]>
git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - zfs/lib/libzpool/taskq.c
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.
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.
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]
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
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.
31 #include <sys/zfs_context.h>
34 taskq_t
*system_taskq
;
36 #define TASKQ_ACTIVE 0x00010000
39 task_alloc(taskq_t
*tq
, int tqflags
)
44 again
: if ((t
= tq
->tq_freelist
) != NULL
&& tq
->tq_nalloc
>= tq
->tq_minalloc
) {
45 ASSERT(!(t
->tqent_flags
& TQENT_FLAG_PREALLOC
));
46 tq
->tq_freelist
= t
->tqent_next
;
48 if (tq
->tq_nalloc
>= tq
->tq_maxalloc
) {
49 if (!(tqflags
& KM_SLEEP
))
53 * We don't want to exceed tq_maxalloc, but we can't
54 * wait for other tasks to complete (and thus free up
55 * task structures) without risking deadlock with
56 * the caller. So, we just delay for one second
57 * to throttle the allocation rate. If we have tasks
58 * complete before one second timeout expires then
59 * taskq_ent_free will signal us and we will
60 * immediately retry the allocation.
62 tq
->tq_maxalloc_wait
++;
63 rv
= cv_timedwait(&tq
->tq_maxalloc_cv
,
64 &tq
->tq_lock
, ddi_get_lbolt() + hz
);
65 tq
->tq_maxalloc_wait
--;
67 goto again
; /* signaled */
69 mutex_exit(&tq
->tq_lock
);
71 t
= kmem_alloc(sizeof (taskq_ent_t
), tqflags
);
73 mutex_enter(&tq
->tq_lock
);
75 /* Make sure we start without any flags */
84 task_free(taskq_t
*tq
, taskq_ent_t
*t
)
86 if (tq
->tq_nalloc
<= tq
->tq_minalloc
) {
87 t
->tqent_next
= tq
->tq_freelist
;
91 mutex_exit(&tq
->tq_lock
);
92 kmem_free(t
, sizeof (taskq_ent_t
));
93 mutex_enter(&tq
->tq_lock
);
96 if (tq
->tq_maxalloc_wait
)
97 cv_signal(&tq
->tq_maxalloc_cv
);
101 taskq_dispatch(taskq_t
*tq
, task_func_t func
, void *arg
, uint_t tqflags
)
110 mutex_enter(&tq
->tq_lock
);
111 ASSERT(tq
->tq_flags
& TASKQ_ACTIVE
);
112 if ((t
= task_alloc(tq
, tqflags
)) == NULL
) {
113 mutex_exit(&tq
->tq_lock
);
116 if (tqflags
& TQ_FRONT
) {
117 t
->tqent_next
= tq
->tq_task
.tqent_next
;
118 t
->tqent_prev
= &tq
->tq_task
;
120 t
->tqent_next
= &tq
->tq_task
;
121 t
->tqent_prev
= tq
->tq_task
.tqent_prev
;
123 t
->tqent_next
->tqent_prev
= t
;
124 t
->tqent_prev
->tqent_next
= t
;
125 t
->tqent_func
= func
;
128 cv_signal(&tq
->tq_dispatch_cv
);
129 mutex_exit(&tq
->tq_lock
);
134 taskq_dispatch_delay(taskq_t
*tq
, task_func_t func
, void *arg
, uint_t tqflags
,
141 taskq_empty_ent(taskq_ent_t
*t
)
143 return (t
->tqent_next
== NULL
);
147 taskq_init_ent(taskq_ent_t
*t
)
149 t
->tqent_next
= NULL
;
150 t
->tqent_prev
= NULL
;
151 t
->tqent_func
= NULL
;
157 taskq_dispatch_ent(taskq_t
*tq
, task_func_t func
, void *arg
, uint_t flags
,
160 ASSERT(func
!= NULL
);
163 * Mark it as a prealloc'd task. This is important
164 * to ensure that we don't free it later.
166 t
->tqent_flags
|= TQENT_FLAG_PREALLOC
;
168 * Enqueue the task to the underlying queue.
170 mutex_enter(&tq
->tq_lock
);
172 if (flags
& TQ_FRONT
) {
173 t
->tqent_next
= tq
->tq_task
.tqent_next
;
174 t
->tqent_prev
= &tq
->tq_task
;
176 t
->tqent_next
= &tq
->tq_task
;
177 t
->tqent_prev
= tq
->tq_task
.tqent_prev
;
179 t
->tqent_next
->tqent_prev
= t
;
180 t
->tqent_prev
->tqent_next
= t
;
181 t
->tqent_func
= func
;
183 cv_signal(&tq
->tq_dispatch_cv
);
184 mutex_exit(&tq
->tq_lock
);
188 taskq_wait(taskq_t
*tq
)
190 mutex_enter(&tq
->tq_lock
);
191 while (tq
->tq_task
.tqent_next
!= &tq
->tq_task
|| tq
->tq_active
!= 0)
192 cv_wait(&tq
->tq_wait_cv
, &tq
->tq_lock
);
193 mutex_exit(&tq
->tq_lock
);
197 taskq_wait_id(taskq_t
*tq
, taskqid_t id
)
203 taskq_wait_outstanding(taskq_t
*tq
, taskqid_t id
)
209 taskq_thread(void *arg
)
215 mutex_enter(&tq
->tq_lock
);
216 while (tq
->tq_flags
& TASKQ_ACTIVE
) {
217 if ((t
= tq
->tq_task
.tqent_next
) == &tq
->tq_task
) {
218 if (--tq
->tq_active
== 0)
219 cv_broadcast(&tq
->tq_wait_cv
);
220 cv_wait(&tq
->tq_dispatch_cv
, &tq
->tq_lock
);
224 t
->tqent_prev
->tqent_next
= t
->tqent_next
;
225 t
->tqent_next
->tqent_prev
= t
->tqent_prev
;
226 t
->tqent_next
= NULL
;
227 t
->tqent_prev
= NULL
;
228 prealloc
= t
->tqent_flags
& TQENT_FLAG_PREALLOC
;
229 mutex_exit(&tq
->tq_lock
);
231 rw_enter(&tq
->tq_threadlock
, RW_READER
);
232 t
->tqent_func(t
->tqent_arg
);
233 rw_exit(&tq
->tq_threadlock
);
235 mutex_enter(&tq
->tq_lock
);
240 cv_broadcast(&tq
->tq_wait_cv
);
241 mutex_exit(&tq
->tq_lock
);
247 taskq_create(const char *name
, int nthreads
, pri_t pri
,
248 int minalloc
, int maxalloc
, uint_t flags
)
250 taskq_t
*tq
= kmem_zalloc(sizeof (taskq_t
), KM_SLEEP
);
253 if (flags
& TASKQ_THREADS_CPU_PCT
) {
255 ASSERT3S(nthreads
, >=, 0);
256 ASSERT3S(nthreads
, <=, 100);
257 pct
= MIN(nthreads
, 100);
260 nthreads
= (sysconf(_SC_NPROCESSORS_ONLN
) * pct
) / 100;
261 nthreads
= MAX(nthreads
, 1); /* need at least 1 thread */
263 ASSERT3S(nthreads
, >=, 1);
266 rw_init(&tq
->tq_threadlock
, NULL
, RW_DEFAULT
, NULL
);
267 mutex_init(&tq
->tq_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
268 cv_init(&tq
->tq_dispatch_cv
, NULL
, CV_DEFAULT
, NULL
);
269 cv_init(&tq
->tq_wait_cv
, NULL
, CV_DEFAULT
, NULL
);
270 cv_init(&tq
->tq_maxalloc_cv
, NULL
, CV_DEFAULT
, NULL
);
271 (void) strncpy(tq
->tq_name
, name
, TASKQ_NAMELEN
);
272 tq
->tq_flags
= flags
| TASKQ_ACTIVE
;
273 tq
->tq_active
= nthreads
;
274 tq
->tq_nthreads
= nthreads
;
275 tq
->tq_minalloc
= minalloc
;
276 tq
->tq_maxalloc
= maxalloc
;
277 tq
->tq_task
.tqent_next
= &tq
->tq_task
;
278 tq
->tq_task
.tqent_prev
= &tq
->tq_task
;
279 tq
->tq_threadlist
= kmem_alloc(nthreads
* sizeof (kthread_t
*),
282 if (flags
& TASKQ_PREPOPULATE
) {
283 mutex_enter(&tq
->tq_lock
);
284 while (minalloc
-- > 0)
285 task_free(tq
, task_alloc(tq
, KM_SLEEP
));
286 mutex_exit(&tq
->tq_lock
);
289 for (t
= 0; t
< nthreads
; t
++)
290 VERIFY((tq
->tq_threadlist
[t
] = thread_create(NULL
, 0,
291 taskq_thread
, tq
, 0, &p0
, TS_RUN
, pri
)) != NULL
);
297 taskq_destroy(taskq_t
*tq
)
299 int nthreads
= tq
->tq_nthreads
;
303 mutex_enter(&tq
->tq_lock
);
305 tq
->tq_flags
&= ~TASKQ_ACTIVE
;
306 cv_broadcast(&tq
->tq_dispatch_cv
);
308 while (tq
->tq_nthreads
!= 0)
309 cv_wait(&tq
->tq_wait_cv
, &tq
->tq_lock
);
312 while (tq
->tq_nalloc
!= 0) {
313 ASSERT(tq
->tq_freelist
!= NULL
);
314 task_free(tq
, task_alloc(tq
, KM_SLEEP
));
317 mutex_exit(&tq
->tq_lock
);
319 kmem_free(tq
->tq_threadlist
, nthreads
* sizeof (kthread_t
*));
321 rw_destroy(&tq
->tq_threadlock
);
322 mutex_destroy(&tq
->tq_lock
);
323 cv_destroy(&tq
->tq_dispatch_cv
);
324 cv_destroy(&tq
->tq_wait_cv
);
325 cv_destroy(&tq
->tq_maxalloc_cv
);
327 kmem_free(tq
, sizeof (taskq_t
));
331 taskq_member(taskq_t
*tq
, kthread_t
*t
)
338 for (i
= 0; i
< tq
->tq_nthreads
; i
++)
339 if (tq
->tq_threadlist
[i
] == t
)
346 taskq_cancel_id(taskq_t
*tq
, taskqid_t id
)
352 system_taskq_init(void)
354 system_taskq
= taskq_create("system_taskq", 64, maxclsyspri
, 4, 512,
355 TASKQ_DYNAMIC
| TASKQ_PREPOPULATE
);
359 system_taskq_fini(void)
361 taskq_destroy(system_taskq
);
362 system_taskq
= NULL
; /* defensive */