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 "sysemu/replay.h"
29 #include "sysemu/cpus.h"
39 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
40 #include <sys/prctl.h>
43 /***********************************************************/
46 typedef struct QEMUClock
{
47 /* We rely on BQL to protect the timerlists */
48 QLIST_HEAD(, QEMUTimerList
) timerlists
;
54 QEMUTimerListGroup main_loop_tlg
;
55 static QEMUClock qemu_clocks
[QEMU_CLOCK_MAX
];
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.
64 struct QEMUTimerList
{
66 QemuMutex active_timers_lock
;
67 QEMUTimer
*active_timers
;
68 QLIST_ENTRY(QEMUTimerList
) list
;
69 QEMUTimerListNotifyCB
*notify_cb
;
72 /* lightweight method to mark the end of timerlist's running */
73 QemuEvent timers_done_ev
;
78 * @type: type of clock
80 * Translate a clock type into a pointer to QEMUClock object.
82 * Returns: a pointer to the QEMUClock object
84 static inline QEMUClock
*qemu_clock_ptr(QEMUClockType type
)
86 return &qemu_clocks
[type
];
89 static bool timer_expired_ns(QEMUTimer
*timer_head
, int64_t current_time
)
91 return timer_head
&& (timer_head
->expire_time
<= current_time
);
94 QEMUTimerList
*timerlist_new(QEMUClockType type
,
95 QEMUTimerListNotifyCB
*cb
,
98 QEMUTimerList
*timer_list
;
99 QEMUClock
*clock
= qemu_clock_ptr(type
);
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
);
111 void timerlist_free(QEMUTimerList
*timer_list
)
113 assert(!timerlist_has_timers(timer_list
));
114 if (timer_list
->clock
) {
115 QLIST_REMOVE(timer_list
, list
);
117 qemu_mutex_destroy(&timer_list
->active_timers_lock
);
121 static void qemu_clock_init(QEMUClockType type
, QEMUTimerListNotifyCB
*notify_cb
)
123 QEMUClock
*clock
= qemu_clock_ptr(type
);
125 /* Assert that the clock of type TYPE has not been initialized yet. */
126 assert(main_loop_tlg
.tl
[type
] == NULL
);
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
);
134 bool qemu_clock_use_for_deadline(QEMUClockType type
)
136 return !(use_icount
&& (type
== QEMU_CLOCK_VIRTUAL
));
139 void qemu_clock_notify(QEMUClockType type
)
141 QEMUTimerList
*timer_list
;
142 QEMUClock
*clock
= qemu_clock_ptr(type
);
143 QLIST_FOREACH(timer_list
, &clock
->timerlists
, list
) {
144 timerlist_notify(timer_list
);
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.
153 * Caller should hold BQL.
155 void qemu_clock_enable(QEMUClockType type
, bool enabled
)
157 QEMUClock
*clock
= qemu_clock_ptr(type
);
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
);
170 bool timerlist_has_timers(QEMUTimerList
*timer_list
)
172 return !!atomic_read(&timer_list
->active_timers
);
175 bool qemu_clock_has_timers(QEMUClockType type
)
177 return timerlist_has_timers(
178 main_loop_tlg
.tl
[type
]);
181 bool timerlist_expired(QEMUTimerList
*timer_list
)
185 if (!atomic_read(&timer_list
->active_timers
)) {
189 qemu_mutex_lock(&timer_list
->active_timers_lock
);
190 if (!timer_list
->active_timers
) {
191 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
194 expire_time
= timer_list
->active_timers
->expire_time
;
195 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
197 return expire_time
<= qemu_clock_get_ns(timer_list
->clock
->type
);
200 bool qemu_clock_expired(QEMUClockType type
)
202 return timerlist_expired(
203 main_loop_tlg
.tl
[type
]);
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.
211 int64_t timerlist_deadline_ns(QEMUTimerList
*timer_list
)
216 if (!atomic_read(&timer_list
->active_timers
)) {
220 if (!timer_list
->clock
->enabled
) {
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.
228 qemu_mutex_lock(&timer_list
->active_timers_lock
);
229 if (!timer_list
->active_timers
) {
230 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
233 expire_time
= timer_list
->active_timers
->expire_time
;
234 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
236 delta
= expire_time
- qemu_clock_get_ns(timer_list
->clock
->type
);
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
250 int64_t qemu_clock_deadline_ns_all(QEMUClockType type
)
252 int64_t deadline
= -1;
253 QEMUTimerList
*timer_list
;
254 QEMUClock
*clock
= qemu_clock_ptr(type
);
255 QLIST_FOREACH(timer_list
, &clock
->timerlists
, list
) {
256 deadline
= qemu_soonest_timeout(deadline
,
257 timerlist_deadline_ns(timer_list
));
262 QEMUClockType
timerlist_get_clock(QEMUTimerList
*timer_list
)
264 return timer_list
->clock
->type
;
267 QEMUTimerList
*qemu_clock_get_main_loop_timerlist(QEMUClockType type
)
269 return main_loop_tlg
.tl
[type
];
272 void timerlist_notify(QEMUTimerList
*timer_list
)
274 if (timer_list
->notify_cb
) {
275 timer_list
->notify_cb(timer_list
->notify_opaque
, timer_list
->clock
->type
);
281 /* Transition function to convert a nanosecond timeout to ms
282 * This is used where a system does not support ppoll
284 int qemu_timeout_ns_to_ms(int64_t ns
)
295 /* Always round up, because it's better to wait too long than to wait too
296 * little and effectively busy-wait
298 ms
= DIV_ROUND_UP(ns
, SCALE_MS
);
300 /* To avoid overflow problems, limit this to 2^31, i.e. approx 25 days */
301 if (ms
> (int64_t) INT32_MAX
) {
309 /* qemu implementation of g_poll which uses a nanosecond timeout but is
310 * otherwise identical to g_poll
312 int qemu_poll_ns(GPollFD
*fds
, guint nfds
, int64_t timeout
)
316 return ppoll((struct pollfd
*)fds
, nfds
, NULL
, NULL
);
319 int64_t tvsec
= timeout
/ 1000000000LL;
320 /* Avoid possibly overflowing and specifying a negative number of
321 * seconds, which would turn a very long timeout into a busy-wait.
323 if (tvsec
> (int64_t)INT32_MAX
) {
327 ts
.tv_nsec
= timeout
% 1000000000LL;
328 return ppoll((struct pollfd
*)fds
, nfds
, &ts
, NULL
);
331 return g_poll(fds
, nfds
, qemu_timeout_ns_to_ms(timeout
));
336 void timer_init_full(QEMUTimer
*ts
,
337 QEMUTimerListGroup
*timer_list_group
, QEMUClockType type
,
338 int scale
, int attributes
,
339 QEMUTimerCB
*cb
, void *opaque
)
341 if (!timer_list_group
) {
342 timer_list_group
= &main_loop_tlg
;
344 ts
->timer_list
= timer_list_group
->tl
[type
];
348 ts
->attributes
= attributes
;
349 ts
->expire_time
= -1;
352 void timer_deinit(QEMUTimer
*ts
)
354 assert(ts
->expire_time
== -1);
355 ts
->timer_list
= NULL
;
358 static void timer_del_locked(QEMUTimerList
*timer_list
, QEMUTimer
*ts
)
362 ts
->expire_time
= -1;
363 pt
= &timer_list
->active_timers
;
369 atomic_set(pt
, t
->next
);
376 static bool timer_mod_ns_locked(QEMUTimerList
*timer_list
,
377 QEMUTimer
*ts
, int64_t expire_time
)
381 /* add the timer in the sorted list */
382 pt
= &timer_list
->active_timers
;
385 if (!timer_expired_ns(t
, expire_time
)) {
390 ts
->expire_time
= MAX(expire_time
, 0);
394 return pt
== &timer_list
->active_timers
;
397 static void timerlist_rearm(QEMUTimerList
*timer_list
)
399 /* Interrupt execution to force deadline recalculation. */
400 if (timer_list
->clock
->type
== QEMU_CLOCK_VIRTUAL
) {
401 qemu_start_warp_timer();
403 timerlist_notify(timer_list
);
406 /* stop a timer, but do not dealloc it */
407 void timer_del(QEMUTimer
*ts
)
409 QEMUTimerList
*timer_list
= ts
->timer_list
;
412 qemu_mutex_lock(&timer_list
->active_timers_lock
);
413 timer_del_locked(timer_list
, ts
);
414 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
418 /* modify the current timer so that it will be fired when current_time
419 >= expire_time. The corresponding callback will be called. */
420 void timer_mod_ns(QEMUTimer
*ts
, int64_t expire_time
)
422 QEMUTimerList
*timer_list
= ts
->timer_list
;
425 qemu_mutex_lock(&timer_list
->active_timers_lock
);
426 timer_del_locked(timer_list
, ts
);
427 rearm
= timer_mod_ns_locked(timer_list
, ts
, expire_time
);
428 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
431 timerlist_rearm(timer_list
);
435 /* modify the current timer so that it will be fired when current_time
436 >= expire_time or the current deadline, whichever comes earlier.
437 The corresponding callback will be called. */
438 void timer_mod_anticipate_ns(QEMUTimer
*ts
, int64_t expire_time
)
440 QEMUTimerList
*timer_list
= ts
->timer_list
;
443 qemu_mutex_lock(&timer_list
->active_timers_lock
);
444 if (ts
->expire_time
== -1 || ts
->expire_time
> expire_time
) {
445 if (ts
->expire_time
!= -1) {
446 timer_del_locked(timer_list
, ts
);
448 rearm
= timer_mod_ns_locked(timer_list
, ts
, expire_time
);
452 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
455 timerlist_rearm(timer_list
);
459 void timer_mod(QEMUTimer
*ts
, int64_t expire_time
)
461 timer_mod_ns(ts
, expire_time
* ts
->scale
);
464 void timer_mod_anticipate(QEMUTimer
*ts
, int64_t expire_time
)
466 timer_mod_anticipate_ns(ts
, expire_time
* ts
->scale
);
469 bool timer_pending(QEMUTimer
*ts
)
471 return ts
->expire_time
>= 0;
474 bool timer_expired(QEMUTimer
*timer_head
, int64_t current_time
)
476 return timer_expired_ns(timer_head
, current_time
* timer_head
->scale
);
479 bool timerlist_run_timers(QEMUTimerList
*timer_list
)
482 int64_t current_time
;
483 bool progress
= false;
486 bool need_replay_checkpoint
= false;
488 if (!atomic_read(&timer_list
->active_timers
)) {
492 qemu_event_reset(&timer_list
->timers_done_ev
);
493 if (!timer_list
->clock
->enabled
) {
497 switch (timer_list
->clock
->type
) {
498 case QEMU_CLOCK_REALTIME
:
501 case QEMU_CLOCK_VIRTUAL
:
502 if (replay_mode
!= REPLAY_MODE_NONE
) {
503 /* Checkpoint for virtual clock is redundant in cases where
504 * it's being triggered with only non-EXTERNAL timers, because
505 * these timers don't change guest state directly.
506 * Since it has conditional dependence on specific timers, it is
507 * subject to race conditions and requires special handling.
510 need_replay_checkpoint
= true;
513 case QEMU_CLOCK_HOST
:
514 if (!replay_checkpoint(CHECKPOINT_CLOCK_HOST
)) {
518 case QEMU_CLOCK_VIRTUAL_RT
:
519 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL_RT
)) {
526 * Extract expired timers from active timers list and and process them.
528 * In rr mode we need "filtered" checkpointing for virtual clock. The
529 * checkpoint must be recorded/replayed before processing any non-EXTERNAL timer,
530 * and that must only be done once since the clock value stays the same. Because
531 * non-EXTERNAL timers may appear in the timers list while it being processed,
532 * the checkpoint can be issued at a time until no timers are left and we are
535 current_time
= qemu_clock_get_ns(timer_list
->clock
->type
);
536 qemu_mutex_lock(&timer_list
->active_timers_lock
);
537 while ((ts
= timer_list
->active_timers
)) {
538 if (!timer_expired_ns(ts
, current_time
)) {
539 /* No expired timers left. The checkpoint can be skipped
540 * if no timers fired or they were all external.
544 if (need_replay_checkpoint
545 && !(ts
->attributes
& QEMU_TIMER_ATTR_EXTERNAL
)) {
546 /* once we got here, checkpoint clock only once */
547 need_replay_checkpoint
= false;
548 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
549 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL
)) {
552 qemu_mutex_lock(&timer_list
->active_timers_lock
);
553 /* The lock was released; start over again in case the list was
559 /* remove timer from the list before calling the callback */
560 timer_list
->active_timers
= ts
->next
;
562 ts
->expire_time
= -1;
566 /* run the callback (the timer list can be modified) */
567 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
569 qemu_mutex_lock(&timer_list
->active_timers_lock
);
573 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
576 qemu_event_set(&timer_list
->timers_done_ev
);
580 bool qemu_clock_run_timers(QEMUClockType type
)
582 return timerlist_run_timers(main_loop_tlg
.tl
[type
]);
585 void timerlistgroup_init(QEMUTimerListGroup
*tlg
,
586 QEMUTimerListNotifyCB
*cb
, void *opaque
)
589 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
590 tlg
->tl
[type
] = timerlist_new(type
, cb
, opaque
);
594 void timerlistgroup_deinit(QEMUTimerListGroup
*tlg
)
597 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
598 timerlist_free(tlg
->tl
[type
]);
602 bool timerlistgroup_run_timers(QEMUTimerListGroup
*tlg
)
605 bool progress
= false;
606 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
607 progress
|= timerlist_run_timers(tlg
->tl
[type
]);
612 int64_t timerlistgroup_deadline_ns(QEMUTimerListGroup
*tlg
)
614 int64_t deadline
= -1;
616 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
617 if (qemu_clock_use_for_deadline(type
)) {
618 deadline
= qemu_soonest_timeout(deadline
,
619 timerlist_deadline_ns(tlg
->tl
[type
]));
625 int64_t qemu_clock_get_ns(QEMUClockType type
)
628 case QEMU_CLOCK_REALTIME
:
631 case QEMU_CLOCK_VIRTUAL
:
633 return cpu_get_icount();
635 return cpu_get_clock();
637 case QEMU_CLOCK_HOST
:
638 return REPLAY_CLOCK(REPLAY_CLOCK_HOST
, get_clock_realtime());
639 case QEMU_CLOCK_VIRTUAL_RT
:
640 return REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT
, cpu_get_clock());
644 void init_clocks(QEMUTimerListNotifyCB
*notify_cb
)
647 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
648 qemu_clock_init(type
, notify_cb
);
651 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
652 prctl(PR_SET_TIMERSLACK
, 1, 0, 0, 0);
656 uint64_t timer_expire_time_ns(QEMUTimer
*ts
)
658 return timer_pending(ts
) ? ts
->expire_time
: -1;
661 bool qemu_clock_run_all_timers(void)
663 bool progress
= false;
666 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
667 if (qemu_clock_use_for_deadline(type
)) {
668 progress
|= qemu_clock_run_timers(type
);