2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kthread.h>
20 #include <linux/list.h>
21 #include <linux/spinlock.h>
22 #include <linux/freezer.h>
23 #include "async-thread.h"
25 #define WORK_QUEUED_BIT 0
26 #define WORK_DONE_BIT 1
27 #define WORK_ORDER_DONE_BIT 2
30 * container for the kthread task pointer and the list of pending work
31 * One of these is allocated per thread.
33 struct btrfs_worker_thread
{
34 /* pool we belong to */
35 struct btrfs_workers
*workers
;
37 /* list of struct btrfs_work that are waiting for service */
38 struct list_head pending
;
40 /* list of worker threads from struct btrfs_workers */
41 struct list_head worker_list
;
44 struct task_struct
*task
;
46 /* number of things on the pending list */
49 unsigned long sequence
;
51 /* protects the pending list. */
54 /* set to non-zero when this thread is already awake and kicking */
57 /* are we currently idle */
62 * helper function to move a thread onto the idle list after it
63 * has finished some requests.
65 static void check_idle_worker(struct btrfs_worker_thread
*worker
)
67 if (!worker
->idle
&& atomic_read(&worker
->num_pending
) <
68 worker
->workers
->idle_thresh
/ 2) {
70 spin_lock_irqsave(&worker
->workers
->lock
, flags
);
72 list_move(&worker
->worker_list
, &worker
->workers
->idle_list
);
73 spin_unlock_irqrestore(&worker
->workers
->lock
, flags
);
78 * helper function to move a thread off the idle list after new
79 * pending work is added.
81 static void check_busy_worker(struct btrfs_worker_thread
*worker
)
83 if (worker
->idle
&& atomic_read(&worker
->num_pending
) >=
84 worker
->workers
->idle_thresh
) {
86 spin_lock_irqsave(&worker
->workers
->lock
, flags
);
88 list_move_tail(&worker
->worker_list
,
89 &worker
->workers
->worker_list
);
90 spin_unlock_irqrestore(&worker
->workers
->lock
, flags
);
94 static noinline
int run_ordered_completions(struct btrfs_workers
*workers
,
95 struct btrfs_work
*work
)
99 if (!workers
->ordered
)
102 set_bit(WORK_DONE_BIT
, &work
->flags
);
104 spin_lock_irqsave(&workers
->lock
, flags
);
106 while (!list_empty(&workers
->order_list
)) {
107 work
= list_entry(workers
->order_list
.next
,
108 struct btrfs_work
, order_list
);
110 if (!test_bit(WORK_DONE_BIT
, &work
->flags
))
113 /* we are going to call the ordered done function, but
114 * we leave the work item on the list as a barrier so
115 * that later work items that are done don't have their
116 * functions called before this one returns
118 if (test_and_set_bit(WORK_ORDER_DONE_BIT
, &work
->flags
))
121 spin_unlock_irqrestore(&workers
->lock
, flags
);
123 work
->ordered_func(work
);
125 /* now take the lock again and call the freeing code */
126 spin_lock_irqsave(&workers
->lock
, flags
);
127 list_del(&work
->order_list
);
128 work
->ordered_free(work
);
131 spin_unlock_irqrestore(&workers
->lock
, flags
);
136 * main loop for servicing work items
138 static int worker_loop(void *arg
)
140 struct btrfs_worker_thread
*worker
= arg
;
141 struct list_head
*cur
;
142 struct btrfs_work
*work
;
144 spin_lock_irq(&worker
->lock
);
146 while (!list_empty(&worker
->pending
)) {
147 cur
= worker
->pending
.next
;
148 work
= list_entry(cur
, struct btrfs_work
, list
);
149 list_del(&work
->list
);
150 clear_bit(WORK_QUEUED_BIT
, &work
->flags
);
152 work
->worker
= worker
;
153 spin_unlock_irq(&worker
->lock
);
157 atomic_dec(&worker
->num_pending
);
159 * unless this is an ordered work queue,
160 * 'work' was probably freed by func above.
162 run_ordered_completions(worker
->workers
, work
);
164 spin_lock_irq(&worker
->lock
);
165 check_idle_worker(worker
);
168 if (freezing(current
)) {
170 spin_unlock_irq(&worker
->lock
);
173 spin_unlock_irq(&worker
->lock
);
174 if (!kthread_should_stop()) {
177 * we've dropped the lock, did someone else
181 if (!list_empty(&worker
->pending
))
185 * this short schedule allows more work to
186 * come in without the queue functions
187 * needing to go through wake_up_process()
189 * worker->working is still 1, so nobody
190 * is going to try and wake us up
194 if (!list_empty(&worker
->pending
))
197 if (kthread_should_stop())
200 /* still no more work?, sleep for real */
201 spin_lock_irq(&worker
->lock
);
202 set_current_state(TASK_INTERRUPTIBLE
);
203 if (!list_empty(&worker
->pending
))
207 * this makes sure we get a wakeup when someone
208 * adds something new to the queue
211 spin_unlock_irq(&worker
->lock
);
213 if (!kthread_should_stop())
216 __set_current_state(TASK_RUNNING
);
218 } while (!kthread_should_stop());
223 * this will wait for all the worker threads to shutdown
225 int btrfs_stop_workers(struct btrfs_workers
*workers
)
227 struct list_head
*cur
;
228 struct btrfs_worker_thread
*worker
;
230 list_splice_init(&workers
->idle_list
, &workers
->worker_list
);
231 while (!list_empty(&workers
->worker_list
)) {
232 cur
= workers
->worker_list
.next
;
233 worker
= list_entry(cur
, struct btrfs_worker_thread
,
235 kthread_stop(worker
->task
);
236 list_del(&worker
->worker_list
);
243 * simple init on struct btrfs_workers
245 void btrfs_init_workers(struct btrfs_workers
*workers
, char *name
, int max
)
247 workers
->num_workers
= 0;
248 INIT_LIST_HEAD(&workers
->worker_list
);
249 INIT_LIST_HEAD(&workers
->idle_list
);
250 INIT_LIST_HEAD(&workers
->order_list
);
251 spin_lock_init(&workers
->lock
);
252 workers
->max_workers
= max
;
253 workers
->idle_thresh
= 32;
254 workers
->name
= name
;
255 workers
->ordered
= 0;
259 * starts new worker threads. This does not enforce the max worker
260 * count in case you need to temporarily go past it.
262 int btrfs_start_workers(struct btrfs_workers
*workers
, int num_workers
)
264 struct btrfs_worker_thread
*worker
;
268 for (i
= 0; i
< num_workers
; i
++) {
269 worker
= kzalloc(sizeof(*worker
), GFP_NOFS
);
275 INIT_LIST_HEAD(&worker
->pending
);
276 INIT_LIST_HEAD(&worker
->worker_list
);
277 spin_lock_init(&worker
->lock
);
278 atomic_set(&worker
->num_pending
, 0);
279 worker
->task
= kthread_run(worker_loop
, worker
,
280 "btrfs-%s-%d", workers
->name
,
281 workers
->num_workers
+ i
);
282 worker
->workers
= workers
;
283 if (IS_ERR(worker
->task
)) {
285 ret
= PTR_ERR(worker
->task
);
289 spin_lock_irq(&workers
->lock
);
290 list_add_tail(&worker
->worker_list
, &workers
->idle_list
);
292 workers
->num_workers
++;
293 spin_unlock_irq(&workers
->lock
);
297 btrfs_stop_workers(workers
);
302 * run through the list and find a worker thread that doesn't have a lot
303 * to do right now. This can return null if we aren't yet at the thread
304 * count limit and all of the threads are busy.
306 static struct btrfs_worker_thread
*next_worker(struct btrfs_workers
*workers
)
308 struct btrfs_worker_thread
*worker
;
309 struct list_head
*next
;
310 int enforce_min
= workers
->num_workers
< workers
->max_workers
;
313 * if we find an idle thread, don't move it to the end of the
314 * idle list. This improves the chance that the next submission
315 * will reuse the same thread, and maybe catch it while it is still
318 if (!list_empty(&workers
->idle_list
)) {
319 next
= workers
->idle_list
.next
;
320 worker
= list_entry(next
, struct btrfs_worker_thread
,
324 if (enforce_min
|| list_empty(&workers
->worker_list
))
328 * if we pick a busy task, move the task to the end of the list.
329 * hopefully this will keep things somewhat evenly balanced.
330 * Do the move in batches based on the sequence number. This groups
331 * requests submitted at roughly the same time onto the same worker.
333 next
= workers
->worker_list
.next
;
334 worker
= list_entry(next
, struct btrfs_worker_thread
, worker_list
);
335 atomic_inc(&worker
->num_pending
);
338 if (worker
->sequence
% workers
->idle_thresh
== 0)
339 list_move_tail(next
, &workers
->worker_list
);
344 * selects a worker thread to take the next job. This will either find
345 * an idle worker, start a new worker up to the max count, or just return
346 * one of the existing busy workers.
348 static struct btrfs_worker_thread
*find_worker(struct btrfs_workers
*workers
)
350 struct btrfs_worker_thread
*worker
;
354 spin_lock_irqsave(&workers
->lock
, flags
);
355 worker
= next_worker(workers
);
356 spin_unlock_irqrestore(&workers
->lock
, flags
);
359 spin_lock_irqsave(&workers
->lock
, flags
);
360 if (workers
->num_workers
>= workers
->max_workers
) {
361 struct list_head
*fallback
= NULL
;
363 * we have failed to find any workers, just
364 * return the force one
366 if (!list_empty(&workers
->worker_list
))
367 fallback
= workers
->worker_list
.next
;
368 if (!list_empty(&workers
->idle_list
))
369 fallback
= workers
->idle_list
.next
;
371 worker
= list_entry(fallback
,
372 struct btrfs_worker_thread
, worker_list
);
373 spin_unlock_irqrestore(&workers
->lock
, flags
);
375 spin_unlock_irqrestore(&workers
->lock
, flags
);
376 /* we're below the limit, start another worker */
377 btrfs_start_workers(workers
, 1);
385 * btrfs_requeue_work just puts the work item back on the tail of the list
386 * it was taken from. It is intended for use with long running work functions
387 * that make some progress and want to give the cpu up for others.
389 int btrfs_requeue_work(struct btrfs_work
*work
)
391 struct btrfs_worker_thread
*worker
= work
->worker
;
395 if (test_and_set_bit(WORK_QUEUED_BIT
, &work
->flags
))
398 spin_lock_irqsave(&worker
->lock
, flags
);
399 list_add_tail(&work
->list
, &worker
->pending
);
400 atomic_inc(&worker
->num_pending
);
402 /* by definition we're busy, take ourselves off the idle
406 spin_lock_irqsave(&worker
->workers
->lock
, flags
);
408 list_move_tail(&worker
->worker_list
,
409 &worker
->workers
->worker_list
);
410 spin_unlock_irqrestore(&worker
->workers
->lock
, flags
);
412 if (!worker
->working
) {
417 spin_unlock_irqrestore(&worker
->lock
, flags
);
419 wake_up_process(worker
->task
);
426 * places a struct btrfs_work into the pending queue of one of the kthreads
428 int btrfs_queue_worker(struct btrfs_workers
*workers
, struct btrfs_work
*work
)
430 struct btrfs_worker_thread
*worker
;
434 /* don't requeue something already on a list */
435 if (test_and_set_bit(WORK_QUEUED_BIT
, &work
->flags
))
438 worker
= find_worker(workers
);
439 if (workers
->ordered
) {
440 spin_lock_irqsave(&workers
->lock
, flags
);
441 list_add_tail(&work
->order_list
, &workers
->order_list
);
442 spin_unlock_irqrestore(&workers
->lock
, flags
);
444 INIT_LIST_HEAD(&work
->order_list
);
447 spin_lock_irqsave(&worker
->lock
, flags
);
449 list_add_tail(&work
->list
, &worker
->pending
);
450 atomic_inc(&worker
->num_pending
);
451 check_busy_worker(worker
);
454 * avoid calling into wake_up_process if this thread has already
457 if (!worker
->working
)
461 spin_unlock_irqrestore(&worker
->lock
, flags
);
464 wake_up_process(worker
->task
);