]>
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 | ||
82387586 BB |
41 | /* |
42 | * NOTE: Must be called with tq->tq_lock held, returns a list_t which | |
bcd68186 | 43 | * is not attached to the free, work, or pending taskq lists. |
f1ca4da6 | 44 | */ |
046a70c9 | 45 | static taskq_ent_t * |
bcd68186 | 46 | task_alloc(taskq_t *tq, uint_t flags) |
47 | { | |
046a70c9 | 48 | taskq_ent_t *t; |
bcd68186 | 49 | int count = 0; |
b17edc10 | 50 | SENTRY; |
bcd68186 | 51 | |
52 | ASSERT(tq); | |
53 | ASSERT(flags & (TQ_SLEEP | TQ_NOSLEEP)); /* One set */ | |
54 | ASSERT(!((flags & TQ_SLEEP) && (flags & TQ_NOSLEEP))); /* Not both */ | |
3d061e9d | 55 | ASSERT(spin_is_locked(&tq->tq_lock)); |
bcd68186 | 56 | retry: |
046a70c9 | 57 | /* Acquire taskq_ent_t's from free list if available */ |
bcd68186 | 58 | if (!list_empty(&tq->tq_free_list) && !(flags & TQ_NEW)) { |
046a70c9 | 59 | t = list_entry(tq->tq_free_list.next, taskq_ent_t, tqent_list); |
44217f7a PS |
60 | |
61 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
62 | ||
046a70c9 | 63 | list_del_init(&t->tqent_list); |
b17edc10 | 64 | SRETURN(t); |
bcd68186 | 65 | } |
66 | ||
7257ec41 | 67 | /* Free list is empty and memory allocations are prohibited */ |
bcd68186 | 68 | if (flags & TQ_NOALLOC) |
b17edc10 | 69 | SRETURN(NULL); |
bcd68186 | 70 | |
046a70c9 | 71 | /* Hit maximum taskq_ent_t pool size */ |
bcd68186 | 72 | if (tq->tq_nalloc >= tq->tq_maxalloc) { |
73 | if (flags & TQ_NOSLEEP) | |
b17edc10 | 74 | SRETURN(NULL); |
bcd68186 | 75 | |
26f7245c RC |
76 | /* |
77 | * Sleep periodically polling the free list for an available | |
046a70c9 PS |
78 | * taskq_ent_t. Dispatching with TQ_SLEEP should always succeed |
79 | * but we cannot block forever waiting for an taskq_entq_t to | |
26f7245c RC |
80 | * show up in the free list, otherwise a deadlock can happen. |
81 | * | |
82 | * Therefore, we need to allocate a new task even if the number | |
83 | * of allocated tasks is above tq->tq_maxalloc, but we still | |
84 | * end up delaying the task allocation by one second, thereby | |
85 | * throttling the task dispatch rate. | |
86 | */ | |
87 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
88 | schedule_timeout(HZ / 100); | |
89 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
90 | if (count < 100) | |
91 | SGOTO(retry, count++); | |
bcd68186 | 92 | } |
93 | ||
26f7245c | 94 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
046a70c9 | 95 | t = kmem_alloc(sizeof(taskq_ent_t), flags & (TQ_SLEEP | TQ_NOSLEEP)); |
749045bb | 96 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 97 | |
26f7245c | 98 | if (t) { |
44217f7a | 99 | taskq_init_ent(t); |
26f7245c RC |
100 | tq->tq_nalloc++; |
101 | } | |
bcd68186 | 102 | |
b17edc10 | 103 | SRETURN(t); |
bcd68186 | 104 | } |
105 | ||
82387586 | 106 | /* |
046a70c9 | 107 | * NOTE: Must be called with tq->tq_lock held, expects the taskq_ent_t |
bcd68186 | 108 | * to already be removed from the free, work, or pending taskq lists. |
109 | */ | |
110 | static void | |
046a70c9 | 111 | task_free(taskq_t *tq, taskq_ent_t *t) |
bcd68186 | 112 | { |
b17edc10 | 113 | SENTRY; |
bcd68186 | 114 | |
115 | ASSERT(tq); | |
116 | ASSERT(t); | |
117 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
046a70c9 | 118 | ASSERT(list_empty(&t->tqent_list)); |
bcd68186 | 119 | |
046a70c9 | 120 | kmem_free(t, sizeof(taskq_ent_t)); |
bcd68186 | 121 | tq->tq_nalloc--; |
f1ca4da6 | 122 | |
b17edc10 | 123 | SEXIT; |
bcd68186 | 124 | } |
125 | ||
82387586 BB |
126 | /* |
127 | * NOTE: Must be called with tq->tq_lock held, either destroys the | |
046a70c9 | 128 | * taskq_ent_t if too many exist or moves it to the free list for later use. |
bcd68186 | 129 | */ |
f1ca4da6 | 130 | static void |
046a70c9 | 131 | task_done(taskq_t *tq, taskq_ent_t *t) |
f1ca4da6 | 132 | { |
b17edc10 | 133 | SENTRY; |
bcd68186 | 134 | ASSERT(tq); |
135 | ASSERT(t); | |
136 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
137 | ||
046a70c9 | 138 | list_del_init(&t->tqent_list); |
f1ca4da6 | 139 | |
bcd68186 | 140 | if (tq->tq_nalloc <= tq->tq_minalloc) { |
046a70c9 PS |
141 | t->tqent_id = 0; |
142 | t->tqent_func = NULL; | |
143 | t->tqent_arg = NULL; | |
44217f7a | 144 | t->tqent_flags = 0; |
8f2503e0 | 145 | |
046a70c9 | 146 | list_add_tail(&t->tqent_list, &tq->tq_free_list); |
bcd68186 | 147 | } else { |
148 | task_free(tq, t); | |
149 | } | |
f1ca4da6 | 150 | |
b17edc10 | 151 | SEXIT; |
f1ca4da6 | 152 | } |
153 | ||
82387586 BB |
154 | /* |
155 | * As tasks are submitted to the task queue they are assigned a | |
f0d8bb26 NB |
156 | * monotonically increasing taskqid and added to the tail of the pending |
157 | * list. As worker threads become available the tasks are removed from | |
158 | * the head of the pending or priority list, giving preference to the | |
159 | * priority list. The tasks are then added to the work list, preserving | |
160 | * the ordering by taskqid. Finally, as tasks complete they are removed | |
161 | * from the work list. This means that the pending and work lists are | |
162 | * always kept sorted by taskqid. Thus the lowest outstanding | |
82387586 | 163 | * incomplete taskqid can be determined simply by checking the min |
f0d8bb26 NB |
164 | * taskqid for each head item on the pending, priority, and work list. |
165 | * This value is stored in tq->tq_lowest_id and only updated to the new | |
166 | * lowest id when the previous lowest id completes. All taskqids lower | |
167 | * than tq->tq_lowest_id must have completed. It is also possible | |
168 | * larger taskqid's have completed because they may be processed in | |
169 | * parallel by several worker threads. However, this is not a problem | |
170 | * because the behavior of taskq_wait_id() is to block until all | |
171 | * previously submitted taskqid's have completed. | |
82387586 BB |
172 | * |
173 | * XXX: Taskqid_t wrapping is not handled. However, taskqid_t's are | |
174 | * 64-bit values so even if a taskq is processing 2^24 (16,777,216) | |
175 | * taskqid_ts per second it will still take 2^40 seconds, 34,865 years, | |
176 | * before the wrap occurs. I can live with that for now. | |
bcd68186 | 177 | */ |
178 | static int | |
179 | taskq_wait_check(taskq_t *tq, taskqid_t id) | |
180 | { | |
7257ec41 BB |
181 | int rc; |
182 | ||
183 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
184 | rc = (id < tq->tq_lowest_id); | |
185 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
186 | ||
b17edc10 | 187 | SRETURN(rc); |
bcd68186 | 188 | } |
189 | ||
bcd68186 | 190 | void |
191 | __taskq_wait_id(taskq_t *tq, taskqid_t id) | |
f1ca4da6 | 192 | { |
b17edc10 | 193 | SENTRY; |
bcd68186 | 194 | ASSERT(tq); |
195 | ||
196 | wait_event(tq->tq_wait_waitq, taskq_wait_check(tq, id)); | |
197 | ||
b17edc10 | 198 | SEXIT; |
bcd68186 | 199 | } |
200 | EXPORT_SYMBOL(__taskq_wait_id); | |
201 | ||
202 | void | |
203 | __taskq_wait(taskq_t *tq) | |
204 | { | |
205 | taskqid_t id; | |
b17edc10 | 206 | SENTRY; |
bcd68186 | 207 | ASSERT(tq); |
208 | ||
7257ec41 | 209 | /* Wait for the largest outstanding taskqid */ |
749045bb | 210 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
7257ec41 | 211 | id = tq->tq_next_id - 1; |
749045bb | 212 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 213 | |
214 | __taskq_wait_id(tq, id); | |
215 | ||
b17edc10 | 216 | SEXIT; |
bcd68186 | 217 | |
218 | } | |
219 | EXPORT_SYMBOL(__taskq_wait); | |
220 | ||
221 | int | |
222 | __taskq_member(taskq_t *tq, void *t) | |
223 | { | |
2c02b71b PS |
224 | struct list_head *l; |
225 | taskq_thread_t *tqt; | |
b17edc10 | 226 | SENTRY; |
bcd68186 | 227 | |
228 | ASSERT(tq); | |
229 | ASSERT(t); | |
230 | ||
2c02b71b PS |
231 | list_for_each(l, &tq->tq_thread_list) { |
232 | tqt = list_entry(l, taskq_thread_t, tqt_thread_list); | |
233 | if (tqt->tqt_thread == (struct task_struct *)t) | |
234 | SRETURN(1); | |
235 | } | |
bcd68186 | 236 | |
b17edc10 | 237 | SRETURN(0); |
bcd68186 | 238 | } |
239 | EXPORT_SYMBOL(__taskq_member); | |
240 | ||
241 | taskqid_t | |
242 | __taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t flags) | |
243 | { | |
046a70c9 | 244 | taskq_ent_t *t; |
bcd68186 | 245 | taskqid_t rc = 0; |
b17edc10 | 246 | SENTRY; |
f1ca4da6 | 247 | |
937879f1 | 248 | ASSERT(tq); |
249 | ASSERT(func); | |
d05ec4b4 BB |
250 | |
251 | /* Solaris assumes TQ_SLEEP if not passed explicitly */ | |
252 | if (!(flags & (TQ_SLEEP | TQ_NOSLEEP))) | |
253 | flags |= TQ_SLEEP; | |
254 | ||
55abb092 BB |
255 | if (unlikely(in_atomic() && (flags & TQ_SLEEP))) |
256 | PANIC("May schedule while atomic: %s/0x%08x/%d\n", | |
257 | current->comm, preempt_count(), current->pid); | |
f1ca4da6 | 258 | |
749045bb | 259 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
f1ca4da6 | 260 | |
bcd68186 | 261 | /* Taskq being destroyed and all tasks drained */ |
262 | if (!(tq->tq_flags & TQ_ACTIVE)) | |
b17edc10 | 263 | SGOTO(out, rc = 0); |
f1ca4da6 | 264 | |
bcd68186 | 265 | /* Do not queue the task unless there is idle thread for it */ |
266 | ASSERT(tq->tq_nactive <= tq->tq_nthreads); | |
267 | if ((flags & TQ_NOQUEUE) && (tq->tq_nactive == tq->tq_nthreads)) | |
b17edc10 | 268 | SGOTO(out, rc = 0); |
bcd68186 | 269 | |
270 | if ((t = task_alloc(tq, flags)) == NULL) | |
b17edc10 | 271 | SGOTO(out, rc = 0); |
f1ca4da6 | 272 | |
046a70c9 | 273 | spin_lock(&t->tqent_lock); |
f0d8bb26 NB |
274 | |
275 | /* Queue to the priority list instead of the pending list */ | |
276 | if (flags & TQ_FRONT) | |
046a70c9 | 277 | list_add_tail(&t->tqent_list, &tq->tq_prio_list); |
f0d8bb26 | 278 | else |
046a70c9 | 279 | list_add_tail(&t->tqent_list, &tq->tq_pend_list); |
f0d8bb26 | 280 | |
046a70c9 | 281 | t->tqent_id = rc = tq->tq_next_id; |
bcd68186 | 282 | tq->tq_next_id++; |
046a70c9 PS |
283 | t->tqent_func = func; |
284 | t->tqent_arg = arg; | |
44217f7a PS |
285 | |
286 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
287 | ||
046a70c9 | 288 | spin_unlock(&t->tqent_lock); |
bcd68186 | 289 | out: |
749045bb | 290 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
ec2b4104 NB |
291 | if (rc > 0) |
292 | wake_up(&tq->tq_work_waitq); | |
293 | ||
b17edc10 | 294 | SRETURN(rc); |
f1ca4da6 | 295 | } |
f1b59d26 | 296 | EXPORT_SYMBOL(__taskq_dispatch); |
44217f7a PS |
297 | |
298 | void | |
299 | __taskq_dispatch_ent(taskq_t *tq, task_func_t func, void *arg, uint_t flags, | |
300 | taskq_ent_t *t) | |
301 | { | |
302 | SENTRY; | |
303 | ||
304 | ASSERT(tq); | |
305 | ASSERT(func); | |
306 | ASSERT(!(tq->tq_flags & TASKQ_DYNAMIC)); | |
307 | ||
308 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
309 | ||
310 | /* Taskq being destroyed and all tasks drained */ | |
311 | if (!(tq->tq_flags & TQ_ACTIVE)) { | |
312 | t->tqent_id = 0; | |
ec2b4104 | 313 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
44217f7a PS |
314 | goto out; |
315 | } | |
316 | ||
317 | spin_lock(&t->tqent_lock); | |
318 | ||
319 | /* | |
320 | * Mark it as a prealloc'd task. This is important | |
321 | * to ensure that we don't free it later. | |
322 | */ | |
323 | t->tqent_flags |= TQENT_FLAG_PREALLOC; | |
324 | ||
325 | /* Queue to the priority list instead of the pending list */ | |
326 | if (flags & TQ_FRONT) | |
327 | list_add_tail(&t->tqent_list, &tq->tq_prio_list); | |
328 | else | |
329 | list_add_tail(&t->tqent_list, &tq->tq_pend_list); | |
330 | ||
331 | t->tqent_id = tq->tq_next_id; | |
332 | tq->tq_next_id++; | |
333 | t->tqent_func = func; | |
334 | t->tqent_arg = arg; | |
335 | ||
336 | spin_unlock(&t->tqent_lock); | |
ec2b4104 | 337 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
44217f7a PS |
338 | |
339 | wake_up(&tq->tq_work_waitq); | |
340 | out: | |
44217f7a PS |
341 | SEXIT; |
342 | } | |
343 | EXPORT_SYMBOL(__taskq_dispatch_ent); | |
344 | ||
345 | int | |
346 | __taskq_empty_ent(taskq_ent_t *t) | |
347 | { | |
348 | return list_empty(&t->tqent_list); | |
349 | } | |
350 | EXPORT_SYMBOL(__taskq_empty_ent); | |
351 | ||
352 | void | |
353 | __taskq_init_ent(taskq_ent_t *t) | |
354 | { | |
355 | spin_lock_init(&t->tqent_lock); | |
356 | INIT_LIST_HEAD(&t->tqent_list); | |
357 | t->tqent_id = 0; | |
358 | t->tqent_func = NULL; | |
359 | t->tqent_arg = NULL; | |
360 | t->tqent_flags = 0; | |
361 | } | |
362 | EXPORT_SYMBOL(__taskq_init_ent); | |
363 | ||
82387586 BB |
364 | /* |
365 | * Returns the lowest incomplete taskqid_t. The taskqid_t may | |
f0d8bb26 NB |
366 | * be queued on the pending list, on the priority list, or on |
367 | * the work list currently being handled, but it is not 100% | |
368 | * complete yet. | |
82387586 | 369 | */ |
bcd68186 | 370 | static taskqid_t |
371 | taskq_lowest_id(taskq_t *tq) | |
372 | { | |
7257ec41 | 373 | taskqid_t lowest_id = tq->tq_next_id; |
046a70c9 | 374 | taskq_ent_t *t; |
2c02b71b | 375 | taskq_thread_t *tqt; |
b17edc10 | 376 | SENTRY; |
bcd68186 | 377 | |
378 | ASSERT(tq); | |
379 | ASSERT(spin_is_locked(&tq->tq_lock)); | |
380 | ||
82387586 | 381 | if (!list_empty(&tq->tq_pend_list)) { |
046a70c9 PS |
382 | t = list_entry(tq->tq_pend_list.next, taskq_ent_t, tqent_list); |
383 | lowest_id = MIN(lowest_id, t->tqent_id); | |
82387586 | 384 | } |
bcd68186 | 385 | |
f0d8bb26 | 386 | if (!list_empty(&tq->tq_prio_list)) { |
046a70c9 PS |
387 | t = list_entry(tq->tq_prio_list.next, taskq_ent_t, tqent_list); |
388 | lowest_id = MIN(lowest_id, t->tqent_id); | |
f0d8bb26 NB |
389 | } |
390 | ||
2c02b71b PS |
391 | if (!list_empty(&tq->tq_active_list)) { |
392 | tqt = list_entry(tq->tq_active_list.next, taskq_thread_t, | |
393 | tqt_active_list); | |
e7e5f78e PS |
394 | ASSERT(tqt->tqt_id != 0); |
395 | lowest_id = MIN(lowest_id, tqt->tqt_id); | |
82387586 | 396 | } |
bcd68186 | 397 | |
b17edc10 | 398 | SRETURN(lowest_id); |
bcd68186 | 399 | } |
400 | ||
f0d8bb26 NB |
401 | /* |
402 | * Insert a task into a list keeping the list sorted by increasing | |
403 | * taskqid. | |
404 | */ | |
405 | static void | |
2c02b71b | 406 | taskq_insert_in_order(taskq_t *tq, taskq_thread_t *tqt) |
f0d8bb26 | 407 | { |
2c02b71b | 408 | taskq_thread_t *w; |
f0d8bb26 NB |
409 | struct list_head *l; |
410 | ||
b17edc10 | 411 | SENTRY; |
f0d8bb26 | 412 | ASSERT(tq); |
2c02b71b | 413 | ASSERT(tqt); |
f0d8bb26 NB |
414 | ASSERT(spin_is_locked(&tq->tq_lock)); |
415 | ||
2c02b71b PS |
416 | list_for_each_prev(l, &tq->tq_active_list) { |
417 | w = list_entry(l, taskq_thread_t, tqt_active_list); | |
e7e5f78e | 418 | if (w->tqt_id < tqt->tqt_id) { |
2c02b71b | 419 | list_add(&tqt->tqt_active_list, l); |
f0d8bb26 NB |
420 | break; |
421 | } | |
422 | } | |
2c02b71b PS |
423 | if (l == &tq->tq_active_list) |
424 | list_add(&tqt->tqt_active_list, &tq->tq_active_list); | |
f0d8bb26 | 425 | |
b17edc10 | 426 | SEXIT; |
f0d8bb26 NB |
427 | } |
428 | ||
bcd68186 | 429 | static int |
430 | taskq_thread(void *args) | |
431 | { | |
432 | DECLARE_WAITQUEUE(wait, current); | |
433 | sigset_t blocked; | |
2c02b71b PS |
434 | taskq_thread_t *tqt = args; |
435 | taskq_t *tq; | |
046a70c9 | 436 | taskq_ent_t *t; |
f0d8bb26 | 437 | struct list_head *pend_list; |
b17edc10 | 438 | SENTRY; |
bcd68186 | 439 | |
2c02b71b PS |
440 | ASSERT(tqt); |
441 | tq = tqt->tqt_tq; | |
bcd68186 | 442 | current->flags |= PF_NOFREEZE; |
443 | ||
372c2572 BB |
444 | /* Disable the direct memory reclaim path */ |
445 | if (tq->tq_flags & TASKQ_NORECLAIM) | |
446 | current->flags |= PF_MEMALLOC; | |
447 | ||
bcd68186 | 448 | sigfillset(&blocked); |
449 | sigprocmask(SIG_BLOCK, &blocked, NULL); | |
450 | flush_signals(current); | |
451 | ||
749045bb | 452 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 453 | tq->tq_nthreads++; |
454 | wake_up(&tq->tq_wait_waitq); | |
455 | set_current_state(TASK_INTERRUPTIBLE); | |
456 | ||
457 | while (!kthread_should_stop()) { | |
458 | ||
f0d8bb26 NB |
459 | if (list_empty(&tq->tq_pend_list) && |
460 | list_empty(&tq->tq_prio_list)) { | |
749045bb | 461 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
ec2b4104 | 462 | add_wait_queue_exclusive(&tq->tq_work_waitq, &wait); |
bcd68186 | 463 | schedule(); |
ec2b4104 | 464 | remove_wait_queue(&tq->tq_work_waitq, &wait); |
749045bb | 465 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 466 | } else { |
467 | __set_current_state(TASK_RUNNING); | |
468 | } | |
469 | ||
f0d8bb26 NB |
470 | |
471 | if (!list_empty(&tq->tq_prio_list)) | |
472 | pend_list = &tq->tq_prio_list; | |
473 | else if (!list_empty(&tq->tq_pend_list)) | |
474 | pend_list = &tq->tq_pend_list; | |
475 | else | |
476 | pend_list = NULL; | |
477 | ||
478 | if (pend_list) { | |
046a70c9 PS |
479 | t = list_entry(pend_list->next, taskq_ent_t, tqent_list); |
480 | list_del_init(&t->tqent_list); | |
8f2503e0 | 481 | |
44217f7a PS |
482 | /* In order to support recursively dispatching a |
483 | * preallocated taskq_ent_t, tqent_id must be | |
484 | * stored prior to executing tqent_func. */ | |
e7e5f78e | 485 | tqt->tqt_id = t->tqent_id; |
8f2503e0 PS |
486 | |
487 | /* We must store a copy of the flags prior to | |
488 | * servicing the task (servicing a prealloc'd task | |
489 | * returns the ownership of the tqent back to | |
490 | * the caller of taskq_dispatch). Thus, | |
491 | * tqent_flags _may_ change within the call. */ | |
492 | tqt->tqt_flags = t->tqent_flags; | |
493 | ||
2c02b71b | 494 | taskq_insert_in_order(tq, tqt); |
bcd68186 | 495 | tq->tq_nactive++; |
749045bb | 496 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 497 | |
498 | /* Perform the requested task */ | |
046a70c9 | 499 | t->tqent_func(t->tqent_arg); |
bcd68186 | 500 | |
749045bb | 501 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 502 | tq->tq_nactive--; |
2c02b71b | 503 | list_del_init(&tqt->tqt_active_list); |
8f2503e0 PS |
504 | |
505 | /* For prealloc'd tasks, we don't free anything. */ | |
506 | if ((tq->tq_flags & TASKQ_DYNAMIC) || | |
507 | !(tqt->tqt_flags & TQENT_FLAG_PREALLOC)) | |
508 | task_done(tq, t); | |
bcd68186 | 509 | |
7257ec41 BB |
510 | /* When the current lowest outstanding taskqid is |
511 | * done calculate the new lowest outstanding id */ | |
e7e5f78e | 512 | if (tq->tq_lowest_id == tqt->tqt_id) { |
bcd68186 | 513 | tq->tq_lowest_id = taskq_lowest_id(tq); |
e7e5f78e | 514 | ASSERT3S(tq->tq_lowest_id, >, tqt->tqt_id); |
bcd68186 | 515 | } |
516 | ||
e7e5f78e | 517 | tqt->tqt_id = 0; |
8f2503e0 | 518 | tqt->tqt_flags = 0; |
bcd68186 | 519 | wake_up_all(&tq->tq_wait_waitq); |
520 | } | |
521 | ||
522 | set_current_state(TASK_INTERRUPTIBLE); | |
523 | ||
524 | } | |
525 | ||
526 | __set_current_state(TASK_RUNNING); | |
527 | tq->tq_nthreads--; | |
2c02b71b PS |
528 | list_del_init(&tqt->tqt_thread_list); |
529 | kmem_free(tqt, sizeof(taskq_thread_t)); | |
530 | ||
749045bb | 531 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 532 | |
b17edc10 | 533 | SRETURN(0); |
bcd68186 | 534 | } |
535 | ||
f1ca4da6 | 536 | taskq_t * |
537 | __taskq_create(const char *name, int nthreads, pri_t pri, | |
538 | int minalloc, int maxalloc, uint_t flags) | |
539 | { | |
bcd68186 | 540 | taskq_t *tq; |
2c02b71b | 541 | taskq_thread_t *tqt; |
bcd68186 | 542 | int rc = 0, i, j = 0; |
b17edc10 | 543 | SENTRY; |
bcd68186 | 544 | |
545 | ASSERT(name != NULL); | |
546 | ASSERT(pri <= maxclsyspri); | |
547 | ASSERT(minalloc >= 0); | |
548 | ASSERT(maxalloc <= INT_MAX); | |
549 | ASSERT(!(flags & (TASKQ_CPR_SAFE | TASKQ_DYNAMIC))); /* Unsupported */ | |
550 | ||
915404bd BB |
551 | /* Scale the number of threads using nthreads as a percentage */ |
552 | if (flags & TASKQ_THREADS_CPU_PCT) { | |
553 | ASSERT(nthreads <= 100); | |
554 | ASSERT(nthreads >= 0); | |
555 | nthreads = MIN(nthreads, 100); | |
556 | nthreads = MAX(nthreads, 0); | |
557 | nthreads = MAX((num_online_cpus() * nthreads) / 100, 1); | |
558 | } | |
559 | ||
bcd68186 | 560 | tq = kmem_alloc(sizeof(*tq), KM_SLEEP); |
561 | if (tq == NULL) | |
b17edc10 | 562 | SRETURN(NULL); |
bcd68186 | 563 | |
bcd68186 | 564 | spin_lock_init(&tq->tq_lock); |
749045bb | 565 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
2c02b71b PS |
566 | INIT_LIST_HEAD(&tq->tq_thread_list); |
567 | INIT_LIST_HEAD(&tq->tq_active_list); | |
bcd68186 | 568 | tq->tq_name = name; |
569 | tq->tq_nactive = 0; | |
570 | tq->tq_nthreads = 0; | |
571 | tq->tq_pri = pri; | |
572 | tq->tq_minalloc = minalloc; | |
573 | tq->tq_maxalloc = maxalloc; | |
574 | tq->tq_nalloc = 0; | |
575 | tq->tq_flags = (flags | TQ_ACTIVE); | |
576 | tq->tq_next_id = 1; | |
577 | tq->tq_lowest_id = 1; | |
578 | INIT_LIST_HEAD(&tq->tq_free_list); | |
bcd68186 | 579 | INIT_LIST_HEAD(&tq->tq_pend_list); |
f0d8bb26 | 580 | INIT_LIST_HEAD(&tq->tq_prio_list); |
bcd68186 | 581 | init_waitqueue_head(&tq->tq_work_waitq); |
582 | init_waitqueue_head(&tq->tq_wait_waitq); | |
583 | ||
584 | if (flags & TASKQ_PREPOPULATE) | |
585 | for (i = 0; i < minalloc; i++) | |
586 | task_done(tq, task_alloc(tq, TQ_SLEEP | TQ_NEW)); | |
6e605b6e | 587 | |
749045bb | 588 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
6e605b6e | 589 | |
2c02b71b PS |
590 | for (i = 0; i < nthreads; i++) { |
591 | tqt = kmem_alloc(sizeof(*tqt), KM_SLEEP); | |
592 | INIT_LIST_HEAD(&tqt->tqt_thread_list); | |
593 | INIT_LIST_HEAD(&tqt->tqt_active_list); | |
594 | tqt->tqt_tq = tq; | |
e7e5f78e | 595 | tqt->tqt_id = 0; |
2c02b71b PS |
596 | |
597 | tqt->tqt_thread = kthread_create(taskq_thread, tqt, | |
598 | "%s/%d", name, i); | |
599 | if (tqt->tqt_thread) { | |
600 | list_add(&tqt->tqt_thread_list, &tq->tq_thread_list); | |
601 | kthread_bind(tqt->tqt_thread, i % num_online_cpus()); | |
602 | set_user_nice(tqt->tqt_thread, PRIO_TO_NICE(pri)); | |
603 | wake_up_process(tqt->tqt_thread); | |
bcd68186 | 604 | j++; |
2c02b71b PS |
605 | } else { |
606 | kmem_free(tqt, sizeof(taskq_thread_t)); | |
607 | rc = 1; | |
608 | } | |
609 | } | |
bcd68186 | 610 | |
611 | /* Wait for all threads to be started before potential destroy */ | |
612 | wait_event(tq->tq_wait_waitq, tq->tq_nthreads == j); | |
613 | ||
614 | if (rc) { | |
615 | __taskq_destroy(tq); | |
616 | tq = NULL; | |
617 | } | |
618 | ||
b17edc10 | 619 | SRETURN(tq); |
f1ca4da6 | 620 | } |
f1b59d26 | 621 | EXPORT_SYMBOL(__taskq_create); |
b123971f | 622 | |
623 | void | |
624 | __taskq_destroy(taskq_t *tq) | |
625 | { | |
2c02b71b PS |
626 | struct task_struct *thread; |
627 | taskq_thread_t *tqt; | |
046a70c9 | 628 | taskq_ent_t *t; |
b17edc10 | 629 | SENTRY; |
b123971f | 630 | |
bcd68186 | 631 | ASSERT(tq); |
749045bb | 632 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 633 | tq->tq_flags &= ~TQ_ACTIVE; |
749045bb | 634 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 635 | |
636 | /* TQ_ACTIVE cleared prevents new tasks being added to pending */ | |
637 | __taskq_wait(tq); | |
638 | ||
749045bb | 639 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); |
bcd68186 | 640 | |
2c02b71b PS |
641 | /* |
642 | * Signal each thread to exit and block until it does. Each thread | |
643 | * is responsible for removing itself from the list and freeing its | |
644 | * taskq_thread_t. This allows for idle threads to opt to remove | |
645 | * themselves from the taskq. They can be recreated as needed. | |
646 | */ | |
647 | while (!list_empty(&tq->tq_thread_list)) { | |
648 | tqt = list_entry(tq->tq_thread_list.next, | |
649 | taskq_thread_t, tqt_thread_list); | |
650 | thread = tqt->tqt_thread; | |
651 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); | |
652 | ||
653 | kthread_stop(thread); | |
654 | ||
655 | spin_lock_irqsave(&tq->tq_lock, tq->tq_lock_flags); | |
656 | } | |
657 | ||
bcd68186 | 658 | while (!list_empty(&tq->tq_free_list)) { |
046a70c9 | 659 | t = list_entry(tq->tq_free_list.next, taskq_ent_t, tqent_list); |
44217f7a PS |
660 | |
661 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); | |
662 | ||
046a70c9 | 663 | list_del_init(&t->tqent_list); |
bcd68186 | 664 | task_free(tq, t); |
665 | } | |
666 | ||
667 | ASSERT(tq->tq_nthreads == 0); | |
668 | ASSERT(tq->tq_nalloc == 0); | |
2c02b71b PS |
669 | ASSERT(list_empty(&tq->tq_thread_list)); |
670 | ASSERT(list_empty(&tq->tq_active_list)); | |
bcd68186 | 671 | ASSERT(list_empty(&tq->tq_free_list)); |
bcd68186 | 672 | ASSERT(list_empty(&tq->tq_pend_list)); |
f0d8bb26 | 673 | ASSERT(list_empty(&tq->tq_prio_list)); |
bcd68186 | 674 | |
749045bb | 675 | spin_unlock_irqrestore(&tq->tq_lock, tq->tq_lock_flags); |
2c02b71b | 676 | |
bcd68186 | 677 | kmem_free(tq, sizeof(taskq_t)); |
678 | ||
b17edc10 | 679 | SEXIT; |
b123971f | 680 | } |
bcd68186 | 681 | EXPORT_SYMBOL(__taskq_destroy); |
e9cb2b4f BB |
682 | |
683 | int | |
684 | spl_taskq_init(void) | |
685 | { | |
b17edc10 | 686 | SENTRY; |
e9cb2b4f | 687 | |
f220894e BB |
688 | /* Solaris creates a dynamic taskq of up to 64 threads, however in |
689 | * a Linux environment 1 thread per-core is usually about right */ | |
690 | system_taskq = taskq_create("spl_system_taskq", num_online_cpus(), | |
691 | minclsyspri, 4, 512, TASKQ_PREPOPULATE); | |
e9cb2b4f | 692 | if (system_taskq == NULL) |
b17edc10 | 693 | SRETURN(1); |
e9cb2b4f | 694 | |
b17edc10 | 695 | SRETURN(0); |
e9cb2b4f BB |
696 | } |
697 | ||
698 | void | |
699 | spl_taskq_fini(void) | |
700 | { | |
b17edc10 | 701 | SENTRY; |
e9cb2b4f | 702 | taskq_destroy(system_taskq); |
b17edc10 | 703 | SEXIT; |
e9cb2b4f | 704 | } |