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/version.h>
20 #include <linux/kthread.h>
21 #include <linux/list.h>
22 #include <linux/spinlock.h>
23 # include <linux/freezer.h>
24 #include "async-thread.h"
27 * container for the kthread task pointer and the list of pending work
28 * One of these is allocated per thread.
30 struct btrfs_worker_thread {
31 /* pool we belong to */
32 struct btrfs_workers *workers;
34 /* list of struct btrfs_work that are waiting for service */
35 struct list_head pending;
37 /* list of worker threads from struct btrfs_workers */
38 struct list_head worker_list;
41 struct task_struct *task;
43 /* number of things on the pending list */
46 unsigned long sequence;
48 /* protects the pending list. */
51 /* set to non-zero when this thread is already awake and kicking */
54 /* are we currently idle */
59 * helper function to move a thread onto the idle list after it
60 * has finished some requests.
62 static void check_idle_worker(struct btrfs_worker_thread *worker)
64 if (!worker->idle && atomic_read(&worker->num_pending) <
65 worker->workers->idle_thresh / 2) {
67 spin_lock_irqsave(&worker->workers->lock, flags);
69 list_move(&worker->worker_list, &worker->workers->idle_list);
70 spin_unlock_irqrestore(&worker->workers->lock, flags);
75 * helper function to move a thread off the idle list after new
76 * pending work is added.
78 static void check_busy_worker(struct btrfs_worker_thread *worker)
80 if (worker->idle && atomic_read(&worker->num_pending) >=
81 worker->workers->idle_thresh) {
83 spin_lock_irqsave(&worker->workers->lock, flags);
85 list_move_tail(&worker->worker_list,
86 &worker->workers->worker_list);
87 spin_unlock_irqrestore(&worker->workers->lock, flags);
92 * main loop for servicing work items
94 static int worker_loop(void *arg)
96 struct btrfs_worker_thread *worker = arg;
97 struct list_head *cur;
98 struct btrfs_work *work;
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);
107 work->worker = worker;
108 spin_unlock_irq(&worker->lock);
112 atomic_dec(&worker->num_pending);
113 spin_lock_irq(&worker->lock);
114 check_idle_worker(worker);
117 if (freezing(current)) {
120 set_current_state(TASK_INTERRUPTIBLE);
121 spin_unlock_irq(&worker->lock);
123 __set_current_state(TASK_RUNNING);
125 } while (!kthread_should_stop());
130 * this will wait for all the worker threads to shutdown
132 int btrfs_stop_workers(struct btrfs_workers *workers)
134 struct list_head *cur;
135 struct btrfs_worker_thread *worker;
137 list_splice_init(&workers->idle_list, &workers->worker_list);
138 while(!list_empty(&workers->worker_list)) {
139 cur = workers->worker_list.next;
140 worker = list_entry(cur, struct btrfs_worker_thread,
142 kthread_stop(worker->task);
143 list_del(&worker->worker_list);
150 * simple init on struct btrfs_workers
152 void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max)
154 workers->num_workers = 0;
155 INIT_LIST_HEAD(&workers->worker_list);
156 INIT_LIST_HEAD(&workers->idle_list);
157 spin_lock_init(&workers->lock);
158 workers->max_workers = max;
159 workers->idle_thresh = 32;
160 workers->name = name;
164 * starts new worker threads. This does not enforce the max worker
165 * count in case you need to temporarily go past it.
167 int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
169 struct btrfs_worker_thread *worker;
173 for (i = 0; i < num_workers; i++) {
174 worker = kzalloc(sizeof(*worker), GFP_NOFS);
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);
184 worker->task = kthread_run(worker_loop, worker,
185 "btrfs-%s-%d", workers->name,
186 workers->num_workers + i);
187 worker->workers = workers;
188 if (IS_ERR(worker->task)) {
190 ret = PTR_ERR(worker->task);
194 spin_lock_irq(&workers->lock);
195 list_add_tail(&worker->worker_list, &workers->idle_list);
197 workers->num_workers++;
198 spin_unlock_irq(&workers->lock);
202 btrfs_stop_workers(workers);
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.
211 static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers)
213 struct btrfs_worker_thread *worker;
214 struct list_head *next;
215 int enforce_min = workers->num_workers < workers->max_workers;
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
223 if (!list_empty(&workers->idle_list)) {
224 next = workers->idle_list.next;
225 worker = list_entry(next, struct btrfs_worker_thread,
229 if (enforce_min || list_empty(&workers->worker_list))
233 * if we pick a busy task, move the task to the end of the list.
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.
238 next = workers->worker_list.next;
239 worker = list_entry(next, struct btrfs_worker_thread, worker_list);
240 atomic_inc(&worker->num_pending);
243 if (worker->sequence % workers->idle_thresh == 0)
244 list_move_tail(next, &workers->worker_list);
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.
253 static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers)
255 struct btrfs_worker_thread *worker;
259 spin_lock_irqsave(&workers->lock, flags);
260 worker = next_worker(workers);
261 spin_unlock_irqrestore(&workers->lock, flags);
264 spin_lock_irqsave(&workers->lock, flags);
265 if (workers->num_workers >= workers->max_workers) {
266 struct list_head *fallback = NULL;
268 * we have failed to find any workers, just
269 * return the force one
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;
276 worker = list_entry(fallback,
277 struct btrfs_worker_thread, worker_list);
278 spin_unlock_irqrestore(&workers->lock, flags);
280 spin_unlock_irqrestore(&workers->lock, flags);
281 /* we're below the limit, start another worker */
282 btrfs_start_workers(workers, 1);
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.
294 int btrfs_requeue_work(struct btrfs_work *work)
296 struct btrfs_worker_thread *worker = work->worker;
299 if (test_and_set_bit(0, &work->flags))
302 spin_lock_irqsave(&worker->lock, flags);
303 atomic_inc(&worker->num_pending);
304 list_add_tail(&work->list, &worker->pending);
305 check_busy_worker(worker);
306 spin_unlock_irqrestore(&worker->lock, flags);
312 * places a struct btrfs_work into the pending queue of one of the kthreads
314 int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
316 struct btrfs_worker_thread *worker;
320 /* don't requeue something already on a list */
321 if (test_and_set_bit(0, &work->flags))
324 worker = find_worker(workers);
326 spin_lock_irqsave(&worker->lock, flags);
327 atomic_inc(&worker->num_pending);
328 check_busy_worker(worker);
329 list_add_tail(&work->list, &worker->pending);
332 * avoid calling into wake_up_process if this thread has already
335 if (!worker->working)
339 spin_unlock_irqrestore(&worker->lock, flags);
342 wake_up_process(worker->task);