cpus.c

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