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