xref: /freebsd/sys/amd64/vmm/vmm.c (revision 40a8ac8f62b535d30349faf28cf47106b7041b83)
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;
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 	vie_init(vie);
1249 
1250 	/* Fetch, decode and emulate the faulting instruction */
1251 	error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip,
1252 	    vme->inst_length, vie);
1253 	if (error == 1)
1254 		return (0);		/* Resume guest to handle page fault */
1255 	else if (error == -1)
1256 		return (EFAULT);
1257 	else if (error != 0)
1258 		panic("%s: vmm_fetch_instruction error %d", __func__, error);
1259 
1260 	if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0)
1261 		return (EFAULT);
1262 
1263 	/* return to userland unless this is an in-kernel emulated device */
1264 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1265 		mread = lapic_mmio_read;
1266 		mwrite = lapic_mmio_write;
1267 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1268 		mread = vioapic_mmio_read;
1269 		mwrite = vioapic_mmio_write;
1270 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1271 		mread = vhpet_mmio_read;
1272 		mwrite = vhpet_mmio_write;
1273 	} else {
1274 		*retu = true;
1275 		return (0);
1276 	}
1277 
1278 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1279 	    mread, mwrite, retu);
1280 
1281 	return (error);
1282 }
1283 
1284 static int
1285 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1286 {
1287 	int i, done;
1288 	struct vcpu *vcpu;
1289 
1290 	done = 0;
1291 	vcpu = &vm->vcpu[vcpuid];
1292 
1293 	CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1294 
1295 	/*
1296 	 * Wait until all 'active_cpus' have suspended themselves.
1297 	 *
1298 	 * Since a VM may be suspended at any time including when one or
1299 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1300 	 * handler while we are waiting to prevent a deadlock.
1301 	 */
1302 	vcpu_lock(vcpu);
1303 	while (1) {
1304 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1305 			VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1306 			break;
1307 		}
1308 
1309 		if (vm->rendezvous_func == NULL) {
1310 			VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1311 			vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1312 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1313 			vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1314 		} else {
1315 			VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1316 			vcpu_unlock(vcpu);
1317 			vm_handle_rendezvous(vm, vcpuid);
1318 			vcpu_lock(vcpu);
1319 		}
1320 	}
1321 	vcpu_unlock(vcpu);
1322 
1323 	/*
1324 	 * Wakeup the other sleeping vcpus and return to userspace.
1325 	 */
1326 	for (i = 0; i < VM_MAXCPU; i++) {
1327 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1328 			vcpu_notify_event(vm, i, false);
1329 		}
1330 	}
1331 
1332 	*retu = true;
1333 	return (0);
1334 }
1335 
1336 int
1337 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1338 {
1339 	int i;
1340 
1341 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1342 		return (EINVAL);
1343 
1344 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1345 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1346 		    vm->suspend, how);
1347 		return (EALREADY);
1348 	}
1349 
1350 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1351 
1352 	/*
1353 	 * Notify all active vcpus that they are now suspended.
1354 	 */
1355 	for (i = 0; i < VM_MAXCPU; i++) {
1356 		if (CPU_ISSET(i, &vm->active_cpus))
1357 			vcpu_notify_event(vm, i, false);
1358 	}
1359 
1360 	return (0);
1361 }
1362 
1363 void
1364 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1365 {
1366 	struct vm_exit *vmexit;
1367 
1368 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1369 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1370 
1371 	vmexit = vm_exitinfo(vm, vcpuid);
1372 	vmexit->rip = rip;
1373 	vmexit->inst_length = 0;
1374 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1375 	vmexit->u.suspended.how = vm->suspend;
1376 }
1377 
1378 void
1379 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1380 {
1381 	struct vm_exit *vmexit;
1382 
1383 	KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1384 
1385 	vmexit = vm_exitinfo(vm, vcpuid);
1386 	vmexit->rip = rip;
1387 	vmexit->inst_length = 0;
1388 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1389 	vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1390 }
1391 
1392 void
1393 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1394 {
1395 	struct vm_exit *vmexit;
1396 
1397 	vmexit = vm_exitinfo(vm, vcpuid);
1398 	vmexit->rip = rip;
1399 	vmexit->inst_length = 0;
1400 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1401 	vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1402 }
1403 
1404 int
1405 vm_run(struct vm *vm, struct vm_run *vmrun)
1406 {
1407 	int error, vcpuid;
1408 	struct vcpu *vcpu;
1409 	struct pcb *pcb;
1410 	uint64_t tscval, rip;
1411 	struct vm_exit *vme;
1412 	bool retu, intr_disabled;
1413 	pmap_t pmap;
1414 	void *rptr, *sptr;
1415 
1416 	vcpuid = vmrun->cpuid;
1417 
1418 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1419 		return (EINVAL);
1420 
1421 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1422 		return (EINVAL);
1423 
1424 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1425 		return (EINVAL);
1426 
1427 	rptr = &vm->rendezvous_func;
1428 	sptr = &vm->suspend;
1429 	pmap = vmspace_pmap(vm->vmspace);
1430 	vcpu = &vm->vcpu[vcpuid];
1431 	vme = &vcpu->exitinfo;
1432 	rip = vmrun->rip;
1433 restart:
1434 	critical_enter();
1435 
1436 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1437 	    ("vm_run: absurd pm_active"));
1438 
1439 	tscval = rdtsc();
1440 
1441 	pcb = PCPU_GET(curpcb);
1442 	set_pcb_flags(pcb, PCB_FULL_IRET);
1443 
1444 	restore_guest_fpustate(vcpu);
1445 
1446 	vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1447 	error = VMRUN(vm->cookie, vcpuid, rip, pmap, rptr, sptr);
1448 	vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1449 
1450 	save_guest_fpustate(vcpu);
1451 
1452 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1453 
1454 	critical_exit();
1455 
1456 	if (error == 0) {
1457 		retu = false;
1458 		switch (vme->exitcode) {
1459 		case VM_EXITCODE_SUSPENDED:
1460 			error = vm_handle_suspend(vm, vcpuid, &retu);
1461 			break;
1462 		case VM_EXITCODE_IOAPIC_EOI:
1463 			vioapic_process_eoi(vm, vcpuid,
1464 			    vme->u.ioapic_eoi.vector);
1465 			break;
1466 		case VM_EXITCODE_RENDEZVOUS:
1467 			vm_handle_rendezvous(vm, vcpuid);
1468 			error = 0;
1469 			break;
1470 		case VM_EXITCODE_HLT:
1471 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1472 			error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1473 			break;
1474 		case VM_EXITCODE_PAGING:
1475 			error = vm_handle_paging(vm, vcpuid, &retu);
1476 			break;
1477 		case VM_EXITCODE_INST_EMUL:
1478 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1479 			break;
1480 		case VM_EXITCODE_INOUT:
1481 		case VM_EXITCODE_INOUT_STR:
1482 			error = vm_handle_inout(vm, vcpuid, vme, &retu);
1483 			break;
1484 		default:
1485 			retu = true;	/* handled in userland */
1486 			break;
1487 		}
1488 	}
1489 
1490 	if (error == 0 && retu == false) {
1491 		rip = vme->rip + vme->inst_length;
1492 		goto restart;
1493 	}
1494 
1495 	/* copy the exit information */
1496 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1497 	return (error);
1498 }
1499 
1500 int
1501 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1502 {
1503 	struct vcpu *vcpu;
1504 	int type, vector;
1505 
1506 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1507 		return (EINVAL);
1508 
1509 	vcpu = &vm->vcpu[vcpuid];
1510 
1511 	if (info & VM_INTINFO_VALID) {
1512 		type = info & VM_INTINFO_TYPE;
1513 		vector = info & 0xff;
1514 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1515 			return (EINVAL);
1516 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1517 			return (EINVAL);
1518 		if (info & VM_INTINFO_RSVD)
1519 			return (EINVAL);
1520 	} else {
1521 		info = 0;
1522 	}
1523 	VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1524 	vcpu->exitintinfo = info;
1525 	return (0);
1526 }
1527 
1528 enum exc_class {
1529 	EXC_BENIGN,
1530 	EXC_CONTRIBUTORY,
1531 	EXC_PAGEFAULT
1532 };
1533 
1534 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
1535 
1536 static enum exc_class
1537 exception_class(uint64_t info)
1538 {
1539 	int type, vector;
1540 
1541 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1542 	type = info & VM_INTINFO_TYPE;
1543 	vector = info & 0xff;
1544 
1545 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1546 	switch (type) {
1547 	case VM_INTINFO_HWINTR:
1548 	case VM_INTINFO_SWINTR:
1549 	case VM_INTINFO_NMI:
1550 		return (EXC_BENIGN);
1551 	default:
1552 		/*
1553 		 * Hardware exception.
1554 		 *
1555 		 * SVM and VT-x use identical type values to represent NMI,
1556 		 * hardware interrupt and software interrupt.
1557 		 *
1558 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1559 		 * for exceptions except #BP and #OF. #BP and #OF use a type
1560 		 * value of '5' or '6'. Therefore we don't check for explicit
1561 		 * values of 'type' to classify 'intinfo' into a hardware
1562 		 * exception.
1563 		 */
1564 		break;
1565 	}
1566 
1567 	switch (vector) {
1568 	case IDT_PF:
1569 	case IDT_VE:
1570 		return (EXC_PAGEFAULT);
1571 	case IDT_DE:
1572 	case IDT_TS:
1573 	case IDT_NP:
1574 	case IDT_SS:
1575 	case IDT_GP:
1576 		return (EXC_CONTRIBUTORY);
1577 	default:
1578 		return (EXC_BENIGN);
1579 	}
1580 }
1581 
1582 static int
1583 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1584     uint64_t *retinfo)
1585 {
1586 	enum exc_class exc1, exc2;
1587 	int type1, vector1;
1588 
1589 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1590 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1591 
1592 	/*
1593 	 * If an exception occurs while attempting to call the double-fault
1594 	 * handler the processor enters shutdown mode (aka triple fault).
1595 	 */
1596 	type1 = info1 & VM_INTINFO_TYPE;
1597 	vector1 = info1 & 0xff;
1598 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1599 		VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1600 		    info1, info2);
1601 		vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1602 		*retinfo = 0;
1603 		return (0);
1604 	}
1605 
1606 	/*
1607 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1608 	 */
1609 	exc1 = exception_class(info1);
1610 	exc2 = exception_class(info2);
1611 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1612 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1613 		/* Convert nested fault into a double fault. */
1614 		*retinfo = IDT_DF;
1615 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1616 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
1617 	} else {
1618 		/* Handle exceptions serially */
1619 		*retinfo = info2;
1620 	}
1621 	return (1);
1622 }
1623 
1624 static uint64_t
1625 vcpu_exception_intinfo(struct vcpu *vcpu)
1626 {
1627 	uint64_t info = 0;
1628 
1629 	if (vcpu->exception_pending) {
1630 		info = vcpu->exception.vector & 0xff;
1631 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1632 		if (vcpu->exception.error_code_valid) {
1633 			info |= VM_INTINFO_DEL_ERRCODE;
1634 			info |= (uint64_t)vcpu->exception.error_code << 32;
1635 		}
1636 	}
1637 	return (info);
1638 }
1639 
1640 int
1641 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1642 {
1643 	struct vcpu *vcpu;
1644 	uint64_t info1, info2;
1645 	int valid;
1646 
1647 	KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid));
1648 
1649 	vcpu = &vm->vcpu[vcpuid];
1650 
1651 	info1 = vcpu->exitintinfo;
1652 	vcpu->exitintinfo = 0;
1653 
1654 	info2 = 0;
1655 	if (vcpu->exception_pending) {
1656 		info2 = vcpu_exception_intinfo(vcpu);
1657 		vcpu->exception_pending = 0;
1658 		VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1659 		    vcpu->exception.vector, info2);
1660 	}
1661 
1662 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
1663 		valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
1664 	} else if (info1 & VM_INTINFO_VALID) {
1665 		*retinfo = info1;
1666 		valid = 1;
1667 	} else if (info2 & VM_INTINFO_VALID) {
1668 		*retinfo = info2;
1669 		valid = 1;
1670 	} else {
1671 		valid = 0;
1672 	}
1673 
1674 	if (valid) {
1675 		VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
1676 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
1677 	}
1678 
1679 	return (valid);
1680 }
1681 
1682 int
1683 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
1684 {
1685 	struct vcpu *vcpu;
1686 
1687 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1688 		return (EINVAL);
1689 
1690 	vcpu = &vm->vcpu[vcpuid];
1691 	*info1 = vcpu->exitintinfo;
1692 	*info2 = vcpu_exception_intinfo(vcpu);
1693 	return (0);
1694 }
1695 
1696 int
1697 vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *exception)
1698 {
1699 	struct vcpu *vcpu;
1700 
1701 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1702 		return (EINVAL);
1703 
1704 	if (exception->vector < 0 || exception->vector >= 32)
1705 		return (EINVAL);
1706 
1707 	/*
1708 	 * A double fault exception should never be injected directly into
1709 	 * the guest. It is a derived exception that results from specific
1710 	 * combinations of nested faults.
1711 	 */
1712 	if (exception->vector == IDT_DF)
1713 		return (EINVAL);
1714 
1715 	vcpu = &vm->vcpu[vcpuid];
1716 
1717 	if (vcpu->exception_pending) {
1718 		VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
1719 		    "pending exception %d", exception->vector,
1720 		    vcpu->exception.vector);
1721 		return (EBUSY);
1722 	}
1723 
1724 	vcpu->exception_pending = 1;
1725 	vcpu->exception = *exception;
1726 	VCPU_CTR1(vm, vcpuid, "Exception %d pending", exception->vector);
1727 	return (0);
1728 }
1729 
1730 void
1731 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
1732     int errcode)
1733 {
1734 	struct vm_exception exception;
1735 	struct vm_exit *vmexit;
1736 	struct vm *vm;
1737 	int error;
1738 
1739 	vm = vmarg;
1740 
1741 	exception.vector = vector;
1742 	exception.error_code = errcode;
1743 	exception.error_code_valid = errcode_valid;
1744 	error = vm_inject_exception(vm, vcpuid, &exception);
1745 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
1746 
1747 	/*
1748 	 * A fault-like exception allows the instruction to be restarted
1749 	 * after the exception handler returns.
1750 	 *
1751 	 * By setting the inst_length to 0 we ensure that the instruction
1752 	 * pointer remains at the faulting instruction.
1753 	 */
1754 	vmexit = vm_exitinfo(vm, vcpuid);
1755 	vmexit->inst_length = 0;
1756 }
1757 
1758 void
1759 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
1760 {
1761 	struct vm *vm;
1762 	int error;
1763 
1764 	vm = vmarg;
1765 	VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
1766 	    error_code, cr2);
1767 
1768 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
1769 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
1770 
1771 	vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
1772 }
1773 
1774 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
1775 
1776 int
1777 vm_inject_nmi(struct vm *vm, int vcpuid)
1778 {
1779 	struct vcpu *vcpu;
1780 
1781 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1782 		return (EINVAL);
1783 
1784 	vcpu = &vm->vcpu[vcpuid];
1785 
1786 	vcpu->nmi_pending = 1;
1787 	vcpu_notify_event(vm, vcpuid, false);
1788 	return (0);
1789 }
1790 
1791 int
1792 vm_nmi_pending(struct vm *vm, int vcpuid)
1793 {
1794 	struct vcpu *vcpu;
1795 
1796 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1797 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1798 
1799 	vcpu = &vm->vcpu[vcpuid];
1800 
1801 	return (vcpu->nmi_pending);
1802 }
1803 
1804 void
1805 vm_nmi_clear(struct vm *vm, int vcpuid)
1806 {
1807 	struct vcpu *vcpu;
1808 
1809 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1810 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1811 
1812 	vcpu = &vm->vcpu[vcpuid];
1813 
1814 	if (vcpu->nmi_pending == 0)
1815 		panic("vm_nmi_clear: inconsistent nmi_pending state");
1816 
1817 	vcpu->nmi_pending = 0;
1818 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
1819 }
1820 
1821 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
1822 
1823 int
1824 vm_inject_extint(struct vm *vm, int vcpuid)
1825 {
1826 	struct vcpu *vcpu;
1827 
1828 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1829 		return (EINVAL);
1830 
1831 	vcpu = &vm->vcpu[vcpuid];
1832 
1833 	vcpu->extint_pending = 1;
1834 	vcpu_notify_event(vm, vcpuid, false);
1835 	return (0);
1836 }
1837 
1838 int
1839 vm_extint_pending(struct vm *vm, int vcpuid)
1840 {
1841 	struct vcpu *vcpu;
1842 
1843 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1844 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1845 
1846 	vcpu = &vm->vcpu[vcpuid];
1847 
1848 	return (vcpu->extint_pending);
1849 }
1850 
1851 void
1852 vm_extint_clear(struct vm *vm, int vcpuid)
1853 {
1854 	struct vcpu *vcpu;
1855 
1856 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1857 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1858 
1859 	vcpu = &vm->vcpu[vcpuid];
1860 
1861 	if (vcpu->extint_pending == 0)
1862 		panic("vm_extint_clear: inconsistent extint_pending state");
1863 
1864 	vcpu->extint_pending = 0;
1865 	vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
1866 }
1867 
1868 int
1869 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
1870 {
1871 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1872 		return (EINVAL);
1873 
1874 	if (type < 0 || type >= VM_CAP_MAX)
1875 		return (EINVAL);
1876 
1877 	return (VMGETCAP(vm->cookie, vcpu, type, retval));
1878 }
1879 
1880 int
1881 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
1882 {
1883 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1884 		return (EINVAL);
1885 
1886 	if (type < 0 || type >= VM_CAP_MAX)
1887 		return (EINVAL);
1888 
1889 	return (VMSETCAP(vm->cookie, vcpu, type, val));
1890 }
1891 
1892 struct vlapic *
1893 vm_lapic(struct vm *vm, int cpu)
1894 {
1895 	return (vm->vcpu[cpu].vlapic);
1896 }
1897 
1898 struct vioapic *
1899 vm_ioapic(struct vm *vm)
1900 {
1901 
1902 	return (vm->vioapic);
1903 }
1904 
1905 struct vhpet *
1906 vm_hpet(struct vm *vm)
1907 {
1908 
1909 	return (vm->vhpet);
1910 }
1911 
1912 boolean_t
1913 vmm_is_pptdev(int bus, int slot, int func)
1914 {
1915 	int found, i, n;
1916 	int b, s, f;
1917 	char *val, *cp, *cp2;
1918 
1919 	/*
1920 	 * XXX
1921 	 * The length of an environment variable is limited to 128 bytes which
1922 	 * puts an upper limit on the number of passthru devices that may be
1923 	 * specified using a single environment variable.
1924 	 *
1925 	 * Work around this by scanning multiple environment variable
1926 	 * names instead of a single one - yuck!
1927 	 */
1928 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
1929 
1930 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
1931 	found = 0;
1932 	for (i = 0; names[i] != NULL && !found; i++) {
1933 		cp = val = getenv(names[i]);
1934 		while (cp != NULL && *cp != '\0') {
1935 			if ((cp2 = strchr(cp, ' ')) != NULL)
1936 				*cp2 = '\0';
1937 
1938 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
1939 			if (n == 3 && bus == b && slot == s && func == f) {
1940 				found = 1;
1941 				break;
1942 			}
1943 
1944 			if (cp2 != NULL)
1945 				*cp2++ = ' ';
1946 
1947 			cp = cp2;
1948 		}
1949 		freeenv(val);
1950 	}
1951 	return (found);
1952 }
1953 
1954 void *
1955 vm_iommu_domain(struct vm *vm)
1956 {
1957 
1958 	return (vm->iommu);
1959 }
1960 
1961 int
1962 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1963     bool from_idle)
1964 {
1965 	int error;
1966 	struct vcpu *vcpu;
1967 
1968 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1969 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
1970 
1971 	vcpu = &vm->vcpu[vcpuid];
1972 
1973 	vcpu_lock(vcpu);
1974 	error = vcpu_set_state_locked(vcpu, newstate, from_idle);
1975 	vcpu_unlock(vcpu);
1976 
1977 	return (error);
1978 }
1979 
1980 enum vcpu_state
1981 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
1982 {
1983 	struct vcpu *vcpu;
1984 	enum vcpu_state state;
1985 
1986 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1987 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
1988 
1989 	vcpu = &vm->vcpu[vcpuid];
1990 
1991 	vcpu_lock(vcpu);
1992 	state = vcpu->state;
1993 	if (hostcpu != NULL)
1994 		*hostcpu = vcpu->hostcpu;
1995 	vcpu_unlock(vcpu);
1996 
1997 	return (state);
1998 }
1999 
2000 int
2001 vm_activate_cpu(struct vm *vm, int vcpuid)
2002 {
2003 
2004 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2005 		return (EINVAL);
2006 
2007 	if (CPU_ISSET(vcpuid, &vm->active_cpus))
2008 		return (EBUSY);
2009 
2010 	VCPU_CTR0(vm, vcpuid, "activated");
2011 	CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2012 	return (0);
2013 }
2014 
2015 cpuset_t
2016 vm_active_cpus(struct vm *vm)
2017 {
2018 
2019 	return (vm->active_cpus);
2020 }
2021 
2022 cpuset_t
2023 vm_suspended_cpus(struct vm *vm)
2024 {
2025 
2026 	return (vm->suspended_cpus);
2027 }
2028 
2029 void *
2030 vcpu_stats(struct vm *vm, int vcpuid)
2031 {
2032 
2033 	return (vm->vcpu[vcpuid].stats);
2034 }
2035 
2036 int
2037 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2038 {
2039 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2040 		return (EINVAL);
2041 
2042 	*state = vm->vcpu[vcpuid].x2apic_state;
2043 
2044 	return (0);
2045 }
2046 
2047 int
2048 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2049 {
2050 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2051 		return (EINVAL);
2052 
2053 	if (state >= X2APIC_STATE_LAST)
2054 		return (EINVAL);
2055 
2056 	vm->vcpu[vcpuid].x2apic_state = state;
2057 
2058 	vlapic_set_x2apic_state(vm, vcpuid, state);
2059 
2060 	return (0);
2061 }
2062 
2063 /*
2064  * This function is called to ensure that a vcpu "sees" a pending event
2065  * as soon as possible:
2066  * - If the vcpu thread is sleeping then it is woken up.
2067  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2068  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2069  */
2070 void
2071 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2072 {
2073 	int hostcpu;
2074 	struct vcpu *vcpu;
2075 
2076 	vcpu = &vm->vcpu[vcpuid];
2077 
2078 	vcpu_lock(vcpu);
2079 	hostcpu = vcpu->hostcpu;
2080 	if (vcpu->state == VCPU_RUNNING) {
2081 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2082 		if (hostcpu != curcpu) {
2083 			if (lapic_intr) {
2084 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2085 				    vmm_ipinum);
2086 			} else {
2087 				ipi_cpu(hostcpu, vmm_ipinum);
2088 			}
2089 		} else {
2090 			/*
2091 			 * If the 'vcpu' is running on 'curcpu' then it must
2092 			 * be sending a notification to itself (e.g. SELF_IPI).
2093 			 * The pending event will be picked up when the vcpu
2094 			 * transitions back to guest context.
2095 			 */
2096 		}
2097 	} else {
2098 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2099 		    "with hostcpu %d", vcpu->state, hostcpu));
2100 		if (vcpu->state == VCPU_SLEEPING)
2101 			wakeup_one(vcpu);
2102 	}
2103 	vcpu_unlock(vcpu);
2104 }
2105 
2106 struct vmspace *
2107 vm_get_vmspace(struct vm *vm)
2108 {
2109 
2110 	return (vm->vmspace);
2111 }
2112 
2113 int
2114 vm_apicid2vcpuid(struct vm *vm, int apicid)
2115 {
2116 	/*
2117 	 * XXX apic id is assumed to be numerically identical to vcpu id
2118 	 */
2119 	return (apicid);
2120 }
2121 
2122 void
2123 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2124     vm_rendezvous_func_t func, void *arg)
2125 {
2126 	int i;
2127 
2128 	/*
2129 	 * Enforce that this function is called without any locks
2130 	 */
2131 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2132 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
2133 	    ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2134 
2135 restart:
2136 	mtx_lock(&vm->rendezvous_mtx);
2137 	if (vm->rendezvous_func != NULL) {
2138 		/*
2139 		 * If a rendezvous is already in progress then we need to
2140 		 * call the rendezvous handler in case this 'vcpuid' is one
2141 		 * of the targets of the rendezvous.
2142 		 */
2143 		RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2144 		mtx_unlock(&vm->rendezvous_mtx);
2145 		vm_handle_rendezvous(vm, vcpuid);
2146 		goto restart;
2147 	}
2148 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2149 	    "rendezvous is still in progress"));
2150 
2151 	RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2152 	vm->rendezvous_req_cpus = dest;
2153 	CPU_ZERO(&vm->rendezvous_done_cpus);
2154 	vm->rendezvous_arg = arg;
2155 	vm_set_rendezvous_func(vm, func);
2156 	mtx_unlock(&vm->rendezvous_mtx);
2157 
2158 	/*
2159 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2160 	 * vcpus so they handle the rendezvous as soon as possible.
2161 	 */
2162 	for (i = 0; i < VM_MAXCPU; i++) {
2163 		if (CPU_ISSET(i, &dest))
2164 			vcpu_notify_event(vm, i, false);
2165 	}
2166 
2167 	vm_handle_rendezvous(vm, vcpuid);
2168 }
2169 
2170 struct vatpic *
2171 vm_atpic(struct vm *vm)
2172 {
2173 	return (vm->vatpic);
2174 }
2175 
2176 struct vatpit *
2177 vm_atpit(struct vm *vm)
2178 {
2179 	return (vm->vatpit);
2180 }
2181 
2182 enum vm_reg_name
2183 vm_segment_name(int seg)
2184 {
2185 	static enum vm_reg_name seg_names[] = {
2186 		VM_REG_GUEST_ES,
2187 		VM_REG_GUEST_CS,
2188 		VM_REG_GUEST_SS,
2189 		VM_REG_GUEST_DS,
2190 		VM_REG_GUEST_FS,
2191 		VM_REG_GUEST_GS
2192 	};
2193 
2194 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2195 	    ("%s: invalid segment encoding %d", __func__, seg));
2196 	return (seg_names[seg]);
2197 }
2198 
2199 void
2200 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2201     int num_copyinfo)
2202 {
2203 	int idx;
2204 
2205 	for (idx = 0; idx < num_copyinfo; idx++) {
2206 		if (copyinfo[idx].cookie != NULL)
2207 			vm_gpa_release(copyinfo[idx].cookie);
2208 	}
2209 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2210 }
2211 
2212 int
2213 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2214     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2215     int num_copyinfo)
2216 {
2217 	int error, idx, nused;
2218 	size_t n, off, remaining;
2219 	void *hva, *cookie;
2220 	uint64_t gpa;
2221 
2222 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2223 
2224 	nused = 0;
2225 	remaining = len;
2226 	while (remaining > 0) {
2227 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2228 		error = vmm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa);
2229 		if (error)
2230 			return (error);
2231 		off = gpa & PAGE_MASK;
2232 		n = min(remaining, PAGE_SIZE - off);
2233 		copyinfo[nused].gpa = gpa;
2234 		copyinfo[nused].len = n;
2235 		remaining -= n;
2236 		gla += n;
2237 		nused++;
2238 	}
2239 
2240 	for (idx = 0; idx < nused; idx++) {
2241 		hva = vm_gpa_hold(vm, copyinfo[idx].gpa, copyinfo[idx].len,
2242 		    prot, &cookie);
2243 		if (hva == NULL)
2244 			break;
2245 		copyinfo[idx].hva = hva;
2246 		copyinfo[idx].cookie = cookie;
2247 	}
2248 
2249 	if (idx != nused) {
2250 		vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2251 		return (-1);
2252 	} else {
2253 		return (0);
2254 	}
2255 }
2256 
2257 void
2258 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2259     size_t len)
2260 {
2261 	char *dst;
2262 	int idx;
2263 
2264 	dst = kaddr;
2265 	idx = 0;
2266 	while (len > 0) {
2267 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2268 		len -= copyinfo[idx].len;
2269 		dst += copyinfo[idx].len;
2270 		idx++;
2271 	}
2272 }
2273 
2274 void
2275 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2276     struct vm_copyinfo *copyinfo, size_t len)
2277 {
2278 	const char *src;
2279 	int idx;
2280 
2281 	src = kaddr;
2282 	idx = 0;
2283 	while (len > 0) {
2284 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2285 		len -= copyinfo[idx].len;
2286 		src += copyinfo[idx].len;
2287 		idx++;
2288 	}
2289 }
2290 
2291 /*
2292  * Return the amount of in-use and wired memory for the VM. Since
2293  * these are global stats, only return the values with for vCPU 0
2294  */
2295 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2296 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2297 
2298 static void
2299 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2300 {
2301 
2302 	if (vcpu == 0) {
2303 		vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2304 	       	    PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2305 	}
2306 }
2307 
2308 static void
2309 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2310 {
2311 
2312 	if (vcpu == 0) {
2313 		vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2314 	      	    PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2315 	}
2316 }
2317 
2318 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2319 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2320