xref: /freebsd/sys/amd64/vmm/vmm.c (revision 313376588638950ba1e93c403dd8c97bc52fd3a2)
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 "vmm_ipi.h"
78 #include "vmm_stat.h"
79 #include "vmm_lapic.h"
80 
81 #include "io/ppt.h"
82 #include "io/iommu.h"
83 
84 struct vlapic;
85 
86 /*
87  * Initialization:
88  * (a) allocated when vcpu is created
89  * (i) initialized when vcpu is created and when it is reinitialized
90  * (o) initialized the first time the vcpu is created
91  * (x) initialized before use
92  */
93 struct vcpu {
94 	struct mtx 	mtx;		/* (o) protects 'state' and 'hostcpu' */
95 	enum vcpu_state	state;		/* (o) vcpu state */
96 	int		hostcpu;	/* (o) vcpu's host cpu */
97 	struct vlapic	*vlapic;	/* (i) APIC device model */
98 	enum x2apic_state x2apic_state;	/* (i) APIC mode */
99 	uint64_t	exitintinfo;	/* (i) events pending at VM exit */
100 	int		nmi_pending;	/* (i) NMI pending */
101 	int		extint_pending;	/* (i) INTR pending */
102 	struct vm_exception exception;	/* (x) exception collateral */
103 	int	exception_pending;	/* (i) exception pending */
104 	struct savefpu	*guestfpu;	/* (a,i) guest fpu state */
105 	uint64_t	guest_xcr0;	/* (i) guest %xcr0 register */
106 	void		*stats;		/* (a,i) statistics */
107 	struct vm_exit	exitinfo;	/* (x) exit reason and collateral */
108 };
109 
110 #define	vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
111 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
112 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
113 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
114 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
115 
116 struct mem_seg {
117 	vm_paddr_t	gpa;
118 	size_t		len;
119 	boolean_t	wired;
120 	vm_object_t	object;
121 };
122 #define	VM_MAX_MEMORY_SEGMENTS	2
123 
124 /*
125  * Initialization:
126  * (o) initialized the first time the VM is created
127  * (i) initialized when VM is created and when it is reinitialized
128  * (x) initialized before use
129  */
130 struct vm {
131 	void		*cookie;		/* (i) cpu-specific data */
132 	void		*iommu;			/* (x) iommu-specific data */
133 	struct vhpet	*vhpet;			/* (i) virtual HPET */
134 	struct vioapic	*vioapic;		/* (i) virtual ioapic */
135 	struct vatpic	*vatpic;		/* (i) virtual atpic */
136 	struct vatpit	*vatpit;		/* (i) virtual atpit */
137 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
138 	int		suspend;		/* (i) stop VM execution */
139 	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
140 	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
141 	cpuset_t	rendezvous_req_cpus;	/* (x) rendezvous requested */
142 	cpuset_t	rendezvous_done_cpus;	/* (x) rendezvous finished */
143 	void		*rendezvous_arg;	/* (x) rendezvous func/arg */
144 	vm_rendezvous_func_t rendezvous_func;
145 	struct mtx	rendezvous_mtx;		/* (o) rendezvous lock */
146 	int		num_mem_segs;		/* (o) guest memory segments */
147 	struct mem_seg	mem_segs[VM_MAX_MEMORY_SEGMENTS];
148 	struct vmspace	*vmspace;		/* (o) guest's address space */
149 	char		name[VM_MAX_NAMELEN];	/* (o) virtual machine name */
150 	struct vcpu	vcpu[VM_MAXCPU];	/* (i) guest vcpus */
151 };
152 
153 static int vmm_initialized;
154 
155 static struct vmm_ops *ops;
156 #define	VMM_INIT(num)	(ops != NULL ? (*ops->init)(num) : 0)
157 #define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
158 #define	VMM_RESUME()	(ops != NULL ? (*ops->resume)() : 0)
159 
160 #define	VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
161 #define	VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \
162 	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO)
163 #define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
164 #define	VMSPACE_ALLOC(min, max) \
165 	(ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
166 #define	VMSPACE_FREE(vmspace) \
167 	(ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
168 #define	VMGETREG(vmi, vcpu, num, retval)		\
169 	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
170 #define	VMSETREG(vmi, vcpu, num, val)		\
171 	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
172 #define	VMGETDESC(vmi, vcpu, num, desc)		\
173 	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
174 #define	VMSETDESC(vmi, vcpu, num, desc)		\
175 	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
176 #define	VMGETCAP(vmi, vcpu, num, retval)	\
177 	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
178 #define	VMSETCAP(vmi, vcpu, num, val)		\
179 	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
180 #define	VLAPIC_INIT(vmi, vcpu)			\
181 	(ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
182 #define	VLAPIC_CLEANUP(vmi, vlapic)		\
183 	(ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
184 
185 #define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
186 #define	fpu_stop_emulating()	clts()
187 
188 static MALLOC_DEFINE(M_VM, "vm", "vm");
189 
190 /* statistics */
191 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
192 
193 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
194 
195 /*
196  * Halt the guest if all vcpus are executing a HLT instruction with
197  * interrupts disabled.
198  */
199 static int halt_detection_enabled = 1;
200 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
201     &halt_detection_enabled, 0,
202     "Halt VM if all vcpus execute HLT with interrupts disabled");
203 
204 static int vmm_ipinum;
205 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
206     "IPI vector used for vcpu notifications");
207 
208 static void
209 vcpu_cleanup(struct vm *vm, int i, bool destroy)
210 {
211 	struct vcpu *vcpu = &vm->vcpu[i];
212 
213 	VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
214 	if (destroy) {
215 		vmm_stat_free(vcpu->stats);
216 		fpu_save_area_free(vcpu->guestfpu);
217 	}
218 }
219 
220 static void
221 vcpu_init(struct vm *vm, int vcpu_id, bool create)
222 {
223 	struct vcpu *vcpu;
224 
225 	KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU,
226 	    ("vcpu_init: invalid vcpu %d", vcpu_id));
227 
228 	vcpu = &vm->vcpu[vcpu_id];
229 
230 	if (create) {
231 		KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
232 		    "initialized", vcpu_id));
233 		vcpu_lock_init(vcpu);
234 		vcpu->state = VCPU_IDLE;
235 		vcpu->hostcpu = NOCPU;
236 		vcpu->guestfpu = fpu_save_area_alloc();
237 		vcpu->stats = vmm_stat_alloc();
238 	}
239 
240 	vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
241 	vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
242 	vcpu->exitintinfo = 0;
243 	vcpu->nmi_pending = 0;
244 	vcpu->extint_pending = 0;
245 	vcpu->exception_pending = 0;
246 	vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
247 	fpu_save_area_reset(vcpu->guestfpu);
248 	vmm_stat_init(vcpu->stats);
249 }
250 
251 struct vm_exit *
252 vm_exitinfo(struct vm *vm, int cpuid)
253 {
254 	struct vcpu *vcpu;
255 
256 	if (cpuid < 0 || cpuid >= VM_MAXCPU)
257 		panic("vm_exitinfo: invalid cpuid %d", cpuid);
258 
259 	vcpu = &vm->vcpu[cpuid];
260 
261 	return (&vcpu->exitinfo);
262 }
263 
264 static void
265 vmm_resume(void)
266 {
267 	VMM_RESUME();
268 }
269 
270 static int
271 vmm_init(void)
272 {
273 	int error;
274 
275 	vmm_host_state_init();
276 
277 	vmm_ipinum = vmm_ipi_alloc();
278 	if (vmm_ipinum == 0)
279 		vmm_ipinum = IPI_AST;
280 
281 	error = vmm_mem_init();
282 	if (error)
283 		return (error);
284 
285 	if (vmm_is_intel())
286 		ops = &vmm_ops_intel;
287 	else if (vmm_is_amd())
288 		ops = &vmm_ops_amd;
289 	else
290 		return (ENXIO);
291 
292 	vmm_resume_p = vmm_resume;
293 
294 	return (VMM_INIT(vmm_ipinum));
295 }
296 
297 static int
298 vmm_handler(module_t mod, int what, void *arg)
299 {
300 	int error;
301 
302 	switch (what) {
303 	case MOD_LOAD:
304 		vmmdev_init();
305 		if (ppt_avail_devices() > 0)
306 			iommu_init();
307 		error = vmm_init();
308 		if (error == 0)
309 			vmm_initialized = 1;
310 		break;
311 	case MOD_UNLOAD:
312 		error = vmmdev_cleanup();
313 		if (error == 0) {
314 			vmm_resume_p = NULL;
315 			iommu_cleanup();
316 			if (vmm_ipinum != IPI_AST)
317 				vmm_ipi_free(vmm_ipinum);
318 			error = VMM_CLEANUP();
319 			/*
320 			 * Something bad happened - prevent new
321 			 * VMs from being created
322 			 */
323 			if (error)
324 				vmm_initialized = 0;
325 		}
326 		break;
327 	default:
328 		error = 0;
329 		break;
330 	}
331 	return (error);
332 }
333 
334 static moduledata_t vmm_kmod = {
335 	"vmm",
336 	vmm_handler,
337 	NULL
338 };
339 
340 /*
341  * vmm initialization has the following dependencies:
342  *
343  * - iommu initialization must happen after the pci passthru driver has had
344  *   a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
345  *
346  * - VT-x initialization requires smp_rendezvous() and therefore must happen
347  *   after SMP is fully functional (after SI_SUB_SMP).
348  */
349 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
350 MODULE_VERSION(vmm, 1);
351 
352 static void
353 vm_init(struct vm *vm, bool create)
354 {
355 	int i;
356 
357 	vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
358 	vm->iommu = NULL;
359 	vm->vioapic = vioapic_init(vm);
360 	vm->vhpet = vhpet_init(vm);
361 	vm->vatpic = vatpic_init(vm);
362 	vm->vatpit = vatpit_init(vm);
363 
364 	CPU_ZERO(&vm->active_cpus);
365 
366 	vm->suspend = 0;
367 	CPU_ZERO(&vm->suspended_cpus);
368 
369 	for (i = 0; i < VM_MAXCPU; i++)
370 		vcpu_init(vm, i, create);
371 }
372 
373 int
374 vm_create(const char *name, struct vm **retvm)
375 {
376 	struct vm *vm;
377 	struct vmspace *vmspace;
378 
379 	/*
380 	 * If vmm.ko could not be successfully initialized then don't attempt
381 	 * to create the virtual machine.
382 	 */
383 	if (!vmm_initialized)
384 		return (ENXIO);
385 
386 	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
387 		return (EINVAL);
388 
389 	vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
390 	if (vmspace == NULL)
391 		return (ENOMEM);
392 
393 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
394 	strcpy(vm->name, name);
395 	vm->num_mem_segs = 0;
396 	vm->vmspace = vmspace;
397 	mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
398 
399 	vm_init(vm, true);
400 
401 	*retvm = vm;
402 	return (0);
403 }
404 
405 static void
406 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg)
407 {
408 
409 	if (seg->object != NULL)
410 		vmm_mem_free(vm->vmspace, seg->gpa, seg->len);
411 
412 	bzero(seg, sizeof(*seg));
413 }
414 
415 static void
416 vm_cleanup(struct vm *vm, bool destroy)
417 {
418 	int i;
419 
420 	ppt_unassign_all(vm);
421 
422 	if (vm->iommu != NULL)
423 		iommu_destroy_domain(vm->iommu);
424 
425 	vatpit_cleanup(vm->vatpit);
426 	vhpet_cleanup(vm->vhpet);
427 	vatpic_cleanup(vm->vatpic);
428 	vioapic_cleanup(vm->vioapic);
429 
430 	for (i = 0; i < VM_MAXCPU; i++)
431 		vcpu_cleanup(vm, i, destroy);
432 
433 	VMCLEANUP(vm->cookie);
434 
435 	if (destroy) {
436 		for (i = 0; i < vm->num_mem_segs; i++)
437 			vm_free_mem_seg(vm, &vm->mem_segs[i]);
438 
439 		vm->num_mem_segs = 0;
440 
441 		VMSPACE_FREE(vm->vmspace);
442 		vm->vmspace = NULL;
443 	}
444 }
445 
446 void
447 vm_destroy(struct vm *vm)
448 {
449 	vm_cleanup(vm, true);
450 	free(vm, M_VM);
451 }
452 
453 int
454 vm_reinit(struct vm *vm)
455 {
456 	int error;
457 
458 	/*
459 	 * A virtual machine can be reset only if all vcpus are suspended.
460 	 */
461 	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
462 		vm_cleanup(vm, false);
463 		vm_init(vm, false);
464 		error = 0;
465 	} else {
466 		error = EBUSY;
467 	}
468 
469 	return (error);
470 }
471 
472 const char *
473 vm_name(struct vm *vm)
474 {
475 	return (vm->name);
476 }
477 
478 int
479 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
480 {
481 	vm_object_t obj;
482 
483 	if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
484 		return (ENOMEM);
485 	else
486 		return (0);
487 }
488 
489 int
490 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
491 {
492 
493 	vmm_mmio_free(vm->vmspace, gpa, len);
494 	return (0);
495 }
496 
497 boolean_t
498 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa)
499 {
500 	int i;
501 	vm_paddr_t gpabase, gpalimit;
502 
503 	for (i = 0; i < vm->num_mem_segs; i++) {
504 		gpabase = vm->mem_segs[i].gpa;
505 		gpalimit = gpabase + vm->mem_segs[i].len;
506 		if (gpa >= gpabase && gpa < gpalimit)
507 			return (TRUE);		/* 'gpa' is regular memory */
508 	}
509 
510 	if (ppt_is_mmio(vm, gpa))
511 		return (TRUE);			/* 'gpa' is pci passthru mmio */
512 
513 	return (FALSE);
514 }
515 
516 int
517 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
518 {
519 	int available, allocated;
520 	struct mem_seg *seg;
521 	vm_object_t object;
522 	vm_paddr_t g;
523 
524 	if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
525 		return (EINVAL);
526 
527 	available = allocated = 0;
528 	g = gpa;
529 	while (g < gpa + len) {
530 		if (vm_mem_allocated(vm, g))
531 			allocated++;
532 		else
533 			available++;
534 
535 		g += PAGE_SIZE;
536 	}
537 
538 	/*
539 	 * If there are some allocated and some available pages in the address
540 	 * range then it is an error.
541 	 */
542 	if (allocated && available)
543 		return (EINVAL);
544 
545 	/*
546 	 * If the entire address range being requested has already been
547 	 * allocated then there isn't anything more to do.
548 	 */
549 	if (allocated && available == 0)
550 		return (0);
551 
552 	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
553 		return (E2BIG);
554 
555 	seg = &vm->mem_segs[vm->num_mem_segs];
556 
557 	if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL)
558 		return (ENOMEM);
559 
560 	seg->gpa = gpa;
561 	seg->len = len;
562 	seg->object = object;
563 	seg->wired = FALSE;
564 
565 	vm->num_mem_segs++;
566 
567 	return (0);
568 }
569 
570 static vm_paddr_t
571 vm_maxmem(struct vm *vm)
572 {
573 	int i;
574 	vm_paddr_t gpa, maxmem;
575 
576 	maxmem = 0;
577 	for (i = 0; i < vm->num_mem_segs; i++) {
578 		gpa = vm->mem_segs[i].gpa + vm->mem_segs[i].len;
579 		if (gpa > maxmem)
580 			maxmem = gpa;
581 	}
582 	return (maxmem);
583 }
584 
585 static void
586 vm_gpa_unwire(struct vm *vm)
587 {
588 	int i, rv;
589 	struct mem_seg *seg;
590 
591 	for (i = 0; i < vm->num_mem_segs; i++) {
592 		seg = &vm->mem_segs[i];
593 		if (!seg->wired)
594 			continue;
595 
596 		rv = vm_map_unwire(&vm->vmspace->vm_map,
597 				   seg->gpa, seg->gpa + seg->len,
598 				   VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
599 		KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment "
600 		    "%#lx/%ld could not be unwired: %d",
601 		    vm_name(vm), seg->gpa, seg->len, rv));
602 
603 		seg->wired = FALSE;
604 	}
605 }
606 
607 static int
608 vm_gpa_wire(struct vm *vm)
609 {
610 	int i, rv;
611 	struct mem_seg *seg;
612 
613 	for (i = 0; i < vm->num_mem_segs; i++) {
614 		seg = &vm->mem_segs[i];
615 		if (seg->wired)
616 			continue;
617 
618 		/* XXX rlimits? */
619 		rv = vm_map_wire(&vm->vmspace->vm_map,
620 				 seg->gpa, seg->gpa + seg->len,
621 				 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
622 		if (rv != KERN_SUCCESS)
623 			break;
624 
625 		seg->wired = TRUE;
626 	}
627 
628 	if (i < vm->num_mem_segs) {
629 		/*
630 		 * Undo the wiring before returning an error.
631 		 */
632 		vm_gpa_unwire(vm);
633 		return (EAGAIN);
634 	}
635 
636 	return (0);
637 }
638 
639 static void
640 vm_iommu_modify(struct vm *vm, boolean_t map)
641 {
642 	int i, sz;
643 	vm_paddr_t gpa, hpa;
644 	struct mem_seg *seg;
645 	void *vp, *cookie, *host_domain;
646 
647 	sz = PAGE_SIZE;
648 	host_domain = iommu_host_domain();
649 
650 	for (i = 0; i < vm->num_mem_segs; i++) {
651 		seg = &vm->mem_segs[i];
652 		KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired",
653 		    vm_name(vm), seg->gpa, seg->len));
654 
655 		gpa = seg->gpa;
656 		while (gpa < seg->gpa + seg->len) {
657 			vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE,
658 					 &cookie);
659 			KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
660 			    vm_name(vm), gpa));
661 
662 			vm_gpa_release(cookie);
663 
664 			hpa = DMAP_TO_PHYS((uintptr_t)vp);
665 			if (map) {
666 				iommu_create_mapping(vm->iommu, gpa, hpa, sz);
667 				iommu_remove_mapping(host_domain, hpa, sz);
668 			} else {
669 				iommu_remove_mapping(vm->iommu, gpa, sz);
670 				iommu_create_mapping(host_domain, hpa, hpa, sz);
671 			}
672 
673 			gpa += PAGE_SIZE;
674 		}
675 	}
676 
677 	/*
678 	 * Invalidate the cached translations associated with the domain
679 	 * from which pages were removed.
680 	 */
681 	if (map)
682 		iommu_invalidate_tlb(host_domain);
683 	else
684 		iommu_invalidate_tlb(vm->iommu);
685 }
686 
687 #define	vm_iommu_unmap(vm)	vm_iommu_modify((vm), FALSE)
688 #define	vm_iommu_map(vm)	vm_iommu_modify((vm), TRUE)
689 
690 int
691 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
692 {
693 	int error;
694 
695 	error = ppt_unassign_device(vm, bus, slot, func);
696 	if (error)
697 		return (error);
698 
699 	if (ppt_assigned_devices(vm) == 0) {
700 		vm_iommu_unmap(vm);
701 		vm_gpa_unwire(vm);
702 	}
703 	return (0);
704 }
705 
706 int
707 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
708 {
709 	int error;
710 	vm_paddr_t maxaddr;
711 
712 	/*
713 	 * Virtual machines with pci passthru devices get special treatment:
714 	 * - the guest physical memory is wired
715 	 * - the iommu is programmed to do the 'gpa' to 'hpa' translation
716 	 *
717 	 * We need to do this before the first pci passthru device is attached.
718 	 */
719 	if (ppt_assigned_devices(vm) == 0) {
720 		KASSERT(vm->iommu == NULL,
721 		    ("vm_assign_pptdev: iommu must be NULL"));
722 		maxaddr = vm_maxmem(vm);
723 		vm->iommu = iommu_create_domain(maxaddr);
724 
725 		error = vm_gpa_wire(vm);
726 		if (error)
727 			return (error);
728 
729 		vm_iommu_map(vm);
730 	}
731 
732 	error = ppt_assign_device(vm, bus, slot, func);
733 	return (error);
734 }
735 
736 void *
737 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
738 	    void **cookie)
739 {
740 	int count, pageoff;
741 	vm_page_t m;
742 
743 	pageoff = gpa & PAGE_MASK;
744 	if (len > PAGE_SIZE - pageoff)
745 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
746 
747 	count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
748 	    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
749 
750 	if (count == 1) {
751 		*cookie = m;
752 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
753 	} else {
754 		*cookie = NULL;
755 		return (NULL);
756 	}
757 }
758 
759 void
760 vm_gpa_release(void *cookie)
761 {
762 	vm_page_t m = cookie;
763 
764 	vm_page_lock(m);
765 	vm_page_unhold(m);
766 	vm_page_unlock(m);
767 }
768 
769 int
770 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
771 		  struct vm_memory_segment *seg)
772 {
773 	int i;
774 
775 	for (i = 0; i < vm->num_mem_segs; i++) {
776 		if (gpabase == vm->mem_segs[i].gpa) {
777 			seg->gpa = vm->mem_segs[i].gpa;
778 			seg->len = vm->mem_segs[i].len;
779 			seg->wired = vm->mem_segs[i].wired;
780 			return (0);
781 		}
782 	}
783 	return (-1);
784 }
785 
786 int
787 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len,
788 	      vm_offset_t *offset, struct vm_object **object)
789 {
790 	int i;
791 	size_t seg_len;
792 	vm_paddr_t seg_gpa;
793 	vm_object_t seg_obj;
794 
795 	for (i = 0; i < vm->num_mem_segs; i++) {
796 		if ((seg_obj = vm->mem_segs[i].object) == NULL)
797 			continue;
798 
799 		seg_gpa = vm->mem_segs[i].gpa;
800 		seg_len = vm->mem_segs[i].len;
801 
802 		if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) {
803 			*offset = gpa - seg_gpa;
804 			*object = seg_obj;
805 			vm_object_reference(seg_obj);
806 			return (0);
807 		}
808 	}
809 
810 	return (EINVAL);
811 }
812 
813 int
814 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
815 {
816 
817 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
818 		return (EINVAL);
819 
820 	if (reg >= VM_REG_LAST)
821 		return (EINVAL);
822 
823 	return (VMGETREG(vm->cookie, vcpu, reg, retval));
824 }
825 
826 int
827 vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
828 {
829 
830 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
831 		return (EINVAL);
832 
833 	if (reg >= VM_REG_LAST)
834 		return (EINVAL);
835 
836 	return (VMSETREG(vm->cookie, vcpu, reg, val));
837 }
838 
839 static boolean_t
840 is_descriptor_table(int reg)
841 {
842 
843 	switch (reg) {
844 	case VM_REG_GUEST_IDTR:
845 	case VM_REG_GUEST_GDTR:
846 		return (TRUE);
847 	default:
848 		return (FALSE);
849 	}
850 }
851 
852 static boolean_t
853 is_segment_register(int reg)
854 {
855 
856 	switch (reg) {
857 	case VM_REG_GUEST_ES:
858 	case VM_REG_GUEST_CS:
859 	case VM_REG_GUEST_SS:
860 	case VM_REG_GUEST_DS:
861 	case VM_REG_GUEST_FS:
862 	case VM_REG_GUEST_GS:
863 	case VM_REG_GUEST_TR:
864 	case VM_REG_GUEST_LDTR:
865 		return (TRUE);
866 	default:
867 		return (FALSE);
868 	}
869 }
870 
871 int
872 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
873 		struct seg_desc *desc)
874 {
875 
876 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
877 		return (EINVAL);
878 
879 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
880 		return (EINVAL);
881 
882 	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
883 }
884 
885 int
886 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
887 		struct seg_desc *desc)
888 {
889 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
890 		return (EINVAL);
891 
892 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
893 		return (EINVAL);
894 
895 	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
896 }
897 
898 static void
899 restore_guest_fpustate(struct vcpu *vcpu)
900 {
901 
902 	/* flush host state to the pcb */
903 	fpuexit(curthread);
904 
905 	/* restore guest FPU state */
906 	fpu_stop_emulating();
907 	fpurestore(vcpu->guestfpu);
908 
909 	/* restore guest XCR0 if XSAVE is enabled in the host */
910 	if (rcr4() & CR4_XSAVE)
911 		load_xcr(0, vcpu->guest_xcr0);
912 
913 	/*
914 	 * The FPU is now "dirty" with the guest's state so turn on emulation
915 	 * to trap any access to the FPU by the host.
916 	 */
917 	fpu_start_emulating();
918 }
919 
920 static void
921 save_guest_fpustate(struct vcpu *vcpu)
922 {
923 
924 	if ((rcr0() & CR0_TS) == 0)
925 		panic("fpu emulation not enabled in host!");
926 
927 	/* save guest XCR0 and restore host XCR0 */
928 	if (rcr4() & CR4_XSAVE) {
929 		vcpu->guest_xcr0 = rxcr(0);
930 		load_xcr(0, vmm_get_host_xcr0());
931 	}
932 
933 	/* save guest FPU state */
934 	fpu_stop_emulating();
935 	fpusave(vcpu->guestfpu);
936 	fpu_start_emulating();
937 }
938 
939 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
940 
941 static int
942 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
943     bool from_idle)
944 {
945 	int error;
946 
947 	vcpu_assert_locked(vcpu);
948 
949 	/*
950 	 * State transitions from the vmmdev_ioctl() must always begin from
951 	 * the VCPU_IDLE state. This guarantees that there is only a single
952 	 * ioctl() operating on a vcpu at any point.
953 	 */
954 	if (from_idle) {
955 		while (vcpu->state != VCPU_IDLE)
956 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
957 	} else {
958 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
959 		    "vcpu idle state"));
960 	}
961 
962 	if (vcpu->state == VCPU_RUNNING) {
963 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
964 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
965 	} else {
966 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
967 		    "vcpu that is not running", vcpu->hostcpu));
968 	}
969 
970 	/*
971 	 * The following state transitions are allowed:
972 	 * IDLE -> FROZEN -> IDLE
973 	 * FROZEN -> RUNNING -> FROZEN
974 	 * FROZEN -> SLEEPING -> FROZEN
975 	 */
976 	switch (vcpu->state) {
977 	case VCPU_IDLE:
978 	case VCPU_RUNNING:
979 	case VCPU_SLEEPING:
980 		error = (newstate != VCPU_FROZEN);
981 		break;
982 	case VCPU_FROZEN:
983 		error = (newstate == VCPU_FROZEN);
984 		break;
985 	default:
986 		error = 1;
987 		break;
988 	}
989 
990 	if (error)
991 		return (EBUSY);
992 
993 	vcpu->state = newstate;
994 	if (newstate == VCPU_RUNNING)
995 		vcpu->hostcpu = curcpu;
996 	else
997 		vcpu->hostcpu = NOCPU;
998 
999 	if (newstate == VCPU_IDLE)
1000 		wakeup(&vcpu->state);
1001 
1002 	return (0);
1003 }
1004 
1005 static void
1006 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1007 {
1008 	int error;
1009 
1010 	if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1011 		panic("Error %d setting state to %d\n", error, newstate);
1012 }
1013 
1014 static void
1015 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
1016 {
1017 	int error;
1018 
1019 	if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
1020 		panic("Error %d setting state to %d", error, newstate);
1021 }
1022 
1023 static void
1024 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func)
1025 {
1026 
1027 	KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked"));
1028 
1029 	/*
1030 	 * Update 'rendezvous_func' and execute a write memory barrier to
1031 	 * ensure that it is visible across all host cpus. This is not needed
1032 	 * for correctness but it does ensure that all the vcpus will notice
1033 	 * that the rendezvous is requested immediately.
1034 	 */
1035 	vm->rendezvous_func = func;
1036 	wmb();
1037 }
1038 
1039 #define	RENDEZVOUS_CTR0(vm, vcpuid, fmt)				\
1040 	do {								\
1041 		if (vcpuid >= 0)					\
1042 			VCPU_CTR0(vm, vcpuid, fmt);			\
1043 		else							\
1044 			VM_CTR0(vm, fmt);				\
1045 	} while (0)
1046 
1047 static void
1048 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1049 {
1050 
1051 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
1052 	    ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1053 
1054 	mtx_lock(&vm->rendezvous_mtx);
1055 	while (vm->rendezvous_func != NULL) {
1056 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1057 		CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1058 
1059 		if (vcpuid != -1 &&
1060 		    CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1061 		    !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1062 			VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1063 			(*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1064 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1065 		}
1066 		if (CPU_CMP(&vm->rendezvous_req_cpus,
1067 		    &vm->rendezvous_done_cpus) == 0) {
1068 			VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1069 			vm_set_rendezvous_func(vm, NULL);
1070 			wakeup(&vm->rendezvous_func);
1071 			break;
1072 		}
1073 		RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1074 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1075 		    "vmrndv", 0);
1076 	}
1077 	mtx_unlock(&vm->rendezvous_mtx);
1078 }
1079 
1080 /*
1081  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1082  */
1083 static int
1084 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1085 {
1086 	struct vcpu *vcpu;
1087 	const char *wmesg;
1088 	int error, t, vcpu_halted, vm_halted;
1089 
1090 	KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1091 
1092 	vcpu = &vm->vcpu[vcpuid];
1093 	vcpu_halted = 0;
1094 	vm_halted = 0;
1095 
1096 	/*
1097 	 * The typical way to halt a cpu is to execute: "sti; hlt"
1098 	 *
1099 	 * STI sets RFLAGS.IF to enable interrupts. However, the processor
1100 	 * remains in an "interrupt shadow" for an additional instruction
1101 	 * following the STI. This guarantees that "sti; hlt" sequence is
1102 	 * atomic and a pending interrupt will be recognized after the HLT.
1103 	 *
1104 	 * After the HLT emulation is done the vcpu is no longer in an
1105 	 * interrupt shadow and a pending interrupt can be injected on
1106 	 * the next entry into the guest.
1107 	 */
1108 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
1109 	KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
1110 	    __func__, error));
1111 
1112 	vcpu_lock(vcpu);
1113 	while (1) {
1114 		/*
1115 		 * Do a final check for pending NMI or interrupts before
1116 		 * really putting this thread to sleep. Also check for
1117 		 * software events that would cause this vcpu to wakeup.
1118 		 *
1119 		 * These interrupts/events could have happened after the
1120 		 * vcpu returned from VMRUN() and before it acquired the
1121 		 * vcpu lock above.
1122 		 */
1123 		if (vm->rendezvous_func != NULL || vm->suspend)
1124 			break;
1125 		if (vm_nmi_pending(vm, vcpuid))
1126 			break;
1127 		if (!intr_disabled) {
1128 			if (vm_extint_pending(vm, vcpuid) ||
1129 			    vlapic_pending_intr(vcpu->vlapic, NULL)) {
1130 				break;
1131 			}
1132 		}
1133 
1134 		/* Don't go to sleep if the vcpu thread needs to yield */
1135 		if (vcpu_should_yield(vm, vcpuid))
1136 			break;
1137 
1138 		/*
1139 		 * Some Linux guests implement "halt" by having all vcpus
1140 		 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1141 		 * track of the vcpus that have entered this state. When all
1142 		 * vcpus enter the halted state the virtual machine is halted.
1143 		 */
1144 		if (intr_disabled) {
1145 			wmesg = "vmhalt";
1146 			VCPU_CTR0(vm, vcpuid, "Halted");
1147 			if (!vcpu_halted && halt_detection_enabled) {
1148 				vcpu_halted = 1;
1149 				CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1150 			}
1151 			if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1152 				vm_halted = 1;
1153 				break;
1154 			}
1155 		} else {
1156 			wmesg = "vmidle";
1157 		}
1158 
1159 		t = ticks;
1160 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1161 		/*
1162 		 * XXX msleep_spin() cannot be interrupted by signals so
1163 		 * wake up periodically to check pending signals.
1164 		 */
1165 		msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1166 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1167 		vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1168 	}
1169 
1170 	if (vcpu_halted)
1171 		CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1172 
1173 	vcpu_unlock(vcpu);
1174 
1175 	if (vm_halted)
1176 		vm_suspend(vm, VM_SUSPEND_HALT);
1177 
1178 	return (0);
1179 }
1180 
1181 static int
1182 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1183 {
1184 	int rv, ftype;
1185 	struct vm_map *map;
1186 	struct vcpu *vcpu;
1187 	struct vm_exit *vme;
1188 
1189 	vcpu = &vm->vcpu[vcpuid];
1190 	vme = &vcpu->exitinfo;
1191 
1192 	ftype = vme->u.paging.fault_type;
1193 	KASSERT(ftype == VM_PROT_READ ||
1194 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1195 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1196 
1197 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1198 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1199 		    vme->u.paging.gpa, ftype);
1200 		if (rv == 0) {
1201 			VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1202 			    ftype == VM_PROT_READ ? "accessed" : "dirty",
1203 			    vme->u.paging.gpa);
1204 			goto done;
1205 		}
1206 	}
1207 
1208 	map = &vm->vmspace->vm_map;
1209 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
1210 
1211 	VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1212 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1213 
1214 	if (rv != KERN_SUCCESS)
1215 		return (EFAULT);
1216 done:
1217 	/* restart execution at the faulting instruction */
1218 	vme->inst_length = 0;
1219 
1220 	return (0);
1221 }
1222 
1223 static int
1224 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1225 {
1226 	struct vie *vie;
1227 	struct vcpu *vcpu;
1228 	struct vm_exit *vme;
1229 	uint64_t gla, gpa;
1230 	struct vm_guest_paging *paging;
1231 	mem_region_read_t mread;
1232 	mem_region_write_t mwrite;
1233 	enum vm_cpu_mode cpu_mode;
1234 	int cs_d, error, length;
1235 
1236 	vcpu = &vm->vcpu[vcpuid];
1237 	vme = &vcpu->exitinfo;
1238 
1239 	gla = vme->u.inst_emul.gla;
1240 	gpa = vme->u.inst_emul.gpa;
1241 	cs_d = vme->u.inst_emul.cs_d;
1242 	vie = &vme->u.inst_emul.vie;
1243 	paging = &vme->u.inst_emul.paging;
1244 	cpu_mode = paging->cpu_mode;
1245 
1246 	VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1247 
1248 	/* Fetch, decode and emulate the faulting instruction */
1249 	if (vie->num_valid == 0) {
1250 		/*
1251 		 * If the instruction length is not known then assume a
1252 		 * maximum size instruction.
1253 		 */
1254 		length = vme->inst_length ? vme->inst_length : VIE_INST_SIZE;
1255 		error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip,
1256 		    length, vie);
1257 	} else {
1258 		/*
1259 		 * The instruction bytes have already been copied into 'vie'
1260 		 */
1261 		error = 0;
1262 	}
1263 	if (error == 1)
1264 		return (0);		/* Resume guest to handle page fault */
1265 	else if (error == -1)
1266 		return (EFAULT);
1267 	else if (error != 0)
1268 		panic("%s: vmm_fetch_instruction error %d", __func__, error);
1269 
1270 	if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0)
1271 		return (EFAULT);
1272 
1273 	/*
1274 	 * If the instruction length is not specified the update it now.
1275 	 */
1276 	if (vme->inst_length == 0)
1277 		vme->inst_length = vie->num_processed;
1278 
1279 	/* return to userland unless this is an in-kernel emulated device */
1280 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1281 		mread = lapic_mmio_read;
1282 		mwrite = lapic_mmio_write;
1283 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1284 		mread = vioapic_mmio_read;
1285 		mwrite = vioapic_mmio_write;
1286 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1287 		mread = vhpet_mmio_read;
1288 		mwrite = vhpet_mmio_write;
1289 	} else {
1290 		*retu = true;
1291 		return (0);
1292 	}
1293 
1294 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1295 	    mread, mwrite, retu);
1296 
1297 	return (error);
1298 }
1299 
1300 static int
1301 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1302 {
1303 	int i, done;
1304 	struct vcpu *vcpu;
1305 
1306 	done = 0;
1307 	vcpu = &vm->vcpu[vcpuid];
1308 
1309 	CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1310 
1311 	/*
1312 	 * Wait until all 'active_cpus' have suspended themselves.
1313 	 *
1314 	 * Since a VM may be suspended at any time including when one or
1315 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1316 	 * handler while we are waiting to prevent a deadlock.
1317 	 */
1318 	vcpu_lock(vcpu);
1319 	while (1) {
1320 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1321 			VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1322 			break;
1323 		}
1324 
1325 		if (vm->rendezvous_func == NULL) {
1326 			VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1327 			vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1328 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1329 			vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1330 		} else {
1331 			VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1332 			vcpu_unlock(vcpu);
1333 			vm_handle_rendezvous(vm, vcpuid);
1334 			vcpu_lock(vcpu);
1335 		}
1336 	}
1337 	vcpu_unlock(vcpu);
1338 
1339 	/*
1340 	 * Wakeup the other sleeping vcpus and return to userspace.
1341 	 */
1342 	for (i = 0; i < VM_MAXCPU; i++) {
1343 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1344 			vcpu_notify_event(vm, i, false);
1345 		}
1346 	}
1347 
1348 	*retu = true;
1349 	return (0);
1350 }
1351 
1352 int
1353 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1354 {
1355 	int i;
1356 
1357 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1358 		return (EINVAL);
1359 
1360 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1361 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1362 		    vm->suspend, how);
1363 		return (EALREADY);
1364 	}
1365 
1366 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1367 
1368 	/*
1369 	 * Notify all active vcpus that they are now suspended.
1370 	 */
1371 	for (i = 0; i < VM_MAXCPU; i++) {
1372 		if (CPU_ISSET(i, &vm->active_cpus))
1373 			vcpu_notify_event(vm, i, false);
1374 	}
1375 
1376 	return (0);
1377 }
1378 
1379 void
1380 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1381 {
1382 	struct vm_exit *vmexit;
1383 
1384 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1385 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1386 
1387 	vmexit = vm_exitinfo(vm, vcpuid);
1388 	vmexit->rip = rip;
1389 	vmexit->inst_length = 0;
1390 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1391 	vmexit->u.suspended.how = vm->suspend;
1392 }
1393 
1394 void
1395 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1396 {
1397 	struct vm_exit *vmexit;
1398 
1399 	KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1400 
1401 	vmexit = vm_exitinfo(vm, vcpuid);
1402 	vmexit->rip = rip;
1403 	vmexit->inst_length = 0;
1404 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1405 	vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1406 }
1407 
1408 void
1409 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1410 {
1411 	struct vm_exit *vmexit;
1412 
1413 	vmexit = vm_exitinfo(vm, vcpuid);
1414 	vmexit->rip = rip;
1415 	vmexit->inst_length = 0;
1416 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1417 	vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1418 }
1419 
1420 int
1421 vm_run(struct vm *vm, struct vm_run *vmrun)
1422 {
1423 	int error, vcpuid;
1424 	struct vcpu *vcpu;
1425 	struct pcb *pcb;
1426 	uint64_t tscval, rip;
1427 	struct vm_exit *vme;
1428 	bool retu, intr_disabled;
1429 	pmap_t pmap;
1430 	void *rptr, *sptr;
1431 
1432 	vcpuid = vmrun->cpuid;
1433 
1434 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1435 		return (EINVAL);
1436 
1437 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1438 		return (EINVAL);
1439 
1440 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1441 		return (EINVAL);
1442 
1443 	rptr = &vm->rendezvous_func;
1444 	sptr = &vm->suspend;
1445 	pmap = vmspace_pmap(vm->vmspace);
1446 	vcpu = &vm->vcpu[vcpuid];
1447 	vme = &vcpu->exitinfo;
1448 	rip = vmrun->rip;
1449 restart:
1450 	critical_enter();
1451 
1452 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1453 	    ("vm_run: absurd pm_active"));
1454 
1455 	tscval = rdtsc();
1456 
1457 	pcb = PCPU_GET(curpcb);
1458 	set_pcb_flags(pcb, PCB_FULL_IRET);
1459 
1460 	restore_guest_fpustate(vcpu);
1461 
1462 	vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1463 	error = VMRUN(vm->cookie, vcpuid, rip, pmap, rptr, sptr);
1464 	vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1465 
1466 	save_guest_fpustate(vcpu);
1467 
1468 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1469 
1470 	critical_exit();
1471 
1472 	if (error == 0) {
1473 		retu = false;
1474 		switch (vme->exitcode) {
1475 		case VM_EXITCODE_SUSPENDED:
1476 			error = vm_handle_suspend(vm, vcpuid, &retu);
1477 			break;
1478 		case VM_EXITCODE_IOAPIC_EOI:
1479 			vioapic_process_eoi(vm, vcpuid,
1480 			    vme->u.ioapic_eoi.vector);
1481 			break;
1482 		case VM_EXITCODE_RENDEZVOUS:
1483 			vm_handle_rendezvous(vm, vcpuid);
1484 			error = 0;
1485 			break;
1486 		case VM_EXITCODE_HLT:
1487 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1488 			error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1489 			break;
1490 		case VM_EXITCODE_PAGING:
1491 			error = vm_handle_paging(vm, vcpuid, &retu);
1492 			break;
1493 		case VM_EXITCODE_INST_EMUL:
1494 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1495 			break;
1496 		case VM_EXITCODE_INOUT:
1497 		case VM_EXITCODE_INOUT_STR:
1498 			error = vm_handle_inout(vm, vcpuid, vme, &retu);
1499 			break;
1500 		case VM_EXITCODE_MONITOR:
1501 		case VM_EXITCODE_MWAIT:
1502 			vm_inject_ud(vm, vcpuid);
1503 			break;
1504 		default:
1505 			retu = true;	/* handled in userland */
1506 			break;
1507 		}
1508 	}
1509 
1510 	if (error == 0 && retu == false) {
1511 		rip = vme->rip + vme->inst_length;
1512 		goto restart;
1513 	}
1514 
1515 	/* copy the exit information */
1516 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1517 	return (error);
1518 }
1519 
1520 int
1521 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1522 {
1523 	struct vcpu *vcpu;
1524 	int type, vector;
1525 
1526 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1527 		return (EINVAL);
1528 
1529 	vcpu = &vm->vcpu[vcpuid];
1530 
1531 	if (info & VM_INTINFO_VALID) {
1532 		type = info & VM_INTINFO_TYPE;
1533 		vector = info & 0xff;
1534 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1535 			return (EINVAL);
1536 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1537 			return (EINVAL);
1538 		if (info & VM_INTINFO_RSVD)
1539 			return (EINVAL);
1540 	} else {
1541 		info = 0;
1542 	}
1543 	VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1544 	vcpu->exitintinfo = info;
1545 	return (0);
1546 }
1547 
1548 enum exc_class {
1549 	EXC_BENIGN,
1550 	EXC_CONTRIBUTORY,
1551 	EXC_PAGEFAULT
1552 };
1553 
1554 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
1555 
1556 static enum exc_class
1557 exception_class(uint64_t info)
1558 {
1559 	int type, vector;
1560 
1561 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1562 	type = info & VM_INTINFO_TYPE;
1563 	vector = info & 0xff;
1564 
1565 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1566 	switch (type) {
1567 	case VM_INTINFO_HWINTR:
1568 	case VM_INTINFO_SWINTR:
1569 	case VM_INTINFO_NMI:
1570 		return (EXC_BENIGN);
1571 	default:
1572 		/*
1573 		 * Hardware exception.
1574 		 *
1575 		 * SVM and VT-x use identical type values to represent NMI,
1576 		 * hardware interrupt and software interrupt.
1577 		 *
1578 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1579 		 * for exceptions except #BP and #OF. #BP and #OF use a type
1580 		 * value of '5' or '6'. Therefore we don't check for explicit
1581 		 * values of 'type' to classify 'intinfo' into a hardware
1582 		 * exception.
1583 		 */
1584 		break;
1585 	}
1586 
1587 	switch (vector) {
1588 	case IDT_PF:
1589 	case IDT_VE:
1590 		return (EXC_PAGEFAULT);
1591 	case IDT_DE:
1592 	case IDT_TS:
1593 	case IDT_NP:
1594 	case IDT_SS:
1595 	case IDT_GP:
1596 		return (EXC_CONTRIBUTORY);
1597 	default:
1598 		return (EXC_BENIGN);
1599 	}
1600 }
1601 
1602 static int
1603 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1604     uint64_t *retinfo)
1605 {
1606 	enum exc_class exc1, exc2;
1607 	int type1, vector1;
1608 
1609 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1610 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1611 
1612 	/*
1613 	 * If an exception occurs while attempting to call the double-fault
1614 	 * handler the processor enters shutdown mode (aka triple fault).
1615 	 */
1616 	type1 = info1 & VM_INTINFO_TYPE;
1617 	vector1 = info1 & 0xff;
1618 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1619 		VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1620 		    info1, info2);
1621 		vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1622 		*retinfo = 0;
1623 		return (0);
1624 	}
1625 
1626 	/*
1627 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1628 	 */
1629 	exc1 = exception_class(info1);
1630 	exc2 = exception_class(info2);
1631 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1632 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1633 		/* Convert nested fault into a double fault. */
1634 		*retinfo = IDT_DF;
1635 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1636 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
1637 	} else {
1638 		/* Handle exceptions serially */
1639 		*retinfo = info2;
1640 	}
1641 	return (1);
1642 }
1643 
1644 static uint64_t
1645 vcpu_exception_intinfo(struct vcpu *vcpu)
1646 {
1647 	uint64_t info = 0;
1648 
1649 	if (vcpu->exception_pending) {
1650 		info = vcpu->exception.vector & 0xff;
1651 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1652 		if (vcpu->exception.error_code_valid) {
1653 			info |= VM_INTINFO_DEL_ERRCODE;
1654 			info |= (uint64_t)vcpu->exception.error_code << 32;
1655 		}
1656 	}
1657 	return (info);
1658 }
1659 
1660 int
1661 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1662 {
1663 	struct vcpu *vcpu;
1664 	uint64_t info1, info2;
1665 	int valid;
1666 
1667 	KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid));
1668 
1669 	vcpu = &vm->vcpu[vcpuid];
1670 
1671 	info1 = vcpu->exitintinfo;
1672 	vcpu->exitintinfo = 0;
1673 
1674 	info2 = 0;
1675 	if (vcpu->exception_pending) {
1676 		info2 = vcpu_exception_intinfo(vcpu);
1677 		vcpu->exception_pending = 0;
1678 		VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1679 		    vcpu->exception.vector, info2);
1680 	}
1681 
1682 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
1683 		valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
1684 	} else if (info1 & VM_INTINFO_VALID) {
1685 		*retinfo = info1;
1686 		valid = 1;
1687 	} else if (info2 & VM_INTINFO_VALID) {
1688 		*retinfo = info2;
1689 		valid = 1;
1690 	} else {
1691 		valid = 0;
1692 	}
1693 
1694 	if (valid) {
1695 		VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
1696 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
1697 	}
1698 
1699 	return (valid);
1700 }
1701 
1702 int
1703 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
1704 {
1705 	struct vcpu *vcpu;
1706 
1707 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1708 		return (EINVAL);
1709 
1710 	vcpu = &vm->vcpu[vcpuid];
1711 	*info1 = vcpu->exitintinfo;
1712 	*info2 = vcpu_exception_intinfo(vcpu);
1713 	return (0);
1714 }
1715 
1716 int
1717 vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *exception)
1718 {
1719 	struct vcpu *vcpu;
1720 
1721 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1722 		return (EINVAL);
1723 
1724 	if (exception->vector < 0 || exception->vector >= 32)
1725 		return (EINVAL);
1726 
1727 	/*
1728 	 * A double fault exception should never be injected directly into
1729 	 * the guest. It is a derived exception that results from specific
1730 	 * combinations of nested faults.
1731 	 */
1732 	if (exception->vector == IDT_DF)
1733 		return (EINVAL);
1734 
1735 	vcpu = &vm->vcpu[vcpuid];
1736 
1737 	if (vcpu->exception_pending) {
1738 		VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
1739 		    "pending exception %d", exception->vector,
1740 		    vcpu->exception.vector);
1741 		return (EBUSY);
1742 	}
1743 
1744 	vcpu->exception_pending = 1;
1745 	vcpu->exception = *exception;
1746 	VCPU_CTR1(vm, vcpuid, "Exception %d pending", exception->vector);
1747 	return (0);
1748 }
1749 
1750 void
1751 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
1752     int errcode)
1753 {
1754 	struct vm_exception exception;
1755 	struct vm_exit *vmexit;
1756 	struct vm *vm;
1757 	int error;
1758 
1759 	vm = vmarg;
1760 
1761 	exception.vector = vector;
1762 	exception.error_code = errcode;
1763 	exception.error_code_valid = errcode_valid;
1764 	error = vm_inject_exception(vm, vcpuid, &exception);
1765 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
1766 
1767 	/*
1768 	 * A fault-like exception allows the instruction to be restarted
1769 	 * after the exception handler returns.
1770 	 *
1771 	 * By setting the inst_length to 0 we ensure that the instruction
1772 	 * pointer remains at the faulting instruction.
1773 	 */
1774 	vmexit = vm_exitinfo(vm, vcpuid);
1775 	vmexit->inst_length = 0;
1776 }
1777 
1778 void
1779 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
1780 {
1781 	struct vm *vm;
1782 	int error;
1783 
1784 	vm = vmarg;
1785 	VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
1786 	    error_code, cr2);
1787 
1788 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
1789 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
1790 
1791 	vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
1792 }
1793 
1794 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
1795 
1796 int
1797 vm_inject_nmi(struct vm *vm, int vcpuid)
1798 {
1799 	struct vcpu *vcpu;
1800 
1801 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1802 		return (EINVAL);
1803 
1804 	vcpu = &vm->vcpu[vcpuid];
1805 
1806 	vcpu->nmi_pending = 1;
1807 	vcpu_notify_event(vm, vcpuid, false);
1808 	return (0);
1809 }
1810 
1811 int
1812 vm_nmi_pending(struct vm *vm, int vcpuid)
1813 {
1814 	struct vcpu *vcpu;
1815 
1816 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1817 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1818 
1819 	vcpu = &vm->vcpu[vcpuid];
1820 
1821 	return (vcpu->nmi_pending);
1822 }
1823 
1824 void
1825 vm_nmi_clear(struct vm *vm, int vcpuid)
1826 {
1827 	struct vcpu *vcpu;
1828 
1829 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1830 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1831 
1832 	vcpu = &vm->vcpu[vcpuid];
1833 
1834 	if (vcpu->nmi_pending == 0)
1835 		panic("vm_nmi_clear: inconsistent nmi_pending state");
1836 
1837 	vcpu->nmi_pending = 0;
1838 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
1839 }
1840 
1841 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
1842 
1843 int
1844 vm_inject_extint(struct vm *vm, int vcpuid)
1845 {
1846 	struct vcpu *vcpu;
1847 
1848 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1849 		return (EINVAL);
1850 
1851 	vcpu = &vm->vcpu[vcpuid];
1852 
1853 	vcpu->extint_pending = 1;
1854 	vcpu_notify_event(vm, vcpuid, false);
1855 	return (0);
1856 }
1857 
1858 int
1859 vm_extint_pending(struct vm *vm, int vcpuid)
1860 {
1861 	struct vcpu *vcpu;
1862 
1863 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1864 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1865 
1866 	vcpu = &vm->vcpu[vcpuid];
1867 
1868 	return (vcpu->extint_pending);
1869 }
1870 
1871 void
1872 vm_extint_clear(struct vm *vm, int vcpuid)
1873 {
1874 	struct vcpu *vcpu;
1875 
1876 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1877 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1878 
1879 	vcpu = &vm->vcpu[vcpuid];
1880 
1881 	if (vcpu->extint_pending == 0)
1882 		panic("vm_extint_clear: inconsistent extint_pending state");
1883 
1884 	vcpu->extint_pending = 0;
1885 	vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
1886 }
1887 
1888 int
1889 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
1890 {
1891 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1892 		return (EINVAL);
1893 
1894 	if (type < 0 || type >= VM_CAP_MAX)
1895 		return (EINVAL);
1896 
1897 	return (VMGETCAP(vm->cookie, vcpu, type, retval));
1898 }
1899 
1900 int
1901 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
1902 {
1903 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1904 		return (EINVAL);
1905 
1906 	if (type < 0 || type >= VM_CAP_MAX)
1907 		return (EINVAL);
1908 
1909 	return (VMSETCAP(vm->cookie, vcpu, type, val));
1910 }
1911 
1912 struct vlapic *
1913 vm_lapic(struct vm *vm, int cpu)
1914 {
1915 	return (vm->vcpu[cpu].vlapic);
1916 }
1917 
1918 struct vioapic *
1919 vm_ioapic(struct vm *vm)
1920 {
1921 
1922 	return (vm->vioapic);
1923 }
1924 
1925 struct vhpet *
1926 vm_hpet(struct vm *vm)
1927 {
1928 
1929 	return (vm->vhpet);
1930 }
1931 
1932 boolean_t
1933 vmm_is_pptdev(int bus, int slot, int func)
1934 {
1935 	int found, i, n;
1936 	int b, s, f;
1937 	char *val, *cp, *cp2;
1938 
1939 	/*
1940 	 * XXX
1941 	 * The length of an environment variable is limited to 128 bytes which
1942 	 * puts an upper limit on the number of passthru devices that may be
1943 	 * specified using a single environment variable.
1944 	 *
1945 	 * Work around this by scanning multiple environment variable
1946 	 * names instead of a single one - yuck!
1947 	 */
1948 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
1949 
1950 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
1951 	found = 0;
1952 	for (i = 0; names[i] != NULL && !found; i++) {
1953 		cp = val = kern_getenv(names[i]);
1954 		while (cp != NULL && *cp != '\0') {
1955 			if ((cp2 = strchr(cp, ' ')) != NULL)
1956 				*cp2 = '\0';
1957 
1958 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
1959 			if (n == 3 && bus == b && slot == s && func == f) {
1960 				found = 1;
1961 				break;
1962 			}
1963 
1964 			if (cp2 != NULL)
1965 				*cp2++ = ' ';
1966 
1967 			cp = cp2;
1968 		}
1969 		freeenv(val);
1970 	}
1971 	return (found);
1972 }
1973 
1974 void *
1975 vm_iommu_domain(struct vm *vm)
1976 {
1977 
1978 	return (vm->iommu);
1979 }
1980 
1981 int
1982 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1983     bool from_idle)
1984 {
1985 	int error;
1986 	struct vcpu *vcpu;
1987 
1988 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1989 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
1990 
1991 	vcpu = &vm->vcpu[vcpuid];
1992 
1993 	vcpu_lock(vcpu);
1994 	error = vcpu_set_state_locked(vcpu, newstate, from_idle);
1995 	vcpu_unlock(vcpu);
1996 
1997 	return (error);
1998 }
1999 
2000 enum vcpu_state
2001 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2002 {
2003 	struct vcpu *vcpu;
2004 	enum vcpu_state state;
2005 
2006 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2007 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2008 
2009 	vcpu = &vm->vcpu[vcpuid];
2010 
2011 	vcpu_lock(vcpu);
2012 	state = vcpu->state;
2013 	if (hostcpu != NULL)
2014 		*hostcpu = vcpu->hostcpu;
2015 	vcpu_unlock(vcpu);
2016 
2017 	return (state);
2018 }
2019 
2020 int
2021 vm_activate_cpu(struct vm *vm, int vcpuid)
2022 {
2023 
2024 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2025 		return (EINVAL);
2026 
2027 	if (CPU_ISSET(vcpuid, &vm->active_cpus))
2028 		return (EBUSY);
2029 
2030 	VCPU_CTR0(vm, vcpuid, "activated");
2031 	CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2032 	return (0);
2033 }
2034 
2035 cpuset_t
2036 vm_active_cpus(struct vm *vm)
2037 {
2038 
2039 	return (vm->active_cpus);
2040 }
2041 
2042 cpuset_t
2043 vm_suspended_cpus(struct vm *vm)
2044 {
2045 
2046 	return (vm->suspended_cpus);
2047 }
2048 
2049 void *
2050 vcpu_stats(struct vm *vm, int vcpuid)
2051 {
2052 
2053 	return (vm->vcpu[vcpuid].stats);
2054 }
2055 
2056 int
2057 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2058 {
2059 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2060 		return (EINVAL);
2061 
2062 	*state = vm->vcpu[vcpuid].x2apic_state;
2063 
2064 	return (0);
2065 }
2066 
2067 int
2068 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2069 {
2070 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2071 		return (EINVAL);
2072 
2073 	if (state >= X2APIC_STATE_LAST)
2074 		return (EINVAL);
2075 
2076 	vm->vcpu[vcpuid].x2apic_state = state;
2077 
2078 	vlapic_set_x2apic_state(vm, vcpuid, state);
2079 
2080 	return (0);
2081 }
2082 
2083 /*
2084  * This function is called to ensure that a vcpu "sees" a pending event
2085  * as soon as possible:
2086  * - If the vcpu thread is sleeping then it is woken up.
2087  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2088  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2089  */
2090 void
2091 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2092 {
2093 	int hostcpu;
2094 	struct vcpu *vcpu;
2095 
2096 	vcpu = &vm->vcpu[vcpuid];
2097 
2098 	vcpu_lock(vcpu);
2099 	hostcpu = vcpu->hostcpu;
2100 	if (vcpu->state == VCPU_RUNNING) {
2101 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2102 		if (hostcpu != curcpu) {
2103 			if (lapic_intr) {
2104 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2105 				    vmm_ipinum);
2106 			} else {
2107 				ipi_cpu(hostcpu, vmm_ipinum);
2108 			}
2109 		} else {
2110 			/*
2111 			 * If the 'vcpu' is running on 'curcpu' then it must
2112 			 * be sending a notification to itself (e.g. SELF_IPI).
2113 			 * The pending event will be picked up when the vcpu
2114 			 * transitions back to guest context.
2115 			 */
2116 		}
2117 	} else {
2118 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2119 		    "with hostcpu %d", vcpu->state, hostcpu));
2120 		if (vcpu->state == VCPU_SLEEPING)
2121 			wakeup_one(vcpu);
2122 	}
2123 	vcpu_unlock(vcpu);
2124 }
2125 
2126 struct vmspace *
2127 vm_get_vmspace(struct vm *vm)
2128 {
2129 
2130 	return (vm->vmspace);
2131 }
2132 
2133 int
2134 vm_apicid2vcpuid(struct vm *vm, int apicid)
2135 {
2136 	/*
2137 	 * XXX apic id is assumed to be numerically identical to vcpu id
2138 	 */
2139 	return (apicid);
2140 }
2141 
2142 void
2143 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2144     vm_rendezvous_func_t func, void *arg)
2145 {
2146 	int i;
2147 
2148 	/*
2149 	 * Enforce that this function is called without any locks
2150 	 */
2151 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2152 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
2153 	    ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2154 
2155 restart:
2156 	mtx_lock(&vm->rendezvous_mtx);
2157 	if (vm->rendezvous_func != NULL) {
2158 		/*
2159 		 * If a rendezvous is already in progress then we need to
2160 		 * call the rendezvous handler in case this 'vcpuid' is one
2161 		 * of the targets of the rendezvous.
2162 		 */
2163 		RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2164 		mtx_unlock(&vm->rendezvous_mtx);
2165 		vm_handle_rendezvous(vm, vcpuid);
2166 		goto restart;
2167 	}
2168 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2169 	    "rendezvous is still in progress"));
2170 
2171 	RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2172 	vm->rendezvous_req_cpus = dest;
2173 	CPU_ZERO(&vm->rendezvous_done_cpus);
2174 	vm->rendezvous_arg = arg;
2175 	vm_set_rendezvous_func(vm, func);
2176 	mtx_unlock(&vm->rendezvous_mtx);
2177 
2178 	/*
2179 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2180 	 * vcpus so they handle the rendezvous as soon as possible.
2181 	 */
2182 	for (i = 0; i < VM_MAXCPU; i++) {
2183 		if (CPU_ISSET(i, &dest))
2184 			vcpu_notify_event(vm, i, false);
2185 	}
2186 
2187 	vm_handle_rendezvous(vm, vcpuid);
2188 }
2189 
2190 struct vatpic *
2191 vm_atpic(struct vm *vm)
2192 {
2193 	return (vm->vatpic);
2194 }
2195 
2196 struct vatpit *
2197 vm_atpit(struct vm *vm)
2198 {
2199 	return (vm->vatpit);
2200 }
2201 
2202 enum vm_reg_name
2203 vm_segment_name(int seg)
2204 {
2205 	static enum vm_reg_name seg_names[] = {
2206 		VM_REG_GUEST_ES,
2207 		VM_REG_GUEST_CS,
2208 		VM_REG_GUEST_SS,
2209 		VM_REG_GUEST_DS,
2210 		VM_REG_GUEST_FS,
2211 		VM_REG_GUEST_GS
2212 	};
2213 
2214 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2215 	    ("%s: invalid segment encoding %d", __func__, seg));
2216 	return (seg_names[seg]);
2217 }
2218 
2219 void
2220 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2221     int num_copyinfo)
2222 {
2223 	int idx;
2224 
2225 	for (idx = 0; idx < num_copyinfo; idx++) {
2226 		if (copyinfo[idx].cookie != NULL)
2227 			vm_gpa_release(copyinfo[idx].cookie);
2228 	}
2229 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2230 }
2231 
2232 int
2233 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2234     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2235     int num_copyinfo)
2236 {
2237 	int error, idx, nused;
2238 	size_t n, off, remaining;
2239 	void *hva, *cookie;
2240 	uint64_t gpa;
2241 
2242 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2243 
2244 	nused = 0;
2245 	remaining = len;
2246 	while (remaining > 0) {
2247 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2248 		error = vmm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa);
2249 		if (error)
2250 			return (error);
2251 		off = gpa & PAGE_MASK;
2252 		n = min(remaining, PAGE_SIZE - off);
2253 		copyinfo[nused].gpa = gpa;
2254 		copyinfo[nused].len = n;
2255 		remaining -= n;
2256 		gla += n;
2257 		nused++;
2258 	}
2259 
2260 	for (idx = 0; idx < nused; idx++) {
2261 		hva = vm_gpa_hold(vm, copyinfo[idx].gpa, copyinfo[idx].len,
2262 		    prot, &cookie);
2263 		if (hva == NULL)
2264 			break;
2265 		copyinfo[idx].hva = hva;
2266 		copyinfo[idx].cookie = cookie;
2267 	}
2268 
2269 	if (idx != nused) {
2270 		vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2271 		return (-1);
2272 	} else {
2273 		return (0);
2274 	}
2275 }
2276 
2277 void
2278 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2279     size_t len)
2280 {
2281 	char *dst;
2282 	int idx;
2283 
2284 	dst = kaddr;
2285 	idx = 0;
2286 	while (len > 0) {
2287 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2288 		len -= copyinfo[idx].len;
2289 		dst += copyinfo[idx].len;
2290 		idx++;
2291 	}
2292 }
2293 
2294 void
2295 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2296     struct vm_copyinfo *copyinfo, size_t len)
2297 {
2298 	const char *src;
2299 	int idx;
2300 
2301 	src = kaddr;
2302 	idx = 0;
2303 	while (len > 0) {
2304 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2305 		len -= copyinfo[idx].len;
2306 		src += copyinfo[idx].len;
2307 		idx++;
2308 	}
2309 }
2310 
2311 /*
2312  * Return the amount of in-use and wired memory for the VM. Since
2313  * these are global stats, only return the values with for vCPU 0
2314  */
2315 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2316 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2317 
2318 static void
2319 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2320 {
2321 
2322 	if (vcpu == 0) {
2323 		vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2324 	       	    PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2325 	}
2326 }
2327 
2328 static void
2329 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2330 {
2331 
2332 	if (vcpu == 0) {
2333 		vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2334 	      	    PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2335 	}
2336 }
2337 
2338 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2339 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2340