xref: /freebsd/sys/amd64/vmm/vmm.c (revision 184c1b943937986c81e1996d999d21626ec7a4ff)
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	3
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];	/* (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 || strlen(name) >= VM_MAX_NAMELEN)
484 		return (EINVAL);
485 
486 	vmspace = vmmops_vmspace_alloc(0, VM_MAXUSER_ADDRESS_LA48);
487 	if (vmspace == NULL)
488 		return (ENOMEM);
489 
490 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
491 	strcpy(vm->name, name);
492 	vm->vmspace = vmspace;
493 	mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
494 
495 	vm->sockets = 1;
496 	vm->cores = cores_per_package;	/* XXX backwards compatibility */
497 	vm->threads = threads_per_core;	/* XXX backwards compatibility */
498 	vm->maxcpus = VM_MAXCPU;	/* XXX temp to keep code working */
499 
500 	vm_init(vm, true);
501 
502 	*retvm = vm;
503 	return (0);
504 }
505 
506 void
507 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
508     uint16_t *threads, uint16_t *maxcpus)
509 {
510 	*sockets = vm->sockets;
511 	*cores = vm->cores;
512 	*threads = vm->threads;
513 	*maxcpus = vm->maxcpus;
514 }
515 
516 uint16_t
517 vm_get_maxcpus(struct vm *vm)
518 {
519 	return (vm->maxcpus);
520 }
521 
522 int
523 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
524     uint16_t threads, uint16_t maxcpus)
525 {
526 	if (maxcpus != 0)
527 		return (EINVAL);	/* XXX remove when supported */
528 	if ((sockets * cores * threads) > vm->maxcpus)
529 		return (EINVAL);
530 	/* XXX need to check sockets * cores * threads == vCPU, how? */
531 	vm->sockets = sockets;
532 	vm->cores = cores;
533 	vm->threads = threads;
534 	vm->maxcpus = VM_MAXCPU;	/* XXX temp to keep code working */
535 	return(0);
536 }
537 
538 static void
539 vm_cleanup(struct vm *vm, bool destroy)
540 {
541 	struct mem_map *mm;
542 	int i;
543 
544 	ppt_unassign_all(vm);
545 
546 	if (vm->iommu != NULL)
547 		iommu_destroy_domain(vm->iommu);
548 
549 	if (destroy)
550 		vrtc_cleanup(vm->vrtc);
551 	else
552 		vrtc_reset(vm->vrtc);
553 	vpmtmr_cleanup(vm->vpmtmr);
554 	vatpit_cleanup(vm->vatpit);
555 	vhpet_cleanup(vm->vhpet);
556 	vatpic_cleanup(vm->vatpic);
557 	vioapic_cleanup(vm->vioapic);
558 
559 	for (i = 0; i < vm->maxcpus; i++)
560 		vcpu_cleanup(vm, i, destroy);
561 
562 	vmmops_cleanup(vm->cookie);
563 
564 	/*
565 	 * System memory is removed from the guest address space only when
566 	 * the VM is destroyed. This is because the mapping remains the same
567 	 * across VM reset.
568 	 *
569 	 * Device memory can be relocated by the guest (e.g. using PCI BARs)
570 	 * so those mappings are removed on a VM reset.
571 	 */
572 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
573 		mm = &vm->mem_maps[i];
574 		if (destroy || !sysmem_mapping(vm, mm))
575 			vm_free_memmap(vm, i);
576 	}
577 
578 	if (destroy) {
579 		for (i = 0; i < VM_MAX_MEMSEGS; i++)
580 			vm_free_memseg(vm, i);
581 
582 		vmmops_vmspace_free(vm->vmspace);
583 		vm->vmspace = NULL;
584 	}
585 }
586 
587 void
588 vm_destroy(struct vm *vm)
589 {
590 	vm_cleanup(vm, true);
591 	free(vm, M_VM);
592 }
593 
594 int
595 vm_reinit(struct vm *vm)
596 {
597 	int error;
598 
599 	/*
600 	 * A virtual machine can be reset only if all vcpus are suspended.
601 	 */
602 	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
603 		vm_cleanup(vm, false);
604 		vm_init(vm, false);
605 		error = 0;
606 	} else {
607 		error = EBUSY;
608 	}
609 
610 	return (error);
611 }
612 
613 const char *
614 vm_name(struct vm *vm)
615 {
616 	return (vm->name);
617 }
618 
619 int
620 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
621 {
622 	vm_object_t obj;
623 
624 	if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
625 		return (ENOMEM);
626 	else
627 		return (0);
628 }
629 
630 int
631 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
632 {
633 
634 	vmm_mmio_free(vm->vmspace, gpa, len);
635 	return (0);
636 }
637 
638 /*
639  * Return 'true' if 'gpa' is allocated in the guest address space.
640  *
641  * This function is called in the context of a running vcpu which acts as
642  * an implicit lock on 'vm->mem_maps[]'.
643  */
644 bool
645 vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa)
646 {
647 	struct mem_map *mm;
648 	int i;
649 
650 #ifdef INVARIANTS
651 	int hostcpu, state;
652 	state = vcpu_get_state(vm, vcpuid, &hostcpu);
653 	KASSERT(state == VCPU_RUNNING && hostcpu == curcpu,
654 	    ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu));
655 #endif
656 
657 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
658 		mm = &vm->mem_maps[i];
659 		if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len)
660 			return (true);		/* 'gpa' is sysmem or devmem */
661 	}
662 
663 	if (ppt_is_mmio(vm, gpa))
664 		return (true);			/* 'gpa' is pci passthru mmio */
665 
666 	return (false);
667 }
668 
669 int
670 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem)
671 {
672 	struct mem_seg *seg;
673 	vm_object_t obj;
674 
675 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
676 		return (EINVAL);
677 
678 	if (len == 0 || (len & PAGE_MASK))
679 		return (EINVAL);
680 
681 	seg = &vm->mem_segs[ident];
682 	if (seg->object != NULL) {
683 		if (seg->len == len && seg->sysmem == sysmem)
684 			return (EEXIST);
685 		else
686 			return (EINVAL);
687 	}
688 
689 	obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT);
690 	if (obj == NULL)
691 		return (ENOMEM);
692 
693 	seg->len = len;
694 	seg->object = obj;
695 	seg->sysmem = sysmem;
696 	return (0);
697 }
698 
699 int
700 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
701     vm_object_t *objptr)
702 {
703 	struct mem_seg *seg;
704 
705 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
706 		return (EINVAL);
707 
708 	seg = &vm->mem_segs[ident];
709 	if (len)
710 		*len = seg->len;
711 	if (sysmem)
712 		*sysmem = seg->sysmem;
713 	if (objptr)
714 		*objptr = seg->object;
715 	return (0);
716 }
717 
718 void
719 vm_free_memseg(struct vm *vm, int ident)
720 {
721 	struct mem_seg *seg;
722 
723 	KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS,
724 	    ("%s: invalid memseg ident %d", __func__, ident));
725 
726 	seg = &vm->mem_segs[ident];
727 	if (seg->object != NULL) {
728 		vm_object_deallocate(seg->object);
729 		bzero(seg, sizeof(struct mem_seg));
730 	}
731 }
732 
733 int
734 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first,
735     size_t len, int prot, int flags)
736 {
737 	struct mem_seg *seg;
738 	struct mem_map *m, *map;
739 	vm_ooffset_t last;
740 	int i, error;
741 
742 	if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0)
743 		return (EINVAL);
744 
745 	if (flags & ~VM_MEMMAP_F_WIRED)
746 		return (EINVAL);
747 
748 	if (segid < 0 || segid >= VM_MAX_MEMSEGS)
749 		return (EINVAL);
750 
751 	seg = &vm->mem_segs[segid];
752 	if (seg->object == NULL)
753 		return (EINVAL);
754 
755 	last = first + len;
756 	if (first < 0 || first >= last || last > seg->len)
757 		return (EINVAL);
758 
759 	if ((gpa | first | last) & PAGE_MASK)
760 		return (EINVAL);
761 
762 	map = NULL;
763 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
764 		m = &vm->mem_maps[i];
765 		if (m->len == 0) {
766 			map = m;
767 			break;
768 		}
769 	}
770 
771 	if (map == NULL)
772 		return (ENOSPC);
773 
774 	error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa,
775 	    len, 0, VMFS_NO_SPACE, prot, prot, 0);
776 	if (error != KERN_SUCCESS)
777 		return (EFAULT);
778 
779 	vm_object_reference(seg->object);
780 
781 	if (flags & VM_MEMMAP_F_WIRED) {
782 		error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len,
783 		    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
784 		if (error != KERN_SUCCESS) {
785 			vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len);
786 			return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM :
787 			    EFAULT);
788 		}
789 	}
790 
791 	map->gpa = gpa;
792 	map->len = len;
793 	map->segoff = first;
794 	map->segid = segid;
795 	map->prot = prot;
796 	map->flags = flags;
797 	return (0);
798 }
799 
800 int
801 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
802     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags)
803 {
804 	struct mem_map *mm, *mmnext;
805 	int i;
806 
807 	mmnext = NULL;
808 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
809 		mm = &vm->mem_maps[i];
810 		if (mm->len == 0 || mm->gpa < *gpa)
811 			continue;
812 		if (mmnext == NULL || mm->gpa < mmnext->gpa)
813 			mmnext = mm;
814 	}
815 
816 	if (mmnext != NULL) {
817 		*gpa = mmnext->gpa;
818 		if (segid)
819 			*segid = mmnext->segid;
820 		if (segoff)
821 			*segoff = mmnext->segoff;
822 		if (len)
823 			*len = mmnext->len;
824 		if (prot)
825 			*prot = mmnext->prot;
826 		if (flags)
827 			*flags = mmnext->flags;
828 		return (0);
829 	} else {
830 		return (ENOENT);
831 	}
832 }
833 
834 static void
835 vm_free_memmap(struct vm *vm, int ident)
836 {
837 	struct mem_map *mm;
838 	int error;
839 
840 	mm = &vm->mem_maps[ident];
841 	if (mm->len) {
842 		error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa,
843 		    mm->gpa + mm->len);
844 		KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d",
845 		    __func__, error));
846 		bzero(mm, sizeof(struct mem_map));
847 	}
848 }
849 
850 static __inline bool
851 sysmem_mapping(struct vm *vm, struct mem_map *mm)
852 {
853 
854 	if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem)
855 		return (true);
856 	else
857 		return (false);
858 }
859 
860 vm_paddr_t
861 vmm_sysmem_maxaddr(struct vm *vm)
862 {
863 	struct mem_map *mm;
864 	vm_paddr_t maxaddr;
865 	int i;
866 
867 	maxaddr = 0;
868 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
869 		mm = &vm->mem_maps[i];
870 		if (sysmem_mapping(vm, mm)) {
871 			if (maxaddr < mm->gpa + mm->len)
872 				maxaddr = mm->gpa + mm->len;
873 		}
874 	}
875 	return (maxaddr);
876 }
877 
878 static void
879 vm_iommu_modify(struct vm *vm, bool map)
880 {
881 	int i, sz;
882 	vm_paddr_t gpa, hpa;
883 	struct mem_map *mm;
884 	void *vp, *cookie, *host_domain;
885 
886 	sz = PAGE_SIZE;
887 	host_domain = iommu_host_domain();
888 
889 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
890 		mm = &vm->mem_maps[i];
891 		if (!sysmem_mapping(vm, mm))
892 			continue;
893 
894 		if (map) {
895 			KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0,
896 			    ("iommu map found invalid memmap %#lx/%#lx/%#x",
897 			    mm->gpa, mm->len, mm->flags));
898 			if ((mm->flags & VM_MEMMAP_F_WIRED) == 0)
899 				continue;
900 			mm->flags |= VM_MEMMAP_F_IOMMU;
901 		} else {
902 			if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0)
903 				continue;
904 			mm->flags &= ~VM_MEMMAP_F_IOMMU;
905 			KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0,
906 			    ("iommu unmap found invalid memmap %#lx/%#lx/%#x",
907 			    mm->gpa, mm->len, mm->flags));
908 		}
909 
910 		gpa = mm->gpa;
911 		while (gpa < mm->gpa + mm->len) {
912 			vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE,
913 					 &cookie);
914 			KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
915 			    vm_name(vm), gpa));
916 
917 			vm_gpa_release(cookie);
918 
919 			hpa = DMAP_TO_PHYS((uintptr_t)vp);
920 			if (map) {
921 				iommu_create_mapping(vm->iommu, gpa, hpa, sz);
922 				iommu_remove_mapping(host_domain, hpa, sz);
923 			} else {
924 				iommu_remove_mapping(vm->iommu, gpa, sz);
925 				iommu_create_mapping(host_domain, hpa, hpa, sz);
926 			}
927 
928 			gpa += PAGE_SIZE;
929 		}
930 	}
931 
932 	/*
933 	 * Invalidate the cached translations associated with the domain
934 	 * from which pages were removed.
935 	 */
936 	if (map)
937 		iommu_invalidate_tlb(host_domain);
938 	else
939 		iommu_invalidate_tlb(vm->iommu);
940 }
941 
942 #define	vm_iommu_unmap(vm)	vm_iommu_modify((vm), false)
943 #define	vm_iommu_map(vm)	vm_iommu_modify((vm), true)
944 
945 int
946 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
947 {
948 	int error;
949 
950 	error = ppt_unassign_device(vm, bus, slot, func);
951 	if (error)
952 		return (error);
953 
954 	if (ppt_assigned_devices(vm) == 0)
955 		vm_iommu_unmap(vm);
956 
957 	return (0);
958 }
959 
960 int
961 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
962 {
963 	int error;
964 	vm_paddr_t maxaddr;
965 
966 	/* Set up the IOMMU to do the 'gpa' to 'hpa' translation */
967 	if (ppt_assigned_devices(vm) == 0) {
968 		KASSERT(vm->iommu == NULL,
969 		    ("vm_assign_pptdev: iommu must be NULL"));
970 		maxaddr = vmm_sysmem_maxaddr(vm);
971 		vm->iommu = iommu_create_domain(maxaddr);
972 		if (vm->iommu == NULL)
973 			return (ENXIO);
974 		vm_iommu_map(vm);
975 	}
976 
977 	error = ppt_assign_device(vm, bus, slot, func);
978 	return (error);
979 }
980 
981 void *
982 vm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot,
983 	    void **cookie)
984 {
985 	int i, count, pageoff;
986 	struct mem_map *mm;
987 	vm_page_t m;
988 #ifdef INVARIANTS
989 	/*
990 	 * All vcpus are frozen by ioctls that modify the memory map
991 	 * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is
992 	 * guaranteed if at least one vcpu is in the VCPU_FROZEN state.
993 	 */
994 	int state;
995 	KASSERT(vcpuid >= -1 && vcpuid < vm->maxcpus, ("%s: invalid vcpuid %d",
996 	    __func__, vcpuid));
997 	for (i = 0; i < vm->maxcpus; i++) {
998 		if (vcpuid != -1 && vcpuid != i)
999 			continue;
1000 		state = vcpu_get_state(vm, i, NULL);
1001 		KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d",
1002 		    __func__, state));
1003 	}
1004 #endif
1005 	pageoff = gpa & PAGE_MASK;
1006 	if (len > PAGE_SIZE - pageoff)
1007 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
1008 
1009 	count = 0;
1010 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
1011 		mm = &vm->mem_maps[i];
1012 		if (gpa >= mm->gpa && gpa < mm->gpa + mm->len) {
1013 			count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
1014 			    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
1015 			break;
1016 		}
1017 	}
1018 
1019 	if (count == 1) {
1020 		*cookie = m;
1021 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
1022 	} else {
1023 		*cookie = NULL;
1024 		return (NULL);
1025 	}
1026 }
1027 
1028 void
1029 vm_gpa_release(void *cookie)
1030 {
1031 	vm_page_t m = cookie;
1032 
1033 	vm_page_unwire(m, PQ_ACTIVE);
1034 }
1035 
1036 int
1037 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
1038 {
1039 
1040 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1041 		return (EINVAL);
1042 
1043 	if (reg >= VM_REG_LAST)
1044 		return (EINVAL);
1045 
1046 	return (vmmops_getreg(vm->cookie, vcpu, reg, retval));
1047 }
1048 
1049 int
1050 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val)
1051 {
1052 	struct vcpu *vcpu;
1053 	int error;
1054 
1055 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1056 		return (EINVAL);
1057 
1058 	if (reg >= VM_REG_LAST)
1059 		return (EINVAL);
1060 
1061 	error = vmmops_setreg(vm->cookie, vcpuid, reg, val);
1062 	if (error || reg != VM_REG_GUEST_RIP)
1063 		return (error);
1064 
1065 	/* Set 'nextrip' to match the value of %rip */
1066 	VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val);
1067 	vcpu = &vm->vcpu[vcpuid];
1068 	vcpu->nextrip = val;
1069 	return (0);
1070 }
1071 
1072 static bool
1073 is_descriptor_table(int reg)
1074 {
1075 
1076 	switch (reg) {
1077 	case VM_REG_GUEST_IDTR:
1078 	case VM_REG_GUEST_GDTR:
1079 		return (true);
1080 	default:
1081 		return (false);
1082 	}
1083 }
1084 
1085 static bool
1086 is_segment_register(int reg)
1087 {
1088 
1089 	switch (reg) {
1090 	case VM_REG_GUEST_ES:
1091 	case VM_REG_GUEST_CS:
1092 	case VM_REG_GUEST_SS:
1093 	case VM_REG_GUEST_DS:
1094 	case VM_REG_GUEST_FS:
1095 	case VM_REG_GUEST_GS:
1096 	case VM_REG_GUEST_TR:
1097 	case VM_REG_GUEST_LDTR:
1098 		return (true);
1099 	default:
1100 		return (false);
1101 	}
1102 }
1103 
1104 int
1105 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
1106 		struct seg_desc *desc)
1107 {
1108 
1109 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1110 		return (EINVAL);
1111 
1112 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1113 		return (EINVAL);
1114 
1115 	return (vmmops_getdesc(vm->cookie, vcpu, reg, desc));
1116 }
1117 
1118 int
1119 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
1120 		struct seg_desc *desc)
1121 {
1122 	if (vcpu < 0 || vcpu >= vm->maxcpus)
1123 		return (EINVAL);
1124 
1125 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1126 		return (EINVAL);
1127 
1128 	return (vmmops_setdesc(vm->cookie, vcpu, reg, desc));
1129 }
1130 
1131 static void
1132 restore_guest_fpustate(struct vcpu *vcpu)
1133 {
1134 
1135 	/* flush host state to the pcb */
1136 	fpuexit(curthread);
1137 
1138 	/* restore guest FPU state */
1139 	fpu_stop_emulating();
1140 	fpurestore(vcpu->guestfpu);
1141 
1142 	/* restore guest XCR0 if XSAVE is enabled in the host */
1143 	if (rcr4() & CR4_XSAVE)
1144 		load_xcr(0, vcpu->guest_xcr0);
1145 
1146 	/*
1147 	 * The FPU is now "dirty" with the guest's state so turn on emulation
1148 	 * to trap any access to the FPU by the host.
1149 	 */
1150 	fpu_start_emulating();
1151 }
1152 
1153 static void
1154 save_guest_fpustate(struct vcpu *vcpu)
1155 {
1156 
1157 	if ((rcr0() & CR0_TS) == 0)
1158 		panic("fpu emulation not enabled in host!");
1159 
1160 	/* save guest XCR0 and restore host XCR0 */
1161 	if (rcr4() & CR4_XSAVE) {
1162 		vcpu->guest_xcr0 = rxcr(0);
1163 		load_xcr(0, vmm_get_host_xcr0());
1164 	}
1165 
1166 	/* save guest FPU state */
1167 	fpu_stop_emulating();
1168 	fpusave(vcpu->guestfpu);
1169 	fpu_start_emulating();
1170 }
1171 
1172 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
1173 
1174 static int
1175 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1176     bool from_idle)
1177 {
1178 	struct vcpu *vcpu;
1179 	int error;
1180 
1181 	vcpu = &vm->vcpu[vcpuid];
1182 	vcpu_assert_locked(vcpu);
1183 
1184 	/*
1185 	 * State transitions from the vmmdev_ioctl() must always begin from
1186 	 * the VCPU_IDLE state. This guarantees that there is only a single
1187 	 * ioctl() operating on a vcpu at any point.
1188 	 */
1189 	if (from_idle) {
1190 		while (vcpu->state != VCPU_IDLE) {
1191 			vcpu->reqidle = 1;
1192 			vcpu_notify_event_locked(vcpu, false);
1193 			VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to "
1194 			    "idle requested", vcpu_state2str(vcpu->state));
1195 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
1196 		}
1197 	} else {
1198 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1199 		    "vcpu idle state"));
1200 	}
1201 
1202 	if (vcpu->state == VCPU_RUNNING) {
1203 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1204 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1205 	} else {
1206 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1207 		    "vcpu that is not running", vcpu->hostcpu));
1208 	}
1209 
1210 	/*
1211 	 * The following state transitions are allowed:
1212 	 * IDLE -> FROZEN -> IDLE
1213 	 * FROZEN -> RUNNING -> FROZEN
1214 	 * FROZEN -> SLEEPING -> FROZEN
1215 	 */
1216 	switch (vcpu->state) {
1217 	case VCPU_IDLE:
1218 	case VCPU_RUNNING:
1219 	case VCPU_SLEEPING:
1220 		error = (newstate != VCPU_FROZEN);
1221 		break;
1222 	case VCPU_FROZEN:
1223 		error = (newstate == VCPU_FROZEN);
1224 		break;
1225 	default:
1226 		error = 1;
1227 		break;
1228 	}
1229 
1230 	if (error)
1231 		return (EBUSY);
1232 
1233 	VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s",
1234 	    vcpu_state2str(vcpu->state), vcpu_state2str(newstate));
1235 
1236 	vcpu->state = newstate;
1237 	if (newstate == VCPU_RUNNING)
1238 		vcpu->hostcpu = curcpu;
1239 	else
1240 		vcpu->hostcpu = NOCPU;
1241 
1242 	if (newstate == VCPU_IDLE)
1243 		wakeup(&vcpu->state);
1244 
1245 	return (0);
1246 }
1247 
1248 static void
1249 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1250 {
1251 	int error;
1252 
1253 	if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1254 		panic("Error %d setting state to %d\n", error, newstate);
1255 }
1256 
1257 static void
1258 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1259 {
1260 	int error;
1261 
1262 	if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0)
1263 		panic("Error %d setting state to %d", error, newstate);
1264 }
1265 
1266 #define	RENDEZVOUS_CTR0(vm, vcpuid, fmt)				\
1267 	do {								\
1268 		if (vcpuid >= 0)					\
1269 			VCPU_CTR0(vm, vcpuid, fmt);			\
1270 		else							\
1271 			VM_CTR0(vm, fmt);				\
1272 	} while (0)
1273 
1274 static int
1275 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1276 {
1277 	struct thread *td;
1278 	int error;
1279 
1280 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
1281 	    ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1282 
1283 	error = 0;
1284 	td = curthread;
1285 	mtx_lock(&vm->rendezvous_mtx);
1286 	while (vm->rendezvous_func != NULL) {
1287 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1288 		CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1289 
1290 		if (vcpuid != -1 &&
1291 		    CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1292 		    !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1293 			VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1294 			(*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1295 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1296 		}
1297 		if (CPU_CMP(&vm->rendezvous_req_cpus,
1298 		    &vm->rendezvous_done_cpus) == 0) {
1299 			VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1300 			vm->rendezvous_func = NULL;
1301 			wakeup(&vm->rendezvous_func);
1302 			break;
1303 		}
1304 		RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1305 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1306 		    "vmrndv", hz);
1307 		if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1308 			mtx_unlock(&vm->rendezvous_mtx);
1309 			error = thread_check_susp(td, true);
1310 			if (error != 0)
1311 				return (error);
1312 			mtx_lock(&vm->rendezvous_mtx);
1313 		}
1314 	}
1315 	mtx_unlock(&vm->rendezvous_mtx);
1316 	return (0);
1317 }
1318 
1319 /*
1320  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1321  */
1322 static int
1323 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1324 {
1325 	struct vcpu *vcpu;
1326 	const char *wmesg;
1327 	struct thread *td;
1328 	int error, t, vcpu_halted, vm_halted;
1329 
1330 	KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1331 
1332 	vcpu = &vm->vcpu[vcpuid];
1333 	vcpu_halted = 0;
1334 	vm_halted = 0;
1335 	error = 0;
1336 	td = curthread;
1337 
1338 	vcpu_lock(vcpu);
1339 	while (1) {
1340 		/*
1341 		 * Do a final check for pending NMI or interrupts before
1342 		 * really putting this thread to sleep. Also check for
1343 		 * software events that would cause this vcpu to wakeup.
1344 		 *
1345 		 * These interrupts/events could have happened after the
1346 		 * vcpu returned from vmmops_run() and before it acquired the
1347 		 * vcpu lock above.
1348 		 */
1349 		if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle)
1350 			break;
1351 		if (vm_nmi_pending(vm, vcpuid))
1352 			break;
1353 		if (!intr_disabled) {
1354 			if (vm_extint_pending(vm, vcpuid) ||
1355 			    vlapic_pending_intr(vcpu->vlapic, NULL)) {
1356 				break;
1357 			}
1358 		}
1359 
1360 		/* Don't go to sleep if the vcpu thread needs to yield */
1361 		if (vcpu_should_yield(vm, vcpuid))
1362 			break;
1363 
1364 		if (vcpu_debugged(vm, vcpuid))
1365 			break;
1366 
1367 		/*
1368 		 * Some Linux guests implement "halt" by having all vcpus
1369 		 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1370 		 * track of the vcpus that have entered this state. When all
1371 		 * vcpus enter the halted state the virtual machine is halted.
1372 		 */
1373 		if (intr_disabled) {
1374 			wmesg = "vmhalt";
1375 			VCPU_CTR0(vm, vcpuid, "Halted");
1376 			if (!vcpu_halted && halt_detection_enabled) {
1377 				vcpu_halted = 1;
1378 				CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1379 			}
1380 			if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1381 				vm_halted = 1;
1382 				break;
1383 			}
1384 		} else {
1385 			wmesg = "vmidle";
1386 		}
1387 
1388 		t = ticks;
1389 		vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1390 		/*
1391 		 * XXX msleep_spin() cannot be interrupted by signals so
1392 		 * wake up periodically to check pending signals.
1393 		 */
1394 		msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1395 		vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1396 		vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1397 		if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1398 			vcpu_unlock(vcpu);
1399 			error = thread_check_susp(td, false);
1400 			if (error != 0)
1401 				return (error);
1402 			vcpu_lock(vcpu);
1403 		}
1404 	}
1405 
1406 	if (vcpu_halted)
1407 		CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1408 
1409 	vcpu_unlock(vcpu);
1410 
1411 	if (vm_halted)
1412 		vm_suspend(vm, VM_SUSPEND_HALT);
1413 
1414 	return (0);
1415 }
1416 
1417 static int
1418 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1419 {
1420 	int rv, ftype;
1421 	struct vm_map *map;
1422 	struct vcpu *vcpu;
1423 	struct vm_exit *vme;
1424 
1425 	vcpu = &vm->vcpu[vcpuid];
1426 	vme = &vcpu->exitinfo;
1427 
1428 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1429 	    __func__, vme->inst_length));
1430 
1431 	ftype = vme->u.paging.fault_type;
1432 	KASSERT(ftype == VM_PROT_READ ||
1433 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1434 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1435 
1436 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1437 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1438 		    vme->u.paging.gpa, ftype);
1439 		if (rv == 0) {
1440 			VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1441 			    ftype == VM_PROT_READ ? "accessed" : "dirty",
1442 			    vme->u.paging.gpa);
1443 			goto done;
1444 		}
1445 	}
1446 
1447 	map = &vm->vmspace->vm_map;
1448 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL);
1449 
1450 	VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1451 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1452 
1453 	if (rv != KERN_SUCCESS)
1454 		return (EFAULT);
1455 done:
1456 	return (0);
1457 }
1458 
1459 static int
1460 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1461 {
1462 	struct vie *vie;
1463 	struct vcpu *vcpu;
1464 	struct vm_exit *vme;
1465 	uint64_t gla, gpa, cs_base;
1466 	struct vm_guest_paging *paging;
1467 	mem_region_read_t mread;
1468 	mem_region_write_t mwrite;
1469 	enum vm_cpu_mode cpu_mode;
1470 	int cs_d, error, fault;
1471 
1472 	vcpu = &vm->vcpu[vcpuid];
1473 	vme = &vcpu->exitinfo;
1474 
1475 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1476 	    __func__, vme->inst_length));
1477 
1478 	gla = vme->u.inst_emul.gla;
1479 	gpa = vme->u.inst_emul.gpa;
1480 	cs_base = vme->u.inst_emul.cs_base;
1481 	cs_d = vme->u.inst_emul.cs_d;
1482 	vie = &vme->u.inst_emul.vie;
1483 	paging = &vme->u.inst_emul.paging;
1484 	cpu_mode = paging->cpu_mode;
1485 
1486 	VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1487 
1488 	/* Fetch, decode and emulate the faulting instruction */
1489 	if (vie->num_valid == 0) {
1490 		error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip +
1491 		    cs_base, VIE_INST_SIZE, vie, &fault);
1492 	} else {
1493 		/*
1494 		 * The instruction bytes have already been copied into 'vie'
1495 		 */
1496 		error = fault = 0;
1497 	}
1498 	if (error || fault)
1499 		return (error);
1500 
1501 	if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) {
1502 		VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx",
1503 		    vme->rip + cs_base);
1504 		*retu = true;	    /* dump instruction bytes in userspace */
1505 		return (0);
1506 	}
1507 
1508 	/*
1509 	 * Update 'nextrip' based on the length of the emulated instruction.
1510 	 */
1511 	vme->inst_length = vie->num_processed;
1512 	vcpu->nextrip += vie->num_processed;
1513 	VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction "
1514 	    "decoding", vcpu->nextrip);
1515 
1516 	/* return to userland unless this is an in-kernel emulated device */
1517 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1518 		mread = lapic_mmio_read;
1519 		mwrite = lapic_mmio_write;
1520 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1521 		mread = vioapic_mmio_read;
1522 		mwrite = vioapic_mmio_write;
1523 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1524 		mread = vhpet_mmio_read;
1525 		mwrite = vhpet_mmio_write;
1526 	} else {
1527 		*retu = true;
1528 		return (0);
1529 	}
1530 
1531 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1532 	    mread, mwrite, retu);
1533 
1534 	return (error);
1535 }
1536 
1537 static int
1538 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1539 {
1540 	int error, i;
1541 	struct vcpu *vcpu;
1542 	struct thread *td;
1543 
1544 	error = 0;
1545 	vcpu = &vm->vcpu[vcpuid];
1546 	td = curthread;
1547 
1548 	CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1549 
1550 	/*
1551 	 * Wait until all 'active_cpus' have suspended themselves.
1552 	 *
1553 	 * Since a VM may be suspended at any time including when one or
1554 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1555 	 * handler while we are waiting to prevent a deadlock.
1556 	 */
1557 	vcpu_lock(vcpu);
1558 	while (error == 0) {
1559 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1560 			VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1561 			break;
1562 		}
1563 
1564 		if (vm->rendezvous_func == NULL) {
1565 			VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1566 			vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1567 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1568 			vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1569 			if ((td->td_flags & TDF_NEEDSUSPCHK) != 0) {
1570 				vcpu_unlock(vcpu);
1571 				error = thread_check_susp(td, false);
1572 				vcpu_lock(vcpu);
1573 			}
1574 		} else {
1575 			VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1576 			vcpu_unlock(vcpu);
1577 			error = vm_handle_rendezvous(vm, vcpuid);
1578 			vcpu_lock(vcpu);
1579 		}
1580 	}
1581 	vcpu_unlock(vcpu);
1582 
1583 	/*
1584 	 * Wakeup the other sleeping vcpus and return to userspace.
1585 	 */
1586 	for (i = 0; i < vm->maxcpus; i++) {
1587 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1588 			vcpu_notify_event(vm, i, false);
1589 		}
1590 	}
1591 
1592 	*retu = true;
1593 	return (error);
1594 }
1595 
1596 static int
1597 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu)
1598 {
1599 	struct vcpu *vcpu = &vm->vcpu[vcpuid];
1600 
1601 	vcpu_lock(vcpu);
1602 	KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle));
1603 	vcpu->reqidle = 0;
1604 	vcpu_unlock(vcpu);
1605 	*retu = true;
1606 	return (0);
1607 }
1608 
1609 int
1610 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1611 {
1612 	int i;
1613 
1614 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1615 		return (EINVAL);
1616 
1617 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1618 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1619 		    vm->suspend, how);
1620 		return (EALREADY);
1621 	}
1622 
1623 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1624 
1625 	/*
1626 	 * Notify all active vcpus that they are now suspended.
1627 	 */
1628 	for (i = 0; i < vm->maxcpus; i++) {
1629 		if (CPU_ISSET(i, &vm->active_cpus))
1630 			vcpu_notify_event(vm, i, false);
1631 	}
1632 
1633 	return (0);
1634 }
1635 
1636 void
1637 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1638 {
1639 	struct vm_exit *vmexit;
1640 
1641 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1642 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1643 
1644 	vmexit = vm_exitinfo(vm, vcpuid);
1645 	vmexit->rip = rip;
1646 	vmexit->inst_length = 0;
1647 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1648 	vmexit->u.suspended.how = vm->suspend;
1649 }
1650 
1651 void
1652 vm_exit_debug(struct vm *vm, int vcpuid, uint64_t rip)
1653 {
1654 	struct vm_exit *vmexit;
1655 
1656 	vmexit = vm_exitinfo(vm, vcpuid);
1657 	vmexit->rip = rip;
1658 	vmexit->inst_length = 0;
1659 	vmexit->exitcode = VM_EXITCODE_DEBUG;
1660 }
1661 
1662 void
1663 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1664 {
1665 	struct vm_exit *vmexit;
1666 
1667 	KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1668 
1669 	vmexit = vm_exitinfo(vm, vcpuid);
1670 	vmexit->rip = rip;
1671 	vmexit->inst_length = 0;
1672 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1673 	vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1674 }
1675 
1676 void
1677 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip)
1678 {
1679 	struct vm_exit *vmexit;
1680 
1681 	vmexit = vm_exitinfo(vm, vcpuid);
1682 	vmexit->rip = rip;
1683 	vmexit->inst_length = 0;
1684 	vmexit->exitcode = VM_EXITCODE_REQIDLE;
1685 	vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1);
1686 }
1687 
1688 void
1689 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1690 {
1691 	struct vm_exit *vmexit;
1692 
1693 	vmexit = vm_exitinfo(vm, vcpuid);
1694 	vmexit->rip = rip;
1695 	vmexit->inst_length = 0;
1696 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1697 	vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1698 }
1699 
1700 int
1701 vm_run(struct vm *vm, struct vm_run *vmrun)
1702 {
1703 	struct vm_eventinfo evinfo;
1704 	int error, vcpuid;
1705 	struct vcpu *vcpu;
1706 	struct pcb *pcb;
1707 	uint64_t tscval;
1708 	struct vm_exit *vme;
1709 	bool retu, intr_disabled;
1710 	pmap_t pmap;
1711 
1712 	vcpuid = vmrun->cpuid;
1713 
1714 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1715 		return (EINVAL);
1716 
1717 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1718 		return (EINVAL);
1719 
1720 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1721 		return (EINVAL);
1722 
1723 	pmap = vmspace_pmap(vm->vmspace);
1724 	vcpu = &vm->vcpu[vcpuid];
1725 	vme = &vcpu->exitinfo;
1726 	evinfo.rptr = &vm->rendezvous_func;
1727 	evinfo.sptr = &vm->suspend;
1728 	evinfo.iptr = &vcpu->reqidle;
1729 restart:
1730 	critical_enter();
1731 
1732 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1733 	    ("vm_run: absurd pm_active"));
1734 
1735 	tscval = rdtsc();
1736 
1737 	pcb = PCPU_GET(curpcb);
1738 	set_pcb_flags(pcb, PCB_FULL_IRET);
1739 
1740 	restore_guest_fpustate(vcpu);
1741 
1742 	vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1743 	error = vmmops_run(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo);
1744 	vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1745 
1746 	save_guest_fpustate(vcpu);
1747 
1748 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1749 
1750 	critical_exit();
1751 
1752 	if (error == 0) {
1753 		retu = false;
1754 		vcpu->nextrip = vme->rip + vme->inst_length;
1755 		switch (vme->exitcode) {
1756 		case VM_EXITCODE_REQIDLE:
1757 			error = vm_handle_reqidle(vm, vcpuid, &retu);
1758 			break;
1759 		case VM_EXITCODE_SUSPENDED:
1760 			error = vm_handle_suspend(vm, vcpuid, &retu);
1761 			break;
1762 		case VM_EXITCODE_IOAPIC_EOI:
1763 			vioapic_process_eoi(vm, vcpuid,
1764 			    vme->u.ioapic_eoi.vector);
1765 			break;
1766 		case VM_EXITCODE_RENDEZVOUS:
1767 			error = vm_handle_rendezvous(vm, vcpuid);
1768 			break;
1769 		case VM_EXITCODE_HLT:
1770 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1771 			error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1772 			break;
1773 		case VM_EXITCODE_PAGING:
1774 			error = vm_handle_paging(vm, vcpuid, &retu);
1775 			break;
1776 		case VM_EXITCODE_INST_EMUL:
1777 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1778 			break;
1779 		case VM_EXITCODE_INOUT:
1780 		case VM_EXITCODE_INOUT_STR:
1781 			error = vm_handle_inout(vm, vcpuid, vme, &retu);
1782 			break;
1783 		case VM_EXITCODE_MONITOR:
1784 		case VM_EXITCODE_MWAIT:
1785 		case VM_EXITCODE_VMINSN:
1786 			vm_inject_ud(vm, vcpuid);
1787 			break;
1788 		default:
1789 			retu = true;	/* handled in userland */
1790 			break;
1791 		}
1792 	}
1793 
1794 	if (error == 0 && retu == false)
1795 		goto restart;
1796 
1797 	VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode);
1798 
1799 	/* copy the exit information */
1800 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1801 	return (error);
1802 }
1803 
1804 int
1805 vm_restart_instruction(void *arg, int vcpuid)
1806 {
1807 	struct vm *vm;
1808 	struct vcpu *vcpu;
1809 	enum vcpu_state state;
1810 	uint64_t rip;
1811 	int error;
1812 
1813 	vm = arg;
1814 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1815 		return (EINVAL);
1816 
1817 	vcpu = &vm->vcpu[vcpuid];
1818 	state = vcpu_get_state(vm, vcpuid, NULL);
1819 	if (state == VCPU_RUNNING) {
1820 		/*
1821 		 * When a vcpu is "running" the next instruction is determined
1822 		 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
1823 		 * Thus setting 'inst_length' to zero will cause the current
1824 		 * instruction to be restarted.
1825 		 */
1826 		vcpu->exitinfo.inst_length = 0;
1827 		VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by "
1828 		    "setting inst_length to zero", vcpu->exitinfo.rip);
1829 	} else if (state == VCPU_FROZEN) {
1830 		/*
1831 		 * When a vcpu is "frozen" it is outside the critical section
1832 		 * around vmmops_run() and 'nextrip' points to the next
1833 		 * instruction. Thus instruction restart is achieved by setting
1834 		 * 'nextrip' to the vcpu's %rip.
1835 		 */
1836 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip);
1837 		KASSERT(!error, ("%s: error %d getting rip", __func__, error));
1838 		VCPU_CTR2(vm, vcpuid, "restarting instruction by updating "
1839 		    "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
1840 		vcpu->nextrip = rip;
1841 	} else {
1842 		panic("%s: invalid state %d", __func__, state);
1843 	}
1844 	return (0);
1845 }
1846 
1847 int
1848 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1849 {
1850 	struct vcpu *vcpu;
1851 	int type, vector;
1852 
1853 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
1854 		return (EINVAL);
1855 
1856 	vcpu = &vm->vcpu[vcpuid];
1857 
1858 	if (info & VM_INTINFO_VALID) {
1859 		type = info & VM_INTINFO_TYPE;
1860 		vector = info & 0xff;
1861 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1862 			return (EINVAL);
1863 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1864 			return (EINVAL);
1865 		if (info & VM_INTINFO_RSVD)
1866 			return (EINVAL);
1867 	} else {
1868 		info = 0;
1869 	}
1870 	VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1871 	vcpu->exitintinfo = info;
1872 	return (0);
1873 }
1874 
1875 enum exc_class {
1876 	EXC_BENIGN,
1877 	EXC_CONTRIBUTORY,
1878 	EXC_PAGEFAULT
1879 };
1880 
1881 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
1882 
1883 static enum exc_class
1884 exception_class(uint64_t info)
1885 {
1886 	int type, vector;
1887 
1888 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1889 	type = info & VM_INTINFO_TYPE;
1890 	vector = info & 0xff;
1891 
1892 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1893 	switch (type) {
1894 	case VM_INTINFO_HWINTR:
1895 	case VM_INTINFO_SWINTR:
1896 	case VM_INTINFO_NMI:
1897 		return (EXC_BENIGN);
1898 	default:
1899 		/*
1900 		 * Hardware exception.
1901 		 *
1902 		 * SVM and VT-x use identical type values to represent NMI,
1903 		 * hardware interrupt and software interrupt.
1904 		 *
1905 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1906 		 * for exceptions except #BP and #OF. #BP and #OF use a type
1907 		 * value of '5' or '6'. Therefore we don't check for explicit
1908 		 * values of 'type' to classify 'intinfo' into a hardware
1909 		 * exception.
1910 		 */
1911 		break;
1912 	}
1913 
1914 	switch (vector) {
1915 	case IDT_PF:
1916 	case IDT_VE:
1917 		return (EXC_PAGEFAULT);
1918 	case IDT_DE:
1919 	case IDT_TS:
1920 	case IDT_NP:
1921 	case IDT_SS:
1922 	case IDT_GP:
1923 		return (EXC_CONTRIBUTORY);
1924 	default:
1925 		return (EXC_BENIGN);
1926 	}
1927 }
1928 
1929 static int
1930 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1931     uint64_t *retinfo)
1932 {
1933 	enum exc_class exc1, exc2;
1934 	int type1, vector1;
1935 
1936 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1937 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1938 
1939 	/*
1940 	 * If an exception occurs while attempting to call the double-fault
1941 	 * handler the processor enters shutdown mode (aka triple fault).
1942 	 */
1943 	type1 = info1 & VM_INTINFO_TYPE;
1944 	vector1 = info1 & 0xff;
1945 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1946 		VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1947 		    info1, info2);
1948 		vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1949 		*retinfo = 0;
1950 		return (0);
1951 	}
1952 
1953 	/*
1954 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1955 	 */
1956 	exc1 = exception_class(info1);
1957 	exc2 = exception_class(info2);
1958 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1959 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1960 		/* Convert nested fault into a double fault. */
1961 		*retinfo = IDT_DF;
1962 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1963 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
1964 	} else {
1965 		/* Handle exceptions serially */
1966 		*retinfo = info2;
1967 	}
1968 	return (1);
1969 }
1970 
1971 static uint64_t
1972 vcpu_exception_intinfo(struct vcpu *vcpu)
1973 {
1974 	uint64_t info = 0;
1975 
1976 	if (vcpu->exception_pending) {
1977 		info = vcpu->exc_vector & 0xff;
1978 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1979 		if (vcpu->exc_errcode_valid) {
1980 			info |= VM_INTINFO_DEL_ERRCODE;
1981 			info |= (uint64_t)vcpu->exc_errcode << 32;
1982 		}
1983 	}
1984 	return (info);
1985 }
1986 
1987 int
1988 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1989 {
1990 	struct vcpu *vcpu;
1991 	uint64_t info1, info2;
1992 	int valid;
1993 
1994 	KASSERT(vcpuid >= 0 &&
1995 	    vcpuid < vm->maxcpus, ("invalid vcpu %d", vcpuid));
1996 
1997 	vcpu = &vm->vcpu[vcpuid];
1998 
1999 	info1 = vcpu->exitintinfo;
2000 	vcpu->exitintinfo = 0;
2001 
2002 	info2 = 0;
2003 	if (vcpu->exception_pending) {
2004 		info2 = vcpu_exception_intinfo(vcpu);
2005 		vcpu->exception_pending = 0;
2006 		VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
2007 		    vcpu->exc_vector, info2);
2008 	}
2009 
2010 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
2011 		valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
2012 	} else if (info1 & VM_INTINFO_VALID) {
2013 		*retinfo = info1;
2014 		valid = 1;
2015 	} else if (info2 & VM_INTINFO_VALID) {
2016 		*retinfo = info2;
2017 		valid = 1;
2018 	} else {
2019 		valid = 0;
2020 	}
2021 
2022 	if (valid) {
2023 		VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
2024 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
2025 	}
2026 
2027 	return (valid);
2028 }
2029 
2030 int
2031 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
2032 {
2033 	struct vcpu *vcpu;
2034 
2035 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2036 		return (EINVAL);
2037 
2038 	vcpu = &vm->vcpu[vcpuid];
2039 	*info1 = vcpu->exitintinfo;
2040 	*info2 = vcpu_exception_intinfo(vcpu);
2041 	return (0);
2042 }
2043 
2044 int
2045 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid,
2046     uint32_t errcode, int restart_instruction)
2047 {
2048 	struct vcpu *vcpu;
2049 	uint64_t regval;
2050 	int error;
2051 
2052 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2053 		return (EINVAL);
2054 
2055 	if (vector < 0 || vector >= 32)
2056 		return (EINVAL);
2057 
2058 	/*
2059 	 * A double fault exception should never be injected directly into
2060 	 * the guest. It is a derived exception that results from specific
2061 	 * combinations of nested faults.
2062 	 */
2063 	if (vector == IDT_DF)
2064 		return (EINVAL);
2065 
2066 	vcpu = &vm->vcpu[vcpuid];
2067 
2068 	if (vcpu->exception_pending) {
2069 		VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
2070 		    "pending exception %d", vector, vcpu->exc_vector);
2071 		return (EBUSY);
2072 	}
2073 
2074 	if (errcode_valid) {
2075 		/*
2076 		 * Exceptions don't deliver an error code in real mode.
2077 		 */
2078 		error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, &regval);
2079 		KASSERT(!error, ("%s: error %d getting CR0", __func__, error));
2080 		if (!(regval & CR0_PE))
2081 			errcode_valid = 0;
2082 	}
2083 
2084 	/*
2085 	 * From section 26.6.1 "Interruptibility State" in Intel SDM:
2086 	 *
2087 	 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
2088 	 * one instruction or incurs an exception.
2089 	 */
2090 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
2091 	KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
2092 	    __func__, error));
2093 
2094 	if (restart_instruction)
2095 		vm_restart_instruction(vm, vcpuid);
2096 
2097 	vcpu->exception_pending = 1;
2098 	vcpu->exc_vector = vector;
2099 	vcpu->exc_errcode = errcode;
2100 	vcpu->exc_errcode_valid = errcode_valid;
2101 	VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector);
2102 	return (0);
2103 }
2104 
2105 void
2106 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
2107     int errcode)
2108 {
2109 	struct vm *vm;
2110 	int error, restart_instruction;
2111 
2112 	vm = vmarg;
2113 	restart_instruction = 1;
2114 
2115 	error = vm_inject_exception(vm, vcpuid, vector, errcode_valid,
2116 	    errcode, restart_instruction);
2117 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
2118 }
2119 
2120 void
2121 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
2122 {
2123 	struct vm *vm;
2124 	int error;
2125 
2126 	vm = vmarg;
2127 	VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
2128 	    error_code, cr2);
2129 
2130 	error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
2131 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
2132 
2133 	vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
2134 }
2135 
2136 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
2137 
2138 int
2139 vm_inject_nmi(struct vm *vm, int vcpuid)
2140 {
2141 	struct vcpu *vcpu;
2142 
2143 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2144 		return (EINVAL);
2145 
2146 	vcpu = &vm->vcpu[vcpuid];
2147 
2148 	vcpu->nmi_pending = 1;
2149 	vcpu_notify_event(vm, vcpuid, false);
2150 	return (0);
2151 }
2152 
2153 int
2154 vm_nmi_pending(struct vm *vm, int vcpuid)
2155 {
2156 	struct vcpu *vcpu;
2157 
2158 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2159 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2160 
2161 	vcpu = &vm->vcpu[vcpuid];
2162 
2163 	return (vcpu->nmi_pending);
2164 }
2165 
2166 void
2167 vm_nmi_clear(struct vm *vm, int vcpuid)
2168 {
2169 	struct vcpu *vcpu;
2170 
2171 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2172 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2173 
2174 	vcpu = &vm->vcpu[vcpuid];
2175 
2176 	if (vcpu->nmi_pending == 0)
2177 		panic("vm_nmi_clear: inconsistent nmi_pending state");
2178 
2179 	vcpu->nmi_pending = 0;
2180 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
2181 }
2182 
2183 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
2184 
2185 int
2186 vm_inject_extint(struct vm *vm, int vcpuid)
2187 {
2188 	struct vcpu *vcpu;
2189 
2190 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2191 		return (EINVAL);
2192 
2193 	vcpu = &vm->vcpu[vcpuid];
2194 
2195 	vcpu->extint_pending = 1;
2196 	vcpu_notify_event(vm, vcpuid, false);
2197 	return (0);
2198 }
2199 
2200 int
2201 vm_extint_pending(struct vm *vm, int vcpuid)
2202 {
2203 	struct vcpu *vcpu;
2204 
2205 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2206 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2207 
2208 	vcpu = &vm->vcpu[vcpuid];
2209 
2210 	return (vcpu->extint_pending);
2211 }
2212 
2213 void
2214 vm_extint_clear(struct vm *vm, int vcpuid)
2215 {
2216 	struct vcpu *vcpu;
2217 
2218 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2219 		panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2220 
2221 	vcpu = &vm->vcpu[vcpuid];
2222 
2223 	if (vcpu->extint_pending == 0)
2224 		panic("vm_extint_clear: inconsistent extint_pending state");
2225 
2226 	vcpu->extint_pending = 0;
2227 	vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
2228 }
2229 
2230 int
2231 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
2232 {
2233 	if (vcpu < 0 || vcpu >= vm->maxcpus)
2234 		return (EINVAL);
2235 
2236 	if (type < 0 || type >= VM_CAP_MAX)
2237 		return (EINVAL);
2238 
2239 	return (vmmops_getcap(vm->cookie, vcpu, type, retval));
2240 }
2241 
2242 int
2243 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
2244 {
2245 	if (vcpu < 0 || vcpu >= vm->maxcpus)
2246 		return (EINVAL);
2247 
2248 	if (type < 0 || type >= VM_CAP_MAX)
2249 		return (EINVAL);
2250 
2251 	return (vmmops_setcap(vm->cookie, vcpu, type, val));
2252 }
2253 
2254 struct vlapic *
2255 vm_lapic(struct vm *vm, int cpu)
2256 {
2257 	return (vm->vcpu[cpu].vlapic);
2258 }
2259 
2260 struct vioapic *
2261 vm_ioapic(struct vm *vm)
2262 {
2263 
2264 	return (vm->vioapic);
2265 }
2266 
2267 struct vhpet *
2268 vm_hpet(struct vm *vm)
2269 {
2270 
2271 	return (vm->vhpet);
2272 }
2273 
2274 bool
2275 vmm_is_pptdev(int bus, int slot, int func)
2276 {
2277 	int b, f, i, n, s;
2278 	char *val, *cp, *cp2;
2279 	bool found;
2280 
2281 	/*
2282 	 * XXX
2283 	 * The length of an environment variable is limited to 128 bytes which
2284 	 * puts an upper limit on the number of passthru devices that may be
2285 	 * specified using a single environment variable.
2286 	 *
2287 	 * Work around this by scanning multiple environment variable
2288 	 * names instead of a single one - yuck!
2289 	 */
2290 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2291 
2292 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2293 	found = false;
2294 	for (i = 0; names[i] != NULL && !found; i++) {
2295 		cp = val = kern_getenv(names[i]);
2296 		while (cp != NULL && *cp != '\0') {
2297 			if ((cp2 = strchr(cp, ' ')) != NULL)
2298 				*cp2 = '\0';
2299 
2300 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2301 			if (n == 3 && bus == b && slot == s && func == f) {
2302 				found = true;
2303 				break;
2304 			}
2305 
2306 			if (cp2 != NULL)
2307 				*cp2++ = ' ';
2308 
2309 			cp = cp2;
2310 		}
2311 		freeenv(val);
2312 	}
2313 	return (found);
2314 }
2315 
2316 void *
2317 vm_iommu_domain(struct vm *vm)
2318 {
2319 
2320 	return (vm->iommu);
2321 }
2322 
2323 int
2324 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
2325     bool from_idle)
2326 {
2327 	int error;
2328 	struct vcpu *vcpu;
2329 
2330 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2331 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
2332 
2333 	vcpu = &vm->vcpu[vcpuid];
2334 
2335 	vcpu_lock(vcpu);
2336 	error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle);
2337 	vcpu_unlock(vcpu);
2338 
2339 	return (error);
2340 }
2341 
2342 enum vcpu_state
2343 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2344 {
2345 	struct vcpu *vcpu;
2346 	enum vcpu_state state;
2347 
2348 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2349 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2350 
2351 	vcpu = &vm->vcpu[vcpuid];
2352 
2353 	vcpu_lock(vcpu);
2354 	state = vcpu->state;
2355 	if (hostcpu != NULL)
2356 		*hostcpu = vcpu->hostcpu;
2357 	vcpu_unlock(vcpu);
2358 
2359 	return (state);
2360 }
2361 
2362 int
2363 vm_activate_cpu(struct vm *vm, int vcpuid)
2364 {
2365 
2366 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2367 		return (EINVAL);
2368 
2369 	if (CPU_ISSET(vcpuid, &vm->active_cpus))
2370 		return (EBUSY);
2371 
2372 	VCPU_CTR0(vm, vcpuid, "activated");
2373 	CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2374 	return (0);
2375 }
2376 
2377 int
2378 vm_suspend_cpu(struct vm *vm, int vcpuid)
2379 {
2380 	int i;
2381 
2382 	if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2383 		return (EINVAL);
2384 
2385 	if (vcpuid == -1) {
2386 		vm->debug_cpus = vm->active_cpus;
2387 		for (i = 0; i < vm->maxcpus; i++) {
2388 			if (CPU_ISSET(i, &vm->active_cpus))
2389 				vcpu_notify_event(vm, i, false);
2390 		}
2391 	} else {
2392 		if (!CPU_ISSET(vcpuid, &vm->active_cpus))
2393 			return (EINVAL);
2394 
2395 		CPU_SET_ATOMIC(vcpuid, &vm->debug_cpus);
2396 		vcpu_notify_event(vm, vcpuid, false);
2397 	}
2398 	return (0);
2399 }
2400 
2401 int
2402 vm_resume_cpu(struct vm *vm, int vcpuid)
2403 {
2404 
2405 	if (vcpuid < -1 || vcpuid >= vm->maxcpus)
2406 		return (EINVAL);
2407 
2408 	if (vcpuid == -1) {
2409 		CPU_ZERO(&vm->debug_cpus);
2410 	} else {
2411 		if (!CPU_ISSET(vcpuid, &vm->debug_cpus))
2412 			return (EINVAL);
2413 
2414 		CPU_CLR_ATOMIC(vcpuid, &vm->debug_cpus);
2415 	}
2416 	return (0);
2417 }
2418 
2419 int
2420 vcpu_debugged(struct vm *vm, int vcpuid)
2421 {
2422 
2423 	return (CPU_ISSET(vcpuid, &vm->debug_cpus));
2424 }
2425 
2426 cpuset_t
2427 vm_active_cpus(struct vm *vm)
2428 {
2429 
2430 	return (vm->active_cpus);
2431 }
2432 
2433 cpuset_t
2434 vm_debug_cpus(struct vm *vm)
2435 {
2436 
2437 	return (vm->debug_cpus);
2438 }
2439 
2440 cpuset_t
2441 vm_suspended_cpus(struct vm *vm)
2442 {
2443 
2444 	return (vm->suspended_cpus);
2445 }
2446 
2447 void *
2448 vcpu_stats(struct vm *vm, int vcpuid)
2449 {
2450 
2451 	return (vm->vcpu[vcpuid].stats);
2452 }
2453 
2454 int
2455 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2456 {
2457 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2458 		return (EINVAL);
2459 
2460 	*state = vm->vcpu[vcpuid].x2apic_state;
2461 
2462 	return (0);
2463 }
2464 
2465 int
2466 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2467 {
2468 	if (vcpuid < 0 || vcpuid >= vm->maxcpus)
2469 		return (EINVAL);
2470 
2471 	if (state >= X2APIC_STATE_LAST)
2472 		return (EINVAL);
2473 
2474 	vm->vcpu[vcpuid].x2apic_state = state;
2475 
2476 	vlapic_set_x2apic_state(vm, vcpuid, state);
2477 
2478 	return (0);
2479 }
2480 
2481 /*
2482  * This function is called to ensure that a vcpu "sees" a pending event
2483  * as soon as possible:
2484  * - If the vcpu thread is sleeping then it is woken up.
2485  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2486  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2487  */
2488 static void
2489 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
2490 {
2491 	int hostcpu;
2492 
2493 	hostcpu = vcpu->hostcpu;
2494 	if (vcpu->state == VCPU_RUNNING) {
2495 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2496 		if (hostcpu != curcpu) {
2497 			if (lapic_intr) {
2498 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2499 				    vmm_ipinum);
2500 			} else {
2501 				ipi_cpu(hostcpu, vmm_ipinum);
2502 			}
2503 		} else {
2504 			/*
2505 			 * If the 'vcpu' is running on 'curcpu' then it must
2506 			 * be sending a notification to itself (e.g. SELF_IPI).
2507 			 * The pending event will be picked up when the vcpu
2508 			 * transitions back to guest context.
2509 			 */
2510 		}
2511 	} else {
2512 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2513 		    "with hostcpu %d", vcpu->state, hostcpu));
2514 		if (vcpu->state == VCPU_SLEEPING)
2515 			wakeup_one(vcpu);
2516 	}
2517 }
2518 
2519 void
2520 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2521 {
2522 	struct vcpu *vcpu = &vm->vcpu[vcpuid];
2523 
2524 	vcpu_lock(vcpu);
2525 	vcpu_notify_event_locked(vcpu, lapic_intr);
2526 	vcpu_unlock(vcpu);
2527 }
2528 
2529 struct vmspace *
2530 vm_get_vmspace(struct vm *vm)
2531 {
2532 
2533 	return (vm->vmspace);
2534 }
2535 
2536 int
2537 vm_apicid2vcpuid(struct vm *vm, int apicid)
2538 {
2539 	/*
2540 	 * XXX apic id is assumed to be numerically identical to vcpu id
2541 	 */
2542 	return (apicid);
2543 }
2544 
2545 int
2546 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2547     vm_rendezvous_func_t func, void *arg)
2548 {
2549 	int error, i;
2550 
2551 	/*
2552 	 * Enforce that this function is called without any locks
2553 	 */
2554 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2555 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < vm->maxcpus),
2556 	    ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2557 
2558 restart:
2559 	mtx_lock(&vm->rendezvous_mtx);
2560 	if (vm->rendezvous_func != NULL) {
2561 		/*
2562 		 * If a rendezvous is already in progress then we need to
2563 		 * call the rendezvous handler in case this 'vcpuid' is one
2564 		 * of the targets of the rendezvous.
2565 		 */
2566 		RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2567 		mtx_unlock(&vm->rendezvous_mtx);
2568 		error = vm_handle_rendezvous(vm, vcpuid);
2569 		if (error != 0)
2570 			return (error);
2571 		goto restart;
2572 	}
2573 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2574 	    "rendezvous is still in progress"));
2575 
2576 	RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2577 	vm->rendezvous_req_cpus = dest;
2578 	CPU_ZERO(&vm->rendezvous_done_cpus);
2579 	vm->rendezvous_arg = arg;
2580 	vm->rendezvous_func = func;
2581 	mtx_unlock(&vm->rendezvous_mtx);
2582 
2583 	/*
2584 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2585 	 * vcpus so they handle the rendezvous as soon as possible.
2586 	 */
2587 	for (i = 0; i < vm->maxcpus; i++) {
2588 		if (CPU_ISSET(i, &dest))
2589 			vcpu_notify_event(vm, i, false);
2590 	}
2591 
2592 	return (vm_handle_rendezvous(vm, vcpuid));
2593 }
2594 
2595 struct vatpic *
2596 vm_atpic(struct vm *vm)
2597 {
2598 	return (vm->vatpic);
2599 }
2600 
2601 struct vatpit *
2602 vm_atpit(struct vm *vm)
2603 {
2604 	return (vm->vatpit);
2605 }
2606 
2607 struct vpmtmr *
2608 vm_pmtmr(struct vm *vm)
2609 {
2610 
2611 	return (vm->vpmtmr);
2612 }
2613 
2614 struct vrtc *
2615 vm_rtc(struct vm *vm)
2616 {
2617 
2618 	return (vm->vrtc);
2619 }
2620 
2621 enum vm_reg_name
2622 vm_segment_name(int seg)
2623 {
2624 	static enum vm_reg_name seg_names[] = {
2625 		VM_REG_GUEST_ES,
2626 		VM_REG_GUEST_CS,
2627 		VM_REG_GUEST_SS,
2628 		VM_REG_GUEST_DS,
2629 		VM_REG_GUEST_FS,
2630 		VM_REG_GUEST_GS
2631 	};
2632 
2633 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2634 	    ("%s: invalid segment encoding %d", __func__, seg));
2635 	return (seg_names[seg]);
2636 }
2637 
2638 void
2639 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2640     int num_copyinfo)
2641 {
2642 	int idx;
2643 
2644 	for (idx = 0; idx < num_copyinfo; idx++) {
2645 		if (copyinfo[idx].cookie != NULL)
2646 			vm_gpa_release(copyinfo[idx].cookie);
2647 	}
2648 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2649 }
2650 
2651 int
2652 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2653     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2654     int num_copyinfo, int *fault)
2655 {
2656 	int error, idx, nused;
2657 	size_t n, off, remaining;
2658 	void *hva, *cookie;
2659 	uint64_t gpa;
2660 
2661 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2662 
2663 	nused = 0;
2664 	remaining = len;
2665 	while (remaining > 0) {
2666 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2667 		error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault);
2668 		if (error || *fault)
2669 			return (error);
2670 		off = gpa & PAGE_MASK;
2671 		n = min(remaining, PAGE_SIZE - off);
2672 		copyinfo[nused].gpa = gpa;
2673 		copyinfo[nused].len = n;
2674 		remaining -= n;
2675 		gla += n;
2676 		nused++;
2677 	}
2678 
2679 	for (idx = 0; idx < nused; idx++) {
2680 		hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa,
2681 		    copyinfo[idx].len, prot, &cookie);
2682 		if (hva == NULL)
2683 			break;
2684 		copyinfo[idx].hva = hva;
2685 		copyinfo[idx].cookie = cookie;
2686 	}
2687 
2688 	if (idx != nused) {
2689 		vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2690 		return (EFAULT);
2691 	} else {
2692 		*fault = 0;
2693 		return (0);
2694 	}
2695 }
2696 
2697 void
2698 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2699     size_t len)
2700 {
2701 	char *dst;
2702 	int idx;
2703 
2704 	dst = kaddr;
2705 	idx = 0;
2706 	while (len > 0) {
2707 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2708 		len -= copyinfo[idx].len;
2709 		dst += copyinfo[idx].len;
2710 		idx++;
2711 	}
2712 }
2713 
2714 void
2715 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2716     struct vm_copyinfo *copyinfo, size_t len)
2717 {
2718 	const char *src;
2719 	int idx;
2720 
2721 	src = kaddr;
2722 	idx = 0;
2723 	while (len > 0) {
2724 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2725 		len -= copyinfo[idx].len;
2726 		src += copyinfo[idx].len;
2727 		idx++;
2728 	}
2729 }
2730 
2731 /*
2732  * Return the amount of in-use and wired memory for the VM. Since
2733  * these are global stats, only return the values with for vCPU 0
2734  */
2735 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2736 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2737 
2738 static void
2739 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2740 {
2741 
2742 	if (vcpu == 0) {
2743 		vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2744 	       	    PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2745 	}
2746 }
2747 
2748 static void
2749 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2750 {
2751 
2752 	if (vcpu == 0) {
2753 		vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2754 	      	    PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2755 	}
2756 }
2757 
2758 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2759 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2760 
2761 #ifdef BHYVE_SNAPSHOT
2762 static int
2763 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta)
2764 {
2765 	int ret;
2766 	int i;
2767 	struct vcpu *vcpu;
2768 
2769 	for (i = 0; i < VM_MAXCPU; i++) {
2770 		vcpu = &vm->vcpu[i];
2771 
2772 		SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done);
2773 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done);
2774 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done);
2775 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done);
2776 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done);
2777 		SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done);
2778 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done);
2779 		SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done);
2780 		/* XXX we're cheating here, since the value of tsc_offset as
2781 		 * saved here is actually the value of the guest's TSC value.
2782 		 *
2783 		 * It will be turned turned back into an actual offset when the
2784 		 * TSC restore function is called
2785 		 */
2786 		SNAPSHOT_VAR_OR_LEAVE(vcpu->tsc_offset, meta, ret, done);
2787 	}
2788 
2789 done:
2790 	return (ret);
2791 }
2792 
2793 static int
2794 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta)
2795 {
2796 	int ret;
2797 	int i;
2798 	uint64_t now;
2799 
2800 	ret = 0;
2801 	now = rdtsc();
2802 
2803 	if (meta->op == VM_SNAPSHOT_SAVE) {
2804 		/* XXX make tsc_offset take the value TSC proper as seen by the
2805 		 * guest
2806 		 */
2807 		for (i = 0; i < VM_MAXCPU; i++)
2808 			vm->vcpu[i].tsc_offset += now;
2809 	}
2810 
2811 	ret = vm_snapshot_vcpus(vm, meta);
2812 	if (ret != 0) {
2813 		printf("%s: failed to copy vm data to user buffer", __func__);
2814 		goto done;
2815 	}
2816 
2817 	if (meta->op == VM_SNAPSHOT_SAVE) {
2818 		/* XXX turn tsc_offset back into an offset; actual value is only
2819 		 * required for restore; using it otherwise would be wrong
2820 		 */
2821 		for (i = 0; i < VM_MAXCPU; i++)
2822 			vm->vcpu[i].tsc_offset -= now;
2823 	}
2824 
2825 done:
2826 	return (ret);
2827 }
2828 
2829 static int
2830 vm_snapshot_vmcx(struct vm *vm, struct vm_snapshot_meta *meta)
2831 {
2832 	int i, error;
2833 
2834 	error = 0;
2835 
2836 	for (i = 0; i < VM_MAXCPU; i++) {
2837 		error = vmmops_vmcx_snapshot(vm->cookie, meta, i);
2838 		if (error != 0) {
2839 			printf("%s: failed to snapshot vmcs/vmcb data for "
2840 			       "vCPU: %d; error: %d\n", __func__, i, error);
2841 			goto done;
2842 		}
2843 	}
2844 
2845 done:
2846 	return (error);
2847 }
2848 
2849 /*
2850  * Save kernel-side structures to user-space for snapshotting.
2851  */
2852 int
2853 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta)
2854 {
2855 	int ret = 0;
2856 
2857 	switch (meta->dev_req) {
2858 	case STRUCT_VMX:
2859 		ret = vmmops_snapshot(vm->cookie, meta);
2860 		break;
2861 	case STRUCT_VMCX:
2862 		ret = vm_snapshot_vmcx(vm, meta);
2863 		break;
2864 	case STRUCT_VM:
2865 		ret = vm_snapshot_vm(vm, meta);
2866 		break;
2867 	case STRUCT_VIOAPIC:
2868 		ret = vioapic_snapshot(vm_ioapic(vm), meta);
2869 		break;
2870 	case STRUCT_VLAPIC:
2871 		ret = vlapic_snapshot(vm, meta);
2872 		break;
2873 	case STRUCT_VHPET:
2874 		ret = vhpet_snapshot(vm_hpet(vm), meta);
2875 		break;
2876 	case STRUCT_VATPIC:
2877 		ret = vatpic_snapshot(vm_atpic(vm), meta);
2878 		break;
2879 	case STRUCT_VATPIT:
2880 		ret = vatpit_snapshot(vm_atpit(vm), meta);
2881 		break;
2882 	case STRUCT_VPMTMR:
2883 		ret = vpmtmr_snapshot(vm_pmtmr(vm), meta);
2884 		break;
2885 	case STRUCT_VRTC:
2886 		ret = vrtc_snapshot(vm_rtc(vm), meta);
2887 		break;
2888 	default:
2889 		printf("%s: failed to find the requested type %#x\n",
2890 		       __func__, meta->dev_req);
2891 		ret = (EINVAL);
2892 	}
2893 	return (ret);
2894 }
2895 
2896 int
2897 vm_set_tsc_offset(struct vm *vm, int vcpuid, uint64_t offset)
2898 {
2899 	struct vcpu *vcpu;
2900 
2901 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2902 		return (EINVAL);
2903 
2904 	vcpu = &vm->vcpu[vcpuid];
2905 	vcpu->tsc_offset = offset;
2906 
2907 	return (0);
2908 }
2909 
2910 int
2911 vm_restore_time(struct vm *vm)
2912 {
2913 	int error, i;
2914 	uint64_t now;
2915 	struct vcpu *vcpu;
2916 
2917 	now = rdtsc();
2918 
2919 	error = vhpet_restore_time(vm_hpet(vm));
2920 	if (error)
2921 		return (error);
2922 
2923 	for (i = 0; i < nitems(vm->vcpu); i++) {
2924 		vcpu = &vm->vcpu[i];
2925 
2926 		error = vmmops_restore_tsc(vm->cookie, i, vcpu->tsc_offset -
2927 		    now);
2928 		if (error)
2929 			return (error);
2930 	}
2931 
2932 	return (0);
2933 }
2934 #endif
2935