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
98int guest_ncpus;
99uint16_t cpu_cores, cpu_sockets, cpu_threads;
100
101int raw_stdio = 0;
102
103#ifdef BHYVE_SNAPSHOT
104char *restore_file;
105#endif
106
107static const int BSP = 0;
108
109static cpuset_t cpumask;
110
111static void vm_loop(struct vmctx *ctx, struct vcpu *vcpu);
112
113static struct vcpu_info {
114	struct vmctx	*ctx;
115	struct vcpu	*vcpu;
116	int		vcpuid;
117} *vcpu_info;
118
119static cpuset_t **vcpumap;
120
121/*
122 * XXX This parser is known to have the following issues:
123 * 1.  It accepts null key=value tokens ",," as setting "cpus" to an
124 *     empty string.
125 *
126 * The acceptance of a null specification ('-c ""') is by design to match the
127 * manual page syntax specification, this results in a topology of 1 vCPU.
128 */
129int
130bhyve_topology_parse(const char *opt)
131{
132	char *cp, *str, *tofree;
133
134	if (*opt == '\0') {
135		set_config_value("sockets", "1");
136		set_config_value("cores", "1");
137		set_config_value("threads", "1");
138		set_config_value("cpus", "1");
139		return (0);
140	}
141
142	tofree = str = strdup(opt);
143	if (str == NULL)
144		errx(4, "Failed to allocate memory");
145
146	while ((cp = strsep(&str, ",")) != NULL) {
147		if (strncmp(cp, "cpus=", strlen("cpus=")) == 0)
148			set_config_value("cpus", cp + strlen("cpus="));
149		else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0)
150			set_config_value("sockets", cp + strlen("sockets="));
151		else if (strncmp(cp, "cores=", strlen("cores=")) == 0)
152			set_config_value("cores", cp + strlen("cores="));
153		else if (strncmp(cp, "threads=", strlen("threads=")) == 0)
154			set_config_value("threads", cp + strlen("threads="));
155		else if (strchr(cp, '=') != NULL)
156			goto out;
157		else
158			set_config_value("cpus", cp);
159	}
160	free(tofree);
161	return (0);
162
163out:
164	free(tofree);
165	return (-1);
166}
167
168static int
169parse_int_value(const char *key, const char *value, int minval, int maxval)
170{
171	char *cp;
172	long lval;
173
174	errno = 0;
175	lval = strtol(value, &cp, 0);
176	if (errno != 0 || *cp != '\0' || cp == value || lval < minval ||
177	    lval > maxval)
178		errx(4, "Invalid value for %s: '%s'", key, value);
179	return (lval);
180}
181
182/*
183 * Set the sockets, cores, threads, and guest_cpus variables based on
184 * the configured topology.
185 *
186 * The limits of UINT16_MAX are due to the types passed to
187 * vm_set_topology().  vmm.ko may enforce tighter limits.
188 */
189static void
190calc_topology(void)
191{
192	const char *value;
193	bool explicit_cpus;
194	uint64_t ncpus;
195
196	value = get_config_value("cpus");
197	if (value != NULL) {
198		guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX);
199		explicit_cpus = true;
200	} else {
201		guest_ncpus = 1;
202		explicit_cpus = false;
203	}
204	value = get_config_value("cores");
205	if (value != NULL)
206		cpu_cores = parse_int_value("cores", value, 1, UINT16_MAX);
207	else
208		cpu_cores = 1;
209	value = get_config_value("threads");
210	if (value != NULL)
211		cpu_threads = parse_int_value("threads", value, 1, UINT16_MAX);
212	else
213		cpu_threads = 1;
214	value = get_config_value("sockets");
215	if (value != NULL)
216		cpu_sockets = parse_int_value("sockets", value, 1, UINT16_MAX);
217	else
218		cpu_sockets = guest_ncpus;
219
220	/*
221	 * Compute sockets * cores * threads avoiding overflow.  The
222	 * range check above insures these are 16 bit values.
223	 */
224	ncpus = (uint64_t)cpu_sockets * cpu_cores * cpu_threads;
225	if (ncpus > UINT16_MAX)
226		errx(4, "Computed number of vCPUs too high: %ju",
227		    (uintmax_t)ncpus);
228
229	if (explicit_cpus) {
230		if (guest_ncpus != (int)ncpus)
231			errx(4, "Topology (%d sockets, %d cores, %d threads) "
232			    "does not match %d vCPUs",
233			    cpu_sockets, cpu_cores, cpu_threads,
234			    guest_ncpus);
235	} else
236		guest_ncpus = ncpus;
237}
238
239int
240bhyve_pincpu_parse(const char *opt)
241{
242	const char *value;
243	char *newval;
244	char key[16];
245	int vcpu, pcpu;
246
247	if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) {
248		fprintf(stderr, "invalid format: %s\n", opt);
249		return (-1);
250	}
251
252	if (vcpu < 0) {
253		fprintf(stderr, "invalid vcpu '%d'\n", vcpu);
254		return (-1);
255	}
256
257	if (pcpu < 0 || pcpu >= CPU_SETSIZE) {
258		fprintf(stderr, "hostcpu '%d' outside valid range from "
259		    "0 to %d\n", pcpu, CPU_SETSIZE - 1);
260		return (-1);
261	}
262
263	snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
264	value = get_config_value(key);
265
266	if (asprintf(&newval, "%s%s%d", value != NULL ? value : "",
267	    value != NULL ? "," : "", pcpu) == -1) {
268		perror("failed to build new cpuset string");
269		return (-1);
270	}
271
272	set_config_value(key, newval);
273	free(newval);
274	return (0);
275}
276
277static void
278parse_cpuset(int vcpu, const char *list, cpuset_t *set)
279{
280	char *cp, *token;
281	int pcpu, start;
282
283	CPU_ZERO(set);
284	start = -1;
285	token = __DECONST(char *, list);
286	for (;;) {
287		pcpu = strtoul(token, &cp, 0);
288		if (cp == token)
289			errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
290		if (pcpu < 0 || pcpu >= CPU_SETSIZE)
291			errx(4, "hostcpu '%d' outside valid range from 0 to %d",
292			    pcpu, CPU_SETSIZE - 1);
293		switch (*cp) {
294		case ',':
295		case '\0':
296			if (start >= 0) {
297				if (start > pcpu)
298					errx(4, "Invalid hostcpu range %d-%d",
299					    start, pcpu);
300				while (start < pcpu) {
301					CPU_SET(start, set);
302					start++;
303				}
304				start = -1;
305			}
306			CPU_SET(pcpu, set);
307			break;
308		case '-':
309			if (start >= 0)
310				errx(4, "invalid cpuset for vcpu %d: '%s'",
311				    vcpu, list);
312			start = pcpu;
313			break;
314		default:
315			errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
316		}
317		if (*cp == '\0')
318			break;
319		token = cp + 1;
320	}
321}
322
323static void
324build_vcpumaps(void)
325{
326	char key[16];
327	const char *value;
328	int vcpu;
329
330	vcpumap = calloc(guest_ncpus, sizeof(*vcpumap));
331	for (vcpu = 0; vcpu < guest_ncpus; vcpu++) {
332		snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
333		value = get_config_value(key);
334		if (value == NULL)
335			continue;
336		vcpumap[vcpu] = malloc(sizeof(cpuset_t));
337		if (vcpumap[vcpu] == NULL)
338			err(4, "Failed to allocate cpuset for vcpu %d", vcpu);
339		parse_cpuset(vcpu, value, vcpumap[vcpu]);
340	}
341}
342
343void *
344paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len)
345{
346
347	return (vm_map_gpa(ctx, gaddr, len));
348}
349
350#ifdef BHYVE_SNAPSHOT
351uintptr_t
352paddr_host2guest(struct vmctx *ctx, void *addr)
353{
354	return (vm_rev_map_gpa(ctx, addr));
355}
356#endif
357
358int
359fbsdrun_virtio_msix(void)
360{
361
362	return (get_config_bool_default("virtio_msix", true));
363}
364
365struct vcpu *
366fbsdrun_vcpu(int vcpuid)
367{
368	return (vcpu_info[vcpuid].vcpu);
369}
370
371static void *
372fbsdrun_start_thread(void *param)
373{
374	char tname[MAXCOMLEN + 1];
375	struct vcpu_info *vi = param;
376	int error;
377
378	snprintf(tname, sizeof(tname), "vcpu %d", vi->vcpuid);
379	pthread_set_name_np(pthread_self(), tname);
380
381	if (vcpumap[vi->vcpuid] != NULL) {
382		error = pthread_setaffinity_np(pthread_self(),
383		    sizeof(cpuset_t), vcpumap[vi->vcpuid]);
384		assert(error == 0);
385	}
386
387#ifdef BHYVE_SNAPSHOT
388	checkpoint_cpu_add(vi->vcpuid);
389#endif
390#ifdef BHYVE_GDB
391	gdb_cpu_add(vi->vcpu);
392#endif
393
394	vm_loop(vi->ctx, vi->vcpu);
395
396	/* not reached */
397	exit(1);
398	return (NULL);
399}
400
401void
402fbsdrun_addcpu(int vcpuid)
403{
404	struct vcpu_info *vi;
405	pthread_t thr;
406	int error;
407
408	vi = &vcpu_info[vcpuid];
409
410	error = vm_activate_cpu(vi->vcpu);
411	if (error != 0)
412		err(EX_OSERR, "could not activate CPU %d", vi->vcpuid);
413
414	CPU_SET_ATOMIC(vcpuid, &cpumask);
415
416	vm_suspend_cpu(vi->vcpu);
417
418	error = pthread_create(&thr, NULL, fbsdrun_start_thread, vi);
419	assert(error == 0);
420}
421
422void
423fbsdrun_deletecpu(int vcpu)
424{
425	static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER;
426	static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER;
427
428	pthread_mutex_lock(&resetcpu_mtx);
429	if (!CPU_ISSET(vcpu, &cpumask)) {
430		EPRINTLN("Attempting to delete unknown cpu %d", vcpu);
431		exit(4);
432	}
433
434	CPU_CLR(vcpu, &cpumask);
435
436	if (vcpu != BSP) {
437		pthread_cond_signal(&resetcpu_cond);
438		pthread_mutex_unlock(&resetcpu_mtx);
439		pthread_exit(NULL);
440		/* NOTREACHED */
441	}
442
443	while (!CPU_EMPTY(&cpumask)) {
444		pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx);
445	}
446	pthread_mutex_unlock(&resetcpu_mtx);
447}
448
449int
450fbsdrun_suspendcpu(int vcpuid)
451{
452	return (vm_suspend_cpu(vcpu_info[vcpuid].vcpu));
453}
454
455static void
456vm_loop(struct vmctx *ctx, struct vcpu *vcpu)
457{
458	struct vm_exit vme;
459	struct vm_run vmrun;
460	int error, rc;
461	enum vm_exitcode exitcode;
462	cpuset_t active_cpus, dmask;
463
464	error = vm_active_cpus(ctx, &active_cpus);
465	assert(CPU_ISSET(vcpu_id(vcpu), &active_cpus));
466
467	vmrun.vm_exit = &vme;
468	vmrun.cpuset = &dmask;
469	vmrun.cpusetsize = sizeof(dmask);
470
471	while (1) {
472		error = vm_run(vcpu, &vmrun);
473		if (error != 0)
474			break;
475
476		exitcode = vme.exitcode;
477		if (exitcode >= VM_EXITCODE_MAX ||
478		    vmexit_handlers[exitcode] == NULL) {
479			warnx("vm_loop: unexpected exitcode 0x%x", exitcode);
480			exit(4);
481		}
482
483		rc = (*vmexit_handlers[exitcode])(ctx, vcpu, &vmrun);
484
485		switch (rc) {
486		case VMEXIT_CONTINUE:
487			break;
488		case VMEXIT_ABORT:
489			abort();
490		default:
491			exit(4);
492		}
493	}
494	EPRINTLN("vm_run error %d, errno %d", error, errno);
495}
496
497static int
498num_vcpus_allowed(struct vmctx *ctx, struct vcpu *vcpu)
499{
500	uint16_t sockets, cores, threads, maxcpus;
501	int tmp, error;
502
503	/*
504	 * The guest is allowed to spinup more than one processor only if the
505	 * UNRESTRICTED_GUEST capability is available.
506	 */
507	error = vm_get_capability(vcpu, VM_CAP_UNRESTRICTED_GUEST, &tmp);
508	if (error != 0)
509		return (1);
510
511	error = vm_get_topology(ctx, &sockets, &cores, &threads, &maxcpus);
512	if (error == 0)
513		return (maxcpus);
514	else
515		return (1);
516}
517
518static struct vmctx *
519do_open(const char *vmname)
520{
521	struct vmctx *ctx;
522	int error;
523	bool reinit, romboot;
524
525	reinit = false;
526
527#ifdef __amd64__
528	romboot = lpc_bootrom() != NULL;
529#else
530	romboot = true;
531#endif
532
533	error = vm_create(vmname);
534	if (error) {
535		if (errno == EEXIST) {
536			if (romboot) {
537				reinit = true;
538			} else {
539				/*
540				 * The virtual machine has been setup by the
541				 * userspace bootloader.
542				 */
543			}
544		} else {
545			perror("vm_create");
546			exit(4);
547		}
548	} else {
549		if (!romboot) {
550			/*
551			 * If the virtual machine was just created then a
552			 * bootrom must be configured to boot it.
553			 */
554			fprintf(stderr, "virtual machine cannot be booted\n");
555			exit(4);
556		}
557	}
558
559	ctx = vm_open(vmname);
560	if (ctx == NULL) {
561		perror("vm_open");
562		exit(4);
563	}
564
565#ifndef WITHOUT_CAPSICUM
566	if (vm_limit_rights(ctx) != 0)
567		err(EX_OSERR, "vm_limit_rights");
568#endif
569
570	if (reinit) {
571		error = vm_reinit(ctx);
572		if (error) {
573			perror("vm_reinit");
574			exit(4);
575		}
576	}
577	error = vm_set_topology(ctx, cpu_sockets, cpu_cores, cpu_threads, 0);
578	if (error)
579		errx(EX_OSERR, "vm_set_topology");
580	return (ctx);
581}
582
583bool
584bhyve_parse_config_option(const char *option)
585{
586	const char *value;
587	char *path;
588
589	value = strchr(option, '=');
590	if (value == NULL || value[1] == '\0')
591		return (false);
592	path = strndup(option, value - option);
593	if (path == NULL)
594		err(4, "Failed to allocate memory");
595	set_config_value(path, value + 1);
596	free(path);
597	return (true);
598}
599
600void
601bhyve_parse_simple_config_file(const char *path)
602{
603	FILE *fp;
604	char *line, *cp;
605	size_t linecap;
606	unsigned int lineno;
607
608	fp = fopen(path, "r");
609	if (fp == NULL)
610		err(4, "Failed to open configuration file %s", path);
611	line = NULL;
612	linecap = 0;
613	lineno = 1;
614	for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) {
615		if (*line == '#' || *line == '\n')
616			continue;
617		cp = strchr(line, '\n');
618		if (cp != NULL)
619			*cp = '\0';
620		if (!bhyve_parse_config_option(line))
621			errx(4, "%s line %u: invalid config option '%s'", path,
622			    lineno, line);
623	}
624	free(line);
625	fclose(fp);
626}
627
628#ifdef BHYVE_GDB
629void
630bhyve_parse_gdb_options(const char *opt)
631{
632	const char *sport;
633	char *colon;
634
635	if (opt[0] == 'w') {
636		set_config_bool("gdb.wait", true);
637		opt++;
638	}
639
640	colon = strrchr(opt, ':');
641	if (colon == NULL) {
642		sport = opt;
643	} else {
644		*colon = '\0';
645		colon++;
646		sport = colon;
647		set_config_value("gdb.address", opt);
648	}
649
650	set_config_value("gdb.port", sport);
651}
652#endif
653
654int
655main(int argc, char *argv[])
656{
657	int error;
658	int max_vcpus, memflags;
659	struct vcpu *bsp;
660	struct vmctx *ctx;
661	size_t memsize;
662	const char *value, *vmname;
663#ifdef BHYVE_SNAPSHOT
664	struct restore_state rstate;
665#endif
666
667	bhyve_init_config();
668	bhyve_optparse(argc, argv);
669	argc -= optind;
670	argv += optind;
671
672	if (argc > 1)
673		bhyve_usage(1);
674
675#ifdef BHYVE_SNAPSHOT
676	if (restore_file != NULL) {
677		error = load_restore_file(restore_file, &rstate);
678		if (error) {
679			fprintf(stderr, "Failed to read checkpoint info from "
680					"file: '%s'.\n", restore_file);
681			exit(1);
682		}
683		vmname = lookup_vmname(&rstate);
684		if (vmname != NULL)
685			set_config_value("name", vmname);
686	}
687#endif
688
689	if (argc == 1)
690		set_config_value("name", argv[0]);
691
692	vmname = get_config_value("name");
693	if (vmname == NULL)
694		bhyve_usage(1);
695
696	if (get_config_bool_default("config.dump", false)) {
697		dump_config();
698		exit(1);
699	}
700
701	calc_topology();
702	build_vcpumaps();
703
704	value = get_config_value("memory.size");
705	error = vm_parse_memsize(value, &memsize);
706	if (error)
707		errx(EX_USAGE, "invalid memsize '%s'", value);
708
709	ctx = do_open(vmname);
710
711#ifdef BHYVE_SNAPSHOT
712	if (restore_file != NULL) {
713		guest_ncpus = lookup_guest_ncpus(&rstate);
714		memflags = lookup_memflags(&rstate);
715		memsize = lookup_memsize(&rstate);
716	}
717
718	if (guest_ncpus < 1) {
719		fprintf(stderr, "Invalid guest vCPUs (%d)\n", guest_ncpus);
720		exit(1);
721	}
722#endif
723
724	bsp = vm_vcpu_open(ctx, BSP);
725	max_vcpus = num_vcpus_allowed(ctx, bsp);
726	if (guest_ncpus > max_vcpus) {
727		fprintf(stderr, "%d vCPUs requested but only %d available\n",
728			guest_ncpus, max_vcpus);
729		exit(4);
730	}
731
732	bhyve_init_vcpu(bsp);
733
734	/* Allocate per-VCPU resources. */
735	vcpu_info = calloc(guest_ncpus, sizeof(*vcpu_info));
736	for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++) {
737		vcpu_info[vcpuid].ctx = ctx;
738		vcpu_info[vcpuid].vcpuid = vcpuid;
739		if (vcpuid == BSP)
740			vcpu_info[vcpuid].vcpu = bsp;
741		else
742			vcpu_info[vcpuid].vcpu = vm_vcpu_open(ctx, vcpuid);
743	}
744
745	memflags = 0;
746	if (get_config_bool_default("memory.wired", false))
747		memflags |= VM_MEM_F_WIRED;
748	if (get_config_bool_default("memory.guest_in_core", false))
749		memflags |= VM_MEM_F_INCORE;
750	vm_set_memflags(ctx, memflags);
751	error = vm_setup_memory(ctx, memsize, VM_MMAP_ALL);
752	if (error) {
753		fprintf(stderr, "Unable to setup memory (%d)\n", errno);
754		exit(4);
755	}
756
757	init_mem(guest_ncpus);
758	init_bootrom(ctx);
759	if (bhyve_init_platform(ctx, bsp) != 0)
760		exit(4);
761
762	if (qemu_fwcfg_init(ctx) != 0) {
763		fprintf(stderr, "qemu fwcfg initialization error\n");
764		exit(4);
765	}
766
767	if (qemu_fwcfg_add_file("opt/bhyve/hw.ncpu", sizeof(guest_ncpus),
768	    &guest_ncpus) != 0) {
769		fprintf(stderr, "Could not add qemu fwcfg opt/bhyve/hw.ncpu\n");
770		exit(4);
771	}
772
773	/*
774	 * Exit if a device emulation finds an error in its initialization
775	 */
776	if (init_pci(ctx) != 0) {
777		EPRINTLN("Device emulation initialization error: %s",
778		    strerror(errno));
779		exit(4);
780	}
781	if (init_tpm(ctx) != 0) {
782		EPRINTLN("Failed to init TPM device");
783		exit(4);
784	}
785
786	/*
787	 * Initialize after PCI, to allow a bootrom file to reserve the high
788	 * region.
789	 */
790	if (get_config_bool("acpi_tables"))
791		vmgenc_init(ctx);
792
793#ifdef BHYVE_GDB
794	init_gdb(ctx);
795#endif
796
797	/*
798	 * Add all vCPUs.
799	 */
800	for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++)
801		bhyve_start_vcpu(vcpu_info[vcpuid].vcpu, vcpuid == BSP);
802
803#ifdef BHYVE_SNAPSHOT
804	if (restore_file != NULL) {
805		FPRINTLN(stdout, "Pausing pci devs...");
806		if (vm_pause_devices() != 0) {
807			EPRINTLN("Failed to pause PCI device state.");
808			exit(1);
809		}
810
811		FPRINTLN(stdout, "Restoring vm mem...");
812		if (restore_vm_mem(ctx, &rstate) != 0) {
813			EPRINTLN("Failed to restore VM memory.");
814			exit(1);
815		}
816
817		FPRINTLN(stdout, "Restoring pci devs...");
818		if (vm_restore_devices(&rstate) != 0) {
819			EPRINTLN("Failed to restore PCI device state.");
820			exit(1);
821		}
822
823		FPRINTLN(stdout, "Restoring kernel structs...");
824		if (vm_restore_kern_structs(ctx, &rstate) != 0) {
825			EPRINTLN("Failed to restore kernel structs.");
826			exit(1);
827		}
828
829		FPRINTLN(stdout, "Resuming pci devs...");
830		if (vm_resume_devices() != 0) {
831			EPRINTLN("Failed to resume PCI device state.");
832			exit(1);
833		}
834	}
835#endif
836
837	if (bhyve_init_platform_late(ctx, bsp) != 0)
838		exit(4);
839
840	/*
841	 * Change the proc title to include the VM name.
842	 */
843	setproctitle("%s", vmname);
844
845#ifdef BHYVE_SNAPSHOT
846	/*
847	 * checkpointing thread for communication with bhyvectl
848	 */
849	if (init_checkpoint_thread(ctx) != 0)
850		errx(EX_OSERR, "Failed to start checkpoint thread");
851#endif
852
853#ifndef WITHOUT_CAPSICUM
854	caph_cache_catpages();
855
856	if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1)
857		errx(EX_OSERR, "Unable to apply rights for sandbox");
858
859	if (caph_enter() == -1)
860		errx(EX_OSERR, "cap_enter() failed");
861#endif
862
863#ifdef BHYVE_SNAPSHOT
864	if (restore_file != NULL) {
865		destroy_restore_state(&rstate);
866		if (vm_restore_time(ctx) < 0)
867			err(EX_OSERR, "Unable to restore time");
868
869		for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++)
870			vm_resume_cpu(vcpu_info[vcpuid].vcpu);
871	} else
872#endif
873		vm_resume_cpu(bsp);
874
875	/*
876	 * Head off to the main event dispatch loop
877	 */
878	mevent_dispatch();
879
880	exit(4);
881}
882