xref: /freebsd/sys/amd64/vmm/intel/vmx.c (revision 1a61beb0549e05b33df31380e427d90f6e46ff7e)
1 /*-
2  * Copyright (c) 2011 NetApp, Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/smp.h>
35 #include <sys/kernel.h>
36 #include <sys/malloc.h>
37 #include <sys/pcpu.h>
38 #include <sys/proc.h>
39 #include <sys/sysctl.h>
40 
41 #include <vm/vm.h>
42 #include <vm/pmap.h>
43 
44 #include <machine/psl.h>
45 #include <machine/cpufunc.h>
46 #include <machine/md_var.h>
47 #include <machine/segments.h>
48 #include <machine/smp.h>
49 #include <machine/specialreg.h>
50 #include <machine/vmparam.h>
51 
52 #include <machine/vmm.h>
53 #include "vmm_host.h"
54 #include "vmm_ipi.h"
55 #include "vmm_msr.h"
56 #include "vmm_ktr.h"
57 #include "vmm_stat.h"
58 #include "vlapic.h"
59 #include "vlapic_priv.h"
60 
61 #include "vmx_msr.h"
62 #include "ept.h"
63 #include "vmx_cpufunc.h"
64 #include "vmx.h"
65 #include "x86.h"
66 #include "vmx_controls.h"
67 
68 #define	PINBASED_CTLS_ONE_SETTING					\
69 	(PINBASED_EXTINT_EXITING	|				\
70 	 PINBASED_NMI_EXITING		|				\
71 	 PINBASED_VIRTUAL_NMI)
72 #define	PINBASED_CTLS_ZERO_SETTING	0
73 
74 #define PROCBASED_CTLS_WINDOW_SETTING					\
75 	(PROCBASED_INT_WINDOW_EXITING	|				\
76 	 PROCBASED_NMI_WINDOW_EXITING)
77 
78 #define	PROCBASED_CTLS_ONE_SETTING 					\
79 	(PROCBASED_SECONDARY_CONTROLS	|				\
80 	 PROCBASED_IO_EXITING		|				\
81 	 PROCBASED_MSR_BITMAPS		|				\
82 	 PROCBASED_CTLS_WINDOW_SETTING)
83 #define	PROCBASED_CTLS_ZERO_SETTING	\
84 	(PROCBASED_CR3_LOAD_EXITING |	\
85 	PROCBASED_CR3_STORE_EXITING |	\
86 	PROCBASED_IO_BITMAPS)
87 
88 #define	PROCBASED_CTLS2_ONE_SETTING	PROCBASED2_ENABLE_EPT
89 #define	PROCBASED_CTLS2_ZERO_SETTING	0
90 
91 #define VM_EXIT_CTLS_ONE_SETTING_NO_PAT					\
92 	(VM_EXIT_HOST_LMA			|			\
93 	VM_EXIT_SAVE_EFER			|			\
94 	VM_EXIT_LOAD_EFER)
95 
96 #define	VM_EXIT_CTLS_ONE_SETTING					\
97 	(VM_EXIT_CTLS_ONE_SETTING_NO_PAT       	|			\
98 	VM_EXIT_ACKNOWLEDGE_INTERRUPT		|			\
99 	VM_EXIT_SAVE_PAT			|			\
100 	VM_EXIT_LOAD_PAT)
101 #define	VM_EXIT_CTLS_ZERO_SETTING	VM_EXIT_SAVE_DEBUG_CONTROLS
102 
103 #define	VM_ENTRY_CTLS_ONE_SETTING_NO_PAT	VM_ENTRY_LOAD_EFER
104 
105 #define	VM_ENTRY_CTLS_ONE_SETTING					\
106 	(VM_ENTRY_CTLS_ONE_SETTING_NO_PAT     	|			\
107 	VM_ENTRY_LOAD_PAT)
108 #define	VM_ENTRY_CTLS_ZERO_SETTING					\
109 	(VM_ENTRY_LOAD_DEBUG_CONTROLS		|			\
110 	VM_ENTRY_INTO_SMM			|			\
111 	VM_ENTRY_DEACTIVATE_DUAL_MONITOR)
112 
113 #define	guest_msr_rw(vmx, msr) \
114 	msr_bitmap_change_access((vmx)->msr_bitmap, (msr), MSR_BITMAP_ACCESS_RW)
115 
116 #define	HANDLED		1
117 #define	UNHANDLED	0
118 
119 static MALLOC_DEFINE(M_VMX, "vmx", "vmx");
120 static MALLOC_DEFINE(M_VLAPIC, "vlapic", "vlapic");
121 
122 SYSCTL_DECL(_hw_vmm);
123 SYSCTL_NODE(_hw_vmm, OID_AUTO, vmx, CTLFLAG_RW, NULL, NULL);
124 
125 int vmxon_enabled[MAXCPU];
126 static char vmxon_region[MAXCPU][PAGE_SIZE] __aligned(PAGE_SIZE);
127 
128 static uint32_t pinbased_ctls, procbased_ctls, procbased_ctls2;
129 static uint32_t exit_ctls, entry_ctls;
130 
131 static uint64_t cr0_ones_mask, cr0_zeros_mask;
132 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_ones_mask, CTLFLAG_RD,
133 	     &cr0_ones_mask, 0, NULL);
134 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_zeros_mask, CTLFLAG_RD,
135 	     &cr0_zeros_mask, 0, NULL);
136 
137 static uint64_t cr4_ones_mask, cr4_zeros_mask;
138 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_ones_mask, CTLFLAG_RD,
139 	     &cr4_ones_mask, 0, NULL);
140 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_zeros_mask, CTLFLAG_RD,
141 	     &cr4_zeros_mask, 0, NULL);
142 
143 static int vmx_no_patmsr;
144 
145 static int vmx_initialized;
146 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, initialized, CTLFLAG_RD,
147 	   &vmx_initialized, 0, "Intel VMX initialized");
148 
149 /*
150  * Optional capabilities
151  */
152 static int cap_halt_exit;
153 static int cap_pause_exit;
154 static int cap_unrestricted_guest;
155 static int cap_monitor_trap;
156 static int cap_invpcid;
157 
158 static int virtual_interrupt_delivery;
159 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, virtual_interrupt_delivery, CTLFLAG_RD,
160     &virtual_interrupt_delivery, 0, "APICv virtual interrupt delivery support");
161 
162 static int posted_interrupts;
163 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, posted_interrupts, CTLFLAG_RD,
164     &posted_interrupts, 0, "APICv posted interrupt support");
165 
166 static int pirvec;
167 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, posted_interrupt_vector, CTLFLAG_RD,
168     &pirvec, 0, "APICv posted interrupt vector");
169 
170 static struct unrhdr *vpid_unr;
171 static u_int vpid_alloc_failed;
172 SYSCTL_UINT(_hw_vmm_vmx, OID_AUTO, vpid_alloc_failed, CTLFLAG_RD,
173 	    &vpid_alloc_failed, 0, NULL);
174 
175 /*
176  * Use the last page below 4GB as the APIC access address. This address is
177  * occupied by the boot firmware so it is guaranteed that it will not conflict
178  * with a page in system memory.
179  */
180 #define	APIC_ACCESS_ADDRESS	0xFFFFF000
181 
182 static void vmx_inject_pir(struct vlapic *vlapic);
183 
184 #ifdef KTR
185 static const char *
186 exit_reason_to_str(int reason)
187 {
188 	static char reasonbuf[32];
189 
190 	switch (reason) {
191 	case EXIT_REASON_EXCEPTION:
192 		return "exception";
193 	case EXIT_REASON_EXT_INTR:
194 		return "extint";
195 	case EXIT_REASON_TRIPLE_FAULT:
196 		return "triplefault";
197 	case EXIT_REASON_INIT:
198 		return "init";
199 	case EXIT_REASON_SIPI:
200 		return "sipi";
201 	case EXIT_REASON_IO_SMI:
202 		return "iosmi";
203 	case EXIT_REASON_SMI:
204 		return "smi";
205 	case EXIT_REASON_INTR_WINDOW:
206 		return "intrwindow";
207 	case EXIT_REASON_NMI_WINDOW:
208 		return "nmiwindow";
209 	case EXIT_REASON_TASK_SWITCH:
210 		return "taskswitch";
211 	case EXIT_REASON_CPUID:
212 		return "cpuid";
213 	case EXIT_REASON_GETSEC:
214 		return "getsec";
215 	case EXIT_REASON_HLT:
216 		return "hlt";
217 	case EXIT_REASON_INVD:
218 		return "invd";
219 	case EXIT_REASON_INVLPG:
220 		return "invlpg";
221 	case EXIT_REASON_RDPMC:
222 		return "rdpmc";
223 	case EXIT_REASON_RDTSC:
224 		return "rdtsc";
225 	case EXIT_REASON_RSM:
226 		return "rsm";
227 	case EXIT_REASON_VMCALL:
228 		return "vmcall";
229 	case EXIT_REASON_VMCLEAR:
230 		return "vmclear";
231 	case EXIT_REASON_VMLAUNCH:
232 		return "vmlaunch";
233 	case EXIT_REASON_VMPTRLD:
234 		return "vmptrld";
235 	case EXIT_REASON_VMPTRST:
236 		return "vmptrst";
237 	case EXIT_REASON_VMREAD:
238 		return "vmread";
239 	case EXIT_REASON_VMRESUME:
240 		return "vmresume";
241 	case EXIT_REASON_VMWRITE:
242 		return "vmwrite";
243 	case EXIT_REASON_VMXOFF:
244 		return "vmxoff";
245 	case EXIT_REASON_VMXON:
246 		return "vmxon";
247 	case EXIT_REASON_CR_ACCESS:
248 		return "craccess";
249 	case EXIT_REASON_DR_ACCESS:
250 		return "draccess";
251 	case EXIT_REASON_INOUT:
252 		return "inout";
253 	case EXIT_REASON_RDMSR:
254 		return "rdmsr";
255 	case EXIT_REASON_WRMSR:
256 		return "wrmsr";
257 	case EXIT_REASON_INVAL_VMCS:
258 		return "invalvmcs";
259 	case EXIT_REASON_INVAL_MSR:
260 		return "invalmsr";
261 	case EXIT_REASON_MWAIT:
262 		return "mwait";
263 	case EXIT_REASON_MTF:
264 		return "mtf";
265 	case EXIT_REASON_MONITOR:
266 		return "monitor";
267 	case EXIT_REASON_PAUSE:
268 		return "pause";
269 	case EXIT_REASON_MCE:
270 		return "mce";
271 	case EXIT_REASON_TPR:
272 		return "tpr";
273 	case EXIT_REASON_APIC_ACCESS:
274 		return "apic-access";
275 	case EXIT_REASON_GDTR_IDTR:
276 		return "gdtridtr";
277 	case EXIT_REASON_LDTR_TR:
278 		return "ldtrtr";
279 	case EXIT_REASON_EPT_FAULT:
280 		return "eptfault";
281 	case EXIT_REASON_EPT_MISCONFIG:
282 		return "eptmisconfig";
283 	case EXIT_REASON_INVEPT:
284 		return "invept";
285 	case EXIT_REASON_RDTSCP:
286 		return "rdtscp";
287 	case EXIT_REASON_VMX_PREEMPT:
288 		return "vmxpreempt";
289 	case EXIT_REASON_INVVPID:
290 		return "invvpid";
291 	case EXIT_REASON_WBINVD:
292 		return "wbinvd";
293 	case EXIT_REASON_XSETBV:
294 		return "xsetbv";
295 	case EXIT_REASON_APIC_WRITE:
296 		return "apic-write";
297 	default:
298 		snprintf(reasonbuf, sizeof(reasonbuf), "%d", reason);
299 		return (reasonbuf);
300 	}
301 }
302 #endif	/* KTR */
303 
304 u_long
305 vmx_fix_cr0(u_long cr0)
306 {
307 
308 	return ((cr0 | cr0_ones_mask) & ~cr0_zeros_mask);
309 }
310 
311 u_long
312 vmx_fix_cr4(u_long cr4)
313 {
314 
315 	return ((cr4 | cr4_ones_mask) & ~cr4_zeros_mask);
316 }
317 
318 static void
319 vpid_free(int vpid)
320 {
321 	if (vpid < 0 || vpid > 0xffff)
322 		panic("vpid_free: invalid vpid %d", vpid);
323 
324 	/*
325 	 * VPIDs [0,VM_MAXCPU] are special and are not allocated from
326 	 * the unit number allocator.
327 	 */
328 
329 	if (vpid > VM_MAXCPU)
330 		free_unr(vpid_unr, vpid);
331 }
332 
333 static void
334 vpid_alloc(uint16_t *vpid, int num)
335 {
336 	int i, x;
337 
338 	if (num <= 0 || num > VM_MAXCPU)
339 		panic("invalid number of vpids requested: %d", num);
340 
341 	/*
342 	 * If the "enable vpid" execution control is not enabled then the
343 	 * VPID is required to be 0 for all vcpus.
344 	 */
345 	if ((procbased_ctls2 & PROCBASED2_ENABLE_VPID) == 0) {
346 		for (i = 0; i < num; i++)
347 			vpid[i] = 0;
348 		return;
349 	}
350 
351 	/*
352 	 * Allocate a unique VPID for each vcpu from the unit number allocator.
353 	 */
354 	for (i = 0; i < num; i++) {
355 		x = alloc_unr(vpid_unr);
356 		if (x == -1)
357 			break;
358 		else
359 			vpid[i] = x;
360 	}
361 
362 	if (i < num) {
363 		atomic_add_int(&vpid_alloc_failed, 1);
364 
365 		/*
366 		 * If the unit number allocator does not have enough unique
367 		 * VPIDs then we need to allocate from the [1,VM_MAXCPU] range.
368 		 *
369 		 * These VPIDs are not be unique across VMs but this does not
370 		 * affect correctness because the combined mappings are also
371 		 * tagged with the EP4TA which is unique for each VM.
372 		 *
373 		 * It is still sub-optimal because the invvpid will invalidate
374 		 * combined mappings for a particular VPID across all EP4TAs.
375 		 */
376 		while (i-- > 0)
377 			vpid_free(vpid[i]);
378 
379 		for (i = 0; i < num; i++)
380 			vpid[i] = i + 1;
381 	}
382 }
383 
384 static void
385 vpid_init(void)
386 {
387 	/*
388 	 * VPID 0 is required when the "enable VPID" execution control is
389 	 * disabled.
390 	 *
391 	 * VPIDs [1,VM_MAXCPU] are used as the "overflow namespace" when the
392 	 * unit number allocator does not have sufficient unique VPIDs to
393 	 * satisfy the allocation.
394 	 *
395 	 * The remaining VPIDs are managed by the unit number allocator.
396 	 */
397 	vpid_unr = new_unrhdr(VM_MAXCPU + 1, 0xffff, NULL);
398 }
399 
400 static void
401 msr_save_area_init(struct msr_entry *g_area, int *g_count)
402 {
403 	int cnt;
404 
405 	static struct msr_entry guest_msrs[] = {
406 		{ MSR_KGSBASE, 0, 0 },
407 	};
408 
409 	cnt = sizeof(guest_msrs) / sizeof(guest_msrs[0]);
410 	if (cnt > GUEST_MSR_MAX_ENTRIES)
411 		panic("guest msr save area overrun");
412 	bcopy(guest_msrs, g_area, sizeof(guest_msrs));
413 	*g_count = cnt;
414 }
415 
416 static void
417 vmx_disable(void *arg __unused)
418 {
419 	struct invvpid_desc invvpid_desc = { 0 };
420 	struct invept_desc invept_desc = { 0 };
421 
422 	if (vmxon_enabled[curcpu]) {
423 		/*
424 		 * See sections 25.3.3.3 and 25.3.3.4 in Intel Vol 3b.
425 		 *
426 		 * VMXON or VMXOFF are not required to invalidate any TLB
427 		 * caching structures. This prevents potential retention of
428 		 * cached information in the TLB between distinct VMX episodes.
429 		 */
430 		invvpid(INVVPID_TYPE_ALL_CONTEXTS, invvpid_desc);
431 		invept(INVEPT_TYPE_ALL_CONTEXTS, invept_desc);
432 		vmxoff();
433 	}
434 	load_cr4(rcr4() & ~CR4_VMXE);
435 }
436 
437 static int
438 vmx_cleanup(void)
439 {
440 
441 	if (pirvec != 0)
442 		vmm_ipi_free(pirvec);
443 
444 	if (vpid_unr != NULL) {
445 		delete_unrhdr(vpid_unr);
446 		vpid_unr = NULL;
447 	}
448 
449 	smp_rendezvous(NULL, vmx_disable, NULL, NULL);
450 
451 	return (0);
452 }
453 
454 static void
455 vmx_enable(void *arg __unused)
456 {
457 	int error;
458 
459 	load_cr4(rcr4() | CR4_VMXE);
460 
461 	*(uint32_t *)vmxon_region[curcpu] = vmx_revision();
462 	error = vmxon(vmxon_region[curcpu]);
463 	if (error == 0)
464 		vmxon_enabled[curcpu] = 1;
465 }
466 
467 static void
468 vmx_restore(void)
469 {
470 
471 	if (vmxon_enabled[curcpu])
472 		vmxon(vmxon_region[curcpu]);
473 }
474 
475 static int
476 vmx_init(int ipinum)
477 {
478 	int error, use_tpr_shadow;
479 	uint64_t fixed0, fixed1, feature_control;
480 	uint32_t tmp, procbased2_vid_bits;
481 
482 	/* CPUID.1:ECX[bit 5] must be 1 for processor to support VMX */
483 	if (!(cpu_feature2 & CPUID2_VMX)) {
484 		printf("vmx_init: processor does not support VMX operation\n");
485 		return (ENXIO);
486 	}
487 
488 	/*
489 	 * Verify that MSR_IA32_FEATURE_CONTROL lock and VMXON enable bits
490 	 * are set (bits 0 and 2 respectively).
491 	 */
492 	feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL);
493 	if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 0 ||
494 	    (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) {
495 		printf("vmx_init: VMX operation disabled by BIOS\n");
496 		return (ENXIO);
497 	}
498 
499 	/* Check support for primary processor-based VM-execution controls */
500 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
501 			       MSR_VMX_TRUE_PROCBASED_CTLS,
502 			       PROCBASED_CTLS_ONE_SETTING,
503 			       PROCBASED_CTLS_ZERO_SETTING, &procbased_ctls);
504 	if (error) {
505 		printf("vmx_init: processor does not support desired primary "
506 		       "processor-based controls\n");
507 		return (error);
508 	}
509 
510 	/* Clear the processor-based ctl bits that are set on demand */
511 	procbased_ctls &= ~PROCBASED_CTLS_WINDOW_SETTING;
512 
513 	/* Check support for secondary processor-based VM-execution controls */
514 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
515 			       MSR_VMX_PROCBASED_CTLS2,
516 			       PROCBASED_CTLS2_ONE_SETTING,
517 			       PROCBASED_CTLS2_ZERO_SETTING, &procbased_ctls2);
518 	if (error) {
519 		printf("vmx_init: processor does not support desired secondary "
520 		       "processor-based controls\n");
521 		return (error);
522 	}
523 
524 	/* Check support for VPID */
525 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2,
526 			       PROCBASED2_ENABLE_VPID, 0, &tmp);
527 	if (error == 0)
528 		procbased_ctls2 |= PROCBASED2_ENABLE_VPID;
529 
530 	/* Check support for pin-based VM-execution controls */
531 	error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS,
532 			       MSR_VMX_TRUE_PINBASED_CTLS,
533 			       PINBASED_CTLS_ONE_SETTING,
534 			       PINBASED_CTLS_ZERO_SETTING, &pinbased_ctls);
535 	if (error) {
536 		printf("vmx_init: processor does not support desired "
537 		       "pin-based controls\n");
538 		return (error);
539 	}
540 
541 	/* Check support for VM-exit controls */
542 	error = vmx_set_ctlreg(MSR_VMX_EXIT_CTLS, MSR_VMX_TRUE_EXIT_CTLS,
543 			       VM_EXIT_CTLS_ONE_SETTING,
544 			       VM_EXIT_CTLS_ZERO_SETTING,
545 			       &exit_ctls);
546 	if (error) {
547 		/* Try again without the PAT MSR bits */
548 		error = vmx_set_ctlreg(MSR_VMX_EXIT_CTLS,
549 				       MSR_VMX_TRUE_EXIT_CTLS,
550 				       VM_EXIT_CTLS_ONE_SETTING_NO_PAT,
551 				       VM_EXIT_CTLS_ZERO_SETTING,
552 				       &exit_ctls);
553 		if (error) {
554 			printf("vmx_init: processor does not support desired "
555 			       "exit controls\n");
556 			return (error);
557 		} else {
558 			if (bootverbose)
559 				printf("vmm: PAT MSR access not supported\n");
560 			guest_msr_valid(MSR_PAT);
561 			vmx_no_patmsr = 1;
562 		}
563 	}
564 
565 	/* Check support for VM-entry controls */
566 	if (!vmx_no_patmsr) {
567 		error = vmx_set_ctlreg(MSR_VMX_ENTRY_CTLS,
568 				       MSR_VMX_TRUE_ENTRY_CTLS,
569 				       VM_ENTRY_CTLS_ONE_SETTING,
570 				       VM_ENTRY_CTLS_ZERO_SETTING,
571 				       &entry_ctls);
572 	} else {
573 		error = vmx_set_ctlreg(MSR_VMX_ENTRY_CTLS,
574 				       MSR_VMX_TRUE_ENTRY_CTLS,
575 				       VM_ENTRY_CTLS_ONE_SETTING_NO_PAT,
576 				       VM_ENTRY_CTLS_ZERO_SETTING,
577 				       &entry_ctls);
578 	}
579 
580 	if (error) {
581 		printf("vmx_init: processor does not support desired "
582 		       "entry controls\n");
583 		       return (error);
584 	}
585 
586 	/*
587 	 * Check support for optional features by testing them
588 	 * as individual bits
589 	 */
590 	cap_halt_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
591 					MSR_VMX_TRUE_PROCBASED_CTLS,
592 					PROCBASED_HLT_EXITING, 0,
593 					&tmp) == 0);
594 
595 	cap_monitor_trap = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
596 					MSR_VMX_PROCBASED_CTLS,
597 					PROCBASED_MTF, 0,
598 					&tmp) == 0);
599 
600 	cap_pause_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
601 					 MSR_VMX_TRUE_PROCBASED_CTLS,
602 					 PROCBASED_PAUSE_EXITING, 0,
603 					 &tmp) == 0);
604 
605 	cap_unrestricted_guest = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
606 					MSR_VMX_PROCBASED_CTLS2,
607 					PROCBASED2_UNRESTRICTED_GUEST, 0,
608 				        &tmp) == 0);
609 
610 	cap_invpcid = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
611 	    MSR_VMX_PROCBASED_CTLS2, PROCBASED2_ENABLE_INVPCID, 0,
612 	    &tmp) == 0);
613 
614 	/*
615 	 * Check support for virtual interrupt delivery.
616 	 */
617 	procbased2_vid_bits = (PROCBASED2_VIRTUALIZE_APIC_ACCESSES |
618 	    PROCBASED2_VIRTUALIZE_X2APIC_MODE |
619 	    PROCBASED2_APIC_REGISTER_VIRTUALIZATION |
620 	    PROCBASED2_VIRTUAL_INTERRUPT_DELIVERY);
621 
622 	use_tpr_shadow = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
623 	    MSR_VMX_TRUE_PROCBASED_CTLS, PROCBASED_USE_TPR_SHADOW, 0,
624 	    &tmp) == 0);
625 
626 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2,
627 	    procbased2_vid_bits, 0, &tmp);
628 	if (error == 0 && use_tpr_shadow) {
629 		virtual_interrupt_delivery = 1;
630 		TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_vid",
631 		    &virtual_interrupt_delivery);
632 	}
633 
634 	if (virtual_interrupt_delivery) {
635 		procbased_ctls |= PROCBASED_USE_TPR_SHADOW;
636 		procbased_ctls2 |= procbased2_vid_bits;
637 		procbased_ctls2 &= ~PROCBASED2_VIRTUALIZE_X2APIC_MODE;
638 
639 		/*
640 		 * Check for Posted Interrupts only if Virtual Interrupt
641 		 * Delivery is enabled.
642 		 */
643 		error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS,
644 		    MSR_VMX_TRUE_PINBASED_CTLS, PINBASED_POSTED_INTERRUPT, 0,
645 		    &tmp);
646 		if (error == 0) {
647 			pirvec = vmm_ipi_alloc();
648 			if (pirvec == 0) {
649 				if (bootverbose) {
650 					printf("vmx_init: unable to allocate "
651 					    "posted interrupt vector\n");
652 				}
653 			} else {
654 				posted_interrupts = 1;
655 				TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_pir",
656 				    &posted_interrupts);
657 			}
658 		}
659 	}
660 
661 	if (posted_interrupts)
662 		    pinbased_ctls |= PINBASED_POSTED_INTERRUPT;
663 
664 	/* Initialize EPT */
665 	error = ept_init(ipinum);
666 	if (error) {
667 		printf("vmx_init: ept initialization failed (%d)\n", error);
668 		return (error);
669 	}
670 
671 	/*
672 	 * Stash the cr0 and cr4 bits that must be fixed to 0 or 1
673 	 */
674 	fixed0 = rdmsr(MSR_VMX_CR0_FIXED0);
675 	fixed1 = rdmsr(MSR_VMX_CR0_FIXED1);
676 	cr0_ones_mask = fixed0 & fixed1;
677 	cr0_zeros_mask = ~fixed0 & ~fixed1;
678 
679 	/*
680 	 * CR0_PE and CR0_PG can be set to zero in VMX non-root operation
681 	 * if unrestricted guest execution is allowed.
682 	 */
683 	if (cap_unrestricted_guest)
684 		cr0_ones_mask &= ~(CR0_PG | CR0_PE);
685 
686 	/*
687 	 * Do not allow the guest to set CR0_NW or CR0_CD.
688 	 */
689 	cr0_zeros_mask |= (CR0_NW | CR0_CD);
690 
691 	fixed0 = rdmsr(MSR_VMX_CR4_FIXED0);
692 	fixed1 = rdmsr(MSR_VMX_CR4_FIXED1);
693 	cr4_ones_mask = fixed0 & fixed1;
694 	cr4_zeros_mask = ~fixed0 & ~fixed1;
695 
696 	vpid_init();
697 
698 	/* enable VMX operation */
699 	smp_rendezvous(NULL, vmx_enable, NULL, NULL);
700 
701 	vmx_initialized = 1;
702 
703 	return (0);
704 }
705 
706 static void
707 vmx_trigger_hostintr(int vector)
708 {
709 	uintptr_t func;
710 	struct gate_descriptor *gd;
711 
712 	gd = &idt[vector];
713 
714 	KASSERT(vector >= 32 && vector <= 255, ("vmx_trigger_hostintr: "
715 	    "invalid vector %d", vector));
716 	KASSERT(gd->gd_p == 1, ("gate descriptor for vector %d not present",
717 	    vector));
718 	KASSERT(gd->gd_type == SDT_SYSIGT, ("gate descriptor for vector %d "
719 	    "has invalid type %d", vector, gd->gd_type));
720 	KASSERT(gd->gd_dpl == SEL_KPL, ("gate descriptor for vector %d "
721 	    "has invalid dpl %d", vector, gd->gd_dpl));
722 	KASSERT(gd->gd_selector == GSEL(GCODE_SEL, SEL_KPL), ("gate descriptor "
723 	    "for vector %d has invalid selector %d", vector, gd->gd_selector));
724 	KASSERT(gd->gd_ist == 0, ("gate descriptor for vector %d has invalid "
725 	    "IST %d", vector, gd->gd_ist));
726 
727 	func = ((long)gd->gd_hioffset << 16 | gd->gd_looffset);
728 	vmx_call_isr(func);
729 }
730 
731 static int
732 vmx_setup_cr_shadow(int which, struct vmcs *vmcs, uint32_t initial)
733 {
734 	int error, mask_ident, shadow_ident;
735 	uint64_t mask_value;
736 
737 	if (which != 0 && which != 4)
738 		panic("vmx_setup_cr_shadow: unknown cr%d", which);
739 
740 	if (which == 0) {
741 		mask_ident = VMCS_CR0_MASK;
742 		mask_value = cr0_ones_mask | cr0_zeros_mask;
743 		shadow_ident = VMCS_CR0_SHADOW;
744 	} else {
745 		mask_ident = VMCS_CR4_MASK;
746 		mask_value = cr4_ones_mask | cr4_zeros_mask;
747 		shadow_ident = VMCS_CR4_SHADOW;
748 	}
749 
750 	error = vmcs_setreg(vmcs, 0, VMCS_IDENT(mask_ident), mask_value);
751 	if (error)
752 		return (error);
753 
754 	error = vmcs_setreg(vmcs, 0, VMCS_IDENT(shadow_ident), initial);
755 	if (error)
756 		return (error);
757 
758 	return (0);
759 }
760 #define	vmx_setup_cr0_shadow(vmcs,init)	vmx_setup_cr_shadow(0, (vmcs), (init))
761 #define	vmx_setup_cr4_shadow(vmcs,init)	vmx_setup_cr_shadow(4, (vmcs), (init))
762 
763 static void *
764 vmx_vminit(struct vm *vm, pmap_t pmap)
765 {
766 	uint16_t vpid[VM_MAXCPU];
767 	int i, error, guest_msr_count;
768 	struct vmx *vmx;
769 	struct vmcs *vmcs;
770 
771 	vmx = malloc(sizeof(struct vmx), M_VMX, M_WAITOK | M_ZERO);
772 	if ((uintptr_t)vmx & PAGE_MASK) {
773 		panic("malloc of struct vmx not aligned on %d byte boundary",
774 		      PAGE_SIZE);
775 	}
776 	vmx->vm = vm;
777 
778 	vmx->eptp = eptp(vtophys((vm_offset_t)pmap->pm_pml4));
779 
780 	/*
781 	 * Clean up EPTP-tagged guest physical and combined mappings
782 	 *
783 	 * VMX transitions are not required to invalidate any guest physical
784 	 * mappings. So, it may be possible for stale guest physical mappings
785 	 * to be present in the processor TLBs.
786 	 *
787 	 * Combined mappings for this EP4TA are also invalidated for all VPIDs.
788 	 */
789 	ept_invalidate_mappings(vmx->eptp);
790 
791 	msr_bitmap_initialize(vmx->msr_bitmap);
792 
793 	/*
794 	 * It is safe to allow direct access to MSR_GSBASE and MSR_FSBASE.
795 	 * The guest FSBASE and GSBASE are saved and restored during
796 	 * vm-exit and vm-entry respectively. The host FSBASE and GSBASE are
797 	 * always restored from the vmcs host state area on vm-exit.
798 	 *
799 	 * The SYSENTER_CS/ESP/EIP MSRs are identical to FS/GSBASE in
800 	 * how they are saved/restored so can be directly accessed by the
801 	 * guest.
802 	 *
803 	 * Guest KGSBASE is saved and restored in the guest MSR save area.
804 	 * Host KGSBASE is restored before returning to userland from the pcb.
805 	 * There will be a window of time when we are executing in the host
806 	 * kernel context with a value of KGSBASE from the guest. This is ok
807 	 * because the value of KGSBASE is inconsequential in kernel context.
808 	 *
809 	 * MSR_EFER is saved and restored in the guest VMCS area on a
810 	 * VM exit and entry respectively. It is also restored from the
811 	 * host VMCS area on a VM exit.
812 	 */
813 	if (guest_msr_rw(vmx, MSR_GSBASE) ||
814 	    guest_msr_rw(vmx, MSR_FSBASE) ||
815 	    guest_msr_rw(vmx, MSR_SYSENTER_CS_MSR) ||
816 	    guest_msr_rw(vmx, MSR_SYSENTER_ESP_MSR) ||
817 	    guest_msr_rw(vmx, MSR_SYSENTER_EIP_MSR) ||
818 	    guest_msr_rw(vmx, MSR_KGSBASE) ||
819 	    guest_msr_rw(vmx, MSR_EFER))
820 		panic("vmx_vminit: error setting guest msr access");
821 
822 	/*
823 	 * MSR_PAT is saved and restored in the guest VMCS are on a VM exit
824 	 * and entry respectively. It is also restored from the host VMCS
825 	 * area on a VM exit. However, if running on a system with no
826 	 * MSR_PAT save/restore support, leave access disabled so accesses
827 	 * will be trapped.
828 	 */
829 	if (!vmx_no_patmsr && guest_msr_rw(vmx, MSR_PAT))
830 		panic("vmx_vminit: error setting guest pat msr access");
831 
832 	vpid_alloc(vpid, VM_MAXCPU);
833 
834 	if (virtual_interrupt_delivery) {
835 		error = vm_map_mmio(vm, DEFAULT_APIC_BASE, PAGE_SIZE,
836 		    APIC_ACCESS_ADDRESS);
837 		/* XXX this should really return an error to the caller */
838 		KASSERT(error == 0, ("vm_map_mmio(apicbase) error %d", error));
839 	}
840 
841 	for (i = 0; i < VM_MAXCPU; i++) {
842 		vmcs = &vmx->vmcs[i];
843 		vmcs->identifier = vmx_revision();
844 		error = vmclear(vmcs);
845 		if (error != 0) {
846 			panic("vmx_vminit: vmclear error %d on vcpu %d\n",
847 			      error, i);
848 		}
849 
850 		error = vmcs_init(vmcs);
851 		KASSERT(error == 0, ("vmcs_init error %d", error));
852 
853 		VMPTRLD(vmcs);
854 		error = 0;
855 		error += vmwrite(VMCS_HOST_RSP, (u_long)&vmx->ctx[i]);
856 		error += vmwrite(VMCS_EPTP, vmx->eptp);
857 		error += vmwrite(VMCS_PIN_BASED_CTLS, pinbased_ctls);
858 		error += vmwrite(VMCS_PRI_PROC_BASED_CTLS, procbased_ctls);
859 		error += vmwrite(VMCS_SEC_PROC_BASED_CTLS, procbased_ctls2);
860 		error += vmwrite(VMCS_EXIT_CTLS, exit_ctls);
861 		error += vmwrite(VMCS_ENTRY_CTLS, entry_ctls);
862 		error += vmwrite(VMCS_MSR_BITMAP, vtophys(vmx->msr_bitmap));
863 		error += vmwrite(VMCS_VPID, vpid[i]);
864 		if (virtual_interrupt_delivery) {
865 			error += vmwrite(VMCS_APIC_ACCESS, APIC_ACCESS_ADDRESS);
866 			error += vmwrite(VMCS_VIRTUAL_APIC,
867 			    vtophys(&vmx->apic_page[i]));
868 			error += vmwrite(VMCS_EOI_EXIT0, 0);
869 			error += vmwrite(VMCS_EOI_EXIT1, 0);
870 			error += vmwrite(VMCS_EOI_EXIT2, 0);
871 			error += vmwrite(VMCS_EOI_EXIT3, 0);
872 		}
873 		if (posted_interrupts) {
874 			error += vmwrite(VMCS_PIR_VECTOR, pirvec);
875 			error += vmwrite(VMCS_PIR_DESC,
876 			    vtophys(&vmx->pir_desc[i]));
877 		}
878 		VMCLEAR(vmcs);
879 		KASSERT(error == 0, ("vmx_vminit: error customizing the vmcs"));
880 
881 		vmx->cap[i].set = 0;
882 		vmx->cap[i].proc_ctls = procbased_ctls;
883 		vmx->cap[i].proc_ctls2 = procbased_ctls2;
884 
885 		vmx->state[i].lastcpu = -1;
886 		vmx->state[i].vpid = vpid[i];
887 
888 		msr_save_area_init(vmx->guest_msrs[i], &guest_msr_count);
889 
890 		error = vmcs_set_msr_save(vmcs, vtophys(vmx->guest_msrs[i]),
891 		    guest_msr_count);
892 		if (error != 0)
893 			panic("vmcs_set_msr_save error %d", error);
894 
895 		/*
896 		 * Set up the CR0/4 shadows, and init the read shadow
897 		 * to the power-on register value from the Intel Sys Arch.
898 		 *  CR0 - 0x60000010
899 		 *  CR4 - 0
900 		 */
901 		error = vmx_setup_cr0_shadow(vmcs, 0x60000010);
902 		if (error != 0)
903 			panic("vmx_setup_cr0_shadow %d", error);
904 
905 		error = vmx_setup_cr4_shadow(vmcs, 0);
906 		if (error != 0)
907 			panic("vmx_setup_cr4_shadow %d", error);
908 
909 		vmx->ctx[i].pmap = pmap;
910 	}
911 
912 	return (vmx);
913 }
914 
915 static int
916 vmx_handle_cpuid(struct vm *vm, int vcpu, struct vmxctx *vmxctx)
917 {
918 	int handled, func;
919 
920 	func = vmxctx->guest_rax;
921 
922 	handled = x86_emulate_cpuid(vm, vcpu,
923 				    (uint32_t*)(&vmxctx->guest_rax),
924 				    (uint32_t*)(&vmxctx->guest_rbx),
925 				    (uint32_t*)(&vmxctx->guest_rcx),
926 				    (uint32_t*)(&vmxctx->guest_rdx));
927 	return (handled);
928 }
929 
930 static __inline void
931 vmx_run_trace(struct vmx *vmx, int vcpu)
932 {
933 #ifdef KTR
934 	VCPU_CTR1(vmx->vm, vcpu, "Resume execution at %#lx", vmcs_guest_rip());
935 #endif
936 }
937 
938 static __inline void
939 vmx_exit_trace(struct vmx *vmx, int vcpu, uint64_t rip, uint32_t exit_reason,
940 	       int handled)
941 {
942 #ifdef KTR
943 	VCPU_CTR3(vmx->vm, vcpu, "%s %s vmexit at 0x%0lx",
944 		 handled ? "handled" : "unhandled",
945 		 exit_reason_to_str(exit_reason), rip);
946 #endif
947 }
948 
949 static __inline void
950 vmx_astpending_trace(struct vmx *vmx, int vcpu, uint64_t rip)
951 {
952 #ifdef KTR
953 	VCPU_CTR1(vmx->vm, vcpu, "astpending vmexit at 0x%0lx", rip);
954 #endif
955 }
956 
957 static VMM_STAT_INTEL(VCPU_INVVPID_SAVED, "Number of vpid invalidations saved");
958 
959 static void
960 vmx_set_pcpu_defaults(struct vmx *vmx, int vcpu, pmap_t pmap)
961 {
962 	struct vmxstate *vmxstate;
963 	struct invvpid_desc invvpid_desc;
964 
965 	vmxstate = &vmx->state[vcpu];
966 	if (vmxstate->lastcpu == curcpu)
967 		return;
968 
969 	vmxstate->lastcpu = curcpu;
970 
971 	vmm_stat_incr(vmx->vm, vcpu, VCPU_MIGRATIONS, 1);
972 
973 	vmcs_write(VMCS_HOST_TR_BASE, vmm_get_host_trbase());
974 	vmcs_write(VMCS_HOST_GDTR_BASE, vmm_get_host_gdtrbase());
975 	vmcs_write(VMCS_HOST_GS_BASE, vmm_get_host_gsbase());
976 
977 	/*
978 	 * If we are using VPIDs then invalidate all mappings tagged with 'vpid'
979 	 *
980 	 * We do this because this vcpu was executing on a different host
981 	 * cpu when it last ran. We do not track whether it invalidated
982 	 * mappings associated with its 'vpid' during that run. So we must
983 	 * assume that the mappings associated with 'vpid' on 'curcpu' are
984 	 * stale and invalidate them.
985 	 *
986 	 * Note that we incur this penalty only when the scheduler chooses to
987 	 * move the thread associated with this vcpu between host cpus.
988 	 *
989 	 * Note also that this will invalidate mappings tagged with 'vpid'
990 	 * for "all" EP4TAs.
991 	 */
992 	if (vmxstate->vpid != 0) {
993 		if (pmap->pm_eptgen == vmx->eptgen[curcpu]) {
994 			invvpid_desc._res1 = 0;
995 			invvpid_desc._res2 = 0;
996 			invvpid_desc.vpid = vmxstate->vpid;
997 			invvpid(INVVPID_TYPE_SINGLE_CONTEXT, invvpid_desc);
998 		} else {
999 			/*
1000 			 * The invvpid can be skipped if an invept is going to
1001 			 * be performed before entering the guest. The invept
1002 			 * will invalidate combined mappings tagged with
1003 			 * 'vmx->eptp' for all vpids.
1004 			 */
1005 			vmm_stat_incr(vmx->vm, vcpu, VCPU_INVVPID_SAVED, 1);
1006 		}
1007 	}
1008 }
1009 
1010 /*
1011  * We depend on 'procbased_ctls' to have the Interrupt Window Exiting bit set.
1012  */
1013 CTASSERT((PROCBASED_CTLS_ONE_SETTING & PROCBASED_INT_WINDOW_EXITING) != 0);
1014 
1015 static void __inline
1016 vmx_set_int_window_exiting(struct vmx *vmx, int vcpu)
1017 {
1018 
1019 	if ((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) == 0) {
1020 		vmx->cap[vcpu].proc_ctls |= PROCBASED_INT_WINDOW_EXITING;
1021 		vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls);
1022 		VCPU_CTR0(vmx->vm, vcpu, "Enabling interrupt window exiting");
1023 	}
1024 }
1025 
1026 static void __inline
1027 vmx_clear_int_window_exiting(struct vmx *vmx, int vcpu)
1028 {
1029 
1030 	KASSERT((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0,
1031 	    ("intr_window_exiting not set: %#x", vmx->cap[vcpu].proc_ctls));
1032 	vmx->cap[vcpu].proc_ctls &= ~PROCBASED_INT_WINDOW_EXITING;
1033 	vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls);
1034 	VCPU_CTR0(vmx->vm, vcpu, "Disabling interrupt window exiting");
1035 }
1036 
1037 static void __inline
1038 vmx_set_nmi_window_exiting(struct vmx *vmx, int vcpu)
1039 {
1040 
1041 	if ((vmx->cap[vcpu].proc_ctls & PROCBASED_NMI_WINDOW_EXITING) == 0) {
1042 		vmx->cap[vcpu].proc_ctls |= PROCBASED_NMI_WINDOW_EXITING;
1043 		vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls);
1044 		VCPU_CTR0(vmx->vm, vcpu, "Enabling NMI window exiting");
1045 	}
1046 }
1047 
1048 static void __inline
1049 vmx_clear_nmi_window_exiting(struct vmx *vmx, int vcpu)
1050 {
1051 
1052 	KASSERT((vmx->cap[vcpu].proc_ctls & PROCBASED_NMI_WINDOW_EXITING) != 0,
1053 	    ("nmi_window_exiting not set %#x", vmx->cap[vcpu].proc_ctls));
1054 	vmx->cap[vcpu].proc_ctls &= ~PROCBASED_NMI_WINDOW_EXITING;
1055 	vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls);
1056 	VCPU_CTR0(vmx->vm, vcpu, "Disabling NMI window exiting");
1057 }
1058 
1059 #define	NMI_BLOCKING	(VMCS_INTERRUPTIBILITY_NMI_BLOCKING |		\
1060 			 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING)
1061 #define	HWINTR_BLOCKING	(VMCS_INTERRUPTIBILITY_STI_BLOCKING |		\
1062 			 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING)
1063 
1064 static void
1065 vmx_inject_nmi(struct vmx *vmx, int vcpu)
1066 {
1067 	uint32_t gi, info;
1068 
1069 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1070 	KASSERT((gi & NMI_BLOCKING) == 0, ("vmx_inject_nmi: invalid guest "
1071 	    "interruptibility-state %#x", gi));
1072 
1073 	info = vmcs_read(VMCS_ENTRY_INTR_INFO);
1074 	KASSERT((info & VMCS_INTR_VALID) == 0, ("vmx_inject_nmi: invalid "
1075 	    "VM-entry interruption information %#x", info));
1076 
1077 	/*
1078 	 * Inject the virtual NMI. The vector must be the NMI IDT entry
1079 	 * or the VMCS entry check will fail.
1080 	 */
1081 	info = IDT_NMI | VMCS_INTR_T_NMI | VMCS_INTR_VALID;
1082 	vmcs_write(VMCS_ENTRY_INTR_INFO, info);
1083 
1084 	VCPU_CTR0(vmx->vm, vcpu, "Injecting vNMI");
1085 
1086 	/* Clear the request */
1087 	vm_nmi_clear(vmx->vm, vcpu);
1088 }
1089 
1090 static void
1091 vmx_inject_interrupts(struct vmx *vmx, int vcpu, struct vlapic *vlapic)
1092 {
1093 	int vector, need_nmi_exiting;
1094 	uint64_t rflags;
1095 	uint32_t gi, info;
1096 
1097 	if (vm_nmi_pending(vmx->vm, vcpu)) {
1098 		/*
1099 		 * If there are no conditions blocking NMI injection then
1100 		 * inject it directly here otherwise enable "NMI window
1101 		 * exiting" to inject it as soon as we can.
1102 		 *
1103 		 * We also check for STI_BLOCKING because some implementations
1104 		 * don't allow NMI injection in this case. If we are running
1105 		 * on a processor that doesn't have this restriction it will
1106 		 * immediately exit and the NMI will be injected in the
1107 		 * "NMI window exiting" handler.
1108 		 */
1109 		need_nmi_exiting = 1;
1110 		gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1111 		if ((gi & (HWINTR_BLOCKING | NMI_BLOCKING)) == 0) {
1112 			info = vmcs_read(VMCS_ENTRY_INTR_INFO);
1113 			if ((info & VMCS_INTR_VALID) == 0) {
1114 				vmx_inject_nmi(vmx, vcpu);
1115 				need_nmi_exiting = 0;
1116 			} else {
1117 				VCPU_CTR1(vmx->vm, vcpu, "Cannot inject NMI "
1118 				    "due to VM-entry intr info %#x", info);
1119 			}
1120 		} else {
1121 			VCPU_CTR1(vmx->vm, vcpu, "Cannot inject NMI due to "
1122 			    "Guest Interruptibility-state %#x", gi);
1123 		}
1124 
1125 		if (need_nmi_exiting)
1126 			vmx_set_nmi_window_exiting(vmx, vcpu);
1127 	}
1128 
1129 	if (virtual_interrupt_delivery) {
1130 		vmx_inject_pir(vlapic);
1131 		return;
1132 	}
1133 
1134 	/*
1135 	 * If interrupt-window exiting is already in effect then don't bother
1136 	 * checking for pending interrupts. This is just an optimization and
1137 	 * not needed for correctness.
1138 	 */
1139 	if ((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0) {
1140 		VCPU_CTR0(vmx->vm, vcpu, "Skip interrupt injection due to "
1141 		    "pending int_window_exiting");
1142 		return;
1143 	}
1144 
1145 	/* Ask the local apic for a vector to inject */
1146 	if (!vlapic_pending_intr(vlapic, &vector))
1147 		return;
1148 
1149 	KASSERT(vector >= 32 && vector <= 255, ("invalid vector %d", vector));
1150 
1151 	/* Check RFLAGS.IF and the interruptibility state of the guest */
1152 	rflags = vmcs_read(VMCS_GUEST_RFLAGS);
1153 	if ((rflags & PSL_I) == 0) {
1154 		VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to "
1155 		    "rflags %#lx", vector, rflags);
1156 		goto cantinject;
1157 	}
1158 
1159 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1160 	if (gi & HWINTR_BLOCKING) {
1161 		VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to "
1162 		    "Guest Interruptibility-state %#x", vector, gi);
1163 		goto cantinject;
1164 	}
1165 
1166 	info = vmcs_read(VMCS_ENTRY_INTR_INFO);
1167 	if (info & VMCS_INTR_VALID) {
1168 		/*
1169 		 * This is expected and could happen for multiple reasons:
1170 		 * - A vectoring VM-entry was aborted due to astpending
1171 		 * - A VM-exit happened during event injection.
1172 		 * - An NMI was injected above or after "NMI window exiting"
1173 		 */
1174 		VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to "
1175 		    "VM-entry intr info %#x", vector, info);
1176 		goto cantinject;
1177 	}
1178 
1179 	/* Inject the interrupt */
1180 	info = VMCS_INTR_T_HWINTR | VMCS_INTR_VALID;
1181 	info |= vector;
1182 	vmcs_write(VMCS_ENTRY_INTR_INFO, info);
1183 
1184 	/* Update the Local APIC ISR */
1185 	vlapic_intr_accepted(vlapic, vector);
1186 
1187 	VCPU_CTR1(vmx->vm, vcpu, "Injecting hwintr at vector %d", vector);
1188 
1189 	return;
1190 
1191 cantinject:
1192 	/*
1193 	 * Set the Interrupt Window Exiting execution control so we can inject
1194 	 * the interrupt as soon as blocking condition goes away.
1195 	 */
1196 	vmx_set_int_window_exiting(vmx, vcpu);
1197 }
1198 
1199 /*
1200  * If the Virtual NMIs execution control is '1' then the logical processor
1201  * tracks virtual-NMI blocking in the Guest Interruptibility-state field of
1202  * the VMCS. An IRET instruction in VMX non-root operation will remove any
1203  * virtual-NMI blocking.
1204  *
1205  * This unblocking occurs even if the IRET causes a fault. In this case the
1206  * hypervisor needs to restore virtual-NMI blocking before resuming the guest.
1207  */
1208 static void
1209 vmx_restore_nmi_blocking(struct vmx *vmx, int vcpuid)
1210 {
1211 	uint32_t gi;
1212 
1213 	VCPU_CTR0(vmx->vm, vcpuid, "Restore Virtual-NMI blocking");
1214 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1215 	gi |= VMCS_INTERRUPTIBILITY_NMI_BLOCKING;
1216 	vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi);
1217 }
1218 
1219 static void
1220 vmx_clear_nmi_blocking(struct vmx *vmx, int vcpuid)
1221 {
1222 	uint32_t gi;
1223 
1224 	VCPU_CTR0(vmx->vm, vcpuid, "Clear Virtual-NMI blocking");
1225 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1226 	gi &= ~VMCS_INTERRUPTIBILITY_NMI_BLOCKING;
1227 	vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi);
1228 }
1229 
1230 static int
1231 vmx_emulate_xsetbv(struct vmx *vmx, int vcpu)
1232 {
1233 	struct vmxctx *vmxctx;
1234 	uint64_t xcrval;
1235 	const struct xsave_limits *limits;
1236 
1237 	vmxctx = &vmx->ctx[vcpu];
1238 	limits = vmm_get_xsave_limits();
1239 
1240 	/* We only handle xcr0 if the host has XSAVE enabled. */
1241 	if (vmxctx->guest_rcx != 0 || !limits->xsave_enabled)
1242 		return (UNHANDLED);
1243 
1244 	xcrval = vmxctx->guest_rdx << 32 | (vmxctx->guest_rax & 0xffffffff);
1245 	if ((xcrval & ~limits->xcr0_allowed) != 0)
1246 		return (UNHANDLED);
1247 
1248 	if (!(xcrval & XFEATURE_ENABLED_X87))
1249 		return (UNHANDLED);
1250 
1251 	if ((xcrval & (XFEATURE_ENABLED_AVX | XFEATURE_ENABLED_SSE)) ==
1252 	    XFEATURE_ENABLED_AVX)
1253 		return (UNHANDLED);
1254 
1255 	/*
1256 	 * This runs "inside" vmrun() with the guest's FPU state, so
1257 	 * modifying xcr0 directly modifies the guest's xcr0, not the
1258 	 * host's.
1259 	 */
1260 	load_xcr(0, xcrval);
1261 	return (HANDLED);
1262 }
1263 
1264 static int
1265 vmx_emulate_cr_access(struct vmx *vmx, int vcpu, uint64_t exitqual)
1266 {
1267 	int cr, vmcs_guest_cr, vmcs_shadow_cr;
1268 	uint64_t crval, regval, ones_mask, zeros_mask;
1269 	const struct vmxctx *vmxctx;
1270 
1271 	/* We only handle mov to %cr0 or %cr4 at this time */
1272 	if ((exitqual & 0xf0) != 0x00)
1273 		return (UNHANDLED);
1274 
1275 	cr = exitqual & 0xf;
1276 	if (cr != 0 && cr != 4)
1277 		return (UNHANDLED);
1278 
1279 	regval = 0; /* silence gcc */
1280 	vmxctx = &vmx->ctx[vcpu];
1281 
1282 	/*
1283 	 * We must use vmcs_write() directly here because vmcs_setreg() will
1284 	 * call vmclear(vmcs) as a side-effect which we certainly don't want.
1285 	 */
1286 	switch ((exitqual >> 8) & 0xf) {
1287 	case 0:
1288 		regval = vmxctx->guest_rax;
1289 		break;
1290 	case 1:
1291 		regval = vmxctx->guest_rcx;
1292 		break;
1293 	case 2:
1294 		regval = vmxctx->guest_rdx;
1295 		break;
1296 	case 3:
1297 		regval = vmxctx->guest_rbx;
1298 		break;
1299 	case 4:
1300 		regval = vmcs_read(VMCS_GUEST_RSP);
1301 		break;
1302 	case 5:
1303 		regval = vmxctx->guest_rbp;
1304 		break;
1305 	case 6:
1306 		regval = vmxctx->guest_rsi;
1307 		break;
1308 	case 7:
1309 		regval = vmxctx->guest_rdi;
1310 		break;
1311 	case 8:
1312 		regval = vmxctx->guest_r8;
1313 		break;
1314 	case 9:
1315 		regval = vmxctx->guest_r9;
1316 		break;
1317 	case 10:
1318 		regval = vmxctx->guest_r10;
1319 		break;
1320 	case 11:
1321 		regval = vmxctx->guest_r11;
1322 		break;
1323 	case 12:
1324 		regval = vmxctx->guest_r12;
1325 		break;
1326 	case 13:
1327 		regval = vmxctx->guest_r13;
1328 		break;
1329 	case 14:
1330 		regval = vmxctx->guest_r14;
1331 		break;
1332 	case 15:
1333 		regval = vmxctx->guest_r15;
1334 		break;
1335 	}
1336 
1337 	if (cr == 0) {
1338 		ones_mask = cr0_ones_mask;
1339 		zeros_mask = cr0_zeros_mask;
1340 		vmcs_guest_cr = VMCS_GUEST_CR0;
1341 		vmcs_shadow_cr = VMCS_CR0_SHADOW;
1342 	} else {
1343 		ones_mask = cr4_ones_mask;
1344 		zeros_mask = cr4_zeros_mask;
1345 		vmcs_guest_cr = VMCS_GUEST_CR4;
1346 		vmcs_shadow_cr = VMCS_CR4_SHADOW;
1347 	}
1348 	vmcs_write(vmcs_shadow_cr, regval);
1349 
1350 	crval = regval | ones_mask;
1351 	crval &= ~zeros_mask;
1352 	vmcs_write(vmcs_guest_cr, crval);
1353 
1354 	if (cr == 0 && regval & CR0_PG) {
1355 		uint64_t efer, entry_ctls;
1356 
1357 		/*
1358 		 * If CR0.PG is 1 and EFER.LME is 1 then EFER.LMA and
1359 		 * the "IA-32e mode guest" bit in VM-entry control must be
1360 		 * equal.
1361 		 */
1362 		efer = vmcs_read(VMCS_GUEST_IA32_EFER);
1363 		if (efer & EFER_LME) {
1364 			efer |= EFER_LMA;
1365 			vmcs_write(VMCS_GUEST_IA32_EFER, efer);
1366 			entry_ctls = vmcs_read(VMCS_ENTRY_CTLS);
1367 			entry_ctls |= VM_ENTRY_GUEST_LMA;
1368 			vmcs_write(VMCS_ENTRY_CTLS, entry_ctls);
1369 		}
1370 	}
1371 
1372 	return (HANDLED);
1373 }
1374 
1375 static enum vie_cpu_mode
1376 vmx_cpu_mode(void)
1377 {
1378 
1379 	if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LMA)
1380 		return (CPU_MODE_64BIT);
1381 	else
1382 		return (CPU_MODE_COMPATIBILITY);
1383 }
1384 
1385 static enum vie_paging_mode
1386 vmx_paging_mode(void)
1387 {
1388 
1389 	if (!(vmcs_read(VMCS_GUEST_CR0) & CR0_PG))
1390 		return (PAGING_MODE_FLAT);
1391 	if (!(vmcs_read(VMCS_GUEST_CR4) & CR4_PAE))
1392 		return (PAGING_MODE_32);
1393 	if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LME)
1394 		return (PAGING_MODE_64);
1395 	else
1396 		return (PAGING_MODE_PAE);
1397 }
1398 
1399 static int
1400 ept_fault_type(uint64_t ept_qual)
1401 {
1402 	int fault_type;
1403 
1404 	if (ept_qual & EPT_VIOLATION_DATA_WRITE)
1405 		fault_type = VM_PROT_WRITE;
1406 	else if (ept_qual & EPT_VIOLATION_INST_FETCH)
1407 		fault_type = VM_PROT_EXECUTE;
1408 	else
1409 		fault_type= VM_PROT_READ;
1410 
1411 	return (fault_type);
1412 }
1413 
1414 static boolean_t
1415 ept_emulation_fault(uint64_t ept_qual)
1416 {
1417 	int read, write;
1418 
1419 	/* EPT fault on an instruction fetch doesn't make sense here */
1420 	if (ept_qual & EPT_VIOLATION_INST_FETCH)
1421 		return (FALSE);
1422 
1423 	/* EPT fault must be a read fault or a write fault */
1424 	read = ept_qual & EPT_VIOLATION_DATA_READ ? 1 : 0;
1425 	write = ept_qual & EPT_VIOLATION_DATA_WRITE ? 1 : 0;
1426 	if ((read | write) == 0)
1427 		return (FALSE);
1428 
1429 	/*
1430 	 * The EPT violation must have been caused by accessing a
1431 	 * guest-physical address that is a translation of a guest-linear
1432 	 * address.
1433 	 */
1434 	if ((ept_qual & EPT_VIOLATION_GLA_VALID) == 0 ||
1435 	    (ept_qual & EPT_VIOLATION_XLAT_VALID) == 0) {
1436 		return (FALSE);
1437 	}
1438 
1439 	return (TRUE);
1440 }
1441 
1442 static int
1443 vmx_handle_apic_write(struct vlapic *vlapic, uint64_t qual)
1444 {
1445 	int error, handled, offset;
1446 	bool retu;
1447 
1448 	if (!virtual_interrupt_delivery)
1449 		return (UNHANDLED);
1450 
1451 	handled = 1;
1452 	offset = APIC_WRITE_OFFSET(qual);
1453 	switch (offset) {
1454 	case APIC_OFFSET_ID:
1455 		vlapic_id_write_handler(vlapic);
1456 		break;
1457 	case APIC_OFFSET_LDR:
1458 		vlapic_ldr_write_handler(vlapic);
1459 		break;
1460 	case APIC_OFFSET_DFR:
1461 		vlapic_dfr_write_handler(vlapic);
1462 		break;
1463 	case APIC_OFFSET_SVR:
1464 		vlapic_svr_write_handler(vlapic);
1465 		break;
1466 	case APIC_OFFSET_ESR:
1467 		vlapic_esr_write_handler(vlapic);
1468 		break;
1469 	case APIC_OFFSET_ICR_LOW:
1470 		retu = false;
1471 		error = vlapic_icrlo_write_handler(vlapic, &retu);
1472 		if (error != 0 || retu)
1473 			handled = 0;
1474 		break;
1475 	case APIC_OFFSET_CMCI_LVT:
1476 	case APIC_OFFSET_TIMER_LVT ... APIC_OFFSET_ERROR_LVT:
1477 		vlapic_lvt_write_handler(vlapic, offset);
1478 		break;
1479 	case APIC_OFFSET_TIMER_ICR:
1480 		vlapic_icrtmr_write_handler(vlapic);
1481 		break;
1482 	case APIC_OFFSET_TIMER_DCR:
1483 		vlapic_dcr_write_handler(vlapic);
1484 		break;
1485 	default:
1486 		handled = 0;
1487 		break;
1488 	}
1489 	return (handled);
1490 }
1491 
1492 static bool
1493 apic_access_fault(uint64_t gpa)
1494 {
1495 
1496 	if (virtual_interrupt_delivery &&
1497 	    (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE))
1498 		return (true);
1499 	else
1500 		return (false);
1501 }
1502 
1503 static int
1504 vmx_handle_apic_access(struct vmx *vmx, int vcpuid, struct vm_exit *vmexit)
1505 {
1506 	uint64_t qual;
1507 	int access_type, offset, allowed;
1508 
1509 	if (!virtual_interrupt_delivery)
1510 		return (UNHANDLED);
1511 
1512 	qual = vmexit->u.vmx.exit_qualification;
1513 	access_type = APIC_ACCESS_TYPE(qual);
1514 	offset = APIC_ACCESS_OFFSET(qual);
1515 
1516 	allowed = 0;
1517 	if (access_type == 0) {
1518 		/*
1519 		 * Read data access to the following registers is expected.
1520 		 */
1521 		switch (offset) {
1522 		case APIC_OFFSET_APR:
1523 		case APIC_OFFSET_PPR:
1524 		case APIC_OFFSET_RRR:
1525 		case APIC_OFFSET_CMCI_LVT:
1526 		case APIC_OFFSET_TIMER_CCR:
1527 			allowed = 1;
1528 			break;
1529 		default:
1530 			break;
1531 		}
1532 	} else if (access_type == 1) {
1533 		/*
1534 		 * Write data access to the following registers is expected.
1535 		 */
1536 		switch (offset) {
1537 		case APIC_OFFSET_VER:
1538 		case APIC_OFFSET_APR:
1539 		case APIC_OFFSET_PPR:
1540 		case APIC_OFFSET_RRR:
1541 		case APIC_OFFSET_ISR0 ... APIC_OFFSET_ISR7:
1542 		case APIC_OFFSET_TMR0 ... APIC_OFFSET_TMR7:
1543 		case APIC_OFFSET_IRR0 ... APIC_OFFSET_IRR7:
1544 		case APIC_OFFSET_CMCI_LVT:
1545 		case APIC_OFFSET_TIMER_CCR:
1546 			allowed = 1;
1547 			break;
1548 		default:
1549 			break;
1550 		}
1551 	}
1552 
1553 	if (allowed) {
1554 		vmexit->exitcode = VM_EXITCODE_INST_EMUL;
1555 		vmexit->u.inst_emul.gpa = DEFAULT_APIC_BASE + offset;
1556 		vmexit->u.inst_emul.gla = VIE_INVALID_GLA;
1557 		vmexit->u.inst_emul.cr3 = vmcs_guest_cr3();
1558 		vmexit->u.inst_emul.cpu_mode = vmx_cpu_mode();
1559 		vmexit->u.inst_emul.paging_mode = vmx_paging_mode();
1560 	}
1561 
1562 	/*
1563 	 * Regardless of whether the APIC-access is allowed this handler
1564 	 * always returns UNHANDLED:
1565 	 * - if the access is allowed then it is handled by emulating the
1566 	 *   instruction that caused the VM-exit (outside the critical section)
1567 	 * - if the access is not allowed then it will be converted to an
1568 	 *   exitcode of VM_EXITCODE_VMX and will be dealt with in userland.
1569 	 */
1570 	return (UNHANDLED);
1571 }
1572 
1573 static int
1574 vmx_exit_process(struct vmx *vmx, int vcpu, struct vm_exit *vmexit)
1575 {
1576 	int error, handled;
1577 	struct vmxctx *vmxctx;
1578 	struct vlapic *vlapic;
1579 	uint32_t eax, ecx, edx, idtvec_info, idtvec_err, intr_info, reason;
1580 	uint64_t qual, gpa;
1581 	bool retu;
1582 
1583 	CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_VIRTUAL_NMI) != 0);
1584 	CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_NMI_EXITING) != 0);
1585 
1586 	handled = 0;
1587 	vmxctx = &vmx->ctx[vcpu];
1588 
1589 	qual = vmexit->u.vmx.exit_qualification;
1590 	reason = vmexit->u.vmx.exit_reason;
1591 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1592 
1593 	vmm_stat_incr(vmx->vm, vcpu, VMEXIT_COUNT, 1);
1594 
1595 	/*
1596 	 * VM exits that could be triggered during event injection on the
1597 	 * previous VM entry need to be handled specially by re-injecting
1598 	 * the event.
1599 	 *
1600 	 * See "Information for VM Exits During Event Delivery" in Intel SDM
1601 	 * for details.
1602 	 */
1603 	switch (reason) {
1604 	case EXIT_REASON_EPT_FAULT:
1605 	case EXIT_REASON_EPT_MISCONFIG:
1606 	case EXIT_REASON_APIC_ACCESS:
1607 	case EXIT_REASON_TASK_SWITCH:
1608 	case EXIT_REASON_EXCEPTION:
1609 		idtvec_info = vmcs_idt_vectoring_info();
1610 		if (idtvec_info & VMCS_IDT_VEC_VALID) {
1611 			idtvec_info &= ~(1 << 12); /* clear undefined bit */
1612 			vmcs_write(VMCS_ENTRY_INTR_INFO, idtvec_info);
1613 			if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) {
1614 				idtvec_err = vmcs_idt_vectoring_err();
1615 				vmcs_write(VMCS_ENTRY_EXCEPTION_ERROR,
1616 				    idtvec_err);
1617 			}
1618 			/*
1619 			 * If 'virtual NMIs' are being used and the VM-exit
1620 			 * happened while injecting an NMI during the previous
1621 			 * VM-entry, then clear "blocking by NMI" in the Guest
1622 			 * Interruptibility-state.
1623 			 */
1624 			if ((idtvec_info & VMCS_INTR_T_MASK) ==
1625 			    VMCS_INTR_T_NMI) {
1626 				 vmx_clear_nmi_blocking(vmx, vcpu);
1627 			}
1628 			vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length);
1629 		}
1630 	default:
1631 		idtvec_info = 0;
1632 		break;
1633 	}
1634 
1635 	switch (reason) {
1636 	case EXIT_REASON_CR_ACCESS:
1637 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CR_ACCESS, 1);
1638 		handled = vmx_emulate_cr_access(vmx, vcpu, qual);
1639 		break;
1640 	case EXIT_REASON_RDMSR:
1641 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_RDMSR, 1);
1642 		retu = false;
1643 		ecx = vmxctx->guest_rcx;
1644 		error = emulate_rdmsr(vmx->vm, vcpu, ecx, &retu);
1645 		if (error) {
1646 			vmexit->exitcode = VM_EXITCODE_RDMSR;
1647 			vmexit->u.msr.code = ecx;
1648 		} else if (!retu) {
1649 			handled = 1;
1650 		} else {
1651 			/* Return to userspace with a valid exitcode */
1652 			KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS,
1653 			    ("emulate_wrmsr retu with bogus exitcode"));
1654 		}
1655 		break;
1656 	case EXIT_REASON_WRMSR:
1657 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_WRMSR, 1);
1658 		retu = false;
1659 		eax = vmxctx->guest_rax;
1660 		ecx = vmxctx->guest_rcx;
1661 		edx = vmxctx->guest_rdx;
1662 		error = emulate_wrmsr(vmx->vm, vcpu, ecx,
1663 		    (uint64_t)edx << 32 | eax, &retu);
1664 		if (error) {
1665 			vmexit->exitcode = VM_EXITCODE_WRMSR;
1666 			vmexit->u.msr.code = ecx;
1667 			vmexit->u.msr.wval = (uint64_t)edx << 32 | eax;
1668 		} else if (!retu) {
1669 			handled = 1;
1670 		} else {
1671 			/* Return to userspace with a valid exitcode */
1672 			KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS,
1673 			    ("emulate_wrmsr retu with bogus exitcode"));
1674 		}
1675 		break;
1676 	case EXIT_REASON_HLT:
1677 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_HLT, 1);
1678 		vmexit->exitcode = VM_EXITCODE_HLT;
1679 		vmexit->u.hlt.rflags = vmcs_read(VMCS_GUEST_RFLAGS);
1680 		break;
1681 	case EXIT_REASON_MTF:
1682 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_MTRAP, 1);
1683 		vmexit->exitcode = VM_EXITCODE_MTRAP;
1684 		break;
1685 	case EXIT_REASON_PAUSE:
1686 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_PAUSE, 1);
1687 		vmexit->exitcode = VM_EXITCODE_PAUSE;
1688 		break;
1689 	case EXIT_REASON_INTR_WINDOW:
1690 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INTR_WINDOW, 1);
1691 		vmx_clear_int_window_exiting(vmx, vcpu);
1692 		return (1);
1693 	case EXIT_REASON_EXT_INTR:
1694 		/*
1695 		 * External interrupts serve only to cause VM exits and allow
1696 		 * the host interrupt handler to run.
1697 		 *
1698 		 * If this external interrupt triggers a virtual interrupt
1699 		 * to a VM, then that state will be recorded by the
1700 		 * host interrupt handler in the VM's softc. We will inject
1701 		 * this virtual interrupt during the subsequent VM enter.
1702 		 */
1703 		intr_info = vmcs_read(VMCS_EXIT_INTR_INFO);
1704 		KASSERT((intr_info & VMCS_INTR_VALID) != 0 &&
1705 		    (intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_HWINTR,
1706 		    ("VM exit interruption info invalid: %#x", intr_info));
1707 		vmx_trigger_hostintr(intr_info & 0xff);
1708 
1709 		/*
1710 		 * This is special. We want to treat this as an 'handled'
1711 		 * VM-exit but not increment the instruction pointer.
1712 		 */
1713 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EXTINT, 1);
1714 		return (1);
1715 	case EXIT_REASON_NMI_WINDOW:
1716 		/* Exit to allow the pending virtual NMI to be injected */
1717 		if (vm_nmi_pending(vmx->vm, vcpu))
1718 			vmx_inject_nmi(vmx, vcpu);
1719 		vmx_clear_nmi_window_exiting(vmx, vcpu);
1720 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_NMI_WINDOW, 1);
1721 		return (1);
1722 	case EXIT_REASON_INOUT:
1723 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INOUT, 1);
1724 		vmexit->exitcode = VM_EXITCODE_INOUT;
1725 		vmexit->u.inout.bytes = (qual & 0x7) + 1;
1726 		vmexit->u.inout.in = (qual & 0x8) ? 1 : 0;
1727 		vmexit->u.inout.string = (qual & 0x10) ? 1 : 0;
1728 		vmexit->u.inout.rep = (qual & 0x20) ? 1 : 0;
1729 		vmexit->u.inout.port = (uint16_t)(qual >> 16);
1730 		vmexit->u.inout.eax = (uint32_t)(vmxctx->guest_rax);
1731 		break;
1732 	case EXIT_REASON_CPUID:
1733 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CPUID, 1);
1734 		handled = vmx_handle_cpuid(vmx->vm, vcpu, vmxctx);
1735 		break;
1736 	case EXIT_REASON_EXCEPTION:
1737 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EXCEPTION, 1);
1738 		intr_info = vmcs_read(VMCS_EXIT_INTR_INFO);
1739 		KASSERT((intr_info & VMCS_INTR_VALID) != 0,
1740 		    ("VM exit interruption info invalid: %#x", intr_info));
1741 
1742 		/*
1743 		 * If Virtual NMIs control is 1 and the VM-exit is due to a
1744 		 * fault encountered during the execution of IRET then we must
1745 		 * restore the state of "virtual-NMI blocking" before resuming
1746 		 * the guest.
1747 		 *
1748 		 * See "Resuming Guest Software after Handling an Exception".
1749 		 */
1750 		if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 &&
1751 		    (intr_info & 0xff) != IDT_DF &&
1752 		    (intr_info & EXIT_QUAL_NMIUDTI) != 0)
1753 			vmx_restore_nmi_blocking(vmx, vcpu);
1754 
1755 		/*
1756 		 * The NMI has already been handled in vmx_exit_handle_nmi().
1757 		 */
1758 		if ((intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_NMI)
1759 			return (1);
1760 		break;
1761 	case EXIT_REASON_EPT_FAULT:
1762 		/*
1763 		 * If 'gpa' lies within the address space allocated to
1764 		 * memory then this must be a nested page fault otherwise
1765 		 * this must be an instruction that accesses MMIO space.
1766 		 */
1767 		gpa = vmcs_gpa();
1768 		if (vm_mem_allocated(vmx->vm, gpa) || apic_access_fault(gpa)) {
1769 			vmexit->exitcode = VM_EXITCODE_PAGING;
1770 			vmexit->u.paging.gpa = gpa;
1771 			vmexit->u.paging.fault_type = ept_fault_type(qual);
1772 			vmm_stat_incr(vmx->vm, vcpu, VMEXIT_NESTED_FAULT, 1);
1773 		} else if (ept_emulation_fault(qual)) {
1774 			vmexit->exitcode = VM_EXITCODE_INST_EMUL;
1775 			vmexit->u.inst_emul.gpa = gpa;
1776 			vmexit->u.inst_emul.gla = vmcs_gla();
1777 			vmexit->u.inst_emul.cr3 = vmcs_guest_cr3();
1778 			vmexit->u.inst_emul.cpu_mode = vmx_cpu_mode();
1779 			vmexit->u.inst_emul.paging_mode = vmx_paging_mode();
1780 			vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INST_EMUL, 1);
1781 		}
1782 		/*
1783 		 * If Virtual NMIs control is 1 and the VM-exit is due to an
1784 		 * EPT fault during the execution of IRET then we must restore
1785 		 * the state of "virtual-NMI blocking" before resuming.
1786 		 *
1787 		 * See description of "NMI unblocking due to IRET" in
1788 		 * "Exit Qualification for EPT Violations".
1789 		 */
1790 		if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 &&
1791 		    (qual & EXIT_QUAL_NMIUDTI) != 0)
1792 			vmx_restore_nmi_blocking(vmx, vcpu);
1793 		break;
1794 	case EXIT_REASON_VIRTUALIZED_EOI:
1795 		vmexit->exitcode = VM_EXITCODE_IOAPIC_EOI;
1796 		vmexit->u.ioapic_eoi.vector = qual & 0xFF;
1797 		vmexit->inst_length = 0;	/* trap-like */
1798 		break;
1799 	case EXIT_REASON_APIC_ACCESS:
1800 		handled = vmx_handle_apic_access(vmx, vcpu, vmexit);
1801 		break;
1802 	case EXIT_REASON_APIC_WRITE:
1803 		/*
1804 		 * APIC-write VM exit is trap-like so the %rip is already
1805 		 * pointing to the next instruction.
1806 		 */
1807 		vmexit->inst_length = 0;
1808 		vlapic = vm_lapic(vmx->vm, vcpu);
1809 		handled = vmx_handle_apic_write(vlapic, qual);
1810 		break;
1811 	case EXIT_REASON_XSETBV:
1812 		handled = vmx_emulate_xsetbv(vmx, vcpu);
1813 		break;
1814 	default:
1815 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_UNKNOWN, 1);
1816 		break;
1817 	}
1818 
1819 	if (handled) {
1820 		/*
1821 		 * It is possible that control is returned to userland
1822 		 * even though we were able to handle the VM exit in the
1823 		 * kernel.
1824 		 *
1825 		 * In such a case we want to make sure that the userland
1826 		 * restarts guest execution at the instruction *after*
1827 		 * the one we just processed. Therefore we update the
1828 		 * guest rip in the VMCS and in 'vmexit'.
1829 		 */
1830 		vmexit->rip += vmexit->inst_length;
1831 		vmexit->inst_length = 0;
1832 		vmcs_write(VMCS_GUEST_RIP, vmexit->rip);
1833 	} else {
1834 		if (vmexit->exitcode == VM_EXITCODE_BOGUS) {
1835 			/*
1836 			 * If this VM exit was not claimed by anybody then
1837 			 * treat it as a generic VMX exit.
1838 			 */
1839 			vmexit->exitcode = VM_EXITCODE_VMX;
1840 			vmexit->u.vmx.status = VM_SUCCESS;
1841 			vmexit->u.vmx.inst_type = 0;
1842 			vmexit->u.vmx.inst_error = 0;
1843 		} else {
1844 			/*
1845 			 * The exitcode and collateral have been populated.
1846 			 * The VM exit will be processed further in userland.
1847 			 */
1848 		}
1849 	}
1850 	return (handled);
1851 }
1852 
1853 static __inline int
1854 vmx_exit_astpending(struct vmx *vmx, int vcpu, struct vm_exit *vmexit)
1855 {
1856 
1857 	vmexit->rip = vmcs_guest_rip();
1858 	vmexit->inst_length = 0;
1859 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1860 	vmx_astpending_trace(vmx, vcpu, vmexit->rip);
1861 	vmm_stat_incr(vmx->vm, vcpu, VMEXIT_ASTPENDING, 1);
1862 
1863 	return (HANDLED);
1864 }
1865 
1866 static __inline int
1867 vmx_exit_rendezvous(struct vmx *vmx, int vcpu, struct vm_exit *vmexit)
1868 {
1869 
1870 	vmexit->rip = vmcs_guest_rip();
1871 	vmexit->inst_length = 0;
1872 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1873 	vmm_stat_incr(vmx->vm, vcpu, VMEXIT_RENDEZVOUS, 1);
1874 
1875 	return (UNHANDLED);
1876 }
1877 
1878 static __inline int
1879 vmx_exit_inst_error(struct vmxctx *vmxctx, int rc, struct vm_exit *vmexit)
1880 {
1881 
1882 	KASSERT(vmxctx->inst_fail_status != VM_SUCCESS,
1883 	    ("vmx_exit_inst_error: invalid inst_fail_status %d",
1884 	    vmxctx->inst_fail_status));
1885 
1886 	vmexit->inst_length = 0;
1887 	vmexit->exitcode = VM_EXITCODE_VMX;
1888 	vmexit->u.vmx.status = vmxctx->inst_fail_status;
1889 	vmexit->u.vmx.inst_error = vmcs_instruction_error();
1890 	vmexit->u.vmx.exit_reason = ~0;
1891 	vmexit->u.vmx.exit_qualification = ~0;
1892 
1893 	switch (rc) {
1894 	case VMX_VMRESUME_ERROR:
1895 	case VMX_VMLAUNCH_ERROR:
1896 	case VMX_INVEPT_ERROR:
1897 		vmexit->u.vmx.inst_type = rc;
1898 		break;
1899 	default:
1900 		panic("vm_exit_inst_error: vmx_enter_guest returned %d", rc);
1901 	}
1902 
1903 	return (UNHANDLED);
1904 }
1905 
1906 /*
1907  * If the NMI-exiting VM execution control is set to '1' then an NMI in
1908  * non-root operation causes a VM-exit. NMI blocking is in effect so it is
1909  * sufficient to simply vector to the NMI handler via a software interrupt.
1910  * However, this must be done before maskable interrupts are enabled
1911  * otherwise the "iret" issued by an interrupt handler will incorrectly
1912  * clear NMI blocking.
1913  */
1914 static __inline void
1915 vmx_exit_handle_nmi(struct vmx *vmx, int vcpuid, struct vm_exit *vmexit)
1916 {
1917 	uint32_t intr_info;
1918 
1919 	KASSERT((read_rflags() & PSL_I) == 0, ("interrupts enabled"));
1920 
1921 	if (vmexit->u.vmx.exit_reason != EXIT_REASON_EXCEPTION)
1922 		return;
1923 
1924 	intr_info = vmcs_read(VMCS_EXIT_INTR_INFO);
1925 	KASSERT((intr_info & VMCS_INTR_VALID) != 0,
1926 	    ("VM exit interruption info invalid: %#x", intr_info));
1927 
1928 	if ((intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_NMI) {
1929 		KASSERT((intr_info & 0xff) == IDT_NMI, ("VM exit due "
1930 		    "to NMI has invalid vector: %#x", intr_info));
1931 		VCPU_CTR0(vmx->vm, vcpuid, "Vectoring to NMI handler");
1932 		__asm __volatile("int $2");
1933 	}
1934 }
1935 
1936 static int
1937 vmx_run(void *arg, int vcpu, register_t startrip, pmap_t pmap,
1938     void *rendezvous_cookie)
1939 {
1940 	int rc, handled, launched;
1941 	struct vmx *vmx;
1942 	struct vm *vm;
1943 	struct vmxctx *vmxctx;
1944 	struct vmcs *vmcs;
1945 	struct vm_exit *vmexit;
1946 	struct vlapic *vlapic;
1947 	uint64_t rip;
1948 	uint32_t exit_reason;
1949 
1950 	vmx = arg;
1951 	vm = vmx->vm;
1952 	vmcs = &vmx->vmcs[vcpu];
1953 	vmxctx = &vmx->ctx[vcpu];
1954 	vlapic = vm_lapic(vm, vcpu);
1955 	vmexit = vm_exitinfo(vm, vcpu);
1956 	launched = 0;
1957 
1958 	KASSERT(vmxctx->pmap == pmap,
1959 	    ("pmap %p different than ctx pmap %p", pmap, vmxctx->pmap));
1960 
1961 	VMPTRLD(vmcs);
1962 
1963 	/*
1964 	 * XXX
1965 	 * We do this every time because we may setup the virtual machine
1966 	 * from a different process than the one that actually runs it.
1967 	 *
1968 	 * If the life of a virtual machine was spent entirely in the context
1969 	 * of a single process we could do this once in vmx_vminit().
1970 	 */
1971 	vmcs_write(VMCS_HOST_CR3, rcr3());
1972 
1973 	vmcs_write(VMCS_GUEST_RIP, startrip);
1974 	vmx_set_pcpu_defaults(vmx, vcpu, pmap);
1975 	do {
1976 		/*
1977 		 * Interrupts are disabled from this point on until the
1978 		 * guest starts executing. This is done for the following
1979 		 * reasons:
1980 		 *
1981 		 * If an AST is asserted on this thread after the check below,
1982 		 * then the IPI_AST notification will not be lost, because it
1983 		 * will cause a VM exit due to external interrupt as soon as
1984 		 * the guest state is loaded.
1985 		 *
1986 		 * A posted interrupt after 'vmx_inject_interrupts()' will
1987 		 * not be "lost" because it will be held pending in the host
1988 		 * APIC because interrupts are disabled. The pending interrupt
1989 		 * will be recognized as soon as the guest state is loaded.
1990 		 *
1991 		 * The same reasoning applies to the IPI generated by
1992 		 * pmap_invalidate_ept().
1993 		 */
1994 		disable_intr();
1995 		if (curthread->td_flags & (TDF_ASTPENDING | TDF_NEEDRESCHED)) {
1996 			enable_intr();
1997 			handled = vmx_exit_astpending(vmx, vcpu, vmexit);
1998 			break;
1999 		}
2000 
2001 		if (vcpu_rendezvous_pending(rendezvous_cookie)) {
2002 			enable_intr();
2003 			handled = vmx_exit_rendezvous(vmx, vcpu, vmexit);
2004 			break;
2005 		}
2006 
2007 		vmx_inject_interrupts(vmx, vcpu, vlapic);
2008 		vmx_run_trace(vmx, vcpu);
2009 		rc = vmx_enter_guest(vmxctx, vmx, launched);
2010 
2011 		/* Collect some information for VM exit processing */
2012 		vmexit->rip = rip = vmcs_guest_rip();
2013 		vmexit->inst_length = vmexit_instruction_length();
2014 		vmexit->u.vmx.exit_reason = exit_reason = vmcs_exit_reason();
2015 		vmexit->u.vmx.exit_qualification = vmcs_exit_qualification();
2016 
2017 		if (rc == VMX_GUEST_VMEXIT) {
2018 			vmx_exit_handle_nmi(vmx, vcpu, vmexit);
2019 			enable_intr();
2020 			handled = vmx_exit_process(vmx, vcpu, vmexit);
2021 		} else {
2022 			enable_intr();
2023 			handled = vmx_exit_inst_error(vmxctx, rc, vmexit);
2024 		}
2025 		launched = 1;
2026 		vmx_exit_trace(vmx, vcpu, rip, exit_reason, handled);
2027 	} while (handled);
2028 
2029 	/*
2030 	 * If a VM exit has been handled then the exitcode must be BOGUS
2031 	 * If a VM exit is not handled then the exitcode must not be BOGUS
2032 	 */
2033 	if ((handled && vmexit->exitcode != VM_EXITCODE_BOGUS) ||
2034 	    (!handled && vmexit->exitcode == VM_EXITCODE_BOGUS)) {
2035 		panic("Mismatch between handled (%d) and exitcode (%d)",
2036 		      handled, vmexit->exitcode);
2037 	}
2038 
2039 	if (!handled)
2040 		vmm_stat_incr(vm, vcpu, VMEXIT_USERSPACE, 1);
2041 
2042 	VCPU_CTR1(vm, vcpu, "returning from vmx_run: exitcode %d",
2043 	    vmexit->exitcode);
2044 
2045 	VMCLEAR(vmcs);
2046 	return (0);
2047 }
2048 
2049 static void
2050 vmx_vmcleanup(void *arg)
2051 {
2052 	int i, error;
2053 	struct vmx *vmx = arg;
2054 
2055 	if (virtual_interrupt_delivery)
2056 		vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE);
2057 
2058 	for (i = 0; i < VM_MAXCPU; i++)
2059 		vpid_free(vmx->state[i].vpid);
2060 
2061 	/*
2062 	 * XXXSMP we also need to clear the VMCS active on the other vcpus.
2063 	 */
2064 	error = vmclear(&vmx->vmcs[0]);
2065 	if (error != 0)
2066 		panic("vmx_vmcleanup: vmclear error %d on vcpu 0", error);
2067 
2068 	free(vmx, M_VMX);
2069 
2070 	return;
2071 }
2072 
2073 static register_t *
2074 vmxctx_regptr(struct vmxctx *vmxctx, int reg)
2075 {
2076 
2077 	switch (reg) {
2078 	case VM_REG_GUEST_RAX:
2079 		return (&vmxctx->guest_rax);
2080 	case VM_REG_GUEST_RBX:
2081 		return (&vmxctx->guest_rbx);
2082 	case VM_REG_GUEST_RCX:
2083 		return (&vmxctx->guest_rcx);
2084 	case VM_REG_GUEST_RDX:
2085 		return (&vmxctx->guest_rdx);
2086 	case VM_REG_GUEST_RSI:
2087 		return (&vmxctx->guest_rsi);
2088 	case VM_REG_GUEST_RDI:
2089 		return (&vmxctx->guest_rdi);
2090 	case VM_REG_GUEST_RBP:
2091 		return (&vmxctx->guest_rbp);
2092 	case VM_REG_GUEST_R8:
2093 		return (&vmxctx->guest_r8);
2094 	case VM_REG_GUEST_R9:
2095 		return (&vmxctx->guest_r9);
2096 	case VM_REG_GUEST_R10:
2097 		return (&vmxctx->guest_r10);
2098 	case VM_REG_GUEST_R11:
2099 		return (&vmxctx->guest_r11);
2100 	case VM_REG_GUEST_R12:
2101 		return (&vmxctx->guest_r12);
2102 	case VM_REG_GUEST_R13:
2103 		return (&vmxctx->guest_r13);
2104 	case VM_REG_GUEST_R14:
2105 		return (&vmxctx->guest_r14);
2106 	case VM_REG_GUEST_R15:
2107 		return (&vmxctx->guest_r15);
2108 	default:
2109 		break;
2110 	}
2111 	return (NULL);
2112 }
2113 
2114 static int
2115 vmxctx_getreg(struct vmxctx *vmxctx, int reg, uint64_t *retval)
2116 {
2117 	register_t *regp;
2118 
2119 	if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) {
2120 		*retval = *regp;
2121 		return (0);
2122 	} else
2123 		return (EINVAL);
2124 }
2125 
2126 static int
2127 vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val)
2128 {
2129 	register_t *regp;
2130 
2131 	if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) {
2132 		*regp = val;
2133 		return (0);
2134 	} else
2135 		return (EINVAL);
2136 }
2137 
2138 static int
2139 vmx_shadow_reg(int reg)
2140 {
2141 	int shreg;
2142 
2143 	shreg = -1;
2144 
2145 	switch (reg) {
2146 	case VM_REG_GUEST_CR0:
2147 		shreg = VMCS_CR0_SHADOW;
2148                 break;
2149         case VM_REG_GUEST_CR4:
2150 		shreg = VMCS_CR4_SHADOW;
2151 		break;
2152 	default:
2153 		break;
2154 	}
2155 
2156 	return (shreg);
2157 }
2158 
2159 static int
2160 vmx_getreg(void *arg, int vcpu, int reg, uint64_t *retval)
2161 {
2162 	int running, hostcpu;
2163 	struct vmx *vmx = arg;
2164 
2165 	running = vcpu_is_running(vmx->vm, vcpu, &hostcpu);
2166 	if (running && hostcpu != curcpu)
2167 		panic("vmx_getreg: %s%d is running", vm_name(vmx->vm), vcpu);
2168 
2169 	if (vmxctx_getreg(&vmx->ctx[vcpu], reg, retval) == 0)
2170 		return (0);
2171 
2172 	return (vmcs_getreg(&vmx->vmcs[vcpu], running, reg, retval));
2173 }
2174 
2175 static int
2176 vmx_setreg(void *arg, int vcpu, int reg, uint64_t val)
2177 {
2178 	int error, hostcpu, running, shadow;
2179 	uint64_t ctls;
2180 	struct vmx *vmx = arg;
2181 
2182 	running = vcpu_is_running(vmx->vm, vcpu, &hostcpu);
2183 	if (running && hostcpu != curcpu)
2184 		panic("vmx_setreg: %s%d is running", vm_name(vmx->vm), vcpu);
2185 
2186 	if (vmxctx_setreg(&vmx->ctx[vcpu], reg, val) == 0)
2187 		return (0);
2188 
2189 	error = vmcs_setreg(&vmx->vmcs[vcpu], running, reg, val);
2190 
2191 	if (error == 0) {
2192 		/*
2193 		 * If the "load EFER" VM-entry control is 1 then the
2194 		 * value of EFER.LMA must be identical to "IA-32e mode guest"
2195 		 * bit in the VM-entry control.
2196 		 */
2197 		if ((entry_ctls & VM_ENTRY_LOAD_EFER) != 0 &&
2198 		    (reg == VM_REG_GUEST_EFER)) {
2199 			vmcs_getreg(&vmx->vmcs[vcpu], running,
2200 				    VMCS_IDENT(VMCS_ENTRY_CTLS), &ctls);
2201 			if (val & EFER_LMA)
2202 				ctls |= VM_ENTRY_GUEST_LMA;
2203 			else
2204 				ctls &= ~VM_ENTRY_GUEST_LMA;
2205 			vmcs_setreg(&vmx->vmcs[vcpu], running,
2206 				    VMCS_IDENT(VMCS_ENTRY_CTLS), ctls);
2207 		}
2208 
2209 		shadow = vmx_shadow_reg(reg);
2210 		if (shadow > 0) {
2211 			/*
2212 			 * Store the unmodified value in the shadow
2213 			 */
2214 			error = vmcs_setreg(&vmx->vmcs[vcpu], running,
2215 				    VMCS_IDENT(shadow), val);
2216 		}
2217 	}
2218 
2219 	return (error);
2220 }
2221 
2222 static int
2223 vmx_getdesc(void *arg, int vcpu, int reg, struct seg_desc *desc)
2224 {
2225 	struct vmx *vmx = arg;
2226 
2227 	return (vmcs_getdesc(&vmx->vmcs[vcpu], reg, desc));
2228 }
2229 
2230 static int
2231 vmx_setdesc(void *arg, int vcpu, int reg, struct seg_desc *desc)
2232 {
2233 	struct vmx *vmx = arg;
2234 
2235 	return (vmcs_setdesc(&vmx->vmcs[vcpu], reg, desc));
2236 }
2237 
2238 static int
2239 vmx_inject(void *arg, int vcpu, int type, int vector, uint32_t code,
2240 	   int code_valid)
2241 {
2242 	int error;
2243 	uint64_t info;
2244 	struct vmx *vmx = arg;
2245 	struct vmcs *vmcs = &vmx->vmcs[vcpu];
2246 
2247 	static uint32_t type_map[VM_EVENT_MAX] = {
2248 		0x1,		/* VM_EVENT_NONE */
2249 		0x0,		/* VM_HW_INTR */
2250 		0x2,		/* VM_NMI */
2251 		0x3,		/* VM_HW_EXCEPTION */
2252 		0x4,		/* VM_SW_INTR */
2253 		0x5,		/* VM_PRIV_SW_EXCEPTION */
2254 		0x6,		/* VM_SW_EXCEPTION */
2255 	};
2256 
2257 	/*
2258 	 * If there is already an exception pending to be delivered to the
2259 	 * vcpu then just return.
2260 	 */
2261 	error = vmcs_getreg(vmcs, 0, VMCS_IDENT(VMCS_ENTRY_INTR_INFO), &info);
2262 	if (error)
2263 		return (error);
2264 
2265 	if (info & VMCS_INTR_VALID)
2266 		return (EAGAIN);
2267 
2268 	info = vector | (type_map[type] << 8) | (code_valid ? 1 << 11 : 0);
2269 	info |= VMCS_INTR_VALID;
2270 	error = vmcs_setreg(vmcs, 0, VMCS_IDENT(VMCS_ENTRY_INTR_INFO), info);
2271 	if (error != 0)
2272 		return (error);
2273 
2274 	if (code_valid) {
2275 		error = vmcs_setreg(vmcs, 0,
2276 				    VMCS_IDENT(VMCS_ENTRY_EXCEPTION_ERROR),
2277 				    code);
2278 	}
2279 	return (error);
2280 }
2281 
2282 static int
2283 vmx_getcap(void *arg, int vcpu, int type, int *retval)
2284 {
2285 	struct vmx *vmx = arg;
2286 	int vcap;
2287 	int ret;
2288 
2289 	ret = ENOENT;
2290 
2291 	vcap = vmx->cap[vcpu].set;
2292 
2293 	switch (type) {
2294 	case VM_CAP_HALT_EXIT:
2295 		if (cap_halt_exit)
2296 			ret = 0;
2297 		break;
2298 	case VM_CAP_PAUSE_EXIT:
2299 		if (cap_pause_exit)
2300 			ret = 0;
2301 		break;
2302 	case VM_CAP_MTRAP_EXIT:
2303 		if (cap_monitor_trap)
2304 			ret = 0;
2305 		break;
2306 	case VM_CAP_UNRESTRICTED_GUEST:
2307 		if (cap_unrestricted_guest)
2308 			ret = 0;
2309 		break;
2310 	case VM_CAP_ENABLE_INVPCID:
2311 		if (cap_invpcid)
2312 			ret = 0;
2313 		break;
2314 	default:
2315 		break;
2316 	}
2317 
2318 	if (ret == 0)
2319 		*retval = (vcap & (1 << type)) ? 1 : 0;
2320 
2321 	return (ret);
2322 }
2323 
2324 static int
2325 vmx_setcap(void *arg, int vcpu, int type, int val)
2326 {
2327 	struct vmx *vmx = arg;
2328 	struct vmcs *vmcs = &vmx->vmcs[vcpu];
2329 	uint32_t baseval;
2330 	uint32_t *pptr;
2331 	int error;
2332 	int flag;
2333 	int reg;
2334 	int retval;
2335 
2336 	retval = ENOENT;
2337 	pptr = NULL;
2338 
2339 	switch (type) {
2340 	case VM_CAP_HALT_EXIT:
2341 		if (cap_halt_exit) {
2342 			retval = 0;
2343 			pptr = &vmx->cap[vcpu].proc_ctls;
2344 			baseval = *pptr;
2345 			flag = PROCBASED_HLT_EXITING;
2346 			reg = VMCS_PRI_PROC_BASED_CTLS;
2347 		}
2348 		break;
2349 	case VM_CAP_MTRAP_EXIT:
2350 		if (cap_monitor_trap) {
2351 			retval = 0;
2352 			pptr = &vmx->cap[vcpu].proc_ctls;
2353 			baseval = *pptr;
2354 			flag = PROCBASED_MTF;
2355 			reg = VMCS_PRI_PROC_BASED_CTLS;
2356 		}
2357 		break;
2358 	case VM_CAP_PAUSE_EXIT:
2359 		if (cap_pause_exit) {
2360 			retval = 0;
2361 			pptr = &vmx->cap[vcpu].proc_ctls;
2362 			baseval = *pptr;
2363 			flag = PROCBASED_PAUSE_EXITING;
2364 			reg = VMCS_PRI_PROC_BASED_CTLS;
2365 		}
2366 		break;
2367 	case VM_CAP_UNRESTRICTED_GUEST:
2368 		if (cap_unrestricted_guest) {
2369 			retval = 0;
2370 			pptr = &vmx->cap[vcpu].proc_ctls2;
2371 			baseval = *pptr;
2372 			flag = PROCBASED2_UNRESTRICTED_GUEST;
2373 			reg = VMCS_SEC_PROC_BASED_CTLS;
2374 		}
2375 		break;
2376 	case VM_CAP_ENABLE_INVPCID:
2377 		if (cap_invpcid) {
2378 			retval = 0;
2379 			pptr = &vmx->cap[vcpu].proc_ctls2;
2380 			baseval = *pptr;
2381 			flag = PROCBASED2_ENABLE_INVPCID;
2382 			reg = VMCS_SEC_PROC_BASED_CTLS;
2383 		}
2384 		break;
2385 	default:
2386 		break;
2387 	}
2388 
2389 	if (retval == 0) {
2390 		if (val) {
2391 			baseval |= flag;
2392 		} else {
2393 			baseval &= ~flag;
2394 		}
2395 		VMPTRLD(vmcs);
2396 		error = vmwrite(reg, baseval);
2397 		VMCLEAR(vmcs);
2398 
2399 		if (error) {
2400 			retval = error;
2401 		} else {
2402 			/*
2403 			 * Update optional stored flags, and record
2404 			 * setting
2405 			 */
2406 			if (pptr != NULL) {
2407 				*pptr = baseval;
2408 			}
2409 
2410 			if (val) {
2411 				vmx->cap[vcpu].set |= (1 << type);
2412 			} else {
2413 				vmx->cap[vcpu].set &= ~(1 << type);
2414 			}
2415 		}
2416 	}
2417 
2418         return (retval);
2419 }
2420 
2421 struct vlapic_vtx {
2422 	struct vlapic	vlapic;
2423 	struct pir_desc	*pir_desc;
2424 	struct vmx	*vmx;
2425 };
2426 
2427 #define	VMX_CTR_PIR(vm, vcpuid, pir_desc, notify, vector, level, msg)	\
2428 do {									\
2429 	VCPU_CTR2(vm, vcpuid, msg " assert %s-triggered vector %d",	\
2430 	    level ? "level" : "edge", vector);				\
2431 	VCPU_CTR1(vm, vcpuid, msg " pir0 0x%016lx", pir_desc->pir[0]);	\
2432 	VCPU_CTR1(vm, vcpuid, msg " pir1 0x%016lx", pir_desc->pir[1]);	\
2433 	VCPU_CTR1(vm, vcpuid, msg " pir2 0x%016lx", pir_desc->pir[2]);	\
2434 	VCPU_CTR1(vm, vcpuid, msg " pir3 0x%016lx", pir_desc->pir[3]);	\
2435 	VCPU_CTR1(vm, vcpuid, msg " notify: %s", notify ? "yes" : "no");\
2436 } while (0)
2437 
2438 /*
2439  * vlapic->ops handlers that utilize the APICv hardware assist described in
2440  * Chapter 29 of the Intel SDM.
2441  */
2442 static int
2443 vmx_set_intr_ready(struct vlapic *vlapic, int vector, bool level)
2444 {
2445 	struct vlapic_vtx *vlapic_vtx;
2446 	struct pir_desc *pir_desc;
2447 	uint64_t mask;
2448 	int idx, notify;
2449 
2450 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
2451 	pir_desc = vlapic_vtx->pir_desc;
2452 
2453 	/*
2454 	 * Keep track of interrupt requests in the PIR descriptor. This is
2455 	 * because the virtual APIC page pointed to by the VMCS cannot be
2456 	 * modified if the vcpu is running.
2457 	 */
2458 	idx = vector / 64;
2459 	mask = 1UL << (vector % 64);
2460 	atomic_set_long(&pir_desc->pir[idx], mask);
2461 	notify = atomic_cmpset_long(&pir_desc->pending, 0, 1);
2462 
2463 	VMX_CTR_PIR(vlapic->vm, vlapic->vcpuid, pir_desc, notify, vector,
2464 	    level, "vmx_set_intr_ready");
2465 	return (notify);
2466 }
2467 
2468 static int
2469 vmx_pending_intr(struct vlapic *vlapic, int *vecptr)
2470 {
2471 	struct vlapic_vtx *vlapic_vtx;
2472 	struct pir_desc *pir_desc;
2473 	struct LAPIC *lapic;
2474 	uint64_t pending, pirval;
2475 	uint32_t ppr, vpr;
2476 	int i;
2477 
2478 	/*
2479 	 * This function is only expected to be called from the 'HLT' exit
2480 	 * handler which does not care about the vector that is pending.
2481 	 */
2482 	KASSERT(vecptr == NULL, ("vmx_pending_intr: vecptr must be NULL"));
2483 
2484 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
2485 	pir_desc = vlapic_vtx->pir_desc;
2486 
2487 	pending = atomic_load_acq_long(&pir_desc->pending);
2488 	if (!pending)
2489 		return (0);	/* common case */
2490 
2491 	/*
2492 	 * If there is an interrupt pending then it will be recognized only
2493 	 * if its priority is greater than the processor priority.
2494 	 *
2495 	 * Special case: if the processor priority is zero then any pending
2496 	 * interrupt will be recognized.
2497 	 */
2498 	lapic = vlapic->apic_page;
2499 	ppr = lapic->ppr & 0xf0;
2500 	if (ppr == 0)
2501 		return (1);
2502 
2503 	VCPU_CTR1(vlapic->vm, vlapic->vcpuid, "HLT with non-zero PPR %d",
2504 	    lapic->ppr);
2505 
2506 	for (i = 3; i >= 0; i--) {
2507 		pirval = pir_desc->pir[i];
2508 		if (pirval != 0) {
2509 			vpr = (i * 64 + flsl(pirval) - 1) & 0xf0;
2510 			return (vpr > ppr);
2511 		}
2512 	}
2513 	return (0);
2514 }
2515 
2516 static void
2517 vmx_intr_accepted(struct vlapic *vlapic, int vector)
2518 {
2519 
2520 	panic("vmx_intr_accepted: not expected to be called");
2521 }
2522 
2523 static void
2524 vmx_set_tmr(struct vlapic *vlapic, int vector, bool level)
2525 {
2526 	struct vlapic_vtx *vlapic_vtx;
2527 	struct vmx *vmx;
2528 	struct vmcs *vmcs;
2529 	uint64_t mask, val;
2530 
2531 	KASSERT(vector >= 0 && vector <= 255, ("invalid vector %d", vector));
2532 	KASSERT(!vcpu_is_running(vlapic->vm, vlapic->vcpuid, NULL),
2533 	    ("vmx_set_tmr: vcpu cannot be running"));
2534 
2535 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
2536 	vmx = vlapic_vtx->vmx;
2537 	vmcs = &vmx->vmcs[vlapic->vcpuid];
2538 	mask = 1UL << (vector % 64);
2539 
2540 	VMPTRLD(vmcs);
2541 	val = vmcs_read(VMCS_EOI_EXIT(vector));
2542 	if (level)
2543 		val |= mask;
2544 	else
2545 		val &= ~mask;
2546 	vmcs_write(VMCS_EOI_EXIT(vector), val);
2547 	VMCLEAR(vmcs);
2548 }
2549 
2550 static void
2551 vmx_post_intr(struct vlapic *vlapic, int hostcpu)
2552 {
2553 
2554 	ipi_cpu(hostcpu, pirvec);
2555 }
2556 
2557 /*
2558  * Transfer the pending interrupts in the PIR descriptor to the IRR
2559  * in the virtual APIC page.
2560  */
2561 static void
2562 vmx_inject_pir(struct vlapic *vlapic)
2563 {
2564 	struct vlapic_vtx *vlapic_vtx;
2565 	struct pir_desc *pir_desc;
2566 	struct LAPIC *lapic;
2567 	uint64_t val, pirval;
2568 	int rvi, pirbase;
2569 	uint16_t intr_status_old, intr_status_new;
2570 
2571 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
2572 	pir_desc = vlapic_vtx->pir_desc;
2573 	if (atomic_cmpset_long(&pir_desc->pending, 1, 0) == 0) {
2574 		VCPU_CTR0(vlapic->vm, vlapic->vcpuid, "vmx_inject_pir: "
2575 		    "no posted interrupt pending");
2576 		return;
2577 	}
2578 
2579 	pirval = 0;
2580 	lapic = vlapic->apic_page;
2581 
2582 	val = atomic_readandclear_long(&pir_desc->pir[0]);
2583 	if (val != 0) {
2584 		lapic->irr0 |= val;
2585 		lapic->irr1 |= val >> 32;
2586 		pirbase = 0;
2587 		pirval = val;
2588 	}
2589 
2590 	val = atomic_readandclear_long(&pir_desc->pir[1]);
2591 	if (val != 0) {
2592 		lapic->irr2 |= val;
2593 		lapic->irr3 |= val >> 32;
2594 		pirbase = 64;
2595 		pirval = val;
2596 	}
2597 
2598 	val = atomic_readandclear_long(&pir_desc->pir[2]);
2599 	if (val != 0) {
2600 		lapic->irr4 |= val;
2601 		lapic->irr5 |= val >> 32;
2602 		pirbase = 128;
2603 		pirval = val;
2604 	}
2605 
2606 	val = atomic_readandclear_long(&pir_desc->pir[3]);
2607 	if (val != 0) {
2608 		lapic->irr6 |= val;
2609 		lapic->irr7 |= val >> 32;
2610 		pirbase = 192;
2611 		pirval = val;
2612 	}
2613 	VLAPIC_CTR_IRR(vlapic, "vmx_inject_pir");
2614 
2615 	/*
2616 	 * Update RVI so the processor can evaluate pending virtual
2617 	 * interrupts on VM-entry.
2618 	 */
2619 	if (pirval != 0) {
2620 		rvi = pirbase + flsl(pirval) - 1;
2621 		intr_status_old = vmcs_read(VMCS_GUEST_INTR_STATUS);
2622 		intr_status_new = (intr_status_old & 0xFF00) | rvi;
2623 		if (intr_status_new > intr_status_old) {
2624 			vmcs_write(VMCS_GUEST_INTR_STATUS, intr_status_new);
2625 			VCPU_CTR2(vlapic->vm, vlapic->vcpuid, "vmx_inject_pir: "
2626 			    "guest_intr_status changed from 0x%04x to 0x%04x",
2627 			    intr_status_old, intr_status_new);
2628 		}
2629 	}
2630 }
2631 
2632 static struct vlapic *
2633 vmx_vlapic_init(void *arg, int vcpuid)
2634 {
2635 	struct vmx *vmx;
2636 	struct vlapic *vlapic;
2637 	struct vlapic_vtx *vlapic_vtx;
2638 
2639 	vmx = arg;
2640 
2641 	vlapic = malloc(sizeof(struct vlapic_vtx), M_VLAPIC, M_WAITOK | M_ZERO);
2642 	vlapic->vm = vmx->vm;
2643 	vlapic->vcpuid = vcpuid;
2644 	vlapic->apic_page = (struct LAPIC *)&vmx->apic_page[vcpuid];
2645 
2646 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
2647 	vlapic_vtx->pir_desc = &vmx->pir_desc[vcpuid];
2648 	vlapic_vtx->vmx = vmx;
2649 
2650 	if (virtual_interrupt_delivery) {
2651 		vlapic->ops.set_intr_ready = vmx_set_intr_ready;
2652 		vlapic->ops.pending_intr = vmx_pending_intr;
2653 		vlapic->ops.intr_accepted = vmx_intr_accepted;
2654 		vlapic->ops.set_tmr = vmx_set_tmr;
2655 	}
2656 
2657 	if (posted_interrupts)
2658 		vlapic->ops.post_intr = vmx_post_intr;
2659 
2660 	vlapic_init(vlapic);
2661 
2662 	return (vlapic);
2663 }
2664 
2665 static void
2666 vmx_vlapic_cleanup(void *arg, struct vlapic *vlapic)
2667 {
2668 
2669 	vlapic_cleanup(vlapic);
2670 	free(vlapic, M_VLAPIC);
2671 }
2672 
2673 struct vmm_ops vmm_ops_intel = {
2674 	vmx_init,
2675 	vmx_cleanup,
2676 	vmx_restore,
2677 	vmx_vminit,
2678 	vmx_run,
2679 	vmx_vmcleanup,
2680 	vmx_getreg,
2681 	vmx_setreg,
2682 	vmx_getdesc,
2683 	vmx_setdesc,
2684 	vmx_inject,
2685 	vmx_getcap,
2686 	vmx_setcap,
2687 	ept_vmspace_alloc,
2688 	ept_vmspace_free,
2689 	vmx_vlapic_init,
2690 	vmx_vlapic_cleanup,
2691 };
2692