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 #ifndef __FreeBSD__ 114 #include "privileges.h" 115 #endif 116 117 #define GUEST_NIO_PORT 0x488 /* guest upcalls via i/o port */ 118 119 #define MB (1024UL * 1024) 120 #define GB (1024UL * MB) 121 122 static const char * const vmx_exit_reason_desc[] = { 123 [EXIT_REASON_EXCEPTION] = "Exception or non-maskable interrupt (NMI)", 124 [EXIT_REASON_EXT_INTR] = "External interrupt", 125 [EXIT_REASON_TRIPLE_FAULT] = "Triple fault", 126 [EXIT_REASON_INIT] = "INIT signal", 127 [EXIT_REASON_SIPI] = "Start-up IPI (SIPI)", 128 [EXIT_REASON_IO_SMI] = "I/O system-management interrupt (SMI)", 129 [EXIT_REASON_SMI] = "Other SMI", 130 [EXIT_REASON_INTR_WINDOW] = "Interrupt window", 131 [EXIT_REASON_NMI_WINDOW] = "NMI window", 132 [EXIT_REASON_TASK_SWITCH] = "Task switch", 133 [EXIT_REASON_CPUID] = "CPUID", 134 [EXIT_REASON_GETSEC] = "GETSEC", 135 [EXIT_REASON_HLT] = "HLT", 136 [EXIT_REASON_INVD] = "INVD", 137 [EXIT_REASON_INVLPG] = "INVLPG", 138 [EXIT_REASON_RDPMC] = "RDPMC", 139 [EXIT_REASON_RDTSC] = "RDTSC", 140 [EXIT_REASON_RSM] = "RSM", 141 [EXIT_REASON_VMCALL] = "VMCALL", 142 [EXIT_REASON_VMCLEAR] = "VMCLEAR", 143 [EXIT_REASON_VMLAUNCH] = "VMLAUNCH", 144 [EXIT_REASON_VMPTRLD] = "VMPTRLD", 145 [EXIT_REASON_VMPTRST] = "VMPTRST", 146 [EXIT_REASON_VMREAD] = "VMREAD", 147 [EXIT_REASON_VMRESUME] = "VMRESUME", 148 [EXIT_REASON_VMWRITE] = "VMWRITE", 149 [EXIT_REASON_VMXOFF] = "VMXOFF", 150 [EXIT_REASON_VMXON] = "VMXON", 151 [EXIT_REASON_CR_ACCESS] = "Control-register accesses", 152 [EXIT_REASON_DR_ACCESS] = "MOV DR", 153 [EXIT_REASON_INOUT] = "I/O instruction", 154 [EXIT_REASON_RDMSR] = "RDMSR", 155 [EXIT_REASON_WRMSR] = "WRMSR", 156 [EXIT_REASON_INVAL_VMCS] = 157 "VM-entry failure due to invalid guest state", 158 [EXIT_REASON_INVAL_MSR] = "VM-entry failure due to MSR loading", 159 [EXIT_REASON_MWAIT] = "MWAIT", 160 [EXIT_REASON_MTF] = "Monitor trap flag", 161 [EXIT_REASON_MONITOR] = "MONITOR", 162 [EXIT_REASON_PAUSE] = "PAUSE", 163 [EXIT_REASON_MCE_DURING_ENTRY] = 164 "VM-entry failure due to machine-check event", 165 [EXIT_REASON_TPR] = "TPR below threshold", 166 [EXIT_REASON_APIC_ACCESS] = "APIC access", 167 [EXIT_REASON_VIRTUALIZED_EOI] = "Virtualized EOI", 168 [EXIT_REASON_GDTR_IDTR] = "Access to GDTR or IDTR", 169 [EXIT_REASON_LDTR_TR] = "Access to LDTR or TR", 170 [EXIT_REASON_EPT_FAULT] = "EPT violation", 171 [EXIT_REASON_EPT_MISCONFIG] = "EPT misconfiguration", 172 [EXIT_REASON_INVEPT] = "INVEPT", 173 [EXIT_REASON_RDTSCP] = "RDTSCP", 174 [EXIT_REASON_VMX_PREEMPT] = "VMX-preemption timer expired", 175 [EXIT_REASON_INVVPID] = "INVVPID", 176 [EXIT_REASON_WBINVD] = "WBINVD", 177 [EXIT_REASON_XSETBV] = "XSETBV", 178 [EXIT_REASON_APIC_WRITE] = "APIC write", 179 [EXIT_REASON_RDRAND] = "RDRAND", 180 [EXIT_REASON_INVPCID] = "INVPCID", 181 [EXIT_REASON_VMFUNC] = "VMFUNC", 182 [EXIT_REASON_ENCLS] = "ENCLS", 183 [EXIT_REASON_RDSEED] = "RDSEED", 184 [EXIT_REASON_PM_LOG_FULL] = "Page-modification log full", 185 [EXIT_REASON_XSAVES] = "XSAVES", 186 [EXIT_REASON_XRSTORS] = "XRSTORS" 187 }; 188 189 typedef int (*vmexit_handler_t)(struct vmctx *, struct vm_exit *, int *vcpu); 190 extern int vmexit_task_switch(struct vmctx *, struct vm_exit *, int *vcpu); 191 192 int guest_ncpus; 193 uint16_t cores, maxcpus, sockets, threads; 194 195 int raw_stdio = 0; 196 197 static char *progname; 198 static const int BSP = 0; 199 200 static cpuset_t cpumask; 201 202 static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip); 203 204 static struct vm_exit vmexit[VM_MAXCPU]; 205 static struct vm_entry vmentry[VM_MAXCPU]; 206 207 struct bhyvestats { 208 uint64_t vmexit_bogus; 209 uint64_t vmexit_reqidle; 210 uint64_t vmexit_hlt; 211 uint64_t vmexit_pause; 212 uint64_t vmexit_mtrap; 213 uint64_t vmexit_mmio; 214 uint64_t vmexit_inout; 215 uint64_t cpu_switch_rotate; 216 uint64_t cpu_switch_direct; 217 uint64_t mmio_unhandled; 218 } stats; 219 220 struct mt_vmm_info { 221 pthread_t mt_thr; 222 struct vmctx *mt_ctx; 223 int mt_vcpu; 224 uint64_t mt_startrip; 225 } mt_vmm_info[VM_MAXCPU]; 226 227 #ifdef __FreeBSD__ 228 static cpuset_t *vcpumap[VM_MAXCPU] = { NULL }; 229 #endif 230 231 static void 232 usage(int code) 233 { 234 235 fprintf(stderr, 236 #ifdef __FreeBSD__ 237 "Usage: %s [-aehuwxACDHPSWY]\n" 238 #else 239 "Usage: %s [-adehuwxACDHPSWY]\n" 240 #endif 241 " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n" 242 " %*s [-k <file>] [-l <lpc>] [-m mem] [-o <var>=<value>]\n" 243 #ifdef __FreeBSD__ 244 " %*s [-p vcpu:hostcpu] [-s <pci>] [-U uuid] [<vm>]\n" 245 #else 246 " %*s [-s <pci>] [-U uuid] [<vm>]\n" 247 #endif 248 " -a: local apic is in xAPIC mode (deprecated)\n" 249 " -A: create ACPI tables\n" 250 " -c: number of cpus and/or topology specification\n" 251 " -C: include guest memory in core file\n" 252 #ifndef __FreeBSD__ 253 " -d: suspend cpu at boot\n" 254 #endif 255 " -D: destroy on power-off\n" 256 " -e: exit on unhandled I/O access\n" 257 " -h: help\n" 258 " -H: vmexit from the guest on hlt\n" 259 " -k: key=value flat config file\n" 260 " -l: LPC device configuration\n" 261 " -m: memory size\n" 262 " -o: set config 'var' to 'value'\n" 263 #ifdef __FreeBSD__ 264 " -p: pin 'vcpu' to 'hostcpu'\n" 265 #endif 266 " -P: vmexit from the guest on pause\n" 267 " -s: <slot,driver,configinfo> PCI slot config\n" 268 " -S: guest memory cannot be swapped\n" 269 " -u: RTC keeps UTC time\n" 270 " -U: uuid\n" 271 " -w: ignore unimplemented MSRs\n" 272 " -W: force virtio to use single-vector MSI\n" 273 " -x: local apic is in x2APIC mode\n" 274 " -Y: disable MPtable generation\n", 275 progname, (int)strlen(progname), "", (int)strlen(progname), "", 276 (int)strlen(progname), ""); 277 278 exit(code); 279 } 280 281 /* 282 * XXX This parser is known to have the following issues: 283 * 1. It accepts null key=value tokens ",," as setting "cpus" to an 284 * empty string. 285 * 286 * The acceptance of a null specification ('-c ""') is by design to match the 287 * manual page syntax specification, this results in a topology of 1 vCPU. 288 */ 289 static int 290 topology_parse(const char *opt) 291 { 292 char *cp, *str; 293 294 if (*opt == '\0') { 295 set_config_value("sockets", "1"); 296 set_config_value("cores", "1"); 297 set_config_value("threads", "1"); 298 set_config_value("cpus", "1"); 299 return (0); 300 } 301 302 str = strdup(opt); 303 if (str == NULL) 304 errx(4, "Failed to allocate memory"); 305 306 while ((cp = strsep(&str, ",")) != NULL) { 307 if (strncmp(cp, "cpus=", strlen("cpus=")) == 0) 308 set_config_value("cpus", cp + strlen("cpus=")); 309 else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0) 310 set_config_value("sockets", cp + strlen("sockets=")); 311 else if (strncmp(cp, "cores=", strlen("cores=")) == 0) 312 set_config_value("cores", cp + strlen("cores=")); 313 else if (strncmp(cp, "threads=", strlen("threads=")) == 0) 314 set_config_value("threads", cp + strlen("threads=")); 315 #ifdef notyet /* Do not expose this until vmm.ko implements it */ 316 else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0) 317 set_config_value("maxcpus", cp + strlen("maxcpus=")); 318 #endif 319 else if (strchr(cp, '=') != NULL) 320 goto out; 321 else 322 set_config_value("cpus", cp); 323 } 324 free(str); 325 return (0); 326 327 out: 328 free(str); 329 return (-1); 330 } 331 332 static int 333 parse_int_value(const char *key, const char *value, int minval, int maxval) 334 { 335 char *cp; 336 long lval; 337 338 errno = 0; 339 lval = strtol(value, &cp, 0); 340 if (errno != 0 || *cp != '\0' || cp == value || lval < minval || 341 lval > maxval) 342 errx(4, "Invalid value for %s: '%s'", key, value); 343 return (lval); 344 } 345 346 /* 347 * Set the sockets, cores, threads, and guest_cpus variables based on 348 * the configured topology. 349 * 350 * The limits of UINT16_MAX are due to the types passed to 351 * vm_set_topology(). vmm.ko may enforce tighter limits. 352 */ 353 static void 354 calc_topolopgy(void) 355 { 356 const char *value; 357 bool explicit_cpus; 358 uint64_t ncpus; 359 360 value = get_config_value("cpus"); 361 if (value != NULL) { 362 guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX); 363 explicit_cpus = true; 364 } else { 365 guest_ncpus = 1; 366 explicit_cpus = false; 367 } 368 value = get_config_value("cores"); 369 if (value != NULL) 370 cores = parse_int_value("cores", value, 1, UINT16_MAX); 371 else 372 cores = 1; 373 value = get_config_value("threads"); 374 if (value != NULL) 375 threads = parse_int_value("threads", value, 1, UINT16_MAX); 376 else 377 threads = 1; 378 value = get_config_value("sockets"); 379 if (value != NULL) 380 sockets = parse_int_value("sockets", value, 1, UINT16_MAX); 381 else 382 sockets = guest_ncpus; 383 384 /* 385 * Compute sockets * cores * threads avoiding overflow. The 386 * range check above insures these are 16 bit values. 387 */ 388 ncpus = (uint64_t)sockets * cores * threads; 389 if (ncpus > UINT16_MAX) 390 errx(4, "Computed number of vCPUs too high: %ju", 391 (uintmax_t)ncpus); 392 393 if (explicit_cpus) { 394 if (guest_ncpus != ncpus) 395 errx(4, "Topology (%d sockets, %d cores, %d threads) " 396 "does not match %d vCPUs", sockets, cores, threads, 397 guest_ncpus); 398 } else 399 guest_ncpus = ncpus; 400 } 401 402 #ifndef WITHOUT_CAPSICUM 403 /* 404 * 11-stable capsicum helpers 405 */ 406 static void 407 bhyve_caph_cache_catpages(void) 408 { 409 410 (void)catopen("libc", NL_CAT_LOCALE); 411 } 412 413 static int 414 bhyve_caph_limit_stdoe(void) 415 { 416 cap_rights_t rights; 417 unsigned long cmds[] = { TIOCGETA, TIOCGWINSZ }; 418 int i, fds[] = { STDOUT_FILENO, STDERR_FILENO }; 419 420 cap_rights_init(&rights, CAP_FCNTL, CAP_FSTAT, CAP_IOCTL); 421 cap_rights_set(&rights, CAP_WRITE); 422 423 for (i = 0; i < nitems(fds); i++) { 424 if (cap_rights_limit(fds[i], &rights) < 0 && errno != ENOSYS) 425 return (-1); 426 427 if (cap_ioctls_limit(fds[i], cmds, nitems(cmds)) < 0 && errno != ENOSYS) 428 return (-1); 429 430 if (cap_fcntls_limit(fds[i], CAP_FCNTL_GETFL) < 0 && errno != ENOSYS) 431 return (-1); 432 } 433 434 return (0); 435 } 436 437 #endif 438 439 #ifdef __FreeBSD__ 440 static int 441 pincpu_parse(const char *opt) 442 { 443 int vcpu, pcpu; 444 445 if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) { 446 fprintf(stderr, "invalid format: %s\n", opt); 447 return (-1); 448 } 449 450 if (vcpu < 0 || vcpu >= VM_MAXCPU) { 451 fprintf(stderr, "vcpu '%d' outside valid range from 0 to %d\n", 452 vcpu, VM_MAXCPU - 1); 453 return (-1); 454 } 455 456 if (pcpu < 0 || pcpu >= CPU_SETSIZE) { 457 fprintf(stderr, "hostcpu '%d' outside valid range from " 458 "0 to %d\n", pcpu, CPU_SETSIZE - 1); 459 return (-1); 460 } 461 462 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 463 value = get_config_value(key); 464 465 if (asprintf(&newval, "%s%s%d", value != NULL ? value : "", 466 value != NULL ? "," : "", pcpu) == -1) { 467 perror("failed to build new cpuset string"); 468 return (-1); 469 } 470 471 set_config_value(key, newval); 472 free(newval); 473 return (0); 474 } 475 476 static void 477 parse_cpuset(int vcpu, const char *list, cpuset_t *set) 478 { 479 char *cp, *token; 480 int pcpu, start; 481 482 CPU_ZERO(set); 483 start = -1; 484 token = __DECONST(char *, list); 485 for (;;) { 486 pcpu = strtoul(token, &cp, 0); 487 if (cp == token) 488 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 489 if (pcpu < 0 || pcpu >= CPU_SETSIZE) 490 errx(4, "hostcpu '%d' outside valid range from 0 to %d", 491 pcpu, CPU_SETSIZE - 1); 492 switch (*cp) { 493 case ',': 494 case '\0': 495 if (start >= 0) { 496 if (start > pcpu) 497 errx(4, "Invalid hostcpu range %d-%d", 498 start, pcpu); 499 while (start < pcpu) { 500 CPU_SET(start, vcpumap[vcpu]); 501 start++; 502 } 503 start = -1; 504 } 505 CPU_SET(pcpu, vcpumap[vcpu]); 506 break; 507 case '-': 508 if (start >= 0) 509 errx(4, "invalid cpuset for vcpu %d: '%s'", 510 vcpu, list); 511 start = pcpu; 512 break; 513 default: 514 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 515 } 516 if (*cp == '\0') 517 break; 518 token = cp + 1; 519 } 520 } 521 522 static void 523 build_vcpumaps(void) 524 { 525 char key[16]; 526 const char *value; 527 int vcpu; 528 529 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) { 530 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 531 value = get_config_value(key); 532 if (value == NULL) 533 continue; 534 vcpumap[vcpu] = malloc(sizeof(cpuset_t)); 535 if (vcpumap[vcpu] == NULL) 536 err(4, "Failed to allocate cpuset for vcpu %d", vcpu); 537 parse_cpuset(vcpu, value, vcpumap[vcpu]); 538 } 539 } 540 541 void 542 vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid, 543 int errcode) 544 { 545 struct vmctx *ctx; 546 int error, restart_instruction; 547 548 ctx = arg; 549 restart_instruction = 1; 550 551 error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode, 552 restart_instruction); 553 assert(error == 0); 554 } 555 #endif /* __FreeBSD__ */ 556 557 void * 558 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len) 559 { 560 561 return (vm_map_gpa(ctx, gaddr, len)); 562 } 563 564 int 565 fbsdrun_virtio_msix(void) 566 { 567 568 return (get_config_bool_default("virtio_msix", true)); 569 } 570 571 static void * 572 fbsdrun_start_thread(void *param) 573 { 574 char tname[MAXCOMLEN + 1]; 575 struct mt_vmm_info *mtp; 576 int vcpu; 577 578 mtp = param; 579 vcpu = mtp->mt_vcpu; 580 581 snprintf(tname, sizeof(tname), "vcpu %d", vcpu); 582 pthread_set_name_np(mtp->mt_thr, tname); 583 584 gdb_cpu_add(vcpu); 585 586 vm_loop(mtp->mt_ctx, vcpu, mtp->mt_startrip); 587 588 /* not reached */ 589 exit(1); 590 return (NULL); 591 } 592 593 #ifdef __FreeBSD__ 594 void 595 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip) 596 #else 597 void 598 fbsdrun_addcpu(struct vmctx *ctx, int newcpu, uint64_t rip, bool suspend) 599 #endif 600 { 601 int error; 602 603 #ifdef __FreeBSD__ 604 assert(fromcpu == BSP); 605 #endif 606 607 /* 608 * The 'newcpu' must be activated in the context of 'fromcpu'. If 609 * vm_activate_cpu() is delayed until newcpu's pthread starts running 610 * then vmm.ko is out-of-sync with bhyve and this can create a race 611 * with vm_suspend(). 612 */ 613 error = vm_activate_cpu(ctx, newcpu); 614 if (error != 0) 615 err(EX_OSERR, "could not activate CPU %d", newcpu); 616 617 CPU_SET_ATOMIC(newcpu, &cpumask); 618 619 #ifndef __FreeBSD__ 620 if (suspend) 621 (void) vm_suspend_cpu(ctx, newcpu); 622 #endif 623 624 /* 625 * Set up the vmexit struct to allow execution to start 626 * at the given RIP 627 */ 628 mt_vmm_info[newcpu].mt_ctx = ctx; 629 mt_vmm_info[newcpu].mt_vcpu = newcpu; 630 mt_vmm_info[newcpu].mt_startrip = rip; 631 632 error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL, 633 fbsdrun_start_thread, &mt_vmm_info[newcpu]); 634 assert(error == 0); 635 } 636 637 static int 638 fbsdrun_deletecpu(struct vmctx *ctx, int vcpu) 639 { 640 641 if (!CPU_ISSET(vcpu, &cpumask)) { 642 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu); 643 exit(4); 644 } 645 646 CPU_CLR_ATOMIC(vcpu, &cpumask); 647 return (CPU_EMPTY(&cpumask)); 648 } 649 650 static void 651 vmentry_mmio_read(int vcpu, uint64_t gpa, uint8_t bytes, uint64_t data) 652 { 653 struct vm_entry *entry = &vmentry[vcpu]; 654 struct vm_mmio *mmio = &entry->u.mmio; 655 656 assert(entry->cmd == VEC_DEFAULT); 657 658 entry->cmd = VEC_FULFILL_MMIO; 659 mmio->bytes = bytes; 660 mmio->read = 1; 661 mmio->gpa = gpa; 662 mmio->data = data; 663 } 664 665 static void 666 vmentry_mmio_write(int vcpu, uint64_t gpa, uint8_t bytes) 667 { 668 struct vm_entry *entry = &vmentry[vcpu]; 669 struct vm_mmio *mmio = &entry->u.mmio; 670 671 assert(entry->cmd == VEC_DEFAULT); 672 673 entry->cmd = VEC_FULFILL_MMIO; 674 mmio->bytes = bytes; 675 mmio->read = 0; 676 mmio->gpa = gpa; 677 mmio->data = 0; 678 } 679 680 static void 681 vmentry_inout_read(int vcpu, uint16_t port, uint8_t bytes, uint32_t data) 682 { 683 struct vm_entry *entry = &vmentry[vcpu]; 684 struct vm_inout *inout = &entry->u.inout; 685 686 assert(entry->cmd == VEC_DEFAULT); 687 688 entry->cmd = VEC_FULFILL_INOUT; 689 inout->bytes = bytes; 690 inout->flags = INOUT_IN; 691 inout->port = port; 692 inout->eax = data; 693 } 694 695 static void 696 vmentry_inout_write(int vcpu, uint16_t port, uint8_t bytes) 697 { 698 struct vm_entry *entry = &vmentry[vcpu]; 699 struct vm_inout *inout = &entry->u.inout; 700 701 assert(entry->cmd == VEC_DEFAULT); 702 703 entry->cmd = VEC_FULFILL_INOUT; 704 inout->bytes = bytes; 705 inout->flags = 0; 706 inout->port = port; 707 inout->eax = 0; 708 } 709 710 static int 711 vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu, 712 uint32_t eax) 713 { 714 #if BHYVE_DEBUG 715 /* 716 * put guest-driven debug here 717 */ 718 #endif 719 return (VMEXIT_CONTINUE); 720 } 721 722 static int 723 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 724 { 725 int error; 726 int vcpu; 727 struct vm_inout inout; 728 bool in; 729 uint8_t bytes; 730 731 stats.vmexit_inout++; 732 733 vcpu = *pvcpu; 734 inout = vme->u.inout; 735 in = (inout.flags & INOUT_IN) != 0; 736 bytes = inout.bytes; 737 738 /* Extra-special case of host notifications */ 739 if (!in && inout.port == GUEST_NIO_PORT) { 740 error = vmexit_handle_notify(ctx, vme, pvcpu, inout.eax); 741 vmentry_inout_write(vcpu, inout.port, bytes); 742 return (error); 743 } 744 745 error = emulate_inout(ctx, vcpu, &inout); 746 if (error) { 747 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n", 748 in ? "in" : "out", 749 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'), 750 inout.port, vmexit->rip); 751 return (VMEXIT_ABORT); 752 } else { 753 /* 754 * Communicate the status of the inout operation back to the 755 * in-kernel instruction emulation. 756 */ 757 if (in) { 758 vmentry_inout_read(vcpu, inout.port, bytes, inout.eax); 759 } else { 760 vmentry_inout_write(vcpu, inout.port, bytes); 761 } 762 return (VMEXIT_CONTINUE); 763 } 764 } 765 766 static int 767 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 768 { 769 uint64_t val; 770 uint32_t eax, edx; 771 int error; 772 773 val = 0; 774 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val); 775 if (error != 0) { 776 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n", 777 vme->u.msr.code, *pvcpu); 778 if (get_config_bool("x86.strictmsr")) { 779 vm_inject_gp(ctx, *pvcpu); 780 return (VMEXIT_CONTINUE); 781 } 782 } 783 784 eax = val; 785 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax); 786 assert(error == 0); 787 788 edx = val >> 32; 789 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx); 790 assert(error == 0); 791 792 return (VMEXIT_CONTINUE); 793 } 794 795 static int 796 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 797 { 798 int error; 799 800 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval); 801 if (error != 0) { 802 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n", 803 vme->u.msr.code, vme->u.msr.wval, *pvcpu); 804 if (get_config_bool("x86.strictmsr")) { 805 vm_inject_gp(ctx, *pvcpu); 806 return (VMEXIT_CONTINUE); 807 } 808 } 809 return (VMEXIT_CONTINUE); 810 } 811 812 #ifdef __FreeBSD__ 813 static int 814 vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 815 { 816 817 (void)spinup_ap(ctx, *pvcpu, 818 vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip); 819 820 return (VMEXIT_CONTINUE); 821 } 822 #else 823 static int 824 vmexit_run_state(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 825 { 826 /* 827 * Run-state transitions (INIT, SIPI, etc) are handled in-kernel, so an 828 * exit to userspace with that code is not expected. 829 */ 830 fprintf(stderr, "unexpected run-state VM exit"); 831 return (VMEXIT_ABORT); 832 } 833 #endif /* __FreeBSD__ */ 834 835 #ifdef __FreeBSD__ 836 #define DEBUG_EPT_MISCONFIG 837 #else 838 /* EPT misconfig debugging not possible now that raw VMCS access is gone */ 839 #endif 840 841 #ifdef DEBUG_EPT_MISCONFIG 842 #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400 843 844 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4]; 845 static int ept_misconfig_ptenum; 846 #endif 847 848 static const char * 849 vmexit_vmx_desc(uint32_t exit_reason) 850 { 851 852 if (exit_reason >= nitems(vmx_exit_reason_desc) || 853 vmx_exit_reason_desc[exit_reason] == NULL) 854 return ("Unknown"); 855 return (vmx_exit_reason_desc[exit_reason]); 856 } 857 858 static int 859 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 860 { 861 862 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 863 fprintf(stderr, "\treason\t\tVMX\n"); 864 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 865 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 866 fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status); 867 fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason, 868 vmexit_vmx_desc(vmexit->u.vmx.exit_reason)); 869 fprintf(stderr, "\tqualification\t0x%016lx\n", 870 vmexit->u.vmx.exit_qualification); 871 fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type); 872 fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error); 873 #ifdef DEBUG_EPT_MISCONFIG 874 if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) { 875 vm_get_register(ctx, *pvcpu, 876 VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS), 877 &ept_misconfig_gpa); 878 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte, 879 &ept_misconfig_ptenum); 880 fprintf(stderr, "\tEPT misconfiguration:\n"); 881 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa); 882 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n", 883 ept_misconfig_ptenum, ept_misconfig_pte[0], 884 ept_misconfig_pte[1], ept_misconfig_pte[2], 885 ept_misconfig_pte[3]); 886 } 887 #endif /* DEBUG_EPT_MISCONFIG */ 888 return (VMEXIT_ABORT); 889 } 890 891 static int 892 vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 893 { 894 895 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 896 fprintf(stderr, "\treason\t\tSVM\n"); 897 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 898 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 899 fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode); 900 fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1); 901 fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2); 902 return (VMEXIT_ABORT); 903 } 904 905 static int 906 vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 907 { 908 909 assert(vmexit->inst_length == 0); 910 911 stats.vmexit_bogus++; 912 913 return (VMEXIT_CONTINUE); 914 } 915 916 static int 917 vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 918 { 919 920 assert(vmexit->inst_length == 0); 921 922 stats.vmexit_reqidle++; 923 924 return (VMEXIT_CONTINUE); 925 } 926 927 static int 928 vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 929 { 930 931 stats.vmexit_hlt++; 932 933 /* 934 * Just continue execution with the next instruction. We use 935 * the HLT VM exit as a way to be friendly with the host 936 * scheduler. 937 */ 938 return (VMEXIT_CONTINUE); 939 } 940 941 static int 942 vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 943 { 944 945 stats.vmexit_pause++; 946 947 return (VMEXIT_CONTINUE); 948 } 949 950 static int 951 vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 952 { 953 954 assert(vmexit->inst_length == 0); 955 956 stats.vmexit_mtrap++; 957 958 gdb_cpu_mtrap(*pvcpu); 959 960 return (VMEXIT_CONTINUE); 961 } 962 963 static int 964 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 965 { 966 uint8_t i, valid; 967 968 fprintf(stderr, "Failed to emulate instruction sequence "); 969 970 valid = vmexit->u.inst_emul.num_valid; 971 if (valid != 0) { 972 assert(valid <= sizeof (vmexit->u.inst_emul.inst)); 973 fprintf(stderr, "["); 974 for (i = 0; i < valid; i++) { 975 if (i == 0) { 976 fprintf(stderr, "%02x", 977 vmexit->u.inst_emul.inst[i]); 978 } else { 979 fprintf(stderr, ", %02x", 980 vmexit->u.inst_emul.inst[i]); 981 } 982 } 983 fprintf(stderr, "] "); 984 } 985 fprintf(stderr, "@ %rip = %x\n", vmexit->rip); 986 987 return (VMEXIT_ABORT); 988 } 989 990 static int 991 vmexit_mmio(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 992 { 993 int vcpu, err; 994 struct vm_mmio mmio; 995 bool is_read; 996 997 stats.vmexit_mmio++; 998 999 vcpu = *pvcpu; 1000 mmio = vmexit->u.mmio; 1001 is_read = (mmio.read != 0); 1002 1003 err = emulate_mem(ctx, vcpu, &mmio); 1004 1005 if (err == ESRCH) { 1006 fprintf(stderr, "Unhandled memory access to 0x%lx\n", mmio.gpa); 1007 stats.mmio_unhandled++; 1008 1009 /* 1010 * Access to non-existent physical addresses is not likely to 1011 * result in fatal errors on hardware machines, but rather reads 1012 * of all-ones or discarded-but-acknowledged writes. 1013 */ 1014 mmio.data = ~0UL; 1015 err = 0; 1016 } 1017 1018 if (err == 0) { 1019 if (is_read) { 1020 vmentry_mmio_read(vcpu, mmio.gpa, mmio.bytes, 1021 mmio.data); 1022 } else { 1023 vmentry_mmio_write(vcpu, mmio.gpa, mmio.bytes); 1024 } 1025 return (VMEXIT_CONTINUE); 1026 } 1027 1028 fprintf(stderr, "Unhandled mmio error to 0x%lx: %d\n", mmio.gpa, err); 1029 return (VMEXIT_ABORT); 1030 } 1031 1032 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER; 1033 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER; 1034 1035 static int 1036 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1037 { 1038 enum vm_suspend_how how; 1039 1040 how = vmexit->u.suspended.how; 1041 1042 fbsdrun_deletecpu(ctx, *pvcpu); 1043 1044 if (*pvcpu != BSP) { 1045 pthread_mutex_lock(&resetcpu_mtx); 1046 pthread_cond_signal(&resetcpu_cond); 1047 pthread_mutex_unlock(&resetcpu_mtx); 1048 pthread_exit(NULL); 1049 } 1050 1051 pthread_mutex_lock(&resetcpu_mtx); 1052 while (!CPU_EMPTY(&cpumask)) { 1053 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx); 1054 } 1055 pthread_mutex_unlock(&resetcpu_mtx); 1056 1057 switch (how) { 1058 case VM_SUSPEND_RESET: 1059 exit(0); 1060 case VM_SUSPEND_POWEROFF: 1061 if (get_config_bool_default("destroy_on_poweroff", false)) 1062 vm_destroy(ctx); 1063 exit(1); 1064 case VM_SUSPEND_HALT: 1065 exit(2); 1066 case VM_SUSPEND_TRIPLEFAULT: 1067 exit(3); 1068 default: 1069 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how); 1070 exit(100); 1071 } 1072 return (0); /* NOTREACHED */ 1073 } 1074 1075 static int 1076 vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1077 { 1078 1079 gdb_cpu_suspend(*pvcpu); 1080 return (VMEXIT_CONTINUE); 1081 } 1082 1083 static int 1084 vmexit_breakpoint(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1085 { 1086 1087 gdb_cpu_breakpoint(*pvcpu, vmexit); 1088 return (VMEXIT_CONTINUE); 1089 } 1090 1091 static vmexit_handler_t handler[VM_EXITCODE_MAX] = { 1092 [VM_EXITCODE_INOUT] = vmexit_inout, 1093 [VM_EXITCODE_MMIO] = vmexit_mmio, 1094 [VM_EXITCODE_VMX] = vmexit_vmx, 1095 [VM_EXITCODE_SVM] = vmexit_svm, 1096 [VM_EXITCODE_BOGUS] = vmexit_bogus, 1097 [VM_EXITCODE_REQIDLE] = vmexit_reqidle, 1098 [VM_EXITCODE_RDMSR] = vmexit_rdmsr, 1099 [VM_EXITCODE_WRMSR] = vmexit_wrmsr, 1100 [VM_EXITCODE_MTRAP] = vmexit_mtrap, 1101 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul, 1102 #ifdef __FreeBSD__ 1103 [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap, 1104 #else 1105 [VM_EXITCODE_RUN_STATE] = vmexit_run_state, 1106 #endif 1107 [VM_EXITCODE_SUSPENDED] = vmexit_suspend, 1108 [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch, 1109 [VM_EXITCODE_DEBUG] = vmexit_debug, 1110 [VM_EXITCODE_BPT] = vmexit_breakpoint, 1111 }; 1112 1113 static void 1114 vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip) 1115 { 1116 int error, rc; 1117 enum vm_exitcode exitcode; 1118 cpuset_t active_cpus; 1119 struct vm_exit *vexit; 1120 struct vm_entry *ventry; 1121 1122 #ifdef __FreeBSD__ 1123 if (vcpumap[vcpu] != NULL) { 1124 error = pthread_setaffinity_np(pthread_self(), 1125 sizeof(cpuset_t), vcpumap[vcpu]); 1126 assert(error == 0); 1127 } 1128 #endif 1129 error = vm_active_cpus(ctx, &active_cpus); 1130 assert(CPU_ISSET(vcpu, &active_cpus)); 1131 1132 error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip); 1133 assert(error == 0); 1134 1135 ventry = &vmentry[vcpu]; 1136 vexit = &vmexit[vcpu]; 1137 1138 while (1) { 1139 error = vm_run(ctx, vcpu, ventry, vexit); 1140 if (error != 0) 1141 break; 1142 1143 if (ventry->cmd != VEC_DEFAULT) { 1144 /* 1145 * Discard any lingering entry state after it has been 1146 * submitted via vm_run(). 1147 */ 1148 bzero(ventry, sizeof (*ventry)); 1149 } 1150 1151 exitcode = vexit->exitcode; 1152 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) { 1153 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n", 1154 exitcode); 1155 exit(4); 1156 } 1157 1158 rc = (*handler[exitcode])(ctx, vexit, &vcpu); 1159 1160 switch (rc) { 1161 case VMEXIT_CONTINUE: 1162 break; 1163 case VMEXIT_ABORT: 1164 abort(); 1165 default: 1166 exit(4); 1167 } 1168 } 1169 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno); 1170 } 1171 1172 static int 1173 num_vcpus_allowed(struct vmctx *ctx) 1174 { 1175 #ifdef __FreeBSD__ 1176 int tmp, error; 1177 1178 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp); 1179 1180 /* 1181 * The guest is allowed to spinup more than one processor only if the 1182 * UNRESTRICTED_GUEST capability is available. 1183 */ 1184 if (error == 0) 1185 return (VM_MAXCPU); 1186 else 1187 return (1); 1188 #else 1189 /* Unrestricted Guest is always enabled on illumos */ 1190 return (VM_MAXCPU); 1191 #endif /* __FreeBSD__ */ 1192 } 1193 1194 void 1195 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu) 1196 { 1197 int err, tmp; 1198 1199 if (get_config_bool_default("x86.vmexit_on_hlt", false)) { 1200 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp); 1201 if (err < 0) { 1202 fprintf(stderr, "VM exit on HLT not supported\n"); 1203 exit(4); 1204 } 1205 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1); 1206 if (cpu == BSP) 1207 handler[VM_EXITCODE_HLT] = vmexit_hlt; 1208 } 1209 1210 if (get_config_bool_default("x86.vmexit_on_pause", false)) { 1211 /* 1212 * pause exit support required for this mode 1213 */ 1214 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp); 1215 if (err < 0) { 1216 fprintf(stderr, 1217 "SMP mux requested, no pause support\n"); 1218 exit(4); 1219 } 1220 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1); 1221 if (cpu == BSP) 1222 handler[VM_EXITCODE_PAUSE] = vmexit_pause; 1223 } 1224 1225 if (get_config_bool_default("x86.x2apic", false)) 1226 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED); 1227 else 1228 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED); 1229 1230 if (err) { 1231 fprintf(stderr, "Unable to set x2apic state (%d)\n", err); 1232 exit(4); 1233 } 1234 1235 #ifdef __FreeBSD__ 1236 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1); 1237 #endif 1238 } 1239 1240 static struct vmctx * 1241 do_open(const char *vmname) 1242 { 1243 struct vmctx *ctx; 1244 int error; 1245 bool reinit, romboot; 1246 #ifndef WITHOUT_CAPSICUM 1247 cap_rights_t rights; 1248 const cap_ioctl_t *cmds; 1249 size_t ncmds; 1250 #endif 1251 1252 reinit = romboot = false; 1253 1254 if (lpc_bootrom()) 1255 romboot = true; 1256 1257 error = vm_create(vmname); 1258 if (error) { 1259 if (errno == EEXIST) { 1260 if (romboot) { 1261 reinit = true; 1262 } else { 1263 /* 1264 * The virtual machine has been setup by the 1265 * userspace bootloader. 1266 */ 1267 } 1268 } else { 1269 perror("vm_create"); 1270 exit(4); 1271 } 1272 } else { 1273 if (!romboot) { 1274 /* 1275 * If the virtual machine was just created then a 1276 * bootrom must be configured to boot it. 1277 */ 1278 fprintf(stderr, "virtual machine cannot be booted\n"); 1279 exit(4); 1280 } 1281 } 1282 1283 ctx = vm_open(vmname); 1284 if (ctx == NULL) { 1285 perror("vm_open"); 1286 exit(4); 1287 } 1288 1289 #ifndef WITHOUT_CAPSICUM 1290 cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW); 1291 if (caph_rights_limit(vm_get_device_fd(ctx), &rights) == -1) 1292 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1293 vm_get_ioctls(&ncmds); 1294 cmds = vm_get_ioctls(NULL); 1295 if (cmds == NULL) 1296 errx(EX_OSERR, "out of memory"); 1297 if (caph_ioctls_limit(vm_get_device_fd(ctx), cmds, ncmds) == -1) 1298 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1299 free((cap_ioctl_t *)cmds); 1300 #endif 1301 1302 if (reinit) { 1303 error = vm_reinit(ctx); 1304 if (error) { 1305 perror("vm_reinit"); 1306 exit(4); 1307 } 1308 } 1309 error = vm_set_topology(ctx, sockets, cores, threads, maxcpus); 1310 if (error) 1311 errx(EX_OSERR, "vm_set_topology"); 1312 return (ctx); 1313 } 1314 1315 static bool 1316 parse_config_option(const char *option) 1317 { 1318 const char *value; 1319 char *path; 1320 1321 value = strchr(option, '='); 1322 if (value == NULL || value[1] == '\0') 1323 return (false); 1324 path = strndup(option, value - option); 1325 if (path == NULL) 1326 err(4, "Failed to allocate memory"); 1327 set_config_value(path, value + 1); 1328 return (true); 1329 } 1330 1331 static void 1332 parse_simple_config_file(const char *path) 1333 { 1334 FILE *fp; 1335 char *line, *cp; 1336 size_t linecap; 1337 unsigned int lineno; 1338 1339 fp = fopen(path, "r"); 1340 if (fp == NULL) 1341 err(4, "Failed to open configuration file %s", path); 1342 line = NULL; 1343 linecap = 0; 1344 lineno = 1; 1345 for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) { 1346 if (*line == '#' || *line == '\n') 1347 continue; 1348 cp = strchr(line, '\n'); 1349 if (cp != NULL) 1350 *cp = '\0'; 1351 if (!parse_config_option(line)) 1352 errx(4, "%s line %u: invalid config option '%s'", path, 1353 lineno, line); 1354 } 1355 free(line); 1356 fclose(fp); 1357 } 1358 1359 static void 1360 set_defaults(void) 1361 { 1362 1363 set_config_bool("acpi_tables", false); 1364 set_config_value("memory.size", "256M"); 1365 set_config_bool("x86.strictmsr", true); 1366 } 1367 1368 int 1369 main(int argc, char *argv[]) 1370 { 1371 int c, error, err; 1372 int max_vcpus, memflags; 1373 struct vmctx *ctx; 1374 uint64_t rip; 1375 size_t memsize; 1376 const char *value, *vmname; 1377 char *optstr; 1378 1379 init_config(); 1380 set_defaults(); 1381 progname = basename(argv[0]); 1382 1383 #ifdef __FreeBSD__ 1384 optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:U:"; 1385 #else 1386 /* +d, +B, -p */ 1387 optstr = "adehuwxACDHIPSWYk:o:G:c:s:m:l:B:U:"; 1388 #endif 1389 while ((c = getopt(argc, argv, optstr)) != -1) { 1390 switch (c) { 1391 case 'a': 1392 set_config_bool("x86.x2apic", false); 1393 break; 1394 case 'A': 1395 set_config_bool("acpi_tables", true); 1396 break; 1397 case 'D': 1398 set_config_bool("destroy_on_poweroff", true); 1399 break; 1400 #ifndef __FreeBSD__ 1401 case 'B': 1402 if (smbios_parse(optarg) != 0) { 1403 errx(EX_USAGE, "invalid SMBIOS " 1404 "configuration '%s'", optarg); 1405 } 1406 break; 1407 case 'd': 1408 set_config_bool("suspend_at_boot", true); 1409 break; 1410 #endif 1411 #ifdef __FreeBSD__ 1412 case 'p': 1413 if (pincpu_parse(optarg) != 0) { 1414 errx(EX_USAGE, "invalid vcpu pinning " 1415 "configuration '%s'", optarg); 1416 } 1417 break; 1418 #endif 1419 case 'c': 1420 if (topology_parse(optarg) != 0) { 1421 errx(EX_USAGE, "invalid cpu topology " 1422 "'%s'", optarg); 1423 } 1424 break; 1425 case 'C': 1426 set_config_bool("memory.guest_in_core", true); 1427 break; 1428 case 'G': 1429 if (optarg[0] == 'w') { 1430 set_config_bool("gdb.wait", true); 1431 optarg++; 1432 } 1433 set_config_value("gdb.port", optarg); 1434 break; 1435 case 'k': 1436 parse_simple_config_file(optarg); 1437 break; 1438 case 'l': 1439 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1440 lpc_print_supported_devices(); 1441 exit(0); 1442 } else if (lpc_device_parse(optarg) != 0) { 1443 errx(EX_USAGE, "invalid lpc device " 1444 "configuration '%s'", optarg); 1445 } 1446 break; 1447 case 's': 1448 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1449 pci_print_supported_devices(); 1450 exit(0); 1451 } else if (pci_parse_slot(optarg) != 0) 1452 exit(4); 1453 else 1454 break; 1455 case 'S': 1456 set_config_bool("memory.wired", true); 1457 break; 1458 case 'm': 1459 set_config_value("memory.size", optarg); 1460 break; 1461 case 'o': 1462 if (!parse_config_option(optarg)) 1463 errx(EX_USAGE, "invalid configuration option '%s'", optarg); 1464 break; 1465 case 'H': 1466 set_config_bool("x86.vmexit_on_hlt", true); 1467 break; 1468 case 'I': 1469 /* 1470 * The "-I" option was used to add an ioapic to the 1471 * virtual machine. 1472 * 1473 * An ioapic is now provided unconditionally for each 1474 * virtual machine and this option is now deprecated. 1475 */ 1476 break; 1477 case 'P': 1478 set_config_bool("x86.vmexit_on_pause", true); 1479 break; 1480 case 'e': 1481 set_config_bool("x86.strictio", true); 1482 break; 1483 case 'u': 1484 set_config_bool("rtc.use_localtime", false); 1485 break; 1486 case 'U': 1487 set_config_value("uuid", optarg); 1488 break; 1489 case 'w': 1490 set_config_bool("x86.strictmsr", false); 1491 break; 1492 case 'W': 1493 set_config_bool("virtio_msix", false); 1494 break; 1495 case 'x': 1496 set_config_bool("x86.x2apic", true); 1497 break; 1498 case 'Y': 1499 set_config_bool("x86.mptable", false); 1500 break; 1501 case 'h': 1502 usage(0); 1503 default: 1504 usage(1); 1505 } 1506 } 1507 argc -= optind; 1508 argv += optind; 1509 1510 if (argc > 1) 1511 usage(1); 1512 1513 if (argc == 1) 1514 set_config_value("name", argv[0]); 1515 1516 vmname = get_config_value("name"); 1517 if (vmname == NULL) 1518 usage(1); 1519 1520 if (get_config_bool_default("config.dump", false)) { 1521 dump_config(); 1522 exit(1); 1523 } 1524 1525 #ifndef __FreeBSD__ 1526 illumos_priv_init(); 1527 #endif 1528 1529 calc_topolopgy(); 1530 #ifdef __FreeBSD__ 1531 build_vcpumaps(); 1532 #endif 1533 1534 value = get_config_value("memory.size"); 1535 error = vm_parse_memsize(value, &memsize); 1536 if (error) 1537 errx(EX_USAGE, "invalid memsize '%s'", value); 1538 1539 ctx = do_open(vmname); 1540 1541 max_vcpus = num_vcpus_allowed(ctx); 1542 if (guest_ncpus > max_vcpus) { 1543 fprintf(stderr, "%d vCPUs requested but only %d available\n", 1544 guest_ncpus, max_vcpus); 1545 exit(4); 1546 } 1547 1548 fbsdrun_set_capabilities(ctx, BSP); 1549 1550 memflags = 0; 1551 if (get_config_bool_default("memory.wired", false)) 1552 memflags |= VM_MEM_F_WIRED; 1553 if (get_config_bool_default("memory.guest_in_core", false)) 1554 memflags |= VM_MEM_F_INCORE; 1555 vm_set_memflags(ctx, memflags); 1556 #ifdef __FreeBSD__ 1557 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1558 #else 1559 do { 1560 errno = 0; 1561 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1562 error = errno; 1563 if (err != 0 && error == ENOMEM) { 1564 (void) fprintf(stderr, "Unable to allocate memory " 1565 "(%llu), retrying in 1 second\n", memsize); 1566 sleep(1); 1567 } 1568 } while (error == ENOMEM); 1569 #endif 1570 if (err) { 1571 fprintf(stderr, "Unable to set up memory (%d)\n", errno); 1572 exit(4); 1573 } 1574 1575 error = init_msr(); 1576 if (error) { 1577 fprintf(stderr, "init_msr error %d", error); 1578 exit(4); 1579 } 1580 1581 init_mem(); 1582 init_inout(); 1583 #ifdef __FreeBSD__ 1584 kernemu_dev_init(); 1585 #endif 1586 init_bootrom(ctx); 1587 atkbdc_init(ctx); 1588 pci_irq_init(ctx); 1589 ioapic_init(ctx); 1590 1591 rtc_init(ctx); 1592 sci_init(ctx); 1593 #ifndef __FreeBSD__ 1594 pmtmr_init(ctx); 1595 #endif 1596 1597 /* 1598 * Exit if a device emulation finds an error in its initilization 1599 */ 1600 if (init_pci(ctx) != 0) { 1601 perror("device emulation initialization error"); 1602 exit(4); 1603 } 1604 1605 /* 1606 * Initialize after PCI, to allow a bootrom file to reserve the high 1607 * region. 1608 */ 1609 if (get_config_bool("acpi_tables")) 1610 vmgenc_init(ctx); 1611 1612 value = get_config_value("gdb.port"); 1613 #ifdef __FreeBSD__ 1614 if (value != NULL) 1615 init_gdb(ctx, atoi(value), get_config_bool_default("gdb.wait", 1616 false)); 1617 #else 1618 if (value != NULL) { 1619 int port = atoi(value); 1620 1621 if (port < 0) { 1622 init_mdb(ctx, 1623 get_config_bool_default("gdb.wait", false)); 1624 } else { 1625 init_gdb(ctx, port, 1626 get_config_bool_default("gdb.wait", false)); 1627 } 1628 } 1629 #endif 1630 1631 vga_init(1); 1632 1633 if (lpc_bootrom()) { 1634 #ifdef __FreeBSD__ 1635 if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) { 1636 fprintf(stderr, "ROM boot failed: unrestricted guest " 1637 "capability not available\n"); 1638 exit(4); 1639 } 1640 #else 1641 /* Unrestricted Guest is always enabled on illumos */ 1642 #endif 1643 error = vcpu_reset(ctx, BSP); 1644 assert(error == 0); 1645 } 1646 1647 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip); 1648 assert(error == 0); 1649 1650 /* 1651 * build the guest tables, MP etc. 1652 */ 1653 if (get_config_bool_default("x86.mptable", true)) { 1654 error = mptable_build(ctx, guest_ncpus); 1655 if (error) { 1656 perror("error to build the guest tables"); 1657 exit(4); 1658 } 1659 } 1660 1661 #ifndef __FreeBSD__ 1662 smbios_apply(); 1663 #endif 1664 error = smbios_build(ctx); 1665 assert(error == 0); 1666 1667 if (get_config_bool("acpi_tables")) { 1668 error = acpi_build(ctx, guest_ncpus); 1669 assert(error == 0); 1670 } 1671 1672 if (lpc_bootrom()) 1673 fwctl_init(); 1674 1675 /* 1676 * Change the proc title to include the VM name. 1677 */ 1678 setproctitle("%s", vmname); 1679 1680 #ifndef WITHOUT_CAPSICUM 1681 caph_cache_catpages(); 1682 1683 if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1) 1684 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1685 1686 if (caph_enter() == -1) 1687 errx(EX_OSERR, "cap_enter() failed"); 1688 #endif 1689 1690 #ifndef __FreeBSD__ 1691 illumos_priv_lock(); 1692 #endif 1693 1694 #ifdef __FreeBSD__ 1695 /* 1696 * Add CPU 0 1697 */ 1698 fbsdrun_addcpu(ctx, BSP, BSP, rip); 1699 #else 1700 /* Set BSP to run (unlike the APs which wait for INIT) */ 1701 error = vm_set_run_state(ctx, BSP, VRS_RUN, 0); 1702 assert(error == 0); 1703 fbsdrun_addcpu(ctx, BSP, rip, 1704 get_config_bool_default("suspend_at_boot", false)); 1705 1706 /* Add subsequent CPUs, which will wait until INIT/SIPI-ed */ 1707 for (uint_t i = 1; i < guest_ncpus; i++) { 1708 spinup_halted_ap(ctx, i); 1709 } 1710 #endif 1711 /* 1712 * Head off to the main event dispatch loop 1713 */ 1714 mevent_dispatch(); 1715 1716 exit(4); 1717 } 1718