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