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     exit_request = 1;
 311 }
 312
 313 void qemu_mutex_lock_iothread(void) {}
 314 void qemu_mutex_unlock_iothread(void) {}
 315
 316 void cpu_stop_current(void)
 317 {
 318 }
 319
 320 void vm_stop(int reason)
 321 {
 322     do_vm_stop(reason);
 323 }
 324
 325 #else /* CONFIG_IOTHREAD */
 326
 327 #include "qemu-thread.h"
 328
 329 QemuMutex qemu_global_mutex;
 330 static QemuMutex qemu_fair_mutex;
 331
 332 static QemuThread io_thread;
 333
 334 static QemuThread *tcg_cpu_thread;
 335 static QemuCond *tcg_halt_cond;
 336
 337 static int qemu_system_ready;
 338 /* cpu creation */
 339 static QemuCond qemu_cpu_cond;
 340 /* system init */
 341 static QemuCond qemu_system_cond;
 342 static QemuCond qemu_pause_cond;
 343 static QemuCond qemu_work_cond;
 344
 345 static void tcg_init_ipi(void);
 346 static void kvm_init_ipi(CPUState *env);
 347 static sigset_t block_io_signals(void);
 348
 349 /* If we have signalfd, we mask out the signals we want to handle and then
 350  * use signalfd to listen for them.  We rely on whatever the current signal
 351  * handler is to dispatch the signals when we receive them.
 352  */
 353 static void sigfd_handler(void *opaque)
 354 {
 355     int fd = (unsigned long) opaque;
 356     struct qemu_signalfd_siginfo info;
 357     struct sigaction action;
 358     ssize_t len;
 359
 360     while (1) {
 361         do {
 362             len = read(fd, &info, sizeof(info));
 363         } while (len == -1 && errno == EINTR);
 364
 365         if (len == -1 && errno == EAGAIN) {
 366             break;
 367         }
 368
 369         if (len != sizeof(info)) {
 370             printf("read from sigfd returned %zd: %m\n", len);
 371             return;
 372         }
 373
 374         sigaction(info.ssi_signo, NULL, &action);
 375         if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
 376             action.sa_sigaction(info.ssi_signo,
 377                                 (siginfo_t *)&info, NULL);
 378         } else if (action.sa_handler) {
 379             action.sa_handler(info.ssi_signo);
 380         }
 381     }
 382 }
 383
 384 static int qemu_signalfd_init(sigset_t mask)
 385 {
 386     int sigfd;
 387
 388     sigfd = qemu_signalfd(&mask);
 389     if (sigfd == -1) {
 390         fprintf(stderr, "failed to create signalfd\n");
 391         return -errno;
 392     }
 393
 394     fcntl_setfl(sigfd, O_NONBLOCK);
 395
 396     qemu_set_fd_handler2(sigfd, NULL, sigfd_handler, NULL,
 397                          (void *)(unsigned long) sigfd);
 398
 399     return 0;
 400 }
 401
 402 int qemu_init_main_loop(void)
 403 {
 404     int ret;
 405     sigset_t blocked_signals;
 406
 407     cpu_set_debug_excp_handler(cpu_debug_handler);
 408
 409     blocked_signals = block_io_signals();
 410
 411     ret = qemu_signalfd_init(blocked_signals);
 412     if (ret)
 413         return ret;
 414
 415     /* Note eventfd must be drained before signalfd handlers run */
 416     ret = qemu_event_init();
 417     if (ret)
 418         return ret;
 419
 420     qemu_cond_init(&qemu_pause_cond);
 421     qemu_cond_init(&qemu_system_cond);
 422     qemu_mutex_init(&qemu_fair_mutex);
 423     qemu_mutex_init(&qemu_global_mutex);
 424     qemu_mutex_lock(&qemu_global_mutex);
 425
 426     qemu_thread_self(&io_thread);
 427
 428     return 0;
 429 }
 430
 431 void qemu_main_loop_start(void)
 432 {
 433     qemu_system_ready = 1;
 434     qemu_cond_broadcast(&qemu_system_cond);
 435 }
 436
 437 void run_on_cpu(CPUState *env, void (*func)(void *data), void *data)
 438 {
 439     struct qemu_work_item wi;
 440
 441     if (qemu_cpu_self(env)) {
 442         func(data);
 443         return;
 444     }
 445
 446     wi.func = func;
 447     wi.data = data;
 448     if (!env->queued_work_first)
 449         env->queued_work_first = &wi;
 450     else
 451         env->queued_work_last->next = &wi;
 452     env->queued_work_last = &wi;
 453     wi.next = NULL;
 454     wi.done = false;
 455
 456     qemu_cpu_kick(env);
 457     while (!wi.done) {
 458         CPUState *self_env = cpu_single_env;
 459
 460         qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
 461         cpu_single_env = self_env;
 462     }
 463 }
 464
 465 static void flush_queued_work(CPUState *env)
 466 {
 467     struct qemu_work_item *wi;
 468
 469     if (!env->queued_work_first)
 470         return;
 471
 472     while ((wi = env->queued_work_first)) {
 473         env->queued_work_first = wi->next;
 474         wi->func(wi->data);
 475         wi->done = true;
 476     }
 477     env->queued_work_last = NULL;
 478     qemu_cond_broadcast(&qemu_work_cond);
 479 }
 480
 481 static void qemu_wait_io_event_common(CPUState *env)
 482 {
 483     if (env->stop) {
 484         env->stop = 0;
 485         env->stopped = 1;
 486         qemu_cond_signal(&qemu_pause_cond);
 487     }
 488     flush_queued_work(env);
 489     env->thread_kicked = false;
 490 }
 491
 492 static void qemu_tcg_wait_io_event(void)
 493 {
 494     CPUState *env;
 495
 496     while (!any_cpu_has_work())
 497         qemu_cond_timedwait(tcg_halt_cond, &qemu_global_mutex, 1000);
 498
 499     qemu_mutex_unlock(&qemu_global_mutex);
 500
 501     /*
 502      * Users of qemu_global_mutex can be starved, having no chance
 503      * to acquire it since this path will get to it first.
 504      * So use another lock to provide fairness.
 505      */
 506     qemu_mutex_lock(&qemu_fair_mutex);
 507     qemu_mutex_unlock(&qemu_fair_mutex);
 508
 509     qemu_mutex_lock(&qemu_global_mutex);
 510
 511     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 512         qemu_wait_io_event_common(env);
 513     }
 514 }
 515
 516 static void sigbus_reraise(void)
 517 {
 518     sigset_t set;
 519     struct sigaction action;
 520
 521     memset(&action, 0, sizeof(action));
 522     action.sa_handler = SIG_DFL;
 523     if (!sigaction(SIGBUS, &action, NULL)) {
 524         raise(SIGBUS);
 525         sigemptyset(&set);
 526         sigaddset(&set, SIGBUS);
 527         sigprocmask(SIG_UNBLOCK, &set, NULL);
 528     }
 529     perror("Failed to re-raise SIGBUS!\n");
 530     abort();
 531 }
 532
 533 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
 534                            void *ctx)
 535 {
 536 #if defined(TARGET_I386)
 537     if (kvm_on_sigbus(siginfo->ssi_code, (void *)(intptr_t)siginfo->ssi_addr))
 538 #endif
 539         sigbus_reraise();
 540 }
 541
 542 static void qemu_kvm_eat_signal(CPUState *env, int timeout)
 543 {
 544     struct timespec ts;
 545     int r, e;
 546     siginfo_t siginfo;
 547     sigset_t waitset;
 548     sigset_t chkset;
 549
 550     ts.tv_sec = timeout / 1000;
 551     ts.tv_nsec = (timeout % 1000) * 1000000;
 552
 553     sigemptyset(&waitset);
 554     sigaddset(&waitset, SIG_IPI);
 555     sigaddset(&waitset, SIGBUS);
 556
 557     do {
 558         qemu_mutex_unlock(&qemu_global_mutex);
 559
 560         r = sigtimedwait(&waitset, &siginfo, &ts);
 561         e = errno;
 562
 563         qemu_mutex_lock(&qemu_global_mutex);
 564
 565         if (r == -1 && !(e == EAGAIN || e == EINTR)) {
 566             fprintf(stderr, "sigtimedwait: %s\n", strerror(e));
 567             exit(1);
 568         }
 569
 570         switch (r) {
 571         case SIGBUS:
 572 #ifdef TARGET_I386
 573             if (kvm_on_sigbus_vcpu(env, siginfo.si_code, siginfo.si_addr))
 574 #endif
 575                 sigbus_reraise();
 576             break;
 577         default:
 578             break;
 579         }
 580
 581         r = sigpending(&chkset);
 582         if (r == -1) {
 583             fprintf(stderr, "sigpending: %s\n", strerror(e));
 584             exit(1);
 585         }
 586     } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
 587 }
 588
 589 static void qemu_kvm_wait_io_event(CPUState *env)
 590 {
 591     while (!cpu_has_work(env))
 592         qemu_cond_timedwait(env->halt_cond, &qemu_global_mutex, 1000);
 593
 594     qemu_kvm_eat_signal(env, 0);
 595     qemu_wait_io_event_common(env);
 596 }
 597
 598 static int qemu_cpu_exec(CPUState *env);
 599
 600 static void *kvm_cpu_thread_fn(void *arg)
 601 {
 602     CPUState *env = arg;
 603
 604     qemu_mutex_lock(&qemu_global_mutex);
 605     qemu_thread_self(env->thread);
 606     if (kvm_enabled())
 607         kvm_init_vcpu(env);
 608
 609     kvm_init_ipi(env);
 610
 611     /* signal CPU creation */
 612     env->created = 1;
 613     qemu_cond_signal(&qemu_cpu_cond);
 614
 615     /* and wait for machine initialization */
 616     while (!qemu_system_ready)
 617         qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
 618
 619     while (1) {
 620         if (cpu_can_run(env))
 621             qemu_cpu_exec(env);
 622         qemu_kvm_wait_io_event(env);
 623     }
 624
 625     return NULL;
 626 }
 627
 628 static void *tcg_cpu_thread_fn(void *arg)
 629 {
 630     CPUState *env = arg;
 631
 632     tcg_init_ipi();
 633     qemu_thread_self(env->thread);
 634
 635     /* signal CPU creation */
 636     qemu_mutex_lock(&qemu_global_mutex);
 637     for (env = first_cpu; env != NULL; env = env->next_cpu)
 638         env->created = 1;
 639     qemu_cond_signal(&qemu_cpu_cond);
 640
 641     /* and wait for machine initialization */
 642     while (!qemu_system_ready)
 643         qemu_cond_timedwait(&qemu_system_cond, &qemu_global_mutex, 100);
 644
 645     while (1) {
 646         cpu_exec_all();
 647         qemu_tcg_wait_io_event();
 648     }
 649
 650     return NULL;
 651 }
 652
 653 void qemu_cpu_kick(void *_env)
 654 {
 655     CPUState *env = _env;
 656     qemu_cond_broadcast(env->halt_cond);
 657     if (!env->thread_kicked) {
 658         qemu_thread_signal(env->thread, SIG_IPI);
 659         env->thread_kicked = true;
 660     }
 661 }
 662
 663 int qemu_cpu_self(void *_env)
 664 {
 665     CPUState *env = _env;
 666     QemuThread this;
 667
 668     qemu_thread_self(&this);
 669
 670     return qemu_thread_equal(&this, env->thread);
 671 }
 672
 673 static void cpu_signal(int sig)
 674 {
 675     if (cpu_single_env)
 676         cpu_exit(cpu_single_env);
 677     exit_request = 1;
 678 }
 679
 680 static void tcg_init_ipi(void)
 681 {
 682     sigset_t set;
 683     struct sigaction sigact;
 684
 685     memset(&sigact, 0, sizeof(sigact));
 686     sigact.sa_handler = cpu_signal;
 687     sigaction(SIG_IPI, &sigact, NULL);
 688
 689     sigemptyset(&set);
 690     sigaddset(&set, SIG_IPI);
 691     pthread_sigmask(SIG_UNBLOCK, &set, NULL);
 692 }
 693
 694 static void dummy_signal(int sig)
 695 {
 696 }
 697
 698 static void kvm_init_ipi(CPUState *env)
 699 {
 700     int r;
 701     sigset_t set;
 702     struct sigaction sigact;
 703
 704     memset(&sigact, 0, sizeof(sigact));
 705     sigact.sa_handler = dummy_signal;
 706     sigaction(SIG_IPI, &sigact, NULL);
 707
 708     pthread_sigmask(SIG_BLOCK, NULL, &set);
 709     sigdelset(&set, SIG_IPI);
 710     sigdelset(&set, SIGBUS);
 711     r = kvm_set_signal_mask(env, &set);
 712     if (r) {
 713         fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(r));
 714         exit(1);
 715     }
 716 }
 717
 718 static sigset_t block_io_signals(void)
 719 {
 720     sigset_t set;
 721     struct sigaction action;
 722
 723     /* SIGUSR2 used by posix-aio-compat.c */
 724     sigemptyset(&set);
 725     sigaddset(&set, SIGUSR2);
 726     pthread_sigmask(SIG_UNBLOCK, &set, NULL);
 727
 728     sigemptyset(&set);
 729     sigaddset(&set, SIGIO);
 730     sigaddset(&set, SIGALRM);
 731     sigaddset(&set, SIG_IPI);
 732     sigaddset(&set, SIGBUS);
 733     pthread_sigmask(SIG_BLOCK, &set, NULL);
 734
 735     memset(&action, 0, sizeof(action));
 736     action.sa_flags = SA_SIGINFO;
 737     action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
 738     sigaction(SIGBUS, &action, NULL);
 739     prctl(PR_MCE_KILL, 1, 1, 0, 0);
 740
 741     return set;
 742 }
 743
 744 void qemu_mutex_lock_iothread(void)
 745 {
 746     if (kvm_enabled()) {
 747         qemu_mutex_lock(&qemu_global_mutex);
 748     } else {
 749         qemu_mutex_lock(&qemu_fair_mutex);
 750         if (qemu_mutex_trylock(&qemu_global_mutex)) {
 751             qemu_thread_signal(tcg_cpu_thread, SIG_IPI);
 752             qemu_mutex_lock(&qemu_global_mutex);
 753         }
 754         qemu_mutex_unlock(&qemu_fair_mutex);
 755     }
 756 }
 757
 758 void qemu_mutex_unlock_iothread(void)
 759 {
 760     qemu_mutex_unlock(&qemu_global_mutex);
 761 }
 762
 763 static int all_vcpus_paused(void)
 764 {
 765     CPUState *penv = first_cpu;
 766
 767     while (penv) {
 768         if (!penv->stopped)
 769             return 0;
 770         penv = (CPUState *)penv->next_cpu;
 771     }
 772
 773     return 1;
 774 }
 775
 776 void pause_all_vcpus(void)
 777 {
 778     CPUState *penv = first_cpu;
 779
 780     while (penv) {
 781         penv->stop = 1;
 782         qemu_cpu_kick(penv);
 783         penv = (CPUState *)penv->next_cpu;
 784     }
 785
 786     while (!all_vcpus_paused()) {
 787         qemu_cond_timedwait(&qemu_pause_cond, &qemu_global_mutex, 100);
 788         penv = first_cpu;
 789         while (penv) {
 790             qemu_cpu_kick(penv);
 791             penv = (CPUState *)penv->next_cpu;
 792         }
 793     }
 794 }
 795
 796 void resume_all_vcpus(void)
 797 {
 798     CPUState *penv = first_cpu;
 799
 800     while (penv) {
 801         penv->stop = 0;
 802         penv->stopped = 0;
 803         qemu_cpu_kick(penv);
 804         penv = (CPUState *)penv->next_cpu;
 805     }
 806 }
 807
 808 static void tcg_init_vcpu(void *_env)
 809 {
 810     CPUState *env = _env;
 811     /* share a single thread for all cpus with TCG */
 812     if (!tcg_cpu_thread) {
 813         env->thread = qemu_mallocz(sizeof(QemuThread));
 814         env->halt_cond = qemu_mallocz(sizeof(QemuCond));
 815         qemu_cond_init(env->halt_cond);
 816         qemu_thread_create(env->thread, tcg_cpu_thread_fn, env);
 817         while (env->created == 0)
 818             qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
 819         tcg_cpu_thread = env->thread;
 820         tcg_halt_cond = env->halt_cond;
 821     } else {
 822         env->thread = tcg_cpu_thread;
 823         env->halt_cond = tcg_halt_cond;
 824     }
 825 }
 826
 827 static void kvm_start_vcpu(CPUState *env)
 828 {
 829     env->thread = qemu_mallocz(sizeof(QemuThread));
 830     env->halt_cond = qemu_mallocz(sizeof(QemuCond));
 831     qemu_cond_init(env->halt_cond);
 832     qemu_thread_create(env->thread, kvm_cpu_thread_fn, env);
 833     while (env->created == 0)
 834         qemu_cond_timedwait(&qemu_cpu_cond, &qemu_global_mutex, 100);
 835 }
 836
 837 void qemu_init_vcpu(void *_env)
 838 {
 839     CPUState *env = _env;
 840
 841     env->nr_cores = smp_cores;
 842     env->nr_threads = smp_threads;
 843     if (kvm_enabled())
 844         kvm_start_vcpu(env);
 845     else
 846         tcg_init_vcpu(env);
 847 }
 848
 849 void qemu_notify_event(void)
 850 {
 851     qemu_event_increment();
 852 }
 853
 854 static void qemu_system_vmstop_request(int reason)
 855 {
 856     vmstop_requested = reason;
 857     qemu_notify_event();
 858 }
 859
 860 void cpu_stop_current(void)
 861 {
 862     if (cpu_single_env) {
 863         cpu_single_env->stopped = 1;
 864         cpu_exit(cpu_single_env);
 865     }
 866 }
 867
 868 void vm_stop(int reason)
 869 {
 870     QemuThread me;
 871     qemu_thread_self(&me);
 872
 873     if (!qemu_thread_equal(&me, &io_thread)) {
 874         qemu_system_vmstop_request(reason);
 875         /*
 876          * FIXME: should not return to device code in case
 877          * vm_stop() has been requested.
 878          */
 879         cpu_stop_current();
 880         return;
 881     }
 882     do_vm_stop(reason);
 883 }
 884
 885 #endif
 886
 887 static int qemu_cpu_exec(CPUState *env)
 888 {
 889     int ret;
 890 #ifdef CONFIG_PROFILER
 891     int64_t ti;
 892 #endif
 893
 894 #ifdef CONFIG_PROFILER
 895     ti = profile_getclock();
 896 #endif
 897     if (use_icount) {
 898         int64_t count;
 899         int decr;
 900         qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
 901         env->icount_decr.u16.low = 0;
 902         env->icount_extra = 0;
 903         count = qemu_icount_round (qemu_next_deadline());
 904         qemu_icount += count;
 905         decr = (count > 0xffff) ? 0xffff : count;
 906         count -= decr;
 907         env->icount_decr.u16.low = decr;
 908         env->icount_extra = count;
 909     }
 910     ret = cpu_exec(env);
 911 #ifdef CONFIG_PROFILER
 912     qemu_time += profile_getclock() - ti;
 913 #endif
 914     if (use_icount) {
 915         /* Fold pending instructions back into the
 916            instruction counter, and clear the interrupt flag.  */
 917         qemu_icount -= (env->icount_decr.u16.low
 918                         + env->icount_extra);
 919         env->icount_decr.u32 = 0;
 920         env->icount_extra = 0;
 921     }
 922     return ret;
 923 }
 924
 925 bool cpu_exec_all(void)
 926 {
 927     if (next_cpu == NULL)
 928         next_cpu = first_cpu;
 929     for (; next_cpu != NULL && !exit_request; next_cpu = next_cpu->next_cpu) {
 930         CPUState *env = next_cpu;
 931
 932         qemu_clock_enable(vm_clock,
 933                           (env->singlestep_enabled & SSTEP_NOTIMER) == 0);
 934
 935         if (qemu_alarm_pending())
 936             break;
 937         if (cpu_can_run(env)) {
 938             if (qemu_cpu_exec(env) == EXCP_DEBUG) {
 939                 break;
 940             }
 941         } else if (env->stop) {
 942             break;
 943         }
 944     }
 945     exit_request = 0;
 946     return any_cpu_has_work();
 947 }
 948
 949 void set_numa_modes(void)
 950 {
 951     CPUState *env;
 952     int i;
 953
 954     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 955         for (i = 0; i < nb_numa_nodes; i++) {
 956             if (node_cpumask[i] & (1 << env->cpu_index)) {
 957                 env->numa_node = i;
 958             }
 959         }
 960     }
 961 }
 962
 963 void set_cpu_log(const char *optarg)
 964 {
 965     int mask;
 966     const CPULogItem *item;
 967
 968     mask = cpu_str_to_log_mask(optarg);
 969     if (!mask) {
 970         printf("Log items (comma separated):\n");
 971         for (item = cpu_log_items; item->mask != 0; item++) {
 972             printf("%-10s %s\n", item->name, item->help);
 973         }
 974         exit(1);
 975     }
 976     cpu_set_log(mask);
 977 }
 978
 979 /* Return the virtual CPU time, based on the instruction counter.  */
 980 int64_t cpu_get_icount(void)
 981 {
 982     int64_t icount;
 983     CPUState *env = cpu_single_env;;
 984
 985     icount = qemu_icount;
 986     if (env) {
 987         if (!can_do_io(env)) {
 988             fprintf(stderr, "Bad clock read\n");
 989         }
 990         icount -= (env->icount_decr.u16.low + env->icount_extra);
 991     }
 992     return qemu_icount_bias + (icount << icount_time_shift);
 993 }
 994
 995 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
 996 {
 997     /* XXX: implement xxx_cpu_list for targets that still miss it */
 998 #if defined(cpu_list_id)
 999     cpu_list_id(f, cpu_fprintf, optarg);
1000 #elif defined(cpu_list)
1001     cpu_list(f, cpu_fprintf); /* deprecated */
1002 #endif
1003 }