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 2021 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 [-AaCDeHhPSuWwxY]\n" 238 #else 239 "Usage: %s [-AaCDdeHhPSuWwxY]\n" 240 #endif 241 " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n" 242 #ifdef __FreeBSD__ 243 " %*s [-G port] [-k file] [-l lpc] [-m mem] [-o var=value]\n" 244 " %*s [-p vcpu:hostcpu] [-r file] [-s pci] [-U uuid] vmname\n" 245 246 #else 247 " %*s [-k <file>] [-l <lpc>] [-m mem] [-o <var>=<value>]\n" 248 " %*s [-s <pci>] [-U uuid] vmname\n" 249 #endif 250 " -A: create ACPI tables\n" 251 " -a: local apic is in xAPIC mode (deprecated)\n" 252 " -C: include guest memory in core file\n" 253 " -c: number of cpus and/or topology specification\n" 254 " -D: destroy on power-off\n" 255 #ifndef __FreeBSD__ 256 " -d: suspend cpu at boot\n" 257 #endif 258 " -e: exit on unhandled I/O access\n" 259 " -H: vmexit from the guest on hlt\n" 260 " -h: help\n" 261 " -k: key=value flat config file\n" 262 " -l: LPC device configuration\n" 263 " -m: memory size\n" 264 " -o: set config 'var' to 'value'\n" 265 " -P: vmexit from the guest on pause\n" 266 #ifdef __FreeBSD__ 267 " -p: pin 'vcpu' to 'hostcpu'\n" 268 #endif 269 " -S: guest memory cannot be swapped\n" 270 " -s: <slot,driver,configinfo> PCI slot config\n" 271 " -U: uuid\n" 272 " -u: RTC keeps UTC time\n" 273 " -W: force virtio to use single-vector MSI\n" 274 " -w: ignore unimplemented MSRs\n" 275 " -x: local apic is in x2APIC mode\n" 276 " -Y: disable MPtable generation\n", 277 progname, (int)strlen(progname), "", (int)strlen(progname), "", 278 (int)strlen(progname), ""); 279 280 exit(code); 281 } 282 283 /* 284 * XXX This parser is known to have the following issues: 285 * 1. It accepts null key=value tokens ",," as setting "cpus" to an 286 * empty string. 287 * 288 * The acceptance of a null specification ('-c ""') is by design to match the 289 * manual page syntax specification, this results in a topology of 1 vCPU. 290 */ 291 static int 292 topology_parse(const char *opt) 293 { 294 char *cp, *str; 295 296 if (*opt == '\0') { 297 set_config_value("sockets", "1"); 298 set_config_value("cores", "1"); 299 set_config_value("threads", "1"); 300 set_config_value("cpus", "1"); 301 return (0); 302 } 303 304 str = strdup(opt); 305 if (str == NULL) 306 errx(4, "Failed to allocate memory"); 307 308 while ((cp = strsep(&str, ",")) != NULL) { 309 if (strncmp(cp, "cpus=", strlen("cpus=")) == 0) 310 set_config_value("cpus", cp + strlen("cpus=")); 311 else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0) 312 set_config_value("sockets", cp + strlen("sockets=")); 313 else if (strncmp(cp, "cores=", strlen("cores=")) == 0) 314 set_config_value("cores", cp + strlen("cores=")); 315 else if (strncmp(cp, "threads=", strlen("threads=")) == 0) 316 set_config_value("threads", cp + strlen("threads=")); 317 #ifdef notyet /* Do not expose this until vmm.ko implements it */ 318 else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0) 319 set_config_value("maxcpus", cp + strlen("maxcpus=")); 320 #endif 321 else if (strchr(cp, '=') != NULL) 322 goto out; 323 else 324 set_config_value("cpus", cp); 325 } 326 free(str); 327 return (0); 328 329 out: 330 free(str); 331 return (-1); 332 } 333 334 static int 335 parse_int_value(const char *key, const char *value, int minval, int maxval) 336 { 337 char *cp; 338 long lval; 339 340 errno = 0; 341 lval = strtol(value, &cp, 0); 342 if (errno != 0 || *cp != '\0' || cp == value || lval < minval || 343 lval > maxval) 344 errx(4, "Invalid value for %s: '%s'", key, value); 345 return (lval); 346 } 347 348 /* 349 * Set the sockets, cores, threads, and guest_cpus variables based on 350 * the configured topology. 351 * 352 * The limits of UINT16_MAX are due to the types passed to 353 * vm_set_topology(). vmm.ko may enforce tighter limits. 354 */ 355 static void 356 calc_topolopgy(void) 357 { 358 const char *value; 359 bool explicit_cpus; 360 uint64_t ncpus; 361 362 value = get_config_value("cpus"); 363 if (value != NULL) { 364 guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX); 365 explicit_cpus = true; 366 } else { 367 guest_ncpus = 1; 368 explicit_cpus = false; 369 } 370 value = get_config_value("cores"); 371 if (value != NULL) 372 cores = parse_int_value("cores", value, 1, UINT16_MAX); 373 else 374 cores = 1; 375 value = get_config_value("threads"); 376 if (value != NULL) 377 threads = parse_int_value("threads", value, 1, UINT16_MAX); 378 else 379 threads = 1; 380 value = get_config_value("sockets"); 381 if (value != NULL) 382 sockets = parse_int_value("sockets", value, 1, UINT16_MAX); 383 else 384 sockets = guest_ncpus; 385 386 /* 387 * Compute sockets * cores * threads avoiding overflow. The 388 * range check above insures these are 16 bit values. 389 */ 390 ncpus = (uint64_t)sockets * cores * threads; 391 if (ncpus > UINT16_MAX) 392 errx(4, "Computed number of vCPUs too high: %ju", 393 (uintmax_t)ncpus); 394 395 if (explicit_cpus) { 396 if (guest_ncpus != ncpus) 397 errx(4, "Topology (%d sockets, %d cores, %d threads) " 398 "does not match %d vCPUs", sockets, cores, threads, 399 guest_ncpus); 400 } else 401 guest_ncpus = ncpus; 402 } 403 404 #ifndef WITHOUT_CAPSICUM 405 /* 406 * 11-stable capsicum helpers 407 */ 408 static void 409 bhyve_caph_cache_catpages(void) 410 { 411 412 (void)catopen("libc", NL_CAT_LOCALE); 413 } 414 415 static int 416 bhyve_caph_limit_stdoe(void) 417 { 418 cap_rights_t rights; 419 unsigned long cmds[] = { TIOCGETA, TIOCGWINSZ }; 420 int i, fds[] = { STDOUT_FILENO, STDERR_FILENO }; 421 422 cap_rights_init(&rights, CAP_FCNTL, CAP_FSTAT, CAP_IOCTL); 423 cap_rights_set(&rights, CAP_WRITE); 424 425 for (i = 0; i < nitems(fds); i++) { 426 if (cap_rights_limit(fds[i], &rights) < 0 && errno != ENOSYS) 427 return (-1); 428 429 if (cap_ioctls_limit(fds[i], cmds, nitems(cmds)) < 0 && errno != ENOSYS) 430 return (-1); 431 432 if (cap_fcntls_limit(fds[i], CAP_FCNTL_GETFL) < 0 && errno != ENOSYS) 433 return (-1); 434 } 435 436 return (0); 437 } 438 439 #endif 440 441 #ifdef __FreeBSD__ 442 static int 443 pincpu_parse(const char *opt) 444 { 445 int vcpu, pcpu; 446 447 if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) { 448 fprintf(stderr, "invalid format: %s\n", opt); 449 return (-1); 450 } 451 452 if (vcpu < 0 || vcpu >= VM_MAXCPU) { 453 fprintf(stderr, "vcpu '%d' outside valid range from 0 to %d\n", 454 vcpu, VM_MAXCPU - 1); 455 return (-1); 456 } 457 458 if (pcpu < 0 || pcpu >= CPU_SETSIZE) { 459 fprintf(stderr, "hostcpu '%d' outside valid range from " 460 "0 to %d\n", pcpu, CPU_SETSIZE - 1); 461 return (-1); 462 } 463 464 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 465 value = get_config_value(key); 466 467 if (asprintf(&newval, "%s%s%d", value != NULL ? value : "", 468 value != NULL ? "," : "", pcpu) == -1) { 469 perror("failed to build new cpuset string"); 470 return (-1); 471 } 472 473 set_config_value(key, newval); 474 free(newval); 475 return (0); 476 } 477 478 static void 479 parse_cpuset(int vcpu, const char *list, cpuset_t *set) 480 { 481 char *cp, *token; 482 int pcpu, start; 483 484 CPU_ZERO(set); 485 start = -1; 486 token = __DECONST(char *, list); 487 for (;;) { 488 pcpu = strtoul(token, &cp, 0); 489 if (cp == token) 490 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 491 if (pcpu < 0 || pcpu >= CPU_SETSIZE) 492 errx(4, "hostcpu '%d' outside valid range from 0 to %d", 493 pcpu, CPU_SETSIZE - 1); 494 switch (*cp) { 495 case ',': 496 case '\0': 497 if (start >= 0) { 498 if (start > pcpu) 499 errx(4, "Invalid hostcpu range %d-%d", 500 start, pcpu); 501 while (start < pcpu) { 502 CPU_SET(start, vcpumap[vcpu]); 503 start++; 504 } 505 start = -1; 506 } 507 CPU_SET(pcpu, vcpumap[vcpu]); 508 break; 509 case '-': 510 if (start >= 0) 511 errx(4, "invalid cpuset for vcpu %d: '%s'", 512 vcpu, list); 513 start = pcpu; 514 break; 515 default: 516 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 517 } 518 if (*cp == '\0') 519 break; 520 token = cp + 1; 521 } 522 } 523 524 static void 525 build_vcpumaps(void) 526 { 527 char key[16]; 528 const char *value; 529 int vcpu; 530 531 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) { 532 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 533 value = get_config_value(key); 534 if (value == NULL) 535 continue; 536 vcpumap[vcpu] = malloc(sizeof(cpuset_t)); 537 if (vcpumap[vcpu] == NULL) 538 err(4, "Failed to allocate cpuset for vcpu %d", vcpu); 539 parse_cpuset(vcpu, value, vcpumap[vcpu]); 540 } 541 } 542 543 void 544 vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid, 545 int errcode) 546 { 547 struct vmctx *ctx; 548 int error, restart_instruction; 549 550 ctx = arg; 551 restart_instruction = 1; 552 553 error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode, 554 restart_instruction); 555 assert(error == 0); 556 } 557 #endif /* __FreeBSD__ */ 558 559 void * 560 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len) 561 { 562 563 return (vm_map_gpa(ctx, gaddr, len)); 564 } 565 566 int 567 fbsdrun_virtio_msix(void) 568 { 569 570 return (get_config_bool_default("virtio_msix", true)); 571 } 572 573 static void * 574 fbsdrun_start_thread(void *param) 575 { 576 char tname[MAXCOMLEN + 1]; 577 struct mt_vmm_info *mtp; 578 int vcpu; 579 580 mtp = param; 581 vcpu = mtp->mt_vcpu; 582 583 snprintf(tname, sizeof(tname), "vcpu %d", vcpu); 584 pthread_set_name_np(mtp->mt_thr, tname); 585 586 gdb_cpu_add(vcpu); 587 588 vm_loop(mtp->mt_ctx, vcpu, mtp->mt_startrip); 589 590 /* not reached */ 591 exit(1); 592 return (NULL); 593 } 594 595 #ifdef __FreeBSD__ 596 void 597 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip) 598 #else 599 void 600 fbsdrun_addcpu(struct vmctx *ctx, int newcpu, uint64_t rip, bool suspend) 601 #endif 602 { 603 int error; 604 605 #ifdef __FreeBSD__ 606 assert(fromcpu == BSP); 607 #endif 608 609 /* 610 * The 'newcpu' must be activated in the context of 'fromcpu'. If 611 * vm_activate_cpu() is delayed until newcpu's pthread starts running 612 * then vmm.ko is out-of-sync with bhyve and this can create a race 613 * with vm_suspend(). 614 */ 615 error = vm_activate_cpu(ctx, newcpu); 616 if (error != 0) 617 err(EX_OSERR, "could not activate CPU %d", newcpu); 618 619 CPU_SET_ATOMIC(newcpu, &cpumask); 620 621 #ifndef __FreeBSD__ 622 if (suspend) 623 (void) vm_suspend_cpu(ctx, newcpu); 624 #endif 625 626 /* 627 * Set up the vmexit struct to allow execution to start 628 * at the given RIP 629 */ 630 mt_vmm_info[newcpu].mt_ctx = ctx; 631 mt_vmm_info[newcpu].mt_vcpu = newcpu; 632 mt_vmm_info[newcpu].mt_startrip = rip; 633 634 error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL, 635 fbsdrun_start_thread, &mt_vmm_info[newcpu]); 636 assert(error == 0); 637 } 638 639 static int 640 fbsdrun_deletecpu(struct vmctx *ctx, int vcpu) 641 { 642 643 if (!CPU_ISSET(vcpu, &cpumask)) { 644 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu); 645 exit(4); 646 } 647 648 CPU_CLR_ATOMIC(vcpu, &cpumask); 649 return (CPU_EMPTY(&cpumask)); 650 } 651 652 static void 653 vmentry_mmio_read(int vcpu, uint64_t gpa, uint8_t bytes, uint64_t data) 654 { 655 struct vm_entry *entry = &vmentry[vcpu]; 656 struct vm_mmio *mmio = &entry->u.mmio; 657 658 assert(entry->cmd == VEC_DEFAULT); 659 660 entry->cmd = VEC_FULFILL_MMIO; 661 mmio->bytes = bytes; 662 mmio->read = 1; 663 mmio->gpa = gpa; 664 mmio->data = data; 665 } 666 667 static void 668 vmentry_mmio_write(int vcpu, uint64_t gpa, uint8_t bytes) 669 { 670 struct vm_entry *entry = &vmentry[vcpu]; 671 struct vm_mmio *mmio = &entry->u.mmio; 672 673 assert(entry->cmd == VEC_DEFAULT); 674 675 entry->cmd = VEC_FULFILL_MMIO; 676 mmio->bytes = bytes; 677 mmio->read = 0; 678 mmio->gpa = gpa; 679 mmio->data = 0; 680 } 681 682 static void 683 vmentry_inout_read(int vcpu, uint16_t port, uint8_t bytes, uint32_t data) 684 { 685 struct vm_entry *entry = &vmentry[vcpu]; 686 struct vm_inout *inout = &entry->u.inout; 687 688 assert(entry->cmd == VEC_DEFAULT); 689 690 entry->cmd = VEC_FULFILL_INOUT; 691 inout->bytes = bytes; 692 inout->flags = INOUT_IN; 693 inout->port = port; 694 inout->eax = data; 695 } 696 697 static void 698 vmentry_inout_write(int vcpu, uint16_t port, uint8_t bytes) 699 { 700 struct vm_entry *entry = &vmentry[vcpu]; 701 struct vm_inout *inout = &entry->u.inout; 702 703 assert(entry->cmd == VEC_DEFAULT); 704 705 entry->cmd = VEC_FULFILL_INOUT; 706 inout->bytes = bytes; 707 inout->flags = 0; 708 inout->port = port; 709 inout->eax = 0; 710 } 711 712 static int 713 vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu, 714 uint32_t eax) 715 { 716 #if BHYVE_DEBUG 717 /* 718 * put guest-driven debug here 719 */ 720 #endif 721 return (VMEXIT_CONTINUE); 722 } 723 724 static int 725 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 726 { 727 int error; 728 int vcpu; 729 struct vm_inout inout; 730 bool in; 731 uint8_t bytes; 732 733 stats.vmexit_inout++; 734 735 vcpu = *pvcpu; 736 inout = vme->u.inout; 737 in = (inout.flags & INOUT_IN) != 0; 738 bytes = inout.bytes; 739 740 /* Extra-special case of host notifications */ 741 if (!in && inout.port == GUEST_NIO_PORT) { 742 error = vmexit_handle_notify(ctx, vme, pvcpu, inout.eax); 743 vmentry_inout_write(vcpu, inout.port, bytes); 744 return (error); 745 } 746 747 error = emulate_inout(ctx, vcpu, &inout); 748 if (error) { 749 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n", 750 in ? "in" : "out", 751 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'), 752 inout.port, vmexit->rip); 753 return (VMEXIT_ABORT); 754 } else { 755 /* 756 * Communicate the status of the inout operation back to the 757 * in-kernel instruction emulation. 758 */ 759 if (in) { 760 vmentry_inout_read(vcpu, inout.port, bytes, inout.eax); 761 } else { 762 vmentry_inout_write(vcpu, inout.port, bytes); 763 } 764 return (VMEXIT_CONTINUE); 765 } 766 } 767 768 static int 769 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 770 { 771 uint64_t val; 772 uint32_t eax, edx; 773 int error; 774 775 val = 0; 776 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val); 777 if (error != 0) { 778 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n", 779 vme->u.msr.code, *pvcpu); 780 if (get_config_bool("x86.strictmsr")) { 781 vm_inject_gp(ctx, *pvcpu); 782 return (VMEXIT_CONTINUE); 783 } 784 } 785 786 eax = val; 787 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax); 788 assert(error == 0); 789 790 edx = val >> 32; 791 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx); 792 assert(error == 0); 793 794 return (VMEXIT_CONTINUE); 795 } 796 797 static int 798 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 799 { 800 int error; 801 802 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval); 803 if (error != 0) { 804 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n", 805 vme->u.msr.code, vme->u.msr.wval, *pvcpu); 806 if (get_config_bool("x86.strictmsr")) { 807 vm_inject_gp(ctx, *pvcpu); 808 return (VMEXIT_CONTINUE); 809 } 810 } 811 return (VMEXIT_CONTINUE); 812 } 813 814 #ifdef __FreeBSD__ 815 static int 816 vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 817 { 818 819 (void)spinup_ap(ctx, *pvcpu, 820 vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip); 821 822 return (VMEXIT_CONTINUE); 823 } 824 #else 825 static int 826 vmexit_run_state(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 827 { 828 /* 829 * Run-state transitions (INIT, SIPI, etc) are handled in-kernel, so an 830 * exit to userspace with that code is not expected. 831 */ 832 fprintf(stderr, "unexpected run-state VM exit"); 833 return (VMEXIT_ABORT); 834 } 835 #endif /* __FreeBSD__ */ 836 837 #ifdef __FreeBSD__ 838 #define DEBUG_EPT_MISCONFIG 839 #else 840 /* EPT misconfig debugging not possible now that raw VMCS access is gone */ 841 #endif 842 843 #ifdef DEBUG_EPT_MISCONFIG 844 #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400 845 846 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4]; 847 static int ept_misconfig_ptenum; 848 #endif 849 850 static const char * 851 vmexit_vmx_desc(uint32_t exit_reason) 852 { 853 854 if (exit_reason >= nitems(vmx_exit_reason_desc) || 855 vmx_exit_reason_desc[exit_reason] == NULL) 856 return ("Unknown"); 857 return (vmx_exit_reason_desc[exit_reason]); 858 } 859 860 static int 861 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 862 { 863 864 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 865 fprintf(stderr, "\treason\t\tVMX\n"); 866 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 867 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 868 fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status); 869 fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason, 870 vmexit_vmx_desc(vmexit->u.vmx.exit_reason)); 871 fprintf(stderr, "\tqualification\t0x%016lx\n", 872 vmexit->u.vmx.exit_qualification); 873 fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type); 874 fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error); 875 #ifdef DEBUG_EPT_MISCONFIG 876 if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) { 877 vm_get_register(ctx, *pvcpu, 878 VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS), 879 &ept_misconfig_gpa); 880 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte, 881 &ept_misconfig_ptenum); 882 fprintf(stderr, "\tEPT misconfiguration:\n"); 883 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa); 884 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n", 885 ept_misconfig_ptenum, ept_misconfig_pte[0], 886 ept_misconfig_pte[1], ept_misconfig_pte[2], 887 ept_misconfig_pte[3]); 888 } 889 #endif /* DEBUG_EPT_MISCONFIG */ 890 return (VMEXIT_ABORT); 891 } 892 893 static int 894 vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 895 { 896 897 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 898 fprintf(stderr, "\treason\t\tSVM\n"); 899 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 900 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 901 fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode); 902 fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1); 903 fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2); 904 return (VMEXIT_ABORT); 905 } 906 907 static int 908 vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 909 { 910 911 assert(vmexit->inst_length == 0); 912 913 stats.vmexit_bogus++; 914 915 return (VMEXIT_CONTINUE); 916 } 917 918 static int 919 vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 920 { 921 922 assert(vmexit->inst_length == 0); 923 924 stats.vmexit_reqidle++; 925 926 return (VMEXIT_CONTINUE); 927 } 928 929 static int 930 vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 931 { 932 933 stats.vmexit_hlt++; 934 935 /* 936 * Just continue execution with the next instruction. We use 937 * the HLT VM exit as a way to be friendly with the host 938 * scheduler. 939 */ 940 return (VMEXIT_CONTINUE); 941 } 942 943 static int 944 vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 945 { 946 947 stats.vmexit_pause++; 948 949 return (VMEXIT_CONTINUE); 950 } 951 952 static int 953 vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 954 { 955 956 assert(vmexit->inst_length == 0); 957 958 stats.vmexit_mtrap++; 959 960 gdb_cpu_mtrap(*pvcpu); 961 962 return (VMEXIT_CONTINUE); 963 } 964 965 static int 966 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 967 { 968 uint8_t i, valid; 969 970 fprintf(stderr, "Failed to emulate instruction sequence "); 971 972 valid = vmexit->u.inst_emul.num_valid; 973 if (valid != 0) { 974 assert(valid <= sizeof (vmexit->u.inst_emul.inst)); 975 fprintf(stderr, "["); 976 for (i = 0; i < valid; i++) { 977 if (i == 0) { 978 fprintf(stderr, "%02x", 979 vmexit->u.inst_emul.inst[i]); 980 } else { 981 fprintf(stderr, ", %02x", 982 vmexit->u.inst_emul.inst[i]); 983 } 984 } 985 fprintf(stderr, "] "); 986 } 987 fprintf(stderr, "@ %rip = %x\n", vmexit->rip); 988 989 return (VMEXIT_ABORT); 990 } 991 992 static int 993 vmexit_mmio(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 994 { 995 int vcpu, err; 996 struct vm_mmio mmio; 997 bool is_read; 998 999 stats.vmexit_mmio++; 1000 1001 vcpu = *pvcpu; 1002 mmio = vmexit->u.mmio; 1003 is_read = (mmio.read != 0); 1004 1005 err = emulate_mem(ctx, vcpu, &mmio); 1006 1007 if (err == ESRCH) { 1008 fprintf(stderr, "Unhandled memory access to 0x%lx\n", mmio.gpa); 1009 stats.mmio_unhandled++; 1010 1011 /* 1012 * Access to non-existent physical addresses is not likely to 1013 * result in fatal errors on hardware machines, but rather reads 1014 * of all-ones or discarded-but-acknowledged writes. 1015 */ 1016 mmio.data = ~0UL; 1017 err = 0; 1018 } 1019 1020 if (err == 0) { 1021 if (is_read) { 1022 vmentry_mmio_read(vcpu, mmio.gpa, mmio.bytes, 1023 mmio.data); 1024 } else { 1025 vmentry_mmio_write(vcpu, mmio.gpa, mmio.bytes); 1026 } 1027 return (VMEXIT_CONTINUE); 1028 } 1029 1030 fprintf(stderr, "Unhandled mmio error to 0x%lx: %d\n", mmio.gpa, err); 1031 return (VMEXIT_ABORT); 1032 } 1033 1034 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER; 1035 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER; 1036 1037 static int 1038 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1039 { 1040 enum vm_suspend_how how; 1041 1042 how = vmexit->u.suspended.how; 1043 1044 fbsdrun_deletecpu(ctx, *pvcpu); 1045 1046 if (*pvcpu != BSP) { 1047 pthread_mutex_lock(&resetcpu_mtx); 1048 pthread_cond_signal(&resetcpu_cond); 1049 pthread_mutex_unlock(&resetcpu_mtx); 1050 pthread_exit(NULL); 1051 } 1052 1053 pthread_mutex_lock(&resetcpu_mtx); 1054 while (!CPU_EMPTY(&cpumask)) { 1055 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx); 1056 } 1057 pthread_mutex_unlock(&resetcpu_mtx); 1058 1059 switch (how) { 1060 case VM_SUSPEND_RESET: 1061 exit(0); 1062 case VM_SUSPEND_POWEROFF: 1063 if (get_config_bool_default("destroy_on_poweroff", false)) 1064 vm_destroy(ctx); 1065 exit(1); 1066 case VM_SUSPEND_HALT: 1067 exit(2); 1068 case VM_SUSPEND_TRIPLEFAULT: 1069 exit(3); 1070 default: 1071 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how); 1072 exit(100); 1073 } 1074 return (0); /* NOTREACHED */ 1075 } 1076 1077 static int 1078 vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1079 { 1080 1081 gdb_cpu_suspend(*pvcpu); 1082 return (VMEXIT_CONTINUE); 1083 } 1084 1085 static int 1086 vmexit_breakpoint(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1087 { 1088 1089 gdb_cpu_breakpoint(*pvcpu, vmexit); 1090 return (VMEXIT_CONTINUE); 1091 } 1092 1093 static vmexit_handler_t handler[VM_EXITCODE_MAX] = { 1094 [VM_EXITCODE_INOUT] = vmexit_inout, 1095 [VM_EXITCODE_MMIO] = vmexit_mmio, 1096 [VM_EXITCODE_VMX] = vmexit_vmx, 1097 [VM_EXITCODE_SVM] = vmexit_svm, 1098 [VM_EXITCODE_BOGUS] = vmexit_bogus, 1099 [VM_EXITCODE_REQIDLE] = vmexit_reqidle, 1100 [VM_EXITCODE_RDMSR] = vmexit_rdmsr, 1101 [VM_EXITCODE_WRMSR] = vmexit_wrmsr, 1102 [VM_EXITCODE_MTRAP] = vmexit_mtrap, 1103 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul, 1104 #ifdef __FreeBSD__ 1105 [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap, 1106 #else 1107 [VM_EXITCODE_RUN_STATE] = vmexit_run_state, 1108 #endif 1109 [VM_EXITCODE_SUSPENDED] = vmexit_suspend, 1110 [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch, 1111 [VM_EXITCODE_DEBUG] = vmexit_debug, 1112 [VM_EXITCODE_BPT] = vmexit_breakpoint, 1113 }; 1114 1115 static void 1116 vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip) 1117 { 1118 int error, rc; 1119 enum vm_exitcode exitcode; 1120 cpuset_t active_cpus; 1121 struct vm_exit *vexit; 1122 struct vm_entry *ventry; 1123 1124 #ifdef __FreeBSD__ 1125 if (vcpumap[vcpu] != NULL) { 1126 error = pthread_setaffinity_np(pthread_self(), 1127 sizeof(cpuset_t), vcpumap[vcpu]); 1128 assert(error == 0); 1129 } 1130 #endif 1131 error = vm_active_cpus(ctx, &active_cpus); 1132 assert(CPU_ISSET(vcpu, &active_cpus)); 1133 1134 error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip); 1135 assert(error == 0); 1136 1137 ventry = &vmentry[vcpu]; 1138 vexit = &vmexit[vcpu]; 1139 1140 while (1) { 1141 error = vm_run(ctx, vcpu, ventry, vexit); 1142 if (error != 0) 1143 break; 1144 1145 if (ventry->cmd != VEC_DEFAULT) { 1146 /* 1147 * Discard any lingering entry state after it has been 1148 * submitted via vm_run(). 1149 */ 1150 bzero(ventry, sizeof (*ventry)); 1151 } 1152 1153 exitcode = vexit->exitcode; 1154 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) { 1155 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n", 1156 exitcode); 1157 exit(4); 1158 } 1159 1160 rc = (*handler[exitcode])(ctx, vexit, &vcpu); 1161 1162 switch (rc) { 1163 case VMEXIT_CONTINUE: 1164 break; 1165 case VMEXIT_ABORT: 1166 abort(); 1167 default: 1168 exit(4); 1169 } 1170 } 1171 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno); 1172 } 1173 1174 static int 1175 num_vcpus_allowed(struct vmctx *ctx) 1176 { 1177 #ifdef __FreeBSD__ 1178 int tmp, error; 1179 1180 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp); 1181 1182 /* 1183 * The guest is allowed to spinup more than one processor only if the 1184 * UNRESTRICTED_GUEST capability is available. 1185 */ 1186 if (error == 0) 1187 return (VM_MAXCPU); 1188 else 1189 return (1); 1190 #else 1191 /* Unrestricted Guest is always enabled on illumos */ 1192 return (VM_MAXCPU); 1193 #endif /* __FreeBSD__ */ 1194 } 1195 1196 void 1197 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu) 1198 { 1199 int err, tmp; 1200 1201 if (get_config_bool_default("x86.vmexit_on_hlt", false)) { 1202 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp); 1203 if (err < 0) { 1204 fprintf(stderr, "VM exit on HLT not supported\n"); 1205 exit(4); 1206 } 1207 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1); 1208 if (cpu == BSP) 1209 handler[VM_EXITCODE_HLT] = vmexit_hlt; 1210 } 1211 1212 if (get_config_bool_default("x86.vmexit_on_pause", false)) { 1213 /* 1214 * pause exit support required for this mode 1215 */ 1216 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp); 1217 if (err < 0) { 1218 fprintf(stderr, 1219 "SMP mux requested, no pause support\n"); 1220 exit(4); 1221 } 1222 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1); 1223 if (cpu == BSP) 1224 handler[VM_EXITCODE_PAUSE] = vmexit_pause; 1225 } 1226 1227 if (get_config_bool_default("x86.x2apic", false)) 1228 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED); 1229 else 1230 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED); 1231 1232 if (err) { 1233 fprintf(stderr, "Unable to set x2apic state (%d)\n", err); 1234 exit(4); 1235 } 1236 1237 #ifdef __FreeBSD__ 1238 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1); 1239 #endif 1240 } 1241 1242 static struct vmctx * 1243 do_open(const char *vmname) 1244 { 1245 struct vmctx *ctx; 1246 int error; 1247 bool reinit, romboot; 1248 #ifndef WITHOUT_CAPSICUM 1249 cap_rights_t rights; 1250 const cap_ioctl_t *cmds; 1251 size_t ncmds; 1252 #endif 1253 1254 reinit = romboot = false; 1255 1256 if (lpc_bootrom()) 1257 romboot = true; 1258 #ifndef __FreeBSD__ 1259 uint64_t create_flags = 0; 1260 if (get_config_bool_default("memory.use_reservoir", false)) { 1261 create_flags |= VCF_RESERVOIR_MEM; 1262 } 1263 error = vm_create(vmname, create_flags); 1264 #else 1265 error = vm_create(vmname); 1266 #endif /* __FreeBSD__ */ 1267 if (error) { 1268 if (errno == EEXIST) { 1269 if (romboot) { 1270 reinit = true; 1271 } else { 1272 /* 1273 * The virtual machine has been setup by the 1274 * userspace bootloader. 1275 */ 1276 } 1277 } else { 1278 perror("vm_create"); 1279 exit(4); 1280 } 1281 } else { 1282 if (!romboot) { 1283 /* 1284 * If the virtual machine was just created then a 1285 * bootrom must be configured to boot it. 1286 */ 1287 fprintf(stderr, "virtual machine cannot be booted\n"); 1288 exit(4); 1289 } 1290 } 1291 1292 ctx = vm_open(vmname); 1293 if (ctx == NULL) { 1294 perror("vm_open"); 1295 exit(4); 1296 } 1297 1298 #ifndef WITHOUT_CAPSICUM 1299 cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW); 1300 if (caph_rights_limit(vm_get_device_fd(ctx), &rights) == -1) 1301 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1302 vm_get_ioctls(&ncmds); 1303 cmds = vm_get_ioctls(NULL); 1304 if (cmds == NULL) 1305 errx(EX_OSERR, "out of memory"); 1306 if (caph_ioctls_limit(vm_get_device_fd(ctx), cmds, ncmds) == -1) 1307 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1308 free((cap_ioctl_t *)cmds); 1309 #endif 1310 1311 if (reinit) { 1312 #ifndef __FreeBSD__ 1313 error = vm_reinit(ctx, 0); 1314 #else 1315 error = vm_reinit(ctx); 1316 #endif 1317 if (error) { 1318 perror("vm_reinit"); 1319 exit(4); 1320 } 1321 } 1322 error = vm_set_topology(ctx, sockets, cores, threads, maxcpus); 1323 if (error) 1324 errx(EX_OSERR, "vm_set_topology"); 1325 return (ctx); 1326 } 1327 1328 static bool 1329 parse_config_option(const char *option) 1330 { 1331 const char *value; 1332 char *path; 1333 1334 value = strchr(option, '='); 1335 if (value == NULL || value[1] == '\0') 1336 return (false); 1337 path = strndup(option, value - option); 1338 if (path == NULL) 1339 err(4, "Failed to allocate memory"); 1340 set_config_value(path, value + 1); 1341 return (true); 1342 } 1343 1344 static void 1345 parse_simple_config_file(const char *path) 1346 { 1347 FILE *fp; 1348 char *line, *cp; 1349 size_t linecap; 1350 unsigned int lineno; 1351 1352 fp = fopen(path, "r"); 1353 if (fp == NULL) 1354 err(4, "Failed to open configuration file %s", path); 1355 line = NULL; 1356 linecap = 0; 1357 lineno = 1; 1358 for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) { 1359 if (*line == '#' || *line == '\n') 1360 continue; 1361 cp = strchr(line, '\n'); 1362 if (cp != NULL) 1363 *cp = '\0'; 1364 if (!parse_config_option(line)) 1365 errx(4, "%s line %u: invalid config option '%s'", path, 1366 lineno, line); 1367 } 1368 free(line); 1369 fclose(fp); 1370 } 1371 1372 static void 1373 parse_gdb_options(char *optarg) 1374 { 1375 const char *sport; 1376 char *colon; 1377 1378 if (optarg[0] == 'w') { 1379 set_config_bool("gdb.wait", true); 1380 optarg++; 1381 } 1382 1383 colon = strrchr(optarg, ':'); 1384 if (colon == NULL) { 1385 sport = optarg; 1386 } else { 1387 *colon = '\0'; 1388 colon++; 1389 sport = colon; 1390 set_config_value("gdb.address", optarg); 1391 } 1392 1393 set_config_value("gdb.port", sport); 1394 } 1395 1396 static void 1397 set_defaults(void) 1398 { 1399 1400 set_config_bool("acpi_tables", false); 1401 set_config_value("memory.size", "256M"); 1402 set_config_bool("x86.strictmsr", true); 1403 } 1404 1405 int 1406 main(int argc, char *argv[]) 1407 { 1408 int c, error, err; 1409 int max_vcpus, memflags; 1410 struct vmctx *ctx; 1411 uint64_t rip; 1412 size_t memsize; 1413 const char *value, *vmname; 1414 char *optstr; 1415 1416 init_config(); 1417 set_defaults(); 1418 progname = basename(argv[0]); 1419 1420 #ifdef __FreeBSD__ 1421 optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:U:"; 1422 #else 1423 /* +d, +B, -p */ 1424 optstr = "adehuwxACDHIPSWYk:o:G:c:s:m:l:B:U:"; 1425 #endif 1426 while ((c = getopt(argc, argv, optstr)) != -1) { 1427 switch (c) { 1428 case 'a': 1429 set_config_bool("x86.x2apic", false); 1430 break; 1431 case 'A': 1432 set_config_bool("acpi_tables", true); 1433 break; 1434 case 'D': 1435 set_config_bool("destroy_on_poweroff", true); 1436 break; 1437 #ifndef __FreeBSD__ 1438 case 'B': 1439 if (smbios_parse(optarg) != 0) { 1440 errx(EX_USAGE, "invalid SMBIOS " 1441 "configuration '%s'", optarg); 1442 } 1443 break; 1444 case 'd': 1445 set_config_bool("suspend_at_boot", true); 1446 break; 1447 #endif 1448 #ifdef __FreeBSD__ 1449 case 'p': 1450 if (pincpu_parse(optarg) != 0) { 1451 errx(EX_USAGE, "invalid vcpu pinning " 1452 "configuration '%s'", optarg); 1453 } 1454 break; 1455 #endif 1456 case 'c': 1457 if (topology_parse(optarg) != 0) { 1458 errx(EX_USAGE, "invalid cpu topology " 1459 "'%s'", optarg); 1460 } 1461 break; 1462 case 'C': 1463 set_config_bool("memory.guest_in_core", true); 1464 break; 1465 case 'G': 1466 parse_gdb_options(optarg); 1467 break; 1468 case 'k': 1469 parse_simple_config_file(optarg); 1470 break; 1471 case 'l': 1472 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1473 lpc_print_supported_devices(); 1474 exit(0); 1475 } else if (lpc_device_parse(optarg) != 0) { 1476 errx(EX_USAGE, "invalid lpc device " 1477 "configuration '%s'", optarg); 1478 } 1479 break; 1480 case 's': 1481 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1482 pci_print_supported_devices(); 1483 exit(0); 1484 } else if (pci_parse_slot(optarg) != 0) 1485 exit(4); 1486 else 1487 break; 1488 case 'S': 1489 set_config_bool("memory.wired", true); 1490 break; 1491 case 'm': 1492 set_config_value("memory.size", optarg); 1493 break; 1494 case 'o': 1495 if (!parse_config_option(optarg)) 1496 errx(EX_USAGE, "invalid configuration option '%s'", optarg); 1497 break; 1498 case 'H': 1499 set_config_bool("x86.vmexit_on_hlt", true); 1500 break; 1501 case 'I': 1502 /* 1503 * The "-I" option was used to add an ioapic to the 1504 * virtual machine. 1505 * 1506 * An ioapic is now provided unconditionally for each 1507 * virtual machine and this option is now deprecated. 1508 */ 1509 break; 1510 case 'P': 1511 set_config_bool("x86.vmexit_on_pause", true); 1512 break; 1513 case 'e': 1514 set_config_bool("x86.strictio", true); 1515 break; 1516 case 'u': 1517 set_config_bool("rtc.use_localtime", false); 1518 break; 1519 case 'U': 1520 set_config_value("uuid", optarg); 1521 break; 1522 case 'w': 1523 set_config_bool("x86.strictmsr", false); 1524 break; 1525 case 'W': 1526 set_config_bool("virtio_msix", false); 1527 break; 1528 case 'x': 1529 set_config_bool("x86.x2apic", true); 1530 break; 1531 case 'Y': 1532 set_config_bool("x86.mptable", false); 1533 break; 1534 case 'h': 1535 usage(0); 1536 default: 1537 usage(1); 1538 } 1539 } 1540 argc -= optind; 1541 argv += optind; 1542 1543 if (argc > 1) 1544 usage(1); 1545 1546 if (argc == 1) 1547 set_config_value("name", argv[0]); 1548 1549 vmname = get_config_value("name"); 1550 if (vmname == NULL) 1551 usage(1); 1552 1553 if (get_config_bool_default("config.dump", false)) { 1554 dump_config(); 1555 exit(1); 1556 } 1557 1558 #ifndef __FreeBSD__ 1559 illumos_priv_init(); 1560 #endif 1561 1562 calc_topolopgy(); 1563 #ifdef __FreeBSD__ 1564 build_vcpumaps(); 1565 #endif 1566 1567 value = get_config_value("memory.size"); 1568 error = vm_parse_memsize(value, &memsize); 1569 if (error) 1570 errx(EX_USAGE, "invalid memsize '%s'", value); 1571 1572 ctx = do_open(vmname); 1573 1574 max_vcpus = num_vcpus_allowed(ctx); 1575 if (guest_ncpus > max_vcpus) { 1576 fprintf(stderr, "%d vCPUs requested but only %d available\n", 1577 guest_ncpus, max_vcpus); 1578 exit(4); 1579 } 1580 1581 fbsdrun_set_capabilities(ctx, BSP); 1582 1583 memflags = 0; 1584 if (get_config_bool_default("memory.wired", false)) 1585 memflags |= VM_MEM_F_WIRED; 1586 if (get_config_bool_default("memory.guest_in_core", false)) 1587 memflags |= VM_MEM_F_INCORE; 1588 vm_set_memflags(ctx, memflags); 1589 #ifdef __FreeBSD__ 1590 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1591 #else 1592 do { 1593 errno = 0; 1594 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1595 error = errno; 1596 if (err != 0 && error == ENOMEM) { 1597 (void) fprintf(stderr, "Unable to allocate memory " 1598 "(%llu), retrying in 1 second\n", memsize); 1599 sleep(1); 1600 } 1601 } while (error == ENOMEM); 1602 #endif 1603 if (err) { 1604 fprintf(stderr, "Unable to set up memory (%d)\n", errno); 1605 exit(4); 1606 } 1607 1608 error = init_msr(); 1609 if (error) { 1610 fprintf(stderr, "init_msr error %d", error); 1611 exit(4); 1612 } 1613 1614 init_mem(); 1615 init_inout(); 1616 #ifdef __FreeBSD__ 1617 kernemu_dev_init(); 1618 #endif 1619 init_bootrom(ctx); 1620 atkbdc_init(ctx); 1621 pci_irq_init(ctx); 1622 ioapic_init(ctx); 1623 1624 rtc_init(ctx); 1625 sci_init(ctx); 1626 #ifndef __FreeBSD__ 1627 pmtmr_init(ctx); 1628 #endif 1629 1630 /* 1631 * Exit if a device emulation finds an error in its initilization 1632 */ 1633 if (init_pci(ctx) != 0) { 1634 perror("device emulation initialization error"); 1635 exit(4); 1636 } 1637 1638 /* 1639 * Initialize after PCI, to allow a bootrom file to reserve the high 1640 * region. 1641 */ 1642 if (get_config_bool("acpi_tables")) 1643 vmgenc_init(ctx); 1644 1645 #ifdef __FreeBSD__ 1646 init_gdb(ctx); 1647 #else 1648 if (value != NULL) { 1649 int port = atoi(value); 1650 1651 if (port < 0) 1652 init_mdb(ctx); 1653 else 1654 init_gdb(ctx); 1655 } 1656 #endif 1657 1658 vga_init(1); 1659 1660 if (lpc_bootrom()) { 1661 #ifdef __FreeBSD__ 1662 if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) { 1663 fprintf(stderr, "ROM boot failed: unrestricted guest " 1664 "capability not available\n"); 1665 exit(4); 1666 } 1667 #else 1668 /* Unrestricted Guest is always enabled on illumos */ 1669 #endif 1670 error = vcpu_reset(ctx, BSP); 1671 assert(error == 0); 1672 } 1673 1674 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip); 1675 assert(error == 0); 1676 1677 /* 1678 * build the guest tables, MP etc. 1679 */ 1680 if (get_config_bool_default("x86.mptable", true)) { 1681 error = mptable_build(ctx, guest_ncpus); 1682 if (error) { 1683 perror("error to build the guest tables"); 1684 exit(4); 1685 } 1686 } 1687 1688 #ifndef __FreeBSD__ 1689 smbios_apply(); 1690 #endif 1691 error = smbios_build(ctx); 1692 assert(error == 0); 1693 1694 if (get_config_bool("acpi_tables")) { 1695 error = acpi_build(ctx, guest_ncpus); 1696 assert(error == 0); 1697 } 1698 1699 if (lpc_bootrom()) 1700 fwctl_init(); 1701 1702 /* 1703 * Change the proc title to include the VM name. 1704 */ 1705 setproctitle("%s", vmname); 1706 1707 #ifndef WITHOUT_CAPSICUM 1708 caph_cache_catpages(); 1709 1710 if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1) 1711 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1712 1713 if (caph_enter() == -1) 1714 errx(EX_OSERR, "cap_enter() failed"); 1715 #endif 1716 1717 #ifndef __FreeBSD__ 1718 illumos_priv_lock(); 1719 #endif 1720 1721 #ifdef __FreeBSD__ 1722 /* 1723 * Add CPU 0 1724 */ 1725 fbsdrun_addcpu(ctx, BSP, BSP, rip); 1726 #else 1727 /* Set BSP to run (unlike the APs which wait for INIT) */ 1728 error = vm_set_run_state(ctx, BSP, VRS_RUN, 0); 1729 assert(error == 0); 1730 fbsdrun_addcpu(ctx, BSP, rip, 1731 get_config_bool_default("suspend_at_boot", false)); 1732 1733 /* Add subsequent CPUs, which will wait until INIT/SIPI-ed */ 1734 for (uint_t i = 1; i < guest_ncpus; i++) { 1735 spinup_halted_ap(ctx, i); 1736 } 1737 #endif 1738 /* 1739 * Head off to the main event dispatch loop 1740 */ 1741 mevent_dispatch(); 1742 1743 exit(4); 1744 } 1745