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"
41 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
42 #include <sys/prctl.h>
45 /***********************************************************/
48 typedef struct QEMUClock
{
49 /* We rely on BQL to protect the timerlists */
50 QLIST_HEAD(, QEMUTimerList
) timerlists
;
56 QEMUTimerListGroup main_loop_tlg
;
57 static QEMUClock qemu_clocks
[QEMU_CLOCK_MAX
];
59 /* A QEMUTimerList is a list of timers attached to a clock. More
60 * than one QEMUTimerList can be attached to each clock, for instance
61 * used by different AioContexts / threads. Each clock also has
62 * a list of the QEMUTimerLists associated with it, in order that
63 * reenabling the clock can call all the notifiers.
66 struct QEMUTimerList
{
68 QemuMutex active_timers_lock
;
69 QEMUTimer
*active_timers
;
70 QLIST_ENTRY(QEMUTimerList
) list
;
71 QEMUTimerListNotifyCB
*notify_cb
;
74 /* lightweight method to mark the end of timerlist's running */
75 QemuEvent timers_done_ev
;
80 * @type: type of clock
82 * Translate a clock type into a pointer to QEMUClock object.
84 * Returns: a pointer to the QEMUClock object
86 static inline QEMUClock
*qemu_clock_ptr(QEMUClockType type
)
88 return &qemu_clocks
[type
];
91 static bool timer_expired_ns(QEMUTimer
*timer_head
, int64_t current_time
)
93 return timer_head
&& (timer_head
->expire_time
<= current_time
);
96 QEMUTimerList
*timerlist_new(QEMUClockType type
,
97 QEMUTimerListNotifyCB
*cb
,
100 QEMUTimerList
*timer_list
;
101 QEMUClock
*clock
= qemu_clock_ptr(type
);
103 timer_list
= g_malloc0(sizeof(QEMUTimerList
));
104 qemu_event_init(&timer_list
->timers_done_ev
, true);
105 timer_list
->clock
= clock
;
106 timer_list
->notify_cb
= cb
;
107 timer_list
->notify_opaque
= opaque
;
108 qemu_mutex_init(&timer_list
->active_timers_lock
);
109 QLIST_INSERT_HEAD(&clock
->timerlists
, timer_list
, list
);
113 void timerlist_free(QEMUTimerList
*timer_list
)
115 assert(!timerlist_has_timers(timer_list
));
116 if (timer_list
->clock
) {
117 QLIST_REMOVE(timer_list
, list
);
119 qemu_mutex_destroy(&timer_list
->active_timers_lock
);
123 static void qemu_clock_init(QEMUClockType type
, QEMUTimerListNotifyCB
*notify_cb
)
125 QEMUClock
*clock
= qemu_clock_ptr(type
);
127 /* Assert that the clock of type TYPE has not been initialized yet. */
128 assert(main_loop_tlg
.tl
[type
] == NULL
);
131 clock
->enabled
= (type
== QEMU_CLOCK_VIRTUAL
? false : true);
132 QLIST_INIT(&clock
->timerlists
);
133 main_loop_tlg
.tl
[type
] = timerlist_new(type
, notify_cb
, NULL
);
136 bool qemu_clock_use_for_deadline(QEMUClockType type
)
138 return !(icount_enabled() && (type
== QEMU_CLOCK_VIRTUAL
));
141 void qemu_clock_notify(QEMUClockType type
)
143 QEMUTimerList
*timer_list
;
144 QEMUClock
*clock
= qemu_clock_ptr(type
);
145 QLIST_FOREACH(timer_list
, &clock
->timerlists
, list
) {
146 timerlist_notify(timer_list
);
150 /* Disabling the clock will wait for related timerlists to stop
151 * executing qemu_run_timers. Thus, this functions should not
152 * be used from the callback of a timer that is based on @clock.
153 * Doing so would cause a deadlock.
155 * Caller should hold BQL.
157 void qemu_clock_enable(QEMUClockType type
, bool enabled
)
159 QEMUClock
*clock
= qemu_clock_ptr(type
);
161 bool old
= clock
->enabled
;
162 clock
->enabled
= enabled
;
163 if (enabled
&& !old
) {
164 qemu_clock_notify(type
);
165 } else if (!enabled
&& old
) {
166 QLIST_FOREACH(tl
, &clock
->timerlists
, list
) {
167 qemu_event_wait(&tl
->timers_done_ev
);
172 bool timerlist_has_timers(QEMUTimerList
*timer_list
)
174 return !!qatomic_read(&timer_list
->active_timers
);
177 bool qemu_clock_has_timers(QEMUClockType type
)
179 return timerlist_has_timers(
180 main_loop_tlg
.tl
[type
]);
183 bool timerlist_expired(QEMUTimerList
*timer_list
)
187 if (!qatomic_read(&timer_list
->active_timers
)) {
191 WITH_QEMU_LOCK_GUARD(&timer_list
->active_timers_lock
) {
192 if (!timer_list
->active_timers
) {
195 expire_time
= timer_list
->active_timers
->expire_time
;
198 return expire_time
<= qemu_clock_get_ns(timer_list
->clock
->type
);
201 bool qemu_clock_expired(QEMUClockType type
)
203 return timerlist_expired(
204 main_loop_tlg
.tl
[type
]);
208 * As above, but return -1 for no deadline, and do not cap to 2^32
209 * as we know the result is always positive.
212 int64_t timerlist_deadline_ns(QEMUTimerList
*timer_list
)
217 if (!qatomic_read(&timer_list
->active_timers
)) {
221 if (!timer_list
->clock
->enabled
) {
225 /* The active timers list may be modified before the caller uses our return
226 * value but ->notify_cb() is called when the deadline changes. Therefore
227 * the caller should notice the change and there is no race condition.
229 WITH_QEMU_LOCK_GUARD(&timer_list
->active_timers_lock
) {
230 if (!timer_list
->active_timers
) {
233 expire_time
= timer_list
->active_timers
->expire_time
;
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
, int attr_mask
)
252 int64_t deadline
= -1;
256 QEMUTimerList
*timer_list
;
257 QEMUClock
*clock
= qemu_clock_ptr(type
);
259 if (!clock
->enabled
) {
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
)) {
271 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
274 expire_time
= ts
->expire_time
;
275 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
277 delta
= expire_time
- qemu_clock_get_ns(type
);
281 deadline
= qemu_soonest_timeout(deadline
, delta
);
286 QEMUClockType
timerlist_get_clock(QEMUTimerList
*timer_list
)
288 return timer_list
->clock
->type
;
291 QEMUTimerList
*qemu_clock_get_main_loop_timerlist(QEMUClockType type
)
293 return main_loop_tlg
.tl
[type
];
296 void timerlist_notify(QEMUTimerList
*timer_list
)
298 if (timer_list
->notify_cb
) {
299 timer_list
->notify_cb(timer_list
->notify_opaque
, timer_list
->clock
->type
);
305 /* Transition function to convert a nanosecond timeout to ms
306 * This is used where a system does not support ppoll
308 int qemu_timeout_ns_to_ms(int64_t ns
)
319 /* Always round up, because it's better to wait too long than to wait too
320 * little and effectively busy-wait
322 ms
= DIV_ROUND_UP(ns
, SCALE_MS
);
324 /* To avoid overflow problems, limit this to 2^31, i.e. approx 25 days */
325 return MIN(ms
, INT32_MAX
);
329 /* qemu implementation of g_poll which uses a nanosecond timeout but is
330 * otherwise identical to g_poll
332 int qemu_poll_ns(GPollFD
*fds
, guint nfds
, int64_t timeout
)
336 return ppoll((struct pollfd
*)fds
, nfds
, NULL
, NULL
);
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.
343 if (tvsec
> (int64_t)INT32_MAX
) {
347 ts
.tv_nsec
= timeout
% 1000000000LL;
348 return ppoll((struct pollfd
*)fds
, nfds
, &ts
, NULL
);
351 return g_poll(fds
, nfds
, qemu_timeout_ns_to_ms(timeout
));
356 void timer_init_full(QEMUTimer
*ts
,
357 QEMUTimerListGroup
*timer_list_group
, QEMUClockType type
,
358 int scale
, int attributes
,
359 QEMUTimerCB
*cb
, void *opaque
)
361 if (!timer_list_group
) {
362 timer_list_group
= &main_loop_tlg
;
364 ts
->timer_list
= timer_list_group
->tl
[type
];
368 ts
->attributes
= attributes
;
369 ts
->expire_time
= -1;
372 void timer_deinit(QEMUTimer
*ts
)
374 assert(ts
->expire_time
== -1);
375 ts
->timer_list
= NULL
;
378 static void timer_del_locked(QEMUTimerList
*timer_list
, QEMUTimer
*ts
)
382 ts
->expire_time
= -1;
383 pt
= &timer_list
->active_timers
;
389 qatomic_set(pt
, t
->next
);
396 static bool timer_mod_ns_locked(QEMUTimerList
*timer_list
,
397 QEMUTimer
*ts
, int64_t expire_time
)
401 /* add the timer in the sorted list */
402 pt
= &timer_list
->active_timers
;
405 if (!timer_expired_ns(t
, expire_time
)) {
410 ts
->expire_time
= MAX(expire_time
, 0);
414 return pt
== &timer_list
->active_timers
;
417 static void timerlist_rearm(QEMUTimerList
*timer_list
)
419 /* Interrupt execution to force deadline recalculation. */
420 if (icount_enabled() && timer_list
->clock
->type
== QEMU_CLOCK_VIRTUAL
) {
421 icount_start_warp_timer();
423 timerlist_notify(timer_list
);
426 /* stop a timer, but do not dealloc it */
427 void timer_del(QEMUTimer
*ts
)
429 QEMUTimerList
*timer_list
= ts
->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
);
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
)
442 QEMUTimerList
*timer_list
= ts
->timer_list
;
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
);
451 timerlist_rearm(timer_list
);
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
)
460 QEMUTimerList
*timer_list
= ts
->timer_list
;
463 WITH_QEMU_LOCK_GUARD(&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
);
468 rearm
= timer_mod_ns_locked(timer_list
, ts
, expire_time
);
474 timerlist_rearm(timer_list
);
478 void timer_mod(QEMUTimer
*ts
, int64_t expire_time
)
480 timer_mod_ns(ts
, expire_time
* ts
->scale
);
483 void timer_mod_anticipate(QEMUTimer
*ts
, int64_t expire_time
)
485 timer_mod_anticipate_ns(ts
, expire_time
* ts
->scale
);
488 bool timer_pending(QEMUTimer
*ts
)
490 return ts
->expire_time
>= 0;
493 bool timer_expired(QEMUTimer
*timer_head
, int64_t current_time
)
495 return timer_expired_ns(timer_head
, current_time
* timer_head
->scale
);
498 bool timerlist_run_timers(QEMUTimerList
*timer_list
)
501 int64_t current_time
;
502 bool progress
= false;
506 if (!qatomic_read(&timer_list
->active_timers
)) {
510 qemu_event_reset(&timer_list
->timers_done_ev
);
511 if (!timer_list
->clock
->enabled
) {
515 switch (timer_list
->clock
->type
) {
516 case QEMU_CLOCK_REALTIME
:
519 case QEMU_CLOCK_VIRTUAL
:
521 case QEMU_CLOCK_HOST
:
522 if (!replay_checkpoint(CHECKPOINT_CLOCK_HOST
)) {
526 case QEMU_CLOCK_VIRTUAL_RT
:
527 if (!replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL_RT
)) {
534 * Extract expired timers from active timers list and process them.
536 * In rr mode we need "filtered" checkpointing for virtual clock. The
537 * checkpoint must be recorded/replayed before processing any non-EXTERNAL timer,
538 * and that must only be done once since the clock value stays the same. Because
539 * non-EXTERNAL timers may appear in the timers list while it being processed,
540 * the checkpoint can be issued at a time until no timers are left and we are
543 current_time
= qemu_clock_get_ns(timer_list
->clock
->type
);
544 qemu_mutex_lock(&timer_list
->active_timers_lock
);
545 while ((ts
= timer_list
->active_timers
)) {
546 if (!timer_expired_ns(ts
, current_time
)) {
547 /* No expired timers left. The checkpoint can be skipped
548 * if no timers fired or they were all external.
552 /* Checkpoint for virtual clock is redundant in cases where
553 * it's being triggered with only non-EXTERNAL timers, because
554 * these timers don't change guest state directly.
556 if (replay_mode
!= REPLAY_MODE_NONE
557 && timer_list
->clock
->type
== QEMU_CLOCK_VIRTUAL
558 && !(ts
->attributes
& QEMU_TIMER_ATTR_EXTERNAL
)
559 && !replay_checkpoint(CHECKPOINT_CLOCK_VIRTUAL
)) {
560 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
564 /* remove timer from the list before calling the callback */
565 timer_list
->active_timers
= ts
->next
;
567 ts
->expire_time
= -1;
571 /* run the callback (the timer list can be modified) */
572 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
574 qemu_mutex_lock(&timer_list
->active_timers_lock
);
578 qemu_mutex_unlock(&timer_list
->active_timers_lock
);
581 qemu_event_set(&timer_list
->timers_done_ev
);
585 bool qemu_clock_run_timers(QEMUClockType type
)
587 return timerlist_run_timers(main_loop_tlg
.tl
[type
]);
590 void timerlistgroup_init(QEMUTimerListGroup
*tlg
,
591 QEMUTimerListNotifyCB
*cb
, void *opaque
)
594 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
595 tlg
->tl
[type
] = timerlist_new(type
, cb
, opaque
);
599 void timerlistgroup_deinit(QEMUTimerListGroup
*tlg
)
602 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
603 timerlist_free(tlg
->tl
[type
]);
607 bool timerlistgroup_run_timers(QEMUTimerListGroup
*tlg
)
610 bool progress
= false;
611 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
612 progress
|= timerlist_run_timers(tlg
->tl
[type
]);
617 int64_t timerlistgroup_deadline_ns(QEMUTimerListGroup
*tlg
)
619 int64_t deadline
= -1;
621 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
622 if (qemu_clock_use_for_deadline(type
)) {
623 deadline
= qemu_soonest_timeout(deadline
,
624 timerlist_deadline_ns(tlg
->tl
[type
]));
630 int64_t qemu_clock_get_ns(QEMUClockType type
)
633 case QEMU_CLOCK_REALTIME
:
636 case QEMU_CLOCK_VIRTUAL
:
637 return cpus_get_virtual_clock();
638 case QEMU_CLOCK_HOST
:
639 return REPLAY_CLOCK(REPLAY_CLOCK_HOST
, get_clock_realtime());
640 case QEMU_CLOCK_VIRTUAL_RT
:
641 return REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT
, cpu_get_clock());
645 void init_clocks(QEMUTimerListNotifyCB
*notify_cb
)
648 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
649 qemu_clock_init(type
, notify_cb
);
652 #ifdef CONFIG_PRCTL_PR_SET_TIMERSLACK
653 prctl(PR_SET_TIMERSLACK
, 1, 0, 0, 0);
657 uint64_t timer_expire_time_ns(QEMUTimer
*ts
)
659 return timer_pending(ts
) ? ts
->expire_time
: -1;
662 bool qemu_clock_run_all_timers(void)
664 bool progress
= false;
667 for (type
= 0; type
< QEMU_CLOCK_MAX
; type
++) {
668 if (qemu_clock_use_for_deadline(type
)) {
669 progress
|= qemu_clock_run_timers(type
);