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(CPUState *cpu)
  66 {
  67     if (cpu->stop || cpu->queued_work_first) {
  68         return false;
  69     }
  70     if (cpu->stopped || !runstate_is_running()) {
  71         return true;
  72     }
  73     if (!cpu->halted || qemu_cpu_has_work(cpu) ||
  74         kvm_async_interrupts_enabled()) {
  75         return false;
  76     }
  77     return true;
  78 }
  79
  80 static bool all_cpu_threads_idle(void)
  81 {
  82     CPUArchState *env;
  83
  84     for (env = first_cpu; env != NULL; env = env->next_cpu) {
  85         if (!cpu_thread_is_idle(ENV_GET_CPU(env))) {
  86             return false;
  87         }
  88     }
  89     return true;
  90 }
  91
  92 /***********************************************************/
  93 /* guest cycle counter */
  94
  95 /* Conversion factor from emulated instructions to virtual clock ticks.  */
  96 static int icount_time_shift;
  97 /* Arbitrarily pick 1MIPS as the minimum allowable speed.  */
  98 #define MAX_ICOUNT_SHIFT 10
  99 /* Compensate for varying guest execution speed.  */
 100 static int64_t qemu_icount_bias;
 101 static QEMUTimer *icount_rt_timer;
 102 static QEMUTimer *icount_vm_timer;
 103 static QEMUTimer *icount_warp_timer;
 104 static int64_t vm_clock_warp_start;
 105 static int64_t qemu_icount;
 106
 107 typedef struct TimersState {
 108     int64_t cpu_ticks_prev;
 109     int64_t cpu_ticks_offset;
 110     int64_t cpu_clock_offset;
 111     int32_t cpu_ticks_enabled;
 112     int64_t dummy;
 113 } TimersState;
 114
 115 TimersState timers_state;
 116
 117 /* Return the virtual CPU time, based on the instruction counter.  */
 118 int64_t cpu_get_icount(void)
 119 {
 120     int64_t icount;
 121     CPUArchState *env = cpu_single_env;
 122
 123     icount = qemu_icount;
 124     if (env) {
 125         if (!can_do_io(env)) {
 126             fprintf(stderr, "Bad clock read\n");
 127         }
 128         icount -= (env->icount_decr.u16.low + env->icount_extra);
 129     }
 130     return qemu_icount_bias + (icount << icount_time_shift);
 131 }
 132
 133 /* return the host CPU cycle counter and handle stop/restart */
 134 int64_t cpu_get_ticks(void)
 135 {
 136     if (use_icount) {
 137         return cpu_get_icount();
 138     }
 139     if (!timers_state.cpu_ticks_enabled) {
 140         return timers_state.cpu_ticks_offset;
 141     } else {
 142         int64_t ticks;
 143         ticks = cpu_get_real_ticks();
 144         if (timers_state.cpu_ticks_prev > ticks) {
 145             /* Note: non increasing ticks may happen if the host uses
 146                software suspend */
 147             timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks;
 148         }
 149         timers_state.cpu_ticks_prev = ticks;
 150         return ticks + timers_state.cpu_ticks_offset;
 151     }
 152 }
 153
 154 /* return the host CPU monotonic timer and handle stop/restart */
 155 int64_t cpu_get_clock(void)
 156 {
 157     int64_t ti;
 158     if (!timers_state.cpu_ticks_enabled) {
 159         return timers_state.cpu_clock_offset;
 160     } else {
 161         ti = get_clock();
 162         return ti + timers_state.cpu_clock_offset;
 163     }
 164 }
 165
 166 /* enable cpu_get_ticks() */
 167 void cpu_enable_ticks(void)
 168 {
 169     if (!timers_state.cpu_ticks_enabled) {
 170         timers_state.cpu_ticks_offset -= cpu_get_real_ticks();
 171         timers_state.cpu_clock_offset -= get_clock();
 172         timers_state.cpu_ticks_enabled = 1;
 173     }
 174 }
 175
 176 /* disable cpu_get_ticks() : the clock is stopped. You must not call
 177    cpu_get_ticks() after that.  */
 178 void cpu_disable_ticks(void)
 179 {
 180     if (timers_state.cpu_ticks_enabled) {
 181         timers_state.cpu_ticks_offset = cpu_get_ticks();
 182         timers_state.cpu_clock_offset = cpu_get_clock();
 183         timers_state.cpu_ticks_enabled = 0;
 184     }
 185 }
 186
 187 /* Correlation between real and virtual time is always going to be
 188    fairly approximate, so ignore small variation.
 189    When the guest is idle real and virtual time will be aligned in
 190    the IO wait loop.  */
 191 #define ICOUNT_WOBBLE (get_ticks_per_sec() / 10)
 192
 193 static void icount_adjust(void)
 194 {
 195     int64_t cur_time;
 196     int64_t cur_icount;
 197     int64_t delta;
 198     static int64_t last_delta;
 199     /* If the VM is not running, then do nothing.  */
 200     if (!runstate_is_running()) {
 201         return;
 202     }
 203     cur_time = cpu_get_clock();
 204     cur_icount = qemu_get_clock_ns(vm_clock);
 205     delta = cur_icount - cur_time;
 206     /* FIXME: This is a very crude algorithm, somewhat prone to oscillation.  */
 207     if (delta > 0
 208         && last_delta + ICOUNT_WOBBLE < delta * 2
 209         && icount_time_shift > 0) {
 210         /* The guest is getting too far ahead.  Slow time down.  */
 211         icount_time_shift--;
 212     }
 213     if (delta < 0
 214         && last_delta - ICOUNT_WOBBLE > delta * 2
 215         && icount_time_shift < MAX_ICOUNT_SHIFT) {
 216         /* The guest is getting too far behind.  Speed time up.  */
 217         icount_time_shift++;
 218     }
 219     last_delta = delta;
 220     qemu_icount_bias = cur_icount - (qemu_icount << icount_time_shift);
 221 }
 222
 223 static void icount_adjust_rt(void *opaque)
 224 {
 225     qemu_mod_timer(icount_rt_timer,
 226                    qemu_get_clock_ms(rt_clock) + 1000);
 227     icount_adjust();
 228 }
 229
 230 static void icount_adjust_vm(void *opaque)
 231 {
 232     qemu_mod_timer(icount_vm_timer,
 233                    qemu_get_clock_ns(vm_clock) + get_ticks_per_sec() / 10);
 234     icount_adjust();
 235 }
 236
 237 static int64_t qemu_icount_round(int64_t count)
 238 {
 239     return (count + (1 << icount_time_shift) - 1) >> icount_time_shift;
 240 }
 241
 242 static void icount_warp_rt(void *opaque)
 243 {
 244     if (vm_clock_warp_start == -1) {
 245         return;
 246     }
 247
 248     if (runstate_is_running()) {
 249         int64_t clock = qemu_get_clock_ns(rt_clock);
 250         int64_t warp_delta = clock - vm_clock_warp_start;
 251         if (use_icount == 1) {
 252             qemu_icount_bias += warp_delta;
 253         } else {
 254             /*
 255              * In adaptive mode, do not let the vm_clock run too
 256              * far ahead of real time.
 257              */
 258             int64_t cur_time = cpu_get_clock();
 259             int64_t cur_icount = qemu_get_clock_ns(vm_clock);
 260             int64_t delta = cur_time - cur_icount;
 261             qemu_icount_bias += MIN(warp_delta, delta);
 262         }
 263         if (qemu_clock_expired(vm_clock)) {
 264             qemu_notify_event();
 265         }
 266     }
 267     vm_clock_warp_start = -1;
 268 }
 269
 270 void qtest_clock_warp(int64_t dest)
 271 {
 272     int64_t clock = qemu_get_clock_ns(vm_clock);
 273     assert(qtest_enabled());
 274     while (clock < dest) {
 275         int64_t deadline = qemu_clock_deadline(vm_clock);
 276         int64_t warp = MIN(dest - clock, deadline);
 277         qemu_icount_bias += warp;
 278         qemu_run_timers(vm_clock);
 279         clock = qemu_get_clock_ns(vm_clock);
 280     }
 281     qemu_notify_event();
 282 }
 283
 284 void qemu_clock_warp(QEMUClock *clock)
 285 {
 286     int64_t deadline;
 287
 288     /*
 289      * There are too many global variables to make the "warp" behavior
 290      * applicable to other clocks.  But a clock argument removes the
 291      * need for if statements all over the place.
 292      */
 293     if (clock != vm_clock || !use_icount) {
 294         return;
 295     }
 296
 297     /*
 298      * If the CPUs have been sleeping, advance the vm_clock timer now.  This
 299      * ensures that the deadline for the timer is computed correctly below.
 300      * This also makes sure that the insn counter is synchronized before the
 301      * CPU starts running, in case the CPU is woken by an event other than
 302      * the earliest vm_clock timer.
 303      */
 304     icount_warp_rt(NULL);
 305     if (!all_cpu_threads_idle() || !qemu_clock_has_timers(vm_clock)) {
 306         qemu_del_timer(icount_warp_timer);
 307         return;
 308     }
 309
 310     if (qtest_enabled()) {
 311         /* When testing, qtest commands advance icount.  */
 312         return;
 313     }
 314
 315     vm_clock_warp_start = qemu_get_clock_ns(rt_clock);
 316     deadline = qemu_clock_deadline(vm_clock);
 317     if (deadline > 0) {
 318         /*
 319          * Ensure the vm_clock proceeds even when the virtual CPU goes to
 320          * sleep.  Otherwise, the CPU might be waiting for a future timer
 321          * interrupt to wake it up, but the interrupt never comes because
 322          * the vCPU isn't running any insns and thus doesn't advance the
 323          * vm_clock.
 324          *
 325          * An extreme solution for this problem would be to never let VCPUs
 326          * sleep in icount mode if there is a pending vm_clock timer; rather
 327          * time could just advance to the next vm_clock event.  Instead, we
 328          * do stop VCPUs and only advance vm_clock after some "real" time,
 329          * (related to the time left until the next event) has passed.  This
 330          * rt_clock timer will do this.  This avoids that the warps are too
 331          * visible externally---for example, you will not be sending network
 332          * packets continuously instead of every 100ms.
 333          */
 334         qemu_mod_timer(icount_warp_timer, vm_clock_warp_start + deadline);
 335     } else {
 336         qemu_notify_event();
 337     }
 338 }
 339
 340 static const VMStateDescription vmstate_timers = {
 341     .name = "timer",
 342     .version_id = 2,
 343     .minimum_version_id = 1,
 344     .minimum_version_id_old = 1,
 345     .fields      = (VMStateField[]) {
 346         VMSTATE_INT64(cpu_ticks_offset, TimersState),
 347         VMSTATE_INT64(dummy, TimersState),
 348         VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2),
 349         VMSTATE_END_OF_LIST()
 350     }
 351 };
 352
 353 void configure_icount(const char *option)
 354 {
 355     vmstate_register(NULL, 0, &vmstate_timers, &timers_state);
 356     if (!option) {
 357         return;
 358     }
 359
 360     icount_warp_timer = qemu_new_timer_ns(rt_clock, icount_warp_rt, NULL);
 361     if (strcmp(option, "auto") != 0) {
 362         icount_time_shift = strtol(option, NULL, 0);
 363         use_icount = 1;
 364         return;
 365     }
 366
 367     use_icount = 2;
 368
 369     /* 125MIPS seems a reasonable initial guess at the guest speed.
 370        It will be corrected fairly quickly anyway.  */
 371     icount_time_shift = 3;
 372
 373     /* Have both realtime and virtual time triggers for speed adjustment.
 374        The realtime trigger catches emulated time passing too slowly,
 375        the virtual time trigger catches emulated time passing too fast.
 376        Realtime triggers occur even when idle, so use them less frequently
 377        than VM triggers.  */
 378     icount_rt_timer = qemu_new_timer_ms(rt_clock, icount_adjust_rt, NULL);
 379     qemu_mod_timer(icount_rt_timer,
 380                    qemu_get_clock_ms(rt_clock) + 1000);
 381     icount_vm_timer = qemu_new_timer_ns(vm_clock, icount_adjust_vm, NULL);
 382     qemu_mod_timer(icount_vm_timer,
 383                    qemu_get_clock_ns(vm_clock) + get_ticks_per_sec() / 10);
 384 }
 385
 386 /***********************************************************/
 387 void hw_error(const char *fmt, ...)
 388 {
 389     va_list ap;
 390     CPUArchState *env;
 391     CPUState *cpu;
 392
 393     va_start(ap, fmt);
 394     fprintf(stderr, "qemu: hardware error: ");
 395     vfprintf(stderr, fmt, ap);
 396     fprintf(stderr, "\n");
 397     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 398         cpu = ENV_GET_CPU(env);
 399         fprintf(stderr, "CPU #%d:\n", cpu->cpu_index);
 400         cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU);
 401     }
 402     va_end(ap);
 403     abort();
 404 }
 405
 406 void cpu_synchronize_all_states(void)
 407 {
 408     CPUArchState *env;
 409
 410     for (env = first_cpu; env; env = env->next_cpu) {
 411         cpu_synchronize_state(ENV_GET_CPU(env));
 412     }
 413 }
 414
 415 void cpu_synchronize_all_post_reset(void)
 416 {
 417     CPUArchState *cpu;
 418
 419     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
 420         cpu_synchronize_post_reset(ENV_GET_CPU(cpu));
 421     }
 422 }
 423
 424 void cpu_synchronize_all_post_init(void)
 425 {
 426     CPUArchState *cpu;
 427
 428     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
 429         cpu_synchronize_post_init(ENV_GET_CPU(cpu));
 430     }
 431 }
 432
 433 bool cpu_is_stopped(CPUState *cpu)
 434 {
 435     return !runstate_is_running() || cpu->stopped;
 436 }
 437
 438 static void do_vm_stop(RunState state)
 439 {
 440     if (runstate_is_running()) {
 441         cpu_disable_ticks();
 442         pause_all_vcpus();
 443         runstate_set(state);
 444         vm_state_notify(0, state);
 445         bdrv_drain_all();
 446         bdrv_flush_all();
 447         monitor_protocol_event(QEVENT_STOP, NULL);
 448     }
 449 }
 450
 451 static bool cpu_can_run(CPUState *cpu)
 452 {
 453     if (cpu->stop) {
 454         return false;
 455     }
 456     if (cpu->stopped || !runstate_is_running()) {
 457         return false;
 458     }
 459     return true;
 460 }
 461
 462 static void cpu_handle_guest_debug(CPUArchState *env)
 463 {
 464     CPUState *cpu = ENV_GET_CPU(env);
 465
 466     gdb_set_stop_cpu(cpu);
 467     qemu_system_debug_request();
 468     cpu->stopped = true;
 469 }
 470
 471 static void cpu_signal(int sig)
 472 {
 473     if (cpu_single_env) {
 474         cpu_exit(ENV_GET_CPU(cpu_single_env));
 475     }
 476     exit_request = 1;
 477 }
 478
 479 #ifdef CONFIG_LINUX
 480 static void sigbus_reraise(void)
 481 {
 482     sigset_t set;
 483     struct sigaction action;
 484
 485     memset(&action, 0, sizeof(action));
 486     action.sa_handler = SIG_DFL;
 487     if (!sigaction(SIGBUS, &action, NULL)) {
 488         raise(SIGBUS);
 489         sigemptyset(&set);
 490         sigaddset(&set, SIGBUS);
 491         sigprocmask(SIG_UNBLOCK, &set, NULL);
 492     }
 493     perror("Failed to re-raise SIGBUS!\n");
 494     abort();
 495 }
 496
 497 static void sigbus_handler(int n, struct qemu_signalfd_siginfo *siginfo,
 498                            void *ctx)
 499 {
 500     if (kvm_on_sigbus(siginfo->ssi_code,
 501                       (void *)(intptr_t)siginfo->ssi_addr)) {
 502         sigbus_reraise();
 503     }
 504 }
 505
 506 static void qemu_init_sigbus(void)
 507 {
 508     struct sigaction action;
 509
 510     memset(&action, 0, sizeof(action));
 511     action.sa_flags = SA_SIGINFO;
 512     action.sa_sigaction = (void (*)(int, siginfo_t*, void*))sigbus_handler;
 513     sigaction(SIGBUS, &action, NULL);
 514
 515     prctl(PR_MCE_KILL, PR_MCE_KILL_SET, PR_MCE_KILL_EARLY, 0, 0);
 516 }
 517
 518 static void qemu_kvm_eat_signals(CPUState *cpu)
 519 {
 520     struct timespec ts = { 0, 0 };
 521     siginfo_t siginfo;
 522     sigset_t waitset;
 523     sigset_t chkset;
 524     int r;
 525
 526     sigemptyset(&waitset);
 527     sigaddset(&waitset, SIG_IPI);
 528     sigaddset(&waitset, SIGBUS);
 529
 530     do {
 531         r = sigtimedwait(&waitset, &siginfo, &ts);
 532         if (r == -1 && !(errno == EAGAIN || errno == EINTR)) {
 533             perror("sigtimedwait");
 534             exit(1);
 535         }
 536
 537         switch (r) {
 538         case SIGBUS:
 539             if (kvm_on_sigbus_vcpu(cpu, siginfo.si_code, siginfo.si_addr)) {
 540                 sigbus_reraise();
 541             }
 542             break;
 543         default:
 544             break;
 545         }
 546
 547         r = sigpending(&chkset);
 548         if (r == -1) {
 549             perror("sigpending");
 550             exit(1);
 551         }
 552     } while (sigismember(&chkset, SIG_IPI) || sigismember(&chkset, SIGBUS));
 553 }
 554
 555 #else /* !CONFIG_LINUX */
 556
 557 static void qemu_init_sigbus(void)
 558 {
 559 }
 560
 561 static void qemu_kvm_eat_signals(CPUState *cpu)
 562 {
 563 }
 564 #endif /* !CONFIG_LINUX */
 565
 566 #ifndef _WIN32
 567 static void dummy_signal(int sig)
 568 {
 569 }
 570
 571 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
 572 {
 573     int r;
 574     sigset_t set;
 575     struct sigaction sigact;
 576
 577     memset(&sigact, 0, sizeof(sigact));
 578     sigact.sa_handler = dummy_signal;
 579     sigaction(SIG_IPI, &sigact, NULL);
 580
 581     pthread_sigmask(SIG_BLOCK, NULL, &set);
 582     sigdelset(&set, SIG_IPI);
 583     sigdelset(&set, SIGBUS);
 584     r = kvm_set_signal_mask(cpu, &set);
 585     if (r) {
 586         fprintf(stderr, "kvm_set_signal_mask: %s\n", strerror(-r));
 587         exit(1);
 588     }
 589 }
 590
 591 static void qemu_tcg_init_cpu_signals(void)
 592 {
 593     sigset_t set;
 594     struct sigaction sigact;
 595
 596     memset(&sigact, 0, sizeof(sigact));
 597     sigact.sa_handler = cpu_signal;
 598     sigaction(SIG_IPI, &sigact, NULL);
 599
 600     sigemptyset(&set);
 601     sigaddset(&set, SIG_IPI);
 602     pthread_sigmask(SIG_UNBLOCK, &set, NULL);
 603 }
 604
 605 #else /* _WIN32 */
 606 static void qemu_kvm_init_cpu_signals(CPUState *cpu)
 607 {
 608     abort();
 609 }
 610
 611 static void qemu_tcg_init_cpu_signals(void)
 612 {
 613 }
 614 #endif /* _WIN32 */
 615
 616 static QemuMutex qemu_global_mutex;
 617 static QemuCond qemu_io_proceeded_cond;
 618 static bool iothread_requesting_mutex;
 619
 620 static QemuThread io_thread;
 621
 622 static QemuThread *tcg_cpu_thread;
 623 static QemuCond *tcg_halt_cond;
 624
 625 /* cpu creation */
 626 static QemuCond qemu_cpu_cond;
 627 /* system init */
 628 static QemuCond qemu_pause_cond;
 629 static QemuCond qemu_work_cond;
 630
 631 void qemu_init_cpu_loop(void)
 632 {
 633     qemu_init_sigbus();
 634     qemu_cond_init(&qemu_cpu_cond);
 635     qemu_cond_init(&qemu_pause_cond);
 636     qemu_cond_init(&qemu_work_cond);
 637     qemu_cond_init(&qemu_io_proceeded_cond);
 638     qemu_mutex_init(&qemu_global_mutex);
 639
 640     qemu_thread_get_self(&io_thread);
 641 }
 642
 643 void run_on_cpu(CPUState *cpu, void (*func)(void *data), void *data)
 644 {
 645     struct qemu_work_item wi;
 646
 647     if (qemu_cpu_is_self(cpu)) {
 648         func(data);
 649         return;
 650     }
 651
 652     wi.func = func;
 653     wi.data = data;
 654     if (cpu->queued_work_first == NULL) {
 655         cpu->queued_work_first = &wi;
 656     } else {
 657         cpu->queued_work_last->next = &wi;
 658     }
 659     cpu->queued_work_last = &wi;
 660     wi.next = NULL;
 661     wi.done = false;
 662
 663     qemu_cpu_kick(cpu);
 664     while (!wi.done) {
 665         CPUArchState *self_env = cpu_single_env;
 666
 667         qemu_cond_wait(&qemu_work_cond, &qemu_global_mutex);
 668         cpu_single_env = self_env;
 669     }
 670 }
 671
 672 static void flush_queued_work(CPUState *cpu)
 673 {
 674     struct qemu_work_item *wi;
 675
 676     if (cpu->queued_work_first == NULL) {
 677         return;
 678     }
 679
 680     while ((wi = cpu->queued_work_first)) {
 681         cpu->queued_work_first = wi->next;
 682         wi->func(wi->data);
 683         wi->done = true;
 684     }
 685     cpu->queued_work_last = NULL;
 686     qemu_cond_broadcast(&qemu_work_cond);
 687 }
 688
 689 static void qemu_wait_io_event_common(CPUState *cpu)
 690 {
 691     if (cpu->stop) {
 692         cpu->stop = false;
 693         cpu->stopped = true;
 694         qemu_cond_signal(&qemu_pause_cond);
 695     }
 696     flush_queued_work(cpu);
 697     cpu->thread_kicked = false;
 698 }
 699
 700 static void qemu_tcg_wait_io_event(void)
 701 {
 702     CPUArchState *env;
 703
 704     while (all_cpu_threads_idle()) {
 705        /* Start accounting real time to the virtual clock if the CPUs
 706           are idle.  */
 707         qemu_clock_warp(vm_clock);
 708         qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
 709     }
 710
 711     while (iothread_requesting_mutex) {
 712         qemu_cond_wait(&qemu_io_proceeded_cond, &qemu_global_mutex);
 713     }
 714
 715     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 716         qemu_wait_io_event_common(ENV_GET_CPU(env));
 717     }
 718 }
 719
 720 static void qemu_kvm_wait_io_event(CPUState *cpu)
 721 {
 722     while (cpu_thread_is_idle(cpu)) {
 723         qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
 724     }
 725
 726     qemu_kvm_eat_signals(cpu);
 727     qemu_wait_io_event_common(cpu);
 728 }
 729
 730 static void *qemu_kvm_cpu_thread_fn(void *arg)
 731 {
 732     CPUArchState *env = arg;
 733     CPUState *cpu = ENV_GET_CPU(env);
 734     int r;
 735
 736     qemu_mutex_lock(&qemu_global_mutex);
 737     qemu_thread_get_self(cpu->thread);
 738     cpu->thread_id = qemu_get_thread_id();
 739     cpu_single_env = env;
 740
 741     r = kvm_init_vcpu(cpu);
 742     if (r < 0) {
 743         fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r));
 744         exit(1);
 745     }
 746
 747     qemu_kvm_init_cpu_signals(cpu);
 748
 749     /* signal CPU creation */
 750     cpu->created = true;
 751     qemu_cond_signal(&qemu_cpu_cond);
 752
 753     while (1) {
 754         if (cpu_can_run(cpu)) {
 755             r = kvm_cpu_exec(cpu);
 756             if (r == EXCP_DEBUG) {
 757                 cpu_handle_guest_debug(env);
 758             }
 759         }
 760         qemu_kvm_wait_io_event(cpu);
 761     }
 762
 763     return NULL;
 764 }
 765
 766 static void *qemu_dummy_cpu_thread_fn(void *arg)
 767 {
 768 #ifdef _WIN32
 769     fprintf(stderr, "qtest is not supported under Windows\n");
 770     exit(1);
 771 #else
 772     CPUArchState *env = arg;
 773     CPUState *cpu = ENV_GET_CPU(env);
 774     sigset_t waitset;
 775     int r;
 776
 777     qemu_mutex_lock_iothread();
 778     qemu_thread_get_self(cpu->thread);
 779     cpu->thread_id = qemu_get_thread_id();
 780
 781     sigemptyset(&waitset);
 782     sigaddset(&waitset, SIG_IPI);
 783
 784     /* signal CPU creation */
 785     cpu->created = true;
 786     qemu_cond_signal(&qemu_cpu_cond);
 787
 788     cpu_single_env = env;
 789     while (1) {
 790         cpu_single_env = NULL;
 791         qemu_mutex_unlock_iothread();
 792         do {
 793             int sig;
 794             r = sigwait(&waitset, &sig);
 795         } while (r == -1 && (errno == EAGAIN || errno == EINTR));
 796         if (r == -1) {
 797             perror("sigwait");
 798             exit(1);
 799         }
 800         qemu_mutex_lock_iothread();
 801         cpu_single_env = env;
 802         qemu_wait_io_event_common(cpu);
 803     }
 804
 805     return NULL;
 806 #endif
 807 }
 808
 809 static void tcg_exec_all(void);
 810
 811 static void tcg_signal_cpu_creation(CPUState *cpu, void *data)
 812 {
 813     cpu->thread_id = qemu_get_thread_id();
 814     cpu->created = true;
 815 }
 816
 817 static void *qemu_tcg_cpu_thread_fn(void *arg)
 818 {
 819     CPUState *cpu = arg;
 820     CPUArchState *env;
 821
 822     qemu_tcg_init_cpu_signals();
 823     qemu_thread_get_self(cpu->thread);
 824
 825     qemu_mutex_lock(&qemu_global_mutex);
 826     qemu_for_each_cpu(tcg_signal_cpu_creation, NULL);
 827     qemu_cond_signal(&qemu_cpu_cond);
 828
 829     /* wait for initial kick-off after machine start */
 830     while (ENV_GET_CPU(first_cpu)->stopped) {
 831         qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
 832
 833         /* process any pending work */
 834         for (env = first_cpu; env != NULL; env = env->next_cpu) {
 835             qemu_wait_io_event_common(ENV_GET_CPU(env));
 836         }
 837     }
 838
 839     while (1) {
 840         tcg_exec_all();
 841         if (use_icount && qemu_clock_deadline(vm_clock) <= 0) {
 842             qemu_notify_event();
 843         }
 844         qemu_tcg_wait_io_event();
 845     }
 846
 847     return NULL;
 848 }
 849
 850 static void qemu_cpu_kick_thread(CPUState *cpu)
 851 {
 852 #ifndef _WIN32
 853     int err;
 854
 855     err = pthread_kill(cpu->thread->thread, SIG_IPI);
 856     if (err) {
 857         fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
 858         exit(1);
 859     }
 860 #else /* _WIN32 */
 861     if (!qemu_cpu_is_self(cpu)) {
 862         CONTEXT tcgContext;
 863
 864         if (SuspendThread(cpu->hThread) == (DWORD)-1) {
 865             fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
 866                     GetLastError());
 867             exit(1);
 868         }
 869
 870         /* On multi-core systems, we are not sure that the thread is actually
 871          * suspended until we can get the context.
 872          */
 873         tcgContext.ContextFlags = CONTEXT_CONTROL;
 874         while (GetThreadContext(cpu->hThread, &tcgContext) != 0) {
 875             continue;
 876         }
 877
 878         cpu_signal(0);
 879
 880         if (ResumeThread(cpu->hThread) == (DWORD)-1) {
 881             fprintf(stderr, "qemu:%s: GetLastError:%lu\n", __func__,
 882                     GetLastError());
 883             exit(1);
 884         }
 885     }
 886 #endif
 887 }
 888
 889 void qemu_cpu_kick(CPUState *cpu)
 890 {
 891     qemu_cond_broadcast(cpu->halt_cond);
 892     if (!tcg_enabled() && !cpu->thread_kicked) {
 893         qemu_cpu_kick_thread(cpu);
 894         cpu->thread_kicked = true;
 895     }
 896 }
 897
 898 void qemu_cpu_kick_self(void)
 899 {
 900 #ifndef _WIN32
 901     assert(cpu_single_env);
 902     CPUState *cpu_single_cpu = ENV_GET_CPU(cpu_single_env);
 903
 904     if (!cpu_single_cpu->thread_kicked) {
 905         qemu_cpu_kick_thread(cpu_single_cpu);
 906         cpu_single_cpu->thread_kicked = true;
 907     }
 908 #else
 909     abort();
 910 #endif
 911 }
 912
 913 bool qemu_cpu_is_self(CPUState *cpu)
 914 {
 915     return qemu_thread_is_self(cpu->thread);
 916 }
 917
 918 static bool qemu_in_vcpu_thread(void)
 919 {
 920     return cpu_single_env && qemu_cpu_is_self(ENV_GET_CPU(cpu_single_env));
 921 }
 922
 923 void qemu_mutex_lock_iothread(void)
 924 {
 925     if (!tcg_enabled()) {
 926         qemu_mutex_lock(&qemu_global_mutex);
 927     } else {
 928         iothread_requesting_mutex = true;
 929         if (qemu_mutex_trylock(&qemu_global_mutex)) {
 930             qemu_cpu_kick_thread(ENV_GET_CPU(first_cpu));
 931             qemu_mutex_lock(&qemu_global_mutex);
 932         }
 933         iothread_requesting_mutex = false;
 934         qemu_cond_broadcast(&qemu_io_proceeded_cond);
 935     }
 936 }
 937
 938 void qemu_mutex_unlock_iothread(void)
 939 {
 940     qemu_mutex_unlock(&qemu_global_mutex);
 941 }
 942
 943 static int all_vcpus_paused(void)
 944 {
 945     CPUArchState *penv = first_cpu;
 946
 947     while (penv) {
 948         CPUState *pcpu = ENV_GET_CPU(penv);
 949         if (!pcpu->stopped) {
 950             return 0;
 951         }
 952         penv = penv->next_cpu;
 953     }
 954
 955     return 1;
 956 }
 957
 958 void pause_all_vcpus(void)
 959 {
 960     CPUArchState *penv = first_cpu;
 961
 962     qemu_clock_enable(vm_clock, false);
 963     while (penv) {
 964         CPUState *pcpu = ENV_GET_CPU(penv);
 965         pcpu->stop = true;
 966         qemu_cpu_kick(pcpu);
 967         penv = penv->next_cpu;
 968     }
 969
 970     if (qemu_in_vcpu_thread()) {
 971         cpu_stop_current();
 972         if (!kvm_enabled()) {
 973             penv = first_cpu;
 974             while (penv) {
 975                 CPUState *pcpu = ENV_GET_CPU(penv);
 976                 pcpu->stop = false;
 977                 pcpu->stopped = true;
 978                 penv = penv->next_cpu;
 979             }
 980             return;
 981         }
 982     }
 983
 984     while (!all_vcpus_paused()) {
 985         qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex);
 986         penv = first_cpu;
 987         while (penv) {
 988             qemu_cpu_kick(ENV_GET_CPU(penv));
 989             penv = penv->next_cpu;
 990         }
 991     }
 992 }
 993
 994 void cpu_resume(CPUState *cpu)
 995 {
 996     cpu->stop = false;
 997     cpu->stopped = false;
 998     qemu_cpu_kick(cpu);
 999 }
1000
1001 void resume_all_vcpus(void)
1002 {
1003     CPUArchState *penv = first_cpu;
1004
1005     qemu_clock_enable(vm_clock, true);
1006     while (penv) {
1007         CPUState *pcpu = ENV_GET_CPU(penv);
1008         cpu_resume(pcpu);
1009         penv = penv->next_cpu;
1010     }
1011 }
1012
1013 static void qemu_tcg_init_vcpu(CPUState *cpu)
1014 {
1015     /* share a single thread for all cpus with TCG */
1016     if (!tcg_cpu_thread) {
1017         cpu->thread = g_malloc0(sizeof(QemuThread));
1018         cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1019         qemu_cond_init(cpu->halt_cond);
1020         tcg_halt_cond = cpu->halt_cond;
1021         qemu_thread_create(cpu->thread, qemu_tcg_cpu_thread_fn, cpu,
1022                            QEMU_THREAD_JOINABLE);
1023 #ifdef _WIN32
1024         cpu->hThread = qemu_thread_get_handle(cpu->thread);
1025 #endif
1026         while (!cpu->created) {
1027             qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1028         }
1029         tcg_cpu_thread = cpu->thread;
1030     } else {
1031         cpu->thread = tcg_cpu_thread;
1032         cpu->halt_cond = tcg_halt_cond;
1033     }
1034 }
1035
1036 static void qemu_kvm_start_vcpu(CPUArchState *env)
1037 {
1038     CPUState *cpu = ENV_GET_CPU(env);
1039
1040     cpu->thread = g_malloc0(sizeof(QemuThread));
1041     cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1042     qemu_cond_init(cpu->halt_cond);
1043     qemu_thread_create(cpu->thread, qemu_kvm_cpu_thread_fn, env,
1044                        QEMU_THREAD_JOINABLE);
1045     while (!cpu->created) {
1046         qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1047     }
1048 }
1049
1050 static void qemu_dummy_start_vcpu(CPUArchState *env)
1051 {
1052     CPUState *cpu = ENV_GET_CPU(env);
1053
1054     cpu->thread = g_malloc0(sizeof(QemuThread));
1055     cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1056     qemu_cond_init(cpu->halt_cond);
1057     qemu_thread_create(cpu->thread, qemu_dummy_cpu_thread_fn, env,
1058                        QEMU_THREAD_JOINABLE);
1059     while (!cpu->created) {
1060         qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1061     }
1062 }
1063
1064 void qemu_init_vcpu(void *_env)
1065 {
1066     CPUArchState *env = _env;
1067     CPUState *cpu = ENV_GET_CPU(env);
1068
1069     cpu->nr_cores = smp_cores;
1070     cpu->nr_threads = smp_threads;
1071     cpu->stopped = true;
1072     if (kvm_enabled()) {
1073         qemu_kvm_start_vcpu(env);
1074     } else if (tcg_enabled()) {
1075         qemu_tcg_init_vcpu(cpu);
1076     } else {
1077         qemu_dummy_start_vcpu(env);
1078     }
1079 }
1080
1081 void cpu_stop_current(void)
1082 {
1083     if (cpu_single_env) {
1084         CPUState *cpu_single_cpu = ENV_GET_CPU(cpu_single_env);
1085         cpu_single_cpu->stop = false;
1086         cpu_single_cpu->stopped = true;
1087         cpu_exit(cpu_single_cpu);
1088         qemu_cond_signal(&qemu_pause_cond);
1089     }
1090 }
1091
1092 void vm_stop(RunState state)
1093 {
1094     if (qemu_in_vcpu_thread()) {
1095         qemu_system_vmstop_request(state);
1096         /*
1097          * FIXME: should not return to device code in case
1098          * vm_stop() has been requested.
1099          */
1100         cpu_stop_current();
1101         return;
1102     }
1103     do_vm_stop(state);
1104 }
1105
1106 /* does a state transition even if the VM is already stopped,
1107    current state is forgotten forever */
1108 void vm_stop_force_state(RunState state)
1109 {
1110     if (runstate_is_running()) {
1111         vm_stop(state);
1112     } else {
1113         runstate_set(state);
1114     }
1115 }
1116
1117 static int tcg_cpu_exec(CPUArchState *env)
1118 {
1119     int ret;
1120 #ifdef CONFIG_PROFILER
1121     int64_t ti;
1122 #endif
1123
1124 #ifdef CONFIG_PROFILER
1125     ti = profile_getclock();
1126 #endif
1127     if (use_icount) {
1128         int64_t count;
1129         int decr;
1130         qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
1131         env->icount_decr.u16.low = 0;
1132         env->icount_extra = 0;
1133         count = qemu_icount_round(qemu_clock_deadline(vm_clock));
1134         qemu_icount += count;
1135         decr = (count > 0xffff) ? 0xffff : count;
1136         count -= decr;
1137         env->icount_decr.u16.low = decr;
1138         env->icount_extra = count;
1139     }
1140     ret = cpu_exec(env);
1141 #ifdef CONFIG_PROFILER
1142     qemu_time += profile_getclock() - ti;
1143 #endif
1144     if (use_icount) {
1145         /* Fold pending instructions back into the
1146            instruction counter, and clear the interrupt flag.  */
1147         qemu_icount -= (env->icount_decr.u16.low
1148                         + env->icount_extra);
1149         env->icount_decr.u32 = 0;
1150         env->icount_extra = 0;
1151     }
1152     return ret;
1153 }
1154
1155 static void tcg_exec_all(void)
1156 {
1157     int r;
1158
1159     /* Account partial waits to the vm_clock.  */
1160     qemu_clock_warp(vm_clock);
1161
1162     if (next_cpu == NULL) {
1163         next_cpu = first_cpu;
1164     }
1165     for (; next_cpu != NULL && !exit_request; next_cpu = next_cpu->next_cpu) {
1166         CPUArchState *env = next_cpu;
1167         CPUState *cpu = ENV_GET_CPU(env);
1168
1169         qemu_clock_enable(vm_clock,
1170                           (env->singlestep_enabled & SSTEP_NOTIMER) == 0);
1171
1172         if (cpu_can_run(cpu)) {
1173             r = tcg_cpu_exec(env);
1174             if (r == EXCP_DEBUG) {
1175                 cpu_handle_guest_debug(env);
1176                 break;
1177             }
1178         } else if (cpu->stop || cpu->stopped) {
1179             break;
1180         }
1181     }
1182     exit_request = 0;
1183 }
1184
1185 void set_numa_modes(void)
1186 {
1187     CPUArchState *env;
1188     CPUState *cpu;
1189     int i;
1190
1191     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1192         cpu = ENV_GET_CPU(env);
1193         for (i = 0; i < nb_numa_nodes; i++) {
1194             if (test_bit(cpu->cpu_index, node_cpumask[i])) {
1195                 cpu->numa_node = i;
1196             }
1197         }
1198     }
1199 }
1200
1201 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
1202 {
1203     /* XXX: implement xxx_cpu_list for targets that still miss it */
1204 #if defined(cpu_list)
1205     cpu_list(f, cpu_fprintf);
1206 #endif
1207 }
1208
1209 CpuInfoList *qmp_query_cpus(Error **errp)
1210 {
1211     CpuInfoList *head = NULL, *cur_item = NULL;
1212     CPUArchState *env;
1213
1214     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1215         CPUState *cpu = ENV_GET_CPU(env);
1216         CpuInfoList *info;
1217
1218         cpu_synchronize_state(cpu);
1219
1220         info = g_malloc0(sizeof(*info));
1221         info->value = g_malloc0(sizeof(*info->value));
1222         info->value->CPU = cpu->cpu_index;
1223         info->value->current = (env == first_cpu);
1224         info->value->halted = cpu->halted;
1225         info->value->thread_id = cpu->thread_id;
1226 #if defined(TARGET_I386)
1227         info->value->has_pc = true;
1228         info->value->pc = env->eip + env->segs[R_CS].base;
1229 #elif defined(TARGET_PPC)
1230         info->value->has_nip = true;
1231         info->value->nip = env->nip;
1232 #elif defined(TARGET_SPARC)
1233         info->value->has_pc = true;
1234         info->value->pc = env->pc;
1235         info->value->has_npc = true;
1236         info->value->npc = env->npc;
1237 #elif defined(TARGET_MIPS)
1238         info->value->has_PC = true;
1239         info->value->PC = env->active_tc.PC;
1240 #endif
1241
1242         /* XXX: waiting for the qapi to support GSList */
1243         if (!cur_item) {
1244             head = cur_item = info;
1245         } else {
1246             cur_item->next = info;
1247             cur_item = info;
1248         }
1249     }
1250
1251     return head;
1252 }
1253
1254 void qmp_memsave(int64_t addr, int64_t size, const char *filename,
1255                  bool has_cpu, int64_t cpu_index, Error **errp)
1256 {
1257     FILE *f;
1258     uint32_t l;
1259     CPUArchState *env;
1260     CPUState *cpu;
1261     uint8_t buf[1024];
1262
1263     if (!has_cpu) {
1264         cpu_index = 0;
1265     }
1266
1267     cpu = qemu_get_cpu(cpu_index);
1268     if (cpu == NULL) {
1269         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
1270                   "a CPU number");
1271         return;
1272     }
1273     env = cpu->env_ptr;
1274
1275     f = fopen(filename, "wb");
1276     if (!f) {
1277         error_setg_file_open(errp, errno, filename);
1278         return;
1279     }
1280
1281     while (size != 0) {
1282         l = sizeof(buf);
1283         if (l > size)
1284             l = size;
1285         cpu_memory_rw_debug(env, addr, buf, l, 0);
1286         if (fwrite(buf, 1, l, f) != l) {
1287             error_set(errp, QERR_IO_ERROR);
1288             goto exit;
1289         }
1290         addr += l;
1291         size -= l;
1292     }
1293
1294 exit:
1295     fclose(f);
1296 }
1297
1298 void qmp_pmemsave(int64_t addr, int64_t size, const char *filename,
1299                   Error **errp)
1300 {
1301     FILE *f;
1302     uint32_t l;
1303     uint8_t buf[1024];
1304
1305     f = fopen(filename, "wb");
1306     if (!f) {
1307         error_setg_file_open(errp, errno, filename);
1308         return;
1309     }
1310
1311     while (size != 0) {
1312         l = sizeof(buf);
1313         if (l > size)
1314             l = size;
1315         cpu_physical_memory_rw(addr, buf, l, 0);
1316         if (fwrite(buf, 1, l, f) != l) {
1317             error_set(errp, QERR_IO_ERROR);
1318             goto exit;
1319         }
1320         addr += l;
1321         size -= l;
1322     }
1323
1324 exit:
1325     fclose(f);
1326 }
1327
1328 void qmp_inject_nmi(Error **errp)
1329 {
1330 #if defined(TARGET_I386)
1331     CPUArchState *env;
1332
1333     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1334         if (!env->apic_state) {
1335             cpu_interrupt(CPU(x86_env_get_cpu(env)), CPU_INTERRUPT_NMI);
1336         } else {
1337             apic_deliver_nmi(env->apic_state);
1338         }
1339     }
1340 #else
1341     error_set(errp, QERR_UNSUPPORTED);
1342 #endif
1343 }