xref: /freebsd/sys/amd64/vmm/vmm.c (revision d9a42747950146bf03cda7f6e25d219253f8a57a)
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/sx.h>
50 #include <sys/vnode.h>
51 
52 #include <vm/vm.h>
53 #include <vm/vm_param.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_page.h>
57 #include <vm/pmap.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_pager.h>
60 #include <vm/vm_kern.h>
61 #include <vm/vnode_pager.h>
62 #include <vm/swap_pager.h>
63 #include <vm/uma.h>
64 
65 #include <machine/cpu.h>
66 #include <machine/pcb.h>
67 #include <machine/smp.h>
68 #include <machine/md_var.h>
69 #include <x86/psl.h>
70 #include <x86/apicreg.h>
71 #include <x86/ifunc.h>
72 
73 #include <machine/vmm.h>
74 #include <machine/vmm_dev.h>
75 #include <machine/vmm_instruction_emul.h>
76 #include <machine/vmm_snapshot.h>
77 
78 #include "vmm_ioport.h"
79 #include "vmm_ktr.h"
80 #include "vmm_host.h"
81 #include "vmm_mem.h"
82 #include "vmm_util.h"
83 #include "vatpic.h"
84 #include "vatpit.h"
85 #include "vhpet.h"
86 #include "vioapic.h"
87 #include "vlapic.h"
88 #include "vpmtmr.h"
89 #include "vrtc.h"
90 #include "vmm_stat.h"
91 #include "vmm_lapic.h"
92 
93 #include "io/ppt.h"
94 #include "io/iommu.h"
95 
96 struct vlapic;
97 
98 /*
99  * Initialization:
100  * (a) allocated when vcpu is created
101  * (i) initialized when vcpu is created and when it is reinitialized
102  * (o) initialized the first time the vcpu is created
103  * (x) initialized before use
104  */
105 struct vcpu {
106 	struct mtx 	mtx;		/* (o) protects 'state' and 'hostcpu' */
107 	enum vcpu_state	state;		/* (o) vcpu state */
108 	int		vcpuid;		/* (o) */
109 	int		hostcpu;	/* (o) vcpu's host cpu */
110 	int		reqidle;	/* (i) request vcpu to idle */
111 	struct vm	*vm;		/* (o) */
112 	void		*cookie;	/* (i) cpu-specific data */
113 	struct vlapic	*vlapic;	/* (i) APIC device model */
114 	enum x2apic_state x2apic_state;	/* (i) APIC mode */
115 	uint64_t	exitintinfo;	/* (i) events pending at VM exit */
116 	int		nmi_pending;	/* (i) NMI pending */
117 	int		extint_pending;	/* (i) INTR pending */
118 	int	exception_pending;	/* (i) exception pending */
119 	int	exc_vector;		/* (x) exception collateral */
120 	int	exc_errcode_valid;
121 	uint32_t exc_errcode;
122 	struct savefpu	*guestfpu;	/* (a,i) guest fpu state */
123 	uint64_t	guest_xcr0;	/* (i) guest %xcr0 register */
124 	void		*stats;		/* (a,i) statistics */
125 	struct vm_exit	exitinfo;	/* (x) exit reason and collateral */
126 	uint64_t	nextrip;	/* (x) next instruction to execute */
127 	uint64_t	tsc_offset;	/* (o) TSC offsetting */
128 };
129 
130 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
131 #define	vcpu_lock_destroy(v)	mtx_destroy(&((v)->mtx))
132 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
133 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
134 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
135 
136 struct mem_seg {
137 	size_t	len;
138 	bool	sysmem;
139 	struct vm_object *object;
140 };
141 #define	VM_MAX_MEMSEGS	4
142 
143 struct mem_map {
144 	vm_paddr_t	gpa;
145 	size_t		len;
146 	vm_ooffset_t	segoff;
147 	int		segid;
148 	int		prot;
149 	int		flags;
150 };
151 #define	VM_MAX_MEMMAPS	8
152 
153 /*
154  * Initialization:
155  * (o) initialized the first time the VM is created
156  * (i) initialized when VM is created and when it is reinitialized
157  * (x) initialized before use
158  *
159  * Locking:
160  * [m] mem_segs_lock
161  * [r] rendezvous_mtx
162  * [v] reads require one frozen vcpu, writes require freezing all vcpus
163  */
164 struct vm {
165 	void		*cookie;		/* (i) cpu-specific data */
166 	void		*iommu;			/* (x) iommu-specific data */
167 	struct vhpet	*vhpet;			/* (i) virtual HPET */
168 	struct vioapic	*vioapic;		/* (i) virtual ioapic */
169 	struct vatpic	*vatpic;		/* (i) virtual atpic */
170 	struct vatpit	*vatpit;		/* (i) virtual atpit */
171 	struct vpmtmr	*vpmtmr;		/* (i) virtual ACPI PM timer */
172 	struct vrtc	*vrtc;			/* (o) virtual RTC */
173 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
174 	volatile cpuset_t debug_cpus;		/* (i) vcpus stopped for debug */
175 	cpuset_t	startup_cpus;		/* (i) [r] waiting for startup */
176 	int		suspend;		/* (i) stop VM execution */
177 	bool		dying;			/* (o) is dying */
178 	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
179 	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
180 	cpuset_t	rendezvous_req_cpus;	/* (x) [r] rendezvous requested */
181 	cpuset_t	rendezvous_done_cpus;	/* (x) [r] rendezvous finished */
182 	void		*rendezvous_arg;	/* (x) [r] rendezvous func/arg */
183 	vm_rendezvous_func_t rendezvous_func;
184 	struct mtx	rendezvous_mtx;		/* (o) rendezvous lock */
185 	struct mem_map	mem_maps[VM_MAX_MEMMAPS]; /* (i) [m+v] guest address space */
186 	struct mem_seg	mem_segs[VM_MAX_MEMSEGS]; /* (o) [m+v] guest memory regions */
187 	struct vmspace	*vmspace;		/* (o) guest's address space */
188 	char		name[VM_MAX_NAMELEN+1];	/* (o) virtual machine name */
189 	struct vcpu	**vcpu;			/* (o) guest vcpus */
190 	/* The following describe the vm cpu topology */
191 	uint16_t	sockets;		/* (o) num of sockets */
192 	uint16_t	cores;			/* (o) num of cores/socket */
193 	uint16_t	threads;		/* (o) num of threads/core */
194 	uint16_t	maxcpus;		/* (o) max pluggable cpus */
195 	struct sx	mem_segs_lock;		/* (o) */
196 	struct sx	vcpus_init_lock;	/* (o) */
197 };
198 
199 #define	VMM_CTR0(vcpu, format)						\
200 	VCPU_CTR0((vcpu)->vm, (vcpu)->vcpuid, format)
201 
202 #define	VMM_CTR1(vcpu, format, p1)					\
203 	VCPU_CTR1((vcpu)->vm, (vcpu)->vcpuid, format, p1)
204 
205 #define	VMM_CTR2(vcpu, format, p1, p2)					\
206 	VCPU_CTR2((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2)
207 
208 #define	VMM_CTR3(vcpu, format, p1, p2, p3)				\
209 	VCPU_CTR3((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3)
210 
211 #define	VMM_CTR4(vcpu, format, p1, p2, p3, p4)				\
212 	VCPU_CTR4((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3, p4)
213 
214 static int vmm_initialized;
215 
216 static void	vmmops_panic(void);
217 
218 static void
219 vmmops_panic(void)
220 {
221 	panic("vmm_ops func called when !vmm_is_intel() && !vmm_is_svm()");
222 }
223 
224 #define	DEFINE_VMMOPS_IFUNC(ret_type, opname, args)			\
225     DEFINE_IFUNC(static, ret_type, vmmops_##opname, args)		\
226     {									\
227     	if (vmm_is_intel())						\
228     		return (vmm_ops_intel.opname);				\
229     	else if (vmm_is_svm())						\
230     		return (vmm_ops_amd.opname);				\
231     	else								\
232     		return ((ret_type (*)args)vmmops_panic);		\
233     }
234 
235 DEFINE_VMMOPS_IFUNC(int, modinit, (int ipinum))
236 DEFINE_VMMOPS_IFUNC(int, modcleanup, (void))
237 DEFINE_VMMOPS_IFUNC(void, modresume, (void))
238 DEFINE_VMMOPS_IFUNC(void *, init, (struct vm *vm, struct pmap *pmap))
239 DEFINE_VMMOPS_IFUNC(int, run, (void *vcpui, register_t rip, struct pmap *pmap,
240     struct vm_eventinfo *info))
241 DEFINE_VMMOPS_IFUNC(void, cleanup, (void *vmi))
242 DEFINE_VMMOPS_IFUNC(void *, vcpu_init, (void *vmi, struct vcpu *vcpu,
243     int vcpu_id))
244 DEFINE_VMMOPS_IFUNC(void, vcpu_cleanup, (void *vcpui))
245 DEFINE_VMMOPS_IFUNC(int, getreg, (void *vcpui, int num, uint64_t *retval))
246 DEFINE_VMMOPS_IFUNC(int, setreg, (void *vcpui, int num, uint64_t val))
247 DEFINE_VMMOPS_IFUNC(int, getdesc, (void *vcpui, int num, struct seg_desc *desc))
248 DEFINE_VMMOPS_IFUNC(int, setdesc, (void *vcpui, int num, struct seg_desc *desc))
249 DEFINE_VMMOPS_IFUNC(int, getcap, (void *vcpui, int num, int *retval))
250 DEFINE_VMMOPS_IFUNC(int, setcap, (void *vcpui, int num, int val))
251 DEFINE_VMMOPS_IFUNC(struct vmspace *, vmspace_alloc, (vm_offset_t min,
252     vm_offset_t max))
253 DEFINE_VMMOPS_IFUNC(void, vmspace_free, (struct vmspace *vmspace))
254 DEFINE_VMMOPS_IFUNC(struct vlapic *, vlapic_init, (void *vcpui))
255 DEFINE_VMMOPS_IFUNC(void, vlapic_cleanup, (struct vlapic *vlapic))
256 #ifdef BHYVE_SNAPSHOT
257 DEFINE_VMMOPS_IFUNC(int, vcpu_snapshot, (void *vcpui,
258     struct vm_snapshot_meta *meta))
259 DEFINE_VMMOPS_IFUNC(int, restore_tsc, (void *vcpui, uint64_t now))
260 #endif
261 
262 #define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
263 #define	fpu_stop_emulating()	clts()
264 
265 SDT_PROVIDER_DEFINE(vmm);
266 
267 static MALLOC_DEFINE(M_VM, "vm", "vm");
268 
269 /* statistics */
270 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
271 
272 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
273     NULL);
274 
275 /*
276  * Halt the guest if all vcpus are executing a HLT instruction with
277  * interrupts disabled.
278  */
279 static int halt_detection_enabled = 1;
280 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
281     &halt_detection_enabled, 0,
282     "Halt VM if all vcpus execute HLT with interrupts disabled");
283 
284 static int vmm_ipinum;
285 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
286     "IPI vector used for vcpu notifications");
287 
288 static int trace_guest_exceptions;
289 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN,
290     &trace_guest_exceptions, 0,
291     "Trap into hypervisor on all guest exceptions and reflect them back");
292 
293 static int trap_wbinvd;
294 SYSCTL_INT(_hw_vmm, OID_AUTO, trap_wbinvd, CTLFLAG_RDTUN, &trap_wbinvd, 0,
295     "WBINVD triggers a VM-exit");
296 
297 u_int vm_maxcpu;
298 SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
299     &vm_maxcpu, 0, "Maximum number of vCPUs");
300 
301 static void vm_free_memmap(struct vm *vm, int ident);
302 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm);
303 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr);
304 
305 /*
306  * Upper limit on vm_maxcpu.  Limited by use of uint16_t types for CPU
307  * counts as well as range of vpid values for VT-x and by the capacity
308  * of cpuset_t masks.  The call to new_unrhdr() in vpid_init() in
309  * vmx.c requires 'vm_maxcpu + 1 <= 0xffff', hence the '- 1' below.
310  */
311 #define	VM_MAXCPU	MIN(0xffff - 1, CPU_SETSIZE)
312 
313 #ifdef KTR
314 static const char *
315 vcpu_state2str(enum vcpu_state state)
316 {
317 
318 	switch (state) {
319 	case VCPU_IDLE:
320 		return ("idle");
321 	case VCPU_FROZEN:
322 		return ("frozen");
323 	case VCPU_RUNNING:
324 		return ("running");
325 	case VCPU_SLEEPING:
326 		return ("sleeping");
327 	default:
328 		return ("unknown");
329 	}
330 }
331 #endif
332 
333 static void
334 vcpu_cleanup(struct vcpu *vcpu, bool destroy)
335 {
336 	vmmops_vlapic_cleanup(vcpu->vlapic);
337 	vmmops_vcpu_cleanup(vcpu->cookie);
338 	vcpu->cookie = NULL;
339 	if (destroy) {
340 		vmm_stat_free(vcpu->stats);
341 		fpu_save_area_free(vcpu->guestfpu);
342 		vcpu_lock_destroy(vcpu);
343 		free(vcpu, M_VM);
344 	}
345 }
346 
347 static struct vcpu *
348 vcpu_alloc(struct vm *vm, int vcpu_id)
349 {
350 	struct vcpu *vcpu;
351 
352 	KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus,
353 	    ("vcpu_init: invalid vcpu %d", vcpu_id));
354 
355 	vcpu = malloc(sizeof(*vcpu), M_VM, M_WAITOK | M_ZERO);
356 	vcpu_lock_init(vcpu);
357 	vcpu->state = VCPU_IDLE;
358 	vcpu->hostcpu = NOCPU;
359 	vcpu->vcpuid = vcpu_id;
360 	vcpu->vm = vm;
361 	vcpu->guestfpu = fpu_save_area_alloc();
362 	vcpu->stats = vmm_stat_alloc();
363 	vcpu->tsc_offset = 0;
364 	return (vcpu);
365 }
366 
367 static void
368 vcpu_init(struct vcpu *vcpu)
369 {
370 	vcpu->cookie = vmmops_vcpu_init(vcpu->vm->cookie, vcpu, vcpu->vcpuid);
371 	vcpu->vlapic = vmmops_vlapic_init(vcpu->cookie);
372 	vm_set_x2apic_state(vcpu, X2APIC_DISABLED);
373 	vcpu->reqidle = 0;
374 	vcpu->exitintinfo = 0;
375 	vcpu->nmi_pending = 0;
376 	vcpu->extint_pending = 0;
377 	vcpu->exception_pending = 0;
378 	vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
379 	fpu_save_area_reset(vcpu->guestfpu);
380 	vmm_stat_init(vcpu->stats);
381 }
382 
383 int
384 vcpu_trace_exceptions(struct vcpu *vcpu)
385 {
386 
387 	return (trace_guest_exceptions);
388 }
389 
390 int
391 vcpu_trap_wbinvd(struct vcpu *vcpu)
392 {
393 	return (trap_wbinvd);
394 }
395 
396 struct vm_exit *
397 vm_exitinfo(struct vcpu *vcpu)
398 {
399 	return (&vcpu->exitinfo);
400 }
401 
402 static int
403 vmm_init(void)
404 {
405 	int error;
406 
407 	if (!vmm_is_hw_supported())
408 		return (ENXIO);
409 
410 	vm_maxcpu = mp_ncpus;
411 	TUNABLE_INT_FETCH("hw.vmm.maxcpu", &vm_maxcpu);
412 
413 	if (vm_maxcpu > VM_MAXCPU) {
414 		printf("vmm: vm_maxcpu clamped to %u\n", VM_MAXCPU);
415 		vm_maxcpu = VM_MAXCPU;
416 	}
417 	if (vm_maxcpu == 0)
418 		vm_maxcpu = 1;
419 
420 	vmm_host_state_init();
421 
422 	vmm_ipinum = lapic_ipi_alloc(pti ? &IDTVEC(justreturn1_pti) :
423 	    &IDTVEC(justreturn));
424 	if (vmm_ipinum < 0)
425 		vmm_ipinum = IPI_AST;
426 
427 	error = vmm_mem_init();
428 	if (error)
429 		return (error);
430 
431 	vmm_resume_p = vmmops_modresume;
432 
433 	return (vmmops_modinit(vmm_ipinum));
434 }
435 
436 static int
437 vmm_handler(module_t mod, int what, void *arg)
438 {
439 	int error;
440 
441 	switch (what) {
442 	case MOD_LOAD:
443 		if (vmm_is_hw_supported()) {
444 			vmmdev_init();
445 			error = vmm_init();
446 			if (error == 0)
447 				vmm_initialized = 1;
448 		} else {
449 			error = ENXIO;
450 		}
451 		break;
452 	case MOD_UNLOAD:
453 		if (vmm_is_hw_supported()) {
454 			error = vmmdev_cleanup();
455 			if (error == 0) {
456 				vmm_resume_p = NULL;
457 				iommu_cleanup();
458 				if (vmm_ipinum != IPI_AST)
459 					lapic_ipi_free(vmm_ipinum);
460 				error = vmmops_modcleanup();
461 				/*
462 				 * Something bad happened - prevent new
463 				 * VMs from being created
464 				 */
465 				if (error)
466 					vmm_initialized = 0;
467 			}
468 		} else {
469 			error = 0;
470 		}
471 		break;
472 	default:
473 		error = 0;
474 		break;
475 	}
476 	return (error);
477 }
478 
479 static moduledata_t vmm_kmod = {
480 	"vmm",
481 	vmm_handler,
482 	NULL
483 };
484 
485 /*
486  * vmm initialization has the following dependencies:
487  *
488  * - VT-x initialization requires smp_rendezvous() and therefore must happen
489  *   after SMP is fully functional (after SI_SUB_SMP).
490  */
491 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
492 MODULE_VERSION(vmm, 1);
493 
494 static void
495 vm_init(struct vm *vm, bool create)
496 {
497 	vm->cookie = vmmops_init(vm, vmspace_pmap(vm->vmspace));
498 	vm->iommu = NULL;
499 	vm->vioapic = vioapic_init(vm);
500 	vm->vhpet = vhpet_init(vm);
501 	vm->vatpic = vatpic_init(vm);
502 	vm->vatpit = vatpit_init(vm);
503 	vm->vpmtmr = vpmtmr_init(vm);
504 	if (create)
505 		vm->vrtc = vrtc_init(vm);
506 
507 	CPU_ZERO(&vm->active_cpus);
508 	CPU_ZERO(&vm->debug_cpus);
509 	CPU_ZERO(&vm->startup_cpus);
510 
511 	vm->suspend = 0;
512 	CPU_ZERO(&vm->suspended_cpus);
513 
514 	if (!create) {
515 		for (int i = 0; i < vm->maxcpus; i++) {
516 			if (vm->vcpu[i] != NULL)
517 				vcpu_init(vm->vcpu[i]);
518 		}
519 	}
520 }
521 
522 void
523 vm_disable_vcpu_creation(struct vm *vm)
524 {
525 	sx_xlock(&vm->vcpus_init_lock);
526 	vm->dying = true;
527 	sx_xunlock(&vm->vcpus_init_lock);
528 }
529 
530 struct vcpu *
531 vm_alloc_vcpu(struct vm *vm, int vcpuid)
532 {
533 	struct vcpu *vcpu;
534 
535 	if (vcpuid < 0 || vcpuid >= vm_get_maxcpus(vm))
536 		return (NULL);
537 
538 	vcpu = atomic_load_ptr(&vm->vcpu[vcpuid]);
539 	if (__predict_true(vcpu != NULL))
540 		return (vcpu);
541 
542 	sx_xlock(&vm->vcpus_init_lock);
543 	vcpu = vm->vcpu[vcpuid];
544 	if (vcpu == NULL && !vm->dying) {
545 		vcpu = vcpu_alloc(vm, vcpuid);
546 		vcpu_init(vcpu);
547 
548 		/*
549 		 * Ensure vCPU is fully created before updating pointer
550 		 * to permit unlocked reads above.
551 		 */
552 		atomic_store_rel_ptr((uintptr_t *)&vm->vcpu[vcpuid],
553 		    (uintptr_t)vcpu);
554 	}
555 	sx_xunlock(&vm->vcpus_init_lock);
556 	return (vcpu);
557 }
558 
559 void
560 vm_slock_vcpus(struct vm *vm)
561 {
562 	sx_slock(&vm->vcpus_init_lock);
563 }
564 
565 void
566 vm_unlock_vcpus(struct vm *vm)
567 {
568 	sx_unlock(&vm->vcpus_init_lock);
569 }
570 
571 /*
572  * The default CPU topology is a single thread per package.
573  */
574 u_int cores_per_package = 1;
575 u_int threads_per_core = 1;
576 
577 int
578 vm_create(const char *name, struct vm **retvm)
579 {
580 	struct vm *vm;
581 	struct vmspace *vmspace;
582 
583 	/*
584 	 * If vmm.ko could not be successfully initialized then don't attempt
585 	 * to create the virtual machine.
586 	 */
587 	if (!vmm_initialized)
588 		return (ENXIO);
589 
590 	if (name == NULL || strnlen(name, VM_MAX_NAMELEN + 1) ==
591 	    VM_MAX_NAMELEN + 1)
592 		return (EINVAL);
593 
594 	vmspace = vmmops_vmspace_alloc(0, VM_MAXUSER_ADDRESS_LA48);
595 	if (vmspace == NULL)
596 		return (ENOMEM);
597 
598 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
599 	strcpy(vm->name, name);
600 	vm->vmspace = vmspace;
601 	mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
602 	sx_init(&vm->mem_segs_lock, "vm mem_segs");
603 	sx_init(&vm->vcpus_init_lock, "vm vcpus");
604 	vm->vcpu = malloc(sizeof(*vm->vcpu) * vm_maxcpu, M_VM, M_WAITOK |
605 	    M_ZERO);
606 
607 	vm->sockets = 1;
608 	vm->cores = cores_per_package;	/* XXX backwards compatibility */
609 	vm->threads = threads_per_core;	/* XXX backwards compatibility */
610 	vm->maxcpus = vm_maxcpu;
611 
612 	vm_init(vm, true);
613 
614 	*retvm = vm;
615 	return (0);
616 }
617 
618 void
619 vm_get_topology(struct vm *vm, uint16_t *sockets, uint16_t *cores,
620     uint16_t *threads, uint16_t *maxcpus)
621 {
622 	*sockets = vm->sockets;
623 	*cores = vm->cores;
624 	*threads = vm->threads;
625 	*maxcpus = vm->maxcpus;
626 }
627 
628 uint16_t
629 vm_get_maxcpus(struct vm *vm)
630 {
631 	return (vm->maxcpus);
632 }
633 
634 int
635 vm_set_topology(struct vm *vm, uint16_t sockets, uint16_t cores,
636     uint16_t threads, uint16_t maxcpus __unused)
637 {
638 	/* Ignore maxcpus. */
639 	if ((sockets * cores * threads) > vm->maxcpus)
640 		return (EINVAL);
641 	vm->sockets = sockets;
642 	vm->cores = cores;
643 	vm->threads = threads;
644 	return(0);
645 }
646 
647 static void
648 vm_cleanup(struct vm *vm, bool destroy)
649 {
650 	struct mem_map *mm;
651 	int i;
652 
653 	if (destroy)
654 		vm_xlock_memsegs(vm);
655 
656 	ppt_unassign_all(vm);
657 
658 	if (vm->iommu != NULL)
659 		iommu_destroy_domain(vm->iommu);
660 
661 	if (destroy)
662 		vrtc_cleanup(vm->vrtc);
663 	else
664 		vrtc_reset(vm->vrtc);
665 	vpmtmr_cleanup(vm->vpmtmr);
666 	vatpit_cleanup(vm->vatpit);
667 	vhpet_cleanup(vm->vhpet);
668 	vatpic_cleanup(vm->vatpic);
669 	vioapic_cleanup(vm->vioapic);
670 
671 	for (i = 0; i < vm->maxcpus; i++) {
672 		if (vm->vcpu[i] != NULL)
673 			vcpu_cleanup(vm->vcpu[i], destroy);
674 	}
675 
676 	vmmops_cleanup(vm->cookie);
677 
678 	/*
679 	 * System memory is removed from the guest address space only when
680 	 * the VM is destroyed. This is because the mapping remains the same
681 	 * across VM reset.
682 	 *
683 	 * Device memory can be relocated by the guest (e.g. using PCI BARs)
684 	 * so those mappings are removed on a VM reset.
685 	 */
686 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
687 		mm = &vm->mem_maps[i];
688 		if (destroy || !sysmem_mapping(vm, mm))
689 			vm_free_memmap(vm, i);
690 	}
691 
692 	if (destroy) {
693 		for (i = 0; i < VM_MAX_MEMSEGS; i++)
694 			vm_free_memseg(vm, i);
695 		vm_unlock_memsegs(vm);
696 
697 		vmmops_vmspace_free(vm->vmspace);
698 		vm->vmspace = NULL;
699 
700 		free(vm->vcpu, M_VM);
701 		sx_destroy(&vm->vcpus_init_lock);
702 		sx_destroy(&vm->mem_segs_lock);
703 		mtx_destroy(&vm->rendezvous_mtx);
704 	}
705 }
706 
707 void
708 vm_destroy(struct vm *vm)
709 {
710 	vm_cleanup(vm, true);
711 	free(vm, M_VM);
712 }
713 
714 int
715 vm_reinit(struct vm *vm)
716 {
717 	int error;
718 
719 	/*
720 	 * A virtual machine can be reset only if all vcpus are suspended.
721 	 */
722 	if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
723 		vm_cleanup(vm, false);
724 		vm_init(vm, false);
725 		error = 0;
726 	} else {
727 		error = EBUSY;
728 	}
729 
730 	return (error);
731 }
732 
733 const char *
734 vm_name(struct vm *vm)
735 {
736 	return (vm->name);
737 }
738 
739 void
740 vm_slock_memsegs(struct vm *vm)
741 {
742 	sx_slock(&vm->mem_segs_lock);
743 }
744 
745 void
746 vm_xlock_memsegs(struct vm *vm)
747 {
748 	sx_xlock(&vm->mem_segs_lock);
749 }
750 
751 void
752 vm_unlock_memsegs(struct vm *vm)
753 {
754 	sx_unlock(&vm->mem_segs_lock);
755 }
756 
757 int
758 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
759 {
760 	vm_object_t obj;
761 
762 	if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
763 		return (ENOMEM);
764 	else
765 		return (0);
766 }
767 
768 int
769 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
770 {
771 
772 	vmm_mmio_free(vm->vmspace, gpa, len);
773 	return (0);
774 }
775 
776 /*
777  * Return 'true' if 'gpa' is allocated in the guest address space.
778  *
779  * This function is called in the context of a running vcpu which acts as
780  * an implicit lock on 'vm->mem_maps[]'.
781  */
782 bool
783 vm_mem_allocated(struct vcpu *vcpu, vm_paddr_t gpa)
784 {
785 	struct vm *vm = vcpu->vm;
786 	struct mem_map *mm;
787 	int i;
788 
789 #ifdef INVARIANTS
790 	int hostcpu, state;
791 	state = vcpu_get_state(vcpu, &hostcpu);
792 	KASSERT(state == VCPU_RUNNING && hostcpu == curcpu,
793 	    ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu));
794 #endif
795 
796 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
797 		mm = &vm->mem_maps[i];
798 		if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len)
799 			return (true);		/* 'gpa' is sysmem or devmem */
800 	}
801 
802 	if (ppt_is_mmio(vm, gpa))
803 		return (true);			/* 'gpa' is pci passthru mmio */
804 
805 	return (false);
806 }
807 
808 int
809 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem)
810 {
811 	struct mem_seg *seg;
812 	vm_object_t obj;
813 
814 	sx_assert(&vm->mem_segs_lock, SX_XLOCKED);
815 
816 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
817 		return (EINVAL);
818 
819 	if (len == 0 || (len & PAGE_MASK))
820 		return (EINVAL);
821 
822 	seg = &vm->mem_segs[ident];
823 	if (seg->object != NULL) {
824 		if (seg->len == len && seg->sysmem == sysmem)
825 			return (EEXIST);
826 		else
827 			return (EINVAL);
828 	}
829 
830 	obj = vm_object_allocate(OBJT_SWAP, len >> PAGE_SHIFT);
831 	if (obj == NULL)
832 		return (ENOMEM);
833 
834 	seg->len = len;
835 	seg->object = obj;
836 	seg->sysmem = sysmem;
837 	return (0);
838 }
839 
840 int
841 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
842     vm_object_t *objptr)
843 {
844 	struct mem_seg *seg;
845 
846 	sx_assert(&vm->mem_segs_lock, SX_LOCKED);
847 
848 	if (ident < 0 || ident >= VM_MAX_MEMSEGS)
849 		return (EINVAL);
850 
851 	seg = &vm->mem_segs[ident];
852 	if (len)
853 		*len = seg->len;
854 	if (sysmem)
855 		*sysmem = seg->sysmem;
856 	if (objptr)
857 		*objptr = seg->object;
858 	return (0);
859 }
860 
861 void
862 vm_free_memseg(struct vm *vm, int ident)
863 {
864 	struct mem_seg *seg;
865 
866 	KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS,
867 	    ("%s: invalid memseg ident %d", __func__, ident));
868 
869 	seg = &vm->mem_segs[ident];
870 	if (seg->object != NULL) {
871 		vm_object_deallocate(seg->object);
872 		bzero(seg, sizeof(struct mem_seg));
873 	}
874 }
875 
876 int
877 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first,
878     size_t len, int prot, int flags)
879 {
880 	struct mem_seg *seg;
881 	struct mem_map *m, *map;
882 	vm_ooffset_t last;
883 	int i, error;
884 
885 	if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0)
886 		return (EINVAL);
887 
888 	if (flags & ~VM_MEMMAP_F_WIRED)
889 		return (EINVAL);
890 
891 	if (segid < 0 || segid >= VM_MAX_MEMSEGS)
892 		return (EINVAL);
893 
894 	seg = &vm->mem_segs[segid];
895 	if (seg->object == NULL)
896 		return (EINVAL);
897 
898 	last = first + len;
899 	if (first < 0 || first >= last || last > seg->len)
900 		return (EINVAL);
901 
902 	if ((gpa | first | last) & PAGE_MASK)
903 		return (EINVAL);
904 
905 	map = NULL;
906 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
907 		m = &vm->mem_maps[i];
908 		if (m->len == 0) {
909 			map = m;
910 			break;
911 		}
912 	}
913 
914 	if (map == NULL)
915 		return (ENOSPC);
916 
917 	error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa,
918 	    len, 0, VMFS_NO_SPACE, prot, prot, 0);
919 	if (error != KERN_SUCCESS)
920 		return (EFAULT);
921 
922 	vm_object_reference(seg->object);
923 
924 	if (flags & VM_MEMMAP_F_WIRED) {
925 		error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len,
926 		    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
927 		if (error != KERN_SUCCESS) {
928 			vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len);
929 			return (error == KERN_RESOURCE_SHORTAGE ? ENOMEM :
930 			    EFAULT);
931 		}
932 	}
933 
934 	map->gpa = gpa;
935 	map->len = len;
936 	map->segoff = first;
937 	map->segid = segid;
938 	map->prot = prot;
939 	map->flags = flags;
940 	return (0);
941 }
942 
943 int
944 vm_munmap_memseg(struct vm *vm, vm_paddr_t gpa, size_t len)
945 {
946 	struct mem_map *m;
947 	int i;
948 
949 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
950 		m = &vm->mem_maps[i];
951 		if (m->gpa == gpa && m->len == len &&
952 		    (m->flags & VM_MEMMAP_F_IOMMU) == 0) {
953 			vm_free_memmap(vm, i);
954 			return (0);
955 		}
956 	}
957 
958 	return (EINVAL);
959 }
960 
961 int
962 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
963     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags)
964 {
965 	struct mem_map *mm, *mmnext;
966 	int i;
967 
968 	mmnext = NULL;
969 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
970 		mm = &vm->mem_maps[i];
971 		if (mm->len == 0 || mm->gpa < *gpa)
972 			continue;
973 		if (mmnext == NULL || mm->gpa < mmnext->gpa)
974 			mmnext = mm;
975 	}
976 
977 	if (mmnext != NULL) {
978 		*gpa = mmnext->gpa;
979 		if (segid)
980 			*segid = mmnext->segid;
981 		if (segoff)
982 			*segoff = mmnext->segoff;
983 		if (len)
984 			*len = mmnext->len;
985 		if (prot)
986 			*prot = mmnext->prot;
987 		if (flags)
988 			*flags = mmnext->flags;
989 		return (0);
990 	} else {
991 		return (ENOENT);
992 	}
993 }
994 
995 static void
996 vm_free_memmap(struct vm *vm, int ident)
997 {
998 	struct mem_map *mm;
999 	int error __diagused;
1000 
1001 	mm = &vm->mem_maps[ident];
1002 	if (mm->len) {
1003 		error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa,
1004 		    mm->gpa + mm->len);
1005 		KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d",
1006 		    __func__, error));
1007 		bzero(mm, sizeof(struct mem_map));
1008 	}
1009 }
1010 
1011 static __inline bool
1012 sysmem_mapping(struct vm *vm, struct mem_map *mm)
1013 {
1014 
1015 	if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem)
1016 		return (true);
1017 	else
1018 		return (false);
1019 }
1020 
1021 vm_paddr_t
1022 vmm_sysmem_maxaddr(struct vm *vm)
1023 {
1024 	struct mem_map *mm;
1025 	vm_paddr_t maxaddr;
1026 	int i;
1027 
1028 	maxaddr = 0;
1029 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
1030 		mm = &vm->mem_maps[i];
1031 		if (sysmem_mapping(vm, mm)) {
1032 			if (maxaddr < mm->gpa + mm->len)
1033 				maxaddr = mm->gpa + mm->len;
1034 		}
1035 	}
1036 	return (maxaddr);
1037 }
1038 
1039 static void
1040 vm_iommu_modify(struct vm *vm, bool map)
1041 {
1042 	int i, sz;
1043 	vm_paddr_t gpa, hpa;
1044 	struct mem_map *mm;
1045 	void *vp, *cookie, *host_domain;
1046 
1047 	sz = PAGE_SIZE;
1048 	host_domain = iommu_host_domain();
1049 
1050 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
1051 		mm = &vm->mem_maps[i];
1052 		if (!sysmem_mapping(vm, mm))
1053 			continue;
1054 
1055 		if (map) {
1056 			KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0,
1057 			    ("iommu map found invalid memmap %#lx/%#lx/%#x",
1058 			    mm->gpa, mm->len, mm->flags));
1059 			if ((mm->flags & VM_MEMMAP_F_WIRED) == 0)
1060 				continue;
1061 			mm->flags |= VM_MEMMAP_F_IOMMU;
1062 		} else {
1063 			if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0)
1064 				continue;
1065 			mm->flags &= ~VM_MEMMAP_F_IOMMU;
1066 			KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0,
1067 			    ("iommu unmap found invalid memmap %#lx/%#lx/%#x",
1068 			    mm->gpa, mm->len, mm->flags));
1069 		}
1070 
1071 		gpa = mm->gpa;
1072 		while (gpa < mm->gpa + mm->len) {
1073 			vp = vm_gpa_hold_global(vm, gpa, PAGE_SIZE,
1074 			    VM_PROT_WRITE, &cookie);
1075 			KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
1076 			    vm_name(vm), gpa));
1077 
1078 			vm_gpa_release(cookie);
1079 
1080 			hpa = DMAP_TO_PHYS((uintptr_t)vp);
1081 			if (map) {
1082 				iommu_create_mapping(vm->iommu, gpa, hpa, sz);
1083 			} else {
1084 				iommu_remove_mapping(vm->iommu, gpa, sz);
1085 			}
1086 
1087 			gpa += PAGE_SIZE;
1088 		}
1089 	}
1090 
1091 	/*
1092 	 * Invalidate the cached translations associated with the domain
1093 	 * from which pages were removed.
1094 	 */
1095 	if (map)
1096 		iommu_invalidate_tlb(host_domain);
1097 	else
1098 		iommu_invalidate_tlb(vm->iommu);
1099 }
1100 
1101 #define	vm_iommu_unmap(vm)	vm_iommu_modify((vm), false)
1102 #define	vm_iommu_map(vm)	vm_iommu_modify((vm), true)
1103 
1104 int
1105 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
1106 {
1107 	int error;
1108 
1109 	error = ppt_unassign_device(vm, bus, slot, func);
1110 	if (error)
1111 		return (error);
1112 
1113 	if (ppt_assigned_devices(vm) == 0)
1114 		vm_iommu_unmap(vm);
1115 
1116 	return (0);
1117 }
1118 
1119 int
1120 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
1121 {
1122 	int error;
1123 	vm_paddr_t maxaddr;
1124 
1125 	/* Set up the IOMMU to do the 'gpa' to 'hpa' translation */
1126 	if (ppt_assigned_devices(vm) == 0) {
1127 		KASSERT(vm->iommu == NULL,
1128 		    ("vm_assign_pptdev: iommu must be NULL"));
1129 		maxaddr = vmm_sysmem_maxaddr(vm);
1130 		vm->iommu = iommu_create_domain(maxaddr);
1131 		if (vm->iommu == NULL)
1132 			return (ENXIO);
1133 		vm_iommu_map(vm);
1134 	}
1135 
1136 	error = ppt_assign_device(vm, bus, slot, func);
1137 	return (error);
1138 }
1139 
1140 static void *
1141 _vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
1142     void **cookie)
1143 {
1144 	int i, count, pageoff;
1145 	struct mem_map *mm;
1146 	vm_page_t m;
1147 
1148 	pageoff = gpa & PAGE_MASK;
1149 	if (len > PAGE_SIZE - pageoff)
1150 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
1151 
1152 	count = 0;
1153 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
1154 		mm = &vm->mem_maps[i];
1155 		if (gpa >= mm->gpa && gpa < mm->gpa + mm->len) {
1156 			count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
1157 			    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
1158 			break;
1159 		}
1160 	}
1161 
1162 	if (count == 1) {
1163 		*cookie = m;
1164 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
1165 	} else {
1166 		*cookie = NULL;
1167 		return (NULL);
1168 	}
1169 }
1170 
1171 void *
1172 vm_gpa_hold(struct vcpu *vcpu, vm_paddr_t gpa, size_t len, int reqprot,
1173     void **cookie)
1174 {
1175 #ifdef INVARIANTS
1176 	/*
1177 	 * The current vcpu should be frozen to ensure 'vm_memmap[]'
1178 	 * stability.
1179 	 */
1180 	int state = vcpu_get_state(vcpu, NULL);
1181 	KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d",
1182 	    __func__, state));
1183 #endif
1184 	return (_vm_gpa_hold(vcpu->vm, gpa, len, reqprot, cookie));
1185 }
1186 
1187 void *
1188 vm_gpa_hold_global(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
1189     void **cookie)
1190 {
1191 	sx_assert(&vm->mem_segs_lock, SX_LOCKED);
1192 	return (_vm_gpa_hold(vm, gpa, len, reqprot, cookie));
1193 }
1194 
1195 void
1196 vm_gpa_release(void *cookie)
1197 {
1198 	vm_page_t m = cookie;
1199 
1200 	vm_page_unwire(m, PQ_ACTIVE);
1201 }
1202 
1203 int
1204 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval)
1205 {
1206 
1207 	if (reg >= VM_REG_LAST)
1208 		return (EINVAL);
1209 
1210 	return (vmmops_getreg(vcpu->cookie, reg, retval));
1211 }
1212 
1213 int
1214 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val)
1215 {
1216 	int error;
1217 
1218 	if (reg >= VM_REG_LAST)
1219 		return (EINVAL);
1220 
1221 	error = vmmops_setreg(vcpu->cookie, reg, val);
1222 	if (error || reg != VM_REG_GUEST_RIP)
1223 		return (error);
1224 
1225 	/* Set 'nextrip' to match the value of %rip */
1226 	VMM_CTR1(vcpu, "Setting nextrip to %#lx", val);
1227 	vcpu->nextrip = val;
1228 	return (0);
1229 }
1230 
1231 static bool
1232 is_descriptor_table(int reg)
1233 {
1234 
1235 	switch (reg) {
1236 	case VM_REG_GUEST_IDTR:
1237 	case VM_REG_GUEST_GDTR:
1238 		return (true);
1239 	default:
1240 		return (false);
1241 	}
1242 }
1243 
1244 static bool
1245 is_segment_register(int reg)
1246 {
1247 
1248 	switch (reg) {
1249 	case VM_REG_GUEST_ES:
1250 	case VM_REG_GUEST_CS:
1251 	case VM_REG_GUEST_SS:
1252 	case VM_REG_GUEST_DS:
1253 	case VM_REG_GUEST_FS:
1254 	case VM_REG_GUEST_GS:
1255 	case VM_REG_GUEST_TR:
1256 	case VM_REG_GUEST_LDTR:
1257 		return (true);
1258 	default:
1259 		return (false);
1260 	}
1261 }
1262 
1263 int
1264 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc)
1265 {
1266 
1267 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1268 		return (EINVAL);
1269 
1270 	return (vmmops_getdesc(vcpu->cookie, reg, desc));
1271 }
1272 
1273 int
1274 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc)
1275 {
1276 
1277 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1278 		return (EINVAL);
1279 
1280 	return (vmmops_setdesc(vcpu->cookie, reg, desc));
1281 }
1282 
1283 static void
1284 restore_guest_fpustate(struct vcpu *vcpu)
1285 {
1286 
1287 	/* flush host state to the pcb */
1288 	fpuexit(curthread);
1289 
1290 	/* restore guest FPU state */
1291 	fpu_stop_emulating();
1292 	fpurestore(vcpu->guestfpu);
1293 
1294 	/* restore guest XCR0 if XSAVE is enabled in the host */
1295 	if (rcr4() & CR4_XSAVE)
1296 		load_xcr(0, vcpu->guest_xcr0);
1297 
1298 	/*
1299 	 * The FPU is now "dirty" with the guest's state so turn on emulation
1300 	 * to trap any access to the FPU by the host.
1301 	 */
1302 	fpu_start_emulating();
1303 }
1304 
1305 static void
1306 save_guest_fpustate(struct vcpu *vcpu)
1307 {
1308 
1309 	if ((rcr0() & CR0_TS) == 0)
1310 		panic("fpu emulation not enabled in host!");
1311 
1312 	/* save guest XCR0 and restore host XCR0 */
1313 	if (rcr4() & CR4_XSAVE) {
1314 		vcpu->guest_xcr0 = rxcr(0);
1315 		load_xcr(0, vmm_get_host_xcr0());
1316 	}
1317 
1318 	/* save guest FPU state */
1319 	fpu_stop_emulating();
1320 	fpusave(vcpu->guestfpu);
1321 	fpu_start_emulating();
1322 }
1323 
1324 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
1325 
1326 static int
1327 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
1328     bool from_idle)
1329 {
1330 	int error;
1331 
1332 	vcpu_assert_locked(vcpu);
1333 
1334 	/*
1335 	 * State transitions from the vmmdev_ioctl() must always begin from
1336 	 * the VCPU_IDLE state. This guarantees that there is only a single
1337 	 * ioctl() operating on a vcpu at any point.
1338 	 */
1339 	if (from_idle) {
1340 		while (vcpu->state != VCPU_IDLE) {
1341 			vcpu->reqidle = 1;
1342 			vcpu_notify_event_locked(vcpu, false);
1343 			VMM_CTR1(vcpu, "vcpu state change from %s to "
1344 			    "idle requested", vcpu_state2str(vcpu->state));
1345 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
1346 		}
1347 	} else {
1348 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1349 		    "vcpu idle state"));
1350 	}
1351 
1352 	if (vcpu->state == VCPU_RUNNING) {
1353 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1354 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1355 	} else {
1356 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1357 		    "vcpu that is not running", vcpu->hostcpu));
1358 	}
1359 
1360 	/*
1361 	 * The following state transitions are allowed:
1362 	 * IDLE -> FROZEN -> IDLE
1363 	 * FROZEN -> RUNNING -> FROZEN
1364 	 * FROZEN -> SLEEPING -> FROZEN
1365 	 */
1366 	switch (vcpu->state) {
1367 	case VCPU_IDLE:
1368 	case VCPU_RUNNING:
1369 	case VCPU_SLEEPING:
1370 		error = (newstate != VCPU_FROZEN);
1371 		break;
1372 	case VCPU_FROZEN:
1373 		error = (newstate == VCPU_FROZEN);
1374 		break;
1375 	default:
1376 		error = 1;
1377 		break;
1378 	}
1379 
1380 	if (error)
1381 		return (EBUSY);
1382 
1383 	VMM_CTR2(vcpu, "vcpu state changed from %s to %s",
1384 	    vcpu_state2str(vcpu->state), vcpu_state2str(newstate));
1385 
1386 	vcpu->state = newstate;
1387 	if (newstate == VCPU_RUNNING)
1388 		vcpu->hostcpu = curcpu;
1389 	else
1390 		vcpu->hostcpu = NOCPU;
1391 
1392 	if (newstate == VCPU_IDLE)
1393 		wakeup(&vcpu->state);
1394 
1395 	return (0);
1396 }
1397 
1398 static void
1399 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate)
1400 {
1401 	int error;
1402 
1403 	if ((error = vcpu_set_state(vcpu, newstate, false)) != 0)
1404 		panic("Error %d setting state to %d\n", error, newstate);
1405 }
1406 
1407 static void
1408 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
1409 {
1410 	int error;
1411 
1412 	if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
1413 		panic("Error %d setting state to %d", error, newstate);
1414 }
1415 
1416 static int
1417 vm_handle_rendezvous(struct vcpu *vcpu)
1418 {
1419 	struct vm *vm = vcpu->vm;
1420 	struct thread *td;
1421 	int error, vcpuid;
1422 
1423 	error = 0;
1424 	vcpuid = vcpu->vcpuid;
1425 	td = curthread;
1426 	mtx_lock(&vm->rendezvous_mtx);
1427 	while (vm->rendezvous_func != NULL) {
1428 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1429 		CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus);
1430 
1431 		if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1432 		    !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1433 			VMM_CTR0(vcpu, "Calling rendezvous func");
1434 			(*vm->rendezvous_func)(vcpu, vm->rendezvous_arg);
1435 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1436 		}
1437 		if (CPU_CMP(&vm->rendezvous_req_cpus,
1438 		    &vm->rendezvous_done_cpus) == 0) {
1439 			VMM_CTR0(vcpu, "Rendezvous completed");
1440 			CPU_ZERO(&vm->rendezvous_req_cpus);
1441 			vm->rendezvous_func = NULL;
1442 			wakeup(&vm->rendezvous_func);
1443 			break;
1444 		}
1445 		VMM_CTR0(vcpu, "Wait for rendezvous completion");
1446 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1447 		    "vmrndv", hz);
1448 		if (td_ast_pending(td, TDA_SUSPEND)) {
1449 			mtx_unlock(&vm->rendezvous_mtx);
1450 			error = thread_check_susp(td, true);
1451 			if (error != 0)
1452 				return (error);
1453 			mtx_lock(&vm->rendezvous_mtx);
1454 		}
1455 	}
1456 	mtx_unlock(&vm->rendezvous_mtx);
1457 	return (0);
1458 }
1459 
1460 /*
1461  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1462  */
1463 static int
1464 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu)
1465 {
1466 	struct vm *vm = vcpu->vm;
1467 	const char *wmesg;
1468 	struct thread *td;
1469 	int error, t, vcpuid, vcpu_halted, vm_halted;
1470 
1471 	vcpuid = vcpu->vcpuid;
1472 	vcpu_halted = 0;
1473 	vm_halted = 0;
1474 	error = 0;
1475 	td = curthread;
1476 
1477 	KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1478 
1479 	vcpu_lock(vcpu);
1480 	while (1) {
1481 		/*
1482 		 * Do a final check for pending NMI or interrupts before
1483 		 * really putting this thread to sleep. Also check for
1484 		 * software events that would cause this vcpu to wakeup.
1485 		 *
1486 		 * These interrupts/events could have happened after the
1487 		 * vcpu returned from vmmops_run() and before it acquired the
1488 		 * vcpu lock above.
1489 		 */
1490 		if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle)
1491 			break;
1492 		if (vm_nmi_pending(vcpu))
1493 			break;
1494 		if (!intr_disabled) {
1495 			if (vm_extint_pending(vcpu) ||
1496 			    vlapic_pending_intr(vcpu->vlapic, NULL)) {
1497 				break;
1498 			}
1499 		}
1500 
1501 		/* Don't go to sleep if the vcpu thread needs to yield */
1502 		if (vcpu_should_yield(vcpu))
1503 			break;
1504 
1505 		if (vcpu_debugged(vcpu))
1506 			break;
1507 
1508 		/*
1509 		 * Some Linux guests implement "halt" by having all vcpus
1510 		 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1511 		 * track of the vcpus that have entered this state. When all
1512 		 * vcpus enter the halted state the virtual machine is halted.
1513 		 */
1514 		if (intr_disabled) {
1515 			wmesg = "vmhalt";
1516 			VMM_CTR0(vcpu, "Halted");
1517 			if (!vcpu_halted && halt_detection_enabled) {
1518 				vcpu_halted = 1;
1519 				CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1520 			}
1521 			if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1522 				vm_halted = 1;
1523 				break;
1524 			}
1525 		} else {
1526 			wmesg = "vmidle";
1527 		}
1528 
1529 		t = ticks;
1530 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1531 		/*
1532 		 * XXX msleep_spin() cannot be interrupted by signals so
1533 		 * wake up periodically to check pending signals.
1534 		 */
1535 		msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1536 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1537 		vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t);
1538 		if (td_ast_pending(td, TDA_SUSPEND)) {
1539 			vcpu_unlock(vcpu);
1540 			error = thread_check_susp(td, false);
1541 			if (error != 0) {
1542 				if (vcpu_halted) {
1543 					CPU_CLR_ATOMIC(vcpuid,
1544 					    &vm->halted_cpus);
1545 				}
1546 				return (error);
1547 			}
1548 			vcpu_lock(vcpu);
1549 		}
1550 	}
1551 
1552 	if (vcpu_halted)
1553 		CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1554 
1555 	vcpu_unlock(vcpu);
1556 
1557 	if (vm_halted)
1558 		vm_suspend(vm, VM_SUSPEND_HALT);
1559 
1560 	return (0);
1561 }
1562 
1563 static int
1564 vm_handle_paging(struct vcpu *vcpu, bool *retu)
1565 {
1566 	struct vm *vm = vcpu->vm;
1567 	int rv, ftype;
1568 	struct vm_map *map;
1569 	struct vm_exit *vme;
1570 
1571 	vme = &vcpu->exitinfo;
1572 
1573 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1574 	    __func__, vme->inst_length));
1575 
1576 	ftype = vme->u.paging.fault_type;
1577 	KASSERT(ftype == VM_PROT_READ ||
1578 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1579 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1580 
1581 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1582 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1583 		    vme->u.paging.gpa, ftype);
1584 		if (rv == 0) {
1585 			VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx",
1586 			    ftype == VM_PROT_READ ? "accessed" : "dirty",
1587 			    vme->u.paging.gpa);
1588 			goto done;
1589 		}
1590 	}
1591 
1592 	map = &vm->vmspace->vm_map;
1593 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL);
1594 
1595 	VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, "
1596 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1597 
1598 	if (rv != KERN_SUCCESS)
1599 		return (EFAULT);
1600 done:
1601 	return (0);
1602 }
1603 
1604 static int
1605 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu)
1606 {
1607 	struct vie *vie;
1608 	struct vm_exit *vme;
1609 	uint64_t gla, gpa, cs_base;
1610 	struct vm_guest_paging *paging;
1611 	mem_region_read_t mread;
1612 	mem_region_write_t mwrite;
1613 	enum vm_cpu_mode cpu_mode;
1614 	int cs_d, error, fault;
1615 
1616 	vme = &vcpu->exitinfo;
1617 
1618 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1619 	    __func__, vme->inst_length));
1620 
1621 	gla = vme->u.inst_emul.gla;
1622 	gpa = vme->u.inst_emul.gpa;
1623 	cs_base = vme->u.inst_emul.cs_base;
1624 	cs_d = vme->u.inst_emul.cs_d;
1625 	vie = &vme->u.inst_emul.vie;
1626 	paging = &vme->u.inst_emul.paging;
1627 	cpu_mode = paging->cpu_mode;
1628 
1629 	VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa);
1630 
1631 	/* Fetch, decode and emulate the faulting instruction */
1632 	if (vie->num_valid == 0) {
1633 		error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base,
1634 		    VIE_INST_SIZE, vie, &fault);
1635 	} else {
1636 		/*
1637 		 * The instruction bytes have already been copied into 'vie'
1638 		 */
1639 		error = fault = 0;
1640 	}
1641 	if (error || fault)
1642 		return (error);
1643 
1644 	if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) {
1645 		VMM_CTR1(vcpu, "Error decoding instruction at %#lx",
1646 		    vme->rip + cs_base);
1647 		*retu = true;	    /* dump instruction bytes in userspace */
1648 		return (0);
1649 	}
1650 
1651 	/*
1652 	 * Update 'nextrip' based on the length of the emulated instruction.
1653 	 */
1654 	vme->inst_length = vie->num_processed;
1655 	vcpu->nextrip += vie->num_processed;
1656 	VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding",
1657 	    vcpu->nextrip);
1658 
1659 	/* return to userland unless this is an in-kernel emulated device */
1660 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1661 		mread = lapic_mmio_read;
1662 		mwrite = lapic_mmio_write;
1663 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1664 		mread = vioapic_mmio_read;
1665 		mwrite = vioapic_mmio_write;
1666 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1667 		mread = vhpet_mmio_read;
1668 		mwrite = vhpet_mmio_write;
1669 	} else {
1670 		*retu = true;
1671 		return (0);
1672 	}
1673 
1674 	error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite,
1675 	    retu);
1676 
1677 	return (error);
1678 }
1679 
1680 static int
1681 vm_handle_suspend(struct vcpu *vcpu, bool *retu)
1682 {
1683 	struct vm *vm = vcpu->vm;
1684 	int error, i;
1685 	struct thread *td;
1686 
1687 	error = 0;
1688 	td = curthread;
1689 
1690 	CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus);
1691 
1692 	/*
1693 	 * Wait until all 'active_cpus' have suspended themselves.
1694 	 *
1695 	 * Since a VM may be suspended at any time including when one or
1696 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1697 	 * handler while we are waiting to prevent a deadlock.
1698 	 */
1699 	vcpu_lock(vcpu);
1700 	while (error == 0) {
1701 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1702 			VMM_CTR0(vcpu, "All vcpus suspended");
1703 			break;
1704 		}
1705 
1706 		if (vm->rendezvous_func == NULL) {
1707 			VMM_CTR0(vcpu, "Sleeping during suspend");
1708 			vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1709 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1710 			vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1711 			if (td_ast_pending(td, TDA_SUSPEND)) {
1712 				vcpu_unlock(vcpu);
1713 				error = thread_check_susp(td, false);
1714 				vcpu_lock(vcpu);
1715 			}
1716 		} else {
1717 			VMM_CTR0(vcpu, "Rendezvous during suspend");
1718 			vcpu_unlock(vcpu);
1719 			error = vm_handle_rendezvous(vcpu);
1720 			vcpu_lock(vcpu);
1721 		}
1722 	}
1723 	vcpu_unlock(vcpu);
1724 
1725 	/*
1726 	 * Wakeup the other sleeping vcpus and return to userspace.
1727 	 */
1728 	for (i = 0; i < vm->maxcpus; i++) {
1729 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1730 			vcpu_notify_event(vm_vcpu(vm, i), false);
1731 		}
1732 	}
1733 
1734 	*retu = true;
1735 	return (error);
1736 }
1737 
1738 static int
1739 vm_handle_reqidle(struct vcpu *vcpu, bool *retu)
1740 {
1741 	vcpu_lock(vcpu);
1742 	KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle));
1743 	vcpu->reqidle = 0;
1744 	vcpu_unlock(vcpu);
1745 	*retu = true;
1746 	return (0);
1747 }
1748 
1749 int
1750 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1751 {
1752 	int i;
1753 
1754 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1755 		return (EINVAL);
1756 
1757 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1758 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1759 		    vm->suspend, how);
1760 		return (EALREADY);
1761 	}
1762 
1763 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1764 
1765 	/*
1766 	 * Notify all active vcpus that they are now suspended.
1767 	 */
1768 	for (i = 0; i < vm->maxcpus; i++) {
1769 		if (CPU_ISSET(i, &vm->active_cpus))
1770 			vcpu_notify_event(vm_vcpu(vm, i), false);
1771 	}
1772 
1773 	return (0);
1774 }
1775 
1776 void
1777 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip)
1778 {
1779 	struct vm *vm = vcpu->vm;
1780 	struct vm_exit *vmexit;
1781 
1782 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1783 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1784 
1785 	vmexit = vm_exitinfo(vcpu);
1786 	vmexit->rip = rip;
1787 	vmexit->inst_length = 0;
1788 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1789 	vmexit->u.suspended.how = vm->suspend;
1790 }
1791 
1792 void
1793 vm_exit_debug(struct vcpu *vcpu, uint64_t rip)
1794 {
1795 	struct vm_exit *vmexit;
1796 
1797 	vmexit = vm_exitinfo(vcpu);
1798 	vmexit->rip = rip;
1799 	vmexit->inst_length = 0;
1800 	vmexit->exitcode = VM_EXITCODE_DEBUG;
1801 }
1802 
1803 void
1804 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip)
1805 {
1806 	struct vm_exit *vmexit;
1807 
1808 	vmexit = vm_exitinfo(vcpu);
1809 	vmexit->rip = rip;
1810 	vmexit->inst_length = 0;
1811 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1812 	vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1);
1813 }
1814 
1815 void
1816 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip)
1817 {
1818 	struct vm_exit *vmexit;
1819 
1820 	vmexit = vm_exitinfo(vcpu);
1821 	vmexit->rip = rip;
1822 	vmexit->inst_length = 0;
1823 	vmexit->exitcode = VM_EXITCODE_REQIDLE;
1824 	vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1);
1825 }
1826 
1827 void
1828 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip)
1829 {
1830 	struct vm_exit *vmexit;
1831 
1832 	vmexit = vm_exitinfo(vcpu);
1833 	vmexit->rip = rip;
1834 	vmexit->inst_length = 0;
1835 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1836 	vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1);
1837 }
1838 
1839 int
1840 vm_run(struct vcpu *vcpu, struct vm_exit *vme_user)
1841 {
1842 	struct vm *vm = vcpu->vm;
1843 	struct vm_eventinfo evinfo;
1844 	int error, vcpuid;
1845 	struct pcb *pcb;
1846 	uint64_t tscval;
1847 	struct vm_exit *vme;
1848 	bool retu, intr_disabled;
1849 	pmap_t pmap;
1850 
1851 	vcpuid = vcpu->vcpuid;
1852 
1853 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1854 		return (EINVAL);
1855 
1856 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1857 		return (EINVAL);
1858 
1859 	pmap = vmspace_pmap(vm->vmspace);
1860 	vme = &vcpu->exitinfo;
1861 	evinfo.rptr = &vm->rendezvous_req_cpus;
1862 	evinfo.sptr = &vm->suspend;
1863 	evinfo.iptr = &vcpu->reqidle;
1864 restart:
1865 	critical_enter();
1866 
1867 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1868 	    ("vm_run: absurd pm_active"));
1869 
1870 	tscval = rdtsc();
1871 
1872 	pcb = PCPU_GET(curpcb);
1873 	set_pcb_flags(pcb, PCB_FULL_IRET);
1874 
1875 	restore_guest_fpustate(vcpu);
1876 
1877 	vcpu_require_state(vcpu, VCPU_RUNNING);
1878 	error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo);
1879 	vcpu_require_state(vcpu, VCPU_FROZEN);
1880 
1881 	save_guest_fpustate(vcpu);
1882 
1883 	vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1884 
1885 	critical_exit();
1886 
1887 	if (error == 0) {
1888 		retu = false;
1889 		vcpu->nextrip = vme->rip + vme->inst_length;
1890 		switch (vme->exitcode) {
1891 		case VM_EXITCODE_REQIDLE:
1892 			error = vm_handle_reqidle(vcpu, &retu);
1893 			break;
1894 		case VM_EXITCODE_SUSPENDED:
1895 			error = vm_handle_suspend(vcpu, &retu);
1896 			break;
1897 		case VM_EXITCODE_IOAPIC_EOI:
1898 			vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector);
1899 			break;
1900 		case VM_EXITCODE_RENDEZVOUS:
1901 			error = vm_handle_rendezvous(vcpu);
1902 			break;
1903 		case VM_EXITCODE_HLT:
1904 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1905 			error = vm_handle_hlt(vcpu, intr_disabled, &retu);
1906 			break;
1907 		case VM_EXITCODE_PAGING:
1908 			error = vm_handle_paging(vcpu, &retu);
1909 			break;
1910 		case VM_EXITCODE_INST_EMUL:
1911 			error = vm_handle_inst_emul(vcpu, &retu);
1912 			break;
1913 		case VM_EXITCODE_INOUT:
1914 		case VM_EXITCODE_INOUT_STR:
1915 			error = vm_handle_inout(vcpu, vme, &retu);
1916 			break;
1917 		case VM_EXITCODE_MONITOR:
1918 		case VM_EXITCODE_MWAIT:
1919 		case VM_EXITCODE_VMINSN:
1920 			vm_inject_ud(vcpu);
1921 			break;
1922 		default:
1923 			retu = true;	/* handled in userland */
1924 			break;
1925 		}
1926 	}
1927 
1928 	/*
1929 	 * VM_EXITCODE_INST_EMUL could access the apic which could transform the
1930 	 * exit code into VM_EXITCODE_IPI.
1931 	 */
1932 	if (error == 0 && vme->exitcode == VM_EXITCODE_IPI)
1933 		error = vm_handle_ipi(vcpu, vme, &retu);
1934 
1935 	if (error == 0 && retu == false)
1936 		goto restart;
1937 
1938 	vmm_stat_incr(vcpu, VMEXIT_USERSPACE, 1);
1939 	VMM_CTR2(vcpu, "retu %d/%d", error, vme->exitcode);
1940 
1941 	/* copy the exit information */
1942 	*vme_user = *vme;
1943 	return (error);
1944 }
1945 
1946 int
1947 vm_restart_instruction(struct vcpu *vcpu)
1948 {
1949 	enum vcpu_state state;
1950 	uint64_t rip;
1951 	int error __diagused;
1952 
1953 	state = vcpu_get_state(vcpu, NULL);
1954 	if (state == VCPU_RUNNING) {
1955 		/*
1956 		 * When a vcpu is "running" the next instruction is determined
1957 		 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
1958 		 * Thus setting 'inst_length' to zero will cause the current
1959 		 * instruction to be restarted.
1960 		 */
1961 		vcpu->exitinfo.inst_length = 0;
1962 		VMM_CTR1(vcpu, "restarting instruction at %#lx by "
1963 		    "setting inst_length to zero", vcpu->exitinfo.rip);
1964 	} else if (state == VCPU_FROZEN) {
1965 		/*
1966 		 * When a vcpu is "frozen" it is outside the critical section
1967 		 * around vmmops_run() and 'nextrip' points to the next
1968 		 * instruction. Thus instruction restart is achieved by setting
1969 		 * 'nextrip' to the vcpu's %rip.
1970 		 */
1971 		error = vm_get_register(vcpu, VM_REG_GUEST_RIP, &rip);
1972 		KASSERT(!error, ("%s: error %d getting rip", __func__, error));
1973 		VMM_CTR2(vcpu, "restarting instruction by updating "
1974 		    "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
1975 		vcpu->nextrip = rip;
1976 	} else {
1977 		panic("%s: invalid state %d", __func__, state);
1978 	}
1979 	return (0);
1980 }
1981 
1982 int
1983 vm_exit_intinfo(struct vcpu *vcpu, uint64_t info)
1984 {
1985 	int type, vector;
1986 
1987 	if (info & VM_INTINFO_VALID) {
1988 		type = info & VM_INTINFO_TYPE;
1989 		vector = info & 0xff;
1990 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1991 			return (EINVAL);
1992 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1993 			return (EINVAL);
1994 		if (info & VM_INTINFO_RSVD)
1995 			return (EINVAL);
1996 	} else {
1997 		info = 0;
1998 	}
1999 	VMM_CTR2(vcpu, "%s: info1(%#lx)", __func__, info);
2000 	vcpu->exitintinfo = info;
2001 	return (0);
2002 }
2003 
2004 enum exc_class {
2005 	EXC_BENIGN,
2006 	EXC_CONTRIBUTORY,
2007 	EXC_PAGEFAULT
2008 };
2009 
2010 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
2011 
2012 static enum exc_class
2013 exception_class(uint64_t info)
2014 {
2015 	int type, vector;
2016 
2017 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
2018 	type = info & VM_INTINFO_TYPE;
2019 	vector = info & 0xff;
2020 
2021 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
2022 	switch (type) {
2023 	case VM_INTINFO_HWINTR:
2024 	case VM_INTINFO_SWINTR:
2025 	case VM_INTINFO_NMI:
2026 		return (EXC_BENIGN);
2027 	default:
2028 		/*
2029 		 * Hardware exception.
2030 		 *
2031 		 * SVM and VT-x use identical type values to represent NMI,
2032 		 * hardware interrupt and software interrupt.
2033 		 *
2034 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
2035 		 * for exceptions except #BP and #OF. #BP and #OF use a type
2036 		 * value of '5' or '6'. Therefore we don't check for explicit
2037 		 * values of 'type' to classify 'intinfo' into a hardware
2038 		 * exception.
2039 		 */
2040 		break;
2041 	}
2042 
2043 	switch (vector) {
2044 	case IDT_PF:
2045 	case IDT_VE:
2046 		return (EXC_PAGEFAULT);
2047 	case IDT_DE:
2048 	case IDT_TS:
2049 	case IDT_NP:
2050 	case IDT_SS:
2051 	case IDT_GP:
2052 		return (EXC_CONTRIBUTORY);
2053 	default:
2054 		return (EXC_BENIGN);
2055 	}
2056 }
2057 
2058 static int
2059 nested_fault(struct vcpu *vcpu, uint64_t info1, uint64_t info2,
2060     uint64_t *retinfo)
2061 {
2062 	enum exc_class exc1, exc2;
2063 	int type1, vector1;
2064 
2065 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
2066 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
2067 
2068 	/*
2069 	 * If an exception occurs while attempting to call the double-fault
2070 	 * handler the processor enters shutdown mode (aka triple fault).
2071 	 */
2072 	type1 = info1 & VM_INTINFO_TYPE;
2073 	vector1 = info1 & 0xff;
2074 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
2075 		VMM_CTR2(vcpu, "triple fault: info1(%#lx), info2(%#lx)",
2076 		    info1, info2);
2077 		vm_suspend(vcpu->vm, VM_SUSPEND_TRIPLEFAULT);
2078 		*retinfo = 0;
2079 		return (0);
2080 	}
2081 
2082 	/*
2083 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
2084 	 */
2085 	exc1 = exception_class(info1);
2086 	exc2 = exception_class(info2);
2087 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
2088 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
2089 		/* Convert nested fault into a double fault. */
2090 		*retinfo = IDT_DF;
2091 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
2092 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
2093 	} else {
2094 		/* Handle exceptions serially */
2095 		*retinfo = info2;
2096 	}
2097 	return (1);
2098 }
2099 
2100 static uint64_t
2101 vcpu_exception_intinfo(struct vcpu *vcpu)
2102 {
2103 	uint64_t info = 0;
2104 
2105 	if (vcpu->exception_pending) {
2106 		info = vcpu->exc_vector & 0xff;
2107 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
2108 		if (vcpu->exc_errcode_valid) {
2109 			info |= VM_INTINFO_DEL_ERRCODE;
2110 			info |= (uint64_t)vcpu->exc_errcode << 32;
2111 		}
2112 	}
2113 	return (info);
2114 }
2115 
2116 int
2117 vm_entry_intinfo(struct vcpu *vcpu, uint64_t *retinfo)
2118 {
2119 	uint64_t info1, info2;
2120 	int valid;
2121 
2122 	info1 = vcpu->exitintinfo;
2123 	vcpu->exitintinfo = 0;
2124 
2125 	info2 = 0;
2126 	if (vcpu->exception_pending) {
2127 		info2 = vcpu_exception_intinfo(vcpu);
2128 		vcpu->exception_pending = 0;
2129 		VMM_CTR2(vcpu, "Exception %d delivered: %#lx",
2130 		    vcpu->exc_vector, info2);
2131 	}
2132 
2133 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
2134 		valid = nested_fault(vcpu, info1, info2, retinfo);
2135 	} else if (info1 & VM_INTINFO_VALID) {
2136 		*retinfo = info1;
2137 		valid = 1;
2138 	} else if (info2 & VM_INTINFO_VALID) {
2139 		*retinfo = info2;
2140 		valid = 1;
2141 	} else {
2142 		valid = 0;
2143 	}
2144 
2145 	if (valid) {
2146 		VMM_CTR4(vcpu, "%s: info1(%#lx), info2(%#lx), "
2147 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
2148 	}
2149 
2150 	return (valid);
2151 }
2152 
2153 int
2154 vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2)
2155 {
2156 	*info1 = vcpu->exitintinfo;
2157 	*info2 = vcpu_exception_intinfo(vcpu);
2158 	return (0);
2159 }
2160 
2161 int
2162 vm_inject_exception(struct vcpu *vcpu, int vector, int errcode_valid,
2163     uint32_t errcode, int restart_instruction)
2164 {
2165 	uint64_t regval;
2166 	int error __diagused;
2167 
2168 	if (vector < 0 || vector >= 32)
2169 		return (EINVAL);
2170 
2171 	/*
2172 	 * A double fault exception should never be injected directly into
2173 	 * the guest. It is a derived exception that results from specific
2174 	 * combinations of nested faults.
2175 	 */
2176 	if (vector == IDT_DF)
2177 		return (EINVAL);
2178 
2179 	if (vcpu->exception_pending) {
2180 		VMM_CTR2(vcpu, "Unable to inject exception %d due to "
2181 		    "pending exception %d", vector, vcpu->exc_vector);
2182 		return (EBUSY);
2183 	}
2184 
2185 	if (errcode_valid) {
2186 		/*
2187 		 * Exceptions don't deliver an error code in real mode.
2188 		 */
2189 		error = vm_get_register(vcpu, VM_REG_GUEST_CR0, &regval);
2190 		KASSERT(!error, ("%s: error %d getting CR0", __func__, error));
2191 		if (!(regval & CR0_PE))
2192 			errcode_valid = 0;
2193 	}
2194 
2195 	/*
2196 	 * From section 26.6.1 "Interruptibility State" in Intel SDM:
2197 	 *
2198 	 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
2199 	 * one instruction or incurs an exception.
2200 	 */
2201 	error = vm_set_register(vcpu, VM_REG_GUEST_INTR_SHADOW, 0);
2202 	KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
2203 	    __func__, error));
2204 
2205 	if (restart_instruction)
2206 		vm_restart_instruction(vcpu);
2207 
2208 	vcpu->exception_pending = 1;
2209 	vcpu->exc_vector = vector;
2210 	vcpu->exc_errcode = errcode;
2211 	vcpu->exc_errcode_valid = errcode_valid;
2212 	VMM_CTR1(vcpu, "Exception %d pending", vector);
2213 	return (0);
2214 }
2215 
2216 void
2217 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode)
2218 {
2219 	int error __diagused, restart_instruction;
2220 
2221 	restart_instruction = 1;
2222 
2223 	error = vm_inject_exception(vcpu, vector, errcode_valid,
2224 	    errcode, restart_instruction);
2225 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
2226 }
2227 
2228 void
2229 vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2)
2230 {
2231 	int error __diagused;
2232 
2233 	VMM_CTR2(vcpu, "Injecting page fault: error_code %#x, cr2 %#lx",
2234 	    error_code, cr2);
2235 
2236 	error = vm_set_register(vcpu, VM_REG_GUEST_CR2, cr2);
2237 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
2238 
2239 	vm_inject_fault(vcpu, IDT_PF, 1, error_code);
2240 }
2241 
2242 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
2243 
2244 int
2245 vm_inject_nmi(struct vcpu *vcpu)
2246 {
2247 
2248 	vcpu->nmi_pending = 1;
2249 	vcpu_notify_event(vcpu, false);
2250 	return (0);
2251 }
2252 
2253 int
2254 vm_nmi_pending(struct vcpu *vcpu)
2255 {
2256 	return (vcpu->nmi_pending);
2257 }
2258 
2259 void
2260 vm_nmi_clear(struct vcpu *vcpu)
2261 {
2262 	if (vcpu->nmi_pending == 0)
2263 		panic("vm_nmi_clear: inconsistent nmi_pending state");
2264 
2265 	vcpu->nmi_pending = 0;
2266 	vmm_stat_incr(vcpu, VCPU_NMI_COUNT, 1);
2267 }
2268 
2269 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
2270 
2271 int
2272 vm_inject_extint(struct vcpu *vcpu)
2273 {
2274 
2275 	vcpu->extint_pending = 1;
2276 	vcpu_notify_event(vcpu, false);
2277 	return (0);
2278 }
2279 
2280 int
2281 vm_extint_pending(struct vcpu *vcpu)
2282 {
2283 	return (vcpu->extint_pending);
2284 }
2285 
2286 void
2287 vm_extint_clear(struct vcpu *vcpu)
2288 {
2289 	if (vcpu->extint_pending == 0)
2290 		panic("vm_extint_clear: inconsistent extint_pending state");
2291 
2292 	vcpu->extint_pending = 0;
2293 	vmm_stat_incr(vcpu, VCPU_EXTINT_COUNT, 1);
2294 }
2295 
2296 int
2297 vm_get_capability(struct vcpu *vcpu, int type, int *retval)
2298 {
2299 	if (type < 0 || type >= VM_CAP_MAX)
2300 		return (EINVAL);
2301 
2302 	return (vmmops_getcap(vcpu->cookie, type, retval));
2303 }
2304 
2305 int
2306 vm_set_capability(struct vcpu *vcpu, int type, int val)
2307 {
2308 	if (type < 0 || type >= VM_CAP_MAX)
2309 		return (EINVAL);
2310 
2311 	return (vmmops_setcap(vcpu->cookie, type, val));
2312 }
2313 
2314 struct vm *
2315 vcpu_vm(struct vcpu *vcpu)
2316 {
2317 	return (vcpu->vm);
2318 }
2319 
2320 int
2321 vcpu_vcpuid(struct vcpu *vcpu)
2322 {
2323 	return (vcpu->vcpuid);
2324 }
2325 
2326 struct vcpu *
2327 vm_vcpu(struct vm *vm, int vcpuid)
2328 {
2329 	return (vm->vcpu[vcpuid]);
2330 }
2331 
2332 struct vlapic *
2333 vm_lapic(struct vcpu *vcpu)
2334 {
2335 	return (vcpu->vlapic);
2336 }
2337 
2338 struct vioapic *
2339 vm_ioapic(struct vm *vm)
2340 {
2341 
2342 	return (vm->vioapic);
2343 }
2344 
2345 struct vhpet *
2346 vm_hpet(struct vm *vm)
2347 {
2348 
2349 	return (vm->vhpet);
2350 }
2351 
2352 bool
2353 vmm_is_pptdev(int bus, int slot, int func)
2354 {
2355 	int b, f, i, n, s;
2356 	char *val, *cp, *cp2;
2357 	bool found;
2358 
2359 	/*
2360 	 * XXX
2361 	 * The length of an environment variable is limited to 128 bytes which
2362 	 * puts an upper limit on the number of passthru devices that may be
2363 	 * specified using a single environment variable.
2364 	 *
2365 	 * Work around this by scanning multiple environment variable
2366 	 * names instead of a single one - yuck!
2367 	 */
2368 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2369 
2370 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2371 	found = false;
2372 	for (i = 0; names[i] != NULL && !found; i++) {
2373 		cp = val = kern_getenv(names[i]);
2374 		while (cp != NULL && *cp != '\0') {
2375 			if ((cp2 = strchr(cp, ' ')) != NULL)
2376 				*cp2 = '\0';
2377 
2378 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2379 			if (n == 3 && bus == b && slot == s && func == f) {
2380 				found = true;
2381 				break;
2382 			}
2383 
2384 			if (cp2 != NULL)
2385 				*cp2++ = ' ';
2386 
2387 			cp = cp2;
2388 		}
2389 		freeenv(val);
2390 	}
2391 	return (found);
2392 }
2393 
2394 void *
2395 vm_iommu_domain(struct vm *vm)
2396 {
2397 
2398 	return (vm->iommu);
2399 }
2400 
2401 int
2402 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle)
2403 {
2404 	int error;
2405 
2406 	vcpu_lock(vcpu);
2407 	error = vcpu_set_state_locked(vcpu, newstate, from_idle);
2408 	vcpu_unlock(vcpu);
2409 
2410 	return (error);
2411 }
2412 
2413 enum vcpu_state
2414 vcpu_get_state(struct vcpu *vcpu, int *hostcpu)
2415 {
2416 	enum vcpu_state state;
2417 
2418 	vcpu_lock(vcpu);
2419 	state = vcpu->state;
2420 	if (hostcpu != NULL)
2421 		*hostcpu = vcpu->hostcpu;
2422 	vcpu_unlock(vcpu);
2423 
2424 	return (state);
2425 }
2426 
2427 int
2428 vm_activate_cpu(struct vcpu *vcpu)
2429 {
2430 	struct vm *vm = vcpu->vm;
2431 
2432 	if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus))
2433 		return (EBUSY);
2434 
2435 	VMM_CTR0(vcpu, "activated");
2436 	CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus);
2437 	return (0);
2438 }
2439 
2440 int
2441 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu)
2442 {
2443 	if (vcpu == NULL) {
2444 		vm->debug_cpus = vm->active_cpus;
2445 		for (int i = 0; i < vm->maxcpus; i++) {
2446 			if (CPU_ISSET(i, &vm->active_cpus))
2447 				vcpu_notify_event(vm_vcpu(vm, i), false);
2448 		}
2449 	} else {
2450 		if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus))
2451 			return (EINVAL);
2452 
2453 		CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus);
2454 		vcpu_notify_event(vcpu, false);
2455 	}
2456 	return (0);
2457 }
2458 
2459 int
2460 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu)
2461 {
2462 
2463 	if (vcpu == NULL) {
2464 		CPU_ZERO(&vm->debug_cpus);
2465 	} else {
2466 		if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus))
2467 			return (EINVAL);
2468 
2469 		CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus);
2470 	}
2471 	return (0);
2472 }
2473 
2474 int
2475 vcpu_debugged(struct vcpu *vcpu)
2476 {
2477 
2478 	return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus));
2479 }
2480 
2481 cpuset_t
2482 vm_active_cpus(struct vm *vm)
2483 {
2484 
2485 	return (vm->active_cpus);
2486 }
2487 
2488 cpuset_t
2489 vm_debug_cpus(struct vm *vm)
2490 {
2491 
2492 	return (vm->debug_cpus);
2493 }
2494 
2495 cpuset_t
2496 vm_suspended_cpus(struct vm *vm)
2497 {
2498 
2499 	return (vm->suspended_cpus);
2500 }
2501 
2502 /*
2503  * Returns the subset of vCPUs in tostart that are awaiting startup.
2504  * These vCPUs are also marked as no longer awaiting startup.
2505  */
2506 cpuset_t
2507 vm_start_cpus(struct vm *vm, const cpuset_t *tostart)
2508 {
2509 	cpuset_t set;
2510 
2511 	mtx_lock(&vm->rendezvous_mtx);
2512 	CPU_AND(&set, &vm->startup_cpus, tostart);
2513 	CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set);
2514 	mtx_unlock(&vm->rendezvous_mtx);
2515 	return (set);
2516 }
2517 
2518 void
2519 vm_await_start(struct vm *vm, const cpuset_t *waiting)
2520 {
2521 	mtx_lock(&vm->rendezvous_mtx);
2522 	CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting);
2523 	mtx_unlock(&vm->rendezvous_mtx);
2524 }
2525 
2526 void *
2527 vcpu_stats(struct vcpu *vcpu)
2528 {
2529 
2530 	return (vcpu->stats);
2531 }
2532 
2533 int
2534 vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state)
2535 {
2536 	*state = vcpu->x2apic_state;
2537 
2538 	return (0);
2539 }
2540 
2541 int
2542 vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state)
2543 {
2544 	if (state >= X2APIC_STATE_LAST)
2545 		return (EINVAL);
2546 
2547 	vcpu->x2apic_state = state;
2548 
2549 	vlapic_set_x2apic_state(vcpu, state);
2550 
2551 	return (0);
2552 }
2553 
2554 /*
2555  * This function is called to ensure that a vcpu "sees" a pending event
2556  * as soon as possible:
2557  * - If the vcpu thread is sleeping then it is woken up.
2558  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2559  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2560  */
2561 static void
2562 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
2563 {
2564 	int hostcpu;
2565 
2566 	hostcpu = vcpu->hostcpu;
2567 	if (vcpu->state == VCPU_RUNNING) {
2568 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2569 		if (hostcpu != curcpu) {
2570 			if (lapic_intr) {
2571 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2572 				    vmm_ipinum);
2573 			} else {
2574 				ipi_cpu(hostcpu, vmm_ipinum);
2575 			}
2576 		} else {
2577 			/*
2578 			 * If the 'vcpu' is running on 'curcpu' then it must
2579 			 * be sending a notification to itself (e.g. SELF_IPI).
2580 			 * The pending event will be picked up when the vcpu
2581 			 * transitions back to guest context.
2582 			 */
2583 		}
2584 	} else {
2585 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2586 		    "with hostcpu %d", vcpu->state, hostcpu));
2587 		if (vcpu->state == VCPU_SLEEPING)
2588 			wakeup_one(vcpu);
2589 	}
2590 }
2591 
2592 void
2593 vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr)
2594 {
2595 	vcpu_lock(vcpu);
2596 	vcpu_notify_event_locked(vcpu, lapic_intr);
2597 	vcpu_unlock(vcpu);
2598 }
2599 
2600 struct vmspace *
2601 vm_get_vmspace(struct vm *vm)
2602 {
2603 
2604 	return (vm->vmspace);
2605 }
2606 
2607 int
2608 vm_apicid2vcpuid(struct vm *vm, int apicid)
2609 {
2610 	/*
2611 	 * XXX apic id is assumed to be numerically identical to vcpu id
2612 	 */
2613 	return (apicid);
2614 }
2615 
2616 int
2617 vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest,
2618     vm_rendezvous_func_t func, void *arg)
2619 {
2620 	struct vm *vm = vcpu->vm;
2621 	int error, i;
2622 
2623 	/*
2624 	 * Enforce that this function is called without any locks
2625 	 */
2626 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2627 
2628 restart:
2629 	mtx_lock(&vm->rendezvous_mtx);
2630 	if (vm->rendezvous_func != NULL) {
2631 		/*
2632 		 * If a rendezvous is already in progress then we need to
2633 		 * call the rendezvous handler in case this 'vcpu' is one
2634 		 * of the targets of the rendezvous.
2635 		 */
2636 		VMM_CTR0(vcpu, "Rendezvous already in progress");
2637 		mtx_unlock(&vm->rendezvous_mtx);
2638 		error = vm_handle_rendezvous(vcpu);
2639 		if (error != 0)
2640 			return (error);
2641 		goto restart;
2642 	}
2643 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2644 	    "rendezvous is still in progress"));
2645 
2646 	VMM_CTR0(vcpu, "Initiating rendezvous");
2647 	vm->rendezvous_req_cpus = dest;
2648 	CPU_ZERO(&vm->rendezvous_done_cpus);
2649 	vm->rendezvous_arg = arg;
2650 	vm->rendezvous_func = func;
2651 	mtx_unlock(&vm->rendezvous_mtx);
2652 
2653 	/*
2654 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2655 	 * vcpus so they handle the rendezvous as soon as possible.
2656 	 */
2657 	for (i = 0; i < vm->maxcpus; i++) {
2658 		if (CPU_ISSET(i, &dest))
2659 			vcpu_notify_event(vm_vcpu(vm, i), false);
2660 	}
2661 
2662 	return (vm_handle_rendezvous(vcpu));
2663 }
2664 
2665 struct vatpic *
2666 vm_atpic(struct vm *vm)
2667 {
2668 	return (vm->vatpic);
2669 }
2670 
2671 struct vatpit *
2672 vm_atpit(struct vm *vm)
2673 {
2674 	return (vm->vatpit);
2675 }
2676 
2677 struct vpmtmr *
2678 vm_pmtmr(struct vm *vm)
2679 {
2680 
2681 	return (vm->vpmtmr);
2682 }
2683 
2684 struct vrtc *
2685 vm_rtc(struct vm *vm)
2686 {
2687 
2688 	return (vm->vrtc);
2689 }
2690 
2691 enum vm_reg_name
2692 vm_segment_name(int seg)
2693 {
2694 	static enum vm_reg_name seg_names[] = {
2695 		VM_REG_GUEST_ES,
2696 		VM_REG_GUEST_CS,
2697 		VM_REG_GUEST_SS,
2698 		VM_REG_GUEST_DS,
2699 		VM_REG_GUEST_FS,
2700 		VM_REG_GUEST_GS
2701 	};
2702 
2703 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2704 	    ("%s: invalid segment encoding %d", __func__, seg));
2705 	return (seg_names[seg]);
2706 }
2707 
2708 void
2709 vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo)
2710 {
2711 	int idx;
2712 
2713 	for (idx = 0; idx < num_copyinfo; idx++) {
2714 		if (copyinfo[idx].cookie != NULL)
2715 			vm_gpa_release(copyinfo[idx].cookie);
2716 	}
2717 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2718 }
2719 
2720 int
2721 vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging,
2722     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2723     int num_copyinfo, int *fault)
2724 {
2725 	int error, idx, nused;
2726 	size_t n, off, remaining;
2727 	void *hva, *cookie;
2728 	uint64_t gpa;
2729 
2730 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2731 
2732 	nused = 0;
2733 	remaining = len;
2734 	while (remaining > 0) {
2735 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2736 		error = vm_gla2gpa(vcpu, paging, gla, prot, &gpa, fault);
2737 		if (error || *fault)
2738 			return (error);
2739 		off = gpa & PAGE_MASK;
2740 		n = min(remaining, PAGE_SIZE - off);
2741 		copyinfo[nused].gpa = gpa;
2742 		copyinfo[nused].len = n;
2743 		remaining -= n;
2744 		gla += n;
2745 		nused++;
2746 	}
2747 
2748 	for (idx = 0; idx < nused; idx++) {
2749 		hva = vm_gpa_hold(vcpu, copyinfo[idx].gpa,
2750 		    copyinfo[idx].len, prot, &cookie);
2751 		if (hva == NULL)
2752 			break;
2753 		copyinfo[idx].hva = hva;
2754 		copyinfo[idx].cookie = cookie;
2755 	}
2756 
2757 	if (idx != nused) {
2758 		vm_copy_teardown(copyinfo, num_copyinfo);
2759 		return (EFAULT);
2760 	} else {
2761 		*fault = 0;
2762 		return (0);
2763 	}
2764 }
2765 
2766 void
2767 vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len)
2768 {
2769 	char *dst;
2770 	int idx;
2771 
2772 	dst = kaddr;
2773 	idx = 0;
2774 	while (len > 0) {
2775 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2776 		len -= copyinfo[idx].len;
2777 		dst += copyinfo[idx].len;
2778 		idx++;
2779 	}
2780 }
2781 
2782 void
2783 vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len)
2784 {
2785 	const char *src;
2786 	int idx;
2787 
2788 	src = kaddr;
2789 	idx = 0;
2790 	while (len > 0) {
2791 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2792 		len -= copyinfo[idx].len;
2793 		src += copyinfo[idx].len;
2794 		idx++;
2795 	}
2796 }
2797 
2798 /*
2799  * Return the amount of in-use and wired memory for the VM. Since
2800  * these are global stats, only return the values with for vCPU 0
2801  */
2802 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2803 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2804 
2805 static void
2806 vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat)
2807 {
2808 
2809 	if (vcpu->vcpuid == 0) {
2810 		vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE *
2811 		    vmspace_resident_count(vcpu->vm->vmspace));
2812 	}
2813 }
2814 
2815 static void
2816 vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat)
2817 {
2818 
2819 	if (vcpu->vcpuid == 0) {
2820 		vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE *
2821 		    pmap_wired_count(vmspace_pmap(vcpu->vm->vmspace)));
2822 	}
2823 }
2824 
2825 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2826 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2827 
2828 #ifdef BHYVE_SNAPSHOT
2829 static int
2830 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta)
2831 {
2832 	uint64_t tsc, now;
2833 	int ret;
2834 	struct vcpu *vcpu;
2835 	uint16_t i, maxcpus;
2836 
2837 	now = rdtsc();
2838 	maxcpus = vm_get_maxcpus(vm);
2839 	for (i = 0; i < maxcpus; i++) {
2840 		vcpu = vm->vcpu[i];
2841 		if (vcpu == NULL)
2842 			continue;
2843 
2844 		SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done);
2845 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done);
2846 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done);
2847 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done);
2848 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done);
2849 		SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done);
2850 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done);
2851 		SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done);
2852 
2853 		/*
2854 		 * Save the absolute TSC value by adding now to tsc_offset.
2855 		 *
2856 		 * It will be turned turned back into an actual offset when the
2857 		 * TSC restore function is called
2858 		 */
2859 		tsc = now + vcpu->tsc_offset;
2860 		SNAPSHOT_VAR_OR_LEAVE(tsc, meta, ret, done);
2861 		if (meta->op == VM_SNAPSHOT_RESTORE)
2862 			vcpu->tsc_offset = tsc;
2863 	}
2864 
2865 done:
2866 	return (ret);
2867 }
2868 
2869 static int
2870 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta)
2871 {
2872 	int ret;
2873 
2874 	ret = vm_snapshot_vcpus(vm, meta);
2875 	if (ret != 0)
2876 		goto done;
2877 
2878 	SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done);
2879 done:
2880 	return (ret);
2881 }
2882 
2883 static int
2884 vm_snapshot_vcpu(struct vm *vm, struct vm_snapshot_meta *meta)
2885 {
2886 	int error;
2887 	struct vcpu *vcpu;
2888 	uint16_t i, maxcpus;
2889 
2890 	error = 0;
2891 
2892 	maxcpus = vm_get_maxcpus(vm);
2893 	for (i = 0; i < maxcpus; i++) {
2894 		vcpu = vm->vcpu[i];
2895 		if (vcpu == NULL)
2896 			continue;
2897 
2898 		error = vmmops_vcpu_snapshot(vcpu->cookie, meta);
2899 		if (error != 0) {
2900 			printf("%s: failed to snapshot vmcs/vmcb data for "
2901 			       "vCPU: %d; error: %d\n", __func__, i, error);
2902 			goto done;
2903 		}
2904 	}
2905 
2906 done:
2907 	return (error);
2908 }
2909 
2910 /*
2911  * Save kernel-side structures to user-space for snapshotting.
2912  */
2913 int
2914 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta)
2915 {
2916 	int ret = 0;
2917 
2918 	switch (meta->dev_req) {
2919 	case STRUCT_VMCX:
2920 		ret = vm_snapshot_vcpu(vm, meta);
2921 		break;
2922 	case STRUCT_VM:
2923 		ret = vm_snapshot_vm(vm, meta);
2924 		break;
2925 	case STRUCT_VIOAPIC:
2926 		ret = vioapic_snapshot(vm_ioapic(vm), meta);
2927 		break;
2928 	case STRUCT_VLAPIC:
2929 		ret = vlapic_snapshot(vm, meta);
2930 		break;
2931 	case STRUCT_VHPET:
2932 		ret = vhpet_snapshot(vm_hpet(vm), meta);
2933 		break;
2934 	case STRUCT_VATPIC:
2935 		ret = vatpic_snapshot(vm_atpic(vm), meta);
2936 		break;
2937 	case STRUCT_VATPIT:
2938 		ret = vatpit_snapshot(vm_atpit(vm), meta);
2939 		break;
2940 	case STRUCT_VPMTMR:
2941 		ret = vpmtmr_snapshot(vm_pmtmr(vm), meta);
2942 		break;
2943 	case STRUCT_VRTC:
2944 		ret = vrtc_snapshot(vm_rtc(vm), meta);
2945 		break;
2946 	default:
2947 		printf("%s: failed to find the requested type %#x\n",
2948 		       __func__, meta->dev_req);
2949 		ret = (EINVAL);
2950 	}
2951 	return (ret);
2952 }
2953 
2954 void
2955 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset)
2956 {
2957 	vcpu->tsc_offset = offset;
2958 }
2959 
2960 int
2961 vm_restore_time(struct vm *vm)
2962 {
2963 	int error;
2964 	uint64_t now;
2965 	struct vcpu *vcpu;
2966 	uint16_t i, maxcpus;
2967 
2968 	now = rdtsc();
2969 
2970 	error = vhpet_restore_time(vm_hpet(vm));
2971 	if (error)
2972 		return (error);
2973 
2974 	maxcpus = vm_get_maxcpus(vm);
2975 	for (i = 0; i < maxcpus; i++) {
2976 		vcpu = vm->vcpu[i];
2977 		if (vcpu == NULL)
2978 			continue;
2979 
2980 		error = vmmops_restore_tsc(vcpu->cookie,
2981 		    vcpu->tsc_offset - now);
2982 		if (error)
2983 			return (error);
2984 	}
2985 
2986 	return (0);
2987 }
2988 #endif
2989