xref: /freebsd/usr.sbin/bhyve/bhyverun.c (revision 045c8f526484cb3b97f5fd693987f4376fa43c5f)
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 
50 #include <machine/atomic.h>
51 #include <machine/segments.h>
52 
53 #ifndef WITHOUT_CAPSICUM
54 #include <capsicum_helpers.h>
55 #endif
56 #include <stdio.h>
57 #include <stdlib.h>
58 #include <string.h>
59 #include <err.h>
60 #include <errno.h>
61 #ifdef BHYVE_SNAPSHOT
62 #include <fcntl.h>
63 #endif
64 #include <libgen.h>
65 #include <unistd.h>
66 #include <assert.h>
67 #include <pthread.h>
68 #include <pthread_np.h>
69 #include <sysexits.h>
70 #include <stdbool.h>
71 #include <stdint.h>
72 #ifdef BHYVE_SNAPSHOT
73 #include <ucl.h>
74 #include <unistd.h>
75 
76 #include <libxo/xo.h>
77 #endif
78 
79 #include <machine/vmm.h>
80 #ifndef WITHOUT_CAPSICUM
81 #include <machine/vmm_dev.h>
82 #endif
83 #include <machine/vmm_instruction_emul.h>
84 #include <vmmapi.h>
85 
86 #include "bhyverun.h"
87 #include "acpi.h"
88 #include "atkbdc.h"
89 #include "bootrom.h"
90 #include "config.h"
91 #include "inout.h"
92 #include "debug.h"
93 #include "fwctl.h"
94 #include "gdb.h"
95 #include "ioapic.h"
96 #include "kernemu_dev.h"
97 #include "mem.h"
98 #include "mevent.h"
99 #include "mptbl.h"
100 #include "pci_emul.h"
101 #include "pci_irq.h"
102 #include "pci_lpc.h"
103 #include "smbiostbl.h"
104 #ifdef BHYVE_SNAPSHOT
105 #include "snapshot.h"
106 #endif
107 #include "xmsr.h"
108 #include "spinup_ap.h"
109 #include "rtc.h"
110 #include "vmgenc.h"
111 
112 #define GUEST_NIO_PORT		0x488	/* guest upcalls via i/o port */
113 
114 #define MB		(1024UL * 1024)
115 #define GB		(1024UL * MB)
116 
117 static const char * const vmx_exit_reason_desc[] = {
118 	[EXIT_REASON_EXCEPTION] = "Exception or non-maskable interrupt (NMI)",
119 	[EXIT_REASON_EXT_INTR] = "External interrupt",
120 	[EXIT_REASON_TRIPLE_FAULT] = "Triple fault",
121 	[EXIT_REASON_INIT] = "INIT signal",
122 	[EXIT_REASON_SIPI] = "Start-up IPI (SIPI)",
123 	[EXIT_REASON_IO_SMI] = "I/O system-management interrupt (SMI)",
124 	[EXIT_REASON_SMI] = "Other SMI",
125 	[EXIT_REASON_INTR_WINDOW] = "Interrupt window",
126 	[EXIT_REASON_NMI_WINDOW] = "NMI window",
127 	[EXIT_REASON_TASK_SWITCH] = "Task switch",
128 	[EXIT_REASON_CPUID] = "CPUID",
129 	[EXIT_REASON_GETSEC] = "GETSEC",
130 	[EXIT_REASON_HLT] = "HLT",
131 	[EXIT_REASON_INVD] = "INVD",
132 	[EXIT_REASON_INVLPG] = "INVLPG",
133 	[EXIT_REASON_RDPMC] = "RDPMC",
134 	[EXIT_REASON_RDTSC] = "RDTSC",
135 	[EXIT_REASON_RSM] = "RSM",
136 	[EXIT_REASON_VMCALL] = "VMCALL",
137 	[EXIT_REASON_VMCLEAR] = "VMCLEAR",
138 	[EXIT_REASON_VMLAUNCH] = "VMLAUNCH",
139 	[EXIT_REASON_VMPTRLD] = "VMPTRLD",
140 	[EXIT_REASON_VMPTRST] = "VMPTRST",
141 	[EXIT_REASON_VMREAD] = "VMREAD",
142 	[EXIT_REASON_VMRESUME] = "VMRESUME",
143 	[EXIT_REASON_VMWRITE] = "VMWRITE",
144 	[EXIT_REASON_VMXOFF] = "VMXOFF",
145 	[EXIT_REASON_VMXON] = "VMXON",
146 	[EXIT_REASON_CR_ACCESS] = "Control-register accesses",
147 	[EXIT_REASON_DR_ACCESS] = "MOV DR",
148 	[EXIT_REASON_INOUT] = "I/O instruction",
149 	[EXIT_REASON_RDMSR] = "RDMSR",
150 	[EXIT_REASON_WRMSR] = "WRMSR",
151 	[EXIT_REASON_INVAL_VMCS] =
152 	    "VM-entry failure due to invalid guest state",
153 	[EXIT_REASON_INVAL_MSR] = "VM-entry failure due to MSR loading",
154 	[EXIT_REASON_MWAIT] = "MWAIT",
155 	[EXIT_REASON_MTF] = "Monitor trap flag",
156 	[EXIT_REASON_MONITOR] = "MONITOR",
157 	[EXIT_REASON_PAUSE] = "PAUSE",
158 	[EXIT_REASON_MCE_DURING_ENTRY] =
159 	    "VM-entry failure due to machine-check event",
160 	[EXIT_REASON_TPR] = "TPR below threshold",
161 	[EXIT_REASON_APIC_ACCESS] = "APIC access",
162 	[EXIT_REASON_VIRTUALIZED_EOI] = "Virtualized EOI",
163 	[EXIT_REASON_GDTR_IDTR] = "Access to GDTR or IDTR",
164 	[EXIT_REASON_LDTR_TR] = "Access to LDTR or TR",
165 	[EXIT_REASON_EPT_FAULT] = "EPT violation",
166 	[EXIT_REASON_EPT_MISCONFIG] = "EPT misconfiguration",
167 	[EXIT_REASON_INVEPT] = "INVEPT",
168 	[EXIT_REASON_RDTSCP] = "RDTSCP",
169 	[EXIT_REASON_VMX_PREEMPT] = "VMX-preemption timer expired",
170 	[EXIT_REASON_INVVPID] = "INVVPID",
171 	[EXIT_REASON_WBINVD] = "WBINVD",
172 	[EXIT_REASON_XSETBV] = "XSETBV",
173 	[EXIT_REASON_APIC_WRITE] = "APIC write",
174 	[EXIT_REASON_RDRAND] = "RDRAND",
175 	[EXIT_REASON_INVPCID] = "INVPCID",
176 	[EXIT_REASON_VMFUNC] = "VMFUNC",
177 	[EXIT_REASON_ENCLS] = "ENCLS",
178 	[EXIT_REASON_RDSEED] = "RDSEED",
179 	[EXIT_REASON_PM_LOG_FULL] = "Page-modification log full",
180 	[EXIT_REASON_XSAVES] = "XSAVES",
181 	[EXIT_REASON_XRSTORS] = "XRSTORS"
182 };
183 
184 typedef int (*vmexit_handler_t)(struct vmctx *, struct vm_exit *, int *vcpu);
185 extern int vmexit_task_switch(struct vmctx *, struct vm_exit *, int *vcpu);
186 
187 int guest_ncpus;
188 uint16_t cores, maxcpus, sockets, threads;
189 
190 int raw_stdio = 0;
191 
192 static char *progname;
193 static const int BSP = 0;
194 
195 static cpuset_t cpumask;
196 
197 static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip);
198 
199 static struct vm_exit vmexit[VM_MAXCPU];
200 
201 struct bhyvestats {
202 	uint64_t	vmexit_bogus;
203 	uint64_t	vmexit_reqidle;
204 	uint64_t	vmexit_hlt;
205 	uint64_t	vmexit_pause;
206 	uint64_t	vmexit_mtrap;
207 	uint64_t	vmexit_inst_emul;
208 	uint64_t	cpu_switch_rotate;
209 	uint64_t	cpu_switch_direct;
210 } stats;
211 
212 struct mt_vmm_info {
213 	pthread_t	mt_thr;
214 	struct vmctx	*mt_ctx;
215 	int		mt_vcpu;
216 } mt_vmm_info[VM_MAXCPU];
217 
218 static cpuset_t *vcpumap[VM_MAXCPU] = { NULL };
219 
220 static void
221 usage(int code)
222 {
223 
224         fprintf(stderr,
225 		"Usage: %s [-AaCDeHhPSuWwxY]\n"
226 		"       %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n"
227 		"       %*s [-G port] [-k config_file] [-l lpc] [-m mem] [-o var=value]\n"
228 		"       %*s [-p vcpu:hostcpu] [-r file] [-s pci] [-U uuid] vmname\n"
229 		"       -A: create ACPI tables\n"
230 		"       -a: local apic is in xAPIC mode (deprecated)\n"
231 		"       -C: include guest memory in core file\n"
232 		"       -c: number of CPUs and/or topology specification\n"
233 		"       -D: destroy on power-off\n"
234 		"       -e: exit on unhandled I/O access\n"
235 		"       -G: start a debug server\n"
236 		"       -H: vmexit from the guest on HLT\n"
237 		"       -h: help\n"
238 		"       -k: key=value flat config file\n"
239 		"       -K: PS2 keyboard layout\n"
240 		"       -l: LPC device configuration\n"
241 		"       -m: memory size in MB\n"
242 		"       -o: set config 'var' to 'value'\n"
243 		"       -P: vmexit from the guest on pause\n"
244 		"       -p: pin 'vcpu' to 'hostcpu'\n"
245 #ifdef BHYVE_SNAPSHOT
246 		"       -r: path to checkpoint file\n"
247 #endif
248 		"       -S: guest memory cannot be swapped\n"
249 		"       -s: <slot,driver,configinfo> PCI slot config\n"
250 		"       -U: UUID\n"
251 		"       -u: RTC keeps UTC time\n"
252 		"       -W: force virtio to use single-vector MSI\n"
253 		"       -w: ignore unimplemented MSRs\n"
254 		"       -x: local APIC is in x2APIC mode\n"
255 		"       -Y: disable MPtable generation\n",
256 		progname, (int)strlen(progname), "", (int)strlen(progname), "",
257 		(int)strlen(progname), "");
258 
259 	exit(code);
260 }
261 
262 /*
263  * XXX This parser is known to have the following issues:
264  * 1.  It accepts null key=value tokens ",," as setting "cpus" to an
265  *     empty string.
266  *
267  * The acceptance of a null specification ('-c ""') is by design to match the
268  * manual page syntax specification, this results in a topology of 1 vCPU.
269  */
270 static int
271 topology_parse(const char *opt)
272 {
273 	char *cp, *str, *tofree;
274 
275 	if (*opt == '\0') {
276 		set_config_value("sockets", "1");
277 		set_config_value("cores", "1");
278 		set_config_value("threads", "1");
279 		set_config_value("cpus", "1");
280 		return (0);
281 	}
282 
283 	tofree = str = strdup(opt);
284 	if (str == NULL)
285 		errx(4, "Failed to allocate memory");
286 
287 	while ((cp = strsep(&str, ",")) != NULL) {
288 		if (strncmp(cp, "cpus=", strlen("cpus=")) == 0)
289 			set_config_value("cpus", cp + strlen("cpus="));
290 		else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0)
291 			set_config_value("sockets", cp + strlen("sockets="));
292 		else if (strncmp(cp, "cores=", strlen("cores=")) == 0)
293 			set_config_value("cores", cp + strlen("cores="));
294 		else if (strncmp(cp, "threads=", strlen("threads=")) == 0)
295 			set_config_value("threads", cp + strlen("threads="));
296 #ifdef notyet  /* Do not expose this until vmm.ko implements it */
297 		else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0)
298 			set_config_value("maxcpus", cp + strlen("maxcpus="));
299 #endif
300 		else if (strchr(cp, '=') != NULL)
301 			goto out;
302 		else
303 			set_config_value("cpus", cp);
304 	}
305 	free(tofree);
306 	return (0);
307 
308 out:
309 	free(tofree);
310 	return (-1);
311 }
312 
313 static int
314 parse_int_value(const char *key, const char *value, int minval, int maxval)
315 {
316 	char *cp;
317 	long lval;
318 
319 	errno = 0;
320 	lval = strtol(value, &cp, 0);
321 	if (errno != 0 || *cp != '\0' || cp == value || lval < minval ||
322 	    lval > maxval)
323 		errx(4, "Invalid value for %s: '%s'", key, value);
324 	return (lval);
325 }
326 
327 /*
328  * Set the sockets, cores, threads, and guest_cpus variables based on
329  * the configured topology.
330  *
331  * The limits of UINT16_MAX are due to the types passed to
332  * vm_set_topology().  vmm.ko may enforce tighter limits.
333  */
334 static void
335 calc_topolopgy(void)
336 {
337 	const char *value;
338 	bool explicit_cpus;
339 	uint64_t ncpus;
340 
341 	value = get_config_value("cpus");
342 	if (value != NULL) {
343 		guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX);
344 		explicit_cpus = true;
345 	} else {
346 		guest_ncpus = 1;
347 		explicit_cpus = false;
348 	}
349 	value = get_config_value("cores");
350 	if (value != NULL)
351 		cores = parse_int_value("cores", value, 1, UINT16_MAX);
352 	else
353 		cores = 1;
354 	value = get_config_value("threads");
355 	if (value != NULL)
356 		threads = parse_int_value("threads", value, 1, UINT16_MAX);
357 	else
358 		threads = 1;
359 	value = get_config_value("sockets");
360 	if (value != NULL)
361 		sockets = parse_int_value("sockets", value, 1, UINT16_MAX);
362 	else
363 		sockets = guest_ncpus;
364 
365 	/*
366 	 * Compute sockets * cores * threads avoiding overflow.  The
367 	 * range check above insures these are 16 bit values.
368 	 */
369 	ncpus = (uint64_t)sockets * cores * threads;
370 	if (ncpus > UINT16_MAX)
371 		errx(4, "Computed number of vCPUs too high: %ju",
372 		    (uintmax_t)ncpus);
373 
374 	if (explicit_cpus) {
375 		if (guest_ncpus != ncpus)
376 			errx(4, "Topology (%d sockets, %d cores, %d threads) "
377 			    "does not match %d vCPUs", sockets, cores, 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 || vcpu >= VM_MAXCPU) {
397 		fprintf(stderr, "vcpu '%d' outside valid range from 0 to %d\n",
398 		    vcpu, VM_MAXCPU - 1);
399 		return (-1);
400 	}
401 
402 	if (pcpu < 0 || pcpu >= CPU_SETSIZE) {
403 		fprintf(stderr, "hostcpu '%d' outside valid range from "
404 		    "0 to %d\n", pcpu, CPU_SETSIZE - 1);
405 		return (-1);
406 	}
407 
408 	snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
409 	value = get_config_value(key);
410 
411 	if (asprintf(&newval, "%s%s%d", value != NULL ? value : "",
412 	    value != NULL ? "," : "", pcpu) == -1) {
413 		perror("failed to build new cpuset string");
414 		return (-1);
415 	}
416 
417 	set_config_value(key, newval);
418 	free(newval);
419 	return (0);
420 }
421 
422 static void
423 parse_cpuset(int vcpu, const char *list, cpuset_t *set)
424 {
425 	char *cp, *token;
426 	int pcpu, start;
427 
428 	CPU_ZERO(set);
429 	start = -1;
430 	token = __DECONST(char *, list);
431 	for (;;) {
432 		pcpu = strtoul(token, &cp, 0);
433 		if (cp == token)
434 			errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
435 		if (pcpu < 0 || pcpu >= CPU_SETSIZE)
436 			errx(4, "hostcpu '%d' outside valid range from 0 to %d",
437 			    pcpu, CPU_SETSIZE - 1);
438 		switch (*cp) {
439 		case ',':
440 		case '\0':
441 			if (start >= 0) {
442 				if (start > pcpu)
443 					errx(4, "Invalid hostcpu range %d-%d",
444 					    start, pcpu);
445 				while (start < pcpu) {
446 					CPU_SET(start, vcpumap[vcpu]);
447 					start++;
448 				}
449 				start = -1;
450 			}
451 			CPU_SET(pcpu, vcpumap[vcpu]);
452 			break;
453 		case '-':
454 			if (start >= 0)
455 				errx(4, "invalid cpuset for vcpu %d: '%s'",
456 				    vcpu, list);
457 			start = pcpu;
458 			break;
459 		default:
460 			errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
461 		}
462 		if (*cp == '\0')
463 			break;
464 		token = cp + 1;
465 	}
466 }
467 
468 static void
469 build_vcpumaps(void)
470 {
471 	char key[16];
472 	const char *value;
473 	int vcpu;
474 
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(void *arg, int vcpu, int vector, int errcode_valid,
489     int errcode)
490 {
491 	struct vmctx *ctx;
492 	int error, restart_instruction;
493 
494 	ctx = arg;
495 	restart_instruction = 1;
496 
497 	error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode,
498 	    restart_instruction);
499 	assert(error == 0);
500 }
501 
502 void *
503 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len)
504 {
505 
506 	return (vm_map_gpa(ctx, gaddr, len));
507 }
508 
509 #ifdef BHYVE_SNAPSHOT
510 uintptr_t
511 paddr_host2guest(struct vmctx *ctx, void *addr)
512 {
513 	return (vm_rev_map_gpa(ctx, addr));
514 }
515 #endif
516 
517 int
518 fbsdrun_virtio_msix(void)
519 {
520 
521 	return (get_config_bool_default("virtio_msix", true));
522 }
523 
524 static void *
525 fbsdrun_start_thread(void *param)
526 {
527 	char tname[MAXCOMLEN + 1];
528 	struct mt_vmm_info *mtp;
529 	int vcpu;
530 
531 	mtp = param;
532 	vcpu = mtp->mt_vcpu;
533 
534 	snprintf(tname, sizeof(tname), "vcpu %d", vcpu);
535 	pthread_set_name_np(mtp->mt_thr, tname);
536 
537 #ifdef BHYVE_SNAPSHOT
538 	checkpoint_cpu_add(vcpu);
539 #endif
540 	gdb_cpu_add(vcpu);
541 
542 	vm_loop(mtp->mt_ctx, vcpu, vmexit[vcpu].rip);
543 
544 	/* not reached */
545 	exit(1);
546 	return (NULL);
547 }
548 
549 void
550 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip)
551 {
552 	int error;
553 
554 	assert(fromcpu == BSP);
555 
556 	/*
557 	 * The 'newcpu' must be activated in the context of 'fromcpu'. If
558 	 * vm_activate_cpu() is delayed until newcpu's pthread starts running
559 	 * then vmm.ko is out-of-sync with bhyve and this can create a race
560 	 * with vm_suspend().
561 	 */
562 	error = vm_activate_cpu(ctx, newcpu);
563 	if (error != 0)
564 		err(EX_OSERR, "could not activate CPU %d", newcpu);
565 
566 	CPU_SET_ATOMIC(newcpu, &cpumask);
567 
568 	/*
569 	 * Set up the vmexit struct to allow execution to start
570 	 * at the given RIP
571 	 */
572 	vmexit[newcpu].rip = rip;
573 	vmexit[newcpu].inst_length = 0;
574 
575 	mt_vmm_info[newcpu].mt_ctx = ctx;
576 	mt_vmm_info[newcpu].mt_vcpu = newcpu;
577 
578 	error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL,
579 	    fbsdrun_start_thread, &mt_vmm_info[newcpu]);
580 	assert(error == 0);
581 }
582 
583 static int
584 fbsdrun_deletecpu(struct vmctx *ctx, int vcpu)
585 {
586 
587 	if (!CPU_ISSET(vcpu, &cpumask)) {
588 		fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu);
589 		exit(4);
590 	}
591 
592 	CPU_CLR_ATOMIC(vcpu, &cpumask);
593 	return (CPU_EMPTY(&cpumask));
594 }
595 
596 static int
597 vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu,
598 		     uint32_t eax)
599 {
600 #if BHYVE_DEBUG
601 	/*
602 	 * put guest-driven debug here
603 	 */
604 #endif
605 	return (VMEXIT_CONTINUE);
606 }
607 
608 static int
609 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
610 {
611 	int error;
612 	int bytes, port, in, out;
613 	int vcpu;
614 
615 	vcpu = *pvcpu;
616 
617 	port = vme->u.inout.port;
618 	bytes = vme->u.inout.bytes;
619 	in = vme->u.inout.in;
620 	out = !in;
621 
622         /* Extra-special case of host notifications */
623         if (out && port == GUEST_NIO_PORT) {
624                 error = vmexit_handle_notify(ctx, vme, pvcpu, vme->u.inout.eax);
625 		return (error);
626 	}
627 
628 	error = emulate_inout(ctx, vcpu, vme);
629 	if (error) {
630 		fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n",
631 		    in ? "in" : "out",
632 		    bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'),
633 		    port, vmexit->rip);
634 		return (VMEXIT_ABORT);
635 	} else {
636 		return (VMEXIT_CONTINUE);
637 	}
638 }
639 
640 static int
641 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
642 {
643 	uint64_t val;
644 	uint32_t eax, edx;
645 	int error;
646 
647 	val = 0;
648 	error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val);
649 	if (error != 0) {
650 		fprintf(stderr, "rdmsr to register %#x on vcpu %d\n",
651 		    vme->u.msr.code, *pvcpu);
652 		if (get_config_bool("x86.strictmsr")) {
653 			vm_inject_gp(ctx, *pvcpu);
654 			return (VMEXIT_CONTINUE);
655 		}
656 	}
657 
658 	eax = val;
659 	error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax);
660 	assert(error == 0);
661 
662 	edx = val >> 32;
663 	error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx);
664 	assert(error == 0);
665 
666 	return (VMEXIT_CONTINUE);
667 }
668 
669 static int
670 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
671 {
672 	int error;
673 
674 	error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval);
675 	if (error != 0) {
676 		fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n",
677 		    vme->u.msr.code, vme->u.msr.wval, *pvcpu);
678 		if (get_config_bool("x86.strictmsr")) {
679 			vm_inject_gp(ctx, *pvcpu);
680 			return (VMEXIT_CONTINUE);
681 		}
682 	}
683 	return (VMEXIT_CONTINUE);
684 }
685 
686 static int
687 vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu)
688 {
689 
690 	(void)spinup_ap(ctx, *pvcpu,
691 		    vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip);
692 
693 	return (VMEXIT_CONTINUE);
694 }
695 
696 #define	DEBUG_EPT_MISCONFIG
697 #ifdef DEBUG_EPT_MISCONFIG
698 #define	VMCS_GUEST_PHYSICAL_ADDRESS	0x00002400
699 
700 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4];
701 static int ept_misconfig_ptenum;
702 #endif
703 
704 static const char *
705 vmexit_vmx_desc(uint32_t exit_reason)
706 {
707 
708 	if (exit_reason >= nitems(vmx_exit_reason_desc) ||
709 	    vmx_exit_reason_desc[exit_reason] == NULL)
710 		return ("Unknown");
711 	return (vmx_exit_reason_desc[exit_reason]);
712 }
713 
714 static int
715 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
716 {
717 
718 	fprintf(stderr, "vm exit[%d]\n", *pvcpu);
719 	fprintf(stderr, "\treason\t\tVMX\n");
720 	fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip);
721 	fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length);
722 	fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status);
723 	fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason,
724 	    vmexit_vmx_desc(vmexit->u.vmx.exit_reason));
725 	fprintf(stderr, "\tqualification\t0x%016lx\n",
726 	    vmexit->u.vmx.exit_qualification);
727 	fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type);
728 	fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error);
729 #ifdef DEBUG_EPT_MISCONFIG
730 	if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) {
731 		vm_get_register(ctx, *pvcpu,
732 		    VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS),
733 		    &ept_misconfig_gpa);
734 		vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte,
735 		    &ept_misconfig_ptenum);
736 		fprintf(stderr, "\tEPT misconfiguration:\n");
737 		fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa);
738 		fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n",
739 		    ept_misconfig_ptenum, ept_misconfig_pte[0],
740 		    ept_misconfig_pte[1], ept_misconfig_pte[2],
741 		    ept_misconfig_pte[3]);
742 	}
743 #endif	/* DEBUG_EPT_MISCONFIG */
744 	return (VMEXIT_ABORT);
745 }
746 
747 static int
748 vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
749 {
750 
751 	fprintf(stderr, "vm exit[%d]\n", *pvcpu);
752 	fprintf(stderr, "\treason\t\tSVM\n");
753 	fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip);
754 	fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length);
755 	fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode);
756 	fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1);
757 	fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2);
758 	return (VMEXIT_ABORT);
759 }
760 
761 static int
762 vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
763 {
764 
765 	assert(vmexit->inst_length == 0);
766 
767 	stats.vmexit_bogus++;
768 
769 	return (VMEXIT_CONTINUE);
770 }
771 
772 static int
773 vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
774 {
775 
776 	assert(vmexit->inst_length == 0);
777 
778 	stats.vmexit_reqidle++;
779 
780 	return (VMEXIT_CONTINUE);
781 }
782 
783 static int
784 vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
785 {
786 
787 	stats.vmexit_hlt++;
788 
789 	/*
790 	 * Just continue execution with the next instruction. We use
791 	 * the HLT VM exit as a way to be friendly with the host
792 	 * scheduler.
793 	 */
794 	return (VMEXIT_CONTINUE);
795 }
796 
797 static int
798 vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
799 {
800 
801 	stats.vmexit_pause++;
802 
803 	return (VMEXIT_CONTINUE);
804 }
805 
806 static int
807 vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
808 {
809 
810 	assert(vmexit->inst_length == 0);
811 
812 	stats.vmexit_mtrap++;
813 
814 #ifdef BHYVE_SNAPSHOT
815 	checkpoint_cpu_suspend(*pvcpu);
816 #endif
817 	gdb_cpu_mtrap(*pvcpu);
818 #ifdef BHYVE_SNAPSHOT
819 	checkpoint_cpu_resume(*pvcpu);
820 #endif
821 
822 	return (VMEXIT_CONTINUE);
823 }
824 
825 static int
826 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
827 {
828 	int err, i, cs_d;
829 	struct vie *vie;
830 	enum vm_cpu_mode mode;
831 
832 	stats.vmexit_inst_emul++;
833 
834 	vie = &vmexit->u.inst_emul.vie;
835 	if (!vie->decoded) {
836 		/*
837 		 * Attempt to decode in userspace as a fallback.  This allows
838 		 * updating instruction decode in bhyve without rebooting the
839 		 * kernel (rapid prototyping), albeit with much slower
840 		 * emulation.
841 		 */
842 		vie_restart(vie);
843 		mode = vmexit->u.inst_emul.paging.cpu_mode;
844 		cs_d = vmexit->u.inst_emul.cs_d;
845 		if (vmm_decode_instruction(mode, cs_d, vie) != 0)
846 			goto fail;
847 		if (vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RIP,
848 		    vmexit->rip + vie->num_processed) != 0)
849 			goto fail;
850 	}
851 
852 	err = emulate_mem(ctx, *pvcpu, vmexit->u.inst_emul.gpa,
853 	    vie, &vmexit->u.inst_emul.paging);
854 
855 	if (err) {
856 		if (err == ESRCH) {
857 			EPRINTLN("Unhandled memory access to 0x%lx\n",
858 			    vmexit->u.inst_emul.gpa);
859 		}
860 		goto fail;
861 	}
862 
863 	return (VMEXIT_CONTINUE);
864 
865 fail:
866 	fprintf(stderr, "Failed to emulate instruction sequence [ ");
867 	for (i = 0; i < vie->num_valid; i++)
868 		fprintf(stderr, "%02x", vie->inst[i]);
869 	FPRINTLN(stderr, " ] at 0x%lx", vmexit->rip);
870 	return (VMEXIT_ABORT);
871 }
872 
873 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER;
874 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER;
875 
876 static int
877 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
878 {
879 	enum vm_suspend_how how;
880 
881 	how = vmexit->u.suspended.how;
882 
883 	fbsdrun_deletecpu(ctx, *pvcpu);
884 
885 	if (*pvcpu != BSP) {
886 		pthread_mutex_lock(&resetcpu_mtx);
887 		pthread_cond_signal(&resetcpu_cond);
888 		pthread_mutex_unlock(&resetcpu_mtx);
889 		pthread_exit(NULL);
890 	}
891 
892 	pthread_mutex_lock(&resetcpu_mtx);
893 	while (!CPU_EMPTY(&cpumask)) {
894 		pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx);
895 	}
896 	pthread_mutex_unlock(&resetcpu_mtx);
897 
898 	switch (how) {
899 	case VM_SUSPEND_RESET:
900 		exit(0);
901 	case VM_SUSPEND_POWEROFF:
902 		if (get_config_bool_default("destroy_on_poweroff", false))
903 			vm_destroy(ctx);
904 		exit(1);
905 	case VM_SUSPEND_HALT:
906 		exit(2);
907 	case VM_SUSPEND_TRIPLEFAULT:
908 		exit(3);
909 	default:
910 		fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how);
911 		exit(100);
912 	}
913 	return (0);	/* NOTREACHED */
914 }
915 
916 static int
917 vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
918 {
919 
920 #ifdef BHYVE_SNAPSHOT
921 	checkpoint_cpu_suspend(*pvcpu);
922 #endif
923 	gdb_cpu_suspend(*pvcpu);
924 #ifdef BHYVE_SNAPSHOT
925 	checkpoint_cpu_resume(*pvcpu);
926 #endif
927 	return (VMEXIT_CONTINUE);
928 }
929 
930 static int
931 vmexit_breakpoint(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
932 {
933 
934 	gdb_cpu_breakpoint(*pvcpu, vmexit);
935 	return (VMEXIT_CONTINUE);
936 }
937 
938 static vmexit_handler_t handler[VM_EXITCODE_MAX] = {
939 	[VM_EXITCODE_INOUT]  = vmexit_inout,
940 	[VM_EXITCODE_INOUT_STR]  = vmexit_inout,
941 	[VM_EXITCODE_VMX]    = vmexit_vmx,
942 	[VM_EXITCODE_SVM]    = vmexit_svm,
943 	[VM_EXITCODE_BOGUS]  = vmexit_bogus,
944 	[VM_EXITCODE_REQIDLE] = vmexit_reqidle,
945 	[VM_EXITCODE_RDMSR]  = vmexit_rdmsr,
946 	[VM_EXITCODE_WRMSR]  = vmexit_wrmsr,
947 	[VM_EXITCODE_MTRAP]  = vmexit_mtrap,
948 	[VM_EXITCODE_INST_EMUL] = vmexit_inst_emul,
949 	[VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap,
950 	[VM_EXITCODE_SUSPENDED] = vmexit_suspend,
951 	[VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch,
952 	[VM_EXITCODE_DEBUG] = vmexit_debug,
953 	[VM_EXITCODE_BPT] = vmexit_breakpoint,
954 };
955 
956 static void
957 vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip)
958 {
959 	int error, rc;
960 	enum vm_exitcode exitcode;
961 	cpuset_t active_cpus;
962 
963 	if (vcpumap[vcpu] != NULL) {
964 		error = pthread_setaffinity_np(pthread_self(),
965 		    sizeof(cpuset_t), vcpumap[vcpu]);
966 		assert(error == 0);
967 	}
968 
969 	error = vm_active_cpus(ctx, &active_cpus);
970 	assert(CPU_ISSET(vcpu, &active_cpus));
971 
972 	error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip);
973 	assert(error == 0);
974 
975 	while (1) {
976 		error = vm_run(ctx, vcpu, &vmexit[vcpu]);
977 		if (error != 0)
978 			break;
979 
980 		exitcode = vmexit[vcpu].exitcode;
981 		if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) {
982 			fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n",
983 			    exitcode);
984 			exit(4);
985 		}
986 
987 		rc = (*handler[exitcode])(ctx, &vmexit[vcpu], &vcpu);
988 
989 		switch (rc) {
990 		case VMEXIT_CONTINUE:
991 			break;
992 		case VMEXIT_ABORT:
993 			abort();
994 		default:
995 			exit(4);
996 		}
997 	}
998 	fprintf(stderr, "vm_run error %d, errno %d\n", error, errno);
999 }
1000 
1001 static int
1002 num_vcpus_allowed(struct vmctx *ctx)
1003 {
1004 	int tmp, error;
1005 
1006 	error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp);
1007 
1008 	/*
1009 	 * The guest is allowed to spinup more than one processor only if the
1010 	 * UNRESTRICTED_GUEST capability is available.
1011 	 */
1012 	if (error == 0)
1013 		return (VM_MAXCPU);
1014 	else
1015 		return (1);
1016 }
1017 
1018 void
1019 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu)
1020 {
1021 	int err, tmp;
1022 
1023 	if (get_config_bool_default("x86.vmexit_on_hlt", false)) {
1024 		err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp);
1025 		if (err < 0) {
1026 			fprintf(stderr, "VM exit on HLT not supported\n");
1027 			exit(4);
1028 		}
1029 		vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1);
1030 		if (cpu == BSP)
1031 			handler[VM_EXITCODE_HLT] = vmexit_hlt;
1032 	}
1033 
1034 	if (get_config_bool_default("x86.vmexit_on_pause", false)) {
1035 		/*
1036 		 * pause exit support required for this mode
1037 		 */
1038 		err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp);
1039 		if (err < 0) {
1040 			fprintf(stderr,
1041 			    "SMP mux requested, no pause support\n");
1042 			exit(4);
1043 		}
1044 		vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1);
1045 		if (cpu == BSP)
1046 			handler[VM_EXITCODE_PAUSE] = vmexit_pause;
1047         }
1048 
1049 	if (get_config_bool_default("x86.x2apic", false))
1050 		err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED);
1051 	else
1052 		err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED);
1053 
1054 	if (err) {
1055 		fprintf(stderr, "Unable to set x2apic state (%d)\n", err);
1056 		exit(4);
1057 	}
1058 
1059 	vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1);
1060 }
1061 
1062 static struct vmctx *
1063 do_open(const char *vmname)
1064 {
1065 	struct vmctx *ctx;
1066 	int error;
1067 	bool reinit, romboot;
1068 #ifndef WITHOUT_CAPSICUM
1069 	cap_rights_t rights;
1070 	const cap_ioctl_t *cmds;
1071 	size_t ncmds;
1072 #endif
1073 
1074 	reinit = romboot = false;
1075 
1076 	if (lpc_bootrom())
1077 		romboot = true;
1078 
1079 	error = vm_create(vmname);
1080 	if (error) {
1081 		if (errno == EEXIST) {
1082 			if (romboot) {
1083 				reinit = true;
1084 			} else {
1085 				/*
1086 				 * The virtual machine has been setup by the
1087 				 * userspace bootloader.
1088 				 */
1089 			}
1090 		} else {
1091 			perror("vm_create");
1092 			exit(4);
1093 		}
1094 	} else {
1095 		if (!romboot) {
1096 			/*
1097 			 * If the virtual machine was just created then a
1098 			 * bootrom must be configured to boot it.
1099 			 */
1100 			fprintf(stderr, "virtual machine cannot be booted\n");
1101 			exit(4);
1102 		}
1103 	}
1104 
1105 	ctx = vm_open(vmname);
1106 	if (ctx == NULL) {
1107 		perror("vm_open");
1108 		exit(4);
1109 	}
1110 
1111 #ifndef WITHOUT_CAPSICUM
1112 	cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW);
1113 	if (caph_rights_limit(vm_get_device_fd(ctx), &rights) == -1)
1114 		errx(EX_OSERR, "Unable to apply rights for sandbox");
1115 	vm_get_ioctls(&ncmds);
1116 	cmds = vm_get_ioctls(NULL);
1117 	if (cmds == NULL)
1118 		errx(EX_OSERR, "out of memory");
1119 	if (caph_ioctls_limit(vm_get_device_fd(ctx), cmds, ncmds) == -1)
1120 		errx(EX_OSERR, "Unable to apply rights for sandbox");
1121 	free((cap_ioctl_t *)cmds);
1122 #endif
1123 
1124 	if (reinit) {
1125 		error = vm_reinit(ctx);
1126 		if (error) {
1127 			perror("vm_reinit");
1128 			exit(4);
1129 		}
1130 	}
1131 	error = vm_set_topology(ctx, sockets, cores, threads, maxcpus);
1132 	if (error)
1133 		errx(EX_OSERR, "vm_set_topology");
1134 	return (ctx);
1135 }
1136 
1137 void
1138 spinup_vcpu(struct vmctx *ctx, int vcpu)
1139 {
1140 	int error;
1141 	uint64_t rip;
1142 
1143 	error = vm_get_register(ctx, vcpu, VM_REG_GUEST_RIP, &rip);
1144 	assert(error == 0);
1145 
1146 	fbsdrun_set_capabilities(ctx, vcpu);
1147 	error = vm_set_capability(ctx, vcpu, VM_CAP_UNRESTRICTED_GUEST, 1);
1148 	assert(error == 0);
1149 
1150 	fbsdrun_addcpu(ctx, BSP, vcpu, rip);
1151 }
1152 
1153 static bool
1154 parse_config_option(const char *option)
1155 {
1156 	const char *value;
1157 	char *path;
1158 
1159 	value = strchr(option, '=');
1160 	if (value == NULL || value[1] == '\0')
1161 		return (false);
1162 	path = strndup(option, value - option);
1163 	if (path == NULL)
1164 		err(4, "Failed to allocate memory");
1165 	set_config_value(path, value + 1);
1166 	return (true);
1167 }
1168 
1169 static void
1170 parse_simple_config_file(const char *path)
1171 {
1172 	FILE *fp;
1173 	char *line, *cp;
1174 	size_t linecap;
1175 	unsigned int lineno;
1176 
1177 	fp = fopen(path, "r");
1178 	if (fp == NULL)
1179 		err(4, "Failed to open configuration file %s", path);
1180 	line = NULL;
1181 	linecap = 0;
1182 	lineno = 1;
1183 	for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) {
1184 		if (*line == '#' || *line == '\n')
1185 			continue;
1186 		cp = strchr(line, '\n');
1187 		if (cp != NULL)
1188 			*cp = '\0';
1189 		if (!parse_config_option(line))
1190 			errx(4, "%s line %u: invalid config option '%s'", path,
1191 			    lineno, line);
1192 	}
1193 	free(line);
1194 	fclose(fp);
1195 }
1196 
1197 static void
1198 parse_gdb_options(char *optarg)
1199 {
1200 	const char *sport;
1201 	char *colon;
1202 
1203 	if (optarg[0] == 'w') {
1204 		set_config_bool("gdb.wait", true);
1205 		optarg++;
1206 	}
1207 
1208 	colon = strrchr(optarg, ':');
1209 	if (colon == NULL) {
1210 		sport = optarg;
1211 	} else {
1212 		*colon = '\0';
1213 		colon++;
1214 		sport = colon;
1215 		set_config_value("gdb.address", optarg);
1216 	}
1217 
1218 	set_config_value("gdb.port", sport);
1219 }
1220 
1221 static void
1222 set_defaults(void)
1223 {
1224 
1225 	set_config_bool("acpi_tables", false);
1226 	set_config_value("memory.size", "256M");
1227 	set_config_bool("x86.strictmsr", true);
1228 }
1229 
1230 int
1231 main(int argc, char *argv[])
1232 {
1233 	int c, error, err;
1234 	int max_vcpus, memflags;
1235 	struct vmctx *ctx;
1236 	uint64_t rip;
1237 	size_t memsize;
1238 	const char *value, *vmname;
1239 	char *optstr;
1240 #ifdef BHYVE_SNAPSHOT
1241 	char *restore_file;
1242 	struct restore_state rstate;
1243 	int vcpu;
1244 
1245 	restore_file = NULL;
1246 #endif
1247 
1248 	init_config();
1249 	set_defaults();
1250 	progname = basename(argv[0]);
1251 
1252 #ifdef BHYVE_SNAPSHOT
1253 	optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:K:U:r:";
1254 #else
1255 	optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:K:U:";
1256 #endif
1257 	while ((c = getopt(argc, argv, optstr)) != -1) {
1258 		switch (c) {
1259 		case 'a':
1260 			set_config_bool("x86.x2apic", false);
1261 			break;
1262 		case 'A':
1263 			set_config_bool("acpi_tables", true);
1264 			break;
1265 		case 'D':
1266 			set_config_bool("destroy_on_poweroff", true);
1267 			break;
1268 		case 'p':
1269                         if (pincpu_parse(optarg) != 0) {
1270                             errx(EX_USAGE, "invalid vcpu pinning "
1271                                  "configuration '%s'", optarg);
1272                         }
1273 			break;
1274                 case 'c':
1275 			if (topology_parse(optarg) != 0) {
1276 			    errx(EX_USAGE, "invalid cpu topology "
1277 				"'%s'", optarg);
1278 			}
1279 			break;
1280 		case 'C':
1281 			set_config_bool("memory.guest_in_core", true);
1282 			break;
1283 		case 'G':
1284 			parse_gdb_options(optarg);
1285 			break;
1286 		case 'k':
1287 			parse_simple_config_file(optarg);
1288 			break;
1289 		case 'K':
1290 			set_config_value("keyboard.layout", optarg);
1291 			break;
1292 		case 'l':
1293 			if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1294 				lpc_print_supported_devices();
1295 				exit(0);
1296 			} else if (lpc_device_parse(optarg) != 0) {
1297 				errx(EX_USAGE, "invalid lpc device "
1298 				    "configuration '%s'", optarg);
1299 			}
1300 			break;
1301 #ifdef BHYVE_SNAPSHOT
1302 		case 'r':
1303 			restore_file = optarg;
1304 			break;
1305 #endif
1306 		case 's':
1307 			if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1308 				pci_print_supported_devices();
1309 				exit(0);
1310 			} else if (pci_parse_slot(optarg) != 0)
1311 				exit(4);
1312 			else
1313 				break;
1314 		case 'S':
1315 			set_config_bool("memory.wired", true);
1316 			break;
1317                 case 'm':
1318 			set_config_value("memory.size", optarg);
1319 			break;
1320 		case 'o':
1321 			if (!parse_config_option(optarg))
1322 				errx(EX_USAGE, "invalid configuration option '%s'", optarg);
1323 			break;
1324 		case 'H':
1325 			set_config_bool("x86.vmexit_on_hlt", true);
1326 			break;
1327 		case 'I':
1328 			/*
1329 			 * The "-I" option was used to add an ioapic to the
1330 			 * virtual machine.
1331 			 *
1332 			 * An ioapic is now provided unconditionally for each
1333 			 * virtual machine and this option is now deprecated.
1334 			 */
1335 			break;
1336 		case 'P':
1337 			set_config_bool("x86.vmexit_on_pause", true);
1338 			break;
1339 		case 'e':
1340 			set_config_bool("x86.strictio", true);
1341 			break;
1342 		case 'u':
1343 			set_config_bool("rtc.use_localtime", false);
1344 			break;
1345 		case 'U':
1346 			set_config_value("uuid", optarg);
1347 			break;
1348 		case 'w':
1349 			set_config_bool("x86.strictmsr", false);
1350 			break;
1351 		case 'W':
1352 			set_config_bool("virtio_msix", false);
1353 			break;
1354 		case 'x':
1355 			set_config_bool("x86.x2apic", true);
1356 			break;
1357 		case 'Y':
1358 			set_config_bool("x86.mptable", false);
1359 			break;
1360 		case 'h':
1361 			usage(0);
1362 		default:
1363 			usage(1);
1364 		}
1365 	}
1366 	argc -= optind;
1367 	argv += optind;
1368 
1369 	if (argc > 1)
1370 		usage(1);
1371 
1372 #ifdef BHYVE_SNAPSHOT
1373 	if (restore_file != NULL) {
1374 		error = load_restore_file(restore_file, &rstate);
1375 		if (error) {
1376 			fprintf(stderr, "Failed to read checkpoint info from "
1377 					"file: '%s'.\n", restore_file);
1378 			exit(1);
1379 		}
1380 		vmname = lookup_vmname(&rstate);
1381 		if (vmname != NULL)
1382 			set_config_value("name", vmname);
1383 	}
1384 #endif
1385 
1386 	if (argc == 1)
1387 		set_config_value("name", argv[0]);
1388 
1389 	vmname = get_config_value("name");
1390 	if (vmname == NULL)
1391 		usage(1);
1392 
1393 	if (get_config_bool_default("config.dump", false)) {
1394 		dump_config();
1395 		exit(1);
1396 	}
1397 
1398 	calc_topolopgy();
1399 	build_vcpumaps();
1400 
1401 	value = get_config_value("memory.size");
1402 	error = vm_parse_memsize(value, &memsize);
1403 	if (error)
1404 		errx(EX_USAGE, "invalid memsize '%s'", value);
1405 
1406 	ctx = do_open(vmname);
1407 
1408 #ifdef BHYVE_SNAPSHOT
1409 	if (restore_file != NULL) {
1410 		guest_ncpus = lookup_guest_ncpus(&rstate);
1411 		memflags = lookup_memflags(&rstate);
1412 		memsize = lookup_memsize(&rstate);
1413 	}
1414 
1415 	if (guest_ncpus < 1) {
1416 		fprintf(stderr, "Invalid guest vCPUs (%d)\n", guest_ncpus);
1417 		exit(1);
1418 	}
1419 #endif
1420 
1421 	max_vcpus = num_vcpus_allowed(ctx);
1422 	if (guest_ncpus > max_vcpus) {
1423 		fprintf(stderr, "%d vCPUs requested but only %d available\n",
1424 			guest_ncpus, max_vcpus);
1425 		exit(4);
1426 	}
1427 
1428 	fbsdrun_set_capabilities(ctx, BSP);
1429 
1430 	memflags = 0;
1431 	if (get_config_bool_default("memory.wired", false))
1432 		memflags |= VM_MEM_F_WIRED;
1433 	if (get_config_bool_default("memory.guest_in_core", false))
1434 		memflags |= VM_MEM_F_INCORE;
1435 	vm_set_memflags(ctx, memflags);
1436 	err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL);
1437 	if (err) {
1438 		fprintf(stderr, "Unable to setup memory (%d)\n", errno);
1439 		exit(4);
1440 	}
1441 
1442 	error = init_msr();
1443 	if (error) {
1444 		fprintf(stderr, "init_msr error %d", error);
1445 		exit(4);
1446 	}
1447 
1448 	init_mem();
1449 	init_inout();
1450 	kernemu_dev_init();
1451 	init_bootrom(ctx);
1452 	atkbdc_init(ctx);
1453 	pci_irq_init(ctx);
1454 	ioapic_init(ctx);
1455 
1456 	rtc_init(ctx);
1457 	sci_init(ctx);
1458 
1459 	/*
1460 	 * Exit if a device emulation finds an error in its initilization
1461 	 */
1462 	if (init_pci(ctx) != 0) {
1463 		perror("device emulation initialization error");
1464 		exit(4);
1465 	}
1466 
1467 	/*
1468 	 * Initialize after PCI, to allow a bootrom file to reserve the high
1469 	 * region.
1470 	 */
1471 	if (get_config_bool("acpi_tables"))
1472 		vmgenc_init(ctx);
1473 
1474 	init_gdb(ctx);
1475 
1476 	if (lpc_bootrom()) {
1477 		if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) {
1478 			fprintf(stderr, "ROM boot failed: unrestricted guest "
1479 			    "capability not available\n");
1480 			exit(4);
1481 		}
1482 		error = vcpu_reset(ctx, BSP);
1483 		assert(error == 0);
1484 	}
1485 
1486 #ifdef BHYVE_SNAPSHOT
1487 	if (restore_file != NULL) {
1488 		fprintf(stdout, "Pausing pci devs...\r\n");
1489 		if (vm_pause_user_devs(ctx) != 0) {
1490 			fprintf(stderr, "Failed to pause PCI device state.\n");
1491 			exit(1);
1492 		}
1493 
1494 		fprintf(stdout, "Restoring vm mem...\r\n");
1495 		if (restore_vm_mem(ctx, &rstate) != 0) {
1496 			fprintf(stderr, "Failed to restore VM memory.\n");
1497 			exit(1);
1498 		}
1499 
1500 		fprintf(stdout, "Restoring pci devs...\r\n");
1501 		if (vm_restore_user_devs(ctx, &rstate) != 0) {
1502 			fprintf(stderr, "Failed to restore PCI device state.\n");
1503 			exit(1);
1504 		}
1505 
1506 		fprintf(stdout, "Restoring kernel structs...\r\n");
1507 		if (vm_restore_kern_structs(ctx, &rstate) != 0) {
1508 			fprintf(stderr, "Failed to restore kernel structs.\n");
1509 			exit(1);
1510 		}
1511 
1512 		fprintf(stdout, "Resuming pci devs...\r\n");
1513 		if (vm_resume_user_devs(ctx) != 0) {
1514 			fprintf(stderr, "Failed to resume PCI device state.\n");
1515 			exit(1);
1516 		}
1517 	}
1518 #endif
1519 
1520 	error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip);
1521 	assert(error == 0);
1522 
1523 	/*
1524 	 * build the guest tables, MP etc.
1525 	 */
1526 	if (get_config_bool_default("x86.mptable", true)) {
1527 		error = mptable_build(ctx, guest_ncpus);
1528 		if (error) {
1529 			perror("error to build the guest tables");
1530 			exit(4);
1531 		}
1532 	}
1533 
1534 	error = smbios_build(ctx);
1535 	assert(error == 0);
1536 
1537 	if (get_config_bool("acpi_tables")) {
1538 		error = acpi_build(ctx, guest_ncpus);
1539 		assert(error == 0);
1540 	}
1541 
1542 	if (lpc_bootrom())
1543 		fwctl_init();
1544 
1545 	/*
1546 	 * Change the proc title to include the VM name.
1547 	 */
1548 	setproctitle("%s", vmname);
1549 
1550 #ifndef WITHOUT_CAPSICUM
1551 	caph_cache_catpages();
1552 
1553 	if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1)
1554 		errx(EX_OSERR, "Unable to apply rights for sandbox");
1555 
1556 	if (caph_enter() == -1)
1557 		errx(EX_OSERR, "cap_enter() failed");
1558 #endif
1559 
1560 #ifdef BHYVE_SNAPSHOT
1561 	if (restore_file != NULL)
1562 		destroy_restore_state(&rstate);
1563 
1564 	/* initialize mutex/cond variables */
1565 	init_snapshot();
1566 
1567 	/*
1568 	 * checkpointing thread for communication with bhyvectl
1569 	 */
1570 	if (init_checkpoint_thread(ctx) < 0)
1571 		printf("Failed to start checkpoint thread!\r\n");
1572 
1573 	if (restore_file != NULL)
1574 		vm_restore_time(ctx);
1575 #endif
1576 
1577 	/*
1578 	 * Add CPU 0
1579 	 */
1580 	fbsdrun_addcpu(ctx, BSP, BSP, rip);
1581 
1582 #ifdef BHYVE_SNAPSHOT
1583 	/*
1584 	 * If we restore a VM, start all vCPUs now (including APs), otherwise,
1585 	 * let the guest OS to spin them up later via vmexits.
1586 	 */
1587 	if (restore_file != NULL) {
1588 		for (vcpu = 0; vcpu < guest_ncpus; vcpu++) {
1589 			if (vcpu == BSP)
1590 				continue;
1591 
1592 			fprintf(stdout, "spinning up vcpu no %d...\r\n", vcpu);
1593 			spinup_vcpu(ctx, vcpu);
1594 		}
1595 	}
1596 #endif
1597 
1598 	/*
1599 	 * Head off to the main event dispatch loop
1600 	 */
1601 	mevent_dispatch();
1602 
1603 	exit(4);
1604 }
1605