xref: /freebsd/sys/kern/subr_taskqueue.c (revision 9d54812421274e490dc5f0fe4722ab8d35d9b258)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2000 Doug Rabson
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/bus.h>
35 #include <sys/cpuset.h>
36 #include <sys/interrupt.h>
37 #include <sys/kernel.h>
38 #include <sys/kthread.h>
39 #include <sys/libkern.h>
40 #include <sys/limits.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mutex.h>
44 #include <sys/proc.h>
45 #include <sys/epoch.h>
46 #include <sys/sched.h>
47 #include <sys/smp.h>
48 #include <sys/taskqueue.h>
49 #include <sys/unistd.h>
50 #include <machine/stdarg.h>
51 
52 static MALLOC_DEFINE(M_TASKQUEUE, "taskqueue", "Task Queues");
53 static void	*taskqueue_giant_ih;
54 static void	*taskqueue_ih;
55 static void	 taskqueue_fast_enqueue(void *);
56 static void	 taskqueue_swi_enqueue(void *);
57 static void	 taskqueue_swi_giant_enqueue(void *);
58 
59 struct taskqueue_busy {
60 	struct task		*tb_running;
61 	u_int			 tb_seq;
62 	bool			 tb_canceling;
63 	LIST_ENTRY(taskqueue_busy) tb_link;
64 };
65 
66 struct taskqueue {
67 	STAILQ_HEAD(, task)	tq_queue;
68 	LIST_HEAD(, taskqueue_busy) tq_active;
69 	struct task		*tq_hint;
70 	u_int			tq_seq;
71 	int			tq_callouts;
72 	struct mtx_padalign	tq_mutex;
73 	taskqueue_enqueue_fn	tq_enqueue;
74 	void			*tq_context;
75 	char			*tq_name;
76 	struct thread		**tq_threads;
77 	int			tq_tcount;
78 	int			tq_spin;
79 	int			tq_flags;
80 	taskqueue_callback_fn	tq_callbacks[TASKQUEUE_NUM_CALLBACKS];
81 	void			*tq_cb_contexts[TASKQUEUE_NUM_CALLBACKS];
82 };
83 
84 #define	TQ_FLAGS_ACTIVE		(1 << 0)
85 #define	TQ_FLAGS_BLOCKED	(1 << 1)
86 #define	TQ_FLAGS_UNLOCKED_ENQUEUE	(1 << 2)
87 
88 #define	DT_CALLOUT_ARMED	(1 << 0)
89 #define	DT_DRAIN_IN_PROGRESS	(1 << 1)
90 
91 #define	TQ_LOCK(tq)							\
92 	do {								\
93 		if ((tq)->tq_spin)					\
94 			mtx_lock_spin(&(tq)->tq_mutex);			\
95 		else							\
96 			mtx_lock(&(tq)->tq_mutex);			\
97 	} while (0)
98 #define	TQ_ASSERT_LOCKED(tq)	mtx_assert(&(tq)->tq_mutex, MA_OWNED)
99 
100 #define	TQ_UNLOCK(tq)							\
101 	do {								\
102 		if ((tq)->tq_spin)					\
103 			mtx_unlock_spin(&(tq)->tq_mutex);		\
104 		else							\
105 			mtx_unlock(&(tq)->tq_mutex);			\
106 	} while (0)
107 #define	TQ_ASSERT_UNLOCKED(tq)	mtx_assert(&(tq)->tq_mutex, MA_NOTOWNED)
108 
109 void
110 _timeout_task_init(struct taskqueue *queue, struct timeout_task *timeout_task,
111     int priority, task_fn_t func, void *context)
112 {
113 
114 	TASK_INIT(&timeout_task->t, priority, func, context);
115 	callout_init_mtx(&timeout_task->c, &queue->tq_mutex,
116 	    CALLOUT_RETURNUNLOCKED);
117 	timeout_task->q = queue;
118 	timeout_task->f = 0;
119 }
120 
121 static __inline int
122 TQ_SLEEP(struct taskqueue *tq, void *p, const char *wm)
123 {
124 	if (tq->tq_spin)
125 		return (msleep_spin(p, (struct mtx *)&tq->tq_mutex, wm, 0));
126 	return (msleep(p, &tq->tq_mutex, 0, wm, 0));
127 }
128 
129 static struct taskqueue_busy *
130 task_get_busy(struct taskqueue *queue, struct task *task)
131 {
132 	struct taskqueue_busy *tb;
133 
134 	TQ_ASSERT_LOCKED(queue);
135 	LIST_FOREACH(tb, &queue->tq_active, tb_link) {
136 		if (tb->tb_running == task)
137 			return (tb);
138 	}
139 	return (NULL);
140 }
141 
142 static struct taskqueue *
143 _taskqueue_create(const char *name, int mflags,
144 		 taskqueue_enqueue_fn enqueue, void *context,
145 		 int mtxflags, const char *mtxname __unused)
146 {
147 	struct taskqueue *queue;
148 	char *tq_name;
149 
150 	tq_name = malloc(TASKQUEUE_NAMELEN, M_TASKQUEUE, mflags | M_ZERO);
151 	if (tq_name == NULL)
152 		return (NULL);
153 
154 	queue = malloc(sizeof(struct taskqueue), M_TASKQUEUE, mflags | M_ZERO);
155 	if (queue == NULL) {
156 		free(tq_name, M_TASKQUEUE);
157 		return (NULL);
158 	}
159 
160 	snprintf(tq_name, TASKQUEUE_NAMELEN, "%s", (name) ? name : "taskqueue");
161 
162 	STAILQ_INIT(&queue->tq_queue);
163 	LIST_INIT(&queue->tq_active);
164 	queue->tq_enqueue = enqueue;
165 	queue->tq_context = context;
166 	queue->tq_name = tq_name;
167 	queue->tq_spin = (mtxflags & MTX_SPIN) != 0;
168 	queue->tq_flags |= TQ_FLAGS_ACTIVE;
169 	if (enqueue == taskqueue_fast_enqueue ||
170 	    enqueue == taskqueue_swi_enqueue ||
171 	    enqueue == taskqueue_swi_giant_enqueue ||
172 	    enqueue == taskqueue_thread_enqueue)
173 		queue->tq_flags |= TQ_FLAGS_UNLOCKED_ENQUEUE;
174 	mtx_init(&queue->tq_mutex, tq_name, NULL, mtxflags);
175 
176 	return (queue);
177 }
178 
179 struct taskqueue *
180 taskqueue_create(const char *name, int mflags,
181 		 taskqueue_enqueue_fn enqueue, void *context)
182 {
183 
184 	return _taskqueue_create(name, mflags, enqueue, context,
185 			MTX_DEF, name);
186 }
187 
188 void
189 taskqueue_set_callback(struct taskqueue *queue,
190     enum taskqueue_callback_type cb_type, taskqueue_callback_fn callback,
191     void *context)
192 {
193 
194 	KASSERT(((cb_type >= TASKQUEUE_CALLBACK_TYPE_MIN) &&
195 	    (cb_type <= TASKQUEUE_CALLBACK_TYPE_MAX)),
196 	    ("Callback type %d not valid, must be %d-%d", cb_type,
197 	    TASKQUEUE_CALLBACK_TYPE_MIN, TASKQUEUE_CALLBACK_TYPE_MAX));
198 	KASSERT((queue->tq_callbacks[cb_type] == NULL),
199 	    ("Re-initialization of taskqueue callback?"));
200 
201 	queue->tq_callbacks[cb_type] = callback;
202 	queue->tq_cb_contexts[cb_type] = context;
203 }
204 
205 /*
206  * Signal a taskqueue thread to terminate.
207  */
208 static void
209 taskqueue_terminate(struct thread **pp, struct taskqueue *tq)
210 {
211 
212 	while (tq->tq_tcount > 0 || tq->tq_callouts > 0) {
213 		wakeup(tq);
214 		TQ_SLEEP(tq, pp, "tq_destroy");
215 	}
216 }
217 
218 void
219 taskqueue_free(struct taskqueue *queue)
220 {
221 
222 	TQ_LOCK(queue);
223 	queue->tq_flags &= ~TQ_FLAGS_ACTIVE;
224 	taskqueue_terminate(queue->tq_threads, queue);
225 	KASSERT(LIST_EMPTY(&queue->tq_active), ("Tasks still running?"));
226 	KASSERT(queue->tq_callouts == 0, ("Armed timeout tasks"));
227 	mtx_destroy(&queue->tq_mutex);
228 	free(queue->tq_threads, M_TASKQUEUE);
229 	free(queue->tq_name, M_TASKQUEUE);
230 	free(queue, M_TASKQUEUE);
231 }
232 
233 static int
234 taskqueue_enqueue_locked(struct taskqueue *queue, struct task *task, int flags)
235 {
236 	struct task *ins;
237 	struct task *prev;
238 	struct taskqueue_busy *tb;
239 
240 	KASSERT(task->ta_func != NULL, ("enqueueing task with NULL func"));
241 	/*
242 	 * Ignore canceling task if requested.
243 	 */
244 	if (__predict_false((flags & TASKQUEUE_FAIL_IF_CANCELING) != 0)) {
245 		tb = task_get_busy(queue, task);
246 		if (tb != NULL && tb->tb_canceling) {
247 			TQ_UNLOCK(queue);
248 			return (ECANCELED);
249 		}
250 	}
251 
252 	/*
253 	 * Count multiple enqueues.
254 	 */
255 	if (task->ta_pending) {
256 		if (__predict_false((flags & TASKQUEUE_FAIL_IF_PENDING) != 0)) {
257 			TQ_UNLOCK(queue);
258 			return (EEXIST);
259 		}
260 		if (task->ta_pending < USHRT_MAX)
261 			task->ta_pending++;
262 		TQ_UNLOCK(queue);
263 		return (0);
264 	}
265 
266 	/*
267 	 * Optimise cases when all tasks use small set of priorities.
268 	 * In case of only one priority we always insert at the end.
269 	 * In case of two tq_hint typically gives the insertion point.
270 	 * In case of more then two tq_hint should halve the search.
271 	 */
272 	prev = STAILQ_LAST(&queue->tq_queue, task, ta_link);
273 	if (!prev || prev->ta_priority >= task->ta_priority) {
274 		STAILQ_INSERT_TAIL(&queue->tq_queue, task, ta_link);
275 	} else {
276 		prev = queue->tq_hint;
277 		if (prev && prev->ta_priority >= task->ta_priority) {
278 			ins = STAILQ_NEXT(prev, ta_link);
279 		} else {
280 			prev = NULL;
281 			ins = STAILQ_FIRST(&queue->tq_queue);
282 		}
283 		for (; ins; prev = ins, ins = STAILQ_NEXT(ins, ta_link))
284 			if (ins->ta_priority < task->ta_priority)
285 				break;
286 
287 		if (prev) {
288 			STAILQ_INSERT_AFTER(&queue->tq_queue, prev, task, ta_link);
289 			queue->tq_hint = task;
290 		} else
291 			STAILQ_INSERT_HEAD(&queue->tq_queue, task, ta_link);
292 	}
293 
294 	task->ta_pending = 1;
295 	if ((queue->tq_flags & TQ_FLAGS_UNLOCKED_ENQUEUE) != 0)
296 		TQ_UNLOCK(queue);
297 	if ((queue->tq_flags & TQ_FLAGS_BLOCKED) == 0)
298 		queue->tq_enqueue(queue->tq_context);
299 	if ((queue->tq_flags & TQ_FLAGS_UNLOCKED_ENQUEUE) == 0)
300 		TQ_UNLOCK(queue);
301 
302 	/* Return with lock released. */
303 	return (0);
304 }
305 
306 int
307 taskqueue_enqueue_flags(struct taskqueue *queue, struct task *task, int flags)
308 {
309 	int res;
310 
311 	TQ_LOCK(queue);
312 	res = taskqueue_enqueue_locked(queue, task, flags);
313 	/* The lock is released inside. */
314 
315 	return (res);
316 }
317 
318 int
319 taskqueue_enqueue(struct taskqueue *queue, struct task *task)
320 {
321 	return (taskqueue_enqueue_flags(queue, task, 0));
322 }
323 
324 static void
325 taskqueue_timeout_func(void *arg)
326 {
327 	struct taskqueue *queue;
328 	struct timeout_task *timeout_task;
329 
330 	timeout_task = arg;
331 	queue = timeout_task->q;
332 	KASSERT((timeout_task->f & DT_CALLOUT_ARMED) != 0, ("Stray timeout"));
333 	timeout_task->f &= ~DT_CALLOUT_ARMED;
334 	queue->tq_callouts--;
335 	taskqueue_enqueue_locked(timeout_task->q, &timeout_task->t, 0);
336 	/* The lock is released inside. */
337 }
338 
339 int
340 taskqueue_enqueue_timeout_sbt(struct taskqueue *queue,
341     struct timeout_task *timeout_task, sbintime_t sbt, sbintime_t pr, int flags)
342 {
343 	int res;
344 
345 	TQ_LOCK(queue);
346 	KASSERT(timeout_task->q == NULL || timeout_task->q == queue,
347 	    ("Migrated queue"));
348 	timeout_task->q = queue;
349 	res = timeout_task->t.ta_pending;
350 	if (timeout_task->f & DT_DRAIN_IN_PROGRESS) {
351 		/* Do nothing */
352 		TQ_UNLOCK(queue);
353 		res = -1;
354 	} else if (sbt == 0) {
355 		taskqueue_enqueue_locked(queue, &timeout_task->t, 0);
356 		/* The lock is released inside. */
357 	} else {
358 		if ((timeout_task->f & DT_CALLOUT_ARMED) != 0) {
359 			res++;
360 		} else {
361 			queue->tq_callouts++;
362 			timeout_task->f |= DT_CALLOUT_ARMED;
363 			if (sbt < 0)
364 				sbt = -sbt; /* Ignore overflow. */
365 		}
366 		if (sbt > 0) {
367 			if (queue->tq_spin)
368 				flags |= C_DIRECT_EXEC;
369 			callout_reset_sbt(&timeout_task->c, sbt, pr,
370 			    taskqueue_timeout_func, timeout_task, flags);
371 		}
372 		TQ_UNLOCK(queue);
373 	}
374 	return (res);
375 }
376 
377 int
378 taskqueue_enqueue_timeout(struct taskqueue *queue,
379     struct timeout_task *ttask, int ticks)
380 {
381 
382 	return (taskqueue_enqueue_timeout_sbt(queue, ttask, ticks * tick_sbt,
383 	    0, C_HARDCLOCK));
384 }
385 
386 static void
387 taskqueue_task_nop_fn(void *context, int pending)
388 {
389 }
390 
391 /*
392  * Block until all currently queued tasks in this taskqueue
393  * have begun execution.  Tasks queued during execution of
394  * this function are ignored.
395  */
396 static int
397 taskqueue_drain_tq_queue(struct taskqueue *queue)
398 {
399 	struct task t_barrier;
400 
401 	if (STAILQ_EMPTY(&queue->tq_queue))
402 		return (0);
403 
404 	/*
405 	 * Enqueue our barrier after all current tasks, but with
406 	 * the highest priority so that newly queued tasks cannot
407 	 * pass it.  Because of the high priority, we can not use
408 	 * taskqueue_enqueue_locked directly (which drops the lock
409 	 * anyway) so just insert it at tail while we have the
410 	 * queue lock.
411 	 */
412 	TASK_INIT(&t_barrier, UCHAR_MAX, taskqueue_task_nop_fn, &t_barrier);
413 	STAILQ_INSERT_TAIL(&queue->tq_queue, &t_barrier, ta_link);
414 	queue->tq_hint = &t_barrier;
415 	t_barrier.ta_pending = 1;
416 
417 	/*
418 	 * Once the barrier has executed, all previously queued tasks
419 	 * have completed or are currently executing.
420 	 */
421 	while (t_barrier.ta_pending != 0)
422 		TQ_SLEEP(queue, &t_barrier, "tq_qdrain");
423 	return (1);
424 }
425 
426 /*
427  * Block until all currently executing tasks for this taskqueue
428  * complete.  Tasks that begin execution during the execution
429  * of this function are ignored.
430  */
431 static int
432 taskqueue_drain_tq_active(struct taskqueue *queue)
433 {
434 	struct taskqueue_busy *tb;
435 	u_int seq;
436 
437 	if (LIST_EMPTY(&queue->tq_active))
438 		return (0);
439 
440 	/* Block taskq_terminate().*/
441 	queue->tq_callouts++;
442 
443 	/* Wait for any active task with sequence from the past. */
444 	seq = queue->tq_seq;
445 restart:
446 	LIST_FOREACH(tb, &queue->tq_active, tb_link) {
447 		if ((int)(tb->tb_seq - seq) <= 0) {
448 			TQ_SLEEP(queue, tb->tb_running, "tq_adrain");
449 			goto restart;
450 		}
451 	}
452 
453 	/* Release taskqueue_terminate(). */
454 	queue->tq_callouts--;
455 	if ((queue->tq_flags & TQ_FLAGS_ACTIVE) == 0)
456 		wakeup_one(queue->tq_threads);
457 	return (1);
458 }
459 
460 void
461 taskqueue_block(struct taskqueue *queue)
462 {
463 
464 	TQ_LOCK(queue);
465 	queue->tq_flags |= TQ_FLAGS_BLOCKED;
466 	TQ_UNLOCK(queue);
467 }
468 
469 void
470 taskqueue_unblock(struct taskqueue *queue)
471 {
472 
473 	TQ_LOCK(queue);
474 	queue->tq_flags &= ~TQ_FLAGS_BLOCKED;
475 	if (!STAILQ_EMPTY(&queue->tq_queue))
476 		queue->tq_enqueue(queue->tq_context);
477 	TQ_UNLOCK(queue);
478 }
479 
480 static void
481 taskqueue_run_locked(struct taskqueue *queue)
482 {
483 	struct epoch_tracker et;
484 	struct taskqueue_busy tb;
485 	struct task *task;
486 	bool in_net_epoch;
487 	int pending;
488 
489 	KASSERT(queue != NULL, ("tq is NULL"));
490 	TQ_ASSERT_LOCKED(queue);
491 	tb.tb_running = NULL;
492 	LIST_INSERT_HEAD(&queue->tq_active, &tb, tb_link);
493 	in_net_epoch = false;
494 
495 	while ((task = STAILQ_FIRST(&queue->tq_queue)) != NULL) {
496 		STAILQ_REMOVE_HEAD(&queue->tq_queue, ta_link);
497 		if (queue->tq_hint == task)
498 			queue->tq_hint = NULL;
499 		pending = task->ta_pending;
500 		task->ta_pending = 0;
501 		tb.tb_running = task;
502 		tb.tb_seq = ++queue->tq_seq;
503 		tb.tb_canceling = false;
504 		TQ_UNLOCK(queue);
505 
506 		KASSERT(task->ta_func != NULL, ("task->ta_func is NULL"));
507 		if (!in_net_epoch && TASK_IS_NET(task)) {
508 			in_net_epoch = true;
509 			NET_EPOCH_ENTER(et);
510 		} else if (in_net_epoch && !TASK_IS_NET(task)) {
511 			NET_EPOCH_EXIT(et);
512 			in_net_epoch = false;
513 		}
514 		task->ta_func(task->ta_context, pending);
515 
516 		TQ_LOCK(queue);
517 		wakeup(task);
518 	}
519 	if (in_net_epoch)
520 		NET_EPOCH_EXIT(et);
521 	LIST_REMOVE(&tb, tb_link);
522 }
523 
524 void
525 taskqueue_run(struct taskqueue *queue)
526 {
527 
528 	TQ_LOCK(queue);
529 	taskqueue_run_locked(queue);
530 	TQ_UNLOCK(queue);
531 }
532 
533 /*
534  * Only use this function in single threaded contexts. It returns
535  * non-zero if the given task is either pending or running. Else the
536  * task is idle and can be queued again or freed.
537  */
538 int
539 taskqueue_poll_is_busy(struct taskqueue *queue, struct task *task)
540 {
541 	int retval;
542 
543 	TQ_LOCK(queue);
544 	retval = task->ta_pending > 0 || task_get_busy(queue, task) != NULL;
545 	TQ_UNLOCK(queue);
546 
547 	return (retval);
548 }
549 
550 static int
551 taskqueue_cancel_locked(struct taskqueue *queue, struct task *task,
552     u_int *pendp)
553 {
554 	struct taskqueue_busy *tb;
555 	int retval = 0;
556 
557 	if (task->ta_pending > 0) {
558 		STAILQ_REMOVE(&queue->tq_queue, task, task, ta_link);
559 		if (queue->tq_hint == task)
560 			queue->tq_hint = NULL;
561 	}
562 	if (pendp != NULL)
563 		*pendp = task->ta_pending;
564 	task->ta_pending = 0;
565 	tb = task_get_busy(queue, task);
566 	if (tb != NULL) {
567 		tb->tb_canceling = true;
568 		retval = EBUSY;
569 	}
570 
571 	return (retval);
572 }
573 
574 int
575 taskqueue_cancel(struct taskqueue *queue, struct task *task, u_int *pendp)
576 {
577 	int error;
578 
579 	TQ_LOCK(queue);
580 	error = taskqueue_cancel_locked(queue, task, pendp);
581 	TQ_UNLOCK(queue);
582 
583 	return (error);
584 }
585 
586 int
587 taskqueue_cancel_timeout(struct taskqueue *queue,
588     struct timeout_task *timeout_task, u_int *pendp)
589 {
590 	u_int pending, pending1;
591 	int error;
592 
593 	TQ_LOCK(queue);
594 	pending = !!(callout_stop(&timeout_task->c) > 0);
595 	error = taskqueue_cancel_locked(queue, &timeout_task->t, &pending1);
596 	if ((timeout_task->f & DT_CALLOUT_ARMED) != 0) {
597 		timeout_task->f &= ~DT_CALLOUT_ARMED;
598 		queue->tq_callouts--;
599 	}
600 	TQ_UNLOCK(queue);
601 
602 	if (pendp != NULL)
603 		*pendp = pending + pending1;
604 	return (error);
605 }
606 
607 void
608 taskqueue_drain(struct taskqueue *queue, struct task *task)
609 {
610 
611 	if (!queue->tq_spin)
612 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__);
613 
614 	TQ_LOCK(queue);
615 	while (task->ta_pending != 0 || task_get_busy(queue, task) != NULL)
616 		TQ_SLEEP(queue, task, "tq_drain");
617 	TQ_UNLOCK(queue);
618 }
619 
620 void
621 taskqueue_drain_all(struct taskqueue *queue)
622 {
623 
624 	if (!queue->tq_spin)
625 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__);
626 
627 	TQ_LOCK(queue);
628 	(void)taskqueue_drain_tq_queue(queue);
629 	(void)taskqueue_drain_tq_active(queue);
630 	TQ_UNLOCK(queue);
631 }
632 
633 void
634 taskqueue_drain_timeout(struct taskqueue *queue,
635     struct timeout_task *timeout_task)
636 {
637 
638 	/*
639 	 * Set flag to prevent timer from re-starting during drain:
640 	 */
641 	TQ_LOCK(queue);
642 	KASSERT((timeout_task->f & DT_DRAIN_IN_PROGRESS) == 0,
643 	    ("Drain already in progress"));
644 	timeout_task->f |= DT_DRAIN_IN_PROGRESS;
645 	TQ_UNLOCK(queue);
646 
647 	callout_drain(&timeout_task->c);
648 	taskqueue_drain(queue, &timeout_task->t);
649 
650 	/*
651 	 * Clear flag to allow timer to re-start:
652 	 */
653 	TQ_LOCK(queue);
654 	timeout_task->f &= ~DT_DRAIN_IN_PROGRESS;
655 	TQ_UNLOCK(queue);
656 }
657 
658 void
659 taskqueue_quiesce(struct taskqueue *queue)
660 {
661 	int ret;
662 
663 	TQ_LOCK(queue);
664 	do {
665 		ret = taskqueue_drain_tq_queue(queue);
666 		if (ret == 0)
667 			ret = taskqueue_drain_tq_active(queue);
668 	} while (ret != 0);
669 	TQ_UNLOCK(queue);
670 }
671 
672 static void
673 taskqueue_swi_enqueue(void *context)
674 {
675 	swi_sched(taskqueue_ih, 0);
676 }
677 
678 static void
679 taskqueue_swi_run(void *dummy)
680 {
681 	taskqueue_run(taskqueue_swi);
682 }
683 
684 static void
685 taskqueue_swi_giant_enqueue(void *context)
686 {
687 	swi_sched(taskqueue_giant_ih, 0);
688 }
689 
690 static void
691 taskqueue_swi_giant_run(void *dummy)
692 {
693 	taskqueue_run(taskqueue_swi_giant);
694 }
695 
696 static int
697 _taskqueue_start_threads(struct taskqueue **tqp, int count, int pri,
698     cpuset_t *mask, struct proc *p, const char *name, va_list ap)
699 {
700 	char ktname[MAXCOMLEN + 1];
701 	struct thread *td;
702 	struct taskqueue *tq;
703 	int i, error;
704 
705 	if (count <= 0)
706 		return (EINVAL);
707 
708 	vsnprintf(ktname, sizeof(ktname), name, ap);
709 	tq = *tqp;
710 
711 	tq->tq_threads = malloc(sizeof(struct thread *) * count, M_TASKQUEUE,
712 	    M_NOWAIT | M_ZERO);
713 	if (tq->tq_threads == NULL) {
714 		printf("%s: no memory for %s threads\n", __func__, ktname);
715 		return (ENOMEM);
716 	}
717 
718 	for (i = 0; i < count; i++) {
719 		if (count == 1)
720 			error = kthread_add(taskqueue_thread_loop, tqp, p,
721 			    &tq->tq_threads[i], RFSTOPPED, 0, "%s", ktname);
722 		else
723 			error = kthread_add(taskqueue_thread_loop, tqp, p,
724 			    &tq->tq_threads[i], RFSTOPPED, 0,
725 			    "%s_%d", ktname, i);
726 		if (error) {
727 			/* should be ok to continue, taskqueue_free will dtrt */
728 			printf("%s: kthread_add(%s): error %d", __func__,
729 			    ktname, error);
730 			tq->tq_threads[i] = NULL;		/* paranoid */
731 		} else
732 			tq->tq_tcount++;
733 	}
734 	if (tq->tq_tcount == 0) {
735 		free(tq->tq_threads, M_TASKQUEUE);
736 		tq->tq_threads = NULL;
737 		return (ENOMEM);
738 	}
739 	for (i = 0; i < count; i++) {
740 		if (tq->tq_threads[i] == NULL)
741 			continue;
742 		td = tq->tq_threads[i];
743 		if (mask) {
744 			error = cpuset_setthread(td->td_tid, mask);
745 			/*
746 			 * Failing to pin is rarely an actual fatal error;
747 			 * it'll just affect performance.
748 			 */
749 			if (error)
750 				printf("%s: curthread=%llu: can't pin; "
751 				    "error=%d\n",
752 				    __func__,
753 				    (unsigned long long) td->td_tid,
754 				    error);
755 		}
756 		thread_lock(td);
757 		sched_prio(td, pri);
758 		sched_add(td, SRQ_BORING);
759 	}
760 
761 	return (0);
762 }
763 
764 int
765 taskqueue_start_threads(struct taskqueue **tqp, int count, int pri,
766     const char *name, ...)
767 {
768 	va_list ap;
769 	int error;
770 
771 	va_start(ap, name);
772 	error = _taskqueue_start_threads(tqp, count, pri, NULL, NULL, name, ap);
773 	va_end(ap);
774 	return (error);
775 }
776 
777 int
778 taskqueue_start_threads_in_proc(struct taskqueue **tqp, int count, int pri,
779     struct proc *proc, const char *name, ...)
780 {
781 	va_list ap;
782 	int error;
783 
784 	va_start(ap, name);
785 	error = _taskqueue_start_threads(tqp, count, pri, NULL, proc, name, ap);
786 	va_end(ap);
787 	return (error);
788 }
789 
790 int
791 taskqueue_start_threads_cpuset(struct taskqueue **tqp, int count, int pri,
792     cpuset_t *mask, const char *name, ...)
793 {
794 	va_list ap;
795 	int error;
796 
797 	va_start(ap, name);
798 	error = _taskqueue_start_threads(tqp, count, pri, mask, NULL, name, ap);
799 	va_end(ap);
800 	return (error);
801 }
802 
803 static inline void
804 taskqueue_run_callback(struct taskqueue *tq,
805     enum taskqueue_callback_type cb_type)
806 {
807 	taskqueue_callback_fn tq_callback;
808 
809 	TQ_ASSERT_UNLOCKED(tq);
810 	tq_callback = tq->tq_callbacks[cb_type];
811 	if (tq_callback != NULL)
812 		tq_callback(tq->tq_cb_contexts[cb_type]);
813 }
814 
815 void
816 taskqueue_thread_loop(void *arg)
817 {
818 	struct taskqueue **tqp, *tq;
819 
820 	tqp = arg;
821 	tq = *tqp;
822 	taskqueue_run_callback(tq, TASKQUEUE_CALLBACK_TYPE_INIT);
823 	TQ_LOCK(tq);
824 	while ((tq->tq_flags & TQ_FLAGS_ACTIVE) != 0) {
825 		/* XXX ? */
826 		taskqueue_run_locked(tq);
827 		/*
828 		 * Because taskqueue_run() can drop tq_mutex, we need to
829 		 * check if the TQ_FLAGS_ACTIVE flag wasn't removed in the
830 		 * meantime, which means we missed a wakeup.
831 		 */
832 		if ((tq->tq_flags & TQ_FLAGS_ACTIVE) == 0)
833 			break;
834 		TQ_SLEEP(tq, tq, "-");
835 	}
836 	taskqueue_run_locked(tq);
837 	/*
838 	 * This thread is on its way out, so just drop the lock temporarily
839 	 * in order to call the shutdown callback.  This allows the callback
840 	 * to look at the taskqueue, even just before it dies.
841 	 */
842 	TQ_UNLOCK(tq);
843 	taskqueue_run_callback(tq, TASKQUEUE_CALLBACK_TYPE_SHUTDOWN);
844 	TQ_LOCK(tq);
845 
846 	/* rendezvous with thread that asked us to terminate */
847 	tq->tq_tcount--;
848 	wakeup_one(tq->tq_threads);
849 	TQ_UNLOCK(tq);
850 	kthread_exit();
851 }
852 
853 void
854 taskqueue_thread_enqueue(void *context)
855 {
856 	struct taskqueue **tqp, *tq;
857 
858 	tqp = context;
859 	tq = *tqp;
860 	wakeup_any(tq);
861 }
862 
863 TASKQUEUE_DEFINE(swi, taskqueue_swi_enqueue, NULL,
864 		 swi_add(NULL, "task queue", taskqueue_swi_run, NULL, SWI_TQ,
865 		     INTR_MPSAFE, &taskqueue_ih));
866 
867 TASKQUEUE_DEFINE(swi_giant, taskqueue_swi_giant_enqueue, NULL,
868 		 swi_add(NULL, "Giant taskq", taskqueue_swi_giant_run,
869 		     NULL, SWI_TQ_GIANT, 0, &taskqueue_giant_ih));
870 
871 TASKQUEUE_DEFINE_THREAD(thread);
872 
873 struct taskqueue *
874 taskqueue_create_fast(const char *name, int mflags,
875 		 taskqueue_enqueue_fn enqueue, void *context)
876 {
877 	return _taskqueue_create(name, mflags, enqueue, context,
878 			MTX_SPIN, "fast_taskqueue");
879 }
880 
881 static void	*taskqueue_fast_ih;
882 
883 static void
884 taskqueue_fast_enqueue(void *context)
885 {
886 	swi_sched(taskqueue_fast_ih, 0);
887 }
888 
889 static void
890 taskqueue_fast_run(void *dummy)
891 {
892 	taskqueue_run(taskqueue_fast);
893 }
894 
895 TASKQUEUE_FAST_DEFINE(fast, taskqueue_fast_enqueue, NULL,
896 	swi_add(NULL, "fast taskq", taskqueue_fast_run, NULL,
897 	SWI_TQ_FAST, INTR_MPSAFE, &taskqueue_fast_ih));
898 
899 int
900 taskqueue_member(struct taskqueue *queue, struct thread *td)
901 {
902 	int i, j, ret = 0;
903 
904 	for (i = 0, j = 0; ; i++) {
905 		if (queue->tq_threads[i] == NULL)
906 			continue;
907 		if (queue->tq_threads[i] == td) {
908 			ret = 1;
909 			break;
910 		}
911 		if (++j >= queue->tq_tcount)
912 			break;
913 	}
914 	return (ret);
915 }
916