2 * Copyright (c) 1982, 1986, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * From: @(#)kern_clock.c 8.5 (Berkeley) 1/21/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/callout.h>
43 #include <sys/condvar.h>
44 #include <sys/kernel.h>
47 #include <sys/mutex.h>
48 #include <sys/sysctl.h>
51 SYSCTL_INT(_debug, OID_AUTO, to_avg_depth, CTLFLAG_RD, &avg_depth, 0,
52 "Average number of items examined per softclock call. Units = 1/1000");
53 static int avg_gcalls;
54 SYSCTL_INT(_debug, OID_AUTO, to_avg_gcalls, CTLFLAG_RD, &avg_gcalls, 0,
55 "Average number of Giant callouts made per softclock call. Units = 1/1000");
56 static int avg_mtxcalls;
57 SYSCTL_INT(_debug, OID_AUTO, to_avg_mtxcalls, CTLFLAG_RD, &avg_mtxcalls, 0,
58 "Average number of mtx callouts made per softclock call. Units = 1/1000");
59 static int avg_mpcalls;
60 SYSCTL_INT(_debug, OID_AUTO, to_avg_mpcalls, CTLFLAG_RD, &avg_mpcalls, 0,
61 "Average number of MP callouts made per softclock call. Units = 1/1000");
64 * allocate more timeout table slots when table overflows.
67 /* Exported to machdep.c and/or kern_clock.c. */
68 struct callout *callout;
69 struct callout_list callfree;
70 int callwheelsize, callwheelbits, callwheelmask;
71 struct callout_tailq *callwheel;
72 int softticks; /* Like ticks, but for softclock(). */
73 struct mtx callout_lock;
75 struct mtx dont_sleep_in_callout;
78 static struct callout *nextsoftcheck; /* Next callout to be checked. */
81 * Locked by callout_lock:
82 * curr_callout - If a callout is in progress, it is curr_callout.
83 * If curr_callout is non-NULL, threads waiting on
84 * callout_wait will be woken up as soon as the
85 * relevant callout completes.
86 * curr_cancelled - Changing to 1 with both callout_lock and c_mtx held
87 * guarantees that the current callout will not run.
88 * The softclock() function sets this to 0 before it
89 * drops callout_lock to acquire c_mtx, and it calls
90 * the handler only if curr_cancelled still 0 when
91 * c_mtx is successfully acquired.
92 * wakeup_ctr - Incremented every time a thread wants to wait
93 * for a callout to complete. Modified only when
94 * curr_callout is non-NULL.
95 * wakeup_needed - If a thread is waiting on callout_wait, then
96 * wakeup_needed is nonzero. Increased only when
97 * cutt_callout is non-NULL.
99 static struct callout *curr_callout;
100 static int curr_cancelled;
101 static int wakeup_ctr;
102 static int wakeup_needed;
105 * Locked by callout_wait_lock:
106 * callout_wait - If wakeup_needed is set, callout_wait will be
107 * triggered after the current callout finishes.
108 * wakeup_done_ctr - Set to the current value of wakeup_ctr after
109 * callout_wait is triggered.
111 static struct mtx callout_wait_lock;
112 static struct cv callout_wait;
113 static int wakeup_done_ctr;
116 * kern_timeout_callwheel_alloc() - kernel low level callwheel initialization
118 * This code is called very early in the kernel initialization sequence,
119 * and may be called more then once.
122 kern_timeout_callwheel_alloc(caddr_t v)
125 * Calculate callout wheel size
127 for (callwheelsize = 1, callwheelbits = 0;
128 callwheelsize < ncallout;
129 callwheelsize <<= 1, ++callwheelbits)
131 callwheelmask = callwheelsize - 1;
133 callout = (struct callout *)v;
134 v = (caddr_t)(callout + ncallout);
135 callwheel = (struct callout_tailq *)v;
136 v = (caddr_t)(callwheel + callwheelsize);
141 * kern_timeout_callwheel_init() - initialize previously reserved callwheel
144 * This code is called just once, after the space reserved for the
145 * callout wheel has been finalized.
148 kern_timeout_callwheel_init(void)
152 SLIST_INIT(&callfree);
153 for (i = 0; i < ncallout; i++) {
154 callout_init(&callout[i], 0);
155 callout[i].c_flags = CALLOUT_LOCAL_ALLOC;
156 SLIST_INSERT_HEAD(&callfree, &callout[i], c_links.sle);
158 for (i = 0; i < callwheelsize; i++) {
159 TAILQ_INIT(&callwheel[i]);
161 mtx_init(&callout_lock, "callout", NULL, MTX_SPIN | MTX_RECURSE);
163 mtx_init(&dont_sleep_in_callout, "dont_sleep_in_callout", NULL, MTX_DEF);
165 mtx_init(&callout_wait_lock, "callout_wait_lock", NULL, MTX_DEF);
166 cv_init(&callout_wait, "callout_wait");
170 * The callout mechanism is based on the work of Adam M. Costello and
171 * George Varghese, published in a technical report entitled "Redesigning
172 * the BSD Callout and Timer Facilities" and modified slightly for inclusion
173 * in FreeBSD by Justin T. Gibbs. The original work on the data structures
174 * used in this implementation was published by G. Varghese and T. Lauck in
175 * the paper "Hashed and Hierarchical Timing Wheels: Data Structures for
176 * the Efficient Implementation of a Timer Facility" in the Proceedings of
177 * the 11th ACM Annual Symposium on Operating Systems Principles,
178 * Austin, Texas Nov 1987.
182 * Software (low priority) clock interrupt.
183 * Run periodic events from timeout queue.
186 softclock(void *dummy)
189 struct callout_tailq *bucket;
191 int steps; /* #steps since we last allowed interrupts */
198 struct bintime bt1, bt2;
200 static uint64_t maxdt = 36893488147419102LL; /* 2 msec */
201 static timeout_t *lastfunc;
204 #ifndef MAX_SOFTCLOCK_STEPS
205 #define MAX_SOFTCLOCK_STEPS 100 /* Maximum allowed value of steps. */
206 #endif /* MAX_SOFTCLOCK_STEPS */
213 mtx_lock_spin(&callout_lock);
214 while (softticks != ticks) {
217 * softticks may be modified by hard clock, so cache
218 * it while we work on a given bucket.
220 curticks = softticks;
221 bucket = &callwheel[curticks & callwheelmask];
222 c = TAILQ_FIRST(bucket);
225 if (c->c_time != curticks) {
226 c = TAILQ_NEXT(c, c_links.tqe);
228 if (steps >= MAX_SOFTCLOCK_STEPS) {
230 /* Give interrupts a chance. */
231 mtx_unlock_spin(&callout_lock);
233 mtx_lock_spin(&callout_lock);
238 void (*c_func)(void *);
243 nextsoftcheck = TAILQ_NEXT(c, c_links.tqe);
244 TAILQ_REMOVE(bucket, c, c_links.tqe);
248 c_flags = c->c_flags;
249 if (c->c_flags & CALLOUT_LOCAL_ALLOC) {
251 c->c_flags = CALLOUT_LOCAL_ALLOC;
252 SLIST_INSERT_HEAD(&callfree, c,
257 (c->c_flags & ~CALLOUT_PENDING);
261 mtx_unlock_spin(&callout_lock);
265 * The callout may have been cancelled
266 * while we switched locks.
268 if (curr_cancelled) {
270 mtx_lock_spin(&callout_lock);
273 /* The callout cannot be stopped now. */
276 if (c_mtx == &Giant) {
278 CTR1(KTR_CALLOUT, "callout %p",
288 CTR1(KTR_CALLOUT, "callout mpsafe %p",
293 mtx_lock(&dont_sleep_in_callout);
297 mtx_unlock(&dont_sleep_in_callout);
299 bintime_sub(&bt2, &bt1);
300 if (bt2.frac > maxdt) {
301 if (lastfunc != c_func ||
302 bt2.frac > maxdt * 2) {
303 bintime2timespec(&bt2, &ts2);
305 "Expensive timeout(9) function: %p(%p) %jd.%09ld s\n",
307 (intmax_t)ts2.tv_sec,
314 if ((c_flags & CALLOUT_RETURNUNLOCKED) == 0)
316 mtx_lock_spin(&callout_lock);
321 * There might be someone waiting
322 * for the callout to complete.
324 wakeup_cookie = wakeup_ctr;
325 mtx_unlock_spin(&callout_lock);
326 mtx_lock(&callout_wait_lock);
327 cv_broadcast(&callout_wait);
328 wakeup_done_ctr = wakeup_cookie;
329 mtx_unlock(&callout_wait_lock);
330 mtx_lock_spin(&callout_lock);
338 avg_depth += (depth * 1000 - avg_depth) >> 8;
339 avg_mpcalls += (mpcalls * 1000 - avg_mpcalls) >> 8;
340 avg_mtxcalls += (mtxcalls * 1000 - avg_mtxcalls) >> 8;
341 avg_gcalls += (gcalls * 1000 - avg_gcalls) >> 8;
342 nextsoftcheck = NULL;
343 mtx_unlock_spin(&callout_lock);
348 * Execute a function after a specified length of time.
351 * Cancel previous timeout function call.
353 * callout_handle_init --
354 * Initialize a handle so that using it with untimeout is benign.
356 * See AT&T BCI Driver Reference Manual for specification. This
357 * implementation differs from that one in that although an
358 * identification value is returned from timeout, the original
359 * arguments to timeout as well as the identifier are used to
360 * identify entries for untimeout.
362 struct callout_handle
363 timeout(ftn, arg, to_ticks)
369 struct callout_handle handle;
371 mtx_lock_spin(&callout_lock);
373 /* Fill in the next free callout structure. */
374 new = SLIST_FIRST(&callfree);
376 /* XXX Attempt to malloc first */
377 panic("timeout table full");
378 SLIST_REMOVE_HEAD(&callfree, c_links.sle);
380 callout_reset(new, to_ticks, ftn, arg);
382 handle.callout = new;
383 mtx_unlock_spin(&callout_lock);
388 untimeout(ftn, arg, handle)
391 struct callout_handle handle;
395 * Check for a handle that was initialized
396 * by callout_handle_init, but never used
397 * for a real timeout.
399 if (handle.callout == NULL)
402 mtx_lock_spin(&callout_lock);
403 if (handle.callout->c_func == ftn && handle.callout->c_arg == arg)
404 callout_stop(handle.callout);
405 mtx_unlock_spin(&callout_lock);
409 callout_handle_init(struct callout_handle *handle)
411 handle->callout = NULL;
415 * New interface; clients allocate their own callout structures.
417 * callout_reset() - establish or change a timeout
418 * callout_stop() - disestablish a timeout
419 * callout_init() - initialize a callout structure so that it can
420 * safely be passed to callout_reset() and callout_stop()
422 * <sys/callout.h> defines three convenience macros:
424 * callout_active() - returns truth if callout has not been stopped,
425 * drained, or deactivated since the last time the callout was
427 * callout_pending() - returns truth if callout is still waiting for timeout
428 * callout_deactivate() - marks the callout as having been serviced
431 callout_reset(c, to_ticks, ftn, arg)
438 #ifdef notyet /* Some callers of timeout() do not hold Giant. */
439 if (c->c_mtx != NULL)
440 mtx_assert(c->c_mtx, MA_OWNED);
443 mtx_lock_spin(&callout_lock);
444 if (c == curr_callout) {
446 * We're being asked to reschedule a callout which is
447 * currently in progress. If there is a mutex then we
448 * can cancel the callout if it has not really started.
450 if (c->c_mtx != NULL && !curr_cancelled)
454 * Someone has called callout_drain to kill this
455 * callout. Don't reschedule.
457 mtx_unlock_spin(&callout_lock);
461 if (c->c_flags & CALLOUT_PENDING) {
462 if (nextsoftcheck == c) {
463 nextsoftcheck = TAILQ_NEXT(c, c_links.tqe);
465 TAILQ_REMOVE(&callwheel[c->c_time & callwheelmask], c,
469 * Part of the normal "stop a pending callout" process
470 * is to clear the CALLOUT_ACTIVE and CALLOUT_PENDING
471 * flags. We're not going to bother doing that here,
472 * because we're going to be setting those flags ten lines
473 * after this point, and we're holding callout_lock
474 * between now and then.
479 * We could unlock callout_lock here and lock it again before the
480 * TAILQ_INSERT_TAIL, but there's no point since doing this setup
481 * doesn't take much time.
487 c->c_flags |= (CALLOUT_ACTIVE | CALLOUT_PENDING);
489 c->c_time = ticks + to_ticks;
490 TAILQ_INSERT_TAIL(&callwheel[c->c_time & callwheelmask],
492 mtx_unlock_spin(&callout_lock);
496 _callout_stop_safe(c, safe)
500 int use_mtx, wakeup_cookie;
502 if (!safe && c->c_mtx != NULL) {
503 #ifdef notyet /* Some callers do not hold Giant for Giant-locked callouts. */
504 mtx_assert(c->c_mtx, MA_OWNED);
507 use_mtx = mtx_owned(c->c_mtx);
513 mtx_lock_spin(&callout_lock);
515 * Don't attempt to delete a callout that's not on the queue.
517 if (!(c->c_flags & CALLOUT_PENDING)) {
518 c->c_flags &= ~CALLOUT_ACTIVE;
519 if (c != curr_callout) {
520 mtx_unlock_spin(&callout_lock);
524 /* We need to wait until the callout is finished. */
526 wakeup_cookie = wakeup_ctr++;
527 mtx_unlock_spin(&callout_lock);
528 mtx_lock(&callout_wait_lock);
531 * Check to make sure that softclock() didn't
532 * do the wakeup in between our dropping
533 * callout_lock and picking up callout_wait_lock
535 if (wakeup_cookie - wakeup_done_ctr > 0)
536 cv_wait(&callout_wait, &callout_wait_lock);
538 mtx_unlock(&callout_wait_lock);
539 } else if (use_mtx && !curr_cancelled) {
540 /* We can stop the callout before it runs. */
542 mtx_unlock_spin(&callout_lock);
545 mtx_unlock_spin(&callout_lock);
548 c->c_flags &= ~(CALLOUT_ACTIVE | CALLOUT_PENDING);
550 if (nextsoftcheck == c) {
551 nextsoftcheck = TAILQ_NEXT(c, c_links.tqe);
553 TAILQ_REMOVE(&callwheel[c->c_time & callwheelmask], c, c_links.tqe);
555 if (c->c_flags & CALLOUT_LOCAL_ALLOC) {
557 SLIST_INSERT_HEAD(&callfree, c, c_links.sle);
559 mtx_unlock_spin(&callout_lock);
564 callout_init(c, mpsafe)
571 c->c_flags = CALLOUT_RETURNUNLOCKED;
579 callout_init_mtx(c, mtx, flags)
586 KASSERT((flags & ~CALLOUT_RETURNUNLOCKED) == 0,
587 ("callout_init_mtx: bad flags %d", flags));
588 /* CALLOUT_RETURNUNLOCKED makes no sense without a mutex. */
589 KASSERT(mtx != NULL || (flags & CALLOUT_RETURNUNLOCKED) == 0,
590 ("callout_init_mtx: CALLOUT_RETURNUNLOCKED with no mutex"));
591 c->c_flags = flags & CALLOUT_RETURNUNLOCKED;
594 #ifdef APM_FIXUP_CALLTODO
596 * Adjust the kernel calltodo timeout list. This routine is used after
597 * an APM resume to recalculate the calltodo timer list values with the
598 * number of hz's we have been sleeping. The next hardclock() will detect
599 * that there are fired timers and run softclock() to execute them.
601 * Please note, I have not done an exhaustive analysis of what code this
602 * might break. I am motivated to have my select()'s and alarm()'s that
603 * have expired during suspend firing upon resume so that the applications
604 * which set the timer can do the maintanence the timer was for as close
605 * as possible to the originally intended time. Testing this code for a
606 * week showed that resuming from a suspend resulted in 22 to 25 timers
607 * firing, which seemed independant on whether the suspend was 2 hours or
608 * 2 days. Your milage may vary. - Ken Key <key@cs.utk.edu>
611 adjust_timeout_calltodo(time_change)
612 struct timeval *time_change;
614 register struct callout *p;
615 unsigned long delta_ticks;
618 * How many ticks were we asleep?
619 * (stolen from tvtohz()).
622 /* Don't do anything */
623 if (time_change->tv_sec < 0)
625 else if (time_change->tv_sec <= LONG_MAX / 1000000)
626 delta_ticks = (time_change->tv_sec * 1000000 +
627 time_change->tv_usec + (tick - 1)) / tick + 1;
628 else if (time_change->tv_sec <= LONG_MAX / hz)
629 delta_ticks = time_change->tv_sec * hz +
630 (time_change->tv_usec + (tick - 1)) / tick + 1;
632 delta_ticks = LONG_MAX;
634 if (delta_ticks > INT_MAX)
635 delta_ticks = INT_MAX;
638 * Now rip through the timer calltodo list looking for timers
642 /* don't collide with softclock() */
643 mtx_lock_spin(&callout_lock);
644 for (p = calltodo.c_next; p != NULL; p = p->c_next) {
645 p->c_time -= delta_ticks;
647 /* Break if the timer had more time on it than delta_ticks */
651 /* take back the ticks the timer didn't use (p->c_time <= 0) */
652 delta_ticks = -p->c_time;
654 mtx_unlock_spin(&callout_lock);
658 #endif /* APM_FIXUP_CALLTODO */