qemu-kvm.c

   1 /*
   2  * qemu/kvm integration
   3  *
   4  * Copyright (C) 2006-2008 Qumranet Technologies
   5  *
   6  * Licensed under the terms of the GNU GPL version 2 or higher.
   7  */
   8 #include "config.h"
   9 #include "config-host.h"
  10
  11 #include <assert.h>
  12 #include <string.h>
  13 #include "hw/hw.h"
  14 #include "sysemu.h"
  15 #include "qemu-common.h"
  16 #include "console.h"
  17 #include "block.h"
  18 #include "compatfd.h"
  19 #include "gdbstub.h"
  20 #include "monitor.h"
  21
  22 #include "qemu-kvm.h"
  23 #include "libkvm.h"
  24
  25 #include <pthread.h>
  26 #include <sys/utsname.h>
  27 #include <sys/syscall.h>
  28 #include <sys/mman.h>
  29 #include <sys/ioctl.h>
  30 #include "compatfd.h"
  31 #include <sys/prctl.h>
  32
  33 #define false 0
  34 #define true 1
  35
  36 #ifndef PR_MCE_KILL
  37 #define PR_MCE_KILL 33
  38 #endif
  39
  40 #ifndef BUS_MCEERR_AR
  41 #define BUS_MCEERR_AR 4
  42 #endif
  43 #ifndef BUS_MCEERR_AO
  44 #define BUS_MCEERR_AO 5
  45 #endif
  46
  47 #define EXPECTED_KVM_API_VERSION 12
  48
  49 #if EXPECTED_KVM_API_VERSION != KVM_API_VERSION
  50 #error libkvm: userspace and kernel version mismatch
  51 #endif
  52
  53 int kvm_irqchip = 1;
  54 int kvm_pit = 1;
  55 int kvm_pit_reinject = 1;
  56 int kvm_nested = 0;
  57
  58
  59 KVMState *kvm_state;
  60 kvm_context_t kvm_context;
  61
  62 pthread_mutex_t qemu_mutex = PTHREAD_MUTEX_INITIALIZER;
  63 pthread_cond_t qemu_vcpu_cond = PTHREAD_COND_INITIALIZER;
  64 pthread_cond_t qemu_system_cond = PTHREAD_COND_INITIALIZER;
  65 pthread_cond_t qemu_pause_cond = PTHREAD_COND_INITIALIZER;
  66 pthread_cond_t qemu_work_cond = PTHREAD_COND_INITIALIZER;
  67 __thread CPUState *current_env;
  68
  69 static int qemu_system_ready;
  70
  71 #define SIG_IPI (SIGRTMIN+4)
  72
  73 pthread_t io_thread;
  74 static int io_thread_sigfd = -1;
  75
  76 static CPUState *kvm_debug_cpu_requested;
  77
  78 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
  79 /* The list of ioperm_data */
  80 static QLIST_HEAD(, ioperm_data) ioperm_head;
  81 #endif
  82
  83 #define ALIGN(x, y) (((x)+(y)-1) & ~((y)-1))
  84
  85 int kvm_abi = EXPECTED_KVM_API_VERSION;
  86 int kvm_page_size;
  87
  88 #ifdef KVM_CAP_SET_GUEST_DEBUG
  89 static int kvm_debug(CPUState *env,
  90                      struct kvm_debug_exit_arch *arch_info)
  91 {
  92     int handle = kvm_arch_debug(arch_info);
  93
  94     if (handle) {
  95         kvm_debug_cpu_requested = env;
  96         env->stopped = 1;
  97     }
  98     return handle;
  99 }
 100 #endif
 101
 102 static int handle_unhandled(uint64_t reason)
 103 {
 104     fprintf(stderr, "kvm: unhandled exit %" PRIx64 "\n", reason);
 105     return -EINVAL;
 106 }
 107
 108 #define VMX_INVALID_GUEST_STATE 0x80000021
 109
 110 static int handle_failed_vmentry(uint64_t reason)
 111 {
 112     fprintf(stderr, "kvm: vm entry failed with error 0x%" PRIx64 "\n\n", reason);
 113
 114     /* Perhaps we will need to check if this machine is intel since exit reason 0x21
 115        has a different interpretation on SVM */
 116     if (reason == VMX_INVALID_GUEST_STATE) {
 117         fprintf(stderr, "If you're runnning a guest on an Intel machine without\n");
 118         fprintf(stderr, "unrestricted mode support, the failure can be most likely\n");
 119         fprintf(stderr, "due to the guest entering an invalid state for Intel VT.\n");
 120         fprintf(stderr, "For example, the guest maybe running in big real mode\n");
 121         fprintf(stderr, "which is not supported on less recent Intel processors.\n\n");
 122     }
 123
 124     return -EINVAL;
 125 }
 126
 127 static inline void set_gsi(kvm_context_t kvm, unsigned int gsi)
 128 {
 129     uint32_t *bitmap = kvm->used_gsi_bitmap;
 130
 131     if (gsi < kvm->max_gsi)
 132         bitmap[gsi / 32] |= 1U << (gsi % 32);
 133     else
 134         DPRINTF("Invalid GSI %u\n", gsi);
 135 }
 136
 137 static inline void clear_gsi(kvm_context_t kvm, unsigned int gsi)
 138 {
 139     uint32_t *bitmap = kvm->used_gsi_bitmap;
 140
 141     if (gsi < kvm->max_gsi)
 142         bitmap[gsi / 32] &= ~(1U << (gsi % 32));
 143     else
 144         DPRINTF("Invalid GSI %u\n", gsi);
 145 }
 146
 147 static int kvm_create_context(void);
 148
 149 int kvm_init(int smp_cpus)
 150 {
 151     int fd;
 152     int r, gsi_count;
 153
 154
 155     fd = open("/dev/kvm", O_RDWR);
 156     if (fd == -1) {
 157         perror("open /dev/kvm");
 158         return -1;
 159     }
 160     r = ioctl(fd, KVM_GET_API_VERSION, 0);
 161     if (r == -1) {
 162         fprintf(stderr,
 163                 "kvm kernel version too old: "
 164                 "KVM_GET_API_VERSION ioctl not supported\n");
 165         goto out_close;
 166     }
 167     if (r < EXPECTED_KVM_API_VERSION) {
 168         fprintf(stderr, "kvm kernel version too old: "
 169                 "We expect API version %d or newer, but got "
 170                 "version %d\n", EXPECTED_KVM_API_VERSION, r);
 171         goto out_close;
 172     }
 173     if (r > EXPECTED_KVM_API_VERSION) {
 174         fprintf(stderr, "kvm userspace version too old\n");
 175         goto out_close;
 176     }
 177     kvm_abi = r;
 178     kvm_page_size = getpagesize();
 179     kvm_state = qemu_mallocz(sizeof(*kvm_state));
 180     kvm_context = &kvm_state->kvm_context;
 181
 182     kvm_state->fd = fd;
 183     kvm_state->vmfd = -1;
 184     kvm_context->opaque = cpu_single_env;
 185     kvm_context->dirty_pages_log_all = 0;
 186     kvm_context->no_irqchip_creation = 0;
 187     kvm_context->no_pit_creation = 0;
 188
 189 #ifdef KVM_CAP_SET_GUEST_DEBUG
 190     QTAILQ_INIT(&kvm_state->kvm_sw_breakpoints);
 191 #endif
 192
 193     gsi_count = kvm_get_gsi_count(kvm_context);
 194     if (gsi_count > 0) {
 195         int gsi_bits, i;
 196
 197         /* Round up so we can search ints using ffs */
 198         gsi_bits = ALIGN(gsi_count, 32);
 199         kvm_context->used_gsi_bitmap = qemu_mallocz(gsi_bits / 8);
 200         kvm_context->max_gsi = gsi_bits;
 201
 202         /* Mark any over-allocated bits as already in use */
 203         for (i = gsi_count; i < gsi_bits; i++) {
 204             set_gsi(kvm_context, i);
 205         }
 206     }
 207
 208     kvm_cpu_register_phys_memory_client();
 209
 210     pthread_mutex_lock(&qemu_mutex);
 211     return kvm_create_context();
 212
 213   out_close:
 214     close(fd);
 215     return -1;
 216 }
 217
 218 static void kvm_finalize(KVMState *s)
 219 {
 220     /* FIXME
 221        if (kvm->vcpu_fd[0] != -1)
 222            close(kvm->vcpu_fd[0]);
 223        if (kvm->vm_fd != -1)
 224            close(kvm->vm_fd);
 225      */
 226     close(s->fd);
 227     free(s);
 228 }
 229
 230 void kvm_disable_irqchip_creation(kvm_context_t kvm)
 231 {
 232     kvm->no_irqchip_creation = 1;
 233 }
 234
 235 void kvm_disable_pit_creation(kvm_context_t kvm)
 236 {
 237     kvm->no_pit_creation = 1;
 238 }
 239
 240 static void kvm_reset_vcpu(void *opaque)
 241 {
 242     CPUState *env = opaque;
 243
 244     kvm_arch_cpu_reset(env);
 245 }
 246
 247 static void kvm_create_vcpu(CPUState *env, int id)
 248 {
 249     long mmap_size;
 250     int r;
 251     KVMState *s = kvm_state;
 252
 253     r = kvm_vm_ioctl(kvm_state, KVM_CREATE_VCPU, id);
 254     if (r < 0) {
 255         fprintf(stderr, "kvm_create_vcpu: %m\n");
 256         fprintf(stderr, "Failed to create vCPU. Check the -smp parameter.\n");
 257         goto err;
 258     }
 259
 260     env->kvm_fd = r;
 261     env->kvm_state = kvm_state;
 262
 263     mmap_size = kvm_ioctl(kvm_state, KVM_GET_VCPU_MMAP_SIZE, 0);
 264     if (mmap_size < 0) {
 265         fprintf(stderr, "get vcpu mmap size: %m\n");
 266         goto err_fd;
 267     }
 268     env->kvm_run =
 269         mmap(NULL, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED, env->kvm_fd,
 270              0);
 271     if (env->kvm_run == MAP_FAILED) {
 272         fprintf(stderr, "mmap vcpu area: %m\n");
 273         goto err_fd;
 274     }
 275
 276 #ifdef KVM_CAP_COALESCED_MMIO
 277     if (s->coalesced_mmio && !s->coalesced_mmio_ring)
 278         s->coalesced_mmio_ring = (void *) env->kvm_run +
 279                s->coalesced_mmio * PAGE_SIZE;
 280 #endif
 281
 282     r = kvm_arch_init_vcpu(env);
 283     if (r == 0) {
 284         qemu_register_reset(kvm_reset_vcpu, env);
 285     }
 286
 287     return;
 288   err_fd:
 289     close(env->kvm_fd);
 290   err:
 291     /* We're no good with semi-broken states. */
 292     abort();
 293 }
 294
 295 static int kvm_set_boot_vcpu_id(kvm_context_t kvm, uint32_t id)
 296 {
 297 #ifdef KVM_CAP_SET_BOOT_CPU_ID
 298     int r = kvm_ioctl(kvm_state, KVM_CHECK_EXTENSION, KVM_CAP_SET_BOOT_CPU_ID);
 299     if (r > 0) {
 300         return kvm_vm_ioctl(kvm_state, KVM_SET_BOOT_CPU_ID, id);
 301     }
 302     return -ENOSYS;
 303 #else
 304     return -ENOSYS;
 305 #endif
 306 }
 307
 308 int kvm_create_vm(kvm_context_t kvm)
 309 {
 310     int fd;
 311 #ifdef KVM_CAP_IRQ_ROUTING
 312     kvm->irq_routes = qemu_mallocz(sizeof(*kvm->irq_routes));
 313     kvm->nr_allocated_irq_routes = 0;
 314 #endif
 315
 316     fd = kvm_ioctl(kvm_state, KVM_CREATE_VM, 0);
 317     if (fd < 0) {
 318         fprintf(stderr, "kvm_create_vm: %m\n");
 319         return -1;
 320     }
 321     kvm_state->vmfd = fd;
 322     return 0;
 323 }
 324
 325 static int kvm_create_default_phys_mem(kvm_context_t kvm,
 326                                        unsigned long phys_mem_bytes,
 327                                        void **vm_mem)
 328 {
 329 #ifdef KVM_CAP_USER_MEMORY
 330     int r = kvm_ioctl(kvm_state, KVM_CHECK_EXTENSION, KVM_CAP_USER_MEMORY);
 331     if (r > 0)
 332         return 0;
 333     fprintf(stderr,
 334             "Hypervisor too old: KVM_CAP_USER_MEMORY extension not supported\n");
 335 #else
 336 #error Hypervisor too old: KVM_CAP_USER_MEMORY extension not supported
 337 #endif
 338     return -1;
 339 }
 340
 341 void kvm_create_irqchip(kvm_context_t kvm)
 342 {
 343     int r;
 344
 345     kvm->irqchip_in_kernel = 0;
 346 #ifdef KVM_CAP_IRQCHIP
 347     if (!kvm->no_irqchip_creation) {
 348         r = kvm_ioctl(kvm_state, KVM_CHECK_EXTENSION, KVM_CAP_IRQCHIP);
 349         if (r > 0) {            /* kernel irqchip supported */
 350             r = kvm_vm_ioctl(kvm_state, KVM_CREATE_IRQCHIP);
 351             if (r >= 0) {
 352                 kvm->irqchip_inject_ioctl = KVM_IRQ_LINE;
 353 #if defined(KVM_CAP_IRQ_INJECT_STATUS) && defined(KVM_IRQ_LINE_STATUS)
 354                 r = kvm_ioctl(kvm_state, KVM_CHECK_EXTENSION,
 355                               KVM_CAP_IRQ_INJECT_STATUS);
 356                 if (r > 0) {
 357                     kvm->irqchip_inject_ioctl = KVM_IRQ_LINE_STATUS;
 358                 }
 359 #endif
 360                 kvm->irqchip_in_kernel = 1;
 361             } else
 362                 fprintf(stderr, "Create kernel PIC irqchip failed\n");
 363         }
 364     }
 365 #endif
 366     kvm_state->irqchip_in_kernel = kvm->irqchip_in_kernel;
 367 }
 368
 369 int kvm_create(kvm_context_t kvm, unsigned long phys_mem_bytes, void **vm_mem)
 370 {
 371     int r, i;
 372
 373     r = kvm_create_vm(kvm);
 374     if (r < 0) {
 375         return r;
 376     }
 377     r = kvm_arch_create(kvm, phys_mem_bytes, vm_mem);
 378     if (r < 0) {
 379         return r;
 380     }
 381     for (i = 0; i < ARRAY_SIZE(kvm_state->slots); i++) {
 382         kvm_state->slots[i].slot = i;
 383     }
 384
 385     r = kvm_create_default_phys_mem(kvm, phys_mem_bytes, vm_mem);
 386     if (r < 0) {
 387         return r;
 388     }
 389
 390     kvm_create_irqchip(kvm);
 391
 392     return 0;
 393 }
 394
 395 #ifdef KVM_CAP_IRQCHIP
 396
 397 int kvm_set_irq_level(kvm_context_t kvm, int irq, int level, int *status)
 398 {
 399     struct kvm_irq_level event;
 400     int r;
 401
 402     if (!kvm->irqchip_in_kernel) {
 403         return 0;
 404     }
 405     event.level = level;
 406     event.irq = irq;
 407     r = kvm_vm_ioctl(kvm_state, kvm->irqchip_inject_ioctl, &event);
 408     if (r < 0) {
 409         perror("kvm_set_irq_level");
 410     }
 411
 412     if (status) {
 413 #ifdef KVM_CAP_IRQ_INJECT_STATUS
 414         *status =
 415             (kvm->irqchip_inject_ioctl == KVM_IRQ_LINE) ? 1 : event.status;
 416 #else
 417         *status = 1;
 418 #endif
 419     }
 420
 421     return 1;
 422 }
 423
 424 int kvm_get_irqchip(kvm_context_t kvm, struct kvm_irqchip *chip)
 425 {
 426     int r;
 427
 428     if (!kvm->irqchip_in_kernel) {
 429         return 0;
 430     }
 431     r = kvm_vm_ioctl(kvm_state, KVM_GET_IRQCHIP, chip);
 432     if (r < 0) {
 433         perror("kvm_get_irqchip\n");
 434     }
 435     return r;
 436 }
 437
 438 int kvm_set_irqchip(kvm_context_t kvm, struct kvm_irqchip *chip)
 439 {
 440     int r;
 441
 442     if (!kvm->irqchip_in_kernel) {
 443         return 0;
 444     }
 445     r = kvm_vm_ioctl(kvm_state, KVM_SET_IRQCHIP, chip);
 446     if (r < 0) {
 447         perror("kvm_set_irqchip\n");
 448     }
 449     return r;
 450 }
 451
 452 #endif
 453
 454 static int handle_debug(CPUState *env)
 455 {
 456 #ifdef KVM_CAP_SET_GUEST_DEBUG
 457     struct kvm_run *run = env->kvm_run;
 458
 459     return kvm_debug(env, &run->debug.arch);
 460 #else
 461     return 0;
 462 #endif
 463 }
 464
 465 int kvm_get_regs(CPUState *env, struct kvm_regs *regs)
 466 {
 467     return kvm_vcpu_ioctl(env, KVM_GET_REGS, regs);
 468 }
 469
 470 int kvm_set_regs(CPUState *env, struct kvm_regs *regs)
 471 {
 472     return kvm_vcpu_ioctl(env, KVM_SET_REGS, regs);
 473 }
 474
 475 int kvm_get_fpu(CPUState *env, struct kvm_fpu *fpu)
 476 {
 477     return kvm_vcpu_ioctl(env, KVM_GET_FPU, fpu);
 478 }
 479
 480 int kvm_set_fpu(CPUState *env, struct kvm_fpu *fpu)
 481 {
 482     return kvm_vcpu_ioctl(env, KVM_SET_FPU, fpu);
 483 }
 484
 485 int kvm_get_sregs(CPUState *env, struct kvm_sregs *sregs)
 486 {
 487     return kvm_vcpu_ioctl(env, KVM_GET_SREGS, sregs);
 488 }
 489
 490 int kvm_set_sregs(CPUState *env, struct kvm_sregs *sregs)
 491 {
 492     return kvm_vcpu_ioctl(env, KVM_SET_SREGS, sregs);
 493 }
 494
 495 #ifdef KVM_CAP_MP_STATE
 496 int kvm_get_mpstate(CPUState *env, struct kvm_mp_state *mp_state)
 497 {
 498     int r;
 499
 500     r = kvm_ioctl(kvm_state, KVM_CHECK_EXTENSION, KVM_CAP_MP_STATE);
 501     if (r > 0) {
 502         return kvm_vcpu_ioctl(env, KVM_GET_MP_STATE, mp_state);
 503     }
 504     return -ENOSYS;
 505 }
 506
 507 int kvm_set_mpstate(CPUState *env, struct kvm_mp_state *mp_state)
 508 {
 509     int r;
 510
 511     r = kvm_ioctl(kvm_state, KVM_CHECK_EXTENSION, KVM_CAP_MP_STATE);
 512     if (r > 0) {
 513         return kvm_vcpu_ioctl(env, KVM_SET_MP_STATE, mp_state);
 514     }
 515     return -ENOSYS;
 516 }
 517 #endif
 518
 519 #ifdef KVM_CAP_XSAVE
 520 int kvm_get_xsave(CPUState *env, struct kvm_xsave *xsave)
 521 {
 522     return kvm_vcpu_ioctl(env, KVM_GET_XSAVE, xsave);
 523 }
 524
 525 int kvm_set_xsave(CPUState *env, struct kvm_xsave *xsave)
 526 {
 527     return kvm_vcpu_ioctl(env, KVM_SET_XSAVE, xsave);
 528 }
 529 #endif
 530
 531 #ifdef KVM_CAP_XCRS
 532 int kvm_get_xcrs(CPUState *env, struct kvm_xcrs *xcrs)
 533 {
 534     return kvm_vcpu_ioctl(env, KVM_GET_XCRS, xcrs);
 535 }
 536
 537 int kvm_set_xcrs(CPUState *env, struct kvm_xcrs *xcrs)
 538 {
 539     return kvm_vcpu_ioctl(env, KVM_SET_XCRS, xcrs);
 540 }
 541 #endif
 542
 543 static int handle_mmio(CPUState *env)
 544 {
 545     unsigned long addr = env->kvm_run->mmio.phys_addr;
 546     struct kvm_run *kvm_run = env->kvm_run;
 547     void *data = kvm_run->mmio.data;
 548
 549     /* hack: Red Hat 7.1 generates these weird accesses. */
 550     if ((addr > 0xa0000 - 4 && addr <= 0xa0000) && kvm_run->mmio.len == 3) {
 551         return 0;
 552     }
 553
 554     cpu_physical_memory_rw(addr, data, kvm_run->mmio.len, kvm_run->mmio.is_write);
 555     return 0;
 556 }
 557
 558 int handle_io_window(kvm_context_t kvm)
 559 {
 560     return 1;
 561 }
 562
 563 int handle_shutdown(kvm_context_t kvm, CPUState *env)
 564 {
 565     /* stop the current vcpu from going back to guest mode */
 566     env->stopped = 1;
 567
 568     qemu_system_reset_request();
 569     return 1;
 570 }
 571
 572 static inline void push_nmi(kvm_context_t kvm)
 573 {
 574 #ifdef KVM_CAP_USER_NMI
 575     kvm_arch_push_nmi(kvm->opaque);
 576 #endif                          /* KVM_CAP_USER_NMI */
 577 }
 578
 579 void post_kvm_run(kvm_context_t kvm, CPUState *env)
 580 {
 581     pthread_mutex_lock(&qemu_mutex);
 582     kvm_arch_post_run(env, env->kvm_run);
 583     cpu_single_env = env;
 584 }
 585
 586 int pre_kvm_run(kvm_context_t kvm, CPUState *env)
 587 {
 588     kvm_arch_pre_run(env, env->kvm_run);
 589
 590     pthread_mutex_unlock(&qemu_mutex);
 591     return 0;
 592 }
 593
 594 int kvm_is_ready_for_interrupt_injection(CPUState *env)
 595 {
 596     return env->kvm_run->ready_for_interrupt_injection;
 597 }
 598
 599 int kvm_run(CPUState *env)
 600 {
 601     int r;
 602     kvm_context_t kvm = &env->kvm_state->kvm_context;
 603     struct kvm_run *run = env->kvm_run;
 604     int fd = env->kvm_fd;
 605
 606   again:
 607     if (env->kvm_vcpu_dirty) {
 608         kvm_arch_load_regs(env, KVM_PUT_RUNTIME_STATE);
 609         env->kvm_vcpu_dirty = 0;
 610     }
 611     push_nmi(kvm);
 612 #if !defined(__s390__)
 613     if (!kvm->irqchip_in_kernel) {
 614         run->request_interrupt_window = kvm_arch_try_push_interrupts(env);
 615     }
 616 #endif
 617
 618     r = pre_kvm_run(kvm, env);
 619     if (r) {
 620         return r;
 621     }
 622     if (env->exit_request) {
 623         env->exit_request = 0;
 624         pthread_kill(env->kvm_cpu_state.thread, SIG_IPI);
 625     }
 626     r = ioctl(fd, KVM_RUN, 0);
 627
 628     if (r == -1 && errno != EINTR && errno != EAGAIN) {
 629         r = -errno;
 630         post_kvm_run(kvm, env);
 631         fprintf(stderr, "kvm_run: %s\n", strerror(-r));
 632         return r;
 633     }
 634
 635     post_kvm_run(kvm, env);
 636
 637     kvm_flush_coalesced_mmio_buffer();
 638
 639 #if !defined(__s390__)
 640     if (r == -1) {
 641         r = handle_io_window(kvm);
 642         goto more;
 643     }
 644 #endif
 645     if (1) {
 646         switch (run->exit_reason) {
 647         case KVM_EXIT_UNKNOWN:
 648             r = handle_unhandled(run->hw.hardware_exit_reason);
 649             break;
 650         case KVM_EXIT_FAIL_ENTRY:
 651             r = handle_failed_vmentry(run->fail_entry.hardware_entry_failure_reason);
 652             break;
 653         case KVM_EXIT_EXCEPTION:
 654             fprintf(stderr, "exception %d (%x)\n", run->ex.exception,
 655                     run->ex.error_code);
 656             kvm_show_regs(env);
 657             kvm_show_code(env);
 658             abort();
 659             break;
 660         case KVM_EXIT_IO:
 661             r = kvm_handle_io(run->io.port,
 662                                 (uint8_t *)run + run->io.data_offset,
 663                                 run->io.direction,
 664                                 run->io.size,
 665                                 run->io.count);
 666             r = 0;
 667             break;
 668         case KVM_EXIT_DEBUG:
 669             r = handle_debug(env);
 670             break;
 671         case KVM_EXIT_MMIO:
 672             r = handle_mmio(env);
 673             break;
 674         case KVM_EXIT_HLT:
 675             r = kvm_arch_halt(env);
 676             break;
 677         case KVM_EXIT_IRQ_WINDOW_OPEN:
 678             break;
 679         case KVM_EXIT_SHUTDOWN:
 680             r = handle_shutdown(kvm, env);
 681             break;
 682 #if defined(__s390__)
 683         case KVM_EXIT_S390_SIEIC:
 684             r = kvm_s390_handle_intercept(kvm, env, run);
 685             break;
 686         case KVM_EXIT_S390_RESET:
 687             r = kvm_s390_handle_reset(kvm, env, run);
 688             break;
 689 #endif
 690         case KVM_EXIT_INTERNAL_ERROR:
 691             kvm_handle_internal_error(env, run);
 692             r = 1;
 693             break;
 694         default:
 695             if (kvm_arch_run(env)) {
 696                 fprintf(stderr, "unhandled vm exit: 0x%x\n", run->exit_reason);
 697                 kvm_show_regs(env);
 698                 abort();
 699             }
 700             break;
 701         }
 702     }
 703 more:
 704     if (!r) {
 705         goto again;
 706     }
 707     return r;
 708 }
 709
 710 int kvm_inject_irq(CPUState *env, unsigned irq)
 711 {
 712     struct kvm_interrupt intr;
 713
 714     intr.irq = irq;
 715     return kvm_vcpu_ioctl(env, KVM_INTERRUPT, &intr);
 716 }
 717
 718 int kvm_inject_nmi(CPUState *env)
 719 {
 720 #ifdef KVM_CAP_USER_NMI
 721     return kvm_vcpu_ioctl(env, KVM_NMI);
 722 #else
 723     return -ENOSYS;
 724 #endif
 725 }
 726
 727 int kvm_init_coalesced_mmio(kvm_context_t kvm)
 728 {
 729     int r = 0;
 730     kvm_state->coalesced_mmio = 0;
 731 #ifdef KVM_CAP_COALESCED_MMIO
 732     r = kvm_ioctl(kvm_state, KVM_CHECK_EXTENSION, KVM_CAP_COALESCED_MMIO);
 733     if (r > 0) {
 734         kvm_state->coalesced_mmio = r;
 735         return 0;
 736     }
 737 #endif
 738     return r;
 739 }
 740
 741 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
 742 int kvm_assign_pci_device(kvm_context_t kvm,
 743                           struct kvm_assigned_pci_dev *assigned_dev)
 744 {
 745     return kvm_vm_ioctl(kvm_state, KVM_ASSIGN_PCI_DEVICE, assigned_dev);
 746 }
 747
 748 static int kvm_old_assign_irq(kvm_context_t kvm,
 749                               struct kvm_assigned_irq *assigned_irq)
 750 {
 751     return kvm_vm_ioctl(kvm_state, KVM_ASSIGN_IRQ, assigned_irq);
 752 }
 753
 754 #ifdef KVM_CAP_ASSIGN_DEV_IRQ
 755 int kvm_assign_irq(kvm_context_t kvm, struct kvm_assigned_irq *assigned_irq)
 756 {
 757     int ret;
 758
 759     ret = kvm_ioctl(kvm_state, KVM_CHECK_EXTENSION, KVM_CAP_ASSIGN_DEV_IRQ);
 760     if (ret > 0) {
 761         return kvm_vm_ioctl(kvm_state, KVM_ASSIGN_DEV_IRQ, assigned_irq);
 762     }
 763
 764     return kvm_old_assign_irq(kvm, assigned_irq);
 765 }
 766
 767 int kvm_deassign_irq(kvm_context_t kvm, struct kvm_assigned_irq *assigned_irq)
 768 {
 769     return kvm_vm_ioctl(kvm_state, KVM_DEASSIGN_DEV_IRQ, assigned_irq);
 770 }
 771 #else
 772 int kvm_assign_irq(kvm_context_t kvm, struct kvm_assigned_irq *assigned_irq)
 773 {
 774     return kvm_old_assign_irq(kvm, assigned_irq);
 775 }
 776 #endif
 777 #endif
 778
 779 #ifdef KVM_CAP_DEVICE_DEASSIGNMENT
 780 int kvm_deassign_pci_device(kvm_context_t kvm,
 781                             struct kvm_assigned_pci_dev *assigned_dev)
 782 {
 783     return kvm_vm_ioctl(kvm_state, KVM_DEASSIGN_PCI_DEVICE, assigned_dev);
 784 }
 785 #endif
 786
 787 int kvm_reinject_control(kvm_context_t kvm, int pit_reinject)
 788 {
 789 #ifdef KVM_CAP_REINJECT_CONTROL
 790     int r;
 791     struct kvm_reinject_control control;
 792
 793     control.pit_reinject = pit_reinject;
 794
 795     r = kvm_ioctl(kvm_state, KVM_CHECK_EXTENSION, KVM_CAP_REINJECT_CONTROL);
 796     if (r > 0) {
 797         return kvm_vm_ioctl(kvm_state, KVM_REINJECT_CONTROL, &control);
 798     }
 799 #endif
 800     return -ENOSYS;
 801 }
 802
 803 int kvm_has_gsi_routing(kvm_context_t kvm)
 804 {
 805     int r = 0;
 806
 807 #ifdef KVM_CAP_IRQ_ROUTING
 808     r = kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
 809 #endif
 810     return r;
 811 }
 812
 813 int kvm_get_gsi_count(kvm_context_t kvm)
 814 {
 815 #ifdef KVM_CAP_IRQ_ROUTING
 816     return kvm_check_extension(kvm_state, KVM_CAP_IRQ_ROUTING);
 817 #else
 818     return -EINVAL;
 819 #endif
 820 }
 821
 822 int kvm_clear_gsi_routes(kvm_context_t kvm)
 823 {
 824 #ifdef KVM_CAP_IRQ_ROUTING
 825     kvm->irq_routes->nr = 0;
 826     return 0;
 827 #else
 828     return -EINVAL;
 829 #endif
 830 }
 831
 832 int kvm_add_routing_entry(kvm_context_t kvm,
 833                           struct kvm_irq_routing_entry *entry)
 834 {
 835 #ifdef KVM_CAP_IRQ_ROUTING
 836     struct kvm_irq_routing *z;
 837     struct kvm_irq_routing_entry *new;
 838     int n, size;
 839
 840     if (kvm->irq_routes->nr == kvm->nr_allocated_irq_routes) {
 841         n = kvm->nr_allocated_irq_routes * 2;
 842         if (n < 64) {
 843             n = 64;
 844         }
 845         size = sizeof(struct kvm_irq_routing);
 846         size += n * sizeof(*new);
 847         z = realloc(kvm->irq_routes, size);
 848         if (!z) {
 849             return -ENOMEM;
 850         }
 851         kvm->nr_allocated_irq_routes = n;
 852         kvm->irq_routes = z;
 853     }
 854     n = kvm->irq_routes->nr++;
 855     new = &kvm->irq_routes->entries[n];
 856     memset(new, 0, sizeof(*new));
 857     new->gsi = entry->gsi;
 858     new->type = entry->type;
 859     new->flags = entry->flags;
 860     new->u = entry->u;
 861
 862     set_gsi(kvm, entry->gsi);
 863
 864     return 0;
 865 #else
 866     return -ENOSYS;
 867 #endif
 868 }
 869
 870 int kvm_add_irq_route(kvm_context_t kvm, int gsi, int irqchip, int pin)
 871 {
 872 #ifdef KVM_CAP_IRQ_ROUTING
 873     struct kvm_irq_routing_entry e;
 874
 875     e.gsi = gsi;
 876     e.type = KVM_IRQ_ROUTING_IRQCHIP;
 877     e.flags = 0;
 878     e.u.irqchip.irqchip = irqchip;
 879     e.u.irqchip.pin = pin;
 880     return kvm_add_routing_entry(kvm, &e);
 881 #else
 882     return -ENOSYS;
 883 #endif
 884 }
 885
 886 int kvm_del_routing_entry(kvm_context_t kvm,
 887                           struct kvm_irq_routing_entry *entry)
 888 {
 889 #ifdef KVM_CAP_IRQ_ROUTING
 890     struct kvm_irq_routing_entry *e, *p;
 891     int i, gsi, found = 0;
 892
 893     gsi = entry->gsi;
 894
 895     for (i = 0; i < kvm->irq_routes->nr; ++i) {
 896         e = &kvm->irq_routes->entries[i];
 897         if (e->type == entry->type && e->gsi == gsi) {
 898             switch (e->type) {
 899             case KVM_IRQ_ROUTING_IRQCHIP:{
 900                     if (e->u.irqchip.irqchip ==
 901                         entry->u.irqchip.irqchip
 902                         && e->u.irqchip.pin == entry->u.irqchip.pin) {
 903                         p = &kvm->irq_routes->entries[--kvm->irq_routes->nr];
 904                         *e = *p;
 905                         found = 1;
 906                     }
 907                     break;
 908                 }
 909             case KVM_IRQ_ROUTING_MSI:{
 910                     if (e->u.msi.address_lo ==
 911                         entry->u.msi.address_lo
 912                         && e->u.msi.address_hi ==
 913                         entry->u.msi.address_hi
 914                         && e->u.msi.data == entry->u.msi.data) {
 915                         p = &kvm->irq_routes->entries[--kvm->irq_routes->nr];
 916                         *e = *p;
 917                         found = 1;
 918                     }
 919                     break;
 920                 }
 921             default:
 922                 break;
 923             }
 924             if (found) {
 925                 /* If there are no other users of this GSI
 926                  * mark it available in the bitmap */
 927                 for (i = 0; i < kvm->irq_routes->nr; i++) {
 928                     e = &kvm->irq_routes->entries[i];
 929                     if (e->gsi == gsi)
 930                         break;
 931                 }
 932                 if (i == kvm->irq_routes->nr) {
 933                     clear_gsi(kvm, gsi);
 934                 }
 935
 936                 return 0;
 937             }
 938         }
 939     }
 940     return -ESRCH;
 941 #else
 942     return -ENOSYS;
 943 #endif
 944 }
 945
 946 int kvm_update_routing_entry(kvm_context_t kvm,
 947                              struct kvm_irq_routing_entry *entry,
 948                              struct kvm_irq_routing_entry *newentry)
 949 {
 950 #ifdef KVM_CAP_IRQ_ROUTING
 951     struct kvm_irq_routing_entry *e;
 952     int i;
 953
 954     if (entry->gsi != newentry->gsi || entry->type != newentry->type) {
 955         return -EINVAL;
 956     }
 957
 958     for (i = 0; i < kvm->irq_routes->nr; ++i) {
 959         e = &kvm->irq_routes->entries[i];
 960         if (e->type != entry->type || e->gsi != entry->gsi) {
 961             continue;
 962         }
 963         switch (e->type) {
 964         case KVM_IRQ_ROUTING_IRQCHIP:
 965             if (e->u.irqchip.irqchip == entry->u.irqchip.irqchip &&
 966                 e->u.irqchip.pin == entry->u.irqchip.pin) {
 967                 memcpy(&e->u.irqchip, &newentry->u.irqchip,
 968                        sizeof e->u.irqchip);
 969                 return 0;
 970             }
 971             break;
 972         case KVM_IRQ_ROUTING_MSI:
 973             if (e->u.msi.address_lo == entry->u.msi.address_lo &&
 974                 e->u.msi.address_hi == entry->u.msi.address_hi &&
 975                 e->u.msi.data == entry->u.msi.data) {
 976                 memcpy(&e->u.msi, &newentry->u.msi, sizeof e->u.msi);
 977                 return 0;
 978             }
 979             break;
 980         default:
 981             break;
 982         }
 983     }
 984     return -ESRCH;
 985 #else
 986     return -ENOSYS;
 987 #endif
 988 }
 989
 990 int kvm_del_irq_route(kvm_context_t kvm, int gsi, int irqchip, int pin)
 991 {
 992 #ifdef KVM_CAP_IRQ_ROUTING
 993     struct kvm_irq_routing_entry e;
 994
 995     e.gsi = gsi;
 996     e.type = KVM_IRQ_ROUTING_IRQCHIP;
 997     e.flags = 0;
 998     e.u.irqchip.irqchip = irqchip;
 999     e.u.irqchip.pin = pin;
1000     return kvm_del_routing_entry(kvm, &e);
1001 #else
1002     return -ENOSYS;
1003 #endif
1004 }
1005
1006 int kvm_commit_irq_routes(kvm_context_t kvm)
1007 {
1008 #ifdef KVM_CAP_IRQ_ROUTING
1009     kvm->irq_routes->flags = 0;
1010     return kvm_vm_ioctl(kvm_state, KVM_SET_GSI_ROUTING, kvm->irq_routes);
1011 #else
1012     return -ENOSYS;
1013 #endif
1014 }
1015
1016 int kvm_get_irq_route_gsi(kvm_context_t kvm)
1017 {
1018     int i, bit;
1019     uint32_t *buf = kvm->used_gsi_bitmap;
1020
1021     /* Return the lowest unused GSI in the bitmap */
1022     for (i = 0; i < kvm->max_gsi / 32; i++) {
1023         bit = ffs(~buf[i]);
1024         if (!bit) {
1025             continue;
1026         }
1027
1028         return bit - 1 + i * 32;
1029     }
1030
1031     return -ENOSPC;
1032 }
1033
1034 #ifdef KVM_CAP_DEVICE_MSIX
1035 int kvm_assign_set_msix_nr(kvm_context_t kvm,
1036                            struct kvm_assigned_msix_nr *msix_nr)
1037 {
1038     return kvm_vm_ioctl(kvm_state, KVM_ASSIGN_SET_MSIX_NR, msix_nr);
1039 }
1040
1041 int kvm_assign_set_msix_entry(kvm_context_t kvm,
1042                               struct kvm_assigned_msix_entry *entry)
1043 {
1044     return kvm_vm_ioctl(kvm_state, KVM_ASSIGN_SET_MSIX_ENTRY, entry);
1045 }
1046 #endif
1047
1048 #if defined(KVM_CAP_IRQFD) && defined(CONFIG_EVENTFD)
1049
1050 #include <sys/eventfd.h>
1051
1052 static int _kvm_irqfd(kvm_context_t kvm, int fd, int gsi, int flags)
1053 {
1054     struct kvm_irqfd data = {
1055         .fd = fd,
1056         .gsi = gsi,
1057         .flags = flags,
1058     };
1059
1060     return kvm_vm_ioctl(kvm_state, KVM_IRQFD, &data);
1061 }
1062
1063 int kvm_irqfd(kvm_context_t kvm, int gsi, int flags)
1064 {
1065     int r;
1066     int fd;
1067
1068     if (!kvm_check_extension(kvm_state, KVM_CAP_IRQFD))
1069         return -ENOENT;
1070
1071     fd = eventfd(0, 0);
1072     if (fd < 0) {
1073         return -errno;
1074     }
1075
1076     r = _kvm_irqfd(kvm, fd, gsi, 0);
1077     if (r < 0) {
1078         close(fd);
1079         return -errno;
1080     }
1081
1082     return fd;
1083 }
1084
1085 #else                           /* KVM_CAP_IRQFD */
1086
1087 int kvm_irqfd(kvm_context_t kvm, int gsi, int flags)
1088 {
1089     return -ENOSYS;
1090 }
1091
1092 #endif                          /* KVM_CAP_IRQFD */
1093 unsigned long kvm_get_thread_id(void)
1094 {
1095     return syscall(SYS_gettid);
1096 }
1097
1098 static void qemu_cond_wait(pthread_cond_t *cond)
1099 {
1100     CPUState *env = cpu_single_env;
1101
1102     pthread_cond_wait(cond, &qemu_mutex);
1103     cpu_single_env = env;
1104 }
1105
1106 static void sig_ipi_handler(int n)
1107 {
1108 }
1109
1110 static void hardware_memory_error(void)
1111 {
1112     fprintf(stderr, "Hardware memory error!\n");
1113     exit(1);
1114 }
1115
1116 static void sigbus_reraise(void)
1117 {
1118     sigset_t set;
1119     struct sigaction action;
1120
1121     memset(&action, 0, sizeof(action));
1122     action.sa_handler = SIG_DFL;
1123     if (!sigaction(SIGBUS, &action, NULL)) {
1124         raise(SIGBUS);
1125         sigemptyset(&set);
1126         sigaddset(&set, SIGBUS);
1127         sigprocmask(SIG_UNBLOCK, &set, NULL);
1128     }
1129     perror("Failed to re-raise SIGBUS!\n");
1130     abort();
1131 }
1132
1133 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
1134                            void *ctx)
1135 {
1136 #if defined(KVM_CAP_MCE) && defined(TARGET_I386)
1137     if (first_cpu->mcg_cap && siginfo->ssi_addr
1138         && siginfo->ssi_code == BUS_MCEERR_AO) {
1139         uint64_t status;
1140         unsigned long paddr;
1141         CPUState *cenv;
1142
1143         /* Hope we are lucky for AO MCE */
1144         if (do_qemu_ram_addr_from_host((void *)(intptr_t)siginfo->ssi_addr,
1145                                        &paddr)) {
1146             fprintf(stderr, "Hardware memory error for memory used by "
1147                     "QEMU itself instead of guest system!: %llx\n",
1148                     (unsigned long long)siginfo->ssi_addr);
1149             return;
1150         }
1151         status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN
1152             | MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S
1153             | 0xc0;
1154         kvm_inject_x86_mce(first_cpu, 9, status,
1155                            MCG_STATUS_MCIP | MCG_STATUS_RIPV, paddr,
1156                            (MCM_ADDR_PHYS << 6) | 0xc, 1);
1157         for (cenv = first_cpu->next_cpu; cenv != NULL; cenv = cenv->next_cpu) {
1158             kvm_inject_x86_mce(cenv, 1, MCI_STATUS_VAL | MCI_STATUS_UC,
1159                                MCG_STATUS_MCIP | MCG_STATUS_RIPV, 0, 0, 1);
1160         }
1161     } else
1162 #endif
1163     {
1164         if (siginfo->ssi_code == BUS_MCEERR_AO) {
1165             return;
1166         } else if (siginfo->ssi_code == BUS_MCEERR_AR) {
1167             hardware_memory_error();
1168         } else {
1169             sigbus_reraise();
1170         }
1171     }
1172 }
1173
1174 static void on_vcpu(CPUState *env, void (*func)(void *data), void *data)
1175 {
1176     struct qemu_work_item wi;
1177
1178     if (env == current_env) {
1179         func(data);
1180         return;
1181     }
1182
1183     wi.func = func;
1184     wi.data = data;
1185     if (!env->kvm_cpu_state.queued_work_first) {
1186         env->kvm_cpu_state.queued_work_first = &wi;
1187     } else {
1188         env->kvm_cpu_state.queued_work_last->next = &wi;
1189     }
1190     env->kvm_cpu_state.queued_work_last = &wi;
1191     wi.next = NULL;
1192     wi.done = false;
1193
1194     pthread_kill(env->kvm_cpu_state.thread, SIG_IPI);
1195     while (!wi.done) {
1196         qemu_cond_wait(&qemu_work_cond);
1197     }
1198 }
1199
1200 static void do_kvm_cpu_synchronize_state(void *_env)
1201 {
1202     CPUState *env = _env;
1203
1204     if (!env->kvm_vcpu_dirty) {
1205         kvm_arch_save_regs(env);
1206         env->kvm_vcpu_dirty = 1;
1207     }
1208 }
1209
1210 void kvm_cpu_synchronize_state(CPUState *env)
1211 {
1212     if (!env->kvm_vcpu_dirty) {
1213         on_vcpu(env, do_kvm_cpu_synchronize_state, env);
1214     }
1215 }
1216
1217 void kvm_cpu_synchronize_post_reset(CPUState *env)
1218 {
1219     kvm_arch_load_regs(env, KVM_PUT_RESET_STATE);
1220     env->kvm_vcpu_dirty = 0;
1221 }
1222
1223 void kvm_cpu_synchronize_post_init(CPUState *env)
1224 {
1225     kvm_arch_load_regs(env, KVM_PUT_FULL_STATE);
1226     env->kvm_vcpu_dirty = 0;
1227 }
1228
1229 static void inject_interrupt(void *data)
1230 {
1231     cpu_interrupt(current_env, (long) data);
1232 }
1233
1234 void kvm_inject_interrupt(CPUState *env, int mask)
1235 {
1236     on_vcpu(env, inject_interrupt, (void *) (long) mask);
1237 }
1238
1239 void kvm_update_interrupt_request(CPUState *env)
1240 {
1241     int signal = 0;
1242
1243     if (env) {
1244         if (!current_env || !current_env->created) {
1245             signal = 1;
1246         }
1247         /*
1248          * Testing for created here is really redundant
1249          */
1250         if (current_env && current_env->created &&
1251             env != current_env && !env->kvm_cpu_state.signalled) {
1252             signal = 1;
1253         }
1254
1255         if (signal) {
1256             env->kvm_cpu_state.signalled = 1;
1257             if (env->kvm_cpu_state.thread) {
1258                 pthread_kill(env->kvm_cpu_state.thread, SIG_IPI);
1259             }
1260         }
1261     }
1262 }
1263
1264 int kvm_cpu_exec(CPUState *env)
1265 {
1266     int r;
1267
1268     r = kvm_run(env);
1269     if (r < 0) {
1270         printf("kvm_run returned %d\n", r);
1271         vm_stop(0);
1272     }
1273
1274     return 0;
1275 }
1276
1277 int kvm_cpu_is_stopped(CPUState *env)
1278 {
1279     return !vm_running || env->stopped;
1280 }
1281
1282 static void flush_queued_work(CPUState *env)
1283 {
1284     struct qemu_work_item *wi;
1285
1286     if (!env->kvm_cpu_state.queued_work_first) {
1287         return;
1288     }
1289
1290     while ((wi = env->kvm_cpu_state.queued_work_first)) {
1291         env->kvm_cpu_state.queued_work_first = wi->next;
1292         wi->func(wi->data);
1293         wi->done = true;
1294     }
1295     env->kvm_cpu_state.queued_work_last = NULL;
1296     pthread_cond_broadcast(&qemu_work_cond);
1297 }
1298
1299 static int kvm_mce_in_exception(CPUState *env)
1300 {
1301     struct kvm_msr_entry msr_mcg_status = {
1302         .index = MSR_MCG_STATUS,
1303     };
1304     int r;
1305
1306     r = kvm_get_msrs(env, &msr_mcg_status, 1);
1307     if (r == -1 || r == 0) {
1308         return -1;
1309     }
1310     return !!(msr_mcg_status.data & MCG_STATUS_MCIP);
1311 }
1312
1313 static void kvm_on_sigbus(CPUState *env, siginfo_t *siginfo)
1314 {
1315 #if defined(KVM_CAP_MCE) && defined(TARGET_I386)
1316     struct kvm_x86_mce mce = {
1317             .bank = 9,
1318     };
1319     unsigned long paddr;
1320     int r;
1321
1322     if (env->mcg_cap && siginfo->si_addr
1323         && (siginfo->si_code == BUS_MCEERR_AR
1324             || siginfo->si_code == BUS_MCEERR_AO)) {
1325         if (siginfo->si_code == BUS_MCEERR_AR) {
1326             /* Fake an Intel architectural Data Load SRAR UCR */
1327             mce.status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN
1328                 | MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S
1329                 | MCI_STATUS_AR | 0x134;
1330             mce.misc = (MCM_ADDR_PHYS << 6) | 0xc;
1331             mce.mcg_status = MCG_STATUS_MCIP | MCG_STATUS_EIPV;
1332         } else {
1333             /*
1334              * If there is an MCE excpetion being processed, ignore
1335              * this SRAO MCE
1336              */
1337             r = kvm_mce_in_exception(env);
1338             if (r == -1) {
1339                 fprintf(stderr, "Failed to get MCE status\n");
1340             } else if (r) {
1341                 return;
1342             }
1343             /* Fake an Intel architectural Memory scrubbing UCR */
1344             mce.status = MCI_STATUS_VAL | MCI_STATUS_UC | MCI_STATUS_EN
1345                 | MCI_STATUS_MISCV | MCI_STATUS_ADDRV | MCI_STATUS_S
1346                 | 0xc0;
1347             mce.misc = (MCM_ADDR_PHYS << 6) | 0xc;
1348             mce.mcg_status = MCG_STATUS_MCIP | MCG_STATUS_RIPV;
1349         }
1350         if (do_qemu_ram_addr_from_host((void *)siginfo->si_addr, &paddr)) {
1351             fprintf(stderr, "Hardware memory error for memory used by "
1352                     "QEMU itself instaed of guest system!\n");
1353             /* Hope we are lucky for AO MCE */
1354             if (siginfo->si_code == BUS_MCEERR_AO) {
1355                 return;
1356             } else {
1357                 hardware_memory_error();
1358             }
1359         }
1360         mce.addr = paddr;
1361         r = kvm_set_mce(env, &mce);
1362         if (r < 0) {
1363             fprintf(stderr, "kvm_set_mce: %s\n", strerror(errno));
1364             abort();
1365         }
1366     } else
1367 #endif
1368     {
1369         if (siginfo->si_code == BUS_MCEERR_AO) {
1370             return;
1371         } else if (siginfo->si_code == BUS_MCEERR_AR) {
1372             hardware_memory_error();
1373         } else {
1374             sigbus_reraise();
1375         }
1376     }
1377 }
1378
1379 static void kvm_main_loop_wait(CPUState *env, int timeout)
1380 {
1381     struct timespec ts;
1382     int r, e;
1383     siginfo_t siginfo;
1384     sigset_t waitset;
1385     sigset_t chkset;
1386
1387     ts.tv_sec = timeout / 1000;
1388     ts.tv_nsec = (timeout % 1000) * 1000000;
1389     sigemptyset(&waitset);
1390     sigaddset(&waitset, SIG_IPI);
1391     sigaddset(&waitset, SIGBUS);
1392
1393     do {
1394         pthread_mutex_unlock(&qemu_mutex);
1395
1396         r = sigtimedwait(&waitset, &siginfo, &ts);
1397         e = errno;
1398
1399         pthread_mutex_lock(&qemu_mutex);
1400
1401         if (r == -1 && !(e == EAGAIN || e == EINTR)) {
1402             printf("sigtimedwait: %s\n", strerror(e));
1403             exit(1);
1404         }
1405
1406         switch (r) {
1407         case SIGBUS:
1408             kvm_on_sigbus(env, &siginfo);
1409             break;
1410         default:
1411             break;
1412         }
1413
1414         r = sigpending(&chkset);
1415         if (r == -1) {
1416             printf("sigpending: %s\n", strerror(e));
1417             exit(1);
1418         }
1419     } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
1420
1421     cpu_single_env = env;
1422     flush_queued_work(env);
1423
1424     if (env->stop) {
1425         env->stop = 0;
1426         env->stopped = 1;
1427         pthread_cond_signal(&qemu_pause_cond);
1428     }
1429
1430     env->kvm_cpu_state.signalled = 0;
1431 }
1432
1433 static int all_threads_paused(void)
1434 {
1435     CPUState *penv = first_cpu;
1436
1437     while (penv) {
1438         if (penv->stop) {
1439             return 0;
1440         }
1441         penv = (CPUState *) penv->next_cpu;
1442     }
1443
1444     return 1;
1445 }
1446
1447 static void pause_all_threads(void)
1448 {
1449     CPUState *penv = first_cpu;
1450
1451     while (penv) {
1452         if (penv != cpu_single_env) {
1453             penv->stop = 1;
1454             pthread_kill(penv->kvm_cpu_state.thread, SIG_IPI);
1455         } else {
1456             penv->stop = 0;
1457             penv->stopped = 1;
1458             cpu_exit(penv);
1459         }
1460         penv = (CPUState *) penv->next_cpu;
1461     }
1462
1463     while (!all_threads_paused()) {
1464         qemu_cond_wait(&qemu_pause_cond);
1465     }
1466 }
1467
1468 static void resume_all_threads(void)
1469 {
1470     CPUState *penv = first_cpu;
1471
1472     assert(!cpu_single_env);
1473
1474     while (penv) {
1475         penv->stop = 0;
1476         penv->stopped = 0;
1477         pthread_kill(penv->kvm_cpu_state.thread, SIG_IPI);
1478         penv = (CPUState *) penv->next_cpu;
1479     }
1480 }
1481
1482 static void kvm_vm_state_change_handler(void *context, int running, int reason)
1483 {
1484     if (running) {
1485         resume_all_threads();
1486     } else {
1487         pause_all_threads();
1488     }
1489 }
1490
1491 static void setup_kernel_sigmask(CPUState *env)
1492 {
1493     sigset_t set;
1494
1495     sigemptyset(&set);
1496     sigaddset(&set, SIGUSR2);
1497     sigaddset(&set, SIGIO);
1498     sigaddset(&set, SIGALRM);
1499     sigprocmask(SIG_BLOCK, &set, NULL);
1500
1501     sigprocmask(SIG_BLOCK, NULL, &set);
1502     sigdelset(&set, SIG_IPI);
1503     sigdelset(&set, SIGBUS);
1504
1505     kvm_set_signal_mask(env, &set);
1506 }
1507
1508 static void qemu_kvm_system_reset(void)
1509 {
1510     pause_all_threads();
1511
1512     qemu_system_reset();
1513
1514     resume_all_threads();
1515 }
1516
1517 static void process_irqchip_events(CPUState *env)
1518 {
1519     kvm_arch_process_irqchip_events(env);
1520     if (kvm_arch_has_work(env))
1521         env->halted = 0;
1522 }
1523
1524 static int kvm_main_loop_cpu(CPUState *env)
1525 {
1526     while (1) {
1527         int run_cpu = !kvm_cpu_is_stopped(env);
1528         if (run_cpu && !kvm_irqchip_in_kernel()) {
1529             process_irqchip_events(env);
1530             run_cpu = !env->halted;
1531         }
1532         if (run_cpu) {
1533             kvm_cpu_exec(env);
1534             kvm_main_loop_wait(env, 0);
1535         } else {
1536             kvm_main_loop_wait(env, 1000);
1537         }
1538     }
1539     pthread_mutex_unlock(&qemu_mutex);
1540     return 0;
1541 }
1542
1543 static void *ap_main_loop(void *_env)
1544 {
1545     CPUState *env = _env;
1546     sigset_t signals;
1547 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
1548     struct ioperm_data *data = NULL;
1549 #endif
1550
1551     current_env = env;
1552     env->thread_id = kvm_get_thread_id();
1553     sigfillset(&signals);
1554     sigprocmask(SIG_BLOCK, &signals, NULL);
1555
1556 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
1557     /* do ioperm for io ports of assigned devices */
1558     QLIST_FOREACH(data, &ioperm_head, entries)
1559         on_vcpu(env, kvm_arch_do_ioperm, data);
1560 #endif
1561
1562     pthread_mutex_lock(&qemu_mutex);
1563     cpu_single_env = env;
1564
1565     kvm_create_vcpu(env, env->cpu_index);
1566     setup_kernel_sigmask(env);
1567
1568     /* signal VCPU creation */
1569     current_env->created = 1;
1570     pthread_cond_signal(&qemu_vcpu_cond);
1571
1572     /* and wait for machine initialization */
1573     while (!qemu_system_ready) {
1574         qemu_cond_wait(&qemu_system_cond);
1575     }
1576
1577     /* re-initialize cpu_single_env after re-acquiring qemu_mutex */
1578     cpu_single_env = env;
1579
1580     kvm_main_loop_cpu(env);
1581     return NULL;
1582 }
1583
1584 int kvm_init_vcpu(CPUState *env)
1585 {
1586     pthread_create(&env->kvm_cpu_state.thread, NULL, ap_main_loop, env);
1587
1588     while (env->created == 0) {
1589         qemu_cond_wait(&qemu_vcpu_cond);
1590     }
1591
1592     return 0;
1593 }
1594
1595 int kvm_vcpu_inited(CPUState *env)
1596 {
1597     return env->created;
1598 }
1599
1600 #ifdef TARGET_I386
1601 void kvm_hpet_disable_kpit(void)
1602 {
1603     struct kvm_pit_state2 ps2;
1604
1605     kvm_get_pit2(kvm_context, &ps2);
1606     ps2.flags |= KVM_PIT_FLAGS_HPET_LEGACY;
1607     kvm_set_pit2(kvm_context, &ps2);
1608 }
1609
1610 void kvm_hpet_enable_kpit(void)
1611 {
1612     struct kvm_pit_state2 ps2;
1613
1614     kvm_get_pit2(kvm_context, &ps2);
1615     ps2.flags &= ~KVM_PIT_FLAGS_HPET_LEGACY;
1616     kvm_set_pit2(kvm_context, &ps2);
1617 }
1618 #endif
1619
1620 int kvm_init_ap(void)
1621 {
1622     struct sigaction action;
1623
1624     qemu_add_vm_change_state_handler(kvm_vm_state_change_handler, NULL);
1625
1626     signal(SIG_IPI, sig_ipi_handler);
1627
1628     memset(&action, 0, sizeof(action));
1629     action.sa_flags = SA_SIGINFO;
1630     action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
1631     sigaction(SIGBUS, &action, NULL);
1632     prctl(PR_MCE_KILL, 1, 1, 0, 0);
1633     return 0;
1634 }
1635
1636 /* If we have signalfd, we mask out the signals we want to handle and then
1637  * use signalfd to listen for them.  We rely on whatever the current signal
1638  * handler is to dispatch the signals when we receive them.
1639  */
1640
1641 static void sigfd_handler(void *opaque)
1642 {
1643     int fd = (unsigned long) opaque;
1644     struct qemu_signalfd_siginfo info;
1645     struct sigaction action;
1646     ssize_t len;
1647
1648     while (1) {
1649         do {
1650             len = read(fd, &info, sizeof(info));
1651         } while (len == -1 && errno == EINTR);
1652
1653         if (len == -1 && errno == EAGAIN) {
1654             break;
1655         }
1656
1657         if (len != sizeof(info)) {
1658             printf("read from sigfd returned %zd: %m\n", len);
1659             return;
1660         }
1661
1662         sigaction(info.ssi_signo, NULL, &action);
1663         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
1664             action.sa_sigaction(info.ssi_signo,
1665                                 (siginfo_t *)&info, NULL);
1666         } else if (action.sa_handler) {
1667             action.sa_handler(info.ssi_signo);
1668         }
1669     }
1670 }
1671
1672 int kvm_main_loop(void)
1673 {
1674     sigset_t mask;
1675     int sigfd;
1676
1677     io_thread = pthread_self();
1678     qemu_system_ready = 1;
1679
1680     sigemptyset(&mask);
1681     sigaddset(&mask, SIGIO);
1682     sigaddset(&mask, SIGALRM);
1683     sigaddset(&mask, SIGBUS);
1684     sigprocmask(SIG_BLOCK, &mask, NULL);
1685
1686     sigfd = qemu_signalfd(&mask);
1687     if (sigfd == -1) {
1688         fprintf(stderr, "failed to create signalfd\n");
1689         return -errno;
1690     }
1691
1692     fcntl(sigfd, F_SETFL, O_NONBLOCK);
1693
1694     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
1695                          (void *)(unsigned long) sigfd);
1696
1697     pthread_cond_broadcast(&qemu_system_cond);
1698
1699     io_thread_sigfd = sigfd;
1700     cpu_single_env = NULL;
1701
1702     while (1) {
1703         main_loop_wait(0);
1704         if (qemu_shutdown_requested()) {
1705             monitor_protocol_event(QEVENT_SHUTDOWN, NULL);
1706             if (qemu_no_shutdown()) {
1707                 vm_stop(0);
1708             } else {
1709                 break;
1710             }
1711         } else if (qemu_powerdown_requested()) {
1712             monitor_protocol_event(QEVENT_POWERDOWN, NULL);
1713             qemu_irq_raise(qemu_system_powerdown);
1714         } else if (qemu_reset_requested()) {
1715             qemu_kvm_system_reset();
1716         } else if (kvm_debug_cpu_requested) {
1717             gdb_set_stop_cpu(kvm_debug_cpu_requested);
1718             vm_stop(EXCP_DEBUG);
1719             kvm_debug_cpu_requested = NULL;
1720         }
1721     }
1722
1723     pause_all_threads();
1724     pthread_mutex_unlock(&qemu_mutex);
1725
1726     return 0;
1727 }
1728
1729 #if !defined(TARGET_I386)
1730 int kvm_arch_init_irq_routing(void)
1731 {
1732     return 0;
1733 }
1734 #endif
1735
1736 extern int no_hpet;
1737
1738 static int kvm_create_context(void)
1739 {
1740     static const char upgrade_note[] =
1741     "Please upgrade to at least kernel 2.6.29 or recent kvm-kmod\n"
1742     "(see http://sourceforge.net/projects/kvm).\n";
1743
1744     int r;
1745
1746     if (!kvm_irqchip) {
1747         kvm_disable_irqchip_creation(kvm_context);
1748     }
1749     if (!kvm_pit) {
1750         kvm_disable_pit_creation(kvm_context);
1751     }
1752     if (kvm_create(kvm_context, 0, NULL) < 0) {
1753         kvm_finalize(kvm_state);
1754         return -1;
1755     }
1756     r = kvm_arch_qemu_create_context();
1757     if (r < 0) {
1758         kvm_finalize(kvm_state);
1759         return -1;
1760     }
1761     if (kvm_pit && !kvm_pit_reinject) {
1762         if (kvm_reinject_control(kvm_context, 0)) {
1763             fprintf(stderr, "failure to disable in-kernel PIT reinjection\n");
1764             return -1;
1765         }
1766     }
1767
1768     /* There was a nasty bug in < kvm-80 that prevents memory slots from being
1769      * destroyed properly.  Since we rely on this capability, refuse to work
1770      * with any kernel without this capability. */
1771     if (!kvm_check_extension(kvm_state, KVM_CAP_DESTROY_MEMORY_REGION_WORKS)) {
1772         fprintf(stderr,
1773                 "KVM kernel module broken (DESTROY_MEMORY_REGION).\n%s",
1774                 upgrade_note);
1775         return -EINVAL;
1776     }
1777
1778     r = kvm_arch_init_irq_routing();
1779     if (r < 0) {
1780         return r;
1781     }
1782
1783     kvm_state->vcpu_events = 0;
1784 #ifdef KVM_CAP_VCPU_EVENTS
1785     kvm_state->vcpu_events = kvm_check_extension(kvm_state, KVM_CAP_VCPU_EVENTS);
1786 #endif
1787
1788     kvm_state->debugregs = 0;
1789 #ifdef KVM_CAP_DEBUGREGS
1790     kvm_state->debugregs = kvm_check_extension(kvm_state, KVM_CAP_DEBUGREGS);
1791 #endif
1792
1793     kvm_init_ap();
1794     if (kvm_irqchip) {
1795         if (!qemu_kvm_has_gsi_routing()) {
1796             irq0override = 0;
1797 #ifdef TARGET_I386
1798             /* if kernel can't do irq routing, interrupt source
1799              * override 0->2 can not be set up as required by hpet,
1800              * so disable hpet.
1801              */
1802             no_hpet = 1;
1803         } else if (!qemu_kvm_has_pit_state2()) {
1804             no_hpet = 1;
1805         }
1806 #else
1807         }
1808 #endif
1809     }
1810
1811     return 0;
1812 }
1813
1814 #ifdef KVM_CAP_IRQCHIP
1815
1816 int kvm_set_irq(int irq, int level, int *status)
1817 {
1818     return kvm_set_irq_level(kvm_context, irq, level, status);
1819 }
1820
1821 #endif
1822
1823 static void kvm_mutex_unlock(void)
1824 {
1825     assert(!cpu_single_env);
1826     pthread_mutex_unlock(&qemu_mutex);
1827 }
1828
1829 static void kvm_mutex_lock(void)
1830 {
1831     pthread_mutex_lock(&qemu_mutex);
1832     cpu_single_env = NULL;
1833 }
1834
1835 void qemu_mutex_unlock_iothread(void)
1836 {
1837     if (kvm_enabled()) {
1838         kvm_mutex_unlock();
1839     }
1840 }
1841
1842 void qemu_mutex_lock_iothread(void)
1843 {
1844     if (kvm_enabled()) {
1845         kvm_mutex_lock();
1846     }
1847 }
1848
1849 #ifdef CONFIG_KVM_DEVICE_ASSIGNMENT
1850 void kvm_add_ioperm_data(struct ioperm_data *data)
1851 {
1852     QLIST_INSERT_HEAD(&ioperm_head, data, entries);
1853 }
1854
1855 void kvm_remove_ioperm_data(unsigned long start_port, unsigned long num)
1856 {
1857     struct ioperm_data *data;
1858
1859     data = QLIST_FIRST(&ioperm_head);
1860     while (data) {
1861         struct ioperm_data *next = QLIST_NEXT(data, entries);
1862
1863         if (data->start_port == start_port && data->num == num) {
1864             QLIST_REMOVE(data, entries);
1865             qemu_free(data);
1866         }
1867
1868         data = next;
1869     }
1870 }
1871
1872 void kvm_ioperm(CPUState *env, void *data)
1873 {
1874     if (kvm_enabled() && qemu_system_ready) {
1875         on_vcpu(env, kvm_arch_do_ioperm, data);
1876     }
1877 }
1878
1879 #endif
1880
1881 int kvm_set_boot_cpu_id(uint32_t id)
1882 {
1883     return kvm_set_boot_vcpu_id(kvm_context, id);
1884 }
1885
1886 #ifdef TARGET_I386
1887 #ifdef KVM_CAP_MCE
1888 struct kvm_x86_mce_data {
1889     CPUState *env;
1890     struct kvm_x86_mce *mce;
1891     int abort_on_error;
1892 };
1893
1894 static void kvm_do_inject_x86_mce(void *_data)
1895 {
1896     struct kvm_x86_mce_data *data = _data;
1897     int r;
1898
1899     /* If there is an MCE excpetion being processed, ignore this SRAO MCE */
1900     r = kvm_mce_in_exception(data->env);
1901     if (r == -1) {
1902         fprintf(stderr, "Failed to get MCE status\n");
1903     } else if (r && !(data->mce->status & MCI_STATUS_AR)) {
1904         return;
1905     }
1906     r = kvm_set_mce(data->env, data->mce);
1907     if (r < 0) {
1908         perror("kvm_set_mce FAILED");
1909         if (data->abort_on_error) {
1910             abort();
1911         }
1912     }
1913 }
1914 #endif
1915
1916 void kvm_inject_x86_mce(CPUState *cenv, int bank, uint64_t status,
1917                         uint64_t mcg_status, uint64_t addr, uint64_t misc,
1918                         int abort_on_error)
1919 {
1920 #ifdef KVM_CAP_MCE
1921     struct kvm_x86_mce mce = {
1922         .bank = bank,
1923         .status = status,
1924         .mcg_status = mcg_status,
1925         .addr = addr,
1926         .misc = misc,
1927     };
1928     struct kvm_x86_mce_data data = {
1929         .env = cenv,
1930         .mce = &mce,
1931         .abort_on_error = abort_on_error,
1932     };
1933
1934     if (!cenv->mcg_cap) {
1935         fprintf(stderr, "MCE support is not enabled!\n");
1936         return;
1937     }
1938     on_vcpu(cenv, kvm_do_inject_x86_mce, &data);
1939 #else
1940     if (abort_on_error) {
1941         abort();
1942     }
1943 #endif
1944 }
1945 #endif