1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 2011 NetApp, Inc.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #include <sys/cdefs.h>
30 #include <sys/param.h>
31 #include <sys/capsicum.h>
32 #include <sys/sysctl.h>
33 #include <sys/ioctl.h>
34 #include <sys/linker.h>
35 #include <sys/mman.h>
36 #include <sys/module.h>
37 #include <sys/_iovec.h>
38 #include <sys/cpuset.h>
39
40 #include <capsicum_helpers.h>
41 #include <errno.h>
42 #include <stdbool.h>
43 #include <stdio.h>
44 #include <stdlib.h>
45 #include <assert.h>
46 #include <string.h>
47 #include <fcntl.h>
48 #include <unistd.h>
49
50 #include <libutil.h>
51
52 #include <vm/vm.h>
53 #include <machine/vmm.h>
54 #include <machine/vmm_dev.h>
55 #include <machine/vmm_snapshot.h>
56
57 #include "vmmapi.h"
58
59 #define MB (1024 * 1024UL)
60 #define GB (1024 * 1024 * 1024UL)
61
62 /*
63 * Size of the guard region before and after the virtual address space
64 * mapping the guest physical memory. This must be a multiple of the
65 * superpage size for performance reasons.
66 */
67 #define VM_MMAP_GUARD_SIZE (4 * MB)
68
69 #define PROT_RW (PROT_READ | PROT_WRITE)
70 #define PROT_ALL (PROT_READ | PROT_WRITE | PROT_EXEC)
71
72 struct vmctx {
73 int fd;
74 uint32_t lowmem_limit;
75 int memflags;
76 size_t lowmem;
77 size_t highmem;
78 char *baseaddr;
79 char *name;
80 };
81
82 #define CREATE(x) sysctlbyname("hw.vmm.create", NULL, NULL, (x), strlen((x)))
83 #define DESTROY(x) sysctlbyname("hw.vmm.destroy", NULL, NULL, (x), strlen((x)))
84
85 static int
vm_device_open(const char * name)86 vm_device_open(const char *name)
87 {
88 int fd, len;
89 char *vmfile;
90
91 len = strlen("/dev/vmm/") + strlen(name) + 1;
92 vmfile = malloc(len);
93 assert(vmfile != NULL);
94 snprintf(vmfile, len, "/dev/vmm/%s", name);
95
96 /* Open the device file */
97 fd = open(vmfile, O_RDWR, 0);
98
99 free(vmfile);
100 return (fd);
101 }
102
103 int
vm_create(const char * name)104 vm_create(const char *name)
105 {
106 /* Try to load vmm(4) module before creating a guest. */
107 if (modfind("vmm") < 0)
108 kldload("vmm");
109 return (CREATE(name));
110 }
111
112 struct vmctx *
vm_open(const char * name)113 vm_open(const char *name)
114 {
115 struct vmctx *vm;
116 int saved_errno;
117
118 vm = malloc(sizeof(struct vmctx) + strlen(name) + 1);
119 assert(vm != NULL);
120
121 vm->fd = -1;
122 vm->memflags = 0;
123 vm->lowmem_limit = 3 * GB;
124 vm->name = (char *)(vm + 1);
125 strcpy(vm->name, name);
126
127 if ((vm->fd = vm_device_open(vm->name)) < 0)
128 goto err;
129
130 return (vm);
131 err:
132 saved_errno = errno;
133 free(vm);
134 errno = saved_errno;
135 return (NULL);
136 }
137
138 void
vm_close(struct vmctx * vm)139 vm_close(struct vmctx *vm)
140 {
141 assert(vm != NULL);
142
143 close(vm->fd);
144 free(vm);
145 }
146
147 void
vm_destroy(struct vmctx * vm)148 vm_destroy(struct vmctx *vm)
149 {
150 assert(vm != NULL);
151
152 if (vm->fd >= 0)
153 close(vm->fd);
154 DESTROY(vm->name);
155
156 free(vm);
157 }
158
159 int
vm_parse_memsize(const char * opt,size_t * ret_memsize)160 vm_parse_memsize(const char *opt, size_t *ret_memsize)
161 {
162 char *endptr;
163 size_t optval;
164 int error;
165
166 optval = strtoul(opt, &endptr, 0);
167 if (*opt != '\0' && *endptr == '\0') {
168 /*
169 * For the sake of backward compatibility if the memory size
170 * specified on the command line is less than a megabyte then
171 * it is interpreted as being in units of MB.
172 */
173 if (optval < MB)
174 optval *= MB;
175 *ret_memsize = optval;
176 error = 0;
177 } else
178 error = expand_number(opt, ret_memsize);
179
180 return (error);
181 }
182
183 uint32_t
vm_get_lowmem_limit(struct vmctx * ctx)184 vm_get_lowmem_limit(struct vmctx *ctx)
185 {
186
187 return (ctx->lowmem_limit);
188 }
189
190 void
vm_set_lowmem_limit(struct vmctx * ctx,uint32_t limit)191 vm_set_lowmem_limit(struct vmctx *ctx, uint32_t limit)
192 {
193
194 ctx->lowmem_limit = limit;
195 }
196
197 void
vm_set_memflags(struct vmctx * ctx,int flags)198 vm_set_memflags(struct vmctx *ctx, int flags)
199 {
200
201 ctx->memflags = flags;
202 }
203
204 int
vm_get_memflags(struct vmctx * ctx)205 vm_get_memflags(struct vmctx *ctx)
206 {
207
208 return (ctx->memflags);
209 }
210
211 /*
212 * Map segment 'segid' starting at 'off' into guest address range [gpa,gpa+len).
213 */
214 int
vm_mmap_memseg(struct vmctx * ctx,vm_paddr_t gpa,int segid,vm_ooffset_t off,size_t len,int prot)215 vm_mmap_memseg(struct vmctx *ctx, vm_paddr_t gpa, int segid, vm_ooffset_t off,
216 size_t len, int prot)
217 {
218 struct vm_memmap memmap;
219 int error, flags;
220
221 memmap.gpa = gpa;
222 memmap.segid = segid;
223 memmap.segoff = off;
224 memmap.len = len;
225 memmap.prot = prot;
226 memmap.flags = 0;
227
228 if (ctx->memflags & VM_MEM_F_WIRED)
229 memmap.flags |= VM_MEMMAP_F_WIRED;
230
231 /*
232 * If this mapping already exists then don't create it again. This
233 * is the common case for SYSMEM mappings created by bhyveload(8).
234 */
235 error = vm_mmap_getnext(ctx, &gpa, &segid, &off, &len, &prot, &flags);
236 if (error == 0 && gpa == memmap.gpa) {
237 if (segid != memmap.segid || off != memmap.segoff ||
238 prot != memmap.prot || flags != memmap.flags) {
239 errno = EEXIST;
240 return (-1);
241 } else {
242 return (0);
243 }
244 }
245
246 error = ioctl(ctx->fd, VM_MMAP_MEMSEG, &memmap);
247 return (error);
248 }
249
250 int
vm_get_guestmem_from_ctx(struct vmctx * ctx,char ** guest_baseaddr,size_t * lowmem_size,size_t * highmem_size)251 vm_get_guestmem_from_ctx(struct vmctx *ctx, char **guest_baseaddr,
252 size_t *lowmem_size, size_t *highmem_size)
253 {
254
255 *guest_baseaddr = ctx->baseaddr;
256 *lowmem_size = ctx->lowmem;
257 *highmem_size = ctx->highmem;
258 return (0);
259 }
260
261 int
vm_munmap_memseg(struct vmctx * ctx,vm_paddr_t gpa,size_t len)262 vm_munmap_memseg(struct vmctx *ctx, vm_paddr_t gpa, size_t len)
263 {
264 struct vm_munmap munmap;
265 int error;
266
267 munmap.gpa = gpa;
268 munmap.len = len;
269
270 error = ioctl(ctx->fd, VM_MUNMAP_MEMSEG, &munmap);
271 return (error);
272 }
273
274 int
vm_mmap_getnext(struct vmctx * ctx,vm_paddr_t * gpa,int * segid,vm_ooffset_t * segoff,size_t * len,int * prot,int * flags)275 vm_mmap_getnext(struct vmctx *ctx, vm_paddr_t *gpa, int *segid,
276 vm_ooffset_t *segoff, size_t *len, int *prot, int *flags)
277 {
278 struct vm_memmap memmap;
279 int error;
280
281 bzero(&memmap, sizeof(struct vm_memmap));
282 memmap.gpa = *gpa;
283 error = ioctl(ctx->fd, VM_MMAP_GETNEXT, &memmap);
284 if (error == 0) {
285 *gpa = memmap.gpa;
286 *segid = memmap.segid;
287 *segoff = memmap.segoff;
288 *len = memmap.len;
289 *prot = memmap.prot;
290 *flags = memmap.flags;
291 }
292 return (error);
293 }
294
295 /*
296 * Return 0 if the segments are identical and non-zero otherwise.
297 *
298 * This is slightly complicated by the fact that only device memory segments
299 * are named.
300 */
301 static int
cmpseg(size_t len,const char * str,size_t len2,const char * str2)302 cmpseg(size_t len, const char *str, size_t len2, const char *str2)
303 {
304
305 if (len == len2) {
306 if ((!str && !str2) || (str && str2 && !strcmp(str, str2)))
307 return (0);
308 }
309 return (-1);
310 }
311
312 static int
vm_alloc_memseg(struct vmctx * ctx,int segid,size_t len,const char * name)313 vm_alloc_memseg(struct vmctx *ctx, int segid, size_t len, const char *name)
314 {
315 struct vm_memseg memseg;
316 size_t n;
317 int error;
318
319 /*
320 * If the memory segment has already been created then just return.
321 * This is the usual case for the SYSMEM segment created by userspace
322 * loaders like bhyveload(8).
323 */
324 error = vm_get_memseg(ctx, segid, &memseg.len, memseg.name,
325 sizeof(memseg.name));
326 if (error)
327 return (error);
328
329 if (memseg.len != 0) {
330 if (cmpseg(len, name, memseg.len, VM_MEMSEG_NAME(&memseg))) {
331 errno = EINVAL;
332 return (-1);
333 } else {
334 return (0);
335 }
336 }
337
338 bzero(&memseg, sizeof(struct vm_memseg));
339 memseg.segid = segid;
340 memseg.len = len;
341 if (name != NULL) {
342 n = strlcpy(memseg.name, name, sizeof(memseg.name));
343 if (n >= sizeof(memseg.name)) {
344 errno = ENAMETOOLONG;
345 return (-1);
346 }
347 }
348
349 error = ioctl(ctx->fd, VM_ALLOC_MEMSEG, &memseg);
350 return (error);
351 }
352
353 int
vm_get_memseg(struct vmctx * ctx,int segid,size_t * lenp,char * namebuf,size_t bufsize)354 vm_get_memseg(struct vmctx *ctx, int segid, size_t *lenp, char *namebuf,
355 size_t bufsize)
356 {
357 struct vm_memseg memseg;
358 size_t n;
359 int error;
360
361 memseg.segid = segid;
362 error = ioctl(ctx->fd, VM_GET_MEMSEG, &memseg);
363 if (error == 0) {
364 *lenp = memseg.len;
365 n = strlcpy(namebuf, memseg.name, bufsize);
366 if (n >= bufsize) {
367 errno = ENAMETOOLONG;
368 error = -1;
369 }
370 }
371 return (error);
372 }
373
374 static int
setup_memory_segment(struct vmctx * ctx,vm_paddr_t gpa,size_t len,char * base)375 setup_memory_segment(struct vmctx *ctx, vm_paddr_t gpa, size_t len, char *base)
376 {
377 char *ptr;
378 int error, flags;
379
380 /* Map 'len' bytes starting at 'gpa' in the guest address space */
381 error = vm_mmap_memseg(ctx, gpa, VM_SYSMEM, gpa, len, PROT_ALL);
382 if (error)
383 return (error);
384
385 flags = MAP_SHARED | MAP_FIXED;
386 if ((ctx->memflags & VM_MEM_F_INCORE) == 0)
387 flags |= MAP_NOCORE;
388
389 /* mmap into the process address space on the host */
390 ptr = mmap(base + gpa, len, PROT_RW, flags, ctx->fd, gpa);
391 if (ptr == MAP_FAILED)
392 return (-1);
393
394 return (0);
395 }
396
397 int
vm_setup_memory(struct vmctx * ctx,size_t memsize,enum vm_mmap_style vms)398 vm_setup_memory(struct vmctx *ctx, size_t memsize, enum vm_mmap_style vms)
399 {
400 size_t objsize, len;
401 vm_paddr_t gpa;
402 char *baseaddr, *ptr;
403 int error;
404
405 assert(vms == VM_MMAP_ALL);
406
407 /*
408 * If 'memsize' cannot fit entirely in the 'lowmem' segment then
409 * create another 'highmem' segment above 4GB for the remainder.
410 */
411 if (memsize > ctx->lowmem_limit) {
412 ctx->lowmem = ctx->lowmem_limit;
413 ctx->highmem = memsize - ctx->lowmem_limit;
414 objsize = 4*GB + ctx->highmem;
415 } else {
416 ctx->lowmem = memsize;
417 ctx->highmem = 0;
418 objsize = ctx->lowmem;
419 }
420
421 error = vm_alloc_memseg(ctx, VM_SYSMEM, objsize, NULL);
422 if (error)
423 return (error);
424
425 /*
426 * Stake out a contiguous region covering the guest physical memory
427 * and the adjoining guard regions.
428 */
429 len = VM_MMAP_GUARD_SIZE + objsize + VM_MMAP_GUARD_SIZE;
430 ptr = mmap(NULL, len, PROT_NONE, MAP_GUARD | MAP_ALIGNED_SUPER, -1, 0);
431 if (ptr == MAP_FAILED)
432 return (-1);
433
434 baseaddr = ptr + VM_MMAP_GUARD_SIZE;
435 if (ctx->highmem > 0) {
436 gpa = 4*GB;
437 len = ctx->highmem;
438 error = setup_memory_segment(ctx, gpa, len, baseaddr);
439 if (error)
440 return (error);
441 }
442
443 if (ctx->lowmem > 0) {
444 gpa = 0;
445 len = ctx->lowmem;
446 error = setup_memory_segment(ctx, gpa, len, baseaddr);
447 if (error)
448 return (error);
449 }
450
451 ctx->baseaddr = baseaddr;
452
453 return (0);
454 }
455
456 /*
457 * Returns a non-NULL pointer if [gaddr, gaddr+len) is entirely contained in
458 * the lowmem or highmem regions.
459 *
460 * In particular return NULL if [gaddr, gaddr+len) falls in guest MMIO region.
461 * The instruction emulation code depends on this behavior.
462 */
463 void *
vm_map_gpa(struct vmctx * ctx,vm_paddr_t gaddr,size_t len)464 vm_map_gpa(struct vmctx *ctx, vm_paddr_t gaddr, size_t len)
465 {
466
467 if (ctx->lowmem > 0) {
468 if (gaddr < ctx->lowmem && len <= ctx->lowmem &&
469 gaddr + len <= ctx->lowmem)
470 return (ctx->baseaddr + gaddr);
471 }
472
473 if (ctx->highmem > 0) {
474 if (gaddr >= 4*GB) {
475 if (gaddr < 4*GB + ctx->highmem &&
476 len <= ctx->highmem &&
477 gaddr + len <= 4*GB + ctx->highmem)
478 return (ctx->baseaddr + gaddr);
479 }
480 }
481
482 return (NULL);
483 }
484
485 vm_paddr_t
vm_rev_map_gpa(struct vmctx * ctx,void * addr)486 vm_rev_map_gpa(struct vmctx *ctx, void *addr)
487 {
488 vm_paddr_t offaddr;
489
490 offaddr = (char *)addr - ctx->baseaddr;
491
492 if (ctx->lowmem > 0)
493 if (offaddr <= ctx->lowmem)
494 return (offaddr);
495
496 if (ctx->highmem > 0)
497 if (offaddr >= 4*GB && offaddr < 4*GB + ctx->highmem)
498 return (offaddr);
499
500 return ((vm_paddr_t)-1);
501 }
502
503 const char *
vm_get_name(struct vmctx * ctx)504 vm_get_name(struct vmctx *ctx)
505 {
506
507 return (ctx->name);
508 }
509
510 size_t
vm_get_lowmem_size(struct vmctx * ctx)511 vm_get_lowmem_size(struct vmctx *ctx)
512 {
513
514 return (ctx->lowmem);
515 }
516
517 size_t
vm_get_highmem_size(struct vmctx * ctx)518 vm_get_highmem_size(struct vmctx *ctx)
519 {
520
521 return (ctx->highmem);
522 }
523
524 void *
vm_create_devmem(struct vmctx * ctx,int segid,const char * name,size_t len)525 vm_create_devmem(struct vmctx *ctx, int segid, const char *name, size_t len)
526 {
527 char pathname[MAXPATHLEN];
528 size_t len2;
529 char *base, *ptr;
530 int fd, error, flags;
531
532 fd = -1;
533 ptr = MAP_FAILED;
534 if (name == NULL || strlen(name) == 0) {
535 errno = EINVAL;
536 goto done;
537 }
538
539 error = vm_alloc_memseg(ctx, segid, len, name);
540 if (error)
541 goto done;
542
543 strlcpy(pathname, "/dev/vmm.io/", sizeof(pathname));
544 strlcat(pathname, ctx->name, sizeof(pathname));
545 strlcat(pathname, ".", sizeof(pathname));
546 strlcat(pathname, name, sizeof(pathname));
547
548 fd = open(pathname, O_RDWR);
549 if (fd < 0)
550 goto done;
551
552 /*
553 * Stake out a contiguous region covering the device memory and the
554 * adjoining guard regions.
555 */
556 len2 = VM_MMAP_GUARD_SIZE + len + VM_MMAP_GUARD_SIZE;
557 base = mmap(NULL, len2, PROT_NONE, MAP_GUARD | MAP_ALIGNED_SUPER, -1,
558 0);
559 if (base == MAP_FAILED)
560 goto done;
561
562 flags = MAP_SHARED | MAP_FIXED;
563 if ((ctx->memflags & VM_MEM_F_INCORE) == 0)
564 flags |= MAP_NOCORE;
565
566 /* mmap the devmem region in the host address space */
567 ptr = mmap(base + VM_MMAP_GUARD_SIZE, len, PROT_RW, flags, fd, 0);
568 done:
569 if (fd >= 0)
570 close(fd);
571 return (ptr);
572 }
573
574 int
vm_set_desc(struct vmctx * ctx,int vcpu,int reg,uint64_t base,uint32_t limit,uint32_t access)575 vm_set_desc(struct vmctx *ctx, int vcpu, int reg,
576 uint64_t base, uint32_t limit, uint32_t access)
577 {
578 int error;
579 struct vm_seg_desc vmsegdesc;
580
581 bzero(&vmsegdesc, sizeof(vmsegdesc));
582 vmsegdesc.cpuid = vcpu;
583 vmsegdesc.regnum = reg;
584 vmsegdesc.desc.base = base;
585 vmsegdesc.desc.limit = limit;
586 vmsegdesc.desc.access = access;
587
588 error = ioctl(ctx->fd, VM_SET_SEGMENT_DESCRIPTOR, &vmsegdesc);
589 return (error);
590 }
591
592 int
vm_get_desc(struct vmctx * ctx,int vcpu,int reg,uint64_t * base,uint32_t * limit,uint32_t * access)593 vm_get_desc(struct vmctx *ctx, int vcpu, int reg,
594 uint64_t *base, uint32_t *limit, uint32_t *access)
595 {
596 int error;
597 struct vm_seg_desc vmsegdesc;
598
599 bzero(&vmsegdesc, sizeof(vmsegdesc));
600 vmsegdesc.cpuid = vcpu;
601 vmsegdesc.regnum = reg;
602
603 error = ioctl(ctx->fd, VM_GET_SEGMENT_DESCRIPTOR, &vmsegdesc);
604 if (error == 0) {
605 *base = vmsegdesc.desc.base;
606 *limit = vmsegdesc.desc.limit;
607 *access = vmsegdesc.desc.access;
608 }
609 return (error);
610 }
611
612 int
vm_get_seg_desc(struct vmctx * ctx,int vcpu,int reg,struct seg_desc * seg_desc)613 vm_get_seg_desc(struct vmctx *ctx, int vcpu, int reg, struct seg_desc *seg_desc)
614 {
615 int error;
616
617 error = vm_get_desc(ctx, vcpu, reg, &seg_desc->base, &seg_desc->limit,
618 &seg_desc->access);
619 return (error);
620 }
621
622 int
vm_set_register(struct vmctx * ctx,int vcpu,int reg,uint64_t val)623 vm_set_register(struct vmctx *ctx, int vcpu, int reg, uint64_t val)
624 {
625 int error;
626 struct vm_register vmreg;
627
628 bzero(&vmreg, sizeof(vmreg));
629 vmreg.cpuid = vcpu;
630 vmreg.regnum = reg;
631 vmreg.regval = val;
632
633 error = ioctl(ctx->fd, VM_SET_REGISTER, &vmreg);
634 return (error);
635 }
636
637 int
vm_get_register(struct vmctx * ctx,int vcpu,int reg,uint64_t * ret_val)638 vm_get_register(struct vmctx *ctx, int vcpu, int reg, uint64_t *ret_val)
639 {
640 int error;
641 struct vm_register vmreg;
642
643 bzero(&vmreg, sizeof(vmreg));
644 vmreg.cpuid = vcpu;
645 vmreg.regnum = reg;
646
647 error = ioctl(ctx->fd, VM_GET_REGISTER, &vmreg);
648 *ret_val = vmreg.regval;
649 return (error);
650 }
651
652 int
vm_set_register_set(struct vmctx * ctx,int vcpu,unsigned int count,const int * regnums,uint64_t * regvals)653 vm_set_register_set(struct vmctx *ctx, int vcpu, unsigned int count,
654 const int *regnums, uint64_t *regvals)
655 {
656 int error;
657 struct vm_register_set vmregset;
658
659 bzero(&vmregset, sizeof(vmregset));
660 vmregset.cpuid = vcpu;
661 vmregset.count = count;
662 vmregset.regnums = regnums;
663 vmregset.regvals = regvals;
664
665 error = ioctl(ctx->fd, VM_SET_REGISTER_SET, &vmregset);
666 return (error);
667 }
668
669 int
vm_get_register_set(struct vmctx * ctx,int vcpu,unsigned int count,const int * regnums,uint64_t * regvals)670 vm_get_register_set(struct vmctx *ctx, int vcpu, unsigned int count,
671 const int *regnums, uint64_t *regvals)
672 {
673 int error;
674 struct vm_register_set vmregset;
675
676 bzero(&vmregset, sizeof(vmregset));
677 vmregset.cpuid = vcpu;
678 vmregset.count = count;
679 vmregset.regnums = regnums;
680 vmregset.regvals = regvals;
681
682 error = ioctl(ctx->fd, VM_GET_REGISTER_SET, &vmregset);
683 return (error);
684 }
685
686 int
vm_run(struct vmctx * ctx,int vcpu,struct vm_exit * vmexit)687 vm_run(struct vmctx *ctx, int vcpu, struct vm_exit *vmexit)
688 {
689 int error;
690 struct vm_run vmrun;
691
692 bzero(&vmrun, sizeof(vmrun));
693 vmrun.cpuid = vcpu;
694
695 error = ioctl(ctx->fd, VM_RUN, &vmrun);
696 bcopy(&vmrun.vm_exit, vmexit, sizeof(struct vm_exit));
697 return (error);
698 }
699
700 int
vm_suspend(struct vmctx * ctx,enum vm_suspend_how how)701 vm_suspend(struct vmctx *ctx, enum vm_suspend_how how)
702 {
703 struct vm_suspend vmsuspend;
704
705 bzero(&vmsuspend, sizeof(vmsuspend));
706 vmsuspend.how = how;
707 return (ioctl(ctx->fd, VM_SUSPEND, &vmsuspend));
708 }
709
710 int
vm_reinit(struct vmctx * ctx)711 vm_reinit(struct vmctx *ctx)
712 {
713
714 return (ioctl(ctx->fd, VM_REINIT, 0));
715 }
716
717 int
vm_inject_exception(struct vmctx * ctx,int vcpu,int vector,int errcode_valid,uint32_t errcode,int restart_instruction)718 vm_inject_exception(struct vmctx *ctx, int vcpu, int vector, int errcode_valid,
719 uint32_t errcode, int restart_instruction)
720 {
721 struct vm_exception exc;
722
723 exc.cpuid = vcpu;
724 exc.vector = vector;
725 exc.error_code = errcode;
726 exc.error_code_valid = errcode_valid;
727 exc.restart_instruction = restart_instruction;
728
729 return (ioctl(ctx->fd, VM_INJECT_EXCEPTION, &exc));
730 }
731
732 int
vm_apicid2vcpu(struct vmctx * ctx __unused,int apicid)733 vm_apicid2vcpu(struct vmctx *ctx __unused, int apicid)
734 {
735 /*
736 * The apic id associated with the 'vcpu' has the same numerical value
737 * as the 'vcpu' itself.
738 */
739 return (apicid);
740 }
741
742 int
vm_lapic_irq(struct vmctx * ctx,int vcpu,int vector)743 vm_lapic_irq(struct vmctx *ctx, int vcpu, int vector)
744 {
745 struct vm_lapic_irq vmirq;
746
747 bzero(&vmirq, sizeof(vmirq));
748 vmirq.cpuid = vcpu;
749 vmirq.vector = vector;
750
751 return (ioctl(ctx->fd, VM_LAPIC_IRQ, &vmirq));
752 }
753
754 int
vm_lapic_local_irq(struct vmctx * ctx,int vcpu,int vector)755 vm_lapic_local_irq(struct vmctx *ctx, int vcpu, int vector)
756 {
757 struct vm_lapic_irq vmirq;
758
759 bzero(&vmirq, sizeof(vmirq));
760 vmirq.cpuid = vcpu;
761 vmirq.vector = vector;
762
763 return (ioctl(ctx->fd, VM_LAPIC_LOCAL_IRQ, &vmirq));
764 }
765
766 int
vm_lapic_msi(struct vmctx * ctx,uint64_t addr,uint64_t msg)767 vm_lapic_msi(struct vmctx *ctx, uint64_t addr, uint64_t msg)
768 {
769 struct vm_lapic_msi vmmsi;
770
771 bzero(&vmmsi, sizeof(vmmsi));
772 vmmsi.addr = addr;
773 vmmsi.msg = msg;
774
775 return (ioctl(ctx->fd, VM_LAPIC_MSI, &vmmsi));
776 }
777
778 int
vm_ioapic_assert_irq(struct vmctx * ctx,int irq)779 vm_ioapic_assert_irq(struct vmctx *ctx, int irq)
780 {
781 struct vm_ioapic_irq ioapic_irq;
782
783 bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq));
784 ioapic_irq.irq = irq;
785
786 return (ioctl(ctx->fd, VM_IOAPIC_ASSERT_IRQ, &ioapic_irq));
787 }
788
789 int
vm_ioapic_deassert_irq(struct vmctx * ctx,int irq)790 vm_ioapic_deassert_irq(struct vmctx *ctx, int irq)
791 {
792 struct vm_ioapic_irq ioapic_irq;
793
794 bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq));
795 ioapic_irq.irq = irq;
796
797 return (ioctl(ctx->fd, VM_IOAPIC_DEASSERT_IRQ, &ioapic_irq));
798 }
799
800 int
vm_ioapic_pulse_irq(struct vmctx * ctx,int irq)801 vm_ioapic_pulse_irq(struct vmctx *ctx, int irq)
802 {
803 struct vm_ioapic_irq ioapic_irq;
804
805 bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq));
806 ioapic_irq.irq = irq;
807
808 return (ioctl(ctx->fd, VM_IOAPIC_PULSE_IRQ, &ioapic_irq));
809 }
810
811 int
vm_ioapic_pincount(struct vmctx * ctx,int * pincount)812 vm_ioapic_pincount(struct vmctx *ctx, int *pincount)
813 {
814
815 return (ioctl(ctx->fd, VM_IOAPIC_PINCOUNT, pincount));
816 }
817
818 int
vm_readwrite_kernemu_device(struct vmctx * ctx,int vcpu,vm_paddr_t gpa,bool write,int size,uint64_t * value)819 vm_readwrite_kernemu_device(struct vmctx *ctx, int vcpu, vm_paddr_t gpa,
820 bool write, int size, uint64_t *value)
821 {
822 struct vm_readwrite_kernemu_device irp = {
823 .vcpuid = vcpu,
824 .access_width = fls(size) - 1,
825 .gpa = gpa,
826 .value = write ? *value : ~0ul,
827 };
828 long cmd = (write ? VM_SET_KERNEMU_DEV : VM_GET_KERNEMU_DEV);
829 int rc;
830
831 rc = ioctl(ctx->fd, cmd, &irp);
832 if (rc == 0 && !write)
833 *value = irp.value;
834 return (rc);
835 }
836
837 int
vm_isa_assert_irq(struct vmctx * ctx,int atpic_irq,int ioapic_irq)838 vm_isa_assert_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq)
839 {
840 struct vm_isa_irq isa_irq;
841
842 bzero(&isa_irq, sizeof(struct vm_isa_irq));
843 isa_irq.atpic_irq = atpic_irq;
844 isa_irq.ioapic_irq = ioapic_irq;
845
846 return (ioctl(ctx->fd, VM_ISA_ASSERT_IRQ, &isa_irq));
847 }
848
849 int
vm_isa_deassert_irq(struct vmctx * ctx,int atpic_irq,int ioapic_irq)850 vm_isa_deassert_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq)
851 {
852 struct vm_isa_irq isa_irq;
853
854 bzero(&isa_irq, sizeof(struct vm_isa_irq));
855 isa_irq.atpic_irq = atpic_irq;
856 isa_irq.ioapic_irq = ioapic_irq;
857
858 return (ioctl(ctx->fd, VM_ISA_DEASSERT_IRQ, &isa_irq));
859 }
860
861 int
vm_isa_pulse_irq(struct vmctx * ctx,int atpic_irq,int ioapic_irq)862 vm_isa_pulse_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq)
863 {
864 struct vm_isa_irq isa_irq;
865
866 bzero(&isa_irq, sizeof(struct vm_isa_irq));
867 isa_irq.atpic_irq = atpic_irq;
868 isa_irq.ioapic_irq = ioapic_irq;
869
870 return (ioctl(ctx->fd, VM_ISA_PULSE_IRQ, &isa_irq));
871 }
872
873 int
vm_isa_set_irq_trigger(struct vmctx * ctx,int atpic_irq,enum vm_intr_trigger trigger)874 vm_isa_set_irq_trigger(struct vmctx *ctx, int atpic_irq,
875 enum vm_intr_trigger trigger)
876 {
877 struct vm_isa_irq_trigger isa_irq_trigger;
878
879 bzero(&isa_irq_trigger, sizeof(struct vm_isa_irq_trigger));
880 isa_irq_trigger.atpic_irq = atpic_irq;
881 isa_irq_trigger.trigger = trigger;
882
883 return (ioctl(ctx->fd, VM_ISA_SET_IRQ_TRIGGER, &isa_irq_trigger));
884 }
885
886 int
vm_inject_nmi(struct vmctx * ctx,int vcpu)887 vm_inject_nmi(struct vmctx *ctx, int vcpu)
888 {
889 struct vm_nmi vmnmi;
890
891 bzero(&vmnmi, sizeof(vmnmi));
892 vmnmi.cpuid = vcpu;
893
894 return (ioctl(ctx->fd, VM_INJECT_NMI, &vmnmi));
895 }
896
897 static const char *capstrmap[] = {
898 [VM_CAP_HALT_EXIT] = "hlt_exit",
899 [VM_CAP_MTRAP_EXIT] = "mtrap_exit",
900 [VM_CAP_PAUSE_EXIT] = "pause_exit",
901 [VM_CAP_UNRESTRICTED_GUEST] = "unrestricted_guest",
902 [VM_CAP_ENABLE_INVPCID] = "enable_invpcid",
903 [VM_CAP_BPT_EXIT] = "bpt_exit",
904 };
905
906 int
vm_capability_name2type(const char * capname)907 vm_capability_name2type(const char *capname)
908 {
909 int i;
910
911 for (i = 0; i < (int)nitems(capstrmap); i++) {
912 if (strcmp(capstrmap[i], capname) == 0)
913 return (i);
914 }
915
916 return (-1);
917 }
918
919 const char *
vm_capability_type2name(int type)920 vm_capability_type2name(int type)
921 {
922 if (type >= 0 && type < (int)nitems(capstrmap))
923 return (capstrmap[type]);
924
925 return (NULL);
926 }
927
928 int
vm_get_capability(struct vmctx * ctx,int vcpu,enum vm_cap_type cap,int * retval)929 vm_get_capability(struct vmctx *ctx, int vcpu, enum vm_cap_type cap,
930 int *retval)
931 {
932 int error;
933 struct vm_capability vmcap;
934
935 bzero(&vmcap, sizeof(vmcap));
936 vmcap.cpuid = vcpu;
937 vmcap.captype = cap;
938
939 error = ioctl(ctx->fd, VM_GET_CAPABILITY, &vmcap);
940 *retval = vmcap.capval;
941 return (error);
942 }
943
944 int
vm_set_capability(struct vmctx * ctx,int vcpu,enum vm_cap_type cap,int val)945 vm_set_capability(struct vmctx *ctx, int vcpu, enum vm_cap_type cap, int val)
946 {
947 struct vm_capability vmcap;
948
949 bzero(&vmcap, sizeof(vmcap));
950 vmcap.cpuid = vcpu;
951 vmcap.captype = cap;
952 vmcap.capval = val;
953
954 return (ioctl(ctx->fd, VM_SET_CAPABILITY, &vmcap));
955 }
956
957 int
vm_assign_pptdev(struct vmctx * ctx,int bus,int slot,int func)958 vm_assign_pptdev(struct vmctx *ctx, int bus, int slot, int func)
959 {
960 struct vm_pptdev pptdev;
961
962 bzero(&pptdev, sizeof(pptdev));
963 pptdev.bus = bus;
964 pptdev.slot = slot;
965 pptdev.func = func;
966
967 return (ioctl(ctx->fd, VM_BIND_PPTDEV, &pptdev));
968 }
969
970 int
vm_unassign_pptdev(struct vmctx * ctx,int bus,int slot,int func)971 vm_unassign_pptdev(struct vmctx *ctx, int bus, int slot, int func)
972 {
973 struct vm_pptdev pptdev;
974
975 bzero(&pptdev, sizeof(pptdev));
976 pptdev.bus = bus;
977 pptdev.slot = slot;
978 pptdev.func = func;
979
980 return (ioctl(ctx->fd, VM_UNBIND_PPTDEV, &pptdev));
981 }
982
983 int
vm_map_pptdev_mmio(struct vmctx * ctx,int bus,int slot,int func,vm_paddr_t gpa,size_t len,vm_paddr_t hpa)984 vm_map_pptdev_mmio(struct vmctx *ctx, int bus, int slot, int func,
985 vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
986 {
987 struct vm_pptdev_mmio pptmmio;
988
989 bzero(&pptmmio, sizeof(pptmmio));
990 pptmmio.bus = bus;
991 pptmmio.slot = slot;
992 pptmmio.func = func;
993 pptmmio.gpa = gpa;
994 pptmmio.len = len;
995 pptmmio.hpa = hpa;
996
997 return (ioctl(ctx->fd, VM_MAP_PPTDEV_MMIO, &pptmmio));
998 }
999
1000 int
vm_unmap_pptdev_mmio(struct vmctx * ctx,int bus,int slot,int func,vm_paddr_t gpa,size_t len)1001 vm_unmap_pptdev_mmio(struct vmctx *ctx, int bus, int slot, int func,
1002 vm_paddr_t gpa, size_t len)
1003 {
1004 struct vm_pptdev_mmio pptmmio;
1005
1006 bzero(&pptmmio, sizeof(pptmmio));
1007 pptmmio.bus = bus;
1008 pptmmio.slot = slot;
1009 pptmmio.func = func;
1010 pptmmio.gpa = gpa;
1011 pptmmio.len = len;
1012
1013 return (ioctl(ctx->fd, VM_UNMAP_PPTDEV_MMIO, &pptmmio));
1014 }
1015
1016 int
vm_setup_pptdev_msi(struct vmctx * ctx,int vcpu,int bus,int slot,int func,uint64_t addr,uint64_t msg,int numvec)1017 vm_setup_pptdev_msi(struct vmctx *ctx, int vcpu, int bus, int slot, int func,
1018 uint64_t addr, uint64_t msg, int numvec)
1019 {
1020 struct vm_pptdev_msi pptmsi;
1021
1022 bzero(&pptmsi, sizeof(pptmsi));
1023 pptmsi.vcpu = vcpu;
1024 pptmsi.bus = bus;
1025 pptmsi.slot = slot;
1026 pptmsi.func = func;
1027 pptmsi.msg = msg;
1028 pptmsi.addr = addr;
1029 pptmsi.numvec = numvec;
1030
1031 return (ioctl(ctx->fd, VM_PPTDEV_MSI, &pptmsi));
1032 }
1033
1034 int
vm_setup_pptdev_msix(struct vmctx * ctx,int vcpu,int bus,int slot,int func,int idx,uint64_t addr,uint64_t msg,uint32_t vector_control)1035 vm_setup_pptdev_msix(struct vmctx *ctx, int vcpu, int bus, int slot, int func,
1036 int idx, uint64_t addr, uint64_t msg, uint32_t vector_control)
1037 {
1038 struct vm_pptdev_msix pptmsix;
1039
1040 bzero(&pptmsix, sizeof(pptmsix));
1041 pptmsix.vcpu = vcpu;
1042 pptmsix.bus = bus;
1043 pptmsix.slot = slot;
1044 pptmsix.func = func;
1045 pptmsix.idx = idx;
1046 pptmsix.msg = msg;
1047 pptmsix.addr = addr;
1048 pptmsix.vector_control = vector_control;
1049
1050 return ioctl(ctx->fd, VM_PPTDEV_MSIX, &pptmsix);
1051 }
1052
1053 int
vm_disable_pptdev_msix(struct vmctx * ctx,int bus,int slot,int func)1054 vm_disable_pptdev_msix(struct vmctx *ctx, int bus, int slot, int func)
1055 {
1056 struct vm_pptdev ppt;
1057
1058 bzero(&ppt, sizeof(ppt));
1059 ppt.bus = bus;
1060 ppt.slot = slot;
1061 ppt.func = func;
1062
1063 return ioctl(ctx->fd, VM_PPTDEV_DISABLE_MSIX, &ppt);
1064 }
1065
1066 uint64_t *
vm_get_stats(struct vmctx * ctx,int vcpu,struct timeval * ret_tv,int * ret_entries)1067 vm_get_stats(struct vmctx *ctx, int vcpu, struct timeval *ret_tv,
1068 int *ret_entries)
1069 {
1070 static _Thread_local uint64_t *stats_buf;
1071 static _Thread_local u_int stats_count;
1072 uint64_t *new_stats;
1073 struct vm_stats vmstats;
1074 u_int count, index;
1075 bool have_stats;
1076
1077 have_stats = false;
1078 vmstats.cpuid = vcpu;
1079 count = 0;
1080 for (index = 0;; index += nitems(vmstats.statbuf)) {
1081 vmstats.index = index;
1082 if (ioctl(ctx->fd, VM_STATS, &vmstats) != 0)
1083 break;
1084 if (stats_count < index + vmstats.num_entries) {
1085 new_stats = realloc(stats_buf,
1086 (index + vmstats.num_entries) * sizeof(uint64_t));
1087 if (new_stats == NULL) {
1088 errno = ENOMEM;
1089 return (NULL);
1090 }
1091 stats_count = index + vmstats.num_entries;
1092 stats_buf = new_stats;
1093 }
1094 memcpy(stats_buf + index, vmstats.statbuf,
1095 vmstats.num_entries * sizeof(uint64_t));
1096 count += vmstats.num_entries;
1097 have_stats = true;
1098
1099 if (vmstats.num_entries != nitems(vmstats.statbuf))
1100 break;
1101 }
1102 if (have_stats) {
1103 if (ret_entries)
1104 *ret_entries = count;
1105 if (ret_tv)
1106 *ret_tv = vmstats.tv;
1107 return (stats_buf);
1108 } else
1109 return (NULL);
1110 }
1111
1112 const char *
vm_get_stat_desc(struct vmctx * ctx,int index)1113 vm_get_stat_desc(struct vmctx *ctx, int index)
1114 {
1115 static struct vm_stat_desc statdesc;
1116
1117 statdesc.index = index;
1118 if (ioctl(ctx->fd, VM_STAT_DESC, &statdesc) == 0)
1119 return (statdesc.desc);
1120 else
1121 return (NULL);
1122 }
1123
1124 int
vm_get_x2apic_state(struct vmctx * ctx,int vcpu,enum x2apic_state * state)1125 vm_get_x2apic_state(struct vmctx *ctx, int vcpu, enum x2apic_state *state)
1126 {
1127 int error;
1128 struct vm_x2apic x2apic;
1129
1130 bzero(&x2apic, sizeof(x2apic));
1131 x2apic.cpuid = vcpu;
1132
1133 error = ioctl(ctx->fd, VM_GET_X2APIC_STATE, &x2apic);
1134 *state = x2apic.state;
1135 return (error);
1136 }
1137
1138 int
vm_set_x2apic_state(struct vmctx * ctx,int vcpu,enum x2apic_state state)1139 vm_set_x2apic_state(struct vmctx *ctx, int vcpu, enum x2apic_state state)
1140 {
1141 int error;
1142 struct vm_x2apic x2apic;
1143
1144 bzero(&x2apic, sizeof(x2apic));
1145 x2apic.cpuid = vcpu;
1146 x2apic.state = state;
1147
1148 error = ioctl(ctx->fd, VM_SET_X2APIC_STATE, &x2apic);
1149
1150 return (error);
1151 }
1152
1153 /*
1154 * From Intel Vol 3a:
1155 * Table 9-1. IA-32 Processor States Following Power-up, Reset or INIT
1156 */
1157 int
vcpu_reset(struct vmctx * vmctx,int vcpu)1158 vcpu_reset(struct vmctx *vmctx, int vcpu)
1159 {
1160 int error;
1161 uint64_t rflags, rip, cr0, cr4, zero, desc_base, rdx;
1162 uint32_t desc_access, desc_limit;
1163 uint16_t sel;
1164
1165 zero = 0;
1166
1167 rflags = 0x2;
1168 error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RFLAGS, rflags);
1169 if (error)
1170 goto done;
1171
1172 rip = 0xfff0;
1173 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RIP, rip)) != 0)
1174 goto done;
1175
1176 /*
1177 * According to Intels Software Developer Manual CR0 should be
1178 * initialized with CR0_ET | CR0_NW | CR0_CD but that crashes some
1179 * guests like Windows.
1180 */
1181 cr0 = CR0_NE;
1182 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR0, cr0)) != 0)
1183 goto done;
1184
1185 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR2, zero)) != 0)
1186 goto done;
1187
1188 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR3, zero)) != 0)
1189 goto done;
1190
1191 cr4 = 0;
1192 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR4, cr4)) != 0)
1193 goto done;
1194
1195 /*
1196 * CS: present, r/w, accessed, 16-bit, byte granularity, usable
1197 */
1198 desc_base = 0xffff0000;
1199 desc_limit = 0xffff;
1200 desc_access = 0x0093;
1201 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_CS,
1202 desc_base, desc_limit, desc_access);
1203 if (error)
1204 goto done;
1205
1206 sel = 0xf000;
1207 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CS, sel)) != 0)
1208 goto done;
1209
1210 /*
1211 * SS,DS,ES,FS,GS: present, r/w, accessed, 16-bit, byte granularity
1212 */
1213 desc_base = 0;
1214 desc_limit = 0xffff;
1215 desc_access = 0x0093;
1216 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_SS,
1217 desc_base, desc_limit, desc_access);
1218 if (error)
1219 goto done;
1220
1221 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_DS,
1222 desc_base, desc_limit, desc_access);
1223 if (error)
1224 goto done;
1225
1226 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_ES,
1227 desc_base, desc_limit, desc_access);
1228 if (error)
1229 goto done;
1230
1231 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_FS,
1232 desc_base, desc_limit, desc_access);
1233 if (error)
1234 goto done;
1235
1236 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_GS,
1237 desc_base, desc_limit, desc_access);
1238 if (error)
1239 goto done;
1240
1241 sel = 0;
1242 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_SS, sel)) != 0)
1243 goto done;
1244 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_DS, sel)) != 0)
1245 goto done;
1246 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_ES, sel)) != 0)
1247 goto done;
1248 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_FS, sel)) != 0)
1249 goto done;
1250 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_GS, sel)) != 0)
1251 goto done;
1252
1253 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_EFER, zero)) != 0)
1254 goto done;
1255
1256 /* General purpose registers */
1257 rdx = 0xf00;
1258 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RAX, zero)) != 0)
1259 goto done;
1260 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RBX, zero)) != 0)
1261 goto done;
1262 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RCX, zero)) != 0)
1263 goto done;
1264 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RDX, rdx)) != 0)
1265 goto done;
1266 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RSI, zero)) != 0)
1267 goto done;
1268 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RDI, zero)) != 0)
1269 goto done;
1270 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RBP, zero)) != 0)
1271 goto done;
1272 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RSP, zero)) != 0)
1273 goto done;
1274 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_R8, zero)) != 0)
1275 goto done;
1276 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_R9, zero)) != 0)
1277 goto done;
1278 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_R10, zero)) != 0)
1279 goto done;
1280 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_R11, zero)) != 0)
1281 goto done;
1282 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_R12, zero)) != 0)
1283 goto done;
1284 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_R13, zero)) != 0)
1285 goto done;
1286 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_R14, zero)) != 0)
1287 goto done;
1288 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_R15, zero)) != 0)
1289 goto done;
1290
1291 /* GDTR, IDTR */
1292 desc_base = 0;
1293 desc_limit = 0xffff;
1294 desc_access = 0;
1295 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_GDTR,
1296 desc_base, desc_limit, desc_access);
1297 if (error != 0)
1298 goto done;
1299
1300 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_IDTR,
1301 desc_base, desc_limit, desc_access);
1302 if (error != 0)
1303 goto done;
1304
1305 /* TR */
1306 desc_base = 0;
1307 desc_limit = 0xffff;
1308 desc_access = 0x0000008b;
1309 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_TR, 0, 0, desc_access);
1310 if (error)
1311 goto done;
1312
1313 sel = 0;
1314 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_TR, sel)) != 0)
1315 goto done;
1316
1317 /* LDTR */
1318 desc_base = 0;
1319 desc_limit = 0xffff;
1320 desc_access = 0x00000082;
1321 error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_LDTR, desc_base,
1322 desc_limit, desc_access);
1323 if (error)
1324 goto done;
1325
1326 sel = 0;
1327 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_LDTR, 0)) != 0)
1328 goto done;
1329
1330 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_DR6,
1331 0xffff0ff0)) != 0)
1332 goto done;
1333 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_DR7, 0x400)) !=
1334 0)
1335 goto done;
1336
1337 if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_INTR_SHADOW,
1338 zero)) != 0)
1339 goto done;
1340
1341 error = 0;
1342 done:
1343 return (error);
1344 }
1345
1346 int
vm_get_gpa_pmap(struct vmctx * ctx,uint64_t gpa,uint64_t * pte,int * num)1347 vm_get_gpa_pmap(struct vmctx *ctx, uint64_t gpa, uint64_t *pte, int *num)
1348 {
1349 int error, i;
1350 struct vm_gpa_pte gpapte;
1351
1352 bzero(&gpapte, sizeof(gpapte));
1353 gpapte.gpa = gpa;
1354
1355 error = ioctl(ctx->fd, VM_GET_GPA_PMAP, &gpapte);
1356
1357 if (error == 0) {
1358 *num = gpapte.ptenum;
1359 for (i = 0; i < gpapte.ptenum; i++)
1360 pte[i] = gpapte.pte[i];
1361 }
1362
1363 return (error);
1364 }
1365
1366 int
vm_get_hpet_capabilities(struct vmctx * ctx,uint32_t * capabilities)1367 vm_get_hpet_capabilities(struct vmctx *ctx, uint32_t *capabilities)
1368 {
1369 int error;
1370 struct vm_hpet_cap cap;
1371
1372 bzero(&cap, sizeof(struct vm_hpet_cap));
1373 error = ioctl(ctx->fd, VM_GET_HPET_CAPABILITIES, &cap);
1374 if (capabilities != NULL)
1375 *capabilities = cap.capabilities;
1376 return (error);
1377 }
1378
1379 int
vm_gla2gpa(struct vmctx * ctx,int vcpu,struct vm_guest_paging * paging,uint64_t gla,int prot,uint64_t * gpa,int * fault)1380 vm_gla2gpa(struct vmctx *ctx, int vcpu, struct vm_guest_paging *paging,
1381 uint64_t gla, int prot, uint64_t *gpa, int *fault)
1382 {
1383 struct vm_gla2gpa gg;
1384 int error;
1385
1386 bzero(&gg, sizeof(struct vm_gla2gpa));
1387 gg.vcpuid = vcpu;
1388 gg.prot = prot;
1389 gg.gla = gla;
1390 gg.paging = *paging;
1391
1392 error = ioctl(ctx->fd, VM_GLA2GPA, &gg);
1393 if (error == 0) {
1394 *fault = gg.fault;
1395 *gpa = gg.gpa;
1396 }
1397 return (error);
1398 }
1399
1400 int
vm_gla2gpa_nofault(struct vmctx * ctx,int vcpu,struct vm_guest_paging * paging,uint64_t gla,int prot,uint64_t * gpa,int * fault)1401 vm_gla2gpa_nofault(struct vmctx *ctx, int vcpu, struct vm_guest_paging *paging,
1402 uint64_t gla, int prot, uint64_t *gpa, int *fault)
1403 {
1404 struct vm_gla2gpa gg;
1405 int error;
1406
1407 bzero(&gg, sizeof(struct vm_gla2gpa));
1408 gg.vcpuid = vcpu;
1409 gg.prot = prot;
1410 gg.gla = gla;
1411 gg.paging = *paging;
1412
1413 error = ioctl(ctx->fd, VM_GLA2GPA_NOFAULT, &gg);
1414 if (error == 0) {
1415 *fault = gg.fault;
1416 *gpa = gg.gpa;
1417 }
1418 return (error);
1419 }
1420
1421 #ifndef min
1422 #define min(a,b) (((a) < (b)) ? (a) : (b))
1423 #endif
1424
1425 int
vm_copy_setup(struct vmctx * ctx,int vcpu,struct vm_guest_paging * paging,uint64_t gla,size_t len,int prot,struct iovec * iov,int iovcnt,int * fault)1426 vm_copy_setup(struct vmctx *ctx, int vcpu, struct vm_guest_paging *paging,
1427 uint64_t gla, size_t len, int prot, struct iovec *iov, int iovcnt,
1428 int *fault)
1429 {
1430 void *va;
1431 uint64_t gpa, off;
1432 int error, i, n;
1433
1434 for (i = 0; i < iovcnt; i++) {
1435 iov[i].iov_base = 0;
1436 iov[i].iov_len = 0;
1437 }
1438
1439 while (len) {
1440 assert(iovcnt > 0);
1441 error = vm_gla2gpa(ctx, vcpu, paging, gla, prot, &gpa, fault);
1442 if (error || *fault)
1443 return (error);
1444
1445 off = gpa & PAGE_MASK;
1446 n = MIN(len, PAGE_SIZE - off);
1447
1448 va = vm_map_gpa(ctx, gpa, n);
1449 if (va == NULL)
1450 return (EFAULT);
1451
1452 iov->iov_base = va;
1453 iov->iov_len = n;
1454 iov++;
1455 iovcnt--;
1456
1457 gla += n;
1458 len -= n;
1459 }
1460 return (0);
1461 }
1462
1463 void
vm_copy_teardown(struct iovec * iov __unused,int iovcnt __unused)1464 vm_copy_teardown(struct iovec *iov __unused, int iovcnt __unused)
1465 {
1466 /*
1467 * Intentionally empty. This is used by the instruction
1468 * emulation code shared with the kernel. The in-kernel
1469 * version of this is non-empty.
1470 */
1471 }
1472
1473 void
vm_copyin(struct iovec * iov,void * vp,size_t len)1474 vm_copyin(struct iovec *iov, void *vp, size_t len)
1475 {
1476 const char *src;
1477 char *dst;
1478 size_t n;
1479
1480 dst = vp;
1481 while (len) {
1482 assert(iov->iov_len);
1483 n = min(len, iov->iov_len);
1484 src = iov->iov_base;
1485 bcopy(src, dst, n);
1486
1487 iov++;
1488 dst += n;
1489 len -= n;
1490 }
1491 }
1492
1493 void
vm_copyout(const void * vp,struct iovec * iov,size_t len)1494 vm_copyout(const void *vp, struct iovec *iov, size_t len)
1495 {
1496 const char *src;
1497 char *dst;
1498 size_t n;
1499
1500 src = vp;
1501 while (len) {
1502 assert(iov->iov_len);
1503 n = min(len, iov->iov_len);
1504 dst = iov->iov_base;
1505 bcopy(src, dst, n);
1506
1507 iov++;
1508 src += n;
1509 len -= n;
1510 }
1511 }
1512
1513 static int
vm_get_cpus(struct vmctx * ctx,int which,cpuset_t * cpus)1514 vm_get_cpus(struct vmctx *ctx, int which, cpuset_t *cpus)
1515 {
1516 struct vm_cpuset vm_cpuset;
1517 int error;
1518
1519 bzero(&vm_cpuset, sizeof(struct vm_cpuset));
1520 vm_cpuset.which = which;
1521 vm_cpuset.cpusetsize = sizeof(cpuset_t);
1522 vm_cpuset.cpus = cpus;
1523
1524 error = ioctl(ctx->fd, VM_GET_CPUS, &vm_cpuset);
1525 return (error);
1526 }
1527
1528 int
vm_active_cpus(struct vmctx * ctx,cpuset_t * cpus)1529 vm_active_cpus(struct vmctx *ctx, cpuset_t *cpus)
1530 {
1531
1532 return (vm_get_cpus(ctx, VM_ACTIVE_CPUS, cpus));
1533 }
1534
1535 int
vm_suspended_cpus(struct vmctx * ctx,cpuset_t * cpus)1536 vm_suspended_cpus(struct vmctx *ctx, cpuset_t *cpus)
1537 {
1538
1539 return (vm_get_cpus(ctx, VM_SUSPENDED_CPUS, cpus));
1540 }
1541
1542 int
vm_debug_cpus(struct vmctx * ctx,cpuset_t * cpus)1543 vm_debug_cpus(struct vmctx *ctx, cpuset_t *cpus)
1544 {
1545
1546 return (vm_get_cpus(ctx, VM_DEBUG_CPUS, cpus));
1547 }
1548
1549 int
vm_activate_cpu(struct vmctx * ctx,int vcpu)1550 vm_activate_cpu(struct vmctx *ctx, int vcpu)
1551 {
1552 struct vm_activate_cpu ac;
1553 int error;
1554
1555 bzero(&ac, sizeof(struct vm_activate_cpu));
1556 ac.vcpuid = vcpu;
1557 error = ioctl(ctx->fd, VM_ACTIVATE_CPU, &ac);
1558 return (error);
1559 }
1560
1561 int
vm_suspend_cpu(struct vmctx * ctx,int vcpu)1562 vm_suspend_cpu(struct vmctx *ctx, int vcpu)
1563 {
1564 struct vm_activate_cpu ac;
1565 int error;
1566
1567 bzero(&ac, sizeof(struct vm_activate_cpu));
1568 ac.vcpuid = vcpu;
1569 error = ioctl(ctx->fd, VM_SUSPEND_CPU, &ac);
1570 return (error);
1571 }
1572
1573 int
vm_resume_cpu(struct vmctx * ctx,int vcpu)1574 vm_resume_cpu(struct vmctx *ctx, int vcpu)
1575 {
1576 struct vm_activate_cpu ac;
1577 int error;
1578
1579 bzero(&ac, sizeof(struct vm_activate_cpu));
1580 ac.vcpuid = vcpu;
1581 error = ioctl(ctx->fd, VM_RESUME_CPU, &ac);
1582 return (error);
1583 }
1584
1585 int
vm_get_intinfo(struct vmctx * ctx,int vcpu,uint64_t * info1,uint64_t * info2)1586 vm_get_intinfo(struct vmctx *ctx, int vcpu, uint64_t *info1, uint64_t *info2)
1587 {
1588 struct vm_intinfo vmii;
1589 int error;
1590
1591 bzero(&vmii, sizeof(struct vm_intinfo));
1592 vmii.vcpuid = vcpu;
1593 error = ioctl(ctx->fd, VM_GET_INTINFO, &vmii);
1594 if (error == 0) {
1595 *info1 = vmii.info1;
1596 *info2 = vmii.info2;
1597 }
1598 return (error);
1599 }
1600
1601 int
vm_set_intinfo(struct vmctx * ctx,int vcpu,uint64_t info1)1602 vm_set_intinfo(struct vmctx *ctx, int vcpu, uint64_t info1)
1603 {
1604 struct vm_intinfo vmii;
1605 int error;
1606
1607 bzero(&vmii, sizeof(struct vm_intinfo));
1608 vmii.vcpuid = vcpu;
1609 vmii.info1 = info1;
1610 error = ioctl(ctx->fd, VM_SET_INTINFO, &vmii);
1611 return (error);
1612 }
1613
1614 int
vm_rtc_write(struct vmctx * ctx,int offset,uint8_t value)1615 vm_rtc_write(struct vmctx *ctx, int offset, uint8_t value)
1616 {
1617 struct vm_rtc_data rtcdata;
1618 int error;
1619
1620 bzero(&rtcdata, sizeof(struct vm_rtc_data));
1621 rtcdata.offset = offset;
1622 rtcdata.value = value;
1623 error = ioctl(ctx->fd, VM_RTC_WRITE, &rtcdata);
1624 return (error);
1625 }
1626
1627 int
vm_rtc_read(struct vmctx * ctx,int offset,uint8_t * retval)1628 vm_rtc_read(struct vmctx *ctx, int offset, uint8_t *retval)
1629 {
1630 struct vm_rtc_data rtcdata;
1631 int error;
1632
1633 bzero(&rtcdata, sizeof(struct vm_rtc_data));
1634 rtcdata.offset = offset;
1635 error = ioctl(ctx->fd, VM_RTC_READ, &rtcdata);
1636 if (error == 0)
1637 *retval = rtcdata.value;
1638 return (error);
1639 }
1640
1641 int
vm_rtc_settime(struct vmctx * ctx,time_t secs)1642 vm_rtc_settime(struct vmctx *ctx, time_t secs)
1643 {
1644 struct vm_rtc_time rtctime;
1645 int error;
1646
1647 bzero(&rtctime, sizeof(struct vm_rtc_time));
1648 rtctime.secs = secs;
1649 error = ioctl(ctx->fd, VM_RTC_SETTIME, &rtctime);
1650 return (error);
1651 }
1652
1653 int
vm_rtc_gettime(struct vmctx * ctx,time_t * secs)1654 vm_rtc_gettime(struct vmctx *ctx, time_t *secs)
1655 {
1656 struct vm_rtc_time rtctime;
1657 int error;
1658
1659 bzero(&rtctime, sizeof(struct vm_rtc_time));
1660 error = ioctl(ctx->fd, VM_RTC_GETTIME, &rtctime);
1661 if (error == 0)
1662 *secs = rtctime.secs;
1663 return (error);
1664 }
1665
1666 int
vm_restart_instruction(struct vmctx * ctx,int vcpu)1667 vm_restart_instruction(struct vmctx *ctx, int vcpu)
1668 {
1669
1670 return (ioctl(ctx->fd, VM_RESTART_INSTRUCTION, &vcpu));
1671 }
1672
1673 int
vm_snapshot_req(struct vm_snapshot_meta * meta)1674 vm_snapshot_req(struct vm_snapshot_meta *meta)
1675 {
1676
1677 if (ioctl(meta->ctx->fd, VM_SNAPSHOT_REQ, meta) == -1) {
1678 #ifdef SNAPSHOT_DEBUG
1679 fprintf(stderr, "%s: snapshot failed for %s: %d\r\n",
1680 __func__, meta->dev_name, errno);
1681 #endif
1682 return (-1);
1683 }
1684 return (0);
1685 }
1686
1687 int
vm_restore_time(struct vmctx * ctx)1688 vm_restore_time(struct vmctx *ctx)
1689 {
1690 int dummy;
1691
1692 dummy = 0;
1693 return (ioctl(ctx->fd, VM_RESTORE_TIME, &dummy));
1694 }
1695
1696 int
vm_set_topology(struct vmctx * ctx,uint16_t sockets,uint16_t cores,uint16_t threads,uint16_t maxcpus)1697 vm_set_topology(struct vmctx *ctx,
1698 uint16_t sockets, uint16_t cores, uint16_t threads, uint16_t maxcpus)
1699 {
1700 struct vm_cpu_topology topology;
1701
1702 bzero(&topology, sizeof (struct vm_cpu_topology));
1703 topology.sockets = sockets;
1704 topology.cores = cores;
1705 topology.threads = threads;
1706 topology.maxcpus = maxcpus;
1707 return (ioctl(ctx->fd, VM_SET_TOPOLOGY, &topology));
1708 }
1709
1710 int
vm_get_topology(struct vmctx * ctx,uint16_t * sockets,uint16_t * cores,uint16_t * threads,uint16_t * maxcpus)1711 vm_get_topology(struct vmctx *ctx,
1712 uint16_t *sockets, uint16_t *cores, uint16_t *threads, uint16_t *maxcpus)
1713 {
1714 struct vm_cpu_topology topology;
1715 int error;
1716
1717 bzero(&topology, sizeof (struct vm_cpu_topology));
1718 error = ioctl(ctx->fd, VM_GET_TOPOLOGY, &topology);
1719 if (error == 0) {
1720 *sockets = topology.sockets;
1721 *cores = topology.cores;
1722 *threads = topology.threads;
1723 *maxcpus = topology.maxcpus;
1724 }
1725 return (error);
1726 }
1727
1728 /* Keep in sync with machine/vmm_dev.h. */
1729 static const cap_ioctl_t vm_ioctl_cmds[] = { VM_RUN, VM_SUSPEND, VM_REINIT,
1730 VM_ALLOC_MEMSEG, VM_GET_MEMSEG, VM_MMAP_MEMSEG, VM_MMAP_MEMSEG,
1731 VM_MMAP_GETNEXT, VM_MUNMAP_MEMSEG, VM_SET_REGISTER, VM_GET_REGISTER,
1732 VM_SET_SEGMENT_DESCRIPTOR, VM_GET_SEGMENT_DESCRIPTOR,
1733 VM_SET_REGISTER_SET, VM_GET_REGISTER_SET,
1734 VM_SET_KERNEMU_DEV, VM_GET_KERNEMU_DEV,
1735 VM_INJECT_EXCEPTION, VM_LAPIC_IRQ, VM_LAPIC_LOCAL_IRQ,
1736 VM_LAPIC_MSI, VM_IOAPIC_ASSERT_IRQ, VM_IOAPIC_DEASSERT_IRQ,
1737 VM_IOAPIC_PULSE_IRQ, VM_IOAPIC_PINCOUNT, VM_ISA_ASSERT_IRQ,
1738 VM_ISA_DEASSERT_IRQ, VM_ISA_PULSE_IRQ, VM_ISA_SET_IRQ_TRIGGER,
1739 VM_SET_CAPABILITY, VM_GET_CAPABILITY, VM_BIND_PPTDEV,
1740 VM_UNBIND_PPTDEV, VM_MAP_PPTDEV_MMIO, VM_PPTDEV_MSI,
1741 VM_PPTDEV_MSIX, VM_UNMAP_PPTDEV_MMIO, VM_PPTDEV_DISABLE_MSIX,
1742 VM_INJECT_NMI, VM_STATS, VM_STAT_DESC,
1743 VM_SET_X2APIC_STATE, VM_GET_X2APIC_STATE,
1744 VM_GET_HPET_CAPABILITIES, VM_GET_GPA_PMAP, VM_GLA2GPA,
1745 VM_GLA2GPA_NOFAULT,
1746 VM_ACTIVATE_CPU, VM_GET_CPUS, VM_SUSPEND_CPU, VM_RESUME_CPU,
1747 VM_SET_INTINFO, VM_GET_INTINFO,
1748 VM_RTC_WRITE, VM_RTC_READ, VM_RTC_SETTIME, VM_RTC_GETTIME,
1749 VM_RESTART_INSTRUCTION, VM_SET_TOPOLOGY, VM_GET_TOPOLOGY,
1750 VM_SNAPSHOT_REQ, VM_RESTORE_TIME
1751 };
1752
1753 int
vm_limit_rights(struct vmctx * ctx)1754 vm_limit_rights(struct vmctx *ctx)
1755 {
1756 cap_rights_t rights;
1757 size_t ncmds;
1758
1759 cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW);
1760 if (caph_rights_limit(ctx->fd, &rights) != 0)
1761 return (-1);
1762 ncmds = nitems(vm_ioctl_cmds);
1763 if (caph_ioctls_limit(ctx->fd, vm_ioctl_cmds, ncmds) != 0)
1764 return (-1);
1765 return (0);
1766 }
1767
1768 /*
1769 * Avoid using in new code. Operations on the fd should be wrapped here so that
1770 * capability rights can be kept in sync.
1771 */
1772 int
vm_get_device_fd(struct vmctx * ctx)1773 vm_get_device_fd(struct vmctx *ctx)
1774 {
1775
1776 return (ctx->fd);
1777 }
1778
1779 /* Legacy interface, do not use. */
1780 const cap_ioctl_t *
vm_get_ioctls(size_t * len)1781 vm_get_ioctls(size_t *len)
1782 {
1783 cap_ioctl_t *cmds;
1784
1785 if (len == NULL) {
1786 cmds = malloc(sizeof(vm_ioctl_cmds));
1787 if (cmds == NULL)
1788 return (NULL);
1789 bcopy(vm_ioctl_cmds, cmds, sizeof(vm_ioctl_cmds));
1790 return (cmds);
1791 }
1792
1793 *len = nitems(vm_ioctl_cmds);
1794 return (NULL);
1795 }
1796