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
46typedef 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
54QEMUTimerListGroup main_loop_tlg;
55static 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
64struct 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 */
84static inline QEMUClock *qemu_clock_ptr(QEMUClockType type)
85{
86 return &qemu_clocks[type];
87}
88
89static bool timer_expired_ns(QEMUTimer *timer_head, int64_t current_time)
90{
91 return timer_head && (timer_head->expire_time <= current_time);
92}
93
94QEMUTimerList *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
111void 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
121static 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
134bool qemu_clock_use_for_deadline(QEMUClockType type)
135{
136 return !(use_icount && (type == QEMU_CLOCK_VIRTUAL));
137}
138
139void 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 */
155void 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
170bool timerlist_has_timers(QEMUTimerList *timer_list)
171{
172 return !!atomic_read(&timer_list->active_timers);
173}
174
175bool qemu_clock_has_timers(QEMUClockType type)
176{
177 return timerlist_has_timers(
178 main_loop_tlg.tl[type]);
179}
180
181bool 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
200bool 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
211int64_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 */
250int64_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
286QEMUClockType timerlist_get_clock(QEMUTimerList *timer_list)
287{
288 return timer_list->clock->type;
289}
290
291QEMUTimerList *qemu_clock_get_main_loop_timerlist(QEMUClockType type)
292{
293 return main_loop_tlg.tl[type];
294}
295
296void 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 */
308int 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 if (ms > (int64_t) INT32_MAX) {
326 ms = INT32_MAX;
327 }
328
329 return (int) ms;
330}
331
332
333/* qemu implementation of g_poll which uses a nanosecond timeout but is
334 * otherwise identical to g_poll
335 */
336int qemu_poll_ns(GPollFD *fds, guint nfds, int64_t timeout)
337{
338#ifdef CONFIG_PPOLL
339 if (timeout < 0) {
340 return ppoll((struct pollfd *)fds, nfds, NULL, NULL);
341 } else {
342 struct timespec ts;
343 int64_t tvsec = timeout / 1000000000LL;
344 /* Avoid possibly overflowing and specifying a negative number of
345 * seconds, which would turn a very long timeout into a busy-wait.
346 */
347 if (tvsec > (int64_t)INT32_MAX) {
348 tvsec = INT32_MAX;
349 }
350 ts.tv_sec = tvsec;
351 ts.tv_nsec = timeout % 1000000000LL;
352 return ppoll((struct pollfd *)fds, nfds, &ts, NULL);
353 }
354#else
355 return g_poll(fds, nfds, qemu_timeout_ns_to_ms(timeout));
356#endif
357}
358
359
360void timer_init_full(QEMUTimer *ts,
361 QEMUTimerListGroup *timer_list_group, QEMUClockType type,
362 int scale, int attributes,
363 QEMUTimerCB *cb, void *opaque)
364{
365 if (!timer_list_group) {
366 timer_list_group = &main_loop_tlg;
367 }
368 ts->timer_list = timer_list_group->tl[type];
369 ts->cb = cb;
370 ts->opaque = opaque;
371 ts->scale = scale;
372 ts->attributes = attributes;
373 ts->expire_time = -1;
374}
375
376void timer_deinit(QEMUTimer *ts)
377{
378 assert(ts->expire_time == -1);
379 ts->timer_list = NULL;
380}
381
382static void timer_del_locked(QEMUTimerList *timer_list, QEMUTimer *ts)
383{
384 QEMUTimer **pt, *t;
385
386 ts->expire_time = -1;
387 pt = &timer_list->active_timers;
388 for(;;) {
389 t = *pt;
390 if (!t)
391 break;
392 if (t == ts) {
393 atomic_set(pt, t->next);
394 break;
395 }
396 pt = &t->next;
397 }
398}
399
400static bool timer_mod_ns_locked(QEMUTimerList *timer_list,
401 QEMUTimer *ts, int64_t expire_time)
402{
403 QEMUTimer **pt, *t;
404
405 /* add the timer in the sorted list */
406 pt = &timer_list->active_timers;
407 for (;;) {
408 t = *pt;
409 if (!timer_expired_ns(t, expire_time)) {
410 break;
411 }
412 pt = &t->next;
413 }
414 ts->expire_time = MAX(expire_time, 0);
415 ts->next = *pt;
416 atomic_set(pt, ts);
417
418 return pt == &timer_list->active_timers;
419}
420
421static void timerlist_rearm(QEMUTimerList *timer_list)
422{
423 /* Interrupt execution to force deadline recalculation. */
424 if (timer_list->clock->type == QEMU_CLOCK_VIRTUAL) {
425 qemu_start_warp_timer();
426 }
427 timerlist_notify(timer_list);
428}
429
430/* stop a timer, but do not dealloc it */
431void timer_del(QEMUTimer *ts)
432{
433 QEMUTimerList *timer_list = ts->timer_list;
434
435 if (timer_list) {
436 qemu_mutex_lock(&timer_list->active_timers_lock);
437 timer_del_locked(timer_list, ts);
438 qemu_mutex_unlock(&timer_list->active_timers_lock);
439 }
440}
441
442/* modify the current timer so that it will be fired when current_time
443 >= expire_time. The corresponding callback will be called. */
444void timer_mod_ns(QEMUTimer *ts, int64_t expire_time)
445{
446 QEMUTimerList *timer_list = ts->timer_list;
447 bool rearm;
448
449 qemu_mutex_lock(&timer_list->active_timers_lock);
450 timer_del_locked(timer_list, ts);
451 rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
452 qemu_mutex_unlock(&timer_list->active_timers_lock);
453
454 if (rearm) {
455 timerlist_rearm(timer_list);
456 }
457}
458
459/* modify the current timer so that it will be fired when current_time
460 >= expire_time or the current deadline, whichever comes earlier.
461 The corresponding callback will be called. */
462void timer_mod_anticipate_ns(QEMUTimer *ts, int64_t expire_time)
463{
464 QEMUTimerList *timer_list = ts->timer_list;
465 bool rearm;
466
467 qemu_mutex_lock(&timer_list->active_timers_lock);
468 if (ts->expire_time == -1 || ts->expire_time > expire_time) {
469 if (ts->expire_time != -1) {
470 timer_del_locked(timer_list, ts);
471 }
472 rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
473 } else {
474 rearm = false;
475 }
476 qemu_mutex_unlock(&timer_list->active_timers_lock);
477
478 if (rearm) {
479 timerlist_rearm(timer_list);
480 }
481}
482
483void timer_mod(QEMUTimer *ts, int64_t expire_time)
484{
485 timer_mod_ns(ts, expire_time * ts->scale);
486}
487
488void timer_mod_anticipate(QEMUTimer *ts, int64_t expire_time)
489{
490 timer_mod_anticipate_ns(ts, expire_time * ts->scale);
491}
492
493bool timer_pending(QEMUTimer *ts)
494{
495 return ts->expire_time >= 0;
496}
497
498bool timer_expired(QEMUTimer *timer_head, int64_t current_time)
499{
500 return timer_expired_ns(timer_head, current_time * timer_head->scale);
501}
502
503bool timerlist_run_timers(QEMUTimerList *timer_list)
504{
505 QEMUTimer *ts;
506 int64_t current_time;
507 bool progress = false;
508 QEMUTimerCB *cb;
509 void *opaque;
510 bool need_replay_checkpoint = false;
511
512 if (!atomic_read(&timer_list->active_timers)) {
513 return false;
514 }
515
516 qemu_event_reset(&timer_list->timers_done_ev);
517 if (!timer_list->clock->enabled) {
518 goto out;
519 }
520
521 switch (timer_list->clock->type) {
522 case QEMU_CLOCK_REALTIME:
523 break;
524 default:
525 case QEMU_CLOCK_VIRTUAL:
526 if (replay_mode != REPLAY_MODE_NONE) {
527 /* Checkpoint for virtual clock is redundant in cases where
528 * it's being triggered with only non-EXTERNAL timers, because
529 * these timers don't change guest state directly.
530 * Since it has conditional dependence on specific timers, it is
531 * subject to race conditions and requires special handling.
532 * See below.
533 */
534 need_replay_checkpoint = true;
535 }
536 break;
537 case QEMU_CLOCK_HOST:
538 if (!replay_checkpoint(CHECKPOINT_CLOCK_HOST)) {
539 goto out;
540 }
541 break;
542 case QEMU_CLOCK_VIRTUAL_RT:
543 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL_RT)) {
544 goto out;
545 }
546 break;
547 }
548
549 /*
550 * Extract expired timers from active timers list and and process them.
551 *
552 * In rr mode we need "filtered" checkpointing for virtual clock. The
553 * checkpoint must be recorded/replayed before processing any non-EXTERNAL timer,
554 * and that must only be done once since the clock value stays the same. Because
555 * non-EXTERNAL timers may appear in the timers list while it being processed,
556 * the checkpoint can be issued at a time until no timers are left and we are
557 * done".
558 */
559 current_time = qemu_clock_get_ns(timer_list->clock->type);
560 qemu_mutex_lock(&timer_list->active_timers_lock);
561 while ((ts = timer_list->active_timers)) {
562 if (!timer_expired_ns(ts, current_time)) {
563 /* No expired timers left. The checkpoint can be skipped
564 * if no timers fired or they were all external.
565 */
566 break;
567 }
568 if (need_replay_checkpoint
569 && !(ts->attributes & QEMU_TIMER_ATTR_EXTERNAL)) {
570 /* once we got here, checkpoint clock only once */
571 need_replay_checkpoint = false;
572 qemu_mutex_unlock(&timer_list->active_timers_lock);
573 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL)) {
574 goto out;
575 }
576 qemu_mutex_lock(&timer_list->active_timers_lock);
577 /* The lock was released; start over again in case the list was
578 * modified.
579 */
580 continue;
581 }
582
583 /* remove timer from the list before calling the callback */
584 timer_list->active_timers = ts->next;
585 ts->next = NULL;
586 ts->expire_time = -1;
587 cb = ts->cb;
588 opaque = ts->opaque;
589
590 /* run the callback (the timer list can be modified) */
591 qemu_mutex_unlock(&timer_list->active_timers_lock);
592 cb(opaque);
593 qemu_mutex_lock(&timer_list->active_timers_lock);
594
595 progress = true;
596 }
597 qemu_mutex_unlock(&timer_list->active_timers_lock);
598
599out:
600 qemu_event_set(&timer_list->timers_done_ev);
601 return progress;
602}
603
604bool qemu_clock_run_timers(QEMUClockType type)
605{
606 return timerlist_run_timers(main_loop_tlg.tl[type]);
607}
608
609void timerlistgroup_init(QEMUTimerListGroup *tlg,
610 QEMUTimerListNotifyCB *cb, void *opaque)
611{
612 QEMUClockType type;
613 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
614 tlg->tl[type] = timerlist_new(type, cb, opaque);
615 }
616}
617
618void timerlistgroup_deinit(QEMUTimerListGroup *tlg)
619{
620 QEMUClockType type;
621 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
622 timerlist_free(tlg->tl[type]);
623 }
624}
625
626bool timerlistgroup_run_timers(QEMUTimerListGroup *tlg)
627{
628 QEMUClockType type;
629 bool progress = false;
630 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
631 progress |= timerlist_run_timers(tlg->tl[type]);
632 }
633 return progress;
634}
635
636int64_t timerlistgroup_deadline_ns(QEMUTimerListGroup *tlg)
637{
638 int64_t deadline = -1;
639 QEMUClockType type;
640 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
641 if (qemu_clock_use_for_deadline(type)) {
642 deadline = qemu_soonest_timeout(deadline,
643 timerlist_deadline_ns(tlg->tl[type]));
644 }
645 }
646 return deadline;
647}
648
649int64_t qemu_clock_get_ns(QEMUClockType type)
650{
651 switch (type) {
652 case QEMU_CLOCK_REALTIME:
653 return get_clock();
654 default:
655 case QEMU_CLOCK_VIRTUAL:
656 if (use_icount) {
657 return cpu_get_icount();
658 } else {
659 return cpu_get_clock();
660 }
661 case QEMU_CLOCK_HOST:
662 return REPLAY_CLOCK(REPLAY_CLOCK_HOST, get_clock_realtime());
663 case QEMU_CLOCK_VIRTUAL_RT:
664 return REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT, cpu_get_clock());
665 }
666}
667
668void init_clocks(QEMUTimerListNotifyCB *notify_cb)
669{
670 QEMUClockType type;
671 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
672 qemu_clock_init(type, notify_cb);
673 }
674
675#ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
676 prctl(PR_SET_TIMERSLACK, 1, 0, 0, 0);
677#endif
678}
679
680uint64_t timer_expire_time_ns(QEMUTimer *ts)
681{
682 return timer_pending(ts) ? ts->expire_time : -1;
683}
684
685bool qemu_clock_run_all_timers(void)
686{
687 bool progress = false;
688 QEMUClockType type;
689
690 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
691 if (qemu_clock_use_for_deadline(type)) {
692 progress |= qemu_clock_run_timers(type);
693 }
694 }
695
696 return progress;
697}
698