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 "cpus.h"
  36 #include "compatfd.h"
  37 #ifdef CONFIG_LINUX
  38 #include <sys/prctl.h>
  39 #endif
  40
  41 #ifdef SIGRTMIN
  42 #define SIG_IPI (SIGRTMIN+4)
  43 #else
  44 #define SIG_IPI SIGUSR1
  45 #endif
  46
  47 #ifndef PR_MCE_KILL
  48 #define PR_MCE_KILL 33
  49 #endif
  50
  51 static CPUState *next_cpu;
  52
  53 /***********************************************************/
  54 void hw_error(const char *fmt, ...)
  55 {
  56     va_list ap;
  57     CPUState *env;
  58
  59     va_start(ap, fmt);
  60     fprintf(stderr, "qemu: hardware error: ");
  61     vfprintf(stderr, fmt, ap);
  62     fprintf(stderr, "\n");
  63     for(env = first_cpu; env != NULL; env = env->next_cpu) {
  64         fprintf(stderr, "CPU #%d:\n", env->cpu_index);
  65 #ifdef TARGET_I386
  66         cpu_dump_state(env, stderr, fprintf, X86_DUMP_FPU);
  67 #else
  68         cpu_dump_state(env, stderr, fprintf, 0);
  69 #endif
  70     }
  71     va_end(ap);
  72     abort();
  73 }
  74
  75 void cpu_synchronize_all_states(void)
  76 {
  77     CPUState *cpu;
  78
  79     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
  80         cpu_synchronize_state(cpu);
  81     }
  82 }
  83
  84 void cpu_synchronize_all_post_reset(void)
  85 {
  86     CPUState *cpu;
  87
  88     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
  89         cpu_synchronize_post_reset(cpu);
  90     }
  91 }
  92
  93 void cpu_synchronize_all_post_init(void)
  94 {
  95     CPUState *cpu;
  96
  97     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
  98         cpu_synchronize_post_init(cpu);
  99     }
 100 }
 101
 102 int cpu_is_stopped(CPUState *env)
 103 {
 104     return !vm_running || env->stopped;
 105 }
 106
 107 static void do_vm_stop(int reason)
 108 {
 109     if (vm_running) {
 110         cpu_disable_ticks();
 111         vm_running = 0;
 112         pause_all_vcpus();
 113         vm_state_notify(0, reason);
 114         qemu_aio_flush();
 115         bdrv_flush_all();
 116         monitor_protocol_event(QEVENT_STOP, NULL);
 117     }
 118 }
 119
 120 static int cpu_can_run(CPUState *env)
 121 {
 122     if (env->stop)
 123         return 0;
 124     if (env->stopped || !vm_running)
 125         return 0;
 126     return 1;
 127 }
 128
 129 static int cpu_has_work(CPUState *env)
 130 {
 131     if (env->stop)
 132         return 1;
 133     if (env->queued_work_first)
 134         return 1;
 135     if (env->stopped || !vm_running)
 136         return 0;
 137     if (!env->halted)
 138         return 1;
 139     if (qemu_cpu_has_work(env))
 140         return 1;
 141     return 0;
 142 }
 143
 144 static int any_cpu_has_work(void)
 145 {
 146     CPUState *env;
 147
 148     for (env = first_cpu; env != NULL; env = env->next_cpu)
 149         if (cpu_has_work(env))
 150             return 1;
 151     return 0;
 152 }
 153
 154 static void cpu_debug_handler(CPUState *env)
 155 {
 156     gdb_set_stop_cpu(env);
 157     debug_requested = EXCP_DEBUG;
 158     vm_stop(EXCP_DEBUG);
 159 }
 160
 161 #ifndef _WIN32
 162 static int io_thread_fd = -1;
 163
 164 static void qemu_event_increment(void)
 165 {
 166     /* Write 8 bytes to be compatible with eventfd.  */
 167     static const uint64_t val = 1;
 168     ssize_t ret;
 169
 170     if (io_thread_fd == -1)
 171         return;
 172
 173     do {
 174         ret = write(io_thread_fd, &val, sizeof(val));
 175     } while (ret < 0 && errno == EINTR);
 176
 177     /* EAGAIN is fine, a read must be pending.  */
 178     if (ret < 0 && errno != EAGAIN) {
 179         fprintf(stderr, "qemu_event_increment: write() filed: %s\n",
 180                 strerror(errno));
 181         exit (1);
 182     }
 183 }
 184
 185 static void qemu_event_read(void *opaque)
 186 {
 187     int fd = (unsigned long)opaque;
 188     ssize_t len;
 189     char buffer[512];
 190
 191     /* Drain the notify pipe.  For eventfd, only 8 bytes will be read.  */
 192     do {
 193         len = read(fd, buffer, sizeof(buffer));
 194     } while ((len == -1 && errno == EINTR) || len == sizeof(buffer));
 195 }
 196
 197 static int qemu_event_init(void)
 198 {
 199     int err;
 200     int fds[2];
 201
 202     err = qemu_eventfd(fds);
 203     if (err == -1)
 204         return -errno;
 205
 206     err = fcntl_setfl(fds[0], O_NONBLOCK);
 207     if (err < 0)
 208         goto fail;
 209
 210     err = fcntl_setfl(fds[1], O_NONBLOCK);
 211     if (err < 0)
 212         goto fail;
 213
 214     qemu_set_fd_handler2(fds[0], NULL, qemu_event_read, NULL,
 215                          (void *)(unsigned long)fds[0]);
 216
 217     io_thread_fd = fds[1];
 218     return 0;
 219
 220 fail:
 221     close(fds[0]);
 222     close(fds[1]);
 223     return err;
 224 }
 225 #else
 226 HANDLE qemu_event_handle;
 227
 228 static void dummy_event_handler(void *opaque)
 229 {
 230 }
 231
 232 static int qemu_event_init(void)
 233 {
 234     qemu_event_handle = CreateEvent(NULL, FALSE, FALSE, NULL);
 235     if (!qemu_event_handle) {
 236         fprintf(stderr, "Failed CreateEvent: %ld\n", GetLastError());
 237         return -1;
 238     }
 239     qemu_add_wait_object(qemu_event_handle, dummy_event_handler, NULL);
 240     return 0;
 241 }
 242
 243 static void qemu_event_increment(void)
 244 {
 245     if (!SetEvent(qemu_event_handle)) {
 246         fprintf(stderr, "qemu_event_increment: SetEvent failed: %ld\n",
 247                 GetLastError());
 248         exit (1);
 249     }
 250 }
 251 #endif
 252
 253 #ifndef CONFIG_IOTHREAD
 254 int qemu_init_main_loop(void)
 255 {
 256     cpu_set_debug_excp_handler(cpu_debug_handler);
 257
 258     return qemu_event_init();
 259 }
 260
 261 void qemu_main_loop_start(void)
 262 {
 263 }
 264
 265 void qemu_init_vcpu(void *_env)
 266 {
 267     CPUState *env = _env;
 268
 269     env->nr_cores = smp_cores;
 270     env->nr_threads = smp_threads;
 271     if (kvm_enabled())
 272         kvm_init_vcpu(env);
 273     return;
 274 }
 275
 276 int qemu_cpu_self(void *env)
 277 {
 278     return 1;
 279 }
 280
 281 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
 282 {
 283     func(data);
 284 }
 285
 286 void resume_all_vcpus(void)
 287 {
 288 }
 289
 290 void pause_all_vcpus(void)
 291 {
 292 }
 293
 294 void qemu_cpu_kick(void *env)
 295 {
 296     return;
 297 }
 298
 299 void qemu_notify_event(void)
 300 {
 301     CPUState *env = cpu_single_env;
 302
 303     qemu_event_increment ();
 304     if (env) {
 305         cpu_exit(env);
 306     }
 307     if (next_cpu && env != next_cpu) {
 308         cpu_exit(next_cpu);
 309     }
 310 }
 311
 312 void qemu_mutex_lock_iothread(void) {}
 313 void qemu_mutex_unlock_iothread(void) {}
 314
 315 void cpu_stop_current(void)
 316 {
 317 }
 318
 319 void vm_stop(int reason)
 320 {
 321     do_vm_stop(reason);
 322 }
 323
 324 #else /* CONFIG_IOTHREAD */
 325
 326 #include "qemu-thread.h"
 327
 328 QemuMutex qemu_global_mutex;
 329 static QemuMutex qemu_fair_mutex;
 330
 331 static QemuThread io_thread;
 332
 333 static QemuThread *tcg_cpu_thread;
 334 static QemuCond *tcg_halt_cond;
 335
 336 static int qemu_system_ready;
 337 /* cpu creation */
 338 static QemuCond qemu_cpu_cond;
 339 /* system init */
 340 static QemuCond qemu_system_cond;
 341 static QemuCond qemu_pause_cond;
 342 static QemuCond qemu_work_cond;
 343
 344 static void tcg_init_ipi(void);
 345 static void kvm_init_ipi(CPUState *env);
 346 static sigset_t block_io_signals(void);
 347
 348 /* If we have signalfd, we mask out the signals we want to handle and then
 349  * use signalfd to listen for them.  We rely on whatever the current signal
 350  * handler is to dispatch the signals when we receive them.
 351  */
 352 static void sigfd_handler(void *opaque)
 353 {
 354     int fd = (unsigned long) opaque;
 355     struct qemu_signalfd_siginfo info;
 356     struct sigaction action;
 357     ssize_t len;
 358
 359     while (1) {
 360         do {
 361             len = read(fd, &info, sizeof(info));
 362         } while (len == -1 && errno == EINTR);
 363
 364         if (len == -1 && errno == EAGAIN) {
 365             break;
 366         }
 367
 368         if (len != sizeof(info)) {
 369             printf("read from sigfd returned %zd: %m\n", len);
 370             return;
 371         }
 372
 373         sigaction(info.ssi_signo, NULL, &action);
 374         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
 375             action.sa_sigaction(info.ssi_signo,
 376                                 (siginfo_t *)&info, NULL);
 377         } else if (action.sa_handler) {
 378             action.sa_handler(info.ssi_signo);
 379         }
 380     }
 381 }
 382
 383 static int qemu_signalfd_init(sigset_t mask)
 384 {
 385     int sigfd;
 386
 387     sigfd = qemu_signalfd(&mask);
 388     if (sigfd == -1) {
 389         fprintf(stderr, "failed to create signalfd\n");
 390         return -errno;
 391     }
 392
 393     fcntl_setfl(sigfd, O_NONBLOCK);
 394
 395     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
 396                          (void *)(unsigned long) sigfd);
 397
 398     return 0;
 399 }
 400
 401 int qemu_init_main_loop(void)
 402 {
 403     int ret;
 404     sigset_t blocked_signals;
 405
 406     cpu_set_debug_excp_handler(cpu_debug_handler);
 407
 408     blocked_signals = block_io_signals();
 409
 410     ret = qemu_signalfd_init(blocked_signals);
 411     if (ret)
 412         return ret;
 413
 414     /* Note eventfd must be drained before signalfd handlers run */
 415     ret = qemu_event_init();
 416     if (ret)
 417         return ret;
 418
 419     qemu_cond_init(&qemu_pause_cond);
 420     qemu_cond_init(&qemu_system_cond);
 421     qemu_mutex_init(&qemu_fair_mutex);
 422     qemu_mutex_init(&qemu_global_mutex);
 423     qemu_mutex_lock(&qemu_global_mutex);
 424
 425     qemu_thread_self(&io_thread);
 426
 427     return 0;
 428 }
 429
 430 void qemu_main_loop_start(void)
 431 {
 432     qemu_system_ready = 1;
 433     qemu_cond_broadcast(&qemu_system_cond);
 434 }
 435
 436 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
 437 {
 438     struct qemu_work_item wi;
 439
 440     if (qemu_cpu_self(env)) {
 441         func(data);
 442         return;
 443     }
 444
 445     wi.func = func;
 446     wi.data = data;
 447     if (!env->queued_work_first)
 448         env->queued_work_first = &wi;
 449     else
 450         env->queued_work_last->next = &wi;
 451     env->queued_work_last = &wi;
 452     wi.next = NULL;
 453     wi.done = false;
 454
 455     qemu_cpu_kick(env);
 456     while (!wi.done) {
 457         CPUState *self_env = cpu_single_env;
 458
 459         qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
 460         cpu_single_env = self_env;
 461     }
 462 }
 463
 464 static void flush_queued_work(CPUState *env)
 465 {
 466     struct qemu_work_item *wi;
 467
 468     if (!env->queued_work_first)
 469         return;
 470
 471     while ((wi = env->queued_work_first)) {
 472         env->queued_work_first = wi->next;
 473         wi->func(wi->data);
 474         wi->done = true;
 475     }
 476     env->queued_work_last = NULL;
 477     qemu_cond_broadcast(&qemu_work_cond);
 478 }
 479
 480 static void qemu_wait_io_event_common(CPUState *env)
 481 {
 482     if (env->stop) {
 483         env->stop = 0;
 484         env->stopped = 1;
 485         qemu_cond_signal(&qemu_pause_cond);
 486     }
 487     flush_queued_work(env);
 488     env->thread_kicked = false;
 489 }
 490
 491 static void qemu_tcg_wait_io_event(void)
 492 {
 493     CPUState *env;
 494
 495     while (!any_cpu_has_work())
 496         qemu_cond_timedwait(tcg_halt_cond, &qemu_global_mutex, 1000);
 497
 498     qemu_mutex_unlock(&qemu_global_mutex);
 499
 500     /*
 501      * Users of qemu_global_mutex can be starved, having no chance
 502      * to acquire it since this path will get to it first.
 503      * So use another lock to provide fairness.
 504      */
 505     qemu_mutex_lock(&qemu_fair_mutex);
 506     qemu_mutex_unlock(&qemu_fair_mutex);
 507
 508     qemu_mutex_lock(&qemu_global_mutex);
 509
 510     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 511         qemu_wait_io_event_common(env);
 512     }
 513 }
 514
 515 static void sigbus_reraise(void)
 516 {
 517     sigset_t set;
 518     struct sigaction action;
 519
 520     memset(&action, 0, sizeof(action));
 521     action.sa_handler = SIG_DFL;
 522     if (!sigaction(SIGBUS, &action, NULL)) {
 523         raise(SIGBUS);
 524         sigemptyset(&set);
 525         sigaddset(&set, SIGBUS);
 526         sigprocmask(SIG_UNBLOCK, &set, NULL);
 527     }
 528     perror("Failed to re-raise SIGBUS!\n");
 529     abort();
 530 }
 531
 532 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
 533                            void *ctx)
 534 {
 535 #if defined(TARGET_I386)
 536     if (kvm_on_sigbus(siginfo->ssi_code, (void *)(intptr_t)siginfo->ssi_addr))
 537 #endif
 538         sigbus_reraise();
 539 }
 540
 541 static void qemu_kvm_eat_signal(CPUState *env, int timeout)
 542 {
 543     struct timespec ts;
 544     int r, e;
 545     siginfo_t siginfo;
 546     sigset_t waitset;
 547     sigset_t chkset;
 548
 549     ts.tv_sec = timeout / 1000;
 550     ts.tv_nsec = (timeout % 1000) * 1000000;
 551
 552     sigemptyset(&waitset);
 553     sigaddset(&waitset, SIG_IPI);
 554     sigaddset(&waitset, SIGBUS);
 555
 556     do {
 557         qemu_mutex_unlock(&qemu_global_mutex);
 558
 559         r = sigtimedwait(&waitset, &siginfo, &ts);
 560         e = errno;
 561
 562         qemu_mutex_lock(&qemu_global_mutex);
 563
 564         if (r == -1 && !(e == EAGAIN || e == EINTR)) {
 565             fprintf(stderr, "sigtimedwait: %s\n", strerror(e));
 566             exit(1);
 567         }
 568
 569         switch (r) {
 570         case SIGBUS:
 571 #ifdef TARGET_I386
 572             if (kvm_on_sigbus_vcpu(env, siginfo.si_code, siginfo.si_addr))
 573 #endif
 574                 sigbus_reraise();
 575             break;
 576         default:
 577             break;
 578         }
 579
 580         r = sigpending(&chkset);
 581         if (r == -1) {
 582             fprintf(stderr, "sigpending: %s\n", strerror(e));
 583             exit(1);
 584         }
 585     } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
 586 }
 587
 588 static void qemu_kvm_wait_io_event(CPUState *env)
 589 {
 590     while (!cpu_has_work(env))
 591         qemu_cond_timedwait(env->halt_cond, &qemu_global_mutex, 1000);
 592
 593     qemu_kvm_eat_signal(env, 0);
 594     qemu_wait_io_event_common(env);
 595 }
 596
 597 static int qemu_cpu_exec(CPUState *env);
 598
 599 static void *kvm_cpu_thread_fn(void *arg)
 600 {
 601     CPUState *env = arg;
 602
 603     qemu_mutex_lock(&qemu_global_mutex);
 604     qemu_thread_self(env->thread);
 605     if (kvm_enabled())
 606         kvm_init_vcpu(env);
 607
 608     kvm_init_ipi(env);
 609
 610     /* signal CPU creation */
 611     env->created = 1;
 612     qemu_cond_signal(&qemu_cpu_cond);
 613
 614     /* and wait for machine initialization */
 615     while (!qemu_system_ready)
 616         qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
 617
 618     while (1) {
 619         if (cpu_can_run(env))
 620             qemu_cpu_exec(env);
 621         qemu_kvm_wait_io_event(env);
 622     }
 623
 624     return NULL;
 625 }
 626
 627 static void *tcg_cpu_thread_fn(void *arg)
 628 {
 629     CPUState *env = arg;
 630
 631     tcg_init_ipi();
 632     qemu_thread_self(env->thread);
 633
 634     /* signal CPU creation */
 635     qemu_mutex_lock(&qemu_global_mutex);
 636     for (env = first_cpu; env != NULL; env = env->next_cpu)
 637         env->created = 1;
 638     qemu_cond_signal(&qemu_cpu_cond);
 639
 640     /* and wait for machine initialization */
 641     while (!qemu_system_ready)
 642         qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
 643
 644     while (1) {
 645         cpu_exec_all();
 646         qemu_tcg_wait_io_event();
 647     }
 648
 649     return NULL;
 650 }
 651
 652 void qemu_cpu_kick(void *_env)
 653 {
 654     CPUState *env = _env;
 655     qemu_cond_broadcast(env->halt_cond);
 656     if (!env->thread_kicked) {
 657         qemu_thread_signal(env->thread, SIG_IPI);
 658         env->thread_kicked = true;
 659     }
 660 }
 661
 662 int qemu_cpu_self(void *_env)
 663 {
 664     CPUState *env = _env;
 665     QemuThread this;
 666
 667     qemu_thread_self(&this);
 668
 669     return qemu_thread_equal(&this, env->thread);
 670 }
 671
 672 static void cpu_signal(int sig)
 673 {
 674     if (cpu_single_env)
 675         cpu_exit(cpu_single_env);
 676     exit_request = 1;
 677 }
 678
 679 static void tcg_init_ipi(void)
 680 {
 681     sigset_t set;
 682     struct sigaction sigact;
 683
 684     memset(&sigact, 0, sizeof(sigact));
 685     sigact.sa_handler = cpu_signal;
 686     sigaction(SIG_IPI, &sigact, NULL);
 687
 688     sigemptyset(&set);
 689     sigaddset(&set, SIG_IPI);
 690     pthread_sigmask(SIG_UNBLOCK, &set, NULL);
 691 }
 692
 693 static void dummy_signal(int sig)
 694 {
 695 }
 696
 697 static void kvm_init_ipi(CPUState *env)
 698 {
 699     int r;
 700     sigset_t set;
 701     struct sigaction sigact;
 702
 703     memset(&sigact, 0, sizeof(sigact));
 704     sigact.sa_handler = dummy_signal;
 705     sigaction(SIG_IPI, &sigact, NULL);
 706
 707     pthread_sigmask(SIG_BLOCK, NULL, &set);
 708     sigdelset(&set, SIG_IPI);
 709     sigdelset(&set, SIGBUS);
 710     r = kvm_set_signal_mask(env, &set);
 711     if (r) {
 712         fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(r));
 713         exit(1);
 714     }
 715 }
 716
 717 static sigset_t block_io_signals(void)
 718 {
 719     sigset_t set;
 720     struct sigaction action;
 721
 722     /* SIGUSR2 used by posix-aio-compat.c */
 723     sigemptyset(&set);
 724     sigaddset(&set, SIGUSR2);
 725     pthread_sigmask(SIG_UNBLOCK, &set, NULL);
 726
 727     sigemptyset(&set);
 728     sigaddset(&set, SIGIO);
 729     sigaddset(&set, SIGALRM);
 730     sigaddset(&set, SIG_IPI);
 731     sigaddset(&set, SIGBUS);
 732     pthread_sigmask(SIG_BLOCK, &set, NULL);
 733
 734     memset(&action, 0, sizeof(action));
 735     action.sa_flags = SA_SIGINFO;
 736     action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
 737     sigaction(SIGBUS, &action, NULL);
 738     prctl(PR_MCE_KILL, 1, 1, 0, 0);
 739
 740     return set;
 741 }
 742
 743 void qemu_mutex_lock_iothread(void)
 744 {
 745     if (kvm_enabled()) {
 746         qemu_mutex_lock(&qemu_global_mutex);
 747     } else {
 748         qemu_mutex_lock(&qemu_fair_mutex);
 749         if (qemu_mutex_trylock(&qemu_global_mutex)) {
 750             qemu_thread_signal(tcg_cpu_thread, SIG_IPI);
 751             qemu_mutex_lock(&qemu_global_mutex);
 752         }
 753         qemu_mutex_unlock(&qemu_fair_mutex);
 754     }
 755 }
 756
 757 void qemu_mutex_unlock_iothread(void)
 758 {
 759     qemu_mutex_unlock(&qemu_global_mutex);
 760 }
 761
 762 static int all_vcpus_paused(void)
 763 {
 764     CPUState *penv = first_cpu;
 765
 766     while (penv) {
 767         if (!penv->stopped)
 768             return 0;
 769         penv = (CPUState *)penv->next_cpu;
 770     }
 771
 772     return 1;
 773 }
 774
 775 void pause_all_vcpus(void)
 776 {
 777     CPUState *penv = first_cpu;
 778
 779     while (penv) {
 780         penv->stop = 1;
 781         qemu_cpu_kick(penv);
 782         penv = (CPUState *)penv->next_cpu;
 783     }
 784
 785     while (!all_vcpus_paused()) {
 786         qemu_cond_timedwait(&qemu_pause_cond, &qemu_global_mutex, 100);
 787         penv = first_cpu;
 788         while (penv) {
 789             qemu_cpu_kick(penv);
 790             penv = (CPUState *)penv->next_cpu;
 791         }
 792     }
 793 }
 794
 795 void resume_all_vcpus(void)
 796 {
 797     CPUState *penv = first_cpu;
 798
 799     while (penv) {
 800         penv->stop = 0;
 801         penv->stopped = 0;
 802         qemu_cpu_kick(penv);
 803         penv = (CPUState *)penv->next_cpu;
 804     }
 805 }
 806
 807 static void tcg_init_vcpu(void *_env)
 808 {
 809     CPUState *env = _env;
 810     /* share a single thread for all cpus with TCG */
 811     if (!tcg_cpu_thread) {
 812         env->thread = qemu_mallocz(sizeof(QemuThread));
 813         env->halt_cond = qemu_mallocz(sizeof(QemuCond));
 814         qemu_cond_init(env->halt_cond);
 815         qemu_thread_create(env->thread, tcg_cpu_thread_fn, env);
 816         while (env->created == 0)
 817             qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
 818         tcg_cpu_thread = env->thread;
 819         tcg_halt_cond = env->halt_cond;
 820     } else {
 821         env->thread = tcg_cpu_thread;
 822         env->halt_cond = tcg_halt_cond;
 823     }
 824 }
 825
 826 static void kvm_start_vcpu(CPUState *env)
 827 {
 828     env->thread = qemu_mallocz(sizeof(QemuThread));
 829     env->halt_cond = qemu_mallocz(sizeof(QemuCond));
 830     qemu_cond_init(env->halt_cond);
 831     qemu_thread_create(env->thread, kvm_cpu_thread_fn, env);
 832     while (env->created == 0)
 833         qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
 834 }
 835
 836 void qemu_init_vcpu(void *_env)
 837 {
 838     CPUState *env = _env;
 839
 840     env->nr_cores = smp_cores;
 841     env->nr_threads = smp_threads;
 842     if (kvm_enabled())
 843         kvm_start_vcpu(env);
 844     else
 845         tcg_init_vcpu(env);
 846 }
 847
 848 void qemu_notify_event(void)
 849 {
 850     qemu_event_increment();
 851 }
 852
 853 static void qemu_system_vmstop_request(int reason)
 854 {
 855     vmstop_requested = reason;
 856     qemu_notify_event();
 857 }
 858
 859 void cpu_stop_current(void)
 860 {
 861     if (cpu_single_env) {
 862         cpu_single_env->stopped = 1;
 863         cpu_exit(cpu_single_env);
 864     }
 865 }
 866
 867 void vm_stop(int reason)
 868 {
 869     QemuThread me;
 870     qemu_thread_self(&me);
 871
 872     if (!qemu_thread_equal(&me, &io_thread)) {
 873         qemu_system_vmstop_request(reason);
 874         /*
 875          * FIXME: should not return to device code in case
 876          * vm_stop() has been requested.
 877          */
 878         cpu_stop_current();
 879         return;
 880     }
 881     do_vm_stop(reason);
 882 }
 883
 884 #endif
 885
 886 static int qemu_cpu_exec(CPUState *env)
 887 {
 888     int ret;
 889 #ifdef CONFIG_PROFILER
 890     int64_t ti;
 891 #endif
 892
 893 #ifdef CONFIG_PROFILER
 894     ti = profile_getclock();
 895 #endif
 896     if (use_icount) {
 897         int64_t count;
 898         int decr;
 899         qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
 900         env->icount_decr.u16.low = 0;
 901         env->icount_extra = 0;
 902         count = qemu_icount_round (qemu_next_deadline());
 903         qemu_icount += count;
 904         decr = (count > 0xffff) ? 0xffff : count;
 905         count -= decr;
 906         env->icount_decr.u16.low = decr;
 907         env->icount_extra = count;
 908     }
 909     ret = cpu_exec(env);
 910 #ifdef CONFIG_PROFILER
 911     qemu_time += profile_getclock() - ti;
 912 #endif
 913     if (use_icount) {
 914         /* Fold pending instructions back into the
 915            instruction counter, and clear the interrupt flag.  */
 916         qemu_icount -= (env->icount_decr.u16.low
 917                         + env->icount_extra);
 918         env->icount_decr.u32 = 0;
 919         env->icount_extra = 0;
 920     }
 921     return ret;
 922 }
 923
 924 bool cpu_exec_all(void)
 925 {
 926     if (next_cpu == NULL)
 927         next_cpu = first_cpu;
 928     for (; next_cpu != NULL && !exit_request; next_cpu = next_cpu->next_cpu) {
 929         CPUState *env = next_cpu;
 930
 931         qemu_clock_enable(vm_clock,
 932                           (env->singlestep_enabled & SSTEP_NOTIMER) == 0);
 933
 934         if (qemu_alarm_pending())
 935             break;
 936         if (cpu_can_run(env)) {
 937             if (qemu_cpu_exec(env) == EXCP_DEBUG) {
 938                 break;
 939             }
 940         } else if (env->stop) {
 941             break;
 942         }
 943     }
 944     exit_request = 0;
 945     return any_cpu_has_work();
 946 }
 947
 948 void set_numa_modes(void)
 949 {
 950     CPUState *env;
 951     int i;
 952
 953     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 954         for (i = 0; i < nb_numa_nodes; i++) {
 955             if (node_cpumask[i] & (1 << env->cpu_index)) {
 956                 env->numa_node = i;
 957             }
 958         }
 959     }
 960 }
 961
 962 void set_cpu_log(const char *optarg)
 963 {
 964     int mask;
 965     const CPULogItem *item;
 966
 967     mask = cpu_str_to_log_mask(optarg);
 968     if (!mask) {
 969         printf("Log items (comma separated):\n");
 970         for (item = cpu_log_items; item->mask != 0; item++) {
 971             printf("%-10s %s\n", item->name, item->help);
 972         }
 973         exit(1);
 974     }
 975     cpu_set_log(mask);
 976 }
 977
 978 /* Return the virtual CPU time, based on the instruction counter.  */
 979 int64_t cpu_get_icount(void)
 980 {
 981     int64_t icount;
 982     CPUState *env = cpu_single_env;;
 983
 984     icount = qemu_icount;
 985     if (env) {
 986         if (!can_do_io(env)) {
 987             fprintf(stderr, "Bad clock read\n");
 988         }
 989         icount -= (env->icount_decr.u16.low + env->icount_extra);
 990     }
 991     return qemu_icount_bias + (icount << icount_time_shift);
 992 }
 993
 994 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
 995 {
 996     /* XXX: implement xxx_cpu_list for targets that still miss it */
 997 #if defined(cpu_list_id)
 998     cpu_list_id(f, cpu_fprintf, optarg);
 999 #elif defined(cpu_list)
1000     cpu_list(f, cpu_fprintf); /* deprecated */
1001 #endif
1002 }