vmm.c revision 248389 
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: head/sys/amd64/vmm/vmm.c 248389 2013-03-16 22:40:20Z neel $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: head/sys/amd64/vmm/vmm.c 248389 2013-03-16 22:40:20Z neel $");
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		hostcpu;	/* host cpuid this vcpu last ran on */
74	uint64_t	guest_msrs[VMM_MSR_NUM];
75	struct vlapic	*vlapic;
76	int		 vcpuid;
77	struct savefpu	*guestfpu;	/* guest fpu state */
78	void		*stats;
79	struct vm_exit	exitinfo;
80	enum x2apic_state x2apic_state;
81	int		nmi_pending;
82};
83
84#define	vcpu_lock_init(v)	mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
85#define	vcpu_lock(v)		mtx_lock_spin(&((v)->mtx))
86#define	vcpu_unlock(v)		mtx_unlock_spin(&((v)->mtx))
87
88#define	VM_MAX_MEMORY_SEGMENTS	2
89
90struct vm {
91	void		*cookie;	/* processor-specific data */
92	void		*iommu;		/* iommu-specific data */
93	struct vcpu	vcpu[VM_MAXCPU];
94	int		num_mem_segs;
95	struct vm_memory_segment mem_segs[VM_MAX_MEMORY_SEGMENTS];
96	char		name[VM_MAX_NAMELEN];
97
98	/*
99	 * Set of active vcpus.
100	 * An active vcpu is one that has been started implicitly (BSP) or
101	 * explicitly (AP) by sending it a startup ipi.
102	 */
103	cpuset_t	active_cpus;
104};
105
106static struct vmm_ops *ops;
107#define	VMM_INIT()	(ops != NULL ? (*ops->init)() : 0)
108#define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
109
110#define	VMINIT(vm)	(ops != NULL ? (*ops->vminit)(vm): NULL)
111#define	VMRUN(vmi, vcpu, rip) \
112	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip) : ENXIO)
113#define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
114#define	VMMMAP_SET(vmi, gpa, hpa, len, attr, prot, spm)			\
115    	(ops != NULL ? 							\
116    	(*ops->vmmmap_set)(vmi, gpa, hpa, len, attr, prot, spm) :	\
117	ENXIO)
118#define	VMMMAP_GET(vmi, gpa) \
119	(ops != NULL ? (*ops->vmmmap_get)(vmi, gpa) : ENXIO)
120#define	VMGETREG(vmi, vcpu, num, retval)		\
121	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
122#define	VMSETREG(vmi, vcpu, num, val)		\
123	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
124#define	VMGETDESC(vmi, vcpu, num, desc)		\
125	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
126#define	VMSETDESC(vmi, vcpu, num, desc)		\
127	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
128#define	VMINJECT(vmi, vcpu, type, vec, ec, ecv)	\
129	(ops != NULL ? (*ops->vminject)(vmi, vcpu, type, vec, ec, ecv) : ENXIO)
130#define	VMGETCAP(vmi, vcpu, num, retval)	\
131	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
132#define	VMSETCAP(vmi, vcpu, num, val)		\
133	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
134
135#define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
136#define	fpu_stop_emulating()	clts()
137
138static MALLOC_DEFINE(M_VM, "vm", "vm");
139CTASSERT(VMM_MSR_NUM <= 64);	/* msr_mask can keep track of up to 64 msrs */
140
141/* statistics */
142static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
143
144static void
145vcpu_cleanup(struct vcpu *vcpu)
146{
147	vlapic_cleanup(vcpu->vlapic);
148	vmm_stat_free(vcpu->stats);
149	fpu_save_area_free(vcpu->guestfpu);
150}
151
152static void
153vcpu_init(struct vm *vm, uint32_t vcpu_id)
154{
155	struct vcpu *vcpu;
156
157	vcpu = &vm->vcpu[vcpu_id];
158
159	vcpu_lock_init(vcpu);
160	vcpu->hostcpu = NOCPU;
161	vcpu->vcpuid = vcpu_id;
162	vcpu->vlapic = vlapic_init(vm, vcpu_id);
163	vm_set_x2apic_state(vm, vcpu_id, X2APIC_ENABLED);
164	vcpu->guestfpu = fpu_save_area_alloc();
165	fpu_save_area_reset(vcpu->guestfpu);
166	vcpu->stats = vmm_stat_alloc();
167}
168
169struct vm_exit *
170vm_exitinfo(struct vm *vm, int cpuid)
171{
172	struct vcpu *vcpu;
173
174	if (cpuid < 0 || cpuid >= VM_MAXCPU)
175		panic("vm_exitinfo: invalid cpuid %d", cpuid);
176
177	vcpu = &vm->vcpu[cpuid];
178
179	return (&vcpu->exitinfo);
180}
181
182static int
183vmm_init(void)
184{
185	int error;
186
187	vmm_host_state_init();
188	vmm_ipi_init();
189
190	error = vmm_mem_init();
191	if (error)
192		return (error);
193
194	if (vmm_is_intel())
195		ops = &vmm_ops_intel;
196	else if (vmm_is_amd())
197		ops = &vmm_ops_amd;
198	else
199		return (ENXIO);
200
201	vmm_msr_init();
202
203	return (VMM_INIT());
204}
205
206static int
207vmm_handler(module_t mod, int what, void *arg)
208{
209	int error;
210
211	switch (what) {
212	case MOD_LOAD:
213		vmmdev_init();
214		iommu_init();
215		error = vmm_init();
216		break;
217	case MOD_UNLOAD:
218		error = vmmdev_cleanup();
219		if (error == 0) {
220			iommu_cleanup();
221			vmm_ipi_cleanup();
222			error = VMM_CLEANUP();
223		}
224		break;
225	default:
226		error = 0;
227		break;
228	}
229	return (error);
230}
231
232static moduledata_t vmm_kmod = {
233	"vmm",
234	vmm_handler,
235	NULL
236};
237
238/*
239 * vmm initialization has the following dependencies:
240 *
241 * - iommu initialization must happen after the pci passthru driver has had
242 *   a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
243 *
244 * - VT-x initialization requires smp_rendezvous() and therefore must happen
245 *   after SMP is fully functional (after SI_SUB_SMP).
246 */
247DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
248MODULE_VERSION(vmm, 1);
249
250SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
251
252struct vm *
253vm_create(const char *name)
254{
255	int i;
256	struct vm *vm;
257	vm_paddr_t maxaddr;
258
259	const int BSP = 0;
260
261	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
262		return (NULL);
263
264	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
265	strcpy(vm->name, name);
266	vm->cookie = VMINIT(vm);
267
268	for (i = 0; i < VM_MAXCPU; i++) {
269		vcpu_init(vm, i);
270		guest_msrs_init(vm, i);
271	}
272
273	maxaddr = vmm_mem_maxaddr();
274	vm->iommu = iommu_create_domain(maxaddr);
275	vm_activate_cpu(vm, BSP);
276
277	return (vm);
278}
279
280static void
281vm_free_mem_seg(struct vm *vm, struct vm_memory_segment *seg)
282{
283	size_t len;
284	vm_paddr_t hpa;
285	void *host_domain;
286
287	host_domain = iommu_host_domain();
288
289	len = 0;
290	while (len < seg->len) {
291		hpa = vm_gpa2hpa(vm, seg->gpa + len, PAGE_SIZE);
292		if (hpa == (vm_paddr_t)-1) {
293			panic("vm_free_mem_segs: cannot free hpa "
294			      "associated with gpa 0x%016lx", seg->gpa + len);
295		}
296
297		/*
298		 * Remove the 'gpa' to 'hpa' mapping in VMs domain.
299		 * And resurrect the 1:1 mapping for 'hpa' in 'host_domain'.
300		 */
301		iommu_remove_mapping(vm->iommu, seg->gpa + len, PAGE_SIZE);
302		iommu_create_mapping(host_domain, hpa, hpa, PAGE_SIZE);
303
304		vmm_mem_free(hpa, PAGE_SIZE);
305
306		len += PAGE_SIZE;
307	}
308
309	/*
310	 * Invalidate cached translations associated with 'vm->iommu' since
311	 * we have now moved some pages from it.
312	 */
313	iommu_invalidate_tlb(vm->iommu);
314
315	bzero(seg, sizeof(struct vm_memory_segment));
316}
317
318void
319vm_destroy(struct vm *vm)
320{
321	int i;
322
323	ppt_unassign_all(vm);
324
325	for (i = 0; i < vm->num_mem_segs; i++)
326		vm_free_mem_seg(vm, &vm->mem_segs[i]);
327
328	vm->num_mem_segs = 0;
329
330	for (i = 0; i < VM_MAXCPU; i++)
331		vcpu_cleanup(&vm->vcpu[i]);
332
333	iommu_destroy_domain(vm->iommu);
334
335	VMCLEANUP(vm->cookie);
336
337	free(vm, M_VM);
338}
339
340const char *
341vm_name(struct vm *vm)
342{
343	return (vm->name);
344}
345
346int
347vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
348{
349	const boolean_t spok = TRUE;	/* superpage mappings are ok */
350
351	return (VMMMAP_SET(vm->cookie, gpa, hpa, len, VM_MEMATTR_UNCACHEABLE,
352			   VM_PROT_RW, spok));
353}
354
355int
356vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
357{
358	const boolean_t spok = TRUE;	/* superpage mappings are ok */
359
360	return (VMMMAP_SET(vm->cookie, gpa, 0, len, 0,
361			   VM_PROT_NONE, spok));
362}
363
364/*
365 * Returns TRUE if 'gpa' is available for allocation and FALSE otherwise
366 */
367static boolean_t
368vm_gpa_available(struct vm *vm, vm_paddr_t gpa)
369{
370	int i;
371	vm_paddr_t gpabase, gpalimit;
372
373	if (gpa & PAGE_MASK)
374		panic("vm_gpa_available: gpa (0x%016lx) not page aligned", gpa);
375
376	for (i = 0; i < vm->num_mem_segs; i++) {
377		gpabase = vm->mem_segs[i].gpa;
378		gpalimit = gpabase + vm->mem_segs[i].len;
379		if (gpa >= gpabase && gpa < gpalimit)
380			return (FALSE);
381	}
382
383	return (TRUE);
384}
385
386int
387vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
388{
389	int error, available, allocated;
390	struct vm_memory_segment *seg;
391	vm_paddr_t g, hpa;
392	void *host_domain;
393
394	const boolean_t spok = TRUE;	/* superpage mappings are ok */
395
396	if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
397		return (EINVAL);
398
399	available = allocated = 0;
400	g = gpa;
401	while (g < gpa + len) {
402		if (vm_gpa_available(vm, g))
403			available++;
404		else
405			allocated++;
406
407		g += PAGE_SIZE;
408	}
409
410	/*
411	 * If there are some allocated and some available pages in the address
412	 * range then it is an error.
413	 */
414	if (allocated && available)
415		return (EINVAL);
416
417	/*
418	 * If the entire address range being requested has already been
419	 * allocated then there isn't anything more to do.
420	 */
421	if (allocated && available == 0)
422		return (0);
423
424	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
425		return (E2BIG);
426
427	host_domain = iommu_host_domain();
428
429	seg = &vm->mem_segs[vm->num_mem_segs];
430
431	error = 0;
432	seg->gpa = gpa;
433	seg->len = 0;
434	while (seg->len < len) {
435		hpa = vmm_mem_alloc(PAGE_SIZE);
436		if (hpa == 0) {
437			error = ENOMEM;
438			break;
439		}
440
441		error = VMMMAP_SET(vm->cookie, gpa + seg->len, hpa, PAGE_SIZE,
442				   VM_MEMATTR_WRITE_BACK, VM_PROT_ALL, spok);
443		if (error)
444			break;
445
446		/*
447		 * Remove the 1:1 mapping for 'hpa' from the 'host_domain'.
448		 * Add mapping for 'gpa + seg->len' to 'hpa' in the VMs domain.
449		 */
450		iommu_remove_mapping(host_domain, hpa, PAGE_SIZE);
451		iommu_create_mapping(vm->iommu, gpa + seg->len, hpa, PAGE_SIZE);
452
453		seg->len += PAGE_SIZE;
454	}
455
456	if (error) {
457		vm_free_mem_seg(vm, seg);
458		return (error);
459	}
460
461	/*
462	 * Invalidate cached translations associated with 'host_domain' since
463	 * we have now moved some pages from it.
464	 */
465	iommu_invalidate_tlb(host_domain);
466
467	vm->num_mem_segs++;
468
469	return (0);
470}
471
472vm_paddr_t
473vm_gpa2hpa(struct vm *vm, vm_paddr_t gpa, size_t len)
474{
475	vm_paddr_t nextpage;
476
477	nextpage = rounddown(gpa + PAGE_SIZE, PAGE_SIZE);
478	if (len > nextpage - gpa)
479		panic("vm_gpa2hpa: invalid gpa/len: 0x%016lx/%lu", gpa, len);
480
481	return (VMMMAP_GET(vm->cookie, gpa));
482}
483
484int
485vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
486		  struct vm_memory_segment *seg)
487{
488	int i;
489
490	for (i = 0; i < vm->num_mem_segs; i++) {
491		if (gpabase == vm->mem_segs[i].gpa) {
492			*seg = vm->mem_segs[i];
493			return (0);
494		}
495	}
496	return (-1);
497}
498
499int
500vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
501{
502
503	if (vcpu < 0 || vcpu >= VM_MAXCPU)
504		return (EINVAL);
505
506	if (reg >= VM_REG_LAST)
507		return (EINVAL);
508
509	return (VMGETREG(vm->cookie, vcpu, reg, retval));
510}
511
512int
513vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
514{
515
516	if (vcpu < 0 || vcpu >= VM_MAXCPU)
517		return (EINVAL);
518
519	if (reg >= VM_REG_LAST)
520		return (EINVAL);
521
522	return (VMSETREG(vm->cookie, vcpu, reg, val));
523}
524
525static boolean_t
526is_descriptor_table(int reg)
527{
528
529	switch (reg) {
530	case VM_REG_GUEST_IDTR:
531	case VM_REG_GUEST_GDTR:
532		return (TRUE);
533	default:
534		return (FALSE);
535	}
536}
537
538static boolean_t
539is_segment_register(int reg)
540{
541
542	switch (reg) {
543	case VM_REG_GUEST_ES:
544	case VM_REG_GUEST_CS:
545	case VM_REG_GUEST_SS:
546	case VM_REG_GUEST_DS:
547	case VM_REG_GUEST_FS:
548	case VM_REG_GUEST_GS:
549	case VM_REG_GUEST_TR:
550	case VM_REG_GUEST_LDTR:
551		return (TRUE);
552	default:
553		return (FALSE);
554	}
555}
556
557int
558vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
559		struct seg_desc *desc)
560{
561
562	if (vcpu < 0 || vcpu >= VM_MAXCPU)
563		return (EINVAL);
564
565	if (!is_segment_register(reg) && !is_descriptor_table(reg))
566		return (EINVAL);
567
568	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
569}
570
571int
572vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
573		struct seg_desc *desc)
574{
575	if (vcpu < 0 || vcpu >= VM_MAXCPU)
576		return (EINVAL);
577
578	if (!is_segment_register(reg) && !is_descriptor_table(reg))
579		return (EINVAL);
580
581	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
582}
583
584static void
585restore_guest_fpustate(struct vcpu *vcpu)
586{
587
588	/* flush host state to the pcb */
589	fpuexit(curthread);
590
591	/* restore guest FPU state */
592	fpu_stop_emulating();
593	fpurestore(vcpu->guestfpu);
594
595	/*
596	 * The FPU is now "dirty" with the guest's state so turn on emulation
597	 * to trap any access to the FPU by the host.
598	 */
599	fpu_start_emulating();
600}
601
602static void
603save_guest_fpustate(struct vcpu *vcpu)
604{
605
606	if ((rcr0() & CR0_TS) == 0)
607		panic("fpu emulation not enabled in host!");
608
609	/* save guest FPU state */
610	fpu_stop_emulating();
611	fpusave(vcpu->guestfpu);
612	fpu_start_emulating();
613}
614
615static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
616
617int
618vm_run(struct vm *vm, struct vm_run *vmrun)
619{
620	int error, vcpuid, sleepticks, t;
621	struct vcpu *vcpu;
622	struct pcb *pcb;
623	uint64_t tscval, rip;
624	struct vm_exit *vme;
625
626	vcpuid = vmrun->cpuid;
627
628	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
629		return (EINVAL);
630
631	vcpu = &vm->vcpu[vcpuid];
632	vme = &vmrun->vm_exit;
633	rip = vmrun->rip;
634restart:
635	critical_enter();
636
637	tscval = rdtsc();
638
639	pcb = PCPU_GET(curpcb);
640	set_pcb_flags(pcb, PCB_FULL_IRET);
641
642	restore_guest_msrs(vm, vcpuid);
643	restore_guest_fpustate(vcpu);
644
645	vcpu->hostcpu = curcpu;
646	error = VMRUN(vm->cookie, vcpuid, rip);
647	vcpu->hostcpu = NOCPU;
648
649	save_guest_fpustate(vcpu);
650	restore_host_msrs(vm, vcpuid);
651
652	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
653
654	/* copy the exit information */
655	bcopy(&vcpu->exitinfo, vme, sizeof(struct vm_exit));
656
657	critical_exit();
658
659	/*
660	 * Oblige the guest's desire to 'hlt' by sleeping until the vcpu
661	 * is ready to run.
662	 */
663	if (error == 0 && vme->exitcode == VM_EXITCODE_HLT) {
664		vcpu_lock(vcpu);
665
666		/*
667		 * Figure out the number of host ticks until the next apic
668		 * timer interrupt in the guest.
669		 */
670		sleepticks = lapic_timer_tick(vm, vcpuid);
671
672		/*
673		 * If the guest local apic timer is disabled then sleep for
674		 * a long time but not forever.
675		 */
676		if (sleepticks < 0)
677			sleepticks = hz;
678
679		/*
680		 * Do a final check for pending NMI or interrupts before
681		 * really putting this thread to sleep.
682		 *
683		 * These interrupts could have happened any time after we
684		 * returned from VMRUN() and before we grabbed the vcpu lock.
685		 */
686		if (!vm_nmi_pending(vm, vcpuid) &&
687		    lapic_pending_intr(vm, vcpuid) < 0) {
688			if (sleepticks <= 0)
689				panic("invalid sleepticks %d", sleepticks);
690			t = ticks;
691			msleep_spin(vcpu, &vcpu->mtx, "vmidle", sleepticks);
692			vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
693		}
694
695		vcpu_unlock(vcpu);
696
697		rip = vme->rip + vme->inst_length;
698		goto restart;
699	}
700
701	return (error);
702}
703
704int
705vm_inject_event(struct vm *vm, int vcpuid, int type,
706		int vector, uint32_t code, int code_valid)
707{
708	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
709		return (EINVAL);
710
711	if ((type > VM_EVENT_NONE && type < VM_EVENT_MAX) == 0)
712		return (EINVAL);
713
714	if (vector < 0 || vector > 255)
715		return (EINVAL);
716
717	return (VMINJECT(vm->cookie, vcpuid, type, vector, code, code_valid));
718}
719
720static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
721
722int
723vm_inject_nmi(struct vm *vm, int vcpuid)
724{
725	struct vcpu *vcpu;
726
727	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
728		return (EINVAL);
729
730	vcpu = &vm->vcpu[vcpuid];
731
732	vcpu->nmi_pending = 1;
733	vm_interrupt_hostcpu(vm, vcpuid);
734	return (0);
735}
736
737int
738vm_nmi_pending(struct vm *vm, int vcpuid)
739{
740	struct vcpu *vcpu;
741
742	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
743		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
744
745	vcpu = &vm->vcpu[vcpuid];
746
747	return (vcpu->nmi_pending);
748}
749
750void
751vm_nmi_clear(struct vm *vm, int vcpuid)
752{
753	struct vcpu *vcpu;
754
755	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
756		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
757
758	vcpu = &vm->vcpu[vcpuid];
759
760	if (vcpu->nmi_pending == 0)
761		panic("vm_nmi_clear: inconsistent nmi_pending state");
762
763	vcpu->nmi_pending = 0;
764	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
765}
766
767int
768vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
769{
770	if (vcpu < 0 || vcpu >= VM_MAXCPU)
771		return (EINVAL);
772
773	if (type < 0 || type >= VM_CAP_MAX)
774		return (EINVAL);
775
776	return (VMGETCAP(vm->cookie, vcpu, type, retval));
777}
778
779int
780vm_set_capability(struct vm *vm, int vcpu, int type, int val)
781{
782	if (vcpu < 0 || vcpu >= VM_MAXCPU)
783		return (EINVAL);
784
785	if (type < 0 || type >= VM_CAP_MAX)
786		return (EINVAL);
787
788	return (VMSETCAP(vm->cookie, vcpu, type, val));
789}
790
791uint64_t *
792vm_guest_msrs(struct vm *vm, int cpu)
793{
794	return (vm->vcpu[cpu].guest_msrs);
795}
796
797struct vlapic *
798vm_lapic(struct vm *vm, int cpu)
799{
800	return (vm->vcpu[cpu].vlapic);
801}
802
803boolean_t
804vmm_is_pptdev(int bus, int slot, int func)
805{
806	int found, i, n;
807	int b, s, f;
808	char *val, *cp, *cp2;
809
810	/*
811	 * XXX
812	 * The length of an environment variable is limited to 128 bytes which
813	 * puts an upper limit on the number of passthru devices that may be
814	 * specified using a single environment variable.
815	 *
816	 * Work around this by scanning multiple environment variable
817	 * names instead of a single one - yuck!
818	 */
819	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
820
821	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
822	found = 0;
823	for (i = 0; names[i] != NULL && !found; i++) {
824		cp = val = getenv(names[i]);
825		while (cp != NULL && *cp != '\0') {
826			if ((cp2 = strchr(cp, ' ')) != NULL)
827				*cp2 = '\0';
828
829			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
830			if (n == 3 && bus == b && slot == s && func == f) {
831				found = 1;
832				break;
833			}
834
835			if (cp2 != NULL)
836				*cp2++ = ' ';
837
838			cp = cp2;
839		}
840		freeenv(val);
841	}
842	return (found);
843}
844
845void *
846vm_iommu_domain(struct vm *vm)
847{
848
849	return (vm->iommu);
850}
851
852int
853vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state state)
854{
855	int error;
856	struct vcpu *vcpu;
857
858	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
859		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
860
861	vcpu = &vm->vcpu[vcpuid];
862
863	vcpu_lock(vcpu);
864
865	/*
866	 * The following state transitions are allowed:
867	 * IDLE -> RUNNING -> IDLE
868	 * IDLE -> CANNOT_RUN -> IDLE
869	 */
870	if ((vcpu->state == VCPU_IDLE && state != VCPU_IDLE) ||
871	    (vcpu->state != VCPU_IDLE && state == VCPU_IDLE)) {
872		error = 0;
873		vcpu->state = state;
874	} else {
875		error = EBUSY;
876	}
877
878	vcpu_unlock(vcpu);
879
880	return (error);
881}
882
883enum vcpu_state
884vcpu_get_state(struct vm *vm, int vcpuid)
885{
886	struct vcpu *vcpu;
887	enum vcpu_state state;
888
889	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
890		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
891
892	vcpu = &vm->vcpu[vcpuid];
893
894	vcpu_lock(vcpu);
895	state = vcpu->state;
896	vcpu_unlock(vcpu);
897
898	return (state);
899}
900
901void
902vm_activate_cpu(struct vm *vm, int vcpuid)
903{
904
905	if (vcpuid >= 0 && vcpuid < VM_MAXCPU)
906		CPU_SET(vcpuid, &vm->active_cpus);
907}
908
909cpuset_t
910vm_active_cpus(struct vm *vm)
911{
912
913	return (vm->active_cpus);
914}
915
916void *
917vcpu_stats(struct vm *vm, int vcpuid)
918{
919
920	return (vm->vcpu[vcpuid].stats);
921}
922
923int
924vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
925{
926	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
927		return (EINVAL);
928
929	*state = vm->vcpu[vcpuid].x2apic_state;
930
931	return (0);
932}
933
934int
935vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
936{
937	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
938		return (EINVAL);
939
940	if (state < 0 || state >= X2APIC_STATE_LAST)
941		return (EINVAL);
942
943	vm->vcpu[vcpuid].x2apic_state = state;
944
945	vlapic_set_x2apic_state(vm, vcpuid, state);
946
947	return (0);
948}
949
950void
951vm_interrupt_hostcpu(struct vm *vm, int vcpuid)
952{
953	int hostcpu;
954	struct vcpu *vcpu;
955
956	vcpu = &vm->vcpu[vcpuid];
957
958	vcpu_lock(vcpu);
959	hostcpu = vcpu->hostcpu;
960	if (hostcpu == NOCPU) {
961		/*
962		 * If the vcpu is 'RUNNING' but without a valid 'hostcpu' then
963		 * the host thread must be sleeping waiting for an event to
964		 * kick the vcpu out of 'hlt'.
965		 *
966		 * XXX this is racy because the condition exists right before
967		 * and after calling VMRUN() in vm_run(). The wakeup() is
968		 * benign in this case.
969		 */
970		if (vcpu->state == VCPU_RUNNING)
971			wakeup_one(vcpu);
972	} else {
973		if (vcpu->state != VCPU_RUNNING)
974			panic("invalid vcpu state %d", vcpu->state);
975		if (hostcpu != curcpu)
976			ipi_cpu(hostcpu, vmm_ipinum);
977	}
978	vcpu_unlock(vcpu);
979}
980