]> git.proxmox.com Git - mirror_zfs.git/blob - lib/libzpool/taskq.c
Switch from _Noreturn to __attribute__((noreturn))
[mirror_zfs.git] / lib / libzpool / taskq.c
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
9 * or http://www.opensolaris.org/os/licensing.
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 /*
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25 /*
26 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
27 * Copyright 2012 Garrett D'Amore <garrett@damore.org>. All rights reserved.
28 * Copyright (c) 2014 by Delphix. All rights reserved.
29 */
30
31 #include <sys/zfs_context.h>
32
33 int taskq_now;
34 taskq_t *system_taskq;
35 taskq_t *system_delay_taskq;
36
37 static pthread_key_t taskq_tsd;
38
39 #define TASKQ_ACTIVE 0x00010000
40
41 static taskq_ent_t *
42 task_alloc(taskq_t *tq, int tqflags)
43 {
44 taskq_ent_t *t;
45 int rv;
46
47 again: if ((t = tq->tq_freelist) != NULL && tq->tq_nalloc >= tq->tq_minalloc) {
48 ASSERT(!(t->tqent_flags & TQENT_FLAG_PREALLOC));
49 tq->tq_freelist = t->tqent_next;
50 } else {
51 if (tq->tq_nalloc >= tq->tq_maxalloc) {
52 if (!(tqflags & KM_SLEEP))
53 return (NULL);
54
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
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.
64 */
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 */
71 }
72 mutex_exit(&tq->tq_lock);
73
74 t = kmem_alloc(sizeof (taskq_ent_t), tqflags);
75
76 mutex_enter(&tq->tq_lock);
77 if (t != NULL) {
78 /* Make sure we start without any flags */
79 t->tqent_flags = 0;
80 tq->tq_nalloc++;
81 }
82 }
83 return (t);
84 }
85
86 static void
87 task_free(taskq_t *tq, taskq_ent_t *t)
88 {
89 if (tq->tq_nalloc <= tq->tq_minalloc) {
90 t->tqent_next = tq->tq_freelist;
91 tq->tq_freelist = t;
92 } else {
93 tq->tq_nalloc--;
94 mutex_exit(&tq->tq_lock);
95 kmem_free(t, sizeof (taskq_ent_t));
96 mutex_enter(&tq->tq_lock);
97 }
98
99 if (tq->tq_maxalloc_wait)
100 cv_signal(&tq->tq_maxalloc_cv);
101 }
102
103 taskqid_t
104 taskq_dispatch(taskq_t *tq, task_func_t func, void *arg, uint_t tqflags)
105 {
106 taskq_ent_t *t;
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 }
119 if (tqflags & TQ_FRONT) {
120 t->tqent_next = tq->tq_task.tqent_next;
121 t->tqent_prev = &tq->tq_task;
122 } else {
123 t->tqent_next = &tq->tq_task;
124 t->tqent_prev = tq->tq_task.tqent_prev;
125 }
126 t->tqent_next->tqent_prev = t;
127 t->tqent_prev->tqent_next = t;
128 t->tqent_func = func;
129 t->tqent_arg = arg;
130 t->tqent_flags = 0;
131 cv_signal(&tq->tq_dispatch_cv);
132 mutex_exit(&tq->tq_lock);
133 return (1);
134 }
135
136 taskqid_t
137 taskq_dispatch_delay(taskq_t *tq, task_func_t func, void *arg, uint_t tqflags,
138 clock_t expire_time)
139 {
140 (void) tq, (void) func, (void) arg, (void) tqflags, (void) expire_time;
141 return (0);
142 }
143
144 int
145 taskq_empty_ent(taskq_ent_t *t)
146 {
147 return (t->tqent_next == NULL);
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);
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
191 void
192 taskq_wait(taskq_t *tq)
193 {
194 mutex_enter(&tq->tq_lock);
195 while (tq->tq_task.tqent_next != &tq->tq_task || tq->tq_active != 0)
196 cv_wait(&tq->tq_wait_cv, &tq->tq_lock);
197 mutex_exit(&tq->tq_lock);
198 }
199
200 void
201 taskq_wait_id(taskq_t *tq, taskqid_t id)
202 {
203 (void) id;
204 taskq_wait(tq);
205 }
206
207 void
208 taskq_wait_outstanding(taskq_t *tq, taskqid_t id)
209 {
210 (void) id;
211 taskq_wait(tq);
212 }
213
214 static __attribute__((noreturn)) void
215 taskq_thread(void *arg)
216 {
217 taskq_t *tq = arg;
218 taskq_ent_t *t;
219 boolean_t prealloc;
220
221 VERIFY0(pthread_setspecific(taskq_tsd, tq));
222
223 mutex_enter(&tq->tq_lock);
224 while (tq->tq_flags & TASKQ_ACTIVE) {
225 if ((t = tq->tq_task.tqent_next) == &tq->tq_task) {
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 }
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;
237 mutex_exit(&tq->tq_lock);
238
239 rw_enter(&tq->tq_threadlock, RW_READER);
240 t->tqent_func(t->tqent_arg);
241 rw_exit(&tq->tq_threadlock);
242
243 mutex_enter(&tq->tq_lock);
244 if (!prealloc)
245 task_free(tq, t);
246 }
247 tq->tq_nthreads--;
248 cv_broadcast(&tq->tq_wait_cv);
249 mutex_exit(&tq->tq_lock);
250 thread_exit();
251 }
252
253 taskq_t *
254 taskq_create(const char *name, int nthreads, pri_t pri,
255 int minalloc, int maxalloc, uint_t flags)
256 {
257 (void) pri;
258 taskq_t *tq = kmem_zalloc(sizeof (taskq_t), KM_SLEEP);
259 int t;
260
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
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);
278 cv_init(&tq->tq_maxalloc_cv, NULL, CV_DEFAULT, NULL);
279 (void) strncpy(tq->tq_name, name, TASKQ_NAMELEN);
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;
285 tq->tq_task.tqent_next = &tq->tq_task;
286 tq->tq_task.tqent_prev = &tq->tq_task;
287 tq->tq_threadlist = kmem_alloc(nthreads * sizeof (kthread_t *),
288 KM_SLEEP);
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++)
298 VERIFY((tq->tq_threadlist[t] = thread_create(NULL, 0,
299 taskq_thread, tq, 0, &p0, TS_RUN, pri)) != NULL);
300
301 return (tq);
302 }
303
304 void
305 taskq_destroy(taskq_t *tq)
306 {
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);
322 task_free(tq, task_alloc(tq, KM_SLEEP));
323 }
324
325 mutex_exit(&tq->tq_lock);
326
327 kmem_free(tq->tq_threadlist, nthreads * sizeof (kthread_t *));
328
329 rw_destroy(&tq->tq_threadlock);
330 mutex_destroy(&tq->tq_lock);
331 cv_destroy(&tq->tq_dispatch_cv);
332 cv_destroy(&tq->tq_wait_cv);
333 cv_destroy(&tq->tq_maxalloc_cv);
334
335 kmem_free(tq, sizeof (taskq_t));
336 }
337
338 int
339 taskq_member(taskq_t *tq, kthread_t *t)
340 {
341 int i;
342
343 if (taskq_now)
344 return (1);
345
346 for (i = 0; i < tq->tq_nthreads; i++)
347 if (tq->tq_threadlist[i] == t)
348 return (1);
349
350 return (0);
351 }
352
353 taskq_t *
354 taskq_of_curthread(void)
355 {
356 return (pthread_getspecific(taskq_tsd));
357 }
358
359 int
360 taskq_cancel_id(taskq_t *tq, taskqid_t id)
361 {
362 (void) tq, (void) id;
363 return (ENOENT);
364 }
365
366 void
367 system_taskq_init(void)
368 {
369 VERIFY0(pthread_key_create(&taskq_tsd, NULL));
370 system_taskq = taskq_create("system_taskq", 64, maxclsyspri, 4, 512,
371 TASKQ_DYNAMIC | TASKQ_PREPOPULATE);
372 system_delay_taskq = taskq_create("delay_taskq", 4, maxclsyspri, 4,
373 512, TASKQ_DYNAMIC | TASKQ_PREPOPULATE);
374 }
375
376 void
377 system_taskq_fini(void)
378 {
379 taskq_destroy(system_taskq);
380 system_taskq = NULL; /* defensive */
381 taskq_destroy(system_delay_taskq);
382 system_delay_taskq = NULL;
383 VERIFY0(pthread_key_delete(taskq_tsd));
384 }