1/*-
2 * Copyright (c) 2011 NetApp, Inc.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD$
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD$");
31
32#include <sys/param.h>
33#include <sys/systm.h>
34#include <sys/kernel.h>
35#include <sys/module.h>
36#include <sys/sysctl.h>
37#include <sys/malloc.h>
38#include <sys/pcpu.h>
39#include <sys/lock.h>
40#include <sys/mutex.h>
41#include <sys/proc.h>
42#include <sys/sched.h>
43#include <sys/smp.h>
44#include <sys/systm.h>
45
46#include <vm/vm.h>
47
48#include <machine/vm.h>
49#include <machine/pcb.h>
50#include <machine/smp.h>
51#include <x86/apicreg.h>
52
53#include <machine/vmm.h>
54#include "vmm_mem.h"
55#include "vmm_util.h"
56#include <machine/vmm_dev.h>
57#include "vlapic.h"
58#include "vmm_msr.h"
59#include "vmm_ipi.h"
60#include "vmm_stat.h"
61
62#include "io/ppt.h"
63#include "io/iommu.h"
64
65struct vlapic;
66
67struct vcpu {
68 int flags;
69 enum vcpu_state state;
70 struct mtx mtx;
71 int pincpu; /* host cpuid this vcpu is bound to */
72 int hostcpu; /* host cpuid this vcpu last ran on */
73 uint64_t guest_msrs[VMM_MSR_NUM];
74 struct vlapic *vlapic;
75 int vcpuid;
76 struct savefpu *guestfpu; /* guest fpu state */
77 void *stats;
78 struct vm_exit exitinfo;
79 enum x2apic_state x2apic_state;
| 1/*-
2 * Copyright (c) 2011 NetApp, Inc.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD$
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD$");
31
32#include <sys/param.h>
33#include <sys/systm.h>
34#include <sys/kernel.h>
35#include <sys/module.h>
36#include <sys/sysctl.h>
37#include <sys/malloc.h>
38#include <sys/pcpu.h>
39#include <sys/lock.h>
40#include <sys/mutex.h>
41#include <sys/proc.h>
42#include <sys/sched.h>
43#include <sys/smp.h>
44#include <sys/systm.h>
45
46#include <vm/vm.h>
47
48#include <machine/vm.h>
49#include <machine/pcb.h>
50#include <machine/smp.h>
51#include <x86/apicreg.h>
52
53#include <machine/vmm.h>
54#include "vmm_mem.h"
55#include "vmm_util.h"
56#include <machine/vmm_dev.h>
57#include "vlapic.h"
58#include "vmm_msr.h"
59#include "vmm_ipi.h"
60#include "vmm_stat.h"
61
62#include "io/ppt.h"
63#include "io/iommu.h"
64
65struct vlapic;
66
67struct vcpu {
68 int flags;
69 enum vcpu_state state;
70 struct mtx mtx;
71 int pincpu; /* host cpuid this vcpu is bound to */
72 int hostcpu; /* host cpuid this vcpu last ran on */
73 uint64_t guest_msrs[VMM_MSR_NUM];
74 struct vlapic *vlapic;
75 int vcpuid;
76 struct savefpu *guestfpu; /* guest fpu state */
77 void *stats;
78 struct vm_exit exitinfo;
79 enum x2apic_state x2apic_state;
|
| 80 int nmi_pending;
|
80};
81#define VCPU_F_PINNED 0x0001
82
83#define VCPU_PINCPU(vm, vcpuid) \
84 ((vm->vcpu[vcpuid].flags & VCPU_F_PINNED) ? vm->vcpu[vcpuid].pincpu : -1)
85
86#define VCPU_UNPIN(vm, vcpuid) (vm->vcpu[vcpuid].flags &= ~VCPU_F_PINNED)
87
88#define VCPU_PIN(vm, vcpuid, host_cpuid) \
89do { \
90 vm->vcpu[vcpuid].flags |= VCPU_F_PINNED; \
91 vm->vcpu[vcpuid].pincpu = host_cpuid; \
92} while(0)
93
94#define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_DEF)
95#define vcpu_lock(v) mtx_lock(&((v)->mtx))
96#define vcpu_unlock(v) mtx_unlock(&((v)->mtx))
97
98#define VM_MAX_MEMORY_SEGMENTS 2
99
100struct vm {
101 void *cookie; /* processor-specific data */
102 void *iommu; /* iommu-specific data */
103 struct vcpu vcpu[VM_MAXCPU];
104 int num_mem_segs;
105 struct vm_memory_segment mem_segs[VM_MAX_MEMORY_SEGMENTS];
106 char name[VM_MAX_NAMELEN];
107
108 /*
109 * Set of active vcpus.
110 * An active vcpu is one that has been started implicitly (BSP) or
111 * explicitly (AP) by sending it a startup ipi.
112 */
113 cpuset_t active_cpus;
114};
115
116static struct vmm_ops *ops;
117#define VMM_INIT() (ops != NULL ? (*ops->init)() : 0)
118#define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0)
119
120#define VMINIT(vm) (ops != NULL ? (*ops->vminit)(vm): NULL)
121#define VMRUN(vmi, vcpu, rip) \
122 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip) : ENXIO)
123#define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
124#define VMMMAP_SET(vmi, gpa, hpa, len, attr, prot, spm) \
125 (ops != NULL ? \
126 (*ops->vmmmap_set)(vmi, gpa, hpa, len, attr, prot, spm) : \
127 ENXIO)
128#define VMMMAP_GET(vmi, gpa) \
129 (ops != NULL ? (*ops->vmmmap_get)(vmi, gpa) : ENXIO)
130#define VMGETREG(vmi, vcpu, num, retval) \
131 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
132#define VMSETREG(vmi, vcpu, num, val) \
133 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
134#define VMGETDESC(vmi, vcpu, num, desc) \
135 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
136#define VMSETDESC(vmi, vcpu, num, desc) \
137 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
138#define VMINJECT(vmi, vcpu, type, vec, ec, ecv) \
139 (ops != NULL ? (*ops->vminject)(vmi, vcpu, type, vec, ec, ecv) : ENXIO)
| 81};
82#define VCPU_F_PINNED 0x0001
83
84#define VCPU_PINCPU(vm, vcpuid) \
85 ((vm->vcpu[vcpuid].flags & VCPU_F_PINNED) ? vm->vcpu[vcpuid].pincpu : -1)
86
87#define VCPU_UNPIN(vm, vcpuid) (vm->vcpu[vcpuid].flags &= ~VCPU_F_PINNED)
88
89#define VCPU_PIN(vm, vcpuid, host_cpuid) \
90do { \
91 vm->vcpu[vcpuid].flags |= VCPU_F_PINNED; \
92 vm->vcpu[vcpuid].pincpu = host_cpuid; \
93} while(0)
94
95#define vcpu_lock_init(v) mtx_init(&((v)->mtx), "vcpu lock", 0, MTX_DEF)
96#define vcpu_lock(v) mtx_lock(&((v)->mtx))
97#define vcpu_unlock(v) mtx_unlock(&((v)->mtx))
98
99#define VM_MAX_MEMORY_SEGMENTS 2
100
101struct vm {
102 void *cookie; /* processor-specific data */
103 void *iommu; /* iommu-specific data */
104 struct vcpu vcpu[VM_MAXCPU];
105 int num_mem_segs;
106 struct vm_memory_segment mem_segs[VM_MAX_MEMORY_SEGMENTS];
107 char name[VM_MAX_NAMELEN];
108
109 /*
110 * Set of active vcpus.
111 * An active vcpu is one that has been started implicitly (BSP) or
112 * explicitly (AP) by sending it a startup ipi.
113 */
114 cpuset_t active_cpus;
115};
116
117static struct vmm_ops *ops;
118#define VMM_INIT() (ops != NULL ? (*ops->init)() : 0)
119#define VMM_CLEANUP() (ops != NULL ? (*ops->cleanup)() : 0)
120
121#define VMINIT(vm) (ops != NULL ? (*ops->vminit)(vm): NULL)
122#define VMRUN(vmi, vcpu, rip) \
123 (ops != NULL ? (*ops->vmrun)(vmi, vcpu, rip) : ENXIO)
124#define VMCLEANUP(vmi) (ops != NULL ? (*ops->vmcleanup)(vmi) : NULL)
125#define VMMMAP_SET(vmi, gpa, hpa, len, attr, prot, spm) \
126 (ops != NULL ? \
127 (*ops->vmmmap_set)(vmi, gpa, hpa, len, attr, prot, spm) : \
128 ENXIO)
129#define VMMMAP_GET(vmi, gpa) \
130 (ops != NULL ? (*ops->vmmmap_get)(vmi, gpa) : ENXIO)
131#define VMGETREG(vmi, vcpu, num, retval) \
132 (ops != NULL ? (*ops->vmgetreg)(vmi, vcpu, num, retval) : ENXIO)
133#define VMSETREG(vmi, vcpu, num, val) \
134 (ops != NULL ? (*ops->vmsetreg)(vmi, vcpu, num, val) : ENXIO)
135#define VMGETDESC(vmi, vcpu, num, desc) \
136 (ops != NULL ? (*ops->vmgetdesc)(vmi, vcpu, num, desc) : ENXIO)
137#define VMSETDESC(vmi, vcpu, num, desc) \
138 (ops != NULL ? (*ops->vmsetdesc)(vmi, vcpu, num, desc) : ENXIO)
139#define VMINJECT(vmi, vcpu, type, vec, ec, ecv) \
140 (ops != NULL ? (*ops->vminject)(vmi, vcpu, type, vec, ec, ecv) : ENXIO)
|
140#define VMNMI(vmi, vcpu) \
141 (ops != NULL ? (*ops->vmnmi)(vmi, vcpu) : ENXIO)
| |
142#define VMGETCAP(vmi, vcpu, num, retval) \
143 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
144#define VMSETCAP(vmi, vcpu, num, val) \
145 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
146
147#define fpu_start_emulating() start_emulating()
148#define fpu_stop_emulating() stop_emulating()
149
150static MALLOC_DEFINE(M_VM, "vm", "vm");
151CTASSERT(VMM_MSR_NUM <= 64); /* msr_mask can keep track of up to 64 msrs */
152
153/* statistics */
154static VMM_STAT_DEFINE(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
155
156static void
157vcpu_cleanup(struct vcpu *vcpu)
158{
159 vlapic_cleanup(vcpu->vlapic);
160 vmm_stat_free(vcpu->stats);
161 fpu_save_area_free(vcpu->guestfpu);
162}
163
164static void
165vcpu_init(struct vm *vm, uint32_t vcpu_id)
166{
167 struct vcpu *vcpu;
168
169 vcpu = &vm->vcpu[vcpu_id];
170
171 vcpu_lock_init(vcpu);
172 vcpu->hostcpu = NOCPU;
173 vcpu->vcpuid = vcpu_id;
174 vcpu->vlapic = vlapic_init(vm, vcpu_id);
175 vm_set_x2apic_state(vm, vcpu_id, X2APIC_ENABLED);
176 vcpu->guestfpu = fpu_save_area_alloc();
177 fpu_save_area_reset(vcpu->guestfpu);
178 vcpu->stats = vmm_stat_alloc();
179}
180
181struct vm_exit *
182vm_exitinfo(struct vm *vm, int cpuid)
183{
184 struct vcpu *vcpu;
185
186 if (cpuid < 0 || cpuid >= VM_MAXCPU)
187 panic("vm_exitinfo: invalid cpuid %d", cpuid);
188
189 vcpu = &vm->vcpu[cpuid];
190
191 return (&vcpu->exitinfo);
192}
193
194static int
195vmm_init(void)
196{
197 int error;
198
199 vmm_ipi_init();
200
201 error = vmm_mem_init();
202 if (error)
203 return (error);
204
205 if (vmm_is_intel())
206 ops = &vmm_ops_intel;
207 else if (vmm_is_amd())
208 ops = &vmm_ops_amd;
209 else
210 return (ENXIO);
211
212 vmm_msr_init();
213
214 return (VMM_INIT());
215}
216
217static int
218vmm_handler(module_t mod, int what, void *arg)
219{
220 int error;
221
222 switch (what) {
223 case MOD_LOAD:
224 vmmdev_init();
225 iommu_init();
226 error = vmm_init();
227 break;
228 case MOD_UNLOAD:
229 error = vmmdev_cleanup();
230 if (error == 0) {
231 iommu_cleanup();
232 vmm_ipi_cleanup();
233 error = VMM_CLEANUP();
234 }
235 break;
236 default:
237 error = 0;
238 break;
239 }
240 return (error);
241}
242
243static moduledata_t vmm_kmod = {
244 "vmm",
245 vmm_handler,
246 NULL
247};
248
249/*
250 * Execute the module load handler after the pci passthru driver has had
251 * a chance to claim devices. We need this information at the time we do
252 * iommu initialization.
253 */
254DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_CONFIGURE + 1, SI_ORDER_ANY);
255MODULE_VERSION(vmm, 1);
256
257SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
258
259struct vm *
260vm_create(const char *name)
261{
262 int i;
263 struct vm *vm;
264 vm_paddr_t maxaddr;
265
266 const int BSP = 0;
267
268 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
269 return (NULL);
270
271 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
272 strcpy(vm->name, name);
273 vm->cookie = VMINIT(vm);
274
275 for (i = 0; i < VM_MAXCPU; i++) {
276 vcpu_init(vm, i);
277 guest_msrs_init(vm, i);
278 }
279
280 maxaddr = vmm_mem_maxaddr();
281 vm->iommu = iommu_create_domain(maxaddr);
282 vm_activate_cpu(vm, BSP);
283
284 return (vm);
285}
286
287static void
288vm_free_mem_seg(struct vm *vm, struct vm_memory_segment *seg)
289{
290 size_t len;
291 vm_paddr_t hpa;
292 void *host_domain;
293
294 host_domain = iommu_host_domain();
295
296 len = 0;
297 while (len < seg->len) {
298 hpa = vm_gpa2hpa(vm, seg->gpa + len, PAGE_SIZE);
299 if (hpa == (vm_paddr_t)-1) {
300 panic("vm_free_mem_segs: cannot free hpa "
301 "associated with gpa 0x%016lx", seg->gpa + len);
302 }
303
304 /*
305 * Remove the 'gpa' to 'hpa' mapping in VMs domain.
306 * And resurrect the 1:1 mapping for 'hpa' in 'host_domain'.
307 */
308 iommu_remove_mapping(vm->iommu, seg->gpa + len, PAGE_SIZE);
309 iommu_create_mapping(host_domain, hpa, hpa, PAGE_SIZE);
310
311 vmm_mem_free(hpa, PAGE_SIZE);
312
313 len += PAGE_SIZE;
314 }
315
316 /*
317 * Invalidate cached translations associated with 'vm->iommu' since
318 * we have now moved some pages from it.
319 */
320 iommu_invalidate_tlb(vm->iommu);
321
322 bzero(seg, sizeof(struct vm_memory_segment));
323}
324
325void
326vm_destroy(struct vm *vm)
327{
328 int i;
329
330 ppt_unassign_all(vm);
331
332 for (i = 0; i < vm->num_mem_segs; i++)
333 vm_free_mem_seg(vm, &vm->mem_segs[i]);
334
335 vm->num_mem_segs = 0;
336
337 for (i = 0; i < VM_MAXCPU; i++)
338 vcpu_cleanup(&vm->vcpu[i]);
339
340 iommu_destroy_domain(vm->iommu);
341
342 VMCLEANUP(vm->cookie);
343
344 free(vm, M_VM);
345}
346
347const char *
348vm_name(struct vm *vm)
349{
350 return (vm->name);
351}
352
353int
354vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
355{
356 const boolean_t spok = TRUE; /* superpage mappings are ok */
357
358 return (VMMMAP_SET(vm->cookie, gpa, hpa, len, VM_MEMATTR_UNCACHEABLE,
359 VM_PROT_RW, spok));
360}
361
362int
363vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
364{
365 const boolean_t spok = TRUE; /* superpage mappings are ok */
366
367 return (VMMMAP_SET(vm->cookie, gpa, 0, len, 0,
368 VM_PROT_NONE, spok));
369}
370
371/*
372 * Returns TRUE if 'gpa' is available for allocation and FALSE otherwise
373 */
374static boolean_t
375vm_gpa_available(struct vm *vm, vm_paddr_t gpa)
376{
377 int i;
378 vm_paddr_t gpabase, gpalimit;
379
380 if (gpa & PAGE_MASK)
381 panic("vm_gpa_available: gpa (0x%016lx) not page aligned", gpa);
382
383 for (i = 0; i < vm->num_mem_segs; i++) {
384 gpabase = vm->mem_segs[i].gpa;
385 gpalimit = gpabase + vm->mem_segs[i].len;
386 if (gpa >= gpabase && gpa < gpalimit)
387 return (FALSE);
388 }
389
390 return (TRUE);
391}
392
393int
394vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
395{
396 int error, available, allocated;
397 struct vm_memory_segment *seg;
398 vm_paddr_t g, hpa;
399 void *host_domain;
400
401 const boolean_t spok = TRUE; /* superpage mappings are ok */
402
403 if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
404 return (EINVAL);
405
406 available = allocated = 0;
407 g = gpa;
408 while (g < gpa + len) {
409 if (vm_gpa_available(vm, g))
410 available++;
411 else
412 allocated++;
413
414 g += PAGE_SIZE;
415 }
416
417 /*
418 * If there are some allocated and some available pages in the address
419 * range then it is an error.
420 */
421 if (allocated && available)
422 return (EINVAL);
423
424 /*
425 * If the entire address range being requested has already been
426 * allocated then there isn't anything more to do.
427 */
428 if (allocated && available == 0)
429 return (0);
430
431 if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
432 return (E2BIG);
433
434 host_domain = iommu_host_domain();
435
436 seg = &vm->mem_segs[vm->num_mem_segs];
437
438 error = 0;
439 seg->gpa = gpa;
440 seg->len = 0;
441 while (seg->len < len) {
442 hpa = vmm_mem_alloc(PAGE_SIZE);
443 if (hpa == 0) {
444 error = ENOMEM;
445 break;
446 }
447
448 error = VMMMAP_SET(vm->cookie, gpa + seg->len, hpa, PAGE_SIZE,
449 VM_MEMATTR_WRITE_BACK, VM_PROT_ALL, spok);
450 if (error)
451 break;
452
453 /*
454 * Remove the 1:1 mapping for 'hpa' from the 'host_domain'.
455 * Add mapping for 'gpa + seg->len' to 'hpa' in the VMs domain.
456 */
457 iommu_remove_mapping(host_domain, hpa, PAGE_SIZE);
458 iommu_create_mapping(vm->iommu, gpa + seg->len, hpa, PAGE_SIZE);
459
460 seg->len += PAGE_SIZE;
461 }
462
463 if (error) {
464 vm_free_mem_seg(vm, seg);
465 return (error);
466 }
467
468 /*
469 * Invalidate cached translations associated with 'host_domain' since
470 * we have now moved some pages from it.
471 */
472 iommu_invalidate_tlb(host_domain);
473
474 vm->num_mem_segs++;
475
476 return (0);
477}
478
479vm_paddr_t
480vm_gpa2hpa(struct vm *vm, vm_paddr_t gpa, size_t len)
481{
482 vm_paddr_t nextpage;
483
484 nextpage = rounddown(gpa + PAGE_SIZE, PAGE_SIZE);
485 if (len > nextpage - gpa)
486 panic("vm_gpa2hpa: invalid gpa/len: 0x%016lx/%lu", gpa, len);
487
488 return (VMMMAP_GET(vm->cookie, gpa));
489}
490
491int
492vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
493 struct vm_memory_segment *seg)
494{
495 int i;
496
497 for (i = 0; i < vm->num_mem_segs; i++) {
498 if (gpabase == vm->mem_segs[i].gpa) {
499 *seg = vm->mem_segs[i];
500 return (0);
501 }
502 }
503 return (-1);
504}
505
506int
507vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
508{
509
510 if (vcpu < 0 || vcpu >= VM_MAXCPU)
511 return (EINVAL);
512
513 if (reg >= VM_REG_LAST)
514 return (EINVAL);
515
516 return (VMGETREG(vm->cookie, vcpu, reg, retval));
517}
518
519int
520vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
521{
522
523 if (vcpu < 0 || vcpu >= VM_MAXCPU)
524 return (EINVAL);
525
526 if (reg >= VM_REG_LAST)
527 return (EINVAL);
528
529 return (VMSETREG(vm->cookie, vcpu, reg, val));
530}
531
532static boolean_t
533is_descriptor_table(int reg)
534{
535
536 switch (reg) {
537 case VM_REG_GUEST_IDTR:
538 case VM_REG_GUEST_GDTR:
539 return (TRUE);
540 default:
541 return (FALSE);
542 }
543}
544
545static boolean_t
546is_segment_register(int reg)
547{
548
549 switch (reg) {
550 case VM_REG_GUEST_ES:
551 case VM_REG_GUEST_CS:
552 case VM_REG_GUEST_SS:
553 case VM_REG_GUEST_DS:
554 case VM_REG_GUEST_FS:
555 case VM_REG_GUEST_GS:
556 case VM_REG_GUEST_TR:
557 case VM_REG_GUEST_LDTR:
558 return (TRUE);
559 default:
560 return (FALSE);
561 }
562}
563
564int
565vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
566 struct seg_desc *desc)
567{
568
569 if (vcpu < 0 || vcpu >= VM_MAXCPU)
570 return (EINVAL);
571
572 if (!is_segment_register(reg) && !is_descriptor_table(reg))
573 return (EINVAL);
574
575 return (VMGETDESC(vm->cookie, vcpu, reg, desc));
576}
577
578int
579vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
580 struct seg_desc *desc)
581{
582 if (vcpu < 0 || vcpu >= VM_MAXCPU)
583 return (EINVAL);
584
585 if (!is_segment_register(reg) && !is_descriptor_table(reg))
586 return (EINVAL);
587
588 return (VMSETDESC(vm->cookie, vcpu, reg, desc));
589}
590
591int
592vm_get_pinning(struct vm *vm, int vcpuid, int *cpuid)
593{
594
595 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
596 return (EINVAL);
597
598 *cpuid = VCPU_PINCPU(vm, vcpuid);
599
600 return (0);
601}
602
603int
604vm_set_pinning(struct vm *vm, int vcpuid, int host_cpuid)
605{
606 struct thread *td;
607
608 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
609 return (EINVAL);
610
611 td = curthread; /* XXXSMP only safe when muxing vcpus */
612
613 /* unpin */
614 if (host_cpuid < 0) {
615 VCPU_UNPIN(vm, vcpuid);
616 thread_lock(td);
617 sched_unbind(td);
618 thread_unlock(td);
619 return (0);
620 }
621
622 if (CPU_ABSENT(host_cpuid))
623 return (EINVAL);
624
625 /*
626 * XXX we should check that 'host_cpuid' has not already been pinned
627 * by another vm.
628 */
629 thread_lock(td);
630 sched_bind(td, host_cpuid);
631 thread_unlock(td);
632 VCPU_PIN(vm, vcpuid, host_cpuid);
633
634 return (0);
635}
636
637static void
638restore_guest_fpustate(struct vcpu *vcpu)
639{
640
641 /* flush host state to the pcb */
642 fpuexit(curthread);
643 fpu_stop_emulating();
644 fpurestore(vcpu->guestfpu);
645}
646
647static void
648save_guest_fpustate(struct vcpu *vcpu)
649{
650
651 fpusave(vcpu->guestfpu);
652 fpu_start_emulating();
653}
654
655int
656vm_run(struct vm *vm, struct vm_run *vmrun)
657{
658 int error, vcpuid;
659 struct vcpu *vcpu;
660 struct pcb *pcb;
661 uint64_t tscval;
662
663 vcpuid = vmrun->cpuid;
664
665 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
666 return (EINVAL);
667
668 vcpu = &vm->vcpu[vcpuid];
669
670 critical_enter();
671
672 tscval = rdtsc();
673
674 pcb = PCPU_GET(curpcb);
675 set_pcb_flags(pcb, PCB_FULL_IRET);
676
677 restore_guest_msrs(vm, vcpuid);
678 restore_guest_fpustate(vcpu);
679
680 vcpu->hostcpu = curcpu;
681 error = VMRUN(vm->cookie, vcpuid, vmrun->rip);
682 vcpu->hostcpu = NOCPU;
683
684 save_guest_fpustate(vcpu);
685 restore_host_msrs(vm, vcpuid);
686
687 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
688
689 /* copy the exit information */
690 bcopy(&vcpu->exitinfo, &vmrun->vm_exit, sizeof(struct vm_exit));
691
692 critical_exit();
693
694 return (error);
695}
696
697int
698vm_inject_event(struct vm *vm, int vcpuid, int type,
699 int vector, uint32_t code, int code_valid)
700{
701 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
702 return (EINVAL);
703
704 if ((type > VM_EVENT_NONE && type < VM_EVENT_MAX) == 0)
705 return (EINVAL);
706
707 if (vector < 0 || vector > 255)
708 return (EINVAL);
709
710 return (VMINJECT(vm->cookie, vcpuid, type, vector, code, code_valid));
711}
712
| 141#define VMGETCAP(vmi, vcpu, num, retval) \
142 (ops != NULL ? (*ops->vmgetcap)(vmi, vcpu, num, retval) : ENXIO)
143#define VMSETCAP(vmi, vcpu, num, val) \
144 (ops != NULL ? (*ops->vmsetcap)(vmi, vcpu, num, val) : ENXIO)
145
146#define fpu_start_emulating() start_emulating()
147#define fpu_stop_emulating() stop_emulating()
148
149static MALLOC_DEFINE(M_VM, "vm", "vm");
150CTASSERT(VMM_MSR_NUM <= 64); /* msr_mask can keep track of up to 64 msrs */
151
152/* statistics */
153static VMM_STAT_DEFINE(VCPU_TOTAL_RUNTIME, "vcpu total runtime");
154
155static void
156vcpu_cleanup(struct vcpu *vcpu)
157{
158 vlapic_cleanup(vcpu->vlapic);
159 vmm_stat_free(vcpu->stats);
160 fpu_save_area_free(vcpu->guestfpu);
161}
162
163static void
164vcpu_init(struct vm *vm, uint32_t vcpu_id)
165{
166 struct vcpu *vcpu;
167
168 vcpu = &vm->vcpu[vcpu_id];
169
170 vcpu_lock_init(vcpu);
171 vcpu->hostcpu = NOCPU;
172 vcpu->vcpuid = vcpu_id;
173 vcpu->vlapic = vlapic_init(vm, vcpu_id);
174 vm_set_x2apic_state(vm, vcpu_id, X2APIC_ENABLED);
175 vcpu->guestfpu = fpu_save_area_alloc();
176 fpu_save_area_reset(vcpu->guestfpu);
177 vcpu->stats = vmm_stat_alloc();
178}
179
180struct vm_exit *
181vm_exitinfo(struct vm *vm, int cpuid)
182{
183 struct vcpu *vcpu;
184
185 if (cpuid < 0 || cpuid >= VM_MAXCPU)
186 panic("vm_exitinfo: invalid cpuid %d", cpuid);
187
188 vcpu = &vm->vcpu[cpuid];
189
190 return (&vcpu->exitinfo);
191}
192
193static int
194vmm_init(void)
195{
196 int error;
197
198 vmm_ipi_init();
199
200 error = vmm_mem_init();
201 if (error)
202 return (error);
203
204 if (vmm_is_intel())
205 ops = &vmm_ops_intel;
206 else if (vmm_is_amd())
207 ops = &vmm_ops_amd;
208 else
209 return (ENXIO);
210
211 vmm_msr_init();
212
213 return (VMM_INIT());
214}
215
216static int
217vmm_handler(module_t mod, int what, void *arg)
218{
219 int error;
220
221 switch (what) {
222 case MOD_LOAD:
223 vmmdev_init();
224 iommu_init();
225 error = vmm_init();
226 break;
227 case MOD_UNLOAD:
228 error = vmmdev_cleanup();
229 if (error == 0) {
230 iommu_cleanup();
231 vmm_ipi_cleanup();
232 error = VMM_CLEANUP();
233 }
234 break;
235 default:
236 error = 0;
237 break;
238 }
239 return (error);
240}
241
242static moduledata_t vmm_kmod = {
243 "vmm",
244 vmm_handler,
245 NULL
246};
247
248/*
249 * Execute the module load handler after the pci passthru driver has had
250 * a chance to claim devices. We need this information at the time we do
251 * iommu initialization.
252 */
253DECLARE_MODULE(vmm, vmm_kmod, SI_SUB_CONFIGURE + 1, SI_ORDER_ANY);
254MODULE_VERSION(vmm, 1);
255
256SYSCTL_NODE(_hw, OID_AUTO, vmm, CTLFLAG_RW, NULL, NULL);
257
258struct vm *
259vm_create(const char *name)
260{
261 int i;
262 struct vm *vm;
263 vm_paddr_t maxaddr;
264
265 const int BSP = 0;
266
267 if (name == NULL || strlen(name) >= VM_MAX_NAMELEN)
268 return (NULL);
269
270 vm = malloc(sizeof(struct vm), M_VM, M_WAITOK | M_ZERO);
271 strcpy(vm->name, name);
272 vm->cookie = VMINIT(vm);
273
274 for (i = 0; i < VM_MAXCPU; i++) {
275 vcpu_init(vm, i);
276 guest_msrs_init(vm, i);
277 }
278
279 maxaddr = vmm_mem_maxaddr();
280 vm->iommu = iommu_create_domain(maxaddr);
281 vm_activate_cpu(vm, BSP);
282
283 return (vm);
284}
285
286static void
287vm_free_mem_seg(struct vm *vm, struct vm_memory_segment *seg)
288{
289 size_t len;
290 vm_paddr_t hpa;
291 void *host_domain;
292
293 host_domain = iommu_host_domain();
294
295 len = 0;
296 while (len < seg->len) {
297 hpa = vm_gpa2hpa(vm, seg->gpa + len, PAGE_SIZE);
298 if (hpa == (vm_paddr_t)-1) {
299 panic("vm_free_mem_segs: cannot free hpa "
300 "associated with gpa 0x%016lx", seg->gpa + len);
301 }
302
303 /*
304 * Remove the 'gpa' to 'hpa' mapping in VMs domain.
305 * And resurrect the 1:1 mapping for 'hpa' in 'host_domain'.
306 */
307 iommu_remove_mapping(vm->iommu, seg->gpa + len, PAGE_SIZE);
308 iommu_create_mapping(host_domain, hpa, hpa, PAGE_SIZE);
309
310 vmm_mem_free(hpa, PAGE_SIZE);
311
312 len += PAGE_SIZE;
313 }
314
315 /*
316 * Invalidate cached translations associated with 'vm->iommu' since
317 * we have now moved some pages from it.
318 */
319 iommu_invalidate_tlb(vm->iommu);
320
321 bzero(seg, sizeof(struct vm_memory_segment));
322}
323
324void
325vm_destroy(struct vm *vm)
326{
327 int i;
328
329 ppt_unassign_all(vm);
330
331 for (i = 0; i < vm->num_mem_segs; i++)
332 vm_free_mem_seg(vm, &vm->mem_segs[i]);
333
334 vm->num_mem_segs = 0;
335
336 for (i = 0; i < VM_MAXCPU; i++)
337 vcpu_cleanup(&vm->vcpu[i]);
338
339 iommu_destroy_domain(vm->iommu);
340
341 VMCLEANUP(vm->cookie);
342
343 free(vm, M_VM);
344}
345
346const char *
347vm_name(struct vm *vm)
348{
349 return (vm->name);
350}
351
352int
353vm_map_mmio(struct vm *vm, vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
354{
355 const boolean_t spok = TRUE; /* superpage mappings are ok */
356
357 return (VMMMAP_SET(vm->cookie, gpa, hpa, len, VM_MEMATTR_UNCACHEABLE,
358 VM_PROT_RW, spok));
359}
360
361int
362vm_unmap_mmio(struct vm *vm, vm_paddr_t gpa, size_t len)
363{
364 const boolean_t spok = TRUE; /* superpage mappings are ok */
365
366 return (VMMMAP_SET(vm->cookie, gpa, 0, len, 0,
367 VM_PROT_NONE, spok));
368}
369
370/*
371 * Returns TRUE if 'gpa' is available for allocation and FALSE otherwise
372 */
373static boolean_t
374vm_gpa_available(struct vm *vm, vm_paddr_t gpa)
375{
376 int i;
377 vm_paddr_t gpabase, gpalimit;
378
379 if (gpa & PAGE_MASK)
380 panic("vm_gpa_available: gpa (0x%016lx) not page aligned", gpa);
381
382 for (i = 0; i < vm->num_mem_segs; i++) {
383 gpabase = vm->mem_segs[i].gpa;
384 gpalimit = gpabase + vm->mem_segs[i].len;
385 if (gpa >= gpabase && gpa < gpalimit)
386 return (FALSE);
387 }
388
389 return (TRUE);
390}
391
392int
393vm_malloc(struct vm *vm, vm_paddr_t gpa, size_t len)
394{
395 int error, available, allocated;
396 struct vm_memory_segment *seg;
397 vm_paddr_t g, hpa;
398 void *host_domain;
399
400 const boolean_t spok = TRUE; /* superpage mappings are ok */
401
402 if ((gpa & PAGE_MASK) || (len & PAGE_MASK) || len == 0)
403 return (EINVAL);
404
405 available = allocated = 0;
406 g = gpa;
407 while (g < gpa + len) {
408 if (vm_gpa_available(vm, g))
409 available++;
410 else
411 allocated++;
412
413 g += PAGE_SIZE;
414 }
415
416 /*
417 * If there are some allocated and some available pages in the address
418 * range then it is an error.
419 */
420 if (allocated && available)
421 return (EINVAL);
422
423 /*
424 * If the entire address range being requested has already been
425 * allocated then there isn't anything more to do.
426 */
427 if (allocated && available == 0)
428 return (0);
429
430 if (vm->num_mem_segs >= VM_MAX_MEMORY_SEGMENTS)
431 return (E2BIG);
432
433 host_domain = iommu_host_domain();
434
435 seg = &vm->mem_segs[vm->num_mem_segs];
436
437 error = 0;
438 seg->gpa = gpa;
439 seg->len = 0;
440 while (seg->len < len) {
441 hpa = vmm_mem_alloc(PAGE_SIZE);
442 if (hpa == 0) {
443 error = ENOMEM;
444 break;
445 }
446
447 error = VMMMAP_SET(vm->cookie, gpa + seg->len, hpa, PAGE_SIZE,
448 VM_MEMATTR_WRITE_BACK, VM_PROT_ALL, spok);
449 if (error)
450 break;
451
452 /*
453 * Remove the 1:1 mapping for 'hpa' from the 'host_domain'.
454 * Add mapping for 'gpa + seg->len' to 'hpa' in the VMs domain.
455 */
456 iommu_remove_mapping(host_domain, hpa, PAGE_SIZE);
457 iommu_create_mapping(vm->iommu, gpa + seg->len, hpa, PAGE_SIZE);
458
459 seg->len += PAGE_SIZE;
460 }
461
462 if (error) {
463 vm_free_mem_seg(vm, seg);
464 return (error);
465 }
466
467 /*
468 * Invalidate cached translations associated with 'host_domain' since
469 * we have now moved some pages from it.
470 */
471 iommu_invalidate_tlb(host_domain);
472
473 vm->num_mem_segs++;
474
475 return (0);
476}
477
478vm_paddr_t
479vm_gpa2hpa(struct vm *vm, vm_paddr_t gpa, size_t len)
480{
481 vm_paddr_t nextpage;
482
483 nextpage = rounddown(gpa + PAGE_SIZE, PAGE_SIZE);
484 if (len > nextpage - gpa)
485 panic("vm_gpa2hpa: invalid gpa/len: 0x%016lx/%lu", gpa, len);
486
487 return (VMMMAP_GET(vm->cookie, gpa));
488}
489
490int
491vm_gpabase2memseg(struct vm *vm, vm_paddr_t gpabase,
492 struct vm_memory_segment *seg)
493{
494 int i;
495
496 for (i = 0; i < vm->num_mem_segs; i++) {
497 if (gpabase == vm->mem_segs[i].gpa) {
498 *seg = vm->mem_segs[i];
499 return (0);
500 }
501 }
502 return (-1);
503}
504
505int
506vm_get_register(struct vm *vm, int vcpu, int reg, uint64_t *retval)
507{
508
509 if (vcpu < 0 || vcpu >= VM_MAXCPU)
510 return (EINVAL);
511
512 if (reg >= VM_REG_LAST)
513 return (EINVAL);
514
515 return (VMGETREG(vm->cookie, vcpu, reg, retval));
516}
517
518int
519vm_set_register(struct vm *vm, int vcpu, int reg, uint64_t val)
520{
521
522 if (vcpu < 0 || vcpu >= VM_MAXCPU)
523 return (EINVAL);
524
525 if (reg >= VM_REG_LAST)
526 return (EINVAL);
527
528 return (VMSETREG(vm->cookie, vcpu, reg, val));
529}
530
531static boolean_t
532is_descriptor_table(int reg)
533{
534
535 switch (reg) {
536 case VM_REG_GUEST_IDTR:
537 case VM_REG_GUEST_GDTR:
538 return (TRUE);
539 default:
540 return (FALSE);
541 }
542}
543
544static boolean_t
545is_segment_register(int reg)
546{
547
548 switch (reg) {
549 case VM_REG_GUEST_ES:
550 case VM_REG_GUEST_CS:
551 case VM_REG_GUEST_SS:
552 case VM_REG_GUEST_DS:
553 case VM_REG_GUEST_FS:
554 case VM_REG_GUEST_GS:
555 case VM_REG_GUEST_TR:
556 case VM_REG_GUEST_LDTR:
557 return (TRUE);
558 default:
559 return (FALSE);
560 }
561}
562
563int
564vm_get_seg_desc(struct vm *vm, int vcpu, int reg,
565 struct seg_desc *desc)
566{
567
568 if (vcpu < 0 || vcpu >= VM_MAXCPU)
569 return (EINVAL);
570
571 if (!is_segment_register(reg) && !is_descriptor_table(reg))
572 return (EINVAL);
573
574 return (VMGETDESC(vm->cookie, vcpu, reg, desc));
575}
576
577int
578vm_set_seg_desc(struct vm *vm, int vcpu, int reg,
579 struct seg_desc *desc)
580{
581 if (vcpu < 0 || vcpu >= VM_MAXCPU)
582 return (EINVAL);
583
584 if (!is_segment_register(reg) && !is_descriptor_table(reg))
585 return (EINVAL);
586
587 return (VMSETDESC(vm->cookie, vcpu, reg, desc));
588}
589
590int
591vm_get_pinning(struct vm *vm, int vcpuid, int *cpuid)
592{
593
594 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
595 return (EINVAL);
596
597 *cpuid = VCPU_PINCPU(vm, vcpuid);
598
599 return (0);
600}
601
602int
603vm_set_pinning(struct vm *vm, int vcpuid, int host_cpuid)
604{
605 struct thread *td;
606
607 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
608 return (EINVAL);
609
610 td = curthread; /* XXXSMP only safe when muxing vcpus */
611
612 /* unpin */
613 if (host_cpuid < 0) {
614 VCPU_UNPIN(vm, vcpuid);
615 thread_lock(td);
616 sched_unbind(td);
617 thread_unlock(td);
618 return (0);
619 }
620
621 if (CPU_ABSENT(host_cpuid))
622 return (EINVAL);
623
624 /*
625 * XXX we should check that 'host_cpuid' has not already been pinned
626 * by another vm.
627 */
628 thread_lock(td);
629 sched_bind(td, host_cpuid);
630 thread_unlock(td);
631 VCPU_PIN(vm, vcpuid, host_cpuid);
632
633 return (0);
634}
635
636static void
637restore_guest_fpustate(struct vcpu *vcpu)
638{
639
640 /* flush host state to the pcb */
641 fpuexit(curthread);
642 fpu_stop_emulating();
643 fpurestore(vcpu->guestfpu);
644}
645
646static void
647save_guest_fpustate(struct vcpu *vcpu)
648{
649
650 fpusave(vcpu->guestfpu);
651 fpu_start_emulating();
652}
653
654int
655vm_run(struct vm *vm, struct vm_run *vmrun)
656{
657 int error, vcpuid;
658 struct vcpu *vcpu;
659 struct pcb *pcb;
660 uint64_t tscval;
661
662 vcpuid = vmrun->cpuid;
663
664 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
665 return (EINVAL);
666
667 vcpu = &vm->vcpu[vcpuid];
668
669 critical_enter();
670
671 tscval = rdtsc();
672
673 pcb = PCPU_GET(curpcb);
674 set_pcb_flags(pcb, PCB_FULL_IRET);
675
676 restore_guest_msrs(vm, vcpuid);
677 restore_guest_fpustate(vcpu);
678
679 vcpu->hostcpu = curcpu;
680 error = VMRUN(vm->cookie, vcpuid, vmrun->rip);
681 vcpu->hostcpu = NOCPU;
682
683 save_guest_fpustate(vcpu);
684 restore_host_msrs(vm, vcpuid);
685
686 vmm_stat_incr(vm, vcpuid, VCPU_TOTAL_RUNTIME, rdtsc() - tscval);
687
688 /* copy the exit information */
689 bcopy(&vcpu->exitinfo, &vmrun->vm_exit, sizeof(struct vm_exit));
690
691 critical_exit();
692
693 return (error);
694}
695
696int
697vm_inject_event(struct vm *vm, int vcpuid, int type,
698 int vector, uint32_t code, int code_valid)
699{
700 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
701 return (EINVAL);
702
703 if ((type > VM_EVENT_NONE && type < VM_EVENT_MAX) == 0)
704 return (EINVAL);
705
706 if (vector < 0 || vector > 255)
707 return (EINVAL);
708
709 return (VMINJECT(vm->cookie, vcpuid, type, vector, code, code_valid));
710}
711
|
| 712VMM_STAT_DEFINE(VCPU_NMI_COUNT, "number of NMIs delivered to vcpu");
713
|
713int
| 714int
|
714vm_inject_nmi(struct vm *vm, int vcpu)
| 715vm_inject_nmi(struct vm *vm, int vcpuid)
|
715{
| 716{
|
716 int error;
| 717 struct vcpu *vcpu;
|
717
| 718
|
718 if (vcpu < 0 || vcpu >= VM_MAXCPU)
| 719 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
|
719 return (EINVAL);
720
| 720 return (EINVAL);
721
|
721 error = VMNMI(vm->cookie, vcpu);
722 vm_interrupt_hostcpu(vm, vcpu);
723 return (error);
| 722 vcpu = &vm->vcpu[vcpuid];
723
724 vcpu->nmi_pending = 1;
725 vm_interrupt_hostcpu(vm, vcpuid);
726 return (0);
|
724}
725
726int
| 727}
728
729int
|
| 730vm_nmi_pending(struct vm *vm, int vcpuid)
731{
732 struct vcpu *vcpu;
733
734 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
735 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
736
737 vcpu = &vm->vcpu[vcpuid];
738
739 return (vcpu->nmi_pending);
740}
741
742void
743vm_nmi_clear(struct vm *vm, int vcpuid)
744{
745 struct vcpu *vcpu;
746
747 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
748 panic("vm_nmi_pending: invalid vcpuid %d", vcpuid);
749
750 vcpu = &vm->vcpu[vcpuid];
751
752 if (vcpu->nmi_pending == 0)
753 panic("vm_nmi_clear: inconsistent nmi_pending state");
754
755 vcpu->nmi_pending = 0;
756 vmm_stat_incr(vm, vcpuid, VCPU_NMI_COUNT, 1);
757}
758
759int
|
727vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
728{
729 if (vcpu < 0 || vcpu >= VM_MAXCPU)
730 return (EINVAL);
731
732 if (type < 0 || type >= VM_CAP_MAX)
733 return (EINVAL);
734
735 return (VMGETCAP(vm->cookie, vcpu, type, retval));
736}
737
738int
739vm_set_capability(struct vm *vm, int vcpu, int type, int val)
740{
741 if (vcpu < 0 || vcpu >= VM_MAXCPU)
742 return (EINVAL);
743
744 if (type < 0 || type >= VM_CAP_MAX)
745 return (EINVAL);
746
747 return (VMSETCAP(vm->cookie, vcpu, type, val));
748}
749
750uint64_t *
751vm_guest_msrs(struct vm *vm, int cpu)
752{
753 return (vm->vcpu[cpu].guest_msrs);
754}
755
756struct vlapic *
757vm_lapic(struct vm *vm, int cpu)
758{
759 return (vm->vcpu[cpu].vlapic);
760}
761
762boolean_t
763vmm_is_pptdev(int bus, int slot, int func)
764{
765 int found, b, s, f, n;
766 char *val, *cp, *cp2;
767
768 /*
769 * setenv pptdevs "1/2/3 4/5/6 7/8/9 10/11/12"
770 */
771 found = 0;
772 cp = val = getenv("pptdevs");
773 while (cp != NULL && *cp != '\0') {
774 if ((cp2 = strchr(cp, ' ')) != NULL)
775 *cp2 = '\0';
776
777 n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
778 if (n == 3 && bus == b && slot == s && func == f) {
779 found = 1;
780 break;
781 }
782
783 if (cp2 != NULL)
784 *cp2++ = ' ';
785
786 cp = cp2;
787 }
788 freeenv(val);
789 return (found);
790}
791
792void *
793vm_iommu_domain(struct vm *vm)
794{
795
796 return (vm->iommu);
797}
798
799int
800vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state state)
801{
802 int error;
803 struct vcpu *vcpu;
804
805 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
806 panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
807
808 vcpu = &vm->vcpu[vcpuid];
809
810 vcpu_lock(vcpu);
811
812 /*
813 * The following state transitions are allowed:
814 * IDLE -> RUNNING -> IDLE
815 * IDLE -> CANNOT_RUN -> IDLE
816 */
817 if ((vcpu->state == VCPU_IDLE && state != VCPU_IDLE) ||
818 (vcpu->state != VCPU_IDLE && state == VCPU_IDLE)) {
819 error = 0;
820 vcpu->state = state;
821 } else {
822 error = EBUSY;
823 }
824
825 vcpu_unlock(vcpu);
826
827 return (error);
828}
829
830enum vcpu_state
831vcpu_get_state(struct vm *vm, int vcpuid)
832{
833 struct vcpu *vcpu;
834 enum vcpu_state state;
835
836 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
837 panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
838
839 vcpu = &vm->vcpu[vcpuid];
840
841 vcpu_lock(vcpu);
842 state = vcpu->state;
843 vcpu_unlock(vcpu);
844
845 return (state);
846}
847
848void
849vm_activate_cpu(struct vm *vm, int vcpuid)
850{
851
852 if (vcpuid >= 0 && vcpuid < VM_MAXCPU)
853 CPU_SET(vcpuid, &vm->active_cpus);
854}
855
856cpuset_t
857vm_active_cpus(struct vm *vm)
858{
859
860 return (vm->active_cpus);
861}
862
863void *
864vcpu_stats(struct vm *vm, int vcpuid)
865{
866
867 return (vm->vcpu[vcpuid].stats);
868}
869
870int
871vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
872{
873 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
874 return (EINVAL);
875
876 *state = vm->vcpu[vcpuid].x2apic_state;
877
878 return (0);
879}
880
881int
882vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
883{
884 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
885 return (EINVAL);
886
887 if (state < 0 || state >= X2APIC_STATE_LAST)
888 return (EINVAL);
889
890 vm->vcpu[vcpuid].x2apic_state = state;
891
892 vlapic_set_x2apic_state(vm, vcpuid, state);
893
894 return (0);
895}
896
897void
898vm_interrupt_hostcpu(struct vm *vm, int vcpuid)
899{
900 int hostcpu;
901 struct vcpu *vcpu;
902
903 vcpu = &vm->vcpu[vcpuid];
904
905 /*
906 * XXX racy but the worst case is that we'll send an unnecessary IPI
907 * to the 'hostcpu'.
908 *
909 * We cannot use vcpu_is_running() here because it acquires vcpu->mtx
910 * which is not allowed inside a critical section.
911 */
912 hostcpu = vcpu->hostcpu;
913 if (hostcpu == NOCPU || hostcpu == curcpu)
914 return;
915
916 ipi_cpu(hostcpu, vmm_ipinum);
917}
| 760vm_get_capability(struct vm *vm, int vcpu, int type, int *retval)
761{
762 if (vcpu < 0 || vcpu >= VM_MAXCPU)
763 return (EINVAL);
764
765 if (type < 0 || type >= VM_CAP_MAX)
766 return (EINVAL);
767
768 return (VMGETCAP(vm->cookie, vcpu, type, retval));
769}
770
771int
772vm_set_capability(struct vm *vm, int vcpu, int type, int val)
773{
774 if (vcpu < 0 || vcpu >= VM_MAXCPU)
775 return (EINVAL);
776
777 if (type < 0 || type >= VM_CAP_MAX)
778 return (EINVAL);
779
780 return (VMSETCAP(vm->cookie, vcpu, type, val));
781}
782
783uint64_t *
784vm_guest_msrs(struct vm *vm, int cpu)
785{
786 return (vm->vcpu[cpu].guest_msrs);
787}
788
789struct vlapic *
790vm_lapic(struct vm *vm, int cpu)
791{
792 return (vm->vcpu[cpu].vlapic);
793}
794
795boolean_t
796vmm_is_pptdev(int bus, int slot, int func)
797{
798 int found, b, s, f, n;
799 char *val, *cp, *cp2;
800
801 /*
802 * setenv pptdevs "1/2/3 4/5/6 7/8/9 10/11/12"
803 */
804 found = 0;
805 cp = val = getenv("pptdevs");
806 while (cp != NULL && *cp != '\0') {
807 if ((cp2 = strchr(cp, ' ')) != NULL)
808 *cp2 = '\0';
809
810 n = sscanf(cp, "%d/%d/%d", &b, &s, &f);
811 if (n == 3 && bus == b && slot == s && func == f) {
812 found = 1;
813 break;
814 }
815
816 if (cp2 != NULL)
817 *cp2++ = ' ';
818
819 cp = cp2;
820 }
821 freeenv(val);
822 return (found);
823}
824
825void *
826vm_iommu_domain(struct vm *vm)
827{
828
829 return (vm->iommu);
830}
831
832int
833vcpu_set_state(struct vm *vm, int vcpuid, enum vcpu_state state)
834{
835 int error;
836 struct vcpu *vcpu;
837
838 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
839 panic("vm_set_run_state: invalid vcpuid %d", vcpuid);
840
841 vcpu = &vm->vcpu[vcpuid];
842
843 vcpu_lock(vcpu);
844
845 /*
846 * The following state transitions are allowed:
847 * IDLE -> RUNNING -> IDLE
848 * IDLE -> CANNOT_RUN -> IDLE
849 */
850 if ((vcpu->state == VCPU_IDLE && state != VCPU_IDLE) ||
851 (vcpu->state != VCPU_IDLE && state == VCPU_IDLE)) {
852 error = 0;
853 vcpu->state = state;
854 } else {
855 error = EBUSY;
856 }
857
858 vcpu_unlock(vcpu);
859
860 return (error);
861}
862
863enum vcpu_state
864vcpu_get_state(struct vm *vm, int vcpuid)
865{
866 struct vcpu *vcpu;
867 enum vcpu_state state;
868
869 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
870 panic("vm_get_run_state: invalid vcpuid %d", vcpuid);
871
872 vcpu = &vm->vcpu[vcpuid];
873
874 vcpu_lock(vcpu);
875 state = vcpu->state;
876 vcpu_unlock(vcpu);
877
878 return (state);
879}
880
881void
882vm_activate_cpu(struct vm *vm, int vcpuid)
883{
884
885 if (vcpuid >= 0 && vcpuid < VM_MAXCPU)
886 CPU_SET(vcpuid, &vm->active_cpus);
887}
888
889cpuset_t
890vm_active_cpus(struct vm *vm)
891{
892
893 return (vm->active_cpus);
894}
895
896void *
897vcpu_stats(struct vm *vm, int vcpuid)
898{
899
900 return (vm->vcpu[vcpuid].stats);
901}
902
903int
904vm_get_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state *state)
905{
906 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
907 return (EINVAL);
908
909 *state = vm->vcpu[vcpuid].x2apic_state;
910
911 return (0);
912}
913
914int
915vm_set_x2apic_state(struct vm *vm, int vcpuid, enum x2apic_state state)
916{
917 if (vcpuid < 0 || vcpuid >= VM_MAXCPU)
918 return (EINVAL);
919
920 if (state < 0 || state >= X2APIC_STATE_LAST)
921 return (EINVAL);
922
923 vm->vcpu[vcpuid].x2apic_state = state;
924
925 vlapic_set_x2apic_state(vm, vcpuid, state);
926
927 return (0);
928}
929
930void
931vm_interrupt_hostcpu(struct vm *vm, int vcpuid)
932{
933 int hostcpu;
934 struct vcpu *vcpu;
935
936 vcpu = &vm->vcpu[vcpuid];
937
938 /*
939 * XXX racy but the worst case is that we'll send an unnecessary IPI
940 * to the 'hostcpu'.
941 *
942 * We cannot use vcpu_is_running() here because it acquires vcpu->mtx
943 * which is not allowed inside a critical section.
944 */
945 hostcpu = vcpu->hostcpu;
946 if (hostcpu == NOCPU || hostcpu == curcpu)
947 return;
948
949 ipi_cpu(hostcpu, vmm_ipinum);
950}
|