cpus.c

   1 /*
   2  * QEMU System Emulator
   3  *
   4  * Copyright (c) 2003-2008 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 /* Needed early for CONFIG_BSD etc. */
  26 #include "config-host.h"
  27
  28 #include "monitor.h"
  29 #include "sysemu.h"
  30 #include "gdbstub.h"
  31 #include "dma.h"
  32 #include "kvm.h"
  33 #include "exec-all.h"
  34
  35 #include "qemu-thread.h"
  36 #include "cpus.h"
  37 #include "compatfd.h"
  38
  39 #ifdef SIGRTMIN
  40 #define SIG_IPI (SIGRTMIN+4)
  41 #else
  42 #define SIG_IPI SIGUSR1
  43 #endif
  44
  45 #ifdef CONFIG_LINUX
  46
  47 #include <sys/prctl.h>
  48
  49 #ifndef PR_MCE_KILL
  50 #define PR_MCE_KILL 33
  51 #endif
  52
  53 #ifndef PR_MCE_KILL_SET
  54 #define PR_MCE_KILL_SET 1
  55 #endif
  56
  57 #ifndef PR_MCE_KILL_EARLY
  58 #define PR_MCE_KILL_EARLY 1
  59 #endif
  60
  61 #endif /* CONFIG_LINUX */
  62
  63 static CPUState *next_cpu;
  64
  65 /***********************************************************/
  66 void hw_error(const char *fmt, ...)
  67 {
  68     va_list ap;
  69     CPUState *env;
  70
  71     va_start(ap, fmt);
  72     fprintf(stderr, "qemu: hardware error: ");
  73     vfprintf(stderr, fmt, ap);
  74     fprintf(stderr, "\n");
  75     for(env = first_cpu; env != NULL; env = env->next_cpu) {
  76         fprintf(stderr, "CPU #%d:\n", env->cpu_index);
  77 #ifdef TARGET_I386
  78         cpu_dump_state(env, stderr, fprintf, X86_DUMP_FPU);
  79 #else
  80         cpu_dump_state(env, stderr, fprintf, 0);
  81 #endif
  82     }
  83     va_end(ap);
  84     abort();
  85 }
  86
  87 void cpu_synchronize_all_states(void)
  88 {
  89     CPUState *cpu;
  90
  91     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
  92         cpu_synchronize_state(cpu);
  93     }
  94 }
  95
  96 void cpu_synchronize_all_post_reset(void)
  97 {
  98     CPUState *cpu;
  99
 100     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
 101         cpu_synchronize_post_reset(cpu);
 102     }
 103 }
 104
 105 void cpu_synchronize_all_post_init(void)
 106 {
 107     CPUState *cpu;
 108
 109     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
 110         cpu_synchronize_post_init(cpu);
 111     }
 112 }
 113
 114 int cpu_is_stopped(CPUState *env)
 115 {
 116     return !vm_running || env->stopped;
 117 }
 118
 119 static void do_vm_stop(int reason)
 120 {
 121     if (vm_running) {
 122         cpu_disable_ticks();
 123         vm_running = 0;
 124         pause_all_vcpus();
 125         vm_state_notify(0, reason);
 126         qemu_aio_flush();
 127         bdrv_flush_all();
 128         monitor_protocol_event(QEVENT_STOP, NULL);
 129     }
 130 }
 131
 132 static int cpu_can_run(CPUState *env)
 133 {
 134     if (env->stop) {
 135         return 0;
 136     }
 137     if (env->stopped || !vm_running) {
 138         return 0;
 139     }
 140     return 1;
 141 }
 142
 143 static bool cpu_thread_is_idle(CPUState *env)
 144 {
 145     if (env->stop || env->queued_work_first) {
 146         return false;
 147     }
 148     if (env->stopped || !vm_running) {
 149         return true;
 150     }
 151     if (!env->halted || qemu_cpu_has_work(env) ||
 152         (kvm_enabled() && kvm_irqchip_in_kernel())) {
 153         return false;
 154     }
 155     return true;
 156 }
 157
 158 bool all_cpu_threads_idle(void)
 159 {
 160     CPUState *env;
 161
 162     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 163         if (!cpu_thread_is_idle(env)) {
 164             return false;
 165         }
 166     }
 167     return true;
 168 }
 169
 170 static void cpu_handle_guest_debug(CPUState *env)
 171 {
 172     gdb_set_stop_cpu(env);
 173     qemu_system_debug_request();
 174 #ifdef CONFIG_IOTHREAD
 175     env->stopped = 1;
 176 #endif
 177 }
 178
 179 #ifdef CONFIG_IOTHREAD
 180 static void cpu_signal(int sig)
 181 {
 182     if (cpu_single_env) {
 183         cpu_exit(cpu_single_env);
 184     }
 185     exit_request = 1;
 186 }
 187 #endif
 188
 189 #ifdef CONFIG_LINUX
 190 static void sigbus_reraise(void)
 191 {
 192     sigset_t set;
 193     struct sigaction action;
 194
 195     memset(&action, 0, sizeof(action));
 196     action.sa_handler = SIG_DFL;
 197     if (!sigaction(SIGBUS, &action, NULL)) {
 198         raise(SIGBUS);
 199         sigemptyset(&set);
 200         sigaddset(&set, SIGBUS);
 201         sigprocmask(SIG_UNBLOCK, &set, NULL);
 202     }
 203     perror("Failed to re-raise SIGBUS!\n");
 204     abort();
 205 }
 206
 207 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
 208                            void *ctx)
 209 {
 210     if (kvm_on_sigbus(siginfo->ssi_code,
 211                       (void *)(intptr_t)siginfo->ssi_addr)) {
 212         sigbus_reraise();
 213     }
 214 }
 215
 216 static void qemu_init_sigbus(void)
 217 {
 218     struct sigaction action;
 219
 220     memset(&action, 0, sizeof(action));
 221     action.sa_flags = SA_SIGINFO;
 222     action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
 223     sigaction(SIGBUS, &action, NULL);
 224
 225     prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0);
 226 }
 227
 228 static void qemu_kvm_eat_signals(CPUState *env)
 229 {
 230     struct timespec ts = { 0, 0 };
 231     siginfo_t siginfo;
 232     sigset_t waitset;
 233     sigset_t chkset;
 234     int r;
 235
 236     sigemptyset(&waitset);
 237     sigaddset(&waitset, SIG_IPI);
 238     sigaddset(&waitset, SIGBUS);
 239
 240     do {
 241         r = sigtimedwait(&waitset, &siginfo, &ts);
 242         if (r == -1 && !(errno == EAGAIN || errno == EINTR)) {
 243             perror("sigtimedwait");
 244             exit(1);
 245         }
 246
 247         switch (r) {
 248         case SIGBUS:
 249             if (kvm_on_sigbus_vcpu(env, siginfo.si_code, siginfo.si_addr)) {
 250                 sigbus_reraise();
 251             }
 252             break;
 253         default:
 254             break;
 255         }
 256
 257         r = sigpending(&chkset);
 258         if (r == -1) {
 259             perror("sigpending");
 260             exit(1);
 261         }
 262     } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
 263
 264 #ifndef CONFIG_IOTHREAD
 265     if (sigismember(&chkset, SIGIO) || sigismember(&chkset, SIGALRM)) {
 266         qemu_notify_event();
 267     }
 268 #endif
 269 }
 270
 271 #else /* !CONFIG_LINUX */
 272
 273 static void qemu_init_sigbus(void)
 274 {
 275 }
 276
 277 static void qemu_kvm_eat_signals(CPUState *env)
 278 {
 279 }
 280 #endif /* !CONFIG_LINUX */
 281
 282 #ifndef _WIN32
 283 static int io_thread_fd = -1;
 284
 285 static void qemu_event_increment(void)
 286 {
 287     /* Write 8 bytes to be compatible with eventfd.  */
 288     static const uint64_t val = 1;
 289     ssize_t ret;
 290
 291     if (io_thread_fd == -1) {
 292         return;
 293     }
 294     do {
 295         ret = write(io_thread_fd, &val, sizeof(val));
 296     } while (ret < 0 && errno == EINTR);
 297
 298     /* EAGAIN is fine, a read must be pending.  */
 299     if (ret < 0 && errno != EAGAIN) {
 300         fprintf(stderr, "qemu_event_increment: write() filed: %s\n",
 301                 strerror(errno));
 302         exit (1);
 303     }
 304 }
 305
 306 static void qemu_event_read(void *opaque)
 307 {
 308     int fd = (intptr_t)opaque;
 309     ssize_t len;
 310     char buffer[512];
 311
 312     /* Drain the notify pipe.  For eventfd, only 8 bytes will be read.  */
 313     do {
 314         len = read(fd, buffer, sizeof(buffer));
 315     } while ((len == -1 && errno == EINTR) || len == sizeof(buffer));
 316 }
 317
 318 static int qemu_event_init(void)
 319 {
 320     int err;
 321     int fds[2];
 322
 323     err = qemu_eventfd(fds);
 324     if (err == -1) {
 325         return -errno;
 326     }
 327     err = fcntl_setfl(fds[0], O_NONBLOCK);
 328     if (err < 0) {
 329         goto fail;
 330     }
 331     err = fcntl_setfl(fds[1], O_NONBLOCK);
 332     if (err < 0) {
 333         goto fail;
 334     }
 335     qemu_set_fd_handler2(fds[0], NULL, qemu_event_read, NULL,
 336                          (void *)(intptr_t)fds[0]);
 337
 338     io_thread_fd = fds[1];
 339     return 0;
 340
 341 fail:
 342     close(fds[0]);
 343     close(fds[1]);
 344     return err;
 345 }
 346
 347 #ifdef OBSOLETE_KVM_IMPL
 348 static void dummy_signal(int sig)
 349 {
 350 }
 351 #endif
 352
 353 /* If we have signalfd, we mask out the signals we want to handle and then
 354  * use signalfd to listen for them.  We rely on whatever the current signal
 355  * handler is to dispatch the signals when we receive them.
 356  */
 357 static void sigfd_handler(void *opaque)
 358 {
 359     int fd = (intptr_t)opaque;
 360     struct qemu_signalfd_siginfo info;
 361     struct sigaction action;
 362     ssize_t len;
 363
 364     while (1) {
 365         do {
 366             len = read(fd, &info, sizeof(info));
 367         } while (len == -1 && errno == EINTR);
 368
 369         if (len == -1 && errno == EAGAIN) {
 370             break;
 371         }
 372
 373         if (len != sizeof(info)) {
 374             printf("read from sigfd returned %zd: %m\n", len);
 375             return;
 376         }
 377
 378         sigaction(info.ssi_signo, NULL, &action);
 379         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
 380             action.sa_sigaction(info.ssi_signo,
 381                                 (siginfo_t *)&info, NULL);
 382         } else if (action.sa_handler) {
 383             action.sa_handler(info.ssi_signo);
 384         }
 385     }
 386 }
 387
 388 static int qemu_signal_init(void)
 389 {
 390     int sigfd;
 391     sigset_t set;
 392
 393 #ifdef CONFIG_IOTHREAD
 394     /* SIGUSR2 used by posix-aio-compat.c */
 395     sigemptyset(&set);
 396     sigaddset(&set, SIGUSR2);
 397     pthread_sigmask(SIG_UNBLOCK, &set, NULL);
 398
 399     sigemptyset(&set);
 400     sigaddset(&set, SIGIO);
 401     sigaddset(&set, SIGALRM);
 402     sigaddset(&set, SIG_IPI);
 403     sigaddset(&set, SIGBUS);
 404     pthread_sigmask(SIG_BLOCK, &set, NULL);
 405 #else
 406     sigemptyset(&set);
 407     sigaddset(&set, SIGBUS);
 408     if (kvm_enabled()) {
 409         /*
 410          * We need to process timer signals synchronously to avoid a race
 411          * between exit_request check and KVM vcpu entry.
 412          */
 413         sigaddset(&set, SIGIO);
 414         sigaddset(&set, SIGALRM);
 415     }
 416 #endif
 417
 418     sigfd = qemu_signalfd(&set);
 419     if (sigfd == -1) {
 420         fprintf(stderr, "failed to create signalfd\n");
 421         return -errno;
 422     }
 423
 424     fcntl_setfl(sigfd, O_NONBLOCK);
 425
 426     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
 427                          (void *)(intptr_t)sigfd);
 428
 429     return 0;
 430 }
 431
 432 static void qemu_kvm_init_cpu_signals(CPUState *env)
 433 {
 434 #if 0 /* Causes regressions: autotest WinXP.64 migrate.tcp */
 435     int r;
 436     sigset_t set;
 437     struct sigaction sigact;
 438
 439     memset(&sigact, 0, sizeof(sigact));
 440     sigact.sa_handler = dummy_signal;
 441     sigaction(SIG_IPI, &sigact, NULL);
 442
 443 #ifdef CONFIG_IOTHREAD
 444     pthread_sigmask(SIG_BLOCK, NULL, &set);
 445     sigdelset(&set, SIG_IPI);
 446     sigdelset(&set, SIGBUS);
 447     r = kvm_set_signal_mask(env, &set);
 448     if (r) {
 449         fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
 450         exit(1);
 451     }
 452 #else
 453     sigemptyset(&set);
 454     sigaddset(&set, SIG_IPI);
 455     sigaddset(&set, SIGIO);
 456     sigaddset(&set, SIGALRM);
 457     pthread_sigmask(SIG_BLOCK, &set, NULL);
 458
 459     pthread_sigmask(SIG_BLOCK, NULL, &set);
 460     sigdelset(&set, SIGIO);
 461     sigdelset(&set, SIGALRM);
 462 #endif
 463     sigdelset(&set, SIG_IPI);
 464     sigdelset(&set, SIGBUS);
 465     r = kvm_set_signal_mask(env, &set);
 466     if (r) {
 467         fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
 468         exit(1);
 469     }
 470 #endif
 471 }
 472
 473 static void qemu_tcg_init_cpu_signals(void)
 474 {
 475 #ifdef CONFIG_IOTHREAD
 476     sigset_t set;
 477     struct sigaction sigact;
 478
 479     memset(&sigact, 0, sizeof(sigact));
 480     sigact.sa_handler = cpu_signal;
 481     sigaction(SIG_IPI, &sigact, NULL);
 482
 483     sigemptyset(&set);
 484     sigaddset(&set, SIG_IPI);
 485     pthread_sigmask(SIG_UNBLOCK, &set, NULL);
 486 #endif
 487 }
 488
 489 #else /* _WIN32 */
 490
 491 HANDLE qemu_event_handle;
 492
 493 static void dummy_event_handler(void *opaque)
 494 {
 495 }
 496
 497 static int qemu_event_init(void)
 498 {
 499     qemu_event_handle = CreateEvent(NULL, FALSE, FALSE, NULL);
 500     if (!qemu_event_handle) {
 501         fprintf(stderr, "Failed CreateEvent: %ld\n", GetLastError());
 502         return -1;
 503     }
 504     qemu_add_wait_object(qemu_event_handle, dummy_event_handler, NULL);
 505     return 0;
 506 }
 507
 508 static void qemu_event_increment(void)
 509 {
 510     if (!SetEvent(qemu_event_handle)) {
 511         fprintf(stderr, "qemu_event_increment: SetEvent failed: %ld\n",
 512                 GetLastError());
 513         exit (1);
 514     }
 515 }
 516
 517 static int qemu_signal_init(void)
 518 {
 519     return 0;
 520 }
 521
 522 static void qemu_kvm_init_cpu_signals(CPUState *env)
 523 {
 524     abort();
 525 }
 526
 527 static void qemu_tcg_init_cpu_signals(void)
 528 {
 529 }
 530 #endif /* _WIN32 */
 531
 532 #ifndef CONFIG_IOTHREAD
 533 int qemu_init_main_loop(void)
 534 {
 535     int ret;
 536
 537     ret = qemu_signal_init();
 538     if (ret) {
 539         return ret;
 540     }
 541
 542     qemu_init_sigbus();
 543
 544     return qemu_event_init();
 545 }
 546
 547 void qemu_main_loop_start(void)
 548 {
 549 }
 550
 551 void qemu_init_vcpu(void *_env)
 552 {
 553     CPUState *env = _env;
 554     int r;
 555
 556     env->nr_cores = smp_cores;
 557     env->nr_threads = smp_threads;
 558
 559     if (kvm_enabled()) {
 560         r = kvm_init_vcpu(env);
 561         if (r < 0) {
 562             fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r));
 563             exit(1);
 564         }
 565         qemu_kvm_init_cpu_signals(env);
 566     } else {
 567         qemu_tcg_init_cpu_signals();
 568     }
 569 }
 570
 571 int qemu_cpu_is_self(void *env)
 572 {
 573     return 1;
 574 }
 575
 576 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
 577 {
 578     func(data);
 579 }
 580
 581 void resume_all_vcpus(void)
 582 {
 583 }
 584
 585 void pause_all_vcpus(void)
 586 {
 587 }
 588
 589 void qemu_cpu_kick(void *env)
 590 {
 591 }
 592
 593 void qemu_cpu_kick_self(void)
 594 {
 595 #ifndef _WIN32
 596     assert(cpu_single_env);
 597
 598     raise(SIG_IPI);
 599 #else
 600     abort();
 601 #endif
 602 }
 603
 604 void qemu_notify_event(void)
 605 {
 606     CPUState *env = cpu_single_env;
 607
 608     qemu_event_increment ();
 609     if (env) {
 610         cpu_exit(env);
 611     }
 612     if (next_cpu && env != next_cpu) {
 613         cpu_exit(next_cpu);
 614     }
 615     exit_request = 1;
 616 }
 617
 618 #if defined(OBSOLETE_KVM_IMPL) || !defined(CONFIG_KVM)
 619 void qemu_mutex_lock_iothread(void) {}
 620 void qemu_mutex_unlock_iothread(void) {}
 621 #endif
 622
 623 void cpu_stop_current(void)
 624 {
 625 }
 626
 627 void vm_stop(int reason)
 628 {
 629     do_vm_stop(reason);
 630 }
 631
 632 #else /* CONFIG_IOTHREAD */
 633
 634 QemuMutex qemu_global_mutex;
 635 static QemuMutex qemu_fair_mutex;
 636
 637 static QemuThread io_thread;
 638
 639 static QemuThread *tcg_cpu_thread;
 640 static QemuCond *tcg_halt_cond;
 641
 642 static int qemu_system_ready;
 643 /* cpu creation */
 644 static QemuCond qemu_cpu_cond;
 645 /* system init */
 646 static QemuCond qemu_system_cond;
 647 static QemuCond qemu_pause_cond;
 648 static QemuCond qemu_work_cond;
 649
 650 int qemu_init_main_loop(void)
 651 {
 652     int ret;
 653
 654     qemu_init_sigbus();
 655
 656     ret = qemu_signal_init();
 657     if (ret) {
 658         return ret;
 659     }
 660
 661     /* Note eventfd must be drained before signalfd handlers run */
 662     ret = qemu_event_init();
 663     if (ret) {
 664         return ret;
 665     }
 666
 667     qemu_cond_init(&qemu_cpu_cond);
 668     qemu_cond_init(&qemu_system_cond);
 669     qemu_cond_init(&qemu_pause_cond);
 670     qemu_cond_init(&qemu_work_cond);
 671     qemu_mutex_init(&qemu_fair_mutex);
 672     qemu_mutex_init(&qemu_global_mutex);
 673     qemu_mutex_lock(&qemu_global_mutex);
 674
 675     qemu_thread_get_self(&io_thread);
 676
 677     return 0;
 678 }
 679
 680 void qemu_main_loop_start(void)
 681 {
 682     qemu_system_ready = 1;
 683     qemu_cond_broadcast(&qemu_system_cond);
 684 }
 685
 686 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
 687 {
 688     struct qemu_work_item wi;
 689
 690     if (qemu_cpu_is_self(env)) {
 691         func(data);
 692         return;
 693     }
 694
 695     wi.func = func;
 696     wi.data = data;
 697     if (!env->queued_work_first) {
 698         env->queued_work_first = &wi;
 699     } else {
 700         env->queued_work_last->next = &wi;
 701     }
 702     env->queued_work_last = &wi;
 703     wi.next = NULL;
 704     wi.done = false;
 705
 706     qemu_cpu_kick(env);
 707     while (!wi.done) {
 708         CPUState *self_env = cpu_single_env;
 709
 710         qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
 711         cpu_single_env = self_env;
 712     }
 713 }
 714
 715 static void flush_queued_work(CPUState *env)
 716 {
 717     struct qemu_work_item *wi;
 718
 719     if (!env->queued_work_first) {
 720         return;
 721     }
 722
 723     while ((wi = env->queued_work_first)) {
 724         env->queued_work_first = wi->next;
 725         wi->func(wi->data);
 726         wi->done = true;
 727     }
 728     env->queued_work_last = NULL;
 729     qemu_cond_broadcast(&qemu_work_cond);
 730 }
 731
 732 static void qemu_wait_io_event_common(CPUState *env)
 733 {
 734     if (env->stop) {
 735         env->stop = 0;
 736         env->stopped = 1;
 737         qemu_cond_signal(&qemu_pause_cond);
 738     }
 739     flush_queued_work(env);
 740     env->thread_kicked = false;
 741 }
 742
 743 static void qemu_tcg_wait_io_event(void)
 744 {
 745     CPUState *env;
 746
 747     while (all_cpu_threads_idle()) {
 748        /* Start accounting real time to the virtual clock if the CPUs
 749           are idle.  */
 750         qemu_clock_warp(vm_clock);
 751         qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
 752     }
 753
 754     qemu_mutex_unlock(&qemu_global_mutex);
 755
 756     /*
 757      * Users of qemu_global_mutex can be starved, having no chance
 758      * to acquire it since this path will get to it first.
 759      * So use another lock to provide fairness.
 760      */
 761     qemu_mutex_lock(&qemu_fair_mutex);
 762     qemu_mutex_unlock(&qemu_fair_mutex);
 763
 764     qemu_mutex_lock(&qemu_global_mutex);
 765
 766     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 767         qemu_wait_io_event_common(env);
 768     }
 769 }
 770
 771 static void qemu_kvm_wait_io_event(CPUState *env)
 772 {
 773     while (cpu_thread_is_idle(env)) {
 774         qemu_cond_wait(env->halt_cond, &qemu_global_mutex);
 775     }
 776
 777     qemu_kvm_eat_signals(env);
 778     qemu_wait_io_event_common(env);
 779 }
 780
 781 static void *qemu_kvm_cpu_thread_fn(void *arg)
 782 {
 783     CPUState *env = arg;
 784     int r;
 785
 786     qemu_mutex_lock(&qemu_global_mutex);
 787     qemu_thread_get_self(env->thread);
 788     env->thread_id = qemu_get_thread_id();
 789
 790     r = kvm_init_vcpu(env);
 791     if (r < 0) {
 792         fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r));
 793         exit(1);
 794     }
 795
 796     qemu_kvm_init_cpu_signals(env);
 797
 798     /* signal CPU creation */
 799     env->created = 1;
 800     qemu_cond_signal(&qemu_cpu_cond);
 801
 802     /* and wait for machine initialization */
 803     while (!qemu_system_ready) {
 804         qemu_cond_wait(&qemu_system_cond, &qemu_global_mutex);
 805     }
 806
 807     while (1) {
 808         if (cpu_can_run(env)) {
 809             r = kvm_cpu_exec(env);
 810             if (r == EXCP_DEBUG) {
 811                 cpu_handle_guest_debug(env);
 812             }
 813         }
 814         qemu_kvm_wait_io_event(env);
 815     }
 816
 817     return NULL;
 818 }
 819
 820 static void *qemu_tcg_cpu_thread_fn(void *arg)
 821 {
 822     CPUState *env = arg;
 823
 824     qemu_tcg_init_cpu_signals();
 825     qemu_thread_get_self(env->thread);
 826
 827     /* signal CPU creation */
 828     qemu_mutex_lock(&qemu_global_mutex);
 829     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 830         env->thread_id = qemu_get_thread_id();
 831         env->created = 1;
 832     }
 833     qemu_cond_signal(&qemu_cpu_cond);
 834
 835     /* and wait for machine initialization */
 836     while (!qemu_system_ready) {
 837         qemu_cond_wait(&qemu_system_cond, &qemu_global_mutex);
 838     }
 839
 840     while (1) {
 841         cpu_exec_all();
 842         if (use_icount && qemu_next_icount_deadline() <= 0) {
 843             qemu_notify_event();
 844         }
 845         qemu_tcg_wait_io_event();
 846     }
 847
 848     return NULL;
 849 }
 850
 851 static void qemu_cpu_kick_thread(CPUState *env)
 852 {
 853 #ifndef _WIN32
 854     int err;
 855
 856     err = pthread_kill(env->thread->thread, SIG_IPI);
 857     if (err) {
 858         fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
 859         exit(1);
 860     }
 861 #else /* _WIN32 */
 862     if (!qemu_cpu_is_self(env)) {
 863         SuspendThread(env->thread->thread);
 864         cpu_signal(0);
 865         ResumeThread(env->thread->thread);
 866     }
 867 #endif
 868 }
 869
 870 void qemu_cpu_kick(void *_env)
 871 {
 872     CPUState *env = _env;
 873
 874     qemu_cond_broadcast(env->halt_cond);
 875     if (!env->thread_kicked) {
 876         qemu_cpu_kick_thread(env);
 877         env->thread_kicked = true;
 878     }
 879 }
 880
 881 void qemu_cpu_kick_self(void)
 882 {
 883 #ifndef _WIN32
 884     assert(cpu_single_env);
 885
 886     if (!cpu_single_env->thread_kicked) {
 887         qemu_cpu_kick_thread(cpu_single_env);
 888         cpu_single_env->thread_kicked = true;
 889     }
 890 #else
 891     abort();
 892 #endif
 893 }
 894
 895 int qemu_cpu_is_self(void *_env)
 896 {
 897     CPUState *env = _env;
 898
 899     return qemu_thread_is_self(env->thread);
 900 }
 901
 902 void qemu_mutex_lock_iothread(void)
 903 {
 904     if (kvm_enabled()) {
 905         qemu_mutex_lock(&qemu_global_mutex);
 906     } else {
 907         qemu_mutex_lock(&qemu_fair_mutex);
 908         if (qemu_mutex_trylock(&qemu_global_mutex)) {
 909             qemu_cpu_kick_thread(first_cpu);
 910             qemu_mutex_lock(&qemu_global_mutex);
 911         }
 912         qemu_mutex_unlock(&qemu_fair_mutex);
 913     }
 914 }
 915
 916 void qemu_mutex_unlock_iothread(void)
 917 {
 918     qemu_mutex_unlock(&qemu_global_mutex);
 919 }
 920
 921 static int all_vcpus_paused(void)
 922 {
 923     CPUState *penv = first_cpu;
 924
 925     while (penv) {
 926         if (!penv->stopped) {
 927             return 0;
 928         }
 929         penv = (CPUState *)penv->next_cpu;
 930     }
 931
 932     return 1;
 933 }
 934
 935 void pause_all_vcpus(void)
 936 {
 937     CPUState *penv = first_cpu;
 938
 939     while (penv) {
 940         penv->stop = 1;
 941         qemu_cpu_kick(penv);
 942         penv = (CPUState *)penv->next_cpu;
 943     }
 944
 945     while (!all_vcpus_paused()) {
 946         qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex);
 947         penv = first_cpu;
 948         while (penv) {
 949             qemu_cpu_kick(penv);
 950             penv = (CPUState *)penv->next_cpu;
 951         }
 952     }
 953 }
 954
 955 void resume_all_vcpus(void)
 956 {
 957     CPUState *penv = first_cpu;
 958
 959     while (penv) {
 960         penv->stop = 0;
 961         penv->stopped = 0;
 962         qemu_cpu_kick(penv);
 963         penv = (CPUState *)penv->next_cpu;
 964     }
 965 }
 966
 967 static void qemu_tcg_init_vcpu(void *_env)
 968 {
 969     CPUState *env = _env;
 970
 971     /* share a single thread for all cpus with TCG */
 972     if (!tcg_cpu_thread) {
 973         env->thread = qemu_mallocz(sizeof(QemuThread));
 974         env->halt_cond = qemu_mallocz(sizeof(QemuCond));
 975         qemu_cond_init(env->halt_cond);
 976         qemu_thread_create(env->thread, qemu_tcg_cpu_thread_fn, env);
 977         while (env->created == 0) {
 978             qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
 979         }
 980         tcg_cpu_thread = env->thread;
 981         tcg_halt_cond = env->halt_cond;
 982     } else {
 983         env->thread = tcg_cpu_thread;
 984         env->halt_cond = tcg_halt_cond;
 985     }
 986 }
 987
 988 static void qemu_kvm_start_vcpu(CPUState *env)
 989 {
 990     env->thread = qemu_mallocz(sizeof(QemuThread));
 991     env->halt_cond = qemu_mallocz(sizeof(QemuCond));
 992     qemu_cond_init(env->halt_cond);
 993     qemu_thread_create(env->thread, qemu_kvm_cpu_thread_fn, env);
 994     while (env->created == 0) {
 995         qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
 996     }
 997 }
 998
 999 void qemu_init_vcpu(void *_env)
1000 {
1001     CPUState *env = _env;
1002
1003     env->nr_cores = smp_cores;
1004     env->nr_threads = smp_threads;
1005     if (kvm_enabled()) {
1006         qemu_kvm_start_vcpu(env);
1007     } else {
1008         qemu_tcg_init_vcpu(env);
1009     }
1010 }
1011
1012 void qemu_notify_event(void)
1013 {
1014     qemu_event_increment();
1015 }
1016
1017 void cpu_stop_current(void)
1018 {
1019     if (cpu_single_env) {
1020         cpu_single_env->stop = 0;
1021         cpu_single_env->stopped = 1;
1022         cpu_exit(cpu_single_env);
1023         qemu_cond_signal(&qemu_pause_cond);
1024     }
1025 }
1026
1027 void vm_stop(int reason)
1028 {
1029     if (!qemu_thread_is_self(&io_thread)) {
1030         qemu_system_vmstop_request(reason);
1031         /*
1032          * FIXME: should not return to device code in case
1033          * vm_stop() has been requested.
1034          */
1035         cpu_stop_current();
1036         return;
1037     }
1038     do_vm_stop(reason);
1039 }
1040
1041 #endif
1042
1043 static int tcg_cpu_exec(CPUState *env)
1044 {
1045     int ret;
1046 #ifdef CONFIG_PROFILER
1047     int64_t ti;
1048 #endif
1049
1050 #ifdef CONFIG_PROFILER
1051     ti = profile_getclock();
1052 #endif
1053     if (use_icount) {
1054         int64_t count;
1055         int decr;
1056         qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
1057         env->icount_decr.u16.low = 0;
1058         env->icount_extra = 0;
1059         count = qemu_icount_round(qemu_next_icount_deadline());
1060         qemu_icount += count;
1061         decr = (count > 0xffff) ? 0xffff : count;
1062         count -= decr;
1063         env->icount_decr.u16.low = decr;
1064         env->icount_extra = count;
1065     }
1066     ret = cpu_exec(env);
1067 #ifdef CONFIG_PROFILER
1068     qemu_time += profile_getclock() - ti;
1069 #endif
1070     if (use_icount) {
1071         /* Fold pending instructions back into the
1072            instruction counter, and clear the interrupt flag.  */
1073         qemu_icount -= (env->icount_decr.u16.low
1074                         + env->icount_extra);
1075         env->icount_decr.u32 = 0;
1076         env->icount_extra = 0;
1077     }
1078     return ret;
1079 }
1080
1081 bool cpu_exec_all(void)
1082 {
1083     int r;
1084
1085     /* Account partial waits to the vm_clock.  */
1086     qemu_clock_warp(vm_clock);
1087
1088     if (next_cpu == NULL) {
1089         next_cpu = first_cpu;
1090     }
1091     for (; next_cpu != NULL && !exit_request; next_cpu = next_cpu->next_cpu) {
1092         CPUState *env = next_cpu;
1093
1094         qemu_clock_enable(vm_clock,
1095                           (env->singlestep_enabled & SSTEP_NOTIMER) == 0);
1096
1097 #ifndef CONFIG_IOTHREAD
1098         if (qemu_alarm_pending()) {
1099             break;
1100         }
1101 #endif
1102         if (cpu_can_run(env)) {
1103             if (kvm_enabled()) {
1104                 r = kvm_cpu_exec(env);
1105                 qemu_kvm_eat_signals(env);
1106             } else {
1107                 r = tcg_cpu_exec(env);
1108             }
1109             if (r == EXCP_DEBUG) {
1110                 cpu_handle_guest_debug(env);
1111                 break;
1112             }
1113         } else if (env->stop || env->stopped) {
1114             break;
1115         }
1116     }
1117     exit_request = 0;
1118     return !all_cpu_threads_idle();
1119 }
1120
1121 void set_numa_modes(void)
1122 {
1123     CPUState *env;
1124     int i;
1125
1126     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1127         for (i = 0; i < nb_numa_nodes; i++) {
1128             if (node_cpumask[i] & (1 << env->cpu_index)) {
1129                 env->numa_node = i;
1130             }
1131         }
1132     }
1133 }
1134
1135 void set_cpu_log(const char *optarg)
1136 {
1137     int mask;
1138     const CPULogItem *item;
1139
1140     mask = cpu_str_to_log_mask(optarg);
1141     if (!mask) {
1142         printf("Log items (comma separated):\n");
1143         for (item = cpu_log_items; item->mask != 0; item++) {
1144             printf("%-10s %s\n", item->name, item->help);
1145         }
1146         exit(1);
1147     }
1148     cpu_set_log(mask);
1149 }
1150
1151 /* Return the virtual CPU time, based on the instruction counter.  */
1152 int64_t cpu_get_icount(void)
1153 {
1154     int64_t icount;
1155     CPUState *env = cpu_single_env;;
1156
1157     icount = qemu_icount;
1158     if (env) {
1159         if (!can_do_io(env)) {
1160             fprintf(stderr, "Bad clock read\n");
1161         }
1162         icount -= (env->icount_decr.u16.low + env->icount_extra);
1163     }
1164     return qemu_icount_bias + (icount << icount_time_shift);
1165 }
1166
1167 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
1168 {
1169     /* XXX: implement xxx_cpu_list for targets that still miss it */
1170 #if defined(cpu_list_id)
1171     cpu_list_id(f, cpu_fprintf, optarg);
1172 #elif defined(cpu_list)
1173     cpu_list(f, cpu_fprintf); /* deprecated */
1174 #endif
1175 }