]>
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> |
937879f1 | 29 | |
703371d8 AV |
30 | int spl_taskq_thread_bind = 0; |
31 | module_param(spl_taskq_thread_bind, int, 0644); | |
32 | MODULE_PARM_DESC(spl_taskq_thread_bind, "Bind taskq thread to CPU by default"); | |
33 | ||
f7a973d9 BB |
34 | |
35 | int spl_taskq_thread_dynamic = 1; | |
36 | module_param(spl_taskq_thread_dynamic, int, 0644); | |
37 | MODULE_PARM_DESC(spl_taskq_thread_dynamic, "Allow dynamic taskq threads"); | |
38 | ||
62aa81a5 BB |
39 | int spl_taskq_thread_priority = 1; |
40 | module_param(spl_taskq_thread_priority, int, 0644); | |
41 | MODULE_PARM_DESC(spl_taskq_thread_priority, | |
42 | "Allow non-default priority for taskq threads"); | |
43 | ||
f7a973d9 BB |
44 | int spl_taskq_thread_sequential = 4; |
45 | module_param(spl_taskq_thread_sequential, int, 0644); | |
46 | MODULE_PARM_DESC(spl_taskq_thread_sequential, | |
47 | "Create new taskq threads after N sequential tasks"); | |
48 | ||
e9cb2b4f BB |
49 | /* Global system-wide dynamic task queue available for all consumers */ |
50 | taskq_t *system_taskq; | |
51 | EXPORT_SYMBOL(system_taskq); | |
52 | ||
f7a973d9 BB |
53 | /* Private dedicated taskq for creating new taskq threads on demand. */ |
54 | static taskq_t *dynamic_taskq; | |
55 | static taskq_thread_t *taskq_thread_create(taskq_t *); | |
56 | ||
9b51f218 BB |
57 | static int |
58 | task_km_flags(uint_t flags) | |
59 | { | |
60 | if (flags & TQ_NOSLEEP) | |
61 | return KM_NOSLEEP; | |
62 | ||
63 | if (flags & TQ_PUSHPAGE) | |
64 | return KM_PUSHPAGE; | |
65 | ||
66 | return KM_SLEEP; | |
67 | } | |
68 | ||
82387586 BB |
69 | /* |
70 | * NOTE: Must be called with tq->tq_lock held, returns a list_t which | |
bcd68186 | 71 | * is not attached to the free, work, or pending taskq lists. |
f1ca4da6 | 72 | */ |
046a70c9 | 73 | static taskq_ent_t * |
bcd68186 | 74 | task_alloc(taskq_t *tq, uint_t flags) |
75 | { | |
472a34ca BB |
76 | taskq_ent_t *t; |
77 | int count = 0; | |
bcd68186 | 78 | |
472a34ca BB |
79 | ASSERT(tq); |
80 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
bcd68186 | 81 | retry: |
472a34ca BB |
82 | /* Acquire taskq_ent_t's from free list if available */ |
83 | if (!list_empty(&tq->tq_free_list) && !(flags & TQ_NEW)) { | |
84 | t = list_entry(tq->tq_free_list.next, taskq_ent_t, tqent_list); | |
85 | ||
86 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
d9acd930 BB |
87 | ASSERT(!(t->tqent_flags & TQENT_FLAG_CANCEL)); |
88 | ASSERT(!timer_pending(&t->tqent_timer)); | |
472a34ca BB |
89 | |
90 | list_del_init(&t->tqent_list); | |
8d9a23e8 | 91 | return (t); |
472a34ca BB |
92 | } |
93 | ||
94 | /* Free list is empty and memory allocations are prohibited */ | |
95 | if (flags & TQ_NOALLOC) | |
8d9a23e8 | 96 | return (NULL); |
472a34ca BB |
97 | |
98 | /* Hit maximum taskq_ent_t pool size */ | |
99 | if (tq->tq_nalloc >= tq->tq_maxalloc) { | |
100 | if (flags & TQ_NOSLEEP) | |
8d9a23e8 | 101 | return (NULL); |
472a34ca BB |
102 | |
103 | /* | |
104 | * Sleep periodically polling the free list for an available | |
105 | * taskq_ent_t. Dispatching with TQ_SLEEP should always succeed | |
106 | * but we cannot block forever waiting for an taskq_ent_t to | |
107 | * show up in the free list, otherwise a deadlock can happen. | |
108 | * | |
109 | * Therefore, we need to allocate a new task even if the number | |
110 | * of allocated tasks is above tq->tq_maxalloc, but we still | |
111 | * end up delaying the task allocation by one second, thereby | |
112 | * throttling the task dispatch rate. | |
113 | */ | |
114 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
115 | schedule_timeout(HZ / 100); | |
116 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
8d9a23e8 BB |
117 | if (count < 100) { |
118 | count++; | |
119 | goto retry; | |
120 | } | |
472a34ca BB |
121 | } |
122 | ||
123 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
124 | t = kmem_alloc(sizeof(taskq_ent_t), task_km_flags(flags)); | |
125 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
126 | ||
127 | if (t) { | |
128 | taskq_init_ent(t); | |
129 | tq->tq_nalloc++; | |
130 | } | |
131 | ||
8d9a23e8 | 132 | return (t); |
bcd68186 | 133 | } |
134 | ||
82387586 | 135 | /* |
046a70c9 | 136 | * NOTE: Must be called with tq->tq_lock held, expects the taskq_ent_t |
bcd68186 | 137 | * to already be removed from the free, work, or pending taskq lists. |
138 | */ | |
139 | static void | |
046a70c9 | 140 | task_free(taskq_t *tq, taskq_ent_t *t) |
bcd68186 | 141 | { |
472a34ca BB |
142 | ASSERT(tq); |
143 | ASSERT(t); | |
bcd68186 | 144 | ASSERT(spin_is_locked(&tq->tq_lock)); |
046a70c9 | 145 | ASSERT(list_empty(&t->tqent_list)); |
d9acd930 | 146 | ASSERT(!timer_pending(&t->tqent_timer)); |
bcd68186 | 147 | |
472a34ca BB |
148 | kmem_free(t, sizeof(taskq_ent_t)); |
149 | tq->tq_nalloc--; | |
bcd68186 | 150 | } |
151 | ||
82387586 BB |
152 | /* |
153 | * NOTE: Must be called with tq->tq_lock held, either destroys the | |
046a70c9 | 154 | * taskq_ent_t if too many exist or moves it to the free list for later use. |
bcd68186 | 155 | */ |
f1ca4da6 | 156 | static void |
046a70c9 | 157 | task_done(taskq_t *tq, taskq_ent_t *t) |
f1ca4da6 | 158 | { |
bcd68186 | 159 | ASSERT(tq); |
160 | ASSERT(t); | |
161 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
162 | ||
d9acd930 BB |
163 | /* Wake tasks blocked in taskq_wait_id() */ |
164 | wake_up_all(&t->tqent_waitq); | |
165 | ||
046a70c9 | 166 | list_del_init(&t->tqent_list); |
f1ca4da6 | 167 | |
472a34ca | 168 | if (tq->tq_nalloc <= tq->tq_minalloc) { |
046a70c9 PS |
169 | t->tqent_id = 0; |
170 | t->tqent_func = NULL; | |
171 | t->tqent_arg = NULL; | |
44217f7a | 172 | t->tqent_flags = 0; |
8f2503e0 | 173 | |
472a34ca | 174 | list_add_tail(&t->tqent_list, &tq->tq_free_list); |
bcd68186 | 175 | } else { |
176 | task_free(tq, t); | |
177 | } | |
f1ca4da6 | 178 | } |
179 | ||
82387586 | 180 | /* |
d9acd930 BB |
181 | * When a delayed task timer expires remove it from the delay list and |
182 | * add it to the priority list in order for immediate processing. | |
bcd68186 | 183 | */ |
d9acd930 BB |
184 | static void |
185 | task_expire(unsigned long data) | |
bcd68186 | 186 | { |
d9acd930 BB |
187 | taskq_ent_t *w, *t = (taskq_ent_t *)data; |
188 | taskq_t *tq = t->tqent_taskq; | |
189 | struct list_head *l; | |
7257ec41 BB |
190 | |
191 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
d9acd930 BB |
192 | |
193 | if (t->tqent_flags & TQENT_FLAG_CANCEL) { | |
194 | ASSERT(list_empty(&t->tqent_list)); | |
195 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
196 | return; | |
197 | } | |
198 | ||
199 | /* | |
200 | * The priority list must be maintained in strict task id order | |
201 | * from lowest to highest for lowest_id to be easily calculable. | |
202 | */ | |
203 | list_del(&t->tqent_list); | |
204 | list_for_each_prev(l, &tq->tq_prio_list) { | |
205 | w = list_entry(l, taskq_ent_t, tqent_list); | |
206 | if (w->tqent_id < t->tqent_id) { | |
207 | list_add(&t->tqent_list, l); | |
208 | break; | |
209 | } | |
210 | } | |
211 | if (l == &tq->tq_prio_list) | |
212 | list_add(&t->tqent_list, &tq->tq_prio_list); | |
213 | ||
7257ec41 BB |
214 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
215 | ||
d9acd930 BB |
216 | wake_up(&tq->tq_work_waitq); |
217 | } | |
218 | ||
219 | /* | |
220 | * Returns the lowest incomplete taskqid_t. The taskqid_t may | |
221 | * be queued on the pending list, on the priority list, on the | |
222 | * delay list, or on the work list currently being handled, but | |
223 | * it is not 100% complete yet. | |
224 | */ | |
225 | static taskqid_t | |
226 | taskq_lowest_id(taskq_t *tq) | |
227 | { | |
228 | taskqid_t lowest_id = tq->tq_next_id; | |
229 | taskq_ent_t *t; | |
230 | taskq_thread_t *tqt; | |
d9acd930 BB |
231 | |
232 | ASSERT(tq); | |
233 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
234 | ||
235 | if (!list_empty(&tq->tq_pend_list)) { | |
236 | t = list_entry(tq->tq_pend_list.next, taskq_ent_t, tqent_list); | |
237 | lowest_id = MIN(lowest_id, t->tqent_id); | |
238 | } | |
239 | ||
240 | if (!list_empty(&tq->tq_prio_list)) { | |
241 | t = list_entry(tq->tq_prio_list.next, taskq_ent_t, tqent_list); | |
242 | lowest_id = MIN(lowest_id, t->tqent_id); | |
243 | } | |
244 | ||
245 | if (!list_empty(&tq->tq_delay_list)) { | |
246 | t = list_entry(tq->tq_delay_list.next, taskq_ent_t, tqent_list); | |
247 | lowest_id = MIN(lowest_id, t->tqent_id); | |
248 | } | |
249 | ||
250 | if (!list_empty(&tq->tq_active_list)) { | |
251 | tqt = list_entry(tq->tq_active_list.next, taskq_thread_t, | |
252 | tqt_active_list); | |
253 | ASSERT(tqt->tqt_id != 0); | |
254 | lowest_id = MIN(lowest_id, tqt->tqt_id); | |
255 | } | |
256 | ||
8d9a23e8 | 257 | return (lowest_id); |
d9acd930 BB |
258 | } |
259 | ||
260 | /* | |
261 | * Insert a task into a list keeping the list sorted by increasing taskqid. | |
262 | */ | |
263 | static void | |
264 | taskq_insert_in_order(taskq_t *tq, taskq_thread_t *tqt) | |
265 | { | |
266 | taskq_thread_t *w; | |
267 | struct list_head *l; | |
268 | ||
d9acd930 BB |
269 | ASSERT(tq); |
270 | ASSERT(tqt); | |
271 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
272 | ||
273 | list_for_each_prev(l, &tq->tq_active_list) { | |
274 | w = list_entry(l, taskq_thread_t, tqt_active_list); | |
275 | if (w->tqt_id < tqt->tqt_id) { | |
276 | list_add(&tqt->tqt_active_list, l); | |
277 | break; | |
278 | } | |
279 | } | |
280 | if (l == &tq->tq_active_list) | |
281 | list_add(&tqt->tqt_active_list, &tq->tq_active_list); | |
d9acd930 BB |
282 | } |
283 | ||
284 | /* | |
285 | * Find and return a task from the given list if it exists. The list | |
286 | * must be in lowest to highest task id order. | |
287 | */ | |
288 | static taskq_ent_t * | |
289 | taskq_find_list(taskq_t *tq, struct list_head *lh, taskqid_t id) | |
290 | { | |
291 | struct list_head *l; | |
292 | taskq_ent_t *t; | |
d9acd930 BB |
293 | |
294 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
295 | ||
296 | list_for_each(l, lh) { | |
297 | t = list_entry(l, taskq_ent_t, tqent_list); | |
298 | ||
299 | if (t->tqent_id == id) | |
8d9a23e8 | 300 | return (t); |
d9acd930 BB |
301 | |
302 | if (t->tqent_id > id) | |
303 | break; | |
304 | } | |
305 | ||
8d9a23e8 | 306 | return (NULL); |
bcd68186 | 307 | } |
308 | ||
d9acd930 BB |
309 | /* |
310 | * Find an already dispatched task given the task id regardless of what | |
311 | * state it is in. If a task is still pending or executing it will be | |
312 | * returned and 'active' set appropriately. If the task has already | |
313 | * been run then NULL is returned. | |
314 | */ | |
315 | static taskq_ent_t * | |
316 | taskq_find(taskq_t *tq, taskqid_t id, int *active) | |
317 | { | |
318 | taskq_thread_t *tqt; | |
319 | struct list_head *l; | |
320 | taskq_ent_t *t; | |
d9acd930 BB |
321 | |
322 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
323 | *active = 0; | |
324 | ||
325 | t = taskq_find_list(tq, &tq->tq_delay_list, id); | |
326 | if (t) | |
8d9a23e8 | 327 | return (t); |
d9acd930 BB |
328 | |
329 | t = taskq_find_list(tq, &tq->tq_prio_list, id); | |
330 | if (t) | |
8d9a23e8 | 331 | return (t); |
d9acd930 BB |
332 | |
333 | t = taskq_find_list(tq, &tq->tq_pend_list, id); | |
334 | if (t) | |
8d9a23e8 | 335 | return (t); |
d9acd930 BB |
336 | |
337 | list_for_each(l, &tq->tq_active_list) { | |
338 | tqt = list_entry(l, taskq_thread_t, tqt_active_list); | |
339 | if (tqt->tqt_id == id) { | |
340 | t = tqt->tqt_task; | |
341 | *active = 1; | |
8d9a23e8 | 342 | return (t); |
d9acd930 BB |
343 | } |
344 | } | |
345 | ||
8d9a23e8 | 346 | return (NULL); |
d9acd930 BB |
347 | } |
348 | ||
a876b030 CD |
349 | /* |
350 | * Theory for the taskq_wait_id(), taskq_wait_outstanding(), and | |
351 | * taskq_wait() functions below. | |
352 | * | |
353 | * Taskq waiting is accomplished by tracking the lowest outstanding task | |
354 | * id and the next available task id. As tasks are dispatched they are | |
355 | * added to the tail of the pending, priority, or delay lists. As worker | |
356 | * threads become available the tasks are removed from the heads of these | |
357 | * lists and linked to the worker threads. This ensures the lists are | |
358 | * kept sorted by lowest to highest task id. | |
359 | * | |
360 | * Therefore the lowest outstanding task id can be quickly determined by | |
361 | * checking the head item from all of these lists. This value is stored | |
362 | * with the taskq as the lowest id. It only needs to be recalculated when | |
363 | * either the task with the current lowest id completes or is canceled. | |
364 | * | |
365 | * By blocking until the lowest task id exceeds the passed task id the | |
366 | * taskq_wait_outstanding() function can be easily implemented. Similarly, | |
367 | * by blocking until the lowest task id matches the next task id taskq_wait() | |
368 | * can be implemented. | |
369 | * | |
370 | * Callers should be aware that when there are multiple worked threads it | |
371 | * is possible for larger task ids to complete before smaller ones. Also | |
372 | * when the taskq contains delay tasks with small task ids callers may | |
373 | * block for a considerable length of time waiting for them to expire and | |
374 | * execute. | |
375 | */ | |
99c452bb BB |
376 | static int |
377 | taskq_wait_id_check(taskq_t *tq, taskqid_t id) | |
f1ca4da6 | 378 | { |
d9acd930 | 379 | int active = 0; |
99c452bb | 380 | int rc; |
bcd68186 | 381 | |
d9acd930 | 382 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
99c452bb | 383 | rc = (taskq_find(tq, id, &active) == NULL); |
d9acd930 BB |
384 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
385 | ||
99c452bb BB |
386 | return (rc); |
387 | } | |
bcd68186 | 388 | |
99c452bb BB |
389 | /* |
390 | * The taskq_wait_id() function blocks until the passed task id completes. | |
391 | * This does not guarantee that all lower task ids have completed. | |
392 | */ | |
393 | void | |
394 | taskq_wait_id(taskq_t *tq, taskqid_t id) | |
395 | { | |
396 | wait_event(tq->tq_wait_waitq, taskq_wait_id_check(tq, id)); | |
bcd68186 | 397 | } |
aed8671c | 398 | EXPORT_SYMBOL(taskq_wait_id); |
bcd68186 | 399 | |
d9acd930 | 400 | static int |
a876b030 | 401 | taskq_wait_outstanding_check(taskq_t *tq, taskqid_t id) |
d9acd930 BB |
402 | { |
403 | int rc; | |
404 | ||
405 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
406 | rc = (id < tq->tq_lowest_id); | |
407 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
408 | ||
8d9a23e8 | 409 | return (rc); |
d9acd930 BB |
410 | } |
411 | ||
a876b030 CD |
412 | /* |
413 | * The taskq_wait_outstanding() function will block until all tasks with a | |
414 | * lower taskqid than the passed 'id' have been completed. Note that all | |
415 | * task id's are assigned monotonically at dispatch time. Zero may be | |
416 | * passed for the id to indicate all tasks dispatch up to this point, | |
417 | * but not after, should be waited for. | |
418 | */ | |
d9acd930 | 419 | void |
a876b030 | 420 | taskq_wait_outstanding(taskq_t *tq, taskqid_t id) |
d9acd930 | 421 | { |
a876b030 CD |
422 | wait_event(tq->tq_wait_waitq, |
423 | taskq_wait_outstanding_check(tq, id ? id : tq->tq_next_id - 1)); | |
d9acd930 | 424 | } |
a876b030 | 425 | EXPORT_SYMBOL(taskq_wait_outstanding); |
d9acd930 | 426 | |
a876b030 CD |
427 | static int |
428 | taskq_wait_check(taskq_t *tq) | |
bcd68186 | 429 | { |
a876b030 | 430 | int rc; |
bcd68186 | 431 | |
749045bb | 432 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
a876b030 | 433 | rc = (tq->tq_lowest_id == tq->tq_next_id); |
749045bb | 434 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 435 | |
a876b030 CD |
436 | return (rc); |
437 | } | |
438 | ||
439 | /* | |
440 | * The taskq_wait() function will block until the taskq is empty. | |
441 | * This means that if a taskq re-dispatches work to itself taskq_wait() | |
442 | * callers will block indefinitely. | |
443 | */ | |
444 | void | |
445 | taskq_wait(taskq_t *tq) | |
446 | { | |
447 | wait_event(tq->tq_wait_waitq, taskq_wait_check(tq)); | |
bcd68186 | 448 | } |
aed8671c | 449 | EXPORT_SYMBOL(taskq_wait); |
bcd68186 | 450 | |
451 | int | |
aed8671c | 452 | taskq_member(taskq_t *tq, void *t) |
bcd68186 | 453 | { |
2c02b71b PS |
454 | struct list_head *l; |
455 | taskq_thread_t *tqt; | |
f7a973d9 | 456 | int found = 0; |
bcd68186 | 457 | |
458 | ASSERT(tq); | |
472a34ca | 459 | ASSERT(t); |
bcd68186 | 460 | |
f7a973d9 | 461 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
2c02b71b PS |
462 | list_for_each(l, &tq->tq_thread_list) { |
463 | tqt = list_entry(l, taskq_thread_t, tqt_thread_list); | |
f7a973d9 BB |
464 | if (tqt->tqt_thread == (struct task_struct *)t) { |
465 | found = 1; | |
466 | break; | |
467 | } | |
2c02b71b | 468 | } |
f7a973d9 | 469 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 470 | |
f7a973d9 | 471 | return (found); |
bcd68186 | 472 | } |
aed8671c | 473 | EXPORT_SYMBOL(taskq_member); |
bcd68186 | 474 | |
d9acd930 BB |
475 | /* |
476 | * Cancel an already dispatched task given the task id. Still pending tasks | |
477 | * will be immediately canceled, and if the task is active the function will | |
478 | * block until it completes. Preallocated tasks which are canceled must be | |
479 | * freed by the caller. | |
480 | */ | |
481 | int | |
482 | taskq_cancel_id(taskq_t *tq, taskqid_t id) | |
483 | { | |
484 | taskq_ent_t *t; | |
485 | int active = 0; | |
486 | int rc = ENOENT; | |
d9acd930 BB |
487 | |
488 | ASSERT(tq); | |
489 | ||
490 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
491 | t = taskq_find(tq, id, &active); | |
492 | if (t && !active) { | |
493 | list_del_init(&t->tqent_list); | |
494 | t->tqent_flags |= TQENT_FLAG_CANCEL; | |
495 | ||
496 | /* | |
497 | * When canceling the lowest outstanding task id we | |
498 | * must recalculate the new lowest outstanding id. | |
499 | */ | |
500 | if (tq->tq_lowest_id == t->tqent_id) { | |
501 | tq->tq_lowest_id = taskq_lowest_id(tq); | |
502 | ASSERT3S(tq->tq_lowest_id, >, t->tqent_id); | |
503 | } | |
504 | ||
505 | /* | |
506 | * The task_expire() function takes the tq->tq_lock so drop | |
507 | * drop the lock before synchronously cancelling the timer. | |
508 | */ | |
509 | if (timer_pending(&t->tqent_timer)) { | |
510 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
511 | del_timer_sync(&t->tqent_timer); | |
512 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
513 | } | |
514 | ||
515 | if (!(t->tqent_flags & TQENT_FLAG_PREALLOC)) | |
516 | task_done(tq, t); | |
517 | ||
518 | rc = 0; | |
519 | } | |
520 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
521 | ||
522 | if (active) { | |
523 | taskq_wait_id(tq, id); | |
524 | rc = EBUSY; | |
525 | } | |
526 | ||
8d9a23e8 | 527 | return (rc); |
d9acd930 BB |
528 | } |
529 | EXPORT_SYMBOL(taskq_cancel_id); | |
530 | ||
bcd68186 | 531 | taskqid_t |
aed8671c | 532 | taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t flags) |
bcd68186 | 533 | { |
472a34ca | 534 | taskq_ent_t *t; |
bcd68186 | 535 | taskqid_t rc = 0; |
f1ca4da6 | 536 | |
472a34ca BB |
537 | ASSERT(tq); |
538 | ASSERT(func); | |
d05ec4b4 | 539 | |
472a34ca | 540 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
f1ca4da6 | 541 | |
bcd68186 | 542 | /* Taskq being destroyed and all tasks drained */ |
f7a973d9 | 543 | if (!(tq->tq_flags & TASKQ_ACTIVE)) |
8d9a23e8 | 544 | goto out; |
f1ca4da6 | 545 | |
bcd68186 | 546 | /* Do not queue the task unless there is idle thread for it */ |
547 | ASSERT(tq->tq_nactive <= tq->tq_nthreads); | |
548 | if ((flags & TQ_NOQUEUE) && (tq->tq_nactive == tq->tq_nthreads)) | |
8d9a23e8 | 549 | goto out; |
bcd68186 | 550 | |
472a34ca | 551 | if ((t = task_alloc(tq, flags)) == NULL) |
8d9a23e8 | 552 | goto out; |
f1ca4da6 | 553 | |
046a70c9 | 554 | spin_lock(&t->tqent_lock); |
f0d8bb26 NB |
555 | |
556 | /* Queue to the priority list instead of the pending list */ | |
557 | if (flags & TQ_FRONT) | |
046a70c9 | 558 | list_add_tail(&t->tqent_list, &tq->tq_prio_list); |
f0d8bb26 | 559 | else |
046a70c9 | 560 | list_add_tail(&t->tqent_list, &tq->tq_pend_list); |
f0d8bb26 | 561 | |
046a70c9 | 562 | t->tqent_id = rc = tq->tq_next_id; |
bcd68186 | 563 | tq->tq_next_id++; |
472a34ca BB |
564 | t->tqent_func = func; |
565 | t->tqent_arg = arg; | |
d9acd930 BB |
566 | t->tqent_taskq = tq; |
567 | t->tqent_timer.data = 0; | |
568 | t->tqent_timer.function = NULL; | |
569 | t->tqent_timer.expires = 0; | |
44217f7a PS |
570 | |
571 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
572 | ||
046a70c9 | 573 | spin_unlock(&t->tqent_lock); |
0bb43ca2 NB |
574 | |
575 | wake_up(&tq->tq_work_waitq); | |
bcd68186 | 576 | out: |
749045bb | 577 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
8d9a23e8 | 578 | return (rc); |
f1ca4da6 | 579 | } |
aed8671c | 580 | EXPORT_SYMBOL(taskq_dispatch); |
44217f7a | 581 | |
d9acd930 BB |
582 | taskqid_t |
583 | taskq_dispatch_delay(taskq_t *tq, task_func_t func, void *arg, | |
584 | uint_t flags, clock_t expire_time) | |
585 | { | |
d9acd930 | 586 | taskqid_t rc = 0; |
8d9a23e8 | 587 | taskq_ent_t *t; |
d9acd930 BB |
588 | |
589 | ASSERT(tq); | |
590 | ASSERT(func); | |
591 | ||
592 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
593 | ||
594 | /* Taskq being destroyed and all tasks drained */ | |
f7a973d9 | 595 | if (!(tq->tq_flags & TASKQ_ACTIVE)) |
8d9a23e8 | 596 | goto out; |
d9acd930 BB |
597 | |
598 | if ((t = task_alloc(tq, flags)) == NULL) | |
8d9a23e8 | 599 | goto out; |
d9acd930 BB |
600 | |
601 | spin_lock(&t->tqent_lock); | |
602 | ||
603 | /* Queue to the delay list for subsequent execution */ | |
604 | list_add_tail(&t->tqent_list, &tq->tq_delay_list); | |
605 | ||
606 | t->tqent_id = rc = tq->tq_next_id; | |
607 | tq->tq_next_id++; | |
608 | t->tqent_func = func; | |
609 | t->tqent_arg = arg; | |
610 | t->tqent_taskq = tq; | |
611 | t->tqent_timer.data = (unsigned long)t; | |
612 | t->tqent_timer.function = task_expire; | |
613 | t->tqent_timer.expires = (unsigned long)expire_time; | |
614 | add_timer(&t->tqent_timer); | |
615 | ||
616 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
617 | ||
618 | spin_unlock(&t->tqent_lock); | |
619 | out: | |
620 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
8d9a23e8 | 621 | return (rc); |
d9acd930 BB |
622 | } |
623 | EXPORT_SYMBOL(taskq_dispatch_delay); | |
624 | ||
44217f7a | 625 | void |
aed8671c | 626 | taskq_dispatch_ent(taskq_t *tq, task_func_t func, void *arg, uint_t flags, |
44217f7a PS |
627 | taskq_ent_t *t) |
628 | { | |
44217f7a PS |
629 | ASSERT(tq); |
630 | ASSERT(func); | |
44217f7a PS |
631 | |
632 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
633 | ||
634 | /* Taskq being destroyed and all tasks drained */ | |
f7a973d9 | 635 | if (!(tq->tq_flags & TASKQ_ACTIVE)) { |
44217f7a PS |
636 | t->tqent_id = 0; |
637 | goto out; | |
638 | } | |
639 | ||
640 | spin_lock(&t->tqent_lock); | |
641 | ||
642 | /* | |
643 | * Mark it as a prealloc'd task. This is important | |
644 | * to ensure that we don't free it later. | |
645 | */ | |
646 | t->tqent_flags |= TQENT_FLAG_PREALLOC; | |
647 | ||
648 | /* Queue to the priority list instead of the pending list */ | |
649 | if (flags & TQ_FRONT) | |
650 | list_add_tail(&t->tqent_list, &tq->tq_prio_list); | |
651 | else | |
652 | list_add_tail(&t->tqent_list, &tq->tq_pend_list); | |
653 | ||
654 | t->tqent_id = tq->tq_next_id; | |
655 | tq->tq_next_id++; | |
656 | t->tqent_func = func; | |
657 | t->tqent_arg = arg; | |
d9acd930 | 658 | t->tqent_taskq = tq; |
44217f7a PS |
659 | |
660 | spin_unlock(&t->tqent_lock); | |
661 | ||
662 | wake_up(&tq->tq_work_waitq); | |
663 | out: | |
0bb43ca2 | 664 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
44217f7a | 665 | } |
aed8671c | 666 | EXPORT_SYMBOL(taskq_dispatch_ent); |
44217f7a PS |
667 | |
668 | int | |
aed8671c | 669 | taskq_empty_ent(taskq_ent_t *t) |
44217f7a PS |
670 | { |
671 | return list_empty(&t->tqent_list); | |
672 | } | |
aed8671c | 673 | EXPORT_SYMBOL(taskq_empty_ent); |
44217f7a PS |
674 | |
675 | void | |
aed8671c | 676 | taskq_init_ent(taskq_ent_t *t) |
44217f7a PS |
677 | { |
678 | spin_lock_init(&t->tqent_lock); | |
d9acd930 BB |
679 | init_waitqueue_head(&t->tqent_waitq); |
680 | init_timer(&t->tqent_timer); | |
44217f7a PS |
681 | INIT_LIST_HEAD(&t->tqent_list); |
682 | t->tqent_id = 0; | |
683 | t->tqent_func = NULL; | |
684 | t->tqent_arg = NULL; | |
685 | t->tqent_flags = 0; | |
d9acd930 | 686 | t->tqent_taskq = NULL; |
44217f7a | 687 | } |
aed8671c | 688 | EXPORT_SYMBOL(taskq_init_ent); |
44217f7a | 689 | |
f7a973d9 BB |
690 | /* |
691 | * Return the next pending task, preference is given to tasks on the | |
692 | * priority list which were dispatched with TQ_FRONT. | |
693 | */ | |
694 | static taskq_ent_t * | |
695 | taskq_next_ent(taskq_t *tq) | |
696 | { | |
697 | struct list_head *list; | |
698 | ||
699 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
700 | ||
701 | if (!list_empty(&tq->tq_prio_list)) | |
702 | list = &tq->tq_prio_list; | |
703 | else if (!list_empty(&tq->tq_pend_list)) | |
704 | list = &tq->tq_pend_list; | |
705 | else | |
706 | return (NULL); | |
707 | ||
708 | return (list_entry(list->next, taskq_ent_t, tqent_list)); | |
709 | } | |
710 | ||
711 | /* | |
712 | * Spawns a new thread for the specified taskq. | |
713 | */ | |
714 | static void | |
715 | taskq_thread_spawn_task(void *arg) | |
716 | { | |
717 | taskq_t *tq = (taskq_t *)arg; | |
718 | ||
719 | (void) taskq_thread_create(tq); | |
720 | ||
721 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
722 | tq->tq_nspawn--; | |
723 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
724 | } | |
725 | ||
726 | /* | |
727 | * Spawn addition threads for dynamic taskqs (TASKQ_DYNMAIC) the current | |
728 | * number of threads is insufficient to handle the pending tasks. These | |
729 | * new threads must be created by the dedicated dynamic_taskq to avoid | |
730 | * deadlocks between thread creation and memory reclaim. The system_taskq | |
731 | * which is also a dynamic taskq cannot be safely used for this. | |
732 | */ | |
733 | static int | |
734 | taskq_thread_spawn(taskq_t *tq, int seq_tasks) | |
735 | { | |
736 | int spawning = 0; | |
737 | ||
738 | if (!(tq->tq_flags & TASKQ_DYNAMIC)) | |
739 | return (0); | |
740 | ||
741 | if ((seq_tasks > spl_taskq_thread_sequential) && | |
742 | (tq->tq_nthreads + tq->tq_nspawn < tq->tq_maxthreads) && | |
743 | (tq->tq_flags & TASKQ_ACTIVE)) { | |
744 | spawning = (++tq->tq_nspawn); | |
745 | taskq_dispatch(dynamic_taskq, taskq_thread_spawn_task, | |
746 | tq, TQ_NOSLEEP); | |
747 | } | |
748 | ||
749 | return (spawning); | |
750 | } | |
751 | ||
752 | /* | |
753 | * Threads in a dynamic taskq should only exit once it has been completely | |
754 | * drained and no other threads are actively servicing tasks. This prevents | |
755 | * threads from being created and destroyed more than is required. | |
756 | * | |
757 | * The first thread is the thread list is treated as the primary thread. | |
758 | * There is nothing special about the primary thread but in order to avoid | |
759 | * all the taskq pids from changing we opt to make it long running. | |
760 | */ | |
761 | static int | |
762 | taskq_thread_should_stop(taskq_t *tq, taskq_thread_t *tqt) | |
763 | { | |
764 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
765 | ||
766 | if (!(tq->tq_flags & TASKQ_DYNAMIC)) | |
767 | return (0); | |
768 | ||
769 | if (list_first_entry(&(tq->tq_thread_list), taskq_thread_t, | |
770 | tqt_thread_list) == tqt) | |
771 | return (0); | |
772 | ||
773 | return | |
774 | ((tq->tq_nspawn == 0) && /* No threads are being spawned */ | |
775 | (tq->tq_nactive == 0) && /* No threads are handling tasks */ | |
776 | (tq->tq_nthreads > 1) && /* More than 1 thread is running */ | |
777 | (!taskq_next_ent(tq)) && /* There are no pending tasks */ | |
778 | (spl_taskq_thread_dynamic));/* Dynamic taskqs are allowed */ | |
779 | } | |
780 | ||
bcd68186 | 781 | static int |
782 | taskq_thread(void *args) | |
783 | { | |
472a34ca BB |
784 | DECLARE_WAITQUEUE(wait, current); |
785 | sigset_t blocked; | |
2c02b71b | 786 | taskq_thread_t *tqt = args; |
472a34ca BB |
787 | taskq_t *tq; |
788 | taskq_ent_t *t; | |
f7a973d9 | 789 | int seq_tasks = 0; |
bcd68186 | 790 | |
472a34ca | 791 | ASSERT(tqt); |
2c02b71b | 792 | tq = tqt->tqt_tq; |
472a34ca | 793 | current->flags |= PF_NOFREEZE; |
bcd68186 | 794 | |
472a34ca BB |
795 | sigfillset(&blocked); |
796 | sigprocmask(SIG_BLOCK, &blocked, NULL); | |
797 | flush_signals(current); | |
bcd68186 | 798 | |
472a34ca | 799 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
f7a973d9 BB |
800 | |
801 | /* Immediately exit if more threads than allowed were created. */ | |
802 | if (tq->tq_nthreads >= tq->tq_maxthreads) | |
803 | goto error; | |
804 | ||
472a34ca | 805 | tq->tq_nthreads++; |
f7a973d9 | 806 | list_add_tail(&tqt->tqt_thread_list, &tq->tq_thread_list); |
472a34ca BB |
807 | wake_up(&tq->tq_wait_waitq); |
808 | set_current_state(TASK_INTERRUPTIBLE); | |
bcd68186 | 809 | |
472a34ca | 810 | while (!kthread_should_stop()) { |
bcd68186 | 811 | |
f0d8bb26 NB |
812 | if (list_empty(&tq->tq_pend_list) && |
813 | list_empty(&tq->tq_prio_list)) { | |
f7a973d9 BB |
814 | |
815 | if (taskq_thread_should_stop(tq, tqt)) { | |
816 | wake_up_all(&tq->tq_wait_waitq); | |
817 | break; | |
818 | } | |
819 | ||
3c6ed541 | 820 | add_wait_queue_exclusive(&tq->tq_work_waitq, &wait); |
749045bb | 821 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
f7a973d9 | 822 | |
bcd68186 | 823 | schedule(); |
f7a973d9 BB |
824 | seq_tasks = 0; |
825 | ||
749045bb | 826 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
3c6ed541 | 827 | remove_wait_queue(&tq->tq_work_waitq, &wait); |
bcd68186 | 828 | } else { |
829 | __set_current_state(TASK_RUNNING); | |
830 | } | |
831 | ||
f7a973d9 | 832 | if ((t = taskq_next_ent(tq)) != NULL) { |
472a34ca | 833 | list_del_init(&t->tqent_list); |
8f2503e0 | 834 | |
44217f7a PS |
835 | /* In order to support recursively dispatching a |
836 | * preallocated taskq_ent_t, tqent_id must be | |
837 | * stored prior to executing tqent_func. */ | |
e7e5f78e | 838 | tqt->tqt_id = t->tqent_id; |
d9acd930 | 839 | tqt->tqt_task = t; |
8f2503e0 PS |
840 | |
841 | /* We must store a copy of the flags prior to | |
842 | * servicing the task (servicing a prealloc'd task | |
843 | * returns the ownership of the tqent back to | |
844 | * the caller of taskq_dispatch). Thus, | |
845 | * tqent_flags _may_ change within the call. */ | |
846 | tqt->tqt_flags = t->tqent_flags; | |
847 | ||
2c02b71b | 848 | taskq_insert_in_order(tq, tqt); |
472a34ca | 849 | tq->tq_nactive++; |
749045bb | 850 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 851 | |
852 | /* Perform the requested task */ | |
472a34ca | 853 | t->tqent_func(t->tqent_arg); |
bcd68186 | 854 | |
749045bb | 855 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
472a34ca | 856 | tq->tq_nactive--; |
2c02b71b | 857 | list_del_init(&tqt->tqt_active_list); |
d9acd930 | 858 | tqt->tqt_task = NULL; |
8f2503e0 PS |
859 | |
860 | /* For prealloc'd tasks, we don't free anything. */ | |
f7a973d9 | 861 | if (!(tqt->tqt_flags & TQENT_FLAG_PREALLOC)) |
8f2503e0 | 862 | task_done(tq, t); |
bcd68186 | 863 | |
7257ec41 BB |
864 | /* When the current lowest outstanding taskqid is |
865 | * done calculate the new lowest outstanding id */ | |
e7e5f78e | 866 | if (tq->tq_lowest_id == tqt->tqt_id) { |
bcd68186 | 867 | tq->tq_lowest_id = taskq_lowest_id(tq); |
e7e5f78e | 868 | ASSERT3S(tq->tq_lowest_id, >, tqt->tqt_id); |
bcd68186 | 869 | } |
870 | ||
f7a973d9 BB |
871 | /* Spawn additional taskq threads if required. */ |
872 | if (taskq_thread_spawn(tq, ++seq_tasks)) | |
873 | seq_tasks = 0; | |
874 | ||
e7e5f78e | 875 | tqt->tqt_id = 0; |
8f2503e0 | 876 | tqt->tqt_flags = 0; |
472a34ca | 877 | wake_up_all(&tq->tq_wait_waitq); |
f7a973d9 BB |
878 | } else { |
879 | if (taskq_thread_should_stop(tq, tqt)) | |
880 | break; | |
bcd68186 | 881 | } |
882 | ||
883 | set_current_state(TASK_INTERRUPTIBLE); | |
884 | ||
472a34ca | 885 | } |
bcd68186 | 886 | |
887 | __set_current_state(TASK_RUNNING); | |
472a34ca | 888 | tq->tq_nthreads--; |
2c02b71b | 889 | list_del_init(&tqt->tqt_thread_list); |
f7a973d9 BB |
890 | error: |
891 | kmem_free(tqt, sizeof (taskq_thread_t)); | |
472a34ca | 892 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 893 | |
8d9a23e8 | 894 | return (0); |
bcd68186 | 895 | } |
896 | ||
f7a973d9 BB |
897 | static taskq_thread_t * |
898 | taskq_thread_create(taskq_t *tq) | |
899 | { | |
900 | static int last_used_cpu = 0; | |
901 | taskq_thread_t *tqt; | |
902 | ||
903 | tqt = kmem_alloc(sizeof (*tqt), KM_PUSHPAGE); | |
904 | INIT_LIST_HEAD(&tqt->tqt_thread_list); | |
905 | INIT_LIST_HEAD(&tqt->tqt_active_list); | |
906 | tqt->tqt_tq = tq; | |
907 | tqt->tqt_id = 0; | |
908 | ||
909 | tqt->tqt_thread = spl_kthread_create(taskq_thread, tqt, | |
910 | "%s", tq->tq_name); | |
911 | if (tqt->tqt_thread == NULL) { | |
912 | kmem_free(tqt, sizeof (taskq_thread_t)); | |
913 | return (NULL); | |
914 | } | |
915 | ||
916 | if (spl_taskq_thread_bind) { | |
917 | last_used_cpu = (last_used_cpu + 1) % num_online_cpus(); | |
918 | kthread_bind(tqt->tqt_thread, last_used_cpu); | |
919 | } | |
920 | ||
62aa81a5 BB |
921 | if (spl_taskq_thread_priority) |
922 | set_user_nice(tqt->tqt_thread, PRIO_TO_NICE(tq->tq_pri)); | |
923 | ||
f7a973d9 BB |
924 | wake_up_process(tqt->tqt_thread); |
925 | ||
926 | return (tqt); | |
927 | } | |
928 | ||
f1ca4da6 | 929 | taskq_t * |
aed8671c | 930 | taskq_create(const char *name, int nthreads, pri_t pri, |
472a34ca | 931 | int minalloc, int maxalloc, uint_t flags) |
f1ca4da6 | 932 | { |
472a34ca | 933 | taskq_t *tq; |
2c02b71b | 934 | taskq_thread_t *tqt; |
f7a973d9 | 935 | int count = 0, rc = 0, i; |
bcd68186 | 936 | |
472a34ca | 937 | ASSERT(name != NULL); |
472a34ca BB |
938 | ASSERT(minalloc >= 0); |
939 | ASSERT(maxalloc <= INT_MAX); | |
f7a973d9 | 940 | ASSERT(!(flags & (TASKQ_CPR_SAFE))); /* Unsupported */ |
bcd68186 | 941 | |
915404bd BB |
942 | /* Scale the number of threads using nthreads as a percentage */ |
943 | if (flags & TASKQ_THREADS_CPU_PCT) { | |
944 | ASSERT(nthreads <= 100); | |
945 | ASSERT(nthreads >= 0); | |
946 | nthreads = MIN(nthreads, 100); | |
947 | nthreads = MAX(nthreads, 0); | |
948 | nthreads = MAX((num_online_cpus() * nthreads) / 100, 1); | |
949 | } | |
950 | ||
f7a973d9 | 951 | tq = kmem_alloc(sizeof (*tq), KM_PUSHPAGE); |
472a34ca | 952 | if (tq == NULL) |
8d9a23e8 | 953 | return (NULL); |
bcd68186 | 954 | |
472a34ca | 955 | spin_lock_init(&tq->tq_lock); |
472a34ca BB |
956 | INIT_LIST_HEAD(&tq->tq_thread_list); |
957 | INIT_LIST_HEAD(&tq->tq_active_list); | |
f7a973d9 BB |
958 | tq->tq_name = strdup(name); |
959 | tq->tq_nactive = 0; | |
960 | tq->tq_nthreads = 0; | |
961 | tq->tq_nspawn = 0; | |
962 | tq->tq_maxthreads = nthreads; | |
963 | tq->tq_pri = pri; | |
964 | tq->tq_minalloc = minalloc; | |
965 | tq->tq_maxalloc = maxalloc; | |
966 | tq->tq_nalloc = 0; | |
967 | tq->tq_flags = (flags | TASKQ_ACTIVE); | |
968 | tq->tq_next_id = 1; | |
969 | tq->tq_lowest_id = 1; | |
472a34ca BB |
970 | INIT_LIST_HEAD(&tq->tq_free_list); |
971 | INIT_LIST_HEAD(&tq->tq_pend_list); | |
972 | INIT_LIST_HEAD(&tq->tq_prio_list); | |
d9acd930 | 973 | INIT_LIST_HEAD(&tq->tq_delay_list); |
472a34ca BB |
974 | init_waitqueue_head(&tq->tq_work_waitq); |
975 | init_waitqueue_head(&tq->tq_wait_waitq); | |
bcd68186 | 976 | |
f7a973d9 BB |
977 | if (flags & TASKQ_PREPOPULATE) { |
978 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
979 | ||
472a34ca BB |
980 | for (i = 0; i < minalloc; i++) |
981 | task_done(tq, task_alloc(tq, TQ_PUSHPAGE | TQ_NEW)); | |
6e605b6e | 982 | |
f7a973d9 BB |
983 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
984 | } | |
985 | ||
986 | if ((flags & TASKQ_DYNAMIC) && spl_taskq_thread_dynamic) | |
987 | nthreads = 1; | |
6e605b6e | 988 | |
2c02b71b | 989 | for (i = 0; i < nthreads; i++) { |
f7a973d9 BB |
990 | tqt = taskq_thread_create(tq); |
991 | if (tqt == NULL) | |
2c02b71b | 992 | rc = 1; |
f7a973d9 BB |
993 | else |
994 | count++; | |
2c02b71b | 995 | } |
bcd68186 | 996 | |
472a34ca | 997 | /* Wait for all threads to be started before potential destroy */ |
f7a973d9 | 998 | wait_event(tq->tq_wait_waitq, tq->tq_nthreads == count); |
bcd68186 | 999 | |
472a34ca | 1000 | if (rc) { |
aed8671c | 1001 | taskq_destroy(tq); |
472a34ca BB |
1002 | tq = NULL; |
1003 | } | |
bcd68186 | 1004 | |
8d9a23e8 | 1005 | return (tq); |
f1ca4da6 | 1006 | } |
aed8671c | 1007 | EXPORT_SYMBOL(taskq_create); |
b123971f | 1008 | |
1009 | void | |
aed8671c | 1010 | taskq_destroy(taskq_t *tq) |
b123971f | 1011 | { |
2c02b71b PS |
1012 | struct task_struct *thread; |
1013 | taskq_thread_t *tqt; | |
046a70c9 | 1014 | taskq_ent_t *t; |
b123971f | 1015 | |
bcd68186 | 1016 | ASSERT(tq); |
749045bb | 1017 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
f7a973d9 | 1018 | tq->tq_flags &= ~TASKQ_ACTIVE; |
749045bb | 1019 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 1020 | |
f7a973d9 BB |
1021 | /* |
1022 | * When TASKQ_ACTIVE is clear new tasks may not be added nor may | |
1023 | * new worker threads be spawned for dynamic taskq. | |
1024 | */ | |
1025 | if (dynamic_taskq != NULL) | |
1026 | taskq_wait_outstanding(dynamic_taskq, 0); | |
1027 | ||
aed8671c | 1028 | taskq_wait(tq); |
bcd68186 | 1029 | |
472a34ca | 1030 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 1031 | |
2c02b71b PS |
1032 | /* |
1033 | * Signal each thread to exit and block until it does. Each thread | |
1034 | * is responsible for removing itself from the list and freeing its | |
1035 | * taskq_thread_t. This allows for idle threads to opt to remove | |
1036 | * themselves from the taskq. They can be recreated as needed. | |
1037 | */ | |
1038 | while (!list_empty(&tq->tq_thread_list)) { | |
1039 | tqt = list_entry(tq->tq_thread_list.next, | |
f7a973d9 | 1040 | taskq_thread_t, tqt_thread_list); |
2c02b71b PS |
1041 | thread = tqt->tqt_thread; |
1042 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
1043 | ||
1044 | kthread_stop(thread); | |
1045 | ||
472a34ca | 1046 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
2c02b71b PS |
1047 | } |
1048 | ||
472a34ca | 1049 | while (!list_empty(&tq->tq_free_list)) { |
046a70c9 | 1050 | t = list_entry(tq->tq_free_list.next, taskq_ent_t, tqent_list); |
44217f7a PS |
1051 | |
1052 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
1053 | ||
472a34ca BB |
1054 | list_del_init(&t->tqent_list); |
1055 | task_free(tq, t); | |
1056 | } | |
bcd68186 | 1057 | |
f7a973d9 BB |
1058 | ASSERT0(tq->tq_nthreads); |
1059 | ASSERT0(tq->tq_nalloc); | |
1060 | ASSERT0(tq->tq_nspawn); | |
472a34ca BB |
1061 | ASSERT(list_empty(&tq->tq_thread_list)); |
1062 | ASSERT(list_empty(&tq->tq_active_list)); | |
1063 | ASSERT(list_empty(&tq->tq_free_list)); | |
1064 | ASSERT(list_empty(&tq->tq_pend_list)); | |
1065 | ASSERT(list_empty(&tq->tq_prio_list)); | |
d9acd930 | 1066 | ASSERT(list_empty(&tq->tq_delay_list)); |
bcd68186 | 1067 | |
472a34ca | 1068 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
2c02b71b | 1069 | |
f7a973d9 BB |
1070 | strfree(tq->tq_name); |
1071 | kmem_free(tq, sizeof (taskq_t)); | |
b123971f | 1072 | } |
aed8671c | 1073 | EXPORT_SYMBOL(taskq_destroy); |
e9cb2b4f BB |
1074 | |
1075 | int | |
1076 | spl_taskq_init(void) | |
1077 | { | |
3c82160f | 1078 | system_taskq = taskq_create("spl_system_taskq", MAX(boot_ncpus, 64), |
9dc5ffbe | 1079 | maxclsyspri, boot_ncpus, INT_MAX, TASKQ_PREPOPULATE|TASKQ_DYNAMIC); |
e9cb2b4f | 1080 | if (system_taskq == NULL) |
8d9a23e8 | 1081 | return (1); |
e9cb2b4f | 1082 | |
f7a973d9 | 1083 | dynamic_taskq = taskq_create("spl_dynamic_taskq", 1, |
9dc5ffbe | 1084 | maxclsyspri, boot_ncpus, INT_MAX, TASKQ_PREPOPULATE); |
f7a973d9 BB |
1085 | if (dynamic_taskq == NULL) { |
1086 | taskq_destroy(system_taskq); | |
1087 | return (1); | |
1088 | } | |
1089 | ||
8d9a23e8 | 1090 | return (0); |
e9cb2b4f BB |
1091 | } |
1092 | ||
1093 | void | |
1094 | spl_taskq_fini(void) | |
1095 | { | |
f7a973d9 BB |
1096 | taskq_destroy(dynamic_taskq); |
1097 | dynamic_taskq = NULL; | |
1098 | ||
e9cb2b4f | 1099 | taskq_destroy(system_taskq); |
f7a973d9 | 1100 | system_taskq = NULL; |
e9cb2b4f | 1101 | } |