Add ioctl(VM_REINIT) to reinitialize the virtual machine state maintained

by vmm.ko. This allows the virtual machine to be restarted without having
to destroy it first.

Reviewed by:	grehan

9 files changed, 177 insertions, 85 deletions

diff --git a/lib/libvmmapi/vmmapi.c b/lib/libvmmapi/vmmapi.c
index 89c7825..5ce3d8e 100644
--- a/lib/libvmmapi/vmmapi.c
+++ b/lib/libvmmapi/vmmapi.c
@@ -367,6 +367,13 @@ vm_suspend(struct vmctx *ctx, enum vm_suspend_how how)
 	return (ioctl(ctx->fd, VM_SUSPEND, &vmsuspend));
 }
 
+int
+vm_reinit(struct vmctx *ctx)
+{
+
+	return (ioctl(ctx->fd, VM_REINIT, 0));
+}
+
 static int
 vm_inject_exception_real(struct vmctx *ctx, int vcpu, int vector,
     int error_code, int error_code_valid)
diff --git a/lib/libvmmapi/vmmapi.h b/lib/libvmmapi/vmmapi.h
index 0f2e3ae..4cc4290 100644
--- a/lib/libvmmapi/vmmapi.h
+++ b/lib/libvmmapi/vmmapi.h
@@ -69,6 +69,7 @@ int	vm_get_register(struct vmctx *ctx, int vcpu, int reg, uint64_t *retval);
 int	vm_run(struct vmctx *ctx, int vcpu, uint64_t rip,
 	       struct vm_exit *ret_vmexit);
 int	vm_suspend(struct vmctx *ctx, enum vm_suspend_how how);
+int	vm_reinit(struct vmctx *ctx);
 int	vm_apicid2vcpu(struct vmctx *ctx, int apicid);
 int	vm_inject_exception(struct vmctx *ctx, int vcpu, int vec);
 int	vm_inject_exception2(struct vmctx *ctx, int vcpu, int vec, int errcode);
diff --git a/sys/amd64/include/vmm.h b/sys/amd64/include/vmm.h
index 05df325..00e1d96 100644
--- a/sys/amd64/include/vmm.h
+++ b/sys/amd64/include/vmm.h
@@ -105,6 +105,7 @@ extern struct vmm_ops vmm_ops_amd;
 
 int vm_create(const char *name, struct vm **retvm);
 void vm_destroy(struct vm *vm);
+int vm_reinit(struct vm *vm);
 const char *vm_name(struct vm *vm);
 int vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len);
 int vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa);
diff --git a/sys/amd64/include/vmm_dev.h b/sys/amd64/include/vmm_dev.h
index a6568dc..9b3b00d 100644
--- a/sys/amd64/include/vmm_dev.h
+++ b/sys/amd64/include/vmm_dev.h
@@ -196,6 +196,7 @@ enum {
 	IOCNUM_SET_CAPABILITY = 2,
 	IOCNUM_GET_CAPABILITY = 3,
 	IOCNUM_SUSPEND = 4,
+	IOCNUM_REINIT = 5,
 
 	/* memory apis */
 	IOCNUM_MAP_MEMORY = 10,
@@ -251,6 +252,8 @@ enum {
 	_IOWR('v', IOCNUM_RUN, struct vm_run)
 #define	VM_SUSPEND	\
 	_IOW('v', IOCNUM_SUSPEND, struct vm_suspend)
+#define	VM_REINIT	\
+	_IO('v', IOCNUM_REINIT)
 #define	VM_MAP_MEMORY	\
 	_IOWR('v', IOCNUM_MAP_MEMORY, struct vm_memory_segment)
 #define	VM_GET_MEMORY_SEG \
diff --git a/sys/amd64/vmm/vmm.c b/sys/amd64/vmm/vmm.c
index e84359d..435ba39 100644
--- a/sys/amd64/vmm/vmm.c
+++ b/sys/amd64/vmm/vmm.c
@@ -84,25 +84,31 @@ __FBSDID("$FreeBSD$");
 
 struct vlapic;
 
+/*
+ * Initialization:
+ * (a) allocated when vcpu is created
+ * (i) initialized when vcpu is created and when it is reinitialized
+ * (o) initialized the first time the vcpu is created
+ * (x) initialized before use
+ */
 struct vcpu {
-	int		flags;
-	enum vcpu_state	state;
-	struct mtx	mtx;
-	int		hostcpu;	/* host cpuid this vcpu last ran on */
-	uint64_t	guest_msrs[VMM_MSR_NUM];
-	struct vlapic	*vlapic;
-	int		 vcpuid;
-	struct savefpu	*guestfpu;	/* guest fpu state */
-	uint64_t	guest_xcr0;
-	void		*stats;
-	struct vm_exit	exitinfo;
-	enum x2apic_state x2apic_state;
-	int		nmi_pending;
-	int		extint_pending;
-	struct vm_exception exception;
-	int		exception_pending;
+	struct mtx 	mtx;		/* (o) protects 'state' and 'hostcpu' */
+	enum vcpu_state	state;		/* (o) vcpu state */
+	int		hostcpu;	/* (o) vcpu's host cpu */
+	struct vlapic	*vlapic;	/* (i) APIC device model */
+	enum x2apic_state x2apic_state;	/* (i) APIC mode */
+	int		nmi_pending;	/* (i) NMI pending */
+	int		extint_pending;	/* (i) INTR pending */
+	struct vm_exception exception;	/* (x) exception collateral */
+	int	exception_pending;	/* (i) exception pending */
+	struct savefpu	*guestfpu;	/* (a,i) guest fpu state */
+	uint64_t	guest_xcr0;	/* (i) guest %xcr0 register */
+	void		*stats;		/* (a,i) statistics */
+	uint64_t guest_msrs[VMM_MSR_NUM]; /* (i) emulated MSRs */
+	struct vm_exit	exitinfo;	/* (x) exit reason and collateral */
 };
 
+#define	vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
@@ -116,36 +122,33 @@ struct mem_seg {
 };
 #define	VM_MAX_MEMORY_SEGMENTS	2
 
+/*
+ * Initialization:
+ * (o) initialized the first time the VM is created
+ * (i) initialized when VM is created and when it is reinitialized
+ * (x) initialized before use
+ */
 struct vm {
-	void		*cookie;	/* processor-specific data */
-	void		*iommu;		/* iommu-specific data */
-	struct vhpet	*vhpet;		/* virtual HPET */
-	struct vioapic	*vioapic;	/* virtual ioapic */
-	struct vatpic	*vatpic;	/* virtual atpic */
-	struct vatpit	*vatpit;	/* virtual atpit */
-	struct vmspace	*vmspace;	/* guest's address space */
-	struct vcpu	vcpu[VM_MAXCPU];
-	int		num_mem_segs;
-	struct mem_seg	mem_segs[VM_MAX_MEMORY_SEGMENTS];
-	char		name[VM_MAX_NAMELEN];
-
-	/*
-	 * Set of active vcpus.
-	 * An active vcpu is one that has been started implicitly (BSP) or
-	 * explicitly (AP) by sending it a startup ipi.
-	 */
-	volatile cpuset_t active_cpus;
-
-	struct mtx	rendezvous_mtx;
-	cpuset_t	rendezvous_req_cpus;
-	cpuset_t	rendezvous_done_cpus;
-	void		*rendezvous_arg;
+	void		*cookie;		/* (i) cpu-specific data */
+	void		*iommu;			/* (x) iommu-specific data */
+	struct vhpet	*vhpet;			/* (i) virtual HPET */
+	struct vioapic	*vioapic;		/* (i) virtual ioapic */
+	struct vatpic	*vatpic;		/* (i) virtual atpic */
+	struct vatpit	*vatpit;		/* (i) virtual atpit */
+	volatile cpuset_t active_cpus;		/* (i) active vcpus */
+	int		suspend;		/* (i) stop VM execution */
+	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
+	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
+	cpuset_t	rendezvous_req_cpus;	/* (x) rendezvous requested */
+	cpuset_t	rendezvous_done_cpus;	/* (x) rendezvous finished */
+	void		*rendezvous_arg;	/* (x) rendezvous func/arg */
 	vm_rendezvous_func_t rendezvous_func;
-
-	int		suspend;
-	volatile cpuset_t suspended_cpus;
-
-	volatile cpuset_t halted_cpus;
+	struct mtx	rendezvous_mtx;		/* (o) rendezvous lock */
+	int		num_mem_segs;		/* (o) guest memory segments */
+	struct mem_seg	mem_segs[VM_MAX_MEMORY_SEGMENTS];
+	struct vmspace	*vmspace;		/* (o) guest's address space */
+	char		name[VM_MAX_NAMELEN];	/* (o) virtual machine name */
+	struct vcpu	vcpu[VM_MAXCPU];	/* (i) guest vcpus */
 };
 
 static int vmm_initialized;
@@ -206,31 +209,46 @@ SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
     "IPI vector used for vcpu notifications");
 
 static void
-vcpu_cleanup(struct vm *vm, int i)
+vcpu_cleanup(struct vm *vm, int i, bool destroy)
 {
 	struct vcpu *vcpu = &vm->vcpu[i];
 
 	VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
-	vmm_stat_free(vcpu->stats);	
-	fpu_save_area_free(vcpu->guestfpu);
+	if (destroy) {
+		vmm_stat_free(vcpu->stats);	
+		fpu_save_area_free(vcpu->guestfpu);
+	}
 }
 
 static void
-vcpu_init(struct vm *vm, uint32_t vcpu_id)
+vcpu_init(struct vm *vm, int vcpu_id, bool create)
 {
 	struct vcpu *vcpu;
-	
+
+	KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU,
+	    ("vcpu_init: invalid vcpu %d", vcpu_id));
+	  
 	vcpu = &vm->vcpu[vcpu_id];
 
-	vcpu_lock_init(vcpu);
-	vcpu->hostcpu = NOCPU;
-	vcpu->vcpuid = vcpu_id;
+	if (create) {
+		KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
+		    "initialized", vcpu_id));
+		vcpu_lock_init(vcpu);
+		vcpu->state = VCPU_IDLE;
+		vcpu->hostcpu = NOCPU;
+		vcpu->guestfpu = fpu_save_area_alloc();
+		vcpu->stats = vmm_stat_alloc();
+	}
+
 	vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
 	vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
+	vcpu->nmi_pending = 0;
+	vcpu->extint_pending = 0;
+	vcpu->exception_pending = 0;
 	vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
-	vcpu->guestfpu = fpu_save_area_alloc();
 	fpu_save_area_reset(vcpu->guestfpu);
-	vcpu->stats = vmm_stat_alloc();
+	vmm_stat_init(vcpu->stats);
+	guest_msrs_init(vm, vcpu_id);
 }
 
 struct vm_exit *
@@ -335,10 +353,30 @@ static moduledata_t vmm_kmod = {
 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
 MODULE_VERSION(vmm, 1);
 
+static void
+vm_init(struct vm *vm, bool create)
+{
+	int i;
+
+	vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
+	vm->iommu = NULL;
+	vm->vioapic = vioapic_init(vm);
+	vm->vhpet = vhpet_init(vm);
+	vm->vatpic = vatpic_init(vm);
+	vm->vatpit = vatpit_init(vm);
+
+	CPU_ZERO(&vm->active_cpus);
+
+	vm->suspend = 0;
+	CPU_ZERO(&vm->suspended_cpus);
+
+	for (i = 0; i < VM_MAXCPU; i++)
+		vcpu_init(vm, i, create);
+}
+
 int
 vm_create(const char *name, struct vm **retvm)
 {
-	int i;
 	struct vm *vm;
 	struct vmspace *vmspace;
 
@@ -358,18 +396,11 @@ vm_create(const char *name, struct vm **retvm)
 
 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
 	strcpy(vm->name, name);
+	vm->num_mem_segs = 0;
 	vm->vmspace = vmspace;
 	mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
-	vm->cookie = VMINIT(vm, vmspace_pmap(vmspace));
-	vm->vioapic = vioapic_init(vm);
-	vm->vhpet = vhpet_init(vm);
-	vm->vatpic = vatpic_init(vm);
-	vm->vatpit = vatpit_init(vm);
 
-	for (i = 0; i < VM_MAXCPU; i++) {
-		vcpu_init(vm, i);
-		guest_msrs_init(vm, i);
-	}
+	vm_init(vm, true);
 
 	*retvm = vm;
 	return (0);
@@ -385,8 +416,8 @@ vm_free_mem_seg(struct vm *vm, struct mem_seg *seg)
 	bzero(seg, sizeof(*seg));
 }
 
-void
-vm_destroy(struct vm *vm)
+static void
+vm_cleanup(struct vm *vm, bool destroy)
 {
 	int i;
 
@@ -400,21 +431,48 @@ vm_destroy(struct vm *vm)
 	vatpic_cleanup(vm->vatpic);
 	vioapic_cleanup(vm->vioapic);
 
-	for (i = 0; i < vm->num_mem_segs; i++)
-		vm_free_mem_seg(vm, &vm->mem_segs[i]);
+	for (i = 0; i < VM_MAXCPU; i++)
+		vcpu_cleanup(vm, i, destroy);
 
-	vm->num_mem_segs = 0;
+	VMCLEANUP(vm->cookie);
 
-	for (i = 0; i < VM_MAXCPU; i++)
-		vcpu_cleanup(vm, i);
+	if (destroy) {
+		for (i = 0; i < vm->num_mem_segs; i++)
+			vm_free_mem_seg(vm, &vm->mem_segs[i]);
 
-	VMSPACE_FREE(vm->vmspace);
+		vm->num_mem_segs = 0;
 
-	VMCLEANUP(vm->cookie);
+		VMSPACE_FREE(vm->vmspace);
+		vm->vmspace = NULL;
+	}
+}
 
+void
+vm_destroy(struct vm *vm)
+{
+	vm_cleanup(vm, true);
 	free(vm, M_VM);
 }
 
+int
+vm_reinit(struct vm *vm)
+{
+	int error;
+
+	/*
+	 * A virtual machine can be reset only if all vcpus are suspended.
+	 */
+	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
+		vm_cleanup(vm, false);
+		vm_init(vm, false);
+		error = 0;
+	} else {
+		error = EBUSY;
+	}
+
+	return (error);
+}
+
 const char *
 vm_name(struct vm *vm)
 {
diff --git a/sys/amd64/vmm/vmm_dev.c b/sys/amd64/vmm/vmm_dev.c
index 824389f..f3e31a3 100644
--- a/sys/amd64/vmm/vmm_dev.c
+++ b/sys/amd64/vmm/vmm_dev.c
@@ -220,6 +220,7 @@ vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 	case VM_BIND_PPTDEV:
 	case VM_UNBIND_PPTDEV:
 	case VM_MAP_MEMORY:
+	case VM_REINIT:
 		/*
 		 * ioctls that operate on the entire virtual machine must
 		 * prevent all vcpus from running.
@@ -253,6 +254,9 @@ vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 		vmsuspend = (struct vm_suspend *)data;
 		error = vm_suspend(sc->vm, vmsuspend->how);
 		break;
+	case VM_REINIT:
+		error = vm_reinit(sc->vm);
+		break;
 	case VM_STAT_DESC: {
 		statdesc = (struct vm_stat_desc *)data;
 		error = vmm_stat_desc_copy(statdesc->index,
diff --git a/sys/amd64/vmm/vmm_stat.c b/sys/amd64/vmm/vmm_stat.c
index e3d6999..ef9f411 100644
--- a/sys/amd64/vmm/vmm_stat.c
+++ b/sys/amd64/vmm/vmm_stat.c
@@ -52,8 +52,10 @@ static struct vmm_stat_type *vsttab[MAX_VMM_STAT_ELEMS];
 
 static MALLOC_DEFINE(M_VMM_STAT, "vmm stat", "vmm stat");
 
+#define	vst_size	((size_t)vst_num_elems * sizeof(uint64_t))
+
 void
-vmm_stat_init(void *arg)
+vmm_stat_register(void *arg)
 {
 	struct vmm_stat_type *vst = arg;
 
@@ -97,11 +99,15 @@ vmm_stat_copy(struct vm *vm, int vcpu, int *num_stats, uint64_t *buf)
 void *
 vmm_stat_alloc(void)
 {
-	u_long size;
-	
-	size = vst_num_elems * sizeof(uint64_t);
 
-	return (malloc(size, M_VMM_STAT, M_ZERO | M_WAITOK));
+	return (malloc(vst_size, M_VMM_STAT, M_WAITOK));
+}
+
+void
+vmm_stat_init(void *vp)
+{
+
+	bzero(vp, vst_size);
 }
 
 void
diff --git a/sys/amd64/vmm/vmm_stat.h b/sys/amd64/vmm/vmm_stat.h
index 9110c8f..6e98965 100644
--- a/sys/amd64/vmm/vmm_stat.h
+++ b/sys/amd64/vmm/vmm_stat.h
@@ -49,13 +49,13 @@ struct vmm_stat_type {
 	enum vmm_stat_scope scope;
 };
 
-void	vmm_stat_init(void *arg);
+void	vmm_stat_register(void *arg);
 
 #define	VMM_STAT_DEFINE(type, nelems, desc, scope)			\
 	struct vmm_stat_type type[1] = {				\
 		{ -1, nelems, desc, scope }				\
 	};								\
-	SYSINIT(type##_stat, SI_SUB_KLD, SI_ORDER_ANY, vmm_stat_init, type)
+	SYSINIT(type##_stat, SI_SUB_KLD, SI_ORDER_ANY, vmm_stat_register, type)
 
 #define	VMM_STAT_DECLARE(type)						\
 	extern struct vmm_stat_type type[1]
@@ -71,6 +71,7 @@ void	vmm_stat_init(void *arg);
 	VMM_STAT_DEFINE(type, nelems, desc, VMM_STAT_SCOPE_ANY)
 
 void	*vmm_stat_alloc(void);
+void	vmm_stat_init(void *vp);
 void 	vmm_stat_free(void *vp);
 
 /*
diff --git a/usr.sbin/bhyveload/bhyveload.c b/usr.sbin/bhyveload/bhyveload.c
index c1a5432..4442496 100644
--- a/usr.sbin/bhyveload/bhyveload.c
+++ b/usr.sbin/bhyveload/bhyveload.c
@@ -642,7 +642,7 @@ main(int argc, char** argv)
 	void *h;
 	void (*func)(struct loader_callbacks *, void *, int, int);
 	uint64_t mem_size;
-	int opt, error;
+	int opt, error, need_reinit;
 
 	progname = basename(argv[0]);
 
@@ -691,11 +691,14 @@ main(int argc, char** argv)
 
 	vmname = argv[0];
 
+	need_reinit = 0;
 	error = vm_create(vmname);
-	if (error != 0 && errno != EEXIST) {
-		perror("vm_create");
-		exit(1);
-
+	if (error) {
+		if (errno != EEXIST) {
+			perror("vm_create");
+			exit(1);
+		}
+		need_reinit = 1;
 	}
 
 	ctx = vm_open(vmname);
@@ -704,6 +707,14 @@ main(int argc, char** argv)
 		exit(1);
 	}
 
+	if (need_reinit) {
+		error = vm_reinit(ctx);
+		if (error) {
+			perror("vm_reinit");
+			exit(1);
+		}
+	}
+
 	error = vm_setup_memory(ctx, mem_size, VM_MMAP_ALL);
 	if (error) {
 		perror("vm_setup_memory");