xref: /netbsd-src/sys/kern/kern_timeout.c (revision f983e71d70cfccf7b3de601eb4d998b2d886ede4)
1 /*	$NetBSD: kern_timeout.c,v 1.37 2008/04/22 12:04:22 ad Exp $	*/
2 
3 /*-
4  * Copyright (c) 2003, 2006, 2007, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe, and by Andrew Doran.
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 NetBSD
21  *	Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 2001 Thomas Nordin <nordin@openbsd.org>
41  * Copyright (c) 2000-2001 Artur Grabowski <art@openbsd.org>
42  * All rights reserved.
43  *
44  * Redistribution and use in source and binary forms, with or without
45  * modification, are permitted provided that the following conditions
46  * are met:
47  *
48  * 1. Redistributions of source code must retain the above copyright
49  *    notice, this list of conditions and the following disclaimer.
50  * 2. Redistributions in binary form must reproduce the above copyright
51  *    notice, this list of conditions and the following disclaimer in the
52  *    documentation and/or other materials provided with the distribution.
53  * 3. The name of the author may not be used to endorse or promote products
54  *    derived from this software without specific prior written permission.
55  *
56  * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES,
57  * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY
58  * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL
59  * THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
60  * EXEMPLARY, OR CONSEQUENTIAL  DAMAGES (INCLUDING, BUT NOT LIMITED TO,
61  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
62  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
63  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
64  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
65  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
66  */
67 
68 #include <sys/cdefs.h>
69 __KERNEL_RCSID(0, "$NetBSD: kern_timeout.c,v 1.37 2008/04/22 12:04:22 ad Exp $");
70 
71 /*
72  * Timeouts are kept in a hierarchical timing wheel.  The c_time is the
73  * value of c_cpu->cc_ticks when the timeout should be called.  There are
74  * four levels with 256 buckets each. See 'Scheme 7' in "Hashed and
75  * Hierarchical Timing Wheels: Efficient Data Structures for Implementing
76  * a Timer Facility" by George Varghese and Tony Lauck.
77  *
78  * Some of the "math" in here is a bit tricky.  We have to beware of
79  * wrapping ints.
80  *
81  * We use the fact that any element added to the queue must be added with
82  * a positive time.  That means that any element `to' on the queue cannot
83  * be scheduled to timeout further in time than INT_MAX, but c->c_time can
84  * be positive or negative so comparing it with anything is dangerous.
85  * The only way we can use the c->c_time value in any predictable way is
86  * when we calculate how far in the future `to' will timeout - "c->c_time
87  * - c->c_cpu->cc_ticks".  The result will always be positive for future
88  * timeouts and 0 or negative for due timeouts.
89  */
90 
91 #define	_CALLOUT_PRIVATE
92 
93 #include <sys/param.h>
94 #include <sys/systm.h>
95 #include <sys/kernel.h>
96 #include <sys/callout.h>
97 #include <sys/mutex.h>
98 #include <sys/proc.h>
99 #include <sys/sleepq.h>
100 #include <sys/syncobj.h>
101 #include <sys/evcnt.h>
102 #include <sys/intr.h>
103 #include <sys/cpu.h>
104 #include <sys/kmem.h>
105 
106 #ifdef DDB
107 #include <machine/db_machdep.h>
108 #include <ddb/db_interface.h>
109 #include <ddb/db_access.h>
110 #include <ddb/db_sym.h>
111 #include <ddb/db_output.h>
112 #endif
113 
114 #define BUCKETS		1024
115 #define WHEELSIZE	256
116 #define WHEELMASK	255
117 #define WHEELBITS	8
118 
119 #define MASKWHEEL(wheel, time) (((time) >> ((wheel)*WHEELBITS)) & WHEELMASK)
120 
121 #define BUCKET(cc, rel, abs)						\
122     (((rel) <= (1 << (2*WHEELBITS)))					\
123     	? ((rel) <= (1 << WHEELBITS))					\
124             ? &(cc)->cc_wheel[MASKWHEEL(0, (abs))]			\
125             : &(cc)->cc_wheel[MASKWHEEL(1, (abs)) + WHEELSIZE]		\
126         : ((rel) <= (1 << (3*WHEELBITS)))				\
127             ? &(cc)->cc_wheel[MASKWHEEL(2, (abs)) + 2*WHEELSIZE]	\
128             : &(cc)->cc_wheel[MASKWHEEL(3, (abs)) + 3*WHEELSIZE])
129 
130 #define MOVEBUCKET(cc, wheel, time)					\
131     CIRCQ_APPEND(&(cc)->cc_todo,					\
132         &(cc)->cc_wheel[MASKWHEEL((wheel), (time)) + (wheel)*WHEELSIZE])
133 
134 /*
135  * Circular queue definitions.
136  */
137 
138 #define CIRCQ_INIT(list)						\
139 do {									\
140         (list)->cq_next_l = (list);					\
141         (list)->cq_prev_l = (list);					\
142 } while (/*CONSTCOND*/0)
143 
144 #define CIRCQ_INSERT(elem, list)					\
145 do {									\
146         (elem)->cq_prev_e = (list)->cq_prev_e;				\
147         (elem)->cq_next_l = (list);					\
148         (list)->cq_prev_l->cq_next_l = (elem);				\
149         (list)->cq_prev_l = (elem);					\
150 } while (/*CONSTCOND*/0)
151 
152 #define CIRCQ_APPEND(fst, snd)						\
153 do {									\
154         if (!CIRCQ_EMPTY(snd)) {					\
155                 (fst)->cq_prev_l->cq_next_l = (snd)->cq_next_l;		\
156                 (snd)->cq_next_l->cq_prev_l = (fst)->cq_prev_l;		\
157                 (snd)->cq_prev_l->cq_next_l = (fst);			\
158                 (fst)->cq_prev_l = (snd)->cq_prev_l;			\
159                 CIRCQ_INIT(snd);					\
160         }								\
161 } while (/*CONSTCOND*/0)
162 
163 #define CIRCQ_REMOVE(elem)						\
164 do {									\
165         (elem)->cq_next_l->cq_prev_e = (elem)->cq_prev_e;		\
166         (elem)->cq_prev_l->cq_next_e = (elem)->cq_next_e;		\
167 } while (/*CONSTCOND*/0)
168 
169 #define CIRCQ_FIRST(list)	((list)->cq_next_e)
170 #define CIRCQ_NEXT(elem)	((elem)->cq_next_e)
171 #define CIRCQ_LAST(elem,list)	((elem)->cq_next_l == (list))
172 #define CIRCQ_EMPTY(list)	((list)->cq_next_l == (list))
173 
174 static void	callout_softclock(void *);
175 
176 struct callout_cpu {
177 	kmutex_t	cc_lock;
178 	sleepq_t	cc_sleepq;
179 	u_int		cc_nwait;
180 	u_int		cc_ticks;
181 	lwp_t		*cc_lwp;
182 	callout_impl_t	*cc_active;
183 	callout_impl_t	*cc_cancel;
184 	struct evcnt	cc_ev_late;
185 	struct evcnt	cc_ev_block;
186 	struct callout_circq cc_todo;		/* Worklist */
187 	struct callout_circq cc_wheel[BUCKETS];	/* Queues of timeouts */
188 	char		cc_name1[12];
189 	char		cc_name2[12];
190 };
191 
192 static struct callout_cpu callout_cpu0;
193 static void *callout_sih;
194 
195 static inline kmutex_t *
196 callout_lock(callout_impl_t *c)
197 {
198 	kmutex_t *lock;
199 
200 	for (;;) {
201 		lock = &c->c_cpu->cc_lock;
202 		mutex_spin_enter(lock);
203 		if (__predict_true(lock == &c->c_cpu->cc_lock))
204 			return lock;
205 		mutex_spin_exit(lock);
206 	}
207 }
208 
209 /*
210  * callout_startup:
211  *
212  *	Initialize the callout facility, called at system startup time.
213  *	Do just enough to allow callouts to be safely registered.
214  */
215 void
216 callout_startup(void)
217 {
218 	struct callout_cpu *cc;
219 	int b;
220 
221 	KASSERT(curcpu()->ci_data.cpu_callout == NULL);
222 
223 	cc = &callout_cpu0;
224 	mutex_init(&cc->cc_lock, MUTEX_DEFAULT, IPL_SCHED);
225 	CIRCQ_INIT(&cc->cc_todo);
226 	for (b = 0; b < BUCKETS; b++)
227 		CIRCQ_INIT(&cc->cc_wheel[b]);
228 	curcpu()->ci_data.cpu_callout = cc;
229 }
230 
231 /*
232  * callout_init_cpu:
233  *
234  *	Per-CPU initialization.
235  */
236 void
237 callout_init_cpu(struct cpu_info *ci)
238 {
239 	struct callout_cpu *cc;
240 	int b;
241 
242 	KASSERT(sizeof(callout_impl_t) <= sizeof(callout_t));
243 
244 	if ((cc = ci->ci_data.cpu_callout) == NULL) {
245 		cc = kmem_zalloc(sizeof(*cc), KM_SLEEP);
246 		if (cc == NULL)
247 			panic("callout_init_cpu (1)");
248 		mutex_init(&cc->cc_lock, MUTEX_DEFAULT, IPL_SCHED);
249 		CIRCQ_INIT(&cc->cc_todo);
250 		for (b = 0; b < BUCKETS; b++)
251 			CIRCQ_INIT(&cc->cc_wheel[b]);
252 	} else {
253 		/* Boot CPU, one time only. */
254 		callout_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE,
255 		    callout_softclock, NULL);
256 		if (callout_sih == NULL)
257 			panic("callout_init_cpu (2)");
258 	}
259 
260 	sleepq_init(&cc->cc_sleepq, &cc->cc_lock);
261 
262 	snprintf(cc->cc_name1, sizeof(cc->cc_name1), "late/%u",
263 	    cpu_index(ci));
264 	evcnt_attach_dynamic(&cc->cc_ev_late, EVCNT_TYPE_MISC,
265 	    NULL, "callout", cc->cc_name1);
266 
267 	snprintf(cc->cc_name2, sizeof(cc->cc_name2), "wait/%u",
268 	    cpu_index(ci));
269 	evcnt_attach_dynamic(&cc->cc_ev_block, EVCNT_TYPE_MISC,
270 	    NULL, "callout", cc->cc_name2);
271 
272 	ci->ci_data.cpu_callout = cc;
273 }
274 
275 /*
276  * callout_init:
277  *
278  *	Initialize a callout structure.  This must be quick, so we fill
279  *	only the minimum number of fields.
280  */
281 void
282 callout_init(callout_t *cs, u_int flags)
283 {
284 	callout_impl_t *c = (callout_impl_t *)cs;
285 	struct callout_cpu *cc;
286 
287 	KASSERT((flags & ~CALLOUT_FLAGMASK) == 0);
288 
289 	cc = curcpu()->ci_data.cpu_callout;
290 	c->c_func = NULL;
291 	c->c_magic = CALLOUT_MAGIC;
292 	if (__predict_true((flags & CALLOUT_MPSAFE) != 0 && cc != NULL)) {
293 		c->c_flags = flags;
294 		c->c_cpu = cc;
295 		return;
296 	}
297 	c->c_flags = flags | CALLOUT_BOUND;
298 	c->c_cpu = &callout_cpu0;
299 }
300 
301 /*
302  * callout_destroy:
303  *
304  *	Destroy a callout structure.  The callout must be stopped.
305  */
306 void
307 callout_destroy(callout_t *cs)
308 {
309 	callout_impl_t *c = (callout_impl_t *)cs;
310 
311 	/*
312 	 * It's not necessary to lock in order to see the correct value
313 	 * of c->c_flags.  If the callout could potentially have been
314 	 * running, the current thread should have stopped it.
315 	 */
316 	KASSERT((c->c_flags & CALLOUT_PENDING) == 0);
317 	KASSERT(c->c_cpu->cc_lwp == curlwp || c->c_cpu->cc_active != c);
318 	KASSERT(c->c_magic == CALLOUT_MAGIC);
319 	c->c_magic = 0;
320 }
321 
322 /*
323  * callout_schedule_locked:
324  *
325  *	Schedule a callout to run.  The function and argument must
326  *	already be set in the callout structure.  Must be called with
327  *	callout_lock.
328  */
329 static void
330 callout_schedule_locked(callout_impl_t *c, kmutex_t *lock, int to_ticks)
331 {
332 	struct callout_cpu *cc, *occ;
333 	int old_time;
334 
335 	KASSERT(to_ticks >= 0);
336 	KASSERT(c->c_func != NULL);
337 
338 	/* Initialize the time here, it won't change. */
339 	occ = c->c_cpu;
340 	c->c_flags &= ~CALLOUT_FIRED;
341 
342 	/*
343 	 * If this timeout is already scheduled and now is moved
344 	 * earlier, reschedule it now.  Otherwise leave it in place
345 	 * and let it be rescheduled later.
346 	 */
347 	if ((c->c_flags & CALLOUT_PENDING) != 0) {
348 		/* Leave on existing CPU. */
349 		old_time = c->c_time;
350 		c->c_time = to_ticks + occ->cc_ticks;
351 		if (c->c_time - old_time < 0) {
352 			CIRCQ_REMOVE(&c->c_list);
353 			CIRCQ_INSERT(&c->c_list, &occ->cc_todo);
354 		}
355 		mutex_spin_exit(lock);
356 		return;
357 	}
358 
359 	cc = curcpu()->ci_data.cpu_callout;
360 	if ((c->c_flags & CALLOUT_BOUND) != 0 || cc == occ ||
361 	    !mutex_tryenter(&cc->cc_lock)) {
362 		/* Leave on existing CPU. */
363 		c->c_time = to_ticks + occ->cc_ticks;
364 		c->c_flags |= CALLOUT_PENDING;
365 		CIRCQ_INSERT(&c->c_list, &occ->cc_todo);
366 	} else {
367 		/* Move to this CPU. */
368 		c->c_cpu = cc;
369 		c->c_time = to_ticks + cc->cc_ticks;
370 		c->c_flags |= CALLOUT_PENDING;
371 		CIRCQ_INSERT(&c->c_list, &cc->cc_todo);
372 		mutex_spin_exit(&cc->cc_lock);
373 	}
374 	mutex_spin_exit(lock);
375 }
376 
377 /*
378  * callout_reset:
379  *
380  *	Reset a callout structure with a new function and argument, and
381  *	schedule it to run.
382  */
383 void
384 callout_reset(callout_t *cs, int to_ticks, void (*func)(void *), void *arg)
385 {
386 	callout_impl_t *c = (callout_impl_t *)cs;
387 	kmutex_t *lock;
388 
389 	KASSERT(c->c_magic == CALLOUT_MAGIC);
390 
391 	lock = callout_lock(c);
392 	c->c_func = func;
393 	c->c_arg = arg;
394 	callout_schedule_locked(c, lock, to_ticks);
395 }
396 
397 /*
398  * callout_schedule:
399  *
400  *	Schedule a callout to run.  The function and argument must
401  *	already be set in the callout structure.
402  */
403 void
404 callout_schedule(callout_t *cs, int to_ticks)
405 {
406 	callout_impl_t *c = (callout_impl_t *)cs;
407 	kmutex_t *lock;
408 
409 	KASSERT(c->c_magic == CALLOUT_MAGIC);
410 
411 	lock = callout_lock(c);
412 	callout_schedule_locked(c, lock, to_ticks);
413 }
414 
415 /*
416  * callout_stop:
417  *
418  *	Try to cancel a pending callout.  It may be too late: the callout
419  *	could be running on another CPU.  If called from interrupt context,
420  *	the callout could already be in progress at a lower priority.
421  */
422 bool
423 callout_stop(callout_t *cs)
424 {
425 	callout_impl_t *c = (callout_impl_t *)cs;
426 	struct callout_cpu *cc;
427 	kmutex_t *lock;
428 	bool expired;
429 
430 	KASSERT(c->c_magic == CALLOUT_MAGIC);
431 
432 	lock = callout_lock(c);
433 
434 	if ((c->c_flags & CALLOUT_PENDING) != 0)
435 		CIRCQ_REMOVE(&c->c_list);
436 	expired = ((c->c_flags & CALLOUT_FIRED) != 0);
437 	c->c_flags &= ~(CALLOUT_PENDING|CALLOUT_FIRED);
438 
439 	cc = c->c_cpu;
440 	if (cc->cc_active == c) {
441 		/*
442 		 * This is for non-MPSAFE callouts only.  To synchronize
443 		 * effectively we must be called with kernel_lock held.
444 		 * It's also taken in callout_softclock.
445 		 */
446 		cc->cc_cancel = c;
447 	}
448 
449 	mutex_spin_exit(lock);
450 
451 	return expired;
452 }
453 
454 /*
455  * callout_halt:
456  *
457  *	Cancel a pending callout.  If in-flight, block until it completes.
458  *	May not be called from a hard interrupt handler.  If the callout
459  * 	can take locks, the caller of callout_halt() must not hold any of
460  *	those locks, otherwise the two could deadlock.  If 'interlock' is
461  *	non-NULL and we must wait for the callout to complete, it will be
462  *	released and re-acquired before returning.
463  */
464 bool
465 callout_halt(callout_t *cs, kmutex_t *interlock)
466 {
467 	callout_impl_t *c = (callout_impl_t *)cs;
468 	struct callout_cpu *cc;
469 	struct lwp *l;
470 	kmutex_t *lock, *relock;
471 	bool expired;
472 
473 	KASSERT(c->c_magic == CALLOUT_MAGIC);
474 	KASSERT(!cpu_intr_p());
475 
476 	lock = callout_lock(c);
477 	relock = NULL;
478 
479 	expired = ((c->c_flags & CALLOUT_FIRED) != 0);
480 	if ((c->c_flags & CALLOUT_PENDING) != 0)
481 		CIRCQ_REMOVE(&c->c_list);
482 	c->c_flags &= ~(CALLOUT_PENDING|CALLOUT_FIRED);
483 
484 	l = curlwp;
485 	for (;;) {
486 		cc = c->c_cpu;
487 		if (__predict_true(cc->cc_active != c || cc->cc_lwp == l))
488 			break;
489 		if (interlock != NULL) {
490 			/*
491 			 * Avoid potential scheduler lock order problems by
492 			 * dropping the interlock without the callout lock
493 			 * held.
494 			 */
495 			mutex_spin_exit(lock);
496 			mutex_exit(interlock);
497 			relock = interlock;
498 			interlock = NULL;
499 		} else {
500 			/* XXX Better to do priority inheritance. */
501 			KASSERT(l->l_wchan == NULL);
502 			cc->cc_nwait++;
503 			cc->cc_ev_block.ev_count++;
504 			l->l_kpriority = true;
505 			sleepq_enter(&cc->cc_sleepq, l);
506 			sleepq_enqueue(&cc->cc_sleepq, cc, "callout",
507 			    &sleep_syncobj);
508 			KERNEL_UNLOCK_ALL(l, &l->l_biglocks);
509 			sleepq_block(0, false);
510 		}
511 		lock = callout_lock(c);
512 	}
513 
514 	mutex_spin_exit(lock);
515 	if (__predict_false(relock != NULL))
516 		mutex_enter(relock);
517 
518 	return expired;
519 }
520 
521 #ifdef notyet
522 /*
523  * callout_bind:
524  *
525  *	Bind a callout so that it will only execute on one CPU.
526  *	The callout must be stopped, and must be MPSAFE.
527  *
528  *	XXX Disabled for now until it is decided how to handle
529  *	offlined CPUs.  We may want weak+strong binding.
530  */
531 void
532 callout_bind(callout_t *cs, struct cpu_info *ci)
533 {
534 	callout_impl_t *c = (callout_impl_t *)cs;
535 	struct callout_cpu *cc;
536 	kmutex_t *lock;
537 
538 	KASSERT((c->c_flags & CALLOUT_PENDING) == 0);
539 	KASSERT(c->c_cpu->cc_active != c);
540 	KASSERT(c->c_magic == CALLOUT_MAGIC);
541 	KASSERT((c->c_flags & CALLOUT_MPSAFE) != 0);
542 
543 	lock = callout_lock(c);
544 	cc = ci->ci_data.cpu_callout;
545 	c->c_flags |= CALLOUT_BOUND;
546 	if (c->c_cpu != cc) {
547 		/*
548 		 * Assigning c_cpu effectively unlocks the callout
549 		 * structure, as we don't hold the new CPU's lock.
550 		 * Issue memory barrier to prevent accesses being
551 		 * reordered.
552 		 */
553 		membar_exit();
554 		c->c_cpu = cc;
555 	}
556 	mutex_spin_exit(lock);
557 }
558 #endif
559 
560 void
561 callout_setfunc(callout_t *cs, void (*func)(void *), void *arg)
562 {
563 	callout_impl_t *c = (callout_impl_t *)cs;
564 	kmutex_t *lock;
565 
566 	KASSERT(c->c_magic == CALLOUT_MAGIC);
567 
568 	lock = callout_lock(c);
569 	c->c_func = func;
570 	c->c_arg = arg;
571 	mutex_spin_exit(lock);
572 }
573 
574 bool
575 callout_expired(callout_t *cs)
576 {
577 	callout_impl_t *c = (callout_impl_t *)cs;
578 	kmutex_t *lock;
579 	bool rv;
580 
581 	KASSERT(c->c_magic == CALLOUT_MAGIC);
582 
583 	lock = callout_lock(c);
584 	rv = ((c->c_flags & CALLOUT_FIRED) != 0);
585 	mutex_spin_exit(lock);
586 
587 	return rv;
588 }
589 
590 bool
591 callout_active(callout_t *cs)
592 {
593 	callout_impl_t *c = (callout_impl_t *)cs;
594 	kmutex_t *lock;
595 	bool rv;
596 
597 	KASSERT(c->c_magic == CALLOUT_MAGIC);
598 
599 	lock = callout_lock(c);
600 	rv = ((c->c_flags & (CALLOUT_PENDING|CALLOUT_FIRED)) != 0);
601 	mutex_spin_exit(lock);
602 
603 	return rv;
604 }
605 
606 bool
607 callout_pending(callout_t *cs)
608 {
609 	callout_impl_t *c = (callout_impl_t *)cs;
610 	kmutex_t *lock;
611 	bool rv;
612 
613 	KASSERT(c->c_magic == CALLOUT_MAGIC);
614 
615 	lock = callout_lock(c);
616 	rv = ((c->c_flags & CALLOUT_PENDING) != 0);
617 	mutex_spin_exit(lock);
618 
619 	return rv;
620 }
621 
622 bool
623 callout_invoking(callout_t *cs)
624 {
625 	callout_impl_t *c = (callout_impl_t *)cs;
626 	kmutex_t *lock;
627 	bool rv;
628 
629 	KASSERT(c->c_magic == CALLOUT_MAGIC);
630 
631 	lock = callout_lock(c);
632 	rv = ((c->c_flags & CALLOUT_INVOKING) != 0);
633 	mutex_spin_exit(lock);
634 
635 	return rv;
636 }
637 
638 void
639 callout_ack(callout_t *cs)
640 {
641 	callout_impl_t *c = (callout_impl_t *)cs;
642 	kmutex_t *lock;
643 
644 	KASSERT(c->c_magic == CALLOUT_MAGIC);
645 
646 	lock = callout_lock(c);
647 	c->c_flags &= ~CALLOUT_INVOKING;
648 	mutex_spin_exit(lock);
649 }
650 
651 /*
652  * callout_hardclock:
653  *
654  *	Called from hardclock() once every tick.  We schedule a soft
655  *	interrupt if there is work to be done.
656  */
657 void
658 callout_hardclock(void)
659 {
660 	struct callout_cpu *cc;
661 	int needsoftclock, ticks;
662 
663 	cc = curcpu()->ci_data.cpu_callout;
664 	mutex_spin_enter(&cc->cc_lock);
665 
666 	ticks = ++cc->cc_ticks;
667 
668 	MOVEBUCKET(cc, 0, ticks);
669 	if (MASKWHEEL(0, ticks) == 0) {
670 		MOVEBUCKET(cc, 1, ticks);
671 		if (MASKWHEEL(1, ticks) == 0) {
672 			MOVEBUCKET(cc, 2, ticks);
673 			if (MASKWHEEL(2, ticks) == 0)
674 				MOVEBUCKET(cc, 3, ticks);
675 		}
676 	}
677 
678 	needsoftclock = !CIRCQ_EMPTY(&cc->cc_todo);
679 	mutex_spin_exit(&cc->cc_lock);
680 
681 	if (needsoftclock)
682 		softint_schedule(callout_sih);
683 }
684 
685 /*
686  * callout_softclock:
687  *
688  *	Soft interrupt handler, scheduled above if there is work to
689  * 	be done.  Callouts are made in soft interrupt context.
690  */
691 static void
692 callout_softclock(void *v)
693 {
694 	callout_impl_t *c;
695 	struct callout_cpu *cc;
696 	void (*func)(void *);
697 	void *arg;
698 	int mpsafe, count, ticks, delta;
699 	lwp_t *l;
700 
701 	l = curlwp;
702 	KASSERT(l->l_cpu == curcpu());
703 	cc = l->l_cpu->ci_data.cpu_callout;
704 
705 	mutex_spin_enter(&cc->cc_lock);
706 	cc->cc_lwp = l;
707 	while (!CIRCQ_EMPTY(&cc->cc_todo)) {
708 		c = CIRCQ_FIRST(&cc->cc_todo);
709 		KASSERT(c->c_magic == CALLOUT_MAGIC);
710 		KASSERT(c->c_func != NULL);
711 		KASSERT(c->c_cpu == cc);
712 		KASSERT((c->c_flags & CALLOUT_PENDING) != 0);
713 		KASSERT((c->c_flags & CALLOUT_FIRED) == 0);
714 		CIRCQ_REMOVE(&c->c_list);
715 
716 		/* If due run it, otherwise insert it into the right bucket. */
717 		ticks = cc->cc_ticks;
718 		delta = c->c_time - ticks;
719 		if (delta > 0) {
720 			CIRCQ_INSERT(&c->c_list, BUCKET(cc, delta, c->c_time));
721 			continue;
722 		}
723 		if (delta < 0)
724 			cc->cc_ev_late.ev_count++;
725 
726 		c->c_flags ^= (CALLOUT_PENDING | CALLOUT_FIRED);
727 		mpsafe = (c->c_flags & CALLOUT_MPSAFE);
728 		func = c->c_func;
729 		arg = c->c_arg;
730 		cc->cc_active = c;
731 
732 		mutex_spin_exit(&cc->cc_lock);
733 		if (!mpsafe) {
734 			KERNEL_LOCK(1, NULL);
735 			(*func)(arg);
736 			KERNEL_UNLOCK_ONE(NULL);
737 		} else
738 			(*func)(arg);
739 		mutex_spin_enter(&cc->cc_lock);
740 
741 		/*
742 		 * We can't touch 'c' here because it might be
743 		 * freed already.  If LWPs waiting for callout
744 		 * to complete, awaken them.
745 		 */
746 		cc->cc_active = NULL;
747 		if ((count = cc->cc_nwait) != 0) {
748 			cc->cc_nwait = 0;
749 			/* sleepq_wake() drops the lock. */
750 			sleepq_wake(&cc->cc_sleepq, cc, count);
751 			mutex_spin_enter(&cc->cc_lock);
752 		}
753 	}
754 	cc->cc_lwp = NULL;
755 	mutex_spin_exit(&cc->cc_lock);
756 }
757 
758 #ifdef DDB
759 static void
760 db_show_callout_bucket(struct callout_cpu *cc, struct callout_circq *bucket)
761 {
762 	callout_impl_t *c;
763 	db_expr_t offset;
764 	const char *name;
765 	static char question[] = "?";
766 	int b;
767 
768 	if (CIRCQ_EMPTY(bucket))
769 		return;
770 
771 	for (c = CIRCQ_FIRST(bucket); /*nothing*/; c = CIRCQ_NEXT(&c->c_list)) {
772 		db_find_sym_and_offset((db_addr_t)(intptr_t)c->c_func, &name,
773 		    &offset);
774 		name = name ? name : question;
775 		b = (bucket - cc->cc_wheel);
776 		if (b < 0)
777 			b = -WHEELSIZE;
778 		db_printf("%9d %2d/%-4d %16lx  %s\n",
779 		    c->c_time - cc->cc_ticks, b / WHEELSIZE, b,
780 		    (u_long)c->c_arg, name);
781 		if (CIRCQ_LAST(&c->c_list, bucket))
782 			break;
783 	}
784 }
785 
786 void
787 db_show_callout(db_expr_t addr, bool haddr, db_expr_t count, const char *modif)
788 {
789 	CPU_INFO_ITERATOR cii;
790 	struct callout_cpu *cc;
791 	struct cpu_info *ci;
792 	int b;
793 
794 	db_printf("hardclock_ticks now: %d\n", hardclock_ticks);
795 	db_printf("    ticks  wheel               arg  func\n");
796 
797 	/*
798 	 * Don't lock the callwheel; all the other CPUs are paused
799 	 * anyhow, and we might be called in a circumstance where
800 	 * some other CPU was paused while holding the lock.
801 	 */
802 	for (CPU_INFO_FOREACH(cii, ci)) {
803 		cc = ci->ci_data.cpu_callout;
804 		db_show_callout_bucket(cc, &cc->cc_todo);
805 	}
806 	for (b = 0; b < BUCKETS; b++) {
807 		for (CPU_INFO_FOREACH(cii, ci)) {
808 			cc = ci->ci_data.cpu_callout;
809 			db_show_callout_bucket(cc, &cc->cc_wheel[b]);
810 		}
811 	}
812 }
813 #endif /* DDB */
814