kern_timeout.c revision 102936
1/*-
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.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
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 * 3. All advertising materials mentioning features or use of this software
19 *    must display the following acknowledgement:
20 *	This product includes software developed by the University of
21 *	California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 *    may be used to endorse or promote products derived from this software
24 *    without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * SUCH DAMAGE.
37 *
38 *	From: @(#)kern_clock.c	8.5 (Berkeley) 1/21/94
39 * $FreeBSD: head/sys/kern/kern_timeout.c 102936 2002-09-04 20:05:00Z phk $
40 */
41
42#include <sys/param.h>
43#include <sys/systm.h>
44#include <sys/callout.h>
45#include <sys/kernel.h>
46#include <sys/lock.h>
47#include <sys/mutex.h>
48
49/*
50 * TODO:
51 *	allocate more timeout table slots when table overflows.
52 */
53
54/* Exported to machdep.c and/or kern_clock.c.  */
55struct callout *callout;
56struct callout_list callfree;
57int callwheelsize, callwheelbits, callwheelmask;
58struct callout_tailq *callwheel;
59int softticks;			/* Like ticks, but for softclock(). */
60struct mtx callout_lock;
61
62static struct callout *nextsoftcheck;	/* Next callout to be checked. */
63
64/*
65 * kern_timeout_callwheel_alloc() - kernel low level callwheel initialization
66 *
67 *	This code is called very early in the kernel initialization sequence,
68 *	and may be called more then once.
69 */
70caddr_t
71kern_timeout_callwheel_alloc(caddr_t v)
72{
73	/*
74	 * Calculate callout wheel size
75	 */
76	for (callwheelsize = 1, callwheelbits = 0;
77	     callwheelsize < ncallout;
78	     callwheelsize <<= 1, ++callwheelbits)
79		;
80	callwheelmask = callwheelsize - 1;
81
82	callout = (struct callout *)v;
83	v = (caddr_t)(callout + ncallout);
84	callwheel = (struct callout_tailq *)v;
85	v = (caddr_t)(callwheel + callwheelsize);
86	return(v);
87}
88
89/*
90 * kern_timeout_callwheel_init() - initialize previously reserved callwheel
91 *				   space.
92 *
93 *	This code is called just once, after the space reserved for the
94 *	callout wheel has been finalized.
95 */
96void
97kern_timeout_callwheel_init(void)
98{
99	int i;
100
101	SLIST_INIT(&callfree);
102	for (i = 0; i < ncallout; i++) {
103		callout_init(&callout[i], 0);
104		callout[i].c_flags = CALLOUT_LOCAL_ALLOC;
105		SLIST_INSERT_HEAD(&callfree, &callout[i], c_links.sle);
106	}
107	for (i = 0; i < callwheelsize; i++) {
108		TAILQ_INIT(&callwheel[i]);
109	}
110	mtx_init(&callout_lock, "callout", NULL, MTX_SPIN | MTX_RECURSE);
111}
112
113/*
114 * The callout mechanism is based on the work of Adam M. Costello and
115 * George Varghese, published in a technical report entitled "Redesigning
116 * the BSD Callout and Timer Facilities" and modified slightly for inclusion
117 * in FreeBSD by Justin T. Gibbs.  The original work on the data structures
118 * used in this implementation was published by G.Varghese and A. Lauck in
119 * the paper "Hashed and Hierarchical Timing Wheels: Data Structures for
120 * the Efficient Implementation of a Timer Facility" in the Proceedings of
121 * the 11th ACM Annual Symposium on Operating Systems Principles,
122 * Austin, Texas Nov 1987.
123 */
124
125/*
126 * Software (low priority) clock interrupt.
127 * Run periodic events from timeout queue.
128 */
129void
130softclock(void *dummy)
131{
132	struct callout *c;
133	struct callout_tailq *bucket;
134	int curticks;
135	int steps;	/* #steps since we last allowed interrupts */
136#ifdef DIAGNOSTIC
137	struct bintime bt1, bt2;
138	struct timespec ts2;
139	static uint64_t maxdt = 18446744073709551LL;	/* 1 msec */
140#endif
141
142#ifndef MAX_SOFTCLOCK_STEPS
143#define MAX_SOFTCLOCK_STEPS 100 /* Maximum allowed value of steps. */
144#endif /* MAX_SOFTCLOCK_STEPS */
145
146	steps = 0;
147	mtx_lock_spin(&callout_lock);
148	while (softticks != ticks) {
149		softticks++;
150		/*
151		 * softticks may be modified by hard clock, so cache
152		 * it while we work on a given bucket.
153		 */
154		curticks = softticks;
155		bucket = &callwheel[curticks & callwheelmask];
156		c = TAILQ_FIRST(bucket);
157		while (c) {
158			if (c->c_time != curticks) {
159				c = TAILQ_NEXT(c, c_links.tqe);
160				++steps;
161				if (steps >= MAX_SOFTCLOCK_STEPS) {
162					nextsoftcheck = c;
163					/* Give interrupts a chance. */
164					mtx_unlock_spin(&callout_lock);
165					;	/* nothing */
166					mtx_lock_spin(&callout_lock);
167					c = nextsoftcheck;
168					steps = 0;
169				}
170			} else {
171				void (*c_func)(void *);
172				void *c_arg;
173				int c_flags;
174
175				nextsoftcheck = TAILQ_NEXT(c, c_links.tqe);
176				TAILQ_REMOVE(bucket, c, c_links.tqe);
177				c_func = c->c_func;
178				c_arg = c->c_arg;
179				c_flags = c->c_flags;
180				c->c_func = NULL;
181				if (c->c_flags & CALLOUT_LOCAL_ALLOC) {
182					c->c_flags = CALLOUT_LOCAL_ALLOC;
183					SLIST_INSERT_HEAD(&callfree, c,
184							  c_links.sle);
185				} else {
186					c->c_flags =
187					    (c->c_flags & ~CALLOUT_PENDING);
188				}
189				mtx_unlock_spin(&callout_lock);
190				if (!(c_flags & CALLOUT_MPSAFE))
191					mtx_lock(&Giant);
192#ifdef DIAGNOSTIC
193				binuptime(&bt1);
194#endif
195				c_func(c_arg);
196#ifdef DIAGNOSTIC
197				binuptime(&bt2);
198				bintime_sub(&bt2, &bt1);
199				if (bt2.frac > maxdt) {
200					bintime2timespec(&bt2, &ts2);
201					printf(
202			"Expensive timeout(9) function: %p(%p) %d.%09d\n",
203					c_func, c_arg,
204					ts2.tv_sec, ts2.tv_nsec);
205				}
206#endif
207				if (!(c_flags & CALLOUT_MPSAFE))
208					mtx_unlock(&Giant);
209				mtx_lock_spin(&callout_lock);
210				steps = 0;
211				c = nextsoftcheck;
212			}
213		}
214	}
215	nextsoftcheck = NULL;
216	mtx_unlock_spin(&callout_lock);
217}
218
219/*
220 * timeout --
221 *	Execute a function after a specified length of time.
222 *
223 * untimeout --
224 *	Cancel previous timeout function call.
225 *
226 * callout_handle_init --
227 *	Initialize a handle so that using it with untimeout is benign.
228 *
229 *	See AT&T BCI Driver Reference Manual for specification.  This
230 *	implementation differs from that one in that although an
231 *	identification value is returned from timeout, the original
232 *	arguments to timeout as well as the identifier are used to
233 *	identify entries for untimeout.
234 */
235struct callout_handle
236timeout(ftn, arg, to_ticks)
237	timeout_t *ftn;
238	void *arg;
239	int to_ticks;
240{
241	struct callout *new;
242	struct callout_handle handle;
243
244	mtx_lock_spin(&callout_lock);
245
246	/* Fill in the next free callout structure. */
247	new = SLIST_FIRST(&callfree);
248	if (new == NULL)
249		/* XXX Attempt to malloc first */
250		panic("timeout table full");
251	SLIST_REMOVE_HEAD(&callfree, c_links.sle);
252
253	callout_reset(new, to_ticks, ftn, arg);
254
255	handle.callout = new;
256	mtx_unlock_spin(&callout_lock);
257	return (handle);
258}
259
260void
261untimeout(ftn, arg, handle)
262	timeout_t *ftn;
263	void *arg;
264	struct callout_handle handle;
265{
266
267	/*
268	 * Check for a handle that was initialized
269	 * by callout_handle_init, but never used
270	 * for a real timeout.
271	 */
272	if (handle.callout == NULL)
273		return;
274
275	mtx_lock_spin(&callout_lock);
276	if (handle.callout->c_func == ftn && handle.callout->c_arg == arg)
277		callout_stop(handle.callout);
278	mtx_unlock_spin(&callout_lock);
279}
280
281void
282callout_handle_init(struct callout_handle *handle)
283{
284	handle->callout = NULL;
285}
286
287/*
288 * New interface; clients allocate their own callout structures.
289 *
290 * callout_reset() - establish or change a timeout
291 * callout_stop() - disestablish a timeout
292 * callout_init() - initialize a callout structure so that it can
293 *	safely be passed to callout_reset() and callout_stop()
294 *
295 * <sys/callout.h> defines three convenience macros:
296 *
297 * callout_active() - returns truth if callout has not been serviced
298 * callout_pending() - returns truth if callout is still waiting for timeout
299 * callout_deactivate() - marks the callout as having been serviced
300 */
301void
302callout_reset(c, to_ticks, ftn, arg)
303	struct	callout *c;
304	int	to_ticks;
305	void	(*ftn)(void *);
306	void	*arg;
307{
308
309	mtx_lock_spin(&callout_lock);
310	if (c->c_flags & CALLOUT_PENDING)
311		callout_stop(c);
312
313	/*
314	 * We could unlock callout_lock here and lock it again before the
315	 * TAILQ_INSERT_TAIL, but there's no point since doing this setup
316	 * doesn't take much time.
317	 */
318	if (to_ticks <= 0)
319		to_ticks = 1;
320
321	c->c_arg = arg;
322	c->c_flags |= (CALLOUT_ACTIVE | CALLOUT_PENDING);
323	c->c_func = ftn;
324	c->c_time = ticks + to_ticks;
325	TAILQ_INSERT_TAIL(&callwheel[c->c_time & callwheelmask],
326			  c, c_links.tqe);
327	mtx_unlock_spin(&callout_lock);
328}
329
330int
331callout_stop(c)
332	struct	callout *c;
333{
334
335	mtx_lock_spin(&callout_lock);
336	/*
337	 * Don't attempt to delete a callout that's not on the queue.
338	 */
339	if (!(c->c_flags & CALLOUT_PENDING)) {
340		c->c_flags &= ~CALLOUT_ACTIVE;
341		mtx_unlock_spin(&callout_lock);
342		return (0);
343	}
344	c->c_flags &= ~(CALLOUT_ACTIVE | CALLOUT_PENDING);
345
346	if (nextsoftcheck == c) {
347		nextsoftcheck = TAILQ_NEXT(c, c_links.tqe);
348	}
349	TAILQ_REMOVE(&callwheel[c->c_time & callwheelmask], c, c_links.tqe);
350	c->c_func = NULL;
351
352	if (c->c_flags & CALLOUT_LOCAL_ALLOC) {
353		SLIST_INSERT_HEAD(&callfree, c, c_links.sle);
354	}
355	mtx_unlock_spin(&callout_lock);
356	return (1);
357}
358
359void
360callout_init(c, mpsafe)
361	struct	callout *c;
362	int mpsafe;
363{
364	bzero(c, sizeof *c);
365	if (mpsafe)
366		c->c_flags |= CALLOUT_MPSAFE;
367}
368
369#ifdef APM_FIXUP_CALLTODO
370/*
371 * Adjust the kernel calltodo timeout list.  This routine is used after
372 * an APM resume to recalculate the calltodo timer list values with the
373 * number of hz's we have been sleeping.  The next hardclock() will detect
374 * that there are fired timers and run softclock() to execute them.
375 *
376 * Please note, I have not done an exhaustive analysis of what code this
377 * might break.  I am motivated to have my select()'s and alarm()'s that
378 * have expired during suspend firing upon resume so that the applications
379 * which set the timer can do the maintanence the timer was for as close
380 * as possible to the originally intended time.  Testing this code for a
381 * week showed that resuming from a suspend resulted in 22 to 25 timers
382 * firing, which seemed independant on whether the suspend was 2 hours or
383 * 2 days.  Your milage may vary.   - Ken Key <key@cs.utk.edu>
384 */
385void
386adjust_timeout_calltodo(time_change)
387    struct timeval *time_change;
388{
389	register struct callout *p;
390	unsigned long delta_ticks;
391
392	/*
393	 * How many ticks were we asleep?
394	 * (stolen from tvtohz()).
395	 */
396
397	/* Don't do anything */
398	if (time_change->tv_sec < 0)
399		return;
400	else if (time_change->tv_sec <= LONG_MAX / 1000000)
401		delta_ticks = (time_change->tv_sec * 1000000 +
402			       time_change->tv_usec + (tick - 1)) / tick + 1;
403	else if (time_change->tv_sec <= LONG_MAX / hz)
404		delta_ticks = time_change->tv_sec * hz +
405			      (time_change->tv_usec + (tick - 1)) / tick + 1;
406	else
407		delta_ticks = LONG_MAX;
408
409	if (delta_ticks > INT_MAX)
410		delta_ticks = INT_MAX;
411
412	/*
413	 * Now rip through the timer calltodo list looking for timers
414	 * to expire.
415	 */
416
417	/* don't collide with softclock() */
418	mtx_lock_spin(&callout_lock);
419	for (p = calltodo.c_next; p != NULL; p = p->c_next) {
420		p->c_time -= delta_ticks;
421
422		/* Break if the timer had more time on it than delta_ticks */
423		if (p->c_time > 0)
424			break;
425
426		/* take back the ticks the timer didn't use (p->c_time <= 0) */
427		delta_ticks = -p->c_time;
428	}
429	mtx_unlock_spin(&callout_lock);
430
431	return;
432}
433#endif /* APM_FIXUP_CALLTODO */
434