Commit | Line | Data |
---|---|---|
8b712842 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
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. | |
7 | * | |
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. | |
12 | * | |
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. | |
17 | */ | |
18 | ||
19 | #include <linux/kthread.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
8b712842 CM |
21 | #include <linux/list.h> |
22 | #include <linux/spinlock.h> | |
b51912c9 | 23 | #include <linux/freezer.h> |
8b712842 CM |
24 | #include "async-thread.h" |
25 | ||
4a69a410 CM |
26 | #define WORK_QUEUED_BIT 0 |
27 | #define WORK_DONE_BIT 1 | |
28 | #define WORK_ORDER_DONE_BIT 2 | |
d313d7a3 | 29 | #define WORK_HIGH_PRIO_BIT 3 |
4a69a410 | 30 | |
8b712842 CM |
31 | /* |
32 | * container for the kthread task pointer and the list of pending work | |
33 | * One of these is allocated per thread. | |
34 | */ | |
35 | struct btrfs_worker_thread { | |
35d8ba66 CM |
36 | /* pool we belong to */ |
37 | struct btrfs_workers *workers; | |
38 | ||
8b712842 CM |
39 | /* list of struct btrfs_work that are waiting for service */ |
40 | struct list_head pending; | |
d313d7a3 | 41 | struct list_head prio_pending; |
8b712842 CM |
42 | |
43 | /* list of worker threads from struct btrfs_workers */ | |
44 | struct list_head worker_list; | |
45 | ||
46 | /* kthread */ | |
47 | struct task_struct *task; | |
48 | ||
49 | /* number of things on the pending list */ | |
50 | atomic_t num_pending; | |
53863232 | 51 | |
9042846b CM |
52 | /* reference counter for this struct */ |
53 | atomic_t refs; | |
54 | ||
4854ddd0 | 55 | unsigned long sequence; |
8b712842 CM |
56 | |
57 | /* protects the pending list. */ | |
58 | spinlock_t lock; | |
59 | ||
60 | /* set to non-zero when this thread is already awake and kicking */ | |
61 | int working; | |
35d8ba66 CM |
62 | |
63 | /* are we currently idle */ | |
64 | int idle; | |
8b712842 CM |
65 | }; |
66 | ||
0dc3b84a JB |
67 | static int __btrfs_start_workers(struct btrfs_workers *workers); |
68 | ||
61d92c32 CM |
69 | /* |
70 | * btrfs_start_workers uses kthread_run, which can block waiting for memory | |
71 | * for a very long time. It will actually throttle on page writeback, | |
72 | * and so it may not make progress until after our btrfs worker threads | |
73 | * process all of the pending work structs in their queue | |
74 | * | |
75 | * This means we can't use btrfs_start_workers from inside a btrfs worker | |
76 | * thread that is used as part of cleaning dirty memory, which pretty much | |
77 | * involves all of the worker threads. | |
78 | * | |
79 | * Instead we have a helper queue who never has more than one thread | |
80 | * where we scheduler thread start operations. This worker_start struct | |
81 | * is used to contain the work and hold a pointer to the queue that needs | |
82 | * another worker. | |
83 | */ | |
84 | struct worker_start { | |
85 | struct btrfs_work work; | |
86 | struct btrfs_workers *queue; | |
87 | }; | |
88 | ||
89 | static void start_new_worker_func(struct btrfs_work *work) | |
90 | { | |
91 | struct worker_start *start; | |
92 | start = container_of(work, struct worker_start, work); | |
0dc3b84a | 93 | __btrfs_start_workers(start->queue); |
61d92c32 CM |
94 | kfree(start); |
95 | } | |
96 | ||
35d8ba66 CM |
97 | /* |
98 | * helper function to move a thread onto the idle list after it | |
99 | * has finished some requests. | |
100 | */ | |
101 | static void check_idle_worker(struct btrfs_worker_thread *worker) | |
102 | { | |
103 | if (!worker->idle && atomic_read(&worker->num_pending) < | |
104 | worker->workers->idle_thresh / 2) { | |
105 | unsigned long flags; | |
106 | spin_lock_irqsave(&worker->workers->lock, flags); | |
107 | worker->idle = 1; | |
3e99d8eb CM |
108 | |
109 | /* the list may be empty if the worker is just starting */ | |
964fb15a ID |
110 | if (!list_empty(&worker->worker_list) && |
111 | !worker->workers->stopping) { | |
3e99d8eb CM |
112 | list_move(&worker->worker_list, |
113 | &worker->workers->idle_list); | |
114 | } | |
35d8ba66 CM |
115 | spin_unlock_irqrestore(&worker->workers->lock, flags); |
116 | } | |
117 | } | |
118 | ||
119 | /* | |
120 | * helper function to move a thread off the idle list after new | |
121 | * pending work is added. | |
122 | */ | |
123 | static void check_busy_worker(struct btrfs_worker_thread *worker) | |
124 | { | |
125 | if (worker->idle && atomic_read(&worker->num_pending) >= | |
126 | worker->workers->idle_thresh) { | |
127 | unsigned long flags; | |
128 | spin_lock_irqsave(&worker->workers->lock, flags); | |
129 | worker->idle = 0; | |
3e99d8eb | 130 | |
964fb15a ID |
131 | if (!list_empty(&worker->worker_list) && |
132 | !worker->workers->stopping) { | |
3e99d8eb CM |
133 | list_move_tail(&worker->worker_list, |
134 | &worker->workers->worker_list); | |
135 | } | |
35d8ba66 CM |
136 | spin_unlock_irqrestore(&worker->workers->lock, flags); |
137 | } | |
138 | } | |
139 | ||
9042846b CM |
140 | static void check_pending_worker_creates(struct btrfs_worker_thread *worker) |
141 | { | |
142 | struct btrfs_workers *workers = worker->workers; | |
0dc3b84a | 143 | struct worker_start *start; |
9042846b CM |
144 | unsigned long flags; |
145 | ||
146 | rmb(); | |
147 | if (!workers->atomic_start_pending) | |
148 | return; | |
149 | ||
0dc3b84a JB |
150 | start = kzalloc(sizeof(*start), GFP_NOFS); |
151 | if (!start) | |
152 | return; | |
153 | ||
154 | start->work.func = start_new_worker_func; | |
155 | start->queue = workers; | |
156 | ||
9042846b CM |
157 | spin_lock_irqsave(&workers->lock, flags); |
158 | if (!workers->atomic_start_pending) | |
159 | goto out; | |
160 | ||
161 | workers->atomic_start_pending = 0; | |
61d92c32 CM |
162 | if (workers->num_workers + workers->num_workers_starting >= |
163 | workers->max_workers) | |
9042846b CM |
164 | goto out; |
165 | ||
61d92c32 | 166 | workers->num_workers_starting += 1; |
9042846b | 167 | spin_unlock_irqrestore(&workers->lock, flags); |
0dc3b84a | 168 | btrfs_queue_worker(workers->atomic_worker_start, &start->work); |
9042846b CM |
169 | return; |
170 | ||
171 | out: | |
0dc3b84a | 172 | kfree(start); |
9042846b CM |
173 | spin_unlock_irqrestore(&workers->lock, flags); |
174 | } | |
175 | ||
143bede5 | 176 | static noinline void run_ordered_completions(struct btrfs_workers *workers, |
4a69a410 CM |
177 | struct btrfs_work *work) |
178 | { | |
4a69a410 | 179 | if (!workers->ordered) |
143bede5 | 180 | return; |
4a69a410 CM |
181 | |
182 | set_bit(WORK_DONE_BIT, &work->flags); | |
183 | ||
4e3f9c50 | 184 | spin_lock(&workers->order_lock); |
4a69a410 | 185 | |
d313d7a3 CM |
186 | while (1) { |
187 | if (!list_empty(&workers->prio_order_list)) { | |
188 | work = list_entry(workers->prio_order_list.next, | |
189 | struct btrfs_work, order_list); | |
190 | } else if (!list_empty(&workers->order_list)) { | |
191 | work = list_entry(workers->order_list.next, | |
192 | struct btrfs_work, order_list); | |
193 | } else { | |
194 | break; | |
195 | } | |
4a69a410 CM |
196 | if (!test_bit(WORK_DONE_BIT, &work->flags)) |
197 | break; | |
198 | ||
199 | /* we are going to call the ordered done function, but | |
200 | * we leave the work item on the list as a barrier so | |
201 | * that later work items that are done don't have their | |
202 | * functions called before this one returns | |
203 | */ | |
204 | if (test_and_set_bit(WORK_ORDER_DONE_BIT, &work->flags)) | |
205 | break; | |
206 | ||
4e3f9c50 | 207 | spin_unlock(&workers->order_lock); |
4a69a410 CM |
208 | |
209 | work->ordered_func(work); | |
210 | ||
e9fbcb42 | 211 | /* now take the lock again and drop our item from the list */ |
4e3f9c50 | 212 | spin_lock(&workers->order_lock); |
4a69a410 | 213 | list_del(&work->order_list); |
e9fbcb42 CM |
214 | spin_unlock(&workers->order_lock); |
215 | ||
216 | /* | |
217 | * we don't want to call the ordered free functions | |
218 | * with the lock held though | |
219 | */ | |
4a69a410 | 220 | work->ordered_free(work); |
e9fbcb42 | 221 | spin_lock(&workers->order_lock); |
4a69a410 CM |
222 | } |
223 | ||
4e3f9c50 | 224 | spin_unlock(&workers->order_lock); |
4a69a410 CM |
225 | } |
226 | ||
9042846b CM |
227 | static void put_worker(struct btrfs_worker_thread *worker) |
228 | { | |
229 | if (atomic_dec_and_test(&worker->refs)) | |
230 | kfree(worker); | |
231 | } | |
232 | ||
233 | static int try_worker_shutdown(struct btrfs_worker_thread *worker) | |
234 | { | |
235 | int freeit = 0; | |
236 | ||
237 | spin_lock_irq(&worker->lock); | |
627e421a | 238 | spin_lock(&worker->workers->lock); |
9042846b CM |
239 | if (worker->workers->num_workers > 1 && |
240 | worker->idle && | |
241 | !worker->working && | |
242 | !list_empty(&worker->worker_list) && | |
243 | list_empty(&worker->prio_pending) && | |
6e74057c CM |
244 | list_empty(&worker->pending) && |
245 | atomic_read(&worker->num_pending) == 0) { | |
9042846b CM |
246 | freeit = 1; |
247 | list_del_init(&worker->worker_list); | |
248 | worker->workers->num_workers--; | |
249 | } | |
627e421a | 250 | spin_unlock(&worker->workers->lock); |
9042846b CM |
251 | spin_unlock_irq(&worker->lock); |
252 | ||
253 | if (freeit) | |
254 | put_worker(worker); | |
255 | return freeit; | |
256 | } | |
257 | ||
4f878e84 CM |
258 | static struct btrfs_work *get_next_work(struct btrfs_worker_thread *worker, |
259 | struct list_head *prio_head, | |
260 | struct list_head *head) | |
261 | { | |
262 | struct btrfs_work *work = NULL; | |
263 | struct list_head *cur = NULL; | |
264 | ||
265 | if(!list_empty(prio_head)) | |
266 | cur = prio_head->next; | |
267 | ||
268 | smp_mb(); | |
269 | if (!list_empty(&worker->prio_pending)) | |
270 | goto refill; | |
271 | ||
272 | if (!list_empty(head)) | |
273 | cur = head->next; | |
274 | ||
275 | if (cur) | |
276 | goto out; | |
277 | ||
278 | refill: | |
279 | spin_lock_irq(&worker->lock); | |
280 | list_splice_tail_init(&worker->prio_pending, prio_head); | |
281 | list_splice_tail_init(&worker->pending, head); | |
282 | ||
283 | if (!list_empty(prio_head)) | |
284 | cur = prio_head->next; | |
285 | else if (!list_empty(head)) | |
286 | cur = head->next; | |
287 | spin_unlock_irq(&worker->lock); | |
288 | ||
289 | if (!cur) | |
290 | goto out_fail; | |
291 | ||
292 | out: | |
293 | work = list_entry(cur, struct btrfs_work, list); | |
294 | ||
295 | out_fail: | |
296 | return work; | |
297 | } | |
298 | ||
8b712842 CM |
299 | /* |
300 | * main loop for servicing work items | |
301 | */ | |
302 | static int worker_loop(void *arg) | |
303 | { | |
304 | struct btrfs_worker_thread *worker = arg; | |
4f878e84 CM |
305 | struct list_head head; |
306 | struct list_head prio_head; | |
8b712842 | 307 | struct btrfs_work *work; |
4f878e84 CM |
308 | |
309 | INIT_LIST_HEAD(&head); | |
310 | INIT_LIST_HEAD(&prio_head); | |
311 | ||
8b712842 | 312 | do { |
4f878e84 | 313 | again: |
d313d7a3 | 314 | while (1) { |
4f878e84 CM |
315 | |
316 | ||
317 | work = get_next_work(worker, &prio_head, &head); | |
318 | if (!work) | |
d313d7a3 CM |
319 | break; |
320 | ||
8b712842 | 321 | list_del(&work->list); |
4a69a410 | 322 | clear_bit(WORK_QUEUED_BIT, &work->flags); |
8b712842 CM |
323 | |
324 | work->worker = worker; | |
8b712842 CM |
325 | |
326 | work->func(work); | |
327 | ||
328 | atomic_dec(&worker->num_pending); | |
4a69a410 CM |
329 | /* |
330 | * unless this is an ordered work queue, | |
331 | * 'work' was probably freed by func above. | |
332 | */ | |
333 | run_ordered_completions(worker->workers, work); | |
334 | ||
9042846b | 335 | check_pending_worker_creates(worker); |
8f3b65a3 | 336 | cond_resched(); |
8b712842 | 337 | } |
4f878e84 CM |
338 | |
339 | spin_lock_irq(&worker->lock); | |
340 | check_idle_worker(worker); | |
341 | ||
8b712842 | 342 | if (freezing(current)) { |
b51912c9 CM |
343 | worker->working = 0; |
344 | spin_unlock_irq(&worker->lock); | |
a0acae0e | 345 | try_to_freeze(); |
8b712842 | 346 | } else { |
8b712842 | 347 | spin_unlock_irq(&worker->lock); |
b51912c9 CM |
348 | if (!kthread_should_stop()) { |
349 | cpu_relax(); | |
350 | /* | |
351 | * we've dropped the lock, did someone else | |
352 | * jump_in? | |
353 | */ | |
354 | smp_mb(); | |
d313d7a3 CM |
355 | if (!list_empty(&worker->pending) || |
356 | !list_empty(&worker->prio_pending)) | |
b51912c9 CM |
357 | continue; |
358 | ||
359 | /* | |
360 | * this short schedule allows more work to | |
361 | * come in without the queue functions | |
362 | * needing to go through wake_up_process() | |
363 | * | |
364 | * worker->working is still 1, so nobody | |
365 | * is going to try and wake us up | |
366 | */ | |
367 | schedule_timeout(1); | |
368 | smp_mb(); | |
d313d7a3 CM |
369 | if (!list_empty(&worker->pending) || |
370 | !list_empty(&worker->prio_pending)) | |
b51912c9 CM |
371 | continue; |
372 | ||
b5555f77 AG |
373 | if (kthread_should_stop()) |
374 | break; | |
375 | ||
b51912c9 CM |
376 | /* still no more work?, sleep for real */ |
377 | spin_lock_irq(&worker->lock); | |
378 | set_current_state(TASK_INTERRUPTIBLE); | |
d313d7a3 | 379 | if (!list_empty(&worker->pending) || |
4f878e84 CM |
380 | !list_empty(&worker->prio_pending)) { |
381 | spin_unlock_irq(&worker->lock); | |
ed3b3d31 | 382 | set_current_state(TASK_RUNNING); |
4f878e84 CM |
383 | goto again; |
384 | } | |
b51912c9 CM |
385 | |
386 | /* | |
387 | * this makes sure we get a wakeup when someone | |
388 | * adds something new to the queue | |
389 | */ | |
390 | worker->working = 0; | |
391 | spin_unlock_irq(&worker->lock); | |
392 | ||
9042846b CM |
393 | if (!kthread_should_stop()) { |
394 | schedule_timeout(HZ * 120); | |
395 | if (!worker->working && | |
396 | try_worker_shutdown(worker)) { | |
397 | return 0; | |
398 | } | |
399 | } | |
b51912c9 | 400 | } |
8b712842 CM |
401 | __set_current_state(TASK_RUNNING); |
402 | } | |
403 | } while (!kthread_should_stop()); | |
404 | return 0; | |
405 | } | |
406 | ||
407 | /* | |
408 | * this will wait for all the worker threads to shutdown | |
409 | */ | |
143bede5 | 410 | void btrfs_stop_workers(struct btrfs_workers *workers) |
8b712842 CM |
411 | { |
412 | struct list_head *cur; | |
413 | struct btrfs_worker_thread *worker; | |
9042846b | 414 | int can_stop; |
8b712842 | 415 | |
9042846b | 416 | spin_lock_irq(&workers->lock); |
964fb15a | 417 | workers->stopping = 1; |
35d8ba66 | 418 | list_splice_init(&workers->idle_list, &workers->worker_list); |
d397712b | 419 | while (!list_empty(&workers->worker_list)) { |
8b712842 CM |
420 | cur = workers->worker_list.next; |
421 | worker = list_entry(cur, struct btrfs_worker_thread, | |
422 | worker_list); | |
9042846b CM |
423 | |
424 | atomic_inc(&worker->refs); | |
425 | workers->num_workers -= 1; | |
426 | if (!list_empty(&worker->worker_list)) { | |
427 | list_del_init(&worker->worker_list); | |
428 | put_worker(worker); | |
429 | can_stop = 1; | |
430 | } else | |
431 | can_stop = 0; | |
432 | spin_unlock_irq(&workers->lock); | |
433 | if (can_stop) | |
434 | kthread_stop(worker->task); | |
435 | spin_lock_irq(&workers->lock); | |
436 | put_worker(worker); | |
8b712842 | 437 | } |
9042846b | 438 | spin_unlock_irq(&workers->lock); |
8b712842 CM |
439 | } |
440 | ||
441 | /* | |
442 | * simple init on struct btrfs_workers | |
443 | */ | |
61d92c32 CM |
444 | void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max, |
445 | struct btrfs_workers *async_helper) | |
8b712842 CM |
446 | { |
447 | workers->num_workers = 0; | |
61d92c32 | 448 | workers->num_workers_starting = 0; |
8b712842 | 449 | INIT_LIST_HEAD(&workers->worker_list); |
35d8ba66 | 450 | INIT_LIST_HEAD(&workers->idle_list); |
4a69a410 | 451 | INIT_LIST_HEAD(&workers->order_list); |
d313d7a3 | 452 | INIT_LIST_HEAD(&workers->prio_order_list); |
8b712842 | 453 | spin_lock_init(&workers->lock); |
4e3f9c50 | 454 | spin_lock_init(&workers->order_lock); |
8b712842 | 455 | workers->max_workers = max; |
61b49440 | 456 | workers->idle_thresh = 32; |
5443be45 | 457 | workers->name = name; |
4a69a410 | 458 | workers->ordered = 0; |
9042846b | 459 | workers->atomic_start_pending = 0; |
61d92c32 | 460 | workers->atomic_worker_start = async_helper; |
964fb15a | 461 | workers->stopping = 0; |
8b712842 CM |
462 | } |
463 | ||
464 | /* | |
465 | * starts new worker threads. This does not enforce the max worker | |
466 | * count in case you need to temporarily go past it. | |
467 | */ | |
0dc3b84a | 468 | static int __btrfs_start_workers(struct btrfs_workers *workers) |
8b712842 CM |
469 | { |
470 | struct btrfs_worker_thread *worker; | |
471 | int ret = 0; | |
8b712842 | 472 | |
0dc3b84a JB |
473 | worker = kzalloc(sizeof(*worker), GFP_NOFS); |
474 | if (!worker) { | |
475 | ret = -ENOMEM; | |
476 | goto fail; | |
477 | } | |
8b712842 | 478 | |
0dc3b84a JB |
479 | INIT_LIST_HEAD(&worker->pending); |
480 | INIT_LIST_HEAD(&worker->prio_pending); | |
481 | INIT_LIST_HEAD(&worker->worker_list); | |
482 | spin_lock_init(&worker->lock); | |
483 | ||
484 | atomic_set(&worker->num_pending, 0); | |
485 | atomic_set(&worker->refs, 1); | |
486 | worker->workers = workers; | |
964fb15a ID |
487 | worker->task = kthread_create(worker_loop, worker, |
488 | "btrfs-%s-%d", workers->name, | |
489 | workers->num_workers + 1); | |
0dc3b84a JB |
490 | if (IS_ERR(worker->task)) { |
491 | ret = PTR_ERR(worker->task); | |
0dc3b84a | 492 | goto fail; |
8b712842 | 493 | } |
964fb15a | 494 | |
0dc3b84a | 495 | spin_lock_irq(&workers->lock); |
964fb15a ID |
496 | if (workers->stopping) { |
497 | spin_unlock_irq(&workers->lock); | |
498 | goto fail_kthread; | |
499 | } | |
0dc3b84a JB |
500 | list_add_tail(&worker->worker_list, &workers->idle_list); |
501 | worker->idle = 1; | |
502 | workers->num_workers++; | |
503 | workers->num_workers_starting--; | |
504 | WARN_ON(workers->num_workers_starting < 0); | |
505 | spin_unlock_irq(&workers->lock); | |
506 | ||
964fb15a | 507 | wake_up_process(worker->task); |
8b712842 | 508 | return 0; |
964fb15a ID |
509 | |
510 | fail_kthread: | |
511 | kthread_stop(worker->task); | |
8b712842 | 512 | fail: |
964fb15a | 513 | kfree(worker); |
0dc3b84a JB |
514 | spin_lock_irq(&workers->lock); |
515 | workers->num_workers_starting--; | |
516 | spin_unlock_irq(&workers->lock); | |
8b712842 CM |
517 | return ret; |
518 | } | |
519 | ||
0dc3b84a | 520 | int btrfs_start_workers(struct btrfs_workers *workers) |
61d92c32 CM |
521 | { |
522 | spin_lock_irq(&workers->lock); | |
0dc3b84a | 523 | workers->num_workers_starting++; |
61d92c32 | 524 | spin_unlock_irq(&workers->lock); |
0dc3b84a | 525 | return __btrfs_start_workers(workers); |
61d92c32 CM |
526 | } |
527 | ||
8b712842 CM |
528 | /* |
529 | * run through the list and find a worker thread that doesn't have a lot | |
530 | * to do right now. This can return null if we aren't yet at the thread | |
531 | * count limit and all of the threads are busy. | |
532 | */ | |
533 | static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers) | |
534 | { | |
535 | struct btrfs_worker_thread *worker; | |
536 | struct list_head *next; | |
61d92c32 CM |
537 | int enforce_min; |
538 | ||
539 | enforce_min = (workers->num_workers + workers->num_workers_starting) < | |
540 | workers->max_workers; | |
8b712842 | 541 | |
8b712842 | 542 | /* |
35d8ba66 CM |
543 | * if we find an idle thread, don't move it to the end of the |
544 | * idle list. This improves the chance that the next submission | |
545 | * will reuse the same thread, and maybe catch it while it is still | |
546 | * working | |
8b712842 | 547 | */ |
35d8ba66 CM |
548 | if (!list_empty(&workers->idle_list)) { |
549 | next = workers->idle_list.next; | |
8b712842 CM |
550 | worker = list_entry(next, struct btrfs_worker_thread, |
551 | worker_list); | |
35d8ba66 | 552 | return worker; |
8b712842 | 553 | } |
35d8ba66 CM |
554 | if (enforce_min || list_empty(&workers->worker_list)) |
555 | return NULL; | |
556 | ||
8b712842 | 557 | /* |
35d8ba66 | 558 | * if we pick a busy task, move the task to the end of the list. |
d352ac68 CM |
559 | * hopefully this will keep things somewhat evenly balanced. |
560 | * Do the move in batches based on the sequence number. This groups | |
561 | * requests submitted at roughly the same time onto the same worker. | |
8b712842 | 562 | */ |
35d8ba66 CM |
563 | next = workers->worker_list.next; |
564 | worker = list_entry(next, struct btrfs_worker_thread, worker_list); | |
4854ddd0 | 565 | worker->sequence++; |
d352ac68 | 566 | |
53863232 | 567 | if (worker->sequence % workers->idle_thresh == 0) |
4854ddd0 | 568 | list_move_tail(next, &workers->worker_list); |
8b712842 CM |
569 | return worker; |
570 | } | |
571 | ||
d352ac68 CM |
572 | /* |
573 | * selects a worker thread to take the next job. This will either find | |
574 | * an idle worker, start a new worker up to the max count, or just return | |
575 | * one of the existing busy workers. | |
576 | */ | |
8b712842 CM |
577 | static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers) |
578 | { | |
579 | struct btrfs_worker_thread *worker; | |
580 | unsigned long flags; | |
9042846b | 581 | struct list_head *fallback; |
0dc3b84a | 582 | int ret; |
8b712842 | 583 | |
8b712842 | 584 | spin_lock_irqsave(&workers->lock, flags); |
8d532b2a | 585 | again: |
8b712842 | 586 | worker = next_worker(workers); |
8b712842 CM |
587 | |
588 | if (!worker) { | |
61d92c32 CM |
589 | if (workers->num_workers + workers->num_workers_starting >= |
590 | workers->max_workers) { | |
9042846b CM |
591 | goto fallback; |
592 | } else if (workers->atomic_worker_start) { | |
593 | workers->atomic_start_pending = 1; | |
594 | goto fallback; | |
8b712842 | 595 | } else { |
61d92c32 | 596 | workers->num_workers_starting++; |
8b712842 CM |
597 | spin_unlock_irqrestore(&workers->lock, flags); |
598 | /* we're below the limit, start another worker */ | |
0dc3b84a | 599 | ret = __btrfs_start_workers(workers); |
8d532b2a | 600 | spin_lock_irqsave(&workers->lock, flags); |
0dc3b84a JB |
601 | if (ret) |
602 | goto fallback; | |
8b712842 CM |
603 | goto again; |
604 | } | |
605 | } | |
6e74057c | 606 | goto found; |
9042846b CM |
607 | |
608 | fallback: | |
609 | fallback = NULL; | |
610 | /* | |
611 | * we have failed to find any workers, just | |
612 | * return the first one we can find. | |
613 | */ | |
614 | if (!list_empty(&workers->worker_list)) | |
615 | fallback = workers->worker_list.next; | |
616 | if (!list_empty(&workers->idle_list)) | |
617 | fallback = workers->idle_list.next; | |
618 | BUG_ON(!fallback); | |
619 | worker = list_entry(fallback, | |
620 | struct btrfs_worker_thread, worker_list); | |
6e74057c CM |
621 | found: |
622 | /* | |
623 | * this makes sure the worker doesn't exit before it is placed | |
624 | * onto a busy/idle list | |
625 | */ | |
626 | atomic_inc(&worker->num_pending); | |
9042846b CM |
627 | spin_unlock_irqrestore(&workers->lock, flags); |
628 | return worker; | |
8b712842 CM |
629 | } |
630 | ||
631 | /* | |
632 | * btrfs_requeue_work just puts the work item back on the tail of the list | |
633 | * it was taken from. It is intended for use with long running work functions | |
634 | * that make some progress and want to give the cpu up for others. | |
635 | */ | |
143bede5 | 636 | void btrfs_requeue_work(struct btrfs_work *work) |
8b712842 CM |
637 | { |
638 | struct btrfs_worker_thread *worker = work->worker; | |
639 | unsigned long flags; | |
a6837051 | 640 | int wake = 0; |
8b712842 | 641 | |
4a69a410 | 642 | if (test_and_set_bit(WORK_QUEUED_BIT, &work->flags)) |
143bede5 | 643 | return; |
8b712842 CM |
644 | |
645 | spin_lock_irqsave(&worker->lock, flags); | |
d313d7a3 CM |
646 | if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags)) |
647 | list_add_tail(&work->list, &worker->prio_pending); | |
648 | else | |
649 | list_add_tail(&work->list, &worker->pending); | |
b51912c9 | 650 | atomic_inc(&worker->num_pending); |
75ccf47d CM |
651 | |
652 | /* by definition we're busy, take ourselves off the idle | |
653 | * list | |
654 | */ | |
655 | if (worker->idle) { | |
29c5e8ce | 656 | spin_lock(&worker->workers->lock); |
75ccf47d CM |
657 | worker->idle = 0; |
658 | list_move_tail(&worker->worker_list, | |
6e74057c | 659 | &worker->workers->worker_list); |
29c5e8ce | 660 | spin_unlock(&worker->workers->lock); |
75ccf47d | 661 | } |
a6837051 CM |
662 | if (!worker->working) { |
663 | wake = 1; | |
664 | worker->working = 1; | |
665 | } | |
75ccf47d | 666 | |
a6837051 CM |
667 | if (wake) |
668 | wake_up_process(worker->task); | |
9042846b | 669 | spin_unlock_irqrestore(&worker->lock, flags); |
8b712842 CM |
670 | } |
671 | ||
d313d7a3 CM |
672 | void btrfs_set_work_high_prio(struct btrfs_work *work) |
673 | { | |
674 | set_bit(WORK_HIGH_PRIO_BIT, &work->flags); | |
675 | } | |
676 | ||
8b712842 CM |
677 | /* |
678 | * places a struct btrfs_work into the pending queue of one of the kthreads | |
679 | */ | |
0dc3b84a | 680 | void btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work) |
8b712842 CM |
681 | { |
682 | struct btrfs_worker_thread *worker; | |
683 | unsigned long flags; | |
684 | int wake = 0; | |
685 | ||
686 | /* don't requeue something already on a list */ | |
4a69a410 | 687 | if (test_and_set_bit(WORK_QUEUED_BIT, &work->flags)) |
0dc3b84a | 688 | return; |
8b712842 CM |
689 | |
690 | worker = find_worker(workers); | |
4a69a410 | 691 | if (workers->ordered) { |
4e3f9c50 CM |
692 | /* |
693 | * you're not allowed to do ordered queues from an | |
694 | * interrupt handler | |
695 | */ | |
696 | spin_lock(&workers->order_lock); | |
d313d7a3 CM |
697 | if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags)) { |
698 | list_add_tail(&work->order_list, | |
699 | &workers->prio_order_list); | |
700 | } else { | |
701 | list_add_tail(&work->order_list, &workers->order_list); | |
702 | } | |
4e3f9c50 | 703 | spin_unlock(&workers->order_lock); |
4a69a410 CM |
704 | } else { |
705 | INIT_LIST_HEAD(&work->order_list); | |
706 | } | |
8b712842 CM |
707 | |
708 | spin_lock_irqsave(&worker->lock, flags); | |
a6837051 | 709 | |
d313d7a3 CM |
710 | if (test_bit(WORK_HIGH_PRIO_BIT, &work->flags)) |
711 | list_add_tail(&work->list, &worker->prio_pending); | |
712 | else | |
713 | list_add_tail(&work->list, &worker->pending); | |
35d8ba66 | 714 | check_busy_worker(worker); |
8b712842 CM |
715 | |
716 | /* | |
717 | * avoid calling into wake_up_process if this thread has already | |
718 | * been kicked | |
719 | */ | |
720 | if (!worker->working) | |
721 | wake = 1; | |
722 | worker->working = 1; | |
723 | ||
8b712842 CM |
724 | if (wake) |
725 | wake_up_process(worker->task); | |
9042846b | 726 | spin_unlock_irqrestore(&worker->lock, flags); |
8b712842 | 727 | } |