xref: /freebsd/sys/amd64/vmm/vmm.c (revision 2a3e3873a1e4cd958f2b0f85d3b10cfa40575d30)
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 
67 struct vlapic;
68 
69 struct 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 
90 struct 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 
106 static int vmm_initialized;
107 
108 static 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 
140 static MALLOC_DEFINE(M_VM, "vm", "vm");
141 CTASSERT(VMM_MSR_NUM <= 64);	/* msr_mask can keep track of up to 64 msrs */
142 
143 /* statistics */
144 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
145 
146 static void
147 vcpu_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 
154 static void
155 vcpu_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 
171 struct vm_exit *
172 vm_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 
184 static int
185 vmm_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 
208 static int
209 vmm_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 		vmm_initialized = 0;
229 		break;
230 	default:
231 		error = 0;
232 		break;
233 	}
234 	return (error);
235 }
236 
237 static moduledata_t vmm_kmod = {
238 	"vmm",
239 	vmm_handler,
240 	NULL
241 };
242 
243 /*
244  * vmm initialization has the following dependencies:
245  *
246  * - iommu initialization must happen after the pci passthru driver has had
247  *   a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
248  *
249  * - VT-x initialization requires smp_rendezvous() and therefore must happen
250  *   after SMP is fully functional (after SI_SUB_SMP).
251  */
252 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
253 MODULE_VERSION(vmm, 1);
254 
255 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
256 
257 int
258 vm_create(const char *name, struct vm **retvm)
259 {
260 	int i;
261 	struct vm *vm;
262 	vm_paddr_t maxaddr;
263 
264 	const int BSP = 0;
265 
266 	/*
267 	 * If vmm.ko could not be successfully initialized then don't attempt
268 	 * to create the virtual machine.
269 	 */
270 	if (!vmm_initialized)
271 		return (ENXIO);
272 
273 	if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
274 		return (EINVAL);
275 
276 	vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
277 	strcpy(vm->name, name);
278 	vm->cookie = VMINIT(vm);
279 
280 	for (i = 0; i < VM_MAXCPU; i++) {
281 		vcpu_init(vm, i);
282 		guest_msrs_init(vm, i);
283 	}
284 
285 	maxaddr = vmm_mem_maxaddr();
286 	vm->iommu = iommu_create_domain(maxaddr);
287 	vm_activate_cpu(vm, BSP);
288 
289 	*retvm = vm;
290 	return (0);
291 }
292 
293 static void
294 vm_free_mem_seg(struct vm *vm, struct vm_memory_segment *seg)
295 {
296 	size_t len;
297 	vm_paddr_t hpa;
298 	void *host_domain;
299 
300 	host_domain = iommu_host_domain();
301 
302 	len = 0;
303 	while (len < seg->len) {
304 		hpa = vm_gpa2hpa(vm, seg->gpa + len, PAGE_SIZE);
305 		if (hpa == (vm_paddr_t)-1) {
306 			panic("vm_free_mem_segs: cannot free hpa "
307 			      "associated with gpa 0x%016lx", seg->gpa + len);
308 		}
309 
310 		/*
311 		 * Remove the 'gpa' to 'hpa' mapping in VMs domain.
312 		 * And resurrect the 1:1 mapping for 'hpa' in 'host_domain'.
313 		 */
314 		iommu_remove_mapping(vm->iommu, seg->gpa + len, PAGE_SIZE);
315 		iommu_create_mapping(host_domain, hpa, hpa, PAGE_SIZE);
316 
317 		vmm_mem_free(hpa, PAGE_SIZE);
318 
319 		len += PAGE_SIZE;
320 	}
321 
322 	/*
323 	 * Invalidate cached translations associated with 'vm->iommu' since
324 	 * we have now moved some pages from it.
325 	 */
326 	iommu_invalidate_tlb(vm->iommu);
327 
328 	bzero(seg, sizeof(struct vm_memory_segment));
329 }
330 
331 void
332 vm_destroy(struct vm *vm)
333 {
334 	int i;
335 
336 	ppt_unassign_all(vm);
337 
338 	for (i = 0; i < vm->num_mem_segs; i++)
339 		vm_free_mem_seg(vm, &vm->mem_segs[i]);
340 
341 	vm->num_mem_segs = 0;
342 
343 	for (i = 0; i < VM_MAXCPU; i++)
344 		vcpu_cleanup(&vm->vcpu[i]);
345 
346 	iommu_destroy_domain(vm->iommu);
347 
348 	VMCLEANUP(vm->cookie);
349 
350 	free(vm, M_VM);
351 }
352 
353 const char *
354 vm_name(struct vm *vm)
355 {
356 	return (vm->name);
357 }
358 
359 int
360 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
361 {
362 	const boolean_t spok = TRUE;	/* superpage mappings are ok */
363 
364 	return (VMMMAP_SET(vm->cookie, gpa, hpa, len, VM_MEMATTR_UNCACHEABLE,
365 			   VM_PROT_RW, spok));
366 }
367 
368 int
369 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
370 {
371 	const boolean_t spok = TRUE;	/* superpage mappings are ok */
372 
373 	return (VMMMAP_SET(vm->cookie, gpa, 0, len, 0,
374 			   VM_PROT_NONE, spok));
375 }
376 
377 /*
378  * Returns TRUE if 'gpa' is available for allocation and FALSE otherwise
379  */
380 static boolean_t
381 vm_gpa_available(struct vm *vm, vm_paddr_t gpa)
382 {
383 	int i;
384 	vm_paddr_t gpabase, gpalimit;
385 
386 	if (gpa & PAGE_MASK)
387 		panic("vm_gpa_available: gpa (0x%016lx) not page aligned", gpa);
388 
389 	for (i = 0; i < vm->num_mem_segs; i++) {
390 		gpabase = vm->mem_segs[i].gpa;
391 		gpalimit = gpabase + vm->mem_segs[i].len;
392 		if (gpa >= gpabase && gpa < gpalimit)
393 			return (FALSE);
394 	}
395 
396 	return (TRUE);
397 }
398 
399 int
400 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
401 {
402 	int error, available, allocated;
403 	struct vm_memory_segment *seg;
404 	vm_paddr_t g, hpa;
405 	void *host_domain;
406 
407 	const boolean_t spok = TRUE;	/* superpage mappings are ok */
408 
409 	if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
410 		return (EINVAL);
411 
412 	available = allocated = 0;
413 	g = gpa;
414 	while (g < gpa + len) {
415 		if (vm_gpa_available(vm, g))
416 			available++;
417 		else
418 			allocated++;
419 
420 		g += PAGE_SIZE;
421 	}
422 
423 	/*
424 	 * If there are some allocated and some available pages in the address
425 	 * range then it is an error.
426 	 */
427 	if (allocated && available)
428 		return (EINVAL);
429 
430 	/*
431 	 * If the entire address range being requested has already been
432 	 * allocated then there isn't anything more to do.
433 	 */
434 	if (allocated && available == 0)
435 		return (0);
436 
437 	if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
438 		return (E2BIG);
439 
440 	host_domain = iommu_host_domain();
441 
442 	seg = &vm->mem_segs[vm->num_mem_segs];
443 
444 	error = 0;
445 	seg->gpa = gpa;
446 	seg->len = 0;
447 	while (seg->len < len) {
448 		hpa = vmm_mem_alloc(PAGE_SIZE);
449 		if (hpa == 0) {
450 			error = ENOMEM;
451 			break;
452 		}
453 
454 		error = VMMMAP_SET(vm->cookie, gpa + seg->len, hpa, PAGE_SIZE,
455 				   VM_MEMATTR_WRITE_BACK, VM_PROT_ALL, spok);
456 		if (error)
457 			break;
458 
459 		/*
460 		 * Remove the 1:1 mapping for 'hpa' from the 'host_domain'.
461 		 * Add mapping for 'gpa + seg->len' to 'hpa' in the VMs domain.
462 		 */
463 		iommu_remove_mapping(host_domain, hpa, PAGE_SIZE);
464 		iommu_create_mapping(vm->iommu, gpa + seg->len, hpa, PAGE_SIZE);
465 
466 		seg->len += PAGE_SIZE;
467 	}
468 
469 	if (error) {
470 		vm_free_mem_seg(vm, seg);
471 		return (error);
472 	}
473 
474 	/*
475 	 * Invalidate cached translations associated with 'host_domain' since
476 	 * we have now moved some pages from it.
477 	 */
478 	iommu_invalidate_tlb(host_domain);
479 
480 	vm->num_mem_segs++;
481 
482 	return (0);
483 }
484 
485 vm_paddr_t
486 vm_gpa2hpa(struct vm *vm, vm_paddr_t gpa, size_t len)
487 {
488 	vm_paddr_t nextpage;
489 
490 	nextpage = rounddown(gpa + PAGE_SIZE, PAGE_SIZE);
491 	if (len > nextpage - gpa)
492 		panic("vm_gpa2hpa: invalid gpa/len: 0x%016lx/%lu", gpa, len);
493 
494 	return (VMMMAP_GET(vm->cookie, gpa));
495 }
496 
497 int
498 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
499 		  struct vm_memory_segment *seg)
500 {
501 	int i;
502 
503 	for (i = 0; i < vm->num_mem_segs; i++) {
504 		if (gpabase == vm->mem_segs[i].gpa) {
505 			*seg = vm->mem_segs[i];
506 			return (0);
507 		}
508 	}
509 	return (-1);
510 }
511 
512 int
513 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
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 (VMGETREG(vm->cookie, vcpu, reg, retval));
523 }
524 
525 int
526 vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
527 {
528 
529 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
530 		return (EINVAL);
531 
532 	if (reg >= VM_REG_LAST)
533 		return (EINVAL);
534 
535 	return (VMSETREG(vm->cookie, vcpu, reg, val));
536 }
537 
538 static boolean_t
539 is_descriptor_table(int reg)
540 {
541 
542 	switch (reg) {
543 	case VM_REG_GUEST_IDTR:
544 	case VM_REG_GUEST_GDTR:
545 		return (TRUE);
546 	default:
547 		return (FALSE);
548 	}
549 }
550 
551 static boolean_t
552 is_segment_register(int reg)
553 {
554 
555 	switch (reg) {
556 	case VM_REG_GUEST_ES:
557 	case VM_REG_GUEST_CS:
558 	case VM_REG_GUEST_SS:
559 	case VM_REG_GUEST_DS:
560 	case VM_REG_GUEST_FS:
561 	case VM_REG_GUEST_GS:
562 	case VM_REG_GUEST_TR:
563 	case VM_REG_GUEST_LDTR:
564 		return (TRUE);
565 	default:
566 		return (FALSE);
567 	}
568 }
569 
570 int
571 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
572 		struct seg_desc *desc)
573 {
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 (VMGETDESC(vm->cookie, vcpu, reg, desc));
582 }
583 
584 int
585 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
586 		struct seg_desc *desc)
587 {
588 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
589 		return (EINVAL);
590 
591 	if (!is_segment_register(reg) && !is_descriptor_table(reg))
592 		return (EINVAL);
593 
594 	return (VMSETDESC(vm->cookie, vcpu, reg, desc));
595 }
596 
597 static void
598 restore_guest_fpustate(struct vcpu *vcpu)
599 {
600 
601 	/* flush host state to the pcb */
602 	fpuexit(curthread);
603 
604 	/* restore guest FPU state */
605 	fpu_stop_emulating();
606 	fpurestore(vcpu->guestfpu);
607 
608 	/*
609 	 * The FPU is now "dirty" with the guest's state so turn on emulation
610 	 * to trap any access to the FPU by the host.
611 	 */
612 	fpu_start_emulating();
613 }
614 
615 static void
616 save_guest_fpustate(struct vcpu *vcpu)
617 {
618 
619 	if ((rcr0() & CR0_TS) == 0)
620 		panic("fpu emulation not enabled in host!");
621 
622 	/* save guest FPU state */
623 	fpu_stop_emulating();
624 	fpusave(vcpu->guestfpu);
625 	fpu_start_emulating();
626 }
627 
628 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
629 
630 int
631 vm_run(struct vm *vm, struct vm_run *vmrun)
632 {
633 	int error, vcpuid, sleepticks, t;
634 	struct vcpu *vcpu;
635 	struct pcb *pcb;
636 	uint64_t tscval, rip;
637 	struct vm_exit *vme;
638 
639 	vcpuid = vmrun->cpuid;
640 
641 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
642 		return (EINVAL);
643 
644 	vcpu = &vm->vcpu[vcpuid];
645 	vme = &vmrun->vm_exit;
646 	rip = vmrun->rip;
647 restart:
648 	critical_enter();
649 
650 	tscval = rdtsc();
651 
652 	pcb = PCPU_GET(curpcb);
653 	set_pcb_flags(pcb, PCB_FULL_IRET);
654 
655 	restore_guest_msrs(vm, vcpuid);
656 	restore_guest_fpustate(vcpu);
657 
658 	vcpu->hostcpu = curcpu;
659 	error = VMRUN(vm->cookie, vcpuid, rip);
660 	vcpu->hostcpu = NOCPU;
661 
662 	save_guest_fpustate(vcpu);
663 	restore_host_msrs(vm, vcpuid);
664 
665 	vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
666 
667 	/* copy the exit information */
668 	bcopy(&vcpu->exitinfo, vme, sizeof(struct vm_exit));
669 
670 	critical_exit();
671 
672 	/*
673 	 * Oblige the guest's desire to 'hlt' by sleeping until the vcpu
674 	 * is ready to run.
675 	 */
676 	if (error == 0 && vme->exitcode == VM_EXITCODE_HLT) {
677 		vcpu_lock(vcpu);
678 
679 		/*
680 		 * Figure out the number of host ticks until the next apic
681 		 * timer interrupt in the guest.
682 		 */
683 		sleepticks = lapic_timer_tick(vm, vcpuid);
684 
685 		/*
686 		 * If the guest local apic timer is disabled then sleep for
687 		 * a long time but not forever.
688 		 */
689 		if (sleepticks < 0)
690 			sleepticks = hz;
691 
692 		/*
693 		 * Do a final check for pending NMI or interrupts before
694 		 * really putting this thread to sleep.
695 		 *
696 		 * These interrupts could have happened any time after we
697 		 * returned from VMRUN() and before we grabbed the vcpu lock.
698 		 */
699 		if (!vm_nmi_pending(vm, vcpuid) &&
700 		    lapic_pending_intr(vm, vcpuid) < 0) {
701 			if (sleepticks <= 0)
702 				panic("invalid sleepticks %d", sleepticks);
703 			t = ticks;
704 			msleep_spin(vcpu, &vcpu->mtx, "vmidle", sleepticks);
705 			vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
706 		}
707 
708 		vcpu_unlock(vcpu);
709 
710 		rip = vme->rip + vme->inst_length;
711 		goto restart;
712 	}
713 
714 	return (error);
715 }
716 
717 int
718 vm_inject_event(struct vm *vm, int vcpuid, int type,
719 		int vector, uint32_t code, int code_valid)
720 {
721 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
722 		return (EINVAL);
723 
724 	if ((type > VM_EVENT_NONE && type < VM_EVENT_MAX) == 0)
725 		return (EINVAL);
726 
727 	if (vector < 0 || vector > 255)
728 		return (EINVAL);
729 
730 	return (VMINJECT(vm->cookie, vcpuid, type, vector, code, code_valid));
731 }
732 
733 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
734 
735 int
736 vm_inject_nmi(struct vm *vm, int vcpuid)
737 {
738 	struct vcpu *vcpu;
739 
740 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
741 		return (EINVAL);
742 
743 	vcpu = &vm->vcpu[vcpuid];
744 
745 	vcpu->nmi_pending = 1;
746 	vm_interrupt_hostcpu(vm, vcpuid);
747 	return (0);
748 }
749 
750 int
751 vm_nmi_pending(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 	return (vcpu->nmi_pending);
761 }
762 
763 void
764 vm_nmi_clear(struct vm *vm, int vcpuid)
765 {
766 	struct vcpu *vcpu;
767 
768 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
769 		panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
770 
771 	vcpu = &vm->vcpu[vcpuid];
772 
773 	if (vcpu->nmi_pending == 0)
774 		panic("vm_nmi_clear: inconsistent nmi_pending state");
775 
776 	vcpu->nmi_pending = 0;
777 	vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
778 }
779 
780 int
781 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
782 {
783 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
784 		return (EINVAL);
785 
786 	if (type < 0 || type >= VM_CAP_MAX)
787 		return (EINVAL);
788 
789 	return (VMGETCAP(vm->cookie, vcpu, type, retval));
790 }
791 
792 int
793 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
794 {
795 	if (vcpu < 0 || vcpu >= VM_MAXCPU)
796 		return (EINVAL);
797 
798 	if (type < 0 || type >= VM_CAP_MAX)
799 		return (EINVAL);
800 
801 	return (VMSETCAP(vm->cookie, vcpu, type, val));
802 }
803 
804 uint64_t *
805 vm_guest_msrs(struct vm *vm, int cpu)
806 {
807 	return (vm->vcpu[cpu].guest_msrs);
808 }
809 
810 struct vlapic *
811 vm_lapic(struct vm *vm, int cpu)
812 {
813 	return (vm->vcpu[cpu].vlapic);
814 }
815 
816 boolean_t
817 vmm_is_pptdev(int bus, int slot, int func)
818 {
819 	int found, i, n;
820 	int b, s, f;
821 	char *val, *cp, *cp2;
822 
823 	/*
824 	 * XXX
825 	 * The length of an environment variable is limited to 128 bytes which
826 	 * puts an upper limit on the number of passthru devices that may be
827 	 * specified using a single environment variable.
828 	 *
829 	 * Work around this by scanning multiple environment variable
830 	 * names instead of a single one - yuck!
831 	 */
832 	const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
833 
834 	/* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
835 	found = 0;
836 	for (i = 0; names[i] != NULL && !found; i++) {
837 		cp = val = getenv(names[i]);
838 		while (cp != NULL && *cp != '\0') {
839 			if ((cp2 = strchr(cp, ' ')) != NULL)
840 				*cp2 = '\0';
841 
842 			n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
843 			if (n == 3 && bus == b && slot == s && func == f) {
844 				found = 1;
845 				break;
846 			}
847 
848 			if (cp2 != NULL)
849 				*cp2++ = ' ';
850 
851 			cp = cp2;
852 		}
853 		freeenv(val);
854 	}
855 	return (found);
856 }
857 
858 void *
859 vm_iommu_domain(struct vm *vm)
860 {
861 
862 	return (vm->iommu);
863 }
864 
865 int
866 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state state)
867 {
868 	int error;
869 	struct vcpu *vcpu;
870 
871 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
872 		panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
873 
874 	vcpu = &vm->vcpu[vcpuid];
875 
876 	vcpu_lock(vcpu);
877 
878 	/*
879 	 * The following state transitions are allowed:
880 	 * IDLE -> RUNNING -> IDLE
881 	 * IDLE -> CANNOT_RUN -> IDLE
882 	 */
883 	if ((vcpu->state == VCPU_IDLE && state != VCPU_IDLE) ||
884 	    (vcpu->state != VCPU_IDLE && state == VCPU_IDLE)) {
885 		error = 0;
886 		vcpu->state = state;
887 	} else {
888 		error = EBUSY;
889 	}
890 
891 	vcpu_unlock(vcpu);
892 
893 	return (error);
894 }
895 
896 enum vcpu_state
897 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
898 {
899 	struct vcpu *vcpu;
900 	enum vcpu_state state;
901 
902 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
903 		panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
904 
905 	vcpu = &vm->vcpu[vcpuid];
906 
907 	vcpu_lock(vcpu);
908 	state = vcpu->state;
909 	if (hostcpu != NULL)
910 		*hostcpu = vcpu->hostcpu;
911 	vcpu_unlock(vcpu);
912 
913 	return (state);
914 }
915 
916 void
917 vm_activate_cpu(struct vm *vm, int vcpuid)
918 {
919 
920 	if (vcpuid >= 0 && vcpuid < VM_MAXCPU)
921 		CPU_SET(vcpuid, &vm->active_cpus);
922 }
923 
924 cpuset_t
925 vm_active_cpus(struct vm *vm)
926 {
927 
928 	return (vm->active_cpus);
929 }
930 
931 void *
932 vcpu_stats(struct vm *vm, int vcpuid)
933 {
934 
935 	return (vm->vcpu[vcpuid].stats);
936 }
937 
938 int
939 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
940 {
941 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
942 		return (EINVAL);
943 
944 	*state = vm->vcpu[vcpuid].x2apic_state;
945 
946 	return (0);
947 }
948 
949 int
950 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
951 {
952 	if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
953 		return (EINVAL);
954 
955 	if (state >= X2APIC_STATE_LAST)
956 		return (EINVAL);
957 
958 	vm->vcpu[vcpuid].x2apic_state = state;
959 
960 	vlapic_set_x2apic_state(vm, vcpuid, state);
961 
962 	return (0);
963 }
964 
965 void
966 vm_interrupt_hostcpu(struct vm *vm, int vcpuid)
967 {
968 	int hostcpu;
969 	struct vcpu *vcpu;
970 
971 	vcpu = &vm->vcpu[vcpuid];
972 
973 	vcpu_lock(vcpu);
974 	hostcpu = vcpu->hostcpu;
975 	if (hostcpu == NOCPU) {
976 		/*
977 		 * If the vcpu is 'RUNNING' but without a valid 'hostcpu' then
978 		 * the host thread must be sleeping waiting for an event to
979 		 * kick the vcpu out of 'hlt'.
980 		 *
981 		 * XXX this is racy because the condition exists right before
982 		 * and after calling VMRUN() in vm_run(). The wakeup() is
983 		 * benign in this case.
984 		 */
985 		if (vcpu->state == VCPU_RUNNING)
986 			wakeup_one(vcpu);
987 	} else {
988 		if (vcpu->state != VCPU_RUNNING)
989 			panic("invalid vcpu state %d", vcpu->state);
990 		if (hostcpu != curcpu)
991 			ipi_cpu(hostcpu, vmm_ipinum);
992 	}
993 	vcpu_unlock(vcpu);
994 }
995