xref: /illumos-gate/usr/src/lib/libvmmapi/common/vmmapi.c (revision a92282e44f968185a6bba094d1e5fece2da819cf)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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  * $FreeBSD$
29  */
30 /*
31  * This file and its contents are supplied under the terms of the
32  * Common Development and Distribution License ("CDDL"), version 1.0.
33  * You may only use this file in accordance with the terms of version
34  * 1.0 of the CDDL.
35  *
36  * A full copy of the text of the CDDL should have accompanied this
37  * source.  A copy of the CDDL is also available via the Internet at
38  * http://www.illumos.org/license/CDDL.
39  *
40  * Copyright 2015 Pluribus Networks Inc.
41  * Copyright 2019 Joyent, Inc.
42  * Copyright 2020 Oxide Computer Company
43  */
44 
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include <sys/param.h>
49 #include <sys/sysctl.h>
50 #include <sys/ioctl.h>
51 #ifdef	__FreeBSD__
52 #include <sys/linker.h>
53 #endif
54 #include <sys/mman.h>
55 #include <sys/module.h>
56 #include <sys/_iovec.h>
57 #include <sys/cpuset.h>
58 
59 #include <x86/segments.h>
60 #include <machine/specialreg.h>
61 
62 #include <errno.h>
63 #include <stdio.h>
64 #include <stdlib.h>
65 #include <assert.h>
66 #include <string.h>
67 #include <fcntl.h>
68 #include <unistd.h>
69 
70 #include <libutil.h>
71 
72 #include <machine/vmm.h>
73 #include <machine/vmm_dev.h>
74 
75 #include "vmmapi.h"
76 
77 #define	MB	(1024 * 1024UL)
78 #define	GB	(1024 * 1024 * 1024UL)
79 
80 #ifndef __FreeBSD__
81 /* shim to no-op for now */
82 #define	MAP_NOCORE		0
83 #define	MAP_ALIGNED_SUPER	0
84 
85 /* Rely on PROT_NONE for guard purposes */
86 #define	MAP_GUARD		(MAP_PRIVATE | MAP_ANON | MAP_NORESERVE)
87 #endif
88 
89 /*
90  * Size of the guard region before and after the virtual address space
91  * mapping the guest physical memory. This must be a multiple of the
92  * superpage size for performance reasons.
93  */
94 #define	VM_MMAP_GUARD_SIZE	(4 * MB)
95 
96 #define	PROT_RW		(PROT_READ | PROT_WRITE)
97 #define	PROT_ALL	(PROT_READ | PROT_WRITE | PROT_EXEC)
98 
99 struct vmctx {
100 	int	fd;
101 	uint32_t lowmem_limit;
102 	int	memflags;
103 	size_t	lowmem;
104 	size_t	highmem;
105 	char	*baseaddr;
106 	char	*name;
107 };
108 
109 #ifdef	__FreeBSD__
110 #define	CREATE(x)  sysctlbyname("hw.vmm.create", NULL, NULL, (x), strlen((x)))
111 #define	DESTROY(x) sysctlbyname("hw.vmm.destroy", NULL, NULL, (x), strlen((x)))
112 #else
113 #define	CREATE(x)	vm_do_ctl(VMM_CREATE_VM, (x))
114 #define	DESTROY(x)	vm_do_ctl(VMM_DESTROY_VM, (x))
115 
116 static int
117 vm_do_ctl(int cmd, const char *name)
118 {
119 	int ctl_fd;
120 
121 	ctl_fd = open(VMM_CTL_DEV, O_EXCL | O_RDWR);
122 	if (ctl_fd < 0) {
123 		return (-1);
124 	}
125 
126 	if (ioctl(ctl_fd, cmd, name) == -1) {
127 		int err = errno;
128 
129 		/* Do not lose ioctl errno through the close(2) */
130 		(void) close(ctl_fd);
131 		errno = err;
132 		return (-1);
133 	}
134 	(void) close(ctl_fd);
135 
136 	return (0);
137 }
138 #endif
139 
140 static int
141 vm_device_open(const char *name)
142 {
143 	int fd, len;
144 	char *vmfile;
145 
146 	len = strlen("/dev/vmm/") + strlen(name) + 1;
147 	vmfile = malloc(len);
148 	assert(vmfile != NULL);
149 	snprintf(vmfile, len, "/dev/vmm/%s", name);
150 
151 	/* Open the device file */
152 	fd = open(vmfile, O_RDWR, 0);
153 
154 	free(vmfile);
155 	return (fd);
156 }
157 
158 int
159 vm_create(const char *name)
160 {
161 #ifdef __FreeBSD__
162 	/* Try to load vmm(4) module before creating a guest. */
163 	if (modfind("vmm") < 0)
164 		kldload("vmm");
165 #endif
166 	return (CREATE((char *)name));
167 }
168 
169 struct vmctx *
170 vm_open(const char *name)
171 {
172 	struct vmctx *vm;
173 
174 	vm = malloc(sizeof(struct vmctx) + strlen(name) + 1);
175 	assert(vm != NULL);
176 
177 	vm->fd = -1;
178 	vm->memflags = 0;
179 	vm->lowmem_limit = 3 * GB;
180 	vm->name = (char *)(vm + 1);
181 	strcpy(vm->name, name);
182 
183 	if ((vm->fd = vm_device_open(vm->name)) < 0)
184 		goto err;
185 
186 	return (vm);
187 err:
188 #ifdef __FreeBSD__
189 	vm_destroy(vm);
190 #else
191 	/*
192 	 * As libvmmapi is used by other programs to query and control bhyve
193 	 * VMs, destroying a VM just because the open failed isn't useful. We
194 	 * have to free what we have allocated, though.
195 	 */
196 	free(vm);
197 #endif
198 	return (NULL);
199 }
200 
201 #ifndef __FreeBSD__
202 void
203 vm_close(struct vmctx *vm)
204 {
205 	assert(vm != NULL);
206 	assert(vm->fd >= 0);
207 
208 	(void) close(vm->fd);
209 
210 	free(vm);
211 }
212 #endif
213 
214 void
215 vm_destroy(struct vmctx *vm)
216 {
217 	assert(vm != NULL);
218 
219 	if (vm->fd >= 0)
220 		close(vm->fd);
221 	DESTROY(vm->name);
222 
223 	free(vm);
224 }
225 
226 int
227 vm_parse_memsize(const char *optarg, size_t *ret_memsize)
228 {
229 	char *endptr;
230 	size_t optval;
231 	int error;
232 
233 	optval = strtoul(optarg, &endptr, 0);
234 	if (*optarg != '\0' && *endptr == '\0') {
235 		/*
236 		 * For the sake of backward compatibility if the memory size
237 		 * specified on the command line is less than a megabyte then
238 		 * it is interpreted as being in units of MB.
239 		 */
240 		if (optval < MB)
241 			optval *= MB;
242 		*ret_memsize = optval;
243 		error = 0;
244 	} else
245 		error = expand_number(optarg, ret_memsize);
246 
247 	return (error);
248 }
249 
250 uint32_t
251 vm_get_lowmem_limit(struct vmctx *ctx)
252 {
253 
254 	return (ctx->lowmem_limit);
255 }
256 
257 void
258 vm_set_lowmem_limit(struct vmctx *ctx, uint32_t limit)
259 {
260 
261 	ctx->lowmem_limit = limit;
262 }
263 
264 void
265 vm_set_memflags(struct vmctx *ctx, int flags)
266 {
267 
268 	ctx->memflags = flags;
269 }
270 
271 int
272 vm_get_memflags(struct vmctx *ctx)
273 {
274 
275 	return (ctx->memflags);
276 }
277 
278 /*
279  * Map segment 'segid' starting at 'off' into guest address range [gpa,gpa+len).
280  */
281 int
282 vm_mmap_memseg(struct vmctx *ctx, vm_paddr_t gpa, int segid, vm_ooffset_t off,
283     size_t len, int prot)
284 {
285 	struct vm_memmap memmap;
286 	int error, flags;
287 
288 	memmap.gpa = gpa;
289 	memmap.segid = segid;
290 	memmap.segoff = off;
291 	memmap.len = len;
292 	memmap.prot = prot;
293 	memmap.flags = 0;
294 
295 	if (ctx->memflags & VM_MEM_F_WIRED)
296 		memmap.flags |= VM_MEMMAP_F_WIRED;
297 
298 	/*
299 	 * If this mapping already exists then don't create it again. This
300 	 * is the common case for SYSMEM mappings created by bhyveload(8).
301 	 */
302 	error = vm_mmap_getnext(ctx, &gpa, &segid, &off, &len, &prot, &flags);
303 	if (error == 0 && gpa == memmap.gpa) {
304 		if (segid != memmap.segid || off != memmap.segoff ||
305 		    prot != memmap.prot || flags != memmap.flags) {
306 			errno = EEXIST;
307 			return (-1);
308 		} else {
309 			return (0);
310 		}
311 	}
312 
313 	error = ioctl(ctx->fd, VM_MMAP_MEMSEG, &memmap);
314 	return (error);
315 }
316 
317 int
318 vm_mmap_getnext(struct vmctx *ctx, vm_paddr_t *gpa, int *segid,
319     vm_ooffset_t *segoff, size_t *len, int *prot, int *flags)
320 {
321 	struct vm_memmap memmap;
322 	int error;
323 
324 	bzero(&memmap, sizeof(struct vm_memmap));
325 	memmap.gpa = *gpa;
326 	error = ioctl(ctx->fd, VM_MMAP_GETNEXT, &memmap);
327 	if (error == 0) {
328 		*gpa = memmap.gpa;
329 		*segid = memmap.segid;
330 		*segoff = memmap.segoff;
331 		*len = memmap.len;
332 		*prot = memmap.prot;
333 		*flags = memmap.flags;
334 	}
335 	return (error);
336 }
337 
338 /*
339  * Return 0 if the segments are identical and non-zero otherwise.
340  *
341  * This is slightly complicated by the fact that only device memory segments
342  * are named.
343  */
344 static int
345 cmpseg(size_t len, const char *str, size_t len2, const char *str2)
346 {
347 
348 	if (len == len2) {
349 		if ((!str && !str2) || (str && str2 && !strcmp(str, str2)))
350 			return (0);
351 	}
352 	return (-1);
353 }
354 
355 static int
356 vm_alloc_memseg(struct vmctx *ctx, int segid, size_t len, const char *name)
357 {
358 	struct vm_memseg memseg;
359 	size_t n;
360 	int error;
361 
362 	/*
363 	 * If the memory segment has already been created then just return.
364 	 * This is the usual case for the SYSMEM segment created by userspace
365 	 * loaders like bhyveload(8).
366 	 */
367 	error = vm_get_memseg(ctx, segid, &memseg.len, memseg.name,
368 	    sizeof(memseg.name));
369 	if (error)
370 		return (error);
371 
372 	if (memseg.len != 0) {
373 		if (cmpseg(len, name, memseg.len, VM_MEMSEG_NAME(&memseg))) {
374 			errno = EINVAL;
375 			return (-1);
376 		} else {
377 			return (0);
378 		}
379 	}
380 
381 	bzero(&memseg, sizeof(struct vm_memseg));
382 	memseg.segid = segid;
383 	memseg.len = len;
384 	if (name != NULL) {
385 		n = strlcpy(memseg.name, name, sizeof(memseg.name));
386 		if (n >= sizeof(memseg.name)) {
387 			errno = ENAMETOOLONG;
388 			return (-1);
389 		}
390 	}
391 
392 	error = ioctl(ctx->fd, VM_ALLOC_MEMSEG, &memseg);
393 	return (error);
394 }
395 
396 int
397 vm_get_memseg(struct vmctx *ctx, int segid, size_t *lenp, char *namebuf,
398     size_t bufsize)
399 {
400 	struct vm_memseg memseg;
401 	size_t n;
402 	int error;
403 
404 	memseg.segid = segid;
405 	error = ioctl(ctx->fd, VM_GET_MEMSEG, &memseg);
406 	if (error == 0) {
407 		*lenp = memseg.len;
408 		n = strlcpy(namebuf, memseg.name, bufsize);
409 		if (n >= bufsize) {
410 			errno = ENAMETOOLONG;
411 			error = -1;
412 		}
413 	}
414 	return (error);
415 }
416 
417 static int
418 #ifdef __FreeBSD__
419 setup_memory_segment(struct vmctx *ctx, vm_paddr_t gpa, size_t len, char *base)
420 #else
421 setup_memory_segment(struct vmctx *ctx, int segid, vm_paddr_t gpa, size_t len,
422     char *base)
423 #endif
424 {
425 	char *ptr;
426 	int error, flags;
427 
428 	/* Map 'len' bytes starting at 'gpa' in the guest address space */
429 #ifdef __FreeBSD__
430 	error = vm_mmap_memseg(ctx, gpa, VM_SYSMEM, gpa, len, PROT_ALL);
431 #else
432 	/*
433 	 * As we use two segments for lowmem/highmem the offset within the
434 	 * segment is 0 on illumos.
435 	 */
436 	error = vm_mmap_memseg(ctx, gpa, segid, 0, len, PROT_ALL);
437 #endif
438 	if (error)
439 		return (error);
440 
441 	flags = MAP_SHARED | MAP_FIXED;
442 	if ((ctx->memflags & VM_MEM_F_INCORE) == 0)
443 		flags |= MAP_NOCORE;
444 
445 	/* mmap into the process address space on the host */
446 	ptr = mmap(base + gpa, len, PROT_RW, flags, ctx->fd, gpa);
447 	if (ptr == MAP_FAILED)
448 		return (-1);
449 
450 	return (0);
451 }
452 
453 int
454 vm_setup_memory(struct vmctx *ctx, size_t memsize, enum vm_mmap_style vms)
455 {
456 	size_t objsize, len;
457 	vm_paddr_t gpa;
458 	char *baseaddr, *ptr;
459 	int error;
460 
461 	assert(vms == VM_MMAP_ALL);
462 
463 	/*
464 	 * If 'memsize' cannot fit entirely in the 'lowmem' segment then
465 	 * create another 'highmem' segment above 4GB for the remainder.
466 	 */
467 	if (memsize > ctx->lowmem_limit) {
468 		ctx->lowmem = ctx->lowmem_limit;
469 		ctx->highmem = memsize - ctx->lowmem_limit;
470 		objsize = 4*GB + ctx->highmem;
471 	} else {
472 		ctx->lowmem = memsize;
473 		ctx->highmem = 0;
474 		objsize = ctx->lowmem;
475 	}
476 
477 #ifdef __FreeBSD__
478 	error = vm_alloc_memseg(ctx, VM_SYSMEM, objsize, NULL);
479 	if (error)
480 		return (error);
481 #endif
482 
483 	/*
484 	 * Stake out a contiguous region covering the guest physical memory
485 	 * and the adjoining guard regions.
486 	 */
487 	len = VM_MMAP_GUARD_SIZE + objsize + VM_MMAP_GUARD_SIZE;
488 	ptr = mmap(NULL, len, PROT_NONE, MAP_GUARD | MAP_ALIGNED_SUPER, -1, 0);
489 	if (ptr == MAP_FAILED)
490 		return (-1);
491 
492 	baseaddr = ptr + VM_MMAP_GUARD_SIZE;
493 
494 #ifdef __FreeBSD__
495 	if (ctx->highmem > 0) {
496 		gpa = 4*GB;
497 		len = ctx->highmem;
498 		error = setup_memory_segment(ctx, gpa, len, baseaddr);
499 		if (error)
500 			return (error);
501 	}
502 
503 	if (ctx->lowmem > 0) {
504 		gpa = 0;
505 		len = ctx->lowmem;
506 		error = setup_memory_segment(ctx, gpa, len, baseaddr);
507 		if (error)
508 			return (error);
509 	}
510 #else
511 	if (ctx->highmem > 0) {
512 		error = vm_alloc_memseg(ctx, VM_HIGHMEM, ctx->highmem, NULL);
513 		if (error)
514 			return (error);
515 		gpa = 4*GB;
516 		len = ctx->highmem;
517 		error = setup_memory_segment(ctx, VM_HIGHMEM, gpa, len, baseaddr);
518 		if (error)
519 			return (error);
520 	}
521 
522 	if (ctx->lowmem > 0) {
523 		error = vm_alloc_memseg(ctx, VM_LOWMEM, ctx->lowmem, NULL);
524 		if (error)
525 			return (error);
526 		gpa = 0;
527 		len = ctx->lowmem;
528 		error = setup_memory_segment(ctx, VM_LOWMEM, gpa, len, baseaddr);
529 		if (error)
530 			return (error);
531 	}
532 #endif
533 
534 	ctx->baseaddr = baseaddr;
535 
536 	return (0);
537 }
538 
539 /*
540  * Returns a non-NULL pointer if [gaddr, gaddr+len) is entirely contained in
541  * the lowmem or highmem regions.
542  *
543  * In particular return NULL if [gaddr, gaddr+len) falls in guest MMIO region.
544  * The instruction emulation code depends on this behavior.
545  */
546 void *
547 vm_map_gpa(struct vmctx *ctx, vm_paddr_t gaddr, size_t len)
548 {
549 
550 	if (ctx->lowmem > 0) {
551 		if (gaddr < ctx->lowmem && len <= ctx->lowmem &&
552 		    gaddr + len <= ctx->lowmem)
553 			return (ctx->baseaddr + gaddr);
554 	}
555 
556 	if (ctx->highmem > 0) {
557                 if (gaddr >= 4*GB) {
558 			if (gaddr < 4*GB + ctx->highmem &&
559 			    len <= ctx->highmem &&
560 			    gaddr + len <= 4*GB + ctx->highmem)
561 				return (ctx->baseaddr + gaddr);
562 		}
563 	}
564 
565 	return (NULL);
566 }
567 
568 size_t
569 vm_get_lowmem_size(struct vmctx *ctx)
570 {
571 
572 	return (ctx->lowmem);
573 }
574 
575 size_t
576 vm_get_highmem_size(struct vmctx *ctx)
577 {
578 
579 	return (ctx->highmem);
580 }
581 
582 #ifndef __FreeBSD__
583 int
584 vm_get_devmem_offset(struct vmctx *ctx, int segid, off_t *mapoff)
585 {
586 	struct vm_devmem_offset vdo;
587 	int error;
588 
589 	vdo.segid = segid;
590 	error = ioctl(ctx->fd, VM_DEVMEM_GETOFFSET, &vdo);
591 	if (error == 0)
592 		*mapoff = vdo.offset;
593 
594 	return (error);
595 }
596 #endif
597 
598 void *
599 vm_create_devmem(struct vmctx *ctx, int segid, const char *name, size_t len)
600 {
601 #ifdef	__FreeBSD__
602 	char pathname[MAXPATHLEN];
603 #endif
604 	size_t len2;
605 	char *base, *ptr;
606 	int fd, error, flags;
607 	off_t mapoff;
608 
609 	fd = -1;
610 	ptr = MAP_FAILED;
611 	if (name == NULL || strlen(name) == 0) {
612 		errno = EINVAL;
613 		goto done;
614 	}
615 
616 	error = vm_alloc_memseg(ctx, segid, len, name);
617 	if (error)
618 		goto done;
619 
620 #ifdef	__FreeBSD__
621 	strlcpy(pathname, "/dev/vmm.io/", sizeof(pathname));
622 	strlcat(pathname, ctx->name, sizeof(pathname));
623 	strlcat(pathname, ".", sizeof(pathname));
624 	strlcat(pathname, name, sizeof(pathname));
625 
626 	fd = open(pathname, O_RDWR);
627 	if (fd < 0)
628 		goto done;
629 #else
630 	if (vm_get_devmem_offset(ctx, segid, &mapoff) != 0)
631 		goto done;
632 #endif
633 
634 	/*
635 	 * Stake out a contiguous region covering the device memory and the
636 	 * adjoining guard regions.
637 	 */
638 	len2 = VM_MMAP_GUARD_SIZE + len + VM_MMAP_GUARD_SIZE;
639 	base = mmap(NULL, len2, PROT_NONE, MAP_GUARD | MAP_ALIGNED_SUPER, -1,
640 	    0);
641 	if (base == MAP_FAILED)
642 		goto done;
643 
644 	flags = MAP_SHARED | MAP_FIXED;
645 	if ((ctx->memflags & VM_MEM_F_INCORE) == 0)
646 		flags |= MAP_NOCORE;
647 
648 #ifdef	__FreeBSD__
649 	/* mmap the devmem region in the host address space */
650 	ptr = mmap(base + VM_MMAP_GUARD_SIZE, len, PROT_RW, flags, fd, 0);
651 #else
652 	/* mmap the devmem region in the host address space */
653 	ptr = mmap(base + VM_MMAP_GUARD_SIZE, len, PROT_RW, flags, ctx->fd,
654 	    mapoff);
655 #endif
656 done:
657 	if (fd >= 0)
658 		close(fd);
659 	return (ptr);
660 }
661 
662 int
663 vm_set_desc(struct vmctx *ctx, int vcpu, int reg,
664 	    uint64_t base, uint32_t limit, uint32_t access)
665 {
666 	int error;
667 	struct vm_seg_desc vmsegdesc;
668 
669 	bzero(&vmsegdesc, sizeof(vmsegdesc));
670 	vmsegdesc.cpuid = vcpu;
671 	vmsegdesc.regnum = reg;
672 	vmsegdesc.desc.base = base;
673 	vmsegdesc.desc.limit = limit;
674 	vmsegdesc.desc.access = access;
675 
676 	error = ioctl(ctx->fd, VM_SET_SEGMENT_DESCRIPTOR, &vmsegdesc);
677 	return (error);
678 }
679 
680 int
681 vm_get_desc(struct vmctx *ctx, int vcpu, int reg,
682 	    uint64_t *base, uint32_t *limit, uint32_t *access)
683 {
684 	int error;
685 	struct vm_seg_desc vmsegdesc;
686 
687 	bzero(&vmsegdesc, sizeof(vmsegdesc));
688 	vmsegdesc.cpuid = vcpu;
689 	vmsegdesc.regnum = reg;
690 
691 	error = ioctl(ctx->fd, VM_GET_SEGMENT_DESCRIPTOR, &vmsegdesc);
692 	if (error == 0) {
693 		*base = vmsegdesc.desc.base;
694 		*limit = vmsegdesc.desc.limit;
695 		*access = vmsegdesc.desc.access;
696 	}
697 	return (error);
698 }
699 
700 int
701 vm_get_seg_desc(struct vmctx *ctx, int vcpu, int reg, struct seg_desc *seg_desc)
702 {
703 	int error;
704 
705 	error = vm_get_desc(ctx, vcpu, reg, &seg_desc->base, &seg_desc->limit,
706 	    &seg_desc->access);
707 	return (error);
708 }
709 
710 int
711 vm_set_register(struct vmctx *ctx, int vcpu, int reg, uint64_t val)
712 {
713 	int error;
714 	struct vm_register vmreg;
715 
716 	bzero(&vmreg, sizeof(vmreg));
717 	vmreg.cpuid = vcpu;
718 	vmreg.regnum = reg;
719 	vmreg.regval = val;
720 
721 	error = ioctl(ctx->fd, VM_SET_REGISTER, &vmreg);
722 	return (error);
723 }
724 
725 int
726 vm_get_register(struct vmctx *ctx, int vcpu, int reg, uint64_t *ret_val)
727 {
728 	int error;
729 	struct vm_register vmreg;
730 
731 	bzero(&vmreg, sizeof(vmreg));
732 	vmreg.cpuid = vcpu;
733 	vmreg.regnum = reg;
734 
735 	error = ioctl(ctx->fd, VM_GET_REGISTER, &vmreg);
736 	*ret_val = vmreg.regval;
737 	return (error);
738 }
739 
740 int
741 vm_set_register_set(struct vmctx *ctx, int vcpu, unsigned int count,
742     const int *regnums, uint64_t *regvals)
743 {
744 	int error;
745 	struct vm_register_set vmregset;
746 
747 	bzero(&vmregset, sizeof(vmregset));
748 	vmregset.cpuid = vcpu;
749 	vmregset.count = count;
750 	vmregset.regnums = regnums;
751 	vmregset.regvals = regvals;
752 
753 	error = ioctl(ctx->fd, VM_SET_REGISTER_SET, &vmregset);
754 	return (error);
755 }
756 
757 int
758 vm_get_register_set(struct vmctx *ctx, int vcpu, unsigned int count,
759     const int *regnums, uint64_t *regvals)
760 {
761 	int error;
762 	struct vm_register_set vmregset;
763 
764 	bzero(&vmregset, sizeof(vmregset));
765 	vmregset.cpuid = vcpu;
766 	vmregset.count = count;
767 	vmregset.regnums = regnums;
768 	vmregset.regvals = regvals;
769 
770 	error = ioctl(ctx->fd, VM_GET_REGISTER_SET, &vmregset);
771 	return (error);
772 }
773 
774 int
775 vm_run(struct vmctx *ctx, int vcpu, const struct vm_entry *vm_entry,
776     struct vm_exit *vm_exit)
777 {
778 	struct vm_entry entry;
779 
780 	bcopy(vm_entry, &entry, sizeof (entry));
781 	entry.cpuid = vcpu;
782 	entry.exit_data = vm_exit;
783 
784 	return (ioctl(ctx->fd, VM_RUN, &entry));
785 }
786 
787 int
788 vm_suspend(struct vmctx *ctx, enum vm_suspend_how how)
789 {
790 	struct vm_suspend vmsuspend;
791 
792 	bzero(&vmsuspend, sizeof(vmsuspend));
793 	vmsuspend.how = how;
794 	return (ioctl(ctx->fd, VM_SUSPEND, &vmsuspend));
795 }
796 
797 int
798 vm_reinit(struct vmctx *ctx)
799 {
800 
801 	return (ioctl(ctx->fd, VM_REINIT, 0));
802 }
803 
804 int
805 vm_inject_exception(struct vmctx *ctx, int vcpu, int vector, int errcode_valid,
806     uint32_t errcode, int restart_instruction)
807 {
808 	struct vm_exception exc;
809 
810 	exc.cpuid = vcpu;
811 	exc.vector = vector;
812 	exc.error_code = errcode;
813 	exc.error_code_valid = errcode_valid;
814 	exc.restart_instruction = restart_instruction;
815 
816 	return (ioctl(ctx->fd, VM_INJECT_EXCEPTION, &exc));
817 }
818 
819 #ifndef __FreeBSD__
820 void
821 vm_inject_fault(struct vmctx *ctx, int vcpu, int vector, int errcode_valid,
822     int errcode)
823 {
824 	int error;
825 	struct vm_exception exc;
826 
827 	exc.cpuid = vcpu;
828 	exc.vector = vector;
829 	exc.error_code = errcode;
830 	exc.error_code_valid = errcode_valid;
831 	exc.restart_instruction = 1;
832 	error = ioctl(ctx->fd, VM_INJECT_EXCEPTION, &exc);
833 
834 	assert(error == 0);
835 }
836 #endif /* __FreeBSD__ */
837 
838 int
839 vm_apicid2vcpu(struct vmctx *ctx, int apicid)
840 {
841 	/*
842 	 * The apic id associated with the 'vcpu' has the same numerical value
843 	 * as the 'vcpu' itself.
844 	 */
845 	return (apicid);
846 }
847 
848 int
849 vm_lapic_irq(struct vmctx *ctx, int vcpu, int vector)
850 {
851 	struct vm_lapic_irq vmirq;
852 
853 	bzero(&vmirq, sizeof(vmirq));
854 	vmirq.cpuid = vcpu;
855 	vmirq.vector = vector;
856 
857 	return (ioctl(ctx->fd, VM_LAPIC_IRQ, &vmirq));
858 }
859 
860 int
861 vm_lapic_local_irq(struct vmctx *ctx, int vcpu, int vector)
862 {
863 	struct vm_lapic_irq vmirq;
864 
865 	bzero(&vmirq, sizeof(vmirq));
866 	vmirq.cpuid = vcpu;
867 	vmirq.vector = vector;
868 
869 	return (ioctl(ctx->fd, VM_LAPIC_LOCAL_IRQ, &vmirq));
870 }
871 
872 int
873 vm_lapic_msi(struct vmctx *ctx, uint64_t addr, uint64_t msg)
874 {
875 	struct vm_lapic_msi vmmsi;
876 
877 	bzero(&vmmsi, sizeof(vmmsi));
878 	vmmsi.addr = addr;
879 	vmmsi.msg = msg;
880 
881 	return (ioctl(ctx->fd, VM_LAPIC_MSI, &vmmsi));
882 }
883 
884 int
885 vm_ioapic_assert_irq(struct vmctx *ctx, int irq)
886 {
887 	struct vm_ioapic_irq ioapic_irq;
888 
889 	bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq));
890 	ioapic_irq.irq = irq;
891 
892 	return (ioctl(ctx->fd, VM_IOAPIC_ASSERT_IRQ, &ioapic_irq));
893 }
894 
895 int
896 vm_ioapic_deassert_irq(struct vmctx *ctx, int irq)
897 {
898 	struct vm_ioapic_irq ioapic_irq;
899 
900 	bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq));
901 	ioapic_irq.irq = irq;
902 
903 	return (ioctl(ctx->fd, VM_IOAPIC_DEASSERT_IRQ, &ioapic_irq));
904 }
905 
906 int
907 vm_ioapic_pulse_irq(struct vmctx *ctx, int irq)
908 {
909 	struct vm_ioapic_irq ioapic_irq;
910 
911 	bzero(&ioapic_irq, sizeof(struct vm_ioapic_irq));
912 	ioapic_irq.irq = irq;
913 
914 	return (ioctl(ctx->fd, VM_IOAPIC_PULSE_IRQ, &ioapic_irq));
915 }
916 
917 int
918 vm_ioapic_pincount(struct vmctx *ctx, int *pincount)
919 {
920 
921 	return (ioctl(ctx->fd, VM_IOAPIC_PINCOUNT, pincount));
922 }
923 
924 int
925 vm_readwrite_kernemu_device(struct vmctx *ctx, int vcpu, vm_paddr_t gpa,
926     bool write, int size, uint64_t *value)
927 {
928 	struct vm_readwrite_kernemu_device irp = {
929 		.vcpuid = vcpu,
930 		.access_width = fls(size) - 1,
931 		.gpa = gpa,
932 		.value = write ? *value : ~0ul,
933 	};
934 	long cmd = (write ? VM_SET_KERNEMU_DEV : VM_GET_KERNEMU_DEV);
935 	int rc;
936 
937 	rc = ioctl(ctx->fd, cmd, &irp);
938 	if (rc == 0 && !write)
939 		*value = irp.value;
940 	return (rc);
941 }
942 
943 int
944 vm_isa_assert_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq)
945 {
946 	struct vm_isa_irq isa_irq;
947 
948 	bzero(&isa_irq, sizeof(struct vm_isa_irq));
949 	isa_irq.atpic_irq = atpic_irq;
950 	isa_irq.ioapic_irq = ioapic_irq;
951 
952 	return (ioctl(ctx->fd, VM_ISA_ASSERT_IRQ, &isa_irq));
953 }
954 
955 int
956 vm_isa_deassert_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq)
957 {
958 	struct vm_isa_irq isa_irq;
959 
960 	bzero(&isa_irq, sizeof(struct vm_isa_irq));
961 	isa_irq.atpic_irq = atpic_irq;
962 	isa_irq.ioapic_irq = ioapic_irq;
963 
964 	return (ioctl(ctx->fd, VM_ISA_DEASSERT_IRQ, &isa_irq));
965 }
966 
967 int
968 vm_isa_pulse_irq(struct vmctx *ctx, int atpic_irq, int ioapic_irq)
969 {
970 	struct vm_isa_irq isa_irq;
971 
972 	bzero(&isa_irq, sizeof(struct vm_isa_irq));
973 	isa_irq.atpic_irq = atpic_irq;
974 	isa_irq.ioapic_irq = ioapic_irq;
975 
976 	return (ioctl(ctx->fd, VM_ISA_PULSE_IRQ, &isa_irq));
977 }
978 
979 int
980 vm_isa_set_irq_trigger(struct vmctx *ctx, int atpic_irq,
981     enum vm_intr_trigger trigger)
982 {
983 	struct vm_isa_irq_trigger isa_irq_trigger;
984 
985 	bzero(&isa_irq_trigger, sizeof(struct vm_isa_irq_trigger));
986 	isa_irq_trigger.atpic_irq = atpic_irq;
987 	isa_irq_trigger.trigger = trigger;
988 
989 	return (ioctl(ctx->fd, VM_ISA_SET_IRQ_TRIGGER, &isa_irq_trigger));
990 }
991 
992 int
993 vm_inject_nmi(struct vmctx *ctx, int vcpu)
994 {
995 	struct vm_nmi vmnmi;
996 
997 	bzero(&vmnmi, sizeof(vmnmi));
998 	vmnmi.cpuid = vcpu;
999 
1000 	return (ioctl(ctx->fd, VM_INJECT_NMI, &vmnmi));
1001 }
1002 
1003 static const char *capstrmap[] = {
1004 	[VM_CAP_HALT_EXIT]  = "hlt_exit",
1005 	[VM_CAP_MTRAP_EXIT] = "mtrap_exit",
1006 	[VM_CAP_PAUSE_EXIT] = "pause_exit",
1007 #ifdef __FreeBSD__
1008 	[VM_CAP_UNRESTRICTED_GUEST] = "unrestricted_guest",
1009 #endif
1010 	[VM_CAP_ENABLE_INVPCID] = "enable_invpcid",
1011 	[VM_CAP_BPT_EXIT] = "bpt_exit",
1012 };
1013 
1014 int
1015 vm_capability_name2type(const char *capname)
1016 {
1017 	int i;
1018 
1019 	for (i = 0; i < nitems(capstrmap); i++) {
1020 		if (strcmp(capstrmap[i], capname) == 0)
1021 			return (i);
1022 	}
1023 
1024 	return (-1);
1025 }
1026 
1027 const char *
1028 vm_capability_type2name(int type)
1029 {
1030 	if (type >= 0 && type < nitems(capstrmap))
1031 		return (capstrmap[type]);
1032 
1033 	return (NULL);
1034 }
1035 
1036 int
1037 vm_get_capability(struct vmctx *ctx, int vcpu, enum vm_cap_type cap,
1038 		  int *retval)
1039 {
1040 	int error;
1041 	struct vm_capability vmcap;
1042 
1043 	bzero(&vmcap, sizeof(vmcap));
1044 	vmcap.cpuid = vcpu;
1045 	vmcap.captype = cap;
1046 
1047 	error = ioctl(ctx->fd, VM_GET_CAPABILITY, &vmcap);
1048 	*retval = vmcap.capval;
1049 	return (error);
1050 }
1051 
1052 int
1053 vm_set_capability(struct vmctx *ctx, int vcpu, enum vm_cap_type cap, int val)
1054 {
1055 	struct vm_capability vmcap;
1056 
1057 	bzero(&vmcap, sizeof(vmcap));
1058 	vmcap.cpuid = vcpu;
1059 	vmcap.captype = cap;
1060 	vmcap.capval = val;
1061 
1062 	return (ioctl(ctx->fd, VM_SET_CAPABILITY, &vmcap));
1063 }
1064 
1065 #ifdef __FreeBSD__
1066 int
1067 vm_assign_pptdev(struct vmctx *ctx, int bus, int slot, int func)
1068 {
1069 	struct vm_pptdev pptdev;
1070 
1071 	bzero(&pptdev, sizeof(pptdev));
1072 	pptdev.bus = bus;
1073 	pptdev.slot = slot;
1074 	pptdev.func = func;
1075 
1076 	return (ioctl(ctx->fd, VM_BIND_PPTDEV, &pptdev));
1077 }
1078 
1079 int
1080 vm_unassign_pptdev(struct vmctx *ctx, int bus, int slot, int func)
1081 {
1082 	struct vm_pptdev pptdev;
1083 
1084 	bzero(&pptdev, sizeof(pptdev));
1085 	pptdev.bus = bus;
1086 	pptdev.slot = slot;
1087 	pptdev.func = func;
1088 
1089 	return (ioctl(ctx->fd, VM_UNBIND_PPTDEV, &pptdev));
1090 }
1091 
1092 int
1093 vm_map_pptdev_mmio(struct vmctx *ctx, int bus, int slot, int func,
1094 		   vm_paddr_t gpa, size_t len, vm_paddr_t hpa)
1095 {
1096 	struct vm_pptdev_mmio pptmmio;
1097 
1098 	bzero(&pptmmio, sizeof(pptmmio));
1099 	pptmmio.bus = bus;
1100 	pptmmio.slot = slot;
1101 	pptmmio.func = func;
1102 	pptmmio.gpa = gpa;
1103 	pptmmio.len = len;
1104 	pptmmio.hpa = hpa;
1105 
1106 	return (ioctl(ctx->fd, VM_MAP_PPTDEV_MMIO, &pptmmio));
1107 }
1108 
1109 int
1110 vm_setup_pptdev_msi(struct vmctx *ctx, int vcpu, int bus, int slot, int func,
1111     uint64_t addr, uint64_t msg, int numvec)
1112 {
1113 	struct vm_pptdev_msi pptmsi;
1114 
1115 	bzero(&pptmsi, sizeof(pptmsi));
1116 	pptmsi.vcpu = vcpu;
1117 	pptmsi.bus = bus;
1118 	pptmsi.slot = slot;
1119 	pptmsi.func = func;
1120 	pptmsi.msg = msg;
1121 	pptmsi.addr = addr;
1122 	pptmsi.numvec = numvec;
1123 
1124 	return (ioctl(ctx->fd, VM_PPTDEV_MSI, &pptmsi));
1125 }
1126 
1127 int
1128 vm_setup_pptdev_msix(struct vmctx *ctx, int vcpu, int bus, int slot, int func,
1129     int idx, uint64_t addr, uint64_t msg, uint32_t vector_control)
1130 {
1131 	struct vm_pptdev_msix pptmsix;
1132 
1133 	bzero(&pptmsix, sizeof(pptmsix));
1134 	pptmsix.vcpu = vcpu;
1135 	pptmsix.bus = bus;
1136 	pptmsix.slot = slot;
1137 	pptmsix.func = func;
1138 	pptmsix.idx = idx;
1139 	pptmsix.msg = msg;
1140 	pptmsix.addr = addr;
1141 	pptmsix.vector_control = vector_control;
1142 
1143 	return ioctl(ctx->fd, VM_PPTDEV_MSIX, &pptmsix);
1144 }
1145 
1146 int
1147 vm_get_pptdev_limits(struct vmctx *ctx, int bus, int slot, int func,
1148     int *msi_limit, int *msix_limit)
1149 {
1150 	struct vm_pptdev_limits pptlimits;
1151 	int error;
1152 
1153 	bzero(&pptlimits, sizeof (pptlimits));
1154 	pptlimits.bus = bus;
1155 	pptlimits.slot = slot;
1156 	pptlimits.func = func;
1157 
1158 	error = ioctl(ctx->fd, VM_GET_PPTDEV_LIMITS, &pptlimits);
1159 
1160 	*msi_limit = pptlimits.msi_limit;
1161 	*msix_limit = pptlimits.msix_limit;
1162 
1163 	return (error);
1164 }
1165 #else /* __FreeBSD__ */
1166 int
1167 vm_assign_pptdev(struct vmctx *ctx, int pptfd)
1168 {
1169 	struct vm_pptdev pptdev;
1170 
1171 	pptdev.pptfd = pptfd;
1172 	return (ioctl(ctx->fd, VM_BIND_PPTDEV, &pptdev));
1173 }
1174 
1175 int
1176 vm_unassign_pptdev(struct vmctx *ctx, int pptfd)
1177 {
1178 	struct vm_pptdev pptdev;
1179 
1180 	pptdev.pptfd = pptfd;
1181 	return (ioctl(ctx->fd, VM_UNBIND_PPTDEV, &pptdev));
1182 }
1183 
1184 int
1185 vm_map_pptdev_mmio(struct vmctx *ctx, int pptfd, vm_paddr_t gpa, size_t len,
1186     vm_paddr_t hpa)
1187 {
1188 	struct vm_pptdev_mmio pptmmio;
1189 
1190 	pptmmio.pptfd = pptfd;
1191 	pptmmio.gpa = gpa;
1192 	pptmmio.len = len;
1193 	pptmmio.hpa = hpa;
1194 	return (ioctl(ctx->fd, VM_MAP_PPTDEV_MMIO, &pptmmio));
1195 }
1196 
1197 int
1198 vm_setup_pptdev_msi(struct vmctx *ctx, int vcpu, int pptfd, uint64_t addr,
1199     uint64_t msg, int numvec)
1200 {
1201 	struct vm_pptdev_msi pptmsi;
1202 
1203 	pptmsi.vcpu = vcpu;
1204 	pptmsi.pptfd = pptfd;
1205 	pptmsi.msg = msg;
1206 	pptmsi.addr = addr;
1207 	pptmsi.numvec = numvec;
1208 	return (ioctl(ctx->fd, VM_PPTDEV_MSI, &pptmsi));
1209 }
1210 
1211 int
1212 vm_setup_pptdev_msix(struct vmctx *ctx, int vcpu, int pptfd, int idx,
1213     uint64_t addr, uint64_t msg, uint32_t vector_control)
1214 {
1215 	struct vm_pptdev_msix pptmsix;
1216 
1217 	pptmsix.vcpu = vcpu;
1218 	pptmsix.pptfd = pptfd;
1219 	pptmsix.idx = idx;
1220 	pptmsix.msg = msg;
1221 	pptmsix.addr = addr;
1222 	pptmsix.vector_control = vector_control;
1223 	return ioctl(ctx->fd, VM_PPTDEV_MSIX, &pptmsix);
1224 }
1225 
1226 int
1227 vm_get_pptdev_limits(struct vmctx *ctx, int pptfd, int *msi_limit,
1228     int *msix_limit)
1229 {
1230 	struct vm_pptdev_limits pptlimits;
1231 	int error;
1232 
1233 	bzero(&pptlimits, sizeof (pptlimits));
1234 	pptlimits.pptfd = pptfd;
1235 	error = ioctl(ctx->fd, VM_GET_PPTDEV_LIMITS, &pptlimits);
1236 
1237 	*msi_limit = pptlimits.msi_limit;
1238 	*msix_limit = pptlimits.msix_limit;
1239 	return (error);
1240 }
1241 #endif /* __FreeBSD__ */
1242 
1243 uint64_t *
1244 vm_get_stats(struct vmctx *ctx, int vcpu, struct timeval *ret_tv,
1245 	     int *ret_entries)
1246 {
1247 	int error;
1248 
1249 	static struct vm_stats vmstats;
1250 
1251 	vmstats.cpuid = vcpu;
1252 
1253 	error = ioctl(ctx->fd, VM_STATS_IOC, &vmstats);
1254 	if (error == 0) {
1255 		if (ret_entries)
1256 			*ret_entries = vmstats.num_entries;
1257 		if (ret_tv)
1258 			*ret_tv = vmstats.tv;
1259 		return (vmstats.statbuf);
1260 	} else
1261 		return (NULL);
1262 }
1263 
1264 const char *
1265 vm_get_stat_desc(struct vmctx *ctx, int index)
1266 {
1267 	static struct vm_stat_desc statdesc;
1268 
1269 	statdesc.index = index;
1270 	if (ioctl(ctx->fd, VM_STAT_DESC, &statdesc) == 0)
1271 		return (statdesc.desc);
1272 	else
1273 		return (NULL);
1274 }
1275 
1276 int
1277 vm_get_x2apic_state(struct vmctx *ctx, int vcpu, enum x2apic_state *state)
1278 {
1279 	int error;
1280 	struct vm_x2apic x2apic;
1281 
1282 	bzero(&x2apic, sizeof(x2apic));
1283 	x2apic.cpuid = vcpu;
1284 
1285 	error = ioctl(ctx->fd, VM_GET_X2APIC_STATE, &x2apic);
1286 	*state = x2apic.state;
1287 	return (error);
1288 }
1289 
1290 int
1291 vm_set_x2apic_state(struct vmctx *ctx, int vcpu, enum x2apic_state state)
1292 {
1293 	int error;
1294 	struct vm_x2apic x2apic;
1295 
1296 	bzero(&x2apic, sizeof(x2apic));
1297 	x2apic.cpuid = vcpu;
1298 	x2apic.state = state;
1299 
1300 	error = ioctl(ctx->fd, VM_SET_X2APIC_STATE, &x2apic);
1301 
1302 	return (error);
1303 }
1304 
1305 /*
1306  * From Intel Vol 3a:
1307  * Table 9-1. IA-32 Processor States Following Power-up, Reset or INIT
1308  */
1309 int
1310 vcpu_reset(struct vmctx *vmctx, int vcpu)
1311 {
1312 	int error;
1313 	uint64_t rflags, rip, cr0, cr4, zero, desc_base, rdx;
1314 	uint32_t desc_access, desc_limit;
1315 	uint16_t sel;
1316 
1317 	zero = 0;
1318 
1319 	rflags = 0x2;
1320 	error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RFLAGS, rflags);
1321 	if (error)
1322 		goto done;
1323 
1324 	rip = 0xfff0;
1325 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RIP, rip)) != 0)
1326 		goto done;
1327 
1328 	cr0 = CR0_NE;
1329 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR0, cr0)) != 0)
1330 		goto done;
1331 
1332 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR3, zero)) != 0)
1333 		goto done;
1334 
1335 	cr4 = 0;
1336 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CR4, cr4)) != 0)
1337 		goto done;
1338 
1339 	/*
1340 	 * CS: present, r/w, accessed, 16-bit, byte granularity, usable
1341 	 */
1342 	desc_base = 0xffff0000;
1343 	desc_limit = 0xffff;
1344 	desc_access = 0x0093;
1345 	error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_CS,
1346 			    desc_base, desc_limit, desc_access);
1347 	if (error)
1348 		goto done;
1349 
1350 	sel = 0xf000;
1351 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_CS, sel)) != 0)
1352 		goto done;
1353 
1354 	/*
1355 	 * SS,DS,ES,FS,GS: present, r/w, accessed, 16-bit, byte granularity
1356 	 */
1357 	desc_base = 0;
1358 	desc_limit = 0xffff;
1359 	desc_access = 0x0093;
1360 	error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_SS,
1361 			    desc_base, desc_limit, desc_access);
1362 	if (error)
1363 		goto done;
1364 
1365 	error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_DS,
1366 			    desc_base, desc_limit, desc_access);
1367 	if (error)
1368 		goto done;
1369 
1370 	error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_ES,
1371 			    desc_base, desc_limit, desc_access);
1372 	if (error)
1373 		goto done;
1374 
1375 	error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_FS,
1376 			    desc_base, desc_limit, desc_access);
1377 	if (error)
1378 		goto done;
1379 
1380 	error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_GS,
1381 			    desc_base, desc_limit, desc_access);
1382 	if (error)
1383 		goto done;
1384 
1385 	sel = 0;
1386 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_SS, sel)) != 0)
1387 		goto done;
1388 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_DS, sel)) != 0)
1389 		goto done;
1390 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_ES, sel)) != 0)
1391 		goto done;
1392 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_FS, sel)) != 0)
1393 		goto done;
1394 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_GS, sel)) != 0)
1395 		goto done;
1396 
1397 	/* General purpose registers */
1398 	rdx = 0xf00;
1399 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RAX, zero)) != 0)
1400 		goto done;
1401 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RBX, zero)) != 0)
1402 		goto done;
1403 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RCX, zero)) != 0)
1404 		goto done;
1405 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RDX, rdx)) != 0)
1406 		goto done;
1407 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RSI, zero)) != 0)
1408 		goto done;
1409 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RDI, zero)) != 0)
1410 		goto done;
1411 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RBP, zero)) != 0)
1412 		goto done;
1413 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_RSP, zero)) != 0)
1414 		goto done;
1415 
1416 	/* GDTR, IDTR */
1417 	desc_base = 0;
1418 	desc_limit = 0xffff;
1419 	desc_access = 0;
1420 	error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_GDTR,
1421 			    desc_base, desc_limit, desc_access);
1422 	if (error != 0)
1423 		goto done;
1424 
1425 	error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_IDTR,
1426 			    desc_base, desc_limit, desc_access);
1427 	if (error != 0)
1428 		goto done;
1429 
1430 	/* TR */
1431 	desc_base = 0;
1432 	desc_limit = 0xffff;
1433 	desc_access = 0x0000008b;
1434 	error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_TR, 0, 0, desc_access);
1435 	if (error)
1436 		goto done;
1437 
1438 	sel = 0;
1439 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_TR, sel)) != 0)
1440 		goto done;
1441 
1442 	/* LDTR */
1443 	desc_base = 0;
1444 	desc_limit = 0xffff;
1445 	desc_access = 0x00000082;
1446 	error = vm_set_desc(vmctx, vcpu, VM_REG_GUEST_LDTR, desc_base,
1447 			    desc_limit, desc_access);
1448 	if (error)
1449 		goto done;
1450 
1451 	sel = 0;
1452 	if ((error = vm_set_register(vmctx, vcpu, VM_REG_GUEST_LDTR, 0)) != 0)
1453 		goto done;
1454 
1455 	/* XXX cr2, debug registers */
1456 
1457 	error = 0;
1458 done:
1459 	return (error);
1460 }
1461 
1462 int
1463 vm_get_gpa_pmap(struct vmctx *ctx, uint64_t gpa, uint64_t *pte, int *num)
1464 {
1465 	int error, i;
1466 	struct vm_gpa_pte gpapte;
1467 
1468 	bzero(&gpapte, sizeof(gpapte));
1469 	gpapte.gpa = gpa;
1470 
1471 	error = ioctl(ctx->fd, VM_GET_GPA_PMAP, &gpapte);
1472 
1473 	if (error == 0) {
1474 		*num = gpapte.ptenum;
1475 		for (i = 0; i < gpapte.ptenum; i++)
1476 			pte[i] = gpapte.pte[i];
1477 	}
1478 
1479 	return (error);
1480 }
1481 
1482 int
1483 vm_get_hpet_capabilities(struct vmctx *ctx, uint32_t *capabilities)
1484 {
1485 	int error;
1486 	struct vm_hpet_cap cap;
1487 
1488 	bzero(&cap, sizeof(struct vm_hpet_cap));
1489 	error = ioctl(ctx->fd, VM_GET_HPET_CAPABILITIES, &cap);
1490 	if (capabilities != NULL)
1491 		*capabilities = cap.capabilities;
1492 	return (error);
1493 }
1494 
1495 int
1496 vm_gla2gpa(struct vmctx *ctx, int vcpu, struct vm_guest_paging *paging,
1497     uint64_t gla, int prot, uint64_t *gpa, int *fault)
1498 {
1499 	struct vm_gla2gpa gg;
1500 	int error;
1501 
1502 	bzero(&gg, sizeof(struct vm_gla2gpa));
1503 	gg.vcpuid = vcpu;
1504 	gg.prot = prot;
1505 	gg.gla = gla;
1506 	gg.paging = *paging;
1507 
1508 	error = ioctl(ctx->fd, VM_GLA2GPA, &gg);
1509 	if (error == 0) {
1510 		*fault = gg.fault;
1511 		*gpa = gg.gpa;
1512 	}
1513 	return (error);
1514 }
1515 
1516 int
1517 vm_gla2gpa_nofault(struct vmctx *ctx, int vcpu, struct vm_guest_paging *paging,
1518     uint64_t gla, int prot, uint64_t *gpa, int *fault)
1519 {
1520 	struct vm_gla2gpa gg;
1521 	int error;
1522 
1523 	bzero(&gg, sizeof(struct vm_gla2gpa));
1524 	gg.vcpuid = vcpu;
1525 	gg.prot = prot;
1526 	gg.gla = gla;
1527 	gg.paging = *paging;
1528 
1529 	error = ioctl(ctx->fd, VM_GLA2GPA_NOFAULT, &gg);
1530 	if (error == 0) {
1531 		*fault = gg.fault;
1532 		*gpa = gg.gpa;
1533 	}
1534 	return (error);
1535 }
1536 
1537 #ifndef min
1538 #define	min(a,b)	(((a) < (b)) ? (a) : (b))
1539 #endif
1540 
1541 int
1542 vm_copy_setup(struct vmctx *ctx, int vcpu, struct vm_guest_paging *paging,
1543     uint64_t gla, size_t len, int prot, struct iovec *iov, int iovcnt,
1544     int *fault)
1545 {
1546 	void *va;
1547 	uint64_t gpa;
1548 	int error, i, n, off;
1549 
1550 	for (i = 0; i < iovcnt; i++) {
1551 		iov[i].iov_base = 0;
1552 		iov[i].iov_len = 0;
1553 	}
1554 
1555 	while (len) {
1556 		assert(iovcnt > 0);
1557 		error = vm_gla2gpa(ctx, vcpu, paging, gla, prot, &gpa, fault);
1558 		if (error || *fault)
1559 			return (error);
1560 
1561 		off = gpa & PAGE_MASK;
1562 		n = min(len, PAGE_SIZE - off);
1563 
1564 		va = vm_map_gpa(ctx, gpa, n);
1565 		if (va == NULL)
1566 			return (EFAULT);
1567 
1568 		iov->iov_base = va;
1569 		iov->iov_len = n;
1570 		iov++;
1571 		iovcnt--;
1572 
1573 		gla += n;
1574 		len -= n;
1575 	}
1576 	return (0);
1577 }
1578 
1579 void
1580 vm_copy_teardown(struct vmctx *ctx, int vcpu, struct iovec *iov, int iovcnt)
1581 {
1582 
1583 	return;
1584 }
1585 
1586 void
1587 vm_copyin(struct vmctx *ctx, int vcpu, struct iovec *iov, void *vp, size_t len)
1588 {
1589 	const char *src;
1590 	char *dst;
1591 	size_t n;
1592 
1593 	dst = vp;
1594 	while (len) {
1595 		assert(iov->iov_len);
1596 		n = min(len, iov->iov_len);
1597 		src = iov->iov_base;
1598 		bcopy(src, dst, n);
1599 
1600 		iov++;
1601 		dst += n;
1602 		len -= n;
1603 	}
1604 }
1605 
1606 void
1607 vm_copyout(struct vmctx *ctx, int vcpu, const void *vp, struct iovec *iov,
1608     size_t len)
1609 {
1610 	const char *src;
1611 	char *dst;
1612 	size_t n;
1613 
1614 	src = vp;
1615 	while (len) {
1616 		assert(iov->iov_len);
1617 		n = min(len, iov->iov_len);
1618 		dst = iov->iov_base;
1619 		bcopy(src, dst, n);
1620 
1621 		iov++;
1622 		src += n;
1623 		len -= n;
1624 	}
1625 }
1626 
1627 static int
1628 vm_get_cpus(struct vmctx *ctx, int which, cpuset_t *cpus)
1629 {
1630 	struct vm_cpuset vm_cpuset;
1631 	int error;
1632 
1633 	bzero(&vm_cpuset, sizeof(struct vm_cpuset));
1634 	vm_cpuset.which = which;
1635 	vm_cpuset.cpusetsize = sizeof(cpuset_t);
1636 	vm_cpuset.cpus = cpus;
1637 
1638 	error = ioctl(ctx->fd, VM_GET_CPUS, &vm_cpuset);
1639 	return (error);
1640 }
1641 
1642 int
1643 vm_active_cpus(struct vmctx *ctx, cpuset_t *cpus)
1644 {
1645 
1646 	return (vm_get_cpus(ctx, VM_ACTIVE_CPUS, cpus));
1647 }
1648 
1649 int
1650 vm_suspended_cpus(struct vmctx *ctx, cpuset_t *cpus)
1651 {
1652 
1653 	return (vm_get_cpus(ctx, VM_SUSPENDED_CPUS, cpus));
1654 }
1655 
1656 int
1657 vm_debug_cpus(struct vmctx *ctx, cpuset_t *cpus)
1658 {
1659 
1660 	return (vm_get_cpus(ctx, VM_DEBUG_CPUS, cpus));
1661 }
1662 
1663 int
1664 vm_activate_cpu(struct vmctx *ctx, int vcpu)
1665 {
1666 	struct vm_activate_cpu ac;
1667 	int error;
1668 
1669 	bzero(&ac, sizeof(struct vm_activate_cpu));
1670 	ac.vcpuid = vcpu;
1671 	error = ioctl(ctx->fd, VM_ACTIVATE_CPU, &ac);
1672 	return (error);
1673 }
1674 
1675 int
1676 vm_suspend_cpu(struct vmctx *ctx, int vcpu)
1677 {
1678 	struct vm_activate_cpu ac;
1679 	int error;
1680 
1681 	bzero(&ac, sizeof(struct vm_activate_cpu));
1682 	ac.vcpuid = vcpu;
1683 	error = ioctl(ctx->fd, VM_SUSPEND_CPU, &ac);
1684 	return (error);
1685 }
1686 
1687 int
1688 vm_resume_cpu(struct vmctx *ctx, int vcpu)
1689 {
1690 	struct vm_activate_cpu ac;
1691 	int error;
1692 
1693 	bzero(&ac, sizeof(struct vm_activate_cpu));
1694 	ac.vcpuid = vcpu;
1695 	error = ioctl(ctx->fd, VM_RESUME_CPU, &ac);
1696 	return (error);
1697 }
1698 
1699 int
1700 vm_get_intinfo(struct vmctx *ctx, int vcpu, uint64_t *info1, uint64_t *info2)
1701 {
1702 	struct vm_intinfo vmii;
1703 	int error;
1704 
1705 	bzero(&vmii, sizeof(struct vm_intinfo));
1706 	vmii.vcpuid = vcpu;
1707 	error = ioctl(ctx->fd, VM_GET_INTINFO, &vmii);
1708 	if (error == 0) {
1709 		*info1 = vmii.info1;
1710 		*info2 = vmii.info2;
1711 	}
1712 	return (error);
1713 }
1714 
1715 int
1716 vm_set_intinfo(struct vmctx *ctx, int vcpu, uint64_t info1)
1717 {
1718 	struct vm_intinfo vmii;
1719 	int error;
1720 
1721 	bzero(&vmii, sizeof(struct vm_intinfo));
1722 	vmii.vcpuid = vcpu;
1723 	vmii.info1 = info1;
1724 	error = ioctl(ctx->fd, VM_SET_INTINFO, &vmii);
1725 	return (error);
1726 }
1727 
1728 int
1729 vm_rtc_write(struct vmctx *ctx, int offset, uint8_t value)
1730 {
1731 	struct vm_rtc_data rtcdata;
1732 	int error;
1733 
1734 	bzero(&rtcdata, sizeof(struct vm_rtc_data));
1735 	rtcdata.offset = offset;
1736 	rtcdata.value = value;
1737 	error = ioctl(ctx->fd, VM_RTC_WRITE, &rtcdata);
1738 	return (error);
1739 }
1740 
1741 int
1742 vm_rtc_read(struct vmctx *ctx, int offset, uint8_t *retval)
1743 {
1744 	struct vm_rtc_data rtcdata;
1745 	int error;
1746 
1747 	bzero(&rtcdata, sizeof(struct vm_rtc_data));
1748 	rtcdata.offset = offset;
1749 	error = ioctl(ctx->fd, VM_RTC_READ, &rtcdata);
1750 	if (error == 0)
1751 		*retval = rtcdata.value;
1752 	return (error);
1753 }
1754 
1755 int
1756 vm_rtc_settime(struct vmctx *ctx, time_t secs)
1757 {
1758 	struct vm_rtc_time rtctime;
1759 	int error;
1760 
1761 	bzero(&rtctime, sizeof(struct vm_rtc_time));
1762 	rtctime.secs = secs;
1763 	error = ioctl(ctx->fd, VM_RTC_SETTIME, &rtctime);
1764 	return (error);
1765 }
1766 
1767 int
1768 vm_rtc_gettime(struct vmctx *ctx, time_t *secs)
1769 {
1770 	struct vm_rtc_time rtctime;
1771 	int error;
1772 
1773 	bzero(&rtctime, sizeof(struct vm_rtc_time));
1774 	error = ioctl(ctx->fd, VM_RTC_GETTIME, &rtctime);
1775 	if (error == 0)
1776 		*secs = rtctime.secs;
1777 	return (error);
1778 }
1779 
1780 int
1781 vm_restart_instruction(void *arg, int vcpu)
1782 {
1783 	struct vmctx *ctx = arg;
1784 
1785 	return (ioctl(ctx->fd, VM_RESTART_INSTRUCTION, &vcpu));
1786 }
1787 
1788 int
1789 vm_set_topology(struct vmctx *ctx,
1790     uint16_t sockets, uint16_t cores, uint16_t threads, uint16_t maxcpus)
1791 {
1792 	struct vm_cpu_topology topology;
1793 
1794 	bzero(&topology, sizeof (struct vm_cpu_topology));
1795 	topology.sockets = sockets;
1796 	topology.cores = cores;
1797 	topology.threads = threads;
1798 	topology.maxcpus = maxcpus;
1799 	return (ioctl(ctx->fd, VM_SET_TOPOLOGY, &topology));
1800 }
1801 
1802 int
1803 vm_get_topology(struct vmctx *ctx,
1804     uint16_t *sockets, uint16_t *cores, uint16_t *threads, uint16_t *maxcpus)
1805 {
1806 	struct vm_cpu_topology topology;
1807 	int error;
1808 
1809 	bzero(&topology, sizeof (struct vm_cpu_topology));
1810 	error = ioctl(ctx->fd, VM_GET_TOPOLOGY, &topology);
1811 	if (error == 0) {
1812 		*sockets = topology.sockets;
1813 		*cores = topology.cores;
1814 		*threads = topology.threads;
1815 		*maxcpus = topology.maxcpus;
1816 	}
1817 	return (error);
1818 }
1819 
1820 int
1821 vm_get_device_fd(struct vmctx *ctx)
1822 {
1823 
1824 	return (ctx->fd);
1825 }
1826 
1827 #ifndef __FreeBSD__
1828 int
1829 vm_pmtmr_set_location(struct vmctx *ctx, uint16_t ioport)
1830 {
1831 	return (ioctl(ctx->fd, VM_PMTMR_LOCATE, ioport));
1832 }
1833 
1834 int
1835 vm_wrlock_cycle(struct vmctx *ctx)
1836 {
1837 	if (ioctl(ctx->fd, VM_WRLOCK_CYCLE, 0) != 0) {
1838 		return (errno);
1839 	}
1840 	return (0);
1841 }
1842 #endif /* __FreeBSD__ */
1843 
1844 #ifdef __FreeBSD__
1845 const cap_ioctl_t *
1846 vm_get_ioctls(size_t *len)
1847 {
1848 	cap_ioctl_t *cmds;
1849 	/* keep in sync with machine/vmm_dev.h */
1850 	static const cap_ioctl_t vm_ioctl_cmds[] = { VM_RUN, VM_SUSPEND, VM_REINIT,
1851 	    VM_ALLOC_MEMSEG, VM_GET_MEMSEG, VM_MMAP_MEMSEG, VM_MMAP_MEMSEG,
1852 	    VM_MMAP_GETNEXT, VM_SET_REGISTER, VM_GET_REGISTER,
1853 	    VM_SET_SEGMENT_DESCRIPTOR, VM_GET_SEGMENT_DESCRIPTOR,
1854 	    VM_SET_REGISTER_SET, VM_GET_REGISTER_SET,
1855 	    VM_SET_KERNEMU_DEV, VM_GET_KERNEMU_DEV,
1856 	    VM_INJECT_EXCEPTION, VM_LAPIC_IRQ, VM_LAPIC_LOCAL_IRQ,
1857 	    VM_LAPIC_MSI, VM_IOAPIC_ASSERT_IRQ, VM_IOAPIC_DEASSERT_IRQ,
1858 	    VM_IOAPIC_PULSE_IRQ, VM_IOAPIC_PINCOUNT, VM_ISA_ASSERT_IRQ,
1859 	    VM_ISA_DEASSERT_IRQ, VM_ISA_PULSE_IRQ, VM_ISA_SET_IRQ_TRIGGER,
1860 	    VM_SET_CAPABILITY, VM_GET_CAPABILITY, VM_BIND_PPTDEV,
1861 	    VM_UNBIND_PPTDEV, VM_MAP_PPTDEV_MMIO, VM_PPTDEV_MSI,
1862 	    VM_PPTDEV_MSIX, VM_INJECT_NMI, VM_STATS, VM_STAT_DESC,
1863 	    VM_SET_X2APIC_STATE, VM_GET_X2APIC_STATE,
1864 	    VM_GET_HPET_CAPABILITIES, VM_GET_GPA_PMAP, VM_GLA2GPA,
1865 	    VM_GLA2GPA_NOFAULT,
1866 	    VM_ACTIVATE_CPU, VM_GET_CPUS, VM_SUSPEND_CPU, VM_RESUME_CPU,
1867 	    VM_SET_INTINFO, VM_GET_INTINFO,
1868 	    VM_RTC_WRITE, VM_RTC_READ, VM_RTC_SETTIME, VM_RTC_GETTIME,
1869 	    VM_RESTART_INSTRUCTION, VM_SET_TOPOLOGY, VM_GET_TOPOLOGY };
1870 
1871 	if (len == NULL) {
1872 		cmds = malloc(sizeof(vm_ioctl_cmds));
1873 		if (cmds == NULL)
1874 			return (NULL);
1875 		bcopy(vm_ioctl_cmds, cmds, sizeof(vm_ioctl_cmds));
1876 		return (cmds);
1877 	}
1878 
1879 	*len = nitems(vm_ioctl_cmds);
1880 	return (NULL);
1881 }
1882 #endif /* __FreeBSD__ */
1883