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 | * | |
8b03ae3c | 43 | * L: gcwq->lock protected. Access with gcwq->lock held. |
4690c4ab | 44 | * |
73f53c4a TH |
45 | * F: wq->flush_mutex protected. |
46 | * | |
4690c4ab TH |
47 | * W: workqueue_lock protected. |
48 | */ | |
49 | ||
8b03ae3c | 50 | struct global_cwq; |
c34056a3 TH |
51 | struct cpu_workqueue_struct; |
52 | ||
53 | struct worker { | |
54 | struct work_struct *current_work; /* L: work being processed */ | |
affee4b2 | 55 | struct list_head scheduled; /* L: scheduled works */ |
c34056a3 | 56 | struct task_struct *task; /* I: worker task */ |
8b03ae3c | 57 | struct global_cwq *gcwq; /* I: the associated gcwq */ |
c34056a3 TH |
58 | struct cpu_workqueue_struct *cwq; /* I: the associated cwq */ |
59 | int id; /* I: worker id */ | |
60 | }; | |
61 | ||
8b03ae3c TH |
62 | /* |
63 | * Global per-cpu workqueue. | |
64 | */ | |
65 | struct global_cwq { | |
66 | spinlock_t lock; /* the gcwq lock */ | |
67 | unsigned int cpu; /* I: the associated cpu */ | |
68 | struct ida worker_ida; /* L: for worker IDs */ | |
69 | } ____cacheline_aligned_in_smp; | |
70 | ||
1da177e4 | 71 | /* |
f756d5e2 | 72 | * The per-CPU workqueue (if single thread, we always use the first |
0f900049 TH |
73 | * possible cpu). The lower WORK_STRUCT_FLAG_BITS of |
74 | * work_struct->data are used for flags and thus cwqs need to be | |
75 | * aligned at two's power of the number of flag bits. | |
1da177e4 LT |
76 | */ |
77 | struct cpu_workqueue_struct { | |
8b03ae3c | 78 | struct global_cwq *gcwq; /* I: the associated gcwq */ |
1da177e4 LT |
79 | struct list_head worklist; |
80 | wait_queue_head_t more_work; | |
c34056a3 | 81 | struct worker *worker; |
4690c4ab | 82 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
83 | int work_color; /* L: current color */ |
84 | int flush_color; /* L: flushing color */ | |
85 | int nr_in_flight[WORK_NR_COLORS]; | |
86 | /* L: nr of in_flight works */ | |
1e19ffc6 | 87 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 88 | int max_active; /* L: max active works */ |
1e19ffc6 | 89 | struct list_head delayed_works; /* L: delayed works */ |
0f900049 | 90 | }; |
1da177e4 | 91 | |
73f53c4a TH |
92 | /* |
93 | * Structure used to wait for workqueue flush. | |
94 | */ | |
95 | struct wq_flusher { | |
96 | struct list_head list; /* F: list of flushers */ | |
97 | int flush_color; /* F: flush color waiting for */ | |
98 | struct completion done; /* flush completion */ | |
99 | }; | |
100 | ||
1da177e4 LT |
101 | /* |
102 | * The externally visible workqueue abstraction is an array of | |
103 | * per-CPU workqueues: | |
104 | */ | |
105 | struct workqueue_struct { | |
97e37d7b | 106 | unsigned int flags; /* I: WQ_* flags */ |
4690c4ab TH |
107 | struct cpu_workqueue_struct *cpu_wq; /* I: cwq's */ |
108 | struct list_head list; /* W: list of all workqueues */ | |
73f53c4a TH |
109 | |
110 | struct mutex flush_mutex; /* protects wq flushing */ | |
111 | int work_color; /* F: current work color */ | |
112 | int flush_color; /* F: current flush color */ | |
113 | atomic_t nr_cwqs_to_flush; /* flush in progress */ | |
114 | struct wq_flusher *first_flusher; /* F: first flusher */ | |
115 | struct list_head flusher_queue; /* F: flush waiters */ | |
116 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
117 | ||
a0a1a5fd | 118 | int saved_max_active; /* I: saved cwq max_active */ |
4690c4ab | 119 | const char *name; /* I: workqueue name */ |
4e6045f1 | 120 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 121 | struct lockdep_map lockdep_map; |
4e6045f1 | 122 | #endif |
1da177e4 LT |
123 | }; |
124 | ||
dc186ad7 TG |
125 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
126 | ||
127 | static struct debug_obj_descr work_debug_descr; | |
128 | ||
129 | /* | |
130 | * fixup_init is called when: | |
131 | * - an active object is initialized | |
132 | */ | |
133 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
134 | { | |
135 | struct work_struct *work = addr; | |
136 | ||
137 | switch (state) { | |
138 | case ODEBUG_STATE_ACTIVE: | |
139 | cancel_work_sync(work); | |
140 | debug_object_init(work, &work_debug_descr); | |
141 | return 1; | |
142 | default: | |
143 | return 0; | |
144 | } | |
145 | } | |
146 | ||
147 | /* | |
148 | * fixup_activate is called when: | |
149 | * - an active object is activated | |
150 | * - an unknown object is activated (might be a statically initialized object) | |
151 | */ | |
152 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
153 | { | |
154 | struct work_struct *work = addr; | |
155 | ||
156 | switch (state) { | |
157 | ||
158 | case ODEBUG_STATE_NOTAVAILABLE: | |
159 | /* | |
160 | * This is not really a fixup. The work struct was | |
161 | * statically initialized. We just make sure that it | |
162 | * is tracked in the object tracker. | |
163 | */ | |
22df02bb | 164 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
165 | debug_object_init(work, &work_debug_descr); |
166 | debug_object_activate(work, &work_debug_descr); | |
167 | return 0; | |
168 | } | |
169 | WARN_ON_ONCE(1); | |
170 | return 0; | |
171 | ||
172 | case ODEBUG_STATE_ACTIVE: | |
173 | WARN_ON(1); | |
174 | ||
175 | default: | |
176 | return 0; | |
177 | } | |
178 | } | |
179 | ||
180 | /* | |
181 | * fixup_free is called when: | |
182 | * - an active object is freed | |
183 | */ | |
184 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
185 | { | |
186 | struct work_struct *work = addr; | |
187 | ||
188 | switch (state) { | |
189 | case ODEBUG_STATE_ACTIVE: | |
190 | cancel_work_sync(work); | |
191 | debug_object_free(work, &work_debug_descr); | |
192 | return 1; | |
193 | default: | |
194 | return 0; | |
195 | } | |
196 | } | |
197 | ||
198 | static struct debug_obj_descr work_debug_descr = { | |
199 | .name = "work_struct", | |
200 | .fixup_init = work_fixup_init, | |
201 | .fixup_activate = work_fixup_activate, | |
202 | .fixup_free = work_fixup_free, | |
203 | }; | |
204 | ||
205 | static inline void debug_work_activate(struct work_struct *work) | |
206 | { | |
207 | debug_object_activate(work, &work_debug_descr); | |
208 | } | |
209 | ||
210 | static inline void debug_work_deactivate(struct work_struct *work) | |
211 | { | |
212 | debug_object_deactivate(work, &work_debug_descr); | |
213 | } | |
214 | ||
215 | void __init_work(struct work_struct *work, int onstack) | |
216 | { | |
217 | if (onstack) | |
218 | debug_object_init_on_stack(work, &work_debug_descr); | |
219 | else | |
220 | debug_object_init(work, &work_debug_descr); | |
221 | } | |
222 | EXPORT_SYMBOL_GPL(__init_work); | |
223 | ||
224 | void destroy_work_on_stack(struct work_struct *work) | |
225 | { | |
226 | debug_object_free(work, &work_debug_descr); | |
227 | } | |
228 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
229 | ||
230 | #else | |
231 | static inline void debug_work_activate(struct work_struct *work) { } | |
232 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
233 | #endif | |
234 | ||
95402b38 GS |
235 | /* Serializes the accesses to the list of workqueues. */ |
236 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 237 | static LIST_HEAD(workqueues); |
a0a1a5fd | 238 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 239 | |
8b03ae3c TH |
240 | static DEFINE_PER_CPU(struct global_cwq, global_cwq); |
241 | ||
c34056a3 | 242 | static int worker_thread(void *__worker); |
1da177e4 | 243 | |
3af24433 | 244 | static int singlethread_cpu __read_mostly; |
1da177e4 | 245 | |
8b03ae3c TH |
246 | static struct global_cwq *get_gcwq(unsigned int cpu) |
247 | { | |
248 | return &per_cpu(global_cwq, cpu); | |
249 | } | |
250 | ||
1537663f TH |
251 | static struct cpu_workqueue_struct *get_cwq(unsigned int cpu, |
252 | struct workqueue_struct *wq) | |
b1f4ec17 | 253 | { |
1537663f | 254 | return per_cpu_ptr(wq->cpu_wq, cpu); |
b1f4ec17 ON |
255 | } |
256 | ||
1537663f TH |
257 | static struct cpu_workqueue_struct *target_cwq(unsigned int cpu, |
258 | struct workqueue_struct *wq) | |
a848e3b6 | 259 | { |
1537663f | 260 | if (unlikely(wq->flags & WQ_SINGLE_THREAD)) |
a848e3b6 | 261 | cpu = singlethread_cpu; |
1537663f | 262 | return get_cwq(cpu, wq); |
a848e3b6 ON |
263 | } |
264 | ||
73f53c4a TH |
265 | static unsigned int work_color_to_flags(int color) |
266 | { | |
267 | return color << WORK_STRUCT_COLOR_SHIFT; | |
268 | } | |
269 | ||
270 | static int get_work_color(struct work_struct *work) | |
271 | { | |
272 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
273 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
274 | } | |
275 | ||
276 | static int work_next_color(int color) | |
277 | { | |
278 | return (color + 1) % WORK_NR_COLORS; | |
279 | } | |
280 | ||
4594bf15 DH |
281 | /* |
282 | * Set the workqueue on which a work item is to be run | |
283 | * - Must *only* be called if the pending flag is set | |
284 | */ | |
ed7c0fee | 285 | static inline void set_wq_data(struct work_struct *work, |
4690c4ab TH |
286 | struct cpu_workqueue_struct *cwq, |
287 | unsigned long extra_flags) | |
365970a1 | 288 | { |
4594bf15 | 289 | BUG_ON(!work_pending(work)); |
365970a1 | 290 | |
4690c4ab | 291 | atomic_long_set(&work->data, (unsigned long)cwq | work_static(work) | |
22df02bb | 292 | WORK_STRUCT_PENDING | extra_flags); |
365970a1 DH |
293 | } |
294 | ||
4d707b9f ON |
295 | /* |
296 | * Clear WORK_STRUCT_PENDING and the workqueue on which it was queued. | |
297 | */ | |
298 | static inline void clear_wq_data(struct work_struct *work) | |
299 | { | |
4690c4ab | 300 | atomic_long_set(&work->data, work_static(work)); |
4d707b9f ON |
301 | } |
302 | ||
64166699 | 303 | static inline struct cpu_workqueue_struct *get_wq_data(struct work_struct *work) |
365970a1 | 304 | { |
64166699 TH |
305 | return (void *)(atomic_long_read(&work->data) & |
306 | WORK_STRUCT_WQ_DATA_MASK); | |
365970a1 DH |
307 | } |
308 | ||
4690c4ab TH |
309 | /** |
310 | * insert_work - insert a work into cwq | |
311 | * @cwq: cwq @work belongs to | |
312 | * @work: work to insert | |
313 | * @head: insertion point | |
314 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
315 | * | |
316 | * Insert @work into @cwq after @head. | |
317 | * | |
318 | * CONTEXT: | |
8b03ae3c | 319 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 320 | */ |
b89deed3 | 321 | static void insert_work(struct cpu_workqueue_struct *cwq, |
4690c4ab TH |
322 | struct work_struct *work, struct list_head *head, |
323 | unsigned int extra_flags) | |
b89deed3 | 324 | { |
4690c4ab TH |
325 | /* we own @work, set data and link */ |
326 | set_wq_data(work, cwq, extra_flags); | |
327 | ||
6e84d644 ON |
328 | /* |
329 | * Ensure that we get the right work->data if we see the | |
330 | * result of list_add() below, see try_to_grab_pending(). | |
331 | */ | |
332 | smp_wmb(); | |
4690c4ab | 333 | |
1a4d9b0a | 334 | list_add_tail(&work->entry, head); |
b89deed3 ON |
335 | wake_up(&cwq->more_work); |
336 | } | |
337 | ||
4690c4ab | 338 | static void __queue_work(unsigned int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
339 | struct work_struct *work) |
340 | { | |
1537663f | 341 | struct cpu_workqueue_struct *cwq = target_cwq(cpu, wq); |
8b03ae3c | 342 | struct global_cwq *gcwq = cwq->gcwq; |
1e19ffc6 | 343 | struct list_head *worklist; |
1da177e4 LT |
344 | unsigned long flags; |
345 | ||
dc186ad7 | 346 | debug_work_activate(work); |
1e19ffc6 | 347 | |
8b03ae3c | 348 | spin_lock_irqsave(&gcwq->lock, flags); |
4690c4ab | 349 | BUG_ON(!list_empty(&work->entry)); |
1e19ffc6 | 350 | |
73f53c4a | 351 | cwq->nr_in_flight[cwq->work_color]++; |
1e19ffc6 TH |
352 | |
353 | if (likely(cwq->nr_active < cwq->max_active)) { | |
354 | cwq->nr_active++; | |
355 | worklist = &cwq->worklist; | |
356 | } else | |
357 | worklist = &cwq->delayed_works; | |
358 | ||
359 | insert_work(cwq, work, worklist, work_color_to_flags(cwq->work_color)); | |
360 | ||
8b03ae3c | 361 | spin_unlock_irqrestore(&gcwq->lock, flags); |
1da177e4 LT |
362 | } |
363 | ||
0fcb78c2 REB |
364 | /** |
365 | * queue_work - queue work on a workqueue | |
366 | * @wq: workqueue to use | |
367 | * @work: work to queue | |
368 | * | |
057647fc | 369 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
1da177e4 | 370 | * |
00dfcaf7 ON |
371 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
372 | * it can be processed by another CPU. | |
1da177e4 | 373 | */ |
7ad5b3a5 | 374 | int queue_work(struct workqueue_struct *wq, struct work_struct *work) |
1da177e4 | 375 | { |
ef1ca236 ON |
376 | int ret; |
377 | ||
378 | ret = queue_work_on(get_cpu(), wq, work); | |
379 | put_cpu(); | |
380 | ||
1da177e4 LT |
381 | return ret; |
382 | } | |
ae90dd5d | 383 | EXPORT_SYMBOL_GPL(queue_work); |
1da177e4 | 384 | |
c1a220e7 ZR |
385 | /** |
386 | * queue_work_on - queue work on specific cpu | |
387 | * @cpu: CPU number to execute work on | |
388 | * @wq: workqueue to use | |
389 | * @work: work to queue | |
390 | * | |
391 | * Returns 0 if @work was already on a queue, non-zero otherwise. | |
392 | * | |
393 | * We queue the work to a specific CPU, the caller must ensure it | |
394 | * can't go away. | |
395 | */ | |
396 | int | |
397 | queue_work_on(int cpu, struct workqueue_struct *wq, struct work_struct *work) | |
398 | { | |
399 | int ret = 0; | |
400 | ||
22df02bb | 401 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 402 | __queue_work(cpu, wq, work); |
c1a220e7 ZR |
403 | ret = 1; |
404 | } | |
405 | return ret; | |
406 | } | |
407 | EXPORT_SYMBOL_GPL(queue_work_on); | |
408 | ||
6d141c3f | 409 | static void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 410 | { |
52bad64d | 411 | struct delayed_work *dwork = (struct delayed_work *)__data; |
ed7c0fee | 412 | struct cpu_workqueue_struct *cwq = get_wq_data(&dwork->work); |
1da177e4 | 413 | |
4690c4ab | 414 | __queue_work(smp_processor_id(), cwq->wq, &dwork->work); |
1da177e4 LT |
415 | } |
416 | ||
0fcb78c2 REB |
417 | /** |
418 | * queue_delayed_work - queue work on a workqueue after delay | |
419 | * @wq: workqueue to use | |
af9997e4 | 420 | * @dwork: delayable work to queue |
0fcb78c2 REB |
421 | * @delay: number of jiffies to wait before queueing |
422 | * | |
057647fc | 423 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 424 | */ |
7ad5b3a5 | 425 | int queue_delayed_work(struct workqueue_struct *wq, |
52bad64d | 426 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 427 | { |
52bad64d | 428 | if (delay == 0) |
63bc0362 | 429 | return queue_work(wq, &dwork->work); |
1da177e4 | 430 | |
63bc0362 | 431 | return queue_delayed_work_on(-1, wq, dwork, delay); |
1da177e4 | 432 | } |
ae90dd5d | 433 | EXPORT_SYMBOL_GPL(queue_delayed_work); |
1da177e4 | 434 | |
0fcb78c2 REB |
435 | /** |
436 | * queue_delayed_work_on - queue work on specific CPU after delay | |
437 | * @cpu: CPU number to execute work on | |
438 | * @wq: workqueue to use | |
af9997e4 | 439 | * @dwork: work to queue |
0fcb78c2 REB |
440 | * @delay: number of jiffies to wait before queueing |
441 | * | |
057647fc | 442 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 443 | */ |
7a6bc1cd | 444 | int queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
52bad64d | 445 | struct delayed_work *dwork, unsigned long delay) |
7a6bc1cd VP |
446 | { |
447 | int ret = 0; | |
52bad64d DH |
448 | struct timer_list *timer = &dwork->timer; |
449 | struct work_struct *work = &dwork->work; | |
7a6bc1cd | 450 | |
22df02bb | 451 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7a6bc1cd VP |
452 | BUG_ON(timer_pending(timer)); |
453 | BUG_ON(!list_empty(&work->entry)); | |
454 | ||
8a3e77cc AL |
455 | timer_stats_timer_set_start_info(&dwork->timer); |
456 | ||
ed7c0fee | 457 | /* This stores cwq for the moment, for the timer_fn */ |
1537663f | 458 | set_wq_data(work, target_cwq(raw_smp_processor_id(), wq), 0); |
7a6bc1cd | 459 | timer->expires = jiffies + delay; |
52bad64d | 460 | timer->data = (unsigned long)dwork; |
7a6bc1cd | 461 | timer->function = delayed_work_timer_fn; |
63bc0362 ON |
462 | |
463 | if (unlikely(cpu >= 0)) | |
464 | add_timer_on(timer, cpu); | |
465 | else | |
466 | add_timer(timer); | |
7a6bc1cd VP |
467 | ret = 1; |
468 | } | |
469 | return ret; | |
470 | } | |
ae90dd5d | 471 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
1da177e4 | 472 | |
c34056a3 TH |
473 | static struct worker *alloc_worker(void) |
474 | { | |
475 | struct worker *worker; | |
476 | ||
477 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
affee4b2 TH |
478 | if (worker) |
479 | INIT_LIST_HEAD(&worker->scheduled); | |
c34056a3 TH |
480 | return worker; |
481 | } | |
482 | ||
483 | /** | |
484 | * create_worker - create a new workqueue worker | |
485 | * @cwq: cwq the new worker will belong to | |
486 | * @bind: whether to set affinity to @cpu or not | |
487 | * | |
488 | * Create a new worker which is bound to @cwq. The returned worker | |
489 | * can be started by calling start_worker() or destroyed using | |
490 | * destroy_worker(). | |
491 | * | |
492 | * CONTEXT: | |
493 | * Might sleep. Does GFP_KERNEL allocations. | |
494 | * | |
495 | * RETURNS: | |
496 | * Pointer to the newly created worker. | |
497 | */ | |
498 | static struct worker *create_worker(struct cpu_workqueue_struct *cwq, bool bind) | |
499 | { | |
8b03ae3c | 500 | struct global_cwq *gcwq = cwq->gcwq; |
c34056a3 TH |
501 | int id = -1; |
502 | struct worker *worker = NULL; | |
503 | ||
8b03ae3c TH |
504 | spin_lock_irq(&gcwq->lock); |
505 | while (ida_get_new(&gcwq->worker_ida, &id)) { | |
506 | spin_unlock_irq(&gcwq->lock); | |
507 | if (!ida_pre_get(&gcwq->worker_ida, GFP_KERNEL)) | |
c34056a3 | 508 | goto fail; |
8b03ae3c | 509 | spin_lock_irq(&gcwq->lock); |
c34056a3 | 510 | } |
8b03ae3c | 511 | spin_unlock_irq(&gcwq->lock); |
c34056a3 TH |
512 | |
513 | worker = alloc_worker(); | |
514 | if (!worker) | |
515 | goto fail; | |
516 | ||
8b03ae3c | 517 | worker->gcwq = gcwq; |
c34056a3 TH |
518 | worker->cwq = cwq; |
519 | worker->id = id; | |
520 | ||
521 | worker->task = kthread_create(worker_thread, worker, "kworker/%u:%d", | |
8b03ae3c | 522 | gcwq->cpu, id); |
c34056a3 TH |
523 | if (IS_ERR(worker->task)) |
524 | goto fail; | |
525 | ||
526 | if (bind) | |
8b03ae3c | 527 | kthread_bind(worker->task, gcwq->cpu); |
c34056a3 TH |
528 | |
529 | return worker; | |
530 | fail: | |
531 | if (id >= 0) { | |
8b03ae3c TH |
532 | spin_lock_irq(&gcwq->lock); |
533 | ida_remove(&gcwq->worker_ida, id); | |
534 | spin_unlock_irq(&gcwq->lock); | |
c34056a3 TH |
535 | } |
536 | kfree(worker); | |
537 | return NULL; | |
538 | } | |
539 | ||
540 | /** | |
541 | * start_worker - start a newly created worker | |
542 | * @worker: worker to start | |
543 | * | |
544 | * Start @worker. | |
545 | * | |
546 | * CONTEXT: | |
8b03ae3c | 547 | * spin_lock_irq(gcwq->lock). |
c34056a3 TH |
548 | */ |
549 | static void start_worker(struct worker *worker) | |
550 | { | |
551 | wake_up_process(worker->task); | |
552 | } | |
553 | ||
554 | /** | |
555 | * destroy_worker - destroy a workqueue worker | |
556 | * @worker: worker to be destroyed | |
557 | * | |
558 | * Destroy @worker. | |
559 | */ | |
560 | static void destroy_worker(struct worker *worker) | |
561 | { | |
8b03ae3c | 562 | struct global_cwq *gcwq = worker->gcwq; |
c34056a3 TH |
563 | int id = worker->id; |
564 | ||
565 | /* sanity check frenzy */ | |
566 | BUG_ON(worker->current_work); | |
affee4b2 | 567 | BUG_ON(!list_empty(&worker->scheduled)); |
c34056a3 TH |
568 | |
569 | kthread_stop(worker->task); | |
570 | kfree(worker); | |
571 | ||
8b03ae3c TH |
572 | spin_lock_irq(&gcwq->lock); |
573 | ida_remove(&gcwq->worker_ida, id); | |
574 | spin_unlock_irq(&gcwq->lock); | |
c34056a3 TH |
575 | } |
576 | ||
affee4b2 TH |
577 | /** |
578 | * move_linked_works - move linked works to a list | |
579 | * @work: start of series of works to be scheduled | |
580 | * @head: target list to append @work to | |
581 | * @nextp: out paramter for nested worklist walking | |
582 | * | |
583 | * Schedule linked works starting from @work to @head. Work series to | |
584 | * be scheduled starts at @work and includes any consecutive work with | |
585 | * WORK_STRUCT_LINKED set in its predecessor. | |
586 | * | |
587 | * If @nextp is not NULL, it's updated to point to the next work of | |
588 | * the last scheduled work. This allows move_linked_works() to be | |
589 | * nested inside outer list_for_each_entry_safe(). | |
590 | * | |
591 | * CONTEXT: | |
8b03ae3c | 592 | * spin_lock_irq(gcwq->lock). |
affee4b2 TH |
593 | */ |
594 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
595 | struct work_struct **nextp) | |
596 | { | |
597 | struct work_struct *n; | |
598 | ||
599 | /* | |
600 | * Linked worklist will always end before the end of the list, | |
601 | * use NULL for list head. | |
602 | */ | |
603 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
604 | list_move_tail(&work->entry, head); | |
605 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
606 | break; | |
607 | } | |
608 | ||
609 | /* | |
610 | * If we're already inside safe list traversal and have moved | |
611 | * multiple works to the scheduled queue, the next position | |
612 | * needs to be updated. | |
613 | */ | |
614 | if (nextp) | |
615 | *nextp = n; | |
616 | } | |
617 | ||
1e19ffc6 TH |
618 | static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq) |
619 | { | |
620 | struct work_struct *work = list_first_entry(&cwq->delayed_works, | |
621 | struct work_struct, entry); | |
622 | ||
623 | move_linked_works(work, &cwq->worklist, NULL); | |
624 | cwq->nr_active++; | |
625 | } | |
626 | ||
73f53c4a TH |
627 | /** |
628 | * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight | |
629 | * @cwq: cwq of interest | |
630 | * @color: color of work which left the queue | |
631 | * | |
632 | * A work either has completed or is removed from pending queue, | |
633 | * decrement nr_in_flight of its cwq and handle workqueue flushing. | |
634 | * | |
635 | * CONTEXT: | |
8b03ae3c | 636 | * spin_lock_irq(gcwq->lock). |
73f53c4a TH |
637 | */ |
638 | static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color) | |
639 | { | |
640 | /* ignore uncolored works */ | |
641 | if (color == WORK_NO_COLOR) | |
642 | return; | |
643 | ||
644 | cwq->nr_in_flight[color]--; | |
1e19ffc6 TH |
645 | cwq->nr_active--; |
646 | ||
647 | /* one down, submit a delayed one */ | |
648 | if (!list_empty(&cwq->delayed_works) && | |
649 | cwq->nr_active < cwq->max_active) | |
650 | cwq_activate_first_delayed(cwq); | |
73f53c4a TH |
651 | |
652 | /* is flush in progress and are we at the flushing tip? */ | |
653 | if (likely(cwq->flush_color != color)) | |
654 | return; | |
655 | ||
656 | /* are there still in-flight works? */ | |
657 | if (cwq->nr_in_flight[color]) | |
658 | return; | |
659 | ||
660 | /* this cwq is done, clear flush_color */ | |
661 | cwq->flush_color = -1; | |
662 | ||
663 | /* | |
664 | * If this was the last cwq, wake up the first flusher. It | |
665 | * will handle the rest. | |
666 | */ | |
667 | if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush)) | |
668 | complete(&cwq->wq->first_flusher->done); | |
669 | } | |
670 | ||
a62428c0 TH |
671 | /** |
672 | * process_one_work - process single work | |
c34056a3 | 673 | * @worker: self |
a62428c0 TH |
674 | * @work: work to process |
675 | * | |
676 | * Process @work. This function contains all the logics necessary to | |
677 | * process a single work including synchronization against and | |
678 | * interaction with other workers on the same cpu, queueing and | |
679 | * flushing. As long as context requirement is met, any worker can | |
680 | * call this function to process a work. | |
681 | * | |
682 | * CONTEXT: | |
8b03ae3c | 683 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. |
a62428c0 | 684 | */ |
c34056a3 | 685 | static void process_one_work(struct worker *worker, struct work_struct *work) |
a62428c0 | 686 | { |
c34056a3 | 687 | struct cpu_workqueue_struct *cwq = worker->cwq; |
8b03ae3c | 688 | struct global_cwq *gcwq = cwq->gcwq; |
a62428c0 | 689 | work_func_t f = work->func; |
73f53c4a | 690 | int work_color; |
a62428c0 TH |
691 | #ifdef CONFIG_LOCKDEP |
692 | /* | |
693 | * It is permissible to free the struct work_struct from | |
694 | * inside the function that is called from it, this we need to | |
695 | * take into account for lockdep too. To avoid bogus "held | |
696 | * lock freed" warnings as well as problems when looking into | |
697 | * work->lockdep_map, make a copy and use that here. | |
698 | */ | |
699 | struct lockdep_map lockdep_map = work->lockdep_map; | |
700 | #endif | |
701 | /* claim and process */ | |
a62428c0 | 702 | debug_work_deactivate(work); |
c34056a3 | 703 | worker->current_work = work; |
73f53c4a | 704 | work_color = get_work_color(work); |
a62428c0 TH |
705 | list_del_init(&work->entry); |
706 | ||
8b03ae3c | 707 | spin_unlock_irq(&gcwq->lock); |
a62428c0 TH |
708 | |
709 | BUG_ON(get_wq_data(work) != cwq); | |
710 | work_clear_pending(work); | |
711 | lock_map_acquire(&cwq->wq->lockdep_map); | |
712 | lock_map_acquire(&lockdep_map); | |
713 | f(work); | |
714 | lock_map_release(&lockdep_map); | |
715 | lock_map_release(&cwq->wq->lockdep_map); | |
716 | ||
717 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
718 | printk(KERN_ERR "BUG: workqueue leaked lock or atomic: " | |
719 | "%s/0x%08x/%d\n", | |
720 | current->comm, preempt_count(), task_pid_nr(current)); | |
721 | printk(KERN_ERR " last function: "); | |
722 | print_symbol("%s\n", (unsigned long)f); | |
723 | debug_show_held_locks(current); | |
724 | dump_stack(); | |
725 | } | |
726 | ||
8b03ae3c | 727 | spin_lock_irq(&gcwq->lock); |
a62428c0 TH |
728 | |
729 | /* we're done with it, release */ | |
c34056a3 | 730 | worker->current_work = NULL; |
73f53c4a | 731 | cwq_dec_nr_in_flight(cwq, work_color); |
a62428c0 TH |
732 | } |
733 | ||
affee4b2 TH |
734 | /** |
735 | * process_scheduled_works - process scheduled works | |
736 | * @worker: self | |
737 | * | |
738 | * Process all scheduled works. Please note that the scheduled list | |
739 | * may change while processing a work, so this function repeatedly | |
740 | * fetches a work from the top and executes it. | |
741 | * | |
742 | * CONTEXT: | |
8b03ae3c | 743 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed |
affee4b2 TH |
744 | * multiple times. |
745 | */ | |
746 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 747 | { |
affee4b2 TH |
748 | while (!list_empty(&worker->scheduled)) { |
749 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 750 | struct work_struct, entry); |
c34056a3 | 751 | process_one_work(worker, work); |
1da177e4 | 752 | } |
1da177e4 LT |
753 | } |
754 | ||
4690c4ab TH |
755 | /** |
756 | * worker_thread - the worker thread function | |
c34056a3 | 757 | * @__worker: self |
4690c4ab TH |
758 | * |
759 | * The cwq worker thread function. | |
760 | */ | |
c34056a3 | 761 | static int worker_thread(void *__worker) |
1da177e4 | 762 | { |
c34056a3 | 763 | struct worker *worker = __worker; |
8b03ae3c | 764 | struct global_cwq *gcwq = worker->gcwq; |
c34056a3 | 765 | struct cpu_workqueue_struct *cwq = worker->cwq; |
3af24433 | 766 | DEFINE_WAIT(wait); |
1da177e4 | 767 | |
3af24433 | 768 | for (;;) { |
3af24433 | 769 | prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE); |
a0a1a5fd | 770 | if (!kthread_should_stop() && |
14441960 | 771 | list_empty(&cwq->worklist)) |
1da177e4 | 772 | schedule(); |
3af24433 ON |
773 | finish_wait(&cwq->more_work, &wait); |
774 | ||
14441960 | 775 | if (kthread_should_stop()) |
3af24433 | 776 | break; |
1da177e4 | 777 | |
c34056a3 | 778 | if (unlikely(!cpumask_equal(&worker->task->cpus_allowed, |
8b03ae3c | 779 | get_cpu_mask(gcwq->cpu)))) |
c34056a3 | 780 | set_cpus_allowed_ptr(worker->task, |
8b03ae3c | 781 | get_cpu_mask(gcwq->cpu)); |
affee4b2 | 782 | |
8b03ae3c | 783 | spin_lock_irq(&gcwq->lock); |
affee4b2 TH |
784 | |
785 | while (!list_empty(&cwq->worklist)) { | |
786 | struct work_struct *work = | |
787 | list_first_entry(&cwq->worklist, | |
788 | struct work_struct, entry); | |
789 | ||
790 | if (likely(!(*work_data_bits(work) & | |
791 | WORK_STRUCT_LINKED))) { | |
792 | /* optimization path, not strictly necessary */ | |
793 | process_one_work(worker, work); | |
794 | if (unlikely(!list_empty(&worker->scheduled))) | |
795 | process_scheduled_works(worker); | |
796 | } else { | |
797 | move_linked_works(work, &worker->scheduled, | |
798 | NULL); | |
799 | process_scheduled_works(worker); | |
800 | } | |
801 | } | |
802 | ||
8b03ae3c | 803 | spin_unlock_irq(&gcwq->lock); |
1da177e4 | 804 | } |
3af24433 | 805 | |
1da177e4 LT |
806 | return 0; |
807 | } | |
808 | ||
fc2e4d70 ON |
809 | struct wq_barrier { |
810 | struct work_struct work; | |
811 | struct completion done; | |
812 | }; | |
813 | ||
814 | static void wq_barrier_func(struct work_struct *work) | |
815 | { | |
816 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
817 | complete(&barr->done); | |
818 | } | |
819 | ||
4690c4ab TH |
820 | /** |
821 | * insert_wq_barrier - insert a barrier work | |
822 | * @cwq: cwq to insert barrier into | |
823 | * @barr: wq_barrier to insert | |
affee4b2 TH |
824 | * @target: target work to attach @barr to |
825 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 826 | * |
affee4b2 TH |
827 | * @barr is linked to @target such that @barr is completed only after |
828 | * @target finishes execution. Please note that the ordering | |
829 | * guarantee is observed only with respect to @target and on the local | |
830 | * cpu. | |
831 | * | |
832 | * Currently, a queued barrier can't be canceled. This is because | |
833 | * try_to_grab_pending() can't determine whether the work to be | |
834 | * grabbed is at the head of the queue and thus can't clear LINKED | |
835 | * flag of the previous work while there must be a valid next work | |
836 | * after a work with LINKED flag set. | |
837 | * | |
838 | * Note that when @worker is non-NULL, @target may be modified | |
839 | * underneath us, so we can't reliably determine cwq from @target. | |
4690c4ab TH |
840 | * |
841 | * CONTEXT: | |
8b03ae3c | 842 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 843 | */ |
83c22520 | 844 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, |
affee4b2 TH |
845 | struct wq_barrier *barr, |
846 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 847 | { |
affee4b2 TH |
848 | struct list_head *head; |
849 | unsigned int linked = 0; | |
850 | ||
dc186ad7 | 851 | /* |
8b03ae3c | 852 | * debugobject calls are safe here even with gcwq->lock locked |
dc186ad7 TG |
853 | * as we know for sure that this will not trigger any of the |
854 | * checks and call back into the fixup functions where we | |
855 | * might deadlock. | |
856 | */ | |
857 | INIT_WORK_ON_STACK(&barr->work, wq_barrier_func); | |
22df02bb | 858 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 859 | init_completion(&barr->done); |
83c22520 | 860 | |
affee4b2 TH |
861 | /* |
862 | * If @target is currently being executed, schedule the | |
863 | * barrier to the worker; otherwise, put it after @target. | |
864 | */ | |
865 | if (worker) | |
866 | head = worker->scheduled.next; | |
867 | else { | |
868 | unsigned long *bits = work_data_bits(target); | |
869 | ||
870 | head = target->entry.next; | |
871 | /* there can already be other linked works, inherit and set */ | |
872 | linked = *bits & WORK_STRUCT_LINKED; | |
873 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
874 | } | |
875 | ||
dc186ad7 | 876 | debug_work_activate(&barr->work); |
affee4b2 TH |
877 | insert_work(cwq, &barr->work, head, |
878 | work_color_to_flags(WORK_NO_COLOR) | linked); | |
fc2e4d70 ON |
879 | } |
880 | ||
73f53c4a TH |
881 | /** |
882 | * flush_workqueue_prep_cwqs - prepare cwqs for workqueue flushing | |
883 | * @wq: workqueue being flushed | |
884 | * @flush_color: new flush color, < 0 for no-op | |
885 | * @work_color: new work color, < 0 for no-op | |
886 | * | |
887 | * Prepare cwqs for workqueue flushing. | |
888 | * | |
889 | * If @flush_color is non-negative, flush_color on all cwqs should be | |
890 | * -1. If no cwq has in-flight commands at the specified color, all | |
891 | * cwq->flush_color's stay at -1 and %false is returned. If any cwq | |
892 | * has in flight commands, its cwq->flush_color is set to | |
893 | * @flush_color, @wq->nr_cwqs_to_flush is updated accordingly, cwq | |
894 | * wakeup logic is armed and %true is returned. | |
895 | * | |
896 | * The caller should have initialized @wq->first_flusher prior to | |
897 | * calling this function with non-negative @flush_color. If | |
898 | * @flush_color is negative, no flush color update is done and %false | |
899 | * is returned. | |
900 | * | |
901 | * If @work_color is non-negative, all cwqs should have the same | |
902 | * work_color which is previous to @work_color and all will be | |
903 | * advanced to @work_color. | |
904 | * | |
905 | * CONTEXT: | |
906 | * mutex_lock(wq->flush_mutex). | |
907 | * | |
908 | * RETURNS: | |
909 | * %true if @flush_color >= 0 and there's something to flush. %false | |
910 | * otherwise. | |
911 | */ | |
912 | static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq, | |
913 | int flush_color, int work_color) | |
1da177e4 | 914 | { |
73f53c4a TH |
915 | bool wait = false; |
916 | unsigned int cpu; | |
1da177e4 | 917 | |
73f53c4a TH |
918 | if (flush_color >= 0) { |
919 | BUG_ON(atomic_read(&wq->nr_cwqs_to_flush)); | |
920 | atomic_set(&wq->nr_cwqs_to_flush, 1); | |
1da177e4 | 921 | } |
2355b70f | 922 | |
73f53c4a TH |
923 | for_each_possible_cpu(cpu) { |
924 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
8b03ae3c | 925 | struct global_cwq *gcwq = cwq->gcwq; |
73f53c4a | 926 | |
8b03ae3c | 927 | spin_lock_irq(&gcwq->lock); |
73f53c4a TH |
928 | |
929 | if (flush_color >= 0) { | |
930 | BUG_ON(cwq->flush_color != -1); | |
931 | ||
932 | if (cwq->nr_in_flight[flush_color]) { | |
933 | cwq->flush_color = flush_color; | |
934 | atomic_inc(&wq->nr_cwqs_to_flush); | |
935 | wait = true; | |
936 | } | |
937 | } | |
938 | ||
939 | if (work_color >= 0) { | |
940 | BUG_ON(work_color != work_next_color(cwq->work_color)); | |
941 | cwq->work_color = work_color; | |
942 | } | |
943 | ||
8b03ae3c | 944 | spin_unlock_irq(&gcwq->lock); |
dc186ad7 | 945 | } |
14441960 | 946 | |
73f53c4a TH |
947 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_cwqs_to_flush)) |
948 | complete(&wq->first_flusher->done); | |
949 | ||
950 | return wait; | |
1da177e4 LT |
951 | } |
952 | ||
0fcb78c2 | 953 | /** |
1da177e4 | 954 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 955 | * @wq: workqueue to flush |
1da177e4 LT |
956 | * |
957 | * Forces execution of the workqueue and blocks until its completion. | |
958 | * This is typically used in driver shutdown handlers. | |
959 | * | |
fc2e4d70 ON |
960 | * We sleep until all works which were queued on entry have been handled, |
961 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 962 | */ |
7ad5b3a5 | 963 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 964 | { |
73f53c4a TH |
965 | struct wq_flusher this_flusher = { |
966 | .list = LIST_HEAD_INIT(this_flusher.list), | |
967 | .flush_color = -1, | |
968 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
969 | }; | |
970 | int next_color; | |
1da177e4 | 971 | |
3295f0ef IM |
972 | lock_map_acquire(&wq->lockdep_map); |
973 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
974 | |
975 | mutex_lock(&wq->flush_mutex); | |
976 | ||
977 | /* | |
978 | * Start-to-wait phase | |
979 | */ | |
980 | next_color = work_next_color(wq->work_color); | |
981 | ||
982 | if (next_color != wq->flush_color) { | |
983 | /* | |
984 | * Color space is not full. The current work_color | |
985 | * becomes our flush_color and work_color is advanced | |
986 | * by one. | |
987 | */ | |
988 | BUG_ON(!list_empty(&wq->flusher_overflow)); | |
989 | this_flusher.flush_color = wq->work_color; | |
990 | wq->work_color = next_color; | |
991 | ||
992 | if (!wq->first_flusher) { | |
993 | /* no flush in progress, become the first flusher */ | |
994 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
995 | ||
996 | wq->first_flusher = &this_flusher; | |
997 | ||
998 | if (!flush_workqueue_prep_cwqs(wq, wq->flush_color, | |
999 | wq->work_color)) { | |
1000 | /* nothing to flush, done */ | |
1001 | wq->flush_color = next_color; | |
1002 | wq->first_flusher = NULL; | |
1003 | goto out_unlock; | |
1004 | } | |
1005 | } else { | |
1006 | /* wait in queue */ | |
1007 | BUG_ON(wq->flush_color == this_flusher.flush_color); | |
1008 | list_add_tail(&this_flusher.list, &wq->flusher_queue); | |
1009 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
1010 | } | |
1011 | } else { | |
1012 | /* | |
1013 | * Oops, color space is full, wait on overflow queue. | |
1014 | * The next flush completion will assign us | |
1015 | * flush_color and transfer to flusher_queue. | |
1016 | */ | |
1017 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
1018 | } | |
1019 | ||
1020 | mutex_unlock(&wq->flush_mutex); | |
1021 | ||
1022 | wait_for_completion(&this_flusher.done); | |
1023 | ||
1024 | /* | |
1025 | * Wake-up-and-cascade phase | |
1026 | * | |
1027 | * First flushers are responsible for cascading flushes and | |
1028 | * handling overflow. Non-first flushers can simply return. | |
1029 | */ | |
1030 | if (wq->first_flusher != &this_flusher) | |
1031 | return; | |
1032 | ||
1033 | mutex_lock(&wq->flush_mutex); | |
1034 | ||
1035 | wq->first_flusher = NULL; | |
1036 | ||
1037 | BUG_ON(!list_empty(&this_flusher.list)); | |
1038 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
1039 | ||
1040 | while (true) { | |
1041 | struct wq_flusher *next, *tmp; | |
1042 | ||
1043 | /* complete all the flushers sharing the current flush color */ | |
1044 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
1045 | if (next->flush_color != wq->flush_color) | |
1046 | break; | |
1047 | list_del_init(&next->list); | |
1048 | complete(&next->done); | |
1049 | } | |
1050 | ||
1051 | BUG_ON(!list_empty(&wq->flusher_overflow) && | |
1052 | wq->flush_color != work_next_color(wq->work_color)); | |
1053 | ||
1054 | /* this flush_color is finished, advance by one */ | |
1055 | wq->flush_color = work_next_color(wq->flush_color); | |
1056 | ||
1057 | /* one color has been freed, handle overflow queue */ | |
1058 | if (!list_empty(&wq->flusher_overflow)) { | |
1059 | /* | |
1060 | * Assign the same color to all overflowed | |
1061 | * flushers, advance work_color and append to | |
1062 | * flusher_queue. This is the start-to-wait | |
1063 | * phase for these overflowed flushers. | |
1064 | */ | |
1065 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
1066 | tmp->flush_color = wq->work_color; | |
1067 | ||
1068 | wq->work_color = work_next_color(wq->work_color); | |
1069 | ||
1070 | list_splice_tail_init(&wq->flusher_overflow, | |
1071 | &wq->flusher_queue); | |
1072 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
1073 | } | |
1074 | ||
1075 | if (list_empty(&wq->flusher_queue)) { | |
1076 | BUG_ON(wq->flush_color != wq->work_color); | |
1077 | break; | |
1078 | } | |
1079 | ||
1080 | /* | |
1081 | * Need to flush more colors. Make the next flusher | |
1082 | * the new first flusher and arm cwqs. | |
1083 | */ | |
1084 | BUG_ON(wq->flush_color == wq->work_color); | |
1085 | BUG_ON(wq->flush_color != next->flush_color); | |
1086 | ||
1087 | list_del_init(&next->list); | |
1088 | wq->first_flusher = next; | |
1089 | ||
1090 | if (flush_workqueue_prep_cwqs(wq, wq->flush_color, -1)) | |
1091 | break; | |
1092 | ||
1093 | /* | |
1094 | * Meh... this color is already done, clear first | |
1095 | * flusher and repeat cascading. | |
1096 | */ | |
1097 | wq->first_flusher = NULL; | |
1098 | } | |
1099 | ||
1100 | out_unlock: | |
1101 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 1102 | } |
ae90dd5d | 1103 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 1104 | |
db700897 ON |
1105 | /** |
1106 | * flush_work - block until a work_struct's callback has terminated | |
1107 | * @work: the work which is to be flushed | |
1108 | * | |
a67da70d ON |
1109 | * Returns false if @work has already terminated. |
1110 | * | |
db700897 ON |
1111 | * It is expected that, prior to calling flush_work(), the caller has |
1112 | * arranged for the work to not be requeued, otherwise it doesn't make | |
1113 | * sense to use this function. | |
1114 | */ | |
1115 | int flush_work(struct work_struct *work) | |
1116 | { | |
affee4b2 | 1117 | struct worker *worker = NULL; |
db700897 | 1118 | struct cpu_workqueue_struct *cwq; |
8b03ae3c | 1119 | struct global_cwq *gcwq; |
db700897 ON |
1120 | struct wq_barrier barr; |
1121 | ||
1122 | might_sleep(); | |
1123 | cwq = get_wq_data(work); | |
1124 | if (!cwq) | |
1125 | return 0; | |
8b03ae3c | 1126 | gcwq = cwq->gcwq; |
db700897 | 1127 | |
3295f0ef IM |
1128 | lock_map_acquire(&cwq->wq->lockdep_map); |
1129 | lock_map_release(&cwq->wq->lockdep_map); | |
a67da70d | 1130 | |
8b03ae3c | 1131 | spin_lock_irq(&gcwq->lock); |
db700897 ON |
1132 | if (!list_empty(&work->entry)) { |
1133 | /* | |
1134 | * See the comment near try_to_grab_pending()->smp_rmb(). | |
1135 | * If it was re-queued under us we are not going to wait. | |
1136 | */ | |
1137 | smp_rmb(); | |
1138 | if (unlikely(cwq != get_wq_data(work))) | |
4690c4ab | 1139 | goto already_gone; |
db700897 | 1140 | } else { |
affee4b2 TH |
1141 | if (cwq->worker && cwq->worker->current_work == work) |
1142 | worker = cwq->worker; | |
1143 | if (!worker) | |
4690c4ab | 1144 | goto already_gone; |
db700897 | 1145 | } |
db700897 | 1146 | |
affee4b2 | 1147 | insert_wq_barrier(cwq, &barr, work, worker); |
8b03ae3c | 1148 | spin_unlock_irq(&gcwq->lock); |
db700897 | 1149 | wait_for_completion(&barr.done); |
dc186ad7 | 1150 | destroy_work_on_stack(&barr.work); |
db700897 | 1151 | return 1; |
4690c4ab | 1152 | already_gone: |
8b03ae3c | 1153 | spin_unlock_irq(&gcwq->lock); |
4690c4ab | 1154 | return 0; |
db700897 ON |
1155 | } |
1156 | EXPORT_SYMBOL_GPL(flush_work); | |
1157 | ||
6e84d644 | 1158 | /* |
1f1f642e | 1159 | * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit, |
6e84d644 ON |
1160 | * so this work can't be re-armed in any way. |
1161 | */ | |
1162 | static int try_to_grab_pending(struct work_struct *work) | |
1163 | { | |
8b03ae3c | 1164 | struct global_cwq *gcwq; |
6e84d644 | 1165 | struct cpu_workqueue_struct *cwq; |
1f1f642e | 1166 | int ret = -1; |
6e84d644 | 1167 | |
22df02bb | 1168 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1f1f642e | 1169 | return 0; |
6e84d644 ON |
1170 | |
1171 | /* | |
1172 | * The queueing is in progress, or it is already queued. Try to | |
1173 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1174 | */ | |
1175 | ||
1176 | cwq = get_wq_data(work); | |
1177 | if (!cwq) | |
1178 | return ret; | |
8b03ae3c | 1179 | gcwq = cwq->gcwq; |
6e84d644 | 1180 | |
8b03ae3c | 1181 | spin_lock_irq(&gcwq->lock); |
6e84d644 ON |
1182 | if (!list_empty(&work->entry)) { |
1183 | /* | |
1184 | * This work is queued, but perhaps we locked the wrong cwq. | |
1185 | * In that case we must see the new value after rmb(), see | |
1186 | * insert_work()->wmb(). | |
1187 | */ | |
1188 | smp_rmb(); | |
1189 | if (cwq == get_wq_data(work)) { | |
dc186ad7 | 1190 | debug_work_deactivate(work); |
6e84d644 | 1191 | list_del_init(&work->entry); |
73f53c4a | 1192 | cwq_dec_nr_in_flight(cwq, get_work_color(work)); |
6e84d644 ON |
1193 | ret = 1; |
1194 | } | |
1195 | } | |
8b03ae3c | 1196 | spin_unlock_irq(&gcwq->lock); |
6e84d644 ON |
1197 | |
1198 | return ret; | |
1199 | } | |
1200 | ||
1201 | static void wait_on_cpu_work(struct cpu_workqueue_struct *cwq, | |
b89deed3 ON |
1202 | struct work_struct *work) |
1203 | { | |
8b03ae3c | 1204 | struct global_cwq *gcwq = cwq->gcwq; |
b89deed3 | 1205 | struct wq_barrier barr; |
affee4b2 | 1206 | struct worker *worker; |
b89deed3 | 1207 | |
8b03ae3c | 1208 | spin_lock_irq(&gcwq->lock); |
affee4b2 TH |
1209 | |
1210 | worker = NULL; | |
c34056a3 | 1211 | if (unlikely(cwq->worker && cwq->worker->current_work == work)) { |
affee4b2 TH |
1212 | worker = cwq->worker; |
1213 | insert_wq_barrier(cwq, &barr, work, worker); | |
b89deed3 | 1214 | } |
affee4b2 | 1215 | |
8b03ae3c | 1216 | spin_unlock_irq(&gcwq->lock); |
b89deed3 | 1217 | |
affee4b2 | 1218 | if (unlikely(worker)) { |
b89deed3 | 1219 | wait_for_completion(&barr.done); |
dc186ad7 TG |
1220 | destroy_work_on_stack(&barr.work); |
1221 | } | |
b89deed3 ON |
1222 | } |
1223 | ||
6e84d644 | 1224 | static void wait_on_work(struct work_struct *work) |
b89deed3 ON |
1225 | { |
1226 | struct cpu_workqueue_struct *cwq; | |
28e53bdd | 1227 | struct workqueue_struct *wq; |
b1f4ec17 | 1228 | int cpu; |
b89deed3 | 1229 | |
f293ea92 ON |
1230 | might_sleep(); |
1231 | ||
3295f0ef IM |
1232 | lock_map_acquire(&work->lockdep_map); |
1233 | lock_map_release(&work->lockdep_map); | |
4e6045f1 | 1234 | |
b89deed3 | 1235 | cwq = get_wq_data(work); |
b89deed3 | 1236 | if (!cwq) |
3af24433 | 1237 | return; |
b89deed3 | 1238 | |
28e53bdd | 1239 | wq = cwq->wq; |
28e53bdd | 1240 | |
1537663f | 1241 | for_each_possible_cpu(cpu) |
4690c4ab | 1242 | wait_on_cpu_work(get_cwq(cpu, wq), work); |
6e84d644 ON |
1243 | } |
1244 | ||
1f1f642e ON |
1245 | static int __cancel_work_timer(struct work_struct *work, |
1246 | struct timer_list* timer) | |
1247 | { | |
1248 | int ret; | |
1249 | ||
1250 | do { | |
1251 | ret = (timer && likely(del_timer(timer))); | |
1252 | if (!ret) | |
1253 | ret = try_to_grab_pending(work); | |
1254 | wait_on_work(work); | |
1255 | } while (unlikely(ret < 0)); | |
1256 | ||
4d707b9f | 1257 | clear_wq_data(work); |
1f1f642e ON |
1258 | return ret; |
1259 | } | |
1260 | ||
6e84d644 ON |
1261 | /** |
1262 | * cancel_work_sync - block until a work_struct's callback has terminated | |
1263 | * @work: the work which is to be flushed | |
1264 | * | |
1f1f642e ON |
1265 | * Returns true if @work was pending. |
1266 | * | |
6e84d644 ON |
1267 | * cancel_work_sync() will cancel the work if it is queued. If the work's |
1268 | * callback appears to be running, cancel_work_sync() will block until it | |
1269 | * has completed. | |
1270 | * | |
1271 | * It is possible to use this function if the work re-queues itself. It can | |
1272 | * cancel the work even if it migrates to another workqueue, however in that | |
1273 | * case it only guarantees that work->func() has completed on the last queued | |
1274 | * workqueue. | |
1275 | * | |
1276 | * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not | |
1277 | * pending, otherwise it goes into a busy-wait loop until the timer expires. | |
1278 | * | |
1279 | * The caller must ensure that workqueue_struct on which this work was last | |
1280 | * queued can't be destroyed before this function returns. | |
1281 | */ | |
1f1f642e | 1282 | int cancel_work_sync(struct work_struct *work) |
6e84d644 | 1283 | { |
1f1f642e | 1284 | return __cancel_work_timer(work, NULL); |
b89deed3 | 1285 | } |
28e53bdd | 1286 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 1287 | |
6e84d644 | 1288 | /** |
f5a421a4 | 1289 | * cancel_delayed_work_sync - reliably kill off a delayed work. |
6e84d644 ON |
1290 | * @dwork: the delayed work struct |
1291 | * | |
1f1f642e ON |
1292 | * Returns true if @dwork was pending. |
1293 | * | |
6e84d644 ON |
1294 | * It is possible to use this function if @dwork rearms itself via queue_work() |
1295 | * or queue_delayed_work(). See also the comment for cancel_work_sync(). | |
1296 | */ | |
1f1f642e | 1297 | int cancel_delayed_work_sync(struct delayed_work *dwork) |
6e84d644 | 1298 | { |
1f1f642e | 1299 | return __cancel_work_timer(&dwork->work, &dwork->timer); |
6e84d644 | 1300 | } |
f5a421a4 | 1301 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 1302 | |
6e84d644 | 1303 | static struct workqueue_struct *keventd_wq __read_mostly; |
1da177e4 | 1304 | |
0fcb78c2 REB |
1305 | /** |
1306 | * schedule_work - put work task in global workqueue | |
1307 | * @work: job to be done | |
1308 | * | |
5b0f437d BVA |
1309 | * Returns zero if @work was already on the kernel-global workqueue and |
1310 | * non-zero otherwise. | |
1311 | * | |
1312 | * This puts a job in the kernel-global workqueue if it was not already | |
1313 | * queued and leaves it in the same position on the kernel-global | |
1314 | * workqueue otherwise. | |
0fcb78c2 | 1315 | */ |
7ad5b3a5 | 1316 | int schedule_work(struct work_struct *work) |
1da177e4 LT |
1317 | { |
1318 | return queue_work(keventd_wq, work); | |
1319 | } | |
ae90dd5d | 1320 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 1321 | |
c1a220e7 ZR |
1322 | /* |
1323 | * schedule_work_on - put work task on a specific cpu | |
1324 | * @cpu: cpu to put the work task on | |
1325 | * @work: job to be done | |
1326 | * | |
1327 | * This puts a job on a specific cpu | |
1328 | */ | |
1329 | int schedule_work_on(int cpu, struct work_struct *work) | |
1330 | { | |
1331 | return queue_work_on(cpu, keventd_wq, work); | |
1332 | } | |
1333 | EXPORT_SYMBOL(schedule_work_on); | |
1334 | ||
0fcb78c2 REB |
1335 | /** |
1336 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
1337 | * @dwork: job to be done |
1338 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
1339 | * |
1340 | * After waiting for a given time this puts a job in the kernel-global | |
1341 | * workqueue. | |
1342 | */ | |
7ad5b3a5 | 1343 | int schedule_delayed_work(struct delayed_work *dwork, |
82f67cd9 | 1344 | unsigned long delay) |
1da177e4 | 1345 | { |
52bad64d | 1346 | return queue_delayed_work(keventd_wq, dwork, delay); |
1da177e4 | 1347 | } |
ae90dd5d | 1348 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 1349 | |
8c53e463 LT |
1350 | /** |
1351 | * flush_delayed_work - block until a dwork_struct's callback has terminated | |
1352 | * @dwork: the delayed work which is to be flushed | |
1353 | * | |
1354 | * Any timeout is cancelled, and any pending work is run immediately. | |
1355 | */ | |
1356 | void flush_delayed_work(struct delayed_work *dwork) | |
1357 | { | |
1358 | if (del_timer_sync(&dwork->timer)) { | |
4690c4ab TH |
1359 | __queue_work(get_cpu(), get_wq_data(&dwork->work)->wq, |
1360 | &dwork->work); | |
8c53e463 LT |
1361 | put_cpu(); |
1362 | } | |
1363 | flush_work(&dwork->work); | |
1364 | } | |
1365 | EXPORT_SYMBOL(flush_delayed_work); | |
1366 | ||
0fcb78c2 REB |
1367 | /** |
1368 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
1369 | * @cpu: cpu to use | |
52bad64d | 1370 | * @dwork: job to be done |
0fcb78c2 REB |
1371 | * @delay: number of jiffies to wait |
1372 | * | |
1373 | * After waiting for a given time this puts a job in the kernel-global | |
1374 | * workqueue on the specified CPU. | |
1375 | */ | |
1da177e4 | 1376 | int schedule_delayed_work_on(int cpu, |
52bad64d | 1377 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 1378 | { |
52bad64d | 1379 | return queue_delayed_work_on(cpu, keventd_wq, dwork, delay); |
1da177e4 | 1380 | } |
ae90dd5d | 1381 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 1382 | |
b6136773 AM |
1383 | /** |
1384 | * schedule_on_each_cpu - call a function on each online CPU from keventd | |
1385 | * @func: the function to call | |
b6136773 AM |
1386 | * |
1387 | * Returns zero on success. | |
1388 | * Returns -ve errno on failure. | |
1389 | * | |
b6136773 AM |
1390 | * schedule_on_each_cpu() is very slow. |
1391 | */ | |
65f27f38 | 1392 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
1393 | { |
1394 | int cpu; | |
65a64464 | 1395 | int orig = -1; |
b6136773 | 1396 | struct work_struct *works; |
15316ba8 | 1397 | |
b6136773 AM |
1398 | works = alloc_percpu(struct work_struct); |
1399 | if (!works) | |
15316ba8 | 1400 | return -ENOMEM; |
b6136773 | 1401 | |
93981800 TH |
1402 | get_online_cpus(); |
1403 | ||
65a64464 | 1404 | /* |
93981800 TH |
1405 | * When running in keventd don't schedule a work item on |
1406 | * itself. Can just call directly because the work queue is | |
1407 | * already bound. This also is faster. | |
65a64464 | 1408 | */ |
93981800 | 1409 | if (current_is_keventd()) |
65a64464 | 1410 | orig = raw_smp_processor_id(); |
65a64464 | 1411 | |
15316ba8 | 1412 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
1413 | struct work_struct *work = per_cpu_ptr(works, cpu); |
1414 | ||
1415 | INIT_WORK(work, func); | |
65a64464 | 1416 | if (cpu != orig) |
93981800 | 1417 | schedule_work_on(cpu, work); |
65a64464 | 1418 | } |
93981800 TH |
1419 | if (orig >= 0) |
1420 | func(per_cpu_ptr(works, orig)); | |
1421 | ||
1422 | for_each_online_cpu(cpu) | |
1423 | flush_work(per_cpu_ptr(works, cpu)); | |
1424 | ||
95402b38 | 1425 | put_online_cpus(); |
b6136773 | 1426 | free_percpu(works); |
15316ba8 CL |
1427 | return 0; |
1428 | } | |
1429 | ||
eef6a7d5 AS |
1430 | /** |
1431 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
1432 | * | |
1433 | * Forces execution of the kernel-global workqueue and blocks until its | |
1434 | * completion. | |
1435 | * | |
1436 | * Think twice before calling this function! It's very easy to get into | |
1437 | * trouble if you don't take great care. Either of the following situations | |
1438 | * will lead to deadlock: | |
1439 | * | |
1440 | * One of the work items currently on the workqueue needs to acquire | |
1441 | * a lock held by your code or its caller. | |
1442 | * | |
1443 | * Your code is running in the context of a work routine. | |
1444 | * | |
1445 | * They will be detected by lockdep when they occur, but the first might not | |
1446 | * occur very often. It depends on what work items are on the workqueue and | |
1447 | * what locks they need, which you have no control over. | |
1448 | * | |
1449 | * In most situations flushing the entire workqueue is overkill; you merely | |
1450 | * need to know that a particular work item isn't queued and isn't running. | |
1451 | * In such cases you should use cancel_delayed_work_sync() or | |
1452 | * cancel_work_sync() instead. | |
1453 | */ | |
1da177e4 LT |
1454 | void flush_scheduled_work(void) |
1455 | { | |
1456 | flush_workqueue(keventd_wq); | |
1457 | } | |
ae90dd5d | 1458 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 1459 | |
1fa44eca JB |
1460 | /** |
1461 | * execute_in_process_context - reliably execute the routine with user context | |
1462 | * @fn: the function to execute | |
1fa44eca JB |
1463 | * @ew: guaranteed storage for the execute work structure (must |
1464 | * be available when the work executes) | |
1465 | * | |
1466 | * Executes the function immediately if process context is available, | |
1467 | * otherwise schedules the function for delayed execution. | |
1468 | * | |
1469 | * Returns: 0 - function was executed | |
1470 | * 1 - function was scheduled for execution | |
1471 | */ | |
65f27f38 | 1472 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
1473 | { |
1474 | if (!in_interrupt()) { | |
65f27f38 | 1475 | fn(&ew->work); |
1fa44eca JB |
1476 | return 0; |
1477 | } | |
1478 | ||
65f27f38 | 1479 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
1480 | schedule_work(&ew->work); |
1481 | ||
1482 | return 1; | |
1483 | } | |
1484 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
1485 | ||
1da177e4 LT |
1486 | int keventd_up(void) |
1487 | { | |
1488 | return keventd_wq != NULL; | |
1489 | } | |
1490 | ||
1491 | int current_is_keventd(void) | |
1492 | { | |
1493 | struct cpu_workqueue_struct *cwq; | |
d243769d | 1494 | int cpu = raw_smp_processor_id(); /* preempt-safe: keventd is per-cpu */ |
1da177e4 LT |
1495 | int ret = 0; |
1496 | ||
1497 | BUG_ON(!keventd_wq); | |
1498 | ||
1537663f | 1499 | cwq = get_cwq(cpu, keventd_wq); |
c34056a3 | 1500 | if (current == cwq->worker->task) |
1da177e4 LT |
1501 | ret = 1; |
1502 | ||
1503 | return ret; | |
1504 | ||
1505 | } | |
1506 | ||
0f900049 TH |
1507 | static struct cpu_workqueue_struct *alloc_cwqs(void) |
1508 | { | |
1509 | /* | |
1510 | * cwqs are forced aligned according to WORK_STRUCT_FLAG_BITS. | |
1511 | * Make sure that the alignment isn't lower than that of | |
1512 | * unsigned long long. | |
1513 | */ | |
1514 | const size_t size = sizeof(struct cpu_workqueue_struct); | |
1515 | const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS, | |
1516 | __alignof__(unsigned long long)); | |
1517 | struct cpu_workqueue_struct *cwqs; | |
1518 | #ifndef CONFIG_SMP | |
1519 | void *ptr; | |
1520 | ||
1521 | /* | |
1522 | * On UP, percpu allocator doesn't honor alignment parameter | |
1523 | * and simply uses arch-dependent default. Allocate enough | |
1524 | * room to align cwq and put an extra pointer at the end | |
1525 | * pointing back to the originally allocated pointer which | |
1526 | * will be used for free. | |
1527 | * | |
1528 | * FIXME: This really belongs to UP percpu code. Update UP | |
1529 | * percpu code to honor alignment and remove this ugliness. | |
1530 | */ | |
1531 | ptr = __alloc_percpu(size + align + sizeof(void *), 1); | |
1532 | cwqs = PTR_ALIGN(ptr, align); | |
1533 | *(void **)per_cpu_ptr(cwqs + 1, 0) = ptr; | |
1534 | #else | |
1535 | /* On SMP, percpu allocator can do it itself */ | |
1536 | cwqs = __alloc_percpu(size, align); | |
1537 | #endif | |
1538 | /* just in case, make sure it's actually aligned */ | |
1539 | BUG_ON(!IS_ALIGNED((unsigned long)cwqs, align)); | |
1540 | return cwqs; | |
1541 | } | |
1542 | ||
1543 | static void free_cwqs(struct cpu_workqueue_struct *cwqs) | |
1544 | { | |
1545 | #ifndef CONFIG_SMP | |
1546 | /* on UP, the pointer to free is stored right after the cwq */ | |
1547 | if (cwqs) | |
1548 | free_percpu(*(void **)per_cpu_ptr(cwqs + 1, 0)); | |
1549 | #else | |
1550 | free_percpu(cwqs); | |
1551 | #endif | |
1552 | } | |
1553 | ||
4e6045f1 | 1554 | struct workqueue_struct *__create_workqueue_key(const char *name, |
97e37d7b | 1555 | unsigned int flags, |
1e19ffc6 | 1556 | int max_active, |
eb13ba87 JB |
1557 | struct lock_class_key *key, |
1558 | const char *lock_name) | |
1da177e4 | 1559 | { |
1537663f | 1560 | bool singlethread = flags & WQ_SINGLE_THREAD; |
1da177e4 | 1561 | struct workqueue_struct *wq; |
c34056a3 TH |
1562 | bool failed = false; |
1563 | unsigned int cpu; | |
1da177e4 | 1564 | |
1e19ffc6 TH |
1565 | max_active = clamp_val(max_active, 1, INT_MAX); |
1566 | ||
3af24433 ON |
1567 | wq = kzalloc(sizeof(*wq), GFP_KERNEL); |
1568 | if (!wq) | |
4690c4ab | 1569 | goto err; |
3af24433 | 1570 | |
0f900049 | 1571 | wq->cpu_wq = alloc_cwqs(); |
4690c4ab TH |
1572 | if (!wq->cpu_wq) |
1573 | goto err; | |
3af24433 | 1574 | |
97e37d7b | 1575 | wq->flags = flags; |
a0a1a5fd | 1576 | wq->saved_max_active = max_active; |
73f53c4a TH |
1577 | mutex_init(&wq->flush_mutex); |
1578 | atomic_set(&wq->nr_cwqs_to_flush, 0); | |
1579 | INIT_LIST_HEAD(&wq->flusher_queue); | |
1580 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
3af24433 | 1581 | wq->name = name; |
eb13ba87 | 1582 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 1583 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 1584 | |
1537663f TH |
1585 | cpu_maps_update_begin(); |
1586 | /* | |
1587 | * We must initialize cwqs for each possible cpu even if we | |
1588 | * are going to call destroy_workqueue() finally. Otherwise | |
1589 | * cpu_up() can hit the uninitialized cwq once we drop the | |
1590 | * lock. | |
1591 | */ | |
1592 | for_each_possible_cpu(cpu) { | |
1593 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
8b03ae3c | 1594 | struct global_cwq *gcwq = get_gcwq(cpu); |
1537663f | 1595 | |
0f900049 | 1596 | BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK); |
8b03ae3c | 1597 | cwq->gcwq = gcwq; |
c34056a3 | 1598 | cwq->wq = wq; |
73f53c4a | 1599 | cwq->flush_color = -1; |
1e19ffc6 | 1600 | cwq->max_active = max_active; |
1537663f | 1601 | INIT_LIST_HEAD(&cwq->worklist); |
1e19ffc6 | 1602 | INIT_LIST_HEAD(&cwq->delayed_works); |
1537663f TH |
1603 | init_waitqueue_head(&cwq->more_work); |
1604 | ||
c34056a3 | 1605 | if (failed) |
1537663f | 1606 | continue; |
c34056a3 TH |
1607 | cwq->worker = create_worker(cwq, |
1608 | cpu_online(cpu) && !singlethread); | |
1609 | if (cwq->worker) | |
1610 | start_worker(cwq->worker); | |
1537663f | 1611 | else |
c34056a3 | 1612 | failed = true; |
3af24433 ON |
1613 | } |
1614 | ||
a0a1a5fd TH |
1615 | /* |
1616 | * workqueue_lock protects global freeze state and workqueues | |
1617 | * list. Grab it, set max_active accordingly and add the new | |
1618 | * workqueue to workqueues list. | |
1619 | */ | |
1537663f | 1620 | spin_lock(&workqueue_lock); |
a0a1a5fd TH |
1621 | |
1622 | if (workqueue_freezing && wq->flags & WQ_FREEZEABLE) | |
1623 | for_each_possible_cpu(cpu) | |
1624 | get_cwq(cpu, wq)->max_active = 0; | |
1625 | ||
1537663f | 1626 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 1627 | |
1537663f TH |
1628 | spin_unlock(&workqueue_lock); |
1629 | ||
1630 | cpu_maps_update_done(); | |
1631 | ||
c34056a3 | 1632 | if (failed) { |
3af24433 ON |
1633 | destroy_workqueue(wq); |
1634 | wq = NULL; | |
1635 | } | |
1636 | return wq; | |
4690c4ab TH |
1637 | err: |
1638 | if (wq) { | |
0f900049 | 1639 | free_cwqs(wq->cpu_wq); |
4690c4ab TH |
1640 | kfree(wq); |
1641 | } | |
1642 | return NULL; | |
3af24433 | 1643 | } |
4e6045f1 | 1644 | EXPORT_SYMBOL_GPL(__create_workqueue_key); |
1da177e4 | 1645 | |
3af24433 ON |
1646 | /** |
1647 | * destroy_workqueue - safely terminate a workqueue | |
1648 | * @wq: target workqueue | |
1649 | * | |
1650 | * Safely destroy a workqueue. All work currently pending will be done first. | |
1651 | */ | |
1652 | void destroy_workqueue(struct workqueue_struct *wq) | |
1653 | { | |
b1f4ec17 | 1654 | int cpu; |
3af24433 | 1655 | |
a0a1a5fd TH |
1656 | flush_workqueue(wq); |
1657 | ||
1658 | /* | |
1659 | * wq list is used to freeze wq, remove from list after | |
1660 | * flushing is complete in case freeze races us. | |
1661 | */ | |
3da1c84c | 1662 | cpu_maps_update_begin(); |
95402b38 | 1663 | spin_lock(&workqueue_lock); |
b1f4ec17 | 1664 | list_del(&wq->list); |
95402b38 | 1665 | spin_unlock(&workqueue_lock); |
1537663f | 1666 | cpu_maps_update_done(); |
3af24433 | 1667 | |
73f53c4a TH |
1668 | for_each_possible_cpu(cpu) { |
1669 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
1670 | int i; | |
1671 | ||
c34056a3 TH |
1672 | if (cwq->worker) { |
1673 | destroy_worker(cwq->worker); | |
1674 | cwq->worker = NULL; | |
73f53c4a TH |
1675 | } |
1676 | ||
1677 | for (i = 0; i < WORK_NR_COLORS; i++) | |
1678 | BUG_ON(cwq->nr_in_flight[i]); | |
1e19ffc6 TH |
1679 | BUG_ON(cwq->nr_active); |
1680 | BUG_ON(!list_empty(&cwq->delayed_works)); | |
73f53c4a | 1681 | } |
9b41ea72 | 1682 | |
0f900049 | 1683 | free_cwqs(wq->cpu_wq); |
3af24433 ON |
1684 | kfree(wq); |
1685 | } | |
1686 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
1687 | ||
1688 | static int __devinit workqueue_cpu_callback(struct notifier_block *nfb, | |
1689 | unsigned long action, | |
1690 | void *hcpu) | |
1691 | { | |
1692 | unsigned int cpu = (unsigned long)hcpu; | |
1693 | struct cpu_workqueue_struct *cwq; | |
1694 | struct workqueue_struct *wq; | |
1695 | ||
8bb78442 RW |
1696 | action &= ~CPU_TASKS_FROZEN; |
1697 | ||
3af24433 | 1698 | list_for_each_entry(wq, &workqueues, list) { |
1537663f TH |
1699 | if (wq->flags & WQ_SINGLE_THREAD) |
1700 | continue; | |
3af24433 | 1701 | |
1537663f | 1702 | cwq = get_cwq(cpu, wq); |
3af24433 | 1703 | |
1537663f | 1704 | switch (action) { |
3da1c84c | 1705 | case CPU_POST_DEAD: |
73f53c4a | 1706 | flush_workqueue(wq); |
3af24433 ON |
1707 | break; |
1708 | } | |
1da177e4 LT |
1709 | } |
1710 | ||
1537663f | 1711 | return notifier_from_errno(0); |
1da177e4 | 1712 | } |
1da177e4 | 1713 | |
2d3854a3 | 1714 | #ifdef CONFIG_SMP |
8ccad40d | 1715 | |
2d3854a3 | 1716 | struct work_for_cpu { |
6b44003e | 1717 | struct completion completion; |
2d3854a3 RR |
1718 | long (*fn)(void *); |
1719 | void *arg; | |
1720 | long ret; | |
1721 | }; | |
1722 | ||
6b44003e | 1723 | static int do_work_for_cpu(void *_wfc) |
2d3854a3 | 1724 | { |
6b44003e | 1725 | struct work_for_cpu *wfc = _wfc; |
2d3854a3 | 1726 | wfc->ret = wfc->fn(wfc->arg); |
6b44003e AM |
1727 | complete(&wfc->completion); |
1728 | return 0; | |
2d3854a3 RR |
1729 | } |
1730 | ||
1731 | /** | |
1732 | * work_on_cpu - run a function in user context on a particular cpu | |
1733 | * @cpu: the cpu to run on | |
1734 | * @fn: the function to run | |
1735 | * @arg: the function arg | |
1736 | * | |
31ad9081 RR |
1737 | * This will return the value @fn returns. |
1738 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 1739 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 RR |
1740 | */ |
1741 | long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) | |
1742 | { | |
6b44003e AM |
1743 | struct task_struct *sub_thread; |
1744 | struct work_for_cpu wfc = { | |
1745 | .completion = COMPLETION_INITIALIZER_ONSTACK(wfc.completion), | |
1746 | .fn = fn, | |
1747 | .arg = arg, | |
1748 | }; | |
1749 | ||
1750 | sub_thread = kthread_create(do_work_for_cpu, &wfc, "work_for_cpu"); | |
1751 | if (IS_ERR(sub_thread)) | |
1752 | return PTR_ERR(sub_thread); | |
1753 | kthread_bind(sub_thread, cpu); | |
1754 | wake_up_process(sub_thread); | |
1755 | wait_for_completion(&wfc.completion); | |
2d3854a3 RR |
1756 | return wfc.ret; |
1757 | } | |
1758 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
1759 | #endif /* CONFIG_SMP */ | |
1760 | ||
a0a1a5fd TH |
1761 | #ifdef CONFIG_FREEZER |
1762 | ||
1763 | /** | |
1764 | * freeze_workqueues_begin - begin freezing workqueues | |
1765 | * | |
1766 | * Start freezing workqueues. After this function returns, all | |
1767 | * freezeable workqueues will queue new works to their frozen_works | |
1768 | * list instead of the cwq ones. | |
1769 | * | |
1770 | * CONTEXT: | |
8b03ae3c | 1771 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
1772 | */ |
1773 | void freeze_workqueues_begin(void) | |
1774 | { | |
1775 | struct workqueue_struct *wq; | |
1776 | unsigned int cpu; | |
1777 | ||
1778 | spin_lock(&workqueue_lock); | |
1779 | ||
1780 | BUG_ON(workqueue_freezing); | |
1781 | workqueue_freezing = true; | |
1782 | ||
1783 | for_each_possible_cpu(cpu) { | |
8b03ae3c TH |
1784 | struct global_cwq *gcwq = get_gcwq(cpu); |
1785 | ||
1786 | spin_lock_irq(&gcwq->lock); | |
1787 | ||
a0a1a5fd TH |
1788 | list_for_each_entry(wq, &workqueues, list) { |
1789 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
1790 | ||
a0a1a5fd TH |
1791 | if (wq->flags & WQ_FREEZEABLE) |
1792 | cwq->max_active = 0; | |
a0a1a5fd | 1793 | } |
8b03ae3c TH |
1794 | |
1795 | spin_unlock_irq(&gcwq->lock); | |
a0a1a5fd TH |
1796 | } |
1797 | ||
1798 | spin_unlock(&workqueue_lock); | |
1799 | } | |
1800 | ||
1801 | /** | |
1802 | * freeze_workqueues_busy - are freezeable workqueues still busy? | |
1803 | * | |
1804 | * Check whether freezing is complete. This function must be called | |
1805 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
1806 | * | |
1807 | * CONTEXT: | |
1808 | * Grabs and releases workqueue_lock. | |
1809 | * | |
1810 | * RETURNS: | |
1811 | * %true if some freezeable workqueues are still busy. %false if | |
1812 | * freezing is complete. | |
1813 | */ | |
1814 | bool freeze_workqueues_busy(void) | |
1815 | { | |
1816 | struct workqueue_struct *wq; | |
1817 | unsigned int cpu; | |
1818 | bool busy = false; | |
1819 | ||
1820 | spin_lock(&workqueue_lock); | |
1821 | ||
1822 | BUG_ON(!workqueue_freezing); | |
1823 | ||
1824 | for_each_possible_cpu(cpu) { | |
1825 | /* | |
1826 | * nr_active is monotonically decreasing. It's safe | |
1827 | * to peek without lock. | |
1828 | */ | |
1829 | list_for_each_entry(wq, &workqueues, list) { | |
1830 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
1831 | ||
1832 | if (!(wq->flags & WQ_FREEZEABLE)) | |
1833 | continue; | |
1834 | ||
1835 | BUG_ON(cwq->nr_active < 0); | |
1836 | if (cwq->nr_active) { | |
1837 | busy = true; | |
1838 | goto out_unlock; | |
1839 | } | |
1840 | } | |
1841 | } | |
1842 | out_unlock: | |
1843 | spin_unlock(&workqueue_lock); | |
1844 | return busy; | |
1845 | } | |
1846 | ||
1847 | /** | |
1848 | * thaw_workqueues - thaw workqueues | |
1849 | * | |
1850 | * Thaw workqueues. Normal queueing is restored and all collected | |
1851 | * frozen works are transferred to their respective cwq worklists. | |
1852 | * | |
1853 | * CONTEXT: | |
8b03ae3c | 1854 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
1855 | */ |
1856 | void thaw_workqueues(void) | |
1857 | { | |
1858 | struct workqueue_struct *wq; | |
1859 | unsigned int cpu; | |
1860 | ||
1861 | spin_lock(&workqueue_lock); | |
1862 | ||
1863 | if (!workqueue_freezing) | |
1864 | goto out_unlock; | |
1865 | ||
1866 | for_each_possible_cpu(cpu) { | |
8b03ae3c TH |
1867 | struct global_cwq *gcwq = get_gcwq(cpu); |
1868 | ||
1869 | spin_lock_irq(&gcwq->lock); | |
1870 | ||
a0a1a5fd TH |
1871 | list_for_each_entry(wq, &workqueues, list) { |
1872 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
1873 | ||
1874 | if (!(wq->flags & WQ_FREEZEABLE)) | |
1875 | continue; | |
1876 | ||
a0a1a5fd TH |
1877 | /* restore max_active and repopulate worklist */ |
1878 | cwq->max_active = wq->saved_max_active; | |
1879 | ||
1880 | while (!list_empty(&cwq->delayed_works) && | |
1881 | cwq->nr_active < cwq->max_active) | |
1882 | cwq_activate_first_delayed(cwq); | |
1883 | ||
1884 | wake_up(&cwq->more_work); | |
a0a1a5fd | 1885 | } |
8b03ae3c TH |
1886 | |
1887 | spin_unlock_irq(&gcwq->lock); | |
a0a1a5fd TH |
1888 | } |
1889 | ||
1890 | workqueue_freezing = false; | |
1891 | out_unlock: | |
1892 | spin_unlock(&workqueue_lock); | |
1893 | } | |
1894 | #endif /* CONFIG_FREEZER */ | |
1895 | ||
c12920d1 | 1896 | void __init init_workqueues(void) |
1da177e4 | 1897 | { |
c34056a3 TH |
1898 | unsigned int cpu; |
1899 | ||
e7577c50 | 1900 | singlethread_cpu = cpumask_first(cpu_possible_mask); |
1da177e4 | 1901 | hotcpu_notifier(workqueue_cpu_callback, 0); |
8b03ae3c TH |
1902 | |
1903 | /* initialize gcwqs */ | |
1904 | for_each_possible_cpu(cpu) { | |
1905 | struct global_cwq *gcwq = get_gcwq(cpu); | |
1906 | ||
1907 | spin_lock_init(&gcwq->lock); | |
1908 | gcwq->cpu = cpu; | |
1909 | ||
1910 | ida_init(&gcwq->worker_ida); | |
1911 | } | |
1912 | ||
1da177e4 LT |
1913 | keventd_wq = create_workqueue("events"); |
1914 | BUG_ON(!keventd_wq); | |
1915 | } |