4 * Copyright (c) 2003-2008 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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
25 #include "qemu/osdep.h"
26 #include "qemu/main-loop.h"
27 #include "qemu/timer.h"
28 #include "qemu/lockable.h"
29 #include "sysemu/cpu-timers.h"
30 #include "sysemu/replay.h"
31 #include "sysemu/cpus.h"
32 #include "sysemu/qtest.h"
42 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
43 #include <sys/prctl.h>
46 /***********************************************************/
49 typedef struct QEMUClock {
50 /* We rely on BQL to protect the timerlists */
51 QLIST_HEAD(, QEMUTimerList) timerlists;
57 QEMUTimerListGroup main_loop_tlg;
58 static QEMUClock qemu_clocks[QEMU_CLOCK_MAX];
60 /* A QEMUTimerList is a list of timers attached to a clock. More
61 * than one QEMUTimerList can be attached to each clock, for instance
62 * used by different AioContexts / threads. Each clock also has
63 * a list of the QEMUTimerLists associated with it, in order that
64 * reenabling the clock can call all the notifiers.
67 struct QEMUTimerList {
69 QemuMutex active_timers_lock;
70 QEMUTimer *active_timers;
71 QLIST_ENTRY(QEMUTimerList) list;
72 QEMUTimerListNotifyCB *notify_cb;
75 /* lightweight method to mark the end of timerlist's running */
76 QemuEvent timers_done_ev;
81 * @type: type of clock
83 * Translate a clock type into a pointer to QEMUClock object.
85 * Returns: a pointer to the QEMUClock object
87 static inline QEMUClock *qemu_clock_ptr(QEMUClockType type)
89 return &qemu_clocks[type];
92 static bool timer_expired_ns(QEMUTimer *timer_head, int64_t current_time)
94 return timer_head && (timer_head->expire_time <= current_time);
97 QEMUTimerList *timerlist_new(QEMUClockType type,
98 QEMUTimerListNotifyCB *cb,
101 QEMUTimerList *timer_list;
102 QEMUClock *clock = qemu_clock_ptr(type);
104 timer_list = g_malloc0(sizeof(QEMUTimerList));
105 qemu_event_init(&timer_list->timers_done_ev, true);
106 timer_list->clock = clock;
107 timer_list->notify_cb = cb;
108 timer_list->notify_opaque = opaque;
109 qemu_mutex_init(&timer_list->active_timers_lock);
110 QLIST_INSERT_HEAD(&clock->timerlists, timer_list, list);
114 void timerlist_free(QEMUTimerList *timer_list)
116 assert(!timerlist_has_timers(timer_list));
117 if (timer_list->clock) {
118 QLIST_REMOVE(timer_list, list);
120 qemu_mutex_destroy(&timer_list->active_timers_lock);
124 static void qemu_clock_init(QEMUClockType type, QEMUTimerListNotifyCB *notify_cb)
126 QEMUClock *clock = qemu_clock_ptr(type);
128 /* Assert that the clock of type TYPE has not been initialized yet. */
129 assert(main_loop_tlg.tl[type] == NULL);
132 clock->enabled = (type == QEMU_CLOCK_VIRTUAL ? false : true);
133 QLIST_INIT(&clock->timerlists);
134 main_loop_tlg.tl[type] = timerlist_new(type, notify_cb, NULL);
137 bool qemu_clock_use_for_deadline(QEMUClockType type)
139 return !(icount_enabled() && (type == QEMU_CLOCK_VIRTUAL));
142 void qemu_clock_notify(QEMUClockType type)
144 QEMUTimerList *timer_list;
145 QEMUClock *clock = qemu_clock_ptr(type);
146 QLIST_FOREACH(timer_list, &clock->timerlists, list) {
147 timerlist_notify(timer_list);
151 /* Disabling the clock will wait for related timerlists to stop
152 * executing qemu_run_timers. Thus, this functions should not
153 * be used from the callback of a timer that is based on @clock.
154 * Doing so would cause a deadlock.
156 * Caller should hold BQL.
158 void qemu_clock_enable(QEMUClockType type, bool enabled)
160 QEMUClock *clock = qemu_clock_ptr(type);
162 bool old = clock->enabled;
163 clock->enabled = enabled;
164 if (enabled && !old) {
165 qemu_clock_notify(type);
166 } else if (!enabled && old) {
167 QLIST_FOREACH(tl, &clock->timerlists, list) {
168 qemu_event_wait(&tl->timers_done_ev);
173 bool timerlist_has_timers(QEMUTimerList *timer_list)
175 return !!qatomic_read(&timer_list->active_timers);
178 bool qemu_clock_has_timers(QEMUClockType type)
180 return timerlist_has_timers(
181 main_loop_tlg.tl[type]);
184 bool timerlist_expired(QEMUTimerList *timer_list)
188 if (!qatomic_read(&timer_list->active_timers)) {
192 WITH_QEMU_LOCK_GUARD(&timer_list->active_timers_lock) {
193 if (!timer_list->active_timers) {
196 expire_time = timer_list->active_timers->expire_time;
199 return expire_time <= qemu_clock_get_ns(timer_list->clock->type);
202 bool qemu_clock_expired(QEMUClockType type)
204 return timerlist_expired(
205 main_loop_tlg.tl[type]);
209 * As above, but return -1 for no deadline, and do not cap to 2^32
210 * as we know the result is always positive.
213 int64_t timerlist_deadline_ns(QEMUTimerList *timer_list)
218 if (!qatomic_read(&timer_list->active_timers)) {
222 if (!timer_list->clock->enabled) {
226 /* The active timers list may be modified before the caller uses our return
227 * value but ->notify_cb() is called when the deadline changes. Therefore
228 * the caller should notice the change and there is no race condition.
230 WITH_QEMU_LOCK_GUARD(&timer_list->active_timers_lock) {
231 if (!timer_list->active_timers) {
234 expire_time = timer_list->active_timers->expire_time;
237 delta = expire_time - qemu_clock_get_ns(timer_list->clock->type);
246 /* Calculate the soonest deadline across all timerlists attached
247 * to the clock. This is used for the icount timeout so we
248 * ignore whether or not the clock should be used in deadline
251 int64_t qemu_clock_deadline_ns_all(QEMUClockType type, int attr_mask)
253 int64_t deadline = -1;
257 QEMUTimerList *timer_list;
258 QEMUClock *clock = qemu_clock_ptr(type);
260 if (!clock->enabled) {
264 QLIST_FOREACH(timer_list, &clock->timerlists, list) {
265 qemu_mutex_lock(&timer_list->active_timers_lock);
266 ts = timer_list->active_timers;
267 /* Skip all external timers */
268 while (ts && (ts->attributes & ~attr_mask)) {
272 qemu_mutex_unlock(&timer_list->active_timers_lock);
275 expire_time = ts->expire_time;
276 qemu_mutex_unlock(&timer_list->active_timers_lock);
278 delta = expire_time - qemu_clock_get_ns(type);
282 deadline = qemu_soonest_timeout(deadline, delta);
287 QEMUClockType timerlist_get_clock(QEMUTimerList *timer_list)
289 return timer_list->clock->type;
292 QEMUTimerList *qemu_clock_get_main_loop_timerlist(QEMUClockType type)
294 return main_loop_tlg.tl[type];
297 void timerlist_notify(QEMUTimerList *timer_list)
299 if (timer_list->notify_cb) {
300 timer_list->notify_cb(timer_list->notify_opaque, timer_list->clock->type);
306 /* Transition function to convert a nanosecond timeout to ms
307 * This is used where a system does not support ppoll
309 int qemu_timeout_ns_to_ms(int64_t ns)
320 /* Always round up, because it's better to wait too long than to wait too
321 * little and effectively busy-wait
323 ms = DIV_ROUND_UP(ns, SCALE_MS);
325 /* To avoid overflow problems, limit this to 2^31, i.e. approx 25 days */
326 return MIN(ms, INT32_MAX);
330 /* qemu implementation of g_poll which uses a nanosecond timeout but is
331 * otherwise identical to g_poll
333 int qemu_poll_ns(GPollFD *fds, guint nfds, int64_t timeout)
337 return ppoll((struct pollfd *)fds, nfds, NULL, NULL);
340 int64_t tvsec = timeout / 1000000000LL;
341 /* Avoid possibly overflowing and specifying a negative number of
342 * seconds, which would turn a very long timeout into a busy-wait.
344 if (tvsec > (int64_t)INT32_MAX) {
348 ts.tv_nsec = timeout % 1000000000LL;
349 return ppoll((struct pollfd *)fds, nfds, &ts, NULL);
352 return g_poll(fds, nfds, qemu_timeout_ns_to_ms(timeout));
357 void timer_init_full(QEMUTimer *ts,
358 QEMUTimerListGroup *timer_list_group, QEMUClockType type,
359 int scale, int attributes,
360 QEMUTimerCB *cb, void *opaque)
362 if (!timer_list_group) {
363 timer_list_group = &main_loop_tlg;
365 ts->timer_list = timer_list_group->tl[type];
369 ts->attributes = attributes;
370 ts->expire_time = -1;
373 void timer_deinit(QEMUTimer *ts)
375 assert(ts->expire_time == -1);
376 ts->timer_list = NULL;
379 static void timer_del_locked(QEMUTimerList *timer_list, QEMUTimer *ts)
383 ts->expire_time = -1;
384 pt = &timer_list->active_timers;
390 qatomic_set(pt, t->next);
397 static bool timer_mod_ns_locked(QEMUTimerList *timer_list,
398 QEMUTimer *ts, int64_t expire_time)
402 /* add the timer in the sorted list */
403 pt = &timer_list->active_timers;
406 if (!timer_expired_ns(t, expire_time)) {
411 ts->expire_time = MAX(expire_time, 0);
415 return pt == &timer_list->active_timers;
418 static void timerlist_rearm(QEMUTimerList *timer_list)
420 /* Interrupt execution to force deadline recalculation. */
421 if (icount_enabled() && timer_list->clock->type == QEMU_CLOCK_VIRTUAL) {
422 icount_start_warp_timer();
424 timerlist_notify(timer_list);
427 /* stop a timer, but do not dealloc it */
428 void timer_del(QEMUTimer *ts)
430 QEMUTimerList *timer_list = ts->timer_list;
433 qemu_mutex_lock(&timer_list->active_timers_lock);
434 timer_del_locked(timer_list, ts);
435 qemu_mutex_unlock(&timer_list->active_timers_lock);
439 /* modify the current timer so that it will be fired when current_time
440 >= expire_time. The corresponding callback will be called. */
441 void timer_mod_ns(QEMUTimer *ts, int64_t expire_time)
443 QEMUTimerList *timer_list = ts->timer_list;
446 qemu_mutex_lock(&timer_list->active_timers_lock);
447 timer_del_locked(timer_list, ts);
448 rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
449 qemu_mutex_unlock(&timer_list->active_timers_lock);
452 timerlist_rearm(timer_list);
456 /* modify the current timer so that it will be fired when current_time
457 >= expire_time or the current deadline, whichever comes earlier.
458 The corresponding callback will be called. */
459 void timer_mod_anticipate_ns(QEMUTimer *ts, int64_t expire_time)
461 QEMUTimerList *timer_list = ts->timer_list;
464 WITH_QEMU_LOCK_GUARD(&timer_list->active_timers_lock) {
465 if (ts->expire_time == -1 || ts->expire_time > expire_time) {
466 if (ts->expire_time != -1) {
467 timer_del_locked(timer_list, ts);
469 rearm = timer_mod_ns_locked(timer_list, ts, expire_time);
475 timerlist_rearm(timer_list);
479 void timer_mod(QEMUTimer *ts, int64_t expire_time)
481 timer_mod_ns(ts, expire_time * ts->scale);
484 void timer_mod_anticipate(QEMUTimer *ts, int64_t expire_time)
486 timer_mod_anticipate_ns(ts, expire_time * ts->scale);
489 bool timer_pending(QEMUTimer *ts)
491 return ts->expire_time >= 0;
494 bool timer_expired(QEMUTimer *timer_head, int64_t current_time)
496 return timer_expired_ns(timer_head, current_time * timer_head->scale);
499 bool timerlist_run_timers(QEMUTimerList *timer_list)
502 int64_t current_time;
503 bool progress = false;
507 if (!qatomic_read(&timer_list->active_timers)) {
511 qemu_event_reset(&timer_list->timers_done_ev);
512 if (!timer_list->clock->enabled) {
516 switch (timer_list->clock->type) {
517 case QEMU_CLOCK_REALTIME:
520 case QEMU_CLOCK_VIRTUAL:
522 case QEMU_CLOCK_HOST:
523 if (!replay_checkpoint(CHECKPOINT_CLOCK_HOST)) {
527 case QEMU_CLOCK_VIRTUAL_RT:
528 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL_RT)) {
535 * Extract expired timers from active timers list and process them.
537 * In rr mode we need "filtered" checkpointing for virtual clock. The
538 * checkpoint must be recorded/replayed before processing any non-EXTERNAL timer,
539 * and that must only be done once since the clock value stays the same. Because
540 * non-EXTERNAL timers may appear in the timers list while it being processed,
541 * the checkpoint can be issued at a time until no timers are left and we are
544 current_time = qemu_clock_get_ns(timer_list->clock->type);
545 qemu_mutex_lock(&timer_list->active_timers_lock);
546 while ((ts = timer_list->active_timers)) {
547 if (!timer_expired_ns(ts, current_time)) {
548 /* No expired timers left. The checkpoint can be skipped
549 * if no timers fired or they were all external.
553 /* Checkpoint for virtual clock is redundant in cases where
554 * it's being triggered with only non-EXTERNAL timers, because
555 * these timers don't change guest state directly.
557 if (replay_mode != REPLAY_MODE_NONE
558 && timer_list->clock->type == QEMU_CLOCK_VIRTUAL
559 && !(ts->attributes & QEMU_TIMER_ATTR_EXTERNAL)
560 && !replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL)) {
561 qemu_mutex_unlock(&timer_list->active_timers_lock);
565 /* remove timer from the list before calling the callback */
566 timer_list->active_timers = ts->next;
568 ts->expire_time = -1;
572 /* run the callback (the timer list can be modified) */
573 qemu_mutex_unlock(&timer_list->active_timers_lock);
575 qemu_mutex_lock(&timer_list->active_timers_lock);
579 qemu_mutex_unlock(&timer_list->active_timers_lock);
582 qemu_event_set(&timer_list->timers_done_ev);
586 bool qemu_clock_run_timers(QEMUClockType type)
588 return timerlist_run_timers(main_loop_tlg.tl[type]);
591 void timerlistgroup_init(QEMUTimerListGroup *tlg,
592 QEMUTimerListNotifyCB *cb, void *opaque)
595 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
596 tlg->tl[type] = timerlist_new(type, cb, opaque);
600 void timerlistgroup_deinit(QEMUTimerListGroup *tlg)
603 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
604 timerlist_free(tlg->tl[type]);
608 bool timerlistgroup_run_timers(QEMUTimerListGroup *tlg)
611 bool progress = false;
612 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
613 progress |= timerlist_run_timers(tlg->tl[type]);
618 int64_t timerlistgroup_deadline_ns(QEMUTimerListGroup *tlg)
620 int64_t deadline = -1;
622 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
623 if (qemu_clock_use_for_deadline(type)) {
624 deadline = qemu_soonest_timeout(deadline,
625 timerlist_deadline_ns(tlg->tl[type]));
631 int64_t qemu_clock_get_ns(QEMUClockType type)
634 case QEMU_CLOCK_REALTIME:
637 case QEMU_CLOCK_VIRTUAL:
638 return cpus_get_virtual_clock();
639 case QEMU_CLOCK_HOST:
640 return REPLAY_CLOCK(REPLAY_CLOCK_HOST, get_clock_realtime());
641 case QEMU_CLOCK_VIRTUAL_RT:
642 return REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT, cpu_get_clock());
646 void init_clocks(QEMUTimerListNotifyCB *notify_cb)
649 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
650 qemu_clock_init(type, notify_cb);
653 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
654 prctl(PR_SET_TIMERSLACK, 1, 0, 0, 0);
658 uint64_t timer_expire_time_ns(QEMUTimer *ts)
660 return timer_pending(ts) ? ts->expire_time : -1;
663 bool qemu_clock_run_all_timers(void)
665 bool progress = false;
668 for (type = 0; type < QEMU_CLOCK_MAX; type++) {
669 if (qemu_clock_use_for_deadline(type)) {
670 progress |= qemu_clock_run_timers(type);