xref: /freebsd/sys/amd64/vmm/vmm.c (revision 5dd76dd0cc19450133aa379ce0ce4a68ae07fb39)
1 /*-
2  * Copyright (c) 2011 NetApp, Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
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/systm.h>
46 
47 #include <vm/vm.h>
48 #include <vm/vm_object.h>
49 #include <vm/vm_page.h>
50 #include <vm/pmap.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_extern.h>
53 #include <vm/vm_param.h>
54 
55 #include <machine/cpu.h>
56 #include <machine/vm.h>
57 #include <machine/pcb.h>
58 #include <machine/smp.h>
59 #include <x86/psl.h>
60 #include <x86/apicreg.h>
61 #include <machine/vmparam.h>
62 
63 #include <machine/vmm.h>
64 #include <machine/vmm_dev.h>
65 
66 #include "vmm_ktr.h"
67 #include "vmm_host.h"
68 #include "vmm_mem.h"
69 #include "vmm_util.h"
70 #include "vhpet.h"
71 #include "vioapic.h"
72 #include "vlapic.h"
73 #include "vmm_msr.h"
74 #include "vmm_ipi.h"
75 #include "vmm_stat.h"
76 #include "vmm_lapic.h"
77 
78 #include "io/ppt.h"
79 #include "io/iommu.h"
80 
81 struct vlapic;
82 
83 struct vcpu {
84 	int		flags;
85 	enum vcpu_state	state;
86 	struct mtx	mtx;
87 	int		hostcpu;	/* host cpuid this vcpu last ran on */
88 	uint64_t	guest_msrs[VMM_MSR_NUM];
89 	struct vlapic	*vlapic;
90 	int		 vcpuid;
91 	struct savefpu	*guestfpu;	/* guest fpu state */
92 	void		*stats;
93 	struct vm_exit	exitinfo;
94 	enum x2apic_state x2apic_state;
95 	int		nmi_pending;
96 };
97 
98 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
99 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
100 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
101 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
102 
103 struct mem_seg {
104 	vm_paddr_t	gpa;
105 	size_t		len;
106 	boolean_t	wired;
107 	vm_object_t	object;
108 };
109 #define	VM_MAX_MEMORY_SEGMENTS	2
110 
111 struct vm {
112 	void		*cookie;	/* processor-specific data */
113 	void		*iommu;		/* iommu-specific data */
114 	struct vhpet	*vhpet;		/* virtual HPET */
115 	struct vioapic	*vioapic;	/* virtual ioapic */
116 	struct vmspace	*vmspace;	/* guest's address space */
117 	struct vcpu	vcpu[VM_MAXCPU];
118 	int		num_mem_segs;
119 	struct mem_seg	mem_segs[VM_MAX_MEMORY_SEGMENTS];
120 	char		name[VM_MAX_NAMELEN];
121 
122 	/*
123 	 * Set of active vcpus.
124 	 * An active vcpu is one that has been started implicitly (BSP) or
125 	 * explicitly (AP) by sending it a startup ipi.
126 	 */
127 	cpuset_t	active_cpus;
128 
129 	struct mtx	rendezvous_mtx;
130 	cpuset_t	rendezvous_req_cpus;
131 	cpuset_t	rendezvous_done_cpus;
132 	void		*rendezvous_arg;
133 	vm_rendezvous_func_t rendezvous_func;
134 };
135 
136 static int vmm_initialized;
137 
138 static struct vmm_ops *ops;
139 #define	VMM_INIT(num)	(ops != NULL ? (*ops->init)(num) : 0)
140 #define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
141 #define	VMM_RESUME()	(ops != NULL ? (*ops->resume)() : 0)
142 
143 #define	VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
144 #define	VMRUN(vmi, vcpu, rip, pmap, rptr) \
145 	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr) : ENXIO)
146 #define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
147 #define	VMSPACE_ALLOC(min, max) \
148 	(ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
149 #define	VMSPACE_FREE(vmspace) \
150 	(ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
151 #define	VMGETREG(vmi, vcpu, num, retval)		\
152 	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
153 #define	VMSETREG(vmi, vcpu, num, val)		\
154 	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
155 #define	VMGETDESC(vmi, vcpu, num, desc)		\
156 	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
157 #define	VMSETDESC(vmi, vcpu, num, desc)		\
158 	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
159 #define	VMINJECT(vmi, vcpu, type, vec, ec, ecv)	\
160 	(ops != NULL ? (*ops->vminject)(vmi, vcpu, type, vec, ec, ecv) : ENXIO)
161 #define	VMGETCAP(vmi, vcpu, num, retval)	\
162 	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
163 #define	VMSETCAP(vmi, vcpu, num, val)		\
164 	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
165 #define	VLAPIC_INIT(vmi, vcpu)			\
166 	(ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
167 #define	VLAPIC_CLEANUP(vmi, vlapic)		\
168 	(ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
169 
170 #define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
171 #define	fpu_stop_emulating()	clts()
172 
173 static MALLOC_DEFINE(M_VM, "vm", "vm");
174 CTASSERT(VMM_MSR_NUM <= 64);	/* msr_mask can keep track of up to 64 msrs */
175 
176 /* statistics */
177 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
178 
179 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
180 
181 static int vmm_ipinum;
182 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
183     "IPI vector used for vcpu notifications");
184 
185 static void vm_deactivate_cpu(struct vm *vm, int vcpuid);
186 
187 static void
188 vcpu_cleanup(struct vm *vm, int i)
189 {
190 	struct vcpu *vcpu = &vm->vcpu[i];
191 
192 	VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
193 	vmm_stat_free(vcpu->stats);
194 	fpu_save_area_free(vcpu->guestfpu);
195 }
196 
197 static void
198 vcpu_init(struct vm *vm, uint32_t vcpu_id)
199 {
200 	struct vcpu *vcpu;
201 
202 	vcpu = &vm->vcpu[vcpu_id];
203 
204 	vcpu_lock_init(vcpu);
205 	vcpu->hostcpu = NOCPU;
206 	vcpu->vcpuid = vcpu_id;
207 	vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
208 	vm_set_x2apic_state(vm, vcpu_id, X2APIC_ENABLED);
209 	vcpu->guestfpu = fpu_save_area_alloc();
210 	fpu_save_area_reset(vcpu->guestfpu);
211 	vcpu->stats = vmm_stat_alloc();
212 }
213 
214 struct vm_exit *
215 vm_exitinfo(struct vm *vm, int cpuid)
216 {
217 	struct vcpu *vcpu;
218 
219 	if (cpuid < 0 || cpuid >= VM_MAXCPU)
220 		panic("vm_exitinfo: invalid cpuid %d", cpuid);
221 
222 	vcpu = &vm->vcpu[cpuid];
223 
224 	return (&vcpu->exitinfo);
225 }
226 
227 static void
228 vmm_resume(void)
229 {
230 	VMM_RESUME();
231 }
232 
233 static int
234 vmm_init(void)
235 {
236 	int error;
237 
238 	vmm_host_state_init();
239 
240 	vmm_ipinum = vmm_ipi_alloc();
241 	if (vmm_ipinum == 0)
242 		vmm_ipinum = IPI_AST;
243 
244 	error = vmm_mem_init();
245 	if (error)
246 		return (error);
247 
248 	if (vmm_is_intel())
249 		ops = &vmm_ops_intel;
250 	else if (vmm_is_amd())
251 		ops = &vmm_ops_amd;
252 	else
253 		return (ENXIO);
254 
255 	vmm_msr_init();
256 	vmm_resume_p = vmm_resume;
257 
258 	return (VMM_INIT(vmm_ipinum));
259 }
260 
261 static int
262 vmm_handler(module_t mod, int what, void *arg)
263 {
264 	int error;
265 
266 	switch (what) {
267 	case MOD_LOAD:
268 		vmmdev_init();
269 		if (ppt_avail_devices() > 0)
270 			iommu_init();
271 		error = vmm_init();
272 		if (error == 0)
273 			vmm_initialized = 1;
274 		break;
275 	case MOD_UNLOAD:
276 		error = vmmdev_cleanup();
277 		if (error == 0) {
278 			vmm_resume_p = NULL;
279 			iommu_cleanup();
280 			if (vmm_ipinum != IPI_AST)
281 				vmm_ipi_free(vmm_ipinum);
282 			error = VMM_CLEANUP();
283 			/*
284 			 * Something bad happened - prevent new
285 			 * VMs from being created
286 			 */
287 			if (error)
288 				vmm_initialized = 0;
289 		}
290 		break;
291 	default:
292 		error = 0;
293 		break;
294 	}
295 	return (error);
296 }
297 
298 static moduledata_t vmm_kmod = {
299 	"vmm",
300 	vmm_handler,
301 	NULL
302 };
303 
304 /*
305  * vmm initialization has the following dependencies:
306  *
307  * - iommu initialization must happen after the pci passthru driver has had
308  *   a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
309  *
310  * - VT-x initialization requires smp_rendezvous() and therefore must happen
311  *   after SMP is fully functional (after SI_SUB_SMP).
312  */
313 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
314 MODULE_VERSION(vmm, 1);
315 
316 int
317 vm_create(const char *name, struct vm **retvm)
318 {
319 	int i;
320 	struct vm *vm;
321 	struct vmspace *vmspace;
322 
323 	const int BSP = 0;
324 
325 	/*
326 	 * If vmm.ko could not be successfully initialized then don't attempt
327 	 * to create the virtual machine.
328 	 */
329 	if (!vmm_initialized)
330 		return (ENXIO);
331 
332 	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
333 		return (EINVAL);
334 
335 	vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
336 	if (vmspace == NULL)
337 		return (ENOMEM);
338 
339 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
340 	strcpy(vm->name, name);
341 	vm->vmspace = vmspace;
342 	mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
343 	vm->cookie = VMINIT(vm, vmspace_pmap(vmspace));
344 	vm->vioapic = vioapic_init(vm);
345 	vm->vhpet = vhpet_init(vm);
346 
347 	for (i = 0; i < VM_MAXCPU; i++) {
348 		vcpu_init(vm, i);
349 		guest_msrs_init(vm, i);
350 	}
351 
352 	vm_activate_cpu(vm, BSP);
353 
354 	*retvm = vm;
355 	return (0);
356 }
357 
358 static void
359 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg)
360 {
361 
362 	if (seg->object != NULL)
363 		vmm_mem_free(vm->vmspace, seg->gpa, seg->len);
364 
365 	bzero(seg, sizeof(*seg));
366 }
367 
368 void
369 vm_destroy(struct vm *vm)
370 {
371 	int i;
372 
373 	ppt_unassign_all(vm);
374 
375 	if (vm->iommu != NULL)
376 		iommu_destroy_domain(vm->iommu);
377 
378 	vhpet_cleanup(vm->vhpet);
379 	vioapic_cleanup(vm->vioapic);
380 
381 	for (i = 0; i < vm->num_mem_segs; i++)
382 		vm_free_mem_seg(vm, &vm->mem_segs[i]);
383 
384 	vm->num_mem_segs = 0;
385 
386 	for (i = 0; i < VM_MAXCPU; i++)
387 		vcpu_cleanup(vm, i);
388 
389 	VMSPACE_FREE(vm->vmspace);
390 
391 	VMCLEANUP(vm->cookie);
392 
393 	free(vm, M_VM);
394 }
395 
396 const char *
397 vm_name(struct vm *vm)
398 {
399 	return (vm->name);
400 }
401 
402 int
403 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
404 {
405 	vm_object_t obj;
406 
407 	if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
408 		return (ENOMEM);
409 	else
410 		return (0);
411 }
412 
413 int
414 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
415 {
416 
417 	vmm_mmio_free(vm->vmspace, gpa, len);
418 	return (0);
419 }
420 
421 boolean_t
422 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa)
423 {
424 	int i;
425 	vm_paddr_t gpabase, gpalimit;
426 
427 	for (i = 0; i < vm->num_mem_segs; i++) {
428 		gpabase = vm->mem_segs[i].gpa;
429 		gpalimit = gpabase + vm->mem_segs[i].len;
430 		if (gpa >= gpabase && gpa < gpalimit)
431 			return (TRUE);		/* 'gpa' is regular memory */
432 	}
433 
434 	if (ppt_is_mmio(vm, gpa))
435 		return (TRUE);			/* 'gpa' is pci passthru mmio */
436 
437 	return (FALSE);
438 }
439 
440 int
441 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
442 {
443 	int available, allocated;
444 	struct mem_seg *seg;
445 	vm_object_t object;
446 	vm_paddr_t g;
447 
448 	if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
449 		return (EINVAL);
450 
451 	available = allocated = 0;
452 	g = gpa;
453 	while (g < gpa + len) {
454 		if (vm_mem_allocated(vm, g))
455 			allocated++;
456 		else
457 			available++;
458 
459 		g += PAGE_SIZE;
460 	}
461 
462 	/*
463 	 * If there are some allocated and some available pages in the address
464 	 * range then it is an error.
465 	 */
466 	if (allocated && available)
467 		return (EINVAL);
468 
469 	/*
470 	 * If the entire address range being requested has already been
471 	 * allocated then there isn't anything more to do.
472 	 */
473 	if (allocated && available == 0)
474 		return (0);
475 
476 	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
477 		return (E2BIG);
478 
479 	seg = &vm->mem_segs[vm->num_mem_segs];
480 
481 	if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL)
482 		return (ENOMEM);
483 
484 	seg->gpa = gpa;
485 	seg->len = len;
486 	seg->object = object;
487 	seg->wired = FALSE;
488 
489 	vm->num_mem_segs++;
490 
491 	return (0);
492 }
493 
494 static void
495 vm_gpa_unwire(struct vm *vm)
496 {
497 	int i, rv;
498 	struct mem_seg *seg;
499 
500 	for (i = 0; i < vm->num_mem_segs; i++) {
501 		seg = &vm->mem_segs[i];
502 		if (!seg->wired)
503 			continue;
504 
505 		rv = vm_map_unwire(&vm->vmspace->vm_map,
506 				   seg->gpa, seg->gpa + seg->len,
507 				   VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
508 		KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment "
509 		    "%#lx/%ld could not be unwired: %d",
510 		    vm_name(vm), seg->gpa, seg->len, rv));
511 
512 		seg->wired = FALSE;
513 	}
514 }
515 
516 static int
517 vm_gpa_wire(struct vm *vm)
518 {
519 	int i, rv;
520 	struct mem_seg *seg;
521 
522 	for (i = 0; i < vm->num_mem_segs; i++) {
523 		seg = &vm->mem_segs[i];
524 		if (seg->wired)
525 			continue;
526 
527 		/* XXX rlimits? */
528 		rv = vm_map_wire(&vm->vmspace->vm_map,
529 				 seg->gpa, seg->gpa + seg->len,
530 				 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
531 		if (rv != KERN_SUCCESS)
532 			break;
533 
534 		seg->wired = TRUE;
535 	}
536 
537 	if (i < vm->num_mem_segs) {
538 		/*
539 		 * Undo the wiring before returning an error.
540 		 */
541 		vm_gpa_unwire(vm);
542 		return (EAGAIN);
543 	}
544 
545 	return (0);
546 }
547 
548 static void
549 vm_iommu_modify(struct vm *vm, boolean_t map)
550 {
551 	int i, sz;
552 	vm_paddr_t gpa, hpa;
553 	struct mem_seg *seg;
554 	void *vp, *cookie, *host_domain;
555 
556 	sz = PAGE_SIZE;
557 	host_domain = iommu_host_domain();
558 
559 	for (i = 0; i < vm->num_mem_segs; i++) {
560 		seg = &vm->mem_segs[i];
561 		KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired",
562 		    vm_name(vm), seg->gpa, seg->len));
563 
564 		gpa = seg->gpa;
565 		while (gpa < seg->gpa + seg->len) {
566 			vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE,
567 					 &cookie);
568 			KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
569 			    vm_name(vm), gpa));
570 
571 			vm_gpa_release(cookie);
572 
573 			hpa = DMAP_TO_PHYS((uintptr_t)vp);
574 			if (map) {
575 				iommu_create_mapping(vm->iommu, gpa, hpa, sz);
576 				iommu_remove_mapping(host_domain, hpa, sz);
577 			} else {
578 				iommu_remove_mapping(vm->iommu, gpa, sz);
579 				iommu_create_mapping(host_domain, hpa, hpa, sz);
580 			}
581 
582 			gpa += PAGE_SIZE;
583 		}
584 	}
585 
586 	/*
587 	 * Invalidate the cached translations associated with the domain
588 	 * from which pages were removed.
589 	 */
590 	if (map)
591 		iommu_invalidate_tlb(host_domain);
592 	else
593 		iommu_invalidate_tlb(vm->iommu);
594 }
595 
596 #define	vm_iommu_unmap(vm)	vm_iommu_modify((vm), FALSE)
597 #define	vm_iommu_map(vm)	vm_iommu_modify((vm), TRUE)
598 
599 int
600 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
601 {
602 	int error;
603 
604 	error = ppt_unassign_device(vm, bus, slot, func);
605 	if (error)
606 		return (error);
607 
608 	if (ppt_assigned_devices(vm) == 0) {
609 		vm_iommu_unmap(vm);
610 		vm_gpa_unwire(vm);
611 	}
612 	return (0);
613 }
614 
615 int
616 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
617 {
618 	int error;
619 	vm_paddr_t maxaddr;
620 
621 	/*
622 	 * Virtual machines with pci passthru devices get special treatment:
623 	 * - the guest physical memory is wired
624 	 * - the iommu is programmed to do the 'gpa' to 'hpa' translation
625 	 *
626 	 * We need to do this before the first pci passthru device is attached.
627 	 */
628 	if (ppt_assigned_devices(vm) == 0) {
629 		KASSERT(vm->iommu == NULL,
630 		    ("vm_assign_pptdev: iommu must be NULL"));
631 		maxaddr = vmm_mem_maxaddr();
632 		vm->iommu = iommu_create_domain(maxaddr);
633 
634 		error = vm_gpa_wire(vm);
635 		if (error)
636 			return (error);
637 
638 		vm_iommu_map(vm);
639 	}
640 
641 	error = ppt_assign_device(vm, bus, slot, func);
642 	return (error);
643 }
644 
645 void *
646 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
647 	    void **cookie)
648 {
649 	int count, pageoff;
650 	vm_page_t m;
651 
652 	pageoff = gpa & PAGE_MASK;
653 	if (len > PAGE_SIZE - pageoff)
654 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
655 
656 	count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
657 	    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
658 
659 	if (count == 1) {
660 		*cookie = m;
661 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
662 	} else {
663 		*cookie = NULL;
664 		return (NULL);
665 	}
666 }
667 
668 void
669 vm_gpa_release(void *cookie)
670 {
671 	vm_page_t m = cookie;
672 
673 	vm_page_lock(m);
674 	vm_page_unhold(m);
675 	vm_page_unlock(m);
676 }
677 
678 int
679 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
680 		  struct vm_memory_segment *seg)
681 {
682 	int i;
683 
684 	for (i = 0; i < vm->num_mem_segs; i++) {
685 		if (gpabase == vm->mem_segs[i].gpa) {
686 			seg->gpa = vm->mem_segs[i].gpa;
687 			seg->len = vm->mem_segs[i].len;
688 			seg->wired = vm->mem_segs[i].wired;
689 			return (0);
690 		}
691 	}
692 	return (-1);
693 }
694 
695 int
696 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len,
697 	      vm_offset_t *offset, struct vm_object **object)
698 {
699 	int i;
700 	size_t seg_len;
701 	vm_paddr_t seg_gpa;
702 	vm_object_t seg_obj;
703 
704 	for (i = 0; i < vm->num_mem_segs; i++) {
705 		if ((seg_obj = vm->mem_segs[i].object) == NULL)
706 			continue;
707 
708 		seg_gpa = vm->mem_segs[i].gpa;
709 		seg_len = vm->mem_segs[i].len;
710 
711 		if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) {
712 			*offset = gpa - seg_gpa;
713 			*object = seg_obj;
714 			vm_object_reference(seg_obj);
715 			return (0);
716 		}
717 	}
718 
719 	return (EINVAL);
720 }
721 
722 int
723 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
724 {
725 
726 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
727 		return (EINVAL);
728 
729 	if (reg >= VM_REG_LAST)
730 		return (EINVAL);
731 
732 	return (VMGETREG(vm->cookie, vcpu, reg, retval));
733 }
734 
735 int
736 vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
737 {
738 
739 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
740 		return (EINVAL);
741 
742 	if (reg >= VM_REG_LAST)
743 		return (EINVAL);
744 
745 	return (VMSETREG(vm->cookie, vcpu, reg, val));
746 }
747 
748 static boolean_t
749 is_descriptor_table(int reg)
750 {
751 
752 	switch (reg) {
753 	case VM_REG_GUEST_IDTR:
754 	case VM_REG_GUEST_GDTR:
755 		return (TRUE);
756 	default:
757 		return (FALSE);
758 	}
759 }
760 
761 static boolean_t
762 is_segment_register(int reg)
763 {
764 
765 	switch (reg) {
766 	case VM_REG_GUEST_ES:
767 	case VM_REG_GUEST_CS:
768 	case VM_REG_GUEST_SS:
769 	case VM_REG_GUEST_DS:
770 	case VM_REG_GUEST_FS:
771 	case VM_REG_GUEST_GS:
772 	case VM_REG_GUEST_TR:
773 	case VM_REG_GUEST_LDTR:
774 		return (TRUE);
775 	default:
776 		return (FALSE);
777 	}
778 }
779 
780 int
781 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
782 		struct seg_desc *desc)
783 {
784 
785 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
786 		return (EINVAL);
787 
788 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
789 		return (EINVAL);
790 
791 	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
792 }
793 
794 int
795 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
796 		struct seg_desc *desc)
797 {
798 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
799 		return (EINVAL);
800 
801 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
802 		return (EINVAL);
803 
804 	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
805 }
806 
807 static void
808 restore_guest_fpustate(struct vcpu *vcpu)
809 {
810 
811 	/* flush host state to the pcb */
812 	fpuexit(curthread);
813 
814 	/* restore guest FPU state */
815 	fpu_stop_emulating();
816 	fpurestore(vcpu->guestfpu);
817 
818 	/*
819 	 * The FPU is now "dirty" with the guest's state so turn on emulation
820 	 * to trap any access to the FPU by the host.
821 	 */
822 	fpu_start_emulating();
823 }
824 
825 static void
826 save_guest_fpustate(struct vcpu *vcpu)
827 {
828 
829 	if ((rcr0() & CR0_TS) == 0)
830 		panic("fpu emulation not enabled in host!");
831 
832 	/* save guest FPU state */
833 	fpu_stop_emulating();
834 	fpusave(vcpu->guestfpu);
835 	fpu_start_emulating();
836 }
837 
838 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
839 
840 static int
841 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
842     bool from_idle)
843 {
844 	int error;
845 
846 	vcpu_assert_locked(vcpu);
847 
848 	/*
849 	 * State transitions from the vmmdev_ioctl() must always begin from
850 	 * the VCPU_IDLE state. This guarantees that there is only a single
851 	 * ioctl() operating on a vcpu at any point.
852 	 */
853 	if (from_idle) {
854 		while (vcpu->state != VCPU_IDLE)
855 			msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
856 	} else {
857 		KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
858 		    "vcpu idle state"));
859 	}
860 
861 	/*
862 	 * The following state transitions are allowed:
863 	 * IDLE -> FROZEN -> IDLE
864 	 * FROZEN -> RUNNING -> FROZEN
865 	 * FROZEN -> SLEEPING -> FROZEN
866 	 */
867 	switch (vcpu->state) {
868 	case VCPU_IDLE:
869 	case VCPU_RUNNING:
870 	case VCPU_SLEEPING:
871 		error = (newstate != VCPU_FROZEN);
872 		break;
873 	case VCPU_FROZEN:
874 		error = (newstate == VCPU_FROZEN);
875 		break;
876 	default:
877 		error = 1;
878 		break;
879 	}
880 
881 	if (error)
882 		return (EBUSY);
883 
884 	vcpu->state = newstate;
885 	if (newstate == VCPU_IDLE)
886 		wakeup(&vcpu->state);
887 
888 	return (0);
889 }
890 
891 static void
892 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
893 {
894 	int error;
895 
896 	if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
897 		panic("Error %d setting state to %d\n", error, newstate);
898 }
899 
900 static void
901 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
902 {
903 	int error;
904 
905 	if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
906 		panic("Error %d setting state to %d", error, newstate);
907 }
908 
909 static void
910 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func)
911 {
912 
913 	KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked"));
914 
915 	/*
916 	 * Update 'rendezvous_func' and execute a write memory barrier to
917 	 * ensure that it is visible across all host cpus. This is not needed
918 	 * for correctness but it does ensure that all the vcpus will notice
919 	 * that the rendezvous is requested immediately.
920 	 */
921 	vm->rendezvous_func = func;
922 	wmb();
923 }
924 
925 #define	RENDEZVOUS_CTR0(vm, vcpuid, fmt)				\
926 	do {								\
927 		if (vcpuid >= 0)					\
928 			VCPU_CTR0(vm, vcpuid, fmt);			\
929 		else							\
930 			VM_CTR0(vm, fmt);				\
931 	} while (0)
932 
933 static void
934 vm_handle_rendezvous(struct vm *vm, int vcpuid)
935 {
936 
937 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
938 	    ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
939 
940 	mtx_lock(&vm->rendezvous_mtx);
941 	while (vm->rendezvous_func != NULL) {
942 		if (vcpuid != -1 &&
943 		    CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus)) {
944 			VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
945 			(*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
946 			CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
947 		}
948 		if (CPU_CMP(&vm->rendezvous_req_cpus,
949 		    &vm->rendezvous_done_cpus) == 0) {
950 			VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
951 			vm_set_rendezvous_func(vm, NULL);
952 			wakeup(&vm->rendezvous_func);
953 			break;
954 		}
955 		RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
956 		mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
957 		    "vmrndv", 0);
958 	}
959 	mtx_unlock(&vm->rendezvous_mtx);
960 }
961 
962 /*
963  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
964  */
965 static int
966 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
967 {
968 	struct vm_exit *vmexit;
969 	struct vcpu *vcpu;
970 	int t, timo;
971 
972 	vcpu = &vm->vcpu[vcpuid];
973 
974 	vcpu_lock(vcpu);
975 
976 	/*
977 	 * Do a final check for pending NMI or interrupts before
978 	 * really putting this thread to sleep.
979 	 *
980 	 * These interrupts could have happened any time after we
981 	 * returned from VMRUN() and before we grabbed the vcpu lock.
982 	 */
983 	if (!vm_nmi_pending(vm, vcpuid) &&
984 	    (intr_disabled || !vlapic_pending_intr(vcpu->vlapic, NULL))) {
985 		t = ticks;
986 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
987 		if (vlapic_enabled(vcpu->vlapic)) {
988 			/*
989 			 * XXX msleep_spin() is not interruptible so use the
990 			 * 'timo' to put an upper bound on the sleep time.
991 			 */
992 			timo = hz;
993 			msleep_spin(vcpu, &vcpu->mtx, "vmidle", timo);
994 		} else {
995 			/*
996 			 * Spindown the vcpu if the apic is disabled and it
997 			 * had entered the halted state.
998 			 */
999 			*retu = true;
1000 			vmexit = vm_exitinfo(vm, vcpuid);
1001 			vmexit->exitcode = VM_EXITCODE_SPINDOWN_CPU;
1002 			vm_deactivate_cpu(vm, vcpuid);
1003 			VCPU_CTR0(vm, vcpuid, "spinning down cpu");
1004 		}
1005 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1006 		vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1007 	}
1008 	vcpu_unlock(vcpu);
1009 
1010 	return (0);
1011 }
1012 
1013 static int
1014 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1015 {
1016 	int rv, ftype;
1017 	struct vm_map *map;
1018 	struct vcpu *vcpu;
1019 	struct vm_exit *vme;
1020 
1021 	vcpu = &vm->vcpu[vcpuid];
1022 	vme = &vcpu->exitinfo;
1023 
1024 	ftype = vme->u.paging.fault_type;
1025 	KASSERT(ftype == VM_PROT_READ ||
1026 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1027 	    ("vm_handle_paging: invalid fault_type %d", ftype));
1028 
1029 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1030 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1031 		    vme->u.paging.gpa, ftype);
1032 		if (rv == 0)
1033 			goto done;
1034 	}
1035 
1036 	map = &vm->vmspace->vm_map;
1037 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
1038 
1039 	VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1040 	    "ftype = %d", rv, vme->u.paging.gpa, ftype);
1041 
1042 	if (rv != KERN_SUCCESS)
1043 		return (EFAULT);
1044 done:
1045 	/* restart execution at the faulting instruction */
1046 	vme->inst_length = 0;
1047 
1048 	return (0);
1049 }
1050 
1051 static int
1052 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1053 {
1054 	struct vie *vie;
1055 	struct vcpu *vcpu;
1056 	struct vm_exit *vme;
1057 	int error, inst_length;
1058 	uint64_t rip, gla, gpa, cr3;
1059 	mem_region_read_t mread;
1060 	mem_region_write_t mwrite;
1061 
1062 	vcpu = &vm->vcpu[vcpuid];
1063 	vme = &vcpu->exitinfo;
1064 
1065 	rip = vme->rip;
1066 	inst_length = vme->inst_length;
1067 
1068 	gla = vme->u.inst_emul.gla;
1069 	gpa = vme->u.inst_emul.gpa;
1070 	cr3 = vme->u.inst_emul.cr3;
1071 	vie = &vme->u.inst_emul.vie;
1072 
1073 	vie_init(vie);
1074 
1075 	/* Fetch, decode and emulate the faulting instruction */
1076 	if (vmm_fetch_instruction(vm, vcpuid, rip, inst_length, cr3, vie) != 0)
1077 		return (EFAULT);
1078 
1079 	if (vmm_decode_instruction(vm, vcpuid, gla, vie) != 0)
1080 		return (EFAULT);
1081 
1082 	/* return to userland unless this is an in-kernel emulated device */
1083 	if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1084 		mread = lapic_mmio_read;
1085 		mwrite = lapic_mmio_write;
1086 	} else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1087 		mread = vioapic_mmio_read;
1088 		mwrite = vioapic_mmio_write;
1089 	} else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1090 		mread = vhpet_mmio_read;
1091 		mwrite = vhpet_mmio_write;
1092 	} else {
1093 		*retu = true;
1094 		return (0);
1095 	}
1096 
1097 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, mread, mwrite,
1098 	    retu);
1099 
1100 	return (error);
1101 }
1102 
1103 int
1104 vm_run(struct vm *vm, struct vm_run *vmrun)
1105 {
1106 	int error, vcpuid;
1107 	struct vcpu *vcpu;
1108 	struct pcb *pcb;
1109 	uint64_t tscval, rip;
1110 	struct vm_exit *vme;
1111 	bool retu, intr_disabled;
1112 	pmap_t pmap;
1113 
1114 	vcpuid = vmrun->cpuid;
1115 
1116 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1117 		return (EINVAL);
1118 
1119 	pmap = vmspace_pmap(vm->vmspace);
1120 	vcpu = &vm->vcpu[vcpuid];
1121 	vme = &vcpu->exitinfo;
1122 	rip = vmrun->rip;
1123 restart:
1124 	critical_enter();
1125 
1126 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1127 	    ("vm_run: absurd pm_active"));
1128 
1129 	tscval = rdtsc();
1130 
1131 	pcb = PCPU_GET(curpcb);
1132 	set_pcb_flags(pcb, PCB_FULL_IRET);
1133 
1134 	restore_guest_msrs(vm, vcpuid);
1135 	restore_guest_fpustate(vcpu);
1136 
1137 	vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1138 	vcpu->hostcpu = curcpu;
1139 	error = VMRUN(vm->cookie, vcpuid, rip, pmap, &vm->rendezvous_func);
1140 	vcpu->hostcpu = NOCPU;
1141 	vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1142 
1143 	save_guest_fpustate(vcpu);
1144 	restore_host_msrs(vm, vcpuid);
1145 
1146 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1147 
1148 	critical_exit();
1149 
1150 	if (error == 0) {
1151 		retu = false;
1152 		switch (vme->exitcode) {
1153 		case VM_EXITCODE_IOAPIC_EOI:
1154 			vioapic_process_eoi(vm, vcpuid,
1155 			    vme->u.ioapic_eoi.vector);
1156 			break;
1157 		case VM_EXITCODE_RENDEZVOUS:
1158 			vm_handle_rendezvous(vm, vcpuid);
1159 			error = 0;
1160 			break;
1161 		case VM_EXITCODE_HLT:
1162 			intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1163 			error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1164 			break;
1165 		case VM_EXITCODE_PAGING:
1166 			error = vm_handle_paging(vm, vcpuid, &retu);
1167 			break;
1168 		case VM_EXITCODE_INST_EMUL:
1169 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1170 			break;
1171 		default:
1172 			retu = true;	/* handled in userland */
1173 			break;
1174 		}
1175 	}
1176 
1177 	if (error == 0 && retu == false) {
1178 		rip = vme->rip + vme->inst_length;
1179 		goto restart;
1180 	}
1181 
1182 	/* copy the exit information */
1183 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1184 	return (error);
1185 }
1186 
1187 int
1188 vm_inject_event(struct vm *vm, int vcpuid, int type,
1189 		int vector, uint32_t code, int code_valid)
1190 {
1191 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1192 		return (EINVAL);
1193 
1194 	if ((type > VM_EVENT_NONE && type < VM_EVENT_MAX) == 0)
1195 		return (EINVAL);
1196 
1197 	if (vector < 0 || vector > 255)
1198 		return (EINVAL);
1199 
1200 	return (VMINJECT(vm->cookie, vcpuid, type, vector, code, code_valid));
1201 }
1202 
1203 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
1204 
1205 int
1206 vm_inject_nmi(struct vm *vm, int vcpuid)
1207 {
1208 	struct vcpu *vcpu;
1209 
1210 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1211 		return (EINVAL);
1212 
1213 	vcpu = &vm->vcpu[vcpuid];
1214 
1215 	vcpu->nmi_pending = 1;
1216 	vcpu_notify_event(vm, vcpuid, false);
1217 	return (0);
1218 }
1219 
1220 int
1221 vm_nmi_pending(struct vm *vm, int vcpuid)
1222 {
1223 	struct vcpu *vcpu;
1224 
1225 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1226 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1227 
1228 	vcpu = &vm->vcpu[vcpuid];
1229 
1230 	return (vcpu->nmi_pending);
1231 }
1232 
1233 void
1234 vm_nmi_clear(struct vm *vm, int vcpuid)
1235 {
1236 	struct vcpu *vcpu;
1237 
1238 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1239 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1240 
1241 	vcpu = &vm->vcpu[vcpuid];
1242 
1243 	if (vcpu->nmi_pending == 0)
1244 		panic("vm_nmi_clear: inconsistent nmi_pending state");
1245 
1246 	vcpu->nmi_pending = 0;
1247 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
1248 }
1249 
1250 int
1251 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
1252 {
1253 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1254 		return (EINVAL);
1255 
1256 	if (type < 0 || type >= VM_CAP_MAX)
1257 		return (EINVAL);
1258 
1259 	return (VMGETCAP(vm->cookie, vcpu, type, retval));
1260 }
1261 
1262 int
1263 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
1264 {
1265 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1266 		return (EINVAL);
1267 
1268 	if (type < 0 || type >= VM_CAP_MAX)
1269 		return (EINVAL);
1270 
1271 	return (VMSETCAP(vm->cookie, vcpu, type, val));
1272 }
1273 
1274 uint64_t *
1275 vm_guest_msrs(struct vm *vm, int cpu)
1276 {
1277 	return (vm->vcpu[cpu].guest_msrs);
1278 }
1279 
1280 struct vlapic *
1281 vm_lapic(struct vm *vm, int cpu)
1282 {
1283 	return (vm->vcpu[cpu].vlapic);
1284 }
1285 
1286 struct vioapic *
1287 vm_ioapic(struct vm *vm)
1288 {
1289 
1290 	return (vm->vioapic);
1291 }
1292 
1293 struct vhpet *
1294 vm_hpet(struct vm *vm)
1295 {
1296 
1297 	return (vm->vhpet);
1298 }
1299 
1300 boolean_t
1301 vmm_is_pptdev(int bus, int slot, int func)
1302 {
1303 	int found, i, n;
1304 	int b, s, f;
1305 	char *val, *cp, *cp2;
1306 
1307 	/*
1308 	 * XXX
1309 	 * The length of an environment variable is limited to 128 bytes which
1310 	 * puts an upper limit on the number of passthru devices that may be
1311 	 * specified using a single environment variable.
1312 	 *
1313 	 * Work around this by scanning multiple environment variable
1314 	 * names instead of a single one - yuck!
1315 	 */
1316 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
1317 
1318 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
1319 	found = 0;
1320 	for (i = 0; names[i] != NULL && !found; i++) {
1321 		cp = val = getenv(names[i]);
1322 		while (cp != NULL && *cp != '\0') {
1323 			if ((cp2 = strchr(cp, ' ')) != NULL)
1324 				*cp2 = '\0';
1325 
1326 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
1327 			if (n == 3 && bus == b && slot == s && func == f) {
1328 				found = 1;
1329 				break;
1330 			}
1331 
1332 			if (cp2 != NULL)
1333 				*cp2++ = ' ';
1334 
1335 			cp = cp2;
1336 		}
1337 		freeenv(val);
1338 	}
1339 	return (found);
1340 }
1341 
1342 void *
1343 vm_iommu_domain(struct vm *vm)
1344 {
1345 
1346 	return (vm->iommu);
1347 }
1348 
1349 int
1350 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1351     bool from_idle)
1352 {
1353 	int error;
1354 	struct vcpu *vcpu;
1355 
1356 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1357 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
1358 
1359 	vcpu = &vm->vcpu[vcpuid];
1360 
1361 	vcpu_lock(vcpu);
1362 	error = vcpu_set_state_locked(vcpu, newstate, from_idle);
1363 	vcpu_unlock(vcpu);
1364 
1365 	return (error);
1366 }
1367 
1368 enum vcpu_state
1369 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
1370 {
1371 	struct vcpu *vcpu;
1372 	enum vcpu_state state;
1373 
1374 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1375 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
1376 
1377 	vcpu = &vm->vcpu[vcpuid];
1378 
1379 	vcpu_lock(vcpu);
1380 	state = vcpu->state;
1381 	if (hostcpu != NULL)
1382 		*hostcpu = vcpu->hostcpu;
1383 	vcpu_unlock(vcpu);
1384 
1385 	return (state);
1386 }
1387 
1388 void
1389 vm_activate_cpu(struct vm *vm, int vcpuid)
1390 {
1391 
1392 	if (vcpuid >= 0 && vcpuid < VM_MAXCPU)
1393 		CPU_SET(vcpuid, &vm->active_cpus);
1394 }
1395 
1396 static void
1397 vm_deactivate_cpu(struct vm *vm, int vcpuid)
1398 {
1399 
1400 	if (vcpuid >= 0 && vcpuid < VM_MAXCPU)
1401 		CPU_CLR(vcpuid, &vm->active_cpus);
1402 }
1403 
1404 cpuset_t
1405 vm_active_cpus(struct vm *vm)
1406 {
1407 
1408 	return (vm->active_cpus);
1409 }
1410 
1411 void *
1412 vcpu_stats(struct vm *vm, int vcpuid)
1413 {
1414 
1415 	return (vm->vcpu[vcpuid].stats);
1416 }
1417 
1418 int
1419 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
1420 {
1421 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1422 		return (EINVAL);
1423 
1424 	*state = vm->vcpu[vcpuid].x2apic_state;
1425 
1426 	return (0);
1427 }
1428 
1429 int
1430 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
1431 {
1432 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1433 		return (EINVAL);
1434 
1435 	if (state >= X2APIC_STATE_LAST)
1436 		return (EINVAL);
1437 
1438 	vm->vcpu[vcpuid].x2apic_state = state;
1439 
1440 	vlapic_set_x2apic_state(vm, vcpuid, state);
1441 
1442 	return (0);
1443 }
1444 
1445 /*
1446  * This function is called to ensure that a vcpu "sees" a pending event
1447  * as soon as possible:
1448  * - If the vcpu thread is sleeping then it is woken up.
1449  * - If the vcpu is running on a different host_cpu then an IPI will be directed
1450  *   to the host_cpu to cause the vcpu to trap into the hypervisor.
1451  */
1452 void
1453 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
1454 {
1455 	int hostcpu;
1456 	struct vcpu *vcpu;
1457 
1458 	vcpu = &vm->vcpu[vcpuid];
1459 
1460 	vcpu_lock(vcpu);
1461 	hostcpu = vcpu->hostcpu;
1462 	if (hostcpu == NOCPU) {
1463 		if (vcpu->state == VCPU_SLEEPING)
1464 			wakeup_one(vcpu);
1465 	} else {
1466 		if (vcpu->state != VCPU_RUNNING)
1467 			panic("invalid vcpu state %d", vcpu->state);
1468 		if (hostcpu != curcpu) {
1469 			if (lapic_intr)
1470 				vlapic_post_intr(vcpu->vlapic, hostcpu,
1471 				    vmm_ipinum);
1472 			else
1473 				ipi_cpu(hostcpu, vmm_ipinum);
1474 		}
1475 	}
1476 	vcpu_unlock(vcpu);
1477 }
1478 
1479 struct vmspace *
1480 vm_get_vmspace(struct vm *vm)
1481 {
1482 
1483 	return (vm->vmspace);
1484 }
1485 
1486 int
1487 vm_apicid2vcpuid(struct vm *vm, int apicid)
1488 {
1489 	/*
1490 	 * XXX apic id is assumed to be numerically identical to vcpu id
1491 	 */
1492 	return (apicid);
1493 }
1494 
1495 void
1496 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
1497     vm_rendezvous_func_t func, void *arg)
1498 {
1499 	/*
1500 	 * Enforce that this function is called without any locks
1501 	 */
1502 	WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
1503 	KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
1504 	    ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
1505 
1506 restart:
1507 	mtx_lock(&vm->rendezvous_mtx);
1508 	if (vm->rendezvous_func != NULL) {
1509 		/*
1510 		 * If a rendezvous is already in progress then we need to
1511 		 * call the rendezvous handler in case this 'vcpuid' is one
1512 		 * of the targets of the rendezvous.
1513 		 */
1514 		RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
1515 		mtx_unlock(&vm->rendezvous_mtx);
1516 		vm_handle_rendezvous(vm, vcpuid);
1517 		goto restart;
1518 	}
1519 	KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
1520 	    "rendezvous is still in progress"));
1521 
1522 	RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
1523 	vm->rendezvous_req_cpus = dest;
1524 	CPU_ZERO(&vm->rendezvous_done_cpus);
1525 	vm->rendezvous_arg = arg;
1526 	vm_set_rendezvous_func(vm, func);
1527 	mtx_unlock(&vm->rendezvous_mtx);
1528 
1529 	vm_handle_rendezvous(vm, vcpuid);
1530 }
1531