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