xref: /freebsd/sys/amd64/vmm/vmm.c (revision 545ddfbe7d4fe8adfb862903b24eac1d5896c1ef)
1 /*-
2  * Copyright (c) 2011 NetApp, Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/module.h>
36 #include <sys/sysctl.h>
37 #include <sys/malloc.h>
38 #include <sys/pcpu.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/proc.h>
42 #include <sys/rwlock.h>
43 #include <sys/sched.h>
44 #include <sys/smp.h>
45 #include <sys/systm.h>
46 
47 #include <vm/vm.h>
48 #include <vm/vm_object.h>
49 #include <vm/vm_page.h>
50 #include <vm/pmap.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_extern.h>
53 #include <vm/vm_param.h>
54 
55 #include <machine/cpu.h>
56 #include <machine/vm.h>
57 #include <machine/pcb.h>
58 #include <machine/smp.h>
59 #include <x86/psl.h>
60 #include <x86/apicreg.h>
61 #include <machine/vmparam.h>
62 
63 #include <machine/vmm.h>
64 #include <machine/vmm_dev.h>
65 #include <machine/vmm_instruction_emul.h>
66 
67 #include "vmm_ioport.h"
68 #include "vmm_ktr.h"
69 #include "vmm_host.h"
70 #include "vmm_mem.h"
71 #include "vmm_util.h"
72 #include "vatpic.h"
73 #include "vatpit.h"
74 #include "vhpet.h"
75 #include "vioapic.h"
76 #include "vlapic.h"
77 #include "vpmtmr.h"
78 #include "vrtc.h"
79 #include "vmm_ipi.h"
80 #include "vmm_stat.h"
81 #include "vmm_lapic.h"
82 
83 #include "io/ppt.h"
84 #include "io/iommu.h"
85 
86 struct vlapic;
87 
88 /*
89  * Initialization:
90  * (a) allocated when vcpu is created
91  * (i) initialized when vcpu is created and when it is reinitialized
92  * (o) initialized the first time the vcpu is created
93  * (x) initialized before use
94  */
95 struct vcpu {
96 	struct mtx 	mtx;		/* (o) protects 'state' and 'hostcpu' */
97 	enum vcpu_state	state;		/* (o) vcpu state */
98 	int		hostcpu;	/* (o) vcpu's host cpu */
99 	struct vlapic	*vlapic;	/* (i) APIC device model */
100 	enum x2apic_state x2apic_state;	/* (i) APIC mode */
101 	uint64_t	exitintinfo;	/* (i) events pending at VM exit */
102 	int		nmi_pending;	/* (i) NMI pending */
103 	int		extint_pending;	/* (i) INTR pending */
104 	int	exception_pending;	/* (i) exception pending */
105 	int	exc_vector;		/* (x) exception collateral */
106 	int	exc_errcode_valid;
107 	uint32_t exc_errcode;
108 	struct savefpu	*guestfpu;	/* (a,i) guest fpu state */
109 	uint64_t	guest_xcr0;	/* (i) guest %xcr0 register */
110 	void		*stats;		/* (a,i) statistics */
111 	struct vm_exit	exitinfo;	/* (x) exit reason and collateral */
112 	uint64_t	nextrip;	/* (x) next instruction to execute */
113 };
114 
115 #define	vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
116 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
117 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
118 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
119 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
120 
121 struct mem_seg {
122 	vm_paddr_t	gpa;
123 	size_t		len;
124 	boolean_t	wired;
125 	vm_object_t	object;
126 };
127 #define	VM_MAX_MEMORY_SEGMENTS	2
128 
129 /*
130  * Initialization:
131  * (o) initialized the first time the VM is created
132  * (i) initialized when VM is created and when it is reinitialized
133  * (x) initialized before use
134  */
135 struct vm {
136 	void		*cookie;		/* (i) cpu-specific data */
137 	void		*iommu;			/* (x) iommu-specific data */
138 	struct vhpet	*vhpet;			/* (i) virtual HPET */
139 	struct vioapic	*vioapic;		/* (i) virtual ioapic */
140 	struct vatpic	*vatpic;		/* (i) virtual atpic */
141 	struct vatpit	*vatpit;		/* (i) virtual atpit */
142 	struct vpmtmr	*vpmtmr;		/* (i) virtual ACPI PM timer */
143 	struct vrtc	*vrtc;			/* (o) virtual RTC */
144 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
145 	int		suspend;		/* (i) stop VM execution */
146 	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
147 	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
148 	cpuset_t	rendezvous_req_cpus;	/* (x) rendezvous requested */
149 	cpuset_t	rendezvous_done_cpus;	/* (x) rendezvous finished */
150 	void		*rendezvous_arg;	/* (x) rendezvous func/arg */
151 	vm_rendezvous_func_t rendezvous_func;
152 	struct mtx	rendezvous_mtx;		/* (o) rendezvous lock */
153 	int		num_mem_segs;		/* (o) guest memory segments */
154 	struct mem_seg	mem_segs[VM_MAX_MEMORY_SEGMENTS];
155 	struct vmspace	*vmspace;		/* (o) guest's address space */
156 	char		name[VM_MAX_NAMELEN];	/* (o) virtual machine name */
157 	struct vcpu	vcpu[VM_MAXCPU];	/* (i) guest vcpus */
158 };
159 
160 static int vmm_initialized;
161 
162 static struct vmm_ops *ops;
163 #define	VMM_INIT(num)	(ops != NULL ? (*ops->init)(num) : 0)
164 #define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
165 #define	VMM_RESUME()	(ops != NULL ? (*ops->resume)() : 0)
166 
167 #define	VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
168 #define	VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \
169 	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO)
170 #define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
171 #define	VMSPACE_ALLOC(min, max) \
172 	(ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
173 #define	VMSPACE_FREE(vmspace) \
174 	(ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
175 #define	VMGETREG(vmi, vcpu, num, retval)		\
176 	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
177 #define	VMSETREG(vmi, vcpu, num, val)		\
178 	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
179 #define	VMGETDESC(vmi, vcpu, num, desc)		\
180 	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
181 #define	VMSETDESC(vmi, vcpu, num, desc)		\
182 	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
183 #define	VMGETCAP(vmi, vcpu, num, retval)	\
184 	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
185 #define	VMSETCAP(vmi, vcpu, num, val)		\
186 	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
187 #define	VLAPIC_INIT(vmi, vcpu)			\
188 	(ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
189 #define	VLAPIC_CLEANUP(vmi, vlapic)		\
190 	(ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
191 
192 #define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
193 #define	fpu_stop_emulating()	clts()
194 
195 static MALLOC_DEFINE(M_VM, "vm", "vm");
196 
197 /* statistics */
198 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
199 
200 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
201 
202 /*
203  * Halt the guest if all vcpus are executing a HLT instruction with
204  * interrupts disabled.
205  */
206 static int halt_detection_enabled = 1;
207 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
208     &halt_detection_enabled, 0,
209     "Halt VM if all vcpus execute HLT with interrupts disabled");
210 
211 static int vmm_ipinum;
212 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
213     "IPI vector used for vcpu notifications");
214 
215 static int trace_guest_exceptions;
216 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN,
217     &trace_guest_exceptions, 0,
218     "Trap into hypervisor on all guest exceptions and reflect them back");
219 
220 static void
221 vcpu_cleanup(struct vm *vm, int i, bool destroy)
222 {
223 	struct vcpu *vcpu = &vm->vcpu[i];
224 
225 	VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
226 	if (destroy) {
227 		vmm_stat_free(vcpu->stats);
228 		fpu_save_area_free(vcpu->guestfpu);
229 	}
230 }
231 
232 static void
233 vcpu_init(struct vm *vm, int vcpu_id, bool create)
234 {
235 	struct vcpu *vcpu;
236 
237 	KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU,
238 	    ("vcpu_init: invalid vcpu %d", vcpu_id));
239 
240 	vcpu = &vm->vcpu[vcpu_id];
241 
242 	if (create) {
243 		KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
244 		    "initialized", vcpu_id));
245 		vcpu_lock_init(vcpu);
246 		vcpu->state = VCPU_IDLE;
247 		vcpu->hostcpu = NOCPU;
248 		vcpu->guestfpu = fpu_save_area_alloc();
249 		vcpu->stats = vmm_stat_alloc();
250 	}
251 
252 	vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
253 	vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
254 	vcpu->exitintinfo = 0;
255 	vcpu->nmi_pending = 0;
256 	vcpu->extint_pending = 0;
257 	vcpu->exception_pending = 0;
258 	vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
259 	fpu_save_area_reset(vcpu->guestfpu);
260 	vmm_stat_init(vcpu->stats);
261 }
262 
263 int
264 vcpu_trace_exceptions(struct vm *vm, int vcpuid)
265 {
266 
267 	return (trace_guest_exceptions);
268 }
269 
270 struct vm_exit *
271 vm_exitinfo(struct vm *vm, int cpuid)
272 {
273 	struct vcpu *vcpu;
274 
275 	if (cpuid < 0 || cpuid >= VM_MAXCPU)
276 		panic("vm_exitinfo: invalid cpuid %d", cpuid);
277 
278 	vcpu = &vm->vcpu[cpuid];
279 
280 	return (&vcpu->exitinfo);
281 }
282 
283 static void
284 vmm_resume(void)
285 {
286 	VMM_RESUME();
287 }
288 
289 static int
290 vmm_init(void)
291 {
292 	int error;
293 
294 	vmm_host_state_init();
295 
296 	vmm_ipinum = vmm_ipi_alloc();
297 	if (vmm_ipinum == 0)
298 		vmm_ipinum = IPI_AST;
299 
300 	error = vmm_mem_init();
301 	if (error)
302 		return (error);
303 
304 	if (vmm_is_intel())
305 		ops = &vmm_ops_intel;
306 	else if (vmm_is_amd())
307 		ops = &vmm_ops_amd;
308 	else
309 		return (ENXIO);
310 
311 	vmm_resume_p = vmm_resume;
312 
313 	return (VMM_INIT(vmm_ipinum));
314 }
315 
316 static int
317 vmm_handler(module_t mod, int what, void *arg)
318 {
319 	int error;
320 
321 	switch (what) {
322 	case MOD_LOAD:
323 		vmmdev_init();
324 		if (ppt_avail_devices() > 0)
325 			iommu_init();
326 		error = vmm_init();
327 		if (error == 0)
328 			vmm_initialized = 1;
329 		break;
330 	case MOD_UNLOAD:
331 		error = vmmdev_cleanup();
332 		if (error == 0) {
333 			vmm_resume_p = NULL;
334 			iommu_cleanup();
335 			if (vmm_ipinum != IPI_AST)
336 				vmm_ipi_free(vmm_ipinum);
337 			error = VMM_CLEANUP();
338 			/*
339 			 * Something bad happened - prevent new
340 			 * VMs from being created
341 			 */
342 			if (error)
343 				vmm_initialized = 0;
344 		}
345 		break;
346 	default:
347 		error = 0;
348 		break;
349 	}
350 	return (error);
351 }
352 
353 static moduledata_t vmm_kmod = {
354 	"vmm",
355 	vmm_handler,
356 	NULL
357 };
358 
359 /*
360  * vmm initialization has the following dependencies:
361  *
362  * - iommu initialization must happen after the pci passthru driver has had
363  *   a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
364  *
365  * - VT-x initialization requires smp_rendezvous() and therefore must happen
366  *   after SMP is fully functional (after SI_SUB_SMP).
367  */
368 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
369 MODULE_VERSION(vmm, 1);
370 
371 static void
372 vm_init(struct vm *vm, bool create)
373 {
374 	int i;
375 
376 	vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
377 	vm->iommu = NULL;
378 	vm->vioapic = vioapic_init(vm);
379 	vm->vhpet = vhpet_init(vm);
380 	vm->vatpic = vatpic_init(vm);
381 	vm->vatpit = vatpit_init(vm);
382 	vm->vpmtmr = vpmtmr_init(vm);
383 	if (create)
384 		vm->vrtc = vrtc_init(vm);
385 
386 	CPU_ZERO(&vm->active_cpus);
387 
388 	vm->suspend = 0;
389 	CPU_ZERO(&vm->suspended_cpus);
390 
391 	for (i = 0; i < VM_MAXCPU; i++)
392 		vcpu_init(vm, i, create);
393 }
394 
395 int
396 vm_create(const char *name, struct vm **retvm)
397 {
398 	struct vm *vm;
399 	struct vmspace *vmspace;
400 
401 	/*
402 	 * If vmm.ko could not be successfully initialized then don't attempt
403 	 * to create the virtual machine.
404 	 */
405 	if (!vmm_initialized)
406 		return (ENXIO);
407 
408 	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
409 		return (EINVAL);
410 
411 	vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS);
412 	if (vmspace == NULL)
413 		return (ENOMEM);
414 
415 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
416 	strcpy(vm->name, name);
417 	vm->num_mem_segs = 0;
418 	vm->vmspace = vmspace;
419 	mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
420 
421 	vm_init(vm, true);
422 
423 	*retvm = vm;
424 	return (0);
425 }
426 
427 static void
428 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg)
429 {
430 
431 	if (seg->object != NULL)
432 		vmm_mem_free(vm->vmspace, seg->gpa, seg->len);
433 
434 	bzero(seg, sizeof(*seg));
435 }
436 
437 static void
438 vm_cleanup(struct vm *vm, bool destroy)
439 {
440 	int i;
441 
442 	ppt_unassign_all(vm);
443 
444 	if (vm->iommu != NULL)
445 		iommu_destroy_domain(vm->iommu);
446 
447 	if (destroy)
448 		vrtc_cleanup(vm->vrtc);
449 	else
450 		vrtc_reset(vm->vrtc);
451 	vpmtmr_cleanup(vm->vpmtmr);
452 	vatpit_cleanup(vm->vatpit);
453 	vhpet_cleanup(vm->vhpet);
454 	vatpic_cleanup(vm->vatpic);
455 	vioapic_cleanup(vm->vioapic);
456 
457 	for (i = 0; i < VM_MAXCPU; i++)
458 		vcpu_cleanup(vm, i, destroy);
459 
460 	VMCLEANUP(vm->cookie);
461 
462 	if (destroy) {
463 		for (i = 0; i < vm->num_mem_segs; i++)
464 			vm_free_mem_seg(vm, &vm->mem_segs[i]);
465 
466 		vm->num_mem_segs = 0;
467 
468 		VMSPACE_FREE(vm->vmspace);
469 		vm->vmspace = NULL;
470 	}
471 }
472 
473 void
474 vm_destroy(struct vm *vm)
475 {
476 	vm_cleanup(vm, true);
477 	free(vm, M_VM);
478 }
479 
480 int
481 vm_reinit(struct vm *vm)
482 {
483 	int error;
484 
485 	/*
486 	 * A virtual machine can be reset only if all vcpus are suspended.
487 	 */
488 	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
489 		vm_cleanup(vm, false);
490 		vm_init(vm, false);
491 		error = 0;
492 	} else {
493 		error = EBUSY;
494 	}
495 
496 	return (error);
497 }
498 
499 const char *
500 vm_name(struct vm *vm)
501 {
502 	return (vm->name);
503 }
504 
505 int
506 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
507 {
508 	vm_object_t obj;
509 
510 	if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
511 		return (ENOMEM);
512 	else
513 		return (0);
514 }
515 
516 int
517 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
518 {
519 
520 	vmm_mmio_free(vm->vmspace, gpa, len);
521 	return (0);
522 }
523 
524 boolean_t
525 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa)
526 {
527 	int i;
528 	vm_paddr_t gpabase, gpalimit;
529 
530 	for (i = 0; i < vm->num_mem_segs; i++) {
531 		gpabase = vm->mem_segs[i].gpa;
532 		gpalimit = gpabase + vm->mem_segs[i].len;
533 		if (gpa >= gpabase && gpa < gpalimit)
534 			return (TRUE);		/* 'gpa' is regular memory */
535 	}
536 
537 	if (ppt_is_mmio(vm, gpa))
538 		return (TRUE);			/* 'gpa' is pci passthru mmio */
539 
540 	return (FALSE);
541 }
542 
543 int
544 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
545 {
546 	int available, allocated;
547 	struct mem_seg *seg;
548 	vm_object_t object;
549 	vm_paddr_t g;
550 
551 	if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
552 		return (EINVAL);
553 
554 	available = allocated = 0;
555 	g = gpa;
556 	while (g < gpa + len) {
557 		if (vm_mem_allocated(vm, g))
558 			allocated++;
559 		else
560 			available++;
561 
562 		g += PAGE_SIZE;
563 	}
564 
565 	/*
566 	 * If there are some allocated and some available pages in the address
567 	 * range then it is an error.
568 	 */
569 	if (allocated && available)
570 		return (EINVAL);
571 
572 	/*
573 	 * If the entire address range being requested has already been
574 	 * allocated then there isn't anything more to do.
575 	 */
576 	if (allocated && available == 0)
577 		return (0);
578 
579 	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
580 		return (E2BIG);
581 
582 	seg = &vm->mem_segs[vm->num_mem_segs];
583 
584 	if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL)
585 		return (ENOMEM);
586 
587 	seg->gpa = gpa;
588 	seg->len = len;
589 	seg->object = object;
590 	seg->wired = FALSE;
591 
592 	vm->num_mem_segs++;
593 
594 	return (0);
595 }
596 
597 static vm_paddr_t
598 vm_maxmem(struct vm *vm)
599 {
600 	int i;
601 	vm_paddr_t gpa, maxmem;
602 
603 	maxmem = 0;
604 	for (i = 0; i < vm->num_mem_segs; i++) {
605 		gpa = vm->mem_segs[i].gpa + vm->mem_segs[i].len;
606 		if (gpa > maxmem)
607 			maxmem = gpa;
608 	}
609 	return (maxmem);
610 }
611 
612 static void
613 vm_gpa_unwire(struct vm *vm)
614 {
615 	int i, rv;
616 	struct mem_seg *seg;
617 
618 	for (i = 0; i < vm->num_mem_segs; i++) {
619 		seg = &vm->mem_segs[i];
620 		if (!seg->wired)
621 			continue;
622 
623 		rv = vm_map_unwire(&vm->vmspace->vm_map,
624 				   seg->gpa, seg->gpa + seg->len,
625 				   VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
626 		KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment "
627 		    "%#lx/%ld could not be unwired: %d",
628 		    vm_name(vm), seg->gpa, seg->len, rv));
629 
630 		seg->wired = FALSE;
631 	}
632 }
633 
634 static int
635 vm_gpa_wire(struct vm *vm)
636 {
637 	int i, rv;
638 	struct mem_seg *seg;
639 
640 	for (i = 0; i < vm->num_mem_segs; i++) {
641 		seg = &vm->mem_segs[i];
642 		if (seg->wired)
643 			continue;
644 
645 		/* XXX rlimits? */
646 		rv = vm_map_wire(&vm->vmspace->vm_map,
647 				 seg->gpa, seg->gpa + seg->len,
648 				 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
649 		if (rv != KERN_SUCCESS)
650 			break;
651 
652 		seg->wired = TRUE;
653 	}
654 
655 	if (i < vm->num_mem_segs) {
656 		/*
657 		 * Undo the wiring before returning an error.
658 		 */
659 		vm_gpa_unwire(vm);
660 		return (EAGAIN);
661 	}
662 
663 	return (0);
664 }
665 
666 static void
667 vm_iommu_modify(struct vm *vm, boolean_t map)
668 {
669 	int i, sz;
670 	vm_paddr_t gpa, hpa;
671 	struct mem_seg *seg;
672 	void *vp, *cookie, *host_domain;
673 
674 	sz = PAGE_SIZE;
675 	host_domain = iommu_host_domain();
676 
677 	for (i = 0; i < vm->num_mem_segs; i++) {
678 		seg = &vm->mem_segs[i];
679 		KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired",
680 		    vm_name(vm), seg->gpa, seg->len));
681 
682 		gpa = seg->gpa;
683 		while (gpa < seg->gpa + seg->len) {
684 			vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE,
685 					 &cookie);
686 			KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
687 			    vm_name(vm), gpa));
688 
689 			vm_gpa_release(cookie);
690 
691 			hpa = DMAP_TO_PHYS((uintptr_t)vp);
692 			if (map) {
693 				iommu_create_mapping(vm->iommu, gpa, hpa, sz);
694 				iommu_remove_mapping(host_domain, hpa, sz);
695 			} else {
696 				iommu_remove_mapping(vm->iommu, gpa, sz);
697 				iommu_create_mapping(host_domain, hpa, hpa, sz);
698 			}
699 
700 			gpa += PAGE_SIZE;
701 		}
702 	}
703 
704 	/*
705 	 * Invalidate the cached translations associated with the domain
706 	 * from which pages were removed.
707 	 */
708 	if (map)
709 		iommu_invalidate_tlb(host_domain);
710 	else
711 		iommu_invalidate_tlb(vm->iommu);
712 }
713 
714 #define	vm_iommu_unmap(vm)	vm_iommu_modify((vm), FALSE)
715 #define	vm_iommu_map(vm)	vm_iommu_modify((vm), TRUE)
716 
717 int
718 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
719 {
720 	int error;
721 
722 	error = ppt_unassign_device(vm, bus, slot, func);
723 	if (error)
724 		return (error);
725 
726 	if (ppt_assigned_devices(vm) == 0) {
727 		vm_iommu_unmap(vm);
728 		vm_gpa_unwire(vm);
729 	}
730 	return (0);
731 }
732 
733 int
734 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
735 {
736 	int error;
737 	vm_paddr_t maxaddr;
738 
739 	/*
740 	 * Virtual machines with pci passthru devices get special treatment:
741 	 * - the guest physical memory is wired
742 	 * - the iommu is programmed to do the 'gpa' to 'hpa' translation
743 	 *
744 	 * We need to do this before the first pci passthru device is attached.
745 	 */
746 	if (ppt_assigned_devices(vm) == 0) {
747 		KASSERT(vm->iommu == NULL,
748 		    ("vm_assign_pptdev: iommu must be NULL"));
749 		maxaddr = vm_maxmem(vm);
750 		vm->iommu = iommu_create_domain(maxaddr);
751 
752 		error = vm_gpa_wire(vm);
753 		if (error)
754 			return (error);
755 
756 		vm_iommu_map(vm);
757 	}
758 
759 	error = ppt_assign_device(vm, bus, slot, func);
760 	return (error);
761 }
762 
763 void *
764 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
765 	    void **cookie)
766 {
767 	int count, pageoff;
768 	vm_page_t m;
769 
770 	pageoff = gpa & PAGE_MASK;
771 	if (len > PAGE_SIZE - pageoff)
772 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
773 
774 	count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
775 	    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
776 
777 	if (count == 1) {
778 		*cookie = m;
779 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
780 	} else {
781 		*cookie = NULL;
782 		return (NULL);
783 	}
784 }
785 
786 void
787 vm_gpa_release(void *cookie)
788 {
789 	vm_page_t m = cookie;
790 
791 	vm_page_lock(m);
792 	vm_page_unhold(m);
793 	vm_page_unlock(m);
794 }
795 
796 int
797 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
798 		  struct vm_memory_segment *seg)
799 {
800 	int i;
801 
802 	for (i = 0; i < vm->num_mem_segs; i++) {
803 		if (gpabase == vm->mem_segs[i].gpa) {
804 			seg->gpa = vm->mem_segs[i].gpa;
805 			seg->len = vm->mem_segs[i].len;
806 			seg->wired = vm->mem_segs[i].wired;
807 			return (0);
808 		}
809 	}
810 	return (-1);
811 }
812 
813 int
814 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len,
815 	      vm_offset_t *offset, struct vm_object **object)
816 {
817 	int i;
818 	size_t seg_len;
819 	vm_paddr_t seg_gpa;
820 	vm_object_t seg_obj;
821 
822 	for (i = 0; i < vm->num_mem_segs; i++) {
823 		if ((seg_obj = vm->mem_segs[i].object) == NULL)
824 			continue;
825 
826 		seg_gpa = vm->mem_segs[i].gpa;
827 		seg_len = vm->mem_segs[i].len;
828 
829 		if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) {
830 			*offset = gpa - seg_gpa;
831 			*object = seg_obj;
832 			vm_object_reference(seg_obj);
833 			return (0);
834 		}
835 	}
836 
837 	return (EINVAL);
838 }
839 
840 int
841 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
842 {
843 
844 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
845 		return (EINVAL);
846 
847 	if (reg >= VM_REG_LAST)
848 		return (EINVAL);
849 
850 	return (VMGETREG(vm->cookie, vcpu, reg, retval));
851 }
852 
853 int
854 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val)
855 {
856 	struct vcpu *vcpu;
857 	int error;
858 
859 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
860 		return (EINVAL);
861 
862 	if (reg >= VM_REG_LAST)
863 		return (EINVAL);
864 
865 	error = VMSETREG(vm->cookie, vcpuid, reg, val);
866 	if (error || reg != VM_REG_GUEST_RIP)
867 		return (error);
868 
869 	/* Set 'nextrip' to match the value of %rip */
870 	VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val);
871 	vcpu = &vm->vcpu[vcpuid];
872 	vcpu->nextrip = val;
873 	return (0);
874 }
875 
876 static boolean_t
877 is_descriptor_table(int reg)
878 {
879 
880 	switch (reg) {
881 	case VM_REG_GUEST_IDTR:
882 	case VM_REG_GUEST_GDTR:
883 		return (TRUE);
884 	default:
885 		return (FALSE);
886 	}
887 }
888 
889 static boolean_t
890 is_segment_register(int reg)
891 {
892 
893 	switch (reg) {
894 	case VM_REG_GUEST_ES:
895 	case VM_REG_GUEST_CS:
896 	case VM_REG_GUEST_SS:
897 	case VM_REG_GUEST_DS:
898 	case VM_REG_GUEST_FS:
899 	case VM_REG_GUEST_GS:
900 	case VM_REG_GUEST_TR:
901 	case VM_REG_GUEST_LDTR:
902 		return (TRUE);
903 	default:
904 		return (FALSE);
905 	}
906 }
907 
908 int
909 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
910 		struct seg_desc *desc)
911 {
912 
913 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
914 		return (EINVAL);
915 
916 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
917 		return (EINVAL);
918 
919 	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
920 }
921 
922 int
923 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
924 		struct seg_desc *desc)
925 {
926 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
927 		return (EINVAL);
928 
929 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
930 		return (EINVAL);
931 
932 	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
933 }
934 
935 static void
936 restore_guest_fpustate(struct vcpu *vcpu)
937 {
938 
939 	/* flush host state to the pcb */
940 	fpuexit(curthread);
941 
942 	/* restore guest FPU state */
943 	fpu_stop_emulating();
944 	fpurestore(vcpu->guestfpu);
945 
946 	/* restore guest XCR0 if XSAVE is enabled in the host */
947 	if (rcr4() & CR4_XSAVE)
948 		load_xcr(0, vcpu->guest_xcr0);
949 
950 	/*
951 	 * The FPU is now "dirty" with the guest's state so turn on emulation
952 	 * to trap any access to the FPU by the host.
953 	 */
954 	fpu_start_emulating();
955 }
956 
957 static void
958 save_guest_fpustate(struct vcpu *vcpu)
959 {
960 
961 	if ((rcr0() & CR0_TS) == 0)
962 		panic("fpu emulation not enabled in host!");
963 
964 	/* save guest XCR0 and restore host XCR0 */
965 	if (rcr4() & CR4_XSAVE) {
966 		vcpu->guest_xcr0 = rxcr(0);
967 		load_xcr(0, vmm_get_host_xcr0());
968 	}
969 
970 	/* save guest FPU state */
971 	fpu_stop_emulating();
972 	fpusave(vcpu->guestfpu);
973 	fpu_start_emulating();
974 }
975 
976 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
977 
978 static int
979 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
980     bool from_idle)
981 {
982 	int error;
983 
984 	vcpu_assert_locked(vcpu);
985 
986 	/*
987 	 * State transitions from the vmmdev_ioctl() must always begin from
988 	 * the VCPU_IDLE state. This guarantees that there is only a single
989 	 * ioctl() operating on a vcpu at any point.
990 	 */
991 	if (from_idle) {
992 		while (vcpu->state != VCPU_IDLE)
993 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
994 	} else {
995 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
996 		    "vcpu idle state"));
997 	}
998 
999 	if (vcpu->state == VCPU_RUNNING) {
1000 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1001 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1002 	} else {
1003 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1004 		    "vcpu that is not running", vcpu->hostcpu));
1005 	}
1006 
1007 	/*
1008 	 * The following state transitions are allowed:
1009 	 * IDLE -> FROZEN -> IDLE
1010 	 * FROZEN -> RUNNING -> FROZEN
1011 	 * FROZEN -> SLEEPING -> FROZEN
1012 	 */
1013 	switch (vcpu->state) {
1014 	case VCPU_IDLE:
1015 	case VCPU_RUNNING:
1016 	case VCPU_SLEEPING:
1017 		error = (newstate != VCPU_FROZEN);
1018 		break;
1019 	case VCPU_FROZEN:
1020 		error = (newstate == VCPU_FROZEN);
1021 		break;
1022 	default:
1023 		error = 1;
1024 		break;
1025 	}
1026 
1027 	if (error)
1028 		return (EBUSY);
1029 
1030 	vcpu->state = newstate;
1031 	if (newstate == VCPU_RUNNING)
1032 		vcpu->hostcpu = curcpu;
1033 	else
1034 		vcpu->hostcpu = NOCPU;
1035 
1036 	if (newstate == VCPU_IDLE)
1037 		wakeup(&vcpu->state);
1038 
1039 	return (0);
1040 }
1041 
1042 static void
1043 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1044 {
1045 	int error;
1046 
1047 	if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1048 		panic("Error %d setting state to %d\n", error, newstate);
1049 }
1050 
1051 static void
1052 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
1053 {
1054 	int error;
1055 
1056 	if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
1057 		panic("Error %d setting state to %d", error, newstate);
1058 }
1059 
1060 static void
1061 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func)
1062 {
1063 
1064 	KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked"));
1065 
1066 	/*
1067 	 * Update 'rendezvous_func' and execute a write memory barrier to
1068 	 * ensure that it is visible across all host cpus. This is not needed
1069 	 * for correctness but it does ensure that all the vcpus will notice
1070 	 * that the rendezvous is requested immediately.
1071 	 */
1072 	vm->rendezvous_func = func;
1073 	wmb();
1074 }
1075 
1076 #define	RENDEZVOUS_CTR0(vm, vcpuid, fmt)				\
1077 	do {								\
1078 		if (vcpuid >= 0)					\
1079 			VCPU_CTR0(vm, vcpuid, fmt);			\
1080 		else							\
1081 			VM_CTR0(vm, fmt);				\
1082 	} while (0)
1083 
1084 static void
1085 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1086 {
1087 
1088 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
1089 	    ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1090 
1091 	mtx_lock(&vm->rendezvous_mtx);
1092 	while (vm->rendezvous_func != NULL) {
1093 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1094 		CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1095 
1096 		if (vcpuid != -1 &&
1097 		    CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1098 		    !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1099 			VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1100 			(*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1101 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1102 		}
1103 		if (CPU_CMP(&vm->rendezvous_req_cpus,
1104 		    &vm->rendezvous_done_cpus) == 0) {
1105 			VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1106 			vm_set_rendezvous_func(vm, NULL);
1107 			wakeup(&vm->rendezvous_func);
1108 			break;
1109 		}
1110 		RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1111 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1112 		    "vmrndv", 0);
1113 	}
1114 	mtx_unlock(&vm->rendezvous_mtx);
1115 }
1116 
1117 /*
1118  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1119  */
1120 static int
1121 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1122 {
1123 	struct vcpu *vcpu;
1124 	const char *wmesg;
1125 	int t, vcpu_halted, vm_halted;
1126 
1127 	KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1128 
1129 	vcpu = &vm->vcpu[vcpuid];
1130 	vcpu_halted = 0;
1131 	vm_halted = 0;
1132 
1133 	vcpu_lock(vcpu);
1134 	while (1) {
1135 		/*
1136 		 * Do a final check for pending NMI or interrupts before
1137 		 * really putting this thread to sleep. Also check for
1138 		 * software events that would cause this vcpu to wakeup.
1139 		 *
1140 		 * These interrupts/events could have happened after the
1141 		 * vcpu returned from VMRUN() and before it acquired the
1142 		 * vcpu lock above.
1143 		 */
1144 		if (vm->rendezvous_func != NULL || vm->suspend)
1145 			break;
1146 		if (vm_nmi_pending(vm, vcpuid))
1147 			break;
1148 		if (!intr_disabled) {
1149 			if (vm_extint_pending(vm, vcpuid) ||
1150 			    vlapic_pending_intr(vcpu->vlapic, NULL)) {
1151 				break;
1152 			}
1153 		}
1154 
1155 		/* Don't go to sleep if the vcpu thread needs to yield */
1156 		if (vcpu_should_yield(vm, vcpuid))
1157 			break;
1158 
1159 		/*
1160 		 * Some Linux guests implement "halt" by having all vcpus
1161 		 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1162 		 * track of the vcpus that have entered this state. When all
1163 		 * vcpus enter the halted state the virtual machine is halted.
1164 		 */
1165 		if (intr_disabled) {
1166 			wmesg = "vmhalt";
1167 			VCPU_CTR0(vm, vcpuid, "Halted");
1168 			if (!vcpu_halted && halt_detection_enabled) {
1169 				vcpu_halted = 1;
1170 				CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1171 			}
1172 			if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1173 				vm_halted = 1;
1174 				break;
1175 			}
1176 		} else {
1177 			wmesg = "vmidle";
1178 		}
1179 
1180 		t = ticks;
1181 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1182 		/*
1183 		 * XXX msleep_spin() cannot be interrupted by signals so
1184 		 * wake up periodically to check pending signals.
1185 		 */
1186 		msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1187 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1188 		vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1189 	}
1190 
1191 	if (vcpu_halted)
1192 		CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1193 
1194 	vcpu_unlock(vcpu);
1195 
1196 	if (vm_halted)
1197 		vm_suspend(vm, VM_SUSPEND_HALT);
1198 
1199 	return (0);
1200 }
1201 
1202 static int
1203 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1204 {
1205 	int rv, ftype;
1206 	struct vm_map *map;
1207 	struct vcpu *vcpu;
1208 	struct vm_exit *vme;
1209 
1210 	vcpu = &vm->vcpu[vcpuid];
1211 	vme = &vcpu->exitinfo;
1212 
1213 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1214 	    __func__, vme->inst_length));
1215 
1216 	ftype = vme->u.paging.fault_type;
1217 	KASSERT(ftype == VM_PROT_READ ||
1218 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1219 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1220 
1221 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1222 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1223 		    vme->u.paging.gpa, ftype);
1224 		if (rv == 0) {
1225 			VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1226 			    ftype == VM_PROT_READ ? "accessed" : "dirty",
1227 			    vme->u.paging.gpa);
1228 			goto done;
1229 		}
1230 	}
1231 
1232 	map = &vm->vmspace->vm_map;
1233 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
1234 
1235 	VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1236 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1237 
1238 	if (rv != KERN_SUCCESS)
1239 		return (EFAULT);
1240 done:
1241 	return (0);
1242 }
1243 
1244 static int
1245 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1246 {
1247 	struct vie *vie;
1248 	struct vcpu *vcpu;
1249 	struct vm_exit *vme;
1250 	uint64_t gla, gpa;
1251 	struct vm_guest_paging *paging;
1252 	mem_region_read_t mread;
1253 	mem_region_write_t mwrite;
1254 	enum vm_cpu_mode cpu_mode;
1255 	int cs_d, error, length;
1256 
1257 	vcpu = &vm->vcpu[vcpuid];
1258 	vme = &vcpu->exitinfo;
1259 
1260 	gla = vme->u.inst_emul.gla;
1261 	gpa = vme->u.inst_emul.gpa;
1262 	cs_d = vme->u.inst_emul.cs_d;
1263 	vie = &vme->u.inst_emul.vie;
1264 	paging = &vme->u.inst_emul.paging;
1265 	cpu_mode = paging->cpu_mode;
1266 
1267 	VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1268 
1269 	/* Fetch, decode and emulate the faulting instruction */
1270 	if (vie->num_valid == 0) {
1271 		/*
1272 		 * If the instruction length is not known then assume a
1273 		 * maximum size instruction.
1274 		 */
1275 		length = vme->inst_length ? vme->inst_length : VIE_INST_SIZE;
1276 		error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip,
1277 		    length, vie);
1278 	} else {
1279 		/*
1280 		 * The instruction bytes have already been copied into 'vie'
1281 		 */
1282 		error = 0;
1283 	}
1284 	if (error == 1)
1285 		return (0);		/* Resume guest to handle page fault */
1286 	else if (error == -1)
1287 		return (EFAULT);
1288 	else if (error != 0)
1289 		panic("%s: vmm_fetch_instruction error %d", __func__, error);
1290 
1291 	if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0)
1292 		return (EFAULT);
1293 
1294 	/*
1295 	 * If the instruction length was not specified then update it now
1296 	 * along with 'nextrip'.
1297 	 */
1298 	if (vme->inst_length == 0) {
1299 		vme->inst_length = vie->num_processed;
1300 		vcpu->nextrip += vie->num_processed;
1301 	}
1302 
1303 	/* return to userland unless this is an in-kernel emulated device */
1304 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1305 		mread = lapic_mmio_read;
1306 		mwrite = lapic_mmio_write;
1307 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1308 		mread = vioapic_mmio_read;
1309 		mwrite = vioapic_mmio_write;
1310 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1311 		mread = vhpet_mmio_read;
1312 		mwrite = vhpet_mmio_write;
1313 	} else {
1314 		*retu = true;
1315 		return (0);
1316 	}
1317 
1318 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1319 	    mread, mwrite, retu);
1320 
1321 	return (error);
1322 }
1323 
1324 static int
1325 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1326 {
1327 	int i, done;
1328 	struct vcpu *vcpu;
1329 
1330 	done = 0;
1331 	vcpu = &vm->vcpu[vcpuid];
1332 
1333 	CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1334 
1335 	/*
1336 	 * Wait until all 'active_cpus' have suspended themselves.
1337 	 *
1338 	 * Since a VM may be suspended at any time including when one or
1339 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1340 	 * handler while we are waiting to prevent a deadlock.
1341 	 */
1342 	vcpu_lock(vcpu);
1343 	while (1) {
1344 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1345 			VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1346 			break;
1347 		}
1348 
1349 		if (vm->rendezvous_func == NULL) {
1350 			VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1351 			vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1352 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1353 			vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1354 		} else {
1355 			VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1356 			vcpu_unlock(vcpu);
1357 			vm_handle_rendezvous(vm, vcpuid);
1358 			vcpu_lock(vcpu);
1359 		}
1360 	}
1361 	vcpu_unlock(vcpu);
1362 
1363 	/*
1364 	 * Wakeup the other sleeping vcpus and return to userspace.
1365 	 */
1366 	for (i = 0; i < VM_MAXCPU; i++) {
1367 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1368 			vcpu_notify_event(vm, i, false);
1369 		}
1370 	}
1371 
1372 	*retu = true;
1373 	return (0);
1374 }
1375 
1376 int
1377 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1378 {
1379 	int i;
1380 
1381 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1382 		return (EINVAL);
1383 
1384 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1385 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1386 		    vm->suspend, how);
1387 		return (EALREADY);
1388 	}
1389 
1390 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1391 
1392 	/*
1393 	 * Notify all active vcpus that they are now suspended.
1394 	 */
1395 	for (i = 0; i < VM_MAXCPU; i++) {
1396 		if (CPU_ISSET(i, &vm->active_cpus))
1397 			vcpu_notify_event(vm, i, false);
1398 	}
1399 
1400 	return (0);
1401 }
1402 
1403 void
1404 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1405 {
1406 	struct vm_exit *vmexit;
1407 
1408 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1409 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1410 
1411 	vmexit = vm_exitinfo(vm, vcpuid);
1412 	vmexit->rip = rip;
1413 	vmexit->inst_length = 0;
1414 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1415 	vmexit->u.suspended.how = vm->suspend;
1416 }
1417 
1418 void
1419 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1420 {
1421 	struct vm_exit *vmexit;
1422 
1423 	KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1424 
1425 	vmexit = vm_exitinfo(vm, vcpuid);
1426 	vmexit->rip = rip;
1427 	vmexit->inst_length = 0;
1428 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1429 	vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1430 }
1431 
1432 void
1433 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1434 {
1435 	struct vm_exit *vmexit;
1436 
1437 	vmexit = vm_exitinfo(vm, vcpuid);
1438 	vmexit->rip = rip;
1439 	vmexit->inst_length = 0;
1440 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1441 	vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1442 }
1443 
1444 int
1445 vm_run(struct vm *vm, struct vm_run *vmrun)
1446 {
1447 	int error, vcpuid;
1448 	struct vcpu *vcpu;
1449 	struct pcb *pcb;
1450 	uint64_t tscval;
1451 	struct vm_exit *vme;
1452 	bool retu, intr_disabled;
1453 	pmap_t pmap;
1454 	void *rptr, *sptr;
1455 
1456 	vcpuid = vmrun->cpuid;
1457 
1458 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1459 		return (EINVAL);
1460 
1461 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1462 		return (EINVAL);
1463 
1464 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1465 		return (EINVAL);
1466 
1467 	rptr = &vm->rendezvous_func;
1468 	sptr = &vm->suspend;
1469 	pmap = vmspace_pmap(vm->vmspace);
1470 	vcpu = &vm->vcpu[vcpuid];
1471 	vme = &vcpu->exitinfo;
1472 restart:
1473 	critical_enter();
1474 
1475 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1476 	    ("vm_run: absurd pm_active"));
1477 
1478 	tscval = rdtsc();
1479 
1480 	pcb = PCPU_GET(curpcb);
1481 	set_pcb_flags(pcb, PCB_FULL_IRET);
1482 
1483 	restore_guest_fpustate(vcpu);
1484 
1485 	vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1486 	error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, rptr, sptr);
1487 	vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1488 
1489 	save_guest_fpustate(vcpu);
1490 
1491 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1492 
1493 	critical_exit();
1494 
1495 	if (error == 0) {
1496 		retu = false;
1497 		vcpu->nextrip = vme->rip + vme->inst_length;
1498 		switch (vme->exitcode) {
1499 		case VM_EXITCODE_SUSPENDED:
1500 			error = vm_handle_suspend(vm, vcpuid, &retu);
1501 			break;
1502 		case VM_EXITCODE_IOAPIC_EOI:
1503 			vioapic_process_eoi(vm, vcpuid,
1504 			    vme->u.ioapic_eoi.vector);
1505 			break;
1506 		case VM_EXITCODE_RENDEZVOUS:
1507 			vm_handle_rendezvous(vm, vcpuid);
1508 			error = 0;
1509 			break;
1510 		case VM_EXITCODE_HLT:
1511 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1512 			error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1513 			break;
1514 		case VM_EXITCODE_PAGING:
1515 			error = vm_handle_paging(vm, vcpuid, &retu);
1516 			break;
1517 		case VM_EXITCODE_INST_EMUL:
1518 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1519 			break;
1520 		case VM_EXITCODE_INOUT:
1521 		case VM_EXITCODE_INOUT_STR:
1522 			error = vm_handle_inout(vm, vcpuid, vme, &retu);
1523 			break;
1524 		case VM_EXITCODE_MONITOR:
1525 		case VM_EXITCODE_MWAIT:
1526 			vm_inject_ud(vm, vcpuid);
1527 			break;
1528 		default:
1529 			retu = true;	/* handled in userland */
1530 			break;
1531 		}
1532 	}
1533 
1534 	if (error == 0 && retu == false)
1535 		goto restart;
1536 
1537 	/* copy the exit information */
1538 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1539 	return (error);
1540 }
1541 
1542 int
1543 vm_restart_instruction(void *arg, int vcpuid)
1544 {
1545 	struct vm *vm;
1546 	struct vcpu *vcpu;
1547 	enum vcpu_state state;
1548 	uint64_t rip;
1549 	int error;
1550 
1551 	vm = arg;
1552 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1553 		return (EINVAL);
1554 
1555 	vcpu = &vm->vcpu[vcpuid];
1556 	state = vcpu_get_state(vm, vcpuid, NULL);
1557 	if (state == VCPU_RUNNING) {
1558 		/*
1559 		 * When a vcpu is "running" the next instruction is determined
1560 		 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
1561 		 * Thus setting 'inst_length' to zero will cause the current
1562 		 * instruction to be restarted.
1563 		 */
1564 		vcpu->exitinfo.inst_length = 0;
1565 		VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by "
1566 		    "setting inst_length to zero", vcpu->exitinfo.rip);
1567 	} else if (state == VCPU_FROZEN) {
1568 		/*
1569 		 * When a vcpu is "frozen" it is outside the critical section
1570 		 * around VMRUN() and 'nextrip' points to the next instruction.
1571 		 * Thus instruction restart is achieved by setting 'nextrip'
1572 		 * to the vcpu's %rip.
1573 		 */
1574 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip);
1575 		KASSERT(!error, ("%s: error %d getting rip", __func__, error));
1576 		VCPU_CTR2(vm, vcpuid, "restarting instruction by updating "
1577 		    "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
1578 		vcpu->nextrip = rip;
1579 	} else {
1580 		panic("%s: invalid state %d", __func__, state);
1581 	}
1582 	return (0);
1583 }
1584 
1585 int
1586 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1587 {
1588 	struct vcpu *vcpu;
1589 	int type, vector;
1590 
1591 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1592 		return (EINVAL);
1593 
1594 	vcpu = &vm->vcpu[vcpuid];
1595 
1596 	if (info & VM_INTINFO_VALID) {
1597 		type = info & VM_INTINFO_TYPE;
1598 		vector = info & 0xff;
1599 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1600 			return (EINVAL);
1601 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1602 			return (EINVAL);
1603 		if (info & VM_INTINFO_RSVD)
1604 			return (EINVAL);
1605 	} else {
1606 		info = 0;
1607 	}
1608 	VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1609 	vcpu->exitintinfo = info;
1610 	return (0);
1611 }
1612 
1613 enum exc_class {
1614 	EXC_BENIGN,
1615 	EXC_CONTRIBUTORY,
1616 	EXC_PAGEFAULT
1617 };
1618 
1619 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
1620 
1621 static enum exc_class
1622 exception_class(uint64_t info)
1623 {
1624 	int type, vector;
1625 
1626 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1627 	type = info & VM_INTINFO_TYPE;
1628 	vector = info & 0xff;
1629 
1630 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1631 	switch (type) {
1632 	case VM_INTINFO_HWINTR:
1633 	case VM_INTINFO_SWINTR:
1634 	case VM_INTINFO_NMI:
1635 		return (EXC_BENIGN);
1636 	default:
1637 		/*
1638 		 * Hardware exception.
1639 		 *
1640 		 * SVM and VT-x use identical type values to represent NMI,
1641 		 * hardware interrupt and software interrupt.
1642 		 *
1643 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1644 		 * for exceptions except #BP and #OF. #BP and #OF use a type
1645 		 * value of '5' or '6'. Therefore we don't check for explicit
1646 		 * values of 'type' to classify 'intinfo' into a hardware
1647 		 * exception.
1648 		 */
1649 		break;
1650 	}
1651 
1652 	switch (vector) {
1653 	case IDT_PF:
1654 	case IDT_VE:
1655 		return (EXC_PAGEFAULT);
1656 	case IDT_DE:
1657 	case IDT_TS:
1658 	case IDT_NP:
1659 	case IDT_SS:
1660 	case IDT_GP:
1661 		return (EXC_CONTRIBUTORY);
1662 	default:
1663 		return (EXC_BENIGN);
1664 	}
1665 }
1666 
1667 static int
1668 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1669     uint64_t *retinfo)
1670 {
1671 	enum exc_class exc1, exc2;
1672 	int type1, vector1;
1673 
1674 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1675 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1676 
1677 	/*
1678 	 * If an exception occurs while attempting to call the double-fault
1679 	 * handler the processor enters shutdown mode (aka triple fault).
1680 	 */
1681 	type1 = info1 & VM_INTINFO_TYPE;
1682 	vector1 = info1 & 0xff;
1683 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1684 		VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1685 		    info1, info2);
1686 		vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1687 		*retinfo = 0;
1688 		return (0);
1689 	}
1690 
1691 	/*
1692 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1693 	 */
1694 	exc1 = exception_class(info1);
1695 	exc2 = exception_class(info2);
1696 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1697 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1698 		/* Convert nested fault into a double fault. */
1699 		*retinfo = IDT_DF;
1700 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1701 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
1702 	} else {
1703 		/* Handle exceptions serially */
1704 		*retinfo = info2;
1705 	}
1706 	return (1);
1707 }
1708 
1709 static uint64_t
1710 vcpu_exception_intinfo(struct vcpu *vcpu)
1711 {
1712 	uint64_t info = 0;
1713 
1714 	if (vcpu->exception_pending) {
1715 		info = vcpu->exc_vector & 0xff;
1716 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1717 		if (vcpu->exc_errcode_valid) {
1718 			info |= VM_INTINFO_DEL_ERRCODE;
1719 			info |= (uint64_t)vcpu->exc_errcode << 32;
1720 		}
1721 	}
1722 	return (info);
1723 }
1724 
1725 int
1726 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1727 {
1728 	struct vcpu *vcpu;
1729 	uint64_t info1, info2;
1730 	int valid;
1731 
1732 	KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid));
1733 
1734 	vcpu = &vm->vcpu[vcpuid];
1735 
1736 	info1 = vcpu->exitintinfo;
1737 	vcpu->exitintinfo = 0;
1738 
1739 	info2 = 0;
1740 	if (vcpu->exception_pending) {
1741 		info2 = vcpu_exception_intinfo(vcpu);
1742 		vcpu->exception_pending = 0;
1743 		VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1744 		    vcpu->exc_vector, info2);
1745 	}
1746 
1747 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
1748 		valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
1749 	} else if (info1 & VM_INTINFO_VALID) {
1750 		*retinfo = info1;
1751 		valid = 1;
1752 	} else if (info2 & VM_INTINFO_VALID) {
1753 		*retinfo = info2;
1754 		valid = 1;
1755 	} else {
1756 		valid = 0;
1757 	}
1758 
1759 	if (valid) {
1760 		VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
1761 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
1762 	}
1763 
1764 	return (valid);
1765 }
1766 
1767 int
1768 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
1769 {
1770 	struct vcpu *vcpu;
1771 
1772 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1773 		return (EINVAL);
1774 
1775 	vcpu = &vm->vcpu[vcpuid];
1776 	*info1 = vcpu->exitintinfo;
1777 	*info2 = vcpu_exception_intinfo(vcpu);
1778 	return (0);
1779 }
1780 
1781 int
1782 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid,
1783     uint32_t errcode, int restart_instruction)
1784 {
1785 	struct vcpu *vcpu;
1786 	int error;
1787 
1788 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1789 		return (EINVAL);
1790 
1791 	if (vector < 0 || vector >= 32)
1792 		return (EINVAL);
1793 
1794 	/*
1795 	 * A double fault exception should never be injected directly into
1796 	 * the guest. It is a derived exception that results from specific
1797 	 * combinations of nested faults.
1798 	 */
1799 	if (vector == IDT_DF)
1800 		return (EINVAL);
1801 
1802 	vcpu = &vm->vcpu[vcpuid];
1803 
1804 	if (vcpu->exception_pending) {
1805 		VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
1806 		    "pending exception %d", vector, vcpu->exc_vector);
1807 		return (EBUSY);
1808 	}
1809 
1810 	/*
1811 	 * From section 26.6.1 "Interruptibility State" in Intel SDM:
1812 	 *
1813 	 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
1814 	 * one instruction or incurs an exception.
1815 	 */
1816 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
1817 	KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
1818 	    __func__, error));
1819 
1820 	if (restart_instruction)
1821 		vm_restart_instruction(vm, vcpuid);
1822 
1823 	vcpu->exception_pending = 1;
1824 	vcpu->exc_vector = vector;
1825 	vcpu->exc_errcode = errcode;
1826 	vcpu->exc_errcode_valid = errcode_valid;
1827 	VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector);
1828 	return (0);
1829 }
1830 
1831 void
1832 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
1833     int errcode)
1834 {
1835 	struct vm *vm;
1836 	int error, restart_instruction;
1837 
1838 	vm = vmarg;
1839 	restart_instruction = 1;
1840 
1841 	error = vm_inject_exception(vm, vcpuid, vector, errcode_valid,
1842 	    errcode, restart_instruction);
1843 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
1844 }
1845 
1846 void
1847 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
1848 {
1849 	struct vm *vm;
1850 	int error;
1851 
1852 	vm = vmarg;
1853 	VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
1854 	    error_code, cr2);
1855 
1856 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
1857 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
1858 
1859 	vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
1860 }
1861 
1862 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
1863 
1864 int
1865 vm_inject_nmi(struct vm *vm, int vcpuid)
1866 {
1867 	struct vcpu *vcpu;
1868 
1869 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1870 		return (EINVAL);
1871 
1872 	vcpu = &vm->vcpu[vcpuid];
1873 
1874 	vcpu->nmi_pending = 1;
1875 	vcpu_notify_event(vm, vcpuid, false);
1876 	return (0);
1877 }
1878 
1879 int
1880 vm_nmi_pending(struct vm *vm, int vcpuid)
1881 {
1882 	struct vcpu *vcpu;
1883 
1884 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1885 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1886 
1887 	vcpu = &vm->vcpu[vcpuid];
1888 
1889 	return (vcpu->nmi_pending);
1890 }
1891 
1892 void
1893 vm_nmi_clear(struct vm *vm, int vcpuid)
1894 {
1895 	struct vcpu *vcpu;
1896 
1897 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1898 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1899 
1900 	vcpu = &vm->vcpu[vcpuid];
1901 
1902 	if (vcpu->nmi_pending == 0)
1903 		panic("vm_nmi_clear: inconsistent nmi_pending state");
1904 
1905 	vcpu->nmi_pending = 0;
1906 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
1907 }
1908 
1909 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
1910 
1911 int
1912 vm_inject_extint(struct vm *vm, int vcpuid)
1913 {
1914 	struct vcpu *vcpu;
1915 
1916 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1917 		return (EINVAL);
1918 
1919 	vcpu = &vm->vcpu[vcpuid];
1920 
1921 	vcpu->extint_pending = 1;
1922 	vcpu_notify_event(vm, vcpuid, false);
1923 	return (0);
1924 }
1925 
1926 int
1927 vm_extint_pending(struct vm *vm, int vcpuid)
1928 {
1929 	struct vcpu *vcpu;
1930 
1931 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1932 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1933 
1934 	vcpu = &vm->vcpu[vcpuid];
1935 
1936 	return (vcpu->extint_pending);
1937 }
1938 
1939 void
1940 vm_extint_clear(struct vm *vm, int vcpuid)
1941 {
1942 	struct vcpu *vcpu;
1943 
1944 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1945 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1946 
1947 	vcpu = &vm->vcpu[vcpuid];
1948 
1949 	if (vcpu->extint_pending == 0)
1950 		panic("vm_extint_clear: inconsistent extint_pending state");
1951 
1952 	vcpu->extint_pending = 0;
1953 	vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
1954 }
1955 
1956 int
1957 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
1958 {
1959 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1960 		return (EINVAL);
1961 
1962 	if (type < 0 || type >= VM_CAP_MAX)
1963 		return (EINVAL);
1964 
1965 	return (VMGETCAP(vm->cookie, vcpu, type, retval));
1966 }
1967 
1968 int
1969 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
1970 {
1971 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1972 		return (EINVAL);
1973 
1974 	if (type < 0 || type >= VM_CAP_MAX)
1975 		return (EINVAL);
1976 
1977 	return (VMSETCAP(vm->cookie, vcpu, type, val));
1978 }
1979 
1980 struct vlapic *
1981 vm_lapic(struct vm *vm, int cpu)
1982 {
1983 	return (vm->vcpu[cpu].vlapic);
1984 }
1985 
1986 struct vioapic *
1987 vm_ioapic(struct vm *vm)
1988 {
1989 
1990 	return (vm->vioapic);
1991 }
1992 
1993 struct vhpet *
1994 vm_hpet(struct vm *vm)
1995 {
1996 
1997 	return (vm->vhpet);
1998 }
1999 
2000 boolean_t
2001 vmm_is_pptdev(int bus, int slot, int func)
2002 {
2003 	int found, i, n;
2004 	int b, s, f;
2005 	char *val, *cp, *cp2;
2006 
2007 	/*
2008 	 * XXX
2009 	 * The length of an environment variable is limited to 128 bytes which
2010 	 * puts an upper limit on the number of passthru devices that may be
2011 	 * specified using a single environment variable.
2012 	 *
2013 	 * Work around this by scanning multiple environment variable
2014 	 * names instead of a single one - yuck!
2015 	 */
2016 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2017 
2018 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2019 	found = 0;
2020 	for (i = 0; names[i] != NULL && !found; i++) {
2021 		cp = val = kern_getenv(names[i]);
2022 		while (cp != NULL && *cp != '\0') {
2023 			if ((cp2 = strchr(cp, ' ')) != NULL)
2024 				*cp2 = '\0';
2025 
2026 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2027 			if (n == 3 && bus == b && slot == s && func == f) {
2028 				found = 1;
2029 				break;
2030 			}
2031 
2032 			if (cp2 != NULL)
2033 				*cp2++ = ' ';
2034 
2035 			cp = cp2;
2036 		}
2037 		freeenv(val);
2038 	}
2039 	return (found);
2040 }
2041 
2042 void *
2043 vm_iommu_domain(struct vm *vm)
2044 {
2045 
2046 	return (vm->iommu);
2047 }
2048 
2049 int
2050 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
2051     bool from_idle)
2052 {
2053 	int error;
2054 	struct vcpu *vcpu;
2055 
2056 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2057 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
2058 
2059 	vcpu = &vm->vcpu[vcpuid];
2060 
2061 	vcpu_lock(vcpu);
2062 	error = vcpu_set_state_locked(vcpu, newstate, from_idle);
2063 	vcpu_unlock(vcpu);
2064 
2065 	return (error);
2066 }
2067 
2068 enum vcpu_state
2069 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2070 {
2071 	struct vcpu *vcpu;
2072 	enum vcpu_state state;
2073 
2074 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2075 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2076 
2077 	vcpu = &vm->vcpu[vcpuid];
2078 
2079 	vcpu_lock(vcpu);
2080 	state = vcpu->state;
2081 	if (hostcpu != NULL)
2082 		*hostcpu = vcpu->hostcpu;
2083 	vcpu_unlock(vcpu);
2084 
2085 	return (state);
2086 }
2087 
2088 int
2089 vm_activate_cpu(struct vm *vm, int vcpuid)
2090 {
2091 
2092 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2093 		return (EINVAL);
2094 
2095 	if (CPU_ISSET(vcpuid, &vm->active_cpus))
2096 		return (EBUSY);
2097 
2098 	VCPU_CTR0(vm, vcpuid, "activated");
2099 	CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2100 	return (0);
2101 }
2102 
2103 cpuset_t
2104 vm_active_cpus(struct vm *vm)
2105 {
2106 
2107 	return (vm->active_cpus);
2108 }
2109 
2110 cpuset_t
2111 vm_suspended_cpus(struct vm *vm)
2112 {
2113 
2114 	return (vm->suspended_cpus);
2115 }
2116 
2117 void *
2118 vcpu_stats(struct vm *vm, int vcpuid)
2119 {
2120 
2121 	return (vm->vcpu[vcpuid].stats);
2122 }
2123 
2124 int
2125 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2126 {
2127 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2128 		return (EINVAL);
2129 
2130 	*state = vm->vcpu[vcpuid].x2apic_state;
2131 
2132 	return (0);
2133 }
2134 
2135 int
2136 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2137 {
2138 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2139 		return (EINVAL);
2140 
2141 	if (state >= X2APIC_STATE_LAST)
2142 		return (EINVAL);
2143 
2144 	vm->vcpu[vcpuid].x2apic_state = state;
2145 
2146 	vlapic_set_x2apic_state(vm, vcpuid, state);
2147 
2148 	return (0);
2149 }
2150 
2151 /*
2152  * This function is called to ensure that a vcpu "sees" a pending event
2153  * as soon as possible:
2154  * - If the vcpu thread is sleeping then it is woken up.
2155  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2156  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2157  */
2158 void
2159 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2160 {
2161 	int hostcpu;
2162 	struct vcpu *vcpu;
2163 
2164 	vcpu = &vm->vcpu[vcpuid];
2165 
2166 	vcpu_lock(vcpu);
2167 	hostcpu = vcpu->hostcpu;
2168 	if (vcpu->state == VCPU_RUNNING) {
2169 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2170 		if (hostcpu != curcpu) {
2171 			if (lapic_intr) {
2172 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2173 				    vmm_ipinum);
2174 			} else {
2175 				ipi_cpu(hostcpu, vmm_ipinum);
2176 			}
2177 		} else {
2178 			/*
2179 			 * If the 'vcpu' is running on 'curcpu' then it must
2180 			 * be sending a notification to itself (e.g. SELF_IPI).
2181 			 * The pending event will be picked up when the vcpu
2182 			 * transitions back to guest context.
2183 			 */
2184 		}
2185 	} else {
2186 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2187 		    "with hostcpu %d", vcpu->state, hostcpu));
2188 		if (vcpu->state == VCPU_SLEEPING)
2189 			wakeup_one(vcpu);
2190 	}
2191 	vcpu_unlock(vcpu);
2192 }
2193 
2194 struct vmspace *
2195 vm_get_vmspace(struct vm *vm)
2196 {
2197 
2198 	return (vm->vmspace);
2199 }
2200 
2201 int
2202 vm_apicid2vcpuid(struct vm *vm, int apicid)
2203 {
2204 	/*
2205 	 * XXX apic id is assumed to be numerically identical to vcpu id
2206 	 */
2207 	return (apicid);
2208 }
2209 
2210 void
2211 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2212     vm_rendezvous_func_t func, void *arg)
2213 {
2214 	int i;
2215 
2216 	/*
2217 	 * Enforce that this function is called without any locks
2218 	 */
2219 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2220 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
2221 	    ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2222 
2223 restart:
2224 	mtx_lock(&vm->rendezvous_mtx);
2225 	if (vm->rendezvous_func != NULL) {
2226 		/*
2227 		 * If a rendezvous is already in progress then we need to
2228 		 * call the rendezvous handler in case this 'vcpuid' is one
2229 		 * of the targets of the rendezvous.
2230 		 */
2231 		RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2232 		mtx_unlock(&vm->rendezvous_mtx);
2233 		vm_handle_rendezvous(vm, vcpuid);
2234 		goto restart;
2235 	}
2236 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2237 	    "rendezvous is still in progress"));
2238 
2239 	RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2240 	vm->rendezvous_req_cpus = dest;
2241 	CPU_ZERO(&vm->rendezvous_done_cpus);
2242 	vm->rendezvous_arg = arg;
2243 	vm_set_rendezvous_func(vm, func);
2244 	mtx_unlock(&vm->rendezvous_mtx);
2245 
2246 	/*
2247 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2248 	 * vcpus so they handle the rendezvous as soon as possible.
2249 	 */
2250 	for (i = 0; i < VM_MAXCPU; i++) {
2251 		if (CPU_ISSET(i, &dest))
2252 			vcpu_notify_event(vm, i, false);
2253 	}
2254 
2255 	vm_handle_rendezvous(vm, vcpuid);
2256 }
2257 
2258 struct vatpic *
2259 vm_atpic(struct vm *vm)
2260 {
2261 	return (vm->vatpic);
2262 }
2263 
2264 struct vatpit *
2265 vm_atpit(struct vm *vm)
2266 {
2267 	return (vm->vatpit);
2268 }
2269 
2270 struct vpmtmr *
2271 vm_pmtmr(struct vm *vm)
2272 {
2273 
2274 	return (vm->vpmtmr);
2275 }
2276 
2277 struct vrtc *
2278 vm_rtc(struct vm *vm)
2279 {
2280 
2281 	return (vm->vrtc);
2282 }
2283 
2284 enum vm_reg_name
2285 vm_segment_name(int seg)
2286 {
2287 	static enum vm_reg_name seg_names[] = {
2288 		VM_REG_GUEST_ES,
2289 		VM_REG_GUEST_CS,
2290 		VM_REG_GUEST_SS,
2291 		VM_REG_GUEST_DS,
2292 		VM_REG_GUEST_FS,
2293 		VM_REG_GUEST_GS
2294 	};
2295 
2296 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2297 	    ("%s: invalid segment encoding %d", __func__, seg));
2298 	return (seg_names[seg]);
2299 }
2300 
2301 void
2302 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2303     int num_copyinfo)
2304 {
2305 	int idx;
2306 
2307 	for (idx = 0; idx < num_copyinfo; idx++) {
2308 		if (copyinfo[idx].cookie != NULL)
2309 			vm_gpa_release(copyinfo[idx].cookie);
2310 	}
2311 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2312 }
2313 
2314 int
2315 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2316     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2317     int num_copyinfo)
2318 {
2319 	int error, idx, nused;
2320 	size_t n, off, remaining;
2321 	void *hva, *cookie;
2322 	uint64_t gpa;
2323 
2324 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2325 
2326 	nused = 0;
2327 	remaining = len;
2328 	while (remaining > 0) {
2329 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2330 		error = vmm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa);
2331 		if (error)
2332 			return (error);
2333 		off = gpa & PAGE_MASK;
2334 		n = min(remaining, PAGE_SIZE - off);
2335 		copyinfo[nused].gpa = gpa;
2336 		copyinfo[nused].len = n;
2337 		remaining -= n;
2338 		gla += n;
2339 		nused++;
2340 	}
2341 
2342 	for (idx = 0; idx < nused; idx++) {
2343 		hva = vm_gpa_hold(vm, copyinfo[idx].gpa, copyinfo[idx].len,
2344 		    prot, &cookie);
2345 		if (hva == NULL)
2346 			break;
2347 		copyinfo[idx].hva = hva;
2348 		copyinfo[idx].cookie = cookie;
2349 	}
2350 
2351 	if (idx != nused) {
2352 		vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2353 		return (-1);
2354 	} else {
2355 		return (0);
2356 	}
2357 }
2358 
2359 void
2360 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2361     size_t len)
2362 {
2363 	char *dst;
2364 	int idx;
2365 
2366 	dst = kaddr;
2367 	idx = 0;
2368 	while (len > 0) {
2369 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2370 		len -= copyinfo[idx].len;
2371 		dst += copyinfo[idx].len;
2372 		idx++;
2373 	}
2374 }
2375 
2376 void
2377 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2378     struct vm_copyinfo *copyinfo, size_t len)
2379 {
2380 	const char *src;
2381 	int idx;
2382 
2383 	src = kaddr;
2384 	idx = 0;
2385 	while (len > 0) {
2386 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2387 		len -= copyinfo[idx].len;
2388 		src += copyinfo[idx].len;
2389 		idx++;
2390 	}
2391 }
2392 
2393 /*
2394  * Return the amount of in-use and wired memory for the VM. Since
2395  * these are global stats, only return the values with for vCPU 0
2396  */
2397 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2398 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2399 
2400 static void
2401 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2402 {
2403 
2404 	if (vcpu == 0) {
2405 		vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2406 	       	    PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2407 	}
2408 }
2409 
2410 static void
2411 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2412 {
2413 
2414 	if (vcpu == 0) {
2415 		vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2416 	      	    PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2417 	}
2418 }
2419 
2420 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2421 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2422