xref: /freebsd/sys/amd64/vmm/vmm.c (revision 6574b8ed19b093f0af09501d2c9676c28993cb97)
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_msr.h"
78 #include "vmm_ipi.h"
79 #include "vmm_stat.h"
80 #include "vmm_lapic.h"
81 
82 #include "io/ppt.h"
83 #include "io/iommu.h"
84 
85 struct vlapic;
86 
87 /*
88  * Initialization:
89  * (a) allocated when vcpu is created
90  * (i) initialized when vcpu is created and when it is reinitialized
91  * (o) initialized the first time the vcpu is created
92  * (x) initialized before use
93  */
94 struct vcpu {
95 	struct mtx 	mtx;		/* (o) protects 'state' and 'hostcpu' */
96 	enum vcpu_state	state;		/* (o) vcpu state */
97 	int		hostcpu;	/* (o) vcpu's host cpu */
98 	struct vlapic	*vlapic;	/* (i) APIC device model */
99 	enum x2apic_state x2apic_state;	/* (i) APIC mode */
100 	uint64_t	exitintinfo;	/* (i) events pending at VM exit */
101 	int		nmi_pending;	/* (i) NMI pending */
102 	int		extint_pending;	/* (i) INTR pending */
103 	struct vm_exception exception;	/* (x) exception collateral */
104 	int	exception_pending;	/* (i) exception pending */
105 	struct savefpu	*guestfpu;	/* (a,i) guest fpu state */
106 	uint64_t	guest_xcr0;	/* (i) guest %xcr0 register */
107 	void		*stats;		/* (a,i) statistics */
108 	uint64_t guest_msrs[VMM_MSR_NUM]; /* (i) emulated MSRs */
109 	struct vm_exit	exitinfo;	/* (x) exit reason and collateral */
110 };
111 
112 #define	vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
113 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
114 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
115 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
116 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
117 
118 struct mem_seg {
119 	vm_paddr_t	gpa;
120 	size_t		len;
121 	boolean_t	wired;
122 	vm_object_t	object;
123 };
124 #define	VM_MAX_MEMORY_SEGMENTS	2
125 
126 /*
127  * Initialization:
128  * (o) initialized the first time the VM is created
129  * (i) initialized when VM is created and when it is reinitialized
130  * (x) initialized before use
131  */
132 struct vm {
133 	void		*cookie;		/* (i) cpu-specific data */
134 	void		*iommu;			/* (x) iommu-specific data */
135 	struct vhpet	*vhpet;			/* (i) virtual HPET */
136 	struct vioapic	*vioapic;		/* (i) virtual ioapic */
137 	struct vatpic	*vatpic;		/* (i) virtual atpic */
138 	struct vatpit	*vatpit;		/* (i) virtual atpit */
139 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
140 	int		suspend;		/* (i) stop VM execution */
141 	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
142 	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
143 	cpuset_t	rendezvous_req_cpus;	/* (x) rendezvous requested */
144 	cpuset_t	rendezvous_done_cpus;	/* (x) rendezvous finished */
145 	void		*rendezvous_arg;	/* (x) rendezvous func/arg */
146 	vm_rendezvous_func_t rendezvous_func;
147 	struct mtx	rendezvous_mtx;		/* (o) rendezvous lock */
148 	int		num_mem_segs;		/* (o) guest memory segments */
149 	struct mem_seg	mem_segs[VM_MAX_MEMORY_SEGMENTS];
150 	struct vmspace	*vmspace;		/* (o) guest's address space */
151 	char		name[VM_MAX_NAMELEN];	/* (o) virtual machine name */
152 	struct vcpu	vcpu[VM_MAXCPU];	/* (i) guest vcpus */
153 };
154 
155 static int vmm_initialized;
156 
157 static struct vmm_ops *ops;
158 #define	VMM_INIT(num)	(ops != NULL ? (*ops->init)(num) : 0)
159 #define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
160 #define	VMM_RESUME()	(ops != NULL ? (*ops->resume)() : 0)
161 
162 #define	VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
163 #define	VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \
164 	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO)
165 #define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
166 #define	VMSPACE_ALLOC(min, max) \
167 	(ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
168 #define	VMSPACE_FREE(vmspace) \
169 	(ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
170 #define	VMGETREG(vmi, vcpu, num, retval)		\
171 	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
172 #define	VMSETREG(vmi, vcpu, num, val)		\
173 	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
174 #define	VMGETDESC(vmi, vcpu, num, desc)		\
175 	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
176 #define	VMSETDESC(vmi, vcpu, num, desc)		\
177 	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
178 #define	VMGETCAP(vmi, vcpu, num, retval)	\
179 	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
180 #define	VMSETCAP(vmi, vcpu, num, val)		\
181 	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
182 #define	VLAPIC_INIT(vmi, vcpu)			\
183 	(ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
184 #define	VLAPIC_CLEANUP(vmi, vlapic)		\
185 	(ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
186 
187 #define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
188 #define	fpu_stop_emulating()	clts()
189 
190 static MALLOC_DEFINE(M_VM, "vm", "vm");
191 CTASSERT(VMM_MSR_NUM <= 64);	/* msr_mask can keep track of up to 64 msrs */
192 
193 /* statistics */
194 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
195 
196 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
197 
198 /*
199  * Halt the guest if all vcpus are executing a HLT instruction with
200  * interrupts disabled.
201  */
202 static int halt_detection_enabled = 1;
203 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
204     &halt_detection_enabled, 0,
205     "Halt VM if all vcpus execute HLT with interrupts disabled");
206 
207 static int vmm_ipinum;
208 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
209     "IPI vector used for vcpu notifications");
210 
211 static void
212 vcpu_cleanup(struct vm *vm, int i, bool destroy)
213 {
214 	struct vcpu *vcpu = &vm->vcpu[i];
215 
216 	VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
217 	if (destroy) {
218 		vmm_stat_free(vcpu->stats);
219 		fpu_save_area_free(vcpu->guestfpu);
220 	}
221 }
222 
223 static void
224 vcpu_init(struct vm *vm, int vcpu_id, bool create)
225 {
226 	struct vcpu *vcpu;
227 
228 	KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU,
229 	    ("vcpu_init: invalid vcpu %d", vcpu_id));
230 
231 	vcpu = &vm->vcpu[vcpu_id];
232 
233 	if (create) {
234 		KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
235 		    "initialized", vcpu_id));
236 		vcpu_lock_init(vcpu);
237 		vcpu->state = VCPU_IDLE;
238 		vcpu->hostcpu = NOCPU;
239 		vcpu->guestfpu = fpu_save_area_alloc();
240 		vcpu->stats = vmm_stat_alloc();
241 	}
242 
243 	vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
244 	vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
245 	vcpu->exitintinfo = 0;
246 	vcpu->nmi_pending = 0;
247 	vcpu->extint_pending = 0;
248 	vcpu->exception_pending = 0;
249 	vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
250 	fpu_save_area_reset(vcpu->guestfpu);
251 	vmm_stat_init(vcpu->stats);
252 	guest_msrs_init(vm, vcpu_id);
253 }
254 
255 struct vm_exit *
256 vm_exitinfo(struct vm *vm, int cpuid)
257 {
258 	struct vcpu *vcpu;
259 
260 	if (cpuid < 0 || cpuid >= VM_MAXCPU)
261 		panic("vm_exitinfo: invalid cpuid %d", cpuid);
262 
263 	vcpu = &vm->vcpu[cpuid];
264 
265 	return (&vcpu->exitinfo);
266 }
267 
268 static void
269 vmm_resume(void)
270 {
271 	VMM_RESUME();
272 }
273 
274 static int
275 vmm_init(void)
276 {
277 	int error;
278 
279 	vmm_host_state_init();
280 
281 	vmm_ipinum = vmm_ipi_alloc();
282 	if (vmm_ipinum == 0)
283 		vmm_ipinum = IPI_AST;
284 
285 	error = vmm_mem_init();
286 	if (error)
287 		return (error);
288 
289 	if (vmm_is_intel())
290 		ops = &vmm_ops_intel;
291 	else if (vmm_is_amd())
292 		ops = &vmm_ops_amd;
293 	else
294 		return (ENXIO);
295 
296 	vmm_msr_init();
297 	vmm_resume_p = vmm_resume;
298 
299 	return (VMM_INIT(vmm_ipinum));
300 }
301 
302 static int
303 vmm_handler(module_t mod, int what, void *arg)
304 {
305 	int error;
306 
307 	switch (what) {
308 	case MOD_LOAD:
309 		vmmdev_init();
310 		if (ppt_avail_devices() > 0)
311 			iommu_init();
312 		error = vmm_init();
313 		if (error == 0)
314 			vmm_initialized = 1;
315 		break;
316 	case MOD_UNLOAD:
317 		error = vmmdev_cleanup();
318 		if (error == 0) {
319 			vmm_resume_p = NULL;
320 			iommu_cleanup();
321 			if (vmm_ipinum != IPI_AST)
322 				vmm_ipi_free(vmm_ipinum);
323 			error = VMM_CLEANUP();
324 			/*
325 			 * Something bad happened - prevent new
326 			 * VMs from being created
327 			 */
328 			if (error)
329 				vmm_initialized = 0;
330 		}
331 		break;
332 	default:
333 		error = 0;
334 		break;
335 	}
336 	return (error);
337 }
338 
339 static moduledata_t vmm_kmod = {
340 	"vmm",
341 	vmm_handler,
342 	NULL
343 };
344 
345 /*
346  * vmm initialization has the following dependencies:
347  *
348  * - iommu initialization must happen after the pci passthru driver has had
349  *   a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
350  *
351  * - VT-x initialization requires smp_rendezvous() and therefore must happen
352  *   after SMP is fully functional (after SI_SUB_SMP).
353  */
354 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
355 MODULE_VERSION(vmm, 1);
356 
357 static void
358 vm_init(struct vm *vm, bool create)
359 {
360 	int i;
361 
362 	vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
363 	vm->iommu = NULL;
364 	vm->vioapic = vioapic_init(vm);
365 	vm->vhpet = vhpet_init(vm);
366 	vm->vatpic = vatpic_init(vm);
367 	vm->vatpit = vatpit_init(vm);
368 
369 	CPU_ZERO(&vm->active_cpus);
370 
371 	vm->suspend = 0;
372 	CPU_ZERO(&vm->suspended_cpus);
373 
374 	for (i = 0; i < VM_MAXCPU; i++)
375 		vcpu_init(vm, i, create);
376 }
377 
378 int
379 vm_create(const char *name, struct vm **retvm)
380 {
381 	struct vm *vm;
382 	struct vmspace *vmspace;
383 
384 	/*
385 	 * If vmm.ko could not be successfully initialized then don't attempt
386 	 * to create the virtual machine.
387 	 */
388 	if (!vmm_initialized)
389 		return (ENXIO);
390 
391 	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
392 		return (EINVAL);
393 
394 	vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
395 	if (vmspace == NULL)
396 		return (ENOMEM);
397 
398 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
399 	strcpy(vm->name, name);
400 	vm->num_mem_segs = 0;
401 	vm->vmspace = vmspace;
402 	mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
403 
404 	vm_init(vm, true);
405 
406 	*retvm = vm;
407 	return (0);
408 }
409 
410 static void
411 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg)
412 {
413 
414 	if (seg->object != NULL)
415 		vmm_mem_free(vm->vmspace, seg->gpa, seg->len);
416 
417 	bzero(seg, sizeof(*seg));
418 }
419 
420 static void
421 vm_cleanup(struct vm *vm, bool destroy)
422 {
423 	int i;
424 
425 	ppt_unassign_all(vm);
426 
427 	if (vm->iommu != NULL)
428 		iommu_destroy_domain(vm->iommu);
429 
430 	vatpit_cleanup(vm->vatpit);
431 	vhpet_cleanup(vm->vhpet);
432 	vatpic_cleanup(vm->vatpic);
433 	vioapic_cleanup(vm->vioapic);
434 
435 	for (i = 0; i < VM_MAXCPU; i++)
436 		vcpu_cleanup(vm, i, destroy);
437 
438 	VMCLEANUP(vm->cookie);
439 
440 	if (destroy) {
441 		for (i = 0; i < vm->num_mem_segs; i++)
442 			vm_free_mem_seg(vm, &vm->mem_segs[i]);
443 
444 		vm->num_mem_segs = 0;
445 
446 		VMSPACE_FREE(vm->vmspace);
447 		vm->vmspace = NULL;
448 	}
449 }
450 
451 void
452 vm_destroy(struct vm *vm)
453 {
454 	vm_cleanup(vm, true);
455 	free(vm, M_VM);
456 }
457 
458 int
459 vm_reinit(struct vm *vm)
460 {
461 	int error;
462 
463 	/*
464 	 * A virtual machine can be reset only if all vcpus are suspended.
465 	 */
466 	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
467 		vm_cleanup(vm, false);
468 		vm_init(vm, false);
469 		error = 0;
470 	} else {
471 		error = EBUSY;
472 	}
473 
474 	return (error);
475 }
476 
477 const char *
478 vm_name(struct vm *vm)
479 {
480 	return (vm->name);
481 }
482 
483 int
484 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
485 {
486 	vm_object_t obj;
487 
488 	if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
489 		return (ENOMEM);
490 	else
491 		return (0);
492 }
493 
494 int
495 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
496 {
497 
498 	vmm_mmio_free(vm->vmspace, gpa, len);
499 	return (0);
500 }
501 
502 boolean_t
503 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa)
504 {
505 	int i;
506 	vm_paddr_t gpabase, gpalimit;
507 
508 	for (i = 0; i < vm->num_mem_segs; i++) {
509 		gpabase = vm->mem_segs[i].gpa;
510 		gpalimit = gpabase + vm->mem_segs[i].len;
511 		if (gpa >= gpabase && gpa < gpalimit)
512 			return (TRUE);		/* 'gpa' is regular memory */
513 	}
514 
515 	if (ppt_is_mmio(vm, gpa))
516 		return (TRUE);			/* 'gpa' is pci passthru mmio */
517 
518 	return (FALSE);
519 }
520 
521 int
522 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
523 {
524 	int available, allocated;
525 	struct mem_seg *seg;
526 	vm_object_t object;
527 	vm_paddr_t g;
528 
529 	if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
530 		return (EINVAL);
531 
532 	available = allocated = 0;
533 	g = gpa;
534 	while (g < gpa + len) {
535 		if (vm_mem_allocated(vm, g))
536 			allocated++;
537 		else
538 			available++;
539 
540 		g += PAGE_SIZE;
541 	}
542 
543 	/*
544 	 * If there are some allocated and some available pages in the address
545 	 * range then it is an error.
546 	 */
547 	if (allocated && available)
548 		return (EINVAL);
549 
550 	/*
551 	 * If the entire address range being requested has already been
552 	 * allocated then there isn't anything more to do.
553 	 */
554 	if (allocated && available == 0)
555 		return (0);
556 
557 	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
558 		return (E2BIG);
559 
560 	seg = &vm->mem_segs[vm->num_mem_segs];
561 
562 	if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL)
563 		return (ENOMEM);
564 
565 	seg->gpa = gpa;
566 	seg->len = len;
567 	seg->object = object;
568 	seg->wired = FALSE;
569 
570 	vm->num_mem_segs++;
571 
572 	return (0);
573 }
574 
575 static vm_paddr_t
576 vm_maxmem(struct vm *vm)
577 {
578 	int i;
579 	vm_paddr_t gpa, maxmem;
580 
581 	maxmem = 0;
582 	for (i = 0; i < vm->num_mem_segs; i++) {
583 		gpa = vm->mem_segs[i].gpa + vm->mem_segs[i].len;
584 		if (gpa > maxmem)
585 			maxmem = gpa;
586 	}
587 	return (maxmem);
588 }
589 
590 static void
591 vm_gpa_unwire(struct vm *vm)
592 {
593 	int i, rv;
594 	struct mem_seg *seg;
595 
596 	for (i = 0; i < vm->num_mem_segs; i++) {
597 		seg = &vm->mem_segs[i];
598 		if (!seg->wired)
599 			continue;
600 
601 		rv = vm_map_unwire(&vm->vmspace->vm_map,
602 				   seg->gpa, seg->gpa + seg->len,
603 				   VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
604 		KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment "
605 		    "%#lx/%ld could not be unwired: %d",
606 		    vm_name(vm), seg->gpa, seg->len, rv));
607 
608 		seg->wired = FALSE;
609 	}
610 }
611 
612 static int
613 vm_gpa_wire(struct vm *vm)
614 {
615 	int i, rv;
616 	struct mem_seg *seg;
617 
618 	for (i = 0; i < vm->num_mem_segs; i++) {
619 		seg = &vm->mem_segs[i];
620 		if (seg->wired)
621 			continue;
622 
623 		/* XXX rlimits? */
624 		rv = vm_map_wire(&vm->vmspace->vm_map,
625 				 seg->gpa, seg->gpa + seg->len,
626 				 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
627 		if (rv != KERN_SUCCESS)
628 			break;
629 
630 		seg->wired = TRUE;
631 	}
632 
633 	if (i < vm->num_mem_segs) {
634 		/*
635 		 * Undo the wiring before returning an error.
636 		 */
637 		vm_gpa_unwire(vm);
638 		return (EAGAIN);
639 	}
640 
641 	return (0);
642 }
643 
644 static void
645 vm_iommu_modify(struct vm *vm, boolean_t map)
646 {
647 	int i, sz;
648 	vm_paddr_t gpa, hpa;
649 	struct mem_seg *seg;
650 	void *vp, *cookie, *host_domain;
651 
652 	sz = PAGE_SIZE;
653 	host_domain = iommu_host_domain();
654 
655 	for (i = 0; i < vm->num_mem_segs; i++) {
656 		seg = &vm->mem_segs[i];
657 		KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired",
658 		    vm_name(vm), seg->gpa, seg->len));
659 
660 		gpa = seg->gpa;
661 		while (gpa < seg->gpa + seg->len) {
662 			vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE,
663 					 &cookie);
664 			KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
665 			    vm_name(vm), gpa));
666 
667 			vm_gpa_release(cookie);
668 
669 			hpa = DMAP_TO_PHYS((uintptr_t)vp);
670 			if (map) {
671 				iommu_create_mapping(vm->iommu, gpa, hpa, sz);
672 				iommu_remove_mapping(host_domain, hpa, sz);
673 			} else {
674 				iommu_remove_mapping(vm->iommu, gpa, sz);
675 				iommu_create_mapping(host_domain, hpa, hpa, sz);
676 			}
677 
678 			gpa += PAGE_SIZE;
679 		}
680 	}
681 
682 	/*
683 	 * Invalidate the cached translations associated with the domain
684 	 * from which pages were removed.
685 	 */
686 	if (map)
687 		iommu_invalidate_tlb(host_domain);
688 	else
689 		iommu_invalidate_tlb(vm->iommu);
690 }
691 
692 #define	vm_iommu_unmap(vm)	vm_iommu_modify((vm), FALSE)
693 #define	vm_iommu_map(vm)	vm_iommu_modify((vm), TRUE)
694 
695 int
696 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
697 {
698 	int error;
699 
700 	error = ppt_unassign_device(vm, bus, slot, func);
701 	if (error)
702 		return (error);
703 
704 	if (ppt_assigned_devices(vm) == 0) {
705 		vm_iommu_unmap(vm);
706 		vm_gpa_unwire(vm);
707 	}
708 	return (0);
709 }
710 
711 int
712 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
713 {
714 	int error;
715 	vm_paddr_t maxaddr;
716 
717 	/*
718 	 * Virtual machines with pci passthru devices get special treatment:
719 	 * - the guest physical memory is wired
720 	 * - the iommu is programmed to do the 'gpa' to 'hpa' translation
721 	 *
722 	 * We need to do this before the first pci passthru device is attached.
723 	 */
724 	if (ppt_assigned_devices(vm) == 0) {
725 		KASSERT(vm->iommu == NULL,
726 		    ("vm_assign_pptdev: iommu must be NULL"));
727 		maxaddr = vm_maxmem(vm);
728 		vm->iommu = iommu_create_domain(maxaddr);
729 
730 		error = vm_gpa_wire(vm);
731 		if (error)
732 			return (error);
733 
734 		vm_iommu_map(vm);
735 	}
736 
737 	error = ppt_assign_device(vm, bus, slot, func);
738 	return (error);
739 }
740 
741 void *
742 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
743 	    void **cookie)
744 {
745 	int count, pageoff;
746 	vm_page_t m;
747 
748 	pageoff = gpa & PAGE_MASK;
749 	if (len > PAGE_SIZE - pageoff)
750 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
751 
752 	count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
753 	    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
754 
755 	if (count == 1) {
756 		*cookie = m;
757 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
758 	} else {
759 		*cookie = NULL;
760 		return (NULL);
761 	}
762 }
763 
764 void
765 vm_gpa_release(void *cookie)
766 {
767 	vm_page_t m = cookie;
768 
769 	vm_page_lock(m);
770 	vm_page_unhold(m);
771 	vm_page_unlock(m);
772 }
773 
774 int
775 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
776 		  struct vm_memory_segment *seg)
777 {
778 	int i;
779 
780 	for (i = 0; i < vm->num_mem_segs; i++) {
781 		if (gpabase == vm->mem_segs[i].gpa) {
782 			seg->gpa = vm->mem_segs[i].gpa;
783 			seg->len = vm->mem_segs[i].len;
784 			seg->wired = vm->mem_segs[i].wired;
785 			return (0);
786 		}
787 	}
788 	return (-1);
789 }
790 
791 int
792 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len,
793 	      vm_offset_t *offset, struct vm_object **object)
794 {
795 	int i;
796 	size_t seg_len;
797 	vm_paddr_t seg_gpa;
798 	vm_object_t seg_obj;
799 
800 	for (i = 0; i < vm->num_mem_segs; i++) {
801 		if ((seg_obj = vm->mem_segs[i].object) == NULL)
802 			continue;
803 
804 		seg_gpa = vm->mem_segs[i].gpa;
805 		seg_len = vm->mem_segs[i].len;
806 
807 		if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) {
808 			*offset = gpa - seg_gpa;
809 			*object = seg_obj;
810 			vm_object_reference(seg_obj);
811 			return (0);
812 		}
813 	}
814 
815 	return (EINVAL);
816 }
817 
818 int
819 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
820 {
821 
822 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
823 		return (EINVAL);
824 
825 	if (reg >= VM_REG_LAST)
826 		return (EINVAL);
827 
828 	return (VMGETREG(vm->cookie, vcpu, reg, retval));
829 }
830 
831 int
832 vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
833 {
834 
835 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
836 		return (EINVAL);
837 
838 	if (reg >= VM_REG_LAST)
839 		return (EINVAL);
840 
841 	return (VMSETREG(vm->cookie, vcpu, reg, val));
842 }
843 
844 static boolean_t
845 is_descriptor_table(int reg)
846 {
847 
848 	switch (reg) {
849 	case VM_REG_GUEST_IDTR:
850 	case VM_REG_GUEST_GDTR:
851 		return (TRUE);
852 	default:
853 		return (FALSE);
854 	}
855 }
856 
857 static boolean_t
858 is_segment_register(int reg)
859 {
860 
861 	switch (reg) {
862 	case VM_REG_GUEST_ES:
863 	case VM_REG_GUEST_CS:
864 	case VM_REG_GUEST_SS:
865 	case VM_REG_GUEST_DS:
866 	case VM_REG_GUEST_FS:
867 	case VM_REG_GUEST_GS:
868 	case VM_REG_GUEST_TR:
869 	case VM_REG_GUEST_LDTR:
870 		return (TRUE);
871 	default:
872 		return (FALSE);
873 	}
874 }
875 
876 int
877 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
878 		struct seg_desc *desc)
879 {
880 
881 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
882 		return (EINVAL);
883 
884 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
885 		return (EINVAL);
886 
887 	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
888 }
889 
890 int
891 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
892 		struct seg_desc *desc)
893 {
894 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
895 		return (EINVAL);
896 
897 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
898 		return (EINVAL);
899 
900 	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
901 }
902 
903 static void
904 restore_guest_fpustate(struct vcpu *vcpu)
905 {
906 
907 	/* flush host state to the pcb */
908 	fpuexit(curthread);
909 
910 	/* restore guest FPU state */
911 	fpu_stop_emulating();
912 	fpurestore(vcpu->guestfpu);
913 
914 	/* restore guest XCR0 if XSAVE is enabled in the host */
915 	if (rcr4() & CR4_XSAVE)
916 		load_xcr(0, vcpu->guest_xcr0);
917 
918 	/*
919 	 * The FPU is now "dirty" with the guest's state so turn on emulation
920 	 * to trap any access to the FPU by the host.
921 	 */
922 	fpu_start_emulating();
923 }
924 
925 static void
926 save_guest_fpustate(struct vcpu *vcpu)
927 {
928 
929 	if ((rcr0() & CR0_TS) == 0)
930 		panic("fpu emulation not enabled in host!");
931 
932 	/* save guest XCR0 and restore host XCR0 */
933 	if (rcr4() & CR4_XSAVE) {
934 		vcpu->guest_xcr0 = rxcr(0);
935 		load_xcr(0, vmm_get_host_xcr0());
936 	}
937 
938 	/* save guest FPU state */
939 	fpu_stop_emulating();
940 	fpusave(vcpu->guestfpu);
941 	fpu_start_emulating();
942 }
943 
944 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
945 
946 static int
947 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
948     bool from_idle)
949 {
950 	int error;
951 
952 	vcpu_assert_locked(vcpu);
953 
954 	/*
955 	 * State transitions from the vmmdev_ioctl() must always begin from
956 	 * the VCPU_IDLE state. This guarantees that there is only a single
957 	 * ioctl() operating on a vcpu at any point.
958 	 */
959 	if (from_idle) {
960 		while (vcpu->state != VCPU_IDLE)
961 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
962 	} else {
963 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
964 		    "vcpu idle state"));
965 	}
966 
967 	if (vcpu->state == VCPU_RUNNING) {
968 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
969 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
970 	} else {
971 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
972 		    "vcpu that is not running", vcpu->hostcpu));
973 	}
974 
975 	/*
976 	 * The following state transitions are allowed:
977 	 * IDLE -> FROZEN -> IDLE
978 	 * FROZEN -> RUNNING -> FROZEN
979 	 * FROZEN -> SLEEPING -> FROZEN
980 	 */
981 	switch (vcpu->state) {
982 	case VCPU_IDLE:
983 	case VCPU_RUNNING:
984 	case VCPU_SLEEPING:
985 		error = (newstate != VCPU_FROZEN);
986 		break;
987 	case VCPU_FROZEN:
988 		error = (newstate == VCPU_FROZEN);
989 		break;
990 	default:
991 		error = 1;
992 		break;
993 	}
994 
995 	if (error)
996 		return (EBUSY);
997 
998 	vcpu->state = newstate;
999 	if (newstate == VCPU_RUNNING)
1000 		vcpu->hostcpu = curcpu;
1001 	else
1002 		vcpu->hostcpu = NOCPU;
1003 
1004 	if (newstate == VCPU_IDLE)
1005 		wakeup(&vcpu->state);
1006 
1007 	return (0);
1008 }
1009 
1010 static void
1011 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1012 {
1013 	int error;
1014 
1015 	if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1016 		panic("Error %d setting state to %d\n", error, newstate);
1017 }
1018 
1019 static void
1020 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
1021 {
1022 	int error;
1023 
1024 	if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
1025 		panic("Error %d setting state to %d", error, newstate);
1026 }
1027 
1028 static void
1029 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func)
1030 {
1031 
1032 	KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked"));
1033 
1034 	/*
1035 	 * Update 'rendezvous_func' and execute a write memory barrier to
1036 	 * ensure that it is visible across all host cpus. This is not needed
1037 	 * for correctness but it does ensure that all the vcpus will notice
1038 	 * that the rendezvous is requested immediately.
1039 	 */
1040 	vm->rendezvous_func = func;
1041 	wmb();
1042 }
1043 
1044 #define	RENDEZVOUS_CTR0(vm, vcpuid, fmt)				\
1045 	do {								\
1046 		if (vcpuid >= 0)					\
1047 			VCPU_CTR0(vm, vcpuid, fmt);			\
1048 		else							\
1049 			VM_CTR0(vm, fmt);				\
1050 	} while (0)
1051 
1052 static void
1053 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1054 {
1055 
1056 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
1057 	    ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1058 
1059 	mtx_lock(&vm->rendezvous_mtx);
1060 	while (vm->rendezvous_func != NULL) {
1061 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1062 		CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1063 
1064 		if (vcpuid != -1 &&
1065 		    CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1066 		    !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1067 			VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1068 			(*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1069 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1070 		}
1071 		if (CPU_CMP(&vm->rendezvous_req_cpus,
1072 		    &vm->rendezvous_done_cpus) == 0) {
1073 			VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1074 			vm_set_rendezvous_func(vm, NULL);
1075 			wakeup(&vm->rendezvous_func);
1076 			break;
1077 		}
1078 		RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1079 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1080 		    "vmrndv", 0);
1081 	}
1082 	mtx_unlock(&vm->rendezvous_mtx);
1083 }
1084 
1085 /*
1086  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1087  */
1088 static int
1089 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1090 {
1091 	struct vcpu *vcpu;
1092 	const char *wmesg;
1093 	int error, t, vcpu_halted, vm_halted;
1094 
1095 	KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1096 
1097 	vcpu = &vm->vcpu[vcpuid];
1098 	vcpu_halted = 0;
1099 	vm_halted = 0;
1100 
1101 	/*
1102 	 * The typical way to halt a cpu is to execute: "sti; hlt"
1103 	 *
1104 	 * STI sets RFLAGS.IF to enable interrupts. However, the processor
1105 	 * remains in an "interrupt shadow" for an additional instruction
1106 	 * following the STI. This guarantees that "sti; hlt" sequence is
1107 	 * atomic and a pending interrupt will be recognized after the HLT.
1108 	 *
1109 	 * After the HLT emulation is done the vcpu is no longer in an
1110 	 * interrupt shadow and a pending interrupt can be injected on
1111 	 * the next entry into the guest.
1112 	 */
1113 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
1114 	KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
1115 	    __func__, error));
1116 
1117 	vcpu_lock(vcpu);
1118 	while (1) {
1119 		/*
1120 		 * Do a final check for pending NMI or interrupts before
1121 		 * really putting this thread to sleep. Also check for
1122 		 * software events that would cause this vcpu to wakeup.
1123 		 *
1124 		 * These interrupts/events could have happened after the
1125 		 * vcpu returned from VMRUN() and before it acquired the
1126 		 * vcpu lock above.
1127 		 */
1128 		if (vm->rendezvous_func != NULL || vm->suspend)
1129 			break;
1130 		if (vm_nmi_pending(vm, vcpuid))
1131 			break;
1132 		if (!intr_disabled) {
1133 			if (vm_extint_pending(vm, vcpuid) ||
1134 			    vlapic_pending_intr(vcpu->vlapic, NULL)) {
1135 				break;
1136 			}
1137 		}
1138 
1139 		/* Don't go to sleep if the vcpu thread needs to yield */
1140 		if (vcpu_should_yield(vm, vcpuid))
1141 			break;
1142 
1143 		/*
1144 		 * Some Linux guests implement "halt" by having all vcpus
1145 		 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1146 		 * track of the vcpus that have entered this state. When all
1147 		 * vcpus enter the halted state the virtual machine is halted.
1148 		 */
1149 		if (intr_disabled) {
1150 			wmesg = "vmhalt";
1151 			VCPU_CTR0(vm, vcpuid, "Halted");
1152 			if (!vcpu_halted && halt_detection_enabled) {
1153 				vcpu_halted = 1;
1154 				CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1155 			}
1156 			if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1157 				vm_halted = 1;
1158 				break;
1159 			}
1160 		} else {
1161 			wmesg = "vmidle";
1162 		}
1163 
1164 		t = ticks;
1165 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1166 		/*
1167 		 * XXX msleep_spin() cannot be interrupted by signals so
1168 		 * wake up periodically to check pending signals.
1169 		 */
1170 		msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1171 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1172 		vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1173 	}
1174 
1175 	if (vcpu_halted)
1176 		CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1177 
1178 	vcpu_unlock(vcpu);
1179 
1180 	if (vm_halted)
1181 		vm_suspend(vm, VM_SUSPEND_HALT);
1182 
1183 	return (0);
1184 }
1185 
1186 static int
1187 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1188 {
1189 	int rv, ftype;
1190 	struct vm_map *map;
1191 	struct vcpu *vcpu;
1192 	struct vm_exit *vme;
1193 
1194 	vcpu = &vm->vcpu[vcpuid];
1195 	vme = &vcpu->exitinfo;
1196 
1197 	ftype = vme->u.paging.fault_type;
1198 	KASSERT(ftype == VM_PROT_READ ||
1199 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1200 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1201 
1202 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1203 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1204 		    vme->u.paging.gpa, ftype);
1205 		if (rv == 0)
1206 			goto done;
1207 	}
1208 
1209 	map = &vm->vmspace->vm_map;
1210 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
1211 
1212 	VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1213 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1214 
1215 	if (rv != KERN_SUCCESS)
1216 		return (EFAULT);
1217 done:
1218 	/* restart execution at the faulting instruction */
1219 	vme->inst_length = 0;
1220 
1221 	return (0);
1222 }
1223 
1224 static int
1225 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1226 {
1227 	struct vie *vie;
1228 	struct vcpu *vcpu;
1229 	struct vm_exit *vme;
1230 	uint64_t gla, gpa;
1231 	struct vm_guest_paging *paging;
1232 	mem_region_read_t mread;
1233 	mem_region_write_t mwrite;
1234 	enum vm_cpu_mode cpu_mode;
1235 	int cs_d, error;
1236 
1237 	vcpu = &vm->vcpu[vcpuid];
1238 	vme = &vcpu->exitinfo;
1239 
1240 	gla = vme->u.inst_emul.gla;
1241 	gpa = vme->u.inst_emul.gpa;
1242 	cs_d = vme->u.inst_emul.cs_d;
1243 	vie = &vme->u.inst_emul.vie;
1244 	paging = &vme->u.inst_emul.paging;
1245 	cpu_mode = paging->cpu_mode;
1246 
1247 	vie_init(vie);
1248 
1249 	/* Fetch, decode and emulate the faulting instruction */
1250 	error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip,
1251 	    vme->inst_length, vie);
1252 	if (error == 1)
1253 		return (0);		/* Resume guest to handle page fault */
1254 	else if (error == -1)
1255 		return (EFAULT);
1256 	else if (error != 0)
1257 		panic("%s: vmm_fetch_instruction error %d", __func__, error);
1258 
1259 	if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0)
1260 		return (EFAULT);
1261 
1262 	/* return to userland unless this is an in-kernel emulated device */
1263 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1264 		mread = lapic_mmio_read;
1265 		mwrite = lapic_mmio_write;
1266 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1267 		mread = vioapic_mmio_read;
1268 		mwrite = vioapic_mmio_write;
1269 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1270 		mread = vhpet_mmio_read;
1271 		mwrite = vhpet_mmio_write;
1272 	} else {
1273 		*retu = true;
1274 		return (0);
1275 	}
1276 
1277 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1278 	    mread, mwrite, retu);
1279 
1280 	return (error);
1281 }
1282 
1283 static int
1284 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1285 {
1286 	int i, done;
1287 	struct vcpu *vcpu;
1288 
1289 	done = 0;
1290 	vcpu = &vm->vcpu[vcpuid];
1291 
1292 	CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1293 
1294 	/*
1295 	 * Wait until all 'active_cpus' have suspended themselves.
1296 	 *
1297 	 * Since a VM may be suspended at any time including when one or
1298 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1299 	 * handler while we are waiting to prevent a deadlock.
1300 	 */
1301 	vcpu_lock(vcpu);
1302 	while (1) {
1303 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1304 			VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1305 			break;
1306 		}
1307 
1308 		if (vm->rendezvous_func == NULL) {
1309 			VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1310 			vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1311 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1312 			vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1313 		} else {
1314 			VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1315 			vcpu_unlock(vcpu);
1316 			vm_handle_rendezvous(vm, vcpuid);
1317 			vcpu_lock(vcpu);
1318 		}
1319 	}
1320 	vcpu_unlock(vcpu);
1321 
1322 	/*
1323 	 * Wakeup the other sleeping vcpus and return to userspace.
1324 	 */
1325 	for (i = 0; i < VM_MAXCPU; i++) {
1326 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1327 			vcpu_notify_event(vm, i, false);
1328 		}
1329 	}
1330 
1331 	*retu = true;
1332 	return (0);
1333 }
1334 
1335 int
1336 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1337 {
1338 	int i;
1339 
1340 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1341 		return (EINVAL);
1342 
1343 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1344 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1345 		    vm->suspend, how);
1346 		return (EALREADY);
1347 	}
1348 
1349 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1350 
1351 	/*
1352 	 * Notify all active vcpus that they are now suspended.
1353 	 */
1354 	for (i = 0; i < VM_MAXCPU; i++) {
1355 		if (CPU_ISSET(i, &vm->active_cpus))
1356 			vcpu_notify_event(vm, i, false);
1357 	}
1358 
1359 	return (0);
1360 }
1361 
1362 void
1363 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1364 {
1365 	struct vm_exit *vmexit;
1366 
1367 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1368 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1369 
1370 	vmexit = vm_exitinfo(vm, vcpuid);
1371 	vmexit->rip = rip;
1372 	vmexit->inst_length = 0;
1373 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1374 	vmexit->u.suspended.how = vm->suspend;
1375 }
1376 
1377 void
1378 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1379 {
1380 	struct vm_exit *vmexit;
1381 
1382 	KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1383 
1384 	vmexit = vm_exitinfo(vm, vcpuid);
1385 	vmexit->rip = rip;
1386 	vmexit->inst_length = 0;
1387 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1388 	vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1389 }
1390 
1391 void
1392 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1393 {
1394 	struct vm_exit *vmexit;
1395 
1396 	vmexit = vm_exitinfo(vm, vcpuid);
1397 	vmexit->rip = rip;
1398 	vmexit->inst_length = 0;
1399 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1400 	vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1401 }
1402 
1403 int
1404 vm_run(struct vm *vm, struct vm_run *vmrun)
1405 {
1406 	int error, vcpuid;
1407 	struct vcpu *vcpu;
1408 	struct pcb *pcb;
1409 	uint64_t tscval, rip;
1410 	struct vm_exit *vme;
1411 	bool retu, intr_disabled;
1412 	pmap_t pmap;
1413 	void *rptr, *sptr;
1414 
1415 	vcpuid = vmrun->cpuid;
1416 
1417 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1418 		return (EINVAL);
1419 
1420 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1421 		return (EINVAL);
1422 
1423 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1424 		return (EINVAL);
1425 
1426 	rptr = &vm->rendezvous_func;
1427 	sptr = &vm->suspend;
1428 	pmap = vmspace_pmap(vm->vmspace);
1429 	vcpu = &vm->vcpu[vcpuid];
1430 	vme = &vcpu->exitinfo;
1431 	rip = vmrun->rip;
1432 restart:
1433 	critical_enter();
1434 
1435 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1436 	    ("vm_run: absurd pm_active"));
1437 
1438 	tscval = rdtsc();
1439 
1440 	pcb = PCPU_GET(curpcb);
1441 	set_pcb_flags(pcb, PCB_FULL_IRET);
1442 
1443 	restore_guest_msrs(vm, vcpuid);
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 	restore_host_msrs(vm, vcpuid);
1452 
1453 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1454 
1455 	critical_exit();
1456 
1457 	if (error == 0) {
1458 		retu = false;
1459 		switch (vme->exitcode) {
1460 		case VM_EXITCODE_SUSPENDED:
1461 			error = vm_handle_suspend(vm, vcpuid, &retu);
1462 			break;
1463 		case VM_EXITCODE_IOAPIC_EOI:
1464 			vioapic_process_eoi(vm, vcpuid,
1465 			    vme->u.ioapic_eoi.vector);
1466 			break;
1467 		case VM_EXITCODE_RENDEZVOUS:
1468 			vm_handle_rendezvous(vm, vcpuid);
1469 			error = 0;
1470 			break;
1471 		case VM_EXITCODE_HLT:
1472 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1473 			error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1474 			break;
1475 		case VM_EXITCODE_PAGING:
1476 			error = vm_handle_paging(vm, vcpuid, &retu);
1477 			break;
1478 		case VM_EXITCODE_INST_EMUL:
1479 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1480 			break;
1481 		case VM_EXITCODE_INOUT:
1482 		case VM_EXITCODE_INOUT_STR:
1483 			error = vm_handle_inout(vm, vcpuid, vme, &retu);
1484 			break;
1485 		default:
1486 			retu = true;	/* handled in userland */
1487 			break;
1488 		}
1489 	}
1490 
1491 	if (error == 0 && retu == false) {
1492 		rip = vme->rip + vme->inst_length;
1493 		goto restart;
1494 	}
1495 
1496 	/* copy the exit information */
1497 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1498 	return (error);
1499 }
1500 
1501 int
1502 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1503 {
1504 	struct vcpu *vcpu;
1505 	int type, vector;
1506 
1507 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1508 		return (EINVAL);
1509 
1510 	vcpu = &vm->vcpu[vcpuid];
1511 
1512 	if (info & VM_INTINFO_VALID) {
1513 		type = info & VM_INTINFO_TYPE;
1514 		vector = info & 0xff;
1515 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1516 			return (EINVAL);
1517 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1518 			return (EINVAL);
1519 		if (info & VM_INTINFO_RSVD)
1520 			return (EINVAL);
1521 	} else {
1522 		info = 0;
1523 	}
1524 	VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1525 	vcpu->exitintinfo = info;
1526 	return (0);
1527 }
1528 
1529 enum exc_class {
1530 	EXC_BENIGN,
1531 	EXC_CONTRIBUTORY,
1532 	EXC_PAGEFAULT
1533 };
1534 
1535 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
1536 
1537 static enum exc_class
1538 exception_class(uint64_t info)
1539 {
1540 	int type, vector;
1541 
1542 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1543 	type = info & VM_INTINFO_TYPE;
1544 	vector = info & 0xff;
1545 
1546 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1547 	switch (type) {
1548 	case VM_INTINFO_HWINTR:
1549 	case VM_INTINFO_SWINTR:
1550 	case VM_INTINFO_NMI:
1551 		return (EXC_BENIGN);
1552 	default:
1553 		/*
1554 		 * Hardware exception.
1555 		 *
1556 		 * SVM and VT-x use identical type values to represent NMI,
1557 		 * hardware interrupt and software interrupt.
1558 		 *
1559 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1560 		 * for exceptions except #BP and #OF. #BP and #OF use a type
1561 		 * value of '5' or '6'. Therefore we don't check for explicit
1562 		 * values of 'type' to classify 'intinfo' into a hardware
1563 		 * exception.
1564 		 */
1565 		break;
1566 	}
1567 
1568 	switch (vector) {
1569 	case IDT_PF:
1570 	case IDT_VE:
1571 		return (EXC_PAGEFAULT);
1572 	case IDT_DE:
1573 	case IDT_TS:
1574 	case IDT_NP:
1575 	case IDT_SS:
1576 	case IDT_GP:
1577 		return (EXC_CONTRIBUTORY);
1578 	default:
1579 		return (EXC_BENIGN);
1580 	}
1581 }
1582 
1583 static int
1584 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1585     uint64_t *retinfo)
1586 {
1587 	enum exc_class exc1, exc2;
1588 	int type1, vector1;
1589 
1590 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1591 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1592 
1593 	/*
1594 	 * If an exception occurs while attempting to call the double-fault
1595 	 * handler the processor enters shutdown mode (aka triple fault).
1596 	 */
1597 	type1 = info1 & VM_INTINFO_TYPE;
1598 	vector1 = info1 & 0xff;
1599 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1600 		VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1601 		    info1, info2);
1602 		vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1603 		*retinfo = 0;
1604 		return (0);
1605 	}
1606 
1607 	/*
1608 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1609 	 */
1610 	exc1 = exception_class(info1);
1611 	exc2 = exception_class(info2);
1612 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1613 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1614 		/* Convert nested fault into a double fault. */
1615 		*retinfo = IDT_DF;
1616 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1617 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
1618 	} else {
1619 		/* Handle exceptions serially */
1620 		*retinfo = info2;
1621 	}
1622 	return (1);
1623 }
1624 
1625 static uint64_t
1626 vcpu_exception_intinfo(struct vcpu *vcpu)
1627 {
1628 	uint64_t info = 0;
1629 
1630 	if (vcpu->exception_pending) {
1631 		info = vcpu->exception.vector & 0xff;
1632 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1633 		if (vcpu->exception.error_code_valid) {
1634 			info |= VM_INTINFO_DEL_ERRCODE;
1635 			info |= (uint64_t)vcpu->exception.error_code << 32;
1636 		}
1637 	}
1638 	return (info);
1639 }
1640 
1641 int
1642 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1643 {
1644 	struct vcpu *vcpu;
1645 	uint64_t info1, info2;
1646 	int valid;
1647 
1648 	KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid));
1649 
1650 	vcpu = &vm->vcpu[vcpuid];
1651 
1652 	info1 = vcpu->exitintinfo;
1653 	vcpu->exitintinfo = 0;
1654 
1655 	info2 = 0;
1656 	if (vcpu->exception_pending) {
1657 		info2 = vcpu_exception_intinfo(vcpu);
1658 		vcpu->exception_pending = 0;
1659 		VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1660 		    vcpu->exception.vector, info2);
1661 	}
1662 
1663 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
1664 		valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
1665 	} else if (info1 & VM_INTINFO_VALID) {
1666 		*retinfo = info1;
1667 		valid = 1;
1668 	} else if (info2 & VM_INTINFO_VALID) {
1669 		*retinfo = info2;
1670 		valid = 1;
1671 	} else {
1672 		valid = 0;
1673 	}
1674 
1675 	if (valid) {
1676 		VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
1677 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
1678 	}
1679 
1680 	return (valid);
1681 }
1682 
1683 int
1684 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
1685 {
1686 	struct vcpu *vcpu;
1687 
1688 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1689 		return (EINVAL);
1690 
1691 	vcpu = &vm->vcpu[vcpuid];
1692 	*info1 = vcpu->exitintinfo;
1693 	*info2 = vcpu_exception_intinfo(vcpu);
1694 	return (0);
1695 }
1696 
1697 int
1698 vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *exception)
1699 {
1700 	struct vcpu *vcpu;
1701 
1702 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1703 		return (EINVAL);
1704 
1705 	if (exception->vector < 0 || exception->vector >= 32)
1706 		return (EINVAL);
1707 
1708 	/*
1709 	 * A double fault exception should never be injected directly into
1710 	 * the guest. It is a derived exception that results from specific
1711 	 * combinations of nested faults.
1712 	 */
1713 	if (exception->vector == IDT_DF)
1714 		return (EINVAL);
1715 
1716 	vcpu = &vm->vcpu[vcpuid];
1717 
1718 	if (vcpu->exception_pending) {
1719 		VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
1720 		    "pending exception %d", exception->vector,
1721 		    vcpu->exception.vector);
1722 		return (EBUSY);
1723 	}
1724 
1725 	vcpu->exception_pending = 1;
1726 	vcpu->exception = *exception;
1727 	VCPU_CTR1(vm, vcpuid, "Exception %d pending", exception->vector);
1728 	return (0);
1729 }
1730 
1731 void
1732 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
1733     int errcode)
1734 {
1735 	struct vm_exception exception;
1736 	struct vm_exit *vmexit;
1737 	struct vm *vm;
1738 	int error;
1739 
1740 	vm = vmarg;
1741 
1742 	exception.vector = vector;
1743 	exception.error_code = errcode;
1744 	exception.error_code_valid = errcode_valid;
1745 	error = vm_inject_exception(vm, vcpuid, &exception);
1746 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
1747 
1748 	/*
1749 	 * A fault-like exception allows the instruction to be restarted
1750 	 * after the exception handler returns.
1751 	 *
1752 	 * By setting the inst_length to 0 we ensure that the instruction
1753 	 * pointer remains at the faulting instruction.
1754 	 */
1755 	vmexit = vm_exitinfo(vm, vcpuid);
1756 	vmexit->inst_length = 0;
1757 }
1758 
1759 void
1760 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
1761 {
1762 	struct vm *vm;
1763 	int error;
1764 
1765 	vm = vmarg;
1766 	VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
1767 	    error_code, cr2);
1768 
1769 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
1770 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
1771 
1772 	vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
1773 }
1774 
1775 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
1776 
1777 int
1778 vm_inject_nmi(struct vm *vm, int vcpuid)
1779 {
1780 	struct vcpu *vcpu;
1781 
1782 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1783 		return (EINVAL);
1784 
1785 	vcpu = &vm->vcpu[vcpuid];
1786 
1787 	vcpu->nmi_pending = 1;
1788 	vcpu_notify_event(vm, vcpuid, false);
1789 	return (0);
1790 }
1791 
1792 int
1793 vm_nmi_pending(struct vm *vm, int vcpuid)
1794 {
1795 	struct vcpu *vcpu;
1796 
1797 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1798 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1799 
1800 	vcpu = &vm->vcpu[vcpuid];
1801 
1802 	return (vcpu->nmi_pending);
1803 }
1804 
1805 void
1806 vm_nmi_clear(struct vm *vm, int vcpuid)
1807 {
1808 	struct vcpu *vcpu;
1809 
1810 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1811 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1812 
1813 	vcpu = &vm->vcpu[vcpuid];
1814 
1815 	if (vcpu->nmi_pending == 0)
1816 		panic("vm_nmi_clear: inconsistent nmi_pending state");
1817 
1818 	vcpu->nmi_pending = 0;
1819 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
1820 }
1821 
1822 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
1823 
1824 int
1825 vm_inject_extint(struct vm *vm, int vcpuid)
1826 {
1827 	struct vcpu *vcpu;
1828 
1829 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1830 		return (EINVAL);
1831 
1832 	vcpu = &vm->vcpu[vcpuid];
1833 
1834 	vcpu->extint_pending = 1;
1835 	vcpu_notify_event(vm, vcpuid, false);
1836 	return (0);
1837 }
1838 
1839 int
1840 vm_extint_pending(struct vm *vm, int vcpuid)
1841 {
1842 	struct vcpu *vcpu;
1843 
1844 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1845 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1846 
1847 	vcpu = &vm->vcpu[vcpuid];
1848 
1849 	return (vcpu->extint_pending);
1850 }
1851 
1852 void
1853 vm_extint_clear(struct vm *vm, int vcpuid)
1854 {
1855 	struct vcpu *vcpu;
1856 
1857 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1858 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1859 
1860 	vcpu = &vm->vcpu[vcpuid];
1861 
1862 	if (vcpu->extint_pending == 0)
1863 		panic("vm_extint_clear: inconsistent extint_pending state");
1864 
1865 	vcpu->extint_pending = 0;
1866 	vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
1867 }
1868 
1869 int
1870 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
1871 {
1872 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1873 		return (EINVAL);
1874 
1875 	if (type < 0 || type >= VM_CAP_MAX)
1876 		return (EINVAL);
1877 
1878 	return (VMGETCAP(vm->cookie, vcpu, type, retval));
1879 }
1880 
1881 int
1882 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
1883 {
1884 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1885 		return (EINVAL);
1886 
1887 	if (type < 0 || type >= VM_CAP_MAX)
1888 		return (EINVAL);
1889 
1890 	return (VMSETCAP(vm->cookie, vcpu, type, val));
1891 }
1892 
1893 uint64_t *
1894 vm_guest_msrs(struct vm *vm, int cpu)
1895 {
1896 	return (vm->vcpu[cpu].guest_msrs);
1897 }
1898 
1899 struct vlapic *
1900 vm_lapic(struct vm *vm, int cpu)
1901 {
1902 	return (vm->vcpu[cpu].vlapic);
1903 }
1904 
1905 struct vioapic *
1906 vm_ioapic(struct vm *vm)
1907 {
1908 
1909 	return (vm->vioapic);
1910 }
1911 
1912 struct vhpet *
1913 vm_hpet(struct vm *vm)
1914 {
1915 
1916 	return (vm->vhpet);
1917 }
1918 
1919 boolean_t
1920 vmm_is_pptdev(int bus, int slot, int func)
1921 {
1922 	int found, i, n;
1923 	int b, s, f;
1924 	char *val, *cp, *cp2;
1925 
1926 	/*
1927 	 * XXX
1928 	 * The length of an environment variable is limited to 128 bytes which
1929 	 * puts an upper limit on the number of passthru devices that may be
1930 	 * specified using a single environment variable.
1931 	 *
1932 	 * Work around this by scanning multiple environment variable
1933 	 * names instead of a single one - yuck!
1934 	 */
1935 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
1936 
1937 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
1938 	found = 0;
1939 	for (i = 0; names[i] != NULL && !found; i++) {
1940 		cp = val = getenv(names[i]);
1941 		while (cp != NULL && *cp != '\0') {
1942 			if ((cp2 = strchr(cp, ' ')) != NULL)
1943 				*cp2 = '\0';
1944 
1945 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
1946 			if (n == 3 && bus == b && slot == s && func == f) {
1947 				found = 1;
1948 				break;
1949 			}
1950 
1951 			if (cp2 != NULL)
1952 				*cp2++ = ' ';
1953 
1954 			cp = cp2;
1955 		}
1956 		freeenv(val);
1957 	}
1958 	return (found);
1959 }
1960 
1961 void *
1962 vm_iommu_domain(struct vm *vm)
1963 {
1964 
1965 	return (vm->iommu);
1966 }
1967 
1968 int
1969 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1970     bool from_idle)
1971 {
1972 	int error;
1973 	struct vcpu *vcpu;
1974 
1975 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1976 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
1977 
1978 	vcpu = &vm->vcpu[vcpuid];
1979 
1980 	vcpu_lock(vcpu);
1981 	error = vcpu_set_state_locked(vcpu, newstate, from_idle);
1982 	vcpu_unlock(vcpu);
1983 
1984 	return (error);
1985 }
1986 
1987 enum vcpu_state
1988 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
1989 {
1990 	struct vcpu *vcpu;
1991 	enum vcpu_state state;
1992 
1993 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1994 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
1995 
1996 	vcpu = &vm->vcpu[vcpuid];
1997 
1998 	vcpu_lock(vcpu);
1999 	state = vcpu->state;
2000 	if (hostcpu != NULL)
2001 		*hostcpu = vcpu->hostcpu;
2002 	vcpu_unlock(vcpu);
2003 
2004 	return (state);
2005 }
2006 
2007 int
2008 vm_activate_cpu(struct vm *vm, int vcpuid)
2009 {
2010 
2011 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2012 		return (EINVAL);
2013 
2014 	if (CPU_ISSET(vcpuid, &vm->active_cpus))
2015 		return (EBUSY);
2016 
2017 	VCPU_CTR0(vm, vcpuid, "activated");
2018 	CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2019 	return (0);
2020 }
2021 
2022 cpuset_t
2023 vm_active_cpus(struct vm *vm)
2024 {
2025 
2026 	return (vm->active_cpus);
2027 }
2028 
2029 cpuset_t
2030 vm_suspended_cpus(struct vm *vm)
2031 {
2032 
2033 	return (vm->suspended_cpus);
2034 }
2035 
2036 void *
2037 vcpu_stats(struct vm *vm, int vcpuid)
2038 {
2039 
2040 	return (vm->vcpu[vcpuid].stats);
2041 }
2042 
2043 int
2044 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2045 {
2046 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2047 		return (EINVAL);
2048 
2049 	*state = vm->vcpu[vcpuid].x2apic_state;
2050 
2051 	return (0);
2052 }
2053 
2054 int
2055 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2056 {
2057 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2058 		return (EINVAL);
2059 
2060 	if (state >= X2APIC_STATE_LAST)
2061 		return (EINVAL);
2062 
2063 	vm->vcpu[vcpuid].x2apic_state = state;
2064 
2065 	vlapic_set_x2apic_state(vm, vcpuid, state);
2066 
2067 	return (0);
2068 }
2069 
2070 /*
2071  * This function is called to ensure that a vcpu "sees" a pending event
2072  * as soon as possible:
2073  * - If the vcpu thread is sleeping then it is woken up.
2074  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2075  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2076  */
2077 void
2078 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2079 {
2080 	int hostcpu;
2081 	struct vcpu *vcpu;
2082 
2083 	vcpu = &vm->vcpu[vcpuid];
2084 
2085 	vcpu_lock(vcpu);
2086 	hostcpu = vcpu->hostcpu;
2087 	if (vcpu->state == VCPU_RUNNING) {
2088 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2089 		if (hostcpu != curcpu) {
2090 			if (lapic_intr) {
2091 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2092 				    vmm_ipinum);
2093 			} else {
2094 				ipi_cpu(hostcpu, vmm_ipinum);
2095 			}
2096 		} else {
2097 			/*
2098 			 * If the 'vcpu' is running on 'curcpu' then it must
2099 			 * be sending a notification to itself (e.g. SELF_IPI).
2100 			 * The pending event will be picked up when the vcpu
2101 			 * transitions back to guest context.
2102 			 */
2103 		}
2104 	} else {
2105 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2106 		    "with hostcpu %d", vcpu->state, hostcpu));
2107 		if (vcpu->state == VCPU_SLEEPING)
2108 			wakeup_one(vcpu);
2109 	}
2110 	vcpu_unlock(vcpu);
2111 }
2112 
2113 struct vmspace *
2114 vm_get_vmspace(struct vm *vm)
2115 {
2116 
2117 	return (vm->vmspace);
2118 }
2119 
2120 int
2121 vm_apicid2vcpuid(struct vm *vm, int apicid)
2122 {
2123 	/*
2124 	 * XXX apic id is assumed to be numerically identical to vcpu id
2125 	 */
2126 	return (apicid);
2127 }
2128 
2129 void
2130 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2131     vm_rendezvous_func_t func, void *arg)
2132 {
2133 	int i;
2134 
2135 	/*
2136 	 * Enforce that this function is called without any locks
2137 	 */
2138 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2139 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
2140 	    ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2141 
2142 restart:
2143 	mtx_lock(&vm->rendezvous_mtx);
2144 	if (vm->rendezvous_func != NULL) {
2145 		/*
2146 		 * If a rendezvous is already in progress then we need to
2147 		 * call the rendezvous handler in case this 'vcpuid' is one
2148 		 * of the targets of the rendezvous.
2149 		 */
2150 		RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2151 		mtx_unlock(&vm->rendezvous_mtx);
2152 		vm_handle_rendezvous(vm, vcpuid);
2153 		goto restart;
2154 	}
2155 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2156 	    "rendezvous is still in progress"));
2157 
2158 	RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2159 	vm->rendezvous_req_cpus = dest;
2160 	CPU_ZERO(&vm->rendezvous_done_cpus);
2161 	vm->rendezvous_arg = arg;
2162 	vm_set_rendezvous_func(vm, func);
2163 	mtx_unlock(&vm->rendezvous_mtx);
2164 
2165 	/*
2166 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2167 	 * vcpus so they handle the rendezvous as soon as possible.
2168 	 */
2169 	for (i = 0; i < VM_MAXCPU; i++) {
2170 		if (CPU_ISSET(i, &dest))
2171 			vcpu_notify_event(vm, i, false);
2172 	}
2173 
2174 	vm_handle_rendezvous(vm, vcpuid);
2175 }
2176 
2177 struct vatpic *
2178 vm_atpic(struct vm *vm)
2179 {
2180 	return (vm->vatpic);
2181 }
2182 
2183 struct vatpit *
2184 vm_atpit(struct vm *vm)
2185 {
2186 	return (vm->vatpit);
2187 }
2188 
2189 enum vm_reg_name
2190 vm_segment_name(int seg)
2191 {
2192 	static enum vm_reg_name seg_names[] = {
2193 		VM_REG_GUEST_ES,
2194 		VM_REG_GUEST_CS,
2195 		VM_REG_GUEST_SS,
2196 		VM_REG_GUEST_DS,
2197 		VM_REG_GUEST_FS,
2198 		VM_REG_GUEST_GS
2199 	};
2200 
2201 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2202 	    ("%s: invalid segment encoding %d", __func__, seg));
2203 	return (seg_names[seg]);
2204 }
2205 
2206 void
2207 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2208     int num_copyinfo)
2209 {
2210 	int idx;
2211 
2212 	for (idx = 0; idx < num_copyinfo; idx++) {
2213 		if (copyinfo[idx].cookie != NULL)
2214 			vm_gpa_release(copyinfo[idx].cookie);
2215 	}
2216 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2217 }
2218 
2219 int
2220 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2221     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2222     int num_copyinfo)
2223 {
2224 	int error, idx, nused;
2225 	size_t n, off, remaining;
2226 	void *hva, *cookie;
2227 	uint64_t gpa;
2228 
2229 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2230 
2231 	nused = 0;
2232 	remaining = len;
2233 	while (remaining > 0) {
2234 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2235 		error = vmm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa);
2236 		if (error)
2237 			return (error);
2238 		off = gpa & PAGE_MASK;
2239 		n = min(remaining, PAGE_SIZE - off);
2240 		copyinfo[nused].gpa = gpa;
2241 		copyinfo[nused].len = n;
2242 		remaining -= n;
2243 		gla += n;
2244 		nused++;
2245 	}
2246 
2247 	for (idx = 0; idx < nused; idx++) {
2248 		hva = vm_gpa_hold(vm, copyinfo[idx].gpa, copyinfo[idx].len,
2249 		    prot, &cookie);
2250 		if (hva == NULL)
2251 			break;
2252 		copyinfo[idx].hva = hva;
2253 		copyinfo[idx].cookie = cookie;
2254 	}
2255 
2256 	if (idx != nused) {
2257 		vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2258 		return (-1);
2259 	} else {
2260 		return (0);
2261 	}
2262 }
2263 
2264 void
2265 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2266     size_t len)
2267 {
2268 	char *dst;
2269 	int idx;
2270 
2271 	dst = kaddr;
2272 	idx = 0;
2273 	while (len > 0) {
2274 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2275 		len -= copyinfo[idx].len;
2276 		dst += copyinfo[idx].len;
2277 		idx++;
2278 	}
2279 }
2280 
2281 void
2282 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2283     struct vm_copyinfo *copyinfo, size_t len)
2284 {
2285 	const char *src;
2286 	int idx;
2287 
2288 	src = kaddr;
2289 	idx = 0;
2290 	while (len > 0) {
2291 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2292 		len -= copyinfo[idx].len;
2293 		src += copyinfo[idx].len;
2294 		idx++;
2295 	}
2296 }
2297 
2298 /*
2299  * Return the amount of in-use and wired memory for the VM. Since
2300  * these are global stats, only return the values with for vCPU 0
2301  */
2302 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2303 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2304 
2305 static void
2306 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2307 {
2308 
2309 	if (vcpu == 0) {
2310 		vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2311 	       	    PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2312 	}
2313 }
2314 
2315 static void
2316 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2317 {
2318 
2319 	if (vcpu == 0) {
2320 		vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2321 	      	    PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2322 	}
2323 }
2324 
2325 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2326 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2327