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 (!env->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
 394     va_start(ap, fmt);
 395     fprintf(stderr, "qemu: hardware error: ");
 396     vfprintf(stderr, fmt, ap);
 397     fprintf(stderr, "\n");
 398     for(env = first_cpu; env != NULL; env = env->next_cpu) {
 399         fprintf(stderr, "CPU #%d:\n", env->cpu_index);
 400         cpu_dump_state(env, stderr, fprintf, CPU_DUMP_FPU);
 401     }
 402     va_end(ap);
 403     abort();
 404 }
 405
 406 void cpu_synchronize_all_states(void)
 407 {
 408     CPUArchState *cpu;
 409
 410     for (cpu = first_cpu; cpu; cpu = cpu->next_cpu) {
 411         cpu_synchronize_state(cpu);
 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(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(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(env);
 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(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(CPUArchState *env)
 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(env, 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(CPUArchState *env)
 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(CPUArchState *env)
 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(env, &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(CPUArchState *env)
 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(CPUArchState *env)
 721 {
 722     CPUState *cpu = ENV_GET_CPU(env);
 723
 724     while (cpu_thread_is_idle(env)) {
 725         qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
 726     }
 727
 728     qemu_kvm_eat_signals(env);
 729     qemu_wait_io_event_common(cpu);
 730 }
 731
 732 static void *qemu_kvm_cpu_thread_fn(void *arg)
 733 {
 734     CPUArchState *env = arg;
 735     CPUState *cpu = ENV_GET_CPU(env);
 736     int r;
 737
 738     qemu_mutex_lock(&qemu_global_mutex);
 739     qemu_thread_get_self(cpu->thread);
 740     cpu->thread_id = qemu_get_thread_id();
 741     cpu_single_env = env;
 742
 743     r = kvm_init_vcpu(env);
 744     if (r < 0) {
 745         fprintf(stderr, "kvm_init_vcpu failed: %s\n", strerror(-r));
 746         exit(1);
 747     }
 748
 749     qemu_kvm_init_cpu_signals(env);
 750
 751     /* signal CPU creation */
 752     cpu->created = true;
 753     qemu_cond_signal(&qemu_cpu_cond);
 754
 755     while (1) {
 756         if (cpu_can_run(cpu)) {
 757             r = kvm_cpu_exec(env);
 758             if (r == EXCP_DEBUG) {
 759                 cpu_handle_guest_debug(env);
 760             }
 761         }
 762         qemu_kvm_wait_io_event(env);
 763     }
 764
 765     return NULL;
 766 }
 767
 768 static void *qemu_dummy_cpu_thread_fn(void *arg)
 769 {
 770 #ifdef _WIN32
 771     fprintf(stderr, "qtest is not supported under Windows\n");
 772     exit(1);
 773 #else
 774     CPUArchState *env = arg;
 775     CPUState *cpu = ENV_GET_CPU(env);
 776     sigset_t waitset;
 777     int r;
 778
 779     qemu_mutex_lock_iothread();
 780     qemu_thread_get_self(cpu->thread);
 781     cpu->thread_id = qemu_get_thread_id();
 782
 783     sigemptyset(&waitset);
 784     sigaddset(&waitset, SIG_IPI);
 785
 786     /* signal CPU creation */
 787     cpu->created = true;
 788     qemu_cond_signal(&qemu_cpu_cond);
 789
 790     cpu_single_env = env;
 791     while (1) {
 792         cpu_single_env = NULL;
 793         qemu_mutex_unlock_iothread();
 794         do {
 795             int sig;
 796             r = sigwait(&waitset, &sig);
 797         } while (r == -1 && (errno == EAGAIN || errno == EINTR));
 798         if (r == -1) {
 799             perror("sigwait");
 800             exit(1);
 801         }
 802         qemu_mutex_lock_iothread();
 803         cpu_single_env = env;
 804         qemu_wait_io_event_common(cpu);
 805     }
 806
 807     return NULL;
 808 #endif
 809 }
 810
 811 static void tcg_exec_all(void);
 812
 813 static void *qemu_tcg_cpu_thread_fn(void *arg)
 814 {
 815     CPUState *cpu = arg;
 816     CPUArchState *env;
 817
 818     qemu_tcg_init_cpu_signals();
 819     qemu_thread_get_self(cpu->thread);
 820
 821     /* signal CPU creation */
 822     qemu_mutex_lock(&qemu_global_mutex);
 823     for (env = first_cpu; env != NULL; env = env->next_cpu) {
 824         cpu = ENV_GET_CPU(env);
 825         cpu->thread_id = qemu_get_thread_id();
 826         cpu->created = true;
 827     }
 828     qemu_cond_signal(&qemu_cpu_cond);
 829
 830     /* wait for initial kick-off after machine start */
 831     while (ENV_GET_CPU(first_cpu)->stopped) {
 832         qemu_cond_wait(tcg_halt_cond, &qemu_global_mutex);
 833
 834         /* process any pending work */
 835         for (env = first_cpu; env != NULL; env = env->next_cpu) {
 836             qemu_wait_io_event_common(ENV_GET_CPU(env));
 837         }
 838     }
 839
 840     while (1) {
 841         tcg_exec_all();
 842         if (use_icount && qemu_clock_deadline(vm_clock) <= 0) {
 843             qemu_notify_event();
 844         }
 845         qemu_tcg_wait_io_event();
 846     }
 847
 848     return NULL;
 849 }
 850
 851 static void qemu_cpu_kick_thread(CPUState *cpu)
 852 {
 853 #ifndef _WIN32
 854     int err;
 855
 856     err = pthread_kill(cpu->thread->thread, SIG_IPI);
 857     if (err) {
 858         fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
 859         exit(1);
 860     }
 861 #else /* _WIN32 */
 862     if (!qemu_cpu_is_self(cpu)) {
 863         SuspendThread(cpu->hThread);
 864         cpu_signal(0);
 865         ResumeThread(cpu->hThread);
 866     }
 867 #endif
 868 }
 869
 870 void qemu_cpu_kick(CPUState *cpu)
 871 {
 872     qemu_cond_broadcast(cpu->halt_cond);
 873     if (!tcg_enabled() && !cpu->thread_kicked) {
 874         qemu_cpu_kick_thread(cpu);
 875         cpu->thread_kicked = true;
 876     }
 877 }
 878
 879 void qemu_cpu_kick_self(void)
 880 {
 881 #ifndef _WIN32
 882     assert(cpu_single_env);
 883     CPUState *cpu_single_cpu = ENV_GET_CPU(cpu_single_env);
 884
 885     if (!cpu_single_cpu->thread_kicked) {
 886         qemu_cpu_kick_thread(cpu_single_cpu);
 887         cpu_single_cpu->thread_kicked = true;
 888     }
 889 #else
 890     abort();
 891 #endif
 892 }
 893
 894 bool qemu_cpu_is_self(CPUState *cpu)
 895 {
 896     return qemu_thread_is_self(cpu->thread);
 897 }
 898
 899 static bool qemu_in_vcpu_thread(void)
 900 {
 901     return cpu_single_env && qemu_cpu_is_self(ENV_GET_CPU(cpu_single_env));
 902 }
 903
 904 void qemu_mutex_lock_iothread(void)
 905 {
 906     if (!tcg_enabled()) {
 907         qemu_mutex_lock(&qemu_global_mutex);
 908     } else {
 909         iothread_requesting_mutex = true;
 910         if (qemu_mutex_trylock(&qemu_global_mutex)) {
 911             qemu_cpu_kick_thread(ENV_GET_CPU(first_cpu));
 912             qemu_mutex_lock(&qemu_global_mutex);
 913         }
 914         iothread_requesting_mutex = false;
 915         qemu_cond_broadcast(&qemu_io_proceeded_cond);
 916     }
 917 }
 918
 919 void qemu_mutex_unlock_iothread(void)
 920 {
 921     qemu_mutex_unlock(&qemu_global_mutex);
 922 }
 923
 924 static int all_vcpus_paused(void)
 925 {
 926     CPUArchState *penv = first_cpu;
 927
 928     while (penv) {
 929         CPUState *pcpu = ENV_GET_CPU(penv);
 930         if (!pcpu->stopped) {
 931             return 0;
 932         }
 933         penv = penv->next_cpu;
 934     }
 935
 936     return 1;
 937 }
 938
 939 void pause_all_vcpus(void)
 940 {
 941     CPUArchState *penv = first_cpu;
 942
 943     qemu_clock_enable(vm_clock, false);
 944     while (penv) {
 945         CPUState *pcpu = ENV_GET_CPU(penv);
 946         pcpu->stop = true;
 947         qemu_cpu_kick(pcpu);
 948         penv = penv->next_cpu;
 949     }
 950
 951     if (qemu_in_vcpu_thread()) {
 952         cpu_stop_current();
 953         if (!kvm_enabled()) {
 954             while (penv) {
 955                 CPUState *pcpu = ENV_GET_CPU(penv);
 956                 pcpu->stop = 0;
 957                 pcpu->stopped = true;
 958                 penv = penv->next_cpu;
 959             }
 960             return;
 961         }
 962     }
 963
 964     while (!all_vcpus_paused()) {
 965         qemu_cond_wait(&qemu_pause_cond, &qemu_global_mutex);
 966         penv = first_cpu;
 967         while (penv) {
 968             qemu_cpu_kick(ENV_GET_CPU(penv));
 969             penv = penv->next_cpu;
 970         }
 971     }
 972 }
 973
 974 void resume_all_vcpus(void)
 975 {
 976     CPUArchState *penv = first_cpu;
 977
 978     qemu_clock_enable(vm_clock, true);
 979     while (penv) {
 980         CPUState *pcpu = ENV_GET_CPU(penv);
 981         pcpu->stop = false;
 982         pcpu->stopped = false;
 983         qemu_cpu_kick(pcpu);
 984         penv = penv->next_cpu;
 985     }
 986 }
 987
 988 static void qemu_tcg_init_vcpu(CPUState *cpu)
 989 {
 990     /* share a single thread for all cpus with TCG */
 991     if (!tcg_cpu_thread) {
 992         cpu->thread = g_malloc0(sizeof(QemuThread));
 993         cpu->halt_cond = g_malloc0(sizeof(QemuCond));
 994         qemu_cond_init(cpu->halt_cond);
 995         tcg_halt_cond = cpu->halt_cond;
 996         qemu_thread_create(cpu->thread, qemu_tcg_cpu_thread_fn, cpu,
 997                            QEMU_THREAD_JOINABLE);
 998 #ifdef _WIN32
 999         cpu->hThread = qemu_thread_get_handle(cpu->thread);
1000 #endif
1001         while (!cpu->created) {
1002             qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1003         }
1004         tcg_cpu_thread = cpu->thread;
1005     } else {
1006         cpu->thread = tcg_cpu_thread;
1007         cpu->halt_cond = tcg_halt_cond;
1008     }
1009 }
1010
1011 static void qemu_kvm_start_vcpu(CPUArchState *env)
1012 {
1013     CPUState *cpu = ENV_GET_CPU(env);
1014
1015     cpu->thread = g_malloc0(sizeof(QemuThread));
1016     cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1017     qemu_cond_init(cpu->halt_cond);
1018     qemu_thread_create(cpu->thread, qemu_kvm_cpu_thread_fn, env,
1019                        QEMU_THREAD_JOINABLE);
1020     while (!cpu->created) {
1021         qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1022     }
1023 }
1024
1025 static void qemu_dummy_start_vcpu(CPUArchState *env)
1026 {
1027     CPUState *cpu = ENV_GET_CPU(env);
1028
1029     cpu->thread = g_malloc0(sizeof(QemuThread));
1030     cpu->halt_cond = g_malloc0(sizeof(QemuCond));
1031     qemu_cond_init(cpu->halt_cond);
1032     qemu_thread_create(cpu->thread, qemu_dummy_cpu_thread_fn, env,
1033                        QEMU_THREAD_JOINABLE);
1034     while (!cpu->created) {
1035         qemu_cond_wait(&qemu_cpu_cond, &qemu_global_mutex);
1036     }
1037 }
1038
1039 void qemu_init_vcpu(void *_env)
1040 {
1041     CPUArchState *env = _env;
1042     CPUState *cpu = ENV_GET_CPU(env);
1043
1044     env->nr_cores = smp_cores;
1045     env->nr_threads = smp_threads;
1046     cpu->stopped = true;
1047     if (kvm_enabled()) {
1048         qemu_kvm_start_vcpu(env);
1049     } else if (tcg_enabled()) {
1050         qemu_tcg_init_vcpu(cpu);
1051     } else {
1052         qemu_dummy_start_vcpu(env);
1053     }
1054 }
1055
1056 void cpu_stop_current(void)
1057 {
1058     if (cpu_single_env) {
1059         CPUState *cpu_single_cpu = ENV_GET_CPU(cpu_single_env);
1060         cpu_single_cpu->stop = false;
1061         cpu_single_cpu->stopped = true;
1062         cpu_exit(cpu_single_env);
1063         qemu_cond_signal(&qemu_pause_cond);
1064     }
1065 }
1066
1067 void vm_stop(RunState state)
1068 {
1069     if (qemu_in_vcpu_thread()) {
1070         qemu_system_vmstop_request(state);
1071         /*
1072          * FIXME: should not return to device code in case
1073          * vm_stop() has been requested.
1074          */
1075         cpu_stop_current();
1076         return;
1077     }
1078     do_vm_stop(state);
1079 }
1080
1081 /* does a state transition even if the VM is already stopped,
1082    current state is forgotten forever */
1083 void vm_stop_force_state(RunState state)
1084 {
1085     if (runstate_is_running()) {
1086         vm_stop(state);
1087     } else {
1088         runstate_set(state);
1089     }
1090 }
1091
1092 static int tcg_cpu_exec(CPUArchState *env)
1093 {
1094     int ret;
1095 #ifdef CONFIG_PROFILER
1096     int64_t ti;
1097 #endif
1098
1099 #ifdef CONFIG_PROFILER
1100     ti = profile_getclock();
1101 #endif
1102     if (use_icount) {
1103         int64_t count;
1104         int decr;
1105         qemu_icount -= (env->icount_decr.u16.low + env->icount_extra);
1106         env->icount_decr.u16.low = 0;
1107         env->icount_extra = 0;
1108         count = qemu_icount_round(qemu_clock_deadline(vm_clock));
1109         qemu_icount += count;
1110         decr = (count > 0xffff) ? 0xffff : count;
1111         count -= decr;
1112         env->icount_decr.u16.low = decr;
1113         env->icount_extra = count;
1114     }
1115     ret = cpu_exec(env);
1116 #ifdef CONFIG_PROFILER
1117     qemu_time += profile_getclock() - ti;
1118 #endif
1119     if (use_icount) {
1120         /* Fold pending instructions back into the
1121            instruction counter, and clear the interrupt flag.  */
1122         qemu_icount -= (env->icount_decr.u16.low
1123                         + env->icount_extra);
1124         env->icount_decr.u32 = 0;
1125         env->icount_extra = 0;
1126     }
1127     return ret;
1128 }
1129
1130 static void tcg_exec_all(void)
1131 {
1132     int r;
1133
1134     /* Account partial waits to the vm_clock.  */
1135     qemu_clock_warp(vm_clock);
1136
1137     if (next_cpu == NULL) {
1138         next_cpu = first_cpu;
1139     }
1140     for (; next_cpu != NULL && !exit_request; next_cpu = next_cpu->next_cpu) {
1141         CPUArchState *env = next_cpu;
1142         CPUState *cpu = ENV_GET_CPU(env);
1143
1144         qemu_clock_enable(vm_clock,
1145                           (env->singlestep_enabled & SSTEP_NOTIMER) == 0);
1146
1147         if (cpu_can_run(cpu)) {
1148             r = tcg_cpu_exec(env);
1149             if (r == EXCP_DEBUG) {
1150                 cpu_handle_guest_debug(env);
1151                 break;
1152             }
1153         } else if (cpu->stop || cpu->stopped) {
1154             break;
1155         }
1156     }
1157     exit_request = 0;
1158 }
1159
1160 void set_numa_modes(void)
1161 {
1162     CPUArchState *env;
1163     int i;
1164
1165     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1166         for (i = 0; i < nb_numa_nodes; i++) {
1167             if (test_bit(env->cpu_index, node_cpumask[i])) {
1168                 env->numa_node = i;
1169             }
1170         }
1171     }
1172 }
1173
1174 void set_cpu_log(const char *optarg)
1175 {
1176     int mask;
1177     const CPULogItem *item;
1178
1179     mask = cpu_str_to_log_mask(optarg);
1180     if (!mask) {
1181         printf("Log items (comma separated):\n");
1182         for (item = cpu_log_items; item->mask != 0; item++) {
1183             printf("%-10s %s\n", item->name, item->help);
1184         }
1185         exit(1);
1186     }
1187     cpu_set_log(mask);
1188 }
1189
1190 void set_cpu_log_filename(const char *optarg)
1191 {
1192     cpu_set_log_filename(optarg);
1193 }
1194
1195 void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
1196 {
1197     /* XXX: implement xxx_cpu_list for targets that still miss it */
1198 #if defined(cpu_list)
1199     cpu_list(f, cpu_fprintf);
1200 #endif
1201 }
1202
1203 CpuInfoList *qmp_query_cpus(Error **errp)
1204 {
1205     CpuInfoList *head = NULL, *cur_item = NULL;
1206     CPUArchState *env;
1207
1208     for (env = first_cpu; env != NULL; env = env->next_cpu) {
1209         CPUState *cpu = ENV_GET_CPU(env);
1210         CpuInfoList *info;
1211
1212         cpu_synchronize_state(env);
1213
1214         info = g_malloc0(sizeof(*info));
1215         info->value = g_malloc0(sizeof(*info->value));
1216         info->value->CPU = env->cpu_index;
1217         info->value->current = (env == first_cpu);
1218         info->value->halted = env->halted;
1219         info->value->thread_id = cpu->thread_id;
1220 #if defined(TARGET_I386)
1221         info->value->has_pc = true;
1222         info->value->pc = env->eip + env->segs[R_CS].base;
1223 #elif defined(TARGET_PPC)
1224         info->value->has_nip = true;
1225         info->value->nip = env->nip;
1226 #elif defined(TARGET_SPARC)
1227         info->value->has_pc = true;
1228         info->value->pc = env->pc;
1229         info->value->has_npc = true;
1230         info->value->npc = env->npc;
1231 #elif defined(TARGET_MIPS)
1232         info->value->has_PC = true;
1233         info->value->PC = env->active_tc.PC;
1234 #endif
1235
1236         /* XXX: waiting for the qapi to support GSList */
1237         if (!cur_item) {
1238             head = cur_item = info;
1239         } else {
1240             cur_item->next = info;
1241             cur_item = info;
1242         }
1243     }
1244
1245     return head;
1246 }
1247
1248 void qmp_memsave(int64_t addr, int64_t size, const char *filename,
1249                  bool has_cpu, int64_t cpu_index, Error **errp)
1250 {
1251     FILE *f;
1252     uint32_t l;
1253     CPUArchState *env;
1254     uint8_t buf[1024];
1255
1256     if (!has_cpu) {
1257         cpu_index = 0;
1258     }
1259
1260     for (env = first_cpu; env; env = env->next_cpu) {
1261         if (cpu_index == env->cpu_index) {
1262             break;
1263         }
1264     }
1265
1266     if (env == NULL) {
1267         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cpu-index",
1268                   "a CPU number");
1269         return;
1270     }
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(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 }