FreeBSD/Linux Kernel Cross Reference
sys/amd64/vmm/vmm.c
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: releng/10.1/sys/amd64/vmm/vmm.c 270159 2014-08-19 01:20:24Z grehan $
27 */
28
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD: releng/10.1/sys/amd64/vmm/vmm.c 270159 2014-08-19 01:20:24Z 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/rwlock.h>
43 #include <sys/sched.h>
44 #include <sys/smp.h>
45 #include <sys/systm.h>
46
47 #include <vm/vm.h>
48 #include <vm/vm_object.h>
49 #include <vm/vm_page.h>
50 #include <vm/pmap.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_extern.h>
53 #include <vm/vm_param.h>
54
55 #include <machine/cpu.h>
56 #include <machine/vm.h>
57 #include <machine/pcb.h>
58 #include <machine/smp.h>
59 #include <x86/psl.h>
60 #include <x86/apicreg.h>
61 #include <machine/vmparam.h>
62
63 #include <machine/vmm.h>
64 #include <machine/vmm_dev.h>
65 #include <machine/vmm_instruction_emul.h>
66
67 #include "vmm_ioport.h"
68 #include "vmm_ktr.h"
69 #include "vmm_host.h"
70 #include "vmm_mem.h"
71 #include "vmm_util.h"
72 #include "vatpic.h"
73 #include "vatpit.h"
74 #include "vhpet.h"
75 #include "vioapic.h"
76 #include "vlapic.h"
77 #include "vmm_msr.h"
78 #include "vmm_ipi.h"
79 #include "vmm_stat.h"
80 #include "vmm_lapic.h"
81
82 #include "io/ppt.h"
83 #include "io/iommu.h"
84
85 struct vlapic;
86
87 /*
88 * Initialization:
89 * (a) allocated when vcpu is created
90 * (i) initialized when vcpu is created and when it is reinitialized
91 * (o) initialized the first time the vcpu is created
92 * (x) initialized before use
93 */
94 struct vcpu {
95 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */
96 enum vcpu_state state; /* (o) vcpu state */
97 int hostcpu; /* (o) vcpu's host cpu */
98 struct vlapic *vlapic; /* (i) APIC device model */
99 enum x2apic_state x2apic_state; /* (i) APIC mode */
100 uint64_t exitintinfo; /* (i) events pending at VM exit */
101 int nmi_pending; /* (i) NMI pending */
102 int extint_pending; /* (i) INTR pending */
103 struct vm_exception exception; /* (x) exception collateral */
104 int exception_pending; /* (i) exception pending */
105 struct savefpu *guestfpu; /* (a,i) guest fpu state */
106 uint64_t guest_xcr0; /* (i) guest %xcr0 register */
107 void *stats; /* (a,i) statistics */
108 uint64_t guest_msrs[VMM_MSR_NUM]; /* (i) emulated MSRs */
109 struct vm_exit exitinfo; /* (x) exit reason and collateral */
110 };
111
112 #define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
113 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
114 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx))
115 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx))
116 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED)
117
118 struct mem_seg {
119 vm_paddr_t gpa;
120 size_t len;
121 boolean_t wired;
122 vm_object_t object;
123 };
124 #define VM_MAX_MEMORY_SEGMENTS 2
125
126 /*
127 * Initialization:
128 * (o) initialized the first time the VM is created
129 * (i) initialized when VM is created and when it is reinitialized
130 * (x) initialized before use
131 */
132 struct vm {
133 void *cookie; /* (i) cpu-specific data */
134 void *iommu; /* (x) iommu-specific data */
135 struct vhpet *vhpet; /* (i) virtual HPET */
136 struct vioapic *vioapic; /* (i) virtual ioapic */
137 struct vatpic *vatpic; /* (i) virtual atpic */
138 struct vatpit *vatpit; /* (i) virtual atpit */
139 volatile cpuset_t active_cpus; /* (i) active vcpus */
140 int suspend; /* (i) stop VM execution */
141 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */
142 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */
143 cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */
144 cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */
145 void *rendezvous_arg; /* (x) rendezvous func/arg */
146 vm_rendezvous_func_t rendezvous_func;
147 struct mtx rendezvous_mtx; /* (o) rendezvous lock */
148 int num_mem_segs; /* (o) guest memory segments */
149 struct mem_seg mem_segs[VM_MAX_MEMORY_SEGMENTS];
150 struct vmspace *vmspace; /* (o) guest's address space */
151 char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */
152 struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */
153 };
154
155 static int vmm_initialized;
156
157 static struct vmm_ops *ops;
158 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0)
159 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0)
160 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0)
161
162 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
163 #define VMRUN(vmi, vcpu, rip, pmap, rptr, sptr) \
164 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, rptr, sptr) : ENXIO)
165 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
166 #define VMSPACE_ALLOC(min, max) \
167 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
168 #define VMSPACE_FREE(vmspace) \
169 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
170 #define VMGETREG(vmi, vcpu, num, retval) \
171 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
172 #define VMSETREG(vmi, vcpu, num, val) \
173 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
174 #define VMGETDESC(vmi, vcpu, num, desc) \
175 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
176 #define VMSETDESC(vmi, vcpu, num, desc) \
177 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
178 #define VMGETCAP(vmi, vcpu, num, retval) \
179 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
180 #define VMSETCAP(vmi, vcpu, num, val) \
181 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
182 #define VLAPIC_INIT(vmi, vcpu) \
183 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
184 #define VLAPIC_CLEANUP(vmi, vlapic) \
185 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
186
187 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS)
188 #define fpu_stop_emulating() clts()
189
190 static MALLOC_DEFINE(M_VM, "vm", "vm");
191 CTASSERT(VMM_MSR_NUM <= 64); /* msr_mask can keep track of up to 64 msrs */
192
193 /* statistics */
194 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
195
196 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
197
198 /*
199 * Halt the guest if all vcpus are executing a HLT instruction with
200 * interrupts disabled.
201 */
202 static int halt_detection_enabled = 1;
203 TUNABLE_INT("hw.vmm.halt_detection", &halt_detection_enabled);
204 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
205 &halt_detection_enabled, 0,
206 "Halt VM if all vcpus execute HLT with interrupts disabled");
207
208 static int vmm_ipinum;
209 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
210 "IPI vector used for vcpu notifications");
211
212 static void
213 vcpu_cleanup(struct vm *vm, int i, bool destroy)
214 {
215 struct vcpu *vcpu = &vm->vcpu[i];
216
217 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
218 if (destroy) {
219 vmm_stat_free(vcpu->stats);
220 fpu_save_area_free(vcpu->guestfpu);
221 }
222 }
223
224 static void
225 vcpu_init(struct vm *vm, int vcpu_id, bool create)
226 {
227 struct vcpu *vcpu;
228
229 KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU,
230 ("vcpu_init: invalid vcpu %d", vcpu_id));
231
232 vcpu = &vm->vcpu[vcpu_id];
233
234 if (create) {
235 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
236 "initialized", vcpu_id));
237 vcpu_lock_init(vcpu);
238 vcpu->state = VCPU_IDLE;
239 vcpu->hostcpu = NOCPU;
240 vcpu->guestfpu = fpu_save_area_alloc();
241 vcpu->stats = vmm_stat_alloc();
242 }
243
244 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
245 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
246 vcpu->exitintinfo = 0;
247 vcpu->nmi_pending = 0;
248 vcpu->extint_pending = 0;
249 vcpu->exception_pending = 0;
250 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
251 fpu_save_area_reset(vcpu->guestfpu);
252 vmm_stat_init(vcpu->stats);
253 guest_msrs_init(vm, vcpu_id);
254 }
255
256 struct vm_exit *
257 vm_exitinfo(struct vm *vm, int cpuid)
258 {
259 struct vcpu *vcpu;
260
261 if (cpuid < 0 || cpuid >= VM_MAXCPU)
262 panic("vm_exitinfo: invalid cpuid %d", cpuid);
263
264 vcpu = &vm->vcpu[cpuid];
265
266 return (&vcpu->exitinfo);
267 }
268
269 static void
270 vmm_resume(void)
271 {
272 VMM_RESUME();
273 }
274
275 static int
276 vmm_init(void)
277 {
278 int error;
279
280 vmm_host_state_init();
281
282 vmm_ipinum = vmm_ipi_alloc();
283 if (vmm_ipinum == 0)
284 vmm_ipinum = IPI_AST;
285
286 error = vmm_mem_init();
287 if (error)
288 return (error);
289
290 if (vmm_is_intel())
291 ops = &vmm_ops_intel;
292 else if (vmm_is_amd())
293 ops = &vmm_ops_amd;
294 else
295 return (ENXIO);
296
297 vmm_msr_init();
298 vmm_resume_p = vmm_resume;
299
300 return (VMM_INIT(vmm_ipinum));
301 }
302
303 static int
304 vmm_handler(module_t mod, int what, void *arg)
305 {
306 int error;
307
308 switch (what) {
309 case MOD_LOAD:
310 vmmdev_init();
311 if (ppt_avail_devices() > 0)
312 iommu_init();
313 error = vmm_init();
314 if (error == 0)
315 vmm_initialized = 1;
316 break;
317 case MOD_UNLOAD:
318 error = vmmdev_cleanup();
319 if (error == 0) {
320 vmm_resume_p = NULL;
321 iommu_cleanup();
322 if (vmm_ipinum != IPI_AST)
323 vmm_ipi_free(vmm_ipinum);
324 error = VMM_CLEANUP();
325 /*
326 * Something bad happened - prevent new
327 * VMs from being created
328 */
329 if (error)
330 vmm_initialized = 0;
331 }
332 break;
333 default:
334 error = 0;
335 break;
336 }
337 return (error);
338 }
339
340 static moduledata_t vmm_kmod = {
341 "vmm",
342 vmm_handler,
343 NULL
344 };
345
346 /*
347 * vmm initialization has the following dependencies:
348 *
349 * - iommu initialization must happen after the pci passthru driver has had
350 * a chance to attach to any passthru devices (after SI_SUB_CONFIGURE).
351 *
352 * - VT-x initialization requires smp_rendezvous() and therefore must happen
353 * after SMP is fully functional (after SI_SUB_SMP).
354 */
355 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
356 MODULE_VERSION(vmm, 1);
357
358 static void
359 vm_init(struct vm *vm, bool create)
360 {
361 int i;
362
363 vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
364 vm->iommu = NULL;
365 vm->vioapic = vioapic_init(vm);
366 vm->vhpet = vhpet_init(vm);
367 vm->vatpic = vatpic_init(vm);
368 vm->vatpit = vatpit_init(vm);
369
370 CPU_ZERO(&vm->active_cpus);
371
372 vm->suspend = 0;
373 CPU_ZERO(&vm->suspended_cpus);
374
375 for (i = 0; i < VM_MAXCPU; i++)
376 vcpu_init(vm, i, create);
377 }
378
379 int
380 vm_create(const char *name, struct vm **retvm)
381 {
382 struct vm *vm;
383 struct vmspace *vmspace;
384
385 /*
386 * If vmm.ko could not be successfully initialized then don't attempt
387 * to create the virtual machine.
388 */
389 if (!vmm_initialized)
390 return (ENXIO);
391
392 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
393 return (EINVAL);
394
395 vmspace = VMSPACE_ALLOC(VM_MIN_ADDRESS, VM_MAXUSER_ADDRESS);
396 if (vmspace == NULL)
397 return (ENOMEM);
398
399 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
400 strcpy(vm->name, name);
401 vm->num_mem_segs = 0;
402 vm->vmspace = vmspace;
403 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
404
405 vm_init(vm, true);
406
407 *retvm = vm;
408 return (0);
409 }
410
411 static void
412 vm_free_mem_seg(struct vm *vm, struct mem_seg *seg)
413 {
414
415 if (seg->object != NULL)
416 vmm_mem_free(vm->vmspace, seg->gpa, seg->len);
417
418 bzero(seg, sizeof(*seg));
419 }
420
421 static void
422 vm_cleanup(struct vm *vm, bool destroy)
423 {
424 int i;
425
426 ppt_unassign_all(vm);
427
428 if (vm->iommu != NULL)
429 iommu_destroy_domain(vm->iommu);
430
431 vatpit_cleanup(vm->vatpit);
432 vhpet_cleanup(vm->vhpet);
433 vatpic_cleanup(vm->vatpic);
434 vioapic_cleanup(vm->vioapic);
435
436 for (i = 0; i < VM_MAXCPU; i++)
437 vcpu_cleanup(vm, i, destroy);
438
439 VMCLEANUP(vm->cookie);
440
441 if (destroy) {
442 for (i = 0; i < vm->num_mem_segs; i++)
443 vm_free_mem_seg(vm, &vm->mem_segs[i]);
444
445 vm->num_mem_segs = 0;
446
447 VMSPACE_FREE(vm->vmspace);
448 vm->vmspace = NULL;
449 }
450 }
451
452 void
453 vm_destroy(struct vm *vm)
454 {
455 vm_cleanup(vm, true);
456 free(vm, M_VM);
457 }
458
459 int
460 vm_reinit(struct vm *vm)
461 {
462 int error;
463
464 /*
465 * A virtual machine can be reset only if all vcpus are suspended.
466 */
467 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
468 vm_cleanup(vm, false);
469 vm_init(vm, false);
470 error = 0;
471 } else {
472 error = EBUSY;
473 }
474
475 return (error);
476 }
477
478 const char *
479 vm_name(struct vm *vm)
480 {
481 return (vm->name);
482 }
483
484 int
485 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
486 {
487 vm_object_t obj;
488
489 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
490 return (ENOMEM);
491 else
492 return (0);
493 }
494
495 int
496 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
497 {
498
499 vmm_mmio_free(vm->vmspace, gpa, len);
500 return (0);
501 }
502
503 boolean_t
504 vm_mem_allocated(struct vm *vm, vm_paddr_t gpa)
505 {
506 int i;
507 vm_paddr_t gpabase, gpalimit;
508
509 for (i = 0; i < vm->num_mem_segs; i++) {
510 gpabase = vm->mem_segs[i].gpa;
511 gpalimit = gpabase + vm->mem_segs[i].len;
512 if (gpa >= gpabase && gpa < gpalimit)
513 return (TRUE); /* 'gpa' is regular memory */
514 }
515
516 if (ppt_is_mmio(vm, gpa))
517 return (TRUE); /* 'gpa' is pci passthru mmio */
518
519 return (FALSE);
520 }
521
522 int
523 vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
524 {
525 int available, allocated;
526 struct mem_seg *seg;
527 vm_object_t object;
528 vm_paddr_t g;
529
530 if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
531 return (EINVAL);
532
533 available = allocated = 0;
534 g = gpa;
535 while (g < gpa + len) {
536 if (vm_mem_allocated(vm, g))
537 allocated++;
538 else
539 available++;
540
541 g += PAGE_SIZE;
542 }
543
544 /*
545 * If there are some allocated and some available pages in the address
546 * range then it is an error.
547 */
548 if (allocated && available)
549 return (EINVAL);
550
551 /*
552 * If the entire address range being requested has already been
553 * allocated then there isn't anything more to do.
554 */
555 if (allocated && available == 0)
556 return (0);
557
558 if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
559 return (E2BIG);
560
561 seg = &vm->mem_segs[vm->num_mem_segs];
562
563 if ((object = vmm_mem_alloc(vm->vmspace, gpa, len)) == NULL)
564 return (ENOMEM);
565
566 seg->gpa = gpa;
567 seg->len = len;
568 seg->object = object;
569 seg->wired = FALSE;
570
571 vm->num_mem_segs++;
572
573 return (0);
574 }
575
576 static vm_paddr_t
577 vm_maxmem(struct vm *vm)
578 {
579 int i;
580 vm_paddr_t gpa, maxmem;
581
582 maxmem = 0;
583 for (i = 0; i < vm->num_mem_segs; i++) {
584 gpa = vm->mem_segs[i].gpa + vm->mem_segs[i].len;
585 if (gpa > maxmem)
586 maxmem = gpa;
587 }
588 return (maxmem);
589 }
590
591 static void
592 vm_gpa_unwire(struct vm *vm)
593 {
594 int i, rv;
595 struct mem_seg *seg;
596
597 for (i = 0; i < vm->num_mem_segs; i++) {
598 seg = &vm->mem_segs[i];
599 if (!seg->wired)
600 continue;
601
602 rv = vm_map_unwire(&vm->vmspace->vm_map,
603 seg->gpa, seg->gpa + seg->len,
604 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
605 KASSERT(rv == KERN_SUCCESS, ("vm(%s) memory segment "
606 "%#lx/%ld could not be unwired: %d",
607 vm_name(vm), seg->gpa, seg->len, rv));
608
609 seg->wired = FALSE;
610 }
611 }
612
613 static int
614 vm_gpa_wire(struct vm *vm)
615 {
616 int i, rv;
617 struct mem_seg *seg;
618
619 for (i = 0; i < vm->num_mem_segs; i++) {
620 seg = &vm->mem_segs[i];
621 if (seg->wired)
622 continue;
623
624 /* XXX rlimits? */
625 rv = vm_map_wire(&vm->vmspace->vm_map,
626 seg->gpa, seg->gpa + seg->len,
627 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
628 if (rv != KERN_SUCCESS)
629 break;
630
631 seg->wired = TRUE;
632 }
633
634 if (i < vm->num_mem_segs) {
635 /*
636 * Undo the wiring before returning an error.
637 */
638 vm_gpa_unwire(vm);
639 return (EAGAIN);
640 }
641
642 return (0);
643 }
644
645 static void
646 vm_iommu_modify(struct vm *vm, boolean_t map)
647 {
648 int i, sz;
649 vm_paddr_t gpa, hpa;
650 struct mem_seg *seg;
651 void *vp, *cookie, *host_domain;
652
653 sz = PAGE_SIZE;
654 host_domain = iommu_host_domain();
655
656 for (i = 0; i < vm->num_mem_segs; i++) {
657 seg = &vm->mem_segs[i];
658 KASSERT(seg->wired, ("vm(%s) memory segment %#lx/%ld not wired",
659 vm_name(vm), seg->gpa, seg->len));
660
661 gpa = seg->gpa;
662 while (gpa < seg->gpa + seg->len) {
663 vp = vm_gpa_hold(vm, gpa, PAGE_SIZE, VM_PROT_WRITE,
664 &cookie);
665 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
666 vm_name(vm), gpa));
667
668 vm_gpa_release(cookie);
669
670 hpa = DMAP_TO_PHYS((uintptr_t)vp);
671 if (map) {
672 iommu_create_mapping(vm->iommu, gpa, hpa, sz);
673 iommu_remove_mapping(host_domain, hpa, sz);
674 } else {
675 iommu_remove_mapping(vm->iommu, gpa, sz);
676 iommu_create_mapping(host_domain, hpa, hpa, sz);
677 }
678
679 gpa += PAGE_SIZE;
680 }
681 }
682
683 /*
684 * Invalidate the cached translations associated with the domain
685 * from which pages were removed.
686 */
687 if (map)
688 iommu_invalidate_tlb(host_domain);
689 else
690 iommu_invalidate_tlb(vm->iommu);
691 }
692
693 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE)
694 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE)
695
696 int
697 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
698 {
699 int error;
700
701 error = ppt_unassign_device(vm, bus, slot, func);
702 if (error)
703 return (error);
704
705 if (ppt_assigned_devices(vm) == 0) {
706 vm_iommu_unmap(vm);
707 vm_gpa_unwire(vm);
708 }
709 return (0);
710 }
711
712 int
713 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
714 {
715 int error;
716 vm_paddr_t maxaddr;
717
718 /*
719 * Virtual machines with pci passthru devices get special treatment:
720 * - the guest physical memory is wired
721 * - the iommu is programmed to do the 'gpa' to 'hpa' translation
722 *
723 * We need to do this before the first pci passthru device is attached.
724 */
725 if (ppt_assigned_devices(vm) == 0) {
726 KASSERT(vm->iommu == NULL,
727 ("vm_assign_pptdev: iommu must be NULL"));
728 maxaddr = vm_maxmem(vm);
729 vm->iommu = iommu_create_domain(maxaddr);
730
731 error = vm_gpa_wire(vm);
732 if (error)
733 return (error);
734
735 vm_iommu_map(vm);
736 }
737
738 error = ppt_assign_device(vm, bus, slot, func);
739 return (error);
740 }
741
742 void *
743 vm_gpa_hold(struct vm *vm, vm_paddr_t gpa, size_t len, int reqprot,
744 void **cookie)
745 {
746 int count, pageoff;
747 vm_page_t m;
748
749 pageoff = gpa & PAGE_MASK;
750 if (len > PAGE_SIZE - pageoff)
751 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
752
753 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
754 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
755
756 if (count == 1) {
757 *cookie = m;
758 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
759 } else {
760 *cookie = NULL;
761 return (NULL);
762 }
763 }
764
765 void
766 vm_gpa_release(void *cookie)
767 {
768 vm_page_t m = cookie;
769
770 vm_page_lock(m);
771 vm_page_unhold(m);
772 vm_page_unlock(m);
773 }
774
775 int
776 vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
777 struct vm_memory_segment *seg)
778 {
779 int i;
780
781 for (i = 0; i < vm->num_mem_segs; i++) {
782 if (gpabase == vm->mem_segs[i].gpa) {
783 seg->gpa = vm->mem_segs[i].gpa;
784 seg->len = vm->mem_segs[i].len;
785 seg->wired = vm->mem_segs[i].wired;
786 return (0);
787 }
788 }
789 return (-1);
790 }
791
792 int
793 vm_get_memobj(struct vm *vm, vm_paddr_t gpa, size_t len,
794 vm_offset_t *offset, struct vm_object **object)
795 {
796 int i;
797 size_t seg_len;
798 vm_paddr_t seg_gpa;
799 vm_object_t seg_obj;
800
801 for (i = 0; i < vm->num_mem_segs; i++) {
802 if ((seg_obj = vm->mem_segs[i].object) == NULL)
803 continue;
804
805 seg_gpa = vm->mem_segs[i].gpa;
806 seg_len = vm->mem_segs[i].len;
807
808 if (gpa >= seg_gpa && gpa < seg_gpa + seg_len) {
809 *offset = gpa - seg_gpa;
810 *object = seg_obj;
811 vm_object_reference(seg_obj);
812 return (0);
813 }
814 }
815
816 return (EINVAL);
817 }
818
819 int
820 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
821 {
822
823 if (vcpu < 0 || vcpu >= VM_MAXCPU)
824 return (EINVAL);
825
826 if (reg >= VM_REG_LAST)
827 return (EINVAL);
828
829 return (VMGETREG(vm->cookie, vcpu, reg, retval));
830 }
831
832 int
833 vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
834 {
835
836 if (vcpu < 0 || vcpu >= VM_MAXCPU)
837 return (EINVAL);
838
839 if (reg >= VM_REG_LAST)
840 return (EINVAL);
841
842 return (VMSETREG(vm->cookie, vcpu, reg, val));
843 }
844
845 static boolean_t
846 is_descriptor_table(int reg)
847 {
848
849 switch (reg) {
850 case VM_REG_GUEST_IDTR:
851 case VM_REG_GUEST_GDTR:
852 return (TRUE);
853 default:
854 return (FALSE);
855 }
856 }
857
858 static boolean_t
859 is_segment_register(int reg)
860 {
861
862 switch (reg) {
863 case VM_REG_GUEST_ES:
864 case VM_REG_GUEST_CS:
865 case VM_REG_GUEST_SS:
866 case VM_REG_GUEST_DS:
867 case VM_REG_GUEST_FS:
868 case VM_REG_GUEST_GS:
869 case VM_REG_GUEST_TR:
870 case VM_REG_GUEST_LDTR:
871 return (TRUE);
872 default:
873 return (FALSE);
874 }
875 }
876
877 int
878 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
879 struct seg_desc *desc)
880 {
881
882 if (vcpu < 0 || vcpu >= VM_MAXCPU)
883 return (EINVAL);
884
885 if (!is_segment_register(reg) && !is_descriptor_table(reg))
886 return (EINVAL);
887
888 return (VMGETDESC(vm->cookie, vcpu, reg, desc));
889 }
890
891 int
892 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
893 struct seg_desc *desc)
894 {
895 if (vcpu < 0 || vcpu >= VM_MAXCPU)
896 return (EINVAL);
897
898 if (!is_segment_register(reg) && !is_descriptor_table(reg))
899 return (EINVAL);
900
901 return (VMSETDESC(vm->cookie, vcpu, reg, desc));
902 }
903
904 static void
905 restore_guest_fpustate(struct vcpu *vcpu)
906 {
907
908 /* flush host state to the pcb */
909 fpuexit(curthread);
910
911 /* restore guest FPU state */
912 fpu_stop_emulating();
913 fpurestore(vcpu->guestfpu);
914
915 /* restore guest XCR0 if XSAVE is enabled in the host */
916 if (rcr4() & CR4_XSAVE)
917 load_xcr(0, vcpu->guest_xcr0);
918
919 /*
920 * The FPU is now "dirty" with the guest's state so turn on emulation
921 * to trap any access to the FPU by the host.
922 */
923 fpu_start_emulating();
924 }
925
926 static void
927 save_guest_fpustate(struct vcpu *vcpu)
928 {
929
930 if ((rcr0() & CR0_TS) == 0)
931 panic("fpu emulation not enabled in host!");
932
933 /* save guest XCR0 and restore host XCR0 */
934 if (rcr4() & CR4_XSAVE) {
935 vcpu->guest_xcr0 = rxcr(0);
936 load_xcr(0, vmm_get_host_xcr0());
937 }
938
939 /* save guest FPU state */
940 fpu_stop_emulating();
941 fpusave(vcpu->guestfpu);
942 fpu_start_emulating();
943 }
944
945 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
946
947 static int
948 vcpu_set_state_locked(struct vcpu *vcpu, enum vcpu_state newstate,
949 bool from_idle)
950 {
951 int error;
952
953 vcpu_assert_locked(vcpu);
954
955 /*
956 * State transitions from the vmmdev_ioctl() must always begin from
957 * the VCPU_IDLE state. This guarantees that there is only a single
958 * ioctl() operating on a vcpu at any point.
959 */
960 if (from_idle) {
961 while (vcpu->state != VCPU_IDLE)
962 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
963 } else {
964 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
965 "vcpu idle state"));
966 }
967
968 if (vcpu->state == VCPU_RUNNING) {
969 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
970 "mismatch for running vcpu", curcpu, vcpu->hostcpu));
971 } else {
972 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
973 "vcpu that is not running", vcpu->hostcpu));
974 }
975
976 /*
977 * The following state transitions are allowed:
978 * IDLE -> FROZEN -> IDLE
979 * FROZEN -> RUNNING -> FROZEN
980 * FROZEN -> SLEEPING -> FROZEN
981 */
982 switch (vcpu->state) {
983 case VCPU_IDLE:
984 case VCPU_RUNNING:
985 case VCPU_SLEEPING:
986 error = (newstate != VCPU_FROZEN);
987 break;
988 case VCPU_FROZEN:
989 error = (newstate == VCPU_FROZEN);
990 break;
991 default:
992 error = 1;
993 break;
994 }
995
996 if (error)
997 return (EBUSY);
998
999 vcpu->state = newstate;
1000 if (newstate == VCPU_RUNNING)
1001 vcpu->hostcpu = curcpu;
1002 else
1003 vcpu->hostcpu = NOCPU;
1004
1005 if (newstate == VCPU_IDLE)
1006 wakeup(&vcpu->state);
1007
1008 return (0);
1009 }
1010
1011 static void
1012 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1013 {
1014 int error;
1015
1016 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1017 panic("Error %d setting state to %d\n", error, newstate);
1018 }
1019
1020 static void
1021 vcpu_require_state_locked(struct vcpu *vcpu, enum vcpu_state newstate)
1022 {
1023 int error;
1024
1025 if ((error = vcpu_set_state_locked(vcpu, newstate, false)) != 0)
1026 panic("Error %d setting state to %d", error, newstate);
1027 }
1028
1029 static void
1030 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func)
1031 {
1032
1033 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked"));
1034
1035 /*
1036 * Update 'rendezvous_func' and execute a write memory barrier to
1037 * ensure that it is visible across all host cpus. This is not needed
1038 * for correctness but it does ensure that all the vcpus will notice
1039 * that the rendezvous is requested immediately.
1040 */
1041 vm->rendezvous_func = func;
1042 wmb();
1043 }
1044
1045 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \
1046 do { \
1047 if (vcpuid >= 0) \
1048 VCPU_CTR0(vm, vcpuid, fmt); \
1049 else \
1050 VM_CTR0(vm, fmt); \
1051 } while (0)
1052
1053 static void
1054 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1055 {
1056
1057 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
1058 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1059
1060 mtx_lock(&vm->rendezvous_mtx);
1061 while (vm->rendezvous_func != NULL) {
1062 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1063 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1064
1065 if (vcpuid != -1 &&
1066 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1067 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1068 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1069 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1070 CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1071 }
1072 if (CPU_CMP(&vm->rendezvous_req_cpus,
1073 &vm->rendezvous_done_cpus) == 0) {
1074 VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1075 vm_set_rendezvous_func(vm, NULL);
1076 wakeup(&vm->rendezvous_func);
1077 break;
1078 }
1079 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1080 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1081 "vmrndv", 0);
1082 }
1083 mtx_unlock(&vm->rendezvous_mtx);
1084 }
1085
1086 /*
1087 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1088 */
1089 static int
1090 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1091 {
1092 struct vcpu *vcpu;
1093 const char *wmesg;
1094 int t, vcpu_halted, vm_halted;
1095
1096 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1097
1098 vcpu = &vm->vcpu[vcpuid];
1099 vcpu_halted = 0;
1100 vm_halted = 0;
1101
1102 vcpu_lock(vcpu);
1103 while (1) {
1104 /*
1105 * Do a final check for pending NMI or interrupts before
1106 * really putting this thread to sleep. Also check for
1107 * software events that would cause this vcpu to wakeup.
1108 *
1109 * These interrupts/events could have happened after the
1110 * vcpu returned from VMRUN() and before it acquired the
1111 * vcpu lock above.
1112 */
1113 if (vm->rendezvous_func != NULL || vm->suspend)
1114 break;
1115 if (vm_nmi_pending(vm, vcpuid))
1116 break;
1117 if (!intr_disabled) {
1118 if (vm_extint_pending(vm, vcpuid) ||
1119 vlapic_pending_intr(vcpu->vlapic, NULL)) {
1120 break;
1121 }
1122 }
1123
1124 /* Don't go to sleep if the vcpu thread needs to yield */
1125 if (vcpu_should_yield(vm, vcpuid))
1126 break;
1127
1128 /*
1129 * Some Linux guests implement "halt" by having all vcpus
1130 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1131 * track of the vcpus that have entered this state. When all
1132 * vcpus enter the halted state the virtual machine is halted.
1133 */
1134 if (intr_disabled) {
1135 wmesg = "vmhalt";
1136 VCPU_CTR0(vm, vcpuid, "Halted");
1137 if (!vcpu_halted && halt_detection_enabled) {
1138 vcpu_halted = 1;
1139 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1140 }
1141 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1142 vm_halted = 1;
1143 break;
1144 }
1145 } else {
1146 wmesg = "vmidle";
1147 }
1148
1149 t = ticks;
1150 vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1151 /*
1152 * XXX msleep_spin() cannot be interrupted by signals so
1153 * wake up periodically to check pending signals.
1154 */
1155 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1156 vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1157 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1158 }
1159
1160 if (vcpu_halted)
1161 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1162
1163 vcpu_unlock(vcpu);
1164
1165 if (vm_halted)
1166 vm_suspend(vm, VM_SUSPEND_HALT);
1167
1168 return (0);
1169 }
1170
1171 static int
1172 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1173 {
1174 int rv, ftype;
1175 struct vm_map *map;
1176 struct vcpu *vcpu;
1177 struct vm_exit *vme;
1178
1179 vcpu = &vm->vcpu[vcpuid];
1180 vme = &vcpu->exitinfo;
1181
1182 ftype = vme->u.paging.fault_type;
1183 KASSERT(ftype == VM_PROT_READ ||
1184 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1185 ("vm_handle_paging: invalid fault_type %d", ftype));
1186
1187 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1188 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1189 vme->u.paging.gpa, ftype);
1190 if (rv == 0)
1191 goto done;
1192 }
1193
1194 map = &vm->vmspace->vm_map;
1195 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
1196
1197 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1198 "ftype = %d", rv, vme->u.paging.gpa, ftype);
1199
1200 if (rv != KERN_SUCCESS)
1201 return (EFAULT);
1202 done:
1203 /* restart execution at the faulting instruction */
1204 vme->inst_length = 0;
1205
1206 return (0);
1207 }
1208
1209 static int
1210 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1211 {
1212 struct vie *vie;
1213 struct vcpu *vcpu;
1214 struct vm_exit *vme;
1215 uint64_t gla, gpa;
1216 struct vm_guest_paging *paging;
1217 mem_region_read_t mread;
1218 mem_region_write_t mwrite;
1219 enum vm_cpu_mode cpu_mode;
1220 int cs_d, error;
1221
1222 vcpu = &vm->vcpu[vcpuid];
1223 vme = &vcpu->exitinfo;
1224
1225 gla = vme->u.inst_emul.gla;
1226 gpa = vme->u.inst_emul.gpa;
1227 cs_d = vme->u.inst_emul.cs_d;
1228 vie = &vme->u.inst_emul.vie;
1229 paging = &vme->u.inst_emul.paging;
1230 cpu_mode = paging->cpu_mode;
1231
1232 vie_init(vie);
1233
1234 /* Fetch, decode and emulate the faulting instruction */
1235 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip,
1236 vme->inst_length, vie);
1237 if (error == 1)
1238 return (0); /* Resume guest to handle page fault */
1239 else if (error == -1)
1240 return (EFAULT);
1241 else if (error != 0)
1242 panic("%s: vmm_fetch_instruction error %d", __func__, error);
1243
1244 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0)
1245 return (EFAULT);
1246
1247 /* return to userland unless this is an in-kernel emulated device */
1248 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1249 mread = lapic_mmio_read;
1250 mwrite = lapic_mmio_write;
1251 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1252 mread = vioapic_mmio_read;
1253 mwrite = vioapic_mmio_write;
1254 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1255 mread = vhpet_mmio_read;
1256 mwrite = vhpet_mmio_write;
1257 } else {
1258 *retu = true;
1259 return (0);
1260 }
1261
1262 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1263 mread, mwrite, retu);
1264
1265 return (error);
1266 }
1267
1268 static int
1269 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1270 {
1271 int i, done;
1272 struct vcpu *vcpu;
1273
1274 done = 0;
1275 vcpu = &vm->vcpu[vcpuid];
1276
1277 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1278
1279 /*
1280 * Wait until all 'active_cpus' have suspended themselves.
1281 *
1282 * Since a VM may be suspended at any time including when one or
1283 * more vcpus are doing a rendezvous we need to call the rendezvous
1284 * handler while we are waiting to prevent a deadlock.
1285 */
1286 vcpu_lock(vcpu);
1287 while (1) {
1288 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1289 VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1290 break;
1291 }
1292
1293 if (vm->rendezvous_func == NULL) {
1294 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1295 vcpu_require_state_locked(vcpu, VCPU_SLEEPING);
1296 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1297 vcpu_require_state_locked(vcpu, VCPU_FROZEN);
1298 } else {
1299 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1300 vcpu_unlock(vcpu);
1301 vm_handle_rendezvous(vm, vcpuid);
1302 vcpu_lock(vcpu);
1303 }
1304 }
1305 vcpu_unlock(vcpu);
1306
1307 /*
1308 * Wakeup the other sleeping vcpus and return to userspace.
1309 */
1310 for (i = 0; i < VM_MAXCPU; i++) {
1311 if (CPU_ISSET(i, &vm->suspended_cpus)) {
1312 vcpu_notify_event(vm, i, false);
1313 }
1314 }
1315
1316 *retu = true;
1317 return (0);
1318 }
1319
1320 int
1321 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1322 {
1323 int i;
1324
1325 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1326 return (EINVAL);
1327
1328 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1329 VM_CTR2(vm, "virtual machine already suspended %d/%d",
1330 vm->suspend, how);
1331 return (EALREADY);
1332 }
1333
1334 VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1335
1336 /*
1337 * Notify all active vcpus that they are now suspended.
1338 */
1339 for (i = 0; i < VM_MAXCPU; i++) {
1340 if (CPU_ISSET(i, &vm->active_cpus))
1341 vcpu_notify_event(vm, i, false);
1342 }
1343
1344 return (0);
1345 }
1346
1347 void
1348 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1349 {
1350 struct vm_exit *vmexit;
1351
1352 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1353 ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1354
1355 vmexit = vm_exitinfo(vm, vcpuid);
1356 vmexit->rip = rip;
1357 vmexit->inst_length = 0;
1358 vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1359 vmexit->u.suspended.how = vm->suspend;
1360 }
1361
1362 void
1363 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1364 {
1365 struct vm_exit *vmexit;
1366
1367 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1368
1369 vmexit = vm_exitinfo(vm, vcpuid);
1370 vmexit->rip = rip;
1371 vmexit->inst_length = 0;
1372 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1373 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1374 }
1375
1376 void
1377 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1378 {
1379 struct vm_exit *vmexit;
1380
1381 vmexit = vm_exitinfo(vm, vcpuid);
1382 vmexit->rip = rip;
1383 vmexit->inst_length = 0;
1384 vmexit->exitcode = VM_EXITCODE_BOGUS;
1385 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1386 }
1387
1388 int
1389 vm_run(struct vm *vm, struct vm_run *vmrun)
1390 {
1391 int error, vcpuid;
1392 struct vcpu *vcpu;
1393 struct pcb *pcb;
1394 uint64_t tscval, rip;
1395 struct vm_exit *vme;
1396 bool retu, intr_disabled;
1397 pmap_t pmap;
1398 void *rptr, *sptr;
1399
1400 vcpuid = vmrun->cpuid;
1401
1402 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1403 return (EINVAL);
1404
1405 if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1406 return (EINVAL);
1407
1408 if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1409 return (EINVAL);
1410
1411 rptr = &vm->rendezvous_func;
1412 sptr = &vm->suspend;
1413 pmap = vmspace_pmap(vm->vmspace);
1414 vcpu = &vm->vcpu[vcpuid];
1415 vme = &vcpu->exitinfo;
1416 rip = vmrun->rip;
1417 restart:
1418 critical_enter();
1419
1420 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1421 ("vm_run: absurd pm_active"));
1422
1423 tscval = rdtsc();
1424
1425 pcb = PCPU_GET(curpcb);
1426 set_pcb_flags(pcb, PCB_FULL_IRET);
1427
1428 restore_guest_msrs(vm, vcpuid);
1429 restore_guest_fpustate(vcpu);
1430
1431 vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1432 error = VMRUN(vm->cookie, vcpuid, rip, pmap, rptr, sptr);
1433 vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1434
1435 save_guest_fpustate(vcpu);
1436 restore_host_msrs(vm, vcpuid);
1437
1438 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1439
1440 critical_exit();
1441
1442 if (error == 0) {
1443 retu = false;
1444 switch (vme->exitcode) {
1445 case VM_EXITCODE_SUSPENDED:
1446 error = vm_handle_suspend(vm, vcpuid, &retu);
1447 break;
1448 case VM_EXITCODE_IOAPIC_EOI:
1449 vioapic_process_eoi(vm, vcpuid,
1450 vme->u.ioapic_eoi.vector);
1451 break;
1452 case VM_EXITCODE_RENDEZVOUS:
1453 vm_handle_rendezvous(vm, vcpuid);
1454 error = 0;
1455 break;
1456 case VM_EXITCODE_HLT:
1457 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1458 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1459 break;
1460 case VM_EXITCODE_PAGING:
1461 error = vm_handle_paging(vm, vcpuid, &retu);
1462 break;
1463 case VM_EXITCODE_INST_EMUL:
1464 error = vm_handle_inst_emul(vm, vcpuid, &retu);
1465 break;
1466 case VM_EXITCODE_INOUT:
1467 case VM_EXITCODE_INOUT_STR:
1468 error = vm_handle_inout(vm, vcpuid, vme, &retu);
1469 break;
1470 default:
1471 retu = true; /* handled in userland */
1472 break;
1473 }
1474 }
1475
1476 if (error == 0 && retu == false) {
1477 rip = vme->rip + vme->inst_length;
1478 goto restart;
1479 }
1480
1481 /* copy the exit information */
1482 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1483 return (error);
1484 }
1485
1486 int
1487 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1488 {
1489 struct vcpu *vcpu;
1490 int type, vector;
1491
1492 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1493 return (EINVAL);
1494
1495 vcpu = &vm->vcpu[vcpuid];
1496
1497 if (info & VM_INTINFO_VALID) {
1498 type = info & VM_INTINFO_TYPE;
1499 vector = info & 0xff;
1500 if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1501 return (EINVAL);
1502 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1503 return (EINVAL);
1504 if (info & VM_INTINFO_RSVD)
1505 return (EINVAL);
1506 } else {
1507 info = 0;
1508 }
1509 VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1510 vcpu->exitintinfo = info;
1511 return (0);
1512 }
1513
1514 enum exc_class {
1515 EXC_BENIGN,
1516 EXC_CONTRIBUTORY,
1517 EXC_PAGEFAULT
1518 };
1519
1520 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */
1521
1522 static enum exc_class
1523 exception_class(uint64_t info)
1524 {
1525 int type, vector;
1526
1527 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1528 type = info & VM_INTINFO_TYPE;
1529 vector = info & 0xff;
1530
1531 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1532 switch (type) {
1533 case VM_INTINFO_HWINTR:
1534 case VM_INTINFO_SWINTR:
1535 case VM_INTINFO_NMI:
1536 return (EXC_BENIGN);
1537 default:
1538 /*
1539 * Hardware exception.
1540 *
1541 * SVM and VT-x use identical type values to represent NMI,
1542 * hardware interrupt and software interrupt.
1543 *
1544 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1545 * for exceptions except #BP and #OF. #BP and #OF use a type
1546 * value of '5' or '6'. Therefore we don't check for explicit
1547 * values of 'type' to classify 'intinfo' into a hardware
1548 * exception.
1549 */
1550 break;
1551 }
1552
1553 switch (vector) {
1554 case IDT_PF:
1555 case IDT_VE:
1556 return (EXC_PAGEFAULT);
1557 case IDT_DE:
1558 case IDT_TS:
1559 case IDT_NP:
1560 case IDT_SS:
1561 case IDT_GP:
1562 return (EXC_CONTRIBUTORY);
1563 default:
1564 return (EXC_BENIGN);
1565 }
1566 }
1567
1568 static int
1569 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1570 uint64_t *retinfo)
1571 {
1572 enum exc_class exc1, exc2;
1573 int type1, vector1;
1574
1575 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1576 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1577
1578 /*
1579 * If an exception occurs while attempting to call the double-fault
1580 * handler the processor enters shutdown mode (aka triple fault).
1581 */
1582 type1 = info1 & VM_INTINFO_TYPE;
1583 vector1 = info1 & 0xff;
1584 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1585 VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1586 info1, info2);
1587 vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1588 *retinfo = 0;
1589 return (0);
1590 }
1591
1592 /*
1593 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1594 */
1595 exc1 = exception_class(info1);
1596 exc2 = exception_class(info2);
1597 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1598 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1599 /* Convert nested fault into a double fault. */
1600 *retinfo = IDT_DF;
1601 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1602 *retinfo |= VM_INTINFO_DEL_ERRCODE;
1603 } else {
1604 /* Handle exceptions serially */
1605 *retinfo = info2;
1606 }
1607 return (1);
1608 }
1609
1610 static uint64_t
1611 vcpu_exception_intinfo(struct vcpu *vcpu)
1612 {
1613 uint64_t info = 0;
1614
1615 if (vcpu->exception_pending) {
1616 info = vcpu->exception.vector & 0xff;
1617 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1618 if (vcpu->exception.error_code_valid) {
1619 info |= VM_INTINFO_DEL_ERRCODE;
1620 info |= (uint64_t)vcpu->exception.error_code << 32;
1621 }
1622 }
1623 return (info);
1624 }
1625
1626 int
1627 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1628 {
1629 struct vcpu *vcpu;
1630 uint64_t info1, info2;
1631 int valid;
1632
1633 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid));
1634
1635 vcpu = &vm->vcpu[vcpuid];
1636
1637 info1 = vcpu->exitintinfo;
1638 vcpu->exitintinfo = 0;
1639
1640 info2 = 0;
1641 if (vcpu->exception_pending) {
1642 info2 = vcpu_exception_intinfo(vcpu);
1643 vcpu->exception_pending = 0;
1644 VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1645 vcpu->exception.vector, info2);
1646 }
1647
1648 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
1649 valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
1650 } else if (info1 & VM_INTINFO_VALID) {
1651 *retinfo = info1;
1652 valid = 1;
1653 } else if (info2 & VM_INTINFO_VALID) {
1654 *retinfo = info2;
1655 valid = 1;
1656 } else {
1657 valid = 0;
1658 }
1659
1660 if (valid) {
1661 VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
1662 "retinfo(%#lx)", __func__, info1, info2, *retinfo);
1663 }
1664
1665 return (valid);
1666 }
1667
1668 int
1669 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
1670 {
1671 struct vcpu *vcpu;
1672
1673 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1674 return (EINVAL);
1675
1676 vcpu = &vm->vcpu[vcpuid];
1677 *info1 = vcpu->exitintinfo;
1678 *info2 = vcpu_exception_intinfo(vcpu);
1679 return (0);
1680 }
1681
1682 int
1683 vm_inject_exception(struct vm *vm, int vcpuid, struct vm_exception *exception)
1684 {
1685 struct vcpu *vcpu;
1686
1687 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1688 return (EINVAL);
1689
1690 if (exception->vector < 0 || exception->vector >= 32)
1691 return (EINVAL);
1692
1693 /*
1694 * A double fault exception should never be injected directly into
1695 * the guest. It is a derived exception that results from specific
1696 * combinations of nested faults.
1697 */
1698 if (exception->vector == IDT_DF)
1699 return (EINVAL);
1700
1701 vcpu = &vm->vcpu[vcpuid];
1702
1703 if (vcpu->exception_pending) {
1704 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
1705 "pending exception %d", exception->vector,
1706 vcpu->exception.vector);
1707 return (EBUSY);
1708 }
1709
1710 vcpu->exception_pending = 1;
1711 vcpu->exception = *exception;
1712 VCPU_CTR1(vm, vcpuid, "Exception %d pending", exception->vector);
1713 return (0);
1714 }
1715
1716 void
1717 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
1718 int errcode)
1719 {
1720 struct vm_exception exception;
1721 struct vm_exit *vmexit;
1722 struct vm *vm;
1723 int error;
1724
1725 vm = vmarg;
1726
1727 exception.vector = vector;
1728 exception.error_code = errcode;
1729 exception.error_code_valid = errcode_valid;
1730 error = vm_inject_exception(vm, vcpuid, &exception);
1731 KASSERT(error == 0, ("vm_inject_exception error %d", error));
1732
1733 /*
1734 * A fault-like exception allows the instruction to be restarted
1735 * after the exception handler returns.
1736 *
1737 * By setting the inst_length to 0 we ensure that the instruction
1738 * pointer remains at the faulting instruction.
1739 */
1740 vmexit = vm_exitinfo(vm, vcpuid);
1741 vmexit->inst_length = 0;
1742 }
1743
1744 void
1745 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
1746 {
1747 struct vm *vm;
1748 int error;
1749
1750 vm = vmarg;
1751 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
1752 error_code, cr2);
1753
1754 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
1755 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
1756
1757 vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
1758 }
1759
1760 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
1761
1762 int
1763 vm_inject_nmi(struct vm *vm, int vcpuid)
1764 {
1765 struct vcpu *vcpu;
1766
1767 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1768 return (EINVAL);
1769
1770 vcpu = &vm->vcpu[vcpuid];
1771
1772 vcpu->nmi_pending = 1;
1773 vcpu_notify_event(vm, vcpuid, false);
1774 return (0);
1775 }
1776
1777 int
1778 vm_nmi_pending(struct vm *vm, int vcpuid)
1779 {
1780 struct vcpu *vcpu;
1781
1782 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1783 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1784
1785 vcpu = &vm->vcpu[vcpuid];
1786
1787 return (vcpu->nmi_pending);
1788 }
1789
1790 void
1791 vm_nmi_clear(struct vm *vm, int vcpuid)
1792 {
1793 struct vcpu *vcpu;
1794
1795 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1796 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
1797
1798 vcpu = &vm->vcpu[vcpuid];
1799
1800 if (vcpu->nmi_pending == 0)
1801 panic("vm_nmi_clear: inconsistent nmi_pending state");
1802
1803 vcpu->nmi_pending = 0;
1804 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
1805 }
1806
1807 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
1808
1809 int
1810 vm_inject_extint(struct vm *vm, int vcpuid)
1811 {
1812 struct vcpu *vcpu;
1813
1814 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1815 return (EINVAL);
1816
1817 vcpu = &vm->vcpu[vcpuid];
1818
1819 vcpu->extint_pending = 1;
1820 vcpu_notify_event(vm, vcpuid, false);
1821 return (0);
1822 }
1823
1824 int
1825 vm_extint_pending(struct vm *vm, int vcpuid)
1826 {
1827 struct vcpu *vcpu;
1828
1829 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1830 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1831
1832 vcpu = &vm->vcpu[vcpuid];
1833
1834 return (vcpu->extint_pending);
1835 }
1836
1837 void
1838 vm_extint_clear(struct vm *vm, int vcpuid)
1839 {
1840 struct vcpu *vcpu;
1841
1842 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1843 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
1844
1845 vcpu = &vm->vcpu[vcpuid];
1846
1847 if (vcpu->extint_pending == 0)
1848 panic("vm_extint_clear: inconsistent extint_pending state");
1849
1850 vcpu->extint_pending = 0;
1851 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
1852 }
1853
1854 int
1855 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
1856 {
1857 if (vcpu < 0 || vcpu >= VM_MAXCPU)
1858 return (EINVAL);
1859
1860 if (type < 0 || type >= VM_CAP_MAX)
1861 return (EINVAL);
1862
1863 return (VMGETCAP(vm->cookie, vcpu, type, retval));
1864 }
1865
1866 int
1867 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
1868 {
1869 if (vcpu < 0 || vcpu >= VM_MAXCPU)
1870 return (EINVAL);
1871
1872 if (type < 0 || type >= VM_CAP_MAX)
1873 return (EINVAL);
1874
1875 return (VMSETCAP(vm->cookie, vcpu, type, val));
1876 }
1877
1878 uint64_t *
1879 vm_guest_msrs(struct vm *vm, int cpu)
1880 {
1881 return (vm->vcpu[cpu].guest_msrs);
1882 }
1883
1884 struct vlapic *
1885 vm_lapic(struct vm *vm, int cpu)
1886 {
1887 return (vm->vcpu[cpu].vlapic);
1888 }
1889
1890 struct vioapic *
1891 vm_ioapic(struct vm *vm)
1892 {
1893
1894 return (vm->vioapic);
1895 }
1896
1897 struct vhpet *
1898 vm_hpet(struct vm *vm)
1899 {
1900
1901 return (vm->vhpet);
1902 }
1903
1904 boolean_t
1905 vmm_is_pptdev(int bus, int slot, int func)
1906 {
1907 int found, i, n;
1908 int b, s, f;
1909 char *val, *cp, *cp2;
1910
1911 /*
1912 * XXX
1913 * The length of an environment variable is limited to 128 bytes which
1914 * puts an upper limit on the number of passthru devices that may be
1915 * specified using a single environment variable.
1916 *
1917 * Work around this by scanning multiple environment variable
1918 * names instead of a single one - yuck!
1919 */
1920 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
1921
1922 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
1923 found = 0;
1924 for (i = 0; names[i] != NULL && !found; i++) {
1925 cp = val = getenv(names[i]);
1926 while (cp != NULL && *cp != '\0') {
1927 if ((cp2 = strchr(cp, ' ')) != NULL)
1928 *cp2 = '\0';
1929
1930 n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
1931 if (n == 3 && bus == b && slot == s && func == f) {
1932 found = 1;
1933 break;
1934 }
1935
1936 if (cp2 != NULL)
1937 *cp2++ = ' ';
1938
1939 cp = cp2;
1940 }
1941 freeenv(val);
1942 }
1943 return (found);
1944 }
1945
1946 void *
1947 vm_iommu_domain(struct vm *vm)
1948 {
1949
1950 return (vm->iommu);
1951 }
1952
1953 int
1954 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1955 bool from_idle)
1956 {
1957 int error;
1958 struct vcpu *vcpu;
1959
1960 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1961 panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
1962
1963 vcpu = &vm->vcpu[vcpuid];
1964
1965 vcpu_lock(vcpu);
1966 error = vcpu_set_state_locked(vcpu, newstate, from_idle);
1967 vcpu_unlock(vcpu);
1968
1969 return (error);
1970 }
1971
1972 enum vcpu_state
1973 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
1974 {
1975 struct vcpu *vcpu;
1976 enum vcpu_state state;
1977
1978 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1979 panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
1980
1981 vcpu = &vm->vcpu[vcpuid];
1982
1983 vcpu_lock(vcpu);
1984 state = vcpu->state;
1985 if (hostcpu != NULL)
1986 *hostcpu = vcpu->hostcpu;
1987 vcpu_unlock(vcpu);
1988
1989 return (state);
1990 }
1991
1992 int
1993 vm_activate_cpu(struct vm *vm, int vcpuid)
1994 {
1995
1996 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1997 return (EINVAL);
1998
1999 if (CPU_ISSET(vcpuid, &vm->active_cpus))
2000 return (EBUSY);
2001
2002 VCPU_CTR0(vm, vcpuid, "activated");
2003 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2004 return (0);
2005 }
2006
2007 cpuset_t
2008 vm_active_cpus(struct vm *vm)
2009 {
2010
2011 return (vm->active_cpus);
2012 }
2013
2014 cpuset_t
2015 vm_suspended_cpus(struct vm *vm)
2016 {
2017
2018 return (vm->suspended_cpus);
2019 }
2020
2021 void *
2022 vcpu_stats(struct vm *vm, int vcpuid)
2023 {
2024
2025 return (vm->vcpu[vcpuid].stats);
2026 }
2027
2028 int
2029 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2030 {
2031 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2032 return (EINVAL);
2033
2034 *state = vm->vcpu[vcpuid].x2apic_state;
2035
2036 return (0);
2037 }
2038
2039 int
2040 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2041 {
2042 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2043 return (EINVAL);
2044
2045 if (state >= X2APIC_STATE_LAST)
2046 return (EINVAL);
2047
2048 vm->vcpu[vcpuid].x2apic_state = state;
2049
2050 vlapic_set_x2apic_state(vm, vcpuid, state);
2051
2052 return (0);
2053 }
2054
2055 /*
2056 * This function is called to ensure that a vcpu "sees" a pending event
2057 * as soon as possible:
2058 * - If the vcpu thread is sleeping then it is woken up.
2059 * - If the vcpu is running on a different host_cpu then an IPI will be directed
2060 * to the host_cpu to cause the vcpu to trap into the hypervisor.
2061 */
2062 void
2063 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2064 {
2065 int hostcpu;
2066 struct vcpu *vcpu;
2067
2068 vcpu = &vm->vcpu[vcpuid];
2069
2070 vcpu_lock(vcpu);
2071 hostcpu = vcpu->hostcpu;
2072 if (vcpu->state == VCPU_RUNNING) {
2073 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2074 if (hostcpu != curcpu) {
2075 if (lapic_intr) {
2076 vlapic_post_intr(vcpu->vlapic, hostcpu,
2077 vmm_ipinum);
2078 } else {
2079 ipi_cpu(hostcpu, vmm_ipinum);
2080 }
2081 } else {
2082 /*
2083 * If the 'vcpu' is running on 'curcpu' then it must
2084 * be sending a notification to itself (e.g. SELF_IPI).
2085 * The pending event will be picked up when the vcpu
2086 * transitions back to guest context.
2087 */
2088 }
2089 } else {
2090 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2091 "with hostcpu %d", vcpu->state, hostcpu));
2092 if (vcpu->state == VCPU_SLEEPING)
2093 wakeup_one(vcpu);
2094 }
2095 vcpu_unlock(vcpu);
2096 }
2097
2098 struct vmspace *
2099 vm_get_vmspace(struct vm *vm)
2100 {
2101
2102 return (vm->vmspace);
2103 }
2104
2105 int
2106 vm_apicid2vcpuid(struct vm *vm, int apicid)
2107 {
2108 /*
2109 * XXX apic id is assumed to be numerically identical to vcpu id
2110 */
2111 return (apicid);
2112 }
2113
2114 void
2115 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2116 vm_rendezvous_func_t func, void *arg)
2117 {
2118 int i;
2119
2120 /*
2121 * Enforce that this function is called without any locks
2122 */
2123 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2124 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
2125 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2126
2127 restart:
2128 mtx_lock(&vm->rendezvous_mtx);
2129 if (vm->rendezvous_func != NULL) {
2130 /*
2131 * If a rendezvous is already in progress then we need to
2132 * call the rendezvous handler in case this 'vcpuid' is one
2133 * of the targets of the rendezvous.
2134 */
2135 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2136 mtx_unlock(&vm->rendezvous_mtx);
2137 vm_handle_rendezvous(vm, vcpuid);
2138 goto restart;
2139 }
2140 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2141 "rendezvous is still in progress"));
2142
2143 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2144 vm->rendezvous_req_cpus = dest;
2145 CPU_ZERO(&vm->rendezvous_done_cpus);
2146 vm->rendezvous_arg = arg;
2147 vm_set_rendezvous_func(vm, func);
2148 mtx_unlock(&vm->rendezvous_mtx);
2149
2150 /*
2151 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2152 * vcpus so they handle the rendezvous as soon as possible.
2153 */
2154 for (i = 0; i < VM_MAXCPU; i++) {
2155 if (CPU_ISSET(i, &dest))
2156 vcpu_notify_event(vm, i, false);
2157 }
2158
2159 vm_handle_rendezvous(vm, vcpuid);
2160 }
2161
2162 struct vatpic *
2163 vm_atpic(struct vm *vm)
2164 {
2165 return (vm->vatpic);
2166 }
2167
2168 struct vatpit *
2169 vm_atpit(struct vm *vm)
2170 {
2171 return (vm->vatpit);
2172 }
2173
2174 enum vm_reg_name
2175 vm_segment_name(int seg)
2176 {
2177 static enum vm_reg_name seg_names[] = {
2178 VM_REG_GUEST_ES,
2179 VM_REG_GUEST_CS,
2180 VM_REG_GUEST_SS,
2181 VM_REG_GUEST_DS,
2182 VM_REG_GUEST_FS,
2183 VM_REG_GUEST_GS
2184 };
2185
2186 KASSERT(seg >= 0 && seg < nitems(seg_names),
2187 ("%s: invalid segment encoding %d", __func__, seg));
2188 return (seg_names[seg]);
2189 }
2190
2191 void
2192 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2193 int num_copyinfo)
2194 {
2195 int idx;
2196
2197 for (idx = 0; idx < num_copyinfo; idx++) {
2198 if (copyinfo[idx].cookie != NULL)
2199 vm_gpa_release(copyinfo[idx].cookie);
2200 }
2201 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2202 }
2203
2204 int
2205 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2206 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2207 int num_copyinfo)
2208 {
2209 int error, idx, nused;
2210 size_t n, off, remaining;
2211 void *hva, *cookie;
2212 uint64_t gpa;
2213
2214 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2215
2216 nused = 0;
2217 remaining = len;
2218 while (remaining > 0) {
2219 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2220 error = vmm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa);
2221 if (error)
2222 return (error);
2223 off = gpa & PAGE_MASK;
2224 n = min(remaining, PAGE_SIZE - off);
2225 copyinfo[nused].gpa = gpa;
2226 copyinfo[nused].len = n;
2227 remaining -= n;
2228 gla += n;
2229 nused++;
2230 }
2231
2232 for (idx = 0; idx < nused; idx++) {
2233 hva = vm_gpa_hold(vm, copyinfo[idx].gpa, copyinfo[idx].len,
2234 prot, &cookie);
2235 if (hva == NULL)
2236 break;
2237 copyinfo[idx].hva = hva;
2238 copyinfo[idx].cookie = cookie;
2239 }
2240
2241 if (idx != nused) {
2242 vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2243 return (-1);
2244 } else {
2245 return (0);
2246 }
2247 }
2248
2249 void
2250 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2251 size_t len)
2252 {
2253 char *dst;
2254 int idx;
2255
2256 dst = kaddr;
2257 idx = 0;
2258 while (len > 0) {
2259 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2260 len -= copyinfo[idx].len;
2261 dst += copyinfo[idx].len;
2262 idx++;
2263 }
2264 }
2265
2266 void
2267 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2268 struct vm_copyinfo *copyinfo, size_t len)
2269 {
2270 const char *src;
2271 int idx;
2272
2273 src = kaddr;
2274 idx = 0;
2275 while (len > 0) {
2276 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2277 len -= copyinfo[idx].len;
2278 src += copyinfo[idx].len;
2279 idx++;
2280 }
2281 }
2282
2283 /*
2284 * Return the amount of in-use and wired memory for the VM. Since
2285 * these are global stats, only return the values with for vCPU 0
2286 */
2287 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2288 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2289
2290 static void
2291 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2292 {
2293
2294 if (vcpu == 0) {
2295 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2296 PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2297 }
2298 }
2299
2300 static void
2301 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2302 {
2303
2304 if (vcpu == 0) {
2305 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2306 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2307 }
2308 }
2309
2310 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2311 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
Cache object: a3c0edd0ebe94907ae765318ca701c45
|