xref: /freebsd/sys/amd64/vmm/vmm.c (revision 489ba2223676ec251ab1bfe2906d2a62959c8ce3)
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_param.h>
53 #include <vm/vm_extern.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
56 #include <vm/pmap.h>
57 #include <vm/vm_map.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 #include <x86/ifunc.h>
71 
72 #include <machine/vmm.h>
73 #include <machine/vmm_dev.h>
74 #include <machine/vmm_instruction_emul.h>
75 #include <machine/vmm_snapshot.h>
76 
77 #include "vmm_ioport.h"
78 #include "vmm_ktr.h"
79 #include "vmm_host.h"
80 #include "vmm_mem.h"
81 #include "vmm_util.h"
82 #include "vatpic.h"
83 #include "vatpit.h"
84 #include "vhpet.h"
85 #include "vioapic.h"
86 #include "vlapic.h"
87 #include "vpmtmr.h"
88 #include "vrtc.h"
89 #include "vmm_stat.h"
90 #include "vmm_lapic.h"
91 
92 #include "io/ppt.h"
93 #include "io/iommu.h"
94 
95 struct vlapic;
96 
97 /*
98  * Initialization:
99  * (a) allocated when vcpu is created
100  * (i) initialized when vcpu is created and when it is reinitialized
101  * (o) initialized the first time the vcpu is created
102  * (x) initialized before use
103  */
104 struct vcpu {
105 	struct mtx 	mtx;		/* (o) protects 'state' and 'hostcpu' */
106 	enum vcpu_state	state;		/* (o) vcpu state */
107 	int		hostcpu;	/* (o) vcpu's host cpu */
108 	int		reqidle;	/* (i) request vcpu to idle */
109 	struct vlapic	*vlapic;	/* (i) APIC device model */
110 	enum x2apic_state x2apic_state;	/* (i) APIC mode */
111 	uint64_t	exitintinfo;	/* (i) events pending at VM exit */
112 	int		nmi_pending;	/* (i) NMI pending */
113 	int		extint_pending;	/* (i) INTR pending */
114 	int	exception_pending;	/* (i) exception pending */
115 	int	exc_vector;		/* (x) exception collateral */
116 	int	exc_errcode_valid;
117 	uint32_t exc_errcode;
118 	struct savefpu	*guestfpu;	/* (a,i) guest fpu state */
119 	uint64_t	guest_xcr0;	/* (i) guest %xcr0 register */
120 	void		*stats;		/* (a,i) statistics */
121 	struct vm_exit	exitinfo;	/* (x) exit reason and collateral */
122 	uint64_t	nextrip;	/* (x) next instruction to execute */
123 	uint64_t	tsc_offset;	/* (o) TSC offsetting */
124 };
125 
126 #define	vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
127 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
128 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
129 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
130 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
131 
132 struct mem_seg {
133 	size_t	len;
134 	bool	sysmem;
135 	struct vm_object *object;
136 };
137 #define	VM_MAX_MEMSEGS	4
138 
139 struct mem_map {
140 	vm_paddr_t	gpa;
141 	size_t		len;
142 	vm_ooffset_t	segoff;
143 	int		segid;
144 	int		prot;
145 	int		flags;
146 };
147 #define	VM_MAX_MEMMAPS	8
148 
149 /*
150  * Initialization:
151  * (o) initialized the first time the VM is created
152  * (i) initialized when VM is created and when it is reinitialized
153  * (x) initialized before use
154  */
155 struct vm {
156 	void		*cookie;		/* (i) cpu-specific data */
157 	void		*iommu;			/* (x) iommu-specific data */
158 	struct vhpet	*vhpet;			/* (i) virtual HPET */
159 	struct vioapic	*vioapic;		/* (i) virtual ioapic */
160 	struct vatpic	*vatpic;		/* (i) virtual atpic */
161 	struct vatpit	*vatpit;		/* (i) virtual atpit */
162 	struct vpmtmr	*vpmtmr;		/* (i) virtual ACPI PM timer */
163 	struct vrtc	*vrtc;			/* (o) virtual RTC */
164 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
165 	volatile cpuset_t debug_cpus;		/* (i) vcpus stopped for debug */
166 	int		suspend;		/* (i) stop VM execution */
167 	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
168 	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
169 	cpuset_t	rendezvous_req_cpus;	/* (x) rendezvous requested */
170 	cpuset_t	rendezvous_done_cpus;	/* (x) rendezvous finished */
171 	void		*rendezvous_arg;	/* (x) rendezvous func/arg */
172 	vm_rendezvous_func_t rendezvous_func;
173 	struct mtx	rendezvous_mtx;		/* (o) rendezvous lock */
174 	struct mem_map	mem_maps[VM_MAX_MEMMAPS]; /* (i) guest address space */
175 	struct mem_seg	mem_segs[VM_MAX_MEMSEGS]; /* (o) guest memory regions */
176 	struct vmspace	*vmspace;		/* (o) guest's address space */
177 	char		name[VM_MAX_NAMELEN+1];	/* (o) virtual machine name */
178 	struct vcpu	vcpu[VM_MAXCPU];	/* (i) guest vcpus */
179 	/* The following describe the vm cpu topology */
180 	uint16_t	sockets;		/* (o) num of sockets */
181 	uint16_t	cores;			/* (o) num of cores/socket */
182 	uint16_t	threads;		/* (o) num of threads/core */
183 	uint16_t	maxcpus;		/* (o) max pluggable cpus */
184 };
185 
186 static int vmm_initialized;
187 
188 static void	vmmops_panic(void);
189 
190 static void
191 vmmops_panic(void)
192 {
193 	panic("vmm_ops func called when !vmm_is_intel() && !vmm_is_svm()");
194 }
195 
196 #define	DEFINE_VMMOPS_IFUNC(ret_type, opname, args)			\
197     DEFINE_IFUNC(static, ret_type, vmmops_##opname, args)		\
198     {									\
199     	if (vmm_is_intel())						\
200     		return (vmm_ops_intel.opname);				\
201     	else if (vmm_is_svm())						\
202     		return (vmm_ops_amd.opname);				\
203     	else								\
204     		return ((ret_type (*)args)vmmops_panic);		\
205     }
206 
207 DEFINE_VMMOPS_IFUNC(int, modinit, (int ipinum))
208 DEFINE_VMMOPS_IFUNC(int, modcleanup, (void))
209 DEFINE_VMMOPS_IFUNC(void, modresume, (void))
210 DEFINE_VMMOPS_IFUNC(void *, init, (struct vm *vm, struct pmap *pmap))
211 DEFINE_VMMOPS_IFUNC(int, run, (void *vmi, int vcpu, register_t rip,
212     struct pmap *pmap, struct vm_eventinfo *info))
213 DEFINE_VMMOPS_IFUNC(void, cleanup, (void *vmi))
214 DEFINE_VMMOPS_IFUNC(int, getreg, (void *vmi, int vcpu, int num,
215     uint64_t *retval))
216 DEFINE_VMMOPS_IFUNC(int, setreg, (void *vmi, int vcpu, int num,
217     uint64_t val))
218 DEFINE_VMMOPS_IFUNC(int, getdesc, (void *vmi, int vcpu, int num,
219     struct seg_desc *desc))
220 DEFINE_VMMOPS_IFUNC(int, setdesc, (void *vmi, int vcpu, int num,
221     struct seg_desc *desc))
222 DEFINE_VMMOPS_IFUNC(int, getcap, (void *vmi, int vcpu, int num, int *retval))
223 DEFINE_VMMOPS_IFUNC(int, setcap, (void *vmi, int vcpu, int num, int val))
224 DEFINE_VMMOPS_IFUNC(struct vmspace *, vmspace_alloc, (vm_offset_t min,
225     vm_offset_t max))
226 DEFINE_VMMOPS_IFUNC(void, vmspace_free, (struct vmspace *vmspace))
227 DEFINE_VMMOPS_IFUNC(struct vlapic *, vlapic_init, (void *vmi, int vcpu))
228 DEFINE_VMMOPS_IFUNC(void, vlapic_cleanup, (void *vmi, struct vlapic *vlapic))
229 #ifdef BHYVE_SNAPSHOT
230 DEFINE_VMMOPS_IFUNC(int, snapshot, (void *vmi, struct vm_snapshot_meta
231     *meta))
232 DEFINE_VMMOPS_IFUNC(int, vmcx_snapshot, (void *vmi, struct vm_snapshot_meta
233     *meta, int vcpu))
234 DEFINE_VMMOPS_IFUNC(int, restore_tsc, (void *vmi, int vcpuid, uint64_t now))
235 #endif
236 
237 #define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
238 #define	fpu_stop_emulating()	clts()
239 
240 SDT_PROVIDER_DEFINE(vmm);
241 
242 static MALLOC_DEFINE(M_VM, "vm", "vm");
243 
244 /* statistics */
245 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
246 
247 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
248     NULL);
249 
250 /*
251  * Halt the guest if all vcpus are executing a HLT instruction with
252  * interrupts disabled.
253  */
254 static int halt_detection_enabled = 1;
255 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
256     &halt_detection_enabled, 0,
257     "Halt VM if all vcpus execute HLT with interrupts disabled");
258 
259 static int vmm_ipinum;
260 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
261     "IPI vector used for vcpu notifications");
262 
263 static int trace_guest_exceptions;
264 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN,
265     &trace_guest_exceptions, 0,
266     "Trap into hypervisor on all guest exceptions and reflect them back");
267 
268 static void vm_free_memmap(struct vm *vm, int ident);
269 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm);
270 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr);
271 
272 #ifdef KTR
273 static const char *
274 vcpu_state2str(enum vcpu_state state)
275 {
276 
277 	switch (state) {
278 	case VCPU_IDLE:
279 		return ("idle");
280 	case VCPU_FROZEN:
281 		return ("frozen");
282 	case VCPU_RUNNING:
283 		return ("running");
284 	case VCPU_SLEEPING:
285 		return ("sleeping");
286 	default:
287 		return ("unknown");
288 	}
289 }
290 #endif
291 
292 static void
293 vcpu_cleanup(struct vm *vm, int i, bool destroy)
294 {
295 	struct vcpu *vcpu = &vm->vcpu[i];
296 
297 	vmmops_vlapic_cleanup(vm->cookie, vcpu->vlapic);
298 	if (destroy) {
299 		vmm_stat_free(vcpu->stats);
300 		fpu_save_area_free(vcpu->guestfpu);
301 	}
302 }
303 
304 static void
305 vcpu_init(struct vm *vm, int vcpu_id, bool create)
306 {
307 	struct vcpu *vcpu;
308 
309 	KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus,
310 	    ("vcpu_init: invalid vcpu %d", vcpu_id));
311 
312 	vcpu = &vm->vcpu[vcpu_id];
313 
314 	if (create) {
315 		KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
316 		    "initialized", vcpu_id));
317 		vcpu_lock_init(vcpu);
318 		vcpu->state = VCPU_IDLE;
319 		vcpu->hostcpu = NOCPU;
320 		vcpu->guestfpu = fpu_save_area_alloc();
321 		vcpu->stats = vmm_stat_alloc();
322 		vcpu->tsc_offset = 0;
323 	}
324 
325 	vcpu->vlapic = vmmops_vlapic_init(vm->cookie, vcpu_id);
326 	vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
327 	vcpu->reqidle = 0;
328 	vcpu->exitintinfo = 0;
329 	vcpu->nmi_pending = 0;
330 	vcpu->extint_pending = 0;
331 	vcpu->exception_pending = 0;
332 	vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
333 	fpu_save_area_reset(vcpu->guestfpu);
334 	vmm_stat_init(vcpu->stats);
335 }
336 
337 int
338 vcpu_trace_exceptions(struct vm *vm, int vcpuid)
339 {
340 
341 	return (trace_guest_exceptions);
342 }
343 
344 struct vm_exit *
345 vm_exitinfo(struct vm *vm, int cpuid)
346 {
347 	struct vcpu *vcpu;
348 
349 	if (cpuid < 0 || cpuid >= vm->maxcpus)
350 		panic("vm_exitinfo: invalid cpuid %d", cpuid);
351 
352 	vcpu = &vm->vcpu[cpuid];
353 
354 	return (&vcpu->exitinfo);
355 }
356 
357 static int
358 vmm_init(void)
359 {
360 	int error;
361 
362 	if (!vmm_is_hw_supported())
363 		return (ENXIO);
364 
365 	vmm_host_state_init();
366 
367 	vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) :
368 	    &IDTVEC(justreturn));
369 	if (vmm_ipinum < 0)
370 		vmm_ipinum = IPI_AST;
371 
372 	error = vmm_mem_init();
373 	if (error)
374 		return (error);
375 
376 	vmm_resume_p = vmmops_modresume;
377 
378 	return (vmmops_modinit(vmm_ipinum));
379 }
380 
381 static int
382 vmm_handler(module_t mod, int what, void *arg)
383 {
384 	int error;
385 
386 	switch (what) {
387 	case MOD_LOAD:
388 		if (vmm_is_hw_supported()) {
389 			vmmdev_init();
390 			error = vmm_init();
391 			if (error == 0)
392 				vmm_initialized = 1;
393 		} else {
394 			error = ENXIO;
395 		}
396 		break;
397 	case MOD_UNLOAD:
398 		if (vmm_is_hw_supported()) {
399 			error = vmmdev_cleanup();
400 			if (error == 0) {
401 				vmm_resume_p = NULL;
402 				iommu_cleanup();
403 				if (vmm_ipinum != IPI_AST)
404 					lapic_ipi_free(vmm_ipinum);
405 				error = vmmops_modcleanup();
406 				/*
407 				 * Something bad happened - prevent new
408 				 * VMs from being created
409 				 */
410 				if (error)
411 					vmm_initialized = 0;
412 			}
413 		} else {
414 			error = 0;
415 		}
416 		break;
417 	default:
418 		error = 0;
419 		break;
420 	}
421 	return (error);
422 }
423 
424 static moduledata_t vmm_kmod = {
425 	"vmm",
426 	vmm_handler,
427 	NULL
428 };
429 
430 /*
431  * vmm initialization has the following dependencies:
432  *
433  * - VT-x initialization requires smp_rendezvous() and therefore must happen
434  *   after SMP is fully functional (after SI_SUB_SMP).
435  */
436 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
437 MODULE_VERSION(vmm, 1);
438 
439 static void
440 vm_init(struct vm *vm, bool create)
441 {
442 	int i;
443 
444 	vm->cookie = vmmops_init(vm, vmspace_pmap(vm->vmspace));
445 	vm->iommu = NULL;
446 	vm->vioapic = vioapic_init(vm);
447 	vm->vhpet = vhpet_init(vm);
448 	vm->vatpic = vatpic_init(vm);
449 	vm->vatpit = vatpit_init(vm);
450 	vm->vpmtmr = vpmtmr_init(vm);
451 	if (create)
452 		vm->vrtc = vrtc_init(vm);
453 
454 	CPU_ZERO(&vm->active_cpus);
455 	CPU_ZERO(&vm->debug_cpus);
456 
457 	vm->suspend = 0;
458 	CPU_ZERO(&vm->suspended_cpus);
459 
460 	for (i = 0; i < vm->maxcpus; i++)
461 		vcpu_init(vm, i, create);
462 }
463 
464 /*
465  * The default CPU topology is a single thread per package.
466  */
467 u_int cores_per_package = 1;
468 u_int threads_per_core = 1;
469 
470 int
471 vm_create(const char *name, struct vm **retvm)
472 {
473 	struct vm *vm;
474 	struct vmspace *vmspace;
475 
476 	/*
477 	 * If vmm.ko could not be successfully initialized then don't attempt
478 	 * to create the virtual machine.
479 	 */
480 	if (!vmm_initialized)
481 		return (ENXIO);
482 
483 	if (name == NULL || strnlen(name, VM_MAX_NAMELEN + 1) ==
484 	    VM_MAX_NAMELEN + 1)
485 		return (EINVAL);
486 
487 	vmspace = vmmops_vmspace_alloc(0, VM_MAXUSER_ADDRESS_LA48);
488 	if (vmspace == NULL)
489 		return (ENOMEM);
490 
491 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
492 	strcpy(vm->name, name);
493 	vm->vmspace = vmspace;
494 	mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
495 
496 	vm->sockets = 1;
497 	vm->cores = cores_per_package;	/* XXX backwards compatibility */
498 	vm->threads = threads_per_core;	/* XXX backwards compatibility */
499 	vm->maxcpus = VM_MAXCPU;	/* XXX temp to keep code working */
500 
501 	vm_init(vm, true);
502 
503 	*retvm = vm;
504 	return (0);
505 }
506 
507 void
508 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
509     uint16_t *threads, uint16_t *maxcpus)
510 {
511 	*sockets = vm->sockets;
512 	*cores = vm->cores;
513 	*threads = vm->threads;
514 	*maxcpus = vm->maxcpus;
515 }
516 
517 uint16_t
518 vm_get_maxcpus(struct vm *vm)
519 {
520 	return (vm->maxcpus);
521 }
522 
523 int
524 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
525     uint16_t threads, uint16_t maxcpus)
526 {
527 	if (maxcpus != 0)
528 		return (EINVAL);	/* XXX remove when supported */
529 	if ((sockets * cores * threads) > vm->maxcpus)
530 		return (EINVAL);
531 	/* XXX need to check sockets * cores * threads == vCPU, how? */
532 	vm->sockets = sockets;
533 	vm->cores = cores;
534 	vm->threads = threads;
535 	vm->maxcpus = VM_MAXCPU;	/* XXX temp to keep code working */
536 	return(0);
537 }
538 
539 static void
540 vm_cleanup(struct vm *vm, bool destroy)
541 {
542 	struct mem_map *mm;
543 	int i;
544 
545 	ppt_unassign_all(vm);
546 
547 	if (vm->iommu != NULL)
548 		iommu_destroy_domain(vm->iommu);
549 
550 	if (destroy)
551 		vrtc_cleanup(vm->vrtc);
552 	else
553 		vrtc_reset(vm->vrtc);
554 	vpmtmr_cleanup(vm->vpmtmr);
555 	vatpit_cleanup(vm->vatpit);
556 	vhpet_cleanup(vm->vhpet);
557 	vatpic_cleanup(vm->vatpic);
558 	vioapic_cleanup(vm->vioapic);
559 
560 	for (i = 0; i < vm->maxcpus; i++)
561 		vcpu_cleanup(vm, i, destroy);
562 
563 	vmmops_cleanup(vm->cookie);
564 
565 	/*
566 	 * System memory is removed from the guest address space only when
567 	 * the VM is destroyed. This is because the mapping remains the same
568 	 * across VM reset.
569 	 *
570 	 * Device memory can be relocated by the guest (e.g. using PCI BARs)
571 	 * so those mappings are removed on a VM reset.
572 	 */
573 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
574 		mm = &vm->mem_maps[i];
575 		if (destroy || !sysmem_mapping(vm, mm))
576 			vm_free_memmap(vm, i);
577 	}
578 
579 	if (destroy) {
580 		for (i = 0; i < VM_MAX_MEMSEGS; i++)
581 			vm_free_memseg(vm, i);
582 
583 		vmmops_vmspace_free(vm->vmspace);
584 		vm->vmspace = NULL;
585 	}
586 }
587 
588 void
589 vm_destroy(struct vm *vm)
590 {
591 	vm_cleanup(vm, true);
592 	free(vm, M_VM);
593 }
594 
595 int
596 vm_reinit(struct vm *vm)
597 {
598 	int error;
599 
600 	/*
601 	 * A virtual machine can be reset only if all vcpus are suspended.
602 	 */
603 	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
604 		vm_cleanup(vm, false);
605 		vm_init(vm, false);
606 		error = 0;
607 	} else {
608 		error = EBUSY;
609 	}
610 
611 	return (error);
612 }
613 
614 const char *
615 vm_name(struct vm *vm)
616 {
617 	return (vm->name);
618 }
619 
620 int
621 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
622 {
623 	vm_object_t obj;
624 
625 	if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
626 		return (ENOMEM);
627 	else
628 		return (0);
629 }
630 
631 int
632 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
633 {
634 
635 	vmm_mmio_free(vm->vmspace, gpa, len);
636 	return (0);
637 }
638 
639 /*
640  * Return 'true' if 'gpa' is allocated in the guest address space.
641  *
642  * This function is called in the context of a running vcpu which acts as
643  * an implicit lock on 'vm->mem_maps[]'.
644  */
645 bool
646 vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa)
647 {
648 	struct mem_map *mm;
649 	int i;
650 
651 #ifdef INVARIANTS
652 	int hostcpu, state;
653 	state = vcpu_get_state(vm, vcpuid, &hostcpu);
654 	KASSERT(state == VCPU_RUNNING && hostcpu == curcpu,
655 	    ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu));
656 #endif
657 
658 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
659 		mm = &vm->mem_maps[i];
660 		if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len)
661 			return (true);		/* 'gpa' is sysmem or devmem */
662 	}
663 
664 	if (ppt_is_mmio(vm, gpa))
665 		return (true);			/* 'gpa' is pci passthru mmio */
666 
667 	return (false);
668 }
669 
670 int
671 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem)
672 {
673 	struct mem_seg *seg;
674 	vm_object_t obj;
675 
676 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
677 		return (EINVAL);
678 
679 	if (len == 0 || (len & PAGE_MASK))
680 		return (EINVAL);
681 
682 	seg = &vm->mem_segs[ident];
683 	if (seg->object != NULL) {
684 		if (seg->len == len && seg->sysmem == sysmem)
685 			return (EEXIST);
686 		else
687 			return (EINVAL);
688 	}
689 
690 	obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT);
691 	if (obj == NULL)
692 		return (ENOMEM);
693 
694 	seg->len = len;
695 	seg->object = obj;
696 	seg->sysmem = sysmem;
697 	return (0);
698 }
699 
700 int
701 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
702     vm_object_t *objptr)
703 {
704 	struct mem_seg *seg;
705 
706 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
707 		return (EINVAL);
708 
709 	seg = &vm->mem_segs[ident];
710 	if (len)
711 		*len = seg->len;
712 	if (sysmem)
713 		*sysmem = seg->sysmem;
714 	if (objptr)
715 		*objptr = seg->object;
716 	return (0);
717 }
718 
719 void
720 vm_free_memseg(struct vm *vm, int ident)
721 {
722 	struct mem_seg *seg;
723 
724 	KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS,
725 	    ("%s: invalid memseg ident %d", __func__, ident));
726 
727 	seg = &vm->mem_segs[ident];
728 	if (seg->object != NULL) {
729 		vm_object_deallocate(seg->object);
730 		bzero(seg, sizeof(struct mem_seg));
731 	}
732 }
733 
734 int
735 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first,
736     size_t len, int prot, int flags)
737 {
738 	struct mem_seg *seg;
739 	struct mem_map *m, *map;
740 	vm_ooffset_t last;
741 	int i, error;
742 
743 	if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0)
744 		return (EINVAL);
745 
746 	if (flags & ~VM_MEMMAP_F_WIRED)
747 		return (EINVAL);
748 
749 	if (segid < 0 || segid >= VM_MAX_MEMSEGS)
750 		return (EINVAL);
751 
752 	seg = &vm->mem_segs[segid];
753 	if (seg->object == NULL)
754 		return (EINVAL);
755 
756 	last = first + len;
757 	if (first < 0 || first >= last || last > seg->len)
758 		return (EINVAL);
759 
760 	if ((gpa | first | last) & PAGE_MASK)
761 		return (EINVAL);
762 
763 	map = NULL;
764 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
765 		m = &vm->mem_maps[i];
766 		if (m->len == 0) {
767 			map = m;
768 			break;
769 		}
770 	}
771 
772 	if (map == NULL)
773 		return (ENOSPC);
774 
775 	error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa,
776 	    len, 0, VMFS_NO_SPACE, prot, prot, 0);
777 	if (error != KERN_SUCCESS)
778 		return (EFAULT);
779 
780 	vm_object_reference(seg->object);
781 
782 	if (flags & VM_MEMMAP_F_WIRED) {
783 		error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len,
784 		    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
785 		if (error != KERN_SUCCESS) {
786 			vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len);
787 			return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM :
788 			    EFAULT);
789 		}
790 	}
791 
792 	map->gpa = gpa;
793 	map->len = len;
794 	map->segoff = first;
795 	map->segid = segid;
796 	map->prot = prot;
797 	map->flags = flags;
798 	return (0);
799 }
800 
801 int
802 vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len)
803 {
804 	struct mem_map *m;
805 	int i;
806 
807 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
808 		m = &vm->mem_maps[i];
809 		if (m->gpa == gpa && m->len == len &&
810 		    (m->flags & VM_MEMMAP_F_IOMMU) == 0) {
811 			vm_free_memmap(vm, i);
812 			return (0);
813 		}
814 	}
815 
816 	return (EINVAL);
817 }
818 
819 int
820 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
821     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags)
822 {
823 	struct mem_map *mm, *mmnext;
824 	int i;
825 
826 	mmnext = NULL;
827 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
828 		mm = &vm->mem_maps[i];
829 		if (mm->len == 0 || mm->gpa < *gpa)
830 			continue;
831 		if (mmnext == NULL || mm->gpa < mmnext->gpa)
832 			mmnext = mm;
833 	}
834 
835 	if (mmnext != NULL) {
836 		*gpa = mmnext->gpa;
837 		if (segid)
838 			*segid = mmnext->segid;
839 		if (segoff)
840 			*segoff = mmnext->segoff;
841 		if (len)
842 			*len = mmnext->len;
843 		if (prot)
844 			*prot = mmnext->prot;
845 		if (flags)
846 			*flags = mmnext->flags;
847 		return (0);
848 	} else {
849 		return (ENOENT);
850 	}
851 }
852 
853 static void
854 vm_free_memmap(struct vm *vm, int ident)
855 {
856 	struct mem_map *mm;
857 	int error __diagused;
858 
859 	mm = &vm->mem_maps[ident];
860 	if (mm->len) {
861 		error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa,
862 		    mm->gpa + mm->len);
863 		KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d",
864 		    __func__, error));
865 		bzero(mm, sizeof(struct mem_map));
866 	}
867 }
868 
869 static __inline bool
870 sysmem_mapping(struct vm *vm, struct mem_map *mm)
871 {
872 
873 	if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem)
874 		return (true);
875 	else
876 		return (false);
877 }
878 
879 vm_paddr_t
880 vmm_sysmem_maxaddr(struct vm *vm)
881 {
882 	struct mem_map *mm;
883 	vm_paddr_t maxaddr;
884 	int i;
885 
886 	maxaddr = 0;
887 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
888 		mm = &vm->mem_maps[i];
889 		if (sysmem_mapping(vm, mm)) {
890 			if (maxaddr < mm->gpa + mm->len)
891 				maxaddr = mm->gpa + mm->len;
892 		}
893 	}
894 	return (maxaddr);
895 }
896 
897 static void
898 vm_iommu_modify(struct vm *vm, bool map)
899 {
900 	int i, sz;
901 	vm_paddr_t gpa, hpa;
902 	struct mem_map *mm;
903 	void *vp, *cookie, *host_domain;
904 
905 	sz = PAGE_SIZE;
906 	host_domain = iommu_host_domain();
907 
908 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
909 		mm = &vm->mem_maps[i];
910 		if (!sysmem_mapping(vm, mm))
911 			continue;
912 
913 		if (map) {
914 			KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0,
915 			    ("iommu map found invalid memmap %#lx/%#lx/%#x",
916 			    mm->gpa, mm->len, mm->flags));
917 			if ((mm->flags & VM_MEMMAP_F_WIRED) == 0)
918 				continue;
919 			mm->flags |= VM_MEMMAP_F_IOMMU;
920 		} else {
921 			if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0)
922 				continue;
923 			mm->flags &= ~VM_MEMMAP_F_IOMMU;
924 			KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0,
925 			    ("iommu unmap found invalid memmap %#lx/%#lx/%#x",
926 			    mm->gpa, mm->len, mm->flags));
927 		}
928 
929 		gpa = mm->gpa;
930 		while (gpa < mm->gpa + mm->len) {
931 			vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE,
932 					 &cookie);
933 			KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
934 			    vm_name(vm), gpa));
935 
936 			vm_gpa_release(cookie);
937 
938 			hpa = DMAP_TO_PHYS((uintptr_t)vp);
939 			if (map) {
940 				iommu_create_mapping(vm->iommu, gpa, hpa, sz);
941 			} else {
942 				iommu_remove_mapping(vm->iommu, gpa, sz);
943 			}
944 
945 			gpa += PAGE_SIZE;
946 		}
947 	}
948 
949 	/*
950 	 * Invalidate the cached translations associated with the domain
951 	 * from which pages were removed.
952 	 */
953 	if (map)
954 		iommu_invalidate_tlb(host_domain);
955 	else
956 		iommu_invalidate_tlb(vm->iommu);
957 }
958 
959 #define	vm_iommu_unmap(vm)	vm_iommu_modify((vm), false)
960 #define	vm_iommu_map(vm)	vm_iommu_modify((vm), true)
961 
962 int
963 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
964 {
965 	int error;
966 
967 	error = ppt_unassign_device(vm, bus, slot, func);
968 	if (error)
969 		return (error);
970 
971 	if (ppt_assigned_devices(vm) == 0)
972 		vm_iommu_unmap(vm);
973 
974 	return (0);
975 }
976 
977 int
978 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
979 {
980 	int error;
981 	vm_paddr_t maxaddr;
982 
983 	/* Set up the IOMMU to do the 'gpa' to 'hpa' translation */
984 	if (ppt_assigned_devices(vm) == 0) {
985 		KASSERT(vm->iommu == NULL,
986 		    ("vm_assign_pptdev: iommu must be NULL"));
987 		maxaddr = vmm_sysmem_maxaddr(vm);
988 		vm->iommu = iommu_create_domain(maxaddr);
989 		if (vm->iommu == NULL)
990 			return (ENXIO);
991 		vm_iommu_map(vm);
992 	}
993 
994 	error = ppt_assign_device(vm, bus, slot, func);
995 	return (error);
996 }
997 
998 void *
999 vm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot,
1000 	    void **cookie)
1001 {
1002 	int i, count, pageoff;
1003 	struct mem_map *mm;
1004 	vm_page_t m;
1005 #ifdef INVARIANTS
1006 	/*
1007 	 * All vcpus are frozen by ioctls that modify the memory map
1008 	 * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is
1009 	 * guaranteed if at least one vcpu is in the VCPU_FROZEN state.
1010 	 */
1011 	int state;
1012 	KASSERT(vcpuid >= -1 && vcpuid < vm->maxcpus, ("%s: invalid vcpuid %d",
1013 	    __func__, vcpuid));
1014 	for (i = 0; i < vm->maxcpus; i++) {
1015 		if (vcpuid != -1 && vcpuid != i)
1016 			continue;
1017 		state = vcpu_get_state(vm, i, NULL);
1018 		KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d",
1019 		    __func__, state));
1020 	}
1021 #endif
1022 	pageoff = gpa & PAGE_MASK;
1023 	if (len > PAGE_SIZE - pageoff)
1024 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
1025 
1026 	count = 0;
1027 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
1028 		mm = &vm->mem_maps[i];
1029 		if (gpa >= mm->gpa && gpa < mm->gpa + mm->len) {
1030 			count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
1031 			    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
1032 			break;
1033 		}
1034 	}
1035 
1036 	if (count == 1) {
1037 		*cookie = m;
1038 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
1039 	} else {
1040 		*cookie = NULL;
1041 		return (NULL);
1042 	}
1043 }
1044 
1045 void
1046 vm_gpa_release(void *cookie)
1047 {
1048 	vm_page_t m = cookie;
1049 
1050 	vm_page_unwire(m, PQ_ACTIVE);
1051 }
1052 
1053 int
1054 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
1055 {
1056 
1057 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1058 		return (EINVAL);
1059 
1060 	if (reg >= VM_REG_LAST)
1061 		return (EINVAL);
1062 
1063 	return (vmmops_getreg(vm->cookie, vcpu, reg, retval));
1064 }
1065 
1066 int
1067 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val)
1068 {
1069 	struct vcpu *vcpu;
1070 	int error;
1071 
1072 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1073 		return (EINVAL);
1074 
1075 	if (reg >= VM_REG_LAST)
1076 		return (EINVAL);
1077 
1078 	error = vmmops_setreg(vm->cookie, vcpuid, reg, val);
1079 	if (error || reg != VM_REG_GUEST_RIP)
1080 		return (error);
1081 
1082 	/* Set 'nextrip' to match the value of %rip */
1083 	VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val);
1084 	vcpu = &vm->vcpu[vcpuid];
1085 	vcpu->nextrip = val;
1086 	return (0);
1087 }
1088 
1089 static bool
1090 is_descriptor_table(int reg)
1091 {
1092 
1093 	switch (reg) {
1094 	case VM_REG_GUEST_IDTR:
1095 	case VM_REG_GUEST_GDTR:
1096 		return (true);
1097 	default:
1098 		return (false);
1099 	}
1100 }
1101 
1102 static bool
1103 is_segment_register(int reg)
1104 {
1105 
1106 	switch (reg) {
1107 	case VM_REG_GUEST_ES:
1108 	case VM_REG_GUEST_CS:
1109 	case VM_REG_GUEST_SS:
1110 	case VM_REG_GUEST_DS:
1111 	case VM_REG_GUEST_FS:
1112 	case VM_REG_GUEST_GS:
1113 	case VM_REG_GUEST_TR:
1114 	case VM_REG_GUEST_LDTR:
1115 		return (true);
1116 	default:
1117 		return (false);
1118 	}
1119 }
1120 
1121 int
1122 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
1123 		struct seg_desc *desc)
1124 {
1125 
1126 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1127 		return (EINVAL);
1128 
1129 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1130 		return (EINVAL);
1131 
1132 	return (vmmops_getdesc(vm->cookie, vcpu, reg, desc));
1133 }
1134 
1135 int
1136 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
1137 		struct seg_desc *desc)
1138 {
1139 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1140 		return (EINVAL);
1141 
1142 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1143 		return (EINVAL);
1144 
1145 	return (vmmops_setdesc(vm->cookie, vcpu, reg, desc));
1146 }
1147 
1148 static void
1149 restore_guest_fpustate(struct vcpu *vcpu)
1150 {
1151 
1152 	/* flush host state to the pcb */
1153 	fpuexit(curthread);
1154 
1155 	/* restore guest FPU state */
1156 	fpu_stop_emulating();
1157 	fpurestore(vcpu->guestfpu);
1158 
1159 	/* restore guest XCR0 if XSAVE is enabled in the host */
1160 	if (rcr4() & CR4_XSAVE)
1161 		load_xcr(0, vcpu->guest_xcr0);
1162 
1163 	/*
1164 	 * The FPU is now "dirty" with the guest's state so turn on emulation
1165 	 * to trap any access to the FPU by the host.
1166 	 */
1167 	fpu_start_emulating();
1168 }
1169 
1170 static void
1171 save_guest_fpustate(struct vcpu *vcpu)
1172 {
1173 
1174 	if ((rcr0() & CR0_TS) == 0)
1175 		panic("fpu emulation not enabled in host!");
1176 
1177 	/* save guest XCR0 and restore host XCR0 */
1178 	if (rcr4() & CR4_XSAVE) {
1179 		vcpu->guest_xcr0 = rxcr(0);
1180 		load_xcr(0, vmm_get_host_xcr0());
1181 	}
1182 
1183 	/* save guest FPU state */
1184 	fpu_stop_emulating();
1185 	fpusave(vcpu->guestfpu);
1186 	fpu_start_emulating();
1187 }
1188 
1189 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
1190 
1191 static int
1192 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1193     bool from_idle)
1194 {
1195 	struct vcpu *vcpu;
1196 	int error;
1197 
1198 	vcpu = &vm->vcpu[vcpuid];
1199 	vcpu_assert_locked(vcpu);
1200 
1201 	/*
1202 	 * State transitions from the vmmdev_ioctl() must always begin from
1203 	 * the VCPU_IDLE state. This guarantees that there is only a single
1204 	 * ioctl() operating on a vcpu at any point.
1205 	 */
1206 	if (from_idle) {
1207 		while (vcpu->state != VCPU_IDLE) {
1208 			vcpu->reqidle = 1;
1209 			vcpu_notify_event_locked(vcpu, false);
1210 			VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to "
1211 			    "idle requested", vcpu_state2str(vcpu->state));
1212 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
1213 		}
1214 	} else {
1215 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1216 		    "vcpu idle state"));
1217 	}
1218 
1219 	if (vcpu->state == VCPU_RUNNING) {
1220 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1221 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1222 	} else {
1223 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1224 		    "vcpu that is not running", vcpu->hostcpu));
1225 	}
1226 
1227 	/*
1228 	 * The following state transitions are allowed:
1229 	 * IDLE -> FROZEN -> IDLE
1230 	 * FROZEN -> RUNNING -> FROZEN
1231 	 * FROZEN -> SLEEPING -> FROZEN
1232 	 */
1233 	switch (vcpu->state) {
1234 	case VCPU_IDLE:
1235 	case VCPU_RUNNING:
1236 	case VCPU_SLEEPING:
1237 		error = (newstate != VCPU_FROZEN);
1238 		break;
1239 	case VCPU_FROZEN:
1240 		error = (newstate == VCPU_FROZEN);
1241 		break;
1242 	default:
1243 		error = 1;
1244 		break;
1245 	}
1246 
1247 	if (error)
1248 		return (EBUSY);
1249 
1250 	VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s",
1251 	    vcpu_state2str(vcpu->state), vcpu_state2str(newstate));
1252 
1253 	vcpu->state = newstate;
1254 	if (newstate == VCPU_RUNNING)
1255 		vcpu->hostcpu = curcpu;
1256 	else
1257 		vcpu->hostcpu = NOCPU;
1258 
1259 	if (newstate == VCPU_IDLE)
1260 		wakeup(&vcpu->state);
1261 
1262 	return (0);
1263 }
1264 
1265 static void
1266 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1267 {
1268 	int error;
1269 
1270 	if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1271 		panic("Error %d setting state to %d\n", error, newstate);
1272 }
1273 
1274 static void
1275 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1276 {
1277 	int error;
1278 
1279 	if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0)
1280 		panic("Error %d setting state to %d", error, newstate);
1281 }
1282 
1283 #define	RENDEZVOUS_CTR0(vm, vcpuid, fmt)				\
1284 	do {								\
1285 		if (vcpuid >= 0)					\
1286 			VCPU_CTR0(vm, vcpuid, fmt);			\
1287 		else							\
1288 			VM_CTR0(vm, fmt);				\
1289 	} while (0)
1290 
1291 static int
1292 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1293 {
1294 	struct thread *td;
1295 	int error;
1296 
1297 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
1298 	    ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1299 
1300 	error = 0;
1301 	td = curthread;
1302 	mtx_lock(&vm->rendezvous_mtx);
1303 	while (vm->rendezvous_func != NULL) {
1304 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1305 		CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus);
1306 
1307 		if (vcpuid != -1 &&
1308 		    CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1309 		    !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1310 			VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1311 			(*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1312 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1313 		}
1314 		if (CPU_CMP(&vm->rendezvous_req_cpus,
1315 		    &vm->rendezvous_done_cpus) == 0) {
1316 			VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1317 			vm->rendezvous_func = NULL;
1318 			wakeup(&vm->rendezvous_func);
1319 			break;
1320 		}
1321 		RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1322 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1323 		    "vmrndv", hz);
1324 		if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1325 			mtx_unlock(&vm->rendezvous_mtx);
1326 			error = thread_check_susp(td, true);
1327 			if (error != 0)
1328 				return (error);
1329 			mtx_lock(&vm->rendezvous_mtx);
1330 		}
1331 	}
1332 	mtx_unlock(&vm->rendezvous_mtx);
1333 	return (0);
1334 }
1335 
1336 /*
1337  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1338  */
1339 static int
1340 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1341 {
1342 	struct vcpu *vcpu;
1343 	const char *wmesg;
1344 	struct thread *td;
1345 	int error, t, vcpu_halted, vm_halted;
1346 
1347 	KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1348 
1349 	vcpu = &vm->vcpu[vcpuid];
1350 	vcpu_halted = 0;
1351 	vm_halted = 0;
1352 	error = 0;
1353 	td = curthread;
1354 
1355 	vcpu_lock(vcpu);
1356 	while (1) {
1357 		/*
1358 		 * Do a final check for pending NMI or interrupts before
1359 		 * really putting this thread to sleep. Also check for
1360 		 * software events that would cause this vcpu to wakeup.
1361 		 *
1362 		 * These interrupts/events could have happened after the
1363 		 * vcpu returned from vmmops_run() and before it acquired the
1364 		 * vcpu lock above.
1365 		 */
1366 		if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle)
1367 			break;
1368 		if (vm_nmi_pending(vm, vcpuid))
1369 			break;
1370 		if (!intr_disabled) {
1371 			if (vm_extint_pending(vm, vcpuid) ||
1372 			    vlapic_pending_intr(vcpu->vlapic, NULL)) {
1373 				break;
1374 			}
1375 		}
1376 
1377 		/* Don't go to sleep if the vcpu thread needs to yield */
1378 		if (vcpu_should_yield(vm, vcpuid))
1379 			break;
1380 
1381 		if (vcpu_debugged(vm, vcpuid))
1382 			break;
1383 
1384 		/*
1385 		 * Some Linux guests implement "halt" by having all vcpus
1386 		 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1387 		 * track of the vcpus that have entered this state. When all
1388 		 * vcpus enter the halted state the virtual machine is halted.
1389 		 */
1390 		if (intr_disabled) {
1391 			wmesg = "vmhalt";
1392 			VCPU_CTR0(vm, vcpuid, "Halted");
1393 			if (!vcpu_halted && halt_detection_enabled) {
1394 				vcpu_halted = 1;
1395 				CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1396 			}
1397 			if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1398 				vm_halted = 1;
1399 				break;
1400 			}
1401 		} else {
1402 			wmesg = "vmidle";
1403 		}
1404 
1405 		t = ticks;
1406 		vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1407 		/*
1408 		 * XXX msleep_spin() cannot be interrupted by signals so
1409 		 * wake up periodically to check pending signals.
1410 		 */
1411 		msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1412 		vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1413 		vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1414 		if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1415 			vcpu_unlock(vcpu);
1416 			error = thread_check_susp(td, false);
1417 			if (error != 0)
1418 				return (error);
1419 			vcpu_lock(vcpu);
1420 		}
1421 	}
1422 
1423 	if (vcpu_halted)
1424 		CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1425 
1426 	vcpu_unlock(vcpu);
1427 
1428 	if (vm_halted)
1429 		vm_suspend(vm, VM_SUSPEND_HALT);
1430 
1431 	return (0);
1432 }
1433 
1434 static int
1435 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1436 {
1437 	int rv, ftype;
1438 	struct vm_map *map;
1439 	struct vcpu *vcpu;
1440 	struct vm_exit *vme;
1441 
1442 	vcpu = &vm->vcpu[vcpuid];
1443 	vme = &vcpu->exitinfo;
1444 
1445 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1446 	    __func__, vme->inst_length));
1447 
1448 	ftype = vme->u.paging.fault_type;
1449 	KASSERT(ftype == VM_PROT_READ ||
1450 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1451 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1452 
1453 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1454 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1455 		    vme->u.paging.gpa, ftype);
1456 		if (rv == 0) {
1457 			VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1458 			    ftype == VM_PROT_READ ? "accessed" : "dirty",
1459 			    vme->u.paging.gpa);
1460 			goto done;
1461 		}
1462 	}
1463 
1464 	map = &vm->vmspace->vm_map;
1465 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL);
1466 
1467 	VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1468 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1469 
1470 	if (rv != KERN_SUCCESS)
1471 		return (EFAULT);
1472 done:
1473 	return (0);
1474 }
1475 
1476 static int
1477 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1478 {
1479 	struct vie *vie;
1480 	struct vcpu *vcpu;
1481 	struct vm_exit *vme;
1482 	uint64_t gla, gpa, cs_base;
1483 	struct vm_guest_paging *paging;
1484 	mem_region_read_t mread;
1485 	mem_region_write_t mwrite;
1486 	enum vm_cpu_mode cpu_mode;
1487 	int cs_d, error, fault;
1488 
1489 	vcpu = &vm->vcpu[vcpuid];
1490 	vme = &vcpu->exitinfo;
1491 
1492 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1493 	    __func__, vme->inst_length));
1494 
1495 	gla = vme->u.inst_emul.gla;
1496 	gpa = vme->u.inst_emul.gpa;
1497 	cs_base = vme->u.inst_emul.cs_base;
1498 	cs_d = vme->u.inst_emul.cs_d;
1499 	vie = &vme->u.inst_emul.vie;
1500 	paging = &vme->u.inst_emul.paging;
1501 	cpu_mode = paging->cpu_mode;
1502 
1503 	VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1504 
1505 	/* Fetch, decode and emulate the faulting instruction */
1506 	if (vie->num_valid == 0) {
1507 		error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip +
1508 		    cs_base, VIE_INST_SIZE, vie, &fault);
1509 	} else {
1510 		/*
1511 		 * The instruction bytes have already been copied into 'vie'
1512 		 */
1513 		error = fault = 0;
1514 	}
1515 	if (error || fault)
1516 		return (error);
1517 
1518 	if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) {
1519 		VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx",
1520 		    vme->rip + cs_base);
1521 		*retu = true;	    /* dump instruction bytes in userspace */
1522 		return (0);
1523 	}
1524 
1525 	/*
1526 	 * Update 'nextrip' based on the length of the emulated instruction.
1527 	 */
1528 	vme->inst_length = vie->num_processed;
1529 	vcpu->nextrip += vie->num_processed;
1530 	VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction "
1531 	    "decoding", vcpu->nextrip);
1532 
1533 	/* return to userland unless this is an in-kernel emulated device */
1534 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1535 		mread = lapic_mmio_read;
1536 		mwrite = lapic_mmio_write;
1537 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1538 		mread = vioapic_mmio_read;
1539 		mwrite = vioapic_mmio_write;
1540 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1541 		mread = vhpet_mmio_read;
1542 		mwrite = vhpet_mmio_write;
1543 	} else {
1544 		*retu = true;
1545 		return (0);
1546 	}
1547 
1548 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1549 	    mread, mwrite, retu);
1550 
1551 	return (error);
1552 }
1553 
1554 static int
1555 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1556 {
1557 	int error, i;
1558 	struct vcpu *vcpu;
1559 	struct thread *td;
1560 
1561 	error = 0;
1562 	vcpu = &vm->vcpu[vcpuid];
1563 	td = curthread;
1564 
1565 	CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1566 
1567 	/*
1568 	 * Wait until all 'active_cpus' have suspended themselves.
1569 	 *
1570 	 * Since a VM may be suspended at any time including when one or
1571 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1572 	 * handler while we are waiting to prevent a deadlock.
1573 	 */
1574 	vcpu_lock(vcpu);
1575 	while (error == 0) {
1576 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1577 			VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1578 			break;
1579 		}
1580 
1581 		if (vm->rendezvous_func == NULL) {
1582 			VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1583 			vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1584 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1585 			vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1586 			if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1587 				vcpu_unlock(vcpu);
1588 				error = thread_check_susp(td, false);
1589 				vcpu_lock(vcpu);
1590 			}
1591 		} else {
1592 			VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1593 			vcpu_unlock(vcpu);
1594 			error = vm_handle_rendezvous(vm, vcpuid);
1595 			vcpu_lock(vcpu);
1596 		}
1597 	}
1598 	vcpu_unlock(vcpu);
1599 
1600 	/*
1601 	 * Wakeup the other sleeping vcpus and return to userspace.
1602 	 */
1603 	for (i = 0; i < vm->maxcpus; i++) {
1604 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1605 			vcpu_notify_event(vm, i, false);
1606 		}
1607 	}
1608 
1609 	*retu = true;
1610 	return (error);
1611 }
1612 
1613 static int
1614 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu)
1615 {
1616 	struct vcpu *vcpu = &vm->vcpu[vcpuid];
1617 
1618 	vcpu_lock(vcpu);
1619 	KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle));
1620 	vcpu->reqidle = 0;
1621 	vcpu_unlock(vcpu);
1622 	*retu = true;
1623 	return (0);
1624 }
1625 
1626 int
1627 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1628 {
1629 	int i;
1630 
1631 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1632 		return (EINVAL);
1633 
1634 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1635 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1636 		    vm->suspend, how);
1637 		return (EALREADY);
1638 	}
1639 
1640 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1641 
1642 	/*
1643 	 * Notify all active vcpus that they are now suspended.
1644 	 */
1645 	for (i = 0; i < vm->maxcpus; i++) {
1646 		if (CPU_ISSET(i, &vm->active_cpus))
1647 			vcpu_notify_event(vm, i, false);
1648 	}
1649 
1650 	return (0);
1651 }
1652 
1653 void
1654 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1655 {
1656 	struct vm_exit *vmexit;
1657 
1658 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1659 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1660 
1661 	vmexit = vm_exitinfo(vm, vcpuid);
1662 	vmexit->rip = rip;
1663 	vmexit->inst_length = 0;
1664 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1665 	vmexit->u.suspended.how = vm->suspend;
1666 }
1667 
1668 void
1669 vm_exit_debug(struct vm *vm, int vcpuid, uint64_t rip)
1670 {
1671 	struct vm_exit *vmexit;
1672 
1673 	vmexit = vm_exitinfo(vm, vcpuid);
1674 	vmexit->rip = rip;
1675 	vmexit->inst_length = 0;
1676 	vmexit->exitcode = VM_EXITCODE_DEBUG;
1677 }
1678 
1679 void
1680 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1681 {
1682 	struct vm_exit *vmexit;
1683 
1684 	KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1685 
1686 	vmexit = vm_exitinfo(vm, vcpuid);
1687 	vmexit->rip = rip;
1688 	vmexit->inst_length = 0;
1689 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1690 	vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1691 }
1692 
1693 void
1694 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip)
1695 {
1696 	struct vm_exit *vmexit;
1697 
1698 	vmexit = vm_exitinfo(vm, vcpuid);
1699 	vmexit->rip = rip;
1700 	vmexit->inst_length = 0;
1701 	vmexit->exitcode = VM_EXITCODE_REQIDLE;
1702 	vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1);
1703 }
1704 
1705 void
1706 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1707 {
1708 	struct vm_exit *vmexit;
1709 
1710 	vmexit = vm_exitinfo(vm, vcpuid);
1711 	vmexit->rip = rip;
1712 	vmexit->inst_length = 0;
1713 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1714 	vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1715 }
1716 
1717 int
1718 vm_run(struct vm *vm, struct vm_run *vmrun)
1719 {
1720 	struct vm_eventinfo evinfo;
1721 	int error, vcpuid;
1722 	struct vcpu *vcpu;
1723 	struct pcb *pcb;
1724 	uint64_t tscval;
1725 	struct vm_exit *vme;
1726 	bool retu, intr_disabled;
1727 	pmap_t pmap;
1728 
1729 	vcpuid = vmrun->cpuid;
1730 
1731 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1732 		return (EINVAL);
1733 
1734 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1735 		return (EINVAL);
1736 
1737 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1738 		return (EINVAL);
1739 
1740 	pmap = vmspace_pmap(vm->vmspace);
1741 	vcpu = &vm->vcpu[vcpuid];
1742 	vme = &vcpu->exitinfo;
1743 	evinfo.rptr = &vm->rendezvous_func;
1744 	evinfo.sptr = &vm->suspend;
1745 	evinfo.iptr = &vcpu->reqidle;
1746 restart:
1747 	critical_enter();
1748 
1749 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1750 	    ("vm_run: absurd pm_active"));
1751 
1752 	tscval = rdtsc();
1753 
1754 	pcb = PCPU_GET(curpcb);
1755 	set_pcb_flags(pcb, PCB_FULL_IRET);
1756 
1757 	restore_guest_fpustate(vcpu);
1758 
1759 	vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1760 	error = vmmops_run(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo);
1761 	vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1762 
1763 	save_guest_fpustate(vcpu);
1764 
1765 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1766 
1767 	critical_exit();
1768 
1769 	if (error == 0) {
1770 		retu = false;
1771 		vcpu->nextrip = vme->rip + vme->inst_length;
1772 		switch (vme->exitcode) {
1773 		case VM_EXITCODE_REQIDLE:
1774 			error = vm_handle_reqidle(vm, vcpuid, &retu);
1775 			break;
1776 		case VM_EXITCODE_SUSPENDED:
1777 			error = vm_handle_suspend(vm, vcpuid, &retu);
1778 			break;
1779 		case VM_EXITCODE_IOAPIC_EOI:
1780 			vioapic_process_eoi(vm, vcpuid,
1781 			    vme->u.ioapic_eoi.vector);
1782 			break;
1783 		case VM_EXITCODE_RENDEZVOUS:
1784 			error = vm_handle_rendezvous(vm, vcpuid);
1785 			break;
1786 		case VM_EXITCODE_HLT:
1787 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1788 			error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1789 			break;
1790 		case VM_EXITCODE_PAGING:
1791 			error = vm_handle_paging(vm, vcpuid, &retu);
1792 			break;
1793 		case VM_EXITCODE_INST_EMUL:
1794 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1795 			break;
1796 		case VM_EXITCODE_INOUT:
1797 		case VM_EXITCODE_INOUT_STR:
1798 			error = vm_handle_inout(vm, vcpuid, vme, &retu);
1799 			break;
1800 		case VM_EXITCODE_MONITOR:
1801 		case VM_EXITCODE_MWAIT:
1802 		case VM_EXITCODE_VMINSN:
1803 			vm_inject_ud(vm, vcpuid);
1804 			break;
1805 		default:
1806 			retu = true;	/* handled in userland */
1807 			break;
1808 		}
1809 	}
1810 
1811 	if (error == 0 && retu == false)
1812 		goto restart;
1813 
1814 	VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode);
1815 
1816 	/* copy the exit information */
1817 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1818 	return (error);
1819 }
1820 
1821 int
1822 vm_restart_instruction(void *arg, int vcpuid)
1823 {
1824 	struct vm *vm;
1825 	struct vcpu *vcpu;
1826 	enum vcpu_state state;
1827 	uint64_t rip;
1828 	int error __diagused;
1829 
1830 	vm = arg;
1831 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1832 		return (EINVAL);
1833 
1834 	vcpu = &vm->vcpu[vcpuid];
1835 	state = vcpu_get_state(vm, vcpuid, NULL);
1836 	if (state == VCPU_RUNNING) {
1837 		/*
1838 		 * When a vcpu is "running" the next instruction is determined
1839 		 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
1840 		 * Thus setting 'inst_length' to zero will cause the current
1841 		 * instruction to be restarted.
1842 		 */
1843 		vcpu->exitinfo.inst_length = 0;
1844 		VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by "
1845 		    "setting inst_length to zero", vcpu->exitinfo.rip);
1846 	} else if (state == VCPU_FROZEN) {
1847 		/*
1848 		 * When a vcpu is "frozen" it is outside the critical section
1849 		 * around vmmops_run() and 'nextrip' points to the next
1850 		 * instruction. Thus instruction restart is achieved by setting
1851 		 * 'nextrip' to the vcpu's %rip.
1852 		 */
1853 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip);
1854 		KASSERT(!error, ("%s: error %d getting rip", __func__, error));
1855 		VCPU_CTR2(vm, vcpuid, "restarting instruction by updating "
1856 		    "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
1857 		vcpu->nextrip = rip;
1858 	} else {
1859 		panic("%s: invalid state %d", __func__, state);
1860 	}
1861 	return (0);
1862 }
1863 
1864 int
1865 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1866 {
1867 	struct vcpu *vcpu;
1868 	int type, vector;
1869 
1870 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1871 		return (EINVAL);
1872 
1873 	vcpu = &vm->vcpu[vcpuid];
1874 
1875 	if (info & VM_INTINFO_VALID) {
1876 		type = info & VM_INTINFO_TYPE;
1877 		vector = info & 0xff;
1878 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1879 			return (EINVAL);
1880 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1881 			return (EINVAL);
1882 		if (info & VM_INTINFO_RSVD)
1883 			return (EINVAL);
1884 	} else {
1885 		info = 0;
1886 	}
1887 	VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1888 	vcpu->exitintinfo = info;
1889 	return (0);
1890 }
1891 
1892 enum exc_class {
1893 	EXC_BENIGN,
1894 	EXC_CONTRIBUTORY,
1895 	EXC_PAGEFAULT
1896 };
1897 
1898 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
1899 
1900 static enum exc_class
1901 exception_class(uint64_t info)
1902 {
1903 	int type, vector;
1904 
1905 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1906 	type = info & VM_INTINFO_TYPE;
1907 	vector = info & 0xff;
1908 
1909 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1910 	switch (type) {
1911 	case VM_INTINFO_HWINTR:
1912 	case VM_INTINFO_SWINTR:
1913 	case VM_INTINFO_NMI:
1914 		return (EXC_BENIGN);
1915 	default:
1916 		/*
1917 		 * Hardware exception.
1918 		 *
1919 		 * SVM and VT-x use identical type values to represent NMI,
1920 		 * hardware interrupt and software interrupt.
1921 		 *
1922 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1923 		 * for exceptions except #BP and #OF. #BP and #OF use a type
1924 		 * value of '5' or '6'. Therefore we don't check for explicit
1925 		 * values of 'type' to classify 'intinfo' into a hardware
1926 		 * exception.
1927 		 */
1928 		break;
1929 	}
1930 
1931 	switch (vector) {
1932 	case IDT_PF:
1933 	case IDT_VE:
1934 		return (EXC_PAGEFAULT);
1935 	case IDT_DE:
1936 	case IDT_TS:
1937 	case IDT_NP:
1938 	case IDT_SS:
1939 	case IDT_GP:
1940 		return (EXC_CONTRIBUTORY);
1941 	default:
1942 		return (EXC_BENIGN);
1943 	}
1944 }
1945 
1946 static int
1947 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1948     uint64_t *retinfo)
1949 {
1950 	enum exc_class exc1, exc2;
1951 	int type1, vector1;
1952 
1953 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1954 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1955 
1956 	/*
1957 	 * If an exception occurs while attempting to call the double-fault
1958 	 * handler the processor enters shutdown mode (aka triple fault).
1959 	 */
1960 	type1 = info1 & VM_INTINFO_TYPE;
1961 	vector1 = info1 & 0xff;
1962 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1963 		VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1964 		    info1, info2);
1965 		vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1966 		*retinfo = 0;
1967 		return (0);
1968 	}
1969 
1970 	/*
1971 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1972 	 */
1973 	exc1 = exception_class(info1);
1974 	exc2 = exception_class(info2);
1975 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1976 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1977 		/* Convert nested fault into a double fault. */
1978 		*retinfo = IDT_DF;
1979 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1980 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
1981 	} else {
1982 		/* Handle exceptions serially */
1983 		*retinfo = info2;
1984 	}
1985 	return (1);
1986 }
1987 
1988 static uint64_t
1989 vcpu_exception_intinfo(struct vcpu *vcpu)
1990 {
1991 	uint64_t info = 0;
1992 
1993 	if (vcpu->exception_pending) {
1994 		info = vcpu->exc_vector & 0xff;
1995 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1996 		if (vcpu->exc_errcode_valid) {
1997 			info |= VM_INTINFO_DEL_ERRCODE;
1998 			info |= (uint64_t)vcpu->exc_errcode << 32;
1999 		}
2000 	}
2001 	return (info);
2002 }
2003 
2004 int
2005 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
2006 {
2007 	struct vcpu *vcpu;
2008 	uint64_t info1, info2;
2009 	int valid;
2010 
2011 	KASSERT(vcpuid >= 0 &&
2012 	    vcpuid < vm->maxcpus, ("invalid vcpu %d", vcpuid));
2013 
2014 	vcpu = &vm->vcpu[vcpuid];
2015 
2016 	info1 = vcpu->exitintinfo;
2017 	vcpu->exitintinfo = 0;
2018 
2019 	info2 = 0;
2020 	if (vcpu->exception_pending) {
2021 		info2 = vcpu_exception_intinfo(vcpu);
2022 		vcpu->exception_pending = 0;
2023 		VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
2024 		    vcpu->exc_vector, info2);
2025 	}
2026 
2027 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
2028 		valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
2029 	} else if (info1 & VM_INTINFO_VALID) {
2030 		*retinfo = info1;
2031 		valid = 1;
2032 	} else if (info2 & VM_INTINFO_VALID) {
2033 		*retinfo = info2;
2034 		valid = 1;
2035 	} else {
2036 		valid = 0;
2037 	}
2038 
2039 	if (valid) {
2040 		VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
2041 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
2042 	}
2043 
2044 	return (valid);
2045 }
2046 
2047 int
2048 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
2049 {
2050 	struct vcpu *vcpu;
2051 
2052 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2053 		return (EINVAL);
2054 
2055 	vcpu = &vm->vcpu[vcpuid];
2056 	*info1 = vcpu->exitintinfo;
2057 	*info2 = vcpu_exception_intinfo(vcpu);
2058 	return (0);
2059 }
2060 
2061 int
2062 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid,
2063     uint32_t errcode, int restart_instruction)
2064 {
2065 	struct vcpu *vcpu;
2066 	uint64_t regval;
2067 	int error __diagused;
2068 
2069 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2070 		return (EINVAL);
2071 
2072 	if (vector < 0 || vector >= 32)
2073 		return (EINVAL);
2074 
2075 	/*
2076 	 * A double fault exception should never be injected directly into
2077 	 * the guest. It is a derived exception that results from specific
2078 	 * combinations of nested faults.
2079 	 */
2080 	if (vector == IDT_DF)
2081 		return (EINVAL);
2082 
2083 	vcpu = &vm->vcpu[vcpuid];
2084 
2085 	if (vcpu->exception_pending) {
2086 		VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
2087 		    "pending exception %d", vector, vcpu->exc_vector);
2088 		return (EBUSY);
2089 	}
2090 
2091 	if (errcode_valid) {
2092 		/*
2093 		 * Exceptions don't deliver an error code in real mode.
2094 		 */
2095 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, &regval);
2096 		KASSERT(!error, ("%s: error %d getting CR0", __func__, error));
2097 		if (!(regval & CR0_PE))
2098 			errcode_valid = 0;
2099 	}
2100 
2101 	/*
2102 	 * From section 26.6.1 "Interruptibility State" in Intel SDM:
2103 	 *
2104 	 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
2105 	 * one instruction or incurs an exception.
2106 	 */
2107 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
2108 	KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
2109 	    __func__, error));
2110 
2111 	if (restart_instruction)
2112 		vm_restart_instruction(vm, vcpuid);
2113 
2114 	vcpu->exception_pending = 1;
2115 	vcpu->exc_vector = vector;
2116 	vcpu->exc_errcode = errcode;
2117 	vcpu->exc_errcode_valid = errcode_valid;
2118 	VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector);
2119 	return (0);
2120 }
2121 
2122 void
2123 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
2124     int errcode)
2125 {
2126 	struct vm *vm;
2127 	int error __diagused, restart_instruction;
2128 
2129 	vm = vmarg;
2130 	restart_instruction = 1;
2131 
2132 	error = vm_inject_exception(vm, vcpuid, vector, errcode_valid,
2133 	    errcode, restart_instruction);
2134 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
2135 }
2136 
2137 void
2138 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
2139 {
2140 	struct vm *vm;
2141 	int error __diagused;
2142 
2143 	vm = vmarg;
2144 	VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
2145 	    error_code, cr2);
2146 
2147 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
2148 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
2149 
2150 	vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
2151 }
2152 
2153 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
2154 
2155 int
2156 vm_inject_nmi(struct vm *vm, int vcpuid)
2157 {
2158 	struct vcpu *vcpu;
2159 
2160 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2161 		return (EINVAL);
2162 
2163 	vcpu = &vm->vcpu[vcpuid];
2164 
2165 	vcpu->nmi_pending = 1;
2166 	vcpu_notify_event(vm, vcpuid, false);
2167 	return (0);
2168 }
2169 
2170 int
2171 vm_nmi_pending(struct vm *vm, int vcpuid)
2172 {
2173 	struct vcpu *vcpu;
2174 
2175 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2176 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2177 
2178 	vcpu = &vm->vcpu[vcpuid];
2179 
2180 	return (vcpu->nmi_pending);
2181 }
2182 
2183 void
2184 vm_nmi_clear(struct vm *vm, int vcpuid)
2185 {
2186 	struct vcpu *vcpu;
2187 
2188 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2189 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2190 
2191 	vcpu = &vm->vcpu[vcpuid];
2192 
2193 	if (vcpu->nmi_pending == 0)
2194 		panic("vm_nmi_clear: inconsistent nmi_pending state");
2195 
2196 	vcpu->nmi_pending = 0;
2197 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
2198 }
2199 
2200 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
2201 
2202 int
2203 vm_inject_extint(struct vm *vm, int vcpuid)
2204 {
2205 	struct vcpu *vcpu;
2206 
2207 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2208 		return (EINVAL);
2209 
2210 	vcpu = &vm->vcpu[vcpuid];
2211 
2212 	vcpu->extint_pending = 1;
2213 	vcpu_notify_event(vm, vcpuid, false);
2214 	return (0);
2215 }
2216 
2217 int
2218 vm_extint_pending(struct vm *vm, int vcpuid)
2219 {
2220 	struct vcpu *vcpu;
2221 
2222 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2223 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2224 
2225 	vcpu = &vm->vcpu[vcpuid];
2226 
2227 	return (vcpu->extint_pending);
2228 }
2229 
2230 void
2231 vm_extint_clear(struct vm *vm, int vcpuid)
2232 {
2233 	struct vcpu *vcpu;
2234 
2235 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2236 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2237 
2238 	vcpu = &vm->vcpu[vcpuid];
2239 
2240 	if (vcpu->extint_pending == 0)
2241 		panic("vm_extint_clear: inconsistent extint_pending state");
2242 
2243 	vcpu->extint_pending = 0;
2244 	vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
2245 }
2246 
2247 int
2248 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
2249 {
2250 	if (vcpu < 0 || vcpu >= vm->maxcpus)
2251 		return (EINVAL);
2252 
2253 	if (type < 0 || type >= VM_CAP_MAX)
2254 		return (EINVAL);
2255 
2256 	return (vmmops_getcap(vm->cookie, vcpu, type, retval));
2257 }
2258 
2259 int
2260 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
2261 {
2262 	if (vcpu < 0 || vcpu >= vm->maxcpus)
2263 		return (EINVAL);
2264 
2265 	if (type < 0 || type >= VM_CAP_MAX)
2266 		return (EINVAL);
2267 
2268 	return (vmmops_setcap(vm->cookie, vcpu, type, val));
2269 }
2270 
2271 struct vlapic *
2272 vm_lapic(struct vm *vm, int cpu)
2273 {
2274 	return (vm->vcpu[cpu].vlapic);
2275 }
2276 
2277 struct vioapic *
2278 vm_ioapic(struct vm *vm)
2279 {
2280 
2281 	return (vm->vioapic);
2282 }
2283 
2284 struct vhpet *
2285 vm_hpet(struct vm *vm)
2286 {
2287 
2288 	return (vm->vhpet);
2289 }
2290 
2291 bool
2292 vmm_is_pptdev(int bus, int slot, int func)
2293 {
2294 	int b, f, i, n, s;
2295 	char *val, *cp, *cp2;
2296 	bool found;
2297 
2298 	/*
2299 	 * XXX
2300 	 * The length of an environment variable is limited to 128 bytes which
2301 	 * puts an upper limit on the number of passthru devices that may be
2302 	 * specified using a single environment variable.
2303 	 *
2304 	 * Work around this by scanning multiple environment variable
2305 	 * names instead of a single one - yuck!
2306 	 */
2307 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2308 
2309 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2310 	found = false;
2311 	for (i = 0; names[i] != NULL && !found; i++) {
2312 		cp = val = kern_getenv(names[i]);
2313 		while (cp != NULL && *cp != '\0') {
2314 			if ((cp2 = strchr(cp, ' ')) != NULL)
2315 				*cp2 = '\0';
2316 
2317 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2318 			if (n == 3 && bus == b && slot == s && func == f) {
2319 				found = true;
2320 				break;
2321 			}
2322 
2323 			if (cp2 != NULL)
2324 				*cp2++ = ' ';
2325 
2326 			cp = cp2;
2327 		}
2328 		freeenv(val);
2329 	}
2330 	return (found);
2331 }
2332 
2333 void *
2334 vm_iommu_domain(struct vm *vm)
2335 {
2336 
2337 	return (vm->iommu);
2338 }
2339 
2340 int
2341 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
2342     bool from_idle)
2343 {
2344 	int error;
2345 	struct vcpu *vcpu;
2346 
2347 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2348 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
2349 
2350 	vcpu = &vm->vcpu[vcpuid];
2351 
2352 	vcpu_lock(vcpu);
2353 	error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle);
2354 	vcpu_unlock(vcpu);
2355 
2356 	return (error);
2357 }
2358 
2359 enum vcpu_state
2360 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2361 {
2362 	struct vcpu *vcpu;
2363 	enum vcpu_state state;
2364 
2365 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2366 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2367 
2368 	vcpu = &vm->vcpu[vcpuid];
2369 
2370 	vcpu_lock(vcpu);
2371 	state = vcpu->state;
2372 	if (hostcpu != NULL)
2373 		*hostcpu = vcpu->hostcpu;
2374 	vcpu_unlock(vcpu);
2375 
2376 	return (state);
2377 }
2378 
2379 int
2380 vm_activate_cpu(struct vm *vm, int vcpuid)
2381 {
2382 
2383 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2384 		return (EINVAL);
2385 
2386 	if (CPU_ISSET(vcpuid, &vm->active_cpus))
2387 		return (EBUSY);
2388 
2389 	VCPU_CTR0(vm, vcpuid, "activated");
2390 	CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2391 	return (0);
2392 }
2393 
2394 int
2395 vm_suspend_cpu(struct vm *vm, int vcpuid)
2396 {
2397 	int i;
2398 
2399 	if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2400 		return (EINVAL);
2401 
2402 	if (vcpuid == -1) {
2403 		vm->debug_cpus = vm->active_cpus;
2404 		for (i = 0; i < vm->maxcpus; i++) {
2405 			if (CPU_ISSET(i, &vm->active_cpus))
2406 				vcpu_notify_event(vm, i, false);
2407 		}
2408 	} else {
2409 		if (!CPU_ISSET(vcpuid, &vm->active_cpus))
2410 			return (EINVAL);
2411 
2412 		CPU_SET_ATOMIC(vcpuid, &vm->debug_cpus);
2413 		vcpu_notify_event(vm, vcpuid, false);
2414 	}
2415 	return (0);
2416 }
2417 
2418 int
2419 vm_resume_cpu(struct vm *vm, int vcpuid)
2420 {
2421 
2422 	if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2423 		return (EINVAL);
2424 
2425 	if (vcpuid == -1) {
2426 		CPU_ZERO(&vm->debug_cpus);
2427 	} else {
2428 		if (!CPU_ISSET(vcpuid, &vm->debug_cpus))
2429 			return (EINVAL);
2430 
2431 		CPU_CLR_ATOMIC(vcpuid, &vm->debug_cpus);
2432 	}
2433 	return (0);
2434 }
2435 
2436 int
2437 vcpu_debugged(struct vm *vm, int vcpuid)
2438 {
2439 
2440 	return (CPU_ISSET(vcpuid, &vm->debug_cpus));
2441 }
2442 
2443 cpuset_t
2444 vm_active_cpus(struct vm *vm)
2445 {
2446 
2447 	return (vm->active_cpus);
2448 }
2449 
2450 cpuset_t
2451 vm_debug_cpus(struct vm *vm)
2452 {
2453 
2454 	return (vm->debug_cpus);
2455 }
2456 
2457 cpuset_t
2458 vm_suspended_cpus(struct vm *vm)
2459 {
2460 
2461 	return (vm->suspended_cpus);
2462 }
2463 
2464 void *
2465 vcpu_stats(struct vm *vm, int vcpuid)
2466 {
2467 
2468 	return (vm->vcpu[vcpuid].stats);
2469 }
2470 
2471 int
2472 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2473 {
2474 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2475 		return (EINVAL);
2476 
2477 	*state = vm->vcpu[vcpuid].x2apic_state;
2478 
2479 	return (0);
2480 }
2481 
2482 int
2483 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2484 {
2485 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2486 		return (EINVAL);
2487 
2488 	if (state >= X2APIC_STATE_LAST)
2489 		return (EINVAL);
2490 
2491 	vm->vcpu[vcpuid].x2apic_state = state;
2492 
2493 	vlapic_set_x2apic_state(vm, vcpuid, state);
2494 
2495 	return (0);
2496 }
2497 
2498 /*
2499  * This function is called to ensure that a vcpu "sees" a pending event
2500  * as soon as possible:
2501  * - If the vcpu thread is sleeping then it is woken up.
2502  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2503  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2504  */
2505 static void
2506 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
2507 {
2508 	int hostcpu;
2509 
2510 	hostcpu = vcpu->hostcpu;
2511 	if (vcpu->state == VCPU_RUNNING) {
2512 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2513 		if (hostcpu != curcpu) {
2514 			if (lapic_intr) {
2515 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2516 				    vmm_ipinum);
2517 			} else {
2518 				ipi_cpu(hostcpu, vmm_ipinum);
2519 			}
2520 		} else {
2521 			/*
2522 			 * If the 'vcpu' is running on 'curcpu' then it must
2523 			 * be sending a notification to itself (e.g. SELF_IPI).
2524 			 * The pending event will be picked up when the vcpu
2525 			 * transitions back to guest context.
2526 			 */
2527 		}
2528 	} else {
2529 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2530 		    "with hostcpu %d", vcpu->state, hostcpu));
2531 		if (vcpu->state == VCPU_SLEEPING)
2532 			wakeup_one(vcpu);
2533 	}
2534 }
2535 
2536 void
2537 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2538 {
2539 	struct vcpu *vcpu = &vm->vcpu[vcpuid];
2540 
2541 	vcpu_lock(vcpu);
2542 	vcpu_notify_event_locked(vcpu, lapic_intr);
2543 	vcpu_unlock(vcpu);
2544 }
2545 
2546 struct vmspace *
2547 vm_get_vmspace(struct vm *vm)
2548 {
2549 
2550 	return (vm->vmspace);
2551 }
2552 
2553 int
2554 vm_apicid2vcpuid(struct vm *vm, int apicid)
2555 {
2556 	/*
2557 	 * XXX apic id is assumed to be numerically identical to vcpu id
2558 	 */
2559 	return (apicid);
2560 }
2561 
2562 int
2563 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2564     vm_rendezvous_func_t func, void *arg)
2565 {
2566 	int error, i;
2567 
2568 	/*
2569 	 * Enforce that this function is called without any locks
2570 	 */
2571 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2572 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
2573 	    ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2574 
2575 restart:
2576 	mtx_lock(&vm->rendezvous_mtx);
2577 	if (vm->rendezvous_func != NULL) {
2578 		/*
2579 		 * If a rendezvous is already in progress then we need to
2580 		 * call the rendezvous handler in case this 'vcpuid' is one
2581 		 * of the targets of the rendezvous.
2582 		 */
2583 		RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2584 		mtx_unlock(&vm->rendezvous_mtx);
2585 		error = vm_handle_rendezvous(vm, vcpuid);
2586 		if (error != 0)
2587 			return (error);
2588 		goto restart;
2589 	}
2590 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2591 	    "rendezvous is still in progress"));
2592 
2593 	RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2594 	vm->rendezvous_req_cpus = dest;
2595 	CPU_ZERO(&vm->rendezvous_done_cpus);
2596 	vm->rendezvous_arg = arg;
2597 	vm->rendezvous_func = func;
2598 	mtx_unlock(&vm->rendezvous_mtx);
2599 
2600 	/*
2601 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2602 	 * vcpus so they handle the rendezvous as soon as possible.
2603 	 */
2604 	for (i = 0; i < vm->maxcpus; i++) {
2605 		if (CPU_ISSET(i, &dest))
2606 			vcpu_notify_event(vm, i, false);
2607 	}
2608 
2609 	return (vm_handle_rendezvous(vm, vcpuid));
2610 }
2611 
2612 struct vatpic *
2613 vm_atpic(struct vm *vm)
2614 {
2615 	return (vm->vatpic);
2616 }
2617 
2618 struct vatpit *
2619 vm_atpit(struct vm *vm)
2620 {
2621 	return (vm->vatpit);
2622 }
2623 
2624 struct vpmtmr *
2625 vm_pmtmr(struct vm *vm)
2626 {
2627 
2628 	return (vm->vpmtmr);
2629 }
2630 
2631 struct vrtc *
2632 vm_rtc(struct vm *vm)
2633 {
2634 
2635 	return (vm->vrtc);
2636 }
2637 
2638 enum vm_reg_name
2639 vm_segment_name(int seg)
2640 {
2641 	static enum vm_reg_name seg_names[] = {
2642 		VM_REG_GUEST_ES,
2643 		VM_REG_GUEST_CS,
2644 		VM_REG_GUEST_SS,
2645 		VM_REG_GUEST_DS,
2646 		VM_REG_GUEST_FS,
2647 		VM_REG_GUEST_GS
2648 	};
2649 
2650 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2651 	    ("%s: invalid segment encoding %d", __func__, seg));
2652 	return (seg_names[seg]);
2653 }
2654 
2655 void
2656 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2657     int num_copyinfo)
2658 {
2659 	int idx;
2660 
2661 	for (idx = 0; idx < num_copyinfo; idx++) {
2662 		if (copyinfo[idx].cookie != NULL)
2663 			vm_gpa_release(copyinfo[idx].cookie);
2664 	}
2665 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2666 }
2667 
2668 int
2669 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2670     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2671     int num_copyinfo, int *fault)
2672 {
2673 	int error, idx, nused;
2674 	size_t n, off, remaining;
2675 	void *hva, *cookie;
2676 	uint64_t gpa;
2677 
2678 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2679 
2680 	nused = 0;
2681 	remaining = len;
2682 	while (remaining > 0) {
2683 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2684 		error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault);
2685 		if (error || *fault)
2686 			return (error);
2687 		off = gpa & PAGE_MASK;
2688 		n = min(remaining, PAGE_SIZE - off);
2689 		copyinfo[nused].gpa = gpa;
2690 		copyinfo[nused].len = n;
2691 		remaining -= n;
2692 		gla += n;
2693 		nused++;
2694 	}
2695 
2696 	for (idx = 0; idx < nused; idx++) {
2697 		hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa,
2698 		    copyinfo[idx].len, prot, &cookie);
2699 		if (hva == NULL)
2700 			break;
2701 		copyinfo[idx].hva = hva;
2702 		copyinfo[idx].cookie = cookie;
2703 	}
2704 
2705 	if (idx != nused) {
2706 		vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2707 		return (EFAULT);
2708 	} else {
2709 		*fault = 0;
2710 		return (0);
2711 	}
2712 }
2713 
2714 void
2715 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2716     size_t len)
2717 {
2718 	char *dst;
2719 	int idx;
2720 
2721 	dst = kaddr;
2722 	idx = 0;
2723 	while (len > 0) {
2724 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2725 		len -= copyinfo[idx].len;
2726 		dst += copyinfo[idx].len;
2727 		idx++;
2728 	}
2729 }
2730 
2731 void
2732 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2733     struct vm_copyinfo *copyinfo, size_t len)
2734 {
2735 	const char *src;
2736 	int idx;
2737 
2738 	src = kaddr;
2739 	idx = 0;
2740 	while (len > 0) {
2741 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2742 		len -= copyinfo[idx].len;
2743 		src += copyinfo[idx].len;
2744 		idx++;
2745 	}
2746 }
2747 
2748 /*
2749  * Return the amount of in-use and wired memory for the VM. Since
2750  * these are global stats, only return the values with for vCPU 0
2751  */
2752 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2753 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2754 
2755 static void
2756 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2757 {
2758 
2759 	if (vcpu == 0) {
2760 		vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2761 	       	    PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2762 	}
2763 }
2764 
2765 static void
2766 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2767 {
2768 
2769 	if (vcpu == 0) {
2770 		vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2771 	      	    PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2772 	}
2773 }
2774 
2775 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2776 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2777 
2778 #ifdef BHYVE_SNAPSHOT
2779 static int
2780 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta)
2781 {
2782 	int ret;
2783 	int i;
2784 	struct vcpu *vcpu;
2785 
2786 	for (i = 0; i < VM_MAXCPU; i++) {
2787 		vcpu = &vm->vcpu[i];
2788 
2789 		SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done);
2790 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done);
2791 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done);
2792 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done);
2793 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done);
2794 		SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done);
2795 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done);
2796 		SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done);
2797 		/* XXX we're cheating here, since the value of tsc_offset as
2798 		 * saved here is actually the value of the guest's TSC value.
2799 		 *
2800 		 * It will be turned turned back into an actual offset when the
2801 		 * TSC restore function is called
2802 		 */
2803 		SNAPSHOT_VAR_OR_LEAVE(vcpu->tsc_offset, meta, ret, done);
2804 	}
2805 
2806 done:
2807 	return (ret);
2808 }
2809 
2810 static int
2811 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta)
2812 {
2813 	int ret;
2814 	int i;
2815 	uint64_t now;
2816 
2817 	ret = 0;
2818 	now = rdtsc();
2819 
2820 	if (meta->op == VM_SNAPSHOT_SAVE) {
2821 		/* XXX make tsc_offset take the value TSC proper as seen by the
2822 		 * guest
2823 		 */
2824 		for (i = 0; i < VM_MAXCPU; i++)
2825 			vm->vcpu[i].tsc_offset += now;
2826 	}
2827 
2828 	ret = vm_snapshot_vcpus(vm, meta);
2829 	if (ret != 0) {
2830 		printf("%s: failed to copy vm data to user buffer", __func__);
2831 		goto done;
2832 	}
2833 
2834 	if (meta->op == VM_SNAPSHOT_SAVE) {
2835 		/* XXX turn tsc_offset back into an offset; actual value is only
2836 		 * required for restore; using it otherwise would be wrong
2837 		 */
2838 		for (i = 0; i < VM_MAXCPU; i++)
2839 			vm->vcpu[i].tsc_offset -= now;
2840 	}
2841 
2842 done:
2843 	return (ret);
2844 }
2845 
2846 static int
2847 vm_snapshot_vmcx(struct vm *vm, struct vm_snapshot_meta *meta)
2848 {
2849 	int i, error;
2850 
2851 	error = 0;
2852 
2853 	for (i = 0; i < VM_MAXCPU; i++) {
2854 		error = vmmops_vmcx_snapshot(vm->cookie, meta, i);
2855 		if (error != 0) {
2856 			printf("%s: failed to snapshot vmcs/vmcb data for "
2857 			       "vCPU: %d; error: %d\n", __func__, i, error);
2858 			goto done;
2859 		}
2860 	}
2861 
2862 done:
2863 	return (error);
2864 }
2865 
2866 /*
2867  * Save kernel-side structures to user-space for snapshotting.
2868  */
2869 int
2870 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta)
2871 {
2872 	int ret = 0;
2873 
2874 	switch (meta->dev_req) {
2875 	case STRUCT_VMX:
2876 		ret = vmmops_snapshot(vm->cookie, meta);
2877 		break;
2878 	case STRUCT_VMCX:
2879 		ret = vm_snapshot_vmcx(vm, meta);
2880 		break;
2881 	case STRUCT_VM:
2882 		ret = vm_snapshot_vm(vm, meta);
2883 		break;
2884 	case STRUCT_VIOAPIC:
2885 		ret = vioapic_snapshot(vm_ioapic(vm), meta);
2886 		break;
2887 	case STRUCT_VLAPIC:
2888 		ret = vlapic_snapshot(vm, meta);
2889 		break;
2890 	case STRUCT_VHPET:
2891 		ret = vhpet_snapshot(vm_hpet(vm), meta);
2892 		break;
2893 	case STRUCT_VATPIC:
2894 		ret = vatpic_snapshot(vm_atpic(vm), meta);
2895 		break;
2896 	case STRUCT_VATPIT:
2897 		ret = vatpit_snapshot(vm_atpit(vm), meta);
2898 		break;
2899 	case STRUCT_VPMTMR:
2900 		ret = vpmtmr_snapshot(vm_pmtmr(vm), meta);
2901 		break;
2902 	case STRUCT_VRTC:
2903 		ret = vrtc_snapshot(vm_rtc(vm), meta);
2904 		break;
2905 	default:
2906 		printf("%s: failed to find the requested type %#x\n",
2907 		       __func__, meta->dev_req);
2908 		ret = (EINVAL);
2909 	}
2910 	return (ret);
2911 }
2912 
2913 int
2914 vm_set_tsc_offset(struct vm *vm, int vcpuid, uint64_t offset)
2915 {
2916 	struct vcpu *vcpu;
2917 
2918 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2919 		return (EINVAL);
2920 
2921 	vcpu = &vm->vcpu[vcpuid];
2922 	vcpu->tsc_offset = offset;
2923 
2924 	return (0);
2925 }
2926 
2927 int
2928 vm_restore_time(struct vm *vm)
2929 {
2930 	int error, i;
2931 	uint64_t now;
2932 	struct vcpu *vcpu;
2933 
2934 	now = rdtsc();
2935 
2936 	error = vhpet_restore_time(vm_hpet(vm));
2937 	if (error)
2938 		return (error);
2939 
2940 	for (i = 0; i < nitems(vm->vcpu); i++) {
2941 		vcpu = &vm->vcpu[i];
2942 
2943 		error = vmmops_restore_tsc(vm->cookie, i, vcpu->tsc_offset -
2944 		    now);
2945 		if (error)
2946 			return (error);
2947 	}
2948 
2949 	return (0);
2950 }
2951 #endif
2952