xref: /freebsd/sys/amd64/vmm/vmm.c (revision ae41709ab46305df80f7f35bb478a3c8ebf22ebb)
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 (gpa >= mm->gpa && gpa < mm->gpa + mm->len) {
1003 			count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
1004 			    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
1005 			break;
1006 		}
1007 	}
1008 
1009 	if (count == 1) {
1010 		*cookie = m;
1011 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
1012 	} else {
1013 		*cookie = NULL;
1014 		return (NULL);
1015 	}
1016 }
1017 
1018 void
1019 vm_gpa_release(void *cookie)
1020 {
1021 	vm_page_t m = cookie;
1022 
1023 	vm_page_unwire(m, PQ_ACTIVE);
1024 }
1025 
1026 int
1027 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
1028 {
1029 
1030 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1031 		return (EINVAL);
1032 
1033 	if (reg >= VM_REG_LAST)
1034 		return (EINVAL);
1035 
1036 	return (VMGETREG(vm->cookie, vcpu, reg, retval));
1037 }
1038 
1039 int
1040 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val)
1041 {
1042 	struct vcpu *vcpu;
1043 	int error;
1044 
1045 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1046 		return (EINVAL);
1047 
1048 	if (reg >= VM_REG_LAST)
1049 		return (EINVAL);
1050 
1051 	error = VMSETREG(vm->cookie, vcpuid, reg, val);
1052 	if (error || reg != VM_REG_GUEST_RIP)
1053 		return (error);
1054 
1055 	/* Set 'nextrip' to match the value of %rip */
1056 	VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val);
1057 	vcpu = &vm->vcpu[vcpuid];
1058 	vcpu->nextrip = val;
1059 	return (0);
1060 }
1061 
1062 static bool
1063 is_descriptor_table(int reg)
1064 {
1065 
1066 	switch (reg) {
1067 	case VM_REG_GUEST_IDTR:
1068 	case VM_REG_GUEST_GDTR:
1069 		return (true);
1070 	default:
1071 		return (false);
1072 	}
1073 }
1074 
1075 static bool
1076 is_segment_register(int reg)
1077 {
1078 
1079 	switch (reg) {
1080 	case VM_REG_GUEST_ES:
1081 	case VM_REG_GUEST_CS:
1082 	case VM_REG_GUEST_SS:
1083 	case VM_REG_GUEST_DS:
1084 	case VM_REG_GUEST_FS:
1085 	case VM_REG_GUEST_GS:
1086 	case VM_REG_GUEST_TR:
1087 	case VM_REG_GUEST_LDTR:
1088 		return (true);
1089 	default:
1090 		return (false);
1091 	}
1092 }
1093 
1094 int
1095 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
1096 		struct seg_desc *desc)
1097 {
1098 
1099 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1100 		return (EINVAL);
1101 
1102 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1103 		return (EINVAL);
1104 
1105 	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
1106 }
1107 
1108 int
1109 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
1110 		struct seg_desc *desc)
1111 {
1112 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1113 		return (EINVAL);
1114 
1115 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1116 		return (EINVAL);
1117 
1118 	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
1119 }
1120 
1121 static void
1122 restore_guest_fpustate(struct vcpu *vcpu)
1123 {
1124 
1125 	/* flush host state to the pcb */
1126 	fpuexit(curthread);
1127 
1128 	/* restore guest FPU state */
1129 	fpu_stop_emulating();
1130 	fpurestore(vcpu->guestfpu);
1131 
1132 	/* restore guest XCR0 if XSAVE is enabled in the host */
1133 	if (rcr4() & CR4_XSAVE)
1134 		load_xcr(0, vcpu->guest_xcr0);
1135 
1136 	/*
1137 	 * The FPU is now "dirty" with the guest's state so turn on emulation
1138 	 * to trap any access to the FPU by the host.
1139 	 */
1140 	fpu_start_emulating();
1141 }
1142 
1143 static void
1144 save_guest_fpustate(struct vcpu *vcpu)
1145 {
1146 
1147 	if ((rcr0() & CR0_TS) == 0)
1148 		panic("fpu emulation not enabled in host!");
1149 
1150 	/* save guest XCR0 and restore host XCR0 */
1151 	if (rcr4() & CR4_XSAVE) {
1152 		vcpu->guest_xcr0 = rxcr(0);
1153 		load_xcr(0, vmm_get_host_xcr0());
1154 	}
1155 
1156 	/* save guest FPU state */
1157 	fpu_stop_emulating();
1158 	fpusave(vcpu->guestfpu);
1159 	fpu_start_emulating();
1160 }
1161 
1162 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
1163 
1164 static int
1165 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1166     bool from_idle)
1167 {
1168 	struct vcpu *vcpu;
1169 	int error;
1170 
1171 	vcpu = &vm->vcpu[vcpuid];
1172 	vcpu_assert_locked(vcpu);
1173 
1174 	/*
1175 	 * State transitions from the vmmdev_ioctl() must always begin from
1176 	 * the VCPU_IDLE state. This guarantees that there is only a single
1177 	 * ioctl() operating on a vcpu at any point.
1178 	 */
1179 	if (from_idle) {
1180 		while (vcpu->state != VCPU_IDLE) {
1181 			vcpu->reqidle = 1;
1182 			vcpu_notify_event_locked(vcpu, false);
1183 			VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to "
1184 			    "idle requested", vcpu_state2str(vcpu->state));
1185 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
1186 		}
1187 	} else {
1188 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1189 		    "vcpu idle state"));
1190 	}
1191 
1192 	if (vcpu->state == VCPU_RUNNING) {
1193 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1194 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1195 	} else {
1196 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1197 		    "vcpu that is not running", vcpu->hostcpu));
1198 	}
1199 
1200 	/*
1201 	 * The following state transitions are allowed:
1202 	 * IDLE -> FROZEN -> IDLE
1203 	 * FROZEN -> RUNNING -> FROZEN
1204 	 * FROZEN -> SLEEPING -> FROZEN
1205 	 */
1206 	switch (vcpu->state) {
1207 	case VCPU_IDLE:
1208 	case VCPU_RUNNING:
1209 	case VCPU_SLEEPING:
1210 		error = (newstate != VCPU_FROZEN);
1211 		break;
1212 	case VCPU_FROZEN:
1213 		error = (newstate == VCPU_FROZEN);
1214 		break;
1215 	default:
1216 		error = 1;
1217 		break;
1218 	}
1219 
1220 	if (error)
1221 		return (EBUSY);
1222 
1223 	VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s",
1224 	    vcpu_state2str(vcpu->state), vcpu_state2str(newstate));
1225 
1226 	vcpu->state = newstate;
1227 	if (newstate == VCPU_RUNNING)
1228 		vcpu->hostcpu = curcpu;
1229 	else
1230 		vcpu->hostcpu = NOCPU;
1231 
1232 	if (newstate == VCPU_IDLE)
1233 		wakeup(&vcpu->state);
1234 
1235 	return (0);
1236 }
1237 
1238 static void
1239 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1240 {
1241 	int error;
1242 
1243 	if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1244 		panic("Error %d setting state to %d\n", error, newstate);
1245 }
1246 
1247 static void
1248 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1249 {
1250 	int error;
1251 
1252 	if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0)
1253 		panic("Error %d setting state to %d", error, newstate);
1254 }
1255 
1256 #define	RENDEZVOUS_CTR0(vm, vcpuid, fmt)				\
1257 	do {								\
1258 		if (vcpuid >= 0)					\
1259 			VCPU_CTR0(vm, vcpuid, fmt);			\
1260 		else							\
1261 			VM_CTR0(vm, fmt);				\
1262 	} while (0)
1263 
1264 static int
1265 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1266 {
1267 	struct thread *td;
1268 	int error;
1269 
1270 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
1271 	    ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1272 
1273 	error = 0;
1274 	td = curthread;
1275 	mtx_lock(&vm->rendezvous_mtx);
1276 	while (vm->rendezvous_func != NULL) {
1277 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1278 		CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1279 
1280 		if (vcpuid != -1 &&
1281 		    CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1282 		    !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1283 			VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1284 			(*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1285 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1286 		}
1287 		if (CPU_CMP(&vm->rendezvous_req_cpus,
1288 		    &vm->rendezvous_done_cpus) == 0) {
1289 			VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1290 			vm->rendezvous_func = NULL;
1291 			wakeup(&vm->rendezvous_func);
1292 			break;
1293 		}
1294 		RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1295 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1296 		    "vmrndv", hz);
1297 		if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1298 			mtx_unlock(&vm->rendezvous_mtx);
1299 			error = thread_check_susp(td, true);
1300 			if (error != 0)
1301 				return (error);
1302 			mtx_lock(&vm->rendezvous_mtx);
1303 		}
1304 	}
1305 	mtx_unlock(&vm->rendezvous_mtx);
1306 	return (0);
1307 }
1308 
1309 /*
1310  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1311  */
1312 static int
1313 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1314 {
1315 	struct vcpu *vcpu;
1316 	const char *wmesg;
1317 	struct thread *td;
1318 	int error, t, vcpu_halted, vm_halted;
1319 
1320 	KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1321 
1322 	vcpu = &vm->vcpu[vcpuid];
1323 	vcpu_halted = 0;
1324 	vm_halted = 0;
1325 	error = 0;
1326 	td = curthread;
1327 
1328 	vcpu_lock(vcpu);
1329 	while (1) {
1330 		/*
1331 		 * Do a final check for pending NMI or interrupts before
1332 		 * really putting this thread to sleep. Also check for
1333 		 * software events that would cause this vcpu to wakeup.
1334 		 *
1335 		 * These interrupts/events could have happened after the
1336 		 * vcpu returned from VMRUN() and before it acquired the
1337 		 * vcpu lock above.
1338 		 */
1339 		if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle)
1340 			break;
1341 		if (vm_nmi_pending(vm, vcpuid))
1342 			break;
1343 		if (!intr_disabled) {
1344 			if (vm_extint_pending(vm, vcpuid) ||
1345 			    vlapic_pending_intr(vcpu->vlapic, NULL)) {
1346 				break;
1347 			}
1348 		}
1349 
1350 		/* Don't go to sleep if the vcpu thread needs to yield */
1351 		if (vcpu_should_yield(vm, vcpuid))
1352 			break;
1353 
1354 		if (vcpu_debugged(vm, vcpuid))
1355 			break;
1356 
1357 		/*
1358 		 * Some Linux guests implement "halt" by having all vcpus
1359 		 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1360 		 * track of the vcpus that have entered this state. When all
1361 		 * vcpus enter the halted state the virtual machine is halted.
1362 		 */
1363 		if (intr_disabled) {
1364 			wmesg = "vmhalt";
1365 			VCPU_CTR0(vm, vcpuid, "Halted");
1366 			if (!vcpu_halted && halt_detection_enabled) {
1367 				vcpu_halted = 1;
1368 				CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1369 			}
1370 			if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1371 				vm_halted = 1;
1372 				break;
1373 			}
1374 		} else {
1375 			wmesg = "vmidle";
1376 		}
1377 
1378 		t = ticks;
1379 		vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1380 		/*
1381 		 * XXX msleep_spin() cannot be interrupted by signals so
1382 		 * wake up periodically to check pending signals.
1383 		 */
1384 		msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1385 		vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1386 		vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1387 		if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1388 			vcpu_unlock(vcpu);
1389 			error = thread_check_susp(td, false);
1390 			if (error != 0)
1391 				return (error);
1392 			vcpu_lock(vcpu);
1393 		}
1394 	}
1395 
1396 	if (vcpu_halted)
1397 		CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1398 
1399 	vcpu_unlock(vcpu);
1400 
1401 	if (vm_halted)
1402 		vm_suspend(vm, VM_SUSPEND_HALT);
1403 
1404 	return (0);
1405 }
1406 
1407 static int
1408 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1409 {
1410 	int rv, ftype;
1411 	struct vm_map *map;
1412 	struct vcpu *vcpu;
1413 	struct vm_exit *vme;
1414 
1415 	vcpu = &vm->vcpu[vcpuid];
1416 	vme = &vcpu->exitinfo;
1417 
1418 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1419 	    __func__, vme->inst_length));
1420 
1421 	ftype = vme->u.paging.fault_type;
1422 	KASSERT(ftype == VM_PROT_READ ||
1423 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1424 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1425 
1426 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1427 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1428 		    vme->u.paging.gpa, ftype);
1429 		if (rv == 0) {
1430 			VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1431 			    ftype == VM_PROT_READ ? "accessed" : "dirty",
1432 			    vme->u.paging.gpa);
1433 			goto done;
1434 		}
1435 	}
1436 
1437 	map = &vm->vmspace->vm_map;
1438 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL);
1439 
1440 	VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1441 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1442 
1443 	if (rv != KERN_SUCCESS)
1444 		return (EFAULT);
1445 done:
1446 	return (0);
1447 }
1448 
1449 static int
1450 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1451 {
1452 	struct vie *vie;
1453 	struct vcpu *vcpu;
1454 	struct vm_exit *vme;
1455 	uint64_t gla, gpa, cs_base;
1456 	struct vm_guest_paging *paging;
1457 	mem_region_read_t mread;
1458 	mem_region_write_t mwrite;
1459 	enum vm_cpu_mode cpu_mode;
1460 	int cs_d, error, fault;
1461 
1462 	vcpu = &vm->vcpu[vcpuid];
1463 	vme = &vcpu->exitinfo;
1464 
1465 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1466 	    __func__, vme->inst_length));
1467 
1468 	gla = vme->u.inst_emul.gla;
1469 	gpa = vme->u.inst_emul.gpa;
1470 	cs_base = vme->u.inst_emul.cs_base;
1471 	cs_d = vme->u.inst_emul.cs_d;
1472 	vie = &vme->u.inst_emul.vie;
1473 	paging = &vme->u.inst_emul.paging;
1474 	cpu_mode = paging->cpu_mode;
1475 
1476 	VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1477 
1478 	/* Fetch, decode and emulate the faulting instruction */
1479 	if (vie->num_valid == 0) {
1480 		error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip +
1481 		    cs_base, VIE_INST_SIZE, vie, &fault);
1482 	} else {
1483 		/*
1484 		 * The instruction bytes have already been copied into 'vie'
1485 		 */
1486 		error = fault = 0;
1487 	}
1488 	if (error || fault)
1489 		return (error);
1490 
1491 	if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) {
1492 		VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx",
1493 		    vme->rip + cs_base);
1494 		*retu = true;	    /* dump instruction bytes in userspace */
1495 		return (0);
1496 	}
1497 
1498 	/*
1499 	 * Update 'nextrip' based on the length of the emulated instruction.
1500 	 */
1501 	vme->inst_length = vie->num_processed;
1502 	vcpu->nextrip += vie->num_processed;
1503 	VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction "
1504 	    "decoding", vcpu->nextrip);
1505 
1506 	/* return to userland unless this is an in-kernel emulated device */
1507 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1508 		mread = lapic_mmio_read;
1509 		mwrite = lapic_mmio_write;
1510 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1511 		mread = vioapic_mmio_read;
1512 		mwrite = vioapic_mmio_write;
1513 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1514 		mread = vhpet_mmio_read;
1515 		mwrite = vhpet_mmio_write;
1516 	} else {
1517 		*retu = true;
1518 		return (0);
1519 	}
1520 
1521 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1522 	    mread, mwrite, retu);
1523 
1524 	return (error);
1525 }
1526 
1527 static int
1528 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1529 {
1530 	int error, i;
1531 	struct vcpu *vcpu;
1532 	struct thread *td;
1533 
1534 	error = 0;
1535 	vcpu = &vm->vcpu[vcpuid];
1536 	td = curthread;
1537 
1538 	CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1539 
1540 	/*
1541 	 * Wait until all 'active_cpus' have suspended themselves.
1542 	 *
1543 	 * Since a VM may be suspended at any time including when one or
1544 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1545 	 * handler while we are waiting to prevent a deadlock.
1546 	 */
1547 	vcpu_lock(vcpu);
1548 	while (error == 0) {
1549 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1550 			VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1551 			break;
1552 		}
1553 
1554 		if (vm->rendezvous_func == NULL) {
1555 			VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1556 			vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1557 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1558 			vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1559 			if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1560 				vcpu_unlock(vcpu);
1561 				error = thread_check_susp(td, false);
1562 				vcpu_lock(vcpu);
1563 			}
1564 		} else {
1565 			VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1566 			vcpu_unlock(vcpu);
1567 			error = vm_handle_rendezvous(vm, vcpuid);
1568 			vcpu_lock(vcpu);
1569 		}
1570 	}
1571 	vcpu_unlock(vcpu);
1572 
1573 	/*
1574 	 * Wakeup the other sleeping vcpus and return to userspace.
1575 	 */
1576 	for (i = 0; i < vm->maxcpus; i++) {
1577 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1578 			vcpu_notify_event(vm, i, false);
1579 		}
1580 	}
1581 
1582 	*retu = true;
1583 	return (error);
1584 }
1585 
1586 static int
1587 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu)
1588 {
1589 	struct vcpu *vcpu = &vm->vcpu[vcpuid];
1590 
1591 	vcpu_lock(vcpu);
1592 	KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle));
1593 	vcpu->reqidle = 0;
1594 	vcpu_unlock(vcpu);
1595 	*retu = true;
1596 	return (0);
1597 }
1598 
1599 int
1600 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1601 {
1602 	int i;
1603 
1604 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1605 		return (EINVAL);
1606 
1607 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1608 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1609 		    vm->suspend, how);
1610 		return (EALREADY);
1611 	}
1612 
1613 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1614 
1615 	/*
1616 	 * Notify all active vcpus that they are now suspended.
1617 	 */
1618 	for (i = 0; i < vm->maxcpus; i++) {
1619 		if (CPU_ISSET(i, &vm->active_cpus))
1620 			vcpu_notify_event(vm, i, false);
1621 	}
1622 
1623 	return (0);
1624 }
1625 
1626 void
1627 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1628 {
1629 	struct vm_exit *vmexit;
1630 
1631 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1632 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1633 
1634 	vmexit = vm_exitinfo(vm, vcpuid);
1635 	vmexit->rip = rip;
1636 	vmexit->inst_length = 0;
1637 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1638 	vmexit->u.suspended.how = vm->suspend;
1639 }
1640 
1641 void
1642 vm_exit_debug(struct vm *vm, int vcpuid, uint64_t rip)
1643 {
1644 	struct vm_exit *vmexit;
1645 
1646 	vmexit = vm_exitinfo(vm, vcpuid);
1647 	vmexit->rip = rip;
1648 	vmexit->inst_length = 0;
1649 	vmexit->exitcode = VM_EXITCODE_DEBUG;
1650 }
1651 
1652 void
1653 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1654 {
1655 	struct vm_exit *vmexit;
1656 
1657 	KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1658 
1659 	vmexit = vm_exitinfo(vm, vcpuid);
1660 	vmexit->rip = rip;
1661 	vmexit->inst_length = 0;
1662 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1663 	vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1664 }
1665 
1666 void
1667 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip)
1668 {
1669 	struct vm_exit *vmexit;
1670 
1671 	vmexit = vm_exitinfo(vm, vcpuid);
1672 	vmexit->rip = rip;
1673 	vmexit->inst_length = 0;
1674 	vmexit->exitcode = VM_EXITCODE_REQIDLE;
1675 	vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1);
1676 }
1677 
1678 void
1679 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1680 {
1681 	struct vm_exit *vmexit;
1682 
1683 	vmexit = vm_exitinfo(vm, vcpuid);
1684 	vmexit->rip = rip;
1685 	vmexit->inst_length = 0;
1686 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1687 	vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1688 }
1689 
1690 int
1691 vm_run(struct vm *vm, struct vm_run *vmrun)
1692 {
1693 	struct vm_eventinfo evinfo;
1694 	int error, vcpuid;
1695 	struct vcpu *vcpu;
1696 	struct pcb *pcb;
1697 	uint64_t tscval;
1698 	struct vm_exit *vme;
1699 	bool retu, intr_disabled;
1700 	pmap_t pmap;
1701 
1702 	vcpuid = vmrun->cpuid;
1703 
1704 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1705 		return (EINVAL);
1706 
1707 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1708 		return (EINVAL);
1709 
1710 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1711 		return (EINVAL);
1712 
1713 	pmap = vmspace_pmap(vm->vmspace);
1714 	vcpu = &vm->vcpu[vcpuid];
1715 	vme = &vcpu->exitinfo;
1716 	evinfo.rptr = &vm->rendezvous_func;
1717 	evinfo.sptr = &vm->suspend;
1718 	evinfo.iptr = &vcpu->reqidle;
1719 restart:
1720 	critical_enter();
1721 
1722 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1723 	    ("vm_run: absurd pm_active"));
1724 
1725 	tscval = rdtsc();
1726 
1727 	pcb = PCPU_GET(curpcb);
1728 	set_pcb_flags(pcb, PCB_FULL_IRET);
1729 
1730 	restore_guest_fpustate(vcpu);
1731 
1732 	vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1733 	error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo);
1734 	vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1735 
1736 	save_guest_fpustate(vcpu);
1737 
1738 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1739 
1740 	critical_exit();
1741 
1742 	if (error == 0) {
1743 		retu = false;
1744 		vcpu->nextrip = vme->rip + vme->inst_length;
1745 		switch (vme->exitcode) {
1746 		case VM_EXITCODE_REQIDLE:
1747 			error = vm_handle_reqidle(vm, vcpuid, &retu);
1748 			break;
1749 		case VM_EXITCODE_SUSPENDED:
1750 			error = vm_handle_suspend(vm, vcpuid, &retu);
1751 			break;
1752 		case VM_EXITCODE_IOAPIC_EOI:
1753 			vioapic_process_eoi(vm, vcpuid,
1754 			    vme->u.ioapic_eoi.vector);
1755 			break;
1756 		case VM_EXITCODE_RENDEZVOUS:
1757 			error = vm_handle_rendezvous(vm, vcpuid);
1758 			break;
1759 		case VM_EXITCODE_HLT:
1760 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1761 			error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1762 			break;
1763 		case VM_EXITCODE_PAGING:
1764 			error = vm_handle_paging(vm, vcpuid, &retu);
1765 			break;
1766 		case VM_EXITCODE_INST_EMUL:
1767 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1768 			break;
1769 		case VM_EXITCODE_INOUT:
1770 		case VM_EXITCODE_INOUT_STR:
1771 			error = vm_handle_inout(vm, vcpuid, vme, &retu);
1772 			break;
1773 		case VM_EXITCODE_MONITOR:
1774 		case VM_EXITCODE_MWAIT:
1775 		case VM_EXITCODE_VMINSN:
1776 			vm_inject_ud(vm, vcpuid);
1777 			break;
1778 		default:
1779 			retu = true;	/* handled in userland */
1780 			break;
1781 		}
1782 	}
1783 
1784 	if (error == 0 && retu == false)
1785 		goto restart;
1786 
1787 	VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode);
1788 
1789 	/* copy the exit information */
1790 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1791 	return (error);
1792 }
1793 
1794 int
1795 vm_restart_instruction(void *arg, int vcpuid)
1796 {
1797 	struct vm *vm;
1798 	struct vcpu *vcpu;
1799 	enum vcpu_state state;
1800 	uint64_t rip;
1801 	int error;
1802 
1803 	vm = arg;
1804 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1805 		return (EINVAL);
1806 
1807 	vcpu = &vm->vcpu[vcpuid];
1808 	state = vcpu_get_state(vm, vcpuid, NULL);
1809 	if (state == VCPU_RUNNING) {
1810 		/*
1811 		 * When a vcpu is "running" the next instruction is determined
1812 		 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
1813 		 * Thus setting 'inst_length' to zero will cause the current
1814 		 * instruction to be restarted.
1815 		 */
1816 		vcpu->exitinfo.inst_length = 0;
1817 		VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by "
1818 		    "setting inst_length to zero", vcpu->exitinfo.rip);
1819 	} else if (state == VCPU_FROZEN) {
1820 		/*
1821 		 * When a vcpu is "frozen" it is outside the critical section
1822 		 * around VMRUN() and 'nextrip' points to the next instruction.
1823 		 * Thus instruction restart is achieved by setting 'nextrip'
1824 		 * to the vcpu's %rip.
1825 		 */
1826 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip);
1827 		KASSERT(!error, ("%s: error %d getting rip", __func__, error));
1828 		VCPU_CTR2(vm, vcpuid, "restarting instruction by updating "
1829 		    "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
1830 		vcpu->nextrip = rip;
1831 	} else {
1832 		panic("%s: invalid state %d", __func__, state);
1833 	}
1834 	return (0);
1835 }
1836 
1837 int
1838 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1839 {
1840 	struct vcpu *vcpu;
1841 	int type, vector;
1842 
1843 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1844 		return (EINVAL);
1845 
1846 	vcpu = &vm->vcpu[vcpuid];
1847 
1848 	if (info & VM_INTINFO_VALID) {
1849 		type = info & VM_INTINFO_TYPE;
1850 		vector = info & 0xff;
1851 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1852 			return (EINVAL);
1853 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1854 			return (EINVAL);
1855 		if (info & VM_INTINFO_RSVD)
1856 			return (EINVAL);
1857 	} else {
1858 		info = 0;
1859 	}
1860 	VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1861 	vcpu->exitintinfo = info;
1862 	return (0);
1863 }
1864 
1865 enum exc_class {
1866 	EXC_BENIGN,
1867 	EXC_CONTRIBUTORY,
1868 	EXC_PAGEFAULT
1869 };
1870 
1871 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
1872 
1873 static enum exc_class
1874 exception_class(uint64_t info)
1875 {
1876 	int type, vector;
1877 
1878 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1879 	type = info & VM_INTINFO_TYPE;
1880 	vector = info & 0xff;
1881 
1882 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1883 	switch (type) {
1884 	case VM_INTINFO_HWINTR:
1885 	case VM_INTINFO_SWINTR:
1886 	case VM_INTINFO_NMI:
1887 		return (EXC_BENIGN);
1888 	default:
1889 		/*
1890 		 * Hardware exception.
1891 		 *
1892 		 * SVM and VT-x use identical type values to represent NMI,
1893 		 * hardware interrupt and software interrupt.
1894 		 *
1895 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1896 		 * for exceptions except #BP and #OF. #BP and #OF use a type
1897 		 * value of '5' or '6'. Therefore we don't check for explicit
1898 		 * values of 'type' to classify 'intinfo' into a hardware
1899 		 * exception.
1900 		 */
1901 		break;
1902 	}
1903 
1904 	switch (vector) {
1905 	case IDT_PF:
1906 	case IDT_VE:
1907 		return (EXC_PAGEFAULT);
1908 	case IDT_DE:
1909 	case IDT_TS:
1910 	case IDT_NP:
1911 	case IDT_SS:
1912 	case IDT_GP:
1913 		return (EXC_CONTRIBUTORY);
1914 	default:
1915 		return (EXC_BENIGN);
1916 	}
1917 }
1918 
1919 static int
1920 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1921     uint64_t *retinfo)
1922 {
1923 	enum exc_class exc1, exc2;
1924 	int type1, vector1;
1925 
1926 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1927 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1928 
1929 	/*
1930 	 * If an exception occurs while attempting to call the double-fault
1931 	 * handler the processor enters shutdown mode (aka triple fault).
1932 	 */
1933 	type1 = info1 & VM_INTINFO_TYPE;
1934 	vector1 = info1 & 0xff;
1935 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1936 		VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1937 		    info1, info2);
1938 		vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1939 		*retinfo = 0;
1940 		return (0);
1941 	}
1942 
1943 	/*
1944 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1945 	 */
1946 	exc1 = exception_class(info1);
1947 	exc2 = exception_class(info2);
1948 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1949 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1950 		/* Convert nested fault into a double fault. */
1951 		*retinfo = IDT_DF;
1952 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1953 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
1954 	} else {
1955 		/* Handle exceptions serially */
1956 		*retinfo = info2;
1957 	}
1958 	return (1);
1959 }
1960 
1961 static uint64_t
1962 vcpu_exception_intinfo(struct vcpu *vcpu)
1963 {
1964 	uint64_t info = 0;
1965 
1966 	if (vcpu->exception_pending) {
1967 		info = vcpu->exc_vector & 0xff;
1968 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1969 		if (vcpu->exc_errcode_valid) {
1970 			info |= VM_INTINFO_DEL_ERRCODE;
1971 			info |= (uint64_t)vcpu->exc_errcode << 32;
1972 		}
1973 	}
1974 	return (info);
1975 }
1976 
1977 int
1978 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1979 {
1980 	struct vcpu *vcpu;
1981 	uint64_t info1, info2;
1982 	int valid;
1983 
1984 	KASSERT(vcpuid >= 0 &&
1985 	    vcpuid < vm->maxcpus, ("invalid vcpu %d", vcpuid));
1986 
1987 	vcpu = &vm->vcpu[vcpuid];
1988 
1989 	info1 = vcpu->exitintinfo;
1990 	vcpu->exitintinfo = 0;
1991 
1992 	info2 = 0;
1993 	if (vcpu->exception_pending) {
1994 		info2 = vcpu_exception_intinfo(vcpu);
1995 		vcpu->exception_pending = 0;
1996 		VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1997 		    vcpu->exc_vector, info2);
1998 	}
1999 
2000 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
2001 		valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
2002 	} else if (info1 & VM_INTINFO_VALID) {
2003 		*retinfo = info1;
2004 		valid = 1;
2005 	} else if (info2 & VM_INTINFO_VALID) {
2006 		*retinfo = info2;
2007 		valid = 1;
2008 	} else {
2009 		valid = 0;
2010 	}
2011 
2012 	if (valid) {
2013 		VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
2014 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
2015 	}
2016 
2017 	return (valid);
2018 }
2019 
2020 int
2021 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
2022 {
2023 	struct vcpu *vcpu;
2024 
2025 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2026 		return (EINVAL);
2027 
2028 	vcpu = &vm->vcpu[vcpuid];
2029 	*info1 = vcpu->exitintinfo;
2030 	*info2 = vcpu_exception_intinfo(vcpu);
2031 	return (0);
2032 }
2033 
2034 int
2035 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid,
2036     uint32_t errcode, int restart_instruction)
2037 {
2038 	struct vcpu *vcpu;
2039 	uint64_t regval;
2040 	int error;
2041 
2042 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2043 		return (EINVAL);
2044 
2045 	if (vector < 0 || vector >= 32)
2046 		return (EINVAL);
2047 
2048 	/*
2049 	 * A double fault exception should never be injected directly into
2050 	 * the guest. It is a derived exception that results from specific
2051 	 * combinations of nested faults.
2052 	 */
2053 	if (vector == IDT_DF)
2054 		return (EINVAL);
2055 
2056 	vcpu = &vm->vcpu[vcpuid];
2057 
2058 	if (vcpu->exception_pending) {
2059 		VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
2060 		    "pending exception %d", vector, vcpu->exc_vector);
2061 		return (EBUSY);
2062 	}
2063 
2064 	if (errcode_valid) {
2065 		/*
2066 		 * Exceptions don't deliver an error code in real mode.
2067 		 */
2068 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, &regval);
2069 		KASSERT(!error, ("%s: error %d getting CR0", __func__, error));
2070 		if (!(regval & CR0_PE))
2071 			errcode_valid = 0;
2072 	}
2073 
2074 	/*
2075 	 * From section 26.6.1 "Interruptibility State" in Intel SDM:
2076 	 *
2077 	 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
2078 	 * one instruction or incurs an exception.
2079 	 */
2080 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
2081 	KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
2082 	    __func__, error));
2083 
2084 	if (restart_instruction)
2085 		vm_restart_instruction(vm, vcpuid);
2086 
2087 	vcpu->exception_pending = 1;
2088 	vcpu->exc_vector = vector;
2089 	vcpu->exc_errcode = errcode;
2090 	vcpu->exc_errcode_valid = errcode_valid;
2091 	VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector);
2092 	return (0);
2093 }
2094 
2095 void
2096 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
2097     int errcode)
2098 {
2099 	struct vm *vm;
2100 	int error, restart_instruction;
2101 
2102 	vm = vmarg;
2103 	restart_instruction = 1;
2104 
2105 	error = vm_inject_exception(vm, vcpuid, vector, errcode_valid,
2106 	    errcode, restart_instruction);
2107 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
2108 }
2109 
2110 void
2111 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
2112 {
2113 	struct vm *vm;
2114 	int error;
2115 
2116 	vm = vmarg;
2117 	VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
2118 	    error_code, cr2);
2119 
2120 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
2121 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
2122 
2123 	vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
2124 }
2125 
2126 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
2127 
2128 int
2129 vm_inject_nmi(struct vm *vm, int vcpuid)
2130 {
2131 	struct vcpu *vcpu;
2132 
2133 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2134 		return (EINVAL);
2135 
2136 	vcpu = &vm->vcpu[vcpuid];
2137 
2138 	vcpu->nmi_pending = 1;
2139 	vcpu_notify_event(vm, vcpuid, false);
2140 	return (0);
2141 }
2142 
2143 int
2144 vm_nmi_pending(struct vm *vm, int vcpuid)
2145 {
2146 	struct vcpu *vcpu;
2147 
2148 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2149 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2150 
2151 	vcpu = &vm->vcpu[vcpuid];
2152 
2153 	return (vcpu->nmi_pending);
2154 }
2155 
2156 void
2157 vm_nmi_clear(struct vm *vm, int vcpuid)
2158 {
2159 	struct vcpu *vcpu;
2160 
2161 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2162 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2163 
2164 	vcpu = &vm->vcpu[vcpuid];
2165 
2166 	if (vcpu->nmi_pending == 0)
2167 		panic("vm_nmi_clear: inconsistent nmi_pending state");
2168 
2169 	vcpu->nmi_pending = 0;
2170 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
2171 }
2172 
2173 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
2174 
2175 int
2176 vm_inject_extint(struct vm *vm, int vcpuid)
2177 {
2178 	struct vcpu *vcpu;
2179 
2180 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2181 		return (EINVAL);
2182 
2183 	vcpu = &vm->vcpu[vcpuid];
2184 
2185 	vcpu->extint_pending = 1;
2186 	vcpu_notify_event(vm, vcpuid, false);
2187 	return (0);
2188 }
2189 
2190 int
2191 vm_extint_pending(struct vm *vm, int vcpuid)
2192 {
2193 	struct vcpu *vcpu;
2194 
2195 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2196 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2197 
2198 	vcpu = &vm->vcpu[vcpuid];
2199 
2200 	return (vcpu->extint_pending);
2201 }
2202 
2203 void
2204 vm_extint_clear(struct vm *vm, int vcpuid)
2205 {
2206 	struct vcpu *vcpu;
2207 
2208 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2209 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2210 
2211 	vcpu = &vm->vcpu[vcpuid];
2212 
2213 	if (vcpu->extint_pending == 0)
2214 		panic("vm_extint_clear: inconsistent extint_pending state");
2215 
2216 	vcpu->extint_pending = 0;
2217 	vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
2218 }
2219 
2220 int
2221 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
2222 {
2223 	if (vcpu < 0 || vcpu >= vm->maxcpus)
2224 		return (EINVAL);
2225 
2226 	if (type < 0 || type >= VM_CAP_MAX)
2227 		return (EINVAL);
2228 
2229 	return (VMGETCAP(vm->cookie, vcpu, type, retval));
2230 }
2231 
2232 int
2233 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
2234 {
2235 	if (vcpu < 0 || vcpu >= vm->maxcpus)
2236 		return (EINVAL);
2237 
2238 	if (type < 0 || type >= VM_CAP_MAX)
2239 		return (EINVAL);
2240 
2241 	return (VMSETCAP(vm->cookie, vcpu, type, val));
2242 }
2243 
2244 struct vlapic *
2245 vm_lapic(struct vm *vm, int cpu)
2246 {
2247 	return (vm->vcpu[cpu].vlapic);
2248 }
2249 
2250 struct vioapic *
2251 vm_ioapic(struct vm *vm)
2252 {
2253 
2254 	return (vm->vioapic);
2255 }
2256 
2257 struct vhpet *
2258 vm_hpet(struct vm *vm)
2259 {
2260 
2261 	return (vm->vhpet);
2262 }
2263 
2264 bool
2265 vmm_is_pptdev(int bus, int slot, int func)
2266 {
2267 	int b, f, i, n, s;
2268 	char *val, *cp, *cp2;
2269 	bool found;
2270 
2271 	/*
2272 	 * XXX
2273 	 * The length of an environment variable is limited to 128 bytes which
2274 	 * puts an upper limit on the number of passthru devices that may be
2275 	 * specified using a single environment variable.
2276 	 *
2277 	 * Work around this by scanning multiple environment variable
2278 	 * names instead of a single one - yuck!
2279 	 */
2280 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2281 
2282 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2283 	found = false;
2284 	for (i = 0; names[i] != NULL && !found; i++) {
2285 		cp = val = kern_getenv(names[i]);
2286 		while (cp != NULL && *cp != '\0') {
2287 			if ((cp2 = strchr(cp, ' ')) != NULL)
2288 				*cp2 = '\0';
2289 
2290 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2291 			if (n == 3 && bus == b && slot == s && func == f) {
2292 				found = true;
2293 				break;
2294 			}
2295 
2296 			if (cp2 != NULL)
2297 				*cp2++ = ' ';
2298 
2299 			cp = cp2;
2300 		}
2301 		freeenv(val);
2302 	}
2303 	return (found);
2304 }
2305 
2306 void *
2307 vm_iommu_domain(struct vm *vm)
2308 {
2309 
2310 	return (vm->iommu);
2311 }
2312 
2313 int
2314 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
2315     bool from_idle)
2316 {
2317 	int error;
2318 	struct vcpu *vcpu;
2319 
2320 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2321 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
2322 
2323 	vcpu = &vm->vcpu[vcpuid];
2324 
2325 	vcpu_lock(vcpu);
2326 	error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle);
2327 	vcpu_unlock(vcpu);
2328 
2329 	return (error);
2330 }
2331 
2332 enum vcpu_state
2333 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2334 {
2335 	struct vcpu *vcpu;
2336 	enum vcpu_state state;
2337 
2338 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2339 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2340 
2341 	vcpu = &vm->vcpu[vcpuid];
2342 
2343 	vcpu_lock(vcpu);
2344 	state = vcpu->state;
2345 	if (hostcpu != NULL)
2346 		*hostcpu = vcpu->hostcpu;
2347 	vcpu_unlock(vcpu);
2348 
2349 	return (state);
2350 }
2351 
2352 int
2353 vm_activate_cpu(struct vm *vm, int vcpuid)
2354 {
2355 
2356 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2357 		return (EINVAL);
2358 
2359 	if (CPU_ISSET(vcpuid, &vm->active_cpus))
2360 		return (EBUSY);
2361 
2362 	VCPU_CTR0(vm, vcpuid, "activated");
2363 	CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2364 	return (0);
2365 }
2366 
2367 int
2368 vm_suspend_cpu(struct vm *vm, int vcpuid)
2369 {
2370 	int i;
2371 
2372 	if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2373 		return (EINVAL);
2374 
2375 	if (vcpuid == -1) {
2376 		vm->debug_cpus = vm->active_cpus;
2377 		for (i = 0; i < vm->maxcpus; i++) {
2378 			if (CPU_ISSET(i, &vm->active_cpus))
2379 				vcpu_notify_event(vm, i, false);
2380 		}
2381 	} else {
2382 		if (!CPU_ISSET(vcpuid, &vm->active_cpus))
2383 			return (EINVAL);
2384 
2385 		CPU_SET_ATOMIC(vcpuid, &vm->debug_cpus);
2386 		vcpu_notify_event(vm, vcpuid, false);
2387 	}
2388 	return (0);
2389 }
2390 
2391 int
2392 vm_resume_cpu(struct vm *vm, int vcpuid)
2393 {
2394 
2395 	if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2396 		return (EINVAL);
2397 
2398 	if (vcpuid == -1) {
2399 		CPU_ZERO(&vm->debug_cpus);
2400 	} else {
2401 		if (!CPU_ISSET(vcpuid, &vm->debug_cpus))
2402 			return (EINVAL);
2403 
2404 		CPU_CLR_ATOMIC(vcpuid, &vm->debug_cpus);
2405 	}
2406 	return (0);
2407 }
2408 
2409 int
2410 vcpu_debugged(struct vm *vm, int vcpuid)
2411 {
2412 
2413 	return (CPU_ISSET(vcpuid, &vm->debug_cpus));
2414 }
2415 
2416 cpuset_t
2417 vm_active_cpus(struct vm *vm)
2418 {
2419 
2420 	return (vm->active_cpus);
2421 }
2422 
2423 cpuset_t
2424 vm_debug_cpus(struct vm *vm)
2425 {
2426 
2427 	return (vm->debug_cpus);
2428 }
2429 
2430 cpuset_t
2431 vm_suspended_cpus(struct vm *vm)
2432 {
2433 
2434 	return (vm->suspended_cpus);
2435 }
2436 
2437 void *
2438 vcpu_stats(struct vm *vm, int vcpuid)
2439 {
2440 
2441 	return (vm->vcpu[vcpuid].stats);
2442 }
2443 
2444 int
2445 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2446 {
2447 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2448 		return (EINVAL);
2449 
2450 	*state = vm->vcpu[vcpuid].x2apic_state;
2451 
2452 	return (0);
2453 }
2454 
2455 int
2456 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2457 {
2458 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2459 		return (EINVAL);
2460 
2461 	if (state >= X2APIC_STATE_LAST)
2462 		return (EINVAL);
2463 
2464 	vm->vcpu[vcpuid].x2apic_state = state;
2465 
2466 	vlapic_set_x2apic_state(vm, vcpuid, state);
2467 
2468 	return (0);
2469 }
2470 
2471 /*
2472  * This function is called to ensure that a vcpu "sees" a pending event
2473  * as soon as possible:
2474  * - If the vcpu thread is sleeping then it is woken up.
2475  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2476  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2477  */
2478 static void
2479 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
2480 {
2481 	int hostcpu;
2482 
2483 	hostcpu = vcpu->hostcpu;
2484 	if (vcpu->state == VCPU_RUNNING) {
2485 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2486 		if (hostcpu != curcpu) {
2487 			if (lapic_intr) {
2488 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2489 				    vmm_ipinum);
2490 			} else {
2491 				ipi_cpu(hostcpu, vmm_ipinum);
2492 			}
2493 		} else {
2494 			/*
2495 			 * If the 'vcpu' is running on 'curcpu' then it must
2496 			 * be sending a notification to itself (e.g. SELF_IPI).
2497 			 * The pending event will be picked up when the vcpu
2498 			 * transitions back to guest context.
2499 			 */
2500 		}
2501 	} else {
2502 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2503 		    "with hostcpu %d", vcpu->state, hostcpu));
2504 		if (vcpu->state == VCPU_SLEEPING)
2505 			wakeup_one(vcpu);
2506 	}
2507 }
2508 
2509 void
2510 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2511 {
2512 	struct vcpu *vcpu = &vm->vcpu[vcpuid];
2513 
2514 	vcpu_lock(vcpu);
2515 	vcpu_notify_event_locked(vcpu, lapic_intr);
2516 	vcpu_unlock(vcpu);
2517 }
2518 
2519 struct vmspace *
2520 vm_get_vmspace(struct vm *vm)
2521 {
2522 
2523 	return (vm->vmspace);
2524 }
2525 
2526 int
2527 vm_apicid2vcpuid(struct vm *vm, int apicid)
2528 {
2529 	/*
2530 	 * XXX apic id is assumed to be numerically identical to vcpu id
2531 	 */
2532 	return (apicid);
2533 }
2534 
2535 int
2536 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2537     vm_rendezvous_func_t func, void *arg)
2538 {
2539 	int error, i;
2540 
2541 	/*
2542 	 * Enforce that this function is called without any locks
2543 	 */
2544 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2545 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
2546 	    ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2547 
2548 restart:
2549 	mtx_lock(&vm->rendezvous_mtx);
2550 	if (vm->rendezvous_func != NULL) {
2551 		/*
2552 		 * If a rendezvous is already in progress then we need to
2553 		 * call the rendezvous handler in case this 'vcpuid' is one
2554 		 * of the targets of the rendezvous.
2555 		 */
2556 		RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2557 		mtx_unlock(&vm->rendezvous_mtx);
2558 		error = vm_handle_rendezvous(vm, vcpuid);
2559 		if (error != 0)
2560 			return (error);
2561 		goto restart;
2562 	}
2563 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2564 	    "rendezvous is still in progress"));
2565 
2566 	RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2567 	vm->rendezvous_req_cpus = dest;
2568 	CPU_ZERO(&vm->rendezvous_done_cpus);
2569 	vm->rendezvous_arg = arg;
2570 	vm->rendezvous_func = func;
2571 	mtx_unlock(&vm->rendezvous_mtx);
2572 
2573 	/*
2574 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2575 	 * vcpus so they handle the rendezvous as soon as possible.
2576 	 */
2577 	for (i = 0; i < vm->maxcpus; i++) {
2578 		if (CPU_ISSET(i, &dest))
2579 			vcpu_notify_event(vm, i, false);
2580 	}
2581 
2582 	return (vm_handle_rendezvous(vm, vcpuid));
2583 }
2584 
2585 struct vatpic *
2586 vm_atpic(struct vm *vm)
2587 {
2588 	return (vm->vatpic);
2589 }
2590 
2591 struct vatpit *
2592 vm_atpit(struct vm *vm)
2593 {
2594 	return (vm->vatpit);
2595 }
2596 
2597 struct vpmtmr *
2598 vm_pmtmr(struct vm *vm)
2599 {
2600 
2601 	return (vm->vpmtmr);
2602 }
2603 
2604 struct vrtc *
2605 vm_rtc(struct vm *vm)
2606 {
2607 
2608 	return (vm->vrtc);
2609 }
2610 
2611 enum vm_reg_name
2612 vm_segment_name(int seg)
2613 {
2614 	static enum vm_reg_name seg_names[] = {
2615 		VM_REG_GUEST_ES,
2616 		VM_REG_GUEST_CS,
2617 		VM_REG_GUEST_SS,
2618 		VM_REG_GUEST_DS,
2619 		VM_REG_GUEST_FS,
2620 		VM_REG_GUEST_GS
2621 	};
2622 
2623 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2624 	    ("%s: invalid segment encoding %d", __func__, seg));
2625 	return (seg_names[seg]);
2626 }
2627 
2628 void
2629 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2630     int num_copyinfo)
2631 {
2632 	int idx;
2633 
2634 	for (idx = 0; idx < num_copyinfo; idx++) {
2635 		if (copyinfo[idx].cookie != NULL)
2636 			vm_gpa_release(copyinfo[idx].cookie);
2637 	}
2638 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2639 }
2640 
2641 int
2642 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2643     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2644     int num_copyinfo, int *fault)
2645 {
2646 	int error, idx, nused;
2647 	size_t n, off, remaining;
2648 	void *hva, *cookie;
2649 	uint64_t gpa;
2650 
2651 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2652 
2653 	nused = 0;
2654 	remaining = len;
2655 	while (remaining > 0) {
2656 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2657 		error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault);
2658 		if (error || *fault)
2659 			return (error);
2660 		off = gpa & PAGE_MASK;
2661 		n = min(remaining, PAGE_SIZE - off);
2662 		copyinfo[nused].gpa = gpa;
2663 		copyinfo[nused].len = n;
2664 		remaining -= n;
2665 		gla += n;
2666 		nused++;
2667 	}
2668 
2669 	for (idx = 0; idx < nused; idx++) {
2670 		hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa,
2671 		    copyinfo[idx].len, prot, &cookie);
2672 		if (hva == NULL)
2673 			break;
2674 		copyinfo[idx].hva = hva;
2675 		copyinfo[idx].cookie = cookie;
2676 	}
2677 
2678 	if (idx != nused) {
2679 		vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2680 		return (EFAULT);
2681 	} else {
2682 		*fault = 0;
2683 		return (0);
2684 	}
2685 }
2686 
2687 void
2688 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2689     size_t len)
2690 {
2691 	char *dst;
2692 	int idx;
2693 
2694 	dst = kaddr;
2695 	idx = 0;
2696 	while (len > 0) {
2697 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2698 		len -= copyinfo[idx].len;
2699 		dst += copyinfo[idx].len;
2700 		idx++;
2701 	}
2702 }
2703 
2704 void
2705 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2706     struct vm_copyinfo *copyinfo, size_t len)
2707 {
2708 	const char *src;
2709 	int idx;
2710 
2711 	src = kaddr;
2712 	idx = 0;
2713 	while (len > 0) {
2714 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2715 		len -= copyinfo[idx].len;
2716 		src += copyinfo[idx].len;
2717 		idx++;
2718 	}
2719 }
2720 
2721 /*
2722  * Return the amount of in-use and wired memory for the VM. Since
2723  * these are global stats, only return the values with for vCPU 0
2724  */
2725 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2726 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2727 
2728 static void
2729 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2730 {
2731 
2732 	if (vcpu == 0) {
2733 		vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2734 	       	    PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2735 	}
2736 }
2737 
2738 static void
2739 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2740 {
2741 
2742 	if (vcpu == 0) {
2743 		vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2744 	      	    PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2745 	}
2746 }
2747 
2748 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2749 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2750 
2751 #ifdef BHYVE_SNAPSHOT
2752 static int
2753 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta)
2754 {
2755 	int ret;
2756 	int i;
2757 	struct vcpu *vcpu;
2758 
2759 	for (i = 0; i < VM_MAXCPU; i++) {
2760 		vcpu = &vm->vcpu[i];
2761 
2762 		SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done);
2763 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done);
2764 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done);
2765 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done);
2766 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done);
2767 		SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done);
2768 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done);
2769 		SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done);
2770 		/* XXX we're cheating here, since the value of tsc_offset as
2771 		 * saved here is actually the value of the guest's TSC value.
2772 		 *
2773 		 * It will be turned turned back into an actual offset when the
2774 		 * TSC restore function is called
2775 		 */
2776 		SNAPSHOT_VAR_OR_LEAVE(vcpu->tsc_offset, meta, ret, done);
2777 	}
2778 
2779 done:
2780 	return (ret);
2781 }
2782 
2783 static int
2784 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta)
2785 {
2786 	int ret;
2787 	int i;
2788 	uint64_t now;
2789 
2790 	ret = 0;
2791 	now = rdtsc();
2792 
2793 	if (meta->op == VM_SNAPSHOT_SAVE) {
2794 		/* XXX make tsc_offset take the value TSC proper as seen by the
2795 		 * guest
2796 		 */
2797 		for (i = 0; i < VM_MAXCPU; i++)
2798 			vm->vcpu[i].tsc_offset += now;
2799 	}
2800 
2801 	ret = vm_snapshot_vcpus(vm, meta);
2802 	if (ret != 0) {
2803 		printf("%s: failed to copy vm data to user buffer", __func__);
2804 		goto done;
2805 	}
2806 
2807 	if (meta->op == VM_SNAPSHOT_SAVE) {
2808 		/* XXX turn tsc_offset back into an offset; actual value is only
2809 		 * required for restore; using it otherwise would be wrong
2810 		 */
2811 		for (i = 0; i < VM_MAXCPU; i++)
2812 			vm->vcpu[i].tsc_offset -= now;
2813 	}
2814 
2815 done:
2816 	return (ret);
2817 }
2818 
2819 static int
2820 vm_snapshot_vmcx(struct vm *vm, struct vm_snapshot_meta *meta)
2821 {
2822 	int i, error;
2823 
2824 	error = 0;
2825 
2826 	for (i = 0; i < VM_MAXCPU; i++) {
2827 		error = VM_SNAPSHOT_VMCX(vm->cookie, meta, i);
2828 		if (error != 0) {
2829 			printf("%s: failed to snapshot vmcs/vmcb data for "
2830 			       "vCPU: %d; error: %d\n", __func__, i, error);
2831 			goto done;
2832 		}
2833 	}
2834 
2835 done:
2836 	return (error);
2837 }
2838 
2839 /*
2840  * Save kernel-side structures to user-space for snapshotting.
2841  */
2842 int
2843 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta)
2844 {
2845 	int ret = 0;
2846 
2847 	switch (meta->dev_req) {
2848 	case STRUCT_VMX:
2849 		ret = VM_SNAPSHOT_VMI(vm->cookie, meta);
2850 		break;
2851 	case STRUCT_VMCX:
2852 		ret = vm_snapshot_vmcx(vm, meta);
2853 		break;
2854 	case STRUCT_VM:
2855 		ret = vm_snapshot_vm(vm, meta);
2856 		break;
2857 	case STRUCT_VIOAPIC:
2858 		ret = vioapic_snapshot(vm_ioapic(vm), meta);
2859 		break;
2860 	case STRUCT_VLAPIC:
2861 		ret = vlapic_snapshot(vm, meta);
2862 		break;
2863 	case STRUCT_VHPET:
2864 		ret = vhpet_snapshot(vm_hpet(vm), meta);
2865 		break;
2866 	case STRUCT_VATPIC:
2867 		ret = vatpic_snapshot(vm_atpic(vm), meta);
2868 		break;
2869 	case STRUCT_VATPIT:
2870 		ret = vatpit_snapshot(vm_atpit(vm), meta);
2871 		break;
2872 	case STRUCT_VPMTMR:
2873 		ret = vpmtmr_snapshot(vm_pmtmr(vm), meta);
2874 		break;
2875 	case STRUCT_VRTC:
2876 		ret = vrtc_snapshot(vm_rtc(vm), meta);
2877 		break;
2878 	default:
2879 		printf("%s: failed to find the requested type %#x\n",
2880 		       __func__, meta->dev_req);
2881 		ret = (EINVAL);
2882 	}
2883 	return (ret);
2884 }
2885 
2886 int
2887 vm_set_tsc_offset(struct vm *vm, int vcpuid, uint64_t offset)
2888 {
2889 	struct vcpu *vcpu;
2890 
2891 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2892 		return (EINVAL);
2893 
2894 	vcpu = &vm->vcpu[vcpuid];
2895 	vcpu->tsc_offset = offset;
2896 
2897 	return (0);
2898 }
2899 
2900 int
2901 vm_restore_time(struct vm *vm)
2902 {
2903 	int error, i;
2904 	uint64_t now;
2905 	struct vcpu *vcpu;
2906 
2907 	now = rdtsc();
2908 
2909 	error = vhpet_restore_time(vm_hpet(vm));
2910 	if (error)
2911 		return (error);
2912 
2913 	for (i = 0; i < nitems(vm->vcpu); i++) {
2914 		vcpu = &vm->vcpu[i];
2915 
2916 		error = VM_RESTORE_TSC(vm->cookie, i, vcpu->tsc_offset - now);
2917 		if (error)
2918 			return (error);
2919 	}
2920 
2921 	return (0);
2922 }
2923 #endif
2924