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/monitor.h"
  29 #include "sysemu/sysemu.h"
  30 #include "exec/gdbstub.h"
  31 #include "sysemu/dma.h"
  32 #include "sysemu/kvm.h"
  33 #include "qmp-commands.h"
  34
  35 #include "qemu/thread.h"
  36 #include "sysemu/cpus.h"
  37 #include "sysemu/qtest.h"
  38 #include "qemu/main-loop.h"
  39 #include "qemu/bitmap.h"
  40
  41 #ifndef _WIN32
  42 #include "qemu/compatfd.h"
  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 CPUArchState *next_cpu;
  64
  65 static bool cpu_thread_is_idle(CPUArchState *env)
  66 {
  67     CPUState *cpu = ENV_GET_CPU(env);
  68
  69     if (cpu->stop || cpu->queued_work_first) {
  70         return false;
  71     }
  72     if (cpu->stopped || !runstate_is_running()) {
  73         return true;
  74     }
  75     if (!cpu->halted || qemu_cpu_has_work(cpu) ||
  76         kvm_async_interrupts_enabled()) {
  77         return false;
  78     }
  79     return true;
  80 }
  81
  82 static bool all_cpu_threads_idle(void)
  83 {
  84     CPUArchState *env;
  85
  86     for (env = first_cpu; env != NULL; env = env->next_cpu) {
  87         if (!cpu_thread_is_idle(env)) {
  88             return false;
  89         }
  90     }
  91     return true;
  92 }
  93
  94 /***********************************************************/
  95 /* guest cycle counter */
  96
  97 /* Conversion factor from emulated instructions to virtual clock ticks.  */
  98 static int icount_time_shift;
  99 /* Arbitrarily pick 1MIPS as the minimum allowable speed.  */
 100 #define MAX_ICOUNT_SHIFT 10
 101 /* Compensate for varying guest execution speed.  */
 102 static int64_t qemu_icount_bias;
 103 static QEMUTimer *icount_rt_timer;
 104 static QEMUTimer *icount_vm_timer;
 105 static QEMUTimer *icount_warp_timer;
 106 static int64_t vm_clock_warp_start;
 107 static int64_t qemu_icount;
 108
 109 typedef struct TimersState {
 110     int64_t cpu_ticks_prev;
 111     int64_t cpu_ticks_offset;
 112     int64_t cpu_clock_offset;
 113     int32_t cpu_ticks_enabled;
 114     int64_t dummy;
 115 } TimersState;
 116
 117 TimersState timers_state;
 118
 119 /* Return the virtual CPU time, based on the instruction counter.  */
 120 int64_t cpu_get_icount(void)
 121 {
 122     int64_t icount;
 123     CPUArchState *env = cpu_single_env;
 124
 125     icount = qemu_icount;
 126     if (env) {
 127         if (!can_do_io(env)) {
 128             fprintf(stderr, "Bad clock read\n");
 129         }
 130         icount -= (env->icount_decr.u16.low + env->icount_extra);
 131     }
 132     return qemu_icount_bias + (icount << icount_time_shift);
 133 }
 134
 135 /* return the host CPU cycle counter and handle stop/restart */
 136 int64_t cpu_get_ticks(void)
 137 {
 138     if (use_icount) {
 139         return cpu_get_icount();
 140     }
 141     if (!timers_state.cpu_ticks_enabled) {
 142         return timers_state.cpu_ticks_offset;
 143     } else {
 144         int64_t ticks;
 145         ticks = cpu_get_real_ticks();
 146         if (timers_state.cpu_ticks_prev > ticks) {
 147             /* Note: non increasing ticks may happen if the host uses
 148                software suspend */
 149             timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks;
 150         }
 151         timers_state.cpu_ticks_prev = ticks;
 152         return ticks + timers_state.cpu_ticks_offset;
 153     }
 154 }
 155
 156 /* return the host CPU monotonic timer and handle stop/restart */
 157 int64_t cpu_get_clock(void)
 158 {
 159     int64_t ti;
 160     if (!timers_state.cpu_ticks_enabled) {
 161         return timers_state.cpu_clock_offset;
 162     } else {
 163         ti = get_clock();
 164         return ti + timers_state.cpu_clock_offset;
 165     }
 166 }
 167
 168 /* enable cpu_get_ticks() */
 169 void cpu_enable_ticks(void)
 170 {
 171     if (!timers_state.cpu_ticks_enabled) {
 172         timers_state.cpu_ticks_offset -= cpu_get_real_ticks();
 173         timers_state.cpu_clock_offset -= get_clock();
 174         timers_state.cpu_ticks_enabled = 1;
 175     }
 176 }
 177
 178 /* disable cpu_get_ticks() : the clock is stopped. You must not call
 179    cpu_get_ticks() after that.  */
 180 void cpu_disable_ticks(void)
 181 {
 182     if (timers_state.cpu_ticks_enabled) {
 183         timers_state.cpu_ticks_offset = cpu_get_ticks();
 184         timers_state.cpu_clock_offset = cpu_get_clock();
 185         timers_state.cpu_ticks_enabled = 0;
 186     }
 187 }
 188
 189 /* Correlation between real and virtual time is always going to be
 190    fairly approximate, so ignore small variation.
 191    When the guest is idle real and virtual time will be aligned in
 192    the IO wait loop.  */
 193 #define ICOUNT_WOBBLE (get_ticks_per_sec() / 10)
 194
 195 static void icount_adjust(void)
 196 {
 197     int64_t cur_time;
 198     int64_t cur_icount;
 199     int64_t delta;
 200     static int64_t last_delta;
 201     /* If the VM is not running, then do nothing.  */
 202     if (!runstate_is_running()) {
 203         return;
 204     }
 205     cur_time = cpu_get_clock();
 206     cur_icount = qemu_get_clock_ns(vm_clock);
 207     delta = cur_icount - cur_time;
 208     /* FIXME: This is a very crude algorithm, somewhat prone to oscillation.  */
 209     if (delta > 0
 210         && last_delta + ICOUNT_WOBBLE < delta * 2
 211         && icount_time_shift > 0) {
 212         /* The guest is getting too far ahead.  Slow time down.  */
 213         icount_time_shift--;
 214     }
 215     if (delta < 0
 216         && last_delta - ICOUNT_WOBBLE > delta * 2
 217         && icount_time_shift < MAX_ICOUNT_SHIFT) {
 218         /* The guest is getting too far behind.  Speed time up.  */
 219         icount_time_shift++;
 220     }
 221     last_delta = delta;
 222     qemu_icount_bias = cur_icount - (qemu_icount << icount_time_shift);
 223 }
 224
 225 static void icount_adjust_rt(void *opaque)
 226 {
 227     qemu_mod_timer(icount_rt_timer,
 228                    qemu_get_clock_ms(rt_clock) + 1000);
 229     icount_adjust();
 230 }
 231
 232 static void icount_adjust_vm(void *opaque)
 233 {
 234     qemu_mod_timer(icount_vm_timer,
 235                    qemu_get_clock_ns(vm_clock) + get_ticks_per_sec() / 10);
 236     icount_adjust();
 237 }
 238
 239 static int64_t qemu_icount_round(int64_t count)
 240 {
 241     return (count + (1 << icount_time_shift) - 1) >> icount_time_shift;
 242 }
 243
 244 static void icount_warp_rt(void *opaque)
 245 {
 246     if (vm_clock_warp_start == -1) {
 247         return;
 248     }
 249
 250     if (runstate_is_running()) {
 251         int64_t clock = qemu_get_clock_ns(rt_clock);
 252         int64_t warp_delta = clock - vm_clock_warp_start;
 253         if (use_icount == 1) {
 254             qemu_icount_bias += warp_delta;
 255         } else {
 256             /*
 257              * In adaptive mode, do not let the vm_clock run too
 258              * far ahead of real time.
 259              */
 260             int64_t cur_time = cpu_get_clock();
 261             int64_t cur_icount = qemu_get_clock_ns(vm_clock);
 262             int64_t delta = cur_time - cur_icount;
 263             qemu_icount_bias += MIN(warp_delta, delta);
 264         }
 265         if (qemu_clock_expired(vm_clock)) {
 266             qemu_notify_event();
 267         }
 268     }
 269     vm_clock_warp_start = -1;
 270 }
 271
 272 void qtest_clock_warp(int64_t dest)
 273 {
 274     int64_t clock = qemu_get_clock_ns(vm_clock);
 275     assert(qtest_enabled());
 276     while (clock < dest) {
 277         int64_t deadline = qemu_clock_deadline(vm_clock);
 278         int64_t warp = MIN(dest - clock, deadline);
 279         qemu_icount_bias += warp;
 280         qemu_run_timers(vm_clock);
 281         clock = qemu_get_clock_ns(vm_clock);
 282     }
 283     qemu_notify_event();
 284 }
 285
 286 void qemu_clock_warp(QEMUClock *clock)
 287 {
 288     int64_t deadline;
 289
 290     /*
 291      * There are too many global variables to make the "warp" behavior
 292      * applicable to other clocks.  But a clock argument removes the
 293      * need for if statements all over the place.
 294      */
 295     if (clock != vm_clock || !use_icount) {
 296         return;
 297     }
 298
 299     /*
 300      * If the CPUs have been sleeping, advance the vm_clock timer now.  This
 301      * ensures that the deadline for the timer is computed correctly below.
 302      * This also makes sure that the insn counter is synchronized before the
 303      * CPU starts running, in case the CPU is woken by an event other than
 304      * the earliest vm_clock timer.
 305      */
 306     icount_warp_rt(NULL);
 307     if (!all_cpu_threads_idle() || !qemu_clock_has_timers(vm_clock)) {
 308         qemu_del_timer(icount_warp_timer);
 309         return;
 310     }
 311
 312     if (qtest_enabled()) {
 313         /* When testing, qtest commands advance icount.  */
 314         return;
 315     }
 316
 317     vm_clock_warp_start = qemu_get_clock_ns(rt_clock);
 318     deadline = qemu_clock_deadline(vm_clock);
 319     if (deadline > 0) {
 320         /*
 321          * Ensure the vm_clock proceeds even when the virtual CPU goes to
 322          * sleep.  Otherwise, the CPU might be waiting for a future timer
 323          * interrupt to wake it up, but the interrupt never comes because
 324          * the vCPU isn't running any insns and thus doesn't advance the
 325          * vm_clock.
 326          *
 327          * An extreme solution for this problem would be to never let VCPUs
 328          * sleep in icount mode if there is a pending vm_clock timer; rather
 329          * time could just advance to the next vm_clock event.  Instead, we
 330          * do stop VCPUs and only advance vm_clock after some "real" time,
 331          * (related to the time left until the next event) has passed.  This
 332          * rt_clock timer will do this.  This avoids that the warps are too
 333          * visible externally---for example, you will not be sending network
 334          * packets continuously instead of every 100ms.
 335          */
 336         qemu_mod_timer(icount_warp_timer, vm_clock_warp_start + deadline);
 337     } else {
 338         qemu_notify_event();
 339     }
 340 }
 341
 342 static const VMStateDescription vmstate_timers = {
 343     .name = "timer",
 344     .version_id = 2,
 345     .minimum_version_id = 1,
 346     .minimum_version_id_old = 1,
 347     .fields      = (VMStateField[]) {
 348         VMSTATE_INT64(cpu_ticks_offset, TimersState),
 349         VMSTATE_INT64(dummy, TimersState),
 350         VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2),
 351         VMSTATE_END_OF_LIST()
 352     }
 353 };
 354
 355 void configure_icount(const char *option)
 356 {
 357     vmstate_register(NULL, 0, &vmstate_timers, &timers_state);
 358     if (!option) {
 359         return;
 360     }
 361
 362     icount_warp_timer = qemu_new_timer_ns(rt_clock, icount_warp_rt, NULL);
 363     if (strcmp(option, "auto") != 0) {
 364         icount_time_shift = strtol(option, NULL, 0);
 365         use_icount = 1;
 366         return;
 367     }
 368
 369     use_icount = 2;
 370
 371     /* 125MIPS seems a reasonable initial guess at the guest speed.
 372        It will be corrected fairly quickly anyway.  */
 373     icount_time_shift = 3;
 374
 375     /* Have both realtime and virtual time triggers for speed adjustment.
 376        The realtime trigger catches emulated time passing too slowly,
 377        the virtual time trigger catches emulated time passing too fast.
 378        Realtime triggers occur even when idle, so use them less frequently
 379        than VM triggers.  */
 380     icount_rt_timer = qemu_new_timer_ms(rt_clock, icount_adjust_rt, NULL);
 381     qemu_mod_timer(icount_rt_timer,
 382                    qemu_get_clock_ms(rt_clock) + 1000);
 383     icount_vm_timer = qemu_new_timer_ns(vm_clock, icount_adjust_vm, NULL);
 384     qemu_mod_timer(icount_vm_timer,
 385                    qemu_get_clock_ns(vm_clock) + get_ticks_per_sec() / 10);
 386 }
 387
 388 /***********************************************************/
 389 void hw_error(const char *fmt, ...)
 390 {
 391     va_list ap;
 392     CPUArchState *env;
 393     CPUState *cpu;
 394
 395     va_start(ap, fmt);
 396     fprintf(stderr, "qemu: hardware error: ");
 397     vfprintf(stderr, fmt, ap);
 398     fprintf(stderr, "\n");
 399     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 400         cpu = ENV_GET_CPU(env);
 401         fprintf(stderr, "CPU #%d:\n", cpu->cpu_index);
 402         cpu_dump_state(env, stderr, fprintf, CPU_DUMP_FPU);
 403     }
 404     va_end(ap);
 405     abort();
 406 }
 407
 408 void cpu_synchronize_all_states(void)
 409 {
 410     CPUArchState *cpu;
 411
 412     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
 413         cpu_synchronize_state(cpu);
 414     }
 415 }
 416
 417 void cpu_synchronize_all_post_reset(void)
 418 {
 419     CPUArchState *cpu;
 420
 421     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
 422         cpu_synchronize_post_reset(cpu);
 423     }
 424 }
 425
 426 void cpu_synchronize_all_post_init(void)
 427 {
 428     CPUArchState *cpu;
 429
 430     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
 431         cpu_synchronize_post_init(cpu);
 432     }
 433 }
 434
 435 bool cpu_is_stopped(CPUState *cpu)
 436 {
 437     return !runstate_is_running() || cpu->stopped;
 438 }
 439
 440 static void do_vm_stop(RunState state)
 441 {
 442     if (runstate_is_running()) {
 443         cpu_disable_ticks();
 444         pause_all_vcpus();
 445         runstate_set(state);
 446         vm_state_notify(0, state);
 447         bdrv_drain_all();
 448         bdrv_flush_all();
 449         monitor_protocol_event(QEVENT_STOP, NULL);
 450     }
 451 }
 452
 453 static bool cpu_can_run(CPUState *cpu)
 454 {
 455     if (cpu->stop) {
 456         return false;
 457     }
 458     if (cpu->stopped || !runstate_is_running()) {
 459         return false;
 460     }
 461     return true;
 462 }
 463
 464 static void cpu_handle_guest_debug(CPUArchState *env)
 465 {
 466     CPUState *cpu = ENV_GET_CPU(env);
 467
 468     gdb_set_stop_cpu(env);
 469     qemu_system_debug_request();
 470     cpu->stopped = true;
 471 }
 472
 473 static void cpu_signal(int sig)
 474 {
 475     if (cpu_single_env) {
 476         cpu_exit(cpu_single_env);
 477     }
 478     exit_request = 1;
 479 }
 480
 481 #ifdef CONFIG_LINUX
 482 static void sigbus_reraise(void)
 483 {
 484     sigset_t set;
 485     struct sigaction action;
 486
 487     memset(&action, 0, sizeof(action));
 488     action.sa_handler = SIG_DFL;
 489     if (!sigaction(SIGBUS, &action, NULL)) {
 490         raise(SIGBUS);
 491         sigemptyset(&set);
 492         sigaddset(&set, SIGBUS);
 493         sigprocmask(SIG_UNBLOCK, &set, NULL);
 494     }
 495     perror("Failed to re-raise SIGBUS!\n");
 496     abort();
 497 }
 498
 499 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
 500                            void *ctx)
 501 {
 502     if (kvm_on_sigbus(siginfo->ssi_code,
 503                       (void *)(intptr_t)siginfo->ssi_addr)) {
 504         sigbus_reraise();
 505     }
 506 }
 507
 508 static void qemu_init_sigbus(void)
 509 {
 510     struct sigaction action;
 511
 512     memset(&action, 0, sizeof(action));
 513     action.sa_flags = SA_SIGINFO;
 514     action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
 515     sigaction(SIGBUS, &action, NULL);
 516
 517     prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0);
 518 }
 519
 520 static void qemu_kvm_eat_signals(CPUState *cpu)
 521 {
 522     struct timespec ts = { 0, 0 };
 523     siginfo_t siginfo;
 524     sigset_t waitset;
 525     sigset_t chkset;
 526     int r;
 527
 528     sigemptyset(&waitset);
 529     sigaddset(&waitset, SIG_IPI);
 530     sigaddset(&waitset, SIGBUS);
 531
 532     do {
 533         r = sigtimedwait(&waitset, &siginfo, &ts);
 534         if (r == -1 && !(errno == EAGAIN || errno == EINTR)) {
 535             perror("sigtimedwait");
 536             exit(1);
 537         }
 538
 539         switch (r) {
 540         case SIGBUS:
 541             if (kvm_on_sigbus_vcpu(cpu, siginfo.si_code, siginfo.si_addr)) {
 542                 sigbus_reraise();
 543             }
 544             break;
 545         default:
 546             break;
 547         }
 548
 549         r = sigpending(&chkset);
 550         if (r == -1) {
 551             perror("sigpending");
 552             exit(1);
 553         }
 554     } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
 555 }
 556
 557 #else /* !CONFIG_LINUX */
 558
 559 static void qemu_init_sigbus(void)
 560 {
 561 }
 562
 563 static void qemu_kvm_eat_signals(CPUState *cpu)
 564 {
 565 }
 566 #endif /* !CONFIG_LINUX */
 567
 568 #ifndef _WIN32
 569 static void dummy_signal(int sig)
 570 {
 571 }
 572
 573 static void qemu_kvm_init_cpu_signals(CPUArchState *env)
 574 {
 575     int r;
 576     sigset_t set;
 577     struct sigaction sigact;
 578
 579     memset(&sigact, 0, sizeof(sigact));
 580     sigact.sa_handler = dummy_signal;
 581     sigaction(SIG_IPI, &sigact, NULL);
 582
 583     pthread_sigmask(SIG_BLOCK, NULL, &set);
 584     sigdelset(&set, SIG_IPI);
 585     sigdelset(&set, SIGBUS);
 586     r = kvm_set_signal_mask(env, &set);
 587     if (r) {
 588         fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
 589         exit(1);
 590     }
 591 }
 592
 593 static void qemu_tcg_init_cpu_signals(void)
 594 {
 595     sigset_t set;
 596     struct sigaction sigact;
 597
 598     memset(&sigact, 0, sizeof(sigact));
 599     sigact.sa_handler = cpu_signal;
 600     sigaction(SIG_IPI, &sigact, NULL);
 601
 602     sigemptyset(&set);
 603     sigaddset(&set, SIG_IPI);
 604     pthread_sigmask(SIG_UNBLOCK, &set, NULL);
 605 }
 606
 607 #else /* _WIN32 */
 608 static void qemu_kvm_init_cpu_signals(CPUArchState *env)
 609 {
 610     abort();
 611 }
 612
 613 static void qemu_tcg_init_cpu_signals(void)
 614 {
 615 }
 616 #endif /* _WIN32 */
 617
 618 static QemuMutex qemu_global_mutex;
 619 static QemuCond qemu_io_proceeded_cond;
 620 static bool iothread_requesting_mutex;
 621
 622 static QemuThread io_thread;
 623
 624 static QemuThread *tcg_cpu_thread;
 625 static QemuCond *tcg_halt_cond;
 626
 627 /* cpu creation */
 628 static QemuCond qemu_cpu_cond;
 629 /* system init */
 630 static QemuCond qemu_pause_cond;
 631 static QemuCond qemu_work_cond;
 632
 633 void qemu_init_cpu_loop(void)
 634 {
 635     qemu_init_sigbus();
 636     qemu_cond_init(&qemu_cpu_cond);
 637     qemu_cond_init(&qemu_pause_cond);
 638     qemu_cond_init(&qemu_work_cond);
 639     qemu_cond_init(&qemu_io_proceeded_cond);
 640     qemu_mutex_init(&qemu_global_mutex);
 641
 642     qemu_thread_get_self(&io_thread);
 643 }
 644
 645 void run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
 646 {
 647     struct qemu_work_item wi;
 648
 649     if (qemu_cpu_is_self(cpu)) {
 650         func(data);
 651         return;
 652     }
 653
 654     wi.func = func;
 655     wi.data = data;
 656     if (cpu->queued_work_first == NULL) {
 657         cpu->queued_work_first = &wi;
 658     } else {
 659         cpu->queued_work_last->next = &wi;
 660     }
 661     cpu->queued_work_last = &wi;
 662     wi.next = NULL;
 663     wi.done = false;
 664
 665     qemu_cpu_kick(cpu);
 666     while (!wi.done) {
 667         CPUArchState *self_env = cpu_single_env;
 668
 669         qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
 670         cpu_single_env = self_env;
 671     }
 672 }
 673
 674 static void flush_queued_work(CPUState *cpu)
 675 {
 676     struct qemu_work_item *wi;
 677
 678     if (cpu->queued_work_first == NULL) {
 679         return;
 680     }
 681
 682     while ((wi = cpu->queued_work_first)) {
 683         cpu->queued_work_first = wi->next;
 684         wi->func(wi->data);
 685         wi->done = true;
 686     }
 687     cpu->queued_work_last = NULL;
 688     qemu_cond_broadcast(&qemu_work_cond);
 689 }
 690
 691 static void qemu_wait_io_event_common(CPUState *cpu)
 692 {
 693     if (cpu->stop) {
 694         cpu->stop = false;
 695         cpu->stopped = true;
 696         qemu_cond_signal(&qemu_pause_cond);
 697     }
 698     flush_queued_work(cpu);
 699     cpu->thread_kicked = false;
 700 }
 701
 702 static void qemu_tcg_wait_io_event(void)
 703 {
 704     CPUArchState *env;
 705
 706     while (all_cpu_threads_idle()) {
 707        /* Start accounting real time to the virtual clock if the CPUs
 708           are idle.  */
 709         qemu_clock_warp(vm_clock);
 710         qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
 711     }
 712
 713     while (iothread_requesting_mutex) {
 714         qemu_cond_wait(&qemu_io_proceeded_cond, &qemu_global_mutex);
 715     }
 716
 717     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 718         qemu_wait_io_event_common(ENV_GET_CPU(env));
 719     }
 720 }
 721
 722 static void qemu_kvm_wait_io_event(CPUArchState *env)
 723 {
 724     CPUState *cpu = ENV_GET_CPU(env);
 725
 726     while (cpu_thread_is_idle(env)) {
 727         qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
 728     }
 729
 730     qemu_kvm_eat_signals(cpu);
 731     qemu_wait_io_event_common(cpu);
 732 }
 733
 734 static void *qemu_kvm_cpu_thread_fn(void *arg)
 735 {
 736     CPUArchState *env = arg;
 737     CPUState *cpu = ENV_GET_CPU(env);
 738     int r;
 739
 740     qemu_mutex_lock(&qemu_global_mutex);
 741     qemu_thread_get_self(cpu->thread);
 742     cpu->thread_id = qemu_get_thread_id();
 743     cpu_single_env = env;
 744
 745     r = kvm_init_vcpu(cpu);
 746     if (r < 0) {
 747         fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r));
 748         exit(1);
 749     }
 750
 751     qemu_kvm_init_cpu_signals(env);
 752
 753     /* signal CPU creation */
 754     cpu->created = true;
 755     qemu_cond_signal(&qemu_cpu_cond);
 756
 757     while (1) {
 758         if (cpu_can_run(cpu)) {
 759             r = kvm_cpu_exec(env);
 760             if (r == EXCP_DEBUG) {
 761                 cpu_handle_guest_debug(env);
 762             }
 763         }
 764         qemu_kvm_wait_io_event(env);
 765     }
 766
 767     return NULL;
 768 }
 769
 770 static void *qemu_dummy_cpu_thread_fn(void *arg)
 771 {
 772 #ifdef _WIN32
 773     fprintf(stderr, "qtest is not supported under Windows\n");
 774     exit(1);
 775 #else
 776     CPUArchState *env = arg;
 777     CPUState *cpu = ENV_GET_CPU(env);
 778     sigset_t waitset;
 779     int r;
 780
 781     qemu_mutex_lock_iothread();
 782     qemu_thread_get_self(cpu->thread);
 783     cpu->thread_id = qemu_get_thread_id();
 784
 785     sigemptyset(&waitset);
 786     sigaddset(&waitset, SIG_IPI);
 787
 788     /* signal CPU creation */
 789     cpu->created = true;
 790     qemu_cond_signal(&qemu_cpu_cond);
 791
 792     cpu_single_env = env;
 793     while (1) {
 794         cpu_single_env = NULL;
 795         qemu_mutex_unlock_iothread();
 796         do {
 797             int sig;
 798             r = sigwait(&waitset, &sig);
 799         } while (r == -1 && (errno == EAGAIN || errno == EINTR));
 800         if (r == -1) {
 801             perror("sigwait");
 802             exit(1);
 803         }
 804         qemu_mutex_lock_iothread();
 805         cpu_single_env = env;
 806         qemu_wait_io_event_common(cpu);
 807     }
 808
 809     return NULL;
 810 #endif
 811 }
 812
 813 static void tcg_exec_all(void);
 814
 815 static void *qemu_tcg_cpu_thread_fn(void *arg)
 816 {
 817     CPUState *cpu = arg;
 818     CPUArchState *env;
 819
 820     qemu_tcg_init_cpu_signals();
 821     qemu_thread_get_self(cpu->thread);
 822
 823     /* signal CPU creation */
 824     qemu_mutex_lock(&qemu_global_mutex);
 825     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 826         cpu = ENV_GET_CPU(env);
 827         cpu->thread_id = qemu_get_thread_id();
 828         cpu->created = true;
 829     }
 830     qemu_cond_signal(&qemu_cpu_cond);
 831
 832     /* wait for initial kick-off after machine start */
 833     while (ENV_GET_CPU(first_cpu)->stopped) {
 834         qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
 835
 836         /* process any pending work */
 837         for (env = first_cpu; env != NULL; env = env->next_cpu) {
 838             qemu_wait_io_event_common(ENV_GET_CPU(env));
 839         }
 840     }
 841
 842     while (1) {
 843         tcg_exec_all();
 844         if (use_icount && qemu_clock_deadline(vm_clock) <= 0) {
 845             qemu_notify_event();
 846         }
 847         qemu_tcg_wait_io_event();
 848     }
 849
 850     return NULL;
 851 }
 852
 853 static void qemu_cpu_kick_thread(CPUState *cpu)
 854 {
 855 #ifndef _WIN32
 856     int err;
 857
 858     err = pthread_kill(cpu->thread->thread, SIG_IPI);
 859     if (err) {
 860         fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
 861         exit(1);
 862     }
 863 #else /* _WIN32 */
 864     if (!qemu_cpu_is_self(cpu)) {
 865         CONTEXT tcgContext;
 866
 867         if (SuspendThread(cpu->hThread) == (DWORD)-1) {
 868             fprintf(stderr, "qemu:%s: GetLastError:%d\n", __func__,
 869                     GetLastError());
 870             exit(1);
 871         }
 872
 873         /* On multi-core systems, we are not sure that the thread is actually
 874          * suspended until we can get the context.
 875          */
 876         tcgContext.ContextFlags = CONTEXT_CONTROL;
 877         while (GetThreadContext(cpu->hThread, &tcgContext) != 0) {
 878             continue;
 879         }
 880
 881         cpu_signal(0);
 882
 883         if (ResumeThread(cpu->hThread) == (DWORD)-1) {
 884             fprintf(stderr, "qemu:%s: GetLastError:%d\n", __func__,
 885                     GetLastError());
 886             exit(1);
 887         }
 888     }
 889 #endif
 890 }
 891
 892 void qemu_cpu_kick(CPUState *cpu)
 893 {
 894     qemu_cond_broadcast(cpu->halt_cond);
 895     if (!tcg_enabled() && !cpu->thread_kicked) {
 896         qemu_cpu_kick_thread(cpu);
 897         cpu->thread_kicked = true;
 898     }
 899 }
 900
 901 void qemu_cpu_kick_self(void)
 902 {
 903 #ifndef _WIN32
 904     assert(cpu_single_env);
 905     CPUState *cpu_single_cpu = ENV_GET_CPU(cpu_single_env);
 906
 907     if (!cpu_single_cpu->thread_kicked) {
 908         qemu_cpu_kick_thread(cpu_single_cpu);
 909         cpu_single_cpu->thread_kicked = true;
 910     }
 911 #else
 912     abort();
 913 #endif
 914 }
 915
 916 bool qemu_cpu_is_self(CPUState *cpu)
 917 {
 918     return qemu_thread_is_self(cpu->thread);
 919 }
 920
 921 static bool qemu_in_vcpu_thread(void)
 922 {
 923     return cpu_single_env && qemu_cpu_is_self(ENV_GET_CPU(cpu_single_env));
 924 }
 925
 926 void qemu_mutex_lock_iothread(void)
 927 {
 928     if (!tcg_enabled()) {
 929         qemu_mutex_lock(&qemu_global_mutex);
 930     } else {
 931         iothread_requesting_mutex = true;
 932         if (qemu_mutex_trylock(&qemu_global_mutex)) {
 933             qemu_cpu_kick_thread(ENV_GET_CPU(first_cpu));
 934             qemu_mutex_lock(&qemu_global_mutex);
 935         }
 936         iothread_requesting_mutex = false;
 937         qemu_cond_broadcast(&qemu_io_proceeded_cond);
 938     }
 939 }
 940
 941 void qemu_mutex_unlock_iothread(void)
 942 {
 943     qemu_mutex_unlock(&qemu_global_mutex);
 944 }
 945
 946 static int all_vcpus_paused(void)
 947 {
 948     CPUArchState *penv = first_cpu;
 949
 950     while (penv) {
 951         CPUState *pcpu = ENV_GET_CPU(penv);
 952         if (!pcpu->stopped) {
 953             return 0;
 954         }
 955         penv = penv->next_cpu;
 956     }
 957
 958     return 1;
 959 }
 960
 961 void pause_all_vcpus(void)
 962 {
 963     CPUArchState *penv = first_cpu;
 964
 965     qemu_clock_enable(vm_clock, false);
 966     while (penv) {
 967         CPUState *pcpu = ENV_GET_CPU(penv);
 968         pcpu->stop = true;
 969         qemu_cpu_kick(pcpu);
 970         penv = penv->next_cpu;
 971     }
 972
 973     if (qemu_in_vcpu_thread()) {
 974         cpu_stop_current();
 975         if (!kvm_enabled()) {
 976             while (penv) {
 977                 CPUState *pcpu = ENV_GET_CPU(penv);
 978                 pcpu->stop = 0;
 979                 pcpu->stopped = true;
 980                 penv = penv->next_cpu;
 981             }
 982             return;
 983         }
 984     }
 985
 986     while (!all_vcpus_paused()) {
 987         qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex);
 988         penv = first_cpu;
 989         while (penv) {
 990             qemu_cpu_kick(ENV_GET_CPU(penv));
 991             penv = penv->next_cpu;
 992         }
 993     }
 994 }
 995
 996 void resume_all_vcpus(void)
 997 {
 998     CPUArchState *penv = first_cpu;
 999
1000     qemu_clock_enable(vm_clock, true);
1001     while (penv) {
1002         CPUState *pcpu = ENV_GET_CPU(penv);
1003         pcpu->stop = false;
1004         pcpu->stopped = false;
1005         qemu_cpu_kick(pcpu);
1006         penv = penv->next_cpu;
1007     }
1008 }
1009
1010 static void qemu_tcg_init_vcpu(CPUState *cpu)
1011 {
1012     /* share a single thread for all cpus with TCG */
1013     if (!tcg_cpu_thread) {
1014         cpu->thread = g_malloc0(sizeof(QemuThread));
1015         cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1016         qemu_cond_init(cpu->halt_cond);
1017         tcg_halt_cond = cpu->halt_cond;
1018         qemu_thread_create(cpu->thread, qemu_tcg_cpu_thread_fn, cpu,
1019                            QEMU_THREAD_JOINABLE);
1020 #ifdef _WIN32
1021         cpu->hThread = qemu_thread_get_handle(cpu->thread);
1022 #endif
1023         while (!cpu->created) {
1024             qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1025         }
1026         tcg_cpu_thread = cpu->thread;
1027     } else {
1028         cpu->thread = tcg_cpu_thread;
1029         cpu->halt_cond = tcg_halt_cond;
1030     }
1031 }
1032
1033 static void qemu_kvm_start_vcpu(CPUArchState *env)
1034 {
1035     CPUState *cpu = ENV_GET_CPU(env);
1036
1037     cpu->thread = g_malloc0(sizeof(QemuThread));
1038     cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1039     qemu_cond_init(cpu->halt_cond);
1040     qemu_thread_create(cpu->thread, qemu_kvm_cpu_thread_fn, env,
1041                        QEMU_THREAD_JOINABLE);
1042     while (!cpu->created) {
1043         qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1044     }
1045 }
1046
1047 static void qemu_dummy_start_vcpu(CPUArchState *env)
1048 {
1049     CPUState *cpu = ENV_GET_CPU(env);
1050
1051     cpu->thread = g_malloc0(sizeof(QemuThread));
1052     cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1053     qemu_cond_init(cpu->halt_cond);
1054     qemu_thread_create(cpu->thread, qemu_dummy_cpu_thread_fn, env,
1055                        QEMU_THREAD_JOINABLE);
1056     while (!cpu->created) {
1057         qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1058     }
1059 }
1060
1061 void qemu_init_vcpu(void *_env)
1062 {
1063     CPUArchState *env = _env;
1064     CPUState *cpu = ENV_GET_CPU(env);
1065
1066     cpu->nr_cores = smp_cores;
1067     cpu->nr_threads = smp_threads;
1068     cpu->stopped = true;
1069     if (kvm_enabled()) {
1070         qemu_kvm_start_vcpu(env);
1071     } else if (tcg_enabled()) {
1072         qemu_tcg_init_vcpu(cpu);
1073     } else {
1074         qemu_dummy_start_vcpu(env);
1075     }
1076 }
1077
1078 void cpu_stop_current(void)
1079 {
1080     if (cpu_single_env) {
1081         CPUState *cpu_single_cpu = ENV_GET_CPU(cpu_single_env);
1082         cpu_single_cpu->stop = false;
1083         cpu_single_cpu->stopped = true;
1084         cpu_exit(cpu_single_env);
1085         qemu_cond_signal(&qemu_pause_cond);
1086     }
1087 }
1088
1089 void vm_stop(RunState state)
1090 {
1091     if (qemu_in_vcpu_thread()) {
1092         qemu_system_vmstop_request(state);
1093         /*
1094          * FIXME: should not return to device code in case
1095          * vm_stop() has been requested.
1096          */
1097         cpu_stop_current();
1098         return;
1099     }
1100     do_vm_stop(state);
1101 }
1102
1103 /* does a state transition even if the VM is already stopped,
1104    current state is forgotten forever */
1105 void vm_stop_force_state(RunState state)
1106 {
1107     if (runstate_is_running()) {
1108         vm_stop(state);
1109     } else {
1110         runstate_set(state);
1111     }
1112 }
1113
1114 static int tcg_cpu_exec(CPUArchState *env)
1115 {
1116     int ret;
1117 #ifdef CONFIG_PROFILER
1118     int64_t ti;
1119 #endif
1120
1121 #ifdef CONFIG_PROFILER
1122     ti = profile_getclock();
1123 #endif
1124     if (use_icount) {
1125         int64_t count;
1126         int decr;
1127         qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
1128         env->icount_decr.u16.low = 0;
1129         env->icount_extra = 0;
1130         count = qemu_icount_round(qemu_clock_deadline(vm_clock));
1131         qemu_icount += count;
1132         decr = (count > 0xffff) ? 0xffff : count;
1133         count -= decr;
1134         env->icount_decr.u16.low = decr;
1135         env->icount_extra = count;
1136     }
1137     ret = cpu_exec(env);
1138 #ifdef CONFIG_PROFILER
1139     qemu_time += profile_getclock() - ti;
1140 #endif
1141     if (use_icount) {
1142         /* Fold pending instructions back into the
1143            instruction counter, and clear the interrupt flag.  */
1144         qemu_icount -= (env->icount_decr.u16.low
1145                         + env->icount_extra);
1146         env->icount_decr.u32 = 0;
1147         env->icount_extra = 0;
1148     }
1149     return ret;
1150 }
1151
1152 static void tcg_exec_all(void)
1153 {
1154     int r;
1155
1156     /* Account partial waits to the vm_clock.  */
1157     qemu_clock_warp(vm_clock);
1158
1159     if (next_cpu == NULL) {
1160         next_cpu = first_cpu;
1161     }
1162     for (; next_cpu != NULL && !exit_request; next_cpu = next_cpu->next_cpu) {
1163         CPUArchState *env = next_cpu;
1164         CPUState *cpu = ENV_GET_CPU(env);
1165
1166         qemu_clock_enable(vm_clock,
1167                           (env->singlestep_enabled & SSTEP_NOTIMER) == 0);
1168
1169         if (cpu_can_run(cpu)) {
1170             r = tcg_cpu_exec(env);
1171             if (r == EXCP_DEBUG) {
1172                 cpu_handle_guest_debug(env);
1173                 break;
1174             }
1175         } else if (cpu->stop || cpu->stopped) {
1176             break;
1177         }
1178     }
1179     exit_request = 0;
1180 }
1181
1182 void set_numa_modes(void)
1183 {
1184     CPUArchState *env;
1185     CPUState *cpu;
1186     int i;
1187
1188     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1189         cpu = ENV_GET_CPU(env);
1190         for (i = 0; i < nb_numa_nodes; i++) {
1191             if (test_bit(cpu->cpu_index, node_cpumask[i])) {
1192                 cpu->numa_node = i;
1193             }
1194         }
1195     }
1196 }
1197
1198 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
1199 {
1200     /* XXX: implement xxx_cpu_list for targets that still miss it */
1201 #if defined(cpu_list)
1202     cpu_list(f, cpu_fprintf);
1203 #endif
1204 }
1205
1206 CpuInfoList *qmp_query_cpus(Error **errp)
1207 {
1208     CpuInfoList *head = NULL, *cur_item = NULL;
1209     CPUArchState *env;
1210
1211     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1212         CPUState *cpu = ENV_GET_CPU(env);
1213         CpuInfoList *info;
1214
1215         cpu_synchronize_state(env);
1216
1217         info = g_malloc0(sizeof(*info));
1218         info->value = g_malloc0(sizeof(*info->value));
1219         info->value->CPU = cpu->cpu_index;
1220         info->value->current = (env == first_cpu);
1221         info->value->halted = cpu->halted;
1222         info->value->thread_id = cpu->thread_id;
1223 #if defined(TARGET_I386)
1224         info->value->has_pc = true;
1225         info->value->pc = env->eip + env->segs[R_CS].base;
1226 #elif defined(TARGET_PPC)
1227         info->value->has_nip = true;
1228         info->value->nip = env->nip;
1229 #elif defined(TARGET_SPARC)
1230         info->value->has_pc = true;
1231         info->value->pc = env->pc;
1232         info->value->has_npc = true;
1233         info->value->npc = env->npc;
1234 #elif defined(TARGET_MIPS)
1235         info->value->has_PC = true;
1236         info->value->PC = env->active_tc.PC;
1237 #endif
1238
1239         /* XXX: waiting for the qapi to support GSList */
1240         if (!cur_item) {
1241             head = cur_item = info;
1242         } else {
1243             cur_item->next = info;
1244             cur_item = info;
1245         }
1246     }
1247
1248     return head;
1249 }
1250
1251 void qmp_memsave(int64_t addr, int64_t size, const char *filename,
1252                  bool has_cpu, int64_t cpu_index, Error **errp)
1253 {
1254     FILE *f;
1255     uint32_t l;
1256     CPUArchState *env;
1257     CPUState *cpu;
1258     uint8_t buf[1024];
1259
1260     if (!has_cpu) {
1261         cpu_index = 0;
1262     }
1263
1264     cpu = qemu_get_cpu(cpu_index);
1265     if (cpu == NULL) {
1266         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
1267                   "a CPU number");
1268         return;
1269     }
1270     env = cpu->env_ptr;
1271
1272     f = fopen(filename, "wb");
1273     if (!f) {
1274         error_set(errp, QERR_OPEN_FILE_FAILED, filename);
1275         return;
1276     }
1277
1278     while (size != 0) {
1279         l = sizeof(buf);
1280         if (l > size)
1281             l = size;
1282         cpu_memory_rw_debug(env, addr, buf, l, 0);
1283         if (fwrite(buf, 1, l, f) != l) {
1284             error_set(errp, QERR_IO_ERROR);
1285             goto exit;
1286         }
1287         addr += l;
1288         size -= l;
1289     }
1290
1291 exit:
1292     fclose(f);
1293 }
1294
1295 void qmp_pmemsave(int64_t addr, int64_t size, const char *filename,
1296                   Error **errp)
1297 {
1298     FILE *f;
1299     uint32_t l;
1300     uint8_t buf[1024];
1301
1302     f = fopen(filename, "wb");
1303     if (!f) {
1304         error_set(errp, QERR_OPEN_FILE_FAILED, filename);
1305         return;
1306     }
1307
1308     while (size != 0) {
1309         l = sizeof(buf);
1310         if (l > size)
1311             l = size;
1312         cpu_physical_memory_rw(addr, buf, l, 0);
1313         if (fwrite(buf, 1, l, f) != l) {
1314             error_set(errp, QERR_IO_ERROR);
1315             goto exit;
1316         }
1317         addr += l;
1318         size -= l;
1319     }
1320
1321 exit:
1322     fclose(f);
1323 }
1324
1325 void qmp_inject_nmi(Error **errp)
1326 {
1327 #if defined(TARGET_I386)
1328     CPUArchState *env;
1329
1330     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1331         if (!env->apic_state) {
1332             cpu_interrupt(CPU(x86_env_get_cpu(env)), CPU_INTERRUPT_NMI);
1333         } else {
1334             apic_deliver_nmi(env->apic_state);
1335         }
1336     }
1337 #else
1338     error_set(errp, QERR_UNSUPPORTED);
1339 #endif
1340 }