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 56 #ifdef __FreeBSD__ 57 #include <amd64/vmm/intel/vmcs.h> 58 #else 59 #include <sys/cpuset.h> 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 #ifdef __FreeBSD__ 88 #include <machine/vmm_instruction_emul.h> 89 #endif 90 #include <vmmapi.h> 91 92 #include "bhyverun.h" 93 #include "acpi.h" 94 #include "atkbdc.h" 95 #include "bootrom.h" 96 #include "config.h" 97 #include "inout.h" 98 #include "debug.h" 99 #include "fwctl.h" 100 #include "gdb.h" 101 #include "ioapic.h" 102 #include "kernemu_dev.h" 103 #include "mem.h" 104 #include "mevent.h" 105 #include "mptbl.h" 106 #include "pci_emul.h" 107 #include "pci_irq.h" 108 #include "pci_lpc.h" 109 #include "smbiostbl.h" 110 #include "xmsr.h" 111 #include "spinup_ap.h" 112 #include "rtc.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); 191 192 int guest_ncpus; 193 uint16_t cpu_cores, cpu_sockets, cpu_threads; 194 195 int raw_stdio = 0; 196 197 static char *progname; 198 static const int BSP = 0; 199 200 static cpuset_t cpumask; 201 202 static void vm_loop(struct vmctx *ctx, int vcpu); 203 204 #ifndef __FreeBSD__ 205 static struct vm_entry *vmentry; 206 #endif 207 208 static 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 #ifdef __FreeBSD__ 215 uint64_t vmexit_inst_emul; 216 #else 217 uint64_t vmexit_mmio; 218 uint64_t vmexit_inout; 219 uint64_t mmio_unhandled; 220 #endif 221 uint64_t cpu_switch_rotate; 222 uint64_t cpu_switch_direct; 223 } stats; 224 225 static struct mt_vmm_info { 226 pthread_t mt_thr; 227 struct vmctx *mt_ctx; 228 int mt_vcpu; 229 } *mt_vmm_info; 230 231 #ifdef __FreeBSD__ 232 static cpuset_t **vcpumap; 233 #endif 234 235 static void 236 usage(int code) 237 { 238 239 fprintf(stderr, 240 #ifdef __FreeBSD__ 241 "Usage: %s [-AaCDeHhPSuWwxY]\n" 242 #else 243 "Usage: %s [-aCDdeHhPSuWwxY]\n" 244 #endif 245 " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n" 246 #ifdef __FreeBSD__ 247 " %*s [-G port] [-k config_file] [-l lpc] [-m mem] [-o var=value]\n" 248 " %*s [-p vcpu:hostcpu] [-r file] [-s pci] [-U uuid] vmname\n" 249 250 " -A: create ACPI tables\n" 251 #else 252 " %*s [-k <config_file>] [-l <lpc>] [-m mem] [-o <var>=<value>]\n" 253 " %*s [-s <pci>] [-U uuid] vmname\n" 254 #endif 255 " -a: local apic is in xAPIC mode (deprecated)\n" 256 #ifndef __FreeBSD__ 257 " -B type,key=value,...: set SMBIOS information\n" 258 #endif 259 " -C: include guest memory in core file\n" 260 " -c: number of CPUs and/or topology specification\n" 261 " -D: destroy on power-off\n" 262 #ifndef __FreeBSD__ 263 " -d: suspend cpu at boot\n" 264 #endif 265 " -e: exit on unhandled I/O access\n" 266 #ifdef __FreeBSD__ 267 " -G: start a debug server\n" 268 #endif 269 " -H: vmexit from the guest on HLT\n" 270 " -h: help\n" 271 " -k: key=value flat config file\n" 272 " -K: PS2 keyboard layout\n" 273 " -l: LPC device configuration\n" 274 " -m: memory size\n" 275 " -o: set config 'var' to 'value'\n" 276 " -P: vmexit from the guest on pause\n" 277 #ifdef __FreeBSD__ 278 " -p: pin 'vcpu' to 'hostcpu'\n" 279 #endif 280 " -S: guest memory cannot be swapped\n" 281 " -s: <slot,driver,configinfo> PCI slot config\n" 282 " -U: UUID\n" 283 " -u: RTC keeps UTC time\n" 284 " -W: force virtio to use single-vector MSI\n" 285 " -w: ignore unimplemented MSRs\n" 286 " -x: local APIC is in x2APIC mode\n" 287 " -Y: disable MPtable generation\n", 288 progname, (int)strlen(progname), "", (int)strlen(progname), "", 289 (int)strlen(progname), ""); 290 291 exit(code); 292 } 293 294 /* 295 * XXX This parser is known to have the following issues: 296 * 1. It accepts null key=value tokens ",," as setting "cpus" to an 297 * empty string. 298 * 299 * The acceptance of a null specification ('-c ""') is by design to match the 300 * manual page syntax specification, this results in a topology of 1 vCPU. 301 */ 302 static int 303 topology_parse(const char *opt) 304 { 305 char *cp, *str, *tofree; 306 307 if (*opt == '\0') { 308 set_config_value("sockets", "1"); 309 set_config_value("cores", "1"); 310 set_config_value("threads", "1"); 311 set_config_value("cpus", "1"); 312 return (0); 313 } 314 315 tofree = str = strdup(opt); 316 if (str == NULL) 317 errx(4, "Failed to allocate memory"); 318 319 while ((cp = strsep(&str, ",")) != NULL) { 320 if (strncmp(cp, "cpus=", strlen("cpus=")) == 0) 321 set_config_value("cpus", cp + strlen("cpus=")); 322 else if (strncmp(cp, "sockets=", strlen("sockets=")) == 0) 323 set_config_value("sockets", cp + strlen("sockets=")); 324 else if (strncmp(cp, "cores=", strlen("cores=")) == 0) 325 set_config_value("cores", cp + strlen("cores=")); 326 else if (strncmp(cp, "threads=", strlen("threads=")) == 0) 327 set_config_value("threads", cp + strlen("threads=")); 328 #ifdef notyet /* Do not expose this until vmm.ko implements it */ 329 else if (strncmp(cp, "maxcpus=", strlen("maxcpus=")) == 0) 330 set_config_value("maxcpus", cp + strlen("maxcpus=")); 331 #endif 332 else if (strchr(cp, '=') != NULL) 333 goto out; 334 else 335 set_config_value("cpus", cp); 336 } 337 free(tofree); 338 return (0); 339 340 out: 341 free(tofree); 342 return (-1); 343 } 344 345 static int 346 parse_int_value(const char *key, const char *value, int minval, int maxval) 347 { 348 char *cp; 349 long lval; 350 351 errno = 0; 352 lval = strtol(value, &cp, 0); 353 if (errno != 0 || *cp != '\0' || cp == value || lval < minval || 354 lval > maxval) 355 errx(4, "Invalid value for %s: '%s'", key, value); 356 return (lval); 357 } 358 359 /* 360 * Set the sockets, cores, threads, and guest_cpus variables based on 361 * the configured topology. 362 * 363 * The limits of UINT16_MAX are due to the types passed to 364 * vm_set_topology(). vmm.ko may enforce tighter limits. 365 */ 366 static void 367 calc_topology(void) 368 { 369 const char *value; 370 bool explicit_cpus; 371 uint64_t ncpus; 372 373 value = get_config_value("cpus"); 374 if (value != NULL) { 375 guest_ncpus = parse_int_value("cpus", value, 1, UINT16_MAX); 376 explicit_cpus = true; 377 } else { 378 guest_ncpus = 1; 379 explicit_cpus = false; 380 } 381 value = get_config_value("cores"); 382 if (value != NULL) 383 cpu_cores = parse_int_value("cores", value, 1, UINT16_MAX); 384 else 385 cpu_cores = 1; 386 value = get_config_value("threads"); 387 if (value != NULL) 388 cpu_threads = parse_int_value("threads", value, 1, UINT16_MAX); 389 else 390 cpu_threads = 1; 391 value = get_config_value("sockets"); 392 if (value != NULL) 393 cpu_sockets = parse_int_value("sockets", value, 1, UINT16_MAX); 394 else 395 cpu_sockets = guest_ncpus; 396 397 /* 398 * Compute sockets * cores * threads avoiding overflow. The 399 * range check above insures these are 16 bit values. 400 */ 401 ncpus = (uint64_t)cpu_sockets * cpu_cores * cpu_threads; 402 if (ncpus > UINT16_MAX) 403 errx(4, "Computed number of vCPUs too high: %ju", 404 (uintmax_t)ncpus); 405 406 if (explicit_cpus) { 407 if (guest_ncpus != (int)ncpus) 408 errx(4, "Topology (%d sockets, %d cores, %d threads) " 409 "does not match %d vCPUs", 410 cpu_sockets, cpu_cores, cpu_threads, 411 guest_ncpus); 412 } else 413 guest_ncpus = ncpus; 414 } 415 416 #ifdef __FreeBSD__ 417 static int 418 pincpu_parse(const char *opt) 419 { 420 int vcpu, pcpu; 421 const char *value; 422 char *newval; 423 char key[16]; 424 425 if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) { 426 fprintf(stderr, "invalid format: %s\n", opt); 427 return (-1); 428 } 429 430 if (vcpu < 0) { 431 fprintf(stderr, "invalid vcpu '%d'\n", vcpu); 432 return (-1); 433 } 434 435 if (pcpu < 0 || pcpu >= CPU_SETSIZE) { 436 fprintf(stderr, "hostcpu '%d' outside valid range from " 437 "0 to %d\n", pcpu, CPU_SETSIZE - 1); 438 return (-1); 439 } 440 441 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 442 value = get_config_value(key); 443 444 if (asprintf(&newval, "%s%s%d", value != NULL ? value : "", 445 value != NULL ? "," : "", pcpu) == -1) { 446 perror("failed to build new cpuset string"); 447 return (-1); 448 } 449 450 set_config_value(key, newval); 451 free(newval); 452 return (0); 453 } 454 455 static void 456 parse_cpuset(int vcpu, const char *list, cpuset_t *set) 457 { 458 char *cp, *token; 459 int pcpu, start; 460 461 CPU_ZERO(set); 462 start = -1; 463 token = __DECONST(char *, list); 464 for (;;) { 465 pcpu = strtoul(token, &cp, 0); 466 if (cp == token) 467 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 468 if (pcpu < 0 || pcpu >= CPU_SETSIZE) 469 errx(4, "hostcpu '%d' outside valid range from 0 to %d", 470 pcpu, CPU_SETSIZE - 1); 471 switch (*cp) { 472 case ',': 473 case '\0': 474 if (start >= 0) { 475 if (start > pcpu) 476 errx(4, "Invalid hostcpu range %d-%d", 477 start, pcpu); 478 while (start < pcpu) { 479 CPU_SET(start, set); 480 start++; 481 } 482 start = -1; 483 } 484 CPU_SET(pcpu, set); 485 break; 486 case '-': 487 if (start >= 0) 488 errx(4, "invalid cpuset for vcpu %d: '%s'", 489 vcpu, list); 490 start = pcpu; 491 break; 492 default: 493 errx(4, "invalid cpuset for vcpu %d: '%s'", vcpu, list); 494 } 495 if (*cp == '\0') 496 break; 497 token = cp + 1; 498 } 499 } 500 501 static void 502 build_vcpumaps(void) 503 { 504 char key[16]; 505 const char *value; 506 int vcpu; 507 508 vcpumap = calloc(guest_ncpus, sizeof(*vcpumap)); 509 for (vcpu = 0; vcpu < guest_ncpus; vcpu++) { 510 snprintf(key, sizeof(key), "vcpu.%d.cpuset", vcpu); 511 value = get_config_value(key); 512 if (value == NULL) 513 continue; 514 vcpumap[vcpu] = malloc(sizeof(cpuset_t)); 515 if (vcpumap[vcpu] == NULL) 516 err(4, "Failed to allocate cpuset for vcpu %d", vcpu); 517 parse_cpuset(vcpu, value, vcpumap[vcpu]); 518 } 519 } 520 521 void 522 vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid, 523 int errcode) 524 { 525 struct vmctx *ctx; 526 int error, restart_instruction; 527 528 ctx = arg; 529 restart_instruction = 1; 530 531 error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode, 532 restart_instruction); 533 assert(error == 0); 534 } 535 #endif /* __FreeBSD__ */ 536 537 void * 538 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len) 539 { 540 541 return (vm_map_gpa(ctx, gaddr, len)); 542 } 543 544 int 545 fbsdrun_virtio_msix(void) 546 { 547 548 return (get_config_bool_default("virtio_msix", true)); 549 } 550 551 static void * 552 fbsdrun_start_thread(void *param) 553 { 554 char tname[MAXCOMLEN + 1]; 555 struct mt_vmm_info *mtp; 556 #ifdef __FreeBSD__ 557 int error, vcpu; 558 #else 559 int vcpu; 560 #endif 561 562 mtp = param; 563 vcpu = mtp->mt_vcpu; 564 565 snprintf(tname, sizeof(tname), "vcpu %d", vcpu); 566 pthread_set_name_np(mtp->mt_thr, tname); 567 568 #ifdef __FreeBSD__ 569 if (vcpumap[vcpu] != NULL) { 570 error = pthread_setaffinity_np(pthread_self(), 571 sizeof(cpuset_t), vcpumap[vcpu]); 572 assert(error == 0); 573 } 574 #endif 575 576 gdb_cpu_add(vcpu); 577 578 vm_loop(mtp->mt_ctx, vcpu); 579 580 /* not reached */ 581 exit(1); 582 return (NULL); 583 } 584 585 void 586 fbsdrun_addcpu(struct vmctx *ctx, int newcpu, bool suspend) 587 { 588 int error; 589 590 error = vm_activate_cpu(ctx, newcpu); 591 if (error != 0) 592 err(EX_OSERR, "could not activate CPU %d", newcpu); 593 594 CPU_SET_ATOMIC(newcpu, &cpumask); 595 596 if (suspend) 597 (void) vm_suspend_cpu(ctx, newcpu); 598 599 mt_vmm_info[newcpu].mt_ctx = ctx; 600 mt_vmm_info[newcpu].mt_vcpu = newcpu; 601 602 error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL, 603 fbsdrun_start_thread, &mt_vmm_info[newcpu]); 604 assert(error == 0); 605 } 606 607 static int 608 fbsdrun_deletecpu(int vcpu) 609 { 610 611 if (!CPU_ISSET(vcpu, &cpumask)) { 612 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu); 613 exit(4); 614 } 615 616 CPU_CLR_ATOMIC(vcpu, &cpumask); 617 return (CPU_EMPTY(&cpumask)); 618 } 619 620 #ifndef __FreeBSD__ 621 static void 622 vmentry_mmio_read(int vcpu, uint64_t gpa, uint8_t bytes, uint64_t data) 623 { 624 struct vm_entry *entry = &vmentry[vcpu]; 625 struct vm_mmio *mmio = &entry->u.mmio; 626 627 assert(entry->cmd == VEC_DEFAULT); 628 629 entry->cmd = VEC_FULFILL_MMIO; 630 mmio->bytes = bytes; 631 mmio->read = 1; 632 mmio->gpa = gpa; 633 mmio->data = data; 634 } 635 636 static void 637 vmentry_mmio_write(int vcpu, uint64_t gpa, uint8_t bytes) 638 { 639 struct vm_entry *entry = &vmentry[vcpu]; 640 struct vm_mmio *mmio = &entry->u.mmio; 641 642 assert(entry->cmd == VEC_DEFAULT); 643 644 entry->cmd = VEC_FULFILL_MMIO; 645 mmio->bytes = bytes; 646 mmio->read = 0; 647 mmio->gpa = gpa; 648 mmio->data = 0; 649 } 650 651 static void 652 vmentry_inout_read(int vcpu, uint16_t port, uint8_t bytes, uint32_t data) 653 { 654 struct vm_entry *entry = &vmentry[vcpu]; 655 struct vm_inout *inout = &entry->u.inout; 656 657 assert(entry->cmd == VEC_DEFAULT); 658 659 entry->cmd = VEC_FULFILL_INOUT; 660 inout->bytes = bytes; 661 inout->flags = INOUT_IN; 662 inout->port = port; 663 inout->eax = data; 664 } 665 666 static void 667 vmentry_inout_write(int vcpu, uint16_t port, uint8_t bytes) 668 { 669 struct vm_entry *entry = &vmentry[vcpu]; 670 struct vm_inout *inout = &entry->u.inout; 671 672 assert(entry->cmd == VEC_DEFAULT); 673 674 entry->cmd = VEC_FULFILL_INOUT; 675 inout->bytes = bytes; 676 inout->flags = 0; 677 inout->port = port; 678 inout->eax = 0; 679 } 680 #endif 681 682 static int 683 vmexit_handle_notify(struct vmctx *ctx __unused, struct vm_exit *vme __unused, 684 int *pvcpu __unused, uint32_t eax __unused) 685 { 686 #if BHYVE_DEBUG 687 /* 688 * put guest-driven debug here 689 */ 690 #endif 691 return (VMEXIT_CONTINUE); 692 } 693 694 static int 695 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 696 { 697 int error; 698 int vcpu; 699 struct vm_inout inout; 700 bool in; 701 uint8_t bytes; 702 703 stats.vmexit_inout++; 704 705 vcpu = *pvcpu; 706 inout = vme->u.inout; 707 in = (inout.flags & INOUT_IN) != 0; 708 bytes = inout.bytes; 709 710 /* Extra-special case of host notifications */ 711 if (!in && inout.port == GUEST_NIO_PORT) { 712 error = vmexit_handle_notify(ctx, vme, pvcpu, inout.eax); 713 vmentry_inout_write(vcpu, inout.port, bytes); 714 return (error); 715 } 716 717 error = emulate_inout(ctx, vcpu, &inout); 718 if (error) { 719 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n", 720 in ? "in" : "out", 721 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'), 722 inout.port, vme->rip); 723 return (VMEXIT_ABORT); 724 } else { 725 /* 726 * Communicate the status of the inout operation back to the 727 * in-kernel instruction emulation. 728 */ 729 if (in) { 730 vmentry_inout_read(vcpu, inout.port, bytes, inout.eax); 731 } else { 732 vmentry_inout_write(vcpu, inout.port, bytes); 733 } 734 return (VMEXIT_CONTINUE); 735 } 736 } 737 738 static int 739 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 740 { 741 uint64_t val; 742 uint32_t eax, edx; 743 int error; 744 745 val = 0; 746 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val); 747 if (error != 0) { 748 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n", 749 vme->u.msr.code, *pvcpu); 750 if (get_config_bool("x86.strictmsr")) { 751 vm_inject_gp(ctx, *pvcpu); 752 return (VMEXIT_CONTINUE); 753 } 754 } 755 756 eax = val; 757 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax); 758 assert(error == 0); 759 760 edx = val >> 32; 761 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx); 762 assert(error == 0); 763 764 return (VMEXIT_CONTINUE); 765 } 766 767 static int 768 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 769 { 770 int error; 771 772 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval); 773 if (error != 0) { 774 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n", 775 vme->u.msr.code, vme->u.msr.wval, *pvcpu); 776 if (get_config_bool("x86.strictmsr")) { 777 vm_inject_gp(ctx, *pvcpu); 778 return (VMEXIT_CONTINUE); 779 } 780 } 781 return (VMEXIT_CONTINUE); 782 } 783 784 #ifndef __FreeBSD__ 785 static int 786 vmexit_run_state(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 787 { 788 /* 789 * Run-state transitions (INIT, SIPI, etc) are handled in-kernel, so an 790 * exit to userspace with that code is not expected. 791 */ 792 fprintf(stderr, "unexpected run-state VM exit"); 793 return (VMEXIT_ABORT); 794 } 795 796 static int 797 vmexit_paging(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 798 { 799 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 800 fprintf(stderr, "\treason\t\tPAGING\n"); 801 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 802 fprintf(stderr, "\tgpa\t\t0x%016lx\n", vmexit->u.paging.gpa); 803 fprintf(stderr, "\tfault_type\t\t%d\n", vmexit->u.paging.fault_type); 804 805 return (VMEXIT_ABORT); 806 } 807 #endif /* __FreeBSD__ */ 808 809 #ifdef __FreeBSD__ 810 #define DEBUG_EPT_MISCONFIG 811 #else 812 /* EPT misconfig debugging not possible now that raw VMCS access is gone */ 813 #endif 814 815 #ifdef DEBUG_EPT_MISCONFIG 816 #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400 817 818 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4]; 819 static int ept_misconfig_ptenum; 820 #endif 821 822 static const char * 823 vmexit_vmx_desc(uint32_t exit_reason) 824 { 825 826 if (exit_reason >= nitems(vmx_exit_reason_desc) || 827 vmx_exit_reason_desc[exit_reason] == NULL) 828 return ("Unknown"); 829 return (vmx_exit_reason_desc[exit_reason]); 830 } 831 832 static int 833 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 834 { 835 836 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 837 fprintf(stderr, "\treason\t\tVMX\n"); 838 fprintf(stderr, "\trip\t\t0x%016lx\n", vme->rip); 839 fprintf(stderr, "\tinst_length\t%d\n", vme->inst_length); 840 fprintf(stderr, "\tstatus\t\t%d\n", vme->u.vmx.status); 841 fprintf(stderr, "\texit_reason\t%u (%s)\n", vme->u.vmx.exit_reason, 842 vmexit_vmx_desc(vme->u.vmx.exit_reason)); 843 fprintf(stderr, "\tqualification\t0x%016lx\n", 844 vme->u.vmx.exit_qualification); 845 fprintf(stderr, "\tinst_type\t\t%d\n", vme->u.vmx.inst_type); 846 fprintf(stderr, "\tinst_error\t\t%d\n", vme->u.vmx.inst_error); 847 #ifdef DEBUG_EPT_MISCONFIG 848 if (vme->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) { 849 vm_get_register(ctx, *pvcpu, 850 VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS), 851 &ept_misconfig_gpa); 852 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte, 853 &ept_misconfig_ptenum); 854 fprintf(stderr, "\tEPT misconfiguration:\n"); 855 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa); 856 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n", 857 ept_misconfig_ptenum, ept_misconfig_pte[0], 858 ept_misconfig_pte[1], ept_misconfig_pte[2], 859 ept_misconfig_pte[3]); 860 } 861 #endif /* DEBUG_EPT_MISCONFIG */ 862 return (VMEXIT_ABORT); 863 } 864 865 static int 866 vmexit_svm(struct vmctx *ctx __unused, struct vm_exit *vme, int *pvcpu) 867 { 868 869 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 870 fprintf(stderr, "\treason\t\tSVM\n"); 871 fprintf(stderr, "\trip\t\t0x%016lx\n", vme->rip); 872 fprintf(stderr, "\tinst_length\t%d\n", vme->inst_length); 873 fprintf(stderr, "\texitcode\t%#lx\n", vme->u.svm.exitcode); 874 fprintf(stderr, "\texitinfo1\t%#lx\n", vme->u.svm.exitinfo1); 875 fprintf(stderr, "\texitinfo2\t%#lx\n", vme->u.svm.exitinfo2); 876 return (VMEXIT_ABORT); 877 } 878 879 static int 880 vmexit_bogus(struct vmctx *ctx __unused, struct vm_exit *vme, 881 int *pvcpu __unused) 882 { 883 884 assert(vme->inst_length == 0); 885 886 stats.vmexit_bogus++; 887 888 return (VMEXIT_CONTINUE); 889 } 890 891 static int 892 vmexit_reqidle(struct vmctx *ctx __unused, struct vm_exit *vme, 893 int *pvcpu __unused) 894 { 895 896 assert(vme->inst_length == 0); 897 898 stats.vmexit_reqidle++; 899 900 return (VMEXIT_CONTINUE); 901 } 902 903 static int 904 vmexit_hlt(struct vmctx *ctx __unused, struct vm_exit *vme __unused, 905 int *pvcpu __unused) 906 { 907 908 stats.vmexit_hlt++; 909 910 /* 911 * Just continue execution with the next instruction. We use 912 * the HLT VM exit as a way to be friendly with the host 913 * scheduler. 914 */ 915 return (VMEXIT_CONTINUE); 916 } 917 918 static int 919 vmexit_pause(struct vmctx *ctx __unused, struct vm_exit *vme __unused, 920 int *pvcpu __unused) 921 { 922 923 stats.vmexit_pause++; 924 925 return (VMEXIT_CONTINUE); 926 } 927 928 static int 929 vmexit_mtrap(struct vmctx *ctx __unused, struct vm_exit *vme, int *pvcpu) 930 { 931 932 assert(vme->inst_length == 0); 933 934 stats.vmexit_mtrap++; 935 936 gdb_cpu_mtrap(*pvcpu); 937 938 return (VMEXIT_CONTINUE); 939 } 940 941 static int 942 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 943 { 944 uint8_t i, valid; 945 946 fprintf(stderr, "Failed to emulate instruction sequence "); 947 948 valid = vme->u.inst_emul.num_valid; 949 if (valid != 0) { 950 assert(valid <= sizeof (vme->u.inst_emul.inst)); 951 fprintf(stderr, "["); 952 for (i = 0; i < valid; i++) { 953 if (i == 0) { 954 fprintf(stderr, "%02x", 955 vme->u.inst_emul.inst[i]); 956 } else { 957 fprintf(stderr, ", %02x", 958 vme->u.inst_emul.inst[i]); 959 } 960 } 961 fprintf(stderr, "] "); 962 } 963 fprintf(stderr, "@ %rip = %x\n", vme->rip); 964 965 return (VMEXIT_ABORT); 966 } 967 968 static int 969 vmexit_mmio(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 970 { 971 int vcpu, err; 972 struct vm_mmio mmio; 973 bool is_read; 974 975 stats.vmexit_mmio++; 976 977 vcpu = *pvcpu; 978 mmio = vmexit->u.mmio; 979 is_read = (mmio.read != 0); 980 981 err = emulate_mem(ctx, vcpu, &mmio); 982 983 if (err == ESRCH) { 984 fprintf(stderr, "Unhandled memory access to 0x%lx\n", mmio.gpa); 985 stats.mmio_unhandled++; 986 987 /* 988 * Access to non-existent physical addresses is not likely to 989 * result in fatal errors on hardware machines, but rather reads 990 * of all-ones or discarded-but-acknowledged writes. 991 */ 992 mmio.data = ~0UL; 993 err = 0; 994 } 995 996 if (err == 0) { 997 if (is_read) { 998 vmentry_mmio_read(vcpu, mmio.gpa, mmio.bytes, 999 mmio.data); 1000 } else { 1001 vmentry_mmio_write(vcpu, mmio.gpa, mmio.bytes); 1002 } 1003 return (VMEXIT_CONTINUE); 1004 } 1005 1006 fprintf(stderr, "Unhandled mmio error to 0x%lx: %d\n", mmio.gpa, err); 1007 return (VMEXIT_ABORT); 1008 } 1009 1010 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER; 1011 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER; 1012 1013 static int 1014 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 1015 { 1016 enum vm_suspend_how how; 1017 1018 how = vme->u.suspended.how; 1019 1020 fbsdrun_deletecpu(*pvcpu); 1021 1022 if (*pvcpu != BSP) { 1023 pthread_mutex_lock(&resetcpu_mtx); 1024 pthread_cond_signal(&resetcpu_cond); 1025 pthread_mutex_unlock(&resetcpu_mtx); 1026 pthread_exit(NULL); 1027 } 1028 1029 pthread_mutex_lock(&resetcpu_mtx); 1030 while (!CPU_EMPTY(&cpumask)) { 1031 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx); 1032 } 1033 pthread_mutex_unlock(&resetcpu_mtx); 1034 1035 switch (how) { 1036 case VM_SUSPEND_RESET: 1037 exit(0); 1038 case VM_SUSPEND_POWEROFF: 1039 if (get_config_bool_default("destroy_on_poweroff", false)) 1040 vm_destroy(ctx); 1041 exit(1); 1042 case VM_SUSPEND_HALT: 1043 exit(2); 1044 case VM_SUSPEND_TRIPLEFAULT: 1045 exit(3); 1046 default: 1047 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how); 1048 exit(100); 1049 } 1050 return (0); /* NOTREACHED */ 1051 } 1052 1053 static int 1054 vmexit_debug(struct vmctx *ctx __unused, struct vm_exit *vme __unused, 1055 int *pvcpu) 1056 { 1057 1058 gdb_cpu_suspend(*pvcpu); 1059 return (VMEXIT_CONTINUE); 1060 } 1061 1062 static int 1063 vmexit_breakpoint(struct vmctx *ctx __unused, struct vm_exit *vme, int *pvcpu) 1064 { 1065 1066 gdb_cpu_breakpoint(*pvcpu, vme); 1067 return (VMEXIT_CONTINUE); 1068 } 1069 1070 #ifdef __FreeBSD__ 1071 static int 1072 vmexit_ipi(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu __unused) 1073 { 1074 int error = -1; 1075 int i; 1076 switch (vme->u.ipi.mode) { 1077 case APIC_DELMODE_INIT: 1078 CPU_FOREACH_ISSET(i, &vme->u.ipi.dmask) { 1079 error = vm_suspend_cpu(ctx, i); 1080 if (error) { 1081 warnx("%s: failed to suspend cpu %d\n", 1082 __func__, i); 1083 break; 1084 } 1085 } 1086 break; 1087 case APIC_DELMODE_STARTUP: 1088 CPU_FOREACH_ISSET(i, &vme->u.ipi.dmask) { 1089 spinup_ap(ctx, i, vme->u.ipi.vector << PAGE_SHIFT); 1090 } 1091 error = 0; 1092 break; 1093 default: 1094 break; 1095 } 1096 1097 return (error); 1098 } 1099 #endif 1100 1101 static vmexit_handler_t handler[VM_EXITCODE_MAX] = { 1102 [VM_EXITCODE_INOUT] = vmexit_inout, 1103 [VM_EXITCODE_MMIO] = vmexit_mmio, 1104 [VM_EXITCODE_VMX] = vmexit_vmx, 1105 [VM_EXITCODE_SVM] = vmexit_svm, 1106 [VM_EXITCODE_BOGUS] = vmexit_bogus, 1107 [VM_EXITCODE_REQIDLE] = vmexit_reqidle, 1108 [VM_EXITCODE_RDMSR] = vmexit_rdmsr, 1109 [VM_EXITCODE_WRMSR] = vmexit_wrmsr, 1110 [VM_EXITCODE_MTRAP] = vmexit_mtrap, 1111 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul, 1112 #ifndef __FreeBSD__ 1113 [VM_EXITCODE_RUN_STATE] = vmexit_run_state, 1114 [VM_EXITCODE_PAGING] = vmexit_paging, 1115 [VM_EXITCODE_HLT] = vmexit_hlt, 1116 #endif 1117 [VM_EXITCODE_SUSPENDED] = vmexit_suspend, 1118 [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch, 1119 [VM_EXITCODE_DEBUG] = vmexit_debug, 1120 [VM_EXITCODE_BPT] = vmexit_breakpoint, 1121 #ifdef __FreeBSD__ 1122 [VM_EXITCODE_IPI] = vmexit_ipi, 1123 #endif 1124 }; 1125 1126 static void 1127 vm_loop(struct vmctx *ctx, int vcpu) 1128 { 1129 struct vm_exit vme; 1130 int error, rc; 1131 enum vm_exitcode exitcode; 1132 cpuset_t active_cpus; 1133 struct vm_entry *ventry; 1134 1135 error = vm_active_cpus(ctx, &active_cpus); 1136 assert(CPU_ISSET(vcpu, &active_cpus)); 1137 1138 ventry = &vmentry[vcpu]; 1139 1140 while (1) { 1141 error = vm_run(ctx, vcpu, ventry, &vme); 1142 if (error != 0) 1143 break; 1144 1145 if (ventry->cmd != VEC_DEFAULT) { 1146 /* 1147 * Discard any lingering entry state after it has been 1148 * submitted via vm_run(). 1149 */ 1150 bzero(ventry, sizeof (*ventry)); 1151 } 1152 1153 exitcode = vme.exitcode; 1154 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) { 1155 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n", 1156 exitcode); 1157 exit(4); 1158 } 1159 1160 rc = (*handler[exitcode])(ctx, &vme, &vcpu); 1161 1162 switch (rc) { 1163 case VMEXIT_CONTINUE: 1164 break; 1165 case VMEXIT_ABORT: 1166 abort(); 1167 default: 1168 exit(4); 1169 } 1170 } 1171 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno); 1172 } 1173 1174 static int 1175 num_vcpus_allowed(struct vmctx *ctx) 1176 { 1177 uint16_t sockets, cores, threads, maxcpus; 1178 #ifdef __FreeBSD__ 1179 int tmp, error; 1180 1181 /* 1182 * The guest is allowed to spinup more than one processor only if the 1183 * UNRESTRICTED_GUEST capability is available. 1184 */ 1185 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp); 1186 if (error != 0) 1187 return (1); 1188 #else 1189 int error; 1190 /* Unrestricted Guest is always enabled on illumos */ 1191 1192 #endif /* __FreeBSD__ */ 1193 1194 error = vm_get_topology(ctx, &sockets, &cores, &threads, &maxcpus); 1195 if (error == 0) 1196 return (maxcpus); 1197 else 1198 return (1); 1199 } 1200 1201 static void 1202 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu) 1203 { 1204 int err, tmp; 1205 1206 #ifdef __FreeBSD__ 1207 if (get_config_bool_default("x86.vmexit_on_hlt", false)) { 1208 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp); 1209 if (err < 0) { 1210 fprintf(stderr, "VM exit on HLT not supported\n"); 1211 exit(4); 1212 } 1213 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1); 1214 if (cpu == BSP) 1215 handler[VM_EXITCODE_HLT] = vmexit_hlt; 1216 } 1217 #else 1218 /* 1219 * We insist that vmexit-on-hlt is available on the host CPU, and enable 1220 * it by default. Configuration of that feature is done with both of 1221 * those facts in mind. 1222 */ 1223 tmp = (int)get_config_bool_default("x86.vmexit_on_hlt", true); 1224 err = vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, tmp); 1225 if (err < 0) { 1226 fprintf(stderr, "VM exit on HLT not supported\n"); 1227 exit(4); 1228 } 1229 #endif /* __FreeBSD__ */ 1230 1231 if (get_config_bool_default("x86.vmexit_on_pause", false)) { 1232 /* 1233 * pause exit support required for this mode 1234 */ 1235 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp); 1236 if (err < 0) { 1237 fprintf(stderr, 1238 "SMP mux requested, no pause support\n"); 1239 exit(4); 1240 } 1241 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1); 1242 if (cpu == BSP) 1243 handler[VM_EXITCODE_PAUSE] = vmexit_pause; 1244 } 1245 1246 if (get_config_bool_default("x86.x2apic", false)) 1247 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED); 1248 else 1249 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED); 1250 1251 if (err) { 1252 fprintf(stderr, "Unable to set x2apic state (%d)\n", err); 1253 exit(4); 1254 } 1255 1256 #ifdef __FreeBSD__ 1257 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1); 1258 1259 err = vm_set_capability(ctx, cpu, VM_CAP_IPI_EXIT, 1); 1260 assert(err == 0); 1261 #endif 1262 } 1263 1264 static struct vmctx * 1265 do_open(const char *vmname) 1266 { 1267 struct vmctx *ctx; 1268 int error; 1269 bool reinit, romboot; 1270 1271 reinit = romboot = false; 1272 1273 if (lpc_bootrom()) 1274 romboot = true; 1275 #ifndef __FreeBSD__ 1276 uint64_t create_flags = 0; 1277 if (get_config_bool_default("memory.use_reservoir", false)) { 1278 create_flags |= VCF_RESERVOIR_MEM; 1279 } 1280 error = vm_create(vmname, create_flags); 1281 #else 1282 error = vm_create(vmname); 1283 #endif /* __FreeBSD__ */ 1284 if (error) { 1285 if (errno == EEXIST) { 1286 if (romboot) { 1287 reinit = true; 1288 } else { 1289 /* 1290 * The virtual machine has been setup by the 1291 * userspace bootloader. 1292 */ 1293 } 1294 } else { 1295 perror("vm_create"); 1296 exit(4); 1297 } 1298 } else { 1299 if (!romboot) { 1300 /* 1301 * If the virtual machine was just created then a 1302 * bootrom must be configured to boot it. 1303 */ 1304 fprintf(stderr, "virtual machine cannot be booted\n"); 1305 exit(4); 1306 } 1307 } 1308 1309 ctx = vm_open(vmname); 1310 if (ctx == NULL) { 1311 perror("vm_open"); 1312 exit(4); 1313 } 1314 1315 #ifndef WITHOUT_CAPSICUM 1316 if (vm_limit_rights(ctx) != 0) 1317 err(EX_OSERR, "vm_limit_rights"); 1318 #endif 1319 1320 if (reinit) { 1321 #ifndef __FreeBSD__ 1322 error = vm_reinit(ctx, 0); 1323 #else 1324 error = vm_reinit(ctx); 1325 #endif 1326 if (error) { 1327 perror("vm_reinit"); 1328 exit(4); 1329 } 1330 } 1331 error = vm_set_topology(ctx, cpu_sockets, cpu_cores, cpu_threads, 1332 0 /* maxcpus, unimplemented */); 1333 if (error) 1334 errx(EX_OSERR, "vm_set_topology"); 1335 return (ctx); 1336 } 1337 1338 static void 1339 spinup_vcpu(struct vmctx *ctx, int vcpu, bool suspend) 1340 { 1341 int error; 1342 1343 if (vcpu != BSP) { 1344 #ifndef __FreeBSD__ 1345 /* 1346 * On illumos, all APs are spun up halted and run-state 1347 * transitions (INIT, SIPI, etc) are handled in-kernel. 1348 */ 1349 spinup_ap(ctx, vcpu, 0); 1350 #endif 1351 1352 fbsdrun_set_capabilities(ctx, vcpu); 1353 1354 #ifdef __FreeBSD__ 1355 /* 1356 * Enable the 'unrestricted guest' mode for APs. 1357 * 1358 * APs startup in power-on 16-bit mode. 1359 */ 1360 error = vm_set_capability(ctx, vcpu, VM_CAP_UNRESTRICTED_GUEST, 1); 1361 assert(error == 0); 1362 #endif 1363 } 1364 1365 #ifndef __FreeBSD__ 1366 /* 1367 * The value of 'suspend' for the BSP depends on whether the -d 1368 * (suspend_at_boot) flag was given to bhyve. Regardless of that 1369 * value we always want to set the BSP to VRS_RUN and all others to 1370 * VRS_HALT. 1371 */ 1372 error = vm_set_run_state(ctx, vcpu, 1373 vcpu == BSP ? VRS_RUN : VRS_HALT, 0); 1374 assert(error == 0); 1375 #endif 1376 1377 fbsdrun_addcpu(ctx, vcpu, suspend); 1378 } 1379 1380 static bool 1381 parse_config_option(const char *option) 1382 { 1383 const char *value; 1384 char *path; 1385 1386 value = strchr(option, '='); 1387 if (value == NULL || value[1] == '\0') 1388 return (false); 1389 path = strndup(option, value - option); 1390 if (path == NULL) 1391 err(4, "Failed to allocate memory"); 1392 set_config_value(path, value + 1); 1393 return (true); 1394 } 1395 1396 static void 1397 parse_simple_config_file(const char *path) 1398 { 1399 FILE *fp; 1400 char *line, *cp; 1401 size_t linecap; 1402 unsigned int lineno; 1403 1404 fp = fopen(path, "r"); 1405 if (fp == NULL) 1406 err(4, "Failed to open configuration file %s", path); 1407 line = NULL; 1408 linecap = 0; 1409 lineno = 1; 1410 for (lineno = 1; getline(&line, &linecap, fp) > 0; lineno++) { 1411 if (*line == '#' || *line == '\n') 1412 continue; 1413 cp = strchr(line, '\n'); 1414 if (cp != NULL) 1415 *cp = '\0'; 1416 if (!parse_config_option(line)) 1417 errx(4, "%s line %u: invalid config option '%s'", path, 1418 lineno, line); 1419 } 1420 free(line); 1421 fclose(fp); 1422 } 1423 1424 static void 1425 parse_gdb_options(const char *opt) 1426 { 1427 const char *sport; 1428 char *colon; 1429 1430 if (opt[0] == 'w') { 1431 set_config_bool("gdb.wait", true); 1432 opt++; 1433 } 1434 1435 colon = strrchr(opt, ':'); 1436 if (colon == NULL) { 1437 sport = opt; 1438 } else { 1439 *colon = '\0'; 1440 colon++; 1441 sport = colon; 1442 set_config_value("gdb.address", opt); 1443 } 1444 1445 set_config_value("gdb.port", sport); 1446 } 1447 1448 static void 1449 set_defaults(void) 1450 { 1451 1452 set_config_bool("acpi_tables", false); 1453 set_config_value("memory.size", "256M"); 1454 set_config_bool("x86.strictmsr", true); 1455 } 1456 1457 int 1458 main(int argc, char *argv[]) 1459 { 1460 int c, error, err; 1461 int max_vcpus, memflags; 1462 struct vmctx *ctx; 1463 uint64_t rip; 1464 size_t memsize; 1465 const char *optstr, *value, *vmname; 1466 1467 init_config(); 1468 set_defaults(); 1469 progname = basename(argv[0]); 1470 1471 #ifdef __FreeBSD__ 1472 optstr = "aehuwxACDHIPSWYk:o:p:G:c:s:m:l:K:U:"; 1473 #else 1474 /* +d, +B, -p */ 1475 optstr = "adehuwxACDHIPSWYk:o:G:c:s:m:l:B:K:U:"; 1476 #endif 1477 while ((c = getopt(argc, argv, optstr)) != -1) { 1478 switch (c) { 1479 case 'a': 1480 set_config_bool("x86.x2apic", false); 1481 break; 1482 case 'A': 1483 #ifdef __FreeBSD__ 1484 /* 1485 * This option is ignored on illumos since the 1486 * generated ACPI tables are not used; the bootroms 1487 * have their own. The option is retained for backwards 1488 * compatibility but does nothing. Note that the 1489 * acpi_tables configuration is still accepted via 1490 * -o if somebody really wants to generate these tables. 1491 */ 1492 set_config_bool("acpi_tables", true); 1493 #endif 1494 break; 1495 case 'D': 1496 set_config_bool("destroy_on_poweroff", true); 1497 break; 1498 #ifndef __FreeBSD__ 1499 case 'B': 1500 if (smbios_parse(optarg) != 0) { 1501 errx(EX_USAGE, "invalid SMBIOS " 1502 "configuration '%s'", optarg); 1503 } 1504 break; 1505 case 'd': 1506 set_config_bool("suspend_at_boot", true); 1507 break; 1508 #endif 1509 #ifdef __FreeBSD__ 1510 case 'p': 1511 if (pincpu_parse(optarg) != 0) { 1512 errx(EX_USAGE, "invalid vcpu pinning " 1513 "configuration '%s'", optarg); 1514 } 1515 break; 1516 #endif 1517 case 'c': 1518 if (topology_parse(optarg) != 0) { 1519 errx(EX_USAGE, "invalid cpu topology " 1520 "'%s'", optarg); 1521 } 1522 break; 1523 case 'C': 1524 set_config_bool("memory.guest_in_core", true); 1525 break; 1526 case 'G': 1527 parse_gdb_options(optarg); 1528 break; 1529 case 'k': 1530 parse_simple_config_file(optarg); 1531 break; 1532 case 'K': 1533 set_config_value("keyboard.layout", optarg); 1534 break; 1535 case 'l': 1536 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1537 lpc_print_supported_devices(); 1538 exit(0); 1539 } else if (lpc_device_parse(optarg) != 0) { 1540 errx(EX_USAGE, "invalid lpc device " 1541 "configuration '%s'", optarg); 1542 } 1543 break; 1544 case 's': 1545 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1546 pci_print_supported_devices(); 1547 exit(0); 1548 } else if (pci_parse_slot(optarg) != 0) 1549 exit(4); 1550 else 1551 break; 1552 case 'S': 1553 set_config_bool("memory.wired", true); 1554 break; 1555 case 'm': 1556 set_config_value("memory.size", optarg); 1557 break; 1558 case 'o': 1559 if (!parse_config_option(optarg)) 1560 errx(EX_USAGE, "invalid configuration option '%s'", optarg); 1561 break; 1562 case 'H': 1563 set_config_bool("x86.vmexit_on_hlt", true); 1564 break; 1565 case 'I': 1566 /* 1567 * The "-I" option was used to add an ioapic to the 1568 * virtual machine. 1569 * 1570 * An ioapic is now provided unconditionally for each 1571 * virtual machine and this option is now deprecated. 1572 */ 1573 break; 1574 case 'P': 1575 set_config_bool("x86.vmexit_on_pause", true); 1576 break; 1577 case 'e': 1578 set_config_bool("x86.strictio", true); 1579 break; 1580 case 'u': 1581 set_config_bool("rtc.use_localtime", false); 1582 break; 1583 case 'U': 1584 set_config_value("uuid", optarg); 1585 break; 1586 case 'w': 1587 set_config_bool("x86.strictmsr", false); 1588 break; 1589 case 'W': 1590 set_config_bool("virtio_msix", false); 1591 break; 1592 case 'x': 1593 set_config_bool("x86.x2apic", true); 1594 break; 1595 case 'Y': 1596 set_config_bool("x86.mptable", false); 1597 break; 1598 case 'h': 1599 usage(0); 1600 default: 1601 usage(1); 1602 } 1603 } 1604 argc -= optind; 1605 argv += optind; 1606 1607 if (argc > 1) 1608 usage(1); 1609 1610 if (argc == 1) 1611 set_config_value("name", argv[0]); 1612 1613 vmname = get_config_value("name"); 1614 if (vmname == NULL) 1615 usage(1); 1616 1617 if (get_config_bool_default("config.dump", false)) { 1618 dump_config(); 1619 exit(1); 1620 } 1621 1622 #ifndef __FreeBSD__ 1623 illumos_priv_init(); 1624 #endif 1625 1626 calc_topology(); 1627 1628 #ifdef __FreeBSD__ 1629 build_vcpumaps(); 1630 #endif 1631 1632 value = get_config_value("memory.size"); 1633 error = vm_parse_memsize(value, &memsize); 1634 if (error) 1635 errx(EX_USAGE, "invalid memsize '%s'", value); 1636 1637 ctx = do_open(vmname); 1638 1639 max_vcpus = num_vcpus_allowed(ctx); 1640 if (guest_ncpus > max_vcpus) { 1641 fprintf(stderr, "%d vCPUs requested but only %d available\n", 1642 guest_ncpus, max_vcpus); 1643 exit(4); 1644 } 1645 1646 fbsdrun_set_capabilities(ctx, BSP); 1647 1648 memflags = 0; 1649 if (get_config_bool_default("memory.wired", false)) 1650 memflags |= VM_MEM_F_WIRED; 1651 if (get_config_bool_default("memory.guest_in_core", false)) 1652 memflags |= VM_MEM_F_INCORE; 1653 vm_set_memflags(ctx, memflags); 1654 #ifdef __FreeBSD__ 1655 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1656 #else 1657 do { 1658 errno = 0; 1659 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1660 error = errno; 1661 if (err != 0 && error == ENOMEM) { 1662 (void) fprintf(stderr, "Unable to allocate memory " 1663 "(%llu), retrying in 1 second\n", memsize); 1664 sleep(1); 1665 } 1666 } while (error == ENOMEM); 1667 #endif 1668 if (err) { 1669 fprintf(stderr, "Unable to set up memory (%d)\n", errno); 1670 exit(4); 1671 } 1672 1673 error = init_msr(); 1674 if (error) { 1675 fprintf(stderr, "init_msr error %d", error); 1676 exit(4); 1677 } 1678 1679 init_mem(guest_ncpus); 1680 init_inout(); 1681 #ifdef __FreeBSD__ 1682 kernemu_dev_init(); 1683 #endif 1684 init_bootrom(ctx); 1685 atkbdc_init(ctx); 1686 pci_irq_init(ctx); 1687 ioapic_init(ctx); 1688 1689 rtc_init(ctx); 1690 sci_init(ctx); 1691 #ifndef __FreeBSD__ 1692 pmtmr_init(ctx); 1693 #endif 1694 1695 /* 1696 * Exit if a device emulation finds an error in its initilization 1697 */ 1698 if (init_pci(ctx) != 0) { 1699 perror("device emulation initialization error"); 1700 exit(4); 1701 } 1702 1703 /* 1704 * Initialize after PCI, to allow a bootrom file to reserve the high 1705 * region. 1706 */ 1707 if (get_config_bool("acpi_tables")) 1708 vmgenc_init(ctx); 1709 1710 #ifdef __FreeBSD__ 1711 init_gdb(ctx); 1712 #else 1713 if (value != NULL) { 1714 int port = atoi(value); 1715 1716 if (port < 0) 1717 init_mdb(ctx); 1718 else 1719 init_gdb(ctx); 1720 } 1721 #endif 1722 1723 if (lpc_bootrom()) { 1724 #ifdef __FreeBSD__ 1725 if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) { 1726 fprintf(stderr, "ROM boot failed: unrestricted guest " 1727 "capability not available\n"); 1728 exit(4); 1729 } 1730 #else 1731 /* Unrestricted Guest is always enabled on illumos */ 1732 #endif 1733 error = vcpu_reset(ctx, BSP); 1734 assert(error == 0); 1735 } 1736 1737 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip); 1738 assert(error == 0); 1739 1740 /* 1741 * build the guest tables, MP etc. 1742 */ 1743 if (get_config_bool_default("x86.mptable", true)) { 1744 error = mptable_build(ctx, guest_ncpus); 1745 if (error) { 1746 perror("error to build the guest tables"); 1747 exit(4); 1748 } 1749 } 1750 1751 error = smbios_build(ctx); 1752 if (error != 0) 1753 exit(4); 1754 1755 if (get_config_bool("acpi_tables")) { 1756 error = acpi_build(ctx, guest_ncpus); 1757 assert(error == 0); 1758 } 1759 1760 if (lpc_bootrom()) 1761 fwctl_init(); 1762 1763 /* 1764 * Change the proc title to include the VM name. 1765 */ 1766 setproctitle("%s", vmname); 1767 1768 #ifndef WITHOUT_CAPSICUM 1769 caph_cache_catpages(); 1770 1771 if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1) 1772 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1773 1774 if (caph_enter() == -1) 1775 errx(EX_OSERR, "cap_enter() failed"); 1776 #endif 1777 1778 #ifndef __FreeBSD__ 1779 illumos_priv_lock(); 1780 #endif 1781 1782 /* Allocate per-VCPU resources. */ 1783 mt_vmm_info = calloc(guest_ncpus, sizeof(*mt_vmm_info)); 1784 #ifndef __FreeBSD__ 1785 vmentry = calloc(guest_ncpus, sizeof(*vmentry)); 1786 #endif 1787 1788 /* 1789 * Add all vCPUs. 1790 */ 1791 for (int vcpu = 0; vcpu < guest_ncpus; vcpu++) { 1792 #ifdef __FreeBSD__ 1793 bool suspend = (vcpu != BSP); 1794 #else 1795 bool suspend = vcpu == BSP && 1796 get_config_bool_default("suspend_at_boot", false); 1797 #endif 1798 spinup_vcpu(ctx, vcpu, suspend); 1799 } 1800 1801 /* 1802 * Head off to the main event dispatch loop 1803 */ 1804 mevent_dispatch(); 1805 1806 exit(4); 1807 } 1808