xref: /freebsd/sys/amd64/vmm/vmm.c (revision e17f5b1d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011 NetApp, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  *
28  * $FreeBSD$
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_bhyve_snapshot.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
39 #include <sys/module.h>
40 #include <sys/sysctl.h>
41 #include <sys/malloc.h>
42 #include <sys/pcpu.h>
43 #include <sys/lock.h>
44 #include <sys/mutex.h>
45 #include <sys/proc.h>
46 #include <sys/rwlock.h>
47 #include <sys/sched.h>
48 #include <sys/smp.h>
49 #include <sys/vnode.h>
50 
51 #include <vm/vm.h>
52 #include <vm/vm_object.h>
53 #include <vm/vm_page.h>
54 #include <vm/pmap.h>
55 #include <vm/vm_map.h>
56 #include <vm/vm_extern.h>
57 #include <vm/vm_param.h>
58 #include <vm/vm_pager.h>
59 #include <vm/vm_kern.h>
60 #include <vm/vnode_pager.h>
61 #include <vm/swap_pager.h>
62 #include <vm/uma.h>
63 
64 #include <machine/cpu.h>
65 #include <machine/pcb.h>
66 #include <machine/smp.h>
67 #include <machine/md_var.h>
68 #include <x86/psl.h>
69 #include <x86/apicreg.h>
70 
71 #include <machine/vmm.h>
72 #include <machine/vmm_dev.h>
73 #include <machine/vmm_instruction_emul.h>
74 #include <machine/vmm_snapshot.h>
75 
76 #include "vmm_ioport.h"
77 #include "vmm_ktr.h"
78 #include "vmm_host.h"
79 #include "vmm_mem.h"
80 #include "vmm_util.h"
81 #include "vatpic.h"
82 #include "vatpit.h"
83 #include "vhpet.h"
84 #include "vioapic.h"
85 #include "vlapic.h"
86 #include "vpmtmr.h"
87 #include "vrtc.h"
88 #include "vmm_stat.h"
89 #include "vmm_lapic.h"
90 
91 #include "io/ppt.h"
92 #include "io/iommu.h"
93 
94 struct vlapic;
95 
96 /*
97  * Initialization:
98  * (a) allocated when vcpu is created
99  * (i) initialized when vcpu is created and when it is reinitialized
100  * (o) initialized the first time the vcpu is created
101  * (x) initialized before use
102  */
103 struct vcpu {
104 	struct mtx 	mtx;		/* (o) protects 'state' and 'hostcpu' */
105 	enum vcpu_state	state;		/* (o) vcpu state */
106 	int		hostcpu;	/* (o) vcpu's host cpu */
107 	int		reqidle;	/* (i) request vcpu to idle */
108 	struct vlapic	*vlapic;	/* (i) APIC device model */
109 	enum x2apic_state x2apic_state;	/* (i) APIC mode */
110 	uint64_t	exitintinfo;	/* (i) events pending at VM exit */
111 	int		nmi_pending;	/* (i) NMI pending */
112 	int		extint_pending;	/* (i) INTR pending */
113 	int	exception_pending;	/* (i) exception pending */
114 	int	exc_vector;		/* (x) exception collateral */
115 	int	exc_errcode_valid;
116 	uint32_t exc_errcode;
117 	struct savefpu	*guestfpu;	/* (a,i) guest fpu state */
118 	uint64_t	guest_xcr0;	/* (i) guest %xcr0 register */
119 	void		*stats;		/* (a,i) statistics */
120 	struct vm_exit	exitinfo;	/* (x) exit reason and collateral */
121 	uint64_t	nextrip;	/* (x) next instruction to execute */
122 	uint64_t	tsc_offset;	/* (o) TSC offsetting */
123 };
124 
125 #define	vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
126 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
127 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
128 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
129 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
130 
131 struct mem_seg {
132 	size_t	len;
133 	bool	sysmem;
134 	struct vm_object *object;
135 };
136 #define	VM_MAX_MEMSEGS	3
137 
138 struct mem_map {
139 	vm_paddr_t	gpa;
140 	size_t		len;
141 	vm_ooffset_t	segoff;
142 	int		segid;
143 	int		prot;
144 	int		flags;
145 };
146 #define	VM_MAX_MEMMAPS	8
147 
148 /*
149  * Initialization:
150  * (o) initialized the first time the VM is created
151  * (i) initialized when VM is created and when it is reinitialized
152  * (x) initialized before use
153  */
154 struct vm {
155 	void		*cookie;		/* (i) cpu-specific data */
156 	void		*iommu;			/* (x) iommu-specific data */
157 	struct vhpet	*vhpet;			/* (i) virtual HPET */
158 	struct vioapic	*vioapic;		/* (i) virtual ioapic */
159 	struct vatpic	*vatpic;		/* (i) virtual atpic */
160 	struct vatpit	*vatpit;		/* (i) virtual atpit */
161 	struct vpmtmr	*vpmtmr;		/* (i) virtual ACPI PM timer */
162 	struct vrtc	*vrtc;			/* (o) virtual RTC */
163 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
164 	volatile cpuset_t debug_cpus;		/* (i) vcpus stopped for debug */
165 	int		suspend;		/* (i) stop VM execution */
166 	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
167 	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
168 	cpuset_t	rendezvous_req_cpus;	/* (x) rendezvous requested */
169 	cpuset_t	rendezvous_done_cpus;	/* (x) rendezvous finished */
170 	void		*rendezvous_arg;	/* (x) rendezvous func/arg */
171 	vm_rendezvous_func_t rendezvous_func;
172 	struct mtx	rendezvous_mtx;		/* (o) rendezvous lock */
173 	struct mem_map	mem_maps[VM_MAX_MEMMAPS]; /* (i) guest address space */
174 	struct mem_seg	mem_segs[VM_MAX_MEMSEGS]; /* (o) guest memory regions */
175 	struct vmspace	*vmspace;		/* (o) guest's address space */
176 	char		name[VM_MAX_NAMELEN];	/* (o) virtual machine name */
177 	struct vcpu	vcpu[VM_MAXCPU];	/* (i) guest vcpus */
178 	/* The following describe the vm cpu topology */
179 	uint16_t	sockets;		/* (o) num of sockets */
180 	uint16_t	cores;			/* (o) num of cores/socket */
181 	uint16_t	threads;		/* (o) num of threads/core */
182 	uint16_t	maxcpus;		/* (o) max pluggable cpus */
183 };
184 
185 static int vmm_initialized;
186 
187 static struct vmm_ops *ops;
188 #define	VMM_INIT(num)	(ops != NULL ? (*ops->init)(num) : 0)
189 #define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
190 #define	VMM_RESUME()	(ops != NULL ? (*ops->resume)() : 0)
191 
192 #define	VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
193 #define	VMRUN(vmi, vcpu, rip, pmap, evinfo) \
194 	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, evinfo) : ENXIO)
195 #define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
196 #define	VMSPACE_ALLOC(min, max) \
197 	(ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
198 #define	VMSPACE_FREE(vmspace) \
199 	(ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
200 #define	VMGETREG(vmi, vcpu, num, retval)		\
201 	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
202 #define	VMSETREG(vmi, vcpu, num, val)		\
203 	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
204 #define	VMGETDESC(vmi, vcpu, num, desc)		\
205 	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
206 #define	VMSETDESC(vmi, vcpu, num, desc)		\
207 	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
208 #define	VMGETCAP(vmi, vcpu, num, retval)	\
209 	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
210 #define	VMSETCAP(vmi, vcpu, num, val)		\
211 	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
212 #define	VLAPIC_INIT(vmi, vcpu)			\
213 	(ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
214 #define	VLAPIC_CLEANUP(vmi, vlapic)		\
215 	(ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
216 #ifdef BHYVE_SNAPSHOT
217 #define	VM_SNAPSHOT_VMI(vmi, meta) \
218 	(ops != NULL ? (*ops->vmsnapshot)(vmi, meta) : ENXIO)
219 #define	VM_SNAPSHOT_VMCX(vmi, meta, vcpuid) \
220 	(ops != NULL ? (*ops->vmcx_snapshot)(vmi, meta, vcpuid) : ENXIO)
221 #define	VM_RESTORE_TSC(vmi, vcpuid, offset) \
222 	(ops != NULL ? (*ops->vm_restore_tsc)(vmi, vcpuid, offset) : ENXIO)
223 #endif
224 
225 #define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
226 #define	fpu_stop_emulating()	clts()
227 
228 SDT_PROVIDER_DEFINE(vmm);
229 
230 static MALLOC_DEFINE(M_VM, "vm", "vm");
231 
232 /* statistics */
233 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
234 
235 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
236     NULL);
237 
238 /*
239  * Halt the guest if all vcpus are executing a HLT instruction with
240  * interrupts disabled.
241  */
242 static int halt_detection_enabled = 1;
243 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
244     &halt_detection_enabled, 0,
245     "Halt VM if all vcpus execute HLT with interrupts disabled");
246 
247 static int vmm_ipinum;
248 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
249     "IPI vector used for vcpu notifications");
250 
251 static int trace_guest_exceptions;
252 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN,
253     &trace_guest_exceptions, 0,
254     "Trap into hypervisor on all guest exceptions and reflect them back");
255 
256 static void vm_free_memmap(struct vm *vm, int ident);
257 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm);
258 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr);
259 
260 #ifdef KTR
261 static const char *
262 vcpu_state2str(enum vcpu_state state)
263 {
264 
265 	switch (state) {
266 	case VCPU_IDLE:
267 		return ("idle");
268 	case VCPU_FROZEN:
269 		return ("frozen");
270 	case VCPU_RUNNING:
271 		return ("running");
272 	case VCPU_SLEEPING:
273 		return ("sleeping");
274 	default:
275 		return ("unknown");
276 	}
277 }
278 #endif
279 
280 static void
281 vcpu_cleanup(struct vm *vm, int i, bool destroy)
282 {
283 	struct vcpu *vcpu = &vm->vcpu[i];
284 
285 	VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
286 	if (destroy) {
287 		vmm_stat_free(vcpu->stats);
288 		fpu_save_area_free(vcpu->guestfpu);
289 	}
290 }
291 
292 static void
293 vcpu_init(struct vm *vm, int vcpu_id, bool create)
294 {
295 	struct vcpu *vcpu;
296 
297 	KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus,
298 	    ("vcpu_init: invalid vcpu %d", vcpu_id));
299 
300 	vcpu = &vm->vcpu[vcpu_id];
301 
302 	if (create) {
303 		KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
304 		    "initialized", vcpu_id));
305 		vcpu_lock_init(vcpu);
306 		vcpu->state = VCPU_IDLE;
307 		vcpu->hostcpu = NOCPU;
308 		vcpu->guestfpu = fpu_save_area_alloc();
309 		vcpu->stats = vmm_stat_alloc();
310 		vcpu->tsc_offset = 0;
311 	}
312 
313 	vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
314 	vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
315 	vcpu->reqidle = 0;
316 	vcpu->exitintinfo = 0;
317 	vcpu->nmi_pending = 0;
318 	vcpu->extint_pending = 0;
319 	vcpu->exception_pending = 0;
320 	vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
321 	fpu_save_area_reset(vcpu->guestfpu);
322 	vmm_stat_init(vcpu->stats);
323 }
324 
325 int
326 vcpu_trace_exceptions(struct vm *vm, int vcpuid)
327 {
328 
329 	return (trace_guest_exceptions);
330 }
331 
332 struct vm_exit *
333 vm_exitinfo(struct vm *vm, int cpuid)
334 {
335 	struct vcpu *vcpu;
336 
337 	if (cpuid < 0 || cpuid >= vm->maxcpus)
338 		panic("vm_exitinfo: invalid cpuid %d", cpuid);
339 
340 	vcpu = &vm->vcpu[cpuid];
341 
342 	return (&vcpu->exitinfo);
343 }
344 
345 static void
346 vmm_resume(void)
347 {
348 	VMM_RESUME();
349 }
350 
351 static int
352 vmm_init(void)
353 {
354 	int error;
355 
356 	vmm_host_state_init();
357 
358 	vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) :
359 	    &IDTVEC(justreturn));
360 	if (vmm_ipinum < 0)
361 		vmm_ipinum = IPI_AST;
362 
363 	error = vmm_mem_init();
364 	if (error)
365 		return (error);
366 
367 	if (vmm_is_intel())
368 		ops = &vmm_ops_intel;
369 	else if (vmm_is_svm())
370 		ops = &vmm_ops_amd;
371 	else
372 		return (ENXIO);
373 
374 	vmm_resume_p = vmm_resume;
375 
376 	return (VMM_INIT(vmm_ipinum));
377 }
378 
379 static int
380 vmm_handler(module_t mod, int what, void *arg)
381 {
382 	int error;
383 
384 	switch (what) {
385 	case MOD_LOAD:
386 		vmmdev_init();
387 		error = vmm_init();
388 		if (error == 0)
389 			vmm_initialized = 1;
390 		break;
391 	case MOD_UNLOAD:
392 		error = vmmdev_cleanup();
393 		if (error == 0) {
394 			vmm_resume_p = NULL;
395 			iommu_cleanup();
396 			if (vmm_ipinum != IPI_AST)
397 				lapic_ipi_free(vmm_ipinum);
398 			error = VMM_CLEANUP();
399 			/*
400 			 * Something bad happened - prevent new
401 			 * VMs from being created
402 			 */
403 			if (error)
404 				vmm_initialized = 0;
405 		}
406 		break;
407 	default:
408 		error = 0;
409 		break;
410 	}
411 	return (error);
412 }
413 
414 static moduledata_t vmm_kmod = {
415 	"vmm",
416 	vmm_handler,
417 	NULL
418 };
419 
420 /*
421  * vmm initialization has the following dependencies:
422  *
423  * - VT-x initialization requires smp_rendezvous() and therefore must happen
424  *   after SMP is fully functional (after SI_SUB_SMP).
425  */
426 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
427 MODULE_VERSION(vmm, 1);
428 
429 static void
430 vm_init(struct vm *vm, bool create)
431 {
432 	int i;
433 
434 	vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
435 	vm->iommu = NULL;
436 	vm->vioapic = vioapic_init(vm);
437 	vm->vhpet = vhpet_init(vm);
438 	vm->vatpic = vatpic_init(vm);
439 	vm->vatpit = vatpit_init(vm);
440 	vm->vpmtmr = vpmtmr_init(vm);
441 	if (create)
442 		vm->vrtc = vrtc_init(vm);
443 
444 	CPU_ZERO(&vm->active_cpus);
445 	CPU_ZERO(&vm->debug_cpus);
446 
447 	vm->suspend = 0;
448 	CPU_ZERO(&vm->suspended_cpus);
449 
450 	for (i = 0; i < vm->maxcpus; i++)
451 		vcpu_init(vm, i, create);
452 }
453 
454 /*
455  * The default CPU topology is a single thread per package.
456  */
457 u_int cores_per_package = 1;
458 u_int threads_per_core = 1;
459 
460 int
461 vm_create(const char *name, struct vm **retvm)
462 {
463 	struct vm *vm;
464 	struct vmspace *vmspace;
465 
466 	/*
467 	 * If vmm.ko could not be successfully initialized then don't attempt
468 	 * to create the virtual machine.
469 	 */
470 	if (!vmm_initialized)
471 		return (ENXIO);
472 
473 	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
474 		return (EINVAL);
475 
476 	vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS);
477 	if (vmspace == NULL)
478 		return (ENOMEM);
479 
480 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
481 	strcpy(vm->name, name);
482 	vm->vmspace = vmspace;
483 	mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
484 
485 	vm->sockets = 1;
486 	vm->cores = cores_per_package;	/* XXX backwards compatibility */
487 	vm->threads = threads_per_core;	/* XXX backwards compatibility */
488 	vm->maxcpus = VM_MAXCPU;	/* XXX temp to keep code working */
489 
490 	vm_init(vm, true);
491 
492 	*retvm = vm;
493 	return (0);
494 }
495 
496 void
497 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
498     uint16_t *threads, uint16_t *maxcpus)
499 {
500 	*sockets = vm->sockets;
501 	*cores = vm->cores;
502 	*threads = vm->threads;
503 	*maxcpus = vm->maxcpus;
504 }
505 
506 uint16_t
507 vm_get_maxcpus(struct vm *vm)
508 {
509 	return (vm->maxcpus);
510 }
511 
512 int
513 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
514     uint16_t threads, uint16_t maxcpus)
515 {
516 	if (maxcpus != 0)
517 		return (EINVAL);	/* XXX remove when supported */
518 	if ((sockets * cores * threads) > vm->maxcpus)
519 		return (EINVAL);
520 	/* XXX need to check sockets * cores * threads == vCPU, how? */
521 	vm->sockets = sockets;
522 	vm->cores = cores;
523 	vm->threads = threads;
524 	vm->maxcpus = VM_MAXCPU;	/* XXX temp to keep code working */
525 	return(0);
526 }
527 
528 static void
529 vm_cleanup(struct vm *vm, bool destroy)
530 {
531 	struct mem_map *mm;
532 	int i;
533 
534 	ppt_unassign_all(vm);
535 
536 	if (vm->iommu != NULL)
537 		iommu_destroy_domain(vm->iommu);
538 
539 	if (destroy)
540 		vrtc_cleanup(vm->vrtc);
541 	else
542 		vrtc_reset(vm->vrtc);
543 	vpmtmr_cleanup(vm->vpmtmr);
544 	vatpit_cleanup(vm->vatpit);
545 	vhpet_cleanup(vm->vhpet);
546 	vatpic_cleanup(vm->vatpic);
547 	vioapic_cleanup(vm->vioapic);
548 
549 	for (i = 0; i < vm->maxcpus; i++)
550 		vcpu_cleanup(vm, i, destroy);
551 
552 	VMCLEANUP(vm->cookie);
553 
554 	/*
555 	 * System memory is removed from the guest address space only when
556 	 * the VM is destroyed. This is because the mapping remains the same
557 	 * across VM reset.
558 	 *
559 	 * Device memory can be relocated by the guest (e.g. using PCI BARs)
560 	 * so those mappings are removed on a VM reset.
561 	 */
562 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
563 		mm = &vm->mem_maps[i];
564 		if (destroy || !sysmem_mapping(vm, mm))
565 			vm_free_memmap(vm, i);
566 	}
567 
568 	if (destroy) {
569 		for (i = 0; i < VM_MAX_MEMSEGS; i++)
570 			vm_free_memseg(vm, i);
571 
572 		VMSPACE_FREE(vm->vmspace);
573 		vm->vmspace = NULL;
574 	}
575 }
576 
577 void
578 vm_destroy(struct vm *vm)
579 {
580 	vm_cleanup(vm, true);
581 	free(vm, M_VM);
582 }
583 
584 int
585 vm_reinit(struct vm *vm)
586 {
587 	int error;
588 
589 	/*
590 	 * A virtual machine can be reset only if all vcpus are suspended.
591 	 */
592 	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
593 		vm_cleanup(vm, false);
594 		vm_init(vm, false);
595 		error = 0;
596 	} else {
597 		error = EBUSY;
598 	}
599 
600 	return (error);
601 }
602 
603 const char *
604 vm_name(struct vm *vm)
605 {
606 	return (vm->name);
607 }
608 
609 int
610 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
611 {
612 	vm_object_t obj;
613 
614 	if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
615 		return (ENOMEM);
616 	else
617 		return (0);
618 }
619 
620 int
621 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
622 {
623 
624 	vmm_mmio_free(vm->vmspace, gpa, len);
625 	return (0);
626 }
627 
628 /*
629  * Return 'true' if 'gpa' is allocated in the guest address space.
630  *
631  * This function is called in the context of a running vcpu which acts as
632  * an implicit lock on 'vm->mem_maps[]'.
633  */
634 bool
635 vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa)
636 {
637 	struct mem_map *mm;
638 	int i;
639 
640 #ifdef INVARIANTS
641 	int hostcpu, state;
642 	state = vcpu_get_state(vm, vcpuid, &hostcpu);
643 	KASSERT(state == VCPU_RUNNING && hostcpu == curcpu,
644 	    ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu));
645 #endif
646 
647 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
648 		mm = &vm->mem_maps[i];
649 		if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len)
650 			return (true);		/* 'gpa' is sysmem or devmem */
651 	}
652 
653 	if (ppt_is_mmio(vm, gpa))
654 		return (true);			/* 'gpa' is pci passthru mmio */
655 
656 	return (false);
657 }
658 
659 int
660 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem)
661 {
662 	struct mem_seg *seg;
663 	vm_object_t obj;
664 
665 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
666 		return (EINVAL);
667 
668 	if (len == 0 || (len & PAGE_MASK))
669 		return (EINVAL);
670 
671 	seg = &vm->mem_segs[ident];
672 	if (seg->object != NULL) {
673 		if (seg->len == len && seg->sysmem == sysmem)
674 			return (EEXIST);
675 		else
676 			return (EINVAL);
677 	}
678 
679 	obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT);
680 	if (obj == NULL)
681 		return (ENOMEM);
682 
683 	seg->len = len;
684 	seg->object = obj;
685 	seg->sysmem = sysmem;
686 	return (0);
687 }
688 
689 int
690 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
691     vm_object_t *objptr)
692 {
693 	struct mem_seg *seg;
694 
695 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
696 		return (EINVAL);
697 
698 	seg = &vm->mem_segs[ident];
699 	if (len)
700 		*len = seg->len;
701 	if (sysmem)
702 		*sysmem = seg->sysmem;
703 	if (objptr)
704 		*objptr = seg->object;
705 	return (0);
706 }
707 
708 void
709 vm_free_memseg(struct vm *vm, int ident)
710 {
711 	struct mem_seg *seg;
712 
713 	KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS,
714 	    ("%s: invalid memseg ident %d", __func__, ident));
715 
716 	seg = &vm->mem_segs[ident];
717 	if (seg->object != NULL) {
718 		vm_object_deallocate(seg->object);
719 		bzero(seg, sizeof(struct mem_seg));
720 	}
721 }
722 
723 int
724 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first,
725     size_t len, int prot, int flags)
726 {
727 	struct mem_seg *seg;
728 	struct mem_map *m, *map;
729 	vm_ooffset_t last;
730 	int i, error;
731 
732 	if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0)
733 		return (EINVAL);
734 
735 	if (flags & ~VM_MEMMAP_F_WIRED)
736 		return (EINVAL);
737 
738 	if (segid < 0 || segid >= VM_MAX_MEMSEGS)
739 		return (EINVAL);
740 
741 	seg = &vm->mem_segs[segid];
742 	if (seg->object == NULL)
743 		return (EINVAL);
744 
745 	last = first + len;
746 	if (first < 0 || first >= last || last > seg->len)
747 		return (EINVAL);
748 
749 	if ((gpa | first | last) & PAGE_MASK)
750 		return (EINVAL);
751 
752 	map = NULL;
753 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
754 		m = &vm->mem_maps[i];
755 		if (m->len == 0) {
756 			map = m;
757 			break;
758 		}
759 	}
760 
761 	if (map == NULL)
762 		return (ENOSPC);
763 
764 	error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa,
765 	    len, 0, VMFS_NO_SPACE, prot, prot, 0);
766 	if (error != KERN_SUCCESS)
767 		return (EFAULT);
768 
769 	vm_object_reference(seg->object);
770 
771 	if (flags & VM_MEMMAP_F_WIRED) {
772 		error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len,
773 		    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
774 		if (error != KERN_SUCCESS) {
775 			vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len);
776 			return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM :
777 			    EFAULT);
778 		}
779 	}
780 
781 	map->gpa = gpa;
782 	map->len = len;
783 	map->segoff = first;
784 	map->segid = segid;
785 	map->prot = prot;
786 	map->flags = flags;
787 	return (0);
788 }
789 
790 int
791 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
792     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags)
793 {
794 	struct mem_map *mm, *mmnext;
795 	int i;
796 
797 	mmnext = NULL;
798 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
799 		mm = &vm->mem_maps[i];
800 		if (mm->len == 0 || mm->gpa < *gpa)
801 			continue;
802 		if (mmnext == NULL || mm->gpa < mmnext->gpa)
803 			mmnext = mm;
804 	}
805 
806 	if (mmnext != NULL) {
807 		*gpa = mmnext->gpa;
808 		if (segid)
809 			*segid = mmnext->segid;
810 		if (segoff)
811 			*segoff = mmnext->segoff;
812 		if (len)
813 			*len = mmnext->len;
814 		if (prot)
815 			*prot = mmnext->prot;
816 		if (flags)
817 			*flags = mmnext->flags;
818 		return (0);
819 	} else {
820 		return (ENOENT);
821 	}
822 }
823 
824 static void
825 vm_free_memmap(struct vm *vm, int ident)
826 {
827 	struct mem_map *mm;
828 	int error;
829 
830 	mm = &vm->mem_maps[ident];
831 	if (mm->len) {
832 		error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa,
833 		    mm->gpa + mm->len);
834 		KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d",
835 		    __func__, error));
836 		bzero(mm, sizeof(struct mem_map));
837 	}
838 }
839 
840 static __inline bool
841 sysmem_mapping(struct vm *vm, struct mem_map *mm)
842 {
843 
844 	if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem)
845 		return (true);
846 	else
847 		return (false);
848 }
849 
850 vm_paddr_t
851 vmm_sysmem_maxaddr(struct vm *vm)
852 {
853 	struct mem_map *mm;
854 	vm_paddr_t maxaddr;
855 	int i;
856 
857 	maxaddr = 0;
858 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
859 		mm = &vm->mem_maps[i];
860 		if (sysmem_mapping(vm, mm)) {
861 			if (maxaddr < mm->gpa + mm->len)
862 				maxaddr = mm->gpa + mm->len;
863 		}
864 	}
865 	return (maxaddr);
866 }
867 
868 static void
869 vm_iommu_modify(struct vm *vm, bool map)
870 {
871 	int i, sz;
872 	vm_paddr_t gpa, hpa;
873 	struct mem_map *mm;
874 	void *vp, *cookie, *host_domain;
875 
876 	sz = PAGE_SIZE;
877 	host_domain = iommu_host_domain();
878 
879 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
880 		mm = &vm->mem_maps[i];
881 		if (!sysmem_mapping(vm, mm))
882 			continue;
883 
884 		if (map) {
885 			KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0,
886 			    ("iommu map found invalid memmap %#lx/%#lx/%#x",
887 			    mm->gpa, mm->len, mm->flags));
888 			if ((mm->flags & VM_MEMMAP_F_WIRED) == 0)
889 				continue;
890 			mm->flags |= VM_MEMMAP_F_IOMMU;
891 		} else {
892 			if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0)
893 				continue;
894 			mm->flags &= ~VM_MEMMAP_F_IOMMU;
895 			KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0,
896 			    ("iommu unmap found invalid memmap %#lx/%#lx/%#x",
897 			    mm->gpa, mm->len, mm->flags));
898 		}
899 
900 		gpa = mm->gpa;
901 		while (gpa < mm->gpa + mm->len) {
902 			vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE,
903 					 &cookie);
904 			KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
905 			    vm_name(vm), gpa));
906 
907 			vm_gpa_release(cookie);
908 
909 			hpa = DMAP_TO_PHYS((uintptr_t)vp);
910 			if (map) {
911 				iommu_create_mapping(vm->iommu, gpa, hpa, sz);
912 				iommu_remove_mapping(host_domain, hpa, sz);
913 			} else {
914 				iommu_remove_mapping(vm->iommu, gpa, sz);
915 				iommu_create_mapping(host_domain, hpa, hpa, sz);
916 			}
917 
918 			gpa += PAGE_SIZE;
919 		}
920 	}
921 
922 	/*
923 	 * Invalidate the cached translations associated with the domain
924 	 * from which pages were removed.
925 	 */
926 	if (map)
927 		iommu_invalidate_tlb(host_domain);
928 	else
929 		iommu_invalidate_tlb(vm->iommu);
930 }
931 
932 #define	vm_iommu_unmap(vm)	vm_iommu_modify((vm), false)
933 #define	vm_iommu_map(vm)	vm_iommu_modify((vm), true)
934 
935 int
936 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
937 {
938 	int error;
939 
940 	error = ppt_unassign_device(vm, bus, slot, func);
941 	if (error)
942 		return (error);
943 
944 	if (ppt_assigned_devices(vm) == 0)
945 		vm_iommu_unmap(vm);
946 
947 	return (0);
948 }
949 
950 int
951 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
952 {
953 	int error;
954 	vm_paddr_t maxaddr;
955 
956 	/* Set up the IOMMU to do the 'gpa' to 'hpa' translation */
957 	if (ppt_assigned_devices(vm) == 0) {
958 		KASSERT(vm->iommu == NULL,
959 		    ("vm_assign_pptdev: iommu must be NULL"));
960 		maxaddr = vmm_sysmem_maxaddr(vm);
961 		vm->iommu = iommu_create_domain(maxaddr);
962 		if (vm->iommu == NULL)
963 			return (ENXIO);
964 		vm_iommu_map(vm);
965 	}
966 
967 	error = ppt_assign_device(vm, bus, slot, func);
968 	return (error);
969 }
970 
971 void *
972 vm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot,
973 	    void **cookie)
974 {
975 	int i, count, pageoff;
976 	struct mem_map *mm;
977 	vm_page_t m;
978 #ifdef INVARIANTS
979 	/*
980 	 * All vcpus are frozen by ioctls that modify the memory map
981 	 * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is
982 	 * guaranteed if at least one vcpu is in the VCPU_FROZEN state.
983 	 */
984 	int state;
985 	KASSERT(vcpuid >= -1 && vcpuid < vm->maxcpus, ("%s: invalid vcpuid %d",
986 	    __func__, vcpuid));
987 	for (i = 0; i < vm->maxcpus; i++) {
988 		if (vcpuid != -1 && vcpuid != i)
989 			continue;
990 		state = vcpu_get_state(vm, i, NULL);
991 		KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d",
992 		    __func__, state));
993 	}
994 #endif
995 	pageoff = gpa & PAGE_MASK;
996 	if (len > PAGE_SIZE - pageoff)
997 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
998 
999 	count = 0;
1000 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
1001 		mm = &vm->mem_maps[i];
1002 		if (sysmem_mapping(vm, mm) && gpa >= mm->gpa &&
1003 		    gpa < mm->gpa + mm->len) {
1004 			count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
1005 			    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
1006 			break;
1007 		}
1008 	}
1009 
1010 	if (count == 1) {
1011 		*cookie = m;
1012 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
1013 	} else {
1014 		*cookie = NULL;
1015 		return (NULL);
1016 	}
1017 }
1018 
1019 void
1020 vm_gpa_release(void *cookie)
1021 {
1022 	vm_page_t m = cookie;
1023 
1024 	vm_page_unwire(m, PQ_ACTIVE);
1025 }
1026 
1027 int
1028 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
1029 {
1030 
1031 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1032 		return (EINVAL);
1033 
1034 	if (reg >= VM_REG_LAST)
1035 		return (EINVAL);
1036 
1037 	return (VMGETREG(vm->cookie, vcpu, reg, retval));
1038 }
1039 
1040 int
1041 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val)
1042 {
1043 	struct vcpu *vcpu;
1044 	int error;
1045 
1046 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1047 		return (EINVAL);
1048 
1049 	if (reg >= VM_REG_LAST)
1050 		return (EINVAL);
1051 
1052 	error = VMSETREG(vm->cookie, vcpuid, reg, val);
1053 	if (error || reg != VM_REG_GUEST_RIP)
1054 		return (error);
1055 
1056 	/* Set 'nextrip' to match the value of %rip */
1057 	VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val);
1058 	vcpu = &vm->vcpu[vcpuid];
1059 	vcpu->nextrip = val;
1060 	return (0);
1061 }
1062 
1063 static bool
1064 is_descriptor_table(int reg)
1065 {
1066 
1067 	switch (reg) {
1068 	case VM_REG_GUEST_IDTR:
1069 	case VM_REG_GUEST_GDTR:
1070 		return (true);
1071 	default:
1072 		return (false);
1073 	}
1074 }
1075 
1076 static bool
1077 is_segment_register(int reg)
1078 {
1079 
1080 	switch (reg) {
1081 	case VM_REG_GUEST_ES:
1082 	case VM_REG_GUEST_CS:
1083 	case VM_REG_GUEST_SS:
1084 	case VM_REG_GUEST_DS:
1085 	case VM_REG_GUEST_FS:
1086 	case VM_REG_GUEST_GS:
1087 	case VM_REG_GUEST_TR:
1088 	case VM_REG_GUEST_LDTR:
1089 		return (true);
1090 	default:
1091 		return (false);
1092 	}
1093 }
1094 
1095 int
1096 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
1097 		struct seg_desc *desc)
1098 {
1099 
1100 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1101 		return (EINVAL);
1102 
1103 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1104 		return (EINVAL);
1105 
1106 	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
1107 }
1108 
1109 int
1110 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
1111 		struct seg_desc *desc)
1112 {
1113 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1114 		return (EINVAL);
1115 
1116 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1117 		return (EINVAL);
1118 
1119 	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
1120 }
1121 
1122 static void
1123 restore_guest_fpustate(struct vcpu *vcpu)
1124 {
1125 
1126 	/* flush host state to the pcb */
1127 	fpuexit(curthread);
1128 
1129 	/* restore guest FPU state */
1130 	fpu_stop_emulating();
1131 	fpurestore(vcpu->guestfpu);
1132 
1133 	/* restore guest XCR0 if XSAVE is enabled in the host */
1134 	if (rcr4() & CR4_XSAVE)
1135 		load_xcr(0, vcpu->guest_xcr0);
1136 
1137 	/*
1138 	 * The FPU is now "dirty" with the guest's state so turn on emulation
1139 	 * to trap any access to the FPU by the host.
1140 	 */
1141 	fpu_start_emulating();
1142 }
1143 
1144 static void
1145 save_guest_fpustate(struct vcpu *vcpu)
1146 {
1147 
1148 	if ((rcr0() & CR0_TS) == 0)
1149 		panic("fpu emulation not enabled in host!");
1150 
1151 	/* save guest XCR0 and restore host XCR0 */
1152 	if (rcr4() & CR4_XSAVE) {
1153 		vcpu->guest_xcr0 = rxcr(0);
1154 		load_xcr(0, vmm_get_host_xcr0());
1155 	}
1156 
1157 	/* save guest FPU state */
1158 	fpu_stop_emulating();
1159 	fpusave(vcpu->guestfpu);
1160 	fpu_start_emulating();
1161 }
1162 
1163 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
1164 
1165 static int
1166 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1167     bool from_idle)
1168 {
1169 	struct vcpu *vcpu;
1170 	int error;
1171 
1172 	vcpu = &vm->vcpu[vcpuid];
1173 	vcpu_assert_locked(vcpu);
1174 
1175 	/*
1176 	 * State transitions from the vmmdev_ioctl() must always begin from
1177 	 * the VCPU_IDLE state. This guarantees that there is only a single
1178 	 * ioctl() operating on a vcpu at any point.
1179 	 */
1180 	if (from_idle) {
1181 		while (vcpu->state != VCPU_IDLE) {
1182 			vcpu->reqidle = 1;
1183 			vcpu_notify_event_locked(vcpu, false);
1184 			VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to "
1185 			    "idle requested", vcpu_state2str(vcpu->state));
1186 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
1187 		}
1188 	} else {
1189 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1190 		    "vcpu idle state"));
1191 	}
1192 
1193 	if (vcpu->state == VCPU_RUNNING) {
1194 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1195 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1196 	} else {
1197 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1198 		    "vcpu that is not running", vcpu->hostcpu));
1199 	}
1200 
1201 	/*
1202 	 * The following state transitions are allowed:
1203 	 * IDLE -> FROZEN -> IDLE
1204 	 * FROZEN -> RUNNING -> FROZEN
1205 	 * FROZEN -> SLEEPING -> FROZEN
1206 	 */
1207 	switch (vcpu->state) {
1208 	case VCPU_IDLE:
1209 	case VCPU_RUNNING:
1210 	case VCPU_SLEEPING:
1211 		error = (newstate != VCPU_FROZEN);
1212 		break;
1213 	case VCPU_FROZEN:
1214 		error = (newstate == VCPU_FROZEN);
1215 		break;
1216 	default:
1217 		error = 1;
1218 		break;
1219 	}
1220 
1221 	if (error)
1222 		return (EBUSY);
1223 
1224 	VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s",
1225 	    vcpu_state2str(vcpu->state), vcpu_state2str(newstate));
1226 
1227 	vcpu->state = newstate;
1228 	if (newstate == VCPU_RUNNING)
1229 		vcpu->hostcpu = curcpu;
1230 	else
1231 		vcpu->hostcpu = NOCPU;
1232 
1233 	if (newstate == VCPU_IDLE)
1234 		wakeup(&vcpu->state);
1235 
1236 	return (0);
1237 }
1238 
1239 static void
1240 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1241 {
1242 	int error;
1243 
1244 	if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1245 		panic("Error %d setting state to %d\n", error, newstate);
1246 }
1247 
1248 static void
1249 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1250 {
1251 	int error;
1252 
1253 	if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0)
1254 		panic("Error %d setting state to %d", error, newstate);
1255 }
1256 
1257 #define	RENDEZVOUS_CTR0(vm, vcpuid, fmt)				\
1258 	do {								\
1259 		if (vcpuid >= 0)					\
1260 			VCPU_CTR0(vm, vcpuid, fmt);			\
1261 		else							\
1262 			VM_CTR0(vm, fmt);				\
1263 	} while (0)
1264 
1265 static int
1266 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1267 {
1268 	struct thread *td;
1269 	int error;
1270 
1271 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
1272 	    ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1273 
1274 	error = 0;
1275 	td = curthread;
1276 	mtx_lock(&vm->rendezvous_mtx);
1277 	while (vm->rendezvous_func != NULL) {
1278 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1279 		CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1280 
1281 		if (vcpuid != -1 &&
1282 		    CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1283 		    !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1284 			VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1285 			(*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1286 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1287 		}
1288 		if (CPU_CMP(&vm->rendezvous_req_cpus,
1289 		    &vm->rendezvous_done_cpus) == 0) {
1290 			VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1291 			vm->rendezvous_func = NULL;
1292 			wakeup(&vm->rendezvous_func);
1293 			break;
1294 		}
1295 		RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1296 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1297 		    "vmrndv", hz);
1298 		if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1299 			mtx_unlock(&vm->rendezvous_mtx);
1300 			error = thread_check_susp(td, true);
1301 			if (error != 0)
1302 				return (error);
1303 			mtx_lock(&vm->rendezvous_mtx);
1304 		}
1305 	}
1306 	mtx_unlock(&vm->rendezvous_mtx);
1307 	return (0);
1308 }
1309 
1310 /*
1311  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1312  */
1313 static int
1314 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1315 {
1316 	struct vcpu *vcpu;
1317 	const char *wmesg;
1318 	struct thread *td;
1319 	int error, t, vcpu_halted, vm_halted;
1320 
1321 	KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1322 
1323 	vcpu = &vm->vcpu[vcpuid];
1324 	vcpu_halted = 0;
1325 	vm_halted = 0;
1326 	error = 0;
1327 	td = curthread;
1328 
1329 	vcpu_lock(vcpu);
1330 	while (1) {
1331 		/*
1332 		 * Do a final check for pending NMI or interrupts before
1333 		 * really putting this thread to sleep. Also check for
1334 		 * software events that would cause this vcpu to wakeup.
1335 		 *
1336 		 * These interrupts/events could have happened after the
1337 		 * vcpu returned from VMRUN() and before it acquired the
1338 		 * vcpu lock above.
1339 		 */
1340 		if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle)
1341 			break;
1342 		if (vm_nmi_pending(vm, vcpuid))
1343 			break;
1344 		if (!intr_disabled) {
1345 			if (vm_extint_pending(vm, vcpuid) ||
1346 			    vlapic_pending_intr(vcpu->vlapic, NULL)) {
1347 				break;
1348 			}
1349 		}
1350 
1351 		/* Don't go to sleep if the vcpu thread needs to yield */
1352 		if (vcpu_should_yield(vm, vcpuid))
1353 			break;
1354 
1355 		if (vcpu_debugged(vm, vcpuid))
1356 			break;
1357 
1358 		/*
1359 		 * Some Linux guests implement "halt" by having all vcpus
1360 		 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1361 		 * track of the vcpus that have entered this state. When all
1362 		 * vcpus enter the halted state the virtual machine is halted.
1363 		 */
1364 		if (intr_disabled) {
1365 			wmesg = "vmhalt";
1366 			VCPU_CTR0(vm, vcpuid, "Halted");
1367 			if (!vcpu_halted && halt_detection_enabled) {
1368 				vcpu_halted = 1;
1369 				CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1370 			}
1371 			if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1372 				vm_halted = 1;
1373 				break;
1374 			}
1375 		} else {
1376 			wmesg = "vmidle";
1377 		}
1378 
1379 		t = ticks;
1380 		vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1381 		/*
1382 		 * XXX msleep_spin() cannot be interrupted by signals so
1383 		 * wake up periodically to check pending signals.
1384 		 */
1385 		msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1386 		vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1387 		vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1388 		if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1389 			vcpu_unlock(vcpu);
1390 			error = thread_check_susp(td, false);
1391 			if (error != 0)
1392 				return (error);
1393 			vcpu_lock(vcpu);
1394 		}
1395 	}
1396 
1397 	if (vcpu_halted)
1398 		CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1399 
1400 	vcpu_unlock(vcpu);
1401 
1402 	if (vm_halted)
1403 		vm_suspend(vm, VM_SUSPEND_HALT);
1404 
1405 	return (0);
1406 }
1407 
1408 static int
1409 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1410 {
1411 	int rv, ftype;
1412 	struct vm_map *map;
1413 	struct vcpu *vcpu;
1414 	struct vm_exit *vme;
1415 
1416 	vcpu = &vm->vcpu[vcpuid];
1417 	vme = &vcpu->exitinfo;
1418 
1419 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1420 	    __func__, vme->inst_length));
1421 
1422 	ftype = vme->u.paging.fault_type;
1423 	KASSERT(ftype == VM_PROT_READ ||
1424 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1425 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1426 
1427 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1428 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1429 		    vme->u.paging.gpa, ftype);
1430 		if (rv == 0) {
1431 			VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1432 			    ftype == VM_PROT_READ ? "accessed" : "dirty",
1433 			    vme->u.paging.gpa);
1434 			goto done;
1435 		}
1436 	}
1437 
1438 	map = &vm->vmspace->vm_map;
1439 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL);
1440 
1441 	VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1442 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1443 
1444 	if (rv != KERN_SUCCESS)
1445 		return (EFAULT);
1446 done:
1447 	return (0);
1448 }
1449 
1450 static int
1451 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1452 {
1453 	struct vie *vie;
1454 	struct vcpu *vcpu;
1455 	struct vm_exit *vme;
1456 	uint64_t gla, gpa, cs_base;
1457 	struct vm_guest_paging *paging;
1458 	mem_region_read_t mread;
1459 	mem_region_write_t mwrite;
1460 	enum vm_cpu_mode cpu_mode;
1461 	int cs_d, error, fault;
1462 
1463 	vcpu = &vm->vcpu[vcpuid];
1464 	vme = &vcpu->exitinfo;
1465 
1466 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1467 	    __func__, vme->inst_length));
1468 
1469 	gla = vme->u.inst_emul.gla;
1470 	gpa = vme->u.inst_emul.gpa;
1471 	cs_base = vme->u.inst_emul.cs_base;
1472 	cs_d = vme->u.inst_emul.cs_d;
1473 	vie = &vme->u.inst_emul.vie;
1474 	paging = &vme->u.inst_emul.paging;
1475 	cpu_mode = paging->cpu_mode;
1476 
1477 	VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1478 
1479 	/* Fetch, decode and emulate the faulting instruction */
1480 	if (vie->num_valid == 0) {
1481 		error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip +
1482 		    cs_base, VIE_INST_SIZE, vie, &fault);
1483 	} else {
1484 		/*
1485 		 * The instruction bytes have already been copied into 'vie'
1486 		 */
1487 		error = fault = 0;
1488 	}
1489 	if (error || fault)
1490 		return (error);
1491 
1492 	if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) {
1493 		VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx",
1494 		    vme->rip + cs_base);
1495 		*retu = true;	    /* dump instruction bytes in userspace */
1496 		return (0);
1497 	}
1498 
1499 	/*
1500 	 * Update 'nextrip' based on the length of the emulated instruction.
1501 	 */
1502 	vme->inst_length = vie->num_processed;
1503 	vcpu->nextrip += vie->num_processed;
1504 	VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction "
1505 	    "decoding", vcpu->nextrip);
1506 
1507 	/* return to userland unless this is an in-kernel emulated device */
1508 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1509 		mread = lapic_mmio_read;
1510 		mwrite = lapic_mmio_write;
1511 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1512 		mread = vioapic_mmio_read;
1513 		mwrite = vioapic_mmio_write;
1514 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1515 		mread = vhpet_mmio_read;
1516 		mwrite = vhpet_mmio_write;
1517 	} else {
1518 		*retu = true;
1519 		return (0);
1520 	}
1521 
1522 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1523 	    mread, mwrite, retu);
1524 
1525 	return (error);
1526 }
1527 
1528 static int
1529 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1530 {
1531 	int error, i;
1532 	struct vcpu *vcpu;
1533 	struct thread *td;
1534 
1535 	error = 0;
1536 	vcpu = &vm->vcpu[vcpuid];
1537 	td = curthread;
1538 
1539 	CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1540 
1541 	/*
1542 	 * Wait until all 'active_cpus' have suspended themselves.
1543 	 *
1544 	 * Since a VM may be suspended at any time including when one or
1545 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1546 	 * handler while we are waiting to prevent a deadlock.
1547 	 */
1548 	vcpu_lock(vcpu);
1549 	while (error == 0) {
1550 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1551 			VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1552 			break;
1553 		}
1554 
1555 		if (vm->rendezvous_func == NULL) {
1556 			VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1557 			vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1558 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1559 			vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1560 			if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1561 				vcpu_unlock(vcpu);
1562 				error = thread_check_susp(td, false);
1563 				vcpu_lock(vcpu);
1564 			}
1565 		} else {
1566 			VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1567 			vcpu_unlock(vcpu);
1568 			error = vm_handle_rendezvous(vm, vcpuid);
1569 			vcpu_lock(vcpu);
1570 		}
1571 	}
1572 	vcpu_unlock(vcpu);
1573 
1574 	/*
1575 	 * Wakeup the other sleeping vcpus and return to userspace.
1576 	 */
1577 	for (i = 0; i < vm->maxcpus; i++) {
1578 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1579 			vcpu_notify_event(vm, i, false);
1580 		}
1581 	}
1582 
1583 	*retu = true;
1584 	return (error);
1585 }
1586 
1587 static int
1588 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu)
1589 {
1590 	struct vcpu *vcpu = &vm->vcpu[vcpuid];
1591 
1592 	vcpu_lock(vcpu);
1593 	KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle));
1594 	vcpu->reqidle = 0;
1595 	vcpu_unlock(vcpu);
1596 	*retu = true;
1597 	return (0);
1598 }
1599 
1600 int
1601 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1602 {
1603 	int i;
1604 
1605 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1606 		return (EINVAL);
1607 
1608 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1609 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1610 		    vm->suspend, how);
1611 		return (EALREADY);
1612 	}
1613 
1614 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1615 
1616 	/*
1617 	 * Notify all active vcpus that they are now suspended.
1618 	 */
1619 	for (i = 0; i < vm->maxcpus; i++) {
1620 		if (CPU_ISSET(i, &vm->active_cpus))
1621 			vcpu_notify_event(vm, i, false);
1622 	}
1623 
1624 	return (0);
1625 }
1626 
1627 void
1628 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1629 {
1630 	struct vm_exit *vmexit;
1631 
1632 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1633 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1634 
1635 	vmexit = vm_exitinfo(vm, vcpuid);
1636 	vmexit->rip = rip;
1637 	vmexit->inst_length = 0;
1638 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1639 	vmexit->u.suspended.how = vm->suspend;
1640 }
1641 
1642 void
1643 vm_exit_debug(struct vm *vm, int vcpuid, uint64_t rip)
1644 {
1645 	struct vm_exit *vmexit;
1646 
1647 	vmexit = vm_exitinfo(vm, vcpuid);
1648 	vmexit->rip = rip;
1649 	vmexit->inst_length = 0;
1650 	vmexit->exitcode = VM_EXITCODE_DEBUG;
1651 }
1652 
1653 void
1654 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1655 {
1656 	struct vm_exit *vmexit;
1657 
1658 	KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1659 
1660 	vmexit = vm_exitinfo(vm, vcpuid);
1661 	vmexit->rip = rip;
1662 	vmexit->inst_length = 0;
1663 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1664 	vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1665 }
1666 
1667 void
1668 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip)
1669 {
1670 	struct vm_exit *vmexit;
1671 
1672 	vmexit = vm_exitinfo(vm, vcpuid);
1673 	vmexit->rip = rip;
1674 	vmexit->inst_length = 0;
1675 	vmexit->exitcode = VM_EXITCODE_REQIDLE;
1676 	vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1);
1677 }
1678 
1679 void
1680 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1681 {
1682 	struct vm_exit *vmexit;
1683 
1684 	vmexit = vm_exitinfo(vm, vcpuid);
1685 	vmexit->rip = rip;
1686 	vmexit->inst_length = 0;
1687 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1688 	vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1689 }
1690 
1691 int
1692 vm_run(struct vm *vm, struct vm_run *vmrun)
1693 {
1694 	struct vm_eventinfo evinfo;
1695 	int error, vcpuid;
1696 	struct vcpu *vcpu;
1697 	struct pcb *pcb;
1698 	uint64_t tscval;
1699 	struct vm_exit *vme;
1700 	bool retu, intr_disabled;
1701 	pmap_t pmap;
1702 
1703 	vcpuid = vmrun->cpuid;
1704 
1705 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1706 		return (EINVAL);
1707 
1708 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1709 		return (EINVAL);
1710 
1711 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1712 		return (EINVAL);
1713 
1714 	pmap = vmspace_pmap(vm->vmspace);
1715 	vcpu = &vm->vcpu[vcpuid];
1716 	vme = &vcpu->exitinfo;
1717 	evinfo.rptr = &vm->rendezvous_func;
1718 	evinfo.sptr = &vm->suspend;
1719 	evinfo.iptr = &vcpu->reqidle;
1720 restart:
1721 	critical_enter();
1722 
1723 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1724 	    ("vm_run: absurd pm_active"));
1725 
1726 	tscval = rdtsc();
1727 
1728 	pcb = PCPU_GET(curpcb);
1729 	set_pcb_flags(pcb, PCB_FULL_IRET);
1730 
1731 	restore_guest_fpustate(vcpu);
1732 
1733 	vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1734 	error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo);
1735 	vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1736 
1737 	save_guest_fpustate(vcpu);
1738 
1739 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1740 
1741 	critical_exit();
1742 
1743 	if (error == 0) {
1744 		retu = false;
1745 		vcpu->nextrip = vme->rip + vme->inst_length;
1746 		switch (vme->exitcode) {
1747 		case VM_EXITCODE_REQIDLE:
1748 			error = vm_handle_reqidle(vm, vcpuid, &retu);
1749 			break;
1750 		case VM_EXITCODE_SUSPENDED:
1751 			error = vm_handle_suspend(vm, vcpuid, &retu);
1752 			break;
1753 		case VM_EXITCODE_IOAPIC_EOI:
1754 			vioapic_process_eoi(vm, vcpuid,
1755 			    vme->u.ioapic_eoi.vector);
1756 			break;
1757 		case VM_EXITCODE_RENDEZVOUS:
1758 			error = vm_handle_rendezvous(vm, vcpuid);
1759 			break;
1760 		case VM_EXITCODE_HLT:
1761 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1762 			error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1763 			break;
1764 		case VM_EXITCODE_PAGING:
1765 			error = vm_handle_paging(vm, vcpuid, &retu);
1766 			break;
1767 		case VM_EXITCODE_INST_EMUL:
1768 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1769 			break;
1770 		case VM_EXITCODE_INOUT:
1771 		case VM_EXITCODE_INOUT_STR:
1772 			error = vm_handle_inout(vm, vcpuid, vme, &retu);
1773 			break;
1774 		case VM_EXITCODE_MONITOR:
1775 		case VM_EXITCODE_MWAIT:
1776 		case VM_EXITCODE_VMINSN:
1777 			vm_inject_ud(vm, vcpuid);
1778 			break;
1779 		default:
1780 			retu = true;	/* handled in userland */
1781 			break;
1782 		}
1783 	}
1784 
1785 	if (error == 0 && retu == false)
1786 		goto restart;
1787 
1788 	VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode);
1789 
1790 	/* copy the exit information */
1791 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1792 	return (error);
1793 }
1794 
1795 int
1796 vm_restart_instruction(void *arg, int vcpuid)
1797 {
1798 	struct vm *vm;
1799 	struct vcpu *vcpu;
1800 	enum vcpu_state state;
1801 	uint64_t rip;
1802 	int error;
1803 
1804 	vm = arg;
1805 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1806 		return (EINVAL);
1807 
1808 	vcpu = &vm->vcpu[vcpuid];
1809 	state = vcpu_get_state(vm, vcpuid, NULL);
1810 	if (state == VCPU_RUNNING) {
1811 		/*
1812 		 * When a vcpu is "running" the next instruction is determined
1813 		 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
1814 		 * Thus setting 'inst_length' to zero will cause the current
1815 		 * instruction to be restarted.
1816 		 */
1817 		vcpu->exitinfo.inst_length = 0;
1818 		VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by "
1819 		    "setting inst_length to zero", vcpu->exitinfo.rip);
1820 	} else if (state == VCPU_FROZEN) {
1821 		/*
1822 		 * When a vcpu is "frozen" it is outside the critical section
1823 		 * around VMRUN() and 'nextrip' points to the next instruction.
1824 		 * Thus instruction restart is achieved by setting 'nextrip'
1825 		 * to the vcpu's %rip.
1826 		 */
1827 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip);
1828 		KASSERT(!error, ("%s: error %d getting rip", __func__, error));
1829 		VCPU_CTR2(vm, vcpuid, "restarting instruction by updating "
1830 		    "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
1831 		vcpu->nextrip = rip;
1832 	} else {
1833 		panic("%s: invalid state %d", __func__, state);
1834 	}
1835 	return (0);
1836 }
1837 
1838 int
1839 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1840 {
1841 	struct vcpu *vcpu;
1842 	int type, vector;
1843 
1844 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1845 		return (EINVAL);
1846 
1847 	vcpu = &vm->vcpu[vcpuid];
1848 
1849 	if (info & VM_INTINFO_VALID) {
1850 		type = info & VM_INTINFO_TYPE;
1851 		vector = info & 0xff;
1852 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1853 			return (EINVAL);
1854 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1855 			return (EINVAL);
1856 		if (info & VM_INTINFO_RSVD)
1857 			return (EINVAL);
1858 	} else {
1859 		info = 0;
1860 	}
1861 	VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1862 	vcpu->exitintinfo = info;
1863 	return (0);
1864 }
1865 
1866 enum exc_class {
1867 	EXC_BENIGN,
1868 	EXC_CONTRIBUTORY,
1869 	EXC_PAGEFAULT
1870 };
1871 
1872 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
1873 
1874 static enum exc_class
1875 exception_class(uint64_t info)
1876 {
1877 	int type, vector;
1878 
1879 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1880 	type = info & VM_INTINFO_TYPE;
1881 	vector = info & 0xff;
1882 
1883 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1884 	switch (type) {
1885 	case VM_INTINFO_HWINTR:
1886 	case VM_INTINFO_SWINTR:
1887 	case VM_INTINFO_NMI:
1888 		return (EXC_BENIGN);
1889 	default:
1890 		/*
1891 		 * Hardware exception.
1892 		 *
1893 		 * SVM and VT-x use identical type values to represent NMI,
1894 		 * hardware interrupt and software interrupt.
1895 		 *
1896 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1897 		 * for exceptions except #BP and #OF. #BP and #OF use a type
1898 		 * value of '5' or '6'. Therefore we don't check for explicit
1899 		 * values of 'type' to classify 'intinfo' into a hardware
1900 		 * exception.
1901 		 */
1902 		break;
1903 	}
1904 
1905 	switch (vector) {
1906 	case IDT_PF:
1907 	case IDT_VE:
1908 		return (EXC_PAGEFAULT);
1909 	case IDT_DE:
1910 	case IDT_TS:
1911 	case IDT_NP:
1912 	case IDT_SS:
1913 	case IDT_GP:
1914 		return (EXC_CONTRIBUTORY);
1915 	default:
1916 		return (EXC_BENIGN);
1917 	}
1918 }
1919 
1920 static int
1921 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1922     uint64_t *retinfo)
1923 {
1924 	enum exc_class exc1, exc2;
1925 	int type1, vector1;
1926 
1927 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1928 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1929 
1930 	/*
1931 	 * If an exception occurs while attempting to call the double-fault
1932 	 * handler the processor enters shutdown mode (aka triple fault).
1933 	 */
1934 	type1 = info1 & VM_INTINFO_TYPE;
1935 	vector1 = info1 & 0xff;
1936 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1937 		VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1938 		    info1, info2);
1939 		vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1940 		*retinfo = 0;
1941 		return (0);
1942 	}
1943 
1944 	/*
1945 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1946 	 */
1947 	exc1 = exception_class(info1);
1948 	exc2 = exception_class(info2);
1949 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1950 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1951 		/* Convert nested fault into a double fault. */
1952 		*retinfo = IDT_DF;
1953 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1954 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
1955 	} else {
1956 		/* Handle exceptions serially */
1957 		*retinfo = info2;
1958 	}
1959 	return (1);
1960 }
1961 
1962 static uint64_t
1963 vcpu_exception_intinfo(struct vcpu *vcpu)
1964 {
1965 	uint64_t info = 0;
1966 
1967 	if (vcpu->exception_pending) {
1968 		info = vcpu->exc_vector & 0xff;
1969 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1970 		if (vcpu->exc_errcode_valid) {
1971 			info |= VM_INTINFO_DEL_ERRCODE;
1972 			info |= (uint64_t)vcpu->exc_errcode << 32;
1973 		}
1974 	}
1975 	return (info);
1976 }
1977 
1978 int
1979 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1980 {
1981 	struct vcpu *vcpu;
1982 	uint64_t info1, info2;
1983 	int valid;
1984 
1985 	KASSERT(vcpuid >= 0 &&
1986 	    vcpuid < vm->maxcpus, ("invalid vcpu %d", vcpuid));
1987 
1988 	vcpu = &vm->vcpu[vcpuid];
1989 
1990 	info1 = vcpu->exitintinfo;
1991 	vcpu->exitintinfo = 0;
1992 
1993 	info2 = 0;
1994 	if (vcpu->exception_pending) {
1995 		info2 = vcpu_exception_intinfo(vcpu);
1996 		vcpu->exception_pending = 0;
1997 		VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1998 		    vcpu->exc_vector, info2);
1999 	}
2000 
2001 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
2002 		valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
2003 	} else if (info1 & VM_INTINFO_VALID) {
2004 		*retinfo = info1;
2005 		valid = 1;
2006 	} else if (info2 & VM_INTINFO_VALID) {
2007 		*retinfo = info2;
2008 		valid = 1;
2009 	} else {
2010 		valid = 0;
2011 	}
2012 
2013 	if (valid) {
2014 		VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
2015 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
2016 	}
2017 
2018 	return (valid);
2019 }
2020 
2021 int
2022 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
2023 {
2024 	struct vcpu *vcpu;
2025 
2026 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2027 		return (EINVAL);
2028 
2029 	vcpu = &vm->vcpu[vcpuid];
2030 	*info1 = vcpu->exitintinfo;
2031 	*info2 = vcpu_exception_intinfo(vcpu);
2032 	return (0);
2033 }
2034 
2035 int
2036 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid,
2037     uint32_t errcode, int restart_instruction)
2038 {
2039 	struct vcpu *vcpu;
2040 	uint64_t regval;
2041 	int error;
2042 
2043 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2044 		return (EINVAL);
2045 
2046 	if (vector < 0 || vector >= 32)
2047 		return (EINVAL);
2048 
2049 	/*
2050 	 * A double fault exception should never be injected directly into
2051 	 * the guest. It is a derived exception that results from specific
2052 	 * combinations of nested faults.
2053 	 */
2054 	if (vector == IDT_DF)
2055 		return (EINVAL);
2056 
2057 	vcpu = &vm->vcpu[vcpuid];
2058 
2059 	if (vcpu->exception_pending) {
2060 		VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
2061 		    "pending exception %d", vector, vcpu->exc_vector);
2062 		return (EBUSY);
2063 	}
2064 
2065 	if (errcode_valid) {
2066 		/*
2067 		 * Exceptions don't deliver an error code in real mode.
2068 		 */
2069 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, &regval);
2070 		KASSERT(!error, ("%s: error %d getting CR0", __func__, error));
2071 		if (!(regval & CR0_PE))
2072 			errcode_valid = 0;
2073 	}
2074 
2075 	/*
2076 	 * From section 26.6.1 "Interruptibility State" in Intel SDM:
2077 	 *
2078 	 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
2079 	 * one instruction or incurs an exception.
2080 	 */
2081 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
2082 	KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
2083 	    __func__, error));
2084 
2085 	if (restart_instruction)
2086 		vm_restart_instruction(vm, vcpuid);
2087 
2088 	vcpu->exception_pending = 1;
2089 	vcpu->exc_vector = vector;
2090 	vcpu->exc_errcode = errcode;
2091 	vcpu->exc_errcode_valid = errcode_valid;
2092 	VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector);
2093 	return (0);
2094 }
2095 
2096 void
2097 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
2098     int errcode)
2099 {
2100 	struct vm *vm;
2101 	int error, restart_instruction;
2102 
2103 	vm = vmarg;
2104 	restart_instruction = 1;
2105 
2106 	error = vm_inject_exception(vm, vcpuid, vector, errcode_valid,
2107 	    errcode, restart_instruction);
2108 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
2109 }
2110 
2111 void
2112 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
2113 {
2114 	struct vm *vm;
2115 	int error;
2116 
2117 	vm = vmarg;
2118 	VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
2119 	    error_code, cr2);
2120 
2121 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
2122 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
2123 
2124 	vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
2125 }
2126 
2127 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
2128 
2129 int
2130 vm_inject_nmi(struct vm *vm, int vcpuid)
2131 {
2132 	struct vcpu *vcpu;
2133 
2134 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2135 		return (EINVAL);
2136 
2137 	vcpu = &vm->vcpu[vcpuid];
2138 
2139 	vcpu->nmi_pending = 1;
2140 	vcpu_notify_event(vm, vcpuid, false);
2141 	return (0);
2142 }
2143 
2144 int
2145 vm_nmi_pending(struct vm *vm, int vcpuid)
2146 {
2147 	struct vcpu *vcpu;
2148 
2149 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2150 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2151 
2152 	vcpu = &vm->vcpu[vcpuid];
2153 
2154 	return (vcpu->nmi_pending);
2155 }
2156 
2157 void
2158 vm_nmi_clear(struct vm *vm, int vcpuid)
2159 {
2160 	struct vcpu *vcpu;
2161 
2162 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2163 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2164 
2165 	vcpu = &vm->vcpu[vcpuid];
2166 
2167 	if (vcpu->nmi_pending == 0)
2168 		panic("vm_nmi_clear: inconsistent nmi_pending state");
2169 
2170 	vcpu->nmi_pending = 0;
2171 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
2172 }
2173 
2174 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
2175 
2176 int
2177 vm_inject_extint(struct vm *vm, int vcpuid)
2178 {
2179 	struct vcpu *vcpu;
2180 
2181 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2182 		return (EINVAL);
2183 
2184 	vcpu = &vm->vcpu[vcpuid];
2185 
2186 	vcpu->extint_pending = 1;
2187 	vcpu_notify_event(vm, vcpuid, false);
2188 	return (0);
2189 }
2190 
2191 int
2192 vm_extint_pending(struct vm *vm, int vcpuid)
2193 {
2194 	struct vcpu *vcpu;
2195 
2196 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2197 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2198 
2199 	vcpu = &vm->vcpu[vcpuid];
2200 
2201 	return (vcpu->extint_pending);
2202 }
2203 
2204 void
2205 vm_extint_clear(struct vm *vm, int vcpuid)
2206 {
2207 	struct vcpu *vcpu;
2208 
2209 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2210 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2211 
2212 	vcpu = &vm->vcpu[vcpuid];
2213 
2214 	if (vcpu->extint_pending == 0)
2215 		panic("vm_extint_clear: inconsistent extint_pending state");
2216 
2217 	vcpu->extint_pending = 0;
2218 	vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
2219 }
2220 
2221 int
2222 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
2223 {
2224 	if (vcpu < 0 || vcpu >= vm->maxcpus)
2225 		return (EINVAL);
2226 
2227 	if (type < 0 || type >= VM_CAP_MAX)
2228 		return (EINVAL);
2229 
2230 	return (VMGETCAP(vm->cookie, vcpu, type, retval));
2231 }
2232 
2233 int
2234 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
2235 {
2236 	if (vcpu < 0 || vcpu >= vm->maxcpus)
2237 		return (EINVAL);
2238 
2239 	if (type < 0 || type >= VM_CAP_MAX)
2240 		return (EINVAL);
2241 
2242 	return (VMSETCAP(vm->cookie, vcpu, type, val));
2243 }
2244 
2245 struct vlapic *
2246 vm_lapic(struct vm *vm, int cpu)
2247 {
2248 	return (vm->vcpu[cpu].vlapic);
2249 }
2250 
2251 struct vioapic *
2252 vm_ioapic(struct vm *vm)
2253 {
2254 
2255 	return (vm->vioapic);
2256 }
2257 
2258 struct vhpet *
2259 vm_hpet(struct vm *vm)
2260 {
2261 
2262 	return (vm->vhpet);
2263 }
2264 
2265 bool
2266 vmm_is_pptdev(int bus, int slot, int func)
2267 {
2268 	int b, f, i, n, s;
2269 	char *val, *cp, *cp2;
2270 	bool found;
2271 
2272 	/*
2273 	 * XXX
2274 	 * The length of an environment variable is limited to 128 bytes which
2275 	 * puts an upper limit on the number of passthru devices that may be
2276 	 * specified using a single environment variable.
2277 	 *
2278 	 * Work around this by scanning multiple environment variable
2279 	 * names instead of a single one - yuck!
2280 	 */
2281 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2282 
2283 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2284 	found = false;
2285 	for (i = 0; names[i] != NULL && !found; i++) {
2286 		cp = val = kern_getenv(names[i]);
2287 		while (cp != NULL && *cp != '\0') {
2288 			if ((cp2 = strchr(cp, ' ')) != NULL)
2289 				*cp2 = '\0';
2290 
2291 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2292 			if (n == 3 && bus == b && slot == s && func == f) {
2293 				found = true;
2294 				break;
2295 			}
2296 
2297 			if (cp2 != NULL)
2298 				*cp2++ = ' ';
2299 
2300 			cp = cp2;
2301 		}
2302 		freeenv(val);
2303 	}
2304 	return (found);
2305 }
2306 
2307 void *
2308 vm_iommu_domain(struct vm *vm)
2309 {
2310 
2311 	return (vm->iommu);
2312 }
2313 
2314 int
2315 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
2316     bool from_idle)
2317 {
2318 	int error;
2319 	struct vcpu *vcpu;
2320 
2321 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2322 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
2323 
2324 	vcpu = &vm->vcpu[vcpuid];
2325 
2326 	vcpu_lock(vcpu);
2327 	error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle);
2328 	vcpu_unlock(vcpu);
2329 
2330 	return (error);
2331 }
2332 
2333 enum vcpu_state
2334 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2335 {
2336 	struct vcpu *vcpu;
2337 	enum vcpu_state state;
2338 
2339 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2340 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2341 
2342 	vcpu = &vm->vcpu[vcpuid];
2343 
2344 	vcpu_lock(vcpu);
2345 	state = vcpu->state;
2346 	if (hostcpu != NULL)
2347 		*hostcpu = vcpu->hostcpu;
2348 	vcpu_unlock(vcpu);
2349 
2350 	return (state);
2351 }
2352 
2353 int
2354 vm_activate_cpu(struct vm *vm, int vcpuid)
2355 {
2356 
2357 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2358 		return (EINVAL);
2359 
2360 	if (CPU_ISSET(vcpuid, &vm->active_cpus))
2361 		return (EBUSY);
2362 
2363 	VCPU_CTR0(vm, vcpuid, "activated");
2364 	CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2365 	return (0);
2366 }
2367 
2368 int
2369 vm_suspend_cpu(struct vm *vm, int vcpuid)
2370 {
2371 	int i;
2372 
2373 	if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2374 		return (EINVAL);
2375 
2376 	if (vcpuid == -1) {
2377 		vm->debug_cpus = vm->active_cpus;
2378 		for (i = 0; i < vm->maxcpus; i++) {
2379 			if (CPU_ISSET(i, &vm->active_cpus))
2380 				vcpu_notify_event(vm, i, false);
2381 		}
2382 	} else {
2383 		if (!CPU_ISSET(vcpuid, &vm->active_cpus))
2384 			return (EINVAL);
2385 
2386 		CPU_SET_ATOMIC(vcpuid, &vm->debug_cpus);
2387 		vcpu_notify_event(vm, vcpuid, false);
2388 	}
2389 	return (0);
2390 }
2391 
2392 int
2393 vm_resume_cpu(struct vm *vm, int vcpuid)
2394 {
2395 
2396 	if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2397 		return (EINVAL);
2398 
2399 	if (vcpuid == -1) {
2400 		CPU_ZERO(&vm->debug_cpus);
2401 	} else {
2402 		if (!CPU_ISSET(vcpuid, &vm->debug_cpus))
2403 			return (EINVAL);
2404 
2405 		CPU_CLR_ATOMIC(vcpuid, &vm->debug_cpus);
2406 	}
2407 	return (0);
2408 }
2409 
2410 int
2411 vcpu_debugged(struct vm *vm, int vcpuid)
2412 {
2413 
2414 	return (CPU_ISSET(vcpuid, &vm->debug_cpus));
2415 }
2416 
2417 cpuset_t
2418 vm_active_cpus(struct vm *vm)
2419 {
2420 
2421 	return (vm->active_cpus);
2422 }
2423 
2424 cpuset_t
2425 vm_debug_cpus(struct vm *vm)
2426 {
2427 
2428 	return (vm->debug_cpus);
2429 }
2430 
2431 cpuset_t
2432 vm_suspended_cpus(struct vm *vm)
2433 {
2434 
2435 	return (vm->suspended_cpus);
2436 }
2437 
2438 void *
2439 vcpu_stats(struct vm *vm, int vcpuid)
2440 {
2441 
2442 	return (vm->vcpu[vcpuid].stats);
2443 }
2444 
2445 int
2446 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2447 {
2448 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2449 		return (EINVAL);
2450 
2451 	*state = vm->vcpu[vcpuid].x2apic_state;
2452 
2453 	return (0);
2454 }
2455 
2456 int
2457 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2458 {
2459 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2460 		return (EINVAL);
2461 
2462 	if (state >= X2APIC_STATE_LAST)
2463 		return (EINVAL);
2464 
2465 	vm->vcpu[vcpuid].x2apic_state = state;
2466 
2467 	vlapic_set_x2apic_state(vm, vcpuid, state);
2468 
2469 	return (0);
2470 }
2471 
2472 /*
2473  * This function is called to ensure that a vcpu "sees" a pending event
2474  * as soon as possible:
2475  * - If the vcpu thread is sleeping then it is woken up.
2476  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2477  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2478  */
2479 static void
2480 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
2481 {
2482 	int hostcpu;
2483 
2484 	hostcpu = vcpu->hostcpu;
2485 	if (vcpu->state == VCPU_RUNNING) {
2486 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2487 		if (hostcpu != curcpu) {
2488 			if (lapic_intr) {
2489 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2490 				    vmm_ipinum);
2491 			} else {
2492 				ipi_cpu(hostcpu, vmm_ipinum);
2493 			}
2494 		} else {
2495 			/*
2496 			 * If the 'vcpu' is running on 'curcpu' then it must
2497 			 * be sending a notification to itself (e.g. SELF_IPI).
2498 			 * The pending event will be picked up when the vcpu
2499 			 * transitions back to guest context.
2500 			 */
2501 		}
2502 	} else {
2503 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2504 		    "with hostcpu %d", vcpu->state, hostcpu));
2505 		if (vcpu->state == VCPU_SLEEPING)
2506 			wakeup_one(vcpu);
2507 	}
2508 }
2509 
2510 void
2511 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2512 {
2513 	struct vcpu *vcpu = &vm->vcpu[vcpuid];
2514 
2515 	vcpu_lock(vcpu);
2516 	vcpu_notify_event_locked(vcpu, lapic_intr);
2517 	vcpu_unlock(vcpu);
2518 }
2519 
2520 struct vmspace *
2521 vm_get_vmspace(struct vm *vm)
2522 {
2523 
2524 	return (vm->vmspace);
2525 }
2526 
2527 int
2528 vm_apicid2vcpuid(struct vm *vm, int apicid)
2529 {
2530 	/*
2531 	 * XXX apic id is assumed to be numerically identical to vcpu id
2532 	 */
2533 	return (apicid);
2534 }
2535 
2536 int
2537 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2538     vm_rendezvous_func_t func, void *arg)
2539 {
2540 	int error, i;
2541 
2542 	/*
2543 	 * Enforce that this function is called without any locks
2544 	 */
2545 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2546 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
2547 	    ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2548 
2549 restart:
2550 	mtx_lock(&vm->rendezvous_mtx);
2551 	if (vm->rendezvous_func != NULL) {
2552 		/*
2553 		 * If a rendezvous is already in progress then we need to
2554 		 * call the rendezvous handler in case this 'vcpuid' is one
2555 		 * of the targets of the rendezvous.
2556 		 */
2557 		RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2558 		mtx_unlock(&vm->rendezvous_mtx);
2559 		error = vm_handle_rendezvous(vm, vcpuid);
2560 		if (error != 0)
2561 			return (error);
2562 		goto restart;
2563 	}
2564 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2565 	    "rendezvous is still in progress"));
2566 
2567 	RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2568 	vm->rendezvous_req_cpus = dest;
2569 	CPU_ZERO(&vm->rendezvous_done_cpus);
2570 	vm->rendezvous_arg = arg;
2571 	vm->rendezvous_func = func;
2572 	mtx_unlock(&vm->rendezvous_mtx);
2573 
2574 	/*
2575 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2576 	 * vcpus so they handle the rendezvous as soon as possible.
2577 	 */
2578 	for (i = 0; i < vm->maxcpus; i++) {
2579 		if (CPU_ISSET(i, &dest))
2580 			vcpu_notify_event(vm, i, false);
2581 	}
2582 
2583 	return (vm_handle_rendezvous(vm, vcpuid));
2584 }
2585 
2586 struct vatpic *
2587 vm_atpic(struct vm *vm)
2588 {
2589 	return (vm->vatpic);
2590 }
2591 
2592 struct vatpit *
2593 vm_atpit(struct vm *vm)
2594 {
2595 	return (vm->vatpit);
2596 }
2597 
2598 struct vpmtmr *
2599 vm_pmtmr(struct vm *vm)
2600 {
2601 
2602 	return (vm->vpmtmr);
2603 }
2604 
2605 struct vrtc *
2606 vm_rtc(struct vm *vm)
2607 {
2608 
2609 	return (vm->vrtc);
2610 }
2611 
2612 enum vm_reg_name
2613 vm_segment_name(int seg)
2614 {
2615 	static enum vm_reg_name seg_names[] = {
2616 		VM_REG_GUEST_ES,
2617 		VM_REG_GUEST_CS,
2618 		VM_REG_GUEST_SS,
2619 		VM_REG_GUEST_DS,
2620 		VM_REG_GUEST_FS,
2621 		VM_REG_GUEST_GS
2622 	};
2623 
2624 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2625 	    ("%s: invalid segment encoding %d", __func__, seg));
2626 	return (seg_names[seg]);
2627 }
2628 
2629 void
2630 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2631     int num_copyinfo)
2632 {
2633 	int idx;
2634 
2635 	for (idx = 0; idx < num_copyinfo; idx++) {
2636 		if (copyinfo[idx].cookie != NULL)
2637 			vm_gpa_release(copyinfo[idx].cookie);
2638 	}
2639 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2640 }
2641 
2642 int
2643 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2644     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2645     int num_copyinfo, int *fault)
2646 {
2647 	int error, idx, nused;
2648 	size_t n, off, remaining;
2649 	void *hva, *cookie;
2650 	uint64_t gpa;
2651 
2652 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2653 
2654 	nused = 0;
2655 	remaining = len;
2656 	while (remaining > 0) {
2657 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2658 		error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault);
2659 		if (error || *fault)
2660 			return (error);
2661 		off = gpa & PAGE_MASK;
2662 		n = min(remaining, PAGE_SIZE - off);
2663 		copyinfo[nused].gpa = gpa;
2664 		copyinfo[nused].len = n;
2665 		remaining -= n;
2666 		gla += n;
2667 		nused++;
2668 	}
2669 
2670 	for (idx = 0; idx < nused; idx++) {
2671 		hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa,
2672 		    copyinfo[idx].len, prot, &cookie);
2673 		if (hva == NULL)
2674 			break;
2675 		copyinfo[idx].hva = hva;
2676 		copyinfo[idx].cookie = cookie;
2677 	}
2678 
2679 	if (idx != nused) {
2680 		vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2681 		return (EFAULT);
2682 	} else {
2683 		*fault = 0;
2684 		return (0);
2685 	}
2686 }
2687 
2688 void
2689 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2690     size_t len)
2691 {
2692 	char *dst;
2693 	int idx;
2694 
2695 	dst = kaddr;
2696 	idx = 0;
2697 	while (len > 0) {
2698 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2699 		len -= copyinfo[idx].len;
2700 		dst += copyinfo[idx].len;
2701 		idx++;
2702 	}
2703 }
2704 
2705 void
2706 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2707     struct vm_copyinfo *copyinfo, size_t len)
2708 {
2709 	const char *src;
2710 	int idx;
2711 
2712 	src = kaddr;
2713 	idx = 0;
2714 	while (len > 0) {
2715 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2716 		len -= copyinfo[idx].len;
2717 		src += copyinfo[idx].len;
2718 		idx++;
2719 	}
2720 }
2721 
2722 /*
2723  * Return the amount of in-use and wired memory for the VM. Since
2724  * these are global stats, only return the values with for vCPU 0
2725  */
2726 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2727 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2728 
2729 static void
2730 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2731 {
2732 
2733 	if (vcpu == 0) {
2734 		vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2735 	       	    PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2736 	}
2737 }
2738 
2739 static void
2740 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2741 {
2742 
2743 	if (vcpu == 0) {
2744 		vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2745 	      	    PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2746 	}
2747 }
2748 
2749 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2750 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2751 
2752 #ifdef BHYVE_SNAPSHOT
2753 static int
2754 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta)
2755 {
2756 	int ret;
2757 	int i;
2758 	struct vcpu *vcpu;
2759 
2760 	for (i = 0; i < VM_MAXCPU; i++) {
2761 		vcpu = &vm->vcpu[i];
2762 
2763 		SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done);
2764 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done);
2765 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done);
2766 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done);
2767 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done);
2768 		SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done);
2769 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done);
2770 		SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done);
2771 		/* XXX we're cheating here, since the value of tsc_offset as
2772 		 * saved here is actually the value of the guest's TSC value.
2773 		 *
2774 		 * It will be turned turned back into an actual offset when the
2775 		 * TSC restore function is called
2776 		 */
2777 		SNAPSHOT_VAR_OR_LEAVE(vcpu->tsc_offset, meta, ret, done);
2778 	}
2779 
2780 done:
2781 	return (ret);
2782 }
2783 
2784 static int
2785 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta)
2786 {
2787 	int ret;
2788 	int i;
2789 	uint64_t now;
2790 
2791 	ret = 0;
2792 	now = rdtsc();
2793 
2794 	if (meta->op == VM_SNAPSHOT_SAVE) {
2795 		/* XXX make tsc_offset take the value TSC proper as seen by the
2796 		 * guest
2797 		 */
2798 		for (i = 0; i < VM_MAXCPU; i++)
2799 			vm->vcpu[i].tsc_offset += now;
2800 	}
2801 
2802 	ret = vm_snapshot_vcpus(vm, meta);
2803 	if (ret != 0) {
2804 		printf("%s: failed to copy vm data to user buffer", __func__);
2805 		goto done;
2806 	}
2807 
2808 	if (meta->op == VM_SNAPSHOT_SAVE) {
2809 		/* XXX turn tsc_offset back into an offset; actual value is only
2810 		 * required for restore; using it otherwise would be wrong
2811 		 */
2812 		for (i = 0; i < VM_MAXCPU; i++)
2813 			vm->vcpu[i].tsc_offset -= now;
2814 	}
2815 
2816 done:
2817 	return (ret);
2818 }
2819 
2820 static int
2821 vm_snapshot_vmcx(struct vm *vm, struct vm_snapshot_meta *meta)
2822 {
2823 	int i, error;
2824 
2825 	error = 0;
2826 
2827 	for (i = 0; i < VM_MAXCPU; i++) {
2828 		error = VM_SNAPSHOT_VMCX(vm->cookie, meta, i);
2829 		if (error != 0) {
2830 			printf("%s: failed to snapshot vmcs/vmcb data for "
2831 			       "vCPU: %d; error: %d\n", __func__, i, error);
2832 			goto done;
2833 		}
2834 	}
2835 
2836 done:
2837 	return (error);
2838 }
2839 
2840 /*
2841  * Save kernel-side structures to user-space for snapshotting.
2842  */
2843 int
2844 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta)
2845 {
2846 	int ret = 0;
2847 
2848 	switch (meta->dev_req) {
2849 	case STRUCT_VMX:
2850 		ret = VM_SNAPSHOT_VMI(vm->cookie, meta);
2851 		break;
2852 	case STRUCT_VMCX:
2853 		ret = vm_snapshot_vmcx(vm, meta);
2854 		break;
2855 	case STRUCT_VM:
2856 		ret = vm_snapshot_vm(vm, meta);
2857 		break;
2858 	case STRUCT_VIOAPIC:
2859 		ret = vioapic_snapshot(vm_ioapic(vm), meta);
2860 		break;
2861 	case STRUCT_VLAPIC:
2862 		ret = vlapic_snapshot(vm, meta);
2863 		break;
2864 	case STRUCT_VHPET:
2865 		ret = vhpet_snapshot(vm_hpet(vm), meta);
2866 		break;
2867 	case STRUCT_VATPIC:
2868 		ret = vatpic_snapshot(vm_atpic(vm), meta);
2869 		break;
2870 	case STRUCT_VATPIT:
2871 		ret = vatpit_snapshot(vm_atpit(vm), meta);
2872 		break;
2873 	case STRUCT_VPMTMR:
2874 		ret = vpmtmr_snapshot(vm_pmtmr(vm), meta);
2875 		break;
2876 	case STRUCT_VRTC:
2877 		ret = vrtc_snapshot(vm_rtc(vm), meta);
2878 		break;
2879 	default:
2880 		printf("%s: failed to find the requested type %#x\n",
2881 		       __func__, meta->dev_req);
2882 		ret = (EINVAL);
2883 	}
2884 	return (ret);
2885 }
2886 
2887 int
2888 vm_set_tsc_offset(struct vm *vm, int vcpuid, uint64_t offset)
2889 {
2890 	struct vcpu *vcpu;
2891 
2892 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2893 		return (EINVAL);
2894 
2895 	vcpu = &vm->vcpu[vcpuid];
2896 	vcpu->tsc_offset = offset;
2897 
2898 	return (0);
2899 }
2900 
2901 int
2902 vm_restore_time(struct vm *vm)
2903 {
2904 	int error, i;
2905 	uint64_t now;
2906 	struct vcpu *vcpu;
2907 
2908 	now = rdtsc();
2909 
2910 	error = vhpet_restore_time(vm_hpet(vm));
2911 	if (error)
2912 		return (error);
2913 
2914 	for (i = 0; i < nitems(vm->vcpu); i++) {
2915 		vcpu = &vm->vcpu[i];
2916 
2917 		error = VM_RESTORE_TSC(vm->cookie, i, vcpu->tsc_offset - now);
2918 		if (error)
2919 			return (error);
2920 	}
2921 
2922 	return (0);
2923 }
2924 #endif
2925