Activate vcpus from bhyve(8) using the ioctl VM_ACTIVATE_CPU instead of doing

it implicitly in vmm.ko.

Add ioctl VM_GET_CPUS to get the current set of 'active' and 'suspended' cpus
and display them via /usr/sbin/bhyvectl using the "--get-active-cpus" and
"--get-suspended-cpus" options.

This is in preparation for being able to reset virtual machine state without
having to destroy and recreate it.

10 files changed, 171 insertions, 18 deletions

diff --git a/lib/libvmmapi/vmmapi.c b/lib/libvmmapi/vmmapi.c
index ba2904c..89c7825 100644
--- a/lib/libvmmapi/vmmapi.c
+++ b/lib/libvmmapi/vmmapi.c
@@ -29,11 +29,12 @@
 #include <sys/cdefs.h>
 __FBSDID("$FreeBSD$");
 
-#include <sys/types.h>
+#include <sys/param.h>
 #include <sys/sysctl.h>
 #include <sys/ioctl.h>
 #include <sys/mman.h>
 #include <sys/_iovec.h>
+#include <sys/cpuset.h>
 
 #include <machine/specialreg.h>
 #include <machine/param.h>
@@ -1043,3 +1044,44 @@ vm_copyout(struct vmctx *ctx, int vcpu, const void *vp, struct iovec *iov,
 		len -= n;
 	}
 }
+
+static int
+vm_get_cpus(struct vmctx *ctx, int which, cpuset_t *cpus)
+{
+	struct vm_cpuset vm_cpuset;
+	int error;
+
+	bzero(&vm_cpuset, sizeof(struct vm_cpuset));
+	vm_cpuset.which = which;
+	vm_cpuset.cpusetsize = sizeof(cpuset_t);
+	vm_cpuset.cpus = cpus;
+
+	error = ioctl(ctx->fd, VM_GET_CPUS, &vm_cpuset);
+	return (error);
+}
+
+int
+vm_active_cpus(struct vmctx *ctx, cpuset_t *cpus)
+{
+
+	return (vm_get_cpus(ctx, VM_ACTIVE_CPUS, cpus));
+}
+
+int
+vm_suspended_cpus(struct vmctx *ctx, cpuset_t *cpus)
+{
+
+	return (vm_get_cpus(ctx, VM_SUSPENDED_CPUS, cpus));
+}
+
+int
+vm_activate_cpu(struct vmctx *ctx, int vcpu)
+{
+	struct vm_activate_cpu ac;
+	int error;
+
+	bzero(&ac, sizeof(struct vm_activate_cpu));
+	ac.vcpuid = vcpu;
+	error = ioctl(ctx->fd, VM_ACTIVATE_CPU, &ac);
+	return (error);
+}
diff --git a/lib/libvmmapi/vmmapi.h b/lib/libvmmapi/vmmapi.h
index bab41da..0f2e3ae 100644
--- a/lib/libvmmapi/vmmapi.h
+++ b/lib/libvmmapi/vmmapi.h
@@ -29,6 +29,9 @@
 #ifndef _VMMAPI_H_
 #define	_VMMAPI_H_
 
+#include <sys/param.h>
+#include <sys/cpuset.h>
+
 struct iovec;
 struct vmctx;
 enum x2apic_state;
@@ -125,6 +128,10 @@ void	vm_copyout(struct vmctx *ctx, int vcpu, const void *host_src,
 /* Reset vcpu register state */
 int	vcpu_reset(struct vmctx *ctx, int vcpu);
 
+int	vm_active_cpus(struct vmctx *ctx, cpuset_t *cpus);
+int	vm_suspended_cpus(struct vmctx *ctx, cpuset_t *cpus);
+int	vm_activate_cpu(struct vmctx *ctx, int vcpu);
+
 /*
  * FreeBSD specific APIs
  */
diff --git a/sys/amd64/include/vmm.h b/sys/amd64/include/vmm.h
index f1902d2..05df325 100644
--- a/sys/amd64/include/vmm.h
+++ b/sys/amd64/include/vmm.h
@@ -140,8 +140,9 @@ int vm_set_capability(struct vm *vm, int vcpu, int type, int val);
 int vm_get_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state *state);
 int vm_set_x2apic_state(struct vm *vm, int vcpu, enum x2apic_state state);
 int vm_apicid2vcpuid(struct vm *vm, int apicid);
-void vm_activate_cpu(struct vm *vm, int vcpu);
+int vm_activate_cpu(struct vm *vm, int vcpu);
 cpuset_t vm_active_cpus(struct vm *vm);
+cpuset_t vm_suspended_cpus(struct vm *vm);
 struct vm_exit *vm_exitinfo(struct vm *vm, int vcpuid);
 void vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip);
 
diff --git a/sys/amd64/include/vmm_dev.h b/sys/amd64/include/vmm_dev.h
index f094d51..a6568dc 100644
--- a/sys/amd64/include/vmm_dev.h
+++ b/sys/amd64/include/vmm_dev.h
@@ -177,6 +177,18 @@ struct vm_gla2gpa {
 	uint64_t	gpa;
 };
 
+struct vm_activate_cpu {
+	int		vcpuid;
+};
+
+struct vm_cpuset {
+	int		which;
+	int		cpusetsize;
+	cpuset_t	*cpus;
+};
+#define	VM_ACTIVE_CPUS		0
+#define	VM_SUSPENDED_CPUS	1
+
 enum {
 	/* general routines */
 	IOCNUM_ABIVERS = 0,
@@ -229,6 +241,10 @@ enum {
 	IOCNUM_ISA_DEASSERT_IRQ = 81,
 	IOCNUM_ISA_PULSE_IRQ = 82,
 	IOCNUM_ISA_SET_IRQ_TRIGGER = 83,
+
+	/* vm_cpuset */
+	IOCNUM_ACTIVATE_CPU = 90,
+	IOCNUM_GET_CPUSET = 91,
 };
 
 #define	VM_RUN		\
@@ -301,4 +317,8 @@ enum {
 	_IOWR('v', IOCNUM_GET_GPA_PMAP, struct vm_gpa_pte)
 #define	VM_GLA2GPA	\
 	_IOWR('v', IOCNUM_GLA2GPA, struct vm_gla2gpa)
+#define	VM_ACTIVATE_CPU	\
+	_IOW('v', IOCNUM_ACTIVATE_CPU, struct vm_activate_cpu)
+#define	VM_GET_CPUS	\
+	_IOW('v', IOCNUM_GET_CPUSET, struct vm_cpuset)
 #endif
diff --git a/sys/amd64/vmm/io/vlapic.c b/sys/amd64/vmm/io/vlapic.c
index d93641c..4034d34 100644
--- a/sys/amd64/vmm/io/vlapic.c
+++ b/sys/amd64/vmm/io/vlapic.c
@@ -1004,11 +1004,7 @@ vlapic_icrlo_write_handler(struct vlapic *vlapic, bool *retu)
 			if (vlapic2->boot_state != BS_SIPI)
 				return (0);
 
-			/*
-			 * XXX this assumes that the startup IPI always succeeds
-			 */
 			vlapic2->boot_state = BS_RUNNING;
-			vm_activate_cpu(vlapic2->vm, dest);
 
 			*retu = true;
 			vmexit = vm_exitinfo(vlapic->vm, vlapic->vcpuid);
diff --git a/sys/amd64/vmm/vmm.c b/sys/amd64/vmm/vmm.c
index 8ebdfd7..e84359d 100644
--- a/sys/amd64/vmm/vmm.c
+++ b/sys/amd64/vmm/vmm.c
@@ -342,8 +342,6 @@ vm_create(const char *name, struct vm **retvm)
 	struct vm *vm;
 	struct vmspace *vmspace;
 
-	const int BSP = 0;
-
 	/*
 	 * If vmm.ko could not be successfully initialized then don't attempt
 	 * to create the virtual machine.
@@ -373,8 +371,6 @@ vm_create(const char *name, struct vm **retvm)
 		guest_msrs_init(vm, i);
 	}
 
-	vm_activate_cpu(vm, BSP);
-
 	*retvm = vm;
 	return (0);
 }
@@ -1294,6 +1290,12 @@ vm_run(struct vm *vm, struct vm_run *vmrun)
 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
 		return (EINVAL);
 
+	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
+		return (EINVAL);
+
+	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
+		return (EINVAL);
+
 	rptr = &vm->rendezvous_func;
 	sptr = &vm->suspend;
 	pmap = vmspace_pmap(vm->vmspace);
@@ -1708,17 +1710,19 @@ vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
 	return (state);
 }
 
-void
+int
 vm_activate_cpu(struct vm *vm, int vcpuid)
 {
 
-	KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU,
-	    ("vm_activate_cpu: invalid vcpuid %d", vcpuid));
-	KASSERT(!CPU_ISSET(vcpuid, &vm->active_cpus),
-	    ("vm_activate_cpu: vcpuid %d is already active", vcpuid));
+	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
+		return (EINVAL);
+
+	if (CPU_ISSET(vcpuid, &vm->active_cpus))
+		return (EBUSY);
 
 	VCPU_CTR0(vm, vcpuid, "activated");
 	CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
+	return (0);
 }
 
 cpuset_t
@@ -1728,6 +1732,13 @@ vm_active_cpus(struct vm *vm)
 	return (vm->active_cpus);
 }
 
+cpuset_t
+vm_suspended_cpus(struct vm *vm)
+{
+
+	return (vm->suspended_cpus);
+}
+
 void *
 vcpu_stats(struct vm *vm, int vcpuid)
 {
diff --git a/sys/amd64/vmm/vmm_dev.c b/sys/amd64/vmm/vmm_dev.c
index 0561785..824389f 100644
--- a/sys/amd64/vmm/vmm_dev.c
+++ b/sys/amd64/vmm/vmm_dev.c
@@ -146,7 +146,8 @@ static int
 vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 	     struct thread *td)
 {
-	int error, vcpu, state_changed;
+	int error, vcpu, state_changed, size;
+	cpuset_t *cpuset;
 	struct vmmdev_softc *sc;
 	struct vm_memory_segment *seg;
 	struct vm_register *vmreg;
@@ -170,6 +171,8 @@ vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 	struct vm_gpa_pte *gpapte;
 	struct vm_suspend *vmsuspend;
 	struct vm_gla2gpa *gg;
+	struct vm_activate_cpu *vac;
+	struct vm_cpuset *vm_cpuset;
 
 	sc = vmmdev_lookup2(cdev);
 	if (sc == NULL)
@@ -195,6 +198,7 @@ vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 	case VM_PPTDEV_MSIX:
 	case VM_SET_X2APIC_STATE:
 	case VM_GLA2GPA:
+	case VM_ACTIVATE_CPU:
 		/*
 		 * XXX fragile, handle with care
 		 * Assumes that the first field of the ioctl data is the vcpu.
@@ -439,6 +443,29 @@ vmmdev_ioctl(struct cdev *cdev, u_long cmd, caddr_t data, int fflag,
 		}
 		break;
 	}
+	case VM_ACTIVATE_CPU:
+		vac = (struct vm_activate_cpu *)data;
+		error = vm_activate_cpu(sc->vm, vac->vcpuid);
+		break;
+	case VM_GET_CPUS:
+		error = 0;
+		vm_cpuset = (struct vm_cpuset *)data;
+		size = vm_cpuset->cpusetsize;
+		if (size < sizeof(cpuset_t) || size > CPU_MAXSIZE / NBBY) {
+			error = ERANGE;
+			break;
+		}
+		cpuset = malloc(size, M_TEMP, M_WAITOK | M_ZERO);
+		if (vm_cpuset->which == VM_ACTIVE_CPUS)
+			*cpuset = vm_active_cpus(sc->vm);
+		else if (vm_cpuset->which == VM_SUSPENDED_CPUS)
+			*cpuset = vm_suspended_cpus(sc->vm);
+		else
+			error = EINVAL;
+		if (error == 0)
+			error = copyout(cpuset, vm_cpuset->cpus, size);
+		free(cpuset, M_TEMP);
+		break;
 	default:
 		error = ENOTTY;
 		break;
diff --git a/usr.sbin/bhyve/bhyverun.c b/usr.sbin/bhyve/bhyverun.c
index f9a67cb..1e5d3b3 100644
--- a/usr.sbin/bhyve/bhyverun.c
+++ b/usr.sbin/bhyve/bhyverun.c
@@ -242,6 +242,15 @@ fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip)
 
 	assert(fromcpu == BSP);
 
+	/*
+	 * The 'newcpu' must be activated in the context of 'fromcpu'. If
+	 * vm_activate_cpu() is delayed until newcpu's pthread starts running
+	 * then vmm.ko is out-of-sync with bhyve and this can create a race
+	 * with vm_suspend().
+	 */
+	error = vm_activate_cpu(ctx, newcpu);
+	assert(error == 0);
+
 	CPU_SET_ATOMIC(newcpu, &cpumask);
 
 	/*
@@ -532,6 +541,7 @@ vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip)
 	int error, rc, prevcpu;
 	enum vm_exitcode exitcode;
 	enum vm_suspend_how how;
+	cpuset_t active_cpus;
 
 	if (vcpumap[vcpu] != NULL) {
 		error = pthread_setaffinity_np(pthread_self(),
@@ -539,6 +549,9 @@ vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip)
 		assert(error == 0);
 	}
 
+	error = vm_active_cpus(ctx, &active_cpus);
+	assert(CPU_ISSET(vcpu, &active_cpus));
+
 	while (1) {
 		error = vm_run(ctx, vcpu, rip, &vmexit[vcpu]);
 		if (error != 0)
diff --git a/usr.sbin/bhyve/pci_lpc.c b/usr.sbin/bhyve/pci_lpc.c
index 6b61b7a..e98b141 100644
--- a/usr.sbin/bhyve/pci_lpc.c
+++ b/usr.sbin/bhyve/pci_lpc.c
@@ -32,7 +32,6 @@ __FBSDID("$FreeBSD$");
 
 #include <sys/types.h>
 #include <machine/vmm.h>
-#include <machine/vmm_dev.h>
 
 #include <stdio.h>
 #include <stdlib.h>
diff --git a/usr.sbin/bhyvectl/bhyvectl.c b/usr.sbin/bhyvectl/bhyvectl.c
index ceee33a..e77f0d7 100644
--- a/usr.sbin/bhyvectl/bhyvectl.c
+++ b/usr.sbin/bhyvectl/bhyvectl.c
@@ -193,7 +193,9 @@ usage(void)
 	"       [--assert-lapic-lvt=<pin>]\n"
 	"       [--inject-nmi]\n"
 	"       [--force-reset]\n"
-	"       [--force-poweroff]\n",
+	"       [--force-poweroff]\n"
+	"       [--get-active-cpus]\n"
+	"       [--get-suspended-cpus]\n",
 	progname);
 	exit(1);
 }
@@ -203,6 +205,7 @@ static int inject_nmi, assert_lapic_lvt;
 static int force_reset, force_poweroff;
 static const char *capname;
 static int create, destroy, get_lowmem, get_highmem;
+static int get_active_cpus, get_suspended_cpus;
 static uint64_t memsize;
 static int set_cr0, get_cr0, set_cr3, get_cr3, set_cr4, get_cr4;
 static int set_efer, get_efer;
@@ -390,6 +393,25 @@ enum {
 	ASSERT_LAPIC_LVT,
 };
 
+static void
+print_cpus(const char *banner, const cpuset_t *cpus)
+{
+	int i, first;
+
+	first = 1;
+	printf("%s:\t", banner);
+	if (!CPU_EMPTY(cpus)) {
+		for (i = 0; i < CPU_SETSIZE; i++) {
+			if (CPU_ISSET(i, cpus)) {
+				printf("%s%d", first ? " " : ", ", i);
+				first = 0;
+			}
+		}
+	} else
+		printf(" (none)");
+	printf("\n");
+}
+
 int
 main(int argc, char *argv[])
 {
@@ -401,6 +423,7 @@ main(int argc, char *argv[])
 	uint64_t ctl, eptp, bm, addr, u64, pteval[4], *pte;
 	struct vmctx *ctx;
 	int wired;
+	cpuset_t cpus;
 
 	uint64_t cr0, cr3, cr4, dr7, rsp, rip, rflags, efer, pat;
 	uint64_t rax, rbx, rcx, rdx, rsi, rdi, rbp;
@@ -570,6 +593,8 @@ main(int argc, char *argv[])
 		{ "inject-nmi",	NO_ARG,		&inject_nmi,	1 },
 		{ "force-reset",	NO_ARG,	&force_reset,	1 },
 		{ "force-poweroff", NO_ARG,	&force_poweroff, 1 },
+		{ "get-active-cpus", NO_ARG,	&get_active_cpus, 1 },
+		{ "get-suspended-cpus", NO_ARG,	&get_suspended_cpus, 1 },
 		{ NULL,		0,		NULL,		0 }
 	};
 
@@ -1529,6 +1554,18 @@ main(int argc, char *argv[])
 		}
 	}
 
+	if (!error && (get_active_cpus || get_all)) {
+		error = vm_active_cpus(ctx, &cpus);
+		if (!error)
+			print_cpus("active cpus", &cpus);
+	}
+
+	if (!error && (get_suspended_cpus || get_all)) {
+		error = vm_suspended_cpus(ctx, &cpus);
+		if (!error)
+			print_cpus("suspended cpus", &cpus);
+	}
+
 	if (!error && run) {
 		error = vm_get_register(ctx, vcpu, VM_REG_GUEST_RIP, &rip);
 		assert(error == 0);