1 .\" $NetBSD: timeout.9,v 1.2 1996/06/23 22:32:34 pk Exp $
3 .\" Copyright (c) 1996 The NetBSD Foundation, Inc.
4 .\" All rights reserved.
6 .\" This code is derived from software contributed to The NetBSD Foundation
7 .\" by Paul Kranenburg.
9 .\" Redistribution and use in source and binary forms, with or without
10 .\" modification, are permitted provided that the following conditions
12 .\" 1. Redistributions of source code must retain the above copyright
13 .\" notice, this list of conditions and the following disclaimer.
14 .\" 2. Redistributions in binary form must reproduce the above copyright
15 .\" notice, this list of conditions and the following disclaimer in the
16 .\" documentation and/or other materials provided with the distribution.
18 .\" THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19 .\" ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20 .\" TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21 .\" PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE
22 .\" LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23 .\" CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24 .\" SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25 .\" INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26 .\" CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27 .\" ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 .\" POSSIBILITY OF SUCH DAMAGE.
37 .Nm callout_deactivate ,
38 .Nm callout_async_drain ,
41 .Nm callout_init_mtx ,
46 .Nm callout_reset_curcpu ,
47 .Nm callout_reset_on ,
48 .Nm callout_reset_sbt ,
49 .Nm callout_reset_sbt_curcpu ,
50 .Nm callout_reset_sbt_on ,
51 .Nm callout_schedule ,
52 .Nm callout_schedule_curcpu ,
53 .Nm callout_schedule_on ,
54 .Nm callout_schedule_sbt ,
55 .Nm callout_schedule_sbt_curcpu ,
56 .Nm callout_schedule_sbt_on ,
59 .Nd execute a function after a specified length of time
64 typedef void callout_func_t (void *);
67 .Fn callout_active "struct callout *c"
69 .Fn callout_deactivate "struct callout *c"
71 .Fn callout_async_drain "struct callout *c" "callout_func_t *drain"
73 .Fn callout_drain "struct callout *c"
75 .Fn callout_init "struct callout *c" "int mpsafe"
77 .Fn callout_init_mtx "struct callout *c" "struct mtx *mtx" "int flags"
79 .Fn callout_init_rm "struct callout *c" "struct rmlock *rm" "int flags"
81 .Fn callout_init_rw "struct callout *c" "struct rwlock *rw" "int flags"
83 .Fn callout_pending "struct callout *c"
86 .Fa "struct callout *c"
88 .Fa "callout_func_t *func"
92 .Fo callout_reset_curcpu
93 .Fa "struct callout *c"
95 .Fa "callout_func_t *func"
100 .Fa "struct callout *c"
102 .Fa "callout_func_t *func"
107 .Fo callout_reset_sbt
108 .Fa "struct callout *c"
111 .Fa "callout_func_t *func"
116 .Fo callout_reset_sbt_curcpu
117 .Fa "struct callout *c"
120 .Fa "callout_func_t *func"
125 .Fo callout_reset_sbt_on
126 .Fa "struct callout *c"
129 .Fa "callout_func_t *func"
135 .Fn callout_schedule "struct callout *c" "int ticks"
137 .Fn callout_schedule_curcpu "struct callout *c" "int ticks"
139 .Fn callout_schedule_on "struct callout *c" "int ticks" "int cpu"
141 .Fo callout_schedule_sbt
142 .Fa "struct callout *c"
148 .Fo callout_schedule_sbt_curcpu
149 .Fa "struct callout *c"
155 .Fo callout_schedule_sbt_on
156 .Fa "struct callout *c"
163 .Fn callout_stop "struct callout *c"
167 .Fa "sbintime_t precision"
169 .Fa "sbintime_t *sbt_res"
170 .Fa "sbintime_t *precision_res"
175 API is used to schedule a call to an arbitrary function at a specific
177 Consumers of this API are required to allocate a callout structure
179 for each pending function invocation.
180 This structure stores state about the pending function invocation including
181 the function to be called and the time at which the function should be invoked.
182 Pending function calls can be cancelled or rescheduled to a different time.
184 a callout structure may be reused to schedule a new function call after a
185 scheduled call is completed.
187 Callouts only provide a single-shot mode.
188 If a consumer requires a periodic timer,
189 it must explicitly reschedule each function call.
190 This is normally done by rescheduling the subsequent call within the called
193 Callout functions must not sleep.
194 They may not acquire sleepable locks,
195 wait on condition variables,
196 perform blocking allocation requests,
197 or invoke any other action that might sleep.
199 Each callout structure must be initialized by
201 .Fn callout_init_mtx ,
202 .Fn callout_init_rm ,
205 before it is passed to any of the other callout functions.
208 function initializes a callout structure in
210 that is not associated with a specific lock.
214 the callout structure is not considered to be
215 .Dq multi-processor safe ;
216 and the Giant lock will be acquired before calling the callout function
217 and released when the callout function returns.
220 .Fn callout_init_mtx ,
221 .Fn callout_init_rm ,
224 functions initialize a callout structure in
226 that is associated with a specific lock.
227 The lock is specified by the
233 The associated lock must be held while stopping or rescheduling the
235 The callout subsystem acquires the associated lock before calling the
236 callout function and releases it after the function returns.
237 If the callout was cancelled while the callout subsystem waited for the
239 the callout function is not called,
240 and the associated lock is released.
241 This ensures that stopping or rescheduling the callout will abort any
242 previously scheduled invocation.
244 A sleepable read-mostly lock
246 one initialized with the
251 .Fn callout_init_rm .
252 Similarly, other sleepable lock types such as
256 cannot be used with callouts because sleeping is not permitted in
257 the callout subsystem.
262 .Fn callout_init_mtx ,
263 .Fn callout_init_rm ,
265 .Fn callout_init_rw :
266 .Bl -tag -width ".Dv CALLOUT_RETURNUNLOCKED"
267 .It Dv CALLOUT_RETURNUNLOCKED
268 The callout function will release the associated lock itself,
269 so the callout subsystem should not attempt to unlock it
270 after the callout function returns.
271 .It Dv CALLOUT_SHAREDLOCK
272 The lock is only acquired in read mode when running the callout handler.
273 This flag is ignored by
274 .Fn callout_init_mtx .
281 if it is currently pending.
282 If the callout is pending and successfully stopped, then
284 returns a value of one.
285 If the callout is not set, or
286 has already been serviced, then
287 negative one is returned.
288 If the callout is currently being serviced and cannot be stopped,
289 then zero will be returned.
290 If the callout is currently being serviced and cannot be stopped, and at the
291 same time a next invocation of the same callout is also scheduled, then
293 unschedules the next run and returns zero.
294 If the callout has an associated lock,
295 then that lock must be held when this function is called.
298 .Fn callout_async_drain
303 .Fn callout_async_drain
304 returns zero it will arrange for the function
306 to be called using the same argument given to the
309 .Fn callout_async_drain
310 If the callout has an associated lock,
311 then that lock must be held when this function is called.
312 Note that when stopping multiple callouts that use the same lock it is possible
313 to get multiple return's of zero and multiple calls to the
315 function, depending upon which CPU's the callouts are running.
318 function itself is called from the context of the completing callout
319 i.e. softclock or hardclock, just like a callout itself.
325 except that it will wait for the callout
327 to complete if it is already in progress.
328 This function MUST NOT be called while holding any
329 locks on which the callout might block, or deadlock will result.
330 Note that if the callout subsystem has already begun processing this
331 callout, then the callout function may be invoked before
334 However, the callout subsystem does guarantee that the callout will be
343 function families schedule a future function invocation for callout
347 already has a pending callout,
348 it is cancelled before the new invocation is scheduled.
349 These functions return a value of one if a pending callout was cancelled
350 and zero if there was no pending callout.
351 If the callout has an associated lock,
352 then that lock must be held when any of these functions are called.
354 The time at which the callout function will be invoked is determined by
366 the callout is scheduled to execute after
369 Non-positive values of
371 are silently converted to the value
379 arguments provide more control over the scheduled time including
380 support for higher resolution times,
381 specifying the precision of the scheduled time,
382 and setting an absolute deadline instead of a relative timeout.
383 The callout is scheduled to execute in a time window which begins at
384 the time specified in
386 and extends for the amount of time specified in
390 specifies a time in the past,
391 the window is adjusted to start at the current time.
394 allows the callout subsystem to coalesce callouts scheduled close to each
395 other into fewer timer interrupts,
396 reducing processing overhead and power consumption.
399 may be specified to adjust the interpretation of
403 .Bl -tag -width ".Dv C_DIRECT_EXEC"
407 argument as an absolute time since boot.
410 is treated as a relative amount of time,
414 Run the handler directly from hardware interrupt context instead of from the
416 This reduces latency and overhead, but puts more constraints on the callout
418 Callout functions run in this context may use only spin mutexes for locking
419 and should be as small as possible because they run with absolute priority.
421 Specifies relative event time precision as binary logarithm of time interval
422 divided by acceptable time deviation: 1 -- 1/2, 2 -- 1/4, etc.
423 Note that the larger of
425 or this value is used as the length of the time window.
427 .Pq which result in larger time intervals
428 allow the callout subsystem to aggregate more events in one timer interrupt.
432 argument specifies the absolute time at which the callout should be run,
435 argument specifies the requested precision, which will not be
436 adjusted during the scheduling process.
441 values should be calculated by an earlier call to
443 which uses the user-supplied
450 Align the timeouts to
459 argument which identifies the function to be called when the time expires.
460 It must be a pointer to a function that takes a single
467 as its only argument.
474 arguments from the previous callout.
477 functions must always be called to initialize
483 functions can be used.
485 The callout subsystem provides a softclock thread for each CPU in the system.
486 Callouts are assigned to a single CPU and are executed by the softclock thread
489 callouts are assigned to CPU 0.
491 .Fn callout_reset_on ,
492 .Fn callout_reset_sbt_on ,
493 .Fn callout_schedule_on
495 .Fn callout_schedule_sbt_on
496 functions assign the callout to CPU
499 .Fn callout_reset_curcpu ,
500 .Fn callout_reset_sbt_curpu ,
501 .Fn callout_schedule_curcpu
503 .Fn callout_schedule_sbt_curcpu
504 functions assign the callout to the current CPU.
507 .Fn callout_reset_sbt ,
510 .Fn callout_schedule_sbt
511 functions schedule the callout to execute in the softclock thread of the CPU
512 to which it is currently assigned.
514 Softclock threads are not pinned to their respective CPUs by default.
515 The softclock thread for CPU 0 can be pinned to CPU 0 by setting the
516 .Va kern.pin_default_swi
517 loader tunable to a non-zero value.
518 Softclock threads for CPUs other than zero can be pinned to their
519 respective CPUs by setting the
520 .Va kern.pin_pcpu_swi
521 loader tunable to a non-zero value.
524 .Fn callout_pending ,
527 .Fn callout_deactivate
528 provide access to the current state of the callout.
531 macro checks whether a callout is
533 a callout is considered
535 when a timeout has been set but the time has not yet arrived.
536 Note that once the timeout time arrives and the callout subsystem
537 starts to process this callout,
541 even though the callout function may not have finished
546 macro checks whether a callout is marked as
549 .Fn callout_deactivate
550 macro clears the callout's
553 The callout subsystem marks a callout as
555 when a timeout is set and it clears the
563 clear it when a callout expires normally via the execution of the
568 function may be used to pre-calculate the absolute time at which the
569 timeout should be run and the precision of the scheduled run time
570 according to the required time
574 and additional adjustments requested by the
577 Flags accepted by the
579 function are the same as flags for the
582 The resulting time is assigned to the variable pointed to by the
584 argument, and the resulting precision is assigned to
586 When passing the results to
592 to avoid incorrect re-adjustment.
593 The function is intended for situations where precise time of the callout
594 run should be known in advance, since
595 trying to read this time from the callout structure itself after a
598 .Ss "Avoiding Race Conditions"
599 The callout subsystem invokes callout functions from its own thread
601 Without some kind of synchronization,
602 it is possible that a callout
603 function will be invoked concurrently with an attempt to stop or reset
604 the callout by another thread.
605 In particular, since callout functions typically acquire a lock as
606 their first action, the callout function may have already been invoked,
607 but is blocked waiting for that lock at the time that another thread
608 tries to reset or stop the callout.
610 There are three main techniques for addressing these
611 synchronization concerns.
612 The first approach is preferred as it is the simplest:
613 .Bl -enum -offset indent
615 Callouts can be associated with a specific lock when they are initialized
617 .Fn callout_init_mtx ,
618 .Fn callout_init_rm ,
620 .Fn callout_init_rw .
621 When a callout is associated with a lock,
622 the callout subsystem acquires the lock before the callout function is
624 This allows the callout subsystem to transparently handle races between
625 callout cancellation,
628 Note that the associated lock must be acquired before calling
634 functions to provide this safety.
636 A callout initialized via
640 set to zero is implicitly associated with the
645 is held when cancelling or rescheduling the callout,
646 then its use will prevent races with the callout function.
648 The return value from
657 indicates whether or not the callout was removed.
658 If it is known that the callout was set and the callout function has
659 not yet executed, then a return value of
661 indicates that the callout function is about to be called.
663 .Bd -literal -offset indent
664 if (sc->sc_flags & SCFLG_CALLOUT_RUNNING) {
665 if (callout_stop(&sc->sc_callout)) {
666 sc->sc_flags &= ~SCFLG_CALLOUT_RUNNING;
667 /* successfully stopped */
670 * callout has expired and callout
671 * function is about to be executed
678 .Fn callout_pending ,
681 .Fn callout_deactivate
682 macros can be used together to work around the race conditions.
683 When a callout's timeout is set, the callout subsystem marks the
688 When the timeout time arrives, the callout subsystem begins processing
689 the callout by first clearing the
692 It then invokes the callout function without changing the
694 flag, and does not clear the
696 flag even after the callout function returns.
697 The mechanism described here requires the callout function itself to
701 .Fn callout_deactivate
707 functions always clear both the
711 flags before returning.
713 The callout function should first check the
715 flag and return without action if
719 This indicates that the callout was rescheduled using
721 just before the callout function was invoked.
726 then the callout function should also return without action.
727 This indicates that the callout has been stopped.
728 Finally, the callout function should call
729 .Fn callout_deactivate
734 .Bd -literal -offset indent
735 mtx_lock(&sc->sc_mtx);
736 if (callout_pending(&sc->sc_callout)) {
737 /* callout was reset */
738 mtx_unlock(&sc->sc_mtx);
741 if (!callout_active(&sc->sc_callout)) {
742 /* callout was stopped */
743 mtx_unlock(&sc->sc_mtx);
746 callout_deactivate(&sc->sc_callout);
747 /* rest of callout function */
750 Together with appropriate synchronization, such as the mutex used above,
751 this approach permits the
755 functions to be used at any time without races.
757 .Bd -literal -offset indent
758 mtx_lock(&sc->sc_mtx);
759 callout_stop(&sc->sc_callout);
760 /* The callout is effectively stopped now. */
763 If the callout is still pending then these functions operate normally,
764 but if processing of the callout has already begun then the tests in
765 the callout function cause it to return without further action.
766 Synchronization between the callout function and other code ensures that
767 stopping or resetting the callout will never be attempted while the
768 callout function is past the
769 .Fn callout_deactivate
772 The above technique additionally ensures that the
774 flag always reflects whether the callout is effectively enabled or
778 returns false, then the callout is effectively disabled, since even if
779 the callout subsystem is actually just about to invoke the callout
780 function, the callout function will return without action.
783 There is one final race condition that must be considered when a
784 callout is being stopped for the last time.
785 In this case it may not be safe to let the callout function itself
786 detect that the callout was stopped, since it may need to access
787 data objects that have already been destroyed or recycled.
788 To ensure that the callout is completely finished, a call to
792 a callout should always be drained prior to destroying its associated lock
793 or releasing the storage for the callout structure.
797 macro returns the state of a callout's
803 macro returns the state of a callout's
811 function families return a value of one if the callout was pending before the new
812 function invocation was scheduled.
818 functions return a value of one if the callout was still pending when it was
819 called, a zero if the callout could not be stopped and a negative one is it
820 was either not running or has already completed.
823 initially used the long standing
826 callout mechanism which offered O(n) insertion and removal running time
827 but did not generate or require handles for untimeout operations.
830 introduced a new set of timeout and untimeout routines from
835 .An George Varghese ,
836 published in a technical report entitled
837 .%T "Redesigning the BSD Callout and Timer Facilities"
838 and modified for inclusion in
841 .An Justin T. Gibbs .
842 The original work on the data structures used in that implementation
848 .%T "Hashed and Hierarchical Timing Wheels: Data Structures for the Efficient Implementation of a Timer Facility"
850 .%B "Proceedings of the 11th ACM Annual Symposium on Operating Systems Principles" .
853 introduced the first implementations of
858 which permitted callers to allocate dedicated storage for callouts.
859 This ensured that a callout would always fire unlike
861 which would silently fail if it was unable to allocate a callout.
864 permitted callout handlers to be tagged as MPSAFE via
873 .Fn callout_init_mtx .
876 introduced per-CPU callout wheels,
877 .Fn callout_init_rw ,
879 .Fn callout_schedule .
882 changed the underlying timer interrupts used to drive callouts to prefer
883 one-shot event timers instead of a periodic timer interrupt.
886 switched the callout wheel to support tickless operation.
887 These changes introduced
890 .Fn callout_reset_sbt*
895 .Fn callout_init_rm .
899 .Fn callout_schedule_sbt*
904 .Fn callout_async_drain .