qemu-timer.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 #include "sysemu.h"
  26 #include "net.h"
  27 #include "monitor.h"
  28 #include "console.h"
  29
  30 #include "hw/hw.h"
  31
  32 #include <unistd.h>
  33 #include <fcntl.h>
  34 #include <time.h>
  35 #include <errno.h>
  36 #include <sys/time.h>
  37 #include <signal.h>
  38 #ifdef __FreeBSD__
  39 #include <sys/param.h>
  40 #endif
  41
  42 #ifdef __linux__
  43 #include <sys/ioctl.h>
  44 #include <linux/rtc.h>
  45 /* For the benefit of older linux systems which don't supply it,
  46    we use a local copy of hpet.h. */
  47 /* #include <linux/hpet.h> */
  48 #include "hpet.h"
  49 #endif
  50
  51 #ifdef _WIN32
  52 #include <windows.h>
  53 #include <mmsystem.h>
  54 #endif
  55
  56 #include "qemu-timer.h"
  57
  58 /* Conversion factor from emulated instructions to virtual clock ticks.  */
  59 int icount_time_shift;
  60 /* Arbitrarily pick 1MIPS as the minimum allowable speed.  */
  61 #define MAX_ICOUNT_SHIFT 10
  62 /* Compensate for varying guest execution speed.  */
  63 int64_t qemu_icount_bias;
  64 static QEMUTimer *icount_rt_timer;
  65 static QEMUTimer *icount_vm_timer;
  66
  67 /***********************************************************/
  68 /* guest cycle counter */
  69
  70 typedef struct TimersState {
  71     int64_t cpu_ticks_prev;
  72     int64_t cpu_ticks_offset;
  73     int64_t cpu_clock_offset;
  74     int32_t cpu_ticks_enabled;
  75     int64_t dummy;
  76 } TimersState;
  77
  78 TimersState timers_state;
  79
  80 /* return the host CPU cycle counter and handle stop/restart */
  81 int64_t cpu_get_ticks(void)
  82 {
  83     if (use_icount) {
  84         return cpu_get_icount();
  85     }
  86     if (!timers_state.cpu_ticks_enabled) {
  87         return timers_state.cpu_ticks_offset;
  88     } else {
  89         int64_t ticks;
  90         ticks = cpu_get_real_ticks();
  91         if (timers_state.cpu_ticks_prev > ticks) {
  92             /* Note: non increasing ticks may happen if the host uses
  93                software suspend */
  94             timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks;
  95         }
  96         timers_state.cpu_ticks_prev = ticks;
  97         return ticks + timers_state.cpu_ticks_offset;
  98     }
  99 }
 100
 101 /* return the host CPU monotonic timer and handle stop/restart */
 102 static int64_t cpu_get_clock(void)
 103 {
 104     int64_t ti;
 105     if (!timers_state.cpu_ticks_enabled) {
 106         return timers_state.cpu_clock_offset;
 107     } else {
 108         ti = get_clock();
 109         return ti + timers_state.cpu_clock_offset;
 110     }
 111 }
 112
 113 static int64_t qemu_icount_delta(void)
 114 {
 115     if (use_icount == 1) {
 116         /* When not using an adaptive execution frequency
 117            we tend to get badly out of sync with real time,
 118            so just delay for a reasonable amount of time.  */
 119         return 0;
 120     } else {
 121         return cpu_get_icount() - cpu_get_clock();
 122     }
 123 }
 124
 125 /* enable cpu_get_ticks() */
 126 void cpu_enable_ticks(void)
 127 {
 128     if (!timers_state.cpu_ticks_enabled) {
 129         timers_state.cpu_ticks_offset -= cpu_get_real_ticks();
 130         timers_state.cpu_clock_offset -= get_clock();
 131         timers_state.cpu_ticks_enabled = 1;
 132     }
 133 }
 134
 135 /* disable cpu_get_ticks() : the clock is stopped. You must not call
 136    cpu_get_ticks() after that.  */
 137 void cpu_disable_ticks(void)
 138 {
 139     if (timers_state.cpu_ticks_enabled) {
 140         timers_state.cpu_ticks_offset = cpu_get_ticks();
 141         timers_state.cpu_clock_offset = cpu_get_clock();
 142         timers_state.cpu_ticks_enabled = 0;
 143     }
 144 }
 145
 146 /***********************************************************/
 147 /* timers */
 148
 149 #define QEMU_CLOCK_REALTIME 0
 150 #define QEMU_CLOCK_VIRTUAL  1
 151 #define QEMU_CLOCK_HOST     2
 152
 153 struct QEMUClock {
 154     int type;
 155     int enabled;
 156     /* XXX: add frequency */
 157 };
 158
 159 struct QEMUTimer {
 160     QEMUClock *clock;
 161     int64_t expire_time;
 162     QEMUTimerCB *cb;
 163     void *opaque;
 164     struct QEMUTimer *next;
 165 };
 166
 167 struct qemu_alarm_timer {
 168     char const *name;
 169     int (*start)(struct qemu_alarm_timer *t);
 170     void (*stop)(struct qemu_alarm_timer *t);
 171     void (*rearm)(struct qemu_alarm_timer *t);
 172     void *priv;
 173
 174     char expired;
 175     char pending;
 176 };
 177
 178 static struct qemu_alarm_timer *alarm_timer;
 179
 180 int qemu_alarm_pending(void)
 181 {
 182     return alarm_timer->pending;
 183 }
 184
 185 static inline int alarm_has_dynticks(struct qemu_alarm_timer *t)
 186 {
 187     return !!t->rearm;
 188 }
 189
 190 static void qemu_rearm_alarm_timer(struct qemu_alarm_timer *t)
 191 {
 192     if (!alarm_has_dynticks(t))
 193         return;
 194
 195     t->rearm(t);
 196 }
 197
 198 /* TODO: MIN_TIMER_REARM_NS should be optimized */
 199 #define MIN_TIMER_REARM_NS 250000
 200
 201 #ifdef _WIN32
 202
 203 static int win32_start_timer(struct qemu_alarm_timer *t);
 204 static void win32_stop_timer(struct qemu_alarm_timer *t);
 205 static void win32_rearm_timer(struct qemu_alarm_timer *t);
 206
 207 #else
 208
 209 static int unix_start_timer(struct qemu_alarm_timer *t);
 210 static void unix_stop_timer(struct qemu_alarm_timer *t);
 211
 212 #ifdef __linux__
 213
 214 static int dynticks_start_timer(struct qemu_alarm_timer *t);
 215 static void dynticks_stop_timer(struct qemu_alarm_timer *t);
 216 static void dynticks_rearm_timer(struct qemu_alarm_timer *t);
 217
 218 static int hpet_start_timer(struct qemu_alarm_timer *t);
 219 static void hpet_stop_timer(struct qemu_alarm_timer *t);
 220
 221 static int rtc_start_timer(struct qemu_alarm_timer *t);
 222 static void rtc_stop_timer(struct qemu_alarm_timer *t);
 223
 224 #endif /* __linux__ */
 225
 226 #endif /* _WIN32 */
 227
 228 /* Correlation between real and virtual time is always going to be
 229    fairly approximate, so ignore small variation.
 230    When the guest is idle real and virtual time will be aligned in
 231    the IO wait loop.  */
 232 #define ICOUNT_WOBBLE (get_ticks_per_sec() / 10)
 233
 234 static void icount_adjust(void)
 235 {
 236     int64_t cur_time;
 237     int64_t cur_icount;
 238     int64_t delta;
 239     static int64_t last_delta;
 240     /* If the VM is not running, then do nothing.  */
 241     if (!vm_running)
 242         return;
 243
 244     cur_time = cpu_get_clock();
 245     cur_icount = qemu_get_clock(vm_clock);
 246     delta = cur_icount - cur_time;
 247     /* FIXME: This is a very crude algorithm, somewhat prone to oscillation.  */
 248     if (delta > 0
 249         && last_delta + ICOUNT_WOBBLE < delta * 2
 250         && icount_time_shift > 0) {
 251         /* The guest is getting too far ahead.  Slow time down.  */
 252         icount_time_shift--;
 253     }
 254     if (delta < 0
 255         && last_delta - ICOUNT_WOBBLE > delta * 2
 256         && icount_time_shift < MAX_ICOUNT_SHIFT) {
 257         /* The guest is getting too far behind.  Speed time up.  */
 258         icount_time_shift++;
 259     }
 260     last_delta = delta;
 261     qemu_icount_bias = cur_icount - (qemu_icount << icount_time_shift);
 262 }
 263
 264 static void icount_adjust_rt(void * opaque)
 265 {
 266     qemu_mod_timer(icount_rt_timer,
 267                    qemu_get_clock(rt_clock) + 1000);
 268     icount_adjust();
 269 }
 270
 271 static void icount_adjust_vm(void * opaque)
 272 {
 273     qemu_mod_timer(icount_vm_timer,
 274                    qemu_get_clock(vm_clock) + get_ticks_per_sec() / 10);
 275     icount_adjust();
 276 }
 277
 278 int64_t qemu_icount_round(int64_t count)
 279 {
 280     return (count + (1 << icount_time_shift) - 1) >> icount_time_shift;
 281 }
 282
 283 static struct qemu_alarm_timer alarm_timers[] = {
 284 #ifndef _WIN32
 285 #ifdef __linux__
 286     {"dynticks", dynticks_start_timer,
 287      dynticks_stop_timer, dynticks_rearm_timer, NULL},
 288     /* HPET - if available - is preferred */
 289     {"hpet", hpet_start_timer, hpet_stop_timer, NULL, NULL},
 290     /* ...otherwise try RTC */
 291     {"rtc", rtc_start_timer, rtc_stop_timer, NULL, NULL},
 292 #endif
 293     {"unix", unix_start_timer, unix_stop_timer, NULL, NULL},
 294 #else
 295     {"dynticks", win32_start_timer,
 296      win32_stop_timer, win32_rearm_timer, NULL},
 297     {"win32", win32_start_timer,
 298      win32_stop_timer, NULL, NULL},
 299 #endif
 300     {NULL, }
 301 };
 302
 303 static void show_available_alarms(void)
 304 {
 305     int i;
 306
 307     printf("Available alarm timers, in order of precedence:\n");
 308     for (i = 0; alarm_timers[i].name; i++)
 309         printf("%s\n", alarm_timers[i].name);
 310 }
 311
 312 void configure_alarms(char const *opt)
 313 {
 314     int i;
 315     int cur = 0;
 316     int count = ARRAY_SIZE(alarm_timers) - 1;
 317     char *arg;
 318     char *name;
 319     struct qemu_alarm_timer tmp;
 320
 321     if (!strcmp(opt, "?")) {
 322         show_available_alarms();
 323         exit(0);
 324     }
 325
 326     arg = qemu_strdup(opt);
 327
 328     /* Reorder the array */
 329     name = strtok(arg, ",");
 330     while (name) {
 331         for (i = 0; i < count && alarm_timers[i].name; i++) {
 332             if (!strcmp(alarm_timers[i].name, name))
 333                 break;
 334         }
 335
 336         if (i == count) {
 337             fprintf(stderr, "Unknown clock %s\n", name);
 338             goto next;
 339         }
 340
 341         if (i < cur)
 342             /* Ignore */
 343             goto next;
 344
 345         /* Swap */
 346         tmp = alarm_timers[i];
 347         alarm_timers[i] = alarm_timers[cur];
 348         alarm_timers[cur] = tmp;
 349
 350         cur++;
 351 next:
 352         name = strtok(NULL, ",");
 353     }
 354
 355     qemu_free(arg);
 356
 357     if (cur) {
 358         /* Disable remaining timers */
 359         for (i = cur; i < count; i++)
 360             alarm_timers[i].name = NULL;
 361     } else {
 362         show_available_alarms();
 363         exit(1);
 364     }
 365 }
 366
 367 #define QEMU_NUM_CLOCKS 3
 368
 369 QEMUClock *rt_clock;
 370 QEMUClock *vm_clock;
 371 QEMUClock *host_clock;
 372
 373 static QEMUTimer *active_timers[QEMU_NUM_CLOCKS];
 374
 375 static QEMUClock *qemu_new_clock(int type)
 376 {
 377     QEMUClock *clock;
 378     clock = qemu_mallocz(sizeof(QEMUClock));
 379     clock->type = type;
 380     clock->enabled = 1;
 381     return clock;
 382 }
 383
 384 void qemu_clock_enable(QEMUClock *clock, int enabled)
 385 {
 386     clock->enabled = enabled;
 387 }
 388
 389 QEMUTimer *qemu_new_timer(QEMUClock *clock, QEMUTimerCB *cb, void *opaque)
 390 {
 391     QEMUTimer *ts;
 392
 393     ts = qemu_mallocz(sizeof(QEMUTimer));
 394     ts->clock = clock;
 395     ts->cb = cb;
 396     ts->opaque = opaque;
 397     return ts;
 398 }
 399
 400 void qemu_free_timer(QEMUTimer *ts)
 401 {
 402     qemu_free(ts);
 403 }
 404
 405 /* stop a timer, but do not dealloc it */
 406 void qemu_del_timer(QEMUTimer *ts)
 407 {
 408     QEMUTimer **pt, *t;
 409
 410     /* NOTE: this code must be signal safe because
 411        qemu_timer_expired() can be called from a signal. */
 412     pt = &active_timers[ts->clock->type];
 413     for(;;) {
 414         t = *pt;
 415         if (!t)
 416             break;
 417         if (t == ts) {
 418             *pt = t->next;
 419             break;
 420         }
 421         pt = &t->next;
 422     }
 423 }
 424
 425 /* modify the current timer so that it will be fired when current_time
 426    >= expire_time. The corresponding callback will be called. */
 427 void qemu_mod_timer(QEMUTimer *ts, int64_t expire_time)
 428 {
 429     QEMUTimer **pt, *t;
 430
 431     qemu_del_timer(ts);
 432
 433     /* add the timer in the sorted list */
 434     /* NOTE: this code must be signal safe because
 435        qemu_timer_expired() can be called from a signal. */
 436     pt = &active_timers[ts->clock->type];
 437     for(;;) {
 438         t = *pt;
 439         if (!t)
 440             break;
 441         if (t->expire_time > expire_time)
 442             break;
 443         pt = &t->next;
 444     }
 445     ts->expire_time = expire_time;
 446     ts->next = *pt;
 447     *pt = ts;
 448
 449     /* Rearm if necessary  */
 450     if (pt == &active_timers[ts->clock->type]) {
 451         if (!alarm_timer->pending) {
 452             qemu_rearm_alarm_timer(alarm_timer);
 453         }
 454         /* Interrupt execution to force deadline recalculation.  */
 455         if (use_icount)
 456             qemu_notify_event();
 457     }
 458 }
 459
 460 int qemu_timer_pending(QEMUTimer *ts)
 461 {
 462     QEMUTimer *t;
 463     for(t = active_timers[ts->clock->type]; t != NULL; t = t->next) {
 464         if (t == ts)
 465             return 1;
 466     }
 467     return 0;
 468 }
 469
 470 int qemu_timer_expired(QEMUTimer *timer_head, int64_t current_time)
 471 {
 472     if (!timer_head)
 473         return 0;
 474     return (timer_head->expire_time <= current_time);
 475 }
 476
 477 static void qemu_run_timers(QEMUClock *clock)
 478 {
 479     QEMUTimer **ptimer_head, *ts;
 480     int64_t current_time;
 481
 482     if (!clock->enabled)
 483         return;
 484
 485     current_time = qemu_get_clock (clock);
 486     ptimer_head = &active_timers[clock->type];
 487     for(;;) {
 488         ts = *ptimer_head;
 489         if (!ts || ts->expire_time > current_time)
 490             break;
 491         /* remove timer from the list before calling the callback */
 492         *ptimer_head = ts->next;
 493         ts->next = NULL;
 494
 495         /* run the callback (the timer list can be modified) */
 496         ts->cb(ts->opaque);
 497     }
 498 }
 499
 500 int64_t qemu_get_clock(QEMUClock *clock)
 501 {
 502     switch(clock->type) {
 503     case QEMU_CLOCK_REALTIME:
 504         return get_clock() / 1000000;
 505     default:
 506     case QEMU_CLOCK_VIRTUAL:
 507         if (use_icount) {
 508             return cpu_get_icount();
 509         } else {
 510             return cpu_get_clock();
 511         }
 512     case QEMU_CLOCK_HOST:
 513         return get_clock_realtime();
 514     }
 515 }
 516
 517 int64_t qemu_get_clock_ns(QEMUClock *clock)
 518 {
 519     switch(clock->type) {
 520     case QEMU_CLOCK_REALTIME:
 521         return get_clock();
 522     default:
 523     case QEMU_CLOCK_VIRTUAL:
 524         if (use_icount) {
 525             return cpu_get_icount();
 526         } else {
 527             return cpu_get_clock();
 528         }
 529     case QEMU_CLOCK_HOST:
 530         return get_clock_realtime();
 531     }
 532 }
 533
 534 void init_clocks(void)
 535 {
 536     rt_clock = qemu_new_clock(QEMU_CLOCK_REALTIME);
 537     vm_clock = qemu_new_clock(QEMU_CLOCK_VIRTUAL);
 538     host_clock = qemu_new_clock(QEMU_CLOCK_HOST);
 539
 540     rtc_clock = host_clock;
 541 }
 542
 543 /* save a timer */
 544 void qemu_put_timer(QEMUFile *f, QEMUTimer *ts)
 545 {
 546     uint64_t expire_time;
 547
 548     if (qemu_timer_pending(ts)) {
 549         expire_time = ts->expire_time;
 550     } else {
 551         expire_time = -1;
 552     }
 553     qemu_put_be64(f, expire_time);
 554 }
 555
 556 void qemu_get_timer(QEMUFile *f, QEMUTimer *ts)
 557 {
 558     uint64_t expire_time;
 559
 560     expire_time = qemu_get_be64(f);
 561     if (expire_time != -1) {
 562         qemu_mod_timer(ts, expire_time);
 563     } else {
 564         qemu_del_timer(ts);
 565     }
 566 }
 567
 568 static const VMStateDescription vmstate_timers = {
 569     .name = "timer",
 570     .version_id = 2,
 571     .minimum_version_id = 1,
 572     .minimum_version_id_old = 1,
 573     .fields      = (VMStateField []) {
 574         VMSTATE_INT64(cpu_ticks_offset, TimersState),
 575         VMSTATE_INT64(dummy, TimersState),
 576         VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2),
 577         VMSTATE_END_OF_LIST()
 578     }
 579 };
 580
 581 void configure_icount(const char *option)
 582 {
 583     vmstate_register(NULL, 0, &vmstate_timers, &timers_state);
 584     if (!option)
 585         return;
 586
 587     if (strcmp(option, "auto") != 0) {
 588         icount_time_shift = strtol(option, NULL, 0);
 589         use_icount = 1;
 590         return;
 591     }
 592
 593     use_icount = 2;
 594
 595     /* 125MIPS seems a reasonable initial guess at the guest speed.
 596        It will be corrected fairly quickly anyway.  */
 597     icount_time_shift = 3;
 598
 599     /* Have both realtime and virtual time triggers for speed adjustment.
 600        The realtime trigger catches emulated time passing too slowly,
 601        the virtual time trigger catches emulated time passing too fast.
 602        Realtime triggers occur even when idle, so use them less frequently
 603        than VM triggers.  */
 604     icount_rt_timer = qemu_new_timer(rt_clock, icount_adjust_rt, NULL);
 605     qemu_mod_timer(icount_rt_timer,
 606                    qemu_get_clock(rt_clock) + 1000);
 607     icount_vm_timer = qemu_new_timer(vm_clock, icount_adjust_vm, NULL);
 608     qemu_mod_timer(icount_vm_timer,
 609                    qemu_get_clock(vm_clock) + get_ticks_per_sec() / 10);
 610 }
 611
 612 void qemu_run_all_timers(void)
 613 {
 614     alarm_timer->pending = 0;
 615
 616     /* rearm timer, if not periodic */
 617     if (alarm_timer->expired) {
 618         alarm_timer->expired = 0;
 619         qemu_rearm_alarm_timer(alarm_timer);
 620     }
 621
 622     /* vm time timers */
 623     if (vm_running) {
 624         qemu_run_timers(vm_clock);
 625     }
 626
 627     qemu_run_timers(rt_clock);
 628     qemu_run_timers(host_clock);
 629 }
 630
 631 static int64_t qemu_next_alarm_deadline(void);
 632
 633 #ifdef _WIN32
 634 static void CALLBACK host_alarm_handler(PVOID lpParam, BOOLEAN unused)
 635 #else
 636 static void host_alarm_handler(int host_signum)
 637 #endif
 638 {
 639     struct qemu_alarm_timer *t = alarm_timer;
 640     if (!t)
 641         return;
 642
 643 #if 0
 644 #define DISP_FREQ 1000
 645     {
 646         static int64_t delta_min = INT64_MAX;
 647         static int64_t delta_max, delta_cum, last_clock, delta, ti;
 648         static int count;
 649         ti = qemu_get_clock(vm_clock);
 650         if (last_clock != 0) {
 651             delta = ti - last_clock;
 652             if (delta < delta_min)
 653                 delta_min = delta;
 654             if (delta > delta_max)
 655                 delta_max = delta;
 656             delta_cum += delta;
 657             if (++count == DISP_FREQ) {
 658                 printf("timer: min=%" PRId64 " us max=%" PRId64 " us avg=%" PRId64 " us avg_freq=%0.3f Hz\n",
 659                        muldiv64(delta_min, 1000000, get_ticks_per_sec()),
 660                        muldiv64(delta_max, 1000000, get_ticks_per_sec()),
 661                        muldiv64(delta_cum, 1000000 / DISP_FREQ, get_ticks_per_sec()),
 662                        (double)get_ticks_per_sec() / ((double)delta_cum / DISP_FREQ));
 663                 count = 0;
 664                 delta_min = INT64_MAX;
 665                 delta_max = 0;
 666                 delta_cum = 0;
 667             }
 668         }
 669         last_clock = ti;
 670     }
 671 #endif
 672     if (alarm_has_dynticks(t) ||
 673         qemu_next_alarm_deadline () <= 0) {
 674         t->expired = alarm_has_dynticks(t);
 675         t->pending = 1;
 676         qemu_notify_event();
 677     }
 678 }
 679
 680 int64_t qemu_next_deadline(void)
 681 {
 682     /* To avoid problems with overflow limit this to 2^32.  */
 683     int64_t delta = INT32_MAX;
 684
 685     if (active_timers[QEMU_CLOCK_VIRTUAL]) {
 686         delta = active_timers[QEMU_CLOCK_VIRTUAL]->expire_time -
 687                      qemu_get_clock_ns(vm_clock);
 688     }
 689     if (active_timers[QEMU_CLOCK_HOST]) {
 690         int64_t hdelta = active_timers[QEMU_CLOCK_HOST]->expire_time -
 691                  qemu_get_clock_ns(host_clock);
 692         if (hdelta < delta)
 693             delta = hdelta;
 694     }
 695
 696     if (delta < 0)
 697         delta = 0;
 698
 699     return delta;
 700 }
 701
 702 static int64_t qemu_next_alarm_deadline(void)
 703 {
 704     int64_t delta;
 705     int64_t rtdelta;
 706
 707     if (!use_icount && active_timers[QEMU_CLOCK_VIRTUAL]) {
 708         delta = active_timers[QEMU_CLOCK_VIRTUAL]->expire_time -
 709                      qemu_get_clock(vm_clock);
 710     } else {
 711         delta = INT32_MAX;
 712     }
 713     if (active_timers[QEMU_CLOCK_HOST]) {
 714         int64_t hdelta = active_timers[QEMU_CLOCK_HOST]->expire_time -
 715                  qemu_get_clock_ns(host_clock);
 716         if (hdelta < delta)
 717             delta = hdelta;
 718     }
 719     if (active_timers[QEMU_CLOCK_REALTIME]) {
 720         rtdelta = (active_timers[QEMU_CLOCK_REALTIME]->expire_time * 1000000 -
 721                  qemu_get_clock_ns(rt_clock));
 722         if (rtdelta < delta)
 723             delta = rtdelta;
 724     }
 725
 726     return delta;
 727 }
 728
 729 #if defined(__linux__)
 730
 731 #define RTC_FREQ 1024
 732
 733 static void enable_sigio_timer(int fd)
 734 {
 735     struct sigaction act;
 736
 737     /* timer signal */
 738     sigfillset(&act.sa_mask);
 739     act.sa_flags = 0;
 740     act.sa_handler = host_alarm_handler;
 741
 742     sigaction(SIGIO, &act, NULL);
 743     fcntl_setfl(fd, O_ASYNC);
 744     fcntl(fd, F_SETOWN, getpid());
 745 }
 746
 747 static int hpet_start_timer(struct qemu_alarm_timer *t)
 748 {
 749     struct hpet_info info;
 750     int r, fd;
 751
 752     fd = qemu_open("/dev/hpet", O_RDONLY);
 753     if (fd < 0)
 754         return -1;
 755
 756     /* Set frequency */
 757     r = ioctl(fd, HPET_IRQFREQ, RTC_FREQ);
 758     if (r < 0) {
 759         fprintf(stderr, "Could not configure '/dev/hpet' to have a 1024Hz timer. This is not a fatal\n"
 760                 "error, but for better emulation accuracy type:\n"
 761                 "'echo 1024 > /proc/sys/dev/hpet/max-user-freq' as root.\n");
 762         goto fail;
 763     }
 764
 765     /* Check capabilities */
 766     r = ioctl(fd, HPET_INFO, &info);
 767     if (r < 0)
 768         goto fail;
 769
 770     /* Enable periodic mode */
 771     r = ioctl(fd, HPET_EPI, 0);
 772     if (info.hi_flags && (r < 0))
 773         goto fail;
 774
 775     /* Enable interrupt */
 776     r = ioctl(fd, HPET_IE_ON, 0);
 777     if (r < 0)
 778         goto fail;
 779
 780     enable_sigio_timer(fd);
 781     t->priv = (void *)(long)fd;
 782
 783     return 0;
 784 fail:
 785     close(fd);
 786     return -1;
 787 }
 788
 789 static void hpet_stop_timer(struct qemu_alarm_timer *t)
 790 {
 791     int fd = (long)t->priv;
 792
 793     close(fd);
 794 }
 795
 796 static int rtc_start_timer(struct qemu_alarm_timer *t)
 797 {
 798     int rtc_fd;
 799     unsigned long current_rtc_freq = 0;
 800
 801     TFR(rtc_fd = qemu_open("/dev/rtc", O_RDONLY));
 802     if (rtc_fd < 0)
 803         return -1;
 804     ioctl(rtc_fd, RTC_IRQP_READ, &current_rtc_freq);
 805     if (current_rtc_freq != RTC_FREQ &&
 806         ioctl(rtc_fd, RTC_IRQP_SET, RTC_FREQ) < 0) {
 807         fprintf(stderr, "Could not configure '/dev/rtc' to have a 1024 Hz timer. This is not a fatal\n"
 808                 "error, but for better emulation accuracy either use a 2.6 host Linux kernel or\n"
 809                 "type 'echo 1024 > /proc/sys/dev/rtc/max-user-freq' as root.\n");
 810         goto fail;
 811     }
 812     if (ioctl(rtc_fd, RTC_PIE_ON, 0) < 0) {
 813     fail:
 814         close(rtc_fd);
 815         return -1;
 816     }
 817
 818     enable_sigio_timer(rtc_fd);
 819
 820     t->priv = (void *)(long)rtc_fd;
 821
 822     return 0;
 823 }
 824
 825 static void rtc_stop_timer(struct qemu_alarm_timer *t)
 826 {
 827     int rtc_fd = (long)t->priv;
 828
 829     close(rtc_fd);
 830 }
 831
 832 static int dynticks_start_timer(struct qemu_alarm_timer *t)
 833 {
 834     struct sigevent ev;
 835     timer_t host_timer;
 836     struct sigaction act;
 837
 838     sigfillset(&act.sa_mask);
 839     act.sa_flags = 0;
 840     act.sa_handler = host_alarm_handler;
 841
 842     sigaction(SIGALRM, &act, NULL);
 843
 844     /*
 845      * Initialize ev struct to 0 to avoid valgrind complaining
 846      * about uninitialized data in timer_create call
 847      */
 848     memset(&ev, 0, sizeof(ev));
 849     ev.sigev_value.sival_int = 0;
 850     ev.sigev_notify = SIGEV_SIGNAL;
 851     ev.sigev_signo = SIGALRM;
 852
 853     if (timer_create(CLOCK_REALTIME, &ev, &host_timer)) {
 854         perror("timer_create");
 855
 856         /* disable dynticks */
 857         fprintf(stderr, "Dynamic Ticks disabled\n");
 858
 859         return -1;
 860     }
 861
 862     t->priv = (void *)(long)host_timer;
 863
 864     return 0;
 865 }
 866
 867 static void dynticks_stop_timer(struct qemu_alarm_timer *t)
 868 {
 869     timer_t host_timer = (timer_t)(long)t->priv;
 870
 871     timer_delete(host_timer);
 872 }
 873
 874 static void dynticks_rearm_timer(struct qemu_alarm_timer *t)
 875 {
 876     timer_t host_timer = (timer_t)(long)t->priv;
 877     struct itimerspec timeout;
 878     int64_t nearest_delta_ns = INT64_MAX;
 879     int64_t current_ns;
 880
 881     assert(alarm_has_dynticks(t));
 882     if (!active_timers[QEMU_CLOCK_REALTIME] &&
 883         !active_timers[QEMU_CLOCK_VIRTUAL] &&
 884         !active_timers[QEMU_CLOCK_HOST])
 885         return;
 886
 887     nearest_delta_ns = qemu_next_alarm_deadline();
 888     if (nearest_delta_ns < MIN_TIMER_REARM_NS)
 889         nearest_delta_ns = MIN_TIMER_REARM_NS;
 890
 891     /* check whether a timer is already running */
 892     if (timer_gettime(host_timer, &timeout)) {
 893         perror("gettime");
 894         fprintf(stderr, "Internal timer error: aborting\n");
 895         exit(1);
 896     }
 897     current_ns = timeout.it_value.tv_sec * 1000000000LL + timeout.it_value.tv_nsec;
 898     if (current_ns && current_ns <= nearest_delta_ns)
 899         return;
 900
 901     timeout.it_interval.tv_sec = 0;
 902     timeout.it_interval.tv_nsec = 0; /* 0 for one-shot timer */
 903     timeout.it_value.tv_sec =  nearest_delta_ns / 1000000000;
 904     timeout.it_value.tv_nsec = nearest_delta_ns % 1000000000;
 905     if (timer_settime(host_timer, 0 /* RELATIVE */, &timeout, NULL)) {
 906         perror("settime");
 907         fprintf(stderr, "Internal timer error: aborting\n");
 908         exit(1);
 909     }
 910 }
 911
 912 #endif /* defined(__linux__) */
 913
 914 #if !defined(_WIN32)
 915
 916 static int unix_start_timer(struct qemu_alarm_timer *t)
 917 {
 918     struct sigaction act;
 919     struct itimerval itv;
 920     int err;
 921
 922     /* timer signal */
 923     sigfillset(&act.sa_mask);
 924     act.sa_flags = 0;
 925     act.sa_handler = host_alarm_handler;
 926
 927     sigaction(SIGALRM, &act, NULL);
 928
 929     itv.it_interval.tv_sec = 0;
 930     /* for i386 kernel 2.6 to get 1 ms */
 931     itv.it_interval.tv_usec = 999;
 932     itv.it_value.tv_sec = 0;
 933     itv.it_value.tv_usec = 10 * 1000;
 934
 935     err = setitimer(ITIMER_REAL, &itv, NULL);
 936     if (err)
 937         return -1;
 938
 939     return 0;
 940 }
 941
 942 static void unix_stop_timer(struct qemu_alarm_timer *t)
 943 {
 944     struct itimerval itv;
 945
 946     memset(&itv, 0, sizeof(itv));
 947     setitimer(ITIMER_REAL, &itv, NULL);
 948 }
 949
 950 #endif /* !defined(_WIN32) */
 951
 952
 953 #ifdef _WIN32
 954
 955 static int win32_start_timer(struct qemu_alarm_timer *t)
 956 {
 957     HANDLE hTimer;
 958     BOOLEAN success;
 959
 960     /* If you call ChangeTimerQueueTimer on a one-shot timer (its period
 961        is zero) that has already expired, the timer is not updated.  Since
 962        creating a new timer is relatively expensive, set a bogus one-hour
 963        interval in the dynticks case.  */
 964     success = CreateTimerQueueTimer(&hTimer,
 965                           NULL,
 966                           host_alarm_handler,
 967                           t,
 968                           1,
 969                           alarm_has_dynticks(t) ? 3600000 : 1,
 970                           WT_EXECUTEINTIMERTHREAD);
 971
 972     if (!success) {
 973         fprintf(stderr, "Failed to initialize win32 alarm timer: %ld\n",
 974                 GetLastError());
 975         return -1;
 976     }
 977
 978     t->priv = (PVOID) hTimer;
 979     return 0;
 980 }
 981
 982 static void win32_stop_timer(struct qemu_alarm_timer *t)
 983 {
 984     HANDLE hTimer = t->priv;
 985
 986     if (hTimer) {
 987         DeleteTimerQueueTimer(NULL, hTimer, NULL);
 988     }
 989 }
 990
 991 static void win32_rearm_timer(struct qemu_alarm_timer *t)
 992 {
 993     HANDLE hTimer = t->priv;
 994     int nearest_delta_ms;
 995     BOOLEAN success;
 996
 997     assert(alarm_has_dynticks(t));
 998     if (!active_timers[QEMU_CLOCK_REALTIME] &&
 999         !active_timers[QEMU_CLOCK_VIRTUAL] &&
1000         !active_timers[QEMU_CLOCK_HOST])
1001         return;
1002
1003     nearest_delta_ms = (qemu_next_alarm_deadline() + 999999) / 1000000;
1004     if (nearest_delta_ms < 1) {
1005         nearest_delta_ms = 1;
1006     }
1007     success = ChangeTimerQueueTimer(NULL,
1008                                     hTimer,
1009                                     nearest_delta_ms,
1010                                     3600000);
1011
1012     if (!success) {
1013         fprintf(stderr, "Failed to rearm win32 alarm timer: %ld\n",
1014                 GetLastError());
1015         exit(-1);
1016     }
1017
1018 }
1019
1020 #endif /* _WIN32 */
1021
1022 static void alarm_timer_on_change_state_rearm(void *opaque, int running, int reason)
1023 {
1024     if (running)
1025         qemu_rearm_alarm_timer((struct qemu_alarm_timer *) opaque);
1026 }
1027
1028 int init_timer_alarm(void)
1029 {
1030     struct qemu_alarm_timer *t = NULL;
1031     int i, err = -1;
1032
1033     for (i = 0; alarm_timers[i].name; i++) {
1034         t = &alarm_timers[i];
1035
1036         err = t->start(t);
1037         if (!err)
1038             break;
1039     }
1040
1041     if (err) {
1042         err = -ENOENT;
1043         goto fail;
1044     }
1045
1046     /* first event is at time 0 */
1047     t->pending = 1;
1048     alarm_timer = t;
1049     qemu_add_vm_change_state_handler(alarm_timer_on_change_state_rearm, t);
1050
1051     return 0;
1052
1053 fail:
1054     return err;
1055 }
1056
1057 void quit_timers(void)
1058 {
1059     struct qemu_alarm_timer *t = alarm_timer;
1060     alarm_timer = NULL;
1061     t->stop(t);
1062 }
1063
1064 int qemu_calculate_timeout(void)
1065 {
1066     int timeout;
1067     int64_t add;
1068     int64_t delta;
1069
1070     /* When using icount, making forward progress with qemu_icount when the
1071        guest CPU is idle is critical. We only use the static io-thread timeout
1072        for non icount runs.  */
1073     if (!use_icount || !vm_running) {
1074         return 5000;
1075     }
1076
1077     /* Advance virtual time to the next event.  */
1078     delta = qemu_icount_delta();
1079     if (delta > 0) {
1080         /* If virtual time is ahead of real time then just
1081            wait for IO.  */
1082         timeout = (delta + 999999) / 1000000;
1083     } else {
1084         /* Wait for either IO to occur or the next
1085            timer event.  */
1086         add = qemu_next_deadline();
1087         /* We advance the timer before checking for IO.
1088            Limit the amount we advance so that early IO
1089            activity won't get the guest too far ahead.  */
1090         if (add > 10000000)
1091             add = 10000000;
1092         delta += add;
1093         qemu_icount += qemu_icount_round (add);
1094         timeout = delta / 1000000;
1095         if (timeout < 0)
1096             timeout = 0;
1097     }
1098
1099     return timeout;
1100 }
1101