]>
Commit | Line | Data |
---|---|---|
716154c5 BB |
1 | /*****************************************************************************\ |
2 | * Copyright (C) 2007-2010 Lawrence Livermore National Security, LLC. | |
3 | * Copyright (C) 2007 The Regents of the University of California. | |
4 | * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER). | |
5 | * Written by Brian Behlendorf <behlendorf1@llnl.gov>. | |
715f6251 | 6 | * UCRL-CODE-235197 |
7 | * | |
716154c5 | 8 | * This file is part of the SPL, Solaris Porting Layer. |
3d6af2dd | 9 | * For details, see <http://zfsonlinux.org/>. |
715f6251 | 10 | * |
716154c5 BB |
11 | * The SPL is free software; you can redistribute it and/or modify it |
12 | * under the terms of the GNU General Public License as published by the | |
13 | * Free Software Foundation; either version 2 of the License, or (at your | |
14 | * option) any later version. | |
15 | * | |
16 | * The SPL is distributed in the hope that it will be useful, but WITHOUT | |
715f6251 | 17 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
18 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
19 | * for more details. | |
20 | * | |
21 | * You should have received a copy of the GNU General Public License along | |
716154c5 BB |
22 | * with the SPL. If not, see <http://www.gnu.org/licenses/>. |
23 | ***************************************************************************** | |
24 | * Solaris Porting Layer (SPL) Task Queue Implementation. | |
25 | \*****************************************************************************/ | |
715f6251 | 26 | |
f4b37741 | 27 | #include <sys/taskq.h> |
3d061e9d | 28 | #include <sys/kmem.h> |
55abb092 | 29 | #include <spl-debug.h> |
f1ca4da6 | 30 | |
b17edc10 BB |
31 | #ifdef SS_DEBUG_SUBSYS |
32 | #undef SS_DEBUG_SUBSYS | |
937879f1 | 33 | #endif |
34 | ||
b17edc10 | 35 | #define SS_DEBUG_SUBSYS SS_TASKQ |
937879f1 | 36 | |
e9cb2b4f BB |
37 | /* Global system-wide dynamic task queue available for all consumers */ |
38 | taskq_t *system_taskq; | |
39 | EXPORT_SYMBOL(system_taskq); | |
40 | ||
9b51f218 BB |
41 | static int |
42 | task_km_flags(uint_t flags) | |
43 | { | |
44 | if (flags & TQ_NOSLEEP) | |
45 | return KM_NOSLEEP; | |
46 | ||
47 | if (flags & TQ_PUSHPAGE) | |
48 | return KM_PUSHPAGE; | |
49 | ||
50 | return KM_SLEEP; | |
51 | } | |
52 | ||
82387586 BB |
53 | /* |
54 | * NOTE: Must be called with tq->tq_lock held, returns a list_t which | |
bcd68186 | 55 | * is not attached to the free, work, or pending taskq lists. |
f1ca4da6 | 56 | */ |
046a70c9 | 57 | static taskq_ent_t * |
bcd68186 | 58 | task_alloc(taskq_t *tq, uint_t flags) |
59 | { | |
472a34ca BB |
60 | taskq_ent_t *t; |
61 | int count = 0; | |
62 | SENTRY; | |
bcd68186 | 63 | |
472a34ca BB |
64 | ASSERT(tq); |
65 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
bcd68186 | 66 | retry: |
472a34ca BB |
67 | /* Acquire taskq_ent_t's from free list if available */ |
68 | if (!list_empty(&tq->tq_free_list) && !(flags & TQ_NEW)) { | |
69 | t = list_entry(tq->tq_free_list.next, taskq_ent_t, tqent_list); | |
70 | ||
71 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
d9acd930 BB |
72 | ASSERT(!(t->tqent_flags & TQENT_FLAG_CANCEL)); |
73 | ASSERT(!timer_pending(&t->tqent_timer)); | |
472a34ca BB |
74 | |
75 | list_del_init(&t->tqent_list); | |
76 | SRETURN(t); | |
77 | } | |
78 | ||
79 | /* Free list is empty and memory allocations are prohibited */ | |
80 | if (flags & TQ_NOALLOC) | |
81 | SRETURN(NULL); | |
82 | ||
83 | /* Hit maximum taskq_ent_t pool size */ | |
84 | if (tq->tq_nalloc >= tq->tq_maxalloc) { | |
85 | if (flags & TQ_NOSLEEP) | |
86 | SRETURN(NULL); | |
87 | ||
88 | /* | |
89 | * Sleep periodically polling the free list for an available | |
90 | * taskq_ent_t. Dispatching with TQ_SLEEP should always succeed | |
91 | * but we cannot block forever waiting for an taskq_ent_t to | |
92 | * show up in the free list, otherwise a deadlock can happen. | |
93 | * | |
94 | * Therefore, we need to allocate a new task even if the number | |
95 | * of allocated tasks is above tq->tq_maxalloc, but we still | |
96 | * end up delaying the task allocation by one second, thereby | |
97 | * throttling the task dispatch rate. | |
98 | */ | |
99 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
100 | schedule_timeout(HZ / 100); | |
101 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
102 | if (count < 100) | |
103 | SGOTO(retry, count++); | |
104 | } | |
105 | ||
106 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
107 | t = kmem_alloc(sizeof(taskq_ent_t), task_km_flags(flags)); | |
108 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
109 | ||
110 | if (t) { | |
111 | taskq_init_ent(t); | |
112 | tq->tq_nalloc++; | |
113 | } | |
114 | ||
115 | SRETURN(t); | |
bcd68186 | 116 | } |
117 | ||
82387586 | 118 | /* |
046a70c9 | 119 | * NOTE: Must be called with tq->tq_lock held, expects the taskq_ent_t |
bcd68186 | 120 | * to already be removed from the free, work, or pending taskq lists. |
121 | */ | |
122 | static void | |
046a70c9 | 123 | task_free(taskq_t *tq, taskq_ent_t *t) |
bcd68186 | 124 | { |
472a34ca | 125 | SENTRY; |
bcd68186 | 126 | |
472a34ca BB |
127 | ASSERT(tq); |
128 | ASSERT(t); | |
bcd68186 | 129 | ASSERT(spin_is_locked(&tq->tq_lock)); |
046a70c9 | 130 | ASSERT(list_empty(&t->tqent_list)); |
d9acd930 | 131 | ASSERT(!timer_pending(&t->tqent_timer)); |
bcd68186 | 132 | |
472a34ca BB |
133 | kmem_free(t, sizeof(taskq_ent_t)); |
134 | tq->tq_nalloc--; | |
f1ca4da6 | 135 | |
b17edc10 | 136 | SEXIT; |
bcd68186 | 137 | } |
138 | ||
82387586 BB |
139 | /* |
140 | * NOTE: Must be called with tq->tq_lock held, either destroys the | |
046a70c9 | 141 | * taskq_ent_t if too many exist or moves it to the free list for later use. |
bcd68186 | 142 | */ |
f1ca4da6 | 143 | static void |
046a70c9 | 144 | task_done(taskq_t *tq, taskq_ent_t *t) |
f1ca4da6 | 145 | { |
b17edc10 | 146 | SENTRY; |
bcd68186 | 147 | ASSERT(tq); |
148 | ASSERT(t); | |
149 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
150 | ||
d9acd930 BB |
151 | /* Wake tasks blocked in taskq_wait_id() */ |
152 | wake_up_all(&t->tqent_waitq); | |
153 | ||
046a70c9 | 154 | list_del_init(&t->tqent_list); |
f1ca4da6 | 155 | |
472a34ca | 156 | if (tq->tq_nalloc <= tq->tq_minalloc) { |
046a70c9 PS |
157 | t->tqent_id = 0; |
158 | t->tqent_func = NULL; | |
159 | t->tqent_arg = NULL; | |
44217f7a | 160 | t->tqent_flags = 0; |
8f2503e0 | 161 | |
472a34ca | 162 | list_add_tail(&t->tqent_list, &tq->tq_free_list); |
bcd68186 | 163 | } else { |
164 | task_free(tq, t); | |
165 | } | |
f1ca4da6 | 166 | |
472a34ca | 167 | SEXIT; |
f1ca4da6 | 168 | } |
169 | ||
82387586 | 170 | /* |
d9acd930 BB |
171 | * When a delayed task timer expires remove it from the delay list and |
172 | * add it to the priority list in order for immediate processing. | |
bcd68186 | 173 | */ |
d9acd930 BB |
174 | static void |
175 | task_expire(unsigned long data) | |
bcd68186 | 176 | { |
d9acd930 BB |
177 | taskq_ent_t *w, *t = (taskq_ent_t *)data; |
178 | taskq_t *tq = t->tqent_taskq; | |
179 | struct list_head *l; | |
7257ec41 BB |
180 | |
181 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
d9acd930 BB |
182 | |
183 | if (t->tqent_flags & TQENT_FLAG_CANCEL) { | |
184 | ASSERT(list_empty(&t->tqent_list)); | |
185 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
186 | return; | |
187 | } | |
188 | ||
189 | /* | |
190 | * The priority list must be maintained in strict task id order | |
191 | * from lowest to highest for lowest_id to be easily calculable. | |
192 | */ | |
193 | list_del(&t->tqent_list); | |
194 | list_for_each_prev(l, &tq->tq_prio_list) { | |
195 | w = list_entry(l, taskq_ent_t, tqent_list); | |
196 | if (w->tqent_id < t->tqent_id) { | |
197 | list_add(&t->tqent_list, l); | |
198 | break; | |
199 | } | |
200 | } | |
201 | if (l == &tq->tq_prio_list) | |
202 | list_add(&t->tqent_list, &tq->tq_prio_list); | |
203 | ||
7257ec41 BB |
204 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
205 | ||
d9acd930 BB |
206 | wake_up(&tq->tq_work_waitq); |
207 | } | |
208 | ||
209 | /* | |
210 | * Returns the lowest incomplete taskqid_t. The taskqid_t may | |
211 | * be queued on the pending list, on the priority list, on the | |
212 | * delay list, or on the work list currently being handled, but | |
213 | * it is not 100% complete yet. | |
214 | */ | |
215 | static taskqid_t | |
216 | taskq_lowest_id(taskq_t *tq) | |
217 | { | |
218 | taskqid_t lowest_id = tq->tq_next_id; | |
219 | taskq_ent_t *t; | |
220 | taskq_thread_t *tqt; | |
221 | SENTRY; | |
222 | ||
223 | ASSERT(tq); | |
224 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
225 | ||
226 | if (!list_empty(&tq->tq_pend_list)) { | |
227 | t = list_entry(tq->tq_pend_list.next, taskq_ent_t, tqent_list); | |
228 | lowest_id = MIN(lowest_id, t->tqent_id); | |
229 | } | |
230 | ||
231 | if (!list_empty(&tq->tq_prio_list)) { | |
232 | t = list_entry(tq->tq_prio_list.next, taskq_ent_t, tqent_list); | |
233 | lowest_id = MIN(lowest_id, t->tqent_id); | |
234 | } | |
235 | ||
236 | if (!list_empty(&tq->tq_delay_list)) { | |
237 | t = list_entry(tq->tq_delay_list.next, taskq_ent_t, tqent_list); | |
238 | lowest_id = MIN(lowest_id, t->tqent_id); | |
239 | } | |
240 | ||
241 | if (!list_empty(&tq->tq_active_list)) { | |
242 | tqt = list_entry(tq->tq_active_list.next, taskq_thread_t, | |
243 | tqt_active_list); | |
244 | ASSERT(tqt->tqt_id != 0); | |
245 | lowest_id = MIN(lowest_id, tqt->tqt_id); | |
246 | } | |
247 | ||
248 | SRETURN(lowest_id); | |
249 | } | |
250 | ||
251 | /* | |
252 | * Insert a task into a list keeping the list sorted by increasing taskqid. | |
253 | */ | |
254 | static void | |
255 | taskq_insert_in_order(taskq_t *tq, taskq_thread_t *tqt) | |
256 | { | |
257 | taskq_thread_t *w; | |
258 | struct list_head *l; | |
259 | ||
260 | SENTRY; | |
261 | ASSERT(tq); | |
262 | ASSERT(tqt); | |
263 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
264 | ||
265 | list_for_each_prev(l, &tq->tq_active_list) { | |
266 | w = list_entry(l, taskq_thread_t, tqt_active_list); | |
267 | if (w->tqt_id < tqt->tqt_id) { | |
268 | list_add(&tqt->tqt_active_list, l); | |
269 | break; | |
270 | } | |
271 | } | |
272 | if (l == &tq->tq_active_list) | |
273 | list_add(&tqt->tqt_active_list, &tq->tq_active_list); | |
274 | ||
275 | SEXIT; | |
276 | } | |
277 | ||
278 | /* | |
279 | * Find and return a task from the given list if it exists. The list | |
280 | * must be in lowest to highest task id order. | |
281 | */ | |
282 | static taskq_ent_t * | |
283 | taskq_find_list(taskq_t *tq, struct list_head *lh, taskqid_t id) | |
284 | { | |
285 | struct list_head *l; | |
286 | taskq_ent_t *t; | |
287 | SENTRY; | |
288 | ||
289 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
290 | ||
291 | list_for_each(l, lh) { | |
292 | t = list_entry(l, taskq_ent_t, tqent_list); | |
293 | ||
294 | if (t->tqent_id == id) | |
295 | SRETURN(t); | |
296 | ||
297 | if (t->tqent_id > id) | |
298 | break; | |
299 | } | |
300 | ||
301 | SRETURN(NULL); | |
bcd68186 | 302 | } |
303 | ||
d9acd930 BB |
304 | /* |
305 | * Find an already dispatched task given the task id regardless of what | |
306 | * state it is in. If a task is still pending or executing it will be | |
307 | * returned and 'active' set appropriately. If the task has already | |
308 | * been run then NULL is returned. | |
309 | */ | |
310 | static taskq_ent_t * | |
311 | taskq_find(taskq_t *tq, taskqid_t id, int *active) | |
312 | { | |
313 | taskq_thread_t *tqt; | |
314 | struct list_head *l; | |
315 | taskq_ent_t *t; | |
316 | SENTRY; | |
317 | ||
318 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
319 | *active = 0; | |
320 | ||
321 | t = taskq_find_list(tq, &tq->tq_delay_list, id); | |
322 | if (t) | |
323 | SRETURN(t); | |
324 | ||
325 | t = taskq_find_list(tq, &tq->tq_prio_list, id); | |
326 | if (t) | |
327 | SRETURN(t); | |
328 | ||
329 | t = taskq_find_list(tq, &tq->tq_pend_list, id); | |
330 | if (t) | |
331 | SRETURN(t); | |
332 | ||
333 | list_for_each(l, &tq->tq_active_list) { | |
334 | tqt = list_entry(l, taskq_thread_t, tqt_active_list); | |
335 | if (tqt->tqt_id == id) { | |
336 | t = tqt->tqt_task; | |
337 | *active = 1; | |
338 | SRETURN(t); | |
339 | } | |
340 | } | |
341 | ||
342 | SRETURN(NULL); | |
343 | } | |
344 | ||
345 | /* | |
346 | * The taskq_wait_id() function blocks until the passed task id completes. | |
347 | * This does not guarantee that all lower task id's have completed. | |
348 | */ | |
bcd68186 | 349 | void |
aed8671c | 350 | taskq_wait_id(taskq_t *tq, taskqid_t id) |
f1ca4da6 | 351 | { |
d9acd930 BB |
352 | DEFINE_WAIT(wait); |
353 | taskq_ent_t *t; | |
354 | int active = 0; | |
b17edc10 | 355 | SENTRY; |
d9acd930 | 356 | |
bcd68186 | 357 | ASSERT(tq); |
d9acd930 | 358 | ASSERT(id > 0); |
bcd68186 | 359 | |
d9acd930 BB |
360 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
361 | t = taskq_find(tq, id, &active); | |
362 | if (t) | |
363 | prepare_to_wait(&t->tqent_waitq, &wait, TASK_UNINTERRUPTIBLE); | |
364 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
365 | ||
366 | /* | |
367 | * We rely on the kernels autoremove_wake_function() function to | |
368 | * remove us from the wait queue in the context of wake_up(). | |
369 | * Once woken the taskq_ent_t pointer must never be accessed. | |
370 | */ | |
371 | if (t) { | |
372 | t = NULL; | |
373 | schedule(); | |
374 | __set_current_state(TASK_RUNNING); | |
375 | } | |
bcd68186 | 376 | |
b17edc10 | 377 | SEXIT; |
bcd68186 | 378 | } |
aed8671c | 379 | EXPORT_SYMBOL(taskq_wait_id); |
bcd68186 | 380 | |
d9acd930 BB |
381 | /* |
382 | * The taskq_wait() function will block until all previously submitted | |
383 | * tasks have been completed. A previously submitted task is defined as | |
384 | * a task with a lower task id than the current task queue id. Note that | |
385 | * all task id's are assigned monotonically at dispatch time. | |
386 | * | |
387 | * Waiting for all previous tasks to complete is accomplished by tracking | |
388 | * the lowest outstanding task id. As tasks are dispatched they are added | |
389 | * added to the tail of the pending, priority, or delay lists. And as | |
390 | * worker threads become available the tasks are removed from the heads | |
391 | * of these lists and linked to the worker threads. This ensures the | |
392 | * lists are kept in lowest to highest task id order. | |
393 | * | |
394 | * Therefore the lowest outstanding task id can be quickly determined by | |
395 | * checking the head item from all of these lists. This value is stored | |
396 | * with the task queue as the lowest id. It only needs to be recalculated | |
397 | * when either the task with the current lowest id completes or is canceled. | |
398 | * | |
399 | * By blocking until the lowest task id exceeds the current task id when | |
400 | * the function was called we ensure all previous tasks have completed. | |
401 | * | |
402 | * NOTE: When there are multiple worked threads it is possible for larger | |
403 | * task ids to complete before smaller ones. Conversely when the task | |
404 | * queue contains delay tasks with small task ids, you may block for a | |
405 | * considerable length of time waiting for them to expire and execute. | |
406 | */ | |
407 | static int | |
408 | taskq_wait_check(taskq_t *tq, taskqid_t id) | |
409 | { | |
410 | int rc; | |
411 | ||
412 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
413 | rc = (id < tq->tq_lowest_id); | |
414 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
415 | ||
416 | SRETURN(rc); | |
417 | } | |
418 | ||
419 | void | |
420 | taskq_wait_all(taskq_t *tq, taskqid_t id) | |
421 | { | |
422 | wait_event(tq->tq_wait_waitq, taskq_wait_check(tq, id)); | |
423 | } | |
424 | EXPORT_SYMBOL(taskq_wait_all); | |
425 | ||
bcd68186 | 426 | void |
aed8671c | 427 | taskq_wait(taskq_t *tq) |
bcd68186 | 428 | { |
429 | taskqid_t id; | |
b17edc10 | 430 | SENTRY; |
bcd68186 | 431 | ASSERT(tq); |
432 | ||
7257ec41 | 433 | /* Wait for the largest outstanding taskqid */ |
749045bb | 434 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
7257ec41 | 435 | id = tq->tq_next_id - 1; |
749045bb | 436 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 437 | |
d9acd930 | 438 | taskq_wait_all(tq, id); |
bcd68186 | 439 | |
b17edc10 | 440 | SEXIT; |
bcd68186 | 441 | |
442 | } | |
aed8671c | 443 | EXPORT_SYMBOL(taskq_wait); |
bcd68186 | 444 | |
445 | int | |
aed8671c | 446 | taskq_member(taskq_t *tq, void *t) |
bcd68186 | 447 | { |
2c02b71b PS |
448 | struct list_head *l; |
449 | taskq_thread_t *tqt; | |
472a34ca | 450 | SENTRY; |
bcd68186 | 451 | |
452 | ASSERT(tq); | |
472a34ca | 453 | ASSERT(t); |
bcd68186 | 454 | |
2c02b71b PS |
455 | list_for_each(l, &tq->tq_thread_list) { |
456 | tqt = list_entry(l, taskq_thread_t, tqt_thread_list); | |
457 | if (tqt->tqt_thread == (struct task_struct *)t) | |
458 | SRETURN(1); | |
459 | } | |
bcd68186 | 460 | |
472a34ca | 461 | SRETURN(0); |
bcd68186 | 462 | } |
aed8671c | 463 | EXPORT_SYMBOL(taskq_member); |
bcd68186 | 464 | |
d9acd930 BB |
465 | /* |
466 | * Cancel an already dispatched task given the task id. Still pending tasks | |
467 | * will be immediately canceled, and if the task is active the function will | |
468 | * block until it completes. Preallocated tasks which are canceled must be | |
469 | * freed by the caller. | |
470 | */ | |
471 | int | |
472 | taskq_cancel_id(taskq_t *tq, taskqid_t id) | |
473 | { | |
474 | taskq_ent_t *t; | |
475 | int active = 0; | |
476 | int rc = ENOENT; | |
477 | SENTRY; | |
478 | ||
479 | ASSERT(tq); | |
480 | ||
481 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
482 | t = taskq_find(tq, id, &active); | |
483 | if (t && !active) { | |
484 | list_del_init(&t->tqent_list); | |
485 | t->tqent_flags |= TQENT_FLAG_CANCEL; | |
486 | ||
487 | /* | |
488 | * When canceling the lowest outstanding task id we | |
489 | * must recalculate the new lowest outstanding id. | |
490 | */ | |
491 | if (tq->tq_lowest_id == t->tqent_id) { | |
492 | tq->tq_lowest_id = taskq_lowest_id(tq); | |
493 | ASSERT3S(tq->tq_lowest_id, >, t->tqent_id); | |
494 | } | |
495 | ||
496 | /* | |
497 | * The task_expire() function takes the tq->tq_lock so drop | |
498 | * drop the lock before synchronously cancelling the timer. | |
499 | */ | |
500 | if (timer_pending(&t->tqent_timer)) { | |
501 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
502 | del_timer_sync(&t->tqent_timer); | |
503 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
504 | } | |
505 | ||
506 | if (!(t->tqent_flags & TQENT_FLAG_PREALLOC)) | |
507 | task_done(tq, t); | |
508 | ||
509 | rc = 0; | |
510 | } | |
511 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
512 | ||
513 | if (active) { | |
514 | taskq_wait_id(tq, id); | |
515 | rc = EBUSY; | |
516 | } | |
517 | ||
518 | SRETURN(rc); | |
519 | } | |
520 | EXPORT_SYMBOL(taskq_cancel_id); | |
521 | ||
bcd68186 | 522 | taskqid_t |
aed8671c | 523 | taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t flags) |
bcd68186 | 524 | { |
472a34ca | 525 | taskq_ent_t *t; |
bcd68186 | 526 | taskqid_t rc = 0; |
472a34ca | 527 | SENTRY; |
f1ca4da6 | 528 | |
472a34ca BB |
529 | ASSERT(tq); |
530 | ASSERT(func); | |
d05ec4b4 | 531 | |
472a34ca | 532 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
f1ca4da6 | 533 | |
bcd68186 | 534 | /* Taskq being destroyed and all tasks drained */ |
535 | if (!(tq->tq_flags & TQ_ACTIVE)) | |
b17edc10 | 536 | SGOTO(out, rc = 0); |
f1ca4da6 | 537 | |
bcd68186 | 538 | /* Do not queue the task unless there is idle thread for it */ |
539 | ASSERT(tq->tq_nactive <= tq->tq_nthreads); | |
540 | if ((flags & TQ_NOQUEUE) && (tq->tq_nactive == tq->tq_nthreads)) | |
b17edc10 | 541 | SGOTO(out, rc = 0); |
bcd68186 | 542 | |
472a34ca | 543 | if ((t = task_alloc(tq, flags)) == NULL) |
b17edc10 | 544 | SGOTO(out, rc = 0); |
f1ca4da6 | 545 | |
046a70c9 | 546 | spin_lock(&t->tqent_lock); |
f0d8bb26 NB |
547 | |
548 | /* Queue to the priority list instead of the pending list */ | |
549 | if (flags & TQ_FRONT) | |
046a70c9 | 550 | list_add_tail(&t->tqent_list, &tq->tq_prio_list); |
f0d8bb26 | 551 | else |
046a70c9 | 552 | list_add_tail(&t->tqent_list, &tq->tq_pend_list); |
f0d8bb26 | 553 | |
046a70c9 | 554 | t->tqent_id = rc = tq->tq_next_id; |
bcd68186 | 555 | tq->tq_next_id++; |
472a34ca BB |
556 | t->tqent_func = func; |
557 | t->tqent_arg = arg; | |
d9acd930 BB |
558 | t->tqent_taskq = tq; |
559 | t->tqent_timer.data = 0; | |
560 | t->tqent_timer.function = NULL; | |
561 | t->tqent_timer.expires = 0; | |
44217f7a PS |
562 | |
563 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
564 | ||
046a70c9 | 565 | spin_unlock(&t->tqent_lock); |
0bb43ca2 NB |
566 | |
567 | wake_up(&tq->tq_work_waitq); | |
bcd68186 | 568 | out: |
749045bb | 569 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
b17edc10 | 570 | SRETURN(rc); |
f1ca4da6 | 571 | } |
aed8671c | 572 | EXPORT_SYMBOL(taskq_dispatch); |
44217f7a | 573 | |
d9acd930 BB |
574 | taskqid_t |
575 | taskq_dispatch_delay(taskq_t *tq, task_func_t func, void *arg, | |
576 | uint_t flags, clock_t expire_time) | |
577 | { | |
578 | taskq_ent_t *t; | |
579 | taskqid_t rc = 0; | |
580 | SENTRY; | |
581 | ||
582 | ASSERT(tq); | |
583 | ASSERT(func); | |
584 | ||
585 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
586 | ||
587 | /* Taskq being destroyed and all tasks drained */ | |
588 | if (!(tq->tq_flags & TQ_ACTIVE)) | |
589 | SGOTO(out, rc = 0); | |
590 | ||
591 | if ((t = task_alloc(tq, flags)) == NULL) | |
592 | SGOTO(out, rc = 0); | |
593 | ||
594 | spin_lock(&t->tqent_lock); | |
595 | ||
596 | /* Queue to the delay list for subsequent execution */ | |
597 | list_add_tail(&t->tqent_list, &tq->tq_delay_list); | |
598 | ||
599 | t->tqent_id = rc = tq->tq_next_id; | |
600 | tq->tq_next_id++; | |
601 | t->tqent_func = func; | |
602 | t->tqent_arg = arg; | |
603 | t->tqent_taskq = tq; | |
604 | t->tqent_timer.data = (unsigned long)t; | |
605 | t->tqent_timer.function = task_expire; | |
606 | t->tqent_timer.expires = (unsigned long)expire_time; | |
607 | add_timer(&t->tqent_timer); | |
608 | ||
609 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
610 | ||
611 | spin_unlock(&t->tqent_lock); | |
612 | out: | |
613 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
614 | SRETURN(rc); | |
615 | } | |
616 | EXPORT_SYMBOL(taskq_dispatch_delay); | |
617 | ||
44217f7a | 618 | void |
aed8671c | 619 | taskq_dispatch_ent(taskq_t *tq, task_func_t func, void *arg, uint_t flags, |
44217f7a PS |
620 | taskq_ent_t *t) |
621 | { | |
622 | SENTRY; | |
623 | ||
624 | ASSERT(tq); | |
625 | ASSERT(func); | |
626 | ASSERT(!(tq->tq_flags & TASKQ_DYNAMIC)); | |
627 | ||
628 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
629 | ||
630 | /* Taskq being destroyed and all tasks drained */ | |
631 | if (!(tq->tq_flags & TQ_ACTIVE)) { | |
632 | t->tqent_id = 0; | |
633 | goto out; | |
634 | } | |
635 | ||
636 | spin_lock(&t->tqent_lock); | |
637 | ||
638 | /* | |
639 | * Mark it as a prealloc'd task. This is important | |
640 | * to ensure that we don't free it later. | |
641 | */ | |
642 | t->tqent_flags |= TQENT_FLAG_PREALLOC; | |
643 | ||
644 | /* Queue to the priority list instead of the pending list */ | |
645 | if (flags & TQ_FRONT) | |
646 | list_add_tail(&t->tqent_list, &tq->tq_prio_list); | |
647 | else | |
648 | list_add_tail(&t->tqent_list, &tq->tq_pend_list); | |
649 | ||
650 | t->tqent_id = tq->tq_next_id; | |
651 | tq->tq_next_id++; | |
652 | t->tqent_func = func; | |
653 | t->tqent_arg = arg; | |
d9acd930 | 654 | t->tqent_taskq = tq; |
44217f7a PS |
655 | |
656 | spin_unlock(&t->tqent_lock); | |
657 | ||
658 | wake_up(&tq->tq_work_waitq); | |
659 | out: | |
0bb43ca2 | 660 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
44217f7a PS |
661 | SEXIT; |
662 | } | |
aed8671c | 663 | EXPORT_SYMBOL(taskq_dispatch_ent); |
44217f7a PS |
664 | |
665 | int | |
aed8671c | 666 | taskq_empty_ent(taskq_ent_t *t) |
44217f7a PS |
667 | { |
668 | return list_empty(&t->tqent_list); | |
669 | } | |
aed8671c | 670 | EXPORT_SYMBOL(taskq_empty_ent); |
44217f7a PS |
671 | |
672 | void | |
aed8671c | 673 | taskq_init_ent(taskq_ent_t *t) |
44217f7a PS |
674 | { |
675 | spin_lock_init(&t->tqent_lock); | |
d9acd930 BB |
676 | init_waitqueue_head(&t->tqent_waitq); |
677 | init_timer(&t->tqent_timer); | |
44217f7a PS |
678 | INIT_LIST_HEAD(&t->tqent_list); |
679 | t->tqent_id = 0; | |
680 | t->tqent_func = NULL; | |
681 | t->tqent_arg = NULL; | |
682 | t->tqent_flags = 0; | |
d9acd930 | 683 | t->tqent_taskq = NULL; |
44217f7a | 684 | } |
aed8671c | 685 | EXPORT_SYMBOL(taskq_init_ent); |
44217f7a | 686 | |
bcd68186 | 687 | static int |
688 | taskq_thread(void *args) | |
689 | { | |
472a34ca BB |
690 | DECLARE_WAITQUEUE(wait, current); |
691 | sigset_t blocked; | |
2c02b71b | 692 | taskq_thread_t *tqt = args; |
472a34ca BB |
693 | taskq_t *tq; |
694 | taskq_ent_t *t; | |
f0d8bb26 | 695 | struct list_head *pend_list; |
b17edc10 | 696 | SENTRY; |
bcd68186 | 697 | |
472a34ca | 698 | ASSERT(tqt); |
2c02b71b | 699 | tq = tqt->tqt_tq; |
472a34ca | 700 | current->flags |= PF_NOFREEZE; |
bcd68186 | 701 | |
472a34ca BB |
702 | sigfillset(&blocked); |
703 | sigprocmask(SIG_BLOCK, &blocked, NULL); | |
704 | flush_signals(current); | |
bcd68186 | 705 | |
472a34ca BB |
706 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
707 | tq->tq_nthreads++; | |
708 | wake_up(&tq->tq_wait_waitq); | |
709 | set_current_state(TASK_INTERRUPTIBLE); | |
bcd68186 | 710 | |
472a34ca | 711 | while (!kthread_should_stop()) { |
bcd68186 | 712 | |
f0d8bb26 NB |
713 | if (list_empty(&tq->tq_pend_list) && |
714 | list_empty(&tq->tq_prio_list)) { | |
3c6ed541 | 715 | add_wait_queue_exclusive(&tq->tq_work_waitq, &wait); |
749045bb | 716 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 717 | schedule(); |
749045bb | 718 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
3c6ed541 | 719 | remove_wait_queue(&tq->tq_work_waitq, &wait); |
bcd68186 | 720 | } else { |
721 | __set_current_state(TASK_RUNNING); | |
722 | } | |
723 | ||
f0d8bb26 NB |
724 | |
725 | if (!list_empty(&tq->tq_prio_list)) | |
726 | pend_list = &tq->tq_prio_list; | |
727 | else if (!list_empty(&tq->tq_pend_list)) | |
728 | pend_list = &tq->tq_pend_list; | |
729 | else | |
730 | pend_list = NULL; | |
731 | ||
732 | if (pend_list) { | |
472a34ca BB |
733 | t = list_entry(pend_list->next,taskq_ent_t,tqent_list); |
734 | list_del_init(&t->tqent_list); | |
8f2503e0 | 735 | |
44217f7a PS |
736 | /* In order to support recursively dispatching a |
737 | * preallocated taskq_ent_t, tqent_id must be | |
738 | * stored prior to executing tqent_func. */ | |
e7e5f78e | 739 | tqt->tqt_id = t->tqent_id; |
d9acd930 | 740 | tqt->tqt_task = t; |
8f2503e0 PS |
741 | |
742 | /* We must store a copy of the flags prior to | |
743 | * servicing the task (servicing a prealloc'd task | |
744 | * returns the ownership of the tqent back to | |
745 | * the caller of taskq_dispatch). Thus, | |
746 | * tqent_flags _may_ change within the call. */ | |
747 | tqt->tqt_flags = t->tqent_flags; | |
748 | ||
2c02b71b | 749 | taskq_insert_in_order(tq, tqt); |
472a34ca | 750 | tq->tq_nactive++; |
749045bb | 751 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 752 | |
753 | /* Perform the requested task */ | |
472a34ca | 754 | t->tqent_func(t->tqent_arg); |
bcd68186 | 755 | |
749045bb | 756 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
472a34ca | 757 | tq->tq_nactive--; |
2c02b71b | 758 | list_del_init(&tqt->tqt_active_list); |
d9acd930 | 759 | tqt->tqt_task = NULL; |
8f2503e0 PS |
760 | |
761 | /* For prealloc'd tasks, we don't free anything. */ | |
762 | if ((tq->tq_flags & TASKQ_DYNAMIC) || | |
763 | !(tqt->tqt_flags & TQENT_FLAG_PREALLOC)) | |
764 | task_done(tq, t); | |
bcd68186 | 765 | |
7257ec41 BB |
766 | /* When the current lowest outstanding taskqid is |
767 | * done calculate the new lowest outstanding id */ | |
e7e5f78e | 768 | if (tq->tq_lowest_id == tqt->tqt_id) { |
bcd68186 | 769 | tq->tq_lowest_id = taskq_lowest_id(tq); |
e7e5f78e | 770 | ASSERT3S(tq->tq_lowest_id, >, tqt->tqt_id); |
bcd68186 | 771 | } |
772 | ||
e7e5f78e | 773 | tqt->tqt_id = 0; |
8f2503e0 | 774 | tqt->tqt_flags = 0; |
472a34ca | 775 | wake_up_all(&tq->tq_wait_waitq); |
bcd68186 | 776 | } |
777 | ||
778 | set_current_state(TASK_INTERRUPTIBLE); | |
779 | ||
472a34ca | 780 | } |
bcd68186 | 781 | |
782 | __set_current_state(TASK_RUNNING); | |
472a34ca | 783 | tq->tq_nthreads--; |
2c02b71b PS |
784 | list_del_init(&tqt->tqt_thread_list); |
785 | kmem_free(tqt, sizeof(taskq_thread_t)); | |
786 | ||
472a34ca | 787 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 788 | |
b17edc10 | 789 | SRETURN(0); |
bcd68186 | 790 | } |
791 | ||
f1ca4da6 | 792 | taskq_t * |
aed8671c | 793 | taskq_create(const char *name, int nthreads, pri_t pri, |
472a34ca | 794 | int minalloc, int maxalloc, uint_t flags) |
f1ca4da6 | 795 | { |
472a34ca | 796 | taskq_t *tq; |
2c02b71b | 797 | taskq_thread_t *tqt; |
472a34ca BB |
798 | int rc = 0, i, j = 0; |
799 | SENTRY; | |
bcd68186 | 800 | |
472a34ca BB |
801 | ASSERT(name != NULL); |
802 | ASSERT(pri <= maxclsyspri); | |
803 | ASSERT(minalloc >= 0); | |
804 | ASSERT(maxalloc <= INT_MAX); | |
805 | ASSERT(!(flags & (TASKQ_CPR_SAFE | TASKQ_DYNAMIC))); /* Unsupported */ | |
bcd68186 | 806 | |
915404bd BB |
807 | /* Scale the number of threads using nthreads as a percentage */ |
808 | if (flags & TASKQ_THREADS_CPU_PCT) { | |
809 | ASSERT(nthreads <= 100); | |
810 | ASSERT(nthreads >= 0); | |
811 | nthreads = MIN(nthreads, 100); | |
812 | nthreads = MAX(nthreads, 0); | |
813 | nthreads = MAX((num_online_cpus() * nthreads) / 100, 1); | |
814 | } | |
815 | ||
472a34ca BB |
816 | tq = kmem_alloc(sizeof(*tq), KM_PUSHPAGE); |
817 | if (tq == NULL) | |
818 | SRETURN(NULL); | |
bcd68186 | 819 | |
472a34ca BB |
820 | spin_lock_init(&tq->tq_lock); |
821 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
822 | INIT_LIST_HEAD(&tq->tq_thread_list); | |
823 | INIT_LIST_HEAD(&tq->tq_active_list); | |
824 | tq->tq_name = name; | |
825 | tq->tq_nactive = 0; | |
bcd68186 | 826 | tq->tq_nthreads = 0; |
472a34ca BB |
827 | tq->tq_pri = pri; |
828 | tq->tq_minalloc = minalloc; | |
829 | tq->tq_maxalloc = maxalloc; | |
bcd68186 | 830 | tq->tq_nalloc = 0; |
472a34ca | 831 | tq->tq_flags = (flags | TQ_ACTIVE); |
bcd68186 | 832 | tq->tq_next_id = 1; |
833 | tq->tq_lowest_id = 1; | |
472a34ca BB |
834 | INIT_LIST_HEAD(&tq->tq_free_list); |
835 | INIT_LIST_HEAD(&tq->tq_pend_list); | |
836 | INIT_LIST_HEAD(&tq->tq_prio_list); | |
d9acd930 | 837 | INIT_LIST_HEAD(&tq->tq_delay_list); |
472a34ca BB |
838 | init_waitqueue_head(&tq->tq_work_waitq); |
839 | init_waitqueue_head(&tq->tq_wait_waitq); | |
bcd68186 | 840 | |
472a34ca BB |
841 | if (flags & TASKQ_PREPOPULATE) |
842 | for (i = 0; i < minalloc; i++) | |
843 | task_done(tq, task_alloc(tq, TQ_PUSHPAGE | TQ_NEW)); | |
6e605b6e | 844 | |
472a34ca | 845 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
6e605b6e | 846 | |
2c02b71b | 847 | for (i = 0; i < nthreads; i++) { |
9b51f218 | 848 | tqt = kmem_alloc(sizeof(*tqt), KM_PUSHPAGE); |
2c02b71b PS |
849 | INIT_LIST_HEAD(&tqt->tqt_thread_list); |
850 | INIT_LIST_HEAD(&tqt->tqt_active_list); | |
851 | tqt->tqt_tq = tq; | |
e7e5f78e | 852 | tqt->tqt_id = 0; |
2c02b71b PS |
853 | |
854 | tqt->tqt_thread = kthread_create(taskq_thread, tqt, | |
472a34ca | 855 | "%s/%d", name, i); |
2c02b71b PS |
856 | if (tqt->tqt_thread) { |
857 | list_add(&tqt->tqt_thread_list, &tq->tq_thread_list); | |
858 | kthread_bind(tqt->tqt_thread, i % num_online_cpus()); | |
859 | set_user_nice(tqt->tqt_thread, PRIO_TO_NICE(pri)); | |
860 | wake_up_process(tqt->tqt_thread); | |
bcd68186 | 861 | j++; |
2c02b71b PS |
862 | } else { |
863 | kmem_free(tqt, sizeof(taskq_thread_t)); | |
864 | rc = 1; | |
865 | } | |
866 | } | |
bcd68186 | 867 | |
472a34ca | 868 | /* Wait for all threads to be started before potential destroy */ |
bcd68186 | 869 | wait_event(tq->tq_wait_waitq, tq->tq_nthreads == j); |
870 | ||
472a34ca | 871 | if (rc) { |
aed8671c | 872 | taskq_destroy(tq); |
472a34ca BB |
873 | tq = NULL; |
874 | } | |
bcd68186 | 875 | |
472a34ca | 876 | SRETURN(tq); |
f1ca4da6 | 877 | } |
aed8671c | 878 | EXPORT_SYMBOL(taskq_create); |
b123971f | 879 | |
880 | void | |
aed8671c | 881 | taskq_destroy(taskq_t *tq) |
b123971f | 882 | { |
2c02b71b PS |
883 | struct task_struct *thread; |
884 | taskq_thread_t *tqt; | |
046a70c9 | 885 | taskq_ent_t *t; |
b17edc10 | 886 | SENTRY; |
b123971f | 887 | |
bcd68186 | 888 | ASSERT(tq); |
749045bb | 889 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
472a34ca | 890 | tq->tq_flags &= ~TQ_ACTIVE; |
749045bb | 891 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 892 | |
893 | /* TQ_ACTIVE cleared prevents new tasks being added to pending */ | |
aed8671c | 894 | taskq_wait(tq); |
bcd68186 | 895 | |
472a34ca | 896 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 897 | |
2c02b71b PS |
898 | /* |
899 | * Signal each thread to exit and block until it does. Each thread | |
900 | * is responsible for removing itself from the list and freeing its | |
901 | * taskq_thread_t. This allows for idle threads to opt to remove | |
902 | * themselves from the taskq. They can be recreated as needed. | |
903 | */ | |
904 | while (!list_empty(&tq->tq_thread_list)) { | |
905 | tqt = list_entry(tq->tq_thread_list.next, | |
906 | taskq_thread_t, tqt_thread_list); | |
907 | thread = tqt->tqt_thread; | |
908 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
909 | ||
910 | kthread_stop(thread); | |
911 | ||
472a34ca | 912 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
2c02b71b PS |
913 | } |
914 | ||
472a34ca | 915 | while (!list_empty(&tq->tq_free_list)) { |
046a70c9 | 916 | t = list_entry(tq->tq_free_list.next, taskq_ent_t, tqent_list); |
44217f7a PS |
917 | |
918 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
919 | ||
472a34ca BB |
920 | list_del_init(&t->tqent_list); |
921 | task_free(tq, t); | |
922 | } | |
bcd68186 | 923 | |
472a34ca BB |
924 | ASSERT(tq->tq_nthreads == 0); |
925 | ASSERT(tq->tq_nalloc == 0); | |
926 | ASSERT(list_empty(&tq->tq_thread_list)); | |
927 | ASSERT(list_empty(&tq->tq_active_list)); | |
928 | ASSERT(list_empty(&tq->tq_free_list)); | |
929 | ASSERT(list_empty(&tq->tq_pend_list)); | |
930 | ASSERT(list_empty(&tq->tq_prio_list)); | |
d9acd930 | 931 | ASSERT(list_empty(&tq->tq_delay_list)); |
bcd68186 | 932 | |
472a34ca | 933 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
2c02b71b | 934 | |
472a34ca | 935 | kmem_free(tq, sizeof(taskq_t)); |
bcd68186 | 936 | |
b17edc10 | 937 | SEXIT; |
b123971f | 938 | } |
aed8671c | 939 | EXPORT_SYMBOL(taskq_destroy); |
e9cb2b4f BB |
940 | |
941 | int | |
942 | spl_taskq_init(void) | |
943 | { | |
472a34ca | 944 | SENTRY; |
e9cb2b4f | 945 | |
f220894e BB |
946 | /* Solaris creates a dynamic taskq of up to 64 threads, however in |
947 | * a Linux environment 1 thread per-core is usually about right */ | |
472a34ca | 948 | system_taskq = taskq_create("spl_system_taskq", num_online_cpus(), |
f220894e | 949 | minclsyspri, 4, 512, TASKQ_PREPOPULATE); |
e9cb2b4f | 950 | if (system_taskq == NULL) |
b17edc10 | 951 | SRETURN(1); |
e9cb2b4f | 952 | |
472a34ca | 953 | SRETURN(0); |
e9cb2b4f BB |
954 | } |
955 | ||
956 | void | |
957 | spl_taskq_fini(void) | |
958 | { | |
472a34ca | 959 | SENTRY; |
e9cb2b4f | 960 | taskq_destroy(system_taskq); |
472a34ca | 961 | SEXIT; |
e9cb2b4f | 962 | } |