xref: /freebsd/usr.sbin/bhyve/bhyverun.c (revision e1c4c8dd8d2d10b6104f06856a77bd5b4813a801)
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/types.h>
30 #ifndef WITHOUT_CAPSICUM
31 #include <sys/capsicum.h>
32 #endif
33 #include <sys/mman.h>
34 #ifdef BHYVE_SNAPSHOT
35 #include <sys/socket.h>
36 #include <sys/stat.h>
37 #endif
38 #include <sys/time.h>
39 #ifdef BHYVE_SNAPSHOT
40 #include <sys/un.h>
41 #endif
42 
43 #include <machine/atomic.h>
44 
45 #ifndef WITHOUT_CAPSICUM
46 #include <capsicum_helpers.h>
47 #endif
48 #include <stdio.h>
49 #include <stdlib.h>
50 #include <string.h>
51 #include <err.h>
52 #include <errno.h>
53 #ifdef BHYVE_SNAPSHOT
54 #include <fcntl.h>
55 #endif
56 #include <libgen.h>
57 #include <unistd.h>
58 #include <assert.h>
59 #include <pthread.h>
60 #include <pthread_np.h>
61 #include <sysexits.h>
62 #include <stdbool.h>
63 #include <stdint.h>
64 #ifdef BHYVE_SNAPSHOT
65 #include <ucl.h>
66 #include <unistd.h>
67 
68 #include <libxo/xo.h>
69 #endif
70 
71 #include <vmmapi.h>
72 
73 #include "acpi.h"
74 #include "bhyverun.h"
75 #include "bootrom.h"
76 #include "config.h"
77 #include "debug.h"
78 #ifdef BHYVE_GDB
79 #include "gdb.h"
80 #endif
81 #include "mem.h"
82 #include "mevent.h"
83 #include "pci_emul.h"
84 #ifdef __amd64__
85 #include "amd64/pci_lpc.h"
86 #endif
87 #include "qemu_fwcfg.h"
88 #ifdef BHYVE_SNAPSHOT
89 #include "snapshot.h"
90 #endif
91 #include "tpm_device.h"
92 #include "vmgenc.h"
93 #include "vmexit.h"
94 
95 #define MB		(1024UL * 1024)
96 #define GB		(1024UL * MB)
97 
98 int guest_ncpus;
99 uint16_t cpu_cores, cpu_sockets, cpu_threads;
100 
101 int raw_stdio = 0;
102 
103 static const int BSP = 0;
104 
105 static cpuset_t cpumask;
106 
107 static void vm_loop(struct vmctx *ctx, struct vcpu *vcpu);
108 
109 static struct vcpu_info {
110 	struct vmctx	*ctx;
111 	struct vcpu	*vcpu;
112 	int		vcpuid;
113 } *vcpu_info;
114 
115 static cpuset_t **vcpumap;
116 
117 /*
118  * XXX This parser is known to have the following issues:
119  * 1.  It accepts null key=value tokens ",," as setting "cpus" to an
120  *     empty string.
121  *
122  * The acceptance of a null specification ('-c ""') is by design to match the
123  * manual page syntax specification, this results in a topology of 1 vCPU.
124  */
125 int
126 bhyve_topology_parse(const char *opt)
127 {
128 	char *cp, *str, *tofree;
129 
130 	if (*opt == '\0') {
131 		set_config_value("sockets", "1");
132 		set_config_value("cores", "1");
133 		set_config_value("threads", "1");
134 		set_config_value("cpus", "1");
135 		return (0);
136 	}
137 
138 	tofree = str = strdup(opt);
139 	if (str == NULL)
140 		errx(4, "Failed to allocate memory");
141 
142 	while ((cp = strsep(&str, ",")) != NULL) {
143 		if (strncmp(cp, "cpus=", strlen("cpus=")) == 0)
144 			set_config_value("cpus", cp + strlen("cpus="));
145 		else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0)
146 			set_config_value("sockets", cp + strlen("sockets="));
147 		else if (strncmp(cp, "cores=", strlen("cores=")) == 0)
148 			set_config_value("cores", cp + strlen("cores="));
149 		else if (strncmp(cp, "threads=", strlen("threads=")) == 0)
150 			set_config_value("threads", cp + strlen("threads="));
151 		else if (strchr(cp, '=') != NULL)
152 			goto out;
153 		else
154 			set_config_value("cpus", cp);
155 	}
156 	free(tofree);
157 	return (0);
158 
159 out:
160 	free(tofree);
161 	return (-1);
162 }
163 
164 static int
165 parse_int_value(const char *key, const char *value, int minval, int maxval)
166 {
167 	char *cp;
168 	long lval;
169 
170 	errno = 0;
171 	lval = strtol(value, &cp, 0);
172 	if (errno != 0 || *cp != '\0' || cp == value || lval < minval ||
173 	    lval > maxval)
174 		errx(4, "Invalid value for %s: '%s'", key, value);
175 	return (lval);
176 }
177 
178 /*
179  * Set the sockets, cores, threads, and guest_cpus variables based on
180  * the configured topology.
181  *
182  * The limits of UINT16_MAX are due to the types passed to
183  * vm_set_topology().  vmm.ko may enforce tighter limits.
184  */
185 static void
186 calc_topology(void)
187 {
188 	const char *value;
189 	bool explicit_cpus;
190 	uint64_t ncpus;
191 
192 	value = get_config_value("cpus");
193 	if (value != NULL) {
194 		guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX);
195 		explicit_cpus = true;
196 	} else {
197 		guest_ncpus = 1;
198 		explicit_cpus = false;
199 	}
200 	value = get_config_value("cores");
201 	if (value != NULL)
202 		cpu_cores = parse_int_value("cores", value, 1, UINT16_MAX);
203 	else
204 		cpu_cores = 1;
205 	value = get_config_value("threads");
206 	if (value != NULL)
207 		cpu_threads = parse_int_value("threads", value, 1, UINT16_MAX);
208 	else
209 		cpu_threads = 1;
210 	value = get_config_value("sockets");
211 	if (value != NULL)
212 		cpu_sockets = parse_int_value("sockets", value, 1, UINT16_MAX);
213 	else
214 		cpu_sockets = guest_ncpus;
215 
216 	/*
217 	 * Compute sockets * cores * threads avoiding overflow.  The
218 	 * range check above insures these are 16 bit values.
219 	 */
220 	ncpus = (uint64_t)cpu_sockets * cpu_cores * cpu_threads;
221 	if (ncpus > UINT16_MAX)
222 		errx(4, "Computed number of vCPUs too high: %ju",
223 		    (uintmax_t)ncpus);
224 
225 	if (explicit_cpus) {
226 		if (guest_ncpus != (int)ncpus)
227 			errx(4, "Topology (%d sockets, %d cores, %d threads) "
228 			    "does not match %d vCPUs",
229 			    cpu_sockets, cpu_cores, cpu_threads,
230 			    guest_ncpus);
231 	} else
232 		guest_ncpus = ncpus;
233 }
234 
235 int
236 bhyve_pincpu_parse(const char *opt)
237 {
238 	const char *value;
239 	char *newval;
240 	char key[16];
241 	int vcpu, pcpu;
242 
243 	if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) {
244 		fprintf(stderr, "invalid format: %s\n", opt);
245 		return (-1);
246 	}
247 
248 	if (vcpu < 0) {
249 		fprintf(stderr, "invalid vcpu '%d'\n", vcpu);
250 		return (-1);
251 	}
252 
253 	if (pcpu < 0 || pcpu >= CPU_SETSIZE) {
254 		fprintf(stderr, "hostcpu '%d' outside valid range from "
255 		    "0 to %d\n", pcpu, CPU_SETSIZE - 1);
256 		return (-1);
257 	}
258 
259 	snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
260 	value = get_config_value(key);
261 
262 	if (asprintf(&newval, "%s%s%d", value != NULL ? value : "",
263 	    value != NULL ? "," : "", pcpu) == -1) {
264 		perror("failed to build new cpuset string");
265 		return (-1);
266 	}
267 
268 	set_config_value(key, newval);
269 	free(newval);
270 	return (0);
271 }
272 
273 static void
274 parse_cpuset(int vcpu, const char *list, cpuset_t *set)
275 {
276 	char *cp, *token;
277 	int pcpu, start;
278 
279 	CPU_ZERO(set);
280 	start = -1;
281 	token = __DECONST(char *, list);
282 	for (;;) {
283 		pcpu = strtoul(token, &cp, 0);
284 		if (cp == token)
285 			errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
286 		if (pcpu < 0 || pcpu >= CPU_SETSIZE)
287 			errx(4, "hostcpu '%d' outside valid range from 0 to %d",
288 			    pcpu, CPU_SETSIZE - 1);
289 		switch (*cp) {
290 		case ',':
291 		case '\0':
292 			if (start >= 0) {
293 				if (start > pcpu)
294 					errx(4, "Invalid hostcpu range %d-%d",
295 					    start, pcpu);
296 				while (start < pcpu) {
297 					CPU_SET(start, set);
298 					start++;
299 				}
300 				start = -1;
301 			}
302 			CPU_SET(pcpu, set);
303 			break;
304 		case '-':
305 			if (start >= 0)
306 				errx(4, "invalid cpuset for vcpu %d: '%s'",
307 				    vcpu, list);
308 			start = pcpu;
309 			break;
310 		default:
311 			errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
312 		}
313 		if (*cp == '\0')
314 			break;
315 		token = cp + 1;
316 	}
317 }
318 
319 static void
320 build_vcpumaps(void)
321 {
322 	char key[16];
323 	const char *value;
324 	int vcpu;
325 
326 	vcpumap = calloc(guest_ncpus, sizeof(*vcpumap));
327 	for (vcpu = 0; vcpu < guest_ncpus; vcpu++) {
328 		snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
329 		value = get_config_value(key);
330 		if (value == NULL)
331 			continue;
332 		vcpumap[vcpu] = malloc(sizeof(cpuset_t));
333 		if (vcpumap[vcpu] == NULL)
334 			err(4, "Failed to allocate cpuset for vcpu %d", vcpu);
335 		parse_cpuset(vcpu, value, vcpumap[vcpu]);
336 	}
337 }
338 
339 void *
340 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len)
341 {
342 
343 	return (vm_map_gpa(ctx, gaddr, len));
344 }
345 
346 #ifdef BHYVE_SNAPSHOT
347 uintptr_t
348 paddr_host2guest(struct vmctx *ctx, void *addr)
349 {
350 	return (vm_rev_map_gpa(ctx, addr));
351 }
352 #endif
353 
354 int
355 fbsdrun_virtio_msix(void)
356 {
357 
358 	return (get_config_bool_default("virtio_msix", true));
359 }
360 
361 struct vcpu *
362 fbsdrun_vcpu(int vcpuid)
363 {
364 	return (vcpu_info[vcpuid].vcpu);
365 }
366 
367 static void *
368 fbsdrun_start_thread(void *param)
369 {
370 	char tname[MAXCOMLEN + 1];
371 	struct vcpu_info *vi = param;
372 	int error;
373 
374 	snprintf(tname, sizeof(tname), "vcpu %d", vi->vcpuid);
375 	pthread_set_name_np(pthread_self(), tname);
376 
377 	if (vcpumap[vi->vcpuid] != NULL) {
378 		error = pthread_setaffinity_np(pthread_self(),
379 		    sizeof(cpuset_t), vcpumap[vi->vcpuid]);
380 		assert(error == 0);
381 	}
382 
383 #ifdef BHYVE_SNAPSHOT
384 	checkpoint_cpu_add(vi->vcpuid);
385 #endif
386 #ifdef BHYVE_GDB
387 	gdb_cpu_add(vi->vcpu);
388 #endif
389 
390 	vm_loop(vi->ctx, vi->vcpu);
391 
392 	/* not reached */
393 	exit(1);
394 	return (NULL);
395 }
396 
397 void
398 fbsdrun_addcpu(int vcpuid)
399 {
400 	struct vcpu_info *vi;
401 	pthread_t thr;
402 	int error;
403 
404 	vi = &vcpu_info[vcpuid];
405 
406 	error = vm_activate_cpu(vi->vcpu);
407 	if (error != 0)
408 		err(EX_OSERR, "could not activate CPU %d", vi->vcpuid);
409 
410 	CPU_SET_ATOMIC(vcpuid, &cpumask);
411 
412 	vm_suspend_cpu(vi->vcpu);
413 
414 	error = pthread_create(&thr, NULL, fbsdrun_start_thread, vi);
415 	assert(error == 0);
416 }
417 
418 void
419 fbsdrun_deletecpu(int vcpu)
420 {
421 	static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER;
422 	static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER;
423 
424 	pthread_mutex_lock(&resetcpu_mtx);
425 	if (!CPU_ISSET(vcpu, &cpumask)) {
426 		EPRINTLN("Attempting to delete unknown cpu %d", vcpu);
427 		exit(4);
428 	}
429 
430 	CPU_CLR(vcpu, &cpumask);
431 
432 	if (vcpu != BSP) {
433 		pthread_cond_signal(&resetcpu_cond);
434 		pthread_mutex_unlock(&resetcpu_mtx);
435 		pthread_exit(NULL);
436 		/* NOTREACHED */
437 	}
438 
439 	while (!CPU_EMPTY(&cpumask)) {
440 		pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx);
441 	}
442 	pthread_mutex_unlock(&resetcpu_mtx);
443 }
444 
445 int
446 fbsdrun_suspendcpu(int vcpuid)
447 {
448 	return (vm_suspend_cpu(vcpu_info[vcpuid].vcpu));
449 }
450 
451 static void
452 vm_loop(struct vmctx *ctx, struct vcpu *vcpu)
453 {
454 	struct vm_exit vme;
455 	struct vm_run vmrun;
456 	int error, rc;
457 	enum vm_exitcode exitcode;
458 	cpuset_t active_cpus, dmask;
459 
460 	error = vm_active_cpus(ctx, &active_cpus);
461 	assert(CPU_ISSET(vcpu_id(vcpu), &active_cpus));
462 
463 	vmrun.vm_exit = &vme;
464 	vmrun.cpuset = &dmask;
465 	vmrun.cpusetsize = sizeof(dmask);
466 
467 	while (1) {
468 		error = vm_run(vcpu, &vmrun);
469 		if (error != 0)
470 			break;
471 
472 		exitcode = vme.exitcode;
473 		if (exitcode >= VM_EXITCODE_MAX ||
474 		    vmexit_handlers[exitcode] == NULL) {
475 			warnx("vm_loop: unexpected exitcode 0x%x", exitcode);
476 			exit(4);
477 		}
478 
479 		rc = (*vmexit_handlers[exitcode])(ctx, vcpu, &vmrun);
480 
481 		switch (rc) {
482 		case VMEXIT_CONTINUE:
483 			break;
484 		case VMEXIT_ABORT:
485 			abort();
486 		default:
487 			exit(4);
488 		}
489 	}
490 	EPRINTLN("vm_run error %d, errno %d", error, errno);
491 }
492 
493 static int
494 num_vcpus_allowed(struct vmctx *ctx, struct vcpu *vcpu)
495 {
496 	uint16_t sockets, cores, threads, maxcpus;
497 	int tmp, error;
498 
499 	/*
500 	 * The guest is allowed to spinup more than one processor only if the
501 	 * UNRESTRICTED_GUEST capability is available.
502 	 */
503 	error = vm_get_capability(vcpu, VM_CAP_UNRESTRICTED_GUEST, &tmp);
504 	if (error != 0)
505 		return (1);
506 
507 	error = vm_get_topology(ctx, &sockets, &cores, &threads, &maxcpus);
508 	if (error == 0)
509 		return (maxcpus);
510 	else
511 		return (1);
512 }
513 
514 static struct vmctx *
515 do_open(const char *vmname)
516 {
517 	struct vmctx *ctx;
518 	int error;
519 	bool reinit, romboot;
520 
521 	reinit = false;
522 
523 #ifdef __amd64__
524 	romboot = lpc_bootrom() != NULL;
525 #else
526 	romboot = true;
527 #endif
528 
529 	error = vm_create(vmname);
530 	if (error) {
531 		if (errno == EEXIST) {
532 			if (romboot) {
533 				reinit = true;
534 			} else {
535 				/*
536 				 * The virtual machine has been setup by the
537 				 * userspace bootloader.
538 				 */
539 			}
540 		} else {
541 			perror("vm_create");
542 			exit(4);
543 		}
544 	} else {
545 		if (!romboot) {
546 			/*
547 			 * If the virtual machine was just created then a
548 			 * bootrom must be configured to boot it.
549 			 */
550 			fprintf(stderr, "virtual machine cannot be booted\n");
551 			exit(4);
552 		}
553 	}
554 
555 	ctx = vm_open(vmname);
556 	if (ctx == NULL) {
557 		perror("vm_open");
558 		exit(4);
559 	}
560 
561 #ifndef WITHOUT_CAPSICUM
562 	if (vm_limit_rights(ctx) != 0)
563 		err(EX_OSERR, "vm_limit_rights");
564 #endif
565 
566 	if (reinit) {
567 		error = vm_reinit(ctx);
568 		if (error) {
569 			perror("vm_reinit");
570 			exit(4);
571 		}
572 	}
573 	error = vm_set_topology(ctx, cpu_sockets, cpu_cores, cpu_threads, 0);
574 	if (error)
575 		errx(EX_OSERR, "vm_set_topology");
576 	return (ctx);
577 }
578 
579 bool
580 bhyve_parse_config_option(const char *option)
581 {
582 	const char *value;
583 	char *path;
584 
585 	value = strchr(option, '=');
586 	if (value == NULL || value[1] == '\0')
587 		return (false);
588 	path = strndup(option, value - option);
589 	if (path == NULL)
590 		err(4, "Failed to allocate memory");
591 	set_config_value(path, value + 1);
592 	return (true);
593 }
594 
595 void
596 bhyve_parse_simple_config_file(const char *path)
597 {
598 	FILE *fp;
599 	char *line, *cp;
600 	size_t linecap;
601 	unsigned int lineno;
602 
603 	fp = fopen(path, "r");
604 	if (fp == NULL)
605 		err(4, "Failed to open configuration file %s", path);
606 	line = NULL;
607 	linecap = 0;
608 	lineno = 1;
609 	for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) {
610 		if (*line == '#' || *line == '\n')
611 			continue;
612 		cp = strchr(line, '\n');
613 		if (cp != NULL)
614 			*cp = '\0';
615 		if (!bhyve_parse_config_option(line))
616 			errx(4, "%s line %u: invalid config option '%s'", path,
617 			    lineno, line);
618 	}
619 	free(line);
620 	fclose(fp);
621 }
622 
623 #ifdef BHYVE_GDB
624 void
625 bhyve_parse_gdb_options(const char *opt)
626 {
627 	const char *sport;
628 	char *colon;
629 
630 	if (opt[0] == 'w') {
631 		set_config_bool("gdb.wait", true);
632 		opt++;
633 	}
634 
635 	colon = strrchr(opt, ':');
636 	if (colon == NULL) {
637 		sport = opt;
638 	} else {
639 		*colon = '\0';
640 		colon++;
641 		sport = colon;
642 		set_config_value("gdb.address", opt);
643 	}
644 
645 	set_config_value("gdb.port", sport);
646 }
647 #endif
648 
649 int
650 main(int argc, char *argv[])
651 {
652 	int error;
653 	int max_vcpus, memflags;
654 	struct vcpu *bsp;
655 	struct vmctx *ctx;
656 	size_t memsize;
657 	const char *value, *vmname;
658 #ifdef BHYVE_SNAPSHOT
659 	char *restore_file;
660 	struct restore_state rstate;
661 
662 	restore_file = NULL;
663 #endif
664 
665 	bhyve_init_config();
666 	bhyve_optparse(argc, argv);
667 	argc -= optind;
668 	argv += optind;
669 
670 	if (argc > 1)
671 		bhyve_usage(1);
672 
673 #ifdef BHYVE_SNAPSHOT
674 	if (restore_file != NULL) {
675 		error = load_restore_file(restore_file, &rstate);
676 		if (error) {
677 			fprintf(stderr, "Failed to read checkpoint info from "
678 					"file: '%s'.\n", restore_file);
679 			exit(1);
680 		}
681 		vmname = lookup_vmname(&rstate);
682 		if (vmname != NULL)
683 			set_config_value("name", vmname);
684 	}
685 #endif
686 
687 	if (argc == 1)
688 		set_config_value("name", argv[0]);
689 
690 	vmname = get_config_value("name");
691 	if (vmname == NULL)
692 		bhyve_usage(1);
693 
694 	if (get_config_bool_default("config.dump", false)) {
695 		dump_config();
696 		exit(1);
697 	}
698 
699 	calc_topology();
700 	build_vcpumaps();
701 
702 	value = get_config_value("memory.size");
703 	error = vm_parse_memsize(value, &memsize);
704 	if (error)
705 		errx(EX_USAGE, "invalid memsize '%s'", value);
706 
707 	ctx = do_open(vmname);
708 
709 #ifdef BHYVE_SNAPSHOT
710 	if (restore_file != NULL) {
711 		guest_ncpus = lookup_guest_ncpus(&rstate);
712 		memflags = lookup_memflags(&rstate);
713 		memsize = lookup_memsize(&rstate);
714 	}
715 
716 	if (guest_ncpus < 1) {
717 		fprintf(stderr, "Invalid guest vCPUs (%d)\n", guest_ncpus);
718 		exit(1);
719 	}
720 #endif
721 
722 	bsp = vm_vcpu_open(ctx, BSP);
723 	max_vcpus = num_vcpus_allowed(ctx, bsp);
724 	if (guest_ncpus > max_vcpus) {
725 		fprintf(stderr, "%d vCPUs requested but only %d available\n",
726 			guest_ncpus, max_vcpus);
727 		exit(4);
728 	}
729 
730 	bhyve_init_vcpu(bsp);
731 
732 	/* Allocate per-VCPU resources. */
733 	vcpu_info = calloc(guest_ncpus, sizeof(*vcpu_info));
734 	for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++) {
735 		vcpu_info[vcpuid].ctx = ctx;
736 		vcpu_info[vcpuid].vcpuid = vcpuid;
737 		if (vcpuid == BSP)
738 			vcpu_info[vcpuid].vcpu = bsp;
739 		else
740 			vcpu_info[vcpuid].vcpu = vm_vcpu_open(ctx, vcpuid);
741 	}
742 
743 	memflags = 0;
744 	if (get_config_bool_default("memory.wired", false))
745 		memflags |= VM_MEM_F_WIRED;
746 	if (get_config_bool_default("memory.guest_in_core", false))
747 		memflags |= VM_MEM_F_INCORE;
748 	vm_set_memflags(ctx, memflags);
749 	error = vm_setup_memory(ctx, memsize, VM_MMAP_ALL);
750 	if (error) {
751 		fprintf(stderr, "Unable to setup memory (%d)\n", errno);
752 		exit(4);
753 	}
754 
755 	init_mem(guest_ncpus);
756 	init_bootrom(ctx);
757 	if (bhyve_init_platform(ctx, bsp) != 0)
758 		exit(4);
759 
760 	if (qemu_fwcfg_init(ctx) != 0) {
761 		fprintf(stderr, "qemu fwcfg initialization error\n");
762 		exit(4);
763 	}
764 
765 	if (qemu_fwcfg_add_file("opt/bhyve/hw.ncpu", sizeof(guest_ncpus),
766 	    &guest_ncpus) != 0) {
767 		fprintf(stderr, "Could not add qemu fwcfg opt/bhyve/hw.ncpu\n");
768 		exit(4);
769 	}
770 
771 	/*
772 	 * Exit if a device emulation finds an error in its initialization
773 	 */
774 	if (init_pci(ctx) != 0) {
775 		EPRINTLN("Device emulation initialization error: %s",
776 		    strerror(errno));
777 		exit(4);
778 	}
779 	if (init_tpm(ctx) != 0) {
780 		EPRINTLN("Failed to init TPM device");
781 		exit(4);
782 	}
783 
784 	/*
785 	 * Initialize after PCI, to allow a bootrom file to reserve the high
786 	 * region.
787 	 */
788 	if (get_config_bool("acpi_tables"))
789 		vmgenc_init(ctx);
790 
791 #ifdef BHYVE_GDB
792 	init_gdb(ctx);
793 #endif
794 
795 	/*
796 	 * Add all vCPUs.
797 	 */
798 	for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++)
799 		bhyve_start_vcpu(vcpu_info[vcpuid].vcpu, vcpuid == BSP);
800 
801 #ifdef BHYVE_SNAPSHOT
802 	if (restore_file != NULL) {
803 		FPRINTLN(stdout, "Pausing pci devs...");
804 		if (vm_pause_devices() != 0) {
805 			EPRINTLN("Failed to pause PCI device state.");
806 			exit(1);
807 		}
808 
809 		FPRINTLN(stdout, "Restoring vm mem...");
810 		if (restore_vm_mem(ctx, &rstate) != 0) {
811 			EPRINTLN("Failed to restore VM memory.");
812 			exit(1);
813 		}
814 
815 		FPRINTLN(stdout, "Restoring pci devs...");
816 		if (vm_restore_devices(&rstate) != 0) {
817 			EPRINTLN("Failed to restore PCI device state.");
818 			exit(1);
819 		}
820 
821 		FPRINTLN(stdout, "Restoring kernel structs...");
822 		if (vm_restore_kern_structs(ctx, &rstate) != 0) {
823 			EPRINTLN("Failed to restore kernel structs.");
824 			exit(1);
825 		}
826 
827 		FPRINTLN(stdout, "Resuming pci devs...");
828 		if (vm_resume_devices() != 0) {
829 			EPRINTLN("Failed to resume PCI device state.");
830 			exit(1);
831 		}
832 	}
833 #endif
834 
835 	if (bhyve_init_platform_late(ctx, bsp) != 0)
836 		exit(4);
837 
838 	/*
839 	 * Change the proc title to include the VM name.
840 	 */
841 	setproctitle("%s", vmname);
842 
843 #ifdef BHYVE_SNAPSHOT
844 	/*
845 	 * checkpointing thread for communication with bhyvectl
846 	 */
847 	if (init_checkpoint_thread(ctx) != 0)
848 		errx(EX_OSERR, "Failed to start checkpoint thread");
849 #endif
850 
851 #ifndef WITHOUT_CAPSICUM
852 	caph_cache_catpages();
853 
854 	if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1)
855 		errx(EX_OSERR, "Unable to apply rights for sandbox");
856 
857 	if (caph_enter() == -1)
858 		errx(EX_OSERR, "cap_enter() failed");
859 #endif
860 
861 #ifdef BHYVE_SNAPSHOT
862 	if (restore_file != NULL) {
863 		destroy_restore_state(&rstate);
864 		if (vm_restore_time(ctx) < 0)
865 			err(EX_OSERR, "Unable to restore time");
866 
867 		for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++)
868 			vm_resume_cpu(vcpu_info[vcpuid].vcpu);
869 	} else
870 #endif
871 		vm_resume_cpu(bsp);
872 
873 	/*
874 	 * Head off to the main event dispatch loop
875 	 */
876 	mevent_dispatch();
877 
878 	exit(4);
879 }
880