diff --git a/sys/amd64/include/vmm.h b/sys/amd64/include/vmm.h
--- a/sys/amd64/include/vmm.h
+++ b/sys/amd64/include/vmm.h
@@ -319,9 +319,12 @@
 typedef void (*vm_rendezvous_func_t)(struct vcpu *vcpu, void *arg);
 int vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest,
     vm_rendezvous_func_t func, void *arg);
+
 cpuset_t vm_active_cpus(struct vm *vm);
 cpuset_t vm_debug_cpus(struct vm *vm);
 cpuset_t vm_suspended_cpus(struct vm *vm);
+cpuset_t vm_start_cpus(struct vm *vm, const cpuset_t *tostart);
+void vm_await_start(struct vm *vm, const cpuset_t *waiting);
 #endif	/* _SYS__CPUSET_H_ */
 
 static __inline int
diff --git a/sys/amd64/vmm/io/vlapic.c b/sys/amd64/vmm/io/vlapic.c
--- a/sys/amd64/vmm/io/vlapic.c
+++ b/sys/amd64/vmm/io/vlapic.c
@@ -1039,7 +1039,6 @@
 	cpuset_t dmask, ipimask;
 	uint64_t icrval;
 	uint32_t dest, vec, mode, shorthand;
-	struct vlapic *vlapic2;
 	struct vcpu *vcpu;
 	struct vm_exit *vmexit;
 	struct LAPIC *lapic;
@@ -1128,14 +1127,9 @@
 			    i == vlapic->vcpuid)
 				break;
 
-			/*
-			 * Userland which doesn't support the IPI exit
-			 * requires that the boot state is set to SIPI
-			 * here.
-			 */
-			vcpu = vm_vcpu(vlapic->vm, i);
-			vlapic2 = vm_lapic(vcpu);
-			vlapic2->boot_state = BS_SIPI;
+			/* vCPU i is waiting for SIPI. */
+			CPU_SETOF(i, &dmask);
+			vm_await_start(vlapic->vm, &dmask);
 			break;
 		}
 
@@ -1158,11 +1152,10 @@
 			/*
 			 * Ignore SIPIs in any state other than wait-for-SIPI
 			 */
-			vcpu = vm_vcpu(vlapic->vm, i);
-			vlapic2 = vm_lapic(vcpu);
-			if (vlapic2->boot_state != BS_SIPI)
+			CPU_SETOF(i, &dmask);
+			dmask = vm_start_cpus(vlapic->vm, &dmask);
+			if (CPU_EMPTY(&dmask))
 				break;
-			vlapic2->boot_state = BS_RUNNING;
 
 			vmexit = vm_exitinfo(vlapic->vcpu);
 			vmexit->exitcode = VM_EXITCODE_SPINUP_AP;
@@ -1173,19 +1166,10 @@
 			break;
 		}
 
-		CPU_FOREACH_ISSET(i, &dmask) {
-			vcpu = vm_vcpu(vlapic->vm, i);
-			vlapic2 = vm_lapic(vcpu);
-
-			/*
-			 * Ignore SIPIs in any state other than wait-for-SIPI
-			 */
-			if (vlapic2->boot_state != BS_SIPI)
-				continue;
-			vlapic2->boot_state = BS_RUNNING;
-			CPU_SET(i, &ipimask);
-		}
-
+		/*
+		 * Ignore SIPIs in any state other than wait-for-SIPI
+		 */
+		ipimask = vm_start_cpus(vlapic->vm, &dmask);
 		break;
 	default:
 		return (1);
@@ -1210,9 +1194,6 @@
 	struct vlapic *vlapic = vm_lapic(vcpu);
 
 	vlapic_reset(vlapic);
-
-	/* vlapic_reset modifies the boot state. */
-	vlapic->boot_state = BS_SIPI;
 }
 
 int
@@ -1223,6 +1204,7 @@
 	case APIC_DELMODE_INIT:
 		vm_smp_rendezvous(vcpu, vme->u.ipi.dmask, vlapic_handle_init,
 		    NULL);
+		vm_await_start(vcpu_vm(vcpu), &vme->u.ipi.dmask);
 		break;
 	case APIC_DELMODE_STARTUP:
 		break;
@@ -1598,11 +1580,6 @@
 	lapic->dcr_timer = 0;
 	vlapic_dcr_write_handler(vlapic);
 
-	if (vlapic->vcpuid == 0)
-		vlapic->boot_state = BS_RUNNING;	/* BSP */
-	else
-		vlapic->boot_state = BS_INIT;		/* AP */
-
 	vlapic->svr_last = lapic->svr;
 }
 
@@ -1900,7 +1877,6 @@
 				      sizeof(vlapic->isrvec_stk),
 				      meta, ret, done);
 		SNAPSHOT_VAR_OR_LEAVE(vlapic->isrvec_stk_top, meta, ret, done);
-		SNAPSHOT_VAR_OR_LEAVE(vlapic->boot_state, meta, ret, done);
 
 		SNAPSHOT_BUF_OR_LEAVE(vlapic->lvt_last,
 				      sizeof(vlapic->lvt_last),
diff --git a/sys/amd64/vmm/io/vlapic_priv.h b/sys/amd64/vmm/io/vlapic_priv.h
--- a/sys/amd64/vmm/io/vlapic_priv.h
+++ b/sys/amd64/vmm/io/vlapic_priv.h
@@ -125,12 +125,6 @@
 	VLAPIC_CTR1((vlapic), msg " isr7 0x%08x", isrptr[7 << 2]);	\
 } while (0)
 
-enum boot_state {
-	BS_INIT,
-	BS_SIPI,
-	BS_RUNNING
-};
-
 /*
  * 16 priority levels with at most one vector injected per level.
  */
@@ -175,7 +169,6 @@
 	int		isrvec_stk_top;
 
 	uint64_t	msr_apicbase;
-	enum boot_state	boot_state;
 
 	/*
 	 * Copies of some registers in the virtual APIC page. We do this for
diff --git a/sys/amd64/vmm/vmm.c b/sys/amd64/vmm/vmm.c
--- a/sys/amd64/vmm/vmm.c
+++ b/sys/amd64/vmm/vmm.c
@@ -173,6 +173,7 @@
 	struct vrtc	*vrtc;			/* (o) virtual RTC */
 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
 	volatile cpuset_t debug_cpus;		/* (i) vcpus stopped for debug */
+	cpuset_t	startup_cpus;		/* (i) [r] waiting for startup */
 	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 */
@@ -486,6 +487,7 @@
 
 	CPU_ZERO(&vm->active_cpus);
 	CPU_ZERO(&vm->debug_cpus);
+	CPU_ZERO(&vm->startup_cpus);
 
 	vm->suspend = 0;
 	CPU_ZERO(&vm->suspended_cpus);
@@ -2421,6 +2423,30 @@
 	return (vm->suspended_cpus);
 }
 
+/*
+ * Returns the subset of vCPUs in tostart that are awaiting startup.
+ * These vCPUs are also marked as no longer awaiting startup.
+ */
+cpuset_t
+vm_start_cpus(struct vm *vm, const cpuset_t *tostart)
+{
+	cpuset_t set;
+
+	mtx_lock(&vm->rendezvous_mtx);
+	CPU_AND(&set, &vm->startup_cpus, tostart);
+	CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set);
+	mtx_unlock(&vm->rendezvous_mtx);
+	return (set);
+}
+
+void
+vm_await_start(struct vm *vm, const cpuset_t *waiting)
+{
+	mtx_lock(&vm->rendezvous_mtx);
+	CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting);
+	mtx_unlock(&vm->rendezvous_mtx);
+}
+
 void *
 vcpu_stats(struct vcpu *vcpu)
 {
@@ -2769,6 +2795,7 @@
 	if (ret != 0)
 		goto done;
 
+	SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done);
 done:
 	return (ret);
 }