usr.sbin/bhyve/bhyverun.c

   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/cdefs.h>
  30 #include <sys/types.h>
  31 #ifndef WITHOUT_CAPSICUM
  32 #include <sys/capsicum.h>
  33 #endif
  34 #include <sys/mman.h>
  35 #ifdef BHYVE_SNAPSHOT
  36 #include <sys/socket.h>
  37 #include <sys/stat.h>
  38 #endif
  39 #include <sys/time.h>
  40 #ifdef BHYVE_SNAPSHOT
  41 #include <sys/un.h>
  42 #endif
  43
  44 #include <amd64/vmm/intel/vmcs.h>
  45 #include <x86/apicreg.h>
  46
  47 #include <machine/atomic.h>
  48 #include <machine/segments.h>
  49
  50 #ifndef WITHOUT_CAPSICUM
  51 #include <capsicum_helpers.h>
  52 #endif
  53 #include <stdio.h>
  54 #include <stdlib.h>
  55 #include <string.h>
  56 #include <err.h>
  57 #include <errno.h>
  58 #ifdef BHYVE_SNAPSHOT
  59 #include <fcntl.h>
  60 #endif
  61 #include <libgen.h>
  62 #include <unistd.h>
  63 #include <assert.h>
  64 #include <pthread.h>
  65 #include <pthread_np.h>
  66 #include <sysexits.h>
  67 #include <stdbool.h>
  68 #include <stdint.h>
  69 #ifdef BHYVE_SNAPSHOT
  70 #include <ucl.h>
  71 #include <unistd.h>
  72
  73 #include <libxo/xo.h>
  74 #endif
  75
  76 #include <machine/vmm.h>
  77 #ifndef WITHOUT_CAPSICUM
  78 #include <machine/vmm_dev.h>
  79 #endif
  80 #include <machine/vmm_instruction_emul.h>
  81 #include <vmmapi.h>
  82
  83 #include "bhyverun.h"
  84 #include "acpi.h"
  85 #include "atkbdc.h"
  86 #include "bootrom.h"
  87 #include "config.h"
  88 #include "inout.h"
  89 #include "debug.h"
  90 #include "e820.h"
  91 #include "fwctl.h"
  92 #include "gdb.h"
  93 #include "ioapic.h"
  94 #include "kernemu_dev.h"
  95 #include "mem.h"
  96 #include "mevent.h"
  97 #include "mptbl.h"
  98 #include "pci_emul.h"
  99 #include "pci_irq.h"
 100 #include "pci_lpc.h"
 101 #include "qemu_fwcfg.h"
 102 #include "smbiostbl.h"
 103 #ifdef BHYVE_SNAPSHOT
 104 #include "snapshot.h"
 105 #endif
 106 #include "tpm_device.h"
 107 #include "xmsr.h"
 108 #include "spinup_ap.h"
 109 #include "rtc.h"
 110 #include "vmgenc.h"
 111
 112 #define MB              (1024UL * 1024)
 113 #define GB              (1024UL * MB)
 114
 115 static const char * const vmx_exit_reason_desc[] = {
 116         [EXIT_REASON_EXCEPTION] = "Exception or non-maskable interrupt (NMI)",
 117         [EXIT_REASON_EXT_INTR] = "External interrupt",
 118         [EXIT_REASON_TRIPLE_FAULT] = "Triple fault",
 119         [EXIT_REASON_INIT] = "INIT signal",
 120         [EXIT_REASON_SIPI] = "Start-up IPI (SIPI)",
 121         [EXIT_REASON_IO_SMI] = "I/O system-management interrupt (SMI)",
 122         [EXIT_REASON_SMI] = "Other SMI",
 123         [EXIT_REASON_INTR_WINDOW] = "Interrupt window",
 124         [EXIT_REASON_NMI_WINDOW] = "NMI window",
 125         [EXIT_REASON_TASK_SWITCH] = "Task switch",
 126         [EXIT_REASON_CPUID] = "CPUID",
 127         [EXIT_REASON_GETSEC] = "GETSEC",
 128         [EXIT_REASON_HLT] = "HLT",
 129         [EXIT_REASON_INVD] = "INVD",
 130         [EXIT_REASON_INVLPG] = "INVLPG",
 131         [EXIT_REASON_RDPMC] = "RDPMC",
 132         [EXIT_REASON_RDTSC] = "RDTSC",
 133         [EXIT_REASON_RSM] = "RSM",
 134         [EXIT_REASON_VMCALL] = "VMCALL",
 135         [EXIT_REASON_VMCLEAR] = "VMCLEAR",
 136         [EXIT_REASON_VMLAUNCH] = "VMLAUNCH",
 137         [EXIT_REASON_VMPTRLD] = "VMPTRLD",
 138         [EXIT_REASON_VMPTRST] = "VMPTRST",
 139         [EXIT_REASON_VMREAD] = "VMREAD",
 140         [EXIT_REASON_VMRESUME] = "VMRESUME",
 141         [EXIT_REASON_VMWRITE] = "VMWRITE",
 142         [EXIT_REASON_VMXOFF] = "VMXOFF",
 143         [EXIT_REASON_VMXON] = "VMXON",
 144         [EXIT_REASON_CR_ACCESS] = "Control-register accesses",
 145         [EXIT_REASON_DR_ACCESS] = "MOV DR",
 146         [EXIT_REASON_INOUT] = "I/O instruction",
 147         [EXIT_REASON_RDMSR] = "RDMSR",
 148         [EXIT_REASON_WRMSR] = "WRMSR",
 149         [EXIT_REASON_INVAL_VMCS] =
 150             "VM-entry failure due to invalid guest state",
 151         [EXIT_REASON_INVAL_MSR] = "VM-entry failure due to MSR loading",
 152         [EXIT_REASON_MWAIT] = "MWAIT",
 153         [EXIT_REASON_MTF] = "Monitor trap flag",
 154         [EXIT_REASON_MONITOR] = "MONITOR",
 155         [EXIT_REASON_PAUSE] = "PAUSE",
 156         [EXIT_REASON_MCE_DURING_ENTRY] =
 157             "VM-entry failure due to machine-check event",
 158         [EXIT_REASON_TPR] = "TPR below threshold",
 159         [EXIT_REASON_APIC_ACCESS] = "APIC access",
 160         [EXIT_REASON_VIRTUALIZED_EOI] = "Virtualized EOI",
 161         [EXIT_REASON_GDTR_IDTR] = "Access to GDTR or IDTR",
 162         [EXIT_REASON_LDTR_TR] = "Access to LDTR or TR",
 163         [EXIT_REASON_EPT_FAULT] = "EPT violation",
 164         [EXIT_REASON_EPT_MISCONFIG] = "EPT misconfiguration",
 165         [EXIT_REASON_INVEPT] = "INVEPT",
 166         [EXIT_REASON_RDTSCP] = "RDTSCP",
 167         [EXIT_REASON_VMX_PREEMPT] = "VMX-preemption timer expired",
 168         [EXIT_REASON_INVVPID] = "INVVPID",
 169         [EXIT_REASON_WBINVD] = "WBINVD",
 170         [EXIT_REASON_XSETBV] = "XSETBV",
 171         [EXIT_REASON_APIC_WRITE] = "APIC write",
 172         [EXIT_REASON_RDRAND] = "RDRAND",
 173         [EXIT_REASON_INVPCID] = "INVPCID",
 174         [EXIT_REASON_VMFUNC] = "VMFUNC",
 175         [EXIT_REASON_ENCLS] = "ENCLS",
 176         [EXIT_REASON_RDSEED] = "RDSEED",
 177         [EXIT_REASON_PM_LOG_FULL] = "Page-modification log full",
 178         [EXIT_REASON_XSAVES] = "XSAVES",
 179         [EXIT_REASON_XRSTORS] = "XRSTORS"
 180 };
 181
 182 typedef int (*vmexit_handler_t)(struct vmctx *, struct vcpu *, struct vm_run *);
 183
 184 int guest_ncpus;
 185 uint16_t cpu_cores, cpu_sockets, cpu_threads;
 186
 187 int raw_stdio = 0;
 188
 189 static char *progname;
 190 static const int BSP = 0;
 191
 192 static cpuset_t cpumask;
 193
 194 static void vm_loop(struct vmctx *ctx, struct vcpu *vcpu);
 195
 196 static struct vcpu_info {
 197         struct vmctx    *ctx;
 198         struct vcpu     *vcpu;
 199         int             vcpuid;
 200 } *vcpu_info;
 201
 202 static cpuset_t **vcpumap;
 203
 204 static void
 205 usage(int code)
 206 {
 207
 208         fprintf(stderr,
 209                 "Usage: %s [-AaCDeHhPSuWwxY]\n"
 210                 "       %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n"
 211                 "       %*s [-G port] [-k config_file] [-l lpc] [-m mem] [-o var=value]\n"
 212                 "       %*s [-p vcpu:hostcpu] [-r file] [-s pci] [-U uuid] vmname\n"
 213                 "       -A: create ACPI tables\n"
 214                 "       -a: local apic is in xAPIC mode (deprecated)\n"
 215                 "       -C: include guest memory in core file\n"
 216                 "       -c: number of CPUs and/or topology specification\n"
 217                 "       -D: destroy on power-off\n"
 218                 "       -e: exit on unhandled I/O access\n"
 219                 "       -G: start a debug server\n"
 220                 "       -H: vmexit from the guest on HLT\n"
 221                 "       -h: help\n"
 222                 "       -k: key=value flat config file\n"
 223                 "       -K: PS2 keyboard layout\n"
 224                 "       -l: LPC device configuration\n"
 225                 "       -m: memory size\n"
 226                 "       -o: set config 'var' to 'value'\n"
 227                 "       -P: vmexit from the guest on pause\n"
 228                 "       -p: pin 'vcpu' to 'hostcpu'\n"
 229 #ifdef BHYVE_SNAPSHOT
 230                 "       -r: path to checkpoint file\n"
 231 #endif
 232                 "       -S: guest memory cannot be swapped\n"
 233                 "       -s: <slot,driver,configinfo> PCI slot config\n"
 234                 "       -U: UUID\n"
 235                 "       -u: RTC keeps UTC time\n"
 236                 "       -W: force virtio to use single-vector MSI\n"
 237                 "       -w: ignore unimplemented MSRs\n"
 238                 "       -x: local APIC is in x2APIC mode\n"
 239                 "       -Y: disable MPtable generation\n",
 240                 progname, (int)strlen(progname), "", (int)strlen(progname), "",
 241                 (int)strlen(progname), "");
 242
 243         exit(code);
 244 }
 245
 246 /*
 247  * XXX This parser is known to have the following issues:
 248  * 1.  It accepts null key=value tokens ",," as setting "cpus" to an
 249  *     empty string.
 250  *
 251  * The acceptance of a null specification ('-c ""') is by design to match the
 252  * manual page syntax specification, this results in a topology of 1 vCPU.
 253  */
 254 static int
 255 topology_parse(const char *opt)
 256 {
 257         char *cp, *str, *tofree;
 258
 259         if (*opt == '\0') {
 260                 set_config_value("sockets", "1");
 261                 set_config_value("cores", "1");
 262                 set_config_value("threads", "1");
 263                 set_config_value("cpus", "1");
 264                 return (0);
 265         }
 266
 267         tofree = str = strdup(opt);
 268         if (str == NULL)
 269                 errx(4, "Failed to allocate memory");
 270
 271         while ((cp = strsep(&str, ",")) != NULL) {
 272                 if (strncmp(cp, "cpus=", strlen("cpus=")) == 0)
 273                         set_config_value("cpus", cp + strlen("cpus="));
 274                 else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0)
 275                         set_config_value("sockets", cp + strlen("sockets="));
 276                 else if (strncmp(cp, "cores=", strlen("cores=")) == 0)
 277                         set_config_value("cores", cp + strlen("cores="));
 278                 else if (strncmp(cp, "threads=", strlen("threads=")) == 0)
 279                         set_config_value("threads", cp + strlen("threads="));
 280                 else if (strchr(cp, '=') != NULL)
 281                         goto out;
 282                 else
 283                         set_config_value("cpus", cp);
 284         }
 285         free(tofree);
 286         return (0);
 287
 288 out:
 289         free(tofree);
 290         return (-1);
 291 }
 292
 293 static int
 294 parse_int_value(const char *key, const char *value, int minval, int maxval)
 295 {
 296         char *cp;
 297         long lval;
 298
 299         errno = 0;
 300         lval = strtol(value, &cp, 0);
 301         if (errno != 0 || *cp != '\0' || cp == value || lval < minval ||
 302             lval > maxval)
 303                 errx(4, "Invalid value for %s: '%s'", key, value);
 304         return (lval);
 305 }
 306
 307 /*
 308  * Set the sockets, cores, threads, and guest_cpus variables based on
 309  * the configured topology.
 310  *
 311  * The limits of UINT16_MAX are due to the types passed to
 312  * vm_set_topology().  vmm.ko may enforce tighter limits.
 313  */
 314 static void
 315 calc_topology(void)
 316 {
 317         const char *value;
 318         bool explicit_cpus;
 319         uint64_t ncpus;
 320
 321         value = get_config_value("cpus");
 322         if (value != NULL) {
 323                 guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX);
 324                 explicit_cpus = true;
 325         } else {
 326                 guest_ncpus = 1;
 327                 explicit_cpus = false;
 328         }
 329         value = get_config_value("cores");
 330         if (value != NULL)
 331                 cpu_cores = parse_int_value("cores", value, 1, UINT16_MAX);
 332         else
 333                 cpu_cores = 1;
 334         value = get_config_value("threads");
 335         if (value != NULL)
 336                 cpu_threads = parse_int_value("threads", value, 1, UINT16_MAX);
 337         else
 338                 cpu_threads = 1;
 339         value = get_config_value("sockets");
 340         if (value != NULL)
 341                 cpu_sockets = parse_int_value("sockets", value, 1, UINT16_MAX);
 342         else
 343                 cpu_sockets = guest_ncpus;
 344
 345         /*
 346          * Compute sockets * cores * threads avoiding overflow.  The
 347          * range check above insures these are 16 bit values.
 348          */
 349         ncpus = (uint64_t)cpu_sockets * cpu_cores * cpu_threads;
 350         if (ncpus > UINT16_MAX)
 351                 errx(4, "Computed number of vCPUs too high: %ju",
 352                     (uintmax_t)ncpus);
 353
 354         if (explicit_cpus) {
 355                 if (guest_ncpus != (int)ncpus)
 356                         errx(4, "Topology (%d sockets, %d cores, %d threads) "
 357                             "does not match %d vCPUs",
 358                             cpu_sockets, cpu_cores, cpu_threads,
 359                             guest_ncpus);
 360         } else
 361                 guest_ncpus = ncpus;
 362 }
 363
 364 static int
 365 pincpu_parse(const char *opt)
 366 {
 367         const char *value;
 368         char *newval;
 369         char key[16];
 370         int vcpu, pcpu;
 371
 372         if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) {
 373                 fprintf(stderr, "invalid format: %s\n", opt);
 374                 return (-1);
 375         }
 376
 377         if (vcpu < 0) {
 378                 fprintf(stderr, "invalid vcpu '%d'\n", vcpu);
 379                 return (-1);
 380         }
 381
 382         if (pcpu < 0 || pcpu >= CPU_SETSIZE) {
 383                 fprintf(stderr, "hostcpu '%d' outside valid range from "
 384                     "0 to %d\n", pcpu, CPU_SETSIZE - 1);
 385                 return (-1);
 386         }
 387
 388         snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
 389         value = get_config_value(key);
 390
 391         if (asprintf(&newval, "%s%s%d", value != NULL ? value : "",
 392             value != NULL ? "," : "", pcpu) == -1) {
 393                 perror("failed to build new cpuset string");
 394                 return (-1);
 395         }
 396
 397         set_config_value(key, newval);
 398         free(newval);
 399         return (0);
 400 }
 401
 402 static void
 403 parse_cpuset(int vcpu, const char *list, cpuset_t *set)
 404 {
 405         char *cp, *token;
 406         int pcpu, start;
 407
 408         CPU_ZERO(set);
 409         start = -1;
 410         token = __DECONST(char *, list);
 411         for (;;) {
 412                 pcpu = strtoul(token, &cp, 0);
 413                 if (cp == token)
 414                         errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
 415                 if (pcpu < 0 || pcpu >= CPU_SETSIZE)
 416                         errx(4, "hostcpu '%d' outside valid range from 0 to %d",
 417                             pcpu, CPU_SETSIZE - 1);
 418                 switch (*cp) {
 419                 case ',':
 420                 case '\0':
 421                         if (start >= 0) {
 422                                 if (start > pcpu)
 423                                         errx(4, "Invalid hostcpu range %d-%d",
 424                                             start, pcpu);
 425                                 while (start < pcpu) {
 426                                         CPU_SET(start, set);
 427                                         start++;
 428                                 }
 429                                 start = -1;
 430                         }
 431                         CPU_SET(pcpu, set);
 432                         break;
 433                 case '-':
 434                         if (start >= 0)
 435                                 errx(4, "invalid cpuset for vcpu %d: '%s'",
 436                                     vcpu, list);
 437                         start = pcpu;
 438                         break;
 439                 default:
 440                         errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list);
 441                 }
 442                 if (*cp == '\0')
 443                         break;
 444                 token = cp + 1;
 445         }
 446 }
 447
 448 static void
 449 build_vcpumaps(void)
 450 {
 451         char key[16];
 452         const char *value;
 453         int vcpu;
 454
 455         vcpumap = calloc(guest_ncpus, sizeof(*vcpumap));
 456         for (vcpu = 0; vcpu < guest_ncpus; vcpu++) {
 457                 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu);
 458                 value = get_config_value(key);
 459                 if (value == NULL)
 460                         continue;
 461                 vcpumap[vcpu] = malloc(sizeof(cpuset_t));
 462                 if (vcpumap[vcpu] == NULL)
 463                         err(4, "Failed to allocate cpuset for vcpu %d", vcpu);
 464                 parse_cpuset(vcpu, value, vcpumap[vcpu]);
 465         }
 466 }
 467
 468 void
 469 vm_inject_fault(struct vcpu *vcpu, int vector, int errcode_valid,
 470     int errcode)
 471 {
 472         int error, restart_instruction;
 473
 474         restart_instruction = 1;
 475
 476         error = vm_inject_exception(vcpu, vector, errcode_valid, errcode,
 477             restart_instruction);
 478         assert(error == 0);
 479 }
 480
 481 void *
 482 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len)
 483 {
 484
 485         return (vm_map_gpa(ctx, gaddr, len));
 486 }
 487
 488 #ifdef BHYVE_SNAPSHOT
 489 uintptr_t
 490 paddr_host2guest(struct vmctx *ctx, void *addr)
 491 {
 492         return (vm_rev_map_gpa(ctx, addr));
 493 }
 494 #endif
 495
 496 int
 497 fbsdrun_virtio_msix(void)
 498 {
 499
 500         return (get_config_bool_default("virtio_msix", true));
 501 }
 502
 503 static void *
 504 fbsdrun_start_thread(void *param)
 505 {
 506         char tname[MAXCOMLEN + 1];
 507         struct vcpu_info *vi = param;
 508         int error;
 509
 510         snprintf(tname, sizeof(tname), "vcpu %d", vi->vcpuid);
 511         pthread_set_name_np(pthread_self(), tname);
 512
 513         if (vcpumap[vi->vcpuid] != NULL) {
 514                 error = pthread_setaffinity_np(pthread_self(),
 515                     sizeof(cpuset_t), vcpumap[vi->vcpuid]);
 516                 assert(error == 0);
 517         }
 518
 519 #ifdef BHYVE_SNAPSHOT
 520         checkpoint_cpu_add(vi->vcpuid);
 521 #endif
 522         gdb_cpu_add(vi->vcpu);
 523
 524         vm_loop(vi->ctx, vi->vcpu);
 525
 526         /* not reached */
 527         exit(1);
 528         return (NULL);
 529 }
 530
 531 static void
 532 fbsdrun_addcpu(struct vcpu_info *vi)
 533 {
 534         pthread_t thr;
 535         int error;
 536
 537         error = vm_activate_cpu(vi->vcpu);
 538         if (error != 0)
 539                 err(EX_OSERR, "could not activate CPU %d", vi->vcpuid);
 540
 541         CPU_SET_ATOMIC(vi->vcpuid, &cpumask);
 542
 543         vm_suspend_cpu(vi->vcpu);
 544
 545         error = pthread_create(&thr, NULL, fbsdrun_start_thread, vi);
 546         assert(error == 0);
 547 }
 548
 549 static void
 550 fbsdrun_deletecpu(int vcpu)
 551 {
 552         static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER;
 553         static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER;
 554
 555         pthread_mutex_lock(&resetcpu_mtx);
 556         if (!CPU_ISSET(vcpu, &cpumask)) {
 557                 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu);
 558                 exit(4);
 559         }
 560
 561         CPU_CLR(vcpu, &cpumask);
 562
 563         if (vcpu != BSP) {
 564                 pthread_cond_signal(&resetcpu_cond);
 565                 pthread_mutex_unlock(&resetcpu_mtx);
 566                 pthread_exit(NULL);
 567                 /* NOTREACHED */
 568         }
 569
 570         while (!CPU_EMPTY(&cpumask)) {
 571                 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx);
 572         }
 573         pthread_mutex_unlock(&resetcpu_mtx);
 574 }
 575
 576 static int
 577 vmexit_inout(struct vmctx *ctx, struct vcpu *vcpu, struct vm_run *vmrun)
 578 {
 579         struct vm_exit *vme;
 580         int error;
 581         int bytes, port, in;
 582
 583         vme = vmrun->vm_exit;
 584         port = vme->u.inout.port;
 585         bytes = vme->u.inout.bytes;
 586         in = vme->u.inout.in;
 587
 588         error = emulate_inout(ctx, vcpu, vme);
 589         if (error) {
 590                 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n",
 591                     in ? "in" : "out",
 592                     bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'),
 593                     port, vme->rip);
 594                 return (VMEXIT_ABORT);
 595         } else {
 596                 return (VMEXIT_CONTINUE);
 597         }
 598 }
 599
 600 static int
 601 vmexit_rdmsr(struct vmctx *ctx __unused, struct vcpu *vcpu,
 602     struct vm_run *vmrun)
 603 {
 604         struct vm_exit *vme;
 605         uint64_t val;
 606         uint32_t eax, edx;
 607         int error;
 608
 609         vme = vmrun->vm_exit;
 610
 611         val = 0;
 612         error = emulate_rdmsr(vcpu, vme->u.msr.code, &val);
 613         if (error != 0) {
 614                 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n",
 615                     vme->u.msr.code, vcpu_id(vcpu));
 616                 if (get_config_bool("x86.strictmsr")) {
 617                         vm_inject_gp(vcpu);
 618                         return (VMEXIT_CONTINUE);
 619                 }
 620         }
 621
 622         eax = val;
 623         error = vm_set_register(vcpu, VM_REG_GUEST_RAX, eax);
 624         assert(error == 0);
 625
 626         edx = val >> 32;
 627         error = vm_set_register(vcpu, VM_REG_GUEST_RDX, edx);
 628         assert(error == 0);
 629
 630         return (VMEXIT_CONTINUE);
 631 }
 632
 633 static int
 634 vmexit_wrmsr(struct vmctx *ctx __unused, struct vcpu *vcpu,
 635     struct vm_run *vmrun)
 636 {
 637         struct vm_exit *vme;
 638         int error;
 639
 640         vme = vmrun->vm_exit;
 641
 642         error = emulate_wrmsr(vcpu, vme->u.msr.code, vme->u.msr.wval);
 643         if (error != 0) {
 644                 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n",
 645                     vme->u.msr.code, vme->u.msr.wval, vcpu_id(vcpu));
 646                 if (get_config_bool("x86.strictmsr")) {
 647                         vm_inject_gp(vcpu);
 648                         return (VMEXIT_CONTINUE);
 649                 }
 650         }
 651         return (VMEXIT_CONTINUE);
 652 }
 653
 654 #define DEBUG_EPT_MISCONFIG
 655 #ifdef DEBUG_EPT_MISCONFIG
 656 #define VMCS_GUEST_PHYSICAL_ADDRESS     0x00002400
 657
 658 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4];
 659 static int ept_misconfig_ptenum;
 660 #endif
 661
 662 static const char *
 663 vmexit_vmx_desc(uint32_t exit_reason)
 664 {
 665
 666         if (exit_reason >= nitems(vmx_exit_reason_desc) ||
 667             vmx_exit_reason_desc[exit_reason] == NULL)
 668                 return ("Unknown");
 669         return (vmx_exit_reason_desc[exit_reason]);
 670 }
 671
 672 static int
 673 vmexit_vmx(struct vmctx *ctx, struct vcpu *vcpu, struct vm_run *vmrun)
 674 {
 675         struct vm_exit *vme;
 676
 677         vme = vmrun->vm_exit;
 678
 679         fprintf(stderr, "vm exit[%d]\n", vcpu_id(vcpu));
 680         fprintf(stderr, "\treason\t\tVMX\n");
 681         fprintf(stderr, "\trip\t\t0x%016lx\n", vme->rip);
 682         fprintf(stderr, "\tinst_length\t%d\n", vme->inst_length);
 683         fprintf(stderr, "\tstatus\t\t%d\n", vme->u.vmx.status);
 684         fprintf(stderr, "\texit_reason\t%u (%s)\n", vme->u.vmx.exit_reason,
 685             vmexit_vmx_desc(vme->u.vmx.exit_reason));
 686         fprintf(stderr, "\tqualification\t0x%016lx\n",
 687             vme->u.vmx.exit_qualification);
 688         fprintf(stderr, "\tinst_type\t\t%d\n", vme->u.vmx.inst_type);
 689         fprintf(stderr, "\tinst_error\t\t%d\n", vme->u.vmx.inst_error);
 690 #ifdef DEBUG_EPT_MISCONFIG
 691         if (vme->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) {
 692                 vm_get_register(vcpu,
 693                     VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS),
 694                     &ept_misconfig_gpa);
 695                 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte,
 696                     &ept_misconfig_ptenum);
 697                 fprintf(stderr, "\tEPT misconfiguration:\n");
 698                 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa);
 699                 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n",
 700                     ept_misconfig_ptenum, ept_misconfig_pte[0],
 701                     ept_misconfig_pte[1], ept_misconfig_pte[2],
 702                     ept_misconfig_pte[3]);
 703         }
 704 #endif  /* DEBUG_EPT_MISCONFIG */
 705         return (VMEXIT_ABORT);
 706 }
 707
 708 static int
 709 vmexit_svm(struct vmctx *ctx __unused, struct vcpu *vcpu, struct vm_run *vmrun)
 710 {
 711         struct vm_exit *vme;
 712
 713         vme = vmrun->vm_exit;
 714
 715         fprintf(stderr, "vm exit[%d]\n", vcpu_id(vcpu));
 716         fprintf(stderr, "\treason\t\tSVM\n");
 717         fprintf(stderr, "\trip\t\t0x%016lx\n", vme->rip);
 718         fprintf(stderr, "\tinst_length\t%d\n", vme->inst_length);
 719         fprintf(stderr, "\texitcode\t%#lx\n", vme->u.svm.exitcode);
 720         fprintf(stderr, "\texitinfo1\t%#lx\n", vme->u.svm.exitinfo1);
 721         fprintf(stderr, "\texitinfo2\t%#lx\n", vme->u.svm.exitinfo2);
 722         return (VMEXIT_ABORT);
 723 }
 724
 725 static int
 726 vmexit_bogus(struct vmctx *ctx __unused, struct vcpu *vcpu __unused,
 727     struct vm_run *vmrun)
 728 {
 729         assert(vmrun->vm_exit->inst_length == 0);
 730
 731         return (VMEXIT_CONTINUE);
 732 }
 733
 734 static int
 735 vmexit_reqidle(struct vmctx *ctx __unused, struct vcpu *vcpu __unused,
 736     struct vm_run *vmrun)
 737 {
 738         assert(vmrun->vm_exit->inst_length == 0);
 739
 740         return (VMEXIT_CONTINUE);
 741 }
 742
 743 static int
 744 vmexit_hlt(struct vmctx *ctx __unused, struct vcpu *vcpu __unused,
 745     struct vm_run *vmrun __unused)
 746 {
 747         /*
 748          * Just continue execution with the next instruction. We use
 749          * the HLT VM exit as a way to be friendly with the host
 750          * scheduler.
 751          */
 752         return (VMEXIT_CONTINUE);
 753 }
 754
 755 static int
 756 vmexit_pause(struct vmctx *ctx __unused, struct vcpu *vcpu __unused,
 757     struct vm_run *vmrun __unused)
 758 {
 759         return (VMEXIT_CONTINUE);
 760 }
 761
 762 static int
 763 vmexit_mtrap(struct vmctx *ctx __unused, struct vcpu *vcpu,
 764     struct vm_run *vmrun)
 765 {
 766         assert(vmrun->vm_exit->inst_length == 0);
 767
 768 #ifdef BHYVE_SNAPSHOT
 769         checkpoint_cpu_suspend(vcpu_id(vcpu));
 770 #endif
 771         gdb_cpu_mtrap(vcpu);
 772 #ifdef BHYVE_SNAPSHOT
 773         checkpoint_cpu_resume(vcpu_id(vcpu));
 774 #endif
 775
 776         return (VMEXIT_CONTINUE);
 777 }
 778
 779 static int
 780 vmexit_inst_emul(struct vmctx *ctx __unused, struct vcpu *vcpu,
 781     struct vm_run *vmrun)
 782 {
 783         struct vm_exit *vme;
 784         struct vie *vie;
 785         int err, i, cs_d;
 786         enum vm_cpu_mode mode;
 787
 788         vme = vmrun->vm_exit;
 789
 790         vie = &vme->u.inst_emul.vie;
 791         if (!vie->decoded) {
 792                 /*
 793                  * Attempt to decode in userspace as a fallback.  This allows
 794                  * updating instruction decode in bhyve without rebooting the
 795                  * kernel (rapid prototyping), albeit with much slower
 796                  * emulation.
 797                  */
 798                 vie_restart(vie);
 799                 mode = vme->u.inst_emul.paging.cpu_mode;
 800                 cs_d = vme->u.inst_emul.cs_d;
 801                 if (vmm_decode_instruction(mode, cs_d, vie) != 0)
 802                         goto fail;
 803                 if (vm_set_register(vcpu, VM_REG_GUEST_RIP,
 804                     vme->rip + vie->num_processed) != 0)
 805                         goto fail;
 806         }
 807
 808         err = emulate_mem(vcpu, vme->u.inst_emul.gpa, vie,
 809             &vme->u.inst_emul.paging);
 810         if (err) {
 811                 if (err == ESRCH) {
 812                         EPRINTLN("Unhandled memory access to 0x%lx\n",
 813                             vme->u.inst_emul.gpa);
 814                 }
 815                 goto fail;
 816         }
 817
 818         return (VMEXIT_CONTINUE);
 819
 820 fail:
 821         fprintf(stderr, "Failed to emulate instruction sequence [ ");
 822         for (i = 0; i < vie->num_valid; i++)
 823                 fprintf(stderr, "%02x", vie->inst[i]);
 824         FPRINTLN(stderr, " ] at 0x%lx", vme->rip);
 825         return (VMEXIT_ABORT);
 826 }
 827
 828 static int
 829 vmexit_suspend(struct vmctx *ctx, struct vcpu *vcpu, struct vm_run *vmrun)
 830 {
 831         struct vm_exit *vme;
 832         enum vm_suspend_how how;
 833         int vcpuid = vcpu_id(vcpu);
 834
 835         vme = vmrun->vm_exit;
 836
 837         how = vme->u.suspended.how;
 838
 839         fbsdrun_deletecpu(vcpuid);
 840
 841         switch (how) {
 842         case VM_SUSPEND_RESET:
 843                 exit(0);
 844         case VM_SUSPEND_POWEROFF:
 845                 if (get_config_bool_default("destroy_on_poweroff", false))
 846                         vm_destroy(ctx);
 847                 exit(1);
 848         case VM_SUSPEND_HALT:
 849                 exit(2);
 850         case VM_SUSPEND_TRIPLEFAULT:
 851                 exit(3);
 852         default:
 853                 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how);
 854                 exit(100);
 855         }
 856         return (0);     /* NOTREACHED */
 857 }
 858
 859 static int
 860 vmexit_debug(struct vmctx *ctx __unused, struct vcpu *vcpu,
 861     struct vm_run *vmrun __unused)
 862 {
 863
 864 #ifdef BHYVE_SNAPSHOT
 865         checkpoint_cpu_suspend(vcpu_id(vcpu));
 866 #endif
 867         gdb_cpu_suspend(vcpu);
 868 #ifdef BHYVE_SNAPSHOT
 869         checkpoint_cpu_resume(vcpu_id(vcpu));
 870 #endif
 871         /*
 872          * XXX-MJ sleep for a short period to avoid chewing up the CPU in the
 873          * window between activation of the vCPU thread and the STARTUP IPI.
 874          */
 875         usleep(1000);
 876         return (VMEXIT_CONTINUE);
 877 }
 878
 879 static int
 880 vmexit_breakpoint(struct vmctx *ctx __unused, struct vcpu *vcpu,
 881     struct vm_run *vmrun)
 882 {
 883         gdb_cpu_breakpoint(vcpu, vmrun->vm_exit);
 884         return (VMEXIT_CONTINUE);
 885 }
 886
 887 static int
 888 vmexit_ipi(struct vmctx *ctx __unused, struct vcpu *vcpu __unused,
 889     struct vm_run *vmrun)
 890 {
 891         struct vm_exit *vme;
 892         cpuset_t *dmask;
 893         int error = -1;
 894         int i;
 895
 896         dmask = vmrun->cpuset;
 897         vme = vmrun->vm_exit;
 898
 899         switch (vme->u.ipi.mode) {
 900         case APIC_DELMODE_INIT:
 901                 CPU_FOREACH_ISSET(i, dmask) {
 902                         error = vm_suspend_cpu(vcpu_info[i].vcpu);
 903                         if (error) {
 904                                 warnx("%s: failed to suspend cpu %d\n",
 905                                     __func__, i);
 906                                 break;
 907                         }
 908                 }
 909                 break;
 910         case APIC_DELMODE_STARTUP:
 911                 CPU_FOREACH_ISSET(i, dmask) {
 912                         spinup_ap(vcpu_info[i].vcpu,
 913                             vme->u.ipi.vector << PAGE_SHIFT);
 914                 }
 915                 error = 0;
 916                 break;
 917         default:
 918                 break;
 919         }
 920
 921         return (error);
 922 }
 923
 924 static const vmexit_handler_t handler[VM_EXITCODE_MAX] = {
 925         [VM_EXITCODE_INOUT]  = vmexit_inout,
 926         [VM_EXITCODE_INOUT_STR]  = vmexit_inout,
 927         [VM_EXITCODE_VMX]    = vmexit_vmx,
 928         [VM_EXITCODE_SVM]    = vmexit_svm,
 929         [VM_EXITCODE_BOGUS]  = vmexit_bogus,
 930         [VM_EXITCODE_REQIDLE] = vmexit_reqidle,
 931         [VM_EXITCODE_RDMSR]  = vmexit_rdmsr,
 932         [VM_EXITCODE_WRMSR]  = vmexit_wrmsr,
 933         [VM_EXITCODE_MTRAP]  = vmexit_mtrap,
 934         [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul,
 935         [VM_EXITCODE_SUSPENDED] = vmexit_suspend,
 936         [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch,
 937         [VM_EXITCODE_DEBUG] = vmexit_debug,
 938         [VM_EXITCODE_BPT] = vmexit_breakpoint,
 939         [VM_EXITCODE_IPI] = vmexit_ipi,
 940         [VM_EXITCODE_HLT] = vmexit_hlt,
 941         [VM_EXITCODE_PAUSE] = vmexit_pause,
 942 };
 943
 944 static void
 945 vm_loop(struct vmctx *ctx, struct vcpu *vcpu)
 946 {
 947         struct vm_exit vme;
 948         struct vm_run vmrun;
 949         int error, rc;
 950         enum vm_exitcode exitcode;
 951         cpuset_t active_cpus, dmask;
 952
 953         error = vm_active_cpus(ctx, &active_cpus);
 954         assert(CPU_ISSET(vcpu_id(vcpu), &active_cpus));
 955
 956         vmrun.vm_exit = &vme;
 957         vmrun.cpuset = &dmask;
 958         vmrun.cpusetsize = sizeof(dmask);
 959
 960         while (1) {
 961                 error = vm_run(vcpu, &vmrun);
 962                 if (error != 0)
 963                         break;
 964
 965                 exitcode = vme.exitcode;
 966                 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) {
 967                         fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n",
 968                             exitcode);
 969                         exit(4);
 970                 }
 971
 972                 rc = (*handler[exitcode])(ctx, vcpu, &vmrun);
 973
 974                 switch (rc) {
 975                 case VMEXIT_CONTINUE:
 976                         break;
 977                 case VMEXIT_ABORT:
 978                         abort();
 979                 default:
 980                         exit(4);
 981                 }
 982         }
 983         fprintf(stderr, "vm_run error %d, errno %d\n", error, errno);
 984 }
 985
 986 static int
 987 num_vcpus_allowed(struct vmctx *ctx, struct vcpu *vcpu)
 988 {
 989         uint16_t sockets, cores, threads, maxcpus;
 990         int tmp, error;
 991
 992         /*
 993          * The guest is allowed to spinup more than one processor only if the
 994          * UNRESTRICTED_GUEST capability is available.
 995          */
 996         error = vm_get_capability(vcpu, VM_CAP_UNRESTRICTED_GUEST, &tmp);
 997         if (error != 0)
 998                 return (1);
 999
1000         error = vm_get_topology(ctx, &sockets, &cores, &threads, &maxcpus);
1001         if (error == 0)
1002                 return (maxcpus);
1003         else
1004                 return (1);
1005 }
1006
1007 static void
1008 fbsdrun_set_capabilities(struct vcpu *vcpu)
1009 {
1010         int err, tmp;
1011
1012         if (get_config_bool_default("x86.vmexit_on_hlt", false)) {
1013                 err = vm_get_capability(vcpu, VM_CAP_HALT_EXIT, &tmp);
1014                 if (err < 0) {
1015                         fprintf(stderr, "VM exit on HLT not supported\n");
1016                         exit(4);
1017                 }
1018                 vm_set_capability(vcpu, VM_CAP_HALT_EXIT, 1);
1019         }
1020
1021         if (get_config_bool_default("x86.vmexit_on_pause", false)) {
1022                 /*
1023                  * pause exit support required for this mode
1024                  */
1025                 err = vm_get_capability(vcpu, VM_CAP_PAUSE_EXIT, &tmp);
1026                 if (err < 0) {
1027                         fprintf(stderr,
1028                             "SMP mux requested, no pause support\n");
1029                         exit(4);
1030                 }
1031                 vm_set_capability(vcpu, VM_CAP_PAUSE_EXIT, 1);
1032         }
1033
1034         if (get_config_bool_default("x86.x2apic", false))
1035                 err = vm_set_x2apic_state(vcpu, X2APIC_ENABLED);
1036         else
1037                 err = vm_set_x2apic_state(vcpu, X2APIC_DISABLED);
1038
1039         if (err) {
1040                 fprintf(stderr, "Unable to set x2apic state (%d)\n", err);
1041                 exit(4);
1042         }
1043
1044         vm_set_capability(vcpu, VM_CAP_ENABLE_INVPCID, 1);
1045
1046         err = vm_set_capability(vcpu, VM_CAP_IPI_EXIT, 1);
1047         assert(err == 0);
1048 }
1049
1050 static struct vmctx *
1051 do_open(const char *vmname)
1052 {
1053         struct vmctx *ctx;
1054         int error;
1055         bool reinit, romboot;
1056
1057         reinit = romboot = false;
1058
1059         if (lpc_bootrom())
1060                 romboot = true;
1061
1062         error = vm_create(vmname);
1063         if (error) {
1064                 if (errno == EEXIST) {
1065                         if (romboot) {
1066                                 reinit = true;
1067                         } else {
1068                                 /*
1069                                  * The virtual machine has been setup by the
1070                                  * userspace bootloader.
1071                                  */
1072                         }
1073                 } else {
1074                         perror("vm_create");
1075                         exit(4);
1076                 }
1077         } else {
1078                 if (!romboot) {
1079                         /*
1080                          * If the virtual machine was just created then a
1081                          * bootrom must be configured to boot it.
1082                          */
1083                         fprintf(stderr, "virtual machine cannot be booted\n");
1084                         exit(4);
1085                 }
1086         }
1087
1088         ctx = vm_open(vmname);
1089         if (ctx == NULL) {
1090                 perror("vm_open");
1091                 exit(4);
1092         }
1093
1094 #ifndef WITHOUT_CAPSICUM
1095         if (vm_limit_rights(ctx) != 0)
1096                 err(EX_OSERR, "vm_limit_rights");
1097 #endif
1098
1099         if (reinit) {
1100                 error = vm_reinit(ctx);
1101                 if (error) {
1102                         perror("vm_reinit");
1103                         exit(4);
1104                 }
1105         }
1106         error = vm_set_topology(ctx, cpu_sockets, cpu_cores, cpu_threads, 0);
1107         if (error)
1108                 errx(EX_OSERR, "vm_set_topology");
1109         return (ctx);
1110 }
1111
1112 static void
1113 spinup_vcpu(struct vcpu_info *vi, bool bsp)
1114 {
1115         int error;
1116
1117         if (!bsp) {
1118                 fbsdrun_set_capabilities(vi->vcpu);
1119
1120                 /*
1121                  * Enable the 'unrestricted guest' mode for APs.
1122                  *
1123                  * APs startup in power-on 16-bit mode.
1124                  */
1125                 error = vm_set_capability(vi->vcpu, VM_CAP_UNRESTRICTED_GUEST, 1);
1126                 assert(error == 0);
1127         }
1128
1129         fbsdrun_addcpu(vi);
1130 }
1131
1132 static bool
1133 parse_config_option(const char *option)
1134 {
1135         const char *value;
1136         char *path;
1137
1138         value = strchr(option, '=');
1139         if (value == NULL || value[1] == '\0')
1140                 return (false);
1141         path = strndup(option, value - option);
1142         if (path == NULL)
1143                 err(4, "Failed to allocate memory");
1144         set_config_value(path, value + 1);
1145         return (true);
1146 }
1147
1148 static void
1149 parse_simple_config_file(const char *path)
1150 {
1151         FILE *fp;
1152         char *line, *cp;
1153         size_t linecap;
1154         unsigned int lineno;
1155
1156         fp = fopen(path, "r");
1157         if (fp == NULL)
1158                 err(4, "Failed to open configuration file %s", path);
1159         line = NULL;
1160         linecap = 0;
1161         lineno = 1;
1162         for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) {
1163                 if (*line == '#' || *line == '\n')
1164                         continue;
1165                 cp = strchr(line, '\n');
1166                 if (cp != NULL)
1167                         *cp = '\0';
1168                 if (!parse_config_option(line))
1169                         errx(4, "%s line %u: invalid config option '%s'", path,
1170                             lineno, line);
1171         }
1172         free(line);
1173         fclose(fp);
1174 }
1175
1176 static void
1177 parse_gdb_options(const char *opt)
1178 {
1179         const char *sport;
1180         char *colon;
1181
1182         if (opt[0] == 'w') {
1183                 set_config_bool("gdb.wait", true);
1184                 opt++;
1185         }
1186
1187         colon = strrchr(opt, ':');
1188         if (colon == NULL) {
1189                 sport = opt;
1190         } else {
1191                 *colon = '\0';
1192                 colon++;
1193                 sport = colon;
1194                 set_config_value("gdb.address", opt);
1195         }
1196
1197         set_config_value("gdb.port", sport);
1198 }
1199
1200 static void
1201 set_defaults(void)
1202 {
1203
1204         set_config_bool("acpi_tables", false);
1205         set_config_bool("acpi_tables_in_memory", true);
1206         set_config_value("memory.size", "256M");
1207         set_config_bool("x86.strictmsr", true);
1208         set_config_value("lpc.fwcfg", "bhyve");
1209 }
1210
1211 int
1212 main(int argc, char *argv[])
1213 {
1214         int c, error;
1215         int max_vcpus, memflags;
1216         struct vcpu *bsp;
1217         struct vmctx *ctx;
1218         struct qemu_fwcfg_item *e820_fwcfg_item;
1219         size_t memsize;
1220         const char *optstr, *value, *vmname;
1221 #ifdef BHYVE_SNAPSHOT
1222         char *restore_file;
1223         struct restore_state rstate;
1224
1225         restore_file = NULL;
1226 #endif
1227
1228         init_config();
1229         set_defaults();
1230         progname = basename(argv[0]);
1231
1232 #ifdef BHYVE_SNAPSHOT
1233         optstr = "aehuwxACDHIPSWYk:f:o:p:G:c:s:m:l:K:U:r:";
1234 #else
1235         optstr = "aehuwxACDHIPSWYk:f:o:p:G:c:s:m:l:K:U:";
1236 #endif
1237         while ((c = getopt(argc, argv, optstr)) != -1) {
1238                 switch (c) {
1239                 case 'a':
1240                         set_config_bool("x86.x2apic", false);
1241                         break;
1242                 case 'A':
1243                         set_config_bool("acpi_tables", true);
1244                         break;
1245                 case 'D':
1246                         set_config_bool("destroy_on_poweroff", true);
1247                         break;
1248                 case 'p':
1249                         if (pincpu_parse(optarg) != 0) {
1250                                 errx(EX_USAGE, "invalid vcpu pinning "
1251                                     "configuration '%s'", optarg);
1252                         }
1253                         break;
1254                 case 'c':
1255                         if (topology_parse(optarg) != 0) {
1256                             errx(EX_USAGE, "invalid cpu topology "
1257                                 "'%s'", optarg);
1258                         }
1259                         break;
1260                 case 'C':
1261                         set_config_bool("memory.guest_in_core", true);
1262                         break;
1263                 case 'f':
1264                         if (qemu_fwcfg_parse_cmdline_arg(optarg) != 0) {
1265                             errx(EX_USAGE, "invalid fwcfg item '%s'", optarg);
1266                         }
1267                         break;
1268                 case 'G':
1269                         parse_gdb_options(optarg);
1270                         break;
1271                 case 'k':
1272                         parse_simple_config_file(optarg);
1273                         break;
1274                 case 'K':
1275                         set_config_value("keyboard.layout", optarg);
1276                         break;
1277                 case 'l':
1278                         if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1279                                 lpc_print_supported_devices();
1280                                 exit(0);
1281                         } else if (lpc_device_parse(optarg) != 0) {
1282                                 errx(EX_USAGE, "invalid lpc device "
1283                                     "configuration '%s'", optarg);
1284                         }
1285                         break;
1286 #ifdef BHYVE_SNAPSHOT
1287                 case 'r':
1288                         restore_file = optarg;
1289                         break;
1290 #endif
1291                 case 's':
1292                         if (strncmp(optarg, "help", strlen(optarg)) == 0) {
1293                                 pci_print_supported_devices();
1294                                 exit(0);
1295                         } else if (pci_parse_slot(optarg) != 0)
1296                                 exit(4);
1297                         else
1298                                 break;
1299                 case 'S':
1300                         set_config_bool("memory.wired", true);
1301                         break;
1302                 case 'm':
1303                         set_config_value("memory.size", optarg);
1304                         break;
1305                 case 'o':
1306                         if (!parse_config_option(optarg))
1307                                 errx(EX_USAGE, "invalid configuration option '%s'", optarg);
1308                         break;
1309                 case 'H':
1310                         set_config_bool("x86.vmexit_on_hlt", true);
1311                         break;
1312                 case 'I':
1313                         /*
1314                          * The "-I" option was used to add an ioapic to the
1315                          * virtual machine.
1316                          *
1317                          * An ioapic is now provided unconditionally for each
1318                          * virtual machine and this option is now deprecated.
1319                          */
1320                         break;
1321                 case 'P':
1322                         set_config_bool("x86.vmexit_on_pause", true);
1323                         break;
1324                 case 'e':
1325                         set_config_bool("x86.strictio", true);
1326                         break;
1327                 case 'u':
1328                         set_config_bool("rtc.use_localtime", false);
1329                         break;
1330                 case 'U':
1331                         set_config_value("uuid", optarg);
1332                         break;
1333                 case 'w':
1334                         set_config_bool("x86.strictmsr", false);
1335                         break;
1336                 case 'W':
1337                         set_config_bool("virtio_msix", false);
1338                         break;
1339                 case 'x':
1340                         set_config_bool("x86.x2apic", true);
1341                         break;
1342                 case 'Y':
1343                         set_config_bool("x86.mptable", false);
1344                         break;
1345                 case 'h':
1346                         usage(0);
1347                 default:
1348                         usage(1);
1349                 }
1350         }
1351         argc -= optind;
1352         argv += optind;
1353
1354         if (argc > 1)
1355                 usage(1);
1356
1357 #ifdef BHYVE_SNAPSHOT
1358         if (restore_file != NULL) {
1359                 error = load_restore_file(restore_file, &rstate);
1360                 if (error) {
1361                         fprintf(stderr, "Failed to read checkpoint info from "
1362                                         "file: '%s'.\n", restore_file);
1363                         exit(1);
1364                 }
1365                 vmname = lookup_vmname(&rstate);
1366                 if (vmname != NULL)
1367                         set_config_value("name", vmname);
1368         }
1369 #endif
1370
1371         if (argc == 1)
1372                 set_config_value("name", argv[0]);
1373
1374         vmname = get_config_value("name");
1375         if (vmname == NULL)
1376                 usage(1);
1377
1378         if (get_config_bool_default("config.dump", false)) {
1379                 dump_config();
1380                 exit(1);
1381         }
1382
1383         calc_topology();
1384         build_vcpumaps();
1385
1386         value = get_config_value("memory.size");
1387         error = vm_parse_memsize(value, &memsize);
1388         if (error)
1389                 errx(EX_USAGE, "invalid memsize '%s'", value);
1390
1391         ctx = do_open(vmname);
1392
1393 #ifdef BHYVE_SNAPSHOT
1394         if (restore_file != NULL) {
1395                 guest_ncpus = lookup_guest_ncpus(&rstate);
1396                 memflags = lookup_memflags(&rstate);
1397                 memsize = lookup_memsize(&rstate);
1398         }
1399
1400         if (guest_ncpus < 1) {
1401                 fprintf(stderr, "Invalid guest vCPUs (%d)\n", guest_ncpus);
1402                 exit(1);
1403         }
1404 #endif
1405
1406         bsp = vm_vcpu_open(ctx, BSP);
1407         max_vcpus = num_vcpus_allowed(ctx, bsp);
1408         if (guest_ncpus > max_vcpus) {
1409                 fprintf(stderr, "%d vCPUs requested but only %d available\n",
1410                         guest_ncpus, max_vcpus);
1411                 exit(4);
1412         }
1413
1414         fbsdrun_set_capabilities(bsp);
1415
1416         /* Allocate per-VCPU resources. */
1417         vcpu_info = calloc(guest_ncpus, sizeof(*vcpu_info));
1418         for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++) {
1419                 vcpu_info[vcpuid].ctx = ctx;
1420                 vcpu_info[vcpuid].vcpuid = vcpuid;
1421                 if (vcpuid == BSP)
1422                         vcpu_info[vcpuid].vcpu = bsp;
1423                 else
1424                         vcpu_info[vcpuid].vcpu = vm_vcpu_open(ctx, vcpuid);
1425         }
1426
1427         memflags = 0;
1428         if (get_config_bool_default("memory.wired", false))
1429                 memflags |= VM_MEM_F_WIRED;
1430         if (get_config_bool_default("memory.guest_in_core", false))
1431                 memflags |= VM_MEM_F_INCORE;
1432         vm_set_memflags(ctx, memflags);
1433         error = vm_setup_memory(ctx, memsize, VM_MMAP_ALL);
1434         if (error) {
1435                 fprintf(stderr, "Unable to setup memory (%d)\n", errno);
1436                 exit(4);
1437         }
1438
1439         error = init_msr();
1440         if (error) {
1441                 fprintf(stderr, "init_msr error %d", error);
1442                 exit(4);
1443         }
1444
1445         init_mem(guest_ncpus);
1446         init_inout();
1447         kernemu_dev_init();
1448         init_bootrom(ctx);
1449         atkbdc_init(ctx);
1450         pci_irq_init(ctx);
1451         ioapic_init(ctx);
1452
1453         rtc_init(ctx);
1454         sci_init(ctx);
1455
1456         if (qemu_fwcfg_init(ctx) != 0) {
1457                 fprintf(stderr, "qemu fwcfg initialization error");
1458                 exit(4);
1459         }
1460
1461         if (qemu_fwcfg_add_file("opt/bhyve/hw.ncpu", sizeof(guest_ncpus),
1462             &guest_ncpus) != 0) {
1463                 fprintf(stderr, "Could not add qemu fwcfg opt/bhyve/hw.ncpu");
1464                 exit(4);
1465         }
1466
1467         if (e820_init(ctx) != 0) {
1468                 fprintf(stderr, "Unable to setup E820");
1469                 exit(4);
1470         }
1471
1472         /*
1473          * Exit if a device emulation finds an error in its initialization
1474          */
1475         if (init_pci(ctx) != 0) {
1476                 perror("device emulation initialization error");
1477                 exit(4);
1478         }
1479         if (init_tpm(ctx) != 0) {
1480                 fprintf(stderr, "Failed to init TPM device");
1481                 exit(4);
1482         }
1483
1484         /*
1485          * Initialize after PCI, to allow a bootrom file to reserve the high
1486          * region.
1487          */
1488         if (get_config_bool("acpi_tables"))
1489                 vmgenc_init(ctx);
1490
1491         init_gdb(ctx);
1492
1493         if (lpc_bootrom()) {
1494                 if (vm_set_capability(bsp, VM_CAP_UNRESTRICTED_GUEST, 1)) {
1495                         fprintf(stderr, "ROM boot failed: unrestricted guest "
1496                             "capability not available\n");
1497                         exit(4);
1498                 }
1499                 error = vcpu_reset(bsp);
1500                 assert(error == 0);
1501         }
1502
1503         /*
1504          * Add all vCPUs.
1505          */
1506         for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++)
1507                 spinup_vcpu(&vcpu_info[vcpuid], vcpuid == BSP);
1508
1509 #ifdef BHYVE_SNAPSHOT
1510         if (restore_file != NULL) {
1511                 fprintf(stdout, "Pausing pci devs...\r\n");
1512                 if (vm_pause_devices() != 0) {
1513                         fprintf(stderr, "Failed to pause PCI device state.\n");
1514                         exit(1);
1515                 }
1516
1517                 fprintf(stdout, "Restoring vm mem...\r\n");
1518                 if (restore_vm_mem(ctx, &rstate) != 0) {
1519                         fprintf(stderr, "Failed to restore VM memory.\n");
1520                         exit(1);
1521                 }
1522
1523                 fprintf(stdout, "Restoring pci devs...\r\n");
1524                 if (vm_restore_devices(&rstate) != 0) {
1525                         fprintf(stderr, "Failed to restore PCI device state.\n");
1526                         exit(1);
1527                 }
1528
1529                 fprintf(stdout, "Restoring kernel structs...\r\n");
1530                 if (vm_restore_kern_structs(ctx, &rstate) != 0) {
1531                         fprintf(stderr, "Failed to restore kernel structs.\n");
1532                         exit(1);
1533                 }
1534
1535                 fprintf(stdout, "Resuming pci devs...\r\n");
1536                 if (vm_resume_devices() != 0) {
1537                         fprintf(stderr, "Failed to resume PCI device state.\n");
1538                         exit(1);
1539                 }
1540         }
1541 #endif
1542
1543         /*
1544          * build the guest tables, MP etc.
1545          */
1546         if (get_config_bool_default("x86.mptable", true)) {
1547                 error = mptable_build(ctx, guest_ncpus);
1548                 if (error) {
1549                         perror("error to build the guest tables");
1550                         exit(4);
1551                 }
1552         }
1553
1554         error = smbios_build(ctx);
1555         if (error != 0)
1556                 exit(4);
1557
1558         if (get_config_bool("acpi_tables")) {
1559                 error = acpi_build(ctx, guest_ncpus);
1560                 assert(error == 0);
1561         }
1562
1563         e820_fwcfg_item = e820_get_fwcfg_item();
1564         if (e820_fwcfg_item == NULL) {
1565             fprintf(stderr, "invalid e820 table");
1566                 exit(4);
1567         }
1568         if (qemu_fwcfg_add_file("etc/e820", e820_fwcfg_item->size,
1569                 e820_fwcfg_item->data) != 0) {
1570                 fprintf(stderr, "could not add qemu fwcfg etc/e820");
1571                 exit(4);
1572         }
1573         free(e820_fwcfg_item);
1574
1575         if (lpc_bootrom() && strcmp(lpc_fwcfg(), "bhyve") == 0) {
1576                 fwctl_init();
1577         }
1578
1579         /*
1580          * Change the proc title to include the VM name.
1581          */
1582         setproctitle("%s", vmname);
1583
1584 #ifdef BHYVE_SNAPSHOT
1585         /* initialize mutex/cond variables */
1586         init_snapshot();
1587
1588         /*
1589          * checkpointing thread for communication with bhyvectl
1590          */
1591         if (init_checkpoint_thread(ctx) != 0)
1592                 errx(EX_OSERR, "Failed to start checkpoint thread");
1593 #endif
1594
1595 #ifndef WITHOUT_CAPSICUM
1596         caph_cache_catpages();
1597
1598         if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1)
1599                 errx(EX_OSERR, "Unable to apply rights for sandbox");
1600
1601         if (caph_enter() == -1)
1602                 errx(EX_OSERR, "cap_enter() failed");
1603 #endif
1604
1605 #ifdef BHYVE_SNAPSHOT
1606         if (restore_file != NULL) {
1607                 destroy_restore_state(&rstate);
1608                 if (vm_restore_time(ctx) < 0)
1609                         err(EX_OSERR, "Unable to restore time");
1610
1611                 for (int vcpuid = 0; vcpuid < guest_ncpus; vcpuid++)
1612                         vm_resume_cpu(vcpu_info[vcpuid].vcpu);
1613         } else
1614 #endif
1615                 vm_resume_cpu(bsp);
1616
1617         /*
1618          * Head off to the main event dispatch loop
1619          */
1620         mevent_dispatch();
1621
1622         exit(4);
1623 }