1/* $NetBSD: kern_timeout.c,v 1.51 2015/11/24 15:48:23 christos Exp $ */
2
3/*-
4 * Copyright (c) 2003, 2006, 2007, 2008, 2009 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 *
19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
30 */
31
32/*
33 * Copyright (c) 2001 Thomas Nordin <nordin@openbsd.org>
34 * Copyright (c) 2000-2001 Artur Grabowski <art@openbsd.org>
35 * All rights reserved.
36 *
37 * Redistribution and use in source and binary forms, with or without
38 * modification, are permitted provided that the following conditions
39 * are met:
40 *
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. The name of the author may not be used to endorse or promote products
47 * derived from this software without specific prior written permission.
48 *
49 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES,
50 * INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY
51 * AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL
52 * THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
53 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
54 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
55 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
56 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
57 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
58 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
59 */
60
61#include <sys/cdefs.h>
62__KERNEL_RCSID(0, "$NetBSD: kern_timeout.c,v 1.51 2015/11/24 15:48:23 christos Exp $");
63
64/*
65 * Timeouts are kept in a hierarchical timing wheel. The c_time is the
66 * value of c_cpu->cc_ticks when the timeout should be called. There are
67 * four levels with 256 buckets each. See 'Scheme 7' in "Hashed and
68 * Hierarchical Timing Wheels: Efficient Data Structures for Implementing
69 * a Timer Facility" by George Varghese and Tony Lauck.
70 *
71 * Some of the "math" in here is a bit tricky. We have to beware of
72 * wrapping ints.
73 *
74 * We use the fact that any element added to the queue must be added with
75 * a positive time. That means that any element `to' on the queue cannot
76 * be scheduled to timeout further in time than INT_MAX, but c->c_time can
77 * be positive or negative so comparing it with anything is dangerous.
78 * The only way we can use the c->c_time value in any predictable way is
79 * when we calculate how far in the future `to' will timeout - "c->c_time
80 * - c->c_cpu->cc_ticks". The result will always be positive for future
81 * timeouts and 0 or negative for due timeouts.
82 */
83
84#define _CALLOUT_PRIVATE
85
86#include <sys/param.h>
87#include <sys/systm.h>
88#include <sys/kernel.h>
89#include <sys/callout.h>
90#include <sys/lwp.h>
91#include <sys/mutex.h>
92#include <sys/proc.h>
93#include <sys/sleepq.h>
94#include <sys/syncobj.h>
95#include <sys/evcnt.h>
96#include <sys/intr.h>
97#include <sys/cpu.h>
98#include <sys/kmem.h>
99
100#ifdef DDB
101#include <machine/db_machdep.h>
102#include <ddb/db_interface.h>
103#include <ddb/db_access.h>
104#include <ddb/db_cpu.h>
105#include <ddb/db_sym.h>
106#include <ddb/db_output.h>
107#endif
108
109#define BUCKETS 1024
110#define WHEELSIZE 256
111#define WHEELMASK 255
112#define WHEELBITS 8
113
114#define MASKWHEEL(wheel, time) (((time) >> ((wheel)*WHEELBITS)) & WHEELMASK)
115
116#define BUCKET(cc, rel, abs) \
117 (((rel) <= (1 << (2*WHEELBITS))) \
118 ? ((rel) <= (1 << WHEELBITS)) \
119 ? &(cc)->cc_wheel[MASKWHEEL(0, (abs))] \
120 : &(cc)->cc_wheel[MASKWHEEL(1, (abs)) + WHEELSIZE] \
121 : ((rel) <= (1 << (3*WHEELBITS))) \
122 ? &(cc)->cc_wheel[MASKWHEEL(2, (abs)) + 2*WHEELSIZE] \
123 : &(cc)->cc_wheel[MASKWHEEL(3, (abs)) + 3*WHEELSIZE])
124
125#define MOVEBUCKET(cc, wheel, time) \
126 CIRCQ_APPEND(&(cc)->cc_todo, \
127 &(cc)->cc_wheel[MASKWHEEL((wheel), (time)) + (wheel)*WHEELSIZE])
128
129/*
130 * Circular queue definitions.
131 */
132
133#define CIRCQ_INIT(list) \
134do { \
135 (list)->cq_next_l = (list); \
136 (list)->cq_prev_l = (list); \
137} while (/*CONSTCOND*/0)
138
139#define CIRCQ_INSERT(elem, list) \
140do { \
141 (elem)->cq_prev_e = (list)->cq_prev_e; \
142 (elem)->cq_next_l = (list); \
143 (list)->cq_prev_l->cq_next_l = (elem); \
144 (list)->cq_prev_l = (elem); \
145} while (/*CONSTCOND*/0)
146
147#define CIRCQ_APPEND(fst, snd) \
148do { \
149 if (!CIRCQ_EMPTY(snd)) { \
150 (fst)->cq_prev_l->cq_next_l = (snd)->cq_next_l; \
151 (snd)->cq_next_l->cq_prev_l = (fst)->cq_prev_l; \
152 (snd)->cq_prev_l->cq_next_l = (fst); \
153 (fst)->cq_prev_l = (snd)->cq_prev_l; \
154 CIRCQ_INIT(snd); \
155 } \
156} while (/*CONSTCOND*/0)
157
158#define CIRCQ_REMOVE(elem) \
159do { \
160 (elem)->cq_next_l->cq_prev_e = (elem)->cq_prev_e; \
161 (elem)->cq_prev_l->cq_next_e = (elem)->cq_next_e; \
162} while (/*CONSTCOND*/0)
163
164#define CIRCQ_FIRST(list) ((list)->cq_next_e)
165#define CIRCQ_NEXT(elem) ((elem)->cq_next_e)
166#define CIRCQ_LAST(elem,list) ((elem)->cq_next_l == (list))
167#define CIRCQ_EMPTY(list) ((list)->cq_next_l == (list))
168
169struct callout_cpu {
170 kmutex_t *cc_lock;
171 sleepq_t cc_sleepq;
172 u_int cc_nwait;
173 u_int cc_ticks;
174 lwp_t *cc_lwp;
175 callout_impl_t *cc_active;
176 callout_impl_t *cc_cancel;
177 struct evcnt cc_ev_late;
178 struct evcnt cc_ev_block;
179 struct callout_circq cc_todo; /* Worklist */
180 struct callout_circq cc_wheel[BUCKETS]; /* Queues of timeouts */
181 char cc_name1[12];
182 char cc_name2[12];
183};
184
185#ifndef CRASH
186
187static void callout_softclock(void *);
188static struct callout_cpu callout_cpu0;
189static void *callout_sih;
190
191static inline kmutex_t *
192callout_lock(callout_impl_t *c)
193{
194 struct callout_cpu *cc;
195 kmutex_t *lock;
196
197 for (;;) {
198 cc = c->c_cpu;
199 lock = cc->cc_lock;
200 mutex_spin_enter(lock);
201 if (__predict_true(cc == c->c_cpu))
202 return lock;
203 mutex_spin_exit(lock);
204 }
205}
206
207/*
208 * callout_startup:
209 *
210 * Initialize the callout facility, called at system startup time.
211 * Do just enough to allow callouts to be safely registered.
212 */
213void
214callout_startup(void)
215{
216 struct callout_cpu *cc;
217 int b;
218
219 KASSERT(curcpu()->ci_data.cpu_callout == NULL);
220
221 cc = &callout_cpu0;
222 cc->cc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SCHED);
223 CIRCQ_INIT(&cc->cc_todo);
224 for (b = 0; b < BUCKETS; b++)
225 CIRCQ_INIT(&cc->cc_wheel[b]);
226 curcpu()->ci_data.cpu_callout = cc;
227}
228
229/*
230 * callout_init_cpu:
231 *
232 * Per-CPU initialization.
233 */
234CTASSERT(sizeof(callout_impl_t) <= sizeof(callout_t));
235
236void
237callout_init_cpu(struct cpu_info *ci)
238{
239 struct callout_cpu *cc;
240 int b;
241
242 if ((cc = ci->ci_data.cpu_callout) == NULL) {
243 cc = kmem_zalloc(sizeof(*cc), KM_SLEEP);
244 if (cc == NULL)
245 panic("callout_init_cpu (1)");
246 cc->cc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SCHED);
247 CIRCQ_INIT(&cc->cc_todo);
248 for (b = 0; b < BUCKETS; b++)
249 CIRCQ_INIT(&cc->cc_wheel[b]);
250 } else {
251 /* Boot CPU, one time only. */
252 callout_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE,
253 callout_softclock, NULL);
254 if (callout_sih == NULL)
255 panic("callout_init_cpu (2)");
256 }
257
258 sleepq_init(&cc->cc_sleepq);
259
260 snprintf(cc->cc_name1, sizeof(cc->cc_name1), "late/%u",
261 cpu_index(ci));
262 evcnt_attach_dynamic(&cc->cc_ev_late, EVCNT_TYPE_MISC,
263 NULL, "callout", cc->cc_name1);
264
265 snprintf(cc->cc_name2, sizeof(cc->cc_name2), "wait/%u",
266 cpu_index(ci));
267 evcnt_attach_dynamic(&cc->cc_ev_block, EVCNT_TYPE_MISC,
268 NULL, "callout", cc->cc_name2);
269
270 ci->ci_data.cpu_callout = cc;
271}
272
273/*
274 * callout_init:
275 *
276 * Initialize a callout structure. This must be quick, so we fill
277 * only the minimum number of fields.
278 */
279void
280callout_init(callout_t *cs, u_int flags)
281{
282 callout_impl_t *c = (callout_impl_t *)cs;
283 struct callout_cpu *cc;
284
285 KASSERT((flags & ~CALLOUT_FLAGMASK) == 0);
286
287 cc = curcpu()->ci_data.cpu_callout;
288 c->c_func = NULL;
289 c->c_magic = CALLOUT_MAGIC;
290 if (__predict_true((flags & CALLOUT_MPSAFE) != 0 && cc != NULL)) {
291 c->c_flags = flags;
292 c->c_cpu = cc;
293 return;
294 }
295 c->c_flags = flags | CALLOUT_BOUND;
296 c->c_cpu = &callout_cpu0;
297}
298
299/*
300 * callout_destroy:
301 *
302 * Destroy a callout structure. The callout must be stopped.
303 */
304void
305callout_destroy(callout_t *cs)
306{
307 callout_impl_t *c = (callout_impl_t *)cs;
308
309 /*
310 * It's not necessary to lock in order to see the correct value
311 * of c->c_flags. If the callout could potentially have been
312 * running, the current thread should have stopped it.
313 */
314 KASSERTMSG((c->c_flags & CALLOUT_PENDING) == 0,
315 "callout %p: c_func (%p) c_flags (%#x) destroyed from %p",
316 c, c->c_func, c->c_flags, __builtin_return_address(0));
317 KASSERT(c->c_cpu->cc_lwp == curlwp || c->c_cpu->cc_active != c);
318 KASSERTMSG(c->c_magic == CALLOUT_MAGIC,
319 "callout %p: c_magic (%#x) != CALLOUT_MAGIC (%#x)",
320 c, c->c_magic, CALLOUT_MAGIC);
321 c->c_magic = 0;
322}
323
324/*
325 * callout_schedule_locked:
326 *
327 * Schedule a callout to run. The function and argument must
328 * already be set in the callout structure. Must be called with
329 * callout_lock.
330 */
331static void
332callout_schedule_locked(callout_impl_t *c, kmutex_t *lock, int to_ticks)
333{
334 struct callout_cpu *cc, *occ;
335 int old_time;
336
337 KASSERT(to_ticks >= 0);
338 KASSERT(c->c_func != NULL);
339
340 /* Initialize the time here, it won't change. */
341 occ = c->c_cpu;
342 c->c_flags &= ~(CALLOUT_FIRED | CALLOUT_INVOKING);
343
344 /*
345 * If this timeout is already scheduled and now is moved
346 * earlier, reschedule it now. Otherwise leave it in place
347 * and let it be rescheduled later.
348 */
349 if ((c->c_flags & CALLOUT_PENDING) != 0) {
350 /* Leave on existing CPU. */
351 old_time = c->c_time;
352 c->c_time = to_ticks + occ->cc_ticks;
353 if (c->c_time - old_time < 0) {
354 CIRCQ_REMOVE(&c->c_list);
355 CIRCQ_INSERT(&c->c_list, &occ->cc_todo);
356 }
357 mutex_spin_exit(lock);
358 return;
359 }
360
361 cc = curcpu()->ci_data.cpu_callout;
362 if ((c->c_flags & CALLOUT_BOUND) != 0 || cc == occ ||
363 !mutex_tryenter(cc->cc_lock)) {
364 /* Leave on existing CPU. */
365 c->c_time = to_ticks + occ->cc_ticks;
366 c->c_flags |= CALLOUT_PENDING;
367 CIRCQ_INSERT(&c->c_list, &occ->cc_todo);
368 } else {
369 /* Move to this CPU. */
370 c->c_cpu = cc;
371 c->c_time = to_ticks + cc->cc_ticks;
372 c->c_flags |= CALLOUT_PENDING;
373 CIRCQ_INSERT(&c->c_list, &cc->cc_todo);
374 mutex_spin_exit(cc->cc_lock);
375 }
376 mutex_spin_exit(lock);
377}
378
379/*
380 * callout_reset:
381 *
382 * Reset a callout structure with a new function and argument, and
383 * schedule it to run.
384 */
385void
386callout_reset(callout_t *cs, int to_ticks, void (*func)(void *), void *arg)
387{
388 callout_impl_t *c = (callout_impl_t *)cs;
389 kmutex_t *lock;
390
391 KASSERT(c->c_magic == CALLOUT_MAGIC);
392 KASSERT(func != NULL);
393
394 lock = callout_lock(c);
395 c->c_func = func;
396 c->c_arg = arg;
397 callout_schedule_locked(c, lock, to_ticks);
398}
399
400/*
401 * callout_schedule:
402 *
403 * Schedule a callout to run. The function and argument must
404 * already be set in the callout structure.
405 */
406void
407callout_schedule(callout_t *cs, int to_ticks)
408{
409 callout_impl_t *c = (callout_impl_t *)cs;
410 kmutex_t *lock;
411
412 KASSERT(c->c_magic == CALLOUT_MAGIC);
413
414 lock = callout_lock(c);
415 callout_schedule_locked(c, lock, to_ticks);
416}
417
418/*
419 * callout_stop:
420 *
421 * Try to cancel a pending callout. It may be too late: the callout
422 * could be running on another CPU. If called from interrupt context,
423 * the callout could already be in progress at a lower priority.
424 */
425bool
426callout_stop(callout_t *cs)
427{
428 callout_impl_t *c = (callout_impl_t *)cs;
429 struct callout_cpu *cc;
430 kmutex_t *lock;
431 bool expired;
432
433 KASSERT(c->c_magic == CALLOUT_MAGIC);
434
435 lock = callout_lock(c);
436
437 if ((c->c_flags & CALLOUT_PENDING) != 0)
438 CIRCQ_REMOVE(&c->c_list);
439 expired = ((c->c_flags & CALLOUT_FIRED) != 0);
440 c->c_flags &= ~(CALLOUT_PENDING|CALLOUT_FIRED);
441
442 cc = c->c_cpu;
443 if (cc->cc_active == c) {
444 /*
445 * This is for non-MPSAFE callouts only. To synchronize
446 * effectively we must be called with kernel_lock held.
447 * It's also taken in callout_softclock.
448 */
449 cc->cc_cancel = c;
450 }
451
452 mutex_spin_exit(lock);
453
454 return expired;
455}
456
457/*
458 * callout_halt:
459 *
460 * Cancel a pending callout. If in-flight, block until it completes.
461 * May not be called from a hard interrupt handler. If the callout
462 * can take locks, the caller of callout_halt() must not hold any of
463 * those locks, otherwise the two could deadlock. If 'interlock' is
464 * non-NULL and we must wait for the callout to complete, it will be
465 * released and re-acquired before returning.
466 */
467bool
468callout_halt(callout_t *cs, void *interlock)
469{
470 callout_impl_t *c = (callout_impl_t *)cs;
471 struct callout_cpu *cc;
472 struct lwp *l;
473 kmutex_t *lock, *relock;
474 bool expired;
475
476 KASSERT(c->c_magic == CALLOUT_MAGIC);
477 KASSERT(!cpu_intr_p());
478
479 lock = callout_lock(c);
480 relock = NULL;
481
482 expired = ((c->c_flags & CALLOUT_FIRED) != 0);
483 if ((c->c_flags & CALLOUT_PENDING) != 0)
484 CIRCQ_REMOVE(&c->c_list);
485 c->c_flags &= ~(CALLOUT_PENDING|CALLOUT_FIRED);
486
487 l = curlwp;
488 for (;;) {
489 cc = c->c_cpu;
490 if (__predict_true(cc->cc_active != c || cc->cc_lwp == l))
491 break;
492 if (interlock != NULL) {
493 /*
494 * Avoid potential scheduler lock order problems by
495 * dropping the interlock without the callout lock
496 * held.
497 */
498 mutex_spin_exit(lock);
499 mutex_exit(interlock);
500 relock = interlock;
501 interlock = NULL;
502 } else {
503 /* XXX Better to do priority inheritance. */
504 KASSERT(l->l_wchan == NULL);
505 cc->cc_nwait++;
506 cc->cc_ev_block.ev_count++;
507 l->l_kpriority = true;
508 sleepq_enter(&cc->cc_sleepq, l, cc->cc_lock);
509 sleepq_enqueue(&cc->cc_sleepq, cc, "callout",
510 &sleep_syncobj);
511 sleepq_block(0, false);
512 }
513 lock = callout_lock(c);
514 }
515
516 mutex_spin_exit(lock);
517 if (__predict_false(relock != NULL))
518 mutex_enter(relock);
519
520 return expired;
521}
522
523#ifdef notyet
524/*
525 * callout_bind:
526 *
527 * Bind a callout so that it will only execute on one CPU.
528 * The callout must be stopped, and must be MPSAFE.
529 *
530 * XXX Disabled for now until it is decided how to handle
531 * offlined CPUs. We may want weak+strong binding.
532 */
533void
534callout_bind(callout_t *cs, struct cpu_info *ci)
535{
536 callout_impl_t *c = (callout_impl_t *)cs;
537 struct callout_cpu *cc;
538 kmutex_t *lock;
539
540 KASSERT((c->c_flags & CALLOUT_PENDING) == 0);
541 KASSERT(c->c_cpu->cc_active != c);
542 KASSERT(c->c_magic == CALLOUT_MAGIC);
543 KASSERT((c->c_flags & CALLOUT_MPSAFE) != 0);
544
545 lock = callout_lock(c);
546 cc = ci->ci_data.cpu_callout;
547 c->c_flags |= CALLOUT_BOUND;
548 if (c->c_cpu != cc) {
549 /*
550 * Assigning c_cpu effectively unlocks the callout
551 * structure, as we don't hold the new CPU's lock.
552 * Issue memory barrier to prevent accesses being
553 * reordered.
554 */
555 membar_exit();
556 c->c_cpu = cc;
557 }
558 mutex_spin_exit(lock);
559}
560#endif
561
562void
563callout_setfunc(callout_t *cs, void (*func)(void *), void *arg)
564{
565 callout_impl_t *c = (callout_impl_t *)cs;
566 kmutex_t *lock;
567
568 KASSERT(c->c_magic == CALLOUT_MAGIC);
569 KASSERT(func != NULL);
570
571 lock = callout_lock(c);
572 c->c_func = func;
573 c->c_arg = arg;
574 mutex_spin_exit(lock);
575}
576
577bool
578callout_expired(callout_t *cs)
579{
580 callout_impl_t *c = (callout_impl_t *)cs;
581 kmutex_t *lock;
582 bool rv;
583
584 KASSERT(c->c_magic == CALLOUT_MAGIC);
585
586 lock = callout_lock(c);
587 rv = ((c->c_flags & CALLOUT_FIRED) != 0);
588 mutex_spin_exit(lock);
589
590 return rv;
591}
592
593bool
594callout_active(callout_t *cs)
595{
596 callout_impl_t *c = (callout_impl_t *)cs;
597 kmutex_t *lock;
598 bool rv;
599
600 KASSERT(c->c_magic == CALLOUT_MAGIC);
601
602 lock = callout_lock(c);
603 rv = ((c->c_flags & (CALLOUT_PENDING|CALLOUT_FIRED)) != 0);
604 mutex_spin_exit(lock);
605
606 return rv;
607}
608
609bool
610callout_pending(callout_t *cs)
611{
612 callout_impl_t *c = (callout_impl_t *)cs;
613 kmutex_t *lock;
614 bool rv;
615
616 KASSERT(c->c_magic == CALLOUT_MAGIC);
617
618 lock = callout_lock(c);
619 rv = ((c->c_flags & CALLOUT_PENDING) != 0);
620 mutex_spin_exit(lock);
621
622 return rv;
623}
624
625bool
626callout_invoking(callout_t *cs)
627{
628 callout_impl_t *c = (callout_impl_t *)cs;
629 kmutex_t *lock;
630 bool rv;
631
632 KASSERT(c->c_magic == CALLOUT_MAGIC);
633
634 lock = callout_lock(c);
635 rv = ((c->c_flags & CALLOUT_INVOKING) != 0);
636 mutex_spin_exit(lock);
637
638 return rv;
639}
640
641void
642callout_ack(callout_t *cs)
643{
644 callout_impl_t *c = (callout_impl_t *)cs;
645 kmutex_t *lock;
646
647 KASSERT(c->c_magic == CALLOUT_MAGIC);
648
649 lock = callout_lock(c);
650 c->c_flags &= ~CALLOUT_INVOKING;
651 mutex_spin_exit(lock);
652}
653
654/*
655 * callout_hardclock:
656 *
657 * Called from hardclock() once every tick. We schedule a soft
658 * interrupt if there is work to be done.
659 */
660void
661callout_hardclock(void)
662{
663 struct callout_cpu *cc;
664 int needsoftclock, ticks;
665
666 cc = curcpu()->ci_data.cpu_callout;
667 mutex_spin_enter(cc->cc_lock);
668
669 ticks = ++cc->cc_ticks;
670
671 MOVEBUCKET(cc, 0, ticks);
672 if (MASKWHEEL(0, ticks) == 0) {
673 MOVEBUCKET(cc, 1, ticks);
674 if (MASKWHEEL(1, ticks) == 0) {
675 MOVEBUCKET(cc, 2, ticks);
676 if (MASKWHEEL(2, ticks) == 0)
677 MOVEBUCKET(cc, 3, ticks);
678 }
679 }
680
681 needsoftclock = !CIRCQ_EMPTY(&cc->cc_todo);
682 mutex_spin_exit(cc->cc_lock);
683
684 if (needsoftclock)
685 softint_schedule(callout_sih);
686}
687
688/*
689 * callout_softclock:
690 *
691 * Soft interrupt handler, scheduled above if there is work to
692 * be done. Callouts are made in soft interrupt context.
693 */
694static void
695callout_softclock(void *v)
696{
697 callout_impl_t *c;
698 struct callout_cpu *cc;
699 void (*func)(void *);
700 void *arg;
701 int mpsafe, count, ticks, delta;
702 lwp_t *l;
703
704 l = curlwp;
705 KASSERT(l->l_cpu == curcpu());
706 cc = l->l_cpu->ci_data.cpu_callout;
707
708 mutex_spin_enter(cc->cc_lock);
709 cc->cc_lwp = l;
710 while (!CIRCQ_EMPTY(&cc->cc_todo)) {
711 c = CIRCQ_FIRST(&cc->cc_todo);
712 KASSERT(c->c_magic == CALLOUT_MAGIC);
713 KASSERT(c->c_func != NULL);
714 KASSERT(c->c_cpu == cc);
715 KASSERT((c->c_flags & CALLOUT_PENDING) != 0);
716 KASSERT((c->c_flags & CALLOUT_FIRED) == 0);
717 CIRCQ_REMOVE(&c->c_list);
718
719 /* If due run it, otherwise insert it into the right bucket. */
720 ticks = cc->cc_ticks;
721 delta = c->c_time - ticks;
722 if (delta > 0) {
723 CIRCQ_INSERT(&c->c_list, BUCKET(cc, delta, c->c_time));
724 continue;
725 }
726 if (delta < 0)
727 cc->cc_ev_late.ev_count++;
728
729 c->c_flags = (c->c_flags & ~CALLOUT_PENDING) |
730 (CALLOUT_FIRED | CALLOUT_INVOKING);
731 mpsafe = (c->c_flags & CALLOUT_MPSAFE);
732 func = c->c_func;
733 arg = c->c_arg;
734 cc->cc_active = c;
735
736 mutex_spin_exit(cc->cc_lock);
737 KASSERT(func != NULL);
738 if (__predict_false(!mpsafe)) {
739 KERNEL_LOCK(1, NULL);
740 (*func)(arg);
741 KERNEL_UNLOCK_ONE(NULL);
742 } else
743 (*func)(arg);
744 mutex_spin_enter(cc->cc_lock);
745
746 /*
747 * We can't touch 'c' here because it might be
748 * freed already. If LWPs waiting for callout
749 * to complete, awaken them.
750 */
751 cc->cc_active = NULL;
752 if ((count = cc->cc_nwait) != 0) {
753 cc->cc_nwait = 0;
754 /* sleepq_wake() drops the lock. */
755 sleepq_wake(&cc->cc_sleepq, cc, count, cc->cc_lock);
756 mutex_spin_enter(cc->cc_lock);
757 }
758 }
759 cc->cc_lwp = NULL;
760 mutex_spin_exit(cc->cc_lock);
761}
762#endif
763
764#ifdef DDB
765static void
766db_show_callout_bucket(struct callout_cpu *cc, struct callout_circq *kbucket,
767 struct callout_circq *bucket)
768{
769 callout_impl_t *c, ci;
770 db_expr_t offset;
771 const char *name;
772 static char question[] = "?";
773 int b;
774
775 if (CIRCQ_LAST(bucket, kbucket))
776 return;
777
778 for (c = CIRCQ_FIRST(bucket); /*nothing*/; c = CIRCQ_NEXT(&c->c_list)) {
779 db_read_bytes((db_addr_t)c, sizeof(ci), (char *)&ci);
780 c = &ci;
781 db_find_sym_and_offset((db_addr_t)(intptr_t)c->c_func, &name,
782 &offset);
783 name = name ? name : question;
784 b = (bucket - cc->cc_wheel);
785 if (b < 0)
786 b = -WHEELSIZE;
787 db_printf("%9d %2d/%-4d %16lx %s\n",
788 c->c_time - cc->cc_ticks, b / WHEELSIZE, b,
789 (u_long)c->c_arg, name);
790 if (CIRCQ_LAST(&c->c_list, kbucket))
791 break;
792 }
793}
794
795void
796db_show_callout(db_expr_t addr, bool haddr, db_expr_t count, const char *modif)
797{
798 struct callout_cpu *cc, ccb;
799 struct cpu_info *ci, cib;
800 int b;
801
802#ifndef CRASH
803 db_printf("hardclock_ticks now: %d\n", hardclock_ticks);
804#endif
805 db_printf(" ticks wheel arg func\n");
806
807 /*
808 * Don't lock the callwheel; all the other CPUs are paused
809 * anyhow, and we might be called in a circumstance where
810 * some other CPU was paused while holding the lock.
811 */
812 for (ci = db_cpu_first(); ci != NULL; ci = db_cpu_next(ci)) {
813 db_read_bytes((db_addr_t)ci, sizeof(cib), (char *)&cib);
814 cc = cib.ci_data.cpu_callout;
815 db_read_bytes((db_addr_t)cc, sizeof(ccb), (char *)&ccb);
816 db_show_callout_bucket(&ccb, &cc->cc_todo, &ccb.cc_todo);
817 }
818 for (b = 0; b < BUCKETS; b++) {
819 for (ci = db_cpu_first(); ci != NULL; ci = db_cpu_next(ci)) {
820 db_read_bytes((db_addr_t)ci, sizeof(cib), (char *)&cib);
821 cc = cib.ci_data.cpu_callout;
822 db_read_bytes((db_addr_t)cc, sizeof(ccb), (char *)&ccb);
823 db_show_callout_bucket(&ccb, &cc->cc_wheel[b],
824 &ccb.cc_wheel[b]);
825 }
826 }
827}
828#endif /* DDB */
829