xref: /freebsd/sys/amd64/vmm/intel/vmx.c (revision 79ac3c12a714bcd3f2354c52d948aed9575c46d6)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011 NetApp, Inc.
5  * All rights reserved.
6  * Copyright (c) 2018 Joyent, Inc.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  * $FreeBSD$
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_bhyve_snapshot.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/smp.h>
40 #include <sys/kernel.h>
41 #include <sys/malloc.h>
42 #include <sys/pcpu.h>
43 #include <sys/proc.h>
44 #include <sys/smr.h>
45 #include <sys/sysctl.h>
46 
47 #include <vm/vm.h>
48 #include <vm/pmap.h>
49 
50 #include <machine/psl.h>
51 #include <machine/cpufunc.h>
52 #include <machine/md_var.h>
53 #include <machine/reg.h>
54 #include <machine/segments.h>
55 #include <machine/smp.h>
56 #include <machine/specialreg.h>
57 #include <machine/vmparam.h>
58 
59 #include <machine/vmm.h>
60 #include <machine/vmm_dev.h>
61 #include <machine/vmm_instruction_emul.h>
62 #include <machine/vmm_snapshot.h>
63 
64 #include "vmm_lapic.h"
65 #include "vmm_host.h"
66 #include "vmm_ioport.h"
67 #include "vmm_ktr.h"
68 #include "vmm_stat.h"
69 #include "vatpic.h"
70 #include "vlapic.h"
71 #include "vlapic_priv.h"
72 
73 #include "ept.h"
74 #include "vmx_cpufunc.h"
75 #include "vmx.h"
76 #include "vmx_msr.h"
77 #include "x86.h"
78 #include "vmx_controls.h"
79 
80 #define	PINBASED_CTLS_ONE_SETTING					\
81 	(PINBASED_EXTINT_EXITING	|				\
82 	 PINBASED_NMI_EXITING		|				\
83 	 PINBASED_VIRTUAL_NMI)
84 #define	PINBASED_CTLS_ZERO_SETTING	0
85 
86 #define PROCBASED_CTLS_WINDOW_SETTING					\
87 	(PROCBASED_INT_WINDOW_EXITING	|				\
88 	 PROCBASED_NMI_WINDOW_EXITING)
89 
90 #define	PROCBASED_CTLS_ONE_SETTING					\
91 	(PROCBASED_SECONDARY_CONTROLS	|				\
92 	 PROCBASED_MWAIT_EXITING	|				\
93 	 PROCBASED_MONITOR_EXITING	|				\
94 	 PROCBASED_IO_EXITING		|				\
95 	 PROCBASED_MSR_BITMAPS		|				\
96 	 PROCBASED_CTLS_WINDOW_SETTING	|				\
97 	 PROCBASED_CR8_LOAD_EXITING	|				\
98 	 PROCBASED_CR8_STORE_EXITING)
99 #define	PROCBASED_CTLS_ZERO_SETTING	\
100 	(PROCBASED_CR3_LOAD_EXITING |	\
101 	PROCBASED_CR3_STORE_EXITING |	\
102 	PROCBASED_IO_BITMAPS)
103 
104 #define	PROCBASED_CTLS2_ONE_SETTING	PROCBASED2_ENABLE_EPT
105 #define	PROCBASED_CTLS2_ZERO_SETTING	0
106 
107 #define	VM_EXIT_CTLS_ONE_SETTING					\
108 	(VM_EXIT_SAVE_DEBUG_CONTROLS		|			\
109 	VM_EXIT_HOST_LMA			|			\
110 	VM_EXIT_SAVE_EFER			|			\
111 	VM_EXIT_LOAD_EFER			|			\
112 	VM_EXIT_ACKNOWLEDGE_INTERRUPT)
113 
114 #define	VM_EXIT_CTLS_ZERO_SETTING	0
115 
116 #define	VM_ENTRY_CTLS_ONE_SETTING					\
117 	(VM_ENTRY_LOAD_DEBUG_CONTROLS		|			\
118 	VM_ENTRY_LOAD_EFER)
119 
120 #define	VM_ENTRY_CTLS_ZERO_SETTING					\
121 	(VM_ENTRY_INTO_SMM			|			\
122 	VM_ENTRY_DEACTIVATE_DUAL_MONITOR)
123 
124 #define	HANDLED		1
125 #define	UNHANDLED	0
126 
127 static MALLOC_DEFINE(M_VMX, "vmx", "vmx");
128 static MALLOC_DEFINE(M_VLAPIC, "vlapic", "vlapic");
129 
130 SYSCTL_DECL(_hw_vmm);
131 SYSCTL_NODE(_hw_vmm, OID_AUTO, vmx, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
132     NULL);
133 
134 int vmxon_enabled[MAXCPU];
135 static char vmxon_region[MAXCPU][PAGE_SIZE] __aligned(PAGE_SIZE);
136 
137 static uint32_t pinbased_ctls, procbased_ctls, procbased_ctls2;
138 static uint32_t exit_ctls, entry_ctls;
139 
140 static uint64_t cr0_ones_mask, cr0_zeros_mask;
141 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_ones_mask, CTLFLAG_RD,
142 	     &cr0_ones_mask, 0, NULL);
143 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_zeros_mask, CTLFLAG_RD,
144 	     &cr0_zeros_mask, 0, NULL);
145 
146 static uint64_t cr4_ones_mask, cr4_zeros_mask;
147 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_ones_mask, CTLFLAG_RD,
148 	     &cr4_ones_mask, 0, NULL);
149 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_zeros_mask, CTLFLAG_RD,
150 	     &cr4_zeros_mask, 0, NULL);
151 
152 static int vmx_initialized;
153 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, initialized, CTLFLAG_RD,
154 	   &vmx_initialized, 0, "Intel VMX initialized");
155 
156 /*
157  * Optional capabilities
158  */
159 static SYSCTL_NODE(_hw_vmm_vmx, OID_AUTO, cap,
160     CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
161     NULL);
162 
163 static int cap_halt_exit;
164 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, halt_exit, CTLFLAG_RD, &cap_halt_exit, 0,
165     "HLT triggers a VM-exit");
166 
167 static int cap_pause_exit;
168 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, pause_exit, CTLFLAG_RD, &cap_pause_exit,
169     0, "PAUSE triggers a VM-exit");
170 
171 static int cap_rdpid;
172 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, rdpid, CTLFLAG_RD, &cap_rdpid, 0,
173     "Guests are allowed to use RDPID");
174 
175 static int cap_rdtscp;
176 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, rdtscp, CTLFLAG_RD, &cap_rdtscp, 0,
177     "Guests are allowed to use RDTSCP");
178 
179 static int cap_unrestricted_guest;
180 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, unrestricted_guest, CTLFLAG_RD,
181     &cap_unrestricted_guest, 0, "Unrestricted guests");
182 
183 static int cap_monitor_trap;
184 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, monitor_trap, CTLFLAG_RD,
185     &cap_monitor_trap, 0, "Monitor trap flag");
186 
187 static int cap_invpcid;
188 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, invpcid, CTLFLAG_RD, &cap_invpcid,
189     0, "Guests are allowed to use INVPCID");
190 
191 static int tpr_shadowing;
192 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, tpr_shadowing, CTLFLAG_RD,
193     &tpr_shadowing, 0, "TPR shadowing support");
194 
195 static int virtual_interrupt_delivery;
196 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, virtual_interrupt_delivery, CTLFLAG_RD,
197     &virtual_interrupt_delivery, 0, "APICv virtual interrupt delivery support");
198 
199 static int posted_interrupts;
200 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, posted_interrupts, CTLFLAG_RD,
201     &posted_interrupts, 0, "APICv posted interrupt support");
202 
203 static int pirvec = -1;
204 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, posted_interrupt_vector, CTLFLAG_RD,
205     &pirvec, 0, "APICv posted interrupt vector");
206 
207 static struct unrhdr *vpid_unr;
208 static u_int vpid_alloc_failed;
209 SYSCTL_UINT(_hw_vmm_vmx, OID_AUTO, vpid_alloc_failed, CTLFLAG_RD,
210 	    &vpid_alloc_failed, 0, NULL);
211 
212 int guest_l1d_flush;
213 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, l1d_flush, CTLFLAG_RD,
214     &guest_l1d_flush, 0, NULL);
215 int guest_l1d_flush_sw;
216 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, l1d_flush_sw, CTLFLAG_RD,
217     &guest_l1d_flush_sw, 0, NULL);
218 
219 static struct msr_entry msr_load_list[1] __aligned(16);
220 
221 /*
222  * The definitions of SDT probes for VMX.
223  */
224 
225 SDT_PROBE_DEFINE3(vmm, vmx, exit, entry,
226     "struct vmx *", "int", "struct vm_exit *");
227 
228 SDT_PROBE_DEFINE4(vmm, vmx, exit, taskswitch,
229     "struct vmx *", "int", "struct vm_exit *", "struct vm_task_switch *");
230 
231 SDT_PROBE_DEFINE4(vmm, vmx, exit, craccess,
232     "struct vmx *", "int", "struct vm_exit *", "uint64_t");
233 
234 SDT_PROBE_DEFINE4(vmm, vmx, exit, rdmsr,
235     "struct vmx *", "int", "struct vm_exit *", "uint32_t");
236 
237 SDT_PROBE_DEFINE5(vmm, vmx, exit, wrmsr,
238     "struct vmx *", "int", "struct vm_exit *", "uint32_t", "uint64_t");
239 
240 SDT_PROBE_DEFINE3(vmm, vmx, exit, halt,
241     "struct vmx *", "int", "struct vm_exit *");
242 
243 SDT_PROBE_DEFINE3(vmm, vmx, exit, mtrap,
244     "struct vmx *", "int", "struct vm_exit *");
245 
246 SDT_PROBE_DEFINE3(vmm, vmx, exit, pause,
247     "struct vmx *", "int", "struct vm_exit *");
248 
249 SDT_PROBE_DEFINE3(vmm, vmx, exit, intrwindow,
250     "struct vmx *", "int", "struct vm_exit *");
251 
252 SDT_PROBE_DEFINE4(vmm, vmx, exit, interrupt,
253     "struct vmx *", "int", "struct vm_exit *", "uint32_t");
254 
255 SDT_PROBE_DEFINE3(vmm, vmx, exit, nmiwindow,
256     "struct vmx *", "int", "struct vm_exit *");
257 
258 SDT_PROBE_DEFINE3(vmm, vmx, exit, inout,
259     "struct vmx *", "int", "struct vm_exit *");
260 
261 SDT_PROBE_DEFINE3(vmm, vmx, exit, cpuid,
262     "struct vmx *", "int", "struct vm_exit *");
263 
264 SDT_PROBE_DEFINE5(vmm, vmx, exit, exception,
265     "struct vmx *", "int", "struct vm_exit *", "uint32_t", "int");
266 
267 SDT_PROBE_DEFINE5(vmm, vmx, exit, nestedfault,
268     "struct vmx *", "int", "struct vm_exit *", "uint64_t", "uint64_t");
269 
270 SDT_PROBE_DEFINE4(vmm, vmx, exit, mmiofault,
271     "struct vmx *", "int", "struct vm_exit *", "uint64_t");
272 
273 SDT_PROBE_DEFINE3(vmm, vmx, exit, eoi,
274     "struct vmx *", "int", "struct vm_exit *");
275 
276 SDT_PROBE_DEFINE3(vmm, vmx, exit, apicaccess,
277     "struct vmx *", "int", "struct vm_exit *");
278 
279 SDT_PROBE_DEFINE4(vmm, vmx, exit, apicwrite,
280     "struct vmx *", "int", "struct vm_exit *", "struct vlapic *");
281 
282 SDT_PROBE_DEFINE3(vmm, vmx, exit, xsetbv,
283     "struct vmx *", "int", "struct vm_exit *");
284 
285 SDT_PROBE_DEFINE3(vmm, vmx, exit, monitor,
286     "struct vmx *", "int", "struct vm_exit *");
287 
288 SDT_PROBE_DEFINE3(vmm, vmx, exit, mwait,
289     "struct vmx *", "int", "struct vm_exit *");
290 
291 SDT_PROBE_DEFINE3(vmm, vmx, exit, vminsn,
292     "struct vmx *", "int", "struct vm_exit *");
293 
294 SDT_PROBE_DEFINE4(vmm, vmx, exit, unknown,
295     "struct vmx *", "int", "struct vm_exit *", "uint32_t");
296 
297 SDT_PROBE_DEFINE4(vmm, vmx, exit, return,
298     "struct vmx *", "int", "struct vm_exit *", "int");
299 
300 /*
301  * Use the last page below 4GB as the APIC access address. This address is
302  * occupied by the boot firmware so it is guaranteed that it will not conflict
303  * with a page in system memory.
304  */
305 #define	APIC_ACCESS_ADDRESS	0xFFFFF000
306 
307 static int vmx_getdesc(void *arg, int vcpu, int reg, struct seg_desc *desc);
308 static int vmx_getreg(void *arg, int vcpu, int reg, uint64_t *retval);
309 static int vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val);
310 static void vmx_inject_pir(struct vlapic *vlapic);
311 #ifdef BHYVE_SNAPSHOT
312 static int vmx_restore_tsc(void *arg, int vcpu, uint64_t now);
313 #endif
314 
315 static inline bool
316 host_has_rdpid(void)
317 {
318 	return ((cpu_stdext_feature2 & CPUID_STDEXT2_RDPID) != 0);
319 }
320 
321 static inline bool
322 host_has_rdtscp(void)
323 {
324 	return ((amd_feature & AMDID_RDTSCP) != 0);
325 }
326 
327 #ifdef KTR
328 static const char *
329 exit_reason_to_str(int reason)
330 {
331 	static char reasonbuf[32];
332 
333 	switch (reason) {
334 	case EXIT_REASON_EXCEPTION:
335 		return "exception";
336 	case EXIT_REASON_EXT_INTR:
337 		return "extint";
338 	case EXIT_REASON_TRIPLE_FAULT:
339 		return "triplefault";
340 	case EXIT_REASON_INIT:
341 		return "init";
342 	case EXIT_REASON_SIPI:
343 		return "sipi";
344 	case EXIT_REASON_IO_SMI:
345 		return "iosmi";
346 	case EXIT_REASON_SMI:
347 		return "smi";
348 	case EXIT_REASON_INTR_WINDOW:
349 		return "intrwindow";
350 	case EXIT_REASON_NMI_WINDOW:
351 		return "nmiwindow";
352 	case EXIT_REASON_TASK_SWITCH:
353 		return "taskswitch";
354 	case EXIT_REASON_CPUID:
355 		return "cpuid";
356 	case EXIT_REASON_GETSEC:
357 		return "getsec";
358 	case EXIT_REASON_HLT:
359 		return "hlt";
360 	case EXIT_REASON_INVD:
361 		return "invd";
362 	case EXIT_REASON_INVLPG:
363 		return "invlpg";
364 	case EXIT_REASON_RDPMC:
365 		return "rdpmc";
366 	case EXIT_REASON_RDTSC:
367 		return "rdtsc";
368 	case EXIT_REASON_RSM:
369 		return "rsm";
370 	case EXIT_REASON_VMCALL:
371 		return "vmcall";
372 	case EXIT_REASON_VMCLEAR:
373 		return "vmclear";
374 	case EXIT_REASON_VMLAUNCH:
375 		return "vmlaunch";
376 	case EXIT_REASON_VMPTRLD:
377 		return "vmptrld";
378 	case EXIT_REASON_VMPTRST:
379 		return "vmptrst";
380 	case EXIT_REASON_VMREAD:
381 		return "vmread";
382 	case EXIT_REASON_VMRESUME:
383 		return "vmresume";
384 	case EXIT_REASON_VMWRITE:
385 		return "vmwrite";
386 	case EXIT_REASON_VMXOFF:
387 		return "vmxoff";
388 	case EXIT_REASON_VMXON:
389 		return "vmxon";
390 	case EXIT_REASON_CR_ACCESS:
391 		return "craccess";
392 	case EXIT_REASON_DR_ACCESS:
393 		return "draccess";
394 	case EXIT_REASON_INOUT:
395 		return "inout";
396 	case EXIT_REASON_RDMSR:
397 		return "rdmsr";
398 	case EXIT_REASON_WRMSR:
399 		return "wrmsr";
400 	case EXIT_REASON_INVAL_VMCS:
401 		return "invalvmcs";
402 	case EXIT_REASON_INVAL_MSR:
403 		return "invalmsr";
404 	case EXIT_REASON_MWAIT:
405 		return "mwait";
406 	case EXIT_REASON_MTF:
407 		return "mtf";
408 	case EXIT_REASON_MONITOR:
409 		return "monitor";
410 	case EXIT_REASON_PAUSE:
411 		return "pause";
412 	case EXIT_REASON_MCE_DURING_ENTRY:
413 		return "mce-during-entry";
414 	case EXIT_REASON_TPR:
415 		return "tpr";
416 	case EXIT_REASON_APIC_ACCESS:
417 		return "apic-access";
418 	case EXIT_REASON_GDTR_IDTR:
419 		return "gdtridtr";
420 	case EXIT_REASON_LDTR_TR:
421 		return "ldtrtr";
422 	case EXIT_REASON_EPT_FAULT:
423 		return "eptfault";
424 	case EXIT_REASON_EPT_MISCONFIG:
425 		return "eptmisconfig";
426 	case EXIT_REASON_INVEPT:
427 		return "invept";
428 	case EXIT_REASON_RDTSCP:
429 		return "rdtscp";
430 	case EXIT_REASON_VMX_PREEMPT:
431 		return "vmxpreempt";
432 	case EXIT_REASON_INVVPID:
433 		return "invvpid";
434 	case EXIT_REASON_WBINVD:
435 		return "wbinvd";
436 	case EXIT_REASON_XSETBV:
437 		return "xsetbv";
438 	case EXIT_REASON_APIC_WRITE:
439 		return "apic-write";
440 	default:
441 		snprintf(reasonbuf, sizeof(reasonbuf), "%d", reason);
442 		return (reasonbuf);
443 	}
444 }
445 #endif	/* KTR */
446 
447 static int
448 vmx_allow_x2apic_msrs(struct vmx *vmx)
449 {
450 	int i, error;
451 
452 	error = 0;
453 
454 	/*
455 	 * Allow readonly access to the following x2APIC MSRs from the guest.
456 	 */
457 	error += guest_msr_ro(vmx, MSR_APIC_ID);
458 	error += guest_msr_ro(vmx, MSR_APIC_VERSION);
459 	error += guest_msr_ro(vmx, MSR_APIC_LDR);
460 	error += guest_msr_ro(vmx, MSR_APIC_SVR);
461 
462 	for (i = 0; i < 8; i++)
463 		error += guest_msr_ro(vmx, MSR_APIC_ISR0 + i);
464 
465 	for (i = 0; i < 8; i++)
466 		error += guest_msr_ro(vmx, MSR_APIC_TMR0 + i);
467 
468 	for (i = 0; i < 8; i++)
469 		error += guest_msr_ro(vmx, MSR_APIC_IRR0 + i);
470 
471 	error += guest_msr_ro(vmx, MSR_APIC_ESR);
472 	error += guest_msr_ro(vmx, MSR_APIC_LVT_TIMER);
473 	error += guest_msr_ro(vmx, MSR_APIC_LVT_THERMAL);
474 	error += guest_msr_ro(vmx, MSR_APIC_LVT_PCINT);
475 	error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT0);
476 	error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT1);
477 	error += guest_msr_ro(vmx, MSR_APIC_LVT_ERROR);
478 	error += guest_msr_ro(vmx, MSR_APIC_ICR_TIMER);
479 	error += guest_msr_ro(vmx, MSR_APIC_DCR_TIMER);
480 	error += guest_msr_ro(vmx, MSR_APIC_ICR);
481 
482 	/*
483 	 * Allow TPR, EOI and SELF_IPI MSRs to be read and written by the guest.
484 	 *
485 	 * These registers get special treatment described in the section
486 	 * "Virtualizing MSR-Based APIC Accesses".
487 	 */
488 	error += guest_msr_rw(vmx, MSR_APIC_TPR);
489 	error += guest_msr_rw(vmx, MSR_APIC_EOI);
490 	error += guest_msr_rw(vmx, MSR_APIC_SELF_IPI);
491 
492 	return (error);
493 }
494 
495 u_long
496 vmx_fix_cr0(u_long cr0)
497 {
498 
499 	return ((cr0 | cr0_ones_mask) & ~cr0_zeros_mask);
500 }
501 
502 u_long
503 vmx_fix_cr4(u_long cr4)
504 {
505 
506 	return ((cr4 | cr4_ones_mask) & ~cr4_zeros_mask);
507 }
508 
509 static void
510 vpid_free(int vpid)
511 {
512 	if (vpid < 0 || vpid > 0xffff)
513 		panic("vpid_free: invalid vpid %d", vpid);
514 
515 	/*
516 	 * VPIDs [0,VM_MAXCPU] are special and are not allocated from
517 	 * the unit number allocator.
518 	 */
519 
520 	if (vpid > VM_MAXCPU)
521 		free_unr(vpid_unr, vpid);
522 }
523 
524 static void
525 vpid_alloc(uint16_t *vpid, int num)
526 {
527 	int i, x;
528 
529 	if (num <= 0 || num > VM_MAXCPU)
530 		panic("invalid number of vpids requested: %d", num);
531 
532 	/*
533 	 * If the "enable vpid" execution control is not enabled then the
534 	 * VPID is required to be 0 for all vcpus.
535 	 */
536 	if ((procbased_ctls2 & PROCBASED2_ENABLE_VPID) == 0) {
537 		for (i = 0; i < num; i++)
538 			vpid[i] = 0;
539 		return;
540 	}
541 
542 	/*
543 	 * Allocate a unique VPID for each vcpu from the unit number allocator.
544 	 */
545 	for (i = 0; i < num; i++) {
546 		x = alloc_unr(vpid_unr);
547 		if (x == -1)
548 			break;
549 		else
550 			vpid[i] = x;
551 	}
552 
553 	if (i < num) {
554 		atomic_add_int(&vpid_alloc_failed, 1);
555 
556 		/*
557 		 * If the unit number allocator does not have enough unique
558 		 * VPIDs then we need to allocate from the [1,VM_MAXCPU] range.
559 		 *
560 		 * These VPIDs are not be unique across VMs but this does not
561 		 * affect correctness because the combined mappings are also
562 		 * tagged with the EP4TA which is unique for each VM.
563 		 *
564 		 * It is still sub-optimal because the invvpid will invalidate
565 		 * combined mappings for a particular VPID across all EP4TAs.
566 		 */
567 		while (i-- > 0)
568 			vpid_free(vpid[i]);
569 
570 		for (i = 0; i < num; i++)
571 			vpid[i] = i + 1;
572 	}
573 }
574 
575 static void
576 vpid_init(void)
577 {
578 	/*
579 	 * VPID 0 is required when the "enable VPID" execution control is
580 	 * disabled.
581 	 *
582 	 * VPIDs [1,VM_MAXCPU] are used as the "overflow namespace" when the
583 	 * unit number allocator does not have sufficient unique VPIDs to
584 	 * satisfy the allocation.
585 	 *
586 	 * The remaining VPIDs are managed by the unit number allocator.
587 	 */
588 	vpid_unr = new_unrhdr(VM_MAXCPU + 1, 0xffff, NULL);
589 }
590 
591 static void
592 vmx_disable(void *arg __unused)
593 {
594 	struct invvpid_desc invvpid_desc = { 0 };
595 	struct invept_desc invept_desc = { 0 };
596 
597 	if (vmxon_enabled[curcpu]) {
598 		/*
599 		 * See sections 25.3.3.3 and 25.3.3.4 in Intel Vol 3b.
600 		 *
601 		 * VMXON or VMXOFF are not required to invalidate any TLB
602 		 * caching structures. This prevents potential retention of
603 		 * cached information in the TLB between distinct VMX episodes.
604 		 */
605 		invvpid(INVVPID_TYPE_ALL_CONTEXTS, invvpid_desc);
606 		invept(INVEPT_TYPE_ALL_CONTEXTS, invept_desc);
607 		vmxoff();
608 	}
609 	load_cr4(rcr4() & ~CR4_VMXE);
610 }
611 
612 static int
613 vmx_modcleanup(void)
614 {
615 
616 	if (pirvec >= 0)
617 		lapic_ipi_free(pirvec);
618 
619 	if (vpid_unr != NULL) {
620 		delete_unrhdr(vpid_unr);
621 		vpid_unr = NULL;
622 	}
623 
624 	if (nmi_flush_l1d_sw == 1)
625 		nmi_flush_l1d_sw = 0;
626 
627 	smp_rendezvous(NULL, vmx_disable, NULL, NULL);
628 
629 	return (0);
630 }
631 
632 static void
633 vmx_enable(void *arg __unused)
634 {
635 	int error;
636 	uint64_t feature_control;
637 
638 	feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL);
639 	if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 0 ||
640 	    (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) {
641 		wrmsr(MSR_IA32_FEATURE_CONTROL,
642 		    feature_control | IA32_FEATURE_CONTROL_VMX_EN |
643 		    IA32_FEATURE_CONTROL_LOCK);
644 	}
645 
646 	load_cr4(rcr4() | CR4_VMXE);
647 
648 	*(uint32_t *)vmxon_region[curcpu] = vmx_revision();
649 	error = vmxon(vmxon_region[curcpu]);
650 	if (error == 0)
651 		vmxon_enabled[curcpu] = 1;
652 }
653 
654 static void
655 vmx_modresume(void)
656 {
657 
658 	if (vmxon_enabled[curcpu])
659 		vmxon(vmxon_region[curcpu]);
660 }
661 
662 static int
663 vmx_modinit(int ipinum)
664 {
665 	int error;
666 	uint64_t basic, fixed0, fixed1, feature_control;
667 	uint32_t tmp, procbased2_vid_bits;
668 
669 	/* CPUID.1:ECX[bit 5] must be 1 for processor to support VMX */
670 	if (!(cpu_feature2 & CPUID2_VMX)) {
671 		printf("vmx_modinit: processor does not support VMX "
672 		    "operation\n");
673 		return (ENXIO);
674 	}
675 
676 	/*
677 	 * Verify that MSR_IA32_FEATURE_CONTROL lock and VMXON enable bits
678 	 * are set (bits 0 and 2 respectively).
679 	 */
680 	feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL);
681 	if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 1 &&
682 	    (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) {
683 		printf("vmx_modinit: VMX operation disabled by BIOS\n");
684 		return (ENXIO);
685 	}
686 
687 	/*
688 	 * Verify capabilities MSR_VMX_BASIC:
689 	 * - bit 54 indicates support for INS/OUTS decoding
690 	 */
691 	basic = rdmsr(MSR_VMX_BASIC);
692 	if ((basic & (1UL << 54)) == 0) {
693 		printf("vmx_modinit: processor does not support desired basic "
694 		    "capabilities\n");
695 		return (EINVAL);
696 	}
697 
698 	/* Check support for primary processor-based VM-execution controls */
699 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
700 			       MSR_VMX_TRUE_PROCBASED_CTLS,
701 			       PROCBASED_CTLS_ONE_SETTING,
702 			       PROCBASED_CTLS_ZERO_SETTING, &procbased_ctls);
703 	if (error) {
704 		printf("vmx_modinit: processor does not support desired "
705 		    "primary processor-based controls\n");
706 		return (error);
707 	}
708 
709 	/* Clear the processor-based ctl bits that are set on demand */
710 	procbased_ctls &= ~PROCBASED_CTLS_WINDOW_SETTING;
711 
712 	/* Check support for secondary processor-based VM-execution controls */
713 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
714 			       MSR_VMX_PROCBASED_CTLS2,
715 			       PROCBASED_CTLS2_ONE_SETTING,
716 			       PROCBASED_CTLS2_ZERO_SETTING, &procbased_ctls2);
717 	if (error) {
718 		printf("vmx_modinit: processor does not support desired "
719 		    "secondary processor-based controls\n");
720 		return (error);
721 	}
722 
723 	/* Check support for VPID */
724 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2,
725 			       PROCBASED2_ENABLE_VPID, 0, &tmp);
726 	if (error == 0)
727 		procbased_ctls2 |= PROCBASED2_ENABLE_VPID;
728 
729 	/* Check support for pin-based VM-execution controls */
730 	error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS,
731 			       MSR_VMX_TRUE_PINBASED_CTLS,
732 			       PINBASED_CTLS_ONE_SETTING,
733 			       PINBASED_CTLS_ZERO_SETTING, &pinbased_ctls);
734 	if (error) {
735 		printf("vmx_modinit: processor does not support desired "
736 		    "pin-based controls\n");
737 		return (error);
738 	}
739 
740 	/* Check support for VM-exit controls */
741 	error = vmx_set_ctlreg(MSR_VMX_EXIT_CTLS, MSR_VMX_TRUE_EXIT_CTLS,
742 			       VM_EXIT_CTLS_ONE_SETTING,
743 			       VM_EXIT_CTLS_ZERO_SETTING,
744 			       &exit_ctls);
745 	if (error) {
746 		printf("vmx_modinit: processor does not support desired "
747 		    "exit controls\n");
748 		return (error);
749 	}
750 
751 	/* Check support for VM-entry controls */
752 	error = vmx_set_ctlreg(MSR_VMX_ENTRY_CTLS, MSR_VMX_TRUE_ENTRY_CTLS,
753 	    VM_ENTRY_CTLS_ONE_SETTING, VM_ENTRY_CTLS_ZERO_SETTING,
754 	    &entry_ctls);
755 	if (error) {
756 		printf("vmx_modinit: processor does not support desired "
757 		    "entry controls\n");
758 		return (error);
759 	}
760 
761 	/*
762 	 * Check support for optional features by testing them
763 	 * as individual bits
764 	 */
765 	cap_halt_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
766 					MSR_VMX_TRUE_PROCBASED_CTLS,
767 					PROCBASED_HLT_EXITING, 0,
768 					&tmp) == 0);
769 
770 	cap_monitor_trap = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
771 					MSR_VMX_PROCBASED_CTLS,
772 					PROCBASED_MTF, 0,
773 					&tmp) == 0);
774 
775 	cap_pause_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
776 					 MSR_VMX_TRUE_PROCBASED_CTLS,
777 					 PROCBASED_PAUSE_EXITING, 0,
778 					 &tmp) == 0);
779 
780 	/*
781 	 * Check support for RDPID and/or RDTSCP.
782 	 *
783 	 * Support a pass-through-based implementation of these via the
784 	 * "enable RDTSCP" VM-execution control and the "RDTSC exiting"
785 	 * VM-execution control.
786 	 *
787 	 * The "enable RDTSCP" VM-execution control applies to both RDPID
788 	 * and RDTSCP (see SDM volume 3, section 25.3, "Changes to
789 	 * Instruction Behavior in VMX Non-root operation"); this is why
790 	 * only this VM-execution control needs to be enabled in order to
791 	 * enable passing through whichever of RDPID and/or RDTSCP are
792 	 * supported by the host.
793 	 *
794 	 * The "RDTSC exiting" VM-execution control applies to both RDTSC
795 	 * and RDTSCP (again, per SDM volume 3, section 25.3), and is
796 	 * already set up for RDTSC and RDTSCP pass-through by the current
797 	 * implementation of RDTSC.
798 	 *
799 	 * Although RDPID and RDTSCP are optional capabilities, since there
800 	 * does not currently seem to be a use case for enabling/disabling
801 	 * these via libvmmapi, choose not to support this and, instead,
802 	 * just statically always enable or always disable this support
803 	 * across all vCPUs on all VMs. (Note that there may be some
804 	 * complications to providing this functionality, e.g., the MSR
805 	 * bitmap is currently per-VM rather than per-vCPU while the
806 	 * capability API wants to be able to control capabilities on a
807 	 * per-vCPU basis).
808 	 */
809 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
810 			       MSR_VMX_PROCBASED_CTLS2,
811 			       PROCBASED2_ENABLE_RDTSCP, 0, &tmp);
812 	cap_rdpid = error == 0 && host_has_rdpid();
813 	cap_rdtscp = error == 0 && host_has_rdtscp();
814 	if (cap_rdpid || cap_rdtscp)
815 		procbased_ctls2 |= PROCBASED2_ENABLE_RDTSCP;
816 
817 	cap_unrestricted_guest = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
818 					MSR_VMX_PROCBASED_CTLS2,
819 					PROCBASED2_UNRESTRICTED_GUEST, 0,
820 				        &tmp) == 0);
821 
822 	cap_invpcid = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
823 	    MSR_VMX_PROCBASED_CTLS2, PROCBASED2_ENABLE_INVPCID, 0,
824 	    &tmp) == 0);
825 
826 	/*
827 	 * Check support for TPR shadow.
828 	 */
829 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
830 	    MSR_VMX_TRUE_PROCBASED_CTLS, PROCBASED_USE_TPR_SHADOW, 0,
831 	    &tmp);
832 	if (error == 0) {
833 		tpr_shadowing = 1;
834 		TUNABLE_INT_FETCH("hw.vmm.vmx.use_tpr_shadowing",
835 		    &tpr_shadowing);
836 	}
837 
838 	if (tpr_shadowing) {
839 		procbased_ctls |= PROCBASED_USE_TPR_SHADOW;
840 		procbased_ctls &= ~PROCBASED_CR8_LOAD_EXITING;
841 		procbased_ctls &= ~PROCBASED_CR8_STORE_EXITING;
842 	}
843 
844 	/*
845 	 * Check support for virtual interrupt delivery.
846 	 */
847 	procbased2_vid_bits = (PROCBASED2_VIRTUALIZE_APIC_ACCESSES |
848 	    PROCBASED2_VIRTUALIZE_X2APIC_MODE |
849 	    PROCBASED2_APIC_REGISTER_VIRTUALIZATION |
850 	    PROCBASED2_VIRTUAL_INTERRUPT_DELIVERY);
851 
852 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2,
853 	    procbased2_vid_bits, 0, &tmp);
854 	if (error == 0 && tpr_shadowing) {
855 		virtual_interrupt_delivery = 1;
856 		TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_vid",
857 		    &virtual_interrupt_delivery);
858 	}
859 
860 	if (virtual_interrupt_delivery) {
861 		procbased_ctls |= PROCBASED_USE_TPR_SHADOW;
862 		procbased_ctls2 |= procbased2_vid_bits;
863 		procbased_ctls2 &= ~PROCBASED2_VIRTUALIZE_X2APIC_MODE;
864 
865 		/*
866 		 * Check for Posted Interrupts only if Virtual Interrupt
867 		 * Delivery is enabled.
868 		 */
869 		error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS,
870 		    MSR_VMX_TRUE_PINBASED_CTLS, PINBASED_POSTED_INTERRUPT, 0,
871 		    &tmp);
872 		if (error == 0) {
873 			pirvec = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) :
874 			    &IDTVEC(justreturn));
875 			if (pirvec < 0) {
876 				if (bootverbose) {
877 					printf("vmx_modinit: unable to "
878 					    "allocate posted interrupt "
879 					    "vector\n");
880 				}
881 			} else {
882 				posted_interrupts = 1;
883 				TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_pir",
884 				    &posted_interrupts);
885 			}
886 		}
887 	}
888 
889 	if (posted_interrupts)
890 		    pinbased_ctls |= PINBASED_POSTED_INTERRUPT;
891 
892 	/* Initialize EPT */
893 	error = ept_init(ipinum);
894 	if (error) {
895 		printf("vmx_modinit: ept initialization failed (%d)\n", error);
896 		return (error);
897 	}
898 
899 	guest_l1d_flush = (cpu_ia32_arch_caps &
900 	    IA32_ARCH_CAP_SKIP_L1DFL_VMENTRY) == 0;
901 	TUNABLE_INT_FETCH("hw.vmm.l1d_flush", &guest_l1d_flush);
902 
903 	/*
904 	 * L1D cache flush is enabled.  Use IA32_FLUSH_CMD MSR when
905 	 * available.  Otherwise fall back to the software flush
906 	 * method which loads enough data from the kernel text to
907 	 * flush existing L1D content, both on VMX entry and on NMI
908 	 * return.
909 	 */
910 	if (guest_l1d_flush) {
911 		if ((cpu_stdext_feature3 & CPUID_STDEXT3_L1D_FLUSH) == 0) {
912 			guest_l1d_flush_sw = 1;
913 			TUNABLE_INT_FETCH("hw.vmm.l1d_flush_sw",
914 			    &guest_l1d_flush_sw);
915 		}
916 		if (guest_l1d_flush_sw) {
917 			if (nmi_flush_l1d_sw <= 1)
918 				nmi_flush_l1d_sw = 1;
919 		} else {
920 			msr_load_list[0].index = MSR_IA32_FLUSH_CMD;
921 			msr_load_list[0].val = IA32_FLUSH_CMD_L1D;
922 		}
923 	}
924 
925 	/*
926 	 * Stash the cr0 and cr4 bits that must be fixed to 0 or 1
927 	 */
928 	fixed0 = rdmsr(MSR_VMX_CR0_FIXED0);
929 	fixed1 = rdmsr(MSR_VMX_CR0_FIXED1);
930 	cr0_ones_mask = fixed0 & fixed1;
931 	cr0_zeros_mask = ~fixed0 & ~fixed1;
932 
933 	/*
934 	 * CR0_PE and CR0_PG can be set to zero in VMX non-root operation
935 	 * if unrestricted guest execution is allowed.
936 	 */
937 	if (cap_unrestricted_guest)
938 		cr0_ones_mask &= ~(CR0_PG | CR0_PE);
939 
940 	/*
941 	 * Do not allow the guest to set CR0_NW or CR0_CD.
942 	 */
943 	cr0_zeros_mask |= (CR0_NW | CR0_CD);
944 
945 	fixed0 = rdmsr(MSR_VMX_CR4_FIXED0);
946 	fixed1 = rdmsr(MSR_VMX_CR4_FIXED1);
947 	cr4_ones_mask = fixed0 & fixed1;
948 	cr4_zeros_mask = ~fixed0 & ~fixed1;
949 
950 	vpid_init();
951 
952 	vmx_msr_init();
953 
954 	/* enable VMX operation */
955 	smp_rendezvous(NULL, vmx_enable, NULL, NULL);
956 
957 	vmx_initialized = 1;
958 
959 	return (0);
960 }
961 
962 static void
963 vmx_trigger_hostintr(int vector)
964 {
965 	uintptr_t func;
966 	struct gate_descriptor *gd;
967 
968 	gd = &idt[vector];
969 
970 	KASSERT(vector >= 32 && vector <= 255, ("vmx_trigger_hostintr: "
971 	    "invalid vector %d", vector));
972 	KASSERT(gd->gd_p == 1, ("gate descriptor for vector %d not present",
973 	    vector));
974 	KASSERT(gd->gd_type == SDT_SYSIGT, ("gate descriptor for vector %d "
975 	    "has invalid type %d", vector, gd->gd_type));
976 	KASSERT(gd->gd_dpl == SEL_KPL, ("gate descriptor for vector %d "
977 	    "has invalid dpl %d", vector, gd->gd_dpl));
978 	KASSERT(gd->gd_selector == GSEL(GCODE_SEL, SEL_KPL), ("gate descriptor "
979 	    "for vector %d has invalid selector %d", vector, gd->gd_selector));
980 	KASSERT(gd->gd_ist == 0, ("gate descriptor for vector %d has invalid "
981 	    "IST %d", vector, gd->gd_ist));
982 
983 	func = ((long)gd->gd_hioffset << 16 | gd->gd_looffset);
984 	vmx_call_isr(func);
985 }
986 
987 static int
988 vmx_setup_cr_shadow(int which, struct vmcs *vmcs, uint32_t initial)
989 {
990 	int error, mask_ident, shadow_ident;
991 	uint64_t mask_value;
992 
993 	if (which != 0 && which != 4)
994 		panic("vmx_setup_cr_shadow: unknown cr%d", which);
995 
996 	if (which == 0) {
997 		mask_ident = VMCS_CR0_MASK;
998 		mask_value = cr0_ones_mask | cr0_zeros_mask;
999 		shadow_ident = VMCS_CR0_SHADOW;
1000 	} else {
1001 		mask_ident = VMCS_CR4_MASK;
1002 		mask_value = cr4_ones_mask | cr4_zeros_mask;
1003 		shadow_ident = VMCS_CR4_SHADOW;
1004 	}
1005 
1006 	error = vmcs_setreg(vmcs, 0, VMCS_IDENT(mask_ident), mask_value);
1007 	if (error)
1008 		return (error);
1009 
1010 	error = vmcs_setreg(vmcs, 0, VMCS_IDENT(shadow_ident), initial);
1011 	if (error)
1012 		return (error);
1013 
1014 	return (0);
1015 }
1016 #define	vmx_setup_cr0_shadow(vmcs,init)	vmx_setup_cr_shadow(0, (vmcs), (init))
1017 #define	vmx_setup_cr4_shadow(vmcs,init)	vmx_setup_cr_shadow(4, (vmcs), (init))
1018 
1019 static void *
1020 vmx_init(struct vm *vm, pmap_t pmap)
1021 {
1022 	uint16_t vpid[VM_MAXCPU];
1023 	int i, error;
1024 	struct vmx *vmx;
1025 	struct vmcs *vmcs;
1026 	uint32_t exc_bitmap;
1027 	uint16_t maxcpus;
1028 
1029 	vmx = malloc(sizeof(struct vmx), M_VMX, M_WAITOK | M_ZERO);
1030 	if ((uintptr_t)vmx & PAGE_MASK) {
1031 		panic("malloc of struct vmx not aligned on %d byte boundary",
1032 		      PAGE_SIZE);
1033 	}
1034 	vmx->vm = vm;
1035 
1036 	vmx->eptp = eptp(vtophys((vm_offset_t)pmap->pm_pmltop));
1037 
1038 	/*
1039 	 * Clean up EPTP-tagged guest physical and combined mappings
1040 	 *
1041 	 * VMX transitions are not required to invalidate any guest physical
1042 	 * mappings. So, it may be possible for stale guest physical mappings
1043 	 * to be present in the processor TLBs.
1044 	 *
1045 	 * Combined mappings for this EP4TA are also invalidated for all VPIDs.
1046 	 */
1047 	ept_invalidate_mappings(vmx->eptp);
1048 
1049 	msr_bitmap_initialize(vmx->msr_bitmap);
1050 
1051 	/*
1052 	 * It is safe to allow direct access to MSR_GSBASE and MSR_FSBASE.
1053 	 * The guest FSBASE and GSBASE are saved and restored during
1054 	 * vm-exit and vm-entry respectively. The host FSBASE and GSBASE are
1055 	 * always restored from the vmcs host state area on vm-exit.
1056 	 *
1057 	 * The SYSENTER_CS/ESP/EIP MSRs are identical to FS/GSBASE in
1058 	 * how they are saved/restored so can be directly accessed by the
1059 	 * guest.
1060 	 *
1061 	 * MSR_EFER is saved and restored in the guest VMCS area on a
1062 	 * VM exit and entry respectively. It is also restored from the
1063 	 * host VMCS area on a VM exit.
1064 	 *
1065 	 * The TSC MSR is exposed read-only. Writes are disallowed as
1066 	 * that will impact the host TSC.  If the guest does a write
1067 	 * the "use TSC offsetting" execution control is enabled and the
1068 	 * difference between the host TSC and the guest TSC is written
1069 	 * into the TSC offset in the VMCS.
1070 	 *
1071 	 * Guest TSC_AUX support is enabled if any of guest RDPID and/or
1072 	 * guest RDTSCP support are enabled (since, as per Table 2-2 in SDM
1073 	 * volume 4, TSC_AUX is supported if any of RDPID and/or RDTSCP are
1074 	 * supported). If guest TSC_AUX support is enabled, TSC_AUX is
1075 	 * exposed read-only so that the VMM can do one fewer MSR read per
1076 	 * exit than if this register were exposed read-write; the guest
1077 	 * restore value can be updated during guest writes (expected to be
1078 	 * rare) instead of during all exits (common).
1079 	 */
1080 	if (guest_msr_rw(vmx, MSR_GSBASE) ||
1081 	    guest_msr_rw(vmx, MSR_FSBASE) ||
1082 	    guest_msr_rw(vmx, MSR_SYSENTER_CS_MSR) ||
1083 	    guest_msr_rw(vmx, MSR_SYSENTER_ESP_MSR) ||
1084 	    guest_msr_rw(vmx, MSR_SYSENTER_EIP_MSR) ||
1085 	    guest_msr_rw(vmx, MSR_EFER) ||
1086 	    guest_msr_ro(vmx, MSR_TSC) ||
1087 	    ((cap_rdpid || cap_rdtscp) && guest_msr_ro(vmx, MSR_TSC_AUX)))
1088 		panic("vmx_init: error setting guest msr access");
1089 
1090 	vpid_alloc(vpid, VM_MAXCPU);
1091 
1092 	if (virtual_interrupt_delivery) {
1093 		error = vm_map_mmio(vm, DEFAULT_APIC_BASE, PAGE_SIZE,
1094 		    APIC_ACCESS_ADDRESS);
1095 		/* XXX this should really return an error to the caller */
1096 		KASSERT(error == 0, ("vm_map_mmio(apicbase) error %d", error));
1097 	}
1098 
1099 	maxcpus = vm_get_maxcpus(vm);
1100 	for (i = 0; i < maxcpus; i++) {
1101 		vmcs = &vmx->vmcs[i];
1102 		vmcs->identifier = vmx_revision();
1103 		error = vmclear(vmcs);
1104 		if (error != 0) {
1105 			panic("vmx_init: vmclear error %d on vcpu %d\n",
1106 			      error, i);
1107 		}
1108 
1109 		vmx_msr_guest_init(vmx, i);
1110 
1111 		error = vmcs_init(vmcs);
1112 		KASSERT(error == 0, ("vmcs_init error %d", error));
1113 
1114 		VMPTRLD(vmcs);
1115 		error = 0;
1116 		error += vmwrite(VMCS_HOST_RSP, (u_long)&vmx->ctx[i]);
1117 		error += vmwrite(VMCS_EPTP, vmx->eptp);
1118 		error += vmwrite(VMCS_PIN_BASED_CTLS, pinbased_ctls);
1119 		error += vmwrite(VMCS_PRI_PROC_BASED_CTLS, procbased_ctls);
1120 		error += vmwrite(VMCS_SEC_PROC_BASED_CTLS, procbased_ctls2);
1121 		error += vmwrite(VMCS_EXIT_CTLS, exit_ctls);
1122 		error += vmwrite(VMCS_ENTRY_CTLS, entry_ctls);
1123 		error += vmwrite(VMCS_MSR_BITMAP, vtophys(vmx->msr_bitmap));
1124 		error += vmwrite(VMCS_VPID, vpid[i]);
1125 
1126 		if (guest_l1d_flush && !guest_l1d_flush_sw) {
1127 			vmcs_write(VMCS_ENTRY_MSR_LOAD, pmap_kextract(
1128 			    (vm_offset_t)&msr_load_list[0]));
1129 			vmcs_write(VMCS_ENTRY_MSR_LOAD_COUNT,
1130 			    nitems(msr_load_list));
1131 			vmcs_write(VMCS_EXIT_MSR_STORE, 0);
1132 			vmcs_write(VMCS_EXIT_MSR_STORE_COUNT, 0);
1133 		}
1134 
1135 		/* exception bitmap */
1136 		if (vcpu_trace_exceptions(vm, i))
1137 			exc_bitmap = 0xffffffff;
1138 		else
1139 			exc_bitmap = 1 << IDT_MC;
1140 		error += vmwrite(VMCS_EXCEPTION_BITMAP, exc_bitmap);
1141 
1142 		vmx->ctx[i].guest_dr6 = DBREG_DR6_RESERVED1;
1143 		error += vmwrite(VMCS_GUEST_DR7, DBREG_DR7_RESERVED1);
1144 
1145 		if (tpr_shadowing) {
1146 			error += vmwrite(VMCS_VIRTUAL_APIC,
1147 			    vtophys(&vmx->apic_page[i]));
1148 		}
1149 
1150 		if (virtual_interrupt_delivery) {
1151 			error += vmwrite(VMCS_APIC_ACCESS, APIC_ACCESS_ADDRESS);
1152 			error += vmwrite(VMCS_EOI_EXIT0, 0);
1153 			error += vmwrite(VMCS_EOI_EXIT1, 0);
1154 			error += vmwrite(VMCS_EOI_EXIT2, 0);
1155 			error += vmwrite(VMCS_EOI_EXIT3, 0);
1156 		}
1157 		if (posted_interrupts) {
1158 			error += vmwrite(VMCS_PIR_VECTOR, pirvec);
1159 			error += vmwrite(VMCS_PIR_DESC,
1160 			    vtophys(&vmx->pir_desc[i]));
1161 		}
1162 		VMCLEAR(vmcs);
1163 		KASSERT(error == 0, ("vmx_init: error customizing the vmcs"));
1164 
1165 		vmx->cap[i].set = 0;
1166 		vmx->cap[i].set |= cap_rdpid != 0 ? 1 << VM_CAP_RDPID : 0;
1167 		vmx->cap[i].set |= cap_rdtscp != 0 ? 1 << VM_CAP_RDTSCP : 0;
1168 		vmx->cap[i].proc_ctls = procbased_ctls;
1169 		vmx->cap[i].proc_ctls2 = procbased_ctls2;
1170 		vmx->cap[i].exc_bitmap = exc_bitmap;
1171 
1172 		vmx->state[i].nextrip = ~0;
1173 		vmx->state[i].lastcpu = NOCPU;
1174 		vmx->state[i].vpid = vpid[i];
1175 
1176 		/*
1177 		 * Set up the CR0/4 shadows, and init the read shadow
1178 		 * to the power-on register value from the Intel Sys Arch.
1179 		 *  CR0 - 0x60000010
1180 		 *  CR4 - 0
1181 		 */
1182 		error = vmx_setup_cr0_shadow(vmcs, 0x60000010);
1183 		if (error != 0)
1184 			panic("vmx_setup_cr0_shadow %d", error);
1185 
1186 		error = vmx_setup_cr4_shadow(vmcs, 0);
1187 		if (error != 0)
1188 			panic("vmx_setup_cr4_shadow %d", error);
1189 
1190 		vmx->ctx[i].pmap = pmap;
1191 	}
1192 
1193 	return (vmx);
1194 }
1195 
1196 static int
1197 vmx_handle_cpuid(struct vm *vm, int vcpu, struct vmxctx *vmxctx)
1198 {
1199 	int handled;
1200 
1201 	handled = x86_emulate_cpuid(vm, vcpu, (uint64_t *)&vmxctx->guest_rax,
1202 	    (uint64_t *)&vmxctx->guest_rbx, (uint64_t *)&vmxctx->guest_rcx,
1203 	    (uint64_t *)&vmxctx->guest_rdx);
1204 	return (handled);
1205 }
1206 
1207 static __inline void
1208 vmx_run_trace(struct vmx *vmx, int vcpu)
1209 {
1210 #ifdef KTR
1211 	VCPU_CTR1(vmx->vm, vcpu, "Resume execution at %#lx", vmcs_guest_rip());
1212 #endif
1213 }
1214 
1215 static __inline void
1216 vmx_exit_trace(struct vmx *vmx, int vcpu, uint64_t rip, uint32_t exit_reason,
1217 	       int handled)
1218 {
1219 #ifdef KTR
1220 	VCPU_CTR3(vmx->vm, vcpu, "%s %s vmexit at 0x%0lx",
1221 		 handled ? "handled" : "unhandled",
1222 		 exit_reason_to_str(exit_reason), rip);
1223 #endif
1224 }
1225 
1226 static __inline void
1227 vmx_astpending_trace(struct vmx *vmx, int vcpu, uint64_t rip)
1228 {
1229 #ifdef KTR
1230 	VCPU_CTR1(vmx->vm, vcpu, "astpending vmexit at 0x%0lx", rip);
1231 #endif
1232 }
1233 
1234 static VMM_STAT_INTEL(VCPU_INVVPID_SAVED, "Number of vpid invalidations saved");
1235 static VMM_STAT_INTEL(VCPU_INVVPID_DONE, "Number of vpid invalidations done");
1236 
1237 /*
1238  * Invalidate guest mappings identified by its vpid from the TLB.
1239  */
1240 static __inline void
1241 vmx_invvpid(struct vmx *vmx, int vcpu, pmap_t pmap, int running)
1242 {
1243 	struct vmxstate *vmxstate;
1244 	struct invvpid_desc invvpid_desc;
1245 
1246 	vmxstate = &vmx->state[vcpu];
1247 	if (vmxstate->vpid == 0)
1248 		return;
1249 
1250 	if (!running) {
1251 		/*
1252 		 * Set the 'lastcpu' to an invalid host cpu.
1253 		 *
1254 		 * This will invalidate TLB entries tagged with the vcpu's
1255 		 * vpid the next time it runs via vmx_set_pcpu_defaults().
1256 		 */
1257 		vmxstate->lastcpu = NOCPU;
1258 		return;
1259 	}
1260 
1261 	KASSERT(curthread->td_critnest > 0, ("%s: vcpu %d running outside "
1262 	    "critical section", __func__, vcpu));
1263 
1264 	/*
1265 	 * Invalidate all mappings tagged with 'vpid'
1266 	 *
1267 	 * We do this because this vcpu was executing on a different host
1268 	 * cpu when it last ran. We do not track whether it invalidated
1269 	 * mappings associated with its 'vpid' during that run. So we must
1270 	 * assume that the mappings associated with 'vpid' on 'curcpu' are
1271 	 * stale and invalidate them.
1272 	 *
1273 	 * Note that we incur this penalty only when the scheduler chooses to
1274 	 * move the thread associated with this vcpu between host cpus.
1275 	 *
1276 	 * Note also that this will invalidate mappings tagged with 'vpid'
1277 	 * for "all" EP4TAs.
1278 	 */
1279 	if (atomic_load_long(&pmap->pm_eptgen) == vmx->eptgen[curcpu]) {
1280 		invvpid_desc._res1 = 0;
1281 		invvpid_desc._res2 = 0;
1282 		invvpid_desc.vpid = vmxstate->vpid;
1283 		invvpid_desc.linear_addr = 0;
1284 		invvpid(INVVPID_TYPE_SINGLE_CONTEXT, invvpid_desc);
1285 		vmm_stat_incr(vmx->vm, vcpu, VCPU_INVVPID_DONE, 1);
1286 	} else {
1287 		/*
1288 		 * The invvpid can be skipped if an invept is going to
1289 		 * be performed before entering the guest. The invept
1290 		 * will invalidate combined mappings tagged with
1291 		 * 'vmx->eptp' for all vpids.
1292 		 */
1293 		vmm_stat_incr(vmx->vm, vcpu, VCPU_INVVPID_SAVED, 1);
1294 	}
1295 }
1296 
1297 static void
1298 vmx_set_pcpu_defaults(struct vmx *vmx, int vcpu, pmap_t pmap)
1299 {
1300 	struct vmxstate *vmxstate;
1301 
1302 	vmxstate = &vmx->state[vcpu];
1303 	if (vmxstate->lastcpu == curcpu)
1304 		return;
1305 
1306 	vmxstate->lastcpu = curcpu;
1307 
1308 	vmm_stat_incr(vmx->vm, vcpu, VCPU_MIGRATIONS, 1);
1309 
1310 	vmcs_write(VMCS_HOST_TR_BASE, vmm_get_host_trbase());
1311 	vmcs_write(VMCS_HOST_GDTR_BASE, vmm_get_host_gdtrbase());
1312 	vmcs_write(VMCS_HOST_GS_BASE, vmm_get_host_gsbase());
1313 	vmx_invvpid(vmx, vcpu, pmap, 1);
1314 }
1315 
1316 /*
1317  * We depend on 'procbased_ctls' to have the Interrupt Window Exiting bit set.
1318  */
1319 CTASSERT((PROCBASED_CTLS_ONE_SETTING & PROCBASED_INT_WINDOW_EXITING) != 0);
1320 
1321 static void __inline
1322 vmx_set_int_window_exiting(struct vmx *vmx, int vcpu)
1323 {
1324 
1325 	if ((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) == 0) {
1326 		vmx->cap[vcpu].proc_ctls |= PROCBASED_INT_WINDOW_EXITING;
1327 		vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls);
1328 		VCPU_CTR0(vmx->vm, vcpu, "Enabling interrupt window exiting");
1329 	}
1330 }
1331 
1332 static void __inline
1333 vmx_clear_int_window_exiting(struct vmx *vmx, int vcpu)
1334 {
1335 
1336 	KASSERT((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0,
1337 	    ("intr_window_exiting not set: %#x", vmx->cap[vcpu].proc_ctls));
1338 	vmx->cap[vcpu].proc_ctls &= ~PROCBASED_INT_WINDOW_EXITING;
1339 	vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls);
1340 	VCPU_CTR0(vmx->vm, vcpu, "Disabling interrupt window exiting");
1341 }
1342 
1343 static void __inline
1344 vmx_set_nmi_window_exiting(struct vmx *vmx, int vcpu)
1345 {
1346 
1347 	if ((vmx->cap[vcpu].proc_ctls & PROCBASED_NMI_WINDOW_EXITING) == 0) {
1348 		vmx->cap[vcpu].proc_ctls |= PROCBASED_NMI_WINDOW_EXITING;
1349 		vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls);
1350 		VCPU_CTR0(vmx->vm, vcpu, "Enabling NMI window exiting");
1351 	}
1352 }
1353 
1354 static void __inline
1355 vmx_clear_nmi_window_exiting(struct vmx *vmx, int vcpu)
1356 {
1357 
1358 	KASSERT((vmx->cap[vcpu].proc_ctls & PROCBASED_NMI_WINDOW_EXITING) != 0,
1359 	    ("nmi_window_exiting not set %#x", vmx->cap[vcpu].proc_ctls));
1360 	vmx->cap[vcpu].proc_ctls &= ~PROCBASED_NMI_WINDOW_EXITING;
1361 	vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls);
1362 	VCPU_CTR0(vmx->vm, vcpu, "Disabling NMI window exiting");
1363 }
1364 
1365 int
1366 vmx_set_tsc_offset(struct vmx *vmx, int vcpu, uint64_t offset)
1367 {
1368 	int error;
1369 
1370 	if ((vmx->cap[vcpu].proc_ctls & PROCBASED_TSC_OFFSET) == 0) {
1371 		vmx->cap[vcpu].proc_ctls |= PROCBASED_TSC_OFFSET;
1372 		vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vmx->cap[vcpu].proc_ctls);
1373 		VCPU_CTR0(vmx->vm, vcpu, "Enabling TSC offsetting");
1374 	}
1375 
1376 	error = vmwrite(VMCS_TSC_OFFSET, offset);
1377 #ifdef BHYVE_SNAPSHOT
1378 	if (error == 0)
1379 		error = vm_set_tsc_offset(vmx->vm, vcpu, offset);
1380 #endif
1381 	return (error);
1382 }
1383 
1384 #define	NMI_BLOCKING	(VMCS_INTERRUPTIBILITY_NMI_BLOCKING |		\
1385 			 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING)
1386 #define	HWINTR_BLOCKING	(VMCS_INTERRUPTIBILITY_STI_BLOCKING |		\
1387 			 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING)
1388 
1389 static void
1390 vmx_inject_nmi(struct vmx *vmx, int vcpu)
1391 {
1392 	uint32_t gi, info;
1393 
1394 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1395 	KASSERT((gi & NMI_BLOCKING) == 0, ("vmx_inject_nmi: invalid guest "
1396 	    "interruptibility-state %#x", gi));
1397 
1398 	info = vmcs_read(VMCS_ENTRY_INTR_INFO);
1399 	KASSERT((info & VMCS_INTR_VALID) == 0, ("vmx_inject_nmi: invalid "
1400 	    "VM-entry interruption information %#x", info));
1401 
1402 	/*
1403 	 * Inject the virtual NMI. The vector must be the NMI IDT entry
1404 	 * or the VMCS entry check will fail.
1405 	 */
1406 	info = IDT_NMI | VMCS_INTR_T_NMI | VMCS_INTR_VALID;
1407 	vmcs_write(VMCS_ENTRY_INTR_INFO, info);
1408 
1409 	VCPU_CTR0(vmx->vm, vcpu, "Injecting vNMI");
1410 
1411 	/* Clear the request */
1412 	vm_nmi_clear(vmx->vm, vcpu);
1413 }
1414 
1415 static void
1416 vmx_inject_interrupts(struct vmx *vmx, int vcpu, struct vlapic *vlapic,
1417     uint64_t guestrip)
1418 {
1419 	int vector, need_nmi_exiting, extint_pending;
1420 	uint64_t rflags, entryinfo;
1421 	uint32_t gi, info;
1422 
1423 	if (vmx->state[vcpu].nextrip != guestrip) {
1424 		gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1425 		if (gi & HWINTR_BLOCKING) {
1426 			VCPU_CTR2(vmx->vm, vcpu, "Guest interrupt blocking "
1427 			    "cleared due to rip change: %#lx/%#lx",
1428 			    vmx->state[vcpu].nextrip, guestrip);
1429 			gi &= ~HWINTR_BLOCKING;
1430 			vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi);
1431 		}
1432 	}
1433 
1434 	if (vm_entry_intinfo(vmx->vm, vcpu, &entryinfo)) {
1435 		KASSERT((entryinfo & VMCS_INTR_VALID) != 0, ("%s: entry "
1436 		    "intinfo is not valid: %#lx", __func__, entryinfo));
1437 
1438 		info = vmcs_read(VMCS_ENTRY_INTR_INFO);
1439 		KASSERT((info & VMCS_INTR_VALID) == 0, ("%s: cannot inject "
1440 		     "pending exception: %#lx/%#x", __func__, entryinfo, info));
1441 
1442 		info = entryinfo;
1443 		vector = info & 0xff;
1444 		if (vector == IDT_BP || vector == IDT_OF) {
1445 			/*
1446 			 * VT-x requires #BP and #OF to be injected as software
1447 			 * exceptions.
1448 			 */
1449 			info &= ~VMCS_INTR_T_MASK;
1450 			info |= VMCS_INTR_T_SWEXCEPTION;
1451 		}
1452 
1453 		if (info & VMCS_INTR_DEL_ERRCODE)
1454 			vmcs_write(VMCS_ENTRY_EXCEPTION_ERROR, entryinfo >> 32);
1455 
1456 		vmcs_write(VMCS_ENTRY_INTR_INFO, info);
1457 	}
1458 
1459 	if (vm_nmi_pending(vmx->vm, vcpu)) {
1460 		/*
1461 		 * If there are no conditions blocking NMI injection then
1462 		 * inject it directly here otherwise enable "NMI window
1463 		 * exiting" to inject it as soon as we can.
1464 		 *
1465 		 * We also check for STI_BLOCKING because some implementations
1466 		 * don't allow NMI injection in this case. If we are running
1467 		 * on a processor that doesn't have this restriction it will
1468 		 * immediately exit and the NMI will be injected in the
1469 		 * "NMI window exiting" handler.
1470 		 */
1471 		need_nmi_exiting = 1;
1472 		gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1473 		if ((gi & (HWINTR_BLOCKING | NMI_BLOCKING)) == 0) {
1474 			info = vmcs_read(VMCS_ENTRY_INTR_INFO);
1475 			if ((info & VMCS_INTR_VALID) == 0) {
1476 				vmx_inject_nmi(vmx, vcpu);
1477 				need_nmi_exiting = 0;
1478 			} else {
1479 				VCPU_CTR1(vmx->vm, vcpu, "Cannot inject NMI "
1480 				    "due to VM-entry intr info %#x", info);
1481 			}
1482 		} else {
1483 			VCPU_CTR1(vmx->vm, vcpu, "Cannot inject NMI due to "
1484 			    "Guest Interruptibility-state %#x", gi);
1485 		}
1486 
1487 		if (need_nmi_exiting)
1488 			vmx_set_nmi_window_exiting(vmx, vcpu);
1489 	}
1490 
1491 	extint_pending = vm_extint_pending(vmx->vm, vcpu);
1492 
1493 	if (!extint_pending && virtual_interrupt_delivery) {
1494 		vmx_inject_pir(vlapic);
1495 		return;
1496 	}
1497 
1498 	/*
1499 	 * If interrupt-window exiting is already in effect then don't bother
1500 	 * checking for pending interrupts. This is just an optimization and
1501 	 * not needed for correctness.
1502 	 */
1503 	if ((vmx->cap[vcpu].proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0) {
1504 		VCPU_CTR0(vmx->vm, vcpu, "Skip interrupt injection due to "
1505 		    "pending int_window_exiting");
1506 		return;
1507 	}
1508 
1509 	if (!extint_pending) {
1510 		/* Ask the local apic for a vector to inject */
1511 		if (!vlapic_pending_intr(vlapic, &vector))
1512 			return;
1513 
1514 		/*
1515 		 * From the Intel SDM, Volume 3, Section "Maskable
1516 		 * Hardware Interrupts":
1517 		 * - maskable interrupt vectors [16,255] can be delivered
1518 		 *   through the local APIC.
1519 		*/
1520 		KASSERT(vector >= 16 && vector <= 255,
1521 		    ("invalid vector %d from local APIC", vector));
1522 	} else {
1523 		/* Ask the legacy pic for a vector to inject */
1524 		vatpic_pending_intr(vmx->vm, &vector);
1525 
1526 		/*
1527 		 * From the Intel SDM, Volume 3, Section "Maskable
1528 		 * Hardware Interrupts":
1529 		 * - maskable interrupt vectors [0,255] can be delivered
1530 		 *   through the INTR pin.
1531 		 */
1532 		KASSERT(vector >= 0 && vector <= 255,
1533 		    ("invalid vector %d from INTR", vector));
1534 	}
1535 
1536 	/* Check RFLAGS.IF and the interruptibility state of the guest */
1537 	rflags = vmcs_read(VMCS_GUEST_RFLAGS);
1538 	if ((rflags & PSL_I) == 0) {
1539 		VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to "
1540 		    "rflags %#lx", vector, rflags);
1541 		goto cantinject;
1542 	}
1543 
1544 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1545 	if (gi & HWINTR_BLOCKING) {
1546 		VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to "
1547 		    "Guest Interruptibility-state %#x", vector, gi);
1548 		goto cantinject;
1549 	}
1550 
1551 	info = vmcs_read(VMCS_ENTRY_INTR_INFO);
1552 	if (info & VMCS_INTR_VALID) {
1553 		/*
1554 		 * This is expected and could happen for multiple reasons:
1555 		 * - A vectoring VM-entry was aborted due to astpending
1556 		 * - A VM-exit happened during event injection.
1557 		 * - An exception was injected above.
1558 		 * - An NMI was injected above or after "NMI window exiting"
1559 		 */
1560 		VCPU_CTR2(vmx->vm, vcpu, "Cannot inject vector %d due to "
1561 		    "VM-entry intr info %#x", vector, info);
1562 		goto cantinject;
1563 	}
1564 
1565 	/* Inject the interrupt */
1566 	info = VMCS_INTR_T_HWINTR | VMCS_INTR_VALID;
1567 	info |= vector;
1568 	vmcs_write(VMCS_ENTRY_INTR_INFO, info);
1569 
1570 	if (!extint_pending) {
1571 		/* Update the Local APIC ISR */
1572 		vlapic_intr_accepted(vlapic, vector);
1573 	} else {
1574 		vm_extint_clear(vmx->vm, vcpu);
1575 		vatpic_intr_accepted(vmx->vm, vector);
1576 
1577 		/*
1578 		 * After we accepted the current ExtINT the PIC may
1579 		 * have posted another one.  If that is the case, set
1580 		 * the Interrupt Window Exiting execution control so
1581 		 * we can inject that one too.
1582 		 *
1583 		 * Also, interrupt window exiting allows us to inject any
1584 		 * pending APIC vector that was preempted by the ExtINT
1585 		 * as soon as possible. This applies both for the software
1586 		 * emulated vlapic and the hardware assisted virtual APIC.
1587 		 */
1588 		vmx_set_int_window_exiting(vmx, vcpu);
1589 	}
1590 
1591 	VCPU_CTR1(vmx->vm, vcpu, "Injecting hwintr at vector %d", vector);
1592 
1593 	return;
1594 
1595 cantinject:
1596 	/*
1597 	 * Set the Interrupt Window Exiting execution control so we can inject
1598 	 * the interrupt as soon as blocking condition goes away.
1599 	 */
1600 	vmx_set_int_window_exiting(vmx, vcpu);
1601 }
1602 
1603 /*
1604  * If the Virtual NMIs execution control is '1' then the logical processor
1605  * tracks virtual-NMI blocking in the Guest Interruptibility-state field of
1606  * the VMCS. An IRET instruction in VMX non-root operation will remove any
1607  * virtual-NMI blocking.
1608  *
1609  * This unblocking occurs even if the IRET causes a fault. In this case the
1610  * hypervisor needs to restore virtual-NMI blocking before resuming the guest.
1611  */
1612 static void
1613 vmx_restore_nmi_blocking(struct vmx *vmx, int vcpuid)
1614 {
1615 	uint32_t gi;
1616 
1617 	VCPU_CTR0(vmx->vm, vcpuid, "Restore Virtual-NMI blocking");
1618 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1619 	gi |= VMCS_INTERRUPTIBILITY_NMI_BLOCKING;
1620 	vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi);
1621 }
1622 
1623 static void
1624 vmx_clear_nmi_blocking(struct vmx *vmx, int vcpuid)
1625 {
1626 	uint32_t gi;
1627 
1628 	VCPU_CTR0(vmx->vm, vcpuid, "Clear Virtual-NMI blocking");
1629 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1630 	gi &= ~VMCS_INTERRUPTIBILITY_NMI_BLOCKING;
1631 	vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi);
1632 }
1633 
1634 static void
1635 vmx_assert_nmi_blocking(struct vmx *vmx, int vcpuid)
1636 {
1637 	uint32_t gi;
1638 
1639 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1640 	KASSERT(gi & VMCS_INTERRUPTIBILITY_NMI_BLOCKING,
1641 	    ("NMI blocking is not in effect %#x", gi));
1642 }
1643 
1644 static int
1645 vmx_emulate_xsetbv(struct vmx *vmx, int vcpu, struct vm_exit *vmexit)
1646 {
1647 	struct vmxctx *vmxctx;
1648 	uint64_t xcrval;
1649 	const struct xsave_limits *limits;
1650 
1651 	vmxctx = &vmx->ctx[vcpu];
1652 	limits = vmm_get_xsave_limits();
1653 
1654 	/*
1655 	 * Note that the processor raises a GP# fault on its own if
1656 	 * xsetbv is executed for CPL != 0, so we do not have to
1657 	 * emulate that fault here.
1658 	 */
1659 
1660 	/* Only xcr0 is supported. */
1661 	if (vmxctx->guest_rcx != 0) {
1662 		vm_inject_gp(vmx->vm, vcpu);
1663 		return (HANDLED);
1664 	}
1665 
1666 	/* We only handle xcr0 if both the host and guest have XSAVE enabled. */
1667 	if (!limits->xsave_enabled || !(vmcs_read(VMCS_GUEST_CR4) & CR4_XSAVE)) {
1668 		vm_inject_ud(vmx->vm, vcpu);
1669 		return (HANDLED);
1670 	}
1671 
1672 	xcrval = vmxctx->guest_rdx << 32 | (vmxctx->guest_rax & 0xffffffff);
1673 	if ((xcrval & ~limits->xcr0_allowed) != 0) {
1674 		vm_inject_gp(vmx->vm, vcpu);
1675 		return (HANDLED);
1676 	}
1677 
1678 	if (!(xcrval & XFEATURE_ENABLED_X87)) {
1679 		vm_inject_gp(vmx->vm, vcpu);
1680 		return (HANDLED);
1681 	}
1682 
1683 	/* AVX (YMM_Hi128) requires SSE. */
1684 	if (xcrval & XFEATURE_ENABLED_AVX &&
1685 	    (xcrval & XFEATURE_AVX) != XFEATURE_AVX) {
1686 		vm_inject_gp(vmx->vm, vcpu);
1687 		return (HANDLED);
1688 	}
1689 
1690 	/*
1691 	 * AVX512 requires base AVX (YMM_Hi128) as well as OpMask,
1692 	 * ZMM_Hi256, and Hi16_ZMM.
1693 	 */
1694 	if (xcrval & XFEATURE_AVX512 &&
1695 	    (xcrval & (XFEATURE_AVX512 | XFEATURE_AVX)) !=
1696 	    (XFEATURE_AVX512 | XFEATURE_AVX)) {
1697 		vm_inject_gp(vmx->vm, vcpu);
1698 		return (HANDLED);
1699 	}
1700 
1701 	/*
1702 	 * Intel MPX requires both bound register state flags to be
1703 	 * set.
1704 	 */
1705 	if (((xcrval & XFEATURE_ENABLED_BNDREGS) != 0) !=
1706 	    ((xcrval & XFEATURE_ENABLED_BNDCSR) != 0)) {
1707 		vm_inject_gp(vmx->vm, vcpu);
1708 		return (HANDLED);
1709 	}
1710 
1711 	/*
1712 	 * This runs "inside" vmrun() with the guest's FPU state, so
1713 	 * modifying xcr0 directly modifies the guest's xcr0, not the
1714 	 * host's.
1715 	 */
1716 	load_xcr(0, xcrval);
1717 	return (HANDLED);
1718 }
1719 
1720 static uint64_t
1721 vmx_get_guest_reg(struct vmx *vmx, int vcpu, int ident)
1722 {
1723 	const struct vmxctx *vmxctx;
1724 
1725 	vmxctx = &vmx->ctx[vcpu];
1726 
1727 	switch (ident) {
1728 	case 0:
1729 		return (vmxctx->guest_rax);
1730 	case 1:
1731 		return (vmxctx->guest_rcx);
1732 	case 2:
1733 		return (vmxctx->guest_rdx);
1734 	case 3:
1735 		return (vmxctx->guest_rbx);
1736 	case 4:
1737 		return (vmcs_read(VMCS_GUEST_RSP));
1738 	case 5:
1739 		return (vmxctx->guest_rbp);
1740 	case 6:
1741 		return (vmxctx->guest_rsi);
1742 	case 7:
1743 		return (vmxctx->guest_rdi);
1744 	case 8:
1745 		return (vmxctx->guest_r8);
1746 	case 9:
1747 		return (vmxctx->guest_r9);
1748 	case 10:
1749 		return (vmxctx->guest_r10);
1750 	case 11:
1751 		return (vmxctx->guest_r11);
1752 	case 12:
1753 		return (vmxctx->guest_r12);
1754 	case 13:
1755 		return (vmxctx->guest_r13);
1756 	case 14:
1757 		return (vmxctx->guest_r14);
1758 	case 15:
1759 		return (vmxctx->guest_r15);
1760 	default:
1761 		panic("invalid vmx register %d", ident);
1762 	}
1763 }
1764 
1765 static void
1766 vmx_set_guest_reg(struct vmx *vmx, int vcpu, int ident, uint64_t regval)
1767 {
1768 	struct vmxctx *vmxctx;
1769 
1770 	vmxctx = &vmx->ctx[vcpu];
1771 
1772 	switch (ident) {
1773 	case 0:
1774 		vmxctx->guest_rax = regval;
1775 		break;
1776 	case 1:
1777 		vmxctx->guest_rcx = regval;
1778 		break;
1779 	case 2:
1780 		vmxctx->guest_rdx = regval;
1781 		break;
1782 	case 3:
1783 		vmxctx->guest_rbx = regval;
1784 		break;
1785 	case 4:
1786 		vmcs_write(VMCS_GUEST_RSP, regval);
1787 		break;
1788 	case 5:
1789 		vmxctx->guest_rbp = regval;
1790 		break;
1791 	case 6:
1792 		vmxctx->guest_rsi = regval;
1793 		break;
1794 	case 7:
1795 		vmxctx->guest_rdi = regval;
1796 		break;
1797 	case 8:
1798 		vmxctx->guest_r8 = regval;
1799 		break;
1800 	case 9:
1801 		vmxctx->guest_r9 = regval;
1802 		break;
1803 	case 10:
1804 		vmxctx->guest_r10 = regval;
1805 		break;
1806 	case 11:
1807 		vmxctx->guest_r11 = regval;
1808 		break;
1809 	case 12:
1810 		vmxctx->guest_r12 = regval;
1811 		break;
1812 	case 13:
1813 		vmxctx->guest_r13 = regval;
1814 		break;
1815 	case 14:
1816 		vmxctx->guest_r14 = regval;
1817 		break;
1818 	case 15:
1819 		vmxctx->guest_r15 = regval;
1820 		break;
1821 	default:
1822 		panic("invalid vmx register %d", ident);
1823 	}
1824 }
1825 
1826 static int
1827 vmx_emulate_cr0_access(struct vmx *vmx, int vcpu, uint64_t exitqual)
1828 {
1829 	uint64_t crval, regval;
1830 
1831 	/* We only handle mov to %cr0 at this time */
1832 	if ((exitqual & 0xf0) != 0x00)
1833 		return (UNHANDLED);
1834 
1835 	regval = vmx_get_guest_reg(vmx, vcpu, (exitqual >> 8) & 0xf);
1836 
1837 	vmcs_write(VMCS_CR0_SHADOW, regval);
1838 
1839 	crval = regval | cr0_ones_mask;
1840 	crval &= ~cr0_zeros_mask;
1841 	vmcs_write(VMCS_GUEST_CR0, crval);
1842 
1843 	if (regval & CR0_PG) {
1844 		uint64_t efer, entry_ctls;
1845 
1846 		/*
1847 		 * If CR0.PG is 1 and EFER.LME is 1 then EFER.LMA and
1848 		 * the "IA-32e mode guest" bit in VM-entry control must be
1849 		 * equal.
1850 		 */
1851 		efer = vmcs_read(VMCS_GUEST_IA32_EFER);
1852 		if (efer & EFER_LME) {
1853 			efer |= EFER_LMA;
1854 			vmcs_write(VMCS_GUEST_IA32_EFER, efer);
1855 			entry_ctls = vmcs_read(VMCS_ENTRY_CTLS);
1856 			entry_ctls |= VM_ENTRY_GUEST_LMA;
1857 			vmcs_write(VMCS_ENTRY_CTLS, entry_ctls);
1858 		}
1859 	}
1860 
1861 	return (HANDLED);
1862 }
1863 
1864 static int
1865 vmx_emulate_cr4_access(struct vmx *vmx, int vcpu, uint64_t exitqual)
1866 {
1867 	uint64_t crval, regval;
1868 
1869 	/* We only handle mov to %cr4 at this time */
1870 	if ((exitqual & 0xf0) != 0x00)
1871 		return (UNHANDLED);
1872 
1873 	regval = vmx_get_guest_reg(vmx, vcpu, (exitqual >> 8) & 0xf);
1874 
1875 	vmcs_write(VMCS_CR4_SHADOW, regval);
1876 
1877 	crval = regval | cr4_ones_mask;
1878 	crval &= ~cr4_zeros_mask;
1879 	vmcs_write(VMCS_GUEST_CR4, crval);
1880 
1881 	return (HANDLED);
1882 }
1883 
1884 static int
1885 vmx_emulate_cr8_access(struct vmx *vmx, int vcpu, uint64_t exitqual)
1886 {
1887 	struct vlapic *vlapic;
1888 	uint64_t cr8;
1889 	int regnum;
1890 
1891 	/* We only handle mov %cr8 to/from a register at this time. */
1892 	if ((exitqual & 0xe0) != 0x00) {
1893 		return (UNHANDLED);
1894 	}
1895 
1896 	vlapic = vm_lapic(vmx->vm, vcpu);
1897 	regnum = (exitqual >> 8) & 0xf;
1898 	if (exitqual & 0x10) {
1899 		cr8 = vlapic_get_cr8(vlapic);
1900 		vmx_set_guest_reg(vmx, vcpu, regnum, cr8);
1901 	} else {
1902 		cr8 = vmx_get_guest_reg(vmx, vcpu, regnum);
1903 		vlapic_set_cr8(vlapic, cr8);
1904 	}
1905 
1906 	return (HANDLED);
1907 }
1908 
1909 /*
1910  * From section "Guest Register State" in the Intel SDM: CPL = SS.DPL
1911  */
1912 static int
1913 vmx_cpl(void)
1914 {
1915 	uint32_t ssar;
1916 
1917 	ssar = vmcs_read(VMCS_GUEST_SS_ACCESS_RIGHTS);
1918 	return ((ssar >> 5) & 0x3);
1919 }
1920 
1921 static enum vm_cpu_mode
1922 vmx_cpu_mode(void)
1923 {
1924 	uint32_t csar;
1925 
1926 	if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LMA) {
1927 		csar = vmcs_read(VMCS_GUEST_CS_ACCESS_RIGHTS);
1928 		if (csar & 0x2000)
1929 			return (CPU_MODE_64BIT);	/* CS.L = 1 */
1930 		else
1931 			return (CPU_MODE_COMPATIBILITY);
1932 	} else if (vmcs_read(VMCS_GUEST_CR0) & CR0_PE) {
1933 		return (CPU_MODE_PROTECTED);
1934 	} else {
1935 		return (CPU_MODE_REAL);
1936 	}
1937 }
1938 
1939 static enum vm_paging_mode
1940 vmx_paging_mode(void)
1941 {
1942 	uint64_t cr4;
1943 
1944 	if (!(vmcs_read(VMCS_GUEST_CR0) & CR0_PG))
1945 		return (PAGING_MODE_FLAT);
1946 	cr4 = vmcs_read(VMCS_GUEST_CR4);
1947 	if (!(cr4 & CR4_PAE))
1948 		return (PAGING_MODE_32);
1949 	if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LME) {
1950 		if (!(cr4 & CR4_LA57))
1951 			return (PAGING_MODE_64);
1952 		return (PAGING_MODE_64_LA57);
1953 	} else
1954 		return (PAGING_MODE_PAE);
1955 }
1956 
1957 static uint64_t
1958 inout_str_index(struct vmx *vmx, int vcpuid, int in)
1959 {
1960 	uint64_t val;
1961 	int error;
1962 	enum vm_reg_name reg;
1963 
1964 	reg = in ? VM_REG_GUEST_RDI : VM_REG_GUEST_RSI;
1965 	error = vmx_getreg(vmx, vcpuid, reg, &val);
1966 	KASSERT(error == 0, ("%s: vmx_getreg error %d", __func__, error));
1967 	return (val);
1968 }
1969 
1970 static uint64_t
1971 inout_str_count(struct vmx *vmx, int vcpuid, int rep)
1972 {
1973 	uint64_t val;
1974 	int error;
1975 
1976 	if (rep) {
1977 		error = vmx_getreg(vmx, vcpuid, VM_REG_GUEST_RCX, &val);
1978 		KASSERT(!error, ("%s: vmx_getreg error %d", __func__, error));
1979 	} else {
1980 		val = 1;
1981 	}
1982 	return (val);
1983 }
1984 
1985 static int
1986 inout_str_addrsize(uint32_t inst_info)
1987 {
1988 	uint32_t size;
1989 
1990 	size = (inst_info >> 7) & 0x7;
1991 	switch (size) {
1992 	case 0:
1993 		return (2);	/* 16 bit */
1994 	case 1:
1995 		return (4);	/* 32 bit */
1996 	case 2:
1997 		return (8);	/* 64 bit */
1998 	default:
1999 		panic("%s: invalid size encoding %d", __func__, size);
2000 	}
2001 }
2002 
2003 static void
2004 inout_str_seginfo(struct vmx *vmx, int vcpuid, uint32_t inst_info, int in,
2005     struct vm_inout_str *vis)
2006 {
2007 	int error, s;
2008 
2009 	if (in) {
2010 		vis->seg_name = VM_REG_GUEST_ES;
2011 	} else {
2012 		s = (inst_info >> 15) & 0x7;
2013 		vis->seg_name = vm_segment_name(s);
2014 	}
2015 
2016 	error = vmx_getdesc(vmx, vcpuid, vis->seg_name, &vis->seg_desc);
2017 	KASSERT(error == 0, ("%s: vmx_getdesc error %d", __func__, error));
2018 }
2019 
2020 static void
2021 vmx_paging_info(struct vm_guest_paging *paging)
2022 {
2023 	paging->cr3 = vmcs_guest_cr3();
2024 	paging->cpl = vmx_cpl();
2025 	paging->cpu_mode = vmx_cpu_mode();
2026 	paging->paging_mode = vmx_paging_mode();
2027 }
2028 
2029 static void
2030 vmexit_inst_emul(struct vm_exit *vmexit, uint64_t gpa, uint64_t gla)
2031 {
2032 	struct vm_guest_paging *paging;
2033 	uint32_t csar;
2034 
2035 	paging = &vmexit->u.inst_emul.paging;
2036 
2037 	vmexit->exitcode = VM_EXITCODE_INST_EMUL;
2038 	vmexit->inst_length = 0;
2039 	vmexit->u.inst_emul.gpa = gpa;
2040 	vmexit->u.inst_emul.gla = gla;
2041 	vmx_paging_info(paging);
2042 	switch (paging->cpu_mode) {
2043 	case CPU_MODE_REAL:
2044 		vmexit->u.inst_emul.cs_base = vmcs_read(VMCS_GUEST_CS_BASE);
2045 		vmexit->u.inst_emul.cs_d = 0;
2046 		break;
2047 	case CPU_MODE_PROTECTED:
2048 	case CPU_MODE_COMPATIBILITY:
2049 		vmexit->u.inst_emul.cs_base = vmcs_read(VMCS_GUEST_CS_BASE);
2050 		csar = vmcs_read(VMCS_GUEST_CS_ACCESS_RIGHTS);
2051 		vmexit->u.inst_emul.cs_d = SEG_DESC_DEF32(csar);
2052 		break;
2053 	default:
2054 		vmexit->u.inst_emul.cs_base = 0;
2055 		vmexit->u.inst_emul.cs_d = 0;
2056 		break;
2057 	}
2058 	vie_init(&vmexit->u.inst_emul.vie, NULL, 0);
2059 }
2060 
2061 static int
2062 ept_fault_type(uint64_t ept_qual)
2063 {
2064 	int fault_type;
2065 
2066 	if (ept_qual & EPT_VIOLATION_DATA_WRITE)
2067 		fault_type = VM_PROT_WRITE;
2068 	else if (ept_qual & EPT_VIOLATION_INST_FETCH)
2069 		fault_type = VM_PROT_EXECUTE;
2070 	else
2071 		fault_type= VM_PROT_READ;
2072 
2073 	return (fault_type);
2074 }
2075 
2076 static bool
2077 ept_emulation_fault(uint64_t ept_qual)
2078 {
2079 	int read, write;
2080 
2081 	/* EPT fault on an instruction fetch doesn't make sense here */
2082 	if (ept_qual & EPT_VIOLATION_INST_FETCH)
2083 		return (false);
2084 
2085 	/* EPT fault must be a read fault or a write fault */
2086 	read = ept_qual & EPT_VIOLATION_DATA_READ ? 1 : 0;
2087 	write = ept_qual & EPT_VIOLATION_DATA_WRITE ? 1 : 0;
2088 	if ((read | write) == 0)
2089 		return (false);
2090 
2091 	/*
2092 	 * The EPT violation must have been caused by accessing a
2093 	 * guest-physical address that is a translation of a guest-linear
2094 	 * address.
2095 	 */
2096 	if ((ept_qual & EPT_VIOLATION_GLA_VALID) == 0 ||
2097 	    (ept_qual & EPT_VIOLATION_XLAT_VALID) == 0) {
2098 		return (false);
2099 	}
2100 
2101 	return (true);
2102 }
2103 
2104 static __inline int
2105 apic_access_virtualization(struct vmx *vmx, int vcpuid)
2106 {
2107 	uint32_t proc_ctls2;
2108 
2109 	proc_ctls2 = vmx->cap[vcpuid].proc_ctls2;
2110 	return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) ? 1 : 0);
2111 }
2112 
2113 static __inline int
2114 x2apic_virtualization(struct vmx *vmx, int vcpuid)
2115 {
2116 	uint32_t proc_ctls2;
2117 
2118 	proc_ctls2 = vmx->cap[vcpuid].proc_ctls2;
2119 	return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_X2APIC_MODE) ? 1 : 0);
2120 }
2121 
2122 static int
2123 vmx_handle_apic_write(struct vmx *vmx, int vcpuid, struct vlapic *vlapic,
2124     uint64_t qual)
2125 {
2126 	int error, handled, offset;
2127 	uint32_t *apic_regs, vector;
2128 	bool retu;
2129 
2130 	handled = HANDLED;
2131 	offset = APIC_WRITE_OFFSET(qual);
2132 
2133 	if (!apic_access_virtualization(vmx, vcpuid)) {
2134 		/*
2135 		 * In general there should not be any APIC write VM-exits
2136 		 * unless APIC-access virtualization is enabled.
2137 		 *
2138 		 * However self-IPI virtualization can legitimately trigger
2139 		 * an APIC-write VM-exit so treat it specially.
2140 		 */
2141 		if (x2apic_virtualization(vmx, vcpuid) &&
2142 		    offset == APIC_OFFSET_SELF_IPI) {
2143 			apic_regs = (uint32_t *)(vlapic->apic_page);
2144 			vector = apic_regs[APIC_OFFSET_SELF_IPI / 4];
2145 			vlapic_self_ipi_handler(vlapic, vector);
2146 			return (HANDLED);
2147 		} else
2148 			return (UNHANDLED);
2149 	}
2150 
2151 	switch (offset) {
2152 	case APIC_OFFSET_ID:
2153 		vlapic_id_write_handler(vlapic);
2154 		break;
2155 	case APIC_OFFSET_LDR:
2156 		vlapic_ldr_write_handler(vlapic);
2157 		break;
2158 	case APIC_OFFSET_DFR:
2159 		vlapic_dfr_write_handler(vlapic);
2160 		break;
2161 	case APIC_OFFSET_SVR:
2162 		vlapic_svr_write_handler(vlapic);
2163 		break;
2164 	case APIC_OFFSET_ESR:
2165 		vlapic_esr_write_handler(vlapic);
2166 		break;
2167 	case APIC_OFFSET_ICR_LOW:
2168 		retu = false;
2169 		error = vlapic_icrlo_write_handler(vlapic, &retu);
2170 		if (error != 0 || retu)
2171 			handled = UNHANDLED;
2172 		break;
2173 	case APIC_OFFSET_CMCI_LVT:
2174 	case APIC_OFFSET_TIMER_LVT ... APIC_OFFSET_ERROR_LVT:
2175 		vlapic_lvt_write_handler(vlapic, offset);
2176 		break;
2177 	case APIC_OFFSET_TIMER_ICR:
2178 		vlapic_icrtmr_write_handler(vlapic);
2179 		break;
2180 	case APIC_OFFSET_TIMER_DCR:
2181 		vlapic_dcr_write_handler(vlapic);
2182 		break;
2183 	default:
2184 		handled = UNHANDLED;
2185 		break;
2186 	}
2187 	return (handled);
2188 }
2189 
2190 static bool
2191 apic_access_fault(struct vmx *vmx, int vcpuid, uint64_t gpa)
2192 {
2193 
2194 	if (apic_access_virtualization(vmx, vcpuid) &&
2195 	    (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE))
2196 		return (true);
2197 	else
2198 		return (false);
2199 }
2200 
2201 static int
2202 vmx_handle_apic_access(struct vmx *vmx, int vcpuid, struct vm_exit *vmexit)
2203 {
2204 	uint64_t qual;
2205 	int access_type, offset, allowed;
2206 
2207 	if (!apic_access_virtualization(vmx, vcpuid))
2208 		return (UNHANDLED);
2209 
2210 	qual = vmexit->u.vmx.exit_qualification;
2211 	access_type = APIC_ACCESS_TYPE(qual);
2212 	offset = APIC_ACCESS_OFFSET(qual);
2213 
2214 	allowed = 0;
2215 	if (access_type == 0) {
2216 		/*
2217 		 * Read data access to the following registers is expected.
2218 		 */
2219 		switch (offset) {
2220 		case APIC_OFFSET_APR:
2221 		case APIC_OFFSET_PPR:
2222 		case APIC_OFFSET_RRR:
2223 		case APIC_OFFSET_CMCI_LVT:
2224 		case APIC_OFFSET_TIMER_CCR:
2225 			allowed = 1;
2226 			break;
2227 		default:
2228 			break;
2229 		}
2230 	} else if (access_type == 1) {
2231 		/*
2232 		 * Write data access to the following registers is expected.
2233 		 */
2234 		switch (offset) {
2235 		case APIC_OFFSET_VER:
2236 		case APIC_OFFSET_APR:
2237 		case APIC_OFFSET_PPR:
2238 		case APIC_OFFSET_RRR:
2239 		case APIC_OFFSET_ISR0 ... APIC_OFFSET_ISR7:
2240 		case APIC_OFFSET_TMR0 ... APIC_OFFSET_TMR7:
2241 		case APIC_OFFSET_IRR0 ... APIC_OFFSET_IRR7:
2242 		case APIC_OFFSET_CMCI_LVT:
2243 		case APIC_OFFSET_TIMER_CCR:
2244 			allowed = 1;
2245 			break;
2246 		default:
2247 			break;
2248 		}
2249 	}
2250 
2251 	if (allowed) {
2252 		vmexit_inst_emul(vmexit, DEFAULT_APIC_BASE + offset,
2253 		    VIE_INVALID_GLA);
2254 	}
2255 
2256 	/*
2257 	 * Regardless of whether the APIC-access is allowed this handler
2258 	 * always returns UNHANDLED:
2259 	 * - if the access is allowed then it is handled by emulating the
2260 	 *   instruction that caused the VM-exit (outside the critical section)
2261 	 * - if the access is not allowed then it will be converted to an
2262 	 *   exitcode of VM_EXITCODE_VMX and will be dealt with in userland.
2263 	 */
2264 	return (UNHANDLED);
2265 }
2266 
2267 static enum task_switch_reason
2268 vmx_task_switch_reason(uint64_t qual)
2269 {
2270 	int reason;
2271 
2272 	reason = (qual >> 30) & 0x3;
2273 	switch (reason) {
2274 	case 0:
2275 		return (TSR_CALL);
2276 	case 1:
2277 		return (TSR_IRET);
2278 	case 2:
2279 		return (TSR_JMP);
2280 	case 3:
2281 		return (TSR_IDT_GATE);
2282 	default:
2283 		panic("%s: invalid reason %d", __func__, reason);
2284 	}
2285 }
2286 
2287 static int
2288 emulate_wrmsr(struct vmx *vmx, int vcpuid, u_int num, uint64_t val, bool *retu)
2289 {
2290 	int error;
2291 
2292 	if (lapic_msr(num))
2293 		error = lapic_wrmsr(vmx->vm, vcpuid, num, val, retu);
2294 	else
2295 		error = vmx_wrmsr(vmx, vcpuid, num, val, retu);
2296 
2297 	return (error);
2298 }
2299 
2300 static int
2301 emulate_rdmsr(struct vmx *vmx, int vcpuid, u_int num, bool *retu)
2302 {
2303 	struct vmxctx *vmxctx;
2304 	uint64_t result;
2305 	uint32_t eax, edx;
2306 	int error;
2307 
2308 	if (lapic_msr(num))
2309 		error = lapic_rdmsr(vmx->vm, vcpuid, num, &result, retu);
2310 	else
2311 		error = vmx_rdmsr(vmx, vcpuid, num, &result, retu);
2312 
2313 	if (error == 0) {
2314 		eax = result;
2315 		vmxctx = &vmx->ctx[vcpuid];
2316 		error = vmxctx_setreg(vmxctx, VM_REG_GUEST_RAX, eax);
2317 		KASSERT(error == 0, ("vmxctx_setreg(rax) error %d", error));
2318 
2319 		edx = result >> 32;
2320 		error = vmxctx_setreg(vmxctx, VM_REG_GUEST_RDX, edx);
2321 		KASSERT(error == 0, ("vmxctx_setreg(rdx) error %d", error));
2322 	}
2323 
2324 	return (error);
2325 }
2326 
2327 static int
2328 vmx_exit_process(struct vmx *vmx, int vcpu, struct vm_exit *vmexit)
2329 {
2330 	int error, errcode, errcode_valid, handled, in;
2331 	struct vmxctx *vmxctx;
2332 	struct vlapic *vlapic;
2333 	struct vm_inout_str *vis;
2334 	struct vm_task_switch *ts;
2335 	uint32_t eax, ecx, edx, idtvec_info, idtvec_err, intr_info, inst_info;
2336 	uint32_t intr_type, intr_vec, reason;
2337 	uint64_t exitintinfo, qual, gpa;
2338 	bool retu;
2339 
2340 	CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_VIRTUAL_NMI) != 0);
2341 	CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_NMI_EXITING) != 0);
2342 
2343 	handled = UNHANDLED;
2344 	vmxctx = &vmx->ctx[vcpu];
2345 
2346 	qual = vmexit->u.vmx.exit_qualification;
2347 	reason = vmexit->u.vmx.exit_reason;
2348 	vmexit->exitcode = VM_EXITCODE_BOGUS;
2349 
2350 	vmm_stat_incr(vmx->vm, vcpu, VMEXIT_COUNT, 1);
2351 	SDT_PROBE3(vmm, vmx, exit, entry, vmx, vcpu, vmexit);
2352 
2353 	/*
2354 	 * VM-entry failures during or after loading guest state.
2355 	 *
2356 	 * These VM-exits are uncommon but must be handled specially
2357 	 * as most VM-exit fields are not populated as usual.
2358 	 */
2359 	if (__predict_false(reason == EXIT_REASON_MCE_DURING_ENTRY)) {
2360 		VCPU_CTR0(vmx->vm, vcpu, "Handling MCE during VM-entry");
2361 		__asm __volatile("int $18");
2362 		return (1);
2363 	}
2364 
2365 	/*
2366 	 * VM exits that can be triggered during event delivery need to
2367 	 * be handled specially by re-injecting the event if the IDT
2368 	 * vectoring information field's valid bit is set.
2369 	 *
2370 	 * See "Information for VM Exits During Event Delivery" in Intel SDM
2371 	 * for details.
2372 	 */
2373 	idtvec_info = vmcs_idt_vectoring_info();
2374 	if (idtvec_info & VMCS_IDT_VEC_VALID) {
2375 		idtvec_info &= ~(1 << 12); /* clear undefined bit */
2376 		exitintinfo = idtvec_info;
2377 		if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) {
2378 			idtvec_err = vmcs_idt_vectoring_err();
2379 			exitintinfo |= (uint64_t)idtvec_err << 32;
2380 		}
2381 		error = vm_exit_intinfo(vmx->vm, vcpu, exitintinfo);
2382 		KASSERT(error == 0, ("%s: vm_set_intinfo error %d",
2383 		    __func__, error));
2384 
2385 		/*
2386 		 * If 'virtual NMIs' are being used and the VM-exit
2387 		 * happened while injecting an NMI during the previous
2388 		 * VM-entry, then clear "blocking by NMI" in the
2389 		 * Guest Interruptibility-State so the NMI can be
2390 		 * reinjected on the subsequent VM-entry.
2391 		 *
2392 		 * However, if the NMI was being delivered through a task
2393 		 * gate, then the new task must start execution with NMIs
2394 		 * blocked so don't clear NMI blocking in this case.
2395 		 */
2396 		intr_type = idtvec_info & VMCS_INTR_T_MASK;
2397 		if (intr_type == VMCS_INTR_T_NMI) {
2398 			if (reason != EXIT_REASON_TASK_SWITCH)
2399 				vmx_clear_nmi_blocking(vmx, vcpu);
2400 			else
2401 				vmx_assert_nmi_blocking(vmx, vcpu);
2402 		}
2403 
2404 		/*
2405 		 * Update VM-entry instruction length if the event being
2406 		 * delivered was a software interrupt or software exception.
2407 		 */
2408 		if (intr_type == VMCS_INTR_T_SWINTR ||
2409 		    intr_type == VMCS_INTR_T_PRIV_SWEXCEPTION ||
2410 		    intr_type == VMCS_INTR_T_SWEXCEPTION) {
2411 			vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length);
2412 		}
2413 	}
2414 
2415 	switch (reason) {
2416 	case EXIT_REASON_TASK_SWITCH:
2417 		ts = &vmexit->u.task_switch;
2418 		ts->tsssel = qual & 0xffff;
2419 		ts->reason = vmx_task_switch_reason(qual);
2420 		ts->ext = 0;
2421 		ts->errcode_valid = 0;
2422 		vmx_paging_info(&ts->paging);
2423 		/*
2424 		 * If the task switch was due to a CALL, JMP, IRET, software
2425 		 * interrupt (INT n) or software exception (INT3, INTO),
2426 		 * then the saved %rip references the instruction that caused
2427 		 * the task switch. The instruction length field in the VMCS
2428 		 * is valid in this case.
2429 		 *
2430 		 * In all other cases (e.g., NMI, hardware exception) the
2431 		 * saved %rip is one that would have been saved in the old TSS
2432 		 * had the task switch completed normally so the instruction
2433 		 * length field is not needed in this case and is explicitly
2434 		 * set to 0.
2435 		 */
2436 		if (ts->reason == TSR_IDT_GATE) {
2437 			KASSERT(idtvec_info & VMCS_IDT_VEC_VALID,
2438 			    ("invalid idtvec_info %#x for IDT task switch",
2439 			    idtvec_info));
2440 			intr_type = idtvec_info & VMCS_INTR_T_MASK;
2441 			if (intr_type != VMCS_INTR_T_SWINTR &&
2442 			    intr_type != VMCS_INTR_T_SWEXCEPTION &&
2443 			    intr_type != VMCS_INTR_T_PRIV_SWEXCEPTION) {
2444 				/* Task switch triggered by external event */
2445 				ts->ext = 1;
2446 				vmexit->inst_length = 0;
2447 				if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) {
2448 					ts->errcode_valid = 1;
2449 					ts->errcode = vmcs_idt_vectoring_err();
2450 				}
2451 			}
2452 		}
2453 		vmexit->exitcode = VM_EXITCODE_TASK_SWITCH;
2454 		SDT_PROBE4(vmm, vmx, exit, taskswitch, vmx, vcpu, vmexit, ts);
2455 		VCPU_CTR4(vmx->vm, vcpu, "task switch reason %d, tss 0x%04x, "
2456 		    "%s errcode 0x%016lx", ts->reason, ts->tsssel,
2457 		    ts->ext ? "external" : "internal",
2458 		    ((uint64_t)ts->errcode << 32) | ts->errcode_valid);
2459 		break;
2460 	case EXIT_REASON_CR_ACCESS:
2461 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CR_ACCESS, 1);
2462 		SDT_PROBE4(vmm, vmx, exit, craccess, vmx, vcpu, vmexit, qual);
2463 		switch (qual & 0xf) {
2464 		case 0:
2465 			handled = vmx_emulate_cr0_access(vmx, vcpu, qual);
2466 			break;
2467 		case 4:
2468 			handled = vmx_emulate_cr4_access(vmx, vcpu, qual);
2469 			break;
2470 		case 8:
2471 			handled = vmx_emulate_cr8_access(vmx, vcpu, qual);
2472 			break;
2473 		}
2474 		break;
2475 	case EXIT_REASON_RDMSR:
2476 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_RDMSR, 1);
2477 		retu = false;
2478 		ecx = vmxctx->guest_rcx;
2479 		VCPU_CTR1(vmx->vm, vcpu, "rdmsr 0x%08x", ecx);
2480 		SDT_PROBE4(vmm, vmx, exit, rdmsr, vmx, vcpu, vmexit, ecx);
2481 		error = emulate_rdmsr(vmx, vcpu, ecx, &retu);
2482 		if (error) {
2483 			vmexit->exitcode = VM_EXITCODE_RDMSR;
2484 			vmexit->u.msr.code = ecx;
2485 		} else if (!retu) {
2486 			handled = HANDLED;
2487 		} else {
2488 			/* Return to userspace with a valid exitcode */
2489 			KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS,
2490 			    ("emulate_rdmsr retu with bogus exitcode"));
2491 		}
2492 		break;
2493 	case EXIT_REASON_WRMSR:
2494 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_WRMSR, 1);
2495 		retu = false;
2496 		eax = vmxctx->guest_rax;
2497 		ecx = vmxctx->guest_rcx;
2498 		edx = vmxctx->guest_rdx;
2499 		VCPU_CTR2(vmx->vm, vcpu, "wrmsr 0x%08x value 0x%016lx",
2500 		    ecx, (uint64_t)edx << 32 | eax);
2501 		SDT_PROBE5(vmm, vmx, exit, wrmsr, vmx, vmexit, vcpu, ecx,
2502 		    (uint64_t)edx << 32 | eax);
2503 		error = emulate_wrmsr(vmx, vcpu, ecx,
2504 		    (uint64_t)edx << 32 | eax, &retu);
2505 		if (error) {
2506 			vmexit->exitcode = VM_EXITCODE_WRMSR;
2507 			vmexit->u.msr.code = ecx;
2508 			vmexit->u.msr.wval = (uint64_t)edx << 32 | eax;
2509 		} else if (!retu) {
2510 			handled = HANDLED;
2511 		} else {
2512 			/* Return to userspace with a valid exitcode */
2513 			KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS,
2514 			    ("emulate_wrmsr retu with bogus exitcode"));
2515 		}
2516 		break;
2517 	case EXIT_REASON_HLT:
2518 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_HLT, 1);
2519 		SDT_PROBE3(vmm, vmx, exit, halt, vmx, vcpu, vmexit);
2520 		vmexit->exitcode = VM_EXITCODE_HLT;
2521 		vmexit->u.hlt.rflags = vmcs_read(VMCS_GUEST_RFLAGS);
2522 		if (virtual_interrupt_delivery)
2523 			vmexit->u.hlt.intr_status =
2524 			    vmcs_read(VMCS_GUEST_INTR_STATUS);
2525 		else
2526 			vmexit->u.hlt.intr_status = 0;
2527 		break;
2528 	case EXIT_REASON_MTF:
2529 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_MTRAP, 1);
2530 		SDT_PROBE3(vmm, vmx, exit, mtrap, vmx, vcpu, vmexit);
2531 		vmexit->exitcode = VM_EXITCODE_MTRAP;
2532 		vmexit->inst_length = 0;
2533 		break;
2534 	case EXIT_REASON_PAUSE:
2535 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_PAUSE, 1);
2536 		SDT_PROBE3(vmm, vmx, exit, pause, vmx, vcpu, vmexit);
2537 		vmexit->exitcode = VM_EXITCODE_PAUSE;
2538 		break;
2539 	case EXIT_REASON_INTR_WINDOW:
2540 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INTR_WINDOW, 1);
2541 		SDT_PROBE3(vmm, vmx, exit, intrwindow, vmx, vcpu, vmexit);
2542 		vmx_clear_int_window_exiting(vmx, vcpu);
2543 		return (1);
2544 	case EXIT_REASON_EXT_INTR:
2545 		/*
2546 		 * External interrupts serve only to cause VM exits and allow
2547 		 * the host interrupt handler to run.
2548 		 *
2549 		 * If this external interrupt triggers a virtual interrupt
2550 		 * to a VM, then that state will be recorded by the
2551 		 * host interrupt handler in the VM's softc. We will inject
2552 		 * this virtual interrupt during the subsequent VM enter.
2553 		 */
2554 		intr_info = vmcs_read(VMCS_EXIT_INTR_INFO);
2555 		SDT_PROBE4(vmm, vmx, exit, interrupt,
2556 		    vmx, vcpu, vmexit, intr_info);
2557 
2558 		/*
2559 		 * XXX: Ignore this exit if VMCS_INTR_VALID is not set.
2560 		 * This appears to be a bug in VMware Fusion?
2561 		 */
2562 		if (!(intr_info & VMCS_INTR_VALID))
2563 			return (1);
2564 		KASSERT((intr_info & VMCS_INTR_VALID) != 0 &&
2565 		    (intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_HWINTR,
2566 		    ("VM exit interruption info invalid: %#x", intr_info));
2567 		vmx_trigger_hostintr(intr_info & 0xff);
2568 
2569 		/*
2570 		 * This is special. We want to treat this as an 'handled'
2571 		 * VM-exit but not increment the instruction pointer.
2572 		 */
2573 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EXTINT, 1);
2574 		return (1);
2575 	case EXIT_REASON_NMI_WINDOW:
2576 		SDT_PROBE3(vmm, vmx, exit, nmiwindow, vmx, vcpu, vmexit);
2577 		/* Exit to allow the pending virtual NMI to be injected */
2578 		if (vm_nmi_pending(vmx->vm, vcpu))
2579 			vmx_inject_nmi(vmx, vcpu);
2580 		vmx_clear_nmi_window_exiting(vmx, vcpu);
2581 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_NMI_WINDOW, 1);
2582 		return (1);
2583 	case EXIT_REASON_INOUT:
2584 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INOUT, 1);
2585 		vmexit->exitcode = VM_EXITCODE_INOUT;
2586 		vmexit->u.inout.bytes = (qual & 0x7) + 1;
2587 		vmexit->u.inout.in = in = (qual & 0x8) ? 1 : 0;
2588 		vmexit->u.inout.string = (qual & 0x10) ? 1 : 0;
2589 		vmexit->u.inout.rep = (qual & 0x20) ? 1 : 0;
2590 		vmexit->u.inout.port = (uint16_t)(qual >> 16);
2591 		vmexit->u.inout.eax = (uint32_t)(vmxctx->guest_rax);
2592 		if (vmexit->u.inout.string) {
2593 			inst_info = vmcs_read(VMCS_EXIT_INSTRUCTION_INFO);
2594 			vmexit->exitcode = VM_EXITCODE_INOUT_STR;
2595 			vis = &vmexit->u.inout_str;
2596 			vmx_paging_info(&vis->paging);
2597 			vis->rflags = vmcs_read(VMCS_GUEST_RFLAGS);
2598 			vis->cr0 = vmcs_read(VMCS_GUEST_CR0);
2599 			vis->index = inout_str_index(vmx, vcpu, in);
2600 			vis->count = inout_str_count(vmx, vcpu, vis->inout.rep);
2601 			vis->addrsize = inout_str_addrsize(inst_info);
2602 			inout_str_seginfo(vmx, vcpu, inst_info, in, vis);
2603 		}
2604 		SDT_PROBE3(vmm, vmx, exit, inout, vmx, vcpu, vmexit);
2605 		break;
2606 	case EXIT_REASON_CPUID:
2607 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_CPUID, 1);
2608 		SDT_PROBE3(vmm, vmx, exit, cpuid, vmx, vcpu, vmexit);
2609 		handled = vmx_handle_cpuid(vmx->vm, vcpu, vmxctx);
2610 		break;
2611 	case EXIT_REASON_EXCEPTION:
2612 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_EXCEPTION, 1);
2613 		intr_info = vmcs_read(VMCS_EXIT_INTR_INFO);
2614 		KASSERT((intr_info & VMCS_INTR_VALID) != 0,
2615 		    ("VM exit interruption info invalid: %#x", intr_info));
2616 
2617 		intr_vec = intr_info & 0xff;
2618 		intr_type = intr_info & VMCS_INTR_T_MASK;
2619 
2620 		/*
2621 		 * If Virtual NMIs control is 1 and the VM-exit is due to a
2622 		 * fault encountered during the execution of IRET then we must
2623 		 * restore the state of "virtual-NMI blocking" before resuming
2624 		 * the guest.
2625 		 *
2626 		 * See "Resuming Guest Software after Handling an Exception".
2627 		 * See "Information for VM Exits Due to Vectored Events".
2628 		 */
2629 		if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 &&
2630 		    (intr_vec != IDT_DF) &&
2631 		    (intr_info & EXIT_QUAL_NMIUDTI) != 0)
2632 			vmx_restore_nmi_blocking(vmx, vcpu);
2633 
2634 		/*
2635 		 * The NMI has already been handled in vmx_exit_handle_nmi().
2636 		 */
2637 		if (intr_type == VMCS_INTR_T_NMI)
2638 			return (1);
2639 
2640 		/*
2641 		 * Call the machine check handler by hand. Also don't reflect
2642 		 * the machine check back into the guest.
2643 		 */
2644 		if (intr_vec == IDT_MC) {
2645 			VCPU_CTR0(vmx->vm, vcpu, "Vectoring to MCE handler");
2646 			__asm __volatile("int $18");
2647 			return (1);
2648 		}
2649 
2650 		/*
2651 		 * If the hypervisor has requested user exits for
2652 		 * debug exceptions, bounce them out to userland.
2653 		 */
2654 		if (intr_type == VMCS_INTR_T_SWEXCEPTION && intr_vec == IDT_BP &&
2655 		    (vmx->cap[vcpu].set & (1 << VM_CAP_BPT_EXIT))) {
2656 			vmexit->exitcode = VM_EXITCODE_BPT;
2657 			vmexit->u.bpt.inst_length = vmexit->inst_length;
2658 			vmexit->inst_length = 0;
2659 			break;
2660 		}
2661 
2662 		if (intr_vec == IDT_PF) {
2663 			error = vmxctx_setreg(vmxctx, VM_REG_GUEST_CR2, qual);
2664 			KASSERT(error == 0, ("%s: vmxctx_setreg(cr2) error %d",
2665 			    __func__, error));
2666 		}
2667 
2668 		/*
2669 		 * Software exceptions exhibit trap-like behavior. This in
2670 		 * turn requires populating the VM-entry instruction length
2671 		 * so that the %rip in the trap frame is past the INT3/INTO
2672 		 * instruction.
2673 		 */
2674 		if (intr_type == VMCS_INTR_T_SWEXCEPTION)
2675 			vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length);
2676 
2677 		/* Reflect all other exceptions back into the guest */
2678 		errcode_valid = errcode = 0;
2679 		if (intr_info & VMCS_INTR_DEL_ERRCODE) {
2680 			errcode_valid = 1;
2681 			errcode = vmcs_read(VMCS_EXIT_INTR_ERRCODE);
2682 		}
2683 		VCPU_CTR2(vmx->vm, vcpu, "Reflecting exception %d/%#x into "
2684 		    "the guest", intr_vec, errcode);
2685 		SDT_PROBE5(vmm, vmx, exit, exception,
2686 		    vmx, vcpu, vmexit, intr_vec, errcode);
2687 		error = vm_inject_exception(vmx->vm, vcpu, intr_vec,
2688 		    errcode_valid, errcode, 0);
2689 		KASSERT(error == 0, ("%s: vm_inject_exception error %d",
2690 		    __func__, error));
2691 		return (1);
2692 
2693 	case EXIT_REASON_EPT_FAULT:
2694 		/*
2695 		 * If 'gpa' lies within the address space allocated to
2696 		 * memory then this must be a nested page fault otherwise
2697 		 * this must be an instruction that accesses MMIO space.
2698 		 */
2699 		gpa = vmcs_gpa();
2700 		if (vm_mem_allocated(vmx->vm, vcpu, gpa) ||
2701 		    apic_access_fault(vmx, vcpu, gpa)) {
2702 			vmexit->exitcode = VM_EXITCODE_PAGING;
2703 			vmexit->inst_length = 0;
2704 			vmexit->u.paging.gpa = gpa;
2705 			vmexit->u.paging.fault_type = ept_fault_type(qual);
2706 			vmm_stat_incr(vmx->vm, vcpu, VMEXIT_NESTED_FAULT, 1);
2707 			SDT_PROBE5(vmm, vmx, exit, nestedfault,
2708 			    vmx, vcpu, vmexit, gpa, qual);
2709 		} else if (ept_emulation_fault(qual)) {
2710 			vmexit_inst_emul(vmexit, gpa, vmcs_gla());
2711 			vmm_stat_incr(vmx->vm, vcpu, VMEXIT_INST_EMUL, 1);
2712 			SDT_PROBE4(vmm, vmx, exit, mmiofault,
2713 			    vmx, vcpu, vmexit, gpa);
2714 		}
2715 		/*
2716 		 * If Virtual NMIs control is 1 and the VM-exit is due to an
2717 		 * EPT fault during the execution of IRET then we must restore
2718 		 * the state of "virtual-NMI blocking" before resuming.
2719 		 *
2720 		 * See description of "NMI unblocking due to IRET" in
2721 		 * "Exit Qualification for EPT Violations".
2722 		 */
2723 		if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 &&
2724 		    (qual & EXIT_QUAL_NMIUDTI) != 0)
2725 			vmx_restore_nmi_blocking(vmx, vcpu);
2726 		break;
2727 	case EXIT_REASON_VIRTUALIZED_EOI:
2728 		vmexit->exitcode = VM_EXITCODE_IOAPIC_EOI;
2729 		vmexit->u.ioapic_eoi.vector = qual & 0xFF;
2730 		SDT_PROBE3(vmm, vmx, exit, eoi, vmx, vcpu, vmexit);
2731 		vmexit->inst_length = 0;	/* trap-like */
2732 		break;
2733 	case EXIT_REASON_APIC_ACCESS:
2734 		SDT_PROBE3(vmm, vmx, exit, apicaccess, vmx, vcpu, vmexit);
2735 		handled = vmx_handle_apic_access(vmx, vcpu, vmexit);
2736 		break;
2737 	case EXIT_REASON_APIC_WRITE:
2738 		/*
2739 		 * APIC-write VM exit is trap-like so the %rip is already
2740 		 * pointing to the next instruction.
2741 		 */
2742 		vmexit->inst_length = 0;
2743 		vlapic = vm_lapic(vmx->vm, vcpu);
2744 		SDT_PROBE4(vmm, vmx, exit, apicwrite,
2745 		    vmx, vcpu, vmexit, vlapic);
2746 		handled = vmx_handle_apic_write(vmx, vcpu, vlapic, qual);
2747 		break;
2748 	case EXIT_REASON_XSETBV:
2749 		SDT_PROBE3(vmm, vmx, exit, xsetbv, vmx, vcpu, vmexit);
2750 		handled = vmx_emulate_xsetbv(vmx, vcpu, vmexit);
2751 		break;
2752 	case EXIT_REASON_MONITOR:
2753 		SDT_PROBE3(vmm, vmx, exit, monitor, vmx, vcpu, vmexit);
2754 		vmexit->exitcode = VM_EXITCODE_MONITOR;
2755 		break;
2756 	case EXIT_REASON_MWAIT:
2757 		SDT_PROBE3(vmm, vmx, exit, mwait, vmx, vcpu, vmexit);
2758 		vmexit->exitcode = VM_EXITCODE_MWAIT;
2759 		break;
2760 	case EXIT_REASON_TPR:
2761 		vlapic = vm_lapic(vmx->vm, vcpu);
2762 		vlapic_sync_tpr(vlapic);
2763 		vmexit->inst_length = 0;
2764 		handled = HANDLED;
2765 		break;
2766 	case EXIT_REASON_VMCALL:
2767 	case EXIT_REASON_VMCLEAR:
2768 	case EXIT_REASON_VMLAUNCH:
2769 	case EXIT_REASON_VMPTRLD:
2770 	case EXIT_REASON_VMPTRST:
2771 	case EXIT_REASON_VMREAD:
2772 	case EXIT_REASON_VMRESUME:
2773 	case EXIT_REASON_VMWRITE:
2774 	case EXIT_REASON_VMXOFF:
2775 	case EXIT_REASON_VMXON:
2776 		SDT_PROBE3(vmm, vmx, exit, vminsn, vmx, vcpu, vmexit);
2777 		vmexit->exitcode = VM_EXITCODE_VMINSN;
2778 		break;
2779 	default:
2780 		SDT_PROBE4(vmm, vmx, exit, unknown,
2781 		    vmx, vcpu, vmexit, reason);
2782 		vmm_stat_incr(vmx->vm, vcpu, VMEXIT_UNKNOWN, 1);
2783 		break;
2784 	}
2785 
2786 	if (handled) {
2787 		/*
2788 		 * It is possible that control is returned to userland
2789 		 * even though we were able to handle the VM exit in the
2790 		 * kernel.
2791 		 *
2792 		 * In such a case we want to make sure that the userland
2793 		 * restarts guest execution at the instruction *after*
2794 		 * the one we just processed. Therefore we update the
2795 		 * guest rip in the VMCS and in 'vmexit'.
2796 		 */
2797 		vmexit->rip += vmexit->inst_length;
2798 		vmexit->inst_length = 0;
2799 		vmcs_write(VMCS_GUEST_RIP, vmexit->rip);
2800 	} else {
2801 		if (vmexit->exitcode == VM_EXITCODE_BOGUS) {
2802 			/*
2803 			 * If this VM exit was not claimed by anybody then
2804 			 * treat it as a generic VMX exit.
2805 			 */
2806 			vmexit->exitcode = VM_EXITCODE_VMX;
2807 			vmexit->u.vmx.status = VM_SUCCESS;
2808 			vmexit->u.vmx.inst_type = 0;
2809 			vmexit->u.vmx.inst_error = 0;
2810 		} else {
2811 			/*
2812 			 * The exitcode and collateral have been populated.
2813 			 * The VM exit will be processed further in userland.
2814 			 */
2815 		}
2816 	}
2817 
2818 	SDT_PROBE4(vmm, vmx, exit, return,
2819 	    vmx, vcpu, vmexit, handled);
2820 	return (handled);
2821 }
2822 
2823 static __inline void
2824 vmx_exit_inst_error(struct vmxctx *vmxctx, int rc, struct vm_exit *vmexit)
2825 {
2826 
2827 	KASSERT(vmxctx->inst_fail_status != VM_SUCCESS,
2828 	    ("vmx_exit_inst_error: invalid inst_fail_status %d",
2829 	    vmxctx->inst_fail_status));
2830 
2831 	vmexit->inst_length = 0;
2832 	vmexit->exitcode = VM_EXITCODE_VMX;
2833 	vmexit->u.vmx.status = vmxctx->inst_fail_status;
2834 	vmexit->u.vmx.inst_error = vmcs_instruction_error();
2835 	vmexit->u.vmx.exit_reason = ~0;
2836 	vmexit->u.vmx.exit_qualification = ~0;
2837 
2838 	switch (rc) {
2839 	case VMX_VMRESUME_ERROR:
2840 	case VMX_VMLAUNCH_ERROR:
2841 		vmexit->u.vmx.inst_type = rc;
2842 		break;
2843 	default:
2844 		panic("vm_exit_inst_error: vmx_enter_guest returned %d", rc);
2845 	}
2846 }
2847 
2848 /*
2849  * If the NMI-exiting VM execution control is set to '1' then an NMI in
2850  * non-root operation causes a VM-exit. NMI blocking is in effect so it is
2851  * sufficient to simply vector to the NMI handler via a software interrupt.
2852  * However, this must be done before maskable interrupts are enabled
2853  * otherwise the "iret" issued by an interrupt handler will incorrectly
2854  * clear NMI blocking.
2855  */
2856 static __inline void
2857 vmx_exit_handle_nmi(struct vmx *vmx, int vcpuid, struct vm_exit *vmexit)
2858 {
2859 	uint32_t intr_info;
2860 
2861 	KASSERT((read_rflags() & PSL_I) == 0, ("interrupts enabled"));
2862 
2863 	if (vmexit->u.vmx.exit_reason != EXIT_REASON_EXCEPTION)
2864 		return;
2865 
2866 	intr_info = vmcs_read(VMCS_EXIT_INTR_INFO);
2867 	KASSERT((intr_info & VMCS_INTR_VALID) != 0,
2868 	    ("VM exit interruption info invalid: %#x", intr_info));
2869 
2870 	if ((intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_NMI) {
2871 		KASSERT((intr_info & 0xff) == IDT_NMI, ("VM exit due "
2872 		    "to NMI has invalid vector: %#x", intr_info));
2873 		VCPU_CTR0(vmx->vm, vcpuid, "Vectoring to NMI handler");
2874 		__asm __volatile("int $2");
2875 	}
2876 }
2877 
2878 static __inline void
2879 vmx_dr_enter_guest(struct vmxctx *vmxctx)
2880 {
2881 	register_t rflags;
2882 
2883 	/* Save host control debug registers. */
2884 	vmxctx->host_dr7 = rdr7();
2885 	vmxctx->host_debugctl = rdmsr(MSR_DEBUGCTLMSR);
2886 
2887 	/*
2888 	 * Disable debugging in DR7 and DEBUGCTL to avoid triggering
2889 	 * exceptions in the host based on the guest DRx values.  The
2890 	 * guest DR7 and DEBUGCTL are saved/restored in the VMCS.
2891 	 */
2892 	load_dr7(0);
2893 	wrmsr(MSR_DEBUGCTLMSR, 0);
2894 
2895 	/*
2896 	 * Disable single stepping the kernel to avoid corrupting the
2897 	 * guest DR6.  A debugger might still be able to corrupt the
2898 	 * guest DR6 by setting a breakpoint after this point and then
2899 	 * single stepping.
2900 	 */
2901 	rflags = read_rflags();
2902 	vmxctx->host_tf = rflags & PSL_T;
2903 	write_rflags(rflags & ~PSL_T);
2904 
2905 	/* Save host debug registers. */
2906 	vmxctx->host_dr0 = rdr0();
2907 	vmxctx->host_dr1 = rdr1();
2908 	vmxctx->host_dr2 = rdr2();
2909 	vmxctx->host_dr3 = rdr3();
2910 	vmxctx->host_dr6 = rdr6();
2911 
2912 	/* Restore guest debug registers. */
2913 	load_dr0(vmxctx->guest_dr0);
2914 	load_dr1(vmxctx->guest_dr1);
2915 	load_dr2(vmxctx->guest_dr2);
2916 	load_dr3(vmxctx->guest_dr3);
2917 	load_dr6(vmxctx->guest_dr6);
2918 }
2919 
2920 static __inline void
2921 vmx_dr_leave_guest(struct vmxctx *vmxctx)
2922 {
2923 
2924 	/* Save guest debug registers. */
2925 	vmxctx->guest_dr0 = rdr0();
2926 	vmxctx->guest_dr1 = rdr1();
2927 	vmxctx->guest_dr2 = rdr2();
2928 	vmxctx->guest_dr3 = rdr3();
2929 	vmxctx->guest_dr6 = rdr6();
2930 
2931 	/*
2932 	 * Restore host debug registers.  Restore DR7, DEBUGCTL, and
2933 	 * PSL_T last.
2934 	 */
2935 	load_dr0(vmxctx->host_dr0);
2936 	load_dr1(vmxctx->host_dr1);
2937 	load_dr2(vmxctx->host_dr2);
2938 	load_dr3(vmxctx->host_dr3);
2939 	load_dr6(vmxctx->host_dr6);
2940 	wrmsr(MSR_DEBUGCTLMSR, vmxctx->host_debugctl);
2941 	load_dr7(vmxctx->host_dr7);
2942 	write_rflags(read_rflags() | vmxctx->host_tf);
2943 }
2944 
2945 static __inline void
2946 vmx_pmap_activate(struct vmx *vmx, pmap_t pmap)
2947 {
2948 	long eptgen;
2949 	int cpu;
2950 
2951 	cpu = curcpu;
2952 
2953 	CPU_SET_ATOMIC(cpu, &pmap->pm_active);
2954 	smr_enter(pmap->pm_eptsmr);
2955 	eptgen = atomic_load_long(&pmap->pm_eptgen);
2956 	if (eptgen != vmx->eptgen[cpu]) {
2957 		vmx->eptgen[cpu] = eptgen;
2958 		invept(INVEPT_TYPE_SINGLE_CONTEXT,
2959 		    (struct invept_desc){ .eptp = vmx->eptp, ._res = 0 });
2960 	}
2961 }
2962 
2963 static __inline void
2964 vmx_pmap_deactivate(struct vmx *vmx, pmap_t pmap)
2965 {
2966 	smr_exit(pmap->pm_eptsmr);
2967 	CPU_CLR_ATOMIC(curcpu, &pmap->pm_active);
2968 }
2969 
2970 static int
2971 vmx_run(void *arg, int vcpu, register_t rip, pmap_t pmap,
2972     struct vm_eventinfo *evinfo)
2973 {
2974 	int rc, handled, launched;
2975 	struct vmx *vmx;
2976 	struct vm *vm;
2977 	struct vmxctx *vmxctx;
2978 	struct vmcs *vmcs;
2979 	struct vm_exit *vmexit;
2980 	struct vlapic *vlapic;
2981 	uint32_t exit_reason;
2982 	struct region_descriptor gdtr, idtr;
2983 	uint16_t ldt_sel;
2984 
2985 	vmx = arg;
2986 	vm = vmx->vm;
2987 	vmcs = &vmx->vmcs[vcpu];
2988 	vmxctx = &vmx->ctx[vcpu];
2989 	vlapic = vm_lapic(vm, vcpu);
2990 	vmexit = vm_exitinfo(vm, vcpu);
2991 	launched = 0;
2992 
2993 	KASSERT(vmxctx->pmap == pmap,
2994 	    ("pmap %p different than ctx pmap %p", pmap, vmxctx->pmap));
2995 
2996 	vmx_msr_guest_enter(vmx, vcpu);
2997 
2998 	VMPTRLD(vmcs);
2999 
3000 	/*
3001 	 * XXX
3002 	 * We do this every time because we may setup the virtual machine
3003 	 * from a different process than the one that actually runs it.
3004 	 *
3005 	 * If the life of a virtual machine was spent entirely in the context
3006 	 * of a single process we could do this once in vmx_init().
3007 	 */
3008 	vmcs_write(VMCS_HOST_CR3, rcr3());
3009 
3010 	vmcs_write(VMCS_GUEST_RIP, rip);
3011 	vmx_set_pcpu_defaults(vmx, vcpu, pmap);
3012 	do {
3013 		KASSERT(vmcs_guest_rip() == rip, ("%s: vmcs guest rip mismatch "
3014 		    "%#lx/%#lx", __func__, vmcs_guest_rip(), rip));
3015 
3016 		handled = UNHANDLED;
3017 		/*
3018 		 * Interrupts are disabled from this point on until the
3019 		 * guest starts executing. This is done for the following
3020 		 * reasons:
3021 		 *
3022 		 * If an AST is asserted on this thread after the check below,
3023 		 * then the IPI_AST notification will not be lost, because it
3024 		 * will cause a VM exit due to external interrupt as soon as
3025 		 * the guest state is loaded.
3026 		 *
3027 		 * A posted interrupt after 'vmx_inject_interrupts()' will
3028 		 * not be "lost" because it will be held pending in the host
3029 		 * APIC because interrupts are disabled. The pending interrupt
3030 		 * will be recognized as soon as the guest state is loaded.
3031 		 *
3032 		 * The same reasoning applies to the IPI generated by
3033 		 * pmap_invalidate_ept().
3034 		 */
3035 		disable_intr();
3036 		vmx_inject_interrupts(vmx, vcpu, vlapic, rip);
3037 
3038 		/*
3039 		 * Check for vcpu suspension after injecting events because
3040 		 * vmx_inject_interrupts() can suspend the vcpu due to a
3041 		 * triple fault.
3042 		 */
3043 		if (vcpu_suspended(evinfo)) {
3044 			enable_intr();
3045 			vm_exit_suspended(vmx->vm, vcpu, rip);
3046 			break;
3047 		}
3048 
3049 		if (vcpu_rendezvous_pending(evinfo)) {
3050 			enable_intr();
3051 			vm_exit_rendezvous(vmx->vm, vcpu, rip);
3052 			break;
3053 		}
3054 
3055 		if (vcpu_reqidle(evinfo)) {
3056 			enable_intr();
3057 			vm_exit_reqidle(vmx->vm, vcpu, rip);
3058 			break;
3059 		}
3060 
3061 		if (vcpu_should_yield(vm, vcpu)) {
3062 			enable_intr();
3063 			vm_exit_astpending(vmx->vm, vcpu, rip);
3064 			vmx_astpending_trace(vmx, vcpu, rip);
3065 			handled = HANDLED;
3066 			break;
3067 		}
3068 
3069 		if (vcpu_debugged(vm, vcpu)) {
3070 			enable_intr();
3071 			vm_exit_debug(vmx->vm, vcpu, rip);
3072 			break;
3073 		}
3074 
3075 		/*
3076 		 * If TPR Shadowing is enabled, the TPR Threshold
3077 		 * must be updated right before entering the guest.
3078 		 */
3079 		if (tpr_shadowing && !virtual_interrupt_delivery) {
3080 			if ((vmx->cap[vcpu].proc_ctls & PROCBASED_USE_TPR_SHADOW) != 0) {
3081 				vmcs_write(VMCS_TPR_THRESHOLD, vlapic_get_cr8(vlapic));
3082 			}
3083 		}
3084 
3085 		/*
3086 		 * VM exits restore the base address but not the
3087 		 * limits of GDTR and IDTR.  The VMCS only stores the
3088 		 * base address, so VM exits set the limits to 0xffff.
3089 		 * Save and restore the full GDTR and IDTR to restore
3090 		 * the limits.
3091 		 *
3092 		 * The VMCS does not save the LDTR at all, and VM
3093 		 * exits clear LDTR as if a NULL selector were loaded.
3094 		 * The userspace hypervisor probably doesn't use a
3095 		 * LDT, but save and restore it to be safe.
3096 		 */
3097 		sgdt(&gdtr);
3098 		sidt(&idtr);
3099 		ldt_sel = sldt();
3100 
3101 		/*
3102 		 * The TSC_AUX MSR must be saved/restored while interrupts
3103 		 * are disabled so that it is not possible for the guest
3104 		 * TSC_AUX MSR value to be overwritten by the resume
3105 		 * portion of the IPI_SUSPEND codepath. This is why the
3106 		 * transition of this MSR is handled separately from those
3107 		 * handled by vmx_msr_guest_{enter,exit}(), which are ok to
3108 		 * be transitioned with preemption disabled but interrupts
3109 		 * enabled.
3110 		 *
3111 		 * These vmx_msr_guest_{enter,exit}_tsc_aux() calls can be
3112 		 * anywhere in this loop so long as they happen with
3113 		 * interrupts disabled. This location is chosen for
3114 		 * simplicity.
3115 		 */
3116 		vmx_msr_guest_enter_tsc_aux(vmx, vcpu);
3117 
3118 		vmx_dr_enter_guest(vmxctx);
3119 
3120 		/*
3121 		 * Mark the EPT as active on this host CPU and invalidate
3122 		 * EPTP-tagged TLB entries if required.
3123 		 */
3124 		vmx_pmap_activate(vmx, pmap);
3125 
3126 		vmx_run_trace(vmx, vcpu);
3127 		rc = vmx_enter_guest(vmxctx, vmx, launched);
3128 
3129 		vmx_pmap_deactivate(vmx, pmap);
3130 		vmx_dr_leave_guest(vmxctx);
3131 		vmx_msr_guest_exit_tsc_aux(vmx, vcpu);
3132 
3133 		bare_lgdt(&gdtr);
3134 		lidt(&idtr);
3135 		lldt(ldt_sel);
3136 
3137 		/* Collect some information for VM exit processing */
3138 		vmexit->rip = rip = vmcs_guest_rip();
3139 		vmexit->inst_length = vmexit_instruction_length();
3140 		vmexit->u.vmx.exit_reason = exit_reason = vmcs_exit_reason();
3141 		vmexit->u.vmx.exit_qualification = vmcs_exit_qualification();
3142 
3143 		/* Update 'nextrip' */
3144 		vmx->state[vcpu].nextrip = rip;
3145 
3146 		if (rc == VMX_GUEST_VMEXIT) {
3147 			vmx_exit_handle_nmi(vmx, vcpu, vmexit);
3148 			enable_intr();
3149 			handled = vmx_exit_process(vmx, vcpu, vmexit);
3150 		} else {
3151 			enable_intr();
3152 			vmx_exit_inst_error(vmxctx, rc, vmexit);
3153 		}
3154 		launched = 1;
3155 		vmx_exit_trace(vmx, vcpu, rip, exit_reason, handled);
3156 		rip = vmexit->rip;
3157 	} while (handled);
3158 
3159 	/*
3160 	 * If a VM exit has been handled then the exitcode must be BOGUS
3161 	 * If a VM exit is not handled then the exitcode must not be BOGUS
3162 	 */
3163 	if ((handled && vmexit->exitcode != VM_EXITCODE_BOGUS) ||
3164 	    (!handled && vmexit->exitcode == VM_EXITCODE_BOGUS)) {
3165 		panic("Mismatch between handled (%d) and exitcode (%d)",
3166 		      handled, vmexit->exitcode);
3167 	}
3168 
3169 	if (!handled)
3170 		vmm_stat_incr(vm, vcpu, VMEXIT_USERSPACE, 1);
3171 
3172 	VCPU_CTR1(vm, vcpu, "returning from vmx_run: exitcode %d",
3173 	    vmexit->exitcode);
3174 
3175 	VMCLEAR(vmcs);
3176 	vmx_msr_guest_exit(vmx, vcpu);
3177 
3178 	return (0);
3179 }
3180 
3181 static void
3182 vmx_cleanup(void *arg)
3183 {
3184 	int i;
3185 	struct vmx *vmx = arg;
3186 	uint16_t maxcpus;
3187 
3188 	if (apic_access_virtualization(vmx, 0))
3189 		vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE);
3190 
3191 	maxcpus = vm_get_maxcpus(vmx->vm);
3192 	for (i = 0; i < maxcpus; i++)
3193 		vpid_free(vmx->state[i].vpid);
3194 
3195 	free(vmx, M_VMX);
3196 
3197 	return;
3198 }
3199 
3200 static register_t *
3201 vmxctx_regptr(struct vmxctx *vmxctx, int reg)
3202 {
3203 
3204 	switch (reg) {
3205 	case VM_REG_GUEST_RAX:
3206 		return (&vmxctx->guest_rax);
3207 	case VM_REG_GUEST_RBX:
3208 		return (&vmxctx->guest_rbx);
3209 	case VM_REG_GUEST_RCX:
3210 		return (&vmxctx->guest_rcx);
3211 	case VM_REG_GUEST_RDX:
3212 		return (&vmxctx->guest_rdx);
3213 	case VM_REG_GUEST_RSI:
3214 		return (&vmxctx->guest_rsi);
3215 	case VM_REG_GUEST_RDI:
3216 		return (&vmxctx->guest_rdi);
3217 	case VM_REG_GUEST_RBP:
3218 		return (&vmxctx->guest_rbp);
3219 	case VM_REG_GUEST_R8:
3220 		return (&vmxctx->guest_r8);
3221 	case VM_REG_GUEST_R9:
3222 		return (&vmxctx->guest_r9);
3223 	case VM_REG_GUEST_R10:
3224 		return (&vmxctx->guest_r10);
3225 	case VM_REG_GUEST_R11:
3226 		return (&vmxctx->guest_r11);
3227 	case VM_REG_GUEST_R12:
3228 		return (&vmxctx->guest_r12);
3229 	case VM_REG_GUEST_R13:
3230 		return (&vmxctx->guest_r13);
3231 	case VM_REG_GUEST_R14:
3232 		return (&vmxctx->guest_r14);
3233 	case VM_REG_GUEST_R15:
3234 		return (&vmxctx->guest_r15);
3235 	case VM_REG_GUEST_CR2:
3236 		return (&vmxctx->guest_cr2);
3237 	case VM_REG_GUEST_DR0:
3238 		return (&vmxctx->guest_dr0);
3239 	case VM_REG_GUEST_DR1:
3240 		return (&vmxctx->guest_dr1);
3241 	case VM_REG_GUEST_DR2:
3242 		return (&vmxctx->guest_dr2);
3243 	case VM_REG_GUEST_DR3:
3244 		return (&vmxctx->guest_dr3);
3245 	case VM_REG_GUEST_DR6:
3246 		return (&vmxctx->guest_dr6);
3247 	default:
3248 		break;
3249 	}
3250 	return (NULL);
3251 }
3252 
3253 static int
3254 vmxctx_getreg(struct vmxctx *vmxctx, int reg, uint64_t *retval)
3255 {
3256 	register_t *regp;
3257 
3258 	if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) {
3259 		*retval = *regp;
3260 		return (0);
3261 	} else
3262 		return (EINVAL);
3263 }
3264 
3265 static int
3266 vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val)
3267 {
3268 	register_t *regp;
3269 
3270 	if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) {
3271 		*regp = val;
3272 		return (0);
3273 	} else
3274 		return (EINVAL);
3275 }
3276 
3277 static int
3278 vmx_get_intr_shadow(struct vmx *vmx, int vcpu, int running, uint64_t *retval)
3279 {
3280 	uint64_t gi;
3281 	int error;
3282 
3283 	error = vmcs_getreg(&vmx->vmcs[vcpu], running,
3284 	    VMCS_IDENT(VMCS_GUEST_INTERRUPTIBILITY), &gi);
3285 	*retval = (gi & HWINTR_BLOCKING) ? 1 : 0;
3286 	return (error);
3287 }
3288 
3289 static int
3290 vmx_modify_intr_shadow(struct vmx *vmx, int vcpu, int running, uint64_t val)
3291 {
3292 	struct vmcs *vmcs;
3293 	uint64_t gi;
3294 	int error, ident;
3295 
3296 	/*
3297 	 * Forcing the vcpu into an interrupt shadow is not supported.
3298 	 */
3299 	if (val) {
3300 		error = EINVAL;
3301 		goto done;
3302 	}
3303 
3304 	vmcs = &vmx->vmcs[vcpu];
3305 	ident = VMCS_IDENT(VMCS_GUEST_INTERRUPTIBILITY);
3306 	error = vmcs_getreg(vmcs, running, ident, &gi);
3307 	if (error == 0) {
3308 		gi &= ~HWINTR_BLOCKING;
3309 		error = vmcs_setreg(vmcs, running, ident, gi);
3310 	}
3311 done:
3312 	VCPU_CTR2(vmx->vm, vcpu, "Setting intr_shadow to %#lx %s", val,
3313 	    error ? "failed" : "succeeded");
3314 	return (error);
3315 }
3316 
3317 static int
3318 vmx_shadow_reg(int reg)
3319 {
3320 	int shreg;
3321 
3322 	shreg = -1;
3323 
3324 	switch (reg) {
3325 	case VM_REG_GUEST_CR0:
3326 		shreg = VMCS_CR0_SHADOW;
3327 		break;
3328 	case VM_REG_GUEST_CR4:
3329 		shreg = VMCS_CR4_SHADOW;
3330 		break;
3331 	default:
3332 		break;
3333 	}
3334 
3335 	return (shreg);
3336 }
3337 
3338 static int
3339 vmx_getreg(void *arg, int vcpu, int reg, uint64_t *retval)
3340 {
3341 	int running, hostcpu;
3342 	struct vmx *vmx = arg;
3343 
3344 	running = vcpu_is_running(vmx->vm, vcpu, &hostcpu);
3345 	if (running && hostcpu != curcpu)
3346 		panic("vmx_getreg: %s%d is running", vm_name(vmx->vm), vcpu);
3347 
3348 	if (reg == VM_REG_GUEST_INTR_SHADOW)
3349 		return (vmx_get_intr_shadow(vmx, vcpu, running, retval));
3350 
3351 	if (vmxctx_getreg(&vmx->ctx[vcpu], reg, retval) == 0)
3352 		return (0);
3353 
3354 	return (vmcs_getreg(&vmx->vmcs[vcpu], running, reg, retval));
3355 }
3356 
3357 static int
3358 vmx_setreg(void *arg, int vcpu, int reg, uint64_t val)
3359 {
3360 	int error, hostcpu, running, shadow;
3361 	uint64_t ctls;
3362 	pmap_t pmap;
3363 	struct vmx *vmx = arg;
3364 
3365 	running = vcpu_is_running(vmx->vm, vcpu, &hostcpu);
3366 	if (running && hostcpu != curcpu)
3367 		panic("vmx_setreg: %s%d is running", vm_name(vmx->vm), vcpu);
3368 
3369 	if (reg == VM_REG_GUEST_INTR_SHADOW)
3370 		return (vmx_modify_intr_shadow(vmx, vcpu, running, val));
3371 
3372 	if (vmxctx_setreg(&vmx->ctx[vcpu], reg, val) == 0)
3373 		return (0);
3374 
3375 	/* Do not permit user write access to VMCS fields by offset. */
3376 	if (reg < 0)
3377 		return (EINVAL);
3378 
3379 	error = vmcs_setreg(&vmx->vmcs[vcpu], running, reg, val);
3380 
3381 	if (error == 0) {
3382 		/*
3383 		 * If the "load EFER" VM-entry control is 1 then the
3384 		 * value of EFER.LMA must be identical to "IA-32e mode guest"
3385 		 * bit in the VM-entry control.
3386 		 */
3387 		if ((entry_ctls & VM_ENTRY_LOAD_EFER) != 0 &&
3388 		    (reg == VM_REG_GUEST_EFER)) {
3389 			vmcs_getreg(&vmx->vmcs[vcpu], running,
3390 				    VMCS_IDENT(VMCS_ENTRY_CTLS), &ctls);
3391 			if (val & EFER_LMA)
3392 				ctls |= VM_ENTRY_GUEST_LMA;
3393 			else
3394 				ctls &= ~VM_ENTRY_GUEST_LMA;
3395 			vmcs_setreg(&vmx->vmcs[vcpu], running,
3396 				    VMCS_IDENT(VMCS_ENTRY_CTLS), ctls);
3397 		}
3398 
3399 		shadow = vmx_shadow_reg(reg);
3400 		if (shadow > 0) {
3401 			/*
3402 			 * Store the unmodified value in the shadow
3403 			 */
3404 			error = vmcs_setreg(&vmx->vmcs[vcpu], running,
3405 				    VMCS_IDENT(shadow), val);
3406 		}
3407 
3408 		if (reg == VM_REG_GUEST_CR3) {
3409 			/*
3410 			 * Invalidate the guest vcpu's TLB mappings to emulate
3411 			 * the behavior of updating %cr3.
3412 			 *
3413 			 * XXX the processor retains global mappings when %cr3
3414 			 * is updated but vmx_invvpid() does not.
3415 			 */
3416 			pmap = vmx->ctx[vcpu].pmap;
3417 			vmx_invvpid(vmx, vcpu, pmap, running);
3418 		}
3419 	}
3420 
3421 	return (error);
3422 }
3423 
3424 static int
3425 vmx_getdesc(void *arg, int vcpu, int reg, struct seg_desc *desc)
3426 {
3427 	int hostcpu, running;
3428 	struct vmx *vmx = arg;
3429 
3430 	running = vcpu_is_running(vmx->vm, vcpu, &hostcpu);
3431 	if (running && hostcpu != curcpu)
3432 		panic("vmx_getdesc: %s%d is running", vm_name(vmx->vm), vcpu);
3433 
3434 	return (vmcs_getdesc(&vmx->vmcs[vcpu], running, reg, desc));
3435 }
3436 
3437 static int
3438 vmx_setdesc(void *arg, int vcpu, int reg, struct seg_desc *desc)
3439 {
3440 	int hostcpu, running;
3441 	struct vmx *vmx = arg;
3442 
3443 	running = vcpu_is_running(vmx->vm, vcpu, &hostcpu);
3444 	if (running && hostcpu != curcpu)
3445 		panic("vmx_setdesc: %s%d is running", vm_name(vmx->vm), vcpu);
3446 
3447 	return (vmcs_setdesc(&vmx->vmcs[vcpu], running, reg, desc));
3448 }
3449 
3450 static int
3451 vmx_getcap(void *arg, int vcpu, int type, int *retval)
3452 {
3453 	struct vmx *vmx = arg;
3454 	int vcap;
3455 	int ret;
3456 
3457 	ret = ENOENT;
3458 
3459 	vcap = vmx->cap[vcpu].set;
3460 
3461 	switch (type) {
3462 	case VM_CAP_HALT_EXIT:
3463 		if (cap_halt_exit)
3464 			ret = 0;
3465 		break;
3466 	case VM_CAP_PAUSE_EXIT:
3467 		if (cap_pause_exit)
3468 			ret = 0;
3469 		break;
3470 	case VM_CAP_MTRAP_EXIT:
3471 		if (cap_monitor_trap)
3472 			ret = 0;
3473 		break;
3474 	case VM_CAP_RDPID:
3475 		if (cap_rdpid)
3476 			ret = 0;
3477 		break;
3478 	case VM_CAP_RDTSCP:
3479 		if (cap_rdtscp)
3480 			ret = 0;
3481 		break;
3482 	case VM_CAP_UNRESTRICTED_GUEST:
3483 		if (cap_unrestricted_guest)
3484 			ret = 0;
3485 		break;
3486 	case VM_CAP_ENABLE_INVPCID:
3487 		if (cap_invpcid)
3488 			ret = 0;
3489 		break;
3490 	case VM_CAP_BPT_EXIT:
3491 		ret = 0;
3492 		break;
3493 	default:
3494 		break;
3495 	}
3496 
3497 	if (ret == 0)
3498 		*retval = (vcap & (1 << type)) ? 1 : 0;
3499 
3500 	return (ret);
3501 }
3502 
3503 static int
3504 vmx_setcap(void *arg, int vcpu, int type, int val)
3505 {
3506 	struct vmx *vmx = arg;
3507 	struct vmcs *vmcs = &vmx->vmcs[vcpu];
3508 	uint32_t baseval;
3509 	uint32_t *pptr;
3510 	int error;
3511 	int flag;
3512 	int reg;
3513 	int retval;
3514 
3515 	retval = ENOENT;
3516 	pptr = NULL;
3517 
3518 	switch (type) {
3519 	case VM_CAP_HALT_EXIT:
3520 		if (cap_halt_exit) {
3521 			retval = 0;
3522 			pptr = &vmx->cap[vcpu].proc_ctls;
3523 			baseval = *pptr;
3524 			flag = PROCBASED_HLT_EXITING;
3525 			reg = VMCS_PRI_PROC_BASED_CTLS;
3526 		}
3527 		break;
3528 	case VM_CAP_MTRAP_EXIT:
3529 		if (cap_monitor_trap) {
3530 			retval = 0;
3531 			pptr = &vmx->cap[vcpu].proc_ctls;
3532 			baseval = *pptr;
3533 			flag = PROCBASED_MTF;
3534 			reg = VMCS_PRI_PROC_BASED_CTLS;
3535 		}
3536 		break;
3537 	case VM_CAP_PAUSE_EXIT:
3538 		if (cap_pause_exit) {
3539 			retval = 0;
3540 			pptr = &vmx->cap[vcpu].proc_ctls;
3541 			baseval = *pptr;
3542 			flag = PROCBASED_PAUSE_EXITING;
3543 			reg = VMCS_PRI_PROC_BASED_CTLS;
3544 		}
3545 		break;
3546 	case VM_CAP_RDPID:
3547 	case VM_CAP_RDTSCP:
3548 		if (cap_rdpid || cap_rdtscp)
3549 			/*
3550 			 * Choose not to support enabling/disabling
3551 			 * RDPID/RDTSCP via libvmmapi since, as per the
3552 			 * discussion in vmx_modinit(), RDPID/RDTSCP are
3553 			 * either always enabled or always disabled.
3554 			 */
3555 			error = EOPNOTSUPP;
3556 		break;
3557 	case VM_CAP_UNRESTRICTED_GUEST:
3558 		if (cap_unrestricted_guest) {
3559 			retval = 0;
3560 			pptr = &vmx->cap[vcpu].proc_ctls2;
3561 			baseval = *pptr;
3562 			flag = PROCBASED2_UNRESTRICTED_GUEST;
3563 			reg = VMCS_SEC_PROC_BASED_CTLS;
3564 		}
3565 		break;
3566 	case VM_CAP_ENABLE_INVPCID:
3567 		if (cap_invpcid) {
3568 			retval = 0;
3569 			pptr = &vmx->cap[vcpu].proc_ctls2;
3570 			baseval = *pptr;
3571 			flag = PROCBASED2_ENABLE_INVPCID;
3572 			reg = VMCS_SEC_PROC_BASED_CTLS;
3573 		}
3574 		break;
3575 	case VM_CAP_BPT_EXIT:
3576 		retval = 0;
3577 
3578 		/* Don't change the bitmap if we are tracing all exceptions. */
3579 		if (vmx->cap[vcpu].exc_bitmap != 0xffffffff) {
3580 			pptr = &vmx->cap[vcpu].exc_bitmap;
3581 			baseval = *pptr;
3582 			flag = (1 << IDT_BP);
3583 			reg = VMCS_EXCEPTION_BITMAP;
3584 		}
3585 		break;
3586 	default:
3587 		break;
3588 	}
3589 
3590 	if (retval)
3591 		return (retval);
3592 
3593 	if (pptr != NULL) {
3594 		if (val) {
3595 			baseval |= flag;
3596 		} else {
3597 			baseval &= ~flag;
3598 		}
3599 		VMPTRLD(vmcs);
3600 		error = vmwrite(reg, baseval);
3601 		VMCLEAR(vmcs);
3602 
3603 		if (error)
3604 			return (error);
3605 
3606 		/*
3607 		 * Update optional stored flags, and record
3608 		 * setting
3609 		 */
3610 		*pptr = baseval;
3611 	}
3612 
3613 	if (val) {
3614 		vmx->cap[vcpu].set |= (1 << type);
3615 	} else {
3616 		vmx->cap[vcpu].set &= ~(1 << type);
3617 	}
3618 
3619 	return (0);
3620 }
3621 
3622 static struct vmspace *
3623 vmx_vmspace_alloc(vm_offset_t min, vm_offset_t max)
3624 {
3625 	return (ept_vmspace_alloc(min, max));
3626 }
3627 
3628 static void
3629 vmx_vmspace_free(struct vmspace *vmspace)
3630 {
3631 	ept_vmspace_free(vmspace);
3632 }
3633 
3634 struct vlapic_vtx {
3635 	struct vlapic	vlapic;
3636 	struct pir_desc	*pir_desc;
3637 	struct vmx	*vmx;
3638 	u_int	pending_prio;
3639 };
3640 
3641 #define VPR_PRIO_BIT(vpr)	(1 << ((vpr) >> 4))
3642 
3643 #define	VMX_CTR_PIR(vm, vcpuid, pir_desc, notify, vector, level, msg)	\
3644 do {									\
3645 	VCPU_CTR2(vm, vcpuid, msg " assert %s-triggered vector %d",	\
3646 	    level ? "level" : "edge", vector);				\
3647 	VCPU_CTR1(vm, vcpuid, msg " pir0 0x%016lx", pir_desc->pir[0]);	\
3648 	VCPU_CTR1(vm, vcpuid, msg " pir1 0x%016lx", pir_desc->pir[1]);	\
3649 	VCPU_CTR1(vm, vcpuid, msg " pir2 0x%016lx", pir_desc->pir[2]);	\
3650 	VCPU_CTR1(vm, vcpuid, msg " pir3 0x%016lx", pir_desc->pir[3]);	\
3651 	VCPU_CTR1(vm, vcpuid, msg " notify: %s", notify ? "yes" : "no");\
3652 } while (0)
3653 
3654 /*
3655  * vlapic->ops handlers that utilize the APICv hardware assist described in
3656  * Chapter 29 of the Intel SDM.
3657  */
3658 static int
3659 vmx_set_intr_ready(struct vlapic *vlapic, int vector, bool level)
3660 {
3661 	struct vlapic_vtx *vlapic_vtx;
3662 	struct pir_desc *pir_desc;
3663 	uint64_t mask;
3664 	int idx, notify = 0;
3665 
3666 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
3667 	pir_desc = vlapic_vtx->pir_desc;
3668 
3669 	/*
3670 	 * Keep track of interrupt requests in the PIR descriptor. This is
3671 	 * because the virtual APIC page pointed to by the VMCS cannot be
3672 	 * modified if the vcpu is running.
3673 	 */
3674 	idx = vector / 64;
3675 	mask = 1UL << (vector % 64);
3676 	atomic_set_long(&pir_desc->pir[idx], mask);
3677 
3678 	/*
3679 	 * A notification is required whenever the 'pending' bit makes a
3680 	 * transition from 0->1.
3681 	 *
3682 	 * Even if the 'pending' bit is already asserted, notification about
3683 	 * the incoming interrupt may still be necessary.  For example, if a
3684 	 * vCPU is HLTed with a high PPR, a low priority interrupt would cause
3685 	 * the 0->1 'pending' transition with a notification, but the vCPU
3686 	 * would ignore the interrupt for the time being.  The same vCPU would
3687 	 * need to then be notified if a high-priority interrupt arrived which
3688 	 * satisfied the PPR.
3689 	 *
3690 	 * The priorities of interrupts injected while 'pending' is asserted
3691 	 * are tracked in a custom bitfield 'pending_prio'.  Should the
3692 	 * to-be-injected interrupt exceed the priorities already present, the
3693 	 * notification is sent.  The priorities recorded in 'pending_prio' are
3694 	 * cleared whenever the 'pending' bit makes another 0->1 transition.
3695 	 */
3696 	if (atomic_cmpset_long(&pir_desc->pending, 0, 1) != 0) {
3697 		notify = 1;
3698 		vlapic_vtx->pending_prio = 0;
3699 	} else {
3700 		const u_int old_prio = vlapic_vtx->pending_prio;
3701 		const u_int prio_bit = VPR_PRIO_BIT(vector & APIC_TPR_INT);
3702 
3703 		if ((old_prio & prio_bit) == 0 && prio_bit > old_prio) {
3704 			atomic_set_int(&vlapic_vtx->pending_prio, prio_bit);
3705 			notify = 1;
3706 		}
3707 	}
3708 
3709 	VMX_CTR_PIR(vlapic->vm, vlapic->vcpuid, pir_desc, notify, vector,
3710 	    level, "vmx_set_intr_ready");
3711 	return (notify);
3712 }
3713 
3714 static int
3715 vmx_pending_intr(struct vlapic *vlapic, int *vecptr)
3716 {
3717 	struct vlapic_vtx *vlapic_vtx;
3718 	struct pir_desc *pir_desc;
3719 	struct LAPIC *lapic;
3720 	uint64_t pending, pirval;
3721 	uint32_t ppr, vpr;
3722 	int i;
3723 
3724 	/*
3725 	 * This function is only expected to be called from the 'HLT' exit
3726 	 * handler which does not care about the vector that is pending.
3727 	 */
3728 	KASSERT(vecptr == NULL, ("vmx_pending_intr: vecptr must be NULL"));
3729 
3730 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
3731 	pir_desc = vlapic_vtx->pir_desc;
3732 
3733 	pending = atomic_load_acq_long(&pir_desc->pending);
3734 	if (!pending) {
3735 		/*
3736 		 * While a virtual interrupt may have already been
3737 		 * processed the actual delivery maybe pending the
3738 		 * interruptibility of the guest.  Recognize a pending
3739 		 * interrupt by reevaluating virtual interrupts
3740 		 * following Section 29.2.1 in the Intel SDM Volume 3.
3741 		 */
3742 		struct vm_exit *vmexit;
3743 		uint8_t rvi, ppr;
3744 
3745 		vmexit = vm_exitinfo(vlapic->vm, vlapic->vcpuid);
3746 		KASSERT(vmexit->exitcode == VM_EXITCODE_HLT,
3747 		    ("vmx_pending_intr: exitcode not 'HLT'"));
3748 		rvi = vmexit->u.hlt.intr_status & APIC_TPR_INT;
3749 		lapic = vlapic->apic_page;
3750 		ppr = lapic->ppr & APIC_TPR_INT;
3751 		if (rvi > ppr) {
3752 			return (1);
3753 		}
3754 
3755 		return (0);
3756 	}
3757 
3758 	/*
3759 	 * If there is an interrupt pending then it will be recognized only
3760 	 * if its priority is greater than the processor priority.
3761 	 *
3762 	 * Special case: if the processor priority is zero then any pending
3763 	 * interrupt will be recognized.
3764 	 */
3765 	lapic = vlapic->apic_page;
3766 	ppr = lapic->ppr & APIC_TPR_INT;
3767 	if (ppr == 0)
3768 		return (1);
3769 
3770 	VCPU_CTR1(vlapic->vm, vlapic->vcpuid, "HLT with non-zero PPR %d",
3771 	    lapic->ppr);
3772 
3773 	vpr = 0;
3774 	for (i = 3; i >= 0; i--) {
3775 		pirval = pir_desc->pir[i];
3776 		if (pirval != 0) {
3777 			vpr = (i * 64 + flsl(pirval) - 1) & APIC_TPR_INT;
3778 			break;
3779 		}
3780 	}
3781 
3782 	/*
3783 	 * If the highest-priority pending interrupt falls short of the
3784 	 * processor priority of this vCPU, ensure that 'pending_prio' does not
3785 	 * have any stale bits which would preclude a higher-priority interrupt
3786 	 * from incurring a notification later.
3787 	 */
3788 	if (vpr <= ppr) {
3789 		const u_int prio_bit = VPR_PRIO_BIT(vpr);
3790 		const u_int old = vlapic_vtx->pending_prio;
3791 
3792 		if (old > prio_bit && (old & prio_bit) == 0) {
3793 			vlapic_vtx->pending_prio = prio_bit;
3794 		}
3795 		return (0);
3796 	}
3797 	return (1);
3798 }
3799 
3800 static void
3801 vmx_intr_accepted(struct vlapic *vlapic, int vector)
3802 {
3803 
3804 	panic("vmx_intr_accepted: not expected to be called");
3805 }
3806 
3807 static void
3808 vmx_set_tmr(struct vlapic *vlapic, int vector, bool level)
3809 {
3810 	struct vlapic_vtx *vlapic_vtx;
3811 	struct vmx *vmx;
3812 	struct vmcs *vmcs;
3813 	uint64_t mask, val;
3814 
3815 	KASSERT(vector >= 0 && vector <= 255, ("invalid vector %d", vector));
3816 	KASSERT(!vcpu_is_running(vlapic->vm, vlapic->vcpuid, NULL),
3817 	    ("vmx_set_tmr: vcpu cannot be running"));
3818 
3819 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
3820 	vmx = vlapic_vtx->vmx;
3821 	vmcs = &vmx->vmcs[vlapic->vcpuid];
3822 	mask = 1UL << (vector % 64);
3823 
3824 	VMPTRLD(vmcs);
3825 	val = vmcs_read(VMCS_EOI_EXIT(vector));
3826 	if (level)
3827 		val |= mask;
3828 	else
3829 		val &= ~mask;
3830 	vmcs_write(VMCS_EOI_EXIT(vector), val);
3831 	VMCLEAR(vmcs);
3832 }
3833 
3834 static void
3835 vmx_enable_x2apic_mode_ts(struct vlapic *vlapic)
3836 {
3837 	struct vmx *vmx;
3838 	struct vmcs *vmcs;
3839 	uint32_t proc_ctls;
3840 	int vcpuid;
3841 
3842 	vcpuid = vlapic->vcpuid;
3843 	vmx = ((struct vlapic_vtx *)vlapic)->vmx;
3844 	vmcs = &vmx->vmcs[vcpuid];
3845 
3846 	proc_ctls = vmx->cap[vcpuid].proc_ctls;
3847 	proc_ctls &= ~PROCBASED_USE_TPR_SHADOW;
3848 	proc_ctls |= PROCBASED_CR8_LOAD_EXITING;
3849 	proc_ctls |= PROCBASED_CR8_STORE_EXITING;
3850 	vmx->cap[vcpuid].proc_ctls = proc_ctls;
3851 
3852 	VMPTRLD(vmcs);
3853 	vmcs_write(VMCS_PRI_PROC_BASED_CTLS, proc_ctls);
3854 	VMCLEAR(vmcs);
3855 }
3856 
3857 static void
3858 vmx_enable_x2apic_mode_vid(struct vlapic *vlapic)
3859 {
3860 	struct vmx *vmx;
3861 	struct vmcs *vmcs;
3862 	uint32_t proc_ctls2;
3863 	int vcpuid, error;
3864 
3865 	vcpuid = vlapic->vcpuid;
3866 	vmx = ((struct vlapic_vtx *)vlapic)->vmx;
3867 	vmcs = &vmx->vmcs[vcpuid];
3868 
3869 	proc_ctls2 = vmx->cap[vcpuid].proc_ctls2;
3870 	KASSERT((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) != 0,
3871 	    ("%s: invalid proc_ctls2 %#x", __func__, proc_ctls2));
3872 
3873 	proc_ctls2 &= ~PROCBASED2_VIRTUALIZE_APIC_ACCESSES;
3874 	proc_ctls2 |= PROCBASED2_VIRTUALIZE_X2APIC_MODE;
3875 	vmx->cap[vcpuid].proc_ctls2 = proc_ctls2;
3876 
3877 	VMPTRLD(vmcs);
3878 	vmcs_write(VMCS_SEC_PROC_BASED_CTLS, proc_ctls2);
3879 	VMCLEAR(vmcs);
3880 
3881 	if (vlapic->vcpuid == 0) {
3882 		/*
3883 		 * The nested page table mappings are shared by all vcpus
3884 		 * so unmap the APIC access page just once.
3885 		 */
3886 		error = vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE);
3887 		KASSERT(error == 0, ("%s: vm_unmap_mmio error %d",
3888 		    __func__, error));
3889 
3890 		/*
3891 		 * The MSR bitmap is shared by all vcpus so modify it only
3892 		 * once in the context of vcpu 0.
3893 		 */
3894 		error = vmx_allow_x2apic_msrs(vmx);
3895 		KASSERT(error == 0, ("%s: vmx_allow_x2apic_msrs error %d",
3896 		    __func__, error));
3897 	}
3898 }
3899 
3900 static void
3901 vmx_post_intr(struct vlapic *vlapic, int hostcpu)
3902 {
3903 
3904 	ipi_cpu(hostcpu, pirvec);
3905 }
3906 
3907 /*
3908  * Transfer the pending interrupts in the PIR descriptor to the IRR
3909  * in the virtual APIC page.
3910  */
3911 static void
3912 vmx_inject_pir(struct vlapic *vlapic)
3913 {
3914 	struct vlapic_vtx *vlapic_vtx;
3915 	struct pir_desc *pir_desc;
3916 	struct LAPIC *lapic;
3917 	uint64_t val, pirval;
3918 	int rvi, pirbase = -1;
3919 	uint16_t intr_status_old, intr_status_new;
3920 
3921 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
3922 	pir_desc = vlapic_vtx->pir_desc;
3923 	if (atomic_cmpset_long(&pir_desc->pending, 1, 0) == 0) {
3924 		VCPU_CTR0(vlapic->vm, vlapic->vcpuid, "vmx_inject_pir: "
3925 		    "no posted interrupt pending");
3926 		return;
3927 	}
3928 
3929 	pirval = 0;
3930 	pirbase = -1;
3931 	lapic = vlapic->apic_page;
3932 
3933 	val = atomic_readandclear_long(&pir_desc->pir[0]);
3934 	if (val != 0) {
3935 		lapic->irr0 |= val;
3936 		lapic->irr1 |= val >> 32;
3937 		pirbase = 0;
3938 		pirval = val;
3939 	}
3940 
3941 	val = atomic_readandclear_long(&pir_desc->pir[1]);
3942 	if (val != 0) {
3943 		lapic->irr2 |= val;
3944 		lapic->irr3 |= val >> 32;
3945 		pirbase = 64;
3946 		pirval = val;
3947 	}
3948 
3949 	val = atomic_readandclear_long(&pir_desc->pir[2]);
3950 	if (val != 0) {
3951 		lapic->irr4 |= val;
3952 		lapic->irr5 |= val >> 32;
3953 		pirbase = 128;
3954 		pirval = val;
3955 	}
3956 
3957 	val = atomic_readandclear_long(&pir_desc->pir[3]);
3958 	if (val != 0) {
3959 		lapic->irr6 |= val;
3960 		lapic->irr7 |= val >> 32;
3961 		pirbase = 192;
3962 		pirval = val;
3963 	}
3964 
3965 	VLAPIC_CTR_IRR(vlapic, "vmx_inject_pir");
3966 
3967 	/*
3968 	 * Update RVI so the processor can evaluate pending virtual
3969 	 * interrupts on VM-entry.
3970 	 *
3971 	 * It is possible for pirval to be 0 here, even though the
3972 	 * pending bit has been set. The scenario is:
3973 	 * CPU-Y is sending a posted interrupt to CPU-X, which
3974 	 * is running a guest and processing posted interrupts in h/w.
3975 	 * CPU-X will eventually exit and the state seen in s/w is
3976 	 * the pending bit set, but no PIR bits set.
3977 	 *
3978 	 *      CPU-X                      CPU-Y
3979 	 *   (vm running)                (host running)
3980 	 *   rx posted interrupt
3981 	 *   CLEAR pending bit
3982 	 *				 SET PIR bit
3983 	 *   READ/CLEAR PIR bits
3984 	 *				 SET pending bit
3985 	 *   (vm exit)
3986 	 *   pending bit set, PIR 0
3987 	 */
3988 	if (pirval != 0) {
3989 		rvi = pirbase + flsl(pirval) - 1;
3990 		intr_status_old = vmcs_read(VMCS_GUEST_INTR_STATUS);
3991 		intr_status_new = (intr_status_old & 0xFF00) | rvi;
3992 		if (intr_status_new > intr_status_old) {
3993 			vmcs_write(VMCS_GUEST_INTR_STATUS, intr_status_new);
3994 			VCPU_CTR2(vlapic->vm, vlapic->vcpuid, "vmx_inject_pir: "
3995 			    "guest_intr_status changed from 0x%04x to 0x%04x",
3996 			    intr_status_old, intr_status_new);
3997 		}
3998 	}
3999 }
4000 
4001 static struct vlapic *
4002 vmx_vlapic_init(void *arg, int vcpuid)
4003 {
4004 	struct vmx *vmx;
4005 	struct vlapic *vlapic;
4006 	struct vlapic_vtx *vlapic_vtx;
4007 
4008 	vmx = arg;
4009 
4010 	vlapic = malloc(sizeof(struct vlapic_vtx), M_VLAPIC, M_WAITOK | M_ZERO);
4011 	vlapic->vm = vmx->vm;
4012 	vlapic->vcpuid = vcpuid;
4013 	vlapic->apic_page = (struct LAPIC *)&vmx->apic_page[vcpuid];
4014 
4015 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
4016 	vlapic_vtx->pir_desc = &vmx->pir_desc[vcpuid];
4017 	vlapic_vtx->vmx = vmx;
4018 
4019 	if (tpr_shadowing) {
4020 		vlapic->ops.enable_x2apic_mode = vmx_enable_x2apic_mode_ts;
4021 	}
4022 
4023 	if (virtual_interrupt_delivery) {
4024 		vlapic->ops.set_intr_ready = vmx_set_intr_ready;
4025 		vlapic->ops.pending_intr = vmx_pending_intr;
4026 		vlapic->ops.intr_accepted = vmx_intr_accepted;
4027 		vlapic->ops.set_tmr = vmx_set_tmr;
4028 		vlapic->ops.enable_x2apic_mode = vmx_enable_x2apic_mode_vid;
4029 	}
4030 
4031 	if (posted_interrupts)
4032 		vlapic->ops.post_intr = vmx_post_intr;
4033 
4034 	vlapic_init(vlapic);
4035 
4036 	return (vlapic);
4037 }
4038 
4039 static void
4040 vmx_vlapic_cleanup(void *arg, struct vlapic *vlapic)
4041 {
4042 
4043 	vlapic_cleanup(vlapic);
4044 	free(vlapic, M_VLAPIC);
4045 }
4046 
4047 #ifdef BHYVE_SNAPSHOT
4048 static int
4049 vmx_snapshot(void *arg, struct vm_snapshot_meta *meta)
4050 {
4051 	struct vmx *vmx;
4052 	struct vmxctx *vmxctx;
4053 	int i;
4054 	int ret;
4055 
4056 	vmx = arg;
4057 
4058 	KASSERT(vmx != NULL, ("%s: arg was NULL", __func__));
4059 
4060 	for (i = 0; i < VM_MAXCPU; i++) {
4061 		SNAPSHOT_BUF_OR_LEAVE(vmx->guest_msrs[i],
4062 		      sizeof(vmx->guest_msrs[i]), meta, ret, done);
4063 
4064 		vmxctx = &vmx->ctx[i];
4065 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rdi, meta, ret, done);
4066 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rsi, meta, ret, done);
4067 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rdx, meta, ret, done);
4068 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rcx, meta, ret, done);
4069 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r8, meta, ret, done);
4070 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r9, meta, ret, done);
4071 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rax, meta, ret, done);
4072 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rbx, meta, ret, done);
4073 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rbp, meta, ret, done);
4074 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r10, meta, ret, done);
4075 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r11, meta, ret, done);
4076 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r12, meta, ret, done);
4077 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r13, meta, ret, done);
4078 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r14, meta, ret, done);
4079 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r15, meta, ret, done);
4080 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_cr2, meta, ret, done);
4081 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr0, meta, ret, done);
4082 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr1, meta, ret, done);
4083 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr2, meta, ret, done);
4084 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr3, meta, ret, done);
4085 		SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr6, meta, ret, done);
4086 	}
4087 
4088 done:
4089 	return (ret);
4090 }
4091 
4092 static int
4093 vmx_vmcx_snapshot(void *arg, struct vm_snapshot_meta *meta, int vcpu)
4094 {
4095 	struct vmcs *vmcs;
4096 	struct vmx *vmx;
4097 	int err, run, hostcpu;
4098 
4099 	vmx = (struct vmx *)arg;
4100 	err = 0;
4101 
4102 	KASSERT(arg != NULL, ("%s: arg was NULL", __func__));
4103 	vmcs = &vmx->vmcs[vcpu];
4104 
4105 	run = vcpu_is_running(vmx->vm, vcpu, &hostcpu);
4106 	if (run && hostcpu != curcpu) {
4107 		printf("%s: %s%d is running", __func__, vm_name(vmx->vm), vcpu);
4108 		return (EINVAL);
4109 	}
4110 
4111 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR0, meta);
4112 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR3, meta);
4113 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR4, meta);
4114 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_DR7, meta);
4115 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RSP, meta);
4116 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RIP, meta);
4117 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RFLAGS, meta);
4118 
4119 	/* Guest segments */
4120 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_ES, meta);
4121 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_ES, meta);
4122 
4123 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CS, meta);
4124 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_CS, meta);
4125 
4126 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_SS, meta);
4127 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_SS, meta);
4128 
4129 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_DS, meta);
4130 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_DS, meta);
4131 
4132 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_FS, meta);
4133 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_FS, meta);
4134 
4135 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_GS, meta);
4136 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_GS, meta);
4137 
4138 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_TR, meta);
4139 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_TR, meta);
4140 
4141 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_LDTR, meta);
4142 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_LDTR, meta);
4143 
4144 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_EFER, meta);
4145 
4146 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_IDTR, meta);
4147 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_GDTR, meta);
4148 
4149 	/* Guest page tables */
4150 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE0, meta);
4151 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE1, meta);
4152 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE2, meta);
4153 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE3, meta);
4154 
4155 	/* Other guest state */
4156 	err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_CS, meta);
4157 	err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_ESP, meta);
4158 	err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_EIP, meta);
4159 	err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_INTERRUPTIBILITY, meta);
4160 	err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_ACTIVITY, meta);
4161 	err += vmcs_snapshot_any(vmcs, run, VMCS_ENTRY_CTLS, meta);
4162 	err += vmcs_snapshot_any(vmcs, run, VMCS_EXIT_CTLS, meta);
4163 
4164 	return (err);
4165 }
4166 
4167 static int
4168 vmx_restore_tsc(void *arg, int vcpu, uint64_t offset)
4169 {
4170 	struct vmcs *vmcs;
4171 	struct vmx *vmx = (struct vmx *)arg;
4172 	int error, running, hostcpu;
4173 
4174 	KASSERT(arg != NULL, ("%s: arg was NULL", __func__));
4175 	vmcs = &vmx->vmcs[vcpu];
4176 
4177 	running = vcpu_is_running(vmx->vm, vcpu, &hostcpu);
4178 	if (running && hostcpu != curcpu) {
4179 		printf("%s: %s%d is running", __func__, vm_name(vmx->vm), vcpu);
4180 		return (EINVAL);
4181 	}
4182 
4183 	if (!running)
4184 		VMPTRLD(vmcs);
4185 
4186 	error = vmx_set_tsc_offset(vmx, vcpu, offset);
4187 
4188 	if (!running)
4189 		VMCLEAR(vmcs);
4190 	return (error);
4191 }
4192 #endif
4193 
4194 const struct vmm_ops vmm_ops_intel = {
4195 	.modinit	= vmx_modinit,
4196 	.modcleanup	= vmx_modcleanup,
4197 	.modresume	= vmx_modresume,
4198 	.init		= vmx_init,
4199 	.run		= vmx_run,
4200 	.cleanup	= vmx_cleanup,
4201 	.getreg		= vmx_getreg,
4202 	.setreg		= vmx_setreg,
4203 	.getdesc	= vmx_getdesc,
4204 	.setdesc	= vmx_setdesc,
4205 	.getcap		= vmx_getcap,
4206 	.setcap		= vmx_setcap,
4207 	.vmspace_alloc	= vmx_vmspace_alloc,
4208 	.vmspace_free	= vmx_vmspace_free,
4209 	.vlapic_init	= vmx_vlapic_init,
4210 	.vlapic_cleanup	= vmx_vlapic_cleanup,
4211 #ifdef BHYVE_SNAPSHOT
4212 	.snapshot	= vmx_snapshot,
4213 	.vmcx_snapshot	= vmx_vmcx_snapshot,
4214 	.restore_tsc	= vmx_restore_tsc,
4215 #endif
4216 };
4217