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>
24 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
25 # include <linux/freezer.h>
27 # include <linux/sched.h>
30 #include "async-thread.h"
33 * container for the kthread task pointer and the list of pending work
34 * One of these is allocated per thread.
36 struct btrfs_worker_thread {
37 /* pool we belong to */
38 struct btrfs_workers *workers;
40 /* list of struct btrfs_work that are waiting for service */
41 struct list_head pending;
43 /* list of worker threads from struct btrfs_workers */
44 struct list_head worker_list;
47 struct task_struct *task;
49 /* number of things on the pending list */
52 unsigned long sequence;
54 /* protects the pending list. */
57 /* set to non-zero when this thread is already awake and kicking */
60 /* are we currently idle */
65 * helper function to move a thread onto the idle list after it
66 * has finished some requests.
68 static void check_idle_worker(struct btrfs_worker_thread *worker)
70 if (!worker->idle && atomic_read(&worker->num_pending) <
71 worker->workers->idle_thresh / 2) {
73 spin_lock_irqsave(&worker->workers->lock, flags);
75 list_move(&worker->worker_list, &worker->workers->idle_list);
76 spin_unlock_irqrestore(&worker->workers->lock, flags);
81 * helper function to move a thread off the idle list after new
82 * pending work is added.
84 static void check_busy_worker(struct btrfs_worker_thread *worker)
86 if (worker->idle && atomic_read(&worker->num_pending) >=
87 worker->workers->idle_thresh) {
89 spin_lock_irqsave(&worker->workers->lock, flags);
91 list_move_tail(&worker->worker_list,
92 &worker->workers->worker_list);
93 spin_unlock_irqrestore(&worker->workers->lock, flags);
98 * main loop for servicing work items
100 static int worker_loop(void *arg)
102 struct btrfs_worker_thread *worker = arg;
103 struct list_head *cur;
104 struct btrfs_work *work;
106 spin_lock_irq(&worker->lock);
107 while(!list_empty(&worker->pending)) {
108 cur = worker->pending.next;
109 work = list_entry(cur, struct btrfs_work, list);
110 list_del(&work->list);
111 clear_bit(0, &work->flags);
113 work->worker = worker;
114 spin_unlock_irq(&worker->lock);
118 atomic_dec(&worker->num_pending);
119 spin_lock_irq(&worker->lock);
120 check_idle_worker(worker);
123 if (freezing(current)) {
126 set_current_state(TASK_INTERRUPTIBLE);
127 spin_unlock_irq(&worker->lock);
129 __set_current_state(TASK_RUNNING);
131 } while (!kthread_should_stop());
136 * this will wait for all the worker threads to shutdown
138 int btrfs_stop_workers(struct btrfs_workers *workers)
140 struct list_head *cur;
141 struct btrfs_worker_thread *worker;
143 list_splice_init(&workers->idle_list, &workers->worker_list);
144 while(!list_empty(&workers->worker_list)) {
145 cur = workers->worker_list.next;
146 worker = list_entry(cur, struct btrfs_worker_thread,
148 kthread_stop(worker->task);
149 list_del(&worker->worker_list);
156 * simple init on struct btrfs_workers
158 void btrfs_init_workers(struct btrfs_workers *workers, char *name, int max)
160 workers->num_workers = 0;
161 INIT_LIST_HEAD(&workers->worker_list);
162 INIT_LIST_HEAD(&workers->idle_list);
163 spin_lock_init(&workers->lock);
164 workers->max_workers = max;
165 workers->idle_thresh = 32;
166 workers->name = name;
170 * starts new worker threads. This does not enforce the max worker
171 * count in case you need to temporarily go past it.
173 int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
175 struct btrfs_worker_thread *worker;
179 for (i = 0; i < num_workers; i++) {
180 worker = kzalloc(sizeof(*worker), GFP_NOFS);
186 INIT_LIST_HEAD(&worker->pending);
187 INIT_LIST_HEAD(&worker->worker_list);
188 spin_lock_init(&worker->lock);
189 atomic_set(&worker->num_pending, 0);
190 worker->task = kthread_run(worker_loop, worker,
191 "btrfs-%s-%d", workers->name,
192 workers->num_workers + i);
193 worker->workers = workers;
194 if (IS_ERR(worker->task)) {
196 ret = PTR_ERR(worker->task);
200 spin_lock_irq(&workers->lock);
201 list_add_tail(&worker->worker_list, &workers->idle_list);
203 workers->num_workers++;
204 spin_unlock_irq(&workers->lock);
208 btrfs_stop_workers(workers);
213 * run through the list and find a worker thread that doesn't have a lot
214 * to do right now. This can return null if we aren't yet at the thread
215 * count limit and all of the threads are busy.
217 static struct btrfs_worker_thread *next_worker(struct btrfs_workers *workers)
219 struct btrfs_worker_thread *worker;
220 struct list_head *next;
221 int enforce_min = workers->num_workers < workers->max_workers;
224 * if we find an idle thread, don't move it to the end of the
225 * idle list. This improves the chance that the next submission
226 * will reuse the same thread, and maybe catch it while it is still
229 if (!list_empty(&workers->idle_list)) {
230 next = workers->idle_list.next;
231 worker = list_entry(next, struct btrfs_worker_thread,
235 if (enforce_min || list_empty(&workers->worker_list))
239 * if we pick a busy task, move the task to the end of the list.
240 * hopefully this will keep things somewhat evenly balanced
242 next = workers->worker_list.next;
243 worker = list_entry(next, struct btrfs_worker_thread, worker_list);
244 atomic_inc(&worker->num_pending);
246 if (worker->sequence % workers->idle_thresh == 0)
247 list_move_tail(next, &workers->worker_list);
251 static struct btrfs_worker_thread *find_worker(struct btrfs_workers *workers)
253 struct btrfs_worker_thread *worker;
257 spin_lock_irqsave(&workers->lock, flags);
258 worker = next_worker(workers);
259 spin_unlock_irqrestore(&workers->lock, flags);
262 spin_lock_irqsave(&workers->lock, flags);
263 if (workers->num_workers >= workers->max_workers) {
264 struct list_head *fallback = NULL;
266 * we have failed to find any workers, just
267 * return the force one
269 if (!list_empty(&workers->worker_list))
270 fallback = workers->worker_list.next;
271 if (!list_empty(&workers->idle_list))
272 fallback = workers->idle_list.next;
274 worker = list_entry(fallback,
275 struct btrfs_worker_thread, worker_list);
276 spin_unlock_irqrestore(&workers->lock, flags);
278 spin_unlock_irqrestore(&workers->lock, flags);
279 /* we're below the limit, start another worker */
280 btrfs_start_workers(workers, 1);
288 * btrfs_requeue_work just puts the work item back on the tail of the list
289 * it was taken from. It is intended for use with long running work functions
290 * that make some progress and want to give the cpu up for others.
292 int btrfs_requeue_work(struct btrfs_work *work)
294 struct btrfs_worker_thread *worker = work->worker;
297 if (test_and_set_bit(0, &work->flags))
300 spin_lock_irqsave(&worker->lock, flags);
301 atomic_inc(&worker->num_pending);
302 list_add_tail(&work->list, &worker->pending);
303 check_busy_worker(worker);
304 spin_unlock_irqrestore(&worker->lock, flags);
310 * places a struct btrfs_work into the pending queue of one of the kthreads
312 int btrfs_queue_worker(struct btrfs_workers *workers, struct btrfs_work *work)
314 struct btrfs_worker_thread *worker;
318 /* don't requeue something already on a list */
319 if (test_and_set_bit(0, &work->flags))
322 worker = find_worker(workers);
324 spin_lock_irqsave(&worker->lock, flags);
325 atomic_inc(&worker->num_pending);
326 check_busy_worker(worker);
327 list_add_tail(&work->list, &worker->pending);
330 * avoid calling into wake_up_process if this thread has already
333 if (!worker->working)
337 spin_unlock_irqrestore(&worker->lock, flags);
340 wake_up_process(worker->task);