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