xref: /freebsd/sys/amd64/vmm/intel/vmx.c (revision 2e3507c25e42292b45a5482e116d278f5515d04d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 
30 #include <sys/cdefs.h>
31 #include "opt_bhyve_snapshot.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/smp.h>
36 #include <sys/kernel.h>
37 #include <sys/malloc.h>
38 #include <sys/pcpu.h>
39 #include <sys/proc.h>
40 #include <sys/reg.h>
41 #include <sys/smr.h>
42 #include <sys/sysctl.h>
43 
44 #include <vm/vm.h>
45 #include <vm/vm_extern.h>
46 #include <vm/pmap.h>
47 
48 #include <machine/psl.h>
49 #include <machine/cpufunc.h>
50 #include <machine/md_var.h>
51 #include <machine/segments.h>
52 #include <machine/smp.h>
53 #include <machine/specialreg.h>
54 #include <machine/vmparam.h>
55 
56 #include <machine/vmm.h>
57 #include <machine/vmm_dev.h>
58 #include <machine/vmm_instruction_emul.h>
59 #include <machine/vmm_snapshot.h>
60 
61 #include "vmm_lapic.h"
62 #include "vmm_host.h"
63 #include "vmm_ioport.h"
64 #include "vmm_ktr.h"
65 #include "vmm_stat.h"
66 #include "vatpic.h"
67 #include "vlapic.h"
68 #include "vlapic_priv.h"
69 
70 #include "ept.h"
71 #include "vmx_cpufunc.h"
72 #include "vmx.h"
73 #include "vmx_msr.h"
74 #include "x86.h"
75 #include "vmx_controls.h"
76 
77 #define	PINBASED_CTLS_ONE_SETTING					\
78 	(PINBASED_EXTINT_EXITING	|				\
79 	 PINBASED_NMI_EXITING		|				\
80 	 PINBASED_VIRTUAL_NMI)
81 #define	PINBASED_CTLS_ZERO_SETTING	0
82 
83 #define PROCBASED_CTLS_WINDOW_SETTING					\
84 	(PROCBASED_INT_WINDOW_EXITING	|				\
85 	 PROCBASED_NMI_WINDOW_EXITING)
86 
87 #define	PROCBASED_CTLS_ONE_SETTING					\
88 	(PROCBASED_SECONDARY_CONTROLS	|				\
89 	 PROCBASED_MWAIT_EXITING	|				\
90 	 PROCBASED_MONITOR_EXITING	|				\
91 	 PROCBASED_IO_EXITING		|				\
92 	 PROCBASED_MSR_BITMAPS		|				\
93 	 PROCBASED_CTLS_WINDOW_SETTING	|				\
94 	 PROCBASED_CR8_LOAD_EXITING	|				\
95 	 PROCBASED_CR8_STORE_EXITING)
96 #define	PROCBASED_CTLS_ZERO_SETTING	\
97 	(PROCBASED_CR3_LOAD_EXITING |	\
98 	PROCBASED_CR3_STORE_EXITING |	\
99 	PROCBASED_IO_BITMAPS)
100 
101 #define	PROCBASED_CTLS2_ONE_SETTING	PROCBASED2_ENABLE_EPT
102 #define	PROCBASED_CTLS2_ZERO_SETTING	0
103 
104 #define	VM_EXIT_CTLS_ONE_SETTING					\
105 	(VM_EXIT_SAVE_DEBUG_CONTROLS		|			\
106 	VM_EXIT_HOST_LMA			|			\
107 	VM_EXIT_SAVE_EFER			|			\
108 	VM_EXIT_LOAD_EFER			|			\
109 	VM_EXIT_ACKNOWLEDGE_INTERRUPT)
110 
111 #define	VM_EXIT_CTLS_ZERO_SETTING	0
112 
113 #define	VM_ENTRY_CTLS_ONE_SETTING					\
114 	(VM_ENTRY_LOAD_DEBUG_CONTROLS		|			\
115 	VM_ENTRY_LOAD_EFER)
116 
117 #define	VM_ENTRY_CTLS_ZERO_SETTING					\
118 	(VM_ENTRY_INTO_SMM			|			\
119 	VM_ENTRY_DEACTIVATE_DUAL_MONITOR)
120 
121 #define	HANDLED		1
122 #define	UNHANDLED	0
123 
124 static MALLOC_DEFINE(M_VMX, "vmx", "vmx");
125 static MALLOC_DEFINE(M_VLAPIC, "vlapic", "vlapic");
126 
127 bool vmx_have_msr_tsc_aux;
128 
129 SYSCTL_DECL(_hw_vmm);
130 SYSCTL_NODE(_hw_vmm, OID_AUTO, vmx, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
131     NULL);
132 
133 int vmxon_enabled[MAXCPU];
134 static uint8_t *vmxon_region;
135 
136 static uint32_t pinbased_ctls, procbased_ctls, procbased_ctls2;
137 static uint32_t exit_ctls, entry_ctls;
138 
139 static uint64_t cr0_ones_mask, cr0_zeros_mask;
140 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_ones_mask, CTLFLAG_RD,
141 	     &cr0_ones_mask, 0, NULL);
142 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr0_zeros_mask, CTLFLAG_RD,
143 	     &cr0_zeros_mask, 0, NULL);
144 
145 static uint64_t cr4_ones_mask, cr4_zeros_mask;
146 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_ones_mask, CTLFLAG_RD,
147 	     &cr4_ones_mask, 0, NULL);
148 SYSCTL_ULONG(_hw_vmm_vmx, OID_AUTO, cr4_zeros_mask, CTLFLAG_RD,
149 	     &cr4_zeros_mask, 0, NULL);
150 
151 static int vmx_initialized;
152 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, initialized, CTLFLAG_RD,
153 	   &vmx_initialized, 0, "Intel VMX initialized");
154 
155 /*
156  * Optional capabilities
157  */
158 static SYSCTL_NODE(_hw_vmm_vmx, OID_AUTO, cap,
159     CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
160     NULL);
161 
162 static int cap_halt_exit;
163 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, halt_exit, CTLFLAG_RD, &cap_halt_exit, 0,
164     "HLT triggers a VM-exit");
165 
166 static int cap_pause_exit;
167 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, pause_exit, CTLFLAG_RD, &cap_pause_exit,
168     0, "PAUSE triggers a VM-exit");
169 
170 static int cap_wbinvd_exit;
171 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, wbinvd_exit, CTLFLAG_RD, &cap_wbinvd_exit,
172     0, "WBINVD triggers a VM-exit");
173 
174 static int cap_rdpid;
175 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, rdpid, CTLFLAG_RD, &cap_rdpid, 0,
176     "Guests are allowed to use RDPID");
177 
178 static int cap_rdtscp;
179 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, rdtscp, CTLFLAG_RD, &cap_rdtscp, 0,
180     "Guests are allowed to use RDTSCP");
181 
182 static int cap_unrestricted_guest;
183 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, unrestricted_guest, CTLFLAG_RD,
184     &cap_unrestricted_guest, 0, "Unrestricted guests");
185 
186 static int cap_monitor_trap;
187 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, monitor_trap, CTLFLAG_RD,
188     &cap_monitor_trap, 0, "Monitor trap flag");
189 
190 static int cap_invpcid;
191 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, invpcid, CTLFLAG_RD, &cap_invpcid,
192     0, "Guests are allowed to use INVPCID");
193 
194 static int tpr_shadowing;
195 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, tpr_shadowing,
196     CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
197     &tpr_shadowing, 0, "TPR shadowing support");
198 
199 static int virtual_interrupt_delivery;
200 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, virtual_interrupt_delivery,
201     CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
202     &virtual_interrupt_delivery, 0, "APICv virtual interrupt delivery support");
203 
204 static int posted_interrupts;
205 SYSCTL_INT(_hw_vmm_vmx_cap, OID_AUTO, posted_interrupts,
206     CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
207     &posted_interrupts, 0, "APICv posted interrupt support");
208 
209 static int pirvec = -1;
210 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, posted_interrupt_vector, CTLFLAG_RD,
211     &pirvec, 0, "APICv posted interrupt vector");
212 
213 static struct unrhdr *vpid_unr;
214 static u_int vpid_alloc_failed;
215 SYSCTL_UINT(_hw_vmm_vmx, OID_AUTO, vpid_alloc_failed, CTLFLAG_RD,
216 	    &vpid_alloc_failed, 0, NULL);
217 
218 int guest_l1d_flush;
219 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, l1d_flush, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
220     &guest_l1d_flush, 0, NULL);
221 int guest_l1d_flush_sw;
222 SYSCTL_INT(_hw_vmm_vmx, OID_AUTO, l1d_flush_sw, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
223     &guest_l1d_flush_sw, 0, NULL);
224 
225 static struct msr_entry msr_load_list[1] __aligned(16);
226 
227 /*
228  * The definitions of SDT probes for VMX.
229  */
230 
231 SDT_PROBE_DEFINE3(vmm, vmx, exit, entry,
232     "struct vmx *", "int", "struct vm_exit *");
233 
234 SDT_PROBE_DEFINE4(vmm, vmx, exit, taskswitch,
235     "struct vmx *", "int", "struct vm_exit *", "struct vm_task_switch *");
236 
237 SDT_PROBE_DEFINE4(vmm, vmx, exit, craccess,
238     "struct vmx *", "int", "struct vm_exit *", "uint64_t");
239 
240 SDT_PROBE_DEFINE4(vmm, vmx, exit, rdmsr,
241     "struct vmx *", "int", "struct vm_exit *", "uint32_t");
242 
243 SDT_PROBE_DEFINE5(vmm, vmx, exit, wrmsr,
244     "struct vmx *", "int", "struct vm_exit *", "uint32_t", "uint64_t");
245 
246 SDT_PROBE_DEFINE3(vmm, vmx, exit, halt,
247     "struct vmx *", "int", "struct vm_exit *");
248 
249 SDT_PROBE_DEFINE3(vmm, vmx, exit, mtrap,
250     "struct vmx *", "int", "struct vm_exit *");
251 
252 SDT_PROBE_DEFINE3(vmm, vmx, exit, pause,
253     "struct vmx *", "int", "struct vm_exit *");
254 
255 SDT_PROBE_DEFINE3(vmm, vmx, exit, intrwindow,
256     "struct vmx *", "int", "struct vm_exit *");
257 
258 SDT_PROBE_DEFINE4(vmm, vmx, exit, interrupt,
259     "struct vmx *", "int", "struct vm_exit *", "uint32_t");
260 
261 SDT_PROBE_DEFINE3(vmm, vmx, exit, nmiwindow,
262     "struct vmx *", "int", "struct vm_exit *");
263 
264 SDT_PROBE_DEFINE3(vmm, vmx, exit, inout,
265     "struct vmx *", "int", "struct vm_exit *");
266 
267 SDT_PROBE_DEFINE3(vmm, vmx, exit, cpuid,
268     "struct vmx *", "int", "struct vm_exit *");
269 
270 SDT_PROBE_DEFINE5(vmm, vmx, exit, exception,
271     "struct vmx *", "int", "struct vm_exit *", "uint32_t", "int");
272 
273 SDT_PROBE_DEFINE5(vmm, vmx, exit, nestedfault,
274     "struct vmx *", "int", "struct vm_exit *", "uint64_t", "uint64_t");
275 
276 SDT_PROBE_DEFINE4(vmm, vmx, exit, mmiofault,
277     "struct vmx *", "int", "struct vm_exit *", "uint64_t");
278 
279 SDT_PROBE_DEFINE3(vmm, vmx, exit, eoi,
280     "struct vmx *", "int", "struct vm_exit *");
281 
282 SDT_PROBE_DEFINE3(vmm, vmx, exit, apicaccess,
283     "struct vmx *", "int", "struct vm_exit *");
284 
285 SDT_PROBE_DEFINE4(vmm, vmx, exit, apicwrite,
286     "struct vmx *", "int", "struct vm_exit *", "struct vlapic *");
287 
288 SDT_PROBE_DEFINE3(vmm, vmx, exit, xsetbv,
289     "struct vmx *", "int", "struct vm_exit *");
290 
291 SDT_PROBE_DEFINE3(vmm, vmx, exit, monitor,
292     "struct vmx *", "int", "struct vm_exit *");
293 
294 SDT_PROBE_DEFINE3(vmm, vmx, exit, mwait,
295     "struct vmx *", "int", "struct vm_exit *");
296 
297 SDT_PROBE_DEFINE3(vmm, vmx, exit, vminsn,
298     "struct vmx *", "int", "struct vm_exit *");
299 
300 SDT_PROBE_DEFINE4(vmm, vmx, exit, unknown,
301     "struct vmx *", "int", "struct vm_exit *", "uint32_t");
302 
303 SDT_PROBE_DEFINE4(vmm, vmx, exit, return,
304     "struct vmx *", "int", "struct vm_exit *", "int");
305 
306 /*
307  * Use the last page below 4GB as the APIC access address. This address is
308  * occupied by the boot firmware so it is guaranteed that it will not conflict
309  * with a page in system memory.
310  */
311 #define	APIC_ACCESS_ADDRESS	0xFFFFF000
312 
313 static int vmx_getdesc(void *vcpui, int reg, struct seg_desc *desc);
314 static int vmx_getreg(void *vcpui, int reg, uint64_t *retval);
315 static int vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val);
316 static void vmx_inject_pir(struct vlapic *vlapic);
317 #ifdef BHYVE_SNAPSHOT
318 static int vmx_restore_tsc(void *vcpui, uint64_t now);
319 #endif
320 
321 static inline bool
322 host_has_rdpid(void)
323 {
324 	return ((cpu_stdext_feature2 & CPUID_STDEXT2_RDPID) != 0);
325 }
326 
327 static inline bool
328 host_has_rdtscp(void)
329 {
330 	return ((amd_feature & AMDID_RDTSCP) != 0);
331 }
332 
333 #ifdef KTR
334 static const char *
335 exit_reason_to_str(int reason)
336 {
337 	static char reasonbuf[32];
338 
339 	switch (reason) {
340 	case EXIT_REASON_EXCEPTION:
341 		return "exception";
342 	case EXIT_REASON_EXT_INTR:
343 		return "extint";
344 	case EXIT_REASON_TRIPLE_FAULT:
345 		return "triplefault";
346 	case EXIT_REASON_INIT:
347 		return "init";
348 	case EXIT_REASON_SIPI:
349 		return "sipi";
350 	case EXIT_REASON_IO_SMI:
351 		return "iosmi";
352 	case EXIT_REASON_SMI:
353 		return "smi";
354 	case EXIT_REASON_INTR_WINDOW:
355 		return "intrwindow";
356 	case EXIT_REASON_NMI_WINDOW:
357 		return "nmiwindow";
358 	case EXIT_REASON_TASK_SWITCH:
359 		return "taskswitch";
360 	case EXIT_REASON_CPUID:
361 		return "cpuid";
362 	case EXIT_REASON_GETSEC:
363 		return "getsec";
364 	case EXIT_REASON_HLT:
365 		return "hlt";
366 	case EXIT_REASON_INVD:
367 		return "invd";
368 	case EXIT_REASON_INVLPG:
369 		return "invlpg";
370 	case EXIT_REASON_RDPMC:
371 		return "rdpmc";
372 	case EXIT_REASON_RDTSC:
373 		return "rdtsc";
374 	case EXIT_REASON_RSM:
375 		return "rsm";
376 	case EXIT_REASON_VMCALL:
377 		return "vmcall";
378 	case EXIT_REASON_VMCLEAR:
379 		return "vmclear";
380 	case EXIT_REASON_VMLAUNCH:
381 		return "vmlaunch";
382 	case EXIT_REASON_VMPTRLD:
383 		return "vmptrld";
384 	case EXIT_REASON_VMPTRST:
385 		return "vmptrst";
386 	case EXIT_REASON_VMREAD:
387 		return "vmread";
388 	case EXIT_REASON_VMRESUME:
389 		return "vmresume";
390 	case EXIT_REASON_VMWRITE:
391 		return "vmwrite";
392 	case EXIT_REASON_VMXOFF:
393 		return "vmxoff";
394 	case EXIT_REASON_VMXON:
395 		return "vmxon";
396 	case EXIT_REASON_CR_ACCESS:
397 		return "craccess";
398 	case EXIT_REASON_DR_ACCESS:
399 		return "draccess";
400 	case EXIT_REASON_INOUT:
401 		return "inout";
402 	case EXIT_REASON_RDMSR:
403 		return "rdmsr";
404 	case EXIT_REASON_WRMSR:
405 		return "wrmsr";
406 	case EXIT_REASON_INVAL_VMCS:
407 		return "invalvmcs";
408 	case EXIT_REASON_INVAL_MSR:
409 		return "invalmsr";
410 	case EXIT_REASON_MWAIT:
411 		return "mwait";
412 	case EXIT_REASON_MTF:
413 		return "mtf";
414 	case EXIT_REASON_MONITOR:
415 		return "monitor";
416 	case EXIT_REASON_PAUSE:
417 		return "pause";
418 	case EXIT_REASON_MCE_DURING_ENTRY:
419 		return "mce-during-entry";
420 	case EXIT_REASON_TPR:
421 		return "tpr";
422 	case EXIT_REASON_APIC_ACCESS:
423 		return "apic-access";
424 	case EXIT_REASON_GDTR_IDTR:
425 		return "gdtridtr";
426 	case EXIT_REASON_LDTR_TR:
427 		return "ldtrtr";
428 	case EXIT_REASON_EPT_FAULT:
429 		return "eptfault";
430 	case EXIT_REASON_EPT_MISCONFIG:
431 		return "eptmisconfig";
432 	case EXIT_REASON_INVEPT:
433 		return "invept";
434 	case EXIT_REASON_RDTSCP:
435 		return "rdtscp";
436 	case EXIT_REASON_VMX_PREEMPT:
437 		return "vmxpreempt";
438 	case EXIT_REASON_INVVPID:
439 		return "invvpid";
440 	case EXIT_REASON_WBINVD:
441 		return "wbinvd";
442 	case EXIT_REASON_XSETBV:
443 		return "xsetbv";
444 	case EXIT_REASON_APIC_WRITE:
445 		return "apic-write";
446 	default:
447 		snprintf(reasonbuf, sizeof(reasonbuf), "%d", reason);
448 		return (reasonbuf);
449 	}
450 }
451 #endif	/* KTR */
452 
453 static int
454 vmx_allow_x2apic_msrs(struct vmx *vmx)
455 {
456 	int i, error;
457 
458 	error = 0;
459 
460 	/*
461 	 * Allow readonly access to the following x2APIC MSRs from the guest.
462 	 */
463 	error += guest_msr_ro(vmx, MSR_APIC_ID);
464 	error += guest_msr_ro(vmx, MSR_APIC_VERSION);
465 	error += guest_msr_ro(vmx, MSR_APIC_LDR);
466 	error += guest_msr_ro(vmx, MSR_APIC_SVR);
467 
468 	for (i = 0; i < 8; i++)
469 		error += guest_msr_ro(vmx, MSR_APIC_ISR0 + i);
470 
471 	for (i = 0; i < 8; i++)
472 		error += guest_msr_ro(vmx, MSR_APIC_TMR0 + i);
473 
474 	for (i = 0; i < 8; i++)
475 		error += guest_msr_ro(vmx, MSR_APIC_IRR0 + i);
476 
477 	error += guest_msr_ro(vmx, MSR_APIC_ESR);
478 	error += guest_msr_ro(vmx, MSR_APIC_LVT_TIMER);
479 	error += guest_msr_ro(vmx, MSR_APIC_LVT_THERMAL);
480 	error += guest_msr_ro(vmx, MSR_APIC_LVT_PCINT);
481 	error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT0);
482 	error += guest_msr_ro(vmx, MSR_APIC_LVT_LINT1);
483 	error += guest_msr_ro(vmx, MSR_APIC_LVT_ERROR);
484 	error += guest_msr_ro(vmx, MSR_APIC_ICR_TIMER);
485 	error += guest_msr_ro(vmx, MSR_APIC_DCR_TIMER);
486 	error += guest_msr_ro(vmx, MSR_APIC_ICR);
487 
488 	/*
489 	 * Allow TPR, EOI and SELF_IPI MSRs to be read and written by the guest.
490 	 *
491 	 * These registers get special treatment described in the section
492 	 * "Virtualizing MSR-Based APIC Accesses".
493 	 */
494 	error += guest_msr_rw(vmx, MSR_APIC_TPR);
495 	error += guest_msr_rw(vmx, MSR_APIC_EOI);
496 	error += guest_msr_rw(vmx, MSR_APIC_SELF_IPI);
497 
498 	return (error);
499 }
500 
501 u_long
502 vmx_fix_cr0(u_long cr0)
503 {
504 
505 	return ((cr0 | cr0_ones_mask) & ~cr0_zeros_mask);
506 }
507 
508 u_long
509 vmx_fix_cr4(u_long cr4)
510 {
511 
512 	return ((cr4 | cr4_ones_mask) & ~cr4_zeros_mask);
513 }
514 
515 static void
516 vpid_free(int vpid)
517 {
518 	if (vpid < 0 || vpid > 0xffff)
519 		panic("vpid_free: invalid vpid %d", vpid);
520 
521 	/*
522 	 * VPIDs [0,vm_maxcpu] are special and are not allocated from
523 	 * the unit number allocator.
524 	 */
525 
526 	if (vpid > vm_maxcpu)
527 		free_unr(vpid_unr, vpid);
528 }
529 
530 static uint16_t
531 vpid_alloc(int vcpuid)
532 {
533 	int x;
534 
535 	/*
536 	 * If the "enable vpid" execution control is not enabled then the
537 	 * VPID is required to be 0 for all vcpus.
538 	 */
539 	if ((procbased_ctls2 & PROCBASED2_ENABLE_VPID) == 0)
540 		return (0);
541 
542 	/*
543 	 * Try to allocate a unique VPID for each from the unit number
544 	 * allocator.
545 	 */
546 	x = alloc_unr(vpid_unr);
547 
548 	if (x == -1) {
549 		atomic_add_int(&vpid_alloc_failed, 1);
550 
551 		/*
552 		 * If the unit number allocator does not have enough unique
553 		 * VPIDs then we need to allocate from the [1,vm_maxcpu] range.
554 		 *
555 		 * These VPIDs are not be unique across VMs but this does not
556 		 * affect correctness because the combined mappings are also
557 		 * tagged with the EP4TA which is unique for each VM.
558 		 *
559 		 * It is still sub-optimal because the invvpid will invalidate
560 		 * combined mappings for a particular VPID across all EP4TAs.
561 		 */
562 		return (vcpuid + 1);
563 	}
564 
565 	return (x);
566 }
567 
568 static void
569 vpid_init(void)
570 {
571 	/*
572 	 * VPID 0 is required when the "enable VPID" execution control is
573 	 * disabled.
574 	 *
575 	 * VPIDs [1,vm_maxcpu] are used as the "overflow namespace" when the
576 	 * unit number allocator does not have sufficient unique VPIDs to
577 	 * satisfy the allocation.
578 	 *
579 	 * The remaining VPIDs are managed by the unit number allocator.
580 	 */
581 	vpid_unr = new_unrhdr(vm_maxcpu + 1, 0xffff, NULL);
582 }
583 
584 static void
585 vmx_disable(void *arg __unused)
586 {
587 	struct invvpid_desc invvpid_desc = { 0 };
588 	struct invept_desc invept_desc = { 0 };
589 
590 	if (vmxon_enabled[curcpu]) {
591 		/*
592 		 * See sections 25.3.3.3 and 25.3.3.4 in Intel Vol 3b.
593 		 *
594 		 * VMXON or VMXOFF are not required to invalidate any TLB
595 		 * caching structures. This prevents potential retention of
596 		 * cached information in the TLB between distinct VMX episodes.
597 		 */
598 		invvpid(INVVPID_TYPE_ALL_CONTEXTS, invvpid_desc);
599 		invept(INVEPT_TYPE_ALL_CONTEXTS, invept_desc);
600 		vmxoff();
601 	}
602 	load_cr4(rcr4() & ~CR4_VMXE);
603 }
604 
605 static int
606 vmx_modcleanup(void)
607 {
608 
609 	if (pirvec >= 0)
610 		lapic_ipi_free(pirvec);
611 
612 	if (vpid_unr != NULL) {
613 		delete_unrhdr(vpid_unr);
614 		vpid_unr = NULL;
615 	}
616 
617 	if (nmi_flush_l1d_sw == 1)
618 		nmi_flush_l1d_sw = 0;
619 
620 	smp_rendezvous(NULL, vmx_disable, NULL, NULL);
621 
622 	if (vmxon_region != NULL)
623 		kmem_free(vmxon_region, (mp_maxid + 1) * PAGE_SIZE);
624 
625 	return (0);
626 }
627 
628 static void
629 vmx_enable(void *arg __unused)
630 {
631 	int error;
632 	uint64_t feature_control;
633 
634 	feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL);
635 	if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 0 ||
636 	    (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) {
637 		wrmsr(MSR_IA32_FEATURE_CONTROL,
638 		    feature_control | IA32_FEATURE_CONTROL_VMX_EN |
639 		    IA32_FEATURE_CONTROL_LOCK);
640 	}
641 
642 	load_cr4(rcr4() | CR4_VMXE);
643 
644 	*(uint32_t *)&vmxon_region[curcpu * PAGE_SIZE] = vmx_revision();
645 	error = vmxon(&vmxon_region[curcpu * PAGE_SIZE]);
646 	if (error == 0)
647 		vmxon_enabled[curcpu] = 1;
648 }
649 
650 static void
651 vmx_modresume(void)
652 {
653 
654 	if (vmxon_enabled[curcpu])
655 		vmxon(&vmxon_region[curcpu * PAGE_SIZE]);
656 }
657 
658 static int
659 vmx_modinit(int ipinum)
660 {
661 	int error;
662 	uint64_t basic, fixed0, fixed1, feature_control;
663 	uint32_t tmp, procbased2_vid_bits;
664 
665 	/* CPUID.1:ECX[bit 5] must be 1 for processor to support VMX */
666 	if (!(cpu_feature2 & CPUID2_VMX)) {
667 		printf("vmx_modinit: processor does not support VMX "
668 		    "operation\n");
669 		return (ENXIO);
670 	}
671 
672 	/*
673 	 * Verify that MSR_IA32_FEATURE_CONTROL lock and VMXON enable bits
674 	 * are set (bits 0 and 2 respectively).
675 	 */
676 	feature_control = rdmsr(MSR_IA32_FEATURE_CONTROL);
677 	if ((feature_control & IA32_FEATURE_CONTROL_LOCK) == 1 &&
678 	    (feature_control & IA32_FEATURE_CONTROL_VMX_EN) == 0) {
679 		printf("vmx_modinit: VMX operation disabled by BIOS\n");
680 		return (ENXIO);
681 	}
682 
683 	/*
684 	 * Verify capabilities MSR_VMX_BASIC:
685 	 * - bit 54 indicates support for INS/OUTS decoding
686 	 */
687 	basic = rdmsr(MSR_VMX_BASIC);
688 	if ((basic & (1UL << 54)) == 0) {
689 		printf("vmx_modinit: processor does not support desired basic "
690 		    "capabilities\n");
691 		return (EINVAL);
692 	}
693 
694 	/* Check support for primary processor-based VM-execution controls */
695 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
696 			       MSR_VMX_TRUE_PROCBASED_CTLS,
697 			       PROCBASED_CTLS_ONE_SETTING,
698 			       PROCBASED_CTLS_ZERO_SETTING, &procbased_ctls);
699 	if (error) {
700 		printf("vmx_modinit: processor does not support desired "
701 		    "primary processor-based controls\n");
702 		return (error);
703 	}
704 
705 	/* Clear the processor-based ctl bits that are set on demand */
706 	procbased_ctls &= ~PROCBASED_CTLS_WINDOW_SETTING;
707 
708 	/* Check support for secondary processor-based VM-execution controls */
709 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
710 			       MSR_VMX_PROCBASED_CTLS2,
711 			       PROCBASED_CTLS2_ONE_SETTING,
712 			       PROCBASED_CTLS2_ZERO_SETTING, &procbased_ctls2);
713 	if (error) {
714 		printf("vmx_modinit: processor does not support desired "
715 		    "secondary processor-based controls\n");
716 		return (error);
717 	}
718 
719 	/* Check support for VPID */
720 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2,
721 			       PROCBASED2_ENABLE_VPID, 0, &tmp);
722 	if (error == 0)
723 		procbased_ctls2 |= PROCBASED2_ENABLE_VPID;
724 
725 	/* Check support for pin-based VM-execution controls */
726 	error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS,
727 			       MSR_VMX_TRUE_PINBASED_CTLS,
728 			       PINBASED_CTLS_ONE_SETTING,
729 			       PINBASED_CTLS_ZERO_SETTING, &pinbased_ctls);
730 	if (error) {
731 		printf("vmx_modinit: processor does not support desired "
732 		    "pin-based controls\n");
733 		return (error);
734 	}
735 
736 	/* Check support for VM-exit controls */
737 	error = vmx_set_ctlreg(MSR_VMX_EXIT_CTLS, MSR_VMX_TRUE_EXIT_CTLS,
738 			       VM_EXIT_CTLS_ONE_SETTING,
739 			       VM_EXIT_CTLS_ZERO_SETTING,
740 			       &exit_ctls);
741 	if (error) {
742 		printf("vmx_modinit: processor does not support desired "
743 		    "exit controls\n");
744 		return (error);
745 	}
746 
747 	/* Check support for VM-entry controls */
748 	error = vmx_set_ctlreg(MSR_VMX_ENTRY_CTLS, MSR_VMX_TRUE_ENTRY_CTLS,
749 	    VM_ENTRY_CTLS_ONE_SETTING, VM_ENTRY_CTLS_ZERO_SETTING,
750 	    &entry_ctls);
751 	if (error) {
752 		printf("vmx_modinit: processor does not support desired "
753 		    "entry controls\n");
754 		return (error);
755 	}
756 
757 	/*
758 	 * Check support for optional features by testing them
759 	 * as individual bits
760 	 */
761 	cap_halt_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
762 					MSR_VMX_TRUE_PROCBASED_CTLS,
763 					PROCBASED_HLT_EXITING, 0,
764 					&tmp) == 0);
765 
766 	cap_monitor_trap = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
767 					MSR_VMX_PROCBASED_CTLS,
768 					PROCBASED_MTF, 0,
769 					&tmp) == 0);
770 
771 	cap_pause_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
772 					 MSR_VMX_TRUE_PROCBASED_CTLS,
773 					 PROCBASED_PAUSE_EXITING, 0,
774 					 &tmp) == 0);
775 
776 	cap_wbinvd_exit = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
777 					MSR_VMX_PROCBASED_CTLS2,
778 					PROCBASED2_WBINVD_EXITING,
779 					0,
780 					&tmp) == 0);
781 
782 	/*
783 	 * Check support for RDPID and/or RDTSCP.
784 	 *
785 	 * Support a pass-through-based implementation of these via the
786 	 * "enable RDTSCP" VM-execution control and the "RDTSC exiting"
787 	 * VM-execution control.
788 	 *
789 	 * The "enable RDTSCP" VM-execution control applies to both RDPID
790 	 * and RDTSCP (see SDM volume 3, section 25.3, "Changes to
791 	 * Instruction Behavior in VMX Non-root operation"); this is why
792 	 * only this VM-execution control needs to be enabled in order to
793 	 * enable passing through whichever of RDPID and/or RDTSCP are
794 	 * supported by the host.
795 	 *
796 	 * The "RDTSC exiting" VM-execution control applies to both RDTSC
797 	 * and RDTSCP (again, per SDM volume 3, section 25.3), and is
798 	 * already set up for RDTSC and RDTSCP pass-through by the current
799 	 * implementation of RDTSC.
800 	 *
801 	 * Although RDPID and RDTSCP are optional capabilities, since there
802 	 * does not currently seem to be a use case for enabling/disabling
803 	 * these via libvmmapi, choose not to support this and, instead,
804 	 * just statically always enable or always disable this support
805 	 * across all vCPUs on all VMs. (Note that there may be some
806 	 * complications to providing this functionality, e.g., the MSR
807 	 * bitmap is currently per-VM rather than per-vCPU while the
808 	 * capability API wants to be able to control capabilities on a
809 	 * per-vCPU basis).
810 	 */
811 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
812 			       MSR_VMX_PROCBASED_CTLS2,
813 			       PROCBASED2_ENABLE_RDTSCP, 0, &tmp);
814 	cap_rdpid = error == 0 && host_has_rdpid();
815 	cap_rdtscp = error == 0 && host_has_rdtscp();
816 	if (cap_rdpid || cap_rdtscp) {
817 		procbased_ctls2 |= PROCBASED2_ENABLE_RDTSCP;
818 		vmx_have_msr_tsc_aux = true;
819 	}
820 
821 	cap_unrestricted_guest = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
822 					MSR_VMX_PROCBASED_CTLS2,
823 					PROCBASED2_UNRESTRICTED_GUEST, 0,
824 				        &tmp) == 0);
825 
826 	cap_invpcid = (vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2,
827 	    MSR_VMX_PROCBASED_CTLS2, PROCBASED2_ENABLE_INVPCID, 0,
828 	    &tmp) == 0);
829 
830 	/*
831 	 * Check support for TPR shadow.
832 	 */
833 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS,
834 	    MSR_VMX_TRUE_PROCBASED_CTLS, PROCBASED_USE_TPR_SHADOW, 0,
835 	    &tmp);
836 	if (error == 0) {
837 		tpr_shadowing = 1;
838 #ifndef BURN_BRIDGES
839 		TUNABLE_INT_FETCH("hw.vmm.vmx.use_tpr_shadowing",
840 		    &tpr_shadowing);
841 #endif
842 		TUNABLE_INT_FETCH("hw.vmm.vmx.cap.tpr_shadowing",
843 		    &tpr_shadowing);
844 	}
845 
846 	if (tpr_shadowing) {
847 		procbased_ctls |= PROCBASED_USE_TPR_SHADOW;
848 		procbased_ctls &= ~PROCBASED_CR8_LOAD_EXITING;
849 		procbased_ctls &= ~PROCBASED_CR8_STORE_EXITING;
850 	}
851 
852 	/*
853 	 * Check support for virtual interrupt delivery.
854 	 */
855 	procbased2_vid_bits = (PROCBASED2_VIRTUALIZE_APIC_ACCESSES |
856 	    PROCBASED2_VIRTUALIZE_X2APIC_MODE |
857 	    PROCBASED2_APIC_REGISTER_VIRTUALIZATION |
858 	    PROCBASED2_VIRTUAL_INTERRUPT_DELIVERY);
859 
860 	error = vmx_set_ctlreg(MSR_VMX_PROCBASED_CTLS2, MSR_VMX_PROCBASED_CTLS2,
861 	    procbased2_vid_bits, 0, &tmp);
862 	if (error == 0 && tpr_shadowing) {
863 		virtual_interrupt_delivery = 1;
864 #ifndef BURN_BRIDGES
865 		TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_vid",
866 		    &virtual_interrupt_delivery);
867 #endif
868 		TUNABLE_INT_FETCH("hw.vmm.vmx.cap.virtual_interrupt_delivery",
869 		    &virtual_interrupt_delivery);
870 	}
871 
872 	if (virtual_interrupt_delivery) {
873 		procbased_ctls |= PROCBASED_USE_TPR_SHADOW;
874 		procbased_ctls2 |= procbased2_vid_bits;
875 		procbased_ctls2 &= ~PROCBASED2_VIRTUALIZE_X2APIC_MODE;
876 
877 		/*
878 		 * Check for Posted Interrupts only if Virtual Interrupt
879 		 * Delivery is enabled.
880 		 */
881 		error = vmx_set_ctlreg(MSR_VMX_PINBASED_CTLS,
882 		    MSR_VMX_TRUE_PINBASED_CTLS, PINBASED_POSTED_INTERRUPT, 0,
883 		    &tmp);
884 		if (error == 0) {
885 			pirvec = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) :
886 			    &IDTVEC(justreturn));
887 			if (pirvec < 0) {
888 				if (bootverbose) {
889 					printf("vmx_modinit: unable to "
890 					    "allocate posted interrupt "
891 					    "vector\n");
892 				}
893 			} else {
894 				posted_interrupts = 1;
895 #ifndef BURN_BRIDGES
896 				TUNABLE_INT_FETCH("hw.vmm.vmx.use_apic_pir",
897 				    &posted_interrupts);
898 #endif
899 				TUNABLE_INT_FETCH("hw.vmm.vmx.cap.posted_interrupts",
900 				    &posted_interrupts);
901 			}
902 		}
903 	}
904 
905 	if (posted_interrupts)
906 		    pinbased_ctls |= PINBASED_POSTED_INTERRUPT;
907 
908 	/* Initialize EPT */
909 	error = ept_init(ipinum);
910 	if (error) {
911 		printf("vmx_modinit: ept initialization failed (%d)\n", error);
912 		return (error);
913 	}
914 
915 	guest_l1d_flush = (cpu_ia32_arch_caps &
916 	    IA32_ARCH_CAP_SKIP_L1DFL_VMENTRY) == 0;
917 #ifndef BURN_BRIDGES
918 	TUNABLE_INT_FETCH("hw.vmm.l1d_flush", &guest_l1d_flush);
919 #endif
920 	TUNABLE_INT_FETCH("hw.vmm.vmx.l1d_flush", &guest_l1d_flush);
921 
922 	/*
923 	 * L1D cache flush is enabled.  Use IA32_FLUSH_CMD MSR when
924 	 * available.  Otherwise fall back to the software flush
925 	 * method which loads enough data from the kernel text to
926 	 * flush existing L1D content, both on VMX entry and on NMI
927 	 * return.
928 	 */
929 	if (guest_l1d_flush) {
930 		if ((cpu_stdext_feature3 & CPUID_STDEXT3_L1D_FLUSH) == 0) {
931 			guest_l1d_flush_sw = 1;
932 #ifndef BURN_BRIDGES
933 			TUNABLE_INT_FETCH("hw.vmm.l1d_flush_sw",
934 			    &guest_l1d_flush_sw);
935 #endif
936 			TUNABLE_INT_FETCH("hw.vmm.vmx.l1d_flush_sw",
937 			    &guest_l1d_flush_sw);
938 		}
939 		if (guest_l1d_flush_sw) {
940 			if (nmi_flush_l1d_sw <= 1)
941 				nmi_flush_l1d_sw = 1;
942 		} else {
943 			msr_load_list[0].index = MSR_IA32_FLUSH_CMD;
944 			msr_load_list[0].val = IA32_FLUSH_CMD_L1D;
945 		}
946 	}
947 
948 	/*
949 	 * Stash the cr0 and cr4 bits that must be fixed to 0 or 1
950 	 */
951 	fixed0 = rdmsr(MSR_VMX_CR0_FIXED0);
952 	fixed1 = rdmsr(MSR_VMX_CR0_FIXED1);
953 	cr0_ones_mask = fixed0 & fixed1;
954 	cr0_zeros_mask = ~fixed0 & ~fixed1;
955 
956 	/*
957 	 * CR0_PE and CR0_PG can be set to zero in VMX non-root operation
958 	 * if unrestricted guest execution is allowed.
959 	 */
960 	if (cap_unrestricted_guest)
961 		cr0_ones_mask &= ~(CR0_PG | CR0_PE);
962 
963 	/*
964 	 * Do not allow the guest to set CR0_NW or CR0_CD.
965 	 */
966 	cr0_zeros_mask |= (CR0_NW | CR0_CD);
967 
968 	fixed0 = rdmsr(MSR_VMX_CR4_FIXED0);
969 	fixed1 = rdmsr(MSR_VMX_CR4_FIXED1);
970 	cr4_ones_mask = fixed0 & fixed1;
971 	cr4_zeros_mask = ~fixed0 & ~fixed1;
972 
973 	vpid_init();
974 
975 	vmx_msr_init();
976 
977 	/* enable VMX operation */
978 	vmxon_region = kmem_malloc((mp_maxid + 1) * PAGE_SIZE,
979 	    M_WAITOK | M_ZERO);
980 	smp_rendezvous(NULL, vmx_enable, NULL, NULL);
981 
982 	vmx_initialized = 1;
983 
984 	return (0);
985 }
986 
987 static void
988 vmx_trigger_hostintr(int vector)
989 {
990 	uintptr_t func;
991 	struct gate_descriptor *gd;
992 
993 	gd = &idt[vector];
994 
995 	KASSERT(vector >= 32 && vector <= 255, ("vmx_trigger_hostintr: "
996 	    "invalid vector %d", vector));
997 	KASSERT(gd->gd_p == 1, ("gate descriptor for vector %d not present",
998 	    vector));
999 	KASSERT(gd->gd_type == SDT_SYSIGT, ("gate descriptor for vector %d "
1000 	    "has invalid type %d", vector, gd->gd_type));
1001 	KASSERT(gd->gd_dpl == SEL_KPL, ("gate descriptor for vector %d "
1002 	    "has invalid dpl %d", vector, gd->gd_dpl));
1003 	KASSERT(gd->gd_selector == GSEL(GCODE_SEL, SEL_KPL), ("gate descriptor "
1004 	    "for vector %d has invalid selector %d", vector, gd->gd_selector));
1005 	KASSERT(gd->gd_ist == 0, ("gate descriptor for vector %d has invalid "
1006 	    "IST %d", vector, gd->gd_ist));
1007 
1008 	func = ((long)gd->gd_hioffset << 16 | gd->gd_looffset);
1009 	vmx_call_isr(func);
1010 }
1011 
1012 static int
1013 vmx_setup_cr_shadow(int which, struct vmcs *vmcs, uint32_t initial)
1014 {
1015 	int error, mask_ident, shadow_ident;
1016 	uint64_t mask_value;
1017 
1018 	if (which != 0 && which != 4)
1019 		panic("vmx_setup_cr_shadow: unknown cr%d", which);
1020 
1021 	if (which == 0) {
1022 		mask_ident = VMCS_CR0_MASK;
1023 		mask_value = cr0_ones_mask | cr0_zeros_mask;
1024 		shadow_ident = VMCS_CR0_SHADOW;
1025 	} else {
1026 		mask_ident = VMCS_CR4_MASK;
1027 		mask_value = cr4_ones_mask | cr4_zeros_mask;
1028 		shadow_ident = VMCS_CR4_SHADOW;
1029 	}
1030 
1031 	error = vmcs_setreg(vmcs, 0, VMCS_IDENT(mask_ident), mask_value);
1032 	if (error)
1033 		return (error);
1034 
1035 	error = vmcs_setreg(vmcs, 0, VMCS_IDENT(shadow_ident), initial);
1036 	if (error)
1037 		return (error);
1038 
1039 	return (0);
1040 }
1041 #define	vmx_setup_cr0_shadow(vmcs,init)	vmx_setup_cr_shadow(0, (vmcs), (init))
1042 #define	vmx_setup_cr4_shadow(vmcs,init)	vmx_setup_cr_shadow(4, (vmcs), (init))
1043 
1044 static void *
1045 vmx_init(struct vm *vm, pmap_t pmap)
1046 {
1047 	int error __diagused;
1048 	struct vmx *vmx;
1049 
1050 	vmx = malloc(sizeof(struct vmx), M_VMX, M_WAITOK | M_ZERO);
1051 	vmx->vm = vm;
1052 
1053 	vmx->eptp = eptp(vtophys((vm_offset_t)pmap->pm_pmltop));
1054 
1055 	/*
1056 	 * Clean up EPTP-tagged guest physical and combined mappings
1057 	 *
1058 	 * VMX transitions are not required to invalidate any guest physical
1059 	 * mappings. So, it may be possible for stale guest physical mappings
1060 	 * to be present in the processor TLBs.
1061 	 *
1062 	 * Combined mappings for this EP4TA are also invalidated for all VPIDs.
1063 	 */
1064 	ept_invalidate_mappings(vmx->eptp);
1065 
1066 	vmx->msr_bitmap = malloc_aligned(PAGE_SIZE, PAGE_SIZE, M_VMX,
1067 	    M_WAITOK | M_ZERO);
1068 	msr_bitmap_initialize(vmx->msr_bitmap);
1069 
1070 	/*
1071 	 * It is safe to allow direct access to MSR_GSBASE and MSR_FSBASE.
1072 	 * The guest FSBASE and GSBASE are saved and restored during
1073 	 * vm-exit and vm-entry respectively. The host FSBASE and GSBASE are
1074 	 * always restored from the vmcs host state area on vm-exit.
1075 	 *
1076 	 * The SYSENTER_CS/ESP/EIP MSRs are identical to FS/GSBASE in
1077 	 * how they are saved/restored so can be directly accessed by the
1078 	 * guest.
1079 	 *
1080 	 * MSR_EFER is saved and restored in the guest VMCS area on a
1081 	 * VM exit and entry respectively. It is also restored from the
1082 	 * host VMCS area on a VM exit.
1083 	 *
1084 	 * The TSC MSR is exposed read-only. Writes are disallowed as
1085 	 * that will impact the host TSC.  If the guest does a write
1086 	 * the "use TSC offsetting" execution control is enabled and the
1087 	 * difference between the host TSC and the guest TSC is written
1088 	 * into the TSC offset in the VMCS.
1089 	 *
1090 	 * Guest TSC_AUX support is enabled if any of guest RDPID and/or
1091 	 * guest RDTSCP support are enabled (since, as per Table 2-2 in SDM
1092 	 * volume 4, TSC_AUX is supported if any of RDPID and/or RDTSCP are
1093 	 * supported). If guest TSC_AUX support is enabled, TSC_AUX is
1094 	 * exposed read-only so that the VMM can do one fewer MSR read per
1095 	 * exit than if this register were exposed read-write; the guest
1096 	 * restore value can be updated during guest writes (expected to be
1097 	 * rare) instead of during all exits (common).
1098 	 */
1099 	if (guest_msr_rw(vmx, MSR_GSBASE) ||
1100 	    guest_msr_rw(vmx, MSR_FSBASE) ||
1101 	    guest_msr_rw(vmx, MSR_SYSENTER_CS_MSR) ||
1102 	    guest_msr_rw(vmx, MSR_SYSENTER_ESP_MSR) ||
1103 	    guest_msr_rw(vmx, MSR_SYSENTER_EIP_MSR) ||
1104 	    guest_msr_rw(vmx, MSR_EFER) ||
1105 	    guest_msr_ro(vmx, MSR_TSC) ||
1106 	    ((cap_rdpid || cap_rdtscp) && guest_msr_ro(vmx, MSR_TSC_AUX)))
1107 		panic("vmx_init: error setting guest msr access");
1108 
1109 	if (virtual_interrupt_delivery) {
1110 		error = vm_map_mmio(vm, DEFAULT_APIC_BASE, PAGE_SIZE,
1111 		    APIC_ACCESS_ADDRESS);
1112 		/* XXX this should really return an error to the caller */
1113 		KASSERT(error == 0, ("vm_map_mmio(apicbase) error %d", error));
1114 	}
1115 
1116 	vmx->pmap = pmap;
1117 	return (vmx);
1118 }
1119 
1120 static void *
1121 vmx_vcpu_init(void *vmi, struct vcpu *vcpu1, int vcpuid)
1122 {
1123 	struct vmx *vmx = vmi;
1124 	struct vmcs *vmcs;
1125 	struct vmx_vcpu *vcpu;
1126 	uint32_t exc_bitmap;
1127 	uint16_t vpid;
1128 	int error;
1129 
1130 	vpid = vpid_alloc(vcpuid);
1131 
1132 	vcpu = malloc(sizeof(*vcpu), M_VMX, M_WAITOK | M_ZERO);
1133 	vcpu->vmx = vmx;
1134 	vcpu->vcpu = vcpu1;
1135 	vcpu->vcpuid = vcpuid;
1136 	vcpu->vmcs = malloc_aligned(sizeof(*vmcs), PAGE_SIZE, M_VMX,
1137 	    M_WAITOK | M_ZERO);
1138 	vcpu->apic_page = malloc_aligned(PAGE_SIZE, PAGE_SIZE, M_VMX,
1139 	    M_WAITOK | M_ZERO);
1140 	vcpu->pir_desc = malloc_aligned(sizeof(*vcpu->pir_desc), 64, M_VMX,
1141 	    M_WAITOK | M_ZERO);
1142 
1143 	vmcs = vcpu->vmcs;
1144 	vmcs->identifier = vmx_revision();
1145 	error = vmclear(vmcs);
1146 	if (error != 0) {
1147 		panic("vmx_init: vmclear error %d on vcpu %d\n",
1148 		    error, vcpuid);
1149 	}
1150 
1151 	vmx_msr_guest_init(vmx, vcpu);
1152 
1153 	error = vmcs_init(vmcs);
1154 	KASSERT(error == 0, ("vmcs_init error %d", error));
1155 
1156 	VMPTRLD(vmcs);
1157 	error = 0;
1158 	error += vmwrite(VMCS_HOST_RSP, (u_long)&vcpu->ctx);
1159 	error += vmwrite(VMCS_EPTP, vmx->eptp);
1160 	error += vmwrite(VMCS_PIN_BASED_CTLS, pinbased_ctls);
1161 	error += vmwrite(VMCS_PRI_PROC_BASED_CTLS, procbased_ctls);
1162 	if (vcpu_trap_wbinvd(vcpu->vcpu)) {
1163 		KASSERT(cap_wbinvd_exit, ("WBINVD trap not available"));
1164 		procbased_ctls2 |= PROCBASED2_WBINVD_EXITING;
1165 	}
1166 	error += vmwrite(VMCS_SEC_PROC_BASED_CTLS, procbased_ctls2);
1167 	error += vmwrite(VMCS_EXIT_CTLS, exit_ctls);
1168 	error += vmwrite(VMCS_ENTRY_CTLS, entry_ctls);
1169 	error += vmwrite(VMCS_MSR_BITMAP, vtophys(vmx->msr_bitmap));
1170 	error += vmwrite(VMCS_VPID, vpid);
1171 
1172 	if (guest_l1d_flush && !guest_l1d_flush_sw) {
1173 		vmcs_write(VMCS_ENTRY_MSR_LOAD, pmap_kextract(
1174 			(vm_offset_t)&msr_load_list[0]));
1175 		vmcs_write(VMCS_ENTRY_MSR_LOAD_COUNT,
1176 		    nitems(msr_load_list));
1177 		vmcs_write(VMCS_EXIT_MSR_STORE, 0);
1178 		vmcs_write(VMCS_EXIT_MSR_STORE_COUNT, 0);
1179 	}
1180 
1181 	/* exception bitmap */
1182 	if (vcpu_trace_exceptions(vcpu->vcpu))
1183 		exc_bitmap = 0xffffffff;
1184 	else
1185 		exc_bitmap = 1 << IDT_MC;
1186 	error += vmwrite(VMCS_EXCEPTION_BITMAP, exc_bitmap);
1187 
1188 	vcpu->ctx.guest_dr6 = DBREG_DR6_RESERVED1;
1189 	error += vmwrite(VMCS_GUEST_DR7, DBREG_DR7_RESERVED1);
1190 
1191 	if (tpr_shadowing) {
1192 		error += vmwrite(VMCS_VIRTUAL_APIC, vtophys(vcpu->apic_page));
1193 	}
1194 
1195 	if (virtual_interrupt_delivery) {
1196 		error += vmwrite(VMCS_APIC_ACCESS, APIC_ACCESS_ADDRESS);
1197 		error += vmwrite(VMCS_EOI_EXIT0, 0);
1198 		error += vmwrite(VMCS_EOI_EXIT1, 0);
1199 		error += vmwrite(VMCS_EOI_EXIT2, 0);
1200 		error += vmwrite(VMCS_EOI_EXIT3, 0);
1201 	}
1202 	if (posted_interrupts) {
1203 		error += vmwrite(VMCS_PIR_VECTOR, pirvec);
1204 		error += vmwrite(VMCS_PIR_DESC, vtophys(vcpu->pir_desc));
1205 	}
1206 	VMCLEAR(vmcs);
1207 	KASSERT(error == 0, ("vmx_init: error customizing the vmcs"));
1208 
1209 	vcpu->cap.set = 0;
1210 	vcpu->cap.set |= cap_rdpid != 0 ? 1 << VM_CAP_RDPID : 0;
1211 	vcpu->cap.set |= cap_rdtscp != 0 ? 1 << VM_CAP_RDTSCP : 0;
1212 	vcpu->cap.proc_ctls = procbased_ctls;
1213 	vcpu->cap.proc_ctls2 = procbased_ctls2;
1214 	vcpu->cap.exc_bitmap = exc_bitmap;
1215 
1216 	vcpu->state.nextrip = ~0;
1217 	vcpu->state.lastcpu = NOCPU;
1218 	vcpu->state.vpid = vpid;
1219 
1220 	/*
1221 	 * Set up the CR0/4 shadows, and init the read shadow
1222 	 * to the power-on register value from the Intel Sys Arch.
1223 	 *  CR0 - 0x60000010
1224 	 *  CR4 - 0
1225 	 */
1226 	error = vmx_setup_cr0_shadow(vmcs, 0x60000010);
1227 	if (error != 0)
1228 		panic("vmx_setup_cr0_shadow %d", error);
1229 
1230 	error = vmx_setup_cr4_shadow(vmcs, 0);
1231 	if (error != 0)
1232 		panic("vmx_setup_cr4_shadow %d", error);
1233 
1234 	vcpu->ctx.pmap = vmx->pmap;
1235 
1236 	return (vcpu);
1237 }
1238 
1239 static int
1240 vmx_handle_cpuid(struct vmx_vcpu *vcpu, struct vmxctx *vmxctx)
1241 {
1242 	int handled;
1243 
1244 	handled = x86_emulate_cpuid(vcpu->vcpu, (uint64_t *)&vmxctx->guest_rax,
1245 	    (uint64_t *)&vmxctx->guest_rbx, (uint64_t *)&vmxctx->guest_rcx,
1246 	    (uint64_t *)&vmxctx->guest_rdx);
1247 	return (handled);
1248 }
1249 
1250 static __inline void
1251 vmx_run_trace(struct vmx_vcpu *vcpu)
1252 {
1253 	VMX_CTR1(vcpu, "Resume execution at %#lx", vmcs_guest_rip());
1254 }
1255 
1256 static __inline void
1257 vmx_exit_trace(struct vmx_vcpu *vcpu, uint64_t rip, uint32_t exit_reason,
1258     int handled)
1259 {
1260 	VMX_CTR3(vcpu, "%s %s vmexit at 0x%0lx",
1261 		 handled ? "handled" : "unhandled",
1262 		 exit_reason_to_str(exit_reason), rip);
1263 }
1264 
1265 static __inline void
1266 vmx_astpending_trace(struct vmx_vcpu *vcpu, uint64_t rip)
1267 {
1268 	VMX_CTR1(vcpu, "astpending vmexit at 0x%0lx", rip);
1269 }
1270 
1271 static VMM_STAT_INTEL(VCPU_INVVPID_SAVED, "Number of vpid invalidations saved");
1272 static VMM_STAT_INTEL(VCPU_INVVPID_DONE, "Number of vpid invalidations done");
1273 
1274 /*
1275  * Invalidate guest mappings identified by its vpid from the TLB.
1276  */
1277 static __inline void
1278 vmx_invvpid(struct vmx *vmx, struct vmx_vcpu *vcpu, pmap_t pmap, int running)
1279 {
1280 	struct vmxstate *vmxstate;
1281 	struct invvpid_desc invvpid_desc;
1282 
1283 	vmxstate = &vcpu->state;
1284 	if (vmxstate->vpid == 0)
1285 		return;
1286 
1287 	if (!running) {
1288 		/*
1289 		 * Set the 'lastcpu' to an invalid host cpu.
1290 		 *
1291 		 * This will invalidate TLB entries tagged with the vcpu's
1292 		 * vpid the next time it runs via vmx_set_pcpu_defaults().
1293 		 */
1294 		vmxstate->lastcpu = NOCPU;
1295 		return;
1296 	}
1297 
1298 	KASSERT(curthread->td_critnest > 0, ("%s: vcpu %d running outside "
1299 	    "critical section", __func__, vcpu->vcpuid));
1300 
1301 	/*
1302 	 * Invalidate all mappings tagged with 'vpid'
1303 	 *
1304 	 * We do this because this vcpu was executing on a different host
1305 	 * cpu when it last ran. We do not track whether it invalidated
1306 	 * mappings associated with its 'vpid' during that run. So we must
1307 	 * assume that the mappings associated with 'vpid' on 'curcpu' are
1308 	 * stale and invalidate them.
1309 	 *
1310 	 * Note that we incur this penalty only when the scheduler chooses to
1311 	 * move the thread associated with this vcpu between host cpus.
1312 	 *
1313 	 * Note also that this will invalidate mappings tagged with 'vpid'
1314 	 * for "all" EP4TAs.
1315 	 */
1316 	if (atomic_load_long(&pmap->pm_eptgen) == vmx->eptgen[curcpu]) {
1317 		invvpid_desc._res1 = 0;
1318 		invvpid_desc._res2 = 0;
1319 		invvpid_desc.vpid = vmxstate->vpid;
1320 		invvpid_desc.linear_addr = 0;
1321 		invvpid(INVVPID_TYPE_SINGLE_CONTEXT, invvpid_desc);
1322 		vmm_stat_incr(vcpu->vcpu, VCPU_INVVPID_DONE, 1);
1323 	} else {
1324 		/*
1325 		 * The invvpid can be skipped if an invept is going to
1326 		 * be performed before entering the guest. The invept
1327 		 * will invalidate combined mappings tagged with
1328 		 * 'vmx->eptp' for all vpids.
1329 		 */
1330 		vmm_stat_incr(vcpu->vcpu, VCPU_INVVPID_SAVED, 1);
1331 	}
1332 }
1333 
1334 static void
1335 vmx_set_pcpu_defaults(struct vmx *vmx, struct vmx_vcpu *vcpu, pmap_t pmap)
1336 {
1337 	struct vmxstate *vmxstate;
1338 
1339 	vmxstate = &vcpu->state;
1340 	if (vmxstate->lastcpu == curcpu)
1341 		return;
1342 
1343 	vmxstate->lastcpu = curcpu;
1344 
1345 	vmm_stat_incr(vcpu->vcpu, VCPU_MIGRATIONS, 1);
1346 
1347 	vmcs_write(VMCS_HOST_TR_BASE, vmm_get_host_trbase());
1348 	vmcs_write(VMCS_HOST_GDTR_BASE, vmm_get_host_gdtrbase());
1349 	vmcs_write(VMCS_HOST_GS_BASE, vmm_get_host_gsbase());
1350 	vmx_invvpid(vmx, vcpu, pmap, 1);
1351 }
1352 
1353 /*
1354  * We depend on 'procbased_ctls' to have the Interrupt Window Exiting bit set.
1355  */
1356 CTASSERT((PROCBASED_CTLS_ONE_SETTING & PROCBASED_INT_WINDOW_EXITING) != 0);
1357 
1358 static void __inline
1359 vmx_set_int_window_exiting(struct vmx_vcpu *vcpu)
1360 {
1361 
1362 	if ((vcpu->cap.proc_ctls & PROCBASED_INT_WINDOW_EXITING) == 0) {
1363 		vcpu->cap.proc_ctls |= PROCBASED_INT_WINDOW_EXITING;
1364 		vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vcpu->cap.proc_ctls);
1365 		VMX_CTR0(vcpu, "Enabling interrupt window exiting");
1366 	}
1367 }
1368 
1369 static void __inline
1370 vmx_clear_int_window_exiting(struct vmx_vcpu *vcpu)
1371 {
1372 
1373 	KASSERT((vcpu->cap.proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0,
1374 	    ("intr_window_exiting not set: %#x", vcpu->cap.proc_ctls));
1375 	vcpu->cap.proc_ctls &= ~PROCBASED_INT_WINDOW_EXITING;
1376 	vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vcpu->cap.proc_ctls);
1377 	VMX_CTR0(vcpu, "Disabling interrupt window exiting");
1378 }
1379 
1380 static void __inline
1381 vmx_set_nmi_window_exiting(struct vmx_vcpu *vcpu)
1382 {
1383 
1384 	if ((vcpu->cap.proc_ctls & PROCBASED_NMI_WINDOW_EXITING) == 0) {
1385 		vcpu->cap.proc_ctls |= PROCBASED_NMI_WINDOW_EXITING;
1386 		vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vcpu->cap.proc_ctls);
1387 		VMX_CTR0(vcpu, "Enabling NMI window exiting");
1388 	}
1389 }
1390 
1391 static void __inline
1392 vmx_clear_nmi_window_exiting(struct vmx_vcpu *vcpu)
1393 {
1394 
1395 	KASSERT((vcpu->cap.proc_ctls & PROCBASED_NMI_WINDOW_EXITING) != 0,
1396 	    ("nmi_window_exiting not set %#x", vcpu->cap.proc_ctls));
1397 	vcpu->cap.proc_ctls &= ~PROCBASED_NMI_WINDOW_EXITING;
1398 	vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vcpu->cap.proc_ctls);
1399 	VMX_CTR0(vcpu, "Disabling NMI window exiting");
1400 }
1401 
1402 int
1403 vmx_set_tsc_offset(struct vmx_vcpu *vcpu, uint64_t offset)
1404 {
1405 	int error;
1406 
1407 	if ((vcpu->cap.proc_ctls & PROCBASED_TSC_OFFSET) == 0) {
1408 		vcpu->cap.proc_ctls |= PROCBASED_TSC_OFFSET;
1409 		vmcs_write(VMCS_PRI_PROC_BASED_CTLS, vcpu->cap.proc_ctls);
1410 		VMX_CTR0(vcpu, "Enabling TSC offsetting");
1411 	}
1412 
1413 	error = vmwrite(VMCS_TSC_OFFSET, offset);
1414 #ifdef BHYVE_SNAPSHOT
1415 	if (error == 0)
1416 		vm_set_tsc_offset(vcpu->vcpu, offset);
1417 #endif
1418 	return (error);
1419 }
1420 
1421 #define	NMI_BLOCKING	(VMCS_INTERRUPTIBILITY_NMI_BLOCKING |		\
1422 			 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING)
1423 #define	HWINTR_BLOCKING	(VMCS_INTERRUPTIBILITY_STI_BLOCKING |		\
1424 			 VMCS_INTERRUPTIBILITY_MOVSS_BLOCKING)
1425 
1426 static void
1427 vmx_inject_nmi(struct vmx_vcpu *vcpu)
1428 {
1429 	uint32_t gi __diagused, info;
1430 
1431 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1432 	KASSERT((gi & NMI_BLOCKING) == 0, ("vmx_inject_nmi: invalid guest "
1433 	    "interruptibility-state %#x", gi));
1434 
1435 	info = vmcs_read(VMCS_ENTRY_INTR_INFO);
1436 	KASSERT((info & VMCS_INTR_VALID) == 0, ("vmx_inject_nmi: invalid "
1437 	    "VM-entry interruption information %#x", info));
1438 
1439 	/*
1440 	 * Inject the virtual NMI. The vector must be the NMI IDT entry
1441 	 * or the VMCS entry check will fail.
1442 	 */
1443 	info = IDT_NMI | VMCS_INTR_T_NMI | VMCS_INTR_VALID;
1444 	vmcs_write(VMCS_ENTRY_INTR_INFO, info);
1445 
1446 	VMX_CTR0(vcpu, "Injecting vNMI");
1447 
1448 	/* Clear the request */
1449 	vm_nmi_clear(vcpu->vcpu);
1450 }
1451 
1452 static void
1453 vmx_inject_interrupts(struct vmx_vcpu *vcpu, struct vlapic *vlapic,
1454     uint64_t guestrip)
1455 {
1456 	int vector, need_nmi_exiting, extint_pending;
1457 	uint64_t rflags, entryinfo;
1458 	uint32_t gi, info;
1459 
1460 	if (vcpu->cap.set & (1 << VM_CAP_MASK_HWINTR)) {
1461 		return;
1462 	}
1463 
1464 	if (vcpu->state.nextrip != guestrip) {
1465 		gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1466 		if (gi & HWINTR_BLOCKING) {
1467 			VMX_CTR2(vcpu, "Guest interrupt blocking "
1468 			    "cleared due to rip change: %#lx/%#lx",
1469 			    vcpu->state.nextrip, guestrip);
1470 			gi &= ~HWINTR_BLOCKING;
1471 			vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi);
1472 		}
1473 	}
1474 
1475 	if (vm_entry_intinfo(vcpu->vcpu, &entryinfo)) {
1476 		KASSERT((entryinfo & VMCS_INTR_VALID) != 0, ("%s: entry "
1477 		    "intinfo is not valid: %#lx", __func__, entryinfo));
1478 
1479 		info = vmcs_read(VMCS_ENTRY_INTR_INFO);
1480 		KASSERT((info & VMCS_INTR_VALID) == 0, ("%s: cannot inject "
1481 		     "pending exception: %#lx/%#x", __func__, entryinfo, info));
1482 
1483 		info = entryinfo;
1484 		vector = info & 0xff;
1485 		if (vector == IDT_BP || vector == IDT_OF) {
1486 			/*
1487 			 * VT-x requires #BP and #OF to be injected as software
1488 			 * exceptions.
1489 			 */
1490 			info &= ~VMCS_INTR_T_MASK;
1491 			info |= VMCS_INTR_T_SWEXCEPTION;
1492 		}
1493 
1494 		if (info & VMCS_INTR_DEL_ERRCODE)
1495 			vmcs_write(VMCS_ENTRY_EXCEPTION_ERROR, entryinfo >> 32);
1496 
1497 		vmcs_write(VMCS_ENTRY_INTR_INFO, info);
1498 	}
1499 
1500 	if (vm_nmi_pending(vcpu->vcpu)) {
1501 		/*
1502 		 * If there are no conditions blocking NMI injection then
1503 		 * inject it directly here otherwise enable "NMI window
1504 		 * exiting" to inject it as soon as we can.
1505 		 *
1506 		 * We also check for STI_BLOCKING because some implementations
1507 		 * don't allow NMI injection in this case. If we are running
1508 		 * on a processor that doesn't have this restriction it will
1509 		 * immediately exit and the NMI will be injected in the
1510 		 * "NMI window exiting" handler.
1511 		 */
1512 		need_nmi_exiting = 1;
1513 		gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1514 		if ((gi & (HWINTR_BLOCKING | NMI_BLOCKING)) == 0) {
1515 			info = vmcs_read(VMCS_ENTRY_INTR_INFO);
1516 			if ((info & VMCS_INTR_VALID) == 0) {
1517 				vmx_inject_nmi(vcpu);
1518 				need_nmi_exiting = 0;
1519 			} else {
1520 				VMX_CTR1(vcpu, "Cannot inject NMI "
1521 				    "due to VM-entry intr info %#x", info);
1522 			}
1523 		} else {
1524 			VMX_CTR1(vcpu, "Cannot inject NMI due to "
1525 			    "Guest Interruptibility-state %#x", gi);
1526 		}
1527 
1528 		if (need_nmi_exiting)
1529 			vmx_set_nmi_window_exiting(vcpu);
1530 	}
1531 
1532 	extint_pending = vm_extint_pending(vcpu->vcpu);
1533 
1534 	if (!extint_pending && virtual_interrupt_delivery) {
1535 		vmx_inject_pir(vlapic);
1536 		return;
1537 	}
1538 
1539 	/*
1540 	 * If interrupt-window exiting is already in effect then don't bother
1541 	 * checking for pending interrupts. This is just an optimization and
1542 	 * not needed for correctness.
1543 	 */
1544 	if ((vcpu->cap.proc_ctls & PROCBASED_INT_WINDOW_EXITING) != 0) {
1545 		VMX_CTR0(vcpu, "Skip interrupt injection due to "
1546 		    "pending int_window_exiting");
1547 		return;
1548 	}
1549 
1550 	if (!extint_pending) {
1551 		/* Ask the local apic for a vector to inject */
1552 		if (!vlapic_pending_intr(vlapic, &vector))
1553 			return;
1554 
1555 		/*
1556 		 * From the Intel SDM, Volume 3, Section "Maskable
1557 		 * Hardware Interrupts":
1558 		 * - maskable interrupt vectors [16,255] can be delivered
1559 		 *   through the local APIC.
1560 		*/
1561 		KASSERT(vector >= 16 && vector <= 255,
1562 		    ("invalid vector %d from local APIC", vector));
1563 	} else {
1564 		/* Ask the legacy pic for a vector to inject */
1565 		vatpic_pending_intr(vcpu->vmx->vm, &vector);
1566 
1567 		/*
1568 		 * From the Intel SDM, Volume 3, Section "Maskable
1569 		 * Hardware Interrupts":
1570 		 * - maskable interrupt vectors [0,255] can be delivered
1571 		 *   through the INTR pin.
1572 		 */
1573 		KASSERT(vector >= 0 && vector <= 255,
1574 		    ("invalid vector %d from INTR", vector));
1575 	}
1576 
1577 	/* Check RFLAGS.IF and the interruptibility state of the guest */
1578 	rflags = vmcs_read(VMCS_GUEST_RFLAGS);
1579 	if ((rflags & PSL_I) == 0) {
1580 		VMX_CTR2(vcpu, "Cannot inject vector %d due to "
1581 		    "rflags %#lx", vector, rflags);
1582 		goto cantinject;
1583 	}
1584 
1585 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1586 	if (gi & HWINTR_BLOCKING) {
1587 		VMX_CTR2(vcpu, "Cannot inject vector %d due to "
1588 		    "Guest Interruptibility-state %#x", vector, gi);
1589 		goto cantinject;
1590 	}
1591 
1592 	info = vmcs_read(VMCS_ENTRY_INTR_INFO);
1593 	if (info & VMCS_INTR_VALID) {
1594 		/*
1595 		 * This is expected and could happen for multiple reasons:
1596 		 * - A vectoring VM-entry was aborted due to astpending
1597 		 * - A VM-exit happened during event injection.
1598 		 * - An exception was injected above.
1599 		 * - An NMI was injected above or after "NMI window exiting"
1600 		 */
1601 		VMX_CTR2(vcpu, "Cannot inject vector %d due to "
1602 		    "VM-entry intr info %#x", vector, info);
1603 		goto cantinject;
1604 	}
1605 
1606 	/* Inject the interrupt */
1607 	info = VMCS_INTR_T_HWINTR | VMCS_INTR_VALID;
1608 	info |= vector;
1609 	vmcs_write(VMCS_ENTRY_INTR_INFO, info);
1610 
1611 	if (!extint_pending) {
1612 		/* Update the Local APIC ISR */
1613 		vlapic_intr_accepted(vlapic, vector);
1614 	} else {
1615 		vm_extint_clear(vcpu->vcpu);
1616 		vatpic_intr_accepted(vcpu->vmx->vm, vector);
1617 
1618 		/*
1619 		 * After we accepted the current ExtINT the PIC may
1620 		 * have posted another one.  If that is the case, set
1621 		 * the Interrupt Window Exiting execution control so
1622 		 * we can inject that one too.
1623 		 *
1624 		 * Also, interrupt window exiting allows us to inject any
1625 		 * pending APIC vector that was preempted by the ExtINT
1626 		 * as soon as possible. This applies both for the software
1627 		 * emulated vlapic and the hardware assisted virtual APIC.
1628 		 */
1629 		vmx_set_int_window_exiting(vcpu);
1630 	}
1631 
1632 	VMX_CTR1(vcpu, "Injecting hwintr at vector %d", vector);
1633 
1634 	return;
1635 
1636 cantinject:
1637 	/*
1638 	 * Set the Interrupt Window Exiting execution control so we can inject
1639 	 * the interrupt as soon as blocking condition goes away.
1640 	 */
1641 	vmx_set_int_window_exiting(vcpu);
1642 }
1643 
1644 /*
1645  * If the Virtual NMIs execution control is '1' then the logical processor
1646  * tracks virtual-NMI blocking in the Guest Interruptibility-state field of
1647  * the VMCS. An IRET instruction in VMX non-root operation will remove any
1648  * virtual-NMI blocking.
1649  *
1650  * This unblocking occurs even if the IRET causes a fault. In this case the
1651  * hypervisor needs to restore virtual-NMI blocking before resuming the guest.
1652  */
1653 static void
1654 vmx_restore_nmi_blocking(struct vmx_vcpu *vcpu)
1655 {
1656 	uint32_t gi;
1657 
1658 	VMX_CTR0(vcpu, "Restore Virtual-NMI blocking");
1659 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1660 	gi |= VMCS_INTERRUPTIBILITY_NMI_BLOCKING;
1661 	vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi);
1662 }
1663 
1664 static void
1665 vmx_clear_nmi_blocking(struct vmx_vcpu *vcpu)
1666 {
1667 	uint32_t gi;
1668 
1669 	VMX_CTR0(vcpu, "Clear Virtual-NMI blocking");
1670 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1671 	gi &= ~VMCS_INTERRUPTIBILITY_NMI_BLOCKING;
1672 	vmcs_write(VMCS_GUEST_INTERRUPTIBILITY, gi);
1673 }
1674 
1675 static void
1676 vmx_assert_nmi_blocking(struct vmx_vcpu *vcpu)
1677 {
1678 	uint32_t gi __diagused;
1679 
1680 	gi = vmcs_read(VMCS_GUEST_INTERRUPTIBILITY);
1681 	KASSERT(gi & VMCS_INTERRUPTIBILITY_NMI_BLOCKING,
1682 	    ("NMI blocking is not in effect %#x", gi));
1683 }
1684 
1685 static int
1686 vmx_emulate_xsetbv(struct vmx *vmx, struct vmx_vcpu *vcpu,
1687     struct vm_exit *vmexit)
1688 {
1689 	struct vmxctx *vmxctx;
1690 	uint64_t xcrval;
1691 	const struct xsave_limits *limits;
1692 
1693 	vmxctx = &vcpu->ctx;
1694 	limits = vmm_get_xsave_limits();
1695 
1696 	/*
1697 	 * Note that the processor raises a GP# fault on its own if
1698 	 * xsetbv is executed for CPL != 0, so we do not have to
1699 	 * emulate that fault here.
1700 	 */
1701 
1702 	/* Only xcr0 is supported. */
1703 	if (vmxctx->guest_rcx != 0) {
1704 		vm_inject_gp(vcpu->vcpu);
1705 		return (HANDLED);
1706 	}
1707 
1708 	/* We only handle xcr0 if both the host and guest have XSAVE enabled. */
1709 	if (!limits->xsave_enabled || !(vmcs_read(VMCS_GUEST_CR4) & CR4_XSAVE)) {
1710 		vm_inject_ud(vcpu->vcpu);
1711 		return (HANDLED);
1712 	}
1713 
1714 	xcrval = vmxctx->guest_rdx << 32 | (vmxctx->guest_rax & 0xffffffff);
1715 	if ((xcrval & ~limits->xcr0_allowed) != 0) {
1716 		vm_inject_gp(vcpu->vcpu);
1717 		return (HANDLED);
1718 	}
1719 
1720 	if (!(xcrval & XFEATURE_ENABLED_X87)) {
1721 		vm_inject_gp(vcpu->vcpu);
1722 		return (HANDLED);
1723 	}
1724 
1725 	/* AVX (YMM_Hi128) requires SSE. */
1726 	if (xcrval & XFEATURE_ENABLED_AVX &&
1727 	    (xcrval & XFEATURE_AVX) != XFEATURE_AVX) {
1728 		vm_inject_gp(vcpu->vcpu);
1729 		return (HANDLED);
1730 	}
1731 
1732 	/*
1733 	 * AVX512 requires base AVX (YMM_Hi128) as well as OpMask,
1734 	 * ZMM_Hi256, and Hi16_ZMM.
1735 	 */
1736 	if (xcrval & XFEATURE_AVX512 &&
1737 	    (xcrval & (XFEATURE_AVX512 | XFEATURE_AVX)) !=
1738 	    (XFEATURE_AVX512 | XFEATURE_AVX)) {
1739 		vm_inject_gp(vcpu->vcpu);
1740 		return (HANDLED);
1741 	}
1742 
1743 	/*
1744 	 * Intel MPX requires both bound register state flags to be
1745 	 * set.
1746 	 */
1747 	if (((xcrval & XFEATURE_ENABLED_BNDREGS) != 0) !=
1748 	    ((xcrval & XFEATURE_ENABLED_BNDCSR) != 0)) {
1749 		vm_inject_gp(vcpu->vcpu);
1750 		return (HANDLED);
1751 	}
1752 
1753 	/*
1754 	 * This runs "inside" vmrun() with the guest's FPU state, so
1755 	 * modifying xcr0 directly modifies the guest's xcr0, not the
1756 	 * host's.
1757 	 */
1758 	load_xcr(0, xcrval);
1759 	return (HANDLED);
1760 }
1761 
1762 static uint64_t
1763 vmx_get_guest_reg(struct vmx_vcpu *vcpu, int ident)
1764 {
1765 	const struct vmxctx *vmxctx;
1766 
1767 	vmxctx = &vcpu->ctx;
1768 
1769 	switch (ident) {
1770 	case 0:
1771 		return (vmxctx->guest_rax);
1772 	case 1:
1773 		return (vmxctx->guest_rcx);
1774 	case 2:
1775 		return (vmxctx->guest_rdx);
1776 	case 3:
1777 		return (vmxctx->guest_rbx);
1778 	case 4:
1779 		return (vmcs_read(VMCS_GUEST_RSP));
1780 	case 5:
1781 		return (vmxctx->guest_rbp);
1782 	case 6:
1783 		return (vmxctx->guest_rsi);
1784 	case 7:
1785 		return (vmxctx->guest_rdi);
1786 	case 8:
1787 		return (vmxctx->guest_r8);
1788 	case 9:
1789 		return (vmxctx->guest_r9);
1790 	case 10:
1791 		return (vmxctx->guest_r10);
1792 	case 11:
1793 		return (vmxctx->guest_r11);
1794 	case 12:
1795 		return (vmxctx->guest_r12);
1796 	case 13:
1797 		return (vmxctx->guest_r13);
1798 	case 14:
1799 		return (vmxctx->guest_r14);
1800 	case 15:
1801 		return (vmxctx->guest_r15);
1802 	default:
1803 		panic("invalid vmx register %d", ident);
1804 	}
1805 }
1806 
1807 static void
1808 vmx_set_guest_reg(struct vmx_vcpu *vcpu, int ident, uint64_t regval)
1809 {
1810 	struct vmxctx *vmxctx;
1811 
1812 	vmxctx = &vcpu->ctx;
1813 
1814 	switch (ident) {
1815 	case 0:
1816 		vmxctx->guest_rax = regval;
1817 		break;
1818 	case 1:
1819 		vmxctx->guest_rcx = regval;
1820 		break;
1821 	case 2:
1822 		vmxctx->guest_rdx = regval;
1823 		break;
1824 	case 3:
1825 		vmxctx->guest_rbx = regval;
1826 		break;
1827 	case 4:
1828 		vmcs_write(VMCS_GUEST_RSP, regval);
1829 		break;
1830 	case 5:
1831 		vmxctx->guest_rbp = regval;
1832 		break;
1833 	case 6:
1834 		vmxctx->guest_rsi = regval;
1835 		break;
1836 	case 7:
1837 		vmxctx->guest_rdi = regval;
1838 		break;
1839 	case 8:
1840 		vmxctx->guest_r8 = regval;
1841 		break;
1842 	case 9:
1843 		vmxctx->guest_r9 = regval;
1844 		break;
1845 	case 10:
1846 		vmxctx->guest_r10 = regval;
1847 		break;
1848 	case 11:
1849 		vmxctx->guest_r11 = regval;
1850 		break;
1851 	case 12:
1852 		vmxctx->guest_r12 = regval;
1853 		break;
1854 	case 13:
1855 		vmxctx->guest_r13 = regval;
1856 		break;
1857 	case 14:
1858 		vmxctx->guest_r14 = regval;
1859 		break;
1860 	case 15:
1861 		vmxctx->guest_r15 = regval;
1862 		break;
1863 	default:
1864 		panic("invalid vmx register %d", ident);
1865 	}
1866 }
1867 
1868 static int
1869 vmx_emulate_cr0_access(struct vmx_vcpu *vcpu, uint64_t exitqual)
1870 {
1871 	uint64_t crval, regval;
1872 
1873 	/* We only handle mov to %cr0 at this time */
1874 	if ((exitqual & 0xf0) != 0x00)
1875 		return (UNHANDLED);
1876 
1877 	regval = vmx_get_guest_reg(vcpu, (exitqual >> 8) & 0xf);
1878 
1879 	vmcs_write(VMCS_CR0_SHADOW, regval);
1880 
1881 	crval = regval | cr0_ones_mask;
1882 	crval &= ~cr0_zeros_mask;
1883 	vmcs_write(VMCS_GUEST_CR0, crval);
1884 
1885 	if (regval & CR0_PG) {
1886 		uint64_t efer, entry_ctls;
1887 
1888 		/*
1889 		 * If CR0.PG is 1 and EFER.LME is 1 then EFER.LMA and
1890 		 * the "IA-32e mode guest" bit in VM-entry control must be
1891 		 * equal.
1892 		 */
1893 		efer = vmcs_read(VMCS_GUEST_IA32_EFER);
1894 		if (efer & EFER_LME) {
1895 			efer |= EFER_LMA;
1896 			vmcs_write(VMCS_GUEST_IA32_EFER, efer);
1897 			entry_ctls = vmcs_read(VMCS_ENTRY_CTLS);
1898 			entry_ctls |= VM_ENTRY_GUEST_LMA;
1899 			vmcs_write(VMCS_ENTRY_CTLS, entry_ctls);
1900 		}
1901 	}
1902 
1903 	return (HANDLED);
1904 }
1905 
1906 static int
1907 vmx_emulate_cr4_access(struct vmx_vcpu *vcpu, uint64_t exitqual)
1908 {
1909 	uint64_t crval, regval;
1910 
1911 	/* We only handle mov to %cr4 at this time */
1912 	if ((exitqual & 0xf0) != 0x00)
1913 		return (UNHANDLED);
1914 
1915 	regval = vmx_get_guest_reg(vcpu, (exitqual >> 8) & 0xf);
1916 
1917 	vmcs_write(VMCS_CR4_SHADOW, regval);
1918 
1919 	crval = regval | cr4_ones_mask;
1920 	crval &= ~cr4_zeros_mask;
1921 	vmcs_write(VMCS_GUEST_CR4, crval);
1922 
1923 	return (HANDLED);
1924 }
1925 
1926 static int
1927 vmx_emulate_cr8_access(struct vmx *vmx, struct vmx_vcpu *vcpu,
1928     uint64_t exitqual)
1929 {
1930 	struct vlapic *vlapic;
1931 	uint64_t cr8;
1932 	int regnum;
1933 
1934 	/* We only handle mov %cr8 to/from a register at this time. */
1935 	if ((exitqual & 0xe0) != 0x00) {
1936 		return (UNHANDLED);
1937 	}
1938 
1939 	vlapic = vm_lapic(vcpu->vcpu);
1940 	regnum = (exitqual >> 8) & 0xf;
1941 	if (exitqual & 0x10) {
1942 		cr8 = vlapic_get_cr8(vlapic);
1943 		vmx_set_guest_reg(vcpu, regnum, cr8);
1944 	} else {
1945 		cr8 = vmx_get_guest_reg(vcpu, regnum);
1946 		vlapic_set_cr8(vlapic, cr8);
1947 	}
1948 
1949 	return (HANDLED);
1950 }
1951 
1952 /*
1953  * From section "Guest Register State" in the Intel SDM: CPL = SS.DPL
1954  */
1955 static int
1956 vmx_cpl(void)
1957 {
1958 	uint32_t ssar;
1959 
1960 	ssar = vmcs_read(VMCS_GUEST_SS_ACCESS_RIGHTS);
1961 	return ((ssar >> 5) & 0x3);
1962 }
1963 
1964 static enum vm_cpu_mode
1965 vmx_cpu_mode(void)
1966 {
1967 	uint32_t csar;
1968 
1969 	if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LMA) {
1970 		csar = vmcs_read(VMCS_GUEST_CS_ACCESS_RIGHTS);
1971 		if (csar & 0x2000)
1972 			return (CPU_MODE_64BIT);	/* CS.L = 1 */
1973 		else
1974 			return (CPU_MODE_COMPATIBILITY);
1975 	} else if (vmcs_read(VMCS_GUEST_CR0) & CR0_PE) {
1976 		return (CPU_MODE_PROTECTED);
1977 	} else {
1978 		return (CPU_MODE_REAL);
1979 	}
1980 }
1981 
1982 static enum vm_paging_mode
1983 vmx_paging_mode(void)
1984 {
1985 	uint64_t cr4;
1986 
1987 	if (!(vmcs_read(VMCS_GUEST_CR0) & CR0_PG))
1988 		return (PAGING_MODE_FLAT);
1989 	cr4 = vmcs_read(VMCS_GUEST_CR4);
1990 	if (!(cr4 & CR4_PAE))
1991 		return (PAGING_MODE_32);
1992 	if (vmcs_read(VMCS_GUEST_IA32_EFER) & EFER_LME) {
1993 		if (!(cr4 & CR4_LA57))
1994 			return (PAGING_MODE_64);
1995 		return (PAGING_MODE_64_LA57);
1996 	} else
1997 		return (PAGING_MODE_PAE);
1998 }
1999 
2000 static uint64_t
2001 inout_str_index(struct vmx_vcpu *vcpu, int in)
2002 {
2003 	uint64_t val;
2004 	int error __diagused;
2005 	enum vm_reg_name reg;
2006 
2007 	reg = in ? VM_REG_GUEST_RDI : VM_REG_GUEST_RSI;
2008 	error = vmx_getreg(vcpu, reg, &val);
2009 	KASSERT(error == 0, ("%s: vmx_getreg error %d", __func__, error));
2010 	return (val);
2011 }
2012 
2013 static uint64_t
2014 inout_str_count(struct vmx_vcpu *vcpu, int rep)
2015 {
2016 	uint64_t val;
2017 	int error __diagused;
2018 
2019 	if (rep) {
2020 		error = vmx_getreg(vcpu, VM_REG_GUEST_RCX, &val);
2021 		KASSERT(!error, ("%s: vmx_getreg error %d", __func__, error));
2022 	} else {
2023 		val = 1;
2024 	}
2025 	return (val);
2026 }
2027 
2028 static int
2029 inout_str_addrsize(uint32_t inst_info)
2030 {
2031 	uint32_t size;
2032 
2033 	size = (inst_info >> 7) & 0x7;
2034 	switch (size) {
2035 	case 0:
2036 		return (2);	/* 16 bit */
2037 	case 1:
2038 		return (4);	/* 32 bit */
2039 	case 2:
2040 		return (8);	/* 64 bit */
2041 	default:
2042 		panic("%s: invalid size encoding %d", __func__, size);
2043 	}
2044 }
2045 
2046 static void
2047 inout_str_seginfo(struct vmx_vcpu *vcpu, uint32_t inst_info, int in,
2048     struct vm_inout_str *vis)
2049 {
2050 	int error __diagused, s;
2051 
2052 	if (in) {
2053 		vis->seg_name = VM_REG_GUEST_ES;
2054 	} else {
2055 		s = (inst_info >> 15) & 0x7;
2056 		vis->seg_name = vm_segment_name(s);
2057 	}
2058 
2059 	error = vmx_getdesc(vcpu, vis->seg_name, &vis->seg_desc);
2060 	KASSERT(error == 0, ("%s: vmx_getdesc error %d", __func__, error));
2061 }
2062 
2063 static void
2064 vmx_paging_info(struct vm_guest_paging *paging)
2065 {
2066 	paging->cr3 = vmcs_guest_cr3();
2067 	paging->cpl = vmx_cpl();
2068 	paging->cpu_mode = vmx_cpu_mode();
2069 	paging->paging_mode = vmx_paging_mode();
2070 }
2071 
2072 static void
2073 vmexit_inst_emul(struct vm_exit *vmexit, uint64_t gpa, uint64_t gla)
2074 {
2075 	struct vm_guest_paging *paging;
2076 	uint32_t csar;
2077 
2078 	paging = &vmexit->u.inst_emul.paging;
2079 
2080 	vmexit->exitcode = VM_EXITCODE_INST_EMUL;
2081 	vmexit->inst_length = 0;
2082 	vmexit->u.inst_emul.gpa = gpa;
2083 	vmexit->u.inst_emul.gla = gla;
2084 	vmx_paging_info(paging);
2085 	switch (paging->cpu_mode) {
2086 	case CPU_MODE_REAL:
2087 		vmexit->u.inst_emul.cs_base = vmcs_read(VMCS_GUEST_CS_BASE);
2088 		vmexit->u.inst_emul.cs_d = 0;
2089 		break;
2090 	case CPU_MODE_PROTECTED:
2091 	case CPU_MODE_COMPATIBILITY:
2092 		vmexit->u.inst_emul.cs_base = vmcs_read(VMCS_GUEST_CS_BASE);
2093 		csar = vmcs_read(VMCS_GUEST_CS_ACCESS_RIGHTS);
2094 		vmexit->u.inst_emul.cs_d = SEG_DESC_DEF32(csar);
2095 		break;
2096 	default:
2097 		vmexit->u.inst_emul.cs_base = 0;
2098 		vmexit->u.inst_emul.cs_d = 0;
2099 		break;
2100 	}
2101 	vie_init(&vmexit->u.inst_emul.vie, NULL, 0);
2102 }
2103 
2104 static int
2105 ept_fault_type(uint64_t ept_qual)
2106 {
2107 	int fault_type;
2108 
2109 	if (ept_qual & EPT_VIOLATION_DATA_WRITE)
2110 		fault_type = VM_PROT_WRITE;
2111 	else if (ept_qual & EPT_VIOLATION_INST_FETCH)
2112 		fault_type = VM_PROT_EXECUTE;
2113 	else
2114 		fault_type= VM_PROT_READ;
2115 
2116 	return (fault_type);
2117 }
2118 
2119 static bool
2120 ept_emulation_fault(uint64_t ept_qual)
2121 {
2122 	int read, write;
2123 
2124 	/* EPT fault on an instruction fetch doesn't make sense here */
2125 	if (ept_qual & EPT_VIOLATION_INST_FETCH)
2126 		return (false);
2127 
2128 	/* EPT fault must be a read fault or a write fault */
2129 	read = ept_qual & EPT_VIOLATION_DATA_READ ? 1 : 0;
2130 	write = ept_qual & EPT_VIOLATION_DATA_WRITE ? 1 : 0;
2131 	if ((read | write) == 0)
2132 		return (false);
2133 
2134 	/*
2135 	 * The EPT violation must have been caused by accessing a
2136 	 * guest-physical address that is a translation of a guest-linear
2137 	 * address.
2138 	 */
2139 	if ((ept_qual & EPT_VIOLATION_GLA_VALID) == 0 ||
2140 	    (ept_qual & EPT_VIOLATION_XLAT_VALID) == 0) {
2141 		return (false);
2142 	}
2143 
2144 	return (true);
2145 }
2146 
2147 static __inline int
2148 apic_access_virtualization(struct vmx_vcpu *vcpu)
2149 {
2150 	uint32_t proc_ctls2;
2151 
2152 	proc_ctls2 = vcpu->cap.proc_ctls2;
2153 	return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) ? 1 : 0);
2154 }
2155 
2156 static __inline int
2157 x2apic_virtualization(struct vmx_vcpu *vcpu)
2158 {
2159 	uint32_t proc_ctls2;
2160 
2161 	proc_ctls2 = vcpu->cap.proc_ctls2;
2162 	return ((proc_ctls2 & PROCBASED2_VIRTUALIZE_X2APIC_MODE) ? 1 : 0);
2163 }
2164 
2165 static int
2166 vmx_handle_apic_write(struct vmx_vcpu *vcpu, struct vlapic *vlapic,
2167     uint64_t qual)
2168 {
2169 	int error, handled, offset;
2170 	uint32_t *apic_regs, vector;
2171 	bool retu;
2172 
2173 	handled = HANDLED;
2174 	offset = APIC_WRITE_OFFSET(qual);
2175 
2176 	if (!apic_access_virtualization(vcpu)) {
2177 		/*
2178 		 * In general there should not be any APIC write VM-exits
2179 		 * unless APIC-access virtualization is enabled.
2180 		 *
2181 		 * However self-IPI virtualization can legitimately trigger
2182 		 * an APIC-write VM-exit so treat it specially.
2183 		 */
2184 		if (x2apic_virtualization(vcpu) &&
2185 		    offset == APIC_OFFSET_SELF_IPI) {
2186 			apic_regs = (uint32_t *)(vlapic->apic_page);
2187 			vector = apic_regs[APIC_OFFSET_SELF_IPI / 4];
2188 			vlapic_self_ipi_handler(vlapic, vector);
2189 			return (HANDLED);
2190 		} else
2191 			return (UNHANDLED);
2192 	}
2193 
2194 	switch (offset) {
2195 	case APIC_OFFSET_ID:
2196 		vlapic_id_write_handler(vlapic);
2197 		break;
2198 	case APIC_OFFSET_LDR:
2199 		vlapic_ldr_write_handler(vlapic);
2200 		break;
2201 	case APIC_OFFSET_DFR:
2202 		vlapic_dfr_write_handler(vlapic);
2203 		break;
2204 	case APIC_OFFSET_SVR:
2205 		vlapic_svr_write_handler(vlapic);
2206 		break;
2207 	case APIC_OFFSET_ESR:
2208 		vlapic_esr_write_handler(vlapic);
2209 		break;
2210 	case APIC_OFFSET_ICR_LOW:
2211 		retu = false;
2212 		error = vlapic_icrlo_write_handler(vlapic, &retu);
2213 		if (error != 0 || retu)
2214 			handled = UNHANDLED;
2215 		break;
2216 	case APIC_OFFSET_CMCI_LVT:
2217 	case APIC_OFFSET_TIMER_LVT ... APIC_OFFSET_ERROR_LVT:
2218 		vlapic_lvt_write_handler(vlapic, offset);
2219 		break;
2220 	case APIC_OFFSET_TIMER_ICR:
2221 		vlapic_icrtmr_write_handler(vlapic);
2222 		break;
2223 	case APIC_OFFSET_TIMER_DCR:
2224 		vlapic_dcr_write_handler(vlapic);
2225 		break;
2226 	default:
2227 		handled = UNHANDLED;
2228 		break;
2229 	}
2230 	return (handled);
2231 }
2232 
2233 static bool
2234 apic_access_fault(struct vmx_vcpu *vcpu, uint64_t gpa)
2235 {
2236 
2237 	if (apic_access_virtualization(vcpu) &&
2238 	    (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE))
2239 		return (true);
2240 	else
2241 		return (false);
2242 }
2243 
2244 static int
2245 vmx_handle_apic_access(struct vmx_vcpu *vcpu, struct vm_exit *vmexit)
2246 {
2247 	uint64_t qual;
2248 	int access_type, offset, allowed;
2249 
2250 	if (!apic_access_virtualization(vcpu))
2251 		return (UNHANDLED);
2252 
2253 	qual = vmexit->u.vmx.exit_qualification;
2254 	access_type = APIC_ACCESS_TYPE(qual);
2255 	offset = APIC_ACCESS_OFFSET(qual);
2256 
2257 	allowed = 0;
2258 	if (access_type == 0) {
2259 		/*
2260 		 * Read data access to the following registers is expected.
2261 		 */
2262 		switch (offset) {
2263 		case APIC_OFFSET_APR:
2264 		case APIC_OFFSET_PPR:
2265 		case APIC_OFFSET_RRR:
2266 		case APIC_OFFSET_CMCI_LVT:
2267 		case APIC_OFFSET_TIMER_CCR:
2268 			allowed = 1;
2269 			break;
2270 		default:
2271 			break;
2272 		}
2273 	} else if (access_type == 1) {
2274 		/*
2275 		 * Write data access to the following registers is expected.
2276 		 */
2277 		switch (offset) {
2278 		case APIC_OFFSET_VER:
2279 		case APIC_OFFSET_APR:
2280 		case APIC_OFFSET_PPR:
2281 		case APIC_OFFSET_RRR:
2282 		case APIC_OFFSET_ISR0 ... APIC_OFFSET_ISR7:
2283 		case APIC_OFFSET_TMR0 ... APIC_OFFSET_TMR7:
2284 		case APIC_OFFSET_IRR0 ... APIC_OFFSET_IRR7:
2285 		case APIC_OFFSET_CMCI_LVT:
2286 		case APIC_OFFSET_TIMER_CCR:
2287 			allowed = 1;
2288 			break;
2289 		default:
2290 			break;
2291 		}
2292 	}
2293 
2294 	if (allowed) {
2295 		vmexit_inst_emul(vmexit, DEFAULT_APIC_BASE + offset,
2296 		    VIE_INVALID_GLA);
2297 	}
2298 
2299 	/*
2300 	 * Regardless of whether the APIC-access is allowed this handler
2301 	 * always returns UNHANDLED:
2302 	 * - if the access is allowed then it is handled by emulating the
2303 	 *   instruction that caused the VM-exit (outside the critical section)
2304 	 * - if the access is not allowed then it will be converted to an
2305 	 *   exitcode of VM_EXITCODE_VMX and will be dealt with in userland.
2306 	 */
2307 	return (UNHANDLED);
2308 }
2309 
2310 static enum task_switch_reason
2311 vmx_task_switch_reason(uint64_t qual)
2312 {
2313 	int reason;
2314 
2315 	reason = (qual >> 30) & 0x3;
2316 	switch (reason) {
2317 	case 0:
2318 		return (TSR_CALL);
2319 	case 1:
2320 		return (TSR_IRET);
2321 	case 2:
2322 		return (TSR_JMP);
2323 	case 3:
2324 		return (TSR_IDT_GATE);
2325 	default:
2326 		panic("%s: invalid reason %d", __func__, reason);
2327 	}
2328 }
2329 
2330 static int
2331 emulate_wrmsr(struct vmx_vcpu *vcpu, u_int num, uint64_t val, bool *retu)
2332 {
2333 	int error;
2334 
2335 	if (lapic_msr(num))
2336 		error = lapic_wrmsr(vcpu->vcpu, num, val, retu);
2337 	else
2338 		error = vmx_wrmsr(vcpu, num, val, retu);
2339 
2340 	return (error);
2341 }
2342 
2343 static int
2344 emulate_rdmsr(struct vmx_vcpu *vcpu, u_int num, bool *retu)
2345 {
2346 	struct vmxctx *vmxctx;
2347 	uint64_t result;
2348 	uint32_t eax, edx;
2349 	int error;
2350 
2351 	if (lapic_msr(num))
2352 		error = lapic_rdmsr(vcpu->vcpu, num, &result, retu);
2353 	else
2354 		error = vmx_rdmsr(vcpu, num, &result, retu);
2355 
2356 	if (error == 0) {
2357 		eax = result;
2358 		vmxctx = &vcpu->ctx;
2359 		error = vmxctx_setreg(vmxctx, VM_REG_GUEST_RAX, eax);
2360 		KASSERT(error == 0, ("vmxctx_setreg(rax) error %d", error));
2361 
2362 		edx = result >> 32;
2363 		error = vmxctx_setreg(vmxctx, VM_REG_GUEST_RDX, edx);
2364 		KASSERT(error == 0, ("vmxctx_setreg(rdx) error %d", error));
2365 	}
2366 
2367 	return (error);
2368 }
2369 
2370 static int
2371 vmx_exit_process(struct vmx *vmx, struct vmx_vcpu *vcpu, struct vm_exit *vmexit)
2372 {
2373 	int error, errcode, errcode_valid, handled, in;
2374 	struct vmxctx *vmxctx;
2375 	struct vlapic *vlapic;
2376 	struct vm_inout_str *vis;
2377 	struct vm_task_switch *ts;
2378 	uint32_t eax, ecx, edx, idtvec_info, idtvec_err, intr_info, inst_info;
2379 	uint32_t intr_type, intr_vec, reason;
2380 	uint64_t exitintinfo, qual, gpa;
2381 #ifdef KDTRACE_HOOKS
2382 	int vcpuid;
2383 #endif
2384 	bool retu;
2385 
2386 	CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_VIRTUAL_NMI) != 0);
2387 	CTASSERT((PINBASED_CTLS_ONE_SETTING & PINBASED_NMI_EXITING) != 0);
2388 
2389 	handled = UNHANDLED;
2390 	vmxctx = &vcpu->ctx;
2391 #ifdef KDTRACE_HOOKS
2392 	vcpuid = vcpu->vcpuid;
2393 #endif
2394 
2395 	qual = vmexit->u.vmx.exit_qualification;
2396 	reason = vmexit->u.vmx.exit_reason;
2397 	vmexit->exitcode = VM_EXITCODE_BOGUS;
2398 
2399 	vmm_stat_incr(vcpu->vcpu, VMEXIT_COUNT, 1);
2400 	SDT_PROBE3(vmm, vmx, exit, entry, vmx, vcpuid, vmexit);
2401 
2402 	/*
2403 	 * VM-entry failures during or after loading guest state.
2404 	 *
2405 	 * These VM-exits are uncommon but must be handled specially
2406 	 * as most VM-exit fields are not populated as usual.
2407 	 */
2408 	if (__predict_false(reason == EXIT_REASON_MCE_DURING_ENTRY)) {
2409 		VMX_CTR0(vcpu, "Handling MCE during VM-entry");
2410 		__asm __volatile("int $18");
2411 		return (1);
2412 	}
2413 
2414 	/*
2415 	 * VM exits that can be triggered during event delivery need to
2416 	 * be handled specially by re-injecting the event if the IDT
2417 	 * vectoring information field's valid bit is set.
2418 	 *
2419 	 * See "Information for VM Exits During Event Delivery" in Intel SDM
2420 	 * for details.
2421 	 */
2422 	idtvec_info = vmcs_idt_vectoring_info();
2423 	if (idtvec_info & VMCS_IDT_VEC_VALID) {
2424 		idtvec_info &= ~(1 << 12); /* clear undefined bit */
2425 		exitintinfo = idtvec_info;
2426 		if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) {
2427 			idtvec_err = vmcs_idt_vectoring_err();
2428 			exitintinfo |= (uint64_t)idtvec_err << 32;
2429 		}
2430 		error = vm_exit_intinfo(vcpu->vcpu, exitintinfo);
2431 		KASSERT(error == 0, ("%s: vm_set_intinfo error %d",
2432 		    __func__, error));
2433 
2434 		/*
2435 		 * If 'virtual NMIs' are being used and the VM-exit
2436 		 * happened while injecting an NMI during the previous
2437 		 * VM-entry, then clear "blocking by NMI" in the
2438 		 * Guest Interruptibility-State so the NMI can be
2439 		 * reinjected on the subsequent VM-entry.
2440 		 *
2441 		 * However, if the NMI was being delivered through a task
2442 		 * gate, then the new task must start execution with NMIs
2443 		 * blocked so don't clear NMI blocking in this case.
2444 		 */
2445 		intr_type = idtvec_info & VMCS_INTR_T_MASK;
2446 		if (intr_type == VMCS_INTR_T_NMI) {
2447 			if (reason != EXIT_REASON_TASK_SWITCH)
2448 				vmx_clear_nmi_blocking(vcpu);
2449 			else
2450 				vmx_assert_nmi_blocking(vcpu);
2451 		}
2452 
2453 		/*
2454 		 * Update VM-entry instruction length if the event being
2455 		 * delivered was a software interrupt or software exception.
2456 		 */
2457 		if (intr_type == VMCS_INTR_T_SWINTR ||
2458 		    intr_type == VMCS_INTR_T_PRIV_SWEXCEPTION ||
2459 		    intr_type == VMCS_INTR_T_SWEXCEPTION) {
2460 			vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length);
2461 		}
2462 	}
2463 
2464 	switch (reason) {
2465 	case EXIT_REASON_TASK_SWITCH:
2466 		ts = &vmexit->u.task_switch;
2467 		ts->tsssel = qual & 0xffff;
2468 		ts->reason = vmx_task_switch_reason(qual);
2469 		ts->ext = 0;
2470 		ts->errcode_valid = 0;
2471 		vmx_paging_info(&ts->paging);
2472 		/*
2473 		 * If the task switch was due to a CALL, JMP, IRET, software
2474 		 * interrupt (INT n) or software exception (INT3, INTO),
2475 		 * then the saved %rip references the instruction that caused
2476 		 * the task switch. The instruction length field in the VMCS
2477 		 * is valid in this case.
2478 		 *
2479 		 * In all other cases (e.g., NMI, hardware exception) the
2480 		 * saved %rip is one that would have been saved in the old TSS
2481 		 * had the task switch completed normally so the instruction
2482 		 * length field is not needed in this case and is explicitly
2483 		 * set to 0.
2484 		 */
2485 		if (ts->reason == TSR_IDT_GATE) {
2486 			KASSERT(idtvec_info & VMCS_IDT_VEC_VALID,
2487 			    ("invalid idtvec_info %#x for IDT task switch",
2488 			    idtvec_info));
2489 			intr_type = idtvec_info & VMCS_INTR_T_MASK;
2490 			if (intr_type != VMCS_INTR_T_SWINTR &&
2491 			    intr_type != VMCS_INTR_T_SWEXCEPTION &&
2492 			    intr_type != VMCS_INTR_T_PRIV_SWEXCEPTION) {
2493 				/* Task switch triggered by external event */
2494 				ts->ext = 1;
2495 				vmexit->inst_length = 0;
2496 				if (idtvec_info & VMCS_IDT_VEC_ERRCODE_VALID) {
2497 					ts->errcode_valid = 1;
2498 					ts->errcode = vmcs_idt_vectoring_err();
2499 				}
2500 			}
2501 		}
2502 		vmexit->exitcode = VM_EXITCODE_TASK_SWITCH;
2503 		SDT_PROBE4(vmm, vmx, exit, taskswitch, vmx, vcpuid, vmexit, ts);
2504 		VMX_CTR4(vcpu, "task switch reason %d, tss 0x%04x, "
2505 		    "%s errcode 0x%016lx", ts->reason, ts->tsssel,
2506 		    ts->ext ? "external" : "internal",
2507 		    ((uint64_t)ts->errcode << 32) | ts->errcode_valid);
2508 		break;
2509 	case EXIT_REASON_CR_ACCESS:
2510 		vmm_stat_incr(vcpu->vcpu, VMEXIT_CR_ACCESS, 1);
2511 		SDT_PROBE4(vmm, vmx, exit, craccess, vmx, vcpuid, vmexit, qual);
2512 		switch (qual & 0xf) {
2513 		case 0:
2514 			handled = vmx_emulate_cr0_access(vcpu, qual);
2515 			break;
2516 		case 4:
2517 			handled = vmx_emulate_cr4_access(vcpu, qual);
2518 			break;
2519 		case 8:
2520 			handled = vmx_emulate_cr8_access(vmx, vcpu, qual);
2521 			break;
2522 		}
2523 		break;
2524 	case EXIT_REASON_RDMSR:
2525 		vmm_stat_incr(vcpu->vcpu, VMEXIT_RDMSR, 1);
2526 		retu = false;
2527 		ecx = vmxctx->guest_rcx;
2528 		VMX_CTR1(vcpu, "rdmsr 0x%08x", ecx);
2529 		SDT_PROBE4(vmm, vmx, exit, rdmsr, vmx, vcpuid, vmexit, ecx);
2530 		error = emulate_rdmsr(vcpu, ecx, &retu);
2531 		if (error) {
2532 			vmexit->exitcode = VM_EXITCODE_RDMSR;
2533 			vmexit->u.msr.code = ecx;
2534 		} else if (!retu) {
2535 			handled = HANDLED;
2536 		} else {
2537 			/* Return to userspace with a valid exitcode */
2538 			KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS,
2539 			    ("emulate_rdmsr retu with bogus exitcode"));
2540 		}
2541 		break;
2542 	case EXIT_REASON_WRMSR:
2543 		vmm_stat_incr(vcpu->vcpu, VMEXIT_WRMSR, 1);
2544 		retu = false;
2545 		eax = vmxctx->guest_rax;
2546 		ecx = vmxctx->guest_rcx;
2547 		edx = vmxctx->guest_rdx;
2548 		VMX_CTR2(vcpu, "wrmsr 0x%08x value 0x%016lx",
2549 		    ecx, (uint64_t)edx << 32 | eax);
2550 		SDT_PROBE5(vmm, vmx, exit, wrmsr, vmx, vmexit, vcpuid, ecx,
2551 		    (uint64_t)edx << 32 | eax);
2552 		error = emulate_wrmsr(vcpu, ecx, (uint64_t)edx << 32 | eax,
2553 		    &retu);
2554 		if (error) {
2555 			vmexit->exitcode = VM_EXITCODE_WRMSR;
2556 			vmexit->u.msr.code = ecx;
2557 			vmexit->u.msr.wval = (uint64_t)edx << 32 | eax;
2558 		} else if (!retu) {
2559 			handled = HANDLED;
2560 		} else {
2561 			/* Return to userspace with a valid exitcode */
2562 			KASSERT(vmexit->exitcode != VM_EXITCODE_BOGUS,
2563 			    ("emulate_wrmsr retu with bogus exitcode"));
2564 		}
2565 		break;
2566 	case EXIT_REASON_HLT:
2567 		vmm_stat_incr(vcpu->vcpu, VMEXIT_HLT, 1);
2568 		SDT_PROBE3(vmm, vmx, exit, halt, vmx, vcpuid, vmexit);
2569 		vmexit->exitcode = VM_EXITCODE_HLT;
2570 		vmexit->u.hlt.rflags = vmcs_read(VMCS_GUEST_RFLAGS);
2571 		if (virtual_interrupt_delivery)
2572 			vmexit->u.hlt.intr_status =
2573 			    vmcs_read(VMCS_GUEST_INTR_STATUS);
2574 		else
2575 			vmexit->u.hlt.intr_status = 0;
2576 		break;
2577 	case EXIT_REASON_MTF:
2578 		vmm_stat_incr(vcpu->vcpu, VMEXIT_MTRAP, 1);
2579 		SDT_PROBE3(vmm, vmx, exit, mtrap, vmx, vcpuid, vmexit);
2580 		vmexit->exitcode = VM_EXITCODE_MTRAP;
2581 		vmexit->inst_length = 0;
2582 		break;
2583 	case EXIT_REASON_PAUSE:
2584 		vmm_stat_incr(vcpu->vcpu, VMEXIT_PAUSE, 1);
2585 		SDT_PROBE3(vmm, vmx, exit, pause, vmx, vcpuid, vmexit);
2586 		vmexit->exitcode = VM_EXITCODE_PAUSE;
2587 		break;
2588 	case EXIT_REASON_INTR_WINDOW:
2589 		vmm_stat_incr(vcpu->vcpu, VMEXIT_INTR_WINDOW, 1);
2590 		SDT_PROBE3(vmm, vmx, exit, intrwindow, vmx, vcpuid, vmexit);
2591 		vmx_clear_int_window_exiting(vcpu);
2592 		return (1);
2593 	case EXIT_REASON_EXT_INTR:
2594 		/*
2595 		 * External interrupts serve only to cause VM exits and allow
2596 		 * the host interrupt handler to run.
2597 		 *
2598 		 * If this external interrupt triggers a virtual interrupt
2599 		 * to a VM, then that state will be recorded by the
2600 		 * host interrupt handler in the VM's softc. We will inject
2601 		 * this virtual interrupt during the subsequent VM enter.
2602 		 */
2603 		intr_info = vmcs_read(VMCS_EXIT_INTR_INFO);
2604 		SDT_PROBE4(vmm, vmx, exit, interrupt,
2605 		    vmx, vcpuid, vmexit, intr_info);
2606 
2607 		/*
2608 		 * XXX: Ignore this exit if VMCS_INTR_VALID is not set.
2609 		 * This appears to be a bug in VMware Fusion?
2610 		 */
2611 		if (!(intr_info & VMCS_INTR_VALID))
2612 			return (1);
2613 		KASSERT((intr_info & VMCS_INTR_VALID) != 0 &&
2614 		    (intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_HWINTR,
2615 		    ("VM exit interruption info invalid: %#x", intr_info));
2616 		vmx_trigger_hostintr(intr_info & 0xff);
2617 
2618 		/*
2619 		 * This is special. We want to treat this as an 'handled'
2620 		 * VM-exit but not increment the instruction pointer.
2621 		 */
2622 		vmm_stat_incr(vcpu->vcpu, VMEXIT_EXTINT, 1);
2623 		return (1);
2624 	case EXIT_REASON_NMI_WINDOW:
2625 		SDT_PROBE3(vmm, vmx, exit, nmiwindow, vmx, vcpuid, vmexit);
2626 		/* Exit to allow the pending virtual NMI to be injected */
2627 		if (vm_nmi_pending(vcpu->vcpu))
2628 			vmx_inject_nmi(vcpu);
2629 		vmx_clear_nmi_window_exiting(vcpu);
2630 		vmm_stat_incr(vcpu->vcpu, VMEXIT_NMI_WINDOW, 1);
2631 		return (1);
2632 	case EXIT_REASON_INOUT:
2633 		vmm_stat_incr(vcpu->vcpu, VMEXIT_INOUT, 1);
2634 		vmexit->exitcode = VM_EXITCODE_INOUT;
2635 		vmexit->u.inout.bytes = (qual & 0x7) + 1;
2636 		vmexit->u.inout.in = in = (qual & 0x8) ? 1 : 0;
2637 		vmexit->u.inout.string = (qual & 0x10) ? 1 : 0;
2638 		vmexit->u.inout.rep = (qual & 0x20) ? 1 : 0;
2639 		vmexit->u.inout.port = (uint16_t)(qual >> 16);
2640 		vmexit->u.inout.eax = (uint32_t)(vmxctx->guest_rax);
2641 		if (vmexit->u.inout.string) {
2642 			inst_info = vmcs_read(VMCS_EXIT_INSTRUCTION_INFO);
2643 			vmexit->exitcode = VM_EXITCODE_INOUT_STR;
2644 			vis = &vmexit->u.inout_str;
2645 			vmx_paging_info(&vis->paging);
2646 			vis->rflags = vmcs_read(VMCS_GUEST_RFLAGS);
2647 			vis->cr0 = vmcs_read(VMCS_GUEST_CR0);
2648 			vis->index = inout_str_index(vcpu, in);
2649 			vis->count = inout_str_count(vcpu, vis->inout.rep);
2650 			vis->addrsize = inout_str_addrsize(inst_info);
2651 			inout_str_seginfo(vcpu, inst_info, in, vis);
2652 		}
2653 		SDT_PROBE3(vmm, vmx, exit, inout, vmx, vcpuid, vmexit);
2654 		break;
2655 	case EXIT_REASON_CPUID:
2656 		vmm_stat_incr(vcpu->vcpu, VMEXIT_CPUID, 1);
2657 		SDT_PROBE3(vmm, vmx, exit, cpuid, vmx, vcpuid, vmexit);
2658 		handled = vmx_handle_cpuid(vcpu, vmxctx);
2659 		break;
2660 	case EXIT_REASON_EXCEPTION:
2661 		vmm_stat_incr(vcpu->vcpu, VMEXIT_EXCEPTION, 1);
2662 		intr_info = vmcs_read(VMCS_EXIT_INTR_INFO);
2663 		KASSERT((intr_info & VMCS_INTR_VALID) != 0,
2664 		    ("VM exit interruption info invalid: %#x", intr_info));
2665 
2666 		intr_vec = intr_info & 0xff;
2667 		intr_type = intr_info & VMCS_INTR_T_MASK;
2668 
2669 		/*
2670 		 * If Virtual NMIs control is 1 and the VM-exit is due to a
2671 		 * fault encountered during the execution of IRET then we must
2672 		 * restore the state of "virtual-NMI blocking" before resuming
2673 		 * the guest.
2674 		 *
2675 		 * See "Resuming Guest Software after Handling an Exception".
2676 		 * See "Information for VM Exits Due to Vectored Events".
2677 		 */
2678 		if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 &&
2679 		    (intr_vec != IDT_DF) &&
2680 		    (intr_info & EXIT_QUAL_NMIUDTI) != 0)
2681 			vmx_restore_nmi_blocking(vcpu);
2682 
2683 		/*
2684 		 * The NMI has already been handled in vmx_exit_handle_nmi().
2685 		 */
2686 		if (intr_type == VMCS_INTR_T_NMI)
2687 			return (1);
2688 
2689 		/*
2690 		 * Call the machine check handler by hand. Also don't reflect
2691 		 * the machine check back into the guest.
2692 		 */
2693 		if (intr_vec == IDT_MC) {
2694 			VMX_CTR0(vcpu, "Vectoring to MCE handler");
2695 			__asm __volatile("int $18");
2696 			return (1);
2697 		}
2698 
2699 		/*
2700 		 * If the hypervisor has requested user exits for
2701 		 * debug exceptions, bounce them out to userland.
2702 		 */
2703 		if (intr_type == VMCS_INTR_T_SWEXCEPTION && intr_vec == IDT_BP &&
2704 		    (vcpu->cap.set & (1 << VM_CAP_BPT_EXIT))) {
2705 			vmexit->exitcode = VM_EXITCODE_BPT;
2706 			vmexit->u.bpt.inst_length = vmexit->inst_length;
2707 			vmexit->inst_length = 0;
2708 			break;
2709 		}
2710 
2711 		if (intr_vec == IDT_PF) {
2712 			error = vmxctx_setreg(vmxctx, VM_REG_GUEST_CR2, qual);
2713 			KASSERT(error == 0, ("%s: vmxctx_setreg(cr2) error %d",
2714 			    __func__, error));
2715 		}
2716 
2717 		/*
2718 		 * Software exceptions exhibit trap-like behavior. This in
2719 		 * turn requires populating the VM-entry instruction length
2720 		 * so that the %rip in the trap frame is past the INT3/INTO
2721 		 * instruction.
2722 		 */
2723 		if (intr_type == VMCS_INTR_T_SWEXCEPTION)
2724 			vmcs_write(VMCS_ENTRY_INST_LENGTH, vmexit->inst_length);
2725 
2726 		/* Reflect all other exceptions back into the guest */
2727 		errcode_valid = errcode = 0;
2728 		if (intr_info & VMCS_INTR_DEL_ERRCODE) {
2729 			errcode_valid = 1;
2730 			errcode = vmcs_read(VMCS_EXIT_INTR_ERRCODE);
2731 		}
2732 		VMX_CTR2(vcpu, "Reflecting exception %d/%#x into "
2733 		    "the guest", intr_vec, errcode);
2734 		SDT_PROBE5(vmm, vmx, exit, exception,
2735 		    vmx, vcpuid, vmexit, intr_vec, errcode);
2736 		error = vm_inject_exception(vcpu->vcpu, intr_vec,
2737 		    errcode_valid, errcode, 0);
2738 		KASSERT(error == 0, ("%s: vm_inject_exception error %d",
2739 		    __func__, error));
2740 		return (1);
2741 
2742 	case EXIT_REASON_EPT_FAULT:
2743 		/*
2744 		 * If 'gpa' lies within the address space allocated to
2745 		 * memory then this must be a nested page fault otherwise
2746 		 * this must be an instruction that accesses MMIO space.
2747 		 */
2748 		gpa = vmcs_gpa();
2749 		if (vm_mem_allocated(vcpu->vcpu, gpa) ||
2750 		    apic_access_fault(vcpu, gpa)) {
2751 			vmexit->exitcode = VM_EXITCODE_PAGING;
2752 			vmexit->inst_length = 0;
2753 			vmexit->u.paging.gpa = gpa;
2754 			vmexit->u.paging.fault_type = ept_fault_type(qual);
2755 			vmm_stat_incr(vcpu->vcpu, VMEXIT_NESTED_FAULT, 1);
2756 			SDT_PROBE5(vmm, vmx, exit, nestedfault,
2757 			    vmx, vcpuid, vmexit, gpa, qual);
2758 		} else if (ept_emulation_fault(qual)) {
2759 			vmexit_inst_emul(vmexit, gpa, vmcs_gla());
2760 			vmm_stat_incr(vcpu->vcpu, VMEXIT_INST_EMUL, 1);
2761 			SDT_PROBE4(vmm, vmx, exit, mmiofault,
2762 			    vmx, vcpuid, vmexit, gpa);
2763 		}
2764 		/*
2765 		 * If Virtual NMIs control is 1 and the VM-exit is due to an
2766 		 * EPT fault during the execution of IRET then we must restore
2767 		 * the state of "virtual-NMI blocking" before resuming.
2768 		 *
2769 		 * See description of "NMI unblocking due to IRET" in
2770 		 * "Exit Qualification for EPT Violations".
2771 		 */
2772 		if ((idtvec_info & VMCS_IDT_VEC_VALID) == 0 &&
2773 		    (qual & EXIT_QUAL_NMIUDTI) != 0)
2774 			vmx_restore_nmi_blocking(vcpu);
2775 		break;
2776 	case EXIT_REASON_VIRTUALIZED_EOI:
2777 		vmexit->exitcode = VM_EXITCODE_IOAPIC_EOI;
2778 		vmexit->u.ioapic_eoi.vector = qual & 0xFF;
2779 		SDT_PROBE3(vmm, vmx, exit, eoi, vmx, vcpuid, vmexit);
2780 		vmexit->inst_length = 0;	/* trap-like */
2781 		break;
2782 	case EXIT_REASON_APIC_ACCESS:
2783 		SDT_PROBE3(vmm, vmx, exit, apicaccess, vmx, vcpuid, vmexit);
2784 		handled = vmx_handle_apic_access(vcpu, vmexit);
2785 		break;
2786 	case EXIT_REASON_APIC_WRITE:
2787 		/*
2788 		 * APIC-write VM exit is trap-like so the %rip is already
2789 		 * pointing to the next instruction.
2790 		 */
2791 		vmexit->inst_length = 0;
2792 		vlapic = vm_lapic(vcpu->vcpu);
2793 		SDT_PROBE4(vmm, vmx, exit, apicwrite,
2794 		    vmx, vcpuid, vmexit, vlapic);
2795 		handled = vmx_handle_apic_write(vcpu, vlapic, qual);
2796 		break;
2797 	case EXIT_REASON_XSETBV:
2798 		SDT_PROBE3(vmm, vmx, exit, xsetbv, vmx, vcpuid, vmexit);
2799 		handled = vmx_emulate_xsetbv(vmx, vcpu, vmexit);
2800 		break;
2801 	case EXIT_REASON_MONITOR:
2802 		SDT_PROBE3(vmm, vmx, exit, monitor, vmx, vcpuid, vmexit);
2803 		vmexit->exitcode = VM_EXITCODE_MONITOR;
2804 		break;
2805 	case EXIT_REASON_MWAIT:
2806 		SDT_PROBE3(vmm, vmx, exit, mwait, vmx, vcpuid, vmexit);
2807 		vmexit->exitcode = VM_EXITCODE_MWAIT;
2808 		break;
2809 	case EXIT_REASON_TPR:
2810 		vlapic = vm_lapic(vcpu->vcpu);
2811 		vlapic_sync_tpr(vlapic);
2812 		vmexit->inst_length = 0;
2813 		handled = HANDLED;
2814 		break;
2815 	case EXIT_REASON_VMCALL:
2816 	case EXIT_REASON_VMCLEAR:
2817 	case EXIT_REASON_VMLAUNCH:
2818 	case EXIT_REASON_VMPTRLD:
2819 	case EXIT_REASON_VMPTRST:
2820 	case EXIT_REASON_VMREAD:
2821 	case EXIT_REASON_VMRESUME:
2822 	case EXIT_REASON_VMWRITE:
2823 	case EXIT_REASON_VMXOFF:
2824 	case EXIT_REASON_VMXON:
2825 		SDT_PROBE3(vmm, vmx, exit, vminsn, vmx, vcpuid, vmexit);
2826 		vmexit->exitcode = VM_EXITCODE_VMINSN;
2827 		break;
2828 	case EXIT_REASON_INVD:
2829 	case EXIT_REASON_WBINVD:
2830 		/* ignore exit */
2831 		handled = HANDLED;
2832 		break;
2833 	default:
2834 		SDT_PROBE4(vmm, vmx, exit, unknown,
2835 		    vmx, vcpuid, vmexit, reason);
2836 		vmm_stat_incr(vcpu->vcpu, VMEXIT_UNKNOWN, 1);
2837 		break;
2838 	}
2839 
2840 	if (handled) {
2841 		/*
2842 		 * It is possible that control is returned to userland
2843 		 * even though we were able to handle the VM exit in the
2844 		 * kernel.
2845 		 *
2846 		 * In such a case we want to make sure that the userland
2847 		 * restarts guest execution at the instruction *after*
2848 		 * the one we just processed. Therefore we update the
2849 		 * guest rip in the VMCS and in 'vmexit'.
2850 		 */
2851 		vmexit->rip += vmexit->inst_length;
2852 		vmexit->inst_length = 0;
2853 		vmcs_write(VMCS_GUEST_RIP, vmexit->rip);
2854 	} else {
2855 		if (vmexit->exitcode == VM_EXITCODE_BOGUS) {
2856 			/*
2857 			 * If this VM exit was not claimed by anybody then
2858 			 * treat it as a generic VMX exit.
2859 			 */
2860 			vmexit->exitcode = VM_EXITCODE_VMX;
2861 			vmexit->u.vmx.status = VM_SUCCESS;
2862 			vmexit->u.vmx.inst_type = 0;
2863 			vmexit->u.vmx.inst_error = 0;
2864 		} else {
2865 			/*
2866 			 * The exitcode and collateral have been populated.
2867 			 * The VM exit will be processed further in userland.
2868 			 */
2869 		}
2870 	}
2871 
2872 	SDT_PROBE4(vmm, vmx, exit, return,
2873 	    vmx, vcpuid, vmexit, handled);
2874 	return (handled);
2875 }
2876 
2877 static __inline void
2878 vmx_exit_inst_error(struct vmxctx *vmxctx, int rc, struct vm_exit *vmexit)
2879 {
2880 
2881 	KASSERT(vmxctx->inst_fail_status != VM_SUCCESS,
2882 	    ("vmx_exit_inst_error: invalid inst_fail_status %d",
2883 	    vmxctx->inst_fail_status));
2884 
2885 	vmexit->inst_length = 0;
2886 	vmexit->exitcode = VM_EXITCODE_VMX;
2887 	vmexit->u.vmx.status = vmxctx->inst_fail_status;
2888 	vmexit->u.vmx.inst_error = vmcs_instruction_error();
2889 	vmexit->u.vmx.exit_reason = ~0;
2890 	vmexit->u.vmx.exit_qualification = ~0;
2891 
2892 	switch (rc) {
2893 	case VMX_VMRESUME_ERROR:
2894 	case VMX_VMLAUNCH_ERROR:
2895 		vmexit->u.vmx.inst_type = rc;
2896 		break;
2897 	default:
2898 		panic("vm_exit_inst_error: vmx_enter_guest returned %d", rc);
2899 	}
2900 }
2901 
2902 /*
2903  * If the NMI-exiting VM execution control is set to '1' then an NMI in
2904  * non-root operation causes a VM-exit. NMI blocking is in effect so it is
2905  * sufficient to simply vector to the NMI handler via a software interrupt.
2906  * However, this must be done before maskable interrupts are enabled
2907  * otherwise the "iret" issued by an interrupt handler will incorrectly
2908  * clear NMI blocking.
2909  */
2910 static __inline void
2911 vmx_exit_handle_nmi(struct vmx_vcpu *vcpu, struct vm_exit *vmexit)
2912 {
2913 	uint32_t intr_info;
2914 
2915 	KASSERT((read_rflags() & PSL_I) == 0, ("interrupts enabled"));
2916 
2917 	if (vmexit->u.vmx.exit_reason != EXIT_REASON_EXCEPTION)
2918 		return;
2919 
2920 	intr_info = vmcs_read(VMCS_EXIT_INTR_INFO);
2921 	KASSERT((intr_info & VMCS_INTR_VALID) != 0,
2922 	    ("VM exit interruption info invalid: %#x", intr_info));
2923 
2924 	if ((intr_info & VMCS_INTR_T_MASK) == VMCS_INTR_T_NMI) {
2925 		KASSERT((intr_info & 0xff) == IDT_NMI, ("VM exit due "
2926 		    "to NMI has invalid vector: %#x", intr_info));
2927 		VMX_CTR0(vcpu, "Vectoring to NMI handler");
2928 		__asm __volatile("int $2");
2929 	}
2930 }
2931 
2932 static __inline void
2933 vmx_dr_enter_guest(struct vmxctx *vmxctx)
2934 {
2935 	register_t rflags;
2936 
2937 	/* Save host control debug registers. */
2938 	vmxctx->host_dr7 = rdr7();
2939 	vmxctx->host_debugctl = rdmsr(MSR_DEBUGCTLMSR);
2940 
2941 	/*
2942 	 * Disable debugging in DR7 and DEBUGCTL to avoid triggering
2943 	 * exceptions in the host based on the guest DRx values.  The
2944 	 * guest DR7 and DEBUGCTL are saved/restored in the VMCS.
2945 	 */
2946 	load_dr7(0);
2947 	wrmsr(MSR_DEBUGCTLMSR, 0);
2948 
2949 	/*
2950 	 * Disable single stepping the kernel to avoid corrupting the
2951 	 * guest DR6.  A debugger might still be able to corrupt the
2952 	 * guest DR6 by setting a breakpoint after this point and then
2953 	 * single stepping.
2954 	 */
2955 	rflags = read_rflags();
2956 	vmxctx->host_tf = rflags & PSL_T;
2957 	write_rflags(rflags & ~PSL_T);
2958 
2959 	/* Save host debug registers. */
2960 	vmxctx->host_dr0 = rdr0();
2961 	vmxctx->host_dr1 = rdr1();
2962 	vmxctx->host_dr2 = rdr2();
2963 	vmxctx->host_dr3 = rdr3();
2964 	vmxctx->host_dr6 = rdr6();
2965 
2966 	/* Restore guest debug registers. */
2967 	load_dr0(vmxctx->guest_dr0);
2968 	load_dr1(vmxctx->guest_dr1);
2969 	load_dr2(vmxctx->guest_dr2);
2970 	load_dr3(vmxctx->guest_dr3);
2971 	load_dr6(vmxctx->guest_dr6);
2972 }
2973 
2974 static __inline void
2975 vmx_dr_leave_guest(struct vmxctx *vmxctx)
2976 {
2977 
2978 	/* Save guest debug registers. */
2979 	vmxctx->guest_dr0 = rdr0();
2980 	vmxctx->guest_dr1 = rdr1();
2981 	vmxctx->guest_dr2 = rdr2();
2982 	vmxctx->guest_dr3 = rdr3();
2983 	vmxctx->guest_dr6 = rdr6();
2984 
2985 	/*
2986 	 * Restore host debug registers.  Restore DR7, DEBUGCTL, and
2987 	 * PSL_T last.
2988 	 */
2989 	load_dr0(vmxctx->host_dr0);
2990 	load_dr1(vmxctx->host_dr1);
2991 	load_dr2(vmxctx->host_dr2);
2992 	load_dr3(vmxctx->host_dr3);
2993 	load_dr6(vmxctx->host_dr6);
2994 	wrmsr(MSR_DEBUGCTLMSR, vmxctx->host_debugctl);
2995 	load_dr7(vmxctx->host_dr7);
2996 	write_rflags(read_rflags() | vmxctx->host_tf);
2997 }
2998 
2999 static __inline void
3000 vmx_pmap_activate(struct vmx *vmx, pmap_t pmap)
3001 {
3002 	long eptgen;
3003 	int cpu;
3004 
3005 	cpu = curcpu;
3006 
3007 	CPU_SET_ATOMIC(cpu, &pmap->pm_active);
3008 	smr_enter(pmap->pm_eptsmr);
3009 	eptgen = atomic_load_long(&pmap->pm_eptgen);
3010 	if (eptgen != vmx->eptgen[cpu]) {
3011 		vmx->eptgen[cpu] = eptgen;
3012 		invept(INVEPT_TYPE_SINGLE_CONTEXT,
3013 		    (struct invept_desc){ .eptp = vmx->eptp, ._res = 0 });
3014 	}
3015 }
3016 
3017 static __inline void
3018 vmx_pmap_deactivate(struct vmx *vmx, pmap_t pmap)
3019 {
3020 	smr_exit(pmap->pm_eptsmr);
3021 	CPU_CLR_ATOMIC(curcpu, &pmap->pm_active);
3022 }
3023 
3024 static int
3025 vmx_run(void *vcpui, register_t rip, pmap_t pmap, struct vm_eventinfo *evinfo)
3026 {
3027 	int rc, handled, launched;
3028 	struct vmx *vmx;
3029 	struct vmx_vcpu *vcpu;
3030 	struct vmxctx *vmxctx;
3031 	struct vmcs *vmcs;
3032 	struct vm_exit *vmexit;
3033 	struct vlapic *vlapic;
3034 	uint32_t exit_reason;
3035 	struct region_descriptor gdtr, idtr;
3036 	uint16_t ldt_sel;
3037 
3038 	vcpu = vcpui;
3039 	vmx = vcpu->vmx;
3040 	vmcs = vcpu->vmcs;
3041 	vmxctx = &vcpu->ctx;
3042 	vlapic = vm_lapic(vcpu->vcpu);
3043 	vmexit = vm_exitinfo(vcpu->vcpu);
3044 	launched = 0;
3045 
3046 	KASSERT(vmxctx->pmap == pmap,
3047 	    ("pmap %p different than ctx pmap %p", pmap, vmxctx->pmap));
3048 
3049 	vmx_msr_guest_enter(vcpu);
3050 
3051 	VMPTRLD(vmcs);
3052 
3053 	/*
3054 	 * XXX
3055 	 * We do this every time because we may setup the virtual machine
3056 	 * from a different process than the one that actually runs it.
3057 	 *
3058 	 * If the life of a virtual machine was spent entirely in the context
3059 	 * of a single process we could do this once in vmx_init().
3060 	 */
3061 	vmcs_write(VMCS_HOST_CR3, rcr3());
3062 
3063 	vmcs_write(VMCS_GUEST_RIP, rip);
3064 	vmx_set_pcpu_defaults(vmx, vcpu, pmap);
3065 	do {
3066 		KASSERT(vmcs_guest_rip() == rip, ("%s: vmcs guest rip mismatch "
3067 		    "%#lx/%#lx", __func__, vmcs_guest_rip(), rip));
3068 
3069 		handled = UNHANDLED;
3070 		/*
3071 		 * Interrupts are disabled from this point on until the
3072 		 * guest starts executing. This is done for the following
3073 		 * reasons:
3074 		 *
3075 		 * If an AST is asserted on this thread after the check below,
3076 		 * then the IPI_AST notification will not be lost, because it
3077 		 * will cause a VM exit due to external interrupt as soon as
3078 		 * the guest state is loaded.
3079 		 *
3080 		 * A posted interrupt after 'vmx_inject_interrupts()' will
3081 		 * not be "lost" because it will be held pending in the host
3082 		 * APIC because interrupts are disabled. The pending interrupt
3083 		 * will be recognized as soon as the guest state is loaded.
3084 		 *
3085 		 * The same reasoning applies to the IPI generated by
3086 		 * pmap_invalidate_ept().
3087 		 */
3088 		disable_intr();
3089 		vmx_inject_interrupts(vcpu, vlapic, rip);
3090 
3091 		/*
3092 		 * Check for vcpu suspension after injecting events because
3093 		 * vmx_inject_interrupts() can suspend the vcpu due to a
3094 		 * triple fault.
3095 		 */
3096 		if (vcpu_suspended(evinfo)) {
3097 			enable_intr();
3098 			vm_exit_suspended(vcpu->vcpu, rip);
3099 			break;
3100 		}
3101 
3102 		if (vcpu_rendezvous_pending(vcpu->vcpu, evinfo)) {
3103 			enable_intr();
3104 			vm_exit_rendezvous(vcpu->vcpu, rip);
3105 			break;
3106 		}
3107 
3108 		if (vcpu_reqidle(evinfo)) {
3109 			enable_intr();
3110 			vm_exit_reqidle(vcpu->vcpu, rip);
3111 			break;
3112 		}
3113 
3114 		if (vcpu_should_yield(vcpu->vcpu)) {
3115 			enable_intr();
3116 			vm_exit_astpending(vcpu->vcpu, rip);
3117 			vmx_astpending_trace(vcpu, rip);
3118 			handled = HANDLED;
3119 			break;
3120 		}
3121 
3122 		if (vcpu_debugged(vcpu->vcpu)) {
3123 			enable_intr();
3124 			vm_exit_debug(vcpu->vcpu, rip);
3125 			break;
3126 		}
3127 
3128 		/*
3129 		 * If TPR Shadowing is enabled, the TPR Threshold
3130 		 * must be updated right before entering the guest.
3131 		 */
3132 		if (tpr_shadowing && !virtual_interrupt_delivery) {
3133 			if ((vcpu->cap.proc_ctls & PROCBASED_USE_TPR_SHADOW) != 0) {
3134 				vmcs_write(VMCS_TPR_THRESHOLD, vlapic_get_cr8(vlapic));
3135 			}
3136 		}
3137 
3138 		/*
3139 		 * VM exits restore the base address but not the
3140 		 * limits of GDTR and IDTR.  The VMCS only stores the
3141 		 * base address, so VM exits set the limits to 0xffff.
3142 		 * Save and restore the full GDTR and IDTR to restore
3143 		 * the limits.
3144 		 *
3145 		 * The VMCS does not save the LDTR at all, and VM
3146 		 * exits clear LDTR as if a NULL selector were loaded.
3147 		 * The userspace hypervisor probably doesn't use a
3148 		 * LDT, but save and restore it to be safe.
3149 		 */
3150 		sgdt(&gdtr);
3151 		sidt(&idtr);
3152 		ldt_sel = sldt();
3153 
3154 		/*
3155 		 * The TSC_AUX MSR must be saved/restored while interrupts
3156 		 * are disabled so that it is not possible for the guest
3157 		 * TSC_AUX MSR value to be overwritten by the resume
3158 		 * portion of the IPI_SUSPEND codepath. This is why the
3159 		 * transition of this MSR is handled separately from those
3160 		 * handled by vmx_msr_guest_{enter,exit}(), which are ok to
3161 		 * be transitioned with preemption disabled but interrupts
3162 		 * enabled.
3163 		 *
3164 		 * These vmx_msr_guest_{enter,exit}_tsc_aux() calls can be
3165 		 * anywhere in this loop so long as they happen with
3166 		 * interrupts disabled. This location is chosen for
3167 		 * simplicity.
3168 		 */
3169 		vmx_msr_guest_enter_tsc_aux(vmx, vcpu);
3170 
3171 		vmx_dr_enter_guest(vmxctx);
3172 
3173 		/*
3174 		 * Mark the EPT as active on this host CPU and invalidate
3175 		 * EPTP-tagged TLB entries if required.
3176 		 */
3177 		vmx_pmap_activate(vmx, pmap);
3178 
3179 		vmx_run_trace(vcpu);
3180 		rc = vmx_enter_guest(vmxctx, vmx, launched);
3181 
3182 		vmx_pmap_deactivate(vmx, pmap);
3183 		vmx_dr_leave_guest(vmxctx);
3184 		vmx_msr_guest_exit_tsc_aux(vmx, vcpu);
3185 
3186 		bare_lgdt(&gdtr);
3187 		lidt(&idtr);
3188 		lldt(ldt_sel);
3189 
3190 		/* Collect some information for VM exit processing */
3191 		vmexit->rip = rip = vmcs_guest_rip();
3192 		vmexit->inst_length = vmexit_instruction_length();
3193 		vmexit->u.vmx.exit_reason = exit_reason = vmcs_exit_reason();
3194 		vmexit->u.vmx.exit_qualification = vmcs_exit_qualification();
3195 
3196 		/* Update 'nextrip' */
3197 		vcpu->state.nextrip = rip;
3198 
3199 		if (rc == VMX_GUEST_VMEXIT) {
3200 			vmx_exit_handle_nmi(vcpu, vmexit);
3201 			enable_intr();
3202 			handled = vmx_exit_process(vmx, vcpu, vmexit);
3203 		} else {
3204 			enable_intr();
3205 			vmx_exit_inst_error(vmxctx, rc, vmexit);
3206 		}
3207 		launched = 1;
3208 		vmx_exit_trace(vcpu, rip, exit_reason, handled);
3209 		rip = vmexit->rip;
3210 	} while (handled);
3211 
3212 	/*
3213 	 * If a VM exit has been handled then the exitcode must be BOGUS
3214 	 * If a VM exit is not handled then the exitcode must not be BOGUS
3215 	 */
3216 	if ((handled && vmexit->exitcode != VM_EXITCODE_BOGUS) ||
3217 	    (!handled && vmexit->exitcode == VM_EXITCODE_BOGUS)) {
3218 		panic("Mismatch between handled (%d) and exitcode (%d)",
3219 		      handled, vmexit->exitcode);
3220 	}
3221 
3222 	VMX_CTR1(vcpu, "returning from vmx_run: exitcode %d",
3223 	    vmexit->exitcode);
3224 
3225 	VMCLEAR(vmcs);
3226 	vmx_msr_guest_exit(vcpu);
3227 
3228 	return (0);
3229 }
3230 
3231 static void
3232 vmx_vcpu_cleanup(void *vcpui)
3233 {
3234 	struct vmx_vcpu *vcpu = vcpui;
3235 
3236 	vpid_free(vcpu->state.vpid);
3237 	free(vcpu->pir_desc, M_VMX);
3238 	free(vcpu->apic_page, M_VMX);
3239 	free(vcpu->vmcs, M_VMX);
3240 	free(vcpu, M_VMX);
3241 }
3242 
3243 static void
3244 vmx_cleanup(void *vmi)
3245 {
3246 	struct vmx *vmx = vmi;
3247 
3248 	if (virtual_interrupt_delivery)
3249 		vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE);
3250 
3251 	free(vmx->msr_bitmap, M_VMX);
3252 	free(vmx, M_VMX);
3253 
3254 	return;
3255 }
3256 
3257 static register_t *
3258 vmxctx_regptr(struct vmxctx *vmxctx, int reg)
3259 {
3260 
3261 	switch (reg) {
3262 	case VM_REG_GUEST_RAX:
3263 		return (&vmxctx->guest_rax);
3264 	case VM_REG_GUEST_RBX:
3265 		return (&vmxctx->guest_rbx);
3266 	case VM_REG_GUEST_RCX:
3267 		return (&vmxctx->guest_rcx);
3268 	case VM_REG_GUEST_RDX:
3269 		return (&vmxctx->guest_rdx);
3270 	case VM_REG_GUEST_RSI:
3271 		return (&vmxctx->guest_rsi);
3272 	case VM_REG_GUEST_RDI:
3273 		return (&vmxctx->guest_rdi);
3274 	case VM_REG_GUEST_RBP:
3275 		return (&vmxctx->guest_rbp);
3276 	case VM_REG_GUEST_R8:
3277 		return (&vmxctx->guest_r8);
3278 	case VM_REG_GUEST_R9:
3279 		return (&vmxctx->guest_r9);
3280 	case VM_REG_GUEST_R10:
3281 		return (&vmxctx->guest_r10);
3282 	case VM_REG_GUEST_R11:
3283 		return (&vmxctx->guest_r11);
3284 	case VM_REG_GUEST_R12:
3285 		return (&vmxctx->guest_r12);
3286 	case VM_REG_GUEST_R13:
3287 		return (&vmxctx->guest_r13);
3288 	case VM_REG_GUEST_R14:
3289 		return (&vmxctx->guest_r14);
3290 	case VM_REG_GUEST_R15:
3291 		return (&vmxctx->guest_r15);
3292 	case VM_REG_GUEST_CR2:
3293 		return (&vmxctx->guest_cr2);
3294 	case VM_REG_GUEST_DR0:
3295 		return (&vmxctx->guest_dr0);
3296 	case VM_REG_GUEST_DR1:
3297 		return (&vmxctx->guest_dr1);
3298 	case VM_REG_GUEST_DR2:
3299 		return (&vmxctx->guest_dr2);
3300 	case VM_REG_GUEST_DR3:
3301 		return (&vmxctx->guest_dr3);
3302 	case VM_REG_GUEST_DR6:
3303 		return (&vmxctx->guest_dr6);
3304 	default:
3305 		break;
3306 	}
3307 	return (NULL);
3308 }
3309 
3310 static int
3311 vmxctx_getreg(struct vmxctx *vmxctx, int reg, uint64_t *retval)
3312 {
3313 	register_t *regp;
3314 
3315 	if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) {
3316 		*retval = *regp;
3317 		return (0);
3318 	} else
3319 		return (EINVAL);
3320 }
3321 
3322 static int
3323 vmxctx_setreg(struct vmxctx *vmxctx, int reg, uint64_t val)
3324 {
3325 	register_t *regp;
3326 
3327 	if ((regp = vmxctx_regptr(vmxctx, reg)) != NULL) {
3328 		*regp = val;
3329 		return (0);
3330 	} else
3331 		return (EINVAL);
3332 }
3333 
3334 static int
3335 vmx_get_intr_shadow(struct vmx_vcpu *vcpu, int running, uint64_t *retval)
3336 {
3337 	uint64_t gi;
3338 	int error;
3339 
3340 	error = vmcs_getreg(vcpu->vmcs, running,
3341 	    VMCS_IDENT(VMCS_GUEST_INTERRUPTIBILITY), &gi);
3342 	*retval = (gi & HWINTR_BLOCKING) ? 1 : 0;
3343 	return (error);
3344 }
3345 
3346 static int
3347 vmx_modify_intr_shadow(struct vmx_vcpu *vcpu, int running, uint64_t val)
3348 {
3349 	struct vmcs *vmcs;
3350 	uint64_t gi;
3351 	int error, ident;
3352 
3353 	/*
3354 	 * Forcing the vcpu into an interrupt shadow is not supported.
3355 	 */
3356 	if (val) {
3357 		error = EINVAL;
3358 		goto done;
3359 	}
3360 
3361 	vmcs = vcpu->vmcs;
3362 	ident = VMCS_IDENT(VMCS_GUEST_INTERRUPTIBILITY);
3363 	error = vmcs_getreg(vmcs, running, ident, &gi);
3364 	if (error == 0) {
3365 		gi &= ~HWINTR_BLOCKING;
3366 		error = vmcs_setreg(vmcs, running, ident, gi);
3367 	}
3368 done:
3369 	VMX_CTR2(vcpu, "Setting intr_shadow to %#lx %s", val,
3370 	    error ? "failed" : "succeeded");
3371 	return (error);
3372 }
3373 
3374 static int
3375 vmx_shadow_reg(int reg)
3376 {
3377 	int shreg;
3378 
3379 	shreg = -1;
3380 
3381 	switch (reg) {
3382 	case VM_REG_GUEST_CR0:
3383 		shreg = VMCS_CR0_SHADOW;
3384 		break;
3385 	case VM_REG_GUEST_CR4:
3386 		shreg = VMCS_CR4_SHADOW;
3387 		break;
3388 	default:
3389 		break;
3390 	}
3391 
3392 	return (shreg);
3393 }
3394 
3395 static int
3396 vmx_getreg(void *vcpui, int reg, uint64_t *retval)
3397 {
3398 	int running, hostcpu;
3399 	struct vmx_vcpu *vcpu = vcpui;
3400 	struct vmx *vmx = vcpu->vmx;
3401 
3402 	running = vcpu_is_running(vcpu->vcpu, &hostcpu);
3403 	if (running && hostcpu != curcpu)
3404 		panic("vmx_getreg: %s%d is running", vm_name(vmx->vm),
3405 		    vcpu->vcpuid);
3406 
3407 	if (reg == VM_REG_GUEST_INTR_SHADOW)
3408 		return (vmx_get_intr_shadow(vcpu, running, retval));
3409 
3410 	if (vmxctx_getreg(&vcpu->ctx, reg, retval) == 0)
3411 		return (0);
3412 
3413 	return (vmcs_getreg(vcpu->vmcs, running, reg, retval));
3414 }
3415 
3416 static int
3417 vmx_setreg(void *vcpui, int reg, uint64_t val)
3418 {
3419 	int error, hostcpu, running, shadow;
3420 	uint64_t ctls;
3421 	pmap_t pmap;
3422 	struct vmx_vcpu *vcpu = vcpui;
3423 	struct vmx *vmx = vcpu->vmx;
3424 
3425 	running = vcpu_is_running(vcpu->vcpu, &hostcpu);
3426 	if (running && hostcpu != curcpu)
3427 		panic("vmx_setreg: %s%d is running", vm_name(vmx->vm),
3428 		    vcpu->vcpuid);
3429 
3430 	if (reg == VM_REG_GUEST_INTR_SHADOW)
3431 		return (vmx_modify_intr_shadow(vcpu, running, val));
3432 
3433 	if (vmxctx_setreg(&vcpu->ctx, reg, val) == 0)
3434 		return (0);
3435 
3436 	/* Do not permit user write access to VMCS fields by offset. */
3437 	if (reg < 0)
3438 		return (EINVAL);
3439 
3440 	error = vmcs_setreg(vcpu->vmcs, running, reg, val);
3441 
3442 	if (error == 0) {
3443 		/*
3444 		 * If the "load EFER" VM-entry control is 1 then the
3445 		 * value of EFER.LMA must be identical to "IA-32e mode guest"
3446 		 * bit in the VM-entry control.
3447 		 */
3448 		if ((entry_ctls & VM_ENTRY_LOAD_EFER) != 0 &&
3449 		    (reg == VM_REG_GUEST_EFER)) {
3450 			vmcs_getreg(vcpu->vmcs, running,
3451 				    VMCS_IDENT(VMCS_ENTRY_CTLS), &ctls);
3452 			if (val & EFER_LMA)
3453 				ctls |= VM_ENTRY_GUEST_LMA;
3454 			else
3455 				ctls &= ~VM_ENTRY_GUEST_LMA;
3456 			vmcs_setreg(vcpu->vmcs, running,
3457 				    VMCS_IDENT(VMCS_ENTRY_CTLS), ctls);
3458 		}
3459 
3460 		shadow = vmx_shadow_reg(reg);
3461 		if (shadow > 0) {
3462 			/*
3463 			 * Store the unmodified value in the shadow
3464 			 */
3465 			error = vmcs_setreg(vcpu->vmcs, running,
3466 				    VMCS_IDENT(shadow), val);
3467 		}
3468 
3469 		if (reg == VM_REG_GUEST_CR3) {
3470 			/*
3471 			 * Invalidate the guest vcpu's TLB mappings to emulate
3472 			 * the behavior of updating %cr3.
3473 			 *
3474 			 * XXX the processor retains global mappings when %cr3
3475 			 * is updated but vmx_invvpid() does not.
3476 			 */
3477 			pmap = vcpu->ctx.pmap;
3478 			vmx_invvpid(vmx, vcpu, pmap, running);
3479 		}
3480 	}
3481 
3482 	return (error);
3483 }
3484 
3485 static int
3486 vmx_getdesc(void *vcpui, int reg, struct seg_desc *desc)
3487 {
3488 	int hostcpu, running;
3489 	struct vmx_vcpu *vcpu = vcpui;
3490 	struct vmx *vmx = vcpu->vmx;
3491 
3492 	running = vcpu_is_running(vcpu->vcpu, &hostcpu);
3493 	if (running && hostcpu != curcpu)
3494 		panic("vmx_getdesc: %s%d is running", vm_name(vmx->vm),
3495 		    vcpu->vcpuid);
3496 
3497 	return (vmcs_getdesc(vcpu->vmcs, running, reg, desc));
3498 }
3499 
3500 static int
3501 vmx_setdesc(void *vcpui, int reg, struct seg_desc *desc)
3502 {
3503 	int hostcpu, running;
3504 	struct vmx_vcpu *vcpu = vcpui;
3505 	struct vmx *vmx = vcpu->vmx;
3506 
3507 	running = vcpu_is_running(vcpu->vcpu, &hostcpu);
3508 	if (running && hostcpu != curcpu)
3509 		panic("vmx_setdesc: %s%d is running", vm_name(vmx->vm),
3510 		    vcpu->vcpuid);
3511 
3512 	return (vmcs_setdesc(vcpu->vmcs, running, reg, desc));
3513 }
3514 
3515 static int
3516 vmx_getcap(void *vcpui, int type, int *retval)
3517 {
3518 	struct vmx_vcpu *vcpu = vcpui;
3519 	int vcap;
3520 	int ret;
3521 
3522 	ret = ENOENT;
3523 
3524 	vcap = vcpu->cap.set;
3525 
3526 	switch (type) {
3527 	case VM_CAP_HALT_EXIT:
3528 		if (cap_halt_exit)
3529 			ret = 0;
3530 		break;
3531 	case VM_CAP_PAUSE_EXIT:
3532 		if (cap_pause_exit)
3533 			ret = 0;
3534 		break;
3535 	case VM_CAP_MTRAP_EXIT:
3536 		if (cap_monitor_trap)
3537 			ret = 0;
3538 		break;
3539 	case VM_CAP_RDPID:
3540 		if (cap_rdpid)
3541 			ret = 0;
3542 		break;
3543 	case VM_CAP_RDTSCP:
3544 		if (cap_rdtscp)
3545 			ret = 0;
3546 		break;
3547 	case VM_CAP_UNRESTRICTED_GUEST:
3548 		if (cap_unrestricted_guest)
3549 			ret = 0;
3550 		break;
3551 	case VM_CAP_ENABLE_INVPCID:
3552 		if (cap_invpcid)
3553 			ret = 0;
3554 		break;
3555 	case VM_CAP_BPT_EXIT:
3556 	case VM_CAP_IPI_EXIT:
3557 		ret = 0;
3558 		break;
3559 	default:
3560 		break;
3561 	}
3562 
3563 	if (ret == 0)
3564 		*retval = (vcap & (1 << type)) ? 1 : 0;
3565 
3566 	return (ret);
3567 }
3568 
3569 static int
3570 vmx_setcap(void *vcpui, int type, int val)
3571 {
3572 	struct vmx_vcpu *vcpu = vcpui;
3573 	struct vmcs *vmcs = vcpu->vmcs;
3574 	struct vlapic *vlapic;
3575 	uint32_t baseval;
3576 	uint32_t *pptr;
3577 	int error;
3578 	int flag;
3579 	int reg;
3580 	int retval;
3581 
3582 	retval = ENOENT;
3583 	pptr = NULL;
3584 
3585 	switch (type) {
3586 	case VM_CAP_HALT_EXIT:
3587 		if (cap_halt_exit) {
3588 			retval = 0;
3589 			pptr = &vcpu->cap.proc_ctls;
3590 			baseval = *pptr;
3591 			flag = PROCBASED_HLT_EXITING;
3592 			reg = VMCS_PRI_PROC_BASED_CTLS;
3593 		}
3594 		break;
3595 	case VM_CAP_MTRAP_EXIT:
3596 		if (cap_monitor_trap) {
3597 			retval = 0;
3598 			pptr = &vcpu->cap.proc_ctls;
3599 			baseval = *pptr;
3600 			flag = PROCBASED_MTF;
3601 			reg = VMCS_PRI_PROC_BASED_CTLS;
3602 		}
3603 		break;
3604 	case VM_CAP_PAUSE_EXIT:
3605 		if (cap_pause_exit) {
3606 			retval = 0;
3607 			pptr = &vcpu->cap.proc_ctls;
3608 			baseval = *pptr;
3609 			flag = PROCBASED_PAUSE_EXITING;
3610 			reg = VMCS_PRI_PROC_BASED_CTLS;
3611 		}
3612 		break;
3613 	case VM_CAP_RDPID:
3614 	case VM_CAP_RDTSCP:
3615 		if (cap_rdpid || cap_rdtscp)
3616 			/*
3617 			 * Choose not to support enabling/disabling
3618 			 * RDPID/RDTSCP via libvmmapi since, as per the
3619 			 * discussion in vmx_modinit(), RDPID/RDTSCP are
3620 			 * either always enabled or always disabled.
3621 			 */
3622 			error = EOPNOTSUPP;
3623 		break;
3624 	case VM_CAP_UNRESTRICTED_GUEST:
3625 		if (cap_unrestricted_guest) {
3626 			retval = 0;
3627 			pptr = &vcpu->cap.proc_ctls2;
3628 			baseval = *pptr;
3629 			flag = PROCBASED2_UNRESTRICTED_GUEST;
3630 			reg = VMCS_SEC_PROC_BASED_CTLS;
3631 		}
3632 		break;
3633 	case VM_CAP_ENABLE_INVPCID:
3634 		if (cap_invpcid) {
3635 			retval = 0;
3636 			pptr = &vcpu->cap.proc_ctls2;
3637 			baseval = *pptr;
3638 			flag = PROCBASED2_ENABLE_INVPCID;
3639 			reg = VMCS_SEC_PROC_BASED_CTLS;
3640 		}
3641 		break;
3642 	case VM_CAP_BPT_EXIT:
3643 		retval = 0;
3644 
3645 		/* Don't change the bitmap if we are tracing all exceptions. */
3646 		if (vcpu->cap.exc_bitmap != 0xffffffff) {
3647 			pptr = &vcpu->cap.exc_bitmap;
3648 			baseval = *pptr;
3649 			flag = (1 << IDT_BP);
3650 			reg = VMCS_EXCEPTION_BITMAP;
3651 		}
3652 		break;
3653 	case VM_CAP_IPI_EXIT:
3654 		retval = 0;
3655 
3656 		vlapic = vm_lapic(vcpu->vcpu);
3657 		vlapic->ipi_exit = val;
3658 		break;
3659 	case VM_CAP_MASK_HWINTR:
3660 		retval = 0;
3661 		break;
3662 	default:
3663 		break;
3664 	}
3665 
3666 	if (retval)
3667 		return (retval);
3668 
3669 	if (pptr != NULL) {
3670 		if (val) {
3671 			baseval |= flag;
3672 		} else {
3673 			baseval &= ~flag;
3674 		}
3675 		VMPTRLD(vmcs);
3676 		error = vmwrite(reg, baseval);
3677 		VMCLEAR(vmcs);
3678 
3679 		if (error)
3680 			return (error);
3681 
3682 		/*
3683 		 * Update optional stored flags, and record
3684 		 * setting
3685 		 */
3686 		*pptr = baseval;
3687 	}
3688 
3689 	if (val) {
3690 		vcpu->cap.set |= (1 << type);
3691 	} else {
3692 		vcpu->cap.set &= ~(1 << type);
3693 	}
3694 
3695 	return (0);
3696 }
3697 
3698 static struct vmspace *
3699 vmx_vmspace_alloc(vm_offset_t min, vm_offset_t max)
3700 {
3701 	return (ept_vmspace_alloc(min, max));
3702 }
3703 
3704 static void
3705 vmx_vmspace_free(struct vmspace *vmspace)
3706 {
3707 	ept_vmspace_free(vmspace);
3708 }
3709 
3710 struct vlapic_vtx {
3711 	struct vlapic	vlapic;
3712 	struct pir_desc	*pir_desc;
3713 	struct vmx_vcpu	*vcpu;
3714 	u_int	pending_prio;
3715 };
3716 
3717 #define VPR_PRIO_BIT(vpr)	(1 << ((vpr) >> 4))
3718 
3719 #define	VMX_CTR_PIR(vlapic, pir_desc, notify, vector, level, msg)	\
3720 do {									\
3721 	VLAPIC_CTR2(vlapic, msg " assert %s-triggered vector %d",	\
3722 	    level ? "level" : "edge", vector);				\
3723 	VLAPIC_CTR1(vlapic, msg " pir0 0x%016lx", pir_desc->pir[0]);	\
3724 	VLAPIC_CTR1(vlapic, msg " pir1 0x%016lx", pir_desc->pir[1]);	\
3725 	VLAPIC_CTR1(vlapic, msg " pir2 0x%016lx", pir_desc->pir[2]);	\
3726 	VLAPIC_CTR1(vlapic, msg " pir3 0x%016lx", pir_desc->pir[3]);	\
3727 	VLAPIC_CTR1(vlapic, msg " notify: %s", notify ? "yes" : "no");	\
3728 } while (0)
3729 
3730 /*
3731  * vlapic->ops handlers that utilize the APICv hardware assist described in
3732  * Chapter 29 of the Intel SDM.
3733  */
3734 static int
3735 vmx_set_intr_ready(struct vlapic *vlapic, int vector, bool level)
3736 {
3737 	struct vlapic_vtx *vlapic_vtx;
3738 	struct pir_desc *pir_desc;
3739 	uint64_t mask;
3740 	int idx, notify = 0;
3741 
3742 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
3743 	pir_desc = vlapic_vtx->pir_desc;
3744 
3745 	/*
3746 	 * Keep track of interrupt requests in the PIR descriptor. This is
3747 	 * because the virtual APIC page pointed to by the VMCS cannot be
3748 	 * modified if the vcpu is running.
3749 	 */
3750 	idx = vector / 64;
3751 	mask = 1UL << (vector % 64);
3752 	atomic_set_long(&pir_desc->pir[idx], mask);
3753 
3754 	/*
3755 	 * A notification is required whenever the 'pending' bit makes a
3756 	 * transition from 0->1.
3757 	 *
3758 	 * Even if the 'pending' bit is already asserted, notification about
3759 	 * the incoming interrupt may still be necessary.  For example, if a
3760 	 * vCPU is HLTed with a high PPR, a low priority interrupt would cause
3761 	 * the 0->1 'pending' transition with a notification, but the vCPU
3762 	 * would ignore the interrupt for the time being.  The same vCPU would
3763 	 * need to then be notified if a high-priority interrupt arrived which
3764 	 * satisfied the PPR.
3765 	 *
3766 	 * The priorities of interrupts injected while 'pending' is asserted
3767 	 * are tracked in a custom bitfield 'pending_prio'.  Should the
3768 	 * to-be-injected interrupt exceed the priorities already present, the
3769 	 * notification is sent.  The priorities recorded in 'pending_prio' are
3770 	 * cleared whenever the 'pending' bit makes another 0->1 transition.
3771 	 */
3772 	if (atomic_cmpset_long(&pir_desc->pending, 0, 1) != 0) {
3773 		notify = 1;
3774 		vlapic_vtx->pending_prio = 0;
3775 	} else {
3776 		const u_int old_prio = vlapic_vtx->pending_prio;
3777 		const u_int prio_bit = VPR_PRIO_BIT(vector & APIC_TPR_INT);
3778 
3779 		if ((old_prio & prio_bit) == 0 && prio_bit > old_prio) {
3780 			atomic_set_int(&vlapic_vtx->pending_prio, prio_bit);
3781 			notify = 1;
3782 		}
3783 	}
3784 
3785 	VMX_CTR_PIR(vlapic, pir_desc, notify, vector, level,
3786 	    "vmx_set_intr_ready");
3787 	return (notify);
3788 }
3789 
3790 static int
3791 vmx_pending_intr(struct vlapic *vlapic, int *vecptr)
3792 {
3793 	struct vlapic_vtx *vlapic_vtx;
3794 	struct pir_desc *pir_desc;
3795 	struct LAPIC *lapic;
3796 	uint64_t pending, pirval;
3797 	uint8_t ppr, vpr, rvi;
3798 	struct vm_exit *vmexit;
3799 	int i;
3800 
3801 	/*
3802 	 * This function is only expected to be called from the 'HLT' exit
3803 	 * handler which does not care about the vector that is pending.
3804 	 */
3805 	KASSERT(vecptr == NULL, ("vmx_pending_intr: vecptr must be NULL"));
3806 
3807 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
3808 	pir_desc = vlapic_vtx->pir_desc;
3809 	lapic = vlapic->apic_page;
3810 
3811 	/*
3812 	 * While a virtual interrupt may have already been
3813 	 * processed the actual delivery maybe pending the
3814 	 * interruptibility of the guest.  Recognize a pending
3815 	 * interrupt by reevaluating virtual interrupts
3816 	 * following Section 30.2.1 in the Intel SDM Volume 3.
3817 	 */
3818 	vmexit = vm_exitinfo(vlapic->vcpu);
3819 	KASSERT(vmexit->exitcode == VM_EXITCODE_HLT,
3820 	    ("vmx_pending_intr: exitcode not 'HLT'"));
3821 	rvi = vmexit->u.hlt.intr_status & APIC_TPR_INT;
3822 	ppr = lapic->ppr & APIC_TPR_INT;
3823 	if (rvi > ppr)
3824 		return (1);
3825 
3826 	pending = atomic_load_acq_long(&pir_desc->pending);
3827 	if (!pending)
3828 		return (0);
3829 
3830 	/*
3831 	 * If there is an interrupt pending then it will be recognized only
3832 	 * if its priority is greater than the processor priority.
3833 	 *
3834 	 * Special case: if the processor priority is zero then any pending
3835 	 * interrupt will be recognized.
3836 	 */
3837 	if (ppr == 0)
3838 		return (1);
3839 
3840 	VLAPIC_CTR1(vlapic, "HLT with non-zero PPR %d", lapic->ppr);
3841 
3842 	vpr = 0;
3843 	for (i = 3; i >= 0; i--) {
3844 		pirval = pir_desc->pir[i];
3845 		if (pirval != 0) {
3846 			vpr = (i * 64 + flsl(pirval) - 1) & APIC_TPR_INT;
3847 			break;
3848 		}
3849 	}
3850 
3851 	/*
3852 	 * If the highest-priority pending interrupt falls short of the
3853 	 * processor priority of this vCPU, ensure that 'pending_prio' does not
3854 	 * have any stale bits which would preclude a higher-priority interrupt
3855 	 * from incurring a notification later.
3856 	 */
3857 	if (vpr <= ppr) {
3858 		const u_int prio_bit = VPR_PRIO_BIT(vpr);
3859 		const u_int old = vlapic_vtx->pending_prio;
3860 
3861 		if (old > prio_bit && (old & prio_bit) == 0) {
3862 			vlapic_vtx->pending_prio = prio_bit;
3863 		}
3864 		return (0);
3865 	}
3866 	return (1);
3867 }
3868 
3869 static void
3870 vmx_intr_accepted(struct vlapic *vlapic, int vector)
3871 {
3872 
3873 	panic("vmx_intr_accepted: not expected to be called");
3874 }
3875 
3876 static void
3877 vmx_set_tmr(struct vlapic *vlapic, int vector, bool level)
3878 {
3879 	struct vlapic_vtx *vlapic_vtx;
3880 	struct vmcs *vmcs;
3881 	uint64_t mask, val;
3882 
3883 	KASSERT(vector >= 0 && vector <= 255, ("invalid vector %d", vector));
3884 	KASSERT(!vcpu_is_running(vlapic->vcpu, NULL),
3885 	    ("vmx_set_tmr: vcpu cannot be running"));
3886 
3887 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
3888 	vmcs = vlapic_vtx->vcpu->vmcs;
3889 	mask = 1UL << (vector % 64);
3890 
3891 	VMPTRLD(vmcs);
3892 	val = vmcs_read(VMCS_EOI_EXIT(vector));
3893 	if (level)
3894 		val |= mask;
3895 	else
3896 		val &= ~mask;
3897 	vmcs_write(VMCS_EOI_EXIT(vector), val);
3898 	VMCLEAR(vmcs);
3899 }
3900 
3901 static void
3902 vmx_enable_x2apic_mode_ts(struct vlapic *vlapic)
3903 {
3904 	struct vlapic_vtx *vlapic_vtx;
3905 	struct vmx_vcpu *vcpu;
3906 	struct vmcs *vmcs;
3907 	uint32_t proc_ctls;
3908 
3909 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
3910 	vcpu = vlapic_vtx->vcpu;
3911 	vmcs = vcpu->vmcs;
3912 
3913 	proc_ctls = vcpu->cap.proc_ctls;
3914 	proc_ctls &= ~PROCBASED_USE_TPR_SHADOW;
3915 	proc_ctls |= PROCBASED_CR8_LOAD_EXITING;
3916 	proc_ctls |= PROCBASED_CR8_STORE_EXITING;
3917 	vcpu->cap.proc_ctls = proc_ctls;
3918 
3919 	VMPTRLD(vmcs);
3920 	vmcs_write(VMCS_PRI_PROC_BASED_CTLS, proc_ctls);
3921 	VMCLEAR(vmcs);
3922 }
3923 
3924 static void
3925 vmx_enable_x2apic_mode_vid(struct vlapic *vlapic)
3926 {
3927 	struct vlapic_vtx *vlapic_vtx;
3928 	struct vmx *vmx;
3929 	struct vmx_vcpu *vcpu;
3930 	struct vmcs *vmcs;
3931 	uint32_t proc_ctls2;
3932 	int error __diagused;
3933 
3934 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
3935 	vcpu = vlapic_vtx->vcpu;
3936 	vmx = vcpu->vmx;
3937 	vmcs = vcpu->vmcs;
3938 
3939 	proc_ctls2 = vcpu->cap.proc_ctls2;
3940 	KASSERT((proc_ctls2 & PROCBASED2_VIRTUALIZE_APIC_ACCESSES) != 0,
3941 	    ("%s: invalid proc_ctls2 %#x", __func__, proc_ctls2));
3942 
3943 	proc_ctls2 &= ~PROCBASED2_VIRTUALIZE_APIC_ACCESSES;
3944 	proc_ctls2 |= PROCBASED2_VIRTUALIZE_X2APIC_MODE;
3945 	vcpu->cap.proc_ctls2 = proc_ctls2;
3946 
3947 	VMPTRLD(vmcs);
3948 	vmcs_write(VMCS_SEC_PROC_BASED_CTLS, proc_ctls2);
3949 	VMCLEAR(vmcs);
3950 
3951 	if (vlapic->vcpuid == 0) {
3952 		/*
3953 		 * The nested page table mappings are shared by all vcpus
3954 		 * so unmap the APIC access page just once.
3955 		 */
3956 		error = vm_unmap_mmio(vmx->vm, DEFAULT_APIC_BASE, PAGE_SIZE);
3957 		KASSERT(error == 0, ("%s: vm_unmap_mmio error %d",
3958 		    __func__, error));
3959 
3960 		/*
3961 		 * The MSR bitmap is shared by all vcpus so modify it only
3962 		 * once in the context of vcpu 0.
3963 		 */
3964 		error = vmx_allow_x2apic_msrs(vmx);
3965 		KASSERT(error == 0, ("%s: vmx_allow_x2apic_msrs error %d",
3966 		    __func__, error));
3967 	}
3968 }
3969 
3970 static void
3971 vmx_post_intr(struct vlapic *vlapic, int hostcpu)
3972 {
3973 
3974 	ipi_cpu(hostcpu, pirvec);
3975 }
3976 
3977 /*
3978  * Transfer the pending interrupts in the PIR descriptor to the IRR
3979  * in the virtual APIC page.
3980  */
3981 static void
3982 vmx_inject_pir(struct vlapic *vlapic)
3983 {
3984 	struct vlapic_vtx *vlapic_vtx;
3985 	struct pir_desc *pir_desc;
3986 	struct LAPIC *lapic;
3987 	uint64_t val, pirval;
3988 	int rvi, pirbase = -1;
3989 	uint16_t intr_status_old, intr_status_new;
3990 
3991 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
3992 	pir_desc = vlapic_vtx->pir_desc;
3993 	if (atomic_cmpset_long(&pir_desc->pending, 1, 0) == 0) {
3994 		VLAPIC_CTR0(vlapic, "vmx_inject_pir: "
3995 		    "no posted interrupt pending");
3996 		return;
3997 	}
3998 
3999 	pirval = 0;
4000 	pirbase = -1;
4001 	lapic = vlapic->apic_page;
4002 
4003 	val = atomic_readandclear_long(&pir_desc->pir[0]);
4004 	if (val != 0) {
4005 		lapic->irr0 |= val;
4006 		lapic->irr1 |= val >> 32;
4007 		pirbase = 0;
4008 		pirval = val;
4009 	}
4010 
4011 	val = atomic_readandclear_long(&pir_desc->pir[1]);
4012 	if (val != 0) {
4013 		lapic->irr2 |= val;
4014 		lapic->irr3 |= val >> 32;
4015 		pirbase = 64;
4016 		pirval = val;
4017 	}
4018 
4019 	val = atomic_readandclear_long(&pir_desc->pir[2]);
4020 	if (val != 0) {
4021 		lapic->irr4 |= val;
4022 		lapic->irr5 |= val >> 32;
4023 		pirbase = 128;
4024 		pirval = val;
4025 	}
4026 
4027 	val = atomic_readandclear_long(&pir_desc->pir[3]);
4028 	if (val != 0) {
4029 		lapic->irr6 |= val;
4030 		lapic->irr7 |= val >> 32;
4031 		pirbase = 192;
4032 		pirval = val;
4033 	}
4034 
4035 	VLAPIC_CTR_IRR(vlapic, "vmx_inject_pir");
4036 
4037 	/*
4038 	 * Update RVI so the processor can evaluate pending virtual
4039 	 * interrupts on VM-entry.
4040 	 *
4041 	 * It is possible for pirval to be 0 here, even though the
4042 	 * pending bit has been set. The scenario is:
4043 	 * CPU-Y is sending a posted interrupt to CPU-X, which
4044 	 * is running a guest and processing posted interrupts in h/w.
4045 	 * CPU-X will eventually exit and the state seen in s/w is
4046 	 * the pending bit set, but no PIR bits set.
4047 	 *
4048 	 *      CPU-X                      CPU-Y
4049 	 *   (vm running)                (host running)
4050 	 *   rx posted interrupt
4051 	 *   CLEAR pending bit
4052 	 *				 SET PIR bit
4053 	 *   READ/CLEAR PIR bits
4054 	 *				 SET pending bit
4055 	 *   (vm exit)
4056 	 *   pending bit set, PIR 0
4057 	 */
4058 	if (pirval != 0) {
4059 		rvi = pirbase + flsl(pirval) - 1;
4060 		intr_status_old = vmcs_read(VMCS_GUEST_INTR_STATUS);
4061 		intr_status_new = (intr_status_old & 0xFF00) | rvi;
4062 		if (intr_status_new > intr_status_old) {
4063 			vmcs_write(VMCS_GUEST_INTR_STATUS, intr_status_new);
4064 			VLAPIC_CTR2(vlapic, "vmx_inject_pir: "
4065 			    "guest_intr_status changed from 0x%04x to 0x%04x",
4066 			    intr_status_old, intr_status_new);
4067 		}
4068 	}
4069 }
4070 
4071 static struct vlapic *
4072 vmx_vlapic_init(void *vcpui)
4073 {
4074 	struct vmx *vmx;
4075 	struct vmx_vcpu *vcpu;
4076 	struct vlapic *vlapic;
4077 	struct vlapic_vtx *vlapic_vtx;
4078 
4079 	vcpu = vcpui;
4080 	vmx = vcpu->vmx;
4081 
4082 	vlapic = malloc(sizeof(struct vlapic_vtx), M_VLAPIC, M_WAITOK | M_ZERO);
4083 	vlapic->vm = vmx->vm;
4084 	vlapic->vcpu = vcpu->vcpu;
4085 	vlapic->vcpuid = vcpu->vcpuid;
4086 	vlapic->apic_page = (struct LAPIC *)vcpu->apic_page;
4087 
4088 	vlapic_vtx = (struct vlapic_vtx *)vlapic;
4089 	vlapic_vtx->pir_desc = vcpu->pir_desc;
4090 	vlapic_vtx->vcpu = vcpu;
4091 
4092 	if (tpr_shadowing) {
4093 		vlapic->ops.enable_x2apic_mode = vmx_enable_x2apic_mode_ts;
4094 	}
4095 
4096 	if (virtual_interrupt_delivery) {
4097 		vlapic->ops.set_intr_ready = vmx_set_intr_ready;
4098 		vlapic->ops.pending_intr = vmx_pending_intr;
4099 		vlapic->ops.intr_accepted = vmx_intr_accepted;
4100 		vlapic->ops.set_tmr = vmx_set_tmr;
4101 		vlapic->ops.enable_x2apic_mode = vmx_enable_x2apic_mode_vid;
4102 	}
4103 
4104 	if (posted_interrupts)
4105 		vlapic->ops.post_intr = vmx_post_intr;
4106 
4107 	vlapic_init(vlapic);
4108 
4109 	return (vlapic);
4110 }
4111 
4112 static void
4113 vmx_vlapic_cleanup(struct vlapic *vlapic)
4114 {
4115 
4116 	vlapic_cleanup(vlapic);
4117 	free(vlapic, M_VLAPIC);
4118 }
4119 
4120 #ifdef BHYVE_SNAPSHOT
4121 static int
4122 vmx_vcpu_snapshot(void *vcpui, struct vm_snapshot_meta *meta)
4123 {
4124 	struct vmcs *vmcs;
4125 	struct vmx *vmx;
4126 	struct vmx_vcpu *vcpu;
4127 	struct vmxctx *vmxctx;
4128 	int err, run, hostcpu;
4129 
4130 	err = 0;
4131 	vcpu = vcpui;
4132 	vmx = vcpu->vmx;
4133 	vmcs = vcpu->vmcs;
4134 
4135 	run = vcpu_is_running(vcpu->vcpu, &hostcpu);
4136 	if (run && hostcpu != curcpu) {
4137 		printf("%s: %s%d is running", __func__, vm_name(vmx->vm),
4138 		    vcpu->vcpuid);
4139 		return (EINVAL);
4140 	}
4141 
4142 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR0, meta);
4143 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR3, meta);
4144 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CR4, meta);
4145 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_DR7, meta);
4146 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RSP, meta);
4147 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RIP, meta);
4148 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_RFLAGS, meta);
4149 
4150 	/* Guest segments */
4151 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_ES, meta);
4152 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_ES, meta);
4153 
4154 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_CS, meta);
4155 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_CS, meta);
4156 
4157 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_SS, meta);
4158 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_SS, meta);
4159 
4160 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_DS, meta);
4161 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_DS, meta);
4162 
4163 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_FS, meta);
4164 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_FS, meta);
4165 
4166 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_GS, meta);
4167 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_GS, meta);
4168 
4169 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_TR, meta);
4170 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_TR, meta);
4171 
4172 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_LDTR, meta);
4173 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_LDTR, meta);
4174 
4175 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_EFER, meta);
4176 
4177 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_IDTR, meta);
4178 	err += vmcs_snapshot_desc(vmcs, run, VM_REG_GUEST_GDTR, meta);
4179 
4180 	/* Guest page tables */
4181 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE0, meta);
4182 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE1, meta);
4183 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE2, meta);
4184 	err += vmcs_snapshot_reg(vmcs, run, VM_REG_GUEST_PDPTE3, meta);
4185 
4186 	/* Other guest state */
4187 	err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_CS, meta);
4188 	err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_ESP, meta);
4189 	err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_IA32_SYSENTER_EIP, meta);
4190 	err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_INTERRUPTIBILITY, meta);
4191 	err += vmcs_snapshot_any(vmcs, run, VMCS_GUEST_ACTIVITY, meta);
4192 	err += vmcs_snapshot_any(vmcs, run, VMCS_ENTRY_CTLS, meta);
4193 	err += vmcs_snapshot_any(vmcs, run, VMCS_EXIT_CTLS, meta);
4194 	if (err != 0)
4195 		goto done;
4196 
4197 	SNAPSHOT_BUF_OR_LEAVE(vcpu->guest_msrs,
4198 	    sizeof(vcpu->guest_msrs), meta, err, done);
4199 
4200 	SNAPSHOT_BUF_OR_LEAVE(vcpu->pir_desc,
4201 	    sizeof(*vcpu->pir_desc), meta, err, done);
4202 
4203 	SNAPSHOT_BUF_OR_LEAVE(&vcpu->mtrr,
4204 	    sizeof(vcpu->mtrr), meta, err, done);
4205 
4206 	vmxctx = &vcpu->ctx;
4207 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rdi, meta, err, done);
4208 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rsi, meta, err, done);
4209 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rdx, meta, err, done);
4210 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rcx, meta, err, done);
4211 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r8, meta, err, done);
4212 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r9, meta, err, done);
4213 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rax, meta, err, done);
4214 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rbx, meta, err, done);
4215 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_rbp, meta, err, done);
4216 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r10, meta, err, done);
4217 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r11, meta, err, done);
4218 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r12, meta, err, done);
4219 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r13, meta, err, done);
4220 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r14, meta, err, done);
4221 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_r15, meta, err, done);
4222 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_cr2, meta, err, done);
4223 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr0, meta, err, done);
4224 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr1, meta, err, done);
4225 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr2, meta, err, done);
4226 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr3, meta, err, done);
4227 	SNAPSHOT_VAR_OR_LEAVE(vmxctx->guest_dr6, meta, err, done);
4228 
4229 done:
4230 	return (err);
4231 }
4232 
4233 static int
4234 vmx_restore_tsc(void *vcpui, uint64_t offset)
4235 {
4236 	struct vmx_vcpu *vcpu = vcpui;
4237 	struct vmcs *vmcs;
4238 	struct vmx *vmx;
4239 	int error, running, hostcpu;
4240 
4241 	vmx = vcpu->vmx;
4242 	vmcs = vcpu->vmcs;
4243 
4244 	running = vcpu_is_running(vcpu->vcpu, &hostcpu);
4245 	if (running && hostcpu != curcpu) {
4246 		printf("%s: %s%d is running", __func__, vm_name(vmx->vm),
4247 		    vcpu->vcpuid);
4248 		return (EINVAL);
4249 	}
4250 
4251 	if (!running)
4252 		VMPTRLD(vmcs);
4253 
4254 	error = vmx_set_tsc_offset(vcpu, offset);
4255 
4256 	if (!running)
4257 		VMCLEAR(vmcs);
4258 	return (error);
4259 }
4260 #endif
4261 
4262 const struct vmm_ops vmm_ops_intel = {
4263 	.modinit	= vmx_modinit,
4264 	.modcleanup	= vmx_modcleanup,
4265 	.modresume	= vmx_modresume,
4266 	.init		= vmx_init,
4267 	.run		= vmx_run,
4268 	.cleanup	= vmx_cleanup,
4269 	.vcpu_init	= vmx_vcpu_init,
4270 	.vcpu_cleanup	= vmx_vcpu_cleanup,
4271 	.getreg		= vmx_getreg,
4272 	.setreg		= vmx_setreg,
4273 	.getdesc	= vmx_getdesc,
4274 	.setdesc	= vmx_setdesc,
4275 	.getcap		= vmx_getcap,
4276 	.setcap		= vmx_setcap,
4277 	.vmspace_alloc	= vmx_vmspace_alloc,
4278 	.vmspace_free	= vmx_vmspace_free,
4279 	.vlapic_init	= vmx_vlapic_init,
4280 	.vlapic_cleanup	= vmx_vlapic_cleanup,
4281 #ifdef BHYVE_SNAPSHOT
4282 	.vcpu_snapshot	= vmx_vcpu_snapshot,
4283 	.restore_tsc	= vmx_restore_tsc,
4284 #endif
4285 };
4286