cpus.c

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