xref: /freebsd/sys/amd64/vmm/vmm.c (revision bd1da0a002e9a43cfb5220835c7a42804d90dc56)
1 /*-
2  * Copyright (c) 2011 NetApp, Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/module.h>
36 #include <sys/sysctl.h>
37 #include <sys/malloc.h>
38 #include <sys/pcpu.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/proc.h>
42 #include <sys/rwlock.h>
43 #include <sys/sched.h>
44 #include <sys/smp.h>
45 #include <sys/systm.h>
46 
47 #include <vm/vm.h>
48 #include <vm/vm_object.h>
49 #include <vm/vm_page.h>
50 #include <vm/pmap.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_extern.h>
53 #include <vm/vm_param.h>
54 
55 #include <machine/cpu.h>
56 #include <machine/vm.h>
57 #include <machine/pcb.h>
58 #include <machine/smp.h>
59 #include <x86/psl.h>
60 #include <x86/apicreg.h>
61 #include <machine/vmparam.h>
62 
63 #include <machine/vmm.h>
64 #include <machine/vmm_dev.h>
65 #include <machine/vmm_instruction_emul.h>
66 
67 #include "vmm_ioport.h"
68 #include "vmm_ktr.h"
69 #include "vmm_host.h"
70 #include "vmm_mem.h"
71 #include "vmm_util.h"
72 #include "vatpic.h"
73 #include "vatpit.h"
74 #include "vhpet.h"
75 #include "vioapic.h"
76 #include "vlapic.h"
77 #include "vpmtmr.h"
78 #include "vrtc.h"
79 #include "vmm_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 	int	exception_pending;	/* (i) exception pending */
104 	int	exc_vector;		/* (x) exception collateral */
105 	int	exc_errcode_valid;
106 	uint32_t exc_errcode;
107 	struct savefpu	*guestfpu;	/* (a,i) guest fpu state */
108 	uint64_t	guest_xcr0;	/* (i) guest %xcr0 register */
109 	void		*stats;		/* (a,i) statistics */
110 	struct vm_exit	exitinfo;	/* (x) exit reason and collateral */
111 	uint64_t	nextrip;	/* (x) next instruction to execute */
112 };
113 
114 #define	vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
115 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
116 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
117 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
118 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
119 
120 struct mem_seg {
121 	vm_paddr_t	gpa;
122 	size_t		len;
123 	boolean_t	wired;
124 	vm_object_t	object;
125 };
126 #define	VM_MAX_MEMORY_SEGMENTS	2
127 
128 /*
129  * Initialization:
130  * (o) initialized the first time the VM is created
131  * (i) initialized when VM is created and when it is reinitialized
132  * (x) initialized before use
133  */
134 struct vm {
135 	void		*cookie;		/* (i) cpu-specific data */
136 	void		*iommu;			/* (x) iommu-specific data */
137 	struct vhpet	*vhpet;			/* (i) virtual HPET */
138 	struct vioapic	*vioapic;		/* (i) virtual ioapic */
139 	struct vatpic	*vatpic;		/* (i) virtual atpic */
140 	struct vatpit	*vatpit;		/* (i) virtual atpit */
141 	struct vpmtmr	*vpmtmr;		/* (i) virtual ACPI PM timer */
142 	struct vrtc	*vrtc;			/* (o) virtual RTC */
143 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
144 	int		suspend;		/* (i) stop VM execution */
145 	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
146 	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
147 	cpuset_t	rendezvous_req_cpus;	/* (x) rendezvous requested */
148 	cpuset_t	rendezvous_done_cpus;	/* (x) rendezvous finished */
149 	void		*rendezvous_arg;	/* (x) rendezvous func/arg */
150 	vm_rendezvous_func_t rendezvous_func;
151 	struct mtx	rendezvous_mtx;		/* (o) rendezvous lock */
152 	int		num_mem_segs;		/* (o) guest memory segments */
153 	struct mem_seg	mem_segs[VM_MAX_MEMORY_SEGMENTS];
154 	struct vmspace	*vmspace;		/* (o) guest's address space */
155 	char		name[VM_MAX_NAMELEN];	/* (o) virtual machine name */
156 	struct vcpu	vcpu[VM_MAXCPU];	/* (i) guest vcpus */
157 };
158 
159 static int vmm_initialized;
160 
161 static struct vmm_ops *ops;
162 #define	VMM_INIT(num)	(ops != NULL ? (*ops->init)(num) : 0)
163 #define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
164 #define	VMM_RESUME()	(ops != NULL ? (*ops->resume)() : 0)
165 
166 #define	VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
167 #define	VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \
168 	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO)
169 #define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
170 #define	VMSPACE_ALLOC(min, max) \
171 	(ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
172 #define	VMSPACE_FREE(vmspace) \
173 	(ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
174 #define	VMGETREG(vmi, vcpu, num, retval)		\
175 	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
176 #define	VMSETREG(vmi, vcpu, num, val)		\
177 	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
178 #define	VMGETDESC(vmi, vcpu, num, desc)		\
179 	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
180 #define	VMSETDESC(vmi, vcpu, num, desc)		\
181 	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
182 #define	VMGETCAP(vmi, vcpu, num, retval)	\
183 	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
184 #define	VMSETCAP(vmi, vcpu, num, val)		\
185 	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
186 #define	VLAPIC_INIT(vmi, vcpu)			\
187 	(ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
188 #define	VLAPIC_CLEANUP(vmi, vlapic)		\
189 	(ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
190 
191 #define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
192 #define	fpu_stop_emulating()	clts()
193 
194 static MALLOC_DEFINE(M_VM, "vm", "vm");
195 
196 /* statistics */
197 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
198 
199 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
200 
201 /*
202  * Halt the guest if all vcpus are executing a HLT instruction with
203  * interrupts disabled.
204  */
205 static int halt_detection_enabled = 1;
206 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
207     &halt_detection_enabled, 0,
208     "Halt VM if all vcpus execute HLT with interrupts disabled");
209 
210 static int vmm_ipinum;
211 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
212     "IPI vector used for vcpu notifications");
213 
214 static int trace_guest_exceptions;
215 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN,
216     &trace_guest_exceptions, 0,
217     "Trap into hypervisor on all guest exceptions and reflect them back");
218 
219 static int vmm_force_iommu = 0;
220 TUNABLE_INT("hw.vmm.force_iommu", &vmm_force_iommu);
221 SYSCTL_INT(_hw_vmm, OID_AUTO, force_iommu, CTLFLAG_RDTUN, &vmm_force_iommu, 0,
222     "Force use of I/O MMU even if no passthrough devices were found.");
223 
224 static void
225 vcpu_cleanup(struct vm *vm, int i, bool destroy)
226 {
227 	struct vcpu *vcpu = &vm->vcpu[i];
228 
229 	VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
230 	if (destroy) {
231 		vmm_stat_free(vcpu->stats);
232 		fpu_save_area_free(vcpu->guestfpu);
233 	}
234 }
235 
236 static void
237 vcpu_init(struct vm *vm, int vcpu_id, bool create)
238 {
239 	struct vcpu *vcpu;
240 
241 	KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU,
242 	    ("vcpu_init: invalid vcpu %d", vcpu_id));
243 
244 	vcpu = &vm->vcpu[vcpu_id];
245 
246 	if (create) {
247 		KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
248 		    "initialized", vcpu_id));
249 		vcpu_lock_init(vcpu);
250 		vcpu->state = VCPU_IDLE;
251 		vcpu->hostcpu = NOCPU;
252 		vcpu->guestfpu = fpu_save_area_alloc();
253 		vcpu->stats = vmm_stat_alloc();
254 	}
255 
256 	vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
257 	vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
258 	vcpu->exitintinfo = 0;
259 	vcpu->nmi_pending = 0;
260 	vcpu->extint_pending = 0;
261 	vcpu->exception_pending = 0;
262 	vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
263 	fpu_save_area_reset(vcpu->guestfpu);
264 	vmm_stat_init(vcpu->stats);
265 }
266 
267 int
268 vcpu_trace_exceptions(struct vm *vm, int vcpuid)
269 {
270 
271 	return (trace_guest_exceptions);
272 }
273 
274 struct vm_exit *
275 vm_exitinfo(struct vm *vm, int cpuid)
276 {
277 	struct vcpu *vcpu;
278 
279 	if (cpuid < 0 || cpuid >= VM_MAXCPU)
280 		panic("vm_exitinfo: invalid cpuid %d", cpuid);
281 
282 	vcpu = &vm->vcpu[cpuid];
283 
284 	return (&vcpu->exitinfo);
285 }
286 
287 static void
288 vmm_resume(void)
289 {
290 	VMM_RESUME();
291 }
292 
293 static int
294 vmm_init(void)
295 {
296 	int error;
297 
298 	vmm_host_state_init();
299 
300 	vmm_ipinum = lapic_ipi_alloc(&IDTVEC(justreturn));
301 	if (vmm_ipinum < 0)
302 		vmm_ipinum = IPI_AST;
303 
304 	error = vmm_mem_init();
305 	if (error)
306 		return (error);
307 
308 	if (vmm_is_intel())
309 		ops = &vmm_ops_intel;
310 	else if (vmm_is_amd())
311 		ops = &vmm_ops_amd;
312 	else
313 		return (ENXIO);
314 
315 	vmm_resume_p = vmm_resume;
316 
317 	return (VMM_INIT(vmm_ipinum));
318 }
319 
320 static int
321 vmm_handler(module_t mod, int what, void *arg)
322 {
323 	int error;
324 
325 	switch (what) {
326 	case MOD_LOAD:
327 		vmmdev_init();
328 		if (vmm_force_iommu || ppt_avail_devices() > 0)
329 			iommu_init();
330 		error = vmm_init();
331 		if (error == 0)
332 			vmm_initialized = 1;
333 		break;
334 	case MOD_UNLOAD:
335 		error = vmmdev_cleanup();
336 		if (error == 0) {
337 			vmm_resume_p = NULL;
338 			iommu_cleanup();
339 			if (vmm_ipinum != IPI_AST)
340 				lapic_ipi_free(vmm_ipinum);
341 			error = VMM_CLEANUP();
342 			/*
343 			 * Something bad happened - prevent new
344 			 * VMs from being created
345 			 */
346 			if (error)
347 				vmm_initialized = 0;
348 		}
349 		break;
350 	default:
351 		error = 0;
352 		break;
353 	}
354 	return (error);
355 }
356 
357 static moduledata_t vmm_kmod = {
358 	"vmm",
359 	vmm_handler,
360 	NULL
361 };
362 
363 /*
364  * vmm initialization has the following dependencies:
365  *
366  * - iommu initialization must happen after the pci passthru driver has had
367  *   a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
368  *
369  * - VT-x initialization requires smp_rendezvous() and therefore must happen
370  *   after SMP is fully functional (after SI_SUB_SMP).
371  */
372 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
373 MODULE_VERSION(vmm, 1);
374 
375 static void
376 vm_init(struct vm *vm, bool create)
377 {
378 	int i;
379 
380 	vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
381 	vm->iommu = NULL;
382 	vm->vioapic = vioapic_init(vm);
383 	vm->vhpet = vhpet_init(vm);
384 	vm->vatpic = vatpic_init(vm);
385 	vm->vatpit = vatpit_init(vm);
386 	vm->vpmtmr = vpmtmr_init(vm);
387 	if (create)
388 		vm->vrtc = vrtc_init(vm);
389 
390 	CPU_ZERO(&vm->active_cpus);
391 
392 	vm->suspend = 0;
393 	CPU_ZERO(&vm->suspended_cpus);
394 
395 	for (i = 0; i < VM_MAXCPU; i++)
396 		vcpu_init(vm, i, create);
397 }
398 
399 int
400 vm_create(const char *name, struct vm **retvm)
401 {
402 	struct vm *vm;
403 	struct vmspace *vmspace;
404 
405 	/*
406 	 * If vmm.ko could not be successfully initialized then don't attempt
407 	 * to create the virtual machine.
408 	 */
409 	if (!vmm_initialized)
410 		return (ENXIO);
411 
412 	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
413 		return (EINVAL);
414 
415 	vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS);
416 	if (vmspace == NULL)
417 		return (ENOMEM);
418 
419 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
420 	strcpy(vm->name, name);
421 	vm->num_mem_segs = 0;
422 	vm->vmspace = vmspace;
423 	mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
424 
425 	vm_init(vm, true);
426 
427 	*retvm = vm;
428 	return (0);
429 }
430 
431 static void
432 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg)
433 {
434 
435 	if (seg->object != NULL)
436 		vmm_mem_free(vm->vmspace, seg->gpa, seg->len);
437 
438 	bzero(seg, sizeof(*seg));
439 }
440 
441 static void
442 vm_cleanup(struct vm *vm, bool destroy)
443 {
444 	int i;
445 
446 	ppt_unassign_all(vm);
447 
448 	if (vm->iommu != NULL)
449 		iommu_destroy_domain(vm->iommu);
450 
451 	if (destroy)
452 		vrtc_cleanup(vm->vrtc);
453 	else
454 		vrtc_reset(vm->vrtc);
455 	vpmtmr_cleanup(vm->vpmtmr);
456 	vatpit_cleanup(vm->vatpit);
457 	vhpet_cleanup(vm->vhpet);
458 	vatpic_cleanup(vm->vatpic);
459 	vioapic_cleanup(vm->vioapic);
460 
461 	for (i = 0; i < VM_MAXCPU; i++)
462 		vcpu_cleanup(vm, i, destroy);
463 
464 	VMCLEANUP(vm->cookie);
465 
466 	if (destroy) {
467 		for (i = 0; i < vm->num_mem_segs; i++)
468 			vm_free_mem_seg(vm, &vm->mem_segs[i]);
469 
470 		vm->num_mem_segs = 0;
471 
472 		VMSPACE_FREE(vm->vmspace);
473 		vm->vmspace = NULL;
474 	}
475 }
476 
477 void
478 vm_destroy(struct vm *vm)
479 {
480 	vm_cleanup(vm, true);
481 	free(vm, M_VM);
482 }
483 
484 int
485 vm_reinit(struct vm *vm)
486 {
487 	int error;
488 
489 	/*
490 	 * A virtual machine can be reset only if all vcpus are suspended.
491 	 */
492 	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
493 		vm_cleanup(vm, false);
494 		vm_init(vm, false);
495 		error = 0;
496 	} else {
497 		error = EBUSY;
498 	}
499 
500 	return (error);
501 }
502 
503 const char *
504 vm_name(struct vm *vm)
505 {
506 	return (vm->name);
507 }
508 
509 int
510 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
511 {
512 	vm_object_t obj;
513 
514 	if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
515 		return (ENOMEM);
516 	else
517 		return (0);
518 }
519 
520 int
521 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
522 {
523 
524 	vmm_mmio_free(vm->vmspace, gpa, len);
525 	return (0);
526 }
527 
528 boolean_t
529 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa)
530 {
531 	int i;
532 	vm_paddr_t gpabase, gpalimit;
533 
534 	for (i = 0; i < vm->num_mem_segs; i++) {
535 		gpabase = vm->mem_segs[i].gpa;
536 		gpalimit = gpabase + vm->mem_segs[i].len;
537 		if (gpa >= gpabase && gpa < gpalimit)
538 			return (TRUE);		/* 'gpa' is regular memory */
539 	}
540 
541 	if (ppt_is_mmio(vm, gpa))
542 		return (TRUE);			/* 'gpa' is pci passthru mmio */
543 
544 	return (FALSE);
545 }
546 
547 int
548 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
549 {
550 	int available, allocated;
551 	struct mem_seg *seg;
552 	vm_object_t object;
553 	vm_paddr_t g;
554 
555 	if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
556 		return (EINVAL);
557 
558 	available = allocated = 0;
559 	g = gpa;
560 	while (g < gpa + len) {
561 		if (vm_mem_allocated(vm, g))
562 			allocated++;
563 		else
564 			available++;
565 
566 		g += PAGE_SIZE;
567 	}
568 
569 	/*
570 	 * If there are some allocated and some available pages in the address
571 	 * range then it is an error.
572 	 */
573 	if (allocated && available)
574 		return (EINVAL);
575 
576 	/*
577 	 * If the entire address range being requested has already been
578 	 * allocated then there isn't anything more to do.
579 	 */
580 	if (allocated && available == 0)
581 		return (0);
582 
583 	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
584 		return (E2BIG);
585 
586 	seg = &vm->mem_segs[vm->num_mem_segs];
587 
588 	if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL)
589 		return (ENOMEM);
590 
591 	seg->gpa = gpa;
592 	seg->len = len;
593 	seg->object = object;
594 	seg->wired = FALSE;
595 
596 	vm->num_mem_segs++;
597 
598 	return (0);
599 }
600 
601 static vm_paddr_t
602 vm_maxmem(struct vm *vm)
603 {
604 	int i;
605 	vm_paddr_t gpa, maxmem;
606 
607 	maxmem = 0;
608 	for (i = 0; i < vm->num_mem_segs; i++) {
609 		gpa = vm->mem_segs[i].gpa + vm->mem_segs[i].len;
610 		if (gpa > maxmem)
611 			maxmem = gpa;
612 	}
613 	return (maxmem);
614 }
615 
616 static void
617 vm_gpa_unwire(struct vm *vm)
618 {
619 	int i, rv;
620 	struct mem_seg *seg;
621 
622 	for (i = 0; i < vm->num_mem_segs; i++) {
623 		seg = &vm->mem_segs[i];
624 		if (!seg->wired)
625 			continue;
626 
627 		rv = vm_map_unwire(&vm->vmspace->vm_map,
628 				   seg->gpa, seg->gpa + seg->len,
629 				   VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
630 		KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment "
631 		    "%#lx/%ld could not be unwired: %d",
632 		    vm_name(vm), seg->gpa, seg->len, rv));
633 
634 		seg->wired = FALSE;
635 	}
636 }
637 
638 static int
639 vm_gpa_wire(struct vm *vm)
640 {
641 	int i, rv;
642 	struct mem_seg *seg;
643 
644 	for (i = 0; i < vm->num_mem_segs; i++) {
645 		seg = &vm->mem_segs[i];
646 		if (seg->wired)
647 			continue;
648 
649 		/* XXX rlimits? */
650 		rv = vm_map_wire(&vm->vmspace->vm_map,
651 				 seg->gpa, seg->gpa + seg->len,
652 				 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
653 		if (rv != KERN_SUCCESS)
654 			break;
655 
656 		seg->wired = TRUE;
657 	}
658 
659 	if (i < vm->num_mem_segs) {
660 		/*
661 		 * Undo the wiring before returning an error.
662 		 */
663 		vm_gpa_unwire(vm);
664 		return (EAGAIN);
665 	}
666 
667 	return (0);
668 }
669 
670 static void
671 vm_iommu_modify(struct vm *vm, boolean_t map)
672 {
673 	int i, sz;
674 	vm_paddr_t gpa, hpa;
675 	struct mem_seg *seg;
676 	void *vp, *cookie, *host_domain;
677 
678 	sz = PAGE_SIZE;
679 	host_domain = iommu_host_domain();
680 
681 	for (i = 0; i < vm->num_mem_segs; i++) {
682 		seg = &vm->mem_segs[i];
683 		KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired",
684 		    vm_name(vm), seg->gpa, seg->len));
685 
686 		gpa = seg->gpa;
687 		while (gpa < seg->gpa + seg->len) {
688 			vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE,
689 					 &cookie);
690 			KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
691 			    vm_name(vm), gpa));
692 
693 			vm_gpa_release(cookie);
694 
695 			hpa = DMAP_TO_PHYS((uintptr_t)vp);
696 			if (map) {
697 				iommu_create_mapping(vm->iommu, gpa, hpa, sz);
698 				iommu_remove_mapping(host_domain, hpa, sz);
699 			} else {
700 				iommu_remove_mapping(vm->iommu, gpa, sz);
701 				iommu_create_mapping(host_domain, hpa, hpa, sz);
702 			}
703 
704 			gpa += PAGE_SIZE;
705 		}
706 	}
707 
708 	/*
709 	 * Invalidate the cached translations associated with the domain
710 	 * from which pages were removed.
711 	 */
712 	if (map)
713 		iommu_invalidate_tlb(host_domain);
714 	else
715 		iommu_invalidate_tlb(vm->iommu);
716 }
717 
718 #define	vm_iommu_unmap(vm)	vm_iommu_modify((vm), FALSE)
719 #define	vm_iommu_map(vm)	vm_iommu_modify((vm), TRUE)
720 
721 int
722 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
723 {
724 	int error;
725 
726 	error = ppt_unassign_device(vm, bus, slot, func);
727 	if (error)
728 		return (error);
729 
730 	if (ppt_assigned_devices(vm) == 0) {
731 		vm_iommu_unmap(vm);
732 		vm_gpa_unwire(vm);
733 	}
734 	return (0);
735 }
736 
737 int
738 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
739 {
740 	int error;
741 	vm_paddr_t maxaddr;
742 
743 	/*
744 	 * Virtual machines with pci passthru devices get special treatment:
745 	 * - the guest physical memory is wired
746 	 * - the iommu is programmed to do the 'gpa' to 'hpa' translation
747 	 *
748 	 * We need to do this before the first pci passthru device is attached.
749 	 */
750 	if (ppt_assigned_devices(vm) == 0) {
751 		KASSERT(vm->iommu == NULL,
752 		    ("vm_assign_pptdev: iommu must be NULL"));
753 		maxaddr = vm_maxmem(vm);
754 		vm->iommu = iommu_create_domain(maxaddr);
755 
756 		error = vm_gpa_wire(vm);
757 		if (error)
758 			return (error);
759 
760 		vm_iommu_map(vm);
761 	}
762 
763 	error = ppt_assign_device(vm, bus, slot, func);
764 	return (error);
765 }
766 
767 void *
768 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
769 	    void **cookie)
770 {
771 	int count, pageoff;
772 	vm_page_t m;
773 
774 	pageoff = gpa & PAGE_MASK;
775 	if (len > PAGE_SIZE - pageoff)
776 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
777 
778 	count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
779 	    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
780 
781 	if (count == 1) {
782 		*cookie = m;
783 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
784 	} else {
785 		*cookie = NULL;
786 		return (NULL);
787 	}
788 }
789 
790 void
791 vm_gpa_release(void *cookie)
792 {
793 	vm_page_t m = cookie;
794 
795 	vm_page_lock(m);
796 	vm_page_unhold(m);
797 	vm_page_unlock(m);
798 }
799 
800 int
801 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
802 		  struct vm_memory_segment *seg)
803 {
804 	int i;
805 
806 	for (i = 0; i < vm->num_mem_segs; i++) {
807 		if (gpabase == vm->mem_segs[i].gpa) {
808 			seg->gpa = vm->mem_segs[i].gpa;
809 			seg->len = vm->mem_segs[i].len;
810 			seg->wired = vm->mem_segs[i].wired;
811 			return (0);
812 		}
813 	}
814 	return (-1);
815 }
816 
817 int
818 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len,
819 	      vm_offset_t *offset, struct vm_object **object)
820 {
821 	int i;
822 	size_t seg_len;
823 	vm_paddr_t seg_gpa;
824 	vm_object_t seg_obj;
825 
826 	for (i = 0; i < vm->num_mem_segs; i++) {
827 		if ((seg_obj = vm->mem_segs[i].object) == NULL)
828 			continue;
829 
830 		seg_gpa = vm->mem_segs[i].gpa;
831 		seg_len = vm->mem_segs[i].len;
832 
833 		if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) {
834 			*offset = gpa - seg_gpa;
835 			*object = seg_obj;
836 			vm_object_reference(seg_obj);
837 			return (0);
838 		}
839 	}
840 
841 	return (EINVAL);
842 }
843 
844 int
845 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
846 {
847 
848 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
849 		return (EINVAL);
850 
851 	if (reg >= VM_REG_LAST)
852 		return (EINVAL);
853 
854 	return (VMGETREG(vm->cookie, vcpu, reg, retval));
855 }
856 
857 int
858 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val)
859 {
860 	struct vcpu *vcpu;
861 	int error;
862 
863 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
864 		return (EINVAL);
865 
866 	if (reg >= VM_REG_LAST)
867 		return (EINVAL);
868 
869 	error = VMSETREG(vm->cookie, vcpuid, reg, val);
870 	if (error || reg != VM_REG_GUEST_RIP)
871 		return (error);
872 
873 	/* Set 'nextrip' to match the value of %rip */
874 	VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val);
875 	vcpu = &vm->vcpu[vcpuid];
876 	vcpu->nextrip = val;
877 	return (0);
878 }
879 
880 static boolean_t
881 is_descriptor_table(int reg)
882 {
883 
884 	switch (reg) {
885 	case VM_REG_GUEST_IDTR:
886 	case VM_REG_GUEST_GDTR:
887 		return (TRUE);
888 	default:
889 		return (FALSE);
890 	}
891 }
892 
893 static boolean_t
894 is_segment_register(int reg)
895 {
896 
897 	switch (reg) {
898 	case VM_REG_GUEST_ES:
899 	case VM_REG_GUEST_CS:
900 	case VM_REG_GUEST_SS:
901 	case VM_REG_GUEST_DS:
902 	case VM_REG_GUEST_FS:
903 	case VM_REG_GUEST_GS:
904 	case VM_REG_GUEST_TR:
905 	case VM_REG_GUEST_LDTR:
906 		return (TRUE);
907 	default:
908 		return (FALSE);
909 	}
910 }
911 
912 int
913 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
914 		struct seg_desc *desc)
915 {
916 
917 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
918 		return (EINVAL);
919 
920 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
921 		return (EINVAL);
922 
923 	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
924 }
925 
926 int
927 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
928 		struct seg_desc *desc)
929 {
930 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
931 		return (EINVAL);
932 
933 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
934 		return (EINVAL);
935 
936 	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
937 }
938 
939 static void
940 restore_guest_fpustate(struct vcpu *vcpu)
941 {
942 
943 	/* flush host state to the pcb */
944 	fpuexit(curthread);
945 
946 	/* restore guest FPU state */
947 	fpu_stop_emulating();
948 	fpurestore(vcpu->guestfpu);
949 
950 	/* restore guest XCR0 if XSAVE is enabled in the host */
951 	if (rcr4() & CR4_XSAVE)
952 		load_xcr(0, vcpu->guest_xcr0);
953 
954 	/*
955 	 * The FPU is now "dirty" with the guest's state so turn on emulation
956 	 * to trap any access to the FPU by the host.
957 	 */
958 	fpu_start_emulating();
959 }
960 
961 static void
962 save_guest_fpustate(struct vcpu *vcpu)
963 {
964 
965 	if ((rcr0() & CR0_TS) == 0)
966 		panic("fpu emulation not enabled in host!");
967 
968 	/* save guest XCR0 and restore host XCR0 */
969 	if (rcr4() & CR4_XSAVE) {
970 		vcpu->guest_xcr0 = rxcr(0);
971 		load_xcr(0, vmm_get_host_xcr0());
972 	}
973 
974 	/* save guest FPU state */
975 	fpu_stop_emulating();
976 	fpusave(vcpu->guestfpu);
977 	fpu_start_emulating();
978 }
979 
980 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
981 
982 static int
983 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
984     bool from_idle)
985 {
986 	int error;
987 
988 	vcpu_assert_locked(vcpu);
989 
990 	/*
991 	 * State transitions from the vmmdev_ioctl() must always begin from
992 	 * the VCPU_IDLE state. This guarantees that there is only a single
993 	 * ioctl() operating on a vcpu at any point.
994 	 */
995 	if (from_idle) {
996 		while (vcpu->state != VCPU_IDLE)
997 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
998 	} else {
999 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1000 		    "vcpu idle state"));
1001 	}
1002 
1003 	if (vcpu->state == VCPU_RUNNING) {
1004 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1005 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1006 	} else {
1007 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1008 		    "vcpu that is not running", vcpu->hostcpu));
1009 	}
1010 
1011 	/*
1012 	 * The following state transitions are allowed:
1013 	 * IDLE -> FROZEN -> IDLE
1014 	 * FROZEN -> RUNNING -> FROZEN
1015 	 * FROZEN -> SLEEPING -> FROZEN
1016 	 */
1017 	switch (vcpu->state) {
1018 	case VCPU_IDLE:
1019 	case VCPU_RUNNING:
1020 	case VCPU_SLEEPING:
1021 		error = (newstate != VCPU_FROZEN);
1022 		break;
1023 	case VCPU_FROZEN:
1024 		error = (newstate == VCPU_FROZEN);
1025 		break;
1026 	default:
1027 		error = 1;
1028 		break;
1029 	}
1030 
1031 	if (error)
1032 		return (EBUSY);
1033 
1034 	vcpu->state = newstate;
1035 	if (newstate == VCPU_RUNNING)
1036 		vcpu->hostcpu = curcpu;
1037 	else
1038 		vcpu->hostcpu = NOCPU;
1039 
1040 	if (newstate == VCPU_IDLE)
1041 		wakeup(&vcpu->state);
1042 
1043 	return (0);
1044 }
1045 
1046 static void
1047 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1048 {
1049 	int error;
1050 
1051 	if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1052 		panic("Error %d setting state to %d\n", error, newstate);
1053 }
1054 
1055 static void
1056 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
1057 {
1058 	int error;
1059 
1060 	if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
1061 		panic("Error %d setting state to %d", error, newstate);
1062 }
1063 
1064 static void
1065 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func)
1066 {
1067 
1068 	KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked"));
1069 
1070 	/*
1071 	 * Update 'rendezvous_func' and execute a write memory barrier to
1072 	 * ensure that it is visible across all host cpus. This is not needed
1073 	 * for correctness but it does ensure that all the vcpus will notice
1074 	 * that the rendezvous is requested immediately.
1075 	 */
1076 	vm->rendezvous_func = func;
1077 	wmb();
1078 }
1079 
1080 #define	RENDEZVOUS_CTR0(vm, vcpuid, fmt)				\
1081 	do {								\
1082 		if (vcpuid >= 0)					\
1083 			VCPU_CTR0(vm, vcpuid, fmt);			\
1084 		else							\
1085 			VM_CTR0(vm, fmt);				\
1086 	} while (0)
1087 
1088 static void
1089 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1090 {
1091 
1092 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
1093 	    ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1094 
1095 	mtx_lock(&vm->rendezvous_mtx);
1096 	while (vm->rendezvous_func != NULL) {
1097 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1098 		CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1099 
1100 		if (vcpuid != -1 &&
1101 		    CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1102 		    !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1103 			VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1104 			(*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1105 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1106 		}
1107 		if (CPU_CMP(&vm->rendezvous_req_cpus,
1108 		    &vm->rendezvous_done_cpus) == 0) {
1109 			VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1110 			vm_set_rendezvous_func(vm, NULL);
1111 			wakeup(&vm->rendezvous_func);
1112 			break;
1113 		}
1114 		RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1115 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1116 		    "vmrndv", 0);
1117 	}
1118 	mtx_unlock(&vm->rendezvous_mtx);
1119 }
1120 
1121 /*
1122  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1123  */
1124 static int
1125 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1126 {
1127 	struct vcpu *vcpu;
1128 	const char *wmesg;
1129 	int t, vcpu_halted, vm_halted;
1130 
1131 	KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1132 
1133 	vcpu = &vm->vcpu[vcpuid];
1134 	vcpu_halted = 0;
1135 	vm_halted = 0;
1136 
1137 	vcpu_lock(vcpu);
1138 	while (1) {
1139 		/*
1140 		 * Do a final check for pending NMI or interrupts before
1141 		 * really putting this thread to sleep. Also check for
1142 		 * software events that would cause this vcpu to wakeup.
1143 		 *
1144 		 * These interrupts/events could have happened after the
1145 		 * vcpu returned from VMRUN() and before it acquired the
1146 		 * vcpu lock above.
1147 		 */
1148 		if (vm->rendezvous_func != NULL || vm->suspend)
1149 			break;
1150 		if (vm_nmi_pending(vm, vcpuid))
1151 			break;
1152 		if (!intr_disabled) {
1153 			if (vm_extint_pending(vm, vcpuid) ||
1154 			    vlapic_pending_intr(vcpu->vlapic, NULL)) {
1155 				break;
1156 			}
1157 		}
1158 
1159 		/* Don't go to sleep if the vcpu thread needs to yield */
1160 		if (vcpu_should_yield(vm, vcpuid))
1161 			break;
1162 
1163 		/*
1164 		 * Some Linux guests implement "halt" by having all vcpus
1165 		 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1166 		 * track of the vcpus that have entered this state. When all
1167 		 * vcpus enter the halted state the virtual machine is halted.
1168 		 */
1169 		if (intr_disabled) {
1170 			wmesg = "vmhalt";
1171 			VCPU_CTR0(vm, vcpuid, "Halted");
1172 			if (!vcpu_halted && halt_detection_enabled) {
1173 				vcpu_halted = 1;
1174 				CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1175 			}
1176 			if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1177 				vm_halted = 1;
1178 				break;
1179 			}
1180 		} else {
1181 			wmesg = "vmidle";
1182 		}
1183 
1184 		t = ticks;
1185 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1186 		/*
1187 		 * XXX msleep_spin() cannot be interrupted by signals so
1188 		 * wake up periodically to check pending signals.
1189 		 */
1190 		msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1191 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1192 		vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1193 	}
1194 
1195 	if (vcpu_halted)
1196 		CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1197 
1198 	vcpu_unlock(vcpu);
1199 
1200 	if (vm_halted)
1201 		vm_suspend(vm, VM_SUSPEND_HALT);
1202 
1203 	return (0);
1204 }
1205 
1206 static int
1207 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1208 {
1209 	int rv, ftype;
1210 	struct vm_map *map;
1211 	struct vcpu *vcpu;
1212 	struct vm_exit *vme;
1213 
1214 	vcpu = &vm->vcpu[vcpuid];
1215 	vme = &vcpu->exitinfo;
1216 
1217 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1218 	    __func__, vme->inst_length));
1219 
1220 	ftype = vme->u.paging.fault_type;
1221 	KASSERT(ftype == VM_PROT_READ ||
1222 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1223 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1224 
1225 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1226 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1227 		    vme->u.paging.gpa, ftype);
1228 		if (rv == 0) {
1229 			VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1230 			    ftype == VM_PROT_READ ? "accessed" : "dirty",
1231 			    vme->u.paging.gpa);
1232 			goto done;
1233 		}
1234 	}
1235 
1236 	map = &vm->vmspace->vm_map;
1237 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
1238 
1239 	VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1240 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1241 
1242 	if (rv != KERN_SUCCESS)
1243 		return (EFAULT);
1244 done:
1245 	return (0);
1246 }
1247 
1248 static int
1249 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1250 {
1251 	struct vie *vie;
1252 	struct vcpu *vcpu;
1253 	struct vm_exit *vme;
1254 	uint64_t gla, gpa, cs_base;
1255 	struct vm_guest_paging *paging;
1256 	mem_region_read_t mread;
1257 	mem_region_write_t mwrite;
1258 	enum vm_cpu_mode cpu_mode;
1259 	int cs_d, error, length;
1260 
1261 	vcpu = &vm->vcpu[vcpuid];
1262 	vme = &vcpu->exitinfo;
1263 
1264 	gla = vme->u.inst_emul.gla;
1265 	gpa = vme->u.inst_emul.gpa;
1266 	cs_base = vme->u.inst_emul.cs_base;
1267 	cs_d = vme->u.inst_emul.cs_d;
1268 	vie = &vme->u.inst_emul.vie;
1269 	paging = &vme->u.inst_emul.paging;
1270 	cpu_mode = paging->cpu_mode;
1271 
1272 	VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1273 
1274 	/* Fetch, decode and emulate the faulting instruction */
1275 	if (vie->num_valid == 0) {
1276 		/*
1277 		 * If the instruction length is not known then assume a
1278 		 * maximum size instruction.
1279 		 */
1280 		length = vme->inst_length ? vme->inst_length : VIE_INST_SIZE;
1281 		error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip +
1282 		    cs_base, length, vie);
1283 	} else {
1284 		/*
1285 		 * The instruction bytes have already been copied into 'vie'
1286 		 */
1287 		error = 0;
1288 	}
1289 	if (error == 1)
1290 		return (0);		/* Resume guest to handle page fault */
1291 	else if (error == -1)
1292 		return (EFAULT);
1293 	else if (error != 0)
1294 		panic("%s: vmm_fetch_instruction error %d", __func__, error);
1295 
1296 	if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) {
1297 		VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx",
1298 		    vme->rip + cs_base);
1299 		*retu = true;	    /* dump instruction bytes in userspace */
1300 		return (0);
1301 	}
1302 
1303 	/*
1304 	 * If the instruction length was not specified then update it now
1305 	 * along with 'nextrip'.
1306 	 */
1307 	if (vme->inst_length == 0) {
1308 		vme->inst_length = vie->num_processed;
1309 		vcpu->nextrip += vie->num_processed;
1310 	}
1311 
1312 	/* return to userland unless this is an in-kernel emulated device */
1313 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1314 		mread = lapic_mmio_read;
1315 		mwrite = lapic_mmio_write;
1316 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1317 		mread = vioapic_mmio_read;
1318 		mwrite = vioapic_mmio_write;
1319 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1320 		mread = vhpet_mmio_read;
1321 		mwrite = vhpet_mmio_write;
1322 	} else {
1323 		*retu = true;
1324 		return (0);
1325 	}
1326 
1327 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1328 	    mread, mwrite, retu);
1329 
1330 	return (error);
1331 }
1332 
1333 static int
1334 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1335 {
1336 	int i, done;
1337 	struct vcpu *vcpu;
1338 
1339 	done = 0;
1340 	vcpu = &vm->vcpu[vcpuid];
1341 
1342 	CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1343 
1344 	/*
1345 	 * Wait until all 'active_cpus' have suspended themselves.
1346 	 *
1347 	 * Since a VM may be suspended at any time including when one or
1348 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1349 	 * handler while we are waiting to prevent a deadlock.
1350 	 */
1351 	vcpu_lock(vcpu);
1352 	while (1) {
1353 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1354 			VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1355 			break;
1356 		}
1357 
1358 		if (vm->rendezvous_func == NULL) {
1359 			VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1360 			vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1361 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1362 			vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1363 		} else {
1364 			VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1365 			vcpu_unlock(vcpu);
1366 			vm_handle_rendezvous(vm, vcpuid);
1367 			vcpu_lock(vcpu);
1368 		}
1369 	}
1370 	vcpu_unlock(vcpu);
1371 
1372 	/*
1373 	 * Wakeup the other sleeping vcpus and return to userspace.
1374 	 */
1375 	for (i = 0; i < VM_MAXCPU; i++) {
1376 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1377 			vcpu_notify_event(vm, i, false);
1378 		}
1379 	}
1380 
1381 	*retu = true;
1382 	return (0);
1383 }
1384 
1385 int
1386 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1387 {
1388 	int i;
1389 
1390 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1391 		return (EINVAL);
1392 
1393 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1394 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1395 		    vm->suspend, how);
1396 		return (EALREADY);
1397 	}
1398 
1399 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1400 
1401 	/*
1402 	 * Notify all active vcpus that they are now suspended.
1403 	 */
1404 	for (i = 0; i < VM_MAXCPU; i++) {
1405 		if (CPU_ISSET(i, &vm->active_cpus))
1406 			vcpu_notify_event(vm, i, false);
1407 	}
1408 
1409 	return (0);
1410 }
1411 
1412 void
1413 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1414 {
1415 	struct vm_exit *vmexit;
1416 
1417 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1418 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1419 
1420 	vmexit = vm_exitinfo(vm, vcpuid);
1421 	vmexit->rip = rip;
1422 	vmexit->inst_length = 0;
1423 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1424 	vmexit->u.suspended.how = vm->suspend;
1425 }
1426 
1427 void
1428 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1429 {
1430 	struct vm_exit *vmexit;
1431 
1432 	KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1433 
1434 	vmexit = vm_exitinfo(vm, vcpuid);
1435 	vmexit->rip = rip;
1436 	vmexit->inst_length = 0;
1437 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1438 	vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1439 }
1440 
1441 void
1442 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1443 {
1444 	struct vm_exit *vmexit;
1445 
1446 	vmexit = vm_exitinfo(vm, vcpuid);
1447 	vmexit->rip = rip;
1448 	vmexit->inst_length = 0;
1449 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1450 	vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1451 }
1452 
1453 int
1454 vm_run(struct vm *vm, struct vm_run *vmrun)
1455 {
1456 	int error, vcpuid;
1457 	struct vcpu *vcpu;
1458 	struct pcb *pcb;
1459 	uint64_t tscval;
1460 	struct vm_exit *vme;
1461 	bool retu, intr_disabled;
1462 	pmap_t pmap;
1463 	void *rptr, *sptr;
1464 
1465 	vcpuid = vmrun->cpuid;
1466 
1467 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1468 		return (EINVAL);
1469 
1470 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1471 		return (EINVAL);
1472 
1473 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1474 		return (EINVAL);
1475 
1476 	rptr = &vm->rendezvous_func;
1477 	sptr = &vm->suspend;
1478 	pmap = vmspace_pmap(vm->vmspace);
1479 	vcpu = &vm->vcpu[vcpuid];
1480 	vme = &vcpu->exitinfo;
1481 restart:
1482 	critical_enter();
1483 
1484 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1485 	    ("vm_run: absurd pm_active"));
1486 
1487 	tscval = rdtsc();
1488 
1489 	pcb = PCPU_GET(curpcb);
1490 	set_pcb_flags(pcb, PCB_FULL_IRET);
1491 
1492 	restore_guest_fpustate(vcpu);
1493 
1494 	vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1495 	error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, rptr, sptr);
1496 	vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1497 
1498 	save_guest_fpustate(vcpu);
1499 
1500 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1501 
1502 	critical_exit();
1503 
1504 	if (error == 0) {
1505 		retu = false;
1506 		vcpu->nextrip = vme->rip + vme->inst_length;
1507 		switch (vme->exitcode) {
1508 		case VM_EXITCODE_SUSPENDED:
1509 			error = vm_handle_suspend(vm, vcpuid, &retu);
1510 			break;
1511 		case VM_EXITCODE_IOAPIC_EOI:
1512 			vioapic_process_eoi(vm, vcpuid,
1513 			    vme->u.ioapic_eoi.vector);
1514 			break;
1515 		case VM_EXITCODE_RENDEZVOUS:
1516 			vm_handle_rendezvous(vm, vcpuid);
1517 			error = 0;
1518 			break;
1519 		case VM_EXITCODE_HLT:
1520 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1521 			error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1522 			break;
1523 		case VM_EXITCODE_PAGING:
1524 			error = vm_handle_paging(vm, vcpuid, &retu);
1525 			break;
1526 		case VM_EXITCODE_INST_EMUL:
1527 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1528 			break;
1529 		case VM_EXITCODE_INOUT:
1530 		case VM_EXITCODE_INOUT_STR:
1531 			error = vm_handle_inout(vm, vcpuid, vme, &retu);
1532 			break;
1533 		case VM_EXITCODE_MONITOR:
1534 		case VM_EXITCODE_MWAIT:
1535 			vm_inject_ud(vm, vcpuid);
1536 			break;
1537 		default:
1538 			retu = true;	/* handled in userland */
1539 			break;
1540 		}
1541 	}
1542 
1543 	if (error == 0 && retu == false)
1544 		goto restart;
1545 
1546 	/* copy the exit information */
1547 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1548 	return (error);
1549 }
1550 
1551 int
1552 vm_restart_instruction(void *arg, int vcpuid)
1553 {
1554 	struct vm *vm;
1555 	struct vcpu *vcpu;
1556 	enum vcpu_state state;
1557 	uint64_t rip;
1558 	int error;
1559 
1560 	vm = arg;
1561 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1562 		return (EINVAL);
1563 
1564 	vcpu = &vm->vcpu[vcpuid];
1565 	state = vcpu_get_state(vm, vcpuid, NULL);
1566 	if (state == VCPU_RUNNING) {
1567 		/*
1568 		 * When a vcpu is "running" the next instruction is determined
1569 		 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
1570 		 * Thus setting 'inst_length' to zero will cause the current
1571 		 * instruction to be restarted.
1572 		 */
1573 		vcpu->exitinfo.inst_length = 0;
1574 		VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by "
1575 		    "setting inst_length to zero", vcpu->exitinfo.rip);
1576 	} else if (state == VCPU_FROZEN) {
1577 		/*
1578 		 * When a vcpu is "frozen" it is outside the critical section
1579 		 * around VMRUN() and 'nextrip' points to the next instruction.
1580 		 * Thus instruction restart is achieved by setting 'nextrip'
1581 		 * to the vcpu's %rip.
1582 		 */
1583 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip);
1584 		KASSERT(!error, ("%s: error %d getting rip", __func__, error));
1585 		VCPU_CTR2(vm, vcpuid, "restarting instruction by updating "
1586 		    "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
1587 		vcpu->nextrip = rip;
1588 	} else {
1589 		panic("%s: invalid state %d", __func__, state);
1590 	}
1591 	return (0);
1592 }
1593 
1594 int
1595 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1596 {
1597 	struct vcpu *vcpu;
1598 	int type, vector;
1599 
1600 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1601 		return (EINVAL);
1602 
1603 	vcpu = &vm->vcpu[vcpuid];
1604 
1605 	if (info & VM_INTINFO_VALID) {
1606 		type = info & VM_INTINFO_TYPE;
1607 		vector = info & 0xff;
1608 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1609 			return (EINVAL);
1610 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1611 			return (EINVAL);
1612 		if (info & VM_INTINFO_RSVD)
1613 			return (EINVAL);
1614 	} else {
1615 		info = 0;
1616 	}
1617 	VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1618 	vcpu->exitintinfo = info;
1619 	return (0);
1620 }
1621 
1622 enum exc_class {
1623 	EXC_BENIGN,
1624 	EXC_CONTRIBUTORY,
1625 	EXC_PAGEFAULT
1626 };
1627 
1628 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
1629 
1630 static enum exc_class
1631 exception_class(uint64_t info)
1632 {
1633 	int type, vector;
1634 
1635 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1636 	type = info & VM_INTINFO_TYPE;
1637 	vector = info & 0xff;
1638 
1639 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1640 	switch (type) {
1641 	case VM_INTINFO_HWINTR:
1642 	case VM_INTINFO_SWINTR:
1643 	case VM_INTINFO_NMI:
1644 		return (EXC_BENIGN);
1645 	default:
1646 		/*
1647 		 * Hardware exception.
1648 		 *
1649 		 * SVM and VT-x use identical type values to represent NMI,
1650 		 * hardware interrupt and software interrupt.
1651 		 *
1652 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1653 		 * for exceptions except #BP and #OF. #BP and #OF use a type
1654 		 * value of '5' or '6'. Therefore we don't check for explicit
1655 		 * values of 'type' to classify 'intinfo' into a hardware
1656 		 * exception.
1657 		 */
1658 		break;
1659 	}
1660 
1661 	switch (vector) {
1662 	case IDT_PF:
1663 	case IDT_VE:
1664 		return (EXC_PAGEFAULT);
1665 	case IDT_DE:
1666 	case IDT_TS:
1667 	case IDT_NP:
1668 	case IDT_SS:
1669 	case IDT_GP:
1670 		return (EXC_CONTRIBUTORY);
1671 	default:
1672 		return (EXC_BENIGN);
1673 	}
1674 }
1675 
1676 static int
1677 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1678     uint64_t *retinfo)
1679 {
1680 	enum exc_class exc1, exc2;
1681 	int type1, vector1;
1682 
1683 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1684 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1685 
1686 	/*
1687 	 * If an exception occurs while attempting to call the double-fault
1688 	 * handler the processor enters shutdown mode (aka triple fault).
1689 	 */
1690 	type1 = info1 & VM_INTINFO_TYPE;
1691 	vector1 = info1 & 0xff;
1692 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1693 		VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1694 		    info1, info2);
1695 		vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1696 		*retinfo = 0;
1697 		return (0);
1698 	}
1699 
1700 	/*
1701 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1702 	 */
1703 	exc1 = exception_class(info1);
1704 	exc2 = exception_class(info2);
1705 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1706 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1707 		/* Convert nested fault into a double fault. */
1708 		*retinfo = IDT_DF;
1709 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1710 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
1711 	} else {
1712 		/* Handle exceptions serially */
1713 		*retinfo = info2;
1714 	}
1715 	return (1);
1716 }
1717 
1718 static uint64_t
1719 vcpu_exception_intinfo(struct vcpu *vcpu)
1720 {
1721 	uint64_t info = 0;
1722 
1723 	if (vcpu->exception_pending) {
1724 		info = vcpu->exc_vector & 0xff;
1725 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1726 		if (vcpu->exc_errcode_valid) {
1727 			info |= VM_INTINFO_DEL_ERRCODE;
1728 			info |= (uint64_t)vcpu->exc_errcode << 32;
1729 		}
1730 	}
1731 	return (info);
1732 }
1733 
1734 int
1735 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1736 {
1737 	struct vcpu *vcpu;
1738 	uint64_t info1, info2;
1739 	int valid;
1740 
1741 	KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid));
1742 
1743 	vcpu = &vm->vcpu[vcpuid];
1744 
1745 	info1 = vcpu->exitintinfo;
1746 	vcpu->exitintinfo = 0;
1747 
1748 	info2 = 0;
1749 	if (vcpu->exception_pending) {
1750 		info2 = vcpu_exception_intinfo(vcpu);
1751 		vcpu->exception_pending = 0;
1752 		VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1753 		    vcpu->exc_vector, info2);
1754 	}
1755 
1756 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
1757 		valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
1758 	} else if (info1 & VM_INTINFO_VALID) {
1759 		*retinfo = info1;
1760 		valid = 1;
1761 	} else if (info2 & VM_INTINFO_VALID) {
1762 		*retinfo = info2;
1763 		valid = 1;
1764 	} else {
1765 		valid = 0;
1766 	}
1767 
1768 	if (valid) {
1769 		VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
1770 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
1771 	}
1772 
1773 	return (valid);
1774 }
1775 
1776 int
1777 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
1778 {
1779 	struct vcpu *vcpu;
1780 
1781 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1782 		return (EINVAL);
1783 
1784 	vcpu = &vm->vcpu[vcpuid];
1785 	*info1 = vcpu->exitintinfo;
1786 	*info2 = vcpu_exception_intinfo(vcpu);
1787 	return (0);
1788 }
1789 
1790 int
1791 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid,
1792     uint32_t errcode, int restart_instruction)
1793 {
1794 	struct vcpu *vcpu;
1795 	int error;
1796 
1797 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1798 		return (EINVAL);
1799 
1800 	if (vector < 0 || vector >= 32)
1801 		return (EINVAL);
1802 
1803 	/*
1804 	 * A double fault exception should never be injected directly into
1805 	 * the guest. It is a derived exception that results from specific
1806 	 * combinations of nested faults.
1807 	 */
1808 	if (vector == IDT_DF)
1809 		return (EINVAL);
1810 
1811 	vcpu = &vm->vcpu[vcpuid];
1812 
1813 	if (vcpu->exception_pending) {
1814 		VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
1815 		    "pending exception %d", vector, vcpu->exc_vector);
1816 		return (EBUSY);
1817 	}
1818 
1819 	/*
1820 	 * From section 26.6.1 "Interruptibility State" in Intel SDM:
1821 	 *
1822 	 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
1823 	 * one instruction or incurs an exception.
1824 	 */
1825 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
1826 	KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
1827 	    __func__, error));
1828 
1829 	if (restart_instruction)
1830 		vm_restart_instruction(vm, vcpuid);
1831 
1832 	vcpu->exception_pending = 1;
1833 	vcpu->exc_vector = vector;
1834 	vcpu->exc_errcode = errcode;
1835 	vcpu->exc_errcode_valid = errcode_valid;
1836 	VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector);
1837 	return (0);
1838 }
1839 
1840 void
1841 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
1842     int errcode)
1843 {
1844 	struct vm *vm;
1845 	int error, restart_instruction;
1846 
1847 	vm = vmarg;
1848 	restart_instruction = 1;
1849 
1850 	error = vm_inject_exception(vm, vcpuid, vector, errcode_valid,
1851 	    errcode, restart_instruction);
1852 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
1853 }
1854 
1855 void
1856 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
1857 {
1858 	struct vm *vm;
1859 	int error;
1860 
1861 	vm = vmarg;
1862 	VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
1863 	    error_code, cr2);
1864 
1865 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
1866 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
1867 
1868 	vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
1869 }
1870 
1871 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
1872 
1873 int
1874 vm_inject_nmi(struct vm *vm, int vcpuid)
1875 {
1876 	struct vcpu *vcpu;
1877 
1878 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1879 		return (EINVAL);
1880 
1881 	vcpu = &vm->vcpu[vcpuid];
1882 
1883 	vcpu->nmi_pending = 1;
1884 	vcpu_notify_event(vm, vcpuid, false);
1885 	return (0);
1886 }
1887 
1888 int
1889 vm_nmi_pending(struct vm *vm, int vcpuid)
1890 {
1891 	struct vcpu *vcpu;
1892 
1893 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1894 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1895 
1896 	vcpu = &vm->vcpu[vcpuid];
1897 
1898 	return (vcpu->nmi_pending);
1899 }
1900 
1901 void
1902 vm_nmi_clear(struct vm *vm, int vcpuid)
1903 {
1904 	struct vcpu *vcpu;
1905 
1906 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1907 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1908 
1909 	vcpu = &vm->vcpu[vcpuid];
1910 
1911 	if (vcpu->nmi_pending == 0)
1912 		panic("vm_nmi_clear: inconsistent nmi_pending state");
1913 
1914 	vcpu->nmi_pending = 0;
1915 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
1916 }
1917 
1918 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
1919 
1920 int
1921 vm_inject_extint(struct vm *vm, int vcpuid)
1922 {
1923 	struct vcpu *vcpu;
1924 
1925 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1926 		return (EINVAL);
1927 
1928 	vcpu = &vm->vcpu[vcpuid];
1929 
1930 	vcpu->extint_pending = 1;
1931 	vcpu_notify_event(vm, vcpuid, false);
1932 	return (0);
1933 }
1934 
1935 int
1936 vm_extint_pending(struct vm *vm, int vcpuid)
1937 {
1938 	struct vcpu *vcpu;
1939 
1940 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1941 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1942 
1943 	vcpu = &vm->vcpu[vcpuid];
1944 
1945 	return (vcpu->extint_pending);
1946 }
1947 
1948 void
1949 vm_extint_clear(struct vm *vm, int vcpuid)
1950 {
1951 	struct vcpu *vcpu;
1952 
1953 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1954 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1955 
1956 	vcpu = &vm->vcpu[vcpuid];
1957 
1958 	if (vcpu->extint_pending == 0)
1959 		panic("vm_extint_clear: inconsistent extint_pending state");
1960 
1961 	vcpu->extint_pending = 0;
1962 	vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
1963 }
1964 
1965 int
1966 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
1967 {
1968 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1969 		return (EINVAL);
1970 
1971 	if (type < 0 || type >= VM_CAP_MAX)
1972 		return (EINVAL);
1973 
1974 	return (VMGETCAP(vm->cookie, vcpu, type, retval));
1975 }
1976 
1977 int
1978 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
1979 {
1980 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1981 		return (EINVAL);
1982 
1983 	if (type < 0 || type >= VM_CAP_MAX)
1984 		return (EINVAL);
1985 
1986 	return (VMSETCAP(vm->cookie, vcpu, type, val));
1987 }
1988 
1989 struct vlapic *
1990 vm_lapic(struct vm *vm, int cpu)
1991 {
1992 	return (vm->vcpu[cpu].vlapic);
1993 }
1994 
1995 struct vioapic *
1996 vm_ioapic(struct vm *vm)
1997 {
1998 
1999 	return (vm->vioapic);
2000 }
2001 
2002 struct vhpet *
2003 vm_hpet(struct vm *vm)
2004 {
2005 
2006 	return (vm->vhpet);
2007 }
2008 
2009 boolean_t
2010 vmm_is_pptdev(int bus, int slot, int func)
2011 {
2012 	int found, i, n;
2013 	int b, s, f;
2014 	char *val, *cp, *cp2;
2015 
2016 	/*
2017 	 * XXX
2018 	 * The length of an environment variable is limited to 128 bytes which
2019 	 * puts an upper limit on the number of passthru devices that may be
2020 	 * specified using a single environment variable.
2021 	 *
2022 	 * Work around this by scanning multiple environment variable
2023 	 * names instead of a single one - yuck!
2024 	 */
2025 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2026 
2027 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2028 	found = 0;
2029 	for (i = 0; names[i] != NULL && !found; i++) {
2030 		cp = val = kern_getenv(names[i]);
2031 		while (cp != NULL && *cp != '\0') {
2032 			if ((cp2 = strchr(cp, ' ')) != NULL)
2033 				*cp2 = '\0';
2034 
2035 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2036 			if (n == 3 && bus == b && slot == s && func == f) {
2037 				found = 1;
2038 				break;
2039 			}
2040 
2041 			if (cp2 != NULL)
2042 				*cp2++ = ' ';
2043 
2044 			cp = cp2;
2045 		}
2046 		freeenv(val);
2047 	}
2048 	return (found);
2049 }
2050 
2051 void *
2052 vm_iommu_domain(struct vm *vm)
2053 {
2054 
2055 	return (vm->iommu);
2056 }
2057 
2058 int
2059 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
2060     bool from_idle)
2061 {
2062 	int error;
2063 	struct vcpu *vcpu;
2064 
2065 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2066 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
2067 
2068 	vcpu = &vm->vcpu[vcpuid];
2069 
2070 	vcpu_lock(vcpu);
2071 	error = vcpu_set_state_locked(vcpu, newstate, from_idle);
2072 	vcpu_unlock(vcpu);
2073 
2074 	return (error);
2075 }
2076 
2077 enum vcpu_state
2078 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2079 {
2080 	struct vcpu *vcpu;
2081 	enum vcpu_state state;
2082 
2083 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2084 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2085 
2086 	vcpu = &vm->vcpu[vcpuid];
2087 
2088 	vcpu_lock(vcpu);
2089 	state = vcpu->state;
2090 	if (hostcpu != NULL)
2091 		*hostcpu = vcpu->hostcpu;
2092 	vcpu_unlock(vcpu);
2093 
2094 	return (state);
2095 }
2096 
2097 int
2098 vm_activate_cpu(struct vm *vm, int vcpuid)
2099 {
2100 
2101 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2102 		return (EINVAL);
2103 
2104 	if (CPU_ISSET(vcpuid, &vm->active_cpus))
2105 		return (EBUSY);
2106 
2107 	VCPU_CTR0(vm, vcpuid, "activated");
2108 	CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2109 	return (0);
2110 }
2111 
2112 cpuset_t
2113 vm_active_cpus(struct vm *vm)
2114 {
2115 
2116 	return (vm->active_cpus);
2117 }
2118 
2119 cpuset_t
2120 vm_suspended_cpus(struct vm *vm)
2121 {
2122 
2123 	return (vm->suspended_cpus);
2124 }
2125 
2126 void *
2127 vcpu_stats(struct vm *vm, int vcpuid)
2128 {
2129 
2130 	return (vm->vcpu[vcpuid].stats);
2131 }
2132 
2133 int
2134 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2135 {
2136 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2137 		return (EINVAL);
2138 
2139 	*state = vm->vcpu[vcpuid].x2apic_state;
2140 
2141 	return (0);
2142 }
2143 
2144 int
2145 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2146 {
2147 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2148 		return (EINVAL);
2149 
2150 	if (state >= X2APIC_STATE_LAST)
2151 		return (EINVAL);
2152 
2153 	vm->vcpu[vcpuid].x2apic_state = state;
2154 
2155 	vlapic_set_x2apic_state(vm, vcpuid, state);
2156 
2157 	return (0);
2158 }
2159 
2160 /*
2161  * This function is called to ensure that a vcpu "sees" a pending event
2162  * as soon as possible:
2163  * - If the vcpu thread is sleeping then it is woken up.
2164  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2165  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2166  */
2167 void
2168 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2169 {
2170 	int hostcpu;
2171 	struct vcpu *vcpu;
2172 
2173 	vcpu = &vm->vcpu[vcpuid];
2174 
2175 	vcpu_lock(vcpu);
2176 	hostcpu = vcpu->hostcpu;
2177 	if (vcpu->state == VCPU_RUNNING) {
2178 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2179 		if (hostcpu != curcpu) {
2180 			if (lapic_intr) {
2181 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2182 				    vmm_ipinum);
2183 			} else {
2184 				ipi_cpu(hostcpu, vmm_ipinum);
2185 			}
2186 		} else {
2187 			/*
2188 			 * If the 'vcpu' is running on 'curcpu' then it must
2189 			 * be sending a notification to itself (e.g. SELF_IPI).
2190 			 * The pending event will be picked up when the vcpu
2191 			 * transitions back to guest context.
2192 			 */
2193 		}
2194 	} else {
2195 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2196 		    "with hostcpu %d", vcpu->state, hostcpu));
2197 		if (vcpu->state == VCPU_SLEEPING)
2198 			wakeup_one(vcpu);
2199 	}
2200 	vcpu_unlock(vcpu);
2201 }
2202 
2203 struct vmspace *
2204 vm_get_vmspace(struct vm *vm)
2205 {
2206 
2207 	return (vm->vmspace);
2208 }
2209 
2210 int
2211 vm_apicid2vcpuid(struct vm *vm, int apicid)
2212 {
2213 	/*
2214 	 * XXX apic id is assumed to be numerically identical to vcpu id
2215 	 */
2216 	return (apicid);
2217 }
2218 
2219 void
2220 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2221     vm_rendezvous_func_t func, void *arg)
2222 {
2223 	int i;
2224 
2225 	/*
2226 	 * Enforce that this function is called without any locks
2227 	 */
2228 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2229 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
2230 	    ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2231 
2232 restart:
2233 	mtx_lock(&vm->rendezvous_mtx);
2234 	if (vm->rendezvous_func != NULL) {
2235 		/*
2236 		 * If a rendezvous is already in progress then we need to
2237 		 * call the rendezvous handler in case this 'vcpuid' is one
2238 		 * of the targets of the rendezvous.
2239 		 */
2240 		RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2241 		mtx_unlock(&vm->rendezvous_mtx);
2242 		vm_handle_rendezvous(vm, vcpuid);
2243 		goto restart;
2244 	}
2245 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2246 	    "rendezvous is still in progress"));
2247 
2248 	RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2249 	vm->rendezvous_req_cpus = dest;
2250 	CPU_ZERO(&vm->rendezvous_done_cpus);
2251 	vm->rendezvous_arg = arg;
2252 	vm_set_rendezvous_func(vm, func);
2253 	mtx_unlock(&vm->rendezvous_mtx);
2254 
2255 	/*
2256 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2257 	 * vcpus so they handle the rendezvous as soon as possible.
2258 	 */
2259 	for (i = 0; i < VM_MAXCPU; i++) {
2260 		if (CPU_ISSET(i, &dest))
2261 			vcpu_notify_event(vm, i, false);
2262 	}
2263 
2264 	vm_handle_rendezvous(vm, vcpuid);
2265 }
2266 
2267 struct vatpic *
2268 vm_atpic(struct vm *vm)
2269 {
2270 	return (vm->vatpic);
2271 }
2272 
2273 struct vatpit *
2274 vm_atpit(struct vm *vm)
2275 {
2276 	return (vm->vatpit);
2277 }
2278 
2279 struct vpmtmr *
2280 vm_pmtmr(struct vm *vm)
2281 {
2282 
2283 	return (vm->vpmtmr);
2284 }
2285 
2286 struct vrtc *
2287 vm_rtc(struct vm *vm)
2288 {
2289 
2290 	return (vm->vrtc);
2291 }
2292 
2293 enum vm_reg_name
2294 vm_segment_name(int seg)
2295 {
2296 	static enum vm_reg_name seg_names[] = {
2297 		VM_REG_GUEST_ES,
2298 		VM_REG_GUEST_CS,
2299 		VM_REG_GUEST_SS,
2300 		VM_REG_GUEST_DS,
2301 		VM_REG_GUEST_FS,
2302 		VM_REG_GUEST_GS
2303 	};
2304 
2305 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2306 	    ("%s: invalid segment encoding %d", __func__, seg));
2307 	return (seg_names[seg]);
2308 }
2309 
2310 void
2311 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2312     int num_copyinfo)
2313 {
2314 	int idx;
2315 
2316 	for (idx = 0; idx < num_copyinfo; idx++) {
2317 		if (copyinfo[idx].cookie != NULL)
2318 			vm_gpa_release(copyinfo[idx].cookie);
2319 	}
2320 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2321 }
2322 
2323 int
2324 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2325     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2326     int num_copyinfo)
2327 {
2328 	int error, idx, nused;
2329 	size_t n, off, remaining;
2330 	void *hva, *cookie;
2331 	uint64_t gpa;
2332 
2333 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2334 
2335 	nused = 0;
2336 	remaining = len;
2337 	while (remaining > 0) {
2338 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2339 		error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa);
2340 		if (error)
2341 			return (error);
2342 		off = gpa & PAGE_MASK;
2343 		n = min(remaining, PAGE_SIZE - off);
2344 		copyinfo[nused].gpa = gpa;
2345 		copyinfo[nused].len = n;
2346 		remaining -= n;
2347 		gla += n;
2348 		nused++;
2349 	}
2350 
2351 	for (idx = 0; idx < nused; idx++) {
2352 		hva = vm_gpa_hold(vm, copyinfo[idx].gpa, copyinfo[idx].len,
2353 		    prot, &cookie);
2354 		if (hva == NULL)
2355 			break;
2356 		copyinfo[idx].hva = hva;
2357 		copyinfo[idx].cookie = cookie;
2358 	}
2359 
2360 	if (idx != nused) {
2361 		vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2362 		return (-1);
2363 	} else {
2364 		return (0);
2365 	}
2366 }
2367 
2368 void
2369 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2370     size_t len)
2371 {
2372 	char *dst;
2373 	int idx;
2374 
2375 	dst = kaddr;
2376 	idx = 0;
2377 	while (len > 0) {
2378 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2379 		len -= copyinfo[idx].len;
2380 		dst += copyinfo[idx].len;
2381 		idx++;
2382 	}
2383 }
2384 
2385 void
2386 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2387     struct vm_copyinfo *copyinfo, size_t len)
2388 {
2389 	const char *src;
2390 	int idx;
2391 
2392 	src = kaddr;
2393 	idx = 0;
2394 	while (len > 0) {
2395 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2396 		len -= copyinfo[idx].len;
2397 		src += copyinfo[idx].len;
2398 		idx++;
2399 	}
2400 }
2401 
2402 /*
2403  * Return the amount of in-use and wired memory for the VM. Since
2404  * these are global stats, only return the values with for vCPU 0
2405  */
2406 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2407 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2408 
2409 static void
2410 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2411 {
2412 
2413 	if (vcpu == 0) {
2414 		vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2415 	       	    PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2416 	}
2417 }
2418 
2419 static void
2420 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2421 {
2422 
2423 	if (vcpu == 0) {
2424 		vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2425 	      	    PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2426 	}
2427 }
2428 
2429 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2430 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2431