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 vm_stop(int reason)
 316 {
 317     do_vm_stop(reason);
 318 }
 319
 320 #else /* CONFIG_IOTHREAD */
 321
 322 #include "qemu-thread.h"
 323
 324 QemuMutex qemu_global_mutex;
 325 static QemuMutex qemu_fair_mutex;
 326
 327 static QemuThread io_thread;
 328
 329 static QemuThread *tcg_cpu_thread;
 330 static QemuCond *tcg_halt_cond;
 331
 332 static int qemu_system_ready;
 333 /* cpu creation */
 334 static QemuCond qemu_cpu_cond;
 335 /* system init */
 336 static QemuCond qemu_system_cond;
 337 static QemuCond qemu_pause_cond;
 338 static QemuCond qemu_work_cond;
 339
 340 static void tcg_init_ipi(void);
 341 static void kvm_init_ipi(CPUState *env);
 342 static sigset_t block_io_signals(void);
 343
 344 /* If we have signalfd, we mask out the signals we want to handle and then
 345  * use signalfd to listen for them.  We rely on whatever the current signal
 346  * handler is to dispatch the signals when we receive them.
 347  */
 348 static void sigfd_handler(void *opaque)
 349 {
 350     int fd = (unsigned long) opaque;
 351     struct qemu_signalfd_siginfo info;
 352     struct sigaction action;
 353     ssize_t len;
 354
 355     while (1) {
 356         do {
 357             len = read(fd, &info, sizeof(info));
 358         } while (len == -1 && errno == EINTR);
 359
 360         if (len == -1 && errno == EAGAIN) {
 361             break;
 362         }
 363
 364         if (len != sizeof(info)) {
 365             printf("read from sigfd returned %zd: %m\n", len);
 366             return;
 367         }
 368
 369         sigaction(info.ssi_signo, NULL, &action);
 370         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
 371             action.sa_sigaction(info.ssi_signo,
 372                                 (siginfo_t *)&info, NULL);
 373         } else if (action.sa_handler) {
 374             action.sa_handler(info.ssi_signo);
 375         }
 376     }
 377 }
 378
 379 static int qemu_signalfd_init(sigset_t mask)
 380 {
 381     int sigfd;
 382
 383     sigfd = qemu_signalfd(&mask);
 384     if (sigfd == -1) {
 385         fprintf(stderr, "failed to create signalfd\n");
 386         return -errno;
 387     }
 388
 389     fcntl_setfl(sigfd, O_NONBLOCK);
 390
 391     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
 392                          (void *)(unsigned long) sigfd);
 393
 394     return 0;
 395 }
 396
 397 int qemu_init_main_loop(void)
 398 {
 399     int ret;
 400     sigset_t blocked_signals;
 401
 402     cpu_set_debug_excp_handler(cpu_debug_handler);
 403
 404     blocked_signals = block_io_signals();
 405
 406     ret = qemu_signalfd_init(blocked_signals);
 407     if (ret)
 408         return ret;
 409
 410     /* Note eventfd must be drained before signalfd handlers run */
 411     ret = qemu_event_init();
 412     if (ret)
 413         return ret;
 414
 415     qemu_cond_init(&qemu_pause_cond);
 416     qemu_cond_init(&qemu_system_cond);
 417     qemu_mutex_init(&qemu_fair_mutex);
 418     qemu_mutex_init(&qemu_global_mutex);
 419     qemu_mutex_lock(&qemu_global_mutex);
 420
 421     qemu_thread_self(&io_thread);
 422
 423     return 0;
 424 }
 425
 426 void qemu_main_loop_start(void)
 427 {
 428     qemu_system_ready = 1;
 429     qemu_cond_broadcast(&qemu_system_cond);
 430 }
 431
 432 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
 433 {
 434     struct qemu_work_item wi;
 435
 436     if (qemu_cpu_self(env)) {
 437         func(data);
 438         return;
 439     }
 440
 441     wi.func = func;
 442     wi.data = data;
 443     if (!env->queued_work_first)
 444         env->queued_work_first = &wi;
 445     else
 446         env->queued_work_last->next = &wi;
 447     env->queued_work_last = &wi;
 448     wi.next = NULL;
 449     wi.done = false;
 450
 451     qemu_cpu_kick(env);
 452     while (!wi.done) {
 453         CPUState *self_env = cpu_single_env;
 454
 455         qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
 456         cpu_single_env = self_env;
 457     }
 458 }
 459
 460 static void flush_queued_work(CPUState *env)
 461 {
 462     struct qemu_work_item *wi;
 463
 464     if (!env->queued_work_first)
 465         return;
 466
 467     while ((wi = env->queued_work_first)) {
 468         env->queued_work_first = wi->next;
 469         wi->func(wi->data);
 470         wi->done = true;
 471     }
 472     env->queued_work_last = NULL;
 473     qemu_cond_broadcast(&qemu_work_cond);
 474 }
 475
 476 static void qemu_wait_io_event_common(CPUState *env)
 477 {
 478     if (env->stop) {
 479         env->stop = 0;
 480         env->stopped = 1;
 481         qemu_cond_signal(&qemu_pause_cond);
 482     }
 483     flush_queued_work(env);
 484     env->thread_kicked = false;
 485 }
 486
 487 static void qemu_tcg_wait_io_event(void)
 488 {
 489     CPUState *env;
 490
 491     while (!any_cpu_has_work())
 492         qemu_cond_timedwait(tcg_halt_cond, &qemu_global_mutex, 1000);
 493
 494     qemu_mutex_unlock(&qemu_global_mutex);
 495
 496     /*
 497      * Users of qemu_global_mutex can be starved, having no chance
 498      * to acquire it since this path will get to it first.
 499      * So use another lock to provide fairness.
 500      */
 501     qemu_mutex_lock(&qemu_fair_mutex);
 502     qemu_mutex_unlock(&qemu_fair_mutex);
 503
 504     qemu_mutex_lock(&qemu_global_mutex);
 505
 506     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 507         qemu_wait_io_event_common(env);
 508     }
 509 }
 510
 511 static void sigbus_reraise(void)
 512 {
 513     sigset_t set;
 514     struct sigaction action;
 515
 516     memset(&action, 0, sizeof(action));
 517     action.sa_handler = SIG_DFL;
 518     if (!sigaction(SIGBUS, &action, NULL)) {
 519         raise(SIGBUS);
 520         sigemptyset(&set);
 521         sigaddset(&set, SIGBUS);
 522         sigprocmask(SIG_UNBLOCK, &set, NULL);
 523     }
 524     perror("Failed to re-raise SIGBUS!\n");
 525     abort();
 526 }
 527
 528 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
 529                            void *ctx)
 530 {
 531 #if defined(TARGET_I386)
 532     if (kvm_on_sigbus(siginfo->ssi_code, (void *)(intptr_t)siginfo->ssi_addr))
 533 #endif
 534         sigbus_reraise();
 535 }
 536
 537 static void qemu_kvm_eat_signal(CPUState *env, int timeout)
 538 {
 539     struct timespec ts;
 540     int r, e;
 541     siginfo_t siginfo;
 542     sigset_t waitset;
 543     sigset_t chkset;
 544
 545     ts.tv_sec = timeout / 1000;
 546     ts.tv_nsec = (timeout % 1000) * 1000000;
 547
 548     sigemptyset(&waitset);
 549     sigaddset(&waitset, SIG_IPI);
 550     sigaddset(&waitset, SIGBUS);
 551
 552     do {
 553         qemu_mutex_unlock(&qemu_global_mutex);
 554
 555         r = sigtimedwait(&waitset, &siginfo, &ts);
 556         e = errno;
 557
 558         qemu_mutex_lock(&qemu_global_mutex);
 559
 560         if (r == -1 && !(e == EAGAIN || e == EINTR)) {
 561             fprintf(stderr, "sigtimedwait: %s\n", strerror(e));
 562             exit(1);
 563         }
 564
 565         switch (r) {
 566         case SIGBUS:
 567 #ifdef TARGET_I386
 568             if (kvm_on_sigbus_vcpu(env, siginfo.si_code, siginfo.si_addr))
 569 #endif
 570                 sigbus_reraise();
 571             break;
 572         default:
 573             break;
 574         }
 575
 576         r = sigpending(&chkset);
 577         if (r == -1) {
 578             fprintf(stderr, "sigpending: %s\n", strerror(e));
 579             exit(1);
 580         }
 581     } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
 582 }
 583
 584 static void qemu_kvm_wait_io_event(CPUState *env)
 585 {
 586     while (!cpu_has_work(env))
 587         qemu_cond_timedwait(env->halt_cond, &qemu_global_mutex, 1000);
 588
 589     qemu_kvm_eat_signal(env, 0);
 590     qemu_wait_io_event_common(env);
 591 }
 592
 593 static int qemu_cpu_exec(CPUState *env);
 594
 595 static void *kvm_cpu_thread_fn(void *arg)
 596 {
 597     CPUState *env = arg;
 598
 599     qemu_mutex_lock(&qemu_global_mutex);
 600     qemu_thread_self(env->thread);
 601     if (kvm_enabled())
 602         kvm_init_vcpu(env);
 603
 604     kvm_init_ipi(env);
 605
 606     /* signal CPU creation */
 607     env->created = 1;
 608     qemu_cond_signal(&qemu_cpu_cond);
 609
 610     /* and wait for machine initialization */
 611     while (!qemu_system_ready)
 612         qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
 613
 614     while (1) {
 615         if (cpu_can_run(env))
 616             qemu_cpu_exec(env);
 617         qemu_kvm_wait_io_event(env);
 618     }
 619
 620     return NULL;
 621 }
 622
 623 static void *tcg_cpu_thread_fn(void *arg)
 624 {
 625     CPUState *env = arg;
 626
 627     tcg_init_ipi();
 628     qemu_thread_self(env->thread);
 629
 630     /* signal CPU creation */
 631     qemu_mutex_lock(&qemu_global_mutex);
 632     for (env = first_cpu; env != NULL; env = env->next_cpu)
 633         env->created = 1;
 634     qemu_cond_signal(&qemu_cpu_cond);
 635
 636     /* and wait for machine initialization */
 637     while (!qemu_system_ready)
 638         qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
 639
 640     while (1) {
 641         cpu_exec_all();
 642         qemu_tcg_wait_io_event();
 643     }
 644
 645     return NULL;
 646 }
 647
 648 void qemu_cpu_kick(void *_env)
 649 {
 650     CPUState *env = _env;
 651     qemu_cond_broadcast(env->halt_cond);
 652     if (!env->thread_kicked) {
 653         qemu_thread_signal(env->thread, SIG_IPI);
 654         env->thread_kicked = true;
 655     }
 656 }
 657
 658 int qemu_cpu_self(void *_env)
 659 {
 660     CPUState *env = _env;
 661     QemuThread this;
 662
 663     qemu_thread_self(&this);
 664
 665     return qemu_thread_equal(&this, env->thread);
 666 }
 667
 668 static void cpu_signal(int sig)
 669 {
 670     if (cpu_single_env)
 671         cpu_exit(cpu_single_env);
 672     exit_request = 1;
 673 }
 674
 675 static void tcg_init_ipi(void)
 676 {
 677     sigset_t set;
 678     struct sigaction sigact;
 679
 680     memset(&sigact, 0, sizeof(sigact));
 681     sigact.sa_handler = cpu_signal;
 682     sigaction(SIG_IPI, &sigact, NULL);
 683
 684     sigemptyset(&set);
 685     sigaddset(&set, SIG_IPI);
 686     pthread_sigmask(SIG_UNBLOCK, &set, NULL);
 687 }
 688
 689 static void dummy_signal(int sig)
 690 {
 691 }
 692
 693 static void kvm_init_ipi(CPUState *env)
 694 {
 695     int r;
 696     sigset_t set;
 697     struct sigaction sigact;
 698
 699     memset(&sigact, 0, sizeof(sigact));
 700     sigact.sa_handler = dummy_signal;
 701     sigaction(SIG_IPI, &sigact, NULL);
 702
 703     pthread_sigmask(SIG_BLOCK, NULL, &set);
 704     sigdelset(&set, SIG_IPI);
 705     sigdelset(&set, SIGBUS);
 706     r = kvm_set_signal_mask(env, &set);
 707     if (r) {
 708         fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(r));
 709         exit(1);
 710     }
 711 }
 712
 713 static sigset_t block_io_signals(void)
 714 {
 715     sigset_t set;
 716     struct sigaction action;
 717
 718     /* SIGUSR2 used by posix-aio-compat.c */
 719     sigemptyset(&set);
 720     sigaddset(&set, SIGUSR2);
 721     pthread_sigmask(SIG_UNBLOCK, &set, NULL);
 722
 723     sigemptyset(&set);
 724     sigaddset(&set, SIGIO);
 725     sigaddset(&set, SIGALRM);
 726     sigaddset(&set, SIG_IPI);
 727     sigaddset(&set, SIGBUS);
 728     pthread_sigmask(SIG_BLOCK, &set, NULL);
 729
 730     memset(&action, 0, sizeof(action));
 731     action.sa_flags = SA_SIGINFO;
 732     action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
 733     sigaction(SIGBUS, &action, NULL);
 734     prctl(PR_MCE_KILL, 1, 1, 0, 0);
 735
 736     return set;
 737 }
 738
 739 void qemu_mutex_lock_iothread(void)
 740 {
 741     if (kvm_enabled()) {
 742         qemu_mutex_lock(&qemu_global_mutex);
 743     } else {
 744         qemu_mutex_lock(&qemu_fair_mutex);
 745         if (qemu_mutex_trylock(&qemu_global_mutex)) {
 746             qemu_thread_signal(tcg_cpu_thread, SIG_IPI);
 747             qemu_mutex_lock(&qemu_global_mutex);
 748         }
 749         qemu_mutex_unlock(&qemu_fair_mutex);
 750     }
 751 }
 752
 753 void qemu_mutex_unlock_iothread(void)
 754 {
 755     qemu_mutex_unlock(&qemu_global_mutex);
 756 }
 757
 758 static int all_vcpus_paused(void)
 759 {
 760     CPUState *penv = first_cpu;
 761
 762     while (penv) {
 763         if (!penv->stopped)
 764             return 0;
 765         penv = (CPUState *)penv->next_cpu;
 766     }
 767
 768     return 1;
 769 }
 770
 771 void pause_all_vcpus(void)
 772 {
 773     CPUState *penv = first_cpu;
 774
 775     while (penv) {
 776         penv->stop = 1;
 777         qemu_cpu_kick(penv);
 778         penv = (CPUState *)penv->next_cpu;
 779     }
 780
 781     while (!all_vcpus_paused()) {
 782         qemu_cond_timedwait(&qemu_pause_cond, &qemu_global_mutex, 100);
 783         penv = first_cpu;
 784         while (penv) {
 785             qemu_cpu_kick(penv);
 786             penv = (CPUState *)penv->next_cpu;
 787         }
 788     }
 789 }
 790
 791 void resume_all_vcpus(void)
 792 {
 793     CPUState *penv = first_cpu;
 794
 795     while (penv) {
 796         penv->stop = 0;
 797         penv->stopped = 0;
 798         qemu_cpu_kick(penv);
 799         penv = (CPUState *)penv->next_cpu;
 800     }
 801 }
 802
 803 static void tcg_init_vcpu(void *_env)
 804 {
 805     CPUState *env = _env;
 806     /* share a single thread for all cpus with TCG */
 807     if (!tcg_cpu_thread) {
 808         env->thread = qemu_mallocz(sizeof(QemuThread));
 809         env->halt_cond = qemu_mallocz(sizeof(QemuCond));
 810         qemu_cond_init(env->halt_cond);
 811         qemu_thread_create(env->thread, tcg_cpu_thread_fn, env);
 812         while (env->created == 0)
 813             qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
 814         tcg_cpu_thread = env->thread;
 815         tcg_halt_cond = env->halt_cond;
 816     } else {
 817         env->thread = tcg_cpu_thread;
 818         env->halt_cond = tcg_halt_cond;
 819     }
 820 }
 821
 822 static void kvm_start_vcpu(CPUState *env)
 823 {
 824     env->thread = qemu_mallocz(sizeof(QemuThread));
 825     env->halt_cond = qemu_mallocz(sizeof(QemuCond));
 826     qemu_cond_init(env->halt_cond);
 827     qemu_thread_create(env->thread, kvm_cpu_thread_fn, env);
 828     while (env->created == 0)
 829         qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
 830 }
 831
 832 void qemu_init_vcpu(void *_env)
 833 {
 834     CPUState *env = _env;
 835
 836     env->nr_cores = smp_cores;
 837     env->nr_threads = smp_threads;
 838     if (kvm_enabled())
 839         kvm_start_vcpu(env);
 840     else
 841         tcg_init_vcpu(env);
 842 }
 843
 844 void qemu_notify_event(void)
 845 {
 846     qemu_event_increment();
 847 }
 848
 849 static void qemu_system_vmstop_request(int reason)
 850 {
 851     vmstop_requested = reason;
 852     qemu_notify_event();
 853 }
 854
 855 void vm_stop(int reason)
 856 {
 857     QemuThread me;
 858     qemu_thread_self(&me);
 859
 860     if (!qemu_thread_equal(&me, &io_thread)) {
 861         qemu_system_vmstop_request(reason);
 862         /*
 863          * FIXME: should not return to device code in case
 864          * vm_stop() has been requested.
 865          */
 866         if (cpu_single_env) {
 867             cpu_exit(cpu_single_env);
 868             cpu_single_env->stop = 1;
 869         }
 870         return;
 871     }
 872     do_vm_stop(reason);
 873 }
 874
 875 #endif
 876
 877 static int qemu_cpu_exec(CPUState *env)
 878 {
 879     int ret;
 880 #ifdef CONFIG_PROFILER
 881     int64_t ti;
 882 #endif
 883
 884 #ifdef CONFIG_PROFILER
 885     ti = profile_getclock();
 886 #endif
 887     if (use_icount) {
 888         int64_t count;
 889         int decr;
 890         qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
 891         env->icount_decr.u16.low = 0;
 892         env->icount_extra = 0;
 893         count = qemu_icount_round (qemu_next_deadline());
 894         qemu_icount += count;
 895         decr = (count > 0xffff) ? 0xffff : count;
 896         count -= decr;
 897         env->icount_decr.u16.low = decr;
 898         env->icount_extra = count;
 899     }
 900     ret = cpu_exec(env);
 901 #ifdef CONFIG_PROFILER
 902     qemu_time += profile_getclock() - ti;
 903 #endif
 904     if (use_icount) {
 905         /* Fold pending instructions back into the
 906            instruction counter, and clear the interrupt flag.  */
 907         qemu_icount -= (env->icount_decr.u16.low
 908                         + env->icount_extra);
 909         env->icount_decr.u32 = 0;
 910         env->icount_extra = 0;
 911     }
 912     return ret;
 913 }
 914
 915 bool cpu_exec_all(void)
 916 {
 917     if (next_cpu == NULL)
 918         next_cpu = first_cpu;
 919     for (; next_cpu != NULL && !exit_request; next_cpu = next_cpu->next_cpu) {
 920         CPUState *env = next_cpu;
 921
 922         qemu_clock_enable(vm_clock,
 923                           (env->singlestep_enabled & SSTEP_NOTIMER) == 0);
 924
 925         if (qemu_alarm_pending())
 926             break;
 927         if (cpu_can_run(env)) {
 928             if (qemu_cpu_exec(env) == EXCP_DEBUG) {
 929                 break;
 930             }
 931         } else if (env->stop) {
 932             break;
 933         }
 934     }
 935     exit_request = 0;
 936     return any_cpu_has_work();
 937 }
 938
 939 void set_numa_modes(void)
 940 {
 941     CPUState *env;
 942     int i;
 943
 944     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 945         for (i = 0; i < nb_numa_nodes; i++) {
 946             if (node_cpumask[i] & (1 << env->cpu_index)) {
 947                 env->numa_node = i;
 948             }
 949         }
 950     }
 951 }
 952
 953 void set_cpu_log(const char *optarg)
 954 {
 955     int mask;
 956     const CPULogItem *item;
 957
 958     mask = cpu_str_to_log_mask(optarg);
 959     if (!mask) {
 960         printf("Log items (comma separated):\n");
 961         for (item = cpu_log_items; item->mask != 0; item++) {
 962             printf("%-10s %s\n", item->name, item->help);
 963         }
 964         exit(1);
 965     }
 966     cpu_set_log(mask);
 967 }
 968
 969 /* Return the virtual CPU time, based on the instruction counter.  */
 970 int64_t cpu_get_icount(void)
 971 {
 972     int64_t icount;
 973     CPUState *env = cpu_single_env;;
 974
 975     icount = qemu_icount;
 976     if (env) {
 977         if (!can_do_io(env)) {
 978             fprintf(stderr, "Bad clock read\n");
 979         }
 980         icount -= (env->icount_decr.u16.low + env->icount_extra);
 981     }
 982     return qemu_icount_bias + (icount << icount_time_shift);
 983 }
 984
 985 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
 986 {
 987     /* XXX: implement xxx_cpu_list for targets that still miss it */
 988 #if defined(cpu_list_id)
 989     cpu_list_id(f, cpu_fprintf, optarg);
 990 #elif defined(cpu_list)
 991     cpu_list(f, cpu_fprintf); /* deprecated */
 992 #endif
 993 }