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 | 80 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 81 | int max_active; /* L: max active works */ |
1e19ffc6 | 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 | ||
a0a1a5fd | 111 | int saved_max_active; /* I: saved cwq max_active */ |
4690c4ab | 112 | const char *name; /* I: workqueue name */ |
4e6045f1 | 113 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 114 | struct lockdep_map lockdep_map; |
4e6045f1 | 115 | #endif |
1da177e4 LT |
116 | }; |
117 | ||
dc186ad7 TG |
118 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
119 | ||
120 | static struct debug_obj_descr work_debug_descr; | |
121 | ||
122 | /* | |
123 | * fixup_init is called when: | |
124 | * - an active object is initialized | |
125 | */ | |
126 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
127 | { | |
128 | struct work_struct *work = addr; | |
129 | ||
130 | switch (state) { | |
131 | case ODEBUG_STATE_ACTIVE: | |
132 | cancel_work_sync(work); | |
133 | debug_object_init(work, &work_debug_descr); | |
134 | return 1; | |
135 | default: | |
136 | return 0; | |
137 | } | |
138 | } | |
139 | ||
140 | /* | |
141 | * fixup_activate is called when: | |
142 | * - an active object is activated | |
143 | * - an unknown object is activated (might be a statically initialized object) | |
144 | */ | |
145 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
146 | { | |
147 | struct work_struct *work = addr; | |
148 | ||
149 | switch (state) { | |
150 | ||
151 | case ODEBUG_STATE_NOTAVAILABLE: | |
152 | /* | |
153 | * This is not really a fixup. The work struct was | |
154 | * statically initialized. We just make sure that it | |
155 | * is tracked in the object tracker. | |
156 | */ | |
22df02bb | 157 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
158 | debug_object_init(work, &work_debug_descr); |
159 | debug_object_activate(work, &work_debug_descr); | |
160 | return 0; | |
161 | } | |
162 | WARN_ON_ONCE(1); | |
163 | return 0; | |
164 | ||
165 | case ODEBUG_STATE_ACTIVE: | |
166 | WARN_ON(1); | |
167 | ||
168 | default: | |
169 | return 0; | |
170 | } | |
171 | } | |
172 | ||
173 | /* | |
174 | * fixup_free is called when: | |
175 | * - an active object is freed | |
176 | */ | |
177 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
178 | { | |
179 | struct work_struct *work = addr; | |
180 | ||
181 | switch (state) { | |
182 | case ODEBUG_STATE_ACTIVE: | |
183 | cancel_work_sync(work); | |
184 | debug_object_free(work, &work_debug_descr); | |
185 | return 1; | |
186 | default: | |
187 | return 0; | |
188 | } | |
189 | } | |
190 | ||
191 | static struct debug_obj_descr work_debug_descr = { | |
192 | .name = "work_struct", | |
193 | .fixup_init = work_fixup_init, | |
194 | .fixup_activate = work_fixup_activate, | |
195 | .fixup_free = work_fixup_free, | |
196 | }; | |
197 | ||
198 | static inline void debug_work_activate(struct work_struct *work) | |
199 | { | |
200 | debug_object_activate(work, &work_debug_descr); | |
201 | } | |
202 | ||
203 | static inline void debug_work_deactivate(struct work_struct *work) | |
204 | { | |
205 | debug_object_deactivate(work, &work_debug_descr); | |
206 | } | |
207 | ||
208 | void __init_work(struct work_struct *work, int onstack) | |
209 | { | |
210 | if (onstack) | |
211 | debug_object_init_on_stack(work, &work_debug_descr); | |
212 | else | |
213 | debug_object_init(work, &work_debug_descr); | |
214 | } | |
215 | EXPORT_SYMBOL_GPL(__init_work); | |
216 | ||
217 | void destroy_work_on_stack(struct work_struct *work) | |
218 | { | |
219 | debug_object_free(work, &work_debug_descr); | |
220 | } | |
221 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
222 | ||
223 | #else | |
224 | static inline void debug_work_activate(struct work_struct *work) { } | |
225 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
226 | #endif | |
227 | ||
95402b38 GS |
228 | /* Serializes the accesses to the list of workqueues. */ |
229 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 230 | static LIST_HEAD(workqueues); |
c34056a3 | 231 | static DEFINE_PER_CPU(struct ida, worker_ida); |
a0a1a5fd | 232 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 TH |
233 | |
234 | static int worker_thread(void *__worker); | |
1da177e4 | 235 | |
3af24433 | 236 | static int singlethread_cpu __read_mostly; |
1da177e4 | 237 | |
1537663f TH |
238 | static struct cpu_workqueue_struct *get_cwq(unsigned int cpu, |
239 | struct workqueue_struct *wq) | |
b1f4ec17 | 240 | { |
1537663f | 241 | return per_cpu_ptr(wq->cpu_wq, cpu); |
b1f4ec17 ON |
242 | } |
243 | ||
1537663f TH |
244 | static struct cpu_workqueue_struct *target_cwq(unsigned int cpu, |
245 | struct workqueue_struct *wq) | |
a848e3b6 | 246 | { |
1537663f | 247 | if (unlikely(wq->flags & WQ_SINGLE_THREAD)) |
a848e3b6 | 248 | cpu = singlethread_cpu; |
1537663f | 249 | return get_cwq(cpu, wq); |
a848e3b6 ON |
250 | } |
251 | ||
73f53c4a TH |
252 | static unsigned int work_color_to_flags(int color) |
253 | { | |
254 | return color << WORK_STRUCT_COLOR_SHIFT; | |
255 | } | |
256 | ||
257 | static int get_work_color(struct work_struct *work) | |
258 | { | |
259 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
260 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
261 | } | |
262 | ||
263 | static int work_next_color(int color) | |
264 | { | |
265 | return (color + 1) % WORK_NR_COLORS; | |
266 | } | |
267 | ||
4594bf15 DH |
268 | /* |
269 | * Set the workqueue on which a work item is to be run | |
270 | * - Must *only* be called if the pending flag is set | |
271 | */ | |
ed7c0fee | 272 | static inline void set_wq_data(struct work_struct *work, |
4690c4ab TH |
273 | struct cpu_workqueue_struct *cwq, |
274 | unsigned long extra_flags) | |
365970a1 | 275 | { |
4594bf15 | 276 | BUG_ON(!work_pending(work)); |
365970a1 | 277 | |
4690c4ab | 278 | atomic_long_set(&work->data, (unsigned long)cwq | work_static(work) | |
22df02bb | 279 | WORK_STRUCT_PENDING | extra_flags); |
365970a1 DH |
280 | } |
281 | ||
4d707b9f ON |
282 | /* |
283 | * Clear WORK_STRUCT_PENDING and the workqueue on which it was queued. | |
284 | */ | |
285 | static inline void clear_wq_data(struct work_struct *work) | |
286 | { | |
4690c4ab | 287 | atomic_long_set(&work->data, work_static(work)); |
4d707b9f ON |
288 | } |
289 | ||
64166699 | 290 | static inline struct cpu_workqueue_struct *get_wq_data(struct work_struct *work) |
365970a1 | 291 | { |
64166699 TH |
292 | return (void *)(atomic_long_read(&work->data) & |
293 | WORK_STRUCT_WQ_DATA_MASK); | |
365970a1 DH |
294 | } |
295 | ||
4690c4ab TH |
296 | /** |
297 | * insert_work - insert a work into cwq | |
298 | * @cwq: cwq @work belongs to | |
299 | * @work: work to insert | |
300 | * @head: insertion point | |
301 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
302 | * | |
303 | * Insert @work into @cwq after @head. | |
304 | * | |
305 | * CONTEXT: | |
306 | * spin_lock_irq(cwq->lock). | |
307 | */ | |
b89deed3 | 308 | static void insert_work(struct cpu_workqueue_struct *cwq, |
4690c4ab TH |
309 | struct work_struct *work, struct list_head *head, |
310 | unsigned int extra_flags) | |
b89deed3 | 311 | { |
4690c4ab TH |
312 | /* we own @work, set data and link */ |
313 | set_wq_data(work, cwq, extra_flags); | |
314 | ||
6e84d644 ON |
315 | /* |
316 | * Ensure that we get the right work->data if we see the | |
317 | * result of list_add() below, see try_to_grab_pending(). | |
318 | */ | |
319 | smp_wmb(); | |
4690c4ab | 320 | |
1a4d9b0a | 321 | list_add_tail(&work->entry, head); |
b89deed3 ON |
322 | wake_up(&cwq->more_work); |
323 | } | |
324 | ||
4690c4ab | 325 | static void __queue_work(unsigned int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
326 | struct work_struct *work) |
327 | { | |
1537663f | 328 | struct cpu_workqueue_struct *cwq = target_cwq(cpu, wq); |
1e19ffc6 | 329 | struct list_head *worklist; |
1da177e4 LT |
330 | unsigned long flags; |
331 | ||
dc186ad7 | 332 | debug_work_activate(work); |
1e19ffc6 | 333 | |
1da177e4 | 334 | spin_lock_irqsave(&cwq->lock, flags); |
4690c4ab | 335 | BUG_ON(!list_empty(&work->entry)); |
1e19ffc6 | 336 | |
73f53c4a | 337 | cwq->nr_in_flight[cwq->work_color]++; |
1e19ffc6 TH |
338 | |
339 | if (likely(cwq->nr_active < cwq->max_active)) { | |
340 | cwq->nr_active++; | |
341 | worklist = &cwq->worklist; | |
342 | } else | |
343 | worklist = &cwq->delayed_works; | |
344 | ||
345 | insert_work(cwq, work, worklist, work_color_to_flags(cwq->work_color)); | |
346 | ||
1da177e4 LT |
347 | spin_unlock_irqrestore(&cwq->lock, flags); |
348 | } | |
349 | ||
0fcb78c2 REB |
350 | /** |
351 | * queue_work - queue work on a workqueue | |
352 | * @wq: workqueue to use | |
353 | * @work: work to queue | |
354 | * | |
057647fc | 355 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
1da177e4 | 356 | * |
00dfcaf7 ON |
357 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
358 | * it can be processed by another CPU. | |
1da177e4 | 359 | */ |
7ad5b3a5 | 360 | int queue_work(struct workqueue_struct *wq, struct work_struct *work) |
1da177e4 | 361 | { |
ef1ca236 ON |
362 | int ret; |
363 | ||
364 | ret = queue_work_on(get_cpu(), wq, work); | |
365 | put_cpu(); | |
366 | ||
1da177e4 LT |
367 | return ret; |
368 | } | |
ae90dd5d | 369 | EXPORT_SYMBOL_GPL(queue_work); |
1da177e4 | 370 | |
c1a220e7 ZR |
371 | /** |
372 | * queue_work_on - queue work on specific cpu | |
373 | * @cpu: CPU number to execute work on | |
374 | * @wq: workqueue to use | |
375 | * @work: work to queue | |
376 | * | |
377 | * Returns 0 if @work was already on a queue, non-zero otherwise. | |
378 | * | |
379 | * We queue the work to a specific CPU, the caller must ensure it | |
380 | * can't go away. | |
381 | */ | |
382 | int | |
383 | queue_work_on(int cpu, struct workqueue_struct *wq, struct work_struct *work) | |
384 | { | |
385 | int ret = 0; | |
386 | ||
22df02bb | 387 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 388 | __queue_work(cpu, wq, work); |
c1a220e7 ZR |
389 | ret = 1; |
390 | } | |
391 | return ret; | |
392 | } | |
393 | EXPORT_SYMBOL_GPL(queue_work_on); | |
394 | ||
6d141c3f | 395 | static void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 396 | { |
52bad64d | 397 | struct delayed_work *dwork = (struct delayed_work *)__data; |
ed7c0fee | 398 | struct cpu_workqueue_struct *cwq = get_wq_data(&dwork->work); |
1da177e4 | 399 | |
4690c4ab | 400 | __queue_work(smp_processor_id(), cwq->wq, &dwork->work); |
1da177e4 LT |
401 | } |
402 | ||
0fcb78c2 REB |
403 | /** |
404 | * queue_delayed_work - queue work on a workqueue after delay | |
405 | * @wq: workqueue to use | |
af9997e4 | 406 | * @dwork: delayable work to queue |
0fcb78c2 REB |
407 | * @delay: number of jiffies to wait before queueing |
408 | * | |
057647fc | 409 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 410 | */ |
7ad5b3a5 | 411 | int queue_delayed_work(struct workqueue_struct *wq, |
52bad64d | 412 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 413 | { |
52bad64d | 414 | if (delay == 0) |
63bc0362 | 415 | return queue_work(wq, &dwork->work); |
1da177e4 | 416 | |
63bc0362 | 417 | return queue_delayed_work_on(-1, wq, dwork, delay); |
1da177e4 | 418 | } |
ae90dd5d | 419 | EXPORT_SYMBOL_GPL(queue_delayed_work); |
1da177e4 | 420 | |
0fcb78c2 REB |
421 | /** |
422 | * queue_delayed_work_on - queue work on specific CPU after delay | |
423 | * @cpu: CPU number to execute work on | |
424 | * @wq: workqueue to use | |
af9997e4 | 425 | * @dwork: work to queue |
0fcb78c2 REB |
426 | * @delay: number of jiffies to wait before queueing |
427 | * | |
057647fc | 428 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 429 | */ |
7a6bc1cd | 430 | int queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
52bad64d | 431 | struct delayed_work *dwork, unsigned long delay) |
7a6bc1cd VP |
432 | { |
433 | int ret = 0; | |
52bad64d DH |
434 | struct timer_list *timer = &dwork->timer; |
435 | struct work_struct *work = &dwork->work; | |
7a6bc1cd | 436 | |
22df02bb | 437 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7a6bc1cd VP |
438 | BUG_ON(timer_pending(timer)); |
439 | BUG_ON(!list_empty(&work->entry)); | |
440 | ||
8a3e77cc AL |
441 | timer_stats_timer_set_start_info(&dwork->timer); |
442 | ||
ed7c0fee | 443 | /* This stores cwq for the moment, for the timer_fn */ |
1537663f | 444 | set_wq_data(work, target_cwq(raw_smp_processor_id(), wq), 0); |
7a6bc1cd | 445 | timer->expires = jiffies + delay; |
52bad64d | 446 | timer->data = (unsigned long)dwork; |
7a6bc1cd | 447 | timer->function = delayed_work_timer_fn; |
63bc0362 ON |
448 | |
449 | if (unlikely(cpu >= 0)) | |
450 | add_timer_on(timer, cpu); | |
451 | else | |
452 | add_timer(timer); | |
7a6bc1cd VP |
453 | ret = 1; |
454 | } | |
455 | return ret; | |
456 | } | |
ae90dd5d | 457 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
1da177e4 | 458 | |
c34056a3 TH |
459 | static struct worker *alloc_worker(void) |
460 | { | |
461 | struct worker *worker; | |
462 | ||
463 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
affee4b2 TH |
464 | if (worker) |
465 | INIT_LIST_HEAD(&worker->scheduled); | |
c34056a3 TH |
466 | return worker; |
467 | } | |
468 | ||
469 | /** | |
470 | * create_worker - create a new workqueue worker | |
471 | * @cwq: cwq the new worker will belong to | |
472 | * @bind: whether to set affinity to @cpu or not | |
473 | * | |
474 | * Create a new worker which is bound to @cwq. The returned worker | |
475 | * can be started by calling start_worker() or destroyed using | |
476 | * destroy_worker(). | |
477 | * | |
478 | * CONTEXT: | |
479 | * Might sleep. Does GFP_KERNEL allocations. | |
480 | * | |
481 | * RETURNS: | |
482 | * Pointer to the newly created worker. | |
483 | */ | |
484 | static struct worker *create_worker(struct cpu_workqueue_struct *cwq, bool bind) | |
485 | { | |
486 | int id = -1; | |
487 | struct worker *worker = NULL; | |
488 | ||
489 | spin_lock(&workqueue_lock); | |
490 | while (ida_get_new(&per_cpu(worker_ida, cwq->cpu), &id)) { | |
491 | spin_unlock(&workqueue_lock); | |
492 | if (!ida_pre_get(&per_cpu(worker_ida, cwq->cpu), GFP_KERNEL)) | |
493 | goto fail; | |
494 | spin_lock(&workqueue_lock); | |
495 | } | |
496 | spin_unlock(&workqueue_lock); | |
497 | ||
498 | worker = alloc_worker(); | |
499 | if (!worker) | |
500 | goto fail; | |
501 | ||
502 | worker->cwq = cwq; | |
503 | worker->id = id; | |
504 | ||
505 | worker->task = kthread_create(worker_thread, worker, "kworker/%u:%d", | |
506 | cwq->cpu, id); | |
507 | if (IS_ERR(worker->task)) | |
508 | goto fail; | |
509 | ||
510 | if (bind) | |
511 | kthread_bind(worker->task, cwq->cpu); | |
512 | ||
513 | return worker; | |
514 | fail: | |
515 | if (id >= 0) { | |
516 | spin_lock(&workqueue_lock); | |
517 | ida_remove(&per_cpu(worker_ida, cwq->cpu), id); | |
518 | spin_unlock(&workqueue_lock); | |
519 | } | |
520 | kfree(worker); | |
521 | return NULL; | |
522 | } | |
523 | ||
524 | /** | |
525 | * start_worker - start a newly created worker | |
526 | * @worker: worker to start | |
527 | * | |
528 | * Start @worker. | |
529 | * | |
530 | * CONTEXT: | |
531 | * spin_lock_irq(cwq->lock). | |
532 | */ | |
533 | static void start_worker(struct worker *worker) | |
534 | { | |
535 | wake_up_process(worker->task); | |
536 | } | |
537 | ||
538 | /** | |
539 | * destroy_worker - destroy a workqueue worker | |
540 | * @worker: worker to be destroyed | |
541 | * | |
542 | * Destroy @worker. | |
543 | */ | |
544 | static void destroy_worker(struct worker *worker) | |
545 | { | |
546 | int cpu = worker->cwq->cpu; | |
547 | int id = worker->id; | |
548 | ||
549 | /* sanity check frenzy */ | |
550 | BUG_ON(worker->current_work); | |
affee4b2 | 551 | BUG_ON(!list_empty(&worker->scheduled)); |
c34056a3 TH |
552 | |
553 | kthread_stop(worker->task); | |
554 | kfree(worker); | |
555 | ||
556 | spin_lock(&workqueue_lock); | |
557 | ida_remove(&per_cpu(worker_ida, cpu), id); | |
558 | spin_unlock(&workqueue_lock); | |
559 | } | |
560 | ||
affee4b2 TH |
561 | /** |
562 | * move_linked_works - move linked works to a list | |
563 | * @work: start of series of works to be scheduled | |
564 | * @head: target list to append @work to | |
565 | * @nextp: out paramter for nested worklist walking | |
566 | * | |
567 | * Schedule linked works starting from @work to @head. Work series to | |
568 | * be scheduled starts at @work and includes any consecutive work with | |
569 | * WORK_STRUCT_LINKED set in its predecessor. | |
570 | * | |
571 | * If @nextp is not NULL, it's updated to point to the next work of | |
572 | * the last scheduled work. This allows move_linked_works() to be | |
573 | * nested inside outer list_for_each_entry_safe(). | |
574 | * | |
575 | * CONTEXT: | |
576 | * spin_lock_irq(cwq->lock). | |
577 | */ | |
578 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
579 | struct work_struct **nextp) | |
580 | { | |
581 | struct work_struct *n; | |
582 | ||
583 | /* | |
584 | * Linked worklist will always end before the end of the list, | |
585 | * use NULL for list head. | |
586 | */ | |
587 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
588 | list_move_tail(&work->entry, head); | |
589 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
590 | break; | |
591 | } | |
592 | ||
593 | /* | |
594 | * If we're already inside safe list traversal and have moved | |
595 | * multiple works to the scheduled queue, the next position | |
596 | * needs to be updated. | |
597 | */ | |
598 | if (nextp) | |
599 | *nextp = n; | |
600 | } | |
601 | ||
1e19ffc6 TH |
602 | static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq) |
603 | { | |
604 | struct work_struct *work = list_first_entry(&cwq->delayed_works, | |
605 | struct work_struct, entry); | |
606 | ||
607 | move_linked_works(work, &cwq->worklist, NULL); | |
608 | cwq->nr_active++; | |
609 | } | |
610 | ||
73f53c4a TH |
611 | /** |
612 | * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight | |
613 | * @cwq: cwq of interest | |
614 | * @color: color of work which left the queue | |
615 | * | |
616 | * A work either has completed or is removed from pending queue, | |
617 | * decrement nr_in_flight of its cwq and handle workqueue flushing. | |
618 | * | |
619 | * CONTEXT: | |
620 | * spin_lock_irq(cwq->lock). | |
621 | */ | |
622 | static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color) | |
623 | { | |
624 | /* ignore uncolored works */ | |
625 | if (color == WORK_NO_COLOR) | |
626 | return; | |
627 | ||
628 | cwq->nr_in_flight[color]--; | |
1e19ffc6 TH |
629 | cwq->nr_active--; |
630 | ||
631 | /* one down, submit a delayed one */ | |
632 | if (!list_empty(&cwq->delayed_works) && | |
633 | cwq->nr_active < cwq->max_active) | |
634 | cwq_activate_first_delayed(cwq); | |
73f53c4a TH |
635 | |
636 | /* is flush in progress and are we at the flushing tip? */ | |
637 | if (likely(cwq->flush_color != color)) | |
638 | return; | |
639 | ||
640 | /* are there still in-flight works? */ | |
641 | if (cwq->nr_in_flight[color]) | |
642 | return; | |
643 | ||
644 | /* this cwq is done, clear flush_color */ | |
645 | cwq->flush_color = -1; | |
646 | ||
647 | /* | |
648 | * If this was the last cwq, wake up the first flusher. It | |
649 | * will handle the rest. | |
650 | */ | |
651 | if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush)) | |
652 | complete(&cwq->wq->first_flusher->done); | |
653 | } | |
654 | ||
a62428c0 TH |
655 | /** |
656 | * process_one_work - process single work | |
c34056a3 | 657 | * @worker: self |
a62428c0 TH |
658 | * @work: work to process |
659 | * | |
660 | * Process @work. This function contains all the logics necessary to | |
661 | * process a single work including synchronization against and | |
662 | * interaction with other workers on the same cpu, queueing and | |
663 | * flushing. As long as context requirement is met, any worker can | |
664 | * call this function to process a work. | |
665 | * | |
666 | * CONTEXT: | |
667 | * spin_lock_irq(cwq->lock) which is released and regrabbed. | |
668 | */ | |
c34056a3 | 669 | static void process_one_work(struct worker *worker, struct work_struct *work) |
a62428c0 | 670 | { |
c34056a3 | 671 | struct cpu_workqueue_struct *cwq = worker->cwq; |
a62428c0 | 672 | work_func_t f = work->func; |
73f53c4a | 673 | int work_color; |
a62428c0 TH |
674 | #ifdef CONFIG_LOCKDEP |
675 | /* | |
676 | * It is permissible to free the struct work_struct from | |
677 | * inside the function that is called from it, this we need to | |
678 | * take into account for lockdep too. To avoid bogus "held | |
679 | * lock freed" warnings as well as problems when looking into | |
680 | * work->lockdep_map, make a copy and use that here. | |
681 | */ | |
682 | struct lockdep_map lockdep_map = work->lockdep_map; | |
683 | #endif | |
684 | /* claim and process */ | |
a62428c0 | 685 | debug_work_deactivate(work); |
c34056a3 | 686 | worker->current_work = work; |
73f53c4a | 687 | work_color = get_work_color(work); |
a62428c0 TH |
688 | list_del_init(&work->entry); |
689 | ||
690 | spin_unlock_irq(&cwq->lock); | |
691 | ||
692 | BUG_ON(get_wq_data(work) != cwq); | |
693 | work_clear_pending(work); | |
694 | lock_map_acquire(&cwq->wq->lockdep_map); | |
695 | lock_map_acquire(&lockdep_map); | |
696 | f(work); | |
697 | lock_map_release(&lockdep_map); | |
698 | lock_map_release(&cwq->wq->lockdep_map); | |
699 | ||
700 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
701 | printk(KERN_ERR "BUG: workqueue leaked lock or atomic: " | |
702 | "%s/0x%08x/%d\n", | |
703 | current->comm, preempt_count(), task_pid_nr(current)); | |
704 | printk(KERN_ERR " last function: "); | |
705 | print_symbol("%s\n", (unsigned long)f); | |
706 | debug_show_held_locks(current); | |
707 | dump_stack(); | |
708 | } | |
709 | ||
710 | spin_lock_irq(&cwq->lock); | |
711 | ||
712 | /* we're done with it, release */ | |
c34056a3 | 713 | worker->current_work = NULL; |
73f53c4a | 714 | cwq_dec_nr_in_flight(cwq, work_color); |
a62428c0 TH |
715 | } |
716 | ||
affee4b2 TH |
717 | /** |
718 | * process_scheduled_works - process scheduled works | |
719 | * @worker: self | |
720 | * | |
721 | * Process all scheduled works. Please note that the scheduled list | |
722 | * may change while processing a work, so this function repeatedly | |
723 | * fetches a work from the top and executes it. | |
724 | * | |
725 | * CONTEXT: | |
726 | * spin_lock_irq(cwq->lock) which may be released and regrabbed | |
727 | * multiple times. | |
728 | */ | |
729 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 730 | { |
affee4b2 TH |
731 | while (!list_empty(&worker->scheduled)) { |
732 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 733 | struct work_struct, entry); |
c34056a3 | 734 | process_one_work(worker, work); |
1da177e4 | 735 | } |
1da177e4 LT |
736 | } |
737 | ||
4690c4ab TH |
738 | /** |
739 | * worker_thread - the worker thread function | |
c34056a3 | 740 | * @__worker: self |
4690c4ab TH |
741 | * |
742 | * The cwq worker thread function. | |
743 | */ | |
c34056a3 | 744 | static int worker_thread(void *__worker) |
1da177e4 | 745 | { |
c34056a3 TH |
746 | struct worker *worker = __worker; |
747 | struct cpu_workqueue_struct *cwq = worker->cwq; | |
3af24433 | 748 | DEFINE_WAIT(wait); |
1da177e4 | 749 | |
3af24433 | 750 | for (;;) { |
3af24433 | 751 | prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE); |
a0a1a5fd | 752 | if (!kthread_should_stop() && |
14441960 | 753 | list_empty(&cwq->worklist)) |
1da177e4 | 754 | schedule(); |
3af24433 ON |
755 | finish_wait(&cwq->more_work, &wait); |
756 | ||
14441960 | 757 | if (kthread_should_stop()) |
3af24433 | 758 | break; |
1da177e4 | 759 | |
c34056a3 | 760 | if (unlikely(!cpumask_equal(&worker->task->cpus_allowed, |
1537663f | 761 | get_cpu_mask(cwq->cpu)))) |
c34056a3 | 762 | set_cpus_allowed_ptr(worker->task, |
1537663f | 763 | get_cpu_mask(cwq->cpu)); |
affee4b2 TH |
764 | |
765 | spin_lock_irq(&cwq->lock); | |
766 | ||
767 | while (!list_empty(&cwq->worklist)) { | |
768 | struct work_struct *work = | |
769 | list_first_entry(&cwq->worklist, | |
770 | struct work_struct, entry); | |
771 | ||
772 | if (likely(!(*work_data_bits(work) & | |
773 | WORK_STRUCT_LINKED))) { | |
774 | /* optimization path, not strictly necessary */ | |
775 | process_one_work(worker, work); | |
776 | if (unlikely(!list_empty(&worker->scheduled))) | |
777 | process_scheduled_works(worker); | |
778 | } else { | |
779 | move_linked_works(work, &worker->scheduled, | |
780 | NULL); | |
781 | process_scheduled_works(worker); | |
782 | } | |
783 | } | |
784 | ||
785 | spin_unlock_irq(&cwq->lock); | |
1da177e4 | 786 | } |
3af24433 | 787 | |
1da177e4 LT |
788 | return 0; |
789 | } | |
790 | ||
fc2e4d70 ON |
791 | struct wq_barrier { |
792 | struct work_struct work; | |
793 | struct completion done; | |
794 | }; | |
795 | ||
796 | static void wq_barrier_func(struct work_struct *work) | |
797 | { | |
798 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
799 | complete(&barr->done); | |
800 | } | |
801 | ||
4690c4ab TH |
802 | /** |
803 | * insert_wq_barrier - insert a barrier work | |
804 | * @cwq: cwq to insert barrier into | |
805 | * @barr: wq_barrier to insert | |
affee4b2 TH |
806 | * @target: target work to attach @barr to |
807 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 808 | * |
affee4b2 TH |
809 | * @barr is linked to @target such that @barr is completed only after |
810 | * @target finishes execution. Please note that the ordering | |
811 | * guarantee is observed only with respect to @target and on the local | |
812 | * cpu. | |
813 | * | |
814 | * Currently, a queued barrier can't be canceled. This is because | |
815 | * try_to_grab_pending() can't determine whether the work to be | |
816 | * grabbed is at the head of the queue and thus can't clear LINKED | |
817 | * flag of the previous work while there must be a valid next work | |
818 | * after a work with LINKED flag set. | |
819 | * | |
820 | * Note that when @worker is non-NULL, @target may be modified | |
821 | * underneath us, so we can't reliably determine cwq from @target. | |
4690c4ab TH |
822 | * |
823 | * CONTEXT: | |
824 | * spin_lock_irq(cwq->lock). | |
825 | */ | |
83c22520 | 826 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, |
affee4b2 TH |
827 | struct wq_barrier *barr, |
828 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 829 | { |
affee4b2 TH |
830 | struct list_head *head; |
831 | unsigned int linked = 0; | |
832 | ||
dc186ad7 TG |
833 | /* |
834 | * debugobject calls are safe here even with cwq->lock locked | |
835 | * as we know for sure that this will not trigger any of the | |
836 | * checks and call back into the fixup functions where we | |
837 | * might deadlock. | |
838 | */ | |
839 | INIT_WORK_ON_STACK(&barr->work, wq_barrier_func); | |
22df02bb | 840 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 841 | init_completion(&barr->done); |
83c22520 | 842 | |
affee4b2 TH |
843 | /* |
844 | * If @target is currently being executed, schedule the | |
845 | * barrier to the worker; otherwise, put it after @target. | |
846 | */ | |
847 | if (worker) | |
848 | head = worker->scheduled.next; | |
849 | else { | |
850 | unsigned long *bits = work_data_bits(target); | |
851 | ||
852 | head = target->entry.next; | |
853 | /* there can already be other linked works, inherit and set */ | |
854 | linked = *bits & WORK_STRUCT_LINKED; | |
855 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
856 | } | |
857 | ||
dc186ad7 | 858 | debug_work_activate(&barr->work); |
affee4b2 TH |
859 | insert_work(cwq, &barr->work, head, |
860 | work_color_to_flags(WORK_NO_COLOR) | linked); | |
fc2e4d70 ON |
861 | } |
862 | ||
73f53c4a TH |
863 | /** |
864 | * flush_workqueue_prep_cwqs - prepare cwqs for workqueue flushing | |
865 | * @wq: workqueue being flushed | |
866 | * @flush_color: new flush color, < 0 for no-op | |
867 | * @work_color: new work color, < 0 for no-op | |
868 | * | |
869 | * Prepare cwqs for workqueue flushing. | |
870 | * | |
871 | * If @flush_color is non-negative, flush_color on all cwqs should be | |
872 | * -1. If no cwq has in-flight commands at the specified color, all | |
873 | * cwq->flush_color's stay at -1 and %false is returned. If any cwq | |
874 | * has in flight commands, its cwq->flush_color is set to | |
875 | * @flush_color, @wq->nr_cwqs_to_flush is updated accordingly, cwq | |
876 | * wakeup logic is armed and %true is returned. | |
877 | * | |
878 | * The caller should have initialized @wq->first_flusher prior to | |
879 | * calling this function with non-negative @flush_color. If | |
880 | * @flush_color is negative, no flush color update is done and %false | |
881 | * is returned. | |
882 | * | |
883 | * If @work_color is non-negative, all cwqs should have the same | |
884 | * work_color which is previous to @work_color and all will be | |
885 | * advanced to @work_color. | |
886 | * | |
887 | * CONTEXT: | |
888 | * mutex_lock(wq->flush_mutex). | |
889 | * | |
890 | * RETURNS: | |
891 | * %true if @flush_color >= 0 and there's something to flush. %false | |
892 | * otherwise. | |
893 | */ | |
894 | static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq, | |
895 | int flush_color, int work_color) | |
1da177e4 | 896 | { |
73f53c4a TH |
897 | bool wait = false; |
898 | unsigned int cpu; | |
1da177e4 | 899 | |
73f53c4a TH |
900 | if (flush_color >= 0) { |
901 | BUG_ON(atomic_read(&wq->nr_cwqs_to_flush)); | |
902 | atomic_set(&wq->nr_cwqs_to_flush, 1); | |
1da177e4 | 903 | } |
2355b70f | 904 | |
73f53c4a TH |
905 | for_each_possible_cpu(cpu) { |
906 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
907 | ||
908 | spin_lock_irq(&cwq->lock); | |
909 | ||
910 | if (flush_color >= 0) { | |
911 | BUG_ON(cwq->flush_color != -1); | |
912 | ||
913 | if (cwq->nr_in_flight[flush_color]) { | |
914 | cwq->flush_color = flush_color; | |
915 | atomic_inc(&wq->nr_cwqs_to_flush); | |
916 | wait = true; | |
917 | } | |
918 | } | |
919 | ||
920 | if (work_color >= 0) { | |
921 | BUG_ON(work_color != work_next_color(cwq->work_color)); | |
922 | cwq->work_color = work_color; | |
923 | } | |
924 | ||
925 | spin_unlock_irq(&cwq->lock); | |
dc186ad7 | 926 | } |
14441960 | 927 | |
73f53c4a TH |
928 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_cwqs_to_flush)) |
929 | complete(&wq->first_flusher->done); | |
930 | ||
931 | return wait; | |
1da177e4 LT |
932 | } |
933 | ||
0fcb78c2 | 934 | /** |
1da177e4 | 935 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 936 | * @wq: workqueue to flush |
1da177e4 LT |
937 | * |
938 | * Forces execution of the workqueue and blocks until its completion. | |
939 | * This is typically used in driver shutdown handlers. | |
940 | * | |
fc2e4d70 ON |
941 | * We sleep until all works which were queued on entry have been handled, |
942 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 943 | */ |
7ad5b3a5 | 944 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 945 | { |
73f53c4a TH |
946 | struct wq_flusher this_flusher = { |
947 | .list = LIST_HEAD_INIT(this_flusher.list), | |
948 | .flush_color = -1, | |
949 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
950 | }; | |
951 | int next_color; | |
1da177e4 | 952 | |
3295f0ef IM |
953 | lock_map_acquire(&wq->lockdep_map); |
954 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
955 | |
956 | mutex_lock(&wq->flush_mutex); | |
957 | ||
958 | /* | |
959 | * Start-to-wait phase | |
960 | */ | |
961 | next_color = work_next_color(wq->work_color); | |
962 | ||
963 | if (next_color != wq->flush_color) { | |
964 | /* | |
965 | * Color space is not full. The current work_color | |
966 | * becomes our flush_color and work_color is advanced | |
967 | * by one. | |
968 | */ | |
969 | BUG_ON(!list_empty(&wq->flusher_overflow)); | |
970 | this_flusher.flush_color = wq->work_color; | |
971 | wq->work_color = next_color; | |
972 | ||
973 | if (!wq->first_flusher) { | |
974 | /* no flush in progress, become the first flusher */ | |
975 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
976 | ||
977 | wq->first_flusher = &this_flusher; | |
978 | ||
979 | if (!flush_workqueue_prep_cwqs(wq, wq->flush_color, | |
980 | wq->work_color)) { | |
981 | /* nothing to flush, done */ | |
982 | wq->flush_color = next_color; | |
983 | wq->first_flusher = NULL; | |
984 | goto out_unlock; | |
985 | } | |
986 | } else { | |
987 | /* wait in queue */ | |
988 | BUG_ON(wq->flush_color == this_flusher.flush_color); | |
989 | list_add_tail(&this_flusher.list, &wq->flusher_queue); | |
990 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
991 | } | |
992 | } else { | |
993 | /* | |
994 | * Oops, color space is full, wait on overflow queue. | |
995 | * The next flush completion will assign us | |
996 | * flush_color and transfer to flusher_queue. | |
997 | */ | |
998 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
999 | } | |
1000 | ||
1001 | mutex_unlock(&wq->flush_mutex); | |
1002 | ||
1003 | wait_for_completion(&this_flusher.done); | |
1004 | ||
1005 | /* | |
1006 | * Wake-up-and-cascade phase | |
1007 | * | |
1008 | * First flushers are responsible for cascading flushes and | |
1009 | * handling overflow. Non-first flushers can simply return. | |
1010 | */ | |
1011 | if (wq->first_flusher != &this_flusher) | |
1012 | return; | |
1013 | ||
1014 | mutex_lock(&wq->flush_mutex); | |
1015 | ||
1016 | wq->first_flusher = NULL; | |
1017 | ||
1018 | BUG_ON(!list_empty(&this_flusher.list)); | |
1019 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
1020 | ||
1021 | while (true) { | |
1022 | struct wq_flusher *next, *tmp; | |
1023 | ||
1024 | /* complete all the flushers sharing the current flush color */ | |
1025 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
1026 | if (next->flush_color != wq->flush_color) | |
1027 | break; | |
1028 | list_del_init(&next->list); | |
1029 | complete(&next->done); | |
1030 | } | |
1031 | ||
1032 | BUG_ON(!list_empty(&wq->flusher_overflow) && | |
1033 | wq->flush_color != work_next_color(wq->work_color)); | |
1034 | ||
1035 | /* this flush_color is finished, advance by one */ | |
1036 | wq->flush_color = work_next_color(wq->flush_color); | |
1037 | ||
1038 | /* one color has been freed, handle overflow queue */ | |
1039 | if (!list_empty(&wq->flusher_overflow)) { | |
1040 | /* | |
1041 | * Assign the same color to all overflowed | |
1042 | * flushers, advance work_color and append to | |
1043 | * flusher_queue. This is the start-to-wait | |
1044 | * phase for these overflowed flushers. | |
1045 | */ | |
1046 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
1047 | tmp->flush_color = wq->work_color; | |
1048 | ||
1049 | wq->work_color = work_next_color(wq->work_color); | |
1050 | ||
1051 | list_splice_tail_init(&wq->flusher_overflow, | |
1052 | &wq->flusher_queue); | |
1053 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
1054 | } | |
1055 | ||
1056 | if (list_empty(&wq->flusher_queue)) { | |
1057 | BUG_ON(wq->flush_color != wq->work_color); | |
1058 | break; | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * Need to flush more colors. Make the next flusher | |
1063 | * the new first flusher and arm cwqs. | |
1064 | */ | |
1065 | BUG_ON(wq->flush_color == wq->work_color); | |
1066 | BUG_ON(wq->flush_color != next->flush_color); | |
1067 | ||
1068 | list_del_init(&next->list); | |
1069 | wq->first_flusher = next; | |
1070 | ||
1071 | if (flush_workqueue_prep_cwqs(wq, wq->flush_color, -1)) | |
1072 | break; | |
1073 | ||
1074 | /* | |
1075 | * Meh... this color is already done, clear first | |
1076 | * flusher and repeat cascading. | |
1077 | */ | |
1078 | wq->first_flusher = NULL; | |
1079 | } | |
1080 | ||
1081 | out_unlock: | |
1082 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 1083 | } |
ae90dd5d | 1084 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 1085 | |
db700897 ON |
1086 | /** |
1087 | * flush_work - block until a work_struct's callback has terminated | |
1088 | * @work: the work which is to be flushed | |
1089 | * | |
a67da70d ON |
1090 | * Returns false if @work has already terminated. |
1091 | * | |
db700897 ON |
1092 | * It is expected that, prior to calling flush_work(), the caller has |
1093 | * arranged for the work to not be requeued, otherwise it doesn't make | |
1094 | * sense to use this function. | |
1095 | */ | |
1096 | int flush_work(struct work_struct *work) | |
1097 | { | |
affee4b2 | 1098 | struct worker *worker = NULL; |
db700897 | 1099 | struct cpu_workqueue_struct *cwq; |
db700897 ON |
1100 | struct wq_barrier barr; |
1101 | ||
1102 | might_sleep(); | |
1103 | cwq = get_wq_data(work); | |
1104 | if (!cwq) | |
1105 | return 0; | |
1106 | ||
3295f0ef IM |
1107 | lock_map_acquire(&cwq->wq->lockdep_map); |
1108 | lock_map_release(&cwq->wq->lockdep_map); | |
a67da70d | 1109 | |
db700897 ON |
1110 | spin_lock_irq(&cwq->lock); |
1111 | if (!list_empty(&work->entry)) { | |
1112 | /* | |
1113 | * See the comment near try_to_grab_pending()->smp_rmb(). | |
1114 | * If it was re-queued under us we are not going to wait. | |
1115 | */ | |
1116 | smp_rmb(); | |
1117 | if (unlikely(cwq != get_wq_data(work))) | |
4690c4ab | 1118 | goto already_gone; |
db700897 | 1119 | } else { |
affee4b2 TH |
1120 | if (cwq->worker && cwq->worker->current_work == work) |
1121 | worker = cwq->worker; | |
1122 | if (!worker) | |
4690c4ab | 1123 | goto already_gone; |
db700897 | 1124 | } |
db700897 | 1125 | |
affee4b2 | 1126 | insert_wq_barrier(cwq, &barr, work, worker); |
4690c4ab | 1127 | spin_unlock_irq(&cwq->lock); |
db700897 | 1128 | wait_for_completion(&barr.done); |
dc186ad7 | 1129 | destroy_work_on_stack(&barr.work); |
db700897 | 1130 | return 1; |
4690c4ab TH |
1131 | already_gone: |
1132 | spin_unlock_irq(&cwq->lock); | |
1133 | return 0; | |
db700897 ON |
1134 | } |
1135 | EXPORT_SYMBOL_GPL(flush_work); | |
1136 | ||
6e84d644 | 1137 | /* |
1f1f642e | 1138 | * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit, |
6e84d644 ON |
1139 | * so this work can't be re-armed in any way. |
1140 | */ | |
1141 | static int try_to_grab_pending(struct work_struct *work) | |
1142 | { | |
1143 | struct cpu_workqueue_struct *cwq; | |
1f1f642e | 1144 | int ret = -1; |
6e84d644 | 1145 | |
22df02bb | 1146 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1f1f642e | 1147 | return 0; |
6e84d644 ON |
1148 | |
1149 | /* | |
1150 | * The queueing is in progress, or it is already queued. Try to | |
1151 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1152 | */ | |
1153 | ||
1154 | cwq = get_wq_data(work); | |
1155 | if (!cwq) | |
1156 | return ret; | |
1157 | ||
1158 | spin_lock_irq(&cwq->lock); | |
1159 | if (!list_empty(&work->entry)) { | |
1160 | /* | |
1161 | * This work is queued, but perhaps we locked the wrong cwq. | |
1162 | * In that case we must see the new value after rmb(), see | |
1163 | * insert_work()->wmb(). | |
1164 | */ | |
1165 | smp_rmb(); | |
1166 | if (cwq == get_wq_data(work)) { | |
dc186ad7 | 1167 | debug_work_deactivate(work); |
6e84d644 | 1168 | list_del_init(&work->entry); |
73f53c4a | 1169 | cwq_dec_nr_in_flight(cwq, get_work_color(work)); |
6e84d644 ON |
1170 | ret = 1; |
1171 | } | |
1172 | } | |
1173 | spin_unlock_irq(&cwq->lock); | |
1174 | ||
1175 | return ret; | |
1176 | } | |
1177 | ||
1178 | static void wait_on_cpu_work(struct cpu_workqueue_struct *cwq, | |
b89deed3 ON |
1179 | struct work_struct *work) |
1180 | { | |
1181 | struct wq_barrier barr; | |
affee4b2 | 1182 | struct worker *worker; |
b89deed3 ON |
1183 | |
1184 | spin_lock_irq(&cwq->lock); | |
affee4b2 TH |
1185 | |
1186 | worker = NULL; | |
c34056a3 | 1187 | if (unlikely(cwq->worker && cwq->worker->current_work == work)) { |
affee4b2 TH |
1188 | worker = cwq->worker; |
1189 | insert_wq_barrier(cwq, &barr, work, worker); | |
b89deed3 | 1190 | } |
affee4b2 | 1191 | |
b89deed3 ON |
1192 | spin_unlock_irq(&cwq->lock); |
1193 | ||
affee4b2 | 1194 | if (unlikely(worker)) { |
b89deed3 | 1195 | wait_for_completion(&barr.done); |
dc186ad7 TG |
1196 | destroy_work_on_stack(&barr.work); |
1197 | } | |
b89deed3 ON |
1198 | } |
1199 | ||
6e84d644 | 1200 | static void wait_on_work(struct work_struct *work) |
b89deed3 ON |
1201 | { |
1202 | struct cpu_workqueue_struct *cwq; | |
28e53bdd | 1203 | struct workqueue_struct *wq; |
b1f4ec17 | 1204 | int cpu; |
b89deed3 | 1205 | |
f293ea92 ON |
1206 | might_sleep(); |
1207 | ||
3295f0ef IM |
1208 | lock_map_acquire(&work->lockdep_map); |
1209 | lock_map_release(&work->lockdep_map); | |
4e6045f1 | 1210 | |
b89deed3 | 1211 | cwq = get_wq_data(work); |
b89deed3 | 1212 | if (!cwq) |
3af24433 | 1213 | return; |
b89deed3 | 1214 | |
28e53bdd | 1215 | wq = cwq->wq; |
28e53bdd | 1216 | |
1537663f | 1217 | for_each_possible_cpu(cpu) |
4690c4ab | 1218 | wait_on_cpu_work(get_cwq(cpu, wq), work); |
6e84d644 ON |
1219 | } |
1220 | ||
1f1f642e ON |
1221 | static int __cancel_work_timer(struct work_struct *work, |
1222 | struct timer_list* timer) | |
1223 | { | |
1224 | int ret; | |
1225 | ||
1226 | do { | |
1227 | ret = (timer && likely(del_timer(timer))); | |
1228 | if (!ret) | |
1229 | ret = try_to_grab_pending(work); | |
1230 | wait_on_work(work); | |
1231 | } while (unlikely(ret < 0)); | |
1232 | ||
4d707b9f | 1233 | clear_wq_data(work); |
1f1f642e ON |
1234 | return ret; |
1235 | } | |
1236 | ||
6e84d644 ON |
1237 | /** |
1238 | * cancel_work_sync - block until a work_struct's callback has terminated | |
1239 | * @work: the work which is to be flushed | |
1240 | * | |
1f1f642e ON |
1241 | * Returns true if @work was pending. |
1242 | * | |
6e84d644 ON |
1243 | * cancel_work_sync() will cancel the work if it is queued. If the work's |
1244 | * callback appears to be running, cancel_work_sync() will block until it | |
1245 | * has completed. | |
1246 | * | |
1247 | * It is possible to use this function if the work re-queues itself. It can | |
1248 | * cancel the work even if it migrates to another workqueue, however in that | |
1249 | * case it only guarantees that work->func() has completed on the last queued | |
1250 | * workqueue. | |
1251 | * | |
1252 | * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not | |
1253 | * pending, otherwise it goes into a busy-wait loop until the timer expires. | |
1254 | * | |
1255 | * The caller must ensure that workqueue_struct on which this work was last | |
1256 | * queued can't be destroyed before this function returns. | |
1257 | */ | |
1f1f642e | 1258 | int cancel_work_sync(struct work_struct *work) |
6e84d644 | 1259 | { |
1f1f642e | 1260 | return __cancel_work_timer(work, NULL); |
b89deed3 | 1261 | } |
28e53bdd | 1262 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 1263 | |
6e84d644 | 1264 | /** |
f5a421a4 | 1265 | * cancel_delayed_work_sync - reliably kill off a delayed work. |
6e84d644 ON |
1266 | * @dwork: the delayed work struct |
1267 | * | |
1f1f642e ON |
1268 | * Returns true if @dwork was pending. |
1269 | * | |
6e84d644 ON |
1270 | * It is possible to use this function if @dwork rearms itself via queue_work() |
1271 | * or queue_delayed_work(). See also the comment for cancel_work_sync(). | |
1272 | */ | |
1f1f642e | 1273 | int cancel_delayed_work_sync(struct delayed_work *dwork) |
6e84d644 | 1274 | { |
1f1f642e | 1275 | return __cancel_work_timer(&dwork->work, &dwork->timer); |
6e84d644 | 1276 | } |
f5a421a4 | 1277 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 1278 | |
6e84d644 | 1279 | static struct workqueue_struct *keventd_wq __read_mostly; |
1da177e4 | 1280 | |
0fcb78c2 REB |
1281 | /** |
1282 | * schedule_work - put work task in global workqueue | |
1283 | * @work: job to be done | |
1284 | * | |
5b0f437d BVA |
1285 | * Returns zero if @work was already on the kernel-global workqueue and |
1286 | * non-zero otherwise. | |
1287 | * | |
1288 | * This puts a job in the kernel-global workqueue if it was not already | |
1289 | * queued and leaves it in the same position on the kernel-global | |
1290 | * workqueue otherwise. | |
0fcb78c2 | 1291 | */ |
7ad5b3a5 | 1292 | int schedule_work(struct work_struct *work) |
1da177e4 LT |
1293 | { |
1294 | return queue_work(keventd_wq, work); | |
1295 | } | |
ae90dd5d | 1296 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 1297 | |
c1a220e7 ZR |
1298 | /* |
1299 | * schedule_work_on - put work task on a specific cpu | |
1300 | * @cpu: cpu to put the work task on | |
1301 | * @work: job to be done | |
1302 | * | |
1303 | * This puts a job on a specific cpu | |
1304 | */ | |
1305 | int schedule_work_on(int cpu, struct work_struct *work) | |
1306 | { | |
1307 | return queue_work_on(cpu, keventd_wq, work); | |
1308 | } | |
1309 | EXPORT_SYMBOL(schedule_work_on); | |
1310 | ||
0fcb78c2 REB |
1311 | /** |
1312 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
1313 | * @dwork: job to be done |
1314 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
1315 | * |
1316 | * After waiting for a given time this puts a job in the kernel-global | |
1317 | * workqueue. | |
1318 | */ | |
7ad5b3a5 | 1319 | int schedule_delayed_work(struct delayed_work *dwork, |
82f67cd9 | 1320 | unsigned long delay) |
1da177e4 | 1321 | { |
52bad64d | 1322 | return queue_delayed_work(keventd_wq, dwork, delay); |
1da177e4 | 1323 | } |
ae90dd5d | 1324 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 1325 | |
8c53e463 LT |
1326 | /** |
1327 | * flush_delayed_work - block until a dwork_struct's callback has terminated | |
1328 | * @dwork: the delayed work which is to be flushed | |
1329 | * | |
1330 | * Any timeout is cancelled, and any pending work is run immediately. | |
1331 | */ | |
1332 | void flush_delayed_work(struct delayed_work *dwork) | |
1333 | { | |
1334 | if (del_timer_sync(&dwork->timer)) { | |
4690c4ab TH |
1335 | __queue_work(get_cpu(), get_wq_data(&dwork->work)->wq, |
1336 | &dwork->work); | |
8c53e463 LT |
1337 | put_cpu(); |
1338 | } | |
1339 | flush_work(&dwork->work); | |
1340 | } | |
1341 | EXPORT_SYMBOL(flush_delayed_work); | |
1342 | ||
0fcb78c2 REB |
1343 | /** |
1344 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
1345 | * @cpu: cpu to use | |
52bad64d | 1346 | * @dwork: job to be done |
0fcb78c2 REB |
1347 | * @delay: number of jiffies to wait |
1348 | * | |
1349 | * After waiting for a given time this puts a job in the kernel-global | |
1350 | * workqueue on the specified CPU. | |
1351 | */ | |
1da177e4 | 1352 | int schedule_delayed_work_on(int cpu, |
52bad64d | 1353 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 1354 | { |
52bad64d | 1355 | return queue_delayed_work_on(cpu, keventd_wq, dwork, delay); |
1da177e4 | 1356 | } |
ae90dd5d | 1357 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 1358 | |
b6136773 AM |
1359 | /** |
1360 | * schedule_on_each_cpu - call a function on each online CPU from keventd | |
1361 | * @func: the function to call | |
b6136773 AM |
1362 | * |
1363 | * Returns zero on success. | |
1364 | * Returns -ve errno on failure. | |
1365 | * | |
b6136773 AM |
1366 | * schedule_on_each_cpu() is very slow. |
1367 | */ | |
65f27f38 | 1368 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
1369 | { |
1370 | int cpu; | |
65a64464 | 1371 | int orig = -1; |
b6136773 | 1372 | struct work_struct *works; |
15316ba8 | 1373 | |
b6136773 AM |
1374 | works = alloc_percpu(struct work_struct); |
1375 | if (!works) | |
15316ba8 | 1376 | return -ENOMEM; |
b6136773 | 1377 | |
93981800 TH |
1378 | get_online_cpus(); |
1379 | ||
65a64464 | 1380 | /* |
93981800 TH |
1381 | * When running in keventd don't schedule a work item on |
1382 | * itself. Can just call directly because the work queue is | |
1383 | * already bound. This also is faster. | |
65a64464 | 1384 | */ |
93981800 | 1385 | if (current_is_keventd()) |
65a64464 | 1386 | orig = raw_smp_processor_id(); |
65a64464 | 1387 | |
15316ba8 | 1388 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
1389 | struct work_struct *work = per_cpu_ptr(works, cpu); |
1390 | ||
1391 | INIT_WORK(work, func); | |
65a64464 | 1392 | if (cpu != orig) |
93981800 | 1393 | schedule_work_on(cpu, work); |
65a64464 | 1394 | } |
93981800 TH |
1395 | if (orig >= 0) |
1396 | func(per_cpu_ptr(works, orig)); | |
1397 | ||
1398 | for_each_online_cpu(cpu) | |
1399 | flush_work(per_cpu_ptr(works, cpu)); | |
1400 | ||
95402b38 | 1401 | put_online_cpus(); |
b6136773 | 1402 | free_percpu(works); |
15316ba8 CL |
1403 | return 0; |
1404 | } | |
1405 | ||
eef6a7d5 AS |
1406 | /** |
1407 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
1408 | * | |
1409 | * Forces execution of the kernel-global workqueue and blocks until its | |
1410 | * completion. | |
1411 | * | |
1412 | * Think twice before calling this function! It's very easy to get into | |
1413 | * trouble if you don't take great care. Either of the following situations | |
1414 | * will lead to deadlock: | |
1415 | * | |
1416 | * One of the work items currently on the workqueue needs to acquire | |
1417 | * a lock held by your code or its caller. | |
1418 | * | |
1419 | * Your code is running in the context of a work routine. | |
1420 | * | |
1421 | * They will be detected by lockdep when they occur, but the first might not | |
1422 | * occur very often. It depends on what work items are on the workqueue and | |
1423 | * what locks they need, which you have no control over. | |
1424 | * | |
1425 | * In most situations flushing the entire workqueue is overkill; you merely | |
1426 | * need to know that a particular work item isn't queued and isn't running. | |
1427 | * In such cases you should use cancel_delayed_work_sync() or | |
1428 | * cancel_work_sync() instead. | |
1429 | */ | |
1da177e4 LT |
1430 | void flush_scheduled_work(void) |
1431 | { | |
1432 | flush_workqueue(keventd_wq); | |
1433 | } | |
ae90dd5d | 1434 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 1435 | |
1fa44eca JB |
1436 | /** |
1437 | * execute_in_process_context - reliably execute the routine with user context | |
1438 | * @fn: the function to execute | |
1fa44eca JB |
1439 | * @ew: guaranteed storage for the execute work structure (must |
1440 | * be available when the work executes) | |
1441 | * | |
1442 | * Executes the function immediately if process context is available, | |
1443 | * otherwise schedules the function for delayed execution. | |
1444 | * | |
1445 | * Returns: 0 - function was executed | |
1446 | * 1 - function was scheduled for execution | |
1447 | */ | |
65f27f38 | 1448 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
1449 | { |
1450 | if (!in_interrupt()) { | |
65f27f38 | 1451 | fn(&ew->work); |
1fa44eca JB |
1452 | return 0; |
1453 | } | |
1454 | ||
65f27f38 | 1455 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
1456 | schedule_work(&ew->work); |
1457 | ||
1458 | return 1; | |
1459 | } | |
1460 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
1461 | ||
1da177e4 LT |
1462 | int keventd_up(void) |
1463 | { | |
1464 | return keventd_wq != NULL; | |
1465 | } | |
1466 | ||
1467 | int current_is_keventd(void) | |
1468 | { | |
1469 | struct cpu_workqueue_struct *cwq; | |
d243769d | 1470 | int cpu = raw_smp_processor_id(); /* preempt-safe: keventd is per-cpu */ |
1da177e4 LT |
1471 | int ret = 0; |
1472 | ||
1473 | BUG_ON(!keventd_wq); | |
1474 | ||
1537663f | 1475 | cwq = get_cwq(cpu, keventd_wq); |
c34056a3 | 1476 | if (current == cwq->worker->task) |
1da177e4 LT |
1477 | ret = 1; |
1478 | ||
1479 | return ret; | |
1480 | ||
1481 | } | |
1482 | ||
0f900049 TH |
1483 | static struct cpu_workqueue_struct *alloc_cwqs(void) |
1484 | { | |
1485 | /* | |
1486 | * cwqs are forced aligned according to WORK_STRUCT_FLAG_BITS. | |
1487 | * Make sure that the alignment isn't lower than that of | |
1488 | * unsigned long long. | |
1489 | */ | |
1490 | const size_t size = sizeof(struct cpu_workqueue_struct); | |
1491 | const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS, | |
1492 | __alignof__(unsigned long long)); | |
1493 | struct cpu_workqueue_struct *cwqs; | |
1494 | #ifndef CONFIG_SMP | |
1495 | void *ptr; | |
1496 | ||
1497 | /* | |
1498 | * On UP, percpu allocator doesn't honor alignment parameter | |
1499 | * and simply uses arch-dependent default. Allocate enough | |
1500 | * room to align cwq and put an extra pointer at the end | |
1501 | * pointing back to the originally allocated pointer which | |
1502 | * will be used for free. | |
1503 | * | |
1504 | * FIXME: This really belongs to UP percpu code. Update UP | |
1505 | * percpu code to honor alignment and remove this ugliness. | |
1506 | */ | |
1507 | ptr = __alloc_percpu(size + align + sizeof(void *), 1); | |
1508 | cwqs = PTR_ALIGN(ptr, align); | |
1509 | *(void **)per_cpu_ptr(cwqs + 1, 0) = ptr; | |
1510 | #else | |
1511 | /* On SMP, percpu allocator can do it itself */ | |
1512 | cwqs = __alloc_percpu(size, align); | |
1513 | #endif | |
1514 | /* just in case, make sure it's actually aligned */ | |
1515 | BUG_ON(!IS_ALIGNED((unsigned long)cwqs, align)); | |
1516 | return cwqs; | |
1517 | } | |
1518 | ||
1519 | static void free_cwqs(struct cpu_workqueue_struct *cwqs) | |
1520 | { | |
1521 | #ifndef CONFIG_SMP | |
1522 | /* on UP, the pointer to free is stored right after the cwq */ | |
1523 | if (cwqs) | |
1524 | free_percpu(*(void **)per_cpu_ptr(cwqs + 1, 0)); | |
1525 | #else | |
1526 | free_percpu(cwqs); | |
1527 | #endif | |
1528 | } | |
1529 | ||
4e6045f1 | 1530 | struct workqueue_struct *__create_workqueue_key(const char *name, |
97e37d7b | 1531 | unsigned int flags, |
1e19ffc6 | 1532 | int max_active, |
eb13ba87 JB |
1533 | struct lock_class_key *key, |
1534 | const char *lock_name) | |
1da177e4 | 1535 | { |
1537663f | 1536 | bool singlethread = flags & WQ_SINGLE_THREAD; |
1da177e4 | 1537 | struct workqueue_struct *wq; |
c34056a3 TH |
1538 | bool failed = false; |
1539 | unsigned int cpu; | |
1da177e4 | 1540 | |
1e19ffc6 TH |
1541 | max_active = clamp_val(max_active, 1, INT_MAX); |
1542 | ||
3af24433 ON |
1543 | wq = kzalloc(sizeof(*wq), GFP_KERNEL); |
1544 | if (!wq) | |
4690c4ab | 1545 | goto err; |
3af24433 | 1546 | |
0f900049 | 1547 | wq->cpu_wq = alloc_cwqs(); |
4690c4ab TH |
1548 | if (!wq->cpu_wq) |
1549 | goto err; | |
3af24433 | 1550 | |
97e37d7b | 1551 | wq->flags = flags; |
a0a1a5fd | 1552 | wq->saved_max_active = max_active; |
73f53c4a TH |
1553 | mutex_init(&wq->flush_mutex); |
1554 | atomic_set(&wq->nr_cwqs_to_flush, 0); | |
1555 | INIT_LIST_HEAD(&wq->flusher_queue); | |
1556 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
3af24433 | 1557 | wq->name = name; |
eb13ba87 | 1558 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 1559 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 1560 | |
1537663f TH |
1561 | cpu_maps_update_begin(); |
1562 | /* | |
1563 | * We must initialize cwqs for each possible cpu even if we | |
1564 | * are going to call destroy_workqueue() finally. Otherwise | |
1565 | * cpu_up() can hit the uninitialized cwq once we drop the | |
1566 | * lock. | |
1567 | */ | |
1568 | for_each_possible_cpu(cpu) { | |
1569 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
1570 | ||
0f900049 | 1571 | BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK); |
1537663f | 1572 | cwq->cpu = cpu; |
c34056a3 | 1573 | cwq->wq = wq; |
73f53c4a | 1574 | cwq->flush_color = -1; |
1e19ffc6 | 1575 | cwq->max_active = max_active; |
1537663f TH |
1576 | spin_lock_init(&cwq->lock); |
1577 | INIT_LIST_HEAD(&cwq->worklist); | |
1e19ffc6 | 1578 | INIT_LIST_HEAD(&cwq->delayed_works); |
1537663f TH |
1579 | init_waitqueue_head(&cwq->more_work); |
1580 | ||
c34056a3 | 1581 | if (failed) |
1537663f | 1582 | continue; |
c34056a3 TH |
1583 | cwq->worker = create_worker(cwq, |
1584 | cpu_online(cpu) && !singlethread); | |
1585 | if (cwq->worker) | |
1586 | start_worker(cwq->worker); | |
1537663f | 1587 | else |
c34056a3 | 1588 | failed = true; |
3af24433 ON |
1589 | } |
1590 | ||
a0a1a5fd TH |
1591 | /* |
1592 | * workqueue_lock protects global freeze state and workqueues | |
1593 | * list. Grab it, set max_active accordingly and add the new | |
1594 | * workqueue to workqueues list. | |
1595 | */ | |
1537663f | 1596 | spin_lock(&workqueue_lock); |
a0a1a5fd TH |
1597 | |
1598 | if (workqueue_freezing && wq->flags & WQ_FREEZEABLE) | |
1599 | for_each_possible_cpu(cpu) | |
1600 | get_cwq(cpu, wq)->max_active = 0; | |
1601 | ||
1537663f | 1602 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 1603 | |
1537663f TH |
1604 | spin_unlock(&workqueue_lock); |
1605 | ||
1606 | cpu_maps_update_done(); | |
1607 | ||
c34056a3 | 1608 | if (failed) { |
3af24433 ON |
1609 | destroy_workqueue(wq); |
1610 | wq = NULL; | |
1611 | } | |
1612 | return wq; | |
4690c4ab TH |
1613 | err: |
1614 | if (wq) { | |
0f900049 | 1615 | free_cwqs(wq->cpu_wq); |
4690c4ab TH |
1616 | kfree(wq); |
1617 | } | |
1618 | return NULL; | |
3af24433 | 1619 | } |
4e6045f1 | 1620 | EXPORT_SYMBOL_GPL(__create_workqueue_key); |
1da177e4 | 1621 | |
3af24433 ON |
1622 | /** |
1623 | * destroy_workqueue - safely terminate a workqueue | |
1624 | * @wq: target workqueue | |
1625 | * | |
1626 | * Safely destroy a workqueue. All work currently pending will be done first. | |
1627 | */ | |
1628 | void destroy_workqueue(struct workqueue_struct *wq) | |
1629 | { | |
b1f4ec17 | 1630 | int cpu; |
3af24433 | 1631 | |
a0a1a5fd TH |
1632 | flush_workqueue(wq); |
1633 | ||
1634 | /* | |
1635 | * wq list is used to freeze wq, remove from list after | |
1636 | * flushing is complete in case freeze races us. | |
1637 | */ | |
3da1c84c | 1638 | cpu_maps_update_begin(); |
95402b38 | 1639 | spin_lock(&workqueue_lock); |
b1f4ec17 | 1640 | list_del(&wq->list); |
95402b38 | 1641 | spin_unlock(&workqueue_lock); |
1537663f | 1642 | cpu_maps_update_done(); |
3af24433 | 1643 | |
73f53c4a TH |
1644 | for_each_possible_cpu(cpu) { |
1645 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
1646 | int i; | |
1647 | ||
c34056a3 TH |
1648 | if (cwq->worker) { |
1649 | destroy_worker(cwq->worker); | |
1650 | cwq->worker = NULL; | |
73f53c4a TH |
1651 | } |
1652 | ||
1653 | for (i = 0; i < WORK_NR_COLORS; i++) | |
1654 | BUG_ON(cwq->nr_in_flight[i]); | |
1e19ffc6 TH |
1655 | BUG_ON(cwq->nr_active); |
1656 | BUG_ON(!list_empty(&cwq->delayed_works)); | |
73f53c4a | 1657 | } |
9b41ea72 | 1658 | |
0f900049 | 1659 | free_cwqs(wq->cpu_wq); |
3af24433 ON |
1660 | kfree(wq); |
1661 | } | |
1662 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
1663 | ||
1664 | static int __devinit workqueue_cpu_callback(struct notifier_block *nfb, | |
1665 | unsigned long action, | |
1666 | void *hcpu) | |
1667 | { | |
1668 | unsigned int cpu = (unsigned long)hcpu; | |
1669 | struct cpu_workqueue_struct *cwq; | |
1670 | struct workqueue_struct *wq; | |
1671 | ||
8bb78442 RW |
1672 | action &= ~CPU_TASKS_FROZEN; |
1673 | ||
3af24433 | 1674 | list_for_each_entry(wq, &workqueues, list) { |
1537663f TH |
1675 | if (wq->flags & WQ_SINGLE_THREAD) |
1676 | continue; | |
3af24433 | 1677 | |
1537663f | 1678 | cwq = get_cwq(cpu, wq); |
3af24433 | 1679 | |
1537663f | 1680 | switch (action) { |
3da1c84c | 1681 | case CPU_POST_DEAD: |
73f53c4a | 1682 | flush_workqueue(wq); |
3af24433 ON |
1683 | break; |
1684 | } | |
1da177e4 LT |
1685 | } |
1686 | ||
1537663f | 1687 | return notifier_from_errno(0); |
1da177e4 | 1688 | } |
1da177e4 | 1689 | |
2d3854a3 | 1690 | #ifdef CONFIG_SMP |
8ccad40d | 1691 | |
2d3854a3 | 1692 | struct work_for_cpu { |
6b44003e | 1693 | struct completion completion; |
2d3854a3 RR |
1694 | long (*fn)(void *); |
1695 | void *arg; | |
1696 | long ret; | |
1697 | }; | |
1698 | ||
6b44003e | 1699 | static int do_work_for_cpu(void *_wfc) |
2d3854a3 | 1700 | { |
6b44003e | 1701 | struct work_for_cpu *wfc = _wfc; |
2d3854a3 | 1702 | wfc->ret = wfc->fn(wfc->arg); |
6b44003e AM |
1703 | complete(&wfc->completion); |
1704 | return 0; | |
2d3854a3 RR |
1705 | } |
1706 | ||
1707 | /** | |
1708 | * work_on_cpu - run a function in user context on a particular cpu | |
1709 | * @cpu: the cpu to run on | |
1710 | * @fn: the function to run | |
1711 | * @arg: the function arg | |
1712 | * | |
31ad9081 RR |
1713 | * This will return the value @fn returns. |
1714 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 1715 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 RR |
1716 | */ |
1717 | long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) | |
1718 | { | |
6b44003e AM |
1719 | struct task_struct *sub_thread; |
1720 | struct work_for_cpu wfc = { | |
1721 | .completion = COMPLETION_INITIALIZER_ONSTACK(wfc.completion), | |
1722 | .fn = fn, | |
1723 | .arg = arg, | |
1724 | }; | |
1725 | ||
1726 | sub_thread = kthread_create(do_work_for_cpu, &wfc, "work_for_cpu"); | |
1727 | if (IS_ERR(sub_thread)) | |
1728 | return PTR_ERR(sub_thread); | |
1729 | kthread_bind(sub_thread, cpu); | |
1730 | wake_up_process(sub_thread); | |
1731 | wait_for_completion(&wfc.completion); | |
2d3854a3 RR |
1732 | return wfc.ret; |
1733 | } | |
1734 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
1735 | #endif /* CONFIG_SMP */ | |
1736 | ||
a0a1a5fd TH |
1737 | #ifdef CONFIG_FREEZER |
1738 | ||
1739 | /** | |
1740 | * freeze_workqueues_begin - begin freezing workqueues | |
1741 | * | |
1742 | * Start freezing workqueues. After this function returns, all | |
1743 | * freezeable workqueues will queue new works to their frozen_works | |
1744 | * list instead of the cwq ones. | |
1745 | * | |
1746 | * CONTEXT: | |
1747 | * Grabs and releases workqueue_lock and cwq->lock's. | |
1748 | */ | |
1749 | void freeze_workqueues_begin(void) | |
1750 | { | |
1751 | struct workqueue_struct *wq; | |
1752 | unsigned int cpu; | |
1753 | ||
1754 | spin_lock(&workqueue_lock); | |
1755 | ||
1756 | BUG_ON(workqueue_freezing); | |
1757 | workqueue_freezing = true; | |
1758 | ||
1759 | for_each_possible_cpu(cpu) { | |
1760 | list_for_each_entry(wq, &workqueues, list) { | |
1761 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
1762 | ||
1763 | spin_lock_irq(&cwq->lock); | |
1764 | ||
1765 | if (wq->flags & WQ_FREEZEABLE) | |
1766 | cwq->max_active = 0; | |
1767 | ||
1768 | spin_unlock_irq(&cwq->lock); | |
1769 | } | |
1770 | } | |
1771 | ||
1772 | spin_unlock(&workqueue_lock); | |
1773 | } | |
1774 | ||
1775 | /** | |
1776 | * freeze_workqueues_busy - are freezeable workqueues still busy? | |
1777 | * | |
1778 | * Check whether freezing is complete. This function must be called | |
1779 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
1780 | * | |
1781 | * CONTEXT: | |
1782 | * Grabs and releases workqueue_lock. | |
1783 | * | |
1784 | * RETURNS: | |
1785 | * %true if some freezeable workqueues are still busy. %false if | |
1786 | * freezing is complete. | |
1787 | */ | |
1788 | bool freeze_workqueues_busy(void) | |
1789 | { | |
1790 | struct workqueue_struct *wq; | |
1791 | unsigned int cpu; | |
1792 | bool busy = false; | |
1793 | ||
1794 | spin_lock(&workqueue_lock); | |
1795 | ||
1796 | BUG_ON(!workqueue_freezing); | |
1797 | ||
1798 | for_each_possible_cpu(cpu) { | |
1799 | /* | |
1800 | * nr_active is monotonically decreasing. It's safe | |
1801 | * to peek without lock. | |
1802 | */ | |
1803 | list_for_each_entry(wq, &workqueues, list) { | |
1804 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
1805 | ||
1806 | if (!(wq->flags & WQ_FREEZEABLE)) | |
1807 | continue; | |
1808 | ||
1809 | BUG_ON(cwq->nr_active < 0); | |
1810 | if (cwq->nr_active) { | |
1811 | busy = true; | |
1812 | goto out_unlock; | |
1813 | } | |
1814 | } | |
1815 | } | |
1816 | out_unlock: | |
1817 | spin_unlock(&workqueue_lock); | |
1818 | return busy; | |
1819 | } | |
1820 | ||
1821 | /** | |
1822 | * thaw_workqueues - thaw workqueues | |
1823 | * | |
1824 | * Thaw workqueues. Normal queueing is restored and all collected | |
1825 | * frozen works are transferred to their respective cwq worklists. | |
1826 | * | |
1827 | * CONTEXT: | |
1828 | * Grabs and releases workqueue_lock and cwq->lock's. | |
1829 | */ | |
1830 | void thaw_workqueues(void) | |
1831 | { | |
1832 | struct workqueue_struct *wq; | |
1833 | unsigned int cpu; | |
1834 | ||
1835 | spin_lock(&workqueue_lock); | |
1836 | ||
1837 | if (!workqueue_freezing) | |
1838 | goto out_unlock; | |
1839 | ||
1840 | for_each_possible_cpu(cpu) { | |
1841 | list_for_each_entry(wq, &workqueues, list) { | |
1842 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
1843 | ||
1844 | if (!(wq->flags & WQ_FREEZEABLE)) | |
1845 | continue; | |
1846 | ||
1847 | spin_lock_irq(&cwq->lock); | |
1848 | ||
1849 | /* restore max_active and repopulate worklist */ | |
1850 | cwq->max_active = wq->saved_max_active; | |
1851 | ||
1852 | while (!list_empty(&cwq->delayed_works) && | |
1853 | cwq->nr_active < cwq->max_active) | |
1854 | cwq_activate_first_delayed(cwq); | |
1855 | ||
1856 | wake_up(&cwq->more_work); | |
1857 | ||
1858 | spin_unlock_irq(&cwq->lock); | |
1859 | } | |
1860 | } | |
1861 | ||
1862 | workqueue_freezing = false; | |
1863 | out_unlock: | |
1864 | spin_unlock(&workqueue_lock); | |
1865 | } | |
1866 | #endif /* CONFIG_FREEZER */ | |
1867 | ||
c12920d1 | 1868 | void __init init_workqueues(void) |
1da177e4 | 1869 | { |
c34056a3 TH |
1870 | unsigned int cpu; |
1871 | ||
1872 | for_each_possible_cpu(cpu) | |
1873 | ida_init(&per_cpu(worker_ida, cpu)); | |
1874 | ||
e7577c50 | 1875 | singlethread_cpu = cpumask_first(cpu_possible_mask); |
1da177e4 LT |
1876 | hotcpu_notifier(workqueue_cpu_callback, 0); |
1877 | keventd_wq = create_workqueue("events"); | |
1878 | BUG_ON(!keventd_wq); | |
1879 | } |