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 * Copyright 2022 OmniOS Community Edition (OmniOSce) Association. 44 */ 45 46 #include <sys/cdefs.h> 47 __FBSDID("$FreeBSD$"); 48 49 #include <sys/types.h> 50 #ifndef WITHOUT_CAPSICUM 51 #include <sys/capsicum.h> 52 #endif 53 #include <sys/mman.h> 54 #include <sys/time.h> 55 #include <sys/cpuset.h> 56 57 #ifdef __FreeBSD__ 58 #include <amd64/vmm/intel/vmcs.h> 59 #else 60 #include <intel/vmcs.h> 61 #endif 62 63 #include <machine/atomic.h> 64 #include <machine/segments.h> 65 66 #ifndef WITHOUT_CAPSICUM 67 #include <capsicum_helpers.h> 68 #endif 69 #include <stdio.h> 70 #include <stdlib.h> 71 #include <string.h> 72 #include <err.h> 73 #include <errno.h> 74 #include <libgen.h> 75 #include <unistd.h> 76 #include <assert.h> 77 #include <pthread.h> 78 #include <pthread_np.h> 79 #include <sysexits.h> 80 #include <stdbool.h> 81 #include <stdint.h> 82 83 #include <machine/vmm.h> 84 #ifndef WITHOUT_CAPSICUM 85 #include <machine/vmm_dev.h> 86 #endif 87 #include <vmmapi.h> 88 89 #include "bhyverun.h" 90 #include "acpi.h" 91 #include "atkbdc.h" 92 #include "console.h" 93 #include "bootrom.h" 94 #include "config.h" 95 #include "inout.h" 96 #include "debug.h" 97 #include "fwctl.h" 98 #include "gdb.h" 99 #include "ioapic.h" 100 #include "kernemu_dev.h" 101 #include "mem.h" 102 #include "mevent.h" 103 #include "mptbl.h" 104 #include "pci_emul.h" 105 #include "pci_irq.h" 106 #include "pci_lpc.h" 107 #include "smbiostbl.h" 108 #include "xmsr.h" 109 #include "spinup_ap.h" 110 #include "rfb.h" 111 #include "rtc.h" 112 #include "vga.h" 113 #include "vmgenc.h" 114 #ifndef __FreeBSD__ 115 #include "privileges.h" 116 #endif 117 118 #define GUEST_NIO_PORT 0x488 /* guest upcalls via i/o port */ 119 120 #define MB (1024UL * 1024) 121 #define GB (1024UL * MB) 122 123 static const char * const vmx_exit_reason_desc[] = { 124 [EXIT_REASON_EXCEPTION] = "Exception or non-maskable interrupt (NMI)", 125 [EXIT_REASON_EXT_INTR] = "External interrupt", 126 [EXIT_REASON_TRIPLE_FAULT] = "Triple fault", 127 [EXIT_REASON_INIT] = "INIT signal", 128 [EXIT_REASON_SIPI] = "Start-up IPI (SIPI)", 129 [EXIT_REASON_IO_SMI] = "I/O system-management interrupt (SMI)", 130 [EXIT_REASON_SMI] = "Other SMI", 131 [EXIT_REASON_INTR_WINDOW] = "Interrupt window", 132 [EXIT_REASON_NMI_WINDOW] = "NMI window", 133 [EXIT_REASON_TASK_SWITCH] = "Task switch", 134 [EXIT_REASON_CPUID] = "CPUID", 135 [EXIT_REASON_GETSEC] = "GETSEC", 136 [EXIT_REASON_HLT] = "HLT", 137 [EXIT_REASON_INVD] = "INVD", 138 [EXIT_REASON_INVLPG] = "INVLPG", 139 [EXIT_REASON_RDPMC] = "RDPMC", 140 [EXIT_REASON_RDTSC] = "RDTSC", 141 [EXIT_REASON_RSM] = "RSM", 142 [EXIT_REASON_VMCALL] = "VMCALL", 143 [EXIT_REASON_VMCLEAR] = "VMCLEAR", 144 [EXIT_REASON_VMLAUNCH] = "VMLAUNCH", 145 [EXIT_REASON_VMPTRLD] = "VMPTRLD", 146 [EXIT_REASON_VMPTRST] = "VMPTRST", 147 [EXIT_REASON_VMREAD] = "VMREAD", 148 [EXIT_REASON_VMRESUME] = "VMRESUME", 149 [EXIT_REASON_VMWRITE] = "VMWRITE", 150 [EXIT_REASON_VMXOFF] = "VMXOFF", 151 [EXIT_REASON_VMXON] = "VMXON", 152 [EXIT_REASON_CR_ACCESS] = "Control-register accesses", 153 [EXIT_REASON_DR_ACCESS] = "MOV DR", 154 [EXIT_REASON_INOUT] = "I/O instruction", 155 [EXIT_REASON_RDMSR] = "RDMSR", 156 [EXIT_REASON_WRMSR] = "WRMSR", 157 [EXIT_REASON_INVAL_VMCS] = 158 "VM-entry failure due to invalid guest state", 159 [EXIT_REASON_INVAL_MSR] = "VM-entry failure due to MSR loading", 160 [EXIT_REASON_MWAIT] = "MWAIT", 161 [EXIT_REASON_MTF] = "Monitor trap flag", 162 [EXIT_REASON_MONITOR] = "MONITOR", 163 [EXIT_REASON_PAUSE] = "PAUSE", 164 [EXIT_REASON_MCE_DURING_ENTRY] = 165 "VM-entry failure due to machine-check event", 166 [EXIT_REASON_TPR] = "TPR below threshold", 167 [EXIT_REASON_APIC_ACCESS] = "APIC access", 168 [EXIT_REASON_VIRTUALIZED_EOI] = "Virtualized EOI", 169 [EXIT_REASON_GDTR_IDTR] = "Access to GDTR or IDTR", 170 [EXIT_REASON_LDTR_TR] = "Access to LDTR or TR", 171 [EXIT_REASON_EPT_FAULT] = "EPT violation", 172 [EXIT_REASON_EPT_MISCONFIG] = "EPT misconfiguration", 173 [EXIT_REASON_INVEPT] = "INVEPT", 174 [EXIT_REASON_RDTSCP] = "RDTSCP", 175 [EXIT_REASON_VMX_PREEMPT] = "VMX-preemption timer expired", 176 [EXIT_REASON_INVVPID] = "INVVPID", 177 [EXIT_REASON_WBINVD] = "WBINVD", 178 [EXIT_REASON_XSETBV] = "XSETBV", 179 [EXIT_REASON_APIC_WRITE] = "APIC write", 180 [EXIT_REASON_RDRAND] = "RDRAND", 181 [EXIT_REASON_INVPCID] = "INVPCID", 182 [EXIT_REASON_VMFUNC] = "VMFUNC", 183 [EXIT_REASON_ENCLS] = "ENCLS", 184 [EXIT_REASON_RDSEED] = "RDSEED", 185 [EXIT_REASON_PM_LOG_FULL] = "Page-modification log full", 186 [EXIT_REASON_XSAVES] = "XSAVES", 187 [EXIT_REASON_XRSTORS] = "XRSTORS" 188 }; 189 190 typedef int (*vmexit_handler_t)(struct vmctx *, struct vm_exit *, int *vcpu); 191 extern int vmexit_task_switch(struct vmctx *, struct vm_exit *, int *vcpu); 192 193 int guest_ncpus; 194 uint16_t cores, maxcpus, sockets, threads; 195 196 int raw_stdio = 0; 197 198 static char *progname; 199 static const int BSP = 0; 200 201 static cpuset_t cpumask; 202 203 static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip); 204 205 static struct vm_exit vmexit[VM_MAXCPU]; 206 static struct vm_entry vmentry[VM_MAXCPU]; 207 208 struct bhyvestats { 209 uint64_t vmexit_bogus; 210 uint64_t vmexit_reqidle; 211 uint64_t vmexit_hlt; 212 uint64_t vmexit_pause; 213 uint64_t vmexit_mtrap; 214 uint64_t vmexit_mmio; 215 uint64_t vmexit_inout; 216 uint64_t cpu_switch_rotate; 217 uint64_t cpu_switch_direct; 218 uint64_t mmio_unhandled; 219 } stats; 220 221 struct mt_vmm_info { 222 pthread_t mt_thr; 223 struct vmctx *mt_ctx; 224 int mt_vcpu; 225 uint64_t mt_startrip; 226 } mt_vmm_info[VM_MAXCPU]; 227 228 #ifdef __FreeBSD__ 229 static cpuset_t *vcpumap[VM_MAXCPU] = { NULL }; 230 #endif 231 232 static void 233 usage(int code) 234 { 235 236 fprintf(stderr, 237 #ifdef __FreeBSD__ 238 "Usage: %s [-AaCDeHhPSuWwxY]\n" 239 #else 240 "Usage: %s [-aCDdeHhPSuWwxY]\n" 241 #endif 242 " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n" 243 #ifdef __FreeBSD__ 244 " %*s [-G port] [-k config_file] [-l lpc] [-m mem] [-o var=value]\n" 245 " %*s [-p vcpu:hostcpu] [-r file] [-s pci] [-U uuid] vmname\n" 246 247 " -A: create ACPI tables\n" 248 #else 249 " %*s [-k <config_file>] [-l <lpc>] [-m mem] [-o <var>=<value>]\n" 250 " %*s [-s <pci>] [-U uuid] vmname\n" 251 #endif 252 " -a: local apic is in xAPIC mode (deprecated)\n" 253 " -C: include guest memory in core file\n" 254 " -c: number of cpus and/or topology specification\n" 255 " -D: destroy on power-off\n" 256 #ifndef __FreeBSD__ 257 " -d: suspend cpu at boot\n" 258 #endif 259 " -e: exit on unhandled I/O access\n" 260 " -H: vmexit from the guest on hlt\n" 261 " -h: help\n" 262 " -k: key=value flat config file\n" 263 " -l: LPC device configuration\n" 264 " -m: memory size\n" 265 " -o: set config 'var' to 'value'\n" 266 " -P: vmexit from the guest on pause\n" 267 #ifdef __FreeBSD__ 268 " -p: pin 'vcpu' to 'hostcpu'\n" 269 #endif 270 " -S: guest memory cannot be swapped\n" 271 " -s: <slot,driver,configinfo> PCI slot config\n" 272 " -U: uuid\n" 273 " -u: RTC keeps UTC time\n" 274 " -W: force virtio to use single-vector MSI\n" 275 " -w: ignore unimplemented MSRs\n" 276 " -x: local apic is in x2APIC mode\n" 277 " -Y: disable MPtable generation\n", 278 progname, (int)strlen(progname), "", (int)strlen(progname), "", 279 (int)strlen(progname), ""); 280 281 exit(code); 282 } 283 284 /* 285 * XXX This parser is known to have the following issues: 286 * 1. It accepts null key=value tokens ",," as setting "cpus" to an 287 * empty string. 288 * 289 * The acceptance of a null specification ('-c ""') is by design to match the 290 * manual page syntax specification, this results in a topology of 1 vCPU. 291 */ 292 static int 293 topology_parse(const char *opt) 294 { 295 char *cp, *str; 296 297 if (*opt == '\0') { 298 set_config_value("sockets", "1"); 299 set_config_value("cores", "1"); 300 set_config_value("threads", "1"); 301 set_config_value("cpus", "1"); 302 return (0); 303 } 304 305 str = strdup(opt); 306 if (str == NULL) 307 errx(4, "Failed to allocate memory"); 308 309 while ((cp = strsep(&str, ",")) != NULL) { 310 if (strncmp(cp, "cpus=", strlen("cpus=")) == 0) 311 set_config_value("cpus", cp + strlen("cpus=")); 312 else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0) 313 set_config_value("sockets", cp + strlen("sockets=")); 314 else if (strncmp(cp, "cores=", strlen("cores=")) == 0) 315 set_config_value("cores", cp + strlen("cores=")); 316 else if (strncmp(cp, "threads=", strlen("threads=")) == 0) 317 set_config_value("threads", cp + strlen("threads=")); 318 #ifdef notyet /* Do not expose this until vmm.ko implements it */ 319 else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0) 320 set_config_value("maxcpus", cp + strlen("maxcpus=")); 321 #endif 322 else if (strchr(cp, '=') != NULL) 323 goto out; 324 else 325 set_config_value("cpus", cp); 326 } 327 free(str); 328 return (0); 329 330 out: 331 free(str); 332 return (-1); 333 } 334 335 static int 336 parse_int_value(const char *key, const char *value, int minval, int maxval) 337 { 338 char *cp; 339 long lval; 340 341 errno = 0; 342 lval = strtol(value, &cp, 0); 343 if (errno != 0 || *cp != '\0' || cp == value || lval < minval || 344 lval > maxval) 345 errx(4, "Invalid value for %s: '%s'", key, value); 346 return (lval); 347 } 348 349 /* 350 * Set the sockets, cores, threads, and guest_cpus variables based on 351 * the configured topology. 352 * 353 * The limits of UINT16_MAX are due to the types passed to 354 * vm_set_topology(). vmm.ko may enforce tighter limits. 355 */ 356 static void 357 calc_topolopgy(void) 358 { 359 const char *value; 360 bool explicit_cpus; 361 uint64_t ncpus; 362 363 value = get_config_value("cpus"); 364 if (value != NULL) { 365 guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX); 366 explicit_cpus = true; 367 } else { 368 guest_ncpus = 1; 369 explicit_cpus = false; 370 } 371 value = get_config_value("cores"); 372 if (value != NULL) 373 cores = parse_int_value("cores", value, 1, UINT16_MAX); 374 else 375 cores = 1; 376 value = get_config_value("threads"); 377 if (value != NULL) 378 threads = parse_int_value("threads", value, 1, UINT16_MAX); 379 else 380 threads = 1; 381 value = get_config_value("sockets"); 382 if (value != NULL) 383 sockets = parse_int_value("sockets", value, 1, UINT16_MAX); 384 else 385 sockets = guest_ncpus; 386 387 /* 388 * Compute sockets * cores * threads avoiding overflow. The 389 * range check above insures these are 16 bit values. 390 */ 391 ncpus = (uint64_t)sockets * cores * threads; 392 if (ncpus > UINT16_MAX) 393 errx(4, "Computed number of vCPUs too high: %ju", 394 (uintmax_t)ncpus); 395 396 if (explicit_cpus) { 397 if (guest_ncpus != ncpus) 398 errx(4, "Topology (%d sockets, %d cores, %d threads) " 399 "does not match %d vCPUs", sockets, cores, threads, 400 guest_ncpus); 401 } else 402 guest_ncpus = ncpus; 403 } 404 405 #ifndef WITHOUT_CAPSICUM 406 /* 407 * 11-stable capsicum helpers 408 */ 409 static void 410 bhyve_caph_cache_catpages(void) 411 { 412 413 (void)catopen("libc", NL_CAT_LOCALE); 414 } 415 416 static int 417 bhyve_caph_limit_stdoe(void) 418 { 419 cap_rights_t rights; 420 unsigned long cmds[] = { TIOCGETA, TIOCGWINSZ }; 421 int i, fds[] = { STDOUT_FILENO, STDERR_FILENO }; 422 423 cap_rights_init(&rights, CAP_FCNTL, CAP_FSTAT, CAP_IOCTL); 424 cap_rights_set(&rights, CAP_WRITE); 425 426 for (i = 0; i < nitems(fds); i++) { 427 if (cap_rights_limit(fds[i], &rights) < 0 && errno != ENOSYS) 428 return (-1); 429 430 if (cap_ioctls_limit(fds[i], cmds, nitems(cmds)) < 0 && errno != ENOSYS) 431 return (-1); 432 433 if (cap_fcntls_limit(fds[i], CAP_FCNTL_GETFL) < 0 && errno != ENOSYS) 434 return (-1); 435 } 436 437 return (0); 438 } 439 440 #endif 441 442 #ifdef __FreeBSD__ 443 static int 444 pincpu_parse(const char *opt) 445 { 446 int vcpu, pcpu; 447 448 if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) { 449 fprintf(stderr, "invalid format: %s\n", opt); 450 return (-1); 451 } 452 453 if (vcpu < 0 || vcpu >= VM_MAXCPU) { 454 fprintf(stderr, "vcpu '%d' outside valid range from 0 to %d\n", 455 vcpu, VM_MAXCPU - 1); 456 return (-1); 457 } 458 459 if (pcpu < 0 || pcpu >= CPU_SETSIZE) { 460 fprintf(stderr, "hostcpu '%d' outside valid range from " 461 "0 to %d\n", pcpu, CPU_SETSIZE - 1); 462 return (-1); 463 } 464 465 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 466 value = get_config_value(key); 467 468 if (asprintf(&newval, "%s%s%d", value != NULL ? value : "", 469 value != NULL ? "," : "", pcpu) == -1) { 470 perror("failed to build new cpuset string"); 471 return (-1); 472 } 473 474 set_config_value(key, newval); 475 free(newval); 476 return (0); 477 } 478 479 static void 480 parse_cpuset(int vcpu, const char *list, cpuset_t *set) 481 { 482 char *cp, *token; 483 int pcpu, start; 484 485 CPU_ZERO(set); 486 start = -1; 487 token = __DECONST(char *, list); 488 for (;;) { 489 pcpu = strtoul(token, &cp, 0); 490 if (cp == token) 491 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 492 if (pcpu < 0 || pcpu >= CPU_SETSIZE) 493 errx(4, "hostcpu '%d' outside valid range from 0 to %d", 494 pcpu, CPU_SETSIZE - 1); 495 switch (*cp) { 496 case ',': 497 case '\0': 498 if (start >= 0) { 499 if (start > pcpu) 500 errx(4, "Invalid hostcpu range %d-%d", 501 start, pcpu); 502 while (start < pcpu) { 503 CPU_SET(start, vcpumap[vcpu]); 504 start++; 505 } 506 start = -1; 507 } 508 CPU_SET(pcpu, vcpumap[vcpu]); 509 break; 510 case '-': 511 if (start >= 0) 512 errx(4, "invalid cpuset for vcpu %d: '%s'", 513 vcpu, list); 514 start = pcpu; 515 break; 516 default: 517 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 518 } 519 if (*cp == '\0') 520 break; 521 token = cp + 1; 522 } 523 } 524 525 static void 526 build_vcpumaps(void) 527 { 528 char key[16]; 529 const char *value; 530 int vcpu; 531 532 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) { 533 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 534 value = get_config_value(key); 535 if (value == NULL) 536 continue; 537 vcpumap[vcpu] = malloc(sizeof(cpuset_t)); 538 if (vcpumap[vcpu] == NULL) 539 err(4, "Failed to allocate cpuset for vcpu %d", vcpu); 540 parse_cpuset(vcpu, value, vcpumap[vcpu]); 541 } 542 } 543 544 void 545 vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid, 546 int errcode) 547 { 548 struct vmctx *ctx; 549 int error, restart_instruction; 550 551 ctx = arg; 552 restart_instruction = 1; 553 554 error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode, 555 restart_instruction); 556 assert(error == 0); 557 } 558 #endif /* __FreeBSD__ */ 559 560 void * 561 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len) 562 { 563 564 return (vm_map_gpa(ctx, gaddr, len)); 565 } 566 567 int 568 fbsdrun_virtio_msix(void) 569 { 570 571 return (get_config_bool_default("virtio_msix", true)); 572 } 573 574 static void * 575 fbsdrun_start_thread(void *param) 576 { 577 char tname[MAXCOMLEN + 1]; 578 struct mt_vmm_info *mtp; 579 int vcpu; 580 581 mtp = param; 582 vcpu = mtp->mt_vcpu; 583 584 snprintf(tname, sizeof(tname), "vcpu %d", vcpu); 585 pthread_set_name_np(mtp->mt_thr, tname); 586 587 gdb_cpu_add(vcpu); 588 589 vm_loop(mtp->mt_ctx, vcpu, mtp->mt_startrip); 590 591 /* not reached */ 592 exit(1); 593 return (NULL); 594 } 595 596 #ifdef __FreeBSD__ 597 void 598 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip) 599 #else 600 void 601 fbsdrun_addcpu(struct vmctx *ctx, int newcpu, uint64_t rip, bool suspend) 602 #endif 603 { 604 int error; 605 606 #ifdef __FreeBSD__ 607 assert(fromcpu == BSP); 608 #endif 609 610 /* 611 * The 'newcpu' must be activated in the context of 'fromcpu'. If 612 * vm_activate_cpu() is delayed until newcpu's pthread starts running 613 * then vmm.ko is out-of-sync with bhyve and this can create a race 614 * with vm_suspend(). 615 */ 616 error = vm_activate_cpu(ctx, newcpu); 617 if (error != 0) 618 err(EX_OSERR, "could not activate CPU %d", newcpu); 619 620 CPU_SET_ATOMIC(newcpu, &cpumask); 621 622 #ifndef __FreeBSD__ 623 if (suspend) 624 (void) vm_suspend_cpu(ctx, newcpu); 625 #endif 626 627 /* 628 * Set up the vmexit struct to allow execution to start 629 * at the given RIP 630 */ 631 mt_vmm_info[newcpu].mt_ctx = ctx; 632 mt_vmm_info[newcpu].mt_vcpu = newcpu; 633 mt_vmm_info[newcpu].mt_startrip = rip; 634 635 error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL, 636 fbsdrun_start_thread, &mt_vmm_info[newcpu]); 637 assert(error == 0); 638 } 639 640 static int 641 fbsdrun_deletecpu(struct vmctx *ctx, int vcpu) 642 { 643 644 if (!CPU_ISSET(vcpu, &cpumask)) { 645 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu); 646 exit(4); 647 } 648 649 CPU_CLR_ATOMIC(vcpu, &cpumask); 650 return (CPU_EMPTY(&cpumask)); 651 } 652 653 static void 654 vmentry_mmio_read(int vcpu, uint64_t gpa, uint8_t bytes, uint64_t data) 655 { 656 struct vm_entry *entry = &vmentry[vcpu]; 657 struct vm_mmio *mmio = &entry->u.mmio; 658 659 assert(entry->cmd == VEC_DEFAULT); 660 661 entry->cmd = VEC_FULFILL_MMIO; 662 mmio->bytes = bytes; 663 mmio->read = 1; 664 mmio->gpa = gpa; 665 mmio->data = data; 666 } 667 668 static void 669 vmentry_mmio_write(int vcpu, uint64_t gpa, uint8_t bytes) 670 { 671 struct vm_entry *entry = &vmentry[vcpu]; 672 struct vm_mmio *mmio = &entry->u.mmio; 673 674 assert(entry->cmd == VEC_DEFAULT); 675 676 entry->cmd = VEC_FULFILL_MMIO; 677 mmio->bytes = bytes; 678 mmio->read = 0; 679 mmio->gpa = gpa; 680 mmio->data = 0; 681 } 682 683 static void 684 vmentry_inout_read(int vcpu, uint16_t port, uint8_t bytes, uint32_t data) 685 { 686 struct vm_entry *entry = &vmentry[vcpu]; 687 struct vm_inout *inout = &entry->u.inout; 688 689 assert(entry->cmd == VEC_DEFAULT); 690 691 entry->cmd = VEC_FULFILL_INOUT; 692 inout->bytes = bytes; 693 inout->flags = INOUT_IN; 694 inout->port = port; 695 inout->eax = data; 696 } 697 698 static void 699 vmentry_inout_write(int vcpu, uint16_t port, uint8_t bytes) 700 { 701 struct vm_entry *entry = &vmentry[vcpu]; 702 struct vm_inout *inout = &entry->u.inout; 703 704 assert(entry->cmd == VEC_DEFAULT); 705 706 entry->cmd = VEC_FULFILL_INOUT; 707 inout->bytes = bytes; 708 inout->flags = 0; 709 inout->port = port; 710 inout->eax = 0; 711 } 712 713 static int 714 vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu, 715 uint32_t eax) 716 { 717 #if BHYVE_DEBUG 718 /* 719 * put guest-driven debug here 720 */ 721 #endif 722 return (VMEXIT_CONTINUE); 723 } 724 725 static int 726 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 727 { 728 int error; 729 int vcpu; 730 struct vm_inout inout; 731 bool in; 732 uint8_t bytes; 733 734 stats.vmexit_inout++; 735 736 vcpu = *pvcpu; 737 inout = vme->u.inout; 738 in = (inout.flags & INOUT_IN) != 0; 739 bytes = inout.bytes; 740 741 /* Extra-special case of host notifications */ 742 if (!in && inout.port == GUEST_NIO_PORT) { 743 error = vmexit_handle_notify(ctx, vme, pvcpu, inout.eax); 744 vmentry_inout_write(vcpu, inout.port, bytes); 745 return (error); 746 } 747 748 error = emulate_inout(ctx, vcpu, &inout); 749 if (error) { 750 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n", 751 in ? "in" : "out", 752 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'), 753 inout.port, vmexit->rip); 754 return (VMEXIT_ABORT); 755 } else { 756 /* 757 * Communicate the status of the inout operation back to the 758 * in-kernel instruction emulation. 759 */ 760 if (in) { 761 vmentry_inout_read(vcpu, inout.port, bytes, inout.eax); 762 } else { 763 vmentry_inout_write(vcpu, inout.port, bytes); 764 } 765 return (VMEXIT_CONTINUE); 766 } 767 } 768 769 static int 770 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 771 { 772 uint64_t val; 773 uint32_t eax, edx; 774 int error; 775 776 val = 0; 777 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val); 778 if (error != 0) { 779 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n", 780 vme->u.msr.code, *pvcpu); 781 if (get_config_bool("x86.strictmsr")) { 782 vm_inject_gp(ctx, *pvcpu); 783 return (VMEXIT_CONTINUE); 784 } 785 } 786 787 eax = val; 788 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax); 789 assert(error == 0); 790 791 edx = val >> 32; 792 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx); 793 assert(error == 0); 794 795 return (VMEXIT_CONTINUE); 796 } 797 798 static int 799 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 800 { 801 int error; 802 803 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval); 804 if (error != 0) { 805 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n", 806 vme->u.msr.code, vme->u.msr.wval, *pvcpu); 807 if (get_config_bool("x86.strictmsr")) { 808 vm_inject_gp(ctx, *pvcpu); 809 return (VMEXIT_CONTINUE); 810 } 811 } 812 return (VMEXIT_CONTINUE); 813 } 814 815 #ifdef __FreeBSD__ 816 static int 817 vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 818 { 819 820 (void)spinup_ap(ctx, *pvcpu, 821 vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip); 822 823 return (VMEXIT_CONTINUE); 824 } 825 #else 826 static int 827 vmexit_run_state(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 828 { 829 /* 830 * Run-state transitions (INIT, SIPI, etc) are handled in-kernel, so an 831 * exit to userspace with that code is not expected. 832 */ 833 fprintf(stderr, "unexpected run-state VM exit"); 834 return (VMEXIT_ABORT); 835 } 836 #endif /* __FreeBSD__ */ 837 838 #ifdef __FreeBSD__ 839 #define DEBUG_EPT_MISCONFIG 840 #else 841 /* EPT misconfig debugging not possible now that raw VMCS access is gone */ 842 #endif 843 844 #ifdef DEBUG_EPT_MISCONFIG 845 #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400 846 847 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4]; 848 static int ept_misconfig_ptenum; 849 #endif 850 851 static const char * 852 vmexit_vmx_desc(uint32_t exit_reason) 853 { 854 855 if (exit_reason >= nitems(vmx_exit_reason_desc) || 856 vmx_exit_reason_desc[exit_reason] == NULL) 857 return ("Unknown"); 858 return (vmx_exit_reason_desc[exit_reason]); 859 } 860 861 static int 862 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 863 { 864 865 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 866 fprintf(stderr, "\treason\t\tVMX\n"); 867 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 868 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 869 fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status); 870 fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason, 871 vmexit_vmx_desc(vmexit->u.vmx.exit_reason)); 872 fprintf(stderr, "\tqualification\t0x%016lx\n", 873 vmexit->u.vmx.exit_qualification); 874 fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type); 875 fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error); 876 #ifdef DEBUG_EPT_MISCONFIG 877 if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) { 878 vm_get_register(ctx, *pvcpu, 879 VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS), 880 &ept_misconfig_gpa); 881 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte, 882 &ept_misconfig_ptenum); 883 fprintf(stderr, "\tEPT misconfiguration:\n"); 884 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa); 885 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n", 886 ept_misconfig_ptenum, ept_misconfig_pte[0], 887 ept_misconfig_pte[1], ept_misconfig_pte[2], 888 ept_misconfig_pte[3]); 889 } 890 #endif /* DEBUG_EPT_MISCONFIG */ 891 return (VMEXIT_ABORT); 892 } 893 894 static int 895 vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 896 { 897 898 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 899 fprintf(stderr, "\treason\t\tSVM\n"); 900 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 901 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 902 fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode); 903 fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1); 904 fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2); 905 return (VMEXIT_ABORT); 906 } 907 908 static int 909 vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 910 { 911 912 assert(vmexit->inst_length == 0); 913 914 stats.vmexit_bogus++; 915 916 return (VMEXIT_CONTINUE); 917 } 918 919 static int 920 vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 921 { 922 923 assert(vmexit->inst_length == 0); 924 925 stats.vmexit_reqidle++; 926 927 return (VMEXIT_CONTINUE); 928 } 929 930 static int 931 vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 932 { 933 934 stats.vmexit_hlt++; 935 936 /* 937 * Just continue execution with the next instruction. We use 938 * the HLT VM exit as a way to be friendly with the host 939 * scheduler. 940 */ 941 return (VMEXIT_CONTINUE); 942 } 943 944 static int 945 vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 946 { 947 948 stats.vmexit_pause++; 949 950 return (VMEXIT_CONTINUE); 951 } 952 953 static int 954 vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 955 { 956 957 assert(vmexit->inst_length == 0); 958 959 stats.vmexit_mtrap++; 960 961 gdb_cpu_mtrap(*pvcpu); 962 963 return (VMEXIT_CONTINUE); 964 } 965 966 static int 967 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 968 { 969 uint8_t i, valid; 970 971 fprintf(stderr, "Failed to emulate instruction sequence "); 972 973 valid = vmexit->u.inst_emul.num_valid; 974 if (valid != 0) { 975 assert(valid <= sizeof (vmexit->u.inst_emul.inst)); 976 fprintf(stderr, "["); 977 for (i = 0; i < valid; i++) { 978 if (i == 0) { 979 fprintf(stderr, "%02x", 980 vmexit->u.inst_emul.inst[i]); 981 } else { 982 fprintf(stderr, ", %02x", 983 vmexit->u.inst_emul.inst[i]); 984 } 985 } 986 fprintf(stderr, "] "); 987 } 988 fprintf(stderr, "@ %rip = %x\n", vmexit->rip); 989 990 return (VMEXIT_ABORT); 991 } 992 993 static int 994 vmexit_mmio(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 995 { 996 int vcpu, err; 997 struct vm_mmio mmio; 998 bool is_read; 999 1000 stats.vmexit_mmio++; 1001 1002 vcpu = *pvcpu; 1003 mmio = vmexit->u.mmio; 1004 is_read = (mmio.read != 0); 1005 1006 err = emulate_mem(ctx, vcpu, &mmio); 1007 1008 if (err == ESRCH) { 1009 fprintf(stderr, "Unhandled memory access to 0x%lx\n", mmio.gpa); 1010 stats.mmio_unhandled++; 1011 1012 /* 1013 * Access to non-existent physical addresses is not likely to 1014 * result in fatal errors on hardware machines, but rather reads 1015 * of all-ones or discarded-but-acknowledged writes. 1016 */ 1017 mmio.data = ~0UL; 1018 err = 0; 1019 } 1020 1021 if (err == 0) { 1022 if (is_read) { 1023 vmentry_mmio_read(vcpu, mmio.gpa, mmio.bytes, 1024 mmio.data); 1025 } else { 1026 vmentry_mmio_write(vcpu, mmio.gpa, mmio.bytes); 1027 } 1028 return (VMEXIT_CONTINUE); 1029 } 1030 1031 fprintf(stderr, "Unhandled mmio error to 0x%lx: %d\n", mmio.gpa, err); 1032 return (VMEXIT_ABORT); 1033 } 1034 1035 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER; 1036 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER; 1037 1038 static int 1039 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1040 { 1041 enum vm_suspend_how how; 1042 1043 how = vmexit->u.suspended.how; 1044 1045 fbsdrun_deletecpu(ctx, *pvcpu); 1046 1047 if (*pvcpu != BSP) { 1048 pthread_mutex_lock(&resetcpu_mtx); 1049 pthread_cond_signal(&resetcpu_cond); 1050 pthread_mutex_unlock(&resetcpu_mtx); 1051 pthread_exit(NULL); 1052 } 1053 1054 pthread_mutex_lock(&resetcpu_mtx); 1055 while (!CPU_EMPTY(&cpumask)) { 1056 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx); 1057 } 1058 pthread_mutex_unlock(&resetcpu_mtx); 1059 1060 switch (how) { 1061 case VM_SUSPEND_RESET: 1062 exit(0); 1063 case VM_SUSPEND_POWEROFF: 1064 if (get_config_bool_default("destroy_on_poweroff", false)) 1065 vm_destroy(ctx); 1066 exit(1); 1067 case VM_SUSPEND_HALT: 1068 exit(2); 1069 case VM_SUSPEND_TRIPLEFAULT: 1070 exit(3); 1071 default: 1072 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how); 1073 exit(100); 1074 } 1075 return (0); /* NOTREACHED */ 1076 } 1077 1078 static int 1079 vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1080 { 1081 1082 gdb_cpu_suspend(*pvcpu); 1083 return (VMEXIT_CONTINUE); 1084 } 1085 1086 static int 1087 vmexit_breakpoint(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1088 { 1089 1090 gdb_cpu_breakpoint(*pvcpu, vmexit); 1091 return (VMEXIT_CONTINUE); 1092 } 1093 1094 static vmexit_handler_t handler[VM_EXITCODE_MAX] = { 1095 [VM_EXITCODE_INOUT] = vmexit_inout, 1096 [VM_EXITCODE_MMIO] = vmexit_mmio, 1097 [VM_EXITCODE_VMX] = vmexit_vmx, 1098 [VM_EXITCODE_SVM] = vmexit_svm, 1099 [VM_EXITCODE_BOGUS] = vmexit_bogus, 1100 [VM_EXITCODE_REQIDLE] = vmexit_reqidle, 1101 [VM_EXITCODE_RDMSR] = vmexit_rdmsr, 1102 [VM_EXITCODE_WRMSR] = vmexit_wrmsr, 1103 [VM_EXITCODE_MTRAP] = vmexit_mtrap, 1104 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul, 1105 #ifdef __FreeBSD__ 1106 [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap, 1107 #else 1108 [VM_EXITCODE_RUN_STATE] = vmexit_run_state, 1109 #endif 1110 [VM_EXITCODE_SUSPENDED] = vmexit_suspend, 1111 [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch, 1112 [VM_EXITCODE_DEBUG] = vmexit_debug, 1113 [VM_EXITCODE_BPT] = vmexit_breakpoint, 1114 }; 1115 1116 static void 1117 vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip) 1118 { 1119 int error, rc; 1120 enum vm_exitcode exitcode; 1121 cpuset_t active_cpus; 1122 struct vm_exit *vexit; 1123 struct vm_entry *ventry; 1124 1125 #ifdef __FreeBSD__ 1126 if (vcpumap[vcpu] != NULL) { 1127 error = pthread_setaffinity_np(pthread_self(), 1128 sizeof(cpuset_t), vcpumap[vcpu]); 1129 assert(error == 0); 1130 } 1131 #endif 1132 error = vm_active_cpus(ctx, &active_cpus); 1133 assert(CPU_ISSET(vcpu, &active_cpus)); 1134 1135 error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip); 1136 assert(error == 0); 1137 1138 ventry = &vmentry[vcpu]; 1139 vexit = &vmexit[vcpu]; 1140 1141 while (1) { 1142 error = vm_run(ctx, vcpu, ventry, vexit); 1143 if (error != 0) 1144 break; 1145 1146 if (ventry->cmd != VEC_DEFAULT) { 1147 /* 1148 * Discard any lingering entry state after it has been 1149 * submitted via vm_run(). 1150 */ 1151 bzero(ventry, sizeof (*ventry)); 1152 } 1153 1154 exitcode = vexit->exitcode; 1155 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) { 1156 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n", 1157 exitcode); 1158 exit(4); 1159 } 1160 1161 rc = (*handler[exitcode])(ctx, vexit, &vcpu); 1162 1163 switch (rc) { 1164 case VMEXIT_CONTINUE: 1165 break; 1166 case VMEXIT_ABORT: 1167 abort(); 1168 default: 1169 exit(4); 1170 } 1171 } 1172 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno); 1173 } 1174 1175 static int 1176 num_vcpus_allowed(struct vmctx *ctx) 1177 { 1178 #ifdef __FreeBSD__ 1179 int tmp, error; 1180 1181 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp); 1182 1183 /* 1184 * The guest is allowed to spinup more than one processor only if the 1185 * UNRESTRICTED_GUEST capability is available. 1186 */ 1187 if (error == 0) 1188 return (VM_MAXCPU); 1189 else 1190 return (1); 1191 #else 1192 /* Unrestricted Guest is always enabled on illumos */ 1193 return (VM_MAXCPU); 1194 #endif /* __FreeBSD__ */ 1195 } 1196 1197 void 1198 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu) 1199 { 1200 int err, tmp; 1201 1202 if (get_config_bool_default("x86.vmexit_on_hlt", false)) { 1203 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp); 1204 if (err < 0) { 1205 fprintf(stderr, "VM exit on HLT not supported\n"); 1206 exit(4); 1207 } 1208 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1); 1209 if (cpu == BSP) 1210 handler[VM_EXITCODE_HLT] = vmexit_hlt; 1211 } 1212 1213 if (get_config_bool_default("x86.vmexit_on_pause", false)) { 1214 /* 1215 * pause exit support required for this mode 1216 */ 1217 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp); 1218 if (err < 0) { 1219 fprintf(stderr, 1220 "SMP mux requested, no pause support\n"); 1221 exit(4); 1222 } 1223 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1); 1224 if (cpu == BSP) 1225 handler[VM_EXITCODE_PAUSE] = vmexit_pause; 1226 } 1227 1228 if (get_config_bool_default("x86.x2apic", false)) 1229 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED); 1230 else 1231 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED); 1232 1233 if (err) { 1234 fprintf(stderr, "Unable to set x2apic state (%d)\n", err); 1235 exit(4); 1236 } 1237 1238 #ifdef __FreeBSD__ 1239 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1); 1240 #endif 1241 } 1242 1243 static struct vmctx * 1244 do_open(const char *vmname) 1245 { 1246 struct vmctx *ctx; 1247 int error; 1248 bool reinit, romboot; 1249 #ifndef WITHOUT_CAPSICUM 1250 cap_rights_t rights; 1251 const cap_ioctl_t *cmds; 1252 size_t ncmds; 1253 #endif 1254 1255 reinit = romboot = false; 1256 1257 if (lpc_bootrom()) 1258 romboot = true; 1259 #ifndef __FreeBSD__ 1260 uint64_t create_flags = 0; 1261 if (get_config_bool_default("memory.use_reservoir", false)) { 1262 create_flags |= VCF_RESERVOIR_MEM; 1263 } 1264 error = vm_create(vmname, create_flags); 1265 #else 1266 error = vm_create(vmname); 1267 #endif /* __FreeBSD__ */ 1268 if (error) { 1269 if (errno == EEXIST) { 1270 if (romboot) { 1271 reinit = true; 1272 } else { 1273 /* 1274 * The virtual machine has been setup by the 1275 * userspace bootloader. 1276 */ 1277 } 1278 } else { 1279 perror("vm_create"); 1280 exit(4); 1281 } 1282 } else { 1283 if (!romboot) { 1284 /* 1285 * If the virtual machine was just created then a 1286 * bootrom must be configured to boot it. 1287 */ 1288 fprintf(stderr, "virtual machine cannot be booted\n"); 1289 exit(4); 1290 } 1291 } 1292 1293 ctx = vm_open(vmname); 1294 if (ctx == NULL) { 1295 perror("vm_open"); 1296 exit(4); 1297 } 1298 1299 #ifndef WITHOUT_CAPSICUM 1300 cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW); 1301 if (caph_rights_limit(vm_get_device_fd(ctx), &rights) == -1) 1302 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1303 vm_get_ioctls(&ncmds); 1304 cmds = vm_get_ioctls(NULL); 1305 if (cmds == NULL) 1306 errx(EX_OSERR, "out of memory"); 1307 if (caph_ioctls_limit(vm_get_device_fd(ctx), cmds, ncmds) == -1) 1308 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1309 free((cap_ioctl_t *)cmds); 1310 #endif 1311 1312 if (reinit) { 1313 #ifndef __FreeBSD__ 1314 error = vm_reinit(ctx, 0); 1315 #else 1316 error = vm_reinit(ctx); 1317 #endif 1318 if (error) { 1319 perror("vm_reinit"); 1320 exit(4); 1321 } 1322 } 1323 error = vm_set_topology(ctx, sockets, cores, threads, maxcpus); 1324 if (error) 1325 errx(EX_OSERR, "vm_set_topology"); 1326 return (ctx); 1327 } 1328 1329 static bool 1330 parse_config_option(const char *option) 1331 { 1332 const char *value; 1333 char *path; 1334 1335 value = strchr(option, '='); 1336 if (value == NULL || value[1] == '\0') 1337 return (false); 1338 path = strndup(option, value - option); 1339 if (path == NULL) 1340 err(4, "Failed to allocate memory"); 1341 set_config_value(path, value + 1); 1342 return (true); 1343 } 1344 1345 static void 1346 parse_simple_config_file(const char *path) 1347 { 1348 FILE *fp; 1349 char *line, *cp; 1350 size_t linecap; 1351 unsigned int lineno; 1352 1353 fp = fopen(path, "r"); 1354 if (fp == NULL) 1355 err(4, "Failed to open configuration file %s", path); 1356 line = NULL; 1357 linecap = 0; 1358 lineno = 1; 1359 for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) { 1360 if (*line == '#' || *line == '\n') 1361 continue; 1362 cp = strchr(line, '\n'); 1363 if (cp != NULL) 1364 *cp = '\0'; 1365 if (!parse_config_option(line)) 1366 errx(4, "%s line %u: invalid config option '%s'", path, 1367 lineno, line); 1368 } 1369 free(line); 1370 fclose(fp); 1371 } 1372 1373 static void 1374 parse_gdb_options(char *optarg) 1375 { 1376 const char *sport; 1377 char *colon; 1378 1379 if (optarg[0] == 'w') { 1380 set_config_bool("gdb.wait", true); 1381 optarg++; 1382 } 1383 1384 colon = strrchr(optarg, ':'); 1385 if (colon == NULL) { 1386 sport = optarg; 1387 } else { 1388 *colon = '\0'; 1389 colon++; 1390 sport = colon; 1391 set_config_value("gdb.address", optarg); 1392 } 1393 1394 set_config_value("gdb.port", sport); 1395 } 1396 1397 static void 1398 set_defaults(void) 1399 { 1400 1401 set_config_bool("acpi_tables", false); 1402 set_config_value("memory.size", "256M"); 1403 set_config_bool("x86.strictmsr", true); 1404 } 1405 1406 int 1407 main(int argc, char *argv[]) 1408 { 1409 int c, error, err; 1410 int max_vcpus, memflags; 1411 struct vmctx *ctx; 1412 uint64_t rip; 1413 size_t memsize; 1414 const char *value, *vmname; 1415 char *optstr; 1416 1417 init_config(); 1418 set_defaults(); 1419 progname = basename(argv[0]); 1420 1421 #ifdef __FreeBSD__ 1422 optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:U:"; 1423 #else 1424 /* +d, +B, -p */ 1425 optstr = "adehuwxACDHIPSWYk:o:G:c:s:m:l:B:U:"; 1426 #endif 1427 while ((c = getopt(argc, argv, optstr)) != -1) { 1428 switch (c) { 1429 case 'a': 1430 set_config_bool("x86.x2apic", false); 1431 break; 1432 case 'A': 1433 #ifdef __FreeBSD__ 1434 /* 1435 * This option is ignored on illumos since the 1436 * generated ACPI tables are not used; the bootroms 1437 * have their own. The option is retained for backwards 1438 * compatibility but does nothing. Note that the 1439 * acpi_tables configuration is still accepted via 1440 * -o if somebody really wants to generate these tables. 1441 */ 1442 set_config_bool("acpi_tables", true); 1443 #endif 1444 break; 1445 case 'D': 1446 set_config_bool("destroy_on_poweroff", true); 1447 break; 1448 #ifndef __FreeBSD__ 1449 case 'B': 1450 if (smbios_parse(optarg) != 0) { 1451 errx(EX_USAGE, "invalid SMBIOS " 1452 "configuration '%s'", optarg); 1453 } 1454 break; 1455 case 'd': 1456 set_config_bool("suspend_at_boot", true); 1457 break; 1458 #endif 1459 #ifdef __FreeBSD__ 1460 case 'p': 1461 if (pincpu_parse(optarg) != 0) { 1462 errx(EX_USAGE, "invalid vcpu pinning " 1463 "configuration '%s'", optarg); 1464 } 1465 break; 1466 #endif 1467 case 'c': 1468 if (topology_parse(optarg) != 0) { 1469 errx(EX_USAGE, "invalid cpu topology " 1470 "'%s'", optarg); 1471 } 1472 break; 1473 case 'C': 1474 set_config_bool("memory.guest_in_core", true); 1475 break; 1476 case 'G': 1477 parse_gdb_options(optarg); 1478 break; 1479 case 'k': 1480 parse_simple_config_file(optarg); 1481 break; 1482 case 'l': 1483 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1484 lpc_print_supported_devices(); 1485 exit(0); 1486 } else if (lpc_device_parse(optarg) != 0) { 1487 errx(EX_USAGE, "invalid lpc device " 1488 "configuration '%s'", optarg); 1489 } 1490 break; 1491 case 's': 1492 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1493 pci_print_supported_devices(); 1494 exit(0); 1495 } else if (pci_parse_slot(optarg) != 0) 1496 exit(4); 1497 else 1498 break; 1499 case 'S': 1500 set_config_bool("memory.wired", true); 1501 break; 1502 case 'm': 1503 set_config_value("memory.size", optarg); 1504 break; 1505 case 'o': 1506 if (!parse_config_option(optarg)) 1507 errx(EX_USAGE, "invalid configuration option '%s'", optarg); 1508 break; 1509 case 'H': 1510 set_config_bool("x86.vmexit_on_hlt", true); 1511 break; 1512 case 'I': 1513 /* 1514 * The "-I" option was used to add an ioapic to the 1515 * virtual machine. 1516 * 1517 * An ioapic is now provided unconditionally for each 1518 * virtual machine and this option is now deprecated. 1519 */ 1520 break; 1521 case 'P': 1522 set_config_bool("x86.vmexit_on_pause", true); 1523 break; 1524 case 'e': 1525 set_config_bool("x86.strictio", true); 1526 break; 1527 case 'u': 1528 set_config_bool("rtc.use_localtime", false); 1529 break; 1530 case 'U': 1531 set_config_value("uuid", optarg); 1532 break; 1533 case 'w': 1534 set_config_bool("x86.strictmsr", false); 1535 break; 1536 case 'W': 1537 set_config_bool("virtio_msix", false); 1538 break; 1539 case 'x': 1540 set_config_bool("x86.x2apic", true); 1541 break; 1542 case 'Y': 1543 set_config_bool("x86.mptable", false); 1544 break; 1545 case 'h': 1546 usage(0); 1547 default: 1548 usage(1); 1549 } 1550 } 1551 argc -= optind; 1552 argv += optind; 1553 1554 if (argc > 1) 1555 usage(1); 1556 1557 if (argc == 1) 1558 set_config_value("name", argv[0]); 1559 1560 vmname = get_config_value("name"); 1561 if (vmname == NULL) 1562 usage(1); 1563 1564 if (get_config_bool_default("config.dump", false)) { 1565 dump_config(); 1566 exit(1); 1567 } 1568 1569 #ifndef __FreeBSD__ 1570 illumos_priv_init(); 1571 #endif 1572 1573 calc_topolopgy(); 1574 #ifdef __FreeBSD__ 1575 build_vcpumaps(); 1576 #endif 1577 1578 value = get_config_value("memory.size"); 1579 error = vm_parse_memsize(value, &memsize); 1580 if (error) 1581 errx(EX_USAGE, "invalid memsize '%s'", value); 1582 1583 ctx = do_open(vmname); 1584 1585 max_vcpus = num_vcpus_allowed(ctx); 1586 if (guest_ncpus > max_vcpus) { 1587 fprintf(stderr, "%d vCPUs requested but only %d available\n", 1588 guest_ncpus, max_vcpus); 1589 exit(4); 1590 } 1591 1592 fbsdrun_set_capabilities(ctx, BSP); 1593 1594 memflags = 0; 1595 if (get_config_bool_default("memory.wired", false)) 1596 memflags |= VM_MEM_F_WIRED; 1597 if (get_config_bool_default("memory.guest_in_core", false)) 1598 memflags |= VM_MEM_F_INCORE; 1599 vm_set_memflags(ctx, memflags); 1600 #ifdef __FreeBSD__ 1601 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1602 #else 1603 do { 1604 errno = 0; 1605 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1606 error = errno; 1607 if (err != 0 && error == ENOMEM) { 1608 (void) fprintf(stderr, "Unable to allocate memory " 1609 "(%llu), retrying in 1 second\n", memsize); 1610 sleep(1); 1611 } 1612 } while (error == ENOMEM); 1613 #endif 1614 if (err) { 1615 fprintf(stderr, "Unable to set up memory (%d)\n", errno); 1616 exit(4); 1617 } 1618 1619 error = init_msr(); 1620 if (error) { 1621 fprintf(stderr, "init_msr error %d", error); 1622 exit(4); 1623 } 1624 1625 init_mem(); 1626 init_inout(); 1627 #ifdef __FreeBSD__ 1628 kernemu_dev_init(); 1629 #endif 1630 init_bootrom(ctx); 1631 atkbdc_init(ctx); 1632 pci_irq_init(ctx); 1633 ioapic_init(ctx); 1634 1635 rtc_init(ctx); 1636 sci_init(ctx); 1637 #ifndef __FreeBSD__ 1638 pmtmr_init(ctx); 1639 #endif 1640 1641 /* 1642 * Exit if a device emulation finds an error in its initilization 1643 */ 1644 if (init_pci(ctx) != 0) { 1645 perror("device emulation initialization error"); 1646 exit(4); 1647 } 1648 1649 /* 1650 * Initialize after PCI, to allow a bootrom file to reserve the high 1651 * region. 1652 */ 1653 if (get_config_bool("acpi_tables")) 1654 vmgenc_init(ctx); 1655 1656 #ifdef __FreeBSD__ 1657 init_gdb(ctx); 1658 #else 1659 if (value != NULL) { 1660 int port = atoi(value); 1661 1662 if (port < 0) 1663 init_mdb(ctx); 1664 else 1665 init_gdb(ctx); 1666 } 1667 #endif 1668 1669 if (lpc_bootrom()) { 1670 #ifdef __FreeBSD__ 1671 if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) { 1672 fprintf(stderr, "ROM boot failed: unrestricted guest " 1673 "capability not available\n"); 1674 exit(4); 1675 } 1676 #else 1677 /* Unrestricted Guest is always enabled on illumos */ 1678 #endif 1679 error = vcpu_reset(ctx, BSP); 1680 assert(error == 0); 1681 } 1682 1683 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip); 1684 assert(error == 0); 1685 1686 /* 1687 * build the guest tables, MP etc. 1688 */ 1689 if (get_config_bool_default("x86.mptable", true)) { 1690 error = mptable_build(ctx, guest_ncpus); 1691 if (error) { 1692 perror("error to build the guest tables"); 1693 exit(4); 1694 } 1695 } 1696 1697 #ifndef __FreeBSD__ 1698 smbios_apply(); 1699 #endif 1700 error = smbios_build(ctx); 1701 assert(error == 0); 1702 1703 if (get_config_bool("acpi_tables")) { 1704 error = acpi_build(ctx, guest_ncpus); 1705 assert(error == 0); 1706 } 1707 1708 if (lpc_bootrom()) 1709 fwctl_init(); 1710 1711 /* 1712 * Change the proc title to include the VM name. 1713 */ 1714 setproctitle("%s", vmname); 1715 1716 #ifndef WITHOUT_CAPSICUM 1717 caph_cache_catpages(); 1718 1719 if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1) 1720 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1721 1722 if (caph_enter() == -1) 1723 errx(EX_OSERR, "cap_enter() failed"); 1724 #endif 1725 1726 #ifndef __FreeBSD__ 1727 illumos_priv_lock(); 1728 #endif 1729 1730 #ifdef __FreeBSD__ 1731 /* 1732 * Add CPU 0 1733 */ 1734 fbsdrun_addcpu(ctx, BSP, BSP, rip); 1735 #else 1736 /* Set BSP to run (unlike the APs which wait for INIT) */ 1737 error = vm_set_run_state(ctx, BSP, VRS_RUN, 0); 1738 assert(error == 0); 1739 fbsdrun_addcpu(ctx, BSP, rip, 1740 get_config_bool_default("suspend_at_boot", false)); 1741 1742 /* Add subsequent CPUs, which will wait until INIT/SIPI-ed */ 1743 for (uint_t i = 1; i < guest_ncpus; i++) { 1744 spinup_halted_ap(ctx, i); 1745 } 1746 #endif 1747 /* 1748 * Head off to the main event dispatch loop 1749 */ 1750 mevent_dispatch(); 1751 1752 exit(4); 1753 } 1754