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 * This file and its contents are supplied under the terms of the 32 * Common Development and Distribution License ("CDDL"), version 1.0. 33 * You may only use this file in accordance with the terms of version 34 * 1.0 of the CDDL. 35 * 36 * A full copy of the text of the CDDL should have accompanied this 37 * source. A copy of the CDDL is also available via the Internet at 38 * http://www.illumos.org/license/CDDL. 39 * 40 * Copyright 2015 Pluribus Networks Inc. 41 * Copyright 2018 Joyent, Inc. 42 * Copyright 2020 Oxide Computer Company 43 */ 44 45 #include <sys/cdefs.h> 46 __FBSDID("$FreeBSD$"); 47 48 #include <sys/types.h> 49 #ifndef WITHOUT_CAPSICUM 50 #include <sys/capsicum.h> 51 #endif 52 #include <sys/mman.h> 53 #include <sys/time.h> 54 #include <sys/cpuset.h> 55 56 #ifdef __FreeBSD__ 57 #include <amd64/vmm/intel/vmcs.h> 58 #else 59 #include <intel/vmcs.h> 60 #endif 61 62 #include <machine/atomic.h> 63 #include <machine/segments.h> 64 65 #ifndef WITHOUT_CAPSICUM 66 #include <capsicum_helpers.h> 67 #endif 68 #include <stdio.h> 69 #include <stdlib.h> 70 #include <string.h> 71 #include <err.h> 72 #include <errno.h> 73 #include <libgen.h> 74 #include <unistd.h> 75 #include <assert.h> 76 #include <pthread.h> 77 #include <pthread_np.h> 78 #include <sysexits.h> 79 #include <stdbool.h> 80 #include <stdint.h> 81 82 #include <machine/vmm.h> 83 #ifndef WITHOUT_CAPSICUM 84 #include <machine/vmm_dev.h> 85 #endif 86 #include <vmmapi.h> 87 88 #include "bhyverun.h" 89 #include "acpi.h" 90 #include "atkbdc.h" 91 #include "console.h" 92 #include "bootrom.h" 93 #include "config.h" 94 #include "inout.h" 95 #include "debug.h" 96 #include "fwctl.h" 97 #include "gdb.h" 98 #include "ioapic.h" 99 #include "kernemu_dev.h" 100 #include "mem.h" 101 #include "mevent.h" 102 #include "mptbl.h" 103 #include "pci_emul.h" 104 #include "pci_irq.h" 105 #include "pci_lpc.h" 106 #include "smbiostbl.h" 107 #include "xmsr.h" 108 #include "spinup_ap.h" 109 #include "rfb.h" 110 #include "rtc.h" 111 #include "vga.h" 112 #include "vmgenc.h" 113 114 #define GUEST_NIO_PORT 0x488 /* guest upcalls via i/o port */ 115 116 #define MB (1024UL * 1024) 117 #define GB (1024UL * MB) 118 119 static const char * const vmx_exit_reason_desc[] = { 120 [EXIT_REASON_EXCEPTION] = "Exception or non-maskable interrupt (NMI)", 121 [EXIT_REASON_EXT_INTR] = "External interrupt", 122 [EXIT_REASON_TRIPLE_FAULT] = "Triple fault", 123 [EXIT_REASON_INIT] = "INIT signal", 124 [EXIT_REASON_SIPI] = "Start-up IPI (SIPI)", 125 [EXIT_REASON_IO_SMI] = "I/O system-management interrupt (SMI)", 126 [EXIT_REASON_SMI] = "Other SMI", 127 [EXIT_REASON_INTR_WINDOW] = "Interrupt window", 128 [EXIT_REASON_NMI_WINDOW] = "NMI window", 129 [EXIT_REASON_TASK_SWITCH] = "Task switch", 130 [EXIT_REASON_CPUID] = "CPUID", 131 [EXIT_REASON_GETSEC] = "GETSEC", 132 [EXIT_REASON_HLT] = "HLT", 133 [EXIT_REASON_INVD] = "INVD", 134 [EXIT_REASON_INVLPG] = "INVLPG", 135 [EXIT_REASON_RDPMC] = "RDPMC", 136 [EXIT_REASON_RDTSC] = "RDTSC", 137 [EXIT_REASON_RSM] = "RSM", 138 [EXIT_REASON_VMCALL] = "VMCALL", 139 [EXIT_REASON_VMCLEAR] = "VMCLEAR", 140 [EXIT_REASON_VMLAUNCH] = "VMLAUNCH", 141 [EXIT_REASON_VMPTRLD] = "VMPTRLD", 142 [EXIT_REASON_VMPTRST] = "VMPTRST", 143 [EXIT_REASON_VMREAD] = "VMREAD", 144 [EXIT_REASON_VMRESUME] = "VMRESUME", 145 [EXIT_REASON_VMWRITE] = "VMWRITE", 146 [EXIT_REASON_VMXOFF] = "VMXOFF", 147 [EXIT_REASON_VMXON] = "VMXON", 148 [EXIT_REASON_CR_ACCESS] = "Control-register accesses", 149 [EXIT_REASON_DR_ACCESS] = "MOV DR", 150 [EXIT_REASON_INOUT] = "I/O instruction", 151 [EXIT_REASON_RDMSR] = "RDMSR", 152 [EXIT_REASON_WRMSR] = "WRMSR", 153 [EXIT_REASON_INVAL_VMCS] = 154 "VM-entry failure due to invalid guest state", 155 [EXIT_REASON_INVAL_MSR] = "VM-entry failure due to MSR loading", 156 [EXIT_REASON_MWAIT] = "MWAIT", 157 [EXIT_REASON_MTF] = "Monitor trap flag", 158 [EXIT_REASON_MONITOR] = "MONITOR", 159 [EXIT_REASON_PAUSE] = "PAUSE", 160 [EXIT_REASON_MCE_DURING_ENTRY] = 161 "VM-entry failure due to machine-check event", 162 [EXIT_REASON_TPR] = "TPR below threshold", 163 [EXIT_REASON_APIC_ACCESS] = "APIC access", 164 [EXIT_REASON_VIRTUALIZED_EOI] = "Virtualized EOI", 165 [EXIT_REASON_GDTR_IDTR] = "Access to GDTR or IDTR", 166 [EXIT_REASON_LDTR_TR] = "Access to LDTR or TR", 167 [EXIT_REASON_EPT_FAULT] = "EPT violation", 168 [EXIT_REASON_EPT_MISCONFIG] = "EPT misconfiguration", 169 [EXIT_REASON_INVEPT] = "INVEPT", 170 [EXIT_REASON_RDTSCP] = "RDTSCP", 171 [EXIT_REASON_VMX_PREEMPT] = "VMX-preemption timer expired", 172 [EXIT_REASON_INVVPID] = "INVVPID", 173 [EXIT_REASON_WBINVD] = "WBINVD", 174 [EXIT_REASON_XSETBV] = "XSETBV", 175 [EXIT_REASON_APIC_WRITE] = "APIC write", 176 [EXIT_REASON_RDRAND] = "RDRAND", 177 [EXIT_REASON_INVPCID] = "INVPCID", 178 [EXIT_REASON_VMFUNC] = "VMFUNC", 179 [EXIT_REASON_ENCLS] = "ENCLS", 180 [EXIT_REASON_RDSEED] = "RDSEED", 181 [EXIT_REASON_PM_LOG_FULL] = "Page-modification log full", 182 [EXIT_REASON_XSAVES] = "XSAVES", 183 [EXIT_REASON_XRSTORS] = "XRSTORS" 184 }; 185 186 typedef int (*vmexit_handler_t)(struct vmctx *, struct vm_exit *, int *vcpu); 187 extern int vmexit_task_switch(struct vmctx *, struct vm_exit *, int *vcpu); 188 189 int guest_ncpus; 190 uint16_t cores, maxcpus, sockets, threads; 191 192 int raw_stdio = 0; 193 194 static char *progname; 195 static const int BSP = 0; 196 197 static cpuset_t cpumask; 198 199 static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip); 200 201 static struct vm_exit vmexit[VM_MAXCPU]; 202 static struct vm_entry vmentry[VM_MAXCPU]; 203 204 struct bhyvestats { 205 uint64_t vmexit_bogus; 206 uint64_t vmexit_reqidle; 207 uint64_t vmexit_hlt; 208 uint64_t vmexit_pause; 209 uint64_t vmexit_mtrap; 210 uint64_t vmexit_mmio; 211 uint64_t vmexit_inout; 212 uint64_t cpu_switch_rotate; 213 uint64_t cpu_switch_direct; 214 uint64_t mmio_unhandled; 215 } stats; 216 217 struct mt_vmm_info { 218 pthread_t mt_thr; 219 struct vmctx *mt_ctx; 220 int mt_vcpu; 221 uint64_t mt_startrip; 222 } mt_vmm_info[VM_MAXCPU]; 223 224 #ifdef __FreeBSD__ 225 static cpuset_t *vcpumap[VM_MAXCPU] = { NULL }; 226 #endif 227 228 static void 229 usage(int code) 230 { 231 232 fprintf(stderr, 233 #ifdef __FreeBSD__ 234 "Usage: %s [-aehuwxACDHPSWY]\n" 235 #else 236 "Usage: %s [-adehuwxACDHPSWY]\n" 237 #endif 238 " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n" 239 " %*s [-k <file>] [-l <lpc>] [-m mem] [-o <var>=<value>]\n" 240 #ifdef __FreeBSD__ 241 " %*s [-p vcpu:hostcpu] [-s <pci>] [-U uuid] [<vm>]\n" 242 #else 243 " %*s [-s <pci>] [-U uuid] [<vm>]\n" 244 #endif 245 " -a: local apic is in xAPIC mode (deprecated)\n" 246 " -A: create ACPI tables\n" 247 " -c: number of cpus and/or topology specification\n" 248 " -C: include guest memory in core file\n" 249 #ifndef __FreeBSD__ 250 " -d: suspend cpu at boot\n" 251 #endif 252 " -D: destroy on power-off\n" 253 " -e: exit on unhandled I/O access\n" 254 " -h: help\n" 255 " -H: vmexit from the guest on hlt\n" 256 " -k: key=value flat config file\n" 257 " -l: LPC device configuration\n" 258 " -m: memory size\n" 259 " -o: set config 'var' to 'value'\n" 260 #ifdef __FreeBSD__ 261 " -p: pin 'vcpu' to 'hostcpu'\n" 262 #endif 263 " -P: vmexit from the guest on pause\n" 264 " -s: <slot,driver,configinfo> PCI slot config\n" 265 " -S: guest memory cannot be swapped\n" 266 " -u: RTC keeps UTC time\n" 267 " -U: uuid\n" 268 " -w: ignore unimplemented MSRs\n" 269 " -W: force virtio to use single-vector MSI\n" 270 " -x: local apic is in x2APIC mode\n" 271 " -Y: disable MPtable generation\n", 272 progname, (int)strlen(progname), "", (int)strlen(progname), "", 273 (int)strlen(progname), ""); 274 275 exit(code); 276 } 277 278 /* 279 * XXX This parser is known to have the following issues: 280 * 1. It accepts null key=value tokens ",," as setting "cpus" to an 281 * empty string. 282 * 283 * The acceptance of a null specification ('-c ""') is by design to match the 284 * manual page syntax specification, this results in a topology of 1 vCPU. 285 */ 286 static int 287 topology_parse(const char *opt) 288 { 289 char *cp, *str; 290 291 if (*opt == '\0') { 292 set_config_value("sockets", "1"); 293 set_config_value("cores", "1"); 294 set_config_value("threads", "1"); 295 set_config_value("cpus", "1"); 296 return (0); 297 } 298 299 str = strdup(opt); 300 if (str == NULL) 301 errx(4, "Failed to allocate memory"); 302 303 while ((cp = strsep(&str, ",")) != NULL) { 304 if (strncmp(cp, "cpus=", strlen("cpus=")) == 0) 305 set_config_value("cpus", cp + strlen("cpus=")); 306 else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0) 307 set_config_value("sockets", cp + strlen("sockets=")); 308 else if (strncmp(cp, "cores=", strlen("cores=")) == 0) 309 set_config_value("cores", cp + strlen("cores=")); 310 else if (strncmp(cp, "threads=", strlen("threads=")) == 0) 311 set_config_value("threads", cp + strlen("threads=")); 312 #ifdef notyet /* Do not expose this until vmm.ko implements it */ 313 else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0) 314 set_config_value("maxcpus", cp + strlen("maxcpus=")); 315 #endif 316 else if (strchr(cp, '=') != NULL) 317 goto out; 318 else 319 set_config_value("cpus", cp); 320 } 321 free(str); 322 return (0); 323 324 out: 325 free(str); 326 return (-1); 327 } 328 329 static int 330 parse_int_value(const char *key, const char *value, int minval, int maxval) 331 { 332 char *cp; 333 long lval; 334 335 errno = 0; 336 lval = strtol(value, &cp, 0); 337 if (errno != 0 || *cp != '\0' || cp == value || lval < minval || 338 lval > maxval) 339 errx(4, "Invalid value for %s: '%s'", key, value); 340 return (lval); 341 } 342 343 /* 344 * Set the sockets, cores, threads, and guest_cpus variables based on 345 * the configured topology. 346 * 347 * The limits of UINT16_MAX are due to the types passed to 348 * vm_set_topology(). vmm.ko may enforce tighter limits. 349 */ 350 static void 351 calc_topolopgy(void) 352 { 353 const char *value; 354 bool explicit_cpus; 355 uint64_t ncpus; 356 357 value = get_config_value("cpus"); 358 if (value != NULL) { 359 guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX); 360 explicit_cpus = true; 361 } else { 362 guest_ncpus = 1; 363 explicit_cpus = false; 364 } 365 value = get_config_value("cores"); 366 if (value != NULL) 367 cores = parse_int_value("cores", value, 1, UINT16_MAX); 368 else 369 cores = 1; 370 value = get_config_value("threads"); 371 if (value != NULL) 372 threads = parse_int_value("threads", value, 1, UINT16_MAX); 373 else 374 threads = 1; 375 value = get_config_value("sockets"); 376 if (value != NULL) 377 sockets = parse_int_value("sockets", value, 1, UINT16_MAX); 378 else 379 sockets = guest_ncpus; 380 381 /* 382 * Compute sockets * cores * threads avoiding overflow. The 383 * range check above insures these are 16 bit values. 384 */ 385 ncpus = (uint64_t)sockets * cores * threads; 386 if (ncpus > UINT16_MAX) 387 errx(4, "Computed number of vCPUs too high: %ju", 388 (uintmax_t)ncpus); 389 390 if (explicit_cpus) { 391 if (guest_ncpus != ncpus) 392 errx(4, "Topology (%d sockets, %d cores, %d threads) " 393 "does not match %d vCPUs", sockets, cores, threads, 394 guest_ncpus); 395 } else 396 guest_ncpus = ncpus; 397 } 398 399 #ifndef WITHOUT_CAPSICUM 400 /* 401 * 11-stable capsicum helpers 402 */ 403 static void 404 bhyve_caph_cache_catpages(void) 405 { 406 407 (void)catopen("libc", NL_CAT_LOCALE); 408 } 409 410 static int 411 bhyve_caph_limit_stdoe(void) 412 { 413 cap_rights_t rights; 414 unsigned long cmds[] = { TIOCGETA, TIOCGWINSZ }; 415 int i, fds[] = { STDOUT_FILENO, STDERR_FILENO }; 416 417 cap_rights_init(&rights, CAP_FCNTL, CAP_FSTAT, CAP_IOCTL); 418 cap_rights_set(&rights, CAP_WRITE); 419 420 for (i = 0; i < nitems(fds); i++) { 421 if (cap_rights_limit(fds[i], &rights) < 0 && errno != ENOSYS) 422 return (-1); 423 424 if (cap_ioctls_limit(fds[i], cmds, nitems(cmds)) < 0 && errno != ENOSYS) 425 return (-1); 426 427 if (cap_fcntls_limit(fds[i], CAP_FCNTL_GETFL) < 0 && errno != ENOSYS) 428 return (-1); 429 } 430 431 return (0); 432 } 433 434 #endif 435 436 #ifdef __FreeBSD__ 437 static int 438 pincpu_parse(const char *opt) 439 { 440 int vcpu, pcpu; 441 442 if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) { 443 fprintf(stderr, "invalid format: %s\n", opt); 444 return (-1); 445 } 446 447 if (vcpu < 0 || vcpu >= VM_MAXCPU) { 448 fprintf(stderr, "vcpu '%d' outside valid range from 0 to %d\n", 449 vcpu, VM_MAXCPU - 1); 450 return (-1); 451 } 452 453 if (pcpu < 0 || pcpu >= CPU_SETSIZE) { 454 fprintf(stderr, "hostcpu '%d' outside valid range from " 455 "0 to %d\n", pcpu, CPU_SETSIZE - 1); 456 return (-1); 457 } 458 459 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 460 value = get_config_value(key); 461 462 if (asprintf(&newval, "%s%s%d", value != NULL ? value : "", 463 value != NULL ? "," : "", pcpu) == -1) { 464 perror("failed to build new cpuset string"); 465 return (-1); 466 } 467 468 set_config_value(key, newval); 469 free(newval); 470 return (0); 471 } 472 473 static void 474 parse_cpuset(int vcpu, const char *list, cpuset_t *set) 475 { 476 char *cp, *token; 477 int pcpu, start; 478 479 CPU_ZERO(set); 480 start = -1; 481 token = __DECONST(char *, list); 482 for (;;) { 483 pcpu = strtoul(token, &cp, 0); 484 if (cp == token) 485 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 486 if (pcpu < 0 || pcpu >= CPU_SETSIZE) 487 errx(4, "hostcpu '%d' outside valid range from 0 to %d", 488 pcpu, CPU_SETSIZE - 1); 489 switch (*cp) { 490 case ',': 491 case '\0': 492 if (start >= 0) { 493 if (start > pcpu) 494 errx(4, "Invalid hostcpu range %d-%d", 495 start, pcpu); 496 while (start < pcpu) { 497 CPU_SET(start, vcpumap[vcpu]); 498 start++; 499 } 500 start = -1; 501 } 502 CPU_SET(pcpu, vcpumap[vcpu]); 503 break; 504 case '-': 505 if (start >= 0) 506 errx(4, "invalid cpuset for vcpu %d: '%s'", 507 vcpu, list); 508 start = pcpu; 509 break; 510 default: 511 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 512 } 513 if (*cp == '\0') 514 break; 515 token = cp + 1; 516 } 517 } 518 519 static void 520 build_vcpumaps(void) 521 { 522 char key[16]; 523 const char *value; 524 int vcpu; 525 526 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) { 527 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 528 value = get_config_value(key); 529 if (value == NULL) 530 continue; 531 vcpumap[vcpu] = malloc(sizeof(cpuset_t)); 532 if (vcpumap[vcpu] == NULL) 533 err(4, "Failed to allocate cpuset for vcpu %d", vcpu); 534 parse_cpuset(vcpu, value, vcpumap[vcpu]); 535 } 536 } 537 538 void 539 vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid, 540 int errcode) 541 { 542 struct vmctx *ctx; 543 int error, restart_instruction; 544 545 ctx = arg; 546 restart_instruction = 1; 547 548 error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode, 549 restart_instruction); 550 assert(error == 0); 551 } 552 #endif /* __FreeBSD__ */ 553 554 void * 555 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len) 556 { 557 558 return (vm_map_gpa(ctx, gaddr, len)); 559 } 560 561 int 562 fbsdrun_virtio_msix(void) 563 { 564 565 return (get_config_bool_default("virtio_msix", true)); 566 } 567 568 static void * 569 fbsdrun_start_thread(void *param) 570 { 571 char tname[MAXCOMLEN + 1]; 572 struct mt_vmm_info *mtp; 573 int vcpu; 574 575 mtp = param; 576 vcpu = mtp->mt_vcpu; 577 578 snprintf(tname, sizeof(tname), "vcpu %d", vcpu); 579 pthread_set_name_np(mtp->mt_thr, tname); 580 581 gdb_cpu_add(vcpu); 582 583 vm_loop(mtp->mt_ctx, vcpu, mtp->mt_startrip); 584 585 /* not reached */ 586 exit(1); 587 return (NULL); 588 } 589 590 #ifdef __FreeBSD__ 591 void 592 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip) 593 #else 594 void 595 fbsdrun_addcpu(struct vmctx *ctx, int newcpu, uint64_t rip, bool suspend) 596 #endif 597 { 598 int error; 599 600 #ifdef __FreeBSD__ 601 assert(fromcpu == BSP); 602 #endif 603 604 /* 605 * The 'newcpu' must be activated in the context of 'fromcpu'. If 606 * vm_activate_cpu() is delayed until newcpu's pthread starts running 607 * then vmm.ko is out-of-sync with bhyve and this can create a race 608 * with vm_suspend(). 609 */ 610 error = vm_activate_cpu(ctx, newcpu); 611 if (error != 0) 612 err(EX_OSERR, "could not activate CPU %d", newcpu); 613 614 CPU_SET_ATOMIC(newcpu, &cpumask); 615 616 #ifndef __FreeBSD__ 617 if (suspend) 618 (void) vm_suspend_cpu(ctx, newcpu); 619 #endif 620 621 /* 622 * Set up the vmexit struct to allow execution to start 623 * at the given RIP 624 */ 625 mt_vmm_info[newcpu].mt_ctx = ctx; 626 mt_vmm_info[newcpu].mt_vcpu = newcpu; 627 mt_vmm_info[newcpu].mt_startrip = rip; 628 629 error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL, 630 fbsdrun_start_thread, &mt_vmm_info[newcpu]); 631 assert(error == 0); 632 } 633 634 static int 635 fbsdrun_deletecpu(struct vmctx *ctx, int vcpu) 636 { 637 638 if (!CPU_ISSET(vcpu, &cpumask)) { 639 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu); 640 exit(4); 641 } 642 643 CPU_CLR_ATOMIC(vcpu, &cpumask); 644 return (CPU_EMPTY(&cpumask)); 645 } 646 647 static void 648 vmentry_mmio_read(int vcpu, uint64_t gpa, uint8_t bytes, uint64_t data) 649 { 650 struct vm_entry *entry = &vmentry[vcpu]; 651 struct vm_mmio *mmio = &entry->u.mmio; 652 653 assert(entry->cmd == VEC_DEFAULT); 654 655 entry->cmd = VEC_FULFILL_MMIO; 656 mmio->bytes = bytes; 657 mmio->read = 1; 658 mmio->gpa = gpa; 659 mmio->data = data; 660 } 661 662 static void 663 vmentry_mmio_write(int vcpu, uint64_t gpa, uint8_t bytes) 664 { 665 struct vm_entry *entry = &vmentry[vcpu]; 666 struct vm_mmio *mmio = &entry->u.mmio; 667 668 assert(entry->cmd == VEC_DEFAULT); 669 670 entry->cmd = VEC_FULFILL_MMIO; 671 mmio->bytes = bytes; 672 mmio->read = 0; 673 mmio->gpa = gpa; 674 mmio->data = 0; 675 } 676 677 static void 678 vmentry_inout_read(int vcpu, uint16_t port, uint8_t bytes, uint32_t data) 679 { 680 struct vm_entry *entry = &vmentry[vcpu]; 681 struct vm_inout *inout = &entry->u.inout; 682 683 assert(entry->cmd == VEC_DEFAULT); 684 685 entry->cmd = VEC_FULFILL_INOUT; 686 inout->bytes = bytes; 687 inout->flags = INOUT_IN; 688 inout->port = port; 689 inout->eax = data; 690 } 691 692 static void 693 vmentry_inout_write(int vcpu, uint16_t port, uint8_t bytes) 694 { 695 struct vm_entry *entry = &vmentry[vcpu]; 696 struct vm_inout *inout = &entry->u.inout; 697 698 assert(entry->cmd == VEC_DEFAULT); 699 700 entry->cmd = VEC_FULFILL_INOUT; 701 inout->bytes = bytes; 702 inout->flags = 0; 703 inout->port = port; 704 inout->eax = 0; 705 } 706 707 static int 708 vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu, 709 uint32_t eax) 710 { 711 #if BHYVE_DEBUG 712 /* 713 * put guest-driven debug here 714 */ 715 #endif 716 return (VMEXIT_CONTINUE); 717 } 718 719 static int 720 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 721 { 722 int error; 723 int vcpu; 724 struct vm_inout inout; 725 bool in; 726 uint8_t bytes; 727 728 stats.vmexit_inout++; 729 730 vcpu = *pvcpu; 731 inout = vme->u.inout; 732 in = (inout.flags & INOUT_IN) != 0; 733 bytes = inout.bytes; 734 735 /* Extra-special case of host notifications */ 736 if (!in && inout.port == GUEST_NIO_PORT) { 737 error = vmexit_handle_notify(ctx, vme, pvcpu, inout.eax); 738 vmentry_inout_write(vcpu, inout.port, bytes); 739 return (error); 740 } 741 742 error = emulate_inout(ctx, vcpu, &inout); 743 if (error) { 744 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n", 745 in ? "in" : "out", 746 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'), 747 inout.port, vmexit->rip); 748 return (VMEXIT_ABORT); 749 } else { 750 /* 751 * Communicate the status of the inout operation back to the 752 * in-kernel instruction emulation. 753 */ 754 if (in) { 755 vmentry_inout_read(vcpu, inout.port, bytes, inout.eax); 756 } else { 757 vmentry_inout_write(vcpu, inout.port, bytes); 758 } 759 return (VMEXIT_CONTINUE); 760 } 761 } 762 763 static int 764 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 765 { 766 uint64_t val; 767 uint32_t eax, edx; 768 int error; 769 770 val = 0; 771 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val); 772 if (error != 0) { 773 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n", 774 vme->u.msr.code, *pvcpu); 775 if (get_config_bool("x86.strictmsr")) { 776 vm_inject_gp(ctx, *pvcpu); 777 return (VMEXIT_CONTINUE); 778 } 779 } 780 781 eax = val; 782 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax); 783 assert(error == 0); 784 785 edx = val >> 32; 786 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx); 787 assert(error == 0); 788 789 return (VMEXIT_CONTINUE); 790 } 791 792 static int 793 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 794 { 795 int error; 796 797 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval); 798 if (error != 0) { 799 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n", 800 vme->u.msr.code, vme->u.msr.wval, *pvcpu); 801 if (get_config_bool("x86.strictmsr")) { 802 vm_inject_gp(ctx, *pvcpu); 803 return (VMEXIT_CONTINUE); 804 } 805 } 806 return (VMEXIT_CONTINUE); 807 } 808 809 #ifdef __FreeBSD__ 810 static int 811 vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 812 { 813 814 (void)spinup_ap(ctx, *pvcpu, 815 vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip); 816 817 return (VMEXIT_CONTINUE); 818 } 819 #else 820 static int 821 vmexit_run_state(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 822 { 823 /* 824 * Run-state transitions (INIT, SIPI, etc) are handled in-kernel, so an 825 * exit to userspace with that code is not expected. 826 */ 827 fprintf(stderr, "unexpected run-state VM exit"); 828 return (VMEXIT_ABORT); 829 } 830 #endif /* __FreeBSD__ */ 831 832 #ifdef __FreeBSD__ 833 #define DEBUG_EPT_MISCONFIG 834 #else 835 /* EPT misconfig debugging not possible now that raw VMCS access is gone */ 836 #endif 837 838 #ifdef DEBUG_EPT_MISCONFIG 839 #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400 840 841 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4]; 842 static int ept_misconfig_ptenum; 843 #endif 844 845 static const char * 846 vmexit_vmx_desc(uint32_t exit_reason) 847 { 848 849 if (exit_reason >= nitems(vmx_exit_reason_desc) || 850 vmx_exit_reason_desc[exit_reason] == NULL) 851 return ("Unknown"); 852 return (vmx_exit_reason_desc[exit_reason]); 853 } 854 855 static int 856 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 857 { 858 859 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 860 fprintf(stderr, "\treason\t\tVMX\n"); 861 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 862 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 863 fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status); 864 fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason, 865 vmexit_vmx_desc(vmexit->u.vmx.exit_reason)); 866 fprintf(stderr, "\tqualification\t0x%016lx\n", 867 vmexit->u.vmx.exit_qualification); 868 fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type); 869 fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error); 870 #ifdef DEBUG_EPT_MISCONFIG 871 if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) { 872 vm_get_register(ctx, *pvcpu, 873 VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS), 874 &ept_misconfig_gpa); 875 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte, 876 &ept_misconfig_ptenum); 877 fprintf(stderr, "\tEPT misconfiguration:\n"); 878 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa); 879 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n", 880 ept_misconfig_ptenum, ept_misconfig_pte[0], 881 ept_misconfig_pte[1], ept_misconfig_pte[2], 882 ept_misconfig_pte[3]); 883 } 884 #endif /* DEBUG_EPT_MISCONFIG */ 885 return (VMEXIT_ABORT); 886 } 887 888 static int 889 vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 890 { 891 892 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 893 fprintf(stderr, "\treason\t\tSVM\n"); 894 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 895 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 896 fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode); 897 fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1); 898 fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2); 899 return (VMEXIT_ABORT); 900 } 901 902 static int 903 vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 904 { 905 906 assert(vmexit->inst_length == 0); 907 908 stats.vmexit_bogus++; 909 910 return (VMEXIT_CONTINUE); 911 } 912 913 static int 914 vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 915 { 916 917 assert(vmexit->inst_length == 0); 918 919 stats.vmexit_reqidle++; 920 921 return (VMEXIT_CONTINUE); 922 } 923 924 static int 925 vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 926 { 927 928 stats.vmexit_hlt++; 929 930 /* 931 * Just continue execution with the next instruction. We use 932 * the HLT VM exit as a way to be friendly with the host 933 * scheduler. 934 */ 935 return (VMEXIT_CONTINUE); 936 } 937 938 static int 939 vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 940 { 941 942 stats.vmexit_pause++; 943 944 return (VMEXIT_CONTINUE); 945 } 946 947 static int 948 vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 949 { 950 951 assert(vmexit->inst_length == 0); 952 953 stats.vmexit_mtrap++; 954 955 gdb_cpu_mtrap(*pvcpu); 956 957 return (VMEXIT_CONTINUE); 958 } 959 960 static int 961 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 962 { 963 uint8_t i, valid; 964 965 fprintf(stderr, "Failed to emulate instruction sequence "); 966 967 valid = vmexit->u.inst_emul.num_valid; 968 if (valid != 0) { 969 assert(valid <= sizeof (vmexit->u.inst_emul.inst)); 970 fprintf(stderr, "["); 971 for (i = 0; i < valid; i++) { 972 if (i == 0) { 973 fprintf(stderr, "%02x", 974 vmexit->u.inst_emul.inst[i]); 975 } else { 976 fprintf(stderr, ", %02x", 977 vmexit->u.inst_emul.inst[i]); 978 } 979 } 980 fprintf(stderr, "] "); 981 } 982 fprintf(stderr, "@ %rip = %x\n", vmexit->rip); 983 984 return (VMEXIT_ABORT); 985 } 986 987 static int 988 vmexit_mmio(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 989 { 990 int vcpu, err; 991 struct vm_mmio mmio; 992 bool is_read; 993 994 stats.vmexit_mmio++; 995 996 vcpu = *pvcpu; 997 mmio = vmexit->u.mmio; 998 is_read = (mmio.read != 0); 999 1000 err = emulate_mem(ctx, vcpu, &mmio); 1001 1002 if (err == ESRCH) { 1003 fprintf(stderr, "Unhandled memory access to 0x%lx\n", mmio.gpa); 1004 stats.mmio_unhandled++; 1005 1006 /* 1007 * Access to non-existent physical addresses is not likely to 1008 * result in fatal errors on hardware machines, but rather reads 1009 * of all-ones or discarded-but-acknowledged writes. 1010 */ 1011 mmio.data = ~0UL; 1012 err = 0; 1013 } 1014 1015 if (err == 0) { 1016 if (is_read) { 1017 vmentry_mmio_read(vcpu, mmio.gpa, mmio.bytes, 1018 mmio.data); 1019 } else { 1020 vmentry_mmio_write(vcpu, mmio.gpa, mmio.bytes); 1021 } 1022 return (VMEXIT_CONTINUE); 1023 } 1024 1025 fprintf(stderr, "Unhandled mmio error to 0x%lx: %d\n", mmio.gpa, err); 1026 return (VMEXIT_ABORT); 1027 } 1028 1029 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER; 1030 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER; 1031 1032 static int 1033 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1034 { 1035 enum vm_suspend_how how; 1036 1037 how = vmexit->u.suspended.how; 1038 1039 fbsdrun_deletecpu(ctx, *pvcpu); 1040 1041 if (*pvcpu != BSP) { 1042 pthread_mutex_lock(&resetcpu_mtx); 1043 pthread_cond_signal(&resetcpu_cond); 1044 pthread_mutex_unlock(&resetcpu_mtx); 1045 pthread_exit(NULL); 1046 } 1047 1048 pthread_mutex_lock(&resetcpu_mtx); 1049 while (!CPU_EMPTY(&cpumask)) { 1050 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx); 1051 } 1052 pthread_mutex_unlock(&resetcpu_mtx); 1053 1054 switch (how) { 1055 case VM_SUSPEND_RESET: 1056 exit(0); 1057 case VM_SUSPEND_POWEROFF: 1058 if (get_config_bool_default("destroy_on_poweroff", false)) 1059 vm_destroy(ctx); 1060 exit(1); 1061 case VM_SUSPEND_HALT: 1062 exit(2); 1063 case VM_SUSPEND_TRIPLEFAULT: 1064 exit(3); 1065 default: 1066 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how); 1067 exit(100); 1068 } 1069 return (0); /* NOTREACHED */ 1070 } 1071 1072 static int 1073 vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1074 { 1075 1076 gdb_cpu_suspend(*pvcpu); 1077 return (VMEXIT_CONTINUE); 1078 } 1079 1080 static int 1081 vmexit_breakpoint(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1082 { 1083 1084 gdb_cpu_breakpoint(*pvcpu, vmexit); 1085 return (VMEXIT_CONTINUE); 1086 } 1087 1088 static vmexit_handler_t handler[VM_EXITCODE_MAX] = { 1089 [VM_EXITCODE_INOUT] = vmexit_inout, 1090 [VM_EXITCODE_MMIO] = vmexit_mmio, 1091 [VM_EXITCODE_VMX] = vmexit_vmx, 1092 [VM_EXITCODE_SVM] = vmexit_svm, 1093 [VM_EXITCODE_BOGUS] = vmexit_bogus, 1094 [VM_EXITCODE_REQIDLE] = vmexit_reqidle, 1095 [VM_EXITCODE_RDMSR] = vmexit_rdmsr, 1096 [VM_EXITCODE_WRMSR] = vmexit_wrmsr, 1097 [VM_EXITCODE_MTRAP] = vmexit_mtrap, 1098 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul, 1099 #ifdef __FreeBSD__ 1100 [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap, 1101 #else 1102 [VM_EXITCODE_RUN_STATE] = vmexit_run_state, 1103 #endif 1104 [VM_EXITCODE_SUSPENDED] = vmexit_suspend, 1105 [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch, 1106 [VM_EXITCODE_DEBUG] = vmexit_debug, 1107 [VM_EXITCODE_BPT] = vmexit_breakpoint, 1108 }; 1109 1110 static void 1111 vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip) 1112 { 1113 int error, rc; 1114 enum vm_exitcode exitcode; 1115 cpuset_t active_cpus; 1116 struct vm_exit *vexit; 1117 struct vm_entry *ventry; 1118 1119 #ifdef __FreeBSD__ 1120 if (vcpumap[vcpu] != NULL) { 1121 error = pthread_setaffinity_np(pthread_self(), 1122 sizeof(cpuset_t), vcpumap[vcpu]); 1123 assert(error == 0); 1124 } 1125 #endif 1126 error = vm_active_cpus(ctx, &active_cpus); 1127 assert(CPU_ISSET(vcpu, &active_cpus)); 1128 1129 error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip); 1130 assert(error == 0); 1131 1132 ventry = &vmentry[vcpu]; 1133 vexit = &vmexit[vcpu]; 1134 1135 while (1) { 1136 error = vm_run(ctx, vcpu, ventry, vexit); 1137 if (error != 0) 1138 break; 1139 1140 if (ventry->cmd != VEC_DEFAULT) { 1141 /* 1142 * Discard any lingering entry state after it has been 1143 * submitted via vm_run(). 1144 */ 1145 bzero(ventry, sizeof (*ventry)); 1146 } 1147 1148 exitcode = vexit->exitcode; 1149 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) { 1150 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n", 1151 exitcode); 1152 exit(4); 1153 } 1154 1155 rc = (*handler[exitcode])(ctx, vexit, &vcpu); 1156 1157 switch (rc) { 1158 case VMEXIT_CONTINUE: 1159 break; 1160 case VMEXIT_ABORT: 1161 abort(); 1162 default: 1163 exit(4); 1164 } 1165 } 1166 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno); 1167 } 1168 1169 static int 1170 num_vcpus_allowed(struct vmctx *ctx) 1171 { 1172 #ifdef __FreeBSD__ 1173 int tmp, error; 1174 1175 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp); 1176 1177 /* 1178 * The guest is allowed to spinup more than one processor only if the 1179 * UNRESTRICTED_GUEST capability is available. 1180 */ 1181 if (error == 0) 1182 return (VM_MAXCPU); 1183 else 1184 return (1); 1185 #else 1186 /* Unrestricted Guest is always enabled on illumos */ 1187 return (VM_MAXCPU); 1188 #endif /* __FreeBSD__ */ 1189 } 1190 1191 void 1192 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu) 1193 { 1194 int err, tmp; 1195 1196 if (get_config_bool_default("x86.vmexit_on_hlt", false)) { 1197 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp); 1198 if (err < 0) { 1199 fprintf(stderr, "VM exit on HLT not supported\n"); 1200 exit(4); 1201 } 1202 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1); 1203 if (cpu == BSP) 1204 handler[VM_EXITCODE_HLT] = vmexit_hlt; 1205 } 1206 1207 if (get_config_bool_default("x86.vmexit_on_pause", false)) { 1208 /* 1209 * pause exit support required for this mode 1210 */ 1211 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp); 1212 if (err < 0) { 1213 fprintf(stderr, 1214 "SMP mux requested, no pause support\n"); 1215 exit(4); 1216 } 1217 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1); 1218 if (cpu == BSP) 1219 handler[VM_EXITCODE_PAUSE] = vmexit_pause; 1220 } 1221 1222 if (get_config_bool_default("x86.x2apic", false)) 1223 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED); 1224 else 1225 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED); 1226 1227 if (err) { 1228 fprintf(stderr, "Unable to set x2apic state (%d)\n", err); 1229 exit(4); 1230 } 1231 1232 #ifdef __FreeBSD__ 1233 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1); 1234 #endif 1235 } 1236 1237 static struct vmctx * 1238 do_open(const char *vmname) 1239 { 1240 struct vmctx *ctx; 1241 int error; 1242 bool reinit, romboot; 1243 #ifndef WITHOUT_CAPSICUM 1244 cap_rights_t rights; 1245 const cap_ioctl_t *cmds; 1246 size_t ncmds; 1247 #endif 1248 1249 reinit = romboot = false; 1250 1251 if (lpc_bootrom()) 1252 romboot = true; 1253 1254 error = vm_create(vmname); 1255 if (error) { 1256 if (errno == EEXIST) { 1257 if (romboot) { 1258 reinit = true; 1259 } else { 1260 /* 1261 * The virtual machine has been setup by the 1262 * userspace bootloader. 1263 */ 1264 } 1265 } else { 1266 perror("vm_create"); 1267 exit(4); 1268 } 1269 } else { 1270 if (!romboot) { 1271 /* 1272 * If the virtual machine was just created then a 1273 * bootrom must be configured to boot it. 1274 */ 1275 fprintf(stderr, "virtual machine cannot be booted\n"); 1276 exit(4); 1277 } 1278 } 1279 1280 ctx = vm_open(vmname); 1281 if (ctx == NULL) { 1282 perror("vm_open"); 1283 exit(4); 1284 } 1285 1286 #ifndef WITHOUT_CAPSICUM 1287 cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW); 1288 if (caph_rights_limit(vm_get_device_fd(ctx), &rights) == -1) 1289 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1290 vm_get_ioctls(&ncmds); 1291 cmds = vm_get_ioctls(NULL); 1292 if (cmds == NULL) 1293 errx(EX_OSERR, "out of memory"); 1294 if (caph_ioctls_limit(vm_get_device_fd(ctx), cmds, ncmds) == -1) 1295 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1296 free((cap_ioctl_t *)cmds); 1297 #endif 1298 1299 if (reinit) { 1300 error = vm_reinit(ctx); 1301 if (error) { 1302 perror("vm_reinit"); 1303 exit(4); 1304 } 1305 } 1306 error = vm_set_topology(ctx, sockets, cores, threads, maxcpus); 1307 if (error) 1308 errx(EX_OSERR, "vm_set_topology"); 1309 return (ctx); 1310 } 1311 1312 static bool 1313 parse_config_option(const char *option) 1314 { 1315 const char *value; 1316 char *path; 1317 1318 value = strchr(option, '='); 1319 if (value == NULL || value[1] == '\0') 1320 return (false); 1321 path = strndup(option, value - option); 1322 if (path == NULL) 1323 err(4, "Failed to allocate memory"); 1324 set_config_value(path, value + 1); 1325 return (true); 1326 } 1327 1328 static void 1329 parse_simple_config_file(const char *path) 1330 { 1331 FILE *fp; 1332 char *line, *cp; 1333 size_t linecap; 1334 unsigned int lineno; 1335 1336 fp = fopen(path, "r"); 1337 if (fp == NULL) 1338 err(4, "Failed to open configuration file %s", path); 1339 line = NULL; 1340 linecap = 0; 1341 lineno = 1; 1342 for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) { 1343 if (*line == '#' || *line == '\n') 1344 continue; 1345 cp = strchr(line, '\n'); 1346 if (cp != NULL) 1347 *cp = '\0'; 1348 if (!parse_config_option(line)) 1349 errx(4, "%s line %u: invalid config option '%s'", path, 1350 lineno, line); 1351 } 1352 free(line); 1353 fclose(fp); 1354 } 1355 1356 static void 1357 set_defaults(void) 1358 { 1359 1360 set_config_bool("acpi_tables", false); 1361 set_config_value("memory.size", "256M"); 1362 set_config_bool("x86.strictmsr", true); 1363 } 1364 1365 int 1366 main(int argc, char *argv[]) 1367 { 1368 int c, error, err; 1369 int max_vcpus, memflags; 1370 struct vmctx *ctx; 1371 uint64_t rip; 1372 size_t memsize; 1373 const char *value, *vmname; 1374 char *optstr; 1375 1376 init_config(); 1377 set_defaults(); 1378 progname = basename(argv[0]); 1379 1380 #ifdef __FreeBSD__ 1381 optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:U:"; 1382 #else 1383 /* +d, +B, -p */ 1384 optstr = "adehuwxACDHIPSWYk:o:G:c:s:m:l:B:U:"; 1385 #endif 1386 while ((c = getopt(argc, argv, optstr)) != -1) { 1387 switch (c) { 1388 case 'a': 1389 set_config_bool("x86.x2apic", false); 1390 break; 1391 case 'A': 1392 set_config_bool("acpi_tables", true); 1393 break; 1394 case 'D': 1395 set_config_bool("destroy_on_poweroff", true); 1396 break; 1397 #ifndef __FreeBSD__ 1398 case 'B': 1399 if (smbios_parse(optarg) != 0) { 1400 errx(EX_USAGE, "invalid SMBIOS " 1401 "configuration '%s'", optarg); 1402 } 1403 break; 1404 case 'd': 1405 set_config_bool("suspend_at_boot", true); 1406 break; 1407 #endif 1408 #ifdef __FreeBSD__ 1409 case 'p': 1410 if (pincpu_parse(optarg) != 0) { 1411 errx(EX_USAGE, "invalid vcpu pinning " 1412 "configuration '%s'", optarg); 1413 } 1414 break; 1415 #endif 1416 case 'c': 1417 if (topology_parse(optarg) != 0) { 1418 errx(EX_USAGE, "invalid cpu topology " 1419 "'%s'", optarg); 1420 } 1421 break; 1422 case 'C': 1423 set_config_bool("memory.guest_in_core", true); 1424 break; 1425 case 'G': 1426 if (optarg[0] == 'w') { 1427 set_config_bool("gdb.wait", true); 1428 optarg++; 1429 } 1430 set_config_value("gdb.port", optarg); 1431 break; 1432 case 'k': 1433 parse_simple_config_file(optarg); 1434 break; 1435 case 'l': 1436 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1437 lpc_print_supported_devices(); 1438 exit(0); 1439 } else if (lpc_device_parse(optarg) != 0) { 1440 errx(EX_USAGE, "invalid lpc device " 1441 "configuration '%s'", optarg); 1442 } 1443 break; 1444 case 's': 1445 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1446 pci_print_supported_devices(); 1447 exit(0); 1448 } else if (pci_parse_slot(optarg) != 0) 1449 exit(4); 1450 else 1451 break; 1452 case 'S': 1453 set_config_bool("memory.wired", true); 1454 break; 1455 case 'm': 1456 set_config_value("memory.size", optarg); 1457 break; 1458 case 'o': 1459 if (!parse_config_option(optarg)) 1460 errx(EX_USAGE, "invalid configuration option '%s'", optarg); 1461 break; 1462 case 'H': 1463 set_config_bool("x86.vmexit_on_hlt", true); 1464 break; 1465 case 'I': 1466 /* 1467 * The "-I" option was used to add an ioapic to the 1468 * virtual machine. 1469 * 1470 * An ioapic is now provided unconditionally for each 1471 * virtual machine and this option is now deprecated. 1472 */ 1473 break; 1474 case 'P': 1475 set_config_bool("x86.vmexit_on_pause", true); 1476 break; 1477 case 'e': 1478 set_config_bool("x86.strictio", true); 1479 break; 1480 case 'u': 1481 set_config_bool("rtc.use_localtime", false); 1482 break; 1483 case 'U': 1484 set_config_value("uuid", optarg); 1485 break; 1486 case 'w': 1487 set_config_bool("x86.strictmsr", false); 1488 break; 1489 case 'W': 1490 set_config_bool("virtio_msix", false); 1491 break; 1492 case 'x': 1493 set_config_bool("x86.x2apic", true); 1494 break; 1495 case 'Y': 1496 set_config_bool("x86.mptable", false); 1497 break; 1498 case 'h': 1499 usage(0); 1500 default: 1501 usage(1); 1502 } 1503 } 1504 argc -= optind; 1505 argv += optind; 1506 1507 if (argc > 1) 1508 usage(1); 1509 1510 if (argc == 1) 1511 set_config_value("name", argv[0]); 1512 1513 vmname = get_config_value("name"); 1514 if (vmname == NULL) 1515 usage(1); 1516 1517 if (get_config_bool_default("config.dump", false)) { 1518 dump_config(); 1519 exit(1); 1520 } 1521 1522 calc_topolopgy(); 1523 #ifdef __FreeBSD__ 1524 build_vcpumaps(); 1525 #endif 1526 1527 value = get_config_value("memory.size"); 1528 error = vm_parse_memsize(value, &memsize); 1529 if (error) 1530 errx(EX_USAGE, "invalid memsize '%s'", value); 1531 1532 ctx = do_open(vmname); 1533 1534 max_vcpus = num_vcpus_allowed(ctx); 1535 if (guest_ncpus > max_vcpus) { 1536 fprintf(stderr, "%d vCPUs requested but only %d available\n", 1537 guest_ncpus, max_vcpus); 1538 exit(4); 1539 } 1540 1541 fbsdrun_set_capabilities(ctx, BSP); 1542 1543 memflags = 0; 1544 if (get_config_bool_default("memory.wired", false)) 1545 memflags |= VM_MEM_F_WIRED; 1546 if (get_config_bool_default("memory.guest_in_core", false)) 1547 memflags |= VM_MEM_F_INCORE; 1548 vm_set_memflags(ctx, memflags); 1549 #ifdef __FreeBSD__ 1550 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1551 #else 1552 do { 1553 errno = 0; 1554 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1555 error = errno; 1556 if (err != 0 && error == ENOMEM) { 1557 (void) fprintf(stderr, "Unable to allocate memory " 1558 "(%llu), retrying in 1 second\n", memsize); 1559 sleep(1); 1560 } 1561 } while (error == ENOMEM); 1562 #endif 1563 if (err) { 1564 fprintf(stderr, "Unable to set up memory (%d)\n", errno); 1565 exit(4); 1566 } 1567 1568 error = init_msr(); 1569 if (error) { 1570 fprintf(stderr, "init_msr error %d", error); 1571 exit(4); 1572 } 1573 1574 init_mem(); 1575 init_inout(); 1576 #ifdef __FreeBSD__ 1577 kernemu_dev_init(); 1578 #endif 1579 init_bootrom(ctx); 1580 atkbdc_init(ctx); 1581 pci_irq_init(ctx); 1582 ioapic_init(ctx); 1583 1584 rtc_init(ctx); 1585 sci_init(ctx); 1586 #ifndef __FreeBSD__ 1587 pmtmr_init(ctx); 1588 #endif 1589 1590 /* 1591 * Exit if a device emulation finds an error in its initilization 1592 */ 1593 if (init_pci(ctx) != 0) { 1594 perror("device emulation initialization error"); 1595 exit(4); 1596 } 1597 1598 /* 1599 * Initialize after PCI, to allow a bootrom file to reserve the high 1600 * region. 1601 */ 1602 if (get_config_bool("acpi_tables")) 1603 vmgenc_init(ctx); 1604 1605 value = get_config_value("gdb.port"); 1606 #ifdef __FreeBSD__ 1607 if (value != NULL) 1608 init_gdb(ctx, atoi(value), get_config_bool_default("gdb.wait", 1609 false)); 1610 #else 1611 if (value != NULL) { 1612 int port = atoi(value); 1613 1614 if (port < 0) { 1615 init_mdb(ctx, 1616 get_config_bool_default("gdb.wait", false)); 1617 } else { 1618 init_gdb(ctx, port, 1619 get_config_bool_default("gdb.wait", false)); 1620 } 1621 } 1622 #endif 1623 1624 vga_init(1); 1625 1626 if (lpc_bootrom()) { 1627 #ifdef __FreeBSD__ 1628 if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) { 1629 fprintf(stderr, "ROM boot failed: unrestricted guest " 1630 "capability not available\n"); 1631 exit(4); 1632 } 1633 #else 1634 /* Unrestricted Guest is always enabled on illumos */ 1635 #endif 1636 error = vcpu_reset(ctx, BSP); 1637 assert(error == 0); 1638 } 1639 1640 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip); 1641 assert(error == 0); 1642 1643 /* 1644 * build the guest tables, MP etc. 1645 */ 1646 if (get_config_bool_default("x86.mptable", true)) { 1647 error = mptable_build(ctx, guest_ncpus); 1648 if (error) { 1649 perror("error to build the guest tables"); 1650 exit(4); 1651 } 1652 } 1653 1654 #ifndef __FreeBSD__ 1655 smbios_apply(); 1656 #endif 1657 error = smbios_build(ctx); 1658 assert(error == 0); 1659 1660 if (get_config_bool("acpi_tables")) { 1661 error = acpi_build(ctx, guest_ncpus); 1662 assert(error == 0); 1663 } 1664 1665 if (lpc_bootrom()) 1666 fwctl_init(); 1667 1668 /* 1669 * Change the proc title to include the VM name. 1670 */ 1671 setproctitle("%s", vmname); 1672 1673 #ifndef WITHOUT_CAPSICUM 1674 caph_cache_catpages(); 1675 1676 if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1) 1677 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1678 1679 if (caph_enter() == -1) 1680 errx(EX_OSERR, "cap_enter() failed"); 1681 #endif 1682 1683 #ifdef __FreeBSD__ 1684 /* 1685 * Add CPU 0 1686 */ 1687 fbsdrun_addcpu(ctx, BSP, BSP, rip); 1688 #else 1689 /* Set BSP to run (unlike the APs which wait for INIT) */ 1690 error = vm_set_run_state(ctx, BSP, VRS_RUN, 0); 1691 assert(error == 0); 1692 fbsdrun_addcpu(ctx, BSP, rip, 1693 get_config_bool_default("suspend_at_boot", false)); 1694 1695 /* Add subsequent CPUs, which will wait until INIT/SIPI-ed */ 1696 for (uint_t i = 1; i < guest_ncpus; i++) { 1697 spinup_halted_ap(ctx, i); 1698 } 1699 #endif 1700 /* 1701 * Head off to the main event dispatch loop 1702 */ 1703 mevent_dispatch(); 1704 1705 exit(4); 1706 } 1707