aboutsummaryrefslogtreecommitdiff
path: root/sys/amd64
diff options
context:
space:
mode:
Diffstat (limited to 'sys/amd64')
-rw-r--r--sys/amd64/acpica/acpi_wakeup.c4
-rw-r--r--sys/amd64/amd64/elf_machdep.c14
-rw-r--r--sys/amd64/include/vmm.h102
-rw-r--r--sys/amd64/linux/linux_sysvec.c12
-rw-r--r--sys/amd64/linux32/linux32_sysvec.c12
-rw-r--r--sys/amd64/pt/pt.c221
-rw-r--r--sys/amd64/vmm/vmm.c224
-rw-r--r--sys/amd64/vmm/vmm_dev_machdep.c1
8 files changed, 366 insertions, 224 deletions
diff --git a/sys/amd64/acpica/acpi_wakeup.c b/sys/amd64/acpica/acpi_wakeup.c
index 99565fbb69ca..8cada2f4f911 100644
--- a/sys/amd64/acpica/acpi_wakeup.c
+++ b/sys/amd64/acpica/acpi_wakeup.c
@@ -74,7 +74,7 @@ extern int acpi_susp_bounce;
extern struct susppcb **susppcbs;
static cpuset_t suspcpus;
-static void acpi_stop_beep(void *);
+static void acpi_stop_beep(void *, enum power_stype);
static int acpi_wakeup_ap(struct acpi_softc *, int);
static void acpi_wakeup_cpus(struct acpi_softc *);
@@ -88,7 +88,7 @@ static void acpi_wakeup_cpus(struct acpi_softc *);
} while (0)
static void
-acpi_stop_beep(void *arg)
+acpi_stop_beep(void *arg, enum power_stype stype)
{
if (acpi_resume_beep != 0)
diff --git a/sys/amd64/amd64/elf_machdep.c b/sys/amd64/amd64/elf_machdep.c
index 6cc2d58bbbcc..933f1ac0051f 100644
--- a/sys/amd64/amd64/elf_machdep.c
+++ b/sys/amd64/amd64/elf_machdep.c
@@ -179,7 +179,7 @@ freebsd_brand_info_la57_img_compat(const struct image_params *imgp,
return (!prefer_uva_la48);
}
-static Elf64_Brandinfo freebsd_brand_info_la48 = {
+static const Elf64_Brandinfo freebsd_brand_info_la48 = {
.brand = ELFOSABI_FREEBSD,
.machine = EM_X86_64,
.compat_3_brand = "FreeBSD",
@@ -190,7 +190,7 @@ static Elf64_Brandinfo freebsd_brand_info_la48 = {
.flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE,
};
-static Elf64_Brandinfo freebsd_brand_info_la57 = {
+static const Elf64_Brandinfo freebsd_brand_info_la57 = {
.brand = ELFOSABI_FREEBSD,
.machine = EM_X86_64,
.compat_3_brand = "FreeBSD",
@@ -216,7 +216,7 @@ sysinit_register_elf64_brand_entries(void *arg __unused)
SYSINIT(elf64, SI_SUB_EXEC, SI_ORDER_FIRST,
sysinit_register_elf64_brand_entries, NULL);
-static Elf64_Brandinfo freebsd_brand_oinfo = {
+static const Elf64_Brandinfo freebsd_brand_oinfo = {
.brand = ELFOSABI_FREEBSD,
.machine = EM_X86_64,
.compat_3_brand = "FreeBSD",
@@ -226,11 +226,10 @@ static Elf64_Brandinfo freebsd_brand_oinfo = {
.brand_note = &elf64_freebsd_brandnote,
.flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
};
-
-SYSINIT(oelf64, SI_SUB_EXEC, SI_ORDER_ANY,
+C_SYSINIT(oelf64, SI_SUB_EXEC, SI_ORDER_ANY,
(sysinit_cfunc_t)elf64_insert_brand_entry, &freebsd_brand_oinfo);
-static Elf64_Brandinfo kfreebsd_brand_info = {
+static const Elf64_Brandinfo kfreebsd_brand_info = {
.brand = ELFOSABI_FREEBSD,
.machine = EM_X86_64,
.compat_3_brand = "FreeBSD",
@@ -240,8 +239,7 @@ static Elf64_Brandinfo kfreebsd_brand_info = {
.brand_note = &elf64_kfreebsd_brandnote,
.flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE_MANDATORY
};
-
-SYSINIT(kelf64, SI_SUB_EXEC, SI_ORDER_ANY,
+C_SYSINIT(kelf64, SI_SUB_EXEC, SI_ORDER_ANY,
(sysinit_cfunc_t)elf64_insert_brand_entry, &kfreebsd_brand_info);
void
diff --git a/sys/amd64/include/vmm.h b/sys/amd64/include/vmm.h
index e35119af8572..ad67510fecf3 100644
--- a/sys/amd64/include/vmm.h
+++ b/sys/amd64/include/vmm.h
@@ -170,55 +170,63 @@ struct vm_eventinfo {
int *iptr; /* reqidle cookie */
};
-typedef int (*vmm_init_func_t)(int ipinum);
-typedef int (*vmm_cleanup_func_t)(void);
-typedef void (*vmm_suspend_func_t)(void);
-typedef void (*vmm_resume_func_t)(void);
-typedef void * (*vmi_init_func_t)(struct vm *vm, struct pmap *pmap);
-typedef int (*vmi_run_func_t)(void *vcpui, register_t rip,
- struct pmap *pmap, struct vm_eventinfo *info);
-typedef void (*vmi_cleanup_func_t)(void *vmi);
-typedef void * (*vmi_vcpu_init_func_t)(void *vmi, struct vcpu *vcpu,
- int vcpu_id);
-typedef void (*vmi_vcpu_cleanup_func_t)(void *vcpui);
-typedef int (*vmi_get_register_t)(void *vcpui, int num, uint64_t *retval);
-typedef int (*vmi_set_register_t)(void *vcpui, int num, uint64_t val);
-typedef int (*vmi_get_desc_t)(void *vcpui, int num, struct seg_desc *desc);
-typedef int (*vmi_set_desc_t)(void *vcpui, int num, struct seg_desc *desc);
-typedef int (*vmi_get_cap_t)(void *vcpui, int num, int *retval);
-typedef int (*vmi_set_cap_t)(void *vcpui, int num, int val);
-typedef struct vmspace * (*vmi_vmspace_alloc)(vm_offset_t min, vm_offset_t max);
-typedef void (*vmi_vmspace_free)(struct vmspace *vmspace);
-typedef struct vlapic * (*vmi_vlapic_init)(void *vcpui);
-typedef void (*vmi_vlapic_cleanup)(struct vlapic *vlapic);
-typedef int (*vmi_snapshot_vcpu_t)(void *vcpui, struct vm_snapshot_meta *meta);
-typedef int (*vmi_restore_tsc_t)(void *vcpui, uint64_t now);
+#define DECLARE_VMMOPS_FUNC(ret_type, opname, args) \
+ typedef ret_type (*vmmops_##opname##_t) args; \
+ ret_type vmmops_##opname args
+
+DECLARE_VMMOPS_FUNC(int, modinit, (int ipinum));
+DECLARE_VMMOPS_FUNC(int, modcleanup, (void));
+DECLARE_VMMOPS_FUNC(void, modresume, (void));
+DECLARE_VMMOPS_FUNC(void, modsuspend, (void));
+DECLARE_VMMOPS_FUNC(void *, init, (struct vm *vm, struct pmap *pmap));
+DECLARE_VMMOPS_FUNC(int, run, (void *vcpui, register_t pc,
+ struct pmap *pmap, struct vm_eventinfo *info));
+DECLARE_VMMOPS_FUNC(void, cleanup, (void *vmi));
+DECLARE_VMMOPS_FUNC(void *, vcpu_init, (void *vmi, struct vcpu *vcpu,
+ int vcpu_id));
+DECLARE_VMMOPS_FUNC(void, vcpu_cleanup, (void *vcpui));
+DECLARE_VMMOPS_FUNC(int, getreg, (void *vcpui, int num, uint64_t *retval));
+DECLARE_VMMOPS_FUNC(int, setreg, (void *vcpui, int num, uint64_t val));
+DECLARE_VMMOPS_FUNC(int, getdesc, (void *vcpui, int num,
+ struct seg_desc *desc));
+DECLARE_VMMOPS_FUNC(int, setdesc, (void *vcpui, int num,
+ struct seg_desc *desc));
+DECLARE_VMMOPS_FUNC(int, getcap, (void *vcpui, int num, int *retval));
+DECLARE_VMMOPS_FUNC(int, setcap, (void *vcpui, int num, int val));
+DECLARE_VMMOPS_FUNC(struct vmspace *, vmspace_alloc,
+ (vm_offset_t min, vm_offset_t max));
+DECLARE_VMMOPS_FUNC(void, vmspace_free, (struct vmspace *vmspace));
+DECLARE_VMMOPS_FUNC(struct vlapic *, vlapic_init, (void *vcpui));
+DECLARE_VMMOPS_FUNC(void, vlapic_cleanup, (struct vlapic *vlapic));
+DECLARE_VMMOPS_FUNC(int, vcpu_snapshot, (void *vcpui,
+ struct vm_snapshot_meta *meta));
+DECLARE_VMMOPS_FUNC(int, restore_tsc, (void *vcpui, uint64_t now));
struct vmm_ops {
- vmm_init_func_t modinit; /* module wide initialization */
- vmm_cleanup_func_t modcleanup;
- vmm_resume_func_t modsuspend;
- vmm_resume_func_t modresume;
-
- vmi_init_func_t init; /* vm-specific initialization */
- vmi_run_func_t run;
- vmi_cleanup_func_t cleanup;
- vmi_vcpu_init_func_t vcpu_init;
- vmi_vcpu_cleanup_func_t vcpu_cleanup;
- vmi_get_register_t getreg;
- vmi_set_register_t setreg;
- vmi_get_desc_t getdesc;
- vmi_set_desc_t setdesc;
- vmi_get_cap_t getcap;
- vmi_set_cap_t setcap;
- vmi_vmspace_alloc vmspace_alloc;
- vmi_vmspace_free vmspace_free;
- vmi_vlapic_init vlapic_init;
- vmi_vlapic_cleanup vlapic_cleanup;
+ vmmops_modinit_t modinit; /* module wide initialization */
+ vmmops_modcleanup_t modcleanup;
+ vmmops_modresume_t modsuspend;
+ vmmops_modresume_t modresume;
+
+ vmmops_init_t init; /* vm-specific initialization */
+ vmmops_run_t run;
+ vmmops_cleanup_t cleanup;
+ vmmops_vcpu_init_t vcpu_init;
+ vmmops_vcpu_cleanup_t vcpu_cleanup;
+ vmmops_getreg_t getreg;
+ vmmops_setreg_t setreg;
+ vmmops_getdesc_t getdesc;
+ vmmops_setdesc_t setdesc;
+ vmmops_getcap_t getcap;
+ vmmops_setcap_t setcap;
+ vmmops_vmspace_alloc_t vmspace_alloc;
+ vmmops_vmspace_free_t vmspace_free;
+ vmmops_vlapic_init_t vlapic_init;
+ vmmops_vlapic_cleanup_t vlapic_cleanup;
/* checkpoint operations */
- vmi_snapshot_vcpu_t vcpu_snapshot;
- vmi_restore_tsc_t restore_tsc;
+ vmmops_vcpu_snapshot_t vcpu_snapshot;
+ vmmops_restore_tsc_t restore_tsc;
};
extern const struct vmm_ops vmm_ops_intel;
@@ -229,7 +237,7 @@ extern u_int vm_maxcpu; /* maximum virtual cpus */
int vm_create(const char *name, struct vm **retvm);
struct vcpu *vm_alloc_vcpu(struct vm *vm, int vcpuid);
void vm_disable_vcpu_creation(struct vm *vm);
-void vm_slock_vcpus(struct vm *vm);
+void vm_lock_vcpus(struct vm *vm);
void vm_unlock_vcpus(struct vm *vm);
void vm_destroy(struct vm *vm);
int vm_reinit(struct vm *vm);
@@ -354,6 +362,7 @@ enum vcpu_state {
};
int vcpu_set_state(struct vcpu *vcpu, enum vcpu_state state, bool from_idle);
+int vcpu_set_state_all(struct vm *vm, enum vcpu_state state);
enum vcpu_state vcpu_get_state(struct vcpu *vcpu, int *hostcpu);
static int __inline
@@ -375,7 +384,6 @@ vcpu_should_yield(struct vcpu *vcpu)
void *vcpu_stats(struct vcpu *vcpu);
void vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr);
-struct vmspace *vm_vmspace(struct vm *vm);
struct vm_mem *vm_mem(struct vm *vm);
struct vatpic *vm_atpic(struct vm *vm);
struct vatpit *vm_atpit(struct vm *vm);
diff --git a/sys/amd64/linux/linux_sysvec.c b/sys/amd64/linux/linux_sysvec.c
index c8579c5da4ad..890cf01c46a0 100644
--- a/sys/amd64/linux/linux_sysvec.c
+++ b/sys/amd64/linux/linux_sysvec.c
@@ -857,7 +857,7 @@ linux_vdso_reloc(char *mapping, Elf_Addr offset)
}
}
-static Elf_Brandnote linux64_brandnote = {
+static const Elf_Brandnote linux64_brandnote = {
.hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
.hdr.n_descsz = 16,
.hdr.n_type = 1,
@@ -866,7 +866,7 @@ static Elf_Brandnote linux64_brandnote = {
.trans_osrel = linux_trans_osrel
};
-static Elf64_Brandinfo linux_glibc2brand = {
+static const Elf64_Brandinfo linux_glibc2brand = {
.brand = ELFOSABI_LINUX,
.machine = EM_X86_64,
.compat_3_brand = "Linux",
@@ -877,7 +877,7 @@ static Elf64_Brandinfo linux_glibc2brand = {
.flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
};
-static Elf64_Brandinfo linux_glibc2brandshort = {
+static const Elf64_Brandinfo linux_glibc2brandshort = {
.brand = ELFOSABI_LINUX,
.machine = EM_X86_64,
.compat_3_brand = "Linux",
@@ -888,7 +888,7 @@ static Elf64_Brandinfo linux_glibc2brandshort = {
.flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
};
-static Elf64_Brandinfo linux_muslbrand = {
+static const Elf64_Brandinfo linux_muslbrand = {
.brand = ELFOSABI_LINUX,
.machine = EM_X86_64,
.compat_3_brand = "Linux",
@@ -900,7 +900,7 @@ static Elf64_Brandinfo linux_muslbrand = {
LINUX_BI_FUTEX_REQUEUE
};
-static Elf64_Brandinfo *linux_brandlist[] = {
+static const Elf64_Brandinfo *linux_brandlist[] = {
&linux_glibc2brand,
&linux_glibc2brandshort,
&linux_muslbrand,
@@ -910,7 +910,7 @@ static Elf64_Brandinfo *linux_brandlist[] = {
static int
linux64_elf_modevent(module_t mod, int type, void *data)
{
- Elf64_Brandinfo **brandinfo;
+ const Elf64_Brandinfo **brandinfo;
int error;
struct linux_ioctl_handler **lihp;
diff --git a/sys/amd64/linux32/linux32_sysvec.c b/sys/amd64/linux32/linux32_sysvec.c
index 8fac626f9053..735ebb151017 100644
--- a/sys/amd64/linux32/linux32_sysvec.c
+++ b/sys/amd64/linux32/linux32_sysvec.c
@@ -954,7 +954,7 @@ linux_vdso_reloc(char *mapping, Elf_Addr offset)
}
}
-static Elf_Brandnote linux32_brandnote = {
+static const Elf_Brandnote linux32_brandnote = {
.hdr.n_namesz = sizeof(GNU_ABI_VENDOR),
.hdr.n_descsz = 16, /* XXX at least 16 */
.hdr.n_type = 1,
@@ -963,7 +963,7 @@ static Elf_Brandnote linux32_brandnote = {
.trans_osrel = linux_trans_osrel
};
-static Elf32_Brandinfo linux_brand = {
+static const Elf32_Brandinfo linux_brand = {
.brand = ELFOSABI_LINUX,
.machine = EM_386,
.compat_3_brand = "Linux",
@@ -974,7 +974,7 @@ static Elf32_Brandinfo linux_brand = {
.flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
};
-static Elf32_Brandinfo linux_glibc2brand = {
+static const Elf32_Brandinfo linux_glibc2brand = {
.brand = ELFOSABI_LINUX,
.machine = EM_386,
.compat_3_brand = "Linux",
@@ -985,7 +985,7 @@ static Elf32_Brandinfo linux_glibc2brand = {
.flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
};
-static Elf32_Brandinfo linux_muslbrand = {
+static const Elf32_Brandinfo linux_muslbrand = {
.brand = ELFOSABI_LINUX,
.machine = EM_386,
.compat_3_brand = "Linux",
@@ -997,7 +997,7 @@ static Elf32_Brandinfo linux_muslbrand = {
LINUX_BI_FUTEX_REQUEUE
};
-static Elf32_Brandinfo *linux_brandlist[] = {
+static const Elf32_Brandinfo *linux_brandlist[] = {
&linux_brand,
&linux_glibc2brand,
&linux_muslbrand,
@@ -1007,7 +1007,7 @@ static Elf32_Brandinfo *linux_brandlist[] = {
static int
linux_elf_modevent(module_t mod, int type, void *data)
{
- Elf32_Brandinfo **brandinfo;
+ const Elf32_Brandinfo **brandinfo;
int error;
struct linux_ioctl_handler **lihp;
diff --git a/sys/amd64/pt/pt.c b/sys/amd64/pt/pt.c
index c7b75767680a..6b2296de049c 100644
--- a/sys/amd64/pt/pt.c
+++ b/sys/amd64/pt/pt.c
@@ -42,15 +42,15 @@
*/
#include <sys/systm.h>
+#include <sys/bus.h>
#include <sys/hwt.h>
+#include <sys/interrupt.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/module.h>
#include <sys/mutex.h>
-#include <sys/sdt.h>
#include <sys/smp.h>
-#include <sys/taskqueue.h>
#include <vm/vm.h>
#include <vm/vm_page.h>
@@ -94,12 +94,7 @@
MALLOC_DEFINE(M_PT, "pt", "Intel Processor Trace");
-SDT_PROVIDER_DEFINE(pt);
-SDT_PROBE_DEFINE(pt, , , topa__intr);
-
-TASKQUEUE_FAST_DEFINE_THREAD(pt);
-
-static void pt_send_buffer_record(void *arg, int pending __unused);
+static void pt_send_buffer_record(void *arg);
static int pt_topa_intr(struct trapframe *tf);
/*
@@ -122,29 +117,24 @@ struct pt_buffer {
size_t size;
struct mtx lock; /* Lock for fields below. */
vm_offset_t offset;
- uint64_t wrap_count;
- int curpage;
};
struct pt_ctx {
int id;
struct pt_buffer buf; /* ToPA buffer metadata */
- struct task task; /* ToPA buffer notification task */
struct hwt_context *hwt_ctx;
uint8_t *save_area; /* PT XSAVE area */
};
/* PT tracing contexts used for CPU mode. */
static struct pt_ctx *pt_pcpu_ctx;
-enum pt_cpu_state {
- PT_DISABLED = 0,
- PT_STOPPED,
- PT_ACTIVE
-};
+enum pt_cpu_state { PT_INACTIVE = 0, PT_ACTIVE };
static struct pt_cpu {
struct pt_ctx *ctx; /* active PT tracing context */
enum pt_cpu_state state; /* used as part of trace stop protocol */
+ void *swi_cookie; /* Software interrupt handler context */
+ int in_pcint_handler;
} *pt_pcpu;
/*
@@ -199,31 +189,28 @@ static __inline void
pt_update_buffer(struct pt_buffer *buf)
{
uint64_t reg;
- int curpage;
+ uint64_t offset;
/* Update buffer offset. */
reg = rdmsr(MSR_IA32_RTIT_OUTPUT_MASK_PTRS);
- curpage = (reg & PT_TOPA_PAGE_MASK) >> PT_TOPA_PAGE_SHIFT;
- mtx_lock_spin(&buf->lock);
- /* Check if the output wrapped. */
- if (buf->curpage > curpage)
- buf->wrap_count++;
- buf->curpage = curpage;
- buf->offset = reg >> 32;
- mtx_unlock_spin(&buf->lock);
-
- dprintf("%s: wrap_cnt: %lu, curpage: %d, offset: %zu\n", __func__,
- buf->wrap_count, buf->curpage, buf->offset);
+ offset = ((reg & PT_TOPA_PAGE_MASK) >> PT_TOPA_PAGE_SHIFT) * PAGE_SIZE;
+ offset += (reg >> 32);
+
+ atomic_store_rel_64(&buf->offset, offset);
}
static __inline void
pt_fill_buffer_record(int id, struct pt_buffer *buf,
struct hwt_record_entry *rec)
{
+ vm_offset_t offset;
+
+ offset = atomic_load_acq_64(&buf->offset);
+
rec->record_type = HWT_RECORD_BUFFER;
rec->buf_id = id;
- rec->curpage = buf->curpage;
- rec->offset = buf->offset + (buf->wrap_count * buf->size);
+ rec->curpage = offset / PAGE_SIZE;
+ rec->offset = offset & PAGE_MASK;
}
/*
@@ -273,9 +260,9 @@ pt_cpu_start(void *dummy)
MPASS(cpu->ctx != NULL);
dprintf("%s: curcpu %d\n", __func__, curcpu);
+ pt_cpu_set_state(curcpu, PT_ACTIVE);
load_cr4(rcr4() | CR4_XSAVE);
wrmsr(MSR_IA32_RTIT_STATUS, 0);
- pt_cpu_set_state(curcpu, PT_ACTIVE);
pt_cpu_toggle_local(cpu->ctx->save_area, true);
}
@@ -291,16 +278,16 @@ pt_cpu_stop(void *dummy)
struct pt_cpu *cpu;
struct pt_ctx *ctx;
- /* Shutdown may occur before PT gets properly configured. */
- if (pt_cpu_get_state(curcpu) == PT_DISABLED)
- return;
-
cpu = &pt_pcpu[curcpu];
ctx = cpu->ctx;
- MPASS(ctx != NULL);
- dprintf("%s: curcpu %d\n", __func__, curcpu);
- pt_cpu_set_state(curcpu, PT_STOPPED);
+ dprintf("%s: curcpu %d\n", __func__, curcpu);
+ /* Shutdown may occur before PT gets properly configured. */
+ if (ctx == NULL) {
+ dprintf("%s: missing context on cpu %d; bailing\n", __func__,
+ curcpu);
+ return;
+ }
pt_cpu_toggle_local(cpu->ctx->save_area, false);
pt_update_buffer(&ctx->buf);
}
@@ -406,13 +393,11 @@ pt_init_ctx(struct pt_ctx *pt_ctx, struct hwt_vm *vm, int ctx_id)
return (ENOMEM);
dprintf("%s: preparing ToPA buffer\n", __func__);
if (pt_topa_prepare(pt_ctx, vm) != 0) {
- dprintf("%s: failed to prepare ToPA buffer\n", __func__);
free(pt_ctx->save_area, M_PT);
return (ENOMEM);
}
pt_ctx->id = ctx_id;
- TASK_INIT(&pt_ctx->task, 0, pt_send_buffer_record, pt_ctx);
return (0);
}
@@ -426,7 +411,6 @@ pt_deinit_ctx(struct pt_ctx *pt_ctx)
if (pt_ctx->save_area != NULL)
free(pt_ctx->save_area, M_PT);
memset(pt_ctx, 0, sizeof(*pt_ctx));
- pt_ctx->buf.topa_hw = NULL;
}
/*
@@ -519,7 +503,6 @@ pt_backend_configure(struct hwt_context *ctx, int cpu_id, int thread_id)
XSTATE_XCOMP_BV_COMPACT;
pt_ext->rtit_ctl |= RTIT_CTL_TRACEEN;
pt_pcpu[cpu_id].ctx = pt_ctx;
- pt_cpu_set_state(cpu_id, PT_STOPPED);
return (0);
}
@@ -549,12 +532,19 @@ pt_backend_disable(struct hwt_context *ctx, int cpu_id)
if (ctx->mode == HWT_MODE_CPU)
return;
-
KASSERT(curcpu == cpu_id,
("%s: attempting to disable PT on another cpu", __func__));
+
+ cpu = &pt_pcpu[cpu_id];
+
+ dprintf("%s: waiting for cpu %d to exit interrupt handler\n", __func__,
+ cpu_id);
+ pt_cpu_set_state(cpu_id, PT_INACTIVE);
+ while (atomic_cmpset_int(&cpu->in_pcint_handler, 1, 0))
+ ;
+
pt_cpu_stop(NULL);
CPU_CLR(cpu_id, &ctx->cpu_map);
- cpu = &pt_pcpu[cpu_id];
cpu->ctx = NULL;
}
@@ -564,14 +554,14 @@ pt_backend_disable(struct hwt_context *ctx, int cpu_id)
static int
pt_backend_enable_smp(struct hwt_context *ctx)
{
-
dprintf("%s\n", __func__);
+
+ KASSERT(ctx->mode == HWT_MODE_CPU,
+ ("%s: should only be used for CPU mode", __func__));
if (ctx->mode == HWT_MODE_CPU &&
atomic_swap_32(&cpu_mode_ctr, 1) != 0)
return (-1);
- KASSERT(ctx->mode == HWT_MODE_CPU,
- ("%s: should only be used for CPU mode", __func__));
smp_rendezvous_cpus(ctx->cpu_map, NULL, pt_cpu_start, NULL, NULL);
return (0);
@@ -583,6 +573,7 @@ pt_backend_enable_smp(struct hwt_context *ctx)
static int
pt_backend_disable_smp(struct hwt_context *ctx)
{
+ struct pt_cpu *cpu;
dprintf("%s\n", __func__);
if (ctx->mode == HWT_MODE_CPU &&
@@ -593,6 +584,14 @@ pt_backend_disable_smp(struct hwt_context *ctx)
dprintf("%s: empty cpu map\n", __func__);
return (-1);
}
+ CPU_FOREACH_ISSET(cpu_id, &ctx->cpu_map) {
+ cpu = &pt_pcpu[cpu_id];
+ dprintf("%s: waiting for cpu %d to exit interrupt handler\n",
+ __func__, cpu_id);
+ pt_cpu_set_state(cpu_id, PT_INACTIVE);
+ while (atomic_cmpset_int(&cpu->in_pcint_handler, 1, 0))
+ ;
+ }
smp_rendezvous_cpus(ctx->cpu_map, NULL, pt_cpu_stop, NULL, NULL);
return (0);
@@ -611,13 +610,13 @@ pt_backend_init(struct hwt_context *ctx)
int error;
dprintf("%s\n", __func__);
- if (ctx->mode == HWT_MODE_CPU) {
- TAILQ_FOREACH(hwt_cpu, &ctx->cpus, next) {
- error = pt_init_ctx(&pt_pcpu_ctx[hwt_cpu->cpu_id],
- hwt_cpu->vm, hwt_cpu->cpu_id);
- if (error)
- return (error);
- }
+ if (ctx->mode != HWT_MODE_CPU)
+ return (0);
+ TAILQ_FOREACH(hwt_cpu, &ctx->cpus, next) {
+ error = pt_init_ctx(&pt_pcpu_ctx[hwt_cpu->cpu_id], hwt_cpu->vm,
+ hwt_cpu->cpu_id);
+ if (error)
+ return (error);
}
return (0);
@@ -647,20 +646,16 @@ pt_backend_deinit(struct hwt_context *ctx)
pt_deinit_ctx(pt_ctx);
}
} else {
- CPU_FOREACH(cpu_id) {
- if (!CPU_ISSET(cpu_id, &ctx->cpu_map))
+ CPU_FOREACH_ISSET(cpu_id, &ctx->cpu_map) {
+ if (pt_pcpu[cpu_id].ctx == NULL)
continue;
- if (pt_pcpu[cpu_id].ctx != NULL) {
- KASSERT(pt_pcpu[cpu_id].ctx ==
- &pt_pcpu_ctx[cpu_id],
- ("%s: CPU mode tracing with non-cpu mode PT"
- "context active",
- __func__));
- pt_pcpu[cpu_id].ctx = NULL;
- }
- pt_ctx = &pt_pcpu_ctx[cpu_id];
- pt_deinit_ctx(pt_ctx);
- memset(&pt_pcpu[cpu_id], 0, sizeof(struct pt_cpu));
+ KASSERT(pt_pcpu[cpu_id].ctx == &pt_pcpu_ctx[cpu_id],
+ ("%s: CPU mode tracing with non-cpu mode PT"
+ "context active",
+ __func__));
+ pt_deinit_ctx(pt_pcpu[cpu_id].ctx);
+ pt_pcpu[cpu_id].ctx = NULL;
+ atomic_set_int(&pt_pcpu[cpu_id].in_pcint_handler, 0);
}
}
@@ -675,15 +670,15 @@ pt_backend_read(struct hwt_vm *vm, int *curpage, vm_offset_t *curpage_offset,
uint64_t *data)
{
struct pt_buffer *buf;
+ uint64_t offset;
if (vm->ctx->mode == HWT_MODE_THREAD)
buf = &((struct pt_ctx *)vm->thr->private)->buf;
else
buf = &pt_pcpu[vm->cpu->cpu_id].ctx->buf;
- mtx_lock_spin(&buf->lock);
- *curpage = buf->curpage;
- *curpage_offset = buf->offset + (buf->wrap_count * vm->ctx->bufsize);
- mtx_unlock_spin(&buf->lock);
+ offset = atomic_load_acq_64(&buf->offset);
+ *curpage = offset / PAGE_SIZE;
+ *curpage_offset = offset & PAGE_MASK;
return (0);
}
@@ -762,15 +757,13 @@ static struct hwt_backend backend = {
* Used as a taskqueue routine from the ToPA interrupt handler.
*/
static void
-pt_send_buffer_record(void *arg, int pending __unused)
+pt_send_buffer_record(void *arg)
{
+ struct pt_cpu *cpu = (struct pt_cpu *)arg;
struct hwt_record_entry record;
- struct pt_ctx *ctx = (struct pt_ctx *)arg;
- /* Prepare buffer record. */
- mtx_lock_spin(&ctx->buf.lock);
+ struct pt_ctx *ctx = cpu->ctx;
pt_fill_buffer_record(ctx->id, &ctx->buf, &record);
- mtx_unlock_spin(&ctx->buf.lock);
hwt_record_ctx(ctx->hwt_ctx, &record, M_ZERO | M_NOWAIT);
}
static void
@@ -795,36 +788,40 @@ static int
pt_topa_intr(struct trapframe *tf)
{
struct pt_buffer *buf;
+ struct pt_cpu *cpu;
struct pt_ctx *ctx;
uint64_t reg;
- SDT_PROBE0(pt, , , topa__intr);
-
- if (pt_cpu_get_state(curcpu) != PT_ACTIVE) {
- return (0);
- }
+ cpu = &pt_pcpu[curcpu];
reg = rdmsr(MSR_IA_GLOBAL_STATUS);
if ((reg & GLOBAL_STATUS_FLAG_TRACETOPAPMI) == 0) {
- /* ACK spurious or leftover interrupt. */
pt_topa_status_clear();
+ return (0);
+ }
+
+ if (pt_cpu_get_state(curcpu) != PT_ACTIVE) {
return (1);
}
+ atomic_set_int(&cpu->in_pcint_handler, 1);
- ctx = pt_pcpu[curcpu].ctx;
+ ctx = cpu->ctx;
+ KASSERT(ctx != NULL,
+ ("%s: cpu %d: ToPA PMI interrupt without an active context",
+ __func__, curcpu));
buf = &ctx->buf;
KASSERT(buf->topa_hw != NULL,
- ("%s: ToPA PMI interrupt with invalid buffer", __func__));
-
+ ("%s: cpu %d: ToPA PMI interrupt with invalid buffer", __func__,
+ curcpu));
pt_cpu_toggle_local(ctx->save_area, false);
pt_update_buffer(buf);
pt_topa_status_clear();
- taskqueue_enqueue_flags(taskqueue_pt, &ctx->task,
- TASKQUEUE_FAIL_IF_PENDING);
if (pt_cpu_get_state(curcpu) == PT_ACTIVE) {
+ swi_sched(cpu->swi_cookie, SWI_FROMNMI);
pt_cpu_toggle_local(ctx->save_area, true);
lapic_reenable_pcint();
}
+ atomic_set_int(&cpu->in_pcint_handler, 0);
return (1);
}
@@ -839,7 +836,7 @@ static int
pt_init(void)
{
u_int cp[4];
- int error;
+ int error, i;
dprintf("pt: Enumerating part 1\n");
cpuid_count(CPUID_PT_LEAF, 0, cp);
@@ -869,20 +866,38 @@ pt_init(void)
pt_pcpu_ctx = mallocarray(mp_ncpus, sizeof(struct pt_ctx), M_PT,
M_ZERO | M_WAITOK);
+ for (i = 0; i < mp_ncpus; i++) {
+ error = swi_add(&clk_intr_event, "pt", pt_send_buffer_record,
+ &pt_pcpu[i], SWI_CLOCK, INTR_MPSAFE,
+ &pt_pcpu[i].swi_cookie);
+ if (error != 0) {
+ dprintf(
+ "%s: failed to add interrupt handler for cpu: %d\n",
+ __func__, error);
+ goto err;
+ }
+ }
+
nmi_register_handler(pt_topa_intr);
- if (!lapic_enable_pcint()) {
- nmi_remove_handler(pt_topa_intr);
- hwt_backend_unregister(&backend);
- free(pt_pcpu, M_PT);
- free(pt_pcpu_ctx, M_PT);
- pt_pcpu = NULL;
- pt_pcpu_ctx = NULL;
+ if (lapic_enable_pcint()) {
+ initialized = true;
+ return (0);
+ } else
printf("pt: failed to setup interrupt line\n");
- return (error);
+err:
+ nmi_remove_handler(pt_topa_intr);
+ hwt_backend_unregister(&backend);
+
+ for (i = 0; i < mp_ncpus; i++) {
+ if (pt_pcpu[i].swi_cookie != 0)
+ swi_remove(pt_pcpu[i].swi_cookie);
}
- initialized = true;
+ free(pt_pcpu, M_PT);
+ free(pt_pcpu_ctx, M_PT);
+ pt_pcpu = NULL;
+ pt_pcpu_ctx = NULL;
- return (0);
+ return (error);
}
/*
@@ -941,14 +956,24 @@ pt_supported(void)
static void
pt_deinit(void)
{
+ int i;
+ struct pt_cpu *cpu;
+
if (!initialized)
return;
nmi_remove_handler(pt_topa_intr);
lapic_disable_pcint();
hwt_backend_unregister(&backend);
+
+ for (i = 0; i < mp_ncpus; i++) {
+ cpu = &pt_pcpu[i];
+ swi_remove(cpu->swi_cookie);
+ }
+
free(pt_pcpu, M_PT);
free(pt_pcpu_ctx, M_PT);
pt_pcpu = NULL;
+ pt_pcpu_ctx = NULL;
initialized = false;
}
diff --git a/sys/amd64/vmm/vmm.c b/sys/amd64/vmm/vmm.c
index c42da02d0bf6..f7c59847140b 100644
--- a/sys/amd64/vmm/vmm.c
+++ b/sys/amd64/vmm/vmm.c
@@ -163,7 +163,6 @@ struct vm {
void *rendezvous_arg; /* (x) [r] rendezvous func/arg */
vm_rendezvous_func_t rendezvous_func;
struct mtx rendezvous_mtx; /* (o) rendezvous lock */
- struct vmspace *vmspace; /* (o) guest's address space */
struct vm_mem mem; /* (i) [m+v] guest memory */
char name[VM_MAX_NAMELEN+1]; /* (o) virtual machine name */
struct vcpu **vcpu; /* (o) guest vcpus */
@@ -201,7 +200,7 @@ vmmops_panic(void)
}
#define DEFINE_VMMOPS_IFUNC(ret_type, opname, args) \
- DEFINE_IFUNC(static, ret_type, vmmops_##opname, args) \
+ DEFINE_IFUNC(, ret_type, vmmops_##opname, args) \
{ \
if (vmm_is_intel()) \
return (vmm_ops_intel.opname); \
@@ -499,7 +498,7 @@ MODULE_VERSION(vmm, 1);
static void
vm_init(struct vm *vm, bool create)
{
- vm->cookie = vmmops_init(vm, vmspace_pmap(vm->vmspace));
+ vm->cookie = vmmops_init(vm, vmspace_pmap(vm_vmspace(vm)));
vm->iommu = NULL;
vm->vioapic = vioapic_init(vm);
vm->vhpet = vhpet_init(vm);
@@ -563,9 +562,9 @@ vm_alloc_vcpu(struct vm *vm, int vcpuid)
}
void
-vm_slock_vcpus(struct vm *vm)
+vm_lock_vcpus(struct vm *vm)
{
- sx_slock(&vm->vcpus_init_lock);
+ sx_xlock(&vm->vcpus_init_lock);
}
void
@@ -584,7 +583,7 @@ int
vm_create(const char *name, struct vm **retvm)
{
struct vm *vm;
- struct vmspace *vmspace;
+ int error;
/*
* If vmm.ko could not be successfully initialized then don't attempt
@@ -597,14 +596,13 @@ vm_create(const char *name, struct vm **retvm)
VM_MAX_NAMELEN + 1)
return (EINVAL);
- vmspace = vmmops_vmspace_alloc(0, VM_MAXUSER_ADDRESS_LA48);
- if (vmspace == NULL)
- return (ENOMEM);
-
vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
+ error = vm_mem_init(&vm->mem, 0, VM_MAXUSER_ADDRESS_LA48);
+ if (error != 0) {
+ free(vm, M_VM);
+ return (error);
+ }
strcpy(vm->name, name);
- vm->vmspace = vmspace;
- vm_mem_init(&vm->mem);
mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
sx_init(&vm->vcpus_init_lock, "vm vcpus");
vm->vcpu = malloc(sizeof(*vm->vcpu) * vm_maxcpu, M_VM, M_WAITOK |
@@ -685,9 +683,6 @@ vm_cleanup(struct vm *vm, bool destroy)
if (destroy) {
vm_mem_destroy(vm);
- vmmops_vmspace_free(vm->vmspace);
- vm->vmspace = NULL;
-
free(vm->vcpu, M_VM);
sx_destroy(&vm->vcpus_init_lock);
mtx_destroy(&vm->rendezvous_mtx);
@@ -731,7 +726,7 @@ vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
{
vm_object_t obj;
- if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
+ if ((obj = vmm_mmio_alloc(vm_vmspace(vm), gpa, len, hpa)) == NULL)
return (ENOMEM);
else
return (0);
@@ -741,19 +736,21 @@ int
vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
{
- vmm_mmio_free(vm->vmspace, gpa, len);
+ vmm_mmio_free(vm_vmspace(vm), gpa, len);
return (0);
}
static int
vm_iommu_map(struct vm *vm)
{
+ pmap_t pmap;
vm_paddr_t gpa, hpa;
struct vm_mem_map *mm;
int error, i;
sx_assert(&vm->mem.mem_segs_lock, SX_LOCKED);
+ pmap = vmspace_pmap(vm_vmspace(vm));
for (i = 0; i < VM_MAX_MEMMAPS; i++) {
if (!vm_memseg_sysmem(vm, i))
continue;
@@ -767,7 +764,7 @@ vm_iommu_map(struct vm *vm)
mm->flags |= VM_MEMMAP_F_IOMMU;
for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) {
- hpa = pmap_extract(vmspace_pmap(vm->vmspace), gpa);
+ hpa = pmap_extract(pmap, gpa);
/*
* All mappings in the vmm vmspace must be
@@ -816,7 +813,7 @@ vm_iommu_unmap(struct vm *vm)
for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) {
KASSERT(vm_page_wired(PHYS_TO_VM_PAGE(pmap_extract(
- vmspace_pmap(vm->vmspace), gpa))),
+ vmspace_pmap(vm_vmspace(vm)), gpa))),
("vm_iommu_unmap: vm %p gpa %jx not wired",
vm, (uintmax_t)gpa));
iommu_remove_mapping(vm->iommu, gpa, PAGE_SIZE);
@@ -993,6 +990,54 @@ save_guest_fpustate(struct vcpu *vcpu)
static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
+/*
+ * Invoke the rendezvous function on the specified vcpu if applicable. Return
+ * true if the rendezvous is finished, false otherwise.
+ */
+static bool
+vm_rendezvous(struct vcpu *vcpu)
+{
+ struct vm *vm = vcpu->vm;
+ int vcpuid;
+
+ mtx_assert(&vcpu->vm->rendezvous_mtx, MA_OWNED);
+ KASSERT(vcpu->vm->rendezvous_func != NULL,
+ ("vm_rendezvous: no rendezvous pending"));
+
+ /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
+ CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus,
+ &vm->active_cpus);
+
+ vcpuid = vcpu->vcpuid;
+ if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
+ !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
+ VMM_CTR0(vcpu, "Calling rendezvous func");
+ (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg);
+ CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
+ }
+ if (CPU_CMP(&vm->rendezvous_req_cpus,
+ &vm->rendezvous_done_cpus) == 0) {
+ VMM_CTR0(vcpu, "Rendezvous completed");
+ CPU_ZERO(&vm->rendezvous_req_cpus);
+ vm->rendezvous_func = NULL;
+ wakeup(&vm->rendezvous_func);
+ return (true);
+ }
+ return (false);
+}
+
+static void
+vcpu_wait_idle(struct vcpu *vcpu)
+{
+ KASSERT(vcpu->state != VCPU_IDLE, ("vcpu already idle"));
+
+ vcpu->reqidle = 1;
+ vcpu_notify_event_locked(vcpu, false);
+ VMM_CTR1(vcpu, "vcpu state change from %s to "
+ "idle requested", vcpu_state2str(vcpu->state));
+ msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
+}
+
static int
vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
bool from_idle)
@@ -1007,13 +1052,8 @@ vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
* ioctl() operating on a vcpu at any point.
*/
if (from_idle) {
- while (vcpu->state != VCPU_IDLE) {
- vcpu->reqidle = 1;
- vcpu_notify_event_locked(vcpu, false);
- VMM_CTR1(vcpu, "vcpu state change from %s to "
- "idle requested", vcpu_state2str(vcpu->state));
- msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
- }
+ while (vcpu->state != VCPU_IDLE)
+ vcpu_wait_idle(vcpu);
} else {
KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
"vcpu idle state"));
@@ -1065,6 +1105,95 @@ vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
return (0);
}
+/*
+ * Try to lock all of the vCPUs in the VM while taking care to avoid deadlocks
+ * with vm_smp_rendezvous().
+ *
+ * The complexity here suggests that the rendezvous mechanism needs a rethink.
+ */
+int
+vcpu_set_state_all(struct vm *vm, enum vcpu_state newstate)
+{
+ cpuset_t locked;
+ struct vcpu *vcpu;
+ int error, i;
+ uint16_t maxcpus;
+
+ KASSERT(newstate != VCPU_IDLE,
+ ("vcpu_set_state_all: invalid target state %d", newstate));
+
+ error = 0;
+ CPU_ZERO(&locked);
+ maxcpus = vm->maxcpus;
+
+ mtx_lock(&vm->rendezvous_mtx);
+restart:
+ if (vm->rendezvous_func != NULL) {
+ /*
+ * If we have a pending rendezvous, then the initiator may be
+ * blocked waiting for other vCPUs to execute the callback. The
+ * current thread may be a vCPU thread so we must not block
+ * waiting for the initiator, otherwise we get a deadlock.
+ * Thus, execute the callback on behalf of any idle vCPUs.
+ */
+ for (i = 0; i < maxcpus; i++) {
+ vcpu = vm_vcpu(vm, i);
+ if (vcpu == NULL)
+ continue;
+ vcpu_lock(vcpu);
+ if (vcpu->state == VCPU_IDLE) {
+ (void)vcpu_set_state_locked(vcpu, VCPU_FROZEN,
+ true);
+ CPU_SET(i, &locked);
+ }
+ if (CPU_ISSET(i, &locked)) {
+ /*
+ * We can safely execute the callback on this
+ * vCPU's behalf.
+ */
+ vcpu_unlock(vcpu);
+ (void)vm_rendezvous(vcpu);
+ vcpu_lock(vcpu);
+ }
+ vcpu_unlock(vcpu);
+ }
+ }
+
+ /*
+ * Now wait for remaining vCPUs to become idle. This may include the
+ * initiator of a rendezvous that is currently blocked on the rendezvous
+ * mutex.
+ */
+ CPU_FOREACH_ISCLR(i, &locked) {
+ if (i >= maxcpus)
+ break;
+ vcpu = vm_vcpu(vm, i);
+ if (vcpu == NULL)
+ continue;
+ vcpu_lock(vcpu);
+ while (vcpu->state != VCPU_IDLE) {
+ mtx_unlock(&vm->rendezvous_mtx);
+ vcpu_wait_idle(vcpu);
+ vcpu_unlock(vcpu);
+ mtx_lock(&vm->rendezvous_mtx);
+ if (vm->rendezvous_func != NULL)
+ goto restart;
+ vcpu_lock(vcpu);
+ }
+ error = vcpu_set_state_locked(vcpu, newstate, true);
+ vcpu_unlock(vcpu);
+ if (error != 0) {
+ /* Roll back state changes. */
+ CPU_FOREACH_ISSET(i, &locked)
+ (void)vcpu_set_state(vcpu, VCPU_IDLE, false);
+ break;
+ }
+ CPU_SET(i, &locked);
+ }
+ mtx_unlock(&vm->rendezvous_mtx);
+ return (error);
+}
+
static void
vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate)
{
@@ -1086,36 +1215,23 @@ vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
static int
vm_handle_rendezvous(struct vcpu *vcpu)
{
- struct vm *vm = vcpu->vm;
+ struct vm *vm;
struct thread *td;
- int error, vcpuid;
- error = 0;
- vcpuid = vcpu->vcpuid;
td = curthread;
+ vm = vcpu->vm;
+
mtx_lock(&vm->rendezvous_mtx);
while (vm->rendezvous_func != NULL) {
- /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
- CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus);
-
- if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
- !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
- VMM_CTR0(vcpu, "Calling rendezvous func");
- (*vm->rendezvous_func)(vcpu, vm->rendezvous_arg);
- CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
- }
- if (CPU_CMP(&vm->rendezvous_req_cpus,
- &vm->rendezvous_done_cpus) == 0) {
- VMM_CTR0(vcpu, "Rendezvous completed");
- CPU_ZERO(&vm->rendezvous_req_cpus);
- vm->rendezvous_func = NULL;
- wakeup(&vm->rendezvous_func);
+ if (vm_rendezvous(vcpu))
break;
- }
+
VMM_CTR0(vcpu, "Wait for rendezvous completion");
mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
"vmrndv", hz);
if (td_ast_pending(td, TDA_SUSPEND)) {
+ int error;
+
mtx_unlock(&vm->rendezvous_mtx);
error = thread_check_susp(td, true);
if (error != 0)
@@ -1249,7 +1365,7 @@ vm_handle_paging(struct vcpu *vcpu, bool *retu)
("vm_handle_paging: invalid fault_type %d", ftype));
if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
- rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
+ rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm_vmspace(vm)),
vme->u.paging.gpa, ftype);
if (rv == 0) {
VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx",
@@ -1259,7 +1375,7 @@ vm_handle_paging(struct vcpu *vcpu, bool *retu)
}
}
- map = &vm->vmspace->vm_map;
+ map = &vm_vmspace(vm)->vm_map;
rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL);
VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, "
@@ -1560,7 +1676,7 @@ vm_run(struct vcpu *vcpu)
if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
return (EINVAL);
- pmap = vmspace_pmap(vm->vmspace);
+ pmap = vmspace_pmap(vm_vmspace(vm));
vme = &vcpu->exitinfo;
evinfo.rptr = &vm->rendezvous_req_cpus;
evinfo.sptr = &vm->suspend;
@@ -2302,12 +2418,6 @@ vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr)
vcpu_unlock(vcpu);
}
-struct vmspace *
-vm_vmspace(struct vm *vm)
-{
- return (vm->vmspace);
-}
-
struct vm_mem *
vm_mem(struct vm *vm)
{
@@ -2519,7 +2629,7 @@ vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat)
if (vcpu->vcpuid == 0) {
vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE *
- vmspace_resident_count(vcpu->vm->vmspace));
+ vmspace_resident_count(vm_vmspace(vcpu->vm)));
}
}
@@ -2529,7 +2639,7 @@ vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat)
if (vcpu->vcpuid == 0) {
vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE *
- pmap_wired_count(vmspace_pmap(vcpu->vm->vmspace)));
+ pmap_wired_count(vmspace_pmap(vm_vmspace(vcpu->vm))));
}
}
diff --git a/sys/amd64/vmm/vmm_dev_machdep.c b/sys/amd64/vmm/vmm_dev_machdep.c
index d8d2b460404c..dfebc9dcadbf 100644
--- a/sys/amd64/vmm/vmm_dev_machdep.c
+++ b/sys/amd64/vmm/vmm_dev_machdep.c
@@ -48,6 +48,7 @@
#include <x86/apicreg.h>
#include <dev/vmm/vmm_dev.h>
+#include <dev/vmm/vmm_mem.h>
#include <dev/vmm/vmm_stat.h>
#include "vmm_lapic.h"