]>
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 BB |
8 | * This file is part of the SPL, Solaris Porting Layer. |
9 | * For details, see <http://github.com/behlendorf/spl/>. | |
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 | { | |
046a70c9 | 60 | taskq_ent_t *t; |
bcd68186 | 61 | int count = 0; |
b17edc10 | 62 | SENTRY; |
bcd68186 | 63 | |
64 | ASSERT(tq); | |
3d061e9d | 65 | ASSERT(spin_is_locked(&tq->tq_lock)); |
bcd68186 | 66 | retry: |
046a70c9 | 67 | /* Acquire taskq_ent_t's from free list if available */ |
bcd68186 | 68 | if (!list_empty(&tq->tq_free_list) && !(flags & TQ_NEW)) { |
046a70c9 | 69 | t = list_entry(tq->tq_free_list.next, taskq_ent_t, tqent_list); |
44217f7a PS |
70 | |
71 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
72 | ||
046a70c9 | 73 | list_del_init(&t->tqent_list); |
b17edc10 | 74 | SRETURN(t); |
bcd68186 | 75 | } |
76 | ||
7257ec41 | 77 | /* Free list is empty and memory allocations are prohibited */ |
bcd68186 | 78 | if (flags & TQ_NOALLOC) |
b17edc10 | 79 | SRETURN(NULL); |
bcd68186 | 80 | |
046a70c9 | 81 | /* Hit maximum taskq_ent_t pool size */ |
bcd68186 | 82 | if (tq->tq_nalloc >= tq->tq_maxalloc) { |
83 | if (flags & TQ_NOSLEEP) | |
b17edc10 | 84 | SRETURN(NULL); |
bcd68186 | 85 | |
26f7245c RC |
86 | /* |
87 | * Sleep periodically polling the free list for an available | |
046a70c9 PS |
88 | * taskq_ent_t. Dispatching with TQ_SLEEP should always succeed |
89 | * but we cannot block forever waiting for an taskq_entq_t to | |
26f7245c RC |
90 | * show up in the free list, otherwise a deadlock can happen. |
91 | * | |
92 | * Therefore, we need to allocate a new task even if the number | |
93 | * of allocated tasks is above tq->tq_maxalloc, but we still | |
94 | * end up delaying the task allocation by one second, thereby | |
95 | * throttling the task dispatch rate. | |
96 | */ | |
97 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
98 | schedule_timeout(HZ / 100); | |
99 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
100 | if (count < 100) | |
101 | SGOTO(retry, count++); | |
bcd68186 | 102 | } |
103 | ||
26f7245c | 104 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
9b51f218 | 105 | t = kmem_alloc(sizeof(taskq_ent_t), task_km_flags(flags)); |
749045bb | 106 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 107 | |
26f7245c | 108 | if (t) { |
44217f7a | 109 | taskq_init_ent(t); |
26f7245c RC |
110 | tq->tq_nalloc++; |
111 | } | |
bcd68186 | 112 | |
b17edc10 | 113 | SRETURN(t); |
bcd68186 | 114 | } |
115 | ||
82387586 | 116 | /* |
046a70c9 | 117 | * NOTE: Must be called with tq->tq_lock held, expects the taskq_ent_t |
bcd68186 | 118 | * to already be removed from the free, work, or pending taskq lists. |
119 | */ | |
120 | static void | |
046a70c9 | 121 | task_free(taskq_t *tq, taskq_ent_t *t) |
bcd68186 | 122 | { |
b17edc10 | 123 | SENTRY; |
bcd68186 | 124 | |
125 | ASSERT(tq); | |
126 | ASSERT(t); | |
127 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
046a70c9 | 128 | ASSERT(list_empty(&t->tqent_list)); |
bcd68186 | 129 | |
046a70c9 | 130 | kmem_free(t, sizeof(taskq_ent_t)); |
bcd68186 | 131 | tq->tq_nalloc--; |
f1ca4da6 | 132 | |
b17edc10 | 133 | SEXIT; |
bcd68186 | 134 | } |
135 | ||
82387586 BB |
136 | /* |
137 | * NOTE: Must be called with tq->tq_lock held, either destroys the | |
046a70c9 | 138 | * taskq_ent_t if too many exist or moves it to the free list for later use. |
bcd68186 | 139 | */ |
f1ca4da6 | 140 | static void |
046a70c9 | 141 | task_done(taskq_t *tq, taskq_ent_t *t) |
f1ca4da6 | 142 | { |
b17edc10 | 143 | SENTRY; |
bcd68186 | 144 | ASSERT(tq); |
145 | ASSERT(t); | |
146 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
147 | ||
046a70c9 | 148 | list_del_init(&t->tqent_list); |
f1ca4da6 | 149 | |
bcd68186 | 150 | if (tq->tq_nalloc <= tq->tq_minalloc) { |
046a70c9 PS |
151 | t->tqent_id = 0; |
152 | t->tqent_func = NULL; | |
153 | t->tqent_arg = NULL; | |
44217f7a | 154 | t->tqent_flags = 0; |
8f2503e0 | 155 | |
046a70c9 | 156 | list_add_tail(&t->tqent_list, &tq->tq_free_list); |
bcd68186 | 157 | } else { |
158 | task_free(tq, t); | |
159 | } | |
f1ca4da6 | 160 | |
b17edc10 | 161 | SEXIT; |
f1ca4da6 | 162 | } |
163 | ||
82387586 BB |
164 | /* |
165 | * As tasks are submitted to the task queue they are assigned a | |
f0d8bb26 NB |
166 | * monotonically increasing taskqid and added to the tail of the pending |
167 | * list. As worker threads become available the tasks are removed from | |
168 | * the head of the pending or priority list, giving preference to the | |
05b8f50c PS |
169 | * priority list. The tasks are then removed from their respective |
170 | * list, and the taskq_thread servicing the task is added to the active | |
171 | * list, preserving the order using the serviced task's taskqid. | |
172 | * Finally, as tasks complete the taskq_thread servicing the task is | |
173 | * removed from the active list. This means that the pending task and | |
174 | * active taskq_thread lists are always kept sorted by taskqid. Thus the | |
175 | * lowest outstanding incomplete taskqid can be determined simply by | |
176 | * checking the min taskqid for each head item on the pending, priority, | |
177 | * and active taskq_thread list. This value is stored in | |
178 | * tq->tq_lowest_id and only updated to the new lowest id when the | |
179 | * previous lowest id completes. All taskqids lower than | |
180 | * tq->tq_lowest_id must have completed. It is also possible larger | |
181 | * taskqid's have completed because they may be processed in parallel by | |
182 | * several worker threads. However, this is not a problem because the | |
183 | * behavior of taskq_wait_id() is to block until all previously | |
184 | * submitted taskqid's have completed. | |
82387586 BB |
185 | * |
186 | * XXX: Taskqid_t wrapping is not handled. However, taskqid_t's are | |
187 | * 64-bit values so even if a taskq is processing 2^24 (16,777,216) | |
188 | * taskqid_ts per second it will still take 2^40 seconds, 34,865 years, | |
189 | * before the wrap occurs. I can live with that for now. | |
bcd68186 | 190 | */ |
191 | static int | |
192 | taskq_wait_check(taskq_t *tq, taskqid_t id) | |
193 | { | |
7257ec41 BB |
194 | int rc; |
195 | ||
196 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
197 | rc = (id < tq->tq_lowest_id); | |
198 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
199 | ||
b17edc10 | 200 | SRETURN(rc); |
bcd68186 | 201 | } |
202 | ||
bcd68186 | 203 | void |
204 | __taskq_wait_id(taskq_t *tq, taskqid_t id) | |
f1ca4da6 | 205 | { |
b17edc10 | 206 | SENTRY; |
bcd68186 | 207 | ASSERT(tq); |
208 | ||
209 | wait_event(tq->tq_wait_waitq, taskq_wait_check(tq, id)); | |
210 | ||
b17edc10 | 211 | SEXIT; |
bcd68186 | 212 | } |
213 | EXPORT_SYMBOL(__taskq_wait_id); | |
214 | ||
215 | void | |
216 | __taskq_wait(taskq_t *tq) | |
217 | { | |
218 | taskqid_t id; | |
b17edc10 | 219 | SENTRY; |
bcd68186 | 220 | ASSERT(tq); |
221 | ||
7257ec41 | 222 | /* Wait for the largest outstanding taskqid */ |
749045bb | 223 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
7257ec41 | 224 | id = tq->tq_next_id - 1; |
749045bb | 225 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 226 | |
227 | __taskq_wait_id(tq, id); | |
228 | ||
b17edc10 | 229 | SEXIT; |
bcd68186 | 230 | |
231 | } | |
232 | EXPORT_SYMBOL(__taskq_wait); | |
233 | ||
234 | int | |
235 | __taskq_member(taskq_t *tq, void *t) | |
236 | { | |
2c02b71b PS |
237 | struct list_head *l; |
238 | taskq_thread_t *tqt; | |
b17edc10 | 239 | SENTRY; |
bcd68186 | 240 | |
241 | ASSERT(tq); | |
242 | ASSERT(t); | |
243 | ||
2c02b71b PS |
244 | list_for_each(l, &tq->tq_thread_list) { |
245 | tqt = list_entry(l, taskq_thread_t, tqt_thread_list); | |
246 | if (tqt->tqt_thread == (struct task_struct *)t) | |
247 | SRETURN(1); | |
248 | } | |
bcd68186 | 249 | |
b17edc10 | 250 | SRETURN(0); |
bcd68186 | 251 | } |
252 | EXPORT_SYMBOL(__taskq_member); | |
253 | ||
254 | taskqid_t | |
255 | __taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t flags) | |
256 | { | |
046a70c9 | 257 | taskq_ent_t *t; |
bcd68186 | 258 | taskqid_t rc = 0; |
b17edc10 | 259 | SENTRY; |
f1ca4da6 | 260 | |
937879f1 | 261 | ASSERT(tq); |
262 | ASSERT(func); | |
d05ec4b4 | 263 | |
749045bb | 264 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
f1ca4da6 | 265 | |
bcd68186 | 266 | /* Taskq being destroyed and all tasks drained */ |
267 | if (!(tq->tq_flags & TQ_ACTIVE)) | |
b17edc10 | 268 | SGOTO(out, rc = 0); |
f1ca4da6 | 269 | |
bcd68186 | 270 | /* Do not queue the task unless there is idle thread for it */ |
271 | ASSERT(tq->tq_nactive <= tq->tq_nthreads); | |
272 | if ((flags & TQ_NOQUEUE) && (tq->tq_nactive == tq->tq_nthreads)) | |
b17edc10 | 273 | SGOTO(out, rc = 0); |
bcd68186 | 274 | |
275 | if ((t = task_alloc(tq, flags)) == NULL) | |
b17edc10 | 276 | SGOTO(out, rc = 0); |
f1ca4da6 | 277 | |
046a70c9 | 278 | spin_lock(&t->tqent_lock); |
f0d8bb26 NB |
279 | |
280 | /* Queue to the priority list instead of the pending list */ | |
281 | if (flags & TQ_FRONT) | |
046a70c9 | 282 | list_add_tail(&t->tqent_list, &tq->tq_prio_list); |
f0d8bb26 | 283 | else |
046a70c9 | 284 | list_add_tail(&t->tqent_list, &tq->tq_pend_list); |
f0d8bb26 | 285 | |
046a70c9 | 286 | t->tqent_id = rc = tq->tq_next_id; |
bcd68186 | 287 | tq->tq_next_id++; |
046a70c9 PS |
288 | t->tqent_func = func; |
289 | t->tqent_arg = arg; | |
44217f7a PS |
290 | |
291 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
292 | ||
046a70c9 | 293 | spin_unlock(&t->tqent_lock); |
0bb43ca2 NB |
294 | |
295 | wake_up(&tq->tq_work_waitq); | |
bcd68186 | 296 | out: |
749045bb | 297 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
b17edc10 | 298 | SRETURN(rc); |
f1ca4da6 | 299 | } |
f1b59d26 | 300 | EXPORT_SYMBOL(__taskq_dispatch); |
44217f7a PS |
301 | |
302 | void | |
303 | __taskq_dispatch_ent(taskq_t *tq, task_func_t func, void *arg, uint_t flags, | |
304 | taskq_ent_t *t) | |
305 | { | |
306 | SENTRY; | |
307 | ||
308 | ASSERT(tq); | |
309 | ASSERT(func); | |
310 | ASSERT(!(tq->tq_flags & TASKQ_DYNAMIC)); | |
311 | ||
312 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
313 | ||
314 | /* Taskq being destroyed and all tasks drained */ | |
315 | if (!(tq->tq_flags & TQ_ACTIVE)) { | |
316 | t->tqent_id = 0; | |
317 | goto out; | |
318 | } | |
319 | ||
320 | spin_lock(&t->tqent_lock); | |
321 | ||
322 | /* | |
323 | * Mark it as a prealloc'd task. This is important | |
324 | * to ensure that we don't free it later. | |
325 | */ | |
326 | t->tqent_flags |= TQENT_FLAG_PREALLOC; | |
327 | ||
328 | /* Queue to the priority list instead of the pending list */ | |
329 | if (flags & TQ_FRONT) | |
330 | list_add_tail(&t->tqent_list, &tq->tq_prio_list); | |
331 | else | |
332 | list_add_tail(&t->tqent_list, &tq->tq_pend_list); | |
333 | ||
334 | t->tqent_id = tq->tq_next_id; | |
335 | tq->tq_next_id++; | |
336 | t->tqent_func = func; | |
337 | t->tqent_arg = arg; | |
338 | ||
339 | spin_unlock(&t->tqent_lock); | |
340 | ||
341 | wake_up(&tq->tq_work_waitq); | |
342 | out: | |
0bb43ca2 | 343 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
44217f7a PS |
344 | SEXIT; |
345 | } | |
346 | EXPORT_SYMBOL(__taskq_dispatch_ent); | |
347 | ||
348 | int | |
349 | __taskq_empty_ent(taskq_ent_t *t) | |
350 | { | |
351 | return list_empty(&t->tqent_list); | |
352 | } | |
353 | EXPORT_SYMBOL(__taskq_empty_ent); | |
354 | ||
355 | void | |
356 | __taskq_init_ent(taskq_ent_t *t) | |
357 | { | |
358 | spin_lock_init(&t->tqent_lock); | |
359 | INIT_LIST_HEAD(&t->tqent_list); | |
360 | t->tqent_id = 0; | |
361 | t->tqent_func = NULL; | |
362 | t->tqent_arg = NULL; | |
363 | t->tqent_flags = 0; | |
364 | } | |
365 | EXPORT_SYMBOL(__taskq_init_ent); | |
366 | ||
82387586 BB |
367 | /* |
368 | * Returns the lowest incomplete taskqid_t. The taskqid_t may | |
f0d8bb26 NB |
369 | * be queued on the pending list, on the priority list, or on |
370 | * the work list currently being handled, but it is not 100% | |
371 | * complete yet. | |
82387586 | 372 | */ |
bcd68186 | 373 | static taskqid_t |
374 | taskq_lowest_id(taskq_t *tq) | |
375 | { | |
7257ec41 | 376 | taskqid_t lowest_id = tq->tq_next_id; |
046a70c9 | 377 | taskq_ent_t *t; |
2c02b71b | 378 | taskq_thread_t *tqt; |
b17edc10 | 379 | SENTRY; |
bcd68186 | 380 | |
381 | ASSERT(tq); | |
382 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
383 | ||
82387586 | 384 | if (!list_empty(&tq->tq_pend_list)) { |
046a70c9 PS |
385 | t = list_entry(tq->tq_pend_list.next, taskq_ent_t, tqent_list); |
386 | lowest_id = MIN(lowest_id, t->tqent_id); | |
82387586 | 387 | } |
bcd68186 | 388 | |
f0d8bb26 | 389 | if (!list_empty(&tq->tq_prio_list)) { |
046a70c9 PS |
390 | t = list_entry(tq->tq_prio_list.next, taskq_ent_t, tqent_list); |
391 | lowest_id = MIN(lowest_id, t->tqent_id); | |
f0d8bb26 NB |
392 | } |
393 | ||
2c02b71b PS |
394 | if (!list_empty(&tq->tq_active_list)) { |
395 | tqt = list_entry(tq->tq_active_list.next, taskq_thread_t, | |
396 | tqt_active_list); | |
e7e5f78e PS |
397 | ASSERT(tqt->tqt_id != 0); |
398 | lowest_id = MIN(lowest_id, tqt->tqt_id); | |
82387586 | 399 | } |
bcd68186 | 400 | |
b17edc10 | 401 | SRETURN(lowest_id); |
bcd68186 | 402 | } |
403 | ||
f0d8bb26 NB |
404 | /* |
405 | * Insert a task into a list keeping the list sorted by increasing | |
406 | * taskqid. | |
407 | */ | |
408 | static void | |
2c02b71b | 409 | taskq_insert_in_order(taskq_t *tq, taskq_thread_t *tqt) |
f0d8bb26 | 410 | { |
2c02b71b | 411 | taskq_thread_t *w; |
f0d8bb26 NB |
412 | struct list_head *l; |
413 | ||
b17edc10 | 414 | SENTRY; |
f0d8bb26 | 415 | ASSERT(tq); |
2c02b71b | 416 | ASSERT(tqt); |
f0d8bb26 NB |
417 | ASSERT(spin_is_locked(&tq->tq_lock)); |
418 | ||
2c02b71b PS |
419 | list_for_each_prev(l, &tq->tq_active_list) { |
420 | w = list_entry(l, taskq_thread_t, tqt_active_list); | |
e7e5f78e | 421 | if (w->tqt_id < tqt->tqt_id) { |
2c02b71b | 422 | list_add(&tqt->tqt_active_list, l); |
f0d8bb26 NB |
423 | break; |
424 | } | |
425 | } | |
2c02b71b PS |
426 | if (l == &tq->tq_active_list) |
427 | list_add(&tqt->tqt_active_list, &tq->tq_active_list); | |
f0d8bb26 | 428 | |
b17edc10 | 429 | SEXIT; |
f0d8bb26 NB |
430 | } |
431 | ||
bcd68186 | 432 | static int |
433 | taskq_thread(void *args) | |
434 | { | |
435 | DECLARE_WAITQUEUE(wait, current); | |
436 | sigset_t blocked; | |
2c02b71b PS |
437 | taskq_thread_t *tqt = args; |
438 | taskq_t *tq; | |
046a70c9 | 439 | taskq_ent_t *t; |
f0d8bb26 | 440 | struct list_head *pend_list; |
b17edc10 | 441 | SENTRY; |
bcd68186 | 442 | |
2c02b71b PS |
443 | ASSERT(tqt); |
444 | tq = tqt->tqt_tq; | |
bcd68186 | 445 | current->flags |= PF_NOFREEZE; |
446 | ||
447 | sigfillset(&blocked); | |
448 | sigprocmask(SIG_BLOCK, &blocked, NULL); | |
449 | flush_signals(current); | |
450 | ||
749045bb | 451 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 452 | tq->tq_nthreads++; |
453 | wake_up(&tq->tq_wait_waitq); | |
454 | set_current_state(TASK_INTERRUPTIBLE); | |
455 | ||
456 | while (!kthread_should_stop()) { | |
457 | ||
f0d8bb26 NB |
458 | if (list_empty(&tq->tq_pend_list) && |
459 | list_empty(&tq->tq_prio_list)) { | |
3c6ed541 | 460 | add_wait_queue_exclusive(&tq->tq_work_waitq, &wait); |
749045bb | 461 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 462 | schedule(); |
749045bb | 463 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
3c6ed541 | 464 | remove_wait_queue(&tq->tq_work_waitq, &wait); |
bcd68186 | 465 | } else { |
466 | __set_current_state(TASK_RUNNING); | |
467 | } | |
468 | ||
f0d8bb26 NB |
469 | |
470 | if (!list_empty(&tq->tq_prio_list)) | |
471 | pend_list = &tq->tq_prio_list; | |
472 | else if (!list_empty(&tq->tq_pend_list)) | |
473 | pend_list = &tq->tq_pend_list; | |
474 | else | |
475 | pend_list = NULL; | |
476 | ||
477 | if (pend_list) { | |
046a70c9 PS |
478 | t = list_entry(pend_list->next, taskq_ent_t, tqent_list); |
479 | list_del_init(&t->tqent_list); | |
8f2503e0 | 480 | |
44217f7a PS |
481 | /* In order to support recursively dispatching a |
482 | * preallocated taskq_ent_t, tqent_id must be | |
483 | * stored prior to executing tqent_func. */ | |
e7e5f78e | 484 | tqt->tqt_id = t->tqent_id; |
8f2503e0 PS |
485 | |
486 | /* We must store a copy of the flags prior to | |
487 | * servicing the task (servicing a prealloc'd task | |
488 | * returns the ownership of the tqent back to | |
489 | * the caller of taskq_dispatch). Thus, | |
490 | * tqent_flags _may_ change within the call. */ | |
491 | tqt->tqt_flags = t->tqent_flags; | |
492 | ||
2c02b71b | 493 | taskq_insert_in_order(tq, tqt); |
bcd68186 | 494 | tq->tq_nactive++; |
749045bb | 495 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 496 | |
497 | /* Perform the requested task */ | |
046a70c9 | 498 | t->tqent_func(t->tqent_arg); |
bcd68186 | 499 | |
749045bb | 500 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 501 | tq->tq_nactive--; |
2c02b71b | 502 | list_del_init(&tqt->tqt_active_list); |
8f2503e0 PS |
503 | |
504 | /* For prealloc'd tasks, we don't free anything. */ | |
505 | if ((tq->tq_flags & TASKQ_DYNAMIC) || | |
506 | !(tqt->tqt_flags & TQENT_FLAG_PREALLOC)) | |
507 | task_done(tq, t); | |
bcd68186 | 508 | |
7257ec41 BB |
509 | /* When the current lowest outstanding taskqid is |
510 | * done calculate the new lowest outstanding id */ | |
e7e5f78e | 511 | if (tq->tq_lowest_id == tqt->tqt_id) { |
bcd68186 | 512 | tq->tq_lowest_id = taskq_lowest_id(tq); |
e7e5f78e | 513 | ASSERT3S(tq->tq_lowest_id, >, tqt->tqt_id); |
bcd68186 | 514 | } |
515 | ||
e7e5f78e | 516 | tqt->tqt_id = 0; |
8f2503e0 | 517 | tqt->tqt_flags = 0; |
bcd68186 | 518 | wake_up_all(&tq->tq_wait_waitq); |
519 | } | |
520 | ||
521 | set_current_state(TASK_INTERRUPTIBLE); | |
522 | ||
523 | } | |
524 | ||
525 | __set_current_state(TASK_RUNNING); | |
526 | tq->tq_nthreads--; | |
2c02b71b PS |
527 | list_del_init(&tqt->tqt_thread_list); |
528 | kmem_free(tqt, sizeof(taskq_thread_t)); | |
529 | ||
749045bb | 530 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 531 | |
b17edc10 | 532 | SRETURN(0); |
bcd68186 | 533 | } |
534 | ||
f1ca4da6 | 535 | taskq_t * |
536 | __taskq_create(const char *name, int nthreads, pri_t pri, | |
537 | int minalloc, int maxalloc, uint_t flags) | |
538 | { | |
bcd68186 | 539 | taskq_t *tq; |
2c02b71b | 540 | taskq_thread_t *tqt; |
bcd68186 | 541 | int rc = 0, i, j = 0; |
b17edc10 | 542 | SENTRY; |
bcd68186 | 543 | |
544 | ASSERT(name != NULL); | |
545 | ASSERT(pri <= maxclsyspri); | |
546 | ASSERT(minalloc >= 0); | |
547 | ASSERT(maxalloc <= INT_MAX); | |
548 | ASSERT(!(flags & (TASKQ_CPR_SAFE | TASKQ_DYNAMIC))); /* Unsupported */ | |
549 | ||
915404bd BB |
550 | /* Scale the number of threads using nthreads as a percentage */ |
551 | if (flags & TASKQ_THREADS_CPU_PCT) { | |
552 | ASSERT(nthreads <= 100); | |
553 | ASSERT(nthreads >= 0); | |
554 | nthreads = MIN(nthreads, 100); | |
555 | nthreads = MAX(nthreads, 0); | |
556 | nthreads = MAX((num_online_cpus() * nthreads) / 100, 1); | |
557 | } | |
558 | ||
9b51f218 | 559 | tq = kmem_alloc(sizeof(*tq), KM_PUSHPAGE); |
bcd68186 | 560 | if (tq == NULL) |
b17edc10 | 561 | SRETURN(NULL); |
bcd68186 | 562 | |
bcd68186 | 563 | spin_lock_init(&tq->tq_lock); |
749045bb | 564 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
2c02b71b PS |
565 | INIT_LIST_HEAD(&tq->tq_thread_list); |
566 | INIT_LIST_HEAD(&tq->tq_active_list); | |
bcd68186 | 567 | tq->tq_name = name; |
568 | tq->tq_nactive = 0; | |
569 | tq->tq_nthreads = 0; | |
570 | tq->tq_pri = pri; | |
571 | tq->tq_minalloc = minalloc; | |
572 | tq->tq_maxalloc = maxalloc; | |
573 | tq->tq_nalloc = 0; | |
574 | tq->tq_flags = (flags | TQ_ACTIVE); | |
575 | tq->tq_next_id = 1; | |
576 | tq->tq_lowest_id = 1; | |
577 | INIT_LIST_HEAD(&tq->tq_free_list); | |
bcd68186 | 578 | INIT_LIST_HEAD(&tq->tq_pend_list); |
f0d8bb26 | 579 | INIT_LIST_HEAD(&tq->tq_prio_list); |
bcd68186 | 580 | init_waitqueue_head(&tq->tq_work_waitq); |
581 | init_waitqueue_head(&tq->tq_wait_waitq); | |
582 | ||
583 | if (flags & TASKQ_PREPOPULATE) | |
584 | for (i = 0; i < minalloc; i++) | |
9b51f218 | 585 | task_done(tq, task_alloc(tq, TQ_PUSHPAGE | TQ_NEW)); |
6e605b6e | 586 | |
749045bb | 587 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
6e605b6e | 588 | |
2c02b71b | 589 | for (i = 0; i < nthreads; i++) { |
9b51f218 | 590 | tqt = kmem_alloc(sizeof(*tqt), KM_PUSHPAGE); |
2c02b71b PS |
591 | INIT_LIST_HEAD(&tqt->tqt_thread_list); |
592 | INIT_LIST_HEAD(&tqt->tqt_active_list); | |
593 | tqt->tqt_tq = tq; | |
e7e5f78e | 594 | tqt->tqt_id = 0; |
2c02b71b PS |
595 | |
596 | tqt->tqt_thread = kthread_create(taskq_thread, tqt, | |
597 | "%s/%d", name, i); | |
598 | if (tqt->tqt_thread) { | |
599 | list_add(&tqt->tqt_thread_list, &tq->tq_thread_list); | |
600 | kthread_bind(tqt->tqt_thread, i % num_online_cpus()); | |
601 | set_user_nice(tqt->tqt_thread, PRIO_TO_NICE(pri)); | |
602 | wake_up_process(tqt->tqt_thread); | |
bcd68186 | 603 | j++; |
2c02b71b PS |
604 | } else { |
605 | kmem_free(tqt, sizeof(taskq_thread_t)); | |
606 | rc = 1; | |
607 | } | |
608 | } | |
bcd68186 | 609 | |
610 | /* Wait for all threads to be started before potential destroy */ | |
611 | wait_event(tq->tq_wait_waitq, tq->tq_nthreads == j); | |
612 | ||
613 | if (rc) { | |
614 | __taskq_destroy(tq); | |
615 | tq = NULL; | |
616 | } | |
617 | ||
b17edc10 | 618 | SRETURN(tq); |
f1ca4da6 | 619 | } |
f1b59d26 | 620 | EXPORT_SYMBOL(__taskq_create); |
b123971f | 621 | |
622 | void | |
623 | __taskq_destroy(taskq_t *tq) | |
624 | { | |
2c02b71b PS |
625 | struct task_struct *thread; |
626 | taskq_thread_t *tqt; | |
046a70c9 | 627 | taskq_ent_t *t; |
b17edc10 | 628 | SENTRY; |
b123971f | 629 | |
bcd68186 | 630 | ASSERT(tq); |
749045bb | 631 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 632 | tq->tq_flags &= ~TQ_ACTIVE; |
749045bb | 633 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 634 | |
635 | /* TQ_ACTIVE cleared prevents new tasks being added to pending */ | |
636 | __taskq_wait(tq); | |
637 | ||
749045bb | 638 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 639 | |
2c02b71b PS |
640 | /* |
641 | * Signal each thread to exit and block until it does. Each thread | |
642 | * is responsible for removing itself from the list and freeing its | |
643 | * taskq_thread_t. This allows for idle threads to opt to remove | |
644 | * themselves from the taskq. They can be recreated as needed. | |
645 | */ | |
646 | while (!list_empty(&tq->tq_thread_list)) { | |
647 | tqt = list_entry(tq->tq_thread_list.next, | |
648 | taskq_thread_t, tqt_thread_list); | |
649 | thread = tqt->tqt_thread; | |
650 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
651 | ||
652 | kthread_stop(thread); | |
653 | ||
654 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
655 | } | |
656 | ||
bcd68186 | 657 | while (!list_empty(&tq->tq_free_list)) { |
046a70c9 | 658 | t = list_entry(tq->tq_free_list.next, taskq_ent_t, tqent_list); |
44217f7a PS |
659 | |
660 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
661 | ||
046a70c9 | 662 | list_del_init(&t->tqent_list); |
bcd68186 | 663 | task_free(tq, t); |
664 | } | |
665 | ||
666 | ASSERT(tq->tq_nthreads == 0); | |
667 | ASSERT(tq->tq_nalloc == 0); | |
2c02b71b PS |
668 | ASSERT(list_empty(&tq->tq_thread_list)); |
669 | ASSERT(list_empty(&tq->tq_active_list)); | |
bcd68186 | 670 | ASSERT(list_empty(&tq->tq_free_list)); |
bcd68186 | 671 | ASSERT(list_empty(&tq->tq_pend_list)); |
f0d8bb26 | 672 | ASSERT(list_empty(&tq->tq_prio_list)); |
bcd68186 | 673 | |
749045bb | 674 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
2c02b71b | 675 | |
bcd68186 | 676 | kmem_free(tq, sizeof(taskq_t)); |
677 | ||
b17edc10 | 678 | SEXIT; |
b123971f | 679 | } |
bcd68186 | 680 | EXPORT_SYMBOL(__taskq_destroy); |
e9cb2b4f BB |
681 | |
682 | int | |
683 | spl_taskq_init(void) | |
684 | { | |
b17edc10 | 685 | SENTRY; |
e9cb2b4f | 686 | |
f220894e BB |
687 | /* Solaris creates a dynamic taskq of up to 64 threads, however in |
688 | * a Linux environment 1 thread per-core is usually about right */ | |
689 | system_taskq = taskq_create("spl_system_taskq", num_online_cpus(), | |
690 | minclsyspri, 4, 512, TASKQ_PREPOPULATE); | |
e9cb2b4f | 691 | if (system_taskq == NULL) |
b17edc10 | 692 | SRETURN(1); |
e9cb2b4f | 693 | |
b17edc10 | 694 | SRETURN(0); |
e9cb2b4f BB |
695 | } |
696 | ||
697 | void | |
698 | spl_taskq_fini(void) | |
699 | { | |
b17edc10 | 700 | SENTRY; |
e9cb2b4f | 701 | taskq_destroy(system_taskq); |
b17edc10 | 702 | SEXIT; |
e9cb2b4f | 703 | } |