bhyverun.c revision 267427
1/*- 2 * Copyright (c) 2011 NetApp, Inc. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 * 26 * $FreeBSD: stable/10/usr.sbin/bhyve/bhyverun.c 267427 2014-06-12 19:58:12Z jhb $ 27 */ 28 29#include <sys/cdefs.h> 30__FBSDID("$FreeBSD: stable/10/usr.sbin/bhyve/bhyverun.c 267427 2014-06-12 19:58:12Z jhb $"); 31 32#include <sys/types.h> 33#include <sys/mman.h> 34#include <sys/time.h> 35 36#include <machine/atomic.h> 37#include <machine/segments.h> 38 39#include <stdio.h> 40#include <stdlib.h> 41#include <string.h> 42#include <err.h> 43#include <libgen.h> 44#include <unistd.h> 45#include <assert.h> 46#include <errno.h> 47#include <pthread.h> 48#include <pthread_np.h> 49#include <sysexits.h> 50 51#include <machine/vmm.h> 52#include <vmmapi.h> 53 54#include "bhyverun.h" 55#include "acpi.h" 56#include "inout.h" 57#include "dbgport.h" 58#include "ioapic.h" 59#include "mem.h" 60#include "mevent.h" 61#include "mptbl.h" 62#include "pci_emul.h" 63#include "pci_lpc.h" 64#include "xmsr.h" 65#include "spinup_ap.h" 66#include "rtc.h" 67 68#define GUEST_NIO_PORT 0x488 /* guest upcalls via i/o port */ 69 70#define VMEXIT_SWITCH 0 /* force vcpu switch in mux mode */ 71#define VMEXIT_CONTINUE 1 /* continue from next instruction */ 72#define VMEXIT_RESTART 2 /* restart current instruction */ 73#define VMEXIT_ABORT 3 /* abort the vm run loop */ 74#define VMEXIT_RESET 4 /* guest machine has reset */ 75#define VMEXIT_POWEROFF 5 /* guest machine has powered off */ 76 77#define MB (1024UL * 1024) 78#define GB (1024UL * MB) 79 80typedef int (*vmexit_handler_t)(struct vmctx *, struct vm_exit *, int *vcpu); 81 82char *vmname; 83 84int guest_ncpus; 85 86static int pincpu = -1; 87static int guest_vmexit_on_hlt, guest_vmexit_on_pause, disable_x2apic; 88static int virtio_msix = 1; 89 90static int strictio; 91static int strictmsr = 1; 92 93static int acpi; 94 95static char *progname; 96static const int BSP = 0; 97 98static int cpumask; 99 100static void vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip); 101 102struct vm_exit vmexit[VM_MAXCPU]; 103 104struct bhyvestats { 105 uint64_t vmexit_bogus; 106 uint64_t vmexit_bogus_switch; 107 uint64_t vmexit_hlt; 108 uint64_t vmexit_pause; 109 uint64_t vmexit_mtrap; 110 uint64_t vmexit_inst_emul; 111 uint64_t cpu_switch_rotate; 112 uint64_t cpu_switch_direct; 113 int io_reset; 114} stats; 115 116struct mt_vmm_info { 117 pthread_t mt_thr; 118 struct vmctx *mt_ctx; 119 int mt_vcpu; 120} mt_vmm_info[VM_MAXCPU]; 121 122static void 123usage(int code) 124{ 125 126 fprintf(stderr, 127 "Usage: %s [-aehwAHIPW] [-g <gdb port>] [-s <pci>]\n" 128 " %*s [-c vcpus] [-p pincpu] [-m mem] [-l <lpc>] <vm>\n" 129 " -a: local apic is in XAPIC mode (default is X2APIC)\n" 130 " -A: create an ACPI table\n" 131 " -g: gdb port\n" 132 " -c: # cpus (default 1)\n" 133 " -p: pin vcpu 'n' to host cpu 'pincpu + n'\n" 134 " -H: vmexit from the guest on hlt\n" 135 " -P: vmexit from the guest on pause\n" 136 " -W: force virtio to use single-vector MSI\n" 137 " -e: exit on unhandled I/O access\n" 138 " -h: help\n" 139 " -s: <slot,driver,configinfo> PCI slot config\n" 140 " -l: LPC device configuration\n" 141 " -m: memory size in MB\n" 142 " -w: ignore unimplemented MSRs\n", 143 progname, (int)strlen(progname), ""); 144 145 exit(code); 146} 147 148void * 149paddr_guest2host(struct vmctx *ctx, uintptr_t gaddr, size_t len) 150{ 151 152 return (vm_map_gpa(ctx, gaddr, len)); 153} 154 155int 156fbsdrun_disable_x2apic(void) 157{ 158 159 return (disable_x2apic); 160} 161 162int 163fbsdrun_vmexit_on_pause(void) 164{ 165 166 return (guest_vmexit_on_pause); 167} 168 169int 170fbsdrun_vmexit_on_hlt(void) 171{ 172 173 return (guest_vmexit_on_hlt); 174} 175 176int 177fbsdrun_virtio_msix(void) 178{ 179 180 return (virtio_msix); 181} 182 183static void * 184fbsdrun_start_thread(void *param) 185{ 186 char tname[MAXCOMLEN + 1]; 187 struct mt_vmm_info *mtp; 188 int vcpu; 189 190 mtp = param; 191 vcpu = mtp->mt_vcpu; 192 193 snprintf(tname, sizeof(tname), "vcpu %d", vcpu); 194 pthread_set_name_np(mtp->mt_thr, tname); 195 196 vm_loop(mtp->mt_ctx, vcpu, vmexit[vcpu].rip); 197 198 /* not reached */ 199 exit(1); 200 return (NULL); 201} 202 203void 204fbsdrun_addcpu(struct vmctx *ctx, int vcpu, uint64_t rip) 205{ 206 int error; 207 208 if (cpumask & (1 << vcpu)) { 209 fprintf(stderr, "addcpu: attempting to add existing cpu %d\n", 210 vcpu); 211 exit(1); 212 } 213 214 atomic_set_int(&cpumask, 1 << vcpu); 215 216 /* 217 * Set up the vmexit struct to allow execution to start 218 * at the given RIP 219 */ 220 vmexit[vcpu].rip = rip; 221 vmexit[vcpu].inst_length = 0; 222 223 mt_vmm_info[vcpu].mt_ctx = ctx; 224 mt_vmm_info[vcpu].mt_vcpu = vcpu; 225 226 error = pthread_create(&mt_vmm_info[vcpu].mt_thr, NULL, 227 fbsdrun_start_thread, &mt_vmm_info[vcpu]); 228 assert(error == 0); 229} 230 231static int 232fbsdrun_deletecpu(struct vmctx *ctx, int vcpu) 233{ 234 235 if ((cpumask & (1 << vcpu)) == 0) { 236 fprintf(stderr, "addcpu: attempting to delete unknown cpu %d\n", 237 vcpu); 238 exit(1); 239 } 240 241 atomic_clear_int(&cpumask, 1 << vcpu); 242 return (cpumask == 0); 243} 244 245static int 246vmexit_catch_reset(void) 247{ 248 stats.io_reset++; 249 return (VMEXIT_RESET); 250} 251 252static int 253vmexit_catch_inout(void) 254{ 255 return (VMEXIT_ABORT); 256} 257 258static int 259vmexit_handle_notify(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu, 260 uint32_t eax) 261{ 262#if BHYVE_DEBUG 263 /* 264 * put guest-driven debug here 265 */ 266#endif 267 return (VMEXIT_CONTINUE); 268} 269 270static int 271vmexit_inout(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 272{ 273 int error; 274 int bytes, port, in, out; 275 uint32_t eax; 276 int vcpu; 277 278 vcpu = *pvcpu; 279 280 port = vme->u.inout.port; 281 bytes = vme->u.inout.bytes; 282 eax = vme->u.inout.eax; 283 in = vme->u.inout.in; 284 out = !in; 285 286 /* We don't deal with these */ 287 if (vme->u.inout.string || vme->u.inout.rep) 288 return (VMEXIT_ABORT); 289 290 /* Special case of guest reset */ 291 if (out && port == 0x64 && (uint8_t)eax == 0xFE) 292 return (vmexit_catch_reset()); 293 294 /* Extra-special case of host notifications */ 295 if (out && port == GUEST_NIO_PORT) 296 return (vmexit_handle_notify(ctx, vme, pvcpu, eax)); 297 298 error = emulate_inout(ctx, vcpu, in, port, bytes, &eax, strictio); 299 if (error == INOUT_OK && in) 300 error = vm_set_register(ctx, vcpu, VM_REG_GUEST_RAX, eax); 301 302 switch (error) { 303 case INOUT_OK: 304 return (VMEXIT_CONTINUE); 305 case INOUT_RESET: 306 return (VMEXIT_RESET); 307 case INOUT_POWEROFF: 308 return (VMEXIT_POWEROFF); 309 default: 310 fprintf(stderr, "Unhandled %s%c 0x%04x\n", 311 in ? "in" : "out", 312 bytes == 1 ? 'b' : (bytes == 2 ? 'w' : 'l'), port); 313 return (vmexit_catch_inout()); 314 } 315} 316 317static int 318vmexit_rdmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 319{ 320 uint64_t val; 321 uint32_t eax, edx; 322 int error; 323 324 val = 0; 325 error = emulate_rdmsr(ctx, *pvcpu, vme->u.msr.code, &val); 326 if (error != 0) { 327 fprintf(stderr, "rdmsr to register %#x on vcpu %d\n", 328 vme->u.msr.code, *pvcpu); 329 if (strictmsr) { 330 error = vm_inject_exception2(ctx, *pvcpu, IDT_GP, 0); 331 assert(error == 0); 332 return (VMEXIT_RESTART); 333 } 334 } 335 336 eax = val; 337 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RAX, eax); 338 assert(error == 0); 339 340 edx = val >> 32; 341 error = vm_set_register(ctx, *pvcpu, VM_REG_GUEST_RDX, edx); 342 assert(error == 0); 343 344 return (VMEXIT_CONTINUE); 345} 346 347static int 348vmexit_wrmsr(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 349{ 350 int error; 351 352 error = emulate_wrmsr(ctx, *pvcpu, vme->u.msr.code, vme->u.msr.wval); 353 if (error != 0) { 354 fprintf(stderr, "wrmsr to register %#x(%#lx) on vcpu %d\n", 355 vme->u.msr.code, vme->u.msr.wval, *pvcpu); 356 if (strictmsr) { 357 error = vm_inject_exception2(ctx, *pvcpu, IDT_GP, 0); 358 assert(error == 0); 359 return (VMEXIT_RESTART); 360 } 361 } 362 return (VMEXIT_CONTINUE); 363} 364 365static int 366vmexit_spinup_ap(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 367{ 368 int newcpu; 369 int retval = VMEXIT_CONTINUE; 370 371 newcpu = spinup_ap(ctx, *pvcpu, 372 vme->u.spinup_ap.vcpu, vme->u.spinup_ap.rip); 373 374 return (retval); 375} 376 377static int 378vmexit_spindown_cpu(struct vmctx *ctx, struct vm_exit *vme, int *pvcpu) 379{ 380 int lastcpu; 381 382 lastcpu = fbsdrun_deletecpu(ctx, *pvcpu); 383 if (!lastcpu) 384 pthread_exit(NULL); 385 return (vmexit_catch_reset()); 386} 387 388static int 389vmexit_vmx(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 390{ 391 392 fprintf(stderr, "vm exit[%d]\n", *pvcpu); 393 fprintf(stderr, "\treason\t\tVMX\n"); 394 fprintf(stderr, "\trip\t\t0x%016lx\n", vmexit->rip); 395 fprintf(stderr, "\tinst_length\t%d\n", vmexit->inst_length); 396 fprintf(stderr, "\tstatus\t\t%d\n", vmexit->u.vmx.status); 397 fprintf(stderr, "\texit_reason\t%u\n", vmexit->u.vmx.exit_reason); 398 fprintf(stderr, "\tqualification\t0x%016lx\n", 399 vmexit->u.vmx.exit_qualification); 400 fprintf(stderr, "\tinst_type\t\t%d\n", vmexit->u.vmx.inst_type); 401 fprintf(stderr, "\tinst_error\t\t%d\n", vmexit->u.vmx.inst_error); 402 403 return (VMEXIT_ABORT); 404} 405 406static int 407vmexit_bogus(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 408{ 409 410 stats.vmexit_bogus++; 411 412 return (VMEXIT_RESTART); 413} 414 415static int 416vmexit_hlt(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 417{ 418 419 stats.vmexit_hlt++; 420 421 /* 422 * Just continue execution with the next instruction. We use 423 * the HLT VM exit as a way to be friendly with the host 424 * scheduler. 425 */ 426 return (VMEXIT_CONTINUE); 427} 428 429static int 430vmexit_pause(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 431{ 432 433 stats.vmexit_pause++; 434 435 return (VMEXIT_CONTINUE); 436} 437 438static int 439vmexit_mtrap(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 440{ 441 442 stats.vmexit_mtrap++; 443 444 return (VMEXIT_RESTART); 445} 446 447static int 448vmexit_inst_emul(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu) 449{ 450 int err; 451 stats.vmexit_inst_emul++; 452 453 err = emulate_mem(ctx, *pvcpu, vmexit->u.inst_emul.gpa, 454 &vmexit->u.inst_emul.vie); 455 456 if (err) { 457 if (err == EINVAL) { 458 fprintf(stderr, 459 "Failed to emulate instruction at 0x%lx\n", 460 vmexit->rip); 461 } else if (err == ESRCH) { 462 fprintf(stderr, "Unhandled memory access to 0x%lx\n", 463 vmexit->u.inst_emul.gpa); 464 } 465 466 return (VMEXIT_ABORT); 467 } 468 469 return (VMEXIT_CONTINUE); 470} 471 472static vmexit_handler_t handler[VM_EXITCODE_MAX] = { 473 [VM_EXITCODE_INOUT] = vmexit_inout, 474 [VM_EXITCODE_VMX] = vmexit_vmx, 475 [VM_EXITCODE_BOGUS] = vmexit_bogus, 476 [VM_EXITCODE_RDMSR] = vmexit_rdmsr, 477 [VM_EXITCODE_WRMSR] = vmexit_wrmsr, 478 [VM_EXITCODE_MTRAP] = vmexit_mtrap, 479 [VM_EXITCODE_INST_EMUL] = vmexit_inst_emul, 480 [VM_EXITCODE_SPINUP_AP] = vmexit_spinup_ap, 481 [VM_EXITCODE_SPINDOWN_CPU] = vmexit_spindown_cpu, 482}; 483 484static void 485vm_loop(struct vmctx *ctx, int vcpu, uint64_t rip) 486{ 487 cpuset_t mask; 488 int error, rc, prevcpu; 489 enum vm_exitcode exitcode; 490 491 if (pincpu >= 0) { 492 CPU_ZERO(&mask); 493 CPU_SET(pincpu + vcpu, &mask); 494 error = pthread_setaffinity_np(pthread_self(), 495 sizeof(mask), &mask); 496 assert(error == 0); 497 } 498 499 while (1) { 500 error = vm_run(ctx, vcpu, rip, &vmexit[vcpu]); 501 if (error != 0) 502 break; 503 504 prevcpu = vcpu; 505 506 exitcode = vmexit[vcpu].exitcode; 507 if (exitcode >= VM_EXITCODE_MAX || handler[exitcode] == NULL) { 508 fprintf(stderr, "vm_loop: unexpected exitcode 0x%x\n", 509 exitcode); 510 exit(1); 511 } 512 513 rc = (*handler[exitcode])(ctx, &vmexit[vcpu], &vcpu); 514 515 switch (rc) { 516 case VMEXIT_CONTINUE: 517 rip = vmexit[vcpu].rip + vmexit[vcpu].inst_length; 518 break; 519 case VMEXIT_RESTART: 520 rip = vmexit[vcpu].rip; 521 break; 522 case VMEXIT_RESET: 523 exit(0); 524 default: 525 exit(1); 526 } 527 } 528 fprintf(stderr, "vm_run error %d, errno %d\n", error, errno); 529} 530 531static int 532num_vcpus_allowed(struct vmctx *ctx) 533{ 534 int tmp, error; 535 536 error = vm_get_capability(ctx, BSP, VM_CAP_UNRESTRICTED_GUEST, &tmp); 537 538 /* 539 * The guest is allowed to spinup more than one processor only if the 540 * UNRESTRICTED_GUEST capability is available. 541 */ 542 if (error == 0) 543 return (VM_MAXCPU); 544 else 545 return (1); 546} 547 548void 549fbsdrun_set_capabilities(struct vmctx *ctx, int cpu) 550{ 551 int err, tmp; 552 553 if (fbsdrun_vmexit_on_hlt()) { 554 err = vm_get_capability(ctx, cpu, VM_CAP_HALT_EXIT, &tmp); 555 if (err < 0) { 556 fprintf(stderr, "VM exit on HLT not supported\n"); 557 exit(1); 558 } 559 vm_set_capability(ctx, cpu, VM_CAP_HALT_EXIT, 1); 560 if (cpu == BSP) 561 handler[VM_EXITCODE_HLT] = vmexit_hlt; 562 } 563 564 if (fbsdrun_vmexit_on_pause()) { 565 /* 566 * pause exit support required for this mode 567 */ 568 err = vm_get_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, &tmp); 569 if (err < 0) { 570 fprintf(stderr, 571 "SMP mux requested, no pause support\n"); 572 exit(1); 573 } 574 vm_set_capability(ctx, cpu, VM_CAP_PAUSE_EXIT, 1); 575 if (cpu == BSP) 576 handler[VM_EXITCODE_PAUSE] = vmexit_pause; 577 } 578 579 if (fbsdrun_disable_x2apic()) 580 err = vm_set_x2apic_state(ctx, cpu, X2APIC_DISABLED); 581 else 582 err = vm_set_x2apic_state(ctx, cpu, X2APIC_ENABLED); 583 584 if (err) { 585 fprintf(stderr, "Unable to set x2apic state (%d)\n", err); 586 exit(1); 587 } 588 589 vm_set_capability(ctx, cpu, VM_CAP_ENABLE_INVPCID, 1); 590} 591 592int 593main(int argc, char *argv[]) 594{ 595 int c, error, gdb_port, err, bvmcons; 596 int max_vcpus; 597 struct vmctx *ctx; 598 uint64_t rip; 599 size_t memsize; 600 601 bvmcons = 0; 602 progname = basename(argv[0]); 603 gdb_port = 0; 604 guest_ncpus = 1; 605 memsize = 256 * MB; 606 607 while ((c = getopt(argc, argv, "abehwAHIPWp:g:c:s:m:l:")) != -1) { 608 switch (c) { 609 case 'a': 610 disable_x2apic = 1; 611 break; 612 case 'A': 613 acpi = 1; 614 break; 615 case 'b': 616 bvmcons = 1; 617 break; 618 case 'p': 619 pincpu = atoi(optarg); 620 break; 621 case 'c': 622 guest_ncpus = atoi(optarg); 623 break; 624 case 'g': 625 gdb_port = atoi(optarg); 626 break; 627 case 'l': 628 if (lpc_device_parse(optarg) != 0) { 629 errx(EX_USAGE, "invalid lpc device " 630 "configuration '%s'", optarg); 631 } 632 break; 633 case 's': 634 if (pci_parse_slot(optarg) != 0) 635 exit(1); 636 else 637 break; 638 case 'm': 639 error = vm_parse_memsize(optarg, &memsize); 640 if (error) 641 errx(EX_USAGE, "invalid memsize '%s'", optarg); 642 break; 643 case 'H': 644 guest_vmexit_on_hlt = 1; 645 break; 646 case 'I': 647 /* 648 * The "-I" option was used to add an ioapic to the 649 * virtual machine. 650 * 651 * An ioapic is now provided unconditionally for each 652 * virtual machine and this option is now deprecated. 653 */ 654 break; 655 case 'P': 656 guest_vmexit_on_pause = 1; 657 break; 658 case 'e': 659 strictio = 1; 660 break; 661 case 'w': 662 strictmsr = 0; 663 break; 664 case 'W': 665 virtio_msix = 0; 666 break; 667 case 'h': 668 usage(0); 669 default: 670 usage(1); 671 } 672 } 673 argc -= optind; 674 argv += optind; 675 676 if (argc != 1) 677 usage(1); 678 679 vmname = argv[0]; 680 681 ctx = vm_open(vmname); 682 if (ctx == NULL) { 683 perror("vm_open"); 684 exit(1); 685 } 686 687 max_vcpus = num_vcpus_allowed(ctx); 688 if (guest_ncpus > max_vcpus) { 689 fprintf(stderr, "%d vCPUs requested but only %d available\n", 690 guest_ncpus, max_vcpus); 691 exit(1); 692 } 693 694 fbsdrun_set_capabilities(ctx, BSP); 695 696 err = vm_setup_memory(ctx, memsize, VM_MMAP_ALL); 697 if (err) { 698 fprintf(stderr, "Unable to setup memory (%d)\n", err); 699 exit(1); 700 } 701 702 init_mem(); 703 init_inout(); 704 ioapic_init(ctx); 705 706 rtc_init(ctx); 707 708 /* 709 * Exit if a device emulation finds an error in it's initilization 710 */ 711 if (init_pci(ctx) != 0) 712 exit(1); 713 714 if (gdb_port != 0) 715 init_dbgport(gdb_port); 716 717 if (bvmcons) 718 init_bvmcons(); 719 720 error = vm_get_register(ctx, BSP, VM_REG_GUEST_RIP, &rip); 721 assert(error == 0); 722 723 /* 724 * build the guest tables, MP etc. 725 */ 726 mptable_build(ctx, guest_ncpus); 727 728 if (acpi) { 729 error = acpi_build(ctx, guest_ncpus); 730 assert(error == 0); 731 } 732 733 /* 734 * Change the proc title to include the VM name. 735 */ 736 setproctitle("%s", vmname); 737 738 /* 739 * Add CPU 0 740 */ 741 fbsdrun_addcpu(ctx, BSP, rip); 742 743 /* 744 * Head off to the main event dispatch loop 745 */ 746 mevent_dispatch(); 747 748 exit(1); 749} 750