]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
1d3ba0bf | 9 | * or https://opensource.org/licenses/CDDL-1.0. |
34dc7c2f BB |
10 | * See the License for the specific language governing permissions |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright 2010 Sun Microsystems, Inc. All rights reserved. |
34dc7c2f BB |
23 | * Use is subject to license terms. |
24 | */ | |
a38718a6 GA |
25 | /* |
26 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. | |
844793c3 | 27 | * Copyright 2012 Garrett D'Amore <garrett@damore.org>. All rights reserved. |
98b25418 | 28 | * Copyright (c) 2014 by Delphix. All rights reserved. |
a38718a6 | 29 | */ |
34dc7c2f | 30 | |
34dc7c2f BB |
31 | #include <sys/zfs_context.h> |
32 | ||
33 | int taskq_now; | |
b128c09f | 34 | taskq_t *system_taskq; |
57ddcda1 | 35 | taskq_t *system_delay_taskq; |
34dc7c2f | 36 | |
b3212d2f MA |
37 | static pthread_key_t taskq_tsd; |
38 | ||
34dc7c2f | 39 | #define TASKQ_ACTIVE 0x00010000 |
34dc7c2f | 40 | |
a38718a6 | 41 | static taskq_ent_t * |
34dc7c2f BB |
42 | task_alloc(taskq_t *tq, int tqflags) |
43 | { | |
a38718a6 | 44 | taskq_ent_t *t; |
428870ff | 45 | int rv; |
34dc7c2f | 46 | |
428870ff | 47 | again: if ((t = tq->tq_freelist) != NULL && tq->tq_nalloc >= tq->tq_minalloc) { |
a38718a6 GA |
48 | ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC)); |
49 | tq->tq_freelist = t->tqent_next; | |
34dc7c2f | 50 | } else { |
34dc7c2f | 51 | if (tq->tq_nalloc >= tq->tq_maxalloc) { |
428870ff | 52 | if (!(tqflags & KM_SLEEP)) |
34dc7c2f | 53 | return (NULL); |
428870ff | 54 | |
34dc7c2f BB |
55 | /* |
56 | * We don't want to exceed tq_maxalloc, but we can't | |
57 | * wait for other tasks to complete (and thus free up | |
58 | * task structures) without risking deadlock with | |
59 | * the caller. So, we just delay for one second | |
428870ff BB |
60 | * to throttle the allocation rate. If we have tasks |
61 | * complete before one second timeout expires then | |
62 | * taskq_ent_free will signal us and we will | |
63 | * immediately retry the allocation. | |
34dc7c2f | 64 | */ |
428870ff BB |
65 | tq->tq_maxalloc_wait++; |
66 | rv = cv_timedwait(&tq->tq_maxalloc_cv, | |
67 | &tq->tq_lock, ddi_get_lbolt() + hz); | |
68 | tq->tq_maxalloc_wait--; | |
69 | if (rv > 0) | |
70 | goto again; /* signaled */ | |
34dc7c2f | 71 | } |
428870ff BB |
72 | mutex_exit(&tq->tq_lock); |
73 | ||
a38718a6 | 74 | t = kmem_alloc(sizeof (taskq_ent_t), tqflags); |
428870ff | 75 | |
34dc7c2f | 76 | mutex_enter(&tq->tq_lock); |
a38718a6 GA |
77 | if (t != NULL) { |
78 | /* Make sure we start without any flags */ | |
79 | t->tqent_flags = 0; | |
34dc7c2f | 80 | tq->tq_nalloc++; |
a38718a6 | 81 | } |
34dc7c2f BB |
82 | } |
83 | return (t); | |
84 | } | |
85 | ||
86 | static void | |
a38718a6 | 87 | task_free(taskq_t *tq, taskq_ent_t *t) |
34dc7c2f BB |
88 | { |
89 | if (tq->tq_nalloc <= tq->tq_minalloc) { | |
a38718a6 | 90 | t->tqent_next = tq->tq_freelist; |
34dc7c2f BB |
91 | tq->tq_freelist = t; |
92 | } else { | |
93 | tq->tq_nalloc--; | |
94 | mutex_exit(&tq->tq_lock); | |
a38718a6 | 95 | kmem_free(t, sizeof (taskq_ent_t)); |
34dc7c2f BB |
96 | mutex_enter(&tq->tq_lock); |
97 | } | |
428870ff BB |
98 | |
99 | if (tq->tq_maxalloc_wait) | |
100 | cv_signal(&tq->tq_maxalloc_cv); | |
34dc7c2f BB |
101 | } |
102 | ||
103 | taskqid_t | |
104 | taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t tqflags) | |
105 | { | |
a38718a6 | 106 | taskq_ent_t *t; |
34dc7c2f BB |
107 | |
108 | if (taskq_now) { | |
109 | func(arg); | |
110 | return (1); | |
111 | } | |
112 | ||
113 | mutex_enter(&tq->tq_lock); | |
114 | ASSERT(tq->tq_flags & TASKQ_ACTIVE); | |
115 | if ((t = task_alloc(tq, tqflags)) == NULL) { | |
116 | mutex_exit(&tq->tq_lock); | |
117 | return (0); | |
118 | } | |
428870ff | 119 | if (tqflags & TQ_FRONT) { |
a38718a6 GA |
120 | t->tqent_next = tq->tq_task.tqent_next; |
121 | t->tqent_prev = &tq->tq_task; | |
428870ff | 122 | } else { |
a38718a6 GA |
123 | t->tqent_next = &tq->tq_task; |
124 | t->tqent_prev = tq->tq_task.tqent_prev; | |
428870ff | 125 | } |
a38718a6 GA |
126 | t->tqent_next->tqent_prev = t; |
127 | t->tqent_prev->tqent_next = t; | |
128 | t->tqent_func = func; | |
129 | t->tqent_arg = arg; | |
844793c3 | 130 | t->tqent_flags = 0; |
34dc7c2f BB |
131 | cv_signal(&tq->tq_dispatch_cv); |
132 | mutex_exit(&tq->tq_lock); | |
133 | return (1); | |
134 | } | |
135 | ||
cc92e9d0 | 136 | taskqid_t |
56050599 | 137 | taskq_dispatch_delay(taskq_t *tq, task_func_t func, void *arg, uint_t tqflags, |
cc92e9d0 GW |
138 | clock_t expire_time) |
139 | { | |
56050599 | 140 | (void) tq, (void) func, (void) arg, (void) tqflags, (void) expire_time; |
cc92e9d0 GW |
141 | return (0); |
142 | } | |
143 | ||
a38718a6 GA |
144 | int |
145 | taskq_empty_ent(taskq_ent_t *t) | |
146 | { | |
d1d7e268 | 147 | return (t->tqent_next == NULL); |
a38718a6 GA |
148 | } |
149 | ||
150 | void | |
151 | taskq_init_ent(taskq_ent_t *t) | |
152 | { | |
153 | t->tqent_next = NULL; | |
154 | t->tqent_prev = NULL; | |
155 | t->tqent_func = NULL; | |
156 | t->tqent_arg = NULL; | |
157 | t->tqent_flags = 0; | |
158 | } | |
159 | ||
160 | void | |
161 | taskq_dispatch_ent(taskq_t *tq, task_func_t func, void *arg, uint_t flags, | |
162 | taskq_ent_t *t) | |
163 | { | |
164 | ASSERT(func != NULL); | |
a38718a6 GA |
165 | |
166 | /* | |
167 | * Mark it as a prealloc'd task. This is important | |
168 | * to ensure that we don't free it later. | |
169 | */ | |
170 | t->tqent_flags |= TQENT_FLAG_PREALLOC; | |
171 | /* | |
172 | * Enqueue the task to the underlying queue. | |
173 | */ | |
174 | mutex_enter(&tq->tq_lock); | |
175 | ||
176 | if (flags & TQ_FRONT) { | |
177 | t->tqent_next = tq->tq_task.tqent_next; | |
178 | t->tqent_prev = &tq->tq_task; | |
179 | } else { | |
180 | t->tqent_next = &tq->tq_task; | |
181 | t->tqent_prev = tq->tq_task.tqent_prev; | |
182 | } | |
183 | t->tqent_next->tqent_prev = t; | |
184 | t->tqent_prev->tqent_next = t; | |
185 | t->tqent_func = func; | |
186 | t->tqent_arg = arg; | |
187 | cv_signal(&tq->tq_dispatch_cv); | |
188 | mutex_exit(&tq->tq_lock); | |
189 | } | |
190 | ||
34dc7c2f BB |
191 | void |
192 | taskq_wait(taskq_t *tq) | |
193 | { | |
194 | mutex_enter(&tq->tq_lock); | |
a38718a6 | 195 | while (tq->tq_task.tqent_next != &tq->tq_task || tq->tq_active != 0) |
34dc7c2f BB |
196 | cv_wait(&tq->tq_wait_cv, &tq->tq_lock); |
197 | mutex_exit(&tq->tq_lock); | |
198 | } | |
199 | ||
044baf00 BB |
200 | void |
201 | taskq_wait_id(taskq_t *tq, taskqid_t id) | |
202 | { | |
56050599 | 203 | (void) id; |
044baf00 BB |
204 | taskq_wait(tq); |
205 | } | |
206 | ||
4f34bd97 BB |
207 | void |
208 | taskq_wait_outstanding(taskq_t *tq, taskqid_t id) | |
209 | { | |
56050599 | 210 | (void) id; |
4f34bd97 BB |
211 | taskq_wait(tq); |
212 | } | |
213 | ||
460748d4 | 214 | static __attribute__((noreturn)) void |
34dc7c2f BB |
215 | taskq_thread(void *arg) |
216 | { | |
217 | taskq_t *tq = arg; | |
a38718a6 GA |
218 | taskq_ent_t *t; |
219 | boolean_t prealloc; | |
34dc7c2f | 220 | |
b3212d2f MA |
221 | VERIFY0(pthread_setspecific(taskq_tsd, tq)); |
222 | ||
34dc7c2f BB |
223 | mutex_enter(&tq->tq_lock); |
224 | while (tq->tq_flags & TASKQ_ACTIVE) { | |
a38718a6 | 225 | if ((t = tq->tq_task.tqent_next) == &tq->tq_task) { |
34dc7c2f BB |
226 | if (--tq->tq_active == 0) |
227 | cv_broadcast(&tq->tq_wait_cv); | |
228 | cv_wait(&tq->tq_dispatch_cv, &tq->tq_lock); | |
229 | tq->tq_active++; | |
230 | continue; | |
231 | } | |
a38718a6 GA |
232 | t->tqent_prev->tqent_next = t->tqent_next; |
233 | t->tqent_next->tqent_prev = t->tqent_prev; | |
234 | t->tqent_next = NULL; | |
235 | t->tqent_prev = NULL; | |
236 | prealloc = t->tqent_flags & TQENT_FLAG_PREALLOC; | |
34dc7c2f BB |
237 | mutex_exit(&tq->tq_lock); |
238 | ||
239 | rw_enter(&tq->tq_threadlock, RW_READER); | |
a38718a6 | 240 | t->tqent_func(t->tqent_arg); |
34dc7c2f BB |
241 | rw_exit(&tq->tq_threadlock); |
242 | ||
243 | mutex_enter(&tq->tq_lock); | |
a38718a6 GA |
244 | if (!prealloc) |
245 | task_free(tq, t); | |
34dc7c2f BB |
246 | } |
247 | tq->tq_nthreads--; | |
248 | cv_broadcast(&tq->tq_wait_cv); | |
249 | mutex_exit(&tq->tq_lock); | |
1e33ac1e | 250 | thread_exit(); |
34dc7c2f BB |
251 | } |
252 | ||
34dc7c2f BB |
253 | taskq_t * |
254 | taskq_create(const char *name, int nthreads, pri_t pri, | |
4ea3f864 | 255 | int minalloc, int maxalloc, uint_t flags) |
34dc7c2f | 256 | { |
56050599 | 257 | (void) pri; |
34dc7c2f BB |
258 | taskq_t *tq = kmem_zalloc(sizeof (taskq_t), KM_SLEEP); |
259 | int t; | |
260 | ||
9babb374 BB |
261 | if (flags & TASKQ_THREADS_CPU_PCT) { |
262 | int pct; | |
263 | ASSERT3S(nthreads, >=, 0); | |
264 | ASSERT3S(nthreads, <=, 100); | |
265 | pct = MIN(nthreads, 100); | |
266 | pct = MAX(pct, 0); | |
267 | ||
268 | nthreads = (sysconf(_SC_NPROCESSORS_ONLN) * pct) / 100; | |
269 | nthreads = MAX(nthreads, 1); /* need at least 1 thread */ | |
270 | } else { | |
271 | ASSERT3S(nthreads, >=, 1); | |
272 | } | |
273 | ||
34dc7c2f BB |
274 | rw_init(&tq->tq_threadlock, NULL, RW_DEFAULT, NULL); |
275 | mutex_init(&tq->tq_lock, NULL, MUTEX_DEFAULT, NULL); | |
276 | cv_init(&tq->tq_dispatch_cv, NULL, CV_DEFAULT, NULL); | |
277 | cv_init(&tq->tq_wait_cv, NULL, CV_DEFAULT, NULL); | |
428870ff | 278 | cv_init(&tq->tq_maxalloc_cv, NULL, CV_DEFAULT, NULL); |
7584fbe8 | 279 | (void) strlcpy(tq->tq_name, name, sizeof (tq->tq_name)); |
34dc7c2f BB |
280 | tq->tq_flags = flags | TASKQ_ACTIVE; |
281 | tq->tq_active = nthreads; | |
282 | tq->tq_nthreads = nthreads; | |
283 | tq->tq_minalloc = minalloc; | |
284 | tq->tq_maxalloc = maxalloc; | |
a38718a6 GA |
285 | tq->tq_task.tqent_next = &tq->tq_task; |
286 | tq->tq_task.tqent_prev = &tq->tq_task; | |
d1d7e268 MK |
287 | tq->tq_threadlist = kmem_alloc(nthreads * sizeof (kthread_t *), |
288 | KM_SLEEP); | |
34dc7c2f BB |
289 | |
290 | if (flags & TASKQ_PREPOPULATE) { | |
291 | mutex_enter(&tq->tq_lock); | |
292 | while (minalloc-- > 0) | |
293 | task_free(tq, task_alloc(tq, KM_SLEEP)); | |
294 | mutex_exit(&tq->tq_lock); | |
295 | } | |
296 | ||
297 | for (t = 0; t < nthreads; t++) | |
1e33ac1e | 298 | VERIFY((tq->tq_threadlist[t] = thread_create(NULL, 0, |
498056ab | 299 | taskq_thread, tq, 0, &p0, TS_RUN, pri)) != NULL); |
34dc7c2f BB |
300 | |
301 | return (tq); | |
302 | } | |
303 | ||
304 | void | |
305 | taskq_destroy(taskq_t *tq) | |
306 | { | |
34dc7c2f BB |
307 | int nthreads = tq->tq_nthreads; |
308 | ||
309 | taskq_wait(tq); | |
310 | ||
311 | mutex_enter(&tq->tq_lock); | |
312 | ||
313 | tq->tq_flags &= ~TASKQ_ACTIVE; | |
314 | cv_broadcast(&tq->tq_dispatch_cv); | |
315 | ||
316 | while (tq->tq_nthreads != 0) | |
317 | cv_wait(&tq->tq_wait_cv, &tq->tq_lock); | |
318 | ||
319 | tq->tq_minalloc = 0; | |
320 | while (tq->tq_nalloc != 0) { | |
321 | ASSERT(tq->tq_freelist != NULL); | |
d62bafee RY |
322 | taskq_ent_t *tqent_nexttq = tq->tq_freelist->tqent_next; |
323 | task_free(tq, tq->tq_freelist); | |
324 | tq->tq_freelist = tqent_nexttq; | |
34dc7c2f BB |
325 | } |
326 | ||
327 | mutex_exit(&tq->tq_lock); | |
328 | ||
1e33ac1e | 329 | kmem_free(tq->tq_threadlist, nthreads * sizeof (kthread_t *)); |
34dc7c2f BB |
330 | |
331 | rw_destroy(&tq->tq_threadlock); | |
332 | mutex_destroy(&tq->tq_lock); | |
333 | cv_destroy(&tq->tq_dispatch_cv); | |
334 | cv_destroy(&tq->tq_wait_cv); | |
428870ff | 335 | cv_destroy(&tq->tq_maxalloc_cv); |
34dc7c2f BB |
336 | |
337 | kmem_free(tq, sizeof (taskq_t)); | |
338 | } | |
339 | ||
3bd4df38 EN |
340 | /* |
341 | * Create a taskq with a specified number of pool threads. Allocate | |
342 | * and return an array of nthreads kthread_t pointers, one for each | |
343 | * thread in the pool. The array is not ordered and must be freed | |
344 | * by the caller. | |
345 | */ | |
346 | taskq_t * | |
347 | taskq_create_synced(const char *name, int nthreads, pri_t pri, | |
348 | int minalloc, int maxalloc, uint_t flags, kthread_t ***ktpp) | |
349 | { | |
350 | taskq_t *tq; | |
351 | kthread_t **kthreads = kmem_zalloc(sizeof (*kthreads) * nthreads, | |
352 | KM_SLEEP); | |
353 | ||
354 | (void) pri; (void) minalloc; (void) maxalloc; | |
355 | ||
356 | flags &= ~(TASKQ_DYNAMIC | TASKQ_THREADS_CPU_PCT | TASKQ_DC_BATCH); | |
357 | ||
358 | tq = taskq_create(name, nthreads, minclsyspri, nthreads, INT_MAX, | |
359 | flags | TASKQ_PREPOPULATE); | |
360 | VERIFY(tq != NULL); | |
361 | VERIFY(tq->tq_nthreads == nthreads); | |
362 | ||
363 | for (int i = 0; i < nthreads; i++) { | |
364 | kthreads[i] = tq->tq_threadlist[i]; | |
365 | } | |
366 | *ktpp = kthreads; | |
367 | return (tq); | |
368 | } | |
369 | ||
34dc7c2f | 370 | int |
1e33ac1e | 371 | taskq_member(taskq_t *tq, kthread_t *t) |
34dc7c2f BB |
372 | { |
373 | int i; | |
374 | ||
375 | if (taskq_now) | |
376 | return (1); | |
377 | ||
378 | for (i = 0; i < tq->tq_nthreads; i++) | |
1e33ac1e | 379 | if (tq->tq_threadlist[i] == t) |
34dc7c2f BB |
380 | return (1); |
381 | ||
382 | return (0); | |
383 | } | |
b128c09f | 384 | |
b3212d2f MA |
385 | taskq_t * |
386 | taskq_of_curthread(void) | |
387 | { | |
388 | return (pthread_getspecific(taskq_tsd)); | |
389 | } | |
390 | ||
cc92e9d0 GW |
391 | int |
392 | taskq_cancel_id(taskq_t *tq, taskqid_t id) | |
393 | { | |
56050599 | 394 | (void) tq, (void) id; |
cc92e9d0 GW |
395 | return (ENOENT); |
396 | } | |
397 | ||
b128c09f BB |
398 | void |
399 | system_taskq_init(void) | |
400 | { | |
b3212d2f | 401 | VERIFY0(pthread_key_create(&taskq_tsd, NULL)); |
1229323d | 402 | system_taskq = taskq_create("system_taskq", 64, maxclsyspri, 4, 512, |
b128c09f | 403 | TASKQ_DYNAMIC | TASKQ_PREPOPULATE); |
57ddcda1 CC |
404 | system_delay_taskq = taskq_create("delay_taskq", 4, maxclsyspri, 4, |
405 | 512, TASKQ_DYNAMIC | TASKQ_PREPOPULATE); | |
b128c09f | 406 | } |
428870ff BB |
407 | |
408 | void | |
409 | system_taskq_fini(void) | |
410 | { | |
411 | taskq_destroy(system_taskq); | |
412 | system_taskq = NULL; /* defensive */ | |
57ddcda1 CC |
413 | taskq_destroy(system_delay_taskq); |
414 | system_delay_taskq = NULL; | |
b3212d2f | 415 | VERIFY0(pthread_key_delete(taskq_tsd)); |
428870ff | 416 | } |