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 | ||
d05e5a4d | 19 | #include <linux/version.h> |
8b712842 CM |
20 | #include <linux/kthread.h> |
21 | #include <linux/list.h> | |
22 | #include <linux/spinlock.h> | |
d05e5a4d | 23 | # include <linux/freezer.h> |
8b712842 CM |
24 | #include "async-thread.h" |
25 | ||
26 | /* | |
27 | * container for the kthread task pointer and the list of pending work | |
28 | * One of these is allocated per thread. | |
29 | */ | |
30 | struct btrfs_worker_thread { | |
35d8ba66 CM |
31 | /* pool we belong to */ |
32 | struct btrfs_workers *workers; | |
33 | ||
8b712842 CM |
34 | /* list of struct btrfs_work that are waiting for service */ |
35 | struct list_head pending; | |
36 | ||
37 | /* list of worker threads from struct btrfs_workers */ | |
38 | struct list_head worker_list; | |
39 | ||
40 | /* kthread */ | |
41 | struct task_struct *task; | |
42 | ||
43 | /* number of things on the pending list */ | |
44 | atomic_t num_pending; | |
53863232 | 45 | |
4854ddd0 | 46 | unsigned long sequence; |
8b712842 CM |
47 | |
48 | /* protects the pending list. */ | |
49 | spinlock_t lock; | |
50 | ||
51 | /* set to non-zero when this thread is already awake and kicking */ | |
52 | int working; | |
35d8ba66 CM |
53 | |
54 | /* are we currently idle */ | |
55 | int idle; | |
8b712842 CM |
56 | }; |
57 | ||
35d8ba66 CM |
58 | /* |
59 | * helper function to move a thread onto the idle list after it | |
60 | * has finished some requests. | |
61 | */ | |
62 | static void check_idle_worker(struct btrfs_worker_thread *worker) | |
63 | { | |
64 | if (!worker->idle && atomic_read(&worker->num_pending) < | |
65 | worker->workers->idle_thresh / 2) { | |
66 | unsigned long flags; | |
67 | spin_lock_irqsave(&worker->workers->lock, flags); | |
68 | worker->idle = 1; | |
69 | list_move(&worker->worker_list, &worker->workers->idle_list); | |
70 | spin_unlock_irqrestore(&worker->workers->lock, flags); | |
71 | } | |
72 | } | |
73 | ||
74 | /* | |
75 | * helper function to move a thread off the idle list after new | |
76 | * pending work is added. | |
77 | */ | |
78 | static void check_busy_worker(struct btrfs_worker_thread *worker) | |
79 | { | |
80 | if (worker->idle && atomic_read(&worker->num_pending) >= | |
81 | worker->workers->idle_thresh) { | |
82 | unsigned long flags; | |
83 | spin_lock_irqsave(&worker->workers->lock, flags); | |
84 | worker->idle = 0; | |
85 | list_move_tail(&worker->worker_list, | |
86 | &worker->workers->worker_list); | |
87 | spin_unlock_irqrestore(&worker->workers->lock, flags); | |
88 | } | |
89 | } | |
90 | ||
8b712842 CM |
91 | /* |
92 | * main loop for servicing work items | |
93 | */ | |
94 | static int worker_loop(void *arg) | |
95 | { | |
96 | struct btrfs_worker_thread *worker = arg; | |
97 | struct list_head *cur; | |
98 | struct btrfs_work *work; | |
99 | do { | |
100 | spin_lock_irq(&worker->lock); | |
101 | while(!list_empty(&worker->pending)) { | |
102 | cur = worker->pending.next; | |
103 | work = list_entry(cur, struct btrfs_work, list); | |
104 | list_del(&work->list); | |
105 | clear_bit(0, &work->flags); | |
106 | ||
107 | work->worker = worker; | |
108 | spin_unlock_irq(&worker->lock); | |
109 | ||
110 | work->func(work); | |
111 | ||
112 | atomic_dec(&worker->num_pending); | |
113 | spin_lock_irq(&worker->lock); | |
35d8ba66 | 114 | check_idle_worker(worker); |
8b712842 CM |
115 | } |
116 | worker->working = 0; | |
117 | if (freezing(current)) { | |
118 | refrigerator(); | |
119 | } else { | |
120 | set_current_state(TASK_INTERRUPTIBLE); | |
121 | spin_unlock_irq(&worker->lock); | |
122 | schedule(); | |
123 | __set_current_state(TASK_RUNNING); | |
124 | } | |
125 | } while (!kthread_should_stop()); | |
126 | return 0; | |
127 | } | |
128 | ||
129 | /* | |
130 | * this will wait for all the worker threads to shutdown | |
131 | */ | |
132 | int btrfs_stop_workers(struct btrfs_workers *workers) | |
133 | { | |
134 | struct list_head *cur; | |
135 | struct btrfs_worker_thread *worker; | |
136 | ||
35d8ba66 | 137 | list_splice_init(&workers->idle_list, &workers->worker_list); |
8b712842 CM |
138 | while(!list_empty(&workers->worker_list)) { |
139 | cur = workers->worker_list.next; | |
140 | worker = list_entry(cur, struct btrfs_worker_thread, | |
141 | worker_list); | |
142 | kthread_stop(worker->task); | |
143 | list_del(&worker->worker_list); | |
144 | kfree(worker); | |
145 | } | |
146 | return 0; | |
147 | } | |
148 | ||
149 | /* | |
150 | * simple init on struct btrfs_workers | |
151 | */ | |
5443be45 | 152 | void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max) |
8b712842 CM |
153 | { |
154 | workers->num_workers = 0; | |
155 | INIT_LIST_HEAD(&workers->worker_list); | |
35d8ba66 | 156 | INIT_LIST_HEAD(&workers->idle_list); |
8b712842 CM |
157 | spin_lock_init(&workers->lock); |
158 | workers->max_workers = max; | |
61b49440 | 159 | workers->idle_thresh = 32; |
5443be45 | 160 | workers->name = name; |
8b712842 CM |
161 | } |
162 | ||
163 | /* | |
164 | * starts new worker threads. This does not enforce the max worker | |
165 | * count in case you need to temporarily go past it. | |
166 | */ | |
167 | int btrfs_start_workers(struct btrfs_workers *workers, int num_workers) | |
168 | { | |
169 | struct btrfs_worker_thread *worker; | |
170 | int ret = 0; | |
171 | int i; | |
172 | ||
173 | for (i = 0; i < num_workers; i++) { | |
174 | worker = kzalloc(sizeof(*worker), GFP_NOFS); | |
175 | if (!worker) { | |
176 | ret = -ENOMEM; | |
177 | goto fail; | |
178 | } | |
179 | ||
180 | INIT_LIST_HEAD(&worker->pending); | |
181 | INIT_LIST_HEAD(&worker->worker_list); | |
182 | spin_lock_init(&worker->lock); | |
183 | atomic_set(&worker->num_pending, 0); | |
5443be45 CM |
184 | worker->task = kthread_run(worker_loop, worker, |
185 | "btrfs-%s-%d", workers->name, | |
186 | workers->num_workers + i); | |
35d8ba66 | 187 | worker->workers = workers; |
8b712842 | 188 | if (IS_ERR(worker->task)) { |
3bf10418 | 189 | kfree(worker); |
8b712842 CM |
190 | ret = PTR_ERR(worker->task); |
191 | goto fail; | |
192 | } | |
193 | ||
194 | spin_lock_irq(&workers->lock); | |
35d8ba66 | 195 | list_add_tail(&worker->worker_list, &workers->idle_list); |
4854ddd0 | 196 | worker->idle = 1; |
8b712842 CM |
197 | workers->num_workers++; |
198 | spin_unlock_irq(&workers->lock); | |
199 | } | |
200 | return 0; | |
201 | fail: | |
202 | btrfs_stop_workers(workers); | |
203 | return ret; | |
204 | } | |
205 | ||
206 | /* | |
207 | * run through the list and find a worker thread that doesn't have a lot | |
208 | * to do right now. This can return null if we aren't yet at the thread | |
209 | * count limit and all of the threads are busy. | |
210 | */ | |
211 | static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers) | |
212 | { | |
213 | struct btrfs_worker_thread *worker; | |
214 | struct list_head *next; | |
8b712842 CM |
215 | int enforce_min = workers->num_workers < workers->max_workers; |
216 | ||
8b712842 | 217 | /* |
35d8ba66 CM |
218 | * if we find an idle thread, don't move it to the end of the |
219 | * idle list. This improves the chance that the next submission | |
220 | * will reuse the same thread, and maybe catch it while it is still | |
221 | * working | |
8b712842 | 222 | */ |
35d8ba66 CM |
223 | if (!list_empty(&workers->idle_list)) { |
224 | next = workers->idle_list.next; | |
8b712842 CM |
225 | worker = list_entry(next, struct btrfs_worker_thread, |
226 | worker_list); | |
35d8ba66 | 227 | return worker; |
8b712842 | 228 | } |
35d8ba66 CM |
229 | if (enforce_min || list_empty(&workers->worker_list)) |
230 | return NULL; | |
231 | ||
8b712842 | 232 | /* |
35d8ba66 | 233 | * if we pick a busy task, move the task to the end of the list. |
d352ac68 CM |
234 | * hopefully this will keep things somewhat evenly balanced. |
235 | * Do the move in batches based on the sequence number. This groups | |
236 | * requests submitted at roughly the same time onto the same worker. | |
8b712842 | 237 | */ |
35d8ba66 CM |
238 | next = workers->worker_list.next; |
239 | worker = list_entry(next, struct btrfs_worker_thread, worker_list); | |
4854ddd0 CM |
240 | atomic_inc(&worker->num_pending); |
241 | worker->sequence++; | |
d352ac68 | 242 | |
53863232 | 243 | if (worker->sequence % workers->idle_thresh == 0) |
4854ddd0 | 244 | list_move_tail(next, &workers->worker_list); |
8b712842 CM |
245 | return worker; |
246 | } | |
247 | ||
d352ac68 CM |
248 | /* |
249 | * selects a worker thread to take the next job. This will either find | |
250 | * an idle worker, start a new worker up to the max count, or just return | |
251 | * one of the existing busy workers. | |
252 | */ | |
8b712842 CM |
253 | static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers) |
254 | { | |
255 | struct btrfs_worker_thread *worker; | |
256 | unsigned long flags; | |
257 | ||
258 | again: | |
259 | spin_lock_irqsave(&workers->lock, flags); | |
260 | worker = next_worker(workers); | |
261 | spin_unlock_irqrestore(&workers->lock, flags); | |
262 | ||
263 | if (!worker) { | |
264 | spin_lock_irqsave(&workers->lock, flags); | |
265 | if (workers->num_workers >= workers->max_workers) { | |
35d8ba66 | 266 | struct list_head *fallback = NULL; |
8b712842 CM |
267 | /* |
268 | * we have failed to find any workers, just | |
269 | * return the force one | |
270 | */ | |
35d8ba66 CM |
271 | if (!list_empty(&workers->worker_list)) |
272 | fallback = workers->worker_list.next; | |
273 | if (!list_empty(&workers->idle_list)) | |
274 | fallback = workers->idle_list.next; | |
275 | BUG_ON(!fallback); | |
276 | worker = list_entry(fallback, | |
8b712842 CM |
277 | struct btrfs_worker_thread, worker_list); |
278 | spin_unlock_irqrestore(&workers->lock, flags); | |
279 | } else { | |
280 | spin_unlock_irqrestore(&workers->lock, flags); | |
281 | /* we're below the limit, start another worker */ | |
282 | btrfs_start_workers(workers, 1); | |
283 | goto again; | |
284 | } | |
285 | } | |
286 | return worker; | |
287 | } | |
288 | ||
289 | /* | |
290 | * btrfs_requeue_work just puts the work item back on the tail of the list | |
291 | * it was taken from. It is intended for use with long running work functions | |
292 | * that make some progress and want to give the cpu up for others. | |
293 | */ | |
294 | int btrfs_requeue_work(struct btrfs_work *work) | |
295 | { | |
296 | struct btrfs_worker_thread *worker = work->worker; | |
297 | unsigned long flags; | |
298 | ||
299 | if (test_and_set_bit(0, &work->flags)) | |
300 | goto out; | |
301 | ||
302 | spin_lock_irqsave(&worker->lock, flags); | |
303 | atomic_inc(&worker->num_pending); | |
304 | list_add_tail(&work->list, &worker->pending); | |
75ccf47d CM |
305 | |
306 | /* by definition we're busy, take ourselves off the idle | |
307 | * list | |
308 | */ | |
309 | if (worker->idle) { | |
310 | spin_lock_irqsave(&worker->workers->lock, flags); | |
311 | worker->idle = 0; | |
312 | list_move_tail(&worker->worker_list, | |
313 | &worker->workers->worker_list); | |
314 | spin_unlock_irqrestore(&worker->workers->lock, flags); | |
315 | } | |
316 | ||
8b712842 | 317 | spin_unlock_irqrestore(&worker->lock, flags); |
75ccf47d | 318 | |
8b712842 CM |
319 | out: |
320 | return 0; | |
321 | } | |
322 | ||
323 | /* | |
324 | * places a struct btrfs_work into the pending queue of one of the kthreads | |
325 | */ | |
326 | int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work) | |
327 | { | |
328 | struct btrfs_worker_thread *worker; | |
329 | unsigned long flags; | |
330 | int wake = 0; | |
331 | ||
332 | /* don't requeue something already on a list */ | |
333 | if (test_and_set_bit(0, &work->flags)) | |
334 | goto out; | |
335 | ||
336 | worker = find_worker(workers); | |
337 | ||
338 | spin_lock_irqsave(&worker->lock, flags); | |
339 | atomic_inc(&worker->num_pending); | |
35d8ba66 | 340 | check_busy_worker(worker); |
8b712842 CM |
341 | list_add_tail(&work->list, &worker->pending); |
342 | ||
343 | /* | |
344 | * avoid calling into wake_up_process if this thread has already | |
345 | * been kicked | |
346 | */ | |
347 | if (!worker->working) | |
348 | wake = 1; | |
349 | worker->working = 1; | |
350 | ||
351 | spin_unlock_irqrestore(&worker->lock, flags); | |
352 | ||
353 | if (wake) | |
354 | wake_up_process(worker->task); | |
355 | out: | |
356 | return 0; | |
357 | } |