xref: /freebsd/sys/amd64/vmm/vmm.c (revision b9f654b163bce26de79705e77b872427c9f2afa1)
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 <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 #include <sys/module.h>
38 #include <sys/sysctl.h>
39 #include <sys/malloc.h>
40 #include <sys/pcpu.h>
41 #include <sys/lock.h>
42 #include <sys/mutex.h>
43 #include <sys/proc.h>
44 #include <sys/rwlock.h>
45 #include <sys/sched.h>
46 #include <sys/smp.h>
47 #include <sys/systm.h>
48 
49 #include <vm/vm.h>
50 #include <vm/vm_object.h>
51 #include <vm/vm_page.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_map.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_param.h>
56 
57 #include <machine/cpu.h>
58 #include <machine/pcb.h>
59 #include <machine/smp.h>
60 #include <machine/md_var.h>
61 #include <x86/psl.h>
62 #include <x86/apicreg.h>
63 
64 #include <machine/vmm.h>
65 #include <machine/vmm_dev.h>
66 #include <machine/vmm_instruction_emul.h>
67 
68 #include "vmm_ioport.h"
69 #include "vmm_ktr.h"
70 #include "vmm_host.h"
71 #include "vmm_mem.h"
72 #include "vmm_util.h"
73 #include "vatpic.h"
74 #include "vatpit.h"
75 #include "vhpet.h"
76 #include "vioapic.h"
77 #include "vlapic.h"
78 #include "vpmtmr.h"
79 #include "vrtc.h"
80 #include "vmm_stat.h"
81 #include "vmm_lapic.h"
82 
83 #include "io/ppt.h"
84 #include "io/iommu.h"
85 
86 struct vlapic;
87 
88 /*
89  * Initialization:
90  * (a) allocated when vcpu is created
91  * (i) initialized when vcpu is created and when it is reinitialized
92  * (o) initialized the first time the vcpu is created
93  * (x) initialized before use
94  */
95 struct vcpu {
96 	struct mtx 	mtx;		/* (o) protects 'state' and 'hostcpu' */
97 	enum vcpu_state	state;		/* (o) vcpu state */
98 	int		hostcpu;	/* (o) vcpu's host cpu */
99 	int		reqidle;	/* (i) request vcpu to idle */
100 	struct vlapic	*vlapic;	/* (i) APIC device model */
101 	enum x2apic_state x2apic_state;	/* (i) APIC mode */
102 	uint64_t	exitintinfo;	/* (i) events pending at VM exit */
103 	int		nmi_pending;	/* (i) NMI pending */
104 	int		extint_pending;	/* (i) INTR pending */
105 	int	exception_pending;	/* (i) exception pending */
106 	int	exc_vector;		/* (x) exception collateral */
107 	int	exc_errcode_valid;
108 	uint32_t exc_errcode;
109 	struct savefpu	*guestfpu;	/* (a,i) guest fpu state */
110 	uint64_t	guest_xcr0;	/* (i) guest %xcr0 register */
111 	void		*stats;		/* (a,i) statistics */
112 	struct vm_exit	exitinfo;	/* (x) exit reason and collateral */
113 	uint64_t	nextrip;	/* (x) next instruction to execute */
114 };
115 
116 #define	vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
117 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
118 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
119 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
120 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
121 
122 struct mem_seg {
123 	size_t	len;
124 	bool	sysmem;
125 	struct vm_object *object;
126 };
127 #define	VM_MAX_MEMSEGS	3
128 
129 struct mem_map {
130 	vm_paddr_t	gpa;
131 	size_t		len;
132 	vm_ooffset_t	segoff;
133 	int		segid;
134 	int		prot;
135 	int		flags;
136 };
137 #define	VM_MAX_MEMMAPS	4
138 
139 /*
140  * Initialization:
141  * (o) initialized the first time the VM is created
142  * (i) initialized when VM is created and when it is reinitialized
143  * (x) initialized before use
144  */
145 struct vm {
146 	void		*cookie;		/* (i) cpu-specific data */
147 	void		*iommu;			/* (x) iommu-specific data */
148 	struct vhpet	*vhpet;			/* (i) virtual HPET */
149 	struct vioapic	*vioapic;		/* (i) virtual ioapic */
150 	struct vatpic	*vatpic;		/* (i) virtual atpic */
151 	struct vatpit	*vatpit;		/* (i) virtual atpit */
152 	struct vpmtmr	*vpmtmr;		/* (i) virtual ACPI PM timer */
153 	struct vrtc	*vrtc;			/* (o) virtual RTC */
154 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
155 	volatile cpuset_t debug_cpus;		/* (i) vcpus stopped for debug */
156 	int		suspend;		/* (i) stop VM execution */
157 	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
158 	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
159 	cpuset_t	rendezvous_req_cpus;	/* (x) rendezvous requested */
160 	cpuset_t	rendezvous_done_cpus;	/* (x) rendezvous finished */
161 	void		*rendezvous_arg;	/* (x) rendezvous func/arg */
162 	vm_rendezvous_func_t rendezvous_func;
163 	struct mtx	rendezvous_mtx;		/* (o) rendezvous lock */
164 	struct mem_map	mem_maps[VM_MAX_MEMMAPS]; /* (i) guest address space */
165 	struct mem_seg	mem_segs[VM_MAX_MEMSEGS]; /* (o) guest memory regions */
166 	struct vmspace	*vmspace;		/* (o) guest's address space */
167 	char		name[VM_MAX_NAMELEN];	/* (o) virtual machine name */
168 	struct vcpu	vcpu[VM_MAXCPU];	/* (i) guest vcpus */
169 	/* The following describe the vm cpu topology */
170 	uint16_t	sockets;		/* (o) num of sockets */
171 	uint16_t	cores;			/* (o) num of cores/socket */
172 	uint16_t	threads;		/* (o) num of threads/core */
173 	uint16_t	maxcpus;		/* (o) max pluggable cpus */
174 };
175 
176 static int vmm_initialized;
177 
178 static struct vmm_ops *ops;
179 #define	VMM_INIT(num)	(ops != NULL ? (*ops->init)(num) : 0)
180 #define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
181 #define	VMM_RESUME()	(ops != NULL ? (*ops->resume)() : 0)
182 
183 #define	VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
184 #define	VMRUN(vmi, vcpu, rip, pmap, evinfo) \
185 	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, evinfo) : ENXIO)
186 #define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
187 #define	VMSPACE_ALLOC(min, max) \
188 	(ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
189 #define	VMSPACE_FREE(vmspace) \
190 	(ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
191 #define	VMGETREG(vmi, vcpu, num, retval)		\
192 	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
193 #define	VMSETREG(vmi, vcpu, num, val)		\
194 	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
195 #define	VMGETDESC(vmi, vcpu, num, desc)		\
196 	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
197 #define	VMSETDESC(vmi, vcpu, num, desc)		\
198 	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
199 #define	VMGETCAP(vmi, vcpu, num, retval)	\
200 	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
201 #define	VMSETCAP(vmi, vcpu, num, val)		\
202 	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
203 #define	VLAPIC_INIT(vmi, vcpu)			\
204 	(ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
205 #define	VLAPIC_CLEANUP(vmi, vlapic)		\
206 	(ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
207 
208 #define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
209 #define	fpu_stop_emulating()	clts()
210 
211 SDT_PROVIDER_DEFINE(vmm);
212 
213 static MALLOC_DEFINE(M_VM, "vm", "vm");
214 
215 /* statistics */
216 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
217 
218 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
219 
220 /*
221  * Halt the guest if all vcpus are executing a HLT instruction with
222  * interrupts disabled.
223  */
224 static int halt_detection_enabled = 1;
225 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
226     &halt_detection_enabled, 0,
227     "Halt VM if all vcpus execute HLT with interrupts disabled");
228 
229 static int vmm_ipinum;
230 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
231     "IPI vector used for vcpu notifications");
232 
233 static int trace_guest_exceptions;
234 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN,
235     &trace_guest_exceptions, 0,
236     "Trap into hypervisor on all guest exceptions and reflect them back");
237 
238 static void vm_free_memmap(struct vm *vm, int ident);
239 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm);
240 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr);
241 
242 #ifdef KTR
243 static const char *
244 vcpu_state2str(enum vcpu_state state)
245 {
246 
247 	switch (state) {
248 	case VCPU_IDLE:
249 		return ("idle");
250 	case VCPU_FROZEN:
251 		return ("frozen");
252 	case VCPU_RUNNING:
253 		return ("running");
254 	case VCPU_SLEEPING:
255 		return ("sleeping");
256 	default:
257 		return ("unknown");
258 	}
259 }
260 #endif
261 
262 static void
263 vcpu_cleanup(struct vm *vm, int i, bool destroy)
264 {
265 	struct vcpu *vcpu = &vm->vcpu[i];
266 
267 	VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
268 	if (destroy) {
269 		vmm_stat_free(vcpu->stats);
270 		fpu_save_area_free(vcpu->guestfpu);
271 	}
272 }
273 
274 static void
275 vcpu_init(struct vm *vm, int vcpu_id, bool create)
276 {
277 	struct vcpu *vcpu;
278 
279 	KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus,
280 	    ("vcpu_init: invalid vcpu %d", vcpu_id));
281 
282 	vcpu = &vm->vcpu[vcpu_id];
283 
284 	if (create) {
285 		KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
286 		    "initialized", vcpu_id));
287 		vcpu_lock_init(vcpu);
288 		vcpu->state = VCPU_IDLE;
289 		vcpu->hostcpu = NOCPU;
290 		vcpu->guestfpu = fpu_save_area_alloc();
291 		vcpu->stats = vmm_stat_alloc();
292 	}
293 
294 	vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
295 	vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
296 	vcpu->reqidle = 0;
297 	vcpu->exitintinfo = 0;
298 	vcpu->nmi_pending = 0;
299 	vcpu->extint_pending = 0;
300 	vcpu->exception_pending = 0;
301 	vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
302 	fpu_save_area_reset(vcpu->guestfpu);
303 	vmm_stat_init(vcpu->stats);
304 }
305 
306 int
307 vcpu_trace_exceptions(struct vm *vm, int vcpuid)
308 {
309 
310 	return (trace_guest_exceptions);
311 }
312 
313 struct vm_exit *
314 vm_exitinfo(struct vm *vm, int cpuid)
315 {
316 	struct vcpu *vcpu;
317 
318 	if (cpuid < 0 || cpuid >= vm->maxcpus)
319 		panic("vm_exitinfo: invalid cpuid %d", cpuid);
320 
321 	vcpu = &vm->vcpu[cpuid];
322 
323 	return (&vcpu->exitinfo);
324 }
325 
326 static void
327 vmm_resume(void)
328 {
329 	VMM_RESUME();
330 }
331 
332 static int
333 vmm_init(void)
334 {
335 	int error;
336 
337 	vmm_host_state_init();
338 
339 	vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) :
340 	    &IDTVEC(justreturn));
341 	if (vmm_ipinum < 0)
342 		vmm_ipinum = IPI_AST;
343 
344 	error = vmm_mem_init();
345 	if (error)
346 		return (error);
347 
348 	if (vmm_is_intel())
349 		ops = &vmm_ops_intel;
350 	else if (vmm_is_amd())
351 		ops = &vmm_ops_amd;
352 	else
353 		return (ENXIO);
354 
355 	vmm_resume_p = vmm_resume;
356 
357 	return (VMM_INIT(vmm_ipinum));
358 }
359 
360 static int
361 vmm_handler(module_t mod, int what, void *arg)
362 {
363 	int error;
364 
365 	switch (what) {
366 	case MOD_LOAD:
367 		vmmdev_init();
368 		error = vmm_init();
369 		if (error == 0)
370 			vmm_initialized = 1;
371 		break;
372 	case MOD_UNLOAD:
373 		error = vmmdev_cleanup();
374 		if (error == 0) {
375 			vmm_resume_p = NULL;
376 			iommu_cleanup();
377 			if (vmm_ipinum != IPI_AST)
378 				lapic_ipi_free(vmm_ipinum);
379 			error = VMM_CLEANUP();
380 			/*
381 			 * Something bad happened - prevent new
382 			 * VMs from being created
383 			 */
384 			if (error)
385 				vmm_initialized = 0;
386 		}
387 		break;
388 	default:
389 		error = 0;
390 		break;
391 	}
392 	return (error);
393 }
394 
395 static moduledata_t vmm_kmod = {
396 	"vmm",
397 	vmm_handler,
398 	NULL
399 };
400 
401 /*
402  * vmm initialization has the following dependencies:
403  *
404  * - VT-x initialization requires smp_rendezvous() and therefore must happen
405  *   after SMP is fully functional (after SI_SUB_SMP).
406  */
407 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
408 MODULE_VERSION(vmm, 1);
409 
410 static void
411 vm_init(struct vm *vm, bool create)
412 {
413 	int i;
414 
415 	vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
416 	vm->iommu = NULL;
417 	vm->vioapic = vioapic_init(vm);
418 	vm->vhpet = vhpet_init(vm);
419 	vm->vatpic = vatpic_init(vm);
420 	vm->vatpit = vatpit_init(vm);
421 	vm->vpmtmr = vpmtmr_init(vm);
422 	if (create)
423 		vm->vrtc = vrtc_init(vm);
424 
425 	CPU_ZERO(&vm->active_cpus);
426 	CPU_ZERO(&vm->debug_cpus);
427 
428 	vm->suspend = 0;
429 	CPU_ZERO(&vm->suspended_cpus);
430 
431 	for (i = 0; i < vm->maxcpus; i++)
432 		vcpu_init(vm, i, create);
433 }
434 
435 /*
436  * The default CPU topology is a single thread per package.
437  */
438 u_int cores_per_package = 1;
439 u_int threads_per_core = 1;
440 
441 int
442 vm_create(const char *name, struct vm **retvm)
443 {
444 	struct vm *vm;
445 	struct vmspace *vmspace;
446 
447 	/*
448 	 * If vmm.ko could not be successfully initialized then don't attempt
449 	 * to create the virtual machine.
450 	 */
451 	if (!vmm_initialized)
452 		return (ENXIO);
453 
454 	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
455 		return (EINVAL);
456 
457 	vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS);
458 	if (vmspace == NULL)
459 		return (ENOMEM);
460 
461 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
462 	strcpy(vm->name, name);
463 	vm->vmspace = vmspace;
464 	mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
465 
466 	vm->sockets = 1;
467 	vm->cores = cores_per_package;	/* XXX backwards compatibility */
468 	vm->threads = threads_per_core;	/* XXX backwards compatibility */
469 	vm->maxcpus = VM_MAXCPU;	/* XXX temp to keep code working */
470 
471 	vm_init(vm, true);
472 
473 	*retvm = vm;
474 	return (0);
475 }
476 
477 void
478 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
479     uint16_t *threads, uint16_t *maxcpus)
480 {
481 	*sockets = vm->sockets;
482 	*cores = vm->cores;
483 	*threads = vm->threads;
484 	*maxcpus = vm->maxcpus;
485 }
486 
487 uint16_t
488 vm_get_maxcpus(struct vm *vm)
489 {
490 	return (vm->maxcpus);
491 }
492 
493 int
494 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
495     uint16_t threads, uint16_t maxcpus)
496 {
497 	if (maxcpus != 0)
498 		return (EINVAL);	/* XXX remove when supported */
499 	if ((sockets * cores * threads) > vm->maxcpus)
500 		return (EINVAL);
501 	/* XXX need to check sockets * cores * threads == vCPU, how? */
502 	vm->sockets = sockets;
503 	vm->cores = cores;
504 	vm->threads = threads;
505 	vm->maxcpus = VM_MAXCPU;	/* XXX temp to keep code working */
506 	return(0);
507 }
508 
509 static void
510 vm_cleanup(struct vm *vm, bool destroy)
511 {
512 	struct mem_map *mm;
513 	int i;
514 
515 	ppt_unassign_all(vm);
516 
517 	if (vm->iommu != NULL)
518 		iommu_destroy_domain(vm->iommu);
519 
520 	if (destroy)
521 		vrtc_cleanup(vm->vrtc);
522 	else
523 		vrtc_reset(vm->vrtc);
524 	vpmtmr_cleanup(vm->vpmtmr);
525 	vatpit_cleanup(vm->vatpit);
526 	vhpet_cleanup(vm->vhpet);
527 	vatpic_cleanup(vm->vatpic);
528 	vioapic_cleanup(vm->vioapic);
529 
530 	for (i = 0; i < vm->maxcpus; i++)
531 		vcpu_cleanup(vm, i, destroy);
532 
533 	VMCLEANUP(vm->cookie);
534 
535 	/*
536 	 * System memory is removed from the guest address space only when
537 	 * the VM is destroyed. This is because the mapping remains the same
538 	 * across VM reset.
539 	 *
540 	 * Device memory can be relocated by the guest (e.g. using PCI BARs)
541 	 * so those mappings are removed on a VM reset.
542 	 */
543 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
544 		mm = &vm->mem_maps[i];
545 		if (destroy || !sysmem_mapping(vm, mm))
546 			vm_free_memmap(vm, i);
547 	}
548 
549 	if (destroy) {
550 		for (i = 0; i < VM_MAX_MEMSEGS; i++)
551 			vm_free_memseg(vm, i);
552 
553 		VMSPACE_FREE(vm->vmspace);
554 		vm->vmspace = NULL;
555 	}
556 }
557 
558 void
559 vm_destroy(struct vm *vm)
560 {
561 	vm_cleanup(vm, true);
562 	free(vm, M_VM);
563 }
564 
565 int
566 vm_reinit(struct vm *vm)
567 {
568 	int error;
569 
570 	/*
571 	 * A virtual machine can be reset only if all vcpus are suspended.
572 	 */
573 	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
574 		vm_cleanup(vm, false);
575 		vm_init(vm, false);
576 		error = 0;
577 	} else {
578 		error = EBUSY;
579 	}
580 
581 	return (error);
582 }
583 
584 const char *
585 vm_name(struct vm *vm)
586 {
587 	return (vm->name);
588 }
589 
590 int
591 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
592 {
593 	vm_object_t obj;
594 
595 	if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
596 		return (ENOMEM);
597 	else
598 		return (0);
599 }
600 
601 int
602 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
603 {
604 
605 	vmm_mmio_free(vm->vmspace, gpa, len);
606 	return (0);
607 }
608 
609 /*
610  * Return 'true' if 'gpa' is allocated in the guest address space.
611  *
612  * This function is called in the context of a running vcpu which acts as
613  * an implicit lock on 'vm->mem_maps[]'.
614  */
615 bool
616 vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa)
617 {
618 	struct mem_map *mm;
619 	int i;
620 
621 #ifdef INVARIANTS
622 	int hostcpu, state;
623 	state = vcpu_get_state(vm, vcpuid, &hostcpu);
624 	KASSERT(state == VCPU_RUNNING && hostcpu == curcpu,
625 	    ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu));
626 #endif
627 
628 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
629 		mm = &vm->mem_maps[i];
630 		if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len)
631 			return (true);		/* 'gpa' is sysmem or devmem */
632 	}
633 
634 	if (ppt_is_mmio(vm, gpa))
635 		return (true);			/* 'gpa' is pci passthru mmio */
636 
637 	return (false);
638 }
639 
640 int
641 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem)
642 {
643 	struct mem_seg *seg;
644 	vm_object_t obj;
645 
646 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
647 		return (EINVAL);
648 
649 	if (len == 0 || (len & PAGE_MASK))
650 		return (EINVAL);
651 
652 	seg = &vm->mem_segs[ident];
653 	if (seg->object != NULL) {
654 		if (seg->len == len && seg->sysmem == sysmem)
655 			return (EEXIST);
656 		else
657 			return (EINVAL);
658 	}
659 
660 	obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT);
661 	if (obj == NULL)
662 		return (ENOMEM);
663 
664 	seg->len = len;
665 	seg->object = obj;
666 	seg->sysmem = sysmem;
667 	return (0);
668 }
669 
670 int
671 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
672     vm_object_t *objptr)
673 {
674 	struct mem_seg *seg;
675 
676 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
677 		return (EINVAL);
678 
679 	seg = &vm->mem_segs[ident];
680 	if (len)
681 		*len = seg->len;
682 	if (sysmem)
683 		*sysmem = seg->sysmem;
684 	if (objptr)
685 		*objptr = seg->object;
686 	return (0);
687 }
688 
689 void
690 vm_free_memseg(struct vm *vm, int ident)
691 {
692 	struct mem_seg *seg;
693 
694 	KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS,
695 	    ("%s: invalid memseg ident %d", __func__, ident));
696 
697 	seg = &vm->mem_segs[ident];
698 	if (seg->object != NULL) {
699 		vm_object_deallocate(seg->object);
700 		bzero(seg, sizeof(struct mem_seg));
701 	}
702 }
703 
704 int
705 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first,
706     size_t len, int prot, int flags)
707 {
708 	struct mem_seg *seg;
709 	struct mem_map *m, *map;
710 	vm_ooffset_t last;
711 	int i, error;
712 
713 	if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0)
714 		return (EINVAL);
715 
716 	if (flags & ~VM_MEMMAP_F_WIRED)
717 		return (EINVAL);
718 
719 	if (segid < 0 || segid >= VM_MAX_MEMSEGS)
720 		return (EINVAL);
721 
722 	seg = &vm->mem_segs[segid];
723 	if (seg->object == NULL)
724 		return (EINVAL);
725 
726 	last = first + len;
727 	if (first < 0 || first >= last || last > seg->len)
728 		return (EINVAL);
729 
730 	if ((gpa | first | last) & PAGE_MASK)
731 		return (EINVAL);
732 
733 	map = NULL;
734 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
735 		m = &vm->mem_maps[i];
736 		if (m->len == 0) {
737 			map = m;
738 			break;
739 		}
740 	}
741 
742 	if (map == NULL)
743 		return (ENOSPC);
744 
745 	error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa,
746 	    len, 0, VMFS_NO_SPACE, prot, prot, 0);
747 	if (error != KERN_SUCCESS)
748 		return (EFAULT);
749 
750 	vm_object_reference(seg->object);
751 
752 	if (flags & VM_MEMMAP_F_WIRED) {
753 		error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len,
754 		    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
755 		if (error != KERN_SUCCESS) {
756 			vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len);
757 			return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM :
758 			    EFAULT);
759 		}
760 	}
761 
762 	map->gpa = gpa;
763 	map->len = len;
764 	map->segoff = first;
765 	map->segid = segid;
766 	map->prot = prot;
767 	map->flags = flags;
768 	return (0);
769 }
770 
771 int
772 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
773     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags)
774 {
775 	struct mem_map *mm, *mmnext;
776 	int i;
777 
778 	mmnext = NULL;
779 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
780 		mm = &vm->mem_maps[i];
781 		if (mm->len == 0 || mm->gpa < *gpa)
782 			continue;
783 		if (mmnext == NULL || mm->gpa < mmnext->gpa)
784 			mmnext = mm;
785 	}
786 
787 	if (mmnext != NULL) {
788 		*gpa = mmnext->gpa;
789 		if (segid)
790 			*segid = mmnext->segid;
791 		if (segoff)
792 			*segoff = mmnext->segoff;
793 		if (len)
794 			*len = mmnext->len;
795 		if (prot)
796 			*prot = mmnext->prot;
797 		if (flags)
798 			*flags = mmnext->flags;
799 		return (0);
800 	} else {
801 		return (ENOENT);
802 	}
803 }
804 
805 static void
806 vm_free_memmap(struct vm *vm, int ident)
807 {
808 	struct mem_map *mm;
809 	int error;
810 
811 	mm = &vm->mem_maps[ident];
812 	if (mm->len) {
813 		error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa,
814 		    mm->gpa + mm->len);
815 		KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d",
816 		    __func__, error));
817 		bzero(mm, sizeof(struct mem_map));
818 	}
819 }
820 
821 static __inline bool
822 sysmem_mapping(struct vm *vm, struct mem_map *mm)
823 {
824 
825 	if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem)
826 		return (true);
827 	else
828 		return (false);
829 }
830 
831 vm_paddr_t
832 vmm_sysmem_maxaddr(struct vm *vm)
833 {
834 	struct mem_map *mm;
835 	vm_paddr_t maxaddr;
836 	int i;
837 
838 	maxaddr = 0;
839 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
840 		mm = &vm->mem_maps[i];
841 		if (sysmem_mapping(vm, mm)) {
842 			if (maxaddr < mm->gpa + mm->len)
843 				maxaddr = mm->gpa + mm->len;
844 		}
845 	}
846 	return (maxaddr);
847 }
848 
849 static void
850 vm_iommu_modify(struct vm *vm, boolean_t map)
851 {
852 	int i, sz;
853 	vm_paddr_t gpa, hpa;
854 	struct mem_map *mm;
855 	void *vp, *cookie, *host_domain;
856 
857 	sz = PAGE_SIZE;
858 	host_domain = iommu_host_domain();
859 
860 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
861 		mm = &vm->mem_maps[i];
862 		if (!sysmem_mapping(vm, mm))
863 			continue;
864 
865 		if (map) {
866 			KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0,
867 			    ("iommu map found invalid memmap %#lx/%#lx/%#x",
868 			    mm->gpa, mm->len, mm->flags));
869 			if ((mm->flags & VM_MEMMAP_F_WIRED) == 0)
870 				continue;
871 			mm->flags |= VM_MEMMAP_F_IOMMU;
872 		} else {
873 			if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0)
874 				continue;
875 			mm->flags &= ~VM_MEMMAP_F_IOMMU;
876 			KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0,
877 			    ("iommu unmap found invalid memmap %#lx/%#lx/%#x",
878 			    mm->gpa, mm->len, mm->flags));
879 		}
880 
881 		gpa = mm->gpa;
882 		while (gpa < mm->gpa + mm->len) {
883 			vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE,
884 					 &cookie);
885 			KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
886 			    vm_name(vm), gpa));
887 
888 			vm_gpa_release(cookie);
889 
890 			hpa = DMAP_TO_PHYS((uintptr_t)vp);
891 			if (map) {
892 				iommu_create_mapping(vm->iommu, gpa, hpa, sz);
893 				iommu_remove_mapping(host_domain, hpa, sz);
894 			} else {
895 				iommu_remove_mapping(vm->iommu, gpa, sz);
896 				iommu_create_mapping(host_domain, hpa, hpa, sz);
897 			}
898 
899 			gpa += PAGE_SIZE;
900 		}
901 	}
902 
903 	/*
904 	 * Invalidate the cached translations associated with the domain
905 	 * from which pages were removed.
906 	 */
907 	if (map)
908 		iommu_invalidate_tlb(host_domain);
909 	else
910 		iommu_invalidate_tlb(vm->iommu);
911 }
912 
913 #define	vm_iommu_unmap(vm)	vm_iommu_modify((vm), FALSE)
914 #define	vm_iommu_map(vm)	vm_iommu_modify((vm), TRUE)
915 
916 int
917 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
918 {
919 	int error;
920 
921 	error = ppt_unassign_device(vm, bus, slot, func);
922 	if (error)
923 		return (error);
924 
925 	if (ppt_assigned_devices(vm) == 0)
926 		vm_iommu_unmap(vm);
927 
928 	return (0);
929 }
930 
931 int
932 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
933 {
934 	int error;
935 	vm_paddr_t maxaddr;
936 
937 	/* Set up the IOMMU to do the 'gpa' to 'hpa' translation */
938 	if (ppt_assigned_devices(vm) == 0) {
939 		KASSERT(vm->iommu == NULL,
940 		    ("vm_assign_pptdev: iommu must be NULL"));
941 		maxaddr = vmm_sysmem_maxaddr(vm);
942 		vm->iommu = iommu_create_domain(maxaddr);
943 		if (vm->iommu == NULL)
944 			return (ENXIO);
945 		vm_iommu_map(vm);
946 	}
947 
948 	error = ppt_assign_device(vm, bus, slot, func);
949 	return (error);
950 }
951 
952 void *
953 vm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot,
954 	    void **cookie)
955 {
956 	int i, count, pageoff;
957 	struct mem_map *mm;
958 	vm_page_t m;
959 #ifdef INVARIANTS
960 	/*
961 	 * All vcpus are frozen by ioctls that modify the memory map
962 	 * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is
963 	 * guaranteed if at least one vcpu is in the VCPU_FROZEN state.
964 	 */
965 	int state;
966 	KASSERT(vcpuid >= -1 && vcpuid < vm->maxcpus, ("%s: invalid vcpuid %d",
967 	    __func__, vcpuid));
968 	for (i = 0; i < vm->maxcpus; i++) {
969 		if (vcpuid != -1 && vcpuid != i)
970 			continue;
971 		state = vcpu_get_state(vm, i, NULL);
972 		KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d",
973 		    __func__, state));
974 	}
975 #endif
976 	pageoff = gpa & PAGE_MASK;
977 	if (len > PAGE_SIZE - pageoff)
978 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
979 
980 	count = 0;
981 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
982 		mm = &vm->mem_maps[i];
983 		if (sysmem_mapping(vm, mm) && gpa >= mm->gpa &&
984 		    gpa < mm->gpa + mm->len) {
985 			count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
986 			    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
987 			break;
988 		}
989 	}
990 
991 	if (count == 1) {
992 		*cookie = m;
993 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
994 	} else {
995 		*cookie = NULL;
996 		return (NULL);
997 	}
998 }
999 
1000 void
1001 vm_gpa_release(void *cookie)
1002 {
1003 	vm_page_t m = cookie;
1004 
1005 	vm_page_lock(m);
1006 	vm_page_unhold(m);
1007 	vm_page_unlock(m);
1008 }
1009 
1010 int
1011 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
1012 {
1013 
1014 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1015 		return (EINVAL);
1016 
1017 	if (reg >= VM_REG_LAST)
1018 		return (EINVAL);
1019 
1020 	return (VMGETREG(vm->cookie, vcpu, reg, retval));
1021 }
1022 
1023 int
1024 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val)
1025 {
1026 	struct vcpu *vcpu;
1027 	int error;
1028 
1029 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1030 		return (EINVAL);
1031 
1032 	if (reg >= VM_REG_LAST)
1033 		return (EINVAL);
1034 
1035 	error = VMSETREG(vm->cookie, vcpuid, reg, val);
1036 	if (error || reg != VM_REG_GUEST_RIP)
1037 		return (error);
1038 
1039 	/* Set 'nextrip' to match the value of %rip */
1040 	VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val);
1041 	vcpu = &vm->vcpu[vcpuid];
1042 	vcpu->nextrip = val;
1043 	return (0);
1044 }
1045 
1046 static boolean_t
1047 is_descriptor_table(int reg)
1048 {
1049 
1050 	switch (reg) {
1051 	case VM_REG_GUEST_IDTR:
1052 	case VM_REG_GUEST_GDTR:
1053 		return (TRUE);
1054 	default:
1055 		return (FALSE);
1056 	}
1057 }
1058 
1059 static boolean_t
1060 is_segment_register(int reg)
1061 {
1062 
1063 	switch (reg) {
1064 	case VM_REG_GUEST_ES:
1065 	case VM_REG_GUEST_CS:
1066 	case VM_REG_GUEST_SS:
1067 	case VM_REG_GUEST_DS:
1068 	case VM_REG_GUEST_FS:
1069 	case VM_REG_GUEST_GS:
1070 	case VM_REG_GUEST_TR:
1071 	case VM_REG_GUEST_LDTR:
1072 		return (TRUE);
1073 	default:
1074 		return (FALSE);
1075 	}
1076 }
1077 
1078 int
1079 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
1080 		struct seg_desc *desc)
1081 {
1082 
1083 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1084 		return (EINVAL);
1085 
1086 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1087 		return (EINVAL);
1088 
1089 	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
1090 }
1091 
1092 int
1093 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
1094 		struct seg_desc *desc)
1095 {
1096 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1097 		return (EINVAL);
1098 
1099 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1100 		return (EINVAL);
1101 
1102 	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
1103 }
1104 
1105 static void
1106 restore_guest_fpustate(struct vcpu *vcpu)
1107 {
1108 
1109 	/* flush host state to the pcb */
1110 	fpuexit(curthread);
1111 
1112 	/* restore guest FPU state */
1113 	fpu_stop_emulating();
1114 	fpurestore(vcpu->guestfpu);
1115 
1116 	/* restore guest XCR0 if XSAVE is enabled in the host */
1117 	if (rcr4() & CR4_XSAVE)
1118 		load_xcr(0, vcpu->guest_xcr0);
1119 
1120 	/*
1121 	 * The FPU is now "dirty" with the guest's state so turn on emulation
1122 	 * to trap any access to the FPU by the host.
1123 	 */
1124 	fpu_start_emulating();
1125 }
1126 
1127 static void
1128 save_guest_fpustate(struct vcpu *vcpu)
1129 {
1130 
1131 	if ((rcr0() & CR0_TS) == 0)
1132 		panic("fpu emulation not enabled in host!");
1133 
1134 	/* save guest XCR0 and restore host XCR0 */
1135 	if (rcr4() & CR4_XSAVE) {
1136 		vcpu->guest_xcr0 = rxcr(0);
1137 		load_xcr(0, vmm_get_host_xcr0());
1138 	}
1139 
1140 	/* save guest FPU state */
1141 	fpu_stop_emulating();
1142 	fpusave(vcpu->guestfpu);
1143 	fpu_start_emulating();
1144 }
1145 
1146 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
1147 
1148 static int
1149 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1150     bool from_idle)
1151 {
1152 	struct vcpu *vcpu;
1153 	int error;
1154 
1155 	vcpu = &vm->vcpu[vcpuid];
1156 	vcpu_assert_locked(vcpu);
1157 
1158 	/*
1159 	 * State transitions from the vmmdev_ioctl() must always begin from
1160 	 * the VCPU_IDLE state. This guarantees that there is only a single
1161 	 * ioctl() operating on a vcpu at any point.
1162 	 */
1163 	if (from_idle) {
1164 		while (vcpu->state != VCPU_IDLE) {
1165 			vcpu->reqidle = 1;
1166 			vcpu_notify_event_locked(vcpu, false);
1167 			VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to "
1168 			    "idle requested", vcpu_state2str(vcpu->state));
1169 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
1170 		}
1171 	} else {
1172 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1173 		    "vcpu idle state"));
1174 	}
1175 
1176 	if (vcpu->state == VCPU_RUNNING) {
1177 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1178 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1179 	} else {
1180 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1181 		    "vcpu that is not running", vcpu->hostcpu));
1182 	}
1183 
1184 	/*
1185 	 * The following state transitions are allowed:
1186 	 * IDLE -> FROZEN -> IDLE
1187 	 * FROZEN -> RUNNING -> FROZEN
1188 	 * FROZEN -> SLEEPING -> FROZEN
1189 	 */
1190 	switch (vcpu->state) {
1191 	case VCPU_IDLE:
1192 	case VCPU_RUNNING:
1193 	case VCPU_SLEEPING:
1194 		error = (newstate != VCPU_FROZEN);
1195 		break;
1196 	case VCPU_FROZEN:
1197 		error = (newstate == VCPU_FROZEN);
1198 		break;
1199 	default:
1200 		error = 1;
1201 		break;
1202 	}
1203 
1204 	if (error)
1205 		return (EBUSY);
1206 
1207 	VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s",
1208 	    vcpu_state2str(vcpu->state), vcpu_state2str(newstate));
1209 
1210 	vcpu->state = newstate;
1211 	if (newstate == VCPU_RUNNING)
1212 		vcpu->hostcpu = curcpu;
1213 	else
1214 		vcpu->hostcpu = NOCPU;
1215 
1216 	if (newstate == VCPU_IDLE)
1217 		wakeup(&vcpu->state);
1218 
1219 	return (0);
1220 }
1221 
1222 static void
1223 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1224 {
1225 	int error;
1226 
1227 	if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1228 		panic("Error %d setting state to %d\n", error, newstate);
1229 }
1230 
1231 static void
1232 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1233 {
1234 	int error;
1235 
1236 	if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0)
1237 		panic("Error %d setting state to %d", error, newstate);
1238 }
1239 
1240 static void
1241 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func)
1242 {
1243 
1244 	KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked"));
1245 
1246 	/*
1247 	 * Update 'rendezvous_func' and execute a write memory barrier to
1248 	 * ensure that it is visible across all host cpus. This is not needed
1249 	 * for correctness but it does ensure that all the vcpus will notice
1250 	 * that the rendezvous is requested immediately.
1251 	 */
1252 	vm->rendezvous_func = func;
1253 	wmb();
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 void
1265 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1266 {
1267 
1268 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
1269 	    ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1270 
1271 	mtx_lock(&vm->rendezvous_mtx);
1272 	while (vm->rendezvous_func != NULL) {
1273 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1274 		CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1275 
1276 		if (vcpuid != -1 &&
1277 		    CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1278 		    !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1279 			VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1280 			(*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1281 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1282 		}
1283 		if (CPU_CMP(&vm->rendezvous_req_cpus,
1284 		    &vm->rendezvous_done_cpus) == 0) {
1285 			VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1286 			vm_set_rendezvous_func(vm, NULL);
1287 			wakeup(&vm->rendezvous_func);
1288 			break;
1289 		}
1290 		RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1291 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1292 		    "vmrndv", 0);
1293 	}
1294 	mtx_unlock(&vm->rendezvous_mtx);
1295 }
1296 
1297 /*
1298  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1299  */
1300 static int
1301 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1302 {
1303 	struct vcpu *vcpu;
1304 	const char *wmesg;
1305 	int t, vcpu_halted, vm_halted;
1306 
1307 	KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1308 
1309 	vcpu = &vm->vcpu[vcpuid];
1310 	vcpu_halted = 0;
1311 	vm_halted = 0;
1312 
1313 	vcpu_lock(vcpu);
1314 	while (1) {
1315 		/*
1316 		 * Do a final check for pending NMI or interrupts before
1317 		 * really putting this thread to sleep. Also check for
1318 		 * software events that would cause this vcpu to wakeup.
1319 		 *
1320 		 * These interrupts/events could have happened after the
1321 		 * vcpu returned from VMRUN() and before it acquired the
1322 		 * vcpu lock above.
1323 		 */
1324 		if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle)
1325 			break;
1326 		if (vm_nmi_pending(vm, vcpuid))
1327 			break;
1328 		if (!intr_disabled) {
1329 			if (vm_extint_pending(vm, vcpuid) ||
1330 			    vlapic_pending_intr(vcpu->vlapic, NULL)) {
1331 				break;
1332 			}
1333 		}
1334 
1335 		/* Don't go to sleep if the vcpu thread needs to yield */
1336 		if (vcpu_should_yield(vm, vcpuid))
1337 			break;
1338 
1339 		if (vcpu_debugged(vm, vcpuid))
1340 			break;
1341 
1342 		/*
1343 		 * Some Linux guests implement "halt" by having all vcpus
1344 		 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1345 		 * track of the vcpus that have entered this state. When all
1346 		 * vcpus enter the halted state the virtual machine is halted.
1347 		 */
1348 		if (intr_disabled) {
1349 			wmesg = "vmhalt";
1350 			VCPU_CTR0(vm, vcpuid, "Halted");
1351 			if (!vcpu_halted && halt_detection_enabled) {
1352 				vcpu_halted = 1;
1353 				CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1354 			}
1355 			if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1356 				vm_halted = 1;
1357 				break;
1358 			}
1359 		} else {
1360 			wmesg = "vmidle";
1361 		}
1362 
1363 		t = ticks;
1364 		vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1365 		/*
1366 		 * XXX msleep_spin() cannot be interrupted by signals so
1367 		 * wake up periodically to check pending signals.
1368 		 */
1369 		msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1370 		vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1371 		vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1372 	}
1373 
1374 	if (vcpu_halted)
1375 		CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1376 
1377 	vcpu_unlock(vcpu);
1378 
1379 	if (vm_halted)
1380 		vm_suspend(vm, VM_SUSPEND_HALT);
1381 
1382 	return (0);
1383 }
1384 
1385 static int
1386 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1387 {
1388 	int rv, ftype;
1389 	struct vm_map *map;
1390 	struct vcpu *vcpu;
1391 	struct vm_exit *vme;
1392 
1393 	vcpu = &vm->vcpu[vcpuid];
1394 	vme = &vcpu->exitinfo;
1395 
1396 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1397 	    __func__, vme->inst_length));
1398 
1399 	ftype = vme->u.paging.fault_type;
1400 	KASSERT(ftype == VM_PROT_READ ||
1401 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1402 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1403 
1404 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1405 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1406 		    vme->u.paging.gpa, ftype);
1407 		if (rv == 0) {
1408 			VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1409 			    ftype == VM_PROT_READ ? "accessed" : "dirty",
1410 			    vme->u.paging.gpa);
1411 			goto done;
1412 		}
1413 	}
1414 
1415 	map = &vm->vmspace->vm_map;
1416 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
1417 
1418 	VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1419 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1420 
1421 	if (rv != KERN_SUCCESS)
1422 		return (EFAULT);
1423 done:
1424 	return (0);
1425 }
1426 
1427 static int
1428 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1429 {
1430 	struct vie *vie;
1431 	struct vcpu *vcpu;
1432 	struct vm_exit *vme;
1433 	uint64_t gla, gpa, cs_base;
1434 	struct vm_guest_paging *paging;
1435 	mem_region_read_t mread;
1436 	mem_region_write_t mwrite;
1437 	enum vm_cpu_mode cpu_mode;
1438 	int cs_d, error, fault;
1439 
1440 	vcpu = &vm->vcpu[vcpuid];
1441 	vme = &vcpu->exitinfo;
1442 
1443 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1444 	    __func__, vme->inst_length));
1445 
1446 	gla = vme->u.inst_emul.gla;
1447 	gpa = vme->u.inst_emul.gpa;
1448 	cs_base = vme->u.inst_emul.cs_base;
1449 	cs_d = vme->u.inst_emul.cs_d;
1450 	vie = &vme->u.inst_emul.vie;
1451 	paging = &vme->u.inst_emul.paging;
1452 	cpu_mode = paging->cpu_mode;
1453 
1454 	VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1455 
1456 	/* Fetch, decode and emulate the faulting instruction */
1457 	if (vie->num_valid == 0) {
1458 		error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip +
1459 		    cs_base, VIE_INST_SIZE, vie, &fault);
1460 	} else {
1461 		/*
1462 		 * The instruction bytes have already been copied into 'vie'
1463 		 */
1464 		error = fault = 0;
1465 	}
1466 	if (error || fault)
1467 		return (error);
1468 
1469 	if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) {
1470 		VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx",
1471 		    vme->rip + cs_base);
1472 		*retu = true;	    /* dump instruction bytes in userspace */
1473 		return (0);
1474 	}
1475 
1476 	/*
1477 	 * Update 'nextrip' based on the length of the emulated instruction.
1478 	 */
1479 	vme->inst_length = vie->num_processed;
1480 	vcpu->nextrip += vie->num_processed;
1481 	VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction "
1482 	    "decoding", vcpu->nextrip);
1483 
1484 	/* return to userland unless this is an in-kernel emulated device */
1485 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1486 		mread = lapic_mmio_read;
1487 		mwrite = lapic_mmio_write;
1488 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1489 		mread = vioapic_mmio_read;
1490 		mwrite = vioapic_mmio_write;
1491 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1492 		mread = vhpet_mmio_read;
1493 		mwrite = vhpet_mmio_write;
1494 	} else {
1495 		*retu = true;
1496 		return (0);
1497 	}
1498 
1499 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1500 	    mread, mwrite, retu);
1501 
1502 	return (error);
1503 }
1504 
1505 static int
1506 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1507 {
1508 	int i, done;
1509 	struct vcpu *vcpu;
1510 
1511 	done = 0;
1512 	vcpu = &vm->vcpu[vcpuid];
1513 
1514 	CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1515 
1516 	/*
1517 	 * Wait until all 'active_cpus' have suspended themselves.
1518 	 *
1519 	 * Since a VM may be suspended at any time including when one or
1520 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1521 	 * handler while we are waiting to prevent a deadlock.
1522 	 */
1523 	vcpu_lock(vcpu);
1524 	while (1) {
1525 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1526 			VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1527 			break;
1528 		}
1529 
1530 		if (vm->rendezvous_func == NULL) {
1531 			VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1532 			vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1533 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1534 			vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1535 		} else {
1536 			VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1537 			vcpu_unlock(vcpu);
1538 			vm_handle_rendezvous(vm, vcpuid);
1539 			vcpu_lock(vcpu);
1540 		}
1541 	}
1542 	vcpu_unlock(vcpu);
1543 
1544 	/*
1545 	 * Wakeup the other sleeping vcpus and return to userspace.
1546 	 */
1547 	for (i = 0; i < vm->maxcpus; i++) {
1548 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1549 			vcpu_notify_event(vm, i, false);
1550 		}
1551 	}
1552 
1553 	*retu = true;
1554 	return (0);
1555 }
1556 
1557 static int
1558 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu)
1559 {
1560 	struct vcpu *vcpu = &vm->vcpu[vcpuid];
1561 
1562 	vcpu_lock(vcpu);
1563 	KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle));
1564 	vcpu->reqidle = 0;
1565 	vcpu_unlock(vcpu);
1566 	*retu = true;
1567 	return (0);
1568 }
1569 
1570 int
1571 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1572 {
1573 	int i;
1574 
1575 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1576 		return (EINVAL);
1577 
1578 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1579 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1580 		    vm->suspend, how);
1581 		return (EALREADY);
1582 	}
1583 
1584 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1585 
1586 	/*
1587 	 * Notify all active vcpus that they are now suspended.
1588 	 */
1589 	for (i = 0; i < vm->maxcpus; i++) {
1590 		if (CPU_ISSET(i, &vm->active_cpus))
1591 			vcpu_notify_event(vm, i, false);
1592 	}
1593 
1594 	return (0);
1595 }
1596 
1597 void
1598 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1599 {
1600 	struct vm_exit *vmexit;
1601 
1602 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1603 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1604 
1605 	vmexit = vm_exitinfo(vm, vcpuid);
1606 	vmexit->rip = rip;
1607 	vmexit->inst_length = 0;
1608 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1609 	vmexit->u.suspended.how = vm->suspend;
1610 }
1611 
1612 void
1613 vm_exit_debug(struct vm *vm, int vcpuid, uint64_t rip)
1614 {
1615 	struct vm_exit *vmexit;
1616 
1617 	vmexit = vm_exitinfo(vm, vcpuid);
1618 	vmexit->rip = rip;
1619 	vmexit->inst_length = 0;
1620 	vmexit->exitcode = VM_EXITCODE_DEBUG;
1621 }
1622 
1623 void
1624 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1625 {
1626 	struct vm_exit *vmexit;
1627 
1628 	KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1629 
1630 	vmexit = vm_exitinfo(vm, vcpuid);
1631 	vmexit->rip = rip;
1632 	vmexit->inst_length = 0;
1633 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1634 	vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1635 }
1636 
1637 void
1638 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip)
1639 {
1640 	struct vm_exit *vmexit;
1641 
1642 	vmexit = vm_exitinfo(vm, vcpuid);
1643 	vmexit->rip = rip;
1644 	vmexit->inst_length = 0;
1645 	vmexit->exitcode = VM_EXITCODE_REQIDLE;
1646 	vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1);
1647 }
1648 
1649 void
1650 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1651 {
1652 	struct vm_exit *vmexit;
1653 
1654 	vmexit = vm_exitinfo(vm, vcpuid);
1655 	vmexit->rip = rip;
1656 	vmexit->inst_length = 0;
1657 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1658 	vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1659 }
1660 
1661 int
1662 vm_run(struct vm *vm, struct vm_run *vmrun)
1663 {
1664 	struct vm_eventinfo evinfo;
1665 	int error, vcpuid;
1666 	struct vcpu *vcpu;
1667 	struct pcb *pcb;
1668 	uint64_t tscval;
1669 	struct vm_exit *vme;
1670 	bool retu, intr_disabled;
1671 	pmap_t pmap;
1672 
1673 	vcpuid = vmrun->cpuid;
1674 
1675 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1676 		return (EINVAL);
1677 
1678 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1679 		return (EINVAL);
1680 
1681 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1682 		return (EINVAL);
1683 
1684 	pmap = vmspace_pmap(vm->vmspace);
1685 	vcpu = &vm->vcpu[vcpuid];
1686 	vme = &vcpu->exitinfo;
1687 	evinfo.rptr = &vm->rendezvous_func;
1688 	evinfo.sptr = &vm->suspend;
1689 	evinfo.iptr = &vcpu->reqidle;
1690 restart:
1691 	critical_enter();
1692 
1693 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1694 	    ("vm_run: absurd pm_active"));
1695 
1696 	tscval = rdtsc();
1697 
1698 	pcb = PCPU_GET(curpcb);
1699 	set_pcb_flags(pcb, PCB_FULL_IRET);
1700 
1701 	restore_guest_fpustate(vcpu);
1702 
1703 	vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1704 	error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo);
1705 	vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1706 
1707 	save_guest_fpustate(vcpu);
1708 
1709 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1710 
1711 	critical_exit();
1712 
1713 	if (error == 0) {
1714 		retu = false;
1715 		vcpu->nextrip = vme->rip + vme->inst_length;
1716 		switch (vme->exitcode) {
1717 		case VM_EXITCODE_REQIDLE:
1718 			error = vm_handle_reqidle(vm, vcpuid, &retu);
1719 			break;
1720 		case VM_EXITCODE_SUSPENDED:
1721 			error = vm_handle_suspend(vm, vcpuid, &retu);
1722 			break;
1723 		case VM_EXITCODE_IOAPIC_EOI:
1724 			vioapic_process_eoi(vm, vcpuid,
1725 			    vme->u.ioapic_eoi.vector);
1726 			break;
1727 		case VM_EXITCODE_RENDEZVOUS:
1728 			vm_handle_rendezvous(vm, vcpuid);
1729 			error = 0;
1730 			break;
1731 		case VM_EXITCODE_HLT:
1732 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1733 			error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1734 			break;
1735 		case VM_EXITCODE_PAGING:
1736 			error = vm_handle_paging(vm, vcpuid, &retu);
1737 			break;
1738 		case VM_EXITCODE_INST_EMUL:
1739 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1740 			break;
1741 		case VM_EXITCODE_INOUT:
1742 		case VM_EXITCODE_INOUT_STR:
1743 			error = vm_handle_inout(vm, vcpuid, vme, &retu);
1744 			break;
1745 		case VM_EXITCODE_MONITOR:
1746 		case VM_EXITCODE_MWAIT:
1747 		case VM_EXITCODE_VMINSN:
1748 			vm_inject_ud(vm, vcpuid);
1749 			break;
1750 		default:
1751 			retu = true;	/* handled in userland */
1752 			break;
1753 		}
1754 	}
1755 
1756 	if (error == 0 && retu == false)
1757 		goto restart;
1758 
1759 	VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode);
1760 
1761 	/* copy the exit information */
1762 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1763 	return (error);
1764 }
1765 
1766 int
1767 vm_restart_instruction(void *arg, int vcpuid)
1768 {
1769 	struct vm *vm;
1770 	struct vcpu *vcpu;
1771 	enum vcpu_state state;
1772 	uint64_t rip;
1773 	int error;
1774 
1775 	vm = arg;
1776 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1777 		return (EINVAL);
1778 
1779 	vcpu = &vm->vcpu[vcpuid];
1780 	state = vcpu_get_state(vm, vcpuid, NULL);
1781 	if (state == VCPU_RUNNING) {
1782 		/*
1783 		 * When a vcpu is "running" the next instruction is determined
1784 		 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
1785 		 * Thus setting 'inst_length' to zero will cause the current
1786 		 * instruction to be restarted.
1787 		 */
1788 		vcpu->exitinfo.inst_length = 0;
1789 		VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by "
1790 		    "setting inst_length to zero", vcpu->exitinfo.rip);
1791 	} else if (state == VCPU_FROZEN) {
1792 		/*
1793 		 * When a vcpu is "frozen" it is outside the critical section
1794 		 * around VMRUN() and 'nextrip' points to the next instruction.
1795 		 * Thus instruction restart is achieved by setting 'nextrip'
1796 		 * to the vcpu's %rip.
1797 		 */
1798 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip);
1799 		KASSERT(!error, ("%s: error %d getting rip", __func__, error));
1800 		VCPU_CTR2(vm, vcpuid, "restarting instruction by updating "
1801 		    "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
1802 		vcpu->nextrip = rip;
1803 	} else {
1804 		panic("%s: invalid state %d", __func__, state);
1805 	}
1806 	return (0);
1807 }
1808 
1809 int
1810 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1811 {
1812 	struct vcpu *vcpu;
1813 	int type, vector;
1814 
1815 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1816 		return (EINVAL);
1817 
1818 	vcpu = &vm->vcpu[vcpuid];
1819 
1820 	if (info & VM_INTINFO_VALID) {
1821 		type = info & VM_INTINFO_TYPE;
1822 		vector = info & 0xff;
1823 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1824 			return (EINVAL);
1825 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1826 			return (EINVAL);
1827 		if (info & VM_INTINFO_RSVD)
1828 			return (EINVAL);
1829 	} else {
1830 		info = 0;
1831 	}
1832 	VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1833 	vcpu->exitintinfo = info;
1834 	return (0);
1835 }
1836 
1837 enum exc_class {
1838 	EXC_BENIGN,
1839 	EXC_CONTRIBUTORY,
1840 	EXC_PAGEFAULT
1841 };
1842 
1843 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
1844 
1845 static enum exc_class
1846 exception_class(uint64_t info)
1847 {
1848 	int type, vector;
1849 
1850 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1851 	type = info & VM_INTINFO_TYPE;
1852 	vector = info & 0xff;
1853 
1854 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1855 	switch (type) {
1856 	case VM_INTINFO_HWINTR:
1857 	case VM_INTINFO_SWINTR:
1858 	case VM_INTINFO_NMI:
1859 		return (EXC_BENIGN);
1860 	default:
1861 		/*
1862 		 * Hardware exception.
1863 		 *
1864 		 * SVM and VT-x use identical type values to represent NMI,
1865 		 * hardware interrupt and software interrupt.
1866 		 *
1867 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1868 		 * for exceptions except #BP and #OF. #BP and #OF use a type
1869 		 * value of '5' or '6'. Therefore we don't check for explicit
1870 		 * values of 'type' to classify 'intinfo' into a hardware
1871 		 * exception.
1872 		 */
1873 		break;
1874 	}
1875 
1876 	switch (vector) {
1877 	case IDT_PF:
1878 	case IDT_VE:
1879 		return (EXC_PAGEFAULT);
1880 	case IDT_DE:
1881 	case IDT_TS:
1882 	case IDT_NP:
1883 	case IDT_SS:
1884 	case IDT_GP:
1885 		return (EXC_CONTRIBUTORY);
1886 	default:
1887 		return (EXC_BENIGN);
1888 	}
1889 }
1890 
1891 static int
1892 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1893     uint64_t *retinfo)
1894 {
1895 	enum exc_class exc1, exc2;
1896 	int type1, vector1;
1897 
1898 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1899 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1900 
1901 	/*
1902 	 * If an exception occurs while attempting to call the double-fault
1903 	 * handler the processor enters shutdown mode (aka triple fault).
1904 	 */
1905 	type1 = info1 & VM_INTINFO_TYPE;
1906 	vector1 = info1 & 0xff;
1907 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1908 		VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1909 		    info1, info2);
1910 		vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1911 		*retinfo = 0;
1912 		return (0);
1913 	}
1914 
1915 	/*
1916 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1917 	 */
1918 	exc1 = exception_class(info1);
1919 	exc2 = exception_class(info2);
1920 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1921 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1922 		/* Convert nested fault into a double fault. */
1923 		*retinfo = IDT_DF;
1924 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1925 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
1926 	} else {
1927 		/* Handle exceptions serially */
1928 		*retinfo = info2;
1929 	}
1930 	return (1);
1931 }
1932 
1933 static uint64_t
1934 vcpu_exception_intinfo(struct vcpu *vcpu)
1935 {
1936 	uint64_t info = 0;
1937 
1938 	if (vcpu->exception_pending) {
1939 		info = vcpu->exc_vector & 0xff;
1940 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1941 		if (vcpu->exc_errcode_valid) {
1942 			info |= VM_INTINFO_DEL_ERRCODE;
1943 			info |= (uint64_t)vcpu->exc_errcode << 32;
1944 		}
1945 	}
1946 	return (info);
1947 }
1948 
1949 int
1950 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1951 {
1952 	struct vcpu *vcpu;
1953 	uint64_t info1, info2;
1954 	int valid;
1955 
1956 	KASSERT(vcpuid >= 0 &&
1957 	    vcpuid < vm->maxcpus, ("invalid vcpu %d", vcpuid));
1958 
1959 	vcpu = &vm->vcpu[vcpuid];
1960 
1961 	info1 = vcpu->exitintinfo;
1962 	vcpu->exitintinfo = 0;
1963 
1964 	info2 = 0;
1965 	if (vcpu->exception_pending) {
1966 		info2 = vcpu_exception_intinfo(vcpu);
1967 		vcpu->exception_pending = 0;
1968 		VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1969 		    vcpu->exc_vector, info2);
1970 	}
1971 
1972 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
1973 		valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
1974 	} else if (info1 & VM_INTINFO_VALID) {
1975 		*retinfo = info1;
1976 		valid = 1;
1977 	} else if (info2 & VM_INTINFO_VALID) {
1978 		*retinfo = info2;
1979 		valid = 1;
1980 	} else {
1981 		valid = 0;
1982 	}
1983 
1984 	if (valid) {
1985 		VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
1986 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
1987 	}
1988 
1989 	return (valid);
1990 }
1991 
1992 int
1993 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
1994 {
1995 	struct vcpu *vcpu;
1996 
1997 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1998 		return (EINVAL);
1999 
2000 	vcpu = &vm->vcpu[vcpuid];
2001 	*info1 = vcpu->exitintinfo;
2002 	*info2 = vcpu_exception_intinfo(vcpu);
2003 	return (0);
2004 }
2005 
2006 int
2007 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid,
2008     uint32_t errcode, int restart_instruction)
2009 {
2010 	struct vcpu *vcpu;
2011 	uint64_t regval;
2012 	int error;
2013 
2014 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2015 		return (EINVAL);
2016 
2017 	if (vector < 0 || vector >= 32)
2018 		return (EINVAL);
2019 
2020 	/*
2021 	 * A double fault exception should never be injected directly into
2022 	 * the guest. It is a derived exception that results from specific
2023 	 * combinations of nested faults.
2024 	 */
2025 	if (vector == IDT_DF)
2026 		return (EINVAL);
2027 
2028 	vcpu = &vm->vcpu[vcpuid];
2029 
2030 	if (vcpu->exception_pending) {
2031 		VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
2032 		    "pending exception %d", vector, vcpu->exc_vector);
2033 		return (EBUSY);
2034 	}
2035 
2036 	if (errcode_valid) {
2037 		/*
2038 		 * Exceptions don't deliver an error code in real mode.
2039 		 */
2040 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, &regval);
2041 		KASSERT(!error, ("%s: error %d getting CR0", __func__, error));
2042 		if (!(regval & CR0_PE))
2043 			errcode_valid = 0;
2044 	}
2045 
2046 	/*
2047 	 * From section 26.6.1 "Interruptibility State" in Intel SDM:
2048 	 *
2049 	 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
2050 	 * one instruction or incurs an exception.
2051 	 */
2052 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
2053 	KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
2054 	    __func__, error));
2055 
2056 	if (restart_instruction)
2057 		vm_restart_instruction(vm, vcpuid);
2058 
2059 	vcpu->exception_pending = 1;
2060 	vcpu->exc_vector = vector;
2061 	vcpu->exc_errcode = errcode;
2062 	vcpu->exc_errcode_valid = errcode_valid;
2063 	VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector);
2064 	return (0);
2065 }
2066 
2067 void
2068 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
2069     int errcode)
2070 {
2071 	struct vm *vm;
2072 	int error, restart_instruction;
2073 
2074 	vm = vmarg;
2075 	restart_instruction = 1;
2076 
2077 	error = vm_inject_exception(vm, vcpuid, vector, errcode_valid,
2078 	    errcode, restart_instruction);
2079 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
2080 }
2081 
2082 void
2083 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
2084 {
2085 	struct vm *vm;
2086 	int error;
2087 
2088 	vm = vmarg;
2089 	VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
2090 	    error_code, cr2);
2091 
2092 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
2093 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
2094 
2095 	vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
2096 }
2097 
2098 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
2099 
2100 int
2101 vm_inject_nmi(struct vm *vm, int vcpuid)
2102 {
2103 	struct vcpu *vcpu;
2104 
2105 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2106 		return (EINVAL);
2107 
2108 	vcpu = &vm->vcpu[vcpuid];
2109 
2110 	vcpu->nmi_pending = 1;
2111 	vcpu_notify_event(vm, vcpuid, false);
2112 	return (0);
2113 }
2114 
2115 int
2116 vm_nmi_pending(struct vm *vm, int vcpuid)
2117 {
2118 	struct vcpu *vcpu;
2119 
2120 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2121 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2122 
2123 	vcpu = &vm->vcpu[vcpuid];
2124 
2125 	return (vcpu->nmi_pending);
2126 }
2127 
2128 void
2129 vm_nmi_clear(struct vm *vm, int vcpuid)
2130 {
2131 	struct vcpu *vcpu;
2132 
2133 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2134 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2135 
2136 	vcpu = &vm->vcpu[vcpuid];
2137 
2138 	if (vcpu->nmi_pending == 0)
2139 		panic("vm_nmi_clear: inconsistent nmi_pending state");
2140 
2141 	vcpu->nmi_pending = 0;
2142 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
2143 }
2144 
2145 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
2146 
2147 int
2148 vm_inject_extint(struct vm *vm, int vcpuid)
2149 {
2150 	struct vcpu *vcpu;
2151 
2152 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2153 		return (EINVAL);
2154 
2155 	vcpu = &vm->vcpu[vcpuid];
2156 
2157 	vcpu->extint_pending = 1;
2158 	vcpu_notify_event(vm, vcpuid, false);
2159 	return (0);
2160 }
2161 
2162 int
2163 vm_extint_pending(struct vm *vm, int vcpuid)
2164 {
2165 	struct vcpu *vcpu;
2166 
2167 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2168 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2169 
2170 	vcpu = &vm->vcpu[vcpuid];
2171 
2172 	return (vcpu->extint_pending);
2173 }
2174 
2175 void
2176 vm_extint_clear(struct vm *vm, int vcpuid)
2177 {
2178 	struct vcpu *vcpu;
2179 
2180 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2181 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2182 
2183 	vcpu = &vm->vcpu[vcpuid];
2184 
2185 	if (vcpu->extint_pending == 0)
2186 		panic("vm_extint_clear: inconsistent extint_pending state");
2187 
2188 	vcpu->extint_pending = 0;
2189 	vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
2190 }
2191 
2192 int
2193 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
2194 {
2195 	if (vcpu < 0 || vcpu >= vm->maxcpus)
2196 		return (EINVAL);
2197 
2198 	if (type < 0 || type >= VM_CAP_MAX)
2199 		return (EINVAL);
2200 
2201 	return (VMGETCAP(vm->cookie, vcpu, type, retval));
2202 }
2203 
2204 int
2205 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
2206 {
2207 	if (vcpu < 0 || vcpu >= vm->maxcpus)
2208 		return (EINVAL);
2209 
2210 	if (type < 0 || type >= VM_CAP_MAX)
2211 		return (EINVAL);
2212 
2213 	return (VMSETCAP(vm->cookie, vcpu, type, val));
2214 }
2215 
2216 struct vlapic *
2217 vm_lapic(struct vm *vm, int cpu)
2218 {
2219 	return (vm->vcpu[cpu].vlapic);
2220 }
2221 
2222 struct vioapic *
2223 vm_ioapic(struct vm *vm)
2224 {
2225 
2226 	return (vm->vioapic);
2227 }
2228 
2229 struct vhpet *
2230 vm_hpet(struct vm *vm)
2231 {
2232 
2233 	return (vm->vhpet);
2234 }
2235 
2236 boolean_t
2237 vmm_is_pptdev(int bus, int slot, int func)
2238 {
2239 	int found, i, n;
2240 	int b, s, f;
2241 	char *val, *cp, *cp2;
2242 
2243 	/*
2244 	 * XXX
2245 	 * The length of an environment variable is limited to 128 bytes which
2246 	 * puts an upper limit on the number of passthru devices that may be
2247 	 * specified using a single environment variable.
2248 	 *
2249 	 * Work around this by scanning multiple environment variable
2250 	 * names instead of a single one - yuck!
2251 	 */
2252 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2253 
2254 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2255 	found = 0;
2256 	for (i = 0; names[i] != NULL && !found; i++) {
2257 		cp = val = kern_getenv(names[i]);
2258 		while (cp != NULL && *cp != '\0') {
2259 			if ((cp2 = strchr(cp, ' ')) != NULL)
2260 				*cp2 = '\0';
2261 
2262 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2263 			if (n == 3 && bus == b && slot == s && func == f) {
2264 				found = 1;
2265 				break;
2266 			}
2267 
2268 			if (cp2 != NULL)
2269 				*cp2++ = ' ';
2270 
2271 			cp = cp2;
2272 		}
2273 		freeenv(val);
2274 	}
2275 	return (found);
2276 }
2277 
2278 void *
2279 vm_iommu_domain(struct vm *vm)
2280 {
2281 
2282 	return (vm->iommu);
2283 }
2284 
2285 int
2286 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
2287     bool from_idle)
2288 {
2289 	int error;
2290 	struct vcpu *vcpu;
2291 
2292 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2293 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
2294 
2295 	vcpu = &vm->vcpu[vcpuid];
2296 
2297 	vcpu_lock(vcpu);
2298 	error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle);
2299 	vcpu_unlock(vcpu);
2300 
2301 	return (error);
2302 }
2303 
2304 enum vcpu_state
2305 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2306 {
2307 	struct vcpu *vcpu;
2308 	enum vcpu_state state;
2309 
2310 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2311 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2312 
2313 	vcpu = &vm->vcpu[vcpuid];
2314 
2315 	vcpu_lock(vcpu);
2316 	state = vcpu->state;
2317 	if (hostcpu != NULL)
2318 		*hostcpu = vcpu->hostcpu;
2319 	vcpu_unlock(vcpu);
2320 
2321 	return (state);
2322 }
2323 
2324 int
2325 vm_activate_cpu(struct vm *vm, int vcpuid)
2326 {
2327 
2328 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2329 		return (EINVAL);
2330 
2331 	if (CPU_ISSET(vcpuid, &vm->active_cpus))
2332 		return (EBUSY);
2333 
2334 	VCPU_CTR0(vm, vcpuid, "activated");
2335 	CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2336 	return (0);
2337 }
2338 
2339 int
2340 vm_suspend_cpu(struct vm *vm, int vcpuid)
2341 {
2342 	int i;
2343 
2344 	if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2345 		return (EINVAL);
2346 
2347 	if (vcpuid == -1) {
2348 		vm->debug_cpus = vm->active_cpus;
2349 		for (i = 0; i < vm->maxcpus; i++) {
2350 			if (CPU_ISSET(i, &vm->active_cpus))
2351 				vcpu_notify_event(vm, i, false);
2352 		}
2353 	} else {
2354 		if (!CPU_ISSET(vcpuid, &vm->active_cpus))
2355 			return (EINVAL);
2356 
2357 		CPU_SET_ATOMIC(vcpuid, &vm->debug_cpus);
2358 		vcpu_notify_event(vm, vcpuid, false);
2359 	}
2360 	return (0);
2361 }
2362 
2363 int
2364 vm_resume_cpu(struct vm *vm, int vcpuid)
2365 {
2366 
2367 	if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2368 		return (EINVAL);
2369 
2370 	if (vcpuid == -1) {
2371 		CPU_ZERO(&vm->debug_cpus);
2372 	} else {
2373 		if (!CPU_ISSET(vcpuid, &vm->debug_cpus))
2374 			return (EINVAL);
2375 
2376 		CPU_CLR_ATOMIC(vcpuid, &vm->debug_cpus);
2377 	}
2378 	return (0);
2379 }
2380 
2381 int
2382 vcpu_debugged(struct vm *vm, int vcpuid)
2383 {
2384 
2385 	return (CPU_ISSET(vcpuid, &vm->debug_cpus));
2386 }
2387 
2388 cpuset_t
2389 vm_active_cpus(struct vm *vm)
2390 {
2391 
2392 	return (vm->active_cpus);
2393 }
2394 
2395 cpuset_t
2396 vm_debug_cpus(struct vm *vm)
2397 {
2398 
2399 	return (vm->debug_cpus);
2400 }
2401 
2402 cpuset_t
2403 vm_suspended_cpus(struct vm *vm)
2404 {
2405 
2406 	return (vm->suspended_cpus);
2407 }
2408 
2409 void *
2410 vcpu_stats(struct vm *vm, int vcpuid)
2411 {
2412 
2413 	return (vm->vcpu[vcpuid].stats);
2414 }
2415 
2416 int
2417 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2418 {
2419 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2420 		return (EINVAL);
2421 
2422 	*state = vm->vcpu[vcpuid].x2apic_state;
2423 
2424 	return (0);
2425 }
2426 
2427 int
2428 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2429 {
2430 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2431 		return (EINVAL);
2432 
2433 	if (state >= X2APIC_STATE_LAST)
2434 		return (EINVAL);
2435 
2436 	vm->vcpu[vcpuid].x2apic_state = state;
2437 
2438 	vlapic_set_x2apic_state(vm, vcpuid, state);
2439 
2440 	return (0);
2441 }
2442 
2443 /*
2444  * This function is called to ensure that a vcpu "sees" a pending event
2445  * as soon as possible:
2446  * - If the vcpu thread is sleeping then it is woken up.
2447  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2448  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2449  */
2450 static void
2451 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
2452 {
2453 	int hostcpu;
2454 
2455 	hostcpu = vcpu->hostcpu;
2456 	if (vcpu->state == VCPU_RUNNING) {
2457 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2458 		if (hostcpu != curcpu) {
2459 			if (lapic_intr) {
2460 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2461 				    vmm_ipinum);
2462 			} else {
2463 				ipi_cpu(hostcpu, vmm_ipinum);
2464 			}
2465 		} else {
2466 			/*
2467 			 * If the 'vcpu' is running on 'curcpu' then it must
2468 			 * be sending a notification to itself (e.g. SELF_IPI).
2469 			 * The pending event will be picked up when the vcpu
2470 			 * transitions back to guest context.
2471 			 */
2472 		}
2473 	} else {
2474 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2475 		    "with hostcpu %d", vcpu->state, hostcpu));
2476 		if (vcpu->state == VCPU_SLEEPING)
2477 			wakeup_one(vcpu);
2478 	}
2479 }
2480 
2481 void
2482 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2483 {
2484 	struct vcpu *vcpu = &vm->vcpu[vcpuid];
2485 
2486 	vcpu_lock(vcpu);
2487 	vcpu_notify_event_locked(vcpu, lapic_intr);
2488 	vcpu_unlock(vcpu);
2489 }
2490 
2491 struct vmspace *
2492 vm_get_vmspace(struct vm *vm)
2493 {
2494 
2495 	return (vm->vmspace);
2496 }
2497 
2498 int
2499 vm_apicid2vcpuid(struct vm *vm, int apicid)
2500 {
2501 	/*
2502 	 * XXX apic id is assumed to be numerically identical to vcpu id
2503 	 */
2504 	return (apicid);
2505 }
2506 
2507 void
2508 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2509     vm_rendezvous_func_t func, void *arg)
2510 {
2511 	int i;
2512 
2513 	/*
2514 	 * Enforce that this function is called without any locks
2515 	 */
2516 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2517 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
2518 	    ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2519 
2520 restart:
2521 	mtx_lock(&vm->rendezvous_mtx);
2522 	if (vm->rendezvous_func != NULL) {
2523 		/*
2524 		 * If a rendezvous is already in progress then we need to
2525 		 * call the rendezvous handler in case this 'vcpuid' is one
2526 		 * of the targets of the rendezvous.
2527 		 */
2528 		RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2529 		mtx_unlock(&vm->rendezvous_mtx);
2530 		vm_handle_rendezvous(vm, vcpuid);
2531 		goto restart;
2532 	}
2533 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2534 	    "rendezvous is still in progress"));
2535 
2536 	RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2537 	vm->rendezvous_req_cpus = dest;
2538 	CPU_ZERO(&vm->rendezvous_done_cpus);
2539 	vm->rendezvous_arg = arg;
2540 	vm_set_rendezvous_func(vm, func);
2541 	mtx_unlock(&vm->rendezvous_mtx);
2542 
2543 	/*
2544 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2545 	 * vcpus so they handle the rendezvous as soon as possible.
2546 	 */
2547 	for (i = 0; i < vm->maxcpus; i++) {
2548 		if (CPU_ISSET(i, &dest))
2549 			vcpu_notify_event(vm, i, false);
2550 	}
2551 
2552 	vm_handle_rendezvous(vm, vcpuid);
2553 }
2554 
2555 struct vatpic *
2556 vm_atpic(struct vm *vm)
2557 {
2558 	return (vm->vatpic);
2559 }
2560 
2561 struct vatpit *
2562 vm_atpit(struct vm *vm)
2563 {
2564 	return (vm->vatpit);
2565 }
2566 
2567 struct vpmtmr *
2568 vm_pmtmr(struct vm *vm)
2569 {
2570 
2571 	return (vm->vpmtmr);
2572 }
2573 
2574 struct vrtc *
2575 vm_rtc(struct vm *vm)
2576 {
2577 
2578 	return (vm->vrtc);
2579 }
2580 
2581 enum vm_reg_name
2582 vm_segment_name(int seg)
2583 {
2584 	static enum vm_reg_name seg_names[] = {
2585 		VM_REG_GUEST_ES,
2586 		VM_REG_GUEST_CS,
2587 		VM_REG_GUEST_SS,
2588 		VM_REG_GUEST_DS,
2589 		VM_REG_GUEST_FS,
2590 		VM_REG_GUEST_GS
2591 	};
2592 
2593 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2594 	    ("%s: invalid segment encoding %d", __func__, seg));
2595 	return (seg_names[seg]);
2596 }
2597 
2598 void
2599 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2600     int num_copyinfo)
2601 {
2602 	int idx;
2603 
2604 	for (idx = 0; idx < num_copyinfo; idx++) {
2605 		if (copyinfo[idx].cookie != NULL)
2606 			vm_gpa_release(copyinfo[idx].cookie);
2607 	}
2608 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2609 }
2610 
2611 int
2612 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2613     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2614     int num_copyinfo, int *fault)
2615 {
2616 	int error, idx, nused;
2617 	size_t n, off, remaining;
2618 	void *hva, *cookie;
2619 	uint64_t gpa;
2620 
2621 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2622 
2623 	nused = 0;
2624 	remaining = len;
2625 	while (remaining > 0) {
2626 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2627 		error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault);
2628 		if (error || *fault)
2629 			return (error);
2630 		off = gpa & PAGE_MASK;
2631 		n = min(remaining, PAGE_SIZE - off);
2632 		copyinfo[nused].gpa = gpa;
2633 		copyinfo[nused].len = n;
2634 		remaining -= n;
2635 		gla += n;
2636 		nused++;
2637 	}
2638 
2639 	for (idx = 0; idx < nused; idx++) {
2640 		hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa,
2641 		    copyinfo[idx].len, prot, &cookie);
2642 		if (hva == NULL)
2643 			break;
2644 		copyinfo[idx].hva = hva;
2645 		copyinfo[idx].cookie = cookie;
2646 	}
2647 
2648 	if (idx != nused) {
2649 		vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2650 		return (EFAULT);
2651 	} else {
2652 		*fault = 0;
2653 		return (0);
2654 	}
2655 }
2656 
2657 void
2658 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2659     size_t len)
2660 {
2661 	char *dst;
2662 	int idx;
2663 
2664 	dst = kaddr;
2665 	idx = 0;
2666 	while (len > 0) {
2667 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2668 		len -= copyinfo[idx].len;
2669 		dst += copyinfo[idx].len;
2670 		idx++;
2671 	}
2672 }
2673 
2674 void
2675 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2676     struct vm_copyinfo *copyinfo, size_t len)
2677 {
2678 	const char *src;
2679 	int idx;
2680 
2681 	src = kaddr;
2682 	idx = 0;
2683 	while (len > 0) {
2684 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2685 		len -= copyinfo[idx].len;
2686 		src += copyinfo[idx].len;
2687 		idx++;
2688 	}
2689 }
2690 
2691 /*
2692  * Return the amount of in-use and wired memory for the VM. Since
2693  * these are global stats, only return the values with for vCPU 0
2694  */
2695 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2696 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2697 
2698 static void
2699 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2700 {
2701 
2702 	if (vcpu == 0) {
2703 		vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2704 	       	    PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2705 	}
2706 }
2707 
2708 static void
2709 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2710 {
2711 
2712 	if (vcpu == 0) {
2713 		vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2714 	      	    PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2715 	}
2716 }
2717 
2718 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2719 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2720