]> git.proxmox.com Git - mirror_qemu.git/blob - util/qemu-timer.c
Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging
[mirror_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/cpus.h"
30
31 #ifdef CONFIG_POSIX
32 #include <pthread.h>
33 #endif
34
35 #ifdef CONFIG_PPOLL
36 #include <poll.h>
37 #endif
38
39 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
40 #include <sys/prctl.h>
41 #endif
42
43 /***********************************************************/
44 /* timers */
45
46 typedef struct QEMUClock {
47 /* We rely on BQL to protect the timerlists */
48 QLIST_HEAD(, QEMUTimerList) timerlists;
49
50 QEMUClockType type;
51 bool enabled;
52 } QEMUClock;
53
54 QEMUTimerListGroup main_loop_tlg;
55 static QEMUClock qemu_clocks[QEMU_CLOCK_MAX];
56
57 /* A QEMUTimerList is a list of timers attached to a clock. More
58 * than one QEMUTimerList can be attached to each clock, for instance
59 * used by different AioContexts / threads. Each clock also has
60 * a list of the QEMUTimerLists associated with it, in order that
61 * reenabling the clock can call all the notifiers.
62 */
63
64 struct QEMUTimerList {
65 QEMUClock *clock;
66 QemuMutex active_timers_lock;
67 QEMUTimer *active_timers;
68 QLIST_ENTRY(QEMUTimerList) list;
69 QEMUTimerListNotifyCB *notify_cb;
70 void *notify_opaque;
71
72 /* lightweight method to mark the end of timerlist's running */
73 QemuEvent timers_done_ev;
74 };
75
76 /**
77 * qemu_clock_ptr:
78 * @type: type of clock
79 *
80 * Translate a clock type into a pointer to QEMUClock object.
81 *
82 * Returns: a pointer to the QEMUClock object
83 */
84 static inline QEMUClock *qemu_clock_ptr(QEMUClockType type)
85 {
86 return &qemu_clocks[type];
87 }
88
89 static bool timer_expired_ns(QEMUTimer *timer_head, int64_t current_time)
90 {
91 return timer_head && (timer_head->expire_time <= current_time);
92 }
93
94 QEMUTimerList *timerlist_new(QEMUClockType type,
95 QEMUTimerListNotifyCB *cb,
96 void *opaque)
97 {
98 QEMUTimerList *timer_list;
99 QEMUClock *clock = qemu_clock_ptr(type);
100
101 timer_list = g_malloc0(sizeof(QEMUTimerList));
102 qemu_event_init(&timer_list->timers_done_ev, true);
103 timer_list->clock = clock;
104 timer_list->notify_cb = cb;
105 timer_list->notify_opaque = opaque;
106 qemu_mutex_init(&timer_list->active_timers_lock);
107 QLIST_INSERT_HEAD(&clock->timerlists, timer_list, list);
108 return timer_list;
109 }
110
111 void timerlist_free(QEMUTimerList *timer_list)
112 {
113 assert(!timerlist_has_timers(timer_list));
114 if (timer_list->clock) {
115 QLIST_REMOVE(timer_list, list);
116 }
117 qemu_mutex_destroy(&timer_list->active_timers_lock);
118 g_free(timer_list);
119 }
120
121 static void qemu_clock_init(QEMUClockType type, QEMUTimerListNotifyCB *notify_cb)
122 {
123 QEMUClock *clock = qemu_clock_ptr(type);
124
125 /* Assert that the clock of type TYPE has not been initialized yet. */
126 assert(main_loop_tlg.tl[type] == NULL);
127
128 clock->type = type;
129 clock->enabled = (type == QEMU_CLOCK_VIRTUAL ? false : true);
130 QLIST_INIT(&clock->timerlists);
131 main_loop_tlg.tl[type] = timerlist_new(type, notify_cb, NULL);
132 }
133
134 bool qemu_clock_use_for_deadline(QEMUClockType type)
135 {
136 return !(use_icount && (type == QEMU_CLOCK_VIRTUAL));
137 }
138
139 void qemu_clock_notify(QEMUClockType type)
140 {
141 QEMUTimerList *timer_list;
142 QEMUClock *clock = qemu_clock_ptr(type);
143 QLIST_FOREACH(timer_list, &clock->timerlists, list) {
144 timerlist_notify(timer_list);
145 }
146 }
147
148 /* Disabling the clock will wait for related timerlists to stop
149 * executing qemu_run_timers. Thus, this functions should not
150 * be used from the callback of a timer that is based on @clock.
151 * Doing so would cause a deadlock.
152 *
153 * Caller should hold BQL.
154 */
155 void qemu_clock_enable(QEMUClockType type, bool enabled)
156 {
157 QEMUClock *clock = qemu_clock_ptr(type);
158 QEMUTimerList *tl;
159 bool old = clock->enabled;
160 clock->enabled = enabled;
161 if (enabled && !old) {
162 qemu_clock_notify(type);
163 } else if (!enabled && old) {
164 QLIST_FOREACH(tl, &clock->timerlists, list) {
165 qemu_event_wait(&tl->timers_done_ev);
166 }
167 }
168 }
169
170 bool timerlist_has_timers(QEMUTimerList *timer_list)
171 {
172 return !!atomic_read(&timer_list->active_timers);
173 }
174
175 bool qemu_clock_has_timers(QEMUClockType type)
176 {
177 return timerlist_has_timers(
178 main_loop_tlg.tl[type]);
179 }
180
181 bool timerlist_expired(QEMUTimerList *timer_list)
182 {
183 int64_t expire_time;
184
185 if (!atomic_read(&timer_list->active_timers)) {
186 return false;
187 }
188
189 qemu_mutex_lock(&timer_list->active_timers_lock);
190 if (!timer_list->active_timers) {
191 qemu_mutex_unlock(&timer_list->active_timers_lock);
192 return false;
193 }
194 expire_time = timer_list->active_timers->expire_time;
195 qemu_mutex_unlock(&timer_list->active_timers_lock);
196
197 return expire_time <= qemu_clock_get_ns(timer_list->clock->type);
198 }
199
200 bool qemu_clock_expired(QEMUClockType type)
201 {
202 return timerlist_expired(
203 main_loop_tlg.tl[type]);
204 }
205
206 /*
207 * As above, but return -1 for no deadline, and do not cap to 2^32
208 * as we know the result is always positive.
209 */
210
211 int64_t timerlist_deadline_ns(QEMUTimerList *timer_list)
212 {
213 int64_t delta;
214 int64_t expire_time;
215
216 if (!atomic_read(&timer_list->active_timers)) {
217 return -1;
218 }
219
220 if (!timer_list->clock->enabled) {
221 return -1;
222 }
223
224 /* The active timers list may be modified before the caller uses our return
225 * value but ->notify_cb() is called when the deadline changes. Therefore
226 * the caller should notice the change and there is no race condition.
227 */
228 qemu_mutex_lock(&timer_list->active_timers_lock);
229 if (!timer_list->active_timers) {
230 qemu_mutex_unlock(&timer_list->active_timers_lock);
231 return -1;
232 }
233 expire_time = timer_list->active_timers->expire_time;
234 qemu_mutex_unlock(&timer_list->active_timers_lock);
235
236 delta = expire_time - qemu_clock_get_ns(timer_list->clock->type);
237
238 if (delta <= 0) {
239 return 0;
240 }
241
242 return delta;
243 }
244
245 /* Calculate the soonest deadline across all timerlists attached
246 * to the clock. This is used for the icount timeout so we
247 * ignore whether or not the clock should be used in deadline
248 * calculations.
249 */
250 int64_t qemu_clock_deadline_ns_all(QEMUClockType type, int attr_mask)
251 {
252 int64_t deadline = -1;
253 int64_t delta;
254 int64_t expire_time;
255 QEMUTimer *ts;
256 QEMUTimerList *timer_list;
257 QEMUClock *clock = qemu_clock_ptr(type);
258
259 if (!clock->enabled) {
260 return -1;
261 }
262
263 QLIST_FOREACH(timer_list, &clock->timerlists, list) {
264 qemu_mutex_lock(&timer_list->active_timers_lock);
265 ts = timer_list->active_timers;
266 /* Skip all external timers */
267 while (ts && (ts->attributes & ~attr_mask)) {
268 ts = ts->next;
269 }
270 if (!ts) {
271 qemu_mutex_unlock(&timer_list->active_timers_lock);
272 continue;
273 }
274 expire_time = ts->expire_time;
275 qemu_mutex_unlock(&timer_list->active_timers_lock);
276
277 delta = expire_time - qemu_clock_get_ns(type);
278 if (delta <= 0) {
279 delta = 0;
280 }
281 deadline = qemu_soonest_timeout(deadline, delta);
282 }
283 return deadline;
284 }
285
286 QEMUClockType timerlist_get_clock(QEMUTimerList *timer_list)
287 {
288 return timer_list->clock->type;
289 }
290
291 QEMUTimerList *qemu_clock_get_main_loop_timerlist(QEMUClockType type)
292 {
293 return main_loop_tlg.tl[type];
294 }
295
296 void timerlist_notify(QEMUTimerList *timer_list)
297 {
298 if (timer_list->notify_cb) {
299 timer_list->notify_cb(timer_list->notify_opaque, timer_list->clock->type);
300 } else {
301 qemu_notify_event();
302 }
303 }
304
305 /* Transition function to convert a nanosecond timeout to ms
306 * This is used where a system does not support ppoll
307 */
308 int qemu_timeout_ns_to_ms(int64_t ns)
309 {
310 int64_t ms;
311 if (ns < 0) {
312 return -1;
313 }
314
315 if (!ns) {
316 return 0;
317 }
318
319 /* Always round up, because it's better to wait too long than to wait too
320 * little and effectively busy-wait
321 */
322 ms = DIV_ROUND_UP(ns, SCALE_MS);
323
324 /* To avoid overflow problems, limit this to 2^31, i.e. approx 25 days */
325 return MIN(ms, INT32_MAX);
326 }
327
328
329 /* qemu implementation of g_poll which uses a nanosecond timeout but is
330 * otherwise identical to g_poll
331 */
332 int qemu_poll_ns(GPollFD *fds, guint nfds, int64_t timeout)
333 {
334 #ifdef CONFIG_PPOLL
335 if (timeout < 0) {
336 return ppoll((struct pollfd *)fds, nfds, NULL, NULL);
337 } else {
338 struct timespec ts;
339 int64_t tvsec = timeout / 1000000000LL;
340 /* Avoid possibly overflowing and specifying a negative number of
341 * seconds, which would turn a very long timeout into a busy-wait.
342 */
343 if (tvsec > (int64_t)INT32_MAX) {
344 tvsec = INT32_MAX;
345 }
346 ts.tv_sec = tvsec;
347 ts.tv_nsec = timeout % 1000000000LL;
348 return ppoll((struct pollfd *)fds, nfds, &ts, NULL);
349 }
350 #else
351 return g_poll(fds, nfds, qemu_timeout_ns_to_ms(timeout));
352 #endif
353 }
354
355
356 void timer_init_full(QEMUTimer *ts,
357 QEMUTimerListGroup *timer_list_group, QEMUClockType type,
358 int scale, int attributes,
359 QEMUTimerCB *cb, void *opaque)
360 {
361 if (!timer_list_group) {
362 timer_list_group = &main_loop_tlg;
363 }
364 ts->timer_list = timer_list_group->tl[type];
365 ts->cb = cb;
366 ts->opaque = opaque;
367 ts->scale = scale;
368 ts->attributes = attributes;
369 ts->expire_time = -1;
370 }
371
372 void timer_deinit(QEMUTimer *ts)
373 {
374 assert(ts->expire_time == -1);
375 ts->timer_list = NULL;
376 }
377
378 static void timer_del_locked(QEMUTimerList *timer_list, QEMUTimer *ts)
379 {
380 QEMUTimer **pt, *t;
381
382 ts->expire_time = -1;
383 pt = &timer_list->active_timers;
384 for(;;) {
385 t = *pt;
386 if (!t)
387 break;
388 if (t == ts) {
389 atomic_set(pt, t->next);
390 break;
391 }
392 pt = &t->next;
393 }
394 }
395
396 static bool timer_mod_ns_locked(QEMUTimerList *timer_list,
397 QEMUTimer *ts, int64_t expire_time)
398 {
399 QEMUTimer **pt, *t;
400
401 /* add the timer in the sorted list */
402 pt = &timer_list->active_timers;
403 for (;;) {
404 t = *pt;
405 if (!timer_expired_ns(t, expire_time)) {
406 break;
407 }
408 pt = &t->next;
409 }
410 ts->expire_time = MAX(expire_time, 0);
411 ts->next = *pt;
412 atomic_set(pt, ts);
413
414 return pt == &timer_list->active_timers;
415 }
416
417 static void timerlist_rearm(QEMUTimerList *timer_list)
418 {
419 /* Interrupt execution to force deadline recalculation. */
420 if (timer_list->clock->type == QEMU_CLOCK_VIRTUAL) {
421 qemu_start_warp_timer();
422 }
423 timerlist_notify(timer_list);
424 }
425
426 /* stop a timer, but do not dealloc it */
427 void timer_del(QEMUTimer *ts)
428 {
429 QEMUTimerList *timer_list = ts->timer_list;
430
431 if (timer_list) {
432 qemu_mutex_lock(&timer_list->active_timers_lock);
433 timer_del_locked(timer_list, ts);
434 qemu_mutex_unlock(&timer_list->active_timers_lock);
435 }
436 }
437
438 /* modify the current timer so that it will be fired when current_time
439 >= expire_time. The corresponding callback will be called. */
440 void timer_mod_ns(QEMUTimer *ts, int64_t expire_time)
441 {
442 QEMUTimerList *timer_list = ts->timer_list;
443 bool rearm;
444
445 qemu_mutex_lock(&timer_list->active_timers_lock);
446 timer_del_locked(timer_list, ts);
447 rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
448 qemu_mutex_unlock(&timer_list->active_timers_lock);
449
450 if (rearm) {
451 timerlist_rearm(timer_list);
452 }
453 }
454
455 /* modify the current timer so that it will be fired when current_time
456 >= expire_time or the current deadline, whichever comes earlier.
457 The corresponding callback will be called. */
458 void timer_mod_anticipate_ns(QEMUTimer *ts, int64_t expire_time)
459 {
460 QEMUTimerList *timer_list = ts->timer_list;
461 bool rearm;
462
463 qemu_mutex_lock(&timer_list->active_timers_lock);
464 if (ts->expire_time == -1 || ts->expire_time > expire_time) {
465 if (ts->expire_time != -1) {
466 timer_del_locked(timer_list, ts);
467 }
468 rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
469 } else {
470 rearm = false;
471 }
472 qemu_mutex_unlock(&timer_list->active_timers_lock);
473
474 if (rearm) {
475 timerlist_rearm(timer_list);
476 }
477 }
478
479 void timer_mod(QEMUTimer *ts, int64_t expire_time)
480 {
481 timer_mod_ns(ts, expire_time * ts->scale);
482 }
483
484 void timer_mod_anticipate(QEMUTimer *ts, int64_t expire_time)
485 {
486 timer_mod_anticipate_ns(ts, expire_time * ts->scale);
487 }
488
489 bool timer_pending(QEMUTimer *ts)
490 {
491 return ts->expire_time >= 0;
492 }
493
494 bool timer_expired(QEMUTimer *timer_head, int64_t current_time)
495 {
496 return timer_expired_ns(timer_head, current_time * timer_head->scale);
497 }
498
499 bool timerlist_run_timers(QEMUTimerList *timer_list)
500 {
501 QEMUTimer *ts;
502 int64_t current_time;
503 bool progress = false;
504 QEMUTimerCB *cb;
505 void *opaque;
506 bool need_replay_checkpoint = false;
507
508 if (!atomic_read(&timer_list->active_timers)) {
509 return false;
510 }
511
512 qemu_event_reset(&timer_list->timers_done_ev);
513 if (!timer_list->clock->enabled) {
514 goto out;
515 }
516
517 switch (timer_list->clock->type) {
518 case QEMU_CLOCK_REALTIME:
519 break;
520 default:
521 case QEMU_CLOCK_VIRTUAL:
522 if (replay_mode != REPLAY_MODE_NONE) {
523 /* Checkpoint for virtual clock is redundant in cases where
524 * it's being triggered with only non-EXTERNAL timers, because
525 * these timers don't change guest state directly.
526 * Since it has conditional dependence on specific timers, it is
527 * subject to race conditions and requires special handling.
528 * See below.
529 */
530 need_replay_checkpoint = true;
531 }
532 break;
533 case QEMU_CLOCK_HOST:
534 if (!replay_checkpoint(CHECKPOINT_CLOCK_HOST)) {
535 goto out;
536 }
537 break;
538 case QEMU_CLOCK_VIRTUAL_RT:
539 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL_RT)) {
540 goto out;
541 }
542 break;
543 }
544
545 /*
546 * Extract expired timers from active timers list and and process them.
547 *
548 * In rr mode we need "filtered" checkpointing for virtual clock. The
549 * checkpoint must be recorded/replayed before processing any non-EXTERNAL timer,
550 * and that must only be done once since the clock value stays the same. Because
551 * non-EXTERNAL timers may appear in the timers list while it being processed,
552 * the checkpoint can be issued at a time until no timers are left and we are
553 * done".
554 */
555 current_time = qemu_clock_get_ns(timer_list->clock->type);
556 qemu_mutex_lock(&timer_list->active_timers_lock);
557 while ((ts = timer_list->active_timers)) {
558 if (!timer_expired_ns(ts, current_time)) {
559 /* No expired timers left. The checkpoint can be skipped
560 * if no timers fired or they were all external.
561 */
562 break;
563 }
564 if (need_replay_checkpoint
565 && !(ts->attributes & QEMU_TIMER_ATTR_EXTERNAL)) {
566 /* once we got here, checkpoint clock only once */
567 need_replay_checkpoint = false;
568 qemu_mutex_unlock(&timer_list->active_timers_lock);
569 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL)) {
570 goto out;
571 }
572 qemu_mutex_lock(&timer_list->active_timers_lock);
573 /* The lock was released; start over again in case the list was
574 * modified.
575 */
576 continue;
577 }
578
579 /* remove timer from the list before calling the callback */
580 timer_list->active_timers = ts->next;
581 ts->next = NULL;
582 ts->expire_time = -1;
583 cb = ts->cb;
584 opaque = ts->opaque;
585
586 /* run the callback (the timer list can be modified) */
587 qemu_mutex_unlock(&timer_list->active_timers_lock);
588 cb(opaque);
589 qemu_mutex_lock(&timer_list->active_timers_lock);
590
591 progress = true;
592 }
593 qemu_mutex_unlock(&timer_list->active_timers_lock);
594
595 out:
596 qemu_event_set(&timer_list->timers_done_ev);
597 return progress;
598 }
599
600 bool qemu_clock_run_timers(QEMUClockType type)
601 {
602 return timerlist_run_timers(main_loop_tlg.tl[type]);
603 }
604
605 void timerlistgroup_init(QEMUTimerListGroup *tlg,
606 QEMUTimerListNotifyCB *cb, void *opaque)
607 {
608 QEMUClockType type;
609 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
610 tlg->tl[type] = timerlist_new(type, cb, opaque);
611 }
612 }
613
614 void timerlistgroup_deinit(QEMUTimerListGroup *tlg)
615 {
616 QEMUClockType type;
617 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
618 timerlist_free(tlg->tl[type]);
619 }
620 }
621
622 bool timerlistgroup_run_timers(QEMUTimerListGroup *tlg)
623 {
624 QEMUClockType type;
625 bool progress = false;
626 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
627 progress |= timerlist_run_timers(tlg->tl[type]);
628 }
629 return progress;
630 }
631
632 int64_t timerlistgroup_deadline_ns(QEMUTimerListGroup *tlg)
633 {
634 int64_t deadline = -1;
635 QEMUClockType type;
636 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
637 if (qemu_clock_use_for_deadline(type)) {
638 deadline = qemu_soonest_timeout(deadline,
639 timerlist_deadline_ns(tlg->tl[type]));
640 }
641 }
642 return deadline;
643 }
644
645 int64_t qemu_clock_get_ns(QEMUClockType type)
646 {
647 switch (type) {
648 case QEMU_CLOCK_REALTIME:
649 return get_clock();
650 default:
651 case QEMU_CLOCK_VIRTUAL:
652 if (use_icount) {
653 return cpu_get_icount();
654 } else {
655 return cpu_get_clock();
656 }
657 case QEMU_CLOCK_HOST:
658 return REPLAY_CLOCK(REPLAY_CLOCK_HOST, get_clock_realtime());
659 case QEMU_CLOCK_VIRTUAL_RT:
660 return REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT, cpu_get_clock());
661 }
662 }
663
664 void init_clocks(QEMUTimerListNotifyCB *notify_cb)
665 {
666 QEMUClockType type;
667 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
668 qemu_clock_init(type, notify_cb);
669 }
670
671 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
672 prctl(PR_SET_TIMERSLACK, 1, 0, 0, 0);
673 #endif
674 }
675
676 uint64_t timer_expire_time_ns(QEMUTimer *ts)
677 {
678 return timer_pending(ts) ? ts->expire_time : -1;
679 }
680
681 bool qemu_clock_run_all_timers(void)
682 {
683 bool progress = false;
684 QEMUClockType type;
685
686 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
687 if (qemu_clock_use_for_deadline(type)) {
688 progress |= qemu_clock_run_timers(type);
689 }
690 }
691
692 return progress;
693 }