xref: /freebsd/usr.sbin/bhyve/bhyverun.c (revision 9088779e3c8b810afb701adb80be154a7b2e0523)
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 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/types.h>
35 #ifndef WITHOUT_CAPSICUM
36 #include <sys/capsicum.h>
37 #endif
38 #include <sys/mman.h>
39 #ifdef BHYVE_SNAPSHOT
40 #include <sys/socket.h>
41 #include <sys/stat.h>
42 #endif
43 #include <sys/time.h>
44 #ifdef BHYVE_SNAPSHOT
45 #include <sys/un.h>
46 #endif
47 
48 #include <amd64/vmm/intel/vmcs.h>
49 #include <x86/apicreg.h>
50 
51 #include <machine/atomic.h>
52 #include <machine/segments.h>
53 
54 #ifndef WITHOUT_CAPSICUM
55 #include <capsicum_helpers.h>
56 #endif
57 #include <stdio.h>
58 #include <stdlib.h>
59 #include <string.h>
60 #include <err.h>
61 #include <errno.h>
62 #ifdef BHYVE_SNAPSHOT
63 #include <fcntl.h>
64 #endif
65 #include <libgen.h>
66 #include <unistd.h>
67 #include <assert.h>
68 #include <pthread.h>
69 #include <pthread_np.h>
70 #include <sysexits.h>
71 #include <stdbool.h>
72 #include <stdint.h>
73 #ifdef BHYVE_SNAPSHOT
74 #include <ucl.h>
75 #include <unistd.h>
76 
77 #include <libxo/xo.h>
78 #endif
79 
80 #include <machine/vmm.h>
81 #ifndef WITHOUT_CAPSICUM
82 #include <machine/vmm_dev.h>
83 #endif
84 #include <machine/vmm_instruction_emul.h>
85 #include <vmmapi.h>
86 
87 #include "bhyverun.h"
88 #include "acpi.h"
89 #include "atkbdc.h"
90 #include "bootrom.h"
91 #include "config.h"
92 #include "inout.h"
93 #include "debug.h"
94 #include "fwctl.h"
95 #include "gdb.h"
96 #include "ioapic.h"
97 #include "kernemu_dev.h"
98 #include "mem.h"
99 #include "mevent.h"
100 #include "mptbl.h"
101 #include "pci_emul.h"
102 #include "pci_irq.h"
103 #include "pci_lpc.h"
104 #include "qemu_fwcfg.h"
105 #include "smbiostbl.h"
106 #ifdef BHYVE_SNAPSHOT
107 #include "snapshot.h"
108 #endif
109 #include "xmsr.h"
110 #include "spinup_ap.h"
111 #include "rtc.h"
112 #include "vmgenc.h"
113 
114 #define GUEST_NIO_PORT		0x488	/* guest upcalls via i/o port */
115 
116 #define MB		(1024UL * 1024)
117 #define GB		(1024UL * MB)
118 
119 static const char * const vmx_exit_reason_desc[] = {
120 	[EXIT_REASON_EXCEPTION] = "Exception or non-maskable interrupt (NMI)",
121 	[EXIT_REASON_EXT_INTR] = "External interrupt",
122 	[EXIT_REASON_TRIPLE_FAULT] = "Triple fault",
123 	[EXIT_REASON_INIT] = "INIT signal",
124 	[EXIT_REASON_SIPI] = "Start-up IPI (SIPI)",
125 	[EXIT_REASON_IO_SMI] = "I/O system-management interrupt (SMI)",
126 	[EXIT_REASON_SMI] = "Other SMI",
127 	[EXIT_REASON_INTR_WINDOW] = "Interrupt window",
128 	[EXIT_REASON_NMI_WINDOW] = "NMI window",
129 	[EXIT_REASON_TASK_SWITCH] = "Task switch",
130 	[EXIT_REASON_CPUID] = "CPUID",
131 	[EXIT_REASON_GETSEC] = "GETSEC",
132 	[EXIT_REASON_HLT] = "HLT",
133 	[EXIT_REASON_INVD] = "INVD",
134 	[EXIT_REASON_INVLPG] = "INVLPG",
135 	[EXIT_REASON_RDPMC] = "RDPMC",
136 	[EXIT_REASON_RDTSC] = "RDTSC",
137 	[EXIT_REASON_RSM] = "RSM",
138 	[EXIT_REASON_VMCALL] = "VMCALL",
139 	[EXIT_REASON_VMCLEAR] = "VMCLEAR",
140 	[EXIT_REASON_VMLAUNCH] = "VMLAUNCH",
141 	[EXIT_REASON_VMPTRLD] = "VMPTRLD",
142 	[EXIT_REASON_VMPTRST] = "VMPTRST",
143 	[EXIT_REASON_VMREAD] = "VMREAD",
144 	[EXIT_REASON_VMRESUME] = "VMRESUME",
145 	[EXIT_REASON_VMWRITE] = "VMWRITE",
146 	[EXIT_REASON_VMXOFF] = "VMXOFF",
147 	[EXIT_REASON_VMXON] = "VMXON",
148 	[EXIT_REASON_CR_ACCESS] = "Control-register accesses",
149 	[EXIT_REASON_DR_ACCESS] = "MOV DR",
150 	[EXIT_REASON_INOUT] = "I/O instruction",
151 	[EXIT_REASON_RDMSR] = "RDMSR",
152 	[EXIT_REASON_WRMSR] = "WRMSR",
153 	[EXIT_REASON_INVAL_VMCS] =
154 	    "VM-entry failure due to invalid guest state",
155 	[EXIT_REASON_INVAL_MSR] = "VM-entry failure due to MSR loading",
156 	[EXIT_REASON_MWAIT] = "MWAIT",
157 	[EXIT_REASON_MTF] = "Monitor trap flag",
158 	[EXIT_REASON_MONITOR] = "MONITOR",
159 	[EXIT_REASON_PAUSE] = "PAUSE",
160 	[EXIT_REASON_MCE_DURING_ENTRY] =
161 	    "VM-entry failure due to machine-check event",
162 	[EXIT_REASON_TPR] = "TPR below threshold",
163 	[EXIT_REASON_APIC_ACCESS] = "APIC access",
164 	[EXIT_REASON_VIRTUALIZED_EOI] = "Virtualized EOI",
165 	[EXIT_REASON_GDTR_IDTR] = "Access to GDTR or IDTR",
166 	[EXIT_REASON_LDTR_TR] = "Access to LDTR or TR",
167 	[EXIT_REASON_EPT_FAULT] = "EPT violation",
168 	[EXIT_REASON_EPT_MISCONFIG] = "EPT misconfiguration",
169 	[EXIT_REASON_INVEPT] = "INVEPT",
170 	[EXIT_REASON_RDTSCP] = "RDTSCP",
171 	[EXIT_REASON_VMX_PREEMPT] = "VMX-preemption timer expired",
172 	[EXIT_REASON_INVVPID] = "INVVPID",
173 	[EXIT_REASON_WBINVD] = "WBINVD",
174 	[EXIT_REASON_XSETBV] = "XSETBV",
175 	[EXIT_REASON_APIC_WRITE] = "APIC write",
176 	[EXIT_REASON_RDRAND] = "RDRAND",
177 	[EXIT_REASON_INVPCID] = "INVPCID",
178 	[EXIT_REASON_VMFUNC] = "VMFUNC",
179 	[EXIT_REASON_ENCLS] = "ENCLS",
180 	[EXIT_REASON_RDSEED] = "RDSEED",
181 	[EXIT_REASON_PM_LOG_FULL] = "Page-modification log full",
182 	[EXIT_REASON_XSAVES] = "XSAVES",
183 	[EXIT_REASON_XRSTORS] = "XRSTORS"
184 };
185 
186 typedef int (*vmexit_handler_t)(struct vmctx *, struct vcpu *, struct vm_exit *);
187 
188 int guest_ncpus;
189 uint16_t cpu_cores, cpu_sockets, cpu_threads;
190 
191 int raw_stdio = 0;
192 
193 static char *progname;
194 static const int BSP = 0;
195 
196 static cpuset_t cpumask;
197 
198 static void vm_loop(struct vmctx *ctx, struct vcpu *vcpu);
199 
200 static struct bhyvestats {
201 	uint64_t	vmexit_bogus;
202 	uint64_t	vmexit_reqidle;
203 	uint64_t	vmexit_hlt;
204 	uint64_t	vmexit_pause;
205 	uint64_t	vmexit_mtrap;
206 	uint64_t	vmexit_inst_emul;
207 	uint64_t	cpu_switch_rotate;
208 	uint64_t	cpu_switch_direct;
209 } stats;
210 
211 static struct vcpu_info {
212 	struct vmctx	*ctx;
213 	struct vcpu	*vcpu;
214 	int		vcpuid;
215 } *vcpu_info;
216 
217 static cpuset_t **vcpumap;
218 
219 static void
220 usage(int code)
221 {
222 
223         fprintf(stderr,
224 		"Usage: %s [-AaCDeHhPSuWwxY]\n"
225 		"       %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n"
226 		"       %*s [-G port] [-k config_file] [-l lpc] [-m mem] [-o var=value]\n"
227 		"       %*s [-p vcpu:hostcpu] [-r file] [-s pci] [-U uuid] vmname\n"
228 		"       -A: create ACPI tables\n"
229 		"       -a: local apic is in xAPIC mode (deprecated)\n"
230 		"       -C: include guest memory in core file\n"
231 		"       -c: number of CPUs and/or topology specification\n"
232 		"       -D: destroy on power-off\n"
233 		"       -e: exit on unhandled I/O access\n"
234 		"       -G: start a debug server\n"
235 		"       -H: vmexit from the guest on HLT\n"
236 		"       -h: help\n"
237 		"       -k: key=value flat config file\n"
238 		"       -K: PS2 keyboard layout\n"
239 		"       -l: LPC device configuration\n"
240 		"       -m: memory size\n"
241 		"       -o: set config 'var' to 'value'\n"
242 		"       -P: vmexit from the guest on pause\n"
243 		"       -p: pin 'vcpu' to 'hostcpu'\n"
244 #ifdef BHYVE_SNAPSHOT
245 		"       -r: path to checkpoint file\n"
246 #endif
247 		"       -S: guest memory cannot be swapped\n"
248 		"       -s: <slot,driver,configinfo> PCI slot config\n"
249 		"       -U: UUID\n"
250 		"       -u: RTC keeps UTC time\n"
251 		"       -W: force virtio to use single-vector MSI\n"
252 		"       -w: ignore unimplemented MSRs\n"
253 		"       -x: local APIC is in x2APIC mode\n"
254 		"       -Y: disable MPtable generation\n",
255 		progname, (int)strlen(progname), "", (int)strlen(progname), "",
256 		(int)strlen(progname), "");
257 
258 	exit(code);
259 }
260 
261 /*
262  * XXX This parser is known to have the following issues:
263  * 1.  It accepts null key=value tokens ",," as setting "cpus" to an
264  *     empty string.
265  *
266  * The acceptance of a null specification ('-c ""') is by design to match the
267  * manual page syntax specification, this results in a topology of 1 vCPU.
268  */
269 static int
270 topology_parse(const char *opt)
271 {
272 	char *cp, *str, *tofree;
273 
274 	if (*opt == '\0') {
275 		set_config_value("sockets", "1");
276 		set_config_value("cores", "1");
277 		set_config_value("threads", "1");
278 		set_config_value("cpus", "1");
279 		return (0);
280 	}
281 
282 	tofree = str = strdup(opt);
283 	if (str == NULL)
284 		errx(4, "Failed to allocate memory");
285 
286 	while ((cp = strsep(&str, ",")) != NULL) {
287 		if (strncmp(cp, "cpus=", strlen("cpus=")) == 0)
288 			set_config_value("cpus", cp + strlen("cpus="));
289 		else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0)
290 			set_config_value("sockets", cp + strlen("sockets="));
291 		else if (strncmp(cp, "cores=", strlen("cores=")) == 0)
292 			set_config_value("cores", cp + strlen("cores="));
293 		else if (strncmp(cp, "threads=", strlen("threads=")) == 0)
294 			set_config_value("threads", cp + strlen("threads="));
295 #ifdef notyet  /* Do not expose this until vmm.ko implements it */
296 		else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0)
297 			set_config_value("maxcpus", cp + strlen("maxcpus="));
298 #endif
299 		else if (strchr(cp, '=') != NULL)
300 			goto out;
301 		else
302 			set_config_value("cpus", cp);
303 	}
304 	free(tofree);
305 	return (0);
306 
307 out:
308 	free(tofree);
309 	return (-1);
310 }
311 
312 static int
313 parse_int_value(const char *key, const char *value, int minval, int maxval)
314 {
315 	char *cp;
316 	long lval;
317 
318 	errno = 0;
319 	lval = strtol(value, &cp, 0);
320 	if (errno != 0 || *cp != '\0' || cp == value || lval < minval ||
321 	    lval > maxval)
322 		errx(4, "Invalid value for %s: '%s'", key, value);
323 	return (lval);
324 }
325 
326 /*
327  * Set the sockets, cores, threads, and guest_cpus variables based on
328  * the configured topology.
329  *
330  * The limits of UINT16_MAX are due to the types passed to
331  * vm_set_topology().  vmm.ko may enforce tighter limits.
332  */
333 static void
334 calc_topology(void)
335 {
336 	const char *value;
337 	bool explicit_cpus;
338 	uint64_t ncpus;
339 
340 	value = get_config_value("cpus");
341 	if (value != NULL) {
342 		guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX);
343 		explicit_cpus = true;
344 	} else {
345 		guest_ncpus = 1;
346 		explicit_cpus = false;
347 	}
348 	value = get_config_value("cores");
349 	if (value != NULL)
350 		cpu_cores = parse_int_value("cores", value, 1, UINT16_MAX);
351 	else
352 		cpu_cores = 1;
353 	value = get_config_value("threads");
354 	if (value != NULL)
355 		cpu_threads = parse_int_value("threads", value, 1, UINT16_MAX);
356 	else
357 		cpu_threads = 1;
358 	value = get_config_value("sockets");
359 	if (value != NULL)
360 		cpu_sockets = parse_int_value("sockets", value, 1, UINT16_MAX);
361 	else
362 		cpu_sockets = guest_ncpus;
363 
364 	/*
365 	 * Compute sockets * cores * threads avoiding overflow.  The
366 	 * range check above insures these are 16 bit values.
367 	 */
368 	ncpus = (uint64_t)cpu_sockets * cpu_cores * cpu_threads;
369 	if (ncpus > UINT16_MAX)
370 		errx(4, "Computed number of vCPUs too high: %ju",
371 		    (uintmax_t)ncpus);
372 
373 	if (explicit_cpus) {
374 		if (guest_ncpus != (int)ncpus)
375 			errx(4, "Topology (%d sockets, %d cores, %d threads) "
376 			    "does not match %d vCPUs",
377 			    cpu_sockets, cpu_cores, cpu_threads,
378 			    guest_ncpus);
379 	} else
380 		guest_ncpus = ncpus;
381 }
382 
383 static int
384 pincpu_parse(const char *opt)
385 {
386 	const char *value;
387 	char *newval;
388 	char key[16];
389 	int vcpu, pcpu;
390 
391 	if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) {
392 		fprintf(stderr, "invalid format: %s\n", opt);
393 		return (-1);
394 	}
395 
396 	if (vcpu < 0) {
397 		fprintf(stderr, "invalid vcpu '%d'\n", vcpu);
398 		return (-1);
399 	}
400 
401 	if (pcpu < 0 || pcpu >= CPU_SETSIZE) {
402 		fprintf(stderr, "hostcpu '%d' outside valid range from "
403 		    "0 to %d\n", pcpu, CPU_SETSIZE - 1);
404 		return (-1);
405 	}
406 
407 	snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
408 	value = get_config_value(key);
409 
410 	if (asprintf(&newval, "%s%s%d", value != NULL ? value : "",
411 	    value != NULL ? "," : "", pcpu) == -1) {
412 		perror("failed to build new cpuset string");
413 		return (-1);
414 	}
415 
416 	set_config_value(key, newval);
417 	free(newval);
418 	return (0);
419 }
420 
421 static void
422 parse_cpuset(int vcpu, const char *list, cpuset_t *set)
423 {
424 	char *cp, *token;
425 	int pcpu, start;
426 
427 	CPU_ZERO(set);
428 	start = -1;
429 	token = __DECONST(char *, list);
430 	for (;;) {
431 		pcpu = strtoul(token, &cp, 0);
432 		if (cp == token)
433 			errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
434 		if (pcpu < 0 || pcpu >= CPU_SETSIZE)
435 			errx(4, "hostcpu '%d' outside valid range from 0 to %d",
436 			    pcpu, CPU_SETSIZE - 1);
437 		switch (*cp) {
438 		case ',':
439 		case '\0':
440 			if (start >= 0) {
441 				if (start > pcpu)
442 					errx(4, "Invalid hostcpu range %d-%d",
443 					    start, pcpu);
444 				while (start < pcpu) {
445 					CPU_SET(start, set);
446 					start++;
447 				}
448 				start = -1;
449 			}
450 			CPU_SET(pcpu, set);
451 			break;
452 		case '-':
453 			if (start >= 0)
454 				errx(4, "invalid cpuset for vcpu %d: '%s'",
455 				    vcpu, list);
456 			start = pcpu;
457 			break;
458 		default:
459 			errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
460 		}
461 		if (*cp == '\0')
462 			break;
463 		token = cp + 1;
464 	}
465 }
466 
467 static void
468 build_vcpumaps(void)
469 {
470 	char key[16];
471 	const char *value;
472 	int vcpu;
473 
474 	vcpumap = calloc(guest_ncpus, sizeof(*vcpumap));
475 	for (vcpu = 0; vcpu < guest_ncpus; vcpu++) {
476 		snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
477 		value = get_config_value(key);
478 		if (value == NULL)
479 			continue;
480 		vcpumap[vcpu] = malloc(sizeof(cpuset_t));
481 		if (vcpumap[vcpu] == NULL)
482 			err(4, "Failed to allocate cpuset for vcpu %d", vcpu);
483 		parse_cpuset(vcpu, value, vcpumap[vcpu]);
484 	}
485 }
486 
487 void
488 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid,
489     int errcode)
490 {
491 	int error, restart_instruction;
492 
493 	restart_instruction = 1;
494 
495 	error = vm_inject_exception(vcpu, vector, errcode_valid, errcode,
496 	    restart_instruction);
497 	assert(error == 0);
498 }
499 
500 void *
501 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len)
502 {
503 
504 	return (vm_map_gpa(ctx, gaddr, len));
505 }
506 
507 #ifdef BHYVE_SNAPSHOT
508 uintptr_t
509 paddr_host2guest(struct vmctx *ctx, void *addr)
510 {
511 	return (vm_rev_map_gpa(ctx, addr));
512 }
513 #endif
514 
515 int
516 fbsdrun_virtio_msix(void)
517 {
518 
519 	return (get_config_bool_default("virtio_msix", true));
520 }
521 
522 static void *
523 fbsdrun_start_thread(void *param)
524 {
525 	char tname[MAXCOMLEN + 1];
526 	struct vcpu_info *vi = param;
527 	int error;
528 
529 	snprintf(tname, sizeof(tname), "vcpu %d", vi->vcpuid);
530 	pthread_set_name_np(pthread_self(), tname);
531 
532 	if (vcpumap[vi->vcpuid] != NULL) {
533 		error = pthread_setaffinity_np(pthread_self(),
534 		    sizeof(cpuset_t), vcpumap[vi->vcpuid]);
535 		assert(error == 0);
536 	}
537 
538 #ifdef BHYVE_SNAPSHOT
539 	checkpoint_cpu_add(vi->vcpuid);
540 #endif
541 	gdb_cpu_add(vi->vcpu);
542 
543 	vm_loop(vi->ctx, vi->vcpu);
544 
545 	/* not reached */
546 	exit(1);
547 	return (NULL);
548 }
549 
550 static void
551 fbsdrun_addcpu(struct vcpu_info *vi)
552 {
553 	pthread_t thr;
554 	int error;
555 
556 	error = vm_activate_cpu(vi->vcpu);
557 	if (error != 0)
558 		err(EX_OSERR, "could not activate CPU %d", vi->vcpuid);
559 
560 	CPU_SET_ATOMIC(vi->vcpuid, &cpumask);
561 
562 	vm_suspend_cpu(vi->vcpu);
563 
564 	error = pthread_create(&thr, NULL, fbsdrun_start_thread, vi);
565 	assert(error == 0);
566 }
567 
568 static int
569 fbsdrun_deletecpu(int vcpu)
570 {
571 
572 	if (!CPU_ISSET(vcpu, &cpumask)) {
573 		fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu);
574 		exit(4);
575 	}
576 
577 	CPU_CLR_ATOMIC(vcpu, &cpumask);
578 	return (CPU_EMPTY(&cpumask));
579 }
580 
581 static int
582 vmexit_handle_notify(struct vmctx *ctx __unused, struct vcpu *vcpu __unused,
583     struct vm_exit *vme __unused, uint32_t eax __unused)
584 {
585 #if BHYVE_DEBUG
586 	/*
587 	 * put guest-driven debug here
588 	 */
589 #endif
590 	return (VMEXIT_CONTINUE);
591 }
592 
593 static int
594 vmexit_inout(struct vmctx *ctx, struct vcpu *vcpu, struct vm_exit *vme)
595 {
596 	int error;
597 	int bytes, port, in, out;
598 
599 	port = vme->u.inout.port;
600 	bytes = vme->u.inout.bytes;
601 	in = vme->u.inout.in;
602 	out = !in;
603 
604         /* Extra-special case of host notifications */
605         if (out && port == GUEST_NIO_PORT) {
606                 error = vmexit_handle_notify(ctx, vcpu, vme, vme->u.inout.eax);
607 		return (error);
608 	}
609 
610 	error = emulate_inout(ctx, vcpu, vme);
611 	if (error) {
612 		fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n",
613 		    in ? "in" : "out",
614 		    bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'),
615 		    port, vme->rip);
616 		return (VMEXIT_ABORT);
617 	} else {
618 		return (VMEXIT_CONTINUE);
619 	}
620 }
621 
622 static int
623 vmexit_rdmsr(struct vmctx *ctx __unused, struct vcpu *vcpu, struct vm_exit *vme)
624 {
625 	uint64_t val;
626 	uint32_t eax, edx;
627 	int error;
628 
629 	val = 0;
630 	error = emulate_rdmsr(vcpu, vme->u.msr.code, &val);
631 	if (error != 0) {
632 		fprintf(stderr, "rdmsr to register %#x on vcpu %d\n",
633 		    vme->u.msr.code, vcpu_id(vcpu));
634 		if (get_config_bool("x86.strictmsr")) {
635 			vm_inject_gp(vcpu);
636 			return (VMEXIT_CONTINUE);
637 		}
638 	}
639 
640 	eax = val;
641 	error = vm_set_register(vcpu, VM_REG_GUEST_RAX, eax);
642 	assert(error == 0);
643 
644 	edx = val >> 32;
645 	error = vm_set_register(vcpu, VM_REG_GUEST_RDX, edx);
646 	assert(error == 0);
647 
648 	return (VMEXIT_CONTINUE);
649 }
650 
651 static int
652 vmexit_wrmsr(struct vmctx *ctx __unused, struct vcpu *vcpu, struct vm_exit *vme)
653 {
654 	int error;
655 
656 	error = emulate_wrmsr(vcpu, vme->u.msr.code, vme->u.msr.wval);
657 	if (error != 0) {
658 		fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n",
659 		    vme->u.msr.code, vme->u.msr.wval, vcpu_id(vcpu));
660 		if (get_config_bool("x86.strictmsr")) {
661 			vm_inject_gp(vcpu);
662 			return (VMEXIT_CONTINUE);
663 		}
664 	}
665 	return (VMEXIT_CONTINUE);
666 }
667 
668 #define	DEBUG_EPT_MISCONFIG
669 #ifdef DEBUG_EPT_MISCONFIG
670 #define	VMCS_GUEST_PHYSICAL_ADDRESS	0x00002400
671 
672 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4];
673 static int ept_misconfig_ptenum;
674 #endif
675 
676 static const char *
677 vmexit_vmx_desc(uint32_t exit_reason)
678 {
679 
680 	if (exit_reason >= nitems(vmx_exit_reason_desc) ||
681 	    vmx_exit_reason_desc[exit_reason] == NULL)
682 		return ("Unknown");
683 	return (vmx_exit_reason_desc[exit_reason]);
684 }
685 
686 static int
687 vmexit_vmx(struct vmctx *ctx, struct vcpu *vcpu, struct vm_exit *vme)
688 {
689 
690 	fprintf(stderr, "vm exit[%d]\n", vcpu_id(vcpu));
691 	fprintf(stderr, "\treason\t\tVMX\n");
692 	fprintf(stderr, "\trip\t\t0x%016lx\n", vme->rip);
693 	fprintf(stderr, "\tinst_length\t%d\n", vme->inst_length);
694 	fprintf(stderr, "\tstatus\t\t%d\n", vme->u.vmx.status);
695 	fprintf(stderr, "\texit_reason\t%u (%s)\n", vme->u.vmx.exit_reason,
696 	    vmexit_vmx_desc(vme->u.vmx.exit_reason));
697 	fprintf(stderr, "\tqualification\t0x%016lx\n",
698 	    vme->u.vmx.exit_qualification);
699 	fprintf(stderr, "\tinst_type\t\t%d\n", vme->u.vmx.inst_type);
700 	fprintf(stderr, "\tinst_error\t\t%d\n", vme->u.vmx.inst_error);
701 #ifdef DEBUG_EPT_MISCONFIG
702 	if (vme->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) {
703 		vm_get_register(vcpu,
704 		    VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS),
705 		    &ept_misconfig_gpa);
706 		vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte,
707 		    &ept_misconfig_ptenum);
708 		fprintf(stderr, "\tEPT misconfiguration:\n");
709 		fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa);
710 		fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n",
711 		    ept_misconfig_ptenum, ept_misconfig_pte[0],
712 		    ept_misconfig_pte[1], ept_misconfig_pte[2],
713 		    ept_misconfig_pte[3]);
714 	}
715 #endif	/* DEBUG_EPT_MISCONFIG */
716 	return (VMEXIT_ABORT);
717 }
718 
719 static int
720 vmexit_svm(struct vmctx *ctx __unused, struct vcpu *vcpu, struct vm_exit *vme)
721 {
722 
723 	fprintf(stderr, "vm exit[%d]\n", vcpu_id(vcpu));
724 	fprintf(stderr, "\treason\t\tSVM\n");
725 	fprintf(stderr, "\trip\t\t0x%016lx\n", vme->rip);
726 	fprintf(stderr, "\tinst_length\t%d\n", vme->inst_length);
727 	fprintf(stderr, "\texitcode\t%#lx\n", vme->u.svm.exitcode);
728 	fprintf(stderr, "\texitinfo1\t%#lx\n", vme->u.svm.exitinfo1);
729 	fprintf(stderr, "\texitinfo2\t%#lx\n", vme->u.svm.exitinfo2);
730 	return (VMEXIT_ABORT);
731 }
732 
733 static int
734 vmexit_bogus(struct vmctx *ctx __unused, struct vcpu *vcpu __unused,
735     struct vm_exit *vme)
736 {
737 
738 	assert(vme->inst_length == 0);
739 
740 	stats.vmexit_bogus++;
741 
742 	return (VMEXIT_CONTINUE);
743 }
744 
745 static int
746 vmexit_reqidle(struct vmctx *ctx __unused, struct vcpu *vcpu __unused,
747     struct vm_exit *vme)
748 {
749 
750 	assert(vme->inst_length == 0);
751 
752 	stats.vmexit_reqidle++;
753 
754 	return (VMEXIT_CONTINUE);
755 }
756 
757 static int
758 vmexit_hlt(struct vmctx *ctx __unused, struct vcpu *vcpu __unused,
759     struct vm_exit *vme __unused)
760 {
761 
762 	stats.vmexit_hlt++;
763 
764 	/*
765 	 * Just continue execution with the next instruction. We use
766 	 * the HLT VM exit as a way to be friendly with the host
767 	 * scheduler.
768 	 */
769 	return (VMEXIT_CONTINUE);
770 }
771 
772 static int
773 vmexit_pause(struct vmctx *ctx __unused, struct vcpu *vcpu __unused,
774     struct vm_exit *vme __unused)
775 {
776 
777 	stats.vmexit_pause++;
778 
779 	return (VMEXIT_CONTINUE);
780 }
781 
782 static int
783 vmexit_mtrap(struct vmctx *ctx __unused, struct vcpu *vcpu,
784     struct vm_exit *vme)
785 {
786 
787 	assert(vme->inst_length == 0);
788 
789 	stats.vmexit_mtrap++;
790 
791 #ifdef BHYVE_SNAPSHOT
792 	checkpoint_cpu_suspend(vcpu_id(vcpu));
793 #endif
794 	gdb_cpu_mtrap(vcpu);
795 #ifdef BHYVE_SNAPSHOT
796 	checkpoint_cpu_resume(vcpu_id(vcpu));
797 #endif
798 
799 	return (VMEXIT_CONTINUE);
800 }
801 
802 static int
803 vmexit_inst_emul(struct vmctx *ctx __unused, struct vcpu *vcpu,
804     struct vm_exit *vme)
805 {
806 	int err, i, cs_d;
807 	struct vie *vie;
808 	enum vm_cpu_mode mode;
809 
810 	stats.vmexit_inst_emul++;
811 
812 	vie = &vme->u.inst_emul.vie;
813 	if (!vie->decoded) {
814 		/*
815 		 * Attempt to decode in userspace as a fallback.  This allows
816 		 * updating instruction decode in bhyve without rebooting the
817 		 * kernel (rapid prototyping), albeit with much slower
818 		 * emulation.
819 		 */
820 		vie_restart(vie);
821 		mode = vme->u.inst_emul.paging.cpu_mode;
822 		cs_d = vme->u.inst_emul.cs_d;
823 		if (vmm_decode_instruction(mode, cs_d, vie) != 0)
824 			goto fail;
825 		if (vm_set_register(vcpu, VM_REG_GUEST_RIP,
826 		    vme->rip + vie->num_processed) != 0)
827 			goto fail;
828 	}
829 
830 	err = emulate_mem(vcpu, vme->u.inst_emul.gpa, vie,
831 	    &vme->u.inst_emul.paging);
832 	if (err) {
833 		if (err == ESRCH) {
834 			EPRINTLN("Unhandled memory access to 0x%lx\n",
835 			    vme->u.inst_emul.gpa);
836 		}
837 		goto fail;
838 	}
839 
840 	return (VMEXIT_CONTINUE);
841 
842 fail:
843 	fprintf(stderr, "Failed to emulate instruction sequence [ ");
844 	for (i = 0; i < vie->num_valid; i++)
845 		fprintf(stderr, "%02x", vie->inst[i]);
846 	FPRINTLN(stderr, " ] at 0x%lx", vme->rip);
847 	return (VMEXIT_ABORT);
848 }
849 
850 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER;
851 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER;
852 
853 static int
854 vmexit_suspend(struct vmctx *ctx, struct vcpu *vcpu, struct vm_exit *vme)
855 {
856 	enum vm_suspend_how how;
857 	int vcpuid = vcpu_id(vcpu);
858 
859 	how = vme->u.suspended.how;
860 
861 	fbsdrun_deletecpu(vcpuid);
862 
863 	if (vcpuid != BSP) {
864 		pthread_mutex_lock(&resetcpu_mtx);
865 		pthread_cond_signal(&resetcpu_cond);
866 		pthread_mutex_unlock(&resetcpu_mtx);
867 		pthread_exit(NULL);
868 	}
869 
870 	pthread_mutex_lock(&resetcpu_mtx);
871 	while (!CPU_EMPTY(&cpumask)) {
872 		pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx);
873 	}
874 	pthread_mutex_unlock(&resetcpu_mtx);
875 
876 	switch (how) {
877 	case VM_SUSPEND_RESET:
878 		exit(0);
879 	case VM_SUSPEND_POWEROFF:
880 		if (get_config_bool_default("destroy_on_poweroff", false))
881 			vm_destroy(ctx);
882 		exit(1);
883 	case VM_SUSPEND_HALT:
884 		exit(2);
885 	case VM_SUSPEND_TRIPLEFAULT:
886 		exit(3);
887 	default:
888 		fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how);
889 		exit(100);
890 	}
891 	return (0);	/* NOTREACHED */
892 }
893 
894 static int
895 vmexit_debug(struct vmctx *ctx __unused, struct vcpu *vcpu,
896     struct vm_exit *vme __unused)
897 {
898 
899 #ifdef BHYVE_SNAPSHOT
900 	checkpoint_cpu_suspend(vcpu_id(vcpu));
901 #endif
902 	gdb_cpu_suspend(vcpu);
903 #ifdef BHYVE_SNAPSHOT
904 	checkpoint_cpu_resume(vcpu_id(vcpu));
905 #endif
906 	/*
907 	 * XXX-MJ sleep for a short period to avoid chewing up the CPU in the
908 	 * window between activation of the vCPU thread and the STARTUP IPI.
909 	 */
910 	usleep(1000);
911 	return (VMEXIT_CONTINUE);
912 }
913 
914 static int
915 vmexit_breakpoint(struct vmctx *ctx __unused, struct vcpu *vcpu,
916     struct vm_exit *vme)
917 {
918 
919 	gdb_cpu_breakpoint(vcpu, vme);
920 	return (VMEXIT_CONTINUE);
921 }
922 
923 static int
924 vmexit_ipi(struct vmctx *ctx __unused, struct vcpu *vcpu __unused,
925     struct vm_exit *vme)
926 {
927 	int error = -1;
928 	int i;
929 	switch (vme->u.ipi.mode) {
930 	case APIC_DELMODE_INIT:
931 		CPU_FOREACH_ISSET(i, &vme->u.ipi.dmask) {
932 			error = vm_suspend_cpu(vcpu_info[i].vcpu);
933 			if (error) {
934 				warnx("%s: failed to suspend cpu %d\n",
935 				    __func__, i);
936 				break;
937 			}
938 		}
939 		break;
940 	case APIC_DELMODE_STARTUP:
941 		CPU_FOREACH_ISSET(i, &vme->u.ipi.dmask) {
942 			spinup_ap(vcpu_info[i].vcpu,
943 			    vme->u.ipi.vector << PAGE_SHIFT);
944 		}
945 		error = 0;
946 		break;
947 	default:
948 		break;
949 	}
950 
951 	return (error);
952 }
953 
954 static vmexit_handler_t handler[VM_EXITCODE_MAX] = {
955 	[VM_EXITCODE_INOUT]  = vmexit_inout,
956 	[VM_EXITCODE_INOUT_STR]  = vmexit_inout,
957 	[VM_EXITCODE_VMX]    = vmexit_vmx,
958 	[VM_EXITCODE_SVM]    = vmexit_svm,
959 	[VM_EXITCODE_BOGUS]  = vmexit_bogus,
960 	[VM_EXITCODE_REQIDLE] = vmexit_reqidle,
961 	[VM_EXITCODE_RDMSR]  = vmexit_rdmsr,
962 	[VM_EXITCODE_WRMSR]  = vmexit_wrmsr,
963 	[VM_EXITCODE_MTRAP]  = vmexit_mtrap,
964 	[VM_EXITCODE_INST_EMUL] = vmexit_inst_emul,
965 	[VM_EXITCODE_SUSPENDED] = vmexit_suspend,
966 	[VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch,
967 	[VM_EXITCODE_DEBUG] = vmexit_debug,
968 	[VM_EXITCODE_BPT] = vmexit_breakpoint,
969 	[VM_EXITCODE_IPI] = vmexit_ipi,
970 };
971 
972 static void
973 vm_loop(struct vmctx *ctx, struct vcpu *vcpu)
974 {
975 	struct vm_exit vme;
976 	int error, rc;
977 	enum vm_exitcode exitcode;
978 	cpuset_t active_cpus;
979 
980 	error = vm_active_cpus(ctx, &active_cpus);
981 	assert(CPU_ISSET(vcpu_id(vcpu), &active_cpus));
982 
983 	while (1) {
984 		error = vm_run(vcpu, &vme);
985 		if (error != 0)
986 			break;
987 
988 		exitcode = vme.exitcode;
989 		if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) {
990 			fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n",
991 			    exitcode);
992 			exit(4);
993 		}
994 
995 		rc = (*handler[exitcode])(ctx, vcpu, &vme);
996 
997 		switch (rc) {
998 		case VMEXIT_CONTINUE:
999 			break;
1000 		case VMEXIT_ABORT:
1001 			abort();
1002 		default:
1003 			exit(4);
1004 		}
1005 	}
1006 	fprintf(stderr, "vm_run error %d, errno %d\n", error, errno);
1007 }
1008 
1009 static int
1010 num_vcpus_allowed(struct vmctx *ctx, struct vcpu *vcpu)
1011 {
1012 	uint16_t sockets, cores, threads, maxcpus;
1013 	int tmp, error;
1014 
1015 	/*
1016 	 * The guest is allowed to spinup more than one processor only if the
1017 	 * UNRESTRICTED_GUEST capability is available.
1018 	 */
1019 	error = vm_get_capability(vcpu, VM_CAP_UNRESTRICTED_GUEST, &tmp);
1020 	if (error != 0)
1021 		return (1);
1022 
1023 	error = vm_get_topology(ctx, &sockets, &cores, &threads, &maxcpus);
1024 	if (error == 0)
1025 		return (maxcpus);
1026 	else
1027 		return (1);
1028 }
1029 
1030 static void
1031 fbsdrun_set_capabilities(struct vcpu *vcpu, bool bsp)
1032 {
1033 	int err, tmp;
1034 
1035 	if (get_config_bool_default("x86.vmexit_on_hlt", false)) {
1036 		err = vm_get_capability(vcpu, VM_CAP_HALT_EXIT, &tmp);
1037 		if (err < 0) {
1038 			fprintf(stderr, "VM exit on HLT not supported\n");
1039 			exit(4);
1040 		}
1041 		vm_set_capability(vcpu, VM_CAP_HALT_EXIT, 1);
1042 		if (bsp)
1043 			handler[VM_EXITCODE_HLT] = vmexit_hlt;
1044 	}
1045 
1046 	if (get_config_bool_default("x86.vmexit_on_pause", false)) {
1047 		/*
1048 		 * pause exit support required for this mode
1049 		 */
1050 		err = vm_get_capability(vcpu, VM_CAP_PAUSE_EXIT, &tmp);
1051 		if (err < 0) {
1052 			fprintf(stderr,
1053 			    "SMP mux requested, no pause support\n");
1054 			exit(4);
1055 		}
1056 		vm_set_capability(vcpu, VM_CAP_PAUSE_EXIT, 1);
1057 		if (bsp)
1058 			handler[VM_EXITCODE_PAUSE] = vmexit_pause;
1059         }
1060 
1061 	if (get_config_bool_default("x86.x2apic", false))
1062 		err = vm_set_x2apic_state(vcpu, X2APIC_ENABLED);
1063 	else
1064 		err = vm_set_x2apic_state(vcpu, X2APIC_DISABLED);
1065 
1066 	if (err) {
1067 		fprintf(stderr, "Unable to set x2apic state (%d)\n", err);
1068 		exit(4);
1069 	}
1070 
1071 	vm_set_capability(vcpu, VM_CAP_ENABLE_INVPCID, 1);
1072 
1073 	err = vm_set_capability(vcpu, VM_CAP_IPI_EXIT, 1);
1074 	assert(err == 0);
1075 }
1076 
1077 static struct vmctx *
1078 do_open(const char *vmname)
1079 {
1080 	struct vmctx *ctx;
1081 	int error;
1082 	bool reinit, romboot;
1083 
1084 	reinit = romboot = false;
1085 
1086 	if (lpc_bootrom())
1087 		romboot = true;
1088 
1089 	error = vm_create(vmname);
1090 	if (error) {
1091 		if (errno == EEXIST) {
1092 			if (romboot) {
1093 				reinit = true;
1094 			} else {
1095 				/*
1096 				 * The virtual machine has been setup by the
1097 				 * userspace bootloader.
1098 				 */
1099 			}
1100 		} else {
1101 			perror("vm_create");
1102 			exit(4);
1103 		}
1104 	} else {
1105 		if (!romboot) {
1106 			/*
1107 			 * If the virtual machine was just created then a
1108 			 * bootrom must be configured to boot it.
1109 			 */
1110 			fprintf(stderr, "virtual machine cannot be booted\n");
1111 			exit(4);
1112 		}
1113 	}
1114 
1115 	ctx = vm_open(vmname);
1116 	if (ctx == NULL) {
1117 		perror("vm_open");
1118 		exit(4);
1119 	}
1120 
1121 #ifndef WITHOUT_CAPSICUM
1122 	if (vm_limit_rights(ctx) != 0)
1123 		err(EX_OSERR, "vm_limit_rights");
1124 #endif
1125 
1126 	if (reinit) {
1127 		error = vm_reinit(ctx);
1128 		if (error) {
1129 			perror("vm_reinit");
1130 			exit(4);
1131 		}
1132 	}
1133 	error = vm_set_topology(ctx, cpu_sockets, cpu_cores, cpu_threads,
1134 	    0 /* maxcpus, unimplemented */);
1135 	if (error)
1136 		errx(EX_OSERR, "vm_set_topology");
1137 	return (ctx);
1138 }
1139 
1140 static void
1141 spinup_vcpu(struct vcpu_info *vi, bool bsp)
1142 {
1143 	int error;
1144 
1145 	if (!bsp) {
1146 		fbsdrun_set_capabilities(vi->vcpu, false);
1147 
1148 		/*
1149 		 * Enable the 'unrestricted guest' mode for APs.
1150 		 *
1151 		 * APs startup in power-on 16-bit mode.
1152 		 */
1153 		error = vm_set_capability(vi->vcpu, VM_CAP_UNRESTRICTED_GUEST, 1);
1154 		assert(error == 0);
1155 	}
1156 
1157 	fbsdrun_addcpu(vi);
1158 }
1159 
1160 static bool
1161 parse_config_option(const char *option)
1162 {
1163 	const char *value;
1164 	char *path;
1165 
1166 	value = strchr(option, '=');
1167 	if (value == NULL || value[1] == '\0')
1168 		return (false);
1169 	path = strndup(option, value - option);
1170 	if (path == NULL)
1171 		err(4, "Failed to allocate memory");
1172 	set_config_value(path, value + 1);
1173 	return (true);
1174 }
1175 
1176 static void
1177 parse_simple_config_file(const char *path)
1178 {
1179 	FILE *fp;
1180 	char *line, *cp;
1181 	size_t linecap;
1182 	unsigned int lineno;
1183 
1184 	fp = fopen(path, "r");
1185 	if (fp == NULL)
1186 		err(4, "Failed to open configuration file %s", path);
1187 	line = NULL;
1188 	linecap = 0;
1189 	lineno = 1;
1190 	for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) {
1191 		if (*line == '#' || *line == '\n')
1192 			continue;
1193 		cp = strchr(line, '\n');
1194 		if (cp != NULL)
1195 			*cp = '\0';
1196 		if (!parse_config_option(line))
1197 			errx(4, "%s line %u: invalid config option '%s'", path,
1198 			    lineno, line);
1199 	}
1200 	free(line);
1201 	fclose(fp);
1202 }
1203 
1204 static void
1205 parse_gdb_options(const char *opt)
1206 {
1207 	const char *sport;
1208 	char *colon;
1209 
1210 	if (opt[0] == 'w') {
1211 		set_config_bool("gdb.wait", true);
1212 		opt++;
1213 	}
1214 
1215 	colon = strrchr(opt, ':');
1216 	if (colon == NULL) {
1217 		sport = opt;
1218 	} else {
1219 		*colon = '\0';
1220 		colon++;
1221 		sport = colon;
1222 		set_config_value("gdb.address", opt);
1223 	}
1224 
1225 	set_config_value("gdb.port", sport);
1226 }
1227 
1228 static void
1229 set_defaults(void)
1230 {
1231 
1232 	set_config_bool("acpi_tables", false);
1233 	set_config_value("memory.size", "256M");
1234 	set_config_bool("x86.strictmsr", true);
1235 	set_config_value("lpc.fwcfg", "bhyve");
1236 }
1237 
1238 int
1239 main(int argc, char *argv[])
1240 {
1241 	int c, error;
1242 	int max_vcpus, memflags;
1243 	struct vcpu *bsp;
1244 	struct vmctx *ctx;
1245 	uint64_t rip;
1246 	size_t memsize;
1247 	const char *optstr, *value, *vmname;
1248 #ifdef BHYVE_SNAPSHOT
1249 	char *restore_file;
1250 	struct restore_state rstate;
1251 
1252 	restore_file = NULL;
1253 #endif
1254 
1255 	init_config();
1256 	set_defaults();
1257 	progname = basename(argv[0]);
1258 
1259 #ifdef BHYVE_SNAPSHOT
1260 	optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:K:U:r:";
1261 #else
1262 	optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:K:U:";
1263 #endif
1264 	while ((c = getopt(argc, argv, optstr)) != -1) {
1265 		switch (c) {
1266 		case 'a':
1267 			set_config_bool("x86.x2apic", false);
1268 			break;
1269 		case 'A':
1270 			set_config_bool("acpi_tables", true);
1271 			break;
1272 		case 'D':
1273 			set_config_bool("destroy_on_poweroff", true);
1274 			break;
1275 		case 'p':
1276                         if (pincpu_parse(optarg) != 0) {
1277                             errx(EX_USAGE, "invalid vcpu pinning "
1278                                  "configuration '%s'", optarg);
1279                         }
1280 			break;
1281                 case 'c':
1282 			if (topology_parse(optarg) != 0) {
1283 			    errx(EX_USAGE, "invalid cpu topology "
1284 				"'%s'", optarg);
1285 			}
1286 			break;
1287 		case 'C':
1288 			set_config_bool("memory.guest_in_core", true);
1289 			break;
1290 		case 'G':
1291 			parse_gdb_options(optarg);
1292 			break;
1293 		case 'k':
1294 			parse_simple_config_file(optarg);
1295 			break;
1296 		case 'K':
1297 			set_config_value("keyboard.layout", optarg);
1298 			break;
1299 		case 'l':
1300 			if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1301 				lpc_print_supported_devices();
1302 				exit(0);
1303 			} else if (lpc_device_parse(optarg) != 0) {
1304 				errx(EX_USAGE, "invalid lpc device "
1305 				    "configuration '%s'", optarg);
1306 			}
1307 			break;
1308 #ifdef BHYVE_SNAPSHOT
1309 		case 'r':
1310 			restore_file = optarg;
1311 			break;
1312 #endif
1313 		case 's':
1314 			if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1315 				pci_print_supported_devices();
1316 				exit(0);
1317 			} else if (pci_parse_slot(optarg) != 0)
1318 				exit(4);
1319 			else
1320 				break;
1321 		case 'S':
1322 			set_config_bool("memory.wired", true);
1323 			break;
1324                 case 'm':
1325 			set_config_value("memory.size", optarg);
1326 			break;
1327 		case 'o':
1328 			if (!parse_config_option(optarg))
1329 				errx(EX_USAGE, "invalid configuration option '%s'", optarg);
1330 			break;
1331 		case 'H':
1332 			set_config_bool("x86.vmexit_on_hlt", true);
1333 			break;
1334 		case 'I':
1335 			/*
1336 			 * The "-I" option was used to add an ioapic to the
1337 			 * virtual machine.
1338 			 *
1339 			 * An ioapic is now provided unconditionally for each
1340 			 * virtual machine and this option is now deprecated.
1341 			 */
1342 			break;
1343 		case 'P':
1344 			set_config_bool("x86.vmexit_on_pause", true);
1345 			break;
1346 		case 'e':
1347 			set_config_bool("x86.strictio", true);
1348 			break;
1349 		case 'u':
1350 			set_config_bool("rtc.use_localtime", false);
1351 			break;
1352 		case 'U':
1353 			set_config_value("uuid", optarg);
1354 			break;
1355 		case 'w':
1356 			set_config_bool("x86.strictmsr", false);
1357 			break;
1358 		case 'W':
1359 			set_config_bool("virtio_msix", false);
1360 			break;
1361 		case 'x':
1362 			set_config_bool("x86.x2apic", true);
1363 			break;
1364 		case 'Y':
1365 			set_config_bool("x86.mptable", false);
1366 			break;
1367 		case 'h':
1368 			usage(0);
1369 		default:
1370 			usage(1);
1371 		}
1372 	}
1373 	argc -= optind;
1374 	argv += optind;
1375 
1376 	if (argc > 1)
1377 		usage(1);
1378 
1379 #ifdef BHYVE_SNAPSHOT
1380 	if (restore_file != NULL) {
1381 		error = load_restore_file(restore_file, &rstate);
1382 		if (error) {
1383 			fprintf(stderr, "Failed to read checkpoint info from "
1384 					"file: '%s'.\n", restore_file);
1385 			exit(1);
1386 		}
1387 		vmname = lookup_vmname(&rstate);
1388 		if (vmname != NULL)
1389 			set_config_value("name", vmname);
1390 	}
1391 #endif
1392 
1393 	if (argc == 1)
1394 		set_config_value("name", argv[0]);
1395 
1396 	vmname = get_config_value("name");
1397 	if (vmname == NULL)
1398 		usage(1);
1399 
1400 	if (get_config_bool_default("config.dump", false)) {
1401 		dump_config();
1402 		exit(1);
1403 	}
1404 
1405 	calc_topology();
1406 	build_vcpumaps();
1407 
1408 	value = get_config_value("memory.size");
1409 	error = vm_parse_memsize(value, &memsize);
1410 	if (error)
1411 		errx(EX_USAGE, "invalid memsize '%s'", value);
1412 
1413 	ctx = do_open(vmname);
1414 
1415 #ifdef BHYVE_SNAPSHOT
1416 	if (restore_file != NULL) {
1417 		guest_ncpus = lookup_guest_ncpus(&rstate);
1418 		memflags = lookup_memflags(&rstate);
1419 		memsize = lookup_memsize(&rstate);
1420 	}
1421 
1422 	if (guest_ncpus < 1) {
1423 		fprintf(stderr, "Invalid guest vCPUs (%d)\n", guest_ncpus);
1424 		exit(1);
1425 	}
1426 #endif
1427 
1428 	bsp = vm_vcpu_open(ctx, BSP);
1429 	max_vcpus = num_vcpus_allowed(ctx, bsp);
1430 	if (guest_ncpus > max_vcpus) {
1431 		fprintf(stderr, "%d vCPUs requested but only %d available\n",
1432 			guest_ncpus, max_vcpus);
1433 		exit(4);
1434 	}
1435 
1436 	fbsdrun_set_capabilities(bsp, true);
1437 
1438 	/* Allocate per-VCPU resources. */
1439 	vcpu_info = calloc(guest_ncpus, sizeof(*vcpu_info));
1440 	for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++) {
1441 		vcpu_info[vcpuid].ctx = ctx;
1442 		vcpu_info[vcpuid].vcpuid = vcpuid;
1443 		if (vcpuid == BSP)
1444 			vcpu_info[vcpuid].vcpu = bsp;
1445 		else
1446 			vcpu_info[vcpuid].vcpu = vm_vcpu_open(ctx, vcpuid);
1447 	}
1448 
1449 	memflags = 0;
1450 	if (get_config_bool_default("memory.wired", false))
1451 		memflags |= VM_MEM_F_WIRED;
1452 	if (get_config_bool_default("memory.guest_in_core", false))
1453 		memflags |= VM_MEM_F_INCORE;
1454 	vm_set_memflags(ctx, memflags);
1455 	error = vm_setup_memory(ctx, memsize, VM_MMAP_ALL);
1456 	if (error) {
1457 		fprintf(stderr, "Unable to setup memory (%d)\n", errno);
1458 		exit(4);
1459 	}
1460 
1461 	error = init_msr();
1462 	if (error) {
1463 		fprintf(stderr, "init_msr error %d", error);
1464 		exit(4);
1465 	}
1466 
1467 	init_mem(guest_ncpus);
1468 	init_inout();
1469 	kernemu_dev_init();
1470 	init_bootrom(ctx);
1471 	atkbdc_init(ctx);
1472 	pci_irq_init(ctx);
1473 	ioapic_init(ctx);
1474 
1475 	rtc_init(ctx);
1476 	sci_init(ctx);
1477 
1478 	if (qemu_fwcfg_init(ctx) != 0) {
1479 		fprintf(stderr, "qemu fwcfg initialization error");
1480 		exit(4);
1481 	}
1482 
1483 	if (qemu_fwcfg_add_file("opt/bhyve/hw.ncpu", sizeof(guest_ncpus),
1484 	    &guest_ncpus) != 0) {
1485 		fprintf(stderr, "Could not add qemu fwcfg opt/bhyve/hw.ncpu");
1486 		exit(4);
1487 	}
1488 
1489 	/*
1490 	 * Exit if a device emulation finds an error in its initilization
1491 	 */
1492 	if (init_pci(ctx) != 0) {
1493 		perror("device emulation initialization error");
1494 		exit(4);
1495 	}
1496 
1497 	/*
1498 	 * Initialize after PCI, to allow a bootrom file to reserve the high
1499 	 * region.
1500 	 */
1501 	if (get_config_bool("acpi_tables"))
1502 		vmgenc_init(ctx);
1503 
1504 	init_gdb(ctx);
1505 
1506 	if (lpc_bootrom()) {
1507 		if (vm_set_capability(bsp, VM_CAP_UNRESTRICTED_GUEST, 1)) {
1508 			fprintf(stderr, "ROM boot failed: unrestricted guest "
1509 			    "capability not available\n");
1510 			exit(4);
1511 		}
1512 		error = vcpu_reset(bsp);
1513 		assert(error == 0);
1514 	}
1515 
1516 	/*
1517 	 * Add all vCPUs.
1518 	 */
1519 	for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++)
1520 		spinup_vcpu(&vcpu_info[vcpuid], vcpuid == BSP);
1521 
1522 #ifdef BHYVE_SNAPSHOT
1523 	if (restore_file != NULL) {
1524 		fprintf(stdout, "Pausing pci devs...\r\n");
1525 		if (vm_pause_user_devs() != 0) {
1526 			fprintf(stderr, "Failed to pause PCI device state.\n");
1527 			exit(1);
1528 		}
1529 
1530 		fprintf(stdout, "Restoring vm mem...\r\n");
1531 		if (restore_vm_mem(ctx, &rstate) != 0) {
1532 			fprintf(stderr, "Failed to restore VM memory.\n");
1533 			exit(1);
1534 		}
1535 
1536 		fprintf(stdout, "Restoring pci devs...\r\n");
1537 		if (vm_restore_user_devs(&rstate) != 0) {
1538 			fprintf(stderr, "Failed to restore PCI device state.\n");
1539 			exit(1);
1540 		}
1541 
1542 		fprintf(stdout, "Restoring kernel structs...\r\n");
1543 		if (vm_restore_kern_structs(ctx, &rstate) != 0) {
1544 			fprintf(stderr, "Failed to restore kernel structs.\n");
1545 			exit(1);
1546 		}
1547 
1548 		fprintf(stdout, "Resuming pci devs...\r\n");
1549 		if (vm_resume_user_devs() != 0) {
1550 			fprintf(stderr, "Failed to resume PCI device state.\n");
1551 			exit(1);
1552 		}
1553 	}
1554 #endif
1555 
1556 	error = vm_get_register(bsp, VM_REG_GUEST_RIP, &rip);
1557 	assert(error == 0);
1558 
1559 	/*
1560 	 * build the guest tables, MP etc.
1561 	 */
1562 	if (get_config_bool_default("x86.mptable", true)) {
1563 		error = mptable_build(ctx, guest_ncpus);
1564 		if (error) {
1565 			perror("error to build the guest tables");
1566 			exit(4);
1567 		}
1568 	}
1569 
1570 	error = smbios_build(ctx);
1571 	if (error != 0)
1572 		exit(4);
1573 
1574 	if (get_config_bool("acpi_tables")) {
1575 		error = acpi_build(ctx, guest_ncpus);
1576 		assert(error == 0);
1577 	}
1578 
1579 	if (lpc_bootrom() && strcmp(lpc_fwcfg(), "bhyve") == 0) {
1580 		fwctl_init();
1581 	}
1582 
1583 	/*
1584 	 * Change the proc title to include the VM name.
1585 	 */
1586 	setproctitle("%s", vmname);
1587 
1588 #ifdef BHYVE_SNAPSHOT
1589 	/* initialize mutex/cond variables */
1590 	init_snapshot();
1591 
1592 	/*
1593 	 * checkpointing thread for communication with bhyvectl
1594 	 */
1595 	if (init_checkpoint_thread(ctx) != 0)
1596 		errx(EX_OSERR, "Failed to start checkpoint thread");
1597 #endif
1598 
1599 #ifndef WITHOUT_CAPSICUM
1600 	caph_cache_catpages();
1601 
1602 	if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1)
1603 		errx(EX_OSERR, "Unable to apply rights for sandbox");
1604 
1605 	if (caph_enter() == -1)
1606 		errx(EX_OSERR, "cap_enter() failed");
1607 #endif
1608 
1609 #ifdef BHYVE_SNAPSHOT
1610 	if (restore_file != NULL) {
1611 		destroy_restore_state(&rstate);
1612 		if (vm_restore_time(ctx) < 0)
1613 			err(EX_OSERR, "Unable to restore time");
1614 
1615 		for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++)
1616 			vm_resume_cpu(vcpu_info[vcpuid].vcpu);
1617 	} else
1618 #endif
1619 		vm_resume_cpu(bsp);
1620 
1621 	/*
1622 	 * Head off to the main event dispatch loop
1623 	 */
1624 	mevent_dispatch();
1625 
1626 	exit(4);
1627 }
1628