]>
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> | |
e1f8e874 | 12 | * Andrew Morton |
1da177e4 LT |
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> |
c34056a3 | 36 | #include <linux/idr.h> |
1da177e4 | 37 | |
4690c4ab TH |
38 | /* |
39 | * Structure fields follow one of the following exclusion rules. | |
40 | * | |
41 | * I: Set during initialization and read-only afterwards. | |
42 | * | |
43 | * L: cwq->lock protected. Access with cwq->lock held. | |
44 | * | |
73f53c4a TH |
45 | * F: wq->flush_mutex protected. |
46 | * | |
4690c4ab TH |
47 | * W: workqueue_lock protected. |
48 | */ | |
49 | ||
c34056a3 TH |
50 | struct cpu_workqueue_struct; |
51 | ||
52 | struct worker { | |
53 | struct work_struct *current_work; /* L: work being processed */ | |
affee4b2 | 54 | struct list_head scheduled; /* L: scheduled works */ |
c34056a3 TH |
55 | struct task_struct *task; /* I: worker task */ |
56 | struct cpu_workqueue_struct *cwq; /* I: the associated cwq */ | |
57 | int id; /* I: worker id */ | |
58 | }; | |
59 | ||
1da177e4 | 60 | /* |
f756d5e2 | 61 | * The per-CPU workqueue (if single thread, we always use the first |
0f900049 TH |
62 | * possible cpu). The lower WORK_STRUCT_FLAG_BITS of |
63 | * work_struct->data are used for flags and thus cwqs need to be | |
64 | * aligned at two's power of the number of flag bits. | |
1da177e4 LT |
65 | */ |
66 | struct cpu_workqueue_struct { | |
67 | ||
68 | spinlock_t lock; | |
69 | ||
1da177e4 LT |
70 | struct list_head worklist; |
71 | wait_queue_head_t more_work; | |
1537663f | 72 | unsigned int cpu; |
c34056a3 | 73 | struct worker *worker; |
1da177e4 | 74 | |
4690c4ab | 75 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
76 | int work_color; /* L: current color */ |
77 | int flush_color; /* L: flushing color */ | |
78 | int nr_in_flight[WORK_NR_COLORS]; | |
79 | /* L: nr of in_flight works */ | |
1e19ffc6 TH |
80 | int nr_active; /* L: nr of active works */ |
81 | int max_active; /* I: max active works */ | |
82 | struct list_head delayed_works; /* L: delayed works */ | |
0f900049 | 83 | }; |
1da177e4 | 84 | |
73f53c4a TH |
85 | /* |
86 | * Structure used to wait for workqueue flush. | |
87 | */ | |
88 | struct wq_flusher { | |
89 | struct list_head list; /* F: list of flushers */ | |
90 | int flush_color; /* F: flush color waiting for */ | |
91 | struct completion done; /* flush completion */ | |
92 | }; | |
93 | ||
1da177e4 LT |
94 | /* |
95 | * The externally visible workqueue abstraction is an array of | |
96 | * per-CPU workqueues: | |
97 | */ | |
98 | struct workqueue_struct { | |
97e37d7b | 99 | unsigned int flags; /* I: WQ_* flags */ |
4690c4ab TH |
100 | struct cpu_workqueue_struct *cpu_wq; /* I: cwq's */ |
101 | struct list_head list; /* W: list of all workqueues */ | |
73f53c4a TH |
102 | |
103 | struct mutex flush_mutex; /* protects wq flushing */ | |
104 | int work_color; /* F: current work color */ | |
105 | int flush_color; /* F: current flush color */ | |
106 | atomic_t nr_cwqs_to_flush; /* flush in progress */ | |
107 | struct wq_flusher *first_flusher; /* F: first flusher */ | |
108 | struct list_head flusher_queue; /* F: flush waiters */ | |
109 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
110 | ||
4690c4ab | 111 | const char *name; /* I: workqueue name */ |
4e6045f1 | 112 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 113 | struct lockdep_map lockdep_map; |
4e6045f1 | 114 | #endif |
1da177e4 LT |
115 | }; |
116 | ||
dc186ad7 TG |
117 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
118 | ||
119 | static struct debug_obj_descr work_debug_descr; | |
120 | ||
121 | /* | |
122 | * fixup_init is called when: | |
123 | * - an active object is initialized | |
124 | */ | |
125 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
126 | { | |
127 | struct work_struct *work = addr; | |
128 | ||
129 | switch (state) { | |
130 | case ODEBUG_STATE_ACTIVE: | |
131 | cancel_work_sync(work); | |
132 | debug_object_init(work, &work_debug_descr); | |
133 | return 1; | |
134 | default: | |
135 | return 0; | |
136 | } | |
137 | } | |
138 | ||
139 | /* | |
140 | * fixup_activate is called when: | |
141 | * - an active object is activated | |
142 | * - an unknown object is activated (might be a statically initialized object) | |
143 | */ | |
144 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
145 | { | |
146 | struct work_struct *work = addr; | |
147 | ||
148 | switch (state) { | |
149 | ||
150 | case ODEBUG_STATE_NOTAVAILABLE: | |
151 | /* | |
152 | * This is not really a fixup. The work struct was | |
153 | * statically initialized. We just make sure that it | |
154 | * is tracked in the object tracker. | |
155 | */ | |
22df02bb | 156 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
157 | debug_object_init(work, &work_debug_descr); |
158 | debug_object_activate(work, &work_debug_descr); | |
159 | return 0; | |
160 | } | |
161 | WARN_ON_ONCE(1); | |
162 | return 0; | |
163 | ||
164 | case ODEBUG_STATE_ACTIVE: | |
165 | WARN_ON(1); | |
166 | ||
167 | default: | |
168 | return 0; | |
169 | } | |
170 | } | |
171 | ||
172 | /* | |
173 | * fixup_free is called when: | |
174 | * - an active object is freed | |
175 | */ | |
176 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
177 | { | |
178 | struct work_struct *work = addr; | |
179 | ||
180 | switch (state) { | |
181 | case ODEBUG_STATE_ACTIVE: | |
182 | cancel_work_sync(work); | |
183 | debug_object_free(work, &work_debug_descr); | |
184 | return 1; | |
185 | default: | |
186 | return 0; | |
187 | } | |
188 | } | |
189 | ||
190 | static struct debug_obj_descr work_debug_descr = { | |
191 | .name = "work_struct", | |
192 | .fixup_init = work_fixup_init, | |
193 | .fixup_activate = work_fixup_activate, | |
194 | .fixup_free = work_fixup_free, | |
195 | }; | |
196 | ||
197 | static inline void debug_work_activate(struct work_struct *work) | |
198 | { | |
199 | debug_object_activate(work, &work_debug_descr); | |
200 | } | |
201 | ||
202 | static inline void debug_work_deactivate(struct work_struct *work) | |
203 | { | |
204 | debug_object_deactivate(work, &work_debug_descr); | |
205 | } | |
206 | ||
207 | void __init_work(struct work_struct *work, int onstack) | |
208 | { | |
209 | if (onstack) | |
210 | debug_object_init_on_stack(work, &work_debug_descr); | |
211 | else | |
212 | debug_object_init(work, &work_debug_descr); | |
213 | } | |
214 | EXPORT_SYMBOL_GPL(__init_work); | |
215 | ||
216 | void destroy_work_on_stack(struct work_struct *work) | |
217 | { | |
218 | debug_object_free(work, &work_debug_descr); | |
219 | } | |
220 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
221 | ||
222 | #else | |
223 | static inline void debug_work_activate(struct work_struct *work) { } | |
224 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
225 | #endif | |
226 | ||
95402b38 GS |
227 | /* Serializes the accesses to the list of workqueues. */ |
228 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 229 | static LIST_HEAD(workqueues); |
c34056a3 TH |
230 | static DEFINE_PER_CPU(struct ida, worker_ida); |
231 | ||
232 | static int worker_thread(void *__worker); | |
1da177e4 | 233 | |
3af24433 | 234 | static int singlethread_cpu __read_mostly; |
1da177e4 | 235 | |
1537663f TH |
236 | static struct cpu_workqueue_struct *get_cwq(unsigned int cpu, |
237 | struct workqueue_struct *wq) | |
b1f4ec17 | 238 | { |
1537663f | 239 | return per_cpu_ptr(wq->cpu_wq, cpu); |
b1f4ec17 ON |
240 | } |
241 | ||
1537663f TH |
242 | static struct cpu_workqueue_struct *target_cwq(unsigned int cpu, |
243 | struct workqueue_struct *wq) | |
a848e3b6 | 244 | { |
1537663f | 245 | if (unlikely(wq->flags & WQ_SINGLE_THREAD)) |
a848e3b6 | 246 | cpu = singlethread_cpu; |
1537663f | 247 | return get_cwq(cpu, wq); |
a848e3b6 ON |
248 | } |
249 | ||
73f53c4a TH |
250 | static unsigned int work_color_to_flags(int color) |
251 | { | |
252 | return color << WORK_STRUCT_COLOR_SHIFT; | |
253 | } | |
254 | ||
255 | static int get_work_color(struct work_struct *work) | |
256 | { | |
257 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
258 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
259 | } | |
260 | ||
261 | static int work_next_color(int color) | |
262 | { | |
263 | return (color + 1) % WORK_NR_COLORS; | |
264 | } | |
265 | ||
4594bf15 DH |
266 | /* |
267 | * Set the workqueue on which a work item is to be run | |
268 | * - Must *only* be called if the pending flag is set | |
269 | */ | |
ed7c0fee | 270 | static inline void set_wq_data(struct work_struct *work, |
4690c4ab TH |
271 | struct cpu_workqueue_struct *cwq, |
272 | unsigned long extra_flags) | |
365970a1 | 273 | { |
4594bf15 | 274 | BUG_ON(!work_pending(work)); |
365970a1 | 275 | |
4690c4ab | 276 | atomic_long_set(&work->data, (unsigned long)cwq | work_static(work) | |
22df02bb | 277 | WORK_STRUCT_PENDING | extra_flags); |
365970a1 DH |
278 | } |
279 | ||
4d707b9f ON |
280 | /* |
281 | * Clear WORK_STRUCT_PENDING and the workqueue on which it was queued. | |
282 | */ | |
283 | static inline void clear_wq_data(struct work_struct *work) | |
284 | { | |
4690c4ab | 285 | atomic_long_set(&work->data, work_static(work)); |
4d707b9f ON |
286 | } |
287 | ||
64166699 | 288 | static inline struct cpu_workqueue_struct *get_wq_data(struct work_struct *work) |
365970a1 | 289 | { |
64166699 TH |
290 | return (void *)(atomic_long_read(&work->data) & |
291 | WORK_STRUCT_WQ_DATA_MASK); | |
365970a1 DH |
292 | } |
293 | ||
4690c4ab TH |
294 | /** |
295 | * insert_work - insert a work into cwq | |
296 | * @cwq: cwq @work belongs to | |
297 | * @work: work to insert | |
298 | * @head: insertion point | |
299 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
300 | * | |
301 | * Insert @work into @cwq after @head. | |
302 | * | |
303 | * CONTEXT: | |
304 | * spin_lock_irq(cwq->lock). | |
305 | */ | |
b89deed3 | 306 | static void insert_work(struct cpu_workqueue_struct *cwq, |
4690c4ab TH |
307 | struct work_struct *work, struct list_head *head, |
308 | unsigned int extra_flags) | |
b89deed3 | 309 | { |
4690c4ab TH |
310 | /* we own @work, set data and link */ |
311 | set_wq_data(work, cwq, extra_flags); | |
312 | ||
6e84d644 ON |
313 | /* |
314 | * Ensure that we get the right work->data if we see the | |
315 | * result of list_add() below, see try_to_grab_pending(). | |
316 | */ | |
317 | smp_wmb(); | |
4690c4ab | 318 | |
1a4d9b0a | 319 | list_add_tail(&work->entry, head); |
b89deed3 ON |
320 | wake_up(&cwq->more_work); |
321 | } | |
322 | ||
4690c4ab | 323 | static void __queue_work(unsigned int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
324 | struct work_struct *work) |
325 | { | |
1537663f | 326 | struct cpu_workqueue_struct *cwq = target_cwq(cpu, wq); |
1e19ffc6 | 327 | struct list_head *worklist; |
1da177e4 LT |
328 | unsigned long flags; |
329 | ||
dc186ad7 | 330 | debug_work_activate(work); |
1e19ffc6 | 331 | |
1da177e4 | 332 | spin_lock_irqsave(&cwq->lock, flags); |
4690c4ab | 333 | BUG_ON(!list_empty(&work->entry)); |
1e19ffc6 | 334 | |
73f53c4a | 335 | cwq->nr_in_flight[cwq->work_color]++; |
1e19ffc6 TH |
336 | |
337 | if (likely(cwq->nr_active < cwq->max_active)) { | |
338 | cwq->nr_active++; | |
339 | worklist = &cwq->worklist; | |
340 | } else | |
341 | worklist = &cwq->delayed_works; | |
342 | ||
343 | insert_work(cwq, work, worklist, work_color_to_flags(cwq->work_color)); | |
344 | ||
1da177e4 LT |
345 | spin_unlock_irqrestore(&cwq->lock, flags); |
346 | } | |
347 | ||
0fcb78c2 REB |
348 | /** |
349 | * queue_work - queue work on a workqueue | |
350 | * @wq: workqueue to use | |
351 | * @work: work to queue | |
352 | * | |
057647fc | 353 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
1da177e4 | 354 | * |
00dfcaf7 ON |
355 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
356 | * it can be processed by another CPU. | |
1da177e4 | 357 | */ |
7ad5b3a5 | 358 | int queue_work(struct workqueue_struct *wq, struct work_struct *work) |
1da177e4 | 359 | { |
ef1ca236 ON |
360 | int ret; |
361 | ||
362 | ret = queue_work_on(get_cpu(), wq, work); | |
363 | put_cpu(); | |
364 | ||
1da177e4 LT |
365 | return ret; |
366 | } | |
ae90dd5d | 367 | EXPORT_SYMBOL_GPL(queue_work); |
1da177e4 | 368 | |
c1a220e7 ZR |
369 | /** |
370 | * queue_work_on - queue work on specific cpu | |
371 | * @cpu: CPU number to execute work on | |
372 | * @wq: workqueue to use | |
373 | * @work: work to queue | |
374 | * | |
375 | * Returns 0 if @work was already on a queue, non-zero otherwise. | |
376 | * | |
377 | * We queue the work to a specific CPU, the caller must ensure it | |
378 | * can't go away. | |
379 | */ | |
380 | int | |
381 | queue_work_on(int cpu, struct workqueue_struct *wq, struct work_struct *work) | |
382 | { | |
383 | int ret = 0; | |
384 | ||
22df02bb | 385 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 386 | __queue_work(cpu, wq, work); |
c1a220e7 ZR |
387 | ret = 1; |
388 | } | |
389 | return ret; | |
390 | } | |
391 | EXPORT_SYMBOL_GPL(queue_work_on); | |
392 | ||
6d141c3f | 393 | static void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 394 | { |
52bad64d | 395 | struct delayed_work *dwork = (struct delayed_work *)__data; |
ed7c0fee | 396 | struct cpu_workqueue_struct *cwq = get_wq_data(&dwork->work); |
1da177e4 | 397 | |
4690c4ab | 398 | __queue_work(smp_processor_id(), cwq->wq, &dwork->work); |
1da177e4 LT |
399 | } |
400 | ||
0fcb78c2 REB |
401 | /** |
402 | * queue_delayed_work - queue work on a workqueue after delay | |
403 | * @wq: workqueue to use | |
af9997e4 | 404 | * @dwork: delayable work to queue |
0fcb78c2 REB |
405 | * @delay: number of jiffies to wait before queueing |
406 | * | |
057647fc | 407 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 408 | */ |
7ad5b3a5 | 409 | int queue_delayed_work(struct workqueue_struct *wq, |
52bad64d | 410 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 411 | { |
52bad64d | 412 | if (delay == 0) |
63bc0362 | 413 | return queue_work(wq, &dwork->work); |
1da177e4 | 414 | |
63bc0362 | 415 | return queue_delayed_work_on(-1, wq, dwork, delay); |
1da177e4 | 416 | } |
ae90dd5d | 417 | EXPORT_SYMBOL_GPL(queue_delayed_work); |
1da177e4 | 418 | |
0fcb78c2 REB |
419 | /** |
420 | * queue_delayed_work_on - queue work on specific CPU after delay | |
421 | * @cpu: CPU number to execute work on | |
422 | * @wq: workqueue to use | |
af9997e4 | 423 | * @dwork: work to queue |
0fcb78c2 REB |
424 | * @delay: number of jiffies to wait before queueing |
425 | * | |
057647fc | 426 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 427 | */ |
7a6bc1cd | 428 | int queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
52bad64d | 429 | struct delayed_work *dwork, unsigned long delay) |
7a6bc1cd VP |
430 | { |
431 | int ret = 0; | |
52bad64d DH |
432 | struct timer_list *timer = &dwork->timer; |
433 | struct work_struct *work = &dwork->work; | |
7a6bc1cd | 434 | |
22df02bb | 435 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7a6bc1cd VP |
436 | BUG_ON(timer_pending(timer)); |
437 | BUG_ON(!list_empty(&work->entry)); | |
438 | ||
8a3e77cc AL |
439 | timer_stats_timer_set_start_info(&dwork->timer); |
440 | ||
ed7c0fee | 441 | /* This stores cwq for the moment, for the timer_fn */ |
1537663f | 442 | set_wq_data(work, target_cwq(raw_smp_processor_id(), wq), 0); |
7a6bc1cd | 443 | timer->expires = jiffies + delay; |
52bad64d | 444 | timer->data = (unsigned long)dwork; |
7a6bc1cd | 445 | timer->function = delayed_work_timer_fn; |
63bc0362 ON |
446 | |
447 | if (unlikely(cpu >= 0)) | |
448 | add_timer_on(timer, cpu); | |
449 | else | |
450 | add_timer(timer); | |
7a6bc1cd VP |
451 | ret = 1; |
452 | } | |
453 | return ret; | |
454 | } | |
ae90dd5d | 455 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
1da177e4 | 456 | |
c34056a3 TH |
457 | static struct worker *alloc_worker(void) |
458 | { | |
459 | struct worker *worker; | |
460 | ||
461 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
affee4b2 TH |
462 | if (worker) |
463 | INIT_LIST_HEAD(&worker->scheduled); | |
c34056a3 TH |
464 | return worker; |
465 | } | |
466 | ||
467 | /** | |
468 | * create_worker - create a new workqueue worker | |
469 | * @cwq: cwq the new worker will belong to | |
470 | * @bind: whether to set affinity to @cpu or not | |
471 | * | |
472 | * Create a new worker which is bound to @cwq. The returned worker | |
473 | * can be started by calling start_worker() or destroyed using | |
474 | * destroy_worker(). | |
475 | * | |
476 | * CONTEXT: | |
477 | * Might sleep. Does GFP_KERNEL allocations. | |
478 | * | |
479 | * RETURNS: | |
480 | * Pointer to the newly created worker. | |
481 | */ | |
482 | static struct worker *create_worker(struct cpu_workqueue_struct *cwq, bool bind) | |
483 | { | |
484 | int id = -1; | |
485 | struct worker *worker = NULL; | |
486 | ||
487 | spin_lock(&workqueue_lock); | |
488 | while (ida_get_new(&per_cpu(worker_ida, cwq->cpu), &id)) { | |
489 | spin_unlock(&workqueue_lock); | |
490 | if (!ida_pre_get(&per_cpu(worker_ida, cwq->cpu), GFP_KERNEL)) | |
491 | goto fail; | |
492 | spin_lock(&workqueue_lock); | |
493 | } | |
494 | spin_unlock(&workqueue_lock); | |
495 | ||
496 | worker = alloc_worker(); | |
497 | if (!worker) | |
498 | goto fail; | |
499 | ||
500 | worker->cwq = cwq; | |
501 | worker->id = id; | |
502 | ||
503 | worker->task = kthread_create(worker_thread, worker, "kworker/%u:%d", | |
504 | cwq->cpu, id); | |
505 | if (IS_ERR(worker->task)) | |
506 | goto fail; | |
507 | ||
508 | if (bind) | |
509 | kthread_bind(worker->task, cwq->cpu); | |
510 | ||
511 | return worker; | |
512 | fail: | |
513 | if (id >= 0) { | |
514 | spin_lock(&workqueue_lock); | |
515 | ida_remove(&per_cpu(worker_ida, cwq->cpu), id); | |
516 | spin_unlock(&workqueue_lock); | |
517 | } | |
518 | kfree(worker); | |
519 | return NULL; | |
520 | } | |
521 | ||
522 | /** | |
523 | * start_worker - start a newly created worker | |
524 | * @worker: worker to start | |
525 | * | |
526 | * Start @worker. | |
527 | * | |
528 | * CONTEXT: | |
529 | * spin_lock_irq(cwq->lock). | |
530 | */ | |
531 | static void start_worker(struct worker *worker) | |
532 | { | |
533 | wake_up_process(worker->task); | |
534 | } | |
535 | ||
536 | /** | |
537 | * destroy_worker - destroy a workqueue worker | |
538 | * @worker: worker to be destroyed | |
539 | * | |
540 | * Destroy @worker. | |
541 | */ | |
542 | static void destroy_worker(struct worker *worker) | |
543 | { | |
544 | int cpu = worker->cwq->cpu; | |
545 | int id = worker->id; | |
546 | ||
547 | /* sanity check frenzy */ | |
548 | BUG_ON(worker->current_work); | |
affee4b2 | 549 | BUG_ON(!list_empty(&worker->scheduled)); |
c34056a3 TH |
550 | |
551 | kthread_stop(worker->task); | |
552 | kfree(worker); | |
553 | ||
554 | spin_lock(&workqueue_lock); | |
555 | ida_remove(&per_cpu(worker_ida, cpu), id); | |
556 | spin_unlock(&workqueue_lock); | |
557 | } | |
558 | ||
affee4b2 TH |
559 | /** |
560 | * move_linked_works - move linked works to a list | |
561 | * @work: start of series of works to be scheduled | |
562 | * @head: target list to append @work to | |
563 | * @nextp: out paramter for nested worklist walking | |
564 | * | |
565 | * Schedule linked works starting from @work to @head. Work series to | |
566 | * be scheduled starts at @work and includes any consecutive work with | |
567 | * WORK_STRUCT_LINKED set in its predecessor. | |
568 | * | |
569 | * If @nextp is not NULL, it's updated to point to the next work of | |
570 | * the last scheduled work. This allows move_linked_works() to be | |
571 | * nested inside outer list_for_each_entry_safe(). | |
572 | * | |
573 | * CONTEXT: | |
574 | * spin_lock_irq(cwq->lock). | |
575 | */ | |
576 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
577 | struct work_struct **nextp) | |
578 | { | |
579 | struct work_struct *n; | |
580 | ||
581 | /* | |
582 | * Linked worklist will always end before the end of the list, | |
583 | * use NULL for list head. | |
584 | */ | |
585 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
586 | list_move_tail(&work->entry, head); | |
587 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
588 | break; | |
589 | } | |
590 | ||
591 | /* | |
592 | * If we're already inside safe list traversal and have moved | |
593 | * multiple works to the scheduled queue, the next position | |
594 | * needs to be updated. | |
595 | */ | |
596 | if (nextp) | |
597 | *nextp = n; | |
598 | } | |
599 | ||
1e19ffc6 TH |
600 | static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq) |
601 | { | |
602 | struct work_struct *work = list_first_entry(&cwq->delayed_works, | |
603 | struct work_struct, entry); | |
604 | ||
605 | move_linked_works(work, &cwq->worklist, NULL); | |
606 | cwq->nr_active++; | |
607 | } | |
608 | ||
73f53c4a TH |
609 | /** |
610 | * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight | |
611 | * @cwq: cwq of interest | |
612 | * @color: color of work which left the queue | |
613 | * | |
614 | * A work either has completed or is removed from pending queue, | |
615 | * decrement nr_in_flight of its cwq and handle workqueue flushing. | |
616 | * | |
617 | * CONTEXT: | |
618 | * spin_lock_irq(cwq->lock). | |
619 | */ | |
620 | static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color) | |
621 | { | |
622 | /* ignore uncolored works */ | |
623 | if (color == WORK_NO_COLOR) | |
624 | return; | |
625 | ||
626 | cwq->nr_in_flight[color]--; | |
1e19ffc6 TH |
627 | cwq->nr_active--; |
628 | ||
629 | /* one down, submit a delayed one */ | |
630 | if (!list_empty(&cwq->delayed_works) && | |
631 | cwq->nr_active < cwq->max_active) | |
632 | cwq_activate_first_delayed(cwq); | |
73f53c4a TH |
633 | |
634 | /* is flush in progress and are we at the flushing tip? */ | |
635 | if (likely(cwq->flush_color != color)) | |
636 | return; | |
637 | ||
638 | /* are there still in-flight works? */ | |
639 | if (cwq->nr_in_flight[color]) | |
640 | return; | |
641 | ||
642 | /* this cwq is done, clear flush_color */ | |
643 | cwq->flush_color = -1; | |
644 | ||
645 | /* | |
646 | * If this was the last cwq, wake up the first flusher. It | |
647 | * will handle the rest. | |
648 | */ | |
649 | if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush)) | |
650 | complete(&cwq->wq->first_flusher->done); | |
651 | } | |
652 | ||
a62428c0 TH |
653 | /** |
654 | * process_one_work - process single work | |
c34056a3 | 655 | * @worker: self |
a62428c0 TH |
656 | * @work: work to process |
657 | * | |
658 | * Process @work. This function contains all the logics necessary to | |
659 | * process a single work including synchronization against and | |
660 | * interaction with other workers on the same cpu, queueing and | |
661 | * flushing. As long as context requirement is met, any worker can | |
662 | * call this function to process a work. | |
663 | * | |
664 | * CONTEXT: | |
665 | * spin_lock_irq(cwq->lock) which is released and regrabbed. | |
666 | */ | |
c34056a3 | 667 | static void process_one_work(struct worker *worker, struct work_struct *work) |
a62428c0 | 668 | { |
c34056a3 | 669 | struct cpu_workqueue_struct *cwq = worker->cwq; |
a62428c0 | 670 | work_func_t f = work->func; |
73f53c4a | 671 | int work_color; |
a62428c0 TH |
672 | #ifdef CONFIG_LOCKDEP |
673 | /* | |
674 | * It is permissible to free the struct work_struct from | |
675 | * inside the function that is called from it, this we need to | |
676 | * take into account for lockdep too. To avoid bogus "held | |
677 | * lock freed" warnings as well as problems when looking into | |
678 | * work->lockdep_map, make a copy and use that here. | |
679 | */ | |
680 | struct lockdep_map lockdep_map = work->lockdep_map; | |
681 | #endif | |
682 | /* claim and process */ | |
a62428c0 | 683 | debug_work_deactivate(work); |
c34056a3 | 684 | worker->current_work = work; |
73f53c4a | 685 | work_color = get_work_color(work); |
a62428c0 TH |
686 | list_del_init(&work->entry); |
687 | ||
688 | spin_unlock_irq(&cwq->lock); | |
689 | ||
690 | BUG_ON(get_wq_data(work) != cwq); | |
691 | work_clear_pending(work); | |
692 | lock_map_acquire(&cwq->wq->lockdep_map); | |
693 | lock_map_acquire(&lockdep_map); | |
694 | f(work); | |
695 | lock_map_release(&lockdep_map); | |
696 | lock_map_release(&cwq->wq->lockdep_map); | |
697 | ||
698 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
699 | printk(KERN_ERR "BUG: workqueue leaked lock or atomic: " | |
700 | "%s/0x%08x/%d\n", | |
701 | current->comm, preempt_count(), task_pid_nr(current)); | |
702 | printk(KERN_ERR " last function: "); | |
703 | print_symbol("%s\n", (unsigned long)f); | |
704 | debug_show_held_locks(current); | |
705 | dump_stack(); | |
706 | } | |
707 | ||
708 | spin_lock_irq(&cwq->lock); | |
709 | ||
710 | /* we're done with it, release */ | |
c34056a3 | 711 | worker->current_work = NULL; |
73f53c4a | 712 | cwq_dec_nr_in_flight(cwq, work_color); |
a62428c0 TH |
713 | } |
714 | ||
affee4b2 TH |
715 | /** |
716 | * process_scheduled_works - process scheduled works | |
717 | * @worker: self | |
718 | * | |
719 | * Process all scheduled works. Please note that the scheduled list | |
720 | * may change while processing a work, so this function repeatedly | |
721 | * fetches a work from the top and executes it. | |
722 | * | |
723 | * CONTEXT: | |
724 | * spin_lock_irq(cwq->lock) which may be released and regrabbed | |
725 | * multiple times. | |
726 | */ | |
727 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 728 | { |
affee4b2 TH |
729 | while (!list_empty(&worker->scheduled)) { |
730 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 731 | struct work_struct, entry); |
c34056a3 | 732 | process_one_work(worker, work); |
1da177e4 | 733 | } |
1da177e4 LT |
734 | } |
735 | ||
4690c4ab TH |
736 | /** |
737 | * worker_thread - the worker thread function | |
c34056a3 | 738 | * @__worker: self |
4690c4ab TH |
739 | * |
740 | * The cwq worker thread function. | |
741 | */ | |
c34056a3 | 742 | static int worker_thread(void *__worker) |
1da177e4 | 743 | { |
c34056a3 TH |
744 | struct worker *worker = __worker; |
745 | struct cpu_workqueue_struct *cwq = worker->cwq; | |
3af24433 | 746 | DEFINE_WAIT(wait); |
1da177e4 | 747 | |
97e37d7b | 748 | if (cwq->wq->flags & WQ_FREEZEABLE) |
83144186 | 749 | set_freezable(); |
1da177e4 | 750 | |
3af24433 | 751 | for (;;) { |
3af24433 | 752 | prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE); |
14441960 ON |
753 | if (!freezing(current) && |
754 | !kthread_should_stop() && | |
755 | list_empty(&cwq->worklist)) | |
1da177e4 | 756 | schedule(); |
3af24433 ON |
757 | finish_wait(&cwq->more_work, &wait); |
758 | ||
85f4186a ON |
759 | try_to_freeze(); |
760 | ||
14441960 | 761 | if (kthread_should_stop()) |
3af24433 | 762 | break; |
1da177e4 | 763 | |
c34056a3 | 764 | if (unlikely(!cpumask_equal(&worker->task->cpus_allowed, |
1537663f | 765 | get_cpu_mask(cwq->cpu)))) |
c34056a3 | 766 | set_cpus_allowed_ptr(worker->task, |
1537663f | 767 | get_cpu_mask(cwq->cpu)); |
affee4b2 TH |
768 | |
769 | spin_lock_irq(&cwq->lock); | |
770 | ||
771 | while (!list_empty(&cwq->worklist)) { | |
772 | struct work_struct *work = | |
773 | list_first_entry(&cwq->worklist, | |
774 | struct work_struct, entry); | |
775 | ||
776 | if (likely(!(*work_data_bits(work) & | |
777 | WORK_STRUCT_LINKED))) { | |
778 | /* optimization path, not strictly necessary */ | |
779 | process_one_work(worker, work); | |
780 | if (unlikely(!list_empty(&worker->scheduled))) | |
781 | process_scheduled_works(worker); | |
782 | } else { | |
783 | move_linked_works(work, &worker->scheduled, | |
784 | NULL); | |
785 | process_scheduled_works(worker); | |
786 | } | |
787 | } | |
788 | ||
789 | spin_unlock_irq(&cwq->lock); | |
1da177e4 | 790 | } |
3af24433 | 791 | |
1da177e4 LT |
792 | return 0; |
793 | } | |
794 | ||
fc2e4d70 ON |
795 | struct wq_barrier { |
796 | struct work_struct work; | |
797 | struct completion done; | |
798 | }; | |
799 | ||
800 | static void wq_barrier_func(struct work_struct *work) | |
801 | { | |
802 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
803 | complete(&barr->done); | |
804 | } | |
805 | ||
4690c4ab TH |
806 | /** |
807 | * insert_wq_barrier - insert a barrier work | |
808 | * @cwq: cwq to insert barrier into | |
809 | * @barr: wq_barrier to insert | |
affee4b2 TH |
810 | * @target: target work to attach @barr to |
811 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 812 | * |
affee4b2 TH |
813 | * @barr is linked to @target such that @barr is completed only after |
814 | * @target finishes execution. Please note that the ordering | |
815 | * guarantee is observed only with respect to @target and on the local | |
816 | * cpu. | |
817 | * | |
818 | * Currently, a queued barrier can't be canceled. This is because | |
819 | * try_to_grab_pending() can't determine whether the work to be | |
820 | * grabbed is at the head of the queue and thus can't clear LINKED | |
821 | * flag of the previous work while there must be a valid next work | |
822 | * after a work with LINKED flag set. | |
823 | * | |
824 | * Note that when @worker is non-NULL, @target may be modified | |
825 | * underneath us, so we can't reliably determine cwq from @target. | |
4690c4ab TH |
826 | * |
827 | * CONTEXT: | |
828 | * spin_lock_irq(cwq->lock). | |
829 | */ | |
83c22520 | 830 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, |
affee4b2 TH |
831 | struct wq_barrier *barr, |
832 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 833 | { |
affee4b2 TH |
834 | struct list_head *head; |
835 | unsigned int linked = 0; | |
836 | ||
dc186ad7 TG |
837 | /* |
838 | * debugobject calls are safe here even with cwq->lock locked | |
839 | * as we know for sure that this will not trigger any of the | |
840 | * checks and call back into the fixup functions where we | |
841 | * might deadlock. | |
842 | */ | |
843 | INIT_WORK_ON_STACK(&barr->work, wq_barrier_func); | |
22df02bb | 844 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 845 | init_completion(&barr->done); |
83c22520 | 846 | |
affee4b2 TH |
847 | /* |
848 | * If @target is currently being executed, schedule the | |
849 | * barrier to the worker; otherwise, put it after @target. | |
850 | */ | |
851 | if (worker) | |
852 | head = worker->scheduled.next; | |
853 | else { | |
854 | unsigned long *bits = work_data_bits(target); | |
855 | ||
856 | head = target->entry.next; | |
857 | /* there can already be other linked works, inherit and set */ | |
858 | linked = *bits & WORK_STRUCT_LINKED; | |
859 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
860 | } | |
861 | ||
dc186ad7 | 862 | debug_work_activate(&barr->work); |
affee4b2 TH |
863 | insert_work(cwq, &barr->work, head, |
864 | work_color_to_flags(WORK_NO_COLOR) | linked); | |
fc2e4d70 ON |
865 | } |
866 | ||
73f53c4a TH |
867 | /** |
868 | * flush_workqueue_prep_cwqs - prepare cwqs for workqueue flushing | |
869 | * @wq: workqueue being flushed | |
870 | * @flush_color: new flush color, < 0 for no-op | |
871 | * @work_color: new work color, < 0 for no-op | |
872 | * | |
873 | * Prepare cwqs for workqueue flushing. | |
874 | * | |
875 | * If @flush_color is non-negative, flush_color on all cwqs should be | |
876 | * -1. If no cwq has in-flight commands at the specified color, all | |
877 | * cwq->flush_color's stay at -1 and %false is returned. If any cwq | |
878 | * has in flight commands, its cwq->flush_color is set to | |
879 | * @flush_color, @wq->nr_cwqs_to_flush is updated accordingly, cwq | |
880 | * wakeup logic is armed and %true is returned. | |
881 | * | |
882 | * The caller should have initialized @wq->first_flusher prior to | |
883 | * calling this function with non-negative @flush_color. If | |
884 | * @flush_color is negative, no flush color update is done and %false | |
885 | * is returned. | |
886 | * | |
887 | * If @work_color is non-negative, all cwqs should have the same | |
888 | * work_color which is previous to @work_color and all will be | |
889 | * advanced to @work_color. | |
890 | * | |
891 | * CONTEXT: | |
892 | * mutex_lock(wq->flush_mutex). | |
893 | * | |
894 | * RETURNS: | |
895 | * %true if @flush_color >= 0 and there's something to flush. %false | |
896 | * otherwise. | |
897 | */ | |
898 | static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq, | |
899 | int flush_color, int work_color) | |
1da177e4 | 900 | { |
73f53c4a TH |
901 | bool wait = false; |
902 | unsigned int cpu; | |
1da177e4 | 903 | |
73f53c4a TH |
904 | if (flush_color >= 0) { |
905 | BUG_ON(atomic_read(&wq->nr_cwqs_to_flush)); | |
906 | atomic_set(&wq->nr_cwqs_to_flush, 1); | |
1da177e4 | 907 | } |
2355b70f | 908 | |
73f53c4a TH |
909 | for_each_possible_cpu(cpu) { |
910 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
911 | ||
912 | spin_lock_irq(&cwq->lock); | |
913 | ||
914 | if (flush_color >= 0) { | |
915 | BUG_ON(cwq->flush_color != -1); | |
916 | ||
917 | if (cwq->nr_in_flight[flush_color]) { | |
918 | cwq->flush_color = flush_color; | |
919 | atomic_inc(&wq->nr_cwqs_to_flush); | |
920 | wait = true; | |
921 | } | |
922 | } | |
923 | ||
924 | if (work_color >= 0) { | |
925 | BUG_ON(work_color != work_next_color(cwq->work_color)); | |
926 | cwq->work_color = work_color; | |
927 | } | |
928 | ||
929 | spin_unlock_irq(&cwq->lock); | |
dc186ad7 | 930 | } |
14441960 | 931 | |
73f53c4a TH |
932 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_cwqs_to_flush)) |
933 | complete(&wq->first_flusher->done); | |
934 | ||
935 | return wait; | |
1da177e4 LT |
936 | } |
937 | ||
0fcb78c2 | 938 | /** |
1da177e4 | 939 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 940 | * @wq: workqueue to flush |
1da177e4 LT |
941 | * |
942 | * Forces execution of the workqueue and blocks until its completion. | |
943 | * This is typically used in driver shutdown handlers. | |
944 | * | |
fc2e4d70 ON |
945 | * We sleep until all works which were queued on entry have been handled, |
946 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 947 | */ |
7ad5b3a5 | 948 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 949 | { |
73f53c4a TH |
950 | struct wq_flusher this_flusher = { |
951 | .list = LIST_HEAD_INIT(this_flusher.list), | |
952 | .flush_color = -1, | |
953 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
954 | }; | |
955 | int next_color; | |
1da177e4 | 956 | |
3295f0ef IM |
957 | lock_map_acquire(&wq->lockdep_map); |
958 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
959 | |
960 | mutex_lock(&wq->flush_mutex); | |
961 | ||
962 | /* | |
963 | * Start-to-wait phase | |
964 | */ | |
965 | next_color = work_next_color(wq->work_color); | |
966 | ||
967 | if (next_color != wq->flush_color) { | |
968 | /* | |
969 | * Color space is not full. The current work_color | |
970 | * becomes our flush_color and work_color is advanced | |
971 | * by one. | |
972 | */ | |
973 | BUG_ON(!list_empty(&wq->flusher_overflow)); | |
974 | this_flusher.flush_color = wq->work_color; | |
975 | wq->work_color = next_color; | |
976 | ||
977 | if (!wq->first_flusher) { | |
978 | /* no flush in progress, become the first flusher */ | |
979 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
980 | ||
981 | wq->first_flusher = &this_flusher; | |
982 | ||
983 | if (!flush_workqueue_prep_cwqs(wq, wq->flush_color, | |
984 | wq->work_color)) { | |
985 | /* nothing to flush, done */ | |
986 | wq->flush_color = next_color; | |
987 | wq->first_flusher = NULL; | |
988 | goto out_unlock; | |
989 | } | |
990 | } else { | |
991 | /* wait in queue */ | |
992 | BUG_ON(wq->flush_color == this_flusher.flush_color); | |
993 | list_add_tail(&this_flusher.list, &wq->flusher_queue); | |
994 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
995 | } | |
996 | } else { | |
997 | /* | |
998 | * Oops, color space is full, wait on overflow queue. | |
999 | * The next flush completion will assign us | |
1000 | * flush_color and transfer to flusher_queue. | |
1001 | */ | |
1002 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
1003 | } | |
1004 | ||
1005 | mutex_unlock(&wq->flush_mutex); | |
1006 | ||
1007 | wait_for_completion(&this_flusher.done); | |
1008 | ||
1009 | /* | |
1010 | * Wake-up-and-cascade phase | |
1011 | * | |
1012 | * First flushers are responsible for cascading flushes and | |
1013 | * handling overflow. Non-first flushers can simply return. | |
1014 | */ | |
1015 | if (wq->first_flusher != &this_flusher) | |
1016 | return; | |
1017 | ||
1018 | mutex_lock(&wq->flush_mutex); | |
1019 | ||
1020 | wq->first_flusher = NULL; | |
1021 | ||
1022 | BUG_ON(!list_empty(&this_flusher.list)); | |
1023 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
1024 | ||
1025 | while (true) { | |
1026 | struct wq_flusher *next, *tmp; | |
1027 | ||
1028 | /* complete all the flushers sharing the current flush color */ | |
1029 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
1030 | if (next->flush_color != wq->flush_color) | |
1031 | break; | |
1032 | list_del_init(&next->list); | |
1033 | complete(&next->done); | |
1034 | } | |
1035 | ||
1036 | BUG_ON(!list_empty(&wq->flusher_overflow) && | |
1037 | wq->flush_color != work_next_color(wq->work_color)); | |
1038 | ||
1039 | /* this flush_color is finished, advance by one */ | |
1040 | wq->flush_color = work_next_color(wq->flush_color); | |
1041 | ||
1042 | /* one color has been freed, handle overflow queue */ | |
1043 | if (!list_empty(&wq->flusher_overflow)) { | |
1044 | /* | |
1045 | * Assign the same color to all overflowed | |
1046 | * flushers, advance work_color and append to | |
1047 | * flusher_queue. This is the start-to-wait | |
1048 | * phase for these overflowed flushers. | |
1049 | */ | |
1050 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
1051 | tmp->flush_color = wq->work_color; | |
1052 | ||
1053 | wq->work_color = work_next_color(wq->work_color); | |
1054 | ||
1055 | list_splice_tail_init(&wq->flusher_overflow, | |
1056 | &wq->flusher_queue); | |
1057 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
1058 | } | |
1059 | ||
1060 | if (list_empty(&wq->flusher_queue)) { | |
1061 | BUG_ON(wq->flush_color != wq->work_color); | |
1062 | break; | |
1063 | } | |
1064 | ||
1065 | /* | |
1066 | * Need to flush more colors. Make the next flusher | |
1067 | * the new first flusher and arm cwqs. | |
1068 | */ | |
1069 | BUG_ON(wq->flush_color == wq->work_color); | |
1070 | BUG_ON(wq->flush_color != next->flush_color); | |
1071 | ||
1072 | list_del_init(&next->list); | |
1073 | wq->first_flusher = next; | |
1074 | ||
1075 | if (flush_workqueue_prep_cwqs(wq, wq->flush_color, -1)) | |
1076 | break; | |
1077 | ||
1078 | /* | |
1079 | * Meh... this color is already done, clear first | |
1080 | * flusher and repeat cascading. | |
1081 | */ | |
1082 | wq->first_flusher = NULL; | |
1083 | } | |
1084 | ||
1085 | out_unlock: | |
1086 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 1087 | } |
ae90dd5d | 1088 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 1089 | |
db700897 ON |
1090 | /** |
1091 | * flush_work - block until a work_struct's callback has terminated | |
1092 | * @work: the work which is to be flushed | |
1093 | * | |
a67da70d ON |
1094 | * Returns false if @work has already terminated. |
1095 | * | |
db700897 ON |
1096 | * It is expected that, prior to calling flush_work(), the caller has |
1097 | * arranged for the work to not be requeued, otherwise it doesn't make | |
1098 | * sense to use this function. | |
1099 | */ | |
1100 | int flush_work(struct work_struct *work) | |
1101 | { | |
affee4b2 | 1102 | struct worker *worker = NULL; |
db700897 | 1103 | struct cpu_workqueue_struct *cwq; |
db700897 ON |
1104 | struct wq_barrier barr; |
1105 | ||
1106 | might_sleep(); | |
1107 | cwq = get_wq_data(work); | |
1108 | if (!cwq) | |
1109 | return 0; | |
1110 | ||
3295f0ef IM |
1111 | lock_map_acquire(&cwq->wq->lockdep_map); |
1112 | lock_map_release(&cwq->wq->lockdep_map); | |
a67da70d | 1113 | |
db700897 ON |
1114 | spin_lock_irq(&cwq->lock); |
1115 | if (!list_empty(&work->entry)) { | |
1116 | /* | |
1117 | * See the comment near try_to_grab_pending()->smp_rmb(). | |
1118 | * If it was re-queued under us we are not going to wait. | |
1119 | */ | |
1120 | smp_rmb(); | |
1121 | if (unlikely(cwq != get_wq_data(work))) | |
4690c4ab | 1122 | goto already_gone; |
db700897 | 1123 | } else { |
affee4b2 TH |
1124 | if (cwq->worker && cwq->worker->current_work == work) |
1125 | worker = cwq->worker; | |
1126 | if (!worker) | |
4690c4ab | 1127 | goto already_gone; |
db700897 | 1128 | } |
db700897 | 1129 | |
affee4b2 | 1130 | insert_wq_barrier(cwq, &barr, work, worker); |
4690c4ab | 1131 | spin_unlock_irq(&cwq->lock); |
db700897 | 1132 | wait_for_completion(&barr.done); |
dc186ad7 | 1133 | destroy_work_on_stack(&barr.work); |
db700897 | 1134 | return 1; |
4690c4ab TH |
1135 | already_gone: |
1136 | spin_unlock_irq(&cwq->lock); | |
1137 | return 0; | |
db700897 ON |
1138 | } |
1139 | EXPORT_SYMBOL_GPL(flush_work); | |
1140 | ||
6e84d644 | 1141 | /* |
1f1f642e | 1142 | * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit, |
6e84d644 ON |
1143 | * so this work can't be re-armed in any way. |
1144 | */ | |
1145 | static int try_to_grab_pending(struct work_struct *work) | |
1146 | { | |
1147 | struct cpu_workqueue_struct *cwq; | |
1f1f642e | 1148 | int ret = -1; |
6e84d644 | 1149 | |
22df02bb | 1150 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1f1f642e | 1151 | return 0; |
6e84d644 ON |
1152 | |
1153 | /* | |
1154 | * The queueing is in progress, or it is already queued. Try to | |
1155 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1156 | */ | |
1157 | ||
1158 | cwq = get_wq_data(work); | |
1159 | if (!cwq) | |
1160 | return ret; | |
1161 | ||
1162 | spin_lock_irq(&cwq->lock); | |
1163 | if (!list_empty(&work->entry)) { | |
1164 | /* | |
1165 | * This work is queued, but perhaps we locked the wrong cwq. | |
1166 | * In that case we must see the new value after rmb(), see | |
1167 | * insert_work()->wmb(). | |
1168 | */ | |
1169 | smp_rmb(); | |
1170 | if (cwq == get_wq_data(work)) { | |
dc186ad7 | 1171 | debug_work_deactivate(work); |
6e84d644 | 1172 | list_del_init(&work->entry); |
73f53c4a | 1173 | cwq_dec_nr_in_flight(cwq, get_work_color(work)); |
6e84d644 ON |
1174 | ret = 1; |
1175 | } | |
1176 | } | |
1177 | spin_unlock_irq(&cwq->lock); | |
1178 | ||
1179 | return ret; | |
1180 | } | |
1181 | ||
1182 | static void wait_on_cpu_work(struct cpu_workqueue_struct *cwq, | |
b89deed3 ON |
1183 | struct work_struct *work) |
1184 | { | |
1185 | struct wq_barrier barr; | |
affee4b2 | 1186 | struct worker *worker; |
b89deed3 ON |
1187 | |
1188 | spin_lock_irq(&cwq->lock); | |
affee4b2 TH |
1189 | |
1190 | worker = NULL; | |
c34056a3 | 1191 | if (unlikely(cwq->worker && cwq->worker->current_work == work)) { |
affee4b2 TH |
1192 | worker = cwq->worker; |
1193 | insert_wq_barrier(cwq, &barr, work, worker); | |
b89deed3 | 1194 | } |
affee4b2 | 1195 | |
b89deed3 ON |
1196 | spin_unlock_irq(&cwq->lock); |
1197 | ||
affee4b2 | 1198 | if (unlikely(worker)) { |
b89deed3 | 1199 | wait_for_completion(&barr.done); |
dc186ad7 TG |
1200 | destroy_work_on_stack(&barr.work); |
1201 | } | |
b89deed3 ON |
1202 | } |
1203 | ||
6e84d644 | 1204 | static void wait_on_work(struct work_struct *work) |
b89deed3 ON |
1205 | { |
1206 | struct cpu_workqueue_struct *cwq; | |
28e53bdd | 1207 | struct workqueue_struct *wq; |
b1f4ec17 | 1208 | int cpu; |
b89deed3 | 1209 | |
f293ea92 ON |
1210 | might_sleep(); |
1211 | ||
3295f0ef IM |
1212 | lock_map_acquire(&work->lockdep_map); |
1213 | lock_map_release(&work->lockdep_map); | |
4e6045f1 | 1214 | |
b89deed3 | 1215 | cwq = get_wq_data(work); |
b89deed3 | 1216 | if (!cwq) |
3af24433 | 1217 | return; |
b89deed3 | 1218 | |
28e53bdd | 1219 | wq = cwq->wq; |
28e53bdd | 1220 | |
1537663f | 1221 | for_each_possible_cpu(cpu) |
4690c4ab | 1222 | wait_on_cpu_work(get_cwq(cpu, wq), work); |
6e84d644 ON |
1223 | } |
1224 | ||
1f1f642e ON |
1225 | static int __cancel_work_timer(struct work_struct *work, |
1226 | struct timer_list* timer) | |
1227 | { | |
1228 | int ret; | |
1229 | ||
1230 | do { | |
1231 | ret = (timer && likely(del_timer(timer))); | |
1232 | if (!ret) | |
1233 | ret = try_to_grab_pending(work); | |
1234 | wait_on_work(work); | |
1235 | } while (unlikely(ret < 0)); | |
1236 | ||
4d707b9f | 1237 | clear_wq_data(work); |
1f1f642e ON |
1238 | return ret; |
1239 | } | |
1240 | ||
6e84d644 ON |
1241 | /** |
1242 | * cancel_work_sync - block until a work_struct's callback has terminated | |
1243 | * @work: the work which is to be flushed | |
1244 | * | |
1f1f642e ON |
1245 | * Returns true if @work was pending. |
1246 | * | |
6e84d644 ON |
1247 | * cancel_work_sync() will cancel the work if it is queued. If the work's |
1248 | * callback appears to be running, cancel_work_sync() will block until it | |
1249 | * has completed. | |
1250 | * | |
1251 | * It is possible to use this function if the work re-queues itself. It can | |
1252 | * cancel the work even if it migrates to another workqueue, however in that | |
1253 | * case it only guarantees that work->func() has completed on the last queued | |
1254 | * workqueue. | |
1255 | * | |
1256 | * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not | |
1257 | * pending, otherwise it goes into a busy-wait loop until the timer expires. | |
1258 | * | |
1259 | * The caller must ensure that workqueue_struct on which this work was last | |
1260 | * queued can't be destroyed before this function returns. | |
1261 | */ | |
1f1f642e | 1262 | int cancel_work_sync(struct work_struct *work) |
6e84d644 | 1263 | { |
1f1f642e | 1264 | return __cancel_work_timer(work, NULL); |
b89deed3 | 1265 | } |
28e53bdd | 1266 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 1267 | |
6e84d644 | 1268 | /** |
f5a421a4 | 1269 | * cancel_delayed_work_sync - reliably kill off a delayed work. |
6e84d644 ON |
1270 | * @dwork: the delayed work struct |
1271 | * | |
1f1f642e ON |
1272 | * Returns true if @dwork was pending. |
1273 | * | |
6e84d644 ON |
1274 | * It is possible to use this function if @dwork rearms itself via queue_work() |
1275 | * or queue_delayed_work(). See also the comment for cancel_work_sync(). | |
1276 | */ | |
1f1f642e | 1277 | int cancel_delayed_work_sync(struct delayed_work *dwork) |
6e84d644 | 1278 | { |
1f1f642e | 1279 | return __cancel_work_timer(&dwork->work, &dwork->timer); |
6e84d644 | 1280 | } |
f5a421a4 | 1281 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 1282 | |
6e84d644 | 1283 | static struct workqueue_struct *keventd_wq __read_mostly; |
1da177e4 | 1284 | |
0fcb78c2 REB |
1285 | /** |
1286 | * schedule_work - put work task in global workqueue | |
1287 | * @work: job to be done | |
1288 | * | |
5b0f437d BVA |
1289 | * Returns zero if @work was already on the kernel-global workqueue and |
1290 | * non-zero otherwise. | |
1291 | * | |
1292 | * This puts a job in the kernel-global workqueue if it was not already | |
1293 | * queued and leaves it in the same position on the kernel-global | |
1294 | * workqueue otherwise. | |
0fcb78c2 | 1295 | */ |
7ad5b3a5 | 1296 | int schedule_work(struct work_struct *work) |
1da177e4 LT |
1297 | { |
1298 | return queue_work(keventd_wq, work); | |
1299 | } | |
ae90dd5d | 1300 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 1301 | |
c1a220e7 ZR |
1302 | /* |
1303 | * schedule_work_on - put work task on a specific cpu | |
1304 | * @cpu: cpu to put the work task on | |
1305 | * @work: job to be done | |
1306 | * | |
1307 | * This puts a job on a specific cpu | |
1308 | */ | |
1309 | int schedule_work_on(int cpu, struct work_struct *work) | |
1310 | { | |
1311 | return queue_work_on(cpu, keventd_wq, work); | |
1312 | } | |
1313 | EXPORT_SYMBOL(schedule_work_on); | |
1314 | ||
0fcb78c2 REB |
1315 | /** |
1316 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
1317 | * @dwork: job to be done |
1318 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
1319 | * |
1320 | * After waiting for a given time this puts a job in the kernel-global | |
1321 | * workqueue. | |
1322 | */ | |
7ad5b3a5 | 1323 | int schedule_delayed_work(struct delayed_work *dwork, |
82f67cd9 | 1324 | unsigned long delay) |
1da177e4 | 1325 | { |
52bad64d | 1326 | return queue_delayed_work(keventd_wq, dwork, delay); |
1da177e4 | 1327 | } |
ae90dd5d | 1328 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 1329 | |
8c53e463 LT |
1330 | /** |
1331 | * flush_delayed_work - block until a dwork_struct's callback has terminated | |
1332 | * @dwork: the delayed work which is to be flushed | |
1333 | * | |
1334 | * Any timeout is cancelled, and any pending work is run immediately. | |
1335 | */ | |
1336 | void flush_delayed_work(struct delayed_work *dwork) | |
1337 | { | |
1338 | if (del_timer_sync(&dwork->timer)) { | |
4690c4ab TH |
1339 | __queue_work(get_cpu(), get_wq_data(&dwork->work)->wq, |
1340 | &dwork->work); | |
8c53e463 LT |
1341 | put_cpu(); |
1342 | } | |
1343 | flush_work(&dwork->work); | |
1344 | } | |
1345 | EXPORT_SYMBOL(flush_delayed_work); | |
1346 | ||
0fcb78c2 REB |
1347 | /** |
1348 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
1349 | * @cpu: cpu to use | |
52bad64d | 1350 | * @dwork: job to be done |
0fcb78c2 REB |
1351 | * @delay: number of jiffies to wait |
1352 | * | |
1353 | * After waiting for a given time this puts a job in the kernel-global | |
1354 | * workqueue on the specified CPU. | |
1355 | */ | |
1da177e4 | 1356 | int schedule_delayed_work_on(int cpu, |
52bad64d | 1357 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 1358 | { |
52bad64d | 1359 | return queue_delayed_work_on(cpu, keventd_wq, dwork, delay); |
1da177e4 | 1360 | } |
ae90dd5d | 1361 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 1362 | |
b6136773 AM |
1363 | /** |
1364 | * schedule_on_each_cpu - call a function on each online CPU from keventd | |
1365 | * @func: the function to call | |
b6136773 AM |
1366 | * |
1367 | * Returns zero on success. | |
1368 | * Returns -ve errno on failure. | |
1369 | * | |
b6136773 AM |
1370 | * schedule_on_each_cpu() is very slow. |
1371 | */ | |
65f27f38 | 1372 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
1373 | { |
1374 | int cpu; | |
65a64464 | 1375 | int orig = -1; |
b6136773 | 1376 | struct work_struct *works; |
15316ba8 | 1377 | |
b6136773 AM |
1378 | works = alloc_percpu(struct work_struct); |
1379 | if (!works) | |
15316ba8 | 1380 | return -ENOMEM; |
b6136773 | 1381 | |
93981800 TH |
1382 | get_online_cpus(); |
1383 | ||
65a64464 | 1384 | /* |
93981800 TH |
1385 | * When running in keventd don't schedule a work item on |
1386 | * itself. Can just call directly because the work queue is | |
1387 | * already bound. This also is faster. | |
65a64464 | 1388 | */ |
93981800 | 1389 | if (current_is_keventd()) |
65a64464 | 1390 | orig = raw_smp_processor_id(); |
65a64464 | 1391 | |
15316ba8 | 1392 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
1393 | struct work_struct *work = per_cpu_ptr(works, cpu); |
1394 | ||
1395 | INIT_WORK(work, func); | |
65a64464 | 1396 | if (cpu != orig) |
93981800 | 1397 | schedule_work_on(cpu, work); |
65a64464 | 1398 | } |
93981800 TH |
1399 | if (orig >= 0) |
1400 | func(per_cpu_ptr(works, orig)); | |
1401 | ||
1402 | for_each_online_cpu(cpu) | |
1403 | flush_work(per_cpu_ptr(works, cpu)); | |
1404 | ||
95402b38 | 1405 | put_online_cpus(); |
b6136773 | 1406 | free_percpu(works); |
15316ba8 CL |
1407 | return 0; |
1408 | } | |
1409 | ||
eef6a7d5 AS |
1410 | /** |
1411 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
1412 | * | |
1413 | * Forces execution of the kernel-global workqueue and blocks until its | |
1414 | * completion. | |
1415 | * | |
1416 | * Think twice before calling this function! It's very easy to get into | |
1417 | * trouble if you don't take great care. Either of the following situations | |
1418 | * will lead to deadlock: | |
1419 | * | |
1420 | * One of the work items currently on the workqueue needs to acquire | |
1421 | * a lock held by your code or its caller. | |
1422 | * | |
1423 | * Your code is running in the context of a work routine. | |
1424 | * | |
1425 | * They will be detected by lockdep when they occur, but the first might not | |
1426 | * occur very often. It depends on what work items are on the workqueue and | |
1427 | * what locks they need, which you have no control over. | |
1428 | * | |
1429 | * In most situations flushing the entire workqueue is overkill; you merely | |
1430 | * need to know that a particular work item isn't queued and isn't running. | |
1431 | * In such cases you should use cancel_delayed_work_sync() or | |
1432 | * cancel_work_sync() instead. | |
1433 | */ | |
1da177e4 LT |
1434 | void flush_scheduled_work(void) |
1435 | { | |
1436 | flush_workqueue(keventd_wq); | |
1437 | } | |
ae90dd5d | 1438 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 1439 | |
1fa44eca JB |
1440 | /** |
1441 | * execute_in_process_context - reliably execute the routine with user context | |
1442 | * @fn: the function to execute | |
1fa44eca JB |
1443 | * @ew: guaranteed storage for the execute work structure (must |
1444 | * be available when the work executes) | |
1445 | * | |
1446 | * Executes the function immediately if process context is available, | |
1447 | * otherwise schedules the function for delayed execution. | |
1448 | * | |
1449 | * Returns: 0 - function was executed | |
1450 | * 1 - function was scheduled for execution | |
1451 | */ | |
65f27f38 | 1452 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
1453 | { |
1454 | if (!in_interrupt()) { | |
65f27f38 | 1455 | fn(&ew->work); |
1fa44eca JB |
1456 | return 0; |
1457 | } | |
1458 | ||
65f27f38 | 1459 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
1460 | schedule_work(&ew->work); |
1461 | ||
1462 | return 1; | |
1463 | } | |
1464 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
1465 | ||
1da177e4 LT |
1466 | int keventd_up(void) |
1467 | { | |
1468 | return keventd_wq != NULL; | |
1469 | } | |
1470 | ||
1471 | int current_is_keventd(void) | |
1472 | { | |
1473 | struct cpu_workqueue_struct *cwq; | |
d243769d | 1474 | int cpu = raw_smp_processor_id(); /* preempt-safe: keventd is per-cpu */ |
1da177e4 LT |
1475 | int ret = 0; |
1476 | ||
1477 | BUG_ON(!keventd_wq); | |
1478 | ||
1537663f | 1479 | cwq = get_cwq(cpu, keventd_wq); |
c34056a3 | 1480 | if (current == cwq->worker->task) |
1da177e4 LT |
1481 | ret = 1; |
1482 | ||
1483 | return ret; | |
1484 | ||
1485 | } | |
1486 | ||
0f900049 TH |
1487 | static struct cpu_workqueue_struct *alloc_cwqs(void) |
1488 | { | |
1489 | /* | |
1490 | * cwqs are forced aligned according to WORK_STRUCT_FLAG_BITS. | |
1491 | * Make sure that the alignment isn't lower than that of | |
1492 | * unsigned long long. | |
1493 | */ | |
1494 | const size_t size = sizeof(struct cpu_workqueue_struct); | |
1495 | const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS, | |
1496 | __alignof__(unsigned long long)); | |
1497 | struct cpu_workqueue_struct *cwqs; | |
1498 | #ifndef CONFIG_SMP | |
1499 | void *ptr; | |
1500 | ||
1501 | /* | |
1502 | * On UP, percpu allocator doesn't honor alignment parameter | |
1503 | * and simply uses arch-dependent default. Allocate enough | |
1504 | * room to align cwq and put an extra pointer at the end | |
1505 | * pointing back to the originally allocated pointer which | |
1506 | * will be used for free. | |
1507 | * | |
1508 | * FIXME: This really belongs to UP percpu code. Update UP | |
1509 | * percpu code to honor alignment and remove this ugliness. | |
1510 | */ | |
1511 | ptr = __alloc_percpu(size + align + sizeof(void *), 1); | |
1512 | cwqs = PTR_ALIGN(ptr, align); | |
1513 | *(void **)per_cpu_ptr(cwqs + 1, 0) = ptr; | |
1514 | #else | |
1515 | /* On SMP, percpu allocator can do it itself */ | |
1516 | cwqs = __alloc_percpu(size, align); | |
1517 | #endif | |
1518 | /* just in case, make sure it's actually aligned */ | |
1519 | BUG_ON(!IS_ALIGNED((unsigned long)cwqs, align)); | |
1520 | return cwqs; | |
1521 | } | |
1522 | ||
1523 | static void free_cwqs(struct cpu_workqueue_struct *cwqs) | |
1524 | { | |
1525 | #ifndef CONFIG_SMP | |
1526 | /* on UP, the pointer to free is stored right after the cwq */ | |
1527 | if (cwqs) | |
1528 | free_percpu(*(void **)per_cpu_ptr(cwqs + 1, 0)); | |
1529 | #else | |
1530 | free_percpu(cwqs); | |
1531 | #endif | |
1532 | } | |
1533 | ||
4e6045f1 | 1534 | struct workqueue_struct *__create_workqueue_key(const char *name, |
97e37d7b | 1535 | unsigned int flags, |
1e19ffc6 | 1536 | int max_active, |
eb13ba87 JB |
1537 | struct lock_class_key *key, |
1538 | const char *lock_name) | |
1da177e4 | 1539 | { |
1537663f | 1540 | bool singlethread = flags & WQ_SINGLE_THREAD; |
1da177e4 | 1541 | struct workqueue_struct *wq; |
c34056a3 TH |
1542 | bool failed = false; |
1543 | unsigned int cpu; | |
1da177e4 | 1544 | |
1e19ffc6 TH |
1545 | max_active = clamp_val(max_active, 1, INT_MAX); |
1546 | ||
3af24433 ON |
1547 | wq = kzalloc(sizeof(*wq), GFP_KERNEL); |
1548 | if (!wq) | |
4690c4ab | 1549 | goto err; |
3af24433 | 1550 | |
0f900049 | 1551 | wq->cpu_wq = alloc_cwqs(); |
4690c4ab TH |
1552 | if (!wq->cpu_wq) |
1553 | goto err; | |
3af24433 | 1554 | |
97e37d7b | 1555 | wq->flags = flags; |
73f53c4a TH |
1556 | mutex_init(&wq->flush_mutex); |
1557 | atomic_set(&wq->nr_cwqs_to_flush, 0); | |
1558 | INIT_LIST_HEAD(&wq->flusher_queue); | |
1559 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
3af24433 | 1560 | wq->name = name; |
eb13ba87 | 1561 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 1562 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 1563 | |
1537663f TH |
1564 | cpu_maps_update_begin(); |
1565 | /* | |
1566 | * We must initialize cwqs for each possible cpu even if we | |
1567 | * are going to call destroy_workqueue() finally. Otherwise | |
1568 | * cpu_up() can hit the uninitialized cwq once we drop the | |
1569 | * lock. | |
1570 | */ | |
1571 | for_each_possible_cpu(cpu) { | |
1572 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
1573 | ||
0f900049 | 1574 | BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK); |
1537663f | 1575 | cwq->cpu = cpu; |
c34056a3 | 1576 | cwq->wq = wq; |
73f53c4a | 1577 | cwq->flush_color = -1; |
1e19ffc6 | 1578 | cwq->max_active = max_active; |
1537663f TH |
1579 | spin_lock_init(&cwq->lock); |
1580 | INIT_LIST_HEAD(&cwq->worklist); | |
1e19ffc6 | 1581 | INIT_LIST_HEAD(&cwq->delayed_works); |
1537663f TH |
1582 | init_waitqueue_head(&cwq->more_work); |
1583 | ||
c34056a3 | 1584 | if (failed) |
1537663f | 1585 | continue; |
c34056a3 TH |
1586 | cwq->worker = create_worker(cwq, |
1587 | cpu_online(cpu) && !singlethread); | |
1588 | if (cwq->worker) | |
1589 | start_worker(cwq->worker); | |
1537663f | 1590 | else |
c34056a3 | 1591 | failed = true; |
3af24433 ON |
1592 | } |
1593 | ||
1537663f TH |
1594 | spin_lock(&workqueue_lock); |
1595 | list_add(&wq->list, &workqueues); | |
1596 | spin_unlock(&workqueue_lock); | |
1597 | ||
1598 | cpu_maps_update_done(); | |
1599 | ||
c34056a3 | 1600 | if (failed) { |
3af24433 ON |
1601 | destroy_workqueue(wq); |
1602 | wq = NULL; | |
1603 | } | |
1604 | return wq; | |
4690c4ab TH |
1605 | err: |
1606 | if (wq) { | |
0f900049 | 1607 | free_cwqs(wq->cpu_wq); |
4690c4ab TH |
1608 | kfree(wq); |
1609 | } | |
1610 | return NULL; | |
3af24433 | 1611 | } |
4e6045f1 | 1612 | EXPORT_SYMBOL_GPL(__create_workqueue_key); |
1da177e4 | 1613 | |
3af24433 ON |
1614 | /** |
1615 | * destroy_workqueue - safely terminate a workqueue | |
1616 | * @wq: target workqueue | |
1617 | * | |
1618 | * Safely destroy a workqueue. All work currently pending will be done first. | |
1619 | */ | |
1620 | void destroy_workqueue(struct workqueue_struct *wq) | |
1621 | { | |
b1f4ec17 | 1622 | int cpu; |
3af24433 | 1623 | |
3da1c84c | 1624 | cpu_maps_update_begin(); |
95402b38 | 1625 | spin_lock(&workqueue_lock); |
b1f4ec17 | 1626 | list_del(&wq->list); |
95402b38 | 1627 | spin_unlock(&workqueue_lock); |
1537663f | 1628 | cpu_maps_update_done(); |
3af24433 | 1629 | |
73f53c4a TH |
1630 | flush_workqueue(wq); |
1631 | ||
1632 | for_each_possible_cpu(cpu) { | |
1633 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
1634 | int i; | |
1635 | ||
c34056a3 TH |
1636 | if (cwq->worker) { |
1637 | destroy_worker(cwq->worker); | |
1638 | cwq->worker = NULL; | |
73f53c4a TH |
1639 | } |
1640 | ||
1641 | for (i = 0; i < WORK_NR_COLORS; i++) | |
1642 | BUG_ON(cwq->nr_in_flight[i]); | |
1e19ffc6 TH |
1643 | BUG_ON(cwq->nr_active); |
1644 | BUG_ON(!list_empty(&cwq->delayed_works)); | |
73f53c4a | 1645 | } |
9b41ea72 | 1646 | |
0f900049 | 1647 | free_cwqs(wq->cpu_wq); |
3af24433 ON |
1648 | kfree(wq); |
1649 | } | |
1650 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
1651 | ||
1652 | static int __devinit workqueue_cpu_callback(struct notifier_block *nfb, | |
1653 | unsigned long action, | |
1654 | void *hcpu) | |
1655 | { | |
1656 | unsigned int cpu = (unsigned long)hcpu; | |
1657 | struct cpu_workqueue_struct *cwq; | |
1658 | struct workqueue_struct *wq; | |
1659 | ||
8bb78442 RW |
1660 | action &= ~CPU_TASKS_FROZEN; |
1661 | ||
3af24433 | 1662 | list_for_each_entry(wq, &workqueues, list) { |
1537663f TH |
1663 | if (wq->flags & WQ_SINGLE_THREAD) |
1664 | continue; | |
3af24433 | 1665 | |
1537663f | 1666 | cwq = get_cwq(cpu, wq); |
3af24433 | 1667 | |
1537663f | 1668 | switch (action) { |
3da1c84c | 1669 | case CPU_POST_DEAD: |
73f53c4a | 1670 | flush_workqueue(wq); |
3af24433 ON |
1671 | break; |
1672 | } | |
1da177e4 LT |
1673 | } |
1674 | ||
1537663f | 1675 | return notifier_from_errno(0); |
1da177e4 | 1676 | } |
1da177e4 | 1677 | |
2d3854a3 | 1678 | #ifdef CONFIG_SMP |
8ccad40d | 1679 | |
2d3854a3 | 1680 | struct work_for_cpu { |
6b44003e | 1681 | struct completion completion; |
2d3854a3 RR |
1682 | long (*fn)(void *); |
1683 | void *arg; | |
1684 | long ret; | |
1685 | }; | |
1686 | ||
6b44003e | 1687 | static int do_work_for_cpu(void *_wfc) |
2d3854a3 | 1688 | { |
6b44003e | 1689 | struct work_for_cpu *wfc = _wfc; |
2d3854a3 | 1690 | wfc->ret = wfc->fn(wfc->arg); |
6b44003e AM |
1691 | complete(&wfc->completion); |
1692 | return 0; | |
2d3854a3 RR |
1693 | } |
1694 | ||
1695 | /** | |
1696 | * work_on_cpu - run a function in user context on a particular cpu | |
1697 | * @cpu: the cpu to run on | |
1698 | * @fn: the function to run | |
1699 | * @arg: the function arg | |
1700 | * | |
31ad9081 RR |
1701 | * This will return the value @fn returns. |
1702 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 1703 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 RR |
1704 | */ |
1705 | long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) | |
1706 | { | |
6b44003e AM |
1707 | struct task_struct *sub_thread; |
1708 | struct work_for_cpu wfc = { | |
1709 | .completion = COMPLETION_INITIALIZER_ONSTACK(wfc.completion), | |
1710 | .fn = fn, | |
1711 | .arg = arg, | |
1712 | }; | |
1713 | ||
1714 | sub_thread = kthread_create(do_work_for_cpu, &wfc, "work_for_cpu"); | |
1715 | if (IS_ERR(sub_thread)) | |
1716 | return PTR_ERR(sub_thread); | |
1717 | kthread_bind(sub_thread, cpu); | |
1718 | wake_up_process(sub_thread); | |
1719 | wait_for_completion(&wfc.completion); | |
2d3854a3 RR |
1720 | return wfc.ret; |
1721 | } | |
1722 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
1723 | #endif /* CONFIG_SMP */ | |
1724 | ||
c12920d1 | 1725 | void __init init_workqueues(void) |
1da177e4 | 1726 | { |
c34056a3 TH |
1727 | unsigned int cpu; |
1728 | ||
1729 | for_each_possible_cpu(cpu) | |
1730 | ida_init(&per_cpu(worker_ida, cpu)); | |
1731 | ||
e7577c50 | 1732 | singlethread_cpu = cpumask_first(cpu_possible_mask); |
1da177e4 LT |
1733 | hotcpu_notifier(workqueue_cpu_callback, 0); |
1734 | keventd_wq = create_workqueue("events"); | |
1735 | BUG_ON(!keventd_wq); | |
1736 | } |