xref: /freebsd/sys/amd64/vmm/vmm.c (revision 09127a36371d5f5ac109ad1063e7fe53ee9bf356)
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/vm.h>
56 #include <machine/pcb.h>
57 #include <machine/smp.h>
58 #include <x86/apicreg.h>
59 #include <machine/pmap.h>
60 #include <machine/vmparam.h>
61 
62 #include <machine/vmm.h>
63 #include "vmm_ktr.h"
64 #include "vmm_host.h"
65 #include "vmm_mem.h"
66 #include "vmm_util.h"
67 #include <machine/vmm_dev.h>
68 #include "vlapic.h"
69 #include "vmm_msr.h"
70 #include "vmm_ipi.h"
71 #include "vmm_stat.h"
72 #include "vmm_lapic.h"
73 
74 #include "io/ppt.h"
75 #include "io/iommu.h"
76 
77 struct vlapic;
78 
79 struct vcpu {
80 	int		flags;
81 	enum vcpu_state	state;
82 	struct mtx	mtx;
83 	int		hostcpu;	/* host cpuid this vcpu last ran on */
84 	uint64_t	guest_msrs[VMM_MSR_NUM];
85 	struct vlapic	*vlapic;
86 	int		 vcpuid;
87 	struct savefpu	*guestfpu;	/* guest fpu state */
88 	void		*stats;
89 	struct vm_exit	exitinfo;
90 	enum x2apic_state x2apic_state;
91 	int		nmi_pending;
92 };
93 
94 #define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
95 #define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
96 #define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
97 #define	vcpu_assert_locked(v)	mtx_assert(&((v)->mtx), MA_OWNED)
98 
99 struct mem_seg {
100 	vm_paddr_t	gpa;
101 	size_t		len;
102 	boolean_t	wired;
103 	vm_object_t	object;
104 };
105 #define	VM_MAX_MEMORY_SEGMENTS	2
106 
107 struct vm {
108 	void		*cookie;	/* processor-specific data */
109 	void		*iommu;		/* iommu-specific data */
110 	struct vmspace	*vmspace;	/* guest's address space */
111 	struct vcpu	vcpu[VM_MAXCPU];
112 	int		num_mem_segs;
113 	struct mem_seg	mem_segs[VM_MAX_MEMORY_SEGMENTS];
114 	char		name[VM_MAX_NAMELEN];
115 
116 	/*
117 	 * Set of active vcpus.
118 	 * An active vcpu is one that has been started implicitly (BSP) or
119 	 * explicitly (AP) by sending it a startup ipi.
120 	 */
121 	cpuset_t	active_cpus;
122 };
123 
124 static int vmm_initialized;
125 
126 static struct vmm_ops *ops;
127 #define	VMM_INIT()	(ops != NULL ? (*ops->init)() : 0)
128 #define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
129 
130 #define	VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
131 #define	VMRUN(vmi, vcpu, rip, pmap) \
132 	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap) : ENXIO)
133 #define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
134 #define	VMSPACE_ALLOC(min, max) \
135 	(ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
136 #define	VMSPACE_FREE(vmspace) \
137 	(ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
138 #define	VMGETREG(vmi, vcpu, num, retval)		\
139 	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
140 #define	VMSETREG(vmi, vcpu, num, val)		\
141 	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
142 #define	VMGETDESC(vmi, vcpu, num, desc)		\
143 	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
144 #define	VMSETDESC(vmi, vcpu, num, desc)		\
145 	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
146 #define	VMINJECT(vmi, vcpu, type, vec, ec, ecv)	\
147 	(ops != NULL ? (*ops->vminject)(vmi, vcpu, type, vec, ec, ecv) : ENXIO)
148 #define	VMGETCAP(vmi, vcpu, num, retval)	\
149 	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
150 #define	VMSETCAP(vmi, vcpu, num, val)		\
151 	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
152 
153 #define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
154 #define	fpu_stop_emulating()	clts()
155 
156 static MALLOC_DEFINE(M_VM, "vm", "vm");
157 CTASSERT(VMM_MSR_NUM <= 64);	/* msr_mask can keep track of up to 64 msrs */
158 
159 /* statistics */
160 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
161 
162 static void
163 vcpu_cleanup(struct vcpu *vcpu)
164 {
165 	vlapic_cleanup(vcpu->vlapic);
166 	vmm_stat_free(vcpu->stats);
167 	fpu_save_area_free(vcpu->guestfpu);
168 }
169 
170 static void
171 vcpu_init(struct vm *vm, uint32_t vcpu_id)
172 {
173 	struct vcpu *vcpu;
174 
175 	vcpu = &vm->vcpu[vcpu_id];
176 
177 	vcpu_lock_init(vcpu);
178 	vcpu->hostcpu = NOCPU;
179 	vcpu->vcpuid = vcpu_id;
180 	vcpu->vlapic = vlapic_init(vm, vcpu_id);
181 	vm_set_x2apic_state(vm, vcpu_id, X2APIC_ENABLED);
182 	vcpu->guestfpu = fpu_save_area_alloc();
183 	fpu_save_area_reset(vcpu->guestfpu);
184 	vcpu->stats = vmm_stat_alloc();
185 }
186 
187 struct vm_exit *
188 vm_exitinfo(struct vm *vm, int cpuid)
189 {
190 	struct vcpu *vcpu;
191 
192 	if (cpuid < 0 || cpuid >= VM_MAXCPU)
193 		panic("vm_exitinfo: invalid cpuid %d", cpuid);
194 
195 	vcpu = &vm->vcpu[cpuid];
196 
197 	return (&vcpu->exitinfo);
198 }
199 
200 static int
201 vmm_init(void)
202 {
203 	int error;
204 
205 	vmm_host_state_init();
206 	vmm_ipi_init();
207 
208 	error = vmm_mem_init();
209 	if (error)
210 		return (error);
211 
212 	if (vmm_is_intel())
213 		ops = &vmm_ops_intel;
214 	else if (vmm_is_amd())
215 		ops = &vmm_ops_amd;
216 	else
217 		return (ENXIO);
218 
219 	vmm_msr_init();
220 
221 	return (VMM_INIT());
222 }
223 
224 static int
225 vmm_handler(module_t mod, int what, void *arg)
226 {
227 	int error;
228 
229 	switch (what) {
230 	case MOD_LOAD:
231 		vmmdev_init();
232 		iommu_init();
233 		error = vmm_init();
234 		if (error == 0)
235 			vmm_initialized = 1;
236 		break;
237 	case MOD_UNLOAD:
238 		error = vmmdev_cleanup();
239 		if (error == 0) {
240 			iommu_cleanup();
241 			vmm_ipi_cleanup();
242 			error = VMM_CLEANUP();
243 			/*
244 			 * Something bad happened - prevent new
245 			 * VMs from being created
246 			 */
247 			if (error)
248 				vmm_initialized = 0;
249 		}
250 		break;
251 	default:
252 		error = 0;
253 		break;
254 	}
255 	return (error);
256 }
257 
258 static moduledata_t vmm_kmod = {
259 	"vmm",
260 	vmm_handler,
261 	NULL
262 };
263 
264 /*
265  * vmm initialization has the following dependencies:
266  *
267  * - iommu initialization must happen after the pci passthru driver has had
268  *   a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
269  *
270  * - VT-x initialization requires smp_rendezvous() and therefore must happen
271  *   after SMP is fully functional (after SI_SUB_SMP).
272  */
273 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
274 MODULE_VERSION(vmm, 1);
275 
276 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
277 
278 int
279 vm_create(const char *name, struct vm **retvm)
280 {
281 	int i;
282 	struct vm *vm;
283 	struct vmspace *vmspace;
284 
285 	const int BSP = 0;
286 
287 	/*
288 	 * If vmm.ko could not be successfully initialized then don't attempt
289 	 * to create the virtual machine.
290 	 */
291 	if (!vmm_initialized)
292 		return (ENXIO);
293 
294 	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
295 		return (EINVAL);
296 
297 	vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
298 	if (vmspace == NULL)
299 		return (ENOMEM);
300 
301 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
302 	strcpy(vm->name, name);
303 	vm->cookie = VMINIT(vm, vmspace_pmap(vmspace));
304 
305 	for (i = 0; i < VM_MAXCPU; i++) {
306 		vcpu_init(vm, i);
307 		guest_msrs_init(vm, i);
308 	}
309 
310 	vm_activate_cpu(vm, BSP);
311 	vm->vmspace = vmspace;
312 
313 	*retvm = vm;
314 	return (0);
315 }
316 
317 static void
318 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg)
319 {
320 
321 	if (seg->object != NULL)
322 		vmm_mem_free(vm->vmspace, seg->gpa, seg->len);
323 
324 	bzero(seg, sizeof(*seg));
325 }
326 
327 void
328 vm_destroy(struct vm *vm)
329 {
330 	int i;
331 
332 	ppt_unassign_all(vm);
333 
334 	if (vm->iommu != NULL)
335 		iommu_destroy_domain(vm->iommu);
336 
337 	for (i = 0; i < vm->num_mem_segs; i++)
338 		vm_free_mem_seg(vm, &vm->mem_segs[i]);
339 
340 	vm->num_mem_segs = 0;
341 
342 	for (i = 0; i < VM_MAXCPU; i++)
343 		vcpu_cleanup(&vm->vcpu[i]);
344 
345 	VMSPACE_FREE(vm->vmspace);
346 
347 	VMCLEANUP(vm->cookie);
348 
349 	free(vm, M_VM);
350 }
351 
352 const char *
353 vm_name(struct vm *vm)
354 {
355 	return (vm->name);
356 }
357 
358 int
359 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
360 {
361 	vm_object_t obj;
362 
363 	if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
364 		return (ENOMEM);
365 	else
366 		return (0);
367 }
368 
369 int
370 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
371 {
372 
373 	vmm_mmio_free(vm->vmspace, gpa, len);
374 	return (0);
375 }
376 
377 boolean_t
378 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa)
379 {
380 	int i;
381 	vm_paddr_t gpabase, gpalimit;
382 
383 	for (i = 0; i < vm->num_mem_segs; i++) {
384 		gpabase = vm->mem_segs[i].gpa;
385 		gpalimit = gpabase + vm->mem_segs[i].len;
386 		if (gpa >= gpabase && gpa < gpalimit)
387 			return (TRUE);		/* 'gpa' is regular memory */
388 	}
389 
390 	if (ppt_is_mmio(vm, gpa))
391 		return (TRUE);			/* 'gpa' is pci passthru mmio */
392 
393 	return (FALSE);
394 }
395 
396 int
397 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
398 {
399 	int available, allocated;
400 	struct mem_seg *seg;
401 	vm_object_t object;
402 	vm_paddr_t g;
403 
404 	if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
405 		return (EINVAL);
406 
407 	available = allocated = 0;
408 	g = gpa;
409 	while (g < gpa + len) {
410 		if (vm_mem_allocated(vm, g))
411 			allocated++;
412 		else
413 			available++;
414 
415 		g += PAGE_SIZE;
416 	}
417 
418 	/*
419 	 * If there are some allocated and some available pages in the address
420 	 * range then it is an error.
421 	 */
422 	if (allocated && available)
423 		return (EINVAL);
424 
425 	/*
426 	 * If the entire address range being requested has already been
427 	 * allocated then there isn't anything more to do.
428 	 */
429 	if (allocated && available == 0)
430 		return (0);
431 
432 	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
433 		return (E2BIG);
434 
435 	seg = &vm->mem_segs[vm->num_mem_segs];
436 
437 	if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL)
438 		return (ENOMEM);
439 
440 	seg->gpa = gpa;
441 	seg->len = len;
442 	seg->object = object;
443 	seg->wired = FALSE;
444 
445 	vm->num_mem_segs++;
446 
447 	return (0);
448 }
449 
450 static void
451 vm_gpa_unwire(struct vm *vm)
452 {
453 	int i, rv;
454 	struct mem_seg *seg;
455 
456 	for (i = 0; i < vm->num_mem_segs; i++) {
457 		seg = &vm->mem_segs[i];
458 		if (!seg->wired)
459 			continue;
460 
461 		rv = vm_map_unwire(&vm->vmspace->vm_map,
462 				   seg->gpa, seg->gpa + seg->len,
463 				   VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
464 		KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment "
465 		    "%#lx/%ld could not be unwired: %d",
466 		    vm_name(vm), seg->gpa, seg->len, rv));
467 
468 		seg->wired = FALSE;
469 	}
470 }
471 
472 static int
473 vm_gpa_wire(struct vm *vm)
474 {
475 	int i, rv;
476 	struct mem_seg *seg;
477 
478 	for (i = 0; i < vm->num_mem_segs; i++) {
479 		seg = &vm->mem_segs[i];
480 		if (seg->wired)
481 			continue;
482 
483 		/* XXX rlimits? */
484 		rv = vm_map_wire(&vm->vmspace->vm_map,
485 				 seg->gpa, seg->gpa + seg->len,
486 				 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
487 		if (rv != KERN_SUCCESS)
488 			break;
489 
490 		seg->wired = TRUE;
491 	}
492 
493 	if (i < vm->num_mem_segs) {
494 		/*
495 		 * Undo the wiring before returning an error.
496 		 */
497 		vm_gpa_unwire(vm);
498 		return (EAGAIN);
499 	}
500 
501 	return (0);
502 }
503 
504 static void
505 vm_iommu_modify(struct vm *vm, boolean_t map)
506 {
507 	int i, sz;
508 	vm_paddr_t gpa, hpa;
509 	struct mem_seg *seg;
510 	void *vp, *cookie, *host_domain;
511 
512 	sz = PAGE_SIZE;
513 	host_domain = iommu_host_domain();
514 
515 	for (i = 0; i < vm->num_mem_segs; i++) {
516 		seg = &vm->mem_segs[i];
517 		KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired",
518 		    vm_name(vm), seg->gpa, seg->len));
519 
520 		gpa = seg->gpa;
521 		while (gpa < seg->gpa + seg->len) {
522 			vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE,
523 					 &cookie);
524 			KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
525 			    vm_name(vm), gpa));
526 
527 			vm_gpa_release(cookie);
528 
529 			hpa = DMAP_TO_PHYS((uintptr_t)vp);
530 			if (map) {
531 				iommu_create_mapping(vm->iommu, gpa, hpa, sz);
532 				iommu_remove_mapping(host_domain, hpa, sz);
533 			} else {
534 				iommu_remove_mapping(vm->iommu, gpa, sz);
535 				iommu_create_mapping(host_domain, hpa, hpa, sz);
536 			}
537 
538 			gpa += PAGE_SIZE;
539 		}
540 	}
541 
542 	/*
543 	 * Invalidate the cached translations associated with the domain
544 	 * from which pages were removed.
545 	 */
546 	if (map)
547 		iommu_invalidate_tlb(host_domain);
548 	else
549 		iommu_invalidate_tlb(vm->iommu);
550 }
551 
552 #define	vm_iommu_unmap(vm)	vm_iommu_modify((vm), FALSE)
553 #define	vm_iommu_map(vm)	vm_iommu_modify((vm), TRUE)
554 
555 int
556 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
557 {
558 	int error;
559 
560 	error = ppt_unassign_device(vm, bus, slot, func);
561 	if (error)
562 		return (error);
563 
564 	if (ppt_num_devices(vm) == 0) {
565 		vm_iommu_unmap(vm);
566 		vm_gpa_unwire(vm);
567 	}
568 	return (0);
569 }
570 
571 int
572 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
573 {
574 	int error;
575 	vm_paddr_t maxaddr;
576 
577 	/*
578 	 * Virtual machines with pci passthru devices get special treatment:
579 	 * - the guest physical memory is wired
580 	 * - the iommu is programmed to do the 'gpa' to 'hpa' translation
581 	 *
582 	 * We need to do this before the first pci passthru device is attached.
583 	 */
584 	if (ppt_num_devices(vm) == 0) {
585 		KASSERT(vm->iommu == NULL,
586 		    ("vm_assign_pptdev: iommu must be NULL"));
587 		maxaddr = vmm_mem_maxaddr();
588 		vm->iommu = iommu_create_domain(maxaddr);
589 
590 		error = vm_gpa_wire(vm);
591 		if (error)
592 			return (error);
593 
594 		vm_iommu_map(vm);
595 	}
596 
597 	error = ppt_assign_device(vm, bus, slot, func);
598 	return (error);
599 }
600 
601 void *
602 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
603 	    void **cookie)
604 {
605 	int count, pageoff;
606 	vm_page_t m;
607 
608 	pageoff = gpa & PAGE_MASK;
609 	if (len > PAGE_SIZE - pageoff)
610 		panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
611 
612 	count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
613 	    trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
614 
615 	if (count == 1) {
616 		*cookie = m;
617 		return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
618 	} else {
619 		*cookie = NULL;
620 		return (NULL);
621 	}
622 }
623 
624 void
625 vm_gpa_release(void *cookie)
626 {
627 	vm_page_t m = cookie;
628 
629 	vm_page_lock(m);
630 	vm_page_unhold(m);
631 	vm_page_unlock(m);
632 }
633 
634 int
635 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
636 		  struct vm_memory_segment *seg)
637 {
638 	int i;
639 
640 	for (i = 0; i < vm->num_mem_segs; i++) {
641 		if (gpabase == vm->mem_segs[i].gpa) {
642 			seg->gpa = vm->mem_segs[i].gpa;
643 			seg->len = vm->mem_segs[i].len;
644 			seg->wired = vm->mem_segs[i].wired;
645 			return (0);
646 		}
647 	}
648 	return (-1);
649 }
650 
651 int
652 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len,
653 	      vm_offset_t *offset, struct vm_object **object)
654 {
655 	int i;
656 	size_t seg_len;
657 	vm_paddr_t seg_gpa;
658 	vm_object_t seg_obj;
659 
660 	for (i = 0; i < vm->num_mem_segs; i++) {
661 		if ((seg_obj = vm->mem_segs[i].object) == NULL)
662 			continue;
663 
664 		seg_gpa = vm->mem_segs[i].gpa;
665 		seg_len = vm->mem_segs[i].len;
666 
667 		if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) {
668 			*offset = gpa - seg_gpa;
669 			*object = seg_obj;
670 			vm_object_reference(seg_obj);
671 			return (0);
672 		}
673 	}
674 
675 	return (EINVAL);
676 }
677 
678 int
679 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
680 {
681 
682 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
683 		return (EINVAL);
684 
685 	if (reg >= VM_REG_LAST)
686 		return (EINVAL);
687 
688 	return (VMGETREG(vm->cookie, vcpu, reg, retval));
689 }
690 
691 int
692 vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
693 {
694 
695 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
696 		return (EINVAL);
697 
698 	if (reg >= VM_REG_LAST)
699 		return (EINVAL);
700 
701 	return (VMSETREG(vm->cookie, vcpu, reg, val));
702 }
703 
704 static boolean_t
705 is_descriptor_table(int reg)
706 {
707 
708 	switch (reg) {
709 	case VM_REG_GUEST_IDTR:
710 	case VM_REG_GUEST_GDTR:
711 		return (TRUE);
712 	default:
713 		return (FALSE);
714 	}
715 }
716 
717 static boolean_t
718 is_segment_register(int reg)
719 {
720 
721 	switch (reg) {
722 	case VM_REG_GUEST_ES:
723 	case VM_REG_GUEST_CS:
724 	case VM_REG_GUEST_SS:
725 	case VM_REG_GUEST_DS:
726 	case VM_REG_GUEST_FS:
727 	case VM_REG_GUEST_GS:
728 	case VM_REG_GUEST_TR:
729 	case VM_REG_GUEST_LDTR:
730 		return (TRUE);
731 	default:
732 		return (FALSE);
733 	}
734 }
735 
736 int
737 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
738 		struct seg_desc *desc)
739 {
740 
741 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
742 		return (EINVAL);
743 
744 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
745 		return (EINVAL);
746 
747 	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
748 }
749 
750 int
751 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
752 		struct seg_desc *desc)
753 {
754 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
755 		return (EINVAL);
756 
757 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
758 		return (EINVAL);
759 
760 	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
761 }
762 
763 static void
764 restore_guest_fpustate(struct vcpu *vcpu)
765 {
766 
767 	/* flush host state to the pcb */
768 	fpuexit(curthread);
769 
770 	/* restore guest FPU state */
771 	fpu_stop_emulating();
772 	fpurestore(vcpu->guestfpu);
773 
774 	/*
775 	 * The FPU is now "dirty" with the guest's state so turn on emulation
776 	 * to trap any access to the FPU by the host.
777 	 */
778 	fpu_start_emulating();
779 }
780 
781 static void
782 save_guest_fpustate(struct vcpu *vcpu)
783 {
784 
785 	if ((rcr0() & CR0_TS) == 0)
786 		panic("fpu emulation not enabled in host!");
787 
788 	/* save guest FPU state */
789 	fpu_stop_emulating();
790 	fpusave(vcpu->guestfpu);
791 	fpu_start_emulating();
792 }
793 
794 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
795 
796 static int
797 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
798 {
799 	int error;
800 
801 	vcpu_assert_locked(vcpu);
802 
803 	/*
804 	 * The following state transitions are allowed:
805 	 * IDLE -> FROZEN -> IDLE
806 	 * FROZEN -> RUNNING -> FROZEN
807 	 * FROZEN -> SLEEPING -> FROZEN
808 	 */
809 	switch (vcpu->state) {
810 	case VCPU_IDLE:
811 	case VCPU_RUNNING:
812 	case VCPU_SLEEPING:
813 		error = (newstate != VCPU_FROZEN);
814 		break;
815 	case VCPU_FROZEN:
816 		error = (newstate == VCPU_FROZEN);
817 		break;
818 	default:
819 		error = 1;
820 		break;
821 	}
822 
823 	if (error == 0)
824 		vcpu->state = newstate;
825 	else
826 		error = EBUSY;
827 
828 	return (error);
829 }
830 
831 static void
832 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
833 {
834 	int error;
835 
836 	if ((error = vcpu_set_state(vm, vcpuid, newstate)) != 0)
837 		panic("Error %d setting state to %d\n", error, newstate);
838 }
839 
840 static void
841 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
842 {
843 	int error;
844 
845 	if ((error = vcpu_set_state_locked(vcpu, newstate)) != 0)
846 		panic("Error %d setting state to %d", error, newstate);
847 }
848 
849 /*
850  * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
851  */
852 static int
853 vm_handle_hlt(struct vm *vm, int vcpuid, boolean_t *retu)
854 {
855 	struct vcpu *vcpu;
856 	int sleepticks, t;
857 
858 	vcpu = &vm->vcpu[vcpuid];
859 
860 	vcpu_lock(vcpu);
861 
862 	/*
863 	 * Figure out the number of host ticks until the next apic
864 	 * timer interrupt in the guest.
865 	 */
866 	sleepticks = lapic_timer_tick(vm, vcpuid);
867 
868 	/*
869 	 * If the guest local apic timer is disabled then sleep for
870 	 * a long time but not forever.
871 	 */
872 	if (sleepticks < 0)
873 		sleepticks = hz;
874 
875 	/*
876 	 * Do a final check for pending NMI or interrupts before
877 	 * really putting this thread to sleep.
878 	 *
879 	 * These interrupts could have happened any time after we
880 	 * returned from VMRUN() and before we grabbed the vcpu lock.
881 	 */
882 	if (!vm_nmi_pending(vm, vcpuid) && lapic_pending_intr(vm, vcpuid) < 0) {
883 		if (sleepticks <= 0)
884 			panic("invalid sleepticks %d", sleepticks);
885 		t = ticks;
886 		vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
887 		msleep_spin(vcpu, &vcpu->mtx, "vmidle", sleepticks);
888 		vcpu_require_state_locked(vcpu, VCPU_FROZEN);
889 		vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
890 	}
891 	vcpu_unlock(vcpu);
892 
893 	return (0);
894 }
895 
896 static int
897 vm_handle_paging(struct vm *vm, int vcpuid, boolean_t *retu)
898 {
899 	int rv, ftype;
900 	struct vm_map *map;
901 	struct vcpu *vcpu;
902 	struct vm_exit *vme;
903 
904 	vcpu = &vm->vcpu[vcpuid];
905 	vme = &vcpu->exitinfo;
906 
907 	ftype = vme->u.paging.fault_type;
908 	KASSERT(ftype == VM_PROT_READ ||
909 	    ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
910 	    ("vm_handle_paging: invalid fault_type %d", ftype));
911 
912 	if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
913 		rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
914 		    vme->u.paging.gpa, ftype);
915 		if (rv == 0)
916 			goto done;
917 	}
918 
919 	map = &vm->vmspace->vm_map;
920 	rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
921 
922 	VMM_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, ftype = %d",
923 		 rv, vme->u.paging.gpa, ftype);
924 
925 	if (rv != KERN_SUCCESS)
926 		return (EFAULT);
927 done:
928 	/* restart execution at the faulting instruction */
929 	vme->inst_length = 0;
930 
931 	return (0);
932 }
933 
934 static int
935 vm_handle_inst_emul(struct vm *vm, int vcpuid, boolean_t *retu)
936 {
937 	struct vie *vie;
938 	struct vcpu *vcpu;
939 	struct vm_exit *vme;
940 	int error, inst_length;
941 	uint64_t rip, gla, gpa, cr3;
942 
943 	vcpu = &vm->vcpu[vcpuid];
944 	vme = &vcpu->exitinfo;
945 
946 	rip = vme->rip;
947 	inst_length = vme->inst_length;
948 
949 	gla = vme->u.inst_emul.gla;
950 	gpa = vme->u.inst_emul.gpa;
951 	cr3 = vme->u.inst_emul.cr3;
952 	vie = &vme->u.inst_emul.vie;
953 
954 	vie_init(vie);
955 
956 	/* Fetch, decode and emulate the faulting instruction */
957 	if (vmm_fetch_instruction(vm, vcpuid, rip, inst_length, cr3, vie) != 0)
958 		return (EFAULT);
959 
960 	if (vmm_decode_instruction(vm, vcpuid, gla, vie) != 0)
961 		return (EFAULT);
962 
963 	/* return to userland unless this is a local apic access */
964 	if (gpa < DEFAULT_APIC_BASE || gpa >= DEFAULT_APIC_BASE + PAGE_SIZE) {
965 		*retu = TRUE;
966 		return (0);
967 	}
968 
969 	error = vmm_emulate_instruction(vm, vcpuid, gpa, vie,
970 					lapic_mmio_read, lapic_mmio_write, 0);
971 
972 	/* return to userland to spin up the AP */
973 	if (error == 0 && vme->exitcode == VM_EXITCODE_SPINUP_AP)
974 		*retu = TRUE;
975 
976 	return (error);
977 }
978 
979 int
980 vm_run(struct vm *vm, struct vm_run *vmrun)
981 {
982 	int error, vcpuid;
983 	struct vcpu *vcpu;
984 	struct pcb *pcb;
985 	uint64_t tscval, rip;
986 	struct vm_exit *vme;
987 	boolean_t retu;
988 	pmap_t pmap;
989 
990 	vcpuid = vmrun->cpuid;
991 
992 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
993 		return (EINVAL);
994 
995 	pmap = vmspace_pmap(vm->vmspace);
996 	vcpu = &vm->vcpu[vcpuid];
997 	vme = &vcpu->exitinfo;
998 	rip = vmrun->rip;
999 restart:
1000 	critical_enter();
1001 
1002 	KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1003 	    ("vm_run: absurd pm_active"));
1004 
1005 	tscval = rdtsc();
1006 
1007 	pcb = PCPU_GET(curpcb);
1008 	set_pcb_flags(pcb, PCB_FULL_IRET);
1009 
1010 	restore_guest_msrs(vm, vcpuid);
1011 	restore_guest_fpustate(vcpu);
1012 
1013 	vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1014 	vcpu->hostcpu = curcpu;
1015 	error = VMRUN(vm->cookie, vcpuid, rip, pmap);
1016 	vcpu->hostcpu = NOCPU;
1017 	vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1018 
1019 	save_guest_fpustate(vcpu);
1020 	restore_host_msrs(vm, vcpuid);
1021 
1022 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1023 
1024 	critical_exit();
1025 
1026 	if (error == 0) {
1027 		retu = FALSE;
1028 		switch (vme->exitcode) {
1029 		case VM_EXITCODE_HLT:
1030 			error = vm_handle_hlt(vm, vcpuid, &retu);
1031 			break;
1032 		case VM_EXITCODE_PAGING:
1033 			error = vm_handle_paging(vm, vcpuid, &retu);
1034 			break;
1035 		case VM_EXITCODE_INST_EMUL:
1036 			error = vm_handle_inst_emul(vm, vcpuid, &retu);
1037 			break;
1038 		default:
1039 			retu = TRUE;	/* handled in userland */
1040 			break;
1041 		}
1042 	}
1043 
1044 	if (error == 0 && retu == FALSE) {
1045 		rip = vme->rip + vme->inst_length;
1046 		goto restart;
1047 	}
1048 
1049 	/* copy the exit information */
1050 	bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1051 	return (error);
1052 }
1053 
1054 int
1055 vm_inject_event(struct vm *vm, int vcpuid, int type,
1056 		int vector, uint32_t code, int code_valid)
1057 {
1058 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1059 		return (EINVAL);
1060 
1061 	if ((type > VM_EVENT_NONE && type < VM_EVENT_MAX) == 0)
1062 		return (EINVAL);
1063 
1064 	if (vector < 0 || vector > 255)
1065 		return (EINVAL);
1066 
1067 	return (VMINJECT(vm->cookie, vcpuid, type, vector, code, code_valid));
1068 }
1069 
1070 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
1071 
1072 int
1073 vm_inject_nmi(struct vm *vm, int vcpuid)
1074 {
1075 	struct vcpu *vcpu;
1076 
1077 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1078 		return (EINVAL);
1079 
1080 	vcpu = &vm->vcpu[vcpuid];
1081 
1082 	vcpu->nmi_pending = 1;
1083 	vm_interrupt_hostcpu(vm, vcpuid);
1084 	return (0);
1085 }
1086 
1087 int
1088 vm_nmi_pending(struct vm *vm, int vcpuid)
1089 {
1090 	struct vcpu *vcpu;
1091 
1092 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1093 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1094 
1095 	vcpu = &vm->vcpu[vcpuid];
1096 
1097 	return (vcpu->nmi_pending);
1098 }
1099 
1100 void
1101 vm_nmi_clear(struct vm *vm, int vcpuid)
1102 {
1103 	struct vcpu *vcpu;
1104 
1105 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1106 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1107 
1108 	vcpu = &vm->vcpu[vcpuid];
1109 
1110 	if (vcpu->nmi_pending == 0)
1111 		panic("vm_nmi_clear: inconsistent nmi_pending state");
1112 
1113 	vcpu->nmi_pending = 0;
1114 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
1115 }
1116 
1117 int
1118 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
1119 {
1120 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1121 		return (EINVAL);
1122 
1123 	if (type < 0 || type >= VM_CAP_MAX)
1124 		return (EINVAL);
1125 
1126 	return (VMGETCAP(vm->cookie, vcpu, type, retval));
1127 }
1128 
1129 int
1130 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
1131 {
1132 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
1133 		return (EINVAL);
1134 
1135 	if (type < 0 || type >= VM_CAP_MAX)
1136 		return (EINVAL);
1137 
1138 	return (VMSETCAP(vm->cookie, vcpu, type, val));
1139 }
1140 
1141 uint64_t *
1142 vm_guest_msrs(struct vm *vm, int cpu)
1143 {
1144 	return (vm->vcpu[cpu].guest_msrs);
1145 }
1146 
1147 struct vlapic *
1148 vm_lapic(struct vm *vm, int cpu)
1149 {
1150 	return (vm->vcpu[cpu].vlapic);
1151 }
1152 
1153 boolean_t
1154 vmm_is_pptdev(int bus, int slot, int func)
1155 {
1156 	int found, i, n;
1157 	int b, s, f;
1158 	char *val, *cp, *cp2;
1159 
1160 	/*
1161 	 * XXX
1162 	 * The length of an environment variable is limited to 128 bytes which
1163 	 * puts an upper limit on the number of passthru devices that may be
1164 	 * specified using a single environment variable.
1165 	 *
1166 	 * Work around this by scanning multiple environment variable
1167 	 * names instead of a single one - yuck!
1168 	 */
1169 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
1170 
1171 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
1172 	found = 0;
1173 	for (i = 0; names[i] != NULL && !found; i++) {
1174 		cp = val = getenv(names[i]);
1175 		while (cp != NULL && *cp != '\0') {
1176 			if ((cp2 = strchr(cp, ' ')) != NULL)
1177 				*cp2 = '\0';
1178 
1179 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
1180 			if (n == 3 && bus == b && slot == s && func == f) {
1181 				found = 1;
1182 				break;
1183 			}
1184 
1185 			if (cp2 != NULL)
1186 				*cp2++ = ' ';
1187 
1188 			cp = cp2;
1189 		}
1190 		freeenv(val);
1191 	}
1192 	return (found);
1193 }
1194 
1195 void *
1196 vm_iommu_domain(struct vm *vm)
1197 {
1198 
1199 	return (vm->iommu);
1200 }
1201 
1202 int
1203 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1204 {
1205 	int error;
1206 	struct vcpu *vcpu;
1207 
1208 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1209 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
1210 
1211 	vcpu = &vm->vcpu[vcpuid];
1212 
1213 	vcpu_lock(vcpu);
1214 	error = vcpu_set_state_locked(vcpu, newstate);
1215 	vcpu_unlock(vcpu);
1216 
1217 	return (error);
1218 }
1219 
1220 enum vcpu_state
1221 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
1222 {
1223 	struct vcpu *vcpu;
1224 	enum vcpu_state state;
1225 
1226 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1227 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
1228 
1229 	vcpu = &vm->vcpu[vcpuid];
1230 
1231 	vcpu_lock(vcpu);
1232 	state = vcpu->state;
1233 	if (hostcpu != NULL)
1234 		*hostcpu = vcpu->hostcpu;
1235 	vcpu_unlock(vcpu);
1236 
1237 	return (state);
1238 }
1239 
1240 void
1241 vm_activate_cpu(struct vm *vm, int vcpuid)
1242 {
1243 
1244 	if (vcpuid >= 0 && vcpuid < VM_MAXCPU)
1245 		CPU_SET(vcpuid, &vm->active_cpus);
1246 }
1247 
1248 cpuset_t
1249 vm_active_cpus(struct vm *vm)
1250 {
1251 
1252 	return (vm->active_cpus);
1253 }
1254 
1255 void *
1256 vcpu_stats(struct vm *vm, int vcpuid)
1257 {
1258 
1259 	return (vm->vcpu[vcpuid].stats);
1260 }
1261 
1262 int
1263 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
1264 {
1265 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1266 		return (EINVAL);
1267 
1268 	*state = vm->vcpu[vcpuid].x2apic_state;
1269 
1270 	return (0);
1271 }
1272 
1273 int
1274 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
1275 {
1276 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1277 		return (EINVAL);
1278 
1279 	if (state >= X2APIC_STATE_LAST)
1280 		return (EINVAL);
1281 
1282 	vm->vcpu[vcpuid].x2apic_state = state;
1283 
1284 	vlapic_set_x2apic_state(vm, vcpuid, state);
1285 
1286 	return (0);
1287 }
1288 
1289 void
1290 vm_interrupt_hostcpu(struct vm *vm, int vcpuid)
1291 {
1292 	int hostcpu;
1293 	struct vcpu *vcpu;
1294 
1295 	vcpu = &vm->vcpu[vcpuid];
1296 
1297 	vcpu_lock(vcpu);
1298 	hostcpu = vcpu->hostcpu;
1299 	if (hostcpu == NOCPU) {
1300 		if (vcpu->state == VCPU_SLEEPING)
1301 			wakeup_one(vcpu);
1302 	} else {
1303 		if (vcpu->state != VCPU_RUNNING)
1304 			panic("invalid vcpu state %d", vcpu->state);
1305 		if (hostcpu != curcpu)
1306 			ipi_cpu(hostcpu, vmm_ipinum);
1307 	}
1308 	vcpu_unlock(vcpu);
1309 }
1310 
1311 struct vmspace *
1312 vm_get_vmspace(struct vm *vm)
1313 {
1314 
1315 	return (vm->vmspace);
1316 }
1317