xref: /freebsd/sys/amd64/vmm/vmm.c (revision f5463265955b829775bbb32e1fd0bc11dafc36ce)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 
29 #include <sys/cdefs.h>
30 #include "opt_bhyve_snapshot.h"
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/module.h>
36 #include <sys/sysctl.h>
37 #include <sys/malloc.h>
38 #include <sys/pcpu.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/proc.h>
42 #include <sys/rwlock.h>
43 #include <sys/sched.h>
44 #include <sys/smp.h>
45 #include <sys/sx.h>
46 #include <sys/vnode.h>
47 
48 #include <vm/vm.h>
49 #include <vm/vm_param.h>
50 #include <vm/vm_extern.h>
51 #include <vm/vm_object.h>
52 #include <vm/vm_page.h>
53 #include <vm/pmap.h>
54 #include <vm/vm_map.h>
55 #include <vm/vm_pager.h>
56 #include <vm/vm_kern.h>
57 #include <vm/vnode_pager.h>
58 #include <vm/swap_pager.h>
59 #include <vm/uma.h>
60 
61 #include <machine/cpu.h>
62 #include <machine/pcb.h>
63 #include <machine/smp.h>
64 #include <machine/md_var.h>
65 #include <x86/psl.h>
66 #include <x86/apicreg.h>
67 #include <x86/ifunc.h>
68 
69 #include <machine/vmm.h>
70 #include <machine/vmm_dev.h>
71 #include <machine/vmm_instruction_emul.h>
72 #include <machine/vmm_snapshot.h>
73 
74 #include "vmm_ioport.h"
75 #include "vmm_ktr.h"
76 #include "vmm_host.h"
77 #include "vmm_mem.h"
78 #include "vmm_util.h"
79 #include "vatpic.h"
80 #include "vatpit.h"
81 #include "vhpet.h"
82 #include "vioapic.h"
83 #include "vlapic.h"
84 #include "vpmtmr.h"
85 #include "vrtc.h"
86 #include "vmm_stat.h"
87 #include "vmm_lapic.h"
88 
89 #include "io/ppt.h"
90 #include "io/iommu.h"
91 
92 struct vlapic;
93 
94 /*
95  * Initialization:
96  * (a) allocated when vcpu is created
97  * (i) initialized when vcpu is created and when it is reinitialized
98  * (o) initialized the first time the vcpu is created
99  * (x) initialized before use
100  */
101 struct vcpu {
102 	struct mtx 	mtx;		/* (o) protects 'state' and 'hostcpu' */
103 	enum vcpu_state	state;		/* (o) vcpu state */
104 	int		vcpuid;		/* (o) */
105 	int		hostcpu;	/* (o) vcpu's host cpu */
106 	int		reqidle;	/* (i) request vcpu to idle */
107 	struct vm	*vm;		/* (o) */
108 	void		*cookie;	/* (i) cpu-specific data */
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 	cpuset_t	exitinfo_cpuset; /* (x) storage for vmexit handlers */
123 	uint64_t	nextrip;	/* (x) next instruction to execute */
124 	uint64_t	tsc_offset;	/* (o) TSC offsetting */
125 };
126 
127 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
128 #define	vcpu_lock_destroy(v)	mtx_destroy(&((v)->mtx))
129 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
130 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
131 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
132 
133 struct mem_seg {
134 	size_t	len;
135 	bool	sysmem;
136 	struct vm_object *object;
137 };
138 #define	VM_MAX_MEMSEGS	4
139 
140 struct mem_map {
141 	vm_paddr_t	gpa;
142 	size_t		len;
143 	vm_ooffset_t	segoff;
144 	int		segid;
145 	int		prot;
146 	int		flags;
147 };
148 #define	VM_MAX_MEMMAPS	8
149 
150 /*
151  * Initialization:
152  * (o) initialized the first time the VM is created
153  * (i) initialized when VM is created and when it is reinitialized
154  * (x) initialized before use
155  *
156  * Locking:
157  * [m] mem_segs_lock
158  * [r] rendezvous_mtx
159  * [v] reads require one frozen vcpu, writes require freezing all vcpus
160  */
161 struct vm {
162 	void		*cookie;		/* (i) cpu-specific data */
163 	void		*iommu;			/* (x) iommu-specific data */
164 	struct vhpet	*vhpet;			/* (i) virtual HPET */
165 	struct vioapic	*vioapic;		/* (i) virtual ioapic */
166 	struct vatpic	*vatpic;		/* (i) virtual atpic */
167 	struct vatpit	*vatpit;		/* (i) virtual atpit */
168 	struct vpmtmr	*vpmtmr;		/* (i) virtual ACPI PM timer */
169 	struct vrtc	*vrtc;			/* (o) virtual RTC */
170 	volatile cpuset_t active_cpus;		/* (i) active vcpus */
171 	volatile cpuset_t debug_cpus;		/* (i) vcpus stopped for debug */
172 	cpuset_t	startup_cpus;		/* (i) [r] waiting for startup */
173 	int		suspend;		/* (i) stop VM execution */
174 	bool		dying;			/* (o) is dying */
175 	volatile cpuset_t suspended_cpus; 	/* (i) suspended vcpus */
176 	volatile cpuset_t halted_cpus;		/* (x) cpus in a hard halt */
177 	cpuset_t	rendezvous_req_cpus;	/* (x) [r] rendezvous requested */
178 	cpuset_t	rendezvous_done_cpus;	/* (x) [r] rendezvous finished */
179 	void		*rendezvous_arg;	/* (x) [r] rendezvous func/arg */
180 	vm_rendezvous_func_t rendezvous_func;
181 	struct mtx	rendezvous_mtx;		/* (o) rendezvous lock */
182 	struct mem_map	mem_maps[VM_MAX_MEMMAPS]; /* (i) [m+v] guest address space */
183 	struct mem_seg	mem_segs[VM_MAX_MEMSEGS]; /* (o) [m+v] guest memory regions */
184 	struct vmspace	*vmspace;		/* (o) guest's address space */
185 	char		name[VM_MAX_NAMELEN+1];	/* (o) virtual machine name */
186 	struct vcpu	**vcpu;			/* (o) guest vcpus */
187 	/* The following describe the vm cpu topology */
188 	uint16_t	sockets;		/* (o) num of sockets */
189 	uint16_t	cores;			/* (o) num of cores/socket */
190 	uint16_t	threads;		/* (o) num of threads/core */
191 	uint16_t	maxcpus;		/* (o) max pluggable cpus */
192 	struct sx	mem_segs_lock;		/* (o) */
193 	struct sx	vcpus_init_lock;	/* (o) */
194 };
195 
196 #define	VMM_CTR0(vcpu, format)						\
197 	VCPU_CTR0((vcpu)->vm, (vcpu)->vcpuid, format)
198 
199 #define	VMM_CTR1(vcpu, format, p1)					\
200 	VCPU_CTR1((vcpu)->vm, (vcpu)->vcpuid, format, p1)
201 
202 #define	VMM_CTR2(vcpu, format, p1, p2)					\
203 	VCPU_CTR2((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2)
204 
205 #define	VMM_CTR3(vcpu, format, p1, p2, p3)				\
206 	VCPU_CTR3((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3)
207 
208 #define	VMM_CTR4(vcpu, format, p1, p2, p3, p4)				\
209 	VCPU_CTR4((vcpu)->vm, (vcpu)->vcpuid, format, p1, p2, p3, p4)
210 
211 static int vmm_initialized;
212 
213 static void	vmmops_panic(void);
214 
215 static void
216 vmmops_panic(void)
217 {
218 	panic("vmm_ops func called when !vmm_is_intel() && !vmm_is_svm()");
219 }
220 
221 #define	DEFINE_VMMOPS_IFUNC(ret_type, opname, args)			\
222     DEFINE_IFUNC(static, ret_type, vmmops_##opname, args)		\
223     {									\
224     	if (vmm_is_intel())						\
225     		return (vmm_ops_intel.opname);				\
226     	else if (vmm_is_svm())						\
227     		return (vmm_ops_amd.opname);				\
228     	else								\
229     		return ((ret_type (*)args)vmmops_panic);		\
230     }
231 
232 DEFINE_VMMOPS_IFUNC(int, modinit, (int ipinum))
233 DEFINE_VMMOPS_IFUNC(int, modcleanup, (void))
234 DEFINE_VMMOPS_IFUNC(void, modresume, (void))
235 DEFINE_VMMOPS_IFUNC(void *, init, (struct vm *vm, struct pmap *pmap))
236 DEFINE_VMMOPS_IFUNC(int, run, (void *vcpui, register_t rip, struct pmap *pmap,
237     struct vm_eventinfo *info))
238 DEFINE_VMMOPS_IFUNC(void, cleanup, (void *vmi))
239 DEFINE_VMMOPS_IFUNC(void *, vcpu_init, (void *vmi, struct vcpu *vcpu,
240     int vcpu_id))
241 DEFINE_VMMOPS_IFUNC(void, vcpu_cleanup, (void *vcpui))
242 DEFINE_VMMOPS_IFUNC(int, getreg, (void *vcpui, int num, uint64_t *retval))
243 DEFINE_VMMOPS_IFUNC(int, setreg, (void *vcpui, int num, uint64_t val))
244 DEFINE_VMMOPS_IFUNC(int, getdesc, (void *vcpui, int num, struct seg_desc *desc))
245 DEFINE_VMMOPS_IFUNC(int, setdesc, (void *vcpui, int num, struct seg_desc *desc))
246 DEFINE_VMMOPS_IFUNC(int, getcap, (void *vcpui, int num, int *retval))
247 DEFINE_VMMOPS_IFUNC(int, setcap, (void *vcpui, int num, int val))
248 DEFINE_VMMOPS_IFUNC(struct vmspace *, vmspace_alloc, (vm_offset_t min,
249     vm_offset_t max))
250 DEFINE_VMMOPS_IFUNC(void, vmspace_free, (struct vmspace *vmspace))
251 DEFINE_VMMOPS_IFUNC(struct vlapic *, vlapic_init, (void *vcpui))
252 DEFINE_VMMOPS_IFUNC(void, vlapic_cleanup, (struct vlapic *vlapic))
253 #ifdef BHYVE_SNAPSHOT
254 DEFINE_VMMOPS_IFUNC(int, vcpu_snapshot, (void *vcpui,
255     struct vm_snapshot_meta *meta))
256 DEFINE_VMMOPS_IFUNC(int, restore_tsc, (void *vcpui, uint64_t now))
257 #endif
258 
259 SDT_PROVIDER_DEFINE(vmm);
260 
261 static MALLOC_DEFINE(M_VM, "vm", "vm");
262 
263 /* statistics */
264 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
265 
266 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL,
267     NULL);
268 
269 /*
270  * Halt the guest if all vcpus are executing a HLT instruction with
271  * interrupts disabled.
272  */
273 static int halt_detection_enabled = 1;
274 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
275     &halt_detection_enabled, 0,
276     "Halt VM if all vcpus execute HLT with interrupts disabled");
277 
278 static int vmm_ipinum;
279 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
280     "IPI vector used for vcpu notifications");
281 
282 static int trace_guest_exceptions;
283 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN,
284     &trace_guest_exceptions, 0,
285     "Trap into hypervisor on all guest exceptions and reflect them back");
286 
287 static int trap_wbinvd;
288 SYSCTL_INT(_hw_vmm, OID_AUTO, trap_wbinvd, CTLFLAG_RDTUN, &trap_wbinvd, 0,
289     "WBINVD triggers a VM-exit");
290 
291 u_int vm_maxcpu;
292 SYSCTL_UINT(_hw_vmm, OID_AUTO, maxcpu, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
293     &vm_maxcpu, 0, "Maximum number of vCPUs");
294 
295 static void vm_free_memmap(struct vm *vm, int ident);
296 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm);
297 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr);
298 
299 /*
300  * Upper limit on vm_maxcpu.  Limited by use of uint16_t types for CPU
301  * counts as well as range of vpid values for VT-x and by the capacity
302  * of cpuset_t masks.  The call to new_unrhdr() in vpid_init() in
303  * vmx.c requires 'vm_maxcpu + 1 <= 0xffff', hence the '- 1' below.
304  */
305 #define	VM_MAXCPU	MIN(0xffff - 1, CPU_SETSIZE)
306 
307 #ifdef KTR
308 static const char *
309 vcpu_state2str(enum vcpu_state state)
310 {
311 
312 	switch (state) {
313 	case VCPU_IDLE:
314 		return ("idle");
315 	case VCPU_FROZEN:
316 		return ("frozen");
317 	case VCPU_RUNNING:
318 		return ("running");
319 	case VCPU_SLEEPING:
320 		return ("sleeping");
321 	default:
322 		return ("unknown");
323 	}
324 }
325 #endif
326 
327 static void
328 vcpu_cleanup(struct vcpu *vcpu, bool destroy)
329 {
330 	vmmops_vlapic_cleanup(vcpu->vlapic);
331 	vmmops_vcpu_cleanup(vcpu->cookie);
332 	vcpu->cookie = NULL;
333 	if (destroy) {
334 		vmm_stat_free(vcpu->stats);
335 		fpu_save_area_free(vcpu->guestfpu);
336 		vcpu_lock_destroy(vcpu);
337 		free(vcpu, M_VM);
338 	}
339 }
340 
341 static struct vcpu *
342 vcpu_alloc(struct vm *vm, int vcpu_id)
343 {
344 	struct vcpu *vcpu;
345 
346 	KASSERT(vcpu_id >= 0 && vcpu_id < vm->maxcpus,
347 	    ("vcpu_init: invalid vcpu %d", vcpu_id));
348 
349 	vcpu = malloc(sizeof(*vcpu), M_VM, M_WAITOK | M_ZERO);
350 	vcpu_lock_init(vcpu);
351 	vcpu->state = VCPU_IDLE;
352 	vcpu->hostcpu = NOCPU;
353 	vcpu->vcpuid = vcpu_id;
354 	vcpu->vm = vm;
355 	vcpu->guestfpu = fpu_save_area_alloc();
356 	vcpu->stats = vmm_stat_alloc();
357 	vcpu->tsc_offset = 0;
358 	return (vcpu);
359 }
360 
361 static void
362 vcpu_init(struct vcpu *vcpu)
363 {
364 	vcpu->cookie = vmmops_vcpu_init(vcpu->vm->cookie, vcpu, vcpu->vcpuid);
365 	vcpu->vlapic = vmmops_vlapic_init(vcpu->cookie);
366 	vm_set_x2apic_state(vcpu, X2APIC_DISABLED);
367 	vcpu->reqidle = 0;
368 	vcpu->exitintinfo = 0;
369 	vcpu->nmi_pending = 0;
370 	vcpu->extint_pending = 0;
371 	vcpu->exception_pending = 0;
372 	vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
373 	fpu_save_area_reset(vcpu->guestfpu);
374 	vmm_stat_init(vcpu->stats);
375 }
376 
377 int
378 vcpu_trace_exceptions(struct vcpu *vcpu)
379 {
380 
381 	return (trace_guest_exceptions);
382 }
383 
384 int
385 vcpu_trap_wbinvd(struct vcpu *vcpu)
386 {
387 	return (trap_wbinvd);
388 }
389 
390 struct vm_exit *
391 vm_exitinfo(struct vcpu *vcpu)
392 {
393 	return (&vcpu->exitinfo);
394 }
395 
396 cpuset_t *
397 vm_exitinfo_cpuset(struct vcpu *vcpu)
398 {
399 	return (&vcpu->exitinfo_cpuset);
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_map(struct vm *vm)
1041 {
1042 	vm_paddr_t gpa, hpa;
1043 	struct mem_map *mm;
1044 	int i;
1045 
1046 	sx_assert(&vm->mem_segs_lock, SX_LOCKED);
1047 
1048 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
1049 		mm = &vm->mem_maps[i];
1050 		if (!sysmem_mapping(vm, mm))
1051 			continue;
1052 
1053 		KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0,
1054 		    ("iommu map found invalid memmap %#lx/%#lx/%#x",
1055 		    mm->gpa, mm->len, mm->flags));
1056 		if ((mm->flags & VM_MEMMAP_F_WIRED) == 0)
1057 			continue;
1058 		mm->flags |= VM_MEMMAP_F_IOMMU;
1059 
1060 		for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) {
1061 			hpa = pmap_extract(vmspace_pmap(vm->vmspace), gpa);
1062 
1063 			/*
1064 			 * All mappings in the vmm vmspace must be
1065 			 * present since they are managed by vmm in this way.
1066 			 * Because we are in pass-through mode, the
1067 			 * mappings must also be wired.  This implies
1068 			 * that all pages must be mapped and wired,
1069 			 * allowing to use pmap_extract() and avoiding the
1070 			 * need to use vm_gpa_hold_global().
1071 			 *
1072 			 * This could change if/when we start
1073 			 * supporting page faults on IOMMU maps.
1074 			 */
1075 			KASSERT(vm_page_wired(PHYS_TO_VM_PAGE(hpa)),
1076 			    ("vm_iommu_map: vm %p gpa %jx hpa %jx not wired",
1077 			    vm, (uintmax_t)gpa, (uintmax_t)hpa));
1078 
1079 			iommu_create_mapping(vm->iommu, gpa, hpa, PAGE_SIZE);
1080 		}
1081 	}
1082 
1083 	iommu_invalidate_tlb(iommu_host_domain());
1084 }
1085 
1086 static void
1087 vm_iommu_unmap(struct vm *vm)
1088 {
1089 	vm_paddr_t gpa;
1090 	struct mem_map *mm;
1091 	int i;
1092 
1093 	sx_assert(&vm->mem_segs_lock, SX_LOCKED);
1094 
1095 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
1096 		mm = &vm->mem_maps[i];
1097 		if (!sysmem_mapping(vm, mm))
1098 			continue;
1099 
1100 		if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0)
1101 			continue;
1102 		mm->flags &= ~VM_MEMMAP_F_IOMMU;
1103 		KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0,
1104 		    ("iommu unmap found invalid memmap %#lx/%#lx/%#x",
1105 		    mm->gpa, mm->len, mm->flags));
1106 
1107 		for (gpa = mm->gpa; gpa < mm->gpa + mm->len; gpa += PAGE_SIZE) {
1108 			KASSERT(vm_page_wired(PHYS_TO_VM_PAGE(pmap_extract(
1109 			    vmspace_pmap(vm->vmspace), gpa))),
1110 			    ("vm_iommu_unmap: vm %p gpa %jx not wired",
1111 			    vm, (uintmax_t)gpa));
1112 			iommu_remove_mapping(vm->iommu, gpa, PAGE_SIZE);
1113 		}
1114 	}
1115 
1116 	/*
1117 	 * Invalidate the cached translations associated with the domain
1118 	 * from which pages were removed.
1119 	 */
1120 	iommu_invalidate_tlb(vm->iommu);
1121 }
1122 
1123 int
1124 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
1125 {
1126 	int error;
1127 
1128 	error = ppt_unassign_device(vm, bus, slot, func);
1129 	if (error)
1130 		return (error);
1131 
1132 	if (ppt_assigned_devices(vm) == 0)
1133 		vm_iommu_unmap(vm);
1134 
1135 	return (0);
1136 }
1137 
1138 int
1139 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
1140 {
1141 	int error;
1142 	vm_paddr_t maxaddr;
1143 
1144 	/* Set up the IOMMU to do the 'gpa' to 'hpa' translation */
1145 	if (ppt_assigned_devices(vm) == 0) {
1146 		KASSERT(vm->iommu == NULL,
1147 		    ("vm_assign_pptdev: iommu must be NULL"));
1148 		maxaddr = vmm_sysmem_maxaddr(vm);
1149 		vm->iommu = iommu_create_domain(maxaddr);
1150 		if (vm->iommu == NULL)
1151 			return (ENXIO);
1152 		vm_iommu_map(vm);
1153 	}
1154 
1155 	error = ppt_assign_device(vm, bus, slot, func);
1156 	return (error);
1157 }
1158 
1159 static void *
1160 _vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
1161     void **cookie)
1162 {
1163 	int i, count, pageoff;
1164 	struct mem_map *mm;
1165 	vm_page_t m;
1166 
1167 	pageoff = gpa & PAGE_MASK;
1168 	if (len > PAGE_SIZE - pageoff)
1169 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
1170 
1171 	count = 0;
1172 	for (i = 0; i < VM_MAX_MEMMAPS; i++) {
1173 		mm = &vm->mem_maps[i];
1174 		if (gpa >= mm->gpa && gpa < mm->gpa + mm->len) {
1175 			count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
1176 			    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
1177 			break;
1178 		}
1179 	}
1180 
1181 	if (count == 1) {
1182 		*cookie = m;
1183 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
1184 	} else {
1185 		*cookie = NULL;
1186 		return (NULL);
1187 	}
1188 }
1189 
1190 void *
1191 vm_gpa_hold(struct vcpu *vcpu, vm_paddr_t gpa, size_t len, int reqprot,
1192     void **cookie)
1193 {
1194 #ifdef INVARIANTS
1195 	/*
1196 	 * The current vcpu should be frozen to ensure 'vm_memmap[]'
1197 	 * stability.
1198 	 */
1199 	int state = vcpu_get_state(vcpu, NULL);
1200 	KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d",
1201 	    __func__, state));
1202 #endif
1203 	return (_vm_gpa_hold(vcpu->vm, gpa, len, reqprot, cookie));
1204 }
1205 
1206 void *
1207 vm_gpa_hold_global(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
1208     void **cookie)
1209 {
1210 	sx_assert(&vm->mem_segs_lock, SX_LOCKED);
1211 	return (_vm_gpa_hold(vm, gpa, len, reqprot, cookie));
1212 }
1213 
1214 void
1215 vm_gpa_release(void *cookie)
1216 {
1217 	vm_page_t m = cookie;
1218 
1219 	vm_page_unwire(m, PQ_ACTIVE);
1220 }
1221 
1222 int
1223 vm_get_register(struct vcpu *vcpu, int reg, uint64_t *retval)
1224 {
1225 
1226 	if (reg >= VM_REG_LAST)
1227 		return (EINVAL);
1228 
1229 	return (vmmops_getreg(vcpu->cookie, reg, retval));
1230 }
1231 
1232 int
1233 vm_set_register(struct vcpu *vcpu, int reg, uint64_t val)
1234 {
1235 	int error;
1236 
1237 	if (reg >= VM_REG_LAST)
1238 		return (EINVAL);
1239 
1240 	error = vmmops_setreg(vcpu->cookie, reg, val);
1241 	if (error || reg != VM_REG_GUEST_RIP)
1242 		return (error);
1243 
1244 	/* Set 'nextrip' to match the value of %rip */
1245 	VMM_CTR1(vcpu, "Setting nextrip to %#lx", val);
1246 	vcpu->nextrip = val;
1247 	return (0);
1248 }
1249 
1250 static bool
1251 is_descriptor_table(int reg)
1252 {
1253 
1254 	switch (reg) {
1255 	case VM_REG_GUEST_IDTR:
1256 	case VM_REG_GUEST_GDTR:
1257 		return (true);
1258 	default:
1259 		return (false);
1260 	}
1261 }
1262 
1263 static bool
1264 is_segment_register(int reg)
1265 {
1266 
1267 	switch (reg) {
1268 	case VM_REG_GUEST_ES:
1269 	case VM_REG_GUEST_CS:
1270 	case VM_REG_GUEST_SS:
1271 	case VM_REG_GUEST_DS:
1272 	case VM_REG_GUEST_FS:
1273 	case VM_REG_GUEST_GS:
1274 	case VM_REG_GUEST_TR:
1275 	case VM_REG_GUEST_LDTR:
1276 		return (true);
1277 	default:
1278 		return (false);
1279 	}
1280 }
1281 
1282 int
1283 vm_get_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc)
1284 {
1285 
1286 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1287 		return (EINVAL);
1288 
1289 	return (vmmops_getdesc(vcpu->cookie, reg, desc));
1290 }
1291 
1292 int
1293 vm_set_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *desc)
1294 {
1295 
1296 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
1297 		return (EINVAL);
1298 
1299 	return (vmmops_setdesc(vcpu->cookie, reg, desc));
1300 }
1301 
1302 static void
1303 restore_guest_fpustate(struct vcpu *vcpu)
1304 {
1305 
1306 	/* flush host state to the pcb */
1307 	fpuexit(curthread);
1308 
1309 	/* restore guest FPU state */
1310 	fpu_enable();
1311 	fpurestore(vcpu->guestfpu);
1312 
1313 	/* restore guest XCR0 if XSAVE is enabled in the host */
1314 	if (rcr4() & CR4_XSAVE)
1315 		load_xcr(0, vcpu->guest_xcr0);
1316 
1317 	/*
1318 	 * The FPU is now "dirty" with the guest's state so disable
1319 	 * the FPU to trap any access by the host.
1320 	 */
1321 	fpu_disable();
1322 }
1323 
1324 static void
1325 save_guest_fpustate(struct vcpu *vcpu)
1326 {
1327 
1328 	if ((rcr0() & CR0_TS) == 0)
1329 		panic("fpu emulation not enabled in host!");
1330 
1331 	/* save guest XCR0 and restore host XCR0 */
1332 	if (rcr4() & CR4_XSAVE) {
1333 		vcpu->guest_xcr0 = rxcr(0);
1334 		load_xcr(0, vmm_get_host_xcr0());
1335 	}
1336 
1337 	/* save guest FPU state */
1338 	fpu_enable();
1339 	fpusave(vcpu->guestfpu);
1340 	fpu_disable();
1341 }
1342 
1343 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
1344 
1345 static int
1346 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
1347     bool from_idle)
1348 {
1349 	int error;
1350 
1351 	vcpu_assert_locked(vcpu);
1352 
1353 	/*
1354 	 * State transitions from the vmmdev_ioctl() must always begin from
1355 	 * the VCPU_IDLE state. This guarantees that there is only a single
1356 	 * ioctl() operating on a vcpu at any point.
1357 	 */
1358 	if (from_idle) {
1359 		while (vcpu->state != VCPU_IDLE) {
1360 			vcpu->reqidle = 1;
1361 			vcpu_notify_event_locked(vcpu, false);
1362 			VMM_CTR1(vcpu, "vcpu state change from %s to "
1363 			    "idle requested", vcpu_state2str(vcpu->state));
1364 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
1365 		}
1366 	} else {
1367 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1368 		    "vcpu idle state"));
1369 	}
1370 
1371 	if (vcpu->state == VCPU_RUNNING) {
1372 		KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1373 		    "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1374 	} else {
1375 		KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1376 		    "vcpu that is not running", vcpu->hostcpu));
1377 	}
1378 
1379 	/*
1380 	 * The following state transitions are allowed:
1381 	 * IDLE -> FROZEN -> IDLE
1382 	 * FROZEN -> RUNNING -> FROZEN
1383 	 * FROZEN -> SLEEPING -> FROZEN
1384 	 */
1385 	switch (vcpu->state) {
1386 	case VCPU_IDLE:
1387 	case VCPU_RUNNING:
1388 	case VCPU_SLEEPING:
1389 		error = (newstate != VCPU_FROZEN);
1390 		break;
1391 	case VCPU_FROZEN:
1392 		error = (newstate == VCPU_FROZEN);
1393 		break;
1394 	default:
1395 		error = 1;
1396 		break;
1397 	}
1398 
1399 	if (error)
1400 		return (EBUSY);
1401 
1402 	VMM_CTR2(vcpu, "vcpu state changed from %s to %s",
1403 	    vcpu_state2str(vcpu->state), vcpu_state2str(newstate));
1404 
1405 	vcpu->state = newstate;
1406 	if (newstate == VCPU_RUNNING)
1407 		vcpu->hostcpu = curcpu;
1408 	else
1409 		vcpu->hostcpu = NOCPU;
1410 
1411 	if (newstate == VCPU_IDLE)
1412 		wakeup(&vcpu->state);
1413 
1414 	return (0);
1415 }
1416 
1417 static void
1418 vcpu_require_state(struct vcpu *vcpu, enum vcpu_state newstate)
1419 {
1420 	int error;
1421 
1422 	if ((error = vcpu_set_state(vcpu, newstate, false)) != 0)
1423 		panic("Error %d setting state to %d\n", error, newstate);
1424 }
1425 
1426 static void
1427 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
1428 {
1429 	int error;
1430 
1431 	if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
1432 		panic("Error %d setting state to %d", error, newstate);
1433 }
1434 
1435 static int
1436 vm_handle_rendezvous(struct vcpu *vcpu)
1437 {
1438 	struct vm *vm = vcpu->vm;
1439 	struct thread *td;
1440 	int error, vcpuid;
1441 
1442 	error = 0;
1443 	vcpuid = vcpu->vcpuid;
1444 	td = curthread;
1445 	mtx_lock(&vm->rendezvous_mtx);
1446 	while (vm->rendezvous_func != NULL) {
1447 		/* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1448 		CPU_AND(&vm->rendezvous_req_cpus, &vm->rendezvous_req_cpus, &vm->active_cpus);
1449 
1450 		if (CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1451 		    !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1452 			VMM_CTR0(vcpu, "Calling rendezvous func");
1453 			(*vm->rendezvous_func)(vcpu, vm->rendezvous_arg);
1454 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1455 		}
1456 		if (CPU_CMP(&vm->rendezvous_req_cpus,
1457 		    &vm->rendezvous_done_cpus) == 0) {
1458 			VMM_CTR0(vcpu, "Rendezvous completed");
1459 			CPU_ZERO(&vm->rendezvous_req_cpus);
1460 			vm->rendezvous_func = NULL;
1461 			wakeup(&vm->rendezvous_func);
1462 			break;
1463 		}
1464 		VMM_CTR0(vcpu, "Wait for rendezvous completion");
1465 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1466 		    "vmrndv", hz);
1467 		if (td_ast_pending(td, TDA_SUSPEND)) {
1468 			mtx_unlock(&vm->rendezvous_mtx);
1469 			error = thread_check_susp(td, true);
1470 			if (error != 0)
1471 				return (error);
1472 			mtx_lock(&vm->rendezvous_mtx);
1473 		}
1474 	}
1475 	mtx_unlock(&vm->rendezvous_mtx);
1476 	return (0);
1477 }
1478 
1479 /*
1480  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1481  */
1482 static int
1483 vm_handle_hlt(struct vcpu *vcpu, bool intr_disabled, bool *retu)
1484 {
1485 	struct vm *vm = vcpu->vm;
1486 	const char *wmesg;
1487 	struct thread *td;
1488 	int error, t, vcpuid, vcpu_halted, vm_halted;
1489 
1490 	vcpuid = vcpu->vcpuid;
1491 	vcpu_halted = 0;
1492 	vm_halted = 0;
1493 	error = 0;
1494 	td = curthread;
1495 
1496 	KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1497 
1498 	vcpu_lock(vcpu);
1499 	while (1) {
1500 		/*
1501 		 * Do a final check for pending NMI or interrupts before
1502 		 * really putting this thread to sleep. Also check for
1503 		 * software events that would cause this vcpu to wakeup.
1504 		 *
1505 		 * These interrupts/events could have happened after the
1506 		 * vcpu returned from vmmops_run() and before it acquired the
1507 		 * vcpu lock above.
1508 		 */
1509 		if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle)
1510 			break;
1511 		if (vm_nmi_pending(vcpu))
1512 			break;
1513 		if (!intr_disabled) {
1514 			if (vm_extint_pending(vcpu) ||
1515 			    vlapic_pending_intr(vcpu->vlapic, NULL)) {
1516 				break;
1517 			}
1518 		}
1519 
1520 		/* Don't go to sleep if the vcpu thread needs to yield */
1521 		if (vcpu_should_yield(vcpu))
1522 			break;
1523 
1524 		if (vcpu_debugged(vcpu))
1525 			break;
1526 
1527 		/*
1528 		 * Some Linux guests implement "halt" by having all vcpus
1529 		 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1530 		 * track of the vcpus that have entered this state. When all
1531 		 * vcpus enter the halted state the virtual machine is halted.
1532 		 */
1533 		if (intr_disabled) {
1534 			wmesg = "vmhalt";
1535 			VMM_CTR0(vcpu, "Halted");
1536 			if (!vcpu_halted && halt_detection_enabled) {
1537 				vcpu_halted = 1;
1538 				CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1539 			}
1540 			if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1541 				vm_halted = 1;
1542 				break;
1543 			}
1544 		} else {
1545 			wmesg = "vmidle";
1546 		}
1547 
1548 		t = ticks;
1549 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1550 		/*
1551 		 * XXX msleep_spin() cannot be interrupted by signals so
1552 		 * wake up periodically to check pending signals.
1553 		 */
1554 		msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1555 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1556 		vmm_stat_incr(vcpu, VCPU_IDLE_TICKS, ticks - t);
1557 		if (td_ast_pending(td, TDA_SUSPEND)) {
1558 			vcpu_unlock(vcpu);
1559 			error = thread_check_susp(td, false);
1560 			if (error != 0) {
1561 				if (vcpu_halted) {
1562 					CPU_CLR_ATOMIC(vcpuid,
1563 					    &vm->halted_cpus);
1564 				}
1565 				return (error);
1566 			}
1567 			vcpu_lock(vcpu);
1568 		}
1569 	}
1570 
1571 	if (vcpu_halted)
1572 		CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1573 
1574 	vcpu_unlock(vcpu);
1575 
1576 	if (vm_halted)
1577 		vm_suspend(vm, VM_SUSPEND_HALT);
1578 
1579 	return (0);
1580 }
1581 
1582 static int
1583 vm_handle_paging(struct vcpu *vcpu, bool *retu)
1584 {
1585 	struct vm *vm = vcpu->vm;
1586 	int rv, ftype;
1587 	struct vm_map *map;
1588 	struct vm_exit *vme;
1589 
1590 	vme = &vcpu->exitinfo;
1591 
1592 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1593 	    __func__, vme->inst_length));
1594 
1595 	ftype = vme->u.paging.fault_type;
1596 	KASSERT(ftype == VM_PROT_READ ||
1597 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1598 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1599 
1600 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1601 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1602 		    vme->u.paging.gpa, ftype);
1603 		if (rv == 0) {
1604 			VMM_CTR2(vcpu, "%s bit emulation for gpa %#lx",
1605 			    ftype == VM_PROT_READ ? "accessed" : "dirty",
1606 			    vme->u.paging.gpa);
1607 			goto done;
1608 		}
1609 	}
1610 
1611 	map = &vm->vmspace->vm_map;
1612 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL, NULL);
1613 
1614 	VMM_CTR3(vcpu, "vm_handle_paging rv = %d, gpa = %#lx, "
1615 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1616 
1617 	if (rv != KERN_SUCCESS)
1618 		return (EFAULT);
1619 done:
1620 	return (0);
1621 }
1622 
1623 static int
1624 vm_handle_inst_emul(struct vcpu *vcpu, bool *retu)
1625 {
1626 	struct vie *vie;
1627 	struct vm_exit *vme;
1628 	uint64_t gla, gpa, cs_base;
1629 	struct vm_guest_paging *paging;
1630 	mem_region_read_t mread;
1631 	mem_region_write_t mwrite;
1632 	enum vm_cpu_mode cpu_mode;
1633 	int cs_d, error, fault;
1634 
1635 	vme = &vcpu->exitinfo;
1636 
1637 	KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1638 	    __func__, vme->inst_length));
1639 
1640 	gla = vme->u.inst_emul.gla;
1641 	gpa = vme->u.inst_emul.gpa;
1642 	cs_base = vme->u.inst_emul.cs_base;
1643 	cs_d = vme->u.inst_emul.cs_d;
1644 	vie = &vme->u.inst_emul.vie;
1645 	paging = &vme->u.inst_emul.paging;
1646 	cpu_mode = paging->cpu_mode;
1647 
1648 	VMM_CTR1(vcpu, "inst_emul fault accessing gpa %#lx", gpa);
1649 
1650 	/* Fetch, decode and emulate the faulting instruction */
1651 	if (vie->num_valid == 0) {
1652 		error = vmm_fetch_instruction(vcpu, paging, vme->rip + cs_base,
1653 		    VIE_INST_SIZE, vie, &fault);
1654 	} else {
1655 		/*
1656 		 * The instruction bytes have already been copied into 'vie'
1657 		 */
1658 		error = fault = 0;
1659 	}
1660 	if (error || fault)
1661 		return (error);
1662 
1663 	if (vmm_decode_instruction(vcpu, gla, cpu_mode, cs_d, vie) != 0) {
1664 		VMM_CTR1(vcpu, "Error decoding instruction at %#lx",
1665 		    vme->rip + cs_base);
1666 		*retu = true;	    /* dump instruction bytes in userspace */
1667 		return (0);
1668 	}
1669 
1670 	/*
1671 	 * Update 'nextrip' based on the length of the emulated instruction.
1672 	 */
1673 	vme->inst_length = vie->num_processed;
1674 	vcpu->nextrip += vie->num_processed;
1675 	VMM_CTR1(vcpu, "nextrip updated to %#lx after instruction decoding",
1676 	    vcpu->nextrip);
1677 
1678 	/* return to userland unless this is an in-kernel emulated device */
1679 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1680 		mread = lapic_mmio_read;
1681 		mwrite = lapic_mmio_write;
1682 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1683 		mread = vioapic_mmio_read;
1684 		mwrite = vioapic_mmio_write;
1685 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1686 		mread = vhpet_mmio_read;
1687 		mwrite = vhpet_mmio_write;
1688 	} else {
1689 		*retu = true;
1690 		return (0);
1691 	}
1692 
1693 	error = vmm_emulate_instruction(vcpu, gpa, vie, paging, mread, mwrite,
1694 	    retu);
1695 
1696 	return (error);
1697 }
1698 
1699 static int
1700 vm_handle_suspend(struct vcpu *vcpu, bool *retu)
1701 {
1702 	struct vm *vm = vcpu->vm;
1703 	int error, i;
1704 	struct thread *td;
1705 
1706 	error = 0;
1707 	td = curthread;
1708 
1709 	CPU_SET_ATOMIC(vcpu->vcpuid, &vm->suspended_cpus);
1710 
1711 	/*
1712 	 * Wait until all 'active_cpus' have suspended themselves.
1713 	 *
1714 	 * Since a VM may be suspended at any time including when one or
1715 	 * more vcpus are doing a rendezvous we need to call the rendezvous
1716 	 * handler while we are waiting to prevent a deadlock.
1717 	 */
1718 	vcpu_lock(vcpu);
1719 	while (error == 0) {
1720 		if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1721 			VMM_CTR0(vcpu, "All vcpus suspended");
1722 			break;
1723 		}
1724 
1725 		if (vm->rendezvous_func == NULL) {
1726 			VMM_CTR0(vcpu, "Sleeping during suspend");
1727 			vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1728 			msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1729 			vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1730 			if (td_ast_pending(td, TDA_SUSPEND)) {
1731 				vcpu_unlock(vcpu);
1732 				error = thread_check_susp(td, false);
1733 				vcpu_lock(vcpu);
1734 			}
1735 		} else {
1736 			VMM_CTR0(vcpu, "Rendezvous during suspend");
1737 			vcpu_unlock(vcpu);
1738 			error = vm_handle_rendezvous(vcpu);
1739 			vcpu_lock(vcpu);
1740 		}
1741 	}
1742 	vcpu_unlock(vcpu);
1743 
1744 	/*
1745 	 * Wakeup the other sleeping vcpus and return to userspace.
1746 	 */
1747 	for (i = 0; i < vm->maxcpus; i++) {
1748 		if (CPU_ISSET(i, &vm->suspended_cpus)) {
1749 			vcpu_notify_event(vm_vcpu(vm, i), false);
1750 		}
1751 	}
1752 
1753 	*retu = true;
1754 	return (error);
1755 }
1756 
1757 static int
1758 vm_handle_reqidle(struct vcpu *vcpu, bool *retu)
1759 {
1760 	vcpu_lock(vcpu);
1761 	KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle));
1762 	vcpu->reqidle = 0;
1763 	vcpu_unlock(vcpu);
1764 	*retu = true;
1765 	return (0);
1766 }
1767 
1768 static int
1769 vm_handle_db(struct vcpu *vcpu, struct vm_exit *vme, bool *retu)
1770 {
1771 	int error, fault;
1772 	uint64_t rsp;
1773 	uint64_t rflags;
1774 	struct vm_copyinfo copyinfo;
1775 
1776 	*retu = true;
1777 	if (!vme->u.dbg.pushf_intercept || vme->u.dbg.tf_shadow_val != 0) {
1778 		return (0);
1779 	}
1780 
1781 	vm_get_register(vcpu, VM_REG_GUEST_RSP, &rsp);
1782 	error = vm_copy_setup(vcpu, &vme->u.dbg.paging, rsp, sizeof(uint64_t),
1783 	    VM_PROT_RW, &copyinfo, 1, &fault);
1784 	if (error != 0 || fault != 0) {
1785 		*retu = false;
1786 		return (EINVAL);
1787 	}
1788 
1789 	/* Read pushed rflags value from top of stack. */
1790 	vm_copyin(&copyinfo, &rflags, sizeof(uint64_t));
1791 
1792 	/* Clear TF bit. */
1793 	rflags &= ~(PSL_T);
1794 
1795 	/* Write updated value back to memory. */
1796 	vm_copyout(&rflags, &copyinfo, sizeof(uint64_t));
1797 	vm_copy_teardown(&copyinfo, 1);
1798 
1799 	return (0);
1800 }
1801 
1802 int
1803 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1804 {
1805 	int i;
1806 
1807 	if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1808 		return (EINVAL);
1809 
1810 	if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1811 		VM_CTR2(vm, "virtual machine already suspended %d/%d",
1812 		    vm->suspend, how);
1813 		return (EALREADY);
1814 	}
1815 
1816 	VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1817 
1818 	/*
1819 	 * Notify all active vcpus that they are now suspended.
1820 	 */
1821 	for (i = 0; i < vm->maxcpus; i++) {
1822 		if (CPU_ISSET(i, &vm->active_cpus))
1823 			vcpu_notify_event(vm_vcpu(vm, i), false);
1824 	}
1825 
1826 	return (0);
1827 }
1828 
1829 void
1830 vm_exit_suspended(struct vcpu *vcpu, uint64_t rip)
1831 {
1832 	struct vm *vm = vcpu->vm;
1833 	struct vm_exit *vmexit;
1834 
1835 	KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1836 	    ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1837 
1838 	vmexit = vm_exitinfo(vcpu);
1839 	vmexit->rip = rip;
1840 	vmexit->inst_length = 0;
1841 	vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1842 	vmexit->u.suspended.how = vm->suspend;
1843 }
1844 
1845 void
1846 vm_exit_debug(struct vcpu *vcpu, uint64_t rip)
1847 {
1848 	struct vm_exit *vmexit;
1849 
1850 	vmexit = vm_exitinfo(vcpu);
1851 	vmexit->rip = rip;
1852 	vmexit->inst_length = 0;
1853 	vmexit->exitcode = VM_EXITCODE_DEBUG;
1854 }
1855 
1856 void
1857 vm_exit_rendezvous(struct vcpu *vcpu, uint64_t rip)
1858 {
1859 	struct vm_exit *vmexit;
1860 
1861 	vmexit = vm_exitinfo(vcpu);
1862 	vmexit->rip = rip;
1863 	vmexit->inst_length = 0;
1864 	vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1865 	vmm_stat_incr(vcpu, VMEXIT_RENDEZVOUS, 1);
1866 }
1867 
1868 void
1869 vm_exit_reqidle(struct vcpu *vcpu, uint64_t rip)
1870 {
1871 	struct vm_exit *vmexit;
1872 
1873 	vmexit = vm_exitinfo(vcpu);
1874 	vmexit->rip = rip;
1875 	vmexit->inst_length = 0;
1876 	vmexit->exitcode = VM_EXITCODE_REQIDLE;
1877 	vmm_stat_incr(vcpu, VMEXIT_REQIDLE, 1);
1878 }
1879 
1880 void
1881 vm_exit_astpending(struct vcpu *vcpu, uint64_t rip)
1882 {
1883 	struct vm_exit *vmexit;
1884 
1885 	vmexit = vm_exitinfo(vcpu);
1886 	vmexit->rip = rip;
1887 	vmexit->inst_length = 0;
1888 	vmexit->exitcode = VM_EXITCODE_BOGUS;
1889 	vmm_stat_incr(vcpu, VMEXIT_ASTPENDING, 1);
1890 }
1891 
1892 int
1893 vm_run(struct vcpu *vcpu)
1894 {
1895 	struct vm *vm = vcpu->vm;
1896 	struct vm_eventinfo evinfo;
1897 	int error, vcpuid;
1898 	struct pcb *pcb;
1899 	uint64_t tscval;
1900 	struct vm_exit *vme;
1901 	bool retu, intr_disabled;
1902 	pmap_t pmap;
1903 
1904 	vcpuid = vcpu->vcpuid;
1905 
1906 	if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1907 		return (EINVAL);
1908 
1909 	if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1910 		return (EINVAL);
1911 
1912 	pmap = vmspace_pmap(vm->vmspace);
1913 	vme = &vcpu->exitinfo;
1914 	evinfo.rptr = &vm->rendezvous_req_cpus;
1915 	evinfo.sptr = &vm->suspend;
1916 	evinfo.iptr = &vcpu->reqidle;
1917 restart:
1918 	critical_enter();
1919 
1920 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1921 	    ("vm_run: absurd pm_active"));
1922 
1923 	tscval = rdtsc();
1924 
1925 	pcb = PCPU_GET(curpcb);
1926 	set_pcb_flags(pcb, PCB_FULL_IRET);
1927 
1928 	restore_guest_fpustate(vcpu);
1929 
1930 	vcpu_require_state(vcpu, VCPU_RUNNING);
1931 	error = vmmops_run(vcpu->cookie, vcpu->nextrip, pmap, &evinfo);
1932 	vcpu_require_state(vcpu, VCPU_FROZEN);
1933 
1934 	save_guest_fpustate(vcpu);
1935 
1936 	vmm_stat_incr(vcpu, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1937 
1938 	critical_exit();
1939 
1940 	if (error == 0) {
1941 		retu = false;
1942 		vcpu->nextrip = vme->rip + vme->inst_length;
1943 		switch (vme->exitcode) {
1944 		case VM_EXITCODE_REQIDLE:
1945 			error = vm_handle_reqidle(vcpu, &retu);
1946 			break;
1947 		case VM_EXITCODE_SUSPENDED:
1948 			error = vm_handle_suspend(vcpu, &retu);
1949 			break;
1950 		case VM_EXITCODE_IOAPIC_EOI:
1951 			vioapic_process_eoi(vm, vme->u.ioapic_eoi.vector);
1952 			break;
1953 		case VM_EXITCODE_RENDEZVOUS:
1954 			error = vm_handle_rendezvous(vcpu);
1955 			break;
1956 		case VM_EXITCODE_HLT:
1957 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1958 			error = vm_handle_hlt(vcpu, intr_disabled, &retu);
1959 			break;
1960 		case VM_EXITCODE_PAGING:
1961 			error = vm_handle_paging(vcpu, &retu);
1962 			break;
1963 		case VM_EXITCODE_INST_EMUL:
1964 			error = vm_handle_inst_emul(vcpu, &retu);
1965 			break;
1966 		case VM_EXITCODE_INOUT:
1967 		case VM_EXITCODE_INOUT_STR:
1968 			error = vm_handle_inout(vcpu, vme, &retu);
1969 			break;
1970 		case VM_EXITCODE_DB:
1971 			error = vm_handle_db(vcpu, vme, &retu);
1972 			break;
1973 		case VM_EXITCODE_MONITOR:
1974 		case VM_EXITCODE_MWAIT:
1975 		case VM_EXITCODE_VMINSN:
1976 			vm_inject_ud(vcpu);
1977 			break;
1978 		default:
1979 			retu = true;	/* handled in userland */
1980 			break;
1981 		}
1982 	}
1983 
1984 	/*
1985 	 * VM_EXITCODE_INST_EMUL could access the apic which could transform the
1986 	 * exit code into VM_EXITCODE_IPI.
1987 	 */
1988 	if (error == 0 && vme->exitcode == VM_EXITCODE_IPI)
1989 		error = vm_handle_ipi(vcpu, vme, &retu);
1990 
1991 	if (error == 0 && retu == false)
1992 		goto restart;
1993 
1994 	vmm_stat_incr(vcpu, VMEXIT_USERSPACE, 1);
1995 	VMM_CTR2(vcpu, "retu %d/%d", error, vme->exitcode);
1996 
1997 	return (error);
1998 }
1999 
2000 int
2001 vm_restart_instruction(struct vcpu *vcpu)
2002 {
2003 	enum vcpu_state state;
2004 	uint64_t rip;
2005 	int error __diagused;
2006 
2007 	state = vcpu_get_state(vcpu, NULL);
2008 	if (state == VCPU_RUNNING) {
2009 		/*
2010 		 * When a vcpu is "running" the next instruction is determined
2011 		 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
2012 		 * Thus setting 'inst_length' to zero will cause the current
2013 		 * instruction to be restarted.
2014 		 */
2015 		vcpu->exitinfo.inst_length = 0;
2016 		VMM_CTR1(vcpu, "restarting instruction at %#lx by "
2017 		    "setting inst_length to zero", vcpu->exitinfo.rip);
2018 	} else if (state == VCPU_FROZEN) {
2019 		/*
2020 		 * When a vcpu is "frozen" it is outside the critical section
2021 		 * around vmmops_run() and 'nextrip' points to the next
2022 		 * instruction. Thus instruction restart is achieved by setting
2023 		 * 'nextrip' to the vcpu's %rip.
2024 		 */
2025 		error = vm_get_register(vcpu, VM_REG_GUEST_RIP, &rip);
2026 		KASSERT(!error, ("%s: error %d getting rip", __func__, error));
2027 		VMM_CTR2(vcpu, "restarting instruction by updating "
2028 		    "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
2029 		vcpu->nextrip = rip;
2030 	} else {
2031 		panic("%s: invalid state %d", __func__, state);
2032 	}
2033 	return (0);
2034 }
2035 
2036 int
2037 vm_exit_intinfo(struct vcpu *vcpu, uint64_t info)
2038 {
2039 	int type, vector;
2040 
2041 	if (info & VM_INTINFO_VALID) {
2042 		type = info & VM_INTINFO_TYPE;
2043 		vector = info & 0xff;
2044 		if (type == VM_INTINFO_NMI && vector != IDT_NMI)
2045 			return (EINVAL);
2046 		if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
2047 			return (EINVAL);
2048 		if (info & VM_INTINFO_RSVD)
2049 			return (EINVAL);
2050 	} else {
2051 		info = 0;
2052 	}
2053 	VMM_CTR2(vcpu, "%s: info1(%#lx)", __func__, info);
2054 	vcpu->exitintinfo = info;
2055 	return (0);
2056 }
2057 
2058 enum exc_class {
2059 	EXC_BENIGN,
2060 	EXC_CONTRIBUTORY,
2061 	EXC_PAGEFAULT
2062 };
2063 
2064 #define	IDT_VE	20	/* Virtualization Exception (Intel specific) */
2065 
2066 static enum exc_class
2067 exception_class(uint64_t info)
2068 {
2069 	int type, vector;
2070 
2071 	KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
2072 	type = info & VM_INTINFO_TYPE;
2073 	vector = info & 0xff;
2074 
2075 	/* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
2076 	switch (type) {
2077 	case VM_INTINFO_HWINTR:
2078 	case VM_INTINFO_SWINTR:
2079 	case VM_INTINFO_NMI:
2080 		return (EXC_BENIGN);
2081 	default:
2082 		/*
2083 		 * Hardware exception.
2084 		 *
2085 		 * SVM and VT-x use identical type values to represent NMI,
2086 		 * hardware interrupt and software interrupt.
2087 		 *
2088 		 * SVM uses type '3' for all exceptions. VT-x uses type '3'
2089 		 * for exceptions except #BP and #OF. #BP and #OF use a type
2090 		 * value of '5' or '6'. Therefore we don't check for explicit
2091 		 * values of 'type' to classify 'intinfo' into a hardware
2092 		 * exception.
2093 		 */
2094 		break;
2095 	}
2096 
2097 	switch (vector) {
2098 	case IDT_PF:
2099 	case IDT_VE:
2100 		return (EXC_PAGEFAULT);
2101 	case IDT_DE:
2102 	case IDT_TS:
2103 	case IDT_NP:
2104 	case IDT_SS:
2105 	case IDT_GP:
2106 		return (EXC_CONTRIBUTORY);
2107 	default:
2108 		return (EXC_BENIGN);
2109 	}
2110 }
2111 
2112 static int
2113 nested_fault(struct vcpu *vcpu, uint64_t info1, uint64_t info2,
2114     uint64_t *retinfo)
2115 {
2116 	enum exc_class exc1, exc2;
2117 	int type1, vector1;
2118 
2119 	KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
2120 	KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
2121 
2122 	/*
2123 	 * If an exception occurs while attempting to call the double-fault
2124 	 * handler the processor enters shutdown mode (aka triple fault).
2125 	 */
2126 	type1 = info1 & VM_INTINFO_TYPE;
2127 	vector1 = info1 & 0xff;
2128 	if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
2129 		VMM_CTR2(vcpu, "triple fault: info1(%#lx), info2(%#lx)",
2130 		    info1, info2);
2131 		vm_suspend(vcpu->vm, VM_SUSPEND_TRIPLEFAULT);
2132 		*retinfo = 0;
2133 		return (0);
2134 	}
2135 
2136 	/*
2137 	 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
2138 	 */
2139 	exc1 = exception_class(info1);
2140 	exc2 = exception_class(info2);
2141 	if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
2142 	    (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
2143 		/* Convert nested fault into a double fault. */
2144 		*retinfo = IDT_DF;
2145 		*retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
2146 		*retinfo |= VM_INTINFO_DEL_ERRCODE;
2147 	} else {
2148 		/* Handle exceptions serially */
2149 		*retinfo = info2;
2150 	}
2151 	return (1);
2152 }
2153 
2154 static uint64_t
2155 vcpu_exception_intinfo(struct vcpu *vcpu)
2156 {
2157 	uint64_t info = 0;
2158 
2159 	if (vcpu->exception_pending) {
2160 		info = vcpu->exc_vector & 0xff;
2161 		info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
2162 		if (vcpu->exc_errcode_valid) {
2163 			info |= VM_INTINFO_DEL_ERRCODE;
2164 			info |= (uint64_t)vcpu->exc_errcode << 32;
2165 		}
2166 	}
2167 	return (info);
2168 }
2169 
2170 int
2171 vm_entry_intinfo(struct vcpu *vcpu, uint64_t *retinfo)
2172 {
2173 	uint64_t info1, info2;
2174 	int valid;
2175 
2176 	info1 = vcpu->exitintinfo;
2177 	vcpu->exitintinfo = 0;
2178 
2179 	info2 = 0;
2180 	if (vcpu->exception_pending) {
2181 		info2 = vcpu_exception_intinfo(vcpu);
2182 		vcpu->exception_pending = 0;
2183 		VMM_CTR2(vcpu, "Exception %d delivered: %#lx",
2184 		    vcpu->exc_vector, info2);
2185 	}
2186 
2187 	if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
2188 		valid = nested_fault(vcpu, info1, info2, retinfo);
2189 	} else if (info1 & VM_INTINFO_VALID) {
2190 		*retinfo = info1;
2191 		valid = 1;
2192 	} else if (info2 & VM_INTINFO_VALID) {
2193 		*retinfo = info2;
2194 		valid = 1;
2195 	} else {
2196 		valid = 0;
2197 	}
2198 
2199 	if (valid) {
2200 		VMM_CTR4(vcpu, "%s: info1(%#lx), info2(%#lx), "
2201 		    "retinfo(%#lx)", __func__, info1, info2, *retinfo);
2202 	}
2203 
2204 	return (valid);
2205 }
2206 
2207 int
2208 vm_get_intinfo(struct vcpu *vcpu, uint64_t *info1, uint64_t *info2)
2209 {
2210 	*info1 = vcpu->exitintinfo;
2211 	*info2 = vcpu_exception_intinfo(vcpu);
2212 	return (0);
2213 }
2214 
2215 int
2216 vm_inject_exception(struct vcpu *vcpu, int vector, int errcode_valid,
2217     uint32_t errcode, int restart_instruction)
2218 {
2219 	uint64_t regval;
2220 	int error __diagused;
2221 
2222 	if (vector < 0 || vector >= 32)
2223 		return (EINVAL);
2224 
2225 	/*
2226 	 * A double fault exception should never be injected directly into
2227 	 * the guest. It is a derived exception that results from specific
2228 	 * combinations of nested faults.
2229 	 */
2230 	if (vector == IDT_DF)
2231 		return (EINVAL);
2232 
2233 	if (vcpu->exception_pending) {
2234 		VMM_CTR2(vcpu, "Unable to inject exception %d due to "
2235 		    "pending exception %d", vector, vcpu->exc_vector);
2236 		return (EBUSY);
2237 	}
2238 
2239 	if (errcode_valid) {
2240 		/*
2241 		 * Exceptions don't deliver an error code in real mode.
2242 		 */
2243 		error = vm_get_register(vcpu, VM_REG_GUEST_CR0, &regval);
2244 		KASSERT(!error, ("%s: error %d getting CR0", __func__, error));
2245 		if (!(regval & CR0_PE))
2246 			errcode_valid = 0;
2247 	}
2248 
2249 	/*
2250 	 * From section 26.6.1 "Interruptibility State" in Intel SDM:
2251 	 *
2252 	 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
2253 	 * one instruction or incurs an exception.
2254 	 */
2255 	error = vm_set_register(vcpu, VM_REG_GUEST_INTR_SHADOW, 0);
2256 	KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
2257 	    __func__, error));
2258 
2259 	if (restart_instruction)
2260 		vm_restart_instruction(vcpu);
2261 
2262 	vcpu->exception_pending = 1;
2263 	vcpu->exc_vector = vector;
2264 	vcpu->exc_errcode = errcode;
2265 	vcpu->exc_errcode_valid = errcode_valid;
2266 	VMM_CTR1(vcpu, "Exception %d pending", vector);
2267 	return (0);
2268 }
2269 
2270 void
2271 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid, int errcode)
2272 {
2273 	int error __diagused, restart_instruction;
2274 
2275 	restart_instruction = 1;
2276 
2277 	error = vm_inject_exception(vcpu, vector, errcode_valid,
2278 	    errcode, restart_instruction);
2279 	KASSERT(error == 0, ("vm_inject_exception error %d", error));
2280 }
2281 
2282 void
2283 vm_inject_pf(struct vcpu *vcpu, int error_code, uint64_t cr2)
2284 {
2285 	int error __diagused;
2286 
2287 	VMM_CTR2(vcpu, "Injecting page fault: error_code %#x, cr2 %#lx",
2288 	    error_code, cr2);
2289 
2290 	error = vm_set_register(vcpu, VM_REG_GUEST_CR2, cr2);
2291 	KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
2292 
2293 	vm_inject_fault(vcpu, IDT_PF, 1, error_code);
2294 }
2295 
2296 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
2297 
2298 int
2299 vm_inject_nmi(struct vcpu *vcpu)
2300 {
2301 
2302 	vcpu->nmi_pending = 1;
2303 	vcpu_notify_event(vcpu, false);
2304 	return (0);
2305 }
2306 
2307 int
2308 vm_nmi_pending(struct vcpu *vcpu)
2309 {
2310 	return (vcpu->nmi_pending);
2311 }
2312 
2313 void
2314 vm_nmi_clear(struct vcpu *vcpu)
2315 {
2316 	if (vcpu->nmi_pending == 0)
2317 		panic("vm_nmi_clear: inconsistent nmi_pending state");
2318 
2319 	vcpu->nmi_pending = 0;
2320 	vmm_stat_incr(vcpu, VCPU_NMI_COUNT, 1);
2321 }
2322 
2323 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
2324 
2325 int
2326 vm_inject_extint(struct vcpu *vcpu)
2327 {
2328 
2329 	vcpu->extint_pending = 1;
2330 	vcpu_notify_event(vcpu, false);
2331 	return (0);
2332 }
2333 
2334 int
2335 vm_extint_pending(struct vcpu *vcpu)
2336 {
2337 	return (vcpu->extint_pending);
2338 }
2339 
2340 void
2341 vm_extint_clear(struct vcpu *vcpu)
2342 {
2343 	if (vcpu->extint_pending == 0)
2344 		panic("vm_extint_clear: inconsistent extint_pending state");
2345 
2346 	vcpu->extint_pending = 0;
2347 	vmm_stat_incr(vcpu, VCPU_EXTINT_COUNT, 1);
2348 }
2349 
2350 int
2351 vm_get_capability(struct vcpu *vcpu, int type, int *retval)
2352 {
2353 	if (type < 0 || type >= VM_CAP_MAX)
2354 		return (EINVAL);
2355 
2356 	return (vmmops_getcap(vcpu->cookie, type, retval));
2357 }
2358 
2359 int
2360 vm_set_capability(struct vcpu *vcpu, int type, int val)
2361 {
2362 	if (type < 0 || type >= VM_CAP_MAX)
2363 		return (EINVAL);
2364 
2365 	return (vmmops_setcap(vcpu->cookie, type, val));
2366 }
2367 
2368 struct vm *
2369 vcpu_vm(struct vcpu *vcpu)
2370 {
2371 	return (vcpu->vm);
2372 }
2373 
2374 int
2375 vcpu_vcpuid(struct vcpu *vcpu)
2376 {
2377 	return (vcpu->vcpuid);
2378 }
2379 
2380 struct vcpu *
2381 vm_vcpu(struct vm *vm, int vcpuid)
2382 {
2383 	return (vm->vcpu[vcpuid]);
2384 }
2385 
2386 struct vlapic *
2387 vm_lapic(struct vcpu *vcpu)
2388 {
2389 	return (vcpu->vlapic);
2390 }
2391 
2392 struct vioapic *
2393 vm_ioapic(struct vm *vm)
2394 {
2395 
2396 	return (vm->vioapic);
2397 }
2398 
2399 struct vhpet *
2400 vm_hpet(struct vm *vm)
2401 {
2402 
2403 	return (vm->vhpet);
2404 }
2405 
2406 bool
2407 vmm_is_pptdev(int bus, int slot, int func)
2408 {
2409 	int b, f, i, n, s;
2410 	char *val, *cp, *cp2;
2411 	bool found;
2412 
2413 	/*
2414 	 * XXX
2415 	 * The length of an environment variable is limited to 128 bytes which
2416 	 * puts an upper limit on the number of passthru devices that may be
2417 	 * specified using a single environment variable.
2418 	 *
2419 	 * Work around this by scanning multiple environment variable
2420 	 * names instead of a single one - yuck!
2421 	 */
2422 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2423 
2424 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2425 	found = false;
2426 	for (i = 0; names[i] != NULL && !found; i++) {
2427 		cp = val = kern_getenv(names[i]);
2428 		while (cp != NULL && *cp != '\0') {
2429 			if ((cp2 = strchr(cp, ' ')) != NULL)
2430 				*cp2 = '\0';
2431 
2432 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2433 			if (n == 3 && bus == b && slot == s && func == f) {
2434 				found = true;
2435 				break;
2436 			}
2437 
2438 			if (cp2 != NULL)
2439 				*cp2++ = ' ';
2440 
2441 			cp = cp2;
2442 		}
2443 		freeenv(val);
2444 	}
2445 	return (found);
2446 }
2447 
2448 void *
2449 vm_iommu_domain(struct vm *vm)
2450 {
2451 
2452 	return (vm->iommu);
2453 }
2454 
2455 int
2456 vcpu_set_state(struct vcpu *vcpu, enum vcpu_state newstate, bool from_idle)
2457 {
2458 	int error;
2459 
2460 	vcpu_lock(vcpu);
2461 	error = vcpu_set_state_locked(vcpu, newstate, from_idle);
2462 	vcpu_unlock(vcpu);
2463 
2464 	return (error);
2465 }
2466 
2467 enum vcpu_state
2468 vcpu_get_state(struct vcpu *vcpu, int *hostcpu)
2469 {
2470 	enum vcpu_state state;
2471 
2472 	vcpu_lock(vcpu);
2473 	state = vcpu->state;
2474 	if (hostcpu != NULL)
2475 		*hostcpu = vcpu->hostcpu;
2476 	vcpu_unlock(vcpu);
2477 
2478 	return (state);
2479 }
2480 
2481 int
2482 vm_activate_cpu(struct vcpu *vcpu)
2483 {
2484 	struct vm *vm = vcpu->vm;
2485 
2486 	if (CPU_ISSET(vcpu->vcpuid, &vm->active_cpus))
2487 		return (EBUSY);
2488 
2489 	VMM_CTR0(vcpu, "activated");
2490 	CPU_SET_ATOMIC(vcpu->vcpuid, &vm->active_cpus);
2491 	return (0);
2492 }
2493 
2494 int
2495 vm_suspend_cpu(struct vm *vm, struct vcpu *vcpu)
2496 {
2497 	if (vcpu == NULL) {
2498 		vm->debug_cpus = vm->active_cpus;
2499 		for (int i = 0; i < vm->maxcpus; i++) {
2500 			if (CPU_ISSET(i, &vm->active_cpus))
2501 				vcpu_notify_event(vm_vcpu(vm, i), false);
2502 		}
2503 	} else {
2504 		if (!CPU_ISSET(vcpu->vcpuid, &vm->active_cpus))
2505 			return (EINVAL);
2506 
2507 		CPU_SET_ATOMIC(vcpu->vcpuid, &vm->debug_cpus);
2508 		vcpu_notify_event(vcpu, false);
2509 	}
2510 	return (0);
2511 }
2512 
2513 int
2514 vm_resume_cpu(struct vm *vm, struct vcpu *vcpu)
2515 {
2516 
2517 	if (vcpu == NULL) {
2518 		CPU_ZERO(&vm->debug_cpus);
2519 	} else {
2520 		if (!CPU_ISSET(vcpu->vcpuid, &vm->debug_cpus))
2521 			return (EINVAL);
2522 
2523 		CPU_CLR_ATOMIC(vcpu->vcpuid, &vm->debug_cpus);
2524 	}
2525 	return (0);
2526 }
2527 
2528 int
2529 vcpu_debugged(struct vcpu *vcpu)
2530 {
2531 
2532 	return (CPU_ISSET(vcpu->vcpuid, &vcpu->vm->debug_cpus));
2533 }
2534 
2535 cpuset_t
2536 vm_active_cpus(struct vm *vm)
2537 {
2538 
2539 	return (vm->active_cpus);
2540 }
2541 
2542 cpuset_t
2543 vm_debug_cpus(struct vm *vm)
2544 {
2545 
2546 	return (vm->debug_cpus);
2547 }
2548 
2549 cpuset_t
2550 vm_suspended_cpus(struct vm *vm)
2551 {
2552 
2553 	return (vm->suspended_cpus);
2554 }
2555 
2556 /*
2557  * Returns the subset of vCPUs in tostart that are awaiting startup.
2558  * These vCPUs are also marked as no longer awaiting startup.
2559  */
2560 cpuset_t
2561 vm_start_cpus(struct vm *vm, const cpuset_t *tostart)
2562 {
2563 	cpuset_t set;
2564 
2565 	mtx_lock(&vm->rendezvous_mtx);
2566 	CPU_AND(&set, &vm->startup_cpus, tostart);
2567 	CPU_ANDNOT(&vm->startup_cpus, &vm->startup_cpus, &set);
2568 	mtx_unlock(&vm->rendezvous_mtx);
2569 	return (set);
2570 }
2571 
2572 void
2573 vm_await_start(struct vm *vm, const cpuset_t *waiting)
2574 {
2575 	mtx_lock(&vm->rendezvous_mtx);
2576 	CPU_OR(&vm->startup_cpus, &vm->startup_cpus, waiting);
2577 	mtx_unlock(&vm->rendezvous_mtx);
2578 }
2579 
2580 void *
2581 vcpu_stats(struct vcpu *vcpu)
2582 {
2583 
2584 	return (vcpu->stats);
2585 }
2586 
2587 int
2588 vm_get_x2apic_state(struct vcpu *vcpu, enum x2apic_state *state)
2589 {
2590 	*state = vcpu->x2apic_state;
2591 
2592 	return (0);
2593 }
2594 
2595 int
2596 vm_set_x2apic_state(struct vcpu *vcpu, enum x2apic_state state)
2597 {
2598 	if (state >= X2APIC_STATE_LAST)
2599 		return (EINVAL);
2600 
2601 	vcpu->x2apic_state = state;
2602 
2603 	vlapic_set_x2apic_state(vcpu, state);
2604 
2605 	return (0);
2606 }
2607 
2608 /*
2609  * This function is called to ensure that a vcpu "sees" a pending event
2610  * as soon as possible:
2611  * - If the vcpu thread is sleeping then it is woken up.
2612  * - If the vcpu is running on a different host_cpu then an IPI will be directed
2613  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
2614  */
2615 static void
2616 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
2617 {
2618 	int hostcpu;
2619 
2620 	hostcpu = vcpu->hostcpu;
2621 	if (vcpu->state == VCPU_RUNNING) {
2622 		KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2623 		if (hostcpu != curcpu) {
2624 			if (lapic_intr) {
2625 				vlapic_post_intr(vcpu->vlapic, hostcpu,
2626 				    vmm_ipinum);
2627 			} else {
2628 				ipi_cpu(hostcpu, vmm_ipinum);
2629 			}
2630 		} else {
2631 			/*
2632 			 * If the 'vcpu' is running on 'curcpu' then it must
2633 			 * be sending a notification to itself (e.g. SELF_IPI).
2634 			 * The pending event will be picked up when the vcpu
2635 			 * transitions back to guest context.
2636 			 */
2637 		}
2638 	} else {
2639 		KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2640 		    "with hostcpu %d", vcpu->state, hostcpu));
2641 		if (vcpu->state == VCPU_SLEEPING)
2642 			wakeup_one(vcpu);
2643 	}
2644 }
2645 
2646 void
2647 vcpu_notify_event(struct vcpu *vcpu, bool lapic_intr)
2648 {
2649 	vcpu_lock(vcpu);
2650 	vcpu_notify_event_locked(vcpu, lapic_intr);
2651 	vcpu_unlock(vcpu);
2652 }
2653 
2654 struct vmspace *
2655 vm_get_vmspace(struct vm *vm)
2656 {
2657 
2658 	return (vm->vmspace);
2659 }
2660 
2661 int
2662 vm_apicid2vcpuid(struct vm *vm, int apicid)
2663 {
2664 	/*
2665 	 * XXX apic id is assumed to be numerically identical to vcpu id
2666 	 */
2667 	return (apicid);
2668 }
2669 
2670 int
2671 vm_smp_rendezvous(struct vcpu *vcpu, cpuset_t dest,
2672     vm_rendezvous_func_t func, void *arg)
2673 {
2674 	struct vm *vm = vcpu->vm;
2675 	int error, i;
2676 
2677 	/*
2678 	 * Enforce that this function is called without any locks
2679 	 */
2680 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2681 
2682 restart:
2683 	mtx_lock(&vm->rendezvous_mtx);
2684 	if (vm->rendezvous_func != NULL) {
2685 		/*
2686 		 * If a rendezvous is already in progress then we need to
2687 		 * call the rendezvous handler in case this 'vcpu' is one
2688 		 * of the targets of the rendezvous.
2689 		 */
2690 		VMM_CTR0(vcpu, "Rendezvous already in progress");
2691 		mtx_unlock(&vm->rendezvous_mtx);
2692 		error = vm_handle_rendezvous(vcpu);
2693 		if (error != 0)
2694 			return (error);
2695 		goto restart;
2696 	}
2697 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2698 	    "rendezvous is still in progress"));
2699 
2700 	VMM_CTR0(vcpu, "Initiating rendezvous");
2701 	vm->rendezvous_req_cpus = dest;
2702 	CPU_ZERO(&vm->rendezvous_done_cpus);
2703 	vm->rendezvous_arg = arg;
2704 	vm->rendezvous_func = func;
2705 	mtx_unlock(&vm->rendezvous_mtx);
2706 
2707 	/*
2708 	 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2709 	 * vcpus so they handle the rendezvous as soon as possible.
2710 	 */
2711 	for (i = 0; i < vm->maxcpus; i++) {
2712 		if (CPU_ISSET(i, &dest))
2713 			vcpu_notify_event(vm_vcpu(vm, i), false);
2714 	}
2715 
2716 	return (vm_handle_rendezvous(vcpu));
2717 }
2718 
2719 struct vatpic *
2720 vm_atpic(struct vm *vm)
2721 {
2722 	return (vm->vatpic);
2723 }
2724 
2725 struct vatpit *
2726 vm_atpit(struct vm *vm)
2727 {
2728 	return (vm->vatpit);
2729 }
2730 
2731 struct vpmtmr *
2732 vm_pmtmr(struct vm *vm)
2733 {
2734 
2735 	return (vm->vpmtmr);
2736 }
2737 
2738 struct vrtc *
2739 vm_rtc(struct vm *vm)
2740 {
2741 
2742 	return (vm->vrtc);
2743 }
2744 
2745 enum vm_reg_name
2746 vm_segment_name(int seg)
2747 {
2748 	static enum vm_reg_name seg_names[] = {
2749 		VM_REG_GUEST_ES,
2750 		VM_REG_GUEST_CS,
2751 		VM_REG_GUEST_SS,
2752 		VM_REG_GUEST_DS,
2753 		VM_REG_GUEST_FS,
2754 		VM_REG_GUEST_GS
2755 	};
2756 
2757 	KASSERT(seg >= 0 && seg < nitems(seg_names),
2758 	    ("%s: invalid segment encoding %d", __func__, seg));
2759 	return (seg_names[seg]);
2760 }
2761 
2762 void
2763 vm_copy_teardown(struct vm_copyinfo *copyinfo, int num_copyinfo)
2764 {
2765 	int idx;
2766 
2767 	for (idx = 0; idx < num_copyinfo; idx++) {
2768 		if (copyinfo[idx].cookie != NULL)
2769 			vm_gpa_release(copyinfo[idx].cookie);
2770 	}
2771 	bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2772 }
2773 
2774 int
2775 vm_copy_setup(struct vcpu *vcpu, struct vm_guest_paging *paging,
2776     uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2777     int num_copyinfo, int *fault)
2778 {
2779 	int error, idx, nused;
2780 	size_t n, off, remaining;
2781 	void *hva, *cookie;
2782 	uint64_t gpa;
2783 
2784 	bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2785 
2786 	nused = 0;
2787 	remaining = len;
2788 	while (remaining > 0) {
2789 		KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2790 		error = vm_gla2gpa(vcpu, paging, gla, prot, &gpa, fault);
2791 		if (error || *fault)
2792 			return (error);
2793 		off = gpa & PAGE_MASK;
2794 		n = min(remaining, PAGE_SIZE - off);
2795 		copyinfo[nused].gpa = gpa;
2796 		copyinfo[nused].len = n;
2797 		remaining -= n;
2798 		gla += n;
2799 		nused++;
2800 	}
2801 
2802 	for (idx = 0; idx < nused; idx++) {
2803 		hva = vm_gpa_hold(vcpu, copyinfo[idx].gpa,
2804 		    copyinfo[idx].len, prot, &cookie);
2805 		if (hva == NULL)
2806 			break;
2807 		copyinfo[idx].hva = hva;
2808 		copyinfo[idx].cookie = cookie;
2809 	}
2810 
2811 	if (idx != nused) {
2812 		vm_copy_teardown(copyinfo, num_copyinfo);
2813 		return (EFAULT);
2814 	} else {
2815 		*fault = 0;
2816 		return (0);
2817 	}
2818 }
2819 
2820 void
2821 vm_copyin(struct vm_copyinfo *copyinfo, void *kaddr, size_t len)
2822 {
2823 	char *dst;
2824 	int idx;
2825 
2826 	dst = kaddr;
2827 	idx = 0;
2828 	while (len > 0) {
2829 		bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2830 		len -= copyinfo[idx].len;
2831 		dst += copyinfo[idx].len;
2832 		idx++;
2833 	}
2834 }
2835 
2836 void
2837 vm_copyout(const void *kaddr, struct vm_copyinfo *copyinfo, size_t len)
2838 {
2839 	const char *src;
2840 	int idx;
2841 
2842 	src = kaddr;
2843 	idx = 0;
2844 	while (len > 0) {
2845 		bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2846 		len -= copyinfo[idx].len;
2847 		src += copyinfo[idx].len;
2848 		idx++;
2849 	}
2850 }
2851 
2852 /*
2853  * Return the amount of in-use and wired memory for the VM. Since
2854  * these are global stats, only return the values with for vCPU 0
2855  */
2856 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2857 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2858 
2859 static void
2860 vm_get_rescnt(struct vcpu *vcpu, struct vmm_stat_type *stat)
2861 {
2862 
2863 	if (vcpu->vcpuid == 0) {
2864 		vmm_stat_set(vcpu, VMM_MEM_RESIDENT, PAGE_SIZE *
2865 		    vmspace_resident_count(vcpu->vm->vmspace));
2866 	}
2867 }
2868 
2869 static void
2870 vm_get_wiredcnt(struct vcpu *vcpu, struct vmm_stat_type *stat)
2871 {
2872 
2873 	if (vcpu->vcpuid == 0) {
2874 		vmm_stat_set(vcpu, VMM_MEM_WIRED, PAGE_SIZE *
2875 		    pmap_wired_count(vmspace_pmap(vcpu->vm->vmspace)));
2876 	}
2877 }
2878 
2879 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2880 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
2881 
2882 #ifdef BHYVE_SNAPSHOT
2883 static int
2884 vm_snapshot_vcpus(struct vm *vm, struct vm_snapshot_meta *meta)
2885 {
2886 	uint64_t tsc, now;
2887 	int ret;
2888 	struct vcpu *vcpu;
2889 	uint16_t i, maxcpus;
2890 
2891 	now = rdtsc();
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 		SNAPSHOT_VAR_OR_LEAVE(vcpu->x2apic_state, meta, ret, done);
2899 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitintinfo, meta, ret, done);
2900 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_vector, meta, ret, done);
2901 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode_valid, meta, ret, done);
2902 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exc_errcode, meta, ret, done);
2903 		SNAPSHOT_VAR_OR_LEAVE(vcpu->guest_xcr0, meta, ret, done);
2904 		SNAPSHOT_VAR_OR_LEAVE(vcpu->exitinfo, meta, ret, done);
2905 		SNAPSHOT_VAR_OR_LEAVE(vcpu->nextrip, meta, ret, done);
2906 
2907 		/*
2908 		 * Save the absolute TSC value by adding now to tsc_offset.
2909 		 *
2910 		 * It will be turned turned back into an actual offset when the
2911 		 * TSC restore function is called
2912 		 */
2913 		tsc = now + vcpu->tsc_offset;
2914 		SNAPSHOT_VAR_OR_LEAVE(tsc, meta, ret, done);
2915 		if (meta->op == VM_SNAPSHOT_RESTORE)
2916 			vcpu->tsc_offset = tsc;
2917 	}
2918 
2919 done:
2920 	return (ret);
2921 }
2922 
2923 static int
2924 vm_snapshot_vm(struct vm *vm, struct vm_snapshot_meta *meta)
2925 {
2926 	int ret;
2927 
2928 	ret = vm_snapshot_vcpus(vm, meta);
2929 	if (ret != 0)
2930 		goto done;
2931 
2932 	SNAPSHOT_VAR_OR_LEAVE(vm->startup_cpus, meta, ret, done);
2933 done:
2934 	return (ret);
2935 }
2936 
2937 static int
2938 vm_snapshot_vcpu(struct vm *vm, struct vm_snapshot_meta *meta)
2939 {
2940 	int error;
2941 	struct vcpu *vcpu;
2942 	uint16_t i, maxcpus;
2943 
2944 	error = 0;
2945 
2946 	maxcpus = vm_get_maxcpus(vm);
2947 	for (i = 0; i < maxcpus; i++) {
2948 		vcpu = vm->vcpu[i];
2949 		if (vcpu == NULL)
2950 			continue;
2951 
2952 		error = vmmops_vcpu_snapshot(vcpu->cookie, meta);
2953 		if (error != 0) {
2954 			printf("%s: failed to snapshot vmcs/vmcb data for "
2955 			       "vCPU: %d; error: %d\n", __func__, i, error);
2956 			goto done;
2957 		}
2958 	}
2959 
2960 done:
2961 	return (error);
2962 }
2963 
2964 /*
2965  * Save kernel-side structures to user-space for snapshotting.
2966  */
2967 int
2968 vm_snapshot_req(struct vm *vm, struct vm_snapshot_meta *meta)
2969 {
2970 	int ret = 0;
2971 
2972 	switch (meta->dev_req) {
2973 	case STRUCT_VMCX:
2974 		ret = vm_snapshot_vcpu(vm, meta);
2975 		break;
2976 	case STRUCT_VM:
2977 		ret = vm_snapshot_vm(vm, meta);
2978 		break;
2979 	case STRUCT_VIOAPIC:
2980 		ret = vioapic_snapshot(vm_ioapic(vm), meta);
2981 		break;
2982 	case STRUCT_VLAPIC:
2983 		ret = vlapic_snapshot(vm, meta);
2984 		break;
2985 	case STRUCT_VHPET:
2986 		ret = vhpet_snapshot(vm_hpet(vm), meta);
2987 		break;
2988 	case STRUCT_VATPIC:
2989 		ret = vatpic_snapshot(vm_atpic(vm), meta);
2990 		break;
2991 	case STRUCT_VATPIT:
2992 		ret = vatpit_snapshot(vm_atpit(vm), meta);
2993 		break;
2994 	case STRUCT_VPMTMR:
2995 		ret = vpmtmr_snapshot(vm_pmtmr(vm), meta);
2996 		break;
2997 	case STRUCT_VRTC:
2998 		ret = vrtc_snapshot(vm_rtc(vm), meta);
2999 		break;
3000 	default:
3001 		printf("%s: failed to find the requested type %#x\n",
3002 		       __func__, meta->dev_req);
3003 		ret = (EINVAL);
3004 	}
3005 	return (ret);
3006 }
3007 
3008 void
3009 vm_set_tsc_offset(struct vcpu *vcpu, uint64_t offset)
3010 {
3011 	vcpu->tsc_offset = offset;
3012 }
3013 
3014 int
3015 vm_restore_time(struct vm *vm)
3016 {
3017 	int error;
3018 	uint64_t now;
3019 	struct vcpu *vcpu;
3020 	uint16_t i, maxcpus;
3021 
3022 	now = rdtsc();
3023 
3024 	error = vhpet_restore_time(vm_hpet(vm));
3025 	if (error)
3026 		return (error);
3027 
3028 	maxcpus = vm_get_maxcpus(vm);
3029 	for (i = 0; i < maxcpus; i++) {
3030 		vcpu = vm->vcpu[i];
3031 		if (vcpu == NULL)
3032 			continue;
3033 
3034 		error = vmmops_restore_tsc(vcpu->cookie,
3035 		    vcpu->tsc_offset - now);
3036 		if (error)
3037 			return (error);
3038 	}
3039 
3040 	return (0);
3041 }
3042 #endif
3043