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 2022 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; 206 #ifndef __FreeBSD__ 207 static struct vm_entry *vmentry; 208 #endif 209 210 struct bhyvestats { 211 uint64_t vmexit_bogus; 212 uint64_t vmexit_reqidle; 213 uint64_t vmexit_hlt; 214 uint64_t vmexit_pause; 215 uint64_t vmexit_mtrap; 216 uint64_t vmexit_mmio; 217 uint64_t vmexit_inout; 218 uint64_t cpu_switch_rotate; 219 uint64_t cpu_switch_direct; 220 uint64_t mmio_unhandled; 221 } stats; 222 223 struct mt_vmm_info { 224 pthread_t mt_thr; 225 struct vmctx *mt_ctx; 226 int mt_vcpu; 227 uint64_t mt_startrip; 228 } *mt_vmm_info; 229 230 #ifdef __FreeBSD__ 231 static cpuset_t **vcpumap; 232 #endif 233 234 static void 235 usage(int code) 236 { 237 238 fprintf(stderr, 239 #ifdef __FreeBSD__ 240 "Usage: %s [-AaCDeHhPSuWwxY]\n" 241 #else 242 "Usage: %s [-aCDdeHhPSuWwxY]\n" 243 #endif 244 " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n" 245 #ifdef __FreeBSD__ 246 " %*s [-G port] [-k config_file] [-l lpc] [-m mem] [-o var=value]\n" 247 " %*s [-p vcpu:hostcpu] [-r file] [-s pci] [-U uuid] vmname\n" 248 249 " -A: create ACPI tables\n" 250 #else 251 " %*s [-k <config_file>] [-l <lpc>] [-m mem] [-o <var>=<value>]\n" 252 " %*s [-s <pci>] [-U uuid] vmname\n" 253 #endif 254 " -a: local apic is in xAPIC mode (deprecated)\n" 255 #ifndef __FreeBSD__ 256 " -B type,key=value,...: set SMBIOS information\n" 257 #endif 258 " -C: include guest memory in core file\n" 259 " -c: number of cpus and/or topology specification\n" 260 " -D: destroy on power-off\n" 261 #ifndef __FreeBSD__ 262 " -d: suspend cpu at boot\n" 263 #endif 264 " -e: exit on unhandled I/O access\n" 265 " -H: vmexit from the guest on hlt\n" 266 " -h: help\n" 267 " -k: key=value flat config file\n" 268 " -K: PS2 keyboard layout\n" 269 " -l: LPC device configuration\n" 270 " -m: memory size\n" 271 " -o: set config 'var' to 'value'\n" 272 " -P: vmexit from the guest on pause\n" 273 #ifdef __FreeBSD__ 274 " -p: pin 'vcpu' to 'hostcpu'\n" 275 #endif 276 " -S: guest memory cannot be swapped\n" 277 " -s: <slot,driver,configinfo> PCI slot config\n" 278 " -U: uuid\n" 279 " -u: RTC keeps UTC time\n" 280 " -W: force virtio to use single-vector MSI\n" 281 " -w: ignore unimplemented MSRs\n" 282 " -x: local apic is in x2APIC mode\n" 283 " -Y: disable MPtable generation\n", 284 progname, (int)strlen(progname), "", (int)strlen(progname), "", 285 (int)strlen(progname), ""); 286 287 exit(code); 288 } 289 290 /* 291 * XXX This parser is known to have the following issues: 292 * 1. It accepts null key=value tokens ",," as setting "cpus" to an 293 * empty string. 294 * 295 * The acceptance of a null specification ('-c ""') is by design to match the 296 * manual page syntax specification, this results in a topology of 1 vCPU. 297 */ 298 static int 299 topology_parse(const char *opt) 300 { 301 char *cp, *str, *tofree; 302 303 if (*opt == '\0') { 304 set_config_value("sockets", "1"); 305 set_config_value("cores", "1"); 306 set_config_value("threads", "1"); 307 set_config_value("cpus", "1"); 308 return (0); 309 } 310 311 tofree = str = strdup(opt); 312 if (str == NULL) 313 errx(4, "Failed to allocate memory"); 314 315 while ((cp = strsep(&str, ",")) != NULL) { 316 if (strncmp(cp, "cpus=", strlen("cpus=")) == 0) 317 set_config_value("cpus", cp + strlen("cpus=")); 318 else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0) 319 set_config_value("sockets", cp + strlen("sockets=")); 320 else if (strncmp(cp, "cores=", strlen("cores=")) == 0) 321 set_config_value("cores", cp + strlen("cores=")); 322 else if (strncmp(cp, "threads=", strlen("threads=")) == 0) 323 set_config_value("threads", cp + strlen("threads=")); 324 #ifdef notyet /* Do not expose this until vmm.ko implements it */ 325 else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0) 326 set_config_value("maxcpus", cp + strlen("maxcpus=")); 327 #endif 328 else if (strchr(cp, '=') != NULL) 329 goto out; 330 else 331 set_config_value("cpus", cp); 332 } 333 free(tofree); 334 return (0); 335 336 out: 337 free(tofree); 338 return (-1); 339 } 340 341 static int 342 parse_int_value(const char *key, const char *value, int minval, int maxval) 343 { 344 char *cp; 345 long lval; 346 347 errno = 0; 348 lval = strtol(value, &cp, 0); 349 if (errno != 0 || *cp != '\0' || cp == value || lval < minval || 350 lval > maxval) 351 errx(4, "Invalid value for %s: '%s'", key, value); 352 return (lval); 353 } 354 355 /* 356 * Set the sockets, cores, threads, and guest_cpus variables based on 357 * the configured topology. 358 * 359 * The limits of UINT16_MAX are due to the types passed to 360 * vm_set_topology(). vmm.ko may enforce tighter limits. 361 */ 362 static void 363 calc_topolopgy(void) 364 { 365 const char *value; 366 bool explicit_cpus; 367 uint64_t ncpus; 368 369 value = get_config_value("cpus"); 370 if (value != NULL) { 371 guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX); 372 explicit_cpus = true; 373 } else { 374 guest_ncpus = 1; 375 explicit_cpus = false; 376 } 377 value = get_config_value("cores"); 378 if (value != NULL) 379 cores = parse_int_value("cores", value, 1, UINT16_MAX); 380 else 381 cores = 1; 382 value = get_config_value("threads"); 383 if (value != NULL) 384 threads = parse_int_value("threads", value, 1, UINT16_MAX); 385 else 386 threads = 1; 387 value = get_config_value("sockets"); 388 if (value != NULL) 389 sockets = parse_int_value("sockets", value, 1, UINT16_MAX); 390 else 391 sockets = guest_ncpus; 392 393 /* 394 * Compute sockets * cores * threads avoiding overflow. The 395 * range check above insures these are 16 bit values. 396 */ 397 ncpus = (uint64_t)sockets * cores * threads; 398 if (ncpus > UINT16_MAX) 399 errx(4, "Computed number of vCPUs too high: %ju", 400 (uintmax_t)ncpus); 401 402 if (explicit_cpus) { 403 if (guest_ncpus != ncpus) 404 errx(4, "Topology (%d sockets, %d cores, %d threads) " 405 "does not match %d vCPUs", sockets, cores, threads, 406 guest_ncpus); 407 } else 408 guest_ncpus = ncpus; 409 } 410 411 #ifndef WITHOUT_CAPSICUM 412 /* 413 * 11-stable capsicum helpers 414 */ 415 static void 416 bhyve_caph_cache_catpages(void) 417 { 418 419 (void)catopen("libc", NL_CAT_LOCALE); 420 } 421 422 static int 423 bhyve_caph_limit_stdoe(void) 424 { 425 cap_rights_t rights; 426 unsigned long cmds[] = { TIOCGETA, TIOCGWINSZ }; 427 int i, fds[] = { STDOUT_FILENO, STDERR_FILENO }; 428 429 cap_rights_init(&rights, CAP_FCNTL, CAP_FSTAT, CAP_IOCTL); 430 cap_rights_set(&rights, CAP_WRITE); 431 432 for (i = 0; i < nitems(fds); i++) { 433 if (cap_rights_limit(fds[i], &rights) < 0 && errno != ENOSYS) 434 return (-1); 435 436 if (cap_ioctls_limit(fds[i], cmds, nitems(cmds)) < 0 && errno != ENOSYS) 437 return (-1); 438 439 if (cap_fcntls_limit(fds[i], CAP_FCNTL_GETFL) < 0 && errno != ENOSYS) 440 return (-1); 441 } 442 443 return (0); 444 } 445 446 #endif 447 448 #ifdef __FreeBSD__ 449 static int 450 pincpu_parse(const char *opt) 451 { 452 int vcpu, pcpu; 453 454 if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) { 455 fprintf(stderr, "invalid format: %s\n", opt); 456 return (-1); 457 } 458 459 if (vcpu < 0) { 460 fprintf(stderr, "invalid vcpu '%d'\n", vcpu); 461 return (-1); 462 } 463 464 if (pcpu < 0 || pcpu >= CPU_SETSIZE) { 465 fprintf(stderr, "hostcpu '%d' outside valid range from " 466 "0 to %d\n", pcpu, CPU_SETSIZE - 1); 467 return (-1); 468 } 469 470 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 471 value = get_config_value(key); 472 473 if (asprintf(&newval, "%s%s%d", value != NULL ? value : "", 474 value != NULL ? "," : "", pcpu) == -1) { 475 perror("failed to build new cpuset string"); 476 return (-1); 477 } 478 479 set_config_value(key, newval); 480 free(newval); 481 return (0); 482 } 483 484 static void 485 parse_cpuset(int vcpu, const char *list, cpuset_t *set) 486 { 487 char *cp, *token; 488 int pcpu, start; 489 490 CPU_ZERO(set); 491 start = -1; 492 token = __DECONST(char *, list); 493 for (;;) { 494 pcpu = strtoul(token, &cp, 0); 495 if (cp == token) 496 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 497 if (pcpu < 0 || pcpu >= CPU_SETSIZE) 498 errx(4, "hostcpu '%d' outside valid range from 0 to %d", 499 pcpu, CPU_SETSIZE - 1); 500 switch (*cp) { 501 case ',': 502 case '\0': 503 if (start >= 0) { 504 if (start > pcpu) 505 errx(4, "Invalid hostcpu range %d-%d", 506 start, pcpu); 507 while (start < pcpu) { 508 CPU_SET(start, vcpumap[vcpu]); 509 start++; 510 } 511 start = -1; 512 } 513 CPU_SET(pcpu, vcpumap[vcpu]); 514 break; 515 case '-': 516 if (start >= 0) 517 errx(4, "invalid cpuset for vcpu %d: '%s'", 518 vcpu, list); 519 start = pcpu; 520 break; 521 default: 522 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 523 } 524 if (*cp == '\0') 525 break; 526 token = cp + 1; 527 } 528 } 529 530 static void 531 build_vcpumaps(void) 532 { 533 char key[16]; 534 const char *value; 535 int vcpu; 536 537 vcpumap = calloc(guest_ncpus, sizeof(*vcpumap)); 538 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) { 539 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 540 value = get_config_value(key); 541 if (value == NULL) 542 continue; 543 vcpumap[vcpu] = malloc(sizeof(cpuset_t)); 544 if (vcpumap[vcpu] == NULL) 545 err(4, "Failed to allocate cpuset for vcpu %d", vcpu); 546 parse_cpuset(vcpu, value, vcpumap[vcpu]); 547 } 548 } 549 550 void 551 vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid, 552 int errcode) 553 { 554 struct vmctx *ctx; 555 int error, restart_instruction; 556 557 ctx = arg; 558 restart_instruction = 1; 559 560 error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode, 561 restart_instruction); 562 assert(error == 0); 563 } 564 #endif /* __FreeBSD__ */ 565 566 void * 567 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len) 568 { 569 570 return (vm_map_gpa(ctx, gaddr, len)); 571 } 572 573 int 574 fbsdrun_virtio_msix(void) 575 { 576 577 return (get_config_bool_default("virtio_msix", true)); 578 } 579 580 static void * 581 fbsdrun_start_thread(void *param) 582 { 583 char tname[MAXCOMLEN + 1]; 584 struct mt_vmm_info *mtp; 585 int vcpu; 586 587 mtp = param; 588 vcpu = mtp->mt_vcpu; 589 590 snprintf(tname, sizeof(tname), "vcpu %d", vcpu); 591 pthread_set_name_np(mtp->mt_thr, tname); 592 593 gdb_cpu_add(vcpu); 594 595 vm_loop(mtp->mt_ctx, vcpu, mtp->mt_startrip); 596 597 /* not reached */ 598 exit(1); 599 return (NULL); 600 } 601 602 #ifdef __FreeBSD__ 603 void 604 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip) 605 #else 606 void 607 fbsdrun_addcpu(struct vmctx *ctx, int newcpu, uint64_t rip, bool suspend) 608 #endif 609 { 610 int error; 611 612 #ifdef __FreeBSD__ 613 assert(fromcpu == BSP); 614 #endif 615 616 /* 617 * The 'newcpu' must be activated in the context of 'fromcpu'. If 618 * vm_activate_cpu() is delayed until newcpu's pthread starts running 619 * then vmm.ko is out-of-sync with bhyve and this can create a race 620 * with vm_suspend(). 621 */ 622 error = vm_activate_cpu(ctx, newcpu); 623 if (error != 0) 624 err(EX_OSERR, "could not activate CPU %d", newcpu); 625 626 CPU_SET_ATOMIC(newcpu, &cpumask); 627 628 #ifndef __FreeBSD__ 629 if (suspend) 630 (void) vm_suspend_cpu(ctx, newcpu); 631 #endif 632 633 /* 634 * Set up the vmexit struct to allow execution to start 635 * at the given RIP 636 */ 637 mt_vmm_info[newcpu].mt_ctx = ctx; 638 mt_vmm_info[newcpu].mt_vcpu = newcpu; 639 mt_vmm_info[newcpu].mt_startrip = rip; 640 641 error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL, 642 fbsdrun_start_thread, &mt_vmm_info[newcpu]); 643 assert(error == 0); 644 } 645 646 static int 647 fbsdrun_deletecpu(struct vmctx *ctx, int vcpu) 648 { 649 650 if (!CPU_ISSET(vcpu, &cpumask)) { 651 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu); 652 exit(4); 653 } 654 655 CPU_CLR_ATOMIC(vcpu, &cpumask); 656 return (CPU_EMPTY(&cpumask)); 657 } 658 659 static void 660 vmentry_mmio_read(int vcpu, uint64_t gpa, uint8_t bytes, uint64_t data) 661 { 662 struct vm_entry *entry = &vmentry[vcpu]; 663 struct vm_mmio *mmio = &entry->u.mmio; 664 665 assert(entry->cmd == VEC_DEFAULT); 666 667 entry->cmd = VEC_FULFILL_MMIO; 668 mmio->bytes = bytes; 669 mmio->read = 1; 670 mmio->gpa = gpa; 671 mmio->data = data; 672 } 673 674 static void 675 vmentry_mmio_write(int vcpu, uint64_t gpa, uint8_t bytes) 676 { 677 struct vm_entry *entry = &vmentry[vcpu]; 678 struct vm_mmio *mmio = &entry->u.mmio; 679 680 assert(entry->cmd == VEC_DEFAULT); 681 682 entry->cmd = VEC_FULFILL_MMIO; 683 mmio->bytes = bytes; 684 mmio->read = 0; 685 mmio->gpa = gpa; 686 mmio->data = 0; 687 } 688 689 static void 690 vmentry_inout_read(int vcpu, uint16_t port, uint8_t bytes, uint32_t data) 691 { 692 struct vm_entry *entry = &vmentry[vcpu]; 693 struct vm_inout *inout = &entry->u.inout; 694 695 assert(entry->cmd == VEC_DEFAULT); 696 697 entry->cmd = VEC_FULFILL_INOUT; 698 inout->bytes = bytes; 699 inout->flags = INOUT_IN; 700 inout->port = port; 701 inout->eax = data; 702 } 703 704 static void 705 vmentry_inout_write(int vcpu, uint16_t port, uint8_t bytes) 706 { 707 struct vm_entry *entry = &vmentry[vcpu]; 708 struct vm_inout *inout = &entry->u.inout; 709 710 assert(entry->cmd == VEC_DEFAULT); 711 712 entry->cmd = VEC_FULFILL_INOUT; 713 inout->bytes = bytes; 714 inout->flags = 0; 715 inout->port = port; 716 inout->eax = 0; 717 } 718 719 static int 720 vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu, 721 uint32_t eax) 722 { 723 #if BHYVE_DEBUG 724 /* 725 * put guest-driven debug here 726 */ 727 #endif 728 return (VMEXIT_CONTINUE); 729 } 730 731 static int 732 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 733 { 734 int error; 735 int vcpu; 736 struct vm_inout inout; 737 bool in; 738 uint8_t bytes; 739 740 stats.vmexit_inout++; 741 742 vcpu = *pvcpu; 743 inout = vme->u.inout; 744 in = (inout.flags & INOUT_IN) != 0; 745 bytes = inout.bytes; 746 747 /* Extra-special case of host notifications */ 748 if (!in && inout.port == GUEST_NIO_PORT) { 749 error = vmexit_handle_notify(ctx, vme, pvcpu, inout.eax); 750 vmentry_inout_write(vcpu, inout.port, bytes); 751 return (error); 752 } 753 754 error = emulate_inout(ctx, vcpu, &inout); 755 if (error) { 756 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n", 757 in ? "in" : "out", 758 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'), 759 inout.port, vmexit->rip); 760 return (VMEXIT_ABORT); 761 } else { 762 /* 763 * Communicate the status of the inout operation back to the 764 * in-kernel instruction emulation. 765 */ 766 if (in) { 767 vmentry_inout_read(vcpu, inout.port, bytes, inout.eax); 768 } else { 769 vmentry_inout_write(vcpu, inout.port, bytes); 770 } 771 return (VMEXIT_CONTINUE); 772 } 773 } 774 775 static int 776 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 777 { 778 uint64_t val; 779 uint32_t eax, edx; 780 int error; 781 782 val = 0; 783 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val); 784 if (error != 0) { 785 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n", 786 vme->u.msr.code, *pvcpu); 787 if (get_config_bool("x86.strictmsr")) { 788 vm_inject_gp(ctx, *pvcpu); 789 return (VMEXIT_CONTINUE); 790 } 791 } 792 793 eax = val; 794 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax); 795 assert(error == 0); 796 797 edx = val >> 32; 798 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx); 799 assert(error == 0); 800 801 return (VMEXIT_CONTINUE); 802 } 803 804 static int 805 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 806 { 807 int error; 808 809 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval); 810 if (error != 0) { 811 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n", 812 vme->u.msr.code, vme->u.msr.wval, *pvcpu); 813 if (get_config_bool("x86.strictmsr")) { 814 vm_inject_gp(ctx, *pvcpu); 815 return (VMEXIT_CONTINUE); 816 } 817 } 818 return (VMEXIT_CONTINUE); 819 } 820 821 #ifdef __FreeBSD__ 822 static int 823 vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 824 { 825 826 (void)spinup_ap(ctx, *pvcpu, 827 vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip); 828 829 return (VMEXIT_CONTINUE); 830 } 831 #else 832 static int 833 vmexit_run_state(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 834 { 835 /* 836 * Run-state transitions (INIT, SIPI, etc) are handled in-kernel, so an 837 * exit to userspace with that code is not expected. 838 */ 839 fprintf(stderr, "unexpected run-state VM exit"); 840 return (VMEXIT_ABORT); 841 } 842 843 static int 844 vmexit_paging(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 845 { 846 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 847 fprintf(stderr, "\treason\t\tPAGING\n"); 848 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 849 fprintf(stderr, "\tgpa\t\t0x%016lx\n", vmexit->u.paging.gpa); 850 fprintf(stderr, "\tfault_type\t\t%d\n", vmexit->u.paging.fault_type); 851 852 return (VMEXIT_ABORT); 853 } 854 #endif /* __FreeBSD__ */ 855 856 #ifdef __FreeBSD__ 857 #define DEBUG_EPT_MISCONFIG 858 #else 859 /* EPT misconfig debugging not possible now that raw VMCS access is gone */ 860 #endif 861 862 #ifdef DEBUG_EPT_MISCONFIG 863 #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400 864 865 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4]; 866 static int ept_misconfig_ptenum; 867 #endif 868 869 static const char * 870 vmexit_vmx_desc(uint32_t exit_reason) 871 { 872 873 if (exit_reason >= nitems(vmx_exit_reason_desc) || 874 vmx_exit_reason_desc[exit_reason] == NULL) 875 return ("Unknown"); 876 return (vmx_exit_reason_desc[exit_reason]); 877 } 878 879 static int 880 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 881 { 882 883 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 884 fprintf(stderr, "\treason\t\tVMX\n"); 885 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 886 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 887 fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status); 888 fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason, 889 vmexit_vmx_desc(vmexit->u.vmx.exit_reason)); 890 fprintf(stderr, "\tqualification\t0x%016lx\n", 891 vmexit->u.vmx.exit_qualification); 892 fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type); 893 fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error); 894 #ifdef DEBUG_EPT_MISCONFIG 895 if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) { 896 vm_get_register(ctx, *pvcpu, 897 VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS), 898 &ept_misconfig_gpa); 899 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte, 900 &ept_misconfig_ptenum); 901 fprintf(stderr, "\tEPT misconfiguration:\n"); 902 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa); 903 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n", 904 ept_misconfig_ptenum, ept_misconfig_pte[0], 905 ept_misconfig_pte[1], ept_misconfig_pte[2], 906 ept_misconfig_pte[3]); 907 } 908 #endif /* DEBUG_EPT_MISCONFIG */ 909 return (VMEXIT_ABORT); 910 } 911 912 static int 913 vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 914 { 915 916 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 917 fprintf(stderr, "\treason\t\tSVM\n"); 918 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 919 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 920 fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode); 921 fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1); 922 fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2); 923 return (VMEXIT_ABORT); 924 } 925 926 static int 927 vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 928 { 929 930 assert(vmexit->inst_length == 0); 931 932 stats.vmexit_bogus++; 933 934 return (VMEXIT_CONTINUE); 935 } 936 937 static int 938 vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 939 { 940 941 assert(vmexit->inst_length == 0); 942 943 stats.vmexit_reqidle++; 944 945 return (VMEXIT_CONTINUE); 946 } 947 948 static int 949 vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 950 { 951 952 stats.vmexit_hlt++; 953 954 /* 955 * Just continue execution with the next instruction. We use 956 * the HLT VM exit as a way to be friendly with the host 957 * scheduler. 958 */ 959 return (VMEXIT_CONTINUE); 960 } 961 962 static int 963 vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 964 { 965 966 stats.vmexit_pause++; 967 968 return (VMEXIT_CONTINUE); 969 } 970 971 static int 972 vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 973 { 974 975 assert(vmexit->inst_length == 0); 976 977 stats.vmexit_mtrap++; 978 979 gdb_cpu_mtrap(*pvcpu); 980 981 return (VMEXIT_CONTINUE); 982 } 983 984 static int 985 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 986 { 987 uint8_t i, valid; 988 989 fprintf(stderr, "Failed to emulate instruction sequence "); 990 991 valid = vmexit->u.inst_emul.num_valid; 992 if (valid != 0) { 993 assert(valid <= sizeof (vmexit->u.inst_emul.inst)); 994 fprintf(stderr, "["); 995 for (i = 0; i < valid; i++) { 996 if (i == 0) { 997 fprintf(stderr, "%02x", 998 vmexit->u.inst_emul.inst[i]); 999 } else { 1000 fprintf(stderr, ", %02x", 1001 vmexit->u.inst_emul.inst[i]); 1002 } 1003 } 1004 fprintf(stderr, "] "); 1005 } 1006 fprintf(stderr, "@ %rip = %x\n", vmexit->rip); 1007 1008 return (VMEXIT_ABORT); 1009 } 1010 1011 static int 1012 vmexit_mmio(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1013 { 1014 int vcpu, err; 1015 struct vm_mmio mmio; 1016 bool is_read; 1017 1018 stats.vmexit_mmio++; 1019 1020 vcpu = *pvcpu; 1021 mmio = vmexit->u.mmio; 1022 is_read = (mmio.read != 0); 1023 1024 err = emulate_mem(ctx, vcpu, &mmio); 1025 1026 if (err == ESRCH) { 1027 fprintf(stderr, "Unhandled memory access to 0x%lx\n", mmio.gpa); 1028 stats.mmio_unhandled++; 1029 1030 /* 1031 * Access to non-existent physical addresses is not likely to 1032 * result in fatal errors on hardware machines, but rather reads 1033 * of all-ones or discarded-but-acknowledged writes. 1034 */ 1035 mmio.data = ~0UL; 1036 err = 0; 1037 } 1038 1039 if (err == 0) { 1040 if (is_read) { 1041 vmentry_mmio_read(vcpu, mmio.gpa, mmio.bytes, 1042 mmio.data); 1043 } else { 1044 vmentry_mmio_write(vcpu, mmio.gpa, mmio.bytes); 1045 } 1046 return (VMEXIT_CONTINUE); 1047 } 1048 1049 fprintf(stderr, "Unhandled mmio error to 0x%lx: %d\n", mmio.gpa, err); 1050 return (VMEXIT_ABORT); 1051 } 1052 1053 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER; 1054 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER; 1055 1056 static int 1057 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1058 { 1059 enum vm_suspend_how how; 1060 1061 how = vmexit->u.suspended.how; 1062 1063 fbsdrun_deletecpu(ctx, *pvcpu); 1064 1065 if (*pvcpu != BSP) { 1066 pthread_mutex_lock(&resetcpu_mtx); 1067 pthread_cond_signal(&resetcpu_cond); 1068 pthread_mutex_unlock(&resetcpu_mtx); 1069 pthread_exit(NULL); 1070 } 1071 1072 pthread_mutex_lock(&resetcpu_mtx); 1073 while (!CPU_EMPTY(&cpumask)) { 1074 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx); 1075 } 1076 pthread_mutex_unlock(&resetcpu_mtx); 1077 1078 switch (how) { 1079 case VM_SUSPEND_RESET: 1080 exit(0); 1081 case VM_SUSPEND_POWEROFF: 1082 if (get_config_bool_default("destroy_on_poweroff", false)) 1083 vm_destroy(ctx); 1084 exit(1); 1085 case VM_SUSPEND_HALT: 1086 exit(2); 1087 case VM_SUSPEND_TRIPLEFAULT: 1088 exit(3); 1089 default: 1090 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how); 1091 exit(100); 1092 } 1093 return (0); /* NOTREACHED */ 1094 } 1095 1096 static int 1097 vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1098 { 1099 1100 gdb_cpu_suspend(*pvcpu); 1101 return (VMEXIT_CONTINUE); 1102 } 1103 1104 static int 1105 vmexit_breakpoint(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 1106 { 1107 1108 gdb_cpu_breakpoint(*pvcpu, vmexit); 1109 return (VMEXIT_CONTINUE); 1110 } 1111 1112 static vmexit_handler_t handler[VM_EXITCODE_MAX] = { 1113 [VM_EXITCODE_INOUT] = vmexit_inout, 1114 [VM_EXITCODE_MMIO] = vmexit_mmio, 1115 [VM_EXITCODE_VMX] = vmexit_vmx, 1116 [VM_EXITCODE_SVM] = vmexit_svm, 1117 [VM_EXITCODE_BOGUS] = vmexit_bogus, 1118 [VM_EXITCODE_REQIDLE] = vmexit_reqidle, 1119 [VM_EXITCODE_RDMSR] = vmexit_rdmsr, 1120 [VM_EXITCODE_WRMSR] = vmexit_wrmsr, 1121 [VM_EXITCODE_MTRAP] = vmexit_mtrap, 1122 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul, 1123 #ifdef __FreeBSD__ 1124 [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap, 1125 #else 1126 [VM_EXITCODE_RUN_STATE] = vmexit_run_state, 1127 [VM_EXITCODE_PAGING] = vmexit_paging, 1128 [VM_EXITCODE_HLT] = vmexit_hlt, 1129 #endif 1130 [VM_EXITCODE_SUSPENDED] = vmexit_suspend, 1131 [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch, 1132 [VM_EXITCODE_DEBUG] = vmexit_debug, 1133 [VM_EXITCODE_BPT] = vmexit_breakpoint, 1134 }; 1135 1136 static void 1137 vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip) 1138 { 1139 int error, rc; 1140 enum vm_exitcode exitcode; 1141 cpuset_t active_cpus; 1142 struct vm_exit *vexit; 1143 struct vm_entry *ventry; 1144 1145 #ifdef __FreeBSD__ 1146 if (vcpumap[vcpu] != NULL) { 1147 error = pthread_setaffinity_np(pthread_self(), 1148 sizeof(cpuset_t), vcpumap[vcpu]); 1149 assert(error == 0); 1150 } 1151 #endif 1152 error = vm_active_cpus(ctx, &active_cpus); 1153 assert(CPU_ISSET(vcpu, &active_cpus)); 1154 1155 error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip); 1156 assert(error == 0); 1157 1158 ventry = &vmentry[vcpu]; 1159 vexit = &vmexit[vcpu]; 1160 1161 while (1) { 1162 error = vm_run(ctx, vcpu, ventry, vexit); 1163 if (error != 0) 1164 break; 1165 1166 if (ventry->cmd != VEC_DEFAULT) { 1167 /* 1168 * Discard any lingering entry state after it has been 1169 * submitted via vm_run(). 1170 */ 1171 bzero(ventry, sizeof (*ventry)); 1172 } 1173 1174 exitcode = vexit->exitcode; 1175 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) { 1176 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n", 1177 exitcode); 1178 exit(4); 1179 } 1180 1181 rc = (*handler[exitcode])(ctx, vexit, &vcpu); 1182 1183 switch (rc) { 1184 case VMEXIT_CONTINUE: 1185 break; 1186 case VMEXIT_ABORT: 1187 abort(); 1188 default: 1189 exit(4); 1190 } 1191 } 1192 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno); 1193 } 1194 1195 static int 1196 num_vcpus_allowed(struct vmctx *ctx) 1197 { 1198 uint16_t sockets, cores, threads, maxcpus; 1199 #ifdef __FreeBSD__ 1200 int tmp, error; 1201 1202 /* 1203 * The guest is allowed to spinup more than one processor only if the 1204 * UNRESTRICTED_GUEST capability is available. 1205 */ 1206 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp); 1207 if (error != 0) 1208 return (1); 1209 #else 1210 int error; 1211 /* Unrestricted Guest is always enabled on illumos */ 1212 1213 #endif /* __FreeBSD__ */ 1214 1215 error = vm_get_topology(ctx, &sockets, &cores, &threads, &maxcpus); 1216 if (error == 0) 1217 return (maxcpus); 1218 else 1219 return (1); 1220 } 1221 1222 void 1223 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu) 1224 { 1225 int err, tmp; 1226 1227 #ifdef __FreeBSD__ 1228 if (get_config_bool_default("x86.vmexit_on_hlt", false)) { 1229 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp); 1230 if (err < 0) { 1231 fprintf(stderr, "VM exit on HLT not supported\n"); 1232 exit(4); 1233 } 1234 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1); 1235 if (cpu == BSP) 1236 handler[VM_EXITCODE_HLT] = vmexit_hlt; 1237 } 1238 #else 1239 /* 1240 * We insist that vmexit-on-hlt is available on the host CPU, and enable 1241 * it by default. Configuration of that feature is done with both of 1242 * those facts in mind. 1243 */ 1244 tmp = (int)get_config_bool_default("x86.vmexit_on_hlt", true); 1245 err = vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, tmp); 1246 if (err < 0) { 1247 fprintf(stderr, "VM exit on HLT not supported\n"); 1248 exit(4); 1249 } 1250 #endif /* __FreeBSD__ */ 1251 1252 if (get_config_bool_default("x86.vmexit_on_pause", false)) { 1253 /* 1254 * pause exit support required for this mode 1255 */ 1256 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp); 1257 if (err < 0) { 1258 fprintf(stderr, 1259 "SMP mux requested, no pause support\n"); 1260 exit(4); 1261 } 1262 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1); 1263 if (cpu == BSP) 1264 handler[VM_EXITCODE_PAUSE] = vmexit_pause; 1265 } 1266 1267 if (get_config_bool_default("x86.x2apic", false)) 1268 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED); 1269 else 1270 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED); 1271 1272 if (err) { 1273 fprintf(stderr, "Unable to set x2apic state (%d)\n", err); 1274 exit(4); 1275 } 1276 1277 #ifdef __FreeBSD__ 1278 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1); 1279 #endif 1280 } 1281 1282 static struct vmctx * 1283 do_open(const char *vmname) 1284 { 1285 struct vmctx *ctx; 1286 int error; 1287 bool reinit, romboot; 1288 #ifndef WITHOUT_CAPSICUM 1289 cap_rights_t rights; 1290 const cap_ioctl_t *cmds; 1291 size_t ncmds; 1292 #endif 1293 1294 reinit = romboot = false; 1295 1296 if (lpc_bootrom()) 1297 romboot = true; 1298 #ifndef __FreeBSD__ 1299 uint64_t create_flags = 0; 1300 if (get_config_bool_default("memory.use_reservoir", false)) { 1301 create_flags |= VCF_RESERVOIR_MEM; 1302 } 1303 error = vm_create(vmname, create_flags); 1304 #else 1305 error = vm_create(vmname); 1306 #endif /* __FreeBSD__ */ 1307 if (error) { 1308 if (errno == EEXIST) { 1309 if (romboot) { 1310 reinit = true; 1311 } else { 1312 /* 1313 * The virtual machine has been setup by the 1314 * userspace bootloader. 1315 */ 1316 } 1317 } else { 1318 perror("vm_create"); 1319 exit(4); 1320 } 1321 } else { 1322 if (!romboot) { 1323 /* 1324 * If the virtual machine was just created then a 1325 * bootrom must be configured to boot it. 1326 */ 1327 fprintf(stderr, "virtual machine cannot be booted\n"); 1328 exit(4); 1329 } 1330 } 1331 1332 ctx = vm_open(vmname); 1333 if (ctx == NULL) { 1334 perror("vm_open"); 1335 exit(4); 1336 } 1337 1338 #ifndef WITHOUT_CAPSICUM 1339 cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW); 1340 if (caph_rights_limit(vm_get_device_fd(ctx), &rights) == -1) 1341 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1342 vm_get_ioctls(&ncmds); 1343 cmds = vm_get_ioctls(NULL); 1344 if (cmds == NULL) 1345 errx(EX_OSERR, "out of memory"); 1346 if (caph_ioctls_limit(vm_get_device_fd(ctx), cmds, ncmds) == -1) 1347 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1348 free((cap_ioctl_t *)cmds); 1349 #endif 1350 1351 if (reinit) { 1352 #ifndef __FreeBSD__ 1353 error = vm_reinit(ctx, 0); 1354 #else 1355 error = vm_reinit(ctx); 1356 #endif 1357 if (error) { 1358 perror("vm_reinit"); 1359 exit(4); 1360 } 1361 } 1362 error = vm_set_topology(ctx, sockets, cores, threads, maxcpus); 1363 if (error) 1364 errx(EX_OSERR, "vm_set_topology"); 1365 return (ctx); 1366 } 1367 1368 static bool 1369 parse_config_option(const char *option) 1370 { 1371 const char *value; 1372 char *path; 1373 1374 value = strchr(option, '='); 1375 if (value == NULL || value[1] == '\0') 1376 return (false); 1377 path = strndup(option, value - option); 1378 if (path == NULL) 1379 err(4, "Failed to allocate memory"); 1380 set_config_value(path, value + 1); 1381 return (true); 1382 } 1383 1384 static void 1385 parse_simple_config_file(const char *path) 1386 { 1387 FILE *fp; 1388 char *line, *cp; 1389 size_t linecap; 1390 unsigned int lineno; 1391 1392 fp = fopen(path, "r"); 1393 if (fp == NULL) 1394 err(4, "Failed to open configuration file %s", path); 1395 line = NULL; 1396 linecap = 0; 1397 lineno = 1; 1398 for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) { 1399 if (*line == '#' || *line == '\n') 1400 continue; 1401 cp = strchr(line, '\n'); 1402 if (cp != NULL) 1403 *cp = '\0'; 1404 if (!parse_config_option(line)) 1405 errx(4, "%s line %u: invalid config option '%s'", path, 1406 lineno, line); 1407 } 1408 free(line); 1409 fclose(fp); 1410 } 1411 1412 static void 1413 parse_gdb_options(char *optarg) 1414 { 1415 const char *sport; 1416 char *colon; 1417 1418 if (optarg[0] == 'w') { 1419 set_config_bool("gdb.wait", true); 1420 optarg++; 1421 } 1422 1423 colon = strrchr(optarg, ':'); 1424 if (colon == NULL) { 1425 sport = optarg; 1426 } else { 1427 *colon = '\0'; 1428 colon++; 1429 sport = colon; 1430 set_config_value("gdb.address", optarg); 1431 } 1432 1433 set_config_value("gdb.port", sport); 1434 } 1435 1436 static void 1437 set_defaults(void) 1438 { 1439 1440 set_config_bool("acpi_tables", false); 1441 set_config_value("memory.size", "256M"); 1442 set_config_bool("x86.strictmsr", true); 1443 } 1444 1445 int 1446 main(int argc, char *argv[]) 1447 { 1448 int c, error, err; 1449 int max_vcpus, memflags; 1450 struct vmctx *ctx; 1451 uint64_t rip; 1452 size_t memsize; 1453 const char *value, *vmname; 1454 char *optstr; 1455 1456 init_config(); 1457 set_defaults(); 1458 progname = basename(argv[0]); 1459 1460 #ifdef __FreeBSD__ 1461 optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:K:U:"; 1462 #else 1463 /* +d, +B, -p */ 1464 optstr = "adehuwxACDHIPSWYk:o:G:c:s:m:l:B:K:U:"; 1465 #endif 1466 while ((c = getopt(argc, argv, optstr)) != -1) { 1467 switch (c) { 1468 case 'a': 1469 set_config_bool("x86.x2apic", false); 1470 break; 1471 case 'A': 1472 #ifdef __FreeBSD__ 1473 /* 1474 * This option is ignored on illumos since the 1475 * generated ACPI tables are not used; the bootroms 1476 * have their own. The option is retained for backwards 1477 * compatibility but does nothing. Note that the 1478 * acpi_tables configuration is still accepted via 1479 * -o if somebody really wants to generate these tables. 1480 */ 1481 set_config_bool("acpi_tables", true); 1482 #endif 1483 break; 1484 case 'D': 1485 set_config_bool("destroy_on_poweroff", true); 1486 break; 1487 #ifndef __FreeBSD__ 1488 case 'B': 1489 if (smbios_parse(optarg) != 0) { 1490 errx(EX_USAGE, "invalid SMBIOS " 1491 "configuration '%s'", optarg); 1492 } 1493 break; 1494 case 'd': 1495 set_config_bool("suspend_at_boot", true); 1496 break; 1497 #endif 1498 #ifdef __FreeBSD__ 1499 case 'p': 1500 if (pincpu_parse(optarg) != 0) { 1501 errx(EX_USAGE, "invalid vcpu pinning " 1502 "configuration '%s'", optarg); 1503 } 1504 break; 1505 #endif 1506 case 'c': 1507 if (topology_parse(optarg) != 0) { 1508 errx(EX_USAGE, "invalid cpu topology " 1509 "'%s'", optarg); 1510 } 1511 break; 1512 case 'C': 1513 set_config_bool("memory.guest_in_core", true); 1514 break; 1515 case 'G': 1516 parse_gdb_options(optarg); 1517 break; 1518 case 'k': 1519 parse_simple_config_file(optarg); 1520 break; 1521 case 'K': 1522 set_config_value("keyboard.layout", optarg); 1523 break; 1524 case 'l': 1525 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1526 lpc_print_supported_devices(); 1527 exit(0); 1528 } else if (lpc_device_parse(optarg) != 0) { 1529 errx(EX_USAGE, "invalid lpc device " 1530 "configuration '%s'", optarg); 1531 } 1532 break; 1533 case 's': 1534 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1535 pci_print_supported_devices(); 1536 exit(0); 1537 } else if (pci_parse_slot(optarg) != 0) 1538 exit(4); 1539 else 1540 break; 1541 case 'S': 1542 set_config_bool("memory.wired", true); 1543 break; 1544 case 'm': 1545 set_config_value("memory.size", optarg); 1546 break; 1547 case 'o': 1548 if (!parse_config_option(optarg)) 1549 errx(EX_USAGE, "invalid configuration option '%s'", optarg); 1550 break; 1551 case 'H': 1552 set_config_bool("x86.vmexit_on_hlt", true); 1553 break; 1554 case 'I': 1555 /* 1556 * The "-I" option was used to add an ioapic to the 1557 * virtual machine. 1558 * 1559 * An ioapic is now provided unconditionally for each 1560 * virtual machine and this option is now deprecated. 1561 */ 1562 break; 1563 case 'P': 1564 set_config_bool("x86.vmexit_on_pause", true); 1565 break; 1566 case 'e': 1567 set_config_bool("x86.strictio", true); 1568 break; 1569 case 'u': 1570 set_config_bool("rtc.use_localtime", false); 1571 break; 1572 case 'U': 1573 set_config_value("uuid", optarg); 1574 break; 1575 case 'w': 1576 set_config_bool("x86.strictmsr", false); 1577 break; 1578 case 'W': 1579 set_config_bool("virtio_msix", false); 1580 break; 1581 case 'x': 1582 set_config_bool("x86.x2apic", true); 1583 break; 1584 case 'Y': 1585 set_config_bool("x86.mptable", false); 1586 break; 1587 case 'h': 1588 usage(0); 1589 default: 1590 usage(1); 1591 } 1592 } 1593 argc -= optind; 1594 argv += optind; 1595 1596 if (argc > 1) 1597 usage(1); 1598 1599 if (argc == 1) 1600 set_config_value("name", argv[0]); 1601 1602 vmname = get_config_value("name"); 1603 if (vmname == NULL) 1604 usage(1); 1605 1606 if (get_config_bool_default("config.dump", false)) { 1607 dump_config(); 1608 exit(1); 1609 } 1610 1611 #ifndef __FreeBSD__ 1612 illumos_priv_init(); 1613 #endif 1614 1615 calc_topolopgy(); 1616 #ifdef __FreeBSD__ 1617 build_vcpumaps(); 1618 #endif 1619 1620 value = get_config_value("memory.size"); 1621 error = vm_parse_memsize(value, &memsize); 1622 if (error) 1623 errx(EX_USAGE, "invalid memsize '%s'", value); 1624 1625 ctx = do_open(vmname); 1626 1627 max_vcpus = num_vcpus_allowed(ctx); 1628 if (guest_ncpus > max_vcpus) { 1629 fprintf(stderr, "%d vCPUs requested but only %d available\n", 1630 guest_ncpus, max_vcpus); 1631 exit(4); 1632 } 1633 1634 fbsdrun_set_capabilities(ctx, BSP); 1635 1636 memflags = 0; 1637 if (get_config_bool_default("memory.wired", false)) 1638 memflags |= VM_MEM_F_WIRED; 1639 if (get_config_bool_default("memory.guest_in_core", false)) 1640 memflags |= VM_MEM_F_INCORE; 1641 vm_set_memflags(ctx, memflags); 1642 #ifdef __FreeBSD__ 1643 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1644 #else 1645 do { 1646 errno = 0; 1647 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1648 error = errno; 1649 if (err != 0 && error == ENOMEM) { 1650 (void) fprintf(stderr, "Unable to allocate memory " 1651 "(%llu), retrying in 1 second\n", memsize); 1652 sleep(1); 1653 } 1654 } while (error == ENOMEM); 1655 #endif 1656 if (err) { 1657 fprintf(stderr, "Unable to set up memory (%d)\n", errno); 1658 exit(4); 1659 } 1660 1661 error = init_msr(); 1662 if (error) { 1663 fprintf(stderr, "init_msr error %d", error); 1664 exit(4); 1665 } 1666 1667 init_mem(guest_ncpus); 1668 init_inout(); 1669 #ifdef __FreeBSD__ 1670 kernemu_dev_init(); 1671 #endif 1672 init_bootrom(ctx); 1673 atkbdc_init(ctx); 1674 pci_irq_init(ctx); 1675 ioapic_init(ctx); 1676 1677 rtc_init(ctx); 1678 sci_init(ctx); 1679 #ifndef __FreeBSD__ 1680 pmtmr_init(ctx); 1681 #endif 1682 1683 /* Allocate per-VCPU resources. */ 1684 vmexit = calloc(guest_ncpus, sizeof(*vmexit)); 1685 mt_vmm_info = calloc(guest_ncpus, sizeof(*mt_vmm_info)); 1686 #ifndef __FreeBSD__ 1687 vmentry = calloc(guest_ncpus, sizeof(*vmentry)); 1688 #endif 1689 1690 /* 1691 * Exit if a device emulation finds an error in its initilization 1692 */ 1693 if (init_pci(ctx) != 0) { 1694 perror("device emulation initialization error"); 1695 exit(4); 1696 } 1697 1698 /* 1699 * Initialize after PCI, to allow a bootrom file to reserve the high 1700 * region. 1701 */ 1702 if (get_config_bool("acpi_tables")) 1703 vmgenc_init(ctx); 1704 1705 #ifdef __FreeBSD__ 1706 init_gdb(ctx); 1707 #else 1708 if (value != NULL) { 1709 int port = atoi(value); 1710 1711 if (port < 0) 1712 init_mdb(ctx); 1713 else 1714 init_gdb(ctx); 1715 } 1716 #endif 1717 1718 if (lpc_bootrom()) { 1719 #ifdef __FreeBSD__ 1720 if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) { 1721 fprintf(stderr, "ROM boot failed: unrestricted guest " 1722 "capability not available\n"); 1723 exit(4); 1724 } 1725 #else 1726 /* Unrestricted Guest is always enabled on illumos */ 1727 #endif 1728 error = vcpu_reset(ctx, BSP); 1729 assert(error == 0); 1730 } 1731 1732 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip); 1733 assert(error == 0); 1734 1735 /* 1736 * build the guest tables, MP etc. 1737 */ 1738 if (get_config_bool_default("x86.mptable", true)) { 1739 error = mptable_build(ctx, guest_ncpus); 1740 if (error) { 1741 perror("error to build the guest tables"); 1742 exit(4); 1743 } 1744 } 1745 1746 error = smbios_build(ctx); 1747 if (error != 0) 1748 exit(4); 1749 1750 if (get_config_bool("acpi_tables")) { 1751 error = acpi_build(ctx, guest_ncpus); 1752 assert(error == 0); 1753 } 1754 1755 if (lpc_bootrom()) 1756 fwctl_init(); 1757 1758 /* 1759 * Change the proc title to include the VM name. 1760 */ 1761 setproctitle("%s", vmname); 1762 1763 #ifndef WITHOUT_CAPSICUM 1764 caph_cache_catpages(); 1765 1766 if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1) 1767 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1768 1769 if (caph_enter() == -1) 1770 errx(EX_OSERR, "cap_enter() failed"); 1771 #endif 1772 1773 #ifndef __FreeBSD__ 1774 illumos_priv_lock(); 1775 #endif 1776 1777 #ifdef __FreeBSD__ 1778 /* 1779 * Add CPU 0 1780 */ 1781 fbsdrun_addcpu(ctx, BSP, BSP, rip); 1782 #else 1783 /* Set BSP to run (unlike the APs which wait for INIT) */ 1784 error = vm_set_run_state(ctx, BSP, VRS_RUN, 0); 1785 assert(error == 0); 1786 fbsdrun_addcpu(ctx, BSP, rip, 1787 get_config_bool_default("suspend_at_boot", false)); 1788 1789 /* Add subsequent CPUs, which will wait until INIT/SIPI-ed */ 1790 for (uint_t i = 1; i < guest_ncpus; i++) { 1791 spinup_halted_ap(ctx, i); 1792 } 1793 #endif 1794 /* 1795 * Head off to the main event dispatch loop 1796 */ 1797 mevent_dispatch(); 1798 1799 exit(4); 1800 } 1801