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 */ 43 44 #include <sys/cdefs.h> 45 __FBSDID("$FreeBSD$"); 46 47 #include <sys/types.h> 48 #ifndef WITHOUT_CAPSICUM 49 #include <sys/capsicum.h> 50 #endif 51 #include <sys/mman.h> 52 #include <sys/time.h> 53 #include <sys/cpuset.h> 54 55 #ifdef __FreeBSD__ 56 #include <amd64/vmm/intel/vmcs.h> 57 #else 58 #include <intel/vmcs.h> 59 #endif 60 61 #include <machine/atomic.h> 62 #include <machine/segments.h> 63 64 #ifndef WITHOUT_CAPSICUM 65 #include <capsicum_helpers.h> 66 #endif 67 #include <stdio.h> 68 #include <stdlib.h> 69 #include <string.h> 70 #include <err.h> 71 #include <errno.h> 72 #include <libgen.h> 73 #include <unistd.h> 74 #include <assert.h> 75 #include <pthread.h> 76 #include <pthread_np.h> 77 #include <sysexits.h> 78 #include <stdbool.h> 79 #include <stdint.h> 80 81 #include <machine/vmm.h> 82 #ifndef WITHOUT_CAPSICUM 83 #include <machine/vmm_dev.h> 84 #endif 85 #include <vmmapi.h> 86 87 #include "bhyverun.h" 88 #include "acpi.h" 89 #include "atkbdc.h" 90 #include "console.h" 91 #include "inout.h" 92 #include "dbgport.h" 93 #include "fwctl.h" 94 #include "gdb.h" 95 #include "ioapic.h" 96 #include "mem.h" 97 #include "mevent.h" 98 #include "mptbl.h" 99 #include "pci_emul.h" 100 #include "pci_irq.h" 101 #include "pci_lpc.h" 102 #include "smbiostbl.h" 103 #include "xmsr.h" 104 #include "spinup_ap.h" 105 #include "rfb.h" 106 #include "rtc.h" 107 #include "vga.h" 108 109 #define GUEST_NIO_PORT 0x488 /* guest upcalls via i/o port */ 110 111 #define MB (1024UL * 1024) 112 #define GB (1024UL * MB) 113 114 static const char * const vmx_exit_reason_desc[] = { 115 [EXIT_REASON_EXCEPTION] = "Exception or non-maskable interrupt (NMI)", 116 [EXIT_REASON_EXT_INTR] = "External interrupt", 117 [EXIT_REASON_TRIPLE_FAULT] = "Triple fault", 118 [EXIT_REASON_INIT] = "INIT signal", 119 [EXIT_REASON_SIPI] = "Start-up IPI (SIPI)", 120 [EXIT_REASON_IO_SMI] = "I/O system-management interrupt (SMI)", 121 [EXIT_REASON_SMI] = "Other SMI", 122 [EXIT_REASON_INTR_WINDOW] = "Interrupt window", 123 [EXIT_REASON_NMI_WINDOW] = "NMI window", 124 [EXIT_REASON_TASK_SWITCH] = "Task switch", 125 [EXIT_REASON_CPUID] = "CPUID", 126 [EXIT_REASON_GETSEC] = "GETSEC", 127 [EXIT_REASON_HLT] = "HLT", 128 [EXIT_REASON_INVD] = "INVD", 129 [EXIT_REASON_INVLPG] = "INVLPG", 130 [EXIT_REASON_RDPMC] = "RDPMC", 131 [EXIT_REASON_RDTSC] = "RDTSC", 132 [EXIT_REASON_RSM] = "RSM", 133 [EXIT_REASON_VMCALL] = "VMCALL", 134 [EXIT_REASON_VMCLEAR] = "VMCLEAR", 135 [EXIT_REASON_VMLAUNCH] = "VMLAUNCH", 136 [EXIT_REASON_VMPTRLD] = "VMPTRLD", 137 [EXIT_REASON_VMPTRST] = "VMPTRST", 138 [EXIT_REASON_VMREAD] = "VMREAD", 139 [EXIT_REASON_VMRESUME] = "VMRESUME", 140 [EXIT_REASON_VMWRITE] = "VMWRITE", 141 [EXIT_REASON_VMXOFF] = "VMXOFF", 142 [EXIT_REASON_VMXON] = "VMXON", 143 [EXIT_REASON_CR_ACCESS] = "Control-register accesses", 144 [EXIT_REASON_DR_ACCESS] = "MOV DR", 145 [EXIT_REASON_INOUT] = "I/O instruction", 146 [EXIT_REASON_RDMSR] = "RDMSR", 147 [EXIT_REASON_WRMSR] = "WRMSR", 148 [EXIT_REASON_INVAL_VMCS] = 149 "VM-entry failure due to invalid guest state", 150 [EXIT_REASON_INVAL_MSR] = "VM-entry failure due to MSR loading", 151 [EXIT_REASON_MWAIT] = "MWAIT", 152 [EXIT_REASON_MTF] = "Monitor trap flag", 153 [EXIT_REASON_MONITOR] = "MONITOR", 154 [EXIT_REASON_PAUSE] = "PAUSE", 155 [EXIT_REASON_MCE_DURING_ENTRY] = 156 "VM-entry failure due to machine-check event", 157 [EXIT_REASON_TPR] = "TPR below threshold", 158 [EXIT_REASON_APIC_ACCESS] = "APIC access", 159 [EXIT_REASON_VIRTUALIZED_EOI] = "Virtualized EOI", 160 [EXIT_REASON_GDTR_IDTR] = "Access to GDTR or IDTR", 161 [EXIT_REASON_LDTR_TR] = "Access to LDTR or TR", 162 [EXIT_REASON_EPT_FAULT] = "EPT violation", 163 [EXIT_REASON_EPT_MISCONFIG] = "EPT misconfiguration", 164 [EXIT_REASON_INVEPT] = "INVEPT", 165 [EXIT_REASON_RDTSCP] = "RDTSCP", 166 [EXIT_REASON_VMX_PREEMPT] = "VMX-preemption timer expired", 167 [EXIT_REASON_INVVPID] = "INVVPID", 168 [EXIT_REASON_WBINVD] = "WBINVD", 169 [EXIT_REASON_XSETBV] = "XSETBV", 170 [EXIT_REASON_APIC_WRITE] = "APIC write", 171 [EXIT_REASON_RDRAND] = "RDRAND", 172 [EXIT_REASON_INVPCID] = "INVPCID", 173 [EXIT_REASON_VMFUNC] = "VMFUNC", 174 [EXIT_REASON_ENCLS] = "ENCLS", 175 [EXIT_REASON_RDSEED] = "RDSEED", 176 [EXIT_REASON_PM_LOG_FULL] = "Page-modification log full", 177 [EXIT_REASON_XSAVES] = "XSAVES", 178 [EXIT_REASON_XRSTORS] = "XRSTORS" 179 }; 180 181 typedef int (*vmexit_handler_t)(struct vmctx *, struct vm_exit *, int *vcpu); 182 extern int vmexit_task_switch(struct vmctx *, struct vm_exit *, int *vcpu); 183 184 char *vmname; 185 186 int guest_ncpus; 187 uint16_t cores, maxcpus, sockets, threads; 188 189 char *guest_uuid_str; 190 191 static int guest_vmexit_on_hlt, guest_vmexit_on_pause; 192 static int virtio_msix = 1; 193 static int x2apic_mode = 0; /* default is xAPIC */ 194 195 static int strictio; 196 static int strictmsr = 1; 197 198 static int acpi; 199 200 static char *progname; 201 static const int BSP = 0; 202 203 static cpuset_t cpumask; 204 205 static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip); 206 207 static struct vm_exit vmexit[VM_MAXCPU]; 208 209 struct bhyvestats { 210 uint64_t vmexit_bogus; 211 uint64_t vmexit_reqidle; 212 uint64_t vmexit_hlt; 213 uint64_t vmexit_pause; 214 uint64_t vmexit_mtrap; 215 uint64_t vmexit_inst_emul; 216 uint64_t cpu_switch_rotate; 217 uint64_t cpu_switch_direct; 218 } stats; 219 220 struct mt_vmm_info { 221 pthread_t mt_thr; 222 struct vmctx *mt_ctx; 223 int mt_vcpu; 224 } mt_vmm_info[VM_MAXCPU]; 225 226 #ifdef __FreeBSD__ 227 static cpuset_t *vcpumap[VM_MAXCPU] = { NULL }; 228 #endif 229 230 static void 231 usage(int code) 232 { 233 234 fprintf(stderr, 235 "Usage: %s [-abehuwxACHPSWY]\n" 236 " %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n" 237 " %*s [-g <gdb port>] [-l <lpc>]\n" 238 #ifdef __FreeBSD__ 239 " %*s [-m mem] [-p vcpu:hostcpu] [-s <pci>] [-U uuid] <vm>\n" 240 #else 241 " %*s [-m mem] [-s <pci>] [-U uuid] <vm>\n" 242 #endif 243 " -a: local apic is in xAPIC mode (deprecated)\n" 244 " -A: create ACPI tables\n" 245 " -c: number of cpus and/or topology specification\n" 246 " -C: include guest memory in core file\n" 247 #ifndef __FreeBSD__ 248 " -d: suspend cpu at boot\n" 249 #endif 250 " -e: exit on unhandled I/O access\n" 251 " -g: gdb port\n" 252 " -h: help\n" 253 " -H: vmexit from the guest on hlt\n" 254 " -l: LPC device configuration\n" 255 " -m: memory size\n" 256 #ifdef __FreeBSD__ 257 " -p: pin 'vcpu' to 'hostcpu'\n" 258 #endif 259 " -P: vmexit from the guest on pause\n" 260 " -s: <slot,driver,configinfo> PCI slot config\n" 261 " -S: guest memory cannot be swapped\n" 262 " -u: RTC keeps UTC time\n" 263 " -U: uuid\n" 264 " -w: ignore unimplemented MSRs\n" 265 " -W: force virtio to use single-vector MSI\n" 266 " -x: local apic is in x2APIC mode\n" 267 " -Y: disable MPtable generation\n", 268 progname, (int)strlen(progname), "", (int)strlen(progname), "", 269 (int)strlen(progname), ""); 270 271 exit(code); 272 } 273 274 /* 275 * XXX This parser is known to have the following issues: 276 * 1. It accepts null key=value tokens ",,". 277 * 2. It accepts whitespace after = and before value. 278 * 3. Values out of range of INT are silently wrapped. 279 * 4. It doesn't check non-final values. 280 * 5. The apparently bogus limits of UINT16_MAX are for future expansion. 281 * 282 * The acceptance of a null specification ('-c ""') is by design to match the 283 * manual page syntax specification, this results in a topology of 1 vCPU. 284 */ 285 static int 286 topology_parse(const char *opt) 287 { 288 uint64_t ncpus; 289 int c, chk, n, s, t, tmp; 290 char *cp, *str; 291 bool ns, scts; 292 293 c = 1, n = 1, s = 1, t = 1; 294 ns = false, scts = false; 295 str = strdup(opt); 296 if (str == NULL) 297 goto out; 298 299 while ((cp = strsep(&str, ",")) != NULL) { 300 if (sscanf(cp, "%i%n", &tmp, &chk) == 1) { 301 n = tmp; 302 ns = true; 303 } else if (sscanf(cp, "cpus=%i%n", &tmp, &chk) == 1) { 304 n = tmp; 305 ns = true; 306 } else if (sscanf(cp, "sockets=%i%n", &tmp, &chk) == 1) { 307 s = tmp; 308 scts = true; 309 } else if (sscanf(cp, "cores=%i%n", &tmp, &chk) == 1) { 310 c = tmp; 311 scts = true; 312 } else if (sscanf(cp, "threads=%i%n", &tmp, &chk) == 1) { 313 t = tmp; 314 scts = true; 315 #ifdef notyet /* Do not expose this until vmm.ko implements it */ 316 } else if (sscanf(cp, "maxcpus=%i%n", &tmp, &chk) == 1) { 317 m = tmp; 318 #endif 319 /* Skip the empty argument case from -c "" */ 320 } else if (cp[0] == '\0') 321 continue; 322 else 323 goto out; 324 /* Any trailing garbage causes an error */ 325 if (cp[chk] != '\0') 326 goto out; 327 } 328 free(str); 329 str = NULL; 330 331 /* 332 * Range check 1 <= n <= UINT16_MAX all values 333 */ 334 if (n < 1 || s < 1 || c < 1 || t < 1 || 335 n > UINT16_MAX || s > UINT16_MAX || c > UINT16_MAX || 336 t > UINT16_MAX) 337 return (-1); 338 339 /* If only the cpus was specified, use that as sockets */ 340 if (!scts) 341 s = n; 342 /* 343 * Compute sockets * cores * threads avoiding overflow 344 * The range check above insures these are 16 bit values 345 * If n was specified check it against computed ncpus 346 */ 347 ncpus = (uint64_t)s * c * t; 348 if (ncpus > UINT16_MAX || (ns && n != ncpus)) 349 return (-1); 350 351 guest_ncpus = ncpus; 352 sockets = s; 353 cores = c; 354 threads = t; 355 return(0); 356 357 out: 358 free(str); 359 return (-1); 360 } 361 362 #ifndef WITHOUT_CAPSICUM 363 /* 364 * 11-stable capsicum helpers 365 */ 366 static void 367 bhyve_caph_cache_catpages(void) 368 { 369 370 (void)catopen("libc", NL_CAT_LOCALE); 371 } 372 373 static int 374 bhyve_caph_limit_stdoe(void) 375 { 376 cap_rights_t rights; 377 unsigned long cmds[] = { TIOCGETA, TIOCGWINSZ }; 378 int i, fds[] = { STDOUT_FILENO, STDERR_FILENO }; 379 380 cap_rights_init(&rights, CAP_FCNTL, CAP_FSTAT, CAP_IOCTL); 381 cap_rights_set(&rights, CAP_WRITE); 382 383 for (i = 0; i < nitems(fds); i++) { 384 if (cap_rights_limit(fds[i], &rights) < 0 && errno != ENOSYS) 385 return (-1); 386 387 if (cap_ioctls_limit(fds[i], cmds, nitems(cmds)) < 0 && errno != ENOSYS) 388 return (-1); 389 390 if (cap_fcntls_limit(fds[i], CAP_FCNTL_GETFL) < 0 && errno != ENOSYS) 391 return (-1); 392 } 393 394 return (0); 395 } 396 397 #endif 398 399 #ifdef __FreeBSD__ 400 static int 401 pincpu_parse(const char *opt) 402 { 403 int vcpu, pcpu; 404 405 if (sscanf(opt, "%d:%d", &vcpu, &pcpu) != 2) { 406 fprintf(stderr, "invalid format: %s\n", opt); 407 return (-1); 408 } 409 410 if (vcpu < 0 || vcpu >= VM_MAXCPU) { 411 fprintf(stderr, "vcpu '%d' outside valid range from 0 to %d\n", 412 vcpu, VM_MAXCPU - 1); 413 return (-1); 414 } 415 416 if (pcpu < 0 || pcpu >= CPU_SETSIZE) { 417 fprintf(stderr, "hostcpu '%d' outside valid range from " 418 "0 to %d\n", pcpu, CPU_SETSIZE - 1); 419 return (-1); 420 } 421 422 if (vcpumap[vcpu] == NULL) { 423 if ((vcpumap[vcpu] = malloc(sizeof(cpuset_t))) == NULL) { 424 perror("malloc"); 425 return (-1); 426 } 427 CPU_ZERO(vcpumap[vcpu]); 428 } 429 CPU_SET(pcpu, vcpumap[vcpu]); 430 return (0); 431 } 432 #endif 433 434 void 435 vm_inject_fault(void *arg, int vcpu, int vector, int errcode_valid, 436 int errcode) 437 { 438 struct vmctx *ctx; 439 int error, restart_instruction; 440 441 ctx = arg; 442 restart_instruction = 1; 443 444 error = vm_inject_exception(ctx, vcpu, vector, errcode_valid, errcode, 445 restart_instruction); 446 assert(error == 0); 447 } 448 449 void * 450 paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len) 451 { 452 453 return (vm_map_gpa(ctx, gaddr, len)); 454 } 455 456 int 457 fbsdrun_vmexit_on_pause(void) 458 { 459 460 return (guest_vmexit_on_pause); 461 } 462 463 int 464 fbsdrun_vmexit_on_hlt(void) 465 { 466 467 return (guest_vmexit_on_hlt); 468 } 469 470 int 471 fbsdrun_virtio_msix(void) 472 { 473 474 return (virtio_msix); 475 } 476 477 static void * 478 fbsdrun_start_thread(void *param) 479 { 480 char tname[MAXCOMLEN + 1]; 481 struct mt_vmm_info *mtp; 482 int vcpu; 483 484 mtp = param; 485 vcpu = mtp->mt_vcpu; 486 487 snprintf(tname, sizeof(tname), "vcpu %d", vcpu); 488 pthread_set_name_np(mtp->mt_thr, tname); 489 490 gdb_cpu_add(vcpu); 491 492 vm_loop(mtp->mt_ctx, vcpu, vmexit[vcpu].rip); 493 494 /* not reached */ 495 exit(1); 496 return (NULL); 497 } 498 499 #ifdef __FreeBSD__ 500 void 501 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip) 502 #else 503 void 504 fbsdrun_addcpu(struct vmctx *ctx, int fromcpu, int newcpu, uint64_t rip, 505 bool suspend) 506 #endif 507 { 508 int error; 509 510 assert(fromcpu == BSP); 511 512 /* 513 * The 'newcpu' must be activated in the context of 'fromcpu'. If 514 * vm_activate_cpu() is delayed until newcpu's pthread starts running 515 * then vmm.ko is out-of-sync with bhyve and this can create a race 516 * with vm_suspend(). 517 */ 518 error = vm_activate_cpu(ctx, newcpu); 519 if (error != 0) 520 err(EX_OSERR, "could not activate CPU %d", newcpu); 521 522 CPU_SET_ATOMIC(newcpu, &cpumask); 523 524 #ifndef __FreeBSD__ 525 if (suspend) 526 (void) vm_suspend_cpu(ctx, newcpu); 527 #endif 528 529 /* 530 * Set up the vmexit struct to allow execution to start 531 * at the given RIP 532 */ 533 vmexit[newcpu].rip = rip; 534 vmexit[newcpu].inst_length = 0; 535 536 mt_vmm_info[newcpu].mt_ctx = ctx; 537 mt_vmm_info[newcpu].mt_vcpu = newcpu; 538 539 error = pthread_create(&mt_vmm_info[newcpu].mt_thr, NULL, 540 fbsdrun_start_thread, &mt_vmm_info[newcpu]); 541 assert(error == 0); 542 } 543 544 static int 545 fbsdrun_deletecpu(struct vmctx *ctx, int vcpu) 546 { 547 548 if (!CPU_ISSET(vcpu, &cpumask)) { 549 fprintf(stderr, "Attempting to delete unknown cpu %d\n", vcpu); 550 exit(4); 551 } 552 553 CPU_CLR_ATOMIC(vcpu, &cpumask); 554 return (CPU_EMPTY(&cpumask)); 555 } 556 557 static int 558 vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu, 559 uint32_t eax) 560 { 561 #if BHYVE_DEBUG 562 /* 563 * put guest-driven debug here 564 */ 565 #endif 566 return (VMEXIT_CONTINUE); 567 } 568 569 static int 570 vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 571 { 572 int error; 573 int bytes, port, in, out; 574 int vcpu; 575 576 vcpu = *pvcpu; 577 578 port = vme->u.inout.port; 579 bytes = vme->u.inout.bytes; 580 in = vme->u.inout.in; 581 out = !in; 582 583 /* Extra-special case of host notifications */ 584 if (out && port == GUEST_NIO_PORT) { 585 error = vmexit_handle_notify(ctx, vme, pvcpu, vme->u.inout.eax); 586 return (error); 587 } 588 589 error = emulate_inout(ctx, vcpu, vme, strictio); 590 if (error) { 591 fprintf(stderr, "Unhandled %s%c 0x%04x at 0x%lx\n", 592 in ? "in" : "out", 593 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'), 594 port, vmexit->rip); 595 return (VMEXIT_ABORT); 596 } else { 597 return (VMEXIT_CONTINUE); 598 } 599 } 600 601 static int 602 vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 603 { 604 uint64_t val; 605 uint32_t eax, edx; 606 int error; 607 608 val = 0; 609 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val); 610 if (error != 0) { 611 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n", 612 vme->u.msr.code, *pvcpu); 613 if (strictmsr) { 614 vm_inject_gp(ctx, *pvcpu); 615 return (VMEXIT_CONTINUE); 616 } 617 } 618 619 eax = val; 620 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax); 621 assert(error == 0); 622 623 edx = val >> 32; 624 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx); 625 assert(error == 0); 626 627 return (VMEXIT_CONTINUE); 628 } 629 630 static int 631 vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 632 { 633 int error; 634 635 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval); 636 if (error != 0) { 637 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n", 638 vme->u.msr.code, vme->u.msr.wval, *pvcpu); 639 if (strictmsr) { 640 vm_inject_gp(ctx, *pvcpu); 641 return (VMEXIT_CONTINUE); 642 } 643 } 644 return (VMEXIT_CONTINUE); 645 } 646 647 static int 648 vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 649 { 650 651 (void)spinup_ap(ctx, *pvcpu, 652 vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip); 653 654 return (VMEXIT_CONTINUE); 655 } 656 657 #define DEBUG_EPT_MISCONFIG 658 #ifdef DEBUG_EPT_MISCONFIG 659 #define VMCS_GUEST_PHYSICAL_ADDRESS 0x00002400 660 661 static uint64_t ept_misconfig_gpa, ept_misconfig_pte[4]; 662 static int ept_misconfig_ptenum; 663 #endif 664 665 static const char * 666 vmexit_vmx_desc(uint32_t exit_reason) 667 { 668 669 if (exit_reason >= nitems(vmx_exit_reason_desc) || 670 vmx_exit_reason_desc[exit_reason] == NULL) 671 return ("Unknown"); 672 return (vmx_exit_reason_desc[exit_reason]); 673 } 674 675 static int 676 vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 677 { 678 679 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 680 fprintf(stderr, "\treason\t\tVMX\n"); 681 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 682 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 683 fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status); 684 fprintf(stderr, "\texit_reason\t%u (%s)\n", vmexit->u.vmx.exit_reason, 685 vmexit_vmx_desc(vmexit->u.vmx.exit_reason)); 686 fprintf(stderr, "\tqualification\t0x%016lx\n", 687 vmexit->u.vmx.exit_qualification); 688 fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type); 689 fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error); 690 #ifdef DEBUG_EPT_MISCONFIG 691 if (vmexit->u.vmx.exit_reason == EXIT_REASON_EPT_MISCONFIG) { 692 vm_get_register(ctx, *pvcpu, 693 VMCS_IDENT(VMCS_GUEST_PHYSICAL_ADDRESS), 694 &ept_misconfig_gpa); 695 vm_get_gpa_pmap(ctx, ept_misconfig_gpa, ept_misconfig_pte, 696 &ept_misconfig_ptenum); 697 fprintf(stderr, "\tEPT misconfiguration:\n"); 698 fprintf(stderr, "\t\tGPA: %#lx\n", ept_misconfig_gpa); 699 fprintf(stderr, "\t\tPTE(%d): %#lx %#lx %#lx %#lx\n", 700 ept_misconfig_ptenum, ept_misconfig_pte[0], 701 ept_misconfig_pte[1], ept_misconfig_pte[2], 702 ept_misconfig_pte[3]); 703 } 704 #endif /* DEBUG_EPT_MISCONFIG */ 705 return (VMEXIT_ABORT); 706 } 707 708 static int 709 vmexit_svm(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 710 { 711 712 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 713 fprintf(stderr, "\treason\t\tSVM\n"); 714 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 715 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 716 fprintf(stderr, "\texitcode\t%#lx\n", vmexit->u.svm.exitcode); 717 fprintf(stderr, "\texitinfo1\t%#lx\n", vmexit->u.svm.exitinfo1); 718 fprintf(stderr, "\texitinfo2\t%#lx\n", vmexit->u.svm.exitinfo2); 719 return (VMEXIT_ABORT); 720 } 721 722 static int 723 vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 724 { 725 726 assert(vmexit->inst_length == 0); 727 728 stats.vmexit_bogus++; 729 730 return (VMEXIT_CONTINUE); 731 } 732 733 static int 734 vmexit_reqidle(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 735 { 736 737 assert(vmexit->inst_length == 0); 738 739 stats.vmexit_reqidle++; 740 741 return (VMEXIT_CONTINUE); 742 } 743 744 static int 745 vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 746 { 747 748 stats.vmexit_hlt++; 749 750 /* 751 * Just continue execution with the next instruction. We use 752 * the HLT VM exit as a way to be friendly with the host 753 * scheduler. 754 */ 755 return (VMEXIT_CONTINUE); 756 } 757 758 static int 759 vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 760 { 761 762 stats.vmexit_pause++; 763 764 return (VMEXIT_CONTINUE); 765 } 766 767 static int 768 vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 769 { 770 771 assert(vmexit->inst_length == 0); 772 773 stats.vmexit_mtrap++; 774 775 gdb_cpu_mtrap(*pvcpu); 776 777 return (VMEXIT_CONTINUE); 778 } 779 780 static int 781 vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 782 { 783 int err, i; 784 struct vie *vie; 785 786 stats.vmexit_inst_emul++; 787 788 vie = &vmexit->u.inst_emul.vie; 789 err = emulate_mem(ctx, *pvcpu, vmexit->u.inst_emul.gpa, 790 vie, &vmexit->u.inst_emul.paging); 791 792 if (err) { 793 if (err == ESRCH) { 794 fprintf(stderr, "Unhandled memory access to 0x%lx\n", 795 vmexit->u.inst_emul.gpa); 796 } 797 798 fprintf(stderr, "Failed to emulate instruction ["); 799 for (i = 0; i < vie->num_valid; i++) { 800 fprintf(stderr, "0x%02x%s", vie->inst[i], 801 i != (vie->num_valid - 1) ? " " : ""); 802 } 803 fprintf(stderr, "] at 0x%lx\n", vmexit->rip); 804 return (VMEXIT_ABORT); 805 } 806 807 return (VMEXIT_CONTINUE); 808 } 809 810 static pthread_mutex_t resetcpu_mtx = PTHREAD_MUTEX_INITIALIZER; 811 static pthread_cond_t resetcpu_cond = PTHREAD_COND_INITIALIZER; 812 813 static int 814 vmexit_suspend(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 815 { 816 enum vm_suspend_how how; 817 818 how = vmexit->u.suspended.how; 819 820 fbsdrun_deletecpu(ctx, *pvcpu); 821 822 if (*pvcpu != BSP) { 823 pthread_mutex_lock(&resetcpu_mtx); 824 pthread_cond_signal(&resetcpu_cond); 825 pthread_mutex_unlock(&resetcpu_mtx); 826 pthread_exit(NULL); 827 } 828 829 pthread_mutex_lock(&resetcpu_mtx); 830 while (!CPU_EMPTY(&cpumask)) { 831 pthread_cond_wait(&resetcpu_cond, &resetcpu_mtx); 832 } 833 pthread_mutex_unlock(&resetcpu_mtx); 834 835 switch (how) { 836 case VM_SUSPEND_RESET: 837 exit(0); 838 case VM_SUSPEND_POWEROFF: 839 exit(1); 840 case VM_SUSPEND_HALT: 841 exit(2); 842 case VM_SUSPEND_TRIPLEFAULT: 843 exit(3); 844 default: 845 fprintf(stderr, "vmexit_suspend: invalid reason %d\n", how); 846 exit(100); 847 } 848 return (0); /* NOTREACHED */ 849 } 850 851 static int 852 vmexit_debug(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 853 { 854 855 gdb_cpu_suspend(*pvcpu); 856 return (VMEXIT_CONTINUE); 857 } 858 859 static vmexit_handler_t handler[VM_EXITCODE_MAX] = { 860 [VM_EXITCODE_INOUT] = vmexit_inout, 861 [VM_EXITCODE_INOUT_STR] = vmexit_inout, 862 [VM_EXITCODE_VMX] = vmexit_vmx, 863 [VM_EXITCODE_SVM] = vmexit_svm, 864 [VM_EXITCODE_BOGUS] = vmexit_bogus, 865 [VM_EXITCODE_REQIDLE] = vmexit_reqidle, 866 [VM_EXITCODE_RDMSR] = vmexit_rdmsr, 867 [VM_EXITCODE_WRMSR] = vmexit_wrmsr, 868 [VM_EXITCODE_MTRAP] = vmexit_mtrap, 869 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul, 870 [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap, 871 [VM_EXITCODE_SUSPENDED] = vmexit_suspend, 872 [VM_EXITCODE_TASK_SWITCH] = vmexit_task_switch, 873 [VM_EXITCODE_DEBUG] = vmexit_debug, 874 }; 875 876 static void 877 vm_loop(struct vmctx *ctx, int vcpu, uint64_t startrip) 878 { 879 int error, rc; 880 enum vm_exitcode exitcode; 881 cpuset_t active_cpus; 882 883 #ifdef __FreeBSD__ 884 if (vcpumap[vcpu] != NULL) { 885 error = pthread_setaffinity_np(pthread_self(), 886 sizeof(cpuset_t), vcpumap[vcpu]); 887 assert(error == 0); 888 } 889 #endif 890 error = vm_active_cpus(ctx, &active_cpus); 891 assert(CPU_ISSET(vcpu, &active_cpus)); 892 893 error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RIP, startrip); 894 assert(error == 0); 895 896 while (1) { 897 error = vm_run(ctx, vcpu, &vmexit[vcpu]); 898 if (error != 0) 899 break; 900 901 exitcode = vmexit[vcpu].exitcode; 902 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) { 903 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n", 904 exitcode); 905 exit(4); 906 } 907 908 rc = (*handler[exitcode])(ctx, &vmexit[vcpu], &vcpu); 909 910 switch (rc) { 911 case VMEXIT_CONTINUE: 912 break; 913 case VMEXIT_ABORT: 914 abort(); 915 default: 916 exit(4); 917 } 918 } 919 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno); 920 } 921 922 static int 923 num_vcpus_allowed(struct vmctx *ctx) 924 { 925 int tmp, error; 926 927 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp); 928 929 /* 930 * The guest is allowed to spinup more than one processor only if the 931 * UNRESTRICTED_GUEST capability is available. 932 */ 933 if (error == 0) 934 return (VM_MAXCPU); 935 else 936 return (1); 937 } 938 939 void 940 fbsdrun_set_capabilities(struct vmctx *ctx, int cpu) 941 { 942 int err, tmp; 943 944 if (fbsdrun_vmexit_on_hlt()) { 945 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp); 946 if (err < 0) { 947 fprintf(stderr, "VM exit on HLT not supported\n"); 948 exit(4); 949 } 950 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1); 951 if (cpu == BSP) 952 handler[VM_EXITCODE_HLT] = vmexit_hlt; 953 } 954 955 if (fbsdrun_vmexit_on_pause()) { 956 /* 957 * pause exit support required for this mode 958 */ 959 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp); 960 if (err < 0) { 961 fprintf(stderr, 962 "SMP mux requested, no pause support\n"); 963 exit(4); 964 } 965 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1); 966 if (cpu == BSP) 967 handler[VM_EXITCODE_PAUSE] = vmexit_pause; 968 } 969 970 if (x2apic_mode) 971 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED); 972 else 973 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED); 974 975 if (err) { 976 fprintf(stderr, "Unable to set x2apic state (%d)\n", err); 977 exit(4); 978 } 979 980 #ifdef __FreeBSD__ 981 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1); 982 #endif 983 } 984 985 static struct vmctx * 986 do_open(const char *vmname) 987 { 988 struct vmctx *ctx; 989 int error; 990 bool reinit, romboot; 991 #ifndef WITHOUT_CAPSICUM 992 cap_rights_t rights; 993 const cap_ioctl_t *cmds; 994 size_t ncmds; 995 #endif 996 997 reinit = romboot = false; 998 999 if (lpc_bootrom()) 1000 romboot = true; 1001 1002 error = vm_create(vmname); 1003 if (error) { 1004 if (errno == EEXIST) { 1005 if (romboot) { 1006 reinit = true; 1007 } else { 1008 /* 1009 * The virtual machine has been setup by the 1010 * userspace bootloader. 1011 */ 1012 } 1013 } else { 1014 perror("vm_create"); 1015 exit(4); 1016 } 1017 } else { 1018 if (!romboot) { 1019 /* 1020 * If the virtual machine was just created then a 1021 * bootrom must be configured to boot it. 1022 */ 1023 fprintf(stderr, "virtual machine cannot be booted\n"); 1024 exit(4); 1025 } 1026 } 1027 1028 ctx = vm_open(vmname); 1029 if (ctx == NULL) { 1030 perror("vm_open"); 1031 exit(4); 1032 } 1033 1034 #ifndef WITHOUT_CAPSICUM 1035 cap_rights_init(&rights, CAP_IOCTL, CAP_MMAP_RW); 1036 if (caph_rights_limit(vm_get_device_fd(ctx), &rights) == -1) 1037 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1038 vm_get_ioctls(&ncmds); 1039 cmds = vm_get_ioctls(NULL); 1040 if (cmds == NULL) 1041 errx(EX_OSERR, "out of memory"); 1042 if (caph_ioctls_limit(vm_get_device_fd(ctx), cmds, ncmds) == -1) 1043 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1044 free((cap_ioctl_t *)cmds); 1045 #endif 1046 1047 if (reinit) { 1048 error = vm_reinit(ctx); 1049 if (error) { 1050 perror("vm_reinit"); 1051 exit(4); 1052 } 1053 } 1054 error = vm_set_topology(ctx, sockets, cores, threads, maxcpus); 1055 if (error) 1056 errx(EX_OSERR, "vm_set_topology"); 1057 return (ctx); 1058 } 1059 1060 int 1061 main(int argc, char *argv[]) 1062 { 1063 int c, error, dbg_port, gdb_port, err, bvmcons; 1064 int max_vcpus, mptgen, memflags; 1065 int rtc_localtime; 1066 bool gdb_stop; 1067 #ifndef __FreeBSD__ 1068 bool suspend = false; 1069 #endif 1070 struct vmctx *ctx; 1071 uint64_t rip; 1072 size_t memsize; 1073 char *optstr; 1074 1075 bvmcons = 0; 1076 progname = basename(argv[0]); 1077 dbg_port = 0; 1078 gdb_port = 0; 1079 gdb_stop = false; 1080 guest_ncpus = 1; 1081 sockets = cores = threads = 1; 1082 maxcpus = 0; 1083 memsize = 256 * MB; 1084 mptgen = 1; 1085 rtc_localtime = 1; 1086 memflags = 0; 1087 1088 #ifdef __FreeBSD__ 1089 optstr = "abehuwxACHIPSWYp:g:G:c:s:m:l:B:U:"; 1090 #else 1091 optstr = "abdehuwxACHIPSWYg:G:c:s:m:l:B:U:"; 1092 #endif 1093 while ((c = getopt(argc, argv, optstr)) != -1) { 1094 switch (c) { 1095 case 'a': 1096 x2apic_mode = 0; 1097 break; 1098 case 'A': 1099 acpi = 1; 1100 break; 1101 case 'b': 1102 bvmcons = 1; 1103 break; 1104 case 'B': 1105 if (smbios_parse(optarg) != 0) { 1106 errx(EX_USAGE, "invalid SMBIOS " 1107 "configuration '%s'", optarg); 1108 } 1109 break; 1110 #ifndef __FreeBSD__ 1111 case 'd': 1112 suspend = true; 1113 break; 1114 #else 1115 case 'p': 1116 if (pincpu_parse(optarg) != 0) { 1117 errx(EX_USAGE, "invalid vcpu pinning " 1118 "configuration '%s'", optarg); 1119 } 1120 break; 1121 #endif 1122 case 'c': 1123 if (topology_parse(optarg) != 0) { 1124 errx(EX_USAGE, "invalid cpu topology " 1125 "'%s'", optarg); 1126 } 1127 break; 1128 case 'C': 1129 memflags |= VM_MEM_F_INCORE; 1130 break; 1131 case 'g': 1132 dbg_port = atoi(optarg); 1133 break; 1134 case 'G': 1135 if (optarg[0] == 'w') { 1136 gdb_stop = true; 1137 optarg++; 1138 } 1139 gdb_port = atoi(optarg); 1140 break; 1141 case 'l': 1142 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1143 lpc_print_supported_devices(); 1144 exit(0); 1145 } else if (lpc_device_parse(optarg) != 0) { 1146 errx(EX_USAGE, "invalid lpc device " 1147 "configuration '%s'", optarg); 1148 } 1149 break; 1150 case 's': 1151 if (strncmp(optarg, "help", strlen(optarg)) == 0) { 1152 pci_print_supported_devices(); 1153 exit(0); 1154 } else if (pci_parse_slot(optarg) != 0) 1155 exit(4); 1156 else 1157 break; 1158 case 'S': 1159 memflags |= VM_MEM_F_WIRED; 1160 break; 1161 case 'm': 1162 error = vm_parse_memsize(optarg, &memsize); 1163 if (error) 1164 errx(EX_USAGE, "invalid memsize '%s'", optarg); 1165 break; 1166 case 'H': 1167 guest_vmexit_on_hlt = 1; 1168 break; 1169 case 'I': 1170 /* 1171 * The "-I" option was used to add an ioapic to the 1172 * virtual machine. 1173 * 1174 * An ioapic is now provided unconditionally for each 1175 * virtual machine and this option is now deprecated. 1176 */ 1177 break; 1178 case 'P': 1179 guest_vmexit_on_pause = 1; 1180 break; 1181 case 'e': 1182 strictio = 1; 1183 break; 1184 case 'u': 1185 rtc_localtime = 0; 1186 break; 1187 case 'U': 1188 guest_uuid_str = optarg; 1189 break; 1190 case 'w': 1191 strictmsr = 0; 1192 break; 1193 case 'W': 1194 virtio_msix = 0; 1195 break; 1196 case 'x': 1197 x2apic_mode = 1; 1198 break; 1199 case 'Y': 1200 mptgen = 0; 1201 break; 1202 case 'h': 1203 usage(0); 1204 default: 1205 usage(1); 1206 } 1207 } 1208 argc -= optind; 1209 argv += optind; 1210 1211 if (argc != 1) 1212 usage(1); 1213 1214 vmname = argv[0]; 1215 ctx = do_open(vmname); 1216 1217 max_vcpus = num_vcpus_allowed(ctx); 1218 if (guest_ncpus > max_vcpus) { 1219 fprintf(stderr, "%d vCPUs requested but only %d available\n", 1220 guest_ncpus, max_vcpus); 1221 exit(4); 1222 } 1223 1224 fbsdrun_set_capabilities(ctx, BSP); 1225 1226 vm_set_memflags(ctx, memflags); 1227 #ifdef __FreeBSD__ 1228 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1229 #else 1230 do { 1231 errno = 0; 1232 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 1233 error = errno; 1234 if (err != 0 && error == ENOMEM) { 1235 (void) fprintf(stderr, "Unable to allocate memory " 1236 "(%llu), retrying in 1 second\n", memsize); 1237 sleep(1); 1238 } 1239 } while (error == ENOMEM); 1240 #endif 1241 if (err) { 1242 fprintf(stderr, "Unable to setup memory (%d)\n", errno); 1243 exit(4); 1244 } 1245 1246 error = init_msr(); 1247 if (error) { 1248 fprintf(stderr, "init_msr error %d", error); 1249 exit(4); 1250 } 1251 1252 init_mem(); 1253 init_inout(); 1254 atkbdc_init(ctx); 1255 pci_irq_init(ctx); 1256 ioapic_init(ctx); 1257 1258 rtc_init(ctx, rtc_localtime); 1259 sci_init(ctx); 1260 1261 /* 1262 * Exit if a device emulation finds an error in its initilization 1263 */ 1264 if (init_pci(ctx) != 0) { 1265 perror("device emulation initialization error"); 1266 exit(4); 1267 } 1268 1269 if (dbg_port != 0) 1270 init_dbgport(dbg_port); 1271 1272 if (gdb_port != 0) 1273 init_gdb(ctx, gdb_port, gdb_stop); 1274 1275 if (bvmcons) 1276 init_bvmcons(); 1277 1278 vga_init(1); 1279 1280 if (lpc_bootrom()) { 1281 if (vm_set_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, 1)) { 1282 fprintf(stderr, "ROM boot failed: unrestricted guest " 1283 "capability not available\n"); 1284 exit(4); 1285 } 1286 error = vcpu_reset(ctx, BSP); 1287 assert(error == 0); 1288 } 1289 1290 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip); 1291 assert(error == 0); 1292 1293 /* 1294 * build the guest tables, MP etc. 1295 */ 1296 if (mptgen) { 1297 error = mptable_build(ctx, guest_ncpus); 1298 if (error) { 1299 perror("error to build the guest tables"); 1300 exit(4); 1301 } 1302 } 1303 1304 error = smbios_build(ctx); 1305 assert(error == 0); 1306 1307 if (acpi) { 1308 error = acpi_build(ctx, guest_ncpus); 1309 assert(error == 0); 1310 } 1311 1312 if (lpc_bootrom()) 1313 fwctl_init(); 1314 1315 /* 1316 * Change the proc title to include the VM name. 1317 */ 1318 setproctitle("%s", vmname); 1319 1320 #ifndef WITHOUT_CAPSICUM 1321 caph_cache_catpages(); 1322 1323 if (caph_limit_stdout() == -1 || caph_limit_stderr() == -1) 1324 errx(EX_OSERR, "Unable to apply rights for sandbox"); 1325 1326 if (caph_enter() == -1) 1327 errx(EX_OSERR, "cap_enter() failed"); 1328 #endif 1329 1330 /* 1331 * Add CPU 0 1332 */ 1333 #ifdef __FreeBSD__ 1334 fbsdrun_addcpu(ctx, BSP, BSP, rip); 1335 #else 1336 fbsdrun_addcpu(ctx, BSP, BSP, rip, suspend); 1337 #endif 1338 /* 1339 * Head off to the main event dispatch loop 1340 */ 1341 mevent_dispatch(); 1342 1343 exit(4); 1344 } 1345