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 /***********************************************************/
  69 /* real time host monotonic timer */
  70
  71
  72 static int64_t get_clock_realtime(void)
  73 {
  74     struct timeval tv;
  75
  76     gettimeofday(&tv, NULL);
  77     return tv.tv_sec * 1000000000LL + (tv.tv_usec * 1000);
  78 }
  79
  80 #ifdef WIN32
  81
  82 static int64_t clock_freq;
  83
  84 static void init_get_clock(void)
  85 {
  86     LARGE_INTEGER freq;
  87     int ret;
  88     ret = QueryPerformanceFrequency(&freq);
  89     if (ret == 0) {
  90         fprintf(stderr, "Could not calibrate ticks\n");
  91         exit(1);
  92     }
  93     clock_freq = freq.QuadPart;
  94 }
  95
  96 static int64_t get_clock(void)
  97 {
  98     LARGE_INTEGER ti;
  99     QueryPerformanceCounter(&ti);
 100     return muldiv64(ti.QuadPart, get_ticks_per_sec(), clock_freq);
 101 }
 102
 103 #else
 104
 105 static int use_rt_clock;
 106
 107 static void init_get_clock(void)
 108 {
 109     use_rt_clock = 0;
 110 #if defined(__linux__) || (defined(__FreeBSD__) && __FreeBSD_version >= 500000) \
 111     || defined(__DragonFly__) || defined(__FreeBSD_kernel__)
 112     {
 113         struct timespec ts;
 114         if (clock_gettime(CLOCK_MONOTONIC, &ts) == 0) {
 115             use_rt_clock = 1;
 116         }
 117     }
 118 #endif
 119 }
 120
 121 static int64_t get_clock(void)
 122 {
 123 #if defined(__linux__) || (defined(__FreeBSD__) && __FreeBSD_version >= 500000) \
 124         || defined(__DragonFly__) || defined(__FreeBSD_kernel__)
 125     if (use_rt_clock) {
 126         struct timespec ts;
 127         clock_gettime(CLOCK_MONOTONIC, &ts);
 128         return ts.tv_sec * 1000000000LL + ts.tv_nsec;
 129     } else
 130 #endif
 131     {
 132         /* XXX: using gettimeofday leads to problems if the date
 133            changes, so it should be avoided. */
 134         return get_clock_realtime();
 135     }
 136 }
 137 #endif
 138
 139 /***********************************************************/
 140 /* guest cycle counter */
 141
 142 typedef struct TimersState {
 143     int64_t cpu_ticks_prev;
 144     int64_t cpu_ticks_offset;
 145     int64_t cpu_clock_offset;
 146     int32_t cpu_ticks_enabled;
 147     int64_t dummy;
 148 } TimersState;
 149
 150 TimersState timers_state;
 151
 152 /* return the host CPU cycle counter and handle stop/restart */
 153 int64_t cpu_get_ticks(void)
 154 {
 155     if (use_icount) {
 156         return cpu_get_icount();
 157     }
 158     if (!timers_state.cpu_ticks_enabled) {
 159         return timers_state.cpu_ticks_offset;
 160     } else {
 161         int64_t ticks;
 162         ticks = cpu_get_real_ticks();
 163         if (timers_state.cpu_ticks_prev > ticks) {
 164             /* Note: non increasing ticks may happen if the host uses
 165                software suspend */
 166             timers_state.cpu_ticks_offset += timers_state.cpu_ticks_prev - ticks;
 167         }
 168         timers_state.cpu_ticks_prev = ticks;
 169         return ticks + timers_state.cpu_ticks_offset;
 170     }
 171 }
 172
 173 /* return the host CPU monotonic timer and handle stop/restart */
 174 static int64_t cpu_get_clock(void)
 175 {
 176     int64_t ti;
 177     if (!timers_state.cpu_ticks_enabled) {
 178         return timers_state.cpu_clock_offset;
 179     } else {
 180         ti = get_clock();
 181         return ti + timers_state.cpu_clock_offset;
 182     }
 183 }
 184
 185 /* FIXME: qemu-kvm hack */
 186 #define CONFIG_IOTHREAD 1
 187 #ifndef CONFIG_IOTHREAD
 188 static int64_t qemu_icount_delta(void)
 189 {
 190     if (!use_icount) {
 191         return 5000 * (int64_t) 1000000;
 192     } else if (use_icount == 1) {
 193         /* When not using an adaptive execution frequency
 194            we tend to get badly out of sync with real time,
 195            so just delay for a reasonable amount of time.  */
 196         return 0;
 197     } else {
 198         return cpu_get_icount() - cpu_get_clock();
 199     }
 200 }
 201 #endif
 202
 203 /* enable cpu_get_ticks() */
 204 void cpu_enable_ticks(void)
 205 {
 206     if (!timers_state.cpu_ticks_enabled) {
 207         timers_state.cpu_ticks_offset -= cpu_get_real_ticks();
 208         timers_state.cpu_clock_offset -= get_clock();
 209         timers_state.cpu_ticks_enabled = 1;
 210     }
 211 }
 212
 213 /* disable cpu_get_ticks() : the clock is stopped. You must not call
 214    cpu_get_ticks() after that.  */
 215 void cpu_disable_ticks(void)
 216 {
 217     if (timers_state.cpu_ticks_enabled) {
 218         timers_state.cpu_ticks_offset = cpu_get_ticks();
 219         timers_state.cpu_clock_offset = cpu_get_clock();
 220         timers_state.cpu_ticks_enabled = 0;
 221     }
 222 }
 223
 224 /***********************************************************/
 225 /* timers */
 226
 227 #define QEMU_CLOCK_REALTIME 0
 228 #define QEMU_CLOCK_VIRTUAL  1
 229 #define QEMU_CLOCK_HOST     2
 230
 231 struct QEMUClock {
 232     int type;
 233     int enabled;
 234     /* XXX: add frequency */
 235 };
 236
 237 struct QEMUTimer {
 238     QEMUClock *clock;
 239     int64_t expire_time;
 240     QEMUTimerCB *cb;
 241     void *opaque;
 242     struct QEMUTimer *next;
 243 };
 244
 245 struct qemu_alarm_timer {
 246     char const *name;
 247     int (*start)(struct qemu_alarm_timer *t);
 248     void (*stop)(struct qemu_alarm_timer *t);
 249     void (*rearm)(struct qemu_alarm_timer *t);
 250     void *priv;
 251
 252     char expired;
 253     char pending;
 254 };
 255
 256 static struct qemu_alarm_timer *alarm_timer;
 257
 258 int qemu_alarm_pending(void)
 259 {
 260     return alarm_timer->pending;
 261 }
 262
 263 static inline int alarm_has_dynticks(struct qemu_alarm_timer *t)
 264 {
 265     return !!t->rearm;
 266 }
 267
 268 static void qemu_rearm_alarm_timer(struct qemu_alarm_timer *t)
 269 {
 270     if (!alarm_has_dynticks(t))
 271         return;
 272
 273     t->rearm(t);
 274 }
 275
 276 /* TODO: MIN_TIMER_REARM_US should be optimized */
 277 #define MIN_TIMER_REARM_US 250
 278
 279 #ifdef _WIN32
 280
 281 struct qemu_alarm_win32 {
 282     MMRESULT timerId;
 283     unsigned int period;
 284 } alarm_win32_data = {0, 0};
 285
 286 static int win32_start_timer(struct qemu_alarm_timer *t);
 287 static void win32_stop_timer(struct qemu_alarm_timer *t);
 288 static void win32_rearm_timer(struct qemu_alarm_timer *t);
 289
 290 #else
 291
 292 static int unix_start_timer(struct qemu_alarm_timer *t);
 293 static void unix_stop_timer(struct qemu_alarm_timer *t);
 294
 295 #ifdef __linux__
 296
 297 static int dynticks_start_timer(struct qemu_alarm_timer *t);
 298 static void dynticks_stop_timer(struct qemu_alarm_timer *t);
 299 static void dynticks_rearm_timer(struct qemu_alarm_timer *t);
 300
 301 static int hpet_start_timer(struct qemu_alarm_timer *t);
 302 static void hpet_stop_timer(struct qemu_alarm_timer *t);
 303
 304 static int rtc_start_timer(struct qemu_alarm_timer *t);
 305 static void rtc_stop_timer(struct qemu_alarm_timer *t);
 306
 307 #endif /* __linux__ */
 308
 309 #endif /* _WIN32 */
 310
 311 /* Correlation between real and virtual time is always going to be
 312    fairly approximate, so ignore small variation.
 313    When the guest is idle real and virtual time will be aligned in
 314    the IO wait loop.  */
 315 #define ICOUNT_WOBBLE (get_ticks_per_sec() / 10)
 316
 317 static void icount_adjust(void)
 318 {
 319     int64_t cur_time;
 320     int64_t cur_icount;
 321     int64_t delta;
 322     static int64_t last_delta;
 323     /* If the VM is not running, then do nothing.  */
 324     if (!vm_running)
 325         return;
 326
 327     cur_time = cpu_get_clock();
 328     cur_icount = qemu_get_clock(vm_clock);
 329     delta = cur_icount - cur_time;
 330     /* FIXME: This is a very crude algorithm, somewhat prone to oscillation.  */
 331     if (delta > 0
 332         && last_delta + ICOUNT_WOBBLE < delta * 2
 333         && icount_time_shift > 0) {
 334         /* The guest is getting too far ahead.  Slow time down.  */
 335         icount_time_shift--;
 336     }
 337     if (delta < 0
 338         && last_delta - ICOUNT_WOBBLE > delta * 2
 339         && icount_time_shift < MAX_ICOUNT_SHIFT) {
 340         /* The guest is getting too far behind.  Speed time up.  */
 341         icount_time_shift++;
 342     }
 343     last_delta = delta;
 344     qemu_icount_bias = cur_icount - (qemu_icount << icount_time_shift);
 345 }
 346
 347 static void icount_adjust_rt(void * opaque)
 348 {
 349     qemu_mod_timer(icount_rt_timer,
 350                    qemu_get_clock(rt_clock) + 1000);
 351     icount_adjust();
 352 }
 353
 354 static void icount_adjust_vm(void * opaque)
 355 {
 356     qemu_mod_timer(icount_vm_timer,
 357                    qemu_get_clock(vm_clock) + get_ticks_per_sec() / 10);
 358     icount_adjust();
 359 }
 360
 361 int64_t qemu_icount_round(int64_t count)
 362 {
 363     return (count + (1 << icount_time_shift) - 1) >> icount_time_shift;
 364 }
 365
 366 static struct qemu_alarm_timer alarm_timers[] = {
 367 #ifndef _WIN32
 368 #ifdef __linux__
 369     {"dynticks", dynticks_start_timer,
 370      dynticks_stop_timer, dynticks_rearm_timer, NULL},
 371     /* HPET - if available - is preferred */
 372     {"hpet", hpet_start_timer, hpet_stop_timer, NULL, NULL},
 373     /* ...otherwise try RTC */
 374     {"rtc", rtc_start_timer, rtc_stop_timer, NULL, NULL},
 375 #endif
 376     {"unix", unix_start_timer, unix_stop_timer, NULL, NULL},
 377 #else
 378     {"dynticks", win32_start_timer,
 379      win32_stop_timer, win32_rearm_timer, &alarm_win32_data},
 380     {"win32", win32_start_timer,
 381      win32_stop_timer, NULL, &alarm_win32_data},
 382 #endif
 383     {NULL, }
 384 };
 385
 386 static void show_available_alarms(void)
 387 {
 388     int i;
 389
 390     printf("Available alarm timers, in order of precedence:\n");
 391     for (i = 0; alarm_timers[i].name; i++)
 392         printf("%s\n", alarm_timers[i].name);
 393 }
 394
 395 void configure_alarms(char const *opt)
 396 {
 397     int i;
 398     int cur = 0;
 399     int count = ARRAY_SIZE(alarm_timers) - 1;
 400     char *arg;
 401     char *name;
 402     struct qemu_alarm_timer tmp;
 403
 404     if (!strcmp(opt, "?")) {
 405         show_available_alarms();
 406         exit(0);
 407     }
 408
 409     arg = qemu_strdup(opt);
 410
 411     /* Reorder the array */
 412     name = strtok(arg, ",");
 413     while (name) {
 414         for (i = 0; i < count && alarm_timers[i].name; i++) {
 415             if (!strcmp(alarm_timers[i].name, name))
 416                 break;
 417         }
 418
 419         if (i == count) {
 420             fprintf(stderr, "Unknown clock %s\n", name);
 421             goto next;
 422         }
 423
 424         if (i < cur)
 425             /* Ignore */
 426             goto next;
 427
 428         /* Swap */
 429         tmp = alarm_timers[i];
 430         alarm_timers[i] = alarm_timers[cur];
 431         alarm_timers[cur] = tmp;
 432
 433         cur++;
 434 next:
 435         name = strtok(NULL, ",");
 436     }
 437
 438     qemu_free(arg);
 439
 440     if (cur) {
 441         /* Disable remaining timers */
 442         for (i = cur; i < count; i++)
 443             alarm_timers[i].name = NULL;
 444     } else {
 445         show_available_alarms();
 446         exit(1);
 447     }
 448 }
 449
 450 #define QEMU_NUM_CLOCKS 3
 451
 452 QEMUClock *rt_clock;
 453 QEMUClock *vm_clock;
 454 QEMUClock *host_clock;
 455
 456 static QEMUTimer *active_timers[QEMU_NUM_CLOCKS];
 457
 458 static QEMUClock *qemu_new_clock(int type)
 459 {
 460     QEMUClock *clock;
 461     clock = qemu_mallocz(sizeof(QEMUClock));
 462     clock->type = type;
 463     clock->enabled = 1;
 464     return clock;
 465 }
 466
 467 void qemu_clock_enable(QEMUClock *clock, int enabled)
 468 {
 469     clock->enabled = enabled;
 470 }
 471
 472 QEMUTimer *qemu_new_timer(QEMUClock *clock, QEMUTimerCB *cb, void *opaque)
 473 {
 474     QEMUTimer *ts;
 475
 476     ts = qemu_mallocz(sizeof(QEMUTimer));
 477     ts->clock = clock;
 478     ts->cb = cb;
 479     ts->opaque = opaque;
 480     return ts;
 481 }
 482
 483 void qemu_free_timer(QEMUTimer *ts)
 484 {
 485     qemu_free(ts);
 486 }
 487
 488 /* stop a timer, but do not dealloc it */
 489 void qemu_del_timer(QEMUTimer *ts)
 490 {
 491     QEMUTimer **pt, *t;
 492
 493     /* NOTE: this code must be signal safe because
 494        qemu_timer_expired() can be called from a signal. */
 495     pt = &active_timers[ts->clock->type];
 496     for(;;) {
 497         t = *pt;
 498         if (!t)
 499             break;
 500         if (t == ts) {
 501             *pt = t->next;
 502             break;
 503         }
 504         pt = &t->next;
 505     }
 506 }
 507
 508 /* modify the current timer so that it will be fired when current_time
 509    >= expire_time. The corresponding callback will be called. */
 510 void qemu_mod_timer(QEMUTimer *ts, int64_t expire_time)
 511 {
 512     QEMUTimer **pt, *t;
 513
 514     qemu_del_timer(ts);
 515
 516     /* add the timer in the sorted list */
 517     /* NOTE: this code must be signal safe because
 518        qemu_timer_expired() can be called from a signal. */
 519     pt = &active_timers[ts->clock->type];
 520     for(;;) {
 521         t = *pt;
 522         if (!t)
 523             break;
 524         if (t->expire_time > expire_time)
 525             break;
 526         pt = &t->next;
 527     }
 528     ts->expire_time = expire_time;
 529     ts->next = *pt;
 530     *pt = ts;
 531
 532     /* Rearm if necessary  */
 533     if (pt == &active_timers[ts->clock->type]) {
 534         if (!alarm_timer->pending) {
 535             qemu_rearm_alarm_timer(alarm_timer);
 536         }
 537         /* Interrupt execution to force deadline recalculation.  */
 538         if (use_icount)
 539             qemu_notify_event();
 540     }
 541 }
 542
 543 int qemu_timer_pending(QEMUTimer *ts)
 544 {
 545     QEMUTimer *t;
 546     for(t = active_timers[ts->clock->type]; t != NULL; t = t->next) {
 547         if (t == ts)
 548             return 1;
 549     }
 550     return 0;
 551 }
 552
 553 int qemu_timer_expired(QEMUTimer *timer_head, int64_t current_time)
 554 {
 555     if (!timer_head)
 556         return 0;
 557     return (timer_head->expire_time <= current_time);
 558 }
 559
 560 static void qemu_run_timers(QEMUClock *clock)
 561 {
 562     QEMUTimer **ptimer_head, *ts;
 563     int64_t current_time;
 564
 565     if (!clock->enabled)
 566         return;
 567
 568     current_time = qemu_get_clock (clock);
 569     ptimer_head = &active_timers[clock->type];
 570     for(;;) {
 571         ts = *ptimer_head;
 572         if (!ts || ts->expire_time > current_time)
 573             break;
 574         /* remove timer from the list before calling the callback */
 575         *ptimer_head = ts->next;
 576         ts->next = NULL;
 577
 578         /* run the callback (the timer list can be modified) */
 579         ts->cb(ts->opaque);
 580     }
 581 }
 582
 583 int64_t qemu_get_clock(QEMUClock *clock)
 584 {
 585     switch(clock->type) {
 586     case QEMU_CLOCK_REALTIME:
 587         return get_clock() / 1000000;
 588     default:
 589     case QEMU_CLOCK_VIRTUAL:
 590         if (use_icount) {
 591             return cpu_get_icount();
 592         } else {
 593             return cpu_get_clock();
 594         }
 595     case QEMU_CLOCK_HOST:
 596         return get_clock_realtime();
 597     }
 598 }
 599
 600 int64_t qemu_get_clock_ns(QEMUClock *clock)
 601 {
 602     switch(clock->type) {
 603     case QEMU_CLOCK_REALTIME:
 604         return get_clock();
 605     default:
 606     case QEMU_CLOCK_VIRTUAL:
 607         if (use_icount) {
 608             return cpu_get_icount();
 609         } else {
 610             return cpu_get_clock();
 611         }
 612     case QEMU_CLOCK_HOST:
 613         return get_clock_realtime();
 614     }
 615 }
 616
 617 void init_clocks(void)
 618 {
 619     init_get_clock();
 620     rt_clock = qemu_new_clock(QEMU_CLOCK_REALTIME);
 621     vm_clock = qemu_new_clock(QEMU_CLOCK_VIRTUAL);
 622     host_clock = qemu_new_clock(QEMU_CLOCK_HOST);
 623
 624     rtc_clock = host_clock;
 625 }
 626
 627 /* save a timer */
 628 void qemu_put_timer(QEMUFile *f, QEMUTimer *ts)
 629 {
 630     uint64_t expire_time;
 631
 632     if (qemu_timer_pending(ts)) {
 633         expire_time = ts->expire_time;
 634     } else {
 635         expire_time = -1;
 636     }
 637     qemu_put_be64(f, expire_time);
 638 }
 639
 640 void qemu_get_timer(QEMUFile *f, QEMUTimer *ts)
 641 {
 642     uint64_t expire_time;
 643
 644     expire_time = qemu_get_be64(f);
 645     if (expire_time != -1) {
 646         qemu_mod_timer(ts, expire_time);
 647     } else {
 648         qemu_del_timer(ts);
 649     }
 650 }
 651
 652 static const VMStateDescription vmstate_timers = {
 653     .name = "timer",
 654     .version_id = 2,
 655     .minimum_version_id = 1,
 656     .minimum_version_id_old = 1,
 657     .fields      = (VMStateField []) {
 658         VMSTATE_INT64(cpu_ticks_offset, TimersState),
 659         VMSTATE_INT64(dummy, TimersState),
 660         VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2),
 661         VMSTATE_END_OF_LIST()
 662     }
 663 };
 664
 665 void configure_icount(const char *option)
 666 {
 667     vmstate_register(0, &vmstate_timers, &timers_state);
 668     if (!option)
 669         return;
 670
 671     if (strcmp(option, "auto") != 0) {
 672         icount_time_shift = strtol(option, NULL, 0);
 673         use_icount = 1;
 674         return;
 675     }
 676
 677     use_icount = 2;
 678
 679     /* 125MIPS seems a reasonable initial guess at the guest speed.
 680        It will be corrected fairly quickly anyway.  */
 681     icount_time_shift = 3;
 682
 683     /* Have both realtime and virtual time triggers for speed adjustment.
 684        The realtime trigger catches emulated time passing too slowly,
 685        the virtual time trigger catches emulated time passing too fast.
 686        Realtime triggers occur even when idle, so use them less frequently
 687        than VM triggers.  */
 688     icount_rt_timer = qemu_new_timer(rt_clock, icount_adjust_rt, NULL);
 689     qemu_mod_timer(icount_rt_timer,
 690                    qemu_get_clock(rt_clock) + 1000);
 691     icount_vm_timer = qemu_new_timer(vm_clock, icount_adjust_vm, NULL);
 692     qemu_mod_timer(icount_vm_timer,
 693                    qemu_get_clock(vm_clock) + get_ticks_per_sec() / 10);
 694 }
 695
 696 void qemu_run_all_timers(void)
 697 {
 698     alarm_timer->pending = 0;
 699
 700     /* rearm timer, if not periodic */
 701     if (alarm_timer->expired) {
 702         alarm_timer->expired = 0;
 703         qemu_rearm_alarm_timer(alarm_timer);
 704     }
 705
 706     /* vm time timers */
 707     if (vm_running) {
 708         qemu_run_timers(vm_clock);
 709     }
 710
 711     qemu_run_timers(rt_clock);
 712     qemu_run_timers(host_clock);
 713 }
 714
 715 #ifdef _WIN32
 716 static void CALLBACK host_alarm_handler(UINT uTimerID, UINT uMsg,
 717                                         DWORD_PTR dwUser, DWORD_PTR dw1,
 718                                         DWORD_PTR dw2)
 719 #else
 720 static void host_alarm_handler(int host_signum)
 721 #endif
 722 {
 723     struct qemu_alarm_timer *t = alarm_timer;
 724     if (!t)
 725         return;
 726
 727 #if 0
 728 #define DISP_FREQ 1000
 729     {
 730         static int64_t delta_min = INT64_MAX;
 731         static int64_t delta_max, delta_cum, last_clock, delta, ti;
 732         static int count;
 733         ti = qemu_get_clock(vm_clock);
 734         if (last_clock != 0) {
 735             delta = ti - last_clock;
 736             if (delta < delta_min)
 737                 delta_min = delta;
 738             if (delta > delta_max)
 739                 delta_max = delta;
 740             delta_cum += delta;
 741             if (++count == DISP_FREQ) {
 742                 printf("timer: min=%" PRId64 " us max=%" PRId64 " us avg=%" PRId64 " us avg_freq=%0.3f Hz\n",
 743                        muldiv64(delta_min, 1000000, get_ticks_per_sec()),
 744                        muldiv64(delta_max, 1000000, get_ticks_per_sec()),
 745                        muldiv64(delta_cum, 1000000 / DISP_FREQ, get_ticks_per_sec()),
 746                        (double)get_ticks_per_sec() / ((double)delta_cum / DISP_FREQ));
 747                 count = 0;
 748                 delta_min = INT64_MAX;
 749                 delta_max = 0;
 750                 delta_cum = 0;
 751             }
 752         }
 753         last_clock = ti;
 754     }
 755 #endif
 756     if (alarm_has_dynticks(t) ||
 757         (!use_icount &&
 758             qemu_timer_expired(active_timers[QEMU_CLOCK_VIRTUAL],
 759                                qemu_get_clock(vm_clock))) ||
 760         qemu_timer_expired(active_timers[QEMU_CLOCK_REALTIME],
 761                            qemu_get_clock(rt_clock)) ||
 762         qemu_timer_expired(active_timers[QEMU_CLOCK_HOST],
 763                            qemu_get_clock(host_clock))) {
 764
 765         t->expired = alarm_has_dynticks(t);
 766         t->pending = 1;
 767         qemu_notify_event();
 768     }
 769 }
 770
 771 int64_t qemu_next_deadline(void)
 772 {
 773     /* To avoid problems with overflow limit this to 2^32.  */
 774     int64_t delta = INT32_MAX;
 775
 776     if (active_timers[QEMU_CLOCK_VIRTUAL]) {
 777         delta = active_timers[QEMU_CLOCK_VIRTUAL]->expire_time -
 778                      qemu_get_clock(vm_clock);
 779     }
 780     if (active_timers[QEMU_CLOCK_HOST]) {
 781         int64_t hdelta = active_timers[QEMU_CLOCK_HOST]->expire_time -
 782                  qemu_get_clock(host_clock);
 783         if (hdelta < delta)
 784             delta = hdelta;
 785     }
 786
 787     if (delta < 0)
 788         delta = 0;
 789
 790     return delta;
 791 }
 792
 793 #ifndef _WIN32
 794
 795 #if defined(__linux__)
 796
 797 #define RTC_FREQ 1024
 798
 799 static uint64_t qemu_next_deadline_dyntick(void)
 800 {
 801     int64_t delta;
 802     int64_t rtdelta;
 803
 804     if (use_icount)
 805         delta = INT32_MAX;
 806     else
 807         delta = (qemu_next_deadline() + 999) / 1000;
 808
 809     if (active_timers[QEMU_CLOCK_REALTIME]) {
 810         rtdelta = (active_timers[QEMU_CLOCK_REALTIME]->expire_time -
 811                  qemu_get_clock(rt_clock))*1000;
 812         if (rtdelta < delta)
 813             delta = rtdelta;
 814     }
 815
 816     if (delta < MIN_TIMER_REARM_US)
 817         delta = MIN_TIMER_REARM_US;
 818
 819     return delta;
 820 }
 821
 822 static void enable_sigio_timer(int fd)
 823 {
 824     struct sigaction act;
 825
 826     /* timer signal */
 827     sigfillset(&act.sa_mask);
 828     act.sa_flags = 0;
 829     act.sa_handler = host_alarm_handler;
 830
 831     sigaction(SIGIO, &act, NULL);
 832     fcntl_setfl(fd, O_ASYNC);
 833     fcntl(fd, F_SETOWN, getpid());
 834 }
 835
 836 static int hpet_start_timer(struct qemu_alarm_timer *t)
 837 {
 838     struct hpet_info info;
 839     int r, fd;
 840
 841     fd = qemu_open("/dev/hpet", O_RDONLY);
 842     if (fd < 0)
 843         return -1;
 844
 845     /* Set frequency */
 846     r = ioctl(fd, HPET_IRQFREQ, RTC_FREQ);
 847     if (r < 0) {
 848         fprintf(stderr, "Could not configure '/dev/hpet' to have a 1024Hz timer. This is not a fatal\n"
 849                 "error, but for better emulation accuracy type:\n"
 850                 "'echo 1024 > /proc/sys/dev/hpet/max-user-freq' as root.\n");
 851         goto fail;
 852     }
 853
 854     /* Check capabilities */
 855     r = ioctl(fd, HPET_INFO, &info);
 856     if (r < 0)
 857         goto fail;
 858
 859     /* Enable periodic mode */
 860     r = ioctl(fd, HPET_EPI, 0);
 861     if (info.hi_flags && (r < 0))
 862         goto fail;
 863
 864     /* Enable interrupt */
 865     r = ioctl(fd, HPET_IE_ON, 0);
 866     if (r < 0)
 867         goto fail;
 868
 869     enable_sigio_timer(fd);
 870     t->priv = (void *)(long)fd;
 871
 872     return 0;
 873 fail:
 874     close(fd);
 875     return -1;
 876 }
 877
 878 static void hpet_stop_timer(struct qemu_alarm_timer *t)
 879 {
 880     int fd = (long)t->priv;
 881
 882     close(fd);
 883 }
 884
 885 static int rtc_start_timer(struct qemu_alarm_timer *t)
 886 {
 887     int rtc_fd;
 888     unsigned long current_rtc_freq = 0;
 889
 890     TFR(rtc_fd = qemu_open("/dev/rtc", O_RDONLY));
 891     if (rtc_fd < 0)
 892         return -1;
 893     ioctl(rtc_fd, RTC_IRQP_READ, &current_rtc_freq);
 894     if (current_rtc_freq != RTC_FREQ &&
 895         ioctl(rtc_fd, RTC_IRQP_SET, RTC_FREQ) < 0) {
 896         fprintf(stderr, "Could not configure '/dev/rtc' to have a 1024 Hz timer. This is not a fatal\n"
 897                 "error, but for better emulation accuracy either use a 2.6 host Linux kernel or\n"
 898                 "type 'echo 1024 > /proc/sys/dev/rtc/max-user-freq' as root.\n");
 899         goto fail;
 900     }
 901     if (ioctl(rtc_fd, RTC_PIE_ON, 0) < 0) {
 902     fail:
 903         close(rtc_fd);
 904         return -1;
 905     }
 906
 907     enable_sigio_timer(rtc_fd);
 908
 909     t->priv = (void *)(long)rtc_fd;
 910
 911     return 0;
 912 }
 913
 914 static void rtc_stop_timer(struct qemu_alarm_timer *t)
 915 {
 916     int rtc_fd = (long)t->priv;
 917
 918     close(rtc_fd);
 919 }
 920
 921 static int dynticks_start_timer(struct qemu_alarm_timer *t)
 922 {
 923     struct sigevent ev;
 924     timer_t host_timer;
 925     struct sigaction act;
 926
 927     sigfillset(&act.sa_mask);
 928     act.sa_flags = 0;
 929     act.sa_handler = host_alarm_handler;
 930
 931     sigaction(SIGALRM, &act, NULL);
 932
 933     /*
 934      * Initialize ev struct to 0 to avoid valgrind complaining
 935      * about uninitialized data in timer_create call
 936      */
 937     memset(&ev, 0, sizeof(ev));
 938     ev.sigev_value.sival_int = 0;
 939     ev.sigev_notify = SIGEV_SIGNAL;
 940     ev.sigev_signo = SIGALRM;
 941
 942     if (timer_create(CLOCK_REALTIME, &ev, &host_timer)) {
 943         perror("timer_create");
 944
 945         /* disable dynticks */
 946         fprintf(stderr, "Dynamic Ticks disabled\n");
 947
 948         return -1;
 949     }
 950
 951     t->priv = (void *)(long)host_timer;
 952
 953     return 0;
 954 }
 955
 956 static void dynticks_stop_timer(struct qemu_alarm_timer *t)
 957 {
 958     timer_t host_timer = (timer_t)(long)t->priv;
 959
 960     timer_delete(host_timer);
 961 }
 962
 963 static void dynticks_rearm_timer(struct qemu_alarm_timer *t)
 964 {
 965     timer_t host_timer = (timer_t)(long)t->priv;
 966     struct itimerspec timeout;
 967     int64_t nearest_delta_us = INT64_MAX;
 968     int64_t current_us;
 969
 970     assert(alarm_has_dynticks(t));
 971     if (!active_timers[QEMU_CLOCK_REALTIME] &&
 972         !active_timers[QEMU_CLOCK_VIRTUAL] &&
 973         !active_timers[QEMU_CLOCK_HOST])
 974         return;
 975
 976     nearest_delta_us = qemu_next_deadline_dyntick();
 977
 978     /* check whether a timer is already running */
 979     if (timer_gettime(host_timer, &timeout)) {
 980         perror("gettime");
 981         fprintf(stderr, "Internal timer error: aborting\n");
 982         exit(1);
 983     }
 984     current_us = timeout.it_value.tv_sec * 1000000 + timeout.it_value.tv_nsec/1000;
 985     if (current_us && current_us <= nearest_delta_us)
 986         return;
 987
 988     timeout.it_interval.tv_sec = 0;
 989     timeout.it_interval.tv_nsec = 0; /* 0 for one-shot timer */
 990     timeout.it_value.tv_sec =  nearest_delta_us / 1000000;
 991     timeout.it_value.tv_nsec = (nearest_delta_us % 1000000) * 1000;
 992     if (timer_settime(host_timer, 0 /* RELATIVE */, &timeout, NULL)) {
 993         perror("settime");
 994         fprintf(stderr, "Internal timer error: aborting\n");
 995         exit(1);
 996     }
 997 }
 998
 999 #endif /* defined(__linux__) */
1000
1001 static int unix_start_timer(struct qemu_alarm_timer *t)
1002 {
1003     struct sigaction act;
1004     struct itimerval itv;
1005     int err;
1006
1007     /* timer signal */
1008     sigfillset(&act.sa_mask);
1009     act.sa_flags = 0;
1010     act.sa_handler = host_alarm_handler;
1011
1012     sigaction(SIGALRM, &act, NULL);
1013
1014     itv.it_interval.tv_sec = 0;
1015     /* for i386 kernel 2.6 to get 1 ms */
1016     itv.it_interval.tv_usec = 999;
1017     itv.it_value.tv_sec = 0;
1018     itv.it_value.tv_usec = 10 * 1000;
1019
1020     err = setitimer(ITIMER_REAL, &itv, NULL);
1021     if (err)
1022         return -1;
1023
1024     return 0;
1025 }
1026
1027 static void unix_stop_timer(struct qemu_alarm_timer *t)
1028 {
1029     struct itimerval itv;
1030
1031     memset(&itv, 0, sizeof(itv));
1032     setitimer(ITIMER_REAL, &itv, NULL);
1033 }
1034
1035 #endif /* !defined(_WIN32) */
1036
1037
1038 #ifdef _WIN32
1039
1040 static int win32_start_timer(struct qemu_alarm_timer *t)
1041 {
1042     TIMECAPS tc;
1043     struct qemu_alarm_win32 *data = t->priv;
1044     UINT flags;
1045
1046     memset(&tc, 0, sizeof(tc));
1047     timeGetDevCaps(&tc, sizeof(tc));
1048
1049     data->period = tc.wPeriodMin;
1050     timeBeginPeriod(data->period);
1051
1052     flags = TIME_CALLBACK_FUNCTION;
1053     if (alarm_has_dynticks(t))
1054         flags |= TIME_ONESHOT;
1055     else
1056         flags |= TIME_PERIODIC;
1057
1058     data->timerId = timeSetEvent(1,         // interval (ms)
1059                         data->period,       // resolution
1060                         host_alarm_handler, // function
1061                         (DWORD)t,           // parameter
1062                         flags);
1063
1064     if (!data->timerId) {
1065         fprintf(stderr, "Failed to initialize win32 alarm timer: %ld\n",
1066                 GetLastError());
1067         timeEndPeriod(data->period);
1068         return -1;
1069     }
1070
1071     return 0;
1072 }
1073
1074 static void win32_stop_timer(struct qemu_alarm_timer *t)
1075 {
1076     struct qemu_alarm_win32 *data = t->priv;
1077
1078     timeKillEvent(data->timerId);
1079     timeEndPeriod(data->period);
1080 }
1081
1082 static void win32_rearm_timer(struct qemu_alarm_timer *t)
1083 {
1084     struct qemu_alarm_win32 *data = t->priv;
1085
1086     assert(alarm_has_dynticks(t));
1087     if (!active_timers[QEMU_CLOCK_REALTIME] &&
1088         !active_timers[QEMU_CLOCK_VIRTUAL] &&
1089         !active_timers[QEMU_CLOCK_HOST])
1090         return;
1091
1092     timeKillEvent(data->timerId);
1093
1094     data->timerId = timeSetEvent(1,
1095                         data->period,
1096                         host_alarm_handler,
1097                         (DWORD)t,
1098                         TIME_ONESHOT | TIME_CALLBACK_FUNCTION);
1099
1100     if (!data->timerId) {
1101         fprintf(stderr, "Failed to re-arm win32 alarm timer %ld\n",
1102                 GetLastError());
1103
1104         timeEndPeriod(data->period);
1105         exit(1);
1106     }
1107 }
1108
1109 #endif /* _WIN32 */
1110
1111 static void alarm_timer_on_change_state_rearm(void *opaque, int running, int reason)
1112 {
1113     if (running)
1114         qemu_rearm_alarm_timer((struct qemu_alarm_timer *) opaque);
1115 }
1116
1117 int init_timer_alarm(void)
1118 {
1119     struct qemu_alarm_timer *t = NULL;
1120     int i, err = -1;
1121
1122     for (i = 0; alarm_timers[i].name; i++) {
1123         t = &alarm_timers[i];
1124
1125         err = t->start(t);
1126         if (!err)
1127             break;
1128     }
1129
1130     if (err) {
1131         err = -ENOENT;
1132         goto fail;
1133     }
1134
1135     /* first event is at time 0 */
1136     t->pending = 1;
1137     alarm_timer = t;
1138     qemu_add_vm_change_state_handler(alarm_timer_on_change_state_rearm, t);
1139
1140     return 0;
1141
1142 fail:
1143     return err;
1144 }
1145
1146 void quit_timers(void)
1147 {
1148     struct qemu_alarm_timer *t = alarm_timer;
1149     alarm_timer = NULL;
1150     t->stop(t);
1151 }
1152
1153 int qemu_calculate_timeout(void)
1154 {
1155 #ifndef CONFIG_IOTHREAD
1156     int timeout;
1157
1158     if (!vm_running)
1159         timeout = 5000;
1160     else {
1161      /* XXX: use timeout computed from timers */
1162         int64_t add;
1163         int64_t delta;
1164         /* Advance virtual time to the next event.  */
1165         delta = qemu_icount_delta();
1166         if (delta > 0) {
1167             /* If virtual time is ahead of real time then just
1168                wait for IO.  */
1169             timeout = (delta + 999999) / 1000000;
1170         } else {
1171             /* Wait for either IO to occur or the next
1172                timer event.  */
1173             add = qemu_next_deadline();
1174             /* We advance the timer before checking for IO.
1175                Limit the amount we advance so that early IO
1176                activity won't get the guest too far ahead.  */
1177             if (add > 10000000)
1178                 add = 10000000;
1179             delta += add;
1180             qemu_icount += qemu_icount_round (add);
1181             timeout = delta / 1000000;
1182             if (timeout < 0)
1183                 timeout = 0;
1184         }
1185     }
1186
1187     return timeout;
1188 #else /* CONFIG_IOTHREAD */
1189     return 1000;
1190 #endif
1191 }
1192