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: stable/11/sys/amd64/vmm/vmm.c 306471 2016-09-30 01:39:18Z jhb $
27 */
28
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD: stable/11/sys/amd64/vmm/vmm.c 306471 2016-09-30 01:39:18Z jhb $");
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/pcb.h>
57 #include <machine/smp.h>
58 #include <x86/psl.h>
59 #include <x86/apicreg.h>
60
61 #include <machine/vmm.h>
62 #include <machine/vmm_dev.h>
63 #include <machine/vmm_instruction_emul.h>
64
65 #include "vmm_ioport.h"
66 #include "vmm_ktr.h"
67 #include "vmm_host.h"
68 #include "vmm_mem.h"
69 #include "vmm_util.h"
70 #include "vatpic.h"
71 #include "vatpit.h"
72 #include "vhpet.h"
73 #include "vioapic.h"
74 #include "vlapic.h"
75 #include "vpmtmr.h"
76 #include "vrtc.h"
77 #include "vmm_stat.h"
78 #include "vmm_lapic.h"
79
80 #include "io/ppt.h"
81 #include "io/iommu.h"
82
83 struct vlapic;
84
85 /*
86 * Initialization:
87 * (a) allocated when vcpu is created
88 * (i) initialized when vcpu is created and when it is reinitialized
89 * (o) initialized the first time the vcpu is created
90 * (x) initialized before use
91 */
92 struct vcpu {
93 struct mtx mtx; /* (o) protects 'state' and 'hostcpu' */
94 enum vcpu_state state; /* (o) vcpu state */
95 int hostcpu; /* (o) vcpu's host cpu */
96 int reqidle; /* (i) request vcpu to idle */
97 struct vlapic *vlapic; /* (i) APIC device model */
98 enum x2apic_state x2apic_state; /* (i) APIC mode */
99 uint64_t exitintinfo; /* (i) events pending at VM exit */
100 int nmi_pending; /* (i) NMI pending */
101 int extint_pending; /* (i) INTR pending */
102 int exception_pending; /* (i) exception pending */
103 int exc_vector; /* (x) exception collateral */
104 int exc_errcode_valid;
105 uint32_t exc_errcode;
106 struct savefpu *guestfpu; /* (a,i) guest fpu state */
107 uint64_t guest_xcr0; /* (i) guest %xcr0 register */
108 void *stats; /* (a,i) statistics */
109 struct vm_exit exitinfo; /* (x) exit reason and collateral */
110 uint64_t nextrip; /* (x) next instruction to execute */
111 };
112
113 #define vcpu_lock_initialized(v) mtx_initialized(&((v)->mtx))
114 #define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_SPIN)
115 #define vcpu_lock(v) mtx_lock_spin(&((v)->mtx))
116 #define vcpu_unlock(v) mtx_unlock_spin(&((v)->mtx))
117 #define vcpu_assert_locked(v) mtx_assert(&((v)->mtx), MA_OWNED)
118
119 struct mem_seg {
120 size_t len;
121 bool sysmem;
122 struct vm_object *object;
123 };
124 #define VM_MAX_MEMSEGS 3
125
126 struct mem_map {
127 vm_paddr_t gpa;
128 size_t len;
129 vm_ooffset_t segoff;
130 int segid;
131 int prot;
132 int flags;
133 };
134 #define VM_MAX_MEMMAPS 4
135
136 /*
137 * Initialization:
138 * (o) initialized the first time the VM is created
139 * (i) initialized when VM is created and when it is reinitialized
140 * (x) initialized before use
141 */
142 struct vm {
143 void *cookie; /* (i) cpu-specific data */
144 void *iommu; /* (x) iommu-specific data */
145 struct vhpet *vhpet; /* (i) virtual HPET */
146 struct vioapic *vioapic; /* (i) virtual ioapic */
147 struct vatpic *vatpic; /* (i) virtual atpic */
148 struct vatpit *vatpit; /* (i) virtual atpit */
149 struct vpmtmr *vpmtmr; /* (i) virtual ACPI PM timer */
150 struct vrtc *vrtc; /* (o) virtual RTC */
151 volatile cpuset_t active_cpus; /* (i) active vcpus */
152 int suspend; /* (i) stop VM execution */
153 volatile cpuset_t suspended_cpus; /* (i) suspended vcpus */
154 volatile cpuset_t halted_cpus; /* (x) cpus in a hard halt */
155 cpuset_t rendezvous_req_cpus; /* (x) rendezvous requested */
156 cpuset_t rendezvous_done_cpus; /* (x) rendezvous finished */
157 void *rendezvous_arg; /* (x) rendezvous func/arg */
158 vm_rendezvous_func_t rendezvous_func;
159 struct mtx rendezvous_mtx; /* (o) rendezvous lock */
160 struct mem_map mem_maps[VM_MAX_MEMMAPS]; /* (i) guest address space */
161 struct mem_seg mem_segs[VM_MAX_MEMSEGS]; /* (o) guest memory regions */
162 struct vmspace *vmspace; /* (o) guest's address space */
163 char name[VM_MAX_NAMELEN]; /* (o) virtual machine name */
164 struct vcpu vcpu[VM_MAXCPU]; /* (i) guest vcpus */
165 };
166
167 static int vmm_initialized;
168
169 static struct vmm_ops *ops;
170 #define VMM_INIT(num) (ops != NULL ? (*ops->init)(num) : 0)
171 #define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0)
172 #define VMM_RESUME() (ops != NULL ? (*ops->resume)() : 0)
173
174 #define VMINIT(vm, pmap) (ops != NULL ? (*ops->vminit)(vm, pmap): NULL)
175 #define VMRUN(vmi, vcpu, rip, pmap, evinfo) \
176 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip, pmap, evinfo) : ENXIO)
177 #define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
178 #define VMSPACE_ALLOC(min, max) \
179 (ops != NULL ? (*ops->vmspace_alloc)(min, max) : NULL)
180 #define VMSPACE_FREE(vmspace) \
181 (ops != NULL ? (*ops->vmspace_free)(vmspace) : ENXIO)
182 #define VMGETREG(vmi, vcpu, num, retval) \
183 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
184 #define VMSETREG(vmi, vcpu, num, val) \
185 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
186 #define VMGETDESC(vmi, vcpu, num, desc) \
187 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
188 #define VMSETDESC(vmi, vcpu, num, desc) \
189 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
190 #define VMGETCAP(vmi, vcpu, num, retval) \
191 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
192 #define VMSETCAP(vmi, vcpu, num, val) \
193 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
194 #define VLAPIC_INIT(vmi, vcpu) \
195 (ops != NULL ? (*ops->vlapic_init)(vmi, vcpu) : NULL)
196 #define VLAPIC_CLEANUP(vmi, vlapic) \
197 (ops != NULL ? (*ops->vlapic_cleanup)(vmi, vlapic) : NULL)
198
199 #define fpu_start_emulating() load_cr0(rcr0() | CR0_TS)
200 #define fpu_stop_emulating() clts()
201
202 static MALLOC_DEFINE(M_VM, "vm", "vm");
203
204 /* statistics */
205 static VMM_STAT(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
206
207 SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
208
209 /*
210 * Halt the guest if all vcpus are executing a HLT instruction with
211 * interrupts disabled.
212 */
213 static int halt_detection_enabled = 1;
214 SYSCTL_INT(_hw_vmm, OID_AUTO, halt_detection, CTLFLAG_RDTUN,
215 &halt_detection_enabled, 0,
216 "Halt VM if all vcpus execute HLT with interrupts disabled");
217
218 static int vmm_ipinum;
219 SYSCTL_INT(_hw_vmm, OID_AUTO, ipinum, CTLFLAG_RD, &vmm_ipinum, 0,
220 "IPI vector used for vcpu notifications");
221
222 static int trace_guest_exceptions;
223 SYSCTL_INT(_hw_vmm, OID_AUTO, trace_guest_exceptions, CTLFLAG_RDTUN,
224 &trace_guest_exceptions, 0,
225 "Trap into hypervisor on all guest exceptions and reflect them back");
226
227 static void vm_free_memmap(struct vm *vm, int ident);
228 static bool sysmem_mapping(struct vm *vm, struct mem_map *mm);
229 static void vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr);
230
231 #ifdef KTR
232 static const char *
233 vcpu_state2str(enum vcpu_state state)
234 {
235
236 switch (state) {
237 case VCPU_IDLE:
238 return ("idle");
239 case VCPU_FROZEN:
240 return ("frozen");
241 case VCPU_RUNNING:
242 return ("running");
243 case VCPU_SLEEPING:
244 return ("sleeping");
245 default:
246 return ("unknown");
247 }
248 }
249 #endif
250
251 static void
252 vcpu_cleanup(struct vm *vm, int i, bool destroy)
253 {
254 struct vcpu *vcpu = &vm->vcpu[i];
255
256 VLAPIC_CLEANUP(vm->cookie, vcpu->vlapic);
257 if (destroy) {
258 vmm_stat_free(vcpu->stats);
259 fpu_save_area_free(vcpu->guestfpu);
260 }
261 }
262
263 static void
264 vcpu_init(struct vm *vm, int vcpu_id, bool create)
265 {
266 struct vcpu *vcpu;
267
268 KASSERT(vcpu_id >= 0 && vcpu_id < VM_MAXCPU,
269 ("vcpu_init: invalid vcpu %d", vcpu_id));
270
271 vcpu = &vm->vcpu[vcpu_id];
272
273 if (create) {
274 KASSERT(!vcpu_lock_initialized(vcpu), ("vcpu %d already "
275 "initialized", vcpu_id));
276 vcpu_lock_init(vcpu);
277 vcpu->state = VCPU_IDLE;
278 vcpu->hostcpu = NOCPU;
279 vcpu->guestfpu = fpu_save_area_alloc();
280 vcpu->stats = vmm_stat_alloc();
281 }
282
283 vcpu->vlapic = VLAPIC_INIT(vm->cookie, vcpu_id);
284 vm_set_x2apic_state(vm, vcpu_id, X2APIC_DISABLED);
285 vcpu->reqidle = 0;
286 vcpu->exitintinfo = 0;
287 vcpu->nmi_pending = 0;
288 vcpu->extint_pending = 0;
289 vcpu->exception_pending = 0;
290 vcpu->guest_xcr0 = XFEATURE_ENABLED_X87;
291 fpu_save_area_reset(vcpu->guestfpu);
292 vmm_stat_init(vcpu->stats);
293 }
294
295 int
296 vcpu_trace_exceptions(struct vm *vm, int vcpuid)
297 {
298
299 return (trace_guest_exceptions);
300 }
301
302 struct vm_exit *
303 vm_exitinfo(struct vm *vm, int cpuid)
304 {
305 struct vcpu *vcpu;
306
307 if (cpuid < 0 || cpuid >= VM_MAXCPU)
308 panic("vm_exitinfo: invalid cpuid %d", cpuid);
309
310 vcpu = &vm->vcpu[cpuid];
311
312 return (&vcpu->exitinfo);
313 }
314
315 static void
316 vmm_resume(void)
317 {
318 VMM_RESUME();
319 }
320
321 static int
322 vmm_init(void)
323 {
324 int error;
325
326 vmm_host_state_init();
327
328 vmm_ipinum = lapic_ipi_alloc(&IDTVEC(justreturn));
329 if (vmm_ipinum < 0)
330 vmm_ipinum = IPI_AST;
331
332 error = vmm_mem_init();
333 if (error)
334 return (error);
335
336 if (vmm_is_intel())
337 ops = &vmm_ops_intel;
338 else if (vmm_is_amd())
339 ops = &vmm_ops_amd;
340 else
341 return (ENXIO);
342
343 vmm_resume_p = vmm_resume;
344
345 return (VMM_INIT(vmm_ipinum));
346 }
347
348 static int
349 vmm_handler(module_t mod, int what, void *arg)
350 {
351 int error;
352
353 switch (what) {
354 case MOD_LOAD:
355 vmmdev_init();
356 error = vmm_init();
357 if (error == 0)
358 vmm_initialized = 1;
359 break;
360 case MOD_UNLOAD:
361 error = vmmdev_cleanup();
362 if (error == 0) {
363 vmm_resume_p = NULL;
364 iommu_cleanup();
365 if (vmm_ipinum != IPI_AST)
366 lapic_ipi_free(vmm_ipinum);
367 error = VMM_CLEANUP();
368 /*
369 * Something bad happened - prevent new
370 * VMs from being created
371 */
372 if (error)
373 vmm_initialized = 0;
374 }
375 break;
376 default:
377 error = 0;
378 break;
379 }
380 return (error);
381 }
382
383 static moduledata_t vmm_kmod = {
384 "vmm",
385 vmm_handler,
386 NULL
387 };
388
389 /*
390 * vmm initialization has the following dependencies:
391 *
392 * - VT-x initialization requires smp_rendezvous() and therefore must happen
393 * after SMP is fully functional (after SI_SUB_SMP).
394 */
395 DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_SMP + 1, SI_ORDER_ANY);
396 MODULE_VERSION(vmm, 1);
397
398 static void
399 vm_init(struct vm *vm, bool create)
400 {
401 int i;
402
403 vm->cookie = VMINIT(vm, vmspace_pmap(vm->vmspace));
404 vm->iommu = NULL;
405 vm->vioapic = vioapic_init(vm);
406 vm->vhpet = vhpet_init(vm);
407 vm->vatpic = vatpic_init(vm);
408 vm->vatpit = vatpit_init(vm);
409 vm->vpmtmr = vpmtmr_init(vm);
410 if (create)
411 vm->vrtc = vrtc_init(vm);
412
413 CPU_ZERO(&vm->active_cpus);
414
415 vm->suspend = 0;
416 CPU_ZERO(&vm->suspended_cpus);
417
418 for (i = 0; i < VM_MAXCPU; i++)
419 vcpu_init(vm, i, create);
420 }
421
422 int
423 vm_create(const char *name, struct vm **retvm)
424 {
425 struct vm *vm;
426 struct vmspace *vmspace;
427
428 /*
429 * If vmm.ko could not be successfully initialized then don't attempt
430 * to create the virtual machine.
431 */
432 if (!vmm_initialized)
433 return (ENXIO);
434
435 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
436 return (EINVAL);
437
438 vmspace = VMSPACE_ALLOC(0, VM_MAXUSER_ADDRESS);
439 if (vmspace == NULL)
440 return (ENOMEM);
441
442 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
443 strcpy(vm->name, name);
444 vm->vmspace = vmspace;
445 mtx_init(&vm->rendezvous_mtx, "vm rendezvous lock", 0, MTX_DEF);
446
447 vm_init(vm, true);
448
449 *retvm = vm;
450 return (0);
451 }
452
453 static void
454 vm_cleanup(struct vm *vm, bool destroy)
455 {
456 struct mem_map *mm;
457 int i;
458
459 ppt_unassign_all(vm);
460
461 if (vm->iommu != NULL)
462 iommu_destroy_domain(vm->iommu);
463
464 if (destroy)
465 vrtc_cleanup(vm->vrtc);
466 else
467 vrtc_reset(vm->vrtc);
468 vpmtmr_cleanup(vm->vpmtmr);
469 vatpit_cleanup(vm->vatpit);
470 vhpet_cleanup(vm->vhpet);
471 vatpic_cleanup(vm->vatpic);
472 vioapic_cleanup(vm->vioapic);
473
474 for (i = 0; i < VM_MAXCPU; i++)
475 vcpu_cleanup(vm, i, destroy);
476
477 VMCLEANUP(vm->cookie);
478
479 /*
480 * System memory is removed from the guest address space only when
481 * the VM is destroyed. This is because the mapping remains the same
482 * across VM reset.
483 *
484 * Device memory can be relocated by the guest (e.g. using PCI BARs)
485 * so those mappings are removed on a VM reset.
486 */
487 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
488 mm = &vm->mem_maps[i];
489 if (destroy || !sysmem_mapping(vm, mm))
490 vm_free_memmap(vm, i);
491 }
492
493 if (destroy) {
494 for (i = 0; i < VM_MAX_MEMSEGS; i++)
495 vm_free_memseg(vm, i);
496
497 VMSPACE_FREE(vm->vmspace);
498 vm->vmspace = NULL;
499 }
500 }
501
502 void
503 vm_destroy(struct vm *vm)
504 {
505 vm_cleanup(vm, true);
506 free(vm, M_VM);
507 }
508
509 int
510 vm_reinit(struct vm *vm)
511 {
512 int error;
513
514 /*
515 * A virtual machine can be reset only if all vcpus are suspended.
516 */
517 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
518 vm_cleanup(vm, false);
519 vm_init(vm, false);
520 error = 0;
521 } else {
522 error = EBUSY;
523 }
524
525 return (error);
526 }
527
528 const char *
529 vm_name(struct vm *vm)
530 {
531 return (vm->name);
532 }
533
534 int
535 vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
536 {
537 vm_object_t obj;
538
539 if ((obj = vmm_mmio_alloc(vm->vmspace, gpa, len, hpa)) == NULL)
540 return (ENOMEM);
541 else
542 return (0);
543 }
544
545 int
546 vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
547 {
548
549 vmm_mmio_free(vm->vmspace, gpa, len);
550 return (0);
551 }
552
553 /*
554 * Return 'true' if 'gpa' is allocated in the guest address space.
555 *
556 * This function is called in the context of a running vcpu which acts as
557 * an implicit lock on 'vm->mem_maps[]'.
558 */
559 bool
560 vm_mem_allocated(struct vm *vm, int vcpuid, vm_paddr_t gpa)
561 {
562 struct mem_map *mm;
563 int i;
564
565 #ifdef INVARIANTS
566 int hostcpu, state;
567 state = vcpu_get_state(vm, vcpuid, &hostcpu);
568 KASSERT(state == VCPU_RUNNING && hostcpu == curcpu,
569 ("%s: invalid vcpu state %d/%d", __func__, state, hostcpu));
570 #endif
571
572 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
573 mm = &vm->mem_maps[i];
574 if (mm->len != 0 && gpa >= mm->gpa && gpa < mm->gpa + mm->len)
575 return (true); /* 'gpa' is sysmem or devmem */
576 }
577
578 if (ppt_is_mmio(vm, gpa))
579 return (true); /* 'gpa' is pci passthru mmio */
580
581 return (false);
582 }
583
584 int
585 vm_alloc_memseg(struct vm *vm, int ident, size_t len, bool sysmem)
586 {
587 struct mem_seg *seg;
588 vm_object_t obj;
589
590 if (ident < 0 || ident >= VM_MAX_MEMSEGS)
591 return (EINVAL);
592
593 if (len == 0 || (len & PAGE_MASK))
594 return (EINVAL);
595
596 seg = &vm->mem_segs[ident];
597 if (seg->object != NULL) {
598 if (seg->len == len && seg->sysmem == sysmem)
599 return (EEXIST);
600 else
601 return (EINVAL);
602 }
603
604 obj = vm_object_allocate(OBJT_DEFAULT, len >> PAGE_SHIFT);
605 if (obj == NULL)
606 return (ENOMEM);
607
608 seg->len = len;
609 seg->object = obj;
610 seg->sysmem = sysmem;
611 return (0);
612 }
613
614 int
615 vm_get_memseg(struct vm *vm, int ident, size_t *len, bool *sysmem,
616 vm_object_t *objptr)
617 {
618 struct mem_seg *seg;
619
620 if (ident < 0 || ident >= VM_MAX_MEMSEGS)
621 return (EINVAL);
622
623 seg = &vm->mem_segs[ident];
624 if (len)
625 *len = seg->len;
626 if (sysmem)
627 *sysmem = seg->sysmem;
628 if (objptr)
629 *objptr = seg->object;
630 return (0);
631 }
632
633 void
634 vm_free_memseg(struct vm *vm, int ident)
635 {
636 struct mem_seg *seg;
637
638 KASSERT(ident >= 0 && ident < VM_MAX_MEMSEGS,
639 ("%s: invalid memseg ident %d", __func__, ident));
640
641 seg = &vm->mem_segs[ident];
642 if (seg->object != NULL) {
643 vm_object_deallocate(seg->object);
644 bzero(seg, sizeof(struct mem_seg));
645 }
646 }
647
648 int
649 vm_mmap_memseg(struct vm *vm, vm_paddr_t gpa, int segid, vm_ooffset_t first,
650 size_t len, int prot, int flags)
651 {
652 struct mem_seg *seg;
653 struct mem_map *m, *map;
654 vm_ooffset_t last;
655 int i, error;
656
657 if (prot == 0 || (prot & ~(VM_PROT_ALL)) != 0)
658 return (EINVAL);
659
660 if (flags & ~VM_MEMMAP_F_WIRED)
661 return (EINVAL);
662
663 if (segid < 0 || segid >= VM_MAX_MEMSEGS)
664 return (EINVAL);
665
666 seg = &vm->mem_segs[segid];
667 if (seg->object == NULL)
668 return (EINVAL);
669
670 last = first + len;
671 if (first < 0 || first >= last || last > seg->len)
672 return (EINVAL);
673
674 if ((gpa | first | last) & PAGE_MASK)
675 return (EINVAL);
676
677 map = NULL;
678 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
679 m = &vm->mem_maps[i];
680 if (m->len == 0) {
681 map = m;
682 break;
683 }
684 }
685
686 if (map == NULL)
687 return (ENOSPC);
688
689 error = vm_map_find(&vm->vmspace->vm_map, seg->object, first, &gpa,
690 len, 0, VMFS_NO_SPACE, prot, prot, 0);
691 if (error != KERN_SUCCESS)
692 return (EFAULT);
693
694 vm_object_reference(seg->object);
695
696 if (flags & VM_MEMMAP_F_WIRED) {
697 error = vm_map_wire(&vm->vmspace->vm_map, gpa, gpa + len,
698 VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
699 if (error != KERN_SUCCESS) {
700 vm_map_remove(&vm->vmspace->vm_map, gpa, gpa + len);
701 return (EFAULT);
702 }
703 }
704
705 map->gpa = gpa;
706 map->len = len;
707 map->segoff = first;
708 map->segid = segid;
709 map->prot = prot;
710 map->flags = flags;
711 return (0);
712 }
713
714 int
715 vm_mmap_getnext(struct vm *vm, vm_paddr_t *gpa, int *segid,
716 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags)
717 {
718 struct mem_map *mm, *mmnext;
719 int i;
720
721 mmnext = NULL;
722 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
723 mm = &vm->mem_maps[i];
724 if (mm->len == 0 || mm->gpa < *gpa)
725 continue;
726 if (mmnext == NULL || mm->gpa < mmnext->gpa)
727 mmnext = mm;
728 }
729
730 if (mmnext != NULL) {
731 *gpa = mmnext->gpa;
732 if (segid)
733 *segid = mmnext->segid;
734 if (segoff)
735 *segoff = mmnext->segoff;
736 if (len)
737 *len = mmnext->len;
738 if (prot)
739 *prot = mmnext->prot;
740 if (flags)
741 *flags = mmnext->flags;
742 return (0);
743 } else {
744 return (ENOENT);
745 }
746 }
747
748 static void
749 vm_free_memmap(struct vm *vm, int ident)
750 {
751 struct mem_map *mm;
752 int error;
753
754 mm = &vm->mem_maps[ident];
755 if (mm->len) {
756 error = vm_map_remove(&vm->vmspace->vm_map, mm->gpa,
757 mm->gpa + mm->len);
758 KASSERT(error == KERN_SUCCESS, ("%s: vm_map_remove error %d",
759 __func__, error));
760 bzero(mm, sizeof(struct mem_map));
761 }
762 }
763
764 static __inline bool
765 sysmem_mapping(struct vm *vm, struct mem_map *mm)
766 {
767
768 if (mm->len != 0 && vm->mem_segs[mm->segid].sysmem)
769 return (true);
770 else
771 return (false);
772 }
773
774 static vm_paddr_t
775 sysmem_maxaddr(struct vm *vm)
776 {
777 struct mem_map *mm;
778 vm_paddr_t maxaddr;
779 int i;
780
781 maxaddr = 0;
782 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
783 mm = &vm->mem_maps[i];
784 if (sysmem_mapping(vm, mm)) {
785 if (maxaddr < mm->gpa + mm->len)
786 maxaddr = mm->gpa + mm->len;
787 }
788 }
789 return (maxaddr);
790 }
791
792 static void
793 vm_iommu_modify(struct vm *vm, boolean_t map)
794 {
795 int i, sz;
796 vm_paddr_t gpa, hpa;
797 struct mem_map *mm;
798 void *vp, *cookie, *host_domain;
799
800 sz = PAGE_SIZE;
801 host_domain = iommu_host_domain();
802
803 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
804 mm = &vm->mem_maps[i];
805 if (!sysmem_mapping(vm, mm))
806 continue;
807
808 if (map) {
809 KASSERT((mm->flags & VM_MEMMAP_F_IOMMU) == 0,
810 ("iommu map found invalid memmap %#lx/%#lx/%#x",
811 mm->gpa, mm->len, mm->flags));
812 if ((mm->flags & VM_MEMMAP_F_WIRED) == 0)
813 continue;
814 mm->flags |= VM_MEMMAP_F_IOMMU;
815 } else {
816 if ((mm->flags & VM_MEMMAP_F_IOMMU) == 0)
817 continue;
818 mm->flags &= ~VM_MEMMAP_F_IOMMU;
819 KASSERT((mm->flags & VM_MEMMAP_F_WIRED) != 0,
820 ("iommu unmap found invalid memmap %#lx/%#lx/%#x",
821 mm->gpa, mm->len, mm->flags));
822 }
823
824 gpa = mm->gpa;
825 while (gpa < mm->gpa + mm->len) {
826 vp = vm_gpa_hold(vm, -1, gpa, PAGE_SIZE, VM_PROT_WRITE,
827 &cookie);
828 KASSERT(vp != NULL, ("vm(%s) could not map gpa %#lx",
829 vm_name(vm), gpa));
830
831 vm_gpa_release(cookie);
832
833 hpa = DMAP_TO_PHYS((uintptr_t)vp);
834 if (map) {
835 iommu_create_mapping(vm->iommu, gpa, hpa, sz);
836 iommu_remove_mapping(host_domain, hpa, sz);
837 } else {
838 iommu_remove_mapping(vm->iommu, gpa, sz);
839 iommu_create_mapping(host_domain, hpa, hpa, sz);
840 }
841
842 gpa += PAGE_SIZE;
843 }
844 }
845
846 /*
847 * Invalidate the cached translations associated with the domain
848 * from which pages were removed.
849 */
850 if (map)
851 iommu_invalidate_tlb(host_domain);
852 else
853 iommu_invalidate_tlb(vm->iommu);
854 }
855
856 #define vm_iommu_unmap(vm) vm_iommu_modify((vm), FALSE)
857 #define vm_iommu_map(vm) vm_iommu_modify((vm), TRUE)
858
859 int
860 vm_unassign_pptdev(struct vm *vm, int bus, int slot, int func)
861 {
862 int error;
863
864 error = ppt_unassign_device(vm, bus, slot, func);
865 if (error)
866 return (error);
867
868 if (ppt_assigned_devices(vm) == 0)
869 vm_iommu_unmap(vm);
870
871 return (0);
872 }
873
874 int
875 vm_assign_pptdev(struct vm *vm, int bus, int slot, int func)
876 {
877 int error;
878 vm_paddr_t maxaddr;
879
880 /* Set up the IOMMU to do the 'gpa' to 'hpa' translation */
881 if (ppt_assigned_devices(vm) == 0) {
882 KASSERT(vm->iommu == NULL,
883 ("vm_assign_pptdev: iommu must be NULL"));
884 maxaddr = sysmem_maxaddr(vm);
885 vm->iommu = iommu_create_domain(maxaddr);
886 if (vm->iommu == NULL)
887 return (ENXIO);
888 vm_iommu_map(vm);
889 }
890
891 error = ppt_assign_device(vm, bus, slot, func);
892 return (error);
893 }
894
895 void *
896 vm_gpa_hold(struct vm *vm, int vcpuid, vm_paddr_t gpa, size_t len, int reqprot,
897 void **cookie)
898 {
899 int i, count, pageoff;
900 struct mem_map *mm;
901 vm_page_t m;
902 #ifdef INVARIANTS
903 /*
904 * All vcpus are frozen by ioctls that modify the memory map
905 * (e.g. VM_MMAP_MEMSEG). Therefore 'vm->memmap[]' stability is
906 * guaranteed if at least one vcpu is in the VCPU_FROZEN state.
907 */
908 int state;
909 KASSERT(vcpuid >= -1 && vcpuid < VM_MAXCPU, ("%s: invalid vcpuid %d",
910 __func__, vcpuid));
911 for (i = 0; i < VM_MAXCPU; i++) {
912 if (vcpuid != -1 && vcpuid != i)
913 continue;
914 state = vcpu_get_state(vm, i, NULL);
915 KASSERT(state == VCPU_FROZEN, ("%s: invalid vcpu state %d",
916 __func__, state));
917 }
918 #endif
919 pageoff = gpa & PAGE_MASK;
920 if (len > PAGE_SIZE - pageoff)
921 panic("vm_gpa_hold: invalid gpa/len: 0x%016lx/%lu", gpa, len);
922
923 count = 0;
924 for (i = 0; i < VM_MAX_MEMMAPS; i++) {
925 mm = &vm->mem_maps[i];
926 if (sysmem_mapping(vm, mm) && gpa >= mm->gpa &&
927 gpa < mm->gpa + mm->len) {
928 count = vm_fault_quick_hold_pages(&vm->vmspace->vm_map,
929 trunc_page(gpa), PAGE_SIZE, reqprot, &m, 1);
930 break;
931 }
932 }
933
934 if (count == 1) {
935 *cookie = m;
936 return ((void *)(PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)) + pageoff));
937 } else {
938 *cookie = NULL;
939 return (NULL);
940 }
941 }
942
943 void
944 vm_gpa_release(void *cookie)
945 {
946 vm_page_t m = cookie;
947
948 vm_page_lock(m);
949 vm_page_unhold(m);
950 vm_page_unlock(m);
951 }
952
953 int
954 vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
955 {
956
957 if (vcpu < 0 || vcpu >= VM_MAXCPU)
958 return (EINVAL);
959
960 if (reg >= VM_REG_LAST)
961 return (EINVAL);
962
963 return (VMGETREG(vm->cookie, vcpu, reg, retval));
964 }
965
966 int
967 vm_set_register(struct vm *vm, int vcpuid, int reg, uint64_t val)
968 {
969 struct vcpu *vcpu;
970 int error;
971
972 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
973 return (EINVAL);
974
975 if (reg >= VM_REG_LAST)
976 return (EINVAL);
977
978 error = VMSETREG(vm->cookie, vcpuid, reg, val);
979 if (error || reg != VM_REG_GUEST_RIP)
980 return (error);
981
982 /* Set 'nextrip' to match the value of %rip */
983 VCPU_CTR1(vm, vcpuid, "Setting nextrip to %#lx", val);
984 vcpu = &vm->vcpu[vcpuid];
985 vcpu->nextrip = val;
986 return (0);
987 }
988
989 static boolean_t
990 is_descriptor_table(int reg)
991 {
992
993 switch (reg) {
994 case VM_REG_GUEST_IDTR:
995 case VM_REG_GUEST_GDTR:
996 return (TRUE);
997 default:
998 return (FALSE);
999 }
1000 }
1001
1002 static boolean_t
1003 is_segment_register(int reg)
1004 {
1005
1006 switch (reg) {
1007 case VM_REG_GUEST_ES:
1008 case VM_REG_GUEST_CS:
1009 case VM_REG_GUEST_SS:
1010 case VM_REG_GUEST_DS:
1011 case VM_REG_GUEST_FS:
1012 case VM_REG_GUEST_GS:
1013 case VM_REG_GUEST_TR:
1014 case VM_REG_GUEST_LDTR:
1015 return (TRUE);
1016 default:
1017 return (FALSE);
1018 }
1019 }
1020
1021 int
1022 vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
1023 struct seg_desc *desc)
1024 {
1025
1026 if (vcpu < 0 || vcpu >= VM_MAXCPU)
1027 return (EINVAL);
1028
1029 if (!is_segment_register(reg) && !is_descriptor_table(reg))
1030 return (EINVAL);
1031
1032 return (VMGETDESC(vm->cookie, vcpu, reg, desc));
1033 }
1034
1035 int
1036 vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
1037 struct seg_desc *desc)
1038 {
1039 if (vcpu < 0 || vcpu >= VM_MAXCPU)
1040 return (EINVAL);
1041
1042 if (!is_segment_register(reg) && !is_descriptor_table(reg))
1043 return (EINVAL);
1044
1045 return (VMSETDESC(vm->cookie, vcpu, reg, desc));
1046 }
1047
1048 static void
1049 restore_guest_fpustate(struct vcpu *vcpu)
1050 {
1051
1052 /* flush host state to the pcb */
1053 fpuexit(curthread);
1054
1055 /* restore guest FPU state */
1056 fpu_stop_emulating();
1057 fpurestore(vcpu->guestfpu);
1058
1059 /* restore guest XCR0 if XSAVE is enabled in the host */
1060 if (rcr4() & CR4_XSAVE)
1061 load_xcr(0, vcpu->guest_xcr0);
1062
1063 /*
1064 * The FPU is now "dirty" with the guest's state so turn on emulation
1065 * to trap any access to the FPU by the host.
1066 */
1067 fpu_start_emulating();
1068 }
1069
1070 static void
1071 save_guest_fpustate(struct vcpu *vcpu)
1072 {
1073
1074 if ((rcr0() & CR0_TS) == 0)
1075 panic("fpu emulation not enabled in host!");
1076
1077 /* save guest XCR0 and restore host XCR0 */
1078 if (rcr4() & CR4_XSAVE) {
1079 vcpu->guest_xcr0 = rxcr(0);
1080 load_xcr(0, vmm_get_host_xcr0());
1081 }
1082
1083 /* save guest FPU state */
1084 fpu_stop_emulating();
1085 fpusave(vcpu->guestfpu);
1086 fpu_start_emulating();
1087 }
1088
1089 static VMM_STAT(VCPU_IDLE_TICKS, "number of ticks vcpu was idle");
1090
1091 static int
1092 vcpu_set_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate,
1093 bool from_idle)
1094 {
1095 struct vcpu *vcpu;
1096 int error;
1097
1098 vcpu = &vm->vcpu[vcpuid];
1099 vcpu_assert_locked(vcpu);
1100
1101 /*
1102 * State transitions from the vmmdev_ioctl() must always begin from
1103 * the VCPU_IDLE state. This guarantees that there is only a single
1104 * ioctl() operating on a vcpu at any point.
1105 */
1106 if (from_idle) {
1107 while (vcpu->state != VCPU_IDLE) {
1108 vcpu->reqidle = 1;
1109 vcpu_notify_event_locked(vcpu, false);
1110 VCPU_CTR1(vm, vcpuid, "vcpu state change from %s to "
1111 "idle requested", vcpu_state2str(vcpu->state));
1112 msleep_spin(&vcpu->state, &vcpu->mtx, "vmstat", hz);
1113 }
1114 } else {
1115 KASSERT(vcpu->state != VCPU_IDLE, ("invalid transition from "
1116 "vcpu idle state"));
1117 }
1118
1119 if (vcpu->state == VCPU_RUNNING) {
1120 KASSERT(vcpu->hostcpu == curcpu, ("curcpu %d and hostcpu %d "
1121 "mismatch for running vcpu", curcpu, vcpu->hostcpu));
1122 } else {
1123 KASSERT(vcpu->hostcpu == NOCPU, ("Invalid hostcpu %d for a "
1124 "vcpu that is not running", vcpu->hostcpu));
1125 }
1126
1127 /*
1128 * The following state transitions are allowed:
1129 * IDLE -> FROZEN -> IDLE
1130 * FROZEN -> RUNNING -> FROZEN
1131 * FROZEN -> SLEEPING -> FROZEN
1132 */
1133 switch (vcpu->state) {
1134 case VCPU_IDLE:
1135 case VCPU_RUNNING:
1136 case VCPU_SLEEPING:
1137 error = (newstate != VCPU_FROZEN);
1138 break;
1139 case VCPU_FROZEN:
1140 error = (newstate == VCPU_FROZEN);
1141 break;
1142 default:
1143 error = 1;
1144 break;
1145 }
1146
1147 if (error)
1148 return (EBUSY);
1149
1150 VCPU_CTR2(vm, vcpuid, "vcpu state changed from %s to %s",
1151 vcpu_state2str(vcpu->state), vcpu_state2str(newstate));
1152
1153 vcpu->state = newstate;
1154 if (newstate == VCPU_RUNNING)
1155 vcpu->hostcpu = curcpu;
1156 else
1157 vcpu->hostcpu = NOCPU;
1158
1159 if (newstate == VCPU_IDLE)
1160 wakeup(&vcpu->state);
1161
1162 return (0);
1163 }
1164
1165 static void
1166 vcpu_require_state(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1167 {
1168 int error;
1169
1170 if ((error = vcpu_set_state(vm, vcpuid, newstate, false)) != 0)
1171 panic("Error %d setting state to %d\n", error, newstate);
1172 }
1173
1174 static void
1175 vcpu_require_state_locked(struct vm *vm, int vcpuid, enum vcpu_state newstate)
1176 {
1177 int error;
1178
1179 if ((error = vcpu_set_state_locked(vm, vcpuid, newstate, false)) != 0)
1180 panic("Error %d setting state to %d", error, newstate);
1181 }
1182
1183 static void
1184 vm_set_rendezvous_func(struct vm *vm, vm_rendezvous_func_t func)
1185 {
1186
1187 KASSERT(mtx_owned(&vm->rendezvous_mtx), ("rendezvous_mtx not locked"));
1188
1189 /*
1190 * Update 'rendezvous_func' and execute a write memory barrier to
1191 * ensure that it is visible across all host cpus. This is not needed
1192 * for correctness but it does ensure that all the vcpus will notice
1193 * that the rendezvous is requested immediately.
1194 */
1195 vm->rendezvous_func = func;
1196 wmb();
1197 }
1198
1199 #define RENDEZVOUS_CTR0(vm, vcpuid, fmt) \
1200 do { \
1201 if (vcpuid >= 0) \
1202 VCPU_CTR0(vm, vcpuid, fmt); \
1203 else \
1204 VM_CTR0(vm, fmt); \
1205 } while (0)
1206
1207 static void
1208 vm_handle_rendezvous(struct vm *vm, int vcpuid)
1209 {
1210
1211 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
1212 ("vm_handle_rendezvous: invalid vcpuid %d", vcpuid));
1213
1214 mtx_lock(&vm->rendezvous_mtx);
1215 while (vm->rendezvous_func != NULL) {
1216 /* 'rendezvous_req_cpus' must be a subset of 'active_cpus' */
1217 CPU_AND(&vm->rendezvous_req_cpus, &vm->active_cpus);
1218
1219 if (vcpuid != -1 &&
1220 CPU_ISSET(vcpuid, &vm->rendezvous_req_cpus) &&
1221 !CPU_ISSET(vcpuid, &vm->rendezvous_done_cpus)) {
1222 VCPU_CTR0(vm, vcpuid, "Calling rendezvous func");
1223 (*vm->rendezvous_func)(vm, vcpuid, vm->rendezvous_arg);
1224 CPU_SET(vcpuid, &vm->rendezvous_done_cpus);
1225 }
1226 if (CPU_CMP(&vm->rendezvous_req_cpus,
1227 &vm->rendezvous_done_cpus) == 0) {
1228 VCPU_CTR0(vm, vcpuid, "Rendezvous completed");
1229 vm_set_rendezvous_func(vm, NULL);
1230 wakeup(&vm->rendezvous_func);
1231 break;
1232 }
1233 RENDEZVOUS_CTR0(vm, vcpuid, "Wait for rendezvous completion");
1234 mtx_sleep(&vm->rendezvous_func, &vm->rendezvous_mtx, 0,
1235 "vmrndv", 0);
1236 }
1237 mtx_unlock(&vm->rendezvous_mtx);
1238 }
1239
1240 /*
1241 * Emulate a guest 'hlt' by sleeping until the vcpu is ready to run.
1242 */
1243 static int
1244 vm_handle_hlt(struct vm *vm, int vcpuid, bool intr_disabled, bool *retu)
1245 {
1246 struct vcpu *vcpu;
1247 const char *wmesg;
1248 int t, vcpu_halted, vm_halted;
1249
1250 KASSERT(!CPU_ISSET(vcpuid, &vm->halted_cpus), ("vcpu already halted"));
1251
1252 vcpu = &vm->vcpu[vcpuid];
1253 vcpu_halted = 0;
1254 vm_halted = 0;
1255
1256 vcpu_lock(vcpu);
1257 while (1) {
1258 /*
1259 * Do a final check for pending NMI or interrupts before
1260 * really putting this thread to sleep. Also check for
1261 * software events that would cause this vcpu to wakeup.
1262 *
1263 * These interrupts/events could have happened after the
1264 * vcpu returned from VMRUN() and before it acquired the
1265 * vcpu lock above.
1266 */
1267 if (vm->rendezvous_func != NULL || vm->suspend || vcpu->reqidle)
1268 break;
1269 if (vm_nmi_pending(vm, vcpuid))
1270 break;
1271 if (!intr_disabled) {
1272 if (vm_extint_pending(vm, vcpuid) ||
1273 vlapic_pending_intr(vcpu->vlapic, NULL)) {
1274 break;
1275 }
1276 }
1277
1278 /* Don't go to sleep if the vcpu thread needs to yield */
1279 if (vcpu_should_yield(vm, vcpuid))
1280 break;
1281
1282 /*
1283 * Some Linux guests implement "halt" by having all vcpus
1284 * execute HLT with interrupts disabled. 'halted_cpus' keeps
1285 * track of the vcpus that have entered this state. When all
1286 * vcpus enter the halted state the virtual machine is halted.
1287 */
1288 if (intr_disabled) {
1289 wmesg = "vmhalt";
1290 VCPU_CTR0(vm, vcpuid, "Halted");
1291 if (!vcpu_halted && halt_detection_enabled) {
1292 vcpu_halted = 1;
1293 CPU_SET_ATOMIC(vcpuid, &vm->halted_cpus);
1294 }
1295 if (CPU_CMP(&vm->halted_cpus, &vm->active_cpus) == 0) {
1296 vm_halted = 1;
1297 break;
1298 }
1299 } else {
1300 wmesg = "vmidle";
1301 }
1302
1303 t = ticks;
1304 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1305 /*
1306 * XXX msleep_spin() cannot be interrupted by signals so
1307 * wake up periodically to check pending signals.
1308 */
1309 msleep_spin(vcpu, &vcpu->mtx, wmesg, hz);
1310 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1311 vmm_stat_incr(vm, vcpuid, VCPU_IDLE_TICKS, ticks - t);
1312 }
1313
1314 if (vcpu_halted)
1315 CPU_CLR_ATOMIC(vcpuid, &vm->halted_cpus);
1316
1317 vcpu_unlock(vcpu);
1318
1319 if (vm_halted)
1320 vm_suspend(vm, VM_SUSPEND_HALT);
1321
1322 return (0);
1323 }
1324
1325 static int
1326 vm_handle_paging(struct vm *vm, int vcpuid, bool *retu)
1327 {
1328 int rv, ftype;
1329 struct vm_map *map;
1330 struct vcpu *vcpu;
1331 struct vm_exit *vme;
1332
1333 vcpu = &vm->vcpu[vcpuid];
1334 vme = &vcpu->exitinfo;
1335
1336 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1337 __func__, vme->inst_length));
1338
1339 ftype = vme->u.paging.fault_type;
1340 KASSERT(ftype == VM_PROT_READ ||
1341 ftype == VM_PROT_WRITE || ftype == VM_PROT_EXECUTE,
1342 ("vm_handle_paging: invalid fault_type %d", ftype));
1343
1344 if (ftype == VM_PROT_READ || ftype == VM_PROT_WRITE) {
1345 rv = pmap_emulate_accessed_dirty(vmspace_pmap(vm->vmspace),
1346 vme->u.paging.gpa, ftype);
1347 if (rv == 0) {
1348 VCPU_CTR2(vm, vcpuid, "%s bit emulation for gpa %#lx",
1349 ftype == VM_PROT_READ ? "accessed" : "dirty",
1350 vme->u.paging.gpa);
1351 goto done;
1352 }
1353 }
1354
1355 map = &vm->vmspace->vm_map;
1356 rv = vm_fault(map, vme->u.paging.gpa, ftype, VM_FAULT_NORMAL);
1357
1358 VCPU_CTR3(vm, vcpuid, "vm_handle_paging rv = %d, gpa = %#lx, "
1359 "ftype = %d", rv, vme->u.paging.gpa, ftype);
1360
1361 if (rv != KERN_SUCCESS)
1362 return (EFAULT);
1363 done:
1364 return (0);
1365 }
1366
1367 static int
1368 vm_handle_inst_emul(struct vm *vm, int vcpuid, bool *retu)
1369 {
1370 struct vie *vie;
1371 struct vcpu *vcpu;
1372 struct vm_exit *vme;
1373 uint64_t gla, gpa, cs_base;
1374 struct vm_guest_paging *paging;
1375 mem_region_read_t mread;
1376 mem_region_write_t mwrite;
1377 enum vm_cpu_mode cpu_mode;
1378 int cs_d, error, fault;
1379
1380 vcpu = &vm->vcpu[vcpuid];
1381 vme = &vcpu->exitinfo;
1382
1383 KASSERT(vme->inst_length == 0, ("%s: invalid inst_length %d",
1384 __func__, vme->inst_length));
1385
1386 gla = vme->u.inst_emul.gla;
1387 gpa = vme->u.inst_emul.gpa;
1388 cs_base = vme->u.inst_emul.cs_base;
1389 cs_d = vme->u.inst_emul.cs_d;
1390 vie = &vme->u.inst_emul.vie;
1391 paging = &vme->u.inst_emul.paging;
1392 cpu_mode = paging->cpu_mode;
1393
1394 VCPU_CTR1(vm, vcpuid, "inst_emul fault accessing gpa %#lx", gpa);
1395
1396 /* Fetch, decode and emulate the faulting instruction */
1397 if (vie->num_valid == 0) {
1398 error = vmm_fetch_instruction(vm, vcpuid, paging, vme->rip +
1399 cs_base, VIE_INST_SIZE, vie, &fault);
1400 } else {
1401 /*
1402 * The instruction bytes have already been copied into 'vie'
1403 */
1404 error = fault = 0;
1405 }
1406 if (error || fault)
1407 return (error);
1408
1409 if (vmm_decode_instruction(vm, vcpuid, gla, cpu_mode, cs_d, vie) != 0) {
1410 VCPU_CTR1(vm, vcpuid, "Error decoding instruction at %#lx",
1411 vme->rip + cs_base);
1412 *retu = true; /* dump instruction bytes in userspace */
1413 return (0);
1414 }
1415
1416 /*
1417 * Update 'nextrip' based on the length of the emulated instruction.
1418 */
1419 vme->inst_length = vie->num_processed;
1420 vcpu->nextrip += vie->num_processed;
1421 VCPU_CTR1(vm, vcpuid, "nextrip updated to %#lx after instruction "
1422 "decoding", vcpu->nextrip);
1423
1424 /* return to userland unless this is an in-kernel emulated device */
1425 if (gpa >= DEFAULT_APIC_BASE && gpa < DEFAULT_APIC_BASE + PAGE_SIZE) {
1426 mread = lapic_mmio_read;
1427 mwrite = lapic_mmio_write;
1428 } else if (gpa >= VIOAPIC_BASE && gpa < VIOAPIC_BASE + VIOAPIC_SIZE) {
1429 mread = vioapic_mmio_read;
1430 mwrite = vioapic_mmio_write;
1431 } else if (gpa >= VHPET_BASE && gpa < VHPET_BASE + VHPET_SIZE) {
1432 mread = vhpet_mmio_read;
1433 mwrite = vhpet_mmio_write;
1434 } else {
1435 *retu = true;
1436 return (0);
1437 }
1438
1439 error = vmm_emulate_instruction(vm, vcpuid, gpa, vie, paging,
1440 mread, mwrite, retu);
1441
1442 return (error);
1443 }
1444
1445 static int
1446 vm_handle_suspend(struct vm *vm, int vcpuid, bool *retu)
1447 {
1448 int i, done;
1449 struct vcpu *vcpu;
1450
1451 done = 0;
1452 vcpu = &vm->vcpu[vcpuid];
1453
1454 CPU_SET_ATOMIC(vcpuid, &vm->suspended_cpus);
1455
1456 /*
1457 * Wait until all 'active_cpus' have suspended themselves.
1458 *
1459 * Since a VM may be suspended at any time including when one or
1460 * more vcpus are doing a rendezvous we need to call the rendezvous
1461 * handler while we are waiting to prevent a deadlock.
1462 */
1463 vcpu_lock(vcpu);
1464 while (1) {
1465 if (CPU_CMP(&vm->suspended_cpus, &vm->active_cpus) == 0) {
1466 VCPU_CTR0(vm, vcpuid, "All vcpus suspended");
1467 break;
1468 }
1469
1470 if (vm->rendezvous_func == NULL) {
1471 VCPU_CTR0(vm, vcpuid, "Sleeping during suspend");
1472 vcpu_require_state_locked(vm, vcpuid, VCPU_SLEEPING);
1473 msleep_spin(vcpu, &vcpu->mtx, "vmsusp", hz);
1474 vcpu_require_state_locked(vm, vcpuid, VCPU_FROZEN);
1475 } else {
1476 VCPU_CTR0(vm, vcpuid, "Rendezvous during suspend");
1477 vcpu_unlock(vcpu);
1478 vm_handle_rendezvous(vm, vcpuid);
1479 vcpu_lock(vcpu);
1480 }
1481 }
1482 vcpu_unlock(vcpu);
1483
1484 /*
1485 * Wakeup the other sleeping vcpus and return to userspace.
1486 */
1487 for (i = 0; i < VM_MAXCPU; i++) {
1488 if (CPU_ISSET(i, &vm->suspended_cpus)) {
1489 vcpu_notify_event(vm, i, false);
1490 }
1491 }
1492
1493 *retu = true;
1494 return (0);
1495 }
1496
1497 static int
1498 vm_handle_reqidle(struct vm *vm, int vcpuid, bool *retu)
1499 {
1500 struct vcpu *vcpu = &vm->vcpu[vcpuid];
1501
1502 vcpu_lock(vcpu);
1503 KASSERT(vcpu->reqidle, ("invalid vcpu reqidle %d", vcpu->reqidle));
1504 vcpu->reqidle = 0;
1505 vcpu_unlock(vcpu);
1506 *retu = true;
1507 return (0);
1508 }
1509
1510 int
1511 vm_suspend(struct vm *vm, enum vm_suspend_how how)
1512 {
1513 int i;
1514
1515 if (how <= VM_SUSPEND_NONE || how >= VM_SUSPEND_LAST)
1516 return (EINVAL);
1517
1518 if (atomic_cmpset_int(&vm->suspend, 0, how) == 0) {
1519 VM_CTR2(vm, "virtual machine already suspended %d/%d",
1520 vm->suspend, how);
1521 return (EALREADY);
1522 }
1523
1524 VM_CTR1(vm, "virtual machine successfully suspended %d", how);
1525
1526 /*
1527 * Notify all active vcpus that they are now suspended.
1528 */
1529 for (i = 0; i < VM_MAXCPU; i++) {
1530 if (CPU_ISSET(i, &vm->active_cpus))
1531 vcpu_notify_event(vm, i, false);
1532 }
1533
1534 return (0);
1535 }
1536
1537 void
1538 vm_exit_suspended(struct vm *vm, int vcpuid, uint64_t rip)
1539 {
1540 struct vm_exit *vmexit;
1541
1542 KASSERT(vm->suspend > VM_SUSPEND_NONE && vm->suspend < VM_SUSPEND_LAST,
1543 ("vm_exit_suspended: invalid suspend type %d", vm->suspend));
1544
1545 vmexit = vm_exitinfo(vm, vcpuid);
1546 vmexit->rip = rip;
1547 vmexit->inst_length = 0;
1548 vmexit->exitcode = VM_EXITCODE_SUSPENDED;
1549 vmexit->u.suspended.how = vm->suspend;
1550 }
1551
1552 void
1553 vm_exit_rendezvous(struct vm *vm, int vcpuid, uint64_t rip)
1554 {
1555 struct vm_exit *vmexit;
1556
1557 KASSERT(vm->rendezvous_func != NULL, ("rendezvous not in progress"));
1558
1559 vmexit = vm_exitinfo(vm, vcpuid);
1560 vmexit->rip = rip;
1561 vmexit->inst_length = 0;
1562 vmexit->exitcode = VM_EXITCODE_RENDEZVOUS;
1563 vmm_stat_incr(vm, vcpuid, VMEXIT_RENDEZVOUS, 1);
1564 }
1565
1566 void
1567 vm_exit_reqidle(struct vm *vm, int vcpuid, uint64_t rip)
1568 {
1569 struct vm_exit *vmexit;
1570
1571 vmexit = vm_exitinfo(vm, vcpuid);
1572 vmexit->rip = rip;
1573 vmexit->inst_length = 0;
1574 vmexit->exitcode = VM_EXITCODE_REQIDLE;
1575 vmm_stat_incr(vm, vcpuid, VMEXIT_REQIDLE, 1);
1576 }
1577
1578 void
1579 vm_exit_astpending(struct vm *vm, int vcpuid, uint64_t rip)
1580 {
1581 struct vm_exit *vmexit;
1582
1583 vmexit = vm_exitinfo(vm, vcpuid);
1584 vmexit->rip = rip;
1585 vmexit->inst_length = 0;
1586 vmexit->exitcode = VM_EXITCODE_BOGUS;
1587 vmm_stat_incr(vm, vcpuid, VMEXIT_ASTPENDING, 1);
1588 }
1589
1590 int
1591 vm_run(struct vm *vm, struct vm_run *vmrun)
1592 {
1593 struct vm_eventinfo evinfo;
1594 int error, vcpuid;
1595 struct vcpu *vcpu;
1596 struct pcb *pcb;
1597 uint64_t tscval;
1598 struct vm_exit *vme;
1599 bool retu, intr_disabled;
1600 pmap_t pmap;
1601
1602 vcpuid = vmrun->cpuid;
1603
1604 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1605 return (EINVAL);
1606
1607 if (!CPU_ISSET(vcpuid, &vm->active_cpus))
1608 return (EINVAL);
1609
1610 if (CPU_ISSET(vcpuid, &vm->suspended_cpus))
1611 return (EINVAL);
1612
1613 pmap = vmspace_pmap(vm->vmspace);
1614 vcpu = &vm->vcpu[vcpuid];
1615 vme = &vcpu->exitinfo;
1616 evinfo.rptr = &vm->rendezvous_func;
1617 evinfo.sptr = &vm->suspend;
1618 evinfo.iptr = &vcpu->reqidle;
1619 restart:
1620 critical_enter();
1621
1622 KASSERT(!CPU_ISSET(curcpu, &pmap->pm_active),
1623 ("vm_run: absurd pm_active"));
1624
1625 tscval = rdtsc();
1626
1627 pcb = PCPU_GET(curpcb);
1628 set_pcb_flags(pcb, PCB_FULL_IRET);
1629
1630 restore_guest_fpustate(vcpu);
1631
1632 vcpu_require_state(vm, vcpuid, VCPU_RUNNING);
1633 error = VMRUN(vm->cookie, vcpuid, vcpu->nextrip, pmap, &evinfo);
1634 vcpu_require_state(vm, vcpuid, VCPU_FROZEN);
1635
1636 save_guest_fpustate(vcpu);
1637
1638 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
1639
1640 critical_exit();
1641
1642 if (error == 0) {
1643 retu = false;
1644 vcpu->nextrip = vme->rip + vme->inst_length;
1645 switch (vme->exitcode) {
1646 case VM_EXITCODE_REQIDLE:
1647 error = vm_handle_reqidle(vm, vcpuid, &retu);
1648 break;
1649 case VM_EXITCODE_SUSPENDED:
1650 error = vm_handle_suspend(vm, vcpuid, &retu);
1651 break;
1652 case VM_EXITCODE_IOAPIC_EOI:
1653 vioapic_process_eoi(vm, vcpuid,
1654 vme->u.ioapic_eoi.vector);
1655 break;
1656 case VM_EXITCODE_RENDEZVOUS:
1657 vm_handle_rendezvous(vm, vcpuid);
1658 error = 0;
1659 break;
1660 case VM_EXITCODE_HLT:
1661 intr_disabled = ((vme->u.hlt.rflags & PSL_I) == 0);
1662 error = vm_handle_hlt(vm, vcpuid, intr_disabled, &retu);
1663 break;
1664 case VM_EXITCODE_PAGING:
1665 error = vm_handle_paging(vm, vcpuid, &retu);
1666 break;
1667 case VM_EXITCODE_INST_EMUL:
1668 error = vm_handle_inst_emul(vm, vcpuid, &retu);
1669 break;
1670 case VM_EXITCODE_INOUT:
1671 case VM_EXITCODE_INOUT_STR:
1672 error = vm_handle_inout(vm, vcpuid, vme, &retu);
1673 break;
1674 case VM_EXITCODE_MONITOR:
1675 case VM_EXITCODE_MWAIT:
1676 vm_inject_ud(vm, vcpuid);
1677 break;
1678 default:
1679 retu = true; /* handled in userland */
1680 break;
1681 }
1682 }
1683
1684 if (error == 0 && retu == false)
1685 goto restart;
1686
1687 VCPU_CTR2(vm, vcpuid, "retu %d/%d", error, vme->exitcode);
1688
1689 /* copy the exit information */
1690 bcopy(vme, &vmrun->vm_exit, sizeof(struct vm_exit));
1691 return (error);
1692 }
1693
1694 int
1695 vm_restart_instruction(void *arg, int vcpuid)
1696 {
1697 struct vm *vm;
1698 struct vcpu *vcpu;
1699 enum vcpu_state state;
1700 uint64_t rip;
1701 int error;
1702
1703 vm = arg;
1704 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1705 return (EINVAL);
1706
1707 vcpu = &vm->vcpu[vcpuid];
1708 state = vcpu_get_state(vm, vcpuid, NULL);
1709 if (state == VCPU_RUNNING) {
1710 /*
1711 * When a vcpu is "running" the next instruction is determined
1712 * by adding 'rip' and 'inst_length' in the vcpu's 'exitinfo'.
1713 * Thus setting 'inst_length' to zero will cause the current
1714 * instruction to be restarted.
1715 */
1716 vcpu->exitinfo.inst_length = 0;
1717 VCPU_CTR1(vm, vcpuid, "restarting instruction at %#lx by "
1718 "setting inst_length to zero", vcpu->exitinfo.rip);
1719 } else if (state == VCPU_FROZEN) {
1720 /*
1721 * When a vcpu is "frozen" it is outside the critical section
1722 * around VMRUN() and 'nextrip' points to the next instruction.
1723 * Thus instruction restart is achieved by setting 'nextrip'
1724 * to the vcpu's %rip.
1725 */
1726 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_RIP, &rip);
1727 KASSERT(!error, ("%s: error %d getting rip", __func__, error));
1728 VCPU_CTR2(vm, vcpuid, "restarting instruction by updating "
1729 "nextrip from %#lx to %#lx", vcpu->nextrip, rip);
1730 vcpu->nextrip = rip;
1731 } else {
1732 panic("%s: invalid state %d", __func__, state);
1733 }
1734 return (0);
1735 }
1736
1737 int
1738 vm_exit_intinfo(struct vm *vm, int vcpuid, uint64_t info)
1739 {
1740 struct vcpu *vcpu;
1741 int type, vector;
1742
1743 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1744 return (EINVAL);
1745
1746 vcpu = &vm->vcpu[vcpuid];
1747
1748 if (info & VM_INTINFO_VALID) {
1749 type = info & VM_INTINFO_TYPE;
1750 vector = info & 0xff;
1751 if (type == VM_INTINFO_NMI && vector != IDT_NMI)
1752 return (EINVAL);
1753 if (type == VM_INTINFO_HWEXCEPTION && vector >= 32)
1754 return (EINVAL);
1755 if (info & VM_INTINFO_RSVD)
1756 return (EINVAL);
1757 } else {
1758 info = 0;
1759 }
1760 VCPU_CTR2(vm, vcpuid, "%s: info1(%#lx)", __func__, info);
1761 vcpu->exitintinfo = info;
1762 return (0);
1763 }
1764
1765 enum exc_class {
1766 EXC_BENIGN,
1767 EXC_CONTRIBUTORY,
1768 EXC_PAGEFAULT
1769 };
1770
1771 #define IDT_VE 20 /* Virtualization Exception (Intel specific) */
1772
1773 static enum exc_class
1774 exception_class(uint64_t info)
1775 {
1776 int type, vector;
1777
1778 KASSERT(info & VM_INTINFO_VALID, ("intinfo must be valid: %#lx", info));
1779 type = info & VM_INTINFO_TYPE;
1780 vector = info & 0xff;
1781
1782 /* Table 6-4, "Interrupt and Exception Classes", Intel SDM, Vol 3 */
1783 switch (type) {
1784 case VM_INTINFO_HWINTR:
1785 case VM_INTINFO_SWINTR:
1786 case VM_INTINFO_NMI:
1787 return (EXC_BENIGN);
1788 default:
1789 /*
1790 * Hardware exception.
1791 *
1792 * SVM and VT-x use identical type values to represent NMI,
1793 * hardware interrupt and software interrupt.
1794 *
1795 * SVM uses type '3' for all exceptions. VT-x uses type '3'
1796 * for exceptions except #BP and #OF. #BP and #OF use a type
1797 * value of '5' or '6'. Therefore we don't check for explicit
1798 * values of 'type' to classify 'intinfo' into a hardware
1799 * exception.
1800 */
1801 break;
1802 }
1803
1804 switch (vector) {
1805 case IDT_PF:
1806 case IDT_VE:
1807 return (EXC_PAGEFAULT);
1808 case IDT_DE:
1809 case IDT_TS:
1810 case IDT_NP:
1811 case IDT_SS:
1812 case IDT_GP:
1813 return (EXC_CONTRIBUTORY);
1814 default:
1815 return (EXC_BENIGN);
1816 }
1817 }
1818
1819 static int
1820 nested_fault(struct vm *vm, int vcpuid, uint64_t info1, uint64_t info2,
1821 uint64_t *retinfo)
1822 {
1823 enum exc_class exc1, exc2;
1824 int type1, vector1;
1825
1826 KASSERT(info1 & VM_INTINFO_VALID, ("info1 %#lx is not valid", info1));
1827 KASSERT(info2 & VM_INTINFO_VALID, ("info2 %#lx is not valid", info2));
1828
1829 /*
1830 * If an exception occurs while attempting to call the double-fault
1831 * handler the processor enters shutdown mode (aka triple fault).
1832 */
1833 type1 = info1 & VM_INTINFO_TYPE;
1834 vector1 = info1 & 0xff;
1835 if (type1 == VM_INTINFO_HWEXCEPTION && vector1 == IDT_DF) {
1836 VCPU_CTR2(vm, vcpuid, "triple fault: info1(%#lx), info2(%#lx)",
1837 info1, info2);
1838 vm_suspend(vm, VM_SUSPEND_TRIPLEFAULT);
1839 *retinfo = 0;
1840 return (0);
1841 }
1842
1843 /*
1844 * Table 6-5 "Conditions for Generating a Double Fault", Intel SDM, Vol3
1845 */
1846 exc1 = exception_class(info1);
1847 exc2 = exception_class(info2);
1848 if ((exc1 == EXC_CONTRIBUTORY && exc2 == EXC_CONTRIBUTORY) ||
1849 (exc1 == EXC_PAGEFAULT && exc2 != EXC_BENIGN)) {
1850 /* Convert nested fault into a double fault. */
1851 *retinfo = IDT_DF;
1852 *retinfo |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1853 *retinfo |= VM_INTINFO_DEL_ERRCODE;
1854 } else {
1855 /* Handle exceptions serially */
1856 *retinfo = info2;
1857 }
1858 return (1);
1859 }
1860
1861 static uint64_t
1862 vcpu_exception_intinfo(struct vcpu *vcpu)
1863 {
1864 uint64_t info = 0;
1865
1866 if (vcpu->exception_pending) {
1867 info = vcpu->exc_vector & 0xff;
1868 info |= VM_INTINFO_VALID | VM_INTINFO_HWEXCEPTION;
1869 if (vcpu->exc_errcode_valid) {
1870 info |= VM_INTINFO_DEL_ERRCODE;
1871 info |= (uint64_t)vcpu->exc_errcode << 32;
1872 }
1873 }
1874 return (info);
1875 }
1876
1877 int
1878 vm_entry_intinfo(struct vm *vm, int vcpuid, uint64_t *retinfo)
1879 {
1880 struct vcpu *vcpu;
1881 uint64_t info1, info2;
1882 int valid;
1883
1884 KASSERT(vcpuid >= 0 && vcpuid < VM_MAXCPU, ("invalid vcpu %d", vcpuid));
1885
1886 vcpu = &vm->vcpu[vcpuid];
1887
1888 info1 = vcpu->exitintinfo;
1889 vcpu->exitintinfo = 0;
1890
1891 info2 = 0;
1892 if (vcpu->exception_pending) {
1893 info2 = vcpu_exception_intinfo(vcpu);
1894 vcpu->exception_pending = 0;
1895 VCPU_CTR2(vm, vcpuid, "Exception %d delivered: %#lx",
1896 vcpu->exc_vector, info2);
1897 }
1898
1899 if ((info1 & VM_INTINFO_VALID) && (info2 & VM_INTINFO_VALID)) {
1900 valid = nested_fault(vm, vcpuid, info1, info2, retinfo);
1901 } else if (info1 & VM_INTINFO_VALID) {
1902 *retinfo = info1;
1903 valid = 1;
1904 } else if (info2 & VM_INTINFO_VALID) {
1905 *retinfo = info2;
1906 valid = 1;
1907 } else {
1908 valid = 0;
1909 }
1910
1911 if (valid) {
1912 VCPU_CTR4(vm, vcpuid, "%s: info1(%#lx), info2(%#lx), "
1913 "retinfo(%#lx)", __func__, info1, info2, *retinfo);
1914 }
1915
1916 return (valid);
1917 }
1918
1919 int
1920 vm_get_intinfo(struct vm *vm, int vcpuid, uint64_t *info1, uint64_t *info2)
1921 {
1922 struct vcpu *vcpu;
1923
1924 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1925 return (EINVAL);
1926
1927 vcpu = &vm->vcpu[vcpuid];
1928 *info1 = vcpu->exitintinfo;
1929 *info2 = vcpu_exception_intinfo(vcpu);
1930 return (0);
1931 }
1932
1933 int
1934 vm_inject_exception(struct vm *vm, int vcpuid, int vector, int errcode_valid,
1935 uint32_t errcode, int restart_instruction)
1936 {
1937 struct vcpu *vcpu;
1938 uint64_t regval;
1939 int error;
1940
1941 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
1942 return (EINVAL);
1943
1944 if (vector < 0 || vector >= 32)
1945 return (EINVAL);
1946
1947 /*
1948 * A double fault exception should never be injected directly into
1949 * the guest. It is a derived exception that results from specific
1950 * combinations of nested faults.
1951 */
1952 if (vector == IDT_DF)
1953 return (EINVAL);
1954
1955 vcpu = &vm->vcpu[vcpuid];
1956
1957 if (vcpu->exception_pending) {
1958 VCPU_CTR2(vm, vcpuid, "Unable to inject exception %d due to "
1959 "pending exception %d", vector, vcpu->exc_vector);
1960 return (EBUSY);
1961 }
1962
1963 if (errcode_valid) {
1964 /*
1965 * Exceptions don't deliver an error code in real mode.
1966 */
1967 error = vm_get_register(vm, vcpuid, VM_REG_GUEST_CR0, ®val);
1968 KASSERT(!error, ("%s: error %d getting CR0", __func__, error));
1969 if (!(regval & CR0_PE))
1970 errcode_valid = 0;
1971 }
1972
1973 /*
1974 * From section 26.6.1 "Interruptibility State" in Intel SDM:
1975 *
1976 * Event blocking by "STI" or "MOV SS" is cleared after guest executes
1977 * one instruction or incurs an exception.
1978 */
1979 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_INTR_SHADOW, 0);
1980 KASSERT(error == 0, ("%s: error %d clearing interrupt shadow",
1981 __func__, error));
1982
1983 if (restart_instruction)
1984 vm_restart_instruction(vm, vcpuid);
1985
1986 vcpu->exception_pending = 1;
1987 vcpu->exc_vector = vector;
1988 vcpu->exc_errcode = errcode;
1989 vcpu->exc_errcode_valid = errcode_valid;
1990 VCPU_CTR1(vm, vcpuid, "Exception %d pending", vector);
1991 return (0);
1992 }
1993
1994 void
1995 vm_inject_fault(void *vmarg, int vcpuid, int vector, int errcode_valid,
1996 int errcode)
1997 {
1998 struct vm *vm;
1999 int error, restart_instruction;
2000
2001 vm = vmarg;
2002 restart_instruction = 1;
2003
2004 error = vm_inject_exception(vm, vcpuid, vector, errcode_valid,
2005 errcode, restart_instruction);
2006 KASSERT(error == 0, ("vm_inject_exception error %d", error));
2007 }
2008
2009 void
2010 vm_inject_pf(void *vmarg, int vcpuid, int error_code, uint64_t cr2)
2011 {
2012 struct vm *vm;
2013 int error;
2014
2015 vm = vmarg;
2016 VCPU_CTR2(vm, vcpuid, "Injecting page fault: error_code %#x, cr2 %#lx",
2017 error_code, cr2);
2018
2019 error = vm_set_register(vm, vcpuid, VM_REG_GUEST_CR2, cr2);
2020 KASSERT(error == 0, ("vm_set_register(cr2) error %d", error));
2021
2022 vm_inject_fault(vm, vcpuid, IDT_PF, 1, error_code);
2023 }
2024
2025 static VMM_STAT(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
2026
2027 int
2028 vm_inject_nmi(struct vm *vm, int vcpuid)
2029 {
2030 struct vcpu *vcpu;
2031
2032 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2033 return (EINVAL);
2034
2035 vcpu = &vm->vcpu[vcpuid];
2036
2037 vcpu->nmi_pending = 1;
2038 vcpu_notify_event(vm, vcpuid, false);
2039 return (0);
2040 }
2041
2042 int
2043 vm_nmi_pending(struct vm *vm, int vcpuid)
2044 {
2045 struct vcpu *vcpu;
2046
2047 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2048 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2049
2050 vcpu = &vm->vcpu[vcpuid];
2051
2052 return (vcpu->nmi_pending);
2053 }
2054
2055 void
2056 vm_nmi_clear(struct vm *vm, int vcpuid)
2057 {
2058 struct vcpu *vcpu;
2059
2060 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2061 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
2062
2063 vcpu = &vm->vcpu[vcpuid];
2064
2065 if (vcpu->nmi_pending == 0)
2066 panic("vm_nmi_clear: inconsistent nmi_pending state");
2067
2068 vcpu->nmi_pending = 0;
2069 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
2070 }
2071
2072 static VMM_STAT(VCPU_EXTINT_COUNT, "number of ExtINTs delivered to vcpu");
2073
2074 int
2075 vm_inject_extint(struct vm *vm, int vcpuid)
2076 {
2077 struct vcpu *vcpu;
2078
2079 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2080 return (EINVAL);
2081
2082 vcpu = &vm->vcpu[vcpuid];
2083
2084 vcpu->extint_pending = 1;
2085 vcpu_notify_event(vm, vcpuid, false);
2086 return (0);
2087 }
2088
2089 int
2090 vm_extint_pending(struct vm *vm, int vcpuid)
2091 {
2092 struct vcpu *vcpu;
2093
2094 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2095 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2096
2097 vcpu = &vm->vcpu[vcpuid];
2098
2099 return (vcpu->extint_pending);
2100 }
2101
2102 void
2103 vm_extint_clear(struct vm *vm, int vcpuid)
2104 {
2105 struct vcpu *vcpu;
2106
2107 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2108 panic("vm_extint_pending: invalid vcpuid %d", vcpuid);
2109
2110 vcpu = &vm->vcpu[vcpuid];
2111
2112 if (vcpu->extint_pending == 0)
2113 panic("vm_extint_clear: inconsistent extint_pending state");
2114
2115 vcpu->extint_pending = 0;
2116 vmm_stat_incr(vm, vcpuid, VCPU_EXTINT_COUNT, 1);
2117 }
2118
2119 int
2120 vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
2121 {
2122 if (vcpu < 0 || vcpu >= VM_MAXCPU)
2123 return (EINVAL);
2124
2125 if (type < 0 || type >= VM_CAP_MAX)
2126 return (EINVAL);
2127
2128 return (VMGETCAP(vm->cookie, vcpu, type, retval));
2129 }
2130
2131 int
2132 vm_set_capability(struct vm *vm, int vcpu, int type, int val)
2133 {
2134 if (vcpu < 0 || vcpu >= VM_MAXCPU)
2135 return (EINVAL);
2136
2137 if (type < 0 || type >= VM_CAP_MAX)
2138 return (EINVAL);
2139
2140 return (VMSETCAP(vm->cookie, vcpu, type, val));
2141 }
2142
2143 struct vlapic *
2144 vm_lapic(struct vm *vm, int cpu)
2145 {
2146 return (vm->vcpu[cpu].vlapic);
2147 }
2148
2149 struct vioapic *
2150 vm_ioapic(struct vm *vm)
2151 {
2152
2153 return (vm->vioapic);
2154 }
2155
2156 struct vhpet *
2157 vm_hpet(struct vm *vm)
2158 {
2159
2160 return (vm->vhpet);
2161 }
2162
2163 boolean_t
2164 vmm_is_pptdev(int bus, int slot, int func)
2165 {
2166 int found, i, n;
2167 int b, s, f;
2168 char *val, *cp, *cp2;
2169
2170 /*
2171 * XXX
2172 * The length of an environment variable is limited to 128 bytes which
2173 * puts an upper limit on the number of passthru devices that may be
2174 * specified using a single environment variable.
2175 *
2176 * Work around this by scanning multiple environment variable
2177 * names instead of a single one - yuck!
2178 */
2179 const char *names[] = { "pptdevs", "pptdevs2", "pptdevs3", NULL };
2180
2181 /* set pptdevs="1/2/3 4/5/6 7/8/9 10/11/12" */
2182 found = 0;
2183 for (i = 0; names[i] != NULL && !found; i++) {
2184 cp = val = kern_getenv(names[i]);
2185 while (cp != NULL && *cp != '\0') {
2186 if ((cp2 = strchr(cp, ' ')) != NULL)
2187 *cp2 = '\0';
2188
2189 n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
2190 if (n == 3 && bus == b && slot == s && func == f) {
2191 found = 1;
2192 break;
2193 }
2194
2195 if (cp2 != NULL)
2196 *cp2++ = ' ';
2197
2198 cp = cp2;
2199 }
2200 freeenv(val);
2201 }
2202 return (found);
2203 }
2204
2205 void *
2206 vm_iommu_domain(struct vm *vm)
2207 {
2208
2209 return (vm->iommu);
2210 }
2211
2212 int
2213 vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state newstate,
2214 bool from_idle)
2215 {
2216 int error;
2217 struct vcpu *vcpu;
2218
2219 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2220 panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
2221
2222 vcpu = &vm->vcpu[vcpuid];
2223
2224 vcpu_lock(vcpu);
2225 error = vcpu_set_state_locked(vm, vcpuid, newstate, from_idle);
2226 vcpu_unlock(vcpu);
2227
2228 return (error);
2229 }
2230
2231 enum vcpu_state
2232 vcpu_get_state(struct vm *vm, int vcpuid, int *hostcpu)
2233 {
2234 struct vcpu *vcpu;
2235 enum vcpu_state state;
2236
2237 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2238 panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
2239
2240 vcpu = &vm->vcpu[vcpuid];
2241
2242 vcpu_lock(vcpu);
2243 state = vcpu->state;
2244 if (hostcpu != NULL)
2245 *hostcpu = vcpu->hostcpu;
2246 vcpu_unlock(vcpu);
2247
2248 return (state);
2249 }
2250
2251 int
2252 vm_activate_cpu(struct vm *vm, int vcpuid)
2253 {
2254
2255 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2256 return (EINVAL);
2257
2258 if (CPU_ISSET(vcpuid, &vm->active_cpus))
2259 return (EBUSY);
2260
2261 VCPU_CTR0(vm, vcpuid, "activated");
2262 CPU_SET_ATOMIC(vcpuid, &vm->active_cpus);
2263 return (0);
2264 }
2265
2266 cpuset_t
2267 vm_active_cpus(struct vm *vm)
2268 {
2269
2270 return (vm->active_cpus);
2271 }
2272
2273 cpuset_t
2274 vm_suspended_cpus(struct vm *vm)
2275 {
2276
2277 return (vm->suspended_cpus);
2278 }
2279
2280 void *
2281 vcpu_stats(struct vm *vm, int vcpuid)
2282 {
2283
2284 return (vm->vcpu[vcpuid].stats);
2285 }
2286
2287 int
2288 vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
2289 {
2290 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2291 return (EINVAL);
2292
2293 *state = vm->vcpu[vcpuid].x2apic_state;
2294
2295 return (0);
2296 }
2297
2298 int
2299 vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
2300 {
2301 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
2302 return (EINVAL);
2303
2304 if (state >= X2APIC_STATE_LAST)
2305 return (EINVAL);
2306
2307 vm->vcpu[vcpuid].x2apic_state = state;
2308
2309 vlapic_set_x2apic_state(vm, vcpuid, state);
2310
2311 return (0);
2312 }
2313
2314 /*
2315 * This function is called to ensure that a vcpu "sees" a pending event
2316 * as soon as possible:
2317 * - If the vcpu thread is sleeping then it is woken up.
2318 * - If the vcpu is running on a different host_cpu then an IPI will be directed
2319 * to the host_cpu to cause the vcpu to trap into the hypervisor.
2320 */
2321 static void
2322 vcpu_notify_event_locked(struct vcpu *vcpu, bool lapic_intr)
2323 {
2324 int hostcpu;
2325
2326 hostcpu = vcpu->hostcpu;
2327 if (vcpu->state == VCPU_RUNNING) {
2328 KASSERT(hostcpu != NOCPU, ("vcpu running on invalid hostcpu"));
2329 if (hostcpu != curcpu) {
2330 if (lapic_intr) {
2331 vlapic_post_intr(vcpu->vlapic, hostcpu,
2332 vmm_ipinum);
2333 } else {
2334 ipi_cpu(hostcpu, vmm_ipinum);
2335 }
2336 } else {
2337 /*
2338 * If the 'vcpu' is running on 'curcpu' then it must
2339 * be sending a notification to itself (e.g. SELF_IPI).
2340 * The pending event will be picked up when the vcpu
2341 * transitions back to guest context.
2342 */
2343 }
2344 } else {
2345 KASSERT(hostcpu == NOCPU, ("vcpu state %d not consistent "
2346 "with hostcpu %d", vcpu->state, hostcpu));
2347 if (vcpu->state == VCPU_SLEEPING)
2348 wakeup_one(vcpu);
2349 }
2350 }
2351
2352 void
2353 vcpu_notify_event(struct vm *vm, int vcpuid, bool lapic_intr)
2354 {
2355 struct vcpu *vcpu = &vm->vcpu[vcpuid];
2356
2357 vcpu_lock(vcpu);
2358 vcpu_notify_event_locked(vcpu, lapic_intr);
2359 vcpu_unlock(vcpu);
2360 }
2361
2362 struct vmspace *
2363 vm_get_vmspace(struct vm *vm)
2364 {
2365
2366 return (vm->vmspace);
2367 }
2368
2369 int
2370 vm_apicid2vcpuid(struct vm *vm, int apicid)
2371 {
2372 /*
2373 * XXX apic id is assumed to be numerically identical to vcpu id
2374 */
2375 return (apicid);
2376 }
2377
2378 void
2379 vm_smp_rendezvous(struct vm *vm, int vcpuid, cpuset_t dest,
2380 vm_rendezvous_func_t func, void *arg)
2381 {
2382 int i;
2383
2384 /*
2385 * Enforce that this function is called without any locks
2386 */
2387 WITNESS_WARN(WARN_PANIC, NULL, "vm_smp_rendezvous");
2388 KASSERT(vcpuid == -1 || (vcpuid >= 0 && vcpuid < VM_MAXCPU),
2389 ("vm_smp_rendezvous: invalid vcpuid %d", vcpuid));
2390
2391 restart:
2392 mtx_lock(&vm->rendezvous_mtx);
2393 if (vm->rendezvous_func != NULL) {
2394 /*
2395 * If a rendezvous is already in progress then we need to
2396 * call the rendezvous handler in case this 'vcpuid' is one
2397 * of the targets of the rendezvous.
2398 */
2399 RENDEZVOUS_CTR0(vm, vcpuid, "Rendezvous already in progress");
2400 mtx_unlock(&vm->rendezvous_mtx);
2401 vm_handle_rendezvous(vm, vcpuid);
2402 goto restart;
2403 }
2404 KASSERT(vm->rendezvous_func == NULL, ("vm_smp_rendezvous: previous "
2405 "rendezvous is still in progress"));
2406
2407 RENDEZVOUS_CTR0(vm, vcpuid, "Initiating rendezvous");
2408 vm->rendezvous_req_cpus = dest;
2409 CPU_ZERO(&vm->rendezvous_done_cpus);
2410 vm->rendezvous_arg = arg;
2411 vm_set_rendezvous_func(vm, func);
2412 mtx_unlock(&vm->rendezvous_mtx);
2413
2414 /*
2415 * Wake up any sleeping vcpus and trigger a VM-exit in any running
2416 * vcpus so they handle the rendezvous as soon as possible.
2417 */
2418 for (i = 0; i < VM_MAXCPU; i++) {
2419 if (CPU_ISSET(i, &dest))
2420 vcpu_notify_event(vm, i, false);
2421 }
2422
2423 vm_handle_rendezvous(vm, vcpuid);
2424 }
2425
2426 struct vatpic *
2427 vm_atpic(struct vm *vm)
2428 {
2429 return (vm->vatpic);
2430 }
2431
2432 struct vatpit *
2433 vm_atpit(struct vm *vm)
2434 {
2435 return (vm->vatpit);
2436 }
2437
2438 struct vpmtmr *
2439 vm_pmtmr(struct vm *vm)
2440 {
2441
2442 return (vm->vpmtmr);
2443 }
2444
2445 struct vrtc *
2446 vm_rtc(struct vm *vm)
2447 {
2448
2449 return (vm->vrtc);
2450 }
2451
2452 enum vm_reg_name
2453 vm_segment_name(int seg)
2454 {
2455 static enum vm_reg_name seg_names[] = {
2456 VM_REG_GUEST_ES,
2457 VM_REG_GUEST_CS,
2458 VM_REG_GUEST_SS,
2459 VM_REG_GUEST_DS,
2460 VM_REG_GUEST_FS,
2461 VM_REG_GUEST_GS
2462 };
2463
2464 KASSERT(seg >= 0 && seg < nitems(seg_names),
2465 ("%s: invalid segment encoding %d", __func__, seg));
2466 return (seg_names[seg]);
2467 }
2468
2469 void
2470 vm_copy_teardown(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo,
2471 int num_copyinfo)
2472 {
2473 int idx;
2474
2475 for (idx = 0; idx < num_copyinfo; idx++) {
2476 if (copyinfo[idx].cookie != NULL)
2477 vm_gpa_release(copyinfo[idx].cookie);
2478 }
2479 bzero(copyinfo, num_copyinfo * sizeof(struct vm_copyinfo));
2480 }
2481
2482 int
2483 vm_copy_setup(struct vm *vm, int vcpuid, struct vm_guest_paging *paging,
2484 uint64_t gla, size_t len, int prot, struct vm_copyinfo *copyinfo,
2485 int num_copyinfo, int *fault)
2486 {
2487 int error, idx, nused;
2488 size_t n, off, remaining;
2489 void *hva, *cookie;
2490 uint64_t gpa;
2491
2492 bzero(copyinfo, sizeof(struct vm_copyinfo) * num_copyinfo);
2493
2494 nused = 0;
2495 remaining = len;
2496 while (remaining > 0) {
2497 KASSERT(nused < num_copyinfo, ("insufficient vm_copyinfo"));
2498 error = vm_gla2gpa(vm, vcpuid, paging, gla, prot, &gpa, fault);
2499 if (error || *fault)
2500 return (error);
2501 off = gpa & PAGE_MASK;
2502 n = min(remaining, PAGE_SIZE - off);
2503 copyinfo[nused].gpa = gpa;
2504 copyinfo[nused].len = n;
2505 remaining -= n;
2506 gla += n;
2507 nused++;
2508 }
2509
2510 for (idx = 0; idx < nused; idx++) {
2511 hva = vm_gpa_hold(vm, vcpuid, copyinfo[idx].gpa,
2512 copyinfo[idx].len, prot, &cookie);
2513 if (hva == NULL)
2514 break;
2515 copyinfo[idx].hva = hva;
2516 copyinfo[idx].cookie = cookie;
2517 }
2518
2519 if (idx != nused) {
2520 vm_copy_teardown(vm, vcpuid, copyinfo, num_copyinfo);
2521 return (EFAULT);
2522 } else {
2523 *fault = 0;
2524 return (0);
2525 }
2526 }
2527
2528 void
2529 vm_copyin(struct vm *vm, int vcpuid, struct vm_copyinfo *copyinfo, void *kaddr,
2530 size_t len)
2531 {
2532 char *dst;
2533 int idx;
2534
2535 dst = kaddr;
2536 idx = 0;
2537 while (len > 0) {
2538 bcopy(copyinfo[idx].hva, dst, copyinfo[idx].len);
2539 len -= copyinfo[idx].len;
2540 dst += copyinfo[idx].len;
2541 idx++;
2542 }
2543 }
2544
2545 void
2546 vm_copyout(struct vm *vm, int vcpuid, const void *kaddr,
2547 struct vm_copyinfo *copyinfo, size_t len)
2548 {
2549 const char *src;
2550 int idx;
2551
2552 src = kaddr;
2553 idx = 0;
2554 while (len > 0) {
2555 bcopy(src, copyinfo[idx].hva, copyinfo[idx].len);
2556 len -= copyinfo[idx].len;
2557 src += copyinfo[idx].len;
2558 idx++;
2559 }
2560 }
2561
2562 /*
2563 * Return the amount of in-use and wired memory for the VM. Since
2564 * these are global stats, only return the values with for vCPU 0
2565 */
2566 VMM_STAT_DECLARE(VMM_MEM_RESIDENT);
2567 VMM_STAT_DECLARE(VMM_MEM_WIRED);
2568
2569 static void
2570 vm_get_rescnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2571 {
2572
2573 if (vcpu == 0) {
2574 vmm_stat_set(vm, vcpu, VMM_MEM_RESIDENT,
2575 PAGE_SIZE * vmspace_resident_count(vm->vmspace));
2576 }
2577 }
2578
2579 static void
2580 vm_get_wiredcnt(struct vm *vm, int vcpu, struct vmm_stat_type *stat)
2581 {
2582
2583 if (vcpu == 0) {
2584 vmm_stat_set(vm, vcpu, VMM_MEM_WIRED,
2585 PAGE_SIZE * pmap_wired_count(vmspace_pmap(vm->vmspace)));
2586 }
2587 }
2588
2589 VMM_STAT_FUNC(VMM_MEM_RESIDENT, "Resident memory", vm_get_rescnt);
2590 VMM_STAT_FUNC(VMM_MEM_WIRED, "Wired memory", vm_get_wiredcnt);
Cache object: 7980b4a7271030dceb7cdd83da7db247
|