = state->dts_nspeculations) == 0) { + ASSERT(state->dts_speculations == NULL); + goto out; + } + + spec = state->dts_speculations; + ASSERT(spec != NULL); + + for (i = 0; i < state->dts_nspeculations; i++) { + if ((buf = spec[i].dtsp_buffer) == NULL) + break; + + dtrace_buffer_free(buf); + kmem_free(buf, bufsize); + } + + kmem_free(spec, nspec * sizeof (dtrace_speculation_t)); + state->dts_nspeculations = 0; + state->dts_speculations = NULL; + +out: + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + + return (rval); +} + +static int +dtrace_state_stop(dtrace_state_t *state, processorid_t *cpu) +{ + dtrace_icookie_t cookie; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE && + state->dts_activity != DTRACE_ACTIVITY_DRAINING) + return (EINVAL); + + /* + * We'll set the activity to DTRACE_ACTIVITY_DRAINING, and issue a sync + * to be sure that every CPU has seen it. See below for the details + * on why this is done. + */ + state->dts_activity = DTRACE_ACTIVITY_DRAINING; + dtrace_sync(); + + /* + * By this point, it is impossible for any CPU to be still processing + * with DTRACE_ACTIVITY_ACTIVE. We can thus set our activity to + * DTRACE_ACTIVITY_COOLDOWN and know that we're not racing with any + * other CPU in dtrace_buffer_reserve(). This allows dtrace_probe() + * and callees to know that the activity is DTRACE_ACTIVITY_COOLDOWN + * iff we're in the END probe. + */ + state->dts_activity = DTRACE_ACTIVITY_COOLDOWN; + dtrace_sync(); + ASSERT(state->dts_activity == DTRACE_ACTIVITY_COOLDOWN); + + /* + * Finally, we can release the reserve and call the END probe. We + * disable interrupts across calling the END probe to allow us to + * return the CPU on which we actually called the END probe. This + * allows user-land to be sure that this CPU's principal buffer is + * processed last. + */ + state->dts_reserve = 0; + + cookie = dtrace_interrupt_disable(); + *cpu = curcpu; + dtrace_probe(dtrace_probeid_end, + (uint64_t)(uintptr_t)state, 0, 0, 0, 0); + dtrace_interrupt_enable(cookie); + + state->dts_activity = DTRACE_ACTIVITY_STOPPED; + dtrace_sync(); + + return (0); +} + +static int +dtrace_state_option(dtrace_state_t *state, dtrace_optid_t option, + dtrace_optval_t val) +{ + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) + return (EBUSY); + + if (option >= DTRACEOPT_MAX) + return (EINVAL); + + if (option != DTRACEOPT_CPU && val < 0) + return (EINVAL); + + switch (option) { + case DTRACEOPT_DESTRUCTIVE: + if (dtrace_destructive_disallow) + return (EACCES); + + state->dts_cred.dcr_destructive = 1; + break; + + case DTRACEOPT_BUFSIZE: + case DTRACEOPT_DYNVARSIZE: + case DTRACEOPT_AGGSIZE: + case DTRACEOPT_SPECSIZE: + case DTRACEOPT_STRSIZE: + if (val < 0) + return (EINVAL); + + if (val >= LONG_MAX) { + /* + * If this is an otherwise negative value, set it to + * the highest multiple of 128m less than LONG_MAX. + * Technically, we're adjusting the size without + * regard to the buffer resizing policy, but in fact, + * this has no effect -- if we set the buffer size to + * ~LONG_MAX and the buffer policy is ultimately set to + * be "manual", the buffer allocation is guaranteed to + * fail, if only because the allocation requires two + * buffers. (We set the the size to the highest + * multiple of 128m because it ensures that the size + * will remain a multiple of a megabyte when + * repeatedly halved -- all the way down to 15m.) + */ + val = LONG_MAX - (1 << 27) + 1; + } + } + + state->dts_options[option] = val; + + return (0); +} + +static void +dtrace_state_destroy(dtrace_state_t *state) +{ + dtrace_ecb_t *ecb; + dtrace_vstate_t *vstate = &state->dts_vstate; +#if defined(sun) + minor_t minor = getminor(state->dts_dev); +#endif + int i, bufsize = NCPU * sizeof (dtrace_buffer_t); + dtrace_speculation_t *spec = state->dts_speculations; + int nspec = state->dts_nspeculations; + uint32_t match; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(MUTEX_HELD(&cpu_lock)); + + /* + * First, retract any retained enablings for this state. + */ + dtrace_enabling_retract(state); + ASSERT(state->dts_nretained == 0); + + if (state->dts_activity == DTRACE_ACTIVITY_ACTIVE || + state->dts_activity == DTRACE_ACTIVITY_DRAINING) { + /* + * We have managed to come into dtrace_state_destroy() on a + * hot enabling -- almost certainly because of a disorderly + * shutdown of a consumer. (That is, a consumer that is + * exiting without having called dtrace_stop().) In this case, + * we're going to set our activity to be KILLED, and then + * issue a sync to be sure that everyone is out of probe + * context before we start blowing away ECBs. + */ + state->dts_activity = DTRACE_ACTIVITY_KILLED; + dtrace_sync(); + } + + /* + * Release the credential hold we took in dtrace_state_create(). + */ + if (state->dts_cred.dcr_cred != NULL) + crfree(state->dts_cred.dcr_cred); + + /* + * Now we can safely disable and destroy any enabled probes. Because + * any DTRACE_PRIV_KERNEL probes may actually be slowing our progress + * (especially if they're all enabled), we take two passes through the + * ECBs: in the first, we disable just DTRACE_PRIV_KERNEL probes, and + * in the second we disable whatever is left over. + */ + for (match = DTRACE_PRIV_KERNEL; ; match = 0) { + for (i = 0; i < state->dts_necbs; i++) { + if ((ecb = state->dts_ecbs[i]) == NULL) + continue; + + if (match && ecb->dte_probe != NULL) { + dtrace_probe_t *probe = ecb->dte_probe; + dtrace_provider_t *prov = probe->dtpr_provider; + + if (!(prov->dtpv_priv.dtpp_flags & match)) + continue; + } + + dtrace_ecb_disable(ecb); + dtrace_ecb_destroy(ecb); + } + + if (!match) + break; + } + + /* + * Before we free the buffers, perform one more sync to assure that + * every CPU is out of probe context. + */ + dtrace_sync(); + + dtrace_buffer_free(state->dts_buffer); + dtrace_buffer_free(state->dts_aggbuffer); + + for (i = 0; i < nspec; i++) + dtrace_buffer_free(spec[i].dtsp_buffer); + +#if defined(sun) + if (state->dts_cleaner != CYCLIC_NONE) + cyclic_remove(state->dts_cleaner); + + if (state->dts_deadman != CYCLIC_NONE) + cyclic_remove(state->dts_deadman); +#else + callout_drain(&state->dts_cleaner); + callout_drain(&state->dts_deadman); +#endif + + dtrace_dstate_fini(&vstate->dtvs_dynvars); + dtrace_vstate_fini(vstate); + if (state->dts_ecbs != NULL) + kmem_free(state->dts_ecbs, state->dts_necbs * sizeof (dtrace_ecb_t *)); + + if (state->dts_aggregations != NULL) { +#ifdef DEBUG + for (i = 0; i < state->dts_naggregations; i++) + ASSERT(state->dts_aggregations[i] == NULL); +#endif + ASSERT(state->dts_naggregations > 0); + kmem_free(state->dts_aggregations, + state->dts_naggregations * sizeof (dtrace_aggregation_t *)); + } + + kmem_free(state->dts_buffer, bufsize); + kmem_free(state->dts_aggbuffer, bufsize); + + for (i = 0; i < nspec; i++) + kmem_free(spec[i].dtsp_buffer, bufsize); + + if (spec != NULL) + kmem_free(spec, nspec * sizeof (dtrace_speculation_t)); + + dtrace_format_destroy(state); + + if (state->dts_aggid_arena != NULL) { +#if defined(sun) + vmem_destroy(state->dts_aggid_arena); +#else + delete_unrhdr(state->dts_aggid_arena); +#endif + state->dts_aggid_arena = NULL; + } +#if defined(sun) + ddi_soft_state_free(dtrace_softstate, minor); + vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1); +#endif +} + +/* + * DTrace Anonymous Enabling Functions + */ +static dtrace_state_t * +dtrace_anon_grab(void) +{ + dtrace_state_t *state; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if ((state = dtrace_anon.dta_state) == NULL) { + ASSERT(dtrace_anon.dta_enabling == NULL); + return (NULL); + } + + ASSERT(dtrace_anon.dta_enabling != NULL); + ASSERT(dtrace_retained != NULL); + + dtrace_enabling_destroy(dtrace_anon.dta_enabling); + dtrace_anon.dta_enabling = NULL; + dtrace_anon.dta_state = NULL; + + return (state); +} + +static void +dtrace_anon_property(void) +{ + int i, rv; + dtrace_state_t *state; + dof_hdr_t *dof; + char c[32]; /* enough for "dof-data-" + digits */ + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(MUTEX_HELD(&cpu_lock)); + + for (i = 0; ; i++) { + (void) snprintf(c, sizeof (c), "dof-data-%d", i); + + dtrace_err_verbose = 1; + + if ((dof = dtrace_dof_property(c)) == NULL) { + dtrace_err_verbose = 0; + break; + } + +#if defined(sun) + /* + * We want to create anonymous state, so we need to transition + * the kernel debugger to indicate that DTrace is active. If + * this fails (e.g. because the debugger has modified text in + * some way), we won't continue with the processing. + */ + if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) { + cmn_err(CE_NOTE, "kernel debugger active; anonymous " + "enabling ignored."); + dtrace_dof_destroy(dof); + break; + } +#endif + + /* + * If we haven't allocated an anonymous state, we'll do so now. + */ + if ((state = dtrace_anon.dta_state) == NULL) { +#if defined(sun) + state = dtrace_state_create(NULL, NULL); +#else + state = dtrace_state_create(NULL); +#endif + dtrace_anon.dta_state = state; + + if (state == NULL) { + /* + * This basically shouldn't happen: the only + * failure mode from dtrace_state_create() is a + * failure of ddi_soft_state_zalloc() that + * itself should never happen. Still, the + * interface allows for a failure mode, and + * we want to fail as gracefully as possible: + * we'll emit an error message and cease + * processing anonymous state in this case. + */ + cmn_err(CE_WARN, "failed to create " + "anonymous state"); + dtrace_dof_destroy(dof); + break; + } + } + + rv = dtrace_dof_slurp(dof, &state->dts_vstate, CRED(), + &dtrace_anon.dta_enabling, 0, B_TRUE); + + if (rv == 0) + rv = dtrace_dof_options(dof, state); + + dtrace_err_verbose = 0; + dtrace_dof_destroy(dof); + + if (rv != 0) { + /* + * This is malformed DOF; chuck any anonymous state + * that we created. + */ + ASSERT(dtrace_anon.dta_enabling == NULL); + dtrace_state_destroy(state); + dtrace_anon.dta_state = NULL; + break; + } + + ASSERT(dtrace_anon.dta_enabling != NULL); + } + + if (dtrace_anon.dta_enabling != NULL) { + int rval; + + /* + * dtrace_enabling_retain() can only fail because we are + * trying to retain more enablings than are allowed -- but + * we only have one anonymous enabling, and we are guaranteed + * to be allowed at least one retained enabling; we assert + * that dtrace_enabling_retain() returns success. + */ + rval = dtrace_enabling_retain(dtrace_anon.dta_enabling); + ASSERT(rval == 0); + + dtrace_enabling_dump(dtrace_anon.dta_enabling); + } +} + +#if defined(sun) +/* + * DTrace Helper Functions + */ +static void +dtrace_helper_trace(dtrace_helper_action_t *helper, + dtrace_mstate_t *mstate, dtrace_vstate_t *vstate, int where) +{ + uint32_t size, next, nnext, i; + dtrace_helptrace_t *ent; + uint16_t flags = cpu_core[curcpu].cpuc_dtrace_flags; + + if (!dtrace_helptrace_enabled) + return; + + ASSERT(vstate->dtvs_nlocals <= dtrace_helptrace_nlocals); + + /* + * What would a tracing framework be without its own tracing + * framework? (Well, a hell of a lot simpler, for starters...) + */ + size = sizeof (dtrace_helptrace_t) + dtrace_helptrace_nlocals * + sizeof (uint64_t) - sizeof (uint64_t); + + /* + * Iterate until we can allocate a slot in the trace buffer. + */ + do { + next = dtrace_helptrace_next; + + if (next + size < dtrace_helptrace_bufsize) { + nnext = next + size; + } else { + nnext = size; + } + } while (dtrace_cas32(&dtrace_helptrace_next, next, nnext) != next); + + /* + * We have our slot; fill it in. + */ + if (nnext == size) + next = 0; + + ent = (dtrace_helptrace_t *)&dtrace_helptrace_buffer[next]; + ent->dtht_helper = helper; + ent->dtht_where = where; + ent->dtht_nlocals = vstate->dtvs_nlocals; + + ent->dtht_fltoffs = (mstate->dtms_present & DTRACE_MSTATE_FLTOFFS) ? + mstate->dtms_fltoffs : -1; + ent->dtht_fault = DTRACE_FLAGS2FLT(flags); + ent->dtht_illval = cpu_core[curcpu].cpuc_dtrace_illval; + + for (i = 0; i < vstate->dtvs_nlocals; i++) { + dtrace_statvar_t *svar; + + if ((svar = vstate->dtvs_locals[i]) == NULL) + continue; + + ASSERT(svar->dtsv_size >= NCPU * sizeof (uint64_t)); + ent->dtht_locals[i] = + ((uint64_t *)(uintptr_t)svar->dtsv_data)[curcpu]; + } +} +#endif + +#if defined(sun) +static uint64_t +dtrace_helper(int which, dtrace_mstate_t *mstate, + dtrace_state_t *state, uint64_t arg0, uint64_t arg1) +{ + uint16_t *flags = &cpu_core[curcpu].cpuc_dtrace_flags; + uint64_t sarg0 = mstate->dtms_arg[0]; + uint64_t sarg1 = mstate->dtms_arg[1]; + uint64_t rval; + dtrace_helpers_t *helpers = curproc->p_dtrace_helpers; + dtrace_helper_action_t *helper; + dtrace_vstate_t *vstate; + dtrace_difo_t *pred; + int i, trace = dtrace_helptrace_enabled; + + ASSERT(which >= 0 && which < DTRACE_NHELPER_ACTIONS); + + if (helpers == NULL) + return (0); + + if ((helper = helpers->dthps_actions[which]) == NULL) + return (0); + + vstate = &helpers->dthps_vstate; + mstate->dtms_arg[0] = arg0; + mstate->dtms_arg[1] = arg1; + + /* + * Now iterate over each helper. If its predicate evaluates to 'true', + * we'll call the corresponding actions. Note that the below calls + * to dtrace_dif_emulate() may set faults in machine state. This is + * okay: our caller (the outer dtrace_dif_emulate()) will simply plow + * the stored DIF offset with its own (which is the desired behavior). + * Also, note the calls to dtrace_dif_emulate() may allocate scratch + * from machine state; this is okay, too. + */ + for (; helper != NULL; helper = helper->dtha_next) { + if ((pred = helper->dtha_predicate) != NULL) { + if (trace) + dtrace_helper_trace(helper, mstate, vstate, 0); + + if (!dtrace_dif_emulate(pred, mstate, vstate, state)) + goto next; + + if (*flags & CPU_DTRACE_FAULT) + goto err; + } + + for (i = 0; i < helper->dtha_nactions; i++) { + if (trace) + dtrace_helper_trace(helper, + mstate, vstate, i + 1); + + rval = dtrace_dif_emulate(helper->dtha_actions[i], + mstate, vstate, state); + + if (*flags & CPU_DTRACE_FAULT) + goto err; + } + +next: + if (trace) + dtrace_helper_trace(helper, mstate, vstate, + DTRACE_HELPTRACE_NEXT); + } + + if (trace) + dtrace_helper_trace(helper, mstate, vstate, + DTRACE_HELPTRACE_DONE); + + /* + * Restore the arg0 that we saved upon entry. + */ + mstate->dtms_arg[0] = sarg0; + mstate->dtms_arg[1] = sarg1; + + return (rval); + +err: + if (trace) + dtrace_helper_trace(helper, mstate, vstate, + DTRACE_HELPTRACE_ERR); + + /* + * Restore the arg0 that we saved upon entry. + */ + mstate->dtms_arg[0] = sarg0; + mstate->dtms_arg[1] = sarg1; + + return (0); +} + +static void +dtrace_helper_action_destroy(dtrace_helper_action_t *helper, + dtrace_vstate_t *vstate) +{ + int i; + + if (helper->dtha_predicate != NULL) + dtrace_difo_release(helper->dtha_predicate, vstate); + + for (i = 0; i < helper->dtha_nactions; i++) { + ASSERT(helper->dtha_actions[i] != NULL); + dtrace_difo_release(helper->dtha_actions[i], vstate); + } + + kmem_free(helper->dtha_actions, + helper->dtha_nactions * sizeof (dtrace_difo_t *)); + kmem_free(helper, sizeof (dtrace_helper_action_t)); +} + +static int +dtrace_helper_destroygen(int gen) +{ + proc_t *p = curproc; + dtrace_helpers_t *help = p->p_dtrace_helpers; + dtrace_vstate_t *vstate; + int i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (help == NULL || gen > help->dthps_generation) + return (EINVAL); + + vstate = &help->dthps_vstate; + + for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { + dtrace_helper_action_t *last = NULL, *h, *next; + + for (h = help->dthps_actions[i]; h != NULL; h = next) { + next = h->dtha_next; + + if (h->dtha_generation == gen) { + if (last != NULL) { + last->dtha_next = next; + } else { + help->dthps_actions[i] = next; + } + + dtrace_helper_action_destroy(h, vstate); + } else { + last = h; + } + } + } + + /* + * Interate until we've cleared out all helper providers with the + * given generation number. + */ + for (;;) { + dtrace_helper_provider_t *prov; + + /* + * Look for a helper provider with the right generation. We + * have to start back at the beginning of the list each time + * because we drop dtrace_lock. It's unlikely that we'll make + * more than two passes. + */ + for (i = 0; i < help->dthps_nprovs; i++) { + prov = help->dthps_provs[i]; + + if (prov->dthp_generation == gen) + break; + } + + /* + * If there were no matches, we're done. + */ + if (i == help->dthps_nprovs) + break; + + /* + * Move the last helper provider into this slot. + */ + help->dthps_nprovs--; + help->dthps_provs[i] = help->dthps_provs[help->dthps_nprovs]; + help->dthps_provs[help->dthps_nprovs] = NULL; + + mutex_exit(&dtrace_lock); + + /* + * If we have a meta provider, remove this helper provider. + */ + mutex_enter(&dtrace_meta_lock); + if (dtrace_meta_pid != NULL) { + ASSERT(dtrace_deferred_pid == NULL); + dtrace_helper_provider_remove(&prov->dthp_prov, + p->p_pid); + } + mutex_exit(&dtrace_meta_lock); + + dtrace_helper_provider_destroy(prov); + + mutex_enter(&dtrace_lock); + } + + return (0); +} +#endif + +#if defined(sun) +static int +dtrace_helper_validate(dtrace_helper_action_t *helper) +{ + int err = 0, i; + dtrace_difo_t *dp; + + if ((dp = helper->dtha_predicate) != NULL) + err += dtrace_difo_validate_helper(dp); + + for (i = 0; i < helper->dtha_nactions; i++) + err += dtrace_difo_validate_helper(helper->dtha_actions[i]); + + return (err == 0); +} +#endif + +#if defined(sun) +static int +dtrace_helper_action_add(int which, dtrace_ecbdesc_t *ep) +{ + dtrace_helpers_t *help; + dtrace_helper_action_t *helper, *last; + dtrace_actdesc_t *act; + dtrace_vstate_t *vstate; + dtrace_predicate_t *pred; + int count = 0, nactions = 0, i; + + if (which < 0 || which >= DTRACE_NHELPER_ACTIONS) + return (EINVAL); + + help = curproc->p_dtrace_helpers; + last = help->dthps_actions[which]; + vstate = &help->dthps_vstate; + + for (count = 0; last != NULL; last = last->dtha_next) { + count++; + if (last->dtha_next == NULL) + break; + } + + /* + * If we already have dtrace_helper_actions_max helper actions for this + * helper action type, we'll refuse to add a new one. + */ + if (count >= dtrace_helper_actions_max) + return (ENOSPC); + + helper = kmem_zalloc(sizeof (dtrace_helper_action_t), KM_SLEEP); + helper->dtha_generation = help->dthps_generation; + + if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) { + ASSERT(pred->dtp_difo != NULL); + dtrace_difo_hold(pred->dtp_difo); + helper->dtha_predicate = pred->dtp_difo; + } + + for (act = ep->dted_action; act != NULL; act = act->dtad_next) { + if (act->dtad_kind != DTRACEACT_DIFEXPR) + goto err; + + if (act->dtad_difo == NULL) + goto err; + + nactions++; + } + + helper->dtha_actions = kmem_zalloc(sizeof (dtrace_difo_t *) * + (helper->dtha_nactions = nactions), KM_SLEEP); + + for (act = ep->dted_action, i = 0; act != NULL; act = act->dtad_next) { + dtrace_difo_hold(act->dtad_difo); + helper->dtha_actions[i++] = act->dtad_difo; + } + + if (!dtrace_helper_validate(helper)) + goto err; + + if (last == NULL) { + help->dthps_actions[which] = helper; + } else { + last->dtha_next = helper; + } + + if (vstate->dtvs_nlocals > dtrace_helptrace_nlocals) { + dtrace_helptrace_nlocals = vstate->dtvs_nlocals; + dtrace_helptrace_next = 0; + } + + return (0); +err: + dtrace_helper_action_destroy(helper, vstate); + return (EINVAL); +} + +static void +dtrace_helper_provider_register(proc_t *p, dtrace_helpers_t *help, + dof_helper_t *dofhp) +{ + ASSERT(MUTEX_NOT_HELD(&dtrace_lock)); + + mutex_enter(&dtrace_meta_lock); + mutex_enter(&dtrace_lock); + + if (!dtrace_attached() || dtrace_meta_pid == NULL) { + /* + * If the dtrace module is loaded but not attached, or if + * there aren't isn't a meta provider registered to deal with + * these provider descriptions, we need to postpone creating + * the actual providers until later. + */ + + if (help->dthps_next == NULL && help->dthps_prev == NULL && + dtrace_deferred_pid != help) { + help->dthps_deferred = 1; + help->dthps_pid = p->p_pid; + help->dthps_next = dtrace_deferred_pid; + help->dthps_prev = NULL; + if (dtrace_deferred_pid != NULL) + dtrace_deferred_pid->dthps_prev = help; + dtrace_deferred_pid = help; + } + + mutex_exit(&dtrace_lock); + + } else if (dofhp != NULL) { + /* + * If the dtrace module is loaded and we have a particular + * helper provider description, pass that off to the + * meta provider. + */ + + mutex_exit(&dtrace_lock); + + dtrace_helper_provide(dofhp, p->p_pid); + + } else { + /* + * Otherwise, just pass all the helper provider descriptions + * off to the meta provider. + */ + + int i; + mutex_exit(&dtrace_lock); + + for (i = 0; i < help->dthps_nprovs; i++) { + dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov, + p->p_pid); + } + } + + mutex_exit(&dtrace_meta_lock); +} + +static int +dtrace_helper_provider_add(dof_helper_t *dofhp, int gen) +{ + dtrace_helpers_t *help; + dtrace_helper_provider_t *hprov, **tmp_provs; + uint_t tmp_maxprovs, i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + help = curproc->p_dtrace_helpers; + ASSERT(help != NULL); + + /* + * If we already have dtrace_helper_providers_max helper providers, + * we're refuse to add a new one. + */ + if (help->dthps_nprovs >= dtrace_helper_providers_max) + return (ENOSPC); + + /* + * Check to make sure this isn't a duplicate. + */ + for (i = 0; i < help->dthps_nprovs; i++) { + if (dofhp->dofhp_addr == + help->dthps_provs[i]->dthp_prov.dofhp_addr) + return (EALREADY); + } + + hprov = kmem_zalloc(sizeof (dtrace_helper_provider_t), KM_SLEEP); + hprov->dthp_prov = *dofhp; + hprov->dthp_ref = 1; + hprov->dthp_generation = gen; + + /* + * Allocate a bigger table for helper providers if it's already full. + */ + if (help->dthps_maxprovs == help->dthps_nprovs) { + tmp_maxprovs = help->dthps_maxprovs; + tmp_provs = help->dthps_provs; + + if (help->dthps_maxprovs == 0) + help->dthps_maxprovs = 2; + else + help->dthps_maxprovs *= 2; + if (help->dthps_maxprovs > dtrace_helper_providers_max) + help->dthps_maxprovs = dtrace_helper_providers_max; + + ASSERT(tmp_maxprovs < help->dthps_maxprovs); + + help->dthps_provs = kmem_zalloc(help->dthps_maxprovs * + sizeof (dtrace_helper_provider_t *), KM_SLEEP); + + if (tmp_provs != NULL) { + bcopy(tmp_provs, help->dthps_provs, tmp_maxprovs * + sizeof (dtrace_helper_provider_t *)); + kmem_free(tmp_provs, tmp_maxprovs * + sizeof (dtrace_helper_provider_t *)); + } + } + + help->dthps_provs[help->dthps_nprovs] = hprov; + help->dthps_nprovs++; + + return (0); +} + +static void +dtrace_helper_provider_destroy(dtrace_helper_provider_t *hprov) +{ + mutex_enter(&dtrace_lock); + + if (--hprov->dthp_ref == 0) { + dof_hdr_t *dof; + mutex_exit(&dtrace_lock); + dof = (dof_hdr_t *)(uintptr_t)hprov->dthp_prov.dofhp_dof; + dtrace_dof_destroy(dof); + kmem_free(hprov, sizeof (dtrace_helper_provider_t)); + } else { + mutex_exit(&dtrace_lock); + } +} + +static int +dtrace_helper_provider_validate(dof_hdr_t *dof, dof_sec_t *sec) +{ + uintptr_t daddr = (uintptr_t)dof; + dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec; + dof_provider_t *provider; + dof_probe_t *probe; + uint8_t *arg; + char *strtab, *typestr; + dof_stridx_t typeidx; + size_t typesz; + uint_t nprobes, j, k; + + ASSERT(sec->dofs_type == DOF_SECT_PROVIDER); + + if (sec->dofs_offset & (sizeof (uint_t) - 1)) { + dtrace_dof_error(dof, "misaligned section offset"); + return (-1); + } + + /* + * The section needs to be large enough to contain the DOF provider + * structure appropriate for the given version. + */ + if (sec->dofs_size < + ((dof->dofh_ident[DOF_ID_VERSION] == DOF_VERSION_1) ? + offsetof(dof_provider_t, dofpv_prenoffs) : + sizeof (dof_provider_t))) { + dtrace_dof_error(dof, "provider section too small"); + return (-1); + } + + provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); + str_sec = dtrace_dof_sect(dof, DOF_SECT_STRTAB, provider->dofpv_strtab); + prb_sec = dtrace_dof_sect(dof, DOF_SECT_PROBES, provider->dofpv_probes); + arg_sec = dtrace_dof_sect(dof, DOF_SECT_PRARGS, provider->dofpv_prargs); + off_sec = dtrace_dof_sect(dof, DOF_SECT_PROFFS, provider->dofpv_proffs); + + if (str_sec == NULL || prb_sec == NULL || + arg_sec == NULL || off_sec == NULL) + return (-1); + + enoff_sec = NULL; + + if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && + provider->dofpv_prenoffs != DOF_SECT_NONE && + (enoff_sec = dtrace_dof_sect(dof, DOF_SECT_PRENOFFS, + provider->dofpv_prenoffs)) == NULL) + return (-1); + + strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); + + if (provider->dofpv_name >= str_sec->dofs_size || + strlen(strtab + provider->dofpv_name) >= DTRACE_PROVNAMELEN) { + dtrace_dof_error(dof, "invalid provider name"); + return (-1); + } + + if (prb_sec->dofs_entsize == 0 || + prb_sec->dofs_entsize > prb_sec->dofs_size) { + dtrace_dof_error(dof, "invalid entry size"); + return (-1); + } + + if (prb_sec->dofs_entsize & (sizeof (uintptr_t) - 1)) { + dtrace_dof_error(dof, "misaligned entry size"); + return (-1); + } + + if (off_sec->dofs_entsize != sizeof (uint32_t)) { + dtrace_dof_error(dof, "invalid entry size"); + return (-1); + } + + if (off_sec->dofs_offset & (sizeof (uint32_t) - 1)) { + dtrace_dof_error(dof, "misaligned section offset"); + return (-1); + } + + if (arg_sec->dofs_entsize != sizeof (uint8_t)) { + dtrace_dof_error(dof, "invalid entry size"); + return (-1); + } + + arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset); + + nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize; + + /* + * Take a pass through the probes to check for errors. + */ + for (j = 0; j < nprobes; j++) { + probe = (dof_probe_t *)(uintptr_t)(daddr + + prb_sec->dofs_offset + j * prb_sec->dofs_entsize); + + if (probe->dofpr_func >= str_sec->dofs_size) { + dtrace_dof_error(dof, "invalid function name"); + return (-1); + } + + if (strlen(strtab + probe->dofpr_func) >= DTRACE_FUNCNAMELEN) { + dtrace_dof_error(dof, "function name too long"); + return (-1); + } + + if (probe->dofpr_name >= str_sec->dofs_size || + strlen(strtab + probe->dofpr_name) >= DTRACE_NAMELEN) { + dtrace_dof_error(dof, "invalid probe name"); + return (-1); + } + + /* + * The offset count must not wrap the index, and the offsets + * must also not overflow the section's data. + */ + if (probe->dofpr_offidx + probe->dofpr_noffs < + probe->dofpr_offidx || + (probe->dofpr_offidx + probe->dofpr_noffs) * + off_sec->dofs_entsize > off_sec->dofs_size) { + dtrace_dof_error(dof, "invalid probe offset"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1) { + /* + * If there's no is-enabled offset section, make sure + * there aren't any is-enabled offsets. Otherwise + * perform the same checks as for probe offsets + * (immediately above). + */ + if (enoff_sec == NULL) { + if (probe->dofpr_enoffidx != 0 || + probe->dofpr_nenoffs != 0) { + dtrace_dof_error(dof, "is-enabled " + "offsets with null section"); + return (-1); + } + } else if (probe->dofpr_enoffidx + + probe->dofpr_nenoffs < probe->dofpr_enoffidx || + (probe->dofpr_enoffidx + probe->dofpr_nenoffs) * + enoff_sec->dofs_entsize > enoff_sec->dofs_size) { + dtrace_dof_error(dof, "invalid is-enabled " + "offset"); + return (-1); + } + + if (probe->dofpr_noffs + probe->dofpr_nenoffs == 0) { + dtrace_dof_error(dof, "zero probe and " + "is-enabled offsets"); + return (-1); + } + } else if (probe->dofpr_noffs == 0) { + dtrace_dof_error(dof, "zero probe offsets"); + return (-1); + } + + if (probe->dofpr_argidx + probe->dofpr_xargc < + probe->dofpr_argidx || + (probe->dofpr_argidx + probe->dofpr_xargc) * + arg_sec->dofs_entsize > arg_sec->dofs_size) { + dtrace_dof_error(dof, "invalid args"); + return (-1); + } + + typeidx = probe->dofpr_nargv; + typestr = strtab + probe->dofpr_nargv; + for (k = 0; k < probe->dofpr_nargc; k++) { + if (typeidx >= str_sec->dofs_size) { + dtrace_dof_error(dof, "bad " + "native argument type"); + return (-1); + } + + typesz = strlen(typestr) + 1; + if (typesz > DTRACE_ARGTYPELEN) { + dtrace_dof_error(dof, "native " + "argument type too long"); + return (-1); + } + typeidx += typesz; + typestr += typesz; + } + + typeidx = probe->dofpr_xargv; + typestr = strtab + probe->dofpr_xargv; + for (k = 0; k < probe->dofpr_xargc; k++) { + if (arg[probe->dofpr_argidx + k] > probe->dofpr_nargc) { + dtrace_dof_error(dof, "bad " + "native argument index"); + return (-1); + } + + if (typeidx >= str_sec->dofs_size) { + dtrace_dof_error(dof, "bad " + "translated argument type"); + return (-1); + } + + typesz = strlen(typestr) + 1; + if (typesz > DTRACE_ARGTYPELEN) { + dtrace_dof_error(dof, "translated argument " + "type too long"); + return (-1); + } + + typeidx += typesz; + typestr += typesz; + } + } + + return (0); +} + +static int +dtrace_helper_slurp(dof_hdr_t *dof, dof_helper_t *dhp) +{ + dtrace_helpers_t *help; + dtrace_vstate_t *vstate; + dtrace_enabling_t *enab = NULL; + int i, gen, rv, nhelpers = 0, nprovs = 0, destroy = 1; + uintptr_t daddr = (uintptr_t)dof; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if ((help = curproc->p_dtrace_helpers) == NULL) + help = dtrace_helpers_create(curproc); + + vstate = &help->dthps_vstate; + + if ((rv = dtrace_dof_slurp(dof, vstate, NULL, &enab, + dhp != NULL ? dhp->dofhp_addr : 0, B_FALSE)) != 0) { + dtrace_dof_destroy(dof); + return (rv); + } + + /* + * Look for helper providers and validate their descriptions. + */ + if (dhp != NULL) { + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + + dof->dofh_secoff + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_PROVIDER) + continue; + + if (dtrace_helper_provider_validate(dof, sec) != 0) { + dtrace_enabling_destroy(enab); + dtrace_dof_destroy(dof); + return (-1); + } + + nprovs++; + } + } + + /* + * Now we need to walk through the ECB descriptions in the enabling. + */ + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_ecbdesc_t *ep = enab->dten_desc[i]; + dtrace_probedesc_t *desc = &ep->dted_probe; + + if (strcmp(desc->dtpd_provider, "dtrace") != 0) + continue; + + if (strcmp(desc->dtpd_mod, "helper") != 0) + continue; + + if (strcmp(desc->dtpd_func, "ustack") != 0) + continue; + + if ((rv = dtrace_helper_action_add(DTRACE_HELPER_ACTION_USTACK, + ep)) != 0) { + /* + * Adding this helper action failed -- we are now going + * to rip out the entire generation and return failure. + */ + (void) dtrace_helper_destroygen(help->dthps_generation); + dtrace_enabling_destroy(enab); + dtrace_dof_destroy(dof); + return (-1); + } + + nhelpers++; + } + + if (nhelpers < enab->dten_ndesc) + dtrace_dof_error(dof, "unmatched helpers"); + + gen = help->dthps_generation++; + dtrace_enabling_destroy(enab); + + if (dhp != NULL && nprovs > 0) { + dhp->dofhp_dof = (uint64_t)(uintptr_t)dof; + if (dtrace_helper_provider_add(dhp, gen) == 0) { + mutex_exit(&dtrace_lock); + dtrace_helper_provider_register(curproc, help, dhp); + mutex_enter(&dtrace_lock); + + destroy = 0; + } + } + + if (destroy) + dtrace_dof_destroy(dof); + + return (gen); +} + +static dtrace_helpers_t * +dtrace_helpers_create(proc_t *p) +{ + dtrace_helpers_t *help; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(p->p_dtrace_helpers == NULL); + + help = kmem_zalloc(sizeof (dtrace_helpers_t), KM_SLEEP); + help->dthps_actions = kmem_zalloc(sizeof (dtrace_helper_action_t *) * + DTRACE_NHELPER_ACTIONS, KM_SLEEP); + + p->p_dtrace_helpers = help; + dtrace_helpers++; + + return (help); +} + +static void +dtrace_helpers_destroy(void) +{ + dtrace_helpers_t *help; + dtrace_vstate_t *vstate; + proc_t *p = curproc; + int i; + + mutex_enter(&dtrace_lock); + + ASSERT(p->p_dtrace_helpers != NULL); + ASSERT(dtrace_helpers > 0); + + help = p->p_dtrace_helpers; + vstate = &help->dthps_vstate; + + /* + * We're now going to lose the help from this process. + */ + p->p_dtrace_helpers = NULL; + dtrace_sync(); + + /* + * Destory the helper actions. + */ + for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { + dtrace_helper_action_t *h, *next; + + for (h = help->dthps_actions[i]; h != NULL; h = next) { + next = h->dtha_next; + dtrace_helper_action_destroy(h, vstate); + h = next; + } + } + + mutex_exit(&dtrace_lock); + + /* + * Destroy the helper providers. + */ + if (help->dthps_maxprovs > 0) { + mutex_enter(&dtrace_meta_lock); + if (dtrace_meta_pid != NULL) { + ASSERT(dtrace_deferred_pid == NULL); + + for (i = 0; i < help->dthps_nprovs; i++) { + dtrace_helper_provider_remove( + &help->dthps_provs[i]->dthp_prov, p->p_pid); + } + } else { + mutex_enter(&dtrace_lock); + ASSERT(help->dthps_deferred == 0 || + help->dthps_next != NULL || + help->dthps_prev != NULL || + help == dtrace_deferred_pid); + + /* + * Remove the helper from the deferred list. + */ + if (help->dthps_next != NULL) + help->dthps_next->dthps_prev = help->dthps_prev; + if (help->dthps_prev != NULL) + help->dthps_prev->dthps_next = help->dthps_next; + if (dtrace_deferred_pid == help) { + dtrace_deferred_pid = help->dthps_next; + ASSERT(help->dthps_prev == NULL); + } + + mutex_exit(&dtrace_lock); + } + + mutex_exit(&dtrace_meta_lock); + + for (i = 0; i < help->dthps_nprovs; i++) { + dtrace_helper_provider_destroy(help->dthps_provs[i]); + } + + kmem_free(help->dthps_provs, help->dthps_maxprovs * + sizeof (dtrace_helper_provider_t *)); + } + + mutex_enter(&dtrace_lock); + + dtrace_vstate_fini(&help->dthps_vstate); + kmem_free(help->dthps_actions, + sizeof (dtrace_helper_action_t *) * DTRACE_NHELPER_ACTIONS); + kmem_free(help, sizeof (dtrace_helpers_t)); + + --dtrace_helpers; + mutex_exit(&dtrace_lock); +} + +static void +dtrace_helpers_duplicate(proc_t *from, proc_t *to) +{ + dtrace_helpers_t *help, *newhelp; + dtrace_helper_action_t *helper, *new, *last; + dtrace_difo_t *dp; + dtrace_vstate_t *vstate; + int i, j, sz, hasprovs = 0; + + mutex_enter(&dtrace_lock); + ASSERT(from->p_dtrace_helpers != NULL); + ASSERT(dtrace_helpers > 0); + + help = from->p_dtrace_helpers; + newhelp = dtrace_helpers_create(to); + ASSERT(to->p_dtrace_helpers != NULL); + + newhelp->dthps_generation = help->dthps_generation; + vstate = &newhelp->dthps_vstate; + + /* + * Duplicate the helper actions. + */ + for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { + if ((helper = help->dthps_actions[i]) == NULL) + continue; + + for (last = NULL; helper != NULL; helper = helper->dtha_next) { + new = kmem_zalloc(sizeof (dtrace_helper_action_t), + KM_SLEEP); + new->dtha_generation = helper->dtha_generation; + + if ((dp = helper->dtha_predicate) != NULL) { + dp = dtrace_difo_duplicate(dp, vstate); + new->dtha_predicate = dp; + } + + new->dtha_nactions = helper->dtha_nactions; + sz = sizeof (dtrace_difo_t *) * new->dtha_nactions; + new->dtha_actions = kmem_alloc(sz, KM_SLEEP); + + for (j = 0; j < new->dtha_nactions; j++) { + dtrace_difo_t *dp = helper->dtha_actions[j]; + + ASSERT(dp != NULL); + dp = dtrace_difo_duplicate(dp, vstate); + new->dtha_actions[j] = dp; + } + + if (last != NULL) { + last->dtha_next = new; + } else { + newhelp->dthps_actions[i] = new; + } + + last = new; + } + } + + /* + * Duplicate the helper providers and register them with the + * DTrace framework. + */ + if (help->dthps_nprovs > 0) { + newhelp->dthps_nprovs = help->dthps_nprovs; + newhelp->dthps_maxprovs = help->dthps_nprovs; + newhelp->dthps_provs = kmem_alloc(newhelp->dthps_nprovs * + sizeof (dtrace_helper_provider_t *), KM_SLEEP); + for (i = 0; i < newhelp->dthps_nprovs; i++) { + newhelp->dthps_provs[i] = help->dthps_provs[i]; + newhelp->dthps_provs[i]->dthp_ref++; + } + + hasprovs = 1; + } + + mutex_exit(&dtrace_lock); + + if (hasprovs) + dtrace_helper_provider_register(to, newhelp, NULL); +} +#endif + +#if defined(sun) +/* + * DTrace Hook Functions + */ +static void +dtrace_module_loaded(modctl_t *ctl) +{ + dtrace_provider_t *prv; + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&mod_lock); + + ASSERT(ctl->mod_busy); + + /* + * We're going to call each providers per-module provide operation + * specifying only this module. + */ + for (prv = dtrace_provider; prv != NULL; prv = prv->dtpv_next) + prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl); + + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + + /* + * If we have any retained enablings, we need to match against them. + * Enabling probes requires that cpu_lock be held, and we cannot hold + * cpu_lock here -- it is legal for cpu_lock to be held when loading a + * module. (In particular, this happens when loading scheduling + * classes.) So if we have any retained enablings, we need to dispatch + * our task queue to do the match for us. + */ + mutex_enter(&dtrace_lock); + + if (dtrace_retained == NULL) { + mutex_exit(&dtrace_lock); + return; + } + + (void) taskq_dispatch(dtrace_taskq, + (task_func_t *)dtrace_enabling_matchall, NULL, TQ_SLEEP); + + mutex_exit(&dtrace_lock); + + /* + * And now, for a little heuristic sleaze: in general, we want to + * match modules as soon as they load. However, we cannot guarantee + * this, because it would lead us to the lock ordering violation + * outlined above. The common case, of course, is that cpu_lock is + * _not_ held -- so we delay here for a clock tick, hoping that that's + * long enough for the task queue to do its work. If it's not, it's + * not a serious problem -- it just means that the module that we + * just loaded may not be immediately instrumentable. + */ + delay(1); +} + +static void +dtrace_module_unloaded(modctl_t *ctl) +{ + dtrace_probe_t template, *probe, *first, *next; + dtrace_provider_t *prov; + + template.dtpr_mod = ctl->mod_modname; + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&mod_lock); + mutex_enter(&dtrace_lock); + + if (dtrace_bymod == NULL) { + /* + * The DTrace module is loaded (obviously) but not attached; + * we don't have any work to do. + */ + mutex_exit(&dtrace_provider_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_lock); + return; + } + + for (probe = first = dtrace_hash_lookup(dtrace_bymod, &template); + probe != NULL; probe = probe->dtpr_nextmod) { + if (probe->dtpr_ecb != NULL) { + mutex_exit(&dtrace_provider_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_lock); + + /* + * This shouldn't _actually_ be possible -- we're + * unloading a module that has an enabled probe in it. + * (It's normally up to the provider to make sure that + * this can't happen.) However, because dtps_enable() + * doesn't have a failure mode, there can be an + * enable/unload race. Upshot: we don't want to + * assert, but we're not going to disable the + * probe, either. + */ + if (dtrace_err_verbose) { + cmn_err(CE_WARN, "unloaded module '%s' had " + "enabled probes", ctl->mod_modname); + } + + return; + } + } + + probe = first; + + for (first = NULL; probe != NULL; probe = next) { + ASSERT(dtrace_probes[probe->dtpr_id - 1] == probe); + + dtrace_probes[probe->dtpr_id - 1] = NULL; + + next = probe->dtpr_nextmod; + dtrace_hash_remove(dtrace_bymod, probe); + dtrace_hash_remove(dtrace_byfunc, probe); + dtrace_hash_remove(dtrace_byname, probe); + + if (first == NULL) { + first = probe; + probe->dtpr_nextmod = NULL; + } else { + probe->dtpr_nextmod = first; + first = probe; + } + } + + /* + * We've removed all of the module's probes from the hash chains and + * from the probe array. Now issue a dtrace_sync() to be sure that + * everyone has cleared out from any probe array processing. + */ + dtrace_sync(); + + for (probe = first; probe != NULL; probe = first) { + first = probe->dtpr_nextmod; + prov = probe->dtpr_provider; + prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, probe->dtpr_id, + probe->dtpr_arg); + kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); + kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); + kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); + vmem_free(dtrace_arena, (void *)(uintptr_t)probe->dtpr_id, 1); + kmem_free(probe, sizeof (dtrace_probe_t)); + } + + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); +} + +static void +dtrace_suspend(void) +{ + dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_suspend)); +} + +static void +dtrace_resume(void) +{ + dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_resume)); +} +#endif + +static int +dtrace_cpu_setup(cpu_setup_t what, processorid_t cpu) +{ + ASSERT(MUTEX_HELD(&cpu_lock)); + mutex_enter(&dtrace_lock); + + switch (what) { + case CPU_CONFIG: { + dtrace_state_t *state; + dtrace_optval_t *opt, rs, c; + + /* + * For now, we only allocate a new buffer for anonymous state. + */ + if ((state = dtrace_anon.dta_state) == NULL) + break; + + if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) + break; + + opt = state->dts_options; + c = opt[DTRACEOPT_CPU]; + + if (c != DTRACE_CPUALL && c != DTRACEOPT_UNSET && c != cpu) + break; + + /* + * Regardless of what the actual policy is, we're going to + * temporarily set our resize policy to be manual. We're + * also going to temporarily set our CPU option to denote + * the newly configured CPU. + */ + rs = opt[DTRACEOPT_BUFRESIZE]; + opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_MANUAL; + opt[DTRACEOPT_CPU] = (dtrace_optval_t)cpu; + + (void) dtrace_state_buffers(state); + + opt[DTRACEOPT_BUFRESIZE] = rs; + opt[DTRACEOPT_CPU] = c; + + break; + } + + case CPU_UNCONFIG: + /* + * We don't free the buffer in the CPU_UNCONFIG case. (The + * buffer will be freed when the consumer exits.) + */ + break; + + default: + break; + } + + mutex_exit(&dtrace_lock); + return (0); +} + +#if defined(sun) +static void +dtrace_cpu_setup_initial(processorid_t cpu) +{ + (void) dtrace_cpu_setup(CPU_CONFIG, cpu); +} +#endif + +static void +dtrace_toxrange_add(uintptr_t base, uintptr_t limit) +{ + if (dtrace_toxranges >= dtrace_toxranges_max) { + int osize, nsize; + dtrace_toxrange_t *range; + + osize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t); + + if (osize == 0) { + ASSERT(dtrace_toxrange == NULL); + ASSERT(dtrace_toxranges_max == 0); + dtrace_toxranges_max = 1; + } else { + dtrace_toxranges_max <<= 1; + } + + nsize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t); + range = kmem_zalloc(nsize, KM_SLEEP); + + if (dtrace_toxrange != NULL) { + ASSERT(osize != 0); + bcopy(dtrace_toxrange, range, osize); + kmem_free(dtrace_toxrange, osize); + } + + dtrace_toxrange = range; + } + + ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_base == 0); + ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_limit == 0); + + dtrace_toxrange[dtrace_toxranges].dtt_base = base; + dtrace_toxrange[dtrace_toxranges].dtt_limit = limit; + dtrace_toxranges++; +} + +/* + * DTrace Driver Cookbook Functions + */ +#if defined(sun) +/*ARGSUSED*/ +static int +dtrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + dtrace_provider_id_t id; + dtrace_state_t *state = NULL; + dtrace_enabling_t *enab; + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + if (ddi_soft_state_init(&dtrace_softstate, + sizeof (dtrace_state_t), 0) != 0) { + cmn_err(CE_NOTE, "/dev/dtrace failed to initialize soft state"); + mutex_exit(&cpu_lock); + mutex_exit(&dtrace_provider_lock); + mutex_exit(&dtrace_lock); + return (DDI_FAILURE); + } + + if (ddi_create_minor_node(devi, DTRACEMNR_DTRACE, S_IFCHR, + DTRACEMNRN_DTRACE, DDI_PSEUDO, NULL) == DDI_FAILURE || + ddi_create_minor_node(devi, DTRACEMNR_HELPER, S_IFCHR, + DTRACEMNRN_HELPER, DDI_PSEUDO, NULL) == DDI_FAILURE) { + cmn_err(CE_NOTE, "/dev/dtrace couldn't create minor nodes"); + ddi_remove_minor_node(devi, NULL); + ddi_soft_state_fini(&dtrace_softstate); + mutex_exit(&cpu_lock); + mutex_exit(&dtrace_provider_lock); + mutex_exit(&dtrace_lock); + return (DDI_FAILURE); + } + + ddi_report_dev(devi); + dtrace_devi = devi; + + dtrace_modload = dtrace_module_loaded; + dtrace_modunload = dtrace_module_unloaded; + dtrace_cpu_init = dtrace_cpu_setup_initial; + dtrace_helpers_cleanup = dtrace_helpers_destroy; + dtrace_helpers_fork = dtrace_helpers_duplicate; + dtrace_cpustart_init = dtrace_suspend; + dtrace_cpustart_fini = dtrace_resume; + dtrace_debugger_init = dtrace_suspend; + dtrace_debugger_fini = dtrace_resume; + + register_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL); + + ASSERT(MUTEX_HELD(&cpu_lock)); + + dtrace_arena = vmem_create("dtrace", (void *)1, UINT32_MAX, 1, + NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER); + dtrace_minor = vmem_create("dtrace_minor", (void *)DTRACEMNRN_CLONE, + UINT32_MAX - DTRACEMNRN_CLONE, 1, NULL, NULL, NULL, 0, + VM_SLEEP | VMC_IDENTIFIER); + dtrace_taskq = taskq_create("dtrace_taskq", 1, maxclsyspri, + 1, INT_MAX, 0); + + dtrace_state_cache = kmem_cache_create("dtrace_state_cache", + sizeof (dtrace_dstate_percpu_t) * NCPU, DTRACE_STATE_ALIGN, + NULL, NULL, NULL, NULL, NULL, 0); + + ASSERT(MUTEX_HELD(&cpu_lock)); + dtrace_bymod = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_mod), + offsetof(dtrace_probe_t, dtpr_nextmod), + offsetof(dtrace_probe_t, dtpr_prevmod)); + + dtrace_byfunc = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_func), + offsetof(dtrace_probe_t, dtpr_nextfunc), + offsetof(dtrace_probe_t, dtpr_prevfunc)); + + dtrace_byname = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_name), + offsetof(dtrace_probe_t, dtpr_nextname), + offsetof(dtrace_probe_t, dtpr_prevname)); + + if (dtrace_retain_max < 1) { + cmn_err(CE_WARN, "illegal value (%lu) for dtrace_retain_max; " + "setting to 1", dtrace_retain_max); + dtrace_retain_max = 1; + } + + /* + * Now discover our toxic ranges. + */ + dtrace_toxic_ranges(dtrace_toxrange_add); + + /* + * Before we register ourselves as a provider to our own framework, + * we would like to assert that dtrace_provider is NULL -- but that's + * not true if we were loaded as a dependency of a DTrace provider. + * Once we've registered, we can assert that dtrace_provider is our + * pseudo provider. + */ + (void) dtrace_register("dtrace", &dtrace_provider_attr, + DTRACE_PRIV_NONE, 0, &dtrace_provider_ops, NULL, &id); + + ASSERT(dtrace_provider != NULL); + ASSERT((dtrace_provider_id_t)dtrace_provider == id); + + dtrace_probeid_begin = dtrace_probe_create((dtrace_provider_id_t) + dtrace_provider, NULL, NULL, "BEGIN", 0, NULL); + dtrace_probeid_end = dtrace_probe_create((dtrace_provider_id_t) + dtrace_provider, NULL, NULL, "END", 0, NULL); + dtrace_probeid_error = dtrace_probe_create((dtrace_provider_id_t) + dtrace_provider, NULL, NULL, "ERROR", 1, NULL); + + dtrace_anon_property(); + mutex_exit(&cpu_lock); + + /* + * If DTrace helper tracing is enabled, we need to allocate the + * trace buffer and initialize the values. + */ + if (dtrace_helptrace_enabled) { + ASSERT(dtrace_helptrace_buffer == NULL); + dtrace_helptrace_buffer = + kmem_zalloc(dtrace_helptrace_bufsize, KM_SLEEP); + dtrace_helptrace_next = 0; + } + + /* + * If there are already providers, we must ask them to provide their + * probes, and then match any anonymous enabling against them. Note + * that there should be no other retained enablings at this time: + * the only retained enablings at this time should be the anonymous + * enabling. + */ + if (dtrace_anon.dta_enabling != NULL) { + ASSERT(dtrace_retained == dtrace_anon.dta_enabling); + + dtrace_enabling_provide(NULL); + state = dtrace_anon.dta_state; + + /* + * We couldn't hold cpu_lock across the above call to + * dtrace_enabling_provide(), but we must hold it to actually + * enable the probes. We have to drop all of our locks, pick + * up cpu_lock, and regain our locks before matching the + * retained anonymous enabling. + */ + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + if ((enab = dtrace_anon.dta_enabling) != NULL) + (void) dtrace_enabling_match(enab, NULL); + + mutex_exit(&cpu_lock); + } + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + if (state != NULL) { + /* + * If we created any anonymous state, set it going now. + */ + (void) dtrace_state_go(state, &dtrace_anon.dta_beganon); + } + + return (DDI_SUCCESS); +} +#endif + +#if !defined(sun) +static void +dtrace_dtr(void *data __unused) +{ +} +#endif + +/*ARGSUSED*/ +static int +#if defined(sun) +dtrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +#else +dtrace_open(struct cdev *dev, int oflags, int devtype, struct thread *td) +#endif +{ + dtrace_state_t *state; + uint32_t priv; + uid_t uid; + zoneid_t zoneid; + +#if defined(sun) + if (getminor(*devp) == DTRACEMNRN_HELPER) + return (0); + + /* + * If this wasn't an open with the "helper" minor, then it must be + * the "dtrace" minor. + */ + ASSERT(getminor(*devp) == DTRACEMNRN_DTRACE); +#else + cred_t *cred_p = NULL; + + cred_p = dev->si_cred; +#endif + + /* + * If no DTRACE_PRIV_* bits are set in the credential, then the + * caller lacks sufficient permission to do anything with DTrace. + */ + dtrace_cred2priv(cred_p, &priv, &uid, &zoneid); + if (priv == DTRACE_PRIV_NONE) { + return (EACCES); + } + + /* + * Ask all providers to provide all their probes. + */ + mutex_enter(&dtrace_provider_lock); + dtrace_probe_provide(NULL, NULL); + mutex_exit(&dtrace_provider_lock); + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_lock); + dtrace_opens++; + dtrace_membar_producer(); + +#if defined(sun) + /* + * If the kernel debugger is active (that is, if the kernel debugger + * modified text in some way), we won't allow the open. + */ + if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) { + dtrace_opens--; + mutex_exit(&cpu_lock); + mutex_exit(&dtrace_lock); + return (EBUSY); + } + + state = dtrace_state_create(devp, cred_p); +#else + state = dtrace_state_create(dev); + devfs_set_cdevpriv(state, dtrace_dtr); +#endif + + mutex_exit(&cpu_lock); + + if (state == NULL) { +#if defined(sun) + if (--dtrace_opens == 0) + (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); +#else + --dtrace_opens; +#endif + mutex_exit(&dtrace_lock); + return (EAGAIN); + } + + mutex_exit(&dtrace_lock); + + return (0); +} + +/*ARGSUSED*/ +static int +#if defined(sun) +dtrace_close(dev_t dev, int flag, int otyp, cred_t *cred_p) +#else +dtrace_close(struct cdev *dev, int flags, int fmt __unused, struct thread *td) +#endif +{ +#if defined(sun) + minor_t minor = getminor(dev); + dtrace_state_t *state; + + if (minor == DTRACEMNRN_HELPER) + return (0); + + state = ddi_get_soft_state(dtrace_softstate, minor); +#else + dtrace_state_t *state; + devfs_get_cdevpriv((void **) &state); + +#endif + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_lock); + + if (state != NULL) { + if (state->dts_anon) { + /* + * There is anonymous state. Destroy that first. + */ + ASSERT(dtrace_anon.dta_state == NULL); + dtrace_state_destroy(state->dts_anon); + } + + dtrace_state_destroy(state); + +#if !defined(sun) + kmem_free(state, 0); + devfs_clear_cdevpriv(); +#endif + } + + ASSERT(dtrace_opens > 0); +#if defined(sun) + if (--dtrace_opens == 0) + (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); +#else + --dtrace_opens; +#endif + + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + + return (0); +} + +#if defined(sun) +/*ARGSUSED*/ +static int +dtrace_ioctl_helper(int cmd, intptr_t arg, int *rv) +{ + int rval; + dof_helper_t help, *dhp = NULL; + + switch (cmd) { + case DTRACEHIOC_ADDDOF: + if (copyin((void *)arg, &help, sizeof (help)) != 0) { + dtrace_dof_error(NULL, "failed to copyin DOF helper"); + return (EFAULT); + } + + dhp = &help; + arg = (intptr_t)help.dofhp_dof; + /*FALLTHROUGH*/ + + case DTRACEHIOC_ADD: { + dof_hdr_t *dof = dtrace_dof_copyin(arg, &rval); + + if (dof == NULL) + return (rval); + + mutex_enter(&dtrace_lock); + + /* + * dtrace_helper_slurp() takes responsibility for the dof -- + * it may free it now or it may save it and free it later. + */ + if ((rval = dtrace_helper_slurp(dof, dhp)) != -1) { + *rv = rval; + rval = 0; + } else { + rval = EINVAL; + } + + mutex_exit(&dtrace_lock); + return (rval); + } + + case DTRACEHIOC_REMOVE: { + mutex_enter(&dtrace_lock); + rval = dtrace_helper_destroygen(arg); + mutex_exit(&dtrace_lock); + + return (rval); + } + + default: + break; + } + + return (ENOTTY); +} + +/*ARGSUSED*/ +static int +dtrace_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv) +{ + minor_t minor = getminor(dev); + dtrace_state_t *state; + int rval; + + if (minor == DTRACEMNRN_HELPER) + return (dtrace_ioctl_helper(cmd, arg, rv)); + + state = ddi_get_soft_state(dtrace_softstate, minor); + + if (state->dts_anon) { + ASSERT(dtrace_anon.dta_state == NULL); + state = state->dts_anon; + } + + switch (cmd) { + case DTRACEIOC_PROVIDER: { + dtrace_providerdesc_t pvd; + dtrace_provider_t *pvp; + + if (copyin((void *)arg, &pvd, sizeof (pvd)) != 0) + return (EFAULT); + + pvd.dtvd_name[DTRACE_PROVNAMELEN - 1] = '\0'; + mutex_enter(&dtrace_provider_lock); + + for (pvp = dtrace_provider; pvp != NULL; pvp = pvp->dtpv_next) { + if (strcmp(pvp->dtpv_name, pvd.dtvd_name) == 0) + break; + } + + mutex_exit(&dtrace_provider_lock); + + if (pvp == NULL) + return (ESRCH); + + bcopy(&pvp->dtpv_priv, &pvd.dtvd_priv, sizeof (dtrace_ppriv_t)); + bcopy(&pvp->dtpv_attr, &pvd.dtvd_attr, sizeof (dtrace_pattr_t)); + + if (copyout(&pvd, (void *)arg, sizeof (pvd)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_EPROBE: { + dtrace_eprobedesc_t epdesc; + dtrace_ecb_t *ecb; + dtrace_action_t *act; + void *buf; + size_t size; + uintptr_t dest; + int nrecs; + + if (copyin((void *)arg, &epdesc, sizeof (epdesc)) != 0) + return (EFAULT); + + mutex_enter(&dtrace_lock); + + if ((ecb = dtrace_epid2ecb(state, epdesc.dtepd_epid)) == NULL) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + if (ecb->dte_probe == NULL) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + epdesc.dtepd_probeid = ecb->dte_probe->dtpr_id; + epdesc.dtepd_uarg = ecb->dte_uarg; + epdesc.dtepd_size = ecb->dte_size; + + nrecs = epdesc.dtepd_nrecs; + epdesc.dtepd_nrecs = 0; + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) + continue; + + epdesc.dtepd_nrecs++; + } + + /* + * Now that we have the size, we need to allocate a temporary + * buffer in which to store the complete description. We need + * the temporary buffer to be able to drop dtrace_lock() + * across the copyout(), below. + */ + size = sizeof (dtrace_eprobedesc_t) + + (epdesc.dtepd_nrecs * sizeof (dtrace_recdesc_t)); + + buf = kmem_alloc(size, KM_SLEEP); + dest = (uintptr_t)buf; + + bcopy(&epdesc, (void *)dest, sizeof (epdesc)); + dest += offsetof(dtrace_eprobedesc_t, dtepd_rec[0]); + + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) + continue; + + if (nrecs-- == 0) + break; + + bcopy(&act->dta_rec, (void *)dest, + sizeof (dtrace_recdesc_t)); + dest += sizeof (dtrace_recdesc_t); + } + + mutex_exit(&dtrace_lock); + + if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) { + kmem_free(buf, size); + return (EFAULT); + } + + kmem_free(buf, size); + return (0); + } + + case DTRACEIOC_AGGDESC: { + dtrace_aggdesc_t aggdesc; + dtrace_action_t *act; + dtrace_aggregation_t *agg; + int nrecs; + uint32_t offs; + dtrace_recdesc_t *lrec; + void *buf; + size_t size; + uintptr_t dest; + + if (copyin((void *)arg, &aggdesc, sizeof (aggdesc)) != 0) + return (EFAULT); + + mutex_enter(&dtrace_lock); + + if ((agg = dtrace_aggid2agg(state, aggdesc.dtagd_id)) == NULL) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + aggdesc.dtagd_epid = agg->dtag_ecb->dte_epid; + + nrecs = aggdesc.dtagd_nrecs; + aggdesc.dtagd_nrecs = 0; + + offs = agg->dtag_base; + lrec = &agg->dtag_action.dta_rec; + aggdesc.dtagd_size = lrec->dtrd_offset + lrec->dtrd_size - offs; + + for (act = agg->dtag_first; ; act = act->dta_next) { + ASSERT(act->dta_intuple || + DTRACEACT_ISAGG(act->dta_kind)); + + /* + * If this action has a record size of zero, it + * denotes an argument to the aggregating action. + * Because the presence of this record doesn't (or + * shouldn't) affect the way the data is interpreted, + * we don't copy it out to save user-level the + * confusion of dealing with a zero-length record. + */ + if (act->dta_rec.dtrd_size == 0) { + ASSERT(agg->dtag_hasarg); + continue; + } + + aggdesc.dtagd_nrecs++; + + if (act == &agg->dtag_action) + break; + } + + /* + * Now that we have the size, we need to allocate a temporary + * buffer in which to store the complete description. We need + * the temporary buffer to be able to drop dtrace_lock() + * across the copyout(), below. + */ + size = sizeof (dtrace_aggdesc_t) + + (aggdesc.dtagd_nrecs * sizeof (dtrace_recdesc_t)); + + buf = kmem_alloc(size, KM_SLEEP); + dest = (uintptr_t)buf; + + bcopy(&aggdesc, (void *)dest, sizeof (aggdesc)); + dest += offsetof(dtrace_aggdesc_t, dtagd_rec[0]); + + for (act = agg->dtag_first; ; act = act->dta_next) { + dtrace_recdesc_t rec = act->dta_rec; + + /* + * See the comment in the above loop for why we pass + * over zero-length records. + */ + if (rec.dtrd_size == 0) { + ASSERT(agg->dtag_hasarg); + continue; + } + + if (nrecs-- == 0) + break; + + rec.dtrd_offset -= offs; + bcopy(&rec, (void *)dest, sizeof (rec)); + dest += sizeof (dtrace_recdesc_t); + + if (act == &agg->dtag_action) + break; + } + + mutex_exit(&dtrace_lock); + + if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) { + kmem_free(buf, size); + return (EFAULT); + } + + kmem_free(buf, size); + return (0); + } + + case DTRACEIOC_ENABLE: { + dof_hdr_t *dof; + dtrace_enabling_t *enab = NULL; + dtrace_vstate_t *vstate; + int err = 0; + + *rv = 0; + + /* + * If a NULL argument has been passed, we take this as our + * cue to reevaluate our enablings. + */ + if (arg == NULL) { + dtrace_enabling_matchall(); + + return (0); + } + + if ((dof = dtrace_dof_copyin(arg, &rval)) == NULL) + return (rval); + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_lock); + vstate = &state->dts_vstate; + + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) { + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + dtrace_dof_destroy(dof); + return (EBUSY); + } + + if (dtrace_dof_slurp(dof, vstate, cr, &enab, 0, B_TRUE) != 0) { + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + dtrace_dof_destroy(dof); + return (EINVAL); + } + + if ((rval = dtrace_dof_options(dof, state)) != 0) { + dtrace_enabling_destroy(enab); + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + dtrace_dof_destroy(dof); + return (rval); + } + + if ((err = dtrace_enabling_match(enab, rv)) == 0) { + err = dtrace_enabling_retain(enab); + } else { + dtrace_enabling_destroy(enab); + } + + mutex_exit(&cpu_lock); + mutex_exit(&dtrace_lock); + dtrace_dof_destroy(dof); + + return (err); + } + + case DTRACEIOC_REPLICATE: { + dtrace_repldesc_t desc; + dtrace_probedesc_t *match = &desc.dtrpd_match; + dtrace_probedesc_t *create = &desc.dtrpd_create; + int err; + + if (copyin((void *)arg, &desc, sizeof (desc)) != 0) + return (EFAULT); + + match->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + match->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + match->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + match->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + create->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + create->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + create->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + create->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + mutex_enter(&dtrace_lock); + err = dtrace_enabling_replicate(state, match, create); + mutex_exit(&dtrace_lock); + + return (err); + } + + case DTRACEIOC_PROBEMATCH: + case DTRACEIOC_PROBES: { + dtrace_probe_t *probe = NULL; + dtrace_probedesc_t desc; + dtrace_probekey_t pkey; + dtrace_id_t i; + int m = 0; + uint32_t priv; + uid_t uid; + zoneid_t zoneid; + + if (copyin((void *)arg, &desc, sizeof (desc)) != 0) + return (EFAULT); + + desc.dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + desc.dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + desc.dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + desc.dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + /* + * Before we attempt to match this probe, we want to give + * all providers the opportunity to provide it. + */ + if (desc.dtpd_id == DTRACE_IDNONE) { + mutex_enter(&dtrace_provider_lock); + dtrace_probe_provide(&desc, NULL); + mutex_exit(&dtrace_provider_lock); + desc.dtpd_id++; + } + + if (cmd == DTRACEIOC_PROBEMATCH) { + dtrace_probekey(&desc, &pkey); + pkey.dtpk_id = DTRACE_IDNONE; + } + + dtrace_cred2priv(cr, &priv, &uid, &zoneid); + + mutex_enter(&dtrace_lock); + + if (cmd == DTRACEIOC_PROBEMATCH) { + for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i - 1]) != NULL && + (m = dtrace_match_probe(probe, &pkey, + priv, uid, zoneid)) != 0) + break; + } + + if (m < 0) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + } else { + for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i - 1]) != NULL && + dtrace_match_priv(probe, priv, uid, zoneid)) + break; + } + } + + if (probe == NULL) { + mutex_exit(&dtrace_lock); + return (ESRCH); + } + + dtrace_probe_description(probe, &desc); + mutex_exit(&dtrace_lock); + + if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_PROBEARG: { + dtrace_argdesc_t desc; + dtrace_probe_t *probe; + dtrace_provider_t *prov; + + if (copyin((void *)arg, &desc, sizeof (desc)) != 0) + return (EFAULT); + + if (desc.dtargd_id == DTRACE_IDNONE) + return (EINVAL); + + if (desc.dtargd_ndx == DTRACE_ARGNONE) + return (EINVAL); + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&mod_lock); + mutex_enter(&dtrace_lock); + + if (desc.dtargd_id > dtrace_nprobes) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + return (EINVAL); + } + + if ((probe = dtrace_probes[desc.dtargd_id - 1]) == NULL) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + return (EINVAL); + } + + mutex_exit(&dtrace_lock); + + prov = probe->dtpr_provider; + + if (prov->dtpv_pops.dtps_getargdesc == NULL) { + /* + * There isn't any typed information for this probe. + * Set the argument number to DTRACE_ARGNONE. + */ + desc.dtargd_ndx = DTRACE_ARGNONE; + } else { + desc.dtargd_native[0] = '\0'; + desc.dtargd_xlate[0] = '\0'; + desc.dtargd_mapping = desc.dtargd_ndx; + + prov->dtpv_pops.dtps_getargdesc(prov->dtpv_arg, + probe->dtpr_id, probe->dtpr_arg, &desc); + } + + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + + if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_GO: { + processorid_t cpuid; + rval = dtrace_state_go(state, &cpuid); + + if (rval != 0) + return (rval); + + if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_STOP: { + processorid_t cpuid; + + mutex_enter(&dtrace_lock); + rval = dtrace_state_stop(state, &cpuid); + mutex_exit(&dtrace_lock); + + if (rval != 0) + return (rval); + + if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_DOFGET: { + dof_hdr_t hdr, *dof; + uint64_t len; + + if (copyin((void *)arg, &hdr, sizeof (hdr)) != 0) + return (EFAULT); + + mutex_enter(&dtrace_lock); + dof = dtrace_dof_create(state); + mutex_exit(&dtrace_lock); + + len = MIN(hdr.dofh_loadsz, dof->dofh_loadsz); + rval = copyout(dof, (void *)arg, len); + dtrace_dof_destroy(dof); + + return (rval == 0 ? 0 : EFAULT); + } + + case DTRACEIOC_AGGSNAP: + case DTRACEIOC_BUFSNAP: { + dtrace_bufdesc_t desc; + caddr_t cached; + dtrace_buffer_t *buf; + + if (copyin((void *)arg, &desc, sizeof (desc)) != 0) + return (EFAULT); + + if (desc.dtbd_cpu < 0 || desc.dtbd_cpu >= NCPU) + return (EINVAL); + + mutex_enter(&dtrace_lock); + + if (cmd == DTRACEIOC_BUFSNAP) { + buf = &state->dts_buffer[desc.dtbd_cpu]; + } else { + buf = &state->dts_aggbuffer[desc.dtbd_cpu]; + } + + if (buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL)) { + size_t sz = buf->dtb_offset; + + if (state->dts_activity != DTRACE_ACTIVITY_STOPPED) { + mutex_exit(&dtrace_lock); + return (EBUSY); + } + + /* + * If this buffer has already been consumed, we're + * going to indicate that there's nothing left here + * to consume. + */ + if (buf->dtb_flags & DTRACEBUF_CONSUMED) { + mutex_exit(&dtrace_lock); + + desc.dtbd_size = 0; + desc.dtbd_drops = 0; + desc.dtbd_errors = 0; + desc.dtbd_oldest = 0; + sz = sizeof (desc); + + if (copyout(&desc, (void *)arg, sz) != 0) + return (EFAULT); + + return (0); + } + + /* + * If this is a ring buffer that has wrapped, we want + * to copy the whole thing out. + */ + if (buf->dtb_flags & DTRACEBUF_WRAPPED) { + dtrace_buffer_polish(buf); + sz = buf->dtb_size; + } + + if (copyout(buf->dtb_tomax, desc.dtbd_data, sz) != 0) { + mutex_exit(&dtrace_lock); + return (EFAULT); + } + + desc.dtbd_size = sz; + desc.dtbd_drops = buf->dtb_drops; + desc.dtbd_errors = buf->dtb_errors; + desc.dtbd_oldest = buf->dtb_xamot_offset; + + mutex_exit(&dtrace_lock); + + if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) + return (EFAULT); + + buf->dtb_flags |= DTRACEBUF_CONSUMED; + + return (0); + } + + if (buf->dtb_tomax == NULL) { + ASSERT(buf->dtb_xamot == NULL); + mutex_exit(&dtrace_lock); + return (ENOENT); + } + + cached = buf->dtb_tomax; + ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); + + dtrace_xcall(desc.dtbd_cpu, + (dtrace_xcall_t)dtrace_buffer_switch, buf); + + state->dts_errors += buf->dtb_xamot_errors; + + /* + * If the buffers did not actually switch, then the cross call + * did not take place -- presumably because the given CPU is + * not in the ready set. If this is the case, we'll return + * ENOENT. + */ + if (buf->dtb_tomax == cached) { + ASSERT(buf->dtb_xamot != cached); + mutex_exit(&dtrace_lock); + return (ENOENT); + } + + ASSERT(cached == buf->dtb_xamot); + + /* + * We have our snapshot; now copy it out. + */ + if (copyout(buf->dtb_xamot, desc.dtbd_data, + buf->dtb_xamot_offset) != 0) { + mutex_exit(&dtrace_lock); + return (EFAULT); + } + + desc.dtbd_size = buf->dtb_xamot_offset; + desc.dtbd_drops = buf->dtb_xamot_drops; + desc.dtbd_errors = buf->dtb_xamot_errors; + desc.dtbd_oldest = 0; + + mutex_exit(&dtrace_lock); + + /* + * Finally, copy out the buffer description. + */ + if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_CONF: { + dtrace_conf_t conf; + + bzero(&conf, sizeof (conf)); + conf.dtc_difversion = DIF_VERSION; + conf.dtc_difintregs = DIF_DIR_NREGS; + conf.dtc_diftupregs = DIF_DTR_NREGS; + conf.dtc_ctfmodel = CTF_MODEL_NATIVE; + + if (copyout(&conf, (void *)arg, sizeof (conf)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_STATUS: { + dtrace_status_t stat; + dtrace_dstate_t *dstate; + int i, j; + uint64_t nerrs; + + /* + * See the comment in dtrace_state_deadman() for the reason + * for setting dts_laststatus to INT64_MAX before setting + * it to the correct value. + */ + state->dts_laststatus = INT64_MAX; + dtrace_membar_producer(); + state->dts_laststatus = dtrace_gethrtime(); + + bzero(&stat, sizeof (stat)); + + mutex_enter(&dtrace_lock); + + if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) { + mutex_exit(&dtrace_lock); + return (ENOENT); + } + + if (state->dts_activity == DTRACE_ACTIVITY_DRAINING) + stat.dtst_exiting = 1; + + nerrs = state->dts_errors; + dstate = &state->dts_vstate.dtvs_dynvars; + + for (i = 0; i < NCPU; i++) { + dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[i]; + + stat.dtst_dyndrops += dcpu->dtdsc_drops; + stat.dtst_dyndrops_dirty += dcpu->dtdsc_dirty_drops; + stat.dtst_dyndrops_rinsing += dcpu->dtdsc_rinsing_drops; + + if (state->dts_buffer[i].dtb_flags & DTRACEBUF_FULL) + stat.dtst_filled++; + + nerrs += state->dts_buffer[i].dtb_errors; + + for (j = 0; j < state->dts_nspeculations; j++) { + dtrace_speculation_t *spec; + dtrace_buffer_t *buf; + + spec = &state->dts_speculations[j]; + buf = &spec->dtsp_buffer[i]; + stat.dtst_specdrops += buf->dtb_xamot_drops; + } + } + + stat.dtst_specdrops_busy = state->dts_speculations_busy; + stat.dtst_specdrops_unavail = state->dts_speculations_unavail; + stat.dtst_stkstroverflows = state->dts_stkstroverflows; + stat.dtst_dblerrors = state->dts_dblerrors; + stat.dtst_killed = + (state->dts_activity == DTRACE_ACTIVITY_KILLED); + stat.dtst_errors = nerrs; + + mutex_exit(&dtrace_lock); + + if (copyout(&stat, (void *)arg, sizeof (stat)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_FORMAT: { + dtrace_fmtdesc_t fmt; + char *str; + int len; + + if (copyin((void *)arg, &fmt, sizeof (fmt)) != 0) + return (EFAULT); + + mutex_enter(&dtrace_lock); + + if (fmt.dtfd_format == 0 || + fmt.dtfd_format > state->dts_nformats) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + /* + * Format strings are allocated contiguously and they are + * never freed; if a format index is less than the number + * of formats, we can assert that the format map is non-NULL + * and that the format for the specified index is non-NULL. + */ + ASSERT(state->dts_formats != NULL); + str = state->dts_formats[fmt.dtfd_format - 1]; + ASSERT(str != NULL); + + len = strlen(str) + 1; + + if (len > fmt.dtfd_length) { + fmt.dtfd_length = len; + + if (copyout(&fmt, (void *)arg, sizeof (fmt)) != 0) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + } else { + if (copyout(str, fmt.dtfd_string, len) != 0) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + } + + mutex_exit(&dtrace_lock); + return (0); + } + + default: + break; + } + + return (ENOTTY); +} + +/*ARGSUSED*/ +static int +dtrace_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) +{ + dtrace_state_t *state; + + switch (cmd) { + case DDI_DETACH: + break; + + case DDI_SUSPEND: + return (DDI_SUCCESS); + + default: + return (DDI_FAILURE); + } + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + ASSERT(dtrace_opens == 0); + + if (dtrace_helpers > 0) { + mutex_exit(&dtrace_provider_lock); + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + return (DDI_FAILURE); + } + + if (dtrace_unregister((dtrace_provider_id_t)dtrace_provider) != 0) { + mutex_exit(&dtrace_provider_lock); + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + return (DDI_FAILURE); + } + + dtrace_provider = NULL; + + if ((state = dtrace_anon_grab()) != NULL) { + /* + * If there were ECBs on this state, the provider should + * have not been allowed to detach; assert that there is + * none. + */ + ASSERT(state->dts_necbs == 0); + dtrace_state_destroy(state); + + /* + * If we're being detached with anonymous state, we need to + * indicate to the kernel debugger that DTrace is now inactive. + */ + (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); + } + + bzero(&dtrace_anon, sizeof (dtrace_anon_t)); + unregister_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL); + dtrace_cpu_init = NULL; + dtrace_helpers_cleanup = NULL; + dtrace_helpers_fork = NULL; + dtrace_cpustart_init = NULL; + dtrace_cpustart_fini = NULL; + dtrace_debugger_init = NULL; + dtrace_debugger_fini = NULL; + dtrace_modload = NULL; + dtrace_modunload = NULL; + + mutex_exit(&cpu_lock); + + if (dtrace_helptrace_enabled) { + kmem_free(dtrace_helptrace_buffer, dtrace_helptrace_bufsize); + dtrace_helptrace_buffer = NULL; + } + + kmem_free(dtrace_probes, dtrace_nprobes * sizeof (dtrace_probe_t *)); + dtrace_probes = NULL; + dtrace_nprobes = 0; + + dtrace_hash_destroy(dtrace_bymod); + dtrace_hash_destroy(dtrace_byfunc); + dtrace_hash_destroy(dtrace_byname); + dtrace_bymod = NULL; + dtrace_byfunc = NULL; + dtrace_byname = NULL; + + kmem_cache_destroy(dtrace_state_cache); + vmem_destroy(dtrace_minor); + vmem_destroy(dtrace_arena); + + if (dtrace_toxrange != NULL) { + kmem_free(dtrace_toxrange, + dtrace_toxranges_max * sizeof (dtrace_toxrange_t)); + dtrace_toxrange = NULL; + dtrace_toxranges = 0; + dtrace_toxranges_max = 0; + } + + ddi_remove_minor_node(dtrace_devi, NULL); + dtrace_devi = NULL; + + ddi_soft_state_fini(&dtrace_softstate); + + ASSERT(dtrace_vtime_references == 0); + ASSERT(dtrace_opens == 0); + ASSERT(dtrace_retained == NULL); + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + /* + * We don't destroy the task queue until after we have dropped our + * locks (taskq_destroy() may block on running tasks). To prevent + * attempting to do work after we have effectively detached but before + * the task queue has been destroyed, all tasks dispatched via the + * task queue must check that DTrace is still attached before + * performing any operation. + */ + taskq_destroy(dtrace_taskq); + dtrace_taskq = NULL; + + return (DDI_SUCCESS); +} +#endif + +#if defined(sun) +/*ARGSUSED*/ +static int +dtrace_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)dtrace_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} +#endif + +#if defined(sun) +static struct cb_ops dtrace_cb_ops = { + dtrace_open, /* open */ + dtrace_close, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + dtrace_ioctl, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_NEW | D_MP /* Driver compatibility flag */ +}; + +static struct dev_ops dtrace_ops = { + DEVO_REV, /* devo_rev */ + 0, /* refcnt */ + dtrace_info, /* get_dev_info */ + nulldev, /* identify */ + nulldev, /* probe */ + dtrace_attach, /* attach */ + dtrace_detach, /* detach */ + nodev, /* reset */ + &dtrace_cb_ops, /* driver operations */ + NULL, /* bus operations */ + nodev /* dev power */ +}; + +static struct modldrv modldrv = { + &mod_driverops, /* module type (this is a pseudo driver) */ + "Dynamic Tracing", /* name of module */ + &dtrace_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, + (void *)&modldrv, + NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} +#else + +static d_ioctl_t dtrace_ioctl; +static void dtrace_load(void *); +static int dtrace_unload(void); +static struct cdev *dtrace_dev; + +void dtrace_invop_init(void); +void dtrace_invop_uninit(void); + +static struct cdevsw dtrace_cdevsw = { + .d_version = D_VERSION, + .d_flags = D_TRACKCLOSE, + .d_close = dtrace_close, + .d_ioctl = dtrace_ioctl, + .d_open = dtrace_open, + .d_name = "dtrace", +}; + +#include <dtrace_anon.c> +#include <dtrace_ioctl.c> +#include <dtrace_load.c> +#include <dtrace_modevent.c> +#include <dtrace_sysctl.c> +#include <dtrace_unload.c> +#include <dtrace_vtime.c> +#include <dtrace_hacks.c> +#include <dtrace_isa.c> + +SYSINIT(dtrace_load, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_load, NULL); +SYSUNINIT(dtrace_unload, SI_SUB_DTRACE, SI_ORDER_FIRST, dtrace_unload, NULL); +SYSINIT(dtrace_anon_init, SI_SUB_DTRACE_ANON, SI_ORDER_FIRST, dtrace_anon_init, NULL); + +DEV_MODULE(dtrace, dtrace_modevent, NULL); +MODULE_VERSION(dtrace, 1); +MODULE_DEPEND(dtrace, cyclic, 1, 1, 1); +MODULE_DEPEND(dtrace, opensolaris, 1, 1, 1); +#endif diff --git a/sys/cddl/contrib/opensolaris/uts/common/dtrace/fasttrap.c b/sys/cddl/contrib/opensolaris/uts/common/dtrace/fasttrap.c new file mode 100644 index 000000000000..45839cbb8f94 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/dtrace/fasttrap.c @@ -0,0 +1,2376 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/atomic.h> +#include <sys/errno.h> +#include <sys/stat.h> +#include <sys/modctl.h> +#include <sys/conf.h> +#include <sys/systm.h> +#include <sys/ddi.h> +#include <sys/sunddi.h> +#include <sys/cpuvar.h> +#include <sys/kmem.h> +#include <sys/strsubr.h> +#include <sys/fasttrap.h> +#include <sys/fasttrap_impl.h> +#include <sys/fasttrap_isa.h> +#include <sys/dtrace.h> +#include <sys/dtrace_impl.h> +#include <sys/sysmacros.h> +#include <sys/proc.h> +#include <sys/priv.h> +#include <sys/policy.h> +#include <util/qsort.h> + +/* + * User-Land Trap-Based Tracing + * ---------------------------- + * + * The fasttrap provider allows DTrace consumers to instrument any user-level + * instruction to gather data; this includes probes with semantic + * signifigance like entry and return as well as simple offsets into the + * function. While the specific techniques used are very ISA specific, the + * methodology is generalizable to any architecture. + * + * + * The General Methodology + * ----------------------- + * + * With the primary goal of tracing every user-land instruction and the + * limitation that we can't trust user space so don't want to rely on much + * information there, we begin by replacing the instructions we want to trace + * with trap instructions. Each instruction we overwrite is saved into a hash + * table keyed by process ID and pc address. When we enter the kernel due to + * this trap instruction, we need the effects of the replaced instruction to + * appear to have occurred before we proceed with the user thread's + * execution. + * + * Each user level thread is represented by a ulwp_t structure which is + * always easily accessible through a register. The most basic way to produce + * the effects of the instruction we replaced is to copy that instruction out + * to a bit of scratch space reserved in the user thread's ulwp_t structure + * (a sort of kernel-private thread local storage), set the PC to that + * scratch space and single step. When we reenter the kernel after single + * stepping the instruction we must then adjust the PC to point to what would + * normally be the next instruction. Of course, special care must be taken + * for branches and jumps, but these represent such a small fraction of any + * instruction set that writing the code to emulate these in the kernel is + * not too difficult. + * + * Return probes may require several tracepoints to trace every return site, + * and, conversely, each tracepoint may activate several probes (the entry + * and offset 0 probes, for example). To solve this muliplexing problem, + * tracepoints contain lists of probes to activate and probes contain lists + * of tracepoints to enable. If a probe is activated, it adds its ID to + * existing tracepoints or creates new ones as necessary. + * + * Most probes are activated _before_ the instruction is executed, but return + * probes are activated _after_ the effects of the last instruction of the + * function are visible. Return probes must be fired _after_ we have + * single-stepped the instruction whereas all other probes are fired + * beforehand. + * + * + * Lock Ordering + * ------------- + * + * The lock ordering below -- both internally and with respect to the DTrace + * framework -- is a little tricky and bears some explanation. Each provider + * has a lock (ftp_mtx) that protects its members including reference counts + * for enabled probes (ftp_rcount), consumers actively creating probes + * (ftp_ccount) and USDT consumers (ftp_mcount); all three prevent a provider + * from being freed. A provider is looked up by taking the bucket lock for the + * provider hash table, and is returned with its lock held. The provider lock + * may be taken in functions invoked by the DTrace framework, but may not be + * held while calling functions in the DTrace framework. + * + * To ensure consistency over multiple calls to the DTrace framework, the + * creation lock (ftp_cmtx) should be held. Naturally, the creation lock may + * not be taken when holding the provider lock as that would create a cyclic + * lock ordering. In situations where one would naturally take the provider + * lock and then the creation lock, we instead up a reference count to prevent + * the provider from disappearing, drop the provider lock, and acquire the + * creation lock. + * + * Briefly: + * bucket lock before provider lock + * DTrace before provider lock + * creation lock before DTrace + * never hold the provider lock and creation lock simultaneously + */ + +static dev_info_t *fasttrap_devi; +static dtrace_meta_provider_id_t fasttrap_meta_id; + +static timeout_id_t fasttrap_timeout; +static kmutex_t fasttrap_cleanup_mtx; +static uint_t fasttrap_cleanup_work; + +/* + * Generation count on modifications to the global tracepoint lookup table. + */ +static volatile uint64_t fasttrap_mod_gen; + +/* + * When the fasttrap provider is loaded, fasttrap_max is set to either + * FASTTRAP_MAX_DEFAULT or the value for fasttrap-max-probes in the + * fasttrap.conf file. Each time a probe is created, fasttrap_total is + * incremented by the number of tracepoints that may be associated with that + * probe; fasttrap_total is capped at fasttrap_max. + */ +#define FASTTRAP_MAX_DEFAULT 250000 +static uint32_t fasttrap_max; +static uint32_t fasttrap_total; + + +#define FASTTRAP_TPOINTS_DEFAULT_SIZE 0x4000 +#define FASTTRAP_PROVIDERS_DEFAULT_SIZE 0x100 +#define FASTTRAP_PROCS_DEFAULT_SIZE 0x100 + +#define FASTTRAP_PID_NAME "pid" + +fasttrap_hash_t fasttrap_tpoints; +static fasttrap_hash_t fasttrap_provs; +static fasttrap_hash_t fasttrap_procs; + +static uint64_t fasttrap_pid_count; /* pid ref count */ +static kmutex_t fasttrap_count_mtx; /* lock on ref count */ + +#define FASTTRAP_ENABLE_FAIL 1 +#define FASTTRAP_ENABLE_PARTIAL 2 + +static int fasttrap_tracepoint_enable(proc_t *, fasttrap_probe_t *, uint_t); +static void fasttrap_tracepoint_disable(proc_t *, fasttrap_probe_t *, uint_t); + +static fasttrap_provider_t *fasttrap_provider_lookup(pid_t, const char *, + const dtrace_pattr_t *); +static void fasttrap_provider_retire(pid_t, const char *, int); +static void fasttrap_provider_free(fasttrap_provider_t *); + +static fasttrap_proc_t *fasttrap_proc_lookup(pid_t); +static void fasttrap_proc_release(fasttrap_proc_t *); + +#define FASTTRAP_PROVS_INDEX(pid, name) \ + ((fasttrap_hash_str(name) + (pid)) & fasttrap_provs.fth_mask) + +#define FASTTRAP_PROCS_INDEX(pid) ((pid) & fasttrap_procs.fth_mask) + +static int +fasttrap_highbit(ulong_t i) +{ + int h = 1; + + if (i == 0) + return (0); +#ifdef _LP64 + if (i & 0xffffffff00000000ul) { + h += 32; i >>= 32; + } +#endif + if (i & 0xffff0000) { + h += 16; i >>= 16; + } + if (i & 0xff00) { + h += 8; i >>= 8; + } + if (i & 0xf0) { + h += 4; i >>= 4; + } + if (i & 0xc) { + h += 2; i >>= 2; + } + if (i & 0x2) { + h += 1; + } + return (h); +} + +static uint_t +fasttrap_hash_str(const char *p) +{ + unsigned int g; + uint_t hval = 0; + + while (*p) { + hval = (hval << 4) + *p++; + if ((g = (hval & 0xf0000000)) != 0) + hval ^= g >> 24; + hval &= ~g; + } + return (hval); +} + +void +fasttrap_sigtrap(proc_t *p, kthread_t *t, uintptr_t pc) +{ + sigqueue_t *sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP); + + sqp->sq_info.si_signo = SIGTRAP; + sqp->sq_info.si_code = TRAP_DTRACE; + sqp->sq_info.si_addr = (caddr_t)pc; + + mutex_enter(&p->p_lock); + sigaddqa(p, t, sqp); + mutex_exit(&p->p_lock); + + if (t != NULL) + aston(t); +} + +/* + * This function ensures that no threads are actively using the memory + * associated with probes that were formerly live. + */ +static void +fasttrap_mod_barrier(uint64_t gen) +{ + int i; + + if (gen < fasttrap_mod_gen) + return; + + fasttrap_mod_gen++; + + for (i = 0; i < NCPU; i++) { + mutex_enter(&cpu_core[i].cpuc_pid_lock); + mutex_exit(&cpu_core[i].cpuc_pid_lock); + } +} + +/* + * This is the timeout's callback for cleaning up the providers and their + * probes. + */ +/*ARGSUSED*/ +static void +fasttrap_pid_cleanup_cb(void *data) +{ + fasttrap_provider_t **fpp, *fp; + fasttrap_bucket_t *bucket; + dtrace_provider_id_t provid; + int i, later; + + static volatile int in = 0; + ASSERT(in == 0); + in = 1; + + mutex_enter(&fasttrap_cleanup_mtx); + while (fasttrap_cleanup_work) { + fasttrap_cleanup_work = 0; + mutex_exit(&fasttrap_cleanup_mtx); + + later = 0; + + /* + * Iterate over all the providers trying to remove the marked + * ones. If a provider is marked but not retired, we just + * have to take a crack at removing it -- it's no big deal if + * we can't. + */ + for (i = 0; i < fasttrap_provs.fth_nent; i++) { + bucket = &fasttrap_provs.fth_table[i]; + mutex_enter(&bucket->ftb_mtx); + fpp = (fasttrap_provider_t **)&bucket->ftb_data; + + while ((fp = *fpp) != NULL) { + if (!fp->ftp_marked) { + fpp = &fp->ftp_next; + continue; + } + + mutex_enter(&fp->ftp_mtx); + + /* + * If this provider has consumers actively + * creating probes (ftp_ccount) or is a USDT + * provider (ftp_mcount), we can't unregister + * or even condense. + */ + if (fp->ftp_ccount != 0 || + fp->ftp_mcount != 0) { + mutex_exit(&fp->ftp_mtx); + fp->ftp_marked = 0; + continue; + } + + if (!fp->ftp_retired || fp->ftp_rcount != 0) + fp->ftp_marked = 0; + + mutex_exit(&fp->ftp_mtx); + + /* + * If we successfully unregister this + * provider we can remove it from the hash + * chain and free the memory. If our attempt + * to unregister fails and this is a retired + * provider, increment our flag to try again + * pretty soon. If we've consumed more than + * half of our total permitted number of + * probes call dtrace_condense() to try to + * clean out the unenabled probes. + */ + provid = fp->ftp_provid; + if (dtrace_unregister(provid) != 0) { + if (fasttrap_total > fasttrap_max / 2) + (void) dtrace_condense(provid); + later += fp->ftp_marked; + fpp = &fp->ftp_next; + } else { + *fpp = fp->ftp_next; + fasttrap_provider_free(fp); + } + } + mutex_exit(&bucket->ftb_mtx); + } + + mutex_enter(&fasttrap_cleanup_mtx); + } + + ASSERT(fasttrap_timeout != 0); + + /* + * If we were unable to remove a retired provider, try again after + * a second. This situation can occur in certain circumstances where + * providers cannot be unregistered even though they have no probes + * enabled because of an execution of dtrace -l or something similar. + * If the timeout has been disabled (set to 1 because we're trying + * to detach), we set fasttrap_cleanup_work to ensure that we'll + * get a chance to do that work if and when the timeout is reenabled + * (if detach fails). + */ + if (later > 0 && fasttrap_timeout != (timeout_id_t)1) + fasttrap_timeout = timeout(&fasttrap_pid_cleanup_cb, NULL, hz); + else if (later > 0) + fasttrap_cleanup_work = 1; + else + fasttrap_timeout = 0; + + mutex_exit(&fasttrap_cleanup_mtx); + in = 0; +} + +/* + * Activates the asynchronous cleanup mechanism. + */ +static void +fasttrap_pid_cleanup(void) +{ + mutex_enter(&fasttrap_cleanup_mtx); + fasttrap_cleanup_work = 1; + if (fasttrap_timeout == 0) + fasttrap_timeout = timeout(&fasttrap_pid_cleanup_cb, NULL, 1); + mutex_exit(&fasttrap_cleanup_mtx); +} + +/* + * This is called from cfork() via dtrace_fasttrap_fork(). The child + * process's address space is (roughly) a copy of the parent process's so + * we have to remove all the instrumentation we had previously enabled in the + * parent. + */ +static void +fasttrap_fork(proc_t *p, proc_t *cp) +{ + pid_t ppid = p->p_pid; + int i; + + ASSERT(curproc == p); + ASSERT(p->p_proc_flag & P_PR_LOCK); + ASSERT(p->p_dtrace_count > 0); + ASSERT(cp->p_dtrace_count == 0); + + /* + * This would be simpler and faster if we maintained per-process + * hash tables of enabled tracepoints. It could, however, potentially + * slow down execution of a tracepoint since we'd need to go + * through two levels of indirection. In the future, we should + * consider either maintaining per-process ancillary lists of + * enabled tracepoints or hanging a pointer to a per-process hash + * table of enabled tracepoints off the proc structure. + */ + + /* + * We don't have to worry about the child process disappearing + * because we're in fork(). + */ + mutex_enter(&cp->p_lock); + sprlock_proc(cp); + mutex_exit(&cp->p_lock); + + /* + * Iterate over every tracepoint looking for ones that belong to the + * parent process, and remove each from the child process. + */ + for (i = 0; i < fasttrap_tpoints.fth_nent; i++) { + fasttrap_tracepoint_t *tp; + fasttrap_bucket_t *bucket = &fasttrap_tpoints.fth_table[i]; + + mutex_enter(&bucket->ftb_mtx); + for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { + if (tp->ftt_pid == ppid && + tp->ftt_proc->ftpc_acount != 0) { + int ret = fasttrap_tracepoint_remove(cp, tp); + ASSERT(ret == 0); + + /* + * The count of active providers can only be + * decremented (i.e. to zero) during exec, + * exit, and removal of a meta provider so it + * should be impossible to drop the count + * mid-fork. + */ + ASSERT(tp->ftt_proc->ftpc_acount != 0); + } + } + mutex_exit(&bucket->ftb_mtx); + } + + mutex_enter(&cp->p_lock); + sprunlock(cp); +} + +/* + * This is called from proc_exit() or from exec_common() if p_dtrace_probes + * is set on the proc structure to indicate that there is a pid provider + * associated with this process. + */ +static void +fasttrap_exec_exit(proc_t *p) +{ + ASSERT(p == curproc); + ASSERT(MUTEX_HELD(&p->p_lock)); + + mutex_exit(&p->p_lock); + + /* + * We clean up the pid provider for this process here; user-land + * static probes are handled by the meta-provider remove entry point. + */ + fasttrap_provider_retire(p->p_pid, FASTTRAP_PID_NAME, 0); + + mutex_enter(&p->p_lock); +} + + +/*ARGSUSED*/ +static void +fasttrap_pid_provide(void *arg, const dtrace_probedesc_t *desc) +{ + /* + * There are no "default" pid probes. + */ +} + +static int +fasttrap_tracepoint_enable(proc_t *p, fasttrap_probe_t *probe, uint_t index) +{ + fasttrap_tracepoint_t *tp, *new_tp = NULL; + fasttrap_bucket_t *bucket; + fasttrap_id_t *id; + pid_t pid; + uintptr_t pc; + + ASSERT(index < probe->ftp_ntps); + + pid = probe->ftp_pid; + pc = probe->ftp_tps[index].fit_tp->ftt_pc; + id = &probe->ftp_tps[index].fit_id; + + ASSERT(probe->ftp_tps[index].fit_tp->ftt_pid == pid); + + ASSERT(!(p->p_flag & SVFORK)); + + /* + * Before we make any modifications, make sure we've imposed a barrier + * on the generation in which this probe was last modified. + */ + fasttrap_mod_barrier(probe->ftp_gen); + + bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; + + /* + * If the tracepoint has already been enabled, just add our id to the + * list of interested probes. This may be our second time through + * this path in which case we'll have constructed the tracepoint we'd + * like to install. If we can't find a match, and have an allocated + * tracepoint ready to go, enable that one now. + * + * A tracepoint whose process is defunct is also considered defunct. + */ +again: + mutex_enter(&bucket->ftb_mtx); + for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { + /* + * Note that it's safe to access the active count on the + * associated proc structure because we know that at least one + * provider (this one) will still be around throughout this + * operation. + */ + if (tp->ftt_pid != pid || tp->ftt_pc != pc || + tp->ftt_proc->ftpc_acount == 0) + continue; + + /* + * Now that we've found a matching tracepoint, it would be + * a decent idea to confirm that the tracepoint is still + * enabled and the trap instruction hasn't been overwritten. + * Since this is a little hairy, we'll punt for now. + */ + + /* + * This can't be the first interested probe. We don't have + * to worry about another thread being in the midst of + * deleting this tracepoint (which would be the only valid + * reason for a tracepoint to have no interested probes) + * since we're holding P_PR_LOCK for this process. + */ + ASSERT(tp->ftt_ids != NULL || tp->ftt_retids != NULL); + + switch (id->fti_ptype) { + case DTFTP_ENTRY: + case DTFTP_OFFSETS: + case DTFTP_IS_ENABLED: + id->fti_next = tp->ftt_ids; + membar_producer(); + tp->ftt_ids = id; + membar_producer(); + break; + + case DTFTP_RETURN: + case DTFTP_POST_OFFSETS: + id->fti_next = tp->ftt_retids; + membar_producer(); + tp->ftt_retids = id; + membar_producer(); + break; + + default: + ASSERT(0); + } + + mutex_exit(&bucket->ftb_mtx); + + if (new_tp != NULL) { + new_tp->ftt_ids = NULL; + new_tp->ftt_retids = NULL; + } + + return (0); + } + + /* + * If we have a good tracepoint ready to go, install it now while + * we have the lock held and no one can screw with us. + */ + if (new_tp != NULL) { + int rc = 0; + + new_tp->ftt_next = bucket->ftb_data; + membar_producer(); + bucket->ftb_data = new_tp; + membar_producer(); + mutex_exit(&bucket->ftb_mtx); + + /* + * Activate the tracepoint in the ISA-specific manner. + * If this fails, we need to report the failure, but + * indicate that this tracepoint must still be disabled + * by calling fasttrap_tracepoint_disable(). + */ + if (fasttrap_tracepoint_install(p, new_tp) != 0) + rc = FASTTRAP_ENABLE_PARTIAL; + + /* + * Increment the count of the number of tracepoints active in + * the victim process. + */ + ASSERT(p->p_proc_flag & P_PR_LOCK); + p->p_dtrace_count++; + + return (rc); + } + + mutex_exit(&bucket->ftb_mtx); + + /* + * Initialize the tracepoint that's been preallocated with the probe. + */ + new_tp = probe->ftp_tps[index].fit_tp; + + ASSERT(new_tp->ftt_pid == pid); + ASSERT(new_tp->ftt_pc == pc); + ASSERT(new_tp->ftt_proc == probe->ftp_prov->ftp_proc); + ASSERT(new_tp->ftt_ids == NULL); + ASSERT(new_tp->ftt_retids == NULL); + + switch (id->fti_ptype) { + case DTFTP_ENTRY: + case DTFTP_OFFSETS: + case DTFTP_IS_ENABLED: + id->fti_next = NULL; + new_tp->ftt_ids = id; + break; + + case DTFTP_RETURN: + case DTFTP_POST_OFFSETS: + id->fti_next = NULL; + new_tp->ftt_retids = id; + break; + + default: + ASSERT(0); + } + + /* + * If the ISA-dependent initialization goes to plan, go back to the + * beginning and try to install this freshly made tracepoint. + */ + if (fasttrap_tracepoint_init(p, new_tp, pc, id->fti_ptype) == 0) + goto again; + + new_tp->ftt_ids = NULL; + new_tp->ftt_retids = NULL; + + return (FASTTRAP_ENABLE_FAIL); +} + +static void +fasttrap_tracepoint_disable(proc_t *p, fasttrap_probe_t *probe, uint_t index) +{ + fasttrap_bucket_t *bucket; + fasttrap_provider_t *provider = probe->ftp_prov; + fasttrap_tracepoint_t **pp, *tp; + fasttrap_id_t *id, **idp; + pid_t pid; + uintptr_t pc; + + ASSERT(index < probe->ftp_ntps); + + pid = probe->ftp_pid; + pc = probe->ftp_tps[index].fit_tp->ftt_pc; + id = &probe->ftp_tps[index].fit_id; + + ASSERT(probe->ftp_tps[index].fit_tp->ftt_pid == pid); + + /* + * Find the tracepoint and make sure that our id is one of the + * ones registered with it. + */ + bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; + mutex_enter(&bucket->ftb_mtx); + for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { + if (tp->ftt_pid == pid && tp->ftt_pc == pc && + tp->ftt_proc == provider->ftp_proc) + break; + } + + /* + * If we somehow lost this tracepoint, we're in a world of hurt. + */ + ASSERT(tp != NULL); + + switch (id->fti_ptype) { + case DTFTP_ENTRY: + case DTFTP_OFFSETS: + case DTFTP_IS_ENABLED: + ASSERT(tp->ftt_ids != NULL); + idp = &tp->ftt_ids; + break; + + case DTFTP_RETURN: + case DTFTP_POST_OFFSETS: + ASSERT(tp->ftt_retids != NULL); + idp = &tp->ftt_retids; + break; + + default: + ASSERT(0); + } + + while ((*idp)->fti_probe != probe) { + idp = &(*idp)->fti_next; + ASSERT(*idp != NULL); + } + + id = *idp; + *idp = id->fti_next; + membar_producer(); + + ASSERT(id->fti_probe == probe); + + /* + * If there are other registered enablings of this tracepoint, we're + * all done, but if this was the last probe assocated with this + * this tracepoint, we need to remove and free it. + */ + if (tp->ftt_ids != NULL || tp->ftt_retids != NULL) { + + /* + * If the current probe's tracepoint is in use, swap it + * for an unused tracepoint. + */ + if (tp == probe->ftp_tps[index].fit_tp) { + fasttrap_probe_t *tmp_probe; + fasttrap_tracepoint_t **tmp_tp; + uint_t tmp_index; + + if (tp->ftt_ids != NULL) { + tmp_probe = tp->ftt_ids->fti_probe; + /* LINTED - alignment */ + tmp_index = FASTTRAP_ID_INDEX(tp->ftt_ids); + tmp_tp = &tmp_probe->ftp_tps[tmp_index].fit_tp; + } else { + tmp_probe = tp->ftt_retids->fti_probe; + /* LINTED - alignment */ + tmp_index = FASTTRAP_ID_INDEX(tp->ftt_retids); + tmp_tp = &tmp_probe->ftp_tps[tmp_index].fit_tp; + } + + ASSERT(*tmp_tp != NULL); + ASSERT(*tmp_tp != probe->ftp_tps[index].fit_tp); + ASSERT((*tmp_tp)->ftt_ids == NULL); + ASSERT((*tmp_tp)->ftt_retids == NULL); + + probe->ftp_tps[index].fit_tp = *tmp_tp; + *tmp_tp = tp; + } + + mutex_exit(&bucket->ftb_mtx); + + /* + * Tag the modified probe with the generation in which it was + * changed. + */ + probe->ftp_gen = fasttrap_mod_gen; + return; + } + + mutex_exit(&bucket->ftb_mtx); + + /* + * We can't safely remove the tracepoint from the set of active + * tracepoints until we've actually removed the fasttrap instruction + * from the process's text. We can, however, operate on this + * tracepoint secure in the knowledge that no other thread is going to + * be looking at it since we hold P_PR_LOCK on the process if it's + * live or we hold the provider lock on the process if it's dead and + * gone. + */ + + /* + * We only need to remove the actual instruction if we're looking + * at an existing process + */ + if (p != NULL) { + /* + * If we fail to restore the instruction we need to kill + * this process since it's in a completely unrecoverable + * state. + */ + if (fasttrap_tracepoint_remove(p, tp) != 0) + fasttrap_sigtrap(p, NULL, pc); + + /* + * Decrement the count of the number of tracepoints active + * in the victim process. + */ + ASSERT(p->p_proc_flag & P_PR_LOCK); + p->p_dtrace_count--; + } + + /* + * Remove the probe from the hash table of active tracepoints. + */ + mutex_enter(&bucket->ftb_mtx); + pp = (fasttrap_tracepoint_t **)&bucket->ftb_data; + ASSERT(*pp != NULL); + while (*pp != tp) { + pp = &(*pp)->ftt_next; + ASSERT(*pp != NULL); + } + + *pp = tp->ftt_next; + membar_producer(); + + mutex_exit(&bucket->ftb_mtx); + + /* + * Tag the modified probe with the generation in which it was changed. + */ + probe->ftp_gen = fasttrap_mod_gen; +} + +static void +fasttrap_enable_callbacks(void) +{ + /* + * We don't have to play the rw lock game here because we're + * providing something rather than taking something away -- + * we can be sure that no threads have tried to follow this + * function pointer yet. + */ + mutex_enter(&fasttrap_count_mtx); + if (fasttrap_pid_count == 0) { + ASSERT(dtrace_pid_probe_ptr == NULL); + ASSERT(dtrace_return_probe_ptr == NULL); + dtrace_pid_probe_ptr = &fasttrap_pid_probe; + dtrace_return_probe_ptr = &fasttrap_return_probe; + } + ASSERT(dtrace_pid_probe_ptr == &fasttrap_pid_probe); + ASSERT(dtrace_return_probe_ptr == &fasttrap_return_probe); + fasttrap_pid_count++; + mutex_exit(&fasttrap_count_mtx); +} + +static void +fasttrap_disable_callbacks(void) +{ + ASSERT(MUTEX_HELD(&cpu_lock)); + + mutex_enter(&fasttrap_count_mtx); + ASSERT(fasttrap_pid_count > 0); + fasttrap_pid_count--; + if (fasttrap_pid_count == 0) { + cpu_t *cur, *cpu = CPU; + + for (cur = cpu->cpu_next_onln; cur != cpu; + cur = cur->cpu_next_onln) { + rw_enter(&cur->cpu_ft_lock, RW_WRITER); + } + + dtrace_pid_probe_ptr = NULL; + dtrace_return_probe_ptr = NULL; + + for (cur = cpu->cpu_next_onln; cur != cpu; + cur = cur->cpu_next_onln) { + rw_exit(&cur->cpu_ft_lock); + } + } + mutex_exit(&fasttrap_count_mtx); +} + +/*ARGSUSED*/ +static void +fasttrap_pid_enable(void *arg, dtrace_id_t id, void *parg) +{ + fasttrap_probe_t *probe = parg; + proc_t *p; + int i, rc; + + ASSERT(probe != NULL); + ASSERT(!probe->ftp_enabled); + ASSERT(id == probe->ftp_id); + ASSERT(MUTEX_HELD(&cpu_lock)); + + /* + * Increment the count of enabled probes on this probe's provider; + * the provider can't go away while the probe still exists. We + * must increment this even if we aren't able to properly enable + * this probe. + */ + mutex_enter(&probe->ftp_prov->ftp_mtx); + probe->ftp_prov->ftp_rcount++; + mutex_exit(&probe->ftp_prov->ftp_mtx); + + /* + * If this probe's provider is retired (meaning it was valid in a + * previously exec'ed incarnation of this address space), bail out. The + * provider can't go away while we're in this code path. + */ + if (probe->ftp_prov->ftp_retired) + return; + + /* + * If we can't find the process, it may be that we're in the context of + * a fork in which the traced process is being born and we're copying + * USDT probes. Otherwise, the process is gone so bail. + */ + if ((p = sprlock(probe->ftp_pid)) == NULL) { + if ((curproc->p_flag & SFORKING) == 0) + return; + + mutex_enter(&pidlock); + p = prfind(probe->ftp_pid); + + /* + * Confirm that curproc is indeed forking the process in which + * we're trying to enable probes. + */ + ASSERT(p != NULL); + ASSERT(p->p_parent == curproc); + ASSERT(p->p_stat == SIDL); + + mutex_enter(&p->p_lock); + mutex_exit(&pidlock); + + sprlock_proc(p); + } + + ASSERT(!(p->p_flag & SVFORK)); + mutex_exit(&p->p_lock); + + /* + * We have to enable the trap entry point before any user threads have + * the chance to execute the trap instruction we're about to place + * in their process's text. + */ + fasttrap_enable_callbacks(); + + /* + * Enable all the tracepoints and add this probe's id to each + * tracepoint's list of active probes. + */ + for (i = 0; i < probe->ftp_ntps; i++) { + if ((rc = fasttrap_tracepoint_enable(p, probe, i)) != 0) { + /* + * If enabling the tracepoint failed completely, + * we don't have to disable it; if the failure + * was only partial we must disable it. + */ + if (rc == FASTTRAP_ENABLE_FAIL) + i--; + else + ASSERT(rc == FASTTRAP_ENABLE_PARTIAL); + + /* + * Back up and pull out all the tracepoints we've + * created so far for this probe. + */ + while (i >= 0) { + fasttrap_tracepoint_disable(p, probe, i); + i--; + } + + mutex_enter(&p->p_lock); + sprunlock(p); + + /* + * Since we're not actually enabling this probe, + * drop our reference on the trap table entry. + */ + fasttrap_disable_callbacks(); + return; + } + } + + mutex_enter(&p->p_lock); + sprunlock(p); + + probe->ftp_enabled = 1; +} + +/*ARGSUSED*/ +static void +fasttrap_pid_disable(void *arg, dtrace_id_t id, void *parg) +{ + fasttrap_probe_t *probe = parg; + fasttrap_provider_t *provider = probe->ftp_prov; + proc_t *p; + int i, whack = 0; + + ASSERT(id == probe->ftp_id); + + /* + * We won't be able to acquire a /proc-esque lock on the process + * iff the process is dead and gone. In this case, we rely on the + * provider lock as a point of mutual exclusion to prevent other + * DTrace consumers from disabling this probe. + */ + if ((p = sprlock(probe->ftp_pid)) != NULL) { + ASSERT(!(p->p_flag & SVFORK)); + mutex_exit(&p->p_lock); + } + + mutex_enter(&provider->ftp_mtx); + + /* + * Disable all the associated tracepoints (for fully enabled probes). + */ + if (probe->ftp_enabled) { + for (i = 0; i < probe->ftp_ntps; i++) { + fasttrap_tracepoint_disable(p, probe, i); + } + } + + ASSERT(provider->ftp_rcount > 0); + provider->ftp_rcount--; + + if (p != NULL) { + /* + * Even though we may not be able to remove it entirely, we + * mark this retired provider to get a chance to remove some + * of the associated probes. + */ + if (provider->ftp_retired && !provider->ftp_marked) + whack = provider->ftp_marked = 1; + mutex_exit(&provider->ftp_mtx); + + mutex_enter(&p->p_lock); + sprunlock(p); + } else { + /* + * If the process is dead, we're just waiting for the + * last probe to be disabled to be able to free it. + */ + if (provider->ftp_rcount == 0 && !provider->ftp_marked) + whack = provider->ftp_marked = 1; + mutex_exit(&provider->ftp_mtx); + } + + if (whack) + fasttrap_pid_cleanup(); + + if (!probe->ftp_enabled) + return; + + probe->ftp_enabled = 0; + + ASSERT(MUTEX_HELD(&cpu_lock)); + fasttrap_disable_callbacks(); +} + +/*ARGSUSED*/ +static void +fasttrap_pid_getargdesc(void *arg, dtrace_id_t id, void *parg, + dtrace_argdesc_t *desc) +{ + fasttrap_probe_t *probe = parg; + char *str; + int i, ndx; + + desc->dtargd_native[0] = '\0'; + desc->dtargd_xlate[0] = '\0'; + + if (probe->ftp_prov->ftp_retired != 0 || + desc->dtargd_ndx >= probe->ftp_nargs) { + desc->dtargd_ndx = DTRACE_ARGNONE; + return; + } + + ndx = (probe->ftp_argmap != NULL) ? + probe->ftp_argmap[desc->dtargd_ndx] : desc->dtargd_ndx; + + str = probe->ftp_ntypes; + for (i = 0; i < ndx; i++) { + str += strlen(str) + 1; + } + + ASSERT(strlen(str + 1) < sizeof (desc->dtargd_native)); + (void) strcpy(desc->dtargd_native, str); + + if (probe->ftp_xtypes == NULL) + return; + + str = probe->ftp_xtypes; + for (i = 0; i < desc->dtargd_ndx; i++) { + str += strlen(str) + 1; + } + + ASSERT(strlen(str + 1) < sizeof (desc->dtargd_xlate)); + (void) strcpy(desc->dtargd_xlate, str); +} + +/*ARGSUSED*/ +static void +fasttrap_pid_destroy(void *arg, dtrace_id_t id, void *parg) +{ + fasttrap_probe_t *probe = parg; + int i; + size_t size; + + ASSERT(probe != NULL); + ASSERT(!probe->ftp_enabled); + ASSERT(fasttrap_total >= probe->ftp_ntps); + + atomic_add_32(&fasttrap_total, -probe->ftp_ntps); + size = offsetof(fasttrap_probe_t, ftp_tps[probe->ftp_ntps]); + + if (probe->ftp_gen + 1 >= fasttrap_mod_gen) + fasttrap_mod_barrier(probe->ftp_gen); + + for (i = 0; i < probe->ftp_ntps; i++) { + kmem_free(probe->ftp_tps[i].fit_tp, + sizeof (fasttrap_tracepoint_t)); + } + + kmem_free(probe, size); +} + + +static const dtrace_pattr_t pid_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +}; + +static dtrace_pops_t pid_pops = { + fasttrap_pid_provide, + NULL, + fasttrap_pid_enable, + fasttrap_pid_disable, + NULL, + NULL, + fasttrap_pid_getargdesc, + fasttrap_pid_getarg, + NULL, + fasttrap_pid_destroy +}; + +static dtrace_pops_t usdt_pops = { + fasttrap_pid_provide, + NULL, + fasttrap_pid_enable, + fasttrap_pid_disable, + NULL, + NULL, + fasttrap_pid_getargdesc, + fasttrap_usdt_getarg, + NULL, + fasttrap_pid_destroy +}; + +static fasttrap_proc_t * +fasttrap_proc_lookup(pid_t pid) +{ + fasttrap_bucket_t *bucket; + fasttrap_proc_t *fprc, *new_fprc; + + bucket = &fasttrap_procs.fth_table[FASTTRAP_PROCS_INDEX(pid)]; + mutex_enter(&bucket->ftb_mtx); + + for (fprc = bucket->ftb_data; fprc != NULL; fprc = fprc->ftpc_next) { + if (fprc->ftpc_pid == pid && fprc->ftpc_acount != 0) { + mutex_enter(&fprc->ftpc_mtx); + mutex_exit(&bucket->ftb_mtx); + fprc->ftpc_rcount++; + atomic_add_64(&fprc->ftpc_acount, 1); + ASSERT(fprc->ftpc_acount <= fprc->ftpc_rcount); + mutex_exit(&fprc->ftpc_mtx); + + return (fprc); + } + } + + /* + * Drop the bucket lock so we don't try to perform a sleeping + * allocation under it. + */ + mutex_exit(&bucket->ftb_mtx); + + new_fprc = kmem_zalloc(sizeof (fasttrap_proc_t), KM_SLEEP); + new_fprc->ftpc_pid = pid; + new_fprc->ftpc_rcount = 1; + new_fprc->ftpc_acount = 1; + + mutex_enter(&bucket->ftb_mtx); + + /* + * Take another lap through the list to make sure a proc hasn't + * been created for this pid while we weren't under the bucket lock. + */ + for (fprc = bucket->ftb_data; fprc != NULL; fprc = fprc->ftpc_next) { + if (fprc->ftpc_pid == pid && fprc->ftpc_acount != 0) { + mutex_enter(&fprc->ftpc_mtx); + mutex_exit(&bucket->ftb_mtx); + fprc->ftpc_rcount++; + atomic_add_64(&fprc->ftpc_acount, 1); + ASSERT(fprc->ftpc_acount <= fprc->ftpc_rcount); + mutex_exit(&fprc->ftpc_mtx); + + kmem_free(new_fprc, sizeof (fasttrap_proc_t)); + + return (fprc); + } + } + + new_fprc->ftpc_next = bucket->ftb_data; + bucket->ftb_data = new_fprc; + + mutex_exit(&bucket->ftb_mtx); + + return (new_fprc); +} + +static void +fasttrap_proc_release(fasttrap_proc_t *proc) +{ + fasttrap_bucket_t *bucket; + fasttrap_proc_t *fprc, **fprcp; + pid_t pid = proc->ftpc_pid; + + mutex_enter(&proc->ftpc_mtx); + + ASSERT(proc->ftpc_rcount != 0); + ASSERT(proc->ftpc_acount <= proc->ftpc_rcount); + + if (--proc->ftpc_rcount != 0) { + mutex_exit(&proc->ftpc_mtx); + return; + } + + mutex_exit(&proc->ftpc_mtx); + + /* + * There should definitely be no live providers associated with this + * process at this point. + */ + ASSERT(proc->ftpc_acount == 0); + + bucket = &fasttrap_procs.fth_table[FASTTRAP_PROCS_INDEX(pid)]; + mutex_enter(&bucket->ftb_mtx); + + fprcp = (fasttrap_proc_t **)&bucket->ftb_data; + while ((fprc = *fprcp) != NULL) { + if (fprc == proc) + break; + + fprcp = &fprc->ftpc_next; + } + + /* + * Something strange has happened if we can't find the proc. + */ + ASSERT(fprc != NULL); + + *fprcp = fprc->ftpc_next; + + mutex_exit(&bucket->ftb_mtx); + + kmem_free(fprc, sizeof (fasttrap_proc_t)); +} + +/* + * Lookup a fasttrap-managed provider based on its name and associated pid. + * If the pattr argument is non-NULL, this function instantiates the provider + * if it doesn't exist otherwise it returns NULL. The provider is returned + * with its lock held. + */ +static fasttrap_provider_t * +fasttrap_provider_lookup(pid_t pid, const char *name, + const dtrace_pattr_t *pattr) +{ + fasttrap_provider_t *fp, *new_fp = NULL; + fasttrap_bucket_t *bucket; + char provname[DTRACE_PROVNAMELEN]; + proc_t *p; + cred_t *cred; + + ASSERT(strlen(name) < sizeof (fp->ftp_name)); + ASSERT(pattr != NULL); + + bucket = &fasttrap_provs.fth_table[FASTTRAP_PROVS_INDEX(pid, name)]; + mutex_enter(&bucket->ftb_mtx); + + /* + * Take a lap through the list and return the match if we find it. + */ + for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { + if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && + !fp->ftp_retired) { + mutex_enter(&fp->ftp_mtx); + mutex_exit(&bucket->ftb_mtx); + return (fp); + } + } + + /* + * Drop the bucket lock so we don't try to perform a sleeping + * allocation under it. + */ + mutex_exit(&bucket->ftb_mtx); + + /* + * Make sure the process exists, isn't a child created as the result + * of a vfork(2), and isn't a zombie (but may be in fork). + */ + mutex_enter(&pidlock); + if ((p = prfind(pid)) == NULL) { + mutex_exit(&pidlock); + return (NULL); + } + mutex_enter(&p->p_lock); + mutex_exit(&pidlock); + if (p->p_flag & (SVFORK | SEXITING)) { + mutex_exit(&p->p_lock); + return (NULL); + } + + /* + * Increment p_dtrace_probes so that the process knows to inform us + * when it exits or execs. fasttrap_provider_free() decrements this + * when we're done with this provider. + */ + p->p_dtrace_probes++; + + /* + * Grab the credentials for this process so we have + * something to pass to dtrace_register(). + */ + mutex_enter(&p->p_crlock); + crhold(p->p_cred); + cred = p->p_cred; + mutex_exit(&p->p_crlock); + mutex_exit(&p->p_lock); + + new_fp = kmem_zalloc(sizeof (fasttrap_provider_t), KM_SLEEP); + new_fp->ftp_pid = pid; + new_fp->ftp_proc = fasttrap_proc_lookup(pid); + + ASSERT(new_fp->ftp_proc != NULL); + + mutex_enter(&bucket->ftb_mtx); + + /* + * Take another lap through the list to make sure a provider hasn't + * been created for this pid while we weren't under the bucket lock. + */ + for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { + if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && + !fp->ftp_retired) { + mutex_enter(&fp->ftp_mtx); + mutex_exit(&bucket->ftb_mtx); + fasttrap_provider_free(new_fp); + crfree(cred); + return (fp); + } + } + + (void) strcpy(new_fp->ftp_name, name); + + /* + * Fail and return NULL if either the provider name is too long + * or we fail to register this new provider with the DTrace + * framework. Note that this is the only place we ever construct + * the full provider name -- we keep it in pieces in the provider + * structure. + */ + if (snprintf(provname, sizeof (provname), "%s%u", name, (uint_t)pid) >= + sizeof (provname) || + dtrace_register(provname, pattr, + DTRACE_PRIV_PROC | DTRACE_PRIV_OWNER | DTRACE_PRIV_ZONEOWNER, cred, + pattr == &pid_attr ? &pid_pops : &usdt_pops, new_fp, + &new_fp->ftp_provid) != 0) { + mutex_exit(&bucket->ftb_mtx); + fasttrap_provider_free(new_fp); + crfree(cred); + return (NULL); + } + + new_fp->ftp_next = bucket->ftb_data; + bucket->ftb_data = new_fp; + + mutex_enter(&new_fp->ftp_mtx); + mutex_exit(&bucket->ftb_mtx); + + crfree(cred); + return (new_fp); +} + +static void +fasttrap_provider_free(fasttrap_provider_t *provider) +{ + pid_t pid = provider->ftp_pid; + proc_t *p; + + /* + * There need to be no associated enabled probes, no consumers + * creating probes, and no meta providers referencing this provider. + */ + ASSERT(provider->ftp_rcount == 0); + ASSERT(provider->ftp_ccount == 0); + ASSERT(provider->ftp_mcount == 0); + + /* + * If this provider hasn't been retired, we need to explicitly drop the + * count of active providers on the associated process structure. + */ + if (!provider->ftp_retired) { + atomic_add_64(&provider->ftp_proc->ftpc_acount, -1); + ASSERT(provider->ftp_proc->ftpc_acount < + provider->ftp_proc->ftpc_rcount); + } + + fasttrap_proc_release(provider->ftp_proc); + + kmem_free(provider, sizeof (fasttrap_provider_t)); + + /* + * Decrement p_dtrace_probes on the process whose provider we're + * freeing. We don't have to worry about clobbering somone else's + * modifications to it because we have locked the bucket that + * corresponds to this process's hash chain in the provider hash + * table. Don't sweat it if we can't find the process. + */ + mutex_enter(&pidlock); + if ((p = prfind(pid)) == NULL) { + mutex_exit(&pidlock); + return; + } + + mutex_enter(&p->p_lock); + mutex_exit(&pidlock); + + p->p_dtrace_probes--; + mutex_exit(&p->p_lock); +} + +static void +fasttrap_provider_retire(pid_t pid, const char *name, int mprov) +{ + fasttrap_provider_t *fp; + fasttrap_bucket_t *bucket; + dtrace_provider_id_t provid; + + ASSERT(strlen(name) < sizeof (fp->ftp_name)); + + bucket = &fasttrap_provs.fth_table[FASTTRAP_PROVS_INDEX(pid, name)]; + mutex_enter(&bucket->ftb_mtx); + + for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { + if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && + !fp->ftp_retired) + break; + } + + if (fp == NULL) { + mutex_exit(&bucket->ftb_mtx); + return; + } + + mutex_enter(&fp->ftp_mtx); + ASSERT(!mprov || fp->ftp_mcount > 0); + if (mprov && --fp->ftp_mcount != 0) { + mutex_exit(&fp->ftp_mtx); + mutex_exit(&bucket->ftb_mtx); + return; + } + + /* + * Mark the provider to be removed in our post-processing step, mark it + * retired, and drop the active count on its proc. Marking it indicates + * that we should try to remove it; setting the retired flag indicates + * that we're done with this provider; dropping the active the proc + * releases our hold, and when this reaches zero (as it will during + * exit or exec) the proc and associated providers become defunct. + * + * We obviously need to take the bucket lock before the provider lock + * to perform the lookup, but we need to drop the provider lock + * before calling into the DTrace framework since we acquire the + * provider lock in callbacks invoked from the DTrace framework. The + * bucket lock therefore protects the integrity of the provider hash + * table. + */ + atomic_add_64(&fp->ftp_proc->ftpc_acount, -1); + ASSERT(fp->ftp_proc->ftpc_acount < fp->ftp_proc->ftpc_rcount); + + fp->ftp_retired = 1; + fp->ftp_marked = 1; + provid = fp->ftp_provid; + mutex_exit(&fp->ftp_mtx); + + /* + * We don't have to worry about invalidating the same provider twice + * since fasttrap_provider_lookup() will ignore provider that have + * been marked as retired. + */ + dtrace_invalidate(provid); + + mutex_exit(&bucket->ftb_mtx); + + fasttrap_pid_cleanup(); +} + +static int +fasttrap_uint32_cmp(const void *ap, const void *bp) +{ + return (*(const uint32_t *)ap - *(const uint32_t *)bp); +} + +static int +fasttrap_uint64_cmp(const void *ap, const void *bp) +{ + return (*(const uint64_t *)ap - *(const uint64_t *)bp); +} + +static int +fasttrap_add_probe(fasttrap_probe_spec_t *pdata) +{ + fasttrap_provider_t *provider; + fasttrap_probe_t *pp; + fasttrap_tracepoint_t *tp; + char *name; + int i, aframes, whack; + + /* + * There needs to be at least one desired trace point. + */ + if (pdata->ftps_noffs == 0) + return (EINVAL); + + switch (pdata->ftps_type) { + case DTFTP_ENTRY: + name = "entry"; + aframes = FASTTRAP_ENTRY_AFRAMES; + break; + case DTFTP_RETURN: + name = "return"; + aframes = FASTTRAP_RETURN_AFRAMES; + break; + case DTFTP_OFFSETS: + name = NULL; + break; + default: + return (EINVAL); + } + + if ((provider = fasttrap_provider_lookup(pdata->ftps_pid, + FASTTRAP_PID_NAME, &pid_attr)) == NULL) + return (ESRCH); + + /* + * Increment this reference count to indicate that a consumer is + * actively adding a new probe associated with this provider. This + * prevents the provider from being deleted -- we'll need to check + * for pending deletions when we drop this reference count. + */ + provider->ftp_ccount++; + mutex_exit(&provider->ftp_mtx); + + /* + * Grab the creation lock to ensure consistency between calls to + * dtrace_probe_lookup() and dtrace_probe_create() in the face of + * other threads creating probes. We must drop the provider lock + * before taking this lock to avoid a three-way deadlock with the + * DTrace framework. + */ + mutex_enter(&provider->ftp_cmtx); + + if (name == NULL) { + for (i = 0; i < pdata->ftps_noffs; i++) { + char name_str[17]; + + (void) sprintf(name_str, "%llx", + (unsigned long long)pdata->ftps_offs[i]); + + if (dtrace_probe_lookup(provider->ftp_provid, + pdata->ftps_mod, pdata->ftps_func, name_str) != 0) + continue; + + atomic_add_32(&fasttrap_total, 1); + + if (fasttrap_total > fasttrap_max) { + atomic_add_32(&fasttrap_total, -1); + goto no_mem; + } + + pp = kmem_zalloc(sizeof (fasttrap_probe_t), KM_SLEEP); + + pp->ftp_prov = provider; + pp->ftp_faddr = pdata->ftps_pc; + pp->ftp_fsize = pdata->ftps_size; + pp->ftp_pid = pdata->ftps_pid; + pp->ftp_ntps = 1; + + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), + KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = pdata->ftps_offs[i] + pdata->ftps_pc; + tp->ftt_pid = pdata->ftps_pid; + + pp->ftp_tps[0].fit_tp = tp; + pp->ftp_tps[0].fit_id.fti_probe = pp; + pp->ftp_tps[0].fit_id.fti_ptype = pdata->ftps_type; + + pp->ftp_id = dtrace_probe_create(provider->ftp_provid, + pdata->ftps_mod, pdata->ftps_func, name_str, + FASTTRAP_OFFSET_AFRAMES, pp); + } + + } else if (dtrace_probe_lookup(provider->ftp_provid, pdata->ftps_mod, + pdata->ftps_func, name) == 0) { + atomic_add_32(&fasttrap_total, pdata->ftps_noffs); + + if (fasttrap_total > fasttrap_max) { + atomic_add_32(&fasttrap_total, -pdata->ftps_noffs); + goto no_mem; + } + + /* + * Make sure all tracepoint program counter values are unique. + * We later assume that each probe has exactly one tracepoint + * for a given pc. + */ + qsort(pdata->ftps_offs, pdata->ftps_noffs, + sizeof (uint64_t), fasttrap_uint64_cmp); + for (i = 1; i < pdata->ftps_noffs; i++) { + if (pdata->ftps_offs[i] > pdata->ftps_offs[i - 1]) + continue; + + atomic_add_32(&fasttrap_total, -pdata->ftps_noffs); + goto no_mem; + } + + ASSERT(pdata->ftps_noffs > 0); + pp = kmem_zalloc(offsetof(fasttrap_probe_t, + ftp_tps[pdata->ftps_noffs]), KM_SLEEP); + + pp->ftp_prov = provider; + pp->ftp_faddr = pdata->ftps_pc; + pp->ftp_fsize = pdata->ftps_size; + pp->ftp_pid = pdata->ftps_pid; + pp->ftp_ntps = pdata->ftps_noffs; + + for (i = 0; i < pdata->ftps_noffs; i++) { + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), + KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = pdata->ftps_offs[i] + pdata->ftps_pc; + tp->ftt_pid = pdata->ftps_pid; + + pp->ftp_tps[i].fit_tp = tp; + pp->ftp_tps[i].fit_id.fti_probe = pp; + pp->ftp_tps[i].fit_id.fti_ptype = pdata->ftps_type; + } + + pp->ftp_id = dtrace_probe_create(provider->ftp_provid, + pdata->ftps_mod, pdata->ftps_func, name, aframes, pp); + } + + mutex_exit(&provider->ftp_cmtx); + + /* + * We know that the provider is still valid since we incremented the + * creation reference count. If someone tried to clean up this provider + * while we were using it (e.g. because the process called exec(2) or + * exit(2)), take note of that and try to clean it up now. + */ + mutex_enter(&provider->ftp_mtx); + provider->ftp_ccount--; + whack = provider->ftp_retired; + mutex_exit(&provider->ftp_mtx); + + if (whack) + fasttrap_pid_cleanup(); + + return (0); + +no_mem: + /* + * If we've exhausted the allowable resources, we'll try to remove + * this provider to free some up. This is to cover the case where + * the user has accidentally created many more probes than was + * intended (e.g. pid123:::). + */ + mutex_exit(&provider->ftp_cmtx); + mutex_enter(&provider->ftp_mtx); + provider->ftp_ccount--; + provider->ftp_marked = 1; + mutex_exit(&provider->ftp_mtx); + + fasttrap_pid_cleanup(); + + return (ENOMEM); +} + +/*ARGSUSED*/ +static void * +fasttrap_meta_provide(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid) +{ + fasttrap_provider_t *provider; + + /* + * A 32-bit unsigned integer (like a pid for example) can be + * expressed in 10 or fewer decimal digits. Make sure that we'll + * have enough space for the provider name. + */ + if (strlen(dhpv->dthpv_provname) + 10 >= + sizeof (provider->ftp_name)) { + cmn_err(CE_WARN, "failed to instantiate provider %s: " + "name too long to accomodate pid", dhpv->dthpv_provname); + return (NULL); + } + + /* + * Don't let folks spoof the true pid provider. + */ + if (strcmp(dhpv->dthpv_provname, FASTTRAP_PID_NAME) == 0) { + cmn_err(CE_WARN, "failed to instantiate provider %s: " + "%s is an invalid name", dhpv->dthpv_provname, + FASTTRAP_PID_NAME); + return (NULL); + } + + /* + * The highest stability class that fasttrap supports is ISA; cap + * the stability of the new provider accordingly. + */ + if (dhpv->dthpv_pattr.dtpa_provider.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_provider.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_mod.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_mod.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_func.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_func.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_name.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_name.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_args.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_args.dtat_class = DTRACE_CLASS_ISA; + + if ((provider = fasttrap_provider_lookup(pid, dhpv->dthpv_provname, + &dhpv->dthpv_pattr)) == NULL) { + cmn_err(CE_WARN, "failed to instantiate provider %s for " + "process %u", dhpv->dthpv_provname, (uint_t)pid); + return (NULL); + } + + /* + * Up the meta provider count so this provider isn't removed until + * the meta provider has been told to remove it. + */ + provider->ftp_mcount++; + + mutex_exit(&provider->ftp_mtx); + + return (provider); +} + +/*ARGSUSED*/ +static void +fasttrap_meta_create_probe(void *arg, void *parg, + dtrace_helper_probedesc_t *dhpb) +{ + fasttrap_provider_t *provider = parg; + fasttrap_probe_t *pp; + fasttrap_tracepoint_t *tp; + int i, j; + uint32_t ntps; + + /* + * Since the meta provider count is non-zero we don't have to worry + * about this provider disappearing. + */ + ASSERT(provider->ftp_mcount > 0); + + /* + * The offsets must be unique. + */ + qsort(dhpb->dthpb_offs, dhpb->dthpb_noffs, sizeof (uint32_t), + fasttrap_uint32_cmp); + for (i = 1; i < dhpb->dthpb_noffs; i++) { + if (dhpb->dthpb_base + dhpb->dthpb_offs[i] <= + dhpb->dthpb_base + dhpb->dthpb_offs[i - 1]) + return; + } + + qsort(dhpb->dthpb_enoffs, dhpb->dthpb_nenoffs, sizeof (uint32_t), + fasttrap_uint32_cmp); + for (i = 1; i < dhpb->dthpb_nenoffs; i++) { + if (dhpb->dthpb_base + dhpb->dthpb_enoffs[i] <= + dhpb->dthpb_base + dhpb->dthpb_enoffs[i - 1]) + return; + } + + /* + * Grab the creation lock to ensure consistency between calls to + * dtrace_probe_lookup() and dtrace_probe_create() in the face of + * other threads creating probes. + */ + mutex_enter(&provider->ftp_cmtx); + + if (dtrace_probe_lookup(provider->ftp_provid, dhpb->dthpb_mod, + dhpb->dthpb_func, dhpb->dthpb_name) != 0) { + mutex_exit(&provider->ftp_cmtx); + return; + } + + ntps = dhpb->dthpb_noffs + dhpb->dthpb_nenoffs; + ASSERT(ntps > 0); + + atomic_add_32(&fasttrap_total, ntps); + + if (fasttrap_total > fasttrap_max) { + atomic_add_32(&fasttrap_total, -ntps); + mutex_exit(&provider->ftp_cmtx); + return; + } + + pp = kmem_zalloc(offsetof(fasttrap_probe_t, ftp_tps[ntps]), KM_SLEEP); + + pp->ftp_prov = provider; + pp->ftp_pid = provider->ftp_pid; + pp->ftp_ntps = ntps; + pp->ftp_nargs = dhpb->dthpb_xargc; + pp->ftp_xtypes = dhpb->dthpb_xtypes; + pp->ftp_ntypes = dhpb->dthpb_ntypes; + + /* + * First create a tracepoint for each actual point of interest. + */ + for (i = 0; i < dhpb->dthpb_noffs; i++) { + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = dhpb->dthpb_base + dhpb->dthpb_offs[i]; + tp->ftt_pid = provider->ftp_pid; + + pp->ftp_tps[i].fit_tp = tp; + pp->ftp_tps[i].fit_id.fti_probe = pp; +#ifdef __sparc + pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_POST_OFFSETS; +#else + pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_OFFSETS; +#endif + } + + /* + * Then create a tracepoint for each is-enabled point. + */ + for (j = 0; i < ntps; i++, j++) { + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = dhpb->dthpb_base + dhpb->dthpb_enoffs[j]; + tp->ftt_pid = provider->ftp_pid; + + pp->ftp_tps[i].fit_tp = tp; + pp->ftp_tps[i].fit_id.fti_probe = pp; + pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_IS_ENABLED; + } + + /* + * If the arguments are shuffled around we set the argument remapping + * table. Later, when the probe fires, we only remap the arguments + * if the table is non-NULL. + */ + for (i = 0; i < dhpb->dthpb_xargc; i++) { + if (dhpb->dthpb_args[i] != i) { + pp->ftp_argmap = dhpb->dthpb_args; + break; + } + } + + /* + * The probe is fully constructed -- register it with DTrace. + */ + pp->ftp_id = dtrace_probe_create(provider->ftp_provid, dhpb->dthpb_mod, + dhpb->dthpb_func, dhpb->dthpb_name, FASTTRAP_OFFSET_AFRAMES, pp); + + mutex_exit(&provider->ftp_cmtx); +} + +/*ARGSUSED*/ +static void +fasttrap_meta_remove(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid) +{ + /* + * Clean up the USDT provider. There may be active consumers of the + * provider busy adding probes, no damage will actually befall the + * provider until that count has dropped to zero. This just puts + * the provider on death row. + */ + fasttrap_provider_retire(pid, dhpv->dthpv_provname, 1); +} + +static dtrace_mops_t fasttrap_mops = { + fasttrap_meta_create_probe, + fasttrap_meta_provide, + fasttrap_meta_remove +}; + +/*ARGSUSED*/ +static int +fasttrap_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +/*ARGSUSED*/ +static int +fasttrap_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv) +{ + if (!dtrace_attached()) + return (EAGAIN); + + if (cmd == FASTTRAPIOC_MAKEPROBE) { + fasttrap_probe_spec_t *uprobe = (void *)arg; + fasttrap_probe_spec_t *probe; + uint64_t noffs; + size_t size; + int ret; + char *c; + + if (copyin(&uprobe->ftps_noffs, &noffs, + sizeof (uprobe->ftps_noffs))) + return (EFAULT); + + /* + * Probes must have at least one tracepoint. + */ + if (noffs == 0) + return (EINVAL); + + size = sizeof (fasttrap_probe_spec_t) + + sizeof (probe->ftps_offs[0]) * (noffs - 1); + + if (size > 1024 * 1024) + return (ENOMEM); + + probe = kmem_alloc(size, KM_SLEEP); + + if (copyin(uprobe, probe, size) != 0) { + kmem_free(probe, size); + return (EFAULT); + } + + /* + * Verify that the function and module strings contain no + * funny characters. + */ + for (c = &probe->ftps_func[0]; *c != '\0'; c++) { + if (*c < 0x20 || 0x7f <= *c) { + ret = EINVAL; + goto err; + } + } + + for (c = &probe->ftps_mod[0]; *c != '\0'; c++) { + if (*c < 0x20 || 0x7f <= *c) { + ret = EINVAL; + goto err; + } + } + + if (!PRIV_POLICY_CHOICE(cr, PRIV_ALL, B_FALSE)) { + proc_t *p; + pid_t pid = probe->ftps_pid; + + mutex_enter(&pidlock); + /* + * Report an error if the process doesn't exist + * or is actively being birthed. + */ + if ((p = prfind(pid)) == NULL || p->p_stat == SIDL) { + mutex_exit(&pidlock); + return (ESRCH); + } + mutex_enter(&p->p_lock); + mutex_exit(&pidlock); + + if ((ret = priv_proc_cred_perm(cr, p, NULL, + VREAD | VWRITE)) != 0) { + mutex_exit(&p->p_lock); + return (ret); + } + + mutex_exit(&p->p_lock); + } + + ret = fasttrap_add_probe(probe); +err: + kmem_free(probe, size); + + return (ret); + + } else if (cmd == FASTTRAPIOC_GETINSTR) { + fasttrap_instr_query_t instr; + fasttrap_tracepoint_t *tp; + uint_t index; + int ret; + + if (copyin((void *)arg, &instr, sizeof (instr)) != 0) + return (EFAULT); + + if (!PRIV_POLICY_CHOICE(cr, PRIV_ALL, B_FALSE)) { + proc_t *p; + pid_t pid = instr.ftiq_pid; + + mutex_enter(&pidlock); + /* + * Report an error if the process doesn't exist + * or is actively being birthed. + */ + if ((p = prfind(pid)) == NULL || p->p_stat == SIDL) { + mutex_exit(&pidlock); + return (ESRCH); + } + mutex_enter(&p->p_lock); + mutex_exit(&pidlock); + + if ((ret = priv_proc_cred_perm(cr, p, NULL, + VREAD)) != 0) { + mutex_exit(&p->p_lock); + return (ret); + } + + mutex_exit(&p->p_lock); + } + + index = FASTTRAP_TPOINTS_INDEX(instr.ftiq_pid, instr.ftiq_pc); + + mutex_enter(&fasttrap_tpoints.fth_table[index].ftb_mtx); + tp = fasttrap_tpoints.fth_table[index].ftb_data; + while (tp != NULL) { + if (instr.ftiq_pid == tp->ftt_pid && + instr.ftiq_pc == tp->ftt_pc && + tp->ftt_proc->ftpc_acount != 0) + break; + + tp = tp->ftt_next; + } + + if (tp == NULL) { + mutex_exit(&fasttrap_tpoints.fth_table[index].ftb_mtx); + return (ENOENT); + } + + bcopy(&tp->ftt_instr, &instr.ftiq_instr, + sizeof (instr.ftiq_instr)); + mutex_exit(&fasttrap_tpoints.fth_table[index].ftb_mtx); + + if (copyout(&instr, (void *)arg, sizeof (instr)) != 0) + return (EFAULT); + + return (0); + } + + return (EINVAL); +} + +static struct cb_ops fasttrap_cb_ops = { + fasttrap_open, /* open */ + nodev, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + fasttrap_ioctl, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_NEW | D_MP /* Driver compatibility flag */ +}; + +/*ARGSUSED*/ +static int +fasttrap_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)fasttrap_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +static int +fasttrap_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + ulong_t nent; + + switch (cmd) { + case DDI_ATTACH: + break; + case DDI_RESUME: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (ddi_create_minor_node(devi, "fasttrap", S_IFCHR, 0, + DDI_PSEUDO, NULL) == DDI_FAILURE) { + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + ddi_report_dev(devi); + fasttrap_devi = devi; + + /* + * Install our hooks into fork(2), exec(2), and exit(2). + */ + dtrace_fasttrap_fork_ptr = &fasttrap_fork; + dtrace_fasttrap_exit_ptr = &fasttrap_exec_exit; + dtrace_fasttrap_exec_ptr = &fasttrap_exec_exit; + + fasttrap_max = ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS, + "fasttrap-max-probes", FASTTRAP_MAX_DEFAULT); + fasttrap_total = 0; + + /* + * Conjure up the tracepoints hashtable... + */ + nent = ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS, + "fasttrap-hash-size", FASTTRAP_TPOINTS_DEFAULT_SIZE); + + if (nent == 0 || nent > 0x1000000) + nent = FASTTRAP_TPOINTS_DEFAULT_SIZE; + + if ((nent & (nent - 1)) == 0) + fasttrap_tpoints.fth_nent = nent; + else + fasttrap_tpoints.fth_nent = 1 << fasttrap_highbit(nent); + ASSERT(fasttrap_tpoints.fth_nent > 0); + fasttrap_tpoints.fth_mask = fasttrap_tpoints.fth_nent - 1; + fasttrap_tpoints.fth_table = kmem_zalloc(fasttrap_tpoints.fth_nent * + sizeof (fasttrap_bucket_t), KM_SLEEP); + + /* + * ... and the providers hash table... + */ + nent = FASTTRAP_PROVIDERS_DEFAULT_SIZE; + if ((nent & (nent - 1)) == 0) + fasttrap_provs.fth_nent = nent; + else + fasttrap_provs.fth_nent = 1 << fasttrap_highbit(nent); + ASSERT(fasttrap_provs.fth_nent > 0); + fasttrap_provs.fth_mask = fasttrap_provs.fth_nent - 1; + fasttrap_provs.fth_table = kmem_zalloc(fasttrap_provs.fth_nent * + sizeof (fasttrap_bucket_t), KM_SLEEP); + + /* + * ... and the procs hash table. + */ + nent = FASTTRAP_PROCS_DEFAULT_SIZE; + if ((nent & (nent - 1)) == 0) + fasttrap_procs.fth_nent = nent; + else + fasttrap_procs.fth_nent = 1 << fasttrap_highbit(nent); + ASSERT(fasttrap_procs.fth_nent > 0); + fasttrap_procs.fth_mask = fasttrap_procs.fth_nent - 1; + fasttrap_procs.fth_table = kmem_zalloc(fasttrap_procs.fth_nent * + sizeof (fasttrap_bucket_t), KM_SLEEP); + + (void) dtrace_meta_register("fasttrap", &fasttrap_mops, NULL, + &fasttrap_meta_id); + + return (DDI_SUCCESS); +} + +static int +fasttrap_detach(dev_info_t *devi, ddi_detach_cmd_t cmd) +{ + int i, fail = 0; + timeout_id_t tmp; + + switch (cmd) { + case DDI_DETACH: + break; + case DDI_SUSPEND: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + /* + * Unregister the meta-provider to make sure no new fasttrap- + * managed providers come along while we're trying to close up + * shop. If we fail to detach, we'll need to re-register as a + * meta-provider. We can fail to unregister as a meta-provider + * if providers we manage still exist. + */ + if (fasttrap_meta_id != DTRACE_METAPROVNONE && + dtrace_meta_unregister(fasttrap_meta_id) != 0) + return (DDI_FAILURE); + + /* + * Prevent any new timeouts from running by setting fasttrap_timeout + * to a non-zero value, and wait for the current timeout to complete. + */ + mutex_enter(&fasttrap_cleanup_mtx); + fasttrap_cleanup_work = 0; + + while (fasttrap_timeout != (timeout_id_t)1) { + tmp = fasttrap_timeout; + fasttrap_timeout = (timeout_id_t)1; + + if (tmp != 0) { + mutex_exit(&fasttrap_cleanup_mtx); + (void) untimeout(tmp); + mutex_enter(&fasttrap_cleanup_mtx); + } + } + + fasttrap_cleanup_work = 0; + mutex_exit(&fasttrap_cleanup_mtx); + + /* + * Iterate over all of our providers. If there's still a process + * that corresponds to that pid, fail to detach. + */ + for (i = 0; i < fasttrap_provs.fth_nent; i++) { + fasttrap_provider_t **fpp, *fp; + fasttrap_bucket_t *bucket = &fasttrap_provs.fth_table[i]; + + mutex_enter(&bucket->ftb_mtx); + fpp = (fasttrap_provider_t **)&bucket->ftb_data; + while ((fp = *fpp) != NULL) { + /* + * Acquire and release the lock as a simple way of + * waiting for any other consumer to finish with + * this provider. A thread must first acquire the + * bucket lock so there's no chance of another thread + * blocking on the provider's lock. + */ + mutex_enter(&fp->ftp_mtx); + mutex_exit(&fp->ftp_mtx); + + if (dtrace_unregister(fp->ftp_provid) != 0) { + fail = 1; + fpp = &fp->ftp_next; + } else { + *fpp = fp->ftp_next; + fasttrap_provider_free(fp); + } + } + + mutex_exit(&bucket->ftb_mtx); + } + + if (fail) { + uint_t work; + /* + * If we're failing to detach, we need to unblock timeouts + * and start a new timeout if any work has accumulated while + * we've been unsuccessfully trying to detach. + */ + mutex_enter(&fasttrap_cleanup_mtx); + fasttrap_timeout = 0; + work = fasttrap_cleanup_work; + mutex_exit(&fasttrap_cleanup_mtx); + + if (work) + fasttrap_pid_cleanup(); + + (void) dtrace_meta_register("fasttrap", &fasttrap_mops, NULL, + &fasttrap_meta_id); + + return (DDI_FAILURE); + } + +#ifdef DEBUG + mutex_enter(&fasttrap_count_mtx); + ASSERT(fasttrap_pid_count == 0); + mutex_exit(&fasttrap_count_mtx); +#endif + + kmem_free(fasttrap_tpoints.fth_table, + fasttrap_tpoints.fth_nent * sizeof (fasttrap_bucket_t)); + fasttrap_tpoints.fth_nent = 0; + + kmem_free(fasttrap_provs.fth_table, + fasttrap_provs.fth_nent * sizeof (fasttrap_bucket_t)); + fasttrap_provs.fth_nent = 0; + + kmem_free(fasttrap_procs.fth_table, + fasttrap_procs.fth_nent * sizeof (fasttrap_bucket_t)); + fasttrap_procs.fth_nent = 0; + + /* + * We know there are no tracepoints in any process anywhere in + * the system so there is no process which has its p_dtrace_count + * greater than zero, therefore we know that no thread can actively + * be executing code in fasttrap_fork(). Similarly for p_dtrace_probes + * and fasttrap_exec() and fasttrap_exit(). + */ + ASSERT(dtrace_fasttrap_fork_ptr == &fasttrap_fork); + dtrace_fasttrap_fork_ptr = NULL; + + ASSERT(dtrace_fasttrap_exec_ptr == &fasttrap_exec_exit); + dtrace_fasttrap_exec_ptr = NULL; + + ASSERT(dtrace_fasttrap_exit_ptr == &fasttrap_exec_exit); + dtrace_fasttrap_exit_ptr = NULL; + + ddi_remove_minor_node(devi, NULL); + + return (DDI_SUCCESS); +} + +static struct dev_ops fasttrap_ops = { + DEVO_REV, /* devo_rev */ + 0, /* refcnt */ + fasttrap_info, /* get_dev_info */ + nulldev, /* identify */ + nulldev, /* probe */ + fasttrap_attach, /* attach */ + fasttrap_detach, /* detach */ + nodev, /* reset */ + &fasttrap_cb_ops, /* driver operations */ + NULL, /* bus operations */ + nodev /* dev power */ +}; + +/* + * Module linkage information for the kernel. + */ +static struct modldrv modldrv = { + &mod_driverops, /* module type (this is a pseudo driver) */ + "Fasttrap Tracing", /* name of module */ + &fasttrap_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, + (void *)&modldrv, + NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} diff --git a/sys/cddl/contrib/opensolaris/uts/common/dtrace/lockstat.c b/sys/cddl/contrib/opensolaris/uts/common/dtrace/lockstat.c new file mode 100644 index 000000000000..3eb76a061d32 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/dtrace/lockstat.c @@ -0,0 +1,341 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/types.h> +#include <sys/param.h> +#include <sys/stat.h> +#include <sys/open.h> +#include <sys/file.h> +#include <sys/conf.h> +#include <sys/modctl.h> +#include <sys/cmn_err.h> +#include <sys/bitmap.h> +#include <sys/debug.h> +#include <sys/kmem.h> +#include <sys/errno.h> +#include <sys/sysmacros.h> +#include <sys/lockstat.h> +#include <sys/atomic.h> +#include <sys/dtrace.h> + +#include <sys/ddi.h> +#include <sys/sunddi.h> + +typedef struct lockstat_probe { + const char *lsp_func; + const char *lsp_name; + int lsp_probe; + dtrace_id_t lsp_id; +} lockstat_probe_t; + +lockstat_probe_t lockstat_probes[] = +{ + { LS_MUTEX_ENTER, LSA_ACQUIRE, LS_MUTEX_ENTER_ACQUIRE }, + { LS_MUTEX_ENTER, LSA_BLOCK, LS_MUTEX_ENTER_BLOCK }, + { LS_MUTEX_ENTER, LSA_SPIN, LS_MUTEX_ENTER_SPIN }, + { LS_MUTEX_EXIT, LSA_RELEASE, LS_MUTEX_EXIT_RELEASE }, + { LS_MUTEX_DESTROY, LSA_RELEASE, LS_MUTEX_DESTROY_RELEASE }, + { LS_MUTEX_TRYENTER, LSA_ACQUIRE, LS_MUTEX_TRYENTER_ACQUIRE }, + { LS_LOCK_SET, LSS_ACQUIRE, LS_LOCK_SET_ACQUIRE }, + { LS_LOCK_SET, LSS_SPIN, LS_LOCK_SET_SPIN }, + { LS_LOCK_SET_SPL, LSS_ACQUIRE, LS_LOCK_SET_SPL_ACQUIRE }, + { LS_LOCK_SET_SPL, LSS_SPIN, LS_LOCK_SET_SPL_SPIN }, + { LS_LOCK_TRY, LSS_ACQUIRE, LS_LOCK_TRY_ACQUIRE }, + { LS_LOCK_CLEAR, LSS_RELEASE, LS_LOCK_CLEAR_RELEASE }, + { LS_LOCK_CLEAR_SPLX, LSS_RELEASE, LS_LOCK_CLEAR_SPLX_RELEASE }, + { LS_CLOCK_UNLOCK, LSS_RELEASE, LS_CLOCK_UNLOCK_RELEASE }, + { LS_RW_ENTER, LSR_ACQUIRE, LS_RW_ENTER_ACQUIRE }, + { LS_RW_ENTER, LSR_BLOCK, LS_RW_ENTER_BLOCK }, + { LS_RW_EXIT, LSR_RELEASE, LS_RW_EXIT_RELEASE }, + { LS_RW_TRYENTER, LSR_ACQUIRE, LS_RW_TRYENTER_ACQUIRE }, + { LS_RW_TRYUPGRADE, LSR_UPGRADE, LS_RW_TRYUPGRADE_UPGRADE }, + { LS_RW_DOWNGRADE, LSR_DOWNGRADE, LS_RW_DOWNGRADE_DOWNGRADE }, + { LS_THREAD_LOCK, LST_SPIN, LS_THREAD_LOCK_SPIN }, + { LS_THREAD_LOCK_HIGH, LST_SPIN, LS_THREAD_LOCK_HIGH_SPIN }, + { NULL } +}; + +static dev_info_t *lockstat_devi; /* saved in xxattach() for xxinfo() */ +static kmutex_t lockstat_test; /* for testing purposes only */ +static dtrace_provider_id_t lockstat_id; + +/*ARGSUSED*/ +static void +lockstat_enable(void *arg, dtrace_id_t id, void *parg) +{ + lockstat_probe_t *probe = parg; + + ASSERT(!lockstat_probemap[probe->lsp_probe]); + + lockstat_probemap[probe->lsp_probe] = id; + membar_producer(); + + lockstat_hot_patch(); + membar_producer(); + + /* + * Immediately generate a record for the lockstat_test mutex + * to verify that the mutex hot-patch code worked as expected. + */ + mutex_enter(&lockstat_test); + mutex_exit(&lockstat_test); +} + +/*ARGSUSED*/ +static void +lockstat_disable(void *arg, dtrace_id_t id, void *parg) +{ + lockstat_probe_t *probe = parg; + int i; + + ASSERT(lockstat_probemap[probe->lsp_probe]); + + lockstat_probemap[probe->lsp_probe] = 0; + lockstat_hot_patch(); + membar_producer(); + + /* + * See if we have any probes left enabled. + */ + for (i = 0; i < LS_NPROBES; i++) { + if (lockstat_probemap[i]) { + /* + * This probe is still enabled. We don't need to deal + * with waiting for all threads to be out of the + * lockstat critical sections; just return. + */ + return; + } + } + + /* + * The delay() here isn't as cheesy as you might think. We don't + * want to busy-loop in the kernel, so we have to give up the + * CPU between calls to lockstat_active_threads(); that much is + * obvious. But the reason it's a do..while loop rather than a + * while loop is subtle. The memory barrier above guarantees that + * no threads will enter the lockstat code from this point forward. + * However, another thread could already be executing lockstat code + * without our knowledge if the update to its t_lockstat field hasn't + * cleared its CPU's store buffer. Delaying for one clock tick + * guarantees that either (1) the thread will have *ample* time to + * complete its work, or (2) the thread will be preempted, in which + * case it will have to grab and release a dispatcher lock, which + * will flush that CPU's store buffer. Either way we're covered. + */ + do { + delay(1); + } while (lockstat_active_threads()); +} + +/*ARGSUSED*/ +static int +lockstat_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +/* ARGSUSED */ +static int +lockstat_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *) lockstat_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +/*ARGSUSED*/ +static void +lockstat_provide(void *arg, const dtrace_probedesc_t *desc) +{ + int i = 0; + + for (i = 0; lockstat_probes[i].lsp_func != NULL; i++) { + lockstat_probe_t *probe = &lockstat_probes[i]; + + if (dtrace_probe_lookup(lockstat_id, "genunix", + probe->lsp_func, probe->lsp_name) != 0) + continue; + + ASSERT(!probe->lsp_id); + probe->lsp_id = dtrace_probe_create(lockstat_id, + "genunix", probe->lsp_func, probe->lsp_name, + 1, probe); + } +} + +/*ARGSUSED*/ +static void +lockstat_destroy(void *arg, dtrace_id_t id, void *parg) +{ + lockstat_probe_t *probe = parg; + + ASSERT(!lockstat_probemap[probe->lsp_probe]); + probe->lsp_id = 0; +} + +static dtrace_pattr_t lockstat_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +}; + +static dtrace_pops_t lockstat_pops = { + lockstat_provide, + NULL, + lockstat_enable, + lockstat_disable, + NULL, + NULL, + NULL, + NULL, + NULL, + lockstat_destroy +}; + +static int +lockstat_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + switch (cmd) { + case DDI_ATTACH: + break; + case DDI_RESUME: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (ddi_create_minor_node(devi, "lockstat", S_IFCHR, 0, + DDI_PSEUDO, 0) == DDI_FAILURE || + dtrace_register("lockstat", &lockstat_attr, DTRACE_PRIV_KERNEL, + NULL, &lockstat_pops, NULL, &lockstat_id) != 0) { + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + lockstat_probe = dtrace_probe; + membar_producer(); + + ddi_report_dev(devi); + lockstat_devi = devi; + return (DDI_SUCCESS); +} + +static int +lockstat_detach(dev_info_t *devi, ddi_detach_cmd_t cmd) +{ + switch (cmd) { + case DDI_DETACH: + break; + case DDI_SUSPEND: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (dtrace_unregister(lockstat_id) != 0) + return (DDI_FAILURE); + + ddi_remove_minor_node(devi, NULL); + return (DDI_SUCCESS); +} + +/* + * Configuration data structures + */ +static struct cb_ops lockstat_cb_ops = { + lockstat_open, /* open */ + nodev, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + nodev, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_MP | D_NEW /* Driver compatibility flag */ +}; + +static struct dev_ops lockstat_ops = { + DEVO_REV, /* devo_rev, */ + 0, /* refcnt */ + lockstat_info, /* getinfo */ + nulldev, /* identify */ + nulldev, /* probe */ + lockstat_attach, /* attach */ + lockstat_detach, /* detach */ + nulldev, /* reset */ + &lockstat_cb_ops, /* cb_ops */ + NULL, /* bus_ops */ +}; + +static struct modldrv modldrv = { + &mod_driverops, /* Type of module. This one is a driver */ + "Lock Statistics %I%", /* name of module */ + &lockstat_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, (void *)&modldrv, NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} diff --git a/sys/cddl/contrib/opensolaris/uts/common/dtrace/profile.c b/sys/cddl/contrib/opensolaris/uts/common/dtrace/profile.c new file mode 100644 index 000000000000..8de919a851a2 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/dtrace/profile.c @@ -0,0 +1,576 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/errno.h> +#include <sys/stat.h> +#include <sys/modctl.h> +#include <sys/conf.h> +#include <sys/systm.h> +#include <sys/ddi.h> +#include <sys/sunddi.h> +#include <sys/cpuvar.h> +#include <sys/kmem.h> +#include <sys/strsubr.h> +#include <sys/dtrace.h> +#include <sys/cyclic.h> +#include <sys/atomic.h> + +static dev_info_t *profile_devi; +static dtrace_provider_id_t profile_id; + +/* + * Regardless of platform, the stack frames look like this in the case of the + * profile provider: + * + * profile_fire + * cyclic_expire + * cyclic_fire + * [ cbe ] + * [ interrupt code ] + * + * On x86, there are five frames from the generic interrupt code; further, the + * interrupted instruction appears as its own stack frame, giving us a total of + * 10. + * + * On SPARC, the picture is further complicated because the compiler + * optimizes away tail-calls -- so the following frames are optimized away: + * + * profile_fire + * cyclic_expire + * + * This gives three frames. However, on DEBUG kernels, the cyclic_expire + * frame cannot be tail-call eliminated, yielding four frames in this case. + * + * All of the above constraints lead to the mess below. Yes, the profile + * provider should ideally figure this out on-the-fly by hitting one of its own + * probes and then walking its own stack trace. This is complicated, however, + * and the static definition doesn't seem to be overly brittle. Still, we + * allow for a manual override in case we get it completely wrong. + */ +#ifdef __x86 +#define PROF_ARTIFICIAL_FRAMES 10 +#else +#ifdef __sparc +#ifdef DEBUG +#define PROF_ARTIFICIAL_FRAMES 4 +#else +#define PROF_ARTIFICIAL_FRAMES 3 +#endif +#endif +#endif + +#define PROF_NAMELEN 15 + +#define PROF_PROFILE 0 +#define PROF_TICK 1 +#define PROF_PREFIX_PROFILE "profile-" +#define PROF_PREFIX_TICK "tick-" + +typedef struct profile_probe { + char prof_name[PROF_NAMELEN]; + dtrace_id_t prof_id; + int prof_kind; + hrtime_t prof_interval; + cyclic_id_t prof_cyclic; +} profile_probe_t; + +typedef struct profile_probe_percpu { + hrtime_t profc_expected; + hrtime_t profc_interval; + profile_probe_t *profc_probe; +} profile_probe_percpu_t; + +hrtime_t profile_interval_min = NANOSEC / 5000; /* 5000 hz */ +int profile_aframes = 0; /* override */ + +static int profile_rates[] = { + 97, 199, 499, 997, 1999, + 4001, 4999, 0, 0, 0, + 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0 +}; + +static int profile_ticks[] = { + 1, 10, 100, 500, 1000, + 5000, 0, 0, 0, 0, + 0, 0, 0, 0, 0 +}; + +/* + * profile_max defines the upper bound on the number of profile probes that + * can exist (this is to prevent malicious or clumsy users from exhausing + * system resources by creating a slew of profile probes). At mod load time, + * this gets its value from PROFILE_MAX_DEFAULT or profile-max-probes if it's + * present in the profile.conf file. + */ +#define PROFILE_MAX_DEFAULT 1000 /* default max. number of probes */ +static uint32_t profile_max; /* maximum number of profile probes */ +static uint32_t profile_total; /* current number of profile probes */ + +static void +profile_fire(void *arg) +{ + profile_probe_percpu_t *pcpu = arg; + profile_probe_t *prof = pcpu->profc_probe; + hrtime_t late; + + late = dtrace_gethrtime() - pcpu->profc_expected; + pcpu->profc_expected += pcpu->profc_interval; + + dtrace_probe(prof->prof_id, CPU->cpu_profile_pc, + CPU->cpu_profile_upc, late, 0, 0); +} + +static void +profile_tick(void *arg) +{ + profile_probe_t *prof = arg; + + dtrace_probe(prof->prof_id, CPU->cpu_profile_pc, + CPU->cpu_profile_upc, 0, 0, 0); +} + +static void +profile_create(hrtime_t interval, const char *name, int kind) +{ + profile_probe_t *prof; + int nr_frames = PROF_ARTIFICIAL_FRAMES + dtrace_mach_aframes(); + + if (profile_aframes) + nr_frames = profile_aframes; + + if (interval < profile_interval_min) + return; + + if (dtrace_probe_lookup(profile_id, NULL, NULL, name) != 0) + return; + + atomic_add_32(&profile_total, 1); + if (profile_total > profile_max) { + atomic_add_32(&profile_total, -1); + return; + } + + prof = kmem_zalloc(sizeof (profile_probe_t), KM_SLEEP); + (void) strcpy(prof->prof_name, name); + prof->prof_interval = interval; + prof->prof_cyclic = CYCLIC_NONE; + prof->prof_kind = kind; + prof->prof_id = dtrace_probe_create(profile_id, + NULL, NULL, name, nr_frames, prof); +} + +/*ARGSUSED*/ +static void +profile_provide(void *arg, const dtrace_probedesc_t *desc) +{ + int i, j, rate, kind; + hrtime_t val = 0, mult = 1, len; + const char *name, *suffix = NULL; + + const struct { + char *prefix; + int kind; + } types[] = { + { PROF_PREFIX_PROFILE, PROF_PROFILE }, + { PROF_PREFIX_TICK, PROF_TICK }, + { NULL, NULL } + }; + + const struct { + char *name; + hrtime_t mult; + } suffixes[] = { + { "ns", NANOSEC / NANOSEC }, + { "nsec", NANOSEC / NANOSEC }, + { "us", NANOSEC / MICROSEC }, + { "usec", NANOSEC / MICROSEC }, + { "ms", NANOSEC / MILLISEC }, + { "msec", NANOSEC / MILLISEC }, + { "s", NANOSEC / SEC }, + { "sec", NANOSEC / SEC }, + { "m", NANOSEC * (hrtime_t)60 }, + { "min", NANOSEC * (hrtime_t)60 }, + { "h", NANOSEC * (hrtime_t)(60 * 60) }, + { "hour", NANOSEC * (hrtime_t)(60 * 60) }, + { "d", NANOSEC * (hrtime_t)(24 * 60 * 60) }, + { "day", NANOSEC * (hrtime_t)(24 * 60 * 60) }, + { "hz", 0 }, + { NULL } + }; + + if (desc == NULL) { + char n[PROF_NAMELEN]; + + /* + * If no description was provided, provide all of our probes. + */ + for (i = 0; i < sizeof (profile_rates) / sizeof (int); i++) { + if ((rate = profile_rates[i]) == 0) + continue; + + (void) snprintf(n, PROF_NAMELEN, "%s%d", + PROF_PREFIX_PROFILE, rate); + profile_create(NANOSEC / rate, n, PROF_PROFILE); + } + + for (i = 0; i < sizeof (profile_ticks) / sizeof (int); i++) { + if ((rate = profile_ticks[i]) == 0) + continue; + + (void) snprintf(n, PROF_NAMELEN, "%s%d", + PROF_PREFIX_TICK, rate); + profile_create(NANOSEC / rate, n, PROF_TICK); + } + + return; + } + + name = desc->dtpd_name; + + for (i = 0; types[i].prefix != NULL; i++) { + len = strlen(types[i].prefix); + + if (strncmp(name, types[i].prefix, len) != 0) + continue; + break; + } + + if (types[i].prefix == NULL) + return; + + kind = types[i].kind; + j = strlen(name) - len; + + /* + * We need to start before any time suffix. + */ + for (j = strlen(name); j >= len; j--) { + if (name[j] >= '0' && name[j] <= '9') + break; + suffix = &name[j]; + } + + ASSERT(suffix != NULL); + + /* + * Now determine the numerical value present in the probe name. + */ + for (; j >= len; j--) { + if (name[j] < '0' || name[j] > '9') + return; + + val += (name[j] - '0') * mult; + mult *= (hrtime_t)10; + } + + if (val == 0) + return; + + /* + * Look-up the suffix to determine the multiplier. + */ + for (i = 0, mult = 0; suffixes[i].name != NULL; i++) { + if (strcasecmp(suffixes[i].name, suffix) == 0) { + mult = suffixes[i].mult; + break; + } + } + + if (suffixes[i].name == NULL && *suffix != '\0') + return; + + if (mult == 0) { + /* + * The default is frequency-per-second. + */ + val = NANOSEC / val; + } else { + val *= mult; + } + + profile_create(val, name, kind); +} + +/*ARGSUSED*/ +static void +profile_destroy(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; + + ASSERT(prof->prof_cyclic == CYCLIC_NONE); + kmem_free(prof, sizeof (profile_probe_t)); + + ASSERT(profile_total >= 1); + atomic_add_32(&profile_total, -1); +} + +/*ARGSUSED*/ +static void +profile_online(void *arg, cpu_t *cpu, cyc_handler_t *hdlr, cyc_time_t *when) +{ + profile_probe_t *prof = arg; + profile_probe_percpu_t *pcpu; + + pcpu = kmem_zalloc(sizeof (profile_probe_percpu_t), KM_SLEEP); + pcpu->profc_probe = prof; + + hdlr->cyh_func = profile_fire; + hdlr->cyh_arg = pcpu; + hdlr->cyh_level = CY_HIGH_LEVEL; + + when->cyt_interval = prof->prof_interval; + when->cyt_when = dtrace_gethrtime() + when->cyt_interval; + + pcpu->profc_expected = when->cyt_when; + pcpu->profc_interval = when->cyt_interval; +} + +/*ARGSUSED*/ +static void +profile_offline(void *arg, cpu_t *cpu, void *oarg) +{ + profile_probe_percpu_t *pcpu = oarg; + + ASSERT(pcpu->profc_probe == arg); + kmem_free(pcpu, sizeof (profile_probe_percpu_t)); +} + +/*ARGSUSED*/ +static void +profile_enable(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; + cyc_omni_handler_t omni; + cyc_handler_t hdlr; + cyc_time_t when; + + ASSERT(prof->prof_interval != 0); + ASSERT(MUTEX_HELD(&cpu_lock)); + + if (prof->prof_kind == PROF_TICK) { + hdlr.cyh_func = profile_tick; + hdlr.cyh_arg = prof; + hdlr.cyh_level = CY_HIGH_LEVEL; + + when.cyt_interval = prof->prof_interval; + when.cyt_when = dtrace_gethrtime() + when.cyt_interval; + } else { + ASSERT(prof->prof_kind == PROF_PROFILE); + omni.cyo_online = profile_online; + omni.cyo_offline = profile_offline; + omni.cyo_arg = prof; + } + + if (prof->prof_kind == PROF_TICK) { + prof->prof_cyclic = cyclic_add(&hdlr, &when); + } else { + prof->prof_cyclic = cyclic_add_omni(&omni); + } +} + +/*ARGSUSED*/ +static void +profile_disable(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; + + ASSERT(prof->prof_cyclic != CYCLIC_NONE); + ASSERT(MUTEX_HELD(&cpu_lock)); + + cyclic_remove(prof->prof_cyclic); + prof->prof_cyclic = CYCLIC_NONE; +} + +/*ARGSUSED*/ +static int +profile_usermode(void *arg, dtrace_id_t id, void *parg) +{ + return (CPU->cpu_profile_pc == 0); +} + +static dtrace_pattr_t profile_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_UNSTABLE, DTRACE_STABILITY_UNSTABLE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +}; + +static dtrace_pops_t profile_pops = { + profile_provide, + NULL, + profile_enable, + profile_disable, + NULL, + NULL, + NULL, + NULL, + profile_usermode, + profile_destroy +}; + +static int +profile_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + switch (cmd) { + case DDI_ATTACH: + break; + case DDI_RESUME: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (ddi_create_minor_node(devi, "profile", S_IFCHR, 0, + DDI_PSEUDO, NULL) == DDI_FAILURE || + dtrace_register("profile", &profile_attr, + DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER, NULL, + &profile_pops, NULL, &profile_id) != 0) { + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + profile_max = ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS, + "profile-max-probes", PROFILE_MAX_DEFAULT); + + ddi_report_dev(devi); + profile_devi = devi; + return (DDI_SUCCESS); +} + +static int +profile_detach(dev_info_t *devi, ddi_detach_cmd_t cmd) +{ + switch (cmd) { + case DDI_DETACH: + break; + case DDI_SUSPEND: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (dtrace_unregister(profile_id) != 0) + return (DDI_FAILURE); + + ddi_remove_minor_node(devi, NULL); + return (DDI_SUCCESS); +} + +/*ARGSUSED*/ +static int +profile_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)profile_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +/*ARGSUSED*/ +static int +profile_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +static struct cb_ops profile_cb_ops = { + profile_open, /* open */ + nodev, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + nodev, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_NEW | D_MP /* Driver compatibility flag */ +}; + +static struct dev_ops profile_ops = { + DEVO_REV, /* devo_rev, */ + 0, /* refcnt */ + profile_info, /* get_dev_info */ + nulldev, /* identify */ + nulldev, /* probe */ + profile_attach, /* attach */ + profile_detach, /* detach */ + nodev, /* reset */ + &profile_cb_ops, /* driver operations */ + NULL, /* bus operations */ + nodev /* dev power */ +}; + +/* + * Module linkage information for the kernel. + */ +static struct modldrv modldrv = { + &mod_driverops, /* module type (this is a pseudo driver) */ + "Profile Interrupt Tracing", /* name of module */ + &profile_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, + (void *)&modldrv, + NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} diff --git a/sys/cddl/contrib/opensolaris/uts/common/dtrace/sdt_subr.c b/sys/cddl/contrib/opensolaris/uts/common/dtrace/sdt_subr.c new file mode 100644 index 000000000000..66ff8a92a01b --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/dtrace/sdt_subr.c @@ -0,0 +1,888 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/sdt_impl.h> + +static dtrace_pattr_t vtrace_attr = { +{ DTRACE_STABILITY_UNSTABLE, DTRACE_STABILITY_UNSTABLE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_UNSTABLE, DTRACE_STABILITY_UNSTABLE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t info_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t fpu_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_CPU }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t fsinfo_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t stab_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t sdt_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t xpv_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_PLATFORM }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_PLATFORM }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_PLATFORM }, +}; + +sdt_provider_t sdt_providers[] = { + { "vtrace", "__vtrace_", &vtrace_attr, 0 }, + { "sysinfo", "__cpu_sysinfo_", &info_attr, 0 }, + { "vminfo", "__cpu_vminfo_", &info_attr, 0 }, + { "fpuinfo", "__fpuinfo_", &fpu_attr, 0 }, + { "sched", "__sched_", &stab_attr, 0 }, + { "proc", "__proc_", &stab_attr, 0 }, + { "io", "__io_", &stab_attr, 0 }, + { "mib", "__mib_", &stab_attr, 0 }, + { "fsinfo", "__fsinfo_", &fsinfo_attr, 0 }, + { "nfsv3", "__nfsv3_", &stab_attr, 0 }, + { "nfsv4", "__nfsv4_", &stab_attr, 0 }, + { "xpv", "__xpv_", &xpv_attr, 0 }, + { "sysevent", "__sysevent_", &stab_attr, 0 }, + { "sdt", NULL, &sdt_attr, 0 }, + { NULL } +}; + +sdt_argdesc_t sdt_args[] = { + { "sched", "wakeup", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "wakeup", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "dequeue", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "dequeue", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "dequeue", 2, 1, "disp_t *", "cpuinfo_t *" }, + { "sched", "enqueue", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "enqueue", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "enqueue", 2, 1, "disp_t *", "cpuinfo_t *" }, + { "sched", "enqueue", 3, 2, "int" }, + { "sched", "off-cpu", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "off-cpu", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "tick", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "tick", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "change-pri", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "change-pri", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "change-pri", 2, 1, "pri_t" }, + { "sched", "schedctl-nopreempt", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "schedctl-nopreempt", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "schedctl-nopreempt", 2, 1, "int" }, + { "sched", "schedctl-preempt", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "schedctl-preempt", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "schedctl-yield", 0, 0, "int" }, + { "sched", "surrender", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "surrender", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "cpucaps-sleep", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "cpucaps-sleep", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "cpucaps-wakeup", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "cpucaps-wakeup", 1, 0, "kthread_t *", "psinfo_t *" }, + + { "proc", "create", 0, 0, "proc_t *", "psinfo_t *" }, + { "proc", "exec", 0, 0, "string" }, + { "proc", "exec-failure", 0, 0, "int" }, + { "proc", "exit", 0, 0, "int" }, + { "proc", "fault", 0, 0, "int" }, + { "proc", "fault", 1, 1, "siginfo_t *" }, + { "proc", "lwp-create", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "proc", "lwp-create", 1, 0, "kthread_t *", "psinfo_t *" }, + { "proc", "signal-clear", 0, 0, "int" }, + { "proc", "signal-clear", 1, 1, "siginfo_t *" }, + { "proc", "signal-discard", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "proc", "signal-discard", 1, 1, "proc_t *", "psinfo_t *" }, + { "proc", "signal-discard", 2, 2, "int" }, + { "proc", "signal-handle", 0, 0, "int" }, + { "proc", "signal-handle", 1, 1, "siginfo_t *" }, + { "proc", "signal-handle", 2, 2, "void (*)(void)" }, + { "proc", "signal-send", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "proc", "signal-send", 1, 0, "kthread_t *", "psinfo_t *" }, + { "proc", "signal-send", 2, 1, "int" }, + + { "io", "start", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "start", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "start", 2, 0, "buf_t *", "fileinfo_t *" }, + { "io", "done", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "done", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "done", 2, 0, "buf_t *", "fileinfo_t *" }, + { "io", "wait-start", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "wait-start", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "wait-start", 2, 0, "buf_t *", "fileinfo_t *" }, + { "io", "wait-done", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "wait-done", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "wait-done", 2, 0, "buf_t *", "fileinfo_t *" }, + + { "mib", NULL, 0, 0, "int" }, + + { "fsinfo", NULL, 0, 0, "vnode_t *", "fileinfo_t *" }, + { "fsinfo", NULL, 1, 1, "int", "int" }, + + { "nfsv3", "op-getattr-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-getattr-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-getattr-start", 2, 3, "GETATTR3args *" }, + { "nfsv3", "op-getattr-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-getattr-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-getattr-done", 2, 3, "GETATTR3res *" }, + { "nfsv3", "op-setattr-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-start", 2, 3, "SETATTR3args *" }, + { "nfsv3", "op-setattr-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-done", 2, 3, "SETATTR3res *" }, + { "nfsv3", "op-lookup-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-lookup-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-lookup-start", 2, 3, "LOOKUP3args *" }, + { "nfsv3", "op-lookup-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-lookup-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-lookup-done", 2, 3, "LOOKUP3res *" }, + { "nfsv3", "op-access-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-access-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-access-start", 2, 3, "ACCESS3args *" }, + { "nfsv3", "op-access-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-access-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-access-done", 2, 3, "ACCESS3res *" }, + { "nfsv3", "op-commit-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-commit-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-commit-start", 2, 3, "COMMIT3args *" }, + { "nfsv3", "op-commit-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-commit-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-commit-done", 2, 3, "COMMIT3res *" }, + { "nfsv3", "op-create-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-create-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-create-start", 2, 3, "CREATE3args *" }, + { "nfsv3", "op-create-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-create-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-create-done", 2, 3, "CREATE3res *" }, + { "nfsv3", "op-fsinfo-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsinfo-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsinfo-start", 2, 3, "FSINFO3args *" }, + { "nfsv3", "op-fsinfo-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsinfo-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsinfo-done", 2, 3, "FSINFO3res *" }, + { "nfsv3", "op-fsstat-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsstat-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsstat-start", 2, 3, "FSSTAT3args *" }, + { "nfsv3", "op-fsstat-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsstat-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsstat-done", 2, 3, "FSSTAT3res *" }, + { "nfsv3", "op-link-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-link-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-link-start", 2, 3, "LINK3args *" }, + { "nfsv3", "op-link-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-link-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-link-done", 2, 3, "LINK3res *" }, + { "nfsv3", "op-mkdir-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mkdir-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mkdir-start", 2, 3, "MKDIR3args *" }, + { "nfsv3", "op-mkdir-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mkdir-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mkdir-done", 2, 3, "MKDIR3res *" }, + { "nfsv3", "op-mknod-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mknod-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mknod-start", 2, 3, "MKNOD3args *" }, + { "nfsv3", "op-mknod-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mknod-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mknod-done", 2, 3, "MKNOD3res *" }, + { "nfsv3", "op-null-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-null-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-null-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-null-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-pathconf-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-pathconf-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-pathconf-start", 2, 3, "PATHCONF3args *" }, + { "nfsv3", "op-pathconf-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-pathconf-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-pathconf-done", 2, 3, "PATHCONF3res *" }, + { "nfsv3", "op-read-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-read-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-read-start", 2, 3, "READ3args *" }, + { "nfsv3", "op-read-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-read-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-read-done", 2, 3, "READ3res *" }, + { "nfsv3", "op-readdir-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdir-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdir-start", 2, 3, "READDIR3args *" }, + { "nfsv3", "op-readdir-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdir-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdir-done", 2, 3, "READDIR3res *" }, + { "nfsv3", "op-readdirplus-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdirplus-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdirplus-start", 2, 3, "READDIRPLUS3args *" }, + { "nfsv3", "op-readdirplus-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdirplus-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdirplus-done", 2, 3, "READDIRPLUS3res *" }, + { "nfsv3", "op-readlink-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readlink-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readlink-start", 2, 3, "READLINK3args *" }, + { "nfsv3", "op-readlink-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readlink-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readlink-done", 2, 3, "READLINK3res *" }, + { "nfsv3", "op-remove-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-remove-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-remove-start", 2, 3, "REMOVE3args *" }, + { "nfsv3", "op-remove-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-remove-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-remove-done", 2, 3, "REMOVE3res *" }, + { "nfsv3", "op-rename-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rename-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rename-start", 2, 3, "RENAME3args *" }, + { "nfsv3", "op-rename-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rename-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rename-done", 2, 3, "RENAME3res *" }, + { "nfsv3", "op-rmdir-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rmdir-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rmdir-start", 2, 3, "RMDIR3args *" }, + { "nfsv3", "op-rmdir-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rmdir-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rmdir-done", 2, 3, "RMDIR3res *" }, + { "nfsv3", "op-setattr-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-start", 2, 3, "SETATTR3args *" }, + { "nfsv3", "op-setattr-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-done", 2, 3, "SETATTR3res *" }, + { "nfsv3", "op-symlink-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-symlink-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-symlink-start", 2, 3, "SYMLINK3args *" }, + { "nfsv3", "op-symlink-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-symlink-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-symlink-done", 2, 3, "SYMLINK3res *" }, + { "nfsv3", "op-write-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-write-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-write-start", 2, 3, "WRITE3args *" }, + { "nfsv3", "op-write-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-write-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-write-done", 2, 3, "WRITE3res *" }, + + { "nfsv4", "null-start", 0, 0, "struct svc_req *", "conninfo_t *" }, + { "nfsv4", "null-done", 0, 0, "struct svc_req *", "conninfo_t *" }, + { "nfsv4", "compound-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "compound-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "compound-start", 2, 1, "COMPOUND4args *" }, + { "nfsv4", "compound-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "compound-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "compound-done", 2, 1, "COMPOUND4res *" }, + { "nfsv4", "op-access-start", 0, 0, "struct compound_state *", + "conninfo_t *"}, + { "nfsv4", "op-access-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-access-start", 2, 1, "ACCESS4args *" }, + { "nfsv4", "op-access-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-access-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-access-done", 2, 1, "ACCESS4res *" }, + { "nfsv4", "op-close-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-close-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-close-start", 2, 1, "CLOSE4args *" }, + { "nfsv4", "op-close-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-close-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-close-done", 2, 1, "CLOSE4res *" }, + { "nfsv4", "op-commit-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-commit-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-commit-start", 2, 1, "COMMIT4args *" }, + { "nfsv4", "op-commit-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-commit-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-commit-done", 2, 1, "COMMIT4res *" }, + { "nfsv4", "op-create-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-create-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-create-start", 2, 1, "CREATE4args *" }, + { "nfsv4", "op-create-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-create-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-create-done", 2, 1, "CREATE4res *" }, + { "nfsv4", "op-delegpurge-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegpurge-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegpurge-start", 2, 1, "DELEGPURGE4args *" }, + { "nfsv4", "op-delegpurge-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegpurge-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegpurge-done", 2, 1, "DELEGPURGE4res *" }, + { "nfsv4", "op-delegreturn-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegreturn-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegreturn-start", 2, 1, "DELEGRETURN4args *" }, + { "nfsv4", "op-delegreturn-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegreturn-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegreturn-done", 2, 1, "DELEGRETURN4res *" }, + { "nfsv4", "op-getattr-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getattr-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getattr-start", 2, 1, "GETATTR4args *" }, + { "nfsv4", "op-getattr-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getattr-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getattr-done", 2, 1, "GETATTR4res *" }, + { "nfsv4", "op-getfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getfh-done", 2, 1, "GETFH4res *" }, + { "nfsv4", "op-link-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-link-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-link-start", 2, 1, "LINK4args *" }, + { "nfsv4", "op-link-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-link-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-link-done", 2, 1, "LINK4res *" }, + { "nfsv4", "op-lock-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lock-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lock-start", 2, 1, "LOCK4args *" }, + { "nfsv4", "op-lock-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lock-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lock-done", 2, 1, "LOCK4res *" }, + { "nfsv4", "op-lockt-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lockt-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lockt-start", 2, 1, "LOCKT4args *" }, + { "nfsv4", "op-lockt-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lockt-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lockt-done", 2, 1, "LOCKT4res *" }, + { "nfsv4", "op-locku-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-locku-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-locku-start", 2, 1, "LOCKU4args *" }, + { "nfsv4", "op-locku-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-locku-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-locku-done", 2, 1, "LOCKU4res *" }, + { "nfsv4", "op-lookup-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookup-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookup-start", 2, 1, "LOOKUP4args *" }, + { "nfsv4", "op-lookup-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookup-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookup-done", 2, 1, "LOOKUP4res *" }, + { "nfsv4", "op-lookupp-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookupp-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookupp-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookupp-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookupp-done", 2, 1, "LOOKUPP4res *" }, + { "nfsv4", "op-nverify-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-nverify-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-nverify-start", 2, 1, "NVERIFY4args *" }, + { "nfsv4", "op-nverify-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-nverify-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-nverify-done", 2, 1, "NVERIFY4res *" }, + { "nfsv4", "op-open-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-start", 2, 1, "OPEN4args *" }, + { "nfsv4", "op-open-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-done", 2, 1, "OPEN4res *" }, + { "nfsv4", "op-open-confirm-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-confirm-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-confirm-start", 2, 1, "OPEN_CONFIRM4args *" }, + { "nfsv4", "op-open-confirm-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-confirm-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-confirm-done", 2, 1, "OPEN_CONFIRM4res *" }, + { "nfsv4", "op-open-downgrade-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-downgrade-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-downgrade-start", 2, 1, "OPEN_DOWNGRADE4args *" }, + { "nfsv4", "op-open-downgrade-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-downgrade-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-downgrade-done", 2, 1, "OPEN_DOWNGRADE4res *" }, + { "nfsv4", "op-openattr-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-openattr-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-openattr-start", 2, 1, "OPENATTR4args *" }, + { "nfsv4", "op-openattr-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-openattr-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-openattr-done", 2, 1, "OPENATTR4res *" }, + { "nfsv4", "op-putfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putfh-start", 2, 1, "PUTFH4args *" }, + { "nfsv4", "op-putfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putfh-done", 2, 1, "PUTFH4res *" }, + { "nfsv4", "op-putpubfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putpubfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putpubfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putpubfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putpubfh-done", 2, 1, "PUTPUBFH4res *" }, + { "nfsv4", "op-putrootfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putrootfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putrootfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putrootfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putrootfh-done", 2, 1, "PUTROOTFH4res *" }, + { "nfsv4", "op-read-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-read-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-read-start", 2, 1, "READ4args *" }, + { "nfsv4", "op-read-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-read-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-read-done", 2, 1, "READ4res *" }, + { "nfsv4", "op-readdir-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readdir-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readdir-start", 2, 1, "READDIR4args *" }, + { "nfsv4", "op-readdir-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readdir-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readdir-done", 2, 1, "READDIR4res *" }, + { "nfsv4", "op-readlink-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readlink-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readlink-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readlink-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readlink-done", 2, 1, "READLINK4res *" }, + { "nfsv4", "op-release-lockowner-start", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-release-lockowner-start", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-release-lockowner-start", 2, 1, + "RELEASE_LOCKOWNER4args *" }, + { "nfsv4", "op-release-lockowner-done", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-release-lockowner-done", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-release-lockowner-done", 2, 1, + "RELEASE_LOCKOWNER4res *" }, + { "nfsv4", "op-remove-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-remove-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-remove-start", 2, 1, "REMOVE4args *" }, + { "nfsv4", "op-remove-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-remove-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-remove-done", 2, 1, "REMOVE4res *" }, + { "nfsv4", "op-rename-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-rename-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-rename-start", 2, 1, "RENAME4args *" }, + { "nfsv4", "op-rename-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-rename-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-rename-done", 2, 1, "RENAME4res *" }, + { "nfsv4", "op-renew-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-renew-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-renew-start", 2, 1, "RENEW4args *" }, + { "nfsv4", "op-renew-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-renew-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-renew-done", 2, 1, "RENEW4res *" }, + { "nfsv4", "op-restorefh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-restorefh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-restorefh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-restorefh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-restorefh-done", 2, 1, "RESTOREFH4res *" }, + { "nfsv4", "op-savefh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-savefh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-savefh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-savefh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-savefh-done", 2, 1, "SAVEFH4res *" }, + { "nfsv4", "op-secinfo-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-secinfo-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-secinfo-start", 2, 1, "SECINFO4args *" }, + { "nfsv4", "op-secinfo-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-secinfo-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-secinfo-done", 2, 1, "SECINFO4res *" }, + { "nfsv4", "op-setattr-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setattr-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setattr-start", 2, 1, "SETATTR4args *" }, + { "nfsv4", "op-setattr-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setattr-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setattr-done", 2, 1, "SETATTR4res *" }, + { "nfsv4", "op-setclientid-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setclientid-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-start", 2, 1, "SETCLIENTID4args *" }, + { "nfsv4", "op-setclientid-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setclientid-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-done", 2, 1, "SETCLIENTID4res *" }, + { "nfsv4", "op-setclientid-confirm-start", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-setclientid-confirm-start", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-confirm-start", 2, 1, + "SETCLIENTID_CONFIRM4args *" }, + { "nfsv4", "op-setclientid-confirm-done", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-setclientid-confirm-done", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-confirm-done", 2, 1, + "SETCLIENTID_CONFIRM4res *" }, + { "nfsv4", "op-verify-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-verify-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-verify-start", 2, 1, "VERIFY4args *" }, + { "nfsv4", "op-verify-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-verify-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-verify-done", 2, 1, "VERIFY4res *" }, + { "nfsv4", "op-write-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-write-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-write-start", 2, 1, "WRITE4args *" }, + { "nfsv4", "op-write-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-write-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-write-done", 2, 1, "WRITE4res *" }, + { "nfsv4", "cb-recall-start", 0, 0, "rfs4_client_t *", + "conninfo_t *" }, + { "nfsv4", "cb-recall-start", 1, 1, "rfs4_deleg_state_t *", + "nfsv4cbinfo_t *" }, + { "nfsv4", "cb-recall-start", 2, 2, "CB_RECALL4args *" }, + { "nfsv4", "cb-recall-done", 0, 0, "rfs4_client_t *", + "conninfo_t *" }, + { "nfsv4", "cb-recall-done", 1, 1, "rfs4_deleg_state_t *", + "nfsv4cbinfo_t *" }, + { "nfsv4", "cb-recall-done", 2, 2, "CB_RECALL4res *" }, + + { "sysevent", "post", 0, 0, "evch_bind_t *", "syseventchaninfo_t *" }, + { "sysevent", "post", 1, 1, "sysevent_impl_t *", "syseventinfo_t *" }, + + { "xpv", "add-to-physmap-end", 0, 0, "int" }, + { "xpv", "add-to-physmap-start", 0, 0, "domid_t" }, + { "xpv", "add-to-physmap-start", 1, 1, "uint_t" }, + { "xpv", "add-to-physmap-start", 2, 2, "ulong_t" }, + { "xpv", "add-to-physmap-start", 3, 3, "ulong_t" }, + { "xpv", "decrease-reservation-end", 0, 0, "int" }, + { "xpv", "decrease-reservation-start", 0, 0, "domid_t" }, + { "xpv", "decrease-reservation-start", 1, 1, "ulong_t" }, + { "xpv", "decrease-reservation-start", 2, 2, "uint_t" }, + { "xpv", "decrease-reservation-start", 3, 3, "ulong_t *" }, + { "xpv", "dom-create-start", 0, 0, "xen_domctl_t *" }, + { "xpv", "dom-destroy-start", 0, 0, "domid_t" }, + { "xpv", "dom-pause-start", 0, 0, "domid_t" }, + { "xpv", "dom-unpause-start", 0, 0, "domid_t" }, + { "xpv", "dom-create-end", 0, 0, "int" }, + { "xpv", "dom-destroy-end", 0, 0, "int" }, + { "xpv", "dom-pause-end", 0, 0, "int" }, + { "xpv", "dom-unpause-end", 0, 0, "int" }, + { "xpv", "evtchn-op-end", 0, 0, "int" }, + { "xpv", "evtchn-op-start", 0, 0, "int" }, + { "xpv", "evtchn-op-start", 1, 1, "void *" }, + { "xpv", "increase-reservation-end", 0, 0, "int" }, + { "xpv", "increase-reservation-start", 0, 0, "domid_t" }, + { "xpv", "increase-reservation-start", 1, 1, "ulong_t" }, + { "xpv", "increase-reservation-start", 2, 2, "uint_t" }, + { "xpv", "increase-reservation-start", 3, 3, "ulong_t *" }, + { "xpv", "mmap-end", 0, 0, "int" }, + { "xpv", "mmap-entry", 0, 0, "ulong_t" }, + { "xpv", "mmap-entry", 1, 1, "ulong_t" }, + { "xpv", "mmap-entry", 2, 2, "ulong_t" }, + { "xpv", "mmap-start", 0, 0, "domid_t" }, + { "xpv", "mmap-start", 1, 1, "int" }, + { "xpv", "mmap-start", 2, 2, "privcmd_mmap_entry_t *" }, + { "xpv", "mmapbatch-end", 0, 0, "int" }, + { "xpv", "mmapbatch-end", 1, 1, "struct seg *" }, + { "xpv", "mmapbatch-end", 2, 2, "caddr_t" }, + { "xpv", "mmapbatch-start", 0, 0, "domid_t" }, + { "xpv", "mmapbatch-start", 1, 1, "int" }, + { "xpv", "mmapbatch-start", 2, 2, "caddr_t" }, + { "xpv", "mmu-ext-op-end", 0, 0, "int" }, + { "xpv", "mmu-ext-op-start", 0, 0, "int" }, + { "xpv", "mmu-ext-op-start", 1, 1, "struct mmuext_op *" }, + { "xpv", "mmu-update-start", 0, 0, "int" }, + { "xpv", "mmu-update-start", 1, 1, "int" }, + { "xpv", "mmu-update-start", 2, 2, "mmu_update_t *" }, + { "xpv", "mmu-update-end", 0, 0, "int" }, + { "xpv", "populate-physmap-end", 0, 0, "int" }, + { "xpv", "populate-physmap-start", 0, 0, "domid_t" }, + { "xpv", "populate-physmap-start", 1, 1, "ulong_t" }, + { "xpv", "populate-physmap-start", 2, 2, "ulong_t *" }, + { "xpv", "set-memory-map-end", 0, 0, "int" }, + { "xpv", "set-memory-map-start", 0, 0, "domid_t" }, + { "xpv", "set-memory-map-start", 1, 1, "int" }, + { "xpv", "set-memory-map-start", 2, 2, "struct xen_memory_map *" }, + { "xpv", "setvcpucontext-end", 0, 0, "int" }, + { "xpv", "setvcpucontext-start", 0, 0, "domid_t" }, + { "xpv", "setvcpucontext-start", 1, 1, "vcpu_guest_context_t *" }, + { NULL } +}; + +/*ARGSUSED*/ +void +sdt_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) +{ + sdt_probe_t *sdp = parg; + int i; + + desc->dtargd_native[0] = '\0'; + desc->dtargd_xlate[0] = '\0'; + + for (i = 0; sdt_args[i].sda_provider != NULL; i++) { + sdt_argdesc_t *a = &sdt_args[i]; + + if (strcmp(sdp->sdp_provider->sdtp_name, a->sda_provider) != 0) + continue; + + if (a->sda_name != NULL && + strcmp(sdp->sdp_name, a->sda_name) != 0) + continue; + + if (desc->dtargd_ndx != a->sda_ndx) + continue; + + if (a->sda_native != NULL) + (void) strcpy(desc->dtargd_native, a->sda_native); + + if (a->sda_xlate != NULL) + (void) strcpy(desc->dtargd_xlate, a->sda_xlate); + + desc->dtargd_mapping = a->sda_mapping; + return; + } + + desc->dtargd_ndx = DTRACE_ARGNONE; +} diff --git a/sys/cddl/contrib/opensolaris/uts/common/dtrace/systrace.c b/sys/cddl/contrib/opensolaris/uts/common/dtrace/systrace.c new file mode 100644 index 000000000000..be14660b04c0 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/dtrace/systrace.c @@ -0,0 +1,373 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/dtrace.h> +#include <sys/systrace.h> +#include <sys/stat.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/ddi.h> +#include <sys/sunddi.h> +#include <sys/atomic.h> + +#define SYSTRACE_ARTIFICIAL_FRAMES 1 + +#define SYSTRACE_SHIFT 16 +#define SYSTRACE_ISENTRY(x) ((int)(x) >> SYSTRACE_SHIFT) +#define SYSTRACE_SYSNUM(x) ((int)(x) & ((1 << SYSTRACE_SHIFT) - 1)) +#define SYSTRACE_ENTRY(id) ((1 << SYSTRACE_SHIFT) | (id)) +#define SYSTRACE_RETURN(id) (id) + +#if ((1 << SYSTRACE_SHIFT) <= NSYSCALL) +#error 1 << SYSTRACE_SHIFT must exceed number of system calls +#endif + +static dev_info_t *systrace_devi; +static dtrace_provider_id_t systrace_id; + +static void +systrace_init(struct sysent *actual, systrace_sysent_t **interposed) +{ + systrace_sysent_t *sysent = *interposed; + int i; + + if (sysent == NULL) { + *interposed = sysent = kmem_zalloc(sizeof (systrace_sysent_t) * + NSYSCALL, KM_SLEEP); + } + + for (i = 0; i < NSYSCALL; i++) { + struct sysent *a = &actual[i]; + systrace_sysent_t *s = &sysent[i]; + + if (LOADABLE_SYSCALL(a) && !LOADED_SYSCALL(a)) + continue; + + if (a->sy_callc == dtrace_systrace_syscall) + continue; + +#ifdef _SYSCALL32_IMPL + if (a->sy_callc == dtrace_systrace_syscall32) + continue; +#endif + + s->stsy_underlying = a->sy_callc; + } +} + +/*ARGSUSED*/ +static void +systrace_provide(void *arg, const dtrace_probedesc_t *desc) +{ + int i; + + if (desc != NULL) + return; + + systrace_init(sysent, &systrace_sysent); +#ifdef _SYSCALL32_IMPL + systrace_init(sysent32, &systrace_sysent32); +#endif + + for (i = 0; i < NSYSCALL; i++) { + if (systrace_sysent[i].stsy_underlying == NULL) + continue; + + if (dtrace_probe_lookup(systrace_id, NULL, + syscallnames[i], "entry") != 0) + continue; + + (void) dtrace_probe_create(systrace_id, NULL, syscallnames[i], + "entry", SYSTRACE_ARTIFICIAL_FRAMES, + (void *)((uintptr_t)SYSTRACE_ENTRY(i))); + (void) dtrace_probe_create(systrace_id, NULL, syscallnames[i], + "return", SYSTRACE_ARTIFICIAL_FRAMES, + (void *)((uintptr_t)SYSTRACE_RETURN(i))); + + systrace_sysent[i].stsy_entry = DTRACE_IDNONE; + systrace_sysent[i].stsy_return = DTRACE_IDNONE; +#ifdef _SYSCALL32_IMPL + systrace_sysent32[i].stsy_entry = DTRACE_IDNONE; + systrace_sysent32[i].stsy_return = DTRACE_IDNONE; +#endif + } +} + +/*ARGSUSED*/ +static void +systrace_destroy(void *arg, dtrace_id_t id, void *parg) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + + /* + * There's nothing to do here but assert that we have actually been + * disabled. + */ + if (SYSTRACE_ISENTRY((uintptr_t)parg)) { + ASSERT(systrace_sysent[sysnum].stsy_entry == DTRACE_IDNONE); +#ifdef _SYSCALL32_IMPL + ASSERT(systrace_sysent32[sysnum].stsy_entry == DTRACE_IDNONE); +#endif + } else { + ASSERT(systrace_sysent[sysnum].stsy_return == DTRACE_IDNONE); +#ifdef _SYSCALL32_IMPL + ASSERT(systrace_sysent32[sysnum].stsy_return == DTRACE_IDNONE); +#endif + } +} + +/*ARGSUSED*/ +static void +systrace_enable(void *arg, dtrace_id_t id, void *parg) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + int enabled = (systrace_sysent[sysnum].stsy_entry != DTRACE_IDNONE || + systrace_sysent[sysnum].stsy_return != DTRACE_IDNONE); + + if (SYSTRACE_ISENTRY((uintptr_t)parg)) { + systrace_sysent[sysnum].stsy_entry = id; +#ifdef _SYSCALL32_IMPL + systrace_sysent32[sysnum].stsy_entry = id; +#endif + } else { + systrace_sysent[sysnum].stsy_return = id; +#ifdef _SYSCALL32_IMPL + systrace_sysent32[sysnum].stsy_return = id; +#endif + } + + if (enabled) { + ASSERT(sysent[sysnum].sy_callc == dtrace_systrace_syscall); + return; + } + + (void) casptr(&sysent[sysnum].sy_callc, + (void *)systrace_sysent[sysnum].stsy_underlying, + (void *)dtrace_systrace_syscall); +#ifdef _SYSCALL32_IMPL + (void) casptr(&sysent32[sysnum].sy_callc, + (void *)systrace_sysent32[sysnum].stsy_underlying, + (void *)dtrace_systrace_syscall32); +#endif +} + +/*ARGSUSED*/ +static void +systrace_disable(void *arg, dtrace_id_t id, void *parg) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + int disable = (systrace_sysent[sysnum].stsy_entry == DTRACE_IDNONE || + systrace_sysent[sysnum].stsy_return == DTRACE_IDNONE); + + if (disable) { + (void) casptr(&sysent[sysnum].sy_callc, + (void *)dtrace_systrace_syscall, + (void *)systrace_sysent[sysnum].stsy_underlying); + +#ifdef _SYSCALL32_IMPL + (void) casptr(&sysent32[sysnum].sy_callc, + (void *)dtrace_systrace_syscall32, + (void *)systrace_sysent32[sysnum].stsy_underlying); +#endif + } + + if (SYSTRACE_ISENTRY((uintptr_t)parg)) { + systrace_sysent[sysnum].stsy_entry = DTRACE_IDNONE; +#ifdef _SYSCALL32_IMPL + systrace_sysent32[sysnum].stsy_entry = DTRACE_IDNONE; +#endif + } else { + systrace_sysent[sysnum].stsy_return = DTRACE_IDNONE; +#ifdef _SYSCALL32_IMPL + systrace_sysent32[sysnum].stsy_return = DTRACE_IDNONE; +#endif + } +} + +static dtrace_pattr_t systrace_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pops_t systrace_pops = { + systrace_provide, + NULL, + systrace_enable, + systrace_disable, + NULL, + NULL, + NULL, + NULL, + NULL, + systrace_destroy +}; + +static int +systrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + switch (cmd) { + case DDI_ATTACH: + break; + case DDI_RESUME: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + systrace_probe = (void (*)())dtrace_probe; + membar_enter(); + + if (ddi_create_minor_node(devi, "systrace", S_IFCHR, 0, + DDI_PSEUDO, NULL) == DDI_FAILURE || + dtrace_register("syscall", &systrace_attr, DTRACE_PRIV_USER, NULL, + &systrace_pops, NULL, &systrace_id) != 0) { + systrace_probe = systrace_stub; + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + ddi_report_dev(devi); + systrace_devi = devi; + + return (DDI_SUCCESS); +} + +static int +systrace_detach(dev_info_t *devi, ddi_detach_cmd_t cmd) +{ + switch (cmd) { + case DDI_DETACH: + break; + case DDI_SUSPEND: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (dtrace_unregister(systrace_id) != 0) + return (DDI_FAILURE); + + ddi_remove_minor_node(devi, NULL); + systrace_probe = systrace_stub; + return (DDI_SUCCESS); +} + +/*ARGSUSED*/ +static int +systrace_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)systrace_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +/*ARGSUSED*/ +static int +systrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +static struct cb_ops systrace_cb_ops = { + systrace_open, /* open */ + nodev, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + nodev, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_NEW | D_MP /* Driver compatibility flag */ +}; + +static struct dev_ops systrace_ops = { + DEVO_REV, /* devo_rev, */ + 0, /* refcnt */ + systrace_info, /* get_dev_info */ + nulldev, /* identify */ + nulldev, /* probe */ + systrace_attach, /* attach */ + systrace_detach, /* detach */ + nodev, /* reset */ + &systrace_cb_ops, /* driver operations */ + NULL, /* bus operations */ + nodev /* dev power */ +}; + +/* + * Module linkage information for the kernel. + */ +static struct modldrv modldrv = { + &mod_driverops, /* module type (this is a pseudo driver) */ + "System Call Tracing", /* name of module */ + &systrace_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, + (void *)&modldrv, + NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c index 55cb12407188..420f802f360d 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c +++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c @@ -1148,7 +1148,7 @@ arc_evict(arc_state_t *state, int64_t bytes, boolean_t recycle, /* prefetch buffers have a minimum lifespan */ if (HDR_IO_IN_PROGRESS(ab) || (ab->b_flags & (ARC_PREFETCH|ARC_INDIRECT) && - lbolt - ab->b_arc_access < arc_min_prefetch_lifespan)) { + LBOLT - ab->b_arc_access < arc_min_prefetch_lifespan)) { skipped++; continue; } @@ -1439,7 +1439,7 @@ arc_reclaim_needed(void) return (1); #endif #else - if (kmem_used() > (kmem_size() * 4) / 5) + if (kmem_used() > (kmem_size() * 3) / 4) return (1); #endif @@ -1525,7 +1525,7 @@ arc_reclaim_thread(void *dummy __unused) } /* reset the growth delay for every reclaim */ - growtime = lbolt + (arc_grow_retry * hz); + growtime = LBOLT + (arc_grow_retry * hz); ASSERT(growtime > 0); if (zfs_needfree && last_reclaim == ARC_RECLAIM_CONS) { @@ -1538,7 +1538,7 @@ arc_reclaim_thread(void *dummy __unused) last_reclaim = ARC_RECLAIM_AGGR; } arc_kmem_reap_now(last_reclaim); - } else if ((growtime > 0) && ((growtime - lbolt) <= 0)) { + } else if ((growtime > 0) && ((growtime - LBOLT) <= 0)) { arc_no_grow = FALSE; } @@ -1757,7 +1757,7 @@ arc_access(arc_buf_hdr_t *buf, kmutex_t *hash_lock) */ ASSERT(buf->b_arc_access == 0); - buf->b_arc_access = lbolt; + buf->b_arc_access = LBOLT; DTRACE_PROBE1(new_state__mru, arc_buf_hdr_t *, buf); arc_change_state(arc_mru, buf, hash_lock); @@ -1781,7 +1781,7 @@ arc_access(arc_buf_hdr_t *buf, kmutex_t *hash_lock) buf->b_flags &= ~ARC_PREFETCH; ARCSTAT_BUMP(arcstat_mru_hits); } - buf->b_arc_access = lbolt; + buf->b_arc_access = LBOLT; return; } @@ -1790,13 +1790,13 @@ arc_access(arc_buf_hdr_t *buf, kmutex_t *hash_lock) * but it is still in the cache. Move it to the MFU * state. */ - if (lbolt > buf->b_arc_access + ARC_MINTIME) { + if (LBOLT > buf->b_arc_access + ARC_MINTIME) { /* * More than 125ms have passed since we * instantiated this buffer. Move it to the * most frequently used state. */ - buf->b_arc_access = lbolt; + buf->b_arc_access = LBOLT; DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, buf); arc_change_state(arc_mfu, buf, hash_lock); } @@ -1819,7 +1819,7 @@ arc_access(arc_buf_hdr_t *buf, kmutex_t *hash_lock) DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, buf); } - buf->b_arc_access = lbolt; + buf->b_arc_access = LBOLT; arc_change_state(new_state, buf, hash_lock); ARCSTAT_BUMP(arcstat_mru_ghost_hits); @@ -1842,7 +1842,7 @@ arc_access(arc_buf_hdr_t *buf, kmutex_t *hash_lock) mutex_exit(&arc_mfu->arcs_mtx); } ARCSTAT_BUMP(arcstat_mfu_hits); - buf->b_arc_access = lbolt; + buf->b_arc_access = LBOLT; } else if (buf->b_state == arc_mfu_ghost) { arc_state_t *new_state = arc_mfu; /* @@ -1860,7 +1860,7 @@ arc_access(arc_buf_hdr_t *buf, kmutex_t *hash_lock) new_state = arc_mru; } - buf->b_arc_access = lbolt; + buf->b_arc_access = LBOLT; DTRACE_PROBE1(new_state__mfu, arc_buf_hdr_t *, buf); arc_change_state(new_state, buf, hash_lock); @@ -2729,7 +2729,7 @@ arc_init(void) arc_c_max = (arc_c * 8) - (1<<30); else arc_c_max = arc_c_min; - arc_c_max = MAX(arc_c * 6, arc_c_max); + arc_c_max = MAX(arc_c * 5, arc_c_max); #ifdef _KERNEL /* * Allow the tunables to override our calculations if they are @@ -2800,16 +2800,17 @@ arc_init(void) arc_dead = FALSE; #ifdef _KERNEL - /* Warn about ZFS memory requirements. */ + /* Warn about ZFS memory and address space requirements. */ if (((uint64_t)physmem * PAGESIZE) < (256 + 128 + 64) * (1 << 20)) { printf("ZFS WARNING: Recommended minimum RAM size is 512MB; " "expect unstable behavior.\n"); - } else if (kmem_size() < 256 * (1 << 20)) { - printf("ZFS WARNING: Recommended minimum kmem_size is 256MB; " + } + if (kmem_size() < 512 * (1 << 20)) { + printf("ZFS WARNING: Recommended minimum kmem_size is 512MB; " "expect unstable behavior.\n"); - printf(" Consider tuning vm.kmem_size or " - "vm.kmem_size_min\n"); - printf(" in /boot/loader.conf.\n"); + printf(" Consider tuning vm.kmem_size and " + "vm.kmem_size_max\n"); + printf(" in /boot/loader.conf.\n"); } #endif } diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_zfetch.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_zfetch.c index 78d625cd41bd..b25cc898c37d 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_zfetch.c +++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_zfetch.c @@ -196,7 +196,7 @@ dmu_zfetch_dofetch(zfetch_t *zf, zstream_t *zs) break; } zs->zst_ph_offset = prefetch_tail; - zs->zst_last = lbolt; + zs->zst_last = LBOLT; } /* @@ -517,7 +517,7 @@ dmu_zfetch_stream_reclaim(zfetch_t *zf) for (zs = list_head(&zf->zf_stream); zs; zs = list_next(&zf->zf_stream, zs)) { - if (((lbolt - zs->zst_last) / hz) > zfetch_min_sec_reap) + if (((LBOLT - zs->zst_last) / hz) > zfetch_min_sec_reap) break; } @@ -639,7 +639,7 @@ dmu_zfetch(zfetch_t *zf, uint64_t offset, uint64_t size, int prefetched) newstream->zst_ph_offset = zst.zst_len + zst.zst_offset; newstream->zst_cap = zst.zst_len; newstream->zst_direction = ZFETCH_FORWARD; - newstream->zst_last = lbolt; + newstream->zst_last = LBOLT; mutex_init(&newstream->zst_lock, NULL, MUTEX_DEFAULT, NULL); diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/refcount.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/refcount.c index 411ed46e13d7..a2f4614fed87 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/refcount.c +++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/refcount.c @@ -42,7 +42,7 @@ static kmem_cache_t *reference_cache; static kmem_cache_t *reference_history_cache; void -refcount_init(void) +refcount_sysinit(void) { reference_cache = kmem_cache_create("reference_cache", sizeof (reference_t), 0, NULL, NULL, NULL, NULL, NULL, 0); diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c index 5da1f96c85c5..1e1f0ee93068 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c +++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c @@ -1103,7 +1103,7 @@ spa_init(int mode) spa_mode = mode; - refcount_init(); + refcount_sysinit(); unique_init(); zio_init(); dmu_init(); diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/refcount.h b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/refcount.h index 4de1cae12fe7..c64c6627f783 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/refcount.h +++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/refcount.h @@ -28,6 +28,7 @@ #pragma ident "%Z%%M% %I% %E% SMI" +#include_next <sys/refcount.h> #include <sys/list.h> #include <sys/zfs_context.h> @@ -70,7 +71,7 @@ int64_t refcount_remove(refcount_t *rc, void *holder_tag); int64_t refcount_add_many(refcount_t *rc, uint64_t number, void *holder_tag); int64_t refcount_remove_many(refcount_t *rc, uint64_t number, void *holder_tag); -void refcount_init(void); +void refcount_sysinit(void); void refcount_fini(void); #else /* DEBUG */ @@ -91,7 +92,7 @@ typedef struct refcount { #define refcount_remove_many(rc, number, holder) \ atomic_add_64_nv(&(rc)->rc_count, -number) -#define refcount_init() +#define refcount_sysinit() #define refcount_fini() #endif /* DEBUG */ diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_ioctl.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_ioctl.c index c9424be1995f..8699922ccf09 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_ioctl.c +++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_ioctl.c @@ -1824,3 +1824,4 @@ static moduledata_t zfs_mod = { 0 }; DECLARE_MODULE(zfsctrl, zfs_mod, SI_SUB_VFS, SI_ORDER_ANY); +MODULE_DEPEND(zfsctrl, opensolaris, 1, 1, 1); diff --git a/sys/cddl/contrib/opensolaris/uts/common/rpc/xdr.c b/sys/cddl/contrib/opensolaris/uts/common/rpc/xdr.c index 65b73b6181a9..e9346681774c 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/rpc/xdr.c +++ b/sys/cddl/contrib/opensolaris/uts/common/rpc/xdr.c @@ -54,11 +54,13 @@ #pragma weak xdr_int64_t = xdr_longlong_t #pragma weak xdr_uint64_t = xdr_u_longlong_t +#if defined(sun) #if !defined(_BIG_ENDIAN) && !defined(_LITTLE_ENDIAN) #error "Exactly one of _BIG_ENDIAN or _LITTLE_ENDIAN must be defined" #elif defined(_BIG_ENDIAN) && defined(_LITTLE_ENDIAN) #error "Only one of _BIG_ENDIAN or _LITTLE_ENDIAN may be defined" #endif +#endif /* * constants specific to the xdr "protocol" @@ -174,12 +176,12 @@ bool_t xdr_longlong_t(XDR *xdrs, longlong_t *hp) { if (xdrs->x_op == XDR_ENCODE) { -#if defined(_LITTLE_ENDIAN) +#if BYTE_ORDER == _LITTLE_ENDIAN if (XDR_PUTINT32(xdrs, (int32_t *)((char *)hp + BYTES_PER_XDR_UNIT)) == TRUE) { return (XDR_PUTINT32(xdrs, (int32_t *)hp)); } -#elif defined(_BIG_ENDIAN) +#else if (XDR_PUTINT32(xdrs, (int32_t *)hp) == TRUE) { return (XDR_PUTINT32(xdrs, (int32_t *)((char *)hp + BYTES_PER_XDR_UNIT))); @@ -189,12 +191,12 @@ xdr_longlong_t(XDR *xdrs, longlong_t *hp) } if (xdrs->x_op == XDR_DECODE) { -#if defined(_LITTLE_ENDIAN) +#if BYTE_ORDER == _LITTLE_ENDIAN if (XDR_GETINT32(xdrs, (int32_t *)((char *)hp + BYTES_PER_XDR_UNIT)) == TRUE) { return (XDR_GETINT32(xdrs, (int32_t *)hp)); } -#elif defined(_BIG_ENDIAN) +#else if (XDR_GETINT32(xdrs, (int32_t *)hp) == TRUE) { return (XDR_GETINT32(xdrs, (int32_t *)((char *)hp + BYTES_PER_XDR_UNIT))); @@ -213,12 +215,12 @@ xdr_u_longlong_t(XDR *xdrs, u_longlong_t *hp) { if (xdrs->x_op == XDR_ENCODE) { -#if defined(_LITTLE_ENDIAN) +#if BYTE_ORDER == _LITTLE_ENDIAN if (XDR_PUTINT32(xdrs, (int32_t *)((char *)hp + BYTES_PER_XDR_UNIT)) == TRUE) { return (XDR_PUTINT32(xdrs, (int32_t *)hp)); } -#elif defined(_BIG_ENDIAN) +#else if (XDR_PUTINT32(xdrs, (int32_t *)hp) == TRUE) { return (XDR_PUTINT32(xdrs, (int32_t *)((char *)hp + BYTES_PER_XDR_UNIT))); @@ -228,12 +230,12 @@ xdr_u_longlong_t(XDR *xdrs, u_longlong_t *hp) } if (xdrs->x_op == XDR_DECODE) { -#if defined(_LITTLE_ENDIAN) +#if BYTE_ORDER == _LITTLE_ENDIAN if (XDR_GETINT32(xdrs, (int32_t *)((char *)hp + BYTES_PER_XDR_UNIT)) == TRUE) { return (XDR_GETINT32(xdrs, (int32_t *)hp)); } -#elif defined(_BIG_ENDIAN) +#else if (XDR_GETINT32(xdrs, (int32_t *)hp) == TRUE) { return (XDR_GETINT32(xdrs, (int32_t *)((char *)hp + BYTES_PER_XDR_UNIT))); diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/cmn_err.h b/sys/cddl/contrib/opensolaris/uts/common/sys/cmn_err.h new file mode 100644 index 000000000000..e710d8e5c30b --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/cmn_err.h @@ -0,0 +1,128 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */ +/* All Rights Reserved */ + + +/* + * Copyright 2004 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _SYS_CMN_ERR_H +#define _SYS_CMN_ERR_H + +#pragma ident "%Z%%M% %I% %E% SMI" + +#if defined(_KERNEL) && !defined(_ASM) +#include <sys/va_list.h> +#endif + +#ifdef __cplusplus +extern "C" { +#endif + +/* Common error handling severity levels */ + +#define CE_CONT 0 /* continuation */ +#define CE_NOTE 1 /* notice */ +#define CE_WARN 2 /* warning */ +#define CE_PANIC 3 /* panic */ +#define CE_IGNORE 4 /* print nothing */ + +#ifndef _ASM + +#ifdef _KERNEL + +/*PRINTFLIKE2*/ +extern void cmn_err(int, const char *, ...) + __KPRINTFLIKE(2); +#pragma rarely_called(cmn_err) + +extern void vzcmn_err(zoneid_t, int, const char *, __va_list) + __KVPRINTFLIKE(3); +#pragma rarely_called(vzcmn_err) + +extern void vcmn_err(int, const char *, __va_list) + __KVPRINTFLIKE(2); +#pragma rarely_called(vcmn_err) + +/*PRINTFLIKE3*/ +extern void zcmn_err(zoneid_t, int, const char *, ...) + __KPRINTFLIKE(3); +#pragma rarely_called(zcmn_err) + +/*PRINTFLIKE1*/ +extern void printf(const char *, ...) + __KPRINTFLIKE(1); +#pragma rarely_called(printf) + +extern void vzprintf(zoneid_t, const char *, __va_list) + __KVPRINTFLIKE(2); +#pragma rarely_called(vzprintf) + +/*PRINTFLIKE2*/ +extern void zprintf(zoneid_t, const char *, ...) + __KPRINTFLIKE(2); +#pragma rarely_called(zprintf) + +extern void vprintf(const char *, __va_list) + __KVPRINTFLIKE(1); +#pragma rarely_called(vprintf) + +/*PRINTFLIKE1*/ +extern void uprintf(const char *, ...) + __KPRINTFLIKE(1); +#pragma rarely_called(uprintf) + +extern void vuprintf(const char *, __va_list) + __KVPRINTFLIKE(1); +#pragma rarely_called(vuprintf) + +/*PRINTFLIKE3*/ +extern size_t snprintf(char *, size_t, const char *, ...) + __KPRINTFLIKE(3); +extern size_t vsnprintf(char *, size_t, const char *, __va_list) + __KVPRINTFLIKE(3); +/*PRINTFLIKE2*/ +extern char *sprintf(char *, const char *, ...) + __KPRINTFLIKE(2); +extern char *vsprintf(char *, const char *, __va_list) + __KVPRINTFLIKE(2); + +/*PRINTFLIKE1*/ +extern void panic(const char *, ...) + __KPRINTFLIKE(1) __NORETURN; +#pragma rarely_called(panic) + +extern void vpanic(const char *, __va_list) + __KVPRINTFLIKE(1) __NORETURN; +#pragma rarely_called(vpanic) + +#endif /* _KERNEL */ +#endif /* !_ASM */ + +#ifdef __cplusplus +} +#endif + +#endif /* _SYS_CMN_ERR_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/cpupart.h b/sys/cddl/contrib/opensolaris/uts/common/sys/cpupart.h new file mode 100644 index 000000000000..b9e0da4e1993 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/cpupart.h @@ -0,0 +1,162 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _SYS_CPUPART_H +#define _SYS_CPUPART_H + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/types.h> +#include <sys/processor.h> +#include <sys/cpuvar.h> +#include <sys/disp.h> +#include <sys/pset.h> +#include <sys/lgrp.h> +#include <sys/lgrp_user.h> +#include <sys/pg.h> +#include <sys/bitset.h> +#include <sys/time.h> + +#ifdef __cplusplus +extern "C" { +#endif + +#ifdef _KERNEL + +typedef int cpupartid_t; + +/* + * Special partition id. + */ +#define CP_DEFAULT 0 + +/* + * Flags for cpupart_list() + */ +#define CP_ALL 0 /* return all cpu partitions */ +#define CP_NONEMPTY 1 /* return only non-empty ones */ + +#if defined(_MACHDEP) +struct mach_cpupart { + cpuset_t mc_haltset; +}; + +extern struct mach_cpupart cp_default_mach; +#else +struct mach_cpupart; +#endif + +typedef struct cpupart { + disp_t cp_kp_queue; /* partition-wide kpreempt queue */ + cpupartid_t cp_id; /* partition ID */ + int cp_ncpus; /* number of online processors */ + struct cpupart *cp_next; /* next partition in list */ + struct cpupart *cp_prev; /* previous partition in list */ + struct cpu *cp_cpulist; /* processor list */ + struct kstat *cp_kstat; /* per-partition statistics */ + + /* + * cp_nrunnable and cp_nrunning are used to calculate load average. + */ + uint_t cp_nrunnable; /* current # of runnable threads */ + uint_t cp_nrunning; /* current # of running threads */ + + /* + * cp_updates, cp_nrunnable_cum, cp_nwaiting_cum, and cp_hp_avenrun + * are used to generate kstat information on an as-needed basis. + */ + uint64_t cp_updates; /* number of statistics updates */ + uint64_t cp_nrunnable_cum; /* cum. # of runnable threads */ + uint64_t cp_nwaiting_cum; /* cum. # of waiting threads */ + + struct loadavg_s cp_loadavg; /* cpupart loadavg */ + + klgrpset_t cp_lgrpset; /* set of lgroups on which this */ + /* partition has cpus */ + lpl_t *cp_lgrploads; /* table of load averages for this */ + /* partition, indexed by lgrp ID */ + int cp_nlgrploads; /* size of cp_lgrploads table */ + uint64_t cp_hp_avenrun[3]; /* high-precision load average */ + uint_t cp_attr; /* bitmask of attributes */ + lgrp_gen_t cp_gen; /* generation number */ + lgrp_id_t cp_lgrp_hint; /* last home lgroup chosen */ + bitset_t cp_cmt_pgs; /* CMT PGs represented */ + + struct mach_cpupart *cp_mach; /* mach-specific */ +} cpupart_t; + +typedef struct cpupart_kstat { + kstat_named_t cpk_updates; /* number of updates */ + kstat_named_t cpk_runnable; /* cum # of runnable threads */ + kstat_named_t cpk_waiting; /* cum # waiting for I/O */ + kstat_named_t cpk_ncpus; /* current # of CPUs */ + kstat_named_t cpk_avenrun_1min; /* 1-minute load average */ + kstat_named_t cpk_avenrun_5min; /* 5-minute load average */ + kstat_named_t cpk_avenrun_15min; /* 15-minute load average */ +} cpupart_kstat_t; + +/* + * Macro to obtain the maximum run priority for the global queue associated + * with given cpu partition. + */ +#define CP_MAXRUNPRI(cp) ((cp)->cp_kp_queue.disp_maxrunpri) + +/* + * This macro is used to determine if the given thread must surrender + * CPU to higher priority runnable threads on one of its dispatch queues. + * This should really be defined in <sys/disp.h> but it is not because + * including <sys/cpupart.h> there would cause recursive includes. + */ +#define DISP_MUST_SURRENDER(t) \ + ((DISP_MAXRUNPRI(t) > DISP_PRIO(t)) || \ + (CP_MAXRUNPRI(t->t_cpupart) > DISP_PRIO(t))) + +extern cpupart_t cp_default; +extern cpupart_t *cp_list_head; +extern uint_t cp_numparts; +extern uint_t cp_numparts_nonempty; + +extern void cpupart_initialize_default(); +extern cpupart_t *cpupart_find(psetid_t); +extern int cpupart_create(psetid_t *); +extern int cpupart_destroy(psetid_t); +extern psetid_t cpupart_query_cpu(cpu_t *); +extern int cpupart_attach_cpu(psetid_t, cpu_t *, int); +extern int cpupart_get_cpus(psetid_t *, processorid_t *, uint_t *); +extern int cpupart_bind_thread(kthread_id_t, psetid_t, int, void *, + void *); +extern void cpupart_kpqalloc(pri_t); +extern int cpupart_get_loadavg(psetid_t, int *, int); +extern uint_t cpupart_list(psetid_t *, uint_t, int); +extern int cpupart_setattr(psetid_t, uint_t); +extern int cpupart_getattr(psetid_t, uint_t *); + +#endif /* _KERNEL */ + +#ifdef __cplusplus +} +#endif + +#endif /* _SYS_CPUPART_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/cpuvar.h b/sys/cddl/contrib/opensolaris/uts/common/sys/cpuvar.h new file mode 100644 index 000000000000..cd0d0278662c --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/cpuvar.h @@ -0,0 +1,737 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _SYS_CPUVAR_H +#define _SYS_CPUVAR_H + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/thread.h> +#include <sys/sysinfo.h> /* has cpu_stat_t definition */ +#include <sys/disp.h> +#include <sys/processor.h> + +#if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP) +#include <sys/machcpuvar.h> +#endif + +#include <sys/types.h> +#include <sys/file.h> +#include <sys/bitmap.h> +#include <sys/rwlock.h> +#include <sys/msacct.h> +#if defined(__GNUC__) && defined(_ASM_INLINES) && defined(_KERNEL) && \ + (defined(__i386) || defined(__amd64)) +#include <asm/cpuvar.h> +#endif + +#ifdef __cplusplus +extern "C" { +#endif + +struct squeue_set_s; + +#define CPU_CACHE_COHERENCE_SIZE 64 +#define S_LOADAVG_SZ 11 +#define S_MOVAVG_SZ 10 + +struct loadavg_s { + int lg_cur; /* current loadavg entry */ + unsigned int lg_len; /* number entries recorded */ + hrtime_t lg_total; /* used to temporarily hold load totals */ + hrtime_t lg_loads[S_LOADAVG_SZ]; /* table of recorded entries */ +}; + +/* + * For fast event tracing. + */ +struct ftrace_record; +typedef struct ftrace_data { + int ftd_state; /* ftrace flags */ + kmutex_t ftd_unused; /* ftrace buffer lock, unused */ + struct ftrace_record *ftd_cur; /* current record */ + struct ftrace_record *ftd_first; /* first record */ + struct ftrace_record *ftd_last; /* last record */ +} ftrace_data_t; + +struct cyc_cpu; +struct nvlist; + +/* + * Per-CPU data. + * + * Be careful adding new members: if they are not the same in all modules (e.g. + * change size depending on a #define), CTF uniquification can fail to work + * properly. Furthermore, this is transitive in that it applies recursively to + * all types pointed to by cpu_t. + */ +typedef struct cpu { + processorid_t cpu_id; /* CPU number */ + processorid_t cpu_seqid; /* sequential CPU id (0..ncpus-1) */ + volatile cpu_flag_t cpu_flags; /* flags indicating CPU state */ + struct cpu *cpu_self; /* pointer to itself */ + kthread_t *cpu_thread; /* current thread */ + kthread_t *cpu_idle_thread; /* idle thread for this CPU */ + kthread_t *cpu_pause_thread; /* pause thread for this CPU */ + klwp_id_t cpu_lwp; /* current lwp (if any) */ + klwp_id_t cpu_fpowner; /* currently loaded fpu owner */ + struct cpupart *cpu_part; /* partition with this CPU */ + struct lgrp_ld *cpu_lpl; /* pointer to this cpu's load */ + int cpu_cache_offset; /* see kmem.c for details */ + + /* + * Links to other CPUs. It is safe to walk these lists if + * one of the following is true: + * - cpu_lock held + * - preemption disabled via kpreempt_disable + * - PIL >= DISP_LEVEL + * - acting thread is an interrupt thread + * - all other CPUs are paused + */ + struct cpu *cpu_next; /* next existing CPU */ + struct cpu *cpu_prev; /* prev existing CPU */ + struct cpu *cpu_next_onln; /* next online (enabled) CPU */ + struct cpu *cpu_prev_onln; /* prev online (enabled) CPU */ + struct cpu *cpu_next_part; /* next CPU in partition */ + struct cpu *cpu_prev_part; /* prev CPU in partition */ + struct cpu *cpu_next_lgrp; /* next CPU in latency group */ + struct cpu *cpu_prev_lgrp; /* prev CPU in latency group */ + struct cpu *cpu_next_lpl; /* next CPU in lgrp partition */ + struct cpu *cpu_prev_lpl; + + struct cpu_pg *cpu_pg; /* cpu's processor groups */ + + void *cpu_reserved[4]; /* reserved for future use */ + + /* + * Scheduling variables. + */ + disp_t *cpu_disp; /* dispatch queue data */ + /* + * Note that cpu_disp is set before the CPU is added to the system + * and is never modified. Hence, no additional locking is needed + * beyond what's necessary to access the cpu_t structure. + */ + char cpu_runrun; /* scheduling flag - set to preempt */ + char cpu_kprunrun; /* force kernel preemption */ + pri_t cpu_chosen_level; /* priority at which cpu */ + /* was chosen for scheduling */ + kthread_t *cpu_dispthread; /* thread selected for dispatch */ + disp_lock_t cpu_thread_lock; /* dispatcher lock on current thread */ + uint8_t cpu_disp_flags; /* flags used by dispatcher */ + /* + * The following field is updated when ever the cpu_dispthread + * changes. Also in places, where the current thread(cpu_dispthread) + * priority changes. This is used in disp_lowpri_cpu() + */ + pri_t cpu_dispatch_pri; /* priority of cpu_dispthread */ + clock_t cpu_last_swtch; /* last time switched to new thread */ + + /* + * Interrupt data. + */ + caddr_t cpu_intr_stack; /* interrupt stack */ + kthread_t *cpu_intr_thread; /* interrupt thread list */ + uint_t cpu_intr_actv; /* interrupt levels active (bitmask) */ + int cpu_base_spl; /* priority for highest rupt active */ + + /* + * Statistics. + */ + cpu_stats_t cpu_stats; /* per-CPU statistics */ + struct kstat *cpu_info_kstat; /* kstat for cpu info */ + + uintptr_t cpu_profile_pc; /* kernel PC in profile interrupt */ + uintptr_t cpu_profile_upc; /* user PC in profile interrupt */ + uintptr_t cpu_profile_pil; /* PIL when profile interrupted */ + + ftrace_data_t cpu_ftrace; /* per cpu ftrace data */ + + clock_t cpu_deadman_lbolt; /* used by deadman() */ + uint_t cpu_deadman_countdown; /* used by deadman() */ + + kmutex_t cpu_cpc_ctxlock; /* protects context for idle thread */ + kcpc_ctx_t *cpu_cpc_ctx; /* performance counter context */ + + /* + * Configuration information for the processor_info system call. + */ + processor_info_t cpu_type_info; /* config info */ + time_t cpu_state_begin; /* when CPU entered current state */ + char cpu_cpr_flags; /* CPR related info */ + struct cyc_cpu *cpu_cyclic; /* per cpu cyclic subsystem data */ + struct squeue_set_s *cpu_squeue_set; /* per cpu squeue set */ + struct nvlist *cpu_props; /* pool-related properties */ + + krwlock_t cpu_ft_lock; /* DTrace: fasttrap lock */ + uintptr_t cpu_dtrace_caller; /* DTrace: caller, if any */ + hrtime_t cpu_dtrace_chillmark; /* DTrace: chill mark time */ + hrtime_t cpu_dtrace_chilled; /* DTrace: total chill time */ + volatile uint16_t cpu_mstate; /* cpu microstate */ + volatile uint16_t cpu_mstate_gen; /* generation counter */ + volatile hrtime_t cpu_mstate_start; /* cpu microstate start time */ + volatile hrtime_t cpu_acct[NCMSTATES]; /* cpu microstate data */ + hrtime_t cpu_intracct[NCMSTATES]; /* interrupt mstate data */ + hrtime_t cpu_waitrq; /* cpu run-queue wait time */ + struct loadavg_s cpu_loadavg; /* loadavg info for this cpu */ + + char *cpu_idstr; /* for printing and debugging */ + char *cpu_brandstr; /* for printing */ + + /* + * Sum of all device interrupt weights that are currently directed at + * this cpu. Cleared at start of interrupt redistribution. + */ + int32_t cpu_intr_weight; + void *cpu_vm_data; + + struct cpu_physid *cpu_physid; /* physical associations */ + + uint64_t cpu_curr_clock; /* current clock freq in Hz */ + char *cpu_supp_freqs; /* supported freqs in Hz */ + + /* + * Interrupt load factor used by dispatcher & softcall + */ + hrtime_t cpu_intrlast; /* total interrupt time (nsec) */ + int cpu_intrload; /* interrupt load factor (0-99%) */ + + /* + * New members must be added /before/ this member, as the CTF tools + * rely on this being the last field before cpu_m, so they can + * correctly calculate the offset when synthetically adding the cpu_m + * member in objects that do not have it. This fixup is required for + * uniquification to work correctly. + */ + uintptr_t cpu_m_pad; + +#if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP) + struct machcpu cpu_m; /* per architecture info */ +#endif +} cpu_t; + +/* + * The cpu_core structure consists of per-CPU state available in any context. + * On some architectures, this may mean that the page(s) containing the + * NCPU-sized array of cpu_core structures must be locked in the TLB -- it + * is up to the platform to assure that this is performed properly. Note that + * the structure is sized to avoid false sharing. + */ +#define CPUC_SIZE (sizeof (uint16_t) + sizeof (uintptr_t) + \ + sizeof (kmutex_t)) +#define CPUC_PADSIZE CPU_CACHE_COHERENCE_SIZE - CPUC_SIZE + +typedef struct cpu_core { + uint16_t cpuc_dtrace_flags; /* DTrace flags */ + uint8_t cpuc_pad[CPUC_PADSIZE]; /* padding */ + uintptr_t cpuc_dtrace_illval; /* DTrace illegal value */ + kmutex_t cpuc_pid_lock; /* DTrace pid provider lock */ +} cpu_core_t; + +#ifdef _KERNEL +extern cpu_core_t cpu_core[]; +#endif /* _KERNEL */ + +/* + * CPU_ON_INTR() macro. Returns non-zero if currently on interrupt stack. + * Note that this isn't a test for a high PIL. For example, cpu_intr_actv + * does not get updated when we go through sys_trap from TL>0 at high PIL. + * getpil() should be used instead to check for PIL levels. + */ +#define CPU_ON_INTR(cpup) ((cpup)->cpu_intr_actv >> (LOCK_LEVEL + 1)) + +#if defined(_KERNEL) || defined(_KMEMUSER) + +#define INTR_STACK_SIZE MAX(DEFAULTSTKSZ, PAGESIZE) + +/* MEMBERS PROTECTED BY "atomicity": cpu_flags */ + +/* + * Flags in the CPU structure. + * + * These are protected by cpu_lock (except during creation). + * + * Offlined-CPUs have three stages of being offline: + * + * CPU_ENABLE indicates that the CPU is participating in I/O interrupts + * that can be directed at a number of different CPUs. If CPU_ENABLE + * is off, the CPU will not be given interrupts that can be sent elsewhere, + * but will still get interrupts from devices associated with that CPU only, + * and from other CPUs. + * + * CPU_OFFLINE indicates that the dispatcher should not allow any threads + * other than interrupt threads to run on that CPU. A CPU will not have + * CPU_OFFLINE set if there are any bound threads (besides interrupts). + * + * CPU_QUIESCED is set if p_offline was able to completely turn idle the + * CPU and it will not have to run interrupt threads. In this case it'll + * stay in the idle loop until CPU_QUIESCED is turned off. + * + * CPU_FROZEN is used only by CPR to mark CPUs that have been successfully + * suspended (in the suspend path), or have yet to be resumed (in the resume + * case). + * + * On some platforms CPUs can be individually powered off. + * The following flags are set for powered off CPUs: CPU_QUIESCED, + * CPU_OFFLINE, and CPU_POWEROFF. The following flags are cleared: + * CPU_RUNNING, CPU_READY, CPU_EXISTS, and CPU_ENABLE. + */ +#define CPU_RUNNING 0x001 /* CPU running */ +#define CPU_READY 0x002 /* CPU ready for cross-calls */ +#define CPU_QUIESCED 0x004 /* CPU will stay in idle */ +#define CPU_EXISTS 0x008 /* CPU is configured */ +#define CPU_ENABLE 0x010 /* CPU enabled for interrupts */ +#define CPU_OFFLINE 0x020 /* CPU offline via p_online */ +#define CPU_POWEROFF 0x040 /* CPU is powered off */ +#define CPU_FROZEN 0x080 /* CPU is frozen via CPR suspend */ +#define CPU_SPARE 0x100 /* CPU offline available for use */ +#define CPU_FAULTED 0x200 /* CPU offline diagnosed faulty */ + +#define FMT_CPU_FLAGS \ + "\20\12fault\11spare\10frozen" \ + "\7poweroff\6offline\5enable\4exist\3quiesced\2ready\1run" + +#define CPU_ACTIVE(cpu) (((cpu)->cpu_flags & CPU_OFFLINE) == 0) + +/* + * Flags for cpu_offline(), cpu_faulted(), and cpu_spare(). + */ +#define CPU_FORCED 0x0001 /* Force CPU offline */ + +/* + * DTrace flags. + */ +#define CPU_DTRACE_NOFAULT 0x0001 /* Don't fault */ +#define CPU_DTRACE_DROP 0x0002 /* Drop this ECB */ +#define CPU_DTRACE_BADADDR 0x0004 /* DTrace fault: bad address */ +#define CPU_DTRACE_BADALIGN 0x0008 /* DTrace fault: bad alignment */ +#define CPU_DTRACE_DIVZERO 0x0010 /* DTrace fault: divide by zero */ +#define CPU_DTRACE_ILLOP 0x0020 /* DTrace fault: illegal operation */ +#define CPU_DTRACE_NOSCRATCH 0x0040 /* DTrace fault: out of scratch */ +#define CPU_DTRACE_KPRIV 0x0080 /* DTrace fault: bad kernel access */ +#define CPU_DTRACE_UPRIV 0x0100 /* DTrace fault: bad user access */ +#define CPU_DTRACE_TUPOFLOW 0x0200 /* DTrace fault: tuple stack overflow */ +#if defined(__sparc) +#define CPU_DTRACE_FAKERESTORE 0x0400 /* pid provider hint to getreg */ +#endif +#define CPU_DTRACE_ENTRY 0x0800 /* pid provider hint to ustack() */ +#define CPU_DTRACE_BADSTACK 0x1000 /* DTrace fault: bad stack */ + +#define CPU_DTRACE_FAULT (CPU_DTRACE_BADADDR | CPU_DTRACE_BADALIGN | \ + CPU_DTRACE_DIVZERO | CPU_DTRACE_ILLOP | \ + CPU_DTRACE_NOSCRATCH | CPU_DTRACE_KPRIV | \ + CPU_DTRACE_UPRIV | CPU_DTRACE_TUPOFLOW | \ + CPU_DTRACE_BADSTACK) +#define CPU_DTRACE_ERROR (CPU_DTRACE_FAULT | CPU_DTRACE_DROP) + +/* + * Dispatcher flags + * These flags must be changed only by the current CPU. + */ +#define CPU_DISP_DONTSTEAL 0x01 /* CPU undergoing context swtch */ +#define CPU_DISP_HALTED 0x02 /* CPU halted waiting for interrupt */ + + +#endif /* _KERNEL || _KMEMUSER */ + +#if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP) + +/* + * Macros for manipulating sets of CPUs as a bitmap. Note that this + * bitmap may vary in size depending on the maximum CPU id a specific + * platform supports. This may be different than the number of CPUs + * the platform supports, since CPU ids can be sparse. We define two + * sets of macros; one for platforms where the maximum CPU id is less + * than the number of bits in a single word (32 in a 32-bit kernel, + * 64 in a 64-bit kernel), and one for platforms that require bitmaps + * of more than one word. + */ + +#define CPUSET_WORDS BT_BITOUL(NCPU) +#define CPUSET_NOTINSET ((uint_t)-1) + +#if CPUSET_WORDS > 1 + +typedef struct cpuset { + ulong_t cpub[CPUSET_WORDS]; +} cpuset_t; + +/* + * Private functions for manipulating cpusets that do not fit in a + * single word. These should not be used directly; instead the + * CPUSET_* macros should be used so the code will be portable + * across different definitions of NCPU. + */ +extern void cpuset_all(cpuset_t *); +extern void cpuset_all_but(cpuset_t *, uint_t); +extern int cpuset_isnull(cpuset_t *); +extern int cpuset_cmp(cpuset_t *, cpuset_t *); +extern void cpuset_only(cpuset_t *, uint_t); +extern uint_t cpuset_find(cpuset_t *); +extern void cpuset_bounds(cpuset_t *, uint_t *, uint_t *); + +#define CPUSET_ALL(set) cpuset_all(&(set)) +#define CPUSET_ALL_BUT(set, cpu) cpuset_all_but(&(set), cpu) +#define CPUSET_ONLY(set, cpu) cpuset_only(&(set), cpu) +#define CPU_IN_SET(set, cpu) BT_TEST((set).cpub, cpu) +#define CPUSET_ADD(set, cpu) BT_SET((set).cpub, cpu) +#define CPUSET_DEL(set, cpu) BT_CLEAR((set).cpub, cpu) +#define CPUSET_ISNULL(set) cpuset_isnull(&(set)) +#define CPUSET_ISEQUAL(set1, set2) cpuset_cmp(&(set1), &(set2)) + +/* + * Find one CPU in the cpuset. + * Sets "cpu" to the id of the found CPU, or CPUSET_NOTINSET if no cpu + * could be found. (i.e. empty set) + */ +#define CPUSET_FIND(set, cpu) { \ + cpu = cpuset_find(&(set)); \ +} + +/* + * Determine the smallest and largest CPU id in the set. Returns + * CPUSET_NOTINSET in smallest and largest when set is empty. + */ +#define CPUSET_BOUNDS(set, smallest, largest) { \ + cpuset_bounds(&(set), &(smallest), &(largest)); \ +} + +/* + * Atomic cpuset operations + * These are safe to use for concurrent cpuset manipulations. + * "xdel" and "xadd" are exclusive operations, that set "result" to "0" + * if the add or del was successful, or "-1" if not successful. + * (e.g. attempting to add a cpu to a cpuset that's already there, or + * deleting a cpu that's not in the cpuset) + */ + +#define CPUSET_ATOMIC_DEL(set, cpu) BT_ATOMIC_CLEAR((set).cpub, (cpu)) +#define CPUSET_ATOMIC_ADD(set, cpu) BT_ATOMIC_SET((set).cpub, (cpu)) + +#define CPUSET_ATOMIC_XADD(set, cpu, result) \ + BT_ATOMIC_SET_EXCL((set).cpub, cpu, result) + +#define CPUSET_ATOMIC_XDEL(set, cpu, result) \ + BT_ATOMIC_CLEAR_EXCL((set).cpub, cpu, result) + + +#define CPUSET_OR(set1, set2) { \ + int _i; \ + for (_i = 0; _i < CPUSET_WORDS; _i++) \ + (set1).cpub[_i] |= (set2).cpub[_i]; \ +} + +#define CPUSET_XOR(set1, set2) { \ + int _i; \ + for (_i = 0; _i < CPUSET_WORDS; _i++) \ + (set1).cpub[_i] ^= (set2).cpub[_i]; \ +} + +#define CPUSET_AND(set1, set2) { \ + int _i; \ + for (_i = 0; _i < CPUSET_WORDS; _i++) \ + (set1).cpub[_i] &= (set2).cpub[_i]; \ +} + +#define CPUSET_ZERO(set) { \ + int _i; \ + for (_i = 0; _i < CPUSET_WORDS; _i++) \ + (set).cpub[_i] = 0; \ +} + +#elif CPUSET_WORDS == 1 + +typedef ulong_t cpuset_t; /* a set of CPUs */ + +#define CPUSET(cpu) (1UL << (cpu)) + +#define CPUSET_ALL(set) ((void)((set) = ~0UL)) +#define CPUSET_ALL_BUT(set, cpu) ((void)((set) = ~CPUSET(cpu))) +#define CPUSET_ONLY(set, cpu) ((void)((set) = CPUSET(cpu))) +#define CPU_IN_SET(set, cpu) ((set) & CPUSET(cpu)) +#define CPUSET_ADD(set, cpu) ((void)((set) |= CPUSET(cpu))) +#define CPUSET_DEL(set, cpu) ((void)((set) &= ~CPUSET(cpu))) +#define CPUSET_ISNULL(set) ((set) == 0) +#define CPUSET_ISEQUAL(set1, set2) ((set1) == (set2)) +#define CPUSET_OR(set1, set2) ((void)((set1) |= (set2))) +#define CPUSET_XOR(set1, set2) ((void)((set1) ^= (set2))) +#define CPUSET_AND(set1, set2) ((void)((set1) &= (set2))) +#define CPUSET_ZERO(set) ((void)((set) = 0)) + +#define CPUSET_FIND(set, cpu) { \ + cpu = (uint_t)(lowbit(set) - 1); \ +} + +#define CPUSET_BOUNDS(set, smallest, largest) { \ + smallest = (uint_t)(lowbit(set) - 1); \ + largest = (uint_t)(highbit(set) - 1); \ +} + +#define CPUSET_ATOMIC_DEL(set, cpu) atomic_and_long(&(set), ~CPUSET(cpu)) +#define CPUSET_ATOMIC_ADD(set, cpu) atomic_or_long(&(set), CPUSET(cpu)) + +#define CPUSET_ATOMIC_XADD(set, cpu, result) \ + { result = atomic_set_long_excl(&(set), (cpu)); } + +#define CPUSET_ATOMIC_XDEL(set, cpu, result) \ + { result = atomic_clear_long_excl(&(set), (cpu)); } + +#else /* CPUSET_WORDS <= 0 */ + +#error NCPU is undefined or invalid + +#endif /* CPUSET_WORDS */ + +extern cpuset_t cpu_seqid_inuse; + +#endif /* (_KERNEL || _KMEMUSER) && _MACHDEP */ + +#define CPU_CPR_OFFLINE 0x0 +#define CPU_CPR_ONLINE 0x1 +#define CPU_CPR_IS_OFFLINE(cpu) (((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE) == 0) +#define CPU_CPR_IS_ONLINE(cpu) ((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE) +#define CPU_SET_CPR_FLAGS(cpu, flag) ((cpu)->cpu_cpr_flags |= flag) + +#if defined(_KERNEL) || defined(_KMEMUSER) + +extern struct cpu *cpu[]; /* indexed by CPU number */ +extern cpu_t *cpu_list; /* list of CPUs */ +extern cpu_t *cpu_active; /* list of active CPUs */ +extern int ncpus; /* number of CPUs present */ +extern int ncpus_online; /* number of CPUs not quiesced */ +extern int max_ncpus; /* max present before ncpus is known */ +extern int boot_max_ncpus; /* like max_ncpus but for real */ +extern processorid_t max_cpuid; /* maximum CPU number */ +extern struct cpu *cpu_inmotion; /* offline or partition move target */ +extern cpu_t *clock_cpu_list; + +#if defined(__i386) || defined(__amd64) +extern struct cpu *curcpup(void); +#define CPU (curcpup()) /* Pointer to current CPU */ +#else +#define CPU (curthread->t_cpu) /* Pointer to current CPU */ +#endif + +/* + * CPU_CURRENT indicates to thread_affinity_set to use CPU->cpu_id + * as the target and to grab cpu_lock instead of requiring the caller + * to grab it. + */ +#define CPU_CURRENT -3 + +/* + * Per-CPU statistics + * + * cpu_stats_t contains numerous system and VM-related statistics, in the form + * of gauges or monotonically-increasing event occurrence counts. + */ + +#define CPU_STATS_ENTER_K() kpreempt_disable() +#define CPU_STATS_EXIT_K() kpreempt_enable() + +#define CPU_STATS_ADD_K(class, stat, amount) \ + { kpreempt_disable(); /* keep from switching CPUs */\ + CPU_STATS_ADDQ(CPU, class, stat, amount); \ + kpreempt_enable(); \ + } + +#define CPU_STATS_ADDQ(cp, class, stat, amount) { \ + extern void __dtrace_probe___cpu_##class##info_##stat(uint_t, \ + uint64_t *, cpu_t *); \ + uint64_t *stataddr = &((cp)->cpu_stats.class.stat); \ + __dtrace_probe___cpu_##class##info_##stat((amount), \ + stataddr, cp); \ + *(stataddr) += (amount); \ +} + +#define CPU_STATS(cp, stat) \ + ((cp)->cpu_stats.stat) + +#endif /* _KERNEL || _KMEMUSER */ + +/* + * CPU support routines. + */ +#if defined(_KERNEL) && defined(__STDC__) /* not for genassym.c */ + +struct zone; + +void cpu_list_init(cpu_t *); +void cpu_add_unit(cpu_t *); +void cpu_del_unit(int cpuid); +void cpu_add_active(cpu_t *); +void cpu_kstat_init(cpu_t *); +void cpu_visibility_add(cpu_t *, struct zone *); +void cpu_visibility_remove(cpu_t *, struct zone *); +void cpu_visibility_configure(cpu_t *, struct zone *); +void cpu_visibility_unconfigure(cpu_t *, struct zone *); +void cpu_visibility_online(cpu_t *, struct zone *); +void cpu_visibility_offline(cpu_t *, struct zone *); +void cpu_create_intrstat(cpu_t *); +void cpu_delete_intrstat(cpu_t *); +int cpu_kstat_intrstat_update(kstat_t *, int); +void cpu_intr_swtch_enter(kthread_t *); +void cpu_intr_swtch_exit(kthread_t *); + +void mbox_lock_init(void); /* initialize cross-call locks */ +void mbox_init(int cpun); /* initialize cross-calls */ +void poke_cpu(int cpun); /* interrupt another CPU (to preempt) */ + +/* + * values for safe_list. Pause state that CPUs are in. + */ +#define PAUSE_IDLE 0 /* normal state */ +#define PAUSE_READY 1 /* paused thread ready to spl */ +#define PAUSE_WAIT 2 /* paused thread is spl-ed high */ +#define PAUSE_DIE 3 /* tell pause thread to leave */ +#define PAUSE_DEAD 4 /* pause thread has left */ + +void mach_cpu_pause(volatile char *); + +void pause_cpus(cpu_t *off_cp); +void start_cpus(void); +int cpus_paused(void); + +void cpu_pause_init(void); +cpu_t *cpu_get(processorid_t cpun); /* get the CPU struct associated */ + +int cpu_online(cpu_t *cp); /* take cpu online */ +int cpu_offline(cpu_t *cp, int flags); /* take cpu offline */ +int cpu_spare(cpu_t *cp, int flags); /* take cpu to spare */ +int cpu_faulted(cpu_t *cp, int flags); /* take cpu to faulted */ +int cpu_poweron(cpu_t *cp); /* take powered-off cpu to offline */ +int cpu_poweroff(cpu_t *cp); /* take offline cpu to powered-off */ + +cpu_t *cpu_intr_next(cpu_t *cp); /* get next online CPU taking intrs */ +int cpu_intr_count(cpu_t *cp); /* count # of CPUs handling intrs */ +int cpu_intr_on(cpu_t *cp); /* CPU taking I/O interrupts? */ +void cpu_intr_enable(cpu_t *cp); /* enable I/O interrupts */ +int cpu_intr_disable(cpu_t *cp); /* disable I/O interrupts */ +void cpu_intr_alloc(cpu_t *cp, int n); /* allocate interrupt threads */ + +/* + * Routines for checking CPU states. + */ +int cpu_is_online(cpu_t *); /* check if CPU is online */ +int cpu_is_nointr(cpu_t *); /* check if CPU can service intrs */ +int cpu_is_active(cpu_t *); /* check if CPU can run threads */ +int cpu_is_offline(cpu_t *); /* check if CPU is offline */ +int cpu_is_poweredoff(cpu_t *); /* check if CPU is powered off */ + +int cpu_flagged_online(cpu_flag_t); /* flags show CPU is online */ +int cpu_flagged_nointr(cpu_flag_t); /* flags show CPU not handling intrs */ +int cpu_flagged_active(cpu_flag_t); /* flags show CPU scheduling threads */ +int cpu_flagged_offline(cpu_flag_t); /* flags show CPU is offline */ +int cpu_flagged_poweredoff(cpu_flag_t); /* flags show CPU is powered off */ + +/* + * The processor_info(2) state of a CPU is a simplified representation suitable + * for use by an application program. Kernel subsystems should utilize the + * internal per-CPU state as given by the cpu_flags member of the cpu structure, + * as this information may include platform- or architecture-specific state + * critical to a subsystem's disposition of a particular CPU. + */ +void cpu_set_state(cpu_t *); /* record/timestamp current state */ +int cpu_get_state(cpu_t *); /* get current cpu state */ +const char *cpu_get_state_str(cpu_t *); /* get current cpu state as string */ + + +void cpu_set_supp_freqs(cpu_t *, const char *); /* set the CPU supported */ + /* frequencies */ + +int cpu_configure(int); +int cpu_unconfigure(int); +void cpu_destroy_bound_threads(cpu_t *cp); + +extern int cpu_bind_thread(kthread_t *tp, processorid_t bind, + processorid_t *obind, int *error); +extern int cpu_unbind(processorid_t cpu_id, boolean_t force); +extern void thread_affinity_set(kthread_t *t, int cpu_id); +extern void thread_affinity_clear(kthread_t *t); +extern void affinity_set(int cpu_id); +extern void affinity_clear(void); +extern void init_cpu_mstate(struct cpu *, int); +extern void term_cpu_mstate(struct cpu *); +extern void new_cpu_mstate(int, hrtime_t); +extern void get_cpu_mstate(struct cpu *, hrtime_t *); +extern void thread_nomigrate(void); +extern void thread_allowmigrate(void); +extern void weakbinding_stop(void); +extern void weakbinding_start(void); + +/* + * The following routines affect the CPUs participation in interrupt processing, + * if that is applicable on the architecture. This only affects interrupts + * which aren't directed at the processor (not cross calls). + * + * cpu_disable_intr returns non-zero if interrupts were previously enabled. + */ +int cpu_disable_intr(struct cpu *cp); /* stop issuing interrupts to cpu */ +void cpu_enable_intr(struct cpu *cp); /* start issuing interrupts to cpu */ + +/* + * The mutex cpu_lock protects cpu_flags for all CPUs, as well as the ncpus + * and ncpus_online counts. + */ +extern kmutex_t cpu_lock; /* lock protecting CPU data */ + +typedef enum { + CPU_INIT, + CPU_CONFIG, + CPU_UNCONFIG, + CPU_ON, + CPU_OFF, + CPU_CPUPART_IN, + CPU_CPUPART_OUT +} cpu_setup_t; + +typedef int cpu_setup_func_t(cpu_setup_t, int, void *); + +/* + * Routines used to register interest in cpu's being added to or removed + * from the system. + */ +extern void register_cpu_setup_func(cpu_setup_func_t *, void *); +extern void unregister_cpu_setup_func(cpu_setup_func_t *, void *); +extern void cpu_state_change_notify(int, cpu_setup_t); + +/* + * Create various strings that describe the given CPU for the + * processor_info system call and configuration-related kstats. + */ +#define CPU_IDSTRLEN 100 + +extern void init_cpu_info(struct cpu *); +extern void cpu_vm_data_init(struct cpu *); +extern void cpu_vm_data_destroy(struct cpu *); + +#endif /* _KERNEL */ + +#ifdef __cplusplus +} +#endif + +#endif /* _SYS_CPUVAR_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/ctf.h b/sys/cddl/contrib/opensolaris/uts/common/sys/ctf.h new file mode 100644 index 000000000000..2d1987be459f --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/ctf.h @@ -0,0 +1,360 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2004 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _CTF_H +#define _CTF_H + +#if defined(sun) +#pragma ident "%Z%%M% %I% %E% SMI" +#endif + +#include <sys/types.h> + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * CTF - Compact ANSI-C Type Format + * + * This file format can be used to compactly represent the information needed + * by a debugger to interpret the ANSI-C types used by a given program. + * Traditionally, this kind of information is generated by the compiler when + * invoked with the -g flag and is stored in "stabs" strings or in the more + * modern DWARF format. CTF provides a representation of only the information + * that is relevant to debugging a complex, optimized C program such as the + * operating system kernel in a form that is significantly more compact than + * the equivalent stabs or DWARF representation. The format is data-model + * independent, so consumers do not need different code depending on whether + * they are 32-bit or 64-bit programs. CTF assumes that a standard ELF symbol + * table is available for use in the debugger, and uses the structure and data + * of the symbol table to avoid storing redundant information. The CTF data + * may be compressed on disk or in memory, indicated by a bit in the header. + * CTF may be interpreted in a raw disk file, or it may be stored in an ELF + * section, typically named .SUNW_ctf. Data structures are aligned so that + * a raw CTF file or CTF ELF section may be manipulated using mmap(2). + * + * The CTF file or section itself has the following structure: + * + * +--------+--------+---------+----------+-------+--------+ + * | file | type | data | function | data | string | + * | header | labels | objects | info | types | table | + * +--------+--------+---------+----------+-------+--------+ + * + * The file header stores a magic number and version information, encoding + * flags, and the byte offset of each of the sections relative to the end of the + * header itself. If the CTF data has been uniquified against another set of + * CTF data, a reference to that data also appears in the the header. This + * reference is the name of the label corresponding to the types uniquified + * against. + * + * Following the header is a list of labels, used to group the types included in + * the data types section. Each label is accompanied by a type ID i. A given + * label refers to the group of types whose IDs are in the range [0, i]. + * + * Data object and function records are stored in the same order as they appear + * in the corresponding symbol table, except that symbols marked SHN_UNDEF are + * not stored and symbols that have no type data are padded out with zeroes. + * For each data object, the type ID (a small integer) is recorded. For each + * function, the type ID of the return type and argument types is recorded. + * + * The data types section is a list of variable size records that represent each + * type, in order by their ID. The types themselves form a directed graph, + * where each node may contain one or more outgoing edges to other type nodes, + * denoted by their ID. + * + * Strings are recorded as a string table ID (0 or 1) and a byte offset into the + * string table. String table 0 is the internal CTF string table. String table + * 1 is the external string table, which is the string table associated with the + * ELF symbol table for this object. CTF does not record any strings that are + * already in the symbol table, and the CTF string table does not contain any + * duplicated strings. + * + * If the CTF data has been merged with another parent CTF object, some outgoing + * edges may refer to type nodes that exist in another CTF object. The debugger + * and libctf library are responsible for connecting the appropriate objects + * together so that the full set of types can be explored and manipulated. + */ + +#define CTF_MAX_TYPE 0xffff /* max type identifier value */ +#define CTF_MAX_NAME 0x7fffffff /* max offset into a string table */ +#define CTF_MAX_VLEN 0x3ff /* max struct, union, enum members or args */ +#define CTF_MAX_INTOFF 0xff /* max offset of intrinsic value in bits */ +#define CTF_MAX_INTBITS 0xffff /* max size of an intrinsic in bits */ + +/* See ctf_type_t */ +#define CTF_MAX_SIZE 0xfffe /* max size of a type in bytes */ +#define CTF_LSIZE_SENT 0xffff /* sentinel for ctt_size */ +#define CTF_MAX_LSIZE UINT64_MAX + +typedef struct ctf_preamble { + ushort_t ctp_magic; /* magic number (CTF_MAGIC) */ + uchar_t ctp_version; /* data format version number (CTF_VERSION) */ + uchar_t ctp_flags; /* flags (see below) */ +} ctf_preamble_t; + +typedef struct ctf_header { + ctf_preamble_t cth_preamble; + uint_t cth_parlabel; /* ref to name of parent lbl uniq'd against */ + uint_t cth_parname; /* ref to basename of parent */ + uint_t cth_lbloff; /* offset of label section */ + uint_t cth_objtoff; /* offset of object section */ + uint_t cth_funcoff; /* offset of function section */ + uint_t cth_typeoff; /* offset of type section */ + uint_t cth_stroff; /* offset of string section */ + uint_t cth_strlen; /* length of string section in bytes */ +} ctf_header_t; + +#define cth_magic cth_preamble.ctp_magic +#define cth_version cth_preamble.ctp_version +#define cth_flags cth_preamble.ctp_flags + +#ifdef CTF_OLD_VERSIONS + +typedef struct ctf_header_v1 { + ctf_preamble_t cth_preamble; + uint_t cth_objtoff; + uint_t cth_funcoff; + uint_t cth_typeoff; + uint_t cth_stroff; + uint_t cth_strlen; +} ctf_header_v1_t; + +#endif /* CTF_OLD_VERSIONS */ + +#define CTF_MAGIC 0xcff1 /* magic number identifying header */ + +/* data format version number */ +#define CTF_VERSION_1 1 +#define CTF_VERSION_2 2 +#define CTF_VERSION CTF_VERSION_2 /* current version */ + +#define CTF_F_COMPRESS 0x1 /* data buffer is compressed */ + +typedef struct ctf_lblent { + uint_t ctl_label; /* ref to name of label */ + uint_t ctl_typeidx; /* last type associated with this label */ +} ctf_lblent_t; + +typedef struct ctf_stype { + uint_t ctt_name; /* reference to name in string table */ + ushort_t ctt_info; /* encoded kind, variant length (see below) */ + union { + ushort_t _size; /* size of entire type in bytes */ + ushort_t _type; /* reference to another type */ + } _u; +} ctf_stype_t; + +/* + * type sizes, measured in bytes, come in two flavors. 99% of them fit within + * (USHRT_MAX - 1), and thus can be stored in the ctt_size member of a + * ctf_stype_t. The maximum value for these sizes is CTF_MAX_SIZE. The sizes + * larger than CTF_MAX_SIZE must be stored in the ctt_lsize member of a + * ctf_type_t. Use of this member is indicated by the presence of + * CTF_LSIZE_SENT in ctt_size. + */ +typedef struct ctf_type { + uint_t ctt_name; /* reference to name in string table */ + ushort_t ctt_info; /* encoded kind, variant length (see below) */ + union { + ushort_t _size; /* always CTF_LSIZE_SENT */ + ushort_t _type; /* do not use */ + } _u; + uint_t ctt_lsizehi; /* high 32 bits of type size in bytes */ + uint_t ctt_lsizelo; /* low 32 bits of type size in bytes */ +} ctf_type_t; + +#define ctt_size _u._size /* for fundamental types that have a size */ +#define ctt_type _u._type /* for types that reference another type */ + +/* + * The following macros compose and decompose values for ctt_info and + * ctt_name, as well as other structures that contain name references. + * + * ------------------------ + * ctt_info: | kind | isroot | vlen | + * ------------------------ + * 15 11 10 9 0 + * + * kind = CTF_INFO_KIND(c.ctt_info); <-- CTF_K_* value (see below) + * vlen = CTF_INFO_VLEN(c.ctt_info); <-- length of variable data list + * + * stid = CTF_NAME_STID(c.ctt_name); <-- string table id number (0 or 1) + * offset = CTF_NAME_OFFSET(c.ctt_name); <-- string table byte offset + * + * c.ctt_info = CTF_TYPE_INFO(kind, vlen); + * c.ctt_name = CTF_TYPE_NAME(stid, offset); + */ + +#define CTF_INFO_KIND(info) (((info) & 0xf800) >> 11) +#define CTF_INFO_ISROOT(info) (((info) & 0x0400) >> 10) +#define CTF_INFO_VLEN(info) (((info) & CTF_MAX_VLEN)) + +#define CTF_NAME_STID(name) ((name) >> 31) +#define CTF_NAME_OFFSET(name) ((name) & 0x7fffffff) + +#define CTF_TYPE_INFO(kind, isroot, vlen) \ + (((kind) << 11) | (((isroot) ? 1 : 0) << 10) | ((vlen) & CTF_MAX_VLEN)) + +#define CTF_TYPE_NAME(stid, offset) \ + (((stid) << 31) | ((offset) & 0x7fffffff)) + +#define CTF_TYPE_ISPARENT(id) ((id) < 0x8000) +#define CTF_TYPE_ISCHILD(id) ((id) > 0x7fff) + +#define CTF_TYPE_TO_INDEX(id) ((id) & 0x7fff) +#define CTF_INDEX_TO_TYPE(id, child) ((child) ? ((id) | 0x8000) : (id)) +#define CTF_PARENT_SHIFT 15 + +#define CTF_STRTAB_0 0 /* symbolic define for string table id 0 */ +#define CTF_STRTAB_1 1 /* symbolic define for string table id 1 */ + +#define CTF_TYPE_LSIZE(cttp) \ + (((uint64_t)(cttp)->ctt_lsizehi) << 32 | (cttp)->ctt_lsizelo) +#define CTF_SIZE_TO_LSIZE_HI(size) ((uint32_t)((uint64_t)(size) >> 32)) +#define CTF_SIZE_TO_LSIZE_LO(size) ((uint32_t)(size)) + +#ifdef CTF_OLD_VERSIONS + +#define CTF_INFO_KIND_V1(info) (((info) & 0xf000) >> 12) +#define CTF_INFO_ISROOT_V1(info) (((info) & 0x0800) >> 11) +#define CTF_INFO_VLEN_V1(info) (((info) & 0x07ff)) + +#define CTF_TYPE_INFO_V1(kind, isroot, vlen) \ + (((kind) << 12) | (((isroot) ? 1 : 0) << 11) | ((vlen) & 0x07ff)) + +#endif /* CTF_OLD_VERSIONS */ + +/* + * Values for CTF_TYPE_KIND(). If the kind has an associated data list, + * CTF_INFO_VLEN() will extract the number of elements in the list, and + * the type of each element is shown in the comments below. + */ +#define CTF_K_UNKNOWN 0 /* unknown type (used for padding) */ +#define CTF_K_INTEGER 1 /* variant data is CTF_INT_DATA() (see below) */ +#define CTF_K_FLOAT 2 /* variant data is CTF_FP_DATA() (see below) */ +#define CTF_K_POINTER 3 /* ctt_type is referenced type */ +#define CTF_K_ARRAY 4 /* variant data is single ctf_array_t */ +#define CTF_K_FUNCTION 5 /* ctt_type is return type, variant data is */ + /* list of argument types (ushort_t's) */ +#define CTF_K_STRUCT 6 /* variant data is list of ctf_member_t's */ +#define CTF_K_UNION 7 /* variant data is list of ctf_member_t's */ +#define CTF_K_ENUM 8 /* variant data is list of ctf_enum_t's */ +#define CTF_K_FORWARD 9 /* no additional data; ctt_name is tag */ +#define CTF_K_TYPEDEF 10 /* ctt_type is referenced type */ +#define CTF_K_VOLATILE 11 /* ctt_type is base type */ +#define CTF_K_CONST 12 /* ctt_type is base type */ +#define CTF_K_RESTRICT 13 /* ctt_type is base type */ + +#define CTF_K_MAX 31 /* Maximum possible CTF_K_* value */ + +/* + * Values for ctt_type when kind is CTF_K_INTEGER. The flags, offset in bits, + * and size in bits are encoded as a single word using the following macros. + */ +#define CTF_INT_ENCODING(data) (((data) & 0xff000000) >> 24) +#define CTF_INT_OFFSET(data) (((data) & 0x00ff0000) >> 16) +#define CTF_INT_BITS(data) (((data) & 0x0000ffff)) + +#define CTF_INT_DATA(encoding, offset, bits) \ + (((encoding) << 24) | ((offset) << 16) | (bits)) + +#define CTF_INT_SIGNED 0x01 /* integer is signed (otherwise unsigned) */ +#define CTF_INT_CHAR 0x02 /* character display format */ +#define CTF_INT_BOOL 0x04 /* boolean display format */ +#define CTF_INT_VARARGS 0x08 /* varargs display format */ + +/* + * Values for ctt_type when kind is CTF_K_FLOAT. The encoding, offset in bits, + * and size in bits are encoded as a single word using the following macros. + */ +#define CTF_FP_ENCODING(data) (((data) & 0xff000000) >> 24) +#define CTF_FP_OFFSET(data) (((data) & 0x00ff0000) >> 16) +#define CTF_FP_BITS(data) (((data) & 0x0000ffff)) + +#define CTF_FP_DATA(encoding, offset, bits) \ + (((encoding) << 24) | ((offset) << 16) | (bits)) + +#define CTF_FP_SINGLE 1 /* IEEE 32-bit float encoding */ +#define CTF_FP_DOUBLE 2 /* IEEE 64-bit float encoding */ +#define CTF_FP_CPLX 3 /* Complex encoding */ +#define CTF_FP_DCPLX 4 /* Double complex encoding */ +#define CTF_FP_LDCPLX 5 /* Long double complex encoding */ +#define CTF_FP_LDOUBLE 6 /* Long double encoding */ +#define CTF_FP_INTRVL 7 /* Interval (2x32-bit) encoding */ +#define CTF_FP_DINTRVL 8 /* Double interval (2x64-bit) encoding */ +#define CTF_FP_LDINTRVL 9 /* Long double interval (2x128-bit) encoding */ +#define CTF_FP_IMAGRY 10 /* Imaginary (32-bit) encoding */ +#define CTF_FP_DIMAGRY 11 /* Long imaginary (64-bit) encoding */ +#define CTF_FP_LDIMAGRY 12 /* Long double imaginary (128-bit) encoding */ + +#define CTF_FP_MAX 12 /* Maximum possible CTF_FP_* value */ + +typedef struct ctf_array { + ushort_t cta_contents; /* reference to type of array contents */ + ushort_t cta_index; /* reference to type of array index */ + uint_t cta_nelems; /* number of elements */ +} ctf_array_t; + +/* + * Most structure members have bit offsets that can be expressed using a + * short. Some don't. ctf_member_t is used for structs which cannot + * contain any of these large offsets, whereas ctf_lmember_t is used in the + * latter case. If ctt_size for a given struct is >= 8192 bytes, all members + * will be stored as type ctf_lmember_t. + */ + +#define CTF_LSTRUCT_THRESH 8192 + +typedef struct ctf_member { + uint_t ctm_name; /* reference to name in string table */ + ushort_t ctm_type; /* reference to type of member */ + ushort_t ctm_offset; /* offset of this member in bits */ +} ctf_member_t; + +typedef struct ctf_lmember { + uint_t ctlm_name; /* reference to name in string table */ + ushort_t ctlm_type; /* reference to type of member */ + ushort_t ctlm_pad; /* padding */ + uint_t ctlm_offsethi; /* high 32 bits of member offset in bits */ + uint_t ctlm_offsetlo; /* low 32 bits of member offset in bits */ +} ctf_lmember_t; + +#define CTF_LMEM_OFFSET(ctlmp) \ + (((uint64_t)(ctlmp)->ctlm_offsethi) << 32 | (ctlmp)->ctlm_offsetlo) +#define CTF_OFFSET_TO_LMEMHI(offset) ((uint32_t)((uint64_t)(offset) >> 32)) +#define CTF_OFFSET_TO_LMEMLO(offset) ((uint32_t)(offset)) + +typedef struct ctf_enum { + uint_t cte_name; /* reference to name in string table */ + int cte_value; /* value associated with this name */ +} ctf_enum_t; + +#ifdef __cplusplus +} +#endif + +#endif /* _CTF_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/ctf_api.h b/sys/cddl/contrib/opensolaris/uts/common/sys/ctf_api.h new file mode 100644 index 000000000000..cd4caaa6f971 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/ctf_api.h @@ -0,0 +1,245 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2005 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +/* + * This header file defines the interfaces available from the CTF debugger + * library, libctf, and an equivalent kernel module. This API can be used by + * a debugger to operate on data in the Compact ANSI-C Type Format (CTF). + * This is NOT a public interface, although it may eventually become one in + * the fullness of time after we gain more experience with the interfaces. + * + * In the meantime, be aware that any program linked with this API in this + * release of Solaris is almost guaranteed to break in the next release. + * + * In short, do not user this header file or the CTF routines for any purpose. + */ + +#ifndef _CTF_API_H +#define _CTF_API_H + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/types.h> +#include <sys/param.h> +#include <sys/elf.h> +#include <sys/ctf.h> + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * Clients can open one or more CTF containers and obtain a pointer to an + * opaque ctf_file_t. Types are identified by an opaque ctf_id_t token. + * These opaque definitions allow libctf to evolve without breaking clients. + */ +typedef struct ctf_file ctf_file_t; +typedef long ctf_id_t; + +/* + * If the debugger needs to provide the CTF library with a set of raw buffers + * for use as the CTF data, symbol table, and string table, it can do so by + * filling in ctf_sect_t structures and passing them to ctf_bufopen(): + */ +typedef struct ctf_sect { + char *cts_name; /* section name (if any) */ + ulong_t cts_type; /* section type (ELF SHT_... value) */ + ulong_t cts_flags; /* section flags (ELF SHF_... value) */ +#if defined(sun) + const void *cts_data; /* pointer to section data */ +#else + void *cts_data; /* pointer to section data */ +#endif + size_t cts_size; /* size of data in bytes */ + size_t cts_entsize; /* size of each section entry (symtab only) */ + off64_t cts_offset; /* file offset of this section (if any) */ +} ctf_sect_t; + +/* + * Encoding information for integers, floating-point values, and certain other + * intrinsics can be obtained by calling ctf_type_encoding(), below. The flags + * field will contain values appropriate for the type defined in <sys/ctf.h>. + */ +typedef struct ctf_encoding { + uint_t cte_format; /* data format (CTF_INT_* or CTF_FP_* flags) */ + uint_t cte_offset; /* offset of value in bits */ + uint_t cte_bits; /* size of storage in bits */ +} ctf_encoding_t; + +typedef struct ctf_membinfo { + ctf_id_t ctm_type; /* type of struct or union member */ + ulong_t ctm_offset; /* offset of member in bits */ +} ctf_membinfo_t; + +typedef struct ctf_arinfo { + ctf_id_t ctr_contents; /* type of array contents */ + ctf_id_t ctr_index; /* type of array index */ + uint_t ctr_nelems; /* number of elements */ +} ctf_arinfo_t; + +typedef struct ctf_funcinfo { + ctf_id_t ctc_return; /* function return type */ + uint_t ctc_argc; /* number of typed arguments to function */ + uint_t ctc_flags; /* function attributes (see below) */ +} ctf_funcinfo_t; + +typedef struct ctf_lblinfo { + ctf_id_t ctb_typeidx; /* last type associated with the label */ +} ctf_lblinfo_t; + +#define CTF_FUNC_VARARG 0x1 /* function arguments end with varargs */ + +/* + * Functions that return integer status or a ctf_id_t use the following value + * to indicate failure. ctf_errno() can be used to obtain an error code. + */ +#define CTF_ERR (-1L) + +/* + * The CTF data model is inferred to be the caller's data model or the data + * model of the given object, unless ctf_setmodel() is explicitly called. + */ +#define CTF_MODEL_ILP32 1 /* object data model is ILP32 */ +#define CTF_MODEL_LP64 2 /* object data model is LP64 */ +#ifdef _LP64 +#define CTF_MODEL_NATIVE CTF_MODEL_LP64 +#else +#define CTF_MODEL_NATIVE CTF_MODEL_ILP32 +#endif + +/* + * Dynamic CTF containers can be created using ctf_create(). The ctf_add_* + * routines can be used to add new definitions to the dynamic container. + * New types are labeled as root or non-root to determine whether they are + * visible at the top-level program scope when subsequently doing a lookup. + */ +#define CTF_ADD_NONROOT 0 /* type only visible in nested scope */ +#define CTF_ADD_ROOT 1 /* type visible at top-level scope */ + +/* + * These typedefs are used to define the signature for callback functions + * that can be used with the iteration and visit functions below: + */ +typedef int ctf_visit_f(const char *, ctf_id_t, ulong_t, int, void *); +typedef int ctf_member_f(const char *, ctf_id_t, ulong_t, void *); +typedef int ctf_enum_f(const char *, int, void *); +typedef int ctf_type_f(ctf_id_t, void *); +typedef int ctf_label_f(const char *, const ctf_lblinfo_t *, void *); + +extern ctf_file_t *ctf_bufopen(const ctf_sect_t *, const ctf_sect_t *, + const ctf_sect_t *, int *); +extern ctf_file_t *ctf_fdopen(int, int *); +extern ctf_file_t *ctf_open(const char *, int *); +extern ctf_file_t *ctf_create(int *); +extern void ctf_close(ctf_file_t *); + +extern ctf_file_t *ctf_parent_file(ctf_file_t *); +extern const char *ctf_parent_name(ctf_file_t *); + +extern int ctf_import(ctf_file_t *, ctf_file_t *); +extern int ctf_setmodel(ctf_file_t *, int); +extern int ctf_getmodel(ctf_file_t *); + +extern void ctf_setspecific(ctf_file_t *, void *); +extern void *ctf_getspecific(ctf_file_t *); + +extern int ctf_errno(ctf_file_t *); +extern const char *ctf_errmsg(int); +extern int ctf_version(int); + +extern int ctf_func_info(ctf_file_t *, ulong_t, ctf_funcinfo_t *); +extern int ctf_func_args(ctf_file_t *, ulong_t, uint_t, ctf_id_t *); + +extern ctf_id_t ctf_lookup_by_name(ctf_file_t *, const char *); +extern ctf_id_t ctf_lookup_by_symbol(ctf_file_t *, ulong_t); + +extern ctf_id_t ctf_type_resolve(ctf_file_t *, ctf_id_t); +extern ssize_t ctf_type_lname(ctf_file_t *, ctf_id_t, char *, size_t); +extern char *ctf_type_name(ctf_file_t *, ctf_id_t, char *, size_t); +extern ssize_t ctf_type_size(ctf_file_t *, ctf_id_t); +extern ssize_t ctf_type_align(ctf_file_t *, ctf_id_t); +extern int ctf_type_kind(ctf_file_t *, ctf_id_t); +extern ctf_id_t ctf_type_reference(ctf_file_t *, ctf_id_t); +extern ctf_id_t ctf_type_pointer(ctf_file_t *, ctf_id_t); +extern int ctf_type_encoding(ctf_file_t *, ctf_id_t, ctf_encoding_t *); +extern int ctf_type_visit(ctf_file_t *, ctf_id_t, ctf_visit_f *, void *); +extern int ctf_type_cmp(ctf_file_t *, ctf_id_t, ctf_file_t *, ctf_id_t); +extern int ctf_type_compat(ctf_file_t *, ctf_id_t, ctf_file_t *, ctf_id_t); + +extern int ctf_member_info(ctf_file_t *, ctf_id_t, const char *, + ctf_membinfo_t *); +extern int ctf_array_info(ctf_file_t *, ctf_id_t, ctf_arinfo_t *); + +extern const char *ctf_enum_name(ctf_file_t *, ctf_id_t, int); +extern int ctf_enum_value(ctf_file_t *, ctf_id_t, const char *, int *); + +extern const char *ctf_label_topmost(ctf_file_t *); +extern int ctf_label_info(ctf_file_t *, const char *, ctf_lblinfo_t *); + +extern int ctf_member_iter(ctf_file_t *, ctf_id_t, ctf_member_f *, void *); +extern int ctf_enum_iter(ctf_file_t *, ctf_id_t, ctf_enum_f *, void *); +extern int ctf_type_iter(ctf_file_t *, ctf_type_f *, void *); +extern int ctf_label_iter(ctf_file_t *, ctf_label_f *, void *); + +extern ctf_id_t ctf_add_array(ctf_file_t *, uint_t, const ctf_arinfo_t *); +extern ctf_id_t ctf_add_const(ctf_file_t *, uint_t, ctf_id_t); +extern ctf_id_t ctf_add_enum(ctf_file_t *, uint_t, const char *); +extern ctf_id_t ctf_add_float(ctf_file_t *, uint_t, + const char *, const ctf_encoding_t *); +extern ctf_id_t ctf_add_forward(ctf_file_t *, uint_t, const char *, uint_t); +extern ctf_id_t ctf_add_function(ctf_file_t *, uint_t, + const ctf_funcinfo_t *, const ctf_id_t *); +extern ctf_id_t ctf_add_integer(ctf_file_t *, uint_t, + const char *, const ctf_encoding_t *); +extern ctf_id_t ctf_add_pointer(ctf_file_t *, uint_t, ctf_id_t); +extern ctf_id_t ctf_add_type(ctf_file_t *, ctf_file_t *, ctf_id_t); +extern ctf_id_t ctf_add_typedef(ctf_file_t *, uint_t, const char *, ctf_id_t); +extern ctf_id_t ctf_add_restrict(ctf_file_t *, uint_t, ctf_id_t); +extern ctf_id_t ctf_add_struct(ctf_file_t *, uint_t, const char *); +extern ctf_id_t ctf_add_union(ctf_file_t *, uint_t, const char *); +extern ctf_id_t ctf_add_volatile(ctf_file_t *, uint_t, ctf_id_t); + +extern int ctf_add_enumerator(ctf_file_t *, ctf_id_t, const char *, int); +extern int ctf_add_member(ctf_file_t *, ctf_id_t, const char *, ctf_id_t); + +extern int ctf_set_array(ctf_file_t *, ctf_id_t, const ctf_arinfo_t *); + +extern int ctf_update(ctf_file_t *); +extern int ctf_discard(ctf_file_t *); +extern int ctf_write(ctf_file_t *, int); + +#ifdef _KERNEL + +struct module; +extern ctf_file_t *ctf_modopen(struct module *, int *); + +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* _CTF_API_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/debug.h b/sys/cddl/contrib/opensolaris/uts/common/sys/debug.h index c87c8848bc03..432e6be94dd5 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/sys/debug.h +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/debug.h @@ -49,7 +49,7 @@ extern "C" { #if defined(__STDC__) extern int assfail(const char *, const char *, int); #define VERIFY(EX) ((void)((EX) || assfail(#EX, __FILE__, __LINE__))) -#if DEBUG +#ifdef DEBUG #define ASSERT(EX) VERIFY(EX) #else #define ASSERT(x) ((void)0) @@ -57,7 +57,7 @@ extern int assfail(const char *, const char *, int); #else /* defined(__STDC__) */ extern int assfail(); #define VERIFY(EX) ((void)((EX) || assfail("EX", __FILE__, __LINE__))) -#if DEBUG +#ifdef DEBUG #define ASSERT(EX) VERIFY(EX) #else #define ASSERT(x) ((void)0) @@ -97,7 +97,7 @@ _NOTE(CONSTCOND) } while (0) #define VERIFY3S(x, y, z) VERIFY3_IMPL(x, y, z, int64_t) #define VERIFY3U(x, y, z) VERIFY3_IMPL(x, y, z, uint64_t) #define VERIFY3P(x, y, z) VERIFY3_IMPL(x, y, z, uintptr_t) -#if DEBUG +#ifdef DEBUG #define ASSERT3S(x, y, z) VERIFY3S(x, y, z) #define ASSERT3U(x, y, z) VERIFY3U(x, y, z) #define ASSERT3P(x, y, z) VERIFY3P(x, y, z) diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/dtrace.h b/sys/cddl/contrib/opensolaris/uts/common/sys/dtrace.h new file mode 100644 index 000000000000..82b97a373682 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/dtrace.h @@ -0,0 +1,2309 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _SYS_DTRACE_H +#define _SYS_DTRACE_H + +#pragma ident "%Z%%M% %I% %E% SMI" + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * DTrace Dynamic Tracing Software: Kernel Interfaces + * + * Note: The contents of this file are private to the implementation of the + * Solaris system and DTrace subsystem and are subject to change at any time + * without notice. Applications and drivers using these interfaces will fail + * to run on future releases. These interfaces should not be used for any + * purpose except those expressly outlined in dtrace(7D) and libdtrace(3LIB). + * Please refer to the "Solaris Dynamic Tracing Guide" for more information. + */ + +#ifndef _ASM + +#include <sys/types.h> +#include <sys/modctl.h> +#include <sys/processor.h> +#if defined(sun) +#include <sys/systm.h> +#else +#include <sys/param.h> +#include <sys/linker.h> +#include <sys/ioccom.h> +#include <sys/ucred.h> +typedef int model_t; +#endif +#include <sys/ctf_api.h> +#include <sys/cyclic.h> +#if defined(sun) +#include <sys/int_limits.h> +#else +#include <sys/stdint.h> +#endif + +/* + * DTrace Universal Constants and Typedefs + */ +#define DTRACE_CPUALL -1 /* all CPUs */ +#define DTRACE_IDNONE 0 /* invalid probe identifier */ +#define DTRACE_EPIDNONE 0 /* invalid enabled probe identifier */ +#define DTRACE_AGGIDNONE 0 /* invalid aggregation identifier */ +#define DTRACE_AGGVARIDNONE 0 /* invalid aggregation variable ID */ +#define DTRACE_CACHEIDNONE 0 /* invalid predicate cache */ +#define DTRACE_PROVNONE 0 /* invalid provider identifier */ +#define DTRACE_METAPROVNONE 0 /* invalid meta-provider identifier */ +#define DTRACE_ARGNONE -1 /* invalid argument index */ + +#define DTRACE_PROVNAMELEN 64 +#define DTRACE_MODNAMELEN 64 +#define DTRACE_FUNCNAMELEN 128 +#define DTRACE_NAMELEN 64 +#define DTRACE_FULLNAMELEN (DTRACE_PROVNAMELEN + DTRACE_MODNAMELEN + \ + DTRACE_FUNCNAMELEN + DTRACE_NAMELEN + 4) +#define DTRACE_ARGTYPELEN 128 + +typedef uint32_t dtrace_id_t; /* probe identifier */ +typedef uint32_t dtrace_epid_t; /* enabled probe identifier */ +typedef uint32_t dtrace_aggid_t; /* aggregation identifier */ +typedef int64_t dtrace_aggvarid_t; /* aggregation variable identifier */ +typedef uint16_t dtrace_actkind_t; /* action kind */ +typedef int64_t dtrace_optval_t; /* option value */ +typedef uint32_t dtrace_cacheid_t; /* predicate cache identifier */ + +typedef enum dtrace_probespec { + DTRACE_PROBESPEC_NONE = -1, + DTRACE_PROBESPEC_PROVIDER = 0, + DTRACE_PROBESPEC_MOD, + DTRACE_PROBESPEC_FUNC, + DTRACE_PROBESPEC_NAME +} dtrace_probespec_t; + +/* + * DTrace Intermediate Format (DIF) + * + * The following definitions describe the DTrace Intermediate Format (DIF), a + * a RISC-like instruction set and program encoding used to represent + * predicates and actions that can be bound to DTrace probes. The constants + * below defining the number of available registers are suggested minimums; the + * compiler should use DTRACEIOC_CONF to dynamically obtain the number of + * registers provided by the current DTrace implementation. + */ +#define DIF_VERSION_1 1 /* DIF version 1: Solaris 10 Beta */ +#define DIF_VERSION_2 2 /* DIF version 2: Solaris 10 FCS */ +#define DIF_VERSION DIF_VERSION_2 /* latest DIF instruction set version */ +#define DIF_DIR_NREGS 8 /* number of DIF integer registers */ +#define DIF_DTR_NREGS 8 /* number of DIF tuple registers */ + +#define DIF_OP_OR 1 /* or r1, r2, rd */ +#define DIF_OP_XOR 2 /* xor r1, r2, rd */ +#define DIF_OP_AND 3 /* and r1, r2, rd */ +#define DIF_OP_SLL 4 /* sll r1, r2, rd */ +#define DIF_OP_SRL 5 /* srl r1, r2, rd */ +#define DIF_OP_SUB 6 /* sub r1, r2, rd */ +#define DIF_OP_ADD 7 /* add r1, r2, rd */ +#define DIF_OP_MUL 8 /* mul r1, r2, rd */ +#define DIF_OP_SDIV 9 /* sdiv r1, r2, rd */ +#define DIF_OP_UDIV 10 /* udiv r1, r2, rd */ +#define DIF_OP_SREM 11 /* srem r1, r2, rd */ +#define DIF_OP_UREM 12 /* urem r1, r2, rd */ +#define DIF_OP_NOT 13 /* not r1, rd */ +#define DIF_OP_MOV 14 /* mov r1, rd */ +#define DIF_OP_CMP 15 /* cmp r1, r2 */ +#define DIF_OP_TST 16 /* tst r1 */ +#define DIF_OP_BA 17 /* ba label */ +#define DIF_OP_BE 18 /* be label */ +#define DIF_OP_BNE 19 /* bne label */ +#define DIF_OP_BG 20 /* bg label */ +#define DIF_OP_BGU 21 /* bgu label */ +#define DIF_OP_BGE 22 /* bge label */ +#define DIF_OP_BGEU 23 /* bgeu label */ +#define DIF_OP_BL 24 /* bl label */ +#define DIF_OP_BLU 25 /* blu label */ +#define DIF_OP_BLE 26 /* ble label */ +#define DIF_OP_BLEU 27 /* bleu label */ +#define DIF_OP_LDSB 28 /* ldsb [r1], rd */ +#define DIF_OP_LDSH 29 /* ldsh [r1], rd */ +#define DIF_OP_LDSW 30 /* ldsw [r1], rd */ +#define DIF_OP_LDUB 31 /* ldub [r1], rd */ +#define DIF_OP_LDUH 32 /* lduh [r1], rd */ +#define DIF_OP_LDUW 33 /* lduw [r1], rd */ +#define DIF_OP_LDX 34 /* ldx [r1], rd */ +#define DIF_OP_RET 35 /* ret rd */ +#define DIF_OP_NOP 36 /* nop */ +#define DIF_OP_SETX 37 /* setx intindex, rd */ +#define DIF_OP_SETS 38 /* sets strindex, rd */ +#define DIF_OP_SCMP 39 /* scmp r1, r2 */ +#define DIF_OP_LDGA 40 /* ldga var, ri, rd */ +#define DIF_OP_LDGS 41 /* ldgs var, rd */ +#define DIF_OP_STGS 42 /* stgs var, rs */ +#define DIF_OP_LDTA 43 /* ldta var, ri, rd */ +#define DIF_OP_LDTS 44 /* ldts var, rd */ +#define DIF_OP_STTS 45 /* stts var, rs */ +#define DIF_OP_SRA 46 /* sra r1, r2, rd */ +#define DIF_OP_CALL 47 /* call subr, rd */ +#define DIF_OP_PUSHTR 48 /* pushtr type, rs, rr */ +#define DIF_OP_PUSHTV 49 /* pushtv type, rs, rv */ +#define DIF_OP_POPTS 50 /* popts */ +#define DIF_OP_FLUSHTS 51 /* flushts */ +#define DIF_OP_LDGAA 52 /* ldgaa var, rd */ +#define DIF_OP_LDTAA 53 /* ldtaa var, rd */ +#define DIF_OP_STGAA 54 /* stgaa var, rs */ +#define DIF_OP_STTAA 55 /* sttaa var, rs */ +#define DIF_OP_LDLS 56 /* ldls var, rd */ +#define DIF_OP_STLS 57 /* stls var, rs */ +#define DIF_OP_ALLOCS 58 /* allocs r1, rd */ +#define DIF_OP_COPYS 59 /* copys r1, r2, rd */ +#define DIF_OP_STB 60 /* stb r1, [rd] */ +#define DIF_OP_STH 61 /* sth r1, [rd] */ +#define DIF_OP_STW 62 /* stw r1, [rd] */ +#define DIF_OP_STX 63 /* stx r1, [rd] */ +#define DIF_OP_ULDSB 64 /* uldsb [r1], rd */ +#define DIF_OP_ULDSH 65 /* uldsh [r1], rd */ +#define DIF_OP_ULDSW 66 /* uldsw [r1], rd */ +#define DIF_OP_ULDUB 67 /* uldub [r1], rd */ +#define DIF_OP_ULDUH 68 /* ulduh [r1], rd */ +#define DIF_OP_ULDUW 69 /* ulduw [r1], rd */ +#define DIF_OP_ULDX 70 /* uldx [r1], rd */ +#define DIF_OP_RLDSB 71 /* rldsb [r1], rd */ +#define DIF_OP_RLDSH 72 /* rldsh [r1], rd */ +#define DIF_OP_RLDSW 73 /* rldsw [r1], rd */ +#define DIF_OP_RLDUB 74 /* rldub [r1], rd */ +#define DIF_OP_RLDUH 75 /* rlduh [r1], rd */ +#define DIF_OP_RLDUW 76 /* rlduw [r1], rd */ +#define DIF_OP_RLDX 77 /* rldx [r1], rd */ +#define DIF_OP_XLATE 78 /* xlate xlrindex, rd */ +#define DIF_OP_XLARG 79 /* xlarg xlrindex, rd */ + +#define DIF_INTOFF_MAX 0xffff /* highest integer table offset */ +#define DIF_STROFF_MAX 0xffff /* highest string table offset */ +#define DIF_REGISTER_MAX 0xff /* highest register number */ +#define DIF_VARIABLE_MAX 0xffff /* highest variable identifier */ +#define DIF_SUBROUTINE_MAX 0xffff /* highest subroutine code */ + +#define DIF_VAR_ARRAY_MIN 0x0000 /* lowest numbered array variable */ +#define DIF_VAR_ARRAY_UBASE 0x0080 /* lowest user-defined array */ +#define DIF_VAR_ARRAY_MAX 0x00ff /* highest numbered array variable */ + +#define DIF_VAR_OTHER_MIN 0x0100 /* lowest numbered scalar or assc */ +#define DIF_VAR_OTHER_UBASE 0x0500 /* lowest user-defined scalar or assc */ +#define DIF_VAR_OTHER_MAX 0xffff /* highest numbered scalar or assc */ + +#define DIF_VAR_ARGS 0x0000 /* arguments array */ +#define DIF_VAR_REGS 0x0001 /* registers array */ +#define DIF_VAR_UREGS 0x0002 /* user registers array */ +#define DIF_VAR_CURTHREAD 0x0100 /* thread pointer */ +#define DIF_VAR_TIMESTAMP 0x0101 /* timestamp */ +#define DIF_VAR_VTIMESTAMP 0x0102 /* virtual timestamp */ +#define DIF_VAR_IPL 0x0103 /* interrupt priority level */ +#define DIF_VAR_EPID 0x0104 /* enabled probe ID */ +#define DIF_VAR_ID 0x0105 /* probe ID */ +#define DIF_VAR_ARG0 0x0106 /* first argument */ +#define DIF_VAR_ARG1 0x0107 /* second argument */ +#define DIF_VAR_ARG2 0x0108 /* third argument */ +#define DIF_VAR_ARG3 0x0109 /* fourth argument */ +#define DIF_VAR_ARG4 0x010a /* fifth argument */ +#define DIF_VAR_ARG5 0x010b /* sixth argument */ +#define DIF_VAR_ARG6 0x010c /* seventh argument */ +#define DIF_VAR_ARG7 0x010d /* eighth argument */ +#define DIF_VAR_ARG8 0x010e /* ninth argument */ +#define DIF_VAR_ARG9 0x010f /* tenth argument */ +#define DIF_VAR_STACKDEPTH 0x0110 /* stack depth */ +#define DIF_VAR_CALLER 0x0111 /* caller */ +#define DIF_VAR_PROBEPROV 0x0112 /* probe provider */ +#define DIF_VAR_PROBEMOD 0x0113 /* probe module */ +#define DIF_VAR_PROBEFUNC 0x0114 /* probe function */ +#define DIF_VAR_PROBENAME 0x0115 /* probe name */ +#define DIF_VAR_PID 0x0116 /* process ID */ +#define DIF_VAR_TID 0x0117 /* (per-process) thread ID */ +#define DIF_VAR_EXECNAME 0x0118 /* name of executable */ +#define DIF_VAR_ZONENAME 0x0119 /* zone name associated with process */ +#define DIF_VAR_WALLTIMESTAMP 0x011a /* wall-clock timestamp */ +#define DIF_VAR_USTACKDEPTH 0x011b /* user-land stack depth */ +#define DIF_VAR_UCALLER 0x011c /* user-level caller */ +#define DIF_VAR_PPID 0x011d /* parent process ID */ +#define DIF_VAR_UID 0x011e /* process user ID */ +#define DIF_VAR_GID 0x011f /* process group ID */ +#define DIF_VAR_ERRNO 0x0120 /* thread errno */ +#define DIF_VAR_EXECARGS 0x0121 /* process arguments */ + +#define DIF_SUBR_RAND 0 +#define DIF_SUBR_MUTEX_OWNED 1 +#define DIF_SUBR_MUTEX_OWNER 2 +#define DIF_SUBR_MUTEX_TYPE_ADAPTIVE 3 +#define DIF_SUBR_MUTEX_TYPE_SPIN 4 +#define DIF_SUBR_RW_READ_HELD 5 +#define DIF_SUBR_RW_WRITE_HELD 6 +#define DIF_SUBR_RW_ISWRITER 7 +#define DIF_SUBR_COPYIN 8 +#define DIF_SUBR_COPYINSTR 9 +#define DIF_SUBR_SPECULATION 10 +#define DIF_SUBR_PROGENYOF 11 +#define DIF_SUBR_STRLEN 12 +#define DIF_SUBR_COPYOUT 13 +#define DIF_SUBR_COPYOUTSTR 14 +#define DIF_SUBR_ALLOCA 15 +#define DIF_SUBR_BCOPY 16 +#define DIF_SUBR_COPYINTO 17 +#define DIF_SUBR_MSGDSIZE 18 +#define DIF_SUBR_MSGSIZE 19 +#define DIF_SUBR_GETMAJOR 20 +#define DIF_SUBR_GETMINOR 21 +#define DIF_SUBR_DDI_PATHNAME 22 +#define DIF_SUBR_STRJOIN 23 +#define DIF_SUBR_LLTOSTR 24 +#define DIF_SUBR_BASENAME 25 +#define DIF_SUBR_DIRNAME 26 +#define DIF_SUBR_CLEANPATH 27 +#define DIF_SUBR_STRCHR 28 +#define DIF_SUBR_STRRCHR 29 +#define DIF_SUBR_STRSTR 30 +#define DIF_SUBR_STRTOK 31 +#define DIF_SUBR_SUBSTR 32 +#define DIF_SUBR_INDEX 33 +#define DIF_SUBR_RINDEX 34 +#define DIF_SUBR_HTONS 35 +#define DIF_SUBR_HTONL 36 +#define DIF_SUBR_HTONLL 37 +#define DIF_SUBR_NTOHS 38 +#define DIF_SUBR_NTOHL 39 +#define DIF_SUBR_NTOHLL 40 +#define DIF_SUBR_INET_NTOP 41 +#define DIF_SUBR_INET_NTOA 42 +#define DIF_SUBR_INET_NTOA6 43 +#define DIF_SUBR_MEMREF 44 +#define DIF_SUBR_TYPEREF 45 +#define DIF_SUBR_SX_SHARED_HELD 46 +#define DIF_SUBR_SX_EXCLUSIVE_HELD 47 +#define DIF_SUBR_SX_ISEXCLUSIVE 48 + +#define DIF_SUBR_MAX 48 /* max subroutine value */ + +typedef uint32_t dif_instr_t; + +#define DIF_INSTR_OP(i) (((i) >> 24) & 0xff) +#define DIF_INSTR_R1(i) (((i) >> 16) & 0xff) +#define DIF_INSTR_R2(i) (((i) >> 8) & 0xff) +#define DIF_INSTR_RD(i) ((i) & 0xff) +#define DIF_INSTR_RS(i) ((i) & 0xff) +#define DIF_INSTR_LABEL(i) ((i) & 0xffffff) +#define DIF_INSTR_VAR(i) (((i) >> 8) & 0xffff) +#define DIF_INSTR_INTEGER(i) (((i) >> 8) & 0xffff) +#define DIF_INSTR_STRING(i) (((i) >> 8) & 0xffff) +#define DIF_INSTR_SUBR(i) (((i) >> 8) & 0xffff) +#define DIF_INSTR_TYPE(i) (((i) >> 16) & 0xff) +#define DIF_INSTR_XLREF(i) (((i) >> 8) & 0xffff) + +#define DIF_INSTR_FMT(op, r1, r2, d) \ + (((op) << 24) | ((r1) << 16) | ((r2) << 8) | (d)) + +#define DIF_INSTR_NOT(r1, d) (DIF_INSTR_FMT(DIF_OP_NOT, r1, 0, d)) +#define DIF_INSTR_MOV(r1, d) (DIF_INSTR_FMT(DIF_OP_MOV, r1, 0, d)) +#define DIF_INSTR_CMP(op, r1, r2) (DIF_INSTR_FMT(op, r1, r2, 0)) +#define DIF_INSTR_TST(r1) (DIF_INSTR_FMT(DIF_OP_TST, r1, 0, 0)) +#define DIF_INSTR_BRANCH(op, label) (((op) << 24) | (label)) +#define DIF_INSTR_LOAD(op, r1, d) (DIF_INSTR_FMT(op, r1, 0, d)) +#define DIF_INSTR_STORE(op, r1, d) (DIF_INSTR_FMT(op, r1, 0, d)) +#define DIF_INSTR_SETX(i, d) ((DIF_OP_SETX << 24) | ((i) << 8) | (d)) +#define DIF_INSTR_SETS(s, d) ((DIF_OP_SETS << 24) | ((s) << 8) | (d)) +#define DIF_INSTR_RET(d) (DIF_INSTR_FMT(DIF_OP_RET, 0, 0, d)) +#define DIF_INSTR_NOP (DIF_OP_NOP << 24) +#define DIF_INSTR_LDA(op, v, r, d) (DIF_INSTR_FMT(op, v, r, d)) +#define DIF_INSTR_LDV(op, v, d) (((op) << 24) | ((v) << 8) | (d)) +#define DIF_INSTR_STV(op, v, rs) (((op) << 24) | ((v) << 8) | (rs)) +#define DIF_INSTR_CALL(s, d) ((DIF_OP_CALL << 24) | ((s) << 8) | (d)) +#define DIF_INSTR_PUSHTS(op, t, r2, rs) (DIF_INSTR_FMT(op, t, r2, rs)) +#define DIF_INSTR_POPTS (DIF_OP_POPTS << 24) +#define DIF_INSTR_FLUSHTS (DIF_OP_FLUSHTS << 24) +#define DIF_INSTR_ALLOCS(r1, d) (DIF_INSTR_FMT(DIF_OP_ALLOCS, r1, 0, d)) +#define DIF_INSTR_COPYS(r1, r2, d) (DIF_INSTR_FMT(DIF_OP_COPYS, r1, r2, d)) +#define DIF_INSTR_XLATE(op, r, d) (((op) << 24) | ((r) << 8) | (d)) + +#define DIF_REG_R0 0 /* %r0 is always set to zero */ + +/* + * A DTrace Intermediate Format Type (DIF Type) is used to represent the types + * of variables, function and associative array arguments, and the return type + * for each DIF object (shown below). It contains a description of the type, + * its size in bytes, and a module identifier. + */ +typedef struct dtrace_diftype { + uint8_t dtdt_kind; /* type kind (see below) */ + uint8_t dtdt_ckind; /* type kind in CTF */ + uint8_t dtdt_flags; /* type flags (see below) */ + uint8_t dtdt_pad; /* reserved for future use */ + uint32_t dtdt_size; /* type size in bytes (unless string) */ +} dtrace_diftype_t; + +#define DIF_TYPE_CTF 0 /* type is a CTF type */ +#define DIF_TYPE_STRING 1 /* type is a D string */ + +#define DIF_TF_BYREF 0x1 /* type is passed by reference */ + +/* + * A DTrace Intermediate Format variable record is used to describe each of the + * variables referenced by a given DIF object. It contains an integer variable + * identifier along with variable scope and properties, as shown below. The + * size of this structure must be sizeof (int) aligned. + */ +typedef struct dtrace_difv { + uint32_t dtdv_name; /* variable name index in dtdo_strtab */ + uint32_t dtdv_id; /* variable reference identifier */ + uint8_t dtdv_kind; /* variable kind (see below) */ + uint8_t dtdv_scope; /* variable scope (see below) */ + uint16_t dtdv_flags; /* variable flags (see below) */ + dtrace_diftype_t dtdv_type; /* variable type (see above) */ +} dtrace_difv_t; + +#define DIFV_KIND_ARRAY 0 /* variable is an array of quantities */ +#define DIFV_KIND_SCALAR 1 /* variable is a scalar quantity */ + +#define DIFV_SCOPE_GLOBAL 0 /* variable has global scope */ +#define DIFV_SCOPE_THREAD 1 /* variable has thread scope */ +#define DIFV_SCOPE_LOCAL 2 /* variable has local scope */ + +#define DIFV_F_REF 0x1 /* variable is referenced by DIFO */ +#define DIFV_F_MOD 0x2 /* variable is written by DIFO */ + +/* + * DTrace Actions + * + * The upper byte determines the class of the action; the low bytes determines + * the specific action within that class. The classes of actions are as + * follows: + * + * [ no class ] <= May record process- or kernel-related data + * DTRACEACT_PROC <= Only records process-related data + * DTRACEACT_PROC_DESTRUCTIVE <= Potentially destructive to processes + * DTRACEACT_KERNEL <= Only records kernel-related data + * DTRACEACT_KERNEL_DESTRUCTIVE <= Potentially destructive to the kernel + * DTRACEACT_SPECULATIVE <= Speculation-related action + * DTRACEACT_AGGREGATION <= Aggregating action + */ +#define DTRACEACT_NONE 0 /* no action */ +#define DTRACEACT_DIFEXPR 1 /* action is DIF expression */ +#define DTRACEACT_EXIT 2 /* exit() action */ +#define DTRACEACT_PRINTF 3 /* printf() action */ +#define DTRACEACT_PRINTA 4 /* printa() action */ +#define DTRACEACT_LIBACT 5 /* library-controlled action */ +#define DTRACEACT_PRINTM 6 /* printm() action */ +#define DTRACEACT_PRINTT 7 /* printt() action */ + +#define DTRACEACT_PROC 0x0100 +#define DTRACEACT_USTACK (DTRACEACT_PROC + 1) +#define DTRACEACT_JSTACK (DTRACEACT_PROC + 2) +#define DTRACEACT_USYM (DTRACEACT_PROC + 3) +#define DTRACEACT_UMOD (DTRACEACT_PROC + 4) +#define DTRACEACT_UADDR (DTRACEACT_PROC + 5) + +#define DTRACEACT_PROC_DESTRUCTIVE 0x0200 +#define DTRACEACT_STOP (DTRACEACT_PROC_DESTRUCTIVE + 1) +#define DTRACEACT_RAISE (DTRACEACT_PROC_DESTRUCTIVE + 2) +#define DTRACEACT_SYSTEM (DTRACEACT_PROC_DESTRUCTIVE + 3) +#define DTRACEACT_FREOPEN (DTRACEACT_PROC_DESTRUCTIVE + 4) + +#define DTRACEACT_PROC_CONTROL 0x0300 + +#define DTRACEACT_KERNEL 0x0400 +#define DTRACEACT_STACK (DTRACEACT_KERNEL + 1) +#define DTRACEACT_SYM (DTRACEACT_KERNEL + 2) +#define DTRACEACT_MOD (DTRACEACT_KERNEL + 3) + +#define DTRACEACT_KERNEL_DESTRUCTIVE 0x0500 +#define DTRACEACT_BREAKPOINT (DTRACEACT_KERNEL_DESTRUCTIVE + 1) +#define DTRACEACT_PANIC (DTRACEACT_KERNEL_DESTRUCTIVE + 2) +#define DTRACEACT_CHILL (DTRACEACT_KERNEL_DESTRUCTIVE + 3) + +#define DTRACEACT_SPECULATIVE 0x0600 +#define DTRACEACT_SPECULATE (DTRACEACT_SPECULATIVE + 1) +#define DTRACEACT_COMMIT (DTRACEACT_SPECULATIVE + 2) +#define DTRACEACT_DISCARD (DTRACEACT_SPECULATIVE + 3) + +#define DTRACEACT_CLASS(x) ((x) & 0xff00) + +#define DTRACEACT_ISDESTRUCTIVE(x) \ + (DTRACEACT_CLASS(x) == DTRACEACT_PROC_DESTRUCTIVE || \ + DTRACEACT_CLASS(x) == DTRACEACT_KERNEL_DESTRUCTIVE) + +#define DTRACEACT_ISSPECULATIVE(x) \ + (DTRACEACT_CLASS(x) == DTRACEACT_SPECULATIVE) + +#define DTRACEACT_ISPRINTFLIKE(x) \ + ((x) == DTRACEACT_PRINTF || (x) == DTRACEACT_PRINTA || \ + (x) == DTRACEACT_SYSTEM || (x) == DTRACEACT_FREOPEN) + +/* + * DTrace Aggregating Actions + * + * These are functions f(x) for which the following is true: + * + * f(f(x_0) U f(x_1) U ... U f(x_n)) = f(x_0 U x_1 U ... U x_n) + * + * where x_n is a set of arbitrary data. Aggregating actions are in their own + * DTrace action class, DTTRACEACT_AGGREGATION. The macros provided here allow + * for easier processing of the aggregation argument and data payload for a few + * aggregating actions (notably: quantize(), lquantize(), and ustack()). + */ +#define DTRACEACT_AGGREGATION 0x0700 +#define DTRACEAGG_COUNT (DTRACEACT_AGGREGATION + 1) +#define DTRACEAGG_MIN (DTRACEACT_AGGREGATION + 2) +#define DTRACEAGG_MAX (DTRACEACT_AGGREGATION + 3) +#define DTRACEAGG_AVG (DTRACEACT_AGGREGATION + 4) +#define DTRACEAGG_SUM (DTRACEACT_AGGREGATION + 5) +#define DTRACEAGG_STDDEV (DTRACEACT_AGGREGATION + 6) +#define DTRACEAGG_QUANTIZE (DTRACEACT_AGGREGATION + 7) +#define DTRACEAGG_LQUANTIZE (DTRACEACT_AGGREGATION + 8) + +#define DTRACEACT_ISAGG(x) \ + (DTRACEACT_CLASS(x) == DTRACEACT_AGGREGATION) + +#define DTRACE_QUANTIZE_NBUCKETS \ + (((sizeof (uint64_t) * NBBY) - 1) * 2 + 1) + +#define DTRACE_QUANTIZE_ZEROBUCKET ((sizeof (uint64_t) * NBBY) - 1) + +#define DTRACE_QUANTIZE_BUCKETVAL(buck) \ + (int64_t)((buck) < DTRACE_QUANTIZE_ZEROBUCKET ? \ + -(1LL << (DTRACE_QUANTIZE_ZEROBUCKET - 1 - (buck))) : \ + (buck) == DTRACE_QUANTIZE_ZEROBUCKET ? 0 : \ + 1LL << ((buck) - DTRACE_QUANTIZE_ZEROBUCKET - 1)) + +#define DTRACE_LQUANTIZE_STEPSHIFT 48 +#define DTRACE_LQUANTIZE_STEPMASK ((uint64_t)UINT16_MAX << 48) +#define DTRACE_LQUANTIZE_LEVELSHIFT 32 +#define DTRACE_LQUANTIZE_LEVELMASK ((uint64_t)UINT16_MAX << 32) +#define DTRACE_LQUANTIZE_BASESHIFT 0 +#define DTRACE_LQUANTIZE_BASEMASK UINT32_MAX + +#define DTRACE_LQUANTIZE_STEP(x) \ + (uint16_t)(((x) & DTRACE_LQUANTIZE_STEPMASK) >> \ + DTRACE_LQUANTIZE_STEPSHIFT) + +#define DTRACE_LQUANTIZE_LEVELS(x) \ + (uint16_t)(((x) & DTRACE_LQUANTIZE_LEVELMASK) >> \ + DTRACE_LQUANTIZE_LEVELSHIFT) + +#define DTRACE_LQUANTIZE_BASE(x) \ + (int32_t)(((x) & DTRACE_LQUANTIZE_BASEMASK) >> \ + DTRACE_LQUANTIZE_BASESHIFT) + +#define DTRACE_USTACK_NFRAMES(x) (uint32_t)((x) & UINT32_MAX) +#define DTRACE_USTACK_STRSIZE(x) (uint32_t)((x) >> 32) +#define DTRACE_USTACK_ARG(x, y) \ + ((((uint64_t)(y)) << 32) | ((x) & UINT32_MAX)) + +#ifndef _LP64 +#if BYTE_ORDER == _BIG_ENDIAN +#define DTRACE_PTR(type, name) uint32_t name##pad; type *name +#else +#define DTRACE_PTR(type, name) type *name; uint32_t name##pad +#endif +#else +#define DTRACE_PTR(type, name) type *name +#endif + +/* + * DTrace Object Format (DOF) + * + * DTrace programs can be persistently encoded in the DOF format so that they + * may be embedded in other programs (for example, in an ELF file) or in the + * dtrace driver configuration file for use in anonymous tracing. The DOF + * format is versioned and extensible so that it can be revised and so that + * internal data structures can be modified or extended compatibly. All DOF + * structures use fixed-size types, so the 32-bit and 64-bit representations + * are identical and consumers can use either data model transparently. + * + * The file layout is structured as follows: + * + * +---------------+-------------------+----- ... ----+---- ... ------+ + * | dof_hdr_t | dof_sec_t[ ... ] | loadable | non-loadable | + * | (file header) | (section headers) | section data | section data | + * +---------------+-------------------+----- ... ----+---- ... ------+ + * |<------------ dof_hdr.dofh_loadsz --------------->| | + * |<------------ dof_hdr.dofh_filesz ------------------------------->| + * + * The file header stores meta-data including a magic number, data model for + * the instrumentation, data encoding, and properties of the DIF code within. + * The header describes its own size and the size of the section headers. By + * convention, an array of section headers follows the file header, and then + * the data for all loadable sections and unloadable sections. This permits + * consumer code to easily download the headers and all loadable data into the + * DTrace driver in one contiguous chunk, omitting other extraneous sections. + * + * The section headers describe the size, offset, alignment, and section type + * for each section. Sections are described using a set of #defines that tell + * the consumer what kind of data is expected. Sections can contain links to + * other sections by storing a dof_secidx_t, an index into the section header + * array, inside of the section data structures. The section header includes + * an entry size so that sections with data arrays can grow their structures. + * + * The DOF data itself can contain many snippets of DIF (i.e. >1 DIFOs), which + * are represented themselves as a collection of related DOF sections. This + * permits us to change the set of sections associated with a DIFO over time, + * and also permits us to encode DIFOs that contain different sets of sections. + * When a DOF section wants to refer to a DIFO, it stores the dof_secidx_t of a + * section of type DOF_SECT_DIFOHDR. This section's data is then an array of + * dof_secidx_t's which in turn denote the sections associated with this DIFO. + * + * This loose coupling of the file structure (header and sections) to the + * structure of the DTrace program itself (ECB descriptions, action + * descriptions, and DIFOs) permits activities such as relocation processing + * to occur in a single pass without having to understand D program structure. + * + * Finally, strings are always stored in ELF-style string tables along with a + * string table section index and string table offset. Therefore strings in + * DOF are always arbitrary-length and not bound to the current implementation. + */ + +#define DOF_ID_SIZE 16 /* total size of dofh_ident[] in bytes */ + +typedef struct dof_hdr { + uint8_t dofh_ident[DOF_ID_SIZE]; /* identification bytes (see below) */ + uint32_t dofh_flags; /* file attribute flags (if any) */ + uint32_t dofh_hdrsize; /* size of file header in bytes */ + uint32_t dofh_secsize; /* size of section header in bytes */ + uint32_t dofh_secnum; /* number of section headers */ + uint64_t dofh_secoff; /* file offset of section headers */ + uint64_t dofh_loadsz; /* file size of loadable portion */ + uint64_t dofh_filesz; /* file size of entire DOF file */ + uint64_t dofh_pad; /* reserved for future use */ +} dof_hdr_t; + +#define DOF_ID_MAG0 0 /* first byte of magic number */ +#define DOF_ID_MAG1 1 /* second byte of magic number */ +#define DOF_ID_MAG2 2 /* third byte of magic number */ +#define DOF_ID_MAG3 3 /* fourth byte of magic number */ +#define DOF_ID_MODEL 4 /* DOF data model (see below) */ +#define DOF_ID_ENCODING 5 /* DOF data encoding (see below) */ +#define DOF_ID_VERSION 6 /* DOF file format major version (see below) */ +#define DOF_ID_DIFVERS 7 /* DIF instruction set version */ +#define DOF_ID_DIFIREG 8 /* DIF integer registers used by compiler */ +#define DOF_ID_DIFTREG 9 /* DIF tuple registers used by compiler */ +#define DOF_ID_PAD 10 /* start of padding bytes (all zeroes) */ + +#define DOF_MAG_MAG0 0x7F /* DOF_ID_MAG[0-3] */ +#define DOF_MAG_MAG1 'D' +#define DOF_MAG_MAG2 'O' +#define DOF_MAG_MAG3 'F' + +#define DOF_MAG_STRING "\177DOF" +#define DOF_MAG_STRLEN 4 + +#define DOF_MODEL_NONE 0 /* DOF_ID_MODEL */ +#define DOF_MODEL_ILP32 1 +#define DOF_MODEL_LP64 2 + +#ifdef _LP64 +#define DOF_MODEL_NATIVE DOF_MODEL_LP64 +#else +#define DOF_MODEL_NATIVE DOF_MODEL_ILP32 +#endif + +#define DOF_ENCODE_NONE 0 /* DOF_ID_ENCODING */ +#define DOF_ENCODE_LSB 1 +#define DOF_ENCODE_MSB 2 + +#if BYTE_ORDER == _BIG_ENDIAN +#define DOF_ENCODE_NATIVE DOF_ENCODE_MSB +#else +#define DOF_ENCODE_NATIVE DOF_ENCODE_LSB +#endif + +#define DOF_VERSION_1 1 /* DOF version 1: Solaris 10 FCS */ +#define DOF_VERSION_2 2 /* DOF version 2: Solaris Express 6/06 */ +#define DOF_VERSION DOF_VERSION_2 /* Latest DOF version */ + +#define DOF_FL_VALID 0 /* mask of all valid dofh_flags bits */ + +typedef uint32_t dof_secidx_t; /* section header table index type */ +typedef uint32_t dof_stridx_t; /* string table index type */ + +#define DOF_SECIDX_NONE (-1U) /* null value for section indices */ +#define DOF_STRIDX_NONE (-1U) /* null value for string indices */ + +typedef struct dof_sec { + uint32_t dofs_type; /* section type (see below) */ + uint32_t dofs_align; /* section data memory alignment */ + uint32_t dofs_flags; /* section flags (if any) */ + uint32_t dofs_entsize; /* size of section entry (if table) */ + uint64_t dofs_offset; /* offset of section data within file */ + uint64_t dofs_size; /* size of section data in bytes */ +} dof_sec_t; + +#define DOF_SECT_NONE 0 /* null section */ +#define DOF_SECT_COMMENTS 1 /* compiler comments */ +#define DOF_SECT_SOURCE 2 /* D program source code */ +#define DOF_SECT_ECBDESC 3 /* dof_ecbdesc_t */ +#define DOF_SECT_PROBEDESC 4 /* dof_probedesc_t */ +#define DOF_SECT_ACTDESC 5 /* dof_actdesc_t array */ +#define DOF_SECT_DIFOHDR 6 /* dof_difohdr_t (variable length) */ +#define DOF_SECT_DIF 7 /* uint32_t array of byte code */ +#define DOF_SECT_STRTAB 8 /* string table */ +#define DOF_SECT_VARTAB 9 /* dtrace_difv_t array */ +#define DOF_SECT_RELTAB 10 /* dof_relodesc_t array */ +#define DOF_SECT_TYPTAB 11 /* dtrace_diftype_t array */ +#define DOF_SECT_URELHDR 12 /* dof_relohdr_t (user relocations) */ +#define DOF_SECT_KRELHDR 13 /* dof_relohdr_t (kernel relocations) */ +#define DOF_SECT_OPTDESC 14 /* dof_optdesc_t array */ +#define DOF_SECT_PROVIDER 15 /* dof_provider_t */ +#define DOF_SECT_PROBES 16 /* dof_probe_t array */ +#define DOF_SECT_PRARGS 17 /* uint8_t array (probe arg mappings) */ +#define DOF_SECT_PROFFS 18 /* uint32_t array (probe arg offsets) */ +#define DOF_SECT_INTTAB 19 /* uint64_t array */ +#define DOF_SECT_UTSNAME 20 /* struct utsname */ +#define DOF_SECT_XLTAB 21 /* dof_xlref_t array */ +#define DOF_SECT_XLMEMBERS 22 /* dof_xlmember_t array */ +#define DOF_SECT_XLIMPORT 23 /* dof_xlator_t */ +#define DOF_SECT_XLEXPORT 24 /* dof_xlator_t */ +#define DOF_SECT_PREXPORT 25 /* dof_secidx_t array (exported objs) */ +#define DOF_SECT_PRENOFFS 26 /* uint32_t array (enabled offsets) */ + +#define DOF_SECF_LOAD 1 /* section should be loaded */ + +typedef struct dof_ecbdesc { + dof_secidx_t dofe_probes; /* link to DOF_SECT_PROBEDESC */ + dof_secidx_t dofe_pred; /* link to DOF_SECT_DIFOHDR */ + dof_secidx_t dofe_actions; /* link to DOF_SECT_ACTDESC */ + uint32_t dofe_pad; /* reserved for future use */ + uint64_t dofe_uarg; /* user-supplied library argument */ +} dof_ecbdesc_t; + +typedef struct dof_probedesc { + dof_secidx_t dofp_strtab; /* link to DOF_SECT_STRTAB section */ + dof_stridx_t dofp_provider; /* provider string */ + dof_stridx_t dofp_mod; /* module string */ + dof_stridx_t dofp_func; /* function string */ + dof_stridx_t dofp_name; /* name string */ + uint32_t dofp_id; /* probe identifier (or zero) */ +} dof_probedesc_t; + +typedef struct dof_actdesc { + dof_secidx_t dofa_difo; /* link to DOF_SECT_DIFOHDR */ + dof_secidx_t dofa_strtab; /* link to DOF_SECT_STRTAB section */ + uint32_t dofa_kind; /* action kind (DTRACEACT_* constant) */ + uint32_t dofa_ntuple; /* number of subsequent tuple actions */ + uint64_t dofa_arg; /* kind-specific argument */ + uint64_t dofa_uarg; /* user-supplied argument */ +} dof_actdesc_t; + +typedef struct dof_difohdr { + dtrace_diftype_t dofd_rtype; /* return type for this fragment */ + dof_secidx_t dofd_links[1]; /* variable length array of indices */ +} dof_difohdr_t; + +typedef struct dof_relohdr { + dof_secidx_t dofr_strtab; /* link to DOF_SECT_STRTAB for names */ + dof_secidx_t dofr_relsec; /* link to DOF_SECT_RELTAB for relos */ + dof_secidx_t dofr_tgtsec; /* link to section we are relocating */ +} dof_relohdr_t; + +typedef struct dof_relodesc { + dof_stridx_t dofr_name; /* string name of relocation symbol */ + uint32_t dofr_type; /* relo type (DOF_RELO_* constant) */ + uint64_t dofr_offset; /* byte offset for relocation */ + uint64_t dofr_data; /* additional type-specific data */ +} dof_relodesc_t; + +#define DOF_RELO_NONE 0 /* empty relocation entry */ +#define DOF_RELO_SETX 1 /* relocate setx value */ + +typedef struct dof_optdesc { + uint32_t dofo_option; /* option identifier */ + dof_secidx_t dofo_strtab; /* string table, if string option */ + uint64_t dofo_value; /* option value or string index */ +} dof_optdesc_t; + +typedef uint32_t dof_attr_t; /* encoded stability attributes */ + +#define DOF_ATTR(n, d, c) (((n) << 24) | ((d) << 16) | ((c) << 8)) +#define DOF_ATTR_NAME(a) (((a) >> 24) & 0xff) +#define DOF_ATTR_DATA(a) (((a) >> 16) & 0xff) +#define DOF_ATTR_CLASS(a) (((a) >> 8) & 0xff) + +typedef struct dof_provider { + dof_secidx_t dofpv_strtab; /* link to DOF_SECT_STRTAB section */ + dof_secidx_t dofpv_probes; /* link to DOF_SECT_PROBES section */ + dof_secidx_t dofpv_prargs; /* link to DOF_SECT_PRARGS section */ + dof_secidx_t dofpv_proffs; /* link to DOF_SECT_PROFFS section */ + dof_stridx_t dofpv_name; /* provider name string */ + dof_attr_t dofpv_provattr; /* provider attributes */ + dof_attr_t dofpv_modattr; /* module attributes */ + dof_attr_t dofpv_funcattr; /* function attributes */ + dof_attr_t dofpv_nameattr; /* name attributes */ + dof_attr_t dofpv_argsattr; /* args attributes */ + dof_secidx_t dofpv_prenoffs; /* link to DOF_SECT_PRENOFFS section */ +} dof_provider_t; + +typedef struct dof_probe { + uint64_t dofpr_addr; /* probe base address or offset */ + dof_stridx_t dofpr_func; /* probe function string */ + dof_stridx_t dofpr_name; /* probe name string */ + dof_stridx_t dofpr_nargv; /* native argument type strings */ + dof_stridx_t dofpr_xargv; /* translated argument type strings */ + uint32_t dofpr_argidx; /* index of first argument mapping */ + uint32_t dofpr_offidx; /* index of first offset entry */ + uint8_t dofpr_nargc; /* native argument count */ + uint8_t dofpr_xargc; /* translated argument count */ + uint16_t dofpr_noffs; /* number of offset entries for probe */ + uint32_t dofpr_enoffidx; /* index of first is-enabled offset */ + uint16_t dofpr_nenoffs; /* number of is-enabled offsets */ + uint16_t dofpr_pad1; /* reserved for future use */ + uint32_t dofpr_pad2; /* reserved for future use */ +} dof_probe_t; + +typedef struct dof_xlator { + dof_secidx_t dofxl_members; /* link to DOF_SECT_XLMEMBERS section */ + dof_secidx_t dofxl_strtab; /* link to DOF_SECT_STRTAB section */ + dof_stridx_t dofxl_argv; /* input parameter type strings */ + uint32_t dofxl_argc; /* input parameter list length */ + dof_stridx_t dofxl_type; /* output type string name */ + dof_attr_t dofxl_attr; /* output stability attributes */ +} dof_xlator_t; + +typedef struct dof_xlmember { + dof_secidx_t dofxm_difo; /* member link to DOF_SECT_DIFOHDR */ + dof_stridx_t dofxm_name; /* member name */ + dtrace_diftype_t dofxm_type; /* member type */ +} dof_xlmember_t; + +typedef struct dof_xlref { + dof_secidx_t dofxr_xlator; /* link to DOF_SECT_XLATORS section */ + uint32_t dofxr_member; /* index of referenced dof_xlmember */ + uint32_t dofxr_argn; /* index of argument for DIF_OP_XLARG */ +} dof_xlref_t; + +/* + * DTrace Intermediate Format Object (DIFO) + * + * A DIFO is used to store the compiled DIF for a D expression, its return + * type, and its string and variable tables. The string table is a single + * buffer of character data into which sets instructions and variable + * references can reference strings using a byte offset. The variable table + * is an array of dtrace_difv_t structures that describe the name and type of + * each variable and the id used in the DIF code. This structure is described + * above in the DIF section of this header file. The DIFO is used at both + * user-level (in the library) and in the kernel, but the structure is never + * passed between the two: the DOF structures form the only interface. As a + * result, the definition can change depending on the presence of _KERNEL. + */ +typedef struct dtrace_difo { + dif_instr_t *dtdo_buf; /* instruction buffer */ + uint64_t *dtdo_inttab; /* integer table (optional) */ + char *dtdo_strtab; /* string table (optional) */ + dtrace_difv_t *dtdo_vartab; /* variable table (optional) */ + uint_t dtdo_len; /* length of instruction buffer */ + uint_t dtdo_intlen; /* length of integer table */ + uint_t dtdo_strlen; /* length of string table */ + uint_t dtdo_varlen; /* length of variable table */ + dtrace_diftype_t dtdo_rtype; /* return type */ + uint_t dtdo_refcnt; /* owner reference count */ + uint_t dtdo_destructive; /* invokes destructive subroutines */ +#ifndef _KERNEL + dof_relodesc_t *dtdo_kreltab; /* kernel relocations */ + dof_relodesc_t *dtdo_ureltab; /* user relocations */ + struct dt_node **dtdo_xlmtab; /* translator references */ + uint_t dtdo_krelen; /* length of krelo table */ + uint_t dtdo_urelen; /* length of urelo table */ + uint_t dtdo_xlmlen; /* length of translator table */ +#endif +} dtrace_difo_t; + +/* + * DTrace Enabling Description Structures + * + * When DTrace is tracking the description of a DTrace enabling entity (probe, + * predicate, action, ECB, record, etc.), it does so in a description + * structure. These structures all end in "desc", and are used at both + * user-level and in the kernel -- but (with the exception of + * dtrace_probedesc_t) they are never passed between them. Typically, + * user-level will use the description structures when assembling an enabling. + * It will then distill those description structures into a DOF object (see + * above), and send it into the kernel. The kernel will again use the + * description structures to create a description of the enabling as it reads + * the DOF. When the description is complete, the enabling will be actually + * created -- turning it into the structures that represent the enabling + * instead of merely describing it. Not surprisingly, the description + * structures bear a strong resemblance to the DOF structures that act as their + * conduit. + */ +struct dtrace_predicate; + +typedef struct dtrace_probedesc { + dtrace_id_t dtpd_id; /* probe identifier */ + char dtpd_provider[DTRACE_PROVNAMELEN]; /* probe provider name */ + char dtpd_mod[DTRACE_MODNAMELEN]; /* probe module name */ + char dtpd_func[DTRACE_FUNCNAMELEN]; /* probe function name */ + char dtpd_name[DTRACE_NAMELEN]; /* probe name */ +} dtrace_probedesc_t; + +typedef struct dtrace_repldesc { + dtrace_probedesc_t dtrpd_match; /* probe descr. to match */ + dtrace_probedesc_t dtrpd_create; /* probe descr. to create */ +} dtrace_repldesc_t; + +typedef struct dtrace_preddesc { + dtrace_difo_t *dtpdd_difo; /* pointer to DIF object */ + struct dtrace_predicate *dtpdd_predicate; /* pointer to predicate */ +} dtrace_preddesc_t; + +typedef struct dtrace_actdesc { + dtrace_difo_t *dtad_difo; /* pointer to DIF object */ + struct dtrace_actdesc *dtad_next; /* next action */ + dtrace_actkind_t dtad_kind; /* kind of action */ + uint32_t dtad_ntuple; /* number in tuple */ + uint64_t dtad_arg; /* action argument */ + uint64_t dtad_uarg; /* user argument */ + int dtad_refcnt; /* reference count */ +} dtrace_actdesc_t; + +typedef struct dtrace_ecbdesc { + dtrace_actdesc_t *dted_action; /* action description(s) */ + dtrace_preddesc_t dted_pred; /* predicate description */ + dtrace_probedesc_t dted_probe; /* probe description */ + uint64_t dted_uarg; /* library argument */ + int dted_refcnt; /* reference count */ +} dtrace_ecbdesc_t; + +/* + * DTrace Metadata Description Structures + * + * DTrace separates the trace data stream from the metadata stream. The only + * metadata tokens placed in the data stream are enabled probe identifiers + * (EPIDs) or (in the case of aggregations) aggregation identifiers. In order + * to determine the structure of the data, DTrace consumers pass the token to + * the kernel, and receive in return a corresponding description of the enabled + * probe (via the dtrace_eprobedesc structure) or the aggregation (via the + * dtrace_aggdesc structure). Both of these structures are expressed in terms + * of record descriptions (via the dtrace_recdesc structure) that describe the + * exact structure of the data. Some record descriptions may also contain a + * format identifier; this additional bit of metadata can be retrieved from the + * kernel, for which a format description is returned via the dtrace_fmtdesc + * structure. Note that all four of these structures must be bitness-neutral + * to allow for a 32-bit DTrace consumer on a 64-bit kernel. + */ +typedef struct dtrace_recdesc { + dtrace_actkind_t dtrd_action; /* kind of action */ + uint32_t dtrd_size; /* size of record */ + uint32_t dtrd_offset; /* offset in ECB's data */ + uint16_t dtrd_alignment; /* required alignment */ + uint16_t dtrd_format; /* format, if any */ + uint64_t dtrd_arg; /* action argument */ + uint64_t dtrd_uarg; /* user argument */ +} dtrace_recdesc_t; + +typedef struct dtrace_eprobedesc { + dtrace_epid_t dtepd_epid; /* enabled probe ID */ + dtrace_id_t dtepd_probeid; /* probe ID */ + uint64_t dtepd_uarg; /* library argument */ + uint32_t dtepd_size; /* total size */ + int dtepd_nrecs; /* number of records */ + dtrace_recdesc_t dtepd_rec[1]; /* records themselves */ +} dtrace_eprobedesc_t; + +typedef struct dtrace_aggdesc { + DTRACE_PTR(char, dtagd_name); /* not filled in by kernel */ + dtrace_aggvarid_t dtagd_varid; /* not filled in by kernel */ + int dtagd_flags; /* not filled in by kernel */ + dtrace_aggid_t dtagd_id; /* aggregation ID */ + dtrace_epid_t dtagd_epid; /* enabled probe ID */ + uint32_t dtagd_size; /* size in bytes */ + int dtagd_nrecs; /* number of records */ + uint32_t dtagd_pad; /* explicit padding */ + dtrace_recdesc_t dtagd_rec[1]; /* record descriptions */ +} dtrace_aggdesc_t; + +typedef struct dtrace_fmtdesc { + DTRACE_PTR(char, dtfd_string); /* format string */ + int dtfd_length; /* length of format string */ + uint16_t dtfd_format; /* format identifier */ +} dtrace_fmtdesc_t; + +#define DTRACE_SIZEOF_EPROBEDESC(desc) \ + (sizeof (dtrace_eprobedesc_t) + ((desc)->dtepd_nrecs ? \ + (((desc)->dtepd_nrecs - 1) * sizeof (dtrace_recdesc_t)) : 0)) + +#define DTRACE_SIZEOF_AGGDESC(desc) \ + (sizeof (dtrace_aggdesc_t) + ((desc)->dtagd_nrecs ? \ + (((desc)->dtagd_nrecs - 1) * sizeof (dtrace_recdesc_t)) : 0)) + +/* + * DTrace Option Interface + * + * Run-time DTrace options are set and retrieved via DOF_SECT_OPTDESC sections + * in a DOF image. The dof_optdesc structure contains an option identifier and + * an option value. The valid option identifiers are found below; the mapping + * between option identifiers and option identifying strings is maintained at + * user-level. Note that the value of DTRACEOPT_UNSET is such that all of the + * following are potentially valid option values: all positive integers, zero + * and negative one. Some options (notably "bufpolicy" and "bufresize") take + * predefined tokens as their values; these are defined with + * DTRACEOPT_{option}_{token}. + */ +#define DTRACEOPT_BUFSIZE 0 /* buffer size */ +#define DTRACEOPT_BUFPOLICY 1 /* buffer policy */ +#define DTRACEOPT_DYNVARSIZE 2 /* dynamic variable size */ +#define DTRACEOPT_AGGSIZE 3 /* aggregation size */ +#define DTRACEOPT_SPECSIZE 4 /* speculation size */ +#define DTRACEOPT_NSPEC 5 /* number of speculations */ +#define DTRACEOPT_STRSIZE 6 /* string size */ +#define DTRACEOPT_CLEANRATE 7 /* dynvar cleaning rate */ +#define DTRACEOPT_CPU 8 /* CPU to trace */ +#define DTRACEOPT_BUFRESIZE 9 /* buffer resizing policy */ +#define DTRACEOPT_GRABANON 10 /* grab anonymous state, if any */ +#define DTRACEOPT_FLOWINDENT 11 /* indent function entry/return */ +#define DTRACEOPT_QUIET 12 /* only output explicitly traced data */ +#define DTRACEOPT_STACKFRAMES 13 /* number of stack frames */ +#define DTRACEOPT_USTACKFRAMES 14 /* number of user stack frames */ +#define DTRACEOPT_AGGRATE 15 /* aggregation snapshot rate */ +#define DTRACEOPT_SWITCHRATE 16 /* buffer switching rate */ +#define DTRACEOPT_STATUSRATE 17 /* status rate */ +#define DTRACEOPT_DESTRUCTIVE 18 /* destructive actions allowed */ +#define DTRACEOPT_STACKINDENT 19 /* output indent for stack traces */ +#define DTRACEOPT_RAWBYTES 20 /* always print bytes in raw form */ +#define DTRACEOPT_JSTACKFRAMES 21 /* number of jstack() frames */ +#define DTRACEOPT_JSTACKSTRSIZE 22 /* size of jstack() string table */ +#define DTRACEOPT_AGGSORTKEY 23 /* sort aggregations by key */ +#define DTRACEOPT_AGGSORTREV 24 /* reverse-sort aggregations */ +#define DTRACEOPT_AGGSORTPOS 25 /* agg. position to sort on */ +#define DTRACEOPT_AGGSORTKEYPOS 26 /* agg. key position to sort on */ +#define DTRACEOPT_MAX 27 /* number of options */ + +#define DTRACEOPT_UNSET (dtrace_optval_t)-2 /* unset option */ + +#define DTRACEOPT_BUFPOLICY_RING 0 /* ring buffer */ +#define DTRACEOPT_BUFPOLICY_FILL 1 /* fill buffer, then stop */ +#define DTRACEOPT_BUFPOLICY_SWITCH 2 /* switch buffers */ + +#define DTRACEOPT_BUFRESIZE_AUTO 0 /* automatic resizing */ +#define DTRACEOPT_BUFRESIZE_MANUAL 1 /* manual resizing */ + +/* + * DTrace Buffer Interface + * + * In order to get a snapshot of the principal or aggregation buffer, + * user-level passes a buffer description to the kernel with the dtrace_bufdesc + * structure. This describes which CPU user-level is interested in, and + * where user-level wishes the kernel to snapshot the buffer to (the + * dtbd_data field). The kernel uses the same structure to pass back some + * information regarding the buffer: the size of data actually copied out, the + * number of drops, the number of errors, and the offset of the oldest record. + * If the buffer policy is a "switch" policy, taking a snapshot of the + * principal buffer has the additional effect of switching the active and + * inactive buffers. Taking a snapshot of the aggregation buffer _always_ has + * the additional effect of switching the active and inactive buffers. + */ +typedef struct dtrace_bufdesc { + uint64_t dtbd_size; /* size of buffer */ + uint32_t dtbd_cpu; /* CPU or DTRACE_CPUALL */ + uint32_t dtbd_errors; /* number of errors */ + uint64_t dtbd_drops; /* number of drops */ + DTRACE_PTR(char, dtbd_data); /* data */ + uint64_t dtbd_oldest; /* offset of oldest record */ +} dtrace_bufdesc_t; + +/* + * DTrace Status + * + * The status of DTrace is relayed via the dtrace_status structure. This + * structure contains members to count drops other than the capacity drops + * available via the buffer interface (see above). This consists of dynamic + * drops (including capacity dynamic drops, rinsing drops and dirty drops), and + * speculative drops (including capacity speculative drops, drops due to busy + * speculative buffers and drops due to unavailable speculative buffers). + * Additionally, the status structure contains a field to indicate the number + * of "fill"-policy buffers have been filled and a boolean field to indicate + * that exit() has been called. If the dtst_exiting field is non-zero, no + * further data will be generated until tracing is stopped (at which time any + * enablings of the END action will be processed); if user-level sees that + * this field is non-zero, tracing should be stopped as soon as possible. + */ +typedef struct dtrace_status { + uint64_t dtst_dyndrops; /* dynamic drops */ + uint64_t dtst_dyndrops_rinsing; /* dyn drops due to rinsing */ + uint64_t dtst_dyndrops_dirty; /* dyn drops due to dirty */ + uint64_t dtst_specdrops; /* speculative drops */ + uint64_t dtst_specdrops_busy; /* spec drops due to busy */ + uint64_t dtst_specdrops_unavail; /* spec drops due to unavail */ + uint64_t dtst_errors; /* total errors */ + uint64_t dtst_filled; /* number of filled bufs */ + uint64_t dtst_stkstroverflows; /* stack string tab overflows */ + uint64_t dtst_dblerrors; /* errors in ERROR probes */ + char dtst_killed; /* non-zero if killed */ + char dtst_exiting; /* non-zero if exit() called */ + char dtst_pad[6]; /* pad out to 64-bit align */ +} dtrace_status_t; + +/* + * DTrace Configuration + * + * User-level may need to understand some elements of the kernel DTrace + * configuration in order to generate correct DIF. This information is + * conveyed via the dtrace_conf structure. + */ +typedef struct dtrace_conf { + uint_t dtc_difversion; /* supported DIF version */ + uint_t dtc_difintregs; /* # of DIF integer registers */ + uint_t dtc_diftupregs; /* # of DIF tuple registers */ + uint_t dtc_ctfmodel; /* CTF data model */ + uint_t dtc_pad[8]; /* reserved for future use */ +} dtrace_conf_t; + +/* + * DTrace Faults + * + * The constants below DTRACEFLT_LIBRARY indicate probe processing faults; + * constants at or above DTRACEFLT_LIBRARY indicate faults in probe + * postprocessing at user-level. Probe processing faults induce an ERROR + * probe and are replicated in unistd.d to allow users' ERROR probes to decode + * the error condition using thse symbolic labels. + */ +#define DTRACEFLT_UNKNOWN 0 /* Unknown fault */ +#define DTRACEFLT_BADADDR 1 /* Bad address */ +#define DTRACEFLT_BADALIGN 2 /* Bad alignment */ +#define DTRACEFLT_ILLOP 3 /* Illegal operation */ +#define DTRACEFLT_DIVZERO 4 /* Divide-by-zero */ +#define DTRACEFLT_NOSCRATCH 5 /* Out of scratch space */ +#define DTRACEFLT_KPRIV 6 /* Illegal kernel access */ +#define DTRACEFLT_UPRIV 7 /* Illegal user access */ +#define DTRACEFLT_TUPOFLOW 8 /* Tuple stack overflow */ +#define DTRACEFLT_BADSTACK 9 /* Bad stack */ + +#define DTRACEFLT_LIBRARY 1000 /* Library-level fault */ + +/* + * DTrace Argument Types + * + * Because it would waste both space and time, argument types do not reside + * with the probe. In order to determine argument types for args[X] + * variables, the D compiler queries for argument types on a probe-by-probe + * basis. (This optimizes for the common case that arguments are either not + * used or used in an untyped fashion.) Typed arguments are specified with a + * string of the type name in the dtragd_native member of the argument + * description structure. Typed arguments may be further translated to types + * of greater stability; the provider indicates such a translated argument by + * filling in the dtargd_xlate member with the string of the translated type. + * Finally, the provider may indicate which argument value a given argument + * maps to by setting the dtargd_mapping member -- allowing a single argument + * to map to multiple args[X] variables. + */ +typedef struct dtrace_argdesc { + dtrace_id_t dtargd_id; /* probe identifier */ + int dtargd_ndx; /* arg number (-1 iff none) */ + int dtargd_mapping; /* value mapping */ + char dtargd_native[DTRACE_ARGTYPELEN]; /* native type name */ + char dtargd_xlate[DTRACE_ARGTYPELEN]; /* translated type name */ +} dtrace_argdesc_t; + +/* + * DTrace Stability Attributes + * + * Each DTrace provider advertises the name and data stability of each of its + * probe description components, as well as its architectural dependencies. + * The D compiler can query the provider attributes (dtrace_pattr_t below) in + * order to compute the properties of an input program and report them. + */ +typedef uint8_t dtrace_stability_t; /* stability code (see attributes(5)) */ +typedef uint8_t dtrace_class_t; /* architectural dependency class */ + +#define DTRACE_STABILITY_INTERNAL 0 /* private to DTrace itself */ +#define DTRACE_STABILITY_PRIVATE 1 /* private to Sun (see docs) */ +#define DTRACE_STABILITY_OBSOLETE 2 /* scheduled for removal */ +#define DTRACE_STABILITY_EXTERNAL 3 /* not controlled by Sun */ +#define DTRACE_STABILITY_UNSTABLE 4 /* new or rapidly changing */ +#define DTRACE_STABILITY_EVOLVING 5 /* less rapidly changing */ +#define DTRACE_STABILITY_STABLE 6 /* mature interface from Sun */ +#define DTRACE_STABILITY_STANDARD 7 /* industry standard */ +#define DTRACE_STABILITY_MAX 7 /* maximum valid stability */ + +#define DTRACE_CLASS_UNKNOWN 0 /* unknown architectural dependency */ +#define DTRACE_CLASS_CPU 1 /* CPU-module-specific */ +#define DTRACE_CLASS_PLATFORM 2 /* platform-specific (uname -i) */ +#define DTRACE_CLASS_GROUP 3 /* hardware-group-specific (uname -m) */ +#define DTRACE_CLASS_ISA 4 /* ISA-specific (uname -p) */ +#define DTRACE_CLASS_COMMON 5 /* common to all systems */ +#define DTRACE_CLASS_MAX 5 /* maximum valid class */ + +#define DTRACE_PRIV_NONE 0x0000 +#define DTRACE_PRIV_KERNEL 0x0001 +#define DTRACE_PRIV_USER 0x0002 +#define DTRACE_PRIV_PROC 0x0004 +#define DTRACE_PRIV_OWNER 0x0008 +#define DTRACE_PRIV_ZONEOWNER 0x0010 + +#define DTRACE_PRIV_ALL \ + (DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER | \ + DTRACE_PRIV_PROC | DTRACE_PRIV_OWNER | DTRACE_PRIV_ZONEOWNER) + +typedef struct dtrace_ppriv { + uint32_t dtpp_flags; /* privilege flags */ + uid_t dtpp_uid; /* user ID */ + zoneid_t dtpp_zoneid; /* zone ID */ +} dtrace_ppriv_t; + +typedef struct dtrace_attribute { + dtrace_stability_t dtat_name; /* entity name stability */ + dtrace_stability_t dtat_data; /* entity data stability */ + dtrace_class_t dtat_class; /* entity data dependency */ +} dtrace_attribute_t; + +typedef struct dtrace_pattr { + dtrace_attribute_t dtpa_provider; /* provider attributes */ + dtrace_attribute_t dtpa_mod; /* module attributes */ + dtrace_attribute_t dtpa_func; /* function attributes */ + dtrace_attribute_t dtpa_name; /* name attributes */ + dtrace_attribute_t dtpa_args; /* args[] attributes */ +} dtrace_pattr_t; + +typedef struct dtrace_providerdesc { + char dtvd_name[DTRACE_PROVNAMELEN]; /* provider name */ + dtrace_pattr_t dtvd_attr; /* stability attributes */ + dtrace_ppriv_t dtvd_priv; /* privileges required */ +} dtrace_providerdesc_t; + +/* + * DTrace Pseudodevice Interface + * + * DTrace is controlled through ioctl(2)'s to the in-kernel dtrace:dtrace + * pseudodevice driver. These ioctls comprise the user-kernel interface to + * DTrace. + */ +#if defined(sun) +#define DTRACEIOC (('d' << 24) | ('t' << 16) | ('r' << 8)) +#define DTRACEIOC_PROVIDER (DTRACEIOC | 1) /* provider query */ +#define DTRACEIOC_PROBES (DTRACEIOC | 2) /* probe query */ +#define DTRACEIOC_BUFSNAP (DTRACEIOC | 4) /* snapshot buffer */ +#define DTRACEIOC_PROBEMATCH (DTRACEIOC | 5) /* match probes */ +#define DTRACEIOC_ENABLE (DTRACEIOC | 6) /* enable probes */ +#define DTRACEIOC_AGGSNAP (DTRACEIOC | 7) /* snapshot agg. */ +#define DTRACEIOC_EPROBE (DTRACEIOC | 8) /* get eprobe desc. */ +#define DTRACEIOC_PROBEARG (DTRACEIOC | 9) /* get probe arg */ +#define DTRACEIOC_CONF (DTRACEIOC | 10) /* get config. */ +#define DTRACEIOC_STATUS (DTRACEIOC | 11) /* get status */ +#define DTRACEIOC_GO (DTRACEIOC | 12) /* start tracing */ +#define DTRACEIOC_STOP (DTRACEIOC | 13) /* stop tracing */ +#define DTRACEIOC_AGGDESC (DTRACEIOC | 15) /* get agg. desc. */ +#define DTRACEIOC_FORMAT (DTRACEIOC | 16) /* get format str */ +#define DTRACEIOC_DOFGET (DTRACEIOC | 17) /* get DOF */ +#define DTRACEIOC_REPLICATE (DTRACEIOC | 18) /* replicate enab */ +#else +#define DTRACEIOC_PROVIDER _IOWR('x',1,dtrace_providerdesc_t) + /* provider query */ +#define DTRACEIOC_PROBES _IOWR('x',2,dtrace_probedesc_t) + /* probe query */ +#define DTRACEIOC_BUFSNAP _IOW('x',4,dtrace_bufdesc_t *) + /* snapshot buffer */ +#define DTRACEIOC_PROBEMATCH _IOWR('x',5,dtrace_probedesc_t) + /* match probes */ +typedef struct { + void *dof; /* DOF userland address written to driver. */ + int n_matched; /* # matches returned by driver. */ +} dtrace_enable_io_t; +#define DTRACEIOC_ENABLE _IOWR('x',6,dtrace_enable_io_t) + /* enable probes */ +#define DTRACEIOC_AGGSNAP _IOW('x',7,dtrace_bufdesc_t *) + /* snapshot agg. */ +#define DTRACEIOC_EPROBE _IOW('x',8,dtrace_eprobedesc_t) + /* get eprobe desc. */ +#define DTRACEIOC_PROBEARG _IOWR('x',9,dtrace_argdesc_t) + /* get probe arg */ +#define DTRACEIOC_CONF _IOR('x',10,dtrace_conf_t) + /* get config. */ +#define DTRACEIOC_STATUS _IOR('x',11,dtrace_status_t) + /* get status */ +#define DTRACEIOC_GO _IOR('x',12,processorid_t) + /* start tracing */ +#define DTRACEIOC_STOP _IOWR('x',13,processorid_t) + /* stop tracing */ +#define DTRACEIOC_AGGDESC _IOW('x',15,dtrace_aggdesc_t *) + /* get agg. desc. */ +#define DTRACEIOC_FORMAT _IOWR('x',16,dtrace_fmtdesc_t) + /* get format str */ +#define DTRACEIOC_DOFGET _IOW('x',17,dof_hdr_t *) + /* get DOF */ +#define DTRACEIOC_REPLICATE _IOW('x',18,dtrace_repldesc_t) + /* replicate enab */ +#endif + +/* + * DTrace Helpers + * + * In general, DTrace establishes probes in processes and takes actions on + * processes without knowing their specific user-level structures. Instead of + * existing in the framework, process-specific knowledge is contained by the + * enabling D program -- which can apply process-specific knowledge by making + * appropriate use of DTrace primitives like copyin() and copyinstr() to + * operate on user-level data. However, there may exist some specific probes + * of particular semantic relevance that the application developer may wish to + * explicitly export. For example, an application may wish to export a probe + * at the point that it begins and ends certain well-defined transactions. In + * addition to providing probes, programs may wish to offer assistance for + * certain actions. For example, in highly dynamic environments (e.g., Java), + * it may be difficult to obtain a stack trace in terms of meaningful symbol + * names (the translation from instruction addresses to corresponding symbol + * names may only be possible in situ); these environments may wish to define + * a series of actions to be applied in situ to obtain a meaningful stack + * trace. + * + * These two mechanisms -- user-level statically defined tracing and assisting + * DTrace actions -- are provided via DTrace _helpers_. Helpers are specified + * via DOF, but unlike enabling DOF, helper DOF may contain definitions of + * providers, probes and their arguments. If a helper wishes to provide + * action assistance, probe descriptions and corresponding DIF actions may be + * specified in the helper DOF. For such helper actions, however, the probe + * description describes the specific helper: all DTrace helpers have the + * provider name "dtrace" and the module name "helper", and the name of the + * helper is contained in the function name (for example, the ustack() helper + * is named "ustack"). Any helper-specific name may be contained in the name + * (for example, if a helper were to have a constructor, it might be named + * "dtrace:helper:<helper>:init"). Helper actions are only called when the + * action that they are helping is taken. Helper actions may only return DIF + * expressions, and may only call the following subroutines: + * + * alloca() <= Allocates memory out of the consumer's scratch space + * bcopy() <= Copies memory to scratch space + * copyin() <= Copies memory from user-level into consumer's scratch + * copyinto() <= Copies memory into a specific location in scratch + * copyinstr() <= Copies a string into a specific location in scratch + * + * Helper actions may only access the following built-in variables: + * + * curthread <= Current kthread_t pointer + * tid <= Current thread identifier + * pid <= Current process identifier + * ppid <= Parent process identifier + * uid <= Current user ID + * gid <= Current group ID + * execname <= Current executable name + * zonename <= Current zone name + * + * Helper actions may not manipulate or allocate dynamic variables, but they + * may have clause-local and statically-allocated global variables. The + * helper action variable state is specific to the helper action -- variables + * used by the helper action may not be accessed outside of the helper + * action, and the helper action may not access variables that like outside + * of it. Helper actions may not load from kernel memory at-large; they are + * restricting to loading current user state (via copyin() and variants) and + * scratch space. As with probe enablings, helper actions are executed in + * program order. The result of the helper action is the result of the last + * executing helper expression. + * + * Helpers -- composed of either providers/probes or probes/actions (or both) + * -- are added by opening the "helper" minor node, and issuing an ioctl(2) + * (DTRACEHIOC_ADDDOF) that specifies the dof_helper_t structure. This + * encapsulates the name and base address of the user-level library or + * executable publishing the helpers and probes as well as the DOF that + * contains the definitions of those helpers and probes. + * + * The DTRACEHIOC_ADD and DTRACEHIOC_REMOVE are left in place for legacy + * helpers and should no longer be used. No other ioctls are valid on the + * helper minor node. + */ +#define DTRACEHIOC (('d' << 24) | ('t' << 16) | ('h' << 8)) +#define DTRACEHIOC_ADD (DTRACEHIOC | 1) /* add helper */ +#define DTRACEHIOC_REMOVE (DTRACEHIOC | 2) /* remove helper */ +#define DTRACEHIOC_ADDDOF (DTRACEHIOC | 3) /* add helper DOF */ + +typedef struct dof_helper { + char dofhp_mod[DTRACE_MODNAMELEN]; /* executable or library name */ + uint64_t dofhp_addr; /* base address of object */ + uint64_t dofhp_dof; /* address of helper DOF */ +} dof_helper_t; + +#define DTRACEMNR_DTRACE "dtrace" /* node for DTrace ops */ +#define DTRACEMNR_HELPER "helper" /* node for helpers */ +#define DTRACEMNRN_DTRACE 0 /* minor for DTrace ops */ +#define DTRACEMNRN_HELPER 1 /* minor for helpers */ +#define DTRACEMNRN_CLONE 2 /* first clone minor */ + +#ifdef _KERNEL + +/* + * DTrace Provider API + * + * The following functions are implemented by the DTrace framework and are + * used to implement separate in-kernel DTrace providers. Common functions + * are provided in uts/common/os/dtrace.c. ISA-dependent subroutines are + * defined in uts/<isa>/dtrace/dtrace_asm.s or uts/<isa>/dtrace/dtrace_isa.c. + * + * The provider API has two halves: the API that the providers consume from + * DTrace, and the API that providers make available to DTrace. + * + * 1 Framework-to-Provider API + * + * 1.1 Overview + * + * The Framework-to-Provider API is represented by the dtrace_pops structure + * that the provider passes to the framework when registering itself. This + * structure consists of the following members: + * + * dtps_provide() <-- Provide all probes, all modules + * dtps_provide_module() <-- Provide all probes in specified module + * dtps_enable() <-- Enable specified probe + * dtps_disable() <-- Disable specified probe + * dtps_suspend() <-- Suspend specified probe + * dtps_resume() <-- Resume specified probe + * dtps_getargdesc() <-- Get the argument description for args[X] + * dtps_getargval() <-- Get the value for an argX or args[X] variable + * dtps_usermode() <-- Find out if the probe was fired in user mode + * dtps_destroy() <-- Destroy all state associated with this probe + * + * 1.2 void dtps_provide(void *arg, const dtrace_probedesc_t *spec) + * + * 1.2.1 Overview + * + * Called to indicate that the provider should provide all probes. If the + * specified description is non-NULL, dtps_provide() is being called because + * no probe matched a specified probe -- if the provider has the ability to + * create custom probes, it may wish to create a probe that matches the + * specified description. + * + * 1.2.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is a pointer to a probe description that the provider may + * wish to consider when creating custom probes. The provider is expected to + * call back into the DTrace framework via dtrace_probe_create() to create + * any necessary probes. dtps_provide() may be called even if the provider + * has made available all probes; the provider should check the return value + * of dtrace_probe_create() to handle this case. Note that the provider need + * not implement both dtps_provide() and dtps_provide_module(); see + * "Arguments and Notes" for dtrace_register(), below. + * + * 1.2.3 Return value + * + * None. + * + * 1.2.4 Caller's context + * + * dtps_provide() is typically called from open() or ioctl() context, but may + * be called from other contexts as well. The DTrace framework is locked in + * such a way that providers may not register or unregister. This means that + * the provider may not call any DTrace API that affects its registration with + * the framework, including dtrace_register(), dtrace_unregister(), + * dtrace_invalidate(), and dtrace_condense(). However, the context is such + * that the provider may (and indeed, is expected to) call probe-related + * DTrace routines, including dtrace_probe_create(), dtrace_probe_lookup(), + * and dtrace_probe_arg(). + * + * 1.3 void dtps_provide_module(void *arg, modctl_t *mp) + * + * 1.3.1 Overview + * + * Called to indicate that the provider should provide all probes in the + * specified module. + * + * 1.3.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is a pointer to a modctl structure that indicates the + * module for which probes should be created. + * + * 1.3.3 Return value + * + * None. + * + * 1.3.4 Caller's context + * + * dtps_provide_module() may be called from open() or ioctl() context, but + * may also be called from a module loading context. mod_lock is held, and + * the DTrace framework is locked in such a way that providers may not + * register or unregister. This means that the provider may not call any + * DTrace API that affects its registration with the framework, including + * dtrace_register(), dtrace_unregister(), dtrace_invalidate(), and + * dtrace_condense(). However, the context is such that the provider may (and + * indeed, is expected to) call probe-related DTrace routines, including + * dtrace_probe_create(), dtrace_probe_lookup(), and dtrace_probe_arg(). Note + * that the provider need not implement both dtps_provide() and + * dtps_provide_module(); see "Arguments and Notes" for dtrace_register(), + * below. + * + * 1.4 void dtps_enable(void *arg, dtrace_id_t id, void *parg) + * + * 1.4.1 Overview + * + * Called to enable the specified probe. + * + * 1.4.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the probe to be enabled. The third + * argument is the probe argument as passed to dtrace_probe_create(). + * dtps_enable() will be called when a probe transitions from not being + * enabled at all to having one or more ECB. The number of ECBs associated + * with the probe may change without subsequent calls into the provider. + * When the number of ECBs drops to zero, the provider will be explicitly + * told to disable the probe via dtps_disable(). dtrace_probe() should never + * be called for a probe identifier that hasn't been explicitly enabled via + * dtps_enable(). + * + * 1.4.3 Return value + * + * None. + * + * 1.4.4 Caller's context + * + * The DTrace framework is locked in such a way that it may not be called + * back into at all. cpu_lock is held. mod_lock is not held and may not + * be acquired. + * + * 1.5 void dtps_disable(void *arg, dtrace_id_t id, void *parg) + * + * 1.5.1 Overview + * + * Called to disable the specified probe. + * + * 1.5.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the probe to be disabled. The third + * argument is the probe argument as passed to dtrace_probe_create(). + * dtps_disable() will be called when a probe transitions from being enabled + * to having zero ECBs. dtrace_probe() should never be called for a probe + * identifier that has been explicitly enabled via dtps_disable(). + * + * 1.5.3 Return value + * + * None. + * + * 1.5.4 Caller's context + * + * The DTrace framework is locked in such a way that it may not be called + * back into at all. cpu_lock is held. mod_lock is not held and may not + * be acquired. + * + * 1.6 void dtps_suspend(void *arg, dtrace_id_t id, void *parg) + * + * 1.6.1 Overview + * + * Called to suspend the specified enabled probe. This entry point is for + * providers that may need to suspend some or all of their probes when CPUs + * are being powered on or when the boot monitor is being entered for a + * prolonged period of time. + * + * 1.6.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the probe to be suspended. The + * third argument is the probe argument as passed to dtrace_probe_create(). + * dtps_suspend will only be called on an enabled probe. Providers that + * provide a dtps_suspend entry point will want to take roughly the action + * that it takes for dtps_disable. + * + * 1.6.3 Return value + * + * None. + * + * 1.6.4 Caller's context + * + * Interrupts are disabled. The DTrace framework is in a state such that the + * specified probe cannot be disabled or destroyed for the duration of + * dtps_suspend(). As interrupts are disabled, the provider is afforded + * little latitude; the provider is expected to do no more than a store to + * memory. + * + * 1.7 void dtps_resume(void *arg, dtrace_id_t id, void *parg) + * + * 1.7.1 Overview + * + * Called to resume the specified enabled probe. This entry point is for + * providers that may need to resume some or all of their probes after the + * completion of an event that induced a call to dtps_suspend(). + * + * 1.7.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the probe to be resumed. The + * third argument is the probe argument as passed to dtrace_probe_create(). + * dtps_resume will only be called on an enabled probe. Providers that + * provide a dtps_resume entry point will want to take roughly the action + * that it takes for dtps_enable. + * + * 1.7.3 Return value + * + * None. + * + * 1.7.4 Caller's context + * + * Interrupts are disabled. The DTrace framework is in a state such that the + * specified probe cannot be disabled or destroyed for the duration of + * dtps_resume(). As interrupts are disabled, the provider is afforded + * little latitude; the provider is expected to do no more than a store to + * memory. + * + * 1.8 void dtps_getargdesc(void *arg, dtrace_id_t id, void *parg, + * dtrace_argdesc_t *desc) + * + * 1.8.1 Overview + * + * Called to retrieve the argument description for an args[X] variable. + * + * 1.8.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the current probe. The third + * argument is the probe argument as passed to dtrace_probe_create(). The + * fourth argument is a pointer to the argument description. This + * description is both an input and output parameter: it contains the + * index of the desired argument in the dtargd_ndx field, and expects + * the other fields to be filled in upon return. If there is no argument + * corresponding to the specified index, the dtargd_ndx field should be set + * to DTRACE_ARGNONE. + * + * 1.8.3 Return value + * + * None. The dtargd_ndx, dtargd_native, dtargd_xlate and dtargd_mapping + * members of the dtrace_argdesc_t structure are all output values. + * + * 1.8.4 Caller's context + * + * dtps_getargdesc() is called from ioctl() context. mod_lock is held, and + * the DTrace framework is locked in such a way that providers may not + * register or unregister. This means that the provider may not call any + * DTrace API that affects its registration with the framework, including + * dtrace_register(), dtrace_unregister(), dtrace_invalidate(), and + * dtrace_condense(). + * + * 1.9 uint64_t dtps_getargval(void *arg, dtrace_id_t id, void *parg, + * int argno, int aframes) + * + * 1.9.1 Overview + * + * Called to retrieve a value for an argX or args[X] variable. + * + * 1.9.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the current probe. The third + * argument is the probe argument as passed to dtrace_probe_create(). The + * fourth argument is the number of the argument (the X in the example in + * 1.9.1). The fifth argument is the number of stack frames that were used + * to get from the actual place in the code that fired the probe to + * dtrace_probe() itself, the so-called artificial frames. This argument may + * be used to descend an appropriate number of frames to find the correct + * values. If this entry point is left NULL, the dtrace_getarg() built-in + * function is used. + * + * 1.9.3 Return value + * + * The value of the argument. + * + * 1.9.4 Caller's context + * + * This is called from within dtrace_probe() meaning that interrupts + * are disabled. No locks should be taken within this entry point. + * + * 1.10 int dtps_usermode(void *arg, dtrace_id_t id, void *parg) + * + * 1.10.1 Overview + * + * Called to determine if the probe was fired in a user context. + * + * 1.10.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the current probe. The third + * argument is the probe argument as passed to dtrace_probe_create(). This + * entry point must not be left NULL for providers whose probes allow for + * mixed mode tracing, that is to say those probes that can fire during + * kernel- _or_ user-mode execution + * + * 1.10.3 Return value + * + * A boolean value. + * + * 1.10.4 Caller's context + * + * This is called from within dtrace_probe() meaning that interrupts + * are disabled. No locks should be taken within this entry point. + * + * 1.11 void dtps_destroy(void *arg, dtrace_id_t id, void *parg) + * + * 1.11.1 Overview + * + * Called to destroy the specified probe. + * + * 1.11.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the probe to be destroyed. The third + * argument is the probe argument as passed to dtrace_probe_create(). The + * provider should free all state associated with the probe. The framework + * guarantees that dtps_destroy() is only called for probes that have either + * been disabled via dtps_disable() or were never enabled via dtps_enable(). + * Once dtps_disable() has been called for a probe, no further call will be + * made specifying the probe. + * + * 1.11.3 Return value + * + * None. + * + * 1.11.4 Caller's context + * + * The DTrace framework is locked in such a way that it may not be called + * back into at all. mod_lock is held. cpu_lock is not held, and may not be + * acquired. + * + * + * 2 Provider-to-Framework API + * + * 2.1 Overview + * + * The Provider-to-Framework API provides the mechanism for the provider to + * register itself with the DTrace framework, to create probes, to lookup + * probes and (most importantly) to fire probes. The Provider-to-Framework + * consists of: + * + * dtrace_register() <-- Register a provider with the DTrace framework + * dtrace_unregister() <-- Remove a provider's DTrace registration + * dtrace_invalidate() <-- Invalidate the specified provider + * dtrace_condense() <-- Remove a provider's unenabled probes + * dtrace_attached() <-- Indicates whether or not DTrace has attached + * dtrace_probe_create() <-- Create a DTrace probe + * dtrace_probe_lookup() <-- Lookup a DTrace probe based on its name + * dtrace_probe_arg() <-- Return the probe argument for a specific probe + * dtrace_probe() <-- Fire the specified probe + * + * 2.2 int dtrace_register(const char *name, const dtrace_pattr_t *pap, + * uint32_t priv, cred_t *cr, const dtrace_pops_t *pops, void *arg, + * dtrace_provider_id_t *idp) + * + * 2.2.1 Overview + * + * dtrace_register() registers the calling provider with the DTrace + * framework. It should generally be called by DTrace providers in their + * attach(9E) entry point. + * + * 2.2.2 Arguments and Notes + * + * The first argument is the name of the provider. The second argument is a + * pointer to the stability attributes for the provider. The third argument + * is the privilege flags for the provider, and must be some combination of: + * + * DTRACE_PRIV_NONE <= All users may enable probes from this provider + * + * DTRACE_PRIV_PROC <= Any user with privilege of PRIV_DTRACE_PROC may + * enable probes from this provider + * + * DTRACE_PRIV_USER <= Any user with privilege of PRIV_DTRACE_USER may + * enable probes from this provider + * + * DTRACE_PRIV_KERNEL <= Any user with privilege of PRIV_DTRACE_KERNEL + * may enable probes from this provider + * + * DTRACE_PRIV_OWNER <= This flag places an additional constraint on + * the privilege requirements above. These probes + * require either (a) a user ID matching the user + * ID of the cred passed in the fourth argument + * or (b) the PRIV_PROC_OWNER privilege. + * + * DTRACE_PRIV_ZONEOWNER<= This flag places an additional constraint on + * the privilege requirements above. These probes + * require either (a) a zone ID matching the zone + * ID of the cred passed in the fourth argument + * or (b) the PRIV_PROC_ZONE privilege. + * + * Note that these flags designate the _visibility_ of the probes, not + * the conditions under which they may or may not fire. + * + * The fourth argument is the credential that is associated with the + * provider. This argument should be NULL if the privilege flags don't + * include DTRACE_PRIV_OWNER or DTRACE_PRIV_ZONEOWNER. If non-NULL, the + * framework stashes the uid and zoneid represented by this credential + * for use at probe-time, in implicit predicates. These limit visibility + * of the probes to users and/or zones which have sufficient privilege to + * access them. + * + * The fifth argument is a DTrace provider operations vector, which provides + * the implementation for the Framework-to-Provider API. (See Section 1, + * above.) This must be non-NULL, and each member must be non-NULL. The + * exceptions to this are (1) the dtps_provide() and dtps_provide_module() + * members (if the provider so desires, _one_ of these members may be left + * NULL -- denoting that the provider only implements the other) and (2) + * the dtps_suspend() and dtps_resume() members, which must either both be + * NULL or both be non-NULL. + * + * The sixth argument is a cookie to be specified as the first argument for + * each function in the Framework-to-Provider API. This argument may have + * any value. + * + * The final argument is a pointer to dtrace_provider_id_t. If + * dtrace_register() successfully completes, the provider identifier will be + * stored in the memory pointed to be this argument. This argument must be + * non-NULL. + * + * 2.2.3 Return value + * + * On success, dtrace_register() returns 0 and stores the new provider's + * identifier into the memory pointed to by the idp argument. On failure, + * dtrace_register() returns an errno: + * + * EINVAL The arguments passed to dtrace_register() were somehow invalid. + * This may because a parameter that must be non-NULL was NULL, + * because the name was invalid (either empty or an illegal + * provider name) or because the attributes were invalid. + * + * No other failure code is returned. + * + * 2.2.4 Caller's context + * + * dtrace_register() may induce calls to dtrace_provide(); the provider must + * hold no locks across dtrace_register() that may also be acquired by + * dtrace_provide(). cpu_lock and mod_lock must not be held. + * + * 2.3 int dtrace_unregister(dtrace_provider_t id) + * + * 2.3.1 Overview + * + * Unregisters the specified provider from the DTrace framework. It should + * generally be called by DTrace providers in their detach(9E) entry point. + * + * 2.3.2 Arguments and Notes + * + * The only argument is the provider identifier, as returned from a + * successful call to dtrace_register(). As a result of calling + * dtrace_unregister(), the DTrace framework will call back into the provider + * via the dtps_destroy() entry point. Once dtrace_unregister() successfully + * completes, however, the DTrace framework will no longer make calls through + * the Framework-to-Provider API. + * + * 2.3.3 Return value + * + * On success, dtrace_unregister returns 0. On failure, dtrace_unregister() + * returns an errno: + * + * EBUSY There are currently processes that have the DTrace pseudodevice + * open, or there exists an anonymous enabling that hasn't yet + * been claimed. + * + * No other failure code is returned. + * + * 2.3.4 Caller's context + * + * Because a call to dtrace_unregister() may induce calls through the + * Framework-to-Provider API, the caller may not hold any lock across + * dtrace_register() that is also acquired in any of the Framework-to- + * Provider API functions. Additionally, mod_lock may not be held. + * + * 2.4 void dtrace_invalidate(dtrace_provider_id_t id) + * + * 2.4.1 Overview + * + * Invalidates the specified provider. All subsequent probe lookups for the + * specified provider will fail, but its probes will not be removed. + * + * 2.4.2 Arguments and note + * + * The only argument is the provider identifier, as returned from a + * successful call to dtrace_register(). In general, a provider's probes + * always remain valid; dtrace_invalidate() is a mechanism for invalidating + * an entire provider, regardless of whether or not probes are enabled or + * not. Note that dtrace_invalidate() will _not_ prevent already enabled + * probes from firing -- it will merely prevent any new enablings of the + * provider's probes. + * + * 2.5 int dtrace_condense(dtrace_provider_id_t id) + * + * 2.5.1 Overview + * + * Removes all the unenabled probes for the given provider. This function is + * not unlike dtrace_unregister(), except that it doesn't remove the + * provider just as many of its associated probes as it can. + * + * 2.5.2 Arguments and Notes + * + * As with dtrace_unregister(), the sole argument is the provider identifier + * as returned from a successful call to dtrace_register(). As a result of + * calling dtrace_condense(), the DTrace framework will call back into the + * given provider's dtps_destroy() entry point for each of the provider's + * unenabled probes. + * + * 2.5.3 Return value + * + * Currently, dtrace_condense() always returns 0. However, consumers of this + * function should check the return value as appropriate; its behavior may + * change in the future. + * + * 2.5.4 Caller's context + * + * As with dtrace_unregister(), the caller may not hold any lock across + * dtrace_condense() that is also acquired in the provider's entry points. + * Also, mod_lock may not be held. + * + * 2.6 int dtrace_attached() + * + * 2.6.1 Overview + * + * Indicates whether or not DTrace has attached. + * + * 2.6.2 Arguments and Notes + * + * For most providers, DTrace makes initial contact beyond registration. + * That is, once a provider has registered with DTrace, it waits to hear + * from DTrace to create probes. However, some providers may wish to + * proactively create probes without first being told by DTrace to do so. + * If providers wish to do this, they must first call dtrace_attached() to + * determine if DTrace itself has attached. If dtrace_attached() returns 0, + * the provider must not make any other Provider-to-Framework API call. + * + * 2.6.3 Return value + * + * dtrace_attached() returns 1 if DTrace has attached, 0 otherwise. + * + * 2.7 int dtrace_probe_create(dtrace_provider_t id, const char *mod, + * const char *func, const char *name, int aframes, void *arg) + * + * 2.7.1 Overview + * + * Creates a probe with specified module name, function name, and name. + * + * 2.7.2 Arguments and Notes + * + * The first argument is the provider identifier, as returned from a + * successful call to dtrace_register(). The second, third, and fourth + * arguments are the module name, function name, and probe name, + * respectively. Of these, module name and function name may both be NULL + * (in which case the probe is considered to be unanchored), or they may both + * be non-NULL. The name must be non-NULL, and must point to a non-empty + * string. + * + * The fifth argument is the number of artificial stack frames that will be + * found on the stack when dtrace_probe() is called for the new probe. These + * artificial frames will be automatically be pruned should the stack() or + * stackdepth() functions be called as part of one of the probe's ECBs. If + * the parameter doesn't add an artificial frame, this parameter should be + * zero. + * + * The final argument is a probe argument that will be passed back to the + * provider when a probe-specific operation is called. (e.g., via + * dtps_enable(), dtps_disable(), etc.) + * + * Note that it is up to the provider to be sure that the probe that it + * creates does not already exist -- if the provider is unsure of the probe's + * existence, it should assure its absence with dtrace_probe_lookup() before + * calling dtrace_probe_create(). + * + * 2.7.3 Return value + * + * dtrace_probe_create() always succeeds, and always returns the identifier + * of the newly-created probe. + * + * 2.7.4 Caller's context + * + * While dtrace_probe_create() is generally expected to be called from + * dtps_provide() and/or dtps_provide_module(), it may be called from other + * non-DTrace contexts. Neither cpu_lock nor mod_lock may be held. + * + * 2.8 dtrace_id_t dtrace_probe_lookup(dtrace_provider_t id, const char *mod, + * const char *func, const char *name) + * + * 2.8.1 Overview + * + * Looks up a probe based on provdider and one or more of module name, + * function name and probe name. + * + * 2.8.2 Arguments and Notes + * + * The first argument is the provider identifier, as returned from a + * successful call to dtrace_register(). The second, third, and fourth + * arguments are the module name, function name, and probe name, + * respectively. Any of these may be NULL; dtrace_probe_lookup() will return + * the identifier of the first probe that is provided by the specified + * provider and matches all of the non-NULL matching criteria. + * dtrace_probe_lookup() is generally used by a provider to be check the + * existence of a probe before creating it with dtrace_probe_create(). + * + * 2.8.3 Return value + * + * If the probe exists, returns its identifier. If the probe does not exist, + * return DTRACE_IDNONE. + * + * 2.8.4 Caller's context + * + * While dtrace_probe_lookup() is generally expected to be called from + * dtps_provide() and/or dtps_provide_module(), it may also be called from + * other non-DTrace contexts. Neither cpu_lock nor mod_lock may be held. + * + * 2.9 void *dtrace_probe_arg(dtrace_provider_t id, dtrace_id_t probe) + * + * 2.9.1 Overview + * + * Returns the probe argument associated with the specified probe. + * + * 2.9.2 Arguments and Notes + * + * The first argument is the provider identifier, as returned from a + * successful call to dtrace_register(). The second argument is a probe + * identifier, as returned from dtrace_probe_lookup() or + * dtrace_probe_create(). This is useful if a probe has multiple + * provider-specific components to it: the provider can create the probe + * once with provider-specific state, and then add to the state by looking + * up the probe based on probe identifier. + * + * 2.9.3 Return value + * + * Returns the argument associated with the specified probe. If the + * specified probe does not exist, or if the specified probe is not provided + * by the specified provider, NULL is returned. + * + * 2.9.4 Caller's context + * + * While dtrace_probe_arg() is generally expected to be called from + * dtps_provide() and/or dtps_provide_module(), it may also be called from + * other non-DTrace contexts. Neither cpu_lock nor mod_lock may be held. + * + * 2.10 void dtrace_probe(dtrace_id_t probe, uintptr_t arg0, uintptr_t arg1, + * uintptr_t arg2, uintptr_t arg3, uintptr_t arg4) + * + * 2.10.1 Overview + * + * The epicenter of DTrace: fires the specified probes with the specified + * arguments. + * + * 2.10.2 Arguments and Notes + * + * The first argument is a probe identifier as returned by + * dtrace_probe_create() or dtrace_probe_lookup(). The second through sixth + * arguments are the values to which the D variables "arg0" through "arg4" + * will be mapped. + * + * dtrace_probe() should be called whenever the specified probe has fired -- + * however the provider defines it. + * + * 2.10.3 Return value + * + * None. + * + * 2.10.4 Caller's context + * + * dtrace_probe() may be called in virtually any context: kernel, user, + * interrupt, high-level interrupt, with arbitrary adaptive locks held, with + * dispatcher locks held, with interrupts disabled, etc. The only latitude + * that must be afforded to DTrace is the ability to make calls within + * itself (and to its in-kernel subroutines) and the ability to access + * arbitrary (but mapped) memory. On some platforms, this constrains + * context. For example, on UltraSPARC, dtrace_probe() cannot be called + * from any context in which TL is greater than zero. dtrace_probe() may + * also not be called from any routine which may be called by dtrace_probe() + * -- which includes functions in the DTrace framework and some in-kernel + * DTrace subroutines. All such functions "dtrace_"; providers that + * instrument the kernel arbitrarily should be sure to not instrument these + * routines. + */ +typedef struct dtrace_pops { + void (*dtps_provide)(void *arg, dtrace_probedesc_t *spec); + void (*dtps_provide_module)(void *arg, modctl_t *mp); + void (*dtps_enable)(void *arg, dtrace_id_t id, void *parg); + void (*dtps_disable)(void *arg, dtrace_id_t id, void *parg); + void (*dtps_suspend)(void *arg, dtrace_id_t id, void *parg); + void (*dtps_resume)(void *arg, dtrace_id_t id, void *parg); + void (*dtps_getargdesc)(void *arg, dtrace_id_t id, void *parg, + dtrace_argdesc_t *desc); + uint64_t (*dtps_getargval)(void *arg, dtrace_id_t id, void *parg, + int argno, int aframes); + int (*dtps_usermode)(void *arg, dtrace_id_t id, void *parg); + void (*dtps_destroy)(void *arg, dtrace_id_t id, void *parg); +} dtrace_pops_t; + +typedef uintptr_t dtrace_provider_id_t; + +extern int dtrace_register(const char *, const dtrace_pattr_t *, uint32_t, + cred_t *, const dtrace_pops_t *, void *, dtrace_provider_id_t *); +extern int dtrace_unregister(dtrace_provider_id_t); +extern int dtrace_condense(dtrace_provider_id_t); +extern void dtrace_invalidate(dtrace_provider_id_t); +extern dtrace_id_t dtrace_probe_lookup(dtrace_provider_id_t, char *, + char *, char *); +extern dtrace_id_t dtrace_probe_create(dtrace_provider_id_t, const char *, + const char *, const char *, int, void *); +extern void *dtrace_probe_arg(dtrace_provider_id_t, dtrace_id_t); +extern void dtrace_probe(dtrace_id_t, uintptr_t arg0, uintptr_t arg1, + uintptr_t arg2, uintptr_t arg3, uintptr_t arg4); + +/* + * DTrace Meta Provider API + * + * The following functions are implemented by the DTrace framework and are + * used to implement meta providers. Meta providers plug into the DTrace + * framework and are used to instantiate new providers on the fly. At + * present, there is only one type of meta provider and only one meta + * provider may be registered with the DTrace framework at a time. The + * sole meta provider type provides user-land static tracing facilities + * by taking meta probe descriptions and adding a corresponding provider + * into the DTrace framework. + * + * 1 Framework-to-Provider + * + * 1.1 Overview + * + * The Framework-to-Provider API is represented by the dtrace_mops structure + * that the meta provider passes to the framework when registering itself as + * a meta provider. This structure consists of the following members: + * + * dtms_create_probe() <-- Add a new probe to a created provider + * dtms_provide_pid() <-- Create a new provider for a given process + * dtms_remove_pid() <-- Remove a previously created provider + * + * 1.2 void dtms_create_probe(void *arg, void *parg, + * dtrace_helper_probedesc_t *probedesc); + * + * 1.2.1 Overview + * + * Called by the DTrace framework to create a new probe in a provider + * created by this meta provider. + * + * 1.2.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_meta_register(). + * The second argument is the provider cookie for the associated provider; + * this is obtained from the return value of dtms_provide_pid(). The third + * argument is the helper probe description. + * + * 1.2.3 Return value + * + * None + * + * 1.2.4 Caller's context + * + * dtms_create_probe() is called from either ioctl() or module load context. + * The DTrace framework is locked in such a way that meta providers may not + * register or unregister. This means that the meta provider cannot call + * dtrace_meta_register() or dtrace_meta_unregister(). However, the context is + * such that the provider may (and is expected to) call provider-related + * DTrace provider APIs including dtrace_probe_create(). + * + * 1.3 void *dtms_provide_pid(void *arg, dtrace_meta_provider_t *mprov, + * pid_t pid) + * + * 1.3.1 Overview + * + * Called by the DTrace framework to instantiate a new provider given the + * description of the provider and probes in the mprov argument. The + * meta provider should call dtrace_register() to insert the new provider + * into the DTrace framework. + * + * 1.3.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_meta_register(). + * The second argument is a pointer to a structure describing the new + * helper provider. The third argument is the process identifier for + * process associated with this new provider. Note that the name of the + * provider as passed to dtrace_register() should be the contatenation of + * the dtmpb_provname member of the mprov argument and the processs + * identifier as a string. + * + * 1.3.3 Return value + * + * The cookie for the provider that the meta provider creates. This is + * the same value that it passed to dtrace_register(). + * + * 1.3.4 Caller's context + * + * dtms_provide_pid() is called from either ioctl() or module load context. + * The DTrace framework is locked in such a way that meta providers may not + * register or unregister. This means that the meta provider cannot call + * dtrace_meta_register() or dtrace_meta_unregister(). However, the context + * is such that the provider may -- and is expected to -- call + * provider-related DTrace provider APIs including dtrace_register(). + * + * 1.4 void dtms_remove_pid(void *arg, dtrace_meta_provider_t *mprov, + * pid_t pid) + * + * 1.4.1 Overview + * + * Called by the DTrace framework to remove a provider that had previously + * been instantiated via the dtms_provide_pid() entry point. The meta + * provider need not remove the provider immediately, but this entry + * point indicates that the provider should be removed as soon as possible + * using the dtrace_unregister() API. + * + * 1.4.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_meta_register(). + * The second argument is a pointer to a structure describing the helper + * provider. The third argument is the process identifier for process + * associated with this new provider. + * + * 1.4.3 Return value + * + * None + * + * 1.4.4 Caller's context + * + * dtms_remove_pid() is called from either ioctl() or exit() context. + * The DTrace framework is locked in such a way that meta providers may not + * register or unregister. This means that the meta provider cannot call + * dtrace_meta_register() or dtrace_meta_unregister(). However, the context + * is such that the provider may -- and is expected to -- call + * provider-related DTrace provider APIs including dtrace_unregister(). + */ +typedef struct dtrace_helper_probedesc { + char *dthpb_mod; /* probe module */ + char *dthpb_func; /* probe function */ + char *dthpb_name; /* probe name */ + uint64_t dthpb_base; /* base address */ + uint32_t *dthpb_offs; /* offsets array */ + uint32_t *dthpb_enoffs; /* is-enabled offsets array */ + uint32_t dthpb_noffs; /* offsets count */ + uint32_t dthpb_nenoffs; /* is-enabled offsets count */ + uint8_t *dthpb_args; /* argument mapping array */ + uint8_t dthpb_xargc; /* translated argument count */ + uint8_t dthpb_nargc; /* native argument count */ + char *dthpb_xtypes; /* translated types strings */ + char *dthpb_ntypes; /* native types strings */ +} dtrace_helper_probedesc_t; + +typedef struct dtrace_helper_provdesc { + char *dthpv_provname; /* provider name */ + dtrace_pattr_t dthpv_pattr; /* stability attributes */ +} dtrace_helper_provdesc_t; + +typedef struct dtrace_mops { + void (*dtms_create_probe)(void *, void *, dtrace_helper_probedesc_t *); + void *(*dtms_provide_pid)(void *, dtrace_helper_provdesc_t *, pid_t); + void (*dtms_remove_pid)(void *, dtrace_helper_provdesc_t *, pid_t); +} dtrace_mops_t; + +typedef uintptr_t dtrace_meta_provider_id_t; + +extern int dtrace_meta_register(const char *, const dtrace_mops_t *, void *, + dtrace_meta_provider_id_t *); +extern int dtrace_meta_unregister(dtrace_meta_provider_id_t); + +/* + * DTrace Kernel Hooks + * + * The following functions are implemented by the base kernel and form a set of + * hooks used by the DTrace framework. DTrace hooks are implemented in either + * uts/common/os/dtrace_subr.c, an ISA-specific assembly file, or in a + * uts/<platform>/os/dtrace_subr.c corresponding to each hardware platform. + */ + +typedef enum dtrace_vtime_state { + DTRACE_VTIME_INACTIVE = 0, /* No DTrace, no TNF */ + DTRACE_VTIME_ACTIVE, /* DTrace virtual time, no TNF */ + DTRACE_VTIME_INACTIVE_TNF, /* No DTrace, TNF active */ + DTRACE_VTIME_ACTIVE_TNF /* DTrace virtual time _and_ TNF */ +} dtrace_vtime_state_t; + +#if defined(sun) +extern dtrace_vtime_state_t dtrace_vtime_active; +#endif +extern void dtrace_vtime_switch(kthread_t *next); +extern void dtrace_vtime_enable_tnf(void); +extern void dtrace_vtime_disable_tnf(void); +extern void dtrace_vtime_enable(void); +extern void dtrace_vtime_disable(void); + +struct regs; + +#if defined(sun) +extern int (*dtrace_pid_probe_ptr)(struct regs *); +extern int (*dtrace_return_probe_ptr)(struct regs *); +extern void (*dtrace_fasttrap_fork_ptr)(proc_t *, proc_t *); +extern void (*dtrace_fasttrap_exec_ptr)(proc_t *); +extern void (*dtrace_fasttrap_exit_ptr)(proc_t *); +extern void dtrace_fasttrap_fork(proc_t *, proc_t *); +#endif + +typedef uintptr_t dtrace_icookie_t; +typedef void (*dtrace_xcall_t)(void *); + +extern dtrace_icookie_t dtrace_interrupt_disable(void); +extern void dtrace_interrupt_enable(dtrace_icookie_t); + +extern void dtrace_membar_producer(void); +extern void dtrace_membar_consumer(void); + +extern void (*dtrace_cpu_init)(processorid_t); +extern void (*dtrace_modload)(modctl_t *); +extern void (*dtrace_modunload)(modctl_t *); +extern void (*dtrace_helpers_cleanup)(void); +extern void (*dtrace_helpers_fork)(proc_t *parent, proc_t *child); +extern void (*dtrace_cpustart_init)(void); +extern void (*dtrace_cpustart_fini)(void); + +extern void (*dtrace_debugger_init)(void); +extern void (*dtrace_debugger_fini)(void); +extern dtrace_cacheid_t dtrace_predcache_id; + +#if defined(sun) +extern hrtime_t dtrace_gethrtime(void); +#else +void dtrace_debug_printf(const char *, ...) __printflike(1, 2); +#endif +extern void dtrace_sync(void); +extern void dtrace_toxic_ranges(void (*)(uintptr_t, uintptr_t)); +extern void dtrace_xcall(processorid_t, dtrace_xcall_t, void *); +extern void dtrace_vpanic(const char *, __va_list); +extern void dtrace_panic(const char *, ...); + +extern int dtrace_safe_defer_signal(void); +extern void dtrace_safe_synchronous_signal(void); + +extern int dtrace_mach_aframes(void); + +#if defined(__i386) || defined(__amd64) +extern int dtrace_instr_size(uchar_t *instr); +extern int dtrace_instr_size_isa(uchar_t *, model_t, int *); +extern void dtrace_invop_add(int (*)(uintptr_t, uintptr_t *, uintptr_t)); +extern void dtrace_invop_remove(int (*)(uintptr_t, uintptr_t *, uintptr_t)); +extern void dtrace_invop_callsite(void); +#endif + +#ifdef __sparc +extern int dtrace_blksuword32(uintptr_t, uint32_t *, int); +extern void dtrace_getfsr(uint64_t *); +#endif + +#define DTRACE_CPUFLAG_ISSET(flag) \ + (cpu_core[curcpu].cpuc_dtrace_flags & (flag)) + +#define DTRACE_CPUFLAG_SET(flag) \ + (cpu_core[curcpu].cpuc_dtrace_flags |= (flag)) + +#define DTRACE_CPUFLAG_CLEAR(flag) \ + (cpu_core[curcpu].cpuc_dtrace_flags &= ~(flag)) + +#endif /* _KERNEL */ + +#endif /* _ASM */ + +#if defined(__i386) || defined(__amd64) + +#define DTRACE_INVOP_PUSHL_EBP 1 +#define DTRACE_INVOP_POPL_EBP 2 +#define DTRACE_INVOP_LEAVE 3 +#define DTRACE_INVOP_NOP 4 +#define DTRACE_INVOP_RET 5 + +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* _SYS_DTRACE_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/dtrace_impl.h b/sys/cddl/contrib/opensolaris/uts/common/sys/dtrace_impl.h new file mode 100644 index 000000000000..0b2a2f061968 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/dtrace_impl.h @@ -0,0 +1,1323 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD: src/sys/cddl/contrib/opensolaris/uts/common/sys/dtrace_impl.h,v 1.3 2008/06/01 01:46:37 jb Exp $ + */ + +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _SYS_DTRACE_IMPL_H +#define _SYS_DTRACE_IMPL_H + +#pragma ident "%Z%%M% %I% %E% SMI" + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * DTrace Dynamic Tracing Software: Kernel Implementation Interfaces + * + * Note: The contents of this file are private to the implementation of the + * Solaris system and DTrace subsystem and are subject to change at any time + * without notice. Applications and drivers using these interfaces will fail + * to run on future releases. These interfaces should not be used for any + * purpose except those expressly outlined in dtrace(7D) and libdtrace(3LIB). + * Please refer to the "Solaris Dynamic Tracing Guide" for more information. + */ + +#include <sys/dtrace.h> +#if !defined(sun) +#ifdef __sparcv9 +typedef uint32_t pc_t; +#else +typedef uintptr_t pc_t; +#endif +typedef u_long greg_t; +#endif + +/* + * DTrace Implementation Constants and Typedefs + */ +#define DTRACE_MAXPROPLEN 128 +#define DTRACE_DYNVAR_CHUNKSIZE 256 + +struct dtrace_probe; +struct dtrace_ecb; +struct dtrace_predicate; +struct dtrace_action; +struct dtrace_provider; +struct dtrace_state; + +typedef struct dtrace_probe dtrace_probe_t; +typedef struct dtrace_ecb dtrace_ecb_t; +typedef struct dtrace_predicate dtrace_predicate_t; +typedef struct dtrace_action dtrace_action_t; +typedef struct dtrace_provider dtrace_provider_t; +typedef struct dtrace_meta dtrace_meta_t; +typedef struct dtrace_state dtrace_state_t; +typedef uint32_t dtrace_optid_t; +typedef uint32_t dtrace_specid_t; +typedef uint64_t dtrace_genid_t; + +/* + * DTrace Probes + * + * The probe is the fundamental unit of the DTrace architecture. Probes are + * created by DTrace providers, and managed by the DTrace framework. A probe + * is identified by a unique <provider, module, function, name> tuple, and has + * a unique probe identifier assigned to it. (Some probes are not associated + * with a specific point in text; these are called _unanchored probes_ and have + * no module or function associated with them.) Probes are represented as a + * dtrace_probe structure. To allow quick lookups based on each element of the + * probe tuple, probes are hashed by each of provider, module, function and + * name. (If a lookup is performed based on a regular expression, a + * dtrace_probekey is prepared, and a linear search is performed.) Each probe + * is additionally pointed to by a linear array indexed by its identifier. The + * identifier is the provider's mechanism for indicating to the DTrace + * framework that a probe has fired: the identifier is passed as the first + * argument to dtrace_probe(), where it is then mapped into the corresponding + * dtrace_probe structure. From the dtrace_probe structure, dtrace_probe() can + * iterate over the probe's list of enabling control blocks; see "DTrace + * Enabling Control Blocks", below.) + */ +struct dtrace_probe { + dtrace_id_t dtpr_id; /* probe identifier */ + dtrace_ecb_t *dtpr_ecb; /* ECB list; see below */ + dtrace_ecb_t *dtpr_ecb_last; /* last ECB in list */ + void *dtpr_arg; /* provider argument */ + dtrace_cacheid_t dtpr_predcache; /* predicate cache ID */ + int dtpr_aframes; /* artificial frames */ + dtrace_provider_t *dtpr_provider; /* pointer to provider */ + char *dtpr_mod; /* probe's module name */ + char *dtpr_func; /* probe's function name */ + char *dtpr_name; /* probe's name */ + dtrace_probe_t *dtpr_nextmod; /* next in module hash */ + dtrace_probe_t *dtpr_prevmod; /* previous in module hash */ + dtrace_probe_t *dtpr_nextfunc; /* next in function hash */ + dtrace_probe_t *dtpr_prevfunc; /* previous in function hash */ + dtrace_probe_t *dtpr_nextname; /* next in name hash */ + dtrace_probe_t *dtpr_prevname; /* previous in name hash */ + dtrace_genid_t dtpr_gen; /* probe generation ID */ +}; + +typedef int dtrace_probekey_f(const char *, const char *, int); + +typedef struct dtrace_probekey { + char *dtpk_prov; /* provider name to match */ + dtrace_probekey_f *dtpk_pmatch; /* provider matching function */ + char *dtpk_mod; /* module name to match */ + dtrace_probekey_f *dtpk_mmatch; /* module matching function */ + char *dtpk_func; /* func name to match */ + dtrace_probekey_f *dtpk_fmatch; /* func matching function */ + char *dtpk_name; /* name to match */ + dtrace_probekey_f *dtpk_nmatch; /* name matching function */ + dtrace_id_t dtpk_id; /* identifier to match */ +} dtrace_probekey_t; + +typedef struct dtrace_hashbucket { + struct dtrace_hashbucket *dthb_next; /* next on hash chain */ + dtrace_probe_t *dthb_chain; /* chain of probes */ + int dthb_len; /* number of probes here */ +} dtrace_hashbucket_t; + +typedef struct dtrace_hash { + dtrace_hashbucket_t **dth_tab; /* hash table */ + int dth_size; /* size of hash table */ + int dth_mask; /* mask to index into table */ + int dth_nbuckets; /* total number of buckets */ + uintptr_t dth_nextoffs; /* offset of next in probe */ + uintptr_t dth_prevoffs; /* offset of prev in probe */ + uintptr_t dth_stroffs; /* offset of str in probe */ +} dtrace_hash_t; + +/* + * DTrace Enabling Control Blocks + * + * When a provider wishes to fire a probe, it calls into dtrace_probe(), + * passing the probe identifier as the first argument. As described above, + * dtrace_probe() maps the identifier into a pointer to a dtrace_probe_t + * structure. This structure contains information about the probe, and a + * pointer to the list of Enabling Control Blocks (ECBs). Each ECB points to + * DTrace consumer state, and contains an optional predicate, and a list of + * actions. (Shown schematically below.) The ECB abstraction allows a single + * probe to be multiplexed across disjoint consumers, or across disjoint + * enablings of a single probe within one consumer. + * + * Enabling Control Block + * dtrace_ecb_t + * +------------------------+ + * | dtrace_epid_t ---------+--------------> Enabled Probe ID (EPID) + * | dtrace_state_t * ------+--------------> State associated with this ECB + * | dtrace_predicate_t * --+---------+ + * | dtrace_action_t * -----+----+ | + * | dtrace_ecb_t * ---+ | | | Predicate (if any) + * +-------------------+----+ | | dtrace_predicate_t + * | | +---> +--------------------+ + * | | | dtrace_difo_t * ---+----> DIFO + * | | +--------------------+ + * | | + * Next ECB | | Action + * (if any) | | dtrace_action_t + * : +--> +-------------------+ + * : | dtrace_actkind_t -+------> kind + * v | dtrace_difo_t * --+------> DIFO (if any) + * | dtrace_recdesc_t -+------> record descr. + * | dtrace_action_t * +------+ + * +-------------------+ | + * | Next action + * +-------------------------------+ (if any) + * | + * | Action + * | dtrace_action_t + * +--> +-------------------+ + * | dtrace_actkind_t -+------> kind + * | dtrace_difo_t * --+------> DIFO (if any) + * | dtrace_action_t * +------+ + * +-------------------+ | + * | Next action + * +-------------------------------+ (if any) + * | + * : + * v + * + * + * dtrace_probe() iterates over the ECB list. If the ECB needs less space + * than is available in the principal buffer, the ECB is processed: if the + * predicate is non-NULL, the DIF object is executed. If the result is + * non-zero, the action list is processed, with each action being executed + * accordingly. When the action list has been completely executed, processing + * advances to the next ECB. processing advances to the next ECB. If the + * result is non-zero; For each ECB, it first determines the The ECB + * abstraction allows disjoint consumers to multiplex on single probes. + */ +struct dtrace_ecb { + dtrace_epid_t dte_epid; /* enabled probe ID */ + uint32_t dte_alignment; /* required alignment */ + size_t dte_needed; /* bytes needed */ + size_t dte_size; /* total size of payload */ + dtrace_predicate_t *dte_predicate; /* predicate, if any */ + dtrace_action_t *dte_action; /* actions, if any */ + dtrace_ecb_t *dte_next; /* next ECB on probe */ + dtrace_state_t *dte_state; /* pointer to state */ + uint32_t dte_cond; /* security condition */ + dtrace_probe_t *dte_probe; /* pointer to probe */ + dtrace_action_t *dte_action_last; /* last action on ECB */ + uint64_t dte_uarg; /* library argument */ +}; + +struct dtrace_predicate { + dtrace_difo_t *dtp_difo; /* DIF object */ + dtrace_cacheid_t dtp_cacheid; /* cache identifier */ + int dtp_refcnt; /* reference count */ +}; + +struct dtrace_action { + dtrace_actkind_t dta_kind; /* kind of action */ + uint16_t dta_intuple; /* boolean: in aggregation */ + uint32_t dta_refcnt; /* reference count */ + dtrace_difo_t *dta_difo; /* pointer to DIFO */ + dtrace_recdesc_t dta_rec; /* record description */ + dtrace_action_t *dta_prev; /* previous action */ + dtrace_action_t *dta_next; /* next action */ +}; + +typedef struct dtrace_aggregation { + dtrace_action_t dtag_action; /* action; must be first */ + dtrace_aggid_t dtag_id; /* identifier */ + dtrace_ecb_t *dtag_ecb; /* corresponding ECB */ + dtrace_action_t *dtag_first; /* first action in tuple */ + uint32_t dtag_base; /* base of aggregation */ + uint8_t dtag_hasarg; /* boolean: has argument */ + uint64_t dtag_initial; /* initial value */ + void (*dtag_aggregate)(uint64_t *, uint64_t, uint64_t); +} dtrace_aggregation_t; + +/* + * DTrace Buffers + * + * Principal buffers, aggregation buffers, and speculative buffers are all + * managed with the dtrace_buffer structure. By default, this structure + * includes twin data buffers -- dtb_tomax and dtb_xamot -- that serve as the + * active and passive buffers, respectively. For speculative buffers, + * dtb_xamot will be NULL; for "ring" and "fill" buffers, dtb_xamot will point + * to a scratch buffer. For all buffer types, the dtrace_buffer structure is + * always allocated on a per-CPU basis; a single dtrace_buffer structure is + * never shared among CPUs. (That is, there is never true sharing of the + * dtrace_buffer structure; to prevent false sharing of the structure, it must + * always be aligned to the coherence granularity -- generally 64 bytes.) + * + * One of the critical design decisions of DTrace is that a given ECB always + * stores the same quantity and type of data. This is done to assure that the + * only metadata required for an ECB's traced data is the EPID. That is, from + * the EPID, the consumer can determine the data layout. (The data buffer + * layout is shown schematically below.) By assuring that one can determine + * data layout from the EPID, the metadata stream can be separated from the + * data stream -- simplifying the data stream enormously. + * + * base of data buffer ---> +------+--------------------+------+ + * | EPID | data | EPID | + * +------+--------+------+----+------+ + * | data | EPID | data | + * +---------------+------+-----------+ + * | data, cont. | + * +------+--------------------+------+ + * | EPID | data | | + * +------+--------------------+ | + * | || | + * | || | + * | \/ | + * : : + * . . + * . . + * . . + * : : + * | | + * limit of data buffer ---> +----------------------------------+ + * + * When evaluating an ECB, dtrace_probe() determines if the ECB's needs of the + * principal buffer (both scratch and payload) exceed the available space. If + * the ECB's needs exceed available space (and if the principal buffer policy + * is the default "switch" policy), the ECB is dropped, the buffer's drop count + * is incremented, and processing advances to the next ECB. If the ECB's needs + * can be met with the available space, the ECB is processed, but the offset in + * the principal buffer is only advanced if the ECB completes processing + * without error. + * + * When a buffer is to be switched (either because the buffer is the principal + * buffer with a "switch" policy or because it is an aggregation buffer), a + * cross call is issued to the CPU associated with the buffer. In the cross + * call context, interrupts are disabled, and the active and the inactive + * buffers are atomically switched. This involves switching the data pointers, + * copying the various state fields (offset, drops, errors, etc.) into their + * inactive equivalents, and clearing the state fields. Because interrupts are + * disabled during this procedure, the switch is guaranteed to appear atomic to + * dtrace_probe(). + * + * DTrace Ring Buffering + * + * To process a ring buffer correctly, one must know the oldest valid record. + * Processing starts at the oldest record in the buffer and continues until + * the end of the buffer is reached. Processing then resumes starting with + * the record stored at offset 0 in the buffer, and continues until the + * youngest record is processed. If trace records are of a fixed-length, + * determining the oldest record is trivial: + * + * - If the ring buffer has not wrapped, the oldest record is the record + * stored at offset 0. + * + * - If the ring buffer has wrapped, the oldest record is the record stored + * at the current offset. + * + * With variable length records, however, just knowing the current offset + * doesn't suffice for determining the oldest valid record: assuming that one + * allows for arbitrary data, one has no way of searching forward from the + * current offset to find the oldest valid record. (That is, one has no way + * of separating data from metadata.) It would be possible to simply refuse to + * process any data in the ring buffer between the current offset and the + * limit, but this leaves (potentially) an enormous amount of otherwise valid + * data unprocessed. + * + * To effect ring buffering, we track two offsets in the buffer: the current + * offset and the _wrapped_ offset. If a request is made to reserve some + * amount of data, and the buffer has wrapped, the wrapped offset is + * incremented until the wrapped offset minus the current offset is greater + * than or equal to the reserve request. This is done by repeatedly looking + * up the ECB corresponding to the EPID at the current wrapped offset, and + * incrementing the wrapped offset by the size of the data payload + * corresponding to that ECB. If this offset is greater than or equal to the + * limit of the data buffer, the wrapped offset is set to 0. Thus, the + * current offset effectively "chases" the wrapped offset around the buffer. + * Schematically: + * + * base of data buffer ---> +------+--------------------+------+ + * | EPID | data | EPID | + * +------+--------+------+----+------+ + * | data | EPID | data | + * +---------------+------+-----------+ + * | data, cont. | + * +------+---------------------------+ + * | EPID | data | + * current offset ---> +------+---------------------------+ + * | invalid data | + * wrapped offset ---> +------+--------------------+------+ + * | EPID | data | EPID | + * +------+--------+------+----+------+ + * | data | EPID | data | + * +---------------+------+-----------+ + * : : + * . . + * . ... valid data ... . + * . . + * : : + * +------+-------------+------+------+ + * | EPID | data | EPID | data | + * +------+------------++------+------+ + * | data, cont. | leftover | + * limit of data buffer ---> +-------------------+--------------+ + * + * If the amount of requested buffer space exceeds the amount of space + * available between the current offset and the end of the buffer: + * + * (1) all words in the data buffer between the current offset and the limit + * of the data buffer (marked "leftover", above) are set to + * DTRACE_EPIDNONE + * + * (2) the wrapped offset is set to zero + * + * (3) the iteration process described above occurs until the wrapped offset + * is greater than the amount of desired space. + * + * The wrapped offset is implemented by (re-)using the inactive offset. + * In a "switch" buffer policy, the inactive offset stores the offset in + * the inactive buffer; in a "ring" buffer policy, it stores the wrapped + * offset. + * + * DTrace Scratch Buffering + * + * Some ECBs may wish to allocate dynamically-sized temporary scratch memory. + * To accommodate such requests easily, scratch memory may be allocated in + * the buffer beyond the current offset plus the needed memory of the current + * ECB. If there isn't sufficient room in the buffer for the requested amount + * of scratch space, the allocation fails and an error is generated. Scratch + * memory is tracked in the dtrace_mstate_t and is automatically freed when + * the ECB ceases processing. Note that ring buffers cannot allocate their + * scratch from the principal buffer -- lest they needlessly overwrite older, + * valid data. Ring buffers therefore have their own dedicated scratch buffer + * from which scratch is allocated. + */ +#define DTRACEBUF_RING 0x0001 /* bufpolicy set to "ring" */ +#define DTRACEBUF_FILL 0x0002 /* bufpolicy set to "fill" */ +#define DTRACEBUF_NOSWITCH 0x0004 /* do not switch buffer */ +#define DTRACEBUF_WRAPPED 0x0008 /* ring buffer has wrapped */ +#define DTRACEBUF_DROPPED 0x0010 /* drops occurred */ +#define DTRACEBUF_ERROR 0x0020 /* errors occurred */ +#define DTRACEBUF_FULL 0x0040 /* "fill" buffer is full */ +#define DTRACEBUF_CONSUMED 0x0080 /* buffer has been consumed */ +#define DTRACEBUF_INACTIVE 0x0100 /* buffer is not yet active */ + +typedef struct dtrace_buffer { + uint64_t dtb_offset; /* current offset in buffer */ + uint64_t dtb_size; /* size of buffer */ + uint32_t dtb_flags; /* flags */ + uint32_t dtb_drops; /* number of drops */ + caddr_t dtb_tomax; /* active buffer */ + caddr_t dtb_xamot; /* inactive buffer */ + uint32_t dtb_xamot_flags; /* inactive flags */ + uint32_t dtb_xamot_drops; /* drops in inactive buffer */ + uint64_t dtb_xamot_offset; /* offset in inactive buffer */ + uint32_t dtb_errors; /* number of errors */ + uint32_t dtb_xamot_errors; /* errors in inactive buffer */ +#ifndef _LP64 + uint64_t dtb_pad1; +#endif +} dtrace_buffer_t; + +/* + * DTrace Aggregation Buffers + * + * Aggregation buffers use much of the same mechanism as described above + * ("DTrace Buffers"). However, because an aggregation is fundamentally a + * hash, there exists dynamic metadata associated with an aggregation buffer + * that is not associated with other kinds of buffers. This aggregation + * metadata is _only_ relevant for the in-kernel implementation of + * aggregations; it is not actually relevant to user-level consumers. To do + * this, we allocate dynamic aggregation data (hash keys and hash buckets) + * starting below the _limit_ of the buffer, and we allocate data from the + * _base_ of the buffer. When the aggregation buffer is copied out, _only_ the + * data is copied out; the metadata is simply discarded. Schematically, + * aggregation buffers look like: + * + * base of data buffer ---> +-------+------+-----------+-------+ + * | aggid | key | value | aggid | + * +-------+------+-----------+-------+ + * | key | + * +-------+-------+-----+------------+ + * | value | aggid | key | value | + * +-------+------++-----+------+-----+ + * | aggid | key | value | | + * +-------+------+-------------+ | + * | || | + * | || | + * | \/ | + * : : + * . . + * . . + * . . + * : : + * | /\ | + * | || +------------+ + * | || | | + * +---------------------+ | + * | hash keys | + * | (dtrace_aggkey structures) | + * | | + * +----------------------------------+ + * | hash buckets | + * | (dtrace_aggbuffer structure) | + * | | + * limit of data buffer ---> +----------------------------------+ + * + * + * As implied above, just as we assure that ECBs always store a constant + * amount of data, we assure that a given aggregation -- identified by its + * aggregation ID -- always stores data of a constant quantity and type. + * As with EPIDs, this allows the aggregation ID to serve as the metadata for a + * given record. + * + * Note that the size of the dtrace_aggkey structure must be sizeof (uintptr_t) + * aligned. (If this the structure changes such that this becomes false, an + * assertion will fail in dtrace_aggregate().) + */ +typedef struct dtrace_aggkey { + uint32_t dtak_hashval; /* hash value */ + uint32_t dtak_action:4; /* action -- 4 bits */ + uint32_t dtak_size:28; /* size -- 28 bits */ + caddr_t dtak_data; /* data pointer */ + struct dtrace_aggkey *dtak_next; /* next in hash chain */ +} dtrace_aggkey_t; + +typedef struct dtrace_aggbuffer { + uintptr_t dtagb_hashsize; /* number of buckets */ + uintptr_t dtagb_free; /* free list of keys */ + dtrace_aggkey_t **dtagb_hash; /* hash table */ +} dtrace_aggbuffer_t; + +/* + * DTrace Speculations + * + * Speculations have a per-CPU buffer and a global state. Once a speculation + * buffer has been comitted or discarded, it cannot be reused until all CPUs + * have taken the same action (commit or discard) on their respective + * speculative buffer. However, because DTrace probes may execute in arbitrary + * context, other CPUs cannot simply be cross-called at probe firing time to + * perform the necessary commit or discard. The speculation states thus + * optimize for the case that a speculative buffer is only active on one CPU at + * the time of a commit() or discard() -- for if this is the case, other CPUs + * need not take action, and the speculation is immediately available for + * reuse. If the speculation is active on multiple CPUs, it must be + * asynchronously cleaned -- potentially leading to a higher rate of dirty + * speculative drops. The speculation states are as follows: + * + * DTRACESPEC_INACTIVE <= Initial state; inactive speculation + * DTRACESPEC_ACTIVE <= Allocated, but not yet speculatively traced to + * DTRACESPEC_ACTIVEONE <= Speculatively traced to on one CPU + * DTRACESPEC_ACTIVEMANY <= Speculatively traced to on more than one CPU + * DTRACESPEC_COMMITTING <= Currently being commited on one CPU + * DTRACESPEC_COMMITTINGMANY <= Currently being commited on many CPUs + * DTRACESPEC_DISCARDING <= Currently being discarded on many CPUs + * + * The state transition diagram is as follows: + * + * +----------------------------------------------------------+ + * | | + * | +------------+ | + * | +-------------------| COMMITTING |<-----------------+ | + * | | +------------+ | | + * | | copied spec. ^ commit() on | | discard() on + * | | into principal | active CPU | | active CPU + * | | | commit() | | + * V V | | | + * +----------+ +--------+ +-----------+ + * | INACTIVE |---------------->| ACTIVE |--------------->| ACTIVEONE | + * +----------+ speculation() +--------+ speculate() +-----------+ + * ^ ^ | | | + * | | | discard() | | + * | | asynchronously | discard() on | | speculate() + * | | cleaned V inactive CPU | | on inactive + * | | +------------+ | | CPU + * | +-------------------| DISCARDING |<-----------------+ | + * | +------------+ | + * | asynchronously ^ | + * | copied spec. | discard() | + * | into principal +------------------------+ | + * | | V + * +----------------+ commit() +------------+ + * | COMMITTINGMANY |<----------------------------------| ACTIVEMANY | + * +----------------+ +------------+ + */ +typedef enum dtrace_speculation_state { + DTRACESPEC_INACTIVE = 0, + DTRACESPEC_ACTIVE, + DTRACESPEC_ACTIVEONE, + DTRACESPEC_ACTIVEMANY, + DTRACESPEC_COMMITTING, + DTRACESPEC_COMMITTINGMANY, + DTRACESPEC_DISCARDING +} dtrace_speculation_state_t; + +typedef struct dtrace_speculation { + dtrace_speculation_state_t dtsp_state; /* current speculation state */ + int dtsp_cleaning; /* non-zero if being cleaned */ + dtrace_buffer_t *dtsp_buffer; /* speculative buffer */ +} dtrace_speculation_t; + +/* + * DTrace Dynamic Variables + * + * The dynamic variable problem is obviously decomposed into two subproblems: + * allocating new dynamic storage, and freeing old dynamic storage. The + * presence of the second problem makes the first much more complicated -- or + * rather, the absence of the second renders the first trivial. This is the + * case with aggregations, for which there is effectively no deallocation of + * dynamic storage. (Or more accurately, all dynamic storage is deallocated + * when a snapshot is taken of the aggregation.) As DTrace dynamic variables + * allow for both dynamic allocation and dynamic deallocation, the + * implementation of dynamic variables is quite a bit more complicated than + * that of their aggregation kin. + * + * We observe that allocating new dynamic storage is tricky only because the + * size can vary -- the allocation problem is much easier if allocation sizes + * are uniform. We further observe that in D, the size of dynamic variables is + * actually _not_ dynamic -- dynamic variable sizes may be determined by static + * analysis of DIF text. (This is true even of putatively dynamically-sized + * objects like strings and stacks, the sizes of which are dictated by the + * "stringsize" and "stackframes" variables, respectively.) We exploit this by + * performing this analysis on all DIF before enabling any probes. For each + * dynamic load or store, we calculate the dynamically-allocated size plus the + * size of the dtrace_dynvar structure plus the storage required to key the + * data. For all DIF, we take the largest value and dub it the _chunksize_. + * We then divide dynamic memory into two parts: a hash table that is wide + * enough to have every chunk in its own bucket, and a larger region of equal + * chunksize units. Whenever we wish to dynamically allocate a variable, we + * always allocate a single chunk of memory. Depending on the uniformity of + * allocation, this will waste some amount of memory -- but it eliminates the + * non-determinism inherent in traditional heap fragmentation. + * + * Dynamic objects are allocated by storing a non-zero value to them; they are + * deallocated by storing a zero value to them. Dynamic variables are + * complicated enormously by being shared between CPUs. In particular, + * consider the following scenario: + * + * CPU A CPU B + * +---------------------------------+ +---------------------------------+ + * | | | | + * | allocates dynamic object a[123] | | | + * | by storing the value 345 to it | | | + * | ---------> | + * | | | wishing to load from object | + * | | | a[123], performs lookup in | + * | | | dynamic variable space | + * | <--------- | + * | deallocates object a[123] by | | | + * | storing 0 to it | | | + * | | | | + * | allocates dynamic object b[567] | | performs load from a[123] | + * | by storing the value 789 to it | | | + * : : : : + * . . . . + * + * This is obviously a race in the D program, but there are nonetheless only + * two valid values for CPU B's load from a[123]: 345 or 0. Most importantly, + * CPU B may _not_ see the value 789 for a[123]. + * + * There are essentially two ways to deal with this: + * + * (1) Explicitly spin-lock variables. That is, if CPU B wishes to load + * from a[123], it needs to lock a[123] and hold the lock for the + * duration that it wishes to manipulate it. + * + * (2) Avoid reusing freed chunks until it is known that no CPU is referring + * to them. + * + * The implementation of (1) is rife with complexity, because it requires the + * user of a dynamic variable to explicitly decree when they are done using it. + * Were all variables by value, this perhaps wouldn't be debilitating -- but + * dynamic variables of non-scalar types are tracked by reference. That is, if + * a dynamic variable is, say, a string, and that variable is to be traced to, + * say, the principal buffer, the DIF emulation code returns to the main + * dtrace_probe() loop a pointer to the underlying storage, not the contents of + * the storage. Further, code calling on DIF emulation would have to be aware + * that the DIF emulation has returned a reference to a dynamic variable that + * has been potentially locked. The variable would have to be unlocked after + * the main dtrace_probe() loop is finished with the variable, and the main + * dtrace_probe() loop would have to be careful to not call any further DIF + * emulation while the variable is locked to avoid deadlock. More generally, + * if one were to implement (1), DIF emulation code dealing with dynamic + * variables could only deal with one dynamic variable at a time (lest deadlock + * result). To sum, (1) exports too much subtlety to the users of dynamic + * variables -- increasing maintenance burden and imposing serious constraints + * on future DTrace development. + * + * The implementation of (2) is also complex, but the complexity is more + * manageable. We need to be sure that when a variable is deallocated, it is + * not placed on a traditional free list, but rather on a _dirty_ list. Once a + * variable is on a dirty list, it cannot be found by CPUs performing a + * subsequent lookup of the variable -- but it may still be in use by other + * CPUs. To assure that all CPUs that may be seeing the old variable have + * cleared out of probe context, a dtrace_sync() can be issued. Once the + * dtrace_sync() has completed, it can be known that all CPUs are done + * manipulating the dynamic variable -- the dirty list can be atomically + * appended to the free list. Unfortunately, there's a slight hiccup in this + * mechanism: dtrace_sync() may not be issued from probe context. The + * dtrace_sync() must be therefore issued asynchronously from non-probe + * context. For this we rely on the DTrace cleaner, a cyclic that runs at the + * "cleanrate" frequency. To ease this implementation, we define several chunk + * lists: + * + * - Dirty. Deallocated chunks, not yet cleaned. Not available. + * + * - Rinsing. Formerly dirty chunks that are currently being asynchronously + * cleaned. Not available, but will be shortly. Dynamic variable + * allocation may not spin or block for availability, however. + * + * - Clean. Clean chunks, ready for allocation -- but not on the free list. + * + * - Free. Available for allocation. + * + * Moreover, to avoid absurd contention, _each_ of these lists is implemented + * on a per-CPU basis. This is only for performance, not correctness; chunks + * may be allocated from another CPU's free list. The algorithm for allocation + * then is this: + * + * (1) Attempt to atomically allocate from current CPU's free list. If list + * is non-empty and allocation is successful, allocation is complete. + * + * (2) If the clean list is non-empty, atomically move it to the free list, + * and reattempt (1). + * + * (3) If the dynamic variable space is in the CLEAN state, look for free + * and clean lists on other CPUs by setting the current CPU to the next + * CPU, and reattempting (1). If the next CPU is the current CPU (that + * is, if all CPUs have been checked), atomically switch the state of + * the dynamic variable space based on the following: + * + * - If no free chunks were found and no dirty chunks were found, + * atomically set the state to EMPTY. + * + * - If dirty chunks were found, atomically set the state to DIRTY. + * + * - If rinsing chunks were found, atomically set the state to RINSING. + * + * (4) Based on state of dynamic variable space state, increment appropriate + * counter to indicate dynamic drops (if in EMPTY state) vs. dynamic + * dirty drops (if in DIRTY state) vs. dynamic rinsing drops (if in + * RINSING state). Fail the allocation. + * + * The cleaning cyclic operates with the following algorithm: for all CPUs + * with a non-empty dirty list, atomically move the dirty list to the rinsing + * list. Perform a dtrace_sync(). For all CPUs with a non-empty rinsing list, + * atomically move the rinsing list to the clean list. Perform another + * dtrace_sync(). By this point, all CPUs have seen the new clean list; the + * state of the dynamic variable space can be restored to CLEAN. + * + * There exist two final races that merit explanation. The first is a simple + * allocation race: + * + * CPU A CPU B + * +---------------------------------+ +---------------------------------+ + * | | | | + * | allocates dynamic object a[123] | | allocates dynamic object a[123] | + * | by storing the value 345 to it | | by storing the value 567 to it | + * | | | | + * : : : : + * . . . . + * + * Again, this is a race in the D program. It can be resolved by having a[123] + * hold the value 345 or a[123] hold the value 567 -- but it must be true that + * a[123] have only _one_ of these values. (That is, the racing CPUs may not + * put the same element twice on the same hash chain.) This is resolved + * simply: before the allocation is undertaken, the start of the new chunk's + * hash chain is noted. Later, after the allocation is complete, the hash + * chain is atomically switched to point to the new element. If this fails + * (because of either concurrent allocations or an allocation concurrent with a + * deletion), the newly allocated chunk is deallocated to the dirty list, and + * the whole process of looking up (and potentially allocating) the dynamic + * variable is reattempted. + * + * The final race is a simple deallocation race: + * + * CPU A CPU B + * +---------------------------------+ +---------------------------------+ + * | | | | + * | deallocates dynamic object | | deallocates dynamic object | + * | a[123] by storing the value 0 | | a[123] by storing the value 0 | + * | to it | | to it | + * | | | | + * : : : : + * . . . . + * + * Once again, this is a race in the D program, but it is one that we must + * handle without corrupting the underlying data structures. Because + * deallocations require the deletion of a chunk from the middle of a hash + * chain, we cannot use a single-word atomic operation to remove it. For this, + * we add a spin lock to the hash buckets that is _only_ used for deallocations + * (allocation races are handled as above). Further, this spin lock is _only_ + * held for the duration of the delete; before control is returned to the DIF + * emulation code, the hash bucket is unlocked. + */ +typedef struct dtrace_key { + uint64_t dttk_value; /* data value or data pointer */ + uint64_t dttk_size; /* 0 if by-val, >0 if by-ref */ +} dtrace_key_t; + +typedef struct dtrace_tuple { + uint32_t dtt_nkeys; /* number of keys in tuple */ + uint32_t dtt_pad; /* padding */ + dtrace_key_t dtt_key[1]; /* array of tuple keys */ +} dtrace_tuple_t; + +typedef struct dtrace_dynvar { + uint64_t dtdv_hashval; /* hash value -- 0 if free */ + struct dtrace_dynvar *dtdv_next; /* next on list or hash chain */ + void *dtdv_data; /* pointer to data */ + dtrace_tuple_t dtdv_tuple; /* tuple key */ +} dtrace_dynvar_t; + +typedef enum dtrace_dynvar_op { + DTRACE_DYNVAR_ALLOC, + DTRACE_DYNVAR_NOALLOC, + DTRACE_DYNVAR_DEALLOC +} dtrace_dynvar_op_t; + +typedef struct dtrace_dynhash { + dtrace_dynvar_t *dtdh_chain; /* hash chain for this bucket */ + uintptr_t dtdh_lock; /* deallocation lock */ +#ifdef _LP64 + uintptr_t dtdh_pad[6]; /* pad to avoid false sharing */ +#else + uintptr_t dtdh_pad[14]; /* pad to avoid false sharing */ +#endif +} dtrace_dynhash_t; + +typedef struct dtrace_dstate_percpu { + dtrace_dynvar_t *dtdsc_free; /* free list for this CPU */ + dtrace_dynvar_t *dtdsc_dirty; /* dirty list for this CPU */ + dtrace_dynvar_t *dtdsc_rinsing; /* rinsing list for this CPU */ + dtrace_dynvar_t *dtdsc_clean; /* clean list for this CPU */ + uint64_t dtdsc_drops; /* number of capacity drops */ + uint64_t dtdsc_dirty_drops; /* number of dirty drops */ + uint64_t dtdsc_rinsing_drops; /* number of rinsing drops */ +#ifdef _LP64 + uint64_t dtdsc_pad; /* pad to avoid false sharing */ +#else + uint64_t dtdsc_pad[2]; /* pad to avoid false sharing */ +#endif +} dtrace_dstate_percpu_t; + +typedef enum dtrace_dstate_state { + DTRACE_DSTATE_CLEAN = 0, + DTRACE_DSTATE_EMPTY, + DTRACE_DSTATE_DIRTY, + DTRACE_DSTATE_RINSING +} dtrace_dstate_state_t; + +typedef struct dtrace_dstate { + void *dtds_base; /* base of dynamic var. space */ + size_t dtds_size; /* size of dynamic var. space */ + size_t dtds_hashsize; /* number of buckets in hash */ + size_t dtds_chunksize; /* size of each chunk */ + dtrace_dynhash_t *dtds_hash; /* pointer to hash table */ + dtrace_dstate_state_t dtds_state; /* current dynamic var. state */ + dtrace_dstate_percpu_t *dtds_percpu; /* per-CPU dyn. var. state */ +} dtrace_dstate_t; + +/* + * DTrace Variable State + * + * The DTrace variable state tracks user-defined variables in its dtrace_vstate + * structure. Each DTrace consumer has exactly one dtrace_vstate structure, + * but some dtrace_vstate structures may exist without a corresponding DTrace + * consumer (see "DTrace Helpers", below). As described in <sys/dtrace.h>, + * user-defined variables can have one of three scopes: + * + * DIFV_SCOPE_GLOBAL => global scope + * DIFV_SCOPE_THREAD => thread-local scope (i.e. "self->" variables) + * DIFV_SCOPE_LOCAL => clause-local scope (i.e. "this->" variables) + * + * The variable state tracks variables by both their scope and their allocation + * type: + * + * - The dtvs_globals and dtvs_locals members each point to an array of + * dtrace_statvar structures. These structures contain both the variable + * metadata (dtrace_difv structures) and the underlying storage for all + * statically allocated variables, including statically allocated + * DIFV_SCOPE_GLOBAL variables and all DIFV_SCOPE_LOCAL variables. + * + * - The dtvs_tlocals member points to an array of dtrace_difv structures for + * DIFV_SCOPE_THREAD variables. As such, this array tracks _only_ the + * variable metadata for DIFV_SCOPE_THREAD variables; the underlying storage + * is allocated out of the dynamic variable space. + * + * - The dtvs_dynvars member is the dynamic variable state associated with the + * variable state. The dynamic variable state (described in "DTrace Dynamic + * Variables", above) tracks all DIFV_SCOPE_THREAD variables and all + * dynamically-allocated DIFV_SCOPE_GLOBAL variables. + */ +typedef struct dtrace_statvar { + uint64_t dtsv_data; /* data or pointer to it */ + size_t dtsv_size; /* size of pointed-to data */ + int dtsv_refcnt; /* reference count */ + dtrace_difv_t dtsv_var; /* variable metadata */ +} dtrace_statvar_t; + +typedef struct dtrace_vstate { + dtrace_state_t *dtvs_state; /* back pointer to state */ + dtrace_statvar_t **dtvs_globals; /* statically-allocated glbls */ + int dtvs_nglobals; /* number of globals */ + dtrace_difv_t *dtvs_tlocals; /* thread-local metadata */ + int dtvs_ntlocals; /* number of thread-locals */ + dtrace_statvar_t **dtvs_locals; /* clause-local data */ + int dtvs_nlocals; /* number of clause-locals */ + dtrace_dstate_t dtvs_dynvars; /* dynamic variable state */ +} dtrace_vstate_t; + +/* + * DTrace Machine State + * + * In the process of processing a fired probe, DTrace needs to track and/or + * cache some per-CPU state associated with that particular firing. This is + * state that is always discarded after the probe firing has completed, and + * much of it is not specific to any DTrace consumer, remaining valid across + * all ECBs. This state is tracked in the dtrace_mstate structure. + */ +#define DTRACE_MSTATE_ARGS 0x00000001 +#define DTRACE_MSTATE_PROBE 0x00000002 +#define DTRACE_MSTATE_EPID 0x00000004 +#define DTRACE_MSTATE_TIMESTAMP 0x00000008 +#define DTRACE_MSTATE_STACKDEPTH 0x00000010 +#define DTRACE_MSTATE_CALLER 0x00000020 +#define DTRACE_MSTATE_IPL 0x00000040 +#define DTRACE_MSTATE_FLTOFFS 0x00000080 +#define DTRACE_MSTATE_WALLTIMESTAMP 0x00000100 +#define DTRACE_MSTATE_USTACKDEPTH 0x00000200 +#define DTRACE_MSTATE_UCALLER 0x00000400 + +typedef struct dtrace_mstate { + uintptr_t dtms_scratch_base; /* base of scratch space */ + uintptr_t dtms_scratch_ptr; /* current scratch pointer */ + size_t dtms_scratch_size; /* scratch size */ + uint32_t dtms_present; /* variables that are present */ + uint64_t dtms_arg[5]; /* cached arguments */ + dtrace_epid_t dtms_epid; /* current EPID */ + uint64_t dtms_timestamp; /* cached timestamp */ + hrtime_t dtms_walltimestamp; /* cached wall timestamp */ + int dtms_stackdepth; /* cached stackdepth */ + int dtms_ustackdepth; /* cached ustackdepth */ + struct dtrace_probe *dtms_probe; /* current probe */ + uintptr_t dtms_caller; /* cached caller */ + uint64_t dtms_ucaller; /* cached user-level caller */ + int dtms_ipl; /* cached interrupt pri lev */ + int dtms_fltoffs; /* faulting DIFO offset */ + uintptr_t dtms_strtok; /* saved strtok() pointer */ + uint32_t dtms_access; /* memory access rights */ + dtrace_difo_t *dtms_difo; /* current dif object */ +} dtrace_mstate_t; + +#define DTRACE_COND_OWNER 0x1 +#define DTRACE_COND_USERMODE 0x2 +#define DTRACE_COND_ZONEOWNER 0x4 + +#define DTRACE_PROBEKEY_MAXDEPTH 8 /* max glob recursion depth */ + +/* + * Access flag used by dtrace_mstate.dtms_access. + */ +#define DTRACE_ACCESS_KERNEL 0x1 /* the priv to read kmem */ + + +/* + * DTrace Activity + * + * Each DTrace consumer is in one of several states, which (for purposes of + * avoiding yet-another overloading of the noun "state") we call the current + * _activity_. The activity transitions on dtrace_go() (from DTRACIOCGO), on + * dtrace_stop() (from DTRACIOCSTOP) and on the exit() action. Activities may + * only transition in one direction; the activity transition diagram is a + * directed acyclic graph. The activity transition diagram is as follows: + * + * + * +----------+ +--------+ +--------+ + * | INACTIVE |------------------>| WARMUP |------------------>| ACTIVE | + * +----------+ dtrace_go(), +--------+ dtrace_go(), +--------+ + * before BEGIN | after BEGIN | | | + * | | | | + * exit() action | | | | + * from BEGIN ECB | | | | + * | | | | + * v | | | + * +----------+ exit() action | | | + * +-----------------------------| DRAINING |<-------------------+ | | + * | +----------+ | | + * | | | | + * | dtrace_stop(), | | | + * | before END | | | + * | | | | + * | v | | + * | +---------+ +----------+ | | + * | | STOPPED |<----------------| COOLDOWN |<----------------------+ | + * | +---------+ dtrace_stop(), +----------+ dtrace_stop(), | + * | after END before END | + * | | + * | +--------+ | + * +----------------------------->| KILLED |<--------------------------+ + * deadman timeout or +--------+ deadman timeout or + * killed consumer killed consumer + * + * Note that once a DTrace consumer has stopped tracing, there is no way to + * restart it; if a DTrace consumer wishes to restart tracing, it must reopen + * the DTrace pseudodevice. + */ +typedef enum dtrace_activity { + DTRACE_ACTIVITY_INACTIVE = 0, /* not yet running */ + DTRACE_ACTIVITY_WARMUP, /* while starting */ + DTRACE_ACTIVITY_ACTIVE, /* running */ + DTRACE_ACTIVITY_DRAINING, /* before stopping */ + DTRACE_ACTIVITY_COOLDOWN, /* while stopping */ + DTRACE_ACTIVITY_STOPPED, /* after stopping */ + DTRACE_ACTIVITY_KILLED /* killed */ +} dtrace_activity_t; + +/* + * DTrace Helper Implementation + * + * A description of the helper architecture may be found in <sys/dtrace.h>. + * Each process contains a pointer to its helpers in its p_dtrace_helpers + * member. This is a pointer to a dtrace_helpers structure, which contains an + * array of pointers to dtrace_helper structures, helper variable state (shared + * among a process's helpers) and a generation count. (The generation count is + * used to provide an identifier when a helper is added so that it may be + * subsequently removed.) The dtrace_helper structure is self-explanatory, + * containing pointers to the objects needed to execute the helper. Note that + * helpers are _duplicated_ across fork(2), and destroyed on exec(2). No more + * than dtrace_helpers_max are allowed per-process. + */ +#define DTRACE_HELPER_ACTION_USTACK 0 +#define DTRACE_NHELPER_ACTIONS 1 + +typedef struct dtrace_helper_action { + int dtha_generation; /* helper action generation */ + int dtha_nactions; /* number of actions */ + dtrace_difo_t *dtha_predicate; /* helper action predicate */ + dtrace_difo_t **dtha_actions; /* array of actions */ + struct dtrace_helper_action *dtha_next; /* next helper action */ +} dtrace_helper_action_t; + +typedef struct dtrace_helper_provider { + int dthp_generation; /* helper provider generation */ + uint32_t dthp_ref; /* reference count */ + dof_helper_t dthp_prov; /* DOF w/ provider and probes */ +} dtrace_helper_provider_t; + +typedef struct dtrace_helpers { + dtrace_helper_action_t **dthps_actions; /* array of helper actions */ + dtrace_vstate_t dthps_vstate; /* helper action var. state */ + dtrace_helper_provider_t **dthps_provs; /* array of providers */ + uint_t dthps_nprovs; /* count of providers */ + uint_t dthps_maxprovs; /* provider array size */ + int dthps_generation; /* current generation */ + pid_t dthps_pid; /* pid of associated proc */ + int dthps_deferred; /* helper in deferred list */ + struct dtrace_helpers *dthps_next; /* next pointer */ + struct dtrace_helpers *dthps_prev; /* prev pointer */ +} dtrace_helpers_t; + +/* + * DTrace Helper Action Tracing + * + * Debugging helper actions can be arduous. To ease the development and + * debugging of helpers, DTrace contains a tracing-framework-within-a-tracing- + * framework: helper tracing. If dtrace_helptrace_enabled is non-zero (which + * it is by default on DEBUG kernels), all helper activity will be traced to a + * global, in-kernel ring buffer. Each entry includes a pointer to the specific + * helper, the location within the helper, and a trace of all local variables. + * The ring buffer may be displayed in a human-readable format with the + * ::dtrace_helptrace mdb(1) dcmd. + */ +#define DTRACE_HELPTRACE_NEXT (-1) +#define DTRACE_HELPTRACE_DONE (-2) +#define DTRACE_HELPTRACE_ERR (-3) + +typedef struct dtrace_helptrace { + dtrace_helper_action_t *dtht_helper; /* helper action */ + int dtht_where; /* where in helper action */ + int dtht_nlocals; /* number of locals */ + int dtht_fault; /* type of fault (if any) */ + int dtht_fltoffs; /* DIF offset */ + uint64_t dtht_illval; /* faulting value */ + uint64_t dtht_locals[1]; /* local variables */ +} dtrace_helptrace_t; + +/* + * DTrace Credentials + * + * In probe context, we have limited flexibility to examine the credentials + * of the DTrace consumer that created a particular enabling. We use + * the Least Privilege interfaces to cache the consumer's cred pointer and + * some facts about that credential in a dtrace_cred_t structure. These + * can limit the consumer's breadth of visibility and what actions the + * consumer may take. + */ +#define DTRACE_CRV_ALLPROC 0x01 +#define DTRACE_CRV_KERNEL 0x02 +#define DTRACE_CRV_ALLZONE 0x04 + +#define DTRACE_CRV_ALL (DTRACE_CRV_ALLPROC | DTRACE_CRV_KERNEL | \ + DTRACE_CRV_ALLZONE) + +#define DTRACE_CRA_PROC 0x0001 +#define DTRACE_CRA_PROC_CONTROL 0x0002 +#define DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER 0x0004 +#define DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE 0x0008 +#define DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG 0x0010 +#define DTRACE_CRA_KERNEL 0x0020 +#define DTRACE_CRA_KERNEL_DESTRUCTIVE 0x0040 + +#define DTRACE_CRA_ALL (DTRACE_CRA_PROC | \ + DTRACE_CRA_PROC_CONTROL | \ + DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER | \ + DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE | \ + DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG | \ + DTRACE_CRA_KERNEL | \ + DTRACE_CRA_KERNEL_DESTRUCTIVE) + +typedef struct dtrace_cred { + cred_t *dcr_cred; + uint8_t dcr_destructive; + uint8_t dcr_visible; + uint16_t dcr_action; +} dtrace_cred_t; + +/* + * DTrace Consumer State + * + * Each DTrace consumer has an associated dtrace_state structure that contains + * its in-kernel DTrace state -- including options, credentials, statistics and + * pointers to ECBs, buffers, speculations and formats. A dtrace_state + * structure is also allocated for anonymous enablings. When anonymous state + * is grabbed, the grabbing consumers dts_anon pointer is set to the grabbed + * dtrace_state structure. + */ +struct dtrace_state { +#if defined(sun) + dev_t dts_dev; /* device */ +#else + struct cdev *dts_dev; /* device */ +#endif + int dts_necbs; /* total number of ECBs */ + dtrace_ecb_t **dts_ecbs; /* array of ECBs */ + dtrace_epid_t dts_epid; /* next EPID to allocate */ + size_t dts_needed; /* greatest needed space */ + struct dtrace_state *dts_anon; /* anon. state, if grabbed */ + dtrace_activity_t dts_activity; /* current activity */ + dtrace_vstate_t dts_vstate; /* variable state */ + dtrace_buffer_t *dts_buffer; /* principal buffer */ + dtrace_buffer_t *dts_aggbuffer; /* aggregation buffer */ + dtrace_speculation_t *dts_speculations; /* speculation array */ + int dts_nspeculations; /* number of speculations */ + int dts_naggregations; /* number of aggregations */ + dtrace_aggregation_t **dts_aggregations; /* aggregation array */ +#if defined(sun) + vmem_t *dts_aggid_arena; /* arena for aggregation IDs */ +#else + struct unrhdr *dts_aggid_arena; /* arena for aggregation IDs */ +#endif + uint64_t dts_errors; /* total number of errors */ + uint32_t dts_speculations_busy; /* number of spec. busy */ + uint32_t dts_speculations_unavail; /* number of spec unavail */ + uint32_t dts_stkstroverflows; /* stack string tab overflows */ + uint32_t dts_dblerrors; /* errors in ERROR probes */ + uint32_t dts_reserve; /* space reserved for END */ + hrtime_t dts_laststatus; /* time of last status */ +#if defined(sun) + cyclic_id_t dts_cleaner; /* cleaning cyclic */ + cyclic_id_t dts_deadman; /* deadman cyclic */ +#else + struct callout dts_cleaner; /* Cleaning callout. */ + struct callout dts_deadman; /* Deadman callout. */ +#endif + hrtime_t dts_alive; /* time last alive */ + char dts_speculates; /* boolean: has speculations */ + char dts_destructive; /* boolean: has dest. actions */ + int dts_nformats; /* number of formats */ + char **dts_formats; /* format string array */ + dtrace_optval_t dts_options[DTRACEOPT_MAX]; /* options */ + dtrace_cred_t dts_cred; /* credentials */ + size_t dts_nretained; /* number of retained enabs */ +}; + +struct dtrace_provider { + dtrace_pattr_t dtpv_attr; /* provider attributes */ + dtrace_ppriv_t dtpv_priv; /* provider privileges */ + dtrace_pops_t dtpv_pops; /* provider operations */ + char *dtpv_name; /* provider name */ + void *dtpv_arg; /* provider argument */ + uint_t dtpv_defunct; /* boolean: defunct provider */ + struct dtrace_provider *dtpv_next; /* next provider */ +}; + +struct dtrace_meta { + dtrace_mops_t dtm_mops; /* meta provider operations */ + char *dtm_name; /* meta provider name */ + void *dtm_arg; /* meta provider user arg */ + uint64_t dtm_count; /* no. of associated provs. */ +}; + +/* + * DTrace Enablings + * + * A dtrace_enabling structure is used to track a collection of ECB + * descriptions -- before they have been turned into actual ECBs. This is + * created as a result of DOF processing, and is generally used to generate + * ECBs immediately thereafter. However, enablings are also generally + * retained should the probes they describe be created at a later time; as + * each new module or provider registers with the framework, the retained + * enablings are reevaluated, with any new match resulting in new ECBs. To + * prevent probes from being matched more than once, the enabling tracks the + * last probe generation matched, and only matches probes from subsequent + * generations. + */ +typedef struct dtrace_enabling { + dtrace_ecbdesc_t **dten_desc; /* all ECB descriptions */ + int dten_ndesc; /* number of ECB descriptions */ + int dten_maxdesc; /* size of ECB array */ + dtrace_vstate_t *dten_vstate; /* associated variable state */ + dtrace_genid_t dten_probegen; /* matched probe generation */ + dtrace_ecbdesc_t *dten_current; /* current ECB description */ + int dten_error; /* current error value */ + int dten_primed; /* boolean: set if primed */ + struct dtrace_enabling *dten_prev; /* previous enabling */ + struct dtrace_enabling *dten_next; /* next enabling */ +} dtrace_enabling_t; + +/* + * DTrace Anonymous Enablings + * + * Anonymous enablings are DTrace enablings that are not associated with a + * controlling process, but rather derive their enabling from DOF stored as + * properties in the dtrace.conf file. If there is an anonymous enabling, a + * DTrace consumer state and enabling are created on attach. The state may be + * subsequently grabbed by the first consumer specifying the "grabanon" + * option. As long as an anonymous DTrace enabling exists, dtrace(7D) will + * refuse to unload. + */ +typedef struct dtrace_anon { + dtrace_state_t *dta_state; /* DTrace consumer state */ + dtrace_enabling_t *dta_enabling; /* pointer to enabling */ + processorid_t dta_beganon; /* which CPU BEGIN ran on */ +} dtrace_anon_t; + +/* + * DTrace Error Debugging + */ +#ifdef DEBUG +#define DTRACE_ERRDEBUG +#endif + +#ifdef DTRACE_ERRDEBUG + +typedef struct dtrace_errhash { + const char *dter_msg; /* error message */ + int dter_count; /* number of times seen */ +} dtrace_errhash_t; + +#define DTRACE_ERRHASHSZ 256 /* must be > number of err msgs */ + +#endif /* DTRACE_ERRDEBUG */ + +/* + * DTrace Toxic Ranges + * + * DTrace supports safe loads from probe context; if the address turns out to + * be invalid, a bit will be set by the kernel indicating that DTrace + * encountered a memory error, and DTrace will propagate the error to the user + * accordingly. However, there may exist some regions of memory in which an + * arbitrary load can change system state, and from which it is impossible to + * recover from such a load after it has been attempted. Examples of this may + * include memory in which programmable I/O registers are mapped (for which a + * read may have some implications for the device) or (in the specific case of + * UltraSPARC-I and -II) the virtual address hole. The platform is required + * to make DTrace aware of these toxic ranges; DTrace will then check that + * target addresses are not in a toxic range before attempting to issue a + * safe load. + */ +typedef struct dtrace_toxrange { + uintptr_t dtt_base; /* base of toxic range */ + uintptr_t dtt_limit; /* limit of toxic range */ +} dtrace_toxrange_t; + +extern uint64_t dtrace_getarg(int, int); +extern greg_t dtrace_getfp(void); +extern int dtrace_getipl(void); +extern uintptr_t dtrace_caller(int); +extern uint32_t dtrace_cas32(uint32_t *, uint32_t, uint32_t); +extern void *dtrace_casptr(volatile void *, volatile void *, volatile void *); +extern void dtrace_copyin(uintptr_t, uintptr_t, size_t, volatile uint16_t *); +extern void dtrace_copyinstr(uintptr_t, uintptr_t, size_t, volatile uint16_t *); +extern void dtrace_copyout(uintptr_t, uintptr_t, size_t, volatile uint16_t *); +extern void dtrace_copyoutstr(uintptr_t, uintptr_t, size_t, + volatile uint16_t *); +extern void dtrace_getpcstack(pc_t *, int, int, uint32_t *); +extern ulong_t dtrace_getreg(struct regs *, uint_t); +extern int dtrace_getstackdepth(int); +extern void dtrace_getupcstack(uint64_t *, int); +extern void dtrace_getufpstack(uint64_t *, uint64_t *, int); +extern int dtrace_getustackdepth(void); +extern uintptr_t dtrace_fulword(void *); +extern uint8_t dtrace_fuword8(void *); +extern uint16_t dtrace_fuword16(void *); +extern uint32_t dtrace_fuword32(void *); +extern uint64_t dtrace_fuword64(void *); +extern void dtrace_probe_error(dtrace_state_t *, dtrace_epid_t, int, int, + int, uintptr_t); +extern int dtrace_assfail(const char *, const char *, int); +extern int dtrace_attached(void); +#if defined(sun) +extern hrtime_t dtrace_gethrestime(void); +#endif + +#ifdef __sparc +extern void dtrace_flush_windows(void); +extern void dtrace_flush_user_windows(void); +extern uint_t dtrace_getotherwin(void); +extern uint_t dtrace_getfprs(void); +#else +extern void dtrace_copy(uintptr_t, uintptr_t, size_t); +extern void dtrace_copystr(uintptr_t, uintptr_t, size_t, volatile uint16_t *); +#endif + +/* + * DTrace Assertions + * + * DTrace calls ASSERT from probe context. To assure that a failed ASSERT + * does not induce a markedly more catastrophic failure (e.g., one from which + * a dump cannot be gleaned), DTrace must define its own ASSERT to be one that + * may safely be called from probe context. This header file must thus be + * included by any DTrace component that calls ASSERT from probe context, and + * _only_ by those components. (The only exception to this is kernel + * debugging infrastructure at user-level that doesn't depend on calling + * ASSERT.) + */ +#undef ASSERT +#ifdef DEBUG +#define ASSERT(EX) ((void)((EX) || \ + dtrace_assfail(#EX, __FILE__, __LINE__))) +#else +#define ASSERT(X) ((void)0) +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* _SYS_DTRACE_IMPL_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/fasttrap.h b/sys/cddl/contrib/opensolaris/uts/common/sys/fasttrap.h new file mode 100644 index 000000000000..7f803144bcf8 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/fasttrap.h @@ -0,0 +1,93 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _SYS_FASTTRAP_H +#define _SYS_FASTTRAP_H + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/fasttrap_isa.h> +#include <sys/dtrace.h> +#include <sys/types.h> + +#ifdef __cplusplus +extern "C" { +#endif + +#define FASTTRAPIOC (('m' << 24) | ('r' << 16) | ('f' << 8)) +#define FASTTRAPIOC_MAKEPROBE (FASTTRAPIOC | 1) +#define FASTTRAPIOC_GETINSTR (FASTTRAPIOC | 2) + +typedef enum fasttrap_probe_type { + DTFTP_NONE = 0, + DTFTP_ENTRY, + DTFTP_RETURN, + DTFTP_OFFSETS, + DTFTP_POST_OFFSETS, + DTFTP_IS_ENABLED +} fasttrap_probe_type_t; + +typedef struct fasttrap_probe_spec { + pid_t ftps_pid; + fasttrap_probe_type_t ftps_type; + + char ftps_func[DTRACE_FUNCNAMELEN]; + char ftps_mod[DTRACE_MODNAMELEN]; + + uint64_t ftps_pc; + uint64_t ftps_size; + uint64_t ftps_noffs; + uint64_t ftps_offs[1]; +} fasttrap_probe_spec_t; + +typedef struct fasttrap_instr_query { + uint64_t ftiq_pc; + pid_t ftiq_pid; + fasttrap_instr_t ftiq_instr; +} fasttrap_instr_query_t; + +/* + * To support the fasttrap provider from very early in a process's life, + * the run-time linker, ld.so.1, has a program header of type PT_SUNWDTRACE + * which points to a data object which must be PT_SUNWDTRACE_SIZE bytes. + * This structure mimics the fasttrap provider section of the ulwp_t structure. + * When the fasttrap provider is changed to require new or different + * instructions, the data object in ld.so.1 and the thread initializers in libc + * (libc_init() and _thrp_create()) need to be updated to include the new + * instructions, and PT_SUNWDTRACE needs to be changed to a new unique number + * (while the old value gets assigned something like PT_SUNWDTRACE_1). Since the + * linker must be backward compatible with old Solaris releases, it must have + * program headers for each of the PT_SUNWDTRACE versions. The kernel's + * elfexec() function only has to look for the latest version of the + * PT_SUNWDTRACE program header. + */ +#define PT_SUNWDTRACE_SIZE FASTTRAP_SUNWDTRACE_SIZE + +#ifdef __cplusplus +} +#endif + +#endif /* _SYS_FASTTRAP_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/fasttrap_impl.h b/sys/cddl/contrib/opensolaris/uts/common/sys/fasttrap_impl.h new file mode 100644 index 000000000000..46b92caba695 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/fasttrap_impl.h @@ -0,0 +1,205 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _FASTTRAP_IMPL_H +#define _FASTTRAP_IMPL_H + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/types.h> +#include <sys/dtrace.h> +#include <sys/proc.h> +#include <sys/fasttrap.h> +#include <sys/fasttrap_isa.h> + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * Fasttrap Providers, Probes and Tracepoints + * + * Each Solaris process can have multiple providers -- the pid provider as + * well as any number of user-level statically defined tracing (USDT) + * providers. Those providers are each represented by a fasttrap_provider_t. + * All providers for a given process have a pointer to a shared + * fasttrap_proc_t. The fasttrap_proc_t has two states: active or defunct. + * When the count of live providers goes to zero it becomes defunct; providers + * drop their count when they are removed individually or en masse when a + * process exits or performs an exec. + * + * Each probe is represented by a fasttrap_probe_t which has a pointer to + * its associated provider as well as a list of fasttrap_id_tp_t structures + * which are tuples combining a fasttrap_id_t and a fasttrap_tracepoint_t. + * A fasttrap_tracepoint_t represents the actual point of instrumentation + * and it contains two lists of fasttrap_id_t structures (to be fired pre- + * and post-instruction emulation) that identify the probes attached to the + * tracepoint. Tracepoints also have a pointer to the fasttrap_proc_t for the + * process they trace which is used when looking up a tracepoint both at + * probe fire time and when enabling and disabling probes. + * + * It's important to note that probes are preallocated with the necessary + * number of tracepoints, but that tracepoints can be shared by probes and + * swapped between probes. If a probe's preallocated tracepoint is enabled + * (and, therefore, the associated probe is enabled), and that probe is + * then disabled, ownership of that tracepoint may be exchanged for an + * unused tracepoint belonging to another probe that was attached to the + * enabled tracepoint. + */ + +typedef struct fasttrap_proc { + pid_t ftpc_pid; /* process ID for this proc */ + uint64_t ftpc_acount; /* count of active providers */ + uint64_t ftpc_rcount; /* provider reference count */ +#if defined(sun) + kmutex_t ftpc_mtx; /* proc lock */ +#else + struct mtx ftpc_mtx; /* proc lock */ +#endif + struct fasttrap_proc *ftpc_next; /* next proc in hash chain */ +} fasttrap_proc_t; + +typedef struct fasttrap_provider { + pid_t ftp_pid; /* process ID for this prov */ + char ftp_name[DTRACE_PROVNAMELEN]; /* prov name (w/o the pid) */ + dtrace_provider_id_t ftp_provid; /* DTrace provider handle */ + uint_t ftp_marked; /* mark for possible removal */ + uint_t ftp_retired; /* mark when retired */ +#if defined(sun) + kmutex_t ftp_mtx; /* provider lock */ + kmutex_t ftp_cmtx; /* lock on creating probes */ +#else + struct mtx ftp_mtx; /* provider lock */ + struct mtx ftp_cmtx; /* lock on creating probes */ +#endif + uint64_t ftp_rcount; /* enabled probes ref count */ + uint64_t ftp_ccount; /* consumers creating probes */ + uint64_t ftp_mcount; /* meta provider count */ + fasttrap_proc_t *ftp_proc; /* shared proc for all provs */ + struct fasttrap_provider *ftp_next; /* next prov in hash chain */ +} fasttrap_provider_t; + +typedef struct fasttrap_id fasttrap_id_t; +typedef struct fasttrap_probe fasttrap_probe_t; +typedef struct fasttrap_tracepoint fasttrap_tracepoint_t; + +struct fasttrap_id { + fasttrap_probe_t *fti_probe; /* referrring probe */ + fasttrap_id_t *fti_next; /* enabled probe list on tp */ + fasttrap_probe_type_t fti_ptype; /* probe type */ +}; + +typedef struct fasttrap_id_tp { + fasttrap_id_t fit_id; + fasttrap_tracepoint_t *fit_tp; +} fasttrap_id_tp_t; + +struct fasttrap_probe { + dtrace_id_t ftp_id; /* DTrace probe identifier */ + pid_t ftp_pid; /* pid for this probe */ + fasttrap_provider_t *ftp_prov; /* this probe's provider */ + uintptr_t ftp_faddr; /* associated function's addr */ + size_t ftp_fsize; /* associated function's size */ + uint64_t ftp_gen; /* modification generation */ + uint64_t ftp_ntps; /* number of tracepoints */ + uint8_t *ftp_argmap; /* native to translated args */ + uint8_t ftp_nargs; /* translated argument count */ + uint8_t ftp_enabled; /* is this probe enabled */ + char *ftp_xtypes; /* translated types index */ + char *ftp_ntypes; /* native types index */ + fasttrap_id_tp_t ftp_tps[1]; /* flexible array */ +}; + +#define FASTTRAP_ID_INDEX(id) \ +((fasttrap_id_tp_t *)(((char *)(id) - offsetof(fasttrap_id_tp_t, fit_id))) - \ +&(id)->fti_probe->ftp_tps[0]) + +struct fasttrap_tracepoint { + fasttrap_proc_t *ftt_proc; /* associated process struct */ + uintptr_t ftt_pc; /* address of tracepoint */ + pid_t ftt_pid; /* pid of tracepoint */ + fasttrap_machtp_t ftt_mtp; /* ISA-specific portion */ + fasttrap_id_t *ftt_ids; /* NULL-terminated list */ + fasttrap_id_t *ftt_retids; /* NULL-terminated list */ + fasttrap_tracepoint_t *ftt_next; /* link in global hash */ +}; + +typedef struct fasttrap_bucket { +#if defined(sun) + kmutex_t ftb_mtx; /* bucket lock */ +#else + struct mtx ftb_mtx; /* bucket lock */ +#endif + void *ftb_data; /* data payload */ + + uint8_t ftb_pad[64 - sizeof (struct mtx) - sizeof (void *)]; +} fasttrap_bucket_t; + +typedef struct fasttrap_hash { + ulong_t fth_nent; /* power-of-2 num. of entries */ + ulong_t fth_mask; /* fth_nent - 1 */ + fasttrap_bucket_t *fth_table; /* array of buckets */ +} fasttrap_hash_t; + +/* + * If at some future point these assembly functions become observable by + * DTrace, then these defines should become separate functions so that the + * fasttrap provider doesn't trigger probes during internal operations. + */ +#define fasttrap_copyout copyout +#define fasttrap_fuword32 fuword32 +#define fasttrap_suword32 suword32 + +#define fasttrap_fulword fulword +#define fasttrap_sulword sulword + +extern void fasttrap_sigtrap(proc_t *, kthread_t *, uintptr_t); + +extern dtrace_id_t fasttrap_probe_id; +extern fasttrap_hash_t fasttrap_tpoints; + +#define FASTTRAP_TPOINTS_INDEX(pid, pc) \ + (((pc) / sizeof (fasttrap_instr_t) + (pid)) & fasttrap_tpoints.fth_mask) + +/* + * Must be implemented by fasttrap_isa.c + */ +extern int fasttrap_tracepoint_init(proc_t *, fasttrap_tracepoint_t *, + uintptr_t, fasttrap_probe_type_t); +extern int fasttrap_tracepoint_install(proc_t *, fasttrap_tracepoint_t *); +extern int fasttrap_tracepoint_remove(proc_t *, fasttrap_tracepoint_t *); + +extern int fasttrap_pid_probe(struct regs *); +extern int fasttrap_return_probe(struct regs *); + +extern uint64_t fasttrap_pid_getarg(void *, dtrace_id_t, void *, int, int); +extern uint64_t fasttrap_usdt_getarg(void *, dtrace_id_t, void *, int, int); + +#ifdef __cplusplus +} +#endif + +#endif /* _FASTTRAP_IMPL_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/isa_defs.h b/sys/cddl/contrib/opensolaris/uts/common/sys/isa_defs.h index 08fcbd02e5c0..b1b99adb8bc4 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/sys/isa_defs.h +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/isa_defs.h @@ -206,6 +206,9 @@ * This indicates that the implementation uses a dynamically * linked unix + krtld to form the core kernel image at boot * time, or (in the absence of this symbol) a prelinked kernel image. + * + * _OBP + * This indicates the firmware interface is OBP. */ #ifdef __cplusplus @@ -216,7 +219,7 @@ extern "C" { * The following set of definitions characterize Solaris on AMD's * 64-bit systems. */ -#if defined(__x86_64) || defined(__amd64) +#if defined(__x86_64) || defined(__amd64) || defined(__ia64__) #if !defined(__amd64) #define __amd64 /* preferred guard */ @@ -229,7 +232,9 @@ extern "C" { /* * Define the appropriate "processor characteristics" */ +#if defined(sun) #define _LITTLE_ENDIAN +#endif #define _STACK_GROWS_DOWNWARD #define _LONG_LONG_LTOH #define _BIT_FIELDS_LTOH @@ -293,7 +298,9 @@ extern "C" { /* * Define the appropriate "processor characteristics" */ +#if defined(sun) #define _LITTLE_ENDIAN +#endif #define _STACK_GROWS_DOWNWARD #define _LONG_LONG_LTOH #define _BIT_FIELDS_LTOH @@ -332,6 +339,51 @@ extern "C" { #define _DONT_USE_1275_GENERIC_NAMES #define _HAVE_CPUID_INSN +#elif defined(__arm__) + +/* + * Define the appropriate "processor characteristics" + */ +#define _STACK_GROWS_DOWNWARD +#define _LONG_LONG_LTOH +#define _BIT_FIELDS_LTOH +#define _IEEE_754 +#define _CHAR_IS_SIGNED +#define _BOOL_ALIGNMENT 1 +#define _CHAR_ALIGNMENT 1 +#define _SHORT_ALIGNMENT 2 +#define _INT_ALIGNMENT 4 +#define _FLOAT_ALIGNMENT 4 +#define _FLOAT_COMPLEX_ALIGNMENT 4 +#define _LONG_ALIGNMENT 4 +#define _LONG_LONG_ALIGNMENT 4 +#define _DOUBLE_ALIGNMENT 4 +#define _DOUBLE_COMPLEX_ALIGNMENT 4 +#define _LONG_DOUBLE_ALIGNMENT 4 +#define _LONG_DOUBLE_COMPLEX_ALIGNMENT 4 +#define _POINTER_ALIGNMENT 4 +#define _MAX_ALIGNMENT 4 +#define _ALIGNMENT_REQUIRED 0 + +#define _LONG_LONG_ALIGNMENT_32 _LONG_LONG_ALIGNMENT + +/* + * Define the appropriate "implementation choices". + */ +#define _ILP32 +#if !defined(_I32LPx) && defined(_KERNEL) +#define _I32LPx +#endif +#define _SUNOS_VTOC_16 +#define _DMA_USES_PHYSADDR +#define _FIRMWARE_NEEDS_FDISK +#define _PSM_MODULES +#define _RTC_CONFIG +#define _DONT_USE_1275_GENERIC_NAMES +#define _HAVE_CPUID_INSN + +#elif defined(__powerpc__) + /* * The following set of definitions characterize the Solaris on SPARC systems. * @@ -378,7 +430,9 @@ extern "C" { * Define the appropriate "processor characteristics" shared between * all Solaris on SPARC systems. */ +#if defined(sun) #define _BIG_ENDIAN +#endif #define _STACK_GROWS_DOWNWARD #define _LONG_LONG_HTOL #define _BIT_FIELDS_HTOL @@ -402,7 +456,7 @@ extern "C" { #define _DMA_USES_VIRTADDR #define _NO_FDISK_PRESENT #define _HAVE_TEM_FIRMWARE -#define _UNIX_KRTLD +#define _OBP /* * The following set of definitions characterize the implementation of diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/procset.h b/sys/cddl/contrib/opensolaris/uts/common/sys/procset.h index c3b58675746e..c367c93966a1 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/sys/procset.h +++ b/sys/cddl/contrib/opensolaris/uts/common/sys/procset.h @@ -139,6 +139,7 @@ typedef struct procset { #endif /* !defined(_XPG4_2) || defined(__EXTENSIONS__) */ +#if defined(sun) #ifdef _KERNEL struct proc; @@ -152,6 +153,7 @@ extern boolean_t cur_inset_only(procset_t *); extern id_t getmyid(idtype_t); #endif /* _KERNEL */ +#endif #ifdef __cplusplus } diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/sdt.h b/sys/cddl/contrib/opensolaris/uts/common/sys/sdt.h deleted file mode 100644 index da695c9203c7..000000000000 --- a/sys/cddl/contrib/opensolaris/uts/common/sys/sdt.h +++ /dev/null @@ -1,176 +0,0 @@ -/* - * CDDL HEADER START - * - * The contents of this file are subject to the terms of the - * Common Development and Distribution License (the "License"). - * You may not use this file except in compliance with the License. - * - * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE - * or http://www.opensolaris.org/os/licensing. - * See the License for the specific language governing permissions - * and limitations under the License. - * - * When distributing Covered Code, include this CDDL HEADER in each - * file and include the License file at usr/src/OPENSOLARIS.LICENSE. - * If applicable, add the following below this CDDL HEADER, with the - * fields enclosed by brackets "[]" replaced with your own identifying - * information: Portions Copyright [yyyy] [name of copyright owner] - * - * CDDL HEADER END - */ -/* - * Copyright 2007 Sun Microsystems, Inc. All rights reserved. - * Use is subject to license terms. - */ - -#ifndef _SYS_SDT_H -#define _SYS_SDT_H - -#pragma ident "%Z%%M% %I% %E% SMI" - -#ifdef __cplusplus -extern "C" { -#endif - -#ifndef _KERNEL - -#define DTRACE_PROBE(provider, name) { \ - extern void __dtrace_##provider##___##name(void); \ - __dtrace_##provider##___##name(); \ -} - -#define DTRACE_PROBE1(provider, name, arg1) { \ - extern void __dtrace_##provider##___##name(unsigned long); \ - __dtrace_##provider##___##name((unsigned long)arg1); \ -} - -#define DTRACE_PROBE2(provider, name, arg1, arg2) { \ - extern void __dtrace_##provider##___##name(unsigned long, \ - unsigned long); \ - __dtrace_##provider##___##name((unsigned long)arg1, \ - (unsigned long)arg2); \ -} - -#define DTRACE_PROBE3(provider, name, arg1, arg2, arg3) { \ - extern void __dtrace_##provider##___##name(unsigned long, \ - unsigned long, unsigned long); \ - __dtrace_##provider##___##name((unsigned long)arg1, \ - (unsigned long)arg2, (unsigned long)arg3); \ -} - -#define DTRACE_PROBE4(provider, name, arg1, arg2, arg3, arg4) { \ - extern void __dtrace_##provider##___##name(unsigned long, \ - unsigned long, unsigned long, unsigned long); \ - __dtrace_##provider##___##name((unsigned long)arg1, \ - (unsigned long)arg2, (unsigned long)arg3, \ - (unsigned long)arg4); \ -} - -#define DTRACE_PROBE5(provider, name, arg1, arg2, arg3, arg4, arg5) { \ - extern void __dtrace_##provider##___##name(unsigned long, \ - unsigned long, unsigned long, unsigned long, unsigned long);\ - __dtrace_##provider##___##name((unsigned long)arg1, \ - (unsigned long)arg2, (unsigned long)arg3, \ - (unsigned long)arg4, (unsigned long)arg5); \ -} - -#else /* _KERNEL */ - -#define DTRACE_PROBE(name) { \ - extern void __dtrace_probe_##name(void); \ - __dtrace_probe_##name(); \ -} - -#define DTRACE_PROBE1(name, type1, arg1) { \ - extern void __dtrace_probe_##name(uintptr_t); \ - __dtrace_probe_##name((uintptr_t)(arg1)); \ -} - -#define DTRACE_PROBE2(name, type1, arg1, type2, arg2) { \ - extern void __dtrace_probe_##name(uintptr_t, uintptr_t); \ - __dtrace_probe_##name((uintptr_t)(arg1), (uintptr_t)(arg2)); \ -} - -#define DTRACE_PROBE3(name, type1, arg1, type2, arg2, type3, arg3) { \ - extern void __dtrace_probe_##name(uintptr_t, uintptr_t, uintptr_t); \ - __dtrace_probe_##name((uintptr_t)(arg1), (uintptr_t)(arg2), \ - (uintptr_t)(arg3)); \ -} - -#define DTRACE_PROBE4(name, type1, arg1, type2, arg2, \ - type3, arg3, type4, arg4) { \ - extern void __dtrace_probe_##name(uintptr_t, uintptr_t, \ - uintptr_t, uintptr_t); \ - __dtrace_probe_##name((uintptr_t)(arg1), (uintptr_t)(arg2), \ - (uintptr_t)(arg3), (uintptr_t)(arg4)); \ -} - -#define DTRACE_SCHED(name) \ - DTRACE_PROBE(__sched_##name); - -#define DTRACE_SCHED1(name, type1, arg1) \ - DTRACE_PROBE1(__sched_##name, type1, arg1); - -#define DTRACE_SCHED2(name, type1, arg1, type2, arg2) \ - DTRACE_PROBE2(__sched_##name, type1, arg1, type2, arg2); - -#define DTRACE_SCHED3(name, type1, arg1, type2, arg2, type3, arg3) \ - DTRACE_PROBE3(__sched_##name, type1, arg1, type2, arg2, type3, arg3); - -#define DTRACE_SCHED4(name, type1, arg1, type2, arg2, \ - type3, arg3, type4, arg4) \ - DTRACE_PROBE4(__sched_##name, type1, arg1, type2, arg2, \ - type3, arg3, type4, arg4); - -#define DTRACE_PROC(name) \ - DTRACE_PROBE(__proc_##name); - -#define DTRACE_PROC1(name, type1, arg1) \ - DTRACE_PROBE1(__proc_##name, type1, arg1); - -#define DTRACE_PROC2(name, type1, arg1, type2, arg2) \ - DTRACE_PROBE2(__proc_##name, type1, arg1, type2, arg2); - -#define DTRACE_PROC3(name, type1, arg1, type2, arg2, type3, arg3) \ - DTRACE_PROBE3(__proc_##name, type1, arg1, type2, arg2, type3, arg3); - -#define DTRACE_PROC4(name, type1, arg1, type2, arg2, \ - type3, arg3, type4, arg4) \ - DTRACE_PROBE4(__proc_##name, type1, arg1, type2, arg2, \ - type3, arg3, type4, arg4); - -#define DTRACE_IO(name) \ - DTRACE_PROBE(__io_##name); - -#define DTRACE_IO1(name, type1, arg1) \ - DTRACE_PROBE1(__io_##name, type1, arg1); - -#define DTRACE_IO2(name, type1, arg1, type2, arg2) \ - DTRACE_PROBE2(__io_##name, type1, arg1, type2, arg2); - -#define DTRACE_IO3(name, type1, arg1, type2, arg2, type3, arg3) \ - DTRACE_PROBE3(__io_##name, type1, arg1, type2, arg2, type3, arg3); - -#define DTRACE_IO4(name, type1, arg1, type2, arg2, \ - type3, arg3, type4, arg4) \ - DTRACE_PROBE4(__io_##name, type1, arg1, type2, arg2, \ - type3, arg3, type4, arg4); - -#define DTRACE_SYSEVENT2(name, type1, arg1, type2, arg2) \ - DTRACE_PROBE2(__sysevent_##name, type1, arg1, type2, arg2); - -#endif /* _KERNEL */ - -extern const char *sdt_prefix; - -typedef struct sdt_probedesc { - char *sdpd_name; /* name of this probe */ - unsigned long sdpd_offset; /* offset of call in text */ - struct sdt_probedesc *sdpd_next; /* next static probe */ -} sdt_probedesc_t; - -#ifdef __cplusplus -} -#endif - -#endif /* _SYS_SDT_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/vmem.h b/sys/cddl/contrib/opensolaris/uts/common/sys/vmem.h deleted file mode 100644 index 1cd2f30e9ba3..000000000000 --- a/sys/cddl/contrib/opensolaris/uts/common/sys/vmem.h +++ /dev/null @@ -1,142 +0,0 @@ -/* - * CDDL HEADER START - * - * The contents of this file are subject to the terms of the - * Common Development and Distribution License, Version 1.0 only - * (the "License"). You may not use this file except in compliance - * with the License. - * - * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE - * or http://www.opensolaris.org/os/licensing. - * See the License for the specific language governing permissions - * and limitations under the License. - * - * When distributing Covered Code, include this CDDL HEADER in each - * file and include the License file at usr/src/OPENSOLARIS.LICENSE. - * If applicable, add the following below this CDDL HEADER, with the - * fields enclosed by brackets "[]" replaced with your own identifying - * information: Portions Copyright [yyyy] [name of copyright owner] - * - * CDDL HEADER END - */ -/* - * Copyright 2005 Sun Microsystems, Inc. All rights reserved. - * Use is subject to license terms. - */ - -#ifndef _SYS_VMEM_H -#define _SYS_VMEM_H - -#pragma ident "%Z%%M% %I% %E% SMI" - -#include <sys/types.h> - -#ifdef __cplusplus -extern "C" { -#endif - - -/* - * Per-allocation flags - */ -#define VM_SLEEP 0x00000000 /* same as KM_SLEEP */ -#define VM_NOSLEEP 0x00000001 /* same as KM_NOSLEEP */ -#define VM_PANIC 0x00000002 /* same as KM_PANIC */ -#define VM_PUSHPAGE 0x00000004 /* same as KM_PUSHPAGE */ -#define VM_KMFLAGS 0x000000ff /* flags that must match KM_* flags */ - -#define VM_BESTFIT 0x00000100 -#define VM_FIRSTFIT 0x00000200 -#define VM_NEXTFIT 0x00000400 - -/* - * The following flags are restricted for use only within the kernel. - * VM_MEMLOAD is for use by the HAT to avoid infinite recursion. - * VM_NORELOC is used by the kernel when static VA->PA mappings are required. - */ -#define VM_MEMLOAD 0x00000800 -#define VM_NORELOC 0x00001000 -/* - * VM_ABORT requests that vmem_alloc() *ignore* the VM_SLEEP/VM_NOSLEEP flags - * and forgo reaping if the allocation or attempted import, fails. This - * flag is a segkmem-specific flag, and should not be used by anyone else. - */ -#define VM_ABORT 0x00002000 - -#define VM_FLAGS 0x0000FFFF - -/* - * Arena creation flags - */ -#define VMC_POPULATOR 0x00010000 -#define VMC_NO_QCACHE 0x00020000 /* cannot use quantum caches */ -#define VMC_IDENTIFIER 0x00040000 /* not backed by memory */ -/* - * internal use only; the import function uses the vmem_ximport_t interface - * and may increase the request size if it so desires - */ -#define VMC_XALLOC 0x00080000 -#define VMC_FLAGS 0xFFFF0000 - -/* - * Public segment types - */ -#define VMEM_ALLOC 0x01 -#define VMEM_FREE 0x02 - -/* - * Implementation-private segment types - */ -#define VMEM_SPAN 0x10 -#define VMEM_ROTOR 0x20 -#define VMEM_WALKER 0x40 - -/* - * VMEM_REENTRANT indicates to vmem_walk() that the callback routine may - * call back into the arena being walked, so vmem_walk() must drop the - * arena lock before each callback. The caveat is that since the arena - * isn't locked, its state can change. Therefore it is up to the callback - * routine to handle cases where the segment isn't of the expected type. - * For example, we use this to walk heap_arena when generating a crash dump; - * see segkmem_dump() for sample usage. - */ -#define VMEM_REENTRANT 0x80000000 - -typedef struct vmem vmem_t; -typedef void *(vmem_alloc_t)(vmem_t *, size_t, int); -typedef void (vmem_free_t)(vmem_t *, void *, size_t); - -/* - * Alternate import style; the requested size is passed in a pointer, - * which can be increased by the import function if desired. - */ -typedef void *(vmem_ximport_t)(vmem_t *, size_t *, int); - -#ifdef _KERNEL -extern vmem_t *vmem_init(const char *, void *, size_t, size_t, - vmem_alloc_t *, vmem_free_t *); -extern void vmem_update(void *); -extern int vmem_is_populator(); -extern size_t vmem_seg_size; -#endif - -extern vmem_t *vmem_create(const char *, void *, size_t, size_t, - vmem_alloc_t *, vmem_free_t *, vmem_t *, size_t, int); -extern vmem_t *vmem_xcreate(const char *, void *, size_t, size_t, - vmem_ximport_t *, vmem_free_t *, vmem_t *, size_t, int); -extern void vmem_destroy(vmem_t *); -extern void *vmem_alloc(vmem_t *, size_t, int); -extern void *vmem_xalloc(vmem_t *, size_t, size_t, size_t, size_t, - void *, void *, int); -extern void vmem_free(vmem_t *, void *, size_t); -extern void vmem_xfree(vmem_t *, void *, size_t); -extern void *vmem_add(vmem_t *, void *, size_t, int); -extern int vmem_contains(vmem_t *, void *, size_t); -extern void vmem_walk(vmem_t *, int, void (*)(void *, void *, size_t), void *); -extern size_t vmem_size(vmem_t *, int); - -#ifdef __cplusplus -} -#endif - -#endif /* _SYS_VMEM_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/intel/sys/fasttrap_isa.h b/sys/cddl/contrib/opensolaris/uts/intel/sys/fasttrap_isa.h new file mode 100644 index 000000000000..9fee8cdb6be0 --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/intel/sys/fasttrap_isa.h @@ -0,0 +1,114 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _FASTTRAP_ISA_H +#define _FASTTRAP_ISA_H + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/types.h> + +#ifdef __cplusplus +extern "C" { +#endif + +#define FASTTRAP_MAX_INSTR_SIZE 15 + +#define FASTTRAP_INSTR 0xcc + +#define FASTTRAP_SUNWDTRACE_SIZE 64 + +typedef uint8_t fasttrap_instr_t; + +typedef struct fasttrap_machtp { + uint8_t ftmt_instr[FASTTRAP_MAX_INSTR_SIZE]; /* orig. instr. */ + uint8_t ftmt_size; /* instruction size */ +#ifdef __amd64 + uint8_t ftmt_ripmode; /* %rip-relative handling mode */ + uint8_t ftmt_modrm; /* saved modrm byte */ +#endif + uint8_t ftmt_type; /* emulation type */ + uint8_t ftmt_code; /* branch condition */ + uint8_t ftmt_base; /* branch base */ + uint8_t ftmt_index; /* branch index */ + uint8_t ftmt_scale; /* branch scale */ + uint8_t ftmt_segment; /* segment for memory accesses */ + uintptr_t ftmt_dest; /* destination of control flow */ +} fasttrap_machtp_t; + +#define ftt_instr ftt_mtp.ftmt_instr +#ifdef __amd64 +#define ftt_ripmode ftt_mtp.ftmt_ripmode +#define ftt_modrm ftt_mtp.ftmt_modrm +#endif +#define ftt_size ftt_mtp.ftmt_size +#define ftt_type ftt_mtp.ftmt_type +#define ftt_code ftt_mtp.ftmt_code +#define ftt_base ftt_mtp.ftmt_base +#define ftt_index ftt_mtp.ftmt_index +#define ftt_scale ftt_mtp.ftmt_scale +#define ftt_segment ftt_mtp.ftmt_segment +#define ftt_dest ftt_mtp.ftmt_dest + +#define FASTTRAP_T_COMMON 0x00 /* common case -- no emulation */ +#define FASTTRAP_T_JCC 0x01 /* near and far conditional jumps */ +#define FASTTRAP_T_LOOP 0x02 /* loop instructions */ +#define FASTTRAP_T_JCXZ 0x03 /* jump if %ecx/%rcx is zero */ +#define FASTTRAP_T_JMP 0x04 /* relative jump */ +#define FASTTRAP_T_CALL 0x05 /* near call (and link) */ +#define FASTTRAP_T_RET 0x06 /* ret */ +#define FASTTRAP_T_RET16 0x07 /* ret <imm16> */ + +/* + * For performance rather than correctness. + */ +#define FASTTRAP_T_PUSHL_EBP 0x10 /* pushl %ebp (for function entry) */ +#define FASTTRAP_T_NOP 0x11 /* nop */ + +#define FASTTRAP_RIP_1 0x1 +#define FASTTRAP_RIP_2 0x2 +#define FASTTRAP_RIP_X 0x4 + +/* + * Segment values. + */ +#define FASTTRAP_SEG_NONE 0 +#define FASTTRAP_SEG_CS 1 +#define FASTTRAP_SEG_DS 2 +#define FASTTRAP_SEG_ES 3 +#define FASTTRAP_SEG_FS 4 +#define FASTTRAP_SEG_GS 5 +#define FASTTRAP_SEG_SS 6 + +#define FASTTRAP_AFRAMES 3 +#define FASTTRAP_RETURN_AFRAMES 4 +#define FASTTRAP_ENTRY_AFRAMES 3 +#define FASTTRAP_OFFSET_AFRAMES 3 + +#ifdef __cplusplus +} +#endif + +#endif /* _FASTTRAP_ISA_H */ diff --git a/sys/cddl/contrib/opensolaris/uts/sparc/sys/fasttrap_isa.h b/sys/cddl/contrib/opensolaris/uts/sparc/sys/fasttrap_isa.h new file mode 100644 index 000000000000..10361cbed8de --- /dev/null +++ b/sys/cddl/contrib/opensolaris/uts/sparc/sys/fasttrap_isa.h @@ -0,0 +1,94 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2005 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _FASTTRAP_ISA_H +#define _FASTTRAP_ISA_H + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/types.h> + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * This is our reserved trap instruction: ta 0x38 + */ +#define FASTTRAP_INSTR 0x91d02038 + +#define FASTTRAP_SUNWDTRACE_SIZE 128 + +typedef uint32_t fasttrap_instr_t; + +typedef struct fasttrap_machtp { + fasttrap_instr_t ftmt_instr; /* original instruction */ + uintptr_t ftmt_dest; /* destination of DCTI */ + uint8_t ftmt_type; /* emulation type */ + uint8_t ftmt_flags; /* emulation flags */ + uint8_t ftmt_cc; /* which cc to look at */ + uint8_t ftmt_code; /* branch condition */ +} fasttrap_machtp_t; + +#define ftt_instr ftt_mtp.ftmt_instr +#define ftt_dest ftt_mtp.ftmt_dest +#define ftt_type ftt_mtp.ftmt_type +#define ftt_flags ftt_mtp.ftmt_flags +#define ftt_cc ftt_mtp.ftmt_cc +#define ftt_code ftt_mtp.ftmt_code + +#define FASTTRAP_T_COMMON 0x00 /* common case -- no emulation */ +#define FASTTRAP_T_CCR 0x01 /* integer condition code branch */ +#define FASTTRAP_T_FCC 0x02 /* floating-point branch */ +#define FASTTRAP_T_REG 0x03 /* register predicated branch */ +#define FASTTRAP_T_ALWAYS 0x04 /* branch always */ +#define FASTTRAP_T_CALL 0x05 /* call instruction */ +#define FASTTRAP_T_JMPL 0x06 /* jmpl instruction */ +#define FASTTRAP_T_RDPC 0x07 /* rdpc instruction */ +#define FASTTRAP_T_RETURN 0x08 /* return instruction */ + +/* + * For performance rather than correctness. + */ +#define FASTTRAP_T_SAVE 0x10 /* save instruction (func entry only) */ +#define FASTTRAP_T_RESTORE 0x11 /* restore instruction */ +#define FASTTRAP_T_OR 0x12 /* mov instruction */ +#define FASTTRAP_T_SETHI 0x13 /* sethi instruction (includes nop) */ + +#define FASTTRAP_F_ANNUL 0x01 /* branch is annulled */ +#define FASTTRAP_F_RETMAYBE 0x02 /* not definitely a return site */ + +#define FASTTRAP_AFRAMES 3 +#define FASTTRAP_RETURN_AFRAMES 4 +#define FASTTRAP_ENTRY_AFRAMES 3 +#define FASTTRAP_OFFSET_AFRAMES 3 + + +#ifdef __cplusplus +} +#endif + +#endif /* _FASTTRAP_ISA_H */ diff --git a/sys/cddl/dev/cyclic/amd64/cyclic_machdep.c b/sys/cddl/dev/cyclic/amd64/cyclic_machdep.c new file mode 100644 index 000000000000..9a15b2cdad03 --- /dev/null +++ b/sys/cddl/dev/cyclic/amd64/cyclic_machdep.c @@ -0,0 +1,133 @@ +/*- + * Copyright 2007 John Birrell <jb@FreeBSD.org> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + */ + +static void enable(cyb_arg_t); +static void disable(cyb_arg_t); +static void reprogram(cyb_arg_t, hrtime_t); +static void xcall(cyb_arg_t, cpu_t *, cyc_func_t, void *); + +static cyc_backend_t be = { + NULL, /* cyb_configure */ + NULL, /* cyb_unconfigure */ + enable, + disable, + reprogram, + xcall, + NULL /* cyb_arg_t cyb_arg */ +}; + +static void +cyclic_ap_start(void *dummy) +{ + /* Initialise the rest of the CPUs. */ + cyclic_mp_init(); +} + +SYSINIT(cyclic_ap_start, SI_SUB_SMP, SI_ORDER_ANY, cyclic_ap_start, NULL); + +/* + * Machine dependent cyclic subsystem initialisation. + */ +static void +cyclic_machdep_init(void) +{ + /* Register the cyclic backend. */ + cyclic_init(&be); +} + +static void +cyclic_machdep_uninit(void) +{ + int i; + + for (i = 0; i <= mp_maxid; i++) + /* Reset the cyclic clock callback hook. */ + lapic_cyclic_clock_func[i] = NULL; + + /* De-register the cyclic backend. */ + cyclic_uninit(); +} + +static hrtime_t exp_due[MAXCPU]; + +/* + * This function is the one registered by the machine dependent + * initialiser as the callback for high speed timer events. + */ +static void +cyclic_clock(struct trapframe *frame) +{ + cpu_t *c = &solaris_cpu[curcpu]; + + if (c->cpu_cyclic != NULL && gethrtime() >= exp_due[curcpu]) { + if (TRAPF_USERMODE(frame)) { + c->cpu_profile_pc = 0; + c->cpu_profile_upc = TRAPF_PC(frame); + } else { + c->cpu_profile_pc = TRAPF_PC(frame); + c->cpu_profile_upc = 0; + } + + c->cpu_intr_actv = 1; + + /* Fire any timers that are due. */ + cyclic_fire(c); + + c->cpu_intr_actv = 0; + } +} + +static void enable(cyb_arg_t arg) +{ + /* Register the cyclic clock callback function. */ + lapic_cyclic_clock_func[curcpu] = cyclic_clock; +} + +static void disable(cyb_arg_t arg) +{ + /* Reset the cyclic clock callback function. */ + lapic_cyclic_clock_func[curcpu] = NULL; +} + +static void reprogram(cyb_arg_t arg, hrtime_t exp) +{ + exp_due[curcpu] = exp; +} + +static void xcall(cyb_arg_t arg, cpu_t *c, cyc_func_t func, void *param) +{ + /* + * If the target CPU is the current one, just call the + * function. This covers the non-SMP case. + */ + if (c == &solaris_cpu[curcpu]) + (*func)(param); + else + smp_rendezvous_cpus((cpumask_t) (1 << c->cpuid), NULL, + func, smp_no_rendevous_barrier, param); +} diff --git a/sys/cddl/dev/cyclic/cyclic.c b/sys/cddl/dev/cyclic/cyclic.c new file mode 100644 index 000000000000..9d4d2d076d97 --- /dev/null +++ b/sys/cddl/dev/cyclic/cyclic.c @@ -0,0 +1,1421 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * Portions Copyright 2008 John Birrell <jb@freebsd.org> + * + * $FreeBSD$ + * + * This is a simplified version of the cyclic timer subsystem from + * OpenSolaris. In the FreeBSD version, we don't use interrupt levels. + */ + +/* + * Copyright 2004 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +/* + * The Cyclic Subsystem + * -------------------- + * + * Prehistory + * + * Historically, most computer architectures have specified interval-based + * timer parts (e.g. SPARCstation's counter/timer; Intel's i8254). While + * these parts deal in relative (i.e. not absolute) time values, they are + * typically used by the operating system to implement the abstraction of + * absolute time. As a result, these parts cannot typically be reprogrammed + * without introducing error in the system's notion of time. + * + * Starting in about 1994, chip architectures began specifying high resolution + * timestamp registers. As of this writing (1999), all major chip families + * (UltraSPARC, PentiumPro, MIPS, PowerPC, Alpha) have high resolution + * timestamp registers, and two (UltraSPARC and MIPS) have added the capacity + * to interrupt based on timestamp values. These timestamp-compare registers + * present a time-based interrupt source which can be reprogrammed arbitrarily + * often without introducing error. Given the low cost of implementing such a + * timestamp-compare register (and the tangible benefit of eliminating + * discrete timer parts), it is reasonable to expect that future chip + * architectures will adopt this feature. + * + * The cyclic subsystem has been designed to take advantage of chip + * architectures with the capacity to interrupt based on absolute, high + * resolution values of time. + * + * Subsystem Overview + * + * The cyclic subsystem is a low-level kernel subsystem designed to provide + * arbitrarily high resolution, per-CPU interval timers (to avoid colliding + * with existing terms, we dub such an interval timer a "cyclic"). + * Alternatively, a cyclic may be specified to be "omnipresent", denoting + * firing on all online CPUs. + * + * Cyclic Subsystem Interface Overview + * ----------------------------------- + * + * The cyclic subsystem has interfaces with the kernel at-large, with other + * kernel subsystems (e.g. the processor management subsystem, the checkpoint + * resume subsystem) and with the platform (the cyclic backend). Each + * of these interfaces is given a brief synopsis here, and is described + * in full above the interface's implementation. + * + * The following diagram displays the cyclic subsystem's interfaces to + * other kernel components. The arrows denote a "calls" relationship, with + * the large arrow indicating the cyclic subsystem's consumer interface. + * Each arrow is labeled with the section in which the corresponding + * interface is described. + * + * Kernel at-large consumers + * -----------++------------ + * || + * || + * _||_ + * \ / + * \/ + * +---------------------+ + * | | + * | Cyclic subsystem |<----------- Other kernel subsystems + * | | + * +---------------------+ + * ^ | + * | | + * | | + * | v + * +---------------------+ + * | | + * | Cyclic backend | + * | (platform specific) | + * | | + * +---------------------+ + * + * + * Kernel At-Large Interfaces + * + * cyclic_add() <-- Creates a cyclic + * cyclic_add_omni() <-- Creates an omnipresent cyclic + * cyclic_remove() <-- Removes a cyclic + * + * Backend Interfaces + * + * cyclic_init() <-- Initializes the cyclic subsystem + * cyclic_fire() <-- Interrupt entry point + * + * The backend-supplied interfaces (through the cyc_backend structure) are + * documented in detail in <sys/cyclic_impl.h> + * + * + * Cyclic Subsystem Implementation Overview + * ---------------------------------------- + * + * The cyclic subsystem is designed to minimize interference between cyclics + * on different CPUs. Thus, all of the cyclic subsystem's data structures + * hang off of a per-CPU structure, cyc_cpu. + * + * Each cyc_cpu has a power-of-two sized array of cyclic structures (the + * cyp_cyclics member of the cyc_cpu structure). If cyclic_add() is called + * and there does not exist a free slot in the cyp_cyclics array, the size of + * the array will be doubled. The array will never shrink. Cyclics are + * referred to by their index in the cyp_cyclics array, which is of type + * cyc_index_t. + * + * The cyclics are kept sorted by expiration time in the cyc_cpu's heap. The + * heap is keyed by cyclic expiration time, with parents expiring earlier + * than their children. + * + * Heap Management + * + * The heap is managed primarily by cyclic_fire(). Upon entry, cyclic_fire() + * compares the root cyclic's expiration time to the current time. If the + * expiration time is in the past, cyclic_expire() is called on the root + * cyclic. Upon return from cyclic_expire(), the cyclic's new expiration time + * is derived by adding its interval to its old expiration time, and a + * downheap operation is performed. After the downheap, cyclic_fire() + * examines the (potentially changed) root cyclic, repeating the + * cyclic_expire()/add interval/cyclic_downheap() sequence until the root + * cyclic has an expiration time in the future. This expiration time + * (guaranteed to be the earliest in the heap) is then communicated to the + * backend via cyb_reprogram. Optimal backends will next call cyclic_fire() + * shortly after the root cyclic's expiration time. + * + * To allow efficient, deterministic downheap operations, we implement the + * heap as an array (the cyp_heap member of the cyc_cpu structure), with each + * element containing an index into the CPU's cyp_cyclics array. + * + * The heap is laid out in the array according to the following: + * + * 1. The root of the heap is always in the 0th element of the heap array + * 2. The left and right children of the nth element are element + * (((n + 1) << 1) - 1) and element ((n + 1) << 1), respectively. + * + * This layout is standard (see, e.g., Cormen's "Algorithms"); the proof + * that these constraints correctly lay out a heap (or indeed, any binary + * tree) is trivial and left to the reader. + * + * To see the heap by example, assume our cyclics array has the following + * members (at time t): + * + * cy_handler cy_expire + * --------------------------------------------- + * [ 0] clock() t+10000000 + * [ 1] deadman() t+1000000000 + * [ 2] clock_highres_fire() t+100 + * [ 3] clock_highres_fire() t+1000 + * [ 4] clock_highres_fire() t+500 + * [ 5] (free) -- + * [ 6] (free) -- + * [ 7] (free) -- + * + * The heap array could be: + * + * [0] [1] [2] [3] [4] [5] [6] [7] + * +-----+-----+-----+-----+-----+-----+-----+-----+ + * | | | | | | | | | + * | 2 | 3 | 4 | 0 | 1 | x | x | x | + * | | | | | | | | | + * +-----+-----+-----+-----+-----+-----+-----+-----+ + * + * Graphically, this array corresponds to the following (excuse the ASCII art): + * + * 2 + * | + * +------------------+------------------+ + * 3 4 + * | + * +---------+--------+ + * 0 1 + * + * Note that the heap is laid out by layer: all nodes at a given depth are + * stored in consecutive elements of the array. Moreover, layers of + * consecutive depths are in adjacent element ranges. This property + * guarantees high locality of reference during downheap operations. + * Specifically, we are guaranteed that we can downheap to a depth of + * + * lg (cache_line_size / sizeof (cyc_index_t)) + * + * nodes with at most one cache miss. On UltraSPARC (64 byte e-cache line + * size), this corresponds to a depth of four nodes. Thus, if there are + * fewer than sixteen cyclics in the heap, downheaps on UltraSPARC miss at + * most once in the e-cache. + * + * Downheaps are required to compare siblings as they proceed down the + * heap. For downheaps proceeding beyond the one-cache-miss depth, every + * access to a left child could potentially miss in the cache. However, + * if we assume + * + * (cache_line_size / sizeof (cyc_index_t)) > 2, + * + * then all siblings are guaranteed to be on the same cache line. Thus, the + * miss on the left child will guarantee a hit on the right child; downheaps + * will incur at most one cache miss per layer beyond the one-cache-miss + * depth. The total number of cache misses for heap management during a + * downheap operation is thus bounded by + * + * lg (n) - lg (cache_line_size / sizeof (cyc_index_t)) + * + * Traditional pointer-based heaps are implemented without regard to + * locality. Downheaps can thus incur two cache misses per layer (one for + * each child), but at most one cache miss at the root. This yields a bound + * of + * + * 2 * lg (n) - 1 + * + * on the total cache misses. + * + * This difference may seem theoretically trivial (the difference is, after + * all, constant), but can become substantial in practice -- especially for + * caches with very large cache lines and high miss penalties (e.g. TLBs). + * + * Heaps must always be full, balanced trees. Heap management must therefore + * track the next point-of-insertion into the heap. In pointer-based heaps, + * recomputing this point takes O(lg (n)). Given the layout of the + * array-based implementation, however, the next point-of-insertion is + * always: + * + * heap[number_of_elements] + * + * We exploit this property by implementing the free-list in the usused + * heap elements. Heap insertion, therefore, consists only of filling in + * the cyclic at cyp_cyclics[cyp_heap[number_of_elements]], incrementing + * the number of elements, and performing an upheap. Heap deletion consists + * of decrementing the number of elements, swapping the to-be-deleted element + * with the element at cyp_heap[number_of_elements], and downheaping. + * + * Filling in more details in our earlier example: + * + * +--- free list head + * | + * V + * + * [0] [1] [2] [3] [4] [5] [6] [7] + * +-----+-----+-----+-----+-----+-----+-----+-----+ + * | | | | | | | | | + * | 2 | 3 | 4 | 0 | 1 | 5 | 6 | 7 | + * | | | | | | | | | + * +-----+-----+-----+-----+-----+-----+-----+-----+ + * + * To insert into this heap, we would just need to fill in the cyclic at + * cyp_cyclics[5], bump the number of elements (from 5 to 6) and perform + * an upheap. + * + * If we wanted to remove, say, cyp_cyclics[3], we would first scan for it + * in the cyp_heap, and discover it at cyp_heap[1]. We would then decrement + * the number of elements (from 5 to 4), swap cyp_heap[1] with cyp_heap[4], + * and perform a downheap from cyp_heap[1]. The linear scan is required + * because the cyclic does not keep a backpointer into the heap. This makes + * heap manipulation (e.g. downheaps) faster at the expense of removal + * operations. + * + * Expiry processing + * + * As alluded to above, cyclic_expire() is called by cyclic_fire() to expire + * a cyclic. Cyclic subsystem consumers are guaranteed that for an arbitrary + * time t in the future, their cyclic handler will have been called + * (t - cyt_when) / cyt_interval times. cyclic_expire() simply needs to call + * the handler. + * + * Resizing + * + * All of the discussion thus far has assumed a static number of cyclics. + * Obviously, static limitations are not practical; we need the capacity + * to resize our data structures dynamically. + * + * We resize our data structures lazily, and only on a per-CPU basis. + * The size of the data structures always doubles and never shrinks. We + * serialize adds (and thus resizes) on cpu_lock; we never need to deal + * with concurrent resizes. Resizes should be rare; they may induce jitter + * on the CPU being resized, but should not affect cyclic operation on other + * CPUs. + * + * Three key cyc_cpu data structures need to be resized: the cyclics array, + * nad the heap array. Resizing is relatively straightforward: + * + * 1. The new, larger arrays are allocated in cyclic_expand() (called + * from cyclic_add()). + * 2. The contents of the old arrays are copied into the new arrays. + * 3. The old cyclics array is bzero()'d + * 4. The pointers are updated. + * + * Removals + * + * Cyclic removals should be rare. To simplify the implementation (and to + * allow optimization for the cyclic_fire()/cyclic_expire() + * path), we force removals and adds to serialize on cpu_lock. + * + */ +#include <sys/cdefs.h> +#include <sys/param.h> +#include <sys/conf.h> +#include <sys/kernel.h> +#include <sys/lock.h> +#include <sys/sx.h> +#include <sys/cyclic_impl.h> +#include <sys/module.h> +#include <sys/systm.h> +#include <sys/atomic.h> +#include <sys/kmem.h> +#include <sys/cmn_err.h> +#include <sys/dtrace_bsd.h> +#include <machine/cpu.h> + +static kmem_cache_t *cyclic_id_cache; +static cyc_id_t *cyclic_id_head; +static cyc_backend_t cyclic_backend; + +MALLOC_DEFINE(M_CYCLIC, "cyclic", "Cyclic timer subsystem"); + +/* + * Returns 1 if the upheap propagated to the root, 0 if it did not. This + * allows the caller to reprogram the backend only when the root has been + * modified. + */ +static int +cyclic_upheap(cyc_cpu_t *cpu, cyc_index_t ndx) +{ + cyclic_t *cyclics; + cyc_index_t *heap; + cyc_index_t heap_parent, heap_current = ndx; + cyc_index_t parent, current; + + if (heap_current == 0) + return (1); + + heap = cpu->cyp_heap; + cyclics = cpu->cyp_cyclics; + heap_parent = CYC_HEAP_PARENT(heap_current); + + for (;;) { + current = heap[heap_current]; + parent = heap[heap_parent]; + + /* + * We have an expiration time later than our parent; we're + * done. + */ + if (cyclics[current].cy_expire >= cyclics[parent].cy_expire) + return (0); + + /* + * We need to swap with our parent, and continue up the heap. + */ + heap[heap_parent] = current; + heap[heap_current] = parent; + + /* + * If we just reached the root, we're done. + */ + if (heap_parent == 0) + return (1); + + heap_current = heap_parent; + heap_parent = CYC_HEAP_PARENT(heap_current); + } +} + +static void +cyclic_downheap(cyc_cpu_t *cpu, cyc_index_t ndx) +{ + cyclic_t *cyclics = cpu->cyp_cyclics; + cyc_index_t *heap = cpu->cyp_heap; + + cyc_index_t heap_left, heap_right, heap_me = ndx; + cyc_index_t left, right, me; + cyc_index_t nelems = cpu->cyp_nelems; + + for (;;) { + /* + * If we don't have a left child (i.e., we're a leaf), we're + * done. + */ + if ((heap_left = CYC_HEAP_LEFT(heap_me)) >= nelems) + return; + + left = heap[heap_left]; + me = heap[heap_me]; + + heap_right = CYC_HEAP_RIGHT(heap_me); + + /* + * Even if we don't have a right child, we still need to compare + * our expiration time against that of our left child. + */ + if (heap_right >= nelems) + goto comp_left; + + right = heap[heap_right]; + + /* + * We have both a left and a right child. We need to compare + * the expiration times of the children to determine which + * expires earlier. + */ + if (cyclics[right].cy_expire < cyclics[left].cy_expire) { + /* + * Our right child is the earlier of our children. + * We'll now compare our expiration time to its; if + * ours is the earlier, we're done. + */ + if (cyclics[me].cy_expire <= cyclics[right].cy_expire) + return; + + /* + * Our right child expires earlier than we do; swap + * with our right child, and descend right. + */ + heap[heap_right] = me; + heap[heap_me] = right; + heap_me = heap_right; + continue; + } + +comp_left: + /* + * Our left child is the earlier of our children (or we have + * no right child). We'll now compare our expiration time + * to its; if ours is the earlier, we're done. + */ + if (cyclics[me].cy_expire <= cyclics[left].cy_expire) + return; + + /* + * Our left child expires earlier than we do; swap with our + * left child, and descend left. + */ + heap[heap_left] = me; + heap[heap_me] = left; + heap_me = heap_left; + } +} + +static void +cyclic_expire(cyc_cpu_t *cpu, cyc_index_t ndx, cyclic_t *cyclic) +{ + cyc_func_t handler = cyclic->cy_handler; + void *arg = cyclic->cy_arg; + + (*handler)(arg); +} + +static void +cyclic_enable_xcall(void *v) +{ + cyc_xcallarg_t *argp = v; + cyc_cpu_t *cpu = argp->cyx_cpu; + cyc_backend_t *be = cpu->cyp_backend; + + be->cyb_enable(be->cyb_arg); +} + +static void +cyclic_enable(cyc_cpu_t *cpu) +{ + cyc_backend_t *be = cpu->cyp_backend; + cyc_xcallarg_t arg; + + arg.cyx_cpu = cpu; + + /* Cross call to the target CPU */ + be->cyb_xcall(be->cyb_arg, cpu->cyp_cpu, cyclic_enable_xcall, &arg); +} + +static void +cyclic_disable_xcall(void *v) +{ + cyc_xcallarg_t *argp = v; + cyc_cpu_t *cpu = argp->cyx_cpu; + cyc_backend_t *be = cpu->cyp_backend; + + be->cyb_disable(be->cyb_arg); +} + +static void +cyclic_disable(cyc_cpu_t *cpu) +{ + cyc_backend_t *be = cpu->cyp_backend; + cyc_xcallarg_t arg; + + arg.cyx_cpu = cpu; + + /* Cross call to the target CPU */ + be->cyb_xcall(be->cyb_arg, cpu->cyp_cpu, cyclic_disable_xcall, &arg); +} + +static void +cyclic_reprogram_xcall(void *v) +{ + cyc_xcallarg_t *argp = v; + cyc_cpu_t *cpu = argp->cyx_cpu; + cyc_backend_t *be = cpu->cyp_backend; + + be->cyb_reprogram(be->cyb_arg, argp->cyx_exp); +} + +static void +cyclic_reprogram(cyc_cpu_t *cpu, hrtime_t exp) +{ + cyc_backend_t *be = cpu->cyp_backend; + cyc_xcallarg_t arg; + + arg.cyx_cpu = cpu; + arg.cyx_exp = exp; + + /* Cross call to the target CPU */ + be->cyb_xcall(be->cyb_arg, cpu->cyp_cpu, cyclic_reprogram_xcall, &arg); +} + +/* + * cyclic_fire(cpu_t *) + * + * Overview + * + * cyclic_fire() is the cyclic subsystem's interrupt handler. + * Called by the cyclic backend. + * + * Arguments and notes + * + * The only argument is the CPU on which the interrupt is executing; + * backends must call into cyclic_fire() on the specified CPU. + * + * cyclic_fire() may be called spuriously without ill effect. Optimal + * backends will call into cyclic_fire() at or shortly after the time + * requested via cyb_reprogram(). However, calling cyclic_fire() + * arbitrarily late will only manifest latency bubbles; the correctness + * of the cyclic subsystem does not rely on the timeliness of the backend. + * + * cyclic_fire() is wait-free; it will not block or spin. + * + * Return values + * + * None. + * + */ +static void +cyclic_fire(cpu_t *c) +{ + cyc_cpu_t *cpu = c->cpu_cyclic; + + mtx_lock_spin(&cpu->cyp_mtx); + + cyc_index_t *heap = cpu->cyp_heap; + cyclic_t *cyclic, *cyclics = cpu->cyp_cyclics; + hrtime_t now = gethrtime(); + hrtime_t exp; + + if (cpu->cyp_nelems == 0) { + /* This is a spurious fire. */ + mtx_unlock_spin(&cpu->cyp_mtx); + return; + } + + for (;;) { + cyc_index_t ndx = heap[0]; + + cyclic = &cyclics[ndx]; + + ASSERT(!(cyclic->cy_flags & CYF_FREE)); + + if ((exp = cyclic->cy_expire) > now) + break; + + cyclic_expire(cpu, ndx, cyclic); + + /* + * If this cyclic will be set to next expire in the distant + * past, we have one of two situations: + * + * a) This is the first firing of a cyclic which had + * cy_expire set to 0. + * + * b) We are tragically late for a cyclic -- most likely + * due to being in the debugger. + * + * In either case, we set the new expiration time to be the + * the next interval boundary. This assures that the + * expiration time modulo the interval is invariant. + * + * We arbitrarily define "distant" to be one second (one second + * is chosen because it's shorter than any foray to the + * debugger while still being longer than any legitimate + * stretch). + */ + exp += cyclic->cy_interval; + + if (now - exp > NANOSEC) { + hrtime_t interval = cyclic->cy_interval; + + exp += ((now - exp) / interval + 1) * interval; + } + + cyclic->cy_expire = exp; + cyclic_downheap(cpu, 0); + } + + /* + * Now we have a cyclic in the root slot which isn't in the past; + * reprogram the interrupt source. + */ + cyclic_reprogram(cpu, exp); + + mtx_unlock_spin(&cpu->cyp_mtx); +} + +/* + * cyclic_expand() will cross call onto the CPU to perform the actual + * expand operation. + */ +static void +cyclic_expand(cyc_cpu_t *cpu) +{ + cyc_index_t new_size, old_size, i; + cyc_index_t *new_heap, *old_heap; + cyclic_t *new_cyclics, *old_cyclics; + + ASSERT(MUTEX_HELD(&cpu_lock)); + + if ((new_size = ((old_size = cpu->cyp_size) << 1)) == 0) + new_size = CY_DEFAULT_PERCPU; + + /* + * Check that the new_size is a power of 2. + */ + ASSERT(((new_size - 1) & new_size) == 0); + + /* Unlock the mutex while allocating memory so we can wait... */ + mtx_unlock_spin(&cpu->cyp_mtx); + + new_heap = malloc(sizeof(cyc_index_t) * new_size, M_CYCLIC, M_WAITOK); + new_cyclics = malloc(sizeof(cyclic_t) * new_size, M_CYCLIC, M_ZERO | M_WAITOK); + + /* Grab the lock again now we've got the memory... */ + mtx_lock_spin(&cpu->cyp_mtx); + + /* Check if another thread beat us while the mutex was unlocked. */ + if (old_size != cpu->cyp_size) { + /* Oh well, he won. */ + mtx_unlock_spin(&cpu->cyp_mtx); + + free(new_heap, M_CYCLIC); + free(new_cyclics, M_CYCLIC); + + mtx_lock_spin(&cpu->cyp_mtx); + return; + } + + old_heap = cpu->cyp_heap; + old_cyclics = cpu->cyp_cyclics; + + bcopy(cpu->cyp_heap, new_heap, sizeof (cyc_index_t) * old_size); + bcopy(old_cyclics, new_cyclics, sizeof (cyclic_t) * old_size); + + /* + * Set up the free list, and set all of the new cyclics to be CYF_FREE. + */ + for (i = old_size; i < new_size; i++) { + new_heap[i] = i; + new_cyclics[i].cy_flags = CYF_FREE; + } + + /* + * We can go ahead and plow the value of cyp_heap and cyp_cyclics; + * cyclic_expand() has kept a copy. + */ + cpu->cyp_heap = new_heap; + cpu->cyp_cyclics = new_cyclics; + cpu->cyp_size = new_size; + + if (old_cyclics != NULL) { + ASSERT(old_heap != NULL); + ASSERT(old_size != 0); + mtx_unlock_spin(&cpu->cyp_mtx); + + free(old_cyclics, M_CYCLIC); + free(old_heap, M_CYCLIC); + + mtx_lock_spin(&cpu->cyp_mtx); + } +} + +static cyc_index_t +cyclic_add_here(cyc_cpu_t *cpu, cyc_handler_t *hdlr, + cyc_time_t *when, uint16_t flags) +{ + cyc_index_t ndx, nelems; + cyclic_t *cyclic; + + ASSERT(MUTEX_HELD(&cpu_lock)); + + mtx_lock_spin(&cpu->cyp_mtx); + + ASSERT(!(cpu->cyp_cpu->cpu_flags & CPU_OFFLINE)); + ASSERT(when->cyt_when >= 0 && when->cyt_interval > 0); + + while (cpu->cyp_nelems == cpu->cyp_size) + cyclic_expand(cpu); + + ASSERT(cpu->cyp_nelems < cpu->cyp_size); + + nelems = cpu->cyp_nelems++; + + if (nelems == 0) + /* + * If this is the first element, we need to enable the + * backend on this CPU. + */ + cyclic_enable(cpu); + + ndx = cpu->cyp_heap[nelems]; + cyclic = &cpu->cyp_cyclics[ndx]; + + ASSERT(cyclic->cy_flags == CYF_FREE); + cyclic->cy_interval = when->cyt_interval; + + if (when->cyt_when == 0) + cyclic->cy_expire = gethrtime() + cyclic->cy_interval; + else + cyclic->cy_expire = when->cyt_when; + + cyclic->cy_handler = hdlr->cyh_func; + cyclic->cy_arg = hdlr->cyh_arg; + cyclic->cy_flags = flags; + + if (cyclic_upheap(cpu, nelems)) { + hrtime_t exp = cyclic->cy_expire; + + /* + * If our upheap propagated to the root, we need to + * reprogram the interrupt source. + */ + cyclic_reprogram(cpu, exp); + } + + mtx_unlock_spin(&cpu->cyp_mtx); + + return (ndx); +} + + +static int +cyclic_remove_here(cyc_cpu_t *cpu, cyc_index_t ndx, cyc_time_t *when, int wait) +{ + cyc_index_t nelems, i; + cyclic_t *cyclic; + cyc_index_t *heap, last; + + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(wait == CY_WAIT || wait == CY_NOWAIT); + + mtx_lock_spin(&cpu->cyp_mtx); + + heap = cpu->cyp_heap; + + nelems = cpu->cyp_nelems; + + cyclic = &cpu->cyp_cyclics[ndx]; + + /* + * Grab the current expiration time. If this cyclic is being + * removed as part of a juggling operation, the expiration time + * will be used when the cyclic is added to the new CPU. + */ + if (when != NULL) { + when->cyt_when = cyclic->cy_expire; + when->cyt_interval = cyclic->cy_interval; + } + + cyclic->cy_flags = CYF_FREE; + + for (i = 0; i < nelems; i++) { + if (heap[i] == ndx) + break; + } + + if (i == nelems) + panic("attempt to remove non-existent cyclic"); + + cpu->cyp_nelems = --nelems; + + if (nelems == 0) + /* + * If we just removed the last element, then we need to + * disable the backend on this CPU. + */ + cyclic_disable(cpu); + + if (i == nelems) + /* + * If we just removed the last element of the heap, then + * we don't have to downheap. + */ + goto done; + + /* + * Swap the last element of the heap with the one we want to + * remove, and downheap (this has the implicit effect of putting + * the newly freed element on the free list). + */ + heap[i] = (last = heap[nelems]); + heap[nelems] = ndx; + + if (i == 0) + cyclic_downheap(cpu, 0); + else { + if (cyclic_upheap(cpu, i) == 0) { + /* + * The upheap didn't propagate to the root; if it + * didn't propagate at all, we need to downheap. + */ + if (heap[i] == last) + cyclic_downheap(cpu, i); + goto done; + } + } + + /* + * We're here because we changed the root; we need to reprogram + * the clock source. + */ + cyclic = &cpu->cyp_cyclics[heap[0]]; + + ASSERT(nelems != 0); + cyclic_reprogram(cpu, cyclic->cy_expire); + +done: + mtx_unlock_spin(&cpu->cyp_mtx); + + return (1); +} + +static void +cyclic_configure(cpu_t *c) +{ + cyc_cpu_t *cpu = malloc(sizeof(cyc_cpu_t), M_CYCLIC, M_ZERO | M_WAITOK); + cyc_backend_t *nbe = malloc(sizeof(cyc_backend_t), M_CYCLIC, M_ZERO | M_WAITOK); + + ASSERT(MUTEX_HELD(&cpu_lock)); + + if (cyclic_id_cache == NULL) + cyclic_id_cache = kmem_cache_create("cyclic_id_cache", + sizeof (cyc_id_t), 0, NULL, NULL, NULL, NULL, NULL, 0); + + cpu->cyp_cpu = c; + + cpu->cyp_size = 1; + cpu->cyp_heap = malloc(sizeof(cyc_index_t), M_CYCLIC, M_ZERO | M_WAITOK); + cpu->cyp_cyclics = malloc(sizeof(cyclic_t), M_CYCLIC, M_ZERO | M_WAITOK); + cpu->cyp_cyclics->cy_flags = CYF_FREE; + + mtx_init(&cpu->cyp_mtx, "cyclic cpu", NULL, MTX_SPIN); + + /* + * Setup the backend for this CPU. + */ + bcopy(&cyclic_backend, nbe, sizeof (cyc_backend_t)); + if (nbe->cyb_configure != NULL) + nbe->cyb_arg = nbe->cyb_configure(c); + cpu->cyp_backend = nbe; + + /* + * On platforms where stray interrupts may be taken during startup, + * the CPU's cpu_cyclic pointer serves as an indicator that the + * cyclic subsystem for this CPU is prepared to field interrupts. + */ + membar_producer(); + + c->cpu_cyclic = cpu; +} + +static void +cyclic_unconfigure(cpu_t *c) +{ + cyc_cpu_t *cpu = c->cpu_cyclic; + cyc_backend_t *be = cpu->cyp_backend; + cyb_arg_t bar = be->cyb_arg; + + ASSERT(MUTEX_HELD(&cpu_lock)); + + c->cpu_cyclic = NULL; + + /* + * Let the backend know that the CPU is being yanked, and free up + * the backend structure. + */ + if (be->cyb_unconfigure != NULL) + be->cyb_unconfigure(bar); + free(be, M_CYCLIC); + cpu->cyp_backend = NULL; + + mtx_destroy(&cpu->cyp_mtx); + + /* Finally, clean up our remaining dynamic structures. */ + free(cpu->cyp_cyclics, M_CYCLIC); + free(cpu->cyp_heap, M_CYCLIC); + free(cpu, M_CYCLIC); +} + +static void +cyclic_omni_start(cyc_id_t *idp, cyc_cpu_t *cpu) +{ + cyc_omni_handler_t *omni = &idp->cyi_omni_hdlr; + cyc_omni_cpu_t *ocpu = malloc(sizeof(cyc_omni_cpu_t), M_CYCLIC , M_WAITOK); + cyc_handler_t hdlr; + cyc_time_t when; + + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(idp->cyi_cpu == NULL); + + hdlr.cyh_func = NULL; + hdlr.cyh_arg = NULL; + + when.cyt_when = 0; + when.cyt_interval = 0; + + omni->cyo_online(omni->cyo_arg, cpu->cyp_cpu, &hdlr, &when); + + ASSERT(hdlr.cyh_func != NULL); + ASSERT(when.cyt_when >= 0 && when.cyt_interval > 0); + + ocpu->cyo_cpu = cpu; + ocpu->cyo_arg = hdlr.cyh_arg; + ocpu->cyo_ndx = cyclic_add_here(cpu, &hdlr, &when, 0); + ocpu->cyo_next = idp->cyi_omni_list; + idp->cyi_omni_list = ocpu; +} + +static void +cyclic_omni_stop(cyc_id_t *idp, cyc_cpu_t *cpu) +{ + cyc_omni_handler_t *omni = &idp->cyi_omni_hdlr; + cyc_omni_cpu_t *ocpu = idp->cyi_omni_list, *prev = NULL; + + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(idp->cyi_cpu == NULL); + ASSERT(ocpu != NULL); + + while (ocpu != NULL && ocpu->cyo_cpu != cpu) { + prev = ocpu; + ocpu = ocpu->cyo_next; + } + + /* + * We _must_ have found an cyc_omni_cpu which corresponds to this + * CPU -- the definition of an omnipresent cyclic is that it runs + * on all online CPUs. + */ + ASSERT(ocpu != NULL); + + if (prev == NULL) { + idp->cyi_omni_list = ocpu->cyo_next; + } else { + prev->cyo_next = ocpu->cyo_next; + } + + (void) cyclic_remove_here(ocpu->cyo_cpu, ocpu->cyo_ndx, NULL, CY_WAIT); + + /* + * The cyclic has been removed from this CPU; time to call the + * omnipresent offline handler. + */ + if (omni->cyo_offline != NULL) + omni->cyo_offline(omni->cyo_arg, cpu->cyp_cpu, ocpu->cyo_arg); + + free(ocpu, M_CYCLIC); +} + +static cyc_id_t * +cyclic_new_id(void) +{ + cyc_id_t *idp; + + ASSERT(MUTEX_HELD(&cpu_lock)); + + idp = kmem_cache_alloc(cyclic_id_cache, KM_SLEEP); + + /* + * The cyi_cpu field of the cyc_id_t structure tracks the CPU + * associated with the cyclic. If and only if this field is NULL, the + * cyc_id_t is an omnipresent cyclic. Note that cyi_omni_list may be + * NULL for an omnipresent cyclic while the cyclic is being created + * or destroyed. + */ + idp->cyi_cpu = NULL; + idp->cyi_ndx = 0; + + idp->cyi_next = cyclic_id_head; + idp->cyi_prev = NULL; + idp->cyi_omni_list = NULL; + + if (cyclic_id_head != NULL) { + ASSERT(cyclic_id_head->cyi_prev == NULL); + cyclic_id_head->cyi_prev = idp; + } + + cyclic_id_head = idp; + + return (idp); +} + +/* + * cyclic_id_t cyclic_add(cyc_handler_t *, cyc_time_t *) + * + * Overview + * + * cyclic_add() will create an unbound cyclic with the specified handler and + * interval. The cyclic will run on a CPU which both has interrupts enabled + * and is in the system CPU partition. + * + * Arguments and notes + * + * As its first argument, cyclic_add() takes a cyc_handler, which has the + * following members: + * + * cyc_func_t cyh_func <-- Cyclic handler + * void *cyh_arg <-- Argument to cyclic handler + * + * In addition to a cyc_handler, cyclic_add() takes a cyc_time, which + * has the following members: + * + * hrtime_t cyt_when <-- Absolute time, in nanoseconds since boot, at + * which to start firing + * hrtime_t cyt_interval <-- Length of interval, in nanoseconds + * + * gethrtime() is the time source for nanoseconds since boot. If cyt_when + * is set to 0, the cyclic will start to fire when cyt_interval next + * divides the number of nanoseconds since boot. + * + * The cyt_interval field _must_ be filled in by the caller; one-shots are + * _not_ explicitly supported by the cyclic subsystem (cyclic_add() will + * assert that cyt_interval is non-zero). The maximum value for either + * field is INT64_MAX; the caller is responsible for assuring that + * cyt_when + cyt_interval <= INT64_MAX. Neither field may be negative. + * + * For an arbitrary time t in the future, the cyclic handler is guaranteed + * to have been called (t - cyt_when) / cyt_interval times. This will + * be true even if interrupts have been disabled for periods greater than + * cyt_interval nanoseconds. In order to compensate for such periods, + * the cyclic handler may be called a finite number of times with an + * arbitrarily small interval. + * + * The cyclic subsystem will not enforce any lower bound on the interval; + * if the interval is less than the time required to process an interrupt, + * the CPU will wedge. It's the responsibility of the caller to assure that + * either the value of the interval is sane, or that its caller has + * sufficient privilege to deny service (i.e. its caller is root). + * + * Return value + * + * cyclic_add() returns a cyclic_id_t, which is guaranteed to be a value + * other than CYCLIC_NONE. cyclic_add() cannot fail. + * + * Caller's context + * + * cpu_lock must be held by the caller, and the caller must not be in + * interrupt context. cyclic_add() will perform a KM_SLEEP kernel + * memory allocation, so the usual rules (e.g. p_lock cannot be held) + * apply. A cyclic may be added even in the presence of CPUs that have + * not been configured with respect to the cyclic subsystem, but only + * configured CPUs will be eligible to run the new cyclic. + * + * Cyclic handler's context + * + * Cyclic handlers will be executed in the interrupt context corresponding + * to the specified level (i.e. either high, lock or low level). The + * usual context rules apply. + * + * A cyclic handler may not grab ANY locks held by the caller of any of + * cyclic_add() or cyclic_remove(); the implementation of these functions + * may require blocking on cyclic handler completion. + * Moreover, cyclic handlers may not make any call back into the cyclic + * subsystem. + */ +cyclic_id_t +cyclic_add(cyc_handler_t *hdlr, cyc_time_t *when) +{ + cyc_id_t *idp = cyclic_new_id(); + solaris_cpu_t *c = &solaris_cpu[curcpu]; + + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(when->cyt_when >= 0 && when->cyt_interval > 0); + + idp->cyi_cpu = c->cpu_cyclic; + idp->cyi_ndx = cyclic_add_here(idp->cyi_cpu, hdlr, when, 0); + + return ((uintptr_t)idp); +} + +/* + * cyclic_id_t cyclic_add_omni(cyc_omni_handler_t *) + * + * Overview + * + * cyclic_add_omni() will create an omnipresent cyclic with the specified + * online and offline handlers. Omnipresent cyclics run on all online + * CPUs, including CPUs which have unbound interrupts disabled. + * + * Arguments + * + * As its only argument, cyclic_add_omni() takes a cyc_omni_handler, which + * has the following members: + * + * void (*cyo_online)() <-- Online handler + * void (*cyo_offline)() <-- Offline handler + * void *cyo_arg <-- Argument to be passed to on/offline handlers + * + * Online handler + * + * The cyo_online member is a pointer to a function which has the following + * four arguments: + * + * void * <-- Argument (cyo_arg) + * cpu_t * <-- Pointer to CPU about to be onlined + * cyc_handler_t * <-- Pointer to cyc_handler_t; must be filled in + * by omni online handler + * cyc_time_t * <-- Pointer to cyc_time_t; must be filled in by + * omni online handler + * + * The omni cyclic online handler is always called _before_ the omni + * cyclic begins to fire on the specified CPU. As the above argument + * description implies, the online handler must fill in the two structures + * passed to it: the cyc_handler_t and the cyc_time_t. These are the + * same two structures passed to cyclic_add(), outlined above. This + * allows the omni cyclic to have maximum flexibility; different CPUs may + * optionally + * + * (a) have different intervals + * (b) be explicitly in or out of phase with one another + * (c) have different handlers + * (d) have different handler arguments + * (e) fire at different levels + * + * Of these, (e) seems somewhat dubious, but is nonetheless allowed. + * + * The omni online handler is called in the same context as cyclic_add(), + * and has the same liberties: omni online handlers may perform KM_SLEEP + * kernel memory allocations, and may grab locks which are also acquired + * by cyclic handlers. However, omni cyclic online handlers may _not_ + * call back into the cyclic subsystem, and should be generally careful + * about calling into arbitrary kernel subsystems. + * + * Offline handler + * + * The cyo_offline member is a pointer to a function which has the following + * three arguments: + * + * void * <-- Argument (cyo_arg) + * cpu_t * <-- Pointer to CPU about to be offlined + * void * <-- CPU's cyclic argument (that is, value + * to which cyh_arg member of the cyc_handler_t + * was set in the omni online handler) + * + * The omni cyclic offline handler is always called _after_ the omni + * cyclic has ceased firing on the specified CPU. Its purpose is to + * allow cleanup of any resources dynamically allocated in the omni cyclic + * online handler. The context of the offline handler is identical to + * that of the online handler; the same constraints and liberties apply. + * + * The offline handler is optional; it may be NULL. + * + * Return value + * + * cyclic_add_omni() returns a cyclic_id_t, which is guaranteed to be a + * value other than CYCLIC_NONE. cyclic_add_omni() cannot fail. + * + * Caller's context + * + * The caller's context is identical to that of cyclic_add(), specified + * above. + */ +cyclic_id_t +cyclic_add_omni(cyc_omni_handler_t *omni) +{ + cyc_id_t *idp = cyclic_new_id(); + cyc_cpu_t *cpu; + cpu_t *c; + int i; + + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(omni != NULL && omni->cyo_online != NULL); + + idp->cyi_omni_hdlr = *omni; + + for (i = 0; i < MAXCPU; i++) { + if (pcpu_find(i) == NULL) + continue; + + c = &solaris_cpu[i]; + + if ((cpu = c->cpu_cyclic) == NULL) + continue; + + cyclic_omni_start(idp, cpu); + } + + /* + * We must have found at least one online CPU on which to run + * this cyclic. + */ + ASSERT(idp->cyi_omni_list != NULL); + ASSERT(idp->cyi_cpu == NULL); + + return ((uintptr_t)idp); +} + +/* + * void cyclic_remove(cyclic_id_t) + * + * Overview + * + * cyclic_remove() will remove the specified cyclic from the system. + * + * Arguments and notes + * + * The only argument is a cyclic_id returned from either cyclic_add() or + * cyclic_add_omni(). + * + * By the time cyclic_remove() returns, the caller is guaranteed that the + * removed cyclic handler has completed execution (this is the same + * semantic that untimeout() provides). As a result, cyclic_remove() may + * need to block, waiting for the removed cyclic to complete execution. + * This leads to an important constraint on the caller: no lock may be + * held across cyclic_remove() that also may be acquired by a cyclic + * handler. + * + * Return value + * + * None; cyclic_remove() always succeeds. + * + * Caller's context + * + * cpu_lock must be held by the caller, and the caller must not be in + * interrupt context. The caller may not hold any locks which are also + * grabbed by any cyclic handler. See "Arguments and notes", above. + */ +void +cyclic_remove(cyclic_id_t id) +{ + cyc_id_t *idp = (cyc_id_t *)id; + cyc_id_t *prev = idp->cyi_prev, *next = idp->cyi_next; + cyc_cpu_t *cpu = idp->cyi_cpu; + + ASSERT(MUTEX_HELD(&cpu_lock)); + + if (cpu != NULL) { + (void) cyclic_remove_here(cpu, idp->cyi_ndx, NULL, CY_WAIT); + } else { + ASSERT(idp->cyi_omni_list != NULL); + while (idp->cyi_omni_list != NULL) + cyclic_omni_stop(idp, idp->cyi_omni_list->cyo_cpu); + } + + if (prev != NULL) { + ASSERT(cyclic_id_head != idp); + prev->cyi_next = next; + } else { + ASSERT(cyclic_id_head == idp); + cyclic_id_head = next; + } + + if (next != NULL) + next->cyi_prev = prev; + + kmem_cache_free(cyclic_id_cache, idp); +} + +static void +cyclic_init(cyc_backend_t *be) +{ + ASSERT(MUTEX_HELD(&cpu_lock)); + + /* + * Copy the passed cyc_backend into the backend template. This must + * be done before the CPU can be configured. + */ + bcopy(be, &cyclic_backend, sizeof (cyc_backend_t)); + + cyclic_configure(&solaris_cpu[curcpu]); +} + +/* + * It is assumed that cyclic_mp_init() is called some time after cyclic + * init (and therefore, after cpu0 has been initialized). We grab cpu_lock, + * find the already initialized CPU, and initialize every other CPU with the + * same backend. + */ +static void +cyclic_mp_init(void) +{ + cpu_t *c; + int i; + + mutex_enter(&cpu_lock); + + for (i = 0; i <= mp_maxid; i++) { + if (pcpu_find(i) == NULL) + continue; + + c = &solaris_cpu[i]; + + if (c->cpu_cyclic == NULL) + cyclic_configure(c); + } + + mutex_exit(&cpu_lock); +} + +static void +cyclic_uninit(void) +{ + struct pcpu *pc; + cpu_t *c; + int id; + + for (id = 0; id <= mp_maxid; id++) { + if ((pc = pcpu_find(id)) == NULL) + continue; + + c = &solaris_cpu[id]; + + if (c->cpu_cyclic == NULL) + continue; + + cyclic_unconfigure(c); + } + + if (cyclic_id_cache != NULL) + kmem_cache_destroy(cyclic_id_cache); +} + +#include "cyclic_machdep.c" + +/* + * Cyclic subsystem initialisation. + */ +static void +cyclic_load(void *dummy) +{ + mutex_enter(&cpu_lock); + + /* Initialise the machine-dependent backend. */ + cyclic_machdep_init(); + + mutex_exit(&cpu_lock); +} + +SYSINIT(cyclic_register, SI_SUB_CYCLIC, SI_ORDER_SECOND, cyclic_load, NULL); + +static void +cyclic_unload(void) +{ + mutex_enter(&cpu_lock); + + /* Uninitialise the machine-dependent backend. */ + cyclic_machdep_uninit(); + + mutex_exit(&cpu_lock); +} + +SYSUNINIT(cyclic_unregister, SI_SUB_CYCLIC, SI_ORDER_SECOND, cyclic_unload, NULL); + +/* ARGSUSED */ +static int +cyclic_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + return (error); +} + +DEV_MODULE(cyclic, cyclic_modevent, NULL); +MODULE_VERSION(cyclic, 1); +MODULE_DEPEND(cyclic, opensolaris, 1, 1, 1); diff --git a/sys/cddl/dev/cyclic/cyclic_test.c b/sys/cddl/dev/cyclic/cyclic_test.c new file mode 100644 index 000000000000..063dbc77e7a3 --- /dev/null +++ b/sys/cddl/dev/cyclic/cyclic_test.c @@ -0,0 +1,301 @@ +/*- + * Copyright 2007 John Birrell <jb@FreeBSD.org> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + */ + +#include <sys/cdefs.h> +#include <sys/systm.h> +#include <sys/kernel.h> +#include <sys/conf.h> +#include <sys/kthread.h> +#include <sys/module.h> +#include <sys/sysctl.h> +#include <sys/cyclic.h> +#include <sys/time.h> + +static struct timespec test_001_start; + +static void +cyclic_test_001_func(void *arg) +{ + struct timespec ts; + + nanotime(&ts); + timespecsub(&ts,&test_001_start); + printf("%s: called after %lu.%09lu on curcpu %d\n",__func__,(u_long) ts.tv_sec,(u_long) ts.tv_nsec, curcpu); +} + +static void +cyclic_test_001(void) +{ + int error = 0; + cyc_handler_t hdlr; + cyc_time_t when; + cyclic_id_t id; + + printf("%s: starting\n",__func__); + + hdlr.cyh_func = (cyc_func_t) cyclic_test_001_func; + hdlr.cyh_arg = 0; + + when.cyt_when = 0; + when.cyt_interval = 1000000000; + + nanotime(&test_001_start); + + mutex_enter(&cpu_lock); + + id = cyclic_add(&hdlr, &when); + + mutex_exit(&cpu_lock); + + DELAY(1200000); + + mutex_enter(&cpu_lock); + + cyclic_remove(id); + + mutex_exit(&cpu_lock); + + printf("%s: %s\n",__func__, error == 0 ? "passed":"failed"); +} + +static struct timespec test_002_start; + +static void +cyclic_test_002_func(void *arg) +{ + struct timespec ts; + + nanotime(&ts); + timespecsub(&ts,&test_002_start); + printf("%s: called after %lu.%09lu on curcpu %d\n",__func__,(u_long) ts.tv_sec,(u_long) ts.tv_nsec, curcpu); +} + +static void +cyclic_test_002_online(void *arg, cpu_t *c, cyc_handler_t *hdlr, cyc_time_t *t) +{ + printf("%s: online on curcpu %d\n",__func__, curcpu); + hdlr->cyh_func = cyclic_test_002_func; + hdlr->cyh_arg = NULL; + t->cyt_when = 0; + t->cyt_interval = 1000000000; +} + +static void +cyclic_test_002_offline(void *arg, cpu_t *c, void *arg1) +{ + printf("%s: offline on curcpu %d\n",__func__, curcpu); +} + +static void +cyclic_test_002(void) +{ + int error = 0; + cyc_omni_handler_t hdlr; + cyclic_id_t id; + + printf("%s: starting\n",__func__); + + hdlr.cyo_online = cyclic_test_002_online; + hdlr.cyo_offline = cyclic_test_002_offline; + hdlr.cyo_arg = NULL; + + nanotime(&test_002_start); + + mutex_enter(&cpu_lock); + + id = cyclic_add_omni(&hdlr); + + mutex_exit(&cpu_lock); + + DELAY(1200000); + + mutex_enter(&cpu_lock); + + cyclic_remove(id); + + mutex_exit(&cpu_lock); + + printf("%s: %s\n",__func__, error == 0 ? "passed":"failed"); +} + +static struct timespec test_003_start; + +static void +cyclic_test_003_func(void *arg) +{ + struct timespec ts; + + nanotime(&ts); + timespecsub(&ts,&test_003_start); + printf("%s: called after %lu.%09lu on curcpu %d id %ju\n",__func__,(u_long) ts.tv_sec,(u_long) ts.tv_nsec, curcpu, (uintmax_t)(uintptr_t) arg); +} + +static void +cyclic_test_003(void) +{ + int error = 0; + cyc_handler_t hdlr; + cyc_time_t when; + cyclic_id_t id; + cyclic_id_t id1; + cyclic_id_t id2; + cyclic_id_t id3; + + printf("%s: starting\n",__func__); + + hdlr.cyh_func = (cyc_func_t) cyclic_test_003_func; + + when.cyt_when = 0; + + nanotime(&test_003_start); + + mutex_enter(&cpu_lock); + + when.cyt_interval = 200000000; + hdlr.cyh_arg = (void *) 0UL; + id = cyclic_add(&hdlr, &when); + + when.cyt_interval = 400000000; + hdlr.cyh_arg = (void *) 1UL; + id1 = cyclic_add(&hdlr, &when); + + hdlr.cyh_arg = (void *) 2UL; + when.cyt_interval = 1000000000; + id2 = cyclic_add(&hdlr, &when); + + hdlr.cyh_arg = (void *) 3UL; + when.cyt_interval = 1300000000; + id3 = cyclic_add(&hdlr, &when); + + mutex_exit(&cpu_lock); + + DELAY(1200000); + + mutex_enter(&cpu_lock); + + cyclic_remove(id); + cyclic_remove(id1); + cyclic_remove(id2); + cyclic_remove(id3); + + mutex_exit(&cpu_lock); + + printf("%s: %s\n",__func__, error == 0 ? "passed":"failed"); +} + +/* Kernel thread command routine. */ +static void +cyclic_run_tests(void *arg) +{ + intptr_t cmd = (intptr_t) arg; + + switch (cmd) { + case 1: + cyclic_test_001(); + break; + case 2: + cyclic_test_002(); + break; + case 3: + cyclic_test_003(); + break; + default: + cyclic_test_001(); + cyclic_test_002(); + cyclic_test_003(); + break; + } + + printf("%s: finished\n",__func__); + + kthread_exit(); +} + +static int +cyclic_test(SYSCTL_HANDLER_ARGS) +{ + int error, cmd = 0; + + error = sysctl_wire_old_buffer(req, sizeof(int)); + if (error == 0) + error = sysctl_handle_int(oidp, &cmd, 0, req); + if (error != 0 || req->newptr == NULL) + return (error); + + /* Check for command validity. */ + switch (cmd) { + case 1: + case 2: + case -1: + /* + * Execute the tests in a kernel thread to avoid blocking + * the sysctl. Look for the results in the syslog. + */ + error = kthread_add(cyclic_run_tests, (void *)(uintptr_t) cmd, + NULL, NULL, 0, 0, "cyctest%d", cmd); + break; + default: + printf("Usage: debug.cyclic.test=(1..9) or -1 for all tests\n"); + error = EINVAL; + break; + } + + return (error); +} + +SYSCTL_NODE(_debug, OID_AUTO, cyclic, CTLFLAG_RW, NULL, "Cyclic nodes"); +SYSCTL_PROC(_debug_cyclic, OID_AUTO, test, CTLTYPE_INT | CTLFLAG_RW, 0, 0, + cyclic_test, "I", "Enables a cyclic test. Use -1 for all tests."); + +static int +cyclic_test_modevent(module_t mod, int type, void *data) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + return (error); +} + +DEV_MODULE(cyclic_test, cyclic_test_modevent, NULL); +MODULE_VERSION(cyclic_test, 1); +MODULE_DEPEND(cyclic_test, cyclic, 1, 1, 1); +MODULE_DEPEND(cyclic_test, opensolaris, 1, 1, 1); diff --git a/sys/cddl/dev/cyclic/i386/cyclic_machdep.c b/sys/cddl/dev/cyclic/i386/cyclic_machdep.c new file mode 100644 index 000000000000..fa40db9329c1 --- /dev/null +++ b/sys/cddl/dev/cyclic/i386/cyclic_machdep.c @@ -0,0 +1,133 @@ +/*- + * Copyright 2006-2008 John Birrell <jb@FreeBSD.org> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + */ + +static void enable(cyb_arg_t); +static void disable(cyb_arg_t); +static void reprogram(cyb_arg_t, hrtime_t); +static void xcall(cyb_arg_t, cpu_t *, cyc_func_t, void *); + +static cyc_backend_t be = { + NULL, /* cyb_configure */ + NULL, /* cyb_unconfigure */ + enable, + disable, + reprogram, + xcall, + NULL /* cyb_arg_t cyb_arg */ +}; + +static void +cyclic_ap_start(void *dummy) +{ + /* Initialise the rest of the CPUs. */ + cyclic_mp_init(); +} + +SYSINIT(cyclic_ap_start, SI_SUB_SMP, SI_ORDER_ANY, cyclic_ap_start, NULL); + +/* + * Machine dependent cyclic subsystem initialisation. + */ +static void +cyclic_machdep_init(void) +{ + /* Register the cyclic backend. */ + cyclic_init(&be); +} + +static void +cyclic_machdep_uninit(void) +{ + int i; + + for (i = 0; i <= mp_maxid; i++) + /* Reset the cyclic clock callback hook. */ + lapic_cyclic_clock_func[i] = NULL; + + /* De-register the cyclic backend. */ + cyclic_uninit(); +} + +static hrtime_t exp_due[MAXCPU]; + +/* + * This function is the one registered by the machine dependent + * initialiser as the callback for high speed timer events. + */ +static void +cyclic_clock(struct trapframe *frame) +{ + cpu_t *c = &solaris_cpu[curcpu]; + + if (c->cpu_cyclic != NULL && gethrtime() >= exp_due[curcpu]) { + if (TRAPF_USERMODE(frame)) { + c->cpu_profile_pc = 0; + c->cpu_profile_upc = TRAPF_PC(frame); + } else { + c->cpu_profile_pc = TRAPF_PC(frame); + c->cpu_profile_upc = 0; + } + + c->cpu_intr_actv = 1; + + /* Fire any timers that are due. */ + cyclic_fire(c); + + c->cpu_intr_actv = 0; + } +} + +static void enable(cyb_arg_t arg) +{ + /* Register the cyclic clock callback function. */ + lapic_cyclic_clock_func[curcpu] = cyclic_clock; +} + +static void disable(cyb_arg_t arg) +{ + /* Reset the cyclic clock callback function. */ + lapic_cyclic_clock_func[curcpu] = NULL; +} + +static void reprogram(cyb_arg_t arg, hrtime_t exp) +{ + exp_due[curcpu] = exp; +} + +static void xcall(cyb_arg_t arg, cpu_t *c, cyc_func_t func, void *param) +{ + /* + * If the target CPU is the current one, just call the + * function. This covers the non-SMP case. + */ + if (c == &solaris_cpu[curcpu]) + (*func)(param); + else + smp_rendezvous_cpus((cpumask_t) (1 << c->cpuid), NULL, + func, smp_no_rendevous_barrier, param); +} diff --git a/sys/cddl/dev/dtmalloc/dtmalloc.c b/sys/cddl/dev/dtmalloc/dtmalloc.c new file mode 100644 index 000000000000..ca822f988668 --- /dev/null +++ b/sys/cddl/dev/dtmalloc/dtmalloc.c @@ -0,0 +1,220 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * Portions Copyright 2006-2008 John Birrell jb@freebsd.org + * + * $FreeBSD$ + * + */ + +#include <sys/cdefs.h> +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/kernel.h> +#include <sys/malloc.h> +#include <sys/module.h> + +#include <sys/dtrace.h> +#include <sys/dtrace_bsd.h> + +static d_open_t dtmalloc_open; +static int dtmalloc_unload(void); +static void dtmalloc_getargdesc(void *, dtrace_id_t, void *, dtrace_argdesc_t *); +static void dtmalloc_provide(void *, dtrace_probedesc_t *); +static void dtmalloc_destroy(void *, dtrace_id_t, void *); +static void dtmalloc_enable(void *, dtrace_id_t, void *); +static void dtmalloc_disable(void *, dtrace_id_t, void *); +static void dtmalloc_load(void *); + +static struct cdevsw dtmalloc_cdevsw = { + .d_version = D_VERSION, + .d_open = dtmalloc_open, + .d_name = "dtmalloc", +}; + +static dtrace_pattr_t dtmalloc_attr = { +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +}; + +static dtrace_pops_t dtmalloc_pops = { + dtmalloc_provide, + NULL, + dtmalloc_enable, + dtmalloc_disable, + NULL, + NULL, + dtmalloc_getargdesc, + NULL, + NULL, + dtmalloc_destroy +}; + +static struct cdev *dtmalloc_cdev; +static dtrace_provider_id_t dtmalloc_id; + +static void +dtmalloc_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) +{ + const char *p = NULL; + + switch (desc->dtargd_ndx) { + case 0: + p = "struct malloc_type *"; + break; + case 1: + p = "struct malloc_type_internal *"; + break; + case 2: + p = "struct malloc_type_stats *"; + break; + case 3: + p = "unsigned long"; + break; + case 4: + p = "int"; + break; + default: + desc->dtargd_ndx = DTRACE_ARGNONE; + break; + } + + if (p != NULL) + strlcpy(desc->dtargd_native, p, sizeof(desc->dtargd_native)); + + return; +} + +static void +dtmalloc_type_cb(struct malloc_type *mtp, void *arg __unused) +{ + char name[DTRACE_FUNCNAMELEN]; + struct malloc_type_internal *mtip = mtp->ks_handle; + + strlcpy(name, mtp->ks_shortdesc, sizeof(name)); + + if (dtrace_probe_lookup(dtmalloc_id, NULL, name, "malloc") != 0) + return; + + (void) dtrace_probe_create(dtmalloc_id, NULL, name, "malloc", 0, + &mtip->mti_probes[DTMALLOC_PROBE_MALLOC]); + (void) dtrace_probe_create(dtmalloc_id, NULL, name, "free", 0, + &mtip->mti_probes[DTMALLOC_PROBE_FREE]); +} + +static void +dtmalloc_provide(void *arg, dtrace_probedesc_t *desc) +{ + if (desc != NULL) + return; + + malloc_type_list(dtmalloc_type_cb, desc); +} + +static void +dtmalloc_destroy(void *arg, dtrace_id_t id, void *parg) +{ +} + +static void +dtmalloc_enable(void *arg, dtrace_id_t id, void *parg) +{ + uint32_t *p = parg; + *p = id; +} + +static void +dtmalloc_disable(void *arg, dtrace_id_t id, void *parg) +{ + uint32_t *p = parg; + *p = 0; +} + +static void +dtmalloc_load(void *dummy) +{ + /* Create the /dev/dtrace/dtmalloc entry. */ + dtmalloc_cdev = make_dev(&dtmalloc_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, + "dtrace/dtmalloc"); + + if (dtrace_register("dtmalloc", &dtmalloc_attr, DTRACE_PRIV_USER, + NULL, &dtmalloc_pops, NULL, &dtmalloc_id) != 0) + return; + + dtrace_malloc_probe = dtrace_probe; +} + + +static int +dtmalloc_unload() +{ + int error = 0; + + dtrace_malloc_probe = NULL; + + if ((error = dtrace_unregister(dtmalloc_id)) != 0) + return (error); + + destroy_dev(dtmalloc_cdev); + + return (error); +} + +static int +dtmalloc_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + + return (error); +} + +static int +dtmalloc_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) +{ + return (0); +} + +SYSINIT(dtmalloc_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, dtmalloc_load, NULL); +SYSUNINIT(dtmalloc_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, dtmalloc_unload, NULL); + +DEV_MODULE(dtmalloc, dtmalloc_modevent, NULL); +MODULE_VERSION(dtmalloc, 1); +MODULE_DEPEND(dtmalloc, dtrace, 1, 1, 1); +MODULE_DEPEND(dtmalloc, opensolaris, 1, 1, 1); diff --git a/sys/cddl/dev/dtrace/amd64/dis_tables.c b/sys/cddl/dev/dtrace/amd64/dis_tables.c new file mode 100644 index 000000000000..5a5bc25c02c3 --- /dev/null +++ b/sys/cddl/dev/dtrace/amd64/dis_tables.c @@ -0,0 +1,3193 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +/* Copyright (c) 1988 AT&T */ +/* All Rights Reserved */ + + +#if defined(sun) +#pragma ident "@(#)dis_tables.c 1.11 06/03/02 SMI" +#endif + +#include "dis_tables.h" + +/* BEGIN CSTYLED */ + +/* + * Disassembly begins in dis_distable, which is equivalent to the One-byte + * Opcode Map in the Intel IA32 ISA Reference (page A-6 in my copy). The + * decoding loops then traverse out through the other tables as necessary to + * decode a given instruction. + * + * The behavior of this file can be controlled by one of the following flags: + * + * DIS_TEXT Include text for disassembly + * DIS_MEM Include memory-size calculations + * + * Either or both of these can be defined. + * + * This file is not, and will never be, cstyled. If anything, the tables should + * be taken out another tab stop or two so nothing overlaps. + */ + +/* + * These functions must be provided for the consumer to do disassembly. + */ +#ifdef DIS_TEXT +extern char *strncpy(char *, const char *, size_t); +extern size_t strlen(const char *); +extern int strcmp(const char *, const char *); +extern int strncmp(const char *, const char *, size_t); +extern size_t strlcat(char *, const char *, size_t); +#endif + + +#define TERM NULL /* used to indicate that the 'indirect' */ + /* field terminates - no pointer. */ + +/* Used to decode instructions. */ +typedef struct instable { + const struct instable *it_indirect; /* for decode op codes */ + uchar_t it_adrmode; +#ifdef DIS_TEXT + char it_name[NCPS]; + uint_t it_suffix:1; /* mneu + "w", "l", or "d" */ +#endif +#ifdef DIS_MEM + uint_t it_size:16; +#endif + uint_t it_invalid64:1; /* opcode invalid in amd64 */ + uint_t it_always64:1; /* 64 bit when in 64 bit mode */ + uint_t it_invalid32:1; /* invalid in IA32 */ + uint_t it_stackop:1; /* push/pop stack operation */ +} instable_t; + +/* + * Instruction formats. + */ +enum { + UNKNOWN, + MRw, + IMlw, + IMw, + IR, + OA, + AO, + MS, + SM, + Mv, + Mw, + M, /* register or memory */ + Mb, /* register or memory, always byte sized */ + MO, /* memory only (no registers) */ + PREF, + SWAPGS, + R, + RA, + SEG, + MR, + RM, + IA, + MA, + SD, + AD, + SA, + D, + INM, + SO, + BD, + I, + P, + V, + DSHIFT, /* for double shift that has an 8-bit immediate */ + U, + OVERRIDE, + NORM, /* instructions w/o ModR/M byte, no memory access */ + IMPLMEM, /* instructions w/o ModR/M byte, implicit mem access */ + O, /* for call */ + JTAB, /* jump table */ + IMUL, /* for 186 iimul instr */ + CBW, /* so data16 can be evaluated for cbw and variants */ + MvI, /* for 186 logicals */ + ENTER, /* for 186 enter instr */ + RMw, /* for 286 arpl instr */ + Ib, /* for push immediate byte */ + F, /* for 287 instructions */ + FF, /* for 287 instructions */ + FFC, /* for 287 instructions */ + DM, /* 16-bit data */ + AM, /* 16-bit addr */ + LSEG, /* for 3-bit seg reg encoding */ + MIb, /* for 386 logicals */ + SREG, /* for 386 special registers */ + PREFIX, /* a REP instruction prefix */ + LOCK, /* a LOCK instruction prefix */ + INT3, /* The int 3 instruction, which has a fake operand */ + INTx, /* The normal int instruction, with explicit int num */ + DSHIFTcl, /* for double shift that implicitly uses %cl */ + CWD, /* so data16 can be evaluated for cwd and variants */ + RET, /* single immediate 16-bit operand */ + MOVZ, /* for movs and movz, with different size operands */ + XADDB, /* for xaddb */ + MOVSXZ, /* AMD64 mov sign extend 32 to 64 bit instruction */ + +/* + * MMX/SIMD addressing modes. + */ + + MMO, /* Prefixable MMX/SIMD-Int mm/mem -> mm */ + MMOIMPL, /* Prefixable MMX/SIMD-Int mm -> mm (mem) */ + MMO3P, /* Prefixable MMX/SIMD-Int mm -> r32,imm8 */ + MMOM3, /* Prefixable MMX/SIMD-Int mm -> r32 */ + MMOS, /* Prefixable MMX/SIMD-Int mm -> mm/mem */ + MMOMS, /* Prefixable MMX/SIMD-Int mm -> mem */ + MMOPM, /* MMX/SIMD-Int mm/mem -> mm,imm8 */ + MMOPRM, /* Prefixable MMX/SIMD-Int r32/mem -> mm,imm8 */ + MMOSH, /* Prefixable MMX mm,imm8 */ + MM, /* MMX/SIMD-Int mm/mem -> mm */ + MMS, /* MMX/SIMD-Int mm -> mm/mem */ + MMSH, /* MMX mm,imm8 */ + XMMO, /* Prefixable SIMD xmm/mem -> xmm */ + XMMOS, /* Prefixable SIMD xmm -> xmm/mem */ + XMMOPM, /* Prefixable SIMD xmm/mem w/to xmm,imm8 */ + XMMOMX, /* Prefixable SIMD mm/mem -> xmm */ + XMMOX3, /* Prefixable SIMD xmm -> r32 */ + XMMOXMM, /* Prefixable SIMD xmm/mem -> mm */ + XMMOM, /* Prefixable SIMD xmm -> mem */ + XMMOMS, /* Prefixable SIMD mem -> xmm */ + XMM, /* SIMD xmm/mem -> xmm */ + XMMXIMPL, /* SIMD xmm -> xmm (mem) */ + XMM3P, /* SIMD xmm -> r32,imm8 */ + XMMP, /* SIMD xmm/mem w/to xmm,imm8 */ + XMMPRM, /* SIMD r32/mem -> xmm,imm8 */ + XMMS, /* SIMD xmm -> xmm/mem */ + XMMM, /* SIMD mem -> xmm */ + XMMMS, /* SIMD xmm -> mem */ + XMM3MX, /* SIMD r32/mem -> xmm */ + XMM3MXS, /* SIMD xmm -> r32/mem */ + XMMSH, /* SIMD xmm,imm8 */ + XMMXM3, /* SIMD xmm/mem -> r32 */ + XMMX3, /* SIMD xmm -> r32 */ + XMMXMM, /* SIMD xmm/mem -> mm */ + XMMMX, /* SIMD mm -> xmm */ + XMMXM, /* SIMD xmm -> mm */ + XMMFENCE, /* SIMD lfence or mfence */ + XMMSFNC /* SIMD sfence (none or mem) */ +}; + +#define FILL 0x90 /* Fill byte used for alignment (nop) */ + +/* +** Register numbers for the i386 +*/ +#define EAX_REGNO 0 +#define ECX_REGNO 1 +#define EDX_REGNO 2 +#define EBX_REGNO 3 +#define ESP_REGNO 4 +#define EBP_REGNO 5 +#define ESI_REGNO 6 +#define EDI_REGNO 7 + +/* + * modes for immediate values + */ +#define MODE_NONE 0 +#define MODE_IPREL 1 /* signed IP relative value */ +#define MODE_SIGNED 2 /* sign extended immediate */ +#define MODE_IMPLIED 3 /* constant value implied from opcode */ +#define MODE_OFFSET 4 /* offset part of an address */ + +/* + * The letters used in these macros are: + * IND - indirect to another to another table + * "T" - means to Terminate indirections (this is the final opcode) + * "S" - means "operand length suffix required" + * "NS" - means "no suffix" which is the operand length suffix of the opcode + * "Z" - means instruction size arg required + * "u" - means the opcode is invalid in IA32 but valid in amd64 + * "x" - means the opcode is invalid in amd64, but not IA32 + * "y" - means the operand size is always 64 bits in 64 bit mode + * "p" - means push/pop stack operation + */ + +#if defined(DIS_TEXT) && defined(DIS_MEM) +#define IND(table) {table, 0, "", 0, 0, 0, 0, 0, 0} +#define INDx(table) {table, 0, "", 0, 0, 1, 0, 0, 0} +#define TNS(name, amode) {TERM, amode, name, 0, 0, 0, 0, 0, 0} +#define TNSu(name, amode) {TERM, amode, name, 0, 0, 0, 0, 1, 0} +#define TNSx(name, amode) {TERM, amode, name, 0, 0, 1, 0, 0, 0} +#define TNSy(name, amode) {TERM, amode, name, 0, 0, 0, 1, 0, 0} +#define TNSyp(name, amode) {TERM, amode, name, 0, 0, 0, 1, 0, 1} +#define TNSZ(name, amode, sz) {TERM, amode, name, 0, sz, 0, 0, 0, 0} +#define TNSZy(name, amode, sz) {TERM, amode, name, 0, sz, 0, 1, 0, 0} +#define TS(name, amode) {TERM, amode, name, 1, 0, 0, 0, 0, 0} +#define TSx(name, amode) {TERM, amode, name, 1, 0, 1, 0, 0, 0} +#define TSy(name, amode) {TERM, amode, name, 1, 0, 0, 1, 0, 0} +#define TSp(name, amode) {TERM, amode, name, 1, 0, 0, 0, 0, 1} +#define TSZ(name, amode, sz) {TERM, amode, name, 1, sz, 0, 0, 0, 0} +#define TSZx(name, amode, sz) {TERM, amode, name, 1, sz, 1, 0, 0, 0} +#define TSZy(name, amode, sz) {TERM, amode, name, 1, sz, 0, 1, 0, 0} +#define INVALID {TERM, UNKNOWN, "", 0, 0, 0, 0, 0} +#elif defined(DIS_TEXT) +#define IND(table) {table, 0, "", 0, 0, 0, 0, 0} +#define INDx(table) {table, 0, "", 0, 1, 0, 0, 0} +#define TNS(name, amode) {TERM, amode, name, 0, 0, 0, 0, 0} +#define TNSu(name, amode) {TERM, amode, name, 0, 0, 0, 1, 0} +#define TNSx(name, amode) {TERM, amode, name, 0, 1, 0, 0, 0} +#define TNSy(name, amode) {TERM, amode, name, 0, 0, 1, 0, 0} +#define TNSyp(name, amode) {TERM, amode, name, 0, 0, 1, 0, 1} +#define TNSZ(name, amode, sz) {TERM, amode, name, 0, 0, 0, 0, 0} +#define TNSZy(name, amode, sz) {TERM, amode, name, 0, 0, 1, 0, 0} +#define TS(name, amode) {TERM, amode, name, 1, 0, 0, 0, 0} +#define TSx(name, amode) {TERM, amode, name, 1, 1, 0, 0, 0} +#define TSy(name, amode) {TERM, amode, name, 1, 0, 1, 0, 0} +#define TSp(name, amode) {TERM, amode, name, 1, 0, 0, 0, 1} +#define TSZ(name, amode, sz) {TERM, amode, name, 1, 0, 0, 0, 0} +#define TSZx(name, amode, sz) {TERM, amode, name, 1, 1, 0, 0, 0} +#define TSZy(name, amode, sz) {TERM, amode, name, 1, 0, 1, 0, 0} +#define INVALID {TERM, UNKNOWN, "", 0, 0, 0, 0, 0} +#elif defined(DIS_MEM) +#define IND(table) {table, 0, 0, 0, 0, 0, 0} +#define INDx(table) {table, 0, 0, 1, 0, 0, 0} +#define TNS(name, amode) {TERM, amode, 0, 0, 0, 0, 0} +#define TNSu(name, amode) {TERM, amode, 0, 0, 0, 1, 0} +#define TNSy(name, amode) {TERM, amode, 0, 0, 1, 0, 0} +#define TNSyp(name, amode) {TERM, amode, 0, 0, 1, 0, 1} +#define TNSx(name, amode) {TERM, amode, 0, 1, 0, 0, 0} +#define TNSZ(name, amode, sz) {TERM, amode, sz, 0, 0, 0, 0} +#define TNSZy(name, amode, sz) {TERM, amode, sz, 0, 1, 0, 0} +#define TS(name, amode) {TERM, amode, 0, 0, 0, 0, 0} +#define TSx(name, amode) {TERM, amode, 0, 1, 0, 0, 0} +#define TSy(name, amode) {TERM, amode, 0, 0, 1, 0, 0} +#define TSp(name, amode) {TERM, amode, 0, 0, 0, 0, 1} +#define TSZ(name, amode, sz) {TERM, amode, sz, 0, 0, 0, 0} +#define TSZx(name, amode, sz) {TERM, amode, sz, 1, 0, 0, 0} +#define TSZy(name, amode, sz) {TERM, amode, sz, 0, 1, 0, 0} +#define INVALID {TERM, UNKNOWN, 0, 0, 0, 0, 0} +#else +#define IND(table) {table[0], 0, 0, 0, 0, 0} +#define INDx(table) {table[0], 0, 1, 0, 0, 0} +#define TNS(name, amode) {TERM, amode, 0, 0, 0, 0} +#define TNSu(name, amode) {TERM, amode, 0, 0, 1, 0} +#define TNSy(name, amode) {TERM, amode, 0, 1, 0, 0} +#define TNSyp(name, amode) {TERM, amode, 0, 1, 0, 1} +#define TNSx(name, amode) {TERM, amode, 1, 0, 0, 0} +#define TNSZ(name, amode, sz) {TERM, amode, 0, 0, 0, 0} +#define TNSZy(name, amode, sz) {TERM, amode, 0, 1, 0, 0} +#define TS(name, amode) {TERM, amode, 0, 0, 0, 0} +#define TSx(name, amode) {TERM, amode, 1, 0, 0, 0} +#define TSy(name, amode) {TERM, amode, 0, 1, 0, 0} +#define TSp(name, amode) {TERM, amode, 0, 0, 0, 1} +#define TSZ(name, amode, sz) {TERM, amode, 0, 0, 0, 0} +#define TSZx(name, amode, sz) {TERM, amode, 1, 0, 0, 0} +#define TSZy(name, amode, sz) {TERM, amode, 0, 1, 0, 0} +#define INVALID {TERM, UNKNOWN, 0, 0, 0, 0} +#endif + +#ifdef DIS_TEXT +/* + * this decodes the r_m field for mode's 0, 1, 2 in 16 bit mode + */ +const char *const dis_addr16[3][8] = { +"(%bx,%si)", "(%bx,%di)", "(%bp,%si)", "(%bp,%di)", "(%si)", "(%di)", "", + "(%bx)", +"(%bx,%si)", "(%bx,%di)", "(%bp,%si)", "(%bp,%di)", "(%si)", "(%di", "(%bp)", + "(%bx)", +"(%bx,%si)", "(%bx,%di)", "(%bp,%si)", "(%bp,%di)", "(%si)", "(%di)", "(%bp)", + "(%bx)", +}; + + +/* + * This decodes 32 bit addressing mode r_m field for modes 0, 1, 2 + */ +const char *const dis_addr32_mode0[16] = { + "(%eax)", "(%ecx)", "(%edx)", "(%ebx)", "", "", "(%esi)", "(%edi)", + "(%r8d)", "(%r9d)", "(%r10d)", "(%r11d)", "", "", "(%r14d)", "(%r15d)" +}; + +const char *const dis_addr32_mode12[16] = { + "(%eax)", "(%ecx)", "(%edx)", "(%ebx)", "", "(%ebp)", "(%esi)", "(%edi)", + "(%r8d)", "(%r9d)", "(%r10d)", "(%r11d)", "", "(%r13d)", "(%r14d)", "(%r15d)" +}; + +/* + * This decodes 64 bit addressing mode r_m field for modes 0, 1, 2 + */ +const char *const dis_addr64_mode0[16] = { + "(%rax)", "(%rcx)", "(%rdx)", "(%rbx)", "", "(%rip)", "(%rsi)", "(%rdi)", + "(%r8)", "(%r9)", "(%r10)", "(%r11)", "(%r12)", "(%rip)", "(%r14)", "(%r15)" +}; +const char *const dis_addr64_mode12[16] = { + "(%rax)", "(%rcx)", "(%rdx)", "(%rbx)", "", "(%rbp)", "(%rsi)", "(%rdi)", + "(%r8)", "(%r9)", "(%r10)", "(%r11)", "(%r12)", "(%r13)", "(%r14)", "(%r15)" +}; + +/* + * decode for scale from SIB byte + */ +const char *const dis_scale_factor[4] = { ")", ",2)", ",4)", ",8)" }; + +/* + * register decoding for normal references to registers (ie. not addressing) + */ +const char *const dis_REG8[16] = { + "%al", "%cl", "%dl", "%bl", "%ah", "%ch", "%dh", "%bh", + "%r8b", "%r9b", "%r10b", "%r11b", "%r12b", "%r13b", "%r14b", "%r15b" +}; + +const char *const dis_REG8_REX[16] = { + "%al", "%cl", "%dl", "%bl", "%spl", "%bpl", "%sil", "%dil", + "%r8b", "%r9b", "%r10b", "%r11b", "%r12b", "%r13b", "%r14b", "%r15b" +}; + +const char *const dis_REG16[16] = { + "%ax", "%cx", "%dx", "%bx", "%sp", "%bp", "%si", "%di", + "%r8w", "%r9w", "%r10w", "%r11w", "%r12w", "%r13w", "%r14w", "%r15w" +}; + +const char *const dis_REG32[16] = { + "%eax", "%ecx", "%edx", "%ebx", "%esp", "%ebp", "%esi", "%edi", + "%r8d", "%r9d", "%r10d", "%r11d", "%r12d", "%r13d", "%r14d", "%r15d" +}; + +const char *const dis_REG64[16] = { + "%rax", "%rcx", "%rdx", "%rbx", "%rsp", "%rbp", "%rsi", "%rdi", + "%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15" +}; + +const char *const dis_DEBUGREG[16] = { + "%db0", "%db1", "%db2", "%db3", "%db4", "%db5", "%db6", "%db7", + "%db8", "%db9", "%db10", "%db11", "%db12", "%db13", "%db14", "%db15" +}; + +const char *const dis_CONTROLREG[16] = { + "%cr0", "%cr1", "%cr2", "%cr3", "%cr4", "%cr5?", "%cr6?", "%cr7?", + "%cr8", "%cr9?", "%cr10?", "%cr11?", "%cr12?", "%cr13?", "%cr14?", "%cr15?" +}; + +const char *const dis_TESTREG[16] = { + "%tr0?", "%tr1?", "%tr2?", "%tr3", "%tr4", "%tr5", "%tr6", "%tr7", + "%tr0?", "%tr1?", "%tr2?", "%tr3", "%tr4", "%tr5", "%tr6", "%tr7" +}; + +const char *const dis_MMREG[16] = { + "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7", + "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7" +}; + +const char *const dis_XMMREG[16] = { + "%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", "%xmm12", "%xmm13", "%xmm14", "%xmm15" +}; + +const char *const dis_SEGREG[16] = { + "%es", "%cs", "%ss", "%ds", "%fs", "%gs", "<reserved>", "<reserved>", + "%es", "%cs", "%ss", "%ds", "%fs", "%gs", "<reserved>", "<reserved>" +}; + +/* + * SIMD predicate suffixes + */ +const char *const dis_PREDSUFFIX[8] = { + "eq", "lt", "le", "unord", "neq", "nlt", "nle", "ord" +}; + + + +#endif /* DIS_TEXT */ + + + + +/* + * "decode table" for 64 bit mode MOVSXD instruction (opcode 0x63) + */ +const instable_t dis_opMOVSLD = TNS("movslq",MOVSXZ); + +/* + * "decode table" for pause and clflush instructions + */ +const instable_t dis_opPause = TNS("pause", NORM); + +/* + * Decode table for 0x0F00 opcodes + */ +const instable_t dis_op0F00[8] = { + +/* [0] */ TNS("sldt",M), TNS("str",M), TNSy("lldt",M), TNSy("ltr",M), +/* [4] */ TNSZ("verr",M,2), TNSZ("verw",M,2), INVALID, INVALID, +}; + + +/* + * Decode table for 0x0F01 opcodes + */ +const instable_t dis_op0F01[8] = { + +/* [0] */ TNSZ("sgdt",MO,6), TNSZ("sidt",MO,6), TNSZ("lgdt",MO,6), TNSZ("lidt",MO,6), +/* [4] */ TNSZ("smsw",M,2), INVALID, TNSZ("lmsw",M,2), TNS("invlpg",SWAPGS), +}; + +/* + * Decode table for 0x0F18 opcodes -- SIMD prefetch + */ +const instable_t dis_op0F18[8] = { + +/* [0] */ TNS("prefetchnta",PREF),TNS("prefetcht0",PREF), TNS("prefetcht1",PREF), TNS("prefetcht2",PREF), +/* [4] */ INVALID, INVALID, INVALID, INVALID, +}; + +/* + * Decode table for 0x0FAE opcodes -- SIMD state save/restore + */ +const instable_t dis_op0FAE[8] = { +/* [0] */ TNSZ("fxsave",M,512), TNSZ("fxrstor",M,512), TNS("ldmxcsr",M), TNS("stmxcsr",M), +/* [4] */ INVALID, TNS("lfence",XMMFENCE), TNS("mfence",XMMFENCE), TNS("sfence",XMMSFNC), +}; + +/* + * Decode table for 0x0FBA opcodes + */ + +const instable_t dis_op0FBA[8] = { + +/* [0] */ INVALID, INVALID, INVALID, INVALID, +/* [4] */ TS("bt",MIb), TS("bts",MIb), TS("btr",MIb), TS("btc",MIb), +}; + +/* + * Decode table for 0x0FC7 opcode + */ + +const instable_t dis_op0FC7[8] = { + +/* [0] */ INVALID, TNS("cmpxchg8b",M), INVALID, INVALID, +/* [4] */ INVALID, INVALID, INVALID, INVALID, +}; + + +/* + * Decode table for 0x0FC8 opcode -- 486 bswap instruction + * + *bit pattern: 0000 1111 1100 1reg + */ +const instable_t dis_op0FC8[4] = { +/* [0] */ TNS("bswap",R), INVALID, INVALID, INVALID, +}; + +/* + * Decode table for 0x0F71, 0x0F72, and 0x0F73 opcodes -- MMX instructions + */ +const instable_t dis_op0F7123[4][8] = { +{ +/* [70].0 */ INVALID, INVALID, INVALID, INVALID, +/* .4 */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [71].0 */ INVALID, INVALID, TNS("psrlw",MMOSH), INVALID, +/* .4 */ TNS("psraw",MMOSH), INVALID, TNS("psllw",MMOSH), INVALID, +}, { +/* [72].0 */ INVALID, INVALID, TNS("psrld",MMOSH), INVALID, +/* .4 */ TNS("psrad",MMOSH), INVALID, TNS("pslld",MMOSH), INVALID, +}, { +/* [73].0 */ INVALID, INVALID, TNS("psrlq",MMOSH), TNS("INVALID",MMOSH), +/* .4 */ INVALID, INVALID, TNS("psllq",MMOSH), TNS("INVALID",MMOSH), +} }; + +/* + * Decode table for SIMD extensions to above 0x0F71-0x0F73 opcodes. + */ +const instable_t dis_opSIMD7123[32] = { +/* [70].0 */ INVALID, INVALID, INVALID, INVALID, +/* .4 */ INVALID, INVALID, INVALID, INVALID, + +/* [71].0 */ INVALID, INVALID, TNS("psrlw",XMMSH), INVALID, +/* .4 */ TNS("psraw",XMMSH), INVALID, TNS("psllw",XMMSH), INVALID, + +/* [72].0 */ INVALID, INVALID, TNS("psrld",XMMSH), INVALID, +/* .4 */ TNS("psrad",XMMSH), INVALID, TNS("pslld",XMMSH), INVALID, + +/* [73].0 */ INVALID, INVALID, TNS("psrlq",XMMSH), TNS("psrldq",XMMSH), +/* .4 */ INVALID, INVALID, TNS("psllq",XMMSH), TNS("pslldq",XMMSH), +}; + +/* + * SIMD instructions have been wedged into the existing IA32 instruction + * set through the use of prefixes. That is, while 0xf0 0x58 may be + * addps, 0xf3 0xf0 0x58 (literally, repz addps) is a completely different + * instruction - addss. At present, three prefixes have been coopted in + * this manner - address size (0x66), repnz (0xf2) and repz (0xf3). The + * following tables are used to provide the prefixed instruction names. + * The arrays are sparse, but they're fast. + */ + +/* + * Decode table for SIMD instructions with the address size (0x66) prefix. + */ +const instable_t dis_opSIMDdata16[256] = { +/* [00] */ INVALID, INVALID, INVALID, INVALID, +/* [04] */ INVALID, INVALID, INVALID, INVALID, +/* [08] */ INVALID, INVALID, INVALID, INVALID, +/* [0C] */ INVALID, INVALID, INVALID, INVALID, + +/* [10] */ TNSZ("movupd",XMM,16), TNSZ("movupd",XMMS,16), TNSZ("movlpd",XMMM,8), TNSZ("movlpd",XMMMS,8), +/* [14] */ TNSZ("unpcklpd",XMM,16),TNSZ("unpckhpd",XMM,16),TNSZ("movhpd",XMMM,8), TNSZ("movhpd",XMMMS,8), +/* [18] */ INVALID, INVALID, INVALID, INVALID, +/* [1C] */ INVALID, INVALID, INVALID, INVALID, + +/* [20] */ INVALID, INVALID, INVALID, INVALID, +/* [24] */ INVALID, INVALID, INVALID, INVALID, +/* [28] */ TNSZ("movapd",XMM,16), TNSZ("movapd",XMMS,16), TNSZ("cvtpi2pd",XMMOMX,8),TNSZ("movntpd",XMMOMS,16), +/* [2C] */ TNSZ("cvttpd2pi",XMMXMM,16),TNSZ("cvtpd2pi",XMMXMM,16),TNSZ("ucomisd",XMM,8),TNSZ("comisd",XMM,8), + +/* [30] */ INVALID, INVALID, INVALID, INVALID, +/* [34] */ INVALID, INVALID, INVALID, INVALID, +/* [38] */ INVALID, INVALID, INVALID, INVALID, +/* [3C] */ INVALID, INVALID, INVALID, INVALID, + +/* [40] */ INVALID, INVALID, INVALID, INVALID, +/* [44] */ INVALID, INVALID, INVALID, INVALID, +/* [48] */ INVALID, INVALID, INVALID, INVALID, +/* [4C] */ INVALID, INVALID, INVALID, INVALID, + +/* [50] */ TNS("movmskpd",XMMOX3), TNSZ("sqrtpd",XMM,16), INVALID, INVALID, +/* [54] */ TNSZ("andpd",XMM,16), TNSZ("andnpd",XMM,16), TNSZ("orpd",XMM,16), TNSZ("xorpd",XMM,16), +/* [58] */ TNSZ("addpd",XMM,16), TNSZ("mulpd",XMM,16), TNSZ("cvtpd2ps",XMM,16),TNSZ("cvtps2dq",XMM,16), +/* [5C] */ TNSZ("subpd",XMM,16), TNSZ("minpd",XMM,16), TNSZ("divpd",XMM,16), TNSZ("maxpd",XMM,16), + +/* [60] */ TNSZ("punpcklbw",XMM,16),TNSZ("punpcklwd",XMM,16),TNSZ("punpckldq",XMM,16),TNSZ("packsswb",XMM,16), +/* [64] */ TNSZ("pcmpgtb",XMM,16), TNSZ("pcmpgtw",XMM,16), TNSZ("pcmpgtd",XMM,16), TNSZ("packuswb",XMM,16), +/* [68] */ TNSZ("punpckhbw",XMM,16),TNSZ("punpckhwd",XMM,16),TNSZ("punpckhdq",XMM,16),TNSZ("packssdw",XMM,16), +/* [6C] */ TNSZ("punpcklqdq",XMM,16),TNSZ("punpckhqdq",XMM,16),TNSZ("movd",XMM3MX,4),TNSZ("movdqa",XMM,16), + +/* [70] */ TNSZ("pshufd",XMMP,16), INVALID, INVALID, INVALID, +/* [74] */ TNSZ("pcmpeqb",XMM,16), TNSZ("pcmpeqw",XMM,16), TNSZ("pcmpeqd",XMM,16), INVALID, +/* [78] */ INVALID, INVALID, INVALID, INVALID, +/* [7C] */ INVALID, INVALID, TNSZ("movd",XMM3MXS,4), TNSZ("movdqa",XMMS,16), + +/* [80] */ INVALID, INVALID, INVALID, INVALID, +/* [84] */ INVALID, INVALID, INVALID, INVALID, +/* [88] */ INVALID, INVALID, INVALID, INVALID, +/* [8C] */ INVALID, INVALID, INVALID, INVALID, + +/* [90] */ INVALID, INVALID, INVALID, INVALID, +/* [94] */ INVALID, INVALID, INVALID, INVALID, +/* [98] */ INVALID, INVALID, INVALID, INVALID, +/* [9C] */ INVALID, INVALID, INVALID, INVALID, + +/* [A0] */ INVALID, INVALID, INVALID, INVALID, +/* [A4] */ INVALID, INVALID, INVALID, INVALID, +/* [A8] */ INVALID, INVALID, INVALID, INVALID, +/* [AC] */ INVALID, INVALID, INVALID, INVALID, + +/* [B0] */ INVALID, INVALID, INVALID, INVALID, +/* [B4] */ INVALID, INVALID, INVALID, INVALID, +/* [B8] */ INVALID, INVALID, INVALID, INVALID, +/* [BC] */ INVALID, INVALID, INVALID, INVALID, + +/* [C0] */ INVALID, INVALID, TNSZ("cmppd",XMMP,16), INVALID, +/* [C4] */ TNSZ("pinsrw",XMMPRM,2),TNS("pextrw",XMM3P), TNSZ("shufpd",XMMP,16), INVALID, +/* [C8] */ INVALID, INVALID, INVALID, INVALID, +/* [CC] */ INVALID, INVALID, INVALID, INVALID, + +/* [D0] */ INVALID, TNSZ("psrlw",XMM,16), TNSZ("psrld",XMM,16), TNSZ("psrlq",XMM,16), +/* [D4] */ TNSZ("paddq",XMM,16), TNSZ("pmullw",XMM,16), TNSZ("movq",XMMS,8), TNS("pmovmskb",XMMX3), +/* [D8] */ TNSZ("psubusb",XMM,16), TNSZ("psubusw",XMM,16), TNSZ("pminub",XMM,16), TNSZ("pand",XMM,16), +/* [DC] */ TNSZ("paddusb",XMM,16), TNSZ("paddusw",XMM,16), TNSZ("pmaxub",XMM,16), TNSZ("pandn",XMM,16), + +/* [E0] */ TNSZ("pavgb",XMM,16), TNSZ("psraw",XMM,16), TNSZ("psrad",XMM,16), TNSZ("pavgw",XMM,16), +/* [E4] */ TNSZ("pmulhuw",XMM,16), TNSZ("pmulhw",XMM,16), TNSZ("cvttpd2dq",XMM,16),TNSZ("movntdq",XMMS,16), +/* [E8] */ TNSZ("psubsb",XMM,16), TNSZ("psubsw",XMM,16), TNSZ("pminsw",XMM,16), TNSZ("por",XMM,16), +/* [EC] */ TNSZ("paddsb",XMM,16), TNSZ("paddsw",XMM,16), TNSZ("pmaxsw",XMM,16), TNSZ("pxor",XMM,16), + +/* [F0] */ INVALID, TNSZ("psllw",XMM,16), TNSZ("pslld",XMM,16), TNSZ("psllq",XMM,16), +/* [F4] */ TNSZ("pmuludq",XMM,16), TNSZ("pmaddwd",XMM,16), TNSZ("psadbw",XMM,16), TNSZ("maskmovdqu", XMMXIMPL,16), +/* [F8] */ TNSZ("psubb",XMM,16), TNSZ("psubw",XMM,16), TNSZ("psubd",XMM,16), TNSZ("psubq",XMM,16), +/* [FC] */ TNSZ("paddb",XMM,16), TNSZ("paddw",XMM,16), TNSZ("paddd",XMM,16), INVALID, +}; + +/* + * Decode table for SIMD instructions with the repnz (0xf2) prefix. + */ +const instable_t dis_opSIMDrepnz[256] = { +/* [00] */ INVALID, INVALID, INVALID, INVALID, +/* [04] */ INVALID, INVALID, INVALID, INVALID, +/* [08] */ INVALID, INVALID, INVALID, INVALID, +/* [0C] */ INVALID, INVALID, INVALID, INVALID, + +/* [10] */ TNSZ("movsd",XMM,8), TNSZ("movsd",XMMS,8), INVALID, INVALID, +/* [14] */ INVALID, INVALID, INVALID, INVALID, +/* [18] */ INVALID, INVALID, INVALID, INVALID, +/* [1C] */ INVALID, INVALID, INVALID, INVALID, + +/* [20] */ INVALID, INVALID, INVALID, INVALID, +/* [24] */ INVALID, INVALID, INVALID, INVALID, +/* [28] */ INVALID, INVALID, TNSZ("cvtsi2sd",XMM3MX,4),INVALID, +/* [2C] */ TNSZ("cvttsd2si",XMMXM3,8),TNSZ("cvtsd2si",XMMXM3,8),INVALID, INVALID, + +/* [30] */ INVALID, INVALID, INVALID, INVALID, +/* [34] */ INVALID, INVALID, INVALID, INVALID, +/* [38] */ INVALID, INVALID, INVALID, INVALID, +/* [3C] */ INVALID, INVALID, INVALID, INVALID, + +/* [40] */ INVALID, INVALID, INVALID, INVALID, +/* [44] */ INVALID, INVALID, INVALID, INVALID, +/* [48] */ INVALID, INVALID, INVALID, INVALID, +/* [4C] */ INVALID, INVALID, INVALID, INVALID, + +/* [50] */ INVALID, TNSZ("sqrtsd",XMM,8), INVALID, INVALID, +/* [54] */ INVALID, INVALID, INVALID, INVALID, +/* [58] */ TNSZ("addsd",XMM,8), TNSZ("mulsd",XMM,8), TNSZ("cvtsd2ss",XMM,8), INVALID, +/* [5C] */ TNSZ("subsd",XMM,8), TNSZ("minsd",XMM,8), TNSZ("divsd",XMM,8), TNSZ("maxsd",XMM,8), + +/* [60] */ INVALID, INVALID, INVALID, INVALID, +/* [64] */ INVALID, INVALID, INVALID, INVALID, +/* [68] */ INVALID, INVALID, INVALID, INVALID, +/* [6C] */ INVALID, INVALID, INVALID, INVALID, + +/* [70] */ TNSZ("pshuflw",XMMP,16),INVALID, INVALID, INVALID, +/* [74] */ INVALID, INVALID, INVALID, INVALID, +/* [78] */ INVALID, INVALID, INVALID, INVALID, +/* [7C] */ INVALID, INVALID, INVALID, INVALID, + +/* [80] */ INVALID, INVALID, INVALID, INVALID, +/* [84] */ INVALID, INVALID, INVALID, INVALID, +/* [88] */ INVALID, INVALID, INVALID, INVALID, +/* [0C] */ INVALID, INVALID, INVALID, INVALID, + +/* [90] */ INVALID, INVALID, INVALID, INVALID, +/* [94] */ INVALID, INVALID, INVALID, INVALID, +/* [98] */ INVALID, INVALID, INVALID, INVALID, +/* [9C] */ INVALID, INVALID, INVALID, INVALID, + +/* [A0] */ INVALID, INVALID, INVALID, INVALID, +/* [A4] */ INVALID, INVALID, INVALID, INVALID, +/* [A8] */ INVALID, INVALID, INVALID, INVALID, +/* [AC] */ INVALID, INVALID, INVALID, INVALID, + +/* [B0] */ INVALID, INVALID, INVALID, INVALID, +/* [B4] */ INVALID, INVALID, INVALID, INVALID, +/* [B8] */ INVALID, INVALID, INVALID, INVALID, +/* [BC] */ INVALID, INVALID, INVALID, INVALID, + +/* [C0] */ INVALID, INVALID, TNSZ("cmpsd",XMMP,8), INVALID, +/* [C4] */ INVALID, INVALID, INVALID, INVALID, +/* [C8] */ INVALID, INVALID, INVALID, INVALID, +/* [CC] */ INVALID, INVALID, INVALID, INVALID, + +/* [D0] */ INVALID, INVALID, INVALID, INVALID, +/* [D4] */ INVALID, INVALID, TNS("movdq2q",XMMXM), INVALID, +/* [D8] */ INVALID, INVALID, INVALID, INVALID, +/* [DC] */ INVALID, INVALID, INVALID, INVALID, + +/* [E0] */ INVALID, INVALID, INVALID, INVALID, +/* [E4] */ INVALID, INVALID, TNSZ("cvtpd2dq",XMM,16),INVALID, +/* [E8] */ INVALID, INVALID, INVALID, INVALID, +/* [EC] */ INVALID, INVALID, INVALID, INVALID, + +/* [F0] */ INVALID, INVALID, INVALID, INVALID, +/* [F4] */ INVALID, INVALID, INVALID, INVALID, +/* [F8] */ INVALID, INVALID, INVALID, INVALID, +/* [FC] */ INVALID, INVALID, INVALID, INVALID, +}; + +/* + * Decode table for SIMD instructions with the repz (0xf3) prefix. + */ +const instable_t dis_opSIMDrepz[256] = { +/* [00] */ INVALID, INVALID, INVALID, INVALID, +/* [04] */ INVALID, INVALID, INVALID, INVALID, +/* [08] */ INVALID, INVALID, INVALID, INVALID, +/* [0C] */ INVALID, INVALID, INVALID, INVALID, + +/* [10] */ TNSZ("movss",XMM,4), TNSZ("movss",XMMS,4), INVALID, INVALID, +/* [14] */ INVALID, INVALID, INVALID, INVALID, +/* [18] */ INVALID, INVALID, INVALID, INVALID, +/* [1C] */ INVALID, INVALID, INVALID, INVALID, + +/* [20] */ INVALID, INVALID, INVALID, INVALID, +/* [24] */ INVALID, INVALID, INVALID, INVALID, +/* [28] */ INVALID, INVALID, TNSZ("cvtsi2ss",XMM3MX,4),INVALID, +/* [2C] */ TNSZ("cvttss2si",XMMXM3,4),TNSZ("cvtss2si",XMMXM3,4),INVALID, INVALID, + +/* [30] */ INVALID, INVALID, INVALID, INVALID, +/* [34] */ INVALID, INVALID, INVALID, INVALID, +/* [38] */ INVALID, INVALID, INVALID, INVALID, +/* [3C] */ INVALID, INVALID, INVALID, INVALID, + +/* [40] */ INVALID, INVALID, INVALID, INVALID, +/* [44] */ INVALID, INVALID, INVALID, INVALID, +/* [48] */ INVALID, INVALID, INVALID, INVALID, +/* [4C] */ INVALID, INVALID, INVALID, INVALID, + +/* [50] */ INVALID, TNSZ("sqrtss",XMM,4), TNSZ("rsqrtss",XMM,4), TNSZ("rcpss",XMM,4), +/* [54] */ INVALID, INVALID, INVALID, INVALID, +/* [58] */ TNSZ("addss",XMM,4), TNSZ("mulss",XMM,4), TNSZ("cvtss2sd",XMM,4), TNSZ("cvttps2dq",XMM,16), +/* [5C] */ TNSZ("subss",XMM,4), TNSZ("minss",XMM,4), TNSZ("divss",XMM,4), TNSZ("maxss",XMM,4), + +/* [60] */ INVALID, INVALID, INVALID, INVALID, +/* [64] */ INVALID, INVALID, INVALID, INVALID, +/* [68] */ INVALID, INVALID, INVALID, INVALID, +/* [6C] */ INVALID, INVALID, INVALID, TNSZ("movdqu",XMM,16), + +/* [70] */ TNSZ("pshufhw",XMMP,16),INVALID, INVALID, INVALID, +/* [74] */ INVALID, INVALID, INVALID, INVALID, +/* [78] */ INVALID, INVALID, INVALID, INVALID, +/* [7C] */ INVALID, INVALID, TNSZ("movq",XMM,8), TNSZ("movdqu",XMMS,16), + +/* [80] */ INVALID, INVALID, INVALID, INVALID, +/* [84] */ INVALID, INVALID, INVALID, INVALID, +/* [88] */ INVALID, INVALID, INVALID, INVALID, +/* [0C] */ INVALID, INVALID, INVALID, INVALID, + +/* [90] */ INVALID, INVALID, INVALID, INVALID, +/* [94] */ INVALID, INVALID, INVALID, INVALID, +/* [98] */ INVALID, INVALID, INVALID, INVALID, +/* [9C] */ INVALID, INVALID, INVALID, INVALID, + +/* [A0] */ INVALID, INVALID, INVALID, INVALID, +/* [A4] */ INVALID, INVALID, INVALID, INVALID, +/* [A8] */ INVALID, INVALID, INVALID, INVALID, +/* [AC] */ INVALID, INVALID, INVALID, INVALID, + +/* [B0] */ INVALID, INVALID, INVALID, INVALID, +/* [B4] */ INVALID, INVALID, INVALID, INVALID, +/* [B8] */ INVALID, INVALID, INVALID, INVALID, +/* [BC] */ INVALID, INVALID, INVALID, INVALID, + +/* [C0] */ INVALID, INVALID, TNSZ("cmpss",XMMP,4), INVALID, +/* [C4] */ INVALID, INVALID, INVALID, INVALID, +/* [C8] */ INVALID, INVALID, INVALID, INVALID, +/* [CC] */ INVALID, INVALID, INVALID, INVALID, + +/* [D0] */ INVALID, INVALID, INVALID, INVALID, +/* [D4] */ INVALID, INVALID, TNS("movq2dq",XMMMX), INVALID, +/* [D8] */ INVALID, INVALID, INVALID, INVALID, +/* [DC] */ INVALID, INVALID, INVALID, INVALID, + +/* [E0] */ INVALID, INVALID, INVALID, INVALID, +/* [E4] */ INVALID, INVALID, TNSZ("cvtdq2pd",XMM,8), INVALID, +/* [E8] */ INVALID, INVALID, INVALID, INVALID, +/* [EC] */ INVALID, INVALID, INVALID, INVALID, + +/* [F0] */ INVALID, INVALID, INVALID, INVALID, +/* [F4] */ INVALID, INVALID, INVALID, INVALID, +/* [F8] */ INVALID, INVALID, INVALID, INVALID, +/* [FC] */ INVALID, INVALID, INVALID, INVALID, +}; + +/* + * Decode table for 0x0F opcodes + */ + +const instable_t dis_op0F[16][16] = { +{ +/* [00] */ IND(dis_op0F00), IND(dis_op0F01), TNS("lar",MR), TNS("lsl",MR), +/* [04] */ INVALID, TNS("syscall",NORM), TNS("clts",NORM), TNS("sysret",NORM), +/* [08] */ TNS("invd",NORM), TNS("wbinvd",NORM), INVALID, TNS("ud2",NORM), +/* [0C] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [10] */ TNSZ("movups",XMMO,16), TNSZ("movups",XMMOS,16),TNSZ("movlps",XMMO,8), TNSZ("movlps",XMMOS,8), +/* [14] */ TNSZ("unpcklps",XMMO,16),TNSZ("unpckhps",XMMO,16),TNSZ("movhps",XMMOM,8),TNSZ("movhps",XMMOMS,8), +/* [18] */ IND(dis_op0F18), INVALID, INVALID, INVALID, +/* [1C] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [20] */ TSy("mov",SREG), TSy("mov",SREG), TSy("mov",SREG), TSy("mov",SREG), +/* [24] */ TSx("mov",SREG), INVALID, TSx("mov",SREG), INVALID, +/* [28] */ TNSZ("movaps",XMMO,16), TNSZ("movaps",XMMOS,16),TNSZ("cvtpi2ps",XMMOMX,8),TNSZ("movntps",XMMOS,16), +/* [2C] */ TNSZ("cvttps2pi",XMMOXMM,8),TNSZ("cvtps2pi",XMMOXMM,8),TNSZ("ucomiss",XMMO,4),TNSZ("comiss",XMMO,4), +}, { +/* [30] */ TNS("wrmsr",NORM), TNS("rdtsc",NORM), TNS("rdmsr",NORM), TNS("rdpmc",NORM), +/* [34] */ TNSx("sysenter",NORM), TNSx("sysexit",NORM), INVALID, INVALID, +/* [38] */ INVALID, INVALID, INVALID, INVALID, +/* [3C] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [40] */ TS("cmovx.o",MR), TS("cmovx.no",MR), TS("cmovx.b",MR), TS("cmovx.ae",MR), +/* [44] */ TS("cmovx.e",MR), TS("cmovx.ne",MR), TS("cmovx.be",MR), TS("cmovx.a",MR), +/* [48] */ TS("cmovx.s",MR), TS("cmovx.ns",MR), TS("cmovx.pe",MR), TS("cmovx.po",MR), +/* [4C] */ TS("cmovx.l",MR), TS("cmovx.ge",MR), TS("cmovx.le",MR), TS("cmovx.g",MR), +}, { +/* [50] */ TNS("movmskps",XMMOX3), TNSZ("sqrtps",XMMO,16), TNSZ("rsqrtps",XMMO,16),TNSZ("rcpps",XMMO,16), +/* [54] */ TNSZ("andps",XMMO,16), TNSZ("andnps",XMMO,16), TNSZ("orps",XMMO,16), TNSZ("xorps",XMMO,16), +/* [58] */ TNSZ("addps",XMMO,16), TNSZ("mulps",XMMO,16), TNSZ("cvtps2pd",XMMO,8),TNSZ("cvtdq2ps",XMMO,16), +/* [5C] */ TNSZ("subps",XMMO,16), TNSZ("minps",XMMO,16), TNSZ("divps",XMMO,16), TNSZ("maxps",XMMO,16), +}, { +/* [60] */ TNSZ("punpcklbw",MMO,4),TNSZ("punpcklwd",MMO,4),TNSZ("punpckldq",MMO,4),TNSZ("packsswb",MMO,8), +/* [64] */ TNSZ("pcmpgtb",MMO,8), TNSZ("pcmpgtw",MMO,8), TNSZ("pcmpgtd",MMO,8), TNSZ("packuswb",MMO,8), +/* [68] */ TNSZ("punpckhbw",MMO,8),TNSZ("punpckhwd",MMO,8),TNSZ("punpckhdq",MMO,8),TNSZ("packssdw",MMO,8), +/* [6C] */ TNSZ("INVALID",MMO,0), TNSZ("INVALID",MMO,0), TNSZ("movd",MMO,4), TNSZ("movq",MMO,8), +}, { +/* [70] */ TNSZ("pshufw",MMOPM,8), TNS("psrXXX",MR), TNS("psrXXX",MR), TNS("psrXXX",MR), +/* [74] */ TNSZ("pcmpeqb",MMO,8), TNSZ("pcmpeqw",MMO,8), TNSZ("pcmpeqd",MMO,8), TNS("emms",NORM), +/* [78] */ INVALID, INVALID, INVALID, INVALID, +/* [7C] */ INVALID, INVALID, TNSZ("movd",MMOS,4), TNSZ("movq",MMOS,8), +}, { +/* [80] */ TNS("jo",D), TNS("jno",D), TNS("jb",D), TNS("jae",D), +/* [84] */ TNS("je",D), TNS("jne",D), TNS("jbe",D), TNS("ja",D), +/* [88] */ TNS("js",D), TNS("jns",D), TNS("jp",D), TNS("jnp",D), +/* [8C] */ TNS("jl",D), TNS("jge",D), TNS("jle",D), TNS("jg",D), +}, { +/* [90] */ TNS("seto",Mb), TNS("setno",Mb), TNS("setb",Mb), TNS("setae",Mb), +/* [94] */ TNS("sete",Mb), TNS("setne",Mb), TNS("setbe",Mb), TNS("seta",Mb), +/* [98] */ TNS("sets",Mb), TNS("setns",Mb), TNS("setp",Mb), TNS("setnp",Mb), +/* [9C] */ TNS("setl",Mb), TNS("setge",Mb), TNS("setle",Mb), TNS("setg",Mb), +}, { +/* [A0] */ TSp("push",LSEG), TSp("pop",LSEG), TNS("cpuid",NORM), TS("bt",RMw), +/* [A4] */ TS("shld",DSHIFT), TS("shld",DSHIFTcl), INVALID, INVALID, +/* [A8] */ TSp("push",LSEG), TSp("pop",LSEG), TNS("rsm",NORM), TS("bts",RMw), +/* [AC] */ TS("shrd",DSHIFT), TS("shrd",DSHIFTcl), IND(dis_op0FAE), TS("imul",MRw), +}, { +/* [B0] */ TNS("cmpxchgb",RMw), TS("cmpxchg",RMw), TS("lss",MR), TS("btr",RMw), +/* [B4] */ TS("lfs",MR), TS("lgs",MR), TS("movzb",MOVZ), TNS("movzwl",MOVZ), +/* [B8] */ INVALID, INVALID, IND(dis_op0FBA), TS("btc",RMw), +/* [BC] */ TS("bsf",MRw), TS("bsr",MRw), TS("movsb",MOVZ), TNS("movswl",MOVZ), +}, { +/* [C0] */ TNS("xaddb",XADDB), TS("xadd",RMw), TNSZ("cmpps",XMMOPM,16),TNS("movnti",RM), +/* [C4] */ TNSZ("pinsrw",MMOPRM,2),TNS("pextrw",MMO3P), TNSZ("shufps",XMMOPM,16),IND(dis_op0FC7), +/* [C8] */ INVALID, INVALID, INVALID, INVALID, +/* [CC] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [D0] */ INVALID, TNSZ("psrlw",MMO,8), TNSZ("psrld",MMO,8), TNSZ("psrlq",MMO,8), +/* [D4] */ TNSZ("paddq",MMO,8), TNSZ("pmullw",MMO,8), TNSZ("INVALID",MMO,0), TNS("pmovmskb",MMOM3), +/* [D8] */ TNSZ("psubusb",MMO,8), TNSZ("psubusw",MMO,8), TNSZ("pminub",MMO,8), TNSZ("pand",MMO,8), +/* [DC] */ TNSZ("paddusb",MMO,8), TNSZ("paddusw",MMO,8), TNSZ("pmaxub",MMO,8), TNSZ("pandn",MMO,8), +}, { +/* [E0] */ TNSZ("pavgb",MMO,8), TNSZ("psraw",MMO,8), TNSZ("psrad",MMO,8), TNSZ("pavgw",MMO,8), +/* [E4] */ TNSZ("pmulhuw",MMO,8), TNSZ("pmulhw",MMO,8), TNS("INVALID",XMMO), TNSZ("movntq",MMOMS,8), +/* [E8] */ TNSZ("psubsb",MMO,8), TNSZ("psubsw",MMO,8), TNSZ("pminsw",MMO,8), TNSZ("por",MMO,8), +/* [EC] */ TNSZ("paddsb",MMO,8), TNSZ("paddsw",MMO,8), TNSZ("pmaxsw",MMO,8), TNSZ("pxor",MMO,8), +}, { +/* [F0] */ INVALID, TNSZ("psllw",MMO,8), TNSZ("pslld",MMO,8), TNSZ("psllq",MMO,8), +/* [F4] */ TNSZ("pmuludq",MMO,8), TNSZ("pmaddwd",MMO,8), TNSZ("psadbw",MMO,8), TNSZ("maskmovq",MMOIMPL,8), +/* [F8] */ TNSZ("psubb",MMO,8), TNSZ("psubw",MMO,8), TNSZ("psubd",MMO,8), TNSZ("psubq",MMO,8), +/* [FC] */ TNSZ("paddb",MMO,8), TNSZ("paddw",MMO,8), TNSZ("paddd",MMO,8), INVALID, +} }; + + +/* + * Decode table for 0x80 opcodes + */ + +const instable_t dis_op80[8] = { + +/* [0] */ TNS("addb",IMlw), TNS("orb",IMw), TNS("adcb",IMlw), TNS("sbbb",IMlw), +/* [4] */ TNS("andb",IMw), TNS("subb",IMlw), TNS("xorb",IMw), TNS("cmpb",IMlw), +}; + + +/* + * Decode table for 0x81 opcodes. + */ + +const instable_t dis_op81[8] = { + +/* [0] */ TS("add",IMlw), TS("or",IMw), TS("adc",IMlw), TS("sbb",IMlw), +/* [4] */ TS("and",IMw), TS("sub",IMlw), TS("xor",IMw), TS("cmp",IMlw), +}; + + +/* + * Decode table for 0x82 opcodes. + */ + +const instable_t dis_op82[8] = { + +/* [0] */ TNSx("addb",IMlw), TNSx("orb",IMlw), TNSx("adcb",IMlw), TNSx("sbbb",IMlw), +/* [4] */ TNSx("andb",IMlw), TNSx("subb",IMlw), TNSx("xorb",IMlw), TNSx("cmpb",IMlw), +}; +/* + * Decode table for 0x83 opcodes. + */ + +const instable_t dis_op83[8] = { + +/* [0] */ TS("add",IMlw), TS("or",IMlw), TS("adc",IMlw), TS("sbb",IMlw), +/* [4] */ TS("and",IMlw), TS("sub",IMlw), TS("xor",IMlw), TS("cmp",IMlw), +}; + +/* + * Decode table for 0xC0 opcodes. + */ + +const instable_t dis_opC0[8] = { + +/* [0] */ TNS("rolb",MvI), TNS("rorb",MvI), TNS("rclb",MvI), TNS("rcrb",MvI), +/* [4] */ TNS("shlb",MvI), TNS("shrb",MvI), INVALID, TNS("sarb",MvI), +}; + +/* + * Decode table for 0xD0 opcodes. + */ + +const instable_t dis_opD0[8] = { + +/* [0] */ TNS("rolb",Mv), TNS("rorb",Mv), TNS("rclb",Mv), TNS("rcrb",Mv), +/* [4] */ TNS("shlb",Mv), TNS("shrb",Mv), TNS("salb",Mv), TNS("sarb",Mv), +}; + +/* + * Decode table for 0xC1 opcodes. + * 186 instruction set + */ + +const instable_t dis_opC1[8] = { + +/* [0] */ TS("rol",MvI), TS("ror",MvI), TS("rcl",MvI), TS("rcr",MvI), +/* [4] */ TS("shl",MvI), TS("shr",MvI), TS("sal",MvI), TS("sar",MvI), +}; + +/* + * Decode table for 0xD1 opcodes. + */ + +const instable_t dis_opD1[8] = { + +/* [0] */ TS("rol",Mv), TS("ror",Mv), TS("rcl",Mv), TS("rcr",Mv), +/* [4] */ TS("shl",Mv), TS("shr",Mv), TS("sal",Mv), TS("sar",Mv), +}; + + +/* + * Decode table for 0xD2 opcodes. + */ + +const instable_t dis_opD2[8] = { + +/* [0] */ TNS("rolb",Mv), TNS("rorb",Mv), TNS("rclb",Mv), TNS("rcrb",Mv), +/* [4] */ TNS("shlb",Mv), TNS("shrb",Mv), TNS("salb",Mv), TNS("sarb",Mv), +}; +/* + * Decode table for 0xD3 opcodes. + */ + +const instable_t dis_opD3[8] = { + +/* [0] */ TS("rol",Mv), TS("ror",Mv), TS("rcl",Mv), TS("rcr",Mv), +/* [4] */ TS("shl",Mv), TS("shr",Mv), TS("salb",Mv), TS("sar",Mv), +}; + + +/* + * Decode table for 0xF6 opcodes. + */ + +const instable_t dis_opF6[8] = { + +/* [0] */ TNS("testb",IMw), TNS("testb",IMw), TNS("notb",Mw), TNS("negb",Mw), +/* [4] */ TNS("mulb",MA), TNS("imulb",MA), TNS("divb",MA), TNS("idivb",MA), +}; + + +/* + * Decode table for 0xF7 opcodes. + */ + +const instable_t dis_opF7[8] = { + +/* [0] */ TS("test",IMw), TS("test",IMw), TS("not",Mw), TS("neg",Mw), +/* [4] */ TS("mul",MA), TS("imul",MA), TS("div",MA), TS("idiv",MA), +}; + + +/* + * Decode table for 0xFE opcodes. + */ + +const instable_t dis_opFE[8] = { + +/* [0] */ TNS("incb",Mw), TNS("decb",Mw), INVALID, INVALID, +/* [4] */ INVALID, INVALID, INVALID, INVALID, +}; +/* + * Decode table for 0xFF opcodes. + */ + +const instable_t dis_opFF[8] = { + +/* [0] */ TS("inc",Mw), TS("dec",Mw), TNSyp("call",INM), TNS("lcall",INM), +/* [4] */ TNSy("jmp",INM), TNS("ljmp",INM), TSp("push",M), INVALID, +}; + +/* for 287 instructions, which are a mess to decode */ + +const instable_t dis_opFP1n2[8][8] = { +{ +/* bit pattern: 1101 1xxx MODxx xR/M */ +/* [0,0] */ TNS("fadds",M), TNS("fmuls",M), TNS("fcoms",M), TNS("fcomps",M), +/* [0,4] */ TNS("fsubs",M), TNS("fsubrs",M), TNS("fdivs",M), TNS("fdivrs",M), +}, { +/* [1,0] */ TNS("flds",M), INVALID, TNS("fsts",M), TNS("fstps",M), +/* [1,4] */ TNSZ("fldenv",M,28), TNSZ("fldcw",M,2), TNSZ("fnstenv",M,28), TNSZ("fnstcw",M,2), +}, { +/* [2,0] */ TNS("fiaddl",M), TNS("fimull",M), TNS("ficoml",M), TNS("ficompl",M), +/* [2,4] */ TNS("fisubl",M), TNS("fisubrl",M), TNS("fidivl",M), TNS("fidivrl",M), +}, { +/* [3,0] */ TNS("fildl",M), INVALID, TNS("fistl",M), TNS("fistpl",M), +/* [3,4] */ INVALID, TNSZ("fldt",M,10), INVALID, TNSZ("fstpt",M,10), +}, { +/* [4,0] */ TNSZ("faddl",M,8), TNSZ("fmull",M,8), TNSZ("fcoml",M,8), TNSZ("fcompl",M,8), +/* [4,1] */ TNSZ("fsubl",M,8), TNSZ("fsubrl",M,8), TNSZ("fdivl",M,8), TNSZ("fdivrl",M,8), +}, { +/* [5,0] */ TNSZ("fldl",M,8), INVALID, TNSZ("fstl",M,8), TNSZ("fstpl",M,8), +/* [5,4] */ TNSZ("frstor",M,108), INVALID, TNSZ("fnsave",M,108), TNSZ("fnstsw",M,2), +}, { +/* [6,0] */ TNSZ("fiadd",M,2), TNSZ("fimul",M,2), TNSZ("ficom",M,2), TNSZ("ficomp",M,2), +/* [6,4] */ TNSZ("fisub",M,2), TNSZ("fisubr",M,2), TNSZ("fidiv",M,2), TNSZ("fidivr",M,2), +}, { +/* [7,0] */ TNSZ("fild",M,2), INVALID, TNSZ("fist",M,2), TNSZ("fistp",M,2), +/* [7,4] */ TNSZ("fbld",M,10), TNSZ("fildll",M,8), TNSZ("fbstp",M,10), TNSZ("fistpll",M,8), +} }; + +const instable_t dis_opFP3[8][8] = { +{ +/* bit pattern: 1101 1xxx 11xx xREG */ +/* [0,0] */ TNS("fadd",FF), TNS("fmul",FF), TNS("fcom",F), TNS("fcomp",F), +/* [0,4] */ TNS("fsub",FF), TNS("fsubr",FF), TNS("fdiv",FF), TNS("fdivr",FF), +}, { +/* [1,0] */ TNS("fld",F), TNS("fxch",F), TNS("fnop",NORM), TNS("fstp",F), +/* [1,4] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [2,0] */ INVALID, INVALID, INVALID, INVALID, +/* [2,4] */ INVALID, TNS("fucompp",NORM), INVALID, INVALID, +}, { +/* [3,0] */ INVALID, INVALID, INVALID, INVALID, +/* [3,4] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [4,0] */ TNS("fadd",FF), TNS("fmul",FF), TNS("fcom",F), TNS("fcomp",F), +/* [4,4] */ TNS("fsub",FF), TNS("fsubr",FF), TNS("fdiv",FF), TNS("fdivr",FF), +}, { +/* [5,0] */ TNS("ffree",F), TNS("fxch",F), TNS("fst",F), TNS("fstp",F), +/* [5,4] */ TNS("fucom",F), TNS("fucomp",F), INVALID, INVALID, +}, { +/* [6,0] */ TNS("faddp",FF), TNS("fmulp",FF), TNS("fcomp",F), TNS("fcompp",NORM), +/* [6,4] */ TNS("fsubp",FF), TNS("fsubrp",FF), TNS("fdivp",FF), TNS("fdivrp",FF), +}, { +/* [7,0] */ TNS("ffree",F), TNS("fxch",F), TNS("fstp",F), TNS("fstp",F), +/* [7,4] */ TNS("fnstsw",M), TNS("fucomip",FFC), TNS("fcomip",FFC), INVALID, +} }; + +const instable_t dis_opFP4[4][8] = { +{ +/* bit pattern: 1101 1001 111x xxxx */ +/* [0,0] */ TNS("fchs",NORM), TNS("fabs",NORM), INVALID, INVALID, +/* [0,4] */ TNS("ftst",NORM), TNS("fxam",NORM), TNS("ftstp",NORM), INVALID, +}, { +/* [1,0] */ TNS("fld1",NORM), TNS("fldl2t",NORM), TNS("fldl2e",NORM), TNS("fldpi",NORM), +/* [1,4] */ TNS("fldlg2",NORM), TNS("fldln2",NORM), TNS("fldz",NORM), INVALID, +}, { +/* [2,0] */ TNS("f2xm1",NORM), TNS("fyl2x",NORM), TNS("fptan",NORM), TNS("fpatan",NORM), +/* [2,4] */ TNS("fxtract",NORM), TNS("fprem1",NORM), TNS("fdecstp",NORM), TNS("fincstp",NORM), +}, { +/* [3,0] */ TNS("fprem",NORM), TNS("fyl2xp1",NORM), TNS("fsqrt",NORM), TNS("fsincos",NORM), +/* [3,4] */ TNS("frndint",NORM), TNS("fscale",NORM), TNS("fsin",NORM), TNS("fcos",NORM), +} }; + +const instable_t dis_opFP5[8] = { +/* bit pattern: 1101 1011 111x xxxx */ +/* [0] */ TNS("feni",NORM), TNS("fdisi",NORM), TNS("fnclex",NORM), TNS("fninit",NORM), +/* [4] */ TNS("fsetpm",NORM), TNS("frstpm",NORM), INVALID, INVALID, +}; + +const instable_t dis_opFP6[8] = { +/* bit pattern: 1101 1011 11yy yxxx */ +/* [00] */ TNS("fcmov.nb",FF), TNS("fcmov.ne",FF), TNS("fcmov.nbe",FF), TNS("fcmov.nu",FF), +/* [04] */ INVALID, TNS("fucomi",F), TNS("fcomi",F), INVALID, +}; + +const instable_t dis_opFP7[8] = { +/* bit pattern: 1101 1010 11yy yxxx */ +/* [00] */ TNS("fcmov.b",FF), TNS("fcmov.e",FF), TNS("fcmov.be",FF), TNS("fcmov.u",FF), +/* [04] */ INVALID, INVALID, INVALID, INVALID, +}; + +/* + * Main decode table for the op codes. The first two nibbles + * will be used as an index into the table. If there is a + * a need to further decode an instruction, the array to be + * referenced is indicated with the other two entries being + * empty. + */ + +const instable_t dis_distable[16][16] = { +{ +/* [0,0] */ TNS("addb",RMw), TS("add",RMw), TNS("addb",MRw), TS("add",MRw), +/* [0,4] */ TNS("addb",IA), TS("add",IA), TSx("push",SEG), TSx("pop",SEG), +/* [0,8] */ TNS("orb",RMw), TS("or",RMw), TNS("orb",MRw), TS("or",MRw), +/* [0,C] */ TNS("orb",IA), TS("or",IA), TSx("push",SEG), IND(&dis_op0F[0][0]), +}, { +/* [1,0] */ TNS("adcb",RMw), TS("adc",RMw), TNS("adcb",MRw), TS("adc",MRw), +/* [1,4] */ TNS("adcb",IA), TS("adc",IA), TSx("push",SEG), TSx("pop",SEG), +/* [1,8] */ TNS("sbbb",RMw), TS("sbb",RMw), TNS("sbbb",MRw), TS("sbb",MRw), +/* [1,C] */ TNS("sbbb",IA), TS("sbb",IA), TSx("push",SEG), TSx("pop",SEG), +}, { +/* [2,0] */ TNS("andb",RMw), TS("and",RMw), TNS("andb",MRw), TS("and",MRw), +/* [2,4] */ TNS("andb",IA), TS("and",IA), TNSx("%es:",OVERRIDE), TNSx("daa",NORM), +/* [2,8] */ TNS("subb",RMw), TS("sub",RMw), TNS("subb",MRw), TS("sub",MRw), +/* [2,C] */ TNS("subb",IA), TS("sub",IA), TNSx("%cs:",OVERRIDE), TNSx("das",NORM), +}, { +/* [3,0] */ TNS("xorb",RMw), TS("xor",RMw), TNS("xorb",MRw), TS("xor",MRw), +/* [3,4] */ TNS("xorb",IA), TS("xor",IA), TNSx("%ss:",OVERRIDE), TNSx("aaa",NORM), +/* [3,8] */ TNS("cmpb",RMw), TS("cmp",RMw), TNS("cmpb",MRw), TS("cmp",MRw), +/* [3,C] */ TNS("cmpb",IA), TS("cmp",IA), TNSx("%ds:",OVERRIDE), TNSx("aas",NORM), +}, { +/* [4,0] */ TSx("inc",R), TSx("inc",R), TSx("inc",R), TSx("inc",R), +/* [4,4] */ TSx("inc",R), TSx("inc",R), TSx("inc",R), TSx("inc",R), +/* [4,8] */ TSx("dec",R), TSx("dec",R), TSx("dec",R), TSx("dec",R), +/* [4,C] */ TSx("dec",R), TSx("dec",R), TSx("dec",R), TSx("dec",R), +}, { +/* [5,0] */ TSp("push",R), TSp("push",R), TSp("push",R), TSp("push",R), +/* [5,4] */ TSp("push",R), TSp("push",R), TSp("push",R), TSp("push",R), +/* [5,8] */ TSp("pop",R), TSp("pop",R), TSp("pop",R), TSp("pop",R), +/* [5,C] */ TSp("pop",R), TSp("pop",R), TSp("pop",R), TSp("pop",R), +}, { +/* [6,0] */ TSZx("pusha",IMPLMEM,28),TSZx("popa",IMPLMEM,28), TSx("bound",MR), TNS("arpl",RMw), +/* [6,4] */ TNS("%fs:",OVERRIDE), TNS("%gs:",OVERRIDE), TNS("data16",DM), TNS("addr16",AM), +/* [6,8] */ TSp("push",I), TS("imul",IMUL), TSp("push",Ib), TS("imul",IMUL), +/* [6,C] */ TNSZ("insb",IMPLMEM,1), TSZ("ins",IMPLMEM,4), TNSZ("outsb",IMPLMEM,1),TSZ("outs",IMPLMEM,4), +}, { +/* [7,0] */ TNSy("jo",BD), TNSy("jno",BD), TNSy("jb",BD), TNSy("jae",BD), +/* [7,4] */ TNSy("je",BD), TNSy("jne",BD), TNSy("jbe",BD), TNSy("ja",BD), +/* [7,8] */ TNSy("js",BD), TNSy("jns",BD), TNSy("jp",BD), TNSy("jnp",BD), +/* [7,C] */ TNSy("jl",BD), TNSy("jge",BD), TNSy("jle",BD), TNSy("jg",BD), +}, { +/* [8,0] */ IND(dis_op80), IND(dis_op81), INDx(dis_op82), IND(dis_op83), +/* [8,4] */ TNS("testb",RMw), TS("test",RMw), TNS("xchgb",RMw), TS("xchg",RMw), +/* [8,8] */ TNS("movb",RMw), TS("mov",RMw), TNS("movb",MRw), TS("mov",MRw), +/* [8,C] */ TNS("movw",SM), TS("lea",MR), TNS("movw",MS), TSp("pop",M), +}, { +/* [9,0] */ TNS("nop",NORM), TS("xchg",RA), TS("xchg",RA), TS("xchg",RA), +/* [9,4] */ TS("xchg",RA), TS("xchg",RA), TS("xchg",RA), TS("xchg",RA), +/* [9,8] */ TNS("cXtX",CBW), TNS("cXtX",CWD), TNSx("lcall",SO), TNS("fwait",NORM), +/* [9,C] */ TSZy("pushf",IMPLMEM,4),TSZy("popf",IMPLMEM,4), TNSx("sahf",NORM), TNSx("lahf",NORM), +}, { +/* [A,0] */ TNS("movb",OA), TS("mov",OA), TNS("movb",AO), TS("mov",AO), +/* [A,4] */ TNSZ("movsb",SD,1), TS("movs",SD), TNSZ("cmpsb",SD,1), TS("cmps",SD), +/* [A,8] */ TNS("testb",IA), TS("test",IA), TNS("stosb",AD), TS("stos",AD), +/* [A,C] */ TNS("lodsb",SA), TS("lods",SA), TNS("scasb",AD), TS("scas",AD), +}, { +/* [B,0] */ TNS("movb",IR), TNS("movb",IR), TNS("movb",IR), TNS("movb",IR), +/* [B,4] */ TNS("movb",IR), TNS("movb",IR), TNS("movb",IR), TNS("movb",IR), +/* [B,8] */ TS("mov",IR), TS("mov",IR), TS("mov",IR), TS("mov",IR), +/* [B,C] */ TS("mov",IR), TS("mov",IR), TS("mov",IR), TS("mov",IR), +}, { +/* [C,0] */ IND(dis_opC0), IND(dis_opC1), TNSyp("ret",RET), TNSyp("ret",NORM), +/* [C,4] */ TNSx("les",MR), TNSx("lds",MR), TNS("movb",IMw), TS("mov",IMw), +/* [C,8] */ TNSyp("enter",ENTER), TNSyp("leave",NORM), TNS("lret",RET), TNS("lret",NORM), +/* [C,C] */ TNS("int",INT3), TNS("int",INTx), TNSx("into",NORM), TNS("iret",NORM), +}, { +/* [D,0] */ IND(dis_opD0), IND(dis_opD1), IND(dis_opD2), IND(dis_opD3), +/* [D,4] */ TNSx("aam",U), TNSx("aad",U), TNSx("falc",NORM), TNSZ("xlat",IMPLMEM,1), + +/* 287 instructions. Note that although the indirect field */ +/* indicates opFP1n2 for further decoding, this is not necessarily */ +/* the case since the opFP arrays are not partitioned according to key1 */ +/* and key2. opFP1n2 is given only to indicate that we haven't */ +/* finished decoding the instruction. */ +/* [D,8] */ IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), +/* [D,C] */ IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), +}, { +/* [E,0] */ TNSy("loopnz",BD), TNSy("loopz",BD), TNSy("loop",BD), TNSy("jcxz",BD), +/* [E,4] */ TNS("inb",P), TS("in",P), TNS("outb",P), TS("out",P), +/* [E,8] */ TNSyp("call",D), TNSy("jmp",D), TNSx("ljmp",SO), TNSy("jmp",BD), +/* [E,C] */ TNS("inb",V), TS("in",V), TNS("outb",V), TS("out",V), +}, { +/* [F,0] */ TNS("lock",LOCK), TNS("icebp", NORM), TNS("repnz",PREFIX), TNS("repz",PREFIX), +/* [F,4] */ TNS("hlt",NORM), TNS("cmc",NORM), IND(dis_opF6), IND(dis_opF7), +/* [F,8] */ TNS("clc",NORM), TNS("stc",NORM), TNS("cli",NORM), TNS("sti",NORM), +/* [F,C] */ TNS("cld",NORM), TNS("std",NORM), IND(dis_opFE), IND(dis_opFF), +} }; + +/* END CSTYLED */ + +/* + * common functions to decode and disassemble an x86 or amd64 instruction + */ + +/* + * These are the individual fields of a REX prefix. Note that a REX + * prefix with none of these set is still needed to: + * - use the MOVSXD (sign extend 32 to 64 bits) instruction + * - access the %sil, %dil, %bpl, %spl registers + */ +#define REX_W 0x08 /* 64 bit operand size when set */ +#define REX_R 0x04 /* high order bit extension of ModRM reg field */ +#define REX_X 0x02 /* high order bit extension of SIB index field */ +#define REX_B 0x01 /* extends ModRM r_m, SIB base, or opcode reg */ + +static uint_t opnd_size; /* SIZE16, SIZE32 or SIZE64 */ +static uint_t addr_size; /* SIZE16, SIZE32 or SIZE64 */ + +/* + * Even in 64 bit mode, usually only 4 byte immediate operands are supported. + */ +static int isize[] = {1, 2, 4, 4}; +static int isize64[] = {1, 2, 4, 8}; + +/* + * Just a bunch of useful macros. + */ +#define WBIT(x) (x & 0x1) /* to get w bit */ +#define REGNO(x) (x & 0x7) /* to get 3 bit register */ +#define VBIT(x) ((x)>>1 & 0x1) /* to get 'v' bit */ +#define OPSIZE(osize, wbit) ((wbit) ? isize[osize] : 1) +#define OPSIZE64(osize, wbit) ((wbit) ? isize64[osize] : 1) + +#define REG_ONLY 3 /* mode to indicate a register operand (not memory) */ + +#define BYTE_OPND 0 /* w-bit value indicating byte register */ +#define LONG_OPND 1 /* w-bit value indicating opnd_size register */ +#define MM_OPND 2 /* "value" used to indicate a mmx reg */ +#define XMM_OPND 3 /* "value" used to indicate a xmm reg */ +#define SEG_OPND 4 /* "value" used to indicate a segment reg */ +#define CONTROL_OPND 5 /* "value" used to indicate a control reg */ +#define DEBUG_OPND 6 /* "value" used to indicate a debug reg */ +#define TEST_OPND 7 /* "value" used to indicate a test reg */ +#define WORD_OPND 8 /* w-bit value indicating word size reg */ + +/* + * Get the next byte and separate the op code into the high and low nibbles. + */ +static int +dtrace_get_opcode(dis86_t *x, uint_t *high, uint_t *low) +{ + int byte; + + /* + * x86 instructions have a maximum length of 15 bytes. Bail out if + * we try to read more. + */ + if (x->d86_len >= 15) + return (x->d86_error = 1); + + if (x->d86_error) + return (1); + byte = x->d86_get_byte(x->d86_data); + if (byte < 0) + return (x->d86_error = 1); + x->d86_bytes[x->d86_len++] = byte; + *low = byte & 0xf; /* ----xxxx low 4 bits */ + *high = byte >> 4 & 0xf; /* xxxx---- bits 7 to 4 */ + return (0); +} + +/* + * Get and decode an SIB (scaled index base) byte + */ +static void +dtrace_get_SIB(dis86_t *x, uint_t *ss, uint_t *index, uint_t *base) +{ + int byte; + + if (x->d86_error) + return; + + byte = x->d86_get_byte(x->d86_data); + if (byte < 0) { + x->d86_error = 1; + return; + } + x->d86_bytes[x->d86_len++] = byte; + + *base = byte & 0x7; + *index = (byte >> 3) & 0x7; + *ss = (byte >> 6) & 0x3; +} + +/* + * Get the byte following the op code and separate it into the + * mode, register, and r/m fields. + */ +static void +dtrace_get_modrm(dis86_t *x, uint_t *mode, uint_t *reg, uint_t *r_m) +{ + if (x->d86_got_modrm == 0) { + if (x->d86_rmindex == -1) + x->d86_rmindex = x->d86_len; + dtrace_get_SIB(x, mode, reg, r_m); + x->d86_got_modrm = 1; + } +} + +/* + * Adjust register selection based on any REX prefix bits present. + */ +/*ARGSUSED*/ +static void +dtrace_rex_adjust(uint_t rex_prefix, uint_t mode, uint_t *reg, uint_t *r_m) +{ + if (reg != NULL && r_m == NULL) { + if (rex_prefix & REX_B) + *reg += 8; + } else { + if (reg != NULL && (REX_R & rex_prefix) != 0) + *reg += 8; + if (r_m != NULL && (REX_B & rex_prefix) != 0) + *r_m += 8; + } +} + +/* + * Get an immediate operand of the given size, with sign extension. + */ +static void +dtrace_imm_opnd(dis86_t *x, int wbit, int size, int opindex) +{ + int i; + int byte; + int valsize = 0; + + if (x->d86_numopnds < opindex + 1) + x->d86_numopnds = opindex + 1; + + switch (wbit) { + case BYTE_OPND: + valsize = 1; + break; + case LONG_OPND: + if (x->d86_opnd_size == SIZE16) + valsize = 2; + else if (x->d86_opnd_size == SIZE32) + valsize = 4; + else + valsize = 8; + break; + case MM_OPND: + case XMM_OPND: + case SEG_OPND: + case CONTROL_OPND: + case DEBUG_OPND: + case TEST_OPND: + valsize = size; + break; + case WORD_OPND: + valsize = 2; + break; + } + if (valsize < size) + valsize = size; + + if (x->d86_error) + return; + x->d86_opnd[opindex].d86_value = 0; + for (i = 0; i < size; ++i) { + byte = x->d86_get_byte(x->d86_data); + if (byte < 0) { + x->d86_error = 1; + return; + } + x->d86_bytes[x->d86_len++] = byte; + x->d86_opnd[opindex].d86_value |= (uint64_t)byte << (i * 8); + } + /* Do sign extension */ + if (x->d86_bytes[x->d86_len - 1] & 0x80) { + for (; i < valsize; i++) + x->d86_opnd[opindex].d86_value |= + (uint64_t)0xff << (i* 8); + } +#ifdef DIS_TEXT + x->d86_opnd[opindex].d86_mode = MODE_SIGNED; + x->d86_opnd[opindex].d86_value_size = valsize; + x->d86_imm_bytes += size; +#endif +} + +/* + * Get an ip relative operand of the given size, with sign extension. + */ +static void +dtrace_disp_opnd(dis86_t *x, int wbit, int size, int opindex) +{ + dtrace_imm_opnd(x, wbit, size, opindex); +#ifdef DIS_TEXT + x->d86_opnd[opindex].d86_mode = MODE_IPREL; +#endif +} + +/* + * Check to see if there is a segment override prefix pending. + * If so, print it in the current 'operand' location and set + * the override flag back to false. + */ +/*ARGSUSED*/ +static void +dtrace_check_override(dis86_t *x, int opindex) +{ +#ifdef DIS_TEXT + if (x->d86_seg_prefix) { + (void) strlcat(x->d86_opnd[opindex].d86_prefix, + x->d86_seg_prefix, PFIXLEN); + } +#endif + x->d86_seg_prefix = NULL; +} + + +/* + * Process a single instruction Register or Memory operand. + * + * mode = addressing mode from ModRM byte + * r_m = r_m (or reg if mode == 3) field from ModRM byte + * wbit = indicates which register (8bit, 16bit, ... MMX, etc.) set to use. + * o = index of operand that we are processing (0, 1 or 2) + * + * the value of reg or r_m must have already been adjusted for any REX prefix. + */ +/*ARGSUSED*/ +static void +dtrace_get_operand(dis86_t *x, uint_t mode, uint_t r_m, int wbit, int opindex) +{ + int have_SIB = 0; /* flag presence of scale-index-byte */ + uint_t ss; /* scale-factor from opcode */ + uint_t index; /* index register number */ + uint_t base; /* base register number */ + int dispsize; /* size of displacement in bytes */ +#ifdef DIS_TEXT + char *opnd = x->d86_opnd[opindex].d86_opnd; +#endif + + if (x->d86_numopnds < opindex + 1) + x->d86_numopnds = opindex + 1; + + if (x->d86_error) + return; + + /* + * first handle a simple register + */ + if (mode == REG_ONLY) { +#ifdef DIS_TEXT + switch (wbit) { + case MM_OPND: + (void) strlcat(opnd, dis_MMREG[r_m], OPLEN); + break; + case XMM_OPND: + (void) strlcat(opnd, dis_XMMREG[r_m], OPLEN); + break; + case SEG_OPND: + (void) strlcat(opnd, dis_SEGREG[r_m], OPLEN); + break; + case CONTROL_OPND: + (void) strlcat(opnd, dis_CONTROLREG[r_m], OPLEN); + break; + case DEBUG_OPND: + (void) strlcat(opnd, dis_DEBUGREG[r_m], OPLEN); + break; + case TEST_OPND: + (void) strlcat(opnd, dis_TESTREG[r_m], OPLEN); + break; + case BYTE_OPND: + if (x->d86_rex_prefix == 0) + (void) strlcat(opnd, dis_REG8[r_m], OPLEN); + else + (void) strlcat(opnd, dis_REG8_REX[r_m], OPLEN); + break; + case WORD_OPND: + (void) strlcat(opnd, dis_REG16[r_m], OPLEN); + break; + case LONG_OPND: + if (x->d86_opnd_size == SIZE16) + (void) strlcat(opnd, dis_REG16[r_m], OPLEN); + else if (x->d86_opnd_size == SIZE32) + (void) strlcat(opnd, dis_REG32[r_m], OPLEN); + else + (void) strlcat(opnd, dis_REG64[r_m], OPLEN); + break; + } +#endif /* DIS_TEXT */ + return; + } + + /* + * if symbolic representation, skip override prefix, if any + */ + dtrace_check_override(x, opindex); + + /* + * Handle 16 bit memory references first, since they decode + * the mode values more simply. + * mode 1 is r_m + 8 bit displacement + * mode 2 is r_m + 16 bit displacement + * mode 0 is just r_m, unless r_m is 6 which is 16 bit disp + */ + if (x->d86_addr_size == SIZE16) { + if ((mode == 0 && r_m == 6) || mode == 2) + dtrace_imm_opnd(x, WORD_OPND, 2, opindex); + else if (mode == 1) + dtrace_imm_opnd(x, BYTE_OPND, 1, opindex); +#ifdef DIS_TEXT + if (mode == 0 && r_m == 6) + x->d86_opnd[opindex].d86_mode = MODE_SIGNED; + else if (mode == 0) + x->d86_opnd[opindex].d86_mode = MODE_NONE; + else + x->d86_opnd[opindex].d86_mode = MODE_OFFSET; + (void) strlcat(opnd, dis_addr16[mode][r_m], OPLEN); +#endif + return; + } + + /* + * 32 and 64 bit addressing modes are more complex since they + * can involve an SIB (scaled index and base) byte to decode. + */ + if (r_m == ESP_REGNO || r_m == ESP_REGNO + 8) { + have_SIB = 1; + dtrace_get_SIB(x, &ss, &index, &base); + if (x->d86_error) + return; + if (base != 5 || mode != 0) + if (x->d86_rex_prefix & REX_B) + base += 8; + if (x->d86_rex_prefix & REX_X) + index += 8; + } else { + base = r_m; + } + + /* + * Compute the displacement size and get its bytes + */ + dispsize = 0; + + if (mode == 1) + dispsize = 1; + else if (mode == 2) + dispsize = 4; + else if ((r_m & 7) == EBP_REGNO || + (have_SIB && (base & 7) == EBP_REGNO)) + dispsize = 4; + + if (dispsize > 0) { + dtrace_imm_opnd(x, dispsize == 4 ? LONG_OPND : BYTE_OPND, + dispsize, opindex); + if (x->d86_error) + return; + } + +#ifdef DIS_TEXT + if (dispsize > 0) + x->d86_opnd[opindex].d86_mode = MODE_OFFSET; + + if (have_SIB == 0) { + if (x->d86_mode == SIZE32) { + if (mode == 0) + (void) strlcat(opnd, dis_addr32_mode0[r_m], + OPLEN); + else + (void) strlcat(opnd, dis_addr32_mode12[r_m], + OPLEN); + } else { + if (mode == 0) + (void) strlcat(opnd, dis_addr64_mode0[r_m], + OPLEN); + else + (void) strlcat(opnd, dis_addr64_mode12[r_m], + OPLEN); + } + } else { + uint_t need_paren = 0; + char **regs; + if (x->d86_mode == SIZE32) /* NOTE this is not addr_size! */ + regs = (char **)dis_REG32; + else + regs = (char **)dis_REG64; + + /* + * print the base (if any) + */ + if (base == EBP_REGNO && mode == 0) { + if (index != ESP_REGNO) { + (void) strlcat(opnd, "(", OPLEN); + need_paren = 1; + } + } else { + (void) strlcat(opnd, "(", OPLEN); + (void) strlcat(opnd, regs[base], OPLEN); + need_paren = 1; + } + + /* + * print the index (if any) + */ + if (index != ESP_REGNO) { + (void) strlcat(opnd, ",", OPLEN); + (void) strlcat(opnd, regs[index], OPLEN); + (void) strlcat(opnd, dis_scale_factor[ss], OPLEN); + } else + if (need_paren) + (void) strlcat(opnd, ")", OPLEN); + } +#endif +} + +/* + * Operand sequence for standard instruction involving one register + * and one register/memory operand. + * wbit indicates a byte(0) or opnd_size(1) operation + * vbit indicates direction (0 for "opcode r,r_m") or (1 for "opcode r_m, r") + */ +#define STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, vbit) { \ + dtrace_get_modrm(x, &mode, ®, &r_m); \ + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); \ + dtrace_get_operand(x, mode, r_m, wbit, vbit); \ + dtrace_get_operand(x, REG_ONLY, reg, wbit, 1 - vbit); \ +} + +/* + * Similar to above, but allows for the two operands to be of different + * classes (ie. wbit). + * wbit is for the r_m operand + * w2 is for the reg operand + */ +#define MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, w2, vbit) { \ + dtrace_get_modrm(x, &mode, ®, &r_m); \ + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); \ + dtrace_get_operand(x, mode, r_m, wbit, vbit); \ + dtrace_get_operand(x, REG_ONLY, reg, w2, 1 - vbit); \ +} + +/* + * Similar, but for 2 operands plus an immediate. + */ +#define THREEOPERAND(x, mode, reg, r_m, rex_prefix, wbit, w2, immsize) { \ + dtrace_get_modrm(x, &mode, ®, &r_m); \ + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); \ + dtrace_get_operand(x, mode, r_m, wbit, 1); \ + dtrace_get_operand(x, REG_ONLY, reg, w2, 2); \ + dtrace_imm_opnd(x, wbit, immsize, 0); \ +} + +/* + * Dissassemble a single x86 or amd64 instruction. + * + * Mode determines the default operating mode (SIZE16, SIZE32 or SIZE64) + * for interpreting instructions. + * + * returns non-zero for bad opcode + */ +int +dtrace_disx86(dis86_t *x, uint_t cpu_mode) +{ + const instable_t *dp = NULL; /* decode table being used */ +#ifdef DIS_TEXT + uint_t i; +#endif +#ifdef DIS_MEM + uint_t nomem = 0; +#define NOMEM (nomem = 1) +#else +#define NOMEM /* nothing */ +#endif + uint_t wbit = 0; /* opcode wbit, 0 is 8 bit, !0 for opnd_size */ + uint_t w2; /* wbit value for second operand */ + uint_t vbit; + uint_t mode = 0; /* mode value from ModRM byte */ + uint_t reg; /* reg value from ModRM byte */ + uint_t r_m; /* r_m value from ModRM byte */ + + uint_t opcode1; /* high nibble of 1st byte */ + uint_t opcode2; /* low nibble of 1st byte */ + uint_t opcode3; /* extra opcode bits usually from ModRM byte */ + uint_t opcode4; /* high nibble of 2nd byte */ + uint_t opcode5; /* low nibble of 2ne byte */ + uint_t opcode6; /* high nibble of 3rd byte */ + uint_t opcode7; /* low nibble of 3rd byte */ + uint_t opcode_bytes = 1; + + /* + * legacy prefixes come in 5 flavors, you should have only one of each + */ + uint_t opnd_size_prefix = 0; + uint_t addr_size_prefix = 0; + uint_t segment_prefix = 0; + uint_t lock_prefix = 0; + uint_t rep_prefix = 0; + uint_t rex_prefix = 0; /* amd64 register extension prefix */ + size_t off; + + x->d86_len = 0; + x->d86_rmindex = -1; + x->d86_error = 0; +#ifdef DIS_TEXT + x->d86_numopnds = 0; + x->d86_seg_prefix = NULL; + x->d86_mneu[0] = 0; + for (i = 0; i < 3; ++i) { + x->d86_opnd[i].d86_opnd[0] = 0; + x->d86_opnd[i].d86_prefix[0] = 0; + x->d86_opnd[i].d86_value_size = 0; + x->d86_opnd[i].d86_value = 0; + x->d86_opnd[i].d86_mode = MODE_NONE; + } +#endif + x->d86_error = 0; + x->d86_memsize = 0; + + if (cpu_mode == SIZE16) { + opnd_size = SIZE16; + addr_size = SIZE16; + } else if (cpu_mode == SIZE32) { + opnd_size = SIZE32; + addr_size = SIZE32; + } else { + opnd_size = SIZE32; + addr_size = SIZE64; + } + + /* + * Get one opcode byte and check for zero padding that follows + * jump tables. + */ + if (dtrace_get_opcode(x, &opcode1, &opcode2) != 0) + goto error; + + if (opcode1 == 0 && opcode2 == 0 && + x->d86_check_func != NULL && x->d86_check_func(x->d86_data)) { +#ifdef DIS_TEXT + (void) strncpy(x->d86_mneu, ".byte\t0", OPLEN); +#endif + goto done; + } + + /* + * Gather up legacy x86 prefix bytes. + */ + for (;;) { + uint_t *which_prefix = NULL; + + dp = &dis_distable[opcode1][opcode2]; + + switch (dp->it_adrmode) { + case PREFIX: + which_prefix = &rep_prefix; + break; + case LOCK: + which_prefix = &lock_prefix; + break; + case OVERRIDE: + which_prefix = &segment_prefix; +#ifdef DIS_TEXT + x->d86_seg_prefix = (char *)dp->it_name; +#endif + if (dp->it_invalid64 && cpu_mode == SIZE64) + goto error; + break; + case AM: + which_prefix = &addr_size_prefix; + break; + case DM: + which_prefix = &opnd_size_prefix; + break; + } + if (which_prefix == NULL) + break; + *which_prefix = (opcode1 << 4) | opcode2; + if (dtrace_get_opcode(x, &opcode1, &opcode2) != 0) + goto error; + } + + /* + * Handle amd64 mode PREFIX values. + * Some of the segment prefixes are no-ops. (only FS/GS actually work) + * We might have a REX prefix (opcodes 0x40-0x4f) + */ + if (cpu_mode == SIZE64) { + if (segment_prefix != 0x64 && segment_prefix != 0x65) + segment_prefix = 0; + + if (opcode1 == 0x4) { + rex_prefix = (opcode1 << 4) | opcode2; + if (dtrace_get_opcode(x, &opcode1, &opcode2) != 0) + goto error; + dp = &dis_distable[opcode1][opcode2]; + } + } + + /* + * Deal with selection of operand and address size now. + * Note that the REX.W bit being set causes opnd_size_prefix to be + * ignored. + */ + if (cpu_mode == SIZE64) { + if (rex_prefix & 0x08) + opnd_size = SIZE64; + else if (opnd_size_prefix) + opnd_size = SIZE16; + + if (addr_size_prefix) + addr_size = SIZE32; + } else if (cpu_mode == SIZE32) { + if (opnd_size_prefix) + opnd_size = SIZE16; + if (addr_size_prefix) + addr_size = SIZE16; + } else { + if (opnd_size_prefix) + opnd_size = SIZE32; + if (addr_size_prefix) + addr_size = SIZE32; + } + + /* + * The pause instruction - a repz'd nop. This doesn't fit + * with any of the other prefix goop added for SSE, so we'll + * special-case it here. + */ + if (rep_prefix == 0xf3 && opcode1 == 0x9 && opcode2 == 0x0) { + rep_prefix = 0; + dp = &dis_opPause; + } + + /* + * Some 386 instructions have 2 bytes of opcode before the mod_r/m + * byte so we may need to perform a table indirection. + */ + if (dp->it_indirect == dis_op0F[0]) { + if (dtrace_get_opcode(x, &opcode4, &opcode5) != 0) + goto error; + opcode_bytes = 2; + if (opcode4 == 0x7 && opcode5 >= 0x1 && opcode5 <= 0x3) { + uint_t subcode; + + if (dtrace_get_opcode(x, &opcode6, &opcode7) != 0) + goto error; + opcode_bytes = 3; + subcode = ((opcode6 & 0x3) << 1) | + ((opcode7 & 0x8) >> 3); + dp = &dis_op0F7123[opcode5][subcode]; + } else if ((opcode4 == 0xc) && (opcode5 >= 0x8)) { + dp = &dis_op0FC8[0]; + } else { + dp = &dis_op0F[opcode4][opcode5]; + } + } + + /* + * If still not at a TERM decode entry, then a ModRM byte + * exists and its fields further decode the instruction. + */ + x->d86_got_modrm = 0; + if (dp->it_indirect != TERM) { + dtrace_get_modrm(x, &mode, &opcode3, &r_m); + if (x->d86_error) + goto error; + reg = opcode3; + + /* + * decode 287 instructions (D8-DF) from opcodeN + */ + if (opcode1 == 0xD && opcode2 >= 0x8) { + if (opcode2 == 0xB && mode == 0x3 && opcode3 == 4) + dp = &dis_opFP5[r_m]; + else if (opcode2 == 0xA && mode == 0x3 && opcode3 < 4) + dp = &dis_opFP7[opcode3]; + else if (opcode2 == 0xB && mode == 0x3) + dp = &dis_opFP6[opcode3]; + else if (opcode2 == 0x9 && mode == 0x3 && opcode3 >= 4) + dp = &dis_opFP4[opcode3 - 4][r_m]; + else if (mode == 0x3) + dp = &dis_opFP3[opcode2 - 8][opcode3]; + else + dp = &dis_opFP1n2[opcode2 - 8][opcode3]; + } else { + dp = dp->it_indirect + opcode3; + } + } + + /* + * In amd64 bit mode, ARPL opcode is changed to MOVSXD + * (sign extend 32bit to 64 bit) + */ + if (cpu_mode == SIZE64 && opcode1 == 0x6 && opcode2 == 0x3) + dp = &dis_opMOVSLD; + + /* + * at this point we should have a correct (or invalid) opcode + */ + if ((cpu_mode == SIZE64 && dp->it_invalid64) || + (cpu_mode != SIZE64 && dp->it_invalid32)) + goto error; + if (dp->it_indirect != TERM) + goto error; + + /* + * deal with MMX/SSE opcodes which are changed by prefixes + */ + switch (dp->it_adrmode) { + case MMO: + case MMOIMPL: + case MMO3P: + case MMOM3: + case MMOMS: + case MMOPM: + case MMOPRM: + case MMOS: + case XMMO: + case XMMOM: + case XMMOMS: + case XMMOPM: + case XMMOS: + case XMMOMX: + case XMMOX3: + case XMMOXMM: + /* + * This is horrible. Some SIMD instructions take the + * form 0x0F 0x?? ..., which is easily decoded using the + * existing tables. Other SIMD instructions use various + * prefix bytes to overload existing instructions. For + * Example, addps is F0, 58, whereas addss is F3 (repz), + * F0, 58. Presumably someone got a raise for this. + * + * If we see one of the instructions which can be + * modified in this way (if we've got one of the SIMDO* + * address modes), we'll check to see if the last prefix + * was a repz. If it was, we strip the prefix from the + * mnemonic, and we indirect using the dis_opSIMDrepz + * table. + */ + + /* + * Calculate our offset in dis_op0F + */ + if ((uintptr_t)dp - (uintptr_t)dis_op0F > sizeof (dis_op0F)) + goto error; + + off = ((uintptr_t)dp - (uintptr_t)dis_op0F) / + sizeof (instable_t); + + /* + * Rewrite if this instruction used one of the magic prefixes. + */ + if (rep_prefix) { + if (rep_prefix == 0xf2) + dp = &dis_opSIMDrepnz[off]; + else + dp = &dis_opSIMDrepz[off]; + rep_prefix = 0; + } else if (opnd_size_prefix) { + dp = &dis_opSIMDdata16[off]; + opnd_size_prefix = 0; + if (opnd_size == SIZE16) + opnd_size = SIZE32; + } + break; + + case MMOSH: + /* + * As with the "normal" SIMD instructions, the MMX + * shuffle instructions are overloaded. These + * instructions, however, are special in that they use + * an extra byte, and thus an extra table. As of this + * writing, they only use the opnd_size prefix. + */ + + /* + * Calculate our offset in dis_op0F7123 + */ + if ((uintptr_t)dp - (uintptr_t)dis_op0F7123 > + sizeof (dis_op0F7123)) + goto error; + + if (opnd_size_prefix) { + off = ((uintptr_t)dp - (uintptr_t)dis_op0F7123) / + sizeof (instable_t); + dp = &dis_opSIMD7123[off]; + opnd_size_prefix = 0; + if (opnd_size == SIZE16) + opnd_size = SIZE32; + } + break; + } + + /* + * In 64 bit mode, some opcodes automatically use opnd_size == SIZE64. + */ + if (cpu_mode == SIZE64) + if (dp->it_always64 || (opnd_size == SIZE32 && dp->it_stackop)) + opnd_size = SIZE64; + +#ifdef DIS_TEXT + /* + * At this point most instructions can format the opcode mnemonic + * including the prefixes. + */ + if (lock_prefix) + (void) strlcat(x->d86_mneu, "lock ", OPLEN); + + if (rep_prefix == 0xf2) + (void) strlcat(x->d86_mneu, "repnz ", OPLEN); + else if (rep_prefix == 0xf3) + (void) strlcat(x->d86_mneu, "repz ", OPLEN); + + if (cpu_mode == SIZE64 && addr_size_prefix) + (void) strlcat(x->d86_mneu, "addr32 ", OPLEN); + + if (dp->it_adrmode != CBW && + dp->it_adrmode != CWD && + dp->it_adrmode != XMMSFNC) { + if (strcmp(dp->it_name, "INVALID") == 0) + goto error; + (void) strlcat(x->d86_mneu, dp->it_name, OPLEN); + if (dp->it_suffix) { + char *types[] = {"", "w", "l", "q"}; + if (opcode_bytes == 2 && opcode4 == 4) { + /* It's a cmovx.yy. Replace the suffix x */ + for (i = 5; i < OPLEN; i++) { + if (x->d86_mneu[i] == '.') + break; + } + x->d86_mneu[i - 1] = *types[opnd_size]; + } else { + (void) strlcat(x->d86_mneu, types[opnd_size], + OPLEN); + } + } + } +#endif + + /* + * Process operands based on the addressing modes. + */ + x->d86_mode = cpu_mode; + x->d86_rex_prefix = rex_prefix; + x->d86_opnd_size = opnd_size; + x->d86_addr_size = addr_size; + vbit = 0; /* initialize for mem/reg -> reg */ + switch (dp->it_adrmode) { + /* + * amd64 instruction to sign extend 32 bit reg/mem operands + * into 64 bit register values + */ + case MOVSXZ: +#ifdef DIS_TEXT + if (rex_prefix == 0) + (void) strncpy(x->d86_mneu, "movzld", OPLEN); +#endif + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + x->d86_opnd_size = SIZE64; + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 1); + x->d86_opnd_size = opnd_size = SIZE32; + wbit = LONG_OPND; + dtrace_get_operand(x, mode, r_m, wbit, 0); + break; + + /* + * movsbl movsbw movsbq (0x0FBE) or movswl movswq (0x0FBF) + * movzbl movzbw movzbq (0x0FB6) or mobzwl movzwq (0x0FB7) + * wbit lives in 2nd byte, note that operands + * are different sized + */ + case MOVZ: + if (rex_prefix & REX_W) { + /* target register size = 64 bit */ + x->d86_mneu[5] = 'q'; + } + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 1); + x->d86_opnd_size = opnd_size = SIZE16; + wbit = WBIT(opcode5); + dtrace_get_operand(x, mode, r_m, wbit, 0); + break; + + /* + * imul instruction, with either 8-bit or longer immediate + * opcode 0x6B for byte, sign-extended displacement, 0x69 for word(s) + */ + case IMUL: + wbit = LONG_OPND; + THREEOPERAND(x, mode, reg, r_m, rex_prefix, wbit, LONG_OPND, + OPSIZE(opnd_size, opcode2 == 0x9)); + break; + + /* memory or register operand to register, with 'w' bit */ + case MRw: + wbit = WBIT(opcode2); + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 0); + break; + + /* register to memory or register operand, with 'w' bit */ + /* arpl happens to fit here also because it is odd */ + case RMw: + if (opcode_bytes == 2) + wbit = WBIT(opcode5); + else + wbit = WBIT(opcode2); + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 1); + break; + + /* xaddb instruction */ + case XADDB: + wbit = 0; + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 1); + break; + + /* MMX register to memory or register operand */ + case MMS: + case MMOS: +#ifdef DIS_TEXT + wbit = strcmp(dp->it_name, "movd") ? MM_OPND : LONG_OPND; +#else + wbit = LONG_OPND; +#endif + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, MM_OPND, 1); + break; + + /* MMX register to memory */ + case MMOMS: + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode == REG_ONLY) + goto error; + wbit = MM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, MM_OPND, 1); + break; + + /* Double shift. Has immediate operand specifying the shift. */ + case DSHIFT: + wbit = LONG_OPND; + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 2); + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 1); + dtrace_imm_opnd(x, wbit, 1, 0); + break; + + /* + * Double shift. With no immediate operand, specifies using %cl. + */ + case DSHIFTcl: + wbit = LONG_OPND; + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 1); + break; + + /* immediate to memory or register operand */ + case IMlw: + wbit = WBIT(opcode2); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 1); + /* + * Have long immediate for opcode 0x81, but not 0x80 nor 0x83 + */ + dtrace_imm_opnd(x, wbit, OPSIZE(opnd_size, opcode2 == 1), 0); + break; + + /* immediate to memory or register operand with the */ + /* 'w' bit present */ + case IMw: + wbit = WBIT(opcode2); + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 1); + dtrace_imm_opnd(x, wbit, OPSIZE(opnd_size, wbit), 0); + break; + + /* immediate to register with register in low 3 bits */ + /* of op code */ + case IR: + /* w-bit here (with regs) is bit 3 */ + wbit = opcode2 >>3 & 0x1; + reg = REGNO(opcode2); + dtrace_rex_adjust(rex_prefix, mode, ®, NULL); + mode = REG_ONLY; + r_m = reg; + dtrace_get_operand(x, mode, r_m, wbit, 1); + dtrace_imm_opnd(x, wbit, OPSIZE64(opnd_size, wbit), 0); + break; + + /* MMX immediate shift of register */ + case MMSH: + case MMOSH: + wbit = MM_OPND; + goto mm_shift; /* in next case */ + + /* SIMD immediate shift of register */ + case XMMSH: + wbit = XMM_OPND; +mm_shift: + reg = REGNO(opcode7); + dtrace_rex_adjust(rex_prefix, mode, ®, NULL); + dtrace_get_operand(x, REG_ONLY, reg, wbit, 1); + dtrace_imm_opnd(x, wbit, 1, 0); + NOMEM; + break; + + /* accumulator to memory operand */ + case AO: + vbit = 1; + /*FALLTHROUGH*/ + + /* memory operand to accumulator */ + case OA: + wbit = WBIT(opcode2); + dtrace_get_operand(x, REG_ONLY, EAX_REGNO, wbit, 1 - vbit); + dtrace_imm_opnd(x, wbit, OPSIZE64(addr_size, LONG_OPND), vbit); +#ifdef DIS_TEXT + x->d86_opnd[vbit].d86_mode = MODE_OFFSET; +#endif + break; + + + /* segment register to memory or register operand */ + case SM: + vbit = 1; + /*FALLTHROUGH*/ + + /* memory or register operand to segment register */ + case MS: + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, LONG_OPND, vbit); + dtrace_get_operand(x, REG_ONLY, reg, SEG_OPND, 1 - vbit); + break; + + /* + * rotate or shift instructions, which may shift by 1 or + * consult the cl register, depending on the 'v' bit + */ + case Mv: + vbit = VBIT(opcode2); + wbit = WBIT(opcode2); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 1); +#ifdef DIS_TEXT + if (vbit) { + (void) strlcat(x->d86_opnd[0].d86_opnd, "%cl", OPLEN); + } else { + x->d86_opnd[0].d86_mode = MODE_SIGNED; + x->d86_opnd[0].d86_value_size = 1; + x->d86_opnd[0].d86_value = 1; + } +#endif + break; + /* + * immediate rotate or shift instructions + */ + case MvI: + wbit = WBIT(opcode2); +normal_imm_mem: + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 1); + dtrace_imm_opnd(x, wbit, 1, 0); + break; + + /* bit test instructions */ + case MIb: + wbit = LONG_OPND; + goto normal_imm_mem; + + /* single memory or register operand with 'w' bit present */ + case Mw: + wbit = WBIT(opcode2); +just_mem: + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 0); + break; + + case SWAPGS: + if (cpu_mode == SIZE64 && mode == 3 && r_m == 0) { +#ifdef DIS_TEXT + (void) strncpy(x->d86_mneu, "swapgs", OPLEN); +#endif + NOMEM; + break; + } + /*FALLTHROUGH*/ + + /* prefetch instruction - memory operand, but no memory acess */ + case PREF: + NOMEM; + /*FALLTHROUGH*/ + + /* single memory or register operand */ + case M: + wbit = LONG_OPND; + goto just_mem; + + /* single memory or register byte operand */ + case Mb: + wbit = BYTE_OPND; + goto just_mem; + + case MO: + /* Similar to M, but only memory (no direct registers) */ + wbit = LONG_OPND; + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode == 3) + goto error; + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 0); + break; + + /* move special register to register or reverse if vbit */ + case SREG: + switch (opcode5) { + + case 2: + vbit = 1; + /*FALLTHROUGH*/ + case 0: + wbit = CONTROL_OPND; + break; + + case 3: + vbit = 1; + /*FALLTHROUGH*/ + case 1: + wbit = DEBUG_OPND; + break; + + case 6: + vbit = 1; + /*FALLTHROUGH*/ + case 4: + wbit = TEST_OPND; + break; + + } + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, REG_ONLY, reg, wbit, vbit); + dtrace_get_operand(x, REG_ONLY, r_m, LONG_OPND, 1 - vbit); + NOMEM; + break; + + /* + * single register operand with register in the low 3 + * bits of op code + */ + case R: + if (opcode_bytes == 2) + reg = REGNO(opcode5); + else + reg = REGNO(opcode2); + dtrace_rex_adjust(rex_prefix, mode, ®, NULL); + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 0); + NOMEM; + break; + + /* + * register to accumulator with register in the low 3 + * bits of op code, xchg instructions + */ + case RA: + NOMEM; + reg = REGNO(opcode2); + dtrace_rex_adjust(rex_prefix, mode, ®, NULL); + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 0); + dtrace_get_operand(x, REG_ONLY, EAX_REGNO, LONG_OPND, 1); + break; + + /* + * single segment register operand, with register in + * bits 3-4 of op code byte + */ + case SEG: + NOMEM; + reg = (x->d86_bytes[x->d86_len - 1] >> 3) & 0x3; + dtrace_get_operand(x, REG_ONLY, reg, SEG_OPND, 0); + break; + + /* + * single segment register operand, with register in + * bits 3-5 of op code + */ + case LSEG: + NOMEM; + /* long seg reg from opcode */ + reg = (x->d86_bytes[x->d86_len - 1] >> 3) & 0x7; + dtrace_get_operand(x, REG_ONLY, reg, SEG_OPND, 0); + break; + + /* memory or register operand to register */ + case MR: + wbit = LONG_OPND; + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 0); + break; + + case RM: + wbit = LONG_OPND; + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 1); + break; + + /* MMX/SIMD-Int memory or mm reg to mm reg */ + case MM: + case MMO: +#ifdef DIS_TEXT + wbit = strcmp(dp->it_name, "movd") ? MM_OPND : LONG_OPND; +#else + wbit = LONG_OPND; +#endif + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, MM_OPND, 0); + break; + + case MMOIMPL: +#ifdef DIS_TEXT + wbit = strcmp(dp->it_name, "movd") ? MM_OPND : LONG_OPND; +#else + wbit = LONG_OPND; +#endif + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode != REG_ONLY) + goto error; + + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 0); + dtrace_get_operand(x, REG_ONLY, reg, MM_OPND, 1); + mode = 0; /* change for memory access size... */ + break; + + /* MMX/SIMD-Int and SIMD-FP predicated mm reg to r32 */ + case MMO3P: + wbit = MM_OPND; + goto xmm3p; + case XMM3P: + wbit = XMM_OPND; +xmm3p: + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode != REG_ONLY) + goto error; + + THREEOPERAND(x, mode, reg, r_m, rex_prefix, wbit, LONG_OPND, 1); + NOMEM; + break; + + /* MMX/SIMD-Int predicated r32/mem to mm reg */ + case MMOPRM: + wbit = LONG_OPND; + w2 = MM_OPND; + goto xmmprm; + case XMMPRM: + wbit = LONG_OPND; + w2 = XMM_OPND; +xmmprm: + THREEOPERAND(x, mode, reg, r_m, rex_prefix, wbit, w2, 1); + break; + + /* MMX/SIMD-Int predicated mm/mem to mm reg */ + case MMOPM: + wbit = w2 = MM_OPND; + goto xmmprm; + + /* MMX/SIMD-Int mm reg to r32 */ + case MMOM3: + NOMEM; + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode != REG_ONLY) + goto error; + wbit = MM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, LONG_OPND, 0); + break; + + /* SIMD memory or xmm reg operand to xmm reg */ + case XMM: + case XMMO: + case XMMXIMPL: + wbit = XMM_OPND; + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 0); + + if (dp->it_adrmode == XMMXIMPL && mode != REG_ONLY) + goto error; + +#ifdef DIS_TEXT + /* + * movlps and movhlps share opcodes. They differ in the + * addressing modes allowed for their operands. + * movhps and movlhps behave similarly. + */ + if (mode == REG_ONLY) { + if (strcmp(dp->it_name, "movlps") == 0) + (void) strncpy(x->d86_mneu, "movhlps", OPLEN); + else if (strcmp(dp->it_name, "movhps") == 0) + (void) strncpy(x->d86_mneu, "movlhps", OPLEN); + } +#endif + if (dp->it_adrmode == XMMXIMPL) + mode = 0; /* change for memory access size... */ + break; + + /* SIMD xmm reg to memory or xmm reg */ + case XMMS: + case XMMOS: + case XMMMS: + case XMMOMS: + dtrace_get_modrm(x, &mode, ®, &r_m); +#ifdef DIS_TEXT + if ((strcmp(dp->it_name, "movlps") == 0 || + strcmp(dp->it_name, "movhps") == 0 || + strcmp(dp->it_name, "movntps") == 0) && + mode == REG_ONLY) + goto error; +#endif + wbit = XMM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 1); + break; + + /* SIMD memory to xmm reg */ + case XMMM: + case XMMOM: + wbit = XMM_OPND; + dtrace_get_modrm(x, &mode, ®, &r_m); +#ifdef DIS_TEXT + if (mode == REG_ONLY) { + if (strcmp(dp->it_name, "movhps") == 0) + (void) strncpy(x->d86_mneu, "movlhps", OPLEN); + else + goto error; + } +#endif + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 0); + break; + + /* SIMD memory or r32 to xmm reg */ + case XMM3MX: + wbit = LONG_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 0); + break; + + case XMM3MXS: + wbit = LONG_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 1); + break; + + /* SIMD memory or mm reg to xmm reg */ + case XMMOMX: + /* SIMD mm to xmm */ + case XMMMX: + wbit = MM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 0); + break; + + /* SIMD memory or xmm reg to mm reg */ + case XMMXMM: + case XMMOXMM: + case XMMXM: + wbit = XMM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, MM_OPND, 0); + break; + + + /* SIMD memory or xmm reg to r32 */ + case XMMXM3: + wbit = XMM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, LONG_OPND, 0); + break; + + /* SIMD xmm to r32 */ + case XMMX3: + case XMMOX3: + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode != REG_ONLY) + goto error; + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, mode, r_m, XMM_OPND, 0); + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 1); + NOMEM; + break; + + /* SIMD predicated memory or xmm reg with/to xmm reg */ + case XMMP: + case XMMOPM: + wbit = XMM_OPND; + THREEOPERAND(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 1); + +#ifdef DIS_TEXT + /* + * cmpps and cmpss vary their instruction name based + * on the value of imm8. Other XMMP instructions, + * such as shufps, require explicit specification of + * the predicate. + */ + if (dp->it_name[0] == 'c' && + dp->it_name[1] == 'm' && + dp->it_name[2] == 'p' && + strlen(dp->it_name) == 5) { + uchar_t pred = x->d86_opnd[0].d86_value & 0xff; + + if (pred >= (sizeof (dis_PREDSUFFIX) / sizeof (char *))) + goto error; + + (void) strncpy(x->d86_mneu, "cmp", OPLEN); + (void) strlcat(x->d86_mneu, dis_PREDSUFFIX[pred], + OPLEN); + (void) strlcat(x->d86_mneu, + dp->it_name + strlen(dp->it_name) - 2, + OPLEN); + x->d86_opnd[0] = x->d86_opnd[1]; + x->d86_opnd[1] = x->d86_opnd[2]; + x->d86_numopnds = 2; + } +#endif + break; + + /* immediate operand to accumulator */ + case IA: + wbit = WBIT(opcode2); + dtrace_get_operand(x, REG_ONLY, EAX_REGNO, wbit, 1); + dtrace_imm_opnd(x, wbit, OPSIZE(opnd_size, wbit), 0); + NOMEM; + break; + + /* memory or register operand to accumulator */ + case MA: + wbit = WBIT(opcode2); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 0); + break; + + /* si register to di register used to reference memory */ + case SD: +#ifdef DIS_TEXT + dtrace_check_override(x, 0); + x->d86_numopnds = 2; + if (addr_size == SIZE64) { + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%rsi)", + OPLEN); + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%rdi)", + OPLEN); + } else if (addr_size == SIZE32) { + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%esi)", + OPLEN); + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%edi)", + OPLEN); + } else { + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%si)", + OPLEN); + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%di)", + OPLEN); + } +#endif + wbit = LONG_OPND; + break; + + /* accumulator to di register */ + case AD: + wbit = WBIT(opcode2); +#ifdef DIS_TEXT + dtrace_check_override(x, 1); + x->d86_numopnds = 2; + dtrace_get_operand(x, REG_ONLY, EAX_REGNO, wbit, 0); + if (addr_size == SIZE64) + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%rdi)", + OPLEN); + else if (addr_size == SIZE32) + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%edi)", + OPLEN); + else + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%di)", + OPLEN); +#endif + break; + + /* si register to accumulator */ + case SA: + wbit = WBIT(opcode2); +#ifdef DIS_TEXT + dtrace_check_override(x, 0); + x->d86_numopnds = 2; + if (addr_size == SIZE64) + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%rsi)", + OPLEN); + else if (addr_size == SIZE32) + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%esi)", + OPLEN); + else + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%si)", + OPLEN); + dtrace_get_operand(x, REG_ONLY, EAX_REGNO, wbit, 1); +#endif + break; + + /* + * single operand, a 16/32 bit displacement + */ + case D: + wbit = LONG_OPND; + dtrace_disp_opnd(x, wbit, OPSIZE(opnd_size, LONG_OPND), 0); + NOMEM; + break; + + /* jmp/call indirect to memory or register operand */ + case INM: +#ifdef DIS_TEXT + (void) strlcat(x->d86_opnd[0].d86_prefix, "*", OPLEN); +#endif + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, LONG_OPND, 0); + wbit = LONG_OPND; + break; + + /* + * for long jumps and long calls -- a new code segment + * register and an offset in IP -- stored in object + * code in reverse order. Note - not valid in amd64 + */ + case SO: + dtrace_check_override(x, 1); + wbit = LONG_OPND; + dtrace_imm_opnd(x, wbit, OPSIZE(opnd_size, LONG_OPND), 1); +#ifdef DIS_TEXT + x->d86_opnd[1].d86_mode = MODE_SIGNED; +#endif + /* will now get segment operand */ + dtrace_imm_opnd(x, wbit, 2, 0); + break; + + /* + * jmp/call. single operand, 8 bit displacement. + * added to current EIP in 'compofff' + */ + case BD: + dtrace_disp_opnd(x, BYTE_OPND, 1, 0); + NOMEM; + break; + + /* single 32/16 bit immediate operand */ + case I: + wbit = LONG_OPND; + dtrace_imm_opnd(x, wbit, OPSIZE(opnd_size, LONG_OPND), 0); + break; + + /* single 8 bit immediate operand */ + case Ib: + wbit = LONG_OPND; + dtrace_imm_opnd(x, wbit, 1, 0); + break; + + case ENTER: + wbit = LONG_OPND; + dtrace_imm_opnd(x, wbit, 2, 0); + dtrace_imm_opnd(x, wbit, 1, 1); + switch (opnd_size) { + case SIZE64: + x->d86_memsize = (x->d86_opnd[1].d86_value + 1) * 8; + break; + case SIZE32: + x->d86_memsize = (x->d86_opnd[1].d86_value + 1) * 4; + break; + case SIZE16: + x->d86_memsize = (x->d86_opnd[1].d86_value + 1) * 2; + break; + } + + break; + + /* 16-bit immediate operand */ + case RET: + wbit = LONG_OPND; + dtrace_imm_opnd(x, wbit, 2, 0); + break; + + /* single 8 bit port operand */ + case P: + dtrace_check_override(x, 0); + dtrace_imm_opnd(x, BYTE_OPND, 1, 0); + NOMEM; + break; + + /* single operand, dx register (variable port instruction) */ + case V: + x->d86_numopnds = 1; + dtrace_check_override(x, 0); +#ifdef DIS_TEXT + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%dx)", OPLEN); +#endif + NOMEM; + break; + + /* + * The int instruction, which has two forms: + * int 3 (breakpoint) or + * int n, where n is indicated in the subsequent + * byte (format Ib). The int 3 instruction (opcode 0xCC), + * where, although the 3 looks like an operand, + * it is implied by the opcode. It must be converted + * to the correct base and output. + */ + case INT3: +#ifdef DIS_TEXT + x->d86_numopnds = 1; + x->d86_opnd[0].d86_mode = MODE_SIGNED; + x->d86_opnd[0].d86_value_size = 1; + x->d86_opnd[0].d86_value = 3; +#endif + NOMEM; + break; + + /* single 8 bit immediate operand */ + case INTx: + dtrace_imm_opnd(x, BYTE_OPND, 1, 0); + NOMEM; + break; + + /* an unused byte must be discarded */ + case U: + if (x->d86_get_byte(x->d86_data) < 0) + goto error; + x->d86_len++; + NOMEM; + break; + + case CBW: +#ifdef DIS_TEXT + if (opnd_size == SIZE16) + (void) strlcat(x->d86_mneu, "cbtw", OPLEN); + else if (opnd_size == SIZE32) + (void) strlcat(x->d86_mneu, "cwtl", OPLEN); + else + (void) strlcat(x->d86_mneu, "cltq", OPLEN); +#endif + wbit = LONG_OPND; + NOMEM; + break; + + case CWD: +#ifdef DIS_TEXT + if (opnd_size == SIZE16) + (void) strlcat(x->d86_mneu, "cwtd", OPLEN); + else if (opnd_size == SIZE32) + (void) strlcat(x->d86_mneu, "cltd", OPLEN); + else + (void) strlcat(x->d86_mneu, "cqtd", OPLEN); +#endif + wbit = LONG_OPND; + NOMEM; + break; + + case XMMSFNC: + /* + * sfence is sfence if mode is REG_ONLY. If mode isn't + * REG_ONLY, mnemonic should be 'clflush'. + */ + dtrace_get_modrm(x, &mode, ®, &r_m); + + /* sfence doesn't take operands */ +#ifdef DIS_TEXT + if (mode == REG_ONLY) { + (void) strlcat(x->d86_mneu, "sfence", OPLEN); + } else { + (void) strlcat(x->d86_mneu, "clflush", OPLEN); + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, mode, r_m, BYTE_OPND, 0); + NOMEM; + } +#else + if (mode != REG_ONLY) { + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, mode, r_m, BYTE_OPND, 0); + NOMEM; + } +#endif + break; + + /* + * no disassembly, the mnemonic was all there was so go on + */ + case NORM: + if (dp->it_invalid32 && cpu_mode != SIZE64) + goto error; + NOMEM; + /*FALLTHROUGH*/ + case IMPLMEM: + break; + + case XMMFENCE: + /* + * Only the following exact byte sequences are allowed: + * + * 0f ae e8 lfence + * 0f ae f0 mfence + */ + if ((uint8_t)x->d86_bytes[x->d86_len - 1] != 0xe8 && + (uint8_t)x->d86_bytes[x->d86_len - 1] != 0xf0) + goto error; + + break; + + + /* float reg */ + case F: +#ifdef DIS_TEXT + x->d86_numopnds = 1; + (void) strlcat(x->d86_opnd[0].d86_opnd, "%st(X)", OPLEN); + x->d86_opnd[0].d86_opnd[4] = r_m + '0'; +#endif + NOMEM; + break; + + /* float reg to float reg, with ret bit present */ + case FF: + vbit = opcode2 >> 2 & 0x1; /* vbit = 1: st -> st(i) */ + /*FALLTHROUGH*/ + case FFC: /* case for vbit always = 0 */ +#ifdef DIS_TEXT + x->d86_numopnds = 2; + (void) strlcat(x->d86_opnd[1 - vbit].d86_opnd, "%st", OPLEN); + (void) strlcat(x->d86_opnd[vbit].d86_opnd, "%st(X)", OPLEN); + x->d86_opnd[vbit].d86_opnd[4] = r_m + '0'; +#endif + NOMEM; + break; + + /* an invalid op code */ + case AM: + case DM: + case OVERRIDE: + case PREFIX: + case UNKNOWN: + NOMEM; + default: + goto error; + } /* end switch */ + if (x->d86_error) + goto error; + +done: +#ifdef DIS_MEM + /* + * compute the size of any memory accessed by the instruction + */ + if (x->d86_memsize != 0) { + return (0); + } else if (dp->it_stackop) { + switch (opnd_size) { + case SIZE16: + x->d86_memsize = 2; + break; + case SIZE32: + x->d86_memsize = 4; + break; + case SIZE64: + x->d86_memsize = 8; + break; + } + } else if (nomem || mode == REG_ONLY) { + x->d86_memsize = 0; + + } else if (dp->it_size != 0) { + /* + * In 64 bit mode descriptor table entries + * go up to 10 bytes and popf/pushf are always 8 bytes + */ + if (x->d86_mode == SIZE64 && dp->it_size == 6) + x->d86_memsize = 10; + else if (x->d86_mode == SIZE64 && opcode1 == 0x9 && + (opcode2 == 0xc || opcode2 == 0xd)) + x->d86_memsize = 8; + else + x->d86_memsize = dp->it_size; + + } else if (wbit == 0) { + x->d86_memsize = 1; + + } else if (wbit == LONG_OPND) { + if (opnd_size == SIZE64) + x->d86_memsize = 8; + else if (opnd_size == SIZE32) + x->d86_memsize = 4; + else + x->d86_memsize = 2; + + } else if (wbit == SEG_OPND) { + x->d86_memsize = 4; + + } else { + x->d86_memsize = 8; + } +#endif + return (0); + +error: +#ifdef DIS_TEXT + (void) strlcat(x->d86_mneu, "undef", OPLEN); +#endif + return (1); +} + +#ifdef DIS_TEXT + +/* + * Some instructions should have immediate operands printed + * as unsigned integers. We compare against this table. + */ +static char *unsigned_ops[] = { + "or", "and", "xor", "test", "in", "out", "lcall", "ljmp", + "rcr", "rcl", "ror", "rol", "shl", "shr", "sal", "psr", "psl", + 0 +}; + +static int +isunsigned_op(char *opcode) +{ + char *where; + int i; + int is_unsigned = 0; + + /* + * Work back to start of last mnemonic, since we may have + * prefixes on some opcodes. + */ + where = opcode + strlen(opcode) - 1; + while (where > opcode && *where != ' ') + --where; + if (*where == ' ') + ++where; + + for (i = 0; unsigned_ops[i]; ++i) { + if (strncmp(where, unsigned_ops[i], + strlen(unsigned_ops[i]))) + continue; + is_unsigned = 1; + break; + } + return (is_unsigned); +} + +/* ARGSUSED */ +void +dtrace_disx86_str(dis86_t *dis, uint_t mode, uintptr_t pc, char *buf, + size_t buflen) +{ + int i; + + dis->d86_sprintf_func(buf, buflen, "%-6s ", dis->d86_mneu); + + /* + * For PC-relative jumps, the pc is really the next pc after executing + * this instruction, so increment it appropriately. + */ + pc += dis->d86_len; + + for (i = 0; i < dis->d86_numopnds; i++) { + d86opnd_t *op = &dis->d86_opnd[i]; + int64_t sv; + uint64_t mask; + + if (i != 0) + (void) strlcat(buf, ",", buflen); + + (void) strlcat(buf, op->d86_prefix, buflen); + + sv = op->d86_value; + + switch (op->d86_mode) { + + case MODE_NONE: + + (void) strlcat(buf, op->d86_opnd, buflen); + break; + + case MODE_SIGNED: + case MODE_IMPLIED: + case MODE_OFFSET: + + if (dis->d86_seg_prefix) + (void) strlcat(buf, dis->d86_seg_prefix, + buflen); + + switch (op->d86_value_size) { + case 1: + sv = (int8_t)sv; + mask = 0xff; + break; + case 2: + sv = (int16_t)sv; + mask = 0xffff; + break; + case 4: + sv = (int32_t)sv; + mask = 0xffffffff; + break; + case 8: + mask = 0xffffffffffffffffULL; + break; + } + + if (op->d86_mode == MODE_SIGNED || + op->d86_mode == MODE_IMPLIED) + (void) strlcat(buf, "$", buflen); + + if (sv < 0 && sv > -0xffff && + !isunsigned_op(dis->d86_mneu)) { + dis->d86_sprintf_func(buf + strlen(buf), + buflen - strlen(buf), + (dis->d86_flags & DIS_OP_OCTAL) ? + "-0%llo" : "-0x%llx", -sv & mask); + } else { + dis->d86_sprintf_func(buf + strlen(buf), + buflen - strlen(buf), + (dis->d86_flags & DIS_OP_OCTAL) ? + "0%llo" : "0x%llx", sv & mask); + } + (void) strlcat(buf, op->d86_opnd, buflen); + break; + + case MODE_IPREL: + + switch (op->d86_value_size) { + case 1: + sv = (int8_t)sv; + break; + case 2: + sv = (int16_t)sv; + break; + case 4: + sv = (int32_t)sv; + break; + } + + if (sv < 0) + dis->d86_sprintf_func(buf + strlen(buf), + buflen - strlen(buf), + (dis->d86_flags & DIS_OP_OCTAL) ? + "-0%llo" : "-0x%llx", -sv - dis->d86_len); + else + dis->d86_sprintf_func(buf + strlen(buf), + buflen - strlen(buf), + (dis->d86_flags & DIS_OP_OCTAL) ? + "+0%llo" : "+0x%llx", sv + dis->d86_len); + + (void) strlcat(buf, "\t<", buflen); + + if (dis->d86_sym_lookup == NULL || + dis->d86_sym_lookup(dis->d86_data, pc + sv, + buf + strlen(buf), buflen - strlen(buf)) != 0) + dis->d86_sprintf_func(buf + strlen(buf), + buflen - strlen(buf), + (dis->d86_flags & DIS_OP_OCTAL) ? + "0%llo" : "0x%llx", pc + sv); + + (void) strlcat(buf, ">", buflen); + + break; + } + } +} + +#endif /* DIS_TEXT */ diff --git a/sys/cddl/dev/dtrace/amd64/dis_tables.h b/sys/cddl/dev/dtrace/amd64/dis_tables.h new file mode 100644 index 000000000000..b45a8c50e738 --- /dev/null +++ b/sys/cddl/dev/dtrace/amd64/dis_tables.h @@ -0,0 +1,112 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +/* Copyright (c) 1988 AT&T */ +/* All Rights Reserved */ + + +#ifndef _DIS_TABLES_H +#define _DIS_TABLES_H + +#if defined(sun) +#pragma ident "@(#)dis_tables.h 1.7 06/03/02 SMI" +#endif + +/* + * Constants and prototypes for the IA32 disassembler backend. See dis_tables.c + * for usage information and documentation. + */ + +#ifdef __cplusplus +extern "C" { +#endif + +#include <sys/types.h> +#include <sys/param.h> + +/* + * values for cpu mode + */ +#define SIZE16 1 +#define SIZE32 2 +#define SIZE64 3 + +#define OPLEN 256 +#define PFIXLEN 8 +#define NCPS 12 /* number of chars per symbol */ + +/* + * data structures that must be provided to dtrace_dis86() + */ +typedef struct d86opnd { + char d86_opnd[OPLEN]; /* symbolic rep of operand */ + char d86_prefix[PFIXLEN]; /* any prefix string or "" */ + uint_t d86_mode; /* mode for immediate */ + uint_t d86_value_size; /* size in bytes of d86_value */ + uint64_t d86_value; /* immediate value of opnd */ +} d86opnd_t; + +typedef struct dis86 { + uint_t d86_mode; + uint_t d86_error; + uint_t d86_len; /* instruction length */ + int d86_rmindex; /* index of modrm byte or -1 */ + uint_t d86_memsize; /* size of memory referenced */ + char d86_bytes[16]; /* bytes of instruction */ + char d86_mneu[OPLEN]; + uint_t d86_numopnds; + uint_t d86_rex_prefix; /* value of REX prefix if !0 */ + char *d86_seg_prefix; /* segment prefix, if any */ + uint_t d86_opnd_size; + uint_t d86_addr_size; + uint_t d86_got_modrm; + struct d86opnd d86_opnd[3]; /* up to 3 operands */ + int (*d86_check_func)(void *); + int (*d86_get_byte)(void *); +#ifdef DIS_TEXT + int (*d86_sym_lookup)(void *, uint64_t, char *, size_t); + int (*d86_sprintf_func)(char *, size_t, const char *, ...); + int d86_flags; + uint_t d86_imm_bytes; +#endif + void *d86_data; +} dis86_t; + +extern int dtrace_disx86(dis86_t *x, uint_t cpu_mode); + +#define DIS_OP_OCTAL 0x1 /* Print all numbers in octal */ + +#ifdef DIS_TEXT +extern void dtrace_disx86_str(dis86_t *x, uint_t cpu_mode, uintptr_t pc, + char *buf, size_t len); +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* _DIS_TABLES_H */ diff --git a/sys/cddl/dev/dtrace/amd64/dtrace_asm.S b/sys/cddl/dev/dtrace/amd64/dtrace_asm.S new file mode 100644 index 000000000000..cf8314c6dceb --- /dev/null +++ b/sys/cddl/dev/dtrace/amd64/dtrace_asm.S @@ -0,0 +1,573 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * Portions Copyright 2008 John Birrell <jb@freebsd.org> + * + * $FreeBSD$ + * + */ +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#define _ASM + +#include <machine/asmacros.h> +#include <sys/cpuvar_defs.h> +#include <sys/dtrace.h> + +#include "assym.s" + +#define INTR_POP \ + MEXITCOUNT; \ + movq TF_RDI(%rsp),%rdi; \ + movq TF_RSI(%rsp),%rsi; \ + movq TF_RDX(%rsp),%rdx; \ + movq TF_RCX(%rsp),%rcx; \ + movq TF_R8(%rsp),%r8; \ + movq TF_R9(%rsp),%r9; \ + movq TF_RAX(%rsp),%rax; \ + movq TF_RBX(%rsp),%rbx; \ + movq TF_RBP(%rsp),%rbp; \ + movq TF_R10(%rsp),%r10; \ + movq TF_R11(%rsp),%r11; \ + movq TF_R12(%rsp),%r12; \ + movq TF_R13(%rsp),%r13; \ + movq TF_R14(%rsp),%r14; \ + movq TF_R15(%rsp),%r15; \ + testb $SEL_RPL_MASK,TF_CS(%rsp); \ + jz 1f; \ + cli; \ + swapgs; \ +1: addq $TF_RIP,%rsp; + + + .globl calltrap + .type calltrap,@function + ENTRY(dtrace_invop_start) + + /* + * #BP traps with %rip set to the next address. We need to decrement + * the value to indicate the address of the int3 (0xcc) instruction + * that we substituted. + */ + movq TF_RIP(%rsp), %rdi + decq %rdi + movq TF_RSP(%rsp), %rsi + movq TF_RAX(%rsp), %rdx + pushq (%rsi) + movq %rsp, %rsi + call dtrace_invop + ALTENTRY(dtrace_invop_callsite) + addq $8, %rsp + cmpl $DTRACE_INVOP_PUSHL_EBP, %eax + je bp_push + cmpl $DTRACE_INVOP_LEAVE, %eax + je bp_leave + cmpl $DTRACE_INVOP_NOP, %eax + je bp_nop + cmpl $DTRACE_INVOP_RET, %eax + je bp_ret + + /* When all else fails handle the trap in the usual way. */ + jmpq *dtrace_invop_calltrap_addr + +bp_push: + /* + * We must emulate a "pushq %rbp". To do this, we pull the stack + * down 8 bytes, and then store the base pointer. + */ + INTR_POP + subq $16, %rsp /* make room for %rbp */ + pushq %rax /* push temp */ + movq 24(%rsp), %rax /* load calling RIP */ + movq %rax, 8(%rsp) /* store calling RIP */ + movq 32(%rsp), %rax /* load calling CS */ + movq %rax, 16(%rsp) /* store calling CS */ + movq 40(%rsp), %rax /* load calling RFLAGS */ + movq %rax, 24(%rsp) /* store calling RFLAGS */ + movq 48(%rsp), %rax /* load calling RSP */ + subq $8, %rax /* make room for %rbp */ + movq %rax, 32(%rsp) /* store calling RSP */ + movq 56(%rsp), %rax /* load calling SS */ + movq %rax, 40(%rsp) /* store calling SS */ + movq 32(%rsp), %rax /* reload calling RSP */ + movq %rbp, (%rax) /* store %rbp there */ + popq %rax /* pop off temp */ + iretq /* return from interrupt */ + /*NOTREACHED*/ + +bp_leave: + /* + * We must emulate a "leave", which is the same as a "movq %rbp, %rsp" + * followed by a "popq %rbp". This is quite a bit simpler on amd64 + * than it is on i386 -- we can exploit the fact that the %rsp is + * explicitly saved to effect the pop without having to reshuffle + * the other data pushed for the trap. + */ + INTR_POP + pushq %rax /* push temp */ + movq 8(%rsp), %rax /* load calling RIP */ + movq %rax, 8(%rsp) /* store calling RIP */ + movq (%rbp), %rax /* get new %rbp */ + addq $8, %rbp /* adjust new %rsp */ + movq %rbp, 32(%rsp) /* store new %rsp */ + movq %rax, %rbp /* set new %rbp */ + popq %rax /* pop off temp */ + iretq /* return from interrupt */ + /*NOTREACHED*/ + +bp_nop: + /* We must emulate a "nop". */ + INTR_POP + iretq + /*NOTREACHED*/ + +bp_ret: + INTR_POP + pushq %rax /* push temp */ + movq 32(%rsp), %rax /* load %rsp */ + movq (%rax), %rax /* load calling RIP */ + movq %rax, 8(%rsp) /* store calling RIP */ + addq $8, 32(%rsp) /* adjust new %rsp */ + popq %rax /* pop off temp */ + iretq /* return from interrupt */ + /*NOTREACHED*/ + + END(dtrace_invop_start) + +/* +void dtrace_invop_init(void) +*/ + ENTRY(dtrace_invop_init) + movq $dtrace_invop_start, dtrace_invop_jump_addr(%rip) + ret + END(dtrace_invop_init) + +/* +void dtrace_invop_uninit(void) +*/ + ENTRY(dtrace_invop_uninit) + movq $0, dtrace_invop_jump_addr(%rip) + ret + END(dtrace_invop_uninit) + +/* +greg_t dtrace_getfp(void) +*/ + ENTRY(dtrace_getfp) + movq %rbp, %rax + ret + END(dtrace_getfp) + +/* +uint32_t +dtrace_cas32(uint32_t *target, uint32_t cmp, uint32_t new) +*/ + ENTRY(dtrace_cas32) + movl %esi, %eax + lock + cmpxchgl %edx, (%rdi) + ret + END(dtrace_cas32) + +/* +void * +dtrace_casptr(void *target, void *cmp, void *new) +*/ + ENTRY(dtrace_casptr) + movq %rsi, %rax + lock + cmpxchgq %rdx, (%rdi) + ret + END(dtrace_casptr) + +/* +uintptr_t +dtrace_caller(int aframes) +*/ + ENTRY(dtrace_caller) + movq $-1, %rax + ret + END(dtrace_caller) + +/* +void +dtrace_copy(uintptr_t src, uintptr_t dest, size_t size) +*/ + ENTRY(dtrace_copy) + pushq %rbp + movq %rsp, %rbp + + xchgq %rdi, %rsi /* make %rsi source, %rdi dest */ + movq %rdx, %rcx /* load count */ + repz /* repeat for count ... */ + smovb /* move from %ds:rsi to %ed:rdi */ + leave + ret + END(dtrace_copy) + +/* +void +dtrace_copystr(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +*/ + ENTRY(dtrace_copystr) + pushq %rbp + movq %rsp, %rbp + +0: + movb (%rdi), %al /* load from source */ + movb %al, (%rsi) /* store to destination */ + addq $1, %rdi /* increment source pointer */ + addq $1, %rsi /* increment destination pointer */ + subq $1, %rdx /* decrement remaining count */ + cmpb $0, %al + je 2f + testq $0xfff, %rdx /* test if count is 4k-aligned */ + jnz 1f /* if not, continue with copying */ + testq $CPU_DTRACE_BADADDR, (%rcx) /* load and test dtrace flags */ + jnz 2f +1: + cmpq $0, %rdx + jne 0b +2: + leave + ret + + END(dtrace_copystr) + +/* +uintptr_t +dtrace_fulword(void *addr) +*/ + ENTRY(dtrace_fulword) + movq (%rdi), %rax + ret + END(dtrace_fulword) + +/* +uint8_t +dtrace_fuword8_nocheck(void *addr) +*/ + ENTRY(dtrace_fuword8_nocheck) + xorq %rax, %rax + movb (%rdi), %al + ret + END(dtrace_fuword8_nocheck) + +/* +uint16_t +dtrace_fuword16_nocheck(void *addr) +*/ + ENTRY(dtrace_fuword16_nocheck) + xorq %rax, %rax + movw (%rdi), %ax + ret + END(dtrace_fuword16_nocheck) + +/* +uint32_t +dtrace_fuword32_nocheck(void *addr) +*/ + ENTRY(dtrace_fuword32_nocheck) + xorq %rax, %rax + movl (%rdi), %eax + ret + END(dtrace_fuword32_nocheck) + +/* +uint64_t +dtrace_fuword64_nocheck(void *addr) +*/ + ENTRY(dtrace_fuword64_nocheck) + movq (%rdi), %rax + ret + END(dtrace_fuword64_nocheck) + +/* +void +dtrace_probe_error(dtrace_state_t *state, dtrace_epid_t epid, int which, + int fault, int fltoffs, uintptr_t illval) +*/ + ENTRY(dtrace_probe_error) + pushq %rbp + movq %rsp, %rbp + subq $0x8, %rsp + movq %r9, (%rsp) + movq %r8, %r9 + movq %rcx, %r8 + movq %rdx, %rcx + movq %rsi, %rdx + movq %rdi, %rsi + movl dtrace_probeid_error(%rip), %edi + call dtrace_probe + addq $0x8, %rsp + leave + ret + END(dtrace_probe_error) + +/* +void +dtrace_membar_producer(void) +*/ + ENTRY(dtrace_membar_producer) + rep; ret /* use 2 byte return instruction when branch target */ + /* AMD Software Optimization Guide - Section 6.2 */ + END(dtrace_membar_producer) + +/* +void +dtrace_membar_consumer(void) +*/ + ENTRY(dtrace_membar_consumer) + rep; ret /* use 2 byte return instruction when branch target */ + /* AMD Software Optimization Guide - Section 6.2 */ + END(dtrace_membar_consumer) + +/* +dtrace_icookie_t +dtrace_interrupt_disable(void) +*/ + ENTRY(dtrace_interrupt_disable) + pushfq + popq %rax + cli + ret + END(dtrace_interrupt_disable) + +/* +void +dtrace_interrupt_enable(dtrace_icookie_t cookie) +*/ + ENTRY(dtrace_interrupt_enable) + pushq %rdi + popfq + ret + END(dtrace_interrupt_enable) + +/* + * The panic() and cmn_err() functions invoke vpanic() as a common entry point + * into the panic code implemented in panicsys(). vpanic() is responsible + * for passing through the format string and arguments, and constructing a + * regs structure on the stack into which it saves the current register + * values. If we are not dying due to a fatal trap, these registers will + * then be preserved in panicbuf as the current processor state. Before + * invoking panicsys(), vpanic() activates the first panic trigger (see + * common/os/panic.c) and switches to the panic_stack if successful. Note that + * DTrace takes a slightly different panic path if it must panic from probe + * context. Instead of calling panic, it calls into dtrace_vpanic(), which + * sets up the initial stack as vpanic does, calls dtrace_panic_trigger(), and + * branches back into vpanic(). + */ + +/* +void +vpanic(const char *format, va_list alist) +*/ + ENTRY(vpanic) /* Initial stack layout: */ + + pushq %rbp /* | %rip | 0x60 */ + movq %rsp, %rbp /* | %rbp | 0x58 */ + pushfq /* | rfl | 0x50 */ + pushq %r11 /* | %r11 | 0x48 */ + pushq %r10 /* | %r10 | 0x40 */ + pushq %rbx /* | %rbx | 0x38 */ + pushq %rax /* | %rax | 0x30 */ + pushq %r9 /* | %r9 | 0x28 */ + pushq %r8 /* | %r8 | 0x20 */ + pushq %rcx /* | %rcx | 0x18 */ + pushq %rdx /* | %rdx | 0x10 */ + pushq %rsi /* | %rsi | 0x8 alist */ + pushq %rdi /* | %rdi | 0x0 format */ + + movq %rsp, %rbx /* %rbx = current %rsp */ + + leaq panic_quiesce(%rip), %rdi /* %rdi = &panic_quiesce */ + call panic_trigger /* %eax = panic_trigger() */ + +vpanic_common: + /* + * The panic_trigger result is in %eax from the call above, and + * dtrace_panic places it in %eax before branching here. + * The rdmsr instructions that follow below will clobber %eax so + * we stash the panic_trigger result in %r11d. + */ + movl %eax, %r11d + cmpl $0, %r11d + je 0f + + /* + * If panic_trigger() was successful, we are the first to initiate a + * panic: we now switch to the reserved panic_stack before continuing. + */ + leaq panic_stack(%rip), %rsp + addq $PANICSTKSIZE, %rsp +0: subq $REGSIZE, %rsp + /* + * Now that we've got everything set up, store the register values as + * they were when we entered vpanic() to the designated location in + * the regs structure we allocated on the stack. + */ +#ifdef notyet + movq 0x0(%rbx), %rcx + movq %rcx, REGOFF_RDI(%rsp) + movq 0x8(%rbx), %rcx + movq %rcx, REGOFF_RSI(%rsp) + movq 0x10(%rbx), %rcx + movq %rcx, REGOFF_RDX(%rsp) + movq 0x18(%rbx), %rcx + movq %rcx, REGOFF_RCX(%rsp) + movq 0x20(%rbx), %rcx + + movq %rcx, REGOFF_R8(%rsp) + movq 0x28(%rbx), %rcx + movq %rcx, REGOFF_R9(%rsp) + movq 0x30(%rbx), %rcx + movq %rcx, REGOFF_RAX(%rsp) + movq 0x38(%rbx), %rcx + movq %rcx, REGOFF_RBX(%rsp) + movq 0x58(%rbx), %rcx + + movq %rcx, REGOFF_RBP(%rsp) + movq 0x40(%rbx), %rcx + movq %rcx, REGOFF_R10(%rsp) + movq 0x48(%rbx), %rcx + movq %rcx, REGOFF_R11(%rsp) + movq %r12, REGOFF_R12(%rsp) + + movq %r13, REGOFF_R13(%rsp) + movq %r14, REGOFF_R14(%rsp) + movq %r15, REGOFF_R15(%rsp) + + xorl %ecx, %ecx + movw %ds, %cx + movq %rcx, REGOFF_DS(%rsp) + movw %es, %cx + movq %rcx, REGOFF_ES(%rsp) + movw %fs, %cx + movq %rcx, REGOFF_FS(%rsp) + movw %gs, %cx + movq %rcx, REGOFF_GS(%rsp) + + movq $0, REGOFF_TRAPNO(%rsp) + + movq $0, REGOFF_ERR(%rsp) + leaq vpanic(%rip), %rcx + movq %rcx, REGOFF_RIP(%rsp) + movw %cs, %cx + movzwq %cx, %rcx + movq %rcx, REGOFF_CS(%rsp) + movq 0x50(%rbx), %rcx + movq %rcx, REGOFF_RFL(%rsp) + movq %rbx, %rcx + addq $0x60, %rcx + movq %rcx, REGOFF_RSP(%rsp) + movw %ss, %cx + movzwq %cx, %rcx + movq %rcx, REGOFF_SS(%rsp) + + /* + * panicsys(format, alist, rp, on_panic_stack) + */ + movq REGOFF_RDI(%rsp), %rdi /* format */ + movq REGOFF_RSI(%rsp), %rsi /* alist */ + movq %rsp, %rdx /* struct regs */ + movl %r11d, %ecx /* on_panic_stack */ + call panicsys + addq $REGSIZE, %rsp +#endif + popq %rdi + popq %rsi + popq %rdx + popq %rcx + popq %r8 + popq %r9 + popq %rax + popq %rbx + popq %r10 + popq %r11 + popfq + leave + ret + END(vpanic) + +/* +void +dtrace_vpanic(const char *format, va_list alist) +*/ + ENTRY(dtrace_vpanic) /* Initial stack layout: */ + + pushq %rbp /* | %rip | 0x60 */ + movq %rsp, %rbp /* | %rbp | 0x58 */ + pushfq /* | rfl | 0x50 */ + pushq %r11 /* | %r11 | 0x48 */ + pushq %r10 /* | %r10 | 0x40 */ + pushq %rbx /* | %rbx | 0x38 */ + pushq %rax /* | %rax | 0x30 */ + pushq %r9 /* | %r9 | 0x28 */ + pushq %r8 /* | %r8 | 0x20 */ + pushq %rcx /* | %rcx | 0x18 */ + pushq %rdx /* | %rdx | 0x10 */ + pushq %rsi /* | %rsi | 0x8 alist */ + pushq %rdi /* | %rdi | 0x0 format */ + + movq %rsp, %rbx /* %rbx = current %rsp */ + + leaq panic_quiesce(%rip), %rdi /* %rdi = &panic_quiesce */ + call dtrace_panic_trigger /* %eax = dtrace_panic_trigger() */ + jmp vpanic_common + + END(dtrace_vpanic) + +/* +int +panic_trigger(int *tp) +*/ + ENTRY(panic_trigger) + xorl %eax, %eax + movl $0xdefacedd, %edx + lock + xchgl %edx, (%rdi) + cmpl $0, %edx + je 0f + movl $0, %eax + ret +0: movl $1, %eax + ret + END(panic_trigger) + +/* +int +dtrace_panic_trigger(int *tp) +*/ + ENTRY(dtrace_panic_trigger) + xorl %eax, %eax + movl $0xdefacedd, %edx + lock + xchgl %edx, (%rdi) + cmpl $0, %edx + je 0f + movl $0, %eax + ret +0: movl $1, %eax + ret + END(dtrace_panic_trigger) diff --git a/sys/cddl/dev/dtrace/amd64/dtrace_isa.c b/sys/cddl/dev/dtrace/amd64/dtrace_isa.c new file mode 100644 index 000000000000..8cd2a0ff649d --- /dev/null +++ b/sys/cddl/dev/dtrace/amd64/dtrace_isa.c @@ -0,0 +1,612 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ +/* + * Copyright 2005 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ +#include <sys/cdefs.h> + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/kernel.h> +#include <sys/stack.h> +#include <sys/pcpu.h> + +#include <machine/frame.h> +#include <machine/md_var.h> +#include <machine/reg.h> +#include <machine/stack.h> + +#include <vm/vm.h> +#include <vm/vm_param.h> +#include <vm/pmap.h> + +extern uintptr_t kernbase; +uintptr_t kernelbase = (uintptr_t) &kernbase; + +uint8_t dtrace_fuword8_nocheck(void *); +uint16_t dtrace_fuword16_nocheck(void *); +uint32_t dtrace_fuword32_nocheck(void *); +uint64_t dtrace_fuword64_nocheck(void *); + +void +dtrace_getpcstack(pc_t *pcstack, int pcstack_limit, int aframes, + uint32_t *intrpc) +{ + int depth = 0; + register_t rbp; + struct amd64_frame *frame; + vm_offset_t callpc; + pc_t caller = (pc_t) solaris_cpu[curcpu].cpu_dtrace_caller; + + if (intrpc != 0) + pcstack[depth++] = (pc_t) intrpc; + + aframes++; + + __asm __volatile("movq %%rbp,%0" : "=r" (rbp)); + + frame = (struct amd64_frame *)rbp; + while (depth < pcstack_limit) { + if (!INKERNEL((long) frame)) + break; + + callpc = frame->f_retaddr; + + if (!INKERNEL(callpc)) + break; + + if (aframes > 0) { + aframes--; + if ((aframes == 0) && (caller != 0)) { + pcstack[depth++] = caller; + } + } + else { + pcstack[depth++] = callpc; + } + + if (frame->f_frame <= frame || + (vm_offset_t)frame->f_frame >= + (vm_offset_t)rbp + KSTACK_PAGES * PAGE_SIZE) + break; + frame = frame->f_frame; + } + + for (; depth < pcstack_limit; depth++) { + pcstack[depth] = 0; + } +} + +static int +dtrace_getustack_common(uint64_t *pcstack, int pcstack_limit, uintptr_t pc, + uintptr_t sp) +{ + volatile uint16_t *flags = + (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; + struct amd64_frame *frame; + int ret = 0; + + ASSERT(pcstack == NULL || pcstack_limit > 0); + + while (pc != 0 && sp != 0) { + ret++; + if (pcstack != NULL) { + *pcstack++ = (uint64_t)pc; + pcstack_limit--; + if (pcstack_limit <= 0) + break; + } + + frame = (struct amd64_frame *) sp; + + pc = dtrace_fulword(&frame->f_retaddr); + sp = dtrace_fulword(&frame->f_frame); + + /* + * This is totally bogus: if we faulted, we're going to clear + * the fault and break. This is to deal with the apparently + * broken Java stacks on x86. + */ + if (*flags & CPU_DTRACE_FAULT) { + *flags &= ~CPU_DTRACE_FAULT; + break; + } + } + + return (ret); +} + +void +dtrace_getupcstack(uint64_t *pcstack, int pcstack_limit) +{ + proc_t *p = curproc; + struct trapframe *tf; + uintptr_t pc, sp; + volatile uint16_t *flags = + (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; + int n; + + if (*flags & CPU_DTRACE_FAULT) + return; + + if (pcstack_limit <= 0) + return; + + /* + * If there's no user context we still need to zero the stack. + */ + if (p == NULL || (tf = curthread->td_frame) == NULL) + goto zero; + + *pcstack++ = (uint64_t)p->p_pid; + pcstack_limit--; + + if (pcstack_limit <= 0) + return; + + pc = tf->tf_rip; + sp = tf->tf_rsp; + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) { + *pcstack++ = (uint64_t)pc; + pcstack_limit--; + if (pcstack_limit <= 0) + return; + + pc = dtrace_fulword((void *) sp); + } + + n = dtrace_getustack_common(pcstack, pcstack_limit, pc, sp); + ASSERT(n >= 0); + ASSERT(n <= pcstack_limit); + + pcstack += n; + pcstack_limit -= n; + +zero: + while (pcstack_limit-- > 0) + *pcstack++ = 0; +} + +int +dtrace_getustackdepth(void) +{ + proc_t *p = curproc; + struct trapframe *tf; + uintptr_t pc, sp; + int n = 0; + + if (p == NULL || (tf = curthread->td_frame) == NULL) + return (0); + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT)) + return (-1); + + pc = tf->tf_rip; + sp = tf->tf_rsp; + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) { + n++; + + pc = dtrace_fulword((void *) sp); + } + + n += dtrace_getustack_common(NULL, 0, pc, sp); + + return (n); +} + +#ifdef notyet +void +dtrace_getufpstack(uint64_t *pcstack, uint64_t *fpstack, int pcstack_limit) +{ + klwp_t *lwp = ttolwp(curthread); + proc_t *p = curproc; + struct regs *rp; + uintptr_t pc, sp, oldcontext; + volatile uint16_t *flags = + (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; + size_t s1, s2; + + if (*flags & CPU_DTRACE_FAULT) + return; + + if (pcstack_limit <= 0) + return; + + /* + * If there's no user context we still need to zero the stack. + */ + if (lwp == NULL || p == NULL || (rp = lwp->lwp_regs) == NULL) + goto zero; + + *pcstack++ = (uint64_t)p->p_pid; + pcstack_limit--; + + if (pcstack_limit <= 0) + return; + + pc = rp->r_pc; + sp = rp->r_fp; + oldcontext = lwp->lwp_oldcontext; + + s1 = sizeof (struct xframe) + 2 * sizeof (long); + s2 = s1 + sizeof (siginfo_t); + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) { + *pcstack++ = (uint64_t)pc; + *fpstack++ = 0; + pcstack_limit--; + if (pcstack_limit <= 0) + return; + + if (p->p_model == DATAMODEL_NATIVE) + pc = dtrace_fulword((void *)rp->r_sp); + else + pc = dtrace_fuword32((void *)rp->r_sp); + } + + while (pc != 0 && sp != 0) { + *pcstack++ = (uint64_t)pc; + *fpstack++ = sp; + pcstack_limit--; + if (pcstack_limit <= 0) + break; + + if (oldcontext == sp + s1 || oldcontext == sp + s2) { + ucontext_t *ucp = (ucontext_t *)oldcontext; + greg_t *gregs = ucp->uc_mcontext.gregs; + + sp = dtrace_fulword(&gregs[REG_FP]); + pc = dtrace_fulword(&gregs[REG_PC]); + + oldcontext = dtrace_fulword(&ucp->uc_link); + } else { + struct xframe *fr = (struct xframe *)sp; + + pc = dtrace_fulword(&fr->fr_savpc); + sp = dtrace_fulword(&fr->fr_savfp); + } + + /* + * This is totally bogus: if we faulted, we're going to clear + * the fault and break. This is to deal with the apparently + * broken Java stacks on x86. + */ + if (*flags & CPU_DTRACE_FAULT) { + *flags &= ~CPU_DTRACE_FAULT; + break; + } + } + +zero: + while (pcstack_limit-- > 0) + *pcstack++ = NULL; +} +#endif + +/*ARGSUSED*/ +uint64_t +dtrace_getarg(int arg, int aframes) +{ + uintptr_t val; + struct amd64_frame *fp = (struct amd64_frame *)dtrace_getfp(); + uintptr_t *stack; + int i; + + /* + * A total of 6 arguments are passed via registers; any argument with + * index of 5 or lower is therefore in a register. + */ + int inreg = 5; + + for (i = 1; i <= aframes; i++) { + fp = fp->f_frame; + + if (fp->f_retaddr == (long)dtrace_invop_callsite) { + /* + * In the case of amd64, we will use the pointer to the + * regs structure that was pushed when we took the + * trap. To get this structure, we must increment + * beyond the frame structure, and then again beyond + * the calling RIP stored in dtrace_invop(). If the + * argument that we're seeking is passed on the stack, + * we'll pull the true stack pointer out of the saved + * registers and decrement our argument by the number + * of arguments passed in registers; if the argument + * we're seeking is passed in regsiters, we can just + * load it directly. + */ + struct reg *rp = (struct reg *)((uintptr_t)&fp[1] + + sizeof (uintptr_t)); + + if (arg <= inreg) { + stack = (uintptr_t *)&rp->r_rdi; + } else { + stack = (uintptr_t *)(rp->r_rsp); + arg -= inreg; + } + goto load; + } + + } + + /* + * We know that we did not come through a trap to get into + * dtrace_probe() -- the provider simply called dtrace_probe() + * directly. As this is the case, we need to shift the argument + * that we're looking for: the probe ID is the first argument to + * dtrace_probe(), so the argument n will actually be found where + * one would expect to find argument (n + 1). + */ + arg++; + + if (arg <= inreg) { + /* + * This shouldn't happen. If the argument is passed in a + * register then it should have been, well, passed in a + * register... + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + + arg -= (inreg + 1); + stack = (uintptr_t *)&fp[1]; + +load: + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + val = stack[arg]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + return (val); + return (0); +} + +int +dtrace_getstackdepth(int aframes) +{ + int depth = 0; + struct amd64_frame *frame; + vm_offset_t rbp; + + aframes++; + rbp = dtrace_getfp(); + frame = (struct amd64_frame *)rbp; + depth++; + for(;;) { + if (!INKERNEL((long) frame)) + break; + if (!INKERNEL((long) frame->f_frame)) + break; + depth++; + if (frame->f_frame <= frame || + (vm_offset_t)frame->f_frame >= + (vm_offset_t)rbp + KSTACK_PAGES * PAGE_SIZE) + break; + frame = frame->f_frame; + } + if (depth < aframes) + return 0; + else + return depth - aframes; +} + +#ifdef notyet +ulong_t +dtrace_getreg(struct regs *rp, uint_t reg) +{ +#if defined(__amd64) + int regmap[] = { + REG_GS, /* GS */ + REG_FS, /* FS */ + REG_ES, /* ES */ + REG_DS, /* DS */ + REG_RDI, /* EDI */ + REG_RSI, /* ESI */ + REG_RBP, /* EBP */ + REG_RSP, /* ESP */ + REG_RBX, /* EBX */ + REG_RDX, /* EDX */ + REG_RCX, /* ECX */ + REG_RAX, /* EAX */ + REG_TRAPNO, /* TRAPNO */ + REG_ERR, /* ERR */ + REG_RIP, /* EIP */ + REG_CS, /* CS */ + REG_RFL, /* EFL */ + REG_RSP, /* UESP */ + REG_SS /* SS */ + }; + + if (reg <= SS) { + if (reg >= sizeof (regmap) / sizeof (int)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + + reg = regmap[reg]; + } else { + reg -= SS + 1; + } + + switch (reg) { + case REG_RDI: + return (rp->r_rdi); + case REG_RSI: + return (rp->r_rsi); + case REG_RDX: + return (rp->r_rdx); + case REG_RCX: + return (rp->r_rcx); + case REG_R8: + return (rp->r_r8); + case REG_R9: + return (rp->r_r9); + case REG_RAX: + return (rp->r_rax); + case REG_RBX: + return (rp->r_rbx); + case REG_RBP: + return (rp->r_rbp); + case REG_R10: + return (rp->r_r10); + case REG_R11: + return (rp->r_r11); + case REG_R12: + return (rp->r_r12); + case REG_R13: + return (rp->r_r13); + case REG_R14: + return (rp->r_r14); + case REG_R15: + return (rp->r_r15); + case REG_DS: + return (rp->r_ds); + case REG_ES: + return (rp->r_es); + case REG_FS: + return (rp->r_fs); + case REG_GS: + return (rp->r_gs); + case REG_TRAPNO: + return (rp->r_trapno); + case REG_ERR: + return (rp->r_err); + case REG_RIP: + return (rp->r_rip); + case REG_CS: + return (rp->r_cs); + case REG_SS: + return (rp->r_ss); + case REG_RFL: + return (rp->r_rfl); + case REG_RSP: + return (rp->r_rsp); + default: + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + +#else + if (reg > SS) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + + return ((&rp->r_gs)[reg]); +#endif +} +#endif + +static int +dtrace_copycheck(uintptr_t uaddr, uintptr_t kaddr, size_t size) +{ + ASSERT(kaddr >= kernelbase && kaddr + size >= kaddr); + + if (uaddr + size >= kernelbase || uaddr + size < uaddr) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = uaddr; + return (0); + } + + return (1); +} + +void +dtrace_copyin(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copy(uaddr, kaddr, size); +} + +void +dtrace_copyout(uintptr_t kaddr, uintptr_t uaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copy(kaddr, uaddr, size); +} + +void +dtrace_copyinstr(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copystr(uaddr, kaddr, size, flags); +} + +void +dtrace_copyoutstr(uintptr_t kaddr, uintptr_t uaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copystr(kaddr, uaddr, size, flags); +} + +uint8_t +dtrace_fuword8(void *uaddr) +{ + if ((uintptr_t)uaddr >= kernelbase) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword8_nocheck(uaddr)); +} + +uint16_t +dtrace_fuword16(void *uaddr) +{ + if ((uintptr_t)uaddr >= kernelbase) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword16_nocheck(uaddr)); +} + +uint32_t +dtrace_fuword32(void *uaddr) +{ + if ((uintptr_t)uaddr >= kernelbase) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword32_nocheck(uaddr)); +} + +uint64_t +dtrace_fuword64(void *uaddr) +{ + if ((uintptr_t)uaddr >= kernelbase) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword64_nocheck(uaddr)); +} diff --git a/sys/cddl/dev/dtrace/amd64/dtrace_subr.c b/sys/cddl/dev/dtrace/amd64/dtrace_subr.c new file mode 100644 index 000000000000..4bf0f039ec9b --- /dev/null +++ b/sys/cddl/dev/dtrace/amd64/dtrace_subr.c @@ -0,0 +1,507 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + */ +/* + * Copyright 2005 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/types.h> +#include <sys/kernel.h> +#include <sys/malloc.h> +#include <sys/kmem.h> +#include <sys/smp.h> +#include <sys/dtrace_impl.h> +#include <sys/dtrace_bsd.h> +#include <machine/clock.h> +#include <machine/frame.h> +#include <vm/pmap.h> + +extern uintptr_t kernelbase; +extern uintptr_t dtrace_in_probe_addr; +extern int dtrace_in_probe; + +int dtrace_invop(uintptr_t, uintptr_t *, uintptr_t); + +typedef struct dtrace_invop_hdlr { + int (*dtih_func)(uintptr_t, uintptr_t *, uintptr_t); + struct dtrace_invop_hdlr *dtih_next; +} dtrace_invop_hdlr_t; + +dtrace_invop_hdlr_t *dtrace_invop_hdlr; + +int +dtrace_invop(uintptr_t addr, uintptr_t *stack, uintptr_t eax) +{ + dtrace_invop_hdlr_t *hdlr; + int rval; + + for (hdlr = dtrace_invop_hdlr; hdlr != NULL; hdlr = hdlr->dtih_next) + if ((rval = hdlr->dtih_func(addr, stack, eax)) != 0) + return (rval); + + return (0); +} + +void +dtrace_invop_add(int (*func)(uintptr_t, uintptr_t *, uintptr_t)) +{ + dtrace_invop_hdlr_t *hdlr; + + hdlr = kmem_alloc(sizeof (dtrace_invop_hdlr_t), KM_SLEEP); + hdlr->dtih_func = func; + hdlr->dtih_next = dtrace_invop_hdlr; + dtrace_invop_hdlr = hdlr; +} + +void +dtrace_invop_remove(int (*func)(uintptr_t, uintptr_t *, uintptr_t)) +{ + dtrace_invop_hdlr_t *hdlr = dtrace_invop_hdlr, *prev = NULL; + + for (;;) { + if (hdlr == NULL) + panic("attempt to remove non-existent invop handler"); + + if (hdlr->dtih_func == func) + break; + + prev = hdlr; + hdlr = hdlr->dtih_next; + } + + if (prev == NULL) { + ASSERT(dtrace_invop_hdlr == hdlr); + dtrace_invop_hdlr = hdlr->dtih_next; + } else { + ASSERT(dtrace_invop_hdlr != hdlr); + prev->dtih_next = hdlr->dtih_next; + } + + kmem_free(hdlr, 0); +} + +/*ARGSUSED*/ +void +dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit)) +{ + (*func)(0, (uintptr_t) addr_PTmap); +} + +void +dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg) +{ + cpumask_t cpus; + + critical_enter(); + + if (cpu == DTRACE_CPUALL) + cpus = all_cpus; + else + cpus = (cpumask_t) (1 << cpu); + + /* If the current CPU is in the set, call the function directly: */ + if ((cpus & (1 << curcpu)) != 0) { + (*func)(arg); + + /* Mask the current CPU from the set */ + cpus &= ~(1 << curcpu); + } + + /* If there are any CPUs in the set, cross-call to those CPUs */ + if (cpus != 0) + smp_rendezvous_cpus(cpus, NULL, func, smp_no_rendevous_barrier, arg); + + critical_exit(); +} + +static void +dtrace_sync_func(void) +{ +} + +void +dtrace_sync(void) +{ + dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL); +} + +#ifdef notyet +int (*dtrace_fasttrap_probe_ptr)(struct regs *); +int (*dtrace_pid_probe_ptr)(struct regs *); +int (*dtrace_return_probe_ptr)(struct regs *); + +void +dtrace_user_probe(struct regs *rp, caddr_t addr, processorid_t cpuid) +{ + krwlock_t *rwp; + proc_t *p = curproc; + extern void trap(struct regs *, caddr_t, processorid_t); + + if (USERMODE(rp->r_cs) || (rp->r_ps & PS_VM)) { + if (curthread->t_cred != p->p_cred) { + cred_t *oldcred = curthread->t_cred; + /* + * DTrace accesses t_cred in probe context. t_cred + * must always be either NULL, or point to a valid, + * allocated cred structure. + */ + curthread->t_cred = crgetcred(); + crfree(oldcred); + } + } + + if (rp->r_trapno == T_DTRACE_RET) { + uint8_t step = curthread->t_dtrace_step; + uint8_t ret = curthread->t_dtrace_ret; + uintptr_t npc = curthread->t_dtrace_npc; + + if (curthread->t_dtrace_ast) { + aston(curthread); + curthread->t_sig_check = 1; + } + + /* + * Clear all user tracing flags. + */ + curthread->t_dtrace_ft = 0; + + /* + * If we weren't expecting to take a return probe trap, kill + * the process as though it had just executed an unassigned + * trap instruction. + */ + if (step == 0) { + tsignal(curthread, SIGILL); + return; + } + + /* + * If we hit this trap unrelated to a return probe, we're + * just here to reset the AST flag since we deferred a signal + * until after we logically single-stepped the instruction we + * copied out. + */ + if (ret == 0) { + rp->r_pc = npc; + return; + } + + /* + * We need to wait until after we've called the + * dtrace_return_probe_ptr function pointer to set %pc. + */ + rwp = &CPU->cpu_ft_lock; + rw_enter(rwp, RW_READER); + if (dtrace_return_probe_ptr != NULL) + (void) (*dtrace_return_probe_ptr)(rp); + rw_exit(rwp); + rp->r_pc = npc; + + } else if (rp->r_trapno == T_DTRACE_PROBE) { + rwp = &CPU->cpu_ft_lock; + rw_enter(rwp, RW_READER); + if (dtrace_fasttrap_probe_ptr != NULL) + (void) (*dtrace_fasttrap_probe_ptr)(rp); + rw_exit(rwp); + + } else if (rp->r_trapno == T_BPTFLT) { + uint8_t instr; + rwp = &CPU->cpu_ft_lock; + + /* + * The DTrace fasttrap provider uses the breakpoint trap + * (int 3). We let DTrace take the first crack at handling + * this trap; if it's not a probe that DTrace knowns about, + * we call into the trap() routine to handle it like a + * breakpoint placed by a conventional debugger. + */ + rw_enter(rwp, RW_READER); + if (dtrace_pid_probe_ptr != NULL && + (*dtrace_pid_probe_ptr)(rp) == 0) { + rw_exit(rwp); + return; + } + rw_exit(rwp); + + /* + * If the instruction that caused the breakpoint trap doesn't + * look like an int 3 anymore, it may be that this tracepoint + * was removed just after the user thread executed it. In + * that case, return to user land to retry the instuction. + */ + if (fuword8((void *)(rp->r_pc - 1), &instr) == 0 && + instr != FASTTRAP_INSTR) { + rp->r_pc--; + return; + } + + trap(rp, addr, cpuid); + + } else { + trap(rp, addr, cpuid); + } +} + +void +dtrace_safe_synchronous_signal(void) +{ + kthread_t *t = curthread; + struct regs *rp = lwptoregs(ttolwp(t)); + size_t isz = t->t_dtrace_npc - t->t_dtrace_pc; + + ASSERT(t->t_dtrace_on); + + /* + * If we're not in the range of scratch addresses, we're not actually + * tracing user instructions so turn off the flags. If the instruction + * we copied out caused a synchonous trap, reset the pc back to its + * original value and turn off the flags. + */ + if (rp->r_pc < t->t_dtrace_scrpc || + rp->r_pc > t->t_dtrace_astpc + isz) { + t->t_dtrace_ft = 0; + } else if (rp->r_pc == t->t_dtrace_scrpc || + rp->r_pc == t->t_dtrace_astpc) { + rp->r_pc = t->t_dtrace_pc; + t->t_dtrace_ft = 0; + } +} + +int +dtrace_safe_defer_signal(void) +{ + kthread_t *t = curthread; + struct regs *rp = lwptoregs(ttolwp(t)); + size_t isz = t->t_dtrace_npc - t->t_dtrace_pc; + + ASSERT(t->t_dtrace_on); + + /* + * If we're not in the range of scratch addresses, we're not actually + * tracing user instructions so turn off the flags. + */ + if (rp->r_pc < t->t_dtrace_scrpc || + rp->r_pc > t->t_dtrace_astpc + isz) { + t->t_dtrace_ft = 0; + return (0); + } + + /* + * If we've executed the original instruction, but haven't performed + * the jmp back to t->t_dtrace_npc or the clean up of any registers + * used to emulate %rip-relative instructions in 64-bit mode, do that + * here and take the signal right away. We detect this condition by + * seeing if the program counter is the range [scrpc + isz, astpc). + */ + if (t->t_dtrace_astpc - rp->r_pc < + t->t_dtrace_astpc - t->t_dtrace_scrpc - isz) { +#ifdef __amd64 + /* + * If there is a scratch register and we're on the + * instruction immediately after the modified instruction, + * restore the value of that scratch register. + */ + if (t->t_dtrace_reg != 0 && + rp->r_pc == t->t_dtrace_scrpc + isz) { + switch (t->t_dtrace_reg) { + case REG_RAX: + rp->r_rax = t->t_dtrace_regv; + break; + case REG_RCX: + rp->r_rcx = t->t_dtrace_regv; + break; + case REG_R8: + rp->r_r8 = t->t_dtrace_regv; + break; + case REG_R9: + rp->r_r9 = t->t_dtrace_regv; + break; + } + } +#endif + rp->r_pc = t->t_dtrace_npc; + t->t_dtrace_ft = 0; + return (0); + } + + /* + * Otherwise, make sure we'll return to the kernel after executing + * the copied out instruction and defer the signal. + */ + if (!t->t_dtrace_step) { + ASSERT(rp->r_pc < t->t_dtrace_astpc); + rp->r_pc += t->t_dtrace_astpc - t->t_dtrace_scrpc; + t->t_dtrace_step = 1; + } + + t->t_dtrace_ast = 1; + + return (1); +} +#endif + +static int64_t tgt_cpu_tsc; +static int64_t hst_cpu_tsc; +static int64_t tsc_skew[MAXCPU]; + +static void +dtrace_gethrtime_init_sync(void *arg) +{ +#ifdef CHECK_SYNC + /* + * Delay this function from returning on one + * of the CPUs to check that the synchronisation + * works. + */ + uintptr_t cpu = (uintptr_t) arg; + + if (cpu == curcpu) { + int i; + for (i = 0; i < 1000000000; i++) + tgt_cpu_tsc = rdtsc(); + tgt_cpu_tsc = 0; + } +#endif +} + +static void +dtrace_gethrtime_init_cpu(void *arg) +{ + uintptr_t cpu = (uintptr_t) arg; + + if (cpu == curcpu) + tgt_cpu_tsc = rdtsc(); + else + hst_cpu_tsc = rdtsc(); +} + +static void +dtrace_gethrtime_init(void *arg) +{ + cpumask_t map; + int i; + struct pcpu *cp; + + /* The current CPU is the reference one. */ + tsc_skew[curcpu] = 0; + + for (i = 0; i <= mp_maxid; i++) { + if (i == curcpu) + continue; + + if ((cp = pcpu_find(i)) == NULL) + continue; + + map = 0; + map |= (1 << curcpu); + map |= (1 << i); + + smp_rendezvous_cpus(map, dtrace_gethrtime_init_sync, + dtrace_gethrtime_init_cpu, + smp_no_rendevous_barrier, (void *)(uintptr_t) i); + + tsc_skew[i] = tgt_cpu_tsc - hst_cpu_tsc; + } +} + +SYSINIT(dtrace_gethrtime_init, SI_SUB_SMP, SI_ORDER_ANY, dtrace_gethrtime_init, NULL); + +/* + * DTrace needs a high resolution time function which can + * be called from a probe context and guaranteed not to have + * instrumented with probes itself. + * + * Returns nanoseconds since boot. + */ +uint64_t +dtrace_gethrtime() +{ + return ((rdtsc() + tsc_skew[curcpu]) * (int64_t) 1000000000 / tsc_freq); +} + +uint64_t +dtrace_gethrestime(void) +{ + printf("%s(%d): XXX\n",__func__,__LINE__); + return (0); +} + +/* Function to handle DTrace traps during probes. See amd64/amd64/trap.c */ +int +dtrace_trap(struct trapframe *frame, u_int type) +{ + /* + * A trap can occur while DTrace executes a probe. Before + * executing the probe, DTrace blocks re-scheduling and sets + * a flag in it's per-cpu flags to indicate that it doesn't + * want to fault. On returning from the the probe, the no-fault + * flag is cleared and finally re-scheduling is enabled. + * + * Check if DTrace has enabled 'no-fault' mode: + * + */ + if ((cpu_core[curcpu].cpuc_dtrace_flags & CPU_DTRACE_NOFAULT) != 0) { + /* + * There are only a couple of trap types that are expected. + * All the rest will be handled in the usual way. + */ + switch (type) { + /* Privilieged instruction fault. */ + case T_PRIVINFLT: + break; + /* General protection fault. */ + case T_PROTFLT: + /* Flag an illegal operation. */ + cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + + /* + * Offset the instruction pointer to the instruction + * following the one causing the fault. + */ + frame->tf_rip += dtrace_instr_size((u_char *) frame->tf_rip); + return (1); + /* Page fault. */ + case T_PAGEFLT: + /* Flag a bad address. */ + cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR; + cpu_core[curcpu].cpuc_dtrace_illval = frame->tf_addr; + + /* + * Offset the instruction pointer to the instruction + * following the one causing the fault. + */ + frame->tf_rip += dtrace_instr_size((u_char *) frame->tf_rip); + return (1); + default: + /* Handle all other traps in the usual way. */ + break; + } + } + + /* Handle the trap in the usual way. */ + return (0); +} diff --git a/sys/cddl/dev/dtrace/amd64/instr_size.c b/sys/cddl/dev/dtrace/amd64/instr_size.c new file mode 100644 index 000000000000..418d9f1fb53b --- /dev/null +++ b/sys/cddl/dev/dtrace/amd64/instr_size.c @@ -0,0 +1,132 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ +/* + * Copyright 2005 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +/* Copyright (c) 1988 AT&T */ +/* All Rights Reserved */ + + +#if defined(sun) +#pragma ident "@(#)instr_size.c 1.14 05/07/08 SMI" +#endif + +#include <sys/types.h> +#include <sys/param.h> +#include <sys/proc.h> +#if defined(sun) +#include <sys/cmn_err.h> +#include <sys/archsystm.h> +#include <sys/copyops.h> +#include <vm/seg_enum.h> +#include <sys/privregs.h> +#else +typedef u_int model_t; +#define DATAMODEL_NATIVE 0 +int dtrace_instr_size(uchar_t *); +#endif + +#include <dis_tables.h> + +/* + * This subsystem (with the minor exception of the instr_size() function) is + * is called from DTrace probe context. This imposes several requirements on + * the implementation: + * + * 1. External subsystems and functions may not be referenced. The one current + * exception is for cmn_err, but only to signal the detection of table + * errors. Assuming the tables are correct, no combination of input is to + * trigger a cmn_err call. + * + * 2. These functions can't be allowed to be traced. To prevent this, + * all functions in the probe path (everything except instr_size()) must + * have names that begin with "dtrace_". + */ + +typedef enum dis_isize { + DIS_ISIZE_INSTR, + DIS_ISIZE_OPERAND +} dis_isize_t; + + +/* + * get a byte from instruction stream + */ +static int +dtrace_dis_get_byte(void *p) +{ + int ret; + uchar_t **instr = p; + + ret = **instr; + *instr += 1; + + return (ret); +} + +/* + * Returns either the size of a given instruction, in bytes, or the size of that + * instruction's memory access (if any), depending on the value of `which'. + * If a programming error in the tables is detected, the system will panic to + * ease diagnosis. Invalid instructions will not be flagged. They will appear + * to have an instruction size between 1 and the actual size, and will be + * reported as having no memory impact. + */ +/* ARGSUSED2 */ +static int +dtrace_dis_isize(uchar_t *instr, dis_isize_t which, model_t model, int *rmindex) +{ + int sz; + dis86_t x; + uint_t mode = SIZE64; + +#if defined(sun) + mode = (model == DATAMODEL_LP64) ? SIZE64 : SIZE32; +#endif + + x.d86_data = (void **)&instr; + x.d86_get_byte = dtrace_dis_get_byte; + x.d86_check_func = NULL; + + if (dtrace_disx86(&x, mode) != 0) + return (-1); + + if (which == DIS_ISIZE_INSTR) + sz = x.d86_len; /* length of the instruction */ + else + sz = x.d86_memsize; /* length of memory operand */ + + if (rmindex != NULL) + *rmindex = x.d86_rmindex; + return (sz); +} + +int +dtrace_instr_size(uchar_t *instr) +{ + return (dtrace_dis_isize(instr, DIS_ISIZE_INSTR, DATAMODEL_NATIVE, + NULL)); +} diff --git a/sys/cddl/dev/dtrace/dtrace_anon.c b/sys/cddl/dev/dtrace/dtrace_anon.c new file mode 100644 index 000000000000..b81ec5be3f11 --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_anon.c @@ -0,0 +1,84 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ + +/* + * DTrace Anonymous Enabling Functions + */ +static void +dtrace_anon_init(void *dummy) +{ + dtrace_state_t *state = NULL; + dtrace_enabling_t *enab; + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + dtrace_anon_property(); + + mutex_exit(&cpu_lock); + + /* + * If there are already providers, we must ask them to provide their + * probes, and then match any anonymous enabling against them. Note + * that there should be no other retained enablings at this time: + * the only retained enablings at this time should be the anonymous + * enabling. + */ + if (dtrace_anon.dta_enabling != NULL) { + ASSERT(dtrace_retained == dtrace_anon.dta_enabling); + + dtrace_enabling_provide(NULL); + state = dtrace_anon.dta_state; + + /* + * We couldn't hold cpu_lock across the above call to + * dtrace_enabling_provide(), but we must hold it to actually + * enable the probes. We have to drop all of our locks, pick + * up cpu_lock, and regain our locks before matching the + * retained anonymous enabling. + */ + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + if ((enab = dtrace_anon.dta_enabling) != NULL) + (void) dtrace_enabling_match(enab, NULL); + + mutex_exit(&cpu_lock); + } + + mutex_exit(&dtrace_provider_lock); + mutex_exit(&dtrace_lock); + + if (state != NULL) { + /* + * If we created any anonymous state, set it going now. + */ + (void) dtrace_state_go(state, &dtrace_anon.dta_beganon); + } +} diff --git a/sys/cddl/dev/dtrace/dtrace_cddl.h b/sys/cddl/dev/dtrace/dtrace_cddl.h new file mode 100644 index 000000000000..75fe8648cc12 --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_cddl.h @@ -0,0 +1,134 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + */ + +#ifndef _DTRACE_CDDL_H_ +#define _DTRACE_CDDL_H_ + +#include <sys/proc.h> + +#define LOCK_LEVEL 10 + +/* + * Kernel DTrace extension to 'struct proc' for FreeBSD. + */ +typedef struct kdtrace_proc { + int p_dtrace_probes; /* Are there probes for this proc? */ + u_int64_t p_dtrace_count; /* Number of DTrace tracepoints */ + void *p_dtrace_helpers; /* DTrace helpers, if any */ + +} kdtrace_proc_t; + +/* + * Kernel DTrace extension to 'struct thread' for FreeBSD. + */ +typedef struct kdtrace_thread { + u_int8_t td_dtrace_stop; /* Indicates a DTrace-desired stop */ + u_int8_t td_dtrace_sig; /* Signal sent via DTrace's raise() */ + u_int td_predcache; /* DTrace predicate cache */ + u_int64_t td_dtrace_vtime; /* DTrace virtual time */ + u_int64_t td_dtrace_start; /* DTrace slice start time */ + + union __tdu { + struct __tds { + u_int8_t _td_dtrace_on; + /* Hit a fasttrap tracepoint. */ + u_int8_t _td_dtrace_step; + /* About to return to kernel. */ + u_int8_t _td_dtrace_ret; + /* Handling a return probe. */ + u_int8_t _td_dtrace_ast; + /* Saved ast flag. */ + } _tds; + u_long _td_dtrace_ft; /* Bitwise or of these flags. */ + } _tdu; +#define td_dtrace_ft _tdu._td_dtrace_ft +#define td_dtrace_on _tdu._tds._td_dtrace_on +#define td_dtrace_step _tdu._tds._td_dtrace_step +#define td_dtrace_ret _tdu._tds._td_dtrace_ret +#define td_dtrace_ast _tdu._tds._td_dtrace_ast + + uintptr_t td_dtrace_pc; /* DTrace saved pc from fasttrap. */ + uintptr_t td_dtrace_npc; /* DTrace next pc from fasttrap. */ + uintptr_t td_dtrace_scrpc; + /* DTrace per-thread scratch location. */ + uintptr_t td_dtrace_astpc; + /* DTrace return sequence location. */ + u_int64_t td_hrtime; /* Last time on cpu. */ + int td_errno; /* Syscall return value. */ +} kdtrace_thread_t; + +/* + * Definitions to reference fields in the FreeBSD DTrace structures defined + * above using the names of fields in similar structures in Solaris. Note + * that the separation on FreeBSD is a licensing constraint designed to + * keep the GENERIC kernel BSD licensed. + */ +#define t_dtrace_vtime td_dtrace->td_dtrace_vtime +#define t_dtrace_start td_dtrace->td_dtrace_start +#define t_dtrace_stop td_dtrace->td_dtrace_stop +#define t_dtrace_sig td_dtrace->td_dtrace_sig +#define t_predcache td_dtrace->td_predcache +#define p_dtrace_helpers p_dtrace->p_dtrace_helpers + +/* + * Definitions for fields in struct proc which are named differntly in FreeBSD. + */ +#define p_cred p_ucred +#define p_parent p_pptr + +/* + * Definitions for fields in struct thread which are named differntly in FreeBSD. + */ +#define t_procp td_proc +#define t_tid td_tid +#define t_did td_tid + + +int priv_policy(const cred_t *, int, boolean_t, int, const char *); +boolean_t priv_policy_only(const cred_t *, int, boolean_t); +boolean_t priv_policy_choice(const cred_t *, int, boolean_t); + +/* + * Test privilege. Audit success or failure, allow privilege debugging. + * Returns 0 for success, err for failure. + */ +#define PRIV_POLICY(cred, priv, all, err, reason) \ + priv_policy((cred), (priv), (all), (err), (reason)) + +/* + * Test privilege. Audit success only, no privilege debugging. + * Returns 1 for success, and 0 for failure. + */ +#define PRIV_POLICY_CHOICE(cred, priv, all) \ + priv_policy_choice((cred), (priv), (all)) + +/* + * Test privilege. No priv_debugging, no auditing. + * Returns 1 for success, and 0 for failure. + */ + +#define PRIV_POLICY_ONLY(cred, priv, all) \ + priv_policy_only((cred), (priv), (all)) + +#endif /* !_DTRACE_CDDL_H_ */ diff --git a/sys/cddl/dev/dtrace/dtrace_clone.c b/sys/cddl/dev/dtrace/dtrace_clone.c new file mode 100644 index 000000000000..52e48ff1bd79 --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_clone.c @@ -0,0 +1,61 @@ +/*- + * Copyright (C) 2006 John Birrell <jb@freebsd.org>. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice(s), this list of conditions and the following disclaimer as + * the first lines of this file unmodified other than the possible + * addition of one or more copyright notices. + * 2. Redistributions in binary form must reproduce the above copyright + * notice(s), this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER + * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH + * DAMAGE. + * + * $FreeBSD$ + * + */ + +static void +dtrace_clone(void *arg, struct ucred *cred, char *name, int namelen, struct cdev **dev) +{ + int u = -1; + size_t len; + + if (*dev != NULL) + return; + + len = strlen(name); + + if (len != 6 && len != 13) + return; + + if (bcmp(name,"dtrace",6) != 0) + return; + + if (len == 13 && bcmp(name,"dtrace/dtrace",13) != 0) + return; + + /* Clone the device to the new minor number. */ + if (clone_create(&dtrace_clones, &dtrace_cdevsw, &u, dev, 0) != 0) + /* Create the /dev/dtrace/dtraceNN entry. */ + *dev = make_dev_cred(&dtrace_cdevsw, unit2minor(u), cred, + UID_ROOT, GID_WHEEL, 0600, "dtrace/dtrace%d", u); + if (*dev != NULL) { + dev_ref(*dev); + (*dev)->si_flags |= SI_CHEAPCLONE; + } +} diff --git a/sys/cddl/dev/dtrace/dtrace_debug.c b/sys/cddl/dev/dtrace/dtrace_debug.c new file mode 100644 index 000000000000..24a7a09a0dfc --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_debug.c @@ -0,0 +1,596 @@ +/*- + * Copyright (C) 2008 John Birrell <jb@freebsd.org>. + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice(s), this list of conditions and the following disclaimer as + * the first lines of this file unmodified other than the possible + * addition of one or more copyright notices. + * 2. Redistributions in binary form must reproduce the above copyright + * notice(s), this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER + * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH + * DAMAGE. + * + * $FreeBSD$ + * + */ + +#ifdef DEBUG + +#if defined(__amd64__) +static __inline int +dtrace_cmpset_long(volatile u_long *dst, u_long exp, u_long src) +{ + u_char res; + + __asm __volatile( + " lock ; " + " cmpxchgq %2,%1 ; " + " sete %0 ; " + "1: " + "# dtrace_cmpset_long" + : "=a" (res), /* 0 */ + "=m" (*dst) /* 1 */ + : "r" (src), /* 2 */ + "a" (exp), /* 3 */ + "m" (*dst) /* 4 */ + : "memory"); + + return (res); +} +#elif defined(__i386__) +static __inline int +dtrace_cmpset_long(volatile u_long *dst, u_long exp, u_long src) +{ + u_char res; + + __asm __volatile( + " lock ; " + " cmpxchgl %2,%1 ; " + " sete %0 ; " + "1: " + "# dtrace_cmpset_long" + : "=a" (res), /* 0 */ + "=m" (*dst) /* 1 */ + : "r" (src), /* 2 */ + "a" (exp), /* 3 */ + "m" (*dst) /* 4 */ + : "memory"); + + return (res); +} +#endif + +#define DTRACE_DEBUG_BUFR_SIZE (32 * 1024) + +struct dtrace_debug_data { + char bufr[DTRACE_DEBUG_BUFR_SIZE]; + char *first; + char *last; + char *next; +} dtrace_debug_data[MAXCPU]; + +static char dtrace_debug_bufr[DTRACE_DEBUG_BUFR_SIZE]; + +static volatile u_long dtrace_debug_flag[MAXCPU]; + +static void +dtrace_debug_lock(int cpu) +{ + while (dtrace_cmpset_long(&dtrace_debug_flag[cpu], 0, 1) == 0) + /* Loop until the lock is obtained. */ + ; +} + +static void +dtrace_debug_unlock(int cpu) +{ + dtrace_debug_flag[cpu] = 0; +} + +static void +dtrace_debug_init(void *dummy) +{ + int i; + struct dtrace_debug_data *d; + + for (i = 0; i <= mp_maxid; i++) { + if (pcpu_find(i) == NULL) + continue; + + d = &dtrace_debug_data[i]; + + if (d->first == NULL) { + d->first = d->bufr; + d->next = d->bufr; + d->last = d->bufr + DTRACE_DEBUG_BUFR_SIZE - 1; + *(d->last) = '\0'; + } + } +} + +SYSINIT(dtrace_debug_init, SI_SUB_KDTRACE, SI_ORDER_ANY, dtrace_debug_init, NULL); +SYSINIT(dtrace_debug_smpinit, SI_SUB_SMP, SI_ORDER_ANY, dtrace_debug_init, NULL); + +static void +dtrace_debug_output(void) +{ + char *p; + int i; + struct dtrace_debug_data *d; + uintptr_t count; + + for (i = 0; i <= mp_maxid; i++) { + if (pcpu_find(i) == NULL) + continue; + + dtrace_debug_lock(i); + + d = &dtrace_debug_data[i]; + + count = 0; + + if (d->first < d->next) { + char *p1 = dtrace_debug_bufr; + + count = (uintptr_t) d->next - (uintptr_t) d->first; + + for (p = d->first; p < d->next; p++) + *p1++ = *p; + } else if (d->next > d->first) { + char *p1 = dtrace_debug_bufr; + + count = (uintptr_t) d->last - (uintptr_t) d->first; + + for (p = d->first; p < d->last; p++) + *p1++ = *p; + + count += (uintptr_t) d->next - (uintptr_t) d->bufr; + + for (p = d->bufr; p < d->next; p++) + *p1++ = *p; + } + + d->first = d->bufr; + d->next = d->bufr; + + dtrace_debug_unlock(i); + + if (count > 0) { + char *last = dtrace_debug_bufr + count; + + p = dtrace_debug_bufr; + + while (p < last) { + if (*p == '\0') { + p++; + continue; + } + + printf("%s", p); + + p += strlen(p); + } + } + } +} + +/* + * Functions below here are called from the probe context, so they can't call + * _any_ functions outside the dtrace module without running foul of the function + * boundary trace provider (fbt). The purpose of these functions is limited to + * buffering debug strings for output when the probe completes on the current CPU. + */ + +static __inline void +dtrace_debug__putc(char c) +{ + struct dtrace_debug_data *d = &dtrace_debug_data[curcpu]; + + *d->next++ = c; + + if (d->next == d->last) + d->next = d->bufr; + + *(d->next) = '\0'; + + if (d->next == d->first) + d->first++; + + if (d->first == d->last) + d->first = d->bufr; +} + +static void __used +dtrace_debug_putc(char c) +{ + dtrace_debug_lock(curcpu); + + dtrace_debug__putc(c); + + dtrace_debug_unlock(curcpu); +} + +static void __used +dtrace_debug_puts(const char *s) +{ + dtrace_debug_lock(curcpu); + + while (*s != '\0') + dtrace_debug__putc(*s++); + + dtrace_debug__putc('\0'); + + dtrace_debug_unlock(curcpu); +} + +/* + * Snaffled from sys/kern/subr_prf.c + * + * Put a NUL-terminated ASCII number (base <= 36) in a buffer in reverse + * order; return an optional length and a pointer to the last character + * written in the buffer (i.e., the first character of the string). + * The buffer pointed to by `nbuf' must have length >= MAXNBUF. + */ +static char * +dtrace_debug_ksprintn(char *nbuf, uintmax_t num, int base, int *lenp, int upper) +{ + char *p, c; + + p = nbuf; + *p = '\0'; + do { + c = hex2ascii(num % base); + *++p = upper ? toupper(c) : c; + } while (num /= base); + if (lenp) + *lenp = p - nbuf; + return (p); +} + +#define MAXNBUF (sizeof(intmax_t) * NBBY + 1) + +static void +dtrace_debug_vprintf(const char *fmt, va_list ap) +{ + char nbuf[MAXNBUF]; + const char *p, *percent, *q; + u_char *up; + int ch, n; + uintmax_t num; + int base, lflag, qflag, tmp, width, ladjust, sharpflag, neg, sign, dot; + int cflag, hflag, jflag, tflag, zflag; + int dwidth, upper; + int radix = 10; + char padc; + int stop = 0, retval = 0; + + num = 0; + + if (fmt == NULL) + fmt = "(fmt null)\n"; + + for (;;) { + padc = ' '; + width = 0; + while ((ch = (u_char)*fmt++) != '%' || stop) { + if (ch == '\0') { + dtrace_debug__putc('\0'); + return; + } + dtrace_debug__putc(ch); + } + percent = fmt - 1; + qflag = 0; lflag = 0; ladjust = 0; sharpflag = 0; neg = 0; + sign = 0; dot = 0; dwidth = 0; upper = 0; + cflag = 0; hflag = 0; jflag = 0; tflag = 0; zflag = 0; +reswitch: switch (ch = (u_char)*fmt++) { + case '.': + dot = 1; + goto reswitch; + case '#': + sharpflag = 1; + goto reswitch; + case '+': + sign = 1; + goto reswitch; + case '-': + ladjust = 1; + goto reswitch; + case '%': + dtrace_debug__putc(ch); + break; + case '*': + if (!dot) { + width = va_arg(ap, int); + if (width < 0) { + ladjust = !ladjust; + width = -width; + } + } else { + dwidth = va_arg(ap, int); + } + goto reswitch; + case '0': + if (!dot) { + padc = '0'; + goto reswitch; + } + case '1': case '2': case '3': case '4': + case '5': case '6': case '7': case '8': case '9': + for (n = 0;; ++fmt) { + n = n * 10 + ch - '0'; + ch = *fmt; + if (ch < '0' || ch > '9') + break; + } + if (dot) + dwidth = n; + else + width = n; + goto reswitch; + case 'b': + num = (u_int)va_arg(ap, int); + p = va_arg(ap, char *); + for (q = dtrace_debug_ksprintn(nbuf, num, *p++, NULL, 0); *q;) + dtrace_debug__putc(*q--); + + if (num == 0) + break; + + for (tmp = 0; *p;) { + n = *p++; + if (num & (1 << (n - 1))) { + dtrace_debug__putc(tmp ? ',' : '<'); + for (; (n = *p) > ' '; ++p) + dtrace_debug__putc(n); + tmp = 1; + } else + for (; *p > ' '; ++p) + continue; + } + if (tmp) + dtrace_debug__putc('>'); + break; + case 'c': + dtrace_debug__putc(va_arg(ap, int)); + break; + case 'D': + up = va_arg(ap, u_char *); + p = va_arg(ap, char *); + if (!width) + width = 16; + while(width--) { + dtrace_debug__putc(hex2ascii(*up >> 4)); + dtrace_debug__putc(hex2ascii(*up & 0x0f)); + up++; + if (width) + for (q=p;*q;q++) + dtrace_debug__putc(*q); + } + break; + case 'd': + case 'i': + base = 10; + sign = 1; + goto handle_sign; + case 'h': + if (hflag) { + hflag = 0; + cflag = 1; + } else + hflag = 1; + goto reswitch; + case 'j': + jflag = 1; + goto reswitch; + case 'l': + if (lflag) { + lflag = 0; + qflag = 1; + } else + lflag = 1; + goto reswitch; + case 'n': + if (jflag) + *(va_arg(ap, intmax_t *)) = retval; + else if (qflag) + *(va_arg(ap, quad_t *)) = retval; + else if (lflag) + *(va_arg(ap, long *)) = retval; + else if (zflag) + *(va_arg(ap, size_t *)) = retval; + else if (hflag) + *(va_arg(ap, short *)) = retval; + else if (cflag) + *(va_arg(ap, char *)) = retval; + else + *(va_arg(ap, int *)) = retval; + break; + case 'o': + base = 8; + goto handle_nosign; + case 'p': + base = 16; + sharpflag = (width == 0); + sign = 0; + num = (uintptr_t)va_arg(ap, void *); + goto number; + case 'q': + qflag = 1; + goto reswitch; + case 'r': + base = radix; + if (sign) + goto handle_sign; + goto handle_nosign; + case 's': + p = va_arg(ap, char *); + if (p == NULL) + p = "(null)"; + if (!dot) + n = strlen (p); + else + for (n = 0; n < dwidth && p[n]; n++) + continue; + + width -= n; + + if (!ladjust && width > 0) + while (width--) + dtrace_debug__putc(padc); + while (n--) + dtrace_debug__putc(*p++); + if (ladjust && width > 0) + while (width--) + dtrace_debug__putc(padc); + break; + case 't': + tflag = 1; + goto reswitch; + case 'u': + base = 10; + goto handle_nosign; + case 'X': + upper = 1; + case 'x': + base = 16; + goto handle_nosign; + case 'y': + base = 16; + sign = 1; + goto handle_sign; + case 'z': + zflag = 1; + goto reswitch; +handle_nosign: + sign = 0; + if (jflag) + num = va_arg(ap, uintmax_t); + else if (qflag) + num = va_arg(ap, u_quad_t); + else if (tflag) + num = va_arg(ap, ptrdiff_t); + else if (lflag) + num = va_arg(ap, u_long); + else if (zflag) + num = va_arg(ap, size_t); + else if (hflag) + num = (u_short)va_arg(ap, int); + else if (cflag) + num = (u_char)va_arg(ap, int); + else + num = va_arg(ap, u_int); + goto number; +handle_sign: + if (jflag) + num = va_arg(ap, intmax_t); + else if (qflag) + num = va_arg(ap, quad_t); + else if (tflag) + num = va_arg(ap, ptrdiff_t); + else if (lflag) + num = va_arg(ap, long); + else if (zflag) + num = va_arg(ap, size_t); + else if (hflag) + num = (short)va_arg(ap, int); + else if (cflag) + num = (char)va_arg(ap, int); + else + num = va_arg(ap, int); +number: + if (sign && (intmax_t)num < 0) { + neg = 1; + num = -(intmax_t)num; + } + p = dtrace_debug_ksprintn(nbuf, num, base, &tmp, upper); + if (sharpflag && num != 0) { + if (base == 8) + tmp++; + else if (base == 16) + tmp += 2; + } + if (neg) + tmp++; + + if (!ladjust && padc != '0' && width + && (width -= tmp) > 0) + while (width--) + dtrace_debug__putc(padc); + if (neg) + dtrace_debug__putc('-'); + if (sharpflag && num != 0) { + if (base == 8) { + dtrace_debug__putc('0'); + } else if (base == 16) { + dtrace_debug__putc('0'); + dtrace_debug__putc('x'); + } + } + if (!ladjust && width && (width -= tmp) > 0) + while (width--) + dtrace_debug__putc(padc); + + while (*p) + dtrace_debug__putc(*p--); + + if (ladjust && width && (width -= tmp) > 0) + while (width--) + dtrace_debug__putc(padc); + + break; + default: + while (percent < fmt) + dtrace_debug__putc(*percent++); + /* + * Since we ignore an formatting argument it is no + * longer safe to obey the remaining formatting + * arguments as the arguments will no longer match + * the format specs. + */ + stop = 1; + break; + } + } + + dtrace_debug__putc('\0'); +} + +void +dtrace_debug_printf(const char *fmt, ...) +{ + va_list ap; + + dtrace_debug_lock(curcpu); + + va_start(ap, fmt); + + dtrace_debug_vprintf(fmt, ap); + + va_end(ap); + + dtrace_debug_unlock(curcpu); +} + +#else + +#define dtrace_debug_output() +#define dtrace_debug_puts(_s) +#define dtrace_debug_printf(fmt, ...) + +#endif diff --git a/sys/cddl/dev/dtrace/dtrace_hacks.c b/sys/cddl/dev/dtrace/dtrace_hacks.c new file mode 100644 index 000000000000..21da9f82e703 --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_hacks.c @@ -0,0 +1,13 @@ +/* $FreeBSD$ */ +/* XXX Hacks.... */ + +dtrace_cacheid_t dtrace_predcache_id; + +int panic_quiesce; +char panic_stack[PANICSTKSIZE]; + +boolean_t +priv_policy_only(const cred_t *a, int b, boolean_t c) +{ + return 0; +} diff --git a/sys/cddl/dev/dtrace/dtrace_ioctl.c b/sys/cddl/dev/dtrace/dtrace_ioctl.c new file mode 100644 index 000000000000..bba8c080c829 --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_ioctl.c @@ -0,0 +1,778 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + */ + +static int dtrace_verbose_ioctl; +SYSCTL_INT(_debug_dtrace, OID_AUTO, verbose_ioctl, CTLFLAG_RW, &dtrace_verbose_ioctl, 0, ""); + +#define DTRACE_IOCTL_PRINTF(fmt, ...) if (dtrace_verbose_ioctl) printf(fmt, ## __VA_ARGS__ ) + +/* ARGSUSED */ +static int +dtrace_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, + int flags __unused, struct thread *td) +{ + dtrace_state_t *state; + devfs_get_cdevpriv((void **) &state); + int error = 0; + if (state == NULL) + return (EINVAL); + + if (state->dts_anon) { + ASSERT(dtrace_anon.dta_state == NULL); + state = state->dts_anon; + } + + switch (cmd) { + case DTRACEIOC_AGGDESC: { + dtrace_aggdesc_t **paggdesc = (dtrace_aggdesc_t **) addr; + dtrace_aggdesc_t aggdesc; + dtrace_action_t *act; + dtrace_aggregation_t *agg; + int nrecs; + uint32_t offs; + dtrace_recdesc_t *lrec; + void *buf; + size_t size; + uintptr_t dest; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_AGGDESC\n",__func__,__LINE__); + + if (copyin((void *) *paggdesc, &aggdesc, sizeof (aggdesc)) != 0) + return (EFAULT); + + mutex_enter(&dtrace_lock); + + if ((agg = dtrace_aggid2agg(state, aggdesc.dtagd_id)) == NULL) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + aggdesc.dtagd_epid = agg->dtag_ecb->dte_epid; + + nrecs = aggdesc.dtagd_nrecs; + aggdesc.dtagd_nrecs = 0; + + offs = agg->dtag_base; + lrec = &agg->dtag_action.dta_rec; + aggdesc.dtagd_size = lrec->dtrd_offset + lrec->dtrd_size - offs; + + for (act = agg->dtag_first; ; act = act->dta_next) { + ASSERT(act->dta_intuple || + DTRACEACT_ISAGG(act->dta_kind)); + + /* + * If this action has a record size of zero, it + * denotes an argument to the aggregating action. + * Because the presence of this record doesn't (or + * shouldn't) affect the way the data is interpreted, + * we don't copy it out to save user-level the + * confusion of dealing with a zero-length record. + */ + if (act->dta_rec.dtrd_size == 0) { + ASSERT(agg->dtag_hasarg); + continue; + } + + aggdesc.dtagd_nrecs++; + + if (act == &agg->dtag_action) + break; + } + + /* + * Now that we have the size, we need to allocate a temporary + * buffer in which to store the complete description. We need + * the temporary buffer to be able to drop dtrace_lock() + * across the copyout(), below. + */ + size = sizeof (dtrace_aggdesc_t) + + (aggdesc.dtagd_nrecs * sizeof (dtrace_recdesc_t)); + + buf = kmem_alloc(size, KM_SLEEP); + dest = (uintptr_t)buf; + + bcopy(&aggdesc, (void *)dest, sizeof (aggdesc)); + dest += offsetof(dtrace_aggdesc_t, dtagd_rec[0]); + + for (act = agg->dtag_first; ; act = act->dta_next) { + dtrace_recdesc_t rec = act->dta_rec; + + /* + * See the comment in the above loop for why we pass + * over zero-length records. + */ + if (rec.dtrd_size == 0) { + ASSERT(agg->dtag_hasarg); + continue; + } + + if (nrecs-- == 0) + break; + + rec.dtrd_offset -= offs; + bcopy(&rec, (void *)dest, sizeof (rec)); + dest += sizeof (dtrace_recdesc_t); + + if (act == &agg->dtag_action) + break; + } + + mutex_exit(&dtrace_lock); + + if (copyout(buf, (void *) *paggdesc, dest - (uintptr_t)buf) != 0) { + kmem_free(buf, size); + return (EFAULT); + } + + kmem_free(buf, size); + return (0); + } + case DTRACEIOC_AGGSNAP: + case DTRACEIOC_BUFSNAP: { + dtrace_bufdesc_t **pdesc = (dtrace_bufdesc_t **) addr; + dtrace_bufdesc_t desc; + caddr_t cached; + dtrace_buffer_t *buf; + + dtrace_debug_output(); + + if (copyin((void *) *pdesc, &desc, sizeof (desc)) != 0) + return (EFAULT); + + DTRACE_IOCTL_PRINTF("%s(%d): %s curcpu %d cpu %d\n", + __func__,__LINE__, + cmd == DTRACEIOC_AGGSNAP ? + "DTRACEIOC_AGGSNAP":"DTRACEIOC_BUFSNAP", + curcpu, desc.dtbd_cpu); + + if (desc.dtbd_cpu < 0 || desc.dtbd_cpu >= NCPU) + return (ENOENT); + if (pcpu_find(desc.dtbd_cpu) == NULL) + return (ENOENT); + + mutex_enter(&dtrace_lock); + + if (cmd == DTRACEIOC_BUFSNAP) { + buf = &state->dts_buffer[desc.dtbd_cpu]; + } else { + buf = &state->dts_aggbuffer[desc.dtbd_cpu]; + } + + if (buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL)) { + size_t sz = buf->dtb_offset; + + if (state->dts_activity != DTRACE_ACTIVITY_STOPPED) { + mutex_exit(&dtrace_lock); + return (EBUSY); + } + + /* + * If this buffer has already been consumed, we're + * going to indicate that there's nothing left here + * to consume. + */ + if (buf->dtb_flags & DTRACEBUF_CONSUMED) { + mutex_exit(&dtrace_lock); + + desc.dtbd_size = 0; + desc.dtbd_drops = 0; + desc.dtbd_errors = 0; + desc.dtbd_oldest = 0; + sz = sizeof (desc); + + if (copyout(&desc, (void *) *pdesc, sz) != 0) + return (EFAULT); + + return (0); + } + + /* + * If this is a ring buffer that has wrapped, we want + * to copy the whole thing out. + */ + if (buf->dtb_flags & DTRACEBUF_WRAPPED) { + dtrace_buffer_polish(buf); + sz = buf->dtb_size; + } + + if (copyout(buf->dtb_tomax, desc.dtbd_data, sz) != 0) { + mutex_exit(&dtrace_lock); + return (EFAULT); + } + + desc.dtbd_size = sz; + desc.dtbd_drops = buf->dtb_drops; + desc.dtbd_errors = buf->dtb_errors; + desc.dtbd_oldest = buf->dtb_xamot_offset; + + mutex_exit(&dtrace_lock); + + if (copyout(&desc, (void *) *pdesc, sizeof (desc)) != 0) + return (EFAULT); + + buf->dtb_flags |= DTRACEBUF_CONSUMED; + + return (0); + } + + if (buf->dtb_tomax == NULL) { + ASSERT(buf->dtb_xamot == NULL); + mutex_exit(&dtrace_lock); + return (ENOENT); + } + + cached = buf->dtb_tomax; + ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); + + dtrace_xcall(desc.dtbd_cpu, + (dtrace_xcall_t)dtrace_buffer_switch, buf); + + state->dts_errors += buf->dtb_xamot_errors; + + /* + * If the buffers did not actually switch, then the cross call + * did not take place -- presumably because the given CPU is + * not in the ready set. If this is the case, we'll return + * ENOENT. + */ + if (buf->dtb_tomax == cached) { + ASSERT(buf->dtb_xamot != cached); + mutex_exit(&dtrace_lock); + return (ENOENT); + } + + ASSERT(cached == buf->dtb_xamot); + + DTRACE_IOCTL_PRINTF("%s(%d): copyout the buffer snapshot\n",__func__,__LINE__); + + /* + * We have our snapshot; now copy it out. + */ + if (copyout(buf->dtb_xamot, desc.dtbd_data, + buf->dtb_xamot_offset) != 0) { + mutex_exit(&dtrace_lock); + return (EFAULT); + } + + desc.dtbd_size = buf->dtb_xamot_offset; + desc.dtbd_drops = buf->dtb_xamot_drops; + desc.dtbd_errors = buf->dtb_xamot_errors; + desc.dtbd_oldest = 0; + + mutex_exit(&dtrace_lock); + + DTRACE_IOCTL_PRINTF("%s(%d): copyout buffer desc: size %zd drops %lu errors %lu\n",__func__,__LINE__,(size_t) desc.dtbd_size,(u_long) desc.dtbd_drops,(u_long) desc.dtbd_errors); + + /* + * Finally, copy out the buffer description. + */ + if (copyout(&desc, (void *) *pdesc, sizeof (desc)) != 0) + return (EFAULT); + + return (0); + } + case DTRACEIOC_CONF: { + dtrace_conf_t conf; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_CONF\n",__func__,__LINE__); + + bzero(&conf, sizeof (conf)); + conf.dtc_difversion = DIF_VERSION; + conf.dtc_difintregs = DIF_DIR_NREGS; + conf.dtc_diftupregs = DIF_DTR_NREGS; + conf.dtc_ctfmodel = CTF_MODEL_NATIVE; + + *((dtrace_conf_t *) addr) = conf; + + return (0); + } + case DTRACEIOC_DOFGET: { + dof_hdr_t **pdof = (dof_hdr_t **) addr; + dof_hdr_t hdr, *dof = *pdof; + int rval; + uint64_t len; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_DOFGET\n",__func__,__LINE__); + + if (copyin((void *)dof, &hdr, sizeof (hdr)) != 0) + return (EFAULT); + + mutex_enter(&dtrace_lock); + dof = dtrace_dof_create(state); + mutex_exit(&dtrace_lock); + + len = MIN(hdr.dofh_loadsz, dof->dofh_loadsz); + rval = copyout(dof, (void *) *pdof, len); + dtrace_dof_destroy(dof); + + return (rval == 0 ? 0 : EFAULT); + } + case DTRACEIOC_ENABLE: { + dof_hdr_t *dof = NULL; + dtrace_enabling_t *enab = NULL; + dtrace_vstate_t *vstate; + int err = 0; + int rval; + dtrace_enable_io_t *p = (dtrace_enable_io_t *) addr; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_ENABLE\n",__func__,__LINE__); + + /* + * If a NULL argument has been passed, we take this as our + * cue to reevaluate our enablings. + */ + if (p->dof == NULL) { + dtrace_enabling_matchall(); + + return (0); + } + + if ((dof = dtrace_dof_copyin((uintptr_t) p->dof, &rval)) == NULL) + return (EINVAL); + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_lock); + vstate = &state->dts_vstate; + + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) { + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + dtrace_dof_destroy(dof); + return (EBUSY); + } + + if (dtrace_dof_slurp(dof, vstate, td->td_ucred, &enab, 0, B_TRUE) != 0) { + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + dtrace_dof_destroy(dof); + return (EINVAL); + } + + if ((rval = dtrace_dof_options(dof, state)) != 0) { + dtrace_enabling_destroy(enab); + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + dtrace_dof_destroy(dof); + return (rval); + } + + if ((err = dtrace_enabling_match(enab, &p->n_matched)) == 0) { + err = dtrace_enabling_retain(enab); + } else { + dtrace_enabling_destroy(enab); + } + + mutex_exit(&cpu_lock); + mutex_exit(&dtrace_lock); + dtrace_dof_destroy(dof); + + return (err); + } + case DTRACEIOC_EPROBE: { + dtrace_eprobedesc_t **pepdesc = (dtrace_eprobedesc_t **) addr; + dtrace_eprobedesc_t epdesc; + dtrace_ecb_t *ecb; + dtrace_action_t *act; + void *buf; + size_t size; + uintptr_t dest; + int nrecs; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_EPROBE\n",__func__,__LINE__); + + if (copyin((void *)*pepdesc, &epdesc, sizeof (epdesc)) != 0) + return (EFAULT); + + mutex_enter(&dtrace_lock); + + if ((ecb = dtrace_epid2ecb(state, epdesc.dtepd_epid)) == NULL) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + if (ecb->dte_probe == NULL) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + epdesc.dtepd_probeid = ecb->dte_probe->dtpr_id; + epdesc.dtepd_uarg = ecb->dte_uarg; + epdesc.dtepd_size = ecb->dte_size; + + nrecs = epdesc.dtepd_nrecs; + epdesc.dtepd_nrecs = 0; + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) + continue; + + epdesc.dtepd_nrecs++; + } + + /* + * Now that we have the size, we need to allocate a temporary + * buffer in which to store the complete description. We need + * the temporary buffer to be able to drop dtrace_lock() + * across the copyout(), below. + */ + size = sizeof (dtrace_eprobedesc_t) + + (epdesc.dtepd_nrecs * sizeof (dtrace_recdesc_t)); + + buf = kmem_alloc(size, KM_SLEEP); + dest = (uintptr_t)buf; + + bcopy(&epdesc, (void *)dest, sizeof (epdesc)); + dest += offsetof(dtrace_eprobedesc_t, dtepd_rec[0]); + + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) + continue; + + if (nrecs-- == 0) + break; + + bcopy(&act->dta_rec, (void *)dest, + sizeof (dtrace_recdesc_t)); + dest += sizeof (dtrace_recdesc_t); + } + + mutex_exit(&dtrace_lock); + + if (copyout(buf, (void *) *pepdesc, dest - (uintptr_t)buf) != 0) { + kmem_free(buf, size); + return (EFAULT); + } + + kmem_free(buf, size); + return (0); + } + case DTRACEIOC_FORMAT: { + dtrace_fmtdesc_t *fmt = (dtrace_fmtdesc_t *) addr; + char *str; + int len; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_FORMAT\n",__func__,__LINE__); + + mutex_enter(&dtrace_lock); + + if (fmt->dtfd_format == 0 || + fmt->dtfd_format > state->dts_nformats) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + /* + * Format strings are allocated contiguously and they are + * never freed; if a format index is less than the number + * of formats, we can assert that the format map is non-NULL + * and that the format for the specified index is non-NULL. + */ + ASSERT(state->dts_formats != NULL); + str = state->dts_formats[fmt->dtfd_format - 1]; + ASSERT(str != NULL); + + len = strlen(str) + 1; + + if (len > fmt->dtfd_length) { + fmt->dtfd_length = len; + } else { + if (copyout(str, fmt->dtfd_string, len) != 0) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + } + + mutex_exit(&dtrace_lock); + return (0); + } + case DTRACEIOC_GO: { + int rval; + processorid_t *cpuid = (processorid_t *) addr; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_GO\n",__func__,__LINE__); + + rval = dtrace_state_go(state, cpuid); + + return (rval); + } + case DTRACEIOC_PROBEARG: { + dtrace_argdesc_t *desc = (dtrace_argdesc_t *) addr; + dtrace_probe_t *probe; + dtrace_provider_t *prov; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_PROBEARG\n",__func__,__LINE__); + + if (desc->dtargd_id == DTRACE_IDNONE) + return (EINVAL); + + if (desc->dtargd_ndx == DTRACE_ARGNONE) + return (EINVAL); + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&mod_lock); + mutex_enter(&dtrace_lock); + + if (desc->dtargd_id > dtrace_nprobes) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + return (EINVAL); + } + + if ((probe = dtrace_probes[desc->dtargd_id - 1]) == NULL) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + return (EINVAL); + } + + mutex_exit(&dtrace_lock); + + prov = probe->dtpr_provider; + + if (prov->dtpv_pops.dtps_getargdesc == NULL) { + /* + * There isn't any typed information for this probe. + * Set the argument number to DTRACE_ARGNONE. + */ + desc->dtargd_ndx = DTRACE_ARGNONE; + } else { + desc->dtargd_native[0] = '\0'; + desc->dtargd_xlate[0] = '\0'; + desc->dtargd_mapping = desc->dtargd_ndx; + + prov->dtpv_pops.dtps_getargdesc(prov->dtpv_arg, + probe->dtpr_id, probe->dtpr_arg, desc); + } + + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + + return (0); + } + case DTRACEIOC_PROBEMATCH: + case DTRACEIOC_PROBES: { + dtrace_probedesc_t *p_desc = (dtrace_probedesc_t *) addr; + dtrace_probe_t *probe = NULL; + dtrace_probekey_t pkey; + dtrace_id_t i; + int m = 0; + uint32_t priv = 0; + uid_t uid = 0; + zoneid_t zoneid = 0; + + DTRACE_IOCTL_PRINTF("%s(%d): %s\n",__func__,__LINE__, + cmd == DTRACEIOC_PROBEMATCH ? + "DTRACEIOC_PROBEMATCH":"DTRACEIOC_PROBES"); + + p_desc->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + p_desc->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + p_desc->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + p_desc->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + /* + * Before we attempt to match this probe, we want to give + * all providers the opportunity to provide it. + */ + if (p_desc->dtpd_id == DTRACE_IDNONE) { + mutex_enter(&dtrace_provider_lock); + dtrace_probe_provide(p_desc, NULL); + mutex_exit(&dtrace_provider_lock); + p_desc->dtpd_id++; + } + + if (cmd == DTRACEIOC_PROBEMATCH) { + dtrace_probekey(p_desc, &pkey); + pkey.dtpk_id = DTRACE_IDNONE; + } + + dtrace_cred2priv(td->td_ucred, &priv, &uid, &zoneid); + + mutex_enter(&dtrace_lock); + + if (cmd == DTRACEIOC_PROBEMATCH) { + for (i = p_desc->dtpd_id; i <= dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i - 1]) != NULL && + (m = dtrace_match_probe(probe, &pkey, + priv, uid, zoneid)) != 0) + break; + } + + if (m < 0) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + } else { + for (i = p_desc->dtpd_id; i <= dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i - 1]) != NULL && + dtrace_match_priv(probe, priv, uid, zoneid)) + break; + } + } + + if (probe == NULL) { + mutex_exit(&dtrace_lock); + return (ESRCH); + } + + dtrace_probe_description(probe, p_desc); + mutex_exit(&dtrace_lock); + + return (0); + } + case DTRACEIOC_PROVIDER: { + dtrace_providerdesc_t *pvd = (dtrace_providerdesc_t *) addr; + dtrace_provider_t *pvp; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_PROVIDER\n",__func__,__LINE__); + + pvd->dtvd_name[DTRACE_PROVNAMELEN - 1] = '\0'; + mutex_enter(&dtrace_provider_lock); + + for (pvp = dtrace_provider; pvp != NULL; pvp = pvp->dtpv_next) { + if (strcmp(pvp->dtpv_name, pvd->dtvd_name) == 0) + break; + } + + mutex_exit(&dtrace_provider_lock); + + if (pvp == NULL) + return (ESRCH); + + bcopy(&pvp->dtpv_priv, &pvd->dtvd_priv, sizeof (dtrace_ppriv_t)); + bcopy(&pvp->dtpv_attr, &pvd->dtvd_attr, sizeof (dtrace_pattr_t)); + + return (0); + } + case DTRACEIOC_REPLICATE: { + dtrace_repldesc_t *desc = (dtrace_repldesc_t *) addr; + dtrace_probedesc_t *match = &desc->dtrpd_match; + dtrace_probedesc_t *create = &desc->dtrpd_create; + int err; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_REPLICATE\n",__func__,__LINE__); + + match->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + match->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + match->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + match->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + create->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + create->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + create->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + create->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + mutex_enter(&dtrace_lock); + err = dtrace_enabling_replicate(state, match, create); + mutex_exit(&dtrace_lock); + + return (err); + } + case DTRACEIOC_STATUS: { + dtrace_status_t *stat = (dtrace_status_t *) addr; + dtrace_dstate_t *dstate; + int i, j; + uint64_t nerrs; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_STATUS\n",__func__,__LINE__); + + /* + * See the comment in dtrace_state_deadman() for the reason + * for setting dts_laststatus to INT64_MAX before setting + * it to the correct value. + */ + state->dts_laststatus = INT64_MAX; + dtrace_membar_producer(); + state->dts_laststatus = dtrace_gethrtime(); + + bzero(stat, sizeof (*stat)); + + mutex_enter(&dtrace_lock); + + if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) { + mutex_exit(&dtrace_lock); + return (ENOENT); + } + + if (state->dts_activity == DTRACE_ACTIVITY_DRAINING) + stat->dtst_exiting = 1; + + nerrs = state->dts_errors; + dstate = &state->dts_vstate.dtvs_dynvars; + + for (i = 0; i < NCPU; i++) { +#if !defined(sun) + if (pcpu_find(i) == NULL) + continue; +#endif + dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[i]; + + stat->dtst_dyndrops += dcpu->dtdsc_drops; + stat->dtst_dyndrops_dirty += dcpu->dtdsc_dirty_drops; + stat->dtst_dyndrops_rinsing += dcpu->dtdsc_rinsing_drops; + + if (state->dts_buffer[i].dtb_flags & DTRACEBUF_FULL) + stat->dtst_filled++; + + nerrs += state->dts_buffer[i].dtb_errors; + + for (j = 0; j < state->dts_nspeculations; j++) { + dtrace_speculation_t *spec; + dtrace_buffer_t *buf; + + spec = &state->dts_speculations[j]; + buf = &spec->dtsp_buffer[i]; + stat->dtst_specdrops += buf->dtb_xamot_drops; + } + } + + stat->dtst_specdrops_busy = state->dts_speculations_busy; + stat->dtst_specdrops_unavail = state->dts_speculations_unavail; + stat->dtst_stkstroverflows = state->dts_stkstroverflows; + stat->dtst_dblerrors = state->dts_dblerrors; + stat->dtst_killed = + (state->dts_activity == DTRACE_ACTIVITY_KILLED); + stat->dtst_errors = nerrs; + + mutex_exit(&dtrace_lock); + + return (0); + } + case DTRACEIOC_STOP: { + int rval; + processorid_t *cpuid = (processorid_t *) addr; + + DTRACE_IOCTL_PRINTF("%s(%d): DTRACEIOC_STOP\n",__func__,__LINE__); + + mutex_enter(&dtrace_lock); + rval = dtrace_state_stop(state, cpuid); + mutex_exit(&dtrace_lock); + + return (rval); + } + default: + error = ENOTTY; + } + return (error); +} diff --git a/sys/cddl/dev/dtrace/dtrace_load.c b/sys/cddl/dev/dtrace/dtrace_load.c new file mode 100644 index 000000000000..08506efcf195 --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_load.c @@ -0,0 +1,160 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + */ + +static void +dtrace_ap_start(void *dummy) +{ + int i; + + mutex_enter(&cpu_lock); + + /* Setup the rest of the CPUs. */ + for (i = 1; i <= mp_maxid; i++) { + if (pcpu_find(i) == NULL) + continue; + + (void) dtrace_cpu_setup(CPU_CONFIG, i); + } + + mutex_exit(&cpu_lock); +} + +SYSINIT(dtrace_ap_start, SI_SUB_SMP, SI_ORDER_ANY, dtrace_ap_start, NULL); + +static void +dtrace_load(void *dummy) +{ + dtrace_provider_id_t id; + + /* Hook into the trap handler. */ + dtrace_trap_func = dtrace_trap; + + /* Hang our hook for thread switches. */ + dtrace_vtime_switch_func = dtrace_vtime_switch; + + /* Hang our hook for exceptions. */ + dtrace_invop_init(); + + /* + * XXX This is a short term hack to avoid having to comment + * out lots and lots of lock/unlock calls. + */ + mutex_init(&mod_lock,"XXX mod_lock hack", MUTEX_DEFAULT, NULL); + + /* + * Initialise the mutexes without 'witness' because the dtrace + * code is mostly written to wait for memory. To have the + * witness code change a malloc() from M_WAITOK to M_NOWAIT + * because a lock is held would surely create a panic in a + * low memory situation. And that low memory situation might be + * the very problem we are trying to trace. + */ + mutex_init(&dtrace_lock,"dtrace probe state", MUTEX_DEFAULT, NULL); + mutex_init(&dtrace_provider_lock,"dtrace provider state", MUTEX_DEFAULT, NULL); + mutex_init(&dtrace_meta_lock,"dtrace meta-provider state", MUTEX_DEFAULT, NULL); + mutex_init(&dtrace_errlock,"dtrace error lock", MUTEX_DEFAULT, NULL); + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + mutex_enter(&cpu_lock); + + ASSERT(MUTEX_HELD(&cpu_lock)); + + dtrace_arena = new_unrhdr(1, INT_MAX, &dtrace_unr_mtx); + + dtrace_state_cache = kmem_cache_create("dtrace_state_cache", + sizeof (dtrace_dstate_percpu_t) * NCPU, DTRACE_STATE_ALIGN, + NULL, NULL, NULL, NULL, NULL, 0); + + ASSERT(MUTEX_HELD(&cpu_lock)); + dtrace_bymod = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_mod), + offsetof(dtrace_probe_t, dtpr_nextmod), + offsetof(dtrace_probe_t, dtpr_prevmod)); + + dtrace_byfunc = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_func), + offsetof(dtrace_probe_t, dtpr_nextfunc), + offsetof(dtrace_probe_t, dtpr_prevfunc)); + + dtrace_byname = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_name), + offsetof(dtrace_probe_t, dtpr_nextname), + offsetof(dtrace_probe_t, dtpr_prevname)); + + if (dtrace_retain_max < 1) { + cmn_err(CE_WARN, "illegal value (%lu) for dtrace_retain_max; " + "setting to 1", dtrace_retain_max); + dtrace_retain_max = 1; + } + + /* + * Now discover our toxic ranges. + */ + dtrace_toxic_ranges(dtrace_toxrange_add); + + /* + * Before we register ourselves as a provider to our own framework, + * we would like to assert that dtrace_provider is NULL -- but that's + * not true if we were loaded as a dependency of a DTrace provider. + * Once we've registered, we can assert that dtrace_provider is our + * pseudo provider. + */ + (void) dtrace_register("dtrace", &dtrace_provider_attr, + DTRACE_PRIV_NONE, 0, &dtrace_provider_ops, NULL, &id); + + ASSERT(dtrace_provider != NULL); + ASSERT((dtrace_provider_id_t)dtrace_provider == id); + + dtrace_probeid_begin = dtrace_probe_create((dtrace_provider_id_t) + dtrace_provider, NULL, NULL, "BEGIN", 0, NULL); + dtrace_probeid_end = dtrace_probe_create((dtrace_provider_id_t) + dtrace_provider, NULL, NULL, "END", 0, NULL); + dtrace_probeid_error = dtrace_probe_create((dtrace_provider_id_t) + dtrace_provider, NULL, NULL, "ERROR", 1, NULL); + + mutex_exit(&cpu_lock); + + /* + * If DTrace helper tracing is enabled, we need to allocate the + * trace buffer and initialize the values. + */ + if (dtrace_helptrace_enabled) { + ASSERT(dtrace_helptrace_buffer == NULL); + dtrace_helptrace_buffer = + kmem_zalloc(dtrace_helptrace_bufsize, KM_SLEEP); + dtrace_helptrace_next = 0; + } + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + mutex_enter(&cpu_lock); + + /* Setup the boot CPU */ + (void) dtrace_cpu_setup(CPU_CONFIG, 0); + + mutex_exit(&cpu_lock); + + dtrace_dev = make_dev(&dtrace_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, "dtrace/dtrace"); + + return; +} diff --git a/sys/cddl/dev/dtrace/dtrace_modevent.c b/sys/cddl/dev/dtrace/dtrace_modevent.c new file mode 100644 index 000000000000..8d318532dee9 --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_modevent.c @@ -0,0 +1,47 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + */ + +/* ARGSUSED */ +static int +dtrace_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + return (error); +} diff --git a/sys/cddl/dev/dtrace/dtrace_sysctl.c b/sys/cddl/dev/dtrace/dtrace_sysctl.c new file mode 100644 index 000000000000..00ed70916010 --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_sysctl.c @@ -0,0 +1,82 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + */ + +int dtrace_debug = 0; +TUNABLE_INT("debug.dtrace.debug", &dtrace_debug); +SYSCTL_INT(_debug_dtrace, OID_AUTO, debug, CTLFLAG_RW, &dtrace_debug, 0, ""); + +/* Report registered DTrace providers. */ +static int +sysctl_dtrace_providers(SYSCTL_HANDLER_ARGS) +{ + char *p_name = NULL; + dtrace_provider_t + *prov = dtrace_provider; + int error = 0; + size_t len = 0; + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + /* Compute the length of the space-separated provider name string. */ + while (prov != NULL) { + len += strlen(prov->dtpv_name) + 1; + prov = prov->dtpv_next; + } + + if ((p_name = kmem_alloc(len, KM_SLEEP)) == NULL) + error = ENOMEM; + else { + /* Start with an empty string. */ + *p_name = '\0'; + + /* Point to the first provider again. */ + prov = dtrace_provider; + + /* Loop through the providers, appending the names. */ + while (prov != NULL) { + if (prov != dtrace_provider) + (void) strlcat(p_name, " ", len); + + (void) strlcat(p_name, prov->dtpv_name, len); + + prov = prov->dtpv_next; + } + } + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + if (p_name != NULL) { + error = sysctl_handle_string(oidp, p_name, len, req); + + kmem_free(p_name, 0); + } + + return (error); +} + +SYSCTL_PROC(_debug_dtrace, OID_AUTO, providers, CTLTYPE_STRING | CTLFLAG_RD, + 0, 0, sysctl_dtrace_providers, "A", ""); + diff --git a/sys/cddl/dev/dtrace/dtrace_test.c b/sys/cddl/dev/dtrace/dtrace_test.c new file mode 100644 index 000000000000..d484fb2c0802 --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_test.c @@ -0,0 +1,73 @@ +/*- + * Copyright 2008 John Birrell <jb@FreeBSD.org> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + */ + +#include <sys/cdefs.h> +#include <sys/types.h> +#include <sys/param.h> +#include <sys/conf.h> +#include <sys/kernel.h> +#include <sys/module.h> +#include <sys/vnode.h> + +/* + * These are variables that the DTrace test suite references in the + * Solaris kernel. We define them here so that the tests function + * unaltered. + */ +int kmem_flags; + +typedef struct vnode vnode_t; +vnode_t dummy; +vnode_t *rootvp = &dummy; + +static int +dtrace_test_modevent(module_t mod, int type, void *data) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + return (error); +} + +DEV_MODULE(dtrace_test, dtrace_test_modevent, NULL); +MODULE_VERSION(dtrace_test, 1); +MODULE_DEPEND(dtrace_test, dtraceall, 1, 1, 1); diff --git a/sys/cddl/dev/dtrace/dtrace_unload.c b/sys/cddl/dev/dtrace/dtrace_unload.c new file mode 100644 index 000000000000..15d41afd29df --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_unload.c @@ -0,0 +1,127 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + */ + +static int +dtrace_unload() +{ + dtrace_state_t *state; + int error = 0; + + destroy_dev(dtrace_dev); + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + mutex_enter(&cpu_lock); + + ASSERT(dtrace_opens == 0); + + if (dtrace_helpers > 0) { + mutex_exit(&cpu_lock); + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + return (EBUSY); + } + + if (dtrace_unregister((dtrace_provider_id_t)dtrace_provider) != 0) { + mutex_exit(&cpu_lock); + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + return (EBUSY); + } + + dtrace_provider = NULL; + + if ((state = dtrace_anon_grab()) != NULL) { + /* + * If there were ECBs on this state, the provider should + * have not been allowed to detach; assert that there is + * none. + */ + ASSERT(state->dts_necbs == 0); + dtrace_state_destroy(state); + } + + bzero(&dtrace_anon, sizeof (dtrace_anon_t)); + + mutex_exit(&cpu_lock); + + if (dtrace_helptrace_enabled) { + kmem_free(dtrace_helptrace_buffer, 0); + dtrace_helptrace_buffer = NULL; + } + + if (dtrace_probes != NULL) { + kmem_free(dtrace_probes, 0); + dtrace_probes = NULL; + dtrace_nprobes = 0; + } + + dtrace_hash_destroy(dtrace_bymod); + dtrace_hash_destroy(dtrace_byfunc); + dtrace_hash_destroy(dtrace_byname); + dtrace_bymod = NULL; + dtrace_byfunc = NULL; + dtrace_byname = NULL; + + kmem_cache_destroy(dtrace_state_cache); + + delete_unrhdr(dtrace_arena); + + if (dtrace_toxrange != NULL) { + kmem_free(dtrace_toxrange, 0); + dtrace_toxrange = NULL; + dtrace_toxranges = 0; + dtrace_toxranges_max = 0; + } + + ASSERT(dtrace_vtime_references == 0); + ASSERT(dtrace_opens == 0); + ASSERT(dtrace_retained == NULL); + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + mutex_destroy(&dtrace_meta_lock); + mutex_destroy(&dtrace_provider_lock); + mutex_destroy(&dtrace_lock); + mutex_destroy(&dtrace_errlock); + + /* XXX Hack */ + mutex_destroy(&mod_lock); + + /* Reset our hook for exceptions. */ + dtrace_invop_uninit(); + + /* + * Reset our hook for thread switches, but ensure that vtime isn't + * active first. + */ + dtrace_vtime_active = 0; + dtrace_vtime_switch_func = NULL; + + /* Unhook from the trap handler. */ + dtrace_trap_func = NULL; + + return (error); +} diff --git a/sys/cddl/dev/dtrace/dtrace_vtime.c b/sys/cddl/dev/dtrace/dtrace_vtime.c new file mode 100644 index 000000000000..a3fa7f77287a --- /dev/null +++ b/sys/cddl/dev/dtrace/dtrace_vtime.c @@ -0,0 +1,101 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ + +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +void +dtrace_vtime_enable(void) +{ + dtrace_vtime_state_t state, nstate = 0; + + do { + state = dtrace_vtime_active; + + switch (state) { + case DTRACE_VTIME_INACTIVE: + nstate = DTRACE_VTIME_ACTIVE; + break; + + case DTRACE_VTIME_INACTIVE_TNF: + nstate = DTRACE_VTIME_ACTIVE_TNF; + break; + + case DTRACE_VTIME_ACTIVE: + case DTRACE_VTIME_ACTIVE_TNF: + panic("DTrace virtual time already enabled"); + /*NOTREACHED*/ + } + + } while (dtrace_cas32((uint32_t *)&dtrace_vtime_active, + state, nstate) != state); +} + +void +dtrace_vtime_disable(void) +{ + dtrace_vtime_state_t state, nstate = 0; + + do { + state = dtrace_vtime_active; + + switch (state) { + case DTRACE_VTIME_ACTIVE: + nstate = DTRACE_VTIME_INACTIVE; + break; + + case DTRACE_VTIME_ACTIVE_TNF: + nstate = DTRACE_VTIME_INACTIVE_TNF; + break; + + case DTRACE_VTIME_INACTIVE: + case DTRACE_VTIME_INACTIVE_TNF: + panic("DTrace virtual time already disabled"); + /*NOTREACHED*/ + } + + } while (dtrace_cas32((uint32_t *)&dtrace_vtime_active, + state, nstate) != state); +} + +void +dtrace_vtime_switch(kthread_t *next) +{ + dtrace_icookie_t cookie; + hrtime_t ts; + + cookie = dtrace_interrupt_disable(); + ts = dtrace_gethrtime(); + + if (curthread->t_dtrace_start != 0) { + curthread->t_dtrace_vtime += ts - curthread->t_dtrace_start; + curthread->t_dtrace_start = 0; + } + + if (next != NULL) + next->t_dtrace_start = ts; + + dtrace_interrupt_enable(cookie); +} diff --git a/sys/cddl/dev/dtrace/i386/dis_tables.c b/sys/cddl/dev/dtrace/i386/dis_tables.c new file mode 100644 index 000000000000..5a5bc25c02c3 --- /dev/null +++ b/sys/cddl/dev/dtrace/i386/dis_tables.c @@ -0,0 +1,3193 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +/* Copyright (c) 1988 AT&T */ +/* All Rights Reserved */ + + +#if defined(sun) +#pragma ident "@(#)dis_tables.c 1.11 06/03/02 SMI" +#endif + +#include "dis_tables.h" + +/* BEGIN CSTYLED */ + +/* + * Disassembly begins in dis_distable, which is equivalent to the One-byte + * Opcode Map in the Intel IA32 ISA Reference (page A-6 in my copy). The + * decoding loops then traverse out through the other tables as necessary to + * decode a given instruction. + * + * The behavior of this file can be controlled by one of the following flags: + * + * DIS_TEXT Include text for disassembly + * DIS_MEM Include memory-size calculations + * + * Either or both of these can be defined. + * + * This file is not, and will never be, cstyled. If anything, the tables should + * be taken out another tab stop or two so nothing overlaps. + */ + +/* + * These functions must be provided for the consumer to do disassembly. + */ +#ifdef DIS_TEXT +extern char *strncpy(char *, const char *, size_t); +extern size_t strlen(const char *); +extern int strcmp(const char *, const char *); +extern int strncmp(const char *, const char *, size_t); +extern size_t strlcat(char *, const char *, size_t); +#endif + + +#define TERM NULL /* used to indicate that the 'indirect' */ + /* field terminates - no pointer. */ + +/* Used to decode instructions. */ +typedef struct instable { + const struct instable *it_indirect; /* for decode op codes */ + uchar_t it_adrmode; +#ifdef DIS_TEXT + char it_name[NCPS]; + uint_t it_suffix:1; /* mneu + "w", "l", or "d" */ +#endif +#ifdef DIS_MEM + uint_t it_size:16; +#endif + uint_t it_invalid64:1; /* opcode invalid in amd64 */ + uint_t it_always64:1; /* 64 bit when in 64 bit mode */ + uint_t it_invalid32:1; /* invalid in IA32 */ + uint_t it_stackop:1; /* push/pop stack operation */ +} instable_t; + +/* + * Instruction formats. + */ +enum { + UNKNOWN, + MRw, + IMlw, + IMw, + IR, + OA, + AO, + MS, + SM, + Mv, + Mw, + M, /* register or memory */ + Mb, /* register or memory, always byte sized */ + MO, /* memory only (no registers) */ + PREF, + SWAPGS, + R, + RA, + SEG, + MR, + RM, + IA, + MA, + SD, + AD, + SA, + D, + INM, + SO, + BD, + I, + P, + V, + DSHIFT, /* for double shift that has an 8-bit immediate */ + U, + OVERRIDE, + NORM, /* instructions w/o ModR/M byte, no memory access */ + IMPLMEM, /* instructions w/o ModR/M byte, implicit mem access */ + O, /* for call */ + JTAB, /* jump table */ + IMUL, /* for 186 iimul instr */ + CBW, /* so data16 can be evaluated for cbw and variants */ + MvI, /* for 186 logicals */ + ENTER, /* for 186 enter instr */ + RMw, /* for 286 arpl instr */ + Ib, /* for push immediate byte */ + F, /* for 287 instructions */ + FF, /* for 287 instructions */ + FFC, /* for 287 instructions */ + DM, /* 16-bit data */ + AM, /* 16-bit addr */ + LSEG, /* for 3-bit seg reg encoding */ + MIb, /* for 386 logicals */ + SREG, /* for 386 special registers */ + PREFIX, /* a REP instruction prefix */ + LOCK, /* a LOCK instruction prefix */ + INT3, /* The int 3 instruction, which has a fake operand */ + INTx, /* The normal int instruction, with explicit int num */ + DSHIFTcl, /* for double shift that implicitly uses %cl */ + CWD, /* so data16 can be evaluated for cwd and variants */ + RET, /* single immediate 16-bit operand */ + MOVZ, /* for movs and movz, with different size operands */ + XADDB, /* for xaddb */ + MOVSXZ, /* AMD64 mov sign extend 32 to 64 bit instruction */ + +/* + * MMX/SIMD addressing modes. + */ + + MMO, /* Prefixable MMX/SIMD-Int mm/mem -> mm */ + MMOIMPL, /* Prefixable MMX/SIMD-Int mm -> mm (mem) */ + MMO3P, /* Prefixable MMX/SIMD-Int mm -> r32,imm8 */ + MMOM3, /* Prefixable MMX/SIMD-Int mm -> r32 */ + MMOS, /* Prefixable MMX/SIMD-Int mm -> mm/mem */ + MMOMS, /* Prefixable MMX/SIMD-Int mm -> mem */ + MMOPM, /* MMX/SIMD-Int mm/mem -> mm,imm8 */ + MMOPRM, /* Prefixable MMX/SIMD-Int r32/mem -> mm,imm8 */ + MMOSH, /* Prefixable MMX mm,imm8 */ + MM, /* MMX/SIMD-Int mm/mem -> mm */ + MMS, /* MMX/SIMD-Int mm -> mm/mem */ + MMSH, /* MMX mm,imm8 */ + XMMO, /* Prefixable SIMD xmm/mem -> xmm */ + XMMOS, /* Prefixable SIMD xmm -> xmm/mem */ + XMMOPM, /* Prefixable SIMD xmm/mem w/to xmm,imm8 */ + XMMOMX, /* Prefixable SIMD mm/mem -> xmm */ + XMMOX3, /* Prefixable SIMD xmm -> r32 */ + XMMOXMM, /* Prefixable SIMD xmm/mem -> mm */ + XMMOM, /* Prefixable SIMD xmm -> mem */ + XMMOMS, /* Prefixable SIMD mem -> xmm */ + XMM, /* SIMD xmm/mem -> xmm */ + XMMXIMPL, /* SIMD xmm -> xmm (mem) */ + XMM3P, /* SIMD xmm -> r32,imm8 */ + XMMP, /* SIMD xmm/mem w/to xmm,imm8 */ + XMMPRM, /* SIMD r32/mem -> xmm,imm8 */ + XMMS, /* SIMD xmm -> xmm/mem */ + XMMM, /* SIMD mem -> xmm */ + XMMMS, /* SIMD xmm -> mem */ + XMM3MX, /* SIMD r32/mem -> xmm */ + XMM3MXS, /* SIMD xmm -> r32/mem */ + XMMSH, /* SIMD xmm,imm8 */ + XMMXM3, /* SIMD xmm/mem -> r32 */ + XMMX3, /* SIMD xmm -> r32 */ + XMMXMM, /* SIMD xmm/mem -> mm */ + XMMMX, /* SIMD mm -> xmm */ + XMMXM, /* SIMD xmm -> mm */ + XMMFENCE, /* SIMD lfence or mfence */ + XMMSFNC /* SIMD sfence (none or mem) */ +}; + +#define FILL 0x90 /* Fill byte used for alignment (nop) */ + +/* +** Register numbers for the i386 +*/ +#define EAX_REGNO 0 +#define ECX_REGNO 1 +#define EDX_REGNO 2 +#define EBX_REGNO 3 +#define ESP_REGNO 4 +#define EBP_REGNO 5 +#define ESI_REGNO 6 +#define EDI_REGNO 7 + +/* + * modes for immediate values + */ +#define MODE_NONE 0 +#define MODE_IPREL 1 /* signed IP relative value */ +#define MODE_SIGNED 2 /* sign extended immediate */ +#define MODE_IMPLIED 3 /* constant value implied from opcode */ +#define MODE_OFFSET 4 /* offset part of an address */ + +/* + * The letters used in these macros are: + * IND - indirect to another to another table + * "T" - means to Terminate indirections (this is the final opcode) + * "S" - means "operand length suffix required" + * "NS" - means "no suffix" which is the operand length suffix of the opcode + * "Z" - means instruction size arg required + * "u" - means the opcode is invalid in IA32 but valid in amd64 + * "x" - means the opcode is invalid in amd64, but not IA32 + * "y" - means the operand size is always 64 bits in 64 bit mode + * "p" - means push/pop stack operation + */ + +#if defined(DIS_TEXT) && defined(DIS_MEM) +#define IND(table) {table, 0, "", 0, 0, 0, 0, 0, 0} +#define INDx(table) {table, 0, "", 0, 0, 1, 0, 0, 0} +#define TNS(name, amode) {TERM, amode, name, 0, 0, 0, 0, 0, 0} +#define TNSu(name, amode) {TERM, amode, name, 0, 0, 0, 0, 1, 0} +#define TNSx(name, amode) {TERM, amode, name, 0, 0, 1, 0, 0, 0} +#define TNSy(name, amode) {TERM, amode, name, 0, 0, 0, 1, 0, 0} +#define TNSyp(name, amode) {TERM, amode, name, 0, 0, 0, 1, 0, 1} +#define TNSZ(name, amode, sz) {TERM, amode, name, 0, sz, 0, 0, 0, 0} +#define TNSZy(name, amode, sz) {TERM, amode, name, 0, sz, 0, 1, 0, 0} +#define TS(name, amode) {TERM, amode, name, 1, 0, 0, 0, 0, 0} +#define TSx(name, amode) {TERM, amode, name, 1, 0, 1, 0, 0, 0} +#define TSy(name, amode) {TERM, amode, name, 1, 0, 0, 1, 0, 0} +#define TSp(name, amode) {TERM, amode, name, 1, 0, 0, 0, 0, 1} +#define TSZ(name, amode, sz) {TERM, amode, name, 1, sz, 0, 0, 0, 0} +#define TSZx(name, amode, sz) {TERM, amode, name, 1, sz, 1, 0, 0, 0} +#define TSZy(name, amode, sz) {TERM, amode, name, 1, sz, 0, 1, 0, 0} +#define INVALID {TERM, UNKNOWN, "", 0, 0, 0, 0, 0} +#elif defined(DIS_TEXT) +#define IND(table) {table, 0, "", 0, 0, 0, 0, 0} +#define INDx(table) {table, 0, "", 0, 1, 0, 0, 0} +#define TNS(name, amode) {TERM, amode, name, 0, 0, 0, 0, 0} +#define TNSu(name, amode) {TERM, amode, name, 0, 0, 0, 1, 0} +#define TNSx(name, amode) {TERM, amode, name, 0, 1, 0, 0, 0} +#define TNSy(name, amode) {TERM, amode, name, 0, 0, 1, 0, 0} +#define TNSyp(name, amode) {TERM, amode, name, 0, 0, 1, 0, 1} +#define TNSZ(name, amode, sz) {TERM, amode, name, 0, 0, 0, 0, 0} +#define TNSZy(name, amode, sz) {TERM, amode, name, 0, 0, 1, 0, 0} +#define TS(name, amode) {TERM, amode, name, 1, 0, 0, 0, 0} +#define TSx(name, amode) {TERM, amode, name, 1, 1, 0, 0, 0} +#define TSy(name, amode) {TERM, amode, name, 1, 0, 1, 0, 0} +#define TSp(name, amode) {TERM, amode, name, 1, 0, 0, 0, 1} +#define TSZ(name, amode, sz) {TERM, amode, name, 1, 0, 0, 0, 0} +#define TSZx(name, amode, sz) {TERM, amode, name, 1, 1, 0, 0, 0} +#define TSZy(name, amode, sz) {TERM, amode, name, 1, 0, 1, 0, 0} +#define INVALID {TERM, UNKNOWN, "", 0, 0, 0, 0, 0} +#elif defined(DIS_MEM) +#define IND(table) {table, 0, 0, 0, 0, 0, 0} +#define INDx(table) {table, 0, 0, 1, 0, 0, 0} +#define TNS(name, amode) {TERM, amode, 0, 0, 0, 0, 0} +#define TNSu(name, amode) {TERM, amode, 0, 0, 0, 1, 0} +#define TNSy(name, amode) {TERM, amode, 0, 0, 1, 0, 0} +#define TNSyp(name, amode) {TERM, amode, 0, 0, 1, 0, 1} +#define TNSx(name, amode) {TERM, amode, 0, 1, 0, 0, 0} +#define TNSZ(name, amode, sz) {TERM, amode, sz, 0, 0, 0, 0} +#define TNSZy(name, amode, sz) {TERM, amode, sz, 0, 1, 0, 0} +#define TS(name, amode) {TERM, amode, 0, 0, 0, 0, 0} +#define TSx(name, amode) {TERM, amode, 0, 1, 0, 0, 0} +#define TSy(name, amode) {TERM, amode, 0, 0, 1, 0, 0} +#define TSp(name, amode) {TERM, amode, 0, 0, 0, 0, 1} +#define TSZ(name, amode, sz) {TERM, amode, sz, 0, 0, 0, 0} +#define TSZx(name, amode, sz) {TERM, amode, sz, 1, 0, 0, 0} +#define TSZy(name, amode, sz) {TERM, amode, sz, 0, 1, 0, 0} +#define INVALID {TERM, UNKNOWN, 0, 0, 0, 0, 0} +#else +#define IND(table) {table[0], 0, 0, 0, 0, 0} +#define INDx(table) {table[0], 0, 1, 0, 0, 0} +#define TNS(name, amode) {TERM, amode, 0, 0, 0, 0} +#define TNSu(name, amode) {TERM, amode, 0, 0, 1, 0} +#define TNSy(name, amode) {TERM, amode, 0, 1, 0, 0} +#define TNSyp(name, amode) {TERM, amode, 0, 1, 0, 1} +#define TNSx(name, amode) {TERM, amode, 1, 0, 0, 0} +#define TNSZ(name, amode, sz) {TERM, amode, 0, 0, 0, 0} +#define TNSZy(name, amode, sz) {TERM, amode, 0, 1, 0, 0} +#define TS(name, amode) {TERM, amode, 0, 0, 0, 0} +#define TSx(name, amode) {TERM, amode, 1, 0, 0, 0} +#define TSy(name, amode) {TERM, amode, 0, 1, 0, 0} +#define TSp(name, amode) {TERM, amode, 0, 0, 0, 1} +#define TSZ(name, amode, sz) {TERM, amode, 0, 0, 0, 0} +#define TSZx(name, amode, sz) {TERM, amode, 1, 0, 0, 0} +#define TSZy(name, amode, sz) {TERM, amode, 0, 1, 0, 0} +#define INVALID {TERM, UNKNOWN, 0, 0, 0, 0} +#endif + +#ifdef DIS_TEXT +/* + * this decodes the r_m field for mode's 0, 1, 2 in 16 bit mode + */ +const char *const dis_addr16[3][8] = { +"(%bx,%si)", "(%bx,%di)", "(%bp,%si)", "(%bp,%di)", "(%si)", "(%di)", "", + "(%bx)", +"(%bx,%si)", "(%bx,%di)", "(%bp,%si)", "(%bp,%di)", "(%si)", "(%di", "(%bp)", + "(%bx)", +"(%bx,%si)", "(%bx,%di)", "(%bp,%si)", "(%bp,%di)", "(%si)", "(%di)", "(%bp)", + "(%bx)", +}; + + +/* + * This decodes 32 bit addressing mode r_m field for modes 0, 1, 2 + */ +const char *const dis_addr32_mode0[16] = { + "(%eax)", "(%ecx)", "(%edx)", "(%ebx)", "", "", "(%esi)", "(%edi)", + "(%r8d)", "(%r9d)", "(%r10d)", "(%r11d)", "", "", "(%r14d)", "(%r15d)" +}; + +const char *const dis_addr32_mode12[16] = { + "(%eax)", "(%ecx)", "(%edx)", "(%ebx)", "", "(%ebp)", "(%esi)", "(%edi)", + "(%r8d)", "(%r9d)", "(%r10d)", "(%r11d)", "", "(%r13d)", "(%r14d)", "(%r15d)" +}; + +/* + * This decodes 64 bit addressing mode r_m field for modes 0, 1, 2 + */ +const char *const dis_addr64_mode0[16] = { + "(%rax)", "(%rcx)", "(%rdx)", "(%rbx)", "", "(%rip)", "(%rsi)", "(%rdi)", + "(%r8)", "(%r9)", "(%r10)", "(%r11)", "(%r12)", "(%rip)", "(%r14)", "(%r15)" +}; +const char *const dis_addr64_mode12[16] = { + "(%rax)", "(%rcx)", "(%rdx)", "(%rbx)", "", "(%rbp)", "(%rsi)", "(%rdi)", + "(%r8)", "(%r9)", "(%r10)", "(%r11)", "(%r12)", "(%r13)", "(%r14)", "(%r15)" +}; + +/* + * decode for scale from SIB byte + */ +const char *const dis_scale_factor[4] = { ")", ",2)", ",4)", ",8)" }; + +/* + * register decoding for normal references to registers (ie. not addressing) + */ +const char *const dis_REG8[16] = { + "%al", "%cl", "%dl", "%bl", "%ah", "%ch", "%dh", "%bh", + "%r8b", "%r9b", "%r10b", "%r11b", "%r12b", "%r13b", "%r14b", "%r15b" +}; + +const char *const dis_REG8_REX[16] = { + "%al", "%cl", "%dl", "%bl", "%spl", "%bpl", "%sil", "%dil", + "%r8b", "%r9b", "%r10b", "%r11b", "%r12b", "%r13b", "%r14b", "%r15b" +}; + +const char *const dis_REG16[16] = { + "%ax", "%cx", "%dx", "%bx", "%sp", "%bp", "%si", "%di", + "%r8w", "%r9w", "%r10w", "%r11w", "%r12w", "%r13w", "%r14w", "%r15w" +}; + +const char *const dis_REG32[16] = { + "%eax", "%ecx", "%edx", "%ebx", "%esp", "%ebp", "%esi", "%edi", + "%r8d", "%r9d", "%r10d", "%r11d", "%r12d", "%r13d", "%r14d", "%r15d" +}; + +const char *const dis_REG64[16] = { + "%rax", "%rcx", "%rdx", "%rbx", "%rsp", "%rbp", "%rsi", "%rdi", + "%r8", "%r9", "%r10", "%r11", "%r12", "%r13", "%r14", "%r15" +}; + +const char *const dis_DEBUGREG[16] = { + "%db0", "%db1", "%db2", "%db3", "%db4", "%db5", "%db6", "%db7", + "%db8", "%db9", "%db10", "%db11", "%db12", "%db13", "%db14", "%db15" +}; + +const char *const dis_CONTROLREG[16] = { + "%cr0", "%cr1", "%cr2", "%cr3", "%cr4", "%cr5?", "%cr6?", "%cr7?", + "%cr8", "%cr9?", "%cr10?", "%cr11?", "%cr12?", "%cr13?", "%cr14?", "%cr15?" +}; + +const char *const dis_TESTREG[16] = { + "%tr0?", "%tr1?", "%tr2?", "%tr3", "%tr4", "%tr5", "%tr6", "%tr7", + "%tr0?", "%tr1?", "%tr2?", "%tr3", "%tr4", "%tr5", "%tr6", "%tr7" +}; + +const char *const dis_MMREG[16] = { + "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7", + "%mm0", "%mm1", "%mm2", "%mm3", "%mm4", "%mm5", "%mm6", "%mm7" +}; + +const char *const dis_XMMREG[16] = { + "%xmm0", "%xmm1", "%xmm2", "%xmm3", "%xmm4", "%xmm5", "%xmm6", "%xmm7", + "%xmm8", "%xmm9", "%xmm10", "%xmm11", "%xmm12", "%xmm13", "%xmm14", "%xmm15" +}; + +const char *const dis_SEGREG[16] = { + "%es", "%cs", "%ss", "%ds", "%fs", "%gs", "<reserved>", "<reserved>", + "%es", "%cs", "%ss", "%ds", "%fs", "%gs", "<reserved>", "<reserved>" +}; + +/* + * SIMD predicate suffixes + */ +const char *const dis_PREDSUFFIX[8] = { + "eq", "lt", "le", "unord", "neq", "nlt", "nle", "ord" +}; + + + +#endif /* DIS_TEXT */ + + + + +/* + * "decode table" for 64 bit mode MOVSXD instruction (opcode 0x63) + */ +const instable_t dis_opMOVSLD = TNS("movslq",MOVSXZ); + +/* + * "decode table" for pause and clflush instructions + */ +const instable_t dis_opPause = TNS("pause", NORM); + +/* + * Decode table for 0x0F00 opcodes + */ +const instable_t dis_op0F00[8] = { + +/* [0] */ TNS("sldt",M), TNS("str",M), TNSy("lldt",M), TNSy("ltr",M), +/* [4] */ TNSZ("verr",M,2), TNSZ("verw",M,2), INVALID, INVALID, +}; + + +/* + * Decode table for 0x0F01 opcodes + */ +const instable_t dis_op0F01[8] = { + +/* [0] */ TNSZ("sgdt",MO,6), TNSZ("sidt",MO,6), TNSZ("lgdt",MO,6), TNSZ("lidt",MO,6), +/* [4] */ TNSZ("smsw",M,2), INVALID, TNSZ("lmsw",M,2), TNS("invlpg",SWAPGS), +}; + +/* + * Decode table for 0x0F18 opcodes -- SIMD prefetch + */ +const instable_t dis_op0F18[8] = { + +/* [0] */ TNS("prefetchnta",PREF),TNS("prefetcht0",PREF), TNS("prefetcht1",PREF), TNS("prefetcht2",PREF), +/* [4] */ INVALID, INVALID, INVALID, INVALID, +}; + +/* + * Decode table for 0x0FAE opcodes -- SIMD state save/restore + */ +const instable_t dis_op0FAE[8] = { +/* [0] */ TNSZ("fxsave",M,512), TNSZ("fxrstor",M,512), TNS("ldmxcsr",M), TNS("stmxcsr",M), +/* [4] */ INVALID, TNS("lfence",XMMFENCE), TNS("mfence",XMMFENCE), TNS("sfence",XMMSFNC), +}; + +/* + * Decode table for 0x0FBA opcodes + */ + +const instable_t dis_op0FBA[8] = { + +/* [0] */ INVALID, INVALID, INVALID, INVALID, +/* [4] */ TS("bt",MIb), TS("bts",MIb), TS("btr",MIb), TS("btc",MIb), +}; + +/* + * Decode table for 0x0FC7 opcode + */ + +const instable_t dis_op0FC7[8] = { + +/* [0] */ INVALID, TNS("cmpxchg8b",M), INVALID, INVALID, +/* [4] */ INVALID, INVALID, INVALID, INVALID, +}; + + +/* + * Decode table for 0x0FC8 opcode -- 486 bswap instruction + * + *bit pattern: 0000 1111 1100 1reg + */ +const instable_t dis_op0FC8[4] = { +/* [0] */ TNS("bswap",R), INVALID, INVALID, INVALID, +}; + +/* + * Decode table for 0x0F71, 0x0F72, and 0x0F73 opcodes -- MMX instructions + */ +const instable_t dis_op0F7123[4][8] = { +{ +/* [70].0 */ INVALID, INVALID, INVALID, INVALID, +/* .4 */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [71].0 */ INVALID, INVALID, TNS("psrlw",MMOSH), INVALID, +/* .4 */ TNS("psraw",MMOSH), INVALID, TNS("psllw",MMOSH), INVALID, +}, { +/* [72].0 */ INVALID, INVALID, TNS("psrld",MMOSH), INVALID, +/* .4 */ TNS("psrad",MMOSH), INVALID, TNS("pslld",MMOSH), INVALID, +}, { +/* [73].0 */ INVALID, INVALID, TNS("psrlq",MMOSH), TNS("INVALID",MMOSH), +/* .4 */ INVALID, INVALID, TNS("psllq",MMOSH), TNS("INVALID",MMOSH), +} }; + +/* + * Decode table for SIMD extensions to above 0x0F71-0x0F73 opcodes. + */ +const instable_t dis_opSIMD7123[32] = { +/* [70].0 */ INVALID, INVALID, INVALID, INVALID, +/* .4 */ INVALID, INVALID, INVALID, INVALID, + +/* [71].0 */ INVALID, INVALID, TNS("psrlw",XMMSH), INVALID, +/* .4 */ TNS("psraw",XMMSH), INVALID, TNS("psllw",XMMSH), INVALID, + +/* [72].0 */ INVALID, INVALID, TNS("psrld",XMMSH), INVALID, +/* .4 */ TNS("psrad",XMMSH), INVALID, TNS("pslld",XMMSH), INVALID, + +/* [73].0 */ INVALID, INVALID, TNS("psrlq",XMMSH), TNS("psrldq",XMMSH), +/* .4 */ INVALID, INVALID, TNS("psllq",XMMSH), TNS("pslldq",XMMSH), +}; + +/* + * SIMD instructions have been wedged into the existing IA32 instruction + * set through the use of prefixes. That is, while 0xf0 0x58 may be + * addps, 0xf3 0xf0 0x58 (literally, repz addps) is a completely different + * instruction - addss. At present, three prefixes have been coopted in + * this manner - address size (0x66), repnz (0xf2) and repz (0xf3). The + * following tables are used to provide the prefixed instruction names. + * The arrays are sparse, but they're fast. + */ + +/* + * Decode table for SIMD instructions with the address size (0x66) prefix. + */ +const instable_t dis_opSIMDdata16[256] = { +/* [00] */ INVALID, INVALID, INVALID, INVALID, +/* [04] */ INVALID, INVALID, INVALID, INVALID, +/* [08] */ INVALID, INVALID, INVALID, INVALID, +/* [0C] */ INVALID, INVALID, INVALID, INVALID, + +/* [10] */ TNSZ("movupd",XMM,16), TNSZ("movupd",XMMS,16), TNSZ("movlpd",XMMM,8), TNSZ("movlpd",XMMMS,8), +/* [14] */ TNSZ("unpcklpd",XMM,16),TNSZ("unpckhpd",XMM,16),TNSZ("movhpd",XMMM,8), TNSZ("movhpd",XMMMS,8), +/* [18] */ INVALID, INVALID, INVALID, INVALID, +/* [1C] */ INVALID, INVALID, INVALID, INVALID, + +/* [20] */ INVALID, INVALID, INVALID, INVALID, +/* [24] */ INVALID, INVALID, INVALID, INVALID, +/* [28] */ TNSZ("movapd",XMM,16), TNSZ("movapd",XMMS,16), TNSZ("cvtpi2pd",XMMOMX,8),TNSZ("movntpd",XMMOMS,16), +/* [2C] */ TNSZ("cvttpd2pi",XMMXMM,16),TNSZ("cvtpd2pi",XMMXMM,16),TNSZ("ucomisd",XMM,8),TNSZ("comisd",XMM,8), + +/* [30] */ INVALID, INVALID, INVALID, INVALID, +/* [34] */ INVALID, INVALID, INVALID, INVALID, +/* [38] */ INVALID, INVALID, INVALID, INVALID, +/* [3C] */ INVALID, INVALID, INVALID, INVALID, + +/* [40] */ INVALID, INVALID, INVALID, INVALID, +/* [44] */ INVALID, INVALID, INVALID, INVALID, +/* [48] */ INVALID, INVALID, INVALID, INVALID, +/* [4C] */ INVALID, INVALID, INVALID, INVALID, + +/* [50] */ TNS("movmskpd",XMMOX3), TNSZ("sqrtpd",XMM,16), INVALID, INVALID, +/* [54] */ TNSZ("andpd",XMM,16), TNSZ("andnpd",XMM,16), TNSZ("orpd",XMM,16), TNSZ("xorpd",XMM,16), +/* [58] */ TNSZ("addpd",XMM,16), TNSZ("mulpd",XMM,16), TNSZ("cvtpd2ps",XMM,16),TNSZ("cvtps2dq",XMM,16), +/* [5C] */ TNSZ("subpd",XMM,16), TNSZ("minpd",XMM,16), TNSZ("divpd",XMM,16), TNSZ("maxpd",XMM,16), + +/* [60] */ TNSZ("punpcklbw",XMM,16),TNSZ("punpcklwd",XMM,16),TNSZ("punpckldq",XMM,16),TNSZ("packsswb",XMM,16), +/* [64] */ TNSZ("pcmpgtb",XMM,16), TNSZ("pcmpgtw",XMM,16), TNSZ("pcmpgtd",XMM,16), TNSZ("packuswb",XMM,16), +/* [68] */ TNSZ("punpckhbw",XMM,16),TNSZ("punpckhwd",XMM,16),TNSZ("punpckhdq",XMM,16),TNSZ("packssdw",XMM,16), +/* [6C] */ TNSZ("punpcklqdq",XMM,16),TNSZ("punpckhqdq",XMM,16),TNSZ("movd",XMM3MX,4),TNSZ("movdqa",XMM,16), + +/* [70] */ TNSZ("pshufd",XMMP,16), INVALID, INVALID, INVALID, +/* [74] */ TNSZ("pcmpeqb",XMM,16), TNSZ("pcmpeqw",XMM,16), TNSZ("pcmpeqd",XMM,16), INVALID, +/* [78] */ INVALID, INVALID, INVALID, INVALID, +/* [7C] */ INVALID, INVALID, TNSZ("movd",XMM3MXS,4), TNSZ("movdqa",XMMS,16), + +/* [80] */ INVALID, INVALID, INVALID, INVALID, +/* [84] */ INVALID, INVALID, INVALID, INVALID, +/* [88] */ INVALID, INVALID, INVALID, INVALID, +/* [8C] */ INVALID, INVALID, INVALID, INVALID, + +/* [90] */ INVALID, INVALID, INVALID, INVALID, +/* [94] */ INVALID, INVALID, INVALID, INVALID, +/* [98] */ INVALID, INVALID, INVALID, INVALID, +/* [9C] */ INVALID, INVALID, INVALID, INVALID, + +/* [A0] */ INVALID, INVALID, INVALID, INVALID, +/* [A4] */ INVALID, INVALID, INVALID, INVALID, +/* [A8] */ INVALID, INVALID, INVALID, INVALID, +/* [AC] */ INVALID, INVALID, INVALID, INVALID, + +/* [B0] */ INVALID, INVALID, INVALID, INVALID, +/* [B4] */ INVALID, INVALID, INVALID, INVALID, +/* [B8] */ INVALID, INVALID, INVALID, INVALID, +/* [BC] */ INVALID, INVALID, INVALID, INVALID, + +/* [C0] */ INVALID, INVALID, TNSZ("cmppd",XMMP,16), INVALID, +/* [C4] */ TNSZ("pinsrw",XMMPRM,2),TNS("pextrw",XMM3P), TNSZ("shufpd",XMMP,16), INVALID, +/* [C8] */ INVALID, INVALID, INVALID, INVALID, +/* [CC] */ INVALID, INVALID, INVALID, INVALID, + +/* [D0] */ INVALID, TNSZ("psrlw",XMM,16), TNSZ("psrld",XMM,16), TNSZ("psrlq",XMM,16), +/* [D4] */ TNSZ("paddq",XMM,16), TNSZ("pmullw",XMM,16), TNSZ("movq",XMMS,8), TNS("pmovmskb",XMMX3), +/* [D8] */ TNSZ("psubusb",XMM,16), TNSZ("psubusw",XMM,16), TNSZ("pminub",XMM,16), TNSZ("pand",XMM,16), +/* [DC] */ TNSZ("paddusb",XMM,16), TNSZ("paddusw",XMM,16), TNSZ("pmaxub",XMM,16), TNSZ("pandn",XMM,16), + +/* [E0] */ TNSZ("pavgb",XMM,16), TNSZ("psraw",XMM,16), TNSZ("psrad",XMM,16), TNSZ("pavgw",XMM,16), +/* [E4] */ TNSZ("pmulhuw",XMM,16), TNSZ("pmulhw",XMM,16), TNSZ("cvttpd2dq",XMM,16),TNSZ("movntdq",XMMS,16), +/* [E8] */ TNSZ("psubsb",XMM,16), TNSZ("psubsw",XMM,16), TNSZ("pminsw",XMM,16), TNSZ("por",XMM,16), +/* [EC] */ TNSZ("paddsb",XMM,16), TNSZ("paddsw",XMM,16), TNSZ("pmaxsw",XMM,16), TNSZ("pxor",XMM,16), + +/* [F0] */ INVALID, TNSZ("psllw",XMM,16), TNSZ("pslld",XMM,16), TNSZ("psllq",XMM,16), +/* [F4] */ TNSZ("pmuludq",XMM,16), TNSZ("pmaddwd",XMM,16), TNSZ("psadbw",XMM,16), TNSZ("maskmovdqu", XMMXIMPL,16), +/* [F8] */ TNSZ("psubb",XMM,16), TNSZ("psubw",XMM,16), TNSZ("psubd",XMM,16), TNSZ("psubq",XMM,16), +/* [FC] */ TNSZ("paddb",XMM,16), TNSZ("paddw",XMM,16), TNSZ("paddd",XMM,16), INVALID, +}; + +/* + * Decode table for SIMD instructions with the repnz (0xf2) prefix. + */ +const instable_t dis_opSIMDrepnz[256] = { +/* [00] */ INVALID, INVALID, INVALID, INVALID, +/* [04] */ INVALID, INVALID, INVALID, INVALID, +/* [08] */ INVALID, INVALID, INVALID, INVALID, +/* [0C] */ INVALID, INVALID, INVALID, INVALID, + +/* [10] */ TNSZ("movsd",XMM,8), TNSZ("movsd",XMMS,8), INVALID, INVALID, +/* [14] */ INVALID, INVALID, INVALID, INVALID, +/* [18] */ INVALID, INVALID, INVALID, INVALID, +/* [1C] */ INVALID, INVALID, INVALID, INVALID, + +/* [20] */ INVALID, INVALID, INVALID, INVALID, +/* [24] */ INVALID, INVALID, INVALID, INVALID, +/* [28] */ INVALID, INVALID, TNSZ("cvtsi2sd",XMM3MX,4),INVALID, +/* [2C] */ TNSZ("cvttsd2si",XMMXM3,8),TNSZ("cvtsd2si",XMMXM3,8),INVALID, INVALID, + +/* [30] */ INVALID, INVALID, INVALID, INVALID, +/* [34] */ INVALID, INVALID, INVALID, INVALID, +/* [38] */ INVALID, INVALID, INVALID, INVALID, +/* [3C] */ INVALID, INVALID, INVALID, INVALID, + +/* [40] */ INVALID, INVALID, INVALID, INVALID, +/* [44] */ INVALID, INVALID, INVALID, INVALID, +/* [48] */ INVALID, INVALID, INVALID, INVALID, +/* [4C] */ INVALID, INVALID, INVALID, INVALID, + +/* [50] */ INVALID, TNSZ("sqrtsd",XMM,8), INVALID, INVALID, +/* [54] */ INVALID, INVALID, INVALID, INVALID, +/* [58] */ TNSZ("addsd",XMM,8), TNSZ("mulsd",XMM,8), TNSZ("cvtsd2ss",XMM,8), INVALID, +/* [5C] */ TNSZ("subsd",XMM,8), TNSZ("minsd",XMM,8), TNSZ("divsd",XMM,8), TNSZ("maxsd",XMM,8), + +/* [60] */ INVALID, INVALID, INVALID, INVALID, +/* [64] */ INVALID, INVALID, INVALID, INVALID, +/* [68] */ INVALID, INVALID, INVALID, INVALID, +/* [6C] */ INVALID, INVALID, INVALID, INVALID, + +/* [70] */ TNSZ("pshuflw",XMMP,16),INVALID, INVALID, INVALID, +/* [74] */ INVALID, INVALID, INVALID, INVALID, +/* [78] */ INVALID, INVALID, INVALID, INVALID, +/* [7C] */ INVALID, INVALID, INVALID, INVALID, + +/* [80] */ INVALID, INVALID, INVALID, INVALID, +/* [84] */ INVALID, INVALID, INVALID, INVALID, +/* [88] */ INVALID, INVALID, INVALID, INVALID, +/* [0C] */ INVALID, INVALID, INVALID, INVALID, + +/* [90] */ INVALID, INVALID, INVALID, INVALID, +/* [94] */ INVALID, INVALID, INVALID, INVALID, +/* [98] */ INVALID, INVALID, INVALID, INVALID, +/* [9C] */ INVALID, INVALID, INVALID, INVALID, + +/* [A0] */ INVALID, INVALID, INVALID, INVALID, +/* [A4] */ INVALID, INVALID, INVALID, INVALID, +/* [A8] */ INVALID, INVALID, INVALID, INVALID, +/* [AC] */ INVALID, INVALID, INVALID, INVALID, + +/* [B0] */ INVALID, INVALID, INVALID, INVALID, +/* [B4] */ INVALID, INVALID, INVALID, INVALID, +/* [B8] */ INVALID, INVALID, INVALID, INVALID, +/* [BC] */ INVALID, INVALID, INVALID, INVALID, + +/* [C0] */ INVALID, INVALID, TNSZ("cmpsd",XMMP,8), INVALID, +/* [C4] */ INVALID, INVALID, INVALID, INVALID, +/* [C8] */ INVALID, INVALID, INVALID, INVALID, +/* [CC] */ INVALID, INVALID, INVALID, INVALID, + +/* [D0] */ INVALID, INVALID, INVALID, INVALID, +/* [D4] */ INVALID, INVALID, TNS("movdq2q",XMMXM), INVALID, +/* [D8] */ INVALID, INVALID, INVALID, INVALID, +/* [DC] */ INVALID, INVALID, INVALID, INVALID, + +/* [E0] */ INVALID, INVALID, INVALID, INVALID, +/* [E4] */ INVALID, INVALID, TNSZ("cvtpd2dq",XMM,16),INVALID, +/* [E8] */ INVALID, INVALID, INVALID, INVALID, +/* [EC] */ INVALID, INVALID, INVALID, INVALID, + +/* [F0] */ INVALID, INVALID, INVALID, INVALID, +/* [F4] */ INVALID, INVALID, INVALID, INVALID, +/* [F8] */ INVALID, INVALID, INVALID, INVALID, +/* [FC] */ INVALID, INVALID, INVALID, INVALID, +}; + +/* + * Decode table for SIMD instructions with the repz (0xf3) prefix. + */ +const instable_t dis_opSIMDrepz[256] = { +/* [00] */ INVALID, INVALID, INVALID, INVALID, +/* [04] */ INVALID, INVALID, INVALID, INVALID, +/* [08] */ INVALID, INVALID, INVALID, INVALID, +/* [0C] */ INVALID, INVALID, INVALID, INVALID, + +/* [10] */ TNSZ("movss",XMM,4), TNSZ("movss",XMMS,4), INVALID, INVALID, +/* [14] */ INVALID, INVALID, INVALID, INVALID, +/* [18] */ INVALID, INVALID, INVALID, INVALID, +/* [1C] */ INVALID, INVALID, INVALID, INVALID, + +/* [20] */ INVALID, INVALID, INVALID, INVALID, +/* [24] */ INVALID, INVALID, INVALID, INVALID, +/* [28] */ INVALID, INVALID, TNSZ("cvtsi2ss",XMM3MX,4),INVALID, +/* [2C] */ TNSZ("cvttss2si",XMMXM3,4),TNSZ("cvtss2si",XMMXM3,4),INVALID, INVALID, + +/* [30] */ INVALID, INVALID, INVALID, INVALID, +/* [34] */ INVALID, INVALID, INVALID, INVALID, +/* [38] */ INVALID, INVALID, INVALID, INVALID, +/* [3C] */ INVALID, INVALID, INVALID, INVALID, + +/* [40] */ INVALID, INVALID, INVALID, INVALID, +/* [44] */ INVALID, INVALID, INVALID, INVALID, +/* [48] */ INVALID, INVALID, INVALID, INVALID, +/* [4C] */ INVALID, INVALID, INVALID, INVALID, + +/* [50] */ INVALID, TNSZ("sqrtss",XMM,4), TNSZ("rsqrtss",XMM,4), TNSZ("rcpss",XMM,4), +/* [54] */ INVALID, INVALID, INVALID, INVALID, +/* [58] */ TNSZ("addss",XMM,4), TNSZ("mulss",XMM,4), TNSZ("cvtss2sd",XMM,4), TNSZ("cvttps2dq",XMM,16), +/* [5C] */ TNSZ("subss",XMM,4), TNSZ("minss",XMM,4), TNSZ("divss",XMM,4), TNSZ("maxss",XMM,4), + +/* [60] */ INVALID, INVALID, INVALID, INVALID, +/* [64] */ INVALID, INVALID, INVALID, INVALID, +/* [68] */ INVALID, INVALID, INVALID, INVALID, +/* [6C] */ INVALID, INVALID, INVALID, TNSZ("movdqu",XMM,16), + +/* [70] */ TNSZ("pshufhw",XMMP,16),INVALID, INVALID, INVALID, +/* [74] */ INVALID, INVALID, INVALID, INVALID, +/* [78] */ INVALID, INVALID, INVALID, INVALID, +/* [7C] */ INVALID, INVALID, TNSZ("movq",XMM,8), TNSZ("movdqu",XMMS,16), + +/* [80] */ INVALID, INVALID, INVALID, INVALID, +/* [84] */ INVALID, INVALID, INVALID, INVALID, +/* [88] */ INVALID, INVALID, INVALID, INVALID, +/* [0C] */ INVALID, INVALID, INVALID, INVALID, + +/* [90] */ INVALID, INVALID, INVALID, INVALID, +/* [94] */ INVALID, INVALID, INVALID, INVALID, +/* [98] */ INVALID, INVALID, INVALID, INVALID, +/* [9C] */ INVALID, INVALID, INVALID, INVALID, + +/* [A0] */ INVALID, INVALID, INVALID, INVALID, +/* [A4] */ INVALID, INVALID, INVALID, INVALID, +/* [A8] */ INVALID, INVALID, INVALID, INVALID, +/* [AC] */ INVALID, INVALID, INVALID, INVALID, + +/* [B0] */ INVALID, INVALID, INVALID, INVALID, +/* [B4] */ INVALID, INVALID, INVALID, INVALID, +/* [B8] */ INVALID, INVALID, INVALID, INVALID, +/* [BC] */ INVALID, INVALID, INVALID, INVALID, + +/* [C0] */ INVALID, INVALID, TNSZ("cmpss",XMMP,4), INVALID, +/* [C4] */ INVALID, INVALID, INVALID, INVALID, +/* [C8] */ INVALID, INVALID, INVALID, INVALID, +/* [CC] */ INVALID, INVALID, INVALID, INVALID, + +/* [D0] */ INVALID, INVALID, INVALID, INVALID, +/* [D4] */ INVALID, INVALID, TNS("movq2dq",XMMMX), INVALID, +/* [D8] */ INVALID, INVALID, INVALID, INVALID, +/* [DC] */ INVALID, INVALID, INVALID, INVALID, + +/* [E0] */ INVALID, INVALID, INVALID, INVALID, +/* [E4] */ INVALID, INVALID, TNSZ("cvtdq2pd",XMM,8), INVALID, +/* [E8] */ INVALID, INVALID, INVALID, INVALID, +/* [EC] */ INVALID, INVALID, INVALID, INVALID, + +/* [F0] */ INVALID, INVALID, INVALID, INVALID, +/* [F4] */ INVALID, INVALID, INVALID, INVALID, +/* [F8] */ INVALID, INVALID, INVALID, INVALID, +/* [FC] */ INVALID, INVALID, INVALID, INVALID, +}; + +/* + * Decode table for 0x0F opcodes + */ + +const instable_t dis_op0F[16][16] = { +{ +/* [00] */ IND(dis_op0F00), IND(dis_op0F01), TNS("lar",MR), TNS("lsl",MR), +/* [04] */ INVALID, TNS("syscall",NORM), TNS("clts",NORM), TNS("sysret",NORM), +/* [08] */ TNS("invd",NORM), TNS("wbinvd",NORM), INVALID, TNS("ud2",NORM), +/* [0C] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [10] */ TNSZ("movups",XMMO,16), TNSZ("movups",XMMOS,16),TNSZ("movlps",XMMO,8), TNSZ("movlps",XMMOS,8), +/* [14] */ TNSZ("unpcklps",XMMO,16),TNSZ("unpckhps",XMMO,16),TNSZ("movhps",XMMOM,8),TNSZ("movhps",XMMOMS,8), +/* [18] */ IND(dis_op0F18), INVALID, INVALID, INVALID, +/* [1C] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [20] */ TSy("mov",SREG), TSy("mov",SREG), TSy("mov",SREG), TSy("mov",SREG), +/* [24] */ TSx("mov",SREG), INVALID, TSx("mov",SREG), INVALID, +/* [28] */ TNSZ("movaps",XMMO,16), TNSZ("movaps",XMMOS,16),TNSZ("cvtpi2ps",XMMOMX,8),TNSZ("movntps",XMMOS,16), +/* [2C] */ TNSZ("cvttps2pi",XMMOXMM,8),TNSZ("cvtps2pi",XMMOXMM,8),TNSZ("ucomiss",XMMO,4),TNSZ("comiss",XMMO,4), +}, { +/* [30] */ TNS("wrmsr",NORM), TNS("rdtsc",NORM), TNS("rdmsr",NORM), TNS("rdpmc",NORM), +/* [34] */ TNSx("sysenter",NORM), TNSx("sysexit",NORM), INVALID, INVALID, +/* [38] */ INVALID, INVALID, INVALID, INVALID, +/* [3C] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [40] */ TS("cmovx.o",MR), TS("cmovx.no",MR), TS("cmovx.b",MR), TS("cmovx.ae",MR), +/* [44] */ TS("cmovx.e",MR), TS("cmovx.ne",MR), TS("cmovx.be",MR), TS("cmovx.a",MR), +/* [48] */ TS("cmovx.s",MR), TS("cmovx.ns",MR), TS("cmovx.pe",MR), TS("cmovx.po",MR), +/* [4C] */ TS("cmovx.l",MR), TS("cmovx.ge",MR), TS("cmovx.le",MR), TS("cmovx.g",MR), +}, { +/* [50] */ TNS("movmskps",XMMOX3), TNSZ("sqrtps",XMMO,16), TNSZ("rsqrtps",XMMO,16),TNSZ("rcpps",XMMO,16), +/* [54] */ TNSZ("andps",XMMO,16), TNSZ("andnps",XMMO,16), TNSZ("orps",XMMO,16), TNSZ("xorps",XMMO,16), +/* [58] */ TNSZ("addps",XMMO,16), TNSZ("mulps",XMMO,16), TNSZ("cvtps2pd",XMMO,8),TNSZ("cvtdq2ps",XMMO,16), +/* [5C] */ TNSZ("subps",XMMO,16), TNSZ("minps",XMMO,16), TNSZ("divps",XMMO,16), TNSZ("maxps",XMMO,16), +}, { +/* [60] */ TNSZ("punpcklbw",MMO,4),TNSZ("punpcklwd",MMO,4),TNSZ("punpckldq",MMO,4),TNSZ("packsswb",MMO,8), +/* [64] */ TNSZ("pcmpgtb",MMO,8), TNSZ("pcmpgtw",MMO,8), TNSZ("pcmpgtd",MMO,8), TNSZ("packuswb",MMO,8), +/* [68] */ TNSZ("punpckhbw",MMO,8),TNSZ("punpckhwd",MMO,8),TNSZ("punpckhdq",MMO,8),TNSZ("packssdw",MMO,8), +/* [6C] */ TNSZ("INVALID",MMO,0), TNSZ("INVALID",MMO,0), TNSZ("movd",MMO,4), TNSZ("movq",MMO,8), +}, { +/* [70] */ TNSZ("pshufw",MMOPM,8), TNS("psrXXX",MR), TNS("psrXXX",MR), TNS("psrXXX",MR), +/* [74] */ TNSZ("pcmpeqb",MMO,8), TNSZ("pcmpeqw",MMO,8), TNSZ("pcmpeqd",MMO,8), TNS("emms",NORM), +/* [78] */ INVALID, INVALID, INVALID, INVALID, +/* [7C] */ INVALID, INVALID, TNSZ("movd",MMOS,4), TNSZ("movq",MMOS,8), +}, { +/* [80] */ TNS("jo",D), TNS("jno",D), TNS("jb",D), TNS("jae",D), +/* [84] */ TNS("je",D), TNS("jne",D), TNS("jbe",D), TNS("ja",D), +/* [88] */ TNS("js",D), TNS("jns",D), TNS("jp",D), TNS("jnp",D), +/* [8C] */ TNS("jl",D), TNS("jge",D), TNS("jle",D), TNS("jg",D), +}, { +/* [90] */ TNS("seto",Mb), TNS("setno",Mb), TNS("setb",Mb), TNS("setae",Mb), +/* [94] */ TNS("sete",Mb), TNS("setne",Mb), TNS("setbe",Mb), TNS("seta",Mb), +/* [98] */ TNS("sets",Mb), TNS("setns",Mb), TNS("setp",Mb), TNS("setnp",Mb), +/* [9C] */ TNS("setl",Mb), TNS("setge",Mb), TNS("setle",Mb), TNS("setg",Mb), +}, { +/* [A0] */ TSp("push",LSEG), TSp("pop",LSEG), TNS("cpuid",NORM), TS("bt",RMw), +/* [A4] */ TS("shld",DSHIFT), TS("shld",DSHIFTcl), INVALID, INVALID, +/* [A8] */ TSp("push",LSEG), TSp("pop",LSEG), TNS("rsm",NORM), TS("bts",RMw), +/* [AC] */ TS("shrd",DSHIFT), TS("shrd",DSHIFTcl), IND(dis_op0FAE), TS("imul",MRw), +}, { +/* [B0] */ TNS("cmpxchgb",RMw), TS("cmpxchg",RMw), TS("lss",MR), TS("btr",RMw), +/* [B4] */ TS("lfs",MR), TS("lgs",MR), TS("movzb",MOVZ), TNS("movzwl",MOVZ), +/* [B8] */ INVALID, INVALID, IND(dis_op0FBA), TS("btc",RMw), +/* [BC] */ TS("bsf",MRw), TS("bsr",MRw), TS("movsb",MOVZ), TNS("movswl",MOVZ), +}, { +/* [C0] */ TNS("xaddb",XADDB), TS("xadd",RMw), TNSZ("cmpps",XMMOPM,16),TNS("movnti",RM), +/* [C4] */ TNSZ("pinsrw",MMOPRM,2),TNS("pextrw",MMO3P), TNSZ("shufps",XMMOPM,16),IND(dis_op0FC7), +/* [C8] */ INVALID, INVALID, INVALID, INVALID, +/* [CC] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [D0] */ INVALID, TNSZ("psrlw",MMO,8), TNSZ("psrld",MMO,8), TNSZ("psrlq",MMO,8), +/* [D4] */ TNSZ("paddq",MMO,8), TNSZ("pmullw",MMO,8), TNSZ("INVALID",MMO,0), TNS("pmovmskb",MMOM3), +/* [D8] */ TNSZ("psubusb",MMO,8), TNSZ("psubusw",MMO,8), TNSZ("pminub",MMO,8), TNSZ("pand",MMO,8), +/* [DC] */ TNSZ("paddusb",MMO,8), TNSZ("paddusw",MMO,8), TNSZ("pmaxub",MMO,8), TNSZ("pandn",MMO,8), +}, { +/* [E0] */ TNSZ("pavgb",MMO,8), TNSZ("psraw",MMO,8), TNSZ("psrad",MMO,8), TNSZ("pavgw",MMO,8), +/* [E4] */ TNSZ("pmulhuw",MMO,8), TNSZ("pmulhw",MMO,8), TNS("INVALID",XMMO), TNSZ("movntq",MMOMS,8), +/* [E8] */ TNSZ("psubsb",MMO,8), TNSZ("psubsw",MMO,8), TNSZ("pminsw",MMO,8), TNSZ("por",MMO,8), +/* [EC] */ TNSZ("paddsb",MMO,8), TNSZ("paddsw",MMO,8), TNSZ("pmaxsw",MMO,8), TNSZ("pxor",MMO,8), +}, { +/* [F0] */ INVALID, TNSZ("psllw",MMO,8), TNSZ("pslld",MMO,8), TNSZ("psllq",MMO,8), +/* [F4] */ TNSZ("pmuludq",MMO,8), TNSZ("pmaddwd",MMO,8), TNSZ("psadbw",MMO,8), TNSZ("maskmovq",MMOIMPL,8), +/* [F8] */ TNSZ("psubb",MMO,8), TNSZ("psubw",MMO,8), TNSZ("psubd",MMO,8), TNSZ("psubq",MMO,8), +/* [FC] */ TNSZ("paddb",MMO,8), TNSZ("paddw",MMO,8), TNSZ("paddd",MMO,8), INVALID, +} }; + + +/* + * Decode table for 0x80 opcodes + */ + +const instable_t dis_op80[8] = { + +/* [0] */ TNS("addb",IMlw), TNS("orb",IMw), TNS("adcb",IMlw), TNS("sbbb",IMlw), +/* [4] */ TNS("andb",IMw), TNS("subb",IMlw), TNS("xorb",IMw), TNS("cmpb",IMlw), +}; + + +/* + * Decode table for 0x81 opcodes. + */ + +const instable_t dis_op81[8] = { + +/* [0] */ TS("add",IMlw), TS("or",IMw), TS("adc",IMlw), TS("sbb",IMlw), +/* [4] */ TS("and",IMw), TS("sub",IMlw), TS("xor",IMw), TS("cmp",IMlw), +}; + + +/* + * Decode table for 0x82 opcodes. + */ + +const instable_t dis_op82[8] = { + +/* [0] */ TNSx("addb",IMlw), TNSx("orb",IMlw), TNSx("adcb",IMlw), TNSx("sbbb",IMlw), +/* [4] */ TNSx("andb",IMlw), TNSx("subb",IMlw), TNSx("xorb",IMlw), TNSx("cmpb",IMlw), +}; +/* + * Decode table for 0x83 opcodes. + */ + +const instable_t dis_op83[8] = { + +/* [0] */ TS("add",IMlw), TS("or",IMlw), TS("adc",IMlw), TS("sbb",IMlw), +/* [4] */ TS("and",IMlw), TS("sub",IMlw), TS("xor",IMlw), TS("cmp",IMlw), +}; + +/* + * Decode table for 0xC0 opcodes. + */ + +const instable_t dis_opC0[8] = { + +/* [0] */ TNS("rolb",MvI), TNS("rorb",MvI), TNS("rclb",MvI), TNS("rcrb",MvI), +/* [4] */ TNS("shlb",MvI), TNS("shrb",MvI), INVALID, TNS("sarb",MvI), +}; + +/* + * Decode table for 0xD0 opcodes. + */ + +const instable_t dis_opD0[8] = { + +/* [0] */ TNS("rolb",Mv), TNS("rorb",Mv), TNS("rclb",Mv), TNS("rcrb",Mv), +/* [4] */ TNS("shlb",Mv), TNS("shrb",Mv), TNS("salb",Mv), TNS("sarb",Mv), +}; + +/* + * Decode table for 0xC1 opcodes. + * 186 instruction set + */ + +const instable_t dis_opC1[8] = { + +/* [0] */ TS("rol",MvI), TS("ror",MvI), TS("rcl",MvI), TS("rcr",MvI), +/* [4] */ TS("shl",MvI), TS("shr",MvI), TS("sal",MvI), TS("sar",MvI), +}; + +/* + * Decode table for 0xD1 opcodes. + */ + +const instable_t dis_opD1[8] = { + +/* [0] */ TS("rol",Mv), TS("ror",Mv), TS("rcl",Mv), TS("rcr",Mv), +/* [4] */ TS("shl",Mv), TS("shr",Mv), TS("sal",Mv), TS("sar",Mv), +}; + + +/* + * Decode table for 0xD2 opcodes. + */ + +const instable_t dis_opD2[8] = { + +/* [0] */ TNS("rolb",Mv), TNS("rorb",Mv), TNS("rclb",Mv), TNS("rcrb",Mv), +/* [4] */ TNS("shlb",Mv), TNS("shrb",Mv), TNS("salb",Mv), TNS("sarb",Mv), +}; +/* + * Decode table for 0xD3 opcodes. + */ + +const instable_t dis_opD3[8] = { + +/* [0] */ TS("rol",Mv), TS("ror",Mv), TS("rcl",Mv), TS("rcr",Mv), +/* [4] */ TS("shl",Mv), TS("shr",Mv), TS("salb",Mv), TS("sar",Mv), +}; + + +/* + * Decode table for 0xF6 opcodes. + */ + +const instable_t dis_opF6[8] = { + +/* [0] */ TNS("testb",IMw), TNS("testb",IMw), TNS("notb",Mw), TNS("negb",Mw), +/* [4] */ TNS("mulb",MA), TNS("imulb",MA), TNS("divb",MA), TNS("idivb",MA), +}; + + +/* + * Decode table for 0xF7 opcodes. + */ + +const instable_t dis_opF7[8] = { + +/* [0] */ TS("test",IMw), TS("test",IMw), TS("not",Mw), TS("neg",Mw), +/* [4] */ TS("mul",MA), TS("imul",MA), TS("div",MA), TS("idiv",MA), +}; + + +/* + * Decode table for 0xFE opcodes. + */ + +const instable_t dis_opFE[8] = { + +/* [0] */ TNS("incb",Mw), TNS("decb",Mw), INVALID, INVALID, +/* [4] */ INVALID, INVALID, INVALID, INVALID, +}; +/* + * Decode table for 0xFF opcodes. + */ + +const instable_t dis_opFF[8] = { + +/* [0] */ TS("inc",Mw), TS("dec",Mw), TNSyp("call",INM), TNS("lcall",INM), +/* [4] */ TNSy("jmp",INM), TNS("ljmp",INM), TSp("push",M), INVALID, +}; + +/* for 287 instructions, which are a mess to decode */ + +const instable_t dis_opFP1n2[8][8] = { +{ +/* bit pattern: 1101 1xxx MODxx xR/M */ +/* [0,0] */ TNS("fadds",M), TNS("fmuls",M), TNS("fcoms",M), TNS("fcomps",M), +/* [0,4] */ TNS("fsubs",M), TNS("fsubrs",M), TNS("fdivs",M), TNS("fdivrs",M), +}, { +/* [1,0] */ TNS("flds",M), INVALID, TNS("fsts",M), TNS("fstps",M), +/* [1,4] */ TNSZ("fldenv",M,28), TNSZ("fldcw",M,2), TNSZ("fnstenv",M,28), TNSZ("fnstcw",M,2), +}, { +/* [2,0] */ TNS("fiaddl",M), TNS("fimull",M), TNS("ficoml",M), TNS("ficompl",M), +/* [2,4] */ TNS("fisubl",M), TNS("fisubrl",M), TNS("fidivl",M), TNS("fidivrl",M), +}, { +/* [3,0] */ TNS("fildl",M), INVALID, TNS("fistl",M), TNS("fistpl",M), +/* [3,4] */ INVALID, TNSZ("fldt",M,10), INVALID, TNSZ("fstpt",M,10), +}, { +/* [4,0] */ TNSZ("faddl",M,8), TNSZ("fmull",M,8), TNSZ("fcoml",M,8), TNSZ("fcompl",M,8), +/* [4,1] */ TNSZ("fsubl",M,8), TNSZ("fsubrl",M,8), TNSZ("fdivl",M,8), TNSZ("fdivrl",M,8), +}, { +/* [5,0] */ TNSZ("fldl",M,8), INVALID, TNSZ("fstl",M,8), TNSZ("fstpl",M,8), +/* [5,4] */ TNSZ("frstor",M,108), INVALID, TNSZ("fnsave",M,108), TNSZ("fnstsw",M,2), +}, { +/* [6,0] */ TNSZ("fiadd",M,2), TNSZ("fimul",M,2), TNSZ("ficom",M,2), TNSZ("ficomp",M,2), +/* [6,4] */ TNSZ("fisub",M,2), TNSZ("fisubr",M,2), TNSZ("fidiv",M,2), TNSZ("fidivr",M,2), +}, { +/* [7,0] */ TNSZ("fild",M,2), INVALID, TNSZ("fist",M,2), TNSZ("fistp",M,2), +/* [7,4] */ TNSZ("fbld",M,10), TNSZ("fildll",M,8), TNSZ("fbstp",M,10), TNSZ("fistpll",M,8), +} }; + +const instable_t dis_opFP3[8][8] = { +{ +/* bit pattern: 1101 1xxx 11xx xREG */ +/* [0,0] */ TNS("fadd",FF), TNS("fmul",FF), TNS("fcom",F), TNS("fcomp",F), +/* [0,4] */ TNS("fsub",FF), TNS("fsubr",FF), TNS("fdiv",FF), TNS("fdivr",FF), +}, { +/* [1,0] */ TNS("fld",F), TNS("fxch",F), TNS("fnop",NORM), TNS("fstp",F), +/* [1,4] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [2,0] */ INVALID, INVALID, INVALID, INVALID, +/* [2,4] */ INVALID, TNS("fucompp",NORM), INVALID, INVALID, +}, { +/* [3,0] */ INVALID, INVALID, INVALID, INVALID, +/* [3,4] */ INVALID, INVALID, INVALID, INVALID, +}, { +/* [4,0] */ TNS("fadd",FF), TNS("fmul",FF), TNS("fcom",F), TNS("fcomp",F), +/* [4,4] */ TNS("fsub",FF), TNS("fsubr",FF), TNS("fdiv",FF), TNS("fdivr",FF), +}, { +/* [5,0] */ TNS("ffree",F), TNS("fxch",F), TNS("fst",F), TNS("fstp",F), +/* [5,4] */ TNS("fucom",F), TNS("fucomp",F), INVALID, INVALID, +}, { +/* [6,0] */ TNS("faddp",FF), TNS("fmulp",FF), TNS("fcomp",F), TNS("fcompp",NORM), +/* [6,4] */ TNS("fsubp",FF), TNS("fsubrp",FF), TNS("fdivp",FF), TNS("fdivrp",FF), +}, { +/* [7,0] */ TNS("ffree",F), TNS("fxch",F), TNS("fstp",F), TNS("fstp",F), +/* [7,4] */ TNS("fnstsw",M), TNS("fucomip",FFC), TNS("fcomip",FFC), INVALID, +} }; + +const instable_t dis_opFP4[4][8] = { +{ +/* bit pattern: 1101 1001 111x xxxx */ +/* [0,0] */ TNS("fchs",NORM), TNS("fabs",NORM), INVALID, INVALID, +/* [0,4] */ TNS("ftst",NORM), TNS("fxam",NORM), TNS("ftstp",NORM), INVALID, +}, { +/* [1,0] */ TNS("fld1",NORM), TNS("fldl2t",NORM), TNS("fldl2e",NORM), TNS("fldpi",NORM), +/* [1,4] */ TNS("fldlg2",NORM), TNS("fldln2",NORM), TNS("fldz",NORM), INVALID, +}, { +/* [2,0] */ TNS("f2xm1",NORM), TNS("fyl2x",NORM), TNS("fptan",NORM), TNS("fpatan",NORM), +/* [2,4] */ TNS("fxtract",NORM), TNS("fprem1",NORM), TNS("fdecstp",NORM), TNS("fincstp",NORM), +}, { +/* [3,0] */ TNS("fprem",NORM), TNS("fyl2xp1",NORM), TNS("fsqrt",NORM), TNS("fsincos",NORM), +/* [3,4] */ TNS("frndint",NORM), TNS("fscale",NORM), TNS("fsin",NORM), TNS("fcos",NORM), +} }; + +const instable_t dis_opFP5[8] = { +/* bit pattern: 1101 1011 111x xxxx */ +/* [0] */ TNS("feni",NORM), TNS("fdisi",NORM), TNS("fnclex",NORM), TNS("fninit",NORM), +/* [4] */ TNS("fsetpm",NORM), TNS("frstpm",NORM), INVALID, INVALID, +}; + +const instable_t dis_opFP6[8] = { +/* bit pattern: 1101 1011 11yy yxxx */ +/* [00] */ TNS("fcmov.nb",FF), TNS("fcmov.ne",FF), TNS("fcmov.nbe",FF), TNS("fcmov.nu",FF), +/* [04] */ INVALID, TNS("fucomi",F), TNS("fcomi",F), INVALID, +}; + +const instable_t dis_opFP7[8] = { +/* bit pattern: 1101 1010 11yy yxxx */ +/* [00] */ TNS("fcmov.b",FF), TNS("fcmov.e",FF), TNS("fcmov.be",FF), TNS("fcmov.u",FF), +/* [04] */ INVALID, INVALID, INVALID, INVALID, +}; + +/* + * Main decode table for the op codes. The first two nibbles + * will be used as an index into the table. If there is a + * a need to further decode an instruction, the array to be + * referenced is indicated with the other two entries being + * empty. + */ + +const instable_t dis_distable[16][16] = { +{ +/* [0,0] */ TNS("addb",RMw), TS("add",RMw), TNS("addb",MRw), TS("add",MRw), +/* [0,4] */ TNS("addb",IA), TS("add",IA), TSx("push",SEG), TSx("pop",SEG), +/* [0,8] */ TNS("orb",RMw), TS("or",RMw), TNS("orb",MRw), TS("or",MRw), +/* [0,C] */ TNS("orb",IA), TS("or",IA), TSx("push",SEG), IND(&dis_op0F[0][0]), +}, { +/* [1,0] */ TNS("adcb",RMw), TS("adc",RMw), TNS("adcb",MRw), TS("adc",MRw), +/* [1,4] */ TNS("adcb",IA), TS("adc",IA), TSx("push",SEG), TSx("pop",SEG), +/* [1,8] */ TNS("sbbb",RMw), TS("sbb",RMw), TNS("sbbb",MRw), TS("sbb",MRw), +/* [1,C] */ TNS("sbbb",IA), TS("sbb",IA), TSx("push",SEG), TSx("pop",SEG), +}, { +/* [2,0] */ TNS("andb",RMw), TS("and",RMw), TNS("andb",MRw), TS("and",MRw), +/* [2,4] */ TNS("andb",IA), TS("and",IA), TNSx("%es:",OVERRIDE), TNSx("daa",NORM), +/* [2,8] */ TNS("subb",RMw), TS("sub",RMw), TNS("subb",MRw), TS("sub",MRw), +/* [2,C] */ TNS("subb",IA), TS("sub",IA), TNSx("%cs:",OVERRIDE), TNSx("das",NORM), +}, { +/* [3,0] */ TNS("xorb",RMw), TS("xor",RMw), TNS("xorb",MRw), TS("xor",MRw), +/* [3,4] */ TNS("xorb",IA), TS("xor",IA), TNSx("%ss:",OVERRIDE), TNSx("aaa",NORM), +/* [3,8] */ TNS("cmpb",RMw), TS("cmp",RMw), TNS("cmpb",MRw), TS("cmp",MRw), +/* [3,C] */ TNS("cmpb",IA), TS("cmp",IA), TNSx("%ds:",OVERRIDE), TNSx("aas",NORM), +}, { +/* [4,0] */ TSx("inc",R), TSx("inc",R), TSx("inc",R), TSx("inc",R), +/* [4,4] */ TSx("inc",R), TSx("inc",R), TSx("inc",R), TSx("inc",R), +/* [4,8] */ TSx("dec",R), TSx("dec",R), TSx("dec",R), TSx("dec",R), +/* [4,C] */ TSx("dec",R), TSx("dec",R), TSx("dec",R), TSx("dec",R), +}, { +/* [5,0] */ TSp("push",R), TSp("push",R), TSp("push",R), TSp("push",R), +/* [5,4] */ TSp("push",R), TSp("push",R), TSp("push",R), TSp("push",R), +/* [5,8] */ TSp("pop",R), TSp("pop",R), TSp("pop",R), TSp("pop",R), +/* [5,C] */ TSp("pop",R), TSp("pop",R), TSp("pop",R), TSp("pop",R), +}, { +/* [6,0] */ TSZx("pusha",IMPLMEM,28),TSZx("popa",IMPLMEM,28), TSx("bound",MR), TNS("arpl",RMw), +/* [6,4] */ TNS("%fs:",OVERRIDE), TNS("%gs:",OVERRIDE), TNS("data16",DM), TNS("addr16",AM), +/* [6,8] */ TSp("push",I), TS("imul",IMUL), TSp("push",Ib), TS("imul",IMUL), +/* [6,C] */ TNSZ("insb",IMPLMEM,1), TSZ("ins",IMPLMEM,4), TNSZ("outsb",IMPLMEM,1),TSZ("outs",IMPLMEM,4), +}, { +/* [7,0] */ TNSy("jo",BD), TNSy("jno",BD), TNSy("jb",BD), TNSy("jae",BD), +/* [7,4] */ TNSy("je",BD), TNSy("jne",BD), TNSy("jbe",BD), TNSy("ja",BD), +/* [7,8] */ TNSy("js",BD), TNSy("jns",BD), TNSy("jp",BD), TNSy("jnp",BD), +/* [7,C] */ TNSy("jl",BD), TNSy("jge",BD), TNSy("jle",BD), TNSy("jg",BD), +}, { +/* [8,0] */ IND(dis_op80), IND(dis_op81), INDx(dis_op82), IND(dis_op83), +/* [8,4] */ TNS("testb",RMw), TS("test",RMw), TNS("xchgb",RMw), TS("xchg",RMw), +/* [8,8] */ TNS("movb",RMw), TS("mov",RMw), TNS("movb",MRw), TS("mov",MRw), +/* [8,C] */ TNS("movw",SM), TS("lea",MR), TNS("movw",MS), TSp("pop",M), +}, { +/* [9,0] */ TNS("nop",NORM), TS("xchg",RA), TS("xchg",RA), TS("xchg",RA), +/* [9,4] */ TS("xchg",RA), TS("xchg",RA), TS("xchg",RA), TS("xchg",RA), +/* [9,8] */ TNS("cXtX",CBW), TNS("cXtX",CWD), TNSx("lcall",SO), TNS("fwait",NORM), +/* [9,C] */ TSZy("pushf",IMPLMEM,4),TSZy("popf",IMPLMEM,4), TNSx("sahf",NORM), TNSx("lahf",NORM), +}, { +/* [A,0] */ TNS("movb",OA), TS("mov",OA), TNS("movb",AO), TS("mov",AO), +/* [A,4] */ TNSZ("movsb",SD,1), TS("movs",SD), TNSZ("cmpsb",SD,1), TS("cmps",SD), +/* [A,8] */ TNS("testb",IA), TS("test",IA), TNS("stosb",AD), TS("stos",AD), +/* [A,C] */ TNS("lodsb",SA), TS("lods",SA), TNS("scasb",AD), TS("scas",AD), +}, { +/* [B,0] */ TNS("movb",IR), TNS("movb",IR), TNS("movb",IR), TNS("movb",IR), +/* [B,4] */ TNS("movb",IR), TNS("movb",IR), TNS("movb",IR), TNS("movb",IR), +/* [B,8] */ TS("mov",IR), TS("mov",IR), TS("mov",IR), TS("mov",IR), +/* [B,C] */ TS("mov",IR), TS("mov",IR), TS("mov",IR), TS("mov",IR), +}, { +/* [C,0] */ IND(dis_opC0), IND(dis_opC1), TNSyp("ret",RET), TNSyp("ret",NORM), +/* [C,4] */ TNSx("les",MR), TNSx("lds",MR), TNS("movb",IMw), TS("mov",IMw), +/* [C,8] */ TNSyp("enter",ENTER), TNSyp("leave",NORM), TNS("lret",RET), TNS("lret",NORM), +/* [C,C] */ TNS("int",INT3), TNS("int",INTx), TNSx("into",NORM), TNS("iret",NORM), +}, { +/* [D,0] */ IND(dis_opD0), IND(dis_opD1), IND(dis_opD2), IND(dis_opD3), +/* [D,4] */ TNSx("aam",U), TNSx("aad",U), TNSx("falc",NORM), TNSZ("xlat",IMPLMEM,1), + +/* 287 instructions. Note that although the indirect field */ +/* indicates opFP1n2 for further decoding, this is not necessarily */ +/* the case since the opFP arrays are not partitioned according to key1 */ +/* and key2. opFP1n2 is given only to indicate that we haven't */ +/* finished decoding the instruction. */ +/* [D,8] */ IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), +/* [D,C] */ IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), IND(&dis_opFP1n2[0][0]), +}, { +/* [E,0] */ TNSy("loopnz",BD), TNSy("loopz",BD), TNSy("loop",BD), TNSy("jcxz",BD), +/* [E,4] */ TNS("inb",P), TS("in",P), TNS("outb",P), TS("out",P), +/* [E,8] */ TNSyp("call",D), TNSy("jmp",D), TNSx("ljmp",SO), TNSy("jmp",BD), +/* [E,C] */ TNS("inb",V), TS("in",V), TNS("outb",V), TS("out",V), +}, { +/* [F,0] */ TNS("lock",LOCK), TNS("icebp", NORM), TNS("repnz",PREFIX), TNS("repz",PREFIX), +/* [F,4] */ TNS("hlt",NORM), TNS("cmc",NORM), IND(dis_opF6), IND(dis_opF7), +/* [F,8] */ TNS("clc",NORM), TNS("stc",NORM), TNS("cli",NORM), TNS("sti",NORM), +/* [F,C] */ TNS("cld",NORM), TNS("std",NORM), IND(dis_opFE), IND(dis_opFF), +} }; + +/* END CSTYLED */ + +/* + * common functions to decode and disassemble an x86 or amd64 instruction + */ + +/* + * These are the individual fields of a REX prefix. Note that a REX + * prefix with none of these set is still needed to: + * - use the MOVSXD (sign extend 32 to 64 bits) instruction + * - access the %sil, %dil, %bpl, %spl registers + */ +#define REX_W 0x08 /* 64 bit operand size when set */ +#define REX_R 0x04 /* high order bit extension of ModRM reg field */ +#define REX_X 0x02 /* high order bit extension of SIB index field */ +#define REX_B 0x01 /* extends ModRM r_m, SIB base, or opcode reg */ + +static uint_t opnd_size; /* SIZE16, SIZE32 or SIZE64 */ +static uint_t addr_size; /* SIZE16, SIZE32 or SIZE64 */ + +/* + * Even in 64 bit mode, usually only 4 byte immediate operands are supported. + */ +static int isize[] = {1, 2, 4, 4}; +static int isize64[] = {1, 2, 4, 8}; + +/* + * Just a bunch of useful macros. + */ +#define WBIT(x) (x & 0x1) /* to get w bit */ +#define REGNO(x) (x & 0x7) /* to get 3 bit register */ +#define VBIT(x) ((x)>>1 & 0x1) /* to get 'v' bit */ +#define OPSIZE(osize, wbit) ((wbit) ? isize[osize] : 1) +#define OPSIZE64(osize, wbit) ((wbit) ? isize64[osize] : 1) + +#define REG_ONLY 3 /* mode to indicate a register operand (not memory) */ + +#define BYTE_OPND 0 /* w-bit value indicating byte register */ +#define LONG_OPND 1 /* w-bit value indicating opnd_size register */ +#define MM_OPND 2 /* "value" used to indicate a mmx reg */ +#define XMM_OPND 3 /* "value" used to indicate a xmm reg */ +#define SEG_OPND 4 /* "value" used to indicate a segment reg */ +#define CONTROL_OPND 5 /* "value" used to indicate a control reg */ +#define DEBUG_OPND 6 /* "value" used to indicate a debug reg */ +#define TEST_OPND 7 /* "value" used to indicate a test reg */ +#define WORD_OPND 8 /* w-bit value indicating word size reg */ + +/* + * Get the next byte and separate the op code into the high and low nibbles. + */ +static int +dtrace_get_opcode(dis86_t *x, uint_t *high, uint_t *low) +{ + int byte; + + /* + * x86 instructions have a maximum length of 15 bytes. Bail out if + * we try to read more. + */ + if (x->d86_len >= 15) + return (x->d86_error = 1); + + if (x->d86_error) + return (1); + byte = x->d86_get_byte(x->d86_data); + if (byte < 0) + return (x->d86_error = 1); + x->d86_bytes[x->d86_len++] = byte; + *low = byte & 0xf; /* ----xxxx low 4 bits */ + *high = byte >> 4 & 0xf; /* xxxx---- bits 7 to 4 */ + return (0); +} + +/* + * Get and decode an SIB (scaled index base) byte + */ +static void +dtrace_get_SIB(dis86_t *x, uint_t *ss, uint_t *index, uint_t *base) +{ + int byte; + + if (x->d86_error) + return; + + byte = x->d86_get_byte(x->d86_data); + if (byte < 0) { + x->d86_error = 1; + return; + } + x->d86_bytes[x->d86_len++] = byte; + + *base = byte & 0x7; + *index = (byte >> 3) & 0x7; + *ss = (byte >> 6) & 0x3; +} + +/* + * Get the byte following the op code and separate it into the + * mode, register, and r/m fields. + */ +static void +dtrace_get_modrm(dis86_t *x, uint_t *mode, uint_t *reg, uint_t *r_m) +{ + if (x->d86_got_modrm == 0) { + if (x->d86_rmindex == -1) + x->d86_rmindex = x->d86_len; + dtrace_get_SIB(x, mode, reg, r_m); + x->d86_got_modrm = 1; + } +} + +/* + * Adjust register selection based on any REX prefix bits present. + */ +/*ARGSUSED*/ +static void +dtrace_rex_adjust(uint_t rex_prefix, uint_t mode, uint_t *reg, uint_t *r_m) +{ + if (reg != NULL && r_m == NULL) { + if (rex_prefix & REX_B) + *reg += 8; + } else { + if (reg != NULL && (REX_R & rex_prefix) != 0) + *reg += 8; + if (r_m != NULL && (REX_B & rex_prefix) != 0) + *r_m += 8; + } +} + +/* + * Get an immediate operand of the given size, with sign extension. + */ +static void +dtrace_imm_opnd(dis86_t *x, int wbit, int size, int opindex) +{ + int i; + int byte; + int valsize = 0; + + if (x->d86_numopnds < opindex + 1) + x->d86_numopnds = opindex + 1; + + switch (wbit) { + case BYTE_OPND: + valsize = 1; + break; + case LONG_OPND: + if (x->d86_opnd_size == SIZE16) + valsize = 2; + else if (x->d86_opnd_size == SIZE32) + valsize = 4; + else + valsize = 8; + break; + case MM_OPND: + case XMM_OPND: + case SEG_OPND: + case CONTROL_OPND: + case DEBUG_OPND: + case TEST_OPND: + valsize = size; + break; + case WORD_OPND: + valsize = 2; + break; + } + if (valsize < size) + valsize = size; + + if (x->d86_error) + return; + x->d86_opnd[opindex].d86_value = 0; + for (i = 0; i < size; ++i) { + byte = x->d86_get_byte(x->d86_data); + if (byte < 0) { + x->d86_error = 1; + return; + } + x->d86_bytes[x->d86_len++] = byte; + x->d86_opnd[opindex].d86_value |= (uint64_t)byte << (i * 8); + } + /* Do sign extension */ + if (x->d86_bytes[x->d86_len - 1] & 0x80) { + for (; i < valsize; i++) + x->d86_opnd[opindex].d86_value |= + (uint64_t)0xff << (i* 8); + } +#ifdef DIS_TEXT + x->d86_opnd[opindex].d86_mode = MODE_SIGNED; + x->d86_opnd[opindex].d86_value_size = valsize; + x->d86_imm_bytes += size; +#endif +} + +/* + * Get an ip relative operand of the given size, with sign extension. + */ +static void +dtrace_disp_opnd(dis86_t *x, int wbit, int size, int opindex) +{ + dtrace_imm_opnd(x, wbit, size, opindex); +#ifdef DIS_TEXT + x->d86_opnd[opindex].d86_mode = MODE_IPREL; +#endif +} + +/* + * Check to see if there is a segment override prefix pending. + * If so, print it in the current 'operand' location and set + * the override flag back to false. + */ +/*ARGSUSED*/ +static void +dtrace_check_override(dis86_t *x, int opindex) +{ +#ifdef DIS_TEXT + if (x->d86_seg_prefix) { + (void) strlcat(x->d86_opnd[opindex].d86_prefix, + x->d86_seg_prefix, PFIXLEN); + } +#endif + x->d86_seg_prefix = NULL; +} + + +/* + * Process a single instruction Register or Memory operand. + * + * mode = addressing mode from ModRM byte + * r_m = r_m (or reg if mode == 3) field from ModRM byte + * wbit = indicates which register (8bit, 16bit, ... MMX, etc.) set to use. + * o = index of operand that we are processing (0, 1 or 2) + * + * the value of reg or r_m must have already been adjusted for any REX prefix. + */ +/*ARGSUSED*/ +static void +dtrace_get_operand(dis86_t *x, uint_t mode, uint_t r_m, int wbit, int opindex) +{ + int have_SIB = 0; /* flag presence of scale-index-byte */ + uint_t ss; /* scale-factor from opcode */ + uint_t index; /* index register number */ + uint_t base; /* base register number */ + int dispsize; /* size of displacement in bytes */ +#ifdef DIS_TEXT + char *opnd = x->d86_opnd[opindex].d86_opnd; +#endif + + if (x->d86_numopnds < opindex + 1) + x->d86_numopnds = opindex + 1; + + if (x->d86_error) + return; + + /* + * first handle a simple register + */ + if (mode == REG_ONLY) { +#ifdef DIS_TEXT + switch (wbit) { + case MM_OPND: + (void) strlcat(opnd, dis_MMREG[r_m], OPLEN); + break; + case XMM_OPND: + (void) strlcat(opnd, dis_XMMREG[r_m], OPLEN); + break; + case SEG_OPND: + (void) strlcat(opnd, dis_SEGREG[r_m], OPLEN); + break; + case CONTROL_OPND: + (void) strlcat(opnd, dis_CONTROLREG[r_m], OPLEN); + break; + case DEBUG_OPND: + (void) strlcat(opnd, dis_DEBUGREG[r_m], OPLEN); + break; + case TEST_OPND: + (void) strlcat(opnd, dis_TESTREG[r_m], OPLEN); + break; + case BYTE_OPND: + if (x->d86_rex_prefix == 0) + (void) strlcat(opnd, dis_REG8[r_m], OPLEN); + else + (void) strlcat(opnd, dis_REG8_REX[r_m], OPLEN); + break; + case WORD_OPND: + (void) strlcat(opnd, dis_REG16[r_m], OPLEN); + break; + case LONG_OPND: + if (x->d86_opnd_size == SIZE16) + (void) strlcat(opnd, dis_REG16[r_m], OPLEN); + else if (x->d86_opnd_size == SIZE32) + (void) strlcat(opnd, dis_REG32[r_m], OPLEN); + else + (void) strlcat(opnd, dis_REG64[r_m], OPLEN); + break; + } +#endif /* DIS_TEXT */ + return; + } + + /* + * if symbolic representation, skip override prefix, if any + */ + dtrace_check_override(x, opindex); + + /* + * Handle 16 bit memory references first, since they decode + * the mode values more simply. + * mode 1 is r_m + 8 bit displacement + * mode 2 is r_m + 16 bit displacement + * mode 0 is just r_m, unless r_m is 6 which is 16 bit disp + */ + if (x->d86_addr_size == SIZE16) { + if ((mode == 0 && r_m == 6) || mode == 2) + dtrace_imm_opnd(x, WORD_OPND, 2, opindex); + else if (mode == 1) + dtrace_imm_opnd(x, BYTE_OPND, 1, opindex); +#ifdef DIS_TEXT + if (mode == 0 && r_m == 6) + x->d86_opnd[opindex].d86_mode = MODE_SIGNED; + else if (mode == 0) + x->d86_opnd[opindex].d86_mode = MODE_NONE; + else + x->d86_opnd[opindex].d86_mode = MODE_OFFSET; + (void) strlcat(opnd, dis_addr16[mode][r_m], OPLEN); +#endif + return; + } + + /* + * 32 and 64 bit addressing modes are more complex since they + * can involve an SIB (scaled index and base) byte to decode. + */ + if (r_m == ESP_REGNO || r_m == ESP_REGNO + 8) { + have_SIB = 1; + dtrace_get_SIB(x, &ss, &index, &base); + if (x->d86_error) + return; + if (base != 5 || mode != 0) + if (x->d86_rex_prefix & REX_B) + base += 8; + if (x->d86_rex_prefix & REX_X) + index += 8; + } else { + base = r_m; + } + + /* + * Compute the displacement size and get its bytes + */ + dispsize = 0; + + if (mode == 1) + dispsize = 1; + else if (mode == 2) + dispsize = 4; + else if ((r_m & 7) == EBP_REGNO || + (have_SIB && (base & 7) == EBP_REGNO)) + dispsize = 4; + + if (dispsize > 0) { + dtrace_imm_opnd(x, dispsize == 4 ? LONG_OPND : BYTE_OPND, + dispsize, opindex); + if (x->d86_error) + return; + } + +#ifdef DIS_TEXT + if (dispsize > 0) + x->d86_opnd[opindex].d86_mode = MODE_OFFSET; + + if (have_SIB == 0) { + if (x->d86_mode == SIZE32) { + if (mode == 0) + (void) strlcat(opnd, dis_addr32_mode0[r_m], + OPLEN); + else + (void) strlcat(opnd, dis_addr32_mode12[r_m], + OPLEN); + } else { + if (mode == 0) + (void) strlcat(opnd, dis_addr64_mode0[r_m], + OPLEN); + else + (void) strlcat(opnd, dis_addr64_mode12[r_m], + OPLEN); + } + } else { + uint_t need_paren = 0; + char **regs; + if (x->d86_mode == SIZE32) /* NOTE this is not addr_size! */ + regs = (char **)dis_REG32; + else + regs = (char **)dis_REG64; + + /* + * print the base (if any) + */ + if (base == EBP_REGNO && mode == 0) { + if (index != ESP_REGNO) { + (void) strlcat(opnd, "(", OPLEN); + need_paren = 1; + } + } else { + (void) strlcat(opnd, "(", OPLEN); + (void) strlcat(opnd, regs[base], OPLEN); + need_paren = 1; + } + + /* + * print the index (if any) + */ + if (index != ESP_REGNO) { + (void) strlcat(opnd, ",", OPLEN); + (void) strlcat(opnd, regs[index], OPLEN); + (void) strlcat(opnd, dis_scale_factor[ss], OPLEN); + } else + if (need_paren) + (void) strlcat(opnd, ")", OPLEN); + } +#endif +} + +/* + * Operand sequence for standard instruction involving one register + * and one register/memory operand. + * wbit indicates a byte(0) or opnd_size(1) operation + * vbit indicates direction (0 for "opcode r,r_m") or (1 for "opcode r_m, r") + */ +#define STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, vbit) { \ + dtrace_get_modrm(x, &mode, ®, &r_m); \ + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); \ + dtrace_get_operand(x, mode, r_m, wbit, vbit); \ + dtrace_get_operand(x, REG_ONLY, reg, wbit, 1 - vbit); \ +} + +/* + * Similar to above, but allows for the two operands to be of different + * classes (ie. wbit). + * wbit is for the r_m operand + * w2 is for the reg operand + */ +#define MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, w2, vbit) { \ + dtrace_get_modrm(x, &mode, ®, &r_m); \ + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); \ + dtrace_get_operand(x, mode, r_m, wbit, vbit); \ + dtrace_get_operand(x, REG_ONLY, reg, w2, 1 - vbit); \ +} + +/* + * Similar, but for 2 operands plus an immediate. + */ +#define THREEOPERAND(x, mode, reg, r_m, rex_prefix, wbit, w2, immsize) { \ + dtrace_get_modrm(x, &mode, ®, &r_m); \ + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); \ + dtrace_get_operand(x, mode, r_m, wbit, 1); \ + dtrace_get_operand(x, REG_ONLY, reg, w2, 2); \ + dtrace_imm_opnd(x, wbit, immsize, 0); \ +} + +/* + * Dissassemble a single x86 or amd64 instruction. + * + * Mode determines the default operating mode (SIZE16, SIZE32 or SIZE64) + * for interpreting instructions. + * + * returns non-zero for bad opcode + */ +int +dtrace_disx86(dis86_t *x, uint_t cpu_mode) +{ + const instable_t *dp = NULL; /* decode table being used */ +#ifdef DIS_TEXT + uint_t i; +#endif +#ifdef DIS_MEM + uint_t nomem = 0; +#define NOMEM (nomem = 1) +#else +#define NOMEM /* nothing */ +#endif + uint_t wbit = 0; /* opcode wbit, 0 is 8 bit, !0 for opnd_size */ + uint_t w2; /* wbit value for second operand */ + uint_t vbit; + uint_t mode = 0; /* mode value from ModRM byte */ + uint_t reg; /* reg value from ModRM byte */ + uint_t r_m; /* r_m value from ModRM byte */ + + uint_t opcode1; /* high nibble of 1st byte */ + uint_t opcode2; /* low nibble of 1st byte */ + uint_t opcode3; /* extra opcode bits usually from ModRM byte */ + uint_t opcode4; /* high nibble of 2nd byte */ + uint_t opcode5; /* low nibble of 2ne byte */ + uint_t opcode6; /* high nibble of 3rd byte */ + uint_t opcode7; /* low nibble of 3rd byte */ + uint_t opcode_bytes = 1; + + /* + * legacy prefixes come in 5 flavors, you should have only one of each + */ + uint_t opnd_size_prefix = 0; + uint_t addr_size_prefix = 0; + uint_t segment_prefix = 0; + uint_t lock_prefix = 0; + uint_t rep_prefix = 0; + uint_t rex_prefix = 0; /* amd64 register extension prefix */ + size_t off; + + x->d86_len = 0; + x->d86_rmindex = -1; + x->d86_error = 0; +#ifdef DIS_TEXT + x->d86_numopnds = 0; + x->d86_seg_prefix = NULL; + x->d86_mneu[0] = 0; + for (i = 0; i < 3; ++i) { + x->d86_opnd[i].d86_opnd[0] = 0; + x->d86_opnd[i].d86_prefix[0] = 0; + x->d86_opnd[i].d86_value_size = 0; + x->d86_opnd[i].d86_value = 0; + x->d86_opnd[i].d86_mode = MODE_NONE; + } +#endif + x->d86_error = 0; + x->d86_memsize = 0; + + if (cpu_mode == SIZE16) { + opnd_size = SIZE16; + addr_size = SIZE16; + } else if (cpu_mode == SIZE32) { + opnd_size = SIZE32; + addr_size = SIZE32; + } else { + opnd_size = SIZE32; + addr_size = SIZE64; + } + + /* + * Get one opcode byte and check for zero padding that follows + * jump tables. + */ + if (dtrace_get_opcode(x, &opcode1, &opcode2) != 0) + goto error; + + if (opcode1 == 0 && opcode2 == 0 && + x->d86_check_func != NULL && x->d86_check_func(x->d86_data)) { +#ifdef DIS_TEXT + (void) strncpy(x->d86_mneu, ".byte\t0", OPLEN); +#endif + goto done; + } + + /* + * Gather up legacy x86 prefix bytes. + */ + for (;;) { + uint_t *which_prefix = NULL; + + dp = &dis_distable[opcode1][opcode2]; + + switch (dp->it_adrmode) { + case PREFIX: + which_prefix = &rep_prefix; + break; + case LOCK: + which_prefix = &lock_prefix; + break; + case OVERRIDE: + which_prefix = &segment_prefix; +#ifdef DIS_TEXT + x->d86_seg_prefix = (char *)dp->it_name; +#endif + if (dp->it_invalid64 && cpu_mode == SIZE64) + goto error; + break; + case AM: + which_prefix = &addr_size_prefix; + break; + case DM: + which_prefix = &opnd_size_prefix; + break; + } + if (which_prefix == NULL) + break; + *which_prefix = (opcode1 << 4) | opcode2; + if (dtrace_get_opcode(x, &opcode1, &opcode2) != 0) + goto error; + } + + /* + * Handle amd64 mode PREFIX values. + * Some of the segment prefixes are no-ops. (only FS/GS actually work) + * We might have a REX prefix (opcodes 0x40-0x4f) + */ + if (cpu_mode == SIZE64) { + if (segment_prefix != 0x64 && segment_prefix != 0x65) + segment_prefix = 0; + + if (opcode1 == 0x4) { + rex_prefix = (opcode1 << 4) | opcode2; + if (dtrace_get_opcode(x, &opcode1, &opcode2) != 0) + goto error; + dp = &dis_distable[opcode1][opcode2]; + } + } + + /* + * Deal with selection of operand and address size now. + * Note that the REX.W bit being set causes opnd_size_prefix to be + * ignored. + */ + if (cpu_mode == SIZE64) { + if (rex_prefix & 0x08) + opnd_size = SIZE64; + else if (opnd_size_prefix) + opnd_size = SIZE16; + + if (addr_size_prefix) + addr_size = SIZE32; + } else if (cpu_mode == SIZE32) { + if (opnd_size_prefix) + opnd_size = SIZE16; + if (addr_size_prefix) + addr_size = SIZE16; + } else { + if (opnd_size_prefix) + opnd_size = SIZE32; + if (addr_size_prefix) + addr_size = SIZE32; + } + + /* + * The pause instruction - a repz'd nop. This doesn't fit + * with any of the other prefix goop added for SSE, so we'll + * special-case it here. + */ + if (rep_prefix == 0xf3 && opcode1 == 0x9 && opcode2 == 0x0) { + rep_prefix = 0; + dp = &dis_opPause; + } + + /* + * Some 386 instructions have 2 bytes of opcode before the mod_r/m + * byte so we may need to perform a table indirection. + */ + if (dp->it_indirect == dis_op0F[0]) { + if (dtrace_get_opcode(x, &opcode4, &opcode5) != 0) + goto error; + opcode_bytes = 2; + if (opcode4 == 0x7 && opcode5 >= 0x1 && opcode5 <= 0x3) { + uint_t subcode; + + if (dtrace_get_opcode(x, &opcode6, &opcode7) != 0) + goto error; + opcode_bytes = 3; + subcode = ((opcode6 & 0x3) << 1) | + ((opcode7 & 0x8) >> 3); + dp = &dis_op0F7123[opcode5][subcode]; + } else if ((opcode4 == 0xc) && (opcode5 >= 0x8)) { + dp = &dis_op0FC8[0]; + } else { + dp = &dis_op0F[opcode4][opcode5]; + } + } + + /* + * If still not at a TERM decode entry, then a ModRM byte + * exists and its fields further decode the instruction. + */ + x->d86_got_modrm = 0; + if (dp->it_indirect != TERM) { + dtrace_get_modrm(x, &mode, &opcode3, &r_m); + if (x->d86_error) + goto error; + reg = opcode3; + + /* + * decode 287 instructions (D8-DF) from opcodeN + */ + if (opcode1 == 0xD && opcode2 >= 0x8) { + if (opcode2 == 0xB && mode == 0x3 && opcode3 == 4) + dp = &dis_opFP5[r_m]; + else if (opcode2 == 0xA && mode == 0x3 && opcode3 < 4) + dp = &dis_opFP7[opcode3]; + else if (opcode2 == 0xB && mode == 0x3) + dp = &dis_opFP6[opcode3]; + else if (opcode2 == 0x9 && mode == 0x3 && opcode3 >= 4) + dp = &dis_opFP4[opcode3 - 4][r_m]; + else if (mode == 0x3) + dp = &dis_opFP3[opcode2 - 8][opcode3]; + else + dp = &dis_opFP1n2[opcode2 - 8][opcode3]; + } else { + dp = dp->it_indirect + opcode3; + } + } + + /* + * In amd64 bit mode, ARPL opcode is changed to MOVSXD + * (sign extend 32bit to 64 bit) + */ + if (cpu_mode == SIZE64 && opcode1 == 0x6 && opcode2 == 0x3) + dp = &dis_opMOVSLD; + + /* + * at this point we should have a correct (or invalid) opcode + */ + if ((cpu_mode == SIZE64 && dp->it_invalid64) || + (cpu_mode != SIZE64 && dp->it_invalid32)) + goto error; + if (dp->it_indirect != TERM) + goto error; + + /* + * deal with MMX/SSE opcodes which are changed by prefixes + */ + switch (dp->it_adrmode) { + case MMO: + case MMOIMPL: + case MMO3P: + case MMOM3: + case MMOMS: + case MMOPM: + case MMOPRM: + case MMOS: + case XMMO: + case XMMOM: + case XMMOMS: + case XMMOPM: + case XMMOS: + case XMMOMX: + case XMMOX3: + case XMMOXMM: + /* + * This is horrible. Some SIMD instructions take the + * form 0x0F 0x?? ..., which is easily decoded using the + * existing tables. Other SIMD instructions use various + * prefix bytes to overload existing instructions. For + * Example, addps is F0, 58, whereas addss is F3 (repz), + * F0, 58. Presumably someone got a raise for this. + * + * If we see one of the instructions which can be + * modified in this way (if we've got one of the SIMDO* + * address modes), we'll check to see if the last prefix + * was a repz. If it was, we strip the prefix from the + * mnemonic, and we indirect using the dis_opSIMDrepz + * table. + */ + + /* + * Calculate our offset in dis_op0F + */ + if ((uintptr_t)dp - (uintptr_t)dis_op0F > sizeof (dis_op0F)) + goto error; + + off = ((uintptr_t)dp - (uintptr_t)dis_op0F) / + sizeof (instable_t); + + /* + * Rewrite if this instruction used one of the magic prefixes. + */ + if (rep_prefix) { + if (rep_prefix == 0xf2) + dp = &dis_opSIMDrepnz[off]; + else + dp = &dis_opSIMDrepz[off]; + rep_prefix = 0; + } else if (opnd_size_prefix) { + dp = &dis_opSIMDdata16[off]; + opnd_size_prefix = 0; + if (opnd_size == SIZE16) + opnd_size = SIZE32; + } + break; + + case MMOSH: + /* + * As with the "normal" SIMD instructions, the MMX + * shuffle instructions are overloaded. These + * instructions, however, are special in that they use + * an extra byte, and thus an extra table. As of this + * writing, they only use the opnd_size prefix. + */ + + /* + * Calculate our offset in dis_op0F7123 + */ + if ((uintptr_t)dp - (uintptr_t)dis_op0F7123 > + sizeof (dis_op0F7123)) + goto error; + + if (opnd_size_prefix) { + off = ((uintptr_t)dp - (uintptr_t)dis_op0F7123) / + sizeof (instable_t); + dp = &dis_opSIMD7123[off]; + opnd_size_prefix = 0; + if (opnd_size == SIZE16) + opnd_size = SIZE32; + } + break; + } + + /* + * In 64 bit mode, some opcodes automatically use opnd_size == SIZE64. + */ + if (cpu_mode == SIZE64) + if (dp->it_always64 || (opnd_size == SIZE32 && dp->it_stackop)) + opnd_size = SIZE64; + +#ifdef DIS_TEXT + /* + * At this point most instructions can format the opcode mnemonic + * including the prefixes. + */ + if (lock_prefix) + (void) strlcat(x->d86_mneu, "lock ", OPLEN); + + if (rep_prefix == 0xf2) + (void) strlcat(x->d86_mneu, "repnz ", OPLEN); + else if (rep_prefix == 0xf3) + (void) strlcat(x->d86_mneu, "repz ", OPLEN); + + if (cpu_mode == SIZE64 && addr_size_prefix) + (void) strlcat(x->d86_mneu, "addr32 ", OPLEN); + + if (dp->it_adrmode != CBW && + dp->it_adrmode != CWD && + dp->it_adrmode != XMMSFNC) { + if (strcmp(dp->it_name, "INVALID") == 0) + goto error; + (void) strlcat(x->d86_mneu, dp->it_name, OPLEN); + if (dp->it_suffix) { + char *types[] = {"", "w", "l", "q"}; + if (opcode_bytes == 2 && opcode4 == 4) { + /* It's a cmovx.yy. Replace the suffix x */ + for (i = 5; i < OPLEN; i++) { + if (x->d86_mneu[i] == '.') + break; + } + x->d86_mneu[i - 1] = *types[opnd_size]; + } else { + (void) strlcat(x->d86_mneu, types[opnd_size], + OPLEN); + } + } + } +#endif + + /* + * Process operands based on the addressing modes. + */ + x->d86_mode = cpu_mode; + x->d86_rex_prefix = rex_prefix; + x->d86_opnd_size = opnd_size; + x->d86_addr_size = addr_size; + vbit = 0; /* initialize for mem/reg -> reg */ + switch (dp->it_adrmode) { + /* + * amd64 instruction to sign extend 32 bit reg/mem operands + * into 64 bit register values + */ + case MOVSXZ: +#ifdef DIS_TEXT + if (rex_prefix == 0) + (void) strncpy(x->d86_mneu, "movzld", OPLEN); +#endif + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + x->d86_opnd_size = SIZE64; + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 1); + x->d86_opnd_size = opnd_size = SIZE32; + wbit = LONG_OPND; + dtrace_get_operand(x, mode, r_m, wbit, 0); + break; + + /* + * movsbl movsbw movsbq (0x0FBE) or movswl movswq (0x0FBF) + * movzbl movzbw movzbq (0x0FB6) or mobzwl movzwq (0x0FB7) + * wbit lives in 2nd byte, note that operands + * are different sized + */ + case MOVZ: + if (rex_prefix & REX_W) { + /* target register size = 64 bit */ + x->d86_mneu[5] = 'q'; + } + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 1); + x->d86_opnd_size = opnd_size = SIZE16; + wbit = WBIT(opcode5); + dtrace_get_operand(x, mode, r_m, wbit, 0); + break; + + /* + * imul instruction, with either 8-bit or longer immediate + * opcode 0x6B for byte, sign-extended displacement, 0x69 for word(s) + */ + case IMUL: + wbit = LONG_OPND; + THREEOPERAND(x, mode, reg, r_m, rex_prefix, wbit, LONG_OPND, + OPSIZE(opnd_size, opcode2 == 0x9)); + break; + + /* memory or register operand to register, with 'w' bit */ + case MRw: + wbit = WBIT(opcode2); + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 0); + break; + + /* register to memory or register operand, with 'w' bit */ + /* arpl happens to fit here also because it is odd */ + case RMw: + if (opcode_bytes == 2) + wbit = WBIT(opcode5); + else + wbit = WBIT(opcode2); + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 1); + break; + + /* xaddb instruction */ + case XADDB: + wbit = 0; + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 1); + break; + + /* MMX register to memory or register operand */ + case MMS: + case MMOS: +#ifdef DIS_TEXT + wbit = strcmp(dp->it_name, "movd") ? MM_OPND : LONG_OPND; +#else + wbit = LONG_OPND; +#endif + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, MM_OPND, 1); + break; + + /* MMX register to memory */ + case MMOMS: + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode == REG_ONLY) + goto error; + wbit = MM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, MM_OPND, 1); + break; + + /* Double shift. Has immediate operand specifying the shift. */ + case DSHIFT: + wbit = LONG_OPND; + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 2); + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 1); + dtrace_imm_opnd(x, wbit, 1, 0); + break; + + /* + * Double shift. With no immediate operand, specifies using %cl. + */ + case DSHIFTcl: + wbit = LONG_OPND; + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 1); + break; + + /* immediate to memory or register operand */ + case IMlw: + wbit = WBIT(opcode2); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 1); + /* + * Have long immediate for opcode 0x81, but not 0x80 nor 0x83 + */ + dtrace_imm_opnd(x, wbit, OPSIZE(opnd_size, opcode2 == 1), 0); + break; + + /* immediate to memory or register operand with the */ + /* 'w' bit present */ + case IMw: + wbit = WBIT(opcode2); + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 1); + dtrace_imm_opnd(x, wbit, OPSIZE(opnd_size, wbit), 0); + break; + + /* immediate to register with register in low 3 bits */ + /* of op code */ + case IR: + /* w-bit here (with regs) is bit 3 */ + wbit = opcode2 >>3 & 0x1; + reg = REGNO(opcode2); + dtrace_rex_adjust(rex_prefix, mode, ®, NULL); + mode = REG_ONLY; + r_m = reg; + dtrace_get_operand(x, mode, r_m, wbit, 1); + dtrace_imm_opnd(x, wbit, OPSIZE64(opnd_size, wbit), 0); + break; + + /* MMX immediate shift of register */ + case MMSH: + case MMOSH: + wbit = MM_OPND; + goto mm_shift; /* in next case */ + + /* SIMD immediate shift of register */ + case XMMSH: + wbit = XMM_OPND; +mm_shift: + reg = REGNO(opcode7); + dtrace_rex_adjust(rex_prefix, mode, ®, NULL); + dtrace_get_operand(x, REG_ONLY, reg, wbit, 1); + dtrace_imm_opnd(x, wbit, 1, 0); + NOMEM; + break; + + /* accumulator to memory operand */ + case AO: + vbit = 1; + /*FALLTHROUGH*/ + + /* memory operand to accumulator */ + case OA: + wbit = WBIT(opcode2); + dtrace_get_operand(x, REG_ONLY, EAX_REGNO, wbit, 1 - vbit); + dtrace_imm_opnd(x, wbit, OPSIZE64(addr_size, LONG_OPND), vbit); +#ifdef DIS_TEXT + x->d86_opnd[vbit].d86_mode = MODE_OFFSET; +#endif + break; + + + /* segment register to memory or register operand */ + case SM: + vbit = 1; + /*FALLTHROUGH*/ + + /* memory or register operand to segment register */ + case MS: + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, LONG_OPND, vbit); + dtrace_get_operand(x, REG_ONLY, reg, SEG_OPND, 1 - vbit); + break; + + /* + * rotate or shift instructions, which may shift by 1 or + * consult the cl register, depending on the 'v' bit + */ + case Mv: + vbit = VBIT(opcode2); + wbit = WBIT(opcode2); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 1); +#ifdef DIS_TEXT + if (vbit) { + (void) strlcat(x->d86_opnd[0].d86_opnd, "%cl", OPLEN); + } else { + x->d86_opnd[0].d86_mode = MODE_SIGNED; + x->d86_opnd[0].d86_value_size = 1; + x->d86_opnd[0].d86_value = 1; + } +#endif + break; + /* + * immediate rotate or shift instructions + */ + case MvI: + wbit = WBIT(opcode2); +normal_imm_mem: + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 1); + dtrace_imm_opnd(x, wbit, 1, 0); + break; + + /* bit test instructions */ + case MIb: + wbit = LONG_OPND; + goto normal_imm_mem; + + /* single memory or register operand with 'w' bit present */ + case Mw: + wbit = WBIT(opcode2); +just_mem: + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 0); + break; + + case SWAPGS: + if (cpu_mode == SIZE64 && mode == 3 && r_m == 0) { +#ifdef DIS_TEXT + (void) strncpy(x->d86_mneu, "swapgs", OPLEN); +#endif + NOMEM; + break; + } + /*FALLTHROUGH*/ + + /* prefetch instruction - memory operand, but no memory acess */ + case PREF: + NOMEM; + /*FALLTHROUGH*/ + + /* single memory or register operand */ + case M: + wbit = LONG_OPND; + goto just_mem; + + /* single memory or register byte operand */ + case Mb: + wbit = BYTE_OPND; + goto just_mem; + + case MO: + /* Similar to M, but only memory (no direct registers) */ + wbit = LONG_OPND; + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode == 3) + goto error; + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 0); + break; + + /* move special register to register or reverse if vbit */ + case SREG: + switch (opcode5) { + + case 2: + vbit = 1; + /*FALLTHROUGH*/ + case 0: + wbit = CONTROL_OPND; + break; + + case 3: + vbit = 1; + /*FALLTHROUGH*/ + case 1: + wbit = DEBUG_OPND; + break; + + case 6: + vbit = 1; + /*FALLTHROUGH*/ + case 4: + wbit = TEST_OPND; + break; + + } + dtrace_get_modrm(x, &mode, ®, &r_m); + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, REG_ONLY, reg, wbit, vbit); + dtrace_get_operand(x, REG_ONLY, r_m, LONG_OPND, 1 - vbit); + NOMEM; + break; + + /* + * single register operand with register in the low 3 + * bits of op code + */ + case R: + if (opcode_bytes == 2) + reg = REGNO(opcode5); + else + reg = REGNO(opcode2); + dtrace_rex_adjust(rex_prefix, mode, ®, NULL); + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 0); + NOMEM; + break; + + /* + * register to accumulator with register in the low 3 + * bits of op code, xchg instructions + */ + case RA: + NOMEM; + reg = REGNO(opcode2); + dtrace_rex_adjust(rex_prefix, mode, ®, NULL); + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 0); + dtrace_get_operand(x, REG_ONLY, EAX_REGNO, LONG_OPND, 1); + break; + + /* + * single segment register operand, with register in + * bits 3-4 of op code byte + */ + case SEG: + NOMEM; + reg = (x->d86_bytes[x->d86_len - 1] >> 3) & 0x3; + dtrace_get_operand(x, REG_ONLY, reg, SEG_OPND, 0); + break; + + /* + * single segment register operand, with register in + * bits 3-5 of op code + */ + case LSEG: + NOMEM; + /* long seg reg from opcode */ + reg = (x->d86_bytes[x->d86_len - 1] >> 3) & 0x7; + dtrace_get_operand(x, REG_ONLY, reg, SEG_OPND, 0); + break; + + /* memory or register operand to register */ + case MR: + wbit = LONG_OPND; + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 0); + break; + + case RM: + wbit = LONG_OPND; + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 1); + break; + + /* MMX/SIMD-Int memory or mm reg to mm reg */ + case MM: + case MMO: +#ifdef DIS_TEXT + wbit = strcmp(dp->it_name, "movd") ? MM_OPND : LONG_OPND; +#else + wbit = LONG_OPND; +#endif + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, MM_OPND, 0); + break; + + case MMOIMPL: +#ifdef DIS_TEXT + wbit = strcmp(dp->it_name, "movd") ? MM_OPND : LONG_OPND; +#else + wbit = LONG_OPND; +#endif + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode != REG_ONLY) + goto error; + + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 0); + dtrace_get_operand(x, REG_ONLY, reg, MM_OPND, 1); + mode = 0; /* change for memory access size... */ + break; + + /* MMX/SIMD-Int and SIMD-FP predicated mm reg to r32 */ + case MMO3P: + wbit = MM_OPND; + goto xmm3p; + case XMM3P: + wbit = XMM_OPND; +xmm3p: + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode != REG_ONLY) + goto error; + + THREEOPERAND(x, mode, reg, r_m, rex_prefix, wbit, LONG_OPND, 1); + NOMEM; + break; + + /* MMX/SIMD-Int predicated r32/mem to mm reg */ + case MMOPRM: + wbit = LONG_OPND; + w2 = MM_OPND; + goto xmmprm; + case XMMPRM: + wbit = LONG_OPND; + w2 = XMM_OPND; +xmmprm: + THREEOPERAND(x, mode, reg, r_m, rex_prefix, wbit, w2, 1); + break; + + /* MMX/SIMD-Int predicated mm/mem to mm reg */ + case MMOPM: + wbit = w2 = MM_OPND; + goto xmmprm; + + /* MMX/SIMD-Int mm reg to r32 */ + case MMOM3: + NOMEM; + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode != REG_ONLY) + goto error; + wbit = MM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, LONG_OPND, 0); + break; + + /* SIMD memory or xmm reg operand to xmm reg */ + case XMM: + case XMMO: + case XMMXIMPL: + wbit = XMM_OPND; + STANDARD_MODRM(x, mode, reg, r_m, rex_prefix, wbit, 0); + + if (dp->it_adrmode == XMMXIMPL && mode != REG_ONLY) + goto error; + +#ifdef DIS_TEXT + /* + * movlps and movhlps share opcodes. They differ in the + * addressing modes allowed for their operands. + * movhps and movlhps behave similarly. + */ + if (mode == REG_ONLY) { + if (strcmp(dp->it_name, "movlps") == 0) + (void) strncpy(x->d86_mneu, "movhlps", OPLEN); + else if (strcmp(dp->it_name, "movhps") == 0) + (void) strncpy(x->d86_mneu, "movlhps", OPLEN); + } +#endif + if (dp->it_adrmode == XMMXIMPL) + mode = 0; /* change for memory access size... */ + break; + + /* SIMD xmm reg to memory or xmm reg */ + case XMMS: + case XMMOS: + case XMMMS: + case XMMOMS: + dtrace_get_modrm(x, &mode, ®, &r_m); +#ifdef DIS_TEXT + if ((strcmp(dp->it_name, "movlps") == 0 || + strcmp(dp->it_name, "movhps") == 0 || + strcmp(dp->it_name, "movntps") == 0) && + mode == REG_ONLY) + goto error; +#endif + wbit = XMM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 1); + break; + + /* SIMD memory to xmm reg */ + case XMMM: + case XMMOM: + wbit = XMM_OPND; + dtrace_get_modrm(x, &mode, ®, &r_m); +#ifdef DIS_TEXT + if (mode == REG_ONLY) { + if (strcmp(dp->it_name, "movhps") == 0) + (void) strncpy(x->d86_mneu, "movlhps", OPLEN); + else + goto error; + } +#endif + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 0); + break; + + /* SIMD memory or r32 to xmm reg */ + case XMM3MX: + wbit = LONG_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 0); + break; + + case XMM3MXS: + wbit = LONG_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 1); + break; + + /* SIMD memory or mm reg to xmm reg */ + case XMMOMX: + /* SIMD mm to xmm */ + case XMMMX: + wbit = MM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 0); + break; + + /* SIMD memory or xmm reg to mm reg */ + case XMMXMM: + case XMMOXMM: + case XMMXM: + wbit = XMM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, MM_OPND, 0); + break; + + + /* SIMD memory or xmm reg to r32 */ + case XMMXM3: + wbit = XMM_OPND; + MIXED_MM(x, mode, reg, r_m, rex_prefix, wbit, LONG_OPND, 0); + break; + + /* SIMD xmm to r32 */ + case XMMX3: + case XMMOX3: + dtrace_get_modrm(x, &mode, ®, &r_m); + if (mode != REG_ONLY) + goto error; + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, mode, r_m, XMM_OPND, 0); + dtrace_get_operand(x, REG_ONLY, reg, LONG_OPND, 1); + NOMEM; + break; + + /* SIMD predicated memory or xmm reg with/to xmm reg */ + case XMMP: + case XMMOPM: + wbit = XMM_OPND; + THREEOPERAND(x, mode, reg, r_m, rex_prefix, wbit, XMM_OPND, 1); + +#ifdef DIS_TEXT + /* + * cmpps and cmpss vary their instruction name based + * on the value of imm8. Other XMMP instructions, + * such as shufps, require explicit specification of + * the predicate. + */ + if (dp->it_name[0] == 'c' && + dp->it_name[1] == 'm' && + dp->it_name[2] == 'p' && + strlen(dp->it_name) == 5) { + uchar_t pred = x->d86_opnd[0].d86_value & 0xff; + + if (pred >= (sizeof (dis_PREDSUFFIX) / sizeof (char *))) + goto error; + + (void) strncpy(x->d86_mneu, "cmp", OPLEN); + (void) strlcat(x->d86_mneu, dis_PREDSUFFIX[pred], + OPLEN); + (void) strlcat(x->d86_mneu, + dp->it_name + strlen(dp->it_name) - 2, + OPLEN); + x->d86_opnd[0] = x->d86_opnd[1]; + x->d86_opnd[1] = x->d86_opnd[2]; + x->d86_numopnds = 2; + } +#endif + break; + + /* immediate operand to accumulator */ + case IA: + wbit = WBIT(opcode2); + dtrace_get_operand(x, REG_ONLY, EAX_REGNO, wbit, 1); + dtrace_imm_opnd(x, wbit, OPSIZE(opnd_size, wbit), 0); + NOMEM; + break; + + /* memory or register operand to accumulator */ + case MA: + wbit = WBIT(opcode2); + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, wbit, 0); + break; + + /* si register to di register used to reference memory */ + case SD: +#ifdef DIS_TEXT + dtrace_check_override(x, 0); + x->d86_numopnds = 2; + if (addr_size == SIZE64) { + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%rsi)", + OPLEN); + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%rdi)", + OPLEN); + } else if (addr_size == SIZE32) { + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%esi)", + OPLEN); + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%edi)", + OPLEN); + } else { + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%si)", + OPLEN); + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%di)", + OPLEN); + } +#endif + wbit = LONG_OPND; + break; + + /* accumulator to di register */ + case AD: + wbit = WBIT(opcode2); +#ifdef DIS_TEXT + dtrace_check_override(x, 1); + x->d86_numopnds = 2; + dtrace_get_operand(x, REG_ONLY, EAX_REGNO, wbit, 0); + if (addr_size == SIZE64) + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%rdi)", + OPLEN); + else if (addr_size == SIZE32) + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%edi)", + OPLEN); + else + (void) strlcat(x->d86_opnd[1].d86_opnd, "(%di)", + OPLEN); +#endif + break; + + /* si register to accumulator */ + case SA: + wbit = WBIT(opcode2); +#ifdef DIS_TEXT + dtrace_check_override(x, 0); + x->d86_numopnds = 2; + if (addr_size == SIZE64) + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%rsi)", + OPLEN); + else if (addr_size == SIZE32) + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%esi)", + OPLEN); + else + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%si)", + OPLEN); + dtrace_get_operand(x, REG_ONLY, EAX_REGNO, wbit, 1); +#endif + break; + + /* + * single operand, a 16/32 bit displacement + */ + case D: + wbit = LONG_OPND; + dtrace_disp_opnd(x, wbit, OPSIZE(opnd_size, LONG_OPND), 0); + NOMEM; + break; + + /* jmp/call indirect to memory or register operand */ + case INM: +#ifdef DIS_TEXT + (void) strlcat(x->d86_opnd[0].d86_prefix, "*", OPLEN); +#endif + dtrace_rex_adjust(rex_prefix, mode, NULL, &r_m); + dtrace_get_operand(x, mode, r_m, LONG_OPND, 0); + wbit = LONG_OPND; + break; + + /* + * for long jumps and long calls -- a new code segment + * register and an offset in IP -- stored in object + * code in reverse order. Note - not valid in amd64 + */ + case SO: + dtrace_check_override(x, 1); + wbit = LONG_OPND; + dtrace_imm_opnd(x, wbit, OPSIZE(opnd_size, LONG_OPND), 1); +#ifdef DIS_TEXT + x->d86_opnd[1].d86_mode = MODE_SIGNED; +#endif + /* will now get segment operand */ + dtrace_imm_opnd(x, wbit, 2, 0); + break; + + /* + * jmp/call. single operand, 8 bit displacement. + * added to current EIP in 'compofff' + */ + case BD: + dtrace_disp_opnd(x, BYTE_OPND, 1, 0); + NOMEM; + break; + + /* single 32/16 bit immediate operand */ + case I: + wbit = LONG_OPND; + dtrace_imm_opnd(x, wbit, OPSIZE(opnd_size, LONG_OPND), 0); + break; + + /* single 8 bit immediate operand */ + case Ib: + wbit = LONG_OPND; + dtrace_imm_opnd(x, wbit, 1, 0); + break; + + case ENTER: + wbit = LONG_OPND; + dtrace_imm_opnd(x, wbit, 2, 0); + dtrace_imm_opnd(x, wbit, 1, 1); + switch (opnd_size) { + case SIZE64: + x->d86_memsize = (x->d86_opnd[1].d86_value + 1) * 8; + break; + case SIZE32: + x->d86_memsize = (x->d86_opnd[1].d86_value + 1) * 4; + break; + case SIZE16: + x->d86_memsize = (x->d86_opnd[1].d86_value + 1) * 2; + break; + } + + break; + + /* 16-bit immediate operand */ + case RET: + wbit = LONG_OPND; + dtrace_imm_opnd(x, wbit, 2, 0); + break; + + /* single 8 bit port operand */ + case P: + dtrace_check_override(x, 0); + dtrace_imm_opnd(x, BYTE_OPND, 1, 0); + NOMEM; + break; + + /* single operand, dx register (variable port instruction) */ + case V: + x->d86_numopnds = 1; + dtrace_check_override(x, 0); +#ifdef DIS_TEXT + (void) strlcat(x->d86_opnd[0].d86_opnd, "(%dx)", OPLEN); +#endif + NOMEM; + break; + + /* + * The int instruction, which has two forms: + * int 3 (breakpoint) or + * int n, where n is indicated in the subsequent + * byte (format Ib). The int 3 instruction (opcode 0xCC), + * where, although the 3 looks like an operand, + * it is implied by the opcode. It must be converted + * to the correct base and output. + */ + case INT3: +#ifdef DIS_TEXT + x->d86_numopnds = 1; + x->d86_opnd[0].d86_mode = MODE_SIGNED; + x->d86_opnd[0].d86_value_size = 1; + x->d86_opnd[0].d86_value = 3; +#endif + NOMEM; + break; + + /* single 8 bit immediate operand */ + case INTx: + dtrace_imm_opnd(x, BYTE_OPND, 1, 0); + NOMEM; + break; + + /* an unused byte must be discarded */ + case U: + if (x->d86_get_byte(x->d86_data) < 0) + goto error; + x->d86_len++; + NOMEM; + break; + + case CBW: +#ifdef DIS_TEXT + if (opnd_size == SIZE16) + (void) strlcat(x->d86_mneu, "cbtw", OPLEN); + else if (opnd_size == SIZE32) + (void) strlcat(x->d86_mneu, "cwtl", OPLEN); + else + (void) strlcat(x->d86_mneu, "cltq", OPLEN); +#endif + wbit = LONG_OPND; + NOMEM; + break; + + case CWD: +#ifdef DIS_TEXT + if (opnd_size == SIZE16) + (void) strlcat(x->d86_mneu, "cwtd", OPLEN); + else if (opnd_size == SIZE32) + (void) strlcat(x->d86_mneu, "cltd", OPLEN); + else + (void) strlcat(x->d86_mneu, "cqtd", OPLEN); +#endif + wbit = LONG_OPND; + NOMEM; + break; + + case XMMSFNC: + /* + * sfence is sfence if mode is REG_ONLY. If mode isn't + * REG_ONLY, mnemonic should be 'clflush'. + */ + dtrace_get_modrm(x, &mode, ®, &r_m); + + /* sfence doesn't take operands */ +#ifdef DIS_TEXT + if (mode == REG_ONLY) { + (void) strlcat(x->d86_mneu, "sfence", OPLEN); + } else { + (void) strlcat(x->d86_mneu, "clflush", OPLEN); + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, mode, r_m, BYTE_OPND, 0); + NOMEM; + } +#else + if (mode != REG_ONLY) { + dtrace_rex_adjust(rex_prefix, mode, ®, &r_m); + dtrace_get_operand(x, mode, r_m, BYTE_OPND, 0); + NOMEM; + } +#endif + break; + + /* + * no disassembly, the mnemonic was all there was so go on + */ + case NORM: + if (dp->it_invalid32 && cpu_mode != SIZE64) + goto error; + NOMEM; + /*FALLTHROUGH*/ + case IMPLMEM: + break; + + case XMMFENCE: + /* + * Only the following exact byte sequences are allowed: + * + * 0f ae e8 lfence + * 0f ae f0 mfence + */ + if ((uint8_t)x->d86_bytes[x->d86_len - 1] != 0xe8 && + (uint8_t)x->d86_bytes[x->d86_len - 1] != 0xf0) + goto error; + + break; + + + /* float reg */ + case F: +#ifdef DIS_TEXT + x->d86_numopnds = 1; + (void) strlcat(x->d86_opnd[0].d86_opnd, "%st(X)", OPLEN); + x->d86_opnd[0].d86_opnd[4] = r_m + '0'; +#endif + NOMEM; + break; + + /* float reg to float reg, with ret bit present */ + case FF: + vbit = opcode2 >> 2 & 0x1; /* vbit = 1: st -> st(i) */ + /*FALLTHROUGH*/ + case FFC: /* case for vbit always = 0 */ +#ifdef DIS_TEXT + x->d86_numopnds = 2; + (void) strlcat(x->d86_opnd[1 - vbit].d86_opnd, "%st", OPLEN); + (void) strlcat(x->d86_opnd[vbit].d86_opnd, "%st(X)", OPLEN); + x->d86_opnd[vbit].d86_opnd[4] = r_m + '0'; +#endif + NOMEM; + break; + + /* an invalid op code */ + case AM: + case DM: + case OVERRIDE: + case PREFIX: + case UNKNOWN: + NOMEM; + default: + goto error; + } /* end switch */ + if (x->d86_error) + goto error; + +done: +#ifdef DIS_MEM + /* + * compute the size of any memory accessed by the instruction + */ + if (x->d86_memsize != 0) { + return (0); + } else if (dp->it_stackop) { + switch (opnd_size) { + case SIZE16: + x->d86_memsize = 2; + break; + case SIZE32: + x->d86_memsize = 4; + break; + case SIZE64: + x->d86_memsize = 8; + break; + } + } else if (nomem || mode == REG_ONLY) { + x->d86_memsize = 0; + + } else if (dp->it_size != 0) { + /* + * In 64 bit mode descriptor table entries + * go up to 10 bytes and popf/pushf are always 8 bytes + */ + if (x->d86_mode == SIZE64 && dp->it_size == 6) + x->d86_memsize = 10; + else if (x->d86_mode == SIZE64 && opcode1 == 0x9 && + (opcode2 == 0xc || opcode2 == 0xd)) + x->d86_memsize = 8; + else + x->d86_memsize = dp->it_size; + + } else if (wbit == 0) { + x->d86_memsize = 1; + + } else if (wbit == LONG_OPND) { + if (opnd_size == SIZE64) + x->d86_memsize = 8; + else if (opnd_size == SIZE32) + x->d86_memsize = 4; + else + x->d86_memsize = 2; + + } else if (wbit == SEG_OPND) { + x->d86_memsize = 4; + + } else { + x->d86_memsize = 8; + } +#endif + return (0); + +error: +#ifdef DIS_TEXT + (void) strlcat(x->d86_mneu, "undef", OPLEN); +#endif + return (1); +} + +#ifdef DIS_TEXT + +/* + * Some instructions should have immediate operands printed + * as unsigned integers. We compare against this table. + */ +static char *unsigned_ops[] = { + "or", "and", "xor", "test", "in", "out", "lcall", "ljmp", + "rcr", "rcl", "ror", "rol", "shl", "shr", "sal", "psr", "psl", + 0 +}; + +static int +isunsigned_op(char *opcode) +{ + char *where; + int i; + int is_unsigned = 0; + + /* + * Work back to start of last mnemonic, since we may have + * prefixes on some opcodes. + */ + where = opcode + strlen(opcode) - 1; + while (where > opcode && *where != ' ') + --where; + if (*where == ' ') + ++where; + + for (i = 0; unsigned_ops[i]; ++i) { + if (strncmp(where, unsigned_ops[i], + strlen(unsigned_ops[i]))) + continue; + is_unsigned = 1; + break; + } + return (is_unsigned); +} + +/* ARGSUSED */ +void +dtrace_disx86_str(dis86_t *dis, uint_t mode, uintptr_t pc, char *buf, + size_t buflen) +{ + int i; + + dis->d86_sprintf_func(buf, buflen, "%-6s ", dis->d86_mneu); + + /* + * For PC-relative jumps, the pc is really the next pc after executing + * this instruction, so increment it appropriately. + */ + pc += dis->d86_len; + + for (i = 0; i < dis->d86_numopnds; i++) { + d86opnd_t *op = &dis->d86_opnd[i]; + int64_t sv; + uint64_t mask; + + if (i != 0) + (void) strlcat(buf, ",", buflen); + + (void) strlcat(buf, op->d86_prefix, buflen); + + sv = op->d86_value; + + switch (op->d86_mode) { + + case MODE_NONE: + + (void) strlcat(buf, op->d86_opnd, buflen); + break; + + case MODE_SIGNED: + case MODE_IMPLIED: + case MODE_OFFSET: + + if (dis->d86_seg_prefix) + (void) strlcat(buf, dis->d86_seg_prefix, + buflen); + + switch (op->d86_value_size) { + case 1: + sv = (int8_t)sv; + mask = 0xff; + break; + case 2: + sv = (int16_t)sv; + mask = 0xffff; + break; + case 4: + sv = (int32_t)sv; + mask = 0xffffffff; + break; + case 8: + mask = 0xffffffffffffffffULL; + break; + } + + if (op->d86_mode == MODE_SIGNED || + op->d86_mode == MODE_IMPLIED) + (void) strlcat(buf, "$", buflen); + + if (sv < 0 && sv > -0xffff && + !isunsigned_op(dis->d86_mneu)) { + dis->d86_sprintf_func(buf + strlen(buf), + buflen - strlen(buf), + (dis->d86_flags & DIS_OP_OCTAL) ? + "-0%llo" : "-0x%llx", -sv & mask); + } else { + dis->d86_sprintf_func(buf + strlen(buf), + buflen - strlen(buf), + (dis->d86_flags & DIS_OP_OCTAL) ? + "0%llo" : "0x%llx", sv & mask); + } + (void) strlcat(buf, op->d86_opnd, buflen); + break; + + case MODE_IPREL: + + switch (op->d86_value_size) { + case 1: + sv = (int8_t)sv; + break; + case 2: + sv = (int16_t)sv; + break; + case 4: + sv = (int32_t)sv; + break; + } + + if (sv < 0) + dis->d86_sprintf_func(buf + strlen(buf), + buflen - strlen(buf), + (dis->d86_flags & DIS_OP_OCTAL) ? + "-0%llo" : "-0x%llx", -sv - dis->d86_len); + else + dis->d86_sprintf_func(buf + strlen(buf), + buflen - strlen(buf), + (dis->d86_flags & DIS_OP_OCTAL) ? + "+0%llo" : "+0x%llx", sv + dis->d86_len); + + (void) strlcat(buf, "\t<", buflen); + + if (dis->d86_sym_lookup == NULL || + dis->d86_sym_lookup(dis->d86_data, pc + sv, + buf + strlen(buf), buflen - strlen(buf)) != 0) + dis->d86_sprintf_func(buf + strlen(buf), + buflen - strlen(buf), + (dis->d86_flags & DIS_OP_OCTAL) ? + "0%llo" : "0x%llx", pc + sv); + + (void) strlcat(buf, ">", buflen); + + break; + } + } +} + +#endif /* DIS_TEXT */ diff --git a/sys/cddl/dev/dtrace/i386/dis_tables.h b/sys/cddl/dev/dtrace/i386/dis_tables.h new file mode 100644 index 000000000000..b45a8c50e738 --- /dev/null +++ b/sys/cddl/dev/dtrace/i386/dis_tables.h @@ -0,0 +1,112 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +/* Copyright (c) 1988 AT&T */ +/* All Rights Reserved */ + + +#ifndef _DIS_TABLES_H +#define _DIS_TABLES_H + +#if defined(sun) +#pragma ident "@(#)dis_tables.h 1.7 06/03/02 SMI" +#endif + +/* + * Constants and prototypes for the IA32 disassembler backend. See dis_tables.c + * for usage information and documentation. + */ + +#ifdef __cplusplus +extern "C" { +#endif + +#include <sys/types.h> +#include <sys/param.h> + +/* + * values for cpu mode + */ +#define SIZE16 1 +#define SIZE32 2 +#define SIZE64 3 + +#define OPLEN 256 +#define PFIXLEN 8 +#define NCPS 12 /* number of chars per symbol */ + +/* + * data structures that must be provided to dtrace_dis86() + */ +typedef struct d86opnd { + char d86_opnd[OPLEN]; /* symbolic rep of operand */ + char d86_prefix[PFIXLEN]; /* any prefix string or "" */ + uint_t d86_mode; /* mode for immediate */ + uint_t d86_value_size; /* size in bytes of d86_value */ + uint64_t d86_value; /* immediate value of opnd */ +} d86opnd_t; + +typedef struct dis86 { + uint_t d86_mode; + uint_t d86_error; + uint_t d86_len; /* instruction length */ + int d86_rmindex; /* index of modrm byte or -1 */ + uint_t d86_memsize; /* size of memory referenced */ + char d86_bytes[16]; /* bytes of instruction */ + char d86_mneu[OPLEN]; + uint_t d86_numopnds; + uint_t d86_rex_prefix; /* value of REX prefix if !0 */ + char *d86_seg_prefix; /* segment prefix, if any */ + uint_t d86_opnd_size; + uint_t d86_addr_size; + uint_t d86_got_modrm; + struct d86opnd d86_opnd[3]; /* up to 3 operands */ + int (*d86_check_func)(void *); + int (*d86_get_byte)(void *); +#ifdef DIS_TEXT + int (*d86_sym_lookup)(void *, uint64_t, char *, size_t); + int (*d86_sprintf_func)(char *, size_t, const char *, ...); + int d86_flags; + uint_t d86_imm_bytes; +#endif + void *d86_data; +} dis86_t; + +extern int dtrace_disx86(dis86_t *x, uint_t cpu_mode); + +#define DIS_OP_OCTAL 0x1 /* Print all numbers in octal */ + +#ifdef DIS_TEXT +extern void dtrace_disx86_str(dis86_t *x, uint_t cpu_mode, uintptr_t pc, + char *buf, size_t len); +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* _DIS_TABLES_H */ diff --git a/sys/cddl/dev/dtrace/i386/dtrace_asm.S b/sys/cddl/dev/dtrace/i386/dtrace_asm.S new file mode 100644 index 000000000000..787a3c8b2ffe --- /dev/null +++ b/sys/cddl/dev/dtrace/i386/dtrace_asm.S @@ -0,0 +1,527 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ +/* + * Copyright 2004 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#define _ASM + +#include <machine/asmacros.h> +#include <sys/cpuvar_defs.h> +#include <sys/dtrace.h> + +#include "assym.s" + + .globl calltrap + .type calltrap,@function + ENTRY(dtrace_invop_start) + + pushl %eax /* push %eax -- may be return value */ + pushl %esp /* push stack pointer */ + addl $48, (%esp) /* adjust to incoming args */ + pushl 40(%esp) /* push calling EIP */ + + /* + * Call dtrace_invop to let it check if the exception was + * a fbt one. The return value in %eax will tell us what + * dtrace_invop wants us to do. + */ + call dtrace_invop + + /* + * We pushed 3 times for the arguments to dtrace_invop, + * so we need to increment the stack pointer to get rid of + * those values. + */ + addl $12, %esp + ALTENTRY(dtrace_invop_callsite) + cmpl $DTRACE_INVOP_PUSHL_EBP, %eax + je invop_push + cmpl $DTRACE_INVOP_POPL_EBP, %eax + je invop_pop + cmpl $DTRACE_INVOP_LEAVE, %eax + je invop_leave + cmpl $DTRACE_INVOP_NOP, %eax + je invop_nop + + /* When all else fails handle the trap in the usual way. */ + jmpl *dtrace_invop_calltrap_addr + +invop_push: + /* + * We must emulate a "pushl %ebp". To do this, we pull the stack + * down 4 bytes, and then store the base pointer. + */ + popal + subl $4, %esp /* make room for %ebp */ + pushl %eax /* push temp */ + movl 8(%esp), %eax /* load calling EIP */ + incl %eax /* increment over LOCK prefix */ + movl %eax, 4(%esp) /* store calling EIP */ + movl 12(%esp), %eax /* load calling CS */ + movl %eax, 8(%esp) /* store calling CS */ + movl 16(%esp), %eax /* load calling EFLAGS */ + movl %eax, 12(%esp) /* store calling EFLAGS */ + movl %ebp, 16(%esp) /* push %ebp */ + popl %eax /* pop off temp */ + iret /* Return from interrupt. */ +invop_pop: + /* + * We must emulate a "popl %ebp". To do this, we do the opposite of + * the above: we remove the %ebp from the stack, and squeeze up the + * saved state from the trap. + */ + popal + pushl %eax /* push temp */ + movl 16(%esp), %ebp /* pop %ebp */ + movl 12(%esp), %eax /* load calling EFLAGS */ + movl %eax, 16(%esp) /* store calling EFLAGS */ + movl 8(%esp), %eax /* load calling CS */ + movl %eax, 12(%esp) /* store calling CS */ + movl 4(%esp), %eax /* load calling EIP */ + incl %eax /* increment over LOCK prefix */ + movl %eax, 8(%esp) /* store calling EIP */ + popl %eax /* pop off temp */ + addl $4, %esp /* adjust stack pointer */ + iret /* Return from interrupt. */ +invop_leave: + /* + * We must emulate a "leave", which is the same as a "movl %ebp, %esp" + * followed by a "popl %ebp". This looks similar to the above, but + * requires two temporaries: one for the new base pointer, and one + * for the staging register. + */ + popa + pushl %eax /* push temp */ + pushl %ebx /* push temp */ + movl %ebp, %ebx /* set temp to old %ebp */ + movl (%ebx), %ebp /* pop %ebp */ + movl 16(%esp), %eax /* load calling EFLAGS */ + movl %eax, (%ebx) /* store calling EFLAGS */ + movl 12(%esp), %eax /* load calling CS */ + movl %eax, -4(%ebx) /* store calling CS */ + movl 8(%esp), %eax /* load calling EIP */ + incl %eax /* increment over LOCK prefix */ + movl %eax, -8(%ebx) /* store calling EIP */ + movl %ebx, -4(%esp) /* temporarily store new %esp */ + popl %ebx /* pop off temp */ + popl %eax /* pop off temp */ + movl -12(%esp), %esp /* set stack pointer */ + subl $8, %esp /* adjust for three pushes, one pop */ + iret /* return from interrupt */ +invop_nop: + /* + * We must emulate a "nop". This is obviously not hard: we need only + * advance the %eip by one. + */ + popa + incl (%esp) + iret /* return from interrupt */ + + END(dtrace_invop_start) + +/* +void dtrace_invop_init(void) +*/ + ENTRY(dtrace_invop_init) + movl $dtrace_invop_start, dtrace_invop_jump_addr + ret + END(dtrace_invop_init) + +/* +void dtrace_invop_uninit(void) +*/ + ENTRY(dtrace_invop_uninit) + movl $0, dtrace_invop_jump_addr + ret + END(dtrace_invop_uninit) + +/* +greg_t dtrace_getfp(void) +*/ + + ENTRY(dtrace_getfp) + movl %ebp, %eax + ret + END(dtrace_getfp) + +/* +uint32_t dtrace_cas32(uint32_t *target, uint32_t cmp, uint32_t new) +*/ + + ENTRY(dtrace_cas32) + ALTENTRY(dtrace_casptr) + movl 4(%esp), %edx + movl 8(%esp), %eax + movl 12(%esp), %ecx + lock + cmpxchgl %ecx, (%edx) + ret + END(dtrace_casptr) + END(dtrace_cas32) + +/* +uintptr_t dtrace_caller(int aframes) +*/ + + ENTRY(dtrace_caller) + movl $-1, %eax + ret + END(dtrace_caller) + +/* +void dtrace_copy(uintptr_t src, uintptr_t dest, size_t size) +*/ + + ENTRY(dtrace_copy) + pushl %ebp + movl %esp, %ebp + pushl %esi + pushl %edi + + movl 8(%ebp), %esi /* Load source address */ + movl 12(%ebp), %edi /* Load destination address */ + movl 16(%ebp), %ecx /* Load count */ + repz /* Repeat for count... */ + smovb /* move from %ds:si to %es:di */ + + popl %edi + popl %esi + movl %ebp, %esp + popl %ebp + ret + END(dtrace_copy) + +/* +void dtrace_copystr(uintptr_t uaddr, uintptr_t kaddr, size_t size) +*/ + + ENTRY(dtrace_copystr) + + pushl %ebp /* Setup stack frame */ + movl %esp, %ebp + pushl %ebx /* Save registers */ + + movl 8(%ebp), %ebx /* Load source address */ + movl 12(%ebp), %edx /* Load destination address */ + movl 16(%ebp), %ecx /* Load count */ + +0: + movb (%ebx), %al /* Load from source */ + movb %al, (%edx) /* Store to destination */ + incl %ebx /* Increment source pointer */ + incl %edx /* Increment destination pointer */ + decl %ecx /* Decrement remaining count */ + cmpb $0, %al + je 1f + cmpl $0, %ecx + jne 0b + +1: + popl %ebx + movl %ebp, %esp + popl %ebp + ret + + END(dtrace_copystr) + +/* +uintptr_t dtrace_fulword(void *addr) +*/ + + ENTRY(dtrace_fulword) + movl 4(%esp), %ecx + xorl %eax, %eax + movl (%ecx), %eax + ret + END(dtrace_fulword) + +/* +uint8_t dtrace_fuword8_nocheck(void *addr) +*/ + + ENTRY(dtrace_fuword8_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + movzbl (%ecx), %eax + ret + END(dtrace_fuword8_nocheck) + +/* +uint16_t dtrace_fuword16_nocheck(void *addr) +*/ + + ENTRY(dtrace_fuword16_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + movzwl (%ecx), %eax + ret + END(dtrace_fuword16_nocheck) + +/* +uint32_t dtrace_fuword32_nocheck(void *addr) +*/ + + ENTRY(dtrace_fuword32_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + movl (%ecx), %eax + ret + END(dtrace_fuword32_nocheck) + +/* +uint64_t dtrace_fuword64_nocheck(void *addr) +*/ + + ENTRY(dtrace_fuword64_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + xorl %edx, %edx + movl (%ecx), %eax + movl 4(%ecx), %edx + ret + END(dtrace_fuword64_nocheck) + +/* +void dtrace_probe_error(dtrace_state_t *state, dtrace_epid_t epid, int which, int fault, int fltoffs, uintptr_t illval) +*/ + + ENTRY(dtrace_probe_error) + pushl %ebp + movl %esp, %ebp + pushl 0x1c(%ebp) + pushl 0x18(%ebp) + pushl 0x14(%ebp) + pushl 0x10(%ebp) + pushl 0xc(%ebp) + pushl 0x8(%ebp) + pushl dtrace_probeid_error + call dtrace_probe + movl %ebp, %esp + popl %ebp + ret + END(dtrace_probe_error) + +/* +void dtrace_membar_producer(void) +*/ + + ENTRY(dtrace_membar_producer) + rep; ret /* use 2 byte return instruction when branch target */ + /* AMD Software Optimization Guide - Section 6.2 */ + END(dtrace_membar_producer) + +/* +void dtrace_membar_consumer(void) +*/ + + ENTRY(dtrace_membar_consumer) + rep; ret /* use 2 byte return instruction when branch target */ + /* AMD Software Optimization Guide - Section 6.2 */ + END(dtrace_membar_consumer) + +/* +dtrace_icookie_t dtrace_interrupt_disable(void) +*/ + ENTRY(dtrace_interrupt_disable) + pushfl + popl %eax + cli + ret + END(dtrace_interrupt_disable) + +/* +void dtrace_interrupt_enable(dtrace_icookie_t cookie) +*/ + ENTRY(dtrace_interrupt_enable) + movl 4(%esp), %eax + pushl %eax + popfl + ret + END(dtrace_interrupt_enable) + +/* + * The panic() and cmn_err() functions invoke vpanic() as a common entry point + * into the panic code implemented in panicsys(). vpanic() is responsible + * for passing through the format string and arguments, and constructing a + * regs structure on the stack into which it saves the current register + * values. If we are not dying due to a fatal trap, these registers will + * then be preserved in panicbuf as the current processor state. Before + * invoking panicsys(), vpanic() activates the first panic trigger (see + * common/os/panic.c) and switches to the panic_stack if successful. Note that + * DTrace takes a slightly different panic path if it must panic from probe + * context. Instead of calling panic, it calls into dtrace_vpanic(), which + * sets up the initial stack as vpanic does, calls dtrace_panic_trigger(), and + * branches back into vpanic(). + */ +/* +void vpanic(const char *format, va_list alist) +*/ + ENTRY(vpanic) /* Initial stack layout: */ + + pushl %ebp /* | %eip | 20 */ + movl %esp, %ebp /* | %ebp | 16 */ + pushl %eax /* | %eax | 12 */ + pushl %ebx /* | %ebx | 8 */ + pushl %ecx /* | %ecx | 4 */ + pushl %edx /* | %edx | 0 */ + + movl %esp, %ebx /* %ebx = current stack pointer */ + + lea panic_quiesce, %eax /* %eax = &panic_quiesce */ + pushl %eax /* push &panic_quiesce */ + call panic_trigger /* %eax = panic_trigger() */ + addl $4, %esp /* reset stack pointer */ + +vpanic_common: + cmpl $0, %eax /* if (%eax == 0) */ + je 0f /* goto 0f; */ + + /* + * If panic_trigger() was successful, we are the first to initiate a + * panic: we now switch to the reserved panic_stack before continuing. + */ + lea panic_stack, %esp /* %esp = panic_stack */ + addl $PANICSTKSIZE, %esp /* %esp += PANICSTKSIZE */ + +0: subl $REGSIZE, %esp /* allocate struct regs */ + + /* + * Now that we've got everything set up, store the register values as + * they were when we entered vpanic() to the designated location in + * the regs structure we allocated on the stack. + */ +#ifdef notyet + mov %gs, %edx + mov %edx, REGOFF_GS(%esp) + mov %fs, %edx + mov %edx, REGOFF_FS(%esp) + mov %es, %edx + mov %edx, REGOFF_ES(%esp) + mov %ds, %edx + mov %edx, REGOFF_DS(%esp) + movl %edi, REGOFF_EDI(%esp) + movl %esi, REGOFF_ESI(%esp) + movl 16(%ebx), %ecx + movl %ecx, REGOFF_EBP(%esp) + movl %ebx, %ecx + addl $20, %ecx + movl %ecx, REGOFF_ESP(%esp) + movl 8(%ebx), %ecx + movl %ecx, REGOFF_EBX(%esp) + movl 0(%ebx), %ecx + movl %ecx, REGOFF_EDX(%esp) + movl 4(%ebx), %ecx + movl %ecx, REGOFF_ECX(%esp) + movl 12(%ebx), %ecx + movl %ecx, REGOFF_EAX(%esp) + movl $0, REGOFF_TRAPNO(%esp) + movl $0, REGOFF_ERR(%esp) + lea vpanic, %ecx + movl %ecx, REGOFF_EIP(%esp) + mov %cs, %edx + movl %edx, REGOFF_CS(%esp) + pushfl + popl %ecx + movl %ecx, REGOFF_EFL(%esp) + movl $0, REGOFF_UESP(%esp) + mov %ss, %edx + movl %edx, REGOFF_SS(%esp) + + movl %esp, %ecx /* %ecx = ®s */ + pushl %eax /* push on_panic_stack */ + pushl %ecx /* push ®s */ + movl 12(%ebp), %ecx /* %ecx = alist */ + pushl %ecx /* push alist */ + movl 8(%ebp), %ecx /* %ecx = format */ + pushl %ecx /* push format */ + call panicsys /* panicsys(); */ + addl $16, %esp /* pop arguments */ + + addl $REGSIZE, %esp +#endif + popl %edx + popl %ecx + popl %ebx + popl %eax + leave + ret + END(vpanic) + +/* +void dtrace_vpanic(const char *format, va_list alist) +*/ + ENTRY(dtrace_vpanic) /* Initial stack layout: */ + + pushl %ebp /* | %eip | 20 */ + movl %esp, %ebp /* | %ebp | 16 */ + pushl %eax /* | %eax | 12 */ + pushl %ebx /* | %ebx | 8 */ + pushl %ecx /* | %ecx | 4 */ + pushl %edx /* | %edx | 0 */ + + movl %esp, %ebx /* %ebx = current stack pointer */ + + lea panic_quiesce, %eax /* %eax = &panic_quiesce */ + pushl %eax /* push &panic_quiesce */ + call dtrace_panic_trigger /* %eax = dtrace_panic_trigger() */ + addl $4, %esp /* reset stack pointer */ + jmp vpanic_common /* jump back to common code */ + + END(dtrace_vpanic) + +/* +int +panic_trigger(int *tp) +*/ + ENTRY(panic_trigger) + xorl %eax, %eax + movl $0xdefacedd, %edx + lock + xchgl %edx, (%edi) + cmpl $0, %edx + je 0f + movl $0, %eax + ret +0: movl $1, %eax + ret + END(panic_trigger) + +/* +int +dtrace_panic_trigger(int *tp) +*/ + ENTRY(dtrace_panic_trigger) + xorl %eax, %eax + movl $0xdefacedd, %edx + lock + xchgl %edx, (%edi) + cmpl $0, %edx + je 0f + movl $0, %eax + ret +0: movl $1, %eax + ret + END(dtrace_panic_trigger) diff --git a/sys/cddl/dev/dtrace/i386/dtrace_isa.c b/sys/cddl/dev/dtrace/i386/dtrace_isa.c new file mode 100644 index 000000000000..bf891aa39e17 --- /dev/null +++ b/sys/cddl/dev/dtrace/i386/dtrace_isa.c @@ -0,0 +1,622 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ +/* + * Copyright 2005 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ +#include <sys/cdefs.h> + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/kernel.h> +#include <sys/stack.h> +#include <sys/pcpu.h> + +#include <machine/md_var.h> +#include <machine/stack.h> + +#include <vm/vm.h> +#include <vm/vm_param.h> +#include <vm/pmap.h> + +extern uintptr_t kernbase; +uintptr_t kernelbase = (uintptr_t) &kernbase; + +#define INKERNEL(va) (((vm_offset_t)(va)) >= USRSTACK && \ + ((vm_offset_t)(va)) < VM_MAX_KERNEL_ADDRESS) + +uint8_t dtrace_fuword8_nocheck(void *); +uint16_t dtrace_fuword16_nocheck(void *); +uint32_t dtrace_fuword32_nocheck(void *); +uint64_t dtrace_fuword64_nocheck(void *); + +void +dtrace_getpcstack(pc_t *pcstack, int pcstack_limit, int aframes, + uint32_t *intrpc) +{ + int depth = 0; + register_t ebp; + struct i386_frame *frame; + vm_offset_t callpc; + pc_t caller = (pc_t) solaris_cpu[curcpu].cpu_dtrace_caller; + + if (intrpc != 0) + pcstack[depth++] = (pc_t) intrpc; + + aframes++; + + __asm __volatile("movl %%ebp,%0" : "=r" (ebp)); + + frame = (struct i386_frame *)ebp; + while (depth < pcstack_limit) { + if (!INKERNEL(frame)) + break; + + callpc = frame->f_retaddr; + + if (!INKERNEL(callpc)) + break; + + if (aframes > 0) { + aframes--; + if ((aframes == 0) && (caller != 0)) { + pcstack[depth++] = caller; + } + } + else { + pcstack[depth++] = callpc; + } + + if (frame->f_frame <= frame || + (vm_offset_t)frame->f_frame >= + (vm_offset_t)ebp + KSTACK_PAGES * PAGE_SIZE) + break; + frame = frame->f_frame; + } + + for (; depth < pcstack_limit; depth++) { + pcstack[depth] = 0; + } +} + +#ifdef notyet +static int +dtrace_getustack_common(uint64_t *pcstack, int pcstack_limit, uintptr_t pc, + uintptr_t sp) +{ + klwp_t *lwp = ttolwp(curthread); + proc_t *p = curproc; + uintptr_t oldcontext = lwp->lwp_oldcontext; + volatile uint16_t *flags = + (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; + size_t s1, s2; + int ret = 0; + + ASSERT(pcstack == NULL || pcstack_limit > 0); + + if (p->p_model == DATAMODEL_NATIVE) { + s1 = sizeof (struct frame) + 2 * sizeof (long); + s2 = s1 + sizeof (siginfo_t); + } else { + s1 = sizeof (struct frame32) + 3 * sizeof (int); + s2 = s1 + sizeof (siginfo32_t); + } + + while (pc != 0 && sp != 0) { + ret++; + if (pcstack != NULL) { + *pcstack++ = (uint64_t)pc; + pcstack_limit--; + if (pcstack_limit <= 0) + break; + } + + if (oldcontext == sp + s1 || oldcontext == sp + s2) { + if (p->p_model == DATAMODEL_NATIVE) { + ucontext_t *ucp = (ucontext_t *)oldcontext; + greg_t *gregs = ucp->uc_mcontext.gregs; + + sp = dtrace_fulword(&gregs[REG_FP]); + pc = dtrace_fulword(&gregs[REG_PC]); + + oldcontext = dtrace_fulword(&ucp->uc_link); + } else { + ucontext32_t *ucp = (ucontext32_t *)oldcontext; + greg32_t *gregs = ucp->uc_mcontext.gregs; + + sp = dtrace_fuword32(&gregs[EBP]); + pc = dtrace_fuword32(&gregs[EIP]); + + oldcontext = dtrace_fuword32(&ucp->uc_link); + } + } else { + if (p->p_model == DATAMODEL_NATIVE) { + struct frame *fr = (struct frame *)sp; + + pc = dtrace_fulword(&fr->fr_savpc); + sp = dtrace_fulword(&fr->fr_savfp); + } else { + struct frame32 *fr = (struct frame32 *)sp; + + pc = dtrace_fuword32(&fr->fr_savpc); + sp = dtrace_fuword32(&fr->fr_savfp); + } + } + + /* + * This is totally bogus: if we faulted, we're going to clear + * the fault and break. This is to deal with the apparently + * broken Java stacks on x86. + */ + if (*flags & CPU_DTRACE_FAULT) { + *flags &= ~CPU_DTRACE_FAULT; + break; + } + } + + return (ret); +} + +void +dtrace_getupcstack(uint64_t *pcstack, int pcstack_limit) +{ + klwp_t *lwp = ttolwp(curthread); + proc_t *p = curproc; + struct regs *rp; + uintptr_t pc, sp; + volatile uint16_t *flags = + (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; + int n; + + if (*flags & CPU_DTRACE_FAULT) + return; + + if (pcstack_limit <= 0) + return; + + /* + * If there's no user context we still need to zero the stack. + */ + if (lwp == NULL || p == NULL || (rp = lwp->lwp_regs) == NULL) + goto zero; + + *pcstack++ = (uint64_t)p->p_pid; + pcstack_limit--; + + if (pcstack_limit <= 0) + return; + + pc = rp->r_pc; + sp = rp->r_fp; + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) { + *pcstack++ = (uint64_t)pc; + pcstack_limit--; + if (pcstack_limit <= 0) + return; + + if (p->p_model == DATAMODEL_NATIVE) + pc = dtrace_fulword((void *)rp->r_sp); + else + pc = dtrace_fuword32((void *)rp->r_sp); + } + + n = dtrace_getustack_common(pcstack, pcstack_limit, pc, sp); + ASSERT(n >= 0); + ASSERT(n <= pcstack_limit); + + pcstack += n; + pcstack_limit -= n; + +zero: + while (pcstack_limit-- > 0) + *pcstack++ = NULL; +} + +int +dtrace_getustackdepth(void) +{ +} + +void +dtrace_getufpstack(uint64_t *pcstack, uint64_t *fpstack, int pcstack_limit) +{ + klwp_t *lwp = ttolwp(curthread); + proc_t *p = curproc; + struct regs *rp; + uintptr_t pc, sp, oldcontext; + volatile uint16_t *flags = + (volatile uint16_t *)&cpu_core[curcpu].cpuc_dtrace_flags; + size_t s1, s2; + + if (*flags & CPU_DTRACE_FAULT) + return; + + if (pcstack_limit <= 0) + return; + + /* + * If there's no user context we still need to zero the stack. + */ + if (lwp == NULL || p == NULL || (rp = lwp->lwp_regs) == NULL) + goto zero; + + *pcstack++ = (uint64_t)p->p_pid; + pcstack_limit--; + + if (pcstack_limit <= 0) + return; + + pc = rp->r_pc; + sp = rp->r_fp; + oldcontext = lwp->lwp_oldcontext; + + if (p->p_model == DATAMODEL_NATIVE) { + s1 = sizeof (struct frame) + 2 * sizeof (long); + s2 = s1 + sizeof (siginfo_t); + } else { + s1 = sizeof (struct frame32) + 3 * sizeof (int); + s2 = s1 + sizeof (siginfo32_t); + } + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) { + *pcstack++ = (uint64_t)pc; + *fpstack++ = 0; + pcstack_limit--; + if (pcstack_limit <= 0) + return; + + if (p->p_model == DATAMODEL_NATIVE) + pc = dtrace_fulword((void *)rp->r_sp); + else + pc = dtrace_fuword32((void *)rp->r_sp); + } + + while (pc != 0 && sp != 0) { + *pcstack++ = (uint64_t)pc; + *fpstack++ = sp; + pcstack_limit--; + if (pcstack_limit <= 0) + break; + + if (oldcontext == sp + s1 || oldcontext == sp + s2) { + if (p->p_model == DATAMODEL_NATIVE) { + ucontext_t *ucp = (ucontext_t *)oldcontext; + greg_t *gregs = ucp->uc_mcontext.gregs; + + sp = dtrace_fulword(&gregs[REG_FP]); + pc = dtrace_fulword(&gregs[REG_PC]); + + oldcontext = dtrace_fulword(&ucp->uc_link); + } else { + ucontext_t *ucp = (ucontext_t *)oldcontext; + greg_t *gregs = ucp->uc_mcontext.gregs; + + sp = dtrace_fuword32(&gregs[EBP]); + pc = dtrace_fuword32(&gregs[EIP]); + + oldcontext = dtrace_fuword32(&ucp->uc_link); + } + } else { + if (p->p_model == DATAMODEL_NATIVE) { + struct frame *fr = (struct frame *)sp; + + pc = dtrace_fulword(&fr->fr_savpc); + sp = dtrace_fulword(&fr->fr_savfp); + } else { + struct frame32 *fr = (struct frame32 *)sp; + + pc = dtrace_fuword32(&fr->fr_savpc); + sp = dtrace_fuword32(&fr->fr_savfp); + } + } + + /* + * This is totally bogus: if we faulted, we're going to clear + * the fault and break. This is to deal with the apparently + * broken Java stacks on x86. + */ + if (*flags & CPU_DTRACE_FAULT) { + *flags &= ~CPU_DTRACE_FAULT; + break; + } + } + +zero: + while (pcstack_limit-- > 0) + *pcstack++ = NULL; +} +#endif + +uint64_t +dtrace_getarg(int arg, int aframes) +{ + uintptr_t val; + struct i386_frame *fp = (struct i386_frame *)dtrace_getfp(); + uintptr_t *stack; + int i; + + for (i = 1; i <= aframes; i++) { + fp = fp->f_frame; + + if (fp->f_retaddr == (long)dtrace_invop_callsite) { + /* + * If we pass through the invalid op handler, we will + * use the pointer that it passed to the stack as the + * second argument to dtrace_invop() as the pointer to + * the stack. When using this stack, we must step + * beyond the EIP/RIP that was pushed when the trap was + * taken -- hence the "+ 1" below. + */ + stack = ((uintptr_t **)&fp[1])[1] + 1; + goto load; + } + + } + + /* + * We know that we did not come through a trap to get into + * dtrace_probe() -- the provider simply called dtrace_probe() + * directly. As this is the case, we need to shift the argument + * that we're looking for: the probe ID is the first argument to + * dtrace_probe(), so the argument n will actually be found where + * one would expect to find argument (n + 1). + */ + arg++; + + stack = (uintptr_t *)&fp[1]; + +load: + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + val = stack[arg]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + return (val); +} + +int +dtrace_getstackdepth(int aframes) +{ + int depth = 0; + struct i386_frame *frame; + vm_offset_t ebp; + + aframes++; + ebp = dtrace_getfp(); + frame = (struct i386_frame *)ebp; + depth++; + for(;;) { + if (!INKERNEL((long) frame)) + break; + if (!INKERNEL((long) frame->f_frame)) + break; + depth++; + if (frame->f_frame <= frame || + (vm_offset_t)frame->f_frame >= + (vm_offset_t)ebp + KSTACK_PAGES * PAGE_SIZE) + break; + frame = frame->f_frame; + } + if (depth < aframes) + return 0; + else + return depth - aframes; +} + +#ifdef notyet +ulong_t +dtrace_getreg(struct regs *rp, uint_t reg) +{ +#if defined(__amd64) + int regmap[] = { + REG_GS, /* GS */ + REG_FS, /* FS */ + REG_ES, /* ES */ + REG_DS, /* DS */ + REG_RDI, /* EDI */ + REG_RSI, /* ESI */ + REG_RBP, /* EBP */ + REG_RSP, /* ESP */ + REG_RBX, /* EBX */ + REG_RDX, /* EDX */ + REG_RCX, /* ECX */ + REG_RAX, /* EAX */ + REG_TRAPNO, /* TRAPNO */ + REG_ERR, /* ERR */ + REG_RIP, /* EIP */ + REG_CS, /* CS */ + REG_RFL, /* EFL */ + REG_RSP, /* UESP */ + REG_SS /* SS */ + }; + + if (reg <= SS) { + if (reg >= sizeof (regmap) / sizeof (int)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + + reg = regmap[reg]; + } else { + reg -= SS + 1; + } + + switch (reg) { + case REG_RDI: + return (rp->r_rdi); + case REG_RSI: + return (rp->r_rsi); + case REG_RDX: + return (rp->r_rdx); + case REG_RCX: + return (rp->r_rcx); + case REG_R8: + return (rp->r_r8); + case REG_R9: + return (rp->r_r9); + case REG_RAX: + return (rp->r_rax); + case REG_RBX: + return (rp->r_rbx); + case REG_RBP: + return (rp->r_rbp); + case REG_R10: + return (rp->r_r10); + case REG_R11: + return (rp->r_r11); + case REG_R12: + return (rp->r_r12); + case REG_R13: + return (rp->r_r13); + case REG_R14: + return (rp->r_r14); + case REG_R15: + return (rp->r_r15); + case REG_DS: + return (rp->r_ds); + case REG_ES: + return (rp->r_es); + case REG_FS: + return (rp->r_fs); + case REG_GS: + return (rp->r_gs); + case REG_TRAPNO: + return (rp->r_trapno); + case REG_ERR: + return (rp->r_err); + case REG_RIP: + return (rp->r_rip); + case REG_CS: + return (rp->r_cs); + case REG_SS: + return (rp->r_ss); + case REG_RFL: + return (rp->r_rfl); + case REG_RSP: + return (rp->r_rsp); + default: + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + +#else + if (reg > SS) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + + return ((&rp->r_gs)[reg]); +#endif +} +#endif + +static int +dtrace_copycheck(uintptr_t uaddr, uintptr_t kaddr, size_t size) +{ + ASSERT(kaddr >= kernelbase && kaddr + size >= kaddr); + + if (uaddr + size >= kernelbase || uaddr + size < uaddr) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = uaddr; + return (0); + } + + return (1); +} + +void +dtrace_copyin(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copy(uaddr, kaddr, size); +} + +void +dtrace_copyout(uintptr_t kaddr, uintptr_t uaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copy(kaddr, uaddr, size); +} + +void +dtrace_copyinstr(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copystr(uaddr, kaddr, size, flags); +} + +void +dtrace_copyoutstr(uintptr_t kaddr, uintptr_t uaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copystr(kaddr, uaddr, size, flags); +} + +uint8_t +dtrace_fuword8(void *uaddr) +{ + if ((uintptr_t)uaddr >= kernelbase) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword8_nocheck(uaddr)); +} + +uint16_t +dtrace_fuword16(void *uaddr) +{ + if ((uintptr_t)uaddr >= kernelbase) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword16_nocheck(uaddr)); +} + +uint32_t +dtrace_fuword32(void *uaddr) +{ + if ((uintptr_t)uaddr >= kernelbase) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword32_nocheck(uaddr)); +} + +uint64_t +dtrace_fuword64(void *uaddr) +{ + if ((uintptr_t)uaddr >= kernelbase) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[curcpu].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword64_nocheck(uaddr)); +} diff --git a/sys/cddl/dev/dtrace/i386/dtrace_subr.c b/sys/cddl/dev/dtrace/i386/dtrace_subr.c new file mode 100644 index 000000000000..50515b643152 --- /dev/null +++ b/sys/cddl/dev/dtrace/i386/dtrace_subr.c @@ -0,0 +1,503 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + * + */ +/* + * Copyright 2005 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/types.h> +#include <sys/kernel.h> +#include <sys/malloc.h> +#include <sys/kmem.h> +#include <sys/smp.h> +#include <sys/dtrace_impl.h> +#include <sys/dtrace_bsd.h> +#include <machine/clock.h> +#include <machine/frame.h> +#include <vm/pmap.h> + +extern uintptr_t kernelbase; +extern uintptr_t dtrace_in_probe_addr; +extern int dtrace_in_probe; + +int dtrace_invop(uintptr_t, uintptr_t *, uintptr_t); + +typedef struct dtrace_invop_hdlr { + int (*dtih_func)(uintptr_t, uintptr_t *, uintptr_t); + struct dtrace_invop_hdlr *dtih_next; +} dtrace_invop_hdlr_t; + +dtrace_invop_hdlr_t *dtrace_invop_hdlr; + +int +dtrace_invop(uintptr_t addr, uintptr_t *stack, uintptr_t eax) +{ + dtrace_invop_hdlr_t *hdlr; + int rval; + + for (hdlr = dtrace_invop_hdlr; hdlr != NULL; hdlr = hdlr->dtih_next) + if ((rval = hdlr->dtih_func(addr, stack, eax)) != 0) + return (rval); + + return (0); +} + +void +dtrace_invop_add(int (*func)(uintptr_t, uintptr_t *, uintptr_t)) +{ + dtrace_invop_hdlr_t *hdlr; + + hdlr = kmem_alloc(sizeof (dtrace_invop_hdlr_t), KM_SLEEP); + hdlr->dtih_func = func; + hdlr->dtih_next = dtrace_invop_hdlr; + dtrace_invop_hdlr = hdlr; +} + +void +dtrace_invop_remove(int (*func)(uintptr_t, uintptr_t *, uintptr_t)) +{ + dtrace_invop_hdlr_t *hdlr = dtrace_invop_hdlr, *prev = NULL; + + for (;;) { + if (hdlr == NULL) + panic("attempt to remove non-existent invop handler"); + + if (hdlr->dtih_func == func) + break; + + prev = hdlr; + hdlr = hdlr->dtih_next; + } + + if (prev == NULL) { + ASSERT(dtrace_invop_hdlr == hdlr); + dtrace_invop_hdlr = hdlr->dtih_next; + } else { + ASSERT(dtrace_invop_hdlr != hdlr); + prev->dtih_next = hdlr->dtih_next; + } + + kmem_free(hdlr, 0); +} + +void +dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit)) +{ + (*func)(0, kernelbase); +} + +void +dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg) +{ + cpumask_t cpus; + + critical_enter(); + + if (cpu == DTRACE_CPUALL) + cpus = all_cpus; + else + cpus = (cpumask_t) (1 << cpu); + + /* If the current CPU is in the set, call the function directly: */ + if ((cpus & (1 << curcpu)) != 0) { + (*func)(arg); + + /* Mask the current CPU from the set */ + cpus &= ~(1 << curcpu); + } + + /* If there are any CPUs in the set, cross-call to those CPUs */ + if (cpus != 0) + smp_rendezvous_cpus(cpus, NULL, func, smp_no_rendevous_barrier, arg); + + critical_exit(); +} + +static void +dtrace_sync_func(void) +{ +} + +void +dtrace_sync(void) +{ + dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL); +} + +#ifdef notyet +int (*dtrace_fasttrap_probe_ptr)(struct regs *); +int (*dtrace_pid_probe_ptr)(struct regs *); +int (*dtrace_return_probe_ptr)(struct regs *); + +void +dtrace_user_probe(struct regs *rp, caddr_t addr, processorid_t cpuid) +{ + krwlock_t *rwp; + proc_t *p = curproc; + extern void trap(struct regs *, caddr_t, processorid_t); + + if (USERMODE(rp->r_cs) || (rp->r_ps & PS_VM)) { + if (curthread->t_cred != p->p_cred) { + cred_t *oldcred = curthread->t_cred; + /* + * DTrace accesses t_cred in probe context. t_cred + * must always be either NULL, or point to a valid, + * allocated cred structure. + */ + curthread->t_cred = crgetcred(); + crfree(oldcred); + } + } + + if (rp->r_trapno == T_DTRACE_RET) { + uint8_t step = curthread->t_dtrace_step; + uint8_t ret = curthread->t_dtrace_ret; + uintptr_t npc = curthread->t_dtrace_npc; + + if (curthread->t_dtrace_ast) { + aston(curthread); + curthread->t_sig_check = 1; + } + + /* + * Clear all user tracing flags. + */ + curthread->t_dtrace_ft = 0; + + /* + * If we weren't expecting to take a return probe trap, kill + * the process as though it had just executed an unassigned + * trap instruction. + */ + if (step == 0) { + tsignal(curthread, SIGILL); + return; + } + + /* + * If we hit this trap unrelated to a return probe, we're + * just here to reset the AST flag since we deferred a signal + * until after we logically single-stepped the instruction we + * copied out. + */ + if (ret == 0) { + rp->r_pc = npc; + return; + } + + /* + * We need to wait until after we've called the + * dtrace_return_probe_ptr function pointer to set %pc. + */ + rwp = &CPU->cpu_ft_lock; + rw_enter(rwp, RW_READER); + if (dtrace_return_probe_ptr != NULL) + (void) (*dtrace_return_probe_ptr)(rp); + rw_exit(rwp); + rp->r_pc = npc; + + } else if (rp->r_trapno == T_DTRACE_PROBE) { + rwp = &CPU->cpu_ft_lock; + rw_enter(rwp, RW_READER); + if (dtrace_fasttrap_probe_ptr != NULL) + (void) (*dtrace_fasttrap_probe_ptr)(rp); + rw_exit(rwp); + + } else if (rp->r_trapno == T_BPTFLT) { + uint8_t instr; + rwp = &CPU->cpu_ft_lock; + + /* + * The DTrace fasttrap provider uses the breakpoint trap + * (int 3). We let DTrace take the first crack at handling + * this trap; if it's not a probe that DTrace knowns about, + * we call into the trap() routine to handle it like a + * breakpoint placed by a conventional debugger. + */ + rw_enter(rwp, RW_READER); + if (dtrace_pid_probe_ptr != NULL && + (*dtrace_pid_probe_ptr)(rp) == 0) { + rw_exit(rwp); + return; + } + rw_exit(rwp); + + /* + * If the instruction that caused the breakpoint trap doesn't + * look like an int 3 anymore, it may be that this tracepoint + * was removed just after the user thread executed it. In + * that case, return to user land to retry the instuction. + */ + if (fuword8((void *)(rp->r_pc - 1), &instr) == 0 && + instr != FASTTRAP_INSTR) { + rp->r_pc--; + return; + } + + trap(rp, addr, cpuid); + + } else { + trap(rp, addr, cpuid); + } +} + +void +dtrace_safe_synchronous_signal(void) +{ + kthread_t *t = curthread; + struct regs *rp = lwptoregs(ttolwp(t)); + size_t isz = t->t_dtrace_npc - t->t_dtrace_pc; + + ASSERT(t->t_dtrace_on); + + /* + * If we're not in the range of scratch addresses, we're not actually + * tracing user instructions so turn off the flags. If the instruction + * we copied out caused a synchonous trap, reset the pc back to its + * original value and turn off the flags. + */ + if (rp->r_pc < t->t_dtrace_scrpc || + rp->r_pc > t->t_dtrace_astpc + isz) { + t->t_dtrace_ft = 0; + } else if (rp->r_pc == t->t_dtrace_scrpc || + rp->r_pc == t->t_dtrace_astpc) { + rp->r_pc = t->t_dtrace_pc; + t->t_dtrace_ft = 0; + } +} + +int +dtrace_safe_defer_signal(void) +{ + kthread_t *t = curthread; + struct regs *rp = lwptoregs(ttolwp(t)); + size_t isz = t->t_dtrace_npc - t->t_dtrace_pc; + + ASSERT(t->t_dtrace_on); + + /* + * If we're not in the range of scratch addresses, we're not actually + * tracing user instructions so turn off the flags. + */ + if (rp->r_pc < t->t_dtrace_scrpc || + rp->r_pc > t->t_dtrace_astpc + isz) { + t->t_dtrace_ft = 0; + return (0); + } + + /* + * If we've executed the original instruction, but haven't performed + * the jmp back to t->t_dtrace_npc or the clean up of any registers + * used to emulate %rip-relative instructions in 64-bit mode, do that + * here and take the signal right away. We detect this condition by + * seeing if the program counter is the range [scrpc + isz, astpc). + */ + if (t->t_dtrace_astpc - rp->r_pc < + t->t_dtrace_astpc - t->t_dtrace_scrpc - isz) { +#ifdef __amd64 + /* + * If there is a scratch register and we're on the + * instruction immediately after the modified instruction, + * restore the value of that scratch register. + */ + if (t->t_dtrace_reg != 0 && + rp->r_pc == t->t_dtrace_scrpc + isz) { + switch (t->t_dtrace_reg) { + case REG_RAX: + rp->r_rax = t->t_dtrace_regv; + break; + case REG_RCX: + rp->r_rcx = t->t_dtrace_regv; + break; + case REG_R8: + rp->r_r8 = t->t_dtrace_regv; + break; + case REG_R9: + rp->r_r9 = t->t_dtrace_regv; + break; + } + } +#endif + rp->r_pc = t->t_dtrace_npc; + t->t_dtrace_ft = 0; + return (0); + } + + /* + * Otherwise, make sure we'll return to the kernel after executing + * the copied out instruction and defer the signal. + */ + if (!t->t_dtrace_step) { + ASSERT(rp->r_pc < t->t_dtrace_astpc); + rp->r_pc += t->t_dtrace_astpc - t->t_dtrace_scrpc; + t->t_dtrace_step = 1; + } + + t->t_dtrace_ast = 1; + + return (1); +} +#endif + +static int64_t tgt_cpu_tsc; +static int64_t hst_cpu_tsc; +static int64_t tsc_skew[MAXCPU]; + +static void +dtrace_gethrtime_init_sync(void *arg) +{ +#ifdef CHECK_SYNC + /* + * Delay this function from returning on one + * of the CPUs to check that the synchronisation + * works. + */ + uintptr_t cpu = (uintptr_t) arg; + + if (cpu == curcpu) { + int i; + for (i = 0; i < 1000000000; i++) + tgt_cpu_tsc = rdtsc(); + tgt_cpu_tsc = 0; + } +#endif +} + +static void +dtrace_gethrtime_init_cpu(void *arg) +{ + uintptr_t cpu = (uintptr_t) arg; + + if (cpu == curcpu) + tgt_cpu_tsc = rdtsc(); + else + hst_cpu_tsc = rdtsc(); +} + +static void +dtrace_gethrtime_init(void *arg) +{ + cpumask_t map; + int i; + struct pcpu *cp; + + /* The current CPU is the reference one. */ + tsc_skew[curcpu] = 0; + + for (i = 0; i <= mp_maxid; i++) { + if (i == curcpu) + continue; + + if ((cp = pcpu_find(i)) == NULL) + continue; + + map = 0; + map |= (1 << curcpu); + map |= (1 << i); + + smp_rendezvous_cpus(map, dtrace_gethrtime_init_sync, + dtrace_gethrtime_init_cpu, + smp_no_rendevous_barrier, (void *)(uintptr_t) i); + + tsc_skew[i] = tgt_cpu_tsc - hst_cpu_tsc; + } +} + +SYSINIT(dtrace_gethrtime_init, SI_SUB_SMP, SI_ORDER_ANY, dtrace_gethrtime_init, NULL); + +/* + * DTrace needs a high resolution time function which can + * be called from a probe context and guaranteed not to have + * instrumented with probes itself. + * + * Returns nanoseconds since boot. + */ +uint64_t +dtrace_gethrtime() +{ + return ((rdtsc() + tsc_skew[curcpu]) * (int64_t) 1000000000 / tsc_freq); +} + +uint64_t +dtrace_gethrestime(void) +{ + printf("%s(%d): XXX\n",__func__,__LINE__); + return (0); +} + +/* Function to handle DTrace traps during probes. See i386/i386/trap.c */ +int +dtrace_trap(struct trapframe *frame, u_int type) +{ + /* + * A trap can occur while DTrace executes a probe. Before + * executing the probe, DTrace blocks re-scheduling and sets + * a flag in it's per-cpu flags to indicate that it doesn't + * want to fault. On returning from the the probe, the no-fault + * flag is cleared and finally re-scheduling is enabled. + * + * Check if DTrace has enabled 'no-fault' mode: + * + */ + if ((cpu_core[curcpu].cpuc_dtrace_flags & CPU_DTRACE_NOFAULT) != 0) { + /* + * There are only a couple of trap types that are expected. + * All the rest will be handled in the usual way. + */ + switch (type) { + /* General protection fault. */ + case T_PROTFLT: + /* Flag an illegal operation. */ + cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + + /* + * Offset the instruction pointer to the instruction + * following the one causing the fault. + */ + frame->tf_eip += dtrace_instr_size((u_char *) frame->tf_eip); + return (1); + /* Page fault. */ + case T_PAGEFLT: + /* Flag a bad address. */ + cpu_core[curcpu].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR; + cpu_core[curcpu].cpuc_dtrace_illval = rcr2(); + + /* + * Offset the instruction pointer to the instruction + * following the one causing the fault. + */ + frame->tf_eip += dtrace_instr_size((u_char *) frame->tf_eip); + return (1); + default: + /* Handle all other traps in the usual way. */ + break; + } + } + + /* Handle the trap in the usual way. */ + return (0); +} diff --git a/sys/cddl/dev/dtrace/i386/instr_size.c b/sys/cddl/dev/dtrace/i386/instr_size.c new file mode 100644 index 000000000000..fb6af2d0abd0 --- /dev/null +++ b/sys/cddl/dev/dtrace/i386/instr_size.c @@ -0,0 +1,132 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * $FreeBSD$ + */ +/* + * Copyright 2005 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +/* Copyright (c) 1988 AT&T */ +/* All Rights Reserved */ + + +#if defined(sun) +#pragma ident "@(#)instr_size.c 1.14 05/07/08 SMI" +#endif + +#include <sys/types.h> +#include <sys/param.h> +#include <sys/proc.h> +#if defined(sun) +#include <sys/cmn_err.h> +#include <sys/archsystm.h> +#include <sys/copyops.h> +#include <vm/seg_enum.h> +#include <sys/privregs.h> +#else +typedef u_int model_t; +#define DATAMODEL_NATIVE 0 +int dtrace_instr_size(uchar_t *); +#endif + +#include <dis_tables.h> + +/* + * This subsystem (with the minor exception of the instr_size() function) is + * is called from DTrace probe context. This imposes several requirements on + * the implementation: + * + * 1. External subsystems and functions may not be referenced. The one current + * exception is for cmn_err, but only to signal the detection of table + * errors. Assuming the tables are correct, no combination of input is to + * trigger a cmn_err call. + * + * 2. These functions can't be allowed to be traced. To prevent this, + * all functions in the probe path (everything except instr_size()) must + * have names that begin with "dtrace_". + */ + +typedef enum dis_isize { + DIS_ISIZE_INSTR, + DIS_ISIZE_OPERAND +} dis_isize_t; + + +/* + * get a byte from instruction stream + */ +static int +dtrace_dis_get_byte(void *p) +{ + int ret; + uchar_t **instr = p; + + ret = **instr; + *instr += 1; + + return (ret); +} + +/* + * Returns either the size of a given instruction, in bytes, or the size of that + * instruction's memory access (if any), depending on the value of `which'. + * If a programming error in the tables is detected, the system will panic to + * ease diagnosis. Invalid instructions will not be flagged. They will appear + * to have an instruction size between 1 and the actual size, and will be + * reported as having no memory impact. + */ +/* ARGSUSED2 */ +static int +dtrace_dis_isize(uchar_t *instr, dis_isize_t which, model_t model, int *rmindex) +{ + int sz; + dis86_t x; + uint_t mode = SIZE32; + +#if defined(sun) + mode = (model == DATAMODEL_LP64) ? SIZE64 : SIZE32; +#endif + + x.d86_data = (void **)&instr; + x.d86_get_byte = dtrace_dis_get_byte; + x.d86_check_func = NULL; + + if (dtrace_disx86(&x, mode) != 0) + return (-1); + + if (which == DIS_ISIZE_INSTR) + sz = x.d86_len; /* length of the instruction */ + else + sz = x.d86_memsize; /* length of memory operand */ + + if (rmindex != NULL) + *rmindex = x.d86_rmindex; + return (sz); +} + +int +dtrace_instr_size(uchar_t *instr) +{ + return (dtrace_dis_isize(instr, DIS_ISIZE_INSTR, DATAMODEL_NATIVE, + NULL)); +} diff --git a/sys/cddl/dev/fasttrap/fasttrap.c b/sys/cddl/dev/fasttrap/fasttrap.c new file mode 100644 index 000000000000..df8309ea282c --- /dev/null +++ b/sys/cddl/dev/fasttrap/fasttrap.c @@ -0,0 +1,2333 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * Portions Copyright 2008 John Birrell <jb@freebsd.org> + * + * $FreeBSD$ + */ + +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#include <sys/cdefs.h> +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/kdb.h> +#include <sys/kernel.h> +#include <sys/limits.h> +#include <sys/linker.h> +#include <sys/lock.h> +#include <sys/module.h> +#include <sys/proc.h> +#include <sys/unistd.h> +#include <sys/dtrace.h> +#include <sys/dtrace_bsd.h> +#include <sys/fasttrap_impl.h> + +/* + * User-Land Trap-Based Tracing + * ---------------------------- + * + * The fasttrap provider allows DTrace consumers to instrument any user-level + * instruction to gather data; this includes probes with semantic + * signifigance like entry and return as well as simple offsets into the + * function. While the specific techniques used are very ISA specific, the + * methodology is generalizable to any architecture. + * + * + * The General Methodology + * ----------------------- + * + * With the primary goal of tracing every user-land instruction and the + * limitation that we can't trust user space so don't want to rely on much + * information there, we begin by replacing the instructions we want to trace + * with trap instructions. Each instruction we overwrite is saved into a hash + * table keyed by process ID and pc address. When we enter the kernel due to + * this trap instruction, we need the effects of the replaced instruction to + * appear to have occurred before we proceed with the user thread's + * execution. + * + * Each user level thread is represented by a ulwp_t structure which is + * always easily accessible through a register. The most basic way to produce + * the effects of the instruction we replaced is to copy that instruction out + * to a bit of scratch space reserved in the user thread's ulwp_t structure + * (a sort of kernel-private thread local storage), set the PC to that + * scratch space and single step. When we reenter the kernel after single + * stepping the instruction we must then adjust the PC to point to what would + * normally be the next instruction. Of course, special care must be taken + * for branches and jumps, but these represent such a small fraction of any + * instruction set that writing the code to emulate these in the kernel is + * not too difficult. + * + * Return probes may require several tracepoints to trace every return site, + * and, conversely, each tracepoint may activate several probes (the entry + * and offset 0 probes, for example). To solve this muliplexing problem, + * tracepoints contain lists of probes to activate and probes contain lists + * of tracepoints to enable. If a probe is activated, it adds its ID to + * existing tracepoints or creates new ones as necessary. + * + * Most probes are activated _before_ the instruction is executed, but return + * probes are activated _after_ the effects of the last instruction of the + * function are visible. Return probes must be fired _after_ we have + * single-stepped the instruction whereas all other probes are fired + * beforehand. + * + * + * Lock Ordering + * ------------- + * + * The lock ordering below -- both internally and with respect to the DTrace + * framework -- is a little tricky and bears some explanation. Each provider + * has a lock (ftp_mtx) that protects its members including reference counts + * for enabled probes (ftp_rcount), consumers actively creating probes + * (ftp_ccount) and USDT consumers (ftp_mcount); all three prevent a provider + * from being freed. A provider is looked up by taking the bucket lock for the + * provider hash table, and is returned with its lock held. The provider lock + * may be taken in functions invoked by the DTrace framework, but may not be + * held while calling functions in the DTrace framework. + * + * To ensure consistency over multiple calls to the DTrace framework, the + * creation lock (ftp_cmtx) should be held. Naturally, the creation lock may + * not be taken when holding the provider lock as that would create a cyclic + * lock ordering. In situations where one would naturally take the provider + * lock and then the creation lock, we instead up a reference count to prevent + * the provider from disappearing, drop the provider lock, and acquire the + * creation lock. + * + * Briefly: + * bucket lock before provider lock + * DTrace before provider lock + * creation lock before DTrace + * never hold the provider lock and creation lock simultaneously + */ + +static d_open_t fasttrap_open; +static d_ioctl_t fasttrap_ioctl; +static int fasttrap_unload(void); +static void fasttrap_load(void *); + +static struct cdevsw fasttrap_cdevsw = { + .d_version = D_VERSION, + .d_open = fasttrap_open, + .d_ioctl = fasttrap_ioctl, + .d_name = "fasttrap", +}; + +static struct cdev *fasttrap_cdev; +static dtrace_provider_id_t fasttrap_id __used; +static dtrace_meta_provider_id_t fasttrap_meta_id; + +/* + * When the fasttrap provider is loaded, fasttrap_max is set to either + * FASTTRAP_MAX_DEFAULT or the value for fasttrap-max-probes in the + * fasttrap.conf file. Each time a probe is created, fasttrap_total is + * incremented by the number of tracepoints that may be associated with that + * probe; fasttrap_total is capped at fasttrap_max. + */ +#define FASTTRAP_MAX_DEFAULT 250000 +static uint32_t fasttrap_max; +static uint32_t fasttrap_total; + +fasttrap_hash_t fasttrap_tpoints; +static fasttrap_hash_t fasttrap_provs; +static fasttrap_hash_t fasttrap_procs; + +#define FASTTRAP_TPOINTS_DEFAULT_SIZE 0x4000 +#define FASTTRAP_PROVIDERS_DEFAULT_SIZE 0x100 +#define FASTTRAP_PROCS_DEFAULT_SIZE 0x100 + +#define FASTTRAP_PID_NAME "pid" + +static struct callout_handle fasttrap_timeout = CALLOUT_HANDLE_INITIALIZER(&fasttrap_timeout); +static struct mtx fasttrap_cleanup_mtx; +MTX_SYSINIT(fasttrap_cleanup_mtx, &fasttrap_cleanup_mtx, "Fasttrap cleanup", MTX_DEF); +static uint_t fasttrap_cleanup_work; + +/* + * Generation count on modifications to the global tracepoint lookup table. + */ +static volatile uint64_t fasttrap_mod_gen; + +static uint64_t fasttrap_pid_count; /* pid ref count */ +static struct mtx fasttrap_count_mtx; /* lock on ref count */ +MTX_SYSINIT(fasttrap_count_mtx, &fasttrap_count_mtx, "Fasttrap ref count", MTX_DEF); + +#define FASTTRAP_ENABLE_FAIL 1 +#define FASTTRAP_ENABLE_PARTIAL 2 + +static int fasttrap_tracepoint_enable(proc_t *, fasttrap_probe_t *, uint_t); +static void fasttrap_tracepoint_disable(proc_t *, fasttrap_probe_t *, uint_t); + +static fasttrap_provider_t *fasttrap_provider_lookup(pid_t, const char *, + const dtrace_pattr_t *); +static void fasttrap_provider_free(fasttrap_provider_t *); +static void fasttrap_provider_retire(pid_t, const char *, int); + +static fasttrap_proc_t *fasttrap_proc_lookup(pid_t); +static void fasttrap_proc_release(fasttrap_proc_t *); + +#define FASTTRAP_PROVS_INDEX(pid, name) \ + ((fasttrap_hash_str(name) + (pid)) & fasttrap_provs.fth_mask) + +#define FASTTRAP_PROCS_INDEX(pid) ((pid) & fasttrap_procs.fth_mask) + +static int +fasttrap_highbit(ulong_t i) +{ + int h = 1; + + if (i == 0) + return (0); +#ifdef _LP64 + if (i & 0xffffffff00000000ul) { + h += 32; i >>= 32; + } +#endif + if (i & 0xffff0000) { + h += 16; i >>= 16; + } + if (i & 0xff00) { + h += 8; i >>= 8; + } + if (i & 0xf0) { + h += 4; i >>= 4; + } + if (i & 0xc) { + h += 2; i >>= 2; + } + if (i & 0x2) { + h += 1; + } + return (h); +} + +static uint_t +fasttrap_hash_str(const char *p) +{ + unsigned int g; + uint_t hval = 0; + + while (*p) { + hval = (hval << 4) + *p++; + if ((g = (hval & 0xf0000000)) != 0) + hval ^= g >> 24; + hval &= ~g; + } + return (hval); +} + +void +fasttrap_sigtrap(proc_t *p, kthread_t *t, uintptr_t pc) +{ +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + sigqueue_t *sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP); + + sqp->sq_info.si_signo = SIGTRAP; + sqp->sq_info.si_code = TRAP_DTRACE; + sqp->sq_info.si_addr = (caddr_t)pc; + + PROC_LOCK(p); + sigaddqa(p, t, sqp); + PROC_UNLOCK(p); + + if (t != NULL) + aston(t); +#endif +} + +/* + * This function ensures that no threads are actively using the memory + * associated with probes that were formerly live. + */ +static void +fasttrap_mod_barrier(uint64_t gen) +{ +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + int i; + + if (gen < fasttrap_mod_gen) + return; + + fasttrap_mod_gen++; + + for (i = 0; i < MAXCPU; i++) { + mtx_lock(&cpu_core[i].cpuc_pid_lock); + mtx_unlock(&cpu_core[i].cpuc_pid_lock); + } +#endif +} + +/* + * This is the timeout's callback for cleaning up the providers and their + * probes. + */ +static void +fasttrap_pid_cleanup_cb(void *data) +{ + fasttrap_provider_t **fpp, *fp; + fasttrap_bucket_t *bucket; + dtrace_provider_id_t provid; + int i, later = 0; + + static volatile int in = 0; + ASSERT(in == 0); + in = 1; + + mtx_lock(&fasttrap_cleanup_mtx); + while (fasttrap_cleanup_work) { + fasttrap_cleanup_work = 0; + mtx_unlock(&fasttrap_cleanup_mtx); + + later = 0; + + /* + * Iterate over all the providers trying to remove the marked + * ones. If a provider is marked but not retired, we just + * have to take a crack at removing it -- it's no big deal if + * we can't. + */ + for (i = 0; i < fasttrap_provs.fth_nent; i++) { + bucket = &fasttrap_provs.fth_table[i]; + mtx_lock(&bucket->ftb_mtx); + fpp = (fasttrap_provider_t **)&bucket->ftb_data; + + while ((fp = *fpp) != NULL) { + if (!fp->ftp_marked) { + fpp = &fp->ftp_next; + continue; + } + + mtx_lock(&fp->ftp_mtx); + + /* + * If this provider has consumers actively + * creating probes (ftp_ccount) or is a USDT + * provider (ftp_mcount), we can't unregister + * or even condense. + */ + if (fp->ftp_ccount != 0 || + fp->ftp_mcount != 0) { + mtx_unlock(&fp->ftp_mtx); + fp->ftp_marked = 0; + continue; + } + + if (!fp->ftp_retired || fp->ftp_rcount != 0) + fp->ftp_marked = 0; + + mtx_unlock(&fp->ftp_mtx); + + /* + * If we successfully unregister this + * provider we can remove it from the hash + * chain and free the memory. If our attempt + * to unregister fails and this is a retired + * provider, increment our flag to try again + * pretty soon. If we've consumed more than + * half of our total permitted number of + * probes call dtrace_condense() to try to + * clean out the unenabled probes. + */ + provid = fp->ftp_provid; + if (dtrace_unregister(provid) != 0) { + if (fasttrap_total > fasttrap_max / 2) + (void) dtrace_condense(provid); + later += fp->ftp_marked; + fpp = &fp->ftp_next; + } else { + *fpp = fp->ftp_next; + fasttrap_provider_free(fp); + } + } + mtx_unlock(&bucket->ftb_mtx); + } + + mtx_lock(&fasttrap_cleanup_mtx); + } + + /* + * If we were unable to remove a retired provider, try again after + * a second. This situation can occur in certain circumstances where + * providers cannot be unregistered even though they have no probes + * enabled because of an execution of dtrace -l or something similar. + * If the timeout has been disabled (set to 1 because we're trying + * to detach), we set fasttrap_cleanup_work to ensure that we'll + * get a chance to do that work if and when the timeout is reenabled + * (if detach fails). + */ + if (later > 0 && fasttrap_timeout.callout != (struct callout *)(uintptr_t)1) + fasttrap_timeout = timeout(&fasttrap_pid_cleanup_cb, NULL, hz); + else if (later > 0) + fasttrap_cleanup_work = 1; + else + fasttrap_timeout.callout = NULL; + + mtx_unlock(&fasttrap_cleanup_mtx); + in = 0; +} + +/* + * Activates the asynchronous cleanup mechanism. + */ +static void +fasttrap_pid_cleanup(void) +{ + mtx_lock(&fasttrap_cleanup_mtx); + fasttrap_cleanup_work = 1; + if (fasttrap_timeout.callout == NULL) + fasttrap_timeout = timeout(&fasttrap_pid_cleanup_cb, NULL, 1); + mtx_unlock(&fasttrap_cleanup_mtx); +} + +/* This function assumes that the process is already locked. */ +static void +fasttrap_fork(proc_t *p, proc_t *cp) +{ + pid_t ppid = p->p_pid; + int i; + +#ifdef DOODAD + ASSERT(curproc == p); +#else + if (curproc != p) printf("%s(%d): Warning curproc != p\n",__func__,__LINE__); +#endif + + /* + * This would be simpler and faster if we maintained per-process + * hash tables of enabled tracepoints. It could, however, potentially + * slow down execution of a tracepoint since we'd need to go + * through two levels of indirection. In the future, we should + * consider either maintaining per-process ancillary lists of + * enabled tracepoints or hanging a pointer to a per-process hash + * table of enabled tracepoints off the proc structure. + */ + + /* + * Iterate over every tracepoint looking for ones that belong to the + * parent process, and remove each from the child process. + */ + for (i = 0; i < fasttrap_tpoints.fth_nent; i++) { + fasttrap_tracepoint_t *tp; + fasttrap_bucket_t *bucket = &fasttrap_tpoints.fth_table[i]; + + mtx_lock(&bucket->ftb_mtx); + for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { + if (tp->ftt_pid == ppid && + tp->ftt_proc->ftpc_acount != 0) { +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + int ret = fasttrap_tracepoint_remove(cp, tp); + ASSERT(ret == 0); +#endif + } + } + mtx_unlock(&bucket->ftb_mtx); + } +} + +/* This function assumes that the process is already locked. */ +static void +fasttrap_exec_exit(proc_t *p) +{ + ASSERT(p == curproc); + + /* + * We clean up the pid provider for this process here; user-land + * static probes are handled by the meta-provider remove entry point. + */ + fasttrap_provider_retire(p->p_pid, FASTTRAP_PID_NAME, 0); +} + + +static void +fasttrap_pid_provide(void *arg, dtrace_probedesc_t *desc) +{ + /* + * There are no "default" pid probes. + */ +printf("%s(%d): There are no default pid probes\n",__func__,__LINE__); +} + +static int +fasttrap_tracepoint_enable(proc_t *p, fasttrap_probe_t *probe, uint_t index) +{ + fasttrap_tracepoint_t *tp, *new_tp = NULL; + fasttrap_bucket_t *bucket; + fasttrap_id_t *id; + pid_t pid; + uintptr_t pc; + + ASSERT(index < probe->ftp_ntps); + + pid = probe->ftp_pid; + pc = probe->ftp_tps[index].fit_tp->ftt_pc; + id = &probe->ftp_tps[index].fit_id; + + ASSERT(probe->ftp_tps[index].fit_tp->ftt_pid == pid); + + /* + * Before we make any modifications, make sure we've imposed a barrier + * on the generation in which this probe was last modified. + */ + fasttrap_mod_barrier(probe->ftp_gen); + + bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; + + /* + * If the tracepoint has already been enabled, just add our id to the + * list of interested probes. This may be our second time through + * this path in which case we'll have constructed the tracepoint we'd + * like to install. If we can't find a match, and have an allocated + * tracepoint ready to go, enable that one now. + * + * A tracepoint whose process is defunct is also considered defunct. + */ +#ifdef DOODAD +again: +#endif + mtx_lock(&bucket->ftb_mtx); + for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { + if (tp->ftt_pid != pid || tp->ftt_pc != pc || + tp->ftt_proc->ftpc_acount == 0) + continue; + + /* + * Now that we've found a matching tracepoint, it would be + * a decent idea to confirm that the tracepoint is still + * enabled and the trap instruction hasn't been overwritten. + * Since this is a little hairy, we'll punt for now. + */ + + /* + * This can't be the first interested probe. We don't have + * to worry about another thread being in the midst of + * deleting this tracepoint (which would be the only valid + * reason for a tracepoint to have no interested probes) + * since we're holding P_PR_LOCK for this process. + */ + ASSERT(tp->ftt_ids != NULL || tp->ftt_retids != NULL); + + switch (id->fti_ptype) { + case DTFTP_ENTRY: + case DTFTP_OFFSETS: + case DTFTP_IS_ENABLED: + id->fti_next = tp->ftt_ids; +#ifdef DOODAD + membar_producer(); +#endif + tp->ftt_ids = id; +#ifdef DOODAD + membar_producer(); +#endif + break; + + case DTFTP_RETURN: + case DTFTP_POST_OFFSETS: + id->fti_next = tp->ftt_retids; +#ifdef DOODAD + membar_producer(); +#endif + tp->ftt_retids = id; +#ifdef DOODAD + membar_producer(); +#endif + break; + + default: + ASSERT(0); + } + + mtx_unlock(&bucket->ftb_mtx); + + if (new_tp != NULL) { + new_tp->ftt_ids = NULL; + new_tp->ftt_retids = NULL; + } + + return (0); + } + + /* + * If we have a good tracepoint ready to go, install it now while + * we have the lock held and no one can screw with us. + */ + if (new_tp != NULL) { + int rc = 0; + + new_tp->ftt_next = bucket->ftb_data; +#ifdef DOODAD + membar_producer(); +#endif + bucket->ftb_data = new_tp; +#ifdef DOODAD + membar_producer(); +#endif + mtx_unlock(&bucket->ftb_mtx); + + /* + * Activate the tracepoint in the ISA-specific manner. + * If this fails, we need to report the failure, but + * indicate that this tracepoint must still be disabled + * by calling fasttrap_tracepoint_disable(). + */ + if (fasttrap_tracepoint_install(p, new_tp) != 0) + rc = FASTTRAP_ENABLE_PARTIAL; + + /* + * Increment the count of the number of tracepoints active in + * the victim process. + */ +#ifdef DOODAD + ASSERT(p->p_proc_flag & P_PR_LOCK); + p->p_dtrace_count++; +#endif + + return (rc); + } + + mtx_unlock(&bucket->ftb_mtx); + + /* + * Initialize the tracepoint that's been preallocated with the probe. + */ + new_tp = probe->ftp_tps[index].fit_tp; + + ASSERT(new_tp->ftt_pid == pid); + ASSERT(new_tp->ftt_pc == pc); + ASSERT(new_tp->ftt_proc == probe->ftp_prov->ftp_proc); + ASSERT(new_tp->ftt_ids == NULL); + ASSERT(new_tp->ftt_retids == NULL); + + switch (id->fti_ptype) { + case DTFTP_ENTRY: + case DTFTP_OFFSETS: + case DTFTP_IS_ENABLED: + id->fti_next = NULL; + new_tp->ftt_ids = id; + break; + + case DTFTP_RETURN: + case DTFTP_POST_OFFSETS: + id->fti_next = NULL; + new_tp->ftt_retids = id; + break; + + default: + ASSERT(0); + } + +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + /* + * If the ISA-dependent initialization goes to plan, go back to the + * beginning and try to install this freshly made tracepoint. + */ + if (fasttrap_tracepoint_init(p, new_tp, pc, id->fti_ptype) == 0) + goto again; +#endif + + new_tp->ftt_ids = NULL; + new_tp->ftt_retids = NULL; + + return (FASTTRAP_ENABLE_FAIL); +} + +static void +fasttrap_tracepoint_disable(proc_t *p, fasttrap_probe_t *probe, uint_t index) +{ + fasttrap_bucket_t *bucket; + fasttrap_provider_t *provider = probe->ftp_prov; + fasttrap_tracepoint_t **pp, *tp; + fasttrap_id_t *id, **idp; + pid_t pid; + uintptr_t pc; + + ASSERT(index < probe->ftp_ntps); + + pid = probe->ftp_pid; + pc = probe->ftp_tps[index].fit_tp->ftt_pc; + id = &probe->ftp_tps[index].fit_id; + + ASSERT(probe->ftp_tps[index].fit_tp->ftt_pid == pid); + + /* + * Find the tracepoint and make sure that our id is one of the + * ones registered with it. + */ + bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; + mtx_lock(&bucket->ftb_mtx); + for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { + if (tp->ftt_pid == pid && tp->ftt_pc == pc && + tp->ftt_proc == provider->ftp_proc) + break; + } + + /* + * If we somehow lost this tracepoint, we're in a world of hurt. + */ + ASSERT(tp != NULL); + + switch (id->fti_ptype) { + case DTFTP_ENTRY: + case DTFTP_OFFSETS: + case DTFTP_IS_ENABLED: + ASSERT(tp->ftt_ids != NULL); + idp = &tp->ftt_ids; + break; + + case DTFTP_RETURN: + case DTFTP_POST_OFFSETS: + ASSERT(tp->ftt_retids != NULL); + idp = &tp->ftt_retids; + break; + + default: + ASSERT(0); + } + + while ((*idp)->fti_probe != probe) { + idp = &(*idp)->fti_next; + ASSERT(*idp != NULL); + } + + id = *idp; + *idp = id->fti_next; +#ifdef DOODAD + membar_producer(); +#endif + + ASSERT(id->fti_probe == probe); + + /* + * If there are other registered enablings of this tracepoint, we're + * all done, but if this was the last probe assocated with this + * this tracepoint, we need to remove and free it. + */ + if (tp->ftt_ids != NULL || tp->ftt_retids != NULL) { + + /* + * If the current probe's tracepoint is in use, swap it + * for an unused tracepoint. + */ + if (tp == probe->ftp_tps[index].fit_tp) { + fasttrap_probe_t *tmp_probe; + fasttrap_tracepoint_t **tmp_tp; + uint_t tmp_index; + + if (tp->ftt_ids != NULL) { + tmp_probe = tp->ftt_ids->fti_probe; + /* LINTED - alignment */ + tmp_index = FASTTRAP_ID_INDEX(tp->ftt_ids); + tmp_tp = &tmp_probe->ftp_tps[tmp_index].fit_tp; + } else { + tmp_probe = tp->ftt_retids->fti_probe; + /* LINTED - alignment */ + tmp_index = FASTTRAP_ID_INDEX(tp->ftt_retids); + tmp_tp = &tmp_probe->ftp_tps[tmp_index].fit_tp; + } + + ASSERT(*tmp_tp != NULL); + ASSERT(*tmp_tp != probe->ftp_tps[index].fit_tp); + ASSERT((*tmp_tp)->ftt_ids == NULL); + ASSERT((*tmp_tp)->ftt_retids == NULL); + + probe->ftp_tps[index].fit_tp = *tmp_tp; + *tmp_tp = tp; + } + + mtx_unlock(&bucket->ftb_mtx); + + /* + * Tag the modified probe with the generation in which it was + * changed. + */ + probe->ftp_gen = fasttrap_mod_gen; + return; + } + + mtx_unlock(&bucket->ftb_mtx); + + /* + * We can't safely remove the tracepoint from the set of active + * tracepoints until we've actually removed the fasttrap instruction + * from the process's text. We can, however, operate on this + * tracepoint secure in the knowledge that no other thread is going to + * be looking at it since we hold P_PR_LOCK on the process if it's + * live or we hold the provider lock on the process if it's dead and + * gone. + */ + + /* + * We only need to remove the actual instruction if we're looking + * at an existing process + */ + if (p != NULL) { + /* + * If we fail to restore the instruction we need to kill + * this process since it's in a completely unrecoverable + * state. + */ +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + if (fasttrap_tracepoint_remove(p, tp) != 0) + fasttrap_sigtrap(p, NULL, pc); +#endif + + /* + * Decrement the count of the number of tracepoints active + * in the victim process. + */ +#ifdef DOODAD + ASSERT(p->p_proc_flag & P_PR_LOCK); + p->p_dtrace_count--; +#endif + } + + /* + * Remove the probe from the hash table of active tracepoints. + */ + mtx_lock(&bucket->ftb_mtx); + pp = (fasttrap_tracepoint_t **)&bucket->ftb_data; + ASSERT(*pp != NULL); + while (*pp != tp) { + pp = &(*pp)->ftt_next; + ASSERT(*pp != NULL); + } + + *pp = tp->ftt_next; +#ifdef DOODAD + membar_producer(); +#endif + + mtx_unlock(&bucket->ftb_mtx); + + /* + * Tag the modified probe with the generation in which it was changed. + */ + probe->ftp_gen = fasttrap_mod_gen; +} + +static void __unused +fasttrap_enable_callbacks(void) +{ + /* + * We don't have to play the rw lock game here because we're + * providing something rather than taking something away -- + * we can be sure that no threads have tried to follow this + * function pointer yet. + */ + mtx_lock(&fasttrap_count_mtx); +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + if (fasttrap_pid_count == 0) { + ASSERT(dtrace_pid_probe_ptr == NULL); + ASSERT(dtrace_return_probe_ptr == NULL); + dtrace_pid_probe_ptr = &fasttrap_pid_probe; + dtrace_return_probe_ptr = &fasttrap_return_probe; + } + ASSERT(dtrace_pid_probe_ptr == &fasttrap_pid_probe); + ASSERT(dtrace_return_probe_ptr == &fasttrap_return_probe); +#endif + fasttrap_pid_count++; + mtx_unlock(&fasttrap_count_mtx); +} + +static void __unused +fasttrap_disable_callbacks(void) +{ + ASSERT(MUTEX_HELD(&cpu_lock)); + + mtx_lock(&fasttrap_count_mtx); +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + ASSERT(fasttrap_pid_count > 0); + fasttrap_pid_count--; + if (fasttrap_pid_count == 0) { + cpu_t *cur, *cpu = CPU; + + for (cur = cpu->cpu_next_onln; cur != cpu; + cur = cur->cpu_next_onln) { + rw_enter(&cur->cpu_ft_lock, RW_WRITER); + } + + dtrace_pid_probe_ptr = NULL; + dtrace_return_probe_ptr = NULL; + + for (cur = cpu->cpu_next_onln; cur != cpu; + cur = cur->cpu_next_onln) { + rw_exit(&cur->cpu_ft_lock); + } + } +#endif + mtx_unlock(&fasttrap_count_mtx); +} + +static void +fasttrap_pid_enable(void *arg, dtrace_id_t id, void *parg) +{ + fasttrap_probe_t *probe = parg; + proc_t *p; + int i, rc; + + ASSERT(probe != NULL); + ASSERT(!probe->ftp_enabled); + ASSERT(id == probe->ftp_id); + ASSERT(MUTEX_HELD(&cpu_lock)); + + /* + * Increment the count of enabled probes on this probe's provider; + * the provider can't go away while the probe still exists. We + * must increment this even if we aren't able to properly enable + * this probe. + */ + mtx_lock(&probe->ftp_prov->ftp_mtx); + probe->ftp_prov->ftp_rcount++; + mtx_unlock(&probe->ftp_prov->ftp_mtx); + + /* + * If this probe's provider is retired (meaning it was valid in a + * previously exec'ed incarnation of this address space), bail out. The + * provider can't go away while we're in this code path. + */ + if (probe->ftp_prov->ftp_retired) + return; + +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + /* + * If we can't find the process, it may be that we're in the context of + * a fork in which the traced process is being born and we're copying + * USDT probes. Otherwise, the process is gone so bail. + */ + if ((p = sprlock(probe->ftp_pid)) == NULL) { + if ((curproc->p_flag & SFORKING) == 0) + return; + + mtx_lock(&pidlock); + p = prfind(probe->ftp_pid); + + /* + * Confirm that curproc is indeed forking the process in which + * we're trying to enable probes. + */ + ASSERT(p != NULL); + ASSERT(p->p_parent == curproc); + ASSERT(p->p_stat == SIDL); + + mtx_lock(&p->p_lock); + mtx_unlock(&pidlock); + + sprlock_proc(p); + } + + ASSERT(!(p->p_flag & SVFORK)); + mtx_unlock(&p->p_lock); +#else + p = pfind(probe->ftp_pid); +#endif + + /* + * We have to enable the trap entry point before any user threads have + * the chance to execute the trap instruction we're about to place + * in their process's text. + */ + fasttrap_enable_callbacks(); + + /* + * Enable all the tracepoints and add this probe's id to each + * tracepoint's list of active probes. + */ + for (i = 0; i < probe->ftp_ntps; i++) { + if ((rc = fasttrap_tracepoint_enable(p, probe, i)) != 0) { + /* + * If enabling the tracepoint failed completely, + * we don't have to disable it; if the failure + * was only partial we must disable it. + */ + if (rc == FASTTRAP_ENABLE_FAIL) + i--; + else + ASSERT(rc == FASTTRAP_ENABLE_PARTIAL); + + /* + * Back up and pull out all the tracepoints we've + * created so far for this probe. + */ + while (i >= 0) { + fasttrap_tracepoint_disable(p, probe, i); + i--; + } + +#ifdef DOODAD + mtx_lock(&p->p_lock); + sprunlock(p); +#endif + + /* + * Since we're not actually enabling this probe, + * drop our reference on the trap table entry. + */ + fasttrap_disable_callbacks(); + return; + } + } + +#ifdef DOODAD + mtx_lock(&p->p_lock); + sprunlock(p); +#endif + + probe->ftp_enabled = 1; +} + +static void +fasttrap_pid_disable(void *arg, dtrace_id_t id, void *parg) +{ +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + fasttrap_probe_t *probe = parg; + fasttrap_provider_t *provider = probe->ftp_prov; + proc_t *p; + int i, whack = 0; + + ASSERT(id == probe->ftp_id); + + /* + * We won't be able to acquire a /proc-esque lock on the process + * iff the process is dead and gone. In this case, we rely on the + * provider lock as a point of mutual exclusion to prevent other + * DTrace consumers from disabling this probe. + */ + if ((p = sprlock(probe->ftp_pid)) != NULL) { + ASSERT(!(p->p_flag & SVFORK)); + mtx_unlock(&p->p_lock); + } + + mtx_lock(&provider->ftp_mtx); + + /* + * Disable all the associated tracepoints (for fully enabled probes). + */ + if (probe->ftp_enabled) { + for (i = 0; i < probe->ftp_ntps; i++) { + fasttrap_tracepoint_disable(p, probe, i); + } + } + + ASSERT(provider->ftp_rcount > 0); + provider->ftp_rcount--; + + if (p != NULL) { + /* + * Even though we may not be able to remove it entirely, we + * mark this retired provider to get a chance to remove some + * of the associated probes. + */ + if (provider->ftp_retired && !provider->ftp_marked) + whack = provider->ftp_marked = 1; + mtx_unlock(&provider->ftp_mtx); + + mtx_lock(&p->p_lock); + sprunlock(p); + } else { + /* + * If the process is dead, we're just waiting for the + * last probe to be disabled to be able to free it. + */ + if (provider->ftp_rcount == 0 && !provider->ftp_marked) + whack = provider->ftp_marked = 1; + mtx_unlock(&provider->ftp_mtx); + } + + if (whack) + fasttrap_pid_cleanup(); + + if (!probe->ftp_enabled) + return; + + probe->ftp_enabled = 0; + + ASSERT(MUTEX_HELD(&cpu_lock)); + fasttrap_disable_callbacks(); +#endif +} + +static void +fasttrap_pid_getargdesc(void *arg, dtrace_id_t id, void *parg, + dtrace_argdesc_t *desc) +{ +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + fasttrap_probe_t *probe = parg; + char *str; + int i, ndx; + + desc->dtargd_native[0] = '\0'; + desc->dtargd_xlate[0] = '\0'; + + if (probe->ftp_prov->ftp_retired != 0 || + desc->dtargd_ndx >= probe->ftp_nargs) { + desc->dtargd_ndx = DTRACE_ARGNONE; + return; + } + + ndx = (probe->ftp_argmap != NULL) ? + probe->ftp_argmap[desc->dtargd_ndx] : desc->dtargd_ndx; + + str = probe->ftp_ntypes; + for (i = 0; i < ndx; i++) { + str += strlen(str) + 1; + } + + ASSERT(strlen(str + 1) < sizeof (desc->dtargd_native)); + (void) strcpy(desc->dtargd_native, str); + + if (probe->ftp_xtypes == NULL) + return; + + str = probe->ftp_xtypes; + for (i = 0; i < desc->dtargd_ndx; i++) { + str += strlen(str) + 1; + } + + ASSERT(strlen(str + 1) < sizeof (desc->dtargd_xlate)); + (void) strcpy(desc->dtargd_xlate, str); +#endif +} + +static void +fasttrap_pid_destroy(void *arg, dtrace_id_t id, void *parg) +{ +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + fasttrap_probe_t *probe = parg; + int i; + size_t size; + + ASSERT(probe != NULL); + ASSERT(!probe->ftp_enabled); + ASSERT(fasttrap_total >= probe->ftp_ntps); + + atomic_add_32(&fasttrap_total, -probe->ftp_ntps); + size = offsetof(fasttrap_probe_t, ftp_tps[probe->ftp_ntps]); + + if (probe->ftp_gen + 1 >= fasttrap_mod_gen) + fasttrap_mod_barrier(probe->ftp_gen); + + for (i = 0; i < probe->ftp_ntps; i++) { + kmem_free(probe->ftp_tps[i].fit_tp, + sizeof (fasttrap_tracepoint_t)); + } + + kmem_free(probe, size); +#endif +} + +static const dtrace_pattr_t pid_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +}; + +static dtrace_pops_t pid_pops = { + fasttrap_pid_provide, + NULL, + fasttrap_pid_enable, + fasttrap_pid_disable, + NULL, + NULL, + fasttrap_pid_getargdesc, + NULL /*fasttrap_pid_getarg*/, + NULL, + fasttrap_pid_destroy +}; + +static dtrace_pops_t usdt_pops = { + fasttrap_pid_provide, + NULL, + fasttrap_pid_enable, + fasttrap_pid_disable, + NULL, + NULL, + fasttrap_pid_getargdesc, + NULL /*fasttrap_usdt_getarg*/, + NULL, + fasttrap_pid_destroy +}; + +static fasttrap_proc_t * +fasttrap_proc_lookup(pid_t pid) +{ + fasttrap_bucket_t *bucket; + fasttrap_proc_t *fprc, *new_fprc; + + bucket = &fasttrap_procs.fth_table[FASTTRAP_PROCS_INDEX(pid)]; + mtx_lock(&bucket->ftb_mtx); + + for (fprc = bucket->ftb_data; fprc != NULL; fprc = fprc->ftpc_next) { + if (fprc->ftpc_pid == pid && fprc->ftpc_acount != 0) { + mtx_lock(&fprc->ftpc_mtx); + mtx_unlock(&bucket->ftb_mtx); + fprc->ftpc_rcount++; + atomic_add_64(&fprc->ftpc_acount, 1); + mtx_unlock(&fprc->ftpc_mtx); + + return (fprc); + } + } + + /* + * Drop the bucket lock so we don't try to perform a sleeping + * allocation under it. + */ + mtx_unlock(&bucket->ftb_mtx); + + new_fprc = kmem_zalloc(sizeof (fasttrap_proc_t), KM_SLEEP); + new_fprc->ftpc_pid = pid; + new_fprc->ftpc_rcount = 1; + new_fprc->ftpc_acount = 1; + + mtx_lock(&bucket->ftb_mtx); + + /* + * Take another lap through the list to make sure a proc hasn't + * been created for this pid while we weren't under the bucket lock. + */ + for (fprc = bucket->ftb_data; fprc != NULL; fprc = fprc->ftpc_next) { + if (fprc->ftpc_pid == pid && fprc->ftpc_acount != 0) { + mtx_lock(&fprc->ftpc_mtx); + mtx_unlock(&bucket->ftb_mtx); + fprc->ftpc_rcount++; + atomic_add_64(&fprc->ftpc_acount, 1); + mtx_unlock(&fprc->ftpc_mtx); + + kmem_free(new_fprc, sizeof (fasttrap_proc_t)); + + return (fprc); + } + } + + new_fprc->ftpc_next = bucket->ftb_data; + bucket->ftb_data = new_fprc; + + mtx_unlock(&bucket->ftb_mtx); + + return (new_fprc); +} + +static void +fasttrap_proc_release(fasttrap_proc_t *proc) +{ + fasttrap_bucket_t *bucket; + fasttrap_proc_t *fprc, **fprcp; + pid_t pid = proc->ftpc_pid; + + mtx_lock(&proc->ftpc_mtx); + + ASSERT(proc->ftpc_rcount != 0); + + if (--proc->ftpc_rcount != 0) { + mtx_unlock(&proc->ftpc_mtx); + return; + } + + mtx_unlock(&proc->ftpc_mtx); + + /* + * There should definitely be no live providers associated with this + * process at this point. + */ + ASSERT(proc->ftpc_acount == 0); + + bucket = &fasttrap_procs.fth_table[FASTTRAP_PROCS_INDEX(pid)]; + mtx_lock(&bucket->ftb_mtx); + + fprcp = (fasttrap_proc_t **)&bucket->ftb_data; + while ((fprc = *fprcp) != NULL) { + if (fprc == proc) + break; + + fprcp = &fprc->ftpc_next; + } + + /* + * Something strange has happened if we can't find the proc. + */ + ASSERT(fprc != NULL); + + *fprcp = fprc->ftpc_next; + + mtx_unlock(&bucket->ftb_mtx); + + kmem_free(fprc, sizeof (fasttrap_proc_t)); +} + +/* + * Lookup a fasttrap-managed provider based on its name and associated pid. + * If the pattr argument is non-NULL, this function instantiates the provider + * if it doesn't exist otherwise it returns NULL. The provider is returned + * with its lock held. + */ +static fasttrap_provider_t * +fasttrap_provider_lookup(pid_t pid, const char *name, + const dtrace_pattr_t *pattr) +{ + fasttrap_provider_t *fp, *new_fp = NULL; + fasttrap_bucket_t *bucket; + char provname[DTRACE_PROVNAMELEN]; + proc_t *p; +#ifdef DOODAD + cred_t *cred; +#endif + + ASSERT(strlen(name) < sizeof (fp->ftp_name)); + ASSERT(pattr != NULL); + + bucket = &fasttrap_provs.fth_table[FASTTRAP_PROVS_INDEX(pid, name)]; + mtx_lock(&bucket->ftb_mtx); + + /* + * Take a lap through the list and return the match if we find it. + */ + for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { + if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && + !fp->ftp_retired) { + mtx_lock(&fp->ftp_mtx); + mtx_unlock(&bucket->ftb_mtx); + return (fp); + } + } + + /* + * Drop the bucket lock so we don't try to perform a sleeping + * allocation under it. + */ + mtx_unlock(&bucket->ftb_mtx); + +#ifdef DOODAD + /* + * Make sure the process exists, isn't a child created as the result + * of a vfork(2), and isn't a zombie (but may be in fork). + */ + mtx_lock(&pidlock); + if ((p = prfind(pid)) == NULL) { + mtx_unlock(&pidlock); + return (NULL); + } + mtx_lock(&p->p_lock); + mtx_unlock(&pidlock); + if (p->p_flag & (SVFORK | SEXITING)) { + mtx_unlock(&p->p_lock); + return (NULL); + } + + /* + * Increment p_dtrace_probes so that the process knows to inform us + * when it exits or execs. fasttrap_provider_free() decrements this + * when we're done with this provider. + */ + p->p_dtrace_probes++; + + /* + * Grab the credentials for this process so we have + * something to pass to dtrace_register(). + */ + mtx_lock(&p->p_crlock); + crhold(p->p_cred); + cred = p->p_cred; + mtx_unlock(&p->p_crlock); + mtx_unlock(&p->p_lock); +#else + p = pfind(pid); +#endif + + new_fp = kmem_zalloc(sizeof (fasttrap_provider_t), KM_SLEEP); + new_fp->ftp_pid = pid; + new_fp->ftp_proc = fasttrap_proc_lookup(pid); + + ASSERT(new_fp->ftp_proc != NULL); + + mtx_lock(&bucket->ftb_mtx); + + /* + * Take another lap through the list to make sure a provider hasn't + * been created for this pid while we weren't under the bucket lock. + */ + for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { + if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && + !fp->ftp_retired) { + mtx_lock(&fp->ftp_mtx); + mtx_unlock(&bucket->ftb_mtx); + fasttrap_provider_free(new_fp); +#ifdef DOODAD + crfree(cred); +#endif + return (fp); + } + } + + (void) strcpy(new_fp->ftp_name, name); + + /* + * Fail and return NULL if either the provider name is too long + * or we fail to register this new provider with the DTrace + * framework. Note that this is the only place we ever construct + * the full provider name -- we keep it in pieces in the provider + * structure. + */ +printf("%s(%d): provname '%s%u'\n", __func__, __LINE__, name, (uint_t)pid); + if (snprintf(provname, sizeof (provname), "%s%u", name, (uint_t)pid) >= + sizeof (provname) || + dtrace_register(provname, pattr, + DTRACE_PRIV_PROC | DTRACE_PRIV_OWNER | DTRACE_PRIV_ZONEOWNER, NULL /*cred*/, + pattr == &pid_attr ? &pid_pops : &usdt_pops, new_fp, + &new_fp->ftp_provid) != 0) { +printf("%s(%d): could not register!\n",__func__,__LINE__); + mtx_unlock(&bucket->ftb_mtx); + fasttrap_provider_free(new_fp); +#ifdef DOODAD + crfree(cred); +#endif + return (NULL); + } + + new_fp->ftp_next = bucket->ftb_data; + bucket->ftb_data = new_fp; + + mtx_lock(&new_fp->ftp_mtx); + mtx_unlock(&bucket->ftb_mtx); + +#ifdef DOODAD + crfree(cred); +#endif + return (new_fp); +} + +static void +fasttrap_provider_free(fasttrap_provider_t *provider) +{ +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + pid_t pid = provider->ftp_pid; + proc_t *p; +#endif + + /* + * There need to be no associated enabled probes, no consumers + * creating probes, and no meta providers referencing this provider. + */ + ASSERT(provider->ftp_rcount == 0); + ASSERT(provider->ftp_ccount == 0); + ASSERT(provider->ftp_mcount == 0); + + fasttrap_proc_release(provider->ftp_proc); + + kmem_free(provider, sizeof (fasttrap_provider_t)); + +#ifdef DOODAD + /* + * Decrement p_dtrace_probes on the process whose provider we're + * freeing. We don't have to worry about clobbering somone else's + * modifications to it because we have locked the bucket that + * corresponds to this process's hash chain in the provider hash + * table. Don't sweat it if we can't find the process. + */ + mtx_lock(&pidlock); + if ((p = prfind(pid)) == NULL) { + mtx_unlock(&pidlock); + return; + } + + mtx_lock(&p->p_lock); + mtx_unlock(&pidlock); + + p->p_dtrace_probes--; + mtx_unlock(&p->p_lock); +#endif +} + +static void +fasttrap_provider_retire(pid_t pid, const char *name, int mprov) +{ + fasttrap_provider_t *fp; + fasttrap_bucket_t *bucket; + dtrace_provider_id_t provid; + + ASSERT(strlen(name) < sizeof (fp->ftp_name)); + + bucket = &fasttrap_provs.fth_table[FASTTRAP_PROVS_INDEX(pid, name)]; + mtx_lock(&bucket->ftb_mtx); + + for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { + if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && + !fp->ftp_retired) + break; + } + + if (fp == NULL) { + mtx_unlock(&bucket->ftb_mtx); + return; + } + + mtx_lock(&fp->ftp_mtx); + ASSERT(!mprov || fp->ftp_mcount > 0); + if (mprov && --fp->ftp_mcount != 0) { + mtx_unlock(&fp->ftp_mtx); + mtx_unlock(&bucket->ftb_mtx); + return; + } + + /* + * Mark the provider to be removed in our post-processing step, mark it + * retired, and drop the active count on its proc. Marking it indicates + * that we should try to remove it; setting the retired flag indicates + * that we're done with this provider; dropping the active the proc + * releases our hold, and when this reaches zero (as it will during + * exit or exec) the proc and associated providers become defunct. + * + * We obviously need to take the bucket lock before the provider lock + * to perform the lookup, but we need to drop the provider lock + * before calling into the DTrace framework since we acquire the + * provider lock in callbacks invoked from the DTrace framework. The + * bucket lock therefore protects the integrity of the provider hash + * table. + */ + atomic_add_64(&fp->ftp_proc->ftpc_acount, -1); + fp->ftp_retired = 1; + fp->ftp_marked = 1; + provid = fp->ftp_provid; + mtx_unlock(&fp->ftp_mtx); + + /* + * We don't have to worry about invalidating the same provider twice + * since fasttrap_provider_lookup() will ignore provider that have + * been marked as retired. + */ + dtrace_invalidate(provid); + + mtx_unlock(&bucket->ftb_mtx); + + fasttrap_pid_cleanup(); +} + +static int __unused +fasttrap_uint32_cmp(const void *ap, const void *bp) +{ + return (*(const uint32_t *)ap - *(const uint32_t *)bp); +} + +static int +fasttrap_uint64_cmp(const void *ap, const void *bp) +{ + return (*(const uint64_t *)ap - *(const uint64_t *)bp); +} + +static int __unused +fasttrap_add_probe (fasttrap_probe_spec_t *pdata) +{ + fasttrap_provider_t *provider; + fasttrap_probe_t *pp; + fasttrap_tracepoint_t *tp; + char *name; + int i, aframes, whack; + + /* + * There needs to be at least one desired trace point. + */ + if (pdata->ftps_noffs == 0) + return (EINVAL); + + switch (pdata->ftps_type) { + case DTFTP_ENTRY: + name = "entry"; + aframes = FASTTRAP_ENTRY_AFRAMES; + break; + case DTFTP_RETURN: + name = "return"; + aframes = FASTTRAP_RETURN_AFRAMES; + break; + case DTFTP_OFFSETS: + name = NULL; + break; + default: + return (EINVAL); + } + + if ((provider = fasttrap_provider_lookup(pdata->ftps_pid, + FASTTRAP_PID_NAME, &pid_attr)) == NULL) + return (ESRCH); + + /* + * Increment this reference count to indicate that a consumer is + * actively adding a new probe associated with this provider. This + * prevents the provider from being deleted -- we'll need to check + * for pending deletions when we drop this reference count. + */ + provider->ftp_ccount++; + mtx_unlock(&provider->ftp_mtx); + + /* + * Grab the creation lock to ensure consistency between calls to + * dtrace_probe_lookup() and dtrace_probe_create() in the face of + * other threads creating probes. We must drop the provider lock + * before taking this lock to avoid a three-way deadlock with the + * DTrace framework. + */ + mtx_lock(&provider->ftp_cmtx); + + if (name == NULL) { + for (i = 0; i < pdata->ftps_noffs; i++) { + char name_str[17]; + + (void) sprintf(name_str, "%llx", + (unsigned long long)pdata->ftps_offs[i]); + + if (dtrace_probe_lookup(provider->ftp_provid, + pdata->ftps_mod, pdata->ftps_func, name_str) != 0) + continue; + + atomic_add_32(&fasttrap_total, 1); + + if (fasttrap_total > fasttrap_max) { + atomic_add_32(&fasttrap_total, -1); + goto no_mem; + } + + pp = kmem_zalloc(sizeof (fasttrap_probe_t), KM_SLEEP); + + pp->ftp_prov = provider; + pp->ftp_faddr = pdata->ftps_pc; + pp->ftp_fsize = pdata->ftps_size; + pp->ftp_pid = pdata->ftps_pid; + pp->ftp_ntps = 1; + + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), + KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = pdata->ftps_offs[i] + pdata->ftps_pc; + tp->ftt_pid = pdata->ftps_pid; + + pp->ftp_tps[0].fit_tp = tp; + pp->ftp_tps[0].fit_id.fti_probe = pp; + pp->ftp_tps[0].fit_id.fti_ptype = pdata->ftps_type; + + pp->ftp_id = dtrace_probe_create(provider->ftp_provid, + pdata->ftps_mod, pdata->ftps_func, name_str, + FASTTRAP_OFFSET_AFRAMES, pp); + } + + } else if (dtrace_probe_lookup(provider->ftp_provid, pdata->ftps_mod, + pdata->ftps_func, name) == 0) { + atomic_add_32(&fasttrap_total, pdata->ftps_noffs); + + if (fasttrap_total > fasttrap_max) { + atomic_add_32(&fasttrap_total, -pdata->ftps_noffs); + goto no_mem; + } + + /* + * Make sure all tracepoint program counter values are unique. + * We later assume that each probe has exactly one tracepoint + * for a given pc. + */ + qsort(pdata->ftps_offs, pdata->ftps_noffs, + sizeof (uint64_t), fasttrap_uint64_cmp); + for (i = 1; i < pdata->ftps_noffs; i++) { + if (pdata->ftps_offs[i] > pdata->ftps_offs[i - 1]) + continue; + + atomic_add_32(&fasttrap_total, -pdata->ftps_noffs); + goto no_mem; + } + + ASSERT(pdata->ftps_noffs > 0); + pp = kmem_zalloc(offsetof(fasttrap_probe_t, + ftp_tps[pdata->ftps_noffs]), KM_SLEEP); + + pp->ftp_prov = provider; + pp->ftp_faddr = pdata->ftps_pc; + pp->ftp_fsize = pdata->ftps_size; + pp->ftp_pid = pdata->ftps_pid; + pp->ftp_ntps = pdata->ftps_noffs; + + for (i = 0; i < pdata->ftps_noffs; i++) { + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), + KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = pdata->ftps_offs[i] + pdata->ftps_pc; + tp->ftt_pid = pdata->ftps_pid; + + pp->ftp_tps[i].fit_tp = tp; + pp->ftp_tps[i].fit_id.fti_probe = pp; + pp->ftp_tps[i].fit_id.fti_ptype = pdata->ftps_type; + } + + pp->ftp_id = dtrace_probe_create(provider->ftp_provid, + pdata->ftps_mod, pdata->ftps_func, name, aframes, pp); + } + + mtx_unlock(&provider->ftp_cmtx); + + /* + * We know that the provider is still valid since we incremented the + * creation reference count. If someone tried to clean up this provider + * while we were using it (e.g. because the process called exec(2) or + * exit(2)), take note of that and try to clean it up now. + */ + mtx_lock(&provider->ftp_mtx); + provider->ftp_ccount--; + whack = provider->ftp_retired; + mtx_unlock(&provider->ftp_mtx); + + if (whack) + fasttrap_pid_cleanup(); + + return (0); + +no_mem: + /* + * If we've exhausted the allowable resources, we'll try to remove + * this provider to free some up. This is to cover the case where + * the user has accidentally created many more probes than was + * intended (e.g. pid123:::). + */ + mtx_unlock(&provider->ftp_cmtx); + mtx_lock(&provider->ftp_mtx); + provider->ftp_ccount--; + provider->ftp_marked = 1; + mtx_unlock(&provider->ftp_mtx); + + fasttrap_pid_cleanup(); + + return (ENOMEM); +} + +static void * +fasttrap_meta_provide(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid) +{ + fasttrap_provider_t *provider; + + /* + * A 32-bit unsigned integer (like a pid for example) can be + * expressed in 10 or fewer decimal digits. Make sure that we'll + * have enough space for the provider name. + */ + if (strlen(dhpv->dthpv_provname) + 10 >= + sizeof (provider->ftp_name)) { + printf("failed to instantiate provider %s: name too long to accomodate pid\n", dhpv->dthpv_provname); + return (NULL); + } + + /* + * Don't let folks spoof the true pid provider. + */ + if (strcmp(dhpv->dthpv_provname, FASTTRAP_PID_NAME) == 0) { + printf("failed to instantiate provider %s: %s is an invalid name\n", dhpv->dthpv_provname, + FASTTRAP_PID_NAME); + return (NULL); + } + + /* + * The highest stability class that fasttrap supports is ISA; cap + * the stability of the new provider accordingly. + */ + if (dhpv->dthpv_pattr.dtpa_provider.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_provider.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_mod.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_mod.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_func.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_func.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_name.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_name.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_args.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_args.dtat_class = DTRACE_CLASS_ISA; + + if ((provider = fasttrap_provider_lookup(pid, dhpv->dthpv_provname, + &dhpv->dthpv_pattr)) == NULL) { + printf("failed to instantiate provider %s for process %u\n", dhpv->dthpv_provname, (uint_t)pid); + return (NULL); + } + + /* + * Up the meta provider count so this provider isn't removed until + * the meta provider has been told to remove it. + */ + provider->ftp_mcount++; + + mtx_unlock(&provider->ftp_mtx); + + return (provider); +} + +static void +fasttrap_meta_create_probe(void *arg, void *parg, + dtrace_helper_probedesc_t *dhpb) +{ +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + fasttrap_provider_t *provider = parg; + fasttrap_probe_t *pp; + fasttrap_tracepoint_t *tp; + int i, j; + uint32_t ntps; + + /* + * Since the meta provider count is non-zero we don't have to worry + * about this provider disappearing. + */ + ASSERT(provider->ftp_mcount > 0); + + /* + * The offsets must be unique. + */ + qsort(dhpb->dthpb_offs, dhpb->dthpb_noffs, sizeof (uint32_t), + fasttrap_uint32_cmp); + for (i = 1; i < dhpb->dthpb_noffs; i++) { + if (dhpb->dthpb_base + dhpb->dthpb_offs[i] <= + dhpb->dthpb_base + dhpb->dthpb_offs[i - 1]) + return; + } + + qsort(dhpb->dthpb_enoffs, dhpb->dthpb_nenoffs, sizeof (uint32_t), + fasttrap_uint32_cmp); + for (i = 1; i < dhpb->dthpb_nenoffs; i++) { + if (dhpb->dthpb_base + dhpb->dthpb_enoffs[i] <= + dhpb->dthpb_base + dhpb->dthpb_enoffs[i - 1]) + return; + } + + /* + * Grab the creation lock to ensure consistency between calls to + * dtrace_probe_lookup() and dtrace_probe_create() in the face of + * other threads creating probes. + */ + mtx_lock(&provider->ftp_cmtx); + + if (dtrace_probe_lookup(provider->ftp_provid, dhpb->dthpb_mod, + dhpb->dthpb_func, dhpb->dthpb_name) != 0) { + mtx_unlock(&provider->ftp_cmtx); + return; + } + + ntps = dhpb->dthpb_noffs + dhpb->dthpb_nenoffs; + ASSERT(ntps > 0); + + atomic_add_32(&fasttrap_total, ntps); + + if (fasttrap_total > fasttrap_max) { + atomic_add_32(&fasttrap_total, -ntps); + mtx_unlock(&provider->ftp_cmtx); + return; + } + + pp = kmem_zalloc(offsetof(fasttrap_probe_t, ftp_tps[ntps]), KM_SLEEP); + + pp->ftp_prov = provider; + pp->ftp_pid = provider->ftp_pid; + pp->ftp_ntps = ntps; + pp->ftp_nargs = dhpb->dthpb_xargc; + pp->ftp_xtypes = dhpb->dthpb_xtypes; + pp->ftp_ntypes = dhpb->dthpb_ntypes; + + /* + * First create a tracepoint for each actual point of interest. + */ + for (i = 0; i < dhpb->dthpb_noffs; i++) { + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = dhpb->dthpb_base + dhpb->dthpb_offs[i]; + tp->ftt_pid = provider->ftp_pid; + + pp->ftp_tps[i].fit_tp = tp; + pp->ftp_tps[i].fit_id.fti_probe = pp; +#ifdef __sparc + pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_POST_OFFSETS; +#else + pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_OFFSETS; +#endif + } + + /* + * Then create a tracepoint for each is-enabled point. + */ + for (j = 0; i < ntps; i++, j++) { + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = dhpb->dthpb_base + dhpb->dthpb_enoffs[j]; + tp->ftt_pid = provider->ftp_pid; + + pp->ftp_tps[i].fit_tp = tp; + pp->ftp_tps[i].fit_id.fti_probe = pp; + pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_IS_ENABLED; + } + + /* + * If the arguments are shuffled around we set the argument remapping + * table. Later, when the probe fires, we only remap the arguments + * if the table is non-NULL. + */ + for (i = 0; i < dhpb->dthpb_xargc; i++) { + if (dhpb->dthpb_args[i] != i) { + pp->ftp_argmap = dhpb->dthpb_args; + break; + } + } + + /* + * The probe is fully constructed -- register it with DTrace. + */ + pp->ftp_id = dtrace_probe_create(provider->ftp_provid, dhpb->dthpb_mod, + dhpb->dthpb_func, dhpb->dthpb_name, FASTTRAP_OFFSET_AFRAMES, pp); + + mtx_unlock(&provider->ftp_cmtx); +#endif +} + +static void +fasttrap_meta_remove(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid) +{ +printf("%s(%d): \n",__func__,__LINE__); + /* + * Clean up the USDT provider. There may be active consumers of the + * provider busy adding probes, no damage will actually befall the + * provider until that count has dropped to zero. This just puts + * the provider on death row. + */ + fasttrap_provider_retire(pid, dhpv->dthpv_provname, 1); +} + +static dtrace_mops_t fasttrap_mops = { + fasttrap_meta_create_probe, + fasttrap_meta_provide, + fasttrap_meta_remove +}; + +static void +fasttrap_load(void *dummy) +{ + int i; + int nent; + + /* Create the /dev/dtrace/fasttrap entry. */ + fasttrap_cdev = make_dev(&fasttrap_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, + "dtrace/fasttrap"); + + /* + * Install our hooks into fork(2), exec(2), and exit(2). + */ + dtrace_fasttrap_fork = fasttrap_fork; + dtrace_fasttrap_exit = fasttrap_exec_exit; + dtrace_fasttrap_exec = fasttrap_exec_exit; + + fasttrap_max = FASTTRAP_MAX_DEFAULT; + fasttrap_total = 0; + + /* + * Conjure up the tracepoints hashtable... + */ + nent = FASTTRAP_TPOINTS_DEFAULT_SIZE; /* XXX sysctl? */ + + if (nent == 0 || nent > 0x1000000) + nent = FASTTRAP_TPOINTS_DEFAULT_SIZE; + + if ((nent & (nent - 1)) == 0) + fasttrap_tpoints.fth_nent = nent; + else + fasttrap_tpoints.fth_nent = 1 << fasttrap_highbit(nent); + ASSERT(fasttrap_tpoints.fth_nent > 0); + fasttrap_tpoints.fth_mask = fasttrap_tpoints.fth_nent - 1; + fasttrap_tpoints.fth_table = kmem_zalloc(fasttrap_tpoints.fth_nent * + sizeof (fasttrap_bucket_t), KM_SLEEP); + + for (i = 0; i < fasttrap_tpoints.fth_nent; i++) + mtx_init(&fasttrap_tpoints.fth_table[i].ftb_mtx, + "Fasttrap tpoints", NULL, MTX_DEF); + + /* + * ... and the providers hash table... + */ + nent = FASTTRAP_PROVIDERS_DEFAULT_SIZE; + if ((nent & (nent - 1)) == 0) + fasttrap_provs.fth_nent = nent; + else + fasttrap_provs.fth_nent = 1 << fasttrap_highbit(nent); + ASSERT(fasttrap_provs.fth_nent > 0); + fasttrap_provs.fth_mask = fasttrap_provs.fth_nent - 1; + fasttrap_provs.fth_table = kmem_zalloc(fasttrap_provs.fth_nent * + sizeof (fasttrap_bucket_t), KM_SLEEP); + + for (i = 0; i < fasttrap_provs.fth_nent; i++) + mtx_init(&fasttrap_provs.fth_table[i].ftb_mtx, + "Fasttrap provs", NULL, MTX_DEF); + + /* + * ... and the procs hash table. + */ + nent = FASTTRAP_PROCS_DEFAULT_SIZE; + if ((nent & (nent - 1)) == 0) + fasttrap_procs.fth_nent = nent; + else + fasttrap_procs.fth_nent = 1 << fasttrap_highbit(nent); + ASSERT(fasttrap_procs.fth_nent > 0); + fasttrap_procs.fth_mask = fasttrap_procs.fth_nent - 1; + fasttrap_procs.fth_table = kmem_zalloc(fasttrap_procs.fth_nent * + sizeof (fasttrap_bucket_t), KM_SLEEP); + + for (i = 0; i < fasttrap_procs.fth_nent; i++) + mtx_init(&fasttrap_procs.fth_table[i].ftb_mtx, + "Fasttrap procs", NULL, MTX_DEF); + + + (void) dtrace_meta_register("fasttrap", &fasttrap_mops, NULL, + &fasttrap_meta_id); +} + +static int +fasttrap_unload() +{ + int error = 0; + int i, fail = 0; + struct callout_handle tmp; + + /* Reset our calback hooks. */ + dtrace_fasttrap_fork = NULL; + dtrace_fasttrap_exec = NULL; + dtrace_fasttrap_exit = NULL; + + /* + * Unregister the meta-provider to make sure no new fasttrap- + * managed providers come along while we're trying to close up + * shop. If we fail to detach, we'll need to re-register as a + * meta-provider. We can fail to unregister as a meta-provider + * if providers we manage still exist. + */ + if (fasttrap_meta_id != DTRACE_METAPROVNONE && + dtrace_meta_unregister(fasttrap_meta_id) != 0) + return (ENOENT); + + /* + * Prevent any new timeouts from running by setting fasttrap_timeout + * to a non-zero value, and wait for the current timeout to complete. + */ + mtx_lock(&fasttrap_cleanup_mtx); + fasttrap_cleanup_work = 0; + + while (fasttrap_timeout.callout != (struct callout *)(uintptr_t)1) { + tmp = fasttrap_timeout; + fasttrap_timeout.callout = (struct callout *)(uintptr_t)1; + + if (tmp.callout != NULL) { + mtx_unlock(&fasttrap_cleanup_mtx); + (void) untimeout(&fasttrap_pid_cleanup_cb, NULL, tmp); + mtx_lock(&fasttrap_cleanup_mtx); + } + } + + fasttrap_cleanup_work = 0; + mtx_unlock(&fasttrap_cleanup_mtx); + + /* + * Iterate over all of our providers. If there's still a process + * that corresponds to that pid, fail to detach. + */ + for (i = 0; i < fasttrap_provs.fth_nent; i++) { + fasttrap_provider_t **fpp, *fp; + fasttrap_bucket_t *bucket = &fasttrap_provs.fth_table[i]; + + mtx_lock(&bucket->ftb_mtx); + fpp = (fasttrap_provider_t **)&bucket->ftb_data; + while ((fp = *fpp) != NULL) { + /* + * Acquire and release the lock as a simple way of + * waiting for any other consumer to finish with + * this provider. A thread must first acquire the + * bucket lock so there's no chance of another thread + * blocking on the provider's lock. + */ + mtx_lock(&fp->ftp_mtx); + mtx_unlock(&fp->ftp_mtx); + + if (dtrace_unregister(fp->ftp_provid) != 0) { + fail = 1; + fpp = &fp->ftp_next; + } else { + *fpp = fp->ftp_next; + fasttrap_provider_free(fp); + } + } + + mtx_unlock(&bucket->ftb_mtx); + } + + if (fail) { + uint_t work; + /* + * If we're failing to detach, we need to unblock timeouts + * and start a new timeout if any work has accumulated while + * we've been unsuccessfully trying to detach. + */ + mtx_lock(&fasttrap_cleanup_mtx); + fasttrap_timeout.callout = NULL; + work = fasttrap_cleanup_work; + mtx_unlock(&fasttrap_cleanup_mtx); + + if (work) + fasttrap_pid_cleanup(); + + (void) dtrace_meta_register("fasttrap", &fasttrap_mops, NULL, + &fasttrap_meta_id); + + return (-1); + } + +#ifdef DEBUG + mtx_lock(&fasttrap_count_mtx); + ASSERT(fasttrap_pid_count == 0); + mtx_unlock(&fasttrap_count_mtx); +#endif + + for (i = 0; i < fasttrap_tpoints.fth_nent; i++) + mtx_destroy(&fasttrap_tpoints.fth_table[i].ftb_mtx); + + kmem_free(fasttrap_tpoints.fth_table, + fasttrap_tpoints.fth_nent * sizeof (fasttrap_bucket_t)); + fasttrap_tpoints.fth_nent = 0; + + for (i = 0; i < fasttrap_provs.fth_nent; i++) + mtx_destroy(&fasttrap_provs.fth_table[i].ftb_mtx); + + kmem_free(fasttrap_provs.fth_table, + fasttrap_provs.fth_nent * sizeof (fasttrap_bucket_t)); + fasttrap_provs.fth_nent = 0; + + for (i = 0; i < fasttrap_procs.fth_nent; i++) + mtx_destroy(&fasttrap_procs.fth_table[i].ftb_mtx); + + kmem_free(fasttrap_procs.fth_table, + fasttrap_procs.fth_nent * sizeof (fasttrap_bucket_t)); + fasttrap_procs.fth_nent = 0; + + destroy_dev(fasttrap_cdev); + + return (error); +} + +static int +fasttrap_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + + return (error); +} + +static int +fasttrap_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) +{ + return (0); +} + +static int +fasttrap_ioctl(struct cdev *dev __unused, u_long cmd __unused, caddr_t addr __unused, int flags __unused, struct thread *td __unused) +{ +printf("%s(%d): DOODAD\n",__func__,__LINE__); +#ifdef DOODAD + if (!dtrace_attached()) + return (EAGAIN); + + if (cmd == FASTTRAPIOC_MAKEPROBE) { + fasttrap_probe_spec_t *uprobe = (void *)arg; + fasttrap_probe_spec_t *probe; + uint64_t noffs; + size_t size; + int ret; + char *c; + + if (copyin(&uprobe->ftps_noffs, &noffs, + sizeof (uprobe->ftps_noffs))) + return (EFAULT); + + /* + * Probes must have at least one tracepoint. + */ + if (noffs == 0) + return (EINVAL); + + size = sizeof (fasttrap_probe_spec_t) + + sizeof (probe->ftps_offs[0]) * (noffs - 1); + + if (size > 1024 * 1024) + return (ENOMEM); + + probe = kmem_alloc(size, KM_SLEEP); + + if (copyin(uprobe, probe, size) != 0) { + kmem_free(probe, size); + return (EFAULT); + } + + /* + * Verify that the function and module strings contain no + * funny characters. + */ + for (c = &probe->ftps_func[0]; *c != '\0'; c++) { + if (*c < 0x20 || 0x7f <= *c) { + ret = EINVAL; + goto err; + } + } + + for (c = &probe->ftps_mod[0]; *c != '\0'; c++) { + if (*c < 0x20 || 0x7f <= *c) { + ret = EINVAL; + goto err; + } + } + + if (!PRIV_POLICY_CHOICE(cr, PRIV_ALL, B_FALSE)) { + proc_t *p; + pid_t pid = probe->ftps_pid; + + mtx_lock(&pidlock); + /* + * Report an error if the process doesn't exist + * or is actively being birthed. + */ + if ((p = prfind(pid)) == NULL || p->p_stat == SIDL) { + mtx_unlock(&pidlock); + return (ESRCH); + } + mtx_lock(&p->p_lock); + mtx_unlock(&pidlock); + + if ((ret = priv_proc_cred_perm(cr, p, NULL, + VREAD | VWRITE)) != 0) { + mtx_unlock(&p->p_lock); + return (ret); + } + + mtx_unlock(&p->p_lock); + } + + ret = fasttrap_add_probe(probe); +err: + kmem_free(probe, size); + + return (ret); + + } else if (cmd == FASTTRAPIOC_GETINSTR) { + fasttrap_instr_query_t instr; + fasttrap_tracepoint_t *tp; + uint_t index; + int ret; + + if (copyin((void *)arg, &instr, sizeof (instr)) != 0) + return (EFAULT); + + if (!PRIV_POLICY_CHOICE(cr, PRIV_ALL, B_FALSE)) { + proc_t *p; + pid_t pid = instr.ftiq_pid; + + mtx_lock(&pidlock); + /* + * Report an error if the process doesn't exist + * or is actively being birthed. + */ + if ((p = prfind(pid)) == NULL || p->p_stat == SIDL) { + mtx_unlock(&pidlock); + return (ESRCH); + } + mtx_lock(&p->p_lock); + mtx_unlock(&pidlock); + + if ((ret = priv_proc_cred_perm(cr, p, NULL, + VREAD)) != 0) { + mtx_unlock(&p->p_lock); + return (ret); + } + + mtx_unlock(&p->p_lock); + } + + index = FASTTRAP_TPOINTS_INDEX(instr.ftiq_pid, instr.ftiq_pc); + + mtx_lock(&fasttrap_tpoints.fth_table[index].ftb_mtx); + tp = fasttrap_tpoints.fth_table[index].ftb_data; + while (tp != NULL) { + if (instr.ftiq_pid == tp->ftt_pid && + instr.ftiq_pc == tp->ftt_pc && + tp->ftt_proc->ftpc_acount != 0) + break; + + tp = tp->ftt_next; + } + + if (tp == NULL) { + mtx_unlock(&fasttrap_tpoints.fth_table[index].ftb_mtx); + return (ENOENT); + } + + bcopy(&tp->ftt_instr, &instr.ftiq_instr, + sizeof (instr.ftiq_instr)); + mtx_unlock(&fasttrap_tpoints.fth_table[index].ftb_mtx); + + if (copyout(&instr, (void *)arg, sizeof (instr)) != 0) + return (EFAULT); + + return (0); + } +#endif + + return (EINVAL); +} + +SYSINIT(fasttrap_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, fasttrap_load, NULL); +SYSUNINIT(fasttrap_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, fasttrap_unload, NULL); + +DEV_MODULE(fasttrap, fasttrap_modevent, NULL); +MODULE_VERSION(fasttrap, 1); +MODULE_DEPEND(fasttrap, dtrace, 1, 1, 1); +MODULE_DEPEND(fasttrap, opensolaris, 1, 1, 1); diff --git a/sys/cddl/dev/fbt/fbt.c b/sys/cddl/dev/fbt/fbt.c new file mode 100644 index 000000000000..f3ddc83e1140 --- /dev/null +++ b/sys/cddl/dev/fbt/fbt.c @@ -0,0 +1,1411 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * Portions Copyright 2006-2008 John Birrell jb@freebsd.org + * + * $FreeBSD$ + * + */ + +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#include <sys/cdefs.h> +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/cpuvar.h> +#include <sys/fcntl.h> +#include <sys/filio.h> +#include <sys/kdb.h> +#include <sys/kernel.h> +#include <sys/kmem.h> +#include <sys/kthread.h> +#include <sys/limits.h> +#include <sys/linker.h> +#include <sys/lock.h> +#include <sys/malloc.h> +#include <sys/module.h> +#include <sys/mutex.h> +#include <sys/pcpu.h> +#include <sys/poll.h> +#include <sys/proc.h> +#include <sys/selinfo.h> +#include <sys/smp.h> +#include <sys/syscall.h> +#include <sys/sysent.h> +#include <sys/sysproto.h> +#include <sys/uio.h> +#include <sys/unistd.h> +#include <machine/stdarg.h> + +#include <sys/dtrace.h> +#include <sys/dtrace_bsd.h> + +MALLOC_DEFINE(M_FBT, "fbt", "Function Boundary Tracing"); + +#define FBT_PUSHL_EBP 0x55 +#define FBT_MOVL_ESP_EBP0_V0 0x8b +#define FBT_MOVL_ESP_EBP1_V0 0xec +#define FBT_MOVL_ESP_EBP0_V1 0x89 +#define FBT_MOVL_ESP_EBP1_V1 0xe5 +#define FBT_REX_RSP_RBP 0x48 + +#define FBT_POPL_EBP 0x5d +#define FBT_RET 0xc3 +#define FBT_RET_IMM16 0xc2 +#define FBT_LEAVE 0xc9 + +#ifdef __amd64__ +#define FBT_PATCHVAL 0xcc +#else +#define FBT_PATCHVAL 0xf0 +#endif + +static d_open_t fbt_open; +static int fbt_unload(void); +static void fbt_getargdesc(void *, dtrace_id_t, void *, dtrace_argdesc_t *); +static void fbt_provide_module(void *, modctl_t *); +static void fbt_destroy(void *, dtrace_id_t, void *); +static void fbt_enable(void *, dtrace_id_t, void *); +static void fbt_disable(void *, dtrace_id_t, void *); +static void fbt_load(void *); +static void fbt_suspend(void *, dtrace_id_t, void *); +static void fbt_resume(void *, dtrace_id_t, void *); + +#define FBT_ENTRY "entry" +#define FBT_RETURN "return" +#define FBT_ADDR2NDX(addr) ((((uintptr_t)(addr)) >> 4) & fbt_probetab_mask) +#define FBT_PROBETAB_SIZE 0x8000 /* 32k entries -- 128K total */ + +static struct cdevsw fbt_cdevsw = { + .d_version = D_VERSION, + .d_open = fbt_open, + .d_name = "fbt", +}; + +static dtrace_pattr_t fbt_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pops_t fbt_pops = { + NULL, + fbt_provide_module, + fbt_enable, + fbt_disable, + fbt_suspend, + fbt_resume, + fbt_getargdesc, + NULL, + NULL, + fbt_destroy +}; + +typedef struct fbt_probe { + struct fbt_probe *fbtp_hashnext; + uint8_t *fbtp_patchpoint; + int8_t fbtp_rval; + uint8_t fbtp_patchval; + uint8_t fbtp_savedval; + uintptr_t fbtp_roffset; + dtrace_id_t fbtp_id; + const char *fbtp_name; + modctl_t *fbtp_ctl; + int fbtp_loadcnt; + int fbtp_primary; + int fbtp_invop_cnt; + int fbtp_symindx; + struct fbt_probe *fbtp_next; +} fbt_probe_t; + +static struct cdev *fbt_cdev; +static dtrace_provider_id_t fbt_id; +static fbt_probe_t **fbt_probetab; +static int fbt_probetab_size; +static int fbt_probetab_mask; +static int fbt_verbose = 0; + +static void +fbt_doubletrap(void) +{ + fbt_probe_t *fbt; + int i; + + for (i = 0; i < fbt_probetab_size; i++) { + fbt = fbt_probetab[i]; + + for (; fbt != NULL; fbt = fbt->fbtp_next) + *fbt->fbtp_patchpoint = fbt->fbtp_savedval; + } +} + +static int +fbt_invop(uintptr_t addr, uintptr_t *stack, uintptr_t rval) +{ + solaris_cpu_t *cpu = &solaris_cpu[curcpu]; + uintptr_t stack0, stack1, stack2, stack3, stack4; + fbt_probe_t *fbt = fbt_probetab[FBT_ADDR2NDX(addr)]; + + for (; fbt != NULL; fbt = fbt->fbtp_hashnext) { + if ((uintptr_t)fbt->fbtp_patchpoint == addr) { + fbt->fbtp_invop_cnt++; + if (fbt->fbtp_roffset == 0) { + int i = 0; + /* + * When accessing the arguments on the stack, + * we must protect against accessing beyond + * the stack. We can safely set NOFAULT here + * -- we know that interrupts are already + * disabled. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + cpu->cpu_dtrace_caller = stack[i++]; + stack0 = stack[i++]; + stack1 = stack[i++]; + stack2 = stack[i++]; + stack3 = stack[i++]; + stack4 = stack[i++]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | + CPU_DTRACE_BADADDR); + + dtrace_probe(fbt->fbtp_id, stack0, stack1, + stack2, stack3, stack4); + + cpu->cpu_dtrace_caller = 0; + } else { +#ifdef __amd64__ + /* + * On amd64, we instrument the ret, not the + * leave. We therefore need to set the caller + * to assure that the top frame of a stack() + * action is correct. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + cpu->cpu_dtrace_caller = stack[0]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | + CPU_DTRACE_BADADDR); +#endif + + dtrace_probe(fbt->fbtp_id, fbt->fbtp_roffset, + rval, 0, 0, 0); + cpu->cpu_dtrace_caller = 0; + } + + return (fbt->fbtp_rval); + } + } + + return (0); +} + +static int +fbt_provide_module_function(linker_file_t lf, int symindx, + linker_symval_t *symval, void *opaque) +{ + char *modname = opaque; + const char *name = symval->name; + fbt_probe_t *fbt, *retfbt; + int j; + int size; + u_int8_t *instr, *limit; + + if (strncmp(name, "dtrace_", 7) == 0 && + strncmp(name, "dtrace_safe_", 12) != 0) { + /* + * Anything beginning with "dtrace_" may be called + * from probe context unless it explicitly indicates + * that it won't be called from probe context by + * using the prefix "dtrace_safe_". + */ + return (0); + } + + if (name[0] == '_' && name[1] == '_') + return (0); + + size = symval->size; + + instr = (u_int8_t *) symval->value; + limit = (u_int8_t *) symval->value + symval->size; + +#ifdef __amd64__ + while (instr < limit) { + if (*instr == FBT_PUSHL_EBP) + break; + + if ((size = dtrace_instr_size(instr)) <= 0) + break; + + instr += size; + } + + if (instr >= limit || *instr != FBT_PUSHL_EBP) { + /* + * We either don't save the frame pointer in this + * function, or we ran into some disassembly + * screw-up. Either way, we bail. + */ + return (0); + } +#else + if (instr[0] != FBT_PUSHL_EBP) + return (0); + + if (!(instr[1] == FBT_MOVL_ESP_EBP0_V0 && + instr[2] == FBT_MOVL_ESP_EBP1_V0) && + !(instr[1] == FBT_MOVL_ESP_EBP0_V1 && + instr[2] == FBT_MOVL_ESP_EBP1_V1)) + return (0); +#endif + + fbt = malloc(sizeof (fbt_probe_t), M_FBT, M_WAITOK | M_ZERO); + fbt->fbtp_name = name; + fbt->fbtp_id = dtrace_probe_create(fbt_id, modname, + name, FBT_ENTRY, 3, fbt); + fbt->fbtp_patchpoint = instr; + fbt->fbtp_ctl = lf; + fbt->fbtp_loadcnt = lf->loadcnt; + fbt->fbtp_rval = DTRACE_INVOP_PUSHL_EBP; + fbt->fbtp_savedval = *instr; + fbt->fbtp_patchval = FBT_PATCHVAL; + fbt->fbtp_symindx = symindx; + + fbt->fbtp_hashnext = fbt_probetab[FBT_ADDR2NDX(instr)]; + fbt_probetab[FBT_ADDR2NDX(instr)] = fbt; + + lf->fbt_nentries++; + + retfbt = NULL; +again: + if (instr >= limit) + return (0); + + /* + * If this disassembly fails, then we've likely walked off into + * a jump table or some other unsuitable area. Bail out of the + * disassembly now. + */ + if ((size = dtrace_instr_size(instr)) <= 0) + return (0); + +#ifdef __amd64__ + /* + * We only instrument "ret" on amd64 -- we don't yet instrument + * ret imm16, largely because the compiler doesn't seem to + * (yet) emit them in the kernel... + */ + if (*instr != FBT_RET) { + instr += size; + goto again; + } +#else + if (!(size == 1 && + (*instr == FBT_POPL_EBP || *instr == FBT_LEAVE) && + (*(instr + 1) == FBT_RET || + *(instr + 1) == FBT_RET_IMM16))) { + instr += size; + goto again; + } +#endif + + /* + * We (desperately) want to avoid erroneously instrumenting a + * jump table, especially given that our markers are pretty + * short: two bytes on x86, and just one byte on amd64. To + * determine if we're looking at a true instruction sequence + * or an inline jump table that happens to contain the same + * byte sequences, we resort to some heuristic sleeze: we + * treat this instruction as being contained within a pointer, + * and see if that pointer points to within the body of the + * function. If it does, we refuse to instrument it. + */ + for (j = 0; j < sizeof (uintptr_t); j++) { + caddr_t check = (caddr_t) instr - j; + uint8_t *ptr; + + if (check < symval->value) + break; + + if (check + sizeof (caddr_t) > (caddr_t)limit) + continue; + + ptr = *(uint8_t **)check; + + if (ptr >= (uint8_t *) symval->value && ptr < limit) { + instr += size; + goto again; + } + } + + /* + * We have a winner! + */ + fbt = malloc(sizeof (fbt_probe_t), M_FBT, M_WAITOK | M_ZERO); + fbt->fbtp_name = name; + + if (retfbt == NULL) { + fbt->fbtp_id = dtrace_probe_create(fbt_id, modname, + name, FBT_RETURN, 3, fbt); + } else { + retfbt->fbtp_next = fbt; + fbt->fbtp_id = retfbt->fbtp_id; + } + + retfbt = fbt; + fbt->fbtp_patchpoint = instr; + fbt->fbtp_ctl = lf; + fbt->fbtp_loadcnt = lf->loadcnt; + fbt->fbtp_symindx = symindx; + +#ifndef __amd64__ + if (*instr == FBT_POPL_EBP) { + fbt->fbtp_rval = DTRACE_INVOP_POPL_EBP; + } else { + ASSERT(*instr == FBT_LEAVE); + fbt->fbtp_rval = DTRACE_INVOP_LEAVE; + } + fbt->fbtp_roffset = + (uintptr_t)(instr - (uint8_t *) symval->value) + 1; + +#else + ASSERT(*instr == FBT_RET); + fbt->fbtp_rval = DTRACE_INVOP_RET; + fbt->fbtp_roffset = + (uintptr_t)(instr - (uint8_t *) symval->value); +#endif + + fbt->fbtp_savedval = *instr; + fbt->fbtp_patchval = FBT_PATCHVAL; + fbt->fbtp_hashnext = fbt_probetab[FBT_ADDR2NDX(instr)]; + fbt_probetab[FBT_ADDR2NDX(instr)] = fbt; + + lf->fbt_nentries++; + + instr += size; + goto again; +} + +static void +fbt_provide_module(void *arg, modctl_t *lf) +{ + char modname[MAXPATHLEN]; + int i; + size_t len; + + strlcpy(modname, lf->filename, sizeof(modname)); + len = strlen(modname); + if (len > 3 && strcmp(modname + len - 3, ".ko") == 0) + modname[len - 3] = '\0'; + + /* + * Employees of dtrace and their families are ineligible. Void + * where prohibited. + */ + if (strcmp(modname, "dtrace") == 0) + return; + + /* + * The cyclic timer subsystem can be built as a module and DTrace + * depends on that, so it is ineligible too. + */ + if (strcmp(modname, "cyclic") == 0) + return; + + /* + * To register with DTrace, a module must list 'dtrace' as a + * dependency in order for the kernel linker to resolve + * symbols like dtrace_register(). All modules with such a + * dependency are ineligible for FBT tracing. + */ + for (i = 0; i < lf->ndeps; i++) + if (strncmp(lf->deps[i]->filename, "dtrace", 6) == 0) + return; + + if (lf->fbt_nentries) { + /* + * This module has some FBT entries allocated; we're afraid + * to screw with it. + */ + return; + } + + /* + * List the functions in the module and the symbol values. + */ + (void) linker_file_function_listall(lf, fbt_provide_module_function, modname); +} + +static void +fbt_destroy(void *arg, dtrace_id_t id, void *parg) +{ + fbt_probe_t *fbt = parg, *next, *hash, *last; + modctl_t *ctl; + int ndx; + + do { + ctl = fbt->fbtp_ctl; + + ctl->fbt_nentries--; + + /* + * Now we need to remove this probe from the fbt_probetab. + */ + ndx = FBT_ADDR2NDX(fbt->fbtp_patchpoint); + last = NULL; + hash = fbt_probetab[ndx]; + + while (hash != fbt) { + ASSERT(hash != NULL); + last = hash; + hash = hash->fbtp_hashnext; + } + + if (last != NULL) { + last->fbtp_hashnext = fbt->fbtp_hashnext; + } else { + fbt_probetab[ndx] = fbt->fbtp_hashnext; + } + + next = fbt->fbtp_next; + free(fbt, M_FBT); + + fbt = next; + } while (fbt != NULL); +} + +static void +fbt_enable(void *arg, dtrace_id_t id, void *parg) +{ + fbt_probe_t *fbt = parg; + modctl_t *ctl = fbt->fbtp_ctl; + + ctl->nenabled++; + + /* + * Now check that our modctl has the expected load count. If it + * doesn't, this module must have been unloaded and reloaded -- and + * we're not going to touch it. + */ + if (ctl->loadcnt != fbt->fbtp_loadcnt) { + if (fbt_verbose) { + printf("fbt is failing for probe %s " + "(module %s reloaded)", + fbt->fbtp_name, ctl->filename); + } + + return; + } + + for (; fbt != NULL; fbt = fbt->fbtp_next) { + *fbt->fbtp_patchpoint = fbt->fbtp_patchval; + } +} + +static void +fbt_disable(void *arg, dtrace_id_t id, void *parg) +{ + fbt_probe_t *fbt = parg; + modctl_t *ctl = fbt->fbtp_ctl; + + ASSERT(ctl->nenabled > 0); + ctl->nenabled--; + + if ((ctl->loadcnt != fbt->fbtp_loadcnt)) + return; + + for (; fbt != NULL; fbt = fbt->fbtp_next) + *fbt->fbtp_patchpoint = fbt->fbtp_savedval; +} + +static void +fbt_suspend(void *arg, dtrace_id_t id, void *parg) +{ + fbt_probe_t *fbt = parg; + modctl_t *ctl = fbt->fbtp_ctl; + + ASSERT(ctl->nenabled > 0); + + if ((ctl->loadcnt != fbt->fbtp_loadcnt)) + return; + + for (; fbt != NULL; fbt = fbt->fbtp_next) + *fbt->fbtp_patchpoint = fbt->fbtp_savedval; +} + +static void +fbt_resume(void *arg, dtrace_id_t id, void *parg) +{ + fbt_probe_t *fbt = parg; + modctl_t *ctl = fbt->fbtp_ctl; + + ASSERT(ctl->nenabled > 0); + + if ((ctl->loadcnt != fbt->fbtp_loadcnt)) + return; + + for (; fbt != NULL; fbt = fbt->fbtp_next) + *fbt->fbtp_patchpoint = fbt->fbtp_patchval; +} + +static int +fbt_ctfoff_init(modctl_t *lf, linker_ctf_t *lc) +{ + const Elf_Sym *symp = lc->symtab;; + const char *name; + const ctf_header_t *hp = (const ctf_header_t *) lc->ctftab; + const uint8_t *ctfdata = lc->ctftab + sizeof(ctf_header_t); + int i; + uint32_t *ctfoff; + uint32_t objtoff = hp->cth_objtoff; + uint32_t funcoff = hp->cth_funcoff; + ushort_t info; + ushort_t vlen; + + /* Sanity check. */ + if (hp->cth_magic != CTF_MAGIC) { + printf("Bad magic value in CTF data of '%s'\n",lf->pathname); + return (EINVAL); + } + + if (lc->symtab == NULL) { + printf("No symbol table in '%s'\n",lf->pathname); + return (EINVAL); + } + + if ((ctfoff = malloc(sizeof(uint32_t) * lc->nsym, M_LINKER, M_WAITOK)) == NULL) + return (ENOMEM); + + *lc->ctfoffp = ctfoff; + + for (i = 0; i < lc->nsym; i++, ctfoff++, symp++) { + if (symp->st_name == 0 || symp->st_shndx == SHN_UNDEF) { + *ctfoff = 0xffffffff; + continue; + } + + if (symp->st_name < lc->strcnt) + name = lc->strtab + symp->st_name; + else + name = "(?)"; + + switch (ELF_ST_TYPE(symp->st_info)) { + case STT_OBJECT: + if (objtoff >= hp->cth_funcoff || + (symp->st_shndx == SHN_ABS && symp->st_value == 0)) { + *ctfoff = 0xffffffff; + break; + } + + *ctfoff = objtoff; + objtoff += sizeof (ushort_t); + break; + + case STT_FUNC: + if (funcoff >= hp->cth_typeoff) { + *ctfoff = 0xffffffff; + break; + } + + *ctfoff = funcoff; + + info = *((const ushort_t *)(ctfdata + funcoff)); + vlen = CTF_INFO_VLEN(info); + + /* + * If we encounter a zero pad at the end, just skip it. + * Otherwise skip over the function and its return type + * (+2) and the argument list (vlen). + */ + if (CTF_INFO_KIND(info) == CTF_K_UNKNOWN && vlen == 0) + funcoff += sizeof (ushort_t); /* skip pad */ + else + funcoff += sizeof (ushort_t) * (vlen + 2); + break; + + default: + *ctfoff = 0xffffffff; + break; + } + } + + return (0); +} + +static ssize_t +fbt_get_ctt_size(uint8_t version, const ctf_type_t *tp, ssize_t *sizep, + ssize_t *incrementp) +{ + ssize_t size, increment; + + if (version > CTF_VERSION_1 && + tp->ctt_size == CTF_LSIZE_SENT) { + size = CTF_TYPE_LSIZE(tp); + increment = sizeof (ctf_type_t); + } else { + size = tp->ctt_size; + increment = sizeof (ctf_stype_t); + } + + if (sizep) + *sizep = size; + if (incrementp) + *incrementp = increment; + + return (size); +} + +static int +fbt_typoff_init(linker_ctf_t *lc) +{ + const ctf_header_t *hp = (const ctf_header_t *) lc->ctftab; + const ctf_type_t *tbuf; + const ctf_type_t *tend; + const ctf_type_t *tp; + const uint8_t *ctfdata = lc->ctftab + sizeof(ctf_header_t); + int ctf_typemax = 0; + uint32_t *xp; + ulong_t pop[CTF_K_MAX + 1] = { 0 }; + + + /* Sanity check. */ + if (hp->cth_magic != CTF_MAGIC) + return (EINVAL); + + tbuf = (const ctf_type_t *) (ctfdata + hp->cth_typeoff); + tend = (const ctf_type_t *) (ctfdata + hp->cth_stroff); + + int child = hp->cth_parname != 0; + + /* + * We make two passes through the entire type section. In this first + * pass, we count the number of each type and the total number of types. + */ + for (tp = tbuf; tp < tend; ctf_typemax++) { + ushort_t kind = CTF_INFO_KIND(tp->ctt_info); + ulong_t vlen = CTF_INFO_VLEN(tp->ctt_info); + ssize_t size, increment; + + size_t vbytes; + uint_t n; + + (void) fbt_get_ctt_size(hp->cth_version, tp, &size, &increment); + + switch (kind) { + case CTF_K_INTEGER: + case CTF_K_FLOAT: + vbytes = sizeof (uint_t); + break; + case CTF_K_ARRAY: + vbytes = sizeof (ctf_array_t); + break; + case CTF_K_FUNCTION: + vbytes = sizeof (ushort_t) * (vlen + (vlen & 1)); + break; + case CTF_K_STRUCT: + case CTF_K_UNION: + if (size < CTF_LSTRUCT_THRESH) { + ctf_member_t *mp = (ctf_member_t *) + ((uintptr_t)tp + increment); + + vbytes = sizeof (ctf_member_t) * vlen; + for (n = vlen; n != 0; n--, mp++) + child |= CTF_TYPE_ISCHILD(mp->ctm_type); + } else { + ctf_lmember_t *lmp = (ctf_lmember_t *) + ((uintptr_t)tp + increment); + + vbytes = sizeof (ctf_lmember_t) * vlen; + for (n = vlen; n != 0; n--, lmp++) + child |= + CTF_TYPE_ISCHILD(lmp->ctlm_type); + } + break; + case CTF_K_ENUM: + vbytes = sizeof (ctf_enum_t) * vlen; + break; + case CTF_K_FORWARD: + /* + * For forward declarations, ctt_type is the CTF_K_* + * kind for the tag, so bump that population count too. + * If ctt_type is unknown, treat the tag as a struct. + */ + if (tp->ctt_type == CTF_K_UNKNOWN || + tp->ctt_type >= CTF_K_MAX) + pop[CTF_K_STRUCT]++; + else + pop[tp->ctt_type]++; + /*FALLTHRU*/ + case CTF_K_UNKNOWN: + vbytes = 0; + break; + case CTF_K_POINTER: + case CTF_K_TYPEDEF: + case CTF_K_VOLATILE: + case CTF_K_CONST: + case CTF_K_RESTRICT: + child |= CTF_TYPE_ISCHILD(tp->ctt_type); + vbytes = 0; + break; + default: + printf("%s(%d): detected invalid CTF kind -- %u\n", __func__, __LINE__, kind); + return (EIO); + } + tp = (ctf_type_t *)((uintptr_t)tp + increment + vbytes); + pop[kind]++; + } + + *lc->typlenp = ctf_typemax; + + if ((xp = malloc(sizeof(uint32_t) * ctf_typemax, M_LINKER, M_ZERO | M_WAITOK)) == NULL) + return (ENOMEM); + + *lc->typoffp = xp; + + /* type id 0 is used as a sentinel value */ + *xp++ = 0; + + /* + * In the second pass, fill in the type offset. + */ + for (tp = tbuf; tp < tend; xp++) { + ushort_t kind = CTF_INFO_KIND(tp->ctt_info); + ulong_t vlen = CTF_INFO_VLEN(tp->ctt_info); + ssize_t size, increment; + + size_t vbytes; + uint_t n; + + (void) fbt_get_ctt_size(hp->cth_version, tp, &size, &increment); + + switch (kind) { + case CTF_K_INTEGER: + case CTF_K_FLOAT: + vbytes = sizeof (uint_t); + break; + case CTF_K_ARRAY: + vbytes = sizeof (ctf_array_t); + break; + case CTF_K_FUNCTION: + vbytes = sizeof (ushort_t) * (vlen + (vlen & 1)); + break; + case CTF_K_STRUCT: + case CTF_K_UNION: + if (size < CTF_LSTRUCT_THRESH) { + ctf_member_t *mp = (ctf_member_t *) + ((uintptr_t)tp + increment); + + vbytes = sizeof (ctf_member_t) * vlen; + for (n = vlen; n != 0; n--, mp++) + child |= CTF_TYPE_ISCHILD(mp->ctm_type); + } else { + ctf_lmember_t *lmp = (ctf_lmember_t *) + ((uintptr_t)tp + increment); + + vbytes = sizeof (ctf_lmember_t) * vlen; + for (n = vlen; n != 0; n--, lmp++) + child |= + CTF_TYPE_ISCHILD(lmp->ctlm_type); + } + break; + case CTF_K_ENUM: + vbytes = sizeof (ctf_enum_t) * vlen; + break; + case CTF_K_FORWARD: + case CTF_K_UNKNOWN: + vbytes = 0; + break; + case CTF_K_POINTER: + case CTF_K_TYPEDEF: + case CTF_K_VOLATILE: + case CTF_K_CONST: + case CTF_K_RESTRICT: + vbytes = 0; + break; + default: + printf("%s(%d): detected invalid CTF kind -- %u\n", __func__, __LINE__, kind); + return (EIO); + } + *xp = (uint32_t)((uintptr_t) tp - (uintptr_t) ctfdata); + tp = (ctf_type_t *)((uintptr_t)tp + increment + vbytes); + } + + return (0); +} + +/* + * CTF Declaration Stack + * + * In order to implement ctf_type_name(), we must convert a type graph back + * into a C type declaration. Unfortunately, a type graph represents a storage + * class ordering of the type whereas a type declaration must obey the C rules + * for operator precedence, and the two orderings are frequently in conflict. + * For example, consider these CTF type graphs and their C declarations: + * + * CTF_K_POINTER -> CTF_K_FUNCTION -> CTF_K_INTEGER : int (*)() + * CTF_K_POINTER -> CTF_K_ARRAY -> CTF_K_INTEGER : int (*)[] + * + * In each case, parentheses are used to raise operator * to higher lexical + * precedence, so the string form of the C declaration cannot be constructed by + * walking the type graph links and forming the string from left to right. + * + * The functions in this file build a set of stacks from the type graph nodes + * corresponding to the C operator precedence levels in the appropriate order. + * The code in ctf_type_name() can then iterate over the levels and nodes in + * lexical precedence order and construct the final C declaration string. + */ +typedef struct ctf_list { + struct ctf_list *l_prev; /* previous pointer or tail pointer */ + struct ctf_list *l_next; /* next pointer or head pointer */ +} ctf_list_t; + +#define ctf_list_prev(elem) ((void *)(((ctf_list_t *)(elem))->l_prev)) +#define ctf_list_next(elem) ((void *)(((ctf_list_t *)(elem))->l_next)) + +typedef enum { + CTF_PREC_BASE, + CTF_PREC_POINTER, + CTF_PREC_ARRAY, + CTF_PREC_FUNCTION, + CTF_PREC_MAX +} ctf_decl_prec_t; + +typedef struct ctf_decl_node { + ctf_list_t cd_list; /* linked list pointers */ + ctf_id_t cd_type; /* type identifier */ + uint_t cd_kind; /* type kind */ + uint_t cd_n; /* type dimension if array */ +} ctf_decl_node_t; + +typedef struct ctf_decl { + ctf_list_t cd_nodes[CTF_PREC_MAX]; /* declaration node stacks */ + int cd_order[CTF_PREC_MAX]; /* storage order of decls */ + ctf_decl_prec_t cd_qualp; /* qualifier precision */ + ctf_decl_prec_t cd_ordp; /* ordered precision */ + char *cd_buf; /* buffer for output */ + char *cd_ptr; /* buffer location */ + char *cd_end; /* buffer limit */ + size_t cd_len; /* buffer space required */ + int cd_err; /* saved error value */ +} ctf_decl_t; + +/* + * Simple doubly-linked list append routine. This implementation assumes that + * each list element contains an embedded ctf_list_t as the first member. + * An additional ctf_list_t is used to store the head (l_next) and tail + * (l_prev) pointers. The current head and tail list elements have their + * previous and next pointers set to NULL, respectively. + */ +static void +ctf_list_append(ctf_list_t *lp, void *new) +{ + ctf_list_t *p = lp->l_prev; /* p = tail list element */ + ctf_list_t *q = new; /* q = new list element */ + + lp->l_prev = q; + q->l_prev = p; + q->l_next = NULL; + + if (p != NULL) + p->l_next = q; + else + lp->l_next = q; +} + +/* + * Prepend the specified existing element to the given ctf_list_t. The + * existing pointer should be pointing at a struct with embedded ctf_list_t. + */ +static void +ctf_list_prepend(ctf_list_t *lp, void *new) +{ + ctf_list_t *p = new; /* p = new list element */ + ctf_list_t *q = lp->l_next; /* q = head list element */ + + lp->l_next = p; + p->l_prev = NULL; + p->l_next = q; + + if (q != NULL) + q->l_prev = p; + else + lp->l_prev = p; +} + +static void +ctf_decl_init(ctf_decl_t *cd, char *buf, size_t len) +{ + int i; + + bzero(cd, sizeof (ctf_decl_t)); + + for (i = CTF_PREC_BASE; i < CTF_PREC_MAX; i++) + cd->cd_order[i] = CTF_PREC_BASE - 1; + + cd->cd_qualp = CTF_PREC_BASE; + cd->cd_ordp = CTF_PREC_BASE; + + cd->cd_buf = buf; + cd->cd_ptr = buf; + cd->cd_end = buf + len; +} + +static void +ctf_decl_fini(ctf_decl_t *cd) +{ + ctf_decl_node_t *cdp, *ndp; + int i; + + for (i = CTF_PREC_BASE; i < CTF_PREC_MAX; i++) { + for (cdp = ctf_list_next(&cd->cd_nodes[i]); + cdp != NULL; cdp = ndp) { + ndp = ctf_list_next(cdp); + free(cdp, M_FBT); + } + } +} + +static const ctf_type_t * +ctf_lookup_by_id(linker_ctf_t *lc, ctf_id_t type) +{ + const ctf_type_t *tp; + uint32_t offset; + uint32_t *typoff = *lc->typoffp; + + if (type >= *lc->typlenp) { + printf("%s(%d): type %d exceeds max %ld\n",__func__,__LINE__,(int) type,*lc->typlenp); + return(NULL); + } + + /* Check if the type isn't cross-referenced. */ + if ((offset = typoff[type]) == 0) { + printf("%s(%d): type %d isn't cross referenced\n",__func__,__LINE__, (int) type); + return(NULL); + } + + tp = (const ctf_type_t *)(lc->ctftab + offset + sizeof(ctf_header_t)); + + return (tp); +} + +static void +fbt_array_info(linker_ctf_t *lc, ctf_id_t type, ctf_arinfo_t *arp) +{ + const ctf_header_t *hp = (const ctf_header_t *) lc->ctftab; + const ctf_type_t *tp; + const ctf_array_t *ap; + ssize_t increment; + + bzero(arp, sizeof(*arp)); + + if ((tp = ctf_lookup_by_id(lc, type)) == NULL) + return; + + if (CTF_INFO_KIND(tp->ctt_info) != CTF_K_ARRAY) + return; + + (void) fbt_get_ctt_size(hp->cth_version, tp, NULL, &increment); + + ap = (const ctf_array_t *)((uintptr_t)tp + increment); + arp->ctr_contents = ap->cta_contents; + arp->ctr_index = ap->cta_index; + arp->ctr_nelems = ap->cta_nelems; +} + +static const char * +ctf_strptr(linker_ctf_t *lc, int name) +{ + const ctf_header_t *hp = (const ctf_header_t *) lc->ctftab;; + const char *strp = ""; + + if (name < 0 || name >= hp->cth_strlen) + return(strp); + + strp = (const char *)(lc->ctftab + hp->cth_stroff + name + sizeof(ctf_header_t)); + + return (strp); +} + +static void +ctf_decl_push(ctf_decl_t *cd, linker_ctf_t *lc, ctf_id_t type) +{ + ctf_decl_node_t *cdp; + ctf_decl_prec_t prec; + uint_t kind, n = 1; + int is_qual = 0; + + const ctf_type_t *tp; + ctf_arinfo_t ar; + + if ((tp = ctf_lookup_by_id(lc, type)) == NULL) { + cd->cd_err = ENOENT; + return; + } + + switch (kind = CTF_INFO_KIND(tp->ctt_info)) { + case CTF_K_ARRAY: + fbt_array_info(lc, type, &ar); + ctf_decl_push(cd, lc, ar.ctr_contents); + n = ar.ctr_nelems; + prec = CTF_PREC_ARRAY; + break; + + case CTF_K_TYPEDEF: + if (ctf_strptr(lc, tp->ctt_name)[0] == '\0') { + ctf_decl_push(cd, lc, tp->ctt_type); + return; + } + prec = CTF_PREC_BASE; + break; + + case CTF_K_FUNCTION: + ctf_decl_push(cd, lc, tp->ctt_type); + prec = CTF_PREC_FUNCTION; + break; + + case CTF_K_POINTER: + ctf_decl_push(cd, lc, tp->ctt_type); + prec = CTF_PREC_POINTER; + break; + + case CTF_K_VOLATILE: + case CTF_K_CONST: + case CTF_K_RESTRICT: + ctf_decl_push(cd, lc, tp->ctt_type); + prec = cd->cd_qualp; + is_qual++; + break; + + default: + prec = CTF_PREC_BASE; + } + + if ((cdp = malloc(sizeof (ctf_decl_node_t), M_FBT, M_WAITOK)) == NULL) { + cd->cd_err = EAGAIN; + return; + } + + cdp->cd_type = type; + cdp->cd_kind = kind; + cdp->cd_n = n; + + if (ctf_list_next(&cd->cd_nodes[prec]) == NULL) + cd->cd_order[prec] = cd->cd_ordp++; + + /* + * Reset cd_qualp to the highest precedence level that we've seen so + * far that can be qualified (CTF_PREC_BASE or CTF_PREC_POINTER). + */ + if (prec > cd->cd_qualp && prec < CTF_PREC_ARRAY) + cd->cd_qualp = prec; + + /* + * C array declarators are ordered inside out so prepend them. Also by + * convention qualifiers of base types precede the type specifier (e.g. + * const int vs. int const) even though the two forms are equivalent. + */ + if (kind == CTF_K_ARRAY || (is_qual && prec == CTF_PREC_BASE)) + ctf_list_prepend(&cd->cd_nodes[prec], cdp); + else + ctf_list_append(&cd->cd_nodes[prec], cdp); +} + +static void +ctf_decl_sprintf(ctf_decl_t *cd, const char *format, ...) +{ + size_t len = (size_t)(cd->cd_end - cd->cd_ptr); + va_list ap; + size_t n; + + va_start(ap, format); + n = vsnprintf(cd->cd_ptr, len, format, ap); + va_end(ap); + + cd->cd_ptr += MIN(n, len); + cd->cd_len += n; +} + +static ssize_t +fbt_type_name(linker_ctf_t *lc, ctf_id_t type, char *buf, size_t len) +{ + ctf_decl_t cd; + ctf_decl_node_t *cdp; + ctf_decl_prec_t prec, lp, rp; + int ptr, arr; + uint_t k; + + if (lc == NULL && type == CTF_ERR) + return (-1); /* simplify caller code by permitting CTF_ERR */ + + ctf_decl_init(&cd, buf, len); + ctf_decl_push(&cd, lc, type); + + if (cd.cd_err != 0) { + ctf_decl_fini(&cd); + return (-1); + } + + /* + * If the type graph's order conflicts with lexical precedence order + * for pointers or arrays, then we need to surround the declarations at + * the corresponding lexical precedence with parentheses. This can + * result in either a parenthesized pointer (*) as in int (*)() or + * int (*)[], or in a parenthesized pointer and array as in int (*[])(). + */ + ptr = cd.cd_order[CTF_PREC_POINTER] > CTF_PREC_POINTER; + arr = cd.cd_order[CTF_PREC_ARRAY] > CTF_PREC_ARRAY; + + rp = arr ? CTF_PREC_ARRAY : ptr ? CTF_PREC_POINTER : -1; + lp = ptr ? CTF_PREC_POINTER : arr ? CTF_PREC_ARRAY : -1; + + k = CTF_K_POINTER; /* avoid leading whitespace (see below) */ + + for (prec = CTF_PREC_BASE; prec < CTF_PREC_MAX; prec++) { + for (cdp = ctf_list_next(&cd.cd_nodes[prec]); + cdp != NULL; cdp = ctf_list_next(cdp)) { + + const ctf_type_t *tp = + ctf_lookup_by_id(lc, cdp->cd_type); + const char *name = ctf_strptr(lc, tp->ctt_name); + + if (k != CTF_K_POINTER && k != CTF_K_ARRAY) + ctf_decl_sprintf(&cd, " "); + + if (lp == prec) { + ctf_decl_sprintf(&cd, "("); + lp = -1; + } + + switch (cdp->cd_kind) { + case CTF_K_INTEGER: + case CTF_K_FLOAT: + case CTF_K_TYPEDEF: + ctf_decl_sprintf(&cd, "%s", name); + break; + case CTF_K_POINTER: + ctf_decl_sprintf(&cd, "*"); + break; + case CTF_K_ARRAY: + ctf_decl_sprintf(&cd, "[%u]", cdp->cd_n); + break; + case CTF_K_FUNCTION: + ctf_decl_sprintf(&cd, "()"); + break; + case CTF_K_STRUCT: + case CTF_K_FORWARD: + ctf_decl_sprintf(&cd, "struct %s", name); + break; + case CTF_K_UNION: + ctf_decl_sprintf(&cd, "union %s", name); + break; + case CTF_K_ENUM: + ctf_decl_sprintf(&cd, "enum %s", name); + break; + case CTF_K_VOLATILE: + ctf_decl_sprintf(&cd, "volatile"); + break; + case CTF_K_CONST: + ctf_decl_sprintf(&cd, "const"); + break; + case CTF_K_RESTRICT: + ctf_decl_sprintf(&cd, "restrict"); + break; + } + + k = cdp->cd_kind; + } + + if (rp == prec) + ctf_decl_sprintf(&cd, ")"); + } + + ctf_decl_fini(&cd); + return (cd.cd_len); +} + +static void +fbt_getargdesc(void *arg __unused, dtrace_id_t id __unused, void *parg, dtrace_argdesc_t *desc) +{ + const ushort_t *dp; + fbt_probe_t *fbt = parg; + linker_ctf_t lc; + modctl_t *ctl = fbt->fbtp_ctl; + int ndx = desc->dtargd_ndx; + int symindx = fbt->fbtp_symindx; + uint32_t *ctfoff; + uint32_t offset; + ushort_t info, kind, n; + + desc->dtargd_ndx = DTRACE_ARGNONE; + + /* Get a pointer to the CTF data and it's length. */ + if (linker_ctf_get(ctl, &lc) != 0) + /* No CTF data? Something wrong? *shrug* */ + return; + + /* Check if this module hasn't been initialised yet. */ + if (*lc.ctfoffp == NULL) { + /* + * Initialise the CTF object and function symindx to + * byte offset array. + */ + if (fbt_ctfoff_init(ctl, &lc) != 0) + return; + + /* Initialise the CTF type to byte offset array. */ + if (fbt_typoff_init(&lc) != 0) + return; + } + + ctfoff = *lc.ctfoffp; + + if (ctfoff == NULL || *lc.typoffp == NULL) + return; + + /* Check if the symbol index is out of range. */ + if (symindx >= lc.nsym) + return; + + /* Check if the symbol isn't cross-referenced. */ + if ((offset = ctfoff[symindx]) == 0xffffffff) + return; + + dp = (const ushort_t *)(lc.ctftab + offset + sizeof(ctf_header_t)); + + info = *dp++; + kind = CTF_INFO_KIND(info); + n = CTF_INFO_VLEN(info); + + if (kind == CTF_K_UNKNOWN && n == 0) { + printf("%s(%d): Unknown function!\n",__func__,__LINE__); + return; + } + + if (kind != CTF_K_FUNCTION) { + printf("%s(%d): Expected a function!\n",__func__,__LINE__); + return; + } + + /* Check if the requested argument doesn't exist. */ + if (ndx >= n) + return; + + /* Skip the return type and arguments up to the one requested. */ + dp += ndx + 1; + + if (fbt_type_name(&lc, *dp, desc->dtargd_native, sizeof(desc->dtargd_native)) > 0) + desc->dtargd_ndx = ndx; + + return; +} + +static void +fbt_load(void *dummy) +{ + /* Create the /dev/dtrace/fbt entry. */ + fbt_cdev = make_dev(&fbt_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, + "dtrace/fbt"); + + /* Default the probe table size if not specified. */ + if (fbt_probetab_size == 0) + fbt_probetab_size = FBT_PROBETAB_SIZE; + + /* Choose the hash mask for the probe table. */ + fbt_probetab_mask = fbt_probetab_size - 1; + + /* Allocate memory for the probe table. */ + fbt_probetab = + malloc(fbt_probetab_size * sizeof (fbt_probe_t *), M_FBT, M_WAITOK | M_ZERO); + + dtrace_doubletrap_func = fbt_doubletrap; + dtrace_invop_add(fbt_invop); + + if (dtrace_register("fbt", &fbt_attr, DTRACE_PRIV_USER, + NULL, &fbt_pops, NULL, &fbt_id) != 0) + return; +} + + +static int +fbt_unload() +{ + int error = 0; + + /* De-register the invalid opcode handler. */ + dtrace_invop_remove(fbt_invop); + + dtrace_doubletrap_func = NULL; + + /* De-register this DTrace provider. */ + if ((error = dtrace_unregister(fbt_id)) != 0) + return (error); + + /* Free the probe table. */ + free(fbt_probetab, M_FBT); + fbt_probetab = NULL; + fbt_probetab_mask = 0; + + destroy_dev(fbt_cdev); + + return (error); +} + +static int +fbt_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + + return (error); +} + +static int +fbt_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) +{ + return (0); +} + +SYSINIT(fbt_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, fbt_load, NULL); +SYSUNINIT(fbt_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, fbt_unload, NULL); + +DEV_MODULE(fbt, fbt_modevent, NULL); +MODULE_VERSION(fbt, 1); +MODULE_DEPEND(fbt, dtrace, 1, 1, 1); +MODULE_DEPEND(fbt, opensolaris, 1, 1, 1); diff --git a/sys/cddl/dev/lockstat/lockstat.c b/sys/cddl/dev/lockstat/lockstat.c new file mode 100644 index 000000000000..5a0bd63bec9e --- /dev/null +++ b/sys/cddl/dev/lockstat/lockstat.c @@ -0,0 +1,289 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * Portions Copyright (c) 2008 Stacey Son <stacey@son.org> + * + * $FreeBSD$ + * + */ + +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#define KDTRACE_HOOKS + +#include <sys/cdefs.h> +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/kernel.h> +#include <sys/limits.h> +#include <sys/lock.h> +#include <sys/linker.h> +#include <sys/module.h> +#include <sys/mutex.h> + +#include <sys/dtrace.h> +#include <sys/lockstat.h> + +#if defined(__i386__) || defined(__amd64__) +#define LOCKSTAT_AFRAMES 1 +#else +#error "architecture not supported" +#endif + +static d_open_t lockstat_open; +static void lockstat_provide(void *, dtrace_probedesc_t *); +static void lockstat_destroy(void *, dtrace_id_t, void *); +static void lockstat_enable(void *, dtrace_id_t, void *); +static void lockstat_disable(void *, dtrace_id_t, void *); +static void lockstat_load(void *); +static int lockstat_unload(void); + + +typedef struct lockstat_probe { + char *lsp_func; + char *lsp_name; + int lsp_probe; + dtrace_id_t lsp_id; +} lockstat_probe_t; + +#ifdef __FreeBSD__ +lockstat_probe_t lockstat_probes[] = +{ + /* Spin Locks */ + { LS_MTX_SPIN_LOCK, LSS_ACQUIRE, LS_MTX_SPIN_LOCK_ACQUIRE, DTRACE_IDNONE }, + { LS_MTX_SPIN_LOCK, LSS_SPIN, LS_MTX_SPIN_LOCK_SPIN, DTRACE_IDNONE }, + { LS_MTX_SPIN_UNLOCK, LSS_RELEASE, LS_MTX_SPIN_UNLOCK_RELEASE, DTRACE_IDNONE }, + /* Adaptive Locks */ + { LS_MTX_LOCK, LSA_ACQUIRE, LS_MTX_LOCK_ACQUIRE, DTRACE_IDNONE }, + { LS_MTX_LOCK, LSA_BLOCK, LS_MTX_LOCK_BLOCK, DTRACE_IDNONE }, + { LS_MTX_LOCK, LSA_SPIN, LS_MTX_LOCK_SPIN, DTRACE_IDNONE }, + { LS_MTX_UNLOCK, LSA_RELEASE, LS_MTX_UNLOCK_RELEASE, DTRACE_IDNONE }, + { LS_MTX_TRYLOCK, LSA_ACQUIRE, LS_MTX_TRYLOCK_ACQUIRE, DTRACE_IDNONE }, + /* Reader/Writer Locks */ + { LS_RW_RLOCK, LSR_ACQUIRE, LS_RW_RLOCK_ACQUIRE, DTRACE_IDNONE }, + { LS_RW_RLOCK, LSR_BLOCK, LS_RW_RLOCK_BLOCK, DTRACE_IDNONE }, + { LS_RW_RLOCK, LSR_SPIN, LS_RW_RLOCK_SPIN, DTRACE_IDNONE }, + { LS_RW_RUNLOCK, LSR_RELEASE, LS_RW_RUNLOCK_RELEASE, DTRACE_IDNONE }, + { LS_RW_WLOCK, LSR_ACQUIRE, LS_RW_WLOCK_ACQUIRE, DTRACE_IDNONE }, + { LS_RW_WLOCK, LSR_BLOCK, LS_RW_WLOCK_BLOCK, DTRACE_IDNONE }, + { LS_RW_WLOCK, LSR_SPIN, LS_RW_WLOCK_SPIN, DTRACE_IDNONE }, + { LS_RW_WUNLOCK, LSR_RELEASE, LS_RW_WUNLOCK_RELEASE, DTRACE_IDNONE }, + { LS_RW_TRYUPGRADE, LSR_UPGRADE, LS_RW_TRYUPGRADE_UPGRADE, DTRACE_IDNONE }, + { LS_RW_DOWNGRADE, LSR_DOWNGRADE, LS_RW_DOWNGRADE_DOWNGRADE, DTRACE_IDNONE }, + /* Shared/Exclusive Locks */ + { LS_SX_SLOCK, LSX_ACQUIRE, LS_SX_SLOCK_ACQUIRE, DTRACE_IDNONE }, + { LS_SX_SLOCK, LSX_BLOCK, LS_SX_SLOCK_BLOCK, DTRACE_IDNONE }, + { LS_SX_SLOCK, LSX_SPIN, LS_SX_SLOCK_SPIN, DTRACE_IDNONE }, + { LS_SX_SUNLOCK, LSX_RELEASE, LS_SX_SUNLOCK_RELEASE, DTRACE_IDNONE }, + { LS_SX_XLOCK, LSX_ACQUIRE, LS_SX_XLOCK_ACQUIRE, DTRACE_IDNONE }, + { LS_SX_XLOCK, LSX_BLOCK, LS_SX_XLOCK_BLOCK, DTRACE_IDNONE }, + { LS_SX_XLOCK, LSX_SPIN, LS_SX_XLOCK_SPIN, DTRACE_IDNONE }, + { LS_SX_XUNLOCK, LSX_RELEASE, LS_SX_XUNLOCK_RELEASE, DTRACE_IDNONE }, + { LS_SX_TRYUPGRADE, LSX_UPGRADE, LS_SX_TRYUPGRADE_UPGRADE, DTRACE_IDNONE }, + { LS_SX_DOWNGRADE, LSX_DOWNGRADE, LS_SX_DOWNGRADE_DOWNGRADE, DTRACE_IDNONE }, + /* Thread Locks */ + { LS_THREAD_LOCK, LST_SPIN, LS_THREAD_LOCK_SPIN, DTRACE_IDNONE }, + { NULL } +}; +#else +#error "OS not supported" +#endif + + +static struct cdevsw lockstat_cdevsw = { + .d_version = D_VERSION, + .d_open = lockstat_open, + .d_name = "lockstat", +}; + +static struct cdev *lockstat_cdev; +static dtrace_provider_id_t lockstat_id; + +/*ARGSUSED*/ +static void +lockstat_enable(void *arg, dtrace_id_t id, void *parg) +{ + lockstat_probe_t *probe = parg; + + ASSERT(!lockstat_probemap[probe->lsp_probe]); + + lockstat_probemap[probe->lsp_probe] = id; +#ifdef DOODAD + membar_producer(); +#endif + + lockstat_probe_func = dtrace_probe; +#ifdef DOODAD + membar_producer(); + + lockstat_hot_patch(); + membar_producer(); +#endif +} + +/*ARGSUSED*/ +static void +lockstat_disable(void *arg, dtrace_id_t id, void *parg) +{ + lockstat_probe_t *probe = parg; + int i; + + ASSERT(lockstat_probemap[probe->lsp_probe]); + + lockstat_probemap[probe->lsp_probe] = 0; +#ifdef DOODAD + lockstat_hot_patch(); + membar_producer(); +#endif + + /* + * See if we have any probes left enabled. + */ + for (i = 0; i < LS_NPROBES; i++) { + if (lockstat_probemap[i]) { + /* + * This probe is still enabled. We don't need to deal + * with waiting for all threads to be out of the + * lockstat critical sections; just return. + */ + return; + } + } + +} + +/*ARGSUSED*/ +static int +lockstat_open(struct cdev *dev __unused, int oflags __unused, + int devtype __unused, struct thread *td __unused) +{ + return (0); +} + +/*ARGSUSED*/ +static void +lockstat_provide(void *arg, dtrace_probedesc_t *desc) +{ + int i = 0; + + for (i = 0; lockstat_probes[i].lsp_func != NULL; i++) { + lockstat_probe_t *probe = &lockstat_probes[i]; + + if (dtrace_probe_lookup(lockstat_id, "kernel", + probe->lsp_func, probe->lsp_name) != 0) + continue; + + ASSERT(!probe->lsp_id); + probe->lsp_id = dtrace_probe_create(lockstat_id, + "kernel", probe->lsp_func, probe->lsp_name, + LOCKSTAT_AFRAMES, probe); + } +} + +/*ARGSUSED*/ +static void +lockstat_destroy(void *arg, dtrace_id_t id, void *parg) +{ + lockstat_probe_t *probe = parg; + + ASSERT(!lockstat_probemap[probe->lsp_probe]); + probe->lsp_id = 0; +} + +static dtrace_pattr_t lockstat_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +}; + +static dtrace_pops_t lockstat_pops = { + lockstat_provide, + NULL, + lockstat_enable, + lockstat_disable, + NULL, + NULL, + NULL, + NULL, + NULL, + lockstat_destroy +}; + +static void +lockstat_load(void *dummy) +{ + /* Create the /dev/dtrace/lockstat entry. */ + lockstat_cdev = make_dev(&lockstat_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, + "dtrace/lockstat"); + + if (dtrace_register("lockstat", &lockstat_attr, DTRACE_PRIV_USER, + NULL, &lockstat_pops, NULL, &lockstat_id) != 0) + return; +} + +static int +lockstat_unload() +{ + int error = 0; + + if ((error = dtrace_unregister(lockstat_id)) != 0) + return (error); + + destroy_dev(lockstat_cdev); + + return (error); +} + +/* ARGSUSED */ +static int +lockstat_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + } + return (error); +} + +SYSINIT(lockstat_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, lockstat_load, NULL); +SYSUNINIT(lockstat_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, lockstat_unload, NULL); + +DEV_MODULE(lockstat, lockstat_modevent, NULL); +MODULE_VERSION(lockstat, 1); +MODULE_DEPEND(lockstat, dtrace, 1, 1, 1); +MODULE_DEPEND(lockstat, opensolaris, 1, 1, 1); diff --git a/sys/cddl/dev/profile/profile.c b/sys/cddl/dev/profile/profile.c new file mode 100644 index 000000000000..ea52e2ea3623 --- /dev/null +++ b/sys/cddl/dev/profile/profile.c @@ -0,0 +1,531 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * Portions Copyright 2006-2008 John Birrell jb@freebsd.org + * + * $FreeBSD$ + * + */ + +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#include <sys/cdefs.h> +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/cpuvar.h> +#include <sys/fcntl.h> +#include <sys/filio.h> +#include <sys/kdb.h> +#include <sys/kernel.h> +#include <sys/kmem.h> +#include <sys/kthread.h> +#include <sys/limits.h> +#include <sys/linker.h> +#include <sys/lock.h> +#include <sys/malloc.h> +#include <sys/module.h> +#include <sys/mutex.h> +#include <sys/poll.h> +#include <sys/proc.h> +#include <sys/selinfo.h> +#include <sys/smp.h> +#include <sys/uio.h> +#include <sys/unistd.h> +#include <machine/stdarg.h> + +#include <sys/cyclic.h> +#include <sys/dtrace.h> +#include <sys/dtrace_bsd.h> + +#define PROF_NAMELEN 15 + +#define PROF_PROFILE 0 +#define PROF_TICK 1 +#define PROF_PREFIX_PROFILE "profile-" +#define PROF_PREFIX_TICK "tick-" + +/* + * Regardless of platform, there are five artificial frames in the case of the + * profile provider: + * + * profile_fire + * cyclic_expire + * cyclic_fire + * [ cbe ] + * [ locore ] + * + * On amd64, there are two frames associated with locore: one in locore, and + * another in common interrupt dispatch code. (i386 has not been modified to + * use this common layer.) Further, on i386, the interrupted instruction + * appears as its own stack frame. All of this means that we need to add one + * frame for amd64, and then take one away for both amd64 and i386. + * + * On SPARC, the picture is further complicated because the compiler + * optimizes away tail-calls -- so the following frames are optimized away: + * + * profile_fire + * cyclic_expire + * + * This gives three frames. However, on DEBUG kernels, the cyclic_expire + * frame cannot be tail-call eliminated, yielding four frames in this case. + * + * All of the above constraints lead to the mess below. Yes, the profile + * provider should ideally figure this out on-the-fly by hiting one of its own + * probes and then walking its own stack trace. This is complicated, however, + * and the static definition doesn't seem to be overly brittle. Still, we + * allow for a manual override in case we get it completely wrong. + */ +#ifdef __amd64 +#define PROF_ARTIFICIAL_FRAMES 7 +#else +#ifdef __i386 +#define PROF_ARTIFICIAL_FRAMES 6 +#else +#ifdef __sparc +#ifdef DEBUG +#define PROF_ARTIFICIAL_FRAMES 4 +#else +#define PROF_ARTIFICIAL_FRAMES 3 +#endif +#endif +#endif +#endif + +typedef struct profile_probe { + char prof_name[PROF_NAMELEN]; + dtrace_id_t prof_id; + int prof_kind; + hrtime_t prof_interval; + cyclic_id_t prof_cyclic; +} profile_probe_t; + +typedef struct profile_probe_percpu { + hrtime_t profc_expected; + hrtime_t profc_interval; + profile_probe_t *profc_probe; +} profile_probe_percpu_t; + +static d_open_t profile_open; +static int profile_unload(void); +static void profile_create(hrtime_t, char *, int); +static void profile_destroy(void *, dtrace_id_t, void *); +static void profile_enable(void *, dtrace_id_t, void *); +static void profile_disable(void *, dtrace_id_t, void *); +static void profile_load(void *); +static void profile_provide(void *, dtrace_probedesc_t *); + +static int profile_rates[] = { + 97, 199, 499, 997, 1999, + 4001, 4999, 0, 0, 0, + 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0 +}; + +static int profile_ticks[] = { + 1, 10, 100, 500, 1000, + 5000, 0, 0, 0, 0, + 0, 0, 0, 0, 0 +}; + +/* + * profile_max defines the upper bound on the number of profile probes that + * can exist (this is to prevent malicious or clumsy users from exhausing + * system resources by creating a slew of profile probes). At mod load time, + * this gets its value from PROFILE_MAX_DEFAULT or profile-max-probes if it's + * present in the profile.conf file. + */ +#define PROFILE_MAX_DEFAULT 1000 /* default max. number of probes */ +static uint32_t profile_max = PROFILE_MAX_DEFAULT; + /* maximum number of profile probes */ +static uint32_t profile_total; /* current number of profile probes */ + +static struct cdevsw profile_cdevsw = { + .d_version = D_VERSION, + .d_open = profile_open, + .d_name = "profile", +}; + +static dtrace_pattr_t profile_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pops_t profile_pops = { + profile_provide, + NULL, + profile_enable, + profile_disable, + NULL, + NULL, + NULL, + NULL, + NULL, + profile_destroy +}; + +static struct cdev *profile_cdev; +static dtrace_provider_id_t profile_id; +static hrtime_t profile_interval_min = NANOSEC / 5000; /* 5000 hz */ +static int profile_aframes = 0; /* override */ + +static void +profile_fire(void *arg) +{ + profile_probe_percpu_t *pcpu = arg; + profile_probe_t *prof = pcpu->profc_probe; + hrtime_t late; + solaris_cpu_t *c = &solaris_cpu[curcpu]; + + late = gethrtime() - pcpu->profc_expected; + pcpu->profc_expected += pcpu->profc_interval; + + dtrace_probe(prof->prof_id, c->cpu_profile_pc, + c->cpu_profile_upc, late, 0, 0); +} + +static void +profile_tick(void *arg) +{ + profile_probe_t *prof = arg; + solaris_cpu_t *c = &solaris_cpu[curcpu]; + + dtrace_probe(prof->prof_id, c->cpu_profile_pc, + c->cpu_profile_upc, 0, 0, 0); +} + +static void +profile_create(hrtime_t interval, char *name, int kind) +{ + profile_probe_t *prof; + + if (interval < profile_interval_min) + return; + + if (dtrace_probe_lookup(profile_id, NULL, NULL, name) != 0) + return; + + atomic_add_32(&profile_total, 1); + if (profile_total > profile_max) { + atomic_add_32(&profile_total, -1); + return; + } + + prof = kmem_zalloc(sizeof (profile_probe_t), KM_SLEEP); + (void) strcpy(prof->prof_name, name); + prof->prof_interval = interval; + prof->prof_cyclic = CYCLIC_NONE; + prof->prof_kind = kind; + prof->prof_id = dtrace_probe_create(profile_id, + NULL, NULL, name, + profile_aframes ? profile_aframes : PROF_ARTIFICIAL_FRAMES, prof); +} + +/*ARGSUSED*/ +static void +profile_provide(void *arg, dtrace_probedesc_t *desc) +{ + int i, j, rate, kind; + hrtime_t val = 0, mult = 1, len = 0; + char *name, *suffix = NULL; + + const struct { + char *prefix; + int kind; + } types[] = { + { PROF_PREFIX_PROFILE, PROF_PROFILE }, + { PROF_PREFIX_TICK, PROF_TICK }, + { 0, 0 } + }; + + const struct { + char *name; + hrtime_t mult; + } suffixes[] = { + { "ns", NANOSEC / NANOSEC }, + { "nsec", NANOSEC / NANOSEC }, + { "us", NANOSEC / MICROSEC }, + { "usec", NANOSEC / MICROSEC }, + { "ms", NANOSEC / MILLISEC }, + { "msec", NANOSEC / MILLISEC }, + { "s", NANOSEC / SEC }, + { "sec", NANOSEC / SEC }, + { "m", NANOSEC * (hrtime_t)60 }, + { "min", NANOSEC * (hrtime_t)60 }, + { "h", NANOSEC * (hrtime_t)(60 * 60) }, + { "hour", NANOSEC * (hrtime_t)(60 * 60) }, + { "d", NANOSEC * (hrtime_t)(24 * 60 * 60) }, + { "day", NANOSEC * (hrtime_t)(24 * 60 * 60) }, + { "hz", 0 }, + { NULL } + }; + + if (desc == NULL) { + char n[PROF_NAMELEN]; + + /* + * If no description was provided, provide all of our probes. + */ + for (i = 0; i < sizeof (profile_rates) / sizeof (int); i++) { + if ((rate = profile_rates[i]) == 0) + continue; + + (void) snprintf(n, PROF_NAMELEN, "%s%d", + PROF_PREFIX_PROFILE, rate); + profile_create(NANOSEC / rate, n, PROF_PROFILE); + } + + for (i = 0; i < sizeof (profile_ticks) / sizeof (int); i++) { + if ((rate = profile_ticks[i]) == 0) + continue; + + (void) snprintf(n, PROF_NAMELEN, "%s%d", + PROF_PREFIX_TICK, rate); + profile_create(NANOSEC / rate, n, PROF_TICK); + } + + return; + } + + name = desc->dtpd_name; + + for (i = 0; types[i].prefix != NULL; i++) { + len = strlen(types[i].prefix); + + if (strncmp(name, types[i].prefix, len) != 0) + continue; + break; + } + + if (types[i].prefix == NULL) + return; + + kind = types[i].kind; + j = strlen(name) - len; + + /* + * We need to start before any time suffix. + */ + for (j = strlen(name); j >= len; j--) { + if (name[j] >= '0' && name[j] <= '9') + break; + suffix = &name[j]; + } + + ASSERT(suffix != NULL); + + /* + * Now determine the numerical value present in the probe name. + */ + for (; j >= len; j--) { + if (name[j] < '0' || name[j] > '9') + return; + + val += (name[j] - '0') * mult; + mult *= (hrtime_t)10; + } + + if (val == 0) + return; + + /* + * Look-up the suffix to determine the multiplier. + */ + for (i = 0, mult = 0; suffixes[i].name != NULL; i++) { + if (strcasecmp(suffixes[i].name, suffix) == 0) { + mult = suffixes[i].mult; + break; + } + } + + if (suffixes[i].name == NULL && *suffix != '\0') + return; + + if (mult == 0) { + /* + * The default is frequency-per-second. + */ + val = NANOSEC / val; + } else { + val *= mult; + } + + profile_create(val, name, kind); +} + +/* ARGSUSED */ +static void +profile_destroy(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; + + ASSERT(prof->prof_cyclic == CYCLIC_NONE); + kmem_free(prof, sizeof (profile_probe_t)); + + ASSERT(profile_total >= 1); + atomic_add_32(&profile_total, -1); +} + +/*ARGSUSED*/ +static void +profile_online(void *arg, cpu_t *cpu, cyc_handler_t *hdlr, cyc_time_t *when) +{ + profile_probe_t *prof = arg; + profile_probe_percpu_t *pcpu; + + pcpu = kmem_zalloc(sizeof (profile_probe_percpu_t), KM_SLEEP); + pcpu->profc_probe = prof; + + hdlr->cyh_func = profile_fire; + hdlr->cyh_arg = pcpu; + + when->cyt_interval = prof->prof_interval; + when->cyt_when = gethrtime() + when->cyt_interval; + + pcpu->profc_expected = when->cyt_when; + pcpu->profc_interval = when->cyt_interval; +} + +/*ARGSUSED*/ +static void +profile_offline(void *arg, cpu_t *cpu, void *oarg) +{ + profile_probe_percpu_t *pcpu = oarg; + + ASSERT(pcpu->profc_probe == arg); + kmem_free(pcpu, sizeof (profile_probe_percpu_t)); +} + +/* ARGSUSED */ +static void +profile_enable(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; + cyc_omni_handler_t omni; + cyc_handler_t hdlr; + cyc_time_t when; + + ASSERT(prof->prof_interval != 0); + ASSERT(MUTEX_HELD(&cpu_lock)); + + if (prof->prof_kind == PROF_TICK) { + hdlr.cyh_func = profile_tick; + hdlr.cyh_arg = prof; + + when.cyt_interval = prof->prof_interval; + when.cyt_when = gethrtime() + when.cyt_interval; + } else { + ASSERT(prof->prof_kind == PROF_PROFILE); + omni.cyo_online = profile_online; + omni.cyo_offline = profile_offline; + omni.cyo_arg = prof; + } + + if (prof->prof_kind == PROF_TICK) { + prof->prof_cyclic = cyclic_add(&hdlr, &when); + } else { + prof->prof_cyclic = cyclic_add_omni(&omni); + } +} + +/* ARGSUSED */ +static void +profile_disable(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; + + ASSERT(prof->prof_cyclic != CYCLIC_NONE); + ASSERT(MUTEX_HELD(&cpu_lock)); + + cyclic_remove(prof->prof_cyclic); + prof->prof_cyclic = CYCLIC_NONE; +} + +static void +profile_load(void *dummy) +{ + /* Create the /dev/dtrace/profile entry. */ + profile_cdev = make_dev(&profile_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, + "dtrace/profile"); + + if (dtrace_register("profile", &profile_attr, DTRACE_PRIV_USER, + NULL, &profile_pops, NULL, &profile_id) != 0) + return; +} + + +static int +profile_unload() +{ + int error = 0; + + if ((error = dtrace_unregister(profile_id)) != 0) + return (error); + + destroy_dev(profile_cdev); + + return (error); +} + +/* ARGSUSED */ +static int +profile_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + return (error); +} + +/* ARGSUSED */ +static int +profile_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) +{ + return (0); +} + +SYSINIT(profile_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, profile_load, NULL); +SYSUNINIT(profile_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, profile_unload, NULL); + +DEV_MODULE(profile, profile_modevent, NULL); +MODULE_VERSION(profile, 1); +MODULE_DEPEND(profile, dtrace, 1, 1, 1); +MODULE_DEPEND(profile, cyclic, 1, 1, 1); +MODULE_DEPEND(profile, opensolaris, 1, 1, 1); diff --git a/sys/cddl/dev/prototype.c b/sys/cddl/dev/prototype.c new file mode 100644 index 000000000000..1f7f8b5f0c2d --- /dev/null +++ b/sys/cddl/dev/prototype.c @@ -0,0 +1,144 @@ +/* + * This file is freeware. You are free to use it and add your own + * license. + * + * $FreeBSD$ + * + */ + +#include <sys/cdefs.h> +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/kernel.h> +#include <sys/module.h> + +#include <sys/dtrace.h> + +static d_open_t prototype_open; +static int prototype_unload(void); +static void prototype_getargdesc(void *, dtrace_id_t, void *, dtrace_argdesc_t *); +static void prototype_provide(void *, dtrace_probedesc_t *); +static void prototype_destroy(void *, dtrace_id_t, void *); +static void prototype_enable(void *, dtrace_id_t, void *); +static void prototype_disable(void *, dtrace_id_t, void *); +static void prototype_load(void *); + +static struct cdevsw prototype_cdevsw = { + .d_version = D_VERSION, + .d_open = prototype_open, + .d_name = "prototype", +}; + +static dtrace_pattr_t prototype_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pops_t prototype_pops = { + prototype_provide, + NULL, + prototype_enable, + prototype_disable, + NULL, + NULL, + prototype_getargdesc, + NULL, + NULL, + prototype_destroy +}; + +static struct cdev *prototype_cdev; +static dtrace_provider_id_t prototype_id; + +static void +prototype_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) +{ +} + +static void +prototype_provide(void *arg, dtrace_probedesc_t *desc) +{ +} + +static void +prototype_destroy(void *arg, dtrace_id_t id, void *parg) +{ +} + +static void +prototype_enable(void *arg, dtrace_id_t id, void *parg) +{ +} + +static void +prototype_disable(void *arg, dtrace_id_t id, void *parg) +{ +} + +static void +prototype_load(void *dummy) +{ + /* Create the /dev/dtrace/prototype entry. */ + prototype_cdev = make_dev(&prototype_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, + "dtrace/prototype"); + + if (dtrace_register("prototype", &prototype_attr, DTRACE_PRIV_USER, + NULL, &prototype_pops, NULL, &prototype_id) != 0) + return; +} + + +static int +prototype_unload() +{ + int error = 0; + + if ((error = dtrace_unregister(prototype_id)) != 0) + return (error); + + destroy_dev(prototype_cdev); + + return (error); +} + +static int +prototype_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + + return (error); +} + +static int +prototype_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) +{ + return (0); +} + +SYSINIT(prototype_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, prototype_load, NULL); +SYSUNINIT(prototype_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, prototype_unload, NULL); + +DEV_MODULE(prototype, prototype_modevent, NULL); +MODULE_VERSION(prototype, 1); +MODULE_DEPEND(prototype, dtrace, 1, 1, 1); +MODULE_DEPEND(prototype, opensolaris, 1, 1, 1); diff --git a/sys/cddl/dev/sdt/sdt.c b/sys/cddl/dev/sdt/sdt.c new file mode 100644 index 000000000000..d5d41724593f --- /dev/null +++ b/sys/cddl/dev/sdt/sdt.c @@ -0,0 +1,254 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * Portions Copyright 2006-2008 John Birrell jb@freebsd.org + * + * $FreeBSD$ + * + */ + +#ifndef KDTRACE_HOOKS +#define KDTRACE_HOOKS +#endif + +#include <sys/cdefs.h> +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/kernel.h> +#include <sys/limits.h> +#include <sys/lock.h> +#include <sys/linker.h> +#include <sys/module.h> +#include <sys/mutex.h> + +#include <sys/dtrace.h> +#include <sys/sdt.h> + +#define SDT_ADDR2NDX(addr) (((uintptr_t)(addr)) >> 4) + +static d_open_t sdt_open; +static int sdt_unload(void); +static void sdt_getargdesc(void *, dtrace_id_t, void *, dtrace_argdesc_t *); +static void sdt_provide_probes(void *, dtrace_probedesc_t *); +static void sdt_destroy(void *, dtrace_id_t, void *); +static void sdt_enable(void *, dtrace_id_t, void *); +static void sdt_disable(void *, dtrace_id_t, void *); +static void sdt_load(void *); + +static struct cdevsw sdt_cdevsw = { + .d_version = D_VERSION, + .d_open = sdt_open, + .d_name = "sdt", +}; + +static dtrace_pattr_t sdt_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pops_t sdt_pops = { + sdt_provide_probes, + NULL, + sdt_enable, + sdt_disable, + NULL, + NULL, + sdt_getargdesc, + NULL, + NULL, + sdt_destroy +}; + +static struct cdev *sdt_cdev; + +static int +sdt_argtype_callback(struct sdt_argtype *argtype, void *arg) +{ + dtrace_argdesc_t *desc = arg; + + if (desc->dtargd_ndx == argtype->ndx) { + desc->dtargd_mapping = desc->dtargd_ndx; /* XXX */ + strlcpy(desc->dtargd_native, argtype->type, + sizeof(desc->dtargd_native)); + desc->dtargd_xlate[0] = '\0'; /* XXX */ + } + + return (0); +} + +static void +sdt_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) +{ + struct sdt_probe *probe = parg; + + if (desc->dtargd_ndx < probe->n_args) + (void) (sdt_argtype_listall(probe, sdt_argtype_callback, desc)); + else + desc->dtargd_ndx = DTRACE_ARGNONE; + + return; +} + +static int +sdt_probe_callback(struct sdt_probe *probe, void *arg __unused) +{ + struct sdt_provider *prov = probe->prov; + char mod[64]; + char func[64]; + char name[64]; + + /* + * Unfortunately this is necessary because the Solaris DTrace + * code mixes consts and non-consts with casts to override + * the incompatibilies. On FreeBSD, we use strict warnings + * in gcc, so we have to respect const vs non-const. + */ + strlcpy(mod, probe->mod, sizeof(mod)); + strlcpy(func, probe->func, sizeof(func)); + strlcpy(name, probe->name, sizeof(name)); + + if (dtrace_probe_lookup(prov->id, mod, func, name) != 0) + return (0); + + (void) dtrace_probe_create(prov->id, probe->mod, probe->func, + probe->name, 0, probe); + + return (0); +} + +static int +sdt_provider_entry(struct sdt_provider *prov, void *arg) +{ + return (sdt_probe_listall(prov, sdt_probe_callback, NULL)); +} + +static void +sdt_provide_probes(void *arg, dtrace_probedesc_t *desc) +{ + if (desc != NULL) + return; + + (void) sdt_provider_listall(sdt_provider_entry, NULL); +} + +static void +sdt_destroy(void *arg, dtrace_id_t id, void *parg) +{ + /* Nothing to do here. */ +} + +static void +sdt_enable(void *arg, dtrace_id_t id, void *parg) +{ + struct sdt_probe *probe = parg; + + probe->id = id; +} + +static void +sdt_disable(void *arg, dtrace_id_t id, void *parg) +{ + struct sdt_probe *probe = parg; + + probe->id = 0; +} + +static int +sdt_provider_reg_callback(struct sdt_provider *prov, void *arg __unused) +{ + return (dtrace_register(prov->name, &sdt_attr, DTRACE_PRIV_USER, + NULL, &sdt_pops, NULL, (dtrace_provider_id_t *) &prov->id)); +} + +static void +sdt_load(void *dummy) +{ + /* Create the /dev/dtrace/sdt entry. */ + sdt_cdev = make_dev(&sdt_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, + "dtrace/sdt"); + + sdt_probe_func = dtrace_probe; + + (void) sdt_provider_listall(sdt_provider_reg_callback, NULL); +} + +static int +sdt_provider_unreg_callback(struct sdt_provider *prov, void *arg __unused) +{ + return (dtrace_unregister(prov->id)); +} + +static int +sdt_unload() +{ + int error = 0; + + sdt_probe_func = sdt_probe_stub; + + (void) sdt_provider_listall(sdt_provider_unreg_callback, NULL); + + destroy_dev(sdt_cdev); + + return (error); +} + +/* ARGSUSED */ +static int +sdt_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + + return (error); +} + +/* ARGSUSED */ +static int +sdt_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) +{ + return (0); +} + +SYSINIT(sdt_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, sdt_load, NULL); +SYSUNINIT(sdt_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, sdt_unload, NULL); + +DEV_MODULE(sdt, sdt_modevent, NULL); +MODULE_VERSION(sdt, 1); +MODULE_DEPEND(sdt, dtrace, 1, 1, 1); +MODULE_DEPEND(sdt, opensolaris, 1, 1, 1); diff --git a/sys/cddl/dev/systrace/systrace.c b/sys/cddl/dev/systrace/systrace.c new file mode 100644 index 000000000000..e54f776576d5 --- /dev/null +++ b/sys/cddl/dev/systrace/systrace.c @@ -0,0 +1,341 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + * + * Portions Copyright 2006-2008 John Birrell jb@freebsd.org + * + * $FreeBSD$ + * + */ + +/* + * Copyright 2006 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#include <sys/cdefs.h> +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/cpuvar.h> +#include <sys/fcntl.h> +#include <sys/filio.h> +#include <sys/kdb.h> +#include <sys/kernel.h> +#include <sys/kmem.h> +#include <sys/kthread.h> +#include <sys/limits.h> +#include <sys/linker.h> +#include <sys/lock.h> +#include <sys/malloc.h> +#include <sys/module.h> +#include <sys/mutex.h> +#include <sys/poll.h> +#include <sys/proc.h> +#include <sys/selinfo.h> +#include <sys/smp.h> +#include <sys/sysproto.h> +#include <sys/sysent.h> +#include <sys/uio.h> +#include <sys/unistd.h> +#include <machine/stdarg.h> + +#include <sys/dtrace.h> + +#ifdef LINUX_SYSTRACE +#include <linux.h> +#include <linux_syscall.h> +#include <linux_proto.h> +#include <linux_syscallnames.c> +#include <linux_systrace.c> +extern struct sysent linux_sysent[]; +#define DEVNAME "dtrace/linsystrace" +#define PROVNAME "linsyscall" +#define MAXSYSCALL LINUX_SYS_MAXSYSCALL +#define SYSCALLNAMES linux_syscallnames +#define SYSENT linux_sysent +#else +/* + * The syscall arguments are processed into a DTrace argument array + * using a generated function. See sys/kern/makesyscalls.sh. + */ +#include <sys/syscall.h> +#include <kern/systrace_args.c> +extern const char *syscallnames[]; +#define DEVNAME "dtrace/systrace" +#define PROVNAME "syscall" +#define MAXSYSCALL SYS_MAXSYSCALL +#define SYSCALLNAMES syscallnames +#define SYSENT sysent +#endif + +#define SYSTRACE_ARTIFICIAL_FRAMES 1 + +#define SYSTRACE_SHIFT 16 +#define SYSTRACE_ISENTRY(x) ((int)(x) >> SYSTRACE_SHIFT) +#define SYSTRACE_SYSNUM(x) ((int)(x) & ((1 << SYSTRACE_SHIFT) - 1)) +#define SYSTRACE_ENTRY(id) ((1 << SYSTRACE_SHIFT) | (id)) +#define SYSTRACE_RETURN(id) (id) + +#if ((1 << SYSTRACE_SHIFT) <= MAXSYSCALL) +#error 1 << SYSTRACE_SHIFT must exceed number of system calls +#endif + +static d_open_t systrace_open; +static int systrace_unload(void); +static void systrace_getargdesc(void *, dtrace_id_t, void *, dtrace_argdesc_t *); +static void systrace_provide(void *, dtrace_probedesc_t *); +static void systrace_destroy(void *, dtrace_id_t, void *); +static void systrace_enable(void *, dtrace_id_t, void *); +static void systrace_disable(void *, dtrace_id_t, void *); +static void systrace_load(void *); + +static struct cdevsw systrace_cdevsw = { + .d_version = D_VERSION, + .d_open = systrace_open, +#ifdef LINUX_SYSTRACE + .d_name = "linsystrace", +#else + .d_name = "systrace", +#endif +}; + +static union { + const char **p_constnames; + char **pp_syscallnames; +} uglyhack = { SYSCALLNAMES }; + +static dtrace_pattr_t systrace_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pops_t systrace_pops = { + systrace_provide, + NULL, + systrace_enable, + systrace_disable, + NULL, + NULL, + systrace_getargdesc, + NULL, + NULL, + systrace_destroy +}; + +static struct cdev *systrace_cdev; +static dtrace_provider_id_t systrace_id; + +#if !defined(LINUX_SYSTRACE) +/* + * Probe callback function. + * + * Note: This function is called for _all_ syscalls, regardless of which sysent + * array the syscall comes from. It could be a standard syscall or a + * compat syscall from something like Linux. + */ +static void +systrace_probe(u_int32_t id, int sysnum, struct sysent *sysent, void *params) +{ + int n_args = 0; + u_int64_t uargs[8]; + + /* + * Check if this syscall has an argument conversion function + * registered. + */ + if (sysent->sy_systrace_args_func != NULL) + /* + * Convert the syscall parameters using the registered + * function. + */ + (*sysent->sy_systrace_args_func)(sysnum, params, uargs, &n_args); + else + /* + * Use the built-in system call argument conversion + * function to translate the syscall structure fields + * into the array of 64-bit values that DTrace + * expects. + */ + systrace_args(sysnum, params, uargs, &n_args); + + /* Process the probe using the converted argments. */ + dtrace_probe(id, uargs[0], uargs[1], uargs[2], uargs[3], uargs[4]); +} +#endif + +static void +systrace_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + + systrace_setargdesc(sysnum, desc->dtargd_ndx, desc->dtargd_native, + sizeof(desc->dtargd_native)); + + if (desc->dtargd_native[0] == '\0') + desc->dtargd_ndx = DTRACE_ARGNONE; + + return; +} + +static void +systrace_provide(void *arg, dtrace_probedesc_t *desc) +{ + int i; + + if (desc != NULL) + return; + + for (i = 0; i < MAXSYSCALL; i++) { + if (dtrace_probe_lookup(systrace_id, NULL, + uglyhack.pp_syscallnames[i], "entry") != 0) + continue; + + (void) dtrace_probe_create(systrace_id, NULL, uglyhack.pp_syscallnames[i], + "entry", SYSTRACE_ARTIFICIAL_FRAMES, + (void *)((uintptr_t)SYSTRACE_ENTRY(i))); + (void) dtrace_probe_create(systrace_id, NULL, uglyhack.pp_syscallnames[i], + "return", SYSTRACE_ARTIFICIAL_FRAMES, + (void *)((uintptr_t)SYSTRACE_RETURN(i))); + } +} + +static void +systrace_destroy(void *arg, dtrace_id_t id, void *parg) +{ +#ifdef DEBUG + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + + /* + * There's nothing to do here but assert that we have actually been + * disabled. + */ + if (SYSTRACE_ISENTRY((uintptr_t)parg)) { + ASSERT(sysent[sysnum].sy_entry == 0); + } else { + ASSERT(sysent[sysnum].sy_return == 0); + } +#endif +} + +static void +systrace_enable(void *arg, dtrace_id_t id, void *parg) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + + if (SYSENT[sysnum].sy_systrace_args_func == NULL) + SYSENT[sysnum].sy_systrace_args_func = systrace_args; + + if (SYSTRACE_ISENTRY((uintptr_t)parg)) + SYSENT[sysnum].sy_entry = id; + else + SYSENT[sysnum].sy_return = id; +} + +static void +systrace_disable(void *arg, dtrace_id_t id, void *parg) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + + SYSENT[sysnum].sy_entry = 0; + SYSENT[sysnum].sy_return = 0; +} + +static void +systrace_load(void *dummy) +{ + /* Create the /dev/dtrace/systrace entry. */ + systrace_cdev = make_dev(&systrace_cdevsw, 0, UID_ROOT, GID_WHEEL, 0600, + DEVNAME); + + if (dtrace_register(PROVNAME, &systrace_attr, DTRACE_PRIV_USER, + NULL, &systrace_pops, NULL, &systrace_id) != 0) + return; + +#if !defined(LINUX_SYSTRACE) + systrace_probe_func = systrace_probe; +#endif +} + + +static int +systrace_unload() +{ + int error = 0; + + if ((error = dtrace_unregister(systrace_id)) != 0) + return (error); + +#if !defined(LINUX_SYSTRACE) + systrace_probe_func = NULL; +#endif + + destroy_dev(systrace_cdev); + + return (error); +} + +static int +systrace_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + return (error); +} + +static int +systrace_open(struct cdev *dev __unused, int oflags __unused, int devtype __unused, struct thread *td __unused) +{ + return (0); +} + +SYSINIT(systrace_load, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_load, NULL); +SYSUNINIT(systrace_unload, SI_SUB_DTRACE_PROVIDER, SI_ORDER_ANY, systrace_unload, NULL); + +#ifdef LINUX_SYSTRACE +DEV_MODULE(linsystrace, systrace_modevent, NULL); +MODULE_VERSION(linsystrace, 1); +MODULE_DEPEND(linsystrace, linux, 1, 1, 1); +MODULE_DEPEND(linsystrace, systrace, 1, 1, 1); +MODULE_DEPEND(linsystrace, dtrace, 1, 1, 1); +MODULE_DEPEND(linsystrace, opensolaris, 1, 1, 1); +#else +DEV_MODULE(systrace, systrace_modevent, NULL); +MODULE_VERSION(systrace, 1); +MODULE_DEPEND(systrace, dtrace, 1, 1, 1); +MODULE_DEPEND(systrace, opensolaris, 1, 1, 1); +#endif diff --git a/sys/conf/Makefile.amd64 b/sys/conf/Makefile.amd64 index 1032a51ac334..6c737eb2b4e7 100644 --- a/sys/conf/Makefile.amd64 +++ b/sys/conf/Makefile.amd64 @@ -32,7 +32,8 @@ S= ../../.. .include "$S/conf/kern.pre.mk" DDB_ENABLED!= grep DDB opt_ddb.h || true -.if !empty(DDB_ENABLED) +DTR_ENABLED!= grep KDTRACE_FRAME opt_kdtrace.h || true +.if !empty(DDB_ENABLED) || !empty(DTR_ENABLED) CFLAGS+= -fno-omit-frame-pointer .endif diff --git a/sys/conf/NOTES b/sys/conf/NOTES index 80c4d0c794b7..f24ae326cad0 100644 --- a/sys/conf/NOTES +++ b/sys/conf/NOTES @@ -356,6 +356,12 @@ options DDB_NUMSYM options GDB # +# Enable the kernel DTrace hooks which are required to load the DTrace +# kernel modules. +# +options KDTRACE_HOOKS + +# # SYSCTL_DEBUG enables a 'sysctl' debug tree that can be used to dump the # contents of the registered sysctl nodes on the console. It is disabled by # default because it generates excessively verbose console output that can diff --git a/sys/conf/files b/sys/conf/files index e2111bb11101..5e53ab8bb03a 100644 --- a/sys/conf/files +++ b/sys/conf/files @@ -1443,6 +1443,7 @@ kern/kern_cpu.c standard kern/kern_cpuset.c standard kern/kern_context.c standard kern/kern_descrip.c standard +kern/kern_dtrace.c optional kdtrace_hooks kern/kern_environment.c standard kern/kern_event.c standard kern/kern_exec.c standard @@ -1526,6 +1527,7 @@ kern/subr_rman.c standard kern/subr_rtc.c optional genclock kern/subr_sbuf.c standard kern/subr_scanf.c standard +kern/kern_sdt.c optional kdtrace_hooks kern/subr_sleepqueue.c standard kern/subr_smp.c standard kern/subr_stack.c optional ddb | stack @@ -1538,7 +1540,7 @@ kern/sys_generic.c standard kern/sys_pipe.c standard kern/sys_process.c standard kern/sys_socket.c standard -kern/syscalls.c optional witness | invariants +kern/syscalls.c optional witness | invariants | kdtrace_hooks kern/sysv_ipc.c standard kern/sysv_msg.c optional sysvmsg kern/sysv_sem.c optional sysvsem @@ -1671,7 +1673,8 @@ net/rtsock.c standard net/slcompress.c optional netgraph_vjc | ppp | sl | sppp | \ netgraph_sppp net/zlib.c optional crypto | geom_uzip | ipsec | \ - mxge | ppp_deflate | netgraph_deflate + mxge | ppp_deflate | netgraph_deflate | \ + ddb_ctf net80211/ieee80211.c optional wlan net80211/ieee80211_acl.c optional wlan_acl net80211/ieee80211_amrr.c optional wlan_amrr diff --git a/sys/conf/kern.post.mk b/sys/conf/kern.post.mk index 1eca596c1694..f712776056f2 100644 --- a/sys/conf/kern.post.mk +++ b/sys/conf/kern.post.mk @@ -86,6 +86,9 @@ ${FULLKERNEL}: ${SYSTEM_DEP} vers.o @rm -f ${.TARGET} @echo linking ${.TARGET} ${SYSTEM_LD} +.if defined(CTFMERGE) + ${SYSTEM_CTFMERGE} +.endif .if !defined(DEBUG) ${OBJCOPY} --strip-debug ${.TARGET} .endif diff --git a/sys/conf/kern.pre.mk b/sys/conf/kern.pre.mk index a5139f14afe4..79ec5b8196c5 100644 --- a/sys/conf/kern.pre.mk +++ b/sys/conf/kern.pre.mk @@ -31,6 +31,8 @@ _MINUS_O= -O2 . endif . if ${MACHINE_ARCH} == "amd64" COPTFLAGS?=-O2 -frename-registers -pipe +. elif ${MACHINE_ARCH} == "sparc64" +COPTFLAGS?=-pipe . else COPTFLAGS?=${_MINUS_O} -pipe . endif @@ -79,6 +81,9 @@ INCLUDES+= -I$S/dev/twa # ... and XFS INCLUDES+= -I$S/gnu/fs/xfs/FreeBSD -I$S/gnu/fs/xfs/FreeBSD/support -I$S/gnu/fs/xfs +# ... and OpenSolaris +INCLUDES+= -I$S/contrib/opensolaris/compat + .endif CFLAGS= ${COPTFLAGS} ${C_DIALECT} ${DEBUG} ${CWARNFLAGS} @@ -126,6 +131,12 @@ NORMAL_C_NOWERROR= ${CC} -c ${CFLAGS} ${PROF} ${.IMPSRC} NORMAL_M= ${AWK} -f $S/tools/makeobjops.awk ${.IMPSRC} -c ; \ ${CC} -c ${CFLAGS} ${WERROR} ${PROF} ${.PREFIX}.c +.if defined(CTFCONVERT) +NORMAL_CTFCONVERT= ${CTFCONVERT} ${CTFFLAGS} ${.TARGET} +.else +NORMAL_CTFCONVERT= +.endif + NORMAL_LINT= ${LINT} ${LINTFLAGS} ${CFLAGS:M-[DIU]*} ${.IMPSRC} GEN_CFILES= $S/$M/$M/genassym.c ${MFILES:T:S/.m$/.c/} @@ -134,6 +145,10 @@ SYSTEM_DEP= Makefile ${SYSTEM_OBJS} SYSTEM_OBJS= locore.o ${MDOBJS} ${OBJS} SYSTEM_OBJS+= ${SYSTEM_CFILES:.c=.o} SYSTEM_OBJS+= hack.So +.if defined(CTFMERGE) +SYSTEM_CTFMERGE= ${CTFMERGE} ${CTFFLAGS} -o ${.TARGET} ${SYSTEM_OBJS} vers.o +LD+= -g +.endif SYSTEM_LD= @${LD} -Bdynamic -T $S/conf/ldscript.$M \ -warn-common -export-dynamic -dynamic-linker /red/herring \ -o ${.TARGET} -X ${SYSTEM_OBJS} vers.o diff --git a/sys/conf/options b/sys/conf/options index 691a93162558..1c0af8214866 100644 --- a/sys/conf/options +++ b/sys/conf/options @@ -50,6 +50,7 @@ TWA_FLASH_FIRMWARE opt_twa.h DDB DDB_CAPTURE_DEFAULTBUFSIZE opt_ddb.h DDB_CAPTURE_MAXBUFSIZE opt_ddb.h +DDB_CTF opt_ddb.h DDB_NUMSYM opt_ddb.h GDB KDB opt_global.h @@ -104,6 +105,7 @@ GEOM_UZIP opt_geom.h GEOM_VIRSTOR opt_geom.h GEOM_VOL opt_geom.h GEOM_ZERO opt_geom.h +KDTRACE_HOOKS opt_kdtrace.h KSE opt_global.h KSTACK_MAX_PAGES KSTACK_PAGES diff --git a/sys/conf/options.amd64 b/sys/conf/options.amd64 index 79bc8162369f..2ea539d79ee8 100644 --- a/sys/conf/options.amd64 +++ b/sys/conf/options.amd64 @@ -64,6 +64,7 @@ DEV_ATPIC opt_atpic.h # Debugging STOP_NMI opt_cpu.h +KDTRACE_FRAME opt_kdtrace.h # BPF just-in-time compiler BPF_JITTER opt_bpf.h diff --git a/sys/i386/conf/GENERIC b/sys/i386/conf/GENERIC index edaa02ecb168..4e62744bb5bb 100644 --- a/sys/i386/conf/GENERIC +++ b/sys/i386/conf/GENERIC @@ -64,6 +64,9 @@ options KBD_INSTALL_CDEV # install a CDEV entry in /dev options ADAPTIVE_GIANT # Giant mutex is adaptive. options STOP_NMI # Stop CPUS using NMI instead of IPI options AUDIT # Security event auditing +options PRINTF_BUFR_SIZE=128 # Prevent printf output being interspersed. +options KDTRACE_HOOKS # DTrace hooks required to load DTrace modules +options DDB_CTF # Load CTF data too. # To make an SMP kernel, the next two lines are needed options SMP # Symmetric MultiProcessor Kernel diff --git a/sys/i386/i386/exception.s b/sys/i386/i386/exception.s index 259327d67207..0ff425e7fed8 100644 --- a/sys/i386/i386/exception.s +++ b/sys/i386/i386/exception.s @@ -31,6 +31,7 @@ */ #include "opt_apic.h" +#include "opt_kdtrace.h" #include "opt_npx.h" #include <machine/asmacros.h> @@ -40,7 +41,24 @@ #include "assym.s" #define SEL_RPL_MASK 0x0003 +#define GSEL_KPL 0x0020 /* GSEL(GCODE_SEL, SEL_KPL) */ + +#ifdef KDTRACE_HOOKS + .bss + .globl dtrace_invop_jump_addr + .align 4 + .type dtrace_invop_jump_addr, @object + .size dtrace_invop_jump_addr, 4 +dtrace_invop_jump_addr: + .zero 4 + .globl dtrace_invop_calltrap_addr + .align 4 + .type dtrace_invop_calltrap_addr, @object + .size dtrace_invop_calltrap_addr, 4 +dtrace_invop_calltrap_addr: + .zero 8 +#endif .text /*****************************************************************************/ @@ -88,8 +106,10 @@ IDTVEC(ofl) pushl $0; TRAP(T_OFLOW) IDTVEC(bnd) pushl $0; TRAP(T_BOUND) +#ifndef KDTRACE_HOOKS IDTVEC(ill) pushl $0; TRAP(T_PRIVINFLT) +#endif IDTVEC(dna) pushl $0; TRAP(T_DNA) IDTVEC(fpusegm) @@ -146,6 +166,45 @@ calltrap: jmp doreti /* + * Privileged instruction fault. + */ +#ifdef KDTRACE_HOOKS + SUPERALIGN_TEXT +IDTVEC(ill) + /* Check if there is no DTrace hook registered. */ + cmpl $0,dtrace_invop_jump_addr + je norm_ill + + /* Check if this is a user fault. */ + cmpl $GSEL_KPL, 4(%esp) /* Check the code segment. */ + + /* If so, just handle it as a normal trap. */ + jne norm_ill + + /* + * This is a kernel instruction fault that might have been caused + * by a DTrace provider. + */ + pushal /* Push all registers onto the stack. */ + + /* + * Set our jump address for the jump back in the event that + * the exception wasn't caused by DTrace at all. + */ + movl $norm_ill, dtrace_invop_calltrap_addr + + /* Jump to the code hooked in by DTrace. */ + jmpl *dtrace_invop_jump_addr + + /* + * Process the instruction fault in the normal way. + */ +norm_ill: + pushl $0 + TRAP(T_PRIVINFLT) +#endif + +/* * SYSCALL CALL GATE (old entry point for a.out binaries) * * The intersegment call has been set up to specify one dummy parameter. diff --git a/sys/i386/i386/local_apic.c b/sys/i386/i386/local_apic.c index f35be2fbb8bb..72cfd1713462 100644 --- a/sys/i386/i386/local_apic.c +++ b/sys/i386/i386/local_apic.c @@ -35,6 +35,7 @@ __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" +#include "opt_kdtrace.h" #include "opt_ddb.h" @@ -65,6 +66,11 @@ __FBSDID("$FreeBSD$"); #include <ddb/ddb.h> #endif +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> +cyclic_clock_func_t lapic_cyclic_clock_func[MAXCPU]; +#endif + /* Sanity checks on IDT vectors. */ CTASSERT(APIC_IO_INTS + APIC_NUM_IOINTS == APIC_TIMER_INT); CTASSERT(APIC_TIMER_INT < APIC_LOCAL_INTS); @@ -670,6 +676,17 @@ lapic_handle_timer(struct trapframe *frame) (*la->la_timer_count)++; critical_enter(); +#ifdef KDTRACE_HOOKS + /* + * If the DTrace hooks are configured and a callback function + * has been registered, then call it to process the high speed + * timers. + */ + int cpu = PCPU_GET(cpuid); + if (lapic_cyclic_clock_func[cpu] != NULL) + (*lapic_cyclic_clock_func[cpu])(frame); +#endif + /* Fire hardclock at hz. */ la->la_hard_ticks += hz; if (la->la_hard_ticks >= lapic_timer_hz) { diff --git a/sys/i386/i386/trap.c b/sys/i386/i386/trap.c index 7be4d499813c..99a9d4c0db9a 100644 --- a/sys/i386/i386/trap.c +++ b/sys/i386/i386/trap.c @@ -49,6 +49,7 @@ __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" #include "opt_isa.h" #include "opt_kdb.h" +#include "opt_kdtrace.h" #include "opt_ktrace.h" #include "opt_npx.h" #include "opt_trap.h" @@ -102,6 +103,26 @@ __FBSDID("$FreeBSD$"); #include <machine/clock.h> #endif +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> + +/* + * This is a hook which is initialised by the dtrace module + * to handle traps which might occur during DTrace probe + * execution. + */ +dtrace_trap_func_t dtrace_trap_func; + +dtrace_doubletrap_func_t dtrace_doubletrap_func; + +/* + * This is a hook which is initialised by the systrace module + * when it is loaded. This keeps the DTrace syscall provider + * implementation opaque. + */ +systrace_probe_func_t systrace_probe_func; +#endif + extern void trap(struct trapframe *frame); extern void syscall(struct trapframe *frame); @@ -219,6 +240,25 @@ trap(struct trapframe *frame) goto out; #endif +#ifdef KDTRACE_HOOKS + /* + * A trap can occur while DTrace executes a probe. Before + * executing the probe, DTrace blocks re-scheduling and sets + * a flag in it's per-cpu flags to indicate that it doesn't + * want to fault. On returning from the the probe, the no-fault + * flag is cleared and finally re-scheduling is enabled. + * + * If the DTrace kernel module has registered a trap handler, + * call it and if it returns non-zero, assume that it has + * handled the trap and modified the trap frame so that this + * function can return normally. + */ + if ((type == T_PROTFLT || type == T_PAGEFLT) && + dtrace_trap_func != NULL) + if ((*dtrace_trap_func)(frame, type)) + goto out; +#endif + if ((frame->tf_eflags & PSL_I) == 0) { /* * Buggy application or kernel code has disabled @@ -916,6 +956,10 @@ trap_fatal(frame, eva) void dblfault_handler() { +#ifdef KDTRACE_HOOKS + if (dtrace_doubletrap_func != NULL) + (*dtrace_doubletrap_func)(); +#endif printf("\nFatal double fault:\n"); printf("eip = 0x%x\n", PCPU_GET(common_tss.tss_eip)); printf("esp = 0x%x\n", PCPU_GET(common_tss.tss_esp)); @@ -1031,9 +1075,34 @@ syscall(struct trapframe *frame) PTRACESTOP_SC(p, td, S_PT_SCE); +#ifdef KDTRACE_HOOKS + /* + * If the systrace module has registered it's probe + * callback and if there is a probe active for the + * syscall 'entry', process the probe. + */ + if (systrace_probe_func != NULL && callp->sy_entry != 0) + (*systrace_probe_func)(callp->sy_entry, code, callp, + args); +#endif + AUDIT_SYSCALL_ENTER(code, td); error = (*callp->sy_call)(td, args); AUDIT_SYSCALL_EXIT(error, td); + + /* Save the latest error return value. */ + td->td_errno = error; + +#ifdef KDTRACE_HOOKS + /* + * If the systrace module has registered it's probe + * callback and if there is a probe active for the + * syscall 'return', process the probe. + */ + if (systrace_probe_func != NULL && callp->sy_return != 0) + (*systrace_probe_func)(callp->sy_return, code, callp, + args); +#endif } switch (error) { diff --git a/sys/i386/include/param.h b/sys/i386/include/param.h index 0f67c271a7a1..09c0855f9fdc 100644 --- a/sys/i386/include/param.h +++ b/sys/i386/include/param.h @@ -65,7 +65,7 @@ #endif #define MID_MACHINE MID_I386 -#ifdef SMP +#if defined(SMP) || defined(KLD_MODULE) #define MAXCPU 16 #else #define MAXCPU 1 diff --git a/sys/i386/linux/Makefile b/sys/i386/linux/Makefile index bd6b32a7d4a8..46d5a319b7c8 100644 --- a/sys/i386/linux/Makefile +++ b/sys/i386/linux/Makefile @@ -12,4 +12,6 @@ linux_sysent.c linux_syscall.h linux_proto.h: ../../kern/makesyscalls.sh \ -mv -f linux_sysent.c linux_sysent.c.bak -mv -f linux_syscall.h linux_syscall.h.bak -mv -f linux_proto.h linux_proto.h.bak + -mv -f linux_systrace.c linux_systrace.c.bak + -mv -f linux_syscallnames.c linux_syscallnames.c.bak sh ../../kern/makesyscalls.sh syscalls.master syscalls.conf diff --git a/sys/i386/linux/syscalls.conf b/sys/i386/linux/syscalls.conf index 87dfbfe7589a..1bc02bafd3c8 100644 --- a/sys/i386/linux/syscalls.conf +++ b/sys/i386/linux/syscalls.conf @@ -1,5 +1,5 @@ # $FreeBSD$ -sysnames="/dev/null" +sysnames="linux_syscallnames.c" sysproto="linux_proto.h" sysproto_h=_LINUX_SYSPROTO_H_ syshdr="linux_syscall.h" @@ -8,4 +8,4 @@ sysmk="/dev/null" syscallprefix="LINUX_SYS_" switchname="linux_sysent" namesname="linux_syscallnames" -systrace="/dev/null" +systrace="linux_systrace.c" diff --git a/sys/ia64/include/param.h b/sys/ia64/include/param.h index c71c6e5a7bf3..cfc198cf99f8 100644 --- a/sys/ia64/include/param.h +++ b/sys/ia64/include/param.h @@ -78,8 +78,8 @@ #define MACHINE_ARCH "ia64" #endif -#ifdef SMP -#define MAXCPU 4 +#if defined(SMP) || defined(KLD_MODULE) +#define MAXCPU 4 #else #define MAXCPU 1 #endif diff --git a/sys/kern/kern_ctf.c b/sys/kern/kern_ctf.c new file mode 100644 index 000000000000..254b8204df39 --- /dev/null +++ b/sys/kern/kern_ctf.c @@ -0,0 +1,325 @@ +/*- + * Copyright (c) 2008 John Birrell <jb@freebsd.org> + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + */ + +/* + * Note this file is included by both link_elf.c and link_elf_obj.c. + * + * The CTF header structure definition can't be used here because it's + * (annoyingly) covered by the CDDL. We will just use a few bytes from + * it as an integer array where we 'know' what they mean. + */ +#define CTF_HDR_SIZE 36 +#define CTF_HDR_STRTAB_U32 7 +#define CTF_HDR_STRLEN_U32 8 + +#ifdef DDB_CTF +static void * +z_alloc(void *nil, u_int items, u_int size) +{ + void *ptr; + + ptr = malloc(items * size, M_TEMP, M_NOWAIT); + return ptr; +} + +static void +z_free(void *nil, void *ptr) +{ + free(ptr, M_TEMP); +} + +#endif + +static int +link_elf_ctf_get(linker_file_t lf, linker_ctf_t *lc) +{ +#ifdef DDB_CTF + Elf_Ehdr *hdr = NULL; + Elf_Shdr *shdr = NULL; + caddr_t ctftab = NULL; + caddr_t raw = NULL; + caddr_t shstrtab = NULL; + elf_file_t ef = (elf_file_t) lf; + int flags; + int i; + int nbytes; + int resid; + int vfslocked; + size_t sz; + struct nameidata nd; + struct thread *td = curthread; + uint8_t ctf_hdr[CTF_HDR_SIZE]; +#endif + int error = 0; + + if (lf == NULL || lc == NULL) + return (EINVAL); + + /* Set the defaults for no CTF present. That's not a crime! */ + bzero(lc, sizeof(*lc)); + +#ifdef DDB_CTF + /* + * First check if we've tried to load CTF data previously and the + * CTF ELF section wasn't found. We flag that condition by setting + * ctfcnt to -1. See below. + */ + if (ef->ctfcnt < 0) + return (0); + + /* Now check if we've already loaded the CTF data.. */ + if (ef->ctfcnt > 0) { + /* We only need to load once. */ + lc->ctftab = ef->ctftab; + lc->ctfcnt = ef->ctfcnt; + lc->symtab = ef->ddbsymtab; + lc->strtab = ef->ddbstrtab; + lc->strcnt = ef->ddbstrcnt; + lc->nsym = ef->ddbsymcnt; + lc->ctfoffp = (uint32_t **) &ef->ctfoff; + lc->typoffp = (uint32_t **) &ef->typoff; + lc->typlenp = &ef->typlen; + return (0); + } + + /* + * We need to try reading the CTF data. Flag no CTF data present + * by default and if we actually succeed in reading it, we'll + * update ctfcnt to the number of bytes read. + */ + ef->ctfcnt = -1; + + NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_SYSSPACE, lf->pathname, td); + flags = FREAD; + error = vn_open(&nd, &flags, 0, NULL); + if (error) + return (error); + vfslocked = NDHASGIANT(&nd); + NDFREE(&nd, NDF_ONLY_PNBUF); + + /* Allocate memory for the FLF header. */ + if ((hdr = malloc(sizeof(*hdr), M_LINKER, M_WAITOK)) == NULL) { + error = ENOMEM; + goto out; + } + + /* Read the ELF header. */ + if ((error = vn_rdwr(UIO_READ, nd.ni_vp, hdr, sizeof(*hdr), + 0, UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED, &resid, + td)) != 0) + goto out; + + /* Sanity check. */ + if (!IS_ELF(*hdr)) { + error = ENOEXEC; + goto out; + } + + nbytes = hdr->e_shnum * hdr->e_shentsize; + if (nbytes == 0 || hdr->e_shoff == 0 || + hdr->e_shentsize != sizeof(Elf_Shdr)) { + error = ENOEXEC; + goto out; + } + + /* Allocate memory for all the section headers */ + if ((shdr = malloc(nbytes, M_LINKER, M_WAITOK)) == NULL) { + error = ENOMEM; + goto out; + } + + /* Read all the section headers */ + if ((error = vn_rdwr(UIO_READ, nd.ni_vp, (caddr_t)shdr, nbytes, + hdr->e_shoff, UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED, + &resid, td)) != 0) + goto out; + + /* + * We need to search for the CTF section by name, so if the + * section names aren't present, then we can't locate the + * .SUNW_ctf section containing the CTF data. + */ + if (hdr->e_shstrndx == 0 || shdr[hdr->e_shstrndx].sh_type != SHT_STRTAB) + goto out; + + /* Allocate memory to buffer the section header strings. */ + if ((shstrtab = malloc(shdr[hdr->e_shstrndx].sh_size, M_LINKER, + M_WAITOK)) == NULL) { + error = ENOMEM; + goto out; + } + + /* Read the section header strings. */ + if ((error = vn_rdwr(UIO_READ, nd.ni_vp, shstrtab, + shdr[hdr->e_shstrndx].sh_size, shdr[hdr->e_shstrndx].sh_offset, + UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, NOCRED, &resid, + td)) != 0) + goto out; + + /* Search for the section containing the CTF data. */ + for (i = 0; i < hdr->e_shnum; i++) + if (strcmp(".SUNW_ctf", shstrtab + shdr[i].sh_name) == 0) + break; + + /* Check if the CTF section wasn't found. */ + if (i >= hdr->e_shnum) + goto out; + + /* Read the CTF header. */ + if ((error = vn_rdwr(UIO_READ, nd.ni_vp, ctf_hdr, sizeof(ctf_hdr), + shdr[i].sh_offset, UIO_SYSSPACE, IO_NODELOCKED, td->td_ucred, + NOCRED, &resid, td)) != 0) + goto out; + + /* Check the CTF magic number. (XXX check for big endian!) */ + if (ctf_hdr[0] != 0xf1 || ctf_hdr[1] != 0xcf) + goto out; + + /* Check if version 2. */ + if (ctf_hdr[2] != 2) + goto out; + + /* Check if the data is compressed. */ + if ((ctf_hdr[3] & 0x1) != 0) { + uint32_t *u32 = (uint32_t *) ctf_hdr; + + /* + * The last two fields in the CTF header are the offset + * from the end of the header to the start of the string + * data and the length of that string data. se this + * information to determine the decompressed CTF data + * buffer required. + */ + sz = u32[CTF_HDR_STRTAB_U32] + u32[CTF_HDR_STRLEN_U32] + + sizeof(ctf_hdr); + + /* + * Allocate memory for the compressed CTF data, including + * the header (which isn't compressed). + */ + if ((raw = malloc(shdr[i].sh_size, M_LINKER, M_WAITOK)) == NULL) { + error = ENOMEM; + goto out; + } + } else { + /* + * The CTF data is not compressed, so the ELF section + * size is the same as the buffer size required. + */ + sz = shdr[i].sh_size; + } + + /* + * Allocate memory to buffer the CTF data in it's decompressed + * form. + */ + if ((ctftab = malloc(sz, M_LINKER, M_WAITOK)) == NULL) { + error = ENOMEM; + goto out; + } + + /* + * Read the CTF data into the raw buffer if compressed, or + * directly into the CTF buffer otherwise. + */ + if ((error = vn_rdwr(UIO_READ, nd.ni_vp, raw == NULL ? ctftab : raw, + shdr[i].sh_size, shdr[i].sh_offset, UIO_SYSSPACE, IO_NODELOCKED, + td->td_ucred, NOCRED, &resid, td)) != 0) + goto out; + + /* Check if decompression is required. */ + if (raw != NULL) { + z_stream zs; + int ret; + + /* + * The header isn't compressed, so copy that into the + * CTF buffer first. + */ + bcopy(ctf_hdr, ctftab, sizeof(ctf_hdr)); + + /* Initialise the zlib structure. */ + bzero(&zs, sizeof(zs)); + zs.zalloc = z_alloc; + zs.zfree = z_free; + + if (inflateInit(&zs) != Z_OK) { + error = EIO; + goto out; + } + + zs.avail_in = shdr[i].sh_size - sizeof(ctf_hdr); + zs.next_in = ((uint8_t *) raw) + sizeof(ctf_hdr); + zs.avail_out = sz - sizeof(ctf_hdr); + zs.next_out = ((uint8_t *) ctftab) + sizeof(ctf_hdr); + if ((ret = inflate(&zs, Z_FINISH)) != Z_STREAM_END) { + printf("%s(%d): zlib inflate returned %d\n", __func__, __LINE__, ret); + error = EIO; + goto out; + } + } + + /* Got the CTF data! */ + ef->ctftab = ctftab; + ef->ctfcnt = shdr[i].sh_size; + + /* We'll retain the memory allocated for the CTF data. */ + ctftab = NULL; + + /* Let the caller use the CTF data read. */ + lc->ctftab = ef->ctftab; + lc->ctfcnt = ef->ctfcnt; + lc->symtab = ef->ddbsymtab; + lc->strtab = ef->ddbstrtab; + lc->strcnt = ef->ddbstrcnt; + lc->nsym = ef->ddbsymcnt; + lc->ctfoffp = (uint32_t **) &ef->ctfoff; + lc->typoffp = (uint32_t **) &ef->typoff; + lc->typlenp = &ef->typlen; + +out: + VOP_UNLOCK(nd.ni_vp, 0, td); + vn_close(nd.ni_vp, FREAD, td->td_ucred, td); + VFS_UNLOCK_GIANT(vfslocked); + + if (hdr != NULL) + free(hdr, M_LINKER); + if (shdr != NULL) + free(shdr, M_LINKER); + if (shstrtab != NULL) + free(shstrtab, M_LINKER); + if (ctftab != NULL) + free(ctftab, M_LINKER); + if (raw != NULL) + free(raw, M_LINKER); +#else + error = EOPNOTSUPP; +#endif + + return (error); +} diff --git a/sys/kern/kern_dtrace.c b/sys/kern/kern_dtrace.c new file mode 100644 index 000000000000..145535201e2f --- /dev/null +++ b/sys/kern/kern_dtrace.c @@ -0,0 +1,108 @@ +/*- + * Copyright (c) 2007-2008 John Birrell <jb@FreeBSD.org> + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +#include <sys/cdefs.h> +__FBSDID("$FreeBSD$"); + +#include "opt_kdb.h" + +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/eventhandler.h> +#include <sys/kdb.h> +#include <sys/kernel.h> +#include <sys/malloc.h> +#include <sys/proc.h> +#include <sys/dtrace_bsd.h> + +#define KDTRACE_PROC_SIZE 64 +#define KDTRACE_PROC_ZERO 8 +#define KDTRACE_THREAD_SIZE 256 +#define KDTRACE_THREAD_ZERO 64 + +MALLOC_DEFINE(M_KDTRACE, "kdtrace", "DTrace hooks"); + +/* Return the DTrace process data size compiled in the kernel hooks. */ +size_t +kdtrace_proc_size() +{ + return(KDTRACE_PROC_SIZE); +} + +static void +kdtrace_proc_ctor(void *arg __unused, struct proc *p) +{ + p->p_dtrace = malloc(KDTRACE_PROC_SIZE, M_KDTRACE, M_WAITOK); + + bzero(p->p_dtrace, KDTRACE_PROC_ZERO); +} + +static void +kdtrace_proc_dtor(void *arg __unused, struct proc *p) +{ + if (p->p_dtrace != NULL) { + free(p->p_dtrace, M_KDTRACE); + p->p_dtrace = NULL; + } +} + +/* Return the DTrace thread data size compiled in the kernel hooks. */ +size_t +kdtrace_thread_size() +{ + return(KDTRACE_THREAD_SIZE); +} + +static void +kdtrace_thread_ctor(void *arg __unused, struct thread *td) +{ + td->td_dtrace = malloc(KDTRACE_THREAD_SIZE, M_KDTRACE, M_WAITOK); + + bzero(td->td_dtrace, KDTRACE_THREAD_ZERO); +} + +static void +kdtrace_thread_dtor(void *arg __unused, struct thread *td) +{ + if (td->td_dtrace != NULL) { + free(td->td_dtrace, M_KDTRACE); + td->td_dtrace = NULL; + } +} + +/* + * Initialise the kernel DTrace hooks. + */ +static void +init_dtrace(void *dummy __unused) +{ + EVENTHANDLER_REGISTER(process_ctor, kdtrace_proc_ctor, NULL, EVENTHANDLER_PRI_ANY); + EVENTHANDLER_REGISTER(process_dtor, kdtrace_proc_dtor, NULL, EVENTHANDLER_PRI_ANY); + EVENTHANDLER_REGISTER(thread_ctor, kdtrace_thread_ctor, NULL, EVENTHANDLER_PRI_ANY); + EVENTHANDLER_REGISTER(thread_dtor, kdtrace_thread_dtor, NULL, EVENTHANDLER_PRI_ANY); +} + +SYSINIT(kdtrace, SI_SUB_KDTRACE, SI_ORDER_FIRST, init_dtrace, NULL); diff --git a/sys/kern/kern_exec.c b/sys/kern/kern_exec.c index 22f1cc173071..81902adfc7d5 100644 --- a/sys/kern/kern_exec.c +++ b/sys/kern/kern_exec.c @@ -28,6 +28,7 @@ __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" +#include "opt_kdtrace.h" #include "opt_ktrace.h" #include "opt_mac.h" @@ -53,6 +54,7 @@ __FBSDID("$FreeBSD$"); #include <sys/pioctl.h> #include <sys/namei.h> #include <sys/resourcevar.h> +#include <sys/sdt.h> #include <sys/sf_buf.h> #include <sys/syscallsubr.h> #include <sys/sysent.h> @@ -82,6 +84,19 @@ __FBSDID("$FreeBSD$"); #include <security/audit/audit.h> #include <security/mac/mac_framework.h> +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> +dtrace_execexit_func_t dtrace_fasttrap_exec; +#endif + +SDT_PROVIDER_DECLARE(proc); +SDT_PROBE_DEFINE(proc, kernel, , exec); +SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *"); +SDT_PROBE_DEFINE(proc, kernel, , exec_failure); +SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int"); +SDT_PROBE_DEFINE(proc, kernel, , exec_success); +SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *"); + MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); @@ -348,6 +363,8 @@ do_execve(td, args, mac_p) imgp->image_header = NULL; + SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 ); + /* * Translate the file name. namei() returns a vnode pointer * in ni_vp amoung other things. @@ -658,6 +675,15 @@ interpret: textvp = p->p_textvp; p->p_textvp = ndp->ni_vp; +#ifdef KDTRACE_HOOKS + /* + * Tell the DTrace fasttrap provider about the exec if it + * has declared an interest. + */ + if (dtrace_fasttrap_exec) + dtrace_fasttrap_exec(p); +#endif + /* * Notify others that we exec'd, and clear the P_INEXEC flag * as we're now a bona fide freshly-execed process. @@ -729,6 +755,9 @@ done1: else crfree(newcred); VOP_UNLOCK(imgp->vp, 0, td); + + SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0); + /* * Handle deferred decrement of ref counts. */ @@ -791,6 +820,8 @@ exec_fail: p->p_flag &= ~P_INEXEC; PROC_UNLOCK(p); + SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0); + done2: #ifdef MAC mac_execve_exit(imgp); diff --git a/sys/kern/kern_exit.c b/sys/kern/kern_exit.c index 4f8a70e4e5e4..1c9bf4c115eb 100644 --- a/sys/kern/kern_exit.c +++ b/sys/kern/kern_exit.c @@ -38,6 +38,7 @@ __FBSDID("$FreeBSD$"); #include "opt_compat.h" +#include "opt_kdtrace.h" #include "opt_ktrace.h" #include "opt_mac.h" @@ -65,6 +66,7 @@ __FBSDID("$FreeBSD$"); #include <sys/ptrace.h> #include <sys/acct.h> /* for acct_process() function prototype */ #include <sys/filedesc.h> +#include <sys/sdt.h> #include <sys/shm.h> #include <sys/sem.h> #ifdef KTRACE @@ -82,6 +84,15 @@ __FBSDID("$FreeBSD$"); #include <vm/vm_page.h> #include <vm/uma.h> +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> +dtrace_execexit_func_t dtrace_fasttrap_exit; +#endif + +SDT_PROVIDER_DECLARE(proc); +SDT_PROBE_DEFINE(proc, kernel, , exit); +SDT_PROBE_ARGTYPE(proc, kernel, , exit, 0, "int"); + /* Required to be non-static for SysVR4 emulator */ MALLOC_DEFINE(M_ZOMBIE, "zombie", "zombie proc status"); @@ -449,11 +460,29 @@ retry: PROC_LOCK(p); p->p_xstat = rv; p->p_xthread = td; + +#ifdef KDTRACE_HOOKS + /* + * Tell the DTrace fasttrap provider about the exit if it + * has declared an interest. + */ + if (dtrace_fasttrap_exit) + dtrace_fasttrap_exit(p); +#endif + /* * Notify interested parties of our demise. */ KNOTE_LOCKED(&p->p_klist, NOTE_EXIT); +#ifdef KDTRACE_HOOKS + int reason = CLD_EXITED; + if (WCOREDUMP(rv)) + reason = CLD_DUMPED; + else if (WIFSIGNALED(rv)) + reason = CLD_KILLED; + SDT_PROBE(proc, kernel, , exit, reason, 0, 0, 0, 0); +#endif /* * Just delete all entries in the p_klist. At this point we won't * report any more events, and there are nasty race conditions that diff --git a/sys/kern/kern_fork.c b/sys/kern/kern_fork.c index 2d15339a4894..71fe803c4653 100644 --- a/sys/kern/kern_fork.c +++ b/sys/kern/kern_fork.c @@ -37,6 +37,7 @@ #include <sys/cdefs.h> __FBSDID("$FreeBSD$"); +#include "opt_kdtrace.h" #include "opt_ktrace.h" #include "opt_mac.h" @@ -63,6 +64,7 @@ __FBSDID("$FreeBSD$"); #include <sys/ktr.h> #include <sys/ktrace.h> #include <sys/unistd.h> +#include <sys/sdt.h> #include <sys/sx.h> #include <sys/signalvar.h> @@ -75,6 +77,16 @@ __FBSDID("$FreeBSD$"); #include <vm/vm_extern.h> #include <vm/uma.h> +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> +dtrace_fork_func_t dtrace_fasttrap_fork; +#endif + +SDT_PROVIDER_DECLARE(proc); +SDT_PROBE_DEFINE(proc, kernel, , create); +SDT_PROBE_ARGTYPE(proc, kernel, , create, 0, "struct proc *"); +SDT_PROBE_ARGTYPE(proc, kernel, , create, 1, "struct proc *"); +SDT_PROBE_ARGTYPE(proc, kernel, , create, 2, "int"); #ifndef _SYS_SYSPROTO_H_ struct fork_args { @@ -620,6 +632,15 @@ again: p2->p_pfsflags = p1->p_pfsflags; } +#ifdef KDTRACE_HOOKS + /* + * Tell the DTrace fasttrap provider about the new process + * if it has registered an interest. + */ + if (dtrace_fasttrap_fork) + dtrace_fasttrap_fork(p1, p2); +#endif + /* * This begins the section where we must prevent the parent * from being swapped. @@ -707,6 +728,8 @@ again: * Tell any interested parties about the new process. */ knote_fork(&p1->p_klist, p2->p_pid); + SDT_PROBE(proc, kernel, , create, p2, p1, flags, 0, 0); + /* * Preserve synchronization semantics of vfork. If waiting for * child to exec or exit, set P_PPWAIT on child, and sleep on our diff --git a/sys/kern/kern_linker.c b/sys/kern/kern_linker.c index de18e9e5cf79..3c114525adcf 100644 --- a/sys/kern/kern_linker.c +++ b/sys/kern/kern_linker.c @@ -97,6 +97,12 @@ linker_file_t linker_kernel_file; static struct sx kld_sx; /* kernel linker lock */ +/* + * Load counter used by clients to determine if a linker file has been + * re-loaded. This counter is incremented for each file load. + */ +static int loadcnt; + static linker_class_list_t classes; static linker_file_list_t linker_files; static int next_file_id = 1; @@ -535,7 +541,7 @@ linker_make_file(const char *pathname, linker_class_t lc) KLD_LOCK_ASSERT(); filename = linker_basename(pathname); - KLD_DPF(FILE, ("linker_make_file: new file, filename=%s\n", filename)); + KLD_DPF(FILE, ("linker_make_file: new file, filename='%s' for pathname='%s'\n", filename, pathname)); lf = (linker_file_t)kobj_create((kobj_class_t)lc, M_LINKER, M_WAITOK); if (lf == NULL) return (NULL); @@ -543,9 +549,13 @@ linker_make_file(const char *pathname, linker_class_t lc) lf->userrefs = 0; lf->flags = 0; lf->filename = linker_strdup(filename); + lf->pathname = linker_strdup(pathname); LINKER_GET_NEXT_FILE_ID(lf->id); lf->ndeps = 0; lf->deps = NULL; + lf->loadcnt = ++loadcnt; + lf->sdt_probes = NULL; + lf->sdt_nprobes = 0; STAILQ_INIT(&lf->common); TAILQ_INIT(&lf->modules); TAILQ_INSERT_TAIL(&linker_files, lf, link); @@ -630,10 +640,20 @@ linker_file_unload(linker_file_t file, int flags) free(file->filename, M_LINKER); file->filename = NULL; } + if (file->pathname) { + free(file->pathname, M_LINKER); + file->pathname = NULL; + } kobj_delete((kobj_t) file, M_LINKER); return (0); } +int +linker_ctf_get(linker_file_t file, linker_ctf_t *lc) +{ + return (LINKER_CTF_GET(file, lc)); +} + static int linker_file_add_dependency(linker_file_t file, linker_file_t dep) { @@ -677,6 +697,16 @@ linker_file_lookup_set(linker_file_t file, const char *name, return (error); } +/* + * List all functions in a file. + */ +int +linker_file_function_listall(linker_file_t lf, + linker_function_nameval_callback_t callback_func, void *arg) +{ + return (LINKER_EACH_FUNCTION_NAMEVAL(lf, callback_func, arg)); +} + caddr_t linker_file_lookup_symbol(linker_file_t file, const char *name, int deps) { @@ -995,7 +1025,13 @@ kern_kldunload(struct thread *td, int fileid, int flags) lf = linker_find_file_by_id(fileid); if (lf) { KLD_DPF(FILE, ("kldunload: lf->userrefs=%d\n", lf->userrefs)); - if (lf->userrefs == 0) { + + /* Check if there are DTrace probes enabled on this file. */ + if (lf->nenabled > 0) { + printf("kldunload: attempt to unload file that has" + " DTrace probes enabled\n"); + error = EBUSY; + } else if (lf->userrefs == 0) { /* * XXX: maybe LINKER_UNLOAD_FORCE should override ? */ @@ -1116,15 +1152,18 @@ kldstat(struct thread *td, struct kldstat_args *uap) { struct kld_file_stat stat; linker_file_t lf; - int error, namelen; + int error, namelen, version, version_num; /* * Check the version of the user's structure. */ - error = copyin(uap->stat, &stat, sizeof(struct kld_file_stat)); - if (error) + if ((error = copyin(&uap->stat->version, &version, sizeof(version))) != 0) return (error); - if (stat.version != sizeof(struct kld_file_stat)) + if (version == sizeof(struct kld_file_stat_1)) + version_num = 1; + else if (version == sizeof(struct kld_file_stat)) + version_num = 2; + else return (EINVAL); #ifdef MAC @@ -1140,6 +1179,7 @@ kldstat(struct thread *td, struct kldstat_args *uap) return (ENOENT); } + /* Version 1 fields: */ namelen = strlen(lf->filename) + 1; if (namelen > MAXPATHLEN) namelen = MAXPATHLEN; @@ -1148,6 +1188,13 @@ kldstat(struct thread *td, struct kldstat_args *uap) stat.id = lf->id; stat.address = lf->address; stat.size = lf->size; + if (version_num > 1) { + /* Version 2 fields: */ + namelen = strlen(lf->pathname) + 1; + if (namelen > MAXPATHLEN) + namelen = MAXPATHLEN; + bcopy(lf->pathname, &stat.pathname[0], namelen); + } KLD_UNLOCK(); td->td_retval[0] = 0; diff --git a/sys/kern/kern_malloc.c b/sys/kern/kern_malloc.c index 20c9fff09dfe..fc7e79409135 100644 --- a/sys/kern/kern_malloc.c +++ b/sys/kern/kern_malloc.c @@ -46,6 +46,7 @@ __FBSDID("$FreeBSD$"); #include "opt_ddb.h" +#include "opt_kdtrace.h" #include "opt_vm.h" #include <sys/param.h> @@ -86,6 +87,12 @@ __FBSDID("$FreeBSD$"); #include <ddb/ddb.h> +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> + +dtrace_malloc_probe_func_t dtrace_malloc_probe; +#endif + /* * When realloc() is called, if the new size is sufficiently smaller than * the old size, realloc() will allocate a new, smaller block to avoid @@ -255,6 +262,17 @@ malloc_type_zone_allocated(struct malloc_type *mtp, unsigned long size, } if (zindx != -1) mtsp->mts_size |= 1 << zindx; + +#ifdef KDTRACE_HOOKS + if (dtrace_malloc_probe != NULL) { + uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_MALLOC]; + if (probe_id != 0) + (dtrace_malloc_probe)(probe_id, + (uintptr_t) mtp, (uintptr_t) mtip, + (uintptr_t) mtsp, size, zindx); + } +#endif + critical_exit(); } @@ -283,6 +301,17 @@ malloc_type_freed(struct malloc_type *mtp, unsigned long size) mtsp = &mtip->mti_stats[curcpu]; mtsp->mts_memfreed += size; mtsp->mts_numfrees++; + +#ifdef KDTRACE_HOOKS + if (dtrace_malloc_probe != NULL) { + uint32_t probe_id = mtip->mti_probes[DTMALLOC_PROBE_FREE]; + if (probe_id != 0) + (dtrace_malloc_probe)(probe_id, + (uintptr_t) mtp, (uintptr_t) mtip, + (uintptr_t) mtsp, size, 0); + } +#endif + critical_exit(); } @@ -804,6 +833,40 @@ SYSCTL_PROC(_kern, OID_AUTO, malloc_stats, CTLFLAG_RD|CTLTYPE_STRUCT, SYSCTL_INT(_kern, OID_AUTO, malloc_count, CTLFLAG_RD, &kmemcount, 0, "Count of kernel malloc types"); +void +malloc_type_list(malloc_type_list_func_t *func, void *arg) +{ + struct malloc_type *mtp, **bufmtp; + int count, i; + size_t buflen; + + mtx_lock(&malloc_mtx); +restart: + mtx_assert(&malloc_mtx, MA_OWNED); + count = kmemcount; + mtx_unlock(&malloc_mtx); + + buflen = sizeof(struct malloc_type *) * count; + bufmtp = malloc(buflen, M_TEMP, M_WAITOK); + + mtx_lock(&malloc_mtx); + + if (count < kmemcount) { + free(bufmtp, M_TEMP); + goto restart; + } + + for (mtp = kmemstatistics, i = 0; mtp != NULL; mtp = mtp->ks_next, i++) + bufmtp[i] = mtp; + + mtx_unlock(&malloc_mtx); + + for (i = 0; i < count; i++) + (func)(bufmtp[i], arg); + + free(bufmtp, M_TEMP); +} + #ifdef DDB DB_SHOW_COMMAND(malloc, db_show_malloc) { diff --git a/sys/kern/kern_proc.c b/sys/kern/kern_proc.c index 20604a6635bc..6debfe2c5d4e 100644 --- a/sys/kern/kern_proc.c +++ b/sys/kern/kern_proc.c @@ -33,6 +33,7 @@ __FBSDID("$FreeBSD$"); #include "opt_ddb.h" +#include "opt_kdtrace.h" #include "opt_ktrace.h" #include "opt_kstack_pages.h" #include "opt_stack.h" @@ -54,6 +55,7 @@ __FBSDID("$FreeBSD$"); #include <sys/sysctl.h> #include <sys/filedesc.h> #include <sys/tty.h> +#include <sys/sdt.h> #include <sys/signalvar.h> #include <sys/sx.h> #include <sys/user.h> @@ -76,6 +78,35 @@ __FBSDID("$FreeBSD$"); #include <vm/vm_object.h> #include <vm/uma.h> +SDT_PROVIDER_DEFINE(proc); +SDT_PROBE_DEFINE(proc, kernel, ctor, entry); +SDT_PROBE_ARGTYPE(proc, kernel, ctor, entry, 0, "struct proc *"); +SDT_PROBE_ARGTYPE(proc, kernel, ctor, entry, 1, "int"); +SDT_PROBE_ARGTYPE(proc, kernel, ctor, entry, 2, "void *"); +SDT_PROBE_ARGTYPE(proc, kernel, ctor, entry, 3, "int"); +SDT_PROBE_DEFINE(proc, kernel, ctor, return); +SDT_PROBE_ARGTYPE(proc, kernel, ctor, return, 0, "struct proc *"); +SDT_PROBE_ARGTYPE(proc, kernel, ctor, return, 1, "int"); +SDT_PROBE_ARGTYPE(proc, kernel, ctor, return, 2, "void *"); +SDT_PROBE_ARGTYPE(proc, kernel, ctor, return, 3, "int"); +SDT_PROBE_DEFINE(proc, kernel, dtor, entry); +SDT_PROBE_ARGTYPE(proc, kernel, dtor, entry, 0, "struct proc *"); +SDT_PROBE_ARGTYPE(proc, kernel, dtor, entry, 1, "int"); +SDT_PROBE_ARGTYPE(proc, kernel, dtor, entry, 2, "void *"); +SDT_PROBE_ARGTYPE(proc, kernel, dtor, entry, 3, "struct thread *"); +SDT_PROBE_DEFINE(proc, kernel, dtor, return); +SDT_PROBE_ARGTYPE(proc, kernel, dtor, return, 0, "struct proc *"); +SDT_PROBE_ARGTYPE(proc, kernel, dtor, return, 1, "int"); +SDT_PROBE_ARGTYPE(proc, kernel, dtor, return, 2, "void *"); +SDT_PROBE_DEFINE(proc, kernel, init, entry); +SDT_PROBE_ARGTYPE(proc, kernel, init, entry, 0, "struct proc *"); +SDT_PROBE_ARGTYPE(proc, kernel, init, entry, 1, "int"); +SDT_PROBE_ARGTYPE(proc, kernel, init, entry, 2, "int"); +SDT_PROBE_DEFINE(proc, kernel, init, return); +SDT_PROBE_ARGTYPE(proc, kernel, init, return, 0, "struct proc *"); +SDT_PROBE_ARGTYPE(proc, kernel, init, return, 1, "int"); +SDT_PROBE_ARGTYPE(proc, kernel, init, return, 2, "int"); + MALLOC_DEFINE(M_PGRP, "pgrp", "process group header"); MALLOC_DEFINE(M_SESSION, "session", "session header"); static MALLOC_DEFINE(M_PROC, "proc", "Proc structures"); @@ -143,7 +174,9 @@ proc_ctor(void *mem, int size, void *arg, int flags) struct proc *p; p = (struct proc *)mem; + SDT_PROBE(proc, kernel, ctor, entry, p, size, arg, flags, 0); EVENTHANDLER_INVOKE(process_ctor, p); + SDT_PROBE(proc, kernel, ctor, return, p, size, arg, flags, 0); return (0); } @@ -158,7 +191,8 @@ proc_dtor(void *mem, int size, void *arg) /* INVARIANTS checks go here */ p = (struct proc *)mem; - td = FIRST_THREAD_IN_PROC(p); + td = FIRST_THREAD_IN_PROC(p); + SDT_PROBE(proc, kernel, dtor, entry, p, size, arg, td, 0); if (td != NULL) { #ifdef INVARIANTS KASSERT((p->p_numthreads == 1), @@ -177,6 +211,7 @@ proc_dtor(void *mem, int size, void *arg) EVENTHANDLER_INVOKE(process_dtor, p); if (p->p_ksi != NULL) KASSERT(! KSI_ONQ(p->p_ksi), ("SIGCHLD queue")); + SDT_PROBE(proc, kernel, dtor, return, p, size, arg, 0, 0); } /* @@ -188,6 +223,7 @@ proc_init(void *mem, int size, int flags) struct proc *p; p = (struct proc *)mem; + SDT_PROBE(proc, kernel, init, entry, p, size, flags, 0, 0); p->p_sched = (struct p_sched *)&p[1]; bzero(&p->p_mtx, sizeof(struct mtx)); mtx_init(&p->p_mtx, "process lock", NULL, MTX_DEF | MTX_DUPOK); @@ -195,6 +231,7 @@ proc_init(void *mem, int size, int flags) TAILQ_INIT(&p->p_threads); /* all threads in proc */ EVENTHANDLER_INVOKE(process_init, p); p->p_stats = pstats_alloc(); + SDT_PROBE(proc, kernel, init, return, p, size, flags, 0, 0); return (0); } diff --git a/sys/kern/kern_sdt.c b/sys/kern/kern_sdt.c new file mode 100644 index 000000000000..495a6a05bc3e --- /dev/null +++ b/sys/kern/kern_sdt.c @@ -0,0 +1,273 @@ +/*- + * Copyright 2006-2008 John Birrell <jb@FreeBSD.org> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + * Backend for the Statically Defined Tracing (SDT) kernel support. This is + * required to allow a module to load even though DTrace kernel support may + * not be present. A module may be built with SDT probes in it which are + * registered and deregistered via SYSINIT/SYSUNINIT. + * + */ + +#include "opt_kdtrace.h" + +#include <sys/cdefs.h> +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/kernel.h> +#include <sys/linker.h> +#include <sys/lock.h> +#include <sys/proc.h> +#include <sys/sx.h> +#include <sys/sdt.h> + +/* + * This is the list of statically defined tracing providers. + */ +static TAILQ_HEAD(sdt_provider_list_head, sdt_provider) sdt_provider_list; + +/* + * Mutex to serialise access to the SDT provider list. + */ +static struct sx sdt_sx; + +/* + * Hook for the DTrace probe function. The 'sdt' provider will set this + * to dtrace_probe when it loads. + */ +sdt_probe_func_t sdt_probe_func = sdt_probe_stub; + +/* + * This is a stub for probe calls in case kernel DTrace support isn't + * compiled in. It should never get called because there is no DTrace + * support to enable it. + */ +void +sdt_probe_stub(u_int32_t id, uintptr_t arg0, uintptr_t arg1, + uintptr_t arg2, uintptr_t arg3, uintptr_t arg4) +{ + printf("sdt_probe_stub: Why did this get called?\n"); +} + +/* + * Called from SYSINIT to register a provider. + */ +void +sdt_provider_register(void *arg) +{ + struct sdt_provider *prov = arg; + + sx_xlock(&sdt_sx); + + TAILQ_INSERT_TAIL(&sdt_provider_list, prov, prov_entry); + + TAILQ_INIT(&prov->probe_list); + + sx_xunlock(&sdt_sx); +} + +/* + * Called from SYSUNINIT to de-register a provider. + */ +void +sdt_provider_deregister(void *arg) +{ + struct sdt_provider *prov = arg; + + sx_xlock(&sdt_sx); + + TAILQ_REMOVE(&sdt_provider_list, prov, prov_entry); + + sx_xunlock(&sdt_sx); +} + +/* + * Called from SYSINIT to register a statically defined trace probe. + */ +void +sdt_probe_register(void *arg) +{ + struct sdt_probe *probe = arg; + + /* + * Check the reference structure version. Only version 1 is + * supported at the moment. + */ + if (probe->version != sizeof(struct sdt_probe)) { + printf("%s:%s:%s has version %d when %d required\n", probe->mod, probe->func, probe->name, probe->version, (int) sizeof(struct sdt_probe)); + return; + } + + sx_xlock(&sdt_sx); + + TAILQ_INSERT_TAIL(&probe->prov->probe_list, probe, probe_entry); + + TAILQ_INIT(&probe->argtype_list); + + probe->state = SDT_INIT; + + sx_xunlock(&sdt_sx); +} + +/* + * Called from SYSUNINIT to de-register a statically defined trace probe. + */ +void +sdt_probe_deregister(void *arg) +{ + struct sdt_probe *probe = arg; + + sx_xlock(&sdt_sx); + + if (probe->state == SDT_INIT) { + TAILQ_REMOVE(&probe->prov->probe_list, probe, probe_entry); + probe->state = SDT_UNINIT; + } + + sx_xunlock(&sdt_sx); +} + +/* + * Called from SYSINIT to register a statically defined trace probe argument. + */ +void +sdt_argtype_register(void *arg) +{ + struct sdt_argtype *argtype = arg; + + sx_xlock(&sdt_sx); + + TAILQ_INSERT_TAIL(&argtype->probe->argtype_list, argtype, argtype_entry); + + argtype->probe->n_args++; + + sx_xunlock(&sdt_sx); +} + +/* + * Called from SYSUNINIT to de-register a statically defined trace probe argument. + */ +void +sdt_argtype_deregister(void *arg) +{ + struct sdt_argtype *argtype = arg; + + sx_xlock(&sdt_sx); + + TAILQ_REMOVE(&argtype->probe->argtype_list, argtype, argtype_entry); + + sx_xunlock(&sdt_sx); +} + +static void +sdt_init(void *arg) +{ + sx_init_flags(&sdt_sx, "Statically Defined Tracing", SX_NOWITNESS); + + TAILQ_INIT(&sdt_provider_list); +} + +SYSINIT(sdt, SI_SUB_KDTRACE, SI_ORDER_FIRST, sdt_init, NULL); + +static void +sdt_uninit(void *arg) +{ + sx_destroy(&sdt_sx); +} + +SYSUNINIT(sdt, SI_SUB_KDTRACE, SI_ORDER_FIRST, sdt_uninit, NULL); + +/* + * List statically defined tracing providers. + */ +int +sdt_provider_listall(sdt_provider_listall_func_t callback_func,void *arg) +{ + int error = 0; + struct sdt_provider *prov; + + sx_xlock(&sdt_sx); + + TAILQ_FOREACH(prov, &sdt_provider_list, prov_entry) { + if ((error = callback_func(prov, arg)) != 0) + break; + } + + sx_xunlock(&sdt_sx); + + return (error); +} + +/* + * List statically defined tracing probes. + */ +int +sdt_probe_listall(struct sdt_provider *prov, + sdt_probe_listall_func_t callback_func,void *arg) +{ + int error = 0; + int locked; + struct sdt_probe *probe; + + locked = sx_xlocked(&sdt_sx); + if (!locked) + sx_xlock(&sdt_sx); + + TAILQ_FOREACH(probe, &prov->probe_list, probe_entry) { + if ((error = callback_func(probe, arg)) != 0) + break; + } + + if (!locked) + sx_xunlock(&sdt_sx); + + return (error); +} + +/* + * List statically defined tracing probe arguments. + */ +int +sdt_argtype_listall(struct sdt_probe *probe, + sdt_argtype_listall_func_t callback_func,void *arg) +{ + int error = 0; + int locked; + struct sdt_argtype *argtype; + + locked = sx_xlocked(&sdt_sx); + if (!locked) + sx_xlock(&sdt_sx); + + TAILQ_FOREACH(argtype, &probe->argtype_list, argtype_entry) { + if ((error = callback_func(argtype, arg)) != 0) + break; + } + + if (!locked) + sx_xunlock(&sdt_sx); + + return (error); +} diff --git a/sys/kern/link_elf.c b/sys/kern/link_elf.c index e32a0a41027e..99e0ff5edfe0 100644 --- a/sys/kern/link_elf.c +++ b/sys/kern/link_elf.c @@ -27,6 +27,7 @@ #include <sys/cdefs.h> __FBSDID("$FreeBSD$"); +#include "opt_ddb.h" #include "opt_gdb.h" #include "opt_mac.h" @@ -62,6 +63,10 @@ __FBSDID("$FreeBSD$"); #include <sys/link_elf.h> +#ifdef DDB_CTF +#include <net/zlib.h> +#endif + #include "linker_if.h" #define MAXSEGS 4 @@ -98,11 +103,18 @@ typedef struct elf_file { long ddbstrcnt; /* number of bytes in string table */ caddr_t symbase; /* malloc'ed symbold base */ caddr_t strbase; /* malloc'ed string base */ + caddr_t ctftab; /* CTF table */ + long ctfcnt; /* number of bytes in CTF table */ + caddr_t ctfoff; /* CTF offset table */ + caddr_t typoff; /* Type offset table */ + long typlen; /* Number of type entries. */ #ifdef GDB struct link_map gdb; /* hooks for gdb */ #endif } *elf_file_t; +#include <kern/kern_ctf.c> + static int link_elf_link_common_finish(linker_file_t); static int link_elf_link_preload(linker_class_t cls, const char*, linker_file_t*); @@ -121,6 +133,9 @@ static int link_elf_lookup_set(linker_file_t, const char *, static int link_elf_each_function_name(linker_file_t, int (*)(const char *, void *), void *); +static int link_elf_each_function_nameval(linker_file_t, + linker_function_nameval_callback_t, + void *); static void link_elf_reloc_local(linker_file_t); static Elf_Addr elf_lookup(linker_file_t lf, Elf_Size symidx, int deps); @@ -134,6 +149,8 @@ static kobj_method_t link_elf_methods[] = { KOBJMETHOD(linker_link_preload_finish, link_elf_link_preload_finish), KOBJMETHOD(linker_lookup_set, link_elf_lookup_set), KOBJMETHOD(linker_each_function_name, link_elf_each_function_name), + KOBJMETHOD(linker_each_function_nameval, link_elf_each_function_nameval), + KOBJMETHOD(linker_ctf_get, link_elf_ctf_get), { 0, 0 } }; @@ -580,7 +597,7 @@ link_elf_load_file(linker_class_t cls, const char* filename, goto out; } #ifdef MAC - error = mac_check_kld_load(curthread->td_ucred, nd.ni_vp); + error = mac_kld_check_load(curthread->td_ucred, nd.ni_vp); if (error) { firstpage = NULL; goto out; @@ -915,6 +932,12 @@ link_elf_unload_file(linker_file_t file) free(ef->symbase, M_LINKER); if (ef->strbase) free(ef->strbase, M_LINKER); + if (ef->ctftab) + free(ef->ctftab, M_LINKER); + if (ef->ctfoff) + free(ef->ctfoff, M_LINKER); + if (ef->typoff) + free(ef->typoff, M_LINKER); } static void @@ -1232,6 +1255,30 @@ link_elf_each_function_name(linker_file_t file, return (0); } +static int +link_elf_each_function_nameval(linker_file_t file, + linker_function_nameval_callback_t callback, void *opaque) +{ + linker_symval_t symval; + elf_file_t ef = (elf_file_t)file; + const Elf_Sym* symp; + int i, error; + + /* Exhaustive search */ + for (i = 0, symp = ef->ddbsymtab; i < ef->ddbsymcnt; i++, symp++) { + if (symp->st_value != 0 && + ELF_ST_TYPE(symp->st_info) == STT_FUNC) { + error = link_elf_symbol_values(file, (c_linker_sym_t) symp, &symval); + if (error) + return (error); + error = callback(file, i, &symval, opaque); + if (error) + return (error); + } + } + return (0); +} + #ifdef __ia64__ /* * Each KLD has its own GP. The GP value for each load module is given by diff --git a/sys/kern/link_elf_obj.c b/sys/kern/link_elf_obj.c index af6defaef80a..61aeebb2a475 100644 --- a/sys/kern/link_elf_obj.c +++ b/sys/kern/link_elf_obj.c @@ -58,6 +58,10 @@ __FBSDID("$FreeBSD$"); #include <sys/link_elf.h> +#ifdef DDB_CTF +#include <net/zlib.h> +#endif + #include "linker_if.h" typedef struct { @@ -106,8 +110,16 @@ typedef struct elf_file { caddr_t shstrtab; /* Section name string table */ long shstrcnt; /* number of bytes in string table */ + caddr_t ctftab; /* CTF table */ + long ctfcnt; /* number of bytes in CTF table */ + caddr_t ctfoff; /* CTF offset table */ + caddr_t typoff; /* Type offset table */ + long typlen; /* Number of type entries. */ + } *elf_file_t; +#include <kern/kern_ctf.c> + static int link_elf_link_preload(linker_class_t cls, const char *, linker_file_t *); static int link_elf_link_preload_finish(linker_file_t); @@ -124,6 +136,9 @@ static int link_elf_lookup_set(linker_file_t, const char *, void ***, void ***, int *); static int link_elf_each_function_name(linker_file_t, int (*)(const char *, void *), void *); +static int link_elf_each_function_nameval(linker_file_t, + linker_function_nameval_callback_t, + void *); static void link_elf_reloc_local(linker_file_t); static Elf_Addr elf_obj_lookup(linker_file_t lf, Elf_Size symidx, int deps); @@ -138,6 +153,8 @@ static kobj_method_t link_elf_methods[] = { KOBJMETHOD(linker_link_preload_finish, link_elf_link_preload_finish), KOBJMETHOD(linker_lookup_set, link_elf_lookup_set), KOBJMETHOD(linker_each_function_name, link_elf_each_function_name), + KOBJMETHOD(linker_each_function_nameval, link_elf_each_function_nameval), + KOBJMETHOD(linker_ctf_get, link_elf_ctf_get), { 0, 0 } }; @@ -417,7 +434,7 @@ link_elf_load_file(linker_class_t cls, const char *filename, goto out; } #ifdef MAC - error = mac_check_kld_load(td->td_ucred, nd.ni_vp); + error = mac_kld_check_load(td->td_ucred, nd.ni_vp); if (error) { goto out; } @@ -820,6 +837,12 @@ link_elf_unload_file(linker_file_t file) free(ef->relatab, M_LINKER); if (ef->progtab) free(ef->progtab, M_LINKER); + if (ef->ctftab) + free(ef->ctftab, M_LINKER); + if (ef->ctfoff) + free(ef->ctfoff, M_LINKER); + if (ef->typoff) + free(ef->typoff, M_LINKER); if (file->filename != NULL) preload_delete_name(file->filename); /* XXX reclaim module memory? */ @@ -852,6 +875,12 @@ link_elf_unload_file(linker_file_t file) free(ef->ddbstrtab, M_LINKER); if (ef->shstrtab) free(ef->shstrtab, M_LINKER); + if (ef->ctftab) + free(ef->ctftab, M_LINKER); + if (ef->ctfoff) + free(ef->ctfoff, M_LINKER); + if (ef->typoff) + free(ef->typoff, M_LINKER); } static const char * @@ -1073,6 +1102,30 @@ link_elf_each_function_name(linker_file_t file, return (0); } +static int +link_elf_each_function_nameval(linker_file_t file, + linker_function_nameval_callback_t callback, void *opaque) +{ + linker_symval_t symval; + elf_file_t ef = (elf_file_t)file; + const Elf_Sym* symp; + int i, error; + + /* Exhaustive search */ + for (i = 0, symp = ef->ddbsymtab; i < ef->ddbsymcnt; i++, symp++) { + if (symp->st_value != 0 && + ELF_ST_TYPE(symp->st_info) == STT_FUNC) { + error = link_elf_symbol_values(file, (c_linker_sym_t) symp, &symval); + if (error) + return (error); + error = callback(file, i, &symval, opaque); + if (error) + return (error); + } + } + return (0); +} + /* * Symbol lookup function that can be used when the symbol index is known (ie * in relocations). It uses the symbol index instead of doing a fully fledged diff --git a/sys/kern/linker_if.m b/sys/kern/linker_if.m index 84d56d2d6f7e..aadcf048bc4b 100644 --- a/sys/kern/linker_if.m +++ b/sys/kern/linker_if.m @@ -64,6 +64,17 @@ METHOD int each_function_name { }; # +# Call the callback with each specified function and it's value +# defined in the file. +# Stop and return the error if the callback returns an error. +# +METHOD int each_function_nameval { + linker_file_t file; + linker_function_nameval_callback_t callback; + void* opaque; +}; + +# # Search for a linker set in a file. Return a pointer to the first # entry (which is itself a pointer), and the number of entries. # "stop" points to the entry beyond the last valid entry. @@ -85,6 +96,15 @@ METHOD void unload { }; # +# Load CTF data if necessary and if there is a .SUNW_ctf section +# in the ELF file, returning info in the linker CTF structure. +# +METHOD int ctf_get { + linker_file_t file; + linker_ctf_t *lc; +}; + +# # Load a file, returning the new linker_file_t in *result. If # the class does not recognise the file type, zero should be # returned, without modifying *result. If the file is diff --git a/sys/kern/makesyscalls.sh b/sys/kern/makesyscalls.sh index e8722abd0767..0b6bbaee7952 100644 --- a/sys/kern/makesyscalls.sh +++ b/sys/kern/makesyscalls.sh @@ -34,10 +34,11 @@ sysent="sysent.switch.$$" sysinc="sysinc.switch.$$" sysarg="sysarg.switch.$$" sysprotoend="sysprotoend.$$" +systracetmp="systrace.$$" -trap "rm $sysaue $sysdcl $syscompat $syscompatdcl $syscompat4 $syscompat4dcl $syscompat6 $syscompat6dcl $sysent $sysinc $sysarg $sysprotoend" 0 +trap "rm $sysaue $sysdcl $syscompat $syscompatdcl $syscompat4 $syscompat4dcl $syscompat6 $syscompat6dcl $sysent $sysinc $sysarg $sysprotoend $systracetmp" 0 -touch $sysaue $sysdcl $syscompat $syscompatdcl $syscompat4 $syscompat4dcl $syscompat6 $syscompat6dcl $sysent $sysinc $sysarg $sysprotoend +touch $sysaue $sysdcl $syscompat $syscompatdcl $syscompat4 $syscompat4dcl $syscompat6 $syscompat6dcl $sysent $sysinc $sysarg $sysprotoend $systracetmp case $# in 0) echo "usage: $0 input-file <config-file>" 1>&2 @@ -82,6 +83,7 @@ s/\$//g syshdr = \"$syshdr\" sysmk = \"$sysmk\" systrace = \"$systrace\" + systracetmp = \"$systracetmp\" compat = \"$compat\" compat4 = \"$compat4\" compat6 = \"$compat6\" @@ -160,6 +162,9 @@ s/\$//g printf "\tint64_t *iarg = (int64_t *) uarg;\n" > systrace printf "\tswitch (sysnum) {\n" > systrace + printf "static void\nsystrace_setargdesc(int sysnum, int ndx, char *desc, size_t descsz)\n{\n\tconst char *p = NULL;\n" > systracetmp + printf "\tswitch (sysnum) {\n" > systracetmp + next } NF == 0 || $1 ~ /^;/ { @@ -317,10 +322,13 @@ s/\$//g || $3 == "NOIMPL" || $3 == "NOSTD" { parseline() printf("\t/* %s */\n\tcase %d: {\n", funcname, syscall) > systrace + printf("\t/* %s */\n\tcase %d:\n", funcname, syscall) > systracetmp if (argc > 0) { + printf("\t\tswitch(ndx) {\n") > systracetmp printf("\t\tstruct %s *p = params;\n", argalias) > systrace for (i = 1; i <= argc; i++) { - if (index(argtype[i], "*") > 0 || argtype[i] == "caddr_t") + printf("\t\tcase %d:\n\t\t\tp = \"%s\";\n\t\t\tbreak;\n", i - 1, argtype[i]) > systracetmp + if (index(argtype[i], "*") > 0 || argtype[i] == "caddr_t" || argtype[i] == "l_handler_t") printf("\t\tuarg[%d] = (intptr_t) p->%s; /* %s */\n", \ i - 1, \ argname[i], argtype[i]) > systrace @@ -333,8 +341,10 @@ s/\$//g i - 1, \ argname[i], argtype[i]) > systrace } + printf("\t\tdefault:\n\t\t\tbreak;\n\t\t};\n") > systracetmp } printf("\t\t*n_args = %d;\n\t\tbreak;\n\t}\n", argc) > systrace + printf("\t\tbreak;\n") > systracetmp if ((!nosys || funcname != "nosys") && \ (funcname != "lkmnosys") && (funcname != "lkmressys")) { if (argc != 0 && $3 != "NOARGS" && $3 != "NOPROTO") { @@ -529,6 +539,7 @@ s/\$//g printf("#define\t%sMAXSYSCALL\t%d\n", syscallprefix, syscall) \ > syshdr printf "\tdefault:\n\t\t*n_args = 0;\n\t\tbreak;\n\t};\n}\n" > systrace + printf "\tdefault:\n\t\tbreak;\n\t};\n\tif (p != NULL)\n\t\tstrlcpy(desc, p, descsz);\n}\n" > systracetmp } ' cat $sysinc $sysent >> $syssw @@ -537,4 +548,5 @@ cat $sysarg $sysdcl \ $syscompat4 $syscompat4dcl \ $syscompat6 $syscompat6dcl \ $sysaue $sysprotoend > $sysproto +cat $systracetmp >> $systrace diff --git a/sys/kern/sched_4bsd.c b/sys/kern/sched_4bsd.c index f844e60539d2..56f69bd34af8 100644 --- a/sys/kern/sched_4bsd.c +++ b/sys/kern/sched_4bsd.c @@ -36,6 +36,7 @@ __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" +#include "opt_kdtrace.h" #include <sys/param.h> #include <sys/systm.h> @@ -60,6 +61,12 @@ __FBSDID("$FreeBSD$"); #include <sys/pmckern.h> #endif +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> +int dtrace_vtime_active; +dtrace_vtime_switch_func_t dtrace_vtime_switch_func; +#endif + /* * INVERSE_ESTCPU_WEIGHT is only suitable for statclock() frequencies in * the range 100-256 Hz (approximately). @@ -890,6 +897,15 @@ sched_switch(struct thread *td, struct thread *newtd, int flags) PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT); #endif +#ifdef KDTRACE_HOOKS + /* + * If DTrace has set the active vtime enum to anything + * other than INACTIVE (0), then it should have set the + * function to call. + */ + if (dtrace_vtime_active) + (*dtrace_vtime_switch_func)(newtd); +#endif /* I feel sleepy */ cpu_switch(td, newtd, td->td_lock); /* diff --git a/sys/kern/sched_ule.c b/sys/kern/sched_ule.c index bd8e655fcb7a..bf8394c33028 100644 --- a/sys/kern/sched_ule.c +++ b/sys/kern/sched_ule.c @@ -39,6 +39,7 @@ __FBSDID("$FreeBSD$"); #include "opt_hwpmc_hooks.h" +#include "opt_kdtrace.h" #include "opt_sched.h" #include <sys/param.h> @@ -69,6 +70,12 @@ __FBSDID("$FreeBSD$"); #include <sys/pmckern.h> #endif +#ifdef KDTRACE_HOOKS +#include <sys/dtrace_bsd.h> +int dtrace_vtime_active; +dtrace_vtime_switch_func_t dtrace_vtime_switch_func; +#endif + #include <machine/cpu.h> #include <machine/smp.h> @@ -1918,6 +1925,16 @@ sched_switch(struct thread *td, struct thread *newtd, int flags) PMC_SWITCH_CONTEXT(td, PMC_FN_CSW_OUT); #endif TDQ_LOCKPTR(tdq)->mtx_lock = (uintptr_t)newtd; + +#ifdef KDTRACE_HOOKS + /* + * If DTrace has set the active vtime enum to anything + * other than INACTIVE (0), then it should have set the + * function to call. + */ + if (dtrace_vtime_active) + (*dtrace_vtime_switch_func)(newtd); +#endif cpu_switch(td, newtd, mtx); /* * We may return from cpu_switch on a different cpu. However, diff --git a/sys/kern/subr_smp.c b/sys/kern/subr_smp.c index 75378f8dad36..375aed7eb4ab 100644 --- a/sys/kern/subr_smp.c +++ b/sys/kern/subr_smp.c @@ -74,6 +74,9 @@ u_int mp_maxid; SYSCTL_NODE(_kern, OID_AUTO, smp, CTLFLAG_RD, NULL, "Kernel SMP"); +SYSCTL_INT(_kern_smp, OID_AUTO, maxid, CTLFLAG_RD, &mp_maxid, 0, + "Max CPU ID."); + SYSCTL_INT(_kern_smp, OID_AUTO, maxcpus, CTLFLAG_RD, &mp_maxcpus, 0, "Max number of CPUs that the system was compiled for."); @@ -104,10 +107,11 @@ SYSCTL_INT(_kern_smp, OID_AUTO, forward_roundrobin_enabled, CTLFLAG_RW, "Forwarding of roundrobin to all other CPUs"); /* Variables needed for SMP rendezvous. */ -static void (*smp_rv_setup_func)(void *arg); -static void (*smp_rv_action_func)(void *arg); -static void (*smp_rv_teardown_func)(void *arg); -static void *smp_rv_func_arg; +static volatile cpumask_t smp_rv_cpumask; +static void (*volatile smp_rv_setup_func)(void *arg); +static void (*volatile smp_rv_action_func)(void *arg); +static void (*volatile smp_rv_teardown_func)(void *arg); +static void * volatile smp_rv_func_arg; static volatile int smp_rv_waiters[3]; /* @@ -286,6 +290,14 @@ restart_cpus(cpumask_t map) return 1; } +void +smp_no_rendevous_barrier(void *dummy) +{ +#ifdef SMP + KASSERT((!smp_started),("smp_no_rendevous called and smp is started")); +#endif +} + /* * All-CPU rendezvous. CPUs are signalled, all execute the setup function * (if specified), rendezvous, execute the action function (if specified), @@ -298,41 +310,57 @@ restart_cpus(cpumask_t map) void smp_rendezvous_action(void) { + cpumask_t map = smp_rv_cpumask; + int i, ncpus = 0; + void* local_func_arg = smp_rv_func_arg; + void (*local_setup_func)(void*) = smp_rv_setup_func; + void (*local_action_func)(void*) = smp_rv_action_func; + void (*local_teardown_func)(void*) = smp_rv_teardown_func; + + for (i = 0; i < MAXCPU; i++) + if (((1 << i) & map) != 0 && pcpu_find(i) != NULL) + ncpus++; /* Ensure we have up-to-date values. */ atomic_add_acq_int(&smp_rv_waiters[0], 1); - while (smp_rv_waiters[0] < mp_ncpus) + while (smp_rv_waiters[0] < ncpus) cpu_spinwait(); /* setup function */ - if (smp_rv_setup_func != NULL) - smp_rv_setup_func(smp_rv_func_arg); + if (local_setup_func != smp_no_rendevous_barrier) { + if (smp_rv_setup_func != NULL) + smp_rv_setup_func(smp_rv_func_arg); - /* spin on entry rendezvous */ - atomic_add_int(&smp_rv_waiters[1], 1); - while (smp_rv_waiters[1] < mp_ncpus) - cpu_spinwait(); + /* spin on entry rendezvous */ + atomic_add_int(&smp_rv_waiters[1], 1); + while (smp_rv_waiters[1] < ncpus) + cpu_spinwait(); + } /* action function */ - if (smp_rv_action_func != NULL) - smp_rv_action_func(smp_rv_func_arg); + if (local_action_func != NULL) + local_action_func(local_func_arg); /* spin on exit rendezvous */ atomic_add_int(&smp_rv_waiters[2], 1); - while (smp_rv_waiters[2] < mp_ncpus) + if (local_teardown_func == smp_no_rendevous_barrier) + return; + while (smp_rv_waiters[2] < ncpus) cpu_spinwait(); /* teardown function */ - if (smp_rv_teardown_func != NULL) - smp_rv_teardown_func(smp_rv_func_arg); + if (local_teardown_func != NULL) + local_teardown_func(local_func_arg); } void -smp_rendezvous(void (* setup_func)(void *), - void (* action_func)(void *), - void (* teardown_func)(void *), - void *arg) +smp_rendezvous_cpus(cpumask_t map, + void (* setup_func)(void *), + void (* action_func)(void *), + void (* teardown_func)(void *), + void *arg) { + int i, ncpus = 0; if (!smp_started) { if (setup_func != NULL) @@ -343,11 +371,16 @@ smp_rendezvous(void (* setup_func)(void *), teardown_func(arg); return; } + + for (i = 0; i < MAXCPU; i++) + if (((1 << i) & map) != 0 && pcpu_find(i) != NULL) + ncpus++; /* obtain rendezvous lock */ mtx_lock_spin(&smp_ipi_mtx); /* set static function pointers */ + smp_rv_cpumask = map & ~(1 << curcpu); smp_rv_setup_func = setup_func; smp_rv_action_func = action_func; smp_rv_teardown_func = teardown_func; @@ -357,14 +390,29 @@ smp_rendezvous(void (* setup_func)(void *), atomic_store_rel_int(&smp_rv_waiters[0], 0); /* signal other processors, which will enter the IPI with interrupts off */ - ipi_all_but_self(IPI_RENDEZVOUS); + ipi_selected(map, IPI_RENDEZVOUS); - /* call executor function */ - smp_rendezvous_action(); + /* Check if the current CPU is in the map */ + if ((map & (1 << curcpu)) != 0) + /* call executor function for the current CPU */ + smp_rendezvous_action(); + + if (teardown_func == smp_no_rendevous_barrier) + while (atomic_load_acq_int(&smp_rv_waiters[2]) < ncpus) + cpu_spinwait(); /* release lock */ mtx_unlock_spin(&smp_ipi_mtx); } + +void +smp_rendezvous(void (* setup_func)(void *), + void (* action_func)(void *), + void (* teardown_func)(void *), + void *arg) +{ + smp_rendezvous_cpus(all_cpus, setup_func, action_func, teardown_func, arg); +} #else /* !SMP */ /* @@ -383,9 +431,24 @@ SYSINIT(cpu_mp_setvariables, SI_SUB_TUNABLES, SI_ORDER_FIRST, mp_setvariables_for_up, NULL); void -smp_rendezvous(void (* setup_func)(void *), - void (* action_func)(void *), - void (* teardown_func)(void *), +smp_rendezvous_cpus(cpumask_t map, + void (*setup_func)(void *), + void (*action_func)(void *), + void (*teardown_func)(void *), + void *arg) +{ + if (setup_func != NULL) + setup_func(arg); + if (action_func != NULL) + action_func(arg); + if (teardown_func != NULL) + teardown_func(arg); +} + +void +smp_rendezvous(void (*setup_func)(void *), + void (*action_func)(void *), + void (*teardown_func)(void *), void *arg) { diff --git a/sys/kern/systrace_args.c b/sys/kern/systrace_args.c index 917c21870f2d..de89d3f03cf0 100644 --- a/sys/kern/systrace_args.c +++ b/sys/kern/systrace_args.c @@ -2941,3 +2941,4720 @@ systrace_args(int sysnum, void *params, u_int64_t *uarg, int *n_args) break; }; } +static void +systrace_setargdesc(int sysnum, int ndx, char *desc, size_t descsz) +{ + const char *p = NULL; + switch (sysnum) { + /* nosys */ + case 0: + break; + /* sys_exit */ + case 1: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* fork */ + case 2: + break; + /* read */ + case 3: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "void *"; + break; + case 2: + p = "size_t"; + break; + default: + break; + }; + break; + /* write */ + case 4: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "const void *"; + break; + case 2: + p = "size_t"; + break; + default: + break; + }; + break; + /* open */ + case 5: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* close */ + case 6: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* wait4 */ + case 7: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int *"; + break; + case 2: + p = "int"; + break; + case 3: + p = "struct rusage *"; + break; + default: + break; + }; + break; + /* link */ + case 9: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "char *"; + break; + default: + break; + }; + break; + /* unlink */ + case 10: + switch(ndx) { + case 0: + p = "char *"; + break; + default: + break; + }; + break; + /* chdir */ + case 12: + switch(ndx) { + case 0: + p = "char *"; + break; + default: + break; + }; + break; + /* fchdir */ + case 13: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* mknod */ + case 14: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* chmod */ + case 15: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* chown */ + case 16: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* obreak */ + case 17: + switch(ndx) { + case 0: + p = "char *"; + break; + default: + break; + }; + break; + /* getpid */ + case 20: + break; + /* mount */ + case 21: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "char *"; + break; + case 2: + p = "int"; + break; + case 3: + p = "caddr_t"; + break; + default: + break; + }; + break; + /* unmount */ + case 22: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* setuid */ + case 23: + switch(ndx) { + case 0: + p = "uid_t"; + break; + default: + break; + }; + break; + /* getuid */ + case 24: + break; + /* geteuid */ + case 25: + break; + /* ptrace */ + case 26: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "pid_t"; + break; + case 2: + p = "caddr_t"; + break; + case 3: + p = "int"; + break; + default: + break; + }; + break; + /* recvmsg */ + case 27: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct msghdr *"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* sendmsg */ + case 28: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct msghdr *"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* recvfrom */ + case 29: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "caddr_t"; + break; + case 2: + p = "size_t"; + break; + case 3: + p = "int"; + break; + case 4: + p = "struct sockaddr *__restrict"; + break; + case 5: + p = "__socklen_t *__restrict"; + break; + default: + break; + }; + break; + /* accept */ + case 30: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct sockaddr *__restrict"; + break; + case 2: + p = "__socklen_t *__restrict"; + break; + default: + break; + }; + break; + /* getpeername */ + case 31: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct sockaddr *__restrict"; + break; + case 2: + p = "__socklen_t *__restrict"; + break; + default: + break; + }; + break; + /* getsockname */ + case 32: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct sockaddr *__restrict"; + break; + case 2: + p = "__socklen_t *__restrict"; + break; + default: + break; + }; + break; + /* access */ + case 33: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* chflags */ + case 34: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* fchflags */ + case 35: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* sync */ + case 36: + break; + /* kill */ + case 37: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* getppid */ + case 39: + break; + /* dup */ + case 41: + switch(ndx) { + case 0: + p = "u_int"; + break; + default: + break; + }; + break; + /* pipe */ + case 42: + break; + /* getegid */ + case 43: + break; + /* profil */ + case 44: + switch(ndx) { + case 0: + p = "caddr_t"; + break; + case 1: + p = "size_t"; + break; + case 2: + p = "size_t"; + break; + case 3: + p = "u_int"; + break; + default: + break; + }; + break; + /* ktrace */ + case 45: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + case 3: + p = "int"; + break; + default: + break; + }; + break; + /* getgid */ + case 47: + break; + /* getlogin */ + case 49: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "u_int"; + break; + default: + break; + }; + break; + /* setlogin */ + case 50: + switch(ndx) { + case 0: + p = "char *"; + break; + default: + break; + }; + break; + /* acct */ + case 51: + switch(ndx) { + case 0: + p = "char *"; + break; + default: + break; + }; + break; + /* sigaltstack */ + case 53: + switch(ndx) { + case 0: + p = "stack_t *"; + break; + case 1: + p = "stack_t *"; + break; + default: + break; + }; + break; + /* ioctl */ + case 54: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "u_long"; + break; + case 2: + p = "caddr_t"; + break; + default: + break; + }; + break; + /* reboot */ + case 55: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* revoke */ + case 56: + switch(ndx) { + case 0: + p = "char *"; + break; + default: + break; + }; + break; + /* symlink */ + case 57: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "char *"; + break; + default: + break; + }; + break; + /* readlink */ + case 58: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "char *"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* execve */ + case 59: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "char **"; + break; + case 2: + p = "char **"; + break; + default: + break; + }; + break; + /* umask */ + case 60: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* chroot */ + case 61: + switch(ndx) { + case 0: + p = "char *"; + break; + default: + break; + }; + break; + /* msync */ + case 65: + switch(ndx) { + case 0: + p = "void *"; + break; + case 1: + p = "size_t"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* vfork */ + case 66: + break; + /* sbrk */ + case 69: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* sstk */ + case 70: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* ovadvise */ + case 72: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* munmap */ + case 73: + switch(ndx) { + case 0: + p = "void *"; + break; + case 1: + p = "size_t"; + break; + default: + break; + }; + break; + /* mprotect */ + case 74: + switch(ndx) { + case 0: + p = "const void *"; + break; + case 1: + p = "size_t"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* madvise */ + case 75: + switch(ndx) { + case 0: + p = "void *"; + break; + case 1: + p = "size_t"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* mincore */ + case 78: + switch(ndx) { + case 0: + p = "const void *"; + break; + case 1: + p = "size_t"; + break; + case 2: + p = "char *"; + break; + default: + break; + }; + break; + /* getgroups */ + case 79: + switch(ndx) { + case 0: + p = "u_int"; + break; + case 1: + p = "gid_t *"; + break; + default: + break; + }; + break; + /* setgroups */ + case 80: + switch(ndx) { + case 0: + p = "u_int"; + break; + case 1: + p = "gid_t *"; + break; + default: + break; + }; + break; + /* getpgrp */ + case 81: + break; + /* setpgid */ + case 82: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* setitimer */ + case 83: + switch(ndx) { + case 0: + p = "u_int"; + break; + case 1: + p = "struct itimerval *"; + break; + case 2: + p = "struct itimerval *"; + break; + default: + break; + }; + break; + /* swapon */ + case 85: + switch(ndx) { + case 0: + p = "char *"; + break; + default: + break; + }; + break; + /* getitimer */ + case 86: + switch(ndx) { + case 0: + p = "u_int"; + break; + case 1: + p = "struct itimerval *"; + break; + default: + break; + }; + break; + /* getdtablesize */ + case 89: + break; + /* dup2 */ + case 90: + switch(ndx) { + case 0: + p = "u_int"; + break; + case 1: + p = "u_int"; + break; + default: + break; + }; + break; + /* fcntl */ + case 92: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "long"; + break; + default: + break; + }; + break; + /* select */ + case 93: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "fd_set *"; + break; + case 2: + p = "fd_set *"; + break; + case 3: + p = "fd_set *"; + break; + case 4: + p = "struct timeval *"; + break; + default: + break; + }; + break; + /* fsync */ + case 95: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* setpriority */ + case 96: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* socket */ + case 97: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* connect */ + case 98: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "caddr_t"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* getpriority */ + case 100: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* bind */ + case 104: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "caddr_t"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* setsockopt */ + case 105: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + case 3: + p = "caddr_t"; + break; + case 4: + p = "int"; + break; + default: + break; + }; + break; + /* listen */ + case 106: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* gettimeofday */ + case 116: + switch(ndx) { + case 0: + p = "struct timeval *"; + break; + case 1: + p = "struct timezone *"; + break; + default: + break; + }; + break; + /* getrusage */ + case 117: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct rusage *"; + break; + default: + break; + }; + break; + /* getsockopt */ + case 118: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + case 3: + p = "caddr_t"; + break; + case 4: + p = "int *"; + break; + default: + break; + }; + break; + /* readv */ + case 120: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct iovec *"; + break; + case 2: + p = "u_int"; + break; + default: + break; + }; + break; + /* writev */ + case 121: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct iovec *"; + break; + case 2: + p = "u_int"; + break; + default: + break; + }; + break; + /* settimeofday */ + case 122: + switch(ndx) { + case 0: + p = "struct timeval *"; + break; + case 1: + p = "struct timezone *"; + break; + default: + break; + }; + break; + /* fchown */ + case 123: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* fchmod */ + case 124: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* setreuid */ + case 126: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* setregid */ + case 127: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* rename */ + case 128: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "char *"; + break; + default: + break; + }; + break; + /* flock */ + case 131: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* mkfifo */ + case 132: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* sendto */ + case 133: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "caddr_t"; + break; + case 2: + p = "size_t"; + break; + case 3: + p = "int"; + break; + case 4: + p = "caddr_t"; + break; + case 5: + p = "int"; + break; + default: + break; + }; + break; + /* shutdown */ + case 134: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* socketpair */ + case 135: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + case 3: + p = "int *"; + break; + default: + break; + }; + break; + /* mkdir */ + case 136: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* rmdir */ + case 137: + switch(ndx) { + case 0: + p = "char *"; + break; + default: + break; + }; + break; + /* utimes */ + case 138: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "struct timeval *"; + break; + default: + break; + }; + break; + /* adjtime */ + case 140: + switch(ndx) { + case 0: + p = "struct timeval *"; + break; + case 1: + p = "struct timeval *"; + break; + default: + break; + }; + break; + /* setsid */ + case 147: + break; + /* quotactl */ + case 148: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + case 3: + p = "caddr_t"; + break; + default: + break; + }; + break; + /* nlm_syscall */ + case 154: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + case 3: + p = "char **"; + break; + default: + break; + }; + break; + /* nfssvc */ + case 155: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "caddr_t"; + break; + default: + break; + }; + break; + /* lgetfh */ + case 160: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "struct fhandle *"; + break; + default: + break; + }; + break; + /* getfh */ + case 161: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "struct fhandle *"; + break; + default: + break; + }; + break; + /* getdomainname */ + case 162: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* setdomainname */ + case 163: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* uname */ + case 164: + switch(ndx) { + case 0: + p = "struct utsname *"; + break; + default: + break; + }; + break; + /* sysarch */ + case 165: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "char *"; + break; + default: + break; + }; + break; + /* rtprio */ + case 166: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "pid_t"; + break; + case 2: + p = "struct rtprio *"; + break; + default: + break; + }; + break; + /* semsys */ + case 169: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + case 3: + p = "int"; + break; + case 4: + p = "int"; + break; + default: + break; + }; + break; + /* msgsys */ + case 170: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + case 3: + p = "int"; + break; + case 4: + p = "int"; + break; + case 5: + p = "int"; + break; + default: + break; + }; + break; + /* shmsys */ + case 171: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + case 3: + p = "int"; + break; + default: + break; + }; + break; + /* freebsd6_pread */ + case 173: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "void *"; + break; + case 2: + p = "size_t"; + break; + case 3: + p = "int"; + break; + case 4: + p = "off_t"; + break; + default: + break; + }; + break; + /* freebsd6_pwrite */ + case 174: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "const void *"; + break; + case 2: + p = "size_t"; + break; + case 3: + p = "int"; + break; + case 4: + p = "off_t"; + break; + default: + break; + }; + break; + /* ntp_adjtime */ + case 176: + switch(ndx) { + case 0: + p = "struct timex *"; + break; + default: + break; + }; + break; + /* setgid */ + case 181: + switch(ndx) { + case 0: + p = "gid_t"; + break; + default: + break; + }; + break; + /* setegid */ + case 182: + switch(ndx) { + case 0: + p = "gid_t"; + break; + default: + break; + }; + break; + /* seteuid */ + case 183: + switch(ndx) { + case 0: + p = "uid_t"; + break; + default: + break; + }; + break; + /* stat */ + case 188: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "struct stat *"; + break; + default: + break; + }; + break; + /* fstat */ + case 189: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct stat *"; + break; + default: + break; + }; + break; + /* lstat */ + case 190: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "struct stat *"; + break; + default: + break; + }; + break; + /* pathconf */ + case 191: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* fpathconf */ + case 192: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* getrlimit */ + case 194: + switch(ndx) { + case 0: + p = "u_int"; + break; + case 1: + p = "struct rlimit *"; + break; + default: + break; + }; + break; + /* setrlimit */ + case 195: + switch(ndx) { + case 0: + p = "u_int"; + break; + case 1: + p = "struct rlimit *"; + break; + default: + break; + }; + break; + /* getdirentries */ + case 196: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "char *"; + break; + case 2: + p = "u_int"; + break; + case 3: + p = "long *"; + break; + default: + break; + }; + break; + /* freebsd6_mmap */ + case 197: + switch(ndx) { + case 0: + p = "caddr_t"; + break; + case 1: + p = "size_t"; + break; + case 2: + p = "int"; + break; + case 3: + p = "int"; + break; + case 4: + p = "int"; + break; + case 5: + p = "int"; + break; + case 6: + p = "off_t"; + break; + default: + break; + }; + break; + /* nosys */ + case 198: + break; + /* freebsd6_lseek */ + case 199: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "off_t"; + break; + case 3: + p = "int"; + break; + default: + break; + }; + break; + /* freebsd6_truncate */ + case 200: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "off_t"; + break; + default: + break; + }; + break; + /* freebsd6_ftruncate */ + case 201: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "off_t"; + break; + default: + break; + }; + break; + /* __sysctl */ + case 202: + switch(ndx) { + case 0: + p = "int *"; + break; + case 1: + p = "u_int"; + break; + case 2: + p = "void *"; + break; + case 3: + p = "size_t *"; + break; + case 4: + p = "void *"; + break; + case 5: + p = "size_t"; + break; + default: + break; + }; + break; + /* mlock */ + case 203: + switch(ndx) { + case 0: + p = "const void *"; + break; + case 1: + p = "size_t"; + break; + default: + break; + }; + break; + /* munlock */ + case 204: + switch(ndx) { + case 0: + p = "const void *"; + break; + case 1: + p = "size_t"; + break; + default: + break; + }; + break; + /* undelete */ + case 205: + switch(ndx) { + case 0: + p = "char *"; + break; + default: + break; + }; + break; + /* futimes */ + case 206: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct timeval *"; + break; + default: + break; + }; + break; + /* getpgid */ + case 207: + switch(ndx) { + case 0: + p = "pid_t"; + break; + default: + break; + }; + break; + /* poll */ + case 209: + switch(ndx) { + case 0: + p = "struct pollfd *"; + break; + case 1: + p = "u_int"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* lkmnosys */ + case 210: + break; + /* lkmnosys */ + case 211: + break; + /* lkmnosys */ + case 212: + break; + /* lkmnosys */ + case 213: + break; + /* lkmnosys */ + case 214: + break; + /* lkmnosys */ + case 215: + break; + /* lkmnosys */ + case 216: + break; + /* lkmnosys */ + case 217: + break; + /* lkmnosys */ + case 218: + break; + /* lkmnosys */ + case 219: + break; + /* __semctl */ + case 220: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + case 3: + p = "union semun *"; + break; + default: + break; + }; + break; + /* semget */ + case 221: + switch(ndx) { + case 0: + p = "key_t"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* semop */ + case 222: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct sembuf *"; + break; + case 2: + p = "size_t"; + break; + default: + break; + }; + break; + /* msgctl */ + case 224: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "struct msqid_ds *"; + break; + default: + break; + }; + break; + /* msgget */ + case 225: + switch(ndx) { + case 0: + p = "key_t"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* msgsnd */ + case 226: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "const void *"; + break; + case 2: + p = "size_t"; + break; + case 3: + p = "int"; + break; + default: + break; + }; + break; + /* msgrcv */ + case 227: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "void *"; + break; + case 2: + p = "size_t"; + break; + case 3: + p = "long"; + break; + case 4: + p = "int"; + break; + default: + break; + }; + break; + /* shmat */ + case 228: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "const void *"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* shmctl */ + case 229: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "struct shmid_ds *"; + break; + default: + break; + }; + break; + /* shmdt */ + case 230: + switch(ndx) { + case 0: + p = "const void *"; + break; + default: + break; + }; + break; + /* shmget */ + case 231: + switch(ndx) { + case 0: + p = "key_t"; + break; + case 1: + p = "size_t"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* clock_gettime */ + case 232: + switch(ndx) { + case 0: + p = "clockid_t"; + break; + case 1: + p = "struct timespec *"; + break; + default: + break; + }; + break; + /* clock_settime */ + case 233: + switch(ndx) { + case 0: + p = "clockid_t"; + break; + case 1: + p = "const struct timespec *"; + break; + default: + break; + }; + break; + /* clock_getres */ + case 234: + switch(ndx) { + case 0: + p = "clockid_t"; + break; + case 1: + p = "struct timespec *"; + break; + default: + break; + }; + break; + /* ktimer_create */ + case 235: + switch(ndx) { + case 0: + p = "clockid_t"; + break; + case 1: + p = "struct sigevent *"; + break; + case 2: + p = "int *"; + break; + default: + break; + }; + break; + /* ktimer_delete */ + case 236: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* ktimer_settime */ + case 237: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const struct itimerspec *"; + break; + case 3: + p = "struct itimerspec *"; + break; + default: + break; + }; + break; + /* ktimer_gettime */ + case 238: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct itimerspec *"; + break; + default: + break; + }; + break; + /* ktimer_getoverrun */ + case 239: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* nanosleep */ + case 240: + switch(ndx) { + case 0: + p = "const struct timespec *"; + break; + case 1: + p = "struct timespec *"; + break; + default: + break; + }; + break; + /* ntp_gettime */ + case 248: + switch(ndx) { + case 0: + p = "struct ntptimeval *"; + break; + default: + break; + }; + break; + /* minherit */ + case 250: + switch(ndx) { + case 0: + p = "void *"; + break; + case 1: + p = "size_t"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* rfork */ + case 251: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* openbsd_poll */ + case 252: + switch(ndx) { + case 0: + p = "struct pollfd *"; + break; + case 1: + p = "u_int"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* issetugid */ + case 253: + break; + /* lchown */ + case 254: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* aio_read */ + case 255: + switch(ndx) { + case 0: + p = "struct aiocb *"; + break; + default: + break; + }; + break; + /* aio_write */ + case 256: + switch(ndx) { + case 0: + p = "struct aiocb *"; + break; + default: + break; + }; + break; + /* lio_listio */ + case 257: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct aiocb *const *"; + break; + case 2: + p = "int"; + break; + case 3: + p = "struct sigevent *"; + break; + default: + break; + }; + break; + /* getdents */ + case 272: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "char *"; + break; + case 2: + p = "size_t"; + break; + default: + break; + }; + break; + /* lchmod */ + case 274: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "mode_t"; + break; + default: + break; + }; + break; + /* lchown */ + case 275: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "uid_t"; + break; + case 2: + p = "gid_t"; + break; + default: + break; + }; + break; + /* lutimes */ + case 276: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "struct timeval *"; + break; + default: + break; + }; + break; + /* msync */ + case 277: + switch(ndx) { + case 0: + p = "void *"; + break; + case 1: + p = "size_t"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* nstat */ + case 278: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "struct nstat *"; + break; + default: + break; + }; + break; + /* nfstat */ + case 279: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct nstat *"; + break; + default: + break; + }; + break; + /* nlstat */ + case 280: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "struct nstat *"; + break; + default: + break; + }; + break; + /* preadv */ + case 289: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct iovec *"; + break; + case 2: + p = "u_int"; + break; + case 3: + p = "off_t"; + break; + default: + break; + }; + break; + /* pwritev */ + case 290: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct iovec *"; + break; + case 2: + p = "u_int"; + break; + case 3: + p = "off_t"; + break; + default: + break; + }; + break; + /* fhopen */ + case 298: + switch(ndx) { + case 0: + p = "const struct fhandle *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* fhstat */ + case 299: + switch(ndx) { + case 0: + p = "const struct fhandle *"; + break; + case 1: + p = "struct stat *"; + break; + default: + break; + }; + break; + /* modnext */ + case 300: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* modstat */ + case 301: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct module_stat *"; + break; + default: + break; + }; + break; + /* modfnext */ + case 302: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* modfind */ + case 303: + switch(ndx) { + case 0: + p = "const char *"; + break; + default: + break; + }; + break; + /* kldload */ + case 304: + switch(ndx) { + case 0: + p = "const char *"; + break; + default: + break; + }; + break; + /* kldunload */ + case 305: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* kldfind */ + case 306: + switch(ndx) { + case 0: + p = "const char *"; + break; + default: + break; + }; + break; + /* kldnext */ + case 307: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* kldstat */ + case 308: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct kld_file_stat *"; + break; + default: + break; + }; + break; + /* kldfirstmod */ + case 309: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* getsid */ + case 310: + switch(ndx) { + case 0: + p = "pid_t"; + break; + default: + break; + }; + break; + /* setresuid */ + case 311: + switch(ndx) { + case 0: + p = "uid_t"; + break; + case 1: + p = "uid_t"; + break; + case 2: + p = "uid_t"; + break; + default: + break; + }; + break; + /* setresgid */ + case 312: + switch(ndx) { + case 0: + p = "gid_t"; + break; + case 1: + p = "gid_t"; + break; + case 2: + p = "gid_t"; + break; + default: + break; + }; + break; + /* aio_return */ + case 314: + switch(ndx) { + case 0: + p = "struct aiocb *"; + break; + default: + break; + }; + break; + /* aio_suspend */ + case 315: + switch(ndx) { + case 0: + p = "struct aiocb *const *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const struct timespec *"; + break; + default: + break; + }; + break; + /* aio_cancel */ + case 316: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct aiocb *"; + break; + default: + break; + }; + break; + /* aio_error */ + case 317: + switch(ndx) { + case 0: + p = "struct aiocb *"; + break; + default: + break; + }; + break; + /* oaio_read */ + case 318: + switch(ndx) { + case 0: + p = "struct oaiocb *"; + break; + default: + break; + }; + break; + /* oaio_write */ + case 319: + switch(ndx) { + case 0: + p = "struct oaiocb *"; + break; + default: + break; + }; + break; + /* olio_listio */ + case 320: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct oaiocb *const *"; + break; + case 2: + p = "int"; + break; + case 3: + p = "struct osigevent *"; + break; + default: + break; + }; + break; + /* yield */ + case 321: + break; + /* mlockall */ + case 324: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* munlockall */ + case 325: + break; + /* __getcwd */ + case 326: + switch(ndx) { + case 0: + p = "u_char *"; + break; + case 1: + p = "u_int"; + break; + default: + break; + }; + break; + /* sched_setparam */ + case 327: + switch(ndx) { + case 0: + p = "pid_t"; + break; + case 1: + p = "const struct sched_param *"; + break; + default: + break; + }; + break; + /* sched_getparam */ + case 328: + switch(ndx) { + case 0: + p = "pid_t"; + break; + case 1: + p = "struct sched_param *"; + break; + default: + break; + }; + break; + /* sched_setscheduler */ + case 329: + switch(ndx) { + case 0: + p = "pid_t"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const struct sched_param *"; + break; + default: + break; + }; + break; + /* sched_getscheduler */ + case 330: + switch(ndx) { + case 0: + p = "pid_t"; + break; + default: + break; + }; + break; + /* sched_yield */ + case 331: + break; + /* sched_get_priority_max */ + case 332: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* sched_get_priority_min */ + case 333: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* sched_rr_get_interval */ + case 334: + switch(ndx) { + case 0: + p = "pid_t"; + break; + case 1: + p = "struct timespec *"; + break; + default: + break; + }; + break; + /* utrace */ + case 335: + switch(ndx) { + case 0: + p = "const void *"; + break; + case 1: + p = "size_t"; + break; + default: + break; + }; + break; + /* kldsym */ + case 337: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "void *"; + break; + default: + break; + }; + break; + /* jail */ + case 338: + switch(ndx) { + case 0: + p = "struct jail *"; + break; + default: + break; + }; + break; + /* sigprocmask */ + case 340: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "const sigset_t *"; + break; + case 2: + p = "sigset_t *"; + break; + default: + break; + }; + break; + /* sigsuspend */ + case 341: + switch(ndx) { + case 0: + p = "const sigset_t *"; + break; + default: + break; + }; + break; + /* sigpending */ + case 343: + switch(ndx) { + case 0: + p = "sigset_t *"; + break; + default: + break; + }; + break; + /* sigtimedwait */ + case 345: + switch(ndx) { + case 0: + p = "const sigset_t *"; + break; + case 1: + p = "siginfo_t *"; + break; + case 2: + p = "const struct timespec *"; + break; + default: + break; + }; + break; + /* sigwaitinfo */ + case 346: + switch(ndx) { + case 0: + p = "const sigset_t *"; + break; + case 1: + p = "siginfo_t *"; + break; + default: + break; + }; + break; + /* __acl_get_file */ + case 347: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "acl_type_t"; + break; + case 2: + p = "struct acl *"; + break; + default: + break; + }; + break; + /* __acl_set_file */ + case 348: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "acl_type_t"; + break; + case 2: + p = "struct acl *"; + break; + default: + break; + }; + break; + /* __acl_get_fd */ + case 349: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "acl_type_t"; + break; + case 2: + p = "struct acl *"; + break; + default: + break; + }; + break; + /* __acl_set_fd */ + case 350: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "acl_type_t"; + break; + case 2: + p = "struct acl *"; + break; + default: + break; + }; + break; + /* __acl_delete_file */ + case 351: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "acl_type_t"; + break; + default: + break; + }; + break; + /* __acl_delete_fd */ + case 352: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "acl_type_t"; + break; + default: + break; + }; + break; + /* __acl_aclcheck_file */ + case 353: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "acl_type_t"; + break; + case 2: + p = "struct acl *"; + break; + default: + break; + }; + break; + /* __acl_aclcheck_fd */ + case 354: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "acl_type_t"; + break; + case 2: + p = "struct acl *"; + break; + default: + break; + }; + break; + /* extattrctl */ + case 355: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const char *"; + break; + case 3: + p = "int"; + break; + case 4: + p = "const char *"; + break; + default: + break; + }; + break; + /* extattr_set_file */ + case 356: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const char *"; + break; + case 3: + p = "void *"; + break; + case 4: + p = "size_t"; + break; + default: + break; + }; + break; + /* extattr_get_file */ + case 357: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const char *"; + break; + case 3: + p = "void *"; + break; + case 4: + p = "size_t"; + break; + default: + break; + }; + break; + /* extattr_delete_file */ + case 358: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const char *"; + break; + default: + break; + }; + break; + /* aio_waitcomplete */ + case 359: + switch(ndx) { + case 0: + p = "struct aiocb **"; + break; + case 1: + p = "struct timespec *"; + break; + default: + break; + }; + break; + /* getresuid */ + case 360: + switch(ndx) { + case 0: + p = "uid_t *"; + break; + case 1: + p = "uid_t *"; + break; + case 2: + p = "uid_t *"; + break; + default: + break; + }; + break; + /* getresgid */ + case 361: + switch(ndx) { + case 0: + p = "gid_t *"; + break; + case 1: + p = "gid_t *"; + break; + case 2: + p = "gid_t *"; + break; + default: + break; + }; + break; + /* kqueue */ + case 362: + break; + /* kevent */ + case 363: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct kevent *"; + break; + case 2: + p = "int"; + break; + case 3: + p = "struct kevent *"; + break; + case 4: + p = "int"; + break; + case 5: + p = "const struct timespec *"; + break; + default: + break; + }; + break; + /* lkmressys */ + case 370: + break; + /* extattr_set_fd */ + case 371: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const char *"; + break; + case 3: + p = "void *"; + break; + case 4: + p = "size_t"; + break; + default: + break; + }; + break; + /* extattr_get_fd */ + case 372: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const char *"; + break; + case 3: + p = "void *"; + break; + case 4: + p = "size_t"; + break; + default: + break; + }; + break; + /* extattr_delete_fd */ + case 373: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const char *"; + break; + default: + break; + }; + break; + /* __setugid */ + case 374: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* nfsclnt */ + case 375: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "caddr_t"; + break; + default: + break; + }; + break; + /* eaccess */ + case 376: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* nmount */ + case 378: + switch(ndx) { + case 0: + p = "struct iovec *"; + break; + case 1: + p = "unsigned int"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* kse_exit */ + case 379: + break; + /* kse_wakeup */ + case 380: + switch(ndx) { + case 0: + p = "struct kse_mailbox *"; + break; + default: + break; + }; + break; + /* kse_create */ + case 381: + switch(ndx) { + case 0: + p = "struct kse_mailbox *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* kse_thr_interrupt */ + case 382: + switch(ndx) { + case 0: + p = "struct kse_thr_mailbox *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "long"; + break; + default: + break; + }; + break; + /* kse_release */ + case 383: + switch(ndx) { + case 0: + p = "struct timespec *"; + break; + default: + break; + }; + break; + /* __mac_get_proc */ + case 384: + switch(ndx) { + case 0: + p = "struct mac *"; + break; + default: + break; + }; + break; + /* __mac_set_proc */ + case 385: + switch(ndx) { + case 0: + p = "struct mac *"; + break; + default: + break; + }; + break; + /* __mac_get_fd */ + case 386: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct mac *"; + break; + default: + break; + }; + break; + /* __mac_get_file */ + case 387: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "struct mac *"; + break; + default: + break; + }; + break; + /* __mac_set_fd */ + case 388: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct mac *"; + break; + default: + break; + }; + break; + /* __mac_set_file */ + case 389: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "struct mac *"; + break; + default: + break; + }; + break; + /* kenv */ + case 390: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "const char *"; + break; + case 2: + p = "char *"; + break; + case 3: + p = "int"; + break; + default: + break; + }; + break; + /* lchflags */ + case 391: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* uuidgen */ + case 392: + switch(ndx) { + case 0: + p = "struct uuid *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* sendfile */ + case 393: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "off_t"; + break; + case 3: + p = "size_t"; + break; + case 4: + p = "struct sf_hdtr *"; + break; + case 5: + p = "off_t *"; + break; + case 6: + p = "int"; + break; + default: + break; + }; + break; + /* mac_syscall */ + case 394: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "void *"; + break; + default: + break; + }; + break; + /* getfsstat */ + case 395: + switch(ndx) { + case 0: + p = "struct statfs *"; + break; + case 1: + p = "long"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* statfs */ + case 396: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "struct statfs *"; + break; + default: + break; + }; + break; + /* fstatfs */ + case 397: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct statfs *"; + break; + default: + break; + }; + break; + /* fhstatfs */ + case 398: + switch(ndx) { + case 0: + p = "const struct fhandle *"; + break; + case 1: + p = "struct statfs *"; + break; + default: + break; + }; + break; + /* ksem_close */ + case 400: + switch(ndx) { + case 0: + p = "semid_t"; + break; + default: + break; + }; + break; + /* ksem_post */ + case 401: + switch(ndx) { + case 0: + p = "semid_t"; + break; + default: + break; + }; + break; + /* ksem_wait */ + case 402: + switch(ndx) { + case 0: + p = "semid_t"; + break; + default: + break; + }; + break; + /* ksem_trywait */ + case 403: + switch(ndx) { + case 0: + p = "semid_t"; + break; + default: + break; + }; + break; + /* ksem_init */ + case 404: + switch(ndx) { + case 0: + p = "semid_t *"; + break; + case 1: + p = "unsigned int"; + break; + default: + break; + }; + break; + /* ksem_open */ + case 405: + switch(ndx) { + case 0: + p = "semid_t *"; + break; + case 1: + p = "const char *"; + break; + case 2: + p = "int"; + break; + case 3: + p = "mode_t"; + break; + case 4: + p = "unsigned int"; + break; + default: + break; + }; + break; + /* ksem_unlink */ + case 406: + switch(ndx) { + case 0: + p = "const char *"; + break; + default: + break; + }; + break; + /* ksem_getvalue */ + case 407: + switch(ndx) { + case 0: + p = "semid_t"; + break; + case 1: + p = "int *"; + break; + default: + break; + }; + break; + /* ksem_destroy */ + case 408: + switch(ndx) { + case 0: + p = "semid_t"; + break; + default: + break; + }; + break; + /* __mac_get_pid */ + case 409: + switch(ndx) { + case 0: + p = "pid_t"; + break; + case 1: + p = "struct mac *"; + break; + default: + break; + }; + break; + /* __mac_get_link */ + case 410: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "struct mac *"; + break; + default: + break; + }; + break; + /* __mac_set_link */ + case 411: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "struct mac *"; + break; + default: + break; + }; + break; + /* extattr_set_link */ + case 412: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const char *"; + break; + case 3: + p = "void *"; + break; + case 4: + p = "size_t"; + break; + default: + break; + }; + break; + /* extattr_get_link */ + case 413: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const char *"; + break; + case 3: + p = "void *"; + break; + case 4: + p = "size_t"; + break; + default: + break; + }; + break; + /* extattr_delete_link */ + case 414: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "const char *"; + break; + default: + break; + }; + break; + /* __mac_execve */ + case 415: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "char **"; + break; + case 2: + p = "char **"; + break; + case 3: + p = "struct mac *"; + break; + default: + break; + }; + break; + /* sigaction */ + case 416: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "const struct sigaction *"; + break; + case 2: + p = "struct sigaction *"; + break; + default: + break; + }; + break; + /* sigreturn */ + case 417: + switch(ndx) { + case 0: + p = "const struct __ucontext *"; + break; + default: + break; + }; + break; + /* getcontext */ + case 421: + switch(ndx) { + case 0: + p = "struct __ucontext *"; + break; + default: + break; + }; + break; + /* setcontext */ + case 422: + switch(ndx) { + case 0: + p = "const struct __ucontext *"; + break; + default: + break; + }; + break; + /* swapcontext */ + case 423: + switch(ndx) { + case 0: + p = "struct __ucontext *"; + break; + case 1: + p = "const struct __ucontext *"; + break; + default: + break; + }; + break; + /* swapoff */ + case 424: + switch(ndx) { + case 0: + p = "const char *"; + break; + default: + break; + }; + break; + /* __acl_get_link */ + case 425: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "acl_type_t"; + break; + case 2: + p = "struct acl *"; + break; + default: + break; + }; + break; + /* __acl_set_link */ + case 426: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "acl_type_t"; + break; + case 2: + p = "struct acl *"; + break; + default: + break; + }; + break; + /* __acl_delete_link */ + case 427: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "acl_type_t"; + break; + default: + break; + }; + break; + /* __acl_aclcheck_link */ + case 428: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "acl_type_t"; + break; + case 2: + p = "struct acl *"; + break; + default: + break; + }; + break; + /* sigwait */ + case 429: + switch(ndx) { + case 0: + p = "const sigset_t *"; + break; + case 1: + p = "int *"; + break; + default: + break; + }; + break; + /* thr_create */ + case 430: + switch(ndx) { + case 0: + p = "ucontext_t *"; + break; + case 1: + p = "long *"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* thr_exit */ + case 431: + switch(ndx) { + case 0: + p = "long *"; + break; + default: + break; + }; + break; + /* thr_self */ + case 432: + switch(ndx) { + case 0: + p = "long *"; + break; + default: + break; + }; + break; + /* thr_kill */ + case 433: + switch(ndx) { + case 0: + p = "long"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* _umtx_lock */ + case 434: + switch(ndx) { + case 0: + p = "struct umtx *"; + break; + default: + break; + }; + break; + /* _umtx_unlock */ + case 435: + switch(ndx) { + case 0: + p = "struct umtx *"; + break; + default: + break; + }; + break; + /* jail_attach */ + case 436: + switch(ndx) { + case 0: + p = "int"; + break; + default: + break; + }; + break; + /* extattr_list_fd */ + case 437: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + case 2: + p = "void *"; + break; + case 3: + p = "size_t"; + break; + default: + break; + }; + break; + /* extattr_list_file */ + case 438: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "void *"; + break; + case 3: + p = "size_t"; + break; + default: + break; + }; + break; + /* extattr_list_link */ + case 439: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "void *"; + break; + case 3: + p = "size_t"; + break; + default: + break; + }; + break; + /* kse_switchin */ + case 440: + switch(ndx) { + case 0: + p = "struct kse_thr_mailbox *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* ksem_timedwait */ + case 441: + switch(ndx) { + case 0: + p = "semid_t"; + break; + case 1: + p = "const struct timespec *"; + break; + default: + break; + }; + break; + /* thr_suspend */ + case 442: + switch(ndx) { + case 0: + p = "const struct timespec *"; + break; + default: + break; + }; + break; + /* thr_wake */ + case 443: + switch(ndx) { + case 0: + p = "long"; + break; + default: + break; + }; + break; + /* kldunloadf */ + case 444: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* audit */ + case 445: + switch(ndx) { + case 0: + p = "const void *"; + break; + case 1: + p = "u_int"; + break; + default: + break; + }; + break; + /* auditon */ + case 446: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "void *"; + break; + case 2: + p = "u_int"; + break; + default: + break; + }; + break; + /* getauid */ + case 447: + switch(ndx) { + case 0: + p = "uid_t *"; + break; + default: + break; + }; + break; + /* setauid */ + case 448: + switch(ndx) { + case 0: + p = "uid_t *"; + break; + default: + break; + }; + break; + /* getaudit */ + case 449: + switch(ndx) { + case 0: + p = "struct auditinfo *"; + break; + default: + break; + }; + break; + /* setaudit */ + case 450: + switch(ndx) { + case 0: + p = "struct auditinfo *"; + break; + default: + break; + }; + break; + /* getaudit_addr */ + case 451: + switch(ndx) { + case 0: + p = "struct auditinfo_addr *"; + break; + case 1: + p = "u_int"; + break; + default: + break; + }; + break; + /* setaudit_addr */ + case 452: + switch(ndx) { + case 0: + p = "struct auditinfo_addr *"; + break; + case 1: + p = "u_int"; + break; + default: + break; + }; + break; + /* auditctl */ + case 453: + switch(ndx) { + case 0: + p = "char *"; + break; + default: + break; + }; + break; + /* _umtx_op */ + case 454: + switch(ndx) { + case 0: + p = "void *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "u_long"; + break; + case 3: + p = "void *"; + break; + case 4: + p = "void *"; + break; + default: + break; + }; + break; + /* thr_new */ + case 455: + switch(ndx) { + case 0: + p = "struct thr_param *"; + break; + case 1: + p = "int"; + break; + default: + break; + }; + break; + /* sigqueue */ + case 456: + switch(ndx) { + case 0: + p = "pid_t"; + break; + case 1: + p = "int"; + break; + case 2: + p = "void *"; + break; + default: + break; + }; + break; + /* kmq_open */ + case 457: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "mode_t"; + break; + case 3: + p = "const struct mq_attr *"; + break; + default: + break; + }; + break; + /* kmq_setattr */ + case 458: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "const struct mq_attr *"; + break; + case 2: + p = "struct mq_attr *"; + break; + default: + break; + }; + break; + /* kmq_timedreceive */ + case 459: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "char *"; + break; + case 2: + p = "size_t"; + break; + case 3: + p = "unsigned *"; + break; + case 4: + p = "const struct timespec *"; + break; + default: + break; + }; + break; + /* kmq_timedsend */ + case 460: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "const char *"; + break; + case 2: + p = "size_t"; + break; + case 3: + p = "unsigned"; + break; + case 4: + p = "const struct timespec *"; + break; + default: + break; + }; + break; + /* kmq_notify */ + case 461: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "const struct sigevent *"; + break; + default: + break; + }; + break; + /* kmq_unlink */ + case 462: + switch(ndx) { + case 0: + p = "const char *"; + break; + default: + break; + }; + break; + /* abort2 */ + case 463: + switch(ndx) { + case 0: + p = "const char *"; + break; + case 1: + p = "int"; + break; + case 2: + p = "void **"; + break; + default: + break; + }; + break; + /* thr_set_name */ + case 464: + switch(ndx) { + case 0: + p = "long"; + break; + case 1: + p = "const char *"; + break; + default: + break; + }; + break; + /* aio_fsync */ + case 465: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct aiocb *"; + break; + default: + break; + }; + break; + /* rtprio_thread */ + case 466: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "lwpid_t"; + break; + case 2: + p = "struct rtprio *"; + break; + default: + break; + }; + break; + /* sctp_peeloff */ + case 471: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "uint32_t"; + break; + default: + break; + }; + break; + /* sctp_generic_sendmsg */ + case 472: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "caddr_t"; + break; + case 2: + p = "int"; + break; + case 3: + p = "caddr_t"; + break; + case 4: + p = "__socklen_t"; + break; + case 5: + p = "struct sctp_sndrcvinfo *"; + break; + case 6: + p = "int"; + break; + default: + break; + }; + break; + /* sctp_generic_sendmsg_iov */ + case 473: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct iovec *"; + break; + case 2: + p = "int"; + break; + case 3: + p = "caddr_t"; + break; + case 4: + p = "__socklen_t"; + break; + case 5: + p = "struct sctp_sndrcvinfo *"; + break; + case 6: + p = "int"; + break; + default: + break; + }; + break; + /* sctp_generic_recvmsg */ + case 474: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "struct iovec *"; + break; + case 2: + p = "int"; + break; + case 3: + p = "struct sockaddr *"; + break; + case 4: + p = "__socklen_t *"; + break; + case 5: + p = "struct sctp_sndrcvinfo *"; + break; + case 6: + p = "int *"; + break; + default: + break; + }; + break; + /* pread */ + case 475: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "void *"; + break; + case 2: + p = "size_t"; + break; + case 3: + p = "off_t"; + break; + default: + break; + }; + break; + /* pwrite */ + case 476: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "const void *"; + break; + case 2: + p = "size_t"; + break; + case 3: + p = "off_t"; + break; + default: + break; + }; + break; + /* mmap */ + case 477: + switch(ndx) { + case 0: + p = "caddr_t"; + break; + case 1: + p = "size_t"; + break; + case 2: + p = "int"; + break; + case 3: + p = "int"; + break; + case 4: + p = "int"; + break; + case 5: + p = "off_t"; + break; + default: + break; + }; + break; + /* lseek */ + case 478: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "off_t"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + /* truncate */ + case 479: + switch(ndx) { + case 0: + p = "char *"; + break; + case 1: + p = "off_t"; + break; + default: + break; + }; + break; + /* ftruncate */ + case 480: + switch(ndx) { + case 0: + p = "int"; + break; + case 1: + p = "off_t"; + break; + default: + break; + }; + break; + /* thr_kill2 */ + case 481: + switch(ndx) { + case 0: + p = "pid_t"; + break; + case 1: + p = "long"; + break; + case 2: + p = "int"; + break; + default: + break; + }; + break; + default: + break; + }; + if (p != NULL) + strlcpy(desc, p, descsz); +} diff --git a/sys/modules/Makefile b/sys/modules/Makefile index 847cb91b3e48..756d2d986451 100644 --- a/sys/modules/Makefile +++ b/sys/modules/Makefile @@ -66,6 +66,7 @@ SUBDIR= ${_3dfx} \ cue \ cxgb \ ${_cx} \ + ${_cyclic} \ dc \ dcons \ dcons_crom \ @@ -73,6 +74,7 @@ SUBDIR= ${_3dfx} \ ${_digi} \ ${_dpt} \ ${_drm} \ + ${_dtrace} \ dummynet \ ${_ed} \ ${_elink} \ @@ -204,6 +206,7 @@ SUBDIR= ${_3dfx} \ ${_nvram} \ ${_nwfs} \ ${_oltr} \ + ${_opensolaris} \ ${_padlock} \ patm \ ${_pccard} \ @@ -390,8 +393,14 @@ _coff= coff _cp= cp _cpufreq= cpufreq _cs= cs +.if ${MK_CDDL} != "no" || defined(ALL_MODULES) +_cyclic= cyclic +.endif _digi= digi _drm= drm +.if ${MK_CDDL} != "no" || defined(ALL_MODULES) +_dtrace= dtrace +.endif _ed= ed _elink= elink _em= em @@ -423,6 +432,9 @@ _nsp= nsp _nwfs= nwfs .endif _oltr= oltr +.if ${MK_CDDL} != "no" || defined(ALL_MODULES) +_opensolaris= opensolaris +.endif _pccard= pccard _pcfclock= pcfclock _pecoff= pecoff @@ -518,8 +530,14 @@ _cmx= cmx _ciss= ciss _coretemp= coretemp _cpufreq= cpufreq +.if ${MK_CDDL} != "no" || defined(ALL_MODULES) +_cyclic= cyclic +.endif _digi= digi _drm= drm +.if ${MK_CDDL} != "no" || defined(ALL_MODULES) +_dtrace= dtrace +.endif _ed= ed _et= et _em= em @@ -550,6 +568,9 @@ _nfe= nfe _nve= nve _nvram= nvram _nxge= nxge +.if ${MK_CDDL} != "no" || defined(ALL_MODULES) +_opensolaris= opensolaris +.endif _pccard= pccard _rdma= rdma _safe= safe diff --git a/sys/modules/cyclic/Makefile b/sys/modules/cyclic/Makefile new file mode 100644 index 000000000000..db99488e138b --- /dev/null +++ b/sys/modules/cyclic/Makefile @@ -0,0 +1,19 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../cddl/dev/cyclic + +KMOD= cyclic +SRCS= cyclic.c + +SRCS+= vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../.. \ + -I${.CURDIR}/../../cddl/dev/cyclic/${MACHINE_ARCH} + +CFLAGS+= -DDEBUG=1 + +IGNORE_PRAGMA= 1 + +.include <bsd.kmod.mk> diff --git a/sys/modules/cyclic_test/Makefile b/sys/modules/cyclic_test/Makefile new file mode 100644 index 000000000000..e4a7c8679adb --- /dev/null +++ b/sys/modules/cyclic_test/Makefile @@ -0,0 +1,16 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../cddl/dev/cyclic + +KMOD= cyclic_test +SRCS= cyclic_test.c + +#SRCS+= bus_if.h device_if.h vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../.. + +IGNORE_PRAGMA= 1 + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/Makefile b/sys/modules/dtrace/Makefile new file mode 100644 index 000000000000..45c2b2f0d406 --- /dev/null +++ b/sys/modules/dtrace/Makefile @@ -0,0 +1,21 @@ +# $FreeBSD$ + +.include <bsd.own.mk> +.include "Makefile.inc" + +SUBDIR= dtmalloc \ + dtrace \ + dtraceall \ + dtrace_test \ + profile \ + prototype \ + sdt \ + systrace + +.if ${MACHINE_ARCH} == "amd64" +SUBDIR+= fbt +.elif ${MACHINE_ARCH} == "i386" +SUBDIR+= fbt +.endif + +.include <bsd.subdir.mk> diff --git a/sys/modules/dtrace/Makefile.inc b/sys/modules/dtrace/Makefile.inc new file mode 100644 index 000000000000..c35764b2998b --- /dev/null +++ b/sys/modules/dtrace/Makefile.inc @@ -0,0 +1,29 @@ +# $FreeBSD$ + +IGNORE_PRAGMA= 1 + +load : + -kldload cyclic + -kldload dtrace +.if ${MACHINE_ARCH} == "i386" + -kldload sdt + -kldload lockstat + -kldload fbt + -kldload prototype +.endif + -kldload profile + -kldload systrace + kldstat + +unload : + -kldunload systrace + -kldunload profile +.if ${MACHINE_ARCH} == "i386" + -kldunload prototype + -kldunload fbt + -kldunload lockstat + -kldunload sdt +.endif + -kldunload dtrace + -kldunload cyclic + kldstat diff --git a/sys/modules/dtrace/dtmalloc/Makefile b/sys/modules/dtrace/dtmalloc/Makefile new file mode 100644 index 000000000000..efb260763fa2 --- /dev/null +++ b/sys/modules/dtrace/dtmalloc/Makefile @@ -0,0 +1,13 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../cddl/dev/dtmalloc + +KMOD= dtmalloc +SRCS= dtmalloc.c +SRCS+= vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../../.. + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/dtrace/Makefile b/sys/modules/dtrace/dtrace/Makefile new file mode 100644 index 000000000000..5d16800a8fd1 --- /dev/null +++ b/sys/modules/dtrace/dtrace/Makefile @@ -0,0 +1,44 @@ +# $FreeBSD$ + +ARCHDIR= ${MACHINE_ARCH} + +.PATH: ${.CURDIR}/../../../cddl/contrib/opensolaris/uts/common/dtrace +.PATH: ${.CURDIR}/../../../cddl/kern +.PATH: ${.CURDIR}/../../../cddl/dev/dtrace +.PATH: ${.CURDIR}/../../../cddl/dev/dtrace/${ARCHDIR} + +KMOD= dtrace +SRCS= dtrace.c \ + dtrace_asm.S \ + dtrace_subr.c + +.if ${MACHINE_ARCH} == "amd64" || ${MACHINE_ARCH} == "i386" +SRCS+= dis_tables.c \ + instr_size.c +.endif + +SRCS+= bus_if.h device_if.h vnode_if.h + +# Needed for dtrace_asm.S +SRCS+= assym.s + +# These are needed for assym.s +SRCS+= opt_compat.h opt_kstack_pages.h opt_nfs.h + +.if ${MACHINE_ARCH} == "i386" +SRCS+= opt_apic.h +.endif + +CFLAGS+= -I${.CURDIR}/../../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../../cddl/dev/dtrace \ + -I${.CURDIR}/../../../cddl/dev/dtrace/${ARCHDIR} \ + -I${.CURDIR}/../../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../../.. -DDIS_MEM + +CFLAGS+= -DSMP -DDEBUG + +EXPORT_SYMS= dtrace_register \ + dtrace_unregister \ + dtrace_probe_lookup + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/dtrace_test/Makefile b/sys/modules/dtrace/dtrace_test/Makefile new file mode 100644 index 000000000000..bc096f3a7398 --- /dev/null +++ b/sys/modules/dtrace/dtrace_test/Makefile @@ -0,0 +1,14 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../cddl/dev/dtrace + +KMOD= dtrace_test +SRCS= dtrace_test.c + +SRCS+= vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../.. + +CFLAGS+= -D_KERNEL + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/dtraceall/Makefile b/sys/modules/dtrace/dtraceall/Makefile new file mode 100644 index 000000000000..6f098024611c --- /dev/null +++ b/sys/modules/dtrace/dtraceall/Makefile @@ -0,0 +1,8 @@ +# $FreeBSD$ + +KMOD= dtraceall +SRCS= dtraceall.c + +CFLAGS+= -I${.CURDIR}/../../.. + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/dtraceall/dtraceall.c b/sys/modules/dtrace/dtraceall/dtraceall.c new file mode 100644 index 000000000000..4f7bdfed5613 --- /dev/null +++ b/sys/modules/dtrace/dtraceall/dtraceall.c @@ -0,0 +1,73 @@ +/* + * Copyright (C) 2008 John Birrell <jb@freebsd.org> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + */ + +#include <sys/cdefs.h> +#include <sys/param.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/kernel.h> +#include <sys/module.h> +#include <sys/errno.h> + +static int +dtraceall_modevent(module_t mod __unused, int type, void *data __unused) +{ + int error = 0; + + switch (type) { + case MOD_LOAD: + break; + + case MOD_UNLOAD: + break; + + case MOD_SHUTDOWN: + break; + + default: + error = EOPNOTSUPP; + break; + + } + + return (error); +} + +DEV_MODULE(dtraceall, dtraceall_modevent, NULL); +MODULE_VERSION(dtraceall, 1); + +/* All the DTrace modules should be dependencies here: */ +MODULE_DEPEND(dtraceall, cyclic, 1, 1, 1); +MODULE_DEPEND(dtraceall, opensolaris, 1, 1, 1); +MODULE_DEPEND(dtraceall, dtrace, 1, 1, 1); +MODULE_DEPEND(dtraceall, dtmalloc, 1, 1, 1); +#if defined(__amd64__) || defined(__i386__) +MODULE_DEPEND(dtraceall, fbt, 1, 1, 1); +#endif +MODULE_DEPEND(dtraceall, sdt, 1, 1, 1); +MODULE_DEPEND(dtraceall, systrace, 1, 1, 1); +MODULE_DEPEND(dtraceall, profile, 1, 1, 1); diff --git a/sys/modules/dtrace/fasttrap/Makefile b/sys/modules/dtrace/fasttrap/Makefile new file mode 100644 index 000000000000..104fd9af4454 --- /dev/null +++ b/sys/modules/dtrace/fasttrap/Makefile @@ -0,0 +1,19 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../cddl/dev/fasttrap + +KMOD= fasttrap +SRCS= fasttrap.c +SRCS+= vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../../.. + +.if ${MACHINE_ARCH} == "amd64" || ${MACHINE_ARCH} == "i386" +CFLAGS+= -I${.CURDIR}/../../../cddl/contrib/opensolaris/uts/intel +.endif + +CFLAGS+= -DSMP -DDEBUG + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/fbt/Makefile b/sys/modules/dtrace/fbt/Makefile new file mode 100644 index 000000000000..9b69b224a227 --- /dev/null +++ b/sys/modules/dtrace/fbt/Makefile @@ -0,0 +1,13 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../cddl/dev/fbt + +KMOD= fbt +SRCS= fbt.c +SRCS+= vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../../.. + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/linsystrace/Makefile b/sys/modules/dtrace/linsystrace/Makefile new file mode 100644 index 000000000000..fb9deb6eeb4e --- /dev/null +++ b/sys/modules/dtrace/linsystrace/Makefile @@ -0,0 +1,16 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../cddl/dev/systrace + +KMOD= linsystrace +SRCS= systrace.c +SRCS+= vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../../../${MACHINE_ARCH}/linux \ + -I${.CURDIR}/../../.. + +CFLAGS+= -DLINUX_SYSTRACE + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/lockstat/Makefile b/sys/modules/dtrace/lockstat/Makefile new file mode 100644 index 000000000000..d387dab96751 --- /dev/null +++ b/sys/modules/dtrace/lockstat/Makefile @@ -0,0 +1,14 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../cddl/dev/lockstat + +KMOD= lockstat +SRCS= lockstat.c + +SRCS+= vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../../.. + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/profile/Makefile b/sys/modules/dtrace/profile/Makefile new file mode 100644 index 000000000000..22cd10d6c9e8 --- /dev/null +++ b/sys/modules/dtrace/profile/Makefile @@ -0,0 +1,13 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../cddl/dev/profile + +KMOD= profile +SRCS= profile.c +SRCS+= vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../../.. + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/prototype/Makefile b/sys/modules/dtrace/prototype/Makefile new file mode 100644 index 000000000000..49cc0a89b273 --- /dev/null +++ b/sys/modules/dtrace/prototype/Makefile @@ -0,0 +1,13 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../cddl/dev + +KMOD= prototype +SRCS= prototype.c +SRCS+= vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../../.. + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/sdt/Makefile b/sys/modules/dtrace/sdt/Makefile new file mode 100644 index 000000000000..d531d789b803 --- /dev/null +++ b/sys/modules/dtrace/sdt/Makefile @@ -0,0 +1,13 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../cddl/dev/sdt + +KMOD= sdt +SRCS= sdt.c +SRCS+= vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../../.. + +.include <bsd.kmod.mk> diff --git a/sys/modules/dtrace/systrace/Makefile b/sys/modules/dtrace/systrace/Makefile new file mode 100644 index 000000000000..d85a365705ee --- /dev/null +++ b/sys/modules/dtrace/systrace/Makefile @@ -0,0 +1,13 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../../cddl/dev/systrace + +KMOD= systrace +SRCS= systrace.c +SRCS+= vnode_if.h + +CFLAGS+= -I${.CURDIR}/../../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../../.. + +.include <bsd.kmod.mk> diff --git a/sys/modules/opensolaris/Makefile b/sys/modules/opensolaris/Makefile new file mode 100644 index 000000000000..5e2195ac8b8f --- /dev/null +++ b/sys/modules/opensolaris/Makefile @@ -0,0 +1,23 @@ +# $FreeBSD$ + +.PATH: ${.CURDIR}/../../cddl/compat/opensolaris/kern + +KMOD= opensolaris +SRCS= opensolaris.c \ + opensolaris_cmn_err.c \ + opensolaris_kmem.c + +.if ${MACHINE_ARCH} == "i386" || ${MACHINE_ARCH} == "amd64" || ${MACHINE_ARCH} == "ia64" +.PATH: ${.CURDIR}/../../cddl/contrib/opensolaris/common/atomic/${MACHINE_ARCH} +SRCS+= atomic.S +.else +SRCS+= opensolaris_atomic.c +.endif + +CFLAGS+= -I${.CURDIR}/../../cddl/compat/opensolaris \ + -I${.CURDIR}/../../cddl/contrib/opensolaris/uts/common \ + -I${.CURDIR}/../.. + +IGNORE_PRAGMA= 1 + +.include <bsd.kmod.mk> diff --git a/sys/modules/zfs/Makefile b/sys/modules/zfs/Makefile index 8ed90e3b5347..0cd4361e4c65 100644 --- a/sys/modules/zfs/Makefile +++ b/sys/modules/zfs/Makefile @@ -4,11 +4,13 @@ KMOD= zfs SRCS= vnode_if.h -.PATH: ${.CURDIR}/../../cddl/contrib/opensolaris/common/acl +SUNW= ${.CURDIR}/../../cddl/contrib/opensolaris + +.PATH: ${SUNW}/common/acl SRCS+= acl_common.c -.PATH: ${.CURDIR}/../../cddl/contrib/opensolaris/common/avl +.PATH: ${SUNW}/common/avl SRCS+= avl.c -.PATH: ${.CURDIR}/../../cddl/contrib/opensolaris/common/nvpair +.PATH: ${SUNW}/common/nvpair SRCS+= nvpair.c .PATH: ${.CURDIR}/../../cddl/compat/opensolaris/kern @@ -22,28 +24,28 @@ SRCS+= opensolaris_vfs.c SRCS+= opensolaris_zone.c .if ${MACHINE_ARCH} == "i386" || ${MACHINE_ARCH} == "amd64" || ${MACHINE_ARCH} == "ia64" || ${MACHINE_ARCH} == "sparc64" -.PATH: ${.CURDIR}/../../cddl/contrib/opensolaris/common/atomic/${MACHINE_ARCH} +.PATH: ${SUNW}/common/atomic/${MACHINE_ARCH} SRCS+= atomic.S .else .PATH: ${.CURDIR}/../../cddl/compat/opensolaris/kern SRCS+= opensolaris_atomic.c .endif -.PATH: ${.CURDIR}/../../cddl/contrib/opensolaris/uts/common/fs +.PATH: ${SUNW}/uts/common/fs SRCS+= gfs.c -.PATH: ${.CURDIR}/../../cddl/contrib/opensolaris/uts/common/os +.PATH: ${SUNW}/uts/common/os SRCS+= callb.c SRCS+= list.c SRCS+= nvpair_alloc_system.c SRCS+= taskq.c -.PATH: ${.CURDIR}/../../cddl/contrib/opensolaris/uts/common/rpc +.PATH: ${SUNW}/uts/common/rpc SRCS+= xdr.c SRCS+= xdr_array.c SRCS+= xdr_mem.c -.PATH: ${.CURDIR}/../../cddl/contrib/opensolaris/uts/common/zmod +.PATH: ${SUNW}/uts/common/zmod SRCS+= adler32.c SRCS+= crc32.c SRCS+= deflate.c @@ -55,9 +57,9 @@ SRCS+= zmod.c SRCS+= zmod_subr.c SRCS+= zutil.c -.PATH: ${.CURDIR}/../../cddl/contrib/opensolaris/common/zfs -.include "${.CURDIR}/../../cddl/contrib/opensolaris/uts/common/Makefile.files" -.PATH: ${.CURDIR}/../../cddl/contrib/opensolaris/uts/common/fs/zfs +.PATH: ${SUNW}/common/zfs +.include "${SUNW}/uts/common/Makefile.files" +.PATH: ${SUNW}/uts/common/fs/zfs ZFS_SRCS= ${ZFS_OBJS:C/.o$/.c/} SRCS+= ${ZFS_SRCS} SRCS+= vdev_geom.c @@ -68,28 +70,31 @@ SRCS+= vdev_geom.c # Use FreeBSD's namecache. CFLAGS+=-DFREEBSD_NAMECACHE -CWARNFLAGS=-Wall -CWARNFLAGS+=-Wno-unknown-pragmas -CWARNFLAGS+=-Wno-missing-braces -CWARNFLAGS+=-Wno-parentheses -CWARNFLAGS+=-Wno-uninitialized -CWARNFLAGS+=-Wno-unused -CWARNFLAGS+=-Wno-switch - -CFLAGS+=-D_SOLARIS_C_SOURCE - CFLAGS+=-I${.CURDIR}/../../cddl/compat/opensolaris -CFLAGS+=-I${.CURDIR}/../../cddl/contrib/opensolaris/uts/common/fs/zfs -CFLAGS+=-I${.CURDIR}/../../cddl/contrib/opensolaris/uts/common/zmod -CFLAGS+=-I${.CURDIR}/../../cddl/contrib/opensolaris/uts/common +CFLAGS+=-I${SUNW}/uts/common/fs/zfs +CFLAGS+=-I${SUNW}/uts/common/zmod +CFLAGS+=-I${SUNW}/uts/common CFLAGS+=-I${.CURDIR}/../.. -CFLAGS+=-I${.CURDIR}/../../cddl/contrib/opensolaris/common/zfs -CFLAGS+=-I${.CURDIR}/../../cddl/contrib/opensolaris/common +CFLAGS+=-I${SUNW}/common/zfs +CFLAGS+=-I${SUNW}/common CFLAGS+=-I${.CURDIR}/../../../include - -WARNS?=1 +CFLAGS+=-DBUILDING_ZFS #CFLAGS+=-DDEBUG=1 #DEBUG_FLAGS=-g .include <bsd.kmod.mk> + +CWARNFLAGS+=-Wno-unknown-pragmas +CWARNFLAGS+=-Wno-missing-prototypes +CWARNFLAGS+=-Wno-undef +CWARNFLAGS+=-Wno-strict-prototypes +CWARNFLAGS+=-Wno-cast-qual +CWARNFLAGS+=-Wno-parentheses +CWARNFLAGS+=-Wno-redundant-decls +CWARNFLAGS+=-Wno-missing-braces +CWARNFLAGS+=-Wno-uninitialized +CWARNFLAGS+=-Wno-unused +CWARNFLAGS+=-Wno-inline +CWARNFLAGS+=-Wno-switch +CWARNFLAGS+=-Wno-pointer-arith diff --git a/sys/powerpc/include/param.h b/sys/powerpc/include/param.h index 6a2018eab0d8..549e0b9da24f 100644 --- a/sys/powerpc/include/param.h +++ b/sys/powerpc/include/param.h @@ -67,8 +67,8 @@ #endif #define MID_MACHINE MID_POWERPC -#ifdef SMP -#define MAXCPU 2 +#if defined(SMP) || defined(KLD_MODULE) +#define MAXCPU 2 #else #define MAXCPU 1 #endif /* SMP */ diff --git a/sys/sparc64/include/param.h b/sys/sparc64/include/param.h index dd86233a943b..dd8fb1cbb431 100644 --- a/sys/sparc64/include/param.h +++ b/sys/sparc64/include/param.h @@ -59,7 +59,7 @@ #endif #define MID_MACHINE MID_SPARC64 -#ifdef SMP +#if defined(SMP) || defined(KLD_MODULE) #define MAXCPU 16 #else #define MAXCPU 1 diff --git a/sys/sun4v/include/param.h b/sys/sun4v/include/param.h index 6e1d389b304c..a4fe98d3cf7b 100644 --- a/sys/sun4v/include/param.h +++ b/sys/sun4v/include/param.h @@ -59,7 +59,7 @@ #endif #define MID_MACHINE MID_SPARC64 -#ifdef SMP +#if defined(SMP) || defined(KLD_MODULE) #define MAXCPU 32 #else #define MAXCPU 1 diff --git a/sys/sys/dtrace_bsd.h b/sys/sys/dtrace_bsd.h new file mode 100644 index 000000000000..e0adbe624a90 --- /dev/null +++ b/sys/sys/dtrace_bsd.h @@ -0,0 +1,111 @@ +/*- + * Copyright (c) 2007-2008 John Birrell (jb@freebsd.org) + * All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + * This file contains BSD shims for Sun's DTrace code. + */ + +#ifndef _SYS_DTRACE_BSD_H +#define _SYS_DTRACE_BSD_H + +/* Forward definitions: */ +struct trapframe; +struct thread; + +/* + * Cyclic clock function type definition used to hook the cyclic + * subsystem into the appropriate timer interrupt. + */ +typedef void (*cyclic_clock_func_t)(struct trapframe *); + +/* + * These external variables are actually machine-dependent, so + * they might not actually exist. + * + * Defining them here avoids a proliferation of header files. + */ +extern cyclic_clock_func_t lapic_cyclic_clock_func[]; + +/* + * The dtrace module handles traps that occur during a DTrace probe. + * This type definition is used in the trap handler to provide a + * hook for the dtrace module to register it's handler with. + */ +typedef int (*dtrace_trap_func_t)(struct trapframe *, u_int); + +int dtrace_trap(struct trapframe *, u_int); + +extern dtrace_trap_func_t dtrace_trap_func; + +/* Used by the machine dependent trap() code. */ +typedef int (*dtrace_invop_func_t)(uintptr_t, uintptr_t *, uintptr_t); +typedef void (*dtrace_doubletrap_func_t)(void); + +/* Global variables in trap.c */ +extern dtrace_invop_func_t dtrace_invop_func; +extern dtrace_doubletrap_func_t dtrace_doubletrap_func; + +/* Virtual time hook function type. */ +typedef void (*dtrace_vtime_switch_func_t)(struct thread *); + +extern int dtrace_vtime_active; +extern dtrace_vtime_switch_func_t dtrace_vtime_switch_func; + +/* The fasttrap module hooks into the fork, exit and exit. */ +typedef void (*dtrace_fork_func_t)(struct proc *, struct proc *); +typedef void (*dtrace_execexit_func_t)(struct proc *); + +/* Global variable in kern_fork.c */ +extern dtrace_fork_func_t dtrace_fasttrap_fork; + +/* Global variable in kern_exec.c */ +extern dtrace_execexit_func_t dtrace_fasttrap_exec; + +/* Global variable in kern_exit.c */ +extern dtrace_execexit_func_t dtrace_fasttrap_exit; + +/* The dtmalloc provider hooks into malloc. */ +typedef void (*dtrace_malloc_probe_func_t)(u_int32_t, uintptr_t arg0, + uintptr_t arg1, uintptr_t arg2, uintptr_t arg3, uintptr_t arg4); + +extern dtrace_malloc_probe_func_t dtrace_malloc_probe; + +/* + * Functions which allow the dtrace module to check that the kernel + * hooks have been compiled with sufficient space for it's private + * structures. + */ +size_t kdtrace_proc_size(void); +size_t kdtrace_thread_size(void); + +/* + * OpenSolaris compatible time functions returning nanoseconds. + * On OpenSolaris these return hrtime_t which we define as uint64_t. + */ +uint64_t dtrace_gethrtime(void); +uint64_t dtrace_gethrestime(void); + +#endif /* _SYS_DTRACE_BSD_H */ diff --git a/sys/sys/elf.h b/sys/sys/elf.h index 54df9336ab9c..b2d6b77c98e2 100644 --- a/sys/sys/elf.h +++ b/sys/sys/elf.h @@ -38,8 +38,4 @@ #include <sys/elf32.h> #include <sys/elf64.h> -#ifdef _SOLARIS_C_SOURCE -#include <sys/_elf_solaris.h> -#endif - #endif /* !_SYS_ELF_H_ */ diff --git a/sys/sys/elf_common.h b/sys/sys/elf_common.h index b941fd70aaf3..ef2ab0fd21da 100644 --- a/sys/sys/elf_common.h +++ b/sys/sys/elf_common.h @@ -431,6 +431,9 @@ typedef struct { #define STV_INTERNAL 0x1 /* Special meaning in relocatable objects. */ #define STV_HIDDEN 0x2 /* Not visible. */ #define STV_PROTECTED 0x3 /* Visible but not preemptible. */ +#define STV_EXPORTED 0x4 +#define STV_SINGLETON 0x5 +#define STV_ELIMINATE 0x6 /* Special symbol table indexes. */ #define STN_UNDEF 0 /* Undefined symbol index. */ diff --git a/sys/sys/kdb.h b/sys/sys/kdb.h index b45cd70d136c..d0bac955b813 100644 --- a/sys/sys/kdb.h +++ b/sys/sys/kdb.h @@ -105,5 +105,6 @@ extern const char * volatile kdb_why; #define KDB_WHY_MAC "mac" /* MAC Framework. */ #define KDB_WHY_POWERPC "powerpc" /* Unhandled powerpc intr. */ #define KDB_WHY_UNIONFS "unionfs" /* Unionfs bug. */ +#define KDB_WHY_DTRACE "dtrace" /* DTrace action entered debugger. */ #endif /* !_SYS_KDB_H_ */ diff --git a/sys/sys/kernel.h b/sys/sys/kernel.h index 9150d8153fe9..90cf72a287fe 100644 --- a/sys/sys/kernel.h +++ b/sys/sys/kernel.h @@ -68,9 +68,7 @@ extern int stathz; /* statistics clock's frequency */ extern int profhz; /* profiling clock's frequency */ extern int profprocs; /* number of process's profiling */ extern int ticks; -#ifndef _SOLARIS_C_SOURCE extern int lbolt; /* once a second sleep address */ -#endif #endif /* _KERNEL */ @@ -115,6 +113,7 @@ enum sysinit_sub_id { SI_SUB_EVENTHANDLER = 0x1C00000, /* eventhandler init */ SI_SUB_KLD = 0x2000000, /* KLD and module setup */ SI_SUB_CPU = 0x2100000, /* CPU resource(s)*/ + SI_SUB_KDTRACE = 0x2140000, /* Kernel dtrace hooks */ SI_SUB_MAC = 0x2180000, /* TrustedBSD MAC subsystem */ SI_SUB_MAC_POLICY = 0x21C0000, /* TrustedBSD MAC policies */ SI_SUB_MAC_LATE = 0x21D0000, /* TrustedBSD MAC subsystem */ @@ -123,6 +122,8 @@ enum sysinit_sub_id { SI_SUB_DDB_SERVICES = 0x2380000, /* capture, scripting, etc. */ SI_SUB_RUN_QUEUE = 0x2400000, /* set up run queue*/ SI_SUB_KTRACE = 0x2480000, /* ktrace */ + SI_SUB_OPENSOLARIS = 0x2490000, /* OpenSolaris compatibility */ + SI_SUB_CYCLIC = 0x24A0000, /* Cyclic timers */ SI_SUB_AUDIT = 0x24C0000, /* audit */ SI_SUB_CREATE_INIT = 0x2500000, /* create init process*/ SI_SUB_SCHED_IDLE = 0x2600000, /* required idle procs */ @@ -133,6 +134,9 @@ enum sysinit_sub_id { SI_SUB_DEVFS = 0x2F00000, /* devfs ready for devices */ SI_SUB_INIT_IF = 0x3000000, /* prep for net interfaces */ SI_SUB_NETGRAPH = 0x3010000, /* Let Netgraph initialize */ + SI_SUB_DTRACE = 0x3020000, /* DTrace subsystem */ + SI_SUB_DTRACE_PROVIDER = 0x3048000, /* DTrace providers */ + SI_SUB_DTRACE_ANON = 0x308C000, /* DTrace anon enabling */ SI_SUB_DRIVERS = 0x3100000, /* Let Drivers initialize */ SI_SUB_CONFIGURE = 0x3800000, /* Configure devices */ SI_SUB_VFS = 0x4000000, /* virtual filesystem*/ @@ -155,7 +159,6 @@ enum sysinit_sub_id { SI_SUB_ROOT_CONF = 0xb000000, /* Find root devices */ SI_SUB_DUMP_CONF = 0xb200000, /* Find dump devices */ SI_SUB_RAID = 0xb380000, /* Configure GEOM classes */ - SI_SUB_MOUNT_ROOT = 0xb400000, /* root mount*/ SI_SUB_SWAP = 0xc000000, /* swap */ SI_SUB_INTRINSIC_POST = 0xd000000, /* proc 0 cleanup*/ SI_SUB_SYSCALLS = 0xd800000, /* register system calls */ diff --git a/sys/sys/linker.h b/sys/sys/linker.h index c4903b09d8ce..4ba7a40072ac 100644 --- a/sys/sys/linker.h +++ b/sys/sys/linker.h @@ -59,6 +59,8 @@ typedef struct linker_symval { size_t size; } linker_symval_t; +typedef int (*linker_function_nameval_callback_t)(linker_file_t, int, linker_symval_t *, void *); + struct common_symbol { STAILQ_ENTRY(common_symbol) link; char* name; @@ -73,6 +75,7 @@ struct linker_file { #define LINKER_FILE_LINKED 0x1 /* file has been fully linked */ TAILQ_ENTRY(linker_file) link; /* list of all loaded files */ char* filename; /* file which was loaded */ + char* pathname; /* file name with full path */ int id; /* unique id */ caddr_t address; /* load address */ size_t size; /* size of file */ @@ -81,6 +84,18 @@ struct linker_file { STAILQ_HEAD(, common_symbol) common; /* list of common symbols */ TAILQ_HEAD(, module) modules; /* modules in this file */ TAILQ_ENTRY(linker_file) loaded; /* preload dependency support */ + int loadcnt; /* load counter value */ + + /* + * Function Boundary Tracing (FBT) or Statically Defined Tracing (SDT) + * fields. + */ + int nenabled; /* number of enabled probes. */ + int fbt_nentries; /* number of fbt entries created. */ + void *sdt_probes; + int sdt_nentries; + size_t sdt_nprobes; + size_t sdt_size; }; /* @@ -141,6 +156,12 @@ int linker_file_lookup_set(linker_file_t _file, const char *_name, void *_start, void *_stop, int *_count); /* + * List all functions in a file. + */ +int linker_file_function_listall(linker_file_t, + linker_function_nameval_callback_t, void *); + +/* * Functions soley for use by the linker class handlers. */ int linker_add_class(linker_class_t _cls); @@ -246,6 +267,20 @@ int elf_reloc_local(linker_file_t _lf, Elf_Addr base, const void *_rel, int _typ const Elf_Sym *elf_get_sym(linker_file_t _lf, Elf_Size _symidx); const char *elf_get_symname(linker_file_t _lf, Elf_Size _symidx); +typedef struct linker_ctf { + const uint8_t *ctftab; /* Decompressed CTF data. */ + int ctfcnt; /* Number of CTF data bytes. */ + const Elf_Sym *symtab; /* Ptr to the symbol table. */ + int nsym; /* Number of symbols. */ + const char *strtab; /* Ptr to the string table. */ + int strcnt; /* Number of string bytes. */ + uint32_t **ctfoffp; /* Ptr to array of obj/fnc offsets. */ + uint32_t **typoffp; /* Ptr to array of type offsets. */ + long *typlenp; /* Ptr to number of type data entries. */ +} linker_ctf_t; + +int linker_ctf_get(linker_file_t, linker_ctf_t *); + int elf_cpu_load_file(linker_file_t); int elf_cpu_unload_file(linker_file_t); @@ -253,15 +288,28 @@ int elf_cpu_unload_file(linker_file_t); #define ELF_RELOC_REL 1 #define ELF_RELOC_RELA 2 +/* + * This is version 1 of the KLD file status structure. It is identified + * by its _size_ in the version field. + */ +struct kld_file_stat_1 { + int version; /* set to sizeof(struct kld_file_stat_1) */ + char name[MAXPATHLEN]; + int refs; + int id; + caddr_t address; /* load address */ + size_t size; /* size in bytes */ +}; #endif /* _KERNEL */ struct kld_file_stat { - int version; /* set to sizeof(linker_file_stat) */ + int version; /* set to sizeof(struct kld_file_stat) */ char name[MAXPATHLEN]; int refs; int id; caddr_t address; /* load address */ size_t size; /* size in bytes */ + char pathname[MAXPATHLEN]; }; struct kld_sym_lookup { diff --git a/sys/sys/malloc.h b/sys/sys/malloc.h index e59b9ac64560..33b287f83d92 100644 --- a/sys/sys/malloc.h +++ b/sys/sys/malloc.h @@ -80,7 +80,16 @@ struct malloc_type_stats { uint64_t _mts_reserved3; /* Reserved field. */ }; +/* + * Index definitions for the mti_probes[] array. + */ +#define DTMALLOC_PROBE_MALLOC 0 +#define DTMALLOC_PROBE_FREE 1 +#define DTMALLOC_PROBE_MAX 2 + struct malloc_type_internal { + uint32_t mti_probes[DTMALLOC_PROBE_MAX]; + /* DTrace probe ID array. */ struct malloc_type_stats mti_stats[MAXCPU]; }; @@ -173,7 +182,11 @@ MALLOC_DECLARE(M_IOV); extern struct mtx malloc_mtx; -/* XXX struct malloc_type is unused for contig*(). */ +/* + * Function type used when iterating over the list of malloc types. + */ +typedef void malloc_type_list_func_t(struct malloc_type *, void *); + void contigfree(void *addr, unsigned long size, struct malloc_type *type); void *contigmalloc(unsigned long size, struct malloc_type *type, int flags, vm_paddr_t low, vm_paddr_t high, unsigned long alignment, @@ -184,6 +197,7 @@ void malloc_init(void *); int malloc_last_fail(void); void malloc_type_allocated(struct malloc_type *type, unsigned long size); void malloc_type_freed(struct malloc_type *type, unsigned long size); +void malloc_type_list(malloc_type_list_func_t *, void *); void malloc_uninit(void *); void *realloc(void *addr, unsigned long size, struct malloc_type *type, int flags); diff --git a/sys/sys/param.h b/sys/sys/param.h index 55d5c673edc0..788c9409bd00 100644 --- a/sys/sys/param.h +++ b/sys/sys/param.h @@ -305,20 +305,4 @@ __END_DECLS #define ctodb(db) /* calculates pages to devblks */ \ ((db) << (PAGE_SHIFT - DEV_BSHIFT)) -/* - * Solaris compatibility definitions. - */ -#ifdef _SOLARIS_C_SOURCE -#define PAGESIZE PAGE_SIZE - -/* - * The OpenSolaris version is set according to the version last imported - * from http://dlc.sun.com/osol/on/downloads/current/. In FreeBSD header - * files it can be used to determine the level of compatibility that the - * FreeBSD headers provide to OpenSolaris code. Perhaps one day there - * will be a really, really Single Unix Specification. - */ -#define __OpenSolaris_version 20060731 -#endif - #endif /* _SYS_PARAM_H_ */ diff --git a/sys/sys/proc.h b/sys/sys/proc.h index 0714c30c77bc..f66f68ad770a 100644 --- a/sys/sys/proc.h +++ b/sys/sys/proc.h @@ -164,6 +164,8 @@ struct trapframe; struct turnstile; struct mqueue_notifier; struct cpuset; +struct kdtrace_proc; +struct kdtrace_thread; /* * Here we define the two structures used for process information. @@ -301,6 +303,8 @@ struct thread { uint64_t td_incruntime; /* (t) Cpu ticks to transfer to proc. */ struct cpuset *td_cpuset; /* (t) CPU affinity mask. */ struct file *td_fpop; /* (k) file referencing cdev under op */ + struct kdtrace_thread *td_dtrace; /* (*) DTrace-specific data. */ + int td_errno; /* Error returned by last syscall. */ }; struct mtx *thread_lock_block(struct thread *); @@ -587,6 +591,7 @@ struct proc { struct p_sched *p_sched; /* (*) Scheduler-specific data. */ STAILQ_HEAD(, ktr_request) p_ktr; /* (o) KTR event queue. */ LIST_HEAD(, mqueue_notifier) p_mqnotifier; /* (c) mqueue notifiers.*/ + struct kdtrace_proc *p_dtrace; /* (*) DTrace-specific data. */ }; #define p_session p_pgrp->pg_session diff --git a/sys/sys/sdt.h b/sys/sys/sdt.h new file mode 100644 index 000000000000..2bf2f2de77af --- /dev/null +++ b/sys/sys/sdt.h @@ -0,0 +1,179 @@ +/*- + * Copyright 2006-2008 John Birrell <jb@FreeBSD.org> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $FreeBSD$ + * + * Statically Defined Tracing (SDT) definitions. + * + */ + +#ifndef _SYS_SDT_H +#define _SYS_SDT_H + +/* Stub these for the time being. */ +#define DTRACE_PROBE(name) +#define DTRACE_PROBE1(name, type1, arg1) +#define DTRACE_PROBE2(name, type1, arg1, type2, arg2) +#define DTRACE_PROBE3(name, type1, arg1, type2, arg2, type3, arg3) +#define DTRACE_PROBE4(name, type1, arg1, type2, arg2, type3, arg3, type4, arg4) + +#ifndef _KERNEL + +/* The promise of things to come. Worlds to explore. People to meet. Things to do. */ + +#else + +#ifndef KDTRACE_HOOKS + +#define SDT_PROVIDER_DEFINE(prov) +#define SDT_PROVIDER_DECLARE(prov) +#define SDT_PROBE_DEFINE(prov, mod, func, name) +#define SDT_PROBE_DECLARE(prov, mod, func, name) +#define SDT_PROBE(prov, mod, func, name, arg0, arg1, arg2, arg3, arg4) +#define SDT_PROBE_ARGTYPE(prov, mod, func, name, num, type) + +#else + +/* + * This type definition must match that of dtrace_probe. It is defined this + * way to avoid having to rely on CDDL code. + */ +typedef void (*sdt_probe_func_t)(u_int32_t, uintptr_t arg0, uintptr_t arg1, + uintptr_t arg2, uintptr_t arg3, uintptr_t arg4); + +/* + * The hook for the probe function. See kern_sdt.c which defaults this to + * it's own stub. The 'sdt' provider will set it to dtrace_probe when it + * loads. + */ +extern sdt_probe_func_t sdt_probe_func; + +typedef enum { + SDT_UNINIT = 1, + SDT_INIT, +} sdt_state_t; + +struct sdt_probe; +struct sdt_provider; + +struct sdt_argtype { + int ndx; /* Argument index. */ + const char *type; /* Argument type string. */ + TAILQ_ENTRY(sdt_argtype) + argtype_entry; /* Argument type list entry. */ + struct sdt_probe + *probe; /* Ptr to the probe structure. */ +}; + +struct sdt_probe { + int version; /* Set to sizeof(struct sdt_ref). */ + sdt_state_t state; + struct sdt_provider + *prov; /* Ptr to the provider structure. */ + TAILQ_ENTRY(sdt_probe) + probe_entry; /* SDT probe list entry. */ + TAILQ_HEAD(argtype_list_head, sdt_argtype) argtype_list; + const char *mod; + const char *func; + const char *name; + id_t id; /* DTrace probe ID. */ + int n_args; /* Number of arguments. */ +}; + +struct sdt_provider { + const char *name; /* Provider name. */ + TAILQ_ENTRY(sdt_provider) + prov_entry; /* SDT provider list entry. */ + TAILQ_HEAD(probe_list_head, sdt_probe) probe_list; + uintptr_t id; /* DTrace provider ID. */ +}; + +#define SDT_PROVIDER_DEFINE(prov) \ + struct sdt_provider sdt_provider_##prov[1] = { \ + { #prov, { NULL, NULL }, { NULL, NULL } } \ + }; \ + SYSINIT(sdt_provider_##prov##_init, SI_SUB_KDTRACE, \ + SI_ORDER_SECOND, sdt_provider_register, \ + sdt_provider_##prov ); \ + SYSUNINIT(sdt_provider_##prov##_uninit, SI_SUB_KDTRACE, \ + SI_ORDER_SECOND, sdt_provider_deregister, \ + sdt_provider_##prov ) + +#define SDT_PROVIDER_DECLARE(prov) \ + extern struct sdt_provider sdt_provider_##prov[1] + +#define SDT_PROBE_DEFINE(prov, mod, func, name) \ + struct sdt_probe sdt_##prov##_##mod##_##func##_##name[1] = { \ + { sizeof(struct sdt_probe), 0, sdt_provider_##prov, \ + { NULL, NULL }, { NULL, NULL }, #mod, #func, #name, 0, 0 } \ + }; \ + SYSINIT(sdt_##prov##_##mod##_##func##_##name##_init, SI_SUB_KDTRACE, \ + SI_ORDER_SECOND + 1, sdt_probe_register, \ + sdt_##prov##_##mod##_##func##_##name ); \ + SYSUNINIT(sdt_##prov##_##mod##_##func##_##name##_uninit, \ + SI_SUB_KDTRACE, SI_ORDER_SECOND + 1, sdt_probe_deregister, \ + sdt_##prov##_##mod##_##func##_##name ) + +#define SDT_PROBE_DECLARE(prov, mod, func, name) \ + extern struct sdt_probe sdt_##prov##_##mod##_##func##_##name[1] + +#define SDT_PROBE(prov, mod, func, name, arg0, arg1, arg2, arg3, arg4) \ + if (sdt_##prov##_##mod##_##func##_##name->id) \ + (*sdt_probe_func)(sdt_##prov##_##mod##_##func##_##name->id, \ + (uintptr_t) arg0, (uintptr_t) arg1, (uintptr_t) arg2, \ + (uintptr_t) arg3, (uintptr_t) arg4) + +#define SDT_PROBE_ARGTYPE(prov, mod, func, name, num, type) \ + struct sdt_argtype sdt_##prov##_##mod##_##func##_##name##num[1] \ + = { { num, type, { NULL, NULL }, \ + sdt_##prov##_##mod##_##func##_##name } \ + }; \ + SYSINIT(sdt_##prov##_##mod##_##func##_##name##num##_init, \ + SI_SUB_KDTRACE, SI_ORDER_SECOND + 2, sdt_argtype_register, \ + sdt_##prov##_##mod##_##func##_##name##num ); \ + SYSUNINIT(sdt_##prov##_##mod##_##func##_##name##num##_uninit, \ + SI_SUB_KDTRACE, SI_ORDER_SECOND + 2, sdt_argtype_deregister, \ + sdt_##prov##_##mod##_##func##_##name##num ) + +typedef int (*sdt_argtype_listall_func_t)(struct sdt_argtype *, void *); +typedef int (*sdt_probe_listall_func_t)(struct sdt_probe *, void *); +typedef int (*sdt_provider_listall_func_t)(struct sdt_provider *, void *); + +void sdt_argtype_deregister(void *); +void sdt_argtype_register(void *); +void sdt_probe_deregister(void *); +void sdt_probe_register(void *); +void sdt_provider_deregister(void *); +void sdt_provider_register(void *); +void sdt_probe_stub(u_int32_t, uintptr_t arg0, uintptr_t arg1, uintptr_t arg2, + uintptr_t arg3, uintptr_t arg4); +int sdt_argtype_listall(struct sdt_probe *, sdt_argtype_listall_func_t, void *); +int sdt_probe_listall(struct sdt_provider *, sdt_probe_listall_func_t, void *); +int sdt_provider_listall(sdt_provider_listall_func_t,void *); + +#endif /* KDTRACE_HOOKS */ + +#endif /* _KERNEL */ + +#endif /* _SYS_SDT_H */ diff --git a/sys/sys/smp.h b/sys/sys/smp.h index f2b2f523213f..2fe44c27820f 100644 --- a/sys/sys/smp.h +++ b/sys/sys/smp.h @@ -100,6 +100,7 @@ void forward_roundrobin(void); int restart_cpus(cpumask_t); int stop_cpus(cpumask_t); void smp_rendezvous_action(void); +void smp_no_rendevous_barrier(void *); extern struct mtx smp_ipi_mtx; #endif /* SMP */ @@ -107,6 +108,11 @@ void smp_rendezvous(void (*)(void *), void (*)(void *), void (*)(void *), void *arg); +void smp_rendezvous_cpus(cpumask_t, + void (*)(void *), + void (*)(void *), + void (*)(void *), + void *arg); #endif /* !LOCORE */ #endif /* _KERNEL */ #endif /* _SYS_SMP_H_ */ diff --git a/sys/sys/stat.h b/sys/sys/stat.h index d3572c34e0ea..10a3739cc440 100644 --- a/sys/sys/stat.h +++ b/sys/sys/stat.h @@ -304,14 +304,6 @@ struct nstat { #endif /* __BSD_VISIBLE */ -/* - * Solaris compatibility definitions. - */ -#ifdef _SOLARIS_C_SOURCE -#define stat64 stat -#define fstat64 fstat -#endif - #ifndef _KERNEL __BEGIN_DECLS #if __BSD_VISIBLE diff --git a/sys/sys/sysent.h b/sys/sys/sysent.h index e0ea4b02fa0d..e712fc54c4a1 100644 --- a/sys/sys/sysent.h +++ b/sys/sys/sysent.h @@ -48,7 +48,7 @@ typedef void (*systrace_probe_func_t)(u_int32_t, int, struct sysent *, void *); * Used by loaded syscalls to convert arguments to a DTrace array * of 64-bit arguments. */ -typedef void (*systrace_args_func_t)(void *, u_int64_t *, int *); +typedef void (*systrace_args_func_t)(int, void *, u_int64_t *, int *); extern systrace_probe_func_t systrace_probe_func; diff --git a/sys/sys/systm.h b/sys/sys/systm.h index 10946db37c9c..9410b8728ab7 100644 --- a/sys/sys/systm.h +++ b/sys/sys/systm.h @@ -217,12 +217,6 @@ u_long casuword(volatile u_long *p, u_long oldval, u_long newval); void realitexpire(void *); -/* - * Cyclic clock function type definition used to hook the cyclic - * subsystem into the appropriate timer interrupt. - */ -typedef void (*cyclic_clock_func_t)(void); - void hardclock(int usermode, uintfptr_t pc); void hardclock_cpu(int usermode); void softclock(void *); diff --git a/sys/sys/time.h b/sys/sys/time.h index 0350c203eeb9..657f0d49b409 100644 --- a/sys/sys/time.h +++ b/sys/sys/time.h @@ -304,8 +304,6 @@ int ratecheck(struct timeval *, const struct timeval *); void timevaladd(struct timeval *t1, const struct timeval *t2); void timevalsub(struct timeval *t1, const struct timeval *t2); int tvtohz(struct timeval *tv); -uint64_t dtrace_gethrtime(void); -uint64_t dtrace_gethrestime(void); #else /* !_KERNEL */ #include <time.h> @@ -324,27 +322,4 @@ __END_DECLS #endif /* !_KERNEL */ -/* - * Solaris compatibility definitions. - */ -#ifdef _SOLARIS_C_SOURCE -/* - * Definitions for commonly used resolutions. - */ -#define SEC 1 -#define MILLISEC 1000 -#define MICROSEC 1000000 -#define NANOSEC 1000000000 - -typedef longlong_t hrtime_t; - -#ifndef _KERNEL -static __inline hrtime_t gethrtime(void) { - struct timespec ts; - clock_gettime(CLOCK_UPTIME,&ts); - return (((u_int64_t) ts.tv_sec) * NANOSEC + ts.tv_nsec); -} -#endif -#endif /* _SOLARIS_C_SOURCE */ - #endif /* !_SYS_TIME_H_ */ diff --git a/sys/sys/types.h b/sys/sys/types.h index 37b8b0534831..929407f49237 100644 --- a/sys/sys/types.h +++ b/sys/sys/types.h @@ -301,36 +301,6 @@ typedef struct vm_page *vm_page_t; #endif /* !_KERNEL */ /* - * Solaris compatibility definitions. - */ -#ifdef _SOLARIS_C_SOURCE -typedef u_int uint_t; -typedef u_char uchar_t; -typedef u_short ushort_t; -typedef u_long ulong_t; - -typedef long long longlong_t; -typedef unsigned long long u_longlong_t; - -typedef off_t off64_t; - -typedef id_t taskid_t; -typedef id_t projid_t; -typedef id_t poolid_t; -typedef id_t zoneid_t; -typedef id_t ctid_t; - -#ifndef _KERNEL -#if defined(__XOPEN_OR_POSIX) -typedef enum { _B_FALSE, _B_TRUE } boolean_t; -#else -typedef enum { B_FALSE, B_TRUE } boolean_t; -#endif /* defined(__XOPEN_OR_POSIX) */ -#endif - -#endif /* _SOLARIS_C_SOURCE */ - -/* * The following are all things that really shouldn't exist in this header, * since its purpose is to provide typedefs, not miscellaneous doodads. */ |