memory: add support getting and using a dirty bitmap copy.
[qemu.git] / util / 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 "qemu/osdep.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/timer.h"
28 #include "sysemu/replay.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/cpus.h"
31
32 #ifdef CONFIG_POSIX
33 #include <pthread.h>
34 #endif
35
36 #ifdef CONFIG_PPOLL
37 #include <poll.h>
38 #endif
39
40 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
41 #include <sys/prctl.h>
42 #endif
43
44 /***********************************************************/
45 /* timers */
46
47 typedef struct QEMUClock {
48 /* We rely on BQL to protect the timerlists */
49 QLIST_HEAD(, QEMUTimerList) timerlists;
50
51 NotifierList reset_notifiers;
52 int64_t last;
53
54 QEMUClockType type;
55 bool enabled;
56 } QEMUClock;
57
58 QEMUTimerListGroup main_loop_tlg;
59 static QEMUClock qemu_clocks[QEMU_CLOCK_MAX];
60
61 /* A QEMUTimerList is a list of timers attached to a clock. More
62 * than one QEMUTimerList can be attached to each clock, for instance
63 * used by different AioContexts / threads. Each clock also has
64 * a list of the QEMUTimerLists associated with it, in order that
65 * reenabling the clock can call all the notifiers.
66 */
67
68 struct QEMUTimerList {
69 QEMUClock *clock;
70 QemuMutex active_timers_lock;
71 QEMUTimer *active_timers;
72 QLIST_ENTRY(QEMUTimerList) list;
73 QEMUTimerListNotifyCB *notify_cb;
74 void *notify_opaque;
75
76 /* lightweight method to mark the end of timerlist's running */
77 QemuEvent timers_done_ev;
78 };
79
80 /**
81 * qemu_clock_ptr:
82 * @type: type of clock
83 *
84 * Translate a clock type into a pointer to QEMUClock object.
85 *
86 * Returns: a pointer to the QEMUClock object
87 */
88 static inline QEMUClock *qemu_clock_ptr(QEMUClockType type)
89 {
90 return &qemu_clocks[type];
91 }
92
93 static bool timer_expired_ns(QEMUTimer *timer_head, int64_t current_time)
94 {
95 return timer_head && (timer_head->expire_time <= current_time);
96 }
97
98 QEMUTimerList *timerlist_new(QEMUClockType type,
99 QEMUTimerListNotifyCB *cb,
100 void *opaque)
101 {
102 QEMUTimerList *timer_list;
103 QEMUClock *clock = qemu_clock_ptr(type);
104
105 timer_list = g_malloc0(sizeof(QEMUTimerList));
106 qemu_event_init(&timer_list->timers_done_ev, true);
107 timer_list->clock = clock;
108 timer_list->notify_cb = cb;
109 timer_list->notify_opaque = opaque;
110 qemu_mutex_init(&timer_list->active_timers_lock);
111 QLIST_INSERT_HEAD(&clock->timerlists, timer_list, list);
112 return timer_list;
113 }
114
115 void timerlist_free(QEMUTimerList *timer_list)
116 {
117 assert(!timerlist_has_timers(timer_list));
118 if (timer_list->clock) {
119 QLIST_REMOVE(timer_list, list);
120 }
121 qemu_mutex_destroy(&timer_list->active_timers_lock);
122 g_free(timer_list);
123 }
124
125 static void qemu_clock_init(QEMUClockType type, QEMUTimerListNotifyCB *notify_cb)
126 {
127 QEMUClock *clock = qemu_clock_ptr(type);
128
129 /* Assert that the clock of type TYPE has not been initialized yet. */
130 assert(main_loop_tlg.tl[type] == NULL);
131
132 clock->type = type;
133 clock->enabled = (type == QEMU_CLOCK_VIRTUAL ? false : true);
134 clock->last = INT64_MIN;
135 QLIST_INIT(&clock->timerlists);
136 notifier_list_init(&clock->reset_notifiers);
137 main_loop_tlg.tl[type] = timerlist_new(type, notify_cb, NULL);
138 }
139
140 bool qemu_clock_use_for_deadline(QEMUClockType type)
141 {
142 return !(use_icount && (type == QEMU_CLOCK_VIRTUAL));
143 }
144
145 void qemu_clock_notify(QEMUClockType type)
146 {
147 QEMUTimerList *timer_list;
148 QEMUClock *clock = qemu_clock_ptr(type);
149 QLIST_FOREACH(timer_list, &clock->timerlists, list) {
150 timerlist_notify(timer_list);
151 }
152 }
153
154 /* Disabling the clock will wait for related timerlists to stop
155 * executing qemu_run_timers. Thus, this functions should not
156 * be used from the callback of a timer that is based on @clock.
157 * Doing so would cause a deadlock.
158 *
159 * Caller should hold BQL.
160 */
161 void qemu_clock_enable(QEMUClockType type, bool enabled)
162 {
163 QEMUClock *clock = qemu_clock_ptr(type);
164 QEMUTimerList *tl;
165 bool old = clock->enabled;
166 clock->enabled = enabled;
167 if (enabled && !old) {
168 qemu_clock_notify(type);
169 } else if (!enabled && old) {
170 QLIST_FOREACH(tl, &clock->timerlists, list) {
171 qemu_event_wait(&tl->timers_done_ev);
172 }
173 }
174 }
175
176 bool timerlist_has_timers(QEMUTimerList *timer_list)
177 {
178 return !!atomic_read(&timer_list->active_timers);
179 }
180
181 bool qemu_clock_has_timers(QEMUClockType type)
182 {
183 return timerlist_has_timers(
184 main_loop_tlg.tl[type]);
185 }
186
187 bool timerlist_expired(QEMUTimerList *timer_list)
188 {
189 int64_t expire_time;
190
191 if (!atomic_read(&timer_list->active_timers)) {
192 return false;
193 }
194
195 qemu_mutex_lock(&timer_list->active_timers_lock);
196 if (!timer_list->active_timers) {
197 qemu_mutex_unlock(&timer_list->active_timers_lock);
198 return false;
199 }
200 expire_time = timer_list->active_timers->expire_time;
201 qemu_mutex_unlock(&timer_list->active_timers_lock);
202
203 return expire_time <= qemu_clock_get_ns(timer_list->clock->type);
204 }
205
206 bool qemu_clock_expired(QEMUClockType type)
207 {
208 return timerlist_expired(
209 main_loop_tlg.tl[type]);
210 }
211
212 /*
213 * As above, but return -1 for no deadline, and do not cap to 2^32
214 * as we know the result is always positive.
215 */
216
217 int64_t timerlist_deadline_ns(QEMUTimerList *timer_list)
218 {
219 int64_t delta;
220 int64_t expire_time;
221
222 if (!atomic_read(&timer_list->active_timers)) {
223 return -1;
224 }
225
226 if (!timer_list->clock->enabled) {
227 return -1;
228 }
229
230 /* The active timers list may be modified before the caller uses our return
231 * value but ->notify_cb() is called when the deadline changes. Therefore
232 * the caller should notice the change and there is no race condition.
233 */
234 qemu_mutex_lock(&timer_list->active_timers_lock);
235 if (!timer_list->active_timers) {
236 qemu_mutex_unlock(&timer_list->active_timers_lock);
237 return -1;
238 }
239 expire_time = timer_list->active_timers->expire_time;
240 qemu_mutex_unlock(&timer_list->active_timers_lock);
241
242 delta = expire_time - qemu_clock_get_ns(timer_list->clock->type);
243
244 if (delta <= 0) {
245 return 0;
246 }
247
248 return delta;
249 }
250
251 /* Calculate the soonest deadline across all timerlists attached
252 * to the clock. This is used for the icount timeout so we
253 * ignore whether or not the clock should be used in deadline
254 * calculations.
255 */
256 int64_t qemu_clock_deadline_ns_all(QEMUClockType type)
257 {
258 int64_t deadline = -1;
259 QEMUTimerList *timer_list;
260 QEMUClock *clock = qemu_clock_ptr(type);
261 QLIST_FOREACH(timer_list, &clock->timerlists, list) {
262 deadline = qemu_soonest_timeout(deadline,
263 timerlist_deadline_ns(timer_list));
264 }
265 return deadline;
266 }
267
268 QEMUClockType timerlist_get_clock(QEMUTimerList *timer_list)
269 {
270 return timer_list->clock->type;
271 }
272
273 QEMUTimerList *qemu_clock_get_main_loop_timerlist(QEMUClockType type)
274 {
275 return main_loop_tlg.tl[type];
276 }
277
278 void timerlist_notify(QEMUTimerList *timer_list)
279 {
280 if (timer_list->notify_cb) {
281 timer_list->notify_cb(timer_list->notify_opaque, timer_list->clock->type);
282 } else {
283 qemu_notify_event();
284 }
285 }
286
287 /* Transition function to convert a nanosecond timeout to ms
288 * This is used where a system does not support ppoll
289 */
290 int qemu_timeout_ns_to_ms(int64_t ns)
291 {
292 int64_t ms;
293 if (ns < 0) {
294 return -1;
295 }
296
297 if (!ns) {
298 return 0;
299 }
300
301 /* Always round up, because it's better to wait too long than to wait too
302 * little and effectively busy-wait
303 */
304 ms = DIV_ROUND_UP(ns, SCALE_MS);
305
306 /* To avoid overflow problems, limit this to 2^31, i.e. approx 25 days */
307 if (ms > (int64_t) INT32_MAX) {
308 ms = INT32_MAX;
309 }
310
311 return (int) ms;
312 }
313
314
315 /* qemu implementation of g_poll which uses a nanosecond timeout but is
316 * otherwise identical to g_poll
317 */
318 int qemu_poll_ns(GPollFD *fds, guint nfds, int64_t timeout)
319 {
320 #ifdef CONFIG_PPOLL
321 if (timeout < 0) {
322 return ppoll((struct pollfd *)fds, nfds, NULL, NULL);
323 } else {
324 struct timespec ts;
325 int64_t tvsec = timeout / 1000000000LL;
326 /* Avoid possibly overflowing and specifying a negative number of
327 * seconds, which would turn a very long timeout into a busy-wait.
328 */
329 if (tvsec > (int64_t)INT32_MAX) {
330 tvsec = INT32_MAX;
331 }
332 ts.tv_sec = tvsec;
333 ts.tv_nsec = timeout % 1000000000LL;
334 return ppoll((struct pollfd *)fds, nfds, &ts, NULL);
335 }
336 #else
337 return g_poll(fds, nfds, qemu_timeout_ns_to_ms(timeout));
338 #endif
339 }
340
341
342 void timer_init_tl(QEMUTimer *ts,
343 QEMUTimerList *timer_list, int scale,
344 QEMUTimerCB *cb, void *opaque)
345 {
346 ts->timer_list = timer_list;
347 ts->cb = cb;
348 ts->opaque = opaque;
349 ts->scale = scale;
350 ts->expire_time = -1;
351 }
352
353 void timer_deinit(QEMUTimer *ts)
354 {
355 assert(ts->expire_time == -1);
356 ts->timer_list = NULL;
357 }
358
359 static void timer_del_locked(QEMUTimerList *timer_list, QEMUTimer *ts)
360 {
361 QEMUTimer **pt, *t;
362
363 ts->expire_time = -1;
364 pt = &timer_list->active_timers;
365 for(;;) {
366 t = *pt;
367 if (!t)
368 break;
369 if (t == ts) {
370 atomic_set(pt, t->next);
371 break;
372 }
373 pt = &t->next;
374 }
375 }
376
377 static bool timer_mod_ns_locked(QEMUTimerList *timer_list,
378 QEMUTimer *ts, int64_t expire_time)
379 {
380 QEMUTimer **pt, *t;
381
382 /* add the timer in the sorted list */
383 pt = &timer_list->active_timers;
384 for (;;) {
385 t = *pt;
386 if (!timer_expired_ns(t, expire_time)) {
387 break;
388 }
389 pt = &t->next;
390 }
391 ts->expire_time = MAX(expire_time, 0);
392 ts->next = *pt;
393 atomic_set(pt, ts);
394
395 return pt == &timer_list->active_timers;
396 }
397
398 static void timerlist_rearm(QEMUTimerList *timer_list)
399 {
400 /* Interrupt execution to force deadline recalculation. */
401 if (timer_list->clock->type == QEMU_CLOCK_VIRTUAL) {
402 qemu_start_warp_timer();
403 }
404 timerlist_notify(timer_list);
405 }
406
407 /* stop a timer, but do not dealloc it */
408 void timer_del(QEMUTimer *ts)
409 {
410 QEMUTimerList *timer_list = ts->timer_list;
411
412 if (timer_list) {
413 qemu_mutex_lock(&timer_list->active_timers_lock);
414 timer_del_locked(timer_list, ts);
415 qemu_mutex_unlock(&timer_list->active_timers_lock);
416 }
417 }
418
419 /* modify the current timer so that it will be fired when current_time
420 >= expire_time. The corresponding callback will be called. */
421 void timer_mod_ns(QEMUTimer *ts, int64_t expire_time)
422 {
423 QEMUTimerList *timer_list = ts->timer_list;
424 bool rearm;
425
426 qemu_mutex_lock(&timer_list->active_timers_lock);
427 timer_del_locked(timer_list, ts);
428 rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
429 qemu_mutex_unlock(&timer_list->active_timers_lock);
430
431 if (rearm) {
432 timerlist_rearm(timer_list);
433 }
434 }
435
436 /* modify the current timer so that it will be fired when current_time
437 >= expire_time or the current deadline, whichever comes earlier.
438 The corresponding callback will be called. */
439 void timer_mod_anticipate_ns(QEMUTimer *ts, int64_t expire_time)
440 {
441 QEMUTimerList *timer_list = ts->timer_list;
442 bool rearm;
443
444 qemu_mutex_lock(&timer_list->active_timers_lock);
445 if (ts->expire_time == -1 || ts->expire_time > expire_time) {
446 if (ts->expire_time != -1) {
447 timer_del_locked(timer_list, ts);
448 }
449 rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
450 } else {
451 rearm = false;
452 }
453 qemu_mutex_unlock(&timer_list->active_timers_lock);
454
455 if (rearm) {
456 timerlist_rearm(timer_list);
457 }
458 }
459
460 void timer_mod(QEMUTimer *ts, int64_t expire_time)
461 {
462 timer_mod_ns(ts, expire_time * ts->scale);
463 }
464
465 void timer_mod_anticipate(QEMUTimer *ts, int64_t expire_time)
466 {
467 timer_mod_anticipate_ns(ts, expire_time * ts->scale);
468 }
469
470 bool timer_pending(QEMUTimer *ts)
471 {
472 return ts->expire_time >= 0;
473 }
474
475 bool timer_expired(QEMUTimer *timer_head, int64_t current_time)
476 {
477 return timer_expired_ns(timer_head, current_time * timer_head->scale);
478 }
479
480 bool timerlist_run_timers(QEMUTimerList *timer_list)
481 {
482 QEMUTimer *ts;
483 int64_t current_time;
484 bool progress = false;
485 QEMUTimerCB *cb;
486 void *opaque;
487
488 if (!atomic_read(&timer_list->active_timers)) {
489 return false;
490 }
491
492 qemu_event_reset(&timer_list->timers_done_ev);
493 if (!timer_list->clock->enabled) {
494 goto out;
495 }
496
497 switch (timer_list->clock->type) {
498 case QEMU_CLOCK_REALTIME:
499 break;
500 default:
501 case QEMU_CLOCK_VIRTUAL:
502 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL)) {
503 goto out;
504 }
505 break;
506 case QEMU_CLOCK_HOST:
507 if (!replay_checkpoint(CHECKPOINT_CLOCK_HOST)) {
508 goto out;
509 }
510 break;
511 case QEMU_CLOCK_VIRTUAL_RT:
512 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL_RT)) {
513 goto out;
514 }
515 break;
516 }
517
518 current_time = qemu_clock_get_ns(timer_list->clock->type);
519 for(;;) {
520 qemu_mutex_lock(&timer_list->active_timers_lock);
521 ts = timer_list->active_timers;
522 if (!timer_expired_ns(ts, current_time)) {
523 qemu_mutex_unlock(&timer_list->active_timers_lock);
524 break;
525 }
526
527 /* remove timer from the list before calling the callback */
528 timer_list->active_timers = ts->next;
529 ts->next = NULL;
530 ts->expire_time = -1;
531 cb = ts->cb;
532 opaque = ts->opaque;
533 qemu_mutex_unlock(&timer_list->active_timers_lock);
534
535 /* run the callback (the timer list can be modified) */
536 cb(opaque);
537 progress = true;
538 }
539
540 out:
541 qemu_event_set(&timer_list->timers_done_ev);
542 return progress;
543 }
544
545 bool qemu_clock_run_timers(QEMUClockType type)
546 {
547 return timerlist_run_timers(main_loop_tlg.tl[type]);
548 }
549
550 void timerlistgroup_init(QEMUTimerListGroup *tlg,
551 QEMUTimerListNotifyCB *cb, void *opaque)
552 {
553 QEMUClockType type;
554 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
555 tlg->tl[type] = timerlist_new(type, cb, opaque);
556 }
557 }
558
559 void timerlistgroup_deinit(QEMUTimerListGroup *tlg)
560 {
561 QEMUClockType type;
562 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
563 timerlist_free(tlg->tl[type]);
564 }
565 }
566
567 bool timerlistgroup_run_timers(QEMUTimerListGroup *tlg)
568 {
569 QEMUClockType type;
570 bool progress = false;
571 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
572 progress |= timerlist_run_timers(tlg->tl[type]);
573 }
574 return progress;
575 }
576
577 int64_t timerlistgroup_deadline_ns(QEMUTimerListGroup *tlg)
578 {
579 int64_t deadline = -1;
580 QEMUClockType type;
581 bool play = replay_mode == REPLAY_MODE_PLAY;
582 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
583 if (qemu_clock_use_for_deadline(type)) {
584 if (!play || type == QEMU_CLOCK_REALTIME) {
585 deadline = qemu_soonest_timeout(deadline,
586 timerlist_deadline_ns(tlg->tl[type]));
587 } else {
588 /* Read clock from the replay file and
589 do not calculate the deadline, based on virtual clock. */
590 qemu_clock_get_ns(type);
591 }
592 }
593 }
594 return deadline;
595 }
596
597 int64_t qemu_clock_get_ns(QEMUClockType type)
598 {
599 int64_t now, last;
600 QEMUClock *clock = qemu_clock_ptr(type);
601
602 switch (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 now = REPLAY_CLOCK(REPLAY_CLOCK_HOST, get_clock_realtime());
614 last = clock->last;
615 clock->last = now;
616 if (now < last || now > (last + get_max_clock_jump())) {
617 notifier_list_notify(&clock->reset_notifiers, &now);
618 }
619 return now;
620 case QEMU_CLOCK_VIRTUAL_RT:
621 return REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT, cpu_get_clock());
622 }
623 }
624
625 void qemu_clock_register_reset_notifier(QEMUClockType type,
626 Notifier *notifier)
627 {
628 QEMUClock *clock = qemu_clock_ptr(type);
629 notifier_list_add(&clock->reset_notifiers, notifier);
630 }
631
632 void qemu_clock_unregister_reset_notifier(QEMUClockType type,
633 Notifier *notifier)
634 {
635 notifier_remove(notifier);
636 }
637
638 void init_clocks(QEMUTimerListNotifyCB *notify_cb)
639 {
640 QEMUClockType type;
641 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
642 qemu_clock_init(type, notify_cb);
643 }
644
645 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
646 prctl(PR_SET_TIMERSLACK, 1, 0, 0, 0);
647 #endif
648 }
649
650 uint64_t timer_expire_time_ns(QEMUTimer *ts)
651 {
652 return timer_pending(ts) ? ts->expire_time : -1;
653 }
654
655 bool qemu_clock_run_all_timers(void)
656 {
657 bool progress = false;
658 QEMUClockType type;
659
660 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
661 if (qemu_clock_use_for_deadline(type)) {
662 progress |= qemu_clock_run_timers(type);
663 }
664 }
665
666 return progress;
667 }