Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/kernel/workqueue.c | |
3 | * | |
4 | * Generic mechanism for defining kernel helper threads for running | |
5 | * arbitrary tasks in process context. | |
6 | * | |
7 | * Started by Ingo Molnar, Copyright (C) 2002 | |
8 | * | |
9 | * Derived from the taskqueue/keventd code by: | |
10 | * | |
11 | * David Woodhouse <dwmw2@infradead.org> | |
12 | * Andrew Morton <andrewm@uow.edu.au> | |
13 | * Kai Petzke <wpp@marie.physik.tu-berlin.de> | |
14 | * Theodore Ts'o <tytso@mit.edu> | |
89ada679 | 15 | * |
cde53535 | 16 | * Made to use alloc_percpu by Christoph Lameter. |
1da177e4 LT |
17 | */ |
18 | ||
19 | #include <linux/module.h> | |
20 | #include <linux/kernel.h> | |
21 | #include <linux/sched.h> | |
22 | #include <linux/init.h> | |
23 | #include <linux/signal.h> | |
24 | #include <linux/completion.h> | |
25 | #include <linux/workqueue.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/cpu.h> | |
28 | #include <linux/notifier.h> | |
29 | #include <linux/kthread.h> | |
1fa44eca | 30 | #include <linux/hardirq.h> |
46934023 | 31 | #include <linux/mempolicy.h> |
341a5958 | 32 | #include <linux/freezer.h> |
d5abe669 PZ |
33 | #include <linux/kallsyms.h> |
34 | #include <linux/debug_locks.h> | |
4e6045f1 | 35 | #include <linux/lockdep.h> |
1da177e4 LT |
36 | |
37 | /* | |
f756d5e2 NL |
38 | * The per-CPU workqueue (if single thread, we always use the first |
39 | * possible cpu). | |
1da177e4 LT |
40 | */ |
41 | struct cpu_workqueue_struct { | |
42 | ||
43 | spinlock_t lock; | |
44 | ||
1da177e4 LT |
45 | struct list_head worklist; |
46 | wait_queue_head_t more_work; | |
3af24433 | 47 | struct work_struct *current_work; |
1da177e4 LT |
48 | |
49 | struct workqueue_struct *wq; | |
36c8b586 | 50 | struct task_struct *thread; |
1da177e4 LT |
51 | |
52 | int run_depth; /* Detect run_workqueue() recursion depth */ | |
53 | } ____cacheline_aligned; | |
54 | ||
55 | /* | |
56 | * The externally visible workqueue abstraction is an array of | |
57 | * per-CPU workqueues: | |
58 | */ | |
59 | struct workqueue_struct { | |
89ada679 | 60 | struct cpu_workqueue_struct *cpu_wq; |
cce1a165 | 61 | struct list_head list; |
1da177e4 | 62 | const char *name; |
cce1a165 | 63 | int singlethread; |
319c2a98 | 64 | int freezeable; /* Freeze threads during suspend */ |
4e6045f1 JB |
65 | #ifdef CONFIG_LOCKDEP |
66 | struct lockdep_map lockdep_map; | |
67 | #endif | |
1da177e4 LT |
68 | }; |
69 | ||
95402b38 GS |
70 | /* Serializes the accesses to the list of workqueues. */ |
71 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 LT |
72 | static LIST_HEAD(workqueues); |
73 | ||
3af24433 | 74 | static int singlethread_cpu __read_mostly; |
b1f4ec17 | 75 | static cpumask_t cpu_singlethread_map __read_mostly; |
14441960 ON |
76 | /* |
77 | * _cpu_down() first removes CPU from cpu_online_map, then CPU_DEAD | |
78 | * flushes cwq->worklist. This means that flush_workqueue/wait_on_work | |
79 | * which comes in between can't use for_each_online_cpu(). We could | |
80 | * use cpu_possible_map, the cpumask below is more a documentation | |
81 | * than optimization. | |
82 | */ | |
3af24433 | 83 | static cpumask_t cpu_populated_map __read_mostly; |
f756d5e2 | 84 | |
1da177e4 LT |
85 | /* If it's single threaded, it isn't in the list of workqueues. */ |
86 | static inline int is_single_threaded(struct workqueue_struct *wq) | |
87 | { | |
cce1a165 | 88 | return wq->singlethread; |
1da177e4 LT |
89 | } |
90 | ||
b1f4ec17 ON |
91 | static const cpumask_t *wq_cpu_map(struct workqueue_struct *wq) |
92 | { | |
93 | return is_single_threaded(wq) | |
94 | ? &cpu_singlethread_map : &cpu_populated_map; | |
95 | } | |
96 | ||
a848e3b6 ON |
97 | static |
98 | struct cpu_workqueue_struct *wq_per_cpu(struct workqueue_struct *wq, int cpu) | |
99 | { | |
100 | if (unlikely(is_single_threaded(wq))) | |
101 | cpu = singlethread_cpu; | |
102 | return per_cpu_ptr(wq->cpu_wq, cpu); | |
103 | } | |
104 | ||
4594bf15 DH |
105 | /* |
106 | * Set the workqueue on which a work item is to be run | |
107 | * - Must *only* be called if the pending flag is set | |
108 | */ | |
ed7c0fee ON |
109 | static inline void set_wq_data(struct work_struct *work, |
110 | struct cpu_workqueue_struct *cwq) | |
365970a1 | 111 | { |
4594bf15 DH |
112 | unsigned long new; |
113 | ||
114 | BUG_ON(!work_pending(work)); | |
365970a1 | 115 | |
ed7c0fee | 116 | new = (unsigned long) cwq | (1UL << WORK_STRUCT_PENDING); |
a08727ba LT |
117 | new |= WORK_STRUCT_FLAG_MASK & *work_data_bits(work); |
118 | atomic_long_set(&work->data, new); | |
365970a1 DH |
119 | } |
120 | ||
ed7c0fee ON |
121 | static inline |
122 | struct cpu_workqueue_struct *get_wq_data(struct work_struct *work) | |
365970a1 | 123 | { |
a08727ba | 124 | return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK); |
365970a1 DH |
125 | } |
126 | ||
b89deed3 | 127 | static void insert_work(struct cpu_workqueue_struct *cwq, |
1a4d9b0a | 128 | struct work_struct *work, struct list_head *head) |
b89deed3 ON |
129 | { |
130 | set_wq_data(work, cwq); | |
6e84d644 ON |
131 | /* |
132 | * Ensure that we get the right work->data if we see the | |
133 | * result of list_add() below, see try_to_grab_pending(). | |
134 | */ | |
135 | smp_wmb(); | |
1a4d9b0a | 136 | list_add_tail(&work->entry, head); |
b89deed3 ON |
137 | wake_up(&cwq->more_work); |
138 | } | |
139 | ||
1da177e4 LT |
140 | static void __queue_work(struct cpu_workqueue_struct *cwq, |
141 | struct work_struct *work) | |
142 | { | |
143 | unsigned long flags; | |
144 | ||
145 | spin_lock_irqsave(&cwq->lock, flags); | |
1a4d9b0a | 146 | insert_work(cwq, work, &cwq->worklist); |
1da177e4 LT |
147 | spin_unlock_irqrestore(&cwq->lock, flags); |
148 | } | |
149 | ||
0fcb78c2 REB |
150 | /** |
151 | * queue_work - queue work on a workqueue | |
152 | * @wq: workqueue to use | |
153 | * @work: work to queue | |
154 | * | |
057647fc | 155 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
1da177e4 | 156 | * |
00dfcaf7 ON |
157 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
158 | * it can be processed by another CPU. | |
1da177e4 | 159 | */ |
7ad5b3a5 | 160 | int queue_work(struct workqueue_struct *wq, struct work_struct *work) |
1da177e4 | 161 | { |
a848e3b6 | 162 | int ret = 0; |
1da177e4 | 163 | |
a08727ba | 164 | if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) { |
1da177e4 | 165 | BUG_ON(!list_empty(&work->entry)); |
a848e3b6 ON |
166 | __queue_work(wq_per_cpu(wq, get_cpu()), work); |
167 | put_cpu(); | |
1da177e4 LT |
168 | ret = 1; |
169 | } | |
1da177e4 LT |
170 | return ret; |
171 | } | |
ae90dd5d | 172 | EXPORT_SYMBOL_GPL(queue_work); |
1da177e4 | 173 | |
c1a220e7 ZR |
174 | /** |
175 | * queue_work_on - queue work on specific cpu | |
176 | * @cpu: CPU number to execute work on | |
177 | * @wq: workqueue to use | |
178 | * @work: work to queue | |
179 | * | |
180 | * Returns 0 if @work was already on a queue, non-zero otherwise. | |
181 | * | |
182 | * We queue the work to a specific CPU, the caller must ensure it | |
183 | * can't go away. | |
184 | */ | |
185 | int | |
186 | queue_work_on(int cpu, struct workqueue_struct *wq, struct work_struct *work) | |
187 | { | |
188 | int ret = 0; | |
189 | ||
190 | if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) { | |
191 | BUG_ON(!list_empty(&work->entry)); | |
192 | __queue_work(wq_per_cpu(wq, cpu), work); | |
193 | ret = 1; | |
194 | } | |
195 | return ret; | |
196 | } | |
197 | EXPORT_SYMBOL_GPL(queue_work_on); | |
198 | ||
6d141c3f | 199 | static void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 200 | { |
52bad64d | 201 | struct delayed_work *dwork = (struct delayed_work *)__data; |
ed7c0fee ON |
202 | struct cpu_workqueue_struct *cwq = get_wq_data(&dwork->work); |
203 | struct workqueue_struct *wq = cwq->wq; | |
1da177e4 | 204 | |
a848e3b6 | 205 | __queue_work(wq_per_cpu(wq, smp_processor_id()), &dwork->work); |
1da177e4 LT |
206 | } |
207 | ||
0fcb78c2 REB |
208 | /** |
209 | * queue_delayed_work - queue work on a workqueue after delay | |
210 | * @wq: workqueue to use | |
af9997e4 | 211 | * @dwork: delayable work to queue |
0fcb78c2 REB |
212 | * @delay: number of jiffies to wait before queueing |
213 | * | |
057647fc | 214 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 215 | */ |
7ad5b3a5 | 216 | int queue_delayed_work(struct workqueue_struct *wq, |
52bad64d | 217 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 218 | { |
52bad64d | 219 | if (delay == 0) |
63bc0362 | 220 | return queue_work(wq, &dwork->work); |
1da177e4 | 221 | |
63bc0362 | 222 | return queue_delayed_work_on(-1, wq, dwork, delay); |
1da177e4 | 223 | } |
ae90dd5d | 224 | EXPORT_SYMBOL_GPL(queue_delayed_work); |
1da177e4 | 225 | |
0fcb78c2 REB |
226 | /** |
227 | * queue_delayed_work_on - queue work on specific CPU after delay | |
228 | * @cpu: CPU number to execute work on | |
229 | * @wq: workqueue to use | |
af9997e4 | 230 | * @dwork: work to queue |
0fcb78c2 REB |
231 | * @delay: number of jiffies to wait before queueing |
232 | * | |
057647fc | 233 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 234 | */ |
7a6bc1cd | 235 | int queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
52bad64d | 236 | struct delayed_work *dwork, unsigned long delay) |
7a6bc1cd VP |
237 | { |
238 | int ret = 0; | |
52bad64d DH |
239 | struct timer_list *timer = &dwork->timer; |
240 | struct work_struct *work = &dwork->work; | |
7a6bc1cd | 241 | |
a08727ba | 242 | if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) { |
7a6bc1cd VP |
243 | BUG_ON(timer_pending(timer)); |
244 | BUG_ON(!list_empty(&work->entry)); | |
245 | ||
8a3e77cc AL |
246 | timer_stats_timer_set_start_info(&dwork->timer); |
247 | ||
ed7c0fee | 248 | /* This stores cwq for the moment, for the timer_fn */ |
a848e3b6 | 249 | set_wq_data(work, wq_per_cpu(wq, raw_smp_processor_id())); |
7a6bc1cd | 250 | timer->expires = jiffies + delay; |
52bad64d | 251 | timer->data = (unsigned long)dwork; |
7a6bc1cd | 252 | timer->function = delayed_work_timer_fn; |
63bc0362 ON |
253 | |
254 | if (unlikely(cpu >= 0)) | |
255 | add_timer_on(timer, cpu); | |
256 | else | |
257 | add_timer(timer); | |
7a6bc1cd VP |
258 | ret = 1; |
259 | } | |
260 | return ret; | |
261 | } | |
ae90dd5d | 262 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
1da177e4 | 263 | |
858119e1 | 264 | static void run_workqueue(struct cpu_workqueue_struct *cwq) |
1da177e4 | 265 | { |
f293ea92 | 266 | spin_lock_irq(&cwq->lock); |
1da177e4 LT |
267 | cwq->run_depth++; |
268 | if (cwq->run_depth > 3) { | |
269 | /* morton gets to eat his hat */ | |
270 | printk("%s: recursion depth exceeded: %d\n", | |
af1f16d0 | 271 | __func__, cwq->run_depth); |
1da177e4 LT |
272 | dump_stack(); |
273 | } | |
274 | while (!list_empty(&cwq->worklist)) { | |
275 | struct work_struct *work = list_entry(cwq->worklist.next, | |
276 | struct work_struct, entry); | |
6bb49e59 | 277 | work_func_t f = work->func; |
4e6045f1 JB |
278 | #ifdef CONFIG_LOCKDEP |
279 | /* | |
280 | * It is permissible to free the struct work_struct | |
281 | * from inside the function that is called from it, | |
282 | * this we need to take into account for lockdep too. | |
283 | * To avoid bogus "held lock freed" warnings as well | |
284 | * as problems when looking into work->lockdep_map, | |
285 | * make a copy and use that here. | |
286 | */ | |
287 | struct lockdep_map lockdep_map = work->lockdep_map; | |
288 | #endif | |
1da177e4 | 289 | |
b89deed3 | 290 | cwq->current_work = work; |
1da177e4 | 291 | list_del_init(cwq->worklist.next); |
f293ea92 | 292 | spin_unlock_irq(&cwq->lock); |
1da177e4 | 293 | |
365970a1 | 294 | BUG_ON(get_wq_data(work) != cwq); |
23b2e599 | 295 | work_clear_pending(work); |
4e6045f1 JB |
296 | lock_acquire(&cwq->wq->lockdep_map, 0, 0, 0, 2, _THIS_IP_); |
297 | lock_acquire(&lockdep_map, 0, 0, 0, 2, _THIS_IP_); | |
65f27f38 | 298 | f(work); |
4e6045f1 JB |
299 | lock_release(&lockdep_map, 1, _THIS_IP_); |
300 | lock_release(&cwq->wq->lockdep_map, 1, _THIS_IP_); | |
1da177e4 | 301 | |
d5abe669 PZ |
302 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { |
303 | printk(KERN_ERR "BUG: workqueue leaked lock or atomic: " | |
304 | "%s/0x%08x/%d\n", | |
305 | current->comm, preempt_count(), | |
ba25f9dc | 306 | task_pid_nr(current)); |
d5abe669 PZ |
307 | printk(KERN_ERR " last function: "); |
308 | print_symbol("%s\n", (unsigned long)f); | |
309 | debug_show_held_locks(current); | |
310 | dump_stack(); | |
311 | } | |
312 | ||
f293ea92 | 313 | spin_lock_irq(&cwq->lock); |
b89deed3 | 314 | cwq->current_work = NULL; |
1da177e4 LT |
315 | } |
316 | cwq->run_depth--; | |
f293ea92 | 317 | spin_unlock_irq(&cwq->lock); |
1da177e4 LT |
318 | } |
319 | ||
320 | static int worker_thread(void *__cwq) | |
321 | { | |
322 | struct cpu_workqueue_struct *cwq = __cwq; | |
3af24433 | 323 | DEFINE_WAIT(wait); |
1da177e4 | 324 | |
83144186 RW |
325 | if (cwq->wq->freezeable) |
326 | set_freezable(); | |
1da177e4 LT |
327 | |
328 | set_user_nice(current, -5); | |
1da177e4 | 329 | |
3af24433 | 330 | for (;;) { |
3af24433 | 331 | prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE); |
14441960 ON |
332 | if (!freezing(current) && |
333 | !kthread_should_stop() && | |
334 | list_empty(&cwq->worklist)) | |
1da177e4 | 335 | schedule(); |
3af24433 ON |
336 | finish_wait(&cwq->more_work, &wait); |
337 | ||
85f4186a ON |
338 | try_to_freeze(); |
339 | ||
14441960 | 340 | if (kthread_should_stop()) |
3af24433 | 341 | break; |
1da177e4 | 342 | |
3af24433 | 343 | run_workqueue(cwq); |
1da177e4 | 344 | } |
3af24433 | 345 | |
1da177e4 LT |
346 | return 0; |
347 | } | |
348 | ||
fc2e4d70 ON |
349 | struct wq_barrier { |
350 | struct work_struct work; | |
351 | struct completion done; | |
352 | }; | |
353 | ||
354 | static void wq_barrier_func(struct work_struct *work) | |
355 | { | |
356 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
357 | complete(&barr->done); | |
358 | } | |
359 | ||
83c22520 | 360 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, |
1a4d9b0a | 361 | struct wq_barrier *barr, struct list_head *head) |
fc2e4d70 ON |
362 | { |
363 | INIT_WORK(&barr->work, wq_barrier_func); | |
364 | __set_bit(WORK_STRUCT_PENDING, work_data_bits(&barr->work)); | |
365 | ||
366 | init_completion(&barr->done); | |
83c22520 | 367 | |
1a4d9b0a | 368 | insert_work(cwq, &barr->work, head); |
fc2e4d70 ON |
369 | } |
370 | ||
14441960 | 371 | static int flush_cpu_workqueue(struct cpu_workqueue_struct *cwq) |
1da177e4 | 372 | { |
14441960 ON |
373 | int active; |
374 | ||
1da177e4 LT |
375 | if (cwq->thread == current) { |
376 | /* | |
377 | * Probably keventd trying to flush its own queue. So simply run | |
378 | * it by hand rather than deadlocking. | |
379 | */ | |
380 | run_workqueue(cwq); | |
14441960 | 381 | active = 1; |
1da177e4 | 382 | } else { |
fc2e4d70 | 383 | struct wq_barrier barr; |
1da177e4 | 384 | |
14441960 | 385 | active = 0; |
83c22520 ON |
386 | spin_lock_irq(&cwq->lock); |
387 | if (!list_empty(&cwq->worklist) || cwq->current_work != NULL) { | |
1a4d9b0a | 388 | insert_wq_barrier(cwq, &barr, &cwq->worklist); |
83c22520 ON |
389 | active = 1; |
390 | } | |
391 | spin_unlock_irq(&cwq->lock); | |
1da177e4 | 392 | |
d721304d | 393 | if (active) |
83c22520 | 394 | wait_for_completion(&barr.done); |
1da177e4 | 395 | } |
14441960 ON |
396 | |
397 | return active; | |
1da177e4 LT |
398 | } |
399 | ||
0fcb78c2 | 400 | /** |
1da177e4 | 401 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 402 | * @wq: workqueue to flush |
1da177e4 LT |
403 | * |
404 | * Forces execution of the workqueue and blocks until its completion. | |
405 | * This is typically used in driver shutdown handlers. | |
406 | * | |
fc2e4d70 ON |
407 | * We sleep until all works which were queued on entry have been handled, |
408 | * but we are not livelocked by new incoming ones. | |
1da177e4 LT |
409 | * |
410 | * This function used to run the workqueues itself. Now we just wait for the | |
411 | * helper threads to do it. | |
412 | */ | |
7ad5b3a5 | 413 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 414 | { |
b1f4ec17 | 415 | const cpumask_t *cpu_map = wq_cpu_map(wq); |
cce1a165 | 416 | int cpu; |
1da177e4 | 417 | |
b1f4ec17 | 418 | might_sleep(); |
4e6045f1 JB |
419 | lock_acquire(&wq->lockdep_map, 0, 0, 0, 2, _THIS_IP_); |
420 | lock_release(&wq->lockdep_map, 1, _THIS_IP_); | |
363ab6f1 | 421 | for_each_cpu_mask_nr(cpu, *cpu_map) |
b1f4ec17 | 422 | flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu)); |
1da177e4 | 423 | } |
ae90dd5d | 424 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 425 | |
db700897 ON |
426 | /** |
427 | * flush_work - block until a work_struct's callback has terminated | |
428 | * @work: the work which is to be flushed | |
429 | * | |
430 | * It is expected that, prior to calling flush_work(), the caller has | |
431 | * arranged for the work to not be requeued, otherwise it doesn't make | |
432 | * sense to use this function. | |
433 | */ | |
434 | int flush_work(struct work_struct *work) | |
435 | { | |
436 | struct cpu_workqueue_struct *cwq; | |
437 | struct list_head *prev; | |
438 | struct wq_barrier barr; | |
439 | ||
440 | might_sleep(); | |
441 | cwq = get_wq_data(work); | |
442 | if (!cwq) | |
443 | return 0; | |
444 | ||
445 | prev = NULL; | |
446 | spin_lock_irq(&cwq->lock); | |
447 | if (!list_empty(&work->entry)) { | |
448 | /* | |
449 | * See the comment near try_to_grab_pending()->smp_rmb(). | |
450 | * If it was re-queued under us we are not going to wait. | |
451 | */ | |
452 | smp_rmb(); | |
453 | if (unlikely(cwq != get_wq_data(work))) | |
454 | goto out; | |
455 | prev = &work->entry; | |
456 | } else { | |
457 | if (cwq->current_work != work) | |
458 | goto out; | |
459 | prev = &cwq->worklist; | |
460 | } | |
461 | insert_wq_barrier(cwq, &barr, prev->next); | |
462 | out: | |
463 | spin_unlock_irq(&cwq->lock); | |
464 | if (!prev) | |
465 | return 0; | |
466 | ||
467 | wait_for_completion(&barr.done); | |
468 | return 1; | |
469 | } | |
470 | EXPORT_SYMBOL_GPL(flush_work); | |
471 | ||
6e84d644 | 472 | /* |
1f1f642e | 473 | * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit, |
6e84d644 ON |
474 | * so this work can't be re-armed in any way. |
475 | */ | |
476 | static int try_to_grab_pending(struct work_struct *work) | |
477 | { | |
478 | struct cpu_workqueue_struct *cwq; | |
1f1f642e | 479 | int ret = -1; |
6e84d644 ON |
480 | |
481 | if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) | |
1f1f642e | 482 | return 0; |
6e84d644 ON |
483 | |
484 | /* | |
485 | * The queueing is in progress, or it is already queued. Try to | |
486 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
487 | */ | |
488 | ||
489 | cwq = get_wq_data(work); | |
490 | if (!cwq) | |
491 | return ret; | |
492 | ||
493 | spin_lock_irq(&cwq->lock); | |
494 | if (!list_empty(&work->entry)) { | |
495 | /* | |
496 | * This work is queued, but perhaps we locked the wrong cwq. | |
497 | * In that case we must see the new value after rmb(), see | |
498 | * insert_work()->wmb(). | |
499 | */ | |
500 | smp_rmb(); | |
501 | if (cwq == get_wq_data(work)) { | |
502 | list_del_init(&work->entry); | |
503 | ret = 1; | |
504 | } | |
505 | } | |
506 | spin_unlock_irq(&cwq->lock); | |
507 | ||
508 | return ret; | |
509 | } | |
510 | ||
511 | static void wait_on_cpu_work(struct cpu_workqueue_struct *cwq, | |
b89deed3 ON |
512 | struct work_struct *work) |
513 | { | |
514 | struct wq_barrier barr; | |
515 | int running = 0; | |
516 | ||
517 | spin_lock_irq(&cwq->lock); | |
518 | if (unlikely(cwq->current_work == work)) { | |
1a4d9b0a | 519 | insert_wq_barrier(cwq, &barr, cwq->worklist.next); |
b89deed3 ON |
520 | running = 1; |
521 | } | |
522 | spin_unlock_irq(&cwq->lock); | |
523 | ||
3af24433 | 524 | if (unlikely(running)) |
b89deed3 | 525 | wait_for_completion(&barr.done); |
b89deed3 ON |
526 | } |
527 | ||
6e84d644 | 528 | static void wait_on_work(struct work_struct *work) |
b89deed3 ON |
529 | { |
530 | struct cpu_workqueue_struct *cwq; | |
28e53bdd ON |
531 | struct workqueue_struct *wq; |
532 | const cpumask_t *cpu_map; | |
b1f4ec17 | 533 | int cpu; |
b89deed3 | 534 | |
f293ea92 ON |
535 | might_sleep(); |
536 | ||
4e6045f1 JB |
537 | lock_acquire(&work->lockdep_map, 0, 0, 0, 2, _THIS_IP_); |
538 | lock_release(&work->lockdep_map, 1, _THIS_IP_); | |
539 | ||
b89deed3 | 540 | cwq = get_wq_data(work); |
b89deed3 | 541 | if (!cwq) |
3af24433 | 542 | return; |
b89deed3 | 543 | |
28e53bdd ON |
544 | wq = cwq->wq; |
545 | cpu_map = wq_cpu_map(wq); | |
546 | ||
363ab6f1 | 547 | for_each_cpu_mask_nr(cpu, *cpu_map) |
6e84d644 ON |
548 | wait_on_cpu_work(per_cpu_ptr(wq->cpu_wq, cpu), work); |
549 | } | |
550 | ||
1f1f642e ON |
551 | static int __cancel_work_timer(struct work_struct *work, |
552 | struct timer_list* timer) | |
553 | { | |
554 | int ret; | |
555 | ||
556 | do { | |
557 | ret = (timer && likely(del_timer(timer))); | |
558 | if (!ret) | |
559 | ret = try_to_grab_pending(work); | |
560 | wait_on_work(work); | |
561 | } while (unlikely(ret < 0)); | |
562 | ||
563 | work_clear_pending(work); | |
564 | return ret; | |
565 | } | |
566 | ||
6e84d644 ON |
567 | /** |
568 | * cancel_work_sync - block until a work_struct's callback has terminated | |
569 | * @work: the work which is to be flushed | |
570 | * | |
1f1f642e ON |
571 | * Returns true if @work was pending. |
572 | * | |
6e84d644 ON |
573 | * cancel_work_sync() will cancel the work if it is queued. If the work's |
574 | * callback appears to be running, cancel_work_sync() will block until it | |
575 | * has completed. | |
576 | * | |
577 | * It is possible to use this function if the work re-queues itself. It can | |
578 | * cancel the work even if it migrates to another workqueue, however in that | |
579 | * case it only guarantees that work->func() has completed on the last queued | |
580 | * workqueue. | |
581 | * | |
582 | * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not | |
583 | * pending, otherwise it goes into a busy-wait loop until the timer expires. | |
584 | * | |
585 | * The caller must ensure that workqueue_struct on which this work was last | |
586 | * queued can't be destroyed before this function returns. | |
587 | */ | |
1f1f642e | 588 | int cancel_work_sync(struct work_struct *work) |
6e84d644 | 589 | { |
1f1f642e | 590 | return __cancel_work_timer(work, NULL); |
b89deed3 | 591 | } |
28e53bdd | 592 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 593 | |
6e84d644 | 594 | /** |
f5a421a4 | 595 | * cancel_delayed_work_sync - reliably kill off a delayed work. |
6e84d644 ON |
596 | * @dwork: the delayed work struct |
597 | * | |
1f1f642e ON |
598 | * Returns true if @dwork was pending. |
599 | * | |
6e84d644 ON |
600 | * It is possible to use this function if @dwork rearms itself via queue_work() |
601 | * or queue_delayed_work(). See also the comment for cancel_work_sync(). | |
602 | */ | |
1f1f642e | 603 | int cancel_delayed_work_sync(struct delayed_work *dwork) |
6e84d644 | 604 | { |
1f1f642e | 605 | return __cancel_work_timer(&dwork->work, &dwork->timer); |
6e84d644 | 606 | } |
f5a421a4 | 607 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 608 | |
6e84d644 | 609 | static struct workqueue_struct *keventd_wq __read_mostly; |
1da177e4 | 610 | |
0fcb78c2 REB |
611 | /** |
612 | * schedule_work - put work task in global workqueue | |
613 | * @work: job to be done | |
614 | * | |
615 | * This puts a job in the kernel-global workqueue. | |
616 | */ | |
7ad5b3a5 | 617 | int schedule_work(struct work_struct *work) |
1da177e4 LT |
618 | { |
619 | return queue_work(keventd_wq, work); | |
620 | } | |
ae90dd5d | 621 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 622 | |
c1a220e7 ZR |
623 | /* |
624 | * schedule_work_on - put work task on a specific cpu | |
625 | * @cpu: cpu to put the work task on | |
626 | * @work: job to be done | |
627 | * | |
628 | * This puts a job on a specific cpu | |
629 | */ | |
630 | int schedule_work_on(int cpu, struct work_struct *work) | |
631 | { | |
632 | return queue_work_on(cpu, keventd_wq, work); | |
633 | } | |
634 | EXPORT_SYMBOL(schedule_work_on); | |
635 | ||
0fcb78c2 REB |
636 | /** |
637 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
638 | * @dwork: job to be done |
639 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
640 | * |
641 | * After waiting for a given time this puts a job in the kernel-global | |
642 | * workqueue. | |
643 | */ | |
7ad5b3a5 | 644 | int schedule_delayed_work(struct delayed_work *dwork, |
82f67cd9 | 645 | unsigned long delay) |
1da177e4 | 646 | { |
52bad64d | 647 | return queue_delayed_work(keventd_wq, dwork, delay); |
1da177e4 | 648 | } |
ae90dd5d | 649 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 650 | |
0fcb78c2 REB |
651 | /** |
652 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
653 | * @cpu: cpu to use | |
52bad64d | 654 | * @dwork: job to be done |
0fcb78c2 REB |
655 | * @delay: number of jiffies to wait |
656 | * | |
657 | * After waiting for a given time this puts a job in the kernel-global | |
658 | * workqueue on the specified CPU. | |
659 | */ | |
1da177e4 | 660 | int schedule_delayed_work_on(int cpu, |
52bad64d | 661 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 662 | { |
52bad64d | 663 | return queue_delayed_work_on(cpu, keventd_wq, dwork, delay); |
1da177e4 | 664 | } |
ae90dd5d | 665 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 666 | |
b6136773 AM |
667 | /** |
668 | * schedule_on_each_cpu - call a function on each online CPU from keventd | |
669 | * @func: the function to call | |
b6136773 AM |
670 | * |
671 | * Returns zero on success. | |
672 | * Returns -ve errno on failure. | |
673 | * | |
b6136773 AM |
674 | * schedule_on_each_cpu() is very slow. |
675 | */ | |
65f27f38 | 676 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
677 | { |
678 | int cpu; | |
b6136773 | 679 | struct work_struct *works; |
15316ba8 | 680 | |
b6136773 AM |
681 | works = alloc_percpu(struct work_struct); |
682 | if (!works) | |
15316ba8 | 683 | return -ENOMEM; |
b6136773 | 684 | |
95402b38 | 685 | get_online_cpus(); |
15316ba8 | 686 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
687 | struct work_struct *work = per_cpu_ptr(works, cpu); |
688 | ||
689 | INIT_WORK(work, func); | |
690 | set_bit(WORK_STRUCT_PENDING, work_data_bits(work)); | |
691 | __queue_work(per_cpu_ptr(keventd_wq->cpu_wq, cpu), work); | |
15316ba8 | 692 | } |
8616a89a ON |
693 | for_each_online_cpu(cpu) |
694 | flush_work(per_cpu_ptr(works, cpu)); | |
95402b38 | 695 | put_online_cpus(); |
b6136773 | 696 | free_percpu(works); |
15316ba8 CL |
697 | return 0; |
698 | } | |
699 | ||
1da177e4 LT |
700 | void flush_scheduled_work(void) |
701 | { | |
702 | flush_workqueue(keventd_wq); | |
703 | } | |
ae90dd5d | 704 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 705 | |
1fa44eca JB |
706 | /** |
707 | * execute_in_process_context - reliably execute the routine with user context | |
708 | * @fn: the function to execute | |
1fa44eca JB |
709 | * @ew: guaranteed storage for the execute work structure (must |
710 | * be available when the work executes) | |
711 | * | |
712 | * Executes the function immediately if process context is available, | |
713 | * otherwise schedules the function for delayed execution. | |
714 | * | |
715 | * Returns: 0 - function was executed | |
716 | * 1 - function was scheduled for execution | |
717 | */ | |
65f27f38 | 718 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
719 | { |
720 | if (!in_interrupt()) { | |
65f27f38 | 721 | fn(&ew->work); |
1fa44eca JB |
722 | return 0; |
723 | } | |
724 | ||
65f27f38 | 725 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
726 | schedule_work(&ew->work); |
727 | ||
728 | return 1; | |
729 | } | |
730 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
731 | ||
1da177e4 LT |
732 | int keventd_up(void) |
733 | { | |
734 | return keventd_wq != NULL; | |
735 | } | |
736 | ||
737 | int current_is_keventd(void) | |
738 | { | |
739 | struct cpu_workqueue_struct *cwq; | |
d243769d | 740 | int cpu = raw_smp_processor_id(); /* preempt-safe: keventd is per-cpu */ |
1da177e4 LT |
741 | int ret = 0; |
742 | ||
743 | BUG_ON(!keventd_wq); | |
744 | ||
89ada679 | 745 | cwq = per_cpu_ptr(keventd_wq->cpu_wq, cpu); |
1da177e4 LT |
746 | if (current == cwq->thread) |
747 | ret = 1; | |
748 | ||
749 | return ret; | |
750 | ||
751 | } | |
752 | ||
3af24433 ON |
753 | static struct cpu_workqueue_struct * |
754 | init_cpu_workqueue(struct workqueue_struct *wq, int cpu) | |
1da177e4 | 755 | { |
89ada679 | 756 | struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu); |
1da177e4 | 757 | |
3af24433 ON |
758 | cwq->wq = wq; |
759 | spin_lock_init(&cwq->lock); | |
760 | INIT_LIST_HEAD(&cwq->worklist); | |
761 | init_waitqueue_head(&cwq->more_work); | |
762 | ||
763 | return cwq; | |
1da177e4 LT |
764 | } |
765 | ||
3af24433 ON |
766 | static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu) |
767 | { | |
768 | struct workqueue_struct *wq = cwq->wq; | |
769 | const char *fmt = is_single_threaded(wq) ? "%s" : "%s/%d"; | |
770 | struct task_struct *p; | |
771 | ||
772 | p = kthread_create(worker_thread, cwq, fmt, wq->name, cpu); | |
773 | /* | |
774 | * Nobody can add the work_struct to this cwq, | |
775 | * if (caller is __create_workqueue) | |
776 | * nobody should see this wq | |
777 | * else // caller is CPU_UP_PREPARE | |
778 | * cpu is not on cpu_online_map | |
779 | * so we can abort safely. | |
780 | */ | |
781 | if (IS_ERR(p)) | |
782 | return PTR_ERR(p); | |
783 | ||
784 | cwq->thread = p; | |
3af24433 ON |
785 | |
786 | return 0; | |
787 | } | |
788 | ||
06ba38a9 ON |
789 | static void start_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu) |
790 | { | |
791 | struct task_struct *p = cwq->thread; | |
792 | ||
793 | if (p != NULL) { | |
794 | if (cpu >= 0) | |
795 | kthread_bind(p, cpu); | |
796 | wake_up_process(p); | |
797 | } | |
798 | } | |
799 | ||
4e6045f1 JB |
800 | struct workqueue_struct *__create_workqueue_key(const char *name, |
801 | int singlethread, | |
802 | int freezeable, | |
eb13ba87 JB |
803 | struct lock_class_key *key, |
804 | const char *lock_name) | |
1da177e4 | 805 | { |
1da177e4 | 806 | struct workqueue_struct *wq; |
3af24433 ON |
807 | struct cpu_workqueue_struct *cwq; |
808 | int err = 0, cpu; | |
1da177e4 | 809 | |
3af24433 ON |
810 | wq = kzalloc(sizeof(*wq), GFP_KERNEL); |
811 | if (!wq) | |
812 | return NULL; | |
813 | ||
814 | wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct); | |
815 | if (!wq->cpu_wq) { | |
816 | kfree(wq); | |
817 | return NULL; | |
818 | } | |
819 | ||
820 | wq->name = name; | |
eb13ba87 | 821 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 822 | wq->singlethread = singlethread; |
3af24433 | 823 | wq->freezeable = freezeable; |
cce1a165 | 824 | INIT_LIST_HEAD(&wq->list); |
3af24433 ON |
825 | |
826 | if (singlethread) { | |
3af24433 ON |
827 | cwq = init_cpu_workqueue(wq, singlethread_cpu); |
828 | err = create_workqueue_thread(cwq, singlethread_cpu); | |
06ba38a9 | 829 | start_workqueue_thread(cwq, -1); |
3af24433 | 830 | } else { |
3da1c84c | 831 | cpu_maps_update_begin(); |
95402b38 | 832 | spin_lock(&workqueue_lock); |
3af24433 | 833 | list_add(&wq->list, &workqueues); |
95402b38 | 834 | spin_unlock(&workqueue_lock); |
3af24433 ON |
835 | |
836 | for_each_possible_cpu(cpu) { | |
837 | cwq = init_cpu_workqueue(wq, cpu); | |
838 | if (err || !cpu_online(cpu)) | |
839 | continue; | |
840 | err = create_workqueue_thread(cwq, cpu); | |
06ba38a9 | 841 | start_workqueue_thread(cwq, cpu); |
1da177e4 | 842 | } |
3da1c84c | 843 | cpu_maps_update_done(); |
3af24433 ON |
844 | } |
845 | ||
846 | if (err) { | |
847 | destroy_workqueue(wq); | |
848 | wq = NULL; | |
849 | } | |
850 | return wq; | |
851 | } | |
4e6045f1 | 852 | EXPORT_SYMBOL_GPL(__create_workqueue_key); |
1da177e4 | 853 | |
1e35eaa2 | 854 | static void cleanup_workqueue_thread(struct cpu_workqueue_struct *cwq) |
3af24433 | 855 | { |
14441960 | 856 | /* |
3da1c84c ON |
857 | * Our caller is either destroy_workqueue() or CPU_POST_DEAD, |
858 | * cpu_add_remove_lock protects cwq->thread. | |
14441960 ON |
859 | */ |
860 | if (cwq->thread == NULL) | |
861 | return; | |
3af24433 | 862 | |
4e6045f1 JB |
863 | lock_acquire(&cwq->wq->lockdep_map, 0, 0, 0, 2, _THIS_IP_); |
864 | lock_release(&cwq->wq->lockdep_map, 1, _THIS_IP_); | |
865 | ||
13c22168 | 866 | flush_cpu_workqueue(cwq); |
14441960 | 867 | /* |
3da1c84c | 868 | * If the caller is CPU_POST_DEAD and cwq->worklist was not empty, |
13c22168 ON |
869 | * a concurrent flush_workqueue() can insert a barrier after us. |
870 | * However, in that case run_workqueue() won't return and check | |
871 | * kthread_should_stop() until it flushes all work_struct's. | |
14441960 ON |
872 | * When ->worklist becomes empty it is safe to exit because no |
873 | * more work_structs can be queued on this cwq: flush_workqueue | |
874 | * checks list_empty(), and a "normal" queue_work() can't use | |
875 | * a dead CPU. | |
876 | */ | |
14441960 ON |
877 | kthread_stop(cwq->thread); |
878 | cwq->thread = NULL; | |
3af24433 ON |
879 | } |
880 | ||
881 | /** | |
882 | * destroy_workqueue - safely terminate a workqueue | |
883 | * @wq: target workqueue | |
884 | * | |
885 | * Safely destroy a workqueue. All work currently pending will be done first. | |
886 | */ | |
887 | void destroy_workqueue(struct workqueue_struct *wq) | |
888 | { | |
b1f4ec17 | 889 | const cpumask_t *cpu_map = wq_cpu_map(wq); |
b1f4ec17 | 890 | int cpu; |
3af24433 | 891 | |
3da1c84c | 892 | cpu_maps_update_begin(); |
95402b38 | 893 | spin_lock(&workqueue_lock); |
b1f4ec17 | 894 | list_del(&wq->list); |
95402b38 | 895 | spin_unlock(&workqueue_lock); |
3af24433 | 896 | |
363ab6f1 | 897 | for_each_cpu_mask_nr(cpu, *cpu_map) |
1e35eaa2 | 898 | cleanup_workqueue_thread(per_cpu_ptr(wq->cpu_wq, cpu)); |
3da1c84c | 899 | cpu_maps_update_done(); |
9b41ea72 | 900 | |
3af24433 ON |
901 | free_percpu(wq->cpu_wq); |
902 | kfree(wq); | |
903 | } | |
904 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
905 | ||
906 | static int __devinit workqueue_cpu_callback(struct notifier_block *nfb, | |
907 | unsigned long action, | |
908 | void *hcpu) | |
909 | { | |
910 | unsigned int cpu = (unsigned long)hcpu; | |
911 | struct cpu_workqueue_struct *cwq; | |
912 | struct workqueue_struct *wq; | |
913 | ||
8bb78442 RW |
914 | action &= ~CPU_TASKS_FROZEN; |
915 | ||
3af24433 | 916 | switch (action) { |
3af24433 ON |
917 | case CPU_UP_PREPARE: |
918 | cpu_set(cpu, cpu_populated_map); | |
919 | } | |
920 | ||
921 | list_for_each_entry(wq, &workqueues, list) { | |
922 | cwq = per_cpu_ptr(wq->cpu_wq, cpu); | |
923 | ||
924 | switch (action) { | |
925 | case CPU_UP_PREPARE: | |
926 | if (!create_workqueue_thread(cwq, cpu)) | |
927 | break; | |
95402b38 GS |
928 | printk(KERN_ERR "workqueue [%s] for %i failed\n", |
929 | wq->name, cpu); | |
3af24433 ON |
930 | return NOTIFY_BAD; |
931 | ||
932 | case CPU_ONLINE: | |
06ba38a9 | 933 | start_workqueue_thread(cwq, cpu); |
3af24433 ON |
934 | break; |
935 | ||
936 | case CPU_UP_CANCELED: | |
06ba38a9 | 937 | start_workqueue_thread(cwq, -1); |
3da1c84c | 938 | case CPU_POST_DEAD: |
1e35eaa2 | 939 | cleanup_workqueue_thread(cwq); |
3af24433 ON |
940 | break; |
941 | } | |
1da177e4 LT |
942 | } |
943 | ||
00dfcaf7 ON |
944 | switch (action) { |
945 | case CPU_UP_CANCELED: | |
3da1c84c | 946 | case CPU_POST_DEAD: |
00dfcaf7 ON |
947 | cpu_clear(cpu, cpu_populated_map); |
948 | } | |
949 | ||
1da177e4 LT |
950 | return NOTIFY_OK; |
951 | } | |
1da177e4 | 952 | |
c12920d1 | 953 | void __init init_workqueues(void) |
1da177e4 | 954 | { |
3af24433 | 955 | cpu_populated_map = cpu_online_map; |
f756d5e2 | 956 | singlethread_cpu = first_cpu(cpu_possible_map); |
b1f4ec17 | 957 | cpu_singlethread_map = cpumask_of_cpu(singlethread_cpu); |
1da177e4 LT |
958 | hotcpu_notifier(workqueue_cpu_callback, 0); |
959 | keventd_wq = create_workqueue("events"); | |
960 | BUG_ON(!keventd_wq); | |
961 | } |