vmm.c revision 253854 
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 253854 2013-08-01 05:59:28Z grehan $
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD: head/sys/amd64/vmm/vmm.c 253854 2013-08-01 05:59:28Z grehan $");
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 int vmm_initialized;
107
108static struct vmm_ops *ops;
109#define	VMM_INIT()	(ops != NULL ? (*ops->init)() : 0)
110#define	VMM_CLEANUP()	(ops != NULL ? (*ops->cleanup)() : 0)
111
112#define	VMINIT(vm)	(ops != NULL ? (*ops->vminit)(vm): NULL)
113#define	VMRUN(vmi, vcpu, rip) \
114	(ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip) : ENXIO)
115#define	VMCLEANUP(vmi)	(ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
116#define	VMMMAP_SET(vmi, gpa, hpa, len, attr, prot, spm)			\
117    	(ops != NULL ? 							\
118    	(*ops->vmmmap_set)(vmi, gpa, hpa, len, attr, prot, spm) :	\
119	ENXIO)
120#define	VMMMAP_GET(vmi, gpa) \
121	(ops != NULL ? (*ops->vmmmap_get)(vmi, gpa) : ENXIO)
122#define	VMGETREG(vmi, vcpu, num, retval)		\
123	(ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
124#define	VMSETREG(vmi, vcpu, num, val)		\
125	(ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
126#define	VMGETDESC(vmi, vcpu, num, desc)		\
127	(ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
128#define	VMSETDESC(vmi, vcpu, num, desc)		\
129	(ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
130#define	VMINJECT(vmi, vcpu, type, vec, ec, ecv)	\
131	(ops != NULL ? (*ops->vminject)(vmi, vcpu, type, vec, ec, ecv) : ENXIO)
132#define	VMGETCAP(vmi, vcpu, num, retval)	\
133	(ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
134#define	VMSETCAP(vmi, vcpu, num, val)		\
135	(ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
136
137#define	fpu_start_emulating()	load_cr0(rcr0() | CR0_TS)
138#define	fpu_stop_emulating()	clts()
139
140static MALLOC_DEFINE(M_VM, "vm", "vm");
141CTASSERT(VMM_MSR_NUM <= 64);	/* msr_mask can keep track of up to 64 msrs */
142
143/* statistics */
144static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
145
146static void
147vcpu_cleanup(struct vcpu *vcpu)
148{
149	vlapic_cleanup(vcpu->vlapic);
150	vmm_stat_free(vcpu->stats);
151	fpu_save_area_free(vcpu->guestfpu);
152}
153
154static void
155vcpu_init(struct vm *vm, uint32_t vcpu_id)
156{
157	struct vcpu *vcpu;
158
159	vcpu = &vm->vcpu[vcpu_id];
160
161	vcpu_lock_init(vcpu);
162	vcpu->hostcpu = NOCPU;
163	vcpu->vcpuid = vcpu_id;
164	vcpu->vlapic = vlapic_init(vm, vcpu_id);
165	vm_set_x2apic_state(vm, vcpu_id, X2APIC_ENABLED);
166	vcpu->guestfpu = fpu_save_area_alloc();
167	fpu_save_area_reset(vcpu->guestfpu);
168	vcpu->stats = vmm_stat_alloc();
169}
170
171struct vm_exit *
172vm_exitinfo(struct vm *vm, int cpuid)
173{
174	struct vcpu *vcpu;
175
176	if (cpuid < 0 || cpuid >= VM_MAXCPU)
177		panic("vm_exitinfo: invalid cpuid %d", cpuid);
178
179	vcpu = &vm->vcpu[cpuid];
180
181	return (&vcpu->exitinfo);
182}
183
184static int
185vmm_init(void)
186{
187	int error;
188
189	vmm_host_state_init();
190	vmm_ipi_init();
191
192	error = vmm_mem_init();
193	if (error)
194		return (error);
195
196	if (vmm_is_intel())
197		ops = &vmm_ops_intel;
198	else if (vmm_is_amd())
199		ops = &vmm_ops_amd;
200	else
201		return (ENXIO);
202
203	vmm_msr_init();
204
205	return (VMM_INIT());
206}
207
208static int
209vmm_handler(module_t mod, int what, void *arg)
210{
211	int error;
212
213	switch (what) {
214	case MOD_LOAD:
215		vmmdev_init();
216		iommu_init();
217		error = vmm_init();
218		if (error == 0)
219			vmm_initialized = 1;
220		break;
221	case MOD_UNLOAD:
222		error = vmmdev_cleanup();
223		if (error == 0) {
224			iommu_cleanup();
225			vmm_ipi_cleanup();
226			error = VMM_CLEANUP();
227			/*
228			 * Something bad happened - prevent new
229			 * VMs from being created
230			 */
231			if (error)
232				vmm_initialized = 0;
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 * vmm initialization has the following dependencies:
250 *
251 * - iommu initialization must happen after the pci passthru driver has had
252 *   a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
253 *
254 * - VT-x initialization requires smp_rendezvous() and therefore must happen
255 *   after SMP is fully functional (after SI_SUB_SMP).
256 */
257DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
258MODULE_VERSION(vmm, 1);
259
260SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
261
262int
263vm_create(const char *name, struct vm **retvm)
264{
265	int i;
266	struct vm *vm;
267	vm_paddr_t maxaddr;
268
269	const int BSP = 0;
270
271	/*
272	 * If vmm.ko could not be successfully initialized then don't attempt
273	 * to create the virtual machine.
274	 */
275	if (!vmm_initialized)
276		return (ENXIO);
277
278	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
279		return (EINVAL);
280
281	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
282	strcpy(vm->name, name);
283	vm->cookie = VMINIT(vm);
284
285	for (i = 0; i < VM_MAXCPU; i++) {
286		vcpu_init(vm, i);
287		guest_msrs_init(vm, i);
288	}
289
290	maxaddr = vmm_mem_maxaddr();
291	vm->iommu = iommu_create_domain(maxaddr);
292	vm_activate_cpu(vm, BSP);
293
294	*retvm = vm;
295	return (0);
296}
297
298static void
299vm_free_mem_seg(struct vm *vm, struct vm_memory_segment *seg)
300{
301	size_t len;
302	vm_paddr_t hpa;
303	void *host_domain;
304
305	host_domain = iommu_host_domain();
306
307	len = 0;
308	while (len < seg->len) {
309		hpa = vm_gpa2hpa(vm, seg->gpa + len, PAGE_SIZE);
310		if (hpa == (vm_paddr_t)-1) {
311			panic("vm_free_mem_segs: cannot free hpa "
312			      "associated with gpa 0x%016lx", seg->gpa + len);
313		}
314
315		/*
316		 * Remove the 'gpa' to 'hpa' mapping in VMs domain.
317		 * And resurrect the 1:1 mapping for 'hpa' in 'host_domain'.
318		 */
319		iommu_remove_mapping(vm->iommu, seg->gpa + len, PAGE_SIZE);
320		iommu_create_mapping(host_domain, hpa, hpa, PAGE_SIZE);
321
322		vmm_mem_free(hpa, PAGE_SIZE);
323
324		len += PAGE_SIZE;
325	}
326
327	/*
328	 * Invalidate cached translations associated with 'vm->iommu' since
329	 * we have now moved some pages from it.
330	 */
331	iommu_invalidate_tlb(vm->iommu);
332
333	bzero(seg, sizeof(struct vm_memory_segment));
334}
335
336void
337vm_destroy(struct vm *vm)
338{
339	int i;
340
341	ppt_unassign_all(vm);
342
343	for (i = 0; i < vm->num_mem_segs; i++)
344		vm_free_mem_seg(vm, &vm->mem_segs[i]);
345
346	vm->num_mem_segs = 0;
347
348	for (i = 0; i < VM_MAXCPU; i++)
349		vcpu_cleanup(&vm->vcpu[i]);
350
351	iommu_destroy_domain(vm->iommu);
352
353	VMCLEANUP(vm->cookie);
354
355	free(vm, M_VM);
356}
357
358const char *
359vm_name(struct vm *vm)
360{
361	return (vm->name);
362}
363
364int
365vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
366{
367	const boolean_t spok = TRUE;	/* superpage mappings are ok */
368
369	return (VMMMAP_SET(vm->cookie, gpa, hpa, len, VM_MEMATTR_UNCACHEABLE,
370			   VM_PROT_RW, spok));
371}
372
373int
374vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
375{
376	const boolean_t spok = TRUE;	/* superpage mappings are ok */
377
378	return (VMMMAP_SET(vm->cookie, gpa, 0, len, 0,
379			   VM_PROT_NONE, spok));
380}
381
382/*
383 * Returns TRUE if 'gpa' is available for allocation and FALSE otherwise
384 */
385static boolean_t
386vm_gpa_available(struct vm *vm, vm_paddr_t gpa)
387{
388	int i;
389	vm_paddr_t gpabase, gpalimit;
390
391	if (gpa & PAGE_MASK)
392		panic("vm_gpa_available: gpa (0x%016lx) not page aligned", gpa);
393
394	for (i = 0; i < vm->num_mem_segs; i++) {
395		gpabase = vm->mem_segs[i].gpa;
396		gpalimit = gpabase + vm->mem_segs[i].len;
397		if (gpa >= gpabase && gpa < gpalimit)
398			return (FALSE);
399	}
400
401	return (TRUE);
402}
403
404int
405vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
406{
407	int error, available, allocated;
408	struct vm_memory_segment *seg;
409	vm_paddr_t g, hpa;
410	void *host_domain;
411
412	const boolean_t spok = TRUE;	/* superpage mappings are ok */
413
414	if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
415		return (EINVAL);
416
417	available = allocated = 0;
418	g = gpa;
419	while (g < gpa + len) {
420		if (vm_gpa_available(vm, g))
421			available++;
422		else
423			allocated++;
424
425		g += PAGE_SIZE;
426	}
427
428	/*
429	 * If there are some allocated and some available pages in the address
430	 * range then it is an error.
431	 */
432	if (allocated && available)
433		return (EINVAL);
434
435	/*
436	 * If the entire address range being requested has already been
437	 * allocated then there isn't anything more to do.
438	 */
439	if (allocated && available == 0)
440		return (0);
441
442	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
443		return (E2BIG);
444
445	host_domain = iommu_host_domain();
446
447	seg = &vm->mem_segs[vm->num_mem_segs];
448
449	error = 0;
450	seg->gpa = gpa;
451	seg->len = 0;
452	while (seg->len < len) {
453		hpa = vmm_mem_alloc(PAGE_SIZE);
454		if (hpa == 0) {
455			error = ENOMEM;
456			break;
457		}
458
459		error = VMMMAP_SET(vm->cookie, gpa + seg->len, hpa, PAGE_SIZE,
460				   VM_MEMATTR_WRITE_BACK, VM_PROT_ALL, spok);
461		if (error)
462			break;
463
464		/*
465		 * Remove the 1:1 mapping for 'hpa' from the 'host_domain'.
466		 * Add mapping for 'gpa + seg->len' to 'hpa' in the VMs domain.
467		 */
468		iommu_remove_mapping(host_domain, hpa, PAGE_SIZE);
469		iommu_create_mapping(vm->iommu, gpa + seg->len, hpa, PAGE_SIZE);
470
471		seg->len += PAGE_SIZE;
472	}
473
474	if (error) {
475		vm_free_mem_seg(vm, seg);
476		return (error);
477	}
478
479	/*
480	 * Invalidate cached translations associated with 'host_domain' since
481	 * we have now moved some pages from it.
482	 */
483	iommu_invalidate_tlb(host_domain);
484
485	vm->num_mem_segs++;
486
487	return (0);
488}
489
490vm_paddr_t
491vm_gpa2hpa(struct vm *vm, vm_paddr_t gpa, size_t len)
492{
493	vm_paddr_t nextpage;
494
495	nextpage = rounddown(gpa + PAGE_SIZE, PAGE_SIZE);
496	if (len > nextpage - gpa)
497		panic("vm_gpa2hpa: invalid gpa/len: 0x%016lx/%lu", gpa, len);
498
499	return (VMMMAP_GET(vm->cookie, gpa));
500}
501
502int
503vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
504		  struct vm_memory_segment *seg)
505{
506	int i;
507
508	for (i = 0; i < vm->num_mem_segs; i++) {
509		if (gpabase == vm->mem_segs[i].gpa) {
510			*seg = vm->mem_segs[i];
511			return (0);
512		}
513	}
514	return (-1);
515}
516
517int
518vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
519{
520
521	if (vcpu < 0 || vcpu >= VM_MAXCPU)
522		return (EINVAL);
523
524	if (reg >= VM_REG_LAST)
525		return (EINVAL);
526
527	return (VMGETREG(vm->cookie, vcpu, reg, retval));
528}
529
530int
531vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
532{
533
534	if (vcpu < 0 || vcpu >= VM_MAXCPU)
535		return (EINVAL);
536
537	if (reg >= VM_REG_LAST)
538		return (EINVAL);
539
540	return (VMSETREG(vm->cookie, vcpu, reg, val));
541}
542
543static boolean_t
544is_descriptor_table(int reg)
545{
546
547	switch (reg) {
548	case VM_REG_GUEST_IDTR:
549	case VM_REG_GUEST_GDTR:
550		return (TRUE);
551	default:
552		return (FALSE);
553	}
554}
555
556static boolean_t
557is_segment_register(int reg)
558{
559
560	switch (reg) {
561	case VM_REG_GUEST_ES:
562	case VM_REG_GUEST_CS:
563	case VM_REG_GUEST_SS:
564	case VM_REG_GUEST_DS:
565	case VM_REG_GUEST_FS:
566	case VM_REG_GUEST_GS:
567	case VM_REG_GUEST_TR:
568	case VM_REG_GUEST_LDTR:
569		return (TRUE);
570	default:
571		return (FALSE);
572	}
573}
574
575int
576vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
577		struct seg_desc *desc)
578{
579
580	if (vcpu < 0 || vcpu >= VM_MAXCPU)
581		return (EINVAL);
582
583	if (!is_segment_register(reg) && !is_descriptor_table(reg))
584		return (EINVAL);
585
586	return (VMGETDESC(vm->cookie, vcpu, reg, desc));
587}
588
589int
590vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
591		struct seg_desc *desc)
592{
593	if (vcpu < 0 || vcpu >= VM_MAXCPU)
594		return (EINVAL);
595
596	if (!is_segment_register(reg) && !is_descriptor_table(reg))
597		return (EINVAL);
598
599	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
600}
601
602static void
603restore_guest_fpustate(struct vcpu *vcpu)
604{
605
606	/* flush host state to the pcb */
607	fpuexit(curthread);
608
609	/* restore guest FPU state */
610	fpu_stop_emulating();
611	fpurestore(vcpu->guestfpu);
612
613	/*
614	 * The FPU is now "dirty" with the guest's state so turn on emulation
615	 * to trap any access to the FPU by the host.
616	 */
617	fpu_start_emulating();
618}
619
620static void
621save_guest_fpustate(struct vcpu *vcpu)
622{
623
624	if ((rcr0() & CR0_TS) == 0)
625		panic("fpu emulation not enabled in host!");
626
627	/* save guest FPU state */
628	fpu_stop_emulating();
629	fpusave(vcpu->guestfpu);
630	fpu_start_emulating();
631}
632
633static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
634
635int
636vm_run(struct vm *vm, struct vm_run *vmrun)
637{
638	int error, vcpuid, sleepticks, t;
639	struct vcpu *vcpu;
640	struct pcb *pcb;
641	uint64_t tscval, rip;
642	struct vm_exit *vme;
643
644	vcpuid = vmrun->cpuid;
645
646	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
647		return (EINVAL);
648
649	vcpu = &vm->vcpu[vcpuid];
650	vme = &vmrun->vm_exit;
651	rip = vmrun->rip;
652restart:
653	critical_enter();
654
655	tscval = rdtsc();
656
657	pcb = PCPU_GET(curpcb);
658	set_pcb_flags(pcb, PCB_FULL_IRET);
659
660	restore_guest_msrs(vm, vcpuid);
661	restore_guest_fpustate(vcpu);
662
663	vcpu->hostcpu = curcpu;
664	error = VMRUN(vm->cookie, vcpuid, rip);
665	vcpu->hostcpu = NOCPU;
666
667	save_guest_fpustate(vcpu);
668	restore_host_msrs(vm, vcpuid);
669
670	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
671
672	/* copy the exit information */
673	bcopy(&vcpu->exitinfo, vme, sizeof(struct vm_exit));
674
675	critical_exit();
676
677	/*
678	 * Oblige the guest's desire to 'hlt' by sleeping until the vcpu
679	 * is ready to run.
680	 */
681	if (error == 0 && vme->exitcode == VM_EXITCODE_HLT) {
682		vcpu_lock(vcpu);
683
684		/*
685		 * Figure out the number of host ticks until the next apic
686		 * timer interrupt in the guest.
687		 */
688		sleepticks = lapic_timer_tick(vm, vcpuid);
689
690		/*
691		 * If the guest local apic timer is disabled then sleep for
692		 * a long time but not forever.
693		 */
694		if (sleepticks < 0)
695			sleepticks = hz;
696
697		/*
698		 * Do a final check for pending NMI or interrupts before
699		 * really putting this thread to sleep.
700		 *
701		 * These interrupts could have happened any time after we
702		 * returned from VMRUN() and before we grabbed the vcpu lock.
703		 */
704		if (!vm_nmi_pending(vm, vcpuid) &&
705		    lapic_pending_intr(vm, vcpuid) < 0) {
706			if (sleepticks <= 0)
707				panic("invalid sleepticks %d", sleepticks);
708			t = ticks;
709			msleep_spin(vcpu, &vcpu->mtx, "vmidle", sleepticks);
710			vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
711		}
712
713		vcpu_unlock(vcpu);
714
715		rip = vme->rip + vme->inst_length;
716		goto restart;
717	}
718
719	return (error);
720}
721
722int
723vm_inject_event(struct vm *vm, int vcpuid, int type,
724		int vector, uint32_t code, int code_valid)
725{
726	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
727		return (EINVAL);
728
729	if ((type > VM_EVENT_NONE && type < VM_EVENT_MAX) == 0)
730		return (EINVAL);
731
732	if (vector < 0 || vector > 255)
733		return (EINVAL);
734
735	return (VMINJECT(vm->cookie, vcpuid, type, vector, code, code_valid));
736}
737
738static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
739
740int
741vm_inject_nmi(struct vm *vm, int vcpuid)
742{
743	struct vcpu *vcpu;
744
745	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
746		return (EINVAL);
747
748	vcpu = &vm->vcpu[vcpuid];
749
750	vcpu->nmi_pending = 1;
751	vm_interrupt_hostcpu(vm, vcpuid);
752	return (0);
753}
754
755int
756vm_nmi_pending(struct vm *vm, int vcpuid)
757{
758	struct vcpu *vcpu;
759
760	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
761		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
762
763	vcpu = &vm->vcpu[vcpuid];
764
765	return (vcpu->nmi_pending);
766}
767
768void
769vm_nmi_clear(struct vm *vm, int vcpuid)
770{
771	struct vcpu *vcpu;
772
773	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
774		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
775
776	vcpu = &vm->vcpu[vcpuid];
777
778	if (vcpu->nmi_pending == 0)
779		panic("vm_nmi_clear: inconsistent nmi_pending state");
780
781	vcpu->nmi_pending = 0;
782	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
783}
784
785int
786vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
787{
788	if (vcpu < 0 || vcpu >= VM_MAXCPU)
789		return (EINVAL);
790
791	if (type < 0 || type >= VM_CAP_MAX)
792		return (EINVAL);
793
794	return (VMGETCAP(vm->cookie, vcpu, type, retval));
795}
796
797int
798vm_set_capability(struct vm *vm, int vcpu, int type, int val)
799{
800	if (vcpu < 0 || vcpu >= VM_MAXCPU)
801		return (EINVAL);
802
803	if (type < 0 || type >= VM_CAP_MAX)
804		return (EINVAL);
805
806	return (VMSETCAP(vm->cookie, vcpu, type, val));
807}
808
809uint64_t *
810vm_guest_msrs(struct vm *vm, int cpu)
811{
812	return (vm->vcpu[cpu].guest_msrs);
813}
814
815struct vlapic *
816vm_lapic(struct vm *vm, int cpu)
817{
818	return (vm->vcpu[cpu].vlapic);
819}
820
821boolean_t
822vmm_is_pptdev(int bus, int slot, int func)
823{
824	int found, i, n;
825	int b, s, f;
826	char *val, *cp, *cp2;
827
828	/*
829	 * XXX
830	 * The length of an environment variable is limited to 128 bytes which
831	 * puts an upper limit on the number of passthru devices that may be
832	 * specified using a single environment variable.
833	 *
834	 * Work around this by scanning multiple environment variable
835	 * names instead of a single one - yuck!
836	 */
837	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
838
839	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
840	found = 0;
841	for (i = 0; names[i] != NULL && !found; i++) {
842		cp = val = getenv(names[i]);
843		while (cp != NULL && *cp != '\0') {
844			if ((cp2 = strchr(cp, ' ')) != NULL)
845				*cp2 = '\0';
846
847			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
848			if (n == 3 && bus == b && slot == s && func == f) {
849				found = 1;
850				break;
851			}
852
853			if (cp2 != NULL)
854				*cp2++ = ' ';
855
856			cp = cp2;
857		}
858		freeenv(val);
859	}
860	return (found);
861}
862
863void *
864vm_iommu_domain(struct vm *vm)
865{
866
867	return (vm->iommu);
868}
869
870int
871vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state state)
872{
873	int error;
874	struct vcpu *vcpu;
875
876	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
877		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
878
879	vcpu = &vm->vcpu[vcpuid];
880
881	vcpu_lock(vcpu);
882
883	/*
884	 * The following state transitions are allowed:
885	 * IDLE -> RUNNING -> IDLE
886	 * IDLE -> CANNOT_RUN -> IDLE
887	 */
888	if ((vcpu->state == VCPU_IDLE && state != VCPU_IDLE) ||
889	    (vcpu->state != VCPU_IDLE && state == VCPU_IDLE)) {
890		error = 0;
891		vcpu->state = state;
892	} else {
893		error = EBUSY;
894	}
895
896	vcpu_unlock(vcpu);
897
898	return (error);
899}
900
901enum vcpu_state
902vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
903{
904	struct vcpu *vcpu;
905	enum vcpu_state state;
906
907	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
908		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
909
910	vcpu = &vm->vcpu[vcpuid];
911
912	vcpu_lock(vcpu);
913	state = vcpu->state;
914	if (hostcpu != NULL)
915		*hostcpu = vcpu->hostcpu;
916	vcpu_unlock(vcpu);
917
918	return (state);
919}
920
921void
922vm_activate_cpu(struct vm *vm, int vcpuid)
923{
924
925	if (vcpuid >= 0 && vcpuid < VM_MAXCPU)
926		CPU_SET(vcpuid, &vm->active_cpus);
927}
928
929cpuset_t
930vm_active_cpus(struct vm *vm)
931{
932
933	return (vm->active_cpus);
934}
935
936void *
937vcpu_stats(struct vm *vm, int vcpuid)
938{
939
940	return (vm->vcpu[vcpuid].stats);
941}
942
943int
944vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
945{
946	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
947		return (EINVAL);
948
949	*state = vm->vcpu[vcpuid].x2apic_state;
950
951	return (0);
952}
953
954int
955vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
956{
957	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
958		return (EINVAL);
959
960	if (state >= X2APIC_STATE_LAST)
961		return (EINVAL);
962
963	vm->vcpu[vcpuid].x2apic_state = state;
964
965	vlapic_set_x2apic_state(vm, vcpuid, state);
966
967	return (0);
968}
969
970void
971vm_interrupt_hostcpu(struct vm *vm, int vcpuid)
972{
973	int hostcpu;
974	struct vcpu *vcpu;
975
976	vcpu = &vm->vcpu[vcpuid];
977
978	vcpu_lock(vcpu);
979	hostcpu = vcpu->hostcpu;
980	if (hostcpu == NOCPU) {
981		/*
982		 * If the vcpu is 'RUNNING' but without a valid 'hostcpu' then
983		 * the host thread must be sleeping waiting for an event to
984		 * kick the vcpu out of 'hlt'.
985		 *
986		 * XXX this is racy because the condition exists right before
987		 * and after calling VMRUN() in vm_run(). The wakeup() is
988		 * benign in this case.
989		 */
990		if (vcpu->state == VCPU_RUNNING)
991			wakeup_one(vcpu);
992	} else {
993		if (vcpu->state != VCPU_RUNNING)
994			panic("invalid vcpu state %d", vcpu->state);
995		if (hostcpu != curcpu)
996			ipi_cpu(hostcpu, vmm_ipinum);
997	}
998	vcpu_unlock(vcpu);
999}
1000