xref: /freebsd/sys/kern/subr_taskqueue.c (revision cfd6422a5217410fbd66f7a7a8a64d9d85e61229)
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 	LIST_ENTRY(taskqueue_busy) tb_link;
63 };
64 
65 struct taskqueue {
66 	STAILQ_HEAD(, task)	tq_queue;
67 	LIST_HEAD(, taskqueue_busy) tq_active;
68 	struct task		*tq_hint;
69 	u_int			tq_seq;
70 	int			tq_callouts;
71 	struct mtx_padalign	tq_mutex;
72 	taskqueue_enqueue_fn	tq_enqueue;
73 	void			*tq_context;
74 	char			*tq_name;
75 	struct thread		**tq_threads;
76 	int			tq_tcount;
77 	int			tq_spin;
78 	int			tq_flags;
79 	taskqueue_callback_fn	tq_callbacks[TASKQUEUE_NUM_CALLBACKS];
80 	void			*tq_cb_contexts[TASKQUEUE_NUM_CALLBACKS];
81 };
82 
83 #define	TQ_FLAGS_ACTIVE		(1 << 0)
84 #define	TQ_FLAGS_BLOCKED	(1 << 1)
85 #define	TQ_FLAGS_UNLOCKED_ENQUEUE	(1 << 2)
86 
87 #define	DT_CALLOUT_ARMED	(1 << 0)
88 #define	DT_DRAIN_IN_PROGRESS	(1 << 1)
89 
90 #define	TQ_LOCK(tq)							\
91 	do {								\
92 		if ((tq)->tq_spin)					\
93 			mtx_lock_spin(&(tq)->tq_mutex);			\
94 		else							\
95 			mtx_lock(&(tq)->tq_mutex);			\
96 	} while (0)
97 #define	TQ_ASSERT_LOCKED(tq)	mtx_assert(&(tq)->tq_mutex, MA_OWNED)
98 
99 #define	TQ_UNLOCK(tq)							\
100 	do {								\
101 		if ((tq)->tq_spin)					\
102 			mtx_unlock_spin(&(tq)->tq_mutex);		\
103 		else							\
104 			mtx_unlock(&(tq)->tq_mutex);			\
105 	} while (0)
106 #define	TQ_ASSERT_UNLOCKED(tq)	mtx_assert(&(tq)->tq_mutex, MA_NOTOWNED)
107 
108 void
109 _timeout_task_init(struct taskqueue *queue, struct timeout_task *timeout_task,
110     int priority, task_fn_t func, void *context)
111 {
112 
113 	TASK_INIT(&timeout_task->t, priority, func, context);
114 	callout_init_mtx(&timeout_task->c, &queue->tq_mutex,
115 	    CALLOUT_RETURNUNLOCKED);
116 	timeout_task->q = queue;
117 	timeout_task->f = 0;
118 }
119 
120 static __inline int
121 TQ_SLEEP(struct taskqueue *tq, void *p, const char *wm)
122 {
123 	if (tq->tq_spin)
124 		return (msleep_spin(p, (struct mtx *)&tq->tq_mutex, wm, 0));
125 	return (msleep(p, &tq->tq_mutex, 0, wm, 0));
126 }
127 
128 static struct taskqueue *
129 _taskqueue_create(const char *name, int mflags,
130 		 taskqueue_enqueue_fn enqueue, void *context,
131 		 int mtxflags, const char *mtxname __unused)
132 {
133 	struct taskqueue *queue;
134 	char *tq_name;
135 
136 	tq_name = malloc(TASKQUEUE_NAMELEN, M_TASKQUEUE, mflags | M_ZERO);
137 	if (tq_name == NULL)
138 		return (NULL);
139 
140 	queue = malloc(sizeof(struct taskqueue), M_TASKQUEUE, mflags | M_ZERO);
141 	if (queue == NULL) {
142 		free(tq_name, M_TASKQUEUE);
143 		return (NULL);
144 	}
145 
146 	snprintf(tq_name, TASKQUEUE_NAMELEN, "%s", (name) ? name : "taskqueue");
147 
148 	STAILQ_INIT(&queue->tq_queue);
149 	LIST_INIT(&queue->tq_active);
150 	queue->tq_enqueue = enqueue;
151 	queue->tq_context = context;
152 	queue->tq_name = tq_name;
153 	queue->tq_spin = (mtxflags & MTX_SPIN) != 0;
154 	queue->tq_flags |= TQ_FLAGS_ACTIVE;
155 	if (enqueue == taskqueue_fast_enqueue ||
156 	    enqueue == taskqueue_swi_enqueue ||
157 	    enqueue == taskqueue_swi_giant_enqueue ||
158 	    enqueue == taskqueue_thread_enqueue)
159 		queue->tq_flags |= TQ_FLAGS_UNLOCKED_ENQUEUE;
160 	mtx_init(&queue->tq_mutex, tq_name, NULL, mtxflags);
161 
162 	return (queue);
163 }
164 
165 struct taskqueue *
166 taskqueue_create(const char *name, int mflags,
167 		 taskqueue_enqueue_fn enqueue, void *context)
168 {
169 
170 	return _taskqueue_create(name, mflags, enqueue, context,
171 			MTX_DEF, name);
172 }
173 
174 void
175 taskqueue_set_callback(struct taskqueue *queue,
176     enum taskqueue_callback_type cb_type, taskqueue_callback_fn callback,
177     void *context)
178 {
179 
180 	KASSERT(((cb_type >= TASKQUEUE_CALLBACK_TYPE_MIN) &&
181 	    (cb_type <= TASKQUEUE_CALLBACK_TYPE_MAX)),
182 	    ("Callback type %d not valid, must be %d-%d", cb_type,
183 	    TASKQUEUE_CALLBACK_TYPE_MIN, TASKQUEUE_CALLBACK_TYPE_MAX));
184 	KASSERT((queue->tq_callbacks[cb_type] == NULL),
185 	    ("Re-initialization of taskqueue callback?"));
186 
187 	queue->tq_callbacks[cb_type] = callback;
188 	queue->tq_cb_contexts[cb_type] = context;
189 }
190 
191 /*
192  * Signal a taskqueue thread to terminate.
193  */
194 static void
195 taskqueue_terminate(struct thread **pp, struct taskqueue *tq)
196 {
197 
198 	while (tq->tq_tcount > 0 || tq->tq_callouts > 0) {
199 		wakeup(tq);
200 		TQ_SLEEP(tq, pp, "tq_destroy");
201 	}
202 }
203 
204 void
205 taskqueue_free(struct taskqueue *queue)
206 {
207 
208 	TQ_LOCK(queue);
209 	queue->tq_flags &= ~TQ_FLAGS_ACTIVE;
210 	taskqueue_terminate(queue->tq_threads, queue);
211 	KASSERT(LIST_EMPTY(&queue->tq_active), ("Tasks still running?"));
212 	KASSERT(queue->tq_callouts == 0, ("Armed timeout tasks"));
213 	mtx_destroy(&queue->tq_mutex);
214 	free(queue->tq_threads, M_TASKQUEUE);
215 	free(queue->tq_name, M_TASKQUEUE);
216 	free(queue, M_TASKQUEUE);
217 }
218 
219 static int
220 taskqueue_enqueue_locked(struct taskqueue *queue, struct task *task)
221 {
222 	struct task *ins;
223 	struct task *prev;
224 
225 	KASSERT(task->ta_func != NULL, ("enqueueing task with NULL func"));
226 	/*
227 	 * Count multiple enqueues.
228 	 */
229 	if (task->ta_pending) {
230 		if (task->ta_pending < USHRT_MAX)
231 			task->ta_pending++;
232 		TQ_UNLOCK(queue);
233 		return (0);
234 	}
235 
236 	/*
237 	 * Optimise cases when all tasks use small set of priorities.
238 	 * In case of only one priority we always insert at the end.
239 	 * In case of two tq_hint typically gives the insertion point.
240 	 * In case of more then two tq_hint should halve the search.
241 	 */
242 	prev = STAILQ_LAST(&queue->tq_queue, task, ta_link);
243 	if (!prev || prev->ta_priority >= task->ta_priority) {
244 		STAILQ_INSERT_TAIL(&queue->tq_queue, task, ta_link);
245 	} else {
246 		prev = queue->tq_hint;
247 		if (prev && prev->ta_priority >= task->ta_priority) {
248 			ins = STAILQ_NEXT(prev, ta_link);
249 		} else {
250 			prev = NULL;
251 			ins = STAILQ_FIRST(&queue->tq_queue);
252 		}
253 		for (; ins; prev = ins, ins = STAILQ_NEXT(ins, ta_link))
254 			if (ins->ta_priority < task->ta_priority)
255 				break;
256 
257 		if (prev) {
258 			STAILQ_INSERT_AFTER(&queue->tq_queue, prev, task, ta_link);
259 			queue->tq_hint = task;
260 		} else
261 			STAILQ_INSERT_HEAD(&queue->tq_queue, task, ta_link);
262 	}
263 
264 	task->ta_pending = 1;
265 	if ((queue->tq_flags & TQ_FLAGS_UNLOCKED_ENQUEUE) != 0)
266 		TQ_UNLOCK(queue);
267 	if ((queue->tq_flags & TQ_FLAGS_BLOCKED) == 0)
268 		queue->tq_enqueue(queue->tq_context);
269 	if ((queue->tq_flags & TQ_FLAGS_UNLOCKED_ENQUEUE) == 0)
270 		TQ_UNLOCK(queue);
271 
272 	/* Return with lock released. */
273 	return (0);
274 }
275 
276 int
277 taskqueue_enqueue(struct taskqueue *queue, struct task *task)
278 {
279 	int res;
280 
281 	TQ_LOCK(queue);
282 	res = taskqueue_enqueue_locked(queue, task);
283 	/* The lock is released inside. */
284 
285 	return (res);
286 }
287 
288 static void
289 taskqueue_timeout_func(void *arg)
290 {
291 	struct taskqueue *queue;
292 	struct timeout_task *timeout_task;
293 
294 	timeout_task = arg;
295 	queue = timeout_task->q;
296 	KASSERT((timeout_task->f & DT_CALLOUT_ARMED) != 0, ("Stray timeout"));
297 	timeout_task->f &= ~DT_CALLOUT_ARMED;
298 	queue->tq_callouts--;
299 	taskqueue_enqueue_locked(timeout_task->q, &timeout_task->t);
300 	/* The lock is released inside. */
301 }
302 
303 int
304 taskqueue_enqueue_timeout_sbt(struct taskqueue *queue,
305     struct timeout_task *timeout_task, sbintime_t sbt, sbintime_t pr, int flags)
306 {
307 	int res;
308 
309 	TQ_LOCK(queue);
310 	KASSERT(timeout_task->q == NULL || timeout_task->q == queue,
311 	    ("Migrated queue"));
312 	KASSERT(!queue->tq_spin, ("Timeout for spin-queue"));
313 	timeout_task->q = queue;
314 	res = timeout_task->t.ta_pending;
315 	if (timeout_task->f & DT_DRAIN_IN_PROGRESS) {
316 		/* Do nothing */
317 		TQ_UNLOCK(queue);
318 		res = -1;
319 	} else if (sbt == 0) {
320 		taskqueue_enqueue_locked(queue, &timeout_task->t);
321 		/* The lock is released inside. */
322 	} else {
323 		if ((timeout_task->f & DT_CALLOUT_ARMED) != 0) {
324 			res++;
325 		} else {
326 			queue->tq_callouts++;
327 			timeout_task->f |= DT_CALLOUT_ARMED;
328 			if (sbt < 0)
329 				sbt = -sbt; /* Ignore overflow. */
330 		}
331 		if (sbt > 0) {
332 			callout_reset_sbt(&timeout_task->c, sbt, pr,
333 			    taskqueue_timeout_func, timeout_task, flags);
334 		}
335 		TQ_UNLOCK(queue);
336 	}
337 	return (res);
338 }
339 
340 int
341 taskqueue_enqueue_timeout(struct taskqueue *queue,
342     struct timeout_task *ttask, int ticks)
343 {
344 
345 	return (taskqueue_enqueue_timeout_sbt(queue, ttask, ticks * tick_sbt,
346 	    0, 0));
347 }
348 
349 static void
350 taskqueue_task_nop_fn(void *context, int pending)
351 {
352 }
353 
354 /*
355  * Block until all currently queued tasks in this taskqueue
356  * have begun execution.  Tasks queued during execution of
357  * this function are ignored.
358  */
359 static int
360 taskqueue_drain_tq_queue(struct taskqueue *queue)
361 {
362 	struct task t_barrier;
363 
364 	if (STAILQ_EMPTY(&queue->tq_queue))
365 		return (0);
366 
367 	/*
368 	 * Enqueue our barrier after all current tasks, but with
369 	 * the highest priority so that newly queued tasks cannot
370 	 * pass it.  Because of the high priority, we can not use
371 	 * taskqueue_enqueue_locked directly (which drops the lock
372 	 * anyway) so just insert it at tail while we have the
373 	 * queue lock.
374 	 */
375 	TASK_INIT(&t_barrier, UCHAR_MAX, taskqueue_task_nop_fn, &t_barrier);
376 	STAILQ_INSERT_TAIL(&queue->tq_queue, &t_barrier, ta_link);
377 	queue->tq_hint = &t_barrier;
378 	t_barrier.ta_pending = 1;
379 
380 	/*
381 	 * Once the barrier has executed, all previously queued tasks
382 	 * have completed or are currently executing.
383 	 */
384 	while (t_barrier.ta_pending != 0)
385 		TQ_SLEEP(queue, &t_barrier, "tq_qdrain");
386 	return (1);
387 }
388 
389 /*
390  * Block until all currently executing tasks for this taskqueue
391  * complete.  Tasks that begin execution during the execution
392  * of this function are ignored.
393  */
394 static int
395 taskqueue_drain_tq_active(struct taskqueue *queue)
396 {
397 	struct taskqueue_busy *tb;
398 	u_int seq;
399 
400 	if (LIST_EMPTY(&queue->tq_active))
401 		return (0);
402 
403 	/* Block taskq_terminate().*/
404 	queue->tq_callouts++;
405 
406 	/* Wait for any active task with sequence from the past. */
407 	seq = queue->tq_seq;
408 restart:
409 	LIST_FOREACH(tb, &queue->tq_active, tb_link) {
410 		if ((int)(tb->tb_seq - seq) <= 0) {
411 			TQ_SLEEP(queue, tb->tb_running, "tq_adrain");
412 			goto restart;
413 		}
414 	}
415 
416 	/* Release taskqueue_terminate(). */
417 	queue->tq_callouts--;
418 	if ((queue->tq_flags & TQ_FLAGS_ACTIVE) == 0)
419 		wakeup_one(queue->tq_threads);
420 	return (1);
421 }
422 
423 void
424 taskqueue_block(struct taskqueue *queue)
425 {
426 
427 	TQ_LOCK(queue);
428 	queue->tq_flags |= TQ_FLAGS_BLOCKED;
429 	TQ_UNLOCK(queue);
430 }
431 
432 void
433 taskqueue_unblock(struct taskqueue *queue)
434 {
435 
436 	TQ_LOCK(queue);
437 	queue->tq_flags &= ~TQ_FLAGS_BLOCKED;
438 	if (!STAILQ_EMPTY(&queue->tq_queue))
439 		queue->tq_enqueue(queue->tq_context);
440 	TQ_UNLOCK(queue);
441 }
442 
443 static void
444 taskqueue_run_locked(struct taskqueue *queue)
445 {
446 	struct epoch_tracker et;
447 	struct taskqueue_busy tb;
448 	struct task *task;
449 	bool in_net_epoch;
450 	int pending;
451 
452 	KASSERT(queue != NULL, ("tq is NULL"));
453 	TQ_ASSERT_LOCKED(queue);
454 	tb.tb_running = NULL;
455 	LIST_INSERT_HEAD(&queue->tq_active, &tb, tb_link);
456 	in_net_epoch = false;
457 
458 	while ((task = STAILQ_FIRST(&queue->tq_queue)) != NULL) {
459 		STAILQ_REMOVE_HEAD(&queue->tq_queue, ta_link);
460 		if (queue->tq_hint == task)
461 			queue->tq_hint = NULL;
462 		pending = task->ta_pending;
463 		task->ta_pending = 0;
464 		tb.tb_running = task;
465 		tb.tb_seq = ++queue->tq_seq;
466 		TQ_UNLOCK(queue);
467 
468 		KASSERT(task->ta_func != NULL, ("task->ta_func is NULL"));
469 		if (!in_net_epoch && TASK_IS_NET(task)) {
470 			in_net_epoch = true;
471 			NET_EPOCH_ENTER(et);
472 		} else if (in_net_epoch && !TASK_IS_NET(task)) {
473 			NET_EPOCH_EXIT(et);
474 			in_net_epoch = false;
475 		}
476 		task->ta_func(task->ta_context, pending);
477 
478 		TQ_LOCK(queue);
479 		wakeup(task);
480 	}
481 	if (in_net_epoch)
482 		NET_EPOCH_EXIT(et);
483 	LIST_REMOVE(&tb, tb_link);
484 }
485 
486 void
487 taskqueue_run(struct taskqueue *queue)
488 {
489 
490 	TQ_LOCK(queue);
491 	taskqueue_run_locked(queue);
492 	TQ_UNLOCK(queue);
493 }
494 
495 static int
496 task_is_running(struct taskqueue *queue, struct task *task)
497 {
498 	struct taskqueue_busy *tb;
499 
500 	TQ_ASSERT_LOCKED(queue);
501 	LIST_FOREACH(tb, &queue->tq_active, tb_link) {
502 		if (tb->tb_running == task)
503 			return (1);
504 	}
505 	return (0);
506 }
507 
508 /*
509  * Only use this function in single threaded contexts. It returns
510  * non-zero if the given task is either pending or running. Else the
511  * task is idle and can be queued again or freed.
512  */
513 int
514 taskqueue_poll_is_busy(struct taskqueue *queue, struct task *task)
515 {
516 	int retval;
517 
518 	TQ_LOCK(queue);
519 	retval = task->ta_pending > 0 || task_is_running(queue, task);
520 	TQ_UNLOCK(queue);
521 
522 	return (retval);
523 }
524 
525 static int
526 taskqueue_cancel_locked(struct taskqueue *queue, struct task *task,
527     u_int *pendp)
528 {
529 
530 	if (task->ta_pending > 0) {
531 		STAILQ_REMOVE(&queue->tq_queue, task, task, ta_link);
532 		if (queue->tq_hint == task)
533 			queue->tq_hint = NULL;
534 	}
535 	if (pendp != NULL)
536 		*pendp = task->ta_pending;
537 	task->ta_pending = 0;
538 	return (task_is_running(queue, task) ? EBUSY : 0);
539 }
540 
541 int
542 taskqueue_cancel(struct taskqueue *queue, struct task *task, u_int *pendp)
543 {
544 	int error;
545 
546 	TQ_LOCK(queue);
547 	error = taskqueue_cancel_locked(queue, task, pendp);
548 	TQ_UNLOCK(queue);
549 
550 	return (error);
551 }
552 
553 int
554 taskqueue_cancel_timeout(struct taskqueue *queue,
555     struct timeout_task *timeout_task, u_int *pendp)
556 {
557 	u_int pending, pending1;
558 	int error;
559 
560 	TQ_LOCK(queue);
561 	pending = !!(callout_stop(&timeout_task->c) > 0);
562 	error = taskqueue_cancel_locked(queue, &timeout_task->t, &pending1);
563 	if ((timeout_task->f & DT_CALLOUT_ARMED) != 0) {
564 		timeout_task->f &= ~DT_CALLOUT_ARMED;
565 		queue->tq_callouts--;
566 	}
567 	TQ_UNLOCK(queue);
568 
569 	if (pendp != NULL)
570 		*pendp = pending + pending1;
571 	return (error);
572 }
573 
574 void
575 taskqueue_drain(struct taskqueue *queue, struct task *task)
576 {
577 
578 	if (!queue->tq_spin)
579 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__);
580 
581 	TQ_LOCK(queue);
582 	while (task->ta_pending != 0 || task_is_running(queue, task))
583 		TQ_SLEEP(queue, task, "tq_drain");
584 	TQ_UNLOCK(queue);
585 }
586 
587 void
588 taskqueue_drain_all(struct taskqueue *queue)
589 {
590 
591 	if (!queue->tq_spin)
592 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, __func__);
593 
594 	TQ_LOCK(queue);
595 	(void)taskqueue_drain_tq_queue(queue);
596 	(void)taskqueue_drain_tq_active(queue);
597 	TQ_UNLOCK(queue);
598 }
599 
600 void
601 taskqueue_drain_timeout(struct taskqueue *queue,
602     struct timeout_task *timeout_task)
603 {
604 
605 	/*
606 	 * Set flag to prevent timer from re-starting during drain:
607 	 */
608 	TQ_LOCK(queue);
609 	KASSERT((timeout_task->f & DT_DRAIN_IN_PROGRESS) == 0,
610 	    ("Drain already in progress"));
611 	timeout_task->f |= DT_DRAIN_IN_PROGRESS;
612 	TQ_UNLOCK(queue);
613 
614 	callout_drain(&timeout_task->c);
615 	taskqueue_drain(queue, &timeout_task->t);
616 
617 	/*
618 	 * Clear flag to allow timer to re-start:
619 	 */
620 	TQ_LOCK(queue);
621 	timeout_task->f &= ~DT_DRAIN_IN_PROGRESS;
622 	TQ_UNLOCK(queue);
623 }
624 
625 void
626 taskqueue_quiesce(struct taskqueue *queue)
627 {
628 	int ret;
629 
630 	TQ_LOCK(queue);
631 	do {
632 		ret = taskqueue_drain_tq_queue(queue);
633 		if (ret == 0)
634 			ret = taskqueue_drain_tq_active(queue);
635 	} while (ret != 0);
636 	TQ_UNLOCK(queue);
637 }
638 
639 static void
640 taskqueue_swi_enqueue(void *context)
641 {
642 	swi_sched(taskqueue_ih, 0);
643 }
644 
645 static void
646 taskqueue_swi_run(void *dummy)
647 {
648 	taskqueue_run(taskqueue_swi);
649 }
650 
651 static void
652 taskqueue_swi_giant_enqueue(void *context)
653 {
654 	swi_sched(taskqueue_giant_ih, 0);
655 }
656 
657 static void
658 taskqueue_swi_giant_run(void *dummy)
659 {
660 	taskqueue_run(taskqueue_swi_giant);
661 }
662 
663 static int
664 _taskqueue_start_threads(struct taskqueue **tqp, int count, int pri,
665     cpuset_t *mask, struct proc *p, const char *name, va_list ap)
666 {
667 	char ktname[MAXCOMLEN + 1];
668 	struct thread *td;
669 	struct taskqueue *tq;
670 	int i, error;
671 
672 	if (count <= 0)
673 		return (EINVAL);
674 
675 	vsnprintf(ktname, sizeof(ktname), name, ap);
676 	tq = *tqp;
677 
678 	tq->tq_threads = malloc(sizeof(struct thread *) * count, M_TASKQUEUE,
679 	    M_NOWAIT | M_ZERO);
680 	if (tq->tq_threads == NULL) {
681 		printf("%s: no memory for %s threads\n", __func__, ktname);
682 		return (ENOMEM);
683 	}
684 
685 	for (i = 0; i < count; i++) {
686 		if (count == 1)
687 			error = kthread_add(taskqueue_thread_loop, tqp, p,
688 			    &tq->tq_threads[i], RFSTOPPED, 0, "%s", ktname);
689 		else
690 			error = kthread_add(taskqueue_thread_loop, tqp, p,
691 			    &tq->tq_threads[i], RFSTOPPED, 0,
692 			    "%s_%d", ktname, i);
693 		if (error) {
694 			/* should be ok to continue, taskqueue_free will dtrt */
695 			printf("%s: kthread_add(%s): error %d", __func__,
696 			    ktname, error);
697 			tq->tq_threads[i] = NULL;		/* paranoid */
698 		} else
699 			tq->tq_tcount++;
700 	}
701 	if (tq->tq_tcount == 0) {
702 		free(tq->tq_threads, M_TASKQUEUE);
703 		tq->tq_threads = NULL;
704 		return (ENOMEM);
705 	}
706 	for (i = 0; i < count; i++) {
707 		if (tq->tq_threads[i] == NULL)
708 			continue;
709 		td = tq->tq_threads[i];
710 		if (mask) {
711 			error = cpuset_setthread(td->td_tid, mask);
712 			/*
713 			 * Failing to pin is rarely an actual fatal error;
714 			 * it'll just affect performance.
715 			 */
716 			if (error)
717 				printf("%s: curthread=%llu: can't pin; "
718 				    "error=%d\n",
719 				    __func__,
720 				    (unsigned long long) td->td_tid,
721 				    error);
722 		}
723 		thread_lock(td);
724 		sched_prio(td, pri);
725 		sched_add(td, SRQ_BORING);
726 	}
727 
728 	return (0);
729 }
730 
731 int
732 taskqueue_start_threads(struct taskqueue **tqp, int count, int pri,
733     const char *name, ...)
734 {
735 	va_list ap;
736 	int error;
737 
738 	va_start(ap, name);
739 	error = _taskqueue_start_threads(tqp, count, pri, NULL, NULL, name, ap);
740 	va_end(ap);
741 	return (error);
742 }
743 
744 int
745 taskqueue_start_threads_in_proc(struct taskqueue **tqp, int count, int pri,
746     struct proc *proc, const char *name, ...)
747 {
748 	va_list ap;
749 	int error;
750 
751 	va_start(ap, name);
752 	error = _taskqueue_start_threads(tqp, count, pri, NULL, proc, name, ap);
753 	va_end(ap);
754 	return (error);
755 }
756 
757 int
758 taskqueue_start_threads_cpuset(struct taskqueue **tqp, int count, int pri,
759     cpuset_t *mask, const char *name, ...)
760 {
761 	va_list ap;
762 	int error;
763 
764 	va_start(ap, name);
765 	error = _taskqueue_start_threads(tqp, count, pri, mask, NULL, name, ap);
766 	va_end(ap);
767 	return (error);
768 }
769 
770 static inline void
771 taskqueue_run_callback(struct taskqueue *tq,
772     enum taskqueue_callback_type cb_type)
773 {
774 	taskqueue_callback_fn tq_callback;
775 
776 	TQ_ASSERT_UNLOCKED(tq);
777 	tq_callback = tq->tq_callbacks[cb_type];
778 	if (tq_callback != NULL)
779 		tq_callback(tq->tq_cb_contexts[cb_type]);
780 }
781 
782 void
783 taskqueue_thread_loop(void *arg)
784 {
785 	struct taskqueue **tqp, *tq;
786 
787 	tqp = arg;
788 	tq = *tqp;
789 	taskqueue_run_callback(tq, TASKQUEUE_CALLBACK_TYPE_INIT);
790 	TQ_LOCK(tq);
791 	while ((tq->tq_flags & TQ_FLAGS_ACTIVE) != 0) {
792 		/* XXX ? */
793 		taskqueue_run_locked(tq);
794 		/*
795 		 * Because taskqueue_run() can drop tq_mutex, we need to
796 		 * check if the TQ_FLAGS_ACTIVE flag wasn't removed in the
797 		 * meantime, which means we missed a wakeup.
798 		 */
799 		if ((tq->tq_flags & TQ_FLAGS_ACTIVE) == 0)
800 			break;
801 		TQ_SLEEP(tq, tq, "-");
802 	}
803 	taskqueue_run_locked(tq);
804 	/*
805 	 * This thread is on its way out, so just drop the lock temporarily
806 	 * in order to call the shutdown callback.  This allows the callback
807 	 * to look at the taskqueue, even just before it dies.
808 	 */
809 	TQ_UNLOCK(tq);
810 	taskqueue_run_callback(tq, TASKQUEUE_CALLBACK_TYPE_SHUTDOWN);
811 	TQ_LOCK(tq);
812 
813 	/* rendezvous with thread that asked us to terminate */
814 	tq->tq_tcount--;
815 	wakeup_one(tq->tq_threads);
816 	TQ_UNLOCK(tq);
817 	kthread_exit();
818 }
819 
820 void
821 taskqueue_thread_enqueue(void *context)
822 {
823 	struct taskqueue **tqp, *tq;
824 
825 	tqp = context;
826 	tq = *tqp;
827 	wakeup_any(tq);
828 }
829 
830 TASKQUEUE_DEFINE(swi, taskqueue_swi_enqueue, NULL,
831 		 swi_add(NULL, "task queue", taskqueue_swi_run, NULL, SWI_TQ,
832 		     INTR_MPSAFE, &taskqueue_ih));
833 
834 TASKQUEUE_DEFINE(swi_giant, taskqueue_swi_giant_enqueue, NULL,
835 		 swi_add(NULL, "Giant taskq", taskqueue_swi_giant_run,
836 		     NULL, SWI_TQ_GIANT, 0, &taskqueue_giant_ih));
837 
838 TASKQUEUE_DEFINE_THREAD(thread);
839 
840 struct taskqueue *
841 taskqueue_create_fast(const char *name, int mflags,
842 		 taskqueue_enqueue_fn enqueue, void *context)
843 {
844 	return _taskqueue_create(name, mflags, enqueue, context,
845 			MTX_SPIN, "fast_taskqueue");
846 }
847 
848 static void	*taskqueue_fast_ih;
849 
850 static void
851 taskqueue_fast_enqueue(void *context)
852 {
853 	swi_sched(taskqueue_fast_ih, 0);
854 }
855 
856 static void
857 taskqueue_fast_run(void *dummy)
858 {
859 	taskqueue_run(taskqueue_fast);
860 }
861 
862 TASKQUEUE_FAST_DEFINE(fast, taskqueue_fast_enqueue, NULL,
863 	swi_add(NULL, "fast taskq", taskqueue_fast_run, NULL,
864 	SWI_TQ_FAST, INTR_MPSAFE, &taskqueue_fast_ih));
865 
866 int
867 taskqueue_member(struct taskqueue *queue, struct thread *td)
868 {
869 	int i, j, ret = 0;
870 
871 	for (i = 0, j = 0; ; i++) {
872 		if (queue->tq_threads[i] == NULL)
873 			continue;
874 		if (queue->tq_threads[i] == td) {
875 			ret = 1;
876 			break;
877 		}
878 		if (++j >= queue->tq_tcount)
879 			break;
880 	}
881 	return (ret);
882 }
883