1 files changed, 36 insertions, 125 deletions

diff --git a/sys/amd64/vmm/vmm.c b/sys/amd64/vmm/vmm.c
index f2bea0d82b5c..2890e990633d 100644
--- a/sys/amd64/vmm/vmm.c
+++ b/sys/amd64/vmm/vmm.c
@@ -31,7 +31,6 @@
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/kernel.h>
-#include <sys/module.h>
 #include <sys/sysctl.h>
 #include <sys/malloc.h>
 #include <sys/pcpu.h>
@@ -189,8 +188,6 @@ struct vm {
 #define	VMM_CTR4(vcpu, format, p1, p2, p3, p4)				\
 	VCPU_CTR4((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3, p4)
 
-static int vmm_initialized;
-
 static void	vmmops_panic(void);
 
 static void
@@ -270,11 +267,7 @@ static int trap_wbinvd;
 SYSCTL_INT(_hw_vmm, OID_AUTO, trap_wbinvd, CTLFLAG_RDTUN, &trap_wbinvd, 0,
     "WBINVD triggers a VM-exit");
 
-u_int vm_maxcpu;
-SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
-    &vm_maxcpu, 0, "Maximum number of vCPUs");
-
-static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr);
+static void vcpu_notify_event_locked(struct vcpu *vcpu);
 
 /* global statistics */
 VMM_STAT(VCPU_MIGRATIONS, "vcpu migration across host cpus");
@@ -299,14 +292,6 @@ VMM_STAT(VMEXIT_USERSPACE, "number of vm exits handled in userspace");
 VMM_STAT(VMEXIT_RENDEZVOUS, "number of times rendezvous pending at exit");
 VMM_STAT(VMEXIT_EXCEPTION, "number of vm exits due to exceptions");
 
-/*
- * Upper limit on vm_maxcpu.  Limited by use of uint16_t types for CPU
- * counts as well as range of vpid values for VT-x and by the capacity
- * of cpuset_t masks.  The call to new_unrhdr() in vpid_init() in
- * vmx.c requires 'vm_maxcpu + 1 <= 0xffff', hence the '- 1' below.
- */
-#define	VM_MAXCPU	MIN(0xffff - 1, CPU_SETSIZE)
-
 #ifdef KTR
 static const char *
 vcpu_state2str(enum vcpu_state state)
@@ -402,22 +387,12 @@ vm_exitinfo_cpuset(struct vcpu *vcpu)
 	return (&vcpu->exitinfo_cpuset);
 }
 
-static int
-vmm_init(void)
+int
+vmm_modinit(void)
 {
 	if (!vmm_is_hw_supported())
 		return (ENXIO);
 
-	vm_maxcpu = mp_ncpus;
-	TUNABLE_INT_FETCH("hw.vmm.maxcpu", &vm_maxcpu);
-
-	if (vm_maxcpu > VM_MAXCPU) {
-		printf("vmm: vm_maxcpu clamped to %u\n", VM_MAXCPU);
-		vm_maxcpu = VM_MAXCPU;
-	}
-	if (vm_maxcpu == 0)
-		vm_maxcpu = 1;
-
 	vmm_host_state_init();
 
 	vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) :
@@ -431,70 +406,17 @@ vmm_init(void)
 	return (vmmops_modinit(vmm_ipinum));
 }
 
-static int
-vmm_handler(module_t mod, int what, void *arg)
+int
+vmm_modcleanup(void)
 {
-	int error;
-
-	switch (what) {
-	case MOD_LOAD:
-		if (vmm_is_hw_supported()) {
-			error = vmmdev_init();
-			if (error != 0)
-				break;
-			error = vmm_init();
-			if (error == 0)
-				vmm_initialized = 1;
-			else
-				(void)vmmdev_cleanup();
-		} else {
-			error = ENXIO;
-		}
-		break;
-	case MOD_UNLOAD:
-		if (vmm_is_hw_supported()) {
-			error = vmmdev_cleanup();
-			if (error == 0) {
-				vmm_suspend_p = NULL;
-				vmm_resume_p = NULL;
-				iommu_cleanup();
-				if (vmm_ipinum != IPI_AST)
-					lapic_ipi_free(vmm_ipinum);
-				error = vmmops_modcleanup();
-				/*
-				 * Something bad happened - prevent new
-				 * VMs from being created
-				 */
-				if (error)
-					vmm_initialized = 0;
-			}
-		} else {
-			error = 0;
-		}
-		break;
-	default:
-		error = 0;
-		break;
-	}
-	return (error);
+	vmm_suspend_p = NULL;
+	vmm_resume_p = NULL;
+	iommu_cleanup();
+	if (vmm_ipinum != IPI_AST)
+		lapic_ipi_free(vmm_ipinum);
+	return (vmmops_modcleanup());
 }
 
-static moduledata_t vmm_kmod = {
-	"vmm",
-	vmm_handler,
-	NULL
-};
-
-/*
- * vmm initialization has the following dependencies:
- *
- * - VT-x initialization requires smp_rendezvous() and therefore must happen
- *   after SMP is fully functional (after SI_SUB_SMP).
- * - vmm device initialization requires an initialized devfs.
- */
-DECLARE_MODULE(vmm, vmm_kmod, MAX(SI_SUB_SMP, SI_SUB_DEVFS) + 1, SI_ORDER_ANY);
-MODULE_VERSION(vmm, 1);
-
 static void
 vm_init(struct vm *vm, bool create)
 {
@@ -573,29 +495,12 @@ vm_unlock_vcpus(struct vm *vm)
 	sx_unlock(&vm->vcpus_init_lock);
 }
 
-/*
- * The default CPU topology is a single thread per package.
- */
-u_int cores_per_package = 1;
-u_int threads_per_core = 1;
-
 int
 vm_create(const char *name, struct vm **retvm)
 {
 	struct vm *vm;
 	int error;
 
-	/*
-	 * If vmm.ko could not be successfully initialized then don't attempt
-	 * to create the virtual machine.
-	 */
-	if (!vmm_initialized)
-		return (ENXIO);
-
-	if (name == NULL || strnlen(name, VM_MAX_NAMELEN + 1) ==
-	    VM_MAX_NAMELEN + 1)
-		return (EINVAL);
-
 	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) {
@@ -609,8 +514,8 @@ vm_create(const char *name, struct vm **retvm)
 	    M_ZERO);
 
 	vm->sockets = 1;
-	vm->cores = cores_per_package;	/* XXX backwards compatibility */
-	vm->threads = threads_per_core;	/* XXX backwards compatibility */
+	vm->cores = 1;		/* XXX backwards compatibility */
+	vm->threads = 1;	/* XXX backwards compatibility */
 	vm->maxcpus = vm_maxcpu;
 
 	vm_init(vm, true);
@@ -1028,7 +933,7 @@ vcpu_wait_idle(struct vcpu *vcpu)
 	KASSERT(vcpu->state != VCPU_IDLE, ("vcpu already idle"));
 
 	vcpu->reqidle = 1;
-	vcpu_notify_event_locked(vcpu, false);
+	vcpu_notify_event_locked(vcpu);
 	VMM_CTR1(vcpu, "vcpu state change from %s to "
 	    "idle requested", vcpu_state2str(vcpu->state));
 	msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
@@ -1509,7 +1414,7 @@ vm_handle_suspend(struct vcpu *vcpu, bool *retu)
 	 */
 	for (i = 0; i < vm->maxcpus; i++) {
 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
-			vcpu_notify_event(vm_vcpu(vm, i), false);
+			vcpu_notify_event(vm_vcpu(vm, i));
 		}
 	}
 
@@ -1583,7 +1488,7 @@ vm_suspend(struct vm *vm, enum vm_suspend_how how)
 	 */
 	for (i = 0; i < vm->maxcpus; i++) {
 		if (CPU_ISSET(i, &vm->active_cpus))
-			vcpu_notify_event(vm_vcpu(vm, i), false);
+			vcpu_notify_event(vm_vcpu(vm, i));
 	}
 
 	return (0);
@@ -2063,7 +1968,7 @@ vm_inject_nmi(struct vcpu *vcpu)
 {
 
 	vcpu->nmi_pending = 1;
-	vcpu_notify_event(vcpu, false);
+	vcpu_notify_event(vcpu);
 	return (0);
 }
 
@@ -2090,7 +1995,7 @@ vm_inject_extint(struct vcpu *vcpu)
 {
 
 	vcpu->extint_pending = 1;
-	vcpu_notify_event(vcpu, false);
+	vcpu_notify_event(vcpu);
 	return (0);
 }
 
@@ -2261,14 +2166,14 @@ vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu)
 		vm->debug_cpus = vm->active_cpus;
 		for (int i = 0; i < vm->maxcpus; i++) {
 			if (CPU_ISSET(i, &vm->active_cpus))
-				vcpu_notify_event(vm_vcpu(vm, i), false);
+				vcpu_notify_event(vm_vcpu(vm, i));
 		}
 	} else {
 		if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus))
 			return (EINVAL);
 
 		CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus);
-		vcpu_notify_event(vcpu, false);
+		vcpu_notify_event(vcpu);
 	}
 	return (0);
 }
@@ -2376,7 +2281,7 @@ vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state)
  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
  */
 static void
-vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
+vcpu_notify_event_locked(struct vcpu *vcpu)
 {
 	int hostcpu;
 
@@ -2384,12 +2289,7 @@ vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
 	if (vcpu->state == VCPU_RUNNING) {
 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
 		if (hostcpu != curcpu) {
-			if (lapic_intr) {
-				vlapic_post_intr(vcpu->vlapic, hostcpu,
-				    vmm_ipinum);
-			} else {
-				ipi_cpu(hostcpu, vmm_ipinum);
-			}
+			ipi_cpu(hostcpu, vmm_ipinum);
 		} else {
 			/*
 			 * If the 'vcpu' is running on 'curcpu' then it must
@@ -2407,10 +2307,21 @@ vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
 }
 
 void
-vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr)
+vcpu_notify_event(struct vcpu *vcpu)
 {
 	vcpu_lock(vcpu);
-	vcpu_notify_event_locked(vcpu, lapic_intr);
+	vcpu_notify_event_locked(vcpu);
+	vcpu_unlock(vcpu);
+}
+
+void
+vcpu_notify_lapic(struct vcpu *vcpu)
+{
+	vcpu_lock(vcpu);
+	if (vcpu->state == VCPU_RUNNING && vcpu->hostcpu != curcpu)
+		vlapic_post_intr(vcpu->vlapic, vcpu->hostcpu, vmm_ipinum);
+	else
+		vcpu_notify_event_locked(vcpu);
 	vcpu_unlock(vcpu);
 }
 
@@ -2472,7 +2383,7 @@ restart:
 	 */
 	for (i = 0; i < vm->maxcpus; i++) {
 		if (CPU_ISSET(i, &dest))
-			vcpu_notify_event(vm_vcpu(vm, i), false);
+			vcpu_notify_event(vm_vcpu(vm, i));
 	}
 
 	return (vm_handle_rendezvous(vcpu));