vmm.c revision 242275 
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/sched.h>
43#include <sys/smp.h>
44#include <sys/systm.h>
45
46#include <vm/vm.h>
47
48#include <machine/vm.h>
49#include <machine/pcb.h>
50#include <machine/smp.h>
51#include <x86/apicreg.h>
52
53#include <machine/vmm.h>
54#include "vmm_host.h"
55#include "vmm_mem.h"
56#include "vmm_util.h"
57#include <machine/vmm_dev.h>
58#include "vlapic.h"
59#include "vmm_msr.h"
60#include "vmm_ipi.h"
61#include "vmm_stat.h"
62#include "vmm_lapic.h"
63
64#include "io/ppt.h"
65#include "io/iommu.h"
66
67struct vlapic;
68
69struct vcpu {
70	int		flags;
71	enum vcpu_state	state;
72	struct mtx	mtx;
73	int		pincpu;		/* host cpuid this vcpu is bound to */
74	int		hostcpu;	/* host cpuid this vcpu last ran on */
75	uint64_t	guest_msrs[VMM_MSR_NUM];
76	struct vlapic	*vlapic;
77	int		 vcpuid;
78	struct savefpu	*guestfpu;	/* guest fpu state */
79	void		*stats;
80	struct vm_exit	exitinfo;
81	enum x2apic_state x2apic_state;
82	int		nmi_pending;
83};
84#define	VCPU_F_PINNED	0x0001
85
86#define	VCPU_PINCPU(vm, vcpuid)	\
87    ((vm->vcpu[vcpuid].flags & VCPU_F_PINNED) ? vm->vcpu[vcpuid].pincpu : -1)
88
89#define	VCPU_UNPIN(vm, vcpuid)	(vm->vcpu[vcpuid].flags &= ~VCPU_F_PINNED)
90
91#define	VCPU_PIN(vm, vcpuid, host_cpuid)				\
92do {									\
93	vm->vcpu[vcpuid].flags |= VCPU_F_PINNED;			\
94	vm->vcpu[vcpuid].pincpu = host_cpuid;				\
95} while(0)
96
97#define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
98#define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
99#define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
100
101#define	VM_MAX_MEMORY_SEGMENTS	2
102
103struct vm {
104	void		*cookie;	/* processor-specific data */
105	void		*iommu;		/* iommu-specific data */
106	struct vcpu	vcpu[VM_MAXCPU];
107	int		num_mem_segs;
108	struct vm_memory_segment mem_segs[VM_MAX_MEMORY_SEGMENTS];
109	char		name[VM_MAX_NAMELEN];
110
111	/*
112	 * Set of active vcpus.
113	 * An active vcpu is one that has been started implicitly (BSP) or
114	 * explicitly (AP) by sending it a startup ipi.
115	 */
116	cpuset_t	active_cpus;
117};
118
119static struct vmm_ops *ops;
120#define	VMM_INIT()	(ops != NULL ? (*ops->init)() : 0)
121#define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
122
123#define	VMINIT(vm)	(ops != NULL ? (*ops->vminit)(vm): NULL)
124#define	VMRUN(vmi, vcpu, rip) \
125	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip) : ENXIO)
126#define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
127#define	VMMMAP_SET(vmi, gpa, hpa, len, attr, prot, spm)			\
128    	(ops != NULL ? 							\
129    	(*ops->vmmmap_set)(vmi, gpa, hpa, len, attr, prot, spm) :	\
130	ENXIO)
131#define	VMMMAP_GET(vmi, gpa) \
132	(ops != NULL ? (*ops->vmmmap_get)(vmi, gpa) : ENXIO)
133#define	VMGETREG(vmi, vcpu, num, retval)		\
134	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
135#define	VMSETREG(vmi, vcpu, num, val)		\
136	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
137#define	VMGETDESC(vmi, vcpu, num, desc)		\
138	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
139#define	VMSETDESC(vmi, vcpu, num, desc)		\
140	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
141#define	VMINJECT(vmi, vcpu, type, vec, ec, ecv)	\
142	(ops != NULL ? (*ops->vminject)(vmi, vcpu, type, vec, ec, ecv) : ENXIO)
143#define	VMGETCAP(vmi, vcpu, num, retval)	\
144	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
145#define	VMSETCAP(vmi, vcpu, num, val)		\
146	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
147
148#define	fpu_start_emulating()	start_emulating()
149#define	fpu_stop_emulating()	stop_emulating()
150
151static MALLOC_DEFINE(M_VM, "vm", "vm");
152CTASSERT(VMM_MSR_NUM <= 64);	/* msr_mask can keep track of up to 64 msrs */
153
154/* statistics */
155static VMM_STAT_DEFINE(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
156
157static void
158vcpu_cleanup(struct vcpu *vcpu)
159{
160	vlapic_cleanup(vcpu->vlapic);
161	vmm_stat_free(vcpu->stats);
162	fpu_save_area_free(vcpu->guestfpu);
163}
164
165static void
166vcpu_init(struct vm *vm, uint32_t vcpu_id)
167{
168	struct vcpu *vcpu;
169
170	vcpu = &vm->vcpu[vcpu_id];
171
172	vcpu_lock_init(vcpu);
173	vcpu->hostcpu = NOCPU;
174	vcpu->vcpuid = vcpu_id;
175	vcpu->vlapic = vlapic_init(vm, vcpu_id);
176	vm_set_x2apic_state(vm, vcpu_id, X2APIC_ENABLED);
177	vcpu->guestfpu = fpu_save_area_alloc();
178	fpu_save_area_reset(vcpu->guestfpu);
179	vcpu->stats = vmm_stat_alloc();
180}
181
182struct vm_exit *
183vm_exitinfo(struct vm *vm, int cpuid)
184{
185	struct vcpu *vcpu;
186
187	if (cpuid < 0 || cpuid >= VM_MAXCPU)
188		panic("vm_exitinfo: invalid cpuid %d", cpuid);
189
190	vcpu = &vm->vcpu[cpuid];
191
192	return (&vcpu->exitinfo);
193}
194
195static int
196vmm_init(void)
197{
198	int error;
199
200	vmm_host_state_init();
201	vmm_ipi_init();
202
203	error = vmm_mem_init();
204	if (error)
205		return (error);
206
207	if (vmm_is_intel())
208		ops = &vmm_ops_intel;
209	else if (vmm_is_amd())
210		ops = &vmm_ops_amd;
211	else
212		return (ENXIO);
213
214	vmm_msr_init();
215
216	return (VMM_INIT());
217}
218
219static int
220vmm_handler(module_t mod, int what, void *arg)
221{
222	int error;
223
224	switch (what) {
225	case MOD_LOAD:
226		vmmdev_init();
227		iommu_init();
228		error = vmm_init();
229		break;
230	case MOD_UNLOAD:
231		error = vmmdev_cleanup();
232		if (error == 0) {
233			iommu_cleanup();
234			vmm_ipi_cleanup();
235			error = VMM_CLEANUP();
236		}
237		break;
238	default:
239		error = 0;
240		break;
241	}
242	return (error);
243}
244
245static moduledata_t vmm_kmod = {
246	"vmm",
247	vmm_handler,
248	NULL
249};
250
251/*
252 * Execute the module load handler after the pci passthru driver has had
253 * a chance to claim devices. We need this information at the time we do
254 * iommu initialization.
255 */
256DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_CONFIGURE + 1, SI_ORDER_ANY);
257MODULE_VERSION(vmm, 1);
258
259SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
260
261struct vm *
262vm_create(const char *name)
263{
264	int i;
265	struct vm *vm;
266	vm_paddr_t maxaddr;
267
268	const int BSP = 0;
269
270	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
271		return (NULL);
272
273	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
274	strcpy(vm->name, name);
275	vm->cookie = VMINIT(vm);
276
277	for (i = 0; i < VM_MAXCPU; i++) {
278		vcpu_init(vm, i);
279		guest_msrs_init(vm, i);
280	}
281
282	maxaddr = vmm_mem_maxaddr();
283	vm->iommu = iommu_create_domain(maxaddr);
284	vm_activate_cpu(vm, BSP);
285
286	return (vm);
287}
288
289static void
290vm_free_mem_seg(struct vm *vm, struct vm_memory_segment *seg)
291{
292	size_t len;
293	vm_paddr_t hpa;
294	void *host_domain;
295
296	host_domain = iommu_host_domain();
297
298	len = 0;
299	while (len < seg->len) {
300		hpa = vm_gpa2hpa(vm, seg->gpa + len, PAGE_SIZE);
301		if (hpa == (vm_paddr_t)-1) {
302			panic("vm_free_mem_segs: cannot free hpa "
303			      "associated with gpa 0x%016lx", seg->gpa + len);
304		}
305
306		/*
307		 * Remove the 'gpa' to 'hpa' mapping in VMs domain.
308		 * And resurrect the 1:1 mapping for 'hpa' in 'host_domain'.
309		 */
310		iommu_remove_mapping(vm->iommu, seg->gpa + len, PAGE_SIZE);
311		iommu_create_mapping(host_domain, hpa, hpa, PAGE_SIZE);
312
313		vmm_mem_free(hpa, PAGE_SIZE);
314
315		len += PAGE_SIZE;
316	}
317
318	/*
319	 * Invalidate cached translations associated with 'vm->iommu' since
320	 * we have now moved some pages from it.
321	 */
322	iommu_invalidate_tlb(vm->iommu);
323
324	bzero(seg, sizeof(struct vm_memory_segment));
325}
326
327void
328vm_destroy(struct vm *vm)
329{
330	int i;
331
332	ppt_unassign_all(vm);
333
334	for (i = 0; i < vm->num_mem_segs; i++)
335		vm_free_mem_seg(vm, &vm->mem_segs[i]);
336
337	vm->num_mem_segs = 0;
338
339	for (i = 0; i < VM_MAXCPU; i++)
340		vcpu_cleanup(&vm->vcpu[i]);
341
342	iommu_destroy_domain(vm->iommu);
343
344	VMCLEANUP(vm->cookie);
345
346	free(vm, M_VM);
347}
348
349const char *
350vm_name(struct vm *vm)
351{
352	return (vm->name);
353}
354
355int
356vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
357{
358	const boolean_t spok = TRUE;	/* superpage mappings are ok */
359
360	return (VMMMAP_SET(vm->cookie, gpa, hpa, len, VM_MEMATTR_UNCACHEABLE,
361			   VM_PROT_RW, spok));
362}
363
364int
365vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
366{
367	const boolean_t spok = TRUE;	/* superpage mappings are ok */
368
369	return (VMMMAP_SET(vm->cookie, gpa, 0, len, 0,
370			   VM_PROT_NONE, spok));
371}
372
373/*
374 * Returns TRUE if 'gpa' is available for allocation and FALSE otherwise
375 */
376static boolean_t
377vm_gpa_available(struct vm *vm, vm_paddr_t gpa)
378{
379	int i;
380	vm_paddr_t gpabase, gpalimit;
381
382	if (gpa & PAGE_MASK)
383		panic("vm_gpa_available: gpa (0x%016lx) not page aligned", gpa);
384
385	for (i = 0; i < vm->num_mem_segs; i++) {
386		gpabase = vm->mem_segs[i].gpa;
387		gpalimit = gpabase + vm->mem_segs[i].len;
388		if (gpa >= gpabase && gpa < gpalimit)
389			return (FALSE);
390	}
391
392	return (TRUE);
393}
394
395int
396vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
397{
398	int error, available, allocated;
399	struct vm_memory_segment *seg;
400	vm_paddr_t g, hpa;
401	void *host_domain;
402
403	const boolean_t spok = TRUE;	/* superpage mappings are ok */
404
405	if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
406		return (EINVAL);
407
408	available = allocated = 0;
409	g = gpa;
410	while (g < gpa + len) {
411		if (vm_gpa_available(vm, g))
412			available++;
413		else
414			allocated++;
415
416		g += PAGE_SIZE;
417	}
418
419	/*
420	 * If there are some allocated and some available pages in the address
421	 * range then it is an error.
422	 */
423	if (allocated && available)
424		return (EINVAL);
425
426	/*
427	 * If the entire address range being requested has already been
428	 * allocated then there isn't anything more to do.
429	 */
430	if (allocated && available == 0)
431		return (0);
432
433	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
434		return (E2BIG);
435
436	host_domain = iommu_host_domain();
437
438	seg = &vm->mem_segs[vm->num_mem_segs];
439
440	error = 0;
441	seg->gpa = gpa;
442	seg->len = 0;
443	while (seg->len < len) {
444		hpa = vmm_mem_alloc(PAGE_SIZE);
445		if (hpa == 0) {
446			error = ENOMEM;
447			break;
448		}
449
450		error = VMMMAP_SET(vm->cookie, gpa + seg->len, hpa, PAGE_SIZE,
451				   VM_MEMATTR_WRITE_BACK, VM_PROT_ALL, spok);
452		if (error)
453			break;
454
455		/*
456		 * Remove the 1:1 mapping for 'hpa' from the 'host_domain'.
457		 * Add mapping for 'gpa + seg->len' to 'hpa' in the VMs domain.
458		 */
459		iommu_remove_mapping(host_domain, hpa, PAGE_SIZE);
460		iommu_create_mapping(vm->iommu, gpa + seg->len, hpa, PAGE_SIZE);
461
462		seg->len += PAGE_SIZE;
463	}
464
465	if (error) {
466		vm_free_mem_seg(vm, seg);
467		return (error);
468	}
469
470	/*
471	 * Invalidate cached translations associated with 'host_domain' since
472	 * we have now moved some pages from it.
473	 */
474	iommu_invalidate_tlb(host_domain);
475
476	vm->num_mem_segs++;
477
478	return (0);
479}
480
481vm_paddr_t
482vm_gpa2hpa(struct vm *vm, vm_paddr_t gpa, size_t len)
483{
484	vm_paddr_t nextpage;
485
486	nextpage = rounddown(gpa + PAGE_SIZE, PAGE_SIZE);
487	if (len > nextpage - gpa)
488		panic("vm_gpa2hpa: invalid gpa/len: 0x%016lx/%lu", gpa, len);
489
490	return (VMMMAP_GET(vm->cookie, gpa));
491}
492
493int
494vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
495		  struct vm_memory_segment *seg)
496{
497	int i;
498
499	for (i = 0; i < vm->num_mem_segs; i++) {
500		if (gpabase == vm->mem_segs[i].gpa) {
501			*seg = vm->mem_segs[i];
502			return (0);
503		}
504	}
505	return (-1);
506}
507
508int
509vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
510{
511
512	if (vcpu < 0 || vcpu >= VM_MAXCPU)
513		return (EINVAL);
514
515	if (reg >= VM_REG_LAST)
516		return (EINVAL);
517
518	return (VMGETREG(vm->cookie, vcpu, reg, retval));
519}
520
521int
522vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
523{
524
525	if (vcpu < 0 || vcpu >= VM_MAXCPU)
526		return (EINVAL);
527
528	if (reg >= VM_REG_LAST)
529		return (EINVAL);
530
531	return (VMSETREG(vm->cookie, vcpu, reg, val));
532}
533
534static boolean_t
535is_descriptor_table(int reg)
536{
537
538	switch (reg) {
539	case VM_REG_GUEST_IDTR:
540	case VM_REG_GUEST_GDTR:
541		return (TRUE);
542	default:
543		return (FALSE);
544	}
545}
546
547static boolean_t
548is_segment_register(int reg)
549{
550
551	switch (reg) {
552	case VM_REG_GUEST_ES:
553	case VM_REG_GUEST_CS:
554	case VM_REG_GUEST_SS:
555	case VM_REG_GUEST_DS:
556	case VM_REG_GUEST_FS:
557	case VM_REG_GUEST_GS:
558	case VM_REG_GUEST_TR:
559	case VM_REG_GUEST_LDTR:
560		return (TRUE);
561	default:
562		return (FALSE);
563	}
564}
565
566int
567vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
568		struct seg_desc *desc)
569{
570
571	if (vcpu < 0 || vcpu >= VM_MAXCPU)
572		return (EINVAL);
573
574	if (!is_segment_register(reg) && !is_descriptor_table(reg))
575		return (EINVAL);
576
577	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
578}
579
580int
581vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
582		struct seg_desc *desc)
583{
584	if (vcpu < 0 || vcpu >= VM_MAXCPU)
585		return (EINVAL);
586
587	if (!is_segment_register(reg) && !is_descriptor_table(reg))
588		return (EINVAL);
589
590	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
591}
592
593int
594vm_get_pinning(struct vm *vm, int vcpuid, int *cpuid)
595{
596
597	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
598		return (EINVAL);
599
600	*cpuid = VCPU_PINCPU(vm, vcpuid);
601
602	return (0);
603}
604
605int
606vm_set_pinning(struct vm *vm, int vcpuid, int host_cpuid)
607{
608	struct thread *td;
609
610	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
611		return (EINVAL);
612
613	td = curthread;		/* XXXSMP only safe when muxing vcpus */
614
615	/* unpin */
616	if (host_cpuid < 0) {
617		VCPU_UNPIN(vm, vcpuid);
618		thread_lock(td);
619		sched_unbind(td);
620		thread_unlock(td);
621		return (0);
622	}
623
624	if (CPU_ABSENT(host_cpuid))
625		return (EINVAL);
626
627	/*
628	 * XXX we should check that 'host_cpuid' has not already been pinned
629	 * by another vm.
630	 */
631	thread_lock(td);
632	sched_bind(td, host_cpuid);
633	thread_unlock(td);
634	VCPU_PIN(vm, vcpuid, host_cpuid);
635
636	return (0);
637}
638
639static void
640restore_guest_fpustate(struct vcpu *vcpu)
641{
642
643	/* flush host state to the pcb */
644	fpuexit(curthread);
645
646	/* restore guest FPU state */
647	fpu_stop_emulating();
648	fpurestore(vcpu->guestfpu);
649
650	/*
651	 * The FPU is now "dirty" with the guest's state so turn on emulation
652	 * to trap any access to the FPU by the host.
653	 */
654	fpu_start_emulating();
655}
656
657static void
658save_guest_fpustate(struct vcpu *vcpu)
659{
660
661	if ((rcr0() & CR0_TS) == 0)
662		panic("fpu emulation not enabled in host!");
663
664	/* save guest FPU state */
665	fpu_stop_emulating();
666	fpusave(vcpu->guestfpu);
667	fpu_start_emulating();
668}
669
670static VMM_STAT_DEFINE(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
671
672int
673vm_run(struct vm *vm, struct vm_run *vmrun)
674{
675	int error, vcpuid, sleepticks, t;
676	struct vcpu *vcpu;
677	struct pcb *pcb;
678	uint64_t tscval, rip;
679	struct vm_exit *vme;
680
681	vcpuid = vmrun->cpuid;
682
683	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
684		return (EINVAL);
685
686	vcpu = &vm->vcpu[vcpuid];
687	vme = &vmrun->vm_exit;
688	rip = vmrun->rip;
689restart:
690	critical_enter();
691
692	tscval = rdtsc();
693
694	pcb = PCPU_GET(curpcb);
695	set_pcb_flags(pcb, PCB_FULL_IRET);
696
697	restore_guest_msrs(vm, vcpuid);
698	restore_guest_fpustate(vcpu);
699
700	vcpu->hostcpu = curcpu;
701	error = VMRUN(vm->cookie, vcpuid, rip);
702	vcpu->hostcpu = NOCPU;
703
704	save_guest_fpustate(vcpu);
705	restore_host_msrs(vm, vcpuid);
706
707	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
708
709	/* copy the exit information */
710	bcopy(&vcpu->exitinfo, vme, sizeof(struct vm_exit));
711
712	critical_exit();
713
714	/*
715	 * Oblige the guest's desire to 'hlt' by sleeping until the vcpu
716	 * is ready to run.
717	 */
718	if (error == 0 && vme->exitcode == VM_EXITCODE_HLT) {
719		vcpu_lock(vcpu);
720
721		/*
722		 * Figure out the number of host ticks until the next apic
723		 * timer interrupt in the guest.
724		 */
725		sleepticks = lapic_timer_tick(vm, vcpuid);
726
727		/*
728		 * If the guest local apic timer is disabled then sleep for
729		 * a long time but not forever.
730		 */
731		if (sleepticks < 0)
732			sleepticks = hz;
733
734		/*
735		 * Do a final check for pending NMI or interrupts before
736		 * really putting this thread to sleep.
737		 *
738		 * These interrupts could have happened any time after we
739		 * returned from VMRUN() and before we grabbed the vcpu lock.
740		 */
741		if (!vm_nmi_pending(vm, vcpuid) &&
742		    lapic_pending_intr(vm, vcpuid) < 0) {
743			if (sleepticks <= 0)
744				panic("invalid sleepticks %d", sleepticks);
745			t = ticks;
746			msleep_spin(vcpu, &vcpu->mtx, "vmidle", sleepticks);
747			vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
748		}
749
750		vcpu_unlock(vcpu);
751
752		rip = vme->rip + vme->inst_length;
753		goto restart;
754	}
755
756	return (error);
757}
758
759int
760vm_inject_event(struct vm *vm, int vcpuid, int type,
761		int vector, uint32_t code, int code_valid)
762{
763	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
764		return (EINVAL);
765
766	if ((type > VM_EVENT_NONE && type < VM_EVENT_MAX) == 0)
767		return (EINVAL);
768
769	if (vector < 0 || vector > 255)
770		return (EINVAL);
771
772	return (VMINJECT(vm->cookie, vcpuid, type, vector, code, code_valid));
773}
774
775static VMM_STAT_DEFINE(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
776
777int
778vm_inject_nmi(struct vm *vm, int vcpuid)
779{
780	struct vcpu *vcpu;
781
782	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
783		return (EINVAL);
784
785	vcpu = &vm->vcpu[vcpuid];
786
787	vcpu->nmi_pending = 1;
788	vm_interrupt_hostcpu(vm, vcpuid);
789	return (0);
790}
791
792int
793vm_nmi_pending(struct vm *vm, int vcpuid)
794{
795	struct vcpu *vcpu;
796
797	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
798		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
799
800	vcpu = &vm->vcpu[vcpuid];
801
802	return (vcpu->nmi_pending);
803}
804
805void
806vm_nmi_clear(struct vm *vm, int vcpuid)
807{
808	struct vcpu *vcpu;
809
810	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
811		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
812
813	vcpu = &vm->vcpu[vcpuid];
814
815	if (vcpu->nmi_pending == 0)
816		panic("vm_nmi_clear: inconsistent nmi_pending state");
817
818	vcpu->nmi_pending = 0;
819	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
820}
821
822int
823vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
824{
825	if (vcpu < 0 || vcpu >= VM_MAXCPU)
826		return (EINVAL);
827
828	if (type < 0 || type >= VM_CAP_MAX)
829		return (EINVAL);
830
831	return (VMGETCAP(vm->cookie, vcpu, type, retval));
832}
833
834int
835vm_set_capability(struct vm *vm, int vcpu, int type, int val)
836{
837	if (vcpu < 0 || vcpu >= VM_MAXCPU)
838		return (EINVAL);
839
840	if (type < 0 || type >= VM_CAP_MAX)
841		return (EINVAL);
842
843	return (VMSETCAP(vm->cookie, vcpu, type, val));
844}
845
846uint64_t *
847vm_guest_msrs(struct vm *vm, int cpu)
848{
849	return (vm->vcpu[cpu].guest_msrs);
850}
851
852struct vlapic *
853vm_lapic(struct vm *vm, int cpu)
854{
855	return (vm->vcpu[cpu].vlapic);
856}
857
858boolean_t
859vmm_is_pptdev(int bus, int slot, int func)
860{
861	int found, b, s, f, n;
862	char *val, *cp, *cp2;
863
864	/*
865	 * setenv pptdevs "1/2/3 4/5/6 7/8/9 10/11/12"
866	 */
867	found = 0;
868	cp = val = getenv("pptdevs");
869	while (cp != NULL && *cp != '\0') {
870		if ((cp2 = strchr(cp, ' ')) != NULL)
871			*cp2 = '\0';
872
873		n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
874		if (n == 3 && bus == b && slot == s && func == f) {
875			found = 1;
876			break;
877		}
878
879		if (cp2 != NULL)
880			*cp2++ = ' ';
881
882		cp = cp2;
883	}
884	freeenv(val);
885	return (found);
886}
887
888void *
889vm_iommu_domain(struct vm *vm)
890{
891
892	return (vm->iommu);
893}
894
895int
896vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state state)
897{
898	int error;
899	struct vcpu *vcpu;
900
901	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
902		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
903
904	vcpu = &vm->vcpu[vcpuid];
905
906	vcpu_lock(vcpu);
907
908	/*
909	 * The following state transitions are allowed:
910	 * IDLE -> RUNNING -> IDLE
911	 * IDLE -> CANNOT_RUN -> IDLE
912	 */
913	if ((vcpu->state == VCPU_IDLE && state != VCPU_IDLE) ||
914	    (vcpu->state != VCPU_IDLE && state == VCPU_IDLE)) {
915		error = 0;
916		vcpu->state = state;
917	} else {
918		error = EBUSY;
919	}
920
921	vcpu_unlock(vcpu);
922
923	return (error);
924}
925
926enum vcpu_state
927vcpu_get_state(struct vm *vm, int vcpuid)
928{
929	struct vcpu *vcpu;
930	enum vcpu_state state;
931
932	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
933		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
934
935	vcpu = &vm->vcpu[vcpuid];
936
937	vcpu_lock(vcpu);
938	state = vcpu->state;
939	vcpu_unlock(vcpu);
940
941	return (state);
942}
943
944void
945vm_activate_cpu(struct vm *vm, int vcpuid)
946{
947
948	if (vcpuid >= 0 && vcpuid < VM_MAXCPU)
949		CPU_SET(vcpuid, &vm->active_cpus);
950}
951
952cpuset_t
953vm_active_cpus(struct vm *vm)
954{
955
956	return (vm->active_cpus);
957}
958
959void *
960vcpu_stats(struct vm *vm, int vcpuid)
961{
962
963	return (vm->vcpu[vcpuid].stats);
964}
965
966int
967vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
968{
969	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
970		return (EINVAL);
971
972	*state = vm->vcpu[vcpuid].x2apic_state;
973
974	return (0);
975}
976
977int
978vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
979{
980	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
981		return (EINVAL);
982
983	if (state < 0 || state >= X2APIC_STATE_LAST)
984		return (EINVAL);
985
986	vm->vcpu[vcpuid].x2apic_state = state;
987
988	vlapic_set_x2apic_state(vm, vcpuid, state);
989
990	return (0);
991}
992
993void
994vm_interrupt_hostcpu(struct vm *vm, int vcpuid)
995{
996	int hostcpu;
997	struct vcpu *vcpu;
998
999	vcpu = &vm->vcpu[vcpuid];
1000
1001	vcpu_lock(vcpu);
1002	hostcpu = vcpu->hostcpu;
1003	if (hostcpu == NOCPU) {
1004		/*
1005		 * If the vcpu is 'RUNNING' but without a valid 'hostcpu' then
1006		 * the host thread must be sleeping waiting for an event to
1007		 * kick the vcpu out of 'hlt'.
1008		 *
1009		 * XXX this is racy because the condition exists right before
1010		 * and after calling VMRUN() in vm_run(). The wakeup() is
1011		 * benign in this case.
1012		 */
1013		if (vcpu->state == VCPU_RUNNING)
1014			wakeup_one(vcpu);
1015	} else {
1016		if (vcpu->state != VCPU_RUNNING)
1017			panic("invalid vcpu state %d", vcpu->state);
1018		if (hostcpu != curcpu)
1019			ipi_cpu(hostcpu, vmm_ipinum);
1020	}
1021	vcpu_unlock(vcpu);
1022}
1023