MFC r256645.

Add a new capability, VM_CAP_ENABLE_INVPCID, that can be enabled to expose
'invpcid' instruction to the guest. Currently bhyve will try to enable this
capability unconditionally if it is available.

Consolidate code in bhyve to set the capabilities so it is no longer
duplicated in BSP and AP bringup.

Add a sysctl 'vm.pmap.invpcid_works' to display whether the 'invpcid'
instruction is available.

Approved by:	re (hrs)

11 files changed, 93 insertions, 53 deletions

diff --git a/lib/libvmmapi/vmmapi.c b/lib/libvmmapi/vmmapi.c
index 810b39e..bb69358 100644
--- a/lib/libvmmapi/vmmapi.c
+++ b/lib/libvmmapi/vmmapi.c
@@ -415,6 +415,7 @@ static struct {
 	{ "mtrap_exit",		VM_CAP_MTRAP_EXIT },
 	{ "pause_exit",		VM_CAP_PAUSE_EXIT },
 	{ "unrestricted_guest",	VM_CAP_UNRESTRICTED_GUEST },
+	{ "enable_invpcid",	VM_CAP_ENABLE_INVPCID },
 	{ 0 }
 };
 
diff --git a/sys/amd64/amd64/pmap.c b/sys/amd64/amd64/pmap.c
index c0af020..d60cbd1 100644
--- a/sys/amd64/amd64/pmap.c
+++ b/sys/amd64/amd64/pmap.c
@@ -371,6 +371,8 @@ int pmap_pcid_enabled = 1;
 SYSCTL_INT(_vm_pmap, OID_AUTO, pcid_enabled, CTLFLAG_RDTUN, &pmap_pcid_enabled,
     0, "Is TLB Context ID enabled ?");
 int invpcid_works = 0;
+SYSCTL_INT(_vm_pmap, OID_AUTO, invpcid_works, CTLFLAG_RD, &invpcid_works, 0,
+    "Is the invpcid instruction available ?");
 
 static int
 pmap_pcid_save_cnt_proc(SYSCTL_HANDLER_ARGS)
diff --git a/sys/amd64/include/vmm.h b/sys/amd64/include/vmm.h
index 64dbc58..e8bc409 100644
--- a/sys/amd64/include/vmm.h
+++ b/sys/amd64/include/vmm.h
@@ -223,6 +223,7 @@ enum vm_cap_type {
 	VM_CAP_MTRAP_EXIT,
 	VM_CAP_PAUSE_EXIT,
 	VM_CAP_UNRESTRICTED_GUEST,
+	VM_CAP_ENABLE_INVPCID,
 	VM_CAP_MAX
 };
 
diff --git a/sys/amd64/vmm/intel/vmx.c b/sys/amd64/vmm/intel/vmx.c
index 5b65de6..1241b14 100644
--- a/sys/amd64/vmm/intel/vmx.c
+++ b/sys/amd64/vmm/intel/vmx.c
@@ -164,6 +164,7 @@ static int cap_halt_exit;
 static int cap_pause_exit;
 static int cap_unrestricted_guest;
 static int cap_monitor_trap;
+static int cap_invpcid;
  
 static struct unrhdr *vpid_unr;
 static u_int vpid_alloc_failed;
@@ -660,6 +661,11 @@ vmx_init(void)
 					PROCBASED2_UNRESTRICTED_GUEST, 0,
 				        &tmp) == 0);
 
+	cap_invpcid = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
+	    MSR_VMX_PROCBASED_CTLS2, PROCBASED2_ENABLE_INVPCID, 0,
+	    &tmp) == 0);
+
+
 	/* Initialize EPT */
 	error = ept_init();
 	if (error) {
@@ -828,6 +834,7 @@ vmx_vminit(struct vm *vm, pmap_t pmap)
 
 		vmx->cap[i].set = 0;
 		vmx->cap[i].proc_ctls = procbased_ctls;
+		vmx->cap[i].proc_ctls2 = procbased_ctls2;
 
 		vmx->state[i].lastcpu = -1;
 		vmx->state[i].vpid = vpid[i];
@@ -1932,6 +1939,10 @@ vmx_getcap(void *arg, int vcpu, int type, int *retval)
 		if (cap_unrestricted_guest)
 			ret = 0;
 		break;
+	case VM_CAP_ENABLE_INVPCID:
+		if (cap_invpcid)
+			ret = 0;
+		break;
 	default:
 		break;
 	}
@@ -1988,11 +1999,21 @@ vmx_setcap(void *arg, int vcpu, int type, int val)
 	case VM_CAP_UNRESTRICTED_GUEST:
 		if (cap_unrestricted_guest) {
 			retval = 0;
-			baseval = procbased_ctls2;
+			pptr = &vmx->cap[vcpu].proc_ctls2;
+			baseval = *pptr;
 			flag = PROCBASED2_UNRESTRICTED_GUEST;
 			reg = VMCS_SEC_PROC_BASED_CTLS;
 		}
 		break;
+	case VM_CAP_ENABLE_INVPCID:
+		if (cap_invpcid) {
+			retval = 0;
+			pptr = &vmx->cap[vcpu].proc_ctls2;
+			baseval = *pptr;
+			flag = PROCBASED2_ENABLE_INVPCID;
+			reg = VMCS_SEC_PROC_BASED_CTLS;
+		}
+		break;
 	default:
 		break;
 	}
diff --git a/sys/amd64/vmm/intel/vmx.h b/sys/amd64/vmm/intel/vmx.h
index 8e57764..450bb66 100644
--- a/sys/amd64/vmm/intel/vmx.h
+++ b/sys/amd64/vmm/intel/vmx.h
@@ -84,6 +84,7 @@ struct vmxctx {
 struct vmxcap {
 	int	set;
 	uint32_t proc_ctls;
+	uint32_t proc_ctls2;
 };
 
 struct vmxstate {
diff --git a/sys/amd64/vmm/intel/vmx_controls.h b/sys/amd64/vmm/intel/vmx_controls.h
index 31f29f8..b37e6d1 100644
--- a/sys/amd64/vmm/intel/vmx_controls.h
+++ b/sys/amd64/vmm/intel/vmx_controls.h
@@ -68,6 +68,7 @@
 #define	PROCBASED2_WBINVD_EXITING	(1 << 6)
 #define	PROCBASED2_UNRESTRICTED_GUEST	(1 << 7)
 #define	PROCBASED2_PAUSE_LOOP_EXITING	(1 << 10)
+#define	PROCBASED2_ENABLE_INVPCID	(1 << 12)
 
 /* VM Exit Controls */
 #define	VM_EXIT_SAVE_DEBUG_CONTROLS	(1 << 2)
diff --git a/sys/amd64/vmm/x86.c b/sys/amd64/vmm/x86.c
index c23f1fc..c1fc006e 100644
--- a/sys/amd64/vmm/x86.c
+++ b/sys/amd64/vmm/x86.c
@@ -53,7 +53,7 @@ int
 x86_emulate_cpuid(struct vm *vm, int vcpu_id,
 		  uint32_t *eax, uint32_t *ebx, uint32_t *ecx, uint32_t *edx)
 {
-	int error;
+	int error, enable_invpcid;
 	unsigned int 	func, regs[4];
 	enum x2apic_state x2apic_state;
 
@@ -202,8 +202,22 @@ x86_emulate_cpuid(struct vm *vm, int vcpu_id,
 			regs[0] |= 0x04008000;
 			break;
 
-		case CPUID_0000_0006:
 		case CPUID_0000_0007:
+			regs[0] = 0;
+			regs[1] = 0;
+			regs[2] = 0;
+			regs[3] = 0;
+
+			/* leaf 0 */
+			if (*ecx == 0) {
+				error = vm_get_capability(vm, vcpu_id,
+				    VM_CAP_ENABLE_INVPCID, &enable_invpcid);
+				if (error == 0 && enable_invpcid)
+					regs[1] |= CPUID_STDEXT_INVPCID;
+			}
+			break;
+
+		case CPUID_0000_0006:
 		case CPUID_0000_000A:
 		case CPUID_0000_000D:
 			/*
diff --git a/usr.sbin/bhyve/bhyverun.c b/usr.sbin/bhyve/bhyverun.c
index 09d043d..bebd186 100644
--- a/usr.sbin/bhyve/bhyverun.c
+++ b/usr.sbin/bhyve/bhyverun.c
@@ -493,10 +493,54 @@ num_vcpus_allowed(struct vmctx *ctx)
 		return (1);
 }
 
+void
+fbsdrun_set_capabilities(struct vmctx *ctx, int cpu)
+{
+	int err, tmp;
+
+	if (fbsdrun_vmexit_on_hlt()) {
+		err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp);
+		if (err < 0) {
+			fprintf(stderr, "VM exit on HLT not supported\n");
+			exit(1);
+		}
+		vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1);
+		if (cpu == BSP)
+			handler[VM_EXITCODE_HLT] = vmexit_hlt;
+	}
+
+        if (fbsdrun_vmexit_on_pause()) {
+		/*
+		 * pause exit support required for this mode
+		 */
+		err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp);
+		if (err < 0) {
+			fprintf(stderr,
+			    "SMP mux requested, no pause support\n");
+			exit(1);
+		}
+		vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1);
+		if (cpu == BSP)
+			handler[VM_EXITCODE_PAUSE] = vmexit_pause;
+        }
+
+	if (fbsdrun_disable_x2apic())
+		err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED);
+	else
+		err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED);
+
+	if (err) {
+		fprintf(stderr, "Unable to set x2apic state (%d)\n", err);
+		exit(1);
+	}
+
+	vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1);
+}
+
 int
 main(int argc, char *argv[])
 {
-	int c, error, gdb_port, tmp, err, ioapic, bvmcons;
+	int c, error, gdb_port, err, ioapic, bvmcons;
 	int max_vcpus;
 	struct vmctx *ctx;
 	uint64_t rip;
@@ -586,39 +630,7 @@ main(int argc, char *argv[])
 		exit(1);
 	}
 
-	if (fbsdrun_vmexit_on_hlt()) {
-		err = vm_get_capability(ctx, BSP, VM_CAP_HALT_EXIT, &tmp);
-		if (err < 0) {
-			fprintf(stderr, "VM exit on HLT not supported\n");
-			exit(1);
-		}
-		vm_set_capability(ctx, BSP, VM_CAP_HALT_EXIT, 1);
-		handler[VM_EXITCODE_HLT] = vmexit_hlt;
-	}
-
-        if (fbsdrun_vmexit_on_pause()) {
-		/*
-		 * pause exit support required for this mode
-		 */
-		err = vm_get_capability(ctx, BSP, VM_CAP_PAUSE_EXIT, &tmp);
-		if (err < 0) {
-			fprintf(stderr,
-			    "SMP mux requested, no pause support\n");
-			exit(1);
-		}
-		vm_set_capability(ctx, BSP, VM_CAP_PAUSE_EXIT, 1);
-		handler[VM_EXITCODE_PAUSE] = vmexit_pause;
-        }
-
-	if (fbsdrun_disable_x2apic())
-		err = vm_set_x2apic_state(ctx, BSP, X2APIC_DISABLED);
-	else
-		err = vm_set_x2apic_state(ctx, BSP, X2APIC_ENABLED);
-
-	if (err) {
-		fprintf(stderr, "Unable to set x2apic state (%d)\n", err);
-		exit(1);
-	}
+	fbsdrun_set_capabilities(ctx, BSP);
 
 	err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL);
 	if (err) {
diff --git a/usr.sbin/bhyve/bhyverun.h b/usr.sbin/bhyve/bhyverun.h
index 27c858e..f5246d4 100644
--- a/usr.sbin/bhyve/bhyverun.h
+++ b/usr.sbin/bhyve/bhyverun.h
@@ -41,6 +41,7 @@ extern char *vmname;
 
 void *paddr_guest2host(struct vmctx *ctx, uintptr_t addr, size_t len);
 
+void fbsdrun_set_capabilities(struct vmctx *ctx, int cpu);
 void fbsdrun_addcpu(struct vmctx *ctx, int cpu, uint64_t rip);
 int  fbsdrun_muxed(void);
 int  fbsdrun_vmexit_on_hlt(void);
diff --git a/usr.sbin/bhyve/spinup_ap.c b/usr.sbin/bhyve/spinup_ap.c
index 2632aed..729108a 100644
--- a/usr.sbin/bhyve/spinup_ap.c
+++ b/usr.sbin/bhyve/spinup_ap.c
@@ -85,22 +85,7 @@ spinup_ap(struct vmctx *ctx, int vcpu, int newcpu, uint64_t rip)
 	error = vcpu_reset(ctx, newcpu);
 	assert(error == 0);
 
-	/* Set up capabilities */
-	if (fbsdrun_vmexit_on_hlt()) {
-		error = vm_set_capability(ctx, newcpu, VM_CAP_HALT_EXIT, 1);
-		assert(error == 0);
-	}
-
-	if (fbsdrun_vmexit_on_pause()) {
-		error = vm_set_capability(ctx, newcpu, VM_CAP_PAUSE_EXIT, 1);
-		assert(error == 0);
-	}
-
-	if (fbsdrun_disable_x2apic())
-		error = vm_set_x2apic_state(ctx, newcpu, X2APIC_DISABLED);
-	else
-		error = vm_set_x2apic_state(ctx, newcpu, X2APIC_ENABLED);
-	assert(error == 0);
+	fbsdrun_set_capabilities(ctx, newcpu);
 
 	/*
 	 * Enable the 'unrestricted guest' mode for 'newcpu'.
diff --git a/usr.sbin/bhyvectl/bhyvectl.c b/usr.sbin/bhyvectl/bhyvectl.c
index d6b32b8..0e92204 100644
--- a/usr.sbin/bhyvectl/bhyvectl.c
+++ b/usr.sbin/bhyvectl/bhyvectl.c
@@ -1495,6 +1495,7 @@ main(int argc, char *argv[])
 					vm_capability_type2name(captype),
 					val ? "set" : "not set", vcpu);
 			} else if (errno == ENOENT) {
+				error = 0;
 				printf("Capability \"%s\" is not available\n",
 					vm_capability_type2name(captype));
 			} else {