vmm.c revision 241982 
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_mem.h"
55#include "vmm_util.h"
56#include <machine/vmm_dev.h>
57#include "vlapic.h"
58#include "vmm_msr.h"
59#include "vmm_ipi.h"
60#include "vmm_stat.h"
61
62#include "io/ppt.h"
63#include "io/iommu.h"
64
65struct vlapic;
66
67struct vcpu {
68	int		flags;
69	enum vcpu_state	state;
70	struct mtx	mtx;
71	int		pincpu;		/* host cpuid this vcpu is bound to */
72	int		hostcpu;	/* host cpuid this vcpu last ran on */
73	uint64_t	guest_msrs[VMM_MSR_NUM];
74	struct vlapic	*vlapic;
75	int		 vcpuid;
76	struct savefpu	*guestfpu;	/* guest fpu state */
77	void		*stats;
78	struct vm_exit	exitinfo;
79	enum x2apic_state x2apic_state;
80	int		nmi_pending;
81};
82#define	VCPU_F_PINNED	0x0001
83
84#define	VCPU_PINCPU(vm, vcpuid)	\
85    ((vm->vcpu[vcpuid].flags & VCPU_F_PINNED) ? vm->vcpu[vcpuid].pincpu : -1)
86
87#define	VCPU_UNPIN(vm, vcpuid)	(vm->vcpu[vcpuid].flags &= ~VCPU_F_PINNED)
88
89#define	VCPU_PIN(vm, vcpuid, host_cpuid)				\
90do {									\
91	vm->vcpu[vcpuid].flags |= VCPU_F_PINNED;			\
92	vm->vcpu[vcpuid].pincpu = host_cpuid;				\
93} while(0)
94
95#define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_DEF)
96#define	vcpu_lock(v)		mtx_lock(&((v)->mtx))
97#define	vcpu_unlock(v)		mtx_unlock(&((v)->mtx))
98
99#define	VM_MAX_MEMORY_SEGMENTS	2
100
101struct vm {
102	void		*cookie;	/* processor-specific data */
103	void		*iommu;		/* iommu-specific data */
104	struct vcpu	vcpu[VM_MAXCPU];
105	int		num_mem_segs;
106	struct vm_memory_segment mem_segs[VM_MAX_MEMORY_SEGMENTS];
107	char		name[VM_MAX_NAMELEN];
108
109	/*
110	 * Set of active vcpus.
111	 * An active vcpu is one that has been started implicitly (BSP) or
112	 * explicitly (AP) by sending it a startup ipi.
113	 */
114	cpuset_t	active_cpus;
115};
116
117static struct vmm_ops *ops;
118#define	VMM_INIT()	(ops != NULL ? (*ops->init)() : 0)
119#define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
120
121#define	VMINIT(vm)	(ops != NULL ? (*ops->vminit)(vm): NULL)
122#define	VMRUN(vmi, vcpu, rip) \
123	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip) : ENXIO)
124#define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
125#define	VMMMAP_SET(vmi, gpa, hpa, len, attr, prot, spm)			\
126    	(ops != NULL ? 							\
127    	(*ops->vmmmap_set)(vmi, gpa, hpa, len, attr, prot, spm) :	\
128	ENXIO)
129#define	VMMMAP_GET(vmi, gpa) \
130	(ops != NULL ? (*ops->vmmmap_get)(vmi, gpa) : ENXIO)
131#define	VMGETREG(vmi, vcpu, num, retval)		\
132	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
133#define	VMSETREG(vmi, vcpu, num, val)		\
134	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
135#define	VMGETDESC(vmi, vcpu, num, desc)		\
136	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
137#define	VMSETDESC(vmi, vcpu, num, desc)		\
138	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
139#define	VMINJECT(vmi, vcpu, type, vec, ec, ecv)	\
140	(ops != NULL ? (*ops->vminject)(vmi, vcpu, type, vec, ec, ecv) : ENXIO)
141#define	VMGETCAP(vmi, vcpu, num, retval)	\
142	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
143#define	VMSETCAP(vmi, vcpu, num, val)		\
144	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
145
146#define	fpu_start_emulating()	start_emulating()
147#define	fpu_stop_emulating()	stop_emulating()
148
149static MALLOC_DEFINE(M_VM, "vm", "vm");
150CTASSERT(VMM_MSR_NUM <= 64);	/* msr_mask can keep track of up to 64 msrs */
151
152/* statistics */
153static VMM_STAT_DEFINE(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
154
155static void
156vcpu_cleanup(struct vcpu *vcpu)
157{
158	vlapic_cleanup(vcpu->vlapic);
159	vmm_stat_free(vcpu->stats);
160	fpu_save_area_free(vcpu->guestfpu);
161}
162
163static void
164vcpu_init(struct vm *vm, uint32_t vcpu_id)
165{
166	struct vcpu *vcpu;
167
168	vcpu = &vm->vcpu[vcpu_id];
169
170	vcpu_lock_init(vcpu);
171	vcpu->hostcpu = NOCPU;
172	vcpu->vcpuid = vcpu_id;
173	vcpu->vlapic = vlapic_init(vm, vcpu_id);
174	vm_set_x2apic_state(vm, vcpu_id, X2APIC_ENABLED);
175	vcpu->guestfpu = fpu_save_area_alloc();
176	fpu_save_area_reset(vcpu->guestfpu);
177	vcpu->stats = vmm_stat_alloc();
178}
179
180struct vm_exit *
181vm_exitinfo(struct vm *vm, int cpuid)
182{
183	struct vcpu *vcpu;
184
185	if (cpuid < 0 || cpuid >= VM_MAXCPU)
186		panic("vm_exitinfo: invalid cpuid %d", cpuid);
187
188	vcpu = &vm->vcpu[cpuid];
189
190	return (&vcpu->exitinfo);
191}
192
193static int
194vmm_init(void)
195{
196	int error;
197
198	vmm_ipi_init();
199
200	error = vmm_mem_init();
201	if (error)
202		return (error);
203
204	if (vmm_is_intel())
205		ops = &vmm_ops_intel;
206	else if (vmm_is_amd())
207		ops = &vmm_ops_amd;
208	else
209		return (ENXIO);
210
211	vmm_msr_init();
212
213	return (VMM_INIT());
214}
215
216static int
217vmm_handler(module_t mod, int what, void *arg)
218{
219	int error;
220
221	switch (what) {
222	case MOD_LOAD:
223		vmmdev_init();
224		iommu_init();
225		error = vmm_init();
226		break;
227	case MOD_UNLOAD:
228		error = vmmdev_cleanup();
229		if (error == 0) {
230			iommu_cleanup();
231			vmm_ipi_cleanup();
232			error = VMM_CLEANUP();
233		}
234		break;
235	default:
236		error = 0;
237		break;
238	}
239	return (error);
240}
241
242static moduledata_t vmm_kmod = {
243	"vmm",
244	vmm_handler,
245	NULL
246};
247
248/*
249 * Execute the module load handler after the pci passthru driver has had
250 * a chance to claim devices. We need this information at the time we do
251 * iommu initialization.
252 */
253DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_CONFIGURE + 1, SI_ORDER_ANY);
254MODULE_VERSION(vmm, 1);
255
256SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
257
258struct vm *
259vm_create(const char *name)
260{
261	int i;
262	struct vm *vm;
263	vm_paddr_t maxaddr;
264
265	const int BSP = 0;
266
267	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
268		return (NULL);
269
270	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
271	strcpy(vm->name, name);
272	vm->cookie = VMINIT(vm);
273
274	for (i = 0; i < VM_MAXCPU; i++) {
275		vcpu_init(vm, i);
276		guest_msrs_init(vm, i);
277	}
278
279	maxaddr = vmm_mem_maxaddr();
280	vm->iommu = iommu_create_domain(maxaddr);
281	vm_activate_cpu(vm, BSP);
282
283	return (vm);
284}
285
286static void
287vm_free_mem_seg(struct vm *vm, struct vm_memory_segment *seg)
288{
289	size_t len;
290	vm_paddr_t hpa;
291	void *host_domain;
292
293	host_domain = iommu_host_domain();
294
295	len = 0;
296	while (len < seg->len) {
297		hpa = vm_gpa2hpa(vm, seg->gpa + len, PAGE_SIZE);
298		if (hpa == (vm_paddr_t)-1) {
299			panic("vm_free_mem_segs: cannot free hpa "
300			      "associated with gpa 0x%016lx", seg->gpa + len);
301		}
302
303		/*
304		 * Remove the 'gpa' to 'hpa' mapping in VMs domain.
305		 * And resurrect the 1:1 mapping for 'hpa' in 'host_domain'.
306		 */
307		iommu_remove_mapping(vm->iommu, seg->gpa + len, PAGE_SIZE);
308		iommu_create_mapping(host_domain, hpa, hpa, PAGE_SIZE);
309
310		vmm_mem_free(hpa, PAGE_SIZE);
311
312		len += PAGE_SIZE;
313	}
314
315	/*
316	 * Invalidate cached translations associated with 'vm->iommu' since
317	 * we have now moved some pages from it.
318	 */
319	iommu_invalidate_tlb(vm->iommu);
320
321	bzero(seg, sizeof(struct vm_memory_segment));
322}
323
324void
325vm_destroy(struct vm *vm)
326{
327	int i;
328
329	ppt_unassign_all(vm);
330
331	for (i = 0; i < vm->num_mem_segs; i++)
332		vm_free_mem_seg(vm, &vm->mem_segs[i]);
333
334	vm->num_mem_segs = 0;
335
336	for (i = 0; i < VM_MAXCPU; i++)
337		vcpu_cleanup(&vm->vcpu[i]);
338
339	iommu_destroy_domain(vm->iommu);
340
341	VMCLEANUP(vm->cookie);
342
343	free(vm, M_VM);
344}
345
346const char *
347vm_name(struct vm *vm)
348{
349	return (vm->name);
350}
351
352int
353vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
354{
355	const boolean_t spok = TRUE;	/* superpage mappings are ok */
356
357	return (VMMMAP_SET(vm->cookie, gpa, hpa, len, VM_MEMATTR_UNCACHEABLE,
358			   VM_PROT_RW, spok));
359}
360
361int
362vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
363{
364	const boolean_t spok = TRUE;	/* superpage mappings are ok */
365
366	return (VMMMAP_SET(vm->cookie, gpa, 0, len, 0,
367			   VM_PROT_NONE, spok));
368}
369
370/*
371 * Returns TRUE if 'gpa' is available for allocation and FALSE otherwise
372 */
373static boolean_t
374vm_gpa_available(struct vm *vm, vm_paddr_t gpa)
375{
376	int i;
377	vm_paddr_t gpabase, gpalimit;
378
379	if (gpa & PAGE_MASK)
380		panic("vm_gpa_available: gpa (0x%016lx) not page aligned", gpa);
381
382	for (i = 0; i < vm->num_mem_segs; i++) {
383		gpabase = vm->mem_segs[i].gpa;
384		gpalimit = gpabase + vm->mem_segs[i].len;
385		if (gpa >= gpabase && gpa < gpalimit)
386			return (FALSE);
387	}
388
389	return (TRUE);
390}
391
392int
393vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
394{
395	int error, available, allocated;
396	struct vm_memory_segment *seg;
397	vm_paddr_t g, hpa;
398	void *host_domain;
399
400	const boolean_t spok = TRUE;	/* superpage mappings are ok */
401
402	if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
403		return (EINVAL);
404
405	available = allocated = 0;
406	g = gpa;
407	while (g < gpa + len) {
408		if (vm_gpa_available(vm, g))
409			available++;
410		else
411			allocated++;
412
413		g += PAGE_SIZE;
414	}
415
416	/*
417	 * If there are some allocated and some available pages in the address
418	 * range then it is an error.
419	 */
420	if (allocated && available)
421		return (EINVAL);
422
423	/*
424	 * If the entire address range being requested has already been
425	 * allocated then there isn't anything more to do.
426	 */
427	if (allocated && available == 0)
428		return (0);
429
430	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
431		return (E2BIG);
432
433	host_domain = iommu_host_domain();
434
435	seg = &vm->mem_segs[vm->num_mem_segs];
436
437	error = 0;
438	seg->gpa = gpa;
439	seg->len = 0;
440	while (seg->len < len) {
441		hpa = vmm_mem_alloc(PAGE_SIZE);
442		if (hpa == 0) {
443			error = ENOMEM;
444			break;
445		}
446
447		error = VMMMAP_SET(vm->cookie, gpa + seg->len, hpa, PAGE_SIZE,
448				   VM_MEMATTR_WRITE_BACK, VM_PROT_ALL, spok);
449		if (error)
450			break;
451
452		/*
453		 * Remove the 1:1 mapping for 'hpa' from the 'host_domain'.
454		 * Add mapping for 'gpa + seg->len' to 'hpa' in the VMs domain.
455		 */
456		iommu_remove_mapping(host_domain, hpa, PAGE_SIZE);
457		iommu_create_mapping(vm->iommu, gpa + seg->len, hpa, PAGE_SIZE);
458
459		seg->len += PAGE_SIZE;
460	}
461
462	if (error) {
463		vm_free_mem_seg(vm, seg);
464		return (error);
465	}
466
467	/*
468	 * Invalidate cached translations associated with 'host_domain' since
469	 * we have now moved some pages from it.
470	 */
471	iommu_invalidate_tlb(host_domain);
472
473	vm->num_mem_segs++;
474
475	return (0);
476}
477
478vm_paddr_t
479vm_gpa2hpa(struct vm *vm, vm_paddr_t gpa, size_t len)
480{
481	vm_paddr_t nextpage;
482
483	nextpage = rounddown(gpa + PAGE_SIZE, PAGE_SIZE);
484	if (len > nextpage - gpa)
485		panic("vm_gpa2hpa: invalid gpa/len: 0x%016lx/%lu", gpa, len);
486
487	return (VMMMAP_GET(vm->cookie, gpa));
488}
489
490int
491vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
492		  struct vm_memory_segment *seg)
493{
494	int i;
495
496	for (i = 0; i < vm->num_mem_segs; i++) {
497		if (gpabase == vm->mem_segs[i].gpa) {
498			*seg = vm->mem_segs[i];
499			return (0);
500		}
501	}
502	return (-1);
503}
504
505int
506vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
507{
508
509	if (vcpu < 0 || vcpu >= VM_MAXCPU)
510		return (EINVAL);
511
512	if (reg >= VM_REG_LAST)
513		return (EINVAL);
514
515	return (VMGETREG(vm->cookie, vcpu, reg, retval));
516}
517
518int
519vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
520{
521
522	if (vcpu < 0 || vcpu >= VM_MAXCPU)
523		return (EINVAL);
524
525	if (reg >= VM_REG_LAST)
526		return (EINVAL);
527
528	return (VMSETREG(vm->cookie, vcpu, reg, val));
529}
530
531static boolean_t
532is_descriptor_table(int reg)
533{
534
535	switch (reg) {
536	case VM_REG_GUEST_IDTR:
537	case VM_REG_GUEST_GDTR:
538		return (TRUE);
539	default:
540		return (FALSE);
541	}
542}
543
544static boolean_t
545is_segment_register(int reg)
546{
547
548	switch (reg) {
549	case VM_REG_GUEST_ES:
550	case VM_REG_GUEST_CS:
551	case VM_REG_GUEST_SS:
552	case VM_REG_GUEST_DS:
553	case VM_REG_GUEST_FS:
554	case VM_REG_GUEST_GS:
555	case VM_REG_GUEST_TR:
556	case VM_REG_GUEST_LDTR:
557		return (TRUE);
558	default:
559		return (FALSE);
560	}
561}
562
563int
564vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
565		struct seg_desc *desc)
566{
567
568	if (vcpu < 0 || vcpu >= VM_MAXCPU)
569		return (EINVAL);
570
571	if (!is_segment_register(reg) && !is_descriptor_table(reg))
572		return (EINVAL);
573
574	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
575}
576
577int
578vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
579		struct seg_desc *desc)
580{
581	if (vcpu < 0 || vcpu >= VM_MAXCPU)
582		return (EINVAL);
583
584	if (!is_segment_register(reg) && !is_descriptor_table(reg))
585		return (EINVAL);
586
587	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
588}
589
590int
591vm_get_pinning(struct vm *vm, int vcpuid, int *cpuid)
592{
593
594	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
595		return (EINVAL);
596
597	*cpuid = VCPU_PINCPU(vm, vcpuid);
598
599	return (0);
600}
601
602int
603vm_set_pinning(struct vm *vm, int vcpuid, int host_cpuid)
604{
605	struct thread *td;
606
607	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
608		return (EINVAL);
609
610	td = curthread;		/* XXXSMP only safe when muxing vcpus */
611
612	/* unpin */
613	if (host_cpuid < 0) {
614		VCPU_UNPIN(vm, vcpuid);
615		thread_lock(td);
616		sched_unbind(td);
617		thread_unlock(td);
618		return (0);
619	}
620
621	if (CPU_ABSENT(host_cpuid))
622		return (EINVAL);
623
624	/*
625	 * XXX we should check that 'host_cpuid' has not already been pinned
626	 * by another vm.
627	 */
628	thread_lock(td);
629	sched_bind(td, host_cpuid);
630	thread_unlock(td);
631	VCPU_PIN(vm, vcpuid, host_cpuid);
632
633	return (0);
634}
635
636static void
637restore_guest_fpustate(struct vcpu *vcpu)
638{
639
640	/* flush host state to the pcb */
641	fpuexit(curthread);
642	fpu_stop_emulating();
643	fpurestore(vcpu->guestfpu);
644}
645
646static void
647save_guest_fpustate(struct vcpu *vcpu)
648{
649
650	fpusave(vcpu->guestfpu);
651	fpu_start_emulating();
652}
653
654int
655vm_run(struct vm *vm, struct vm_run *vmrun)
656{
657	int error, vcpuid;
658	struct vcpu *vcpu;
659	struct pcb *pcb;
660	uint64_t tscval;
661
662	vcpuid = vmrun->cpuid;
663
664	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
665		return (EINVAL);
666
667	vcpu = &vm->vcpu[vcpuid];
668
669	critical_enter();
670
671	tscval = rdtsc();
672
673	pcb = PCPU_GET(curpcb);
674	set_pcb_flags(pcb, PCB_FULL_IRET);
675
676	restore_guest_msrs(vm, vcpuid);
677	restore_guest_fpustate(vcpu);
678
679	vcpu->hostcpu = curcpu;
680	error = VMRUN(vm->cookie, vcpuid, vmrun->rip);
681	vcpu->hostcpu = NOCPU;
682
683	save_guest_fpustate(vcpu);
684	restore_host_msrs(vm, vcpuid);
685
686	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
687
688	/* copy the exit information */
689	bcopy(&vcpu->exitinfo, &vmrun->vm_exit, sizeof(struct vm_exit));
690
691	critical_exit();
692
693	return (error);
694}
695
696int
697vm_inject_event(struct vm *vm, int vcpuid, int type,
698		int vector, uint32_t code, int code_valid)
699{
700	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
701		return (EINVAL);
702
703	if ((type > VM_EVENT_NONE && type < VM_EVENT_MAX) == 0)
704		return (EINVAL);
705
706	if (vector < 0 || vector > 255)
707		return (EINVAL);
708
709	return (VMINJECT(vm->cookie, vcpuid, type, vector, code, code_valid));
710}
711
712VMM_STAT_DEFINE(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
713
714int
715vm_inject_nmi(struct vm *vm, int vcpuid)
716{
717	struct vcpu *vcpu;
718
719	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
720		return (EINVAL);
721
722	vcpu = &vm->vcpu[vcpuid];
723
724	vcpu->nmi_pending = 1;
725	vm_interrupt_hostcpu(vm, vcpuid);
726	return (0);
727}
728
729int
730vm_nmi_pending(struct vm *vm, int vcpuid)
731{
732	struct vcpu *vcpu;
733
734	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
735		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
736
737	vcpu = &vm->vcpu[vcpuid];
738
739	return (vcpu->nmi_pending);
740}
741
742void
743vm_nmi_clear(struct vm *vm, int vcpuid)
744{
745	struct vcpu *vcpu;
746
747	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
748		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
749
750	vcpu = &vm->vcpu[vcpuid];
751
752	if (vcpu->nmi_pending == 0)
753		panic("vm_nmi_clear: inconsistent nmi_pending state");
754
755	vcpu->nmi_pending = 0;
756	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
757}
758
759int
760vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
761{
762	if (vcpu < 0 || vcpu >= VM_MAXCPU)
763		return (EINVAL);
764
765	if (type < 0 || type >= VM_CAP_MAX)
766		return (EINVAL);
767
768	return (VMGETCAP(vm->cookie, vcpu, type, retval));
769}
770
771int
772vm_set_capability(struct vm *vm, int vcpu, int type, int val)
773{
774	if (vcpu < 0 || vcpu >= VM_MAXCPU)
775		return (EINVAL);
776
777	if (type < 0 || type >= VM_CAP_MAX)
778		return (EINVAL);
779
780	return (VMSETCAP(vm->cookie, vcpu, type, val));
781}
782
783uint64_t *
784vm_guest_msrs(struct vm *vm, int cpu)
785{
786	return (vm->vcpu[cpu].guest_msrs);
787}
788
789struct vlapic *
790vm_lapic(struct vm *vm, int cpu)
791{
792	return (vm->vcpu[cpu].vlapic);
793}
794
795boolean_t
796vmm_is_pptdev(int bus, int slot, int func)
797{
798	int found, b, s, f, n;
799	char *val, *cp, *cp2;
800
801	/*
802	 * setenv pptdevs "1/2/3 4/5/6 7/8/9 10/11/12"
803	 */
804	found = 0;
805	cp = val = getenv("pptdevs");
806	while (cp != NULL && *cp != '\0') {
807		if ((cp2 = strchr(cp, ' ')) != NULL)
808			*cp2 = '\0';
809
810		n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
811		if (n == 3 && bus == b && slot == s && func == f) {
812			found = 1;
813			break;
814		}
815
816		if (cp2 != NULL)
817			*cp2++ = ' ';
818
819		cp = cp2;
820	}
821	freeenv(val);
822	return (found);
823}
824
825void *
826vm_iommu_domain(struct vm *vm)
827{
828
829	return (vm->iommu);
830}
831
832int
833vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state state)
834{
835	int error;
836	struct vcpu *vcpu;
837
838	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
839		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
840
841	vcpu = &vm->vcpu[vcpuid];
842
843	vcpu_lock(vcpu);
844
845	/*
846	 * The following state transitions are allowed:
847	 * IDLE -> RUNNING -> IDLE
848	 * IDLE -> CANNOT_RUN -> IDLE
849	 */
850	if ((vcpu->state == VCPU_IDLE && state != VCPU_IDLE) ||
851	    (vcpu->state != VCPU_IDLE && state == VCPU_IDLE)) {
852		error = 0;
853		vcpu->state = state;
854	} else {
855		error = EBUSY;
856	}
857
858	vcpu_unlock(vcpu);
859
860	return (error);
861}
862
863enum vcpu_state
864vcpu_get_state(struct vm *vm, int vcpuid)
865{
866	struct vcpu *vcpu;
867	enum vcpu_state state;
868
869	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
870		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
871
872	vcpu = &vm->vcpu[vcpuid];
873
874	vcpu_lock(vcpu);
875	state = vcpu->state;
876	vcpu_unlock(vcpu);
877
878	return (state);
879}
880
881void
882vm_activate_cpu(struct vm *vm, int vcpuid)
883{
884
885	if (vcpuid >= 0 && vcpuid < VM_MAXCPU)
886		CPU_SET(vcpuid, &vm->active_cpus);
887}
888
889cpuset_t
890vm_active_cpus(struct vm *vm)
891{
892
893	return (vm->active_cpus);
894}
895
896void *
897vcpu_stats(struct vm *vm, int vcpuid)
898{
899
900	return (vm->vcpu[vcpuid].stats);
901}
902
903int
904vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
905{
906	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
907		return (EINVAL);
908
909	*state = vm->vcpu[vcpuid].x2apic_state;
910
911	return (0);
912}
913
914int
915vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
916{
917	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
918		return (EINVAL);
919
920	if (state < 0 || state >= X2APIC_STATE_LAST)
921		return (EINVAL);
922
923	vm->vcpu[vcpuid].x2apic_state = state;
924
925	vlapic_set_x2apic_state(vm, vcpuid, state);
926
927	return (0);
928}
929
930void
931vm_interrupt_hostcpu(struct vm *vm, int vcpuid)
932{
933	int hostcpu;
934	struct vcpu *vcpu;
935
936	vcpu = &vm->vcpu[vcpuid];
937
938	/*
939	 * XXX racy but the worst case is that we'll send an unnecessary IPI
940	 * to the 'hostcpu'.
941	 *
942	 * We cannot use vcpu_is_running() here because it acquires vcpu->mtx
943	 * which is not allowed inside a critical section.
944	 */
945	hostcpu = vcpu->hostcpu;
946	if (hostcpu == NOCPU || hostcpu == curcpu)
947		return;
948
949	ipi_cpu(hostcpu, vmm_ipinum);
950}
951