]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
c54fce6e | 2 | * kernel/workqueue.c - generic async execution with shared worker pool |
1da177e4 | 3 | * |
c54fce6e | 4 | * Copyright (C) 2002 Ingo Molnar |
1da177e4 | 5 | * |
c54fce6e TH |
6 | * Derived from the taskqueue/keventd code by: |
7 | * David Woodhouse <dwmw2@infradead.org> | |
8 | * Andrew Morton | |
9 | * Kai Petzke <wpp@marie.physik.tu-berlin.de> | |
10 | * Theodore Ts'o <tytso@mit.edu> | |
1da177e4 | 11 | * |
c54fce6e | 12 | * Made to use alloc_percpu by Christoph Lameter. |
1da177e4 | 13 | * |
c54fce6e TH |
14 | * Copyright (C) 2010 SUSE Linux Products GmbH |
15 | * Copyright (C) 2010 Tejun Heo <tj@kernel.org> | |
89ada679 | 16 | * |
c54fce6e TH |
17 | * This is the generic async execution mechanism. Work items as are |
18 | * executed in process context. The worker pool is shared and | |
19 | * automatically managed. There is one worker pool for each CPU and | |
20 | * one extra for works which are better served by workers which are | |
21 | * not bound to any specific CPU. | |
22 | * | |
23 | * Please read Documentation/workqueue.txt for details. | |
1da177e4 LT |
24 | */ |
25 | ||
9984de1a | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/kernel.h> |
28 | #include <linux/sched.h> | |
29 | #include <linux/init.h> | |
30 | #include <linux/signal.h> | |
31 | #include <linux/completion.h> | |
32 | #include <linux/workqueue.h> | |
33 | #include <linux/slab.h> | |
34 | #include <linux/cpu.h> | |
35 | #include <linux/notifier.h> | |
36 | #include <linux/kthread.h> | |
1fa44eca | 37 | #include <linux/hardirq.h> |
46934023 | 38 | #include <linux/mempolicy.h> |
341a5958 | 39 | #include <linux/freezer.h> |
d5abe669 PZ |
40 | #include <linux/kallsyms.h> |
41 | #include <linux/debug_locks.h> | |
4e6045f1 | 42 | #include <linux/lockdep.h> |
c34056a3 | 43 | #include <linux/idr.h> |
e22bee78 TH |
44 | |
45 | #include "workqueue_sched.h" | |
1da177e4 | 46 | |
c8e55f36 | 47 | enum { |
bc2ae0f5 TH |
48 | /* |
49 | * global_cwq flags | |
50 | * | |
51 | * A bound gcwq is either associated or disassociated with its CPU. | |
52 | * While associated (!DISASSOCIATED), all workers are bound to the | |
53 | * CPU and none has %WORKER_UNBOUND set and concurrency management | |
54 | * is in effect. | |
55 | * | |
56 | * While DISASSOCIATED, the cpu may be offline and all workers have | |
57 | * %WORKER_UNBOUND set and concurrency management disabled, and may | |
58 | * be executing on any CPU. The gcwq behaves as an unbound one. | |
59 | * | |
60 | * Note that DISASSOCIATED can be flipped only while holding | |
b2eb83d1 | 61 | * assoc_mutex of all pools on the gcwq to avoid changing binding |
bc2ae0f5 TH |
62 | * state while create_worker() is in progress. |
63 | */ | |
11ebea50 TH |
64 | GCWQ_DISASSOCIATED = 1 << 0, /* cpu can't serve workers */ |
65 | GCWQ_FREEZING = 1 << 1, /* freeze in progress */ | |
66 | ||
67 | /* pool flags */ | |
68 | POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */ | |
552a37e9 | 69 | POOL_MANAGING_WORKERS = 1 << 1, /* managing workers */ |
db7bccf4 | 70 | |
c8e55f36 TH |
71 | /* worker flags */ |
72 | WORKER_STARTED = 1 << 0, /* started */ | |
73 | WORKER_DIE = 1 << 1, /* die die die */ | |
74 | WORKER_IDLE = 1 << 2, /* is idle */ | |
e22bee78 | 75 | WORKER_PREP = 1 << 3, /* preparing to run works */ |
fb0e7beb | 76 | WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ |
f3421797 | 77 | WORKER_UNBOUND = 1 << 7, /* worker is unbound */ |
e22bee78 | 78 | |
5f7dabfd | 79 | WORKER_NOT_RUNNING = WORKER_PREP | WORKER_UNBOUND | |
403c821d | 80 | WORKER_CPU_INTENSIVE, |
db7bccf4 | 81 | |
3270476a | 82 | NR_WORKER_POOLS = 2, /* # worker pools per gcwq */ |
4ce62e9e | 83 | |
c8e55f36 TH |
84 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ |
85 | BUSY_WORKER_HASH_SIZE = 1 << BUSY_WORKER_HASH_ORDER, | |
86 | BUSY_WORKER_HASH_MASK = BUSY_WORKER_HASH_SIZE - 1, | |
db7bccf4 | 87 | |
e22bee78 TH |
88 | MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */ |
89 | IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */ | |
90 | ||
3233cdbd TH |
91 | MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2, |
92 | /* call for help after 10ms | |
93 | (min two ticks) */ | |
e22bee78 TH |
94 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ |
95 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
e22bee78 TH |
96 | |
97 | /* | |
98 | * Rescue workers are used only on emergencies and shared by | |
99 | * all cpus. Give -20. | |
100 | */ | |
101 | RESCUER_NICE_LEVEL = -20, | |
3270476a | 102 | HIGHPRI_NICE_LEVEL = -20, |
c8e55f36 | 103 | }; |
1da177e4 LT |
104 | |
105 | /* | |
4690c4ab TH |
106 | * Structure fields follow one of the following exclusion rules. |
107 | * | |
e41e704b TH |
108 | * I: Modifiable by initialization/destruction paths and read-only for |
109 | * everyone else. | |
4690c4ab | 110 | * |
e22bee78 TH |
111 | * P: Preemption protected. Disabling preemption is enough and should |
112 | * only be modified and accessed from the local cpu. | |
113 | * | |
8b03ae3c | 114 | * L: gcwq->lock protected. Access with gcwq->lock held. |
4690c4ab | 115 | * |
e22bee78 TH |
116 | * X: During normal operation, modification requires gcwq->lock and |
117 | * should be done only from local cpu. Either disabling preemption | |
118 | * on local cpu or grabbing gcwq->lock is enough for read access. | |
f3421797 | 119 | * If GCWQ_DISASSOCIATED is set, it's identical to L. |
e22bee78 | 120 | * |
73f53c4a TH |
121 | * F: wq->flush_mutex protected. |
122 | * | |
4690c4ab | 123 | * W: workqueue_lock protected. |
1da177e4 | 124 | */ |
1da177e4 | 125 | |
8b03ae3c | 126 | struct global_cwq; |
bd7bdd43 | 127 | struct worker_pool; |
1da177e4 | 128 | |
e22bee78 TH |
129 | /* |
130 | * The poor guys doing the actual heavy lifting. All on-duty workers | |
131 | * are either serving the manager role, on idle list or on busy hash. | |
132 | */ | |
c34056a3 | 133 | struct worker { |
c8e55f36 TH |
134 | /* on idle list while idle, on busy hash table while busy */ |
135 | union { | |
136 | struct list_head entry; /* L: while idle */ | |
137 | struct hlist_node hentry; /* L: while busy */ | |
138 | }; | |
1da177e4 | 139 | |
c34056a3 | 140 | struct work_struct *current_work; /* L: work being processed */ |
8cca0eea | 141 | struct cpu_workqueue_struct *current_cwq; /* L: current_work's cwq */ |
affee4b2 | 142 | struct list_head scheduled; /* L: scheduled works */ |
c34056a3 | 143 | struct task_struct *task; /* I: worker task */ |
bd7bdd43 | 144 | struct worker_pool *pool; /* I: the associated pool */ |
e22bee78 TH |
145 | /* 64 bytes boundary on 64bit, 32 on 32bit */ |
146 | unsigned long last_active; /* L: last active timestamp */ | |
147 | unsigned int flags; /* X: flags */ | |
c34056a3 | 148 | int id; /* I: worker id */ |
25511a47 TH |
149 | |
150 | /* for rebinding worker to CPU */ | |
25511a47 | 151 | struct work_struct rebind_work; /* L: for busy worker */ |
c34056a3 TH |
152 | }; |
153 | ||
bd7bdd43 TH |
154 | struct worker_pool { |
155 | struct global_cwq *gcwq; /* I: the owning gcwq */ | |
11ebea50 | 156 | unsigned int flags; /* X: flags */ |
bd7bdd43 TH |
157 | |
158 | struct list_head worklist; /* L: list of pending works */ | |
159 | int nr_workers; /* L: total number of workers */ | |
ea1abd61 LJ |
160 | |
161 | /* nr_idle includes the ones off idle_list for rebinding */ | |
bd7bdd43 TH |
162 | int nr_idle; /* L: currently idle ones */ |
163 | ||
164 | struct list_head idle_list; /* X: list of idle workers */ | |
165 | struct timer_list idle_timer; /* L: worker idle timeout */ | |
166 | struct timer_list mayday_timer; /* L: SOS timer for workers */ | |
167 | ||
b2eb83d1 | 168 | struct mutex assoc_mutex; /* protect GCWQ_DISASSOCIATED */ |
bd7bdd43 | 169 | struct ida worker_ida; /* L: for worker IDs */ |
bd7bdd43 TH |
170 | }; |
171 | ||
8b03ae3c | 172 | /* |
e22bee78 TH |
173 | * Global per-cpu workqueue. There's one and only one for each cpu |
174 | * and all works are queued and processed here regardless of their | |
175 | * target workqueues. | |
8b03ae3c TH |
176 | */ |
177 | struct global_cwq { | |
178 | spinlock_t lock; /* the gcwq lock */ | |
179 | unsigned int cpu; /* I: the associated cpu */ | |
db7bccf4 | 180 | unsigned int flags; /* L: GCWQ_* flags */ |
c8e55f36 | 181 | |
bd7bdd43 | 182 | /* workers are chained either in busy_hash or pool idle_list */ |
c8e55f36 TH |
183 | struct hlist_head busy_hash[BUSY_WORKER_HASH_SIZE]; |
184 | /* L: hash of busy workers */ | |
185 | ||
330dad5b JK |
186 | struct worker_pool pools[NR_WORKER_POOLS]; |
187 | /* normal and highpri pools */ | |
8b03ae3c TH |
188 | } ____cacheline_aligned_in_smp; |
189 | ||
1da177e4 | 190 | /* |
502ca9d8 | 191 | * The per-CPU workqueue. The lower WORK_STRUCT_FLAG_BITS of |
0f900049 TH |
192 | * work_struct->data are used for flags and thus cwqs need to be |
193 | * aligned at two's power of the number of flag bits. | |
1da177e4 LT |
194 | */ |
195 | struct cpu_workqueue_struct { | |
bd7bdd43 | 196 | struct worker_pool *pool; /* I: the associated pool */ |
4690c4ab | 197 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
198 | int work_color; /* L: current color */ |
199 | int flush_color; /* L: flushing color */ | |
200 | int nr_in_flight[WORK_NR_COLORS]; | |
201 | /* L: nr of in_flight works */ | |
1e19ffc6 | 202 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 203 | int max_active; /* L: max active works */ |
1e19ffc6 | 204 | struct list_head delayed_works; /* L: delayed works */ |
0f900049 | 205 | }; |
1da177e4 | 206 | |
73f53c4a TH |
207 | /* |
208 | * Structure used to wait for workqueue flush. | |
209 | */ | |
210 | struct wq_flusher { | |
211 | struct list_head list; /* F: list of flushers */ | |
212 | int flush_color; /* F: flush color waiting for */ | |
213 | struct completion done; /* flush completion */ | |
214 | }; | |
215 | ||
f2e005aa TH |
216 | /* |
217 | * All cpumasks are assumed to be always set on UP and thus can't be | |
218 | * used to determine whether there's something to be done. | |
219 | */ | |
220 | #ifdef CONFIG_SMP | |
221 | typedef cpumask_var_t mayday_mask_t; | |
222 | #define mayday_test_and_set_cpu(cpu, mask) \ | |
223 | cpumask_test_and_set_cpu((cpu), (mask)) | |
224 | #define mayday_clear_cpu(cpu, mask) cpumask_clear_cpu((cpu), (mask)) | |
225 | #define for_each_mayday_cpu(cpu, mask) for_each_cpu((cpu), (mask)) | |
9c37547a | 226 | #define alloc_mayday_mask(maskp, gfp) zalloc_cpumask_var((maskp), (gfp)) |
f2e005aa TH |
227 | #define free_mayday_mask(mask) free_cpumask_var((mask)) |
228 | #else | |
229 | typedef unsigned long mayday_mask_t; | |
230 | #define mayday_test_and_set_cpu(cpu, mask) test_and_set_bit(0, &(mask)) | |
231 | #define mayday_clear_cpu(cpu, mask) clear_bit(0, &(mask)) | |
232 | #define for_each_mayday_cpu(cpu, mask) if ((cpu) = 0, (mask)) | |
233 | #define alloc_mayday_mask(maskp, gfp) true | |
234 | #define free_mayday_mask(mask) do { } while (0) | |
235 | #endif | |
1da177e4 LT |
236 | |
237 | /* | |
238 | * The externally visible workqueue abstraction is an array of | |
239 | * per-CPU workqueues: | |
240 | */ | |
241 | struct workqueue_struct { | |
9c5a2ba7 | 242 | unsigned int flags; /* W: WQ_* flags */ |
bdbc5dd7 TH |
243 | union { |
244 | struct cpu_workqueue_struct __percpu *pcpu; | |
245 | struct cpu_workqueue_struct *single; | |
246 | unsigned long v; | |
247 | } cpu_wq; /* I: cwq's */ | |
4690c4ab | 248 | struct list_head list; /* W: list of all workqueues */ |
73f53c4a TH |
249 | |
250 | struct mutex flush_mutex; /* protects wq flushing */ | |
251 | int work_color; /* F: current work color */ | |
252 | int flush_color; /* F: current flush color */ | |
253 | atomic_t nr_cwqs_to_flush; /* flush in progress */ | |
254 | struct wq_flusher *first_flusher; /* F: first flusher */ | |
255 | struct list_head flusher_queue; /* F: flush waiters */ | |
256 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
257 | ||
f2e005aa | 258 | mayday_mask_t mayday_mask; /* cpus requesting rescue */ |
e22bee78 TH |
259 | struct worker *rescuer; /* I: rescue worker */ |
260 | ||
9c5a2ba7 | 261 | int nr_drainers; /* W: drain in progress */ |
dcd989cb | 262 | int saved_max_active; /* W: saved cwq max_active */ |
4e6045f1 | 263 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 264 | struct lockdep_map lockdep_map; |
4e6045f1 | 265 | #endif |
b196be89 | 266 | char name[]; /* I: workqueue name */ |
1da177e4 LT |
267 | }; |
268 | ||
d320c038 | 269 | struct workqueue_struct *system_wq __read_mostly; |
d320c038 | 270 | EXPORT_SYMBOL_GPL(system_wq); |
044c782c | 271 | struct workqueue_struct *system_highpri_wq __read_mostly; |
1aabe902 | 272 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
044c782c | 273 | struct workqueue_struct *system_long_wq __read_mostly; |
d320c038 | 274 | EXPORT_SYMBOL_GPL(system_long_wq); |
044c782c | 275 | struct workqueue_struct *system_unbound_wq __read_mostly; |
f3421797 | 276 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
044c782c | 277 | struct workqueue_struct *system_freezable_wq __read_mostly; |
24d51add | 278 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
d320c038 | 279 | |
97bd2347 TH |
280 | #define CREATE_TRACE_POINTS |
281 | #include <trace/events/workqueue.h> | |
282 | ||
4ce62e9e | 283 | #define for_each_worker_pool(pool, gcwq) \ |
3270476a TH |
284 | for ((pool) = &(gcwq)->pools[0]; \ |
285 | (pool) < &(gcwq)->pools[NR_WORKER_POOLS]; (pool)++) | |
4ce62e9e | 286 | |
db7bccf4 TH |
287 | #define for_each_busy_worker(worker, i, pos, gcwq) \ |
288 | for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++) \ | |
289 | hlist_for_each_entry(worker, pos, &gcwq->busy_hash[i], hentry) | |
290 | ||
f3421797 TH |
291 | static inline int __next_gcwq_cpu(int cpu, const struct cpumask *mask, |
292 | unsigned int sw) | |
293 | { | |
294 | if (cpu < nr_cpu_ids) { | |
295 | if (sw & 1) { | |
296 | cpu = cpumask_next(cpu, mask); | |
297 | if (cpu < nr_cpu_ids) | |
298 | return cpu; | |
299 | } | |
300 | if (sw & 2) | |
301 | return WORK_CPU_UNBOUND; | |
302 | } | |
303 | return WORK_CPU_NONE; | |
304 | } | |
305 | ||
306 | static inline int __next_wq_cpu(int cpu, const struct cpumask *mask, | |
307 | struct workqueue_struct *wq) | |
308 | { | |
309 | return __next_gcwq_cpu(cpu, mask, !(wq->flags & WQ_UNBOUND) ? 1 : 2); | |
310 | } | |
311 | ||
09884951 TH |
312 | /* |
313 | * CPU iterators | |
314 | * | |
315 | * An extra gcwq is defined for an invalid cpu number | |
316 | * (WORK_CPU_UNBOUND) to host workqueues which are not bound to any | |
317 | * specific CPU. The following iterators are similar to | |
318 | * for_each_*_cpu() iterators but also considers the unbound gcwq. | |
319 | * | |
320 | * for_each_gcwq_cpu() : possible CPUs + WORK_CPU_UNBOUND | |
321 | * for_each_online_gcwq_cpu() : online CPUs + WORK_CPU_UNBOUND | |
322 | * for_each_cwq_cpu() : possible CPUs for bound workqueues, | |
323 | * WORK_CPU_UNBOUND for unbound workqueues | |
324 | */ | |
f3421797 TH |
325 | #define for_each_gcwq_cpu(cpu) \ |
326 | for ((cpu) = __next_gcwq_cpu(-1, cpu_possible_mask, 3); \ | |
327 | (cpu) < WORK_CPU_NONE; \ | |
328 | (cpu) = __next_gcwq_cpu((cpu), cpu_possible_mask, 3)) | |
329 | ||
330 | #define for_each_online_gcwq_cpu(cpu) \ | |
331 | for ((cpu) = __next_gcwq_cpu(-1, cpu_online_mask, 3); \ | |
332 | (cpu) < WORK_CPU_NONE; \ | |
333 | (cpu) = __next_gcwq_cpu((cpu), cpu_online_mask, 3)) | |
334 | ||
335 | #define for_each_cwq_cpu(cpu, wq) \ | |
336 | for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, (wq)); \ | |
337 | (cpu) < WORK_CPU_NONE; \ | |
338 | (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, (wq))) | |
339 | ||
dc186ad7 TG |
340 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
341 | ||
342 | static struct debug_obj_descr work_debug_descr; | |
343 | ||
99777288 SG |
344 | static void *work_debug_hint(void *addr) |
345 | { | |
346 | return ((struct work_struct *) addr)->func; | |
347 | } | |
348 | ||
dc186ad7 TG |
349 | /* |
350 | * fixup_init is called when: | |
351 | * - an active object is initialized | |
352 | */ | |
353 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
354 | { | |
355 | struct work_struct *work = addr; | |
356 | ||
357 | switch (state) { | |
358 | case ODEBUG_STATE_ACTIVE: | |
359 | cancel_work_sync(work); | |
360 | debug_object_init(work, &work_debug_descr); | |
361 | return 1; | |
362 | default: | |
363 | return 0; | |
364 | } | |
365 | } | |
366 | ||
367 | /* | |
368 | * fixup_activate is called when: | |
369 | * - an active object is activated | |
370 | * - an unknown object is activated (might be a statically initialized object) | |
371 | */ | |
372 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
373 | { | |
374 | struct work_struct *work = addr; | |
375 | ||
376 | switch (state) { | |
377 | ||
378 | case ODEBUG_STATE_NOTAVAILABLE: | |
379 | /* | |
380 | * This is not really a fixup. The work struct was | |
381 | * statically initialized. We just make sure that it | |
382 | * is tracked in the object tracker. | |
383 | */ | |
22df02bb | 384 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
385 | debug_object_init(work, &work_debug_descr); |
386 | debug_object_activate(work, &work_debug_descr); | |
387 | return 0; | |
388 | } | |
389 | WARN_ON_ONCE(1); | |
390 | return 0; | |
391 | ||
392 | case ODEBUG_STATE_ACTIVE: | |
393 | WARN_ON(1); | |
394 | ||
395 | default: | |
396 | return 0; | |
397 | } | |
398 | } | |
399 | ||
400 | /* | |
401 | * fixup_free is called when: | |
402 | * - an active object is freed | |
403 | */ | |
404 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
405 | { | |
406 | struct work_struct *work = addr; | |
407 | ||
408 | switch (state) { | |
409 | case ODEBUG_STATE_ACTIVE: | |
410 | cancel_work_sync(work); | |
411 | debug_object_free(work, &work_debug_descr); | |
412 | return 1; | |
413 | default: | |
414 | return 0; | |
415 | } | |
416 | } | |
417 | ||
418 | static struct debug_obj_descr work_debug_descr = { | |
419 | .name = "work_struct", | |
99777288 | 420 | .debug_hint = work_debug_hint, |
dc186ad7 TG |
421 | .fixup_init = work_fixup_init, |
422 | .fixup_activate = work_fixup_activate, | |
423 | .fixup_free = work_fixup_free, | |
424 | }; | |
425 | ||
426 | static inline void debug_work_activate(struct work_struct *work) | |
427 | { | |
428 | debug_object_activate(work, &work_debug_descr); | |
429 | } | |
430 | ||
431 | static inline void debug_work_deactivate(struct work_struct *work) | |
432 | { | |
433 | debug_object_deactivate(work, &work_debug_descr); | |
434 | } | |
435 | ||
436 | void __init_work(struct work_struct *work, int onstack) | |
437 | { | |
438 | if (onstack) | |
439 | debug_object_init_on_stack(work, &work_debug_descr); | |
440 | else | |
441 | debug_object_init(work, &work_debug_descr); | |
442 | } | |
443 | EXPORT_SYMBOL_GPL(__init_work); | |
444 | ||
445 | void destroy_work_on_stack(struct work_struct *work) | |
446 | { | |
447 | debug_object_free(work, &work_debug_descr); | |
448 | } | |
449 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
450 | ||
451 | #else | |
452 | static inline void debug_work_activate(struct work_struct *work) { } | |
453 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
454 | #endif | |
455 | ||
95402b38 GS |
456 | /* Serializes the accesses to the list of workqueues. */ |
457 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 458 | static LIST_HEAD(workqueues); |
a0a1a5fd | 459 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 460 | |
e22bee78 TH |
461 | /* |
462 | * The almighty global cpu workqueues. nr_running is the only field | |
463 | * which is expected to be used frequently by other cpus via | |
464 | * try_to_wake_up(). Put it in a separate cacheline. | |
465 | */ | |
8b03ae3c | 466 | static DEFINE_PER_CPU(struct global_cwq, global_cwq); |
4ce62e9e | 467 | static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_WORKER_POOLS]); |
8b03ae3c | 468 | |
f3421797 TH |
469 | /* |
470 | * Global cpu workqueue and nr_running counter for unbound gcwq. The | |
471 | * gcwq is always online, has GCWQ_DISASSOCIATED set, and all its | |
472 | * workers have WORKER_UNBOUND set. | |
473 | */ | |
474 | static struct global_cwq unbound_global_cwq; | |
4ce62e9e TH |
475 | static atomic_t unbound_pool_nr_running[NR_WORKER_POOLS] = { |
476 | [0 ... NR_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */ | |
477 | }; | |
f3421797 | 478 | |
c34056a3 | 479 | static int worker_thread(void *__worker); |
1da177e4 | 480 | |
3270476a TH |
481 | static int worker_pool_pri(struct worker_pool *pool) |
482 | { | |
483 | return pool - pool->gcwq->pools; | |
484 | } | |
485 | ||
8b03ae3c TH |
486 | static struct global_cwq *get_gcwq(unsigned int cpu) |
487 | { | |
f3421797 TH |
488 | if (cpu != WORK_CPU_UNBOUND) |
489 | return &per_cpu(global_cwq, cpu); | |
490 | else | |
491 | return &unbound_global_cwq; | |
8b03ae3c TH |
492 | } |
493 | ||
63d95a91 | 494 | static atomic_t *get_pool_nr_running(struct worker_pool *pool) |
e22bee78 | 495 | { |
63d95a91 | 496 | int cpu = pool->gcwq->cpu; |
3270476a | 497 | int idx = worker_pool_pri(pool); |
63d95a91 | 498 | |
f3421797 | 499 | if (cpu != WORK_CPU_UNBOUND) |
4ce62e9e | 500 | return &per_cpu(pool_nr_running, cpu)[idx]; |
f3421797 | 501 | else |
4ce62e9e | 502 | return &unbound_pool_nr_running[idx]; |
e22bee78 TH |
503 | } |
504 | ||
1537663f TH |
505 | static struct cpu_workqueue_struct *get_cwq(unsigned int cpu, |
506 | struct workqueue_struct *wq) | |
b1f4ec17 | 507 | { |
f3421797 | 508 | if (!(wq->flags & WQ_UNBOUND)) { |
e06ffa1e | 509 | if (likely(cpu < nr_cpu_ids)) |
f3421797 | 510 | return per_cpu_ptr(wq->cpu_wq.pcpu, cpu); |
f3421797 TH |
511 | } else if (likely(cpu == WORK_CPU_UNBOUND)) |
512 | return wq->cpu_wq.single; | |
513 | return NULL; | |
b1f4ec17 ON |
514 | } |
515 | ||
73f53c4a TH |
516 | static unsigned int work_color_to_flags(int color) |
517 | { | |
518 | return color << WORK_STRUCT_COLOR_SHIFT; | |
519 | } | |
520 | ||
521 | static int get_work_color(struct work_struct *work) | |
522 | { | |
523 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
524 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
525 | } | |
526 | ||
527 | static int work_next_color(int color) | |
528 | { | |
529 | return (color + 1) % WORK_NR_COLORS; | |
530 | } | |
1da177e4 | 531 | |
14441960 | 532 | /* |
b5490077 TH |
533 | * While queued, %WORK_STRUCT_CWQ is set and non flag bits of a work's data |
534 | * contain the pointer to the queued cwq. Once execution starts, the flag | |
535 | * is cleared and the high bits contain OFFQ flags and CPU number. | |
7a22ad75 | 536 | * |
bbb68dfa TH |
537 | * set_work_cwq(), set_work_cpu_and_clear_pending(), mark_work_canceling() |
538 | * and clear_work_data() can be used to set the cwq, cpu or clear | |
539 | * work->data. These functions should only be called while the work is | |
540 | * owned - ie. while the PENDING bit is set. | |
7a22ad75 | 541 | * |
bbb68dfa TH |
542 | * get_work_[g]cwq() can be used to obtain the gcwq or cwq corresponding to |
543 | * a work. gcwq is available once the work has been queued anywhere after | |
544 | * initialization until it is sync canceled. cwq is available only while | |
545 | * the work item is queued. | |
7a22ad75 | 546 | * |
bbb68dfa TH |
547 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
548 | * canceled. While being canceled, a work item may have its PENDING set | |
549 | * but stay off timer and worklist for arbitrarily long and nobody should | |
550 | * try to steal the PENDING bit. | |
14441960 | 551 | */ |
7a22ad75 TH |
552 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
553 | unsigned long flags) | |
365970a1 | 554 | { |
4594bf15 | 555 | BUG_ON(!work_pending(work)); |
7a22ad75 TH |
556 | atomic_long_set(&work->data, data | flags | work_static(work)); |
557 | } | |
365970a1 | 558 | |
7a22ad75 TH |
559 | static void set_work_cwq(struct work_struct *work, |
560 | struct cpu_workqueue_struct *cwq, | |
561 | unsigned long extra_flags) | |
562 | { | |
563 | set_work_data(work, (unsigned long)cwq, | |
e120153d | 564 | WORK_STRUCT_PENDING | WORK_STRUCT_CWQ | extra_flags); |
365970a1 DH |
565 | } |
566 | ||
8930caba TH |
567 | static void set_work_cpu_and_clear_pending(struct work_struct *work, |
568 | unsigned int cpu) | |
7a22ad75 | 569 | { |
23657bb1 TH |
570 | /* |
571 | * The following wmb is paired with the implied mb in | |
572 | * test_and_set_bit(PENDING) and ensures all updates to @work made | |
573 | * here are visible to and precede any updates by the next PENDING | |
574 | * owner. | |
575 | */ | |
576 | smp_wmb(); | |
b5490077 | 577 | set_work_data(work, (unsigned long)cpu << WORK_OFFQ_CPU_SHIFT, 0); |
7a22ad75 | 578 | } |
f756d5e2 | 579 | |
7a22ad75 | 580 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 581 | { |
23657bb1 | 582 | smp_wmb(); /* see set_work_cpu_and_clear_pending() */ |
7a22ad75 | 583 | set_work_data(work, WORK_STRUCT_NO_CPU, 0); |
1da177e4 LT |
584 | } |
585 | ||
7a22ad75 | 586 | static struct cpu_workqueue_struct *get_work_cwq(struct work_struct *work) |
b1f4ec17 | 587 | { |
e120153d | 588 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 589 | |
e120153d TH |
590 | if (data & WORK_STRUCT_CWQ) |
591 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); | |
592 | else | |
593 | return NULL; | |
4d707b9f ON |
594 | } |
595 | ||
7a22ad75 | 596 | static struct global_cwq *get_work_gcwq(struct work_struct *work) |
365970a1 | 597 | { |
e120153d | 598 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 TH |
599 | unsigned int cpu; |
600 | ||
e120153d TH |
601 | if (data & WORK_STRUCT_CWQ) |
602 | return ((struct cpu_workqueue_struct *) | |
bd7bdd43 | 603 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->gcwq; |
7a22ad75 | 604 | |
b5490077 | 605 | cpu = data >> WORK_OFFQ_CPU_SHIFT; |
bdbc5dd7 | 606 | if (cpu == WORK_CPU_NONE) |
7a22ad75 TH |
607 | return NULL; |
608 | ||
f3421797 | 609 | BUG_ON(cpu >= nr_cpu_ids && cpu != WORK_CPU_UNBOUND); |
7a22ad75 | 610 | return get_gcwq(cpu); |
b1f4ec17 ON |
611 | } |
612 | ||
bbb68dfa TH |
613 | static void mark_work_canceling(struct work_struct *work) |
614 | { | |
615 | struct global_cwq *gcwq = get_work_gcwq(work); | |
616 | unsigned long cpu = gcwq ? gcwq->cpu : WORK_CPU_NONE; | |
617 | ||
618 | set_work_data(work, (cpu << WORK_OFFQ_CPU_SHIFT) | WORK_OFFQ_CANCELING, | |
619 | WORK_STRUCT_PENDING); | |
620 | } | |
621 | ||
622 | static bool work_is_canceling(struct work_struct *work) | |
623 | { | |
624 | unsigned long data = atomic_long_read(&work->data); | |
625 | ||
626 | return !(data & WORK_STRUCT_CWQ) && (data & WORK_OFFQ_CANCELING); | |
627 | } | |
628 | ||
e22bee78 | 629 | /* |
3270476a TH |
630 | * Policy functions. These define the policies on how the global worker |
631 | * pools are managed. Unless noted otherwise, these functions assume that | |
632 | * they're being called with gcwq->lock held. | |
e22bee78 TH |
633 | */ |
634 | ||
63d95a91 | 635 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 636 | { |
3270476a | 637 | return !atomic_read(get_pool_nr_running(pool)); |
a848e3b6 ON |
638 | } |
639 | ||
4594bf15 | 640 | /* |
e22bee78 TH |
641 | * Need to wake up a worker? Called from anything but currently |
642 | * running workers. | |
974271c4 TH |
643 | * |
644 | * Note that, because unbound workers never contribute to nr_running, this | |
645 | * function will always return %true for unbound gcwq as long as the | |
646 | * worklist isn't empty. | |
4594bf15 | 647 | */ |
63d95a91 | 648 | static bool need_more_worker(struct worker_pool *pool) |
365970a1 | 649 | { |
63d95a91 | 650 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
e22bee78 | 651 | } |
4594bf15 | 652 | |
e22bee78 | 653 | /* Can I start working? Called from busy but !running workers. */ |
63d95a91 | 654 | static bool may_start_working(struct worker_pool *pool) |
e22bee78 | 655 | { |
63d95a91 | 656 | return pool->nr_idle; |
e22bee78 TH |
657 | } |
658 | ||
659 | /* Do I need to keep working? Called from currently running workers. */ | |
63d95a91 | 660 | static bool keep_working(struct worker_pool *pool) |
e22bee78 | 661 | { |
63d95a91 | 662 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 | 663 | |
3270476a | 664 | return !list_empty(&pool->worklist) && atomic_read(nr_running) <= 1; |
e22bee78 TH |
665 | } |
666 | ||
667 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 668 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 669 | { |
63d95a91 | 670 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 671 | } |
365970a1 | 672 | |
e22bee78 | 673 | /* Do I need to be the manager? */ |
63d95a91 | 674 | static bool need_to_manage_workers(struct worker_pool *pool) |
e22bee78 | 675 | { |
63d95a91 | 676 | return need_to_create_worker(pool) || |
11ebea50 | 677 | (pool->flags & POOL_MANAGE_WORKERS); |
e22bee78 TH |
678 | } |
679 | ||
680 | /* Do we have too many workers and should some go away? */ | |
63d95a91 | 681 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 682 | { |
552a37e9 | 683 | bool managing = pool->flags & POOL_MANAGING_WORKERS; |
63d95a91 TH |
684 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
685 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 | 686 | |
ea1abd61 LJ |
687 | /* |
688 | * nr_idle and idle_list may disagree if idle rebinding is in | |
689 | * progress. Never return %true if idle_list is empty. | |
690 | */ | |
691 | if (list_empty(&pool->idle_list)) | |
692 | return false; | |
693 | ||
e22bee78 | 694 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
365970a1 DH |
695 | } |
696 | ||
4d707b9f | 697 | /* |
e22bee78 TH |
698 | * Wake up functions. |
699 | */ | |
700 | ||
7e11629d | 701 | /* Return the first worker. Safe with preemption disabled */ |
63d95a91 | 702 | static struct worker *first_worker(struct worker_pool *pool) |
7e11629d | 703 | { |
63d95a91 | 704 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
705 | return NULL; |
706 | ||
63d95a91 | 707 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
708 | } |
709 | ||
710 | /** | |
711 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 712 | * @pool: worker pool to wake worker from |
7e11629d | 713 | * |
63d95a91 | 714 | * Wake up the first idle worker of @pool. |
7e11629d TH |
715 | * |
716 | * CONTEXT: | |
717 | * spin_lock_irq(gcwq->lock). | |
718 | */ | |
63d95a91 | 719 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 720 | { |
63d95a91 | 721 | struct worker *worker = first_worker(pool); |
7e11629d TH |
722 | |
723 | if (likely(worker)) | |
724 | wake_up_process(worker->task); | |
725 | } | |
726 | ||
d302f017 | 727 | /** |
e22bee78 TH |
728 | * wq_worker_waking_up - a worker is waking up |
729 | * @task: task waking up | |
730 | * @cpu: CPU @task is waking up to | |
731 | * | |
732 | * This function is called during try_to_wake_up() when a worker is | |
733 | * being awoken. | |
734 | * | |
735 | * CONTEXT: | |
736 | * spin_lock_irq(rq->lock) | |
737 | */ | |
738 | void wq_worker_waking_up(struct task_struct *task, unsigned int cpu) | |
739 | { | |
740 | struct worker *worker = kthread_data(task); | |
741 | ||
2d64672e | 742 | if (!(worker->flags & WORKER_NOT_RUNNING)) |
63d95a91 | 743 | atomic_inc(get_pool_nr_running(worker->pool)); |
e22bee78 TH |
744 | } |
745 | ||
746 | /** | |
747 | * wq_worker_sleeping - a worker is going to sleep | |
748 | * @task: task going to sleep | |
749 | * @cpu: CPU in question, must be the current CPU number | |
750 | * | |
751 | * This function is called during schedule() when a busy worker is | |
752 | * going to sleep. Worker on the same cpu can be woken up by | |
753 | * returning pointer to its task. | |
754 | * | |
755 | * CONTEXT: | |
756 | * spin_lock_irq(rq->lock) | |
757 | * | |
758 | * RETURNS: | |
759 | * Worker task on @cpu to wake up, %NULL if none. | |
760 | */ | |
761 | struct task_struct *wq_worker_sleeping(struct task_struct *task, | |
762 | unsigned int cpu) | |
763 | { | |
764 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
bd7bdd43 | 765 | struct worker_pool *pool = worker->pool; |
63d95a91 | 766 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 | 767 | |
2d64672e | 768 | if (worker->flags & WORKER_NOT_RUNNING) |
e22bee78 TH |
769 | return NULL; |
770 | ||
771 | /* this can only happen on the local cpu */ | |
772 | BUG_ON(cpu != raw_smp_processor_id()); | |
773 | ||
774 | /* | |
775 | * The counterpart of the following dec_and_test, implied mb, | |
776 | * worklist not empty test sequence is in insert_work(). | |
777 | * Please read comment there. | |
778 | * | |
628c78e7 TH |
779 | * NOT_RUNNING is clear. This means that we're bound to and |
780 | * running on the local cpu w/ rq lock held and preemption | |
781 | * disabled, which in turn means that none else could be | |
782 | * manipulating idle_list, so dereferencing idle_list without gcwq | |
783 | * lock is safe. | |
e22bee78 | 784 | */ |
bd7bdd43 | 785 | if (atomic_dec_and_test(nr_running) && !list_empty(&pool->worklist)) |
63d95a91 | 786 | to_wakeup = first_worker(pool); |
e22bee78 TH |
787 | return to_wakeup ? to_wakeup->task : NULL; |
788 | } | |
789 | ||
790 | /** | |
791 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 792 | * @worker: self |
d302f017 TH |
793 | * @flags: flags to set |
794 | * @wakeup: wakeup an idle worker if necessary | |
795 | * | |
e22bee78 TH |
796 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
797 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
798 | * woken up. | |
d302f017 | 799 | * |
cb444766 TH |
800 | * CONTEXT: |
801 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
802 | */ |
803 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
804 | bool wakeup) | |
805 | { | |
bd7bdd43 | 806 | struct worker_pool *pool = worker->pool; |
e22bee78 | 807 | |
cb444766 TH |
808 | WARN_ON_ONCE(worker->task != current); |
809 | ||
e22bee78 TH |
810 | /* |
811 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
812 | * wake up an idle worker as necessary if requested by | |
813 | * @wakeup. | |
814 | */ | |
815 | if ((flags & WORKER_NOT_RUNNING) && | |
816 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
63d95a91 | 817 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 TH |
818 | |
819 | if (wakeup) { | |
820 | if (atomic_dec_and_test(nr_running) && | |
bd7bdd43 | 821 | !list_empty(&pool->worklist)) |
63d95a91 | 822 | wake_up_worker(pool); |
e22bee78 TH |
823 | } else |
824 | atomic_dec(nr_running); | |
825 | } | |
826 | ||
d302f017 TH |
827 | worker->flags |= flags; |
828 | } | |
829 | ||
830 | /** | |
e22bee78 | 831 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 832 | * @worker: self |
d302f017 TH |
833 | * @flags: flags to clear |
834 | * | |
e22bee78 | 835 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 836 | * |
cb444766 TH |
837 | * CONTEXT: |
838 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
839 | */ |
840 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
841 | { | |
63d95a91 | 842 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
843 | unsigned int oflags = worker->flags; |
844 | ||
cb444766 TH |
845 | WARN_ON_ONCE(worker->task != current); |
846 | ||
d302f017 | 847 | worker->flags &= ~flags; |
e22bee78 | 848 | |
42c025f3 TH |
849 | /* |
850 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
851 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
852 | * of multiple flags, not a single flag. | |
853 | */ | |
e22bee78 TH |
854 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
855 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
63d95a91 | 856 | atomic_inc(get_pool_nr_running(pool)); |
d302f017 TH |
857 | } |
858 | ||
c8e55f36 TH |
859 | /** |
860 | * busy_worker_head - return the busy hash head for a work | |
861 | * @gcwq: gcwq of interest | |
862 | * @work: work to be hashed | |
863 | * | |
864 | * Return hash head of @gcwq for @work. | |
865 | * | |
866 | * CONTEXT: | |
867 | * spin_lock_irq(gcwq->lock). | |
868 | * | |
869 | * RETURNS: | |
870 | * Pointer to the hash head. | |
871 | */ | |
872 | static struct hlist_head *busy_worker_head(struct global_cwq *gcwq, | |
873 | struct work_struct *work) | |
874 | { | |
875 | const int base_shift = ilog2(sizeof(struct work_struct)); | |
876 | unsigned long v = (unsigned long)work; | |
877 | ||
878 | /* simple shift and fold hash, do we need something better? */ | |
879 | v >>= base_shift; | |
880 | v += v >> BUSY_WORKER_HASH_ORDER; | |
881 | v &= BUSY_WORKER_HASH_MASK; | |
882 | ||
883 | return &gcwq->busy_hash[v]; | |
884 | } | |
885 | ||
8cca0eea TH |
886 | /** |
887 | * __find_worker_executing_work - find worker which is executing a work | |
888 | * @gcwq: gcwq of interest | |
889 | * @bwh: hash head as returned by busy_worker_head() | |
890 | * @work: work to find worker for | |
891 | * | |
892 | * Find a worker which is executing @work on @gcwq. @bwh should be | |
893 | * the hash head obtained by calling busy_worker_head() with the same | |
894 | * work. | |
895 | * | |
896 | * CONTEXT: | |
897 | * spin_lock_irq(gcwq->lock). | |
898 | * | |
899 | * RETURNS: | |
900 | * Pointer to worker which is executing @work if found, NULL | |
901 | * otherwise. | |
902 | */ | |
903 | static struct worker *__find_worker_executing_work(struct global_cwq *gcwq, | |
904 | struct hlist_head *bwh, | |
905 | struct work_struct *work) | |
906 | { | |
907 | struct worker *worker; | |
908 | struct hlist_node *tmp; | |
909 | ||
910 | hlist_for_each_entry(worker, tmp, bwh, hentry) | |
911 | if (worker->current_work == work) | |
912 | return worker; | |
913 | return NULL; | |
914 | } | |
915 | ||
916 | /** | |
917 | * find_worker_executing_work - find worker which is executing a work | |
918 | * @gcwq: gcwq of interest | |
919 | * @work: work to find worker for | |
920 | * | |
921 | * Find a worker which is executing @work on @gcwq. This function is | |
922 | * identical to __find_worker_executing_work() except that this | |
923 | * function calculates @bwh itself. | |
924 | * | |
925 | * CONTEXT: | |
926 | * spin_lock_irq(gcwq->lock). | |
927 | * | |
928 | * RETURNS: | |
929 | * Pointer to worker which is executing @work if found, NULL | |
930 | * otherwise. | |
4d707b9f | 931 | */ |
8cca0eea TH |
932 | static struct worker *find_worker_executing_work(struct global_cwq *gcwq, |
933 | struct work_struct *work) | |
4d707b9f | 934 | { |
8cca0eea TH |
935 | return __find_worker_executing_work(gcwq, busy_worker_head(gcwq, work), |
936 | work); | |
4d707b9f ON |
937 | } |
938 | ||
bf4ede01 TH |
939 | /** |
940 | * move_linked_works - move linked works to a list | |
941 | * @work: start of series of works to be scheduled | |
942 | * @head: target list to append @work to | |
943 | * @nextp: out paramter for nested worklist walking | |
944 | * | |
945 | * Schedule linked works starting from @work to @head. Work series to | |
946 | * be scheduled starts at @work and includes any consecutive work with | |
947 | * WORK_STRUCT_LINKED set in its predecessor. | |
948 | * | |
949 | * If @nextp is not NULL, it's updated to point to the next work of | |
950 | * the last scheduled work. This allows move_linked_works() to be | |
951 | * nested inside outer list_for_each_entry_safe(). | |
952 | * | |
953 | * CONTEXT: | |
954 | * spin_lock_irq(gcwq->lock). | |
955 | */ | |
956 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
957 | struct work_struct **nextp) | |
958 | { | |
959 | struct work_struct *n; | |
960 | ||
961 | /* | |
962 | * Linked worklist will always end before the end of the list, | |
963 | * use NULL for list head. | |
964 | */ | |
965 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
966 | list_move_tail(&work->entry, head); | |
967 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
968 | break; | |
969 | } | |
970 | ||
971 | /* | |
972 | * If we're already inside safe list traversal and have moved | |
973 | * multiple works to the scheduled queue, the next position | |
974 | * needs to be updated. | |
975 | */ | |
976 | if (nextp) | |
977 | *nextp = n; | |
978 | } | |
979 | ||
3aa62497 | 980 | static void cwq_activate_delayed_work(struct work_struct *work) |
bf4ede01 | 981 | { |
3aa62497 | 982 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); |
bf4ede01 TH |
983 | |
984 | trace_workqueue_activate_work(work); | |
985 | move_linked_works(work, &cwq->pool->worklist, NULL); | |
986 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); | |
987 | cwq->nr_active++; | |
988 | } | |
989 | ||
3aa62497 LJ |
990 | static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq) |
991 | { | |
992 | struct work_struct *work = list_first_entry(&cwq->delayed_works, | |
993 | struct work_struct, entry); | |
994 | ||
995 | cwq_activate_delayed_work(work); | |
996 | } | |
997 | ||
bf4ede01 TH |
998 | /** |
999 | * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight | |
1000 | * @cwq: cwq of interest | |
1001 | * @color: color of work which left the queue | |
bf4ede01 TH |
1002 | * |
1003 | * A work either has completed or is removed from pending queue, | |
1004 | * decrement nr_in_flight of its cwq and handle workqueue flushing. | |
1005 | * | |
1006 | * CONTEXT: | |
1007 | * spin_lock_irq(gcwq->lock). | |
1008 | */ | |
b3f9f405 | 1009 | static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color) |
bf4ede01 TH |
1010 | { |
1011 | /* ignore uncolored works */ | |
1012 | if (color == WORK_NO_COLOR) | |
1013 | return; | |
1014 | ||
1015 | cwq->nr_in_flight[color]--; | |
1016 | ||
b3f9f405 LJ |
1017 | cwq->nr_active--; |
1018 | if (!list_empty(&cwq->delayed_works)) { | |
1019 | /* one down, submit a delayed one */ | |
1020 | if (cwq->nr_active < cwq->max_active) | |
1021 | cwq_activate_first_delayed(cwq); | |
bf4ede01 TH |
1022 | } |
1023 | ||
1024 | /* is flush in progress and are we at the flushing tip? */ | |
1025 | if (likely(cwq->flush_color != color)) | |
1026 | return; | |
1027 | ||
1028 | /* are there still in-flight works? */ | |
1029 | if (cwq->nr_in_flight[color]) | |
1030 | return; | |
1031 | ||
1032 | /* this cwq is done, clear flush_color */ | |
1033 | cwq->flush_color = -1; | |
1034 | ||
1035 | /* | |
1036 | * If this was the last cwq, wake up the first flusher. It | |
1037 | * will handle the rest. | |
1038 | */ | |
1039 | if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush)) | |
1040 | complete(&cwq->wq->first_flusher->done); | |
1041 | } | |
1042 | ||
36e227d2 | 1043 | /** |
bbb68dfa | 1044 | * try_to_grab_pending - steal work item from worklist and disable irq |
36e227d2 TH |
1045 | * @work: work item to steal |
1046 | * @is_dwork: @work is a delayed_work | |
bbb68dfa | 1047 | * @flags: place to store irq state |
36e227d2 TH |
1048 | * |
1049 | * Try to grab PENDING bit of @work. This function can handle @work in any | |
1050 | * stable state - idle, on timer or on worklist. Return values are | |
1051 | * | |
1052 | * 1 if @work was pending and we successfully stole PENDING | |
1053 | * 0 if @work was idle and we claimed PENDING | |
1054 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry | |
bbb68dfa TH |
1055 | * -ENOENT if someone else is canceling @work, this state may persist |
1056 | * for arbitrarily long | |
36e227d2 | 1057 | * |
bbb68dfa | 1058 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
e0aecdd8 TH |
1059 | * interrupted while holding PENDING and @work off queue, irq must be |
1060 | * disabled on entry. This, combined with delayed_work->timer being | |
1061 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. | |
bbb68dfa TH |
1062 | * |
1063 | * On successful return, >= 0, irq is disabled and the caller is | |
1064 | * responsible for releasing it using local_irq_restore(*@flags). | |
1065 | * | |
e0aecdd8 | 1066 | * This function is safe to call from any context including IRQ handler. |
bf4ede01 | 1067 | */ |
bbb68dfa TH |
1068 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
1069 | unsigned long *flags) | |
bf4ede01 TH |
1070 | { |
1071 | struct global_cwq *gcwq; | |
bf4ede01 | 1072 | |
bbb68dfa TH |
1073 | local_irq_save(*flags); |
1074 | ||
36e227d2 TH |
1075 | /* try to steal the timer if it exists */ |
1076 | if (is_dwork) { | |
1077 | struct delayed_work *dwork = to_delayed_work(work); | |
1078 | ||
e0aecdd8 TH |
1079 | /* |
1080 | * dwork->timer is irqsafe. If del_timer() fails, it's | |
1081 | * guaranteed that the timer is not queued anywhere and not | |
1082 | * running on the local CPU. | |
1083 | */ | |
36e227d2 TH |
1084 | if (likely(del_timer(&dwork->timer))) |
1085 | return 1; | |
1086 | } | |
1087 | ||
1088 | /* try to claim PENDING the normal way */ | |
bf4ede01 TH |
1089 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1090 | return 0; | |
1091 | ||
1092 | /* | |
1093 | * The queueing is in progress, or it is already queued. Try to | |
1094 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1095 | */ | |
1096 | gcwq = get_work_gcwq(work); | |
1097 | if (!gcwq) | |
bbb68dfa | 1098 | goto fail; |
bf4ede01 | 1099 | |
bbb68dfa | 1100 | spin_lock(&gcwq->lock); |
bf4ede01 TH |
1101 | if (!list_empty(&work->entry)) { |
1102 | /* | |
1103 | * This work is queued, but perhaps we locked the wrong gcwq. | |
1104 | * In that case we must see the new value after rmb(), see | |
1105 | * insert_work()->wmb(). | |
1106 | */ | |
1107 | smp_rmb(); | |
1108 | if (gcwq == get_work_gcwq(work)) { | |
1109 | debug_work_deactivate(work); | |
3aa62497 LJ |
1110 | |
1111 | /* | |
1112 | * A delayed work item cannot be grabbed directly | |
1113 | * because it might have linked NO_COLOR work items | |
1114 | * which, if left on the delayed_list, will confuse | |
1115 | * cwq->nr_active management later on and cause | |
1116 | * stall. Make sure the work item is activated | |
1117 | * before grabbing. | |
1118 | */ | |
1119 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) | |
1120 | cwq_activate_delayed_work(work); | |
1121 | ||
bf4ede01 TH |
1122 | list_del_init(&work->entry); |
1123 | cwq_dec_nr_in_flight(get_work_cwq(work), | |
b3f9f405 | 1124 | get_work_color(work)); |
36e227d2 | 1125 | |
bbb68dfa | 1126 | spin_unlock(&gcwq->lock); |
36e227d2 | 1127 | return 1; |
bf4ede01 TH |
1128 | } |
1129 | } | |
bbb68dfa TH |
1130 | spin_unlock(&gcwq->lock); |
1131 | fail: | |
1132 | local_irq_restore(*flags); | |
1133 | if (work_is_canceling(work)) | |
1134 | return -ENOENT; | |
1135 | cpu_relax(); | |
36e227d2 | 1136 | return -EAGAIN; |
bf4ede01 TH |
1137 | } |
1138 | ||
4690c4ab | 1139 | /** |
7e11629d | 1140 | * insert_work - insert a work into gcwq |
4690c4ab TH |
1141 | * @cwq: cwq @work belongs to |
1142 | * @work: work to insert | |
1143 | * @head: insertion point | |
1144 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
1145 | * | |
7e11629d TH |
1146 | * Insert @work which belongs to @cwq into @gcwq after @head. |
1147 | * @extra_flags is or'd to work_struct flags. | |
4690c4ab TH |
1148 | * |
1149 | * CONTEXT: | |
8b03ae3c | 1150 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 1151 | */ |
b89deed3 | 1152 | static void insert_work(struct cpu_workqueue_struct *cwq, |
4690c4ab TH |
1153 | struct work_struct *work, struct list_head *head, |
1154 | unsigned int extra_flags) | |
b89deed3 | 1155 | { |
63d95a91 | 1156 | struct worker_pool *pool = cwq->pool; |
e22bee78 | 1157 | |
4690c4ab | 1158 | /* we own @work, set data and link */ |
7a22ad75 | 1159 | set_work_cwq(work, cwq, extra_flags); |
e1d8aa9f | 1160 | |
6e84d644 ON |
1161 | /* |
1162 | * Ensure that we get the right work->data if we see the | |
1163 | * result of list_add() below, see try_to_grab_pending(). | |
1164 | */ | |
1165 | smp_wmb(); | |
4690c4ab | 1166 | |
1a4d9b0a | 1167 | list_add_tail(&work->entry, head); |
e22bee78 TH |
1168 | |
1169 | /* | |
1170 | * Ensure either worker_sched_deactivated() sees the above | |
1171 | * list_add_tail() or we see zero nr_running to avoid workers | |
1172 | * lying around lazily while there are works to be processed. | |
1173 | */ | |
1174 | smp_mb(); | |
1175 | ||
63d95a91 TH |
1176 | if (__need_more_worker(pool)) |
1177 | wake_up_worker(pool); | |
b89deed3 ON |
1178 | } |
1179 | ||
c8efcc25 TH |
1180 | /* |
1181 | * Test whether @work is being queued from another work executing on the | |
1182 | * same workqueue. This is rather expensive and should only be used from | |
1183 | * cold paths. | |
1184 | */ | |
1185 | static bool is_chained_work(struct workqueue_struct *wq) | |
1186 | { | |
1187 | unsigned long flags; | |
1188 | unsigned int cpu; | |
1189 | ||
1190 | for_each_gcwq_cpu(cpu) { | |
1191 | struct global_cwq *gcwq = get_gcwq(cpu); | |
1192 | struct worker *worker; | |
1193 | struct hlist_node *pos; | |
1194 | int i; | |
1195 | ||
1196 | spin_lock_irqsave(&gcwq->lock, flags); | |
1197 | for_each_busy_worker(worker, i, pos, gcwq) { | |
1198 | if (worker->task != current) | |
1199 | continue; | |
1200 | spin_unlock_irqrestore(&gcwq->lock, flags); | |
1201 | /* | |
1202 | * I'm @worker, no locking necessary. See if @work | |
1203 | * is headed to the same workqueue. | |
1204 | */ | |
1205 | return worker->current_cwq->wq == wq; | |
1206 | } | |
1207 | spin_unlock_irqrestore(&gcwq->lock, flags); | |
1208 | } | |
1209 | return false; | |
1210 | } | |
1211 | ||
4690c4ab | 1212 | static void __queue_work(unsigned int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
1213 | struct work_struct *work) |
1214 | { | |
502ca9d8 TH |
1215 | struct global_cwq *gcwq; |
1216 | struct cpu_workqueue_struct *cwq; | |
1e19ffc6 | 1217 | struct list_head *worklist; |
8a2e8e5d | 1218 | unsigned int work_flags; |
b75cac93 | 1219 | unsigned int req_cpu = cpu; |
8930caba TH |
1220 | |
1221 | /* | |
1222 | * While a work item is PENDING && off queue, a task trying to | |
1223 | * steal the PENDING will busy-loop waiting for it to either get | |
1224 | * queued or lose PENDING. Grabbing PENDING and queueing should | |
1225 | * happen with IRQ disabled. | |
1226 | */ | |
1227 | WARN_ON_ONCE(!irqs_disabled()); | |
1da177e4 | 1228 | |
dc186ad7 | 1229 | debug_work_activate(work); |
1e19ffc6 | 1230 | |
c8efcc25 | 1231 | /* if dying, only works from the same workqueue are allowed */ |
9c5a2ba7 | 1232 | if (unlikely(wq->flags & WQ_DRAINING) && |
c8efcc25 | 1233 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b TH |
1234 | return; |
1235 | ||
c7fc77f7 TH |
1236 | /* determine gcwq to use */ |
1237 | if (!(wq->flags & WQ_UNBOUND)) { | |
18aa9eff TH |
1238 | struct global_cwq *last_gcwq; |
1239 | ||
57469821 | 1240 | if (cpu == WORK_CPU_UNBOUND) |
c7fc77f7 TH |
1241 | cpu = raw_smp_processor_id(); |
1242 | ||
18aa9eff | 1243 | /* |
dbf2576e TH |
1244 | * It's multi cpu. If @work was previously on a different |
1245 | * cpu, it might still be running there, in which case the | |
1246 | * work needs to be queued on that cpu to guarantee | |
1247 | * non-reentrancy. | |
18aa9eff | 1248 | */ |
502ca9d8 | 1249 | gcwq = get_gcwq(cpu); |
dbf2576e TH |
1250 | last_gcwq = get_work_gcwq(work); |
1251 | ||
1252 | if (last_gcwq && last_gcwq != gcwq) { | |
18aa9eff TH |
1253 | struct worker *worker; |
1254 | ||
8930caba | 1255 | spin_lock(&last_gcwq->lock); |
18aa9eff TH |
1256 | |
1257 | worker = find_worker_executing_work(last_gcwq, work); | |
1258 | ||
1259 | if (worker && worker->current_cwq->wq == wq) | |
1260 | gcwq = last_gcwq; | |
1261 | else { | |
1262 | /* meh... not running there, queue here */ | |
8930caba TH |
1263 | spin_unlock(&last_gcwq->lock); |
1264 | spin_lock(&gcwq->lock); | |
18aa9eff | 1265 | } |
8930caba TH |
1266 | } else { |
1267 | spin_lock(&gcwq->lock); | |
1268 | } | |
f3421797 TH |
1269 | } else { |
1270 | gcwq = get_gcwq(WORK_CPU_UNBOUND); | |
8930caba | 1271 | spin_lock(&gcwq->lock); |
502ca9d8 TH |
1272 | } |
1273 | ||
1274 | /* gcwq determined, get cwq and queue */ | |
1275 | cwq = get_cwq(gcwq->cpu, wq); | |
b75cac93 | 1276 | trace_workqueue_queue_work(req_cpu, cwq, work); |
502ca9d8 | 1277 | |
f5b2552b | 1278 | if (WARN_ON(!list_empty(&work->entry))) { |
8930caba | 1279 | spin_unlock(&gcwq->lock); |
f5b2552b DC |
1280 | return; |
1281 | } | |
1e19ffc6 | 1282 | |
73f53c4a | 1283 | cwq->nr_in_flight[cwq->work_color]++; |
8a2e8e5d | 1284 | work_flags = work_color_to_flags(cwq->work_color); |
1e19ffc6 TH |
1285 | |
1286 | if (likely(cwq->nr_active < cwq->max_active)) { | |
cdadf009 | 1287 | trace_workqueue_activate_work(work); |
1e19ffc6 | 1288 | cwq->nr_active++; |
3270476a | 1289 | worklist = &cwq->pool->worklist; |
8a2e8e5d TH |
1290 | } else { |
1291 | work_flags |= WORK_STRUCT_DELAYED; | |
1e19ffc6 | 1292 | worklist = &cwq->delayed_works; |
8a2e8e5d | 1293 | } |
1e19ffc6 | 1294 | |
8a2e8e5d | 1295 | insert_work(cwq, work, worklist, work_flags); |
1e19ffc6 | 1296 | |
8930caba | 1297 | spin_unlock(&gcwq->lock); |
1da177e4 LT |
1298 | } |
1299 | ||
0fcb78c2 | 1300 | /** |
c1a220e7 ZR |
1301 | * queue_work_on - queue work on specific cpu |
1302 | * @cpu: CPU number to execute work on | |
0fcb78c2 REB |
1303 | * @wq: workqueue to use |
1304 | * @work: work to queue | |
1305 | * | |
d4283e93 | 1306 | * Returns %false if @work was already on a queue, %true otherwise. |
1da177e4 | 1307 | * |
c1a220e7 ZR |
1308 | * We queue the work to a specific CPU, the caller must ensure it |
1309 | * can't go away. | |
1da177e4 | 1310 | */ |
d4283e93 TH |
1311 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
1312 | struct work_struct *work) | |
1da177e4 | 1313 | { |
d4283e93 | 1314 | bool ret = false; |
8930caba | 1315 | unsigned long flags; |
ef1ca236 | 1316 | |
8930caba | 1317 | local_irq_save(flags); |
c1a220e7 | 1318 | |
22df02bb | 1319 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1320 | __queue_work(cpu, wq, work); |
d4283e93 | 1321 | ret = true; |
c1a220e7 | 1322 | } |
ef1ca236 | 1323 | |
8930caba | 1324 | local_irq_restore(flags); |
1da177e4 LT |
1325 | return ret; |
1326 | } | |
c1a220e7 | 1327 | EXPORT_SYMBOL_GPL(queue_work_on); |
1da177e4 | 1328 | |
c1a220e7 | 1329 | /** |
0a13c00e | 1330 | * queue_work - queue work on a workqueue |
c1a220e7 ZR |
1331 | * @wq: workqueue to use |
1332 | * @work: work to queue | |
1333 | * | |
d4283e93 | 1334 | * Returns %false if @work was already on a queue, %true otherwise. |
c1a220e7 | 1335 | * |
0a13c00e TH |
1336 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
1337 | * it can be processed by another CPU. | |
c1a220e7 | 1338 | */ |
d4283e93 | 1339 | bool queue_work(struct workqueue_struct *wq, struct work_struct *work) |
c1a220e7 | 1340 | { |
57469821 | 1341 | return queue_work_on(WORK_CPU_UNBOUND, wq, work); |
c1a220e7 | 1342 | } |
0a13c00e | 1343 | EXPORT_SYMBOL_GPL(queue_work); |
c1a220e7 | 1344 | |
d8e794df | 1345 | void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1346 | { |
52bad64d | 1347 | struct delayed_work *dwork = (struct delayed_work *)__data; |
7a22ad75 | 1348 | struct cpu_workqueue_struct *cwq = get_work_cwq(&dwork->work); |
1da177e4 | 1349 | |
e0aecdd8 | 1350 | /* should have been called from irqsafe timer with irq already off */ |
1265057f | 1351 | __queue_work(dwork->cpu, cwq->wq, &dwork->work); |
1da177e4 | 1352 | } |
d8e794df | 1353 | EXPORT_SYMBOL_GPL(delayed_work_timer_fn); |
1da177e4 | 1354 | |
7beb2edf TH |
1355 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
1356 | struct delayed_work *dwork, unsigned long delay) | |
1da177e4 | 1357 | { |
7beb2edf TH |
1358 | struct timer_list *timer = &dwork->timer; |
1359 | struct work_struct *work = &dwork->work; | |
1360 | unsigned int lcpu; | |
1361 | ||
1362 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || | |
1363 | timer->data != (unsigned long)dwork); | |
1364 | BUG_ON(timer_pending(timer)); | |
1365 | BUG_ON(!list_empty(&work->entry)); | |
1366 | ||
8852aac2 TH |
1367 | /* |
1368 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1369 | * both optimization and correctness. The earliest @timer can | |
1370 | * expire is on the closest next tick and delayed_work users depend | |
1371 | * on that there's no such delay when @delay is 0. | |
1372 | */ | |
1373 | if (!delay) { | |
1374 | __queue_work(cpu, wq, &dwork->work); | |
1375 | return; | |
1376 | } | |
1377 | ||
7beb2edf | 1378 | timer_stats_timer_set_start_info(&dwork->timer); |
1da177e4 | 1379 | |
7beb2edf TH |
1380 | /* |
1381 | * This stores cwq for the moment, for the timer_fn. Note that the | |
1382 | * work's gcwq is preserved to allow reentrance detection for | |
1383 | * delayed works. | |
1384 | */ | |
1385 | if (!(wq->flags & WQ_UNBOUND)) { | |
1386 | struct global_cwq *gcwq = get_work_gcwq(work); | |
1387 | ||
e42986de JK |
1388 | /* |
1389 | * If we cannot get the last gcwq from @work directly, | |
1390 | * select the last CPU such that it avoids unnecessarily | |
1391 | * triggering non-reentrancy check in __queue_work(). | |
1392 | */ | |
1393 | lcpu = cpu; | |
1394 | if (gcwq) | |
7beb2edf | 1395 | lcpu = gcwq->cpu; |
e42986de | 1396 | if (lcpu == WORK_CPU_UNBOUND) |
7beb2edf TH |
1397 | lcpu = raw_smp_processor_id(); |
1398 | } else { | |
1399 | lcpu = WORK_CPU_UNBOUND; | |
1400 | } | |
1401 | ||
1402 | set_work_cwq(work, get_cwq(lcpu, wq), 0); | |
1403 | ||
1265057f | 1404 | dwork->cpu = cpu; |
7beb2edf TH |
1405 | timer->expires = jiffies + delay; |
1406 | ||
1407 | if (unlikely(cpu != WORK_CPU_UNBOUND)) | |
1408 | add_timer_on(timer, cpu); | |
1409 | else | |
1410 | add_timer(timer); | |
1da177e4 LT |
1411 | } |
1412 | ||
0fcb78c2 REB |
1413 | /** |
1414 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1415 | * @cpu: CPU number to execute work on | |
1416 | * @wq: workqueue to use | |
af9997e4 | 1417 | * @dwork: work to queue |
0fcb78c2 REB |
1418 | * @delay: number of jiffies to wait before queueing |
1419 | * | |
715f1300 TH |
1420 | * Returns %false if @work was already on a queue, %true otherwise. If |
1421 | * @delay is zero and @dwork is idle, it will be scheduled for immediate | |
1422 | * execution. | |
0fcb78c2 | 1423 | */ |
d4283e93 TH |
1424 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
1425 | struct delayed_work *dwork, unsigned long delay) | |
7a6bc1cd | 1426 | { |
52bad64d | 1427 | struct work_struct *work = &dwork->work; |
d4283e93 | 1428 | bool ret = false; |
8930caba | 1429 | unsigned long flags; |
7a6bc1cd | 1430 | |
8930caba TH |
1431 | /* read the comment in __queue_work() */ |
1432 | local_irq_save(flags); | |
7a6bc1cd | 1433 | |
22df02bb | 1434 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7beb2edf | 1435 | __queue_delayed_work(cpu, wq, dwork, delay); |
d4283e93 | 1436 | ret = true; |
7a6bc1cd | 1437 | } |
8a3e77cc | 1438 | |
8930caba | 1439 | local_irq_restore(flags); |
7a6bc1cd VP |
1440 | return ret; |
1441 | } | |
ae90dd5d | 1442 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
c7fc77f7 | 1443 | |
0a13c00e TH |
1444 | /** |
1445 | * queue_delayed_work - queue work on a workqueue after delay | |
1446 | * @wq: workqueue to use | |
1447 | * @dwork: delayable work to queue | |
1448 | * @delay: number of jiffies to wait before queueing | |
1449 | * | |
715f1300 | 1450 | * Equivalent to queue_delayed_work_on() but tries to use the local CPU. |
0a13c00e | 1451 | */ |
d4283e93 | 1452 | bool queue_delayed_work(struct workqueue_struct *wq, |
0a13c00e TH |
1453 | struct delayed_work *dwork, unsigned long delay) |
1454 | { | |
57469821 | 1455 | return queue_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); |
0a13c00e TH |
1456 | } |
1457 | EXPORT_SYMBOL_GPL(queue_delayed_work); | |
c7fc77f7 | 1458 | |
8376fe22 TH |
1459 | /** |
1460 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU | |
1461 | * @cpu: CPU number to execute work on | |
1462 | * @wq: workqueue to use | |
1463 | * @dwork: work to queue | |
1464 | * @delay: number of jiffies to wait before queueing | |
1465 | * | |
1466 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, | |
1467 | * modify @dwork's timer so that it expires after @delay. If @delay is | |
1468 | * zero, @work is guaranteed to be scheduled immediately regardless of its | |
1469 | * current state. | |
1470 | * | |
1471 | * Returns %false if @dwork was idle and queued, %true if @dwork was | |
1472 | * pending and its timer was modified. | |
1473 | * | |
e0aecdd8 | 1474 | * This function is safe to call from any context including IRQ handler. |
8376fe22 TH |
1475 | * See try_to_grab_pending() for details. |
1476 | */ | |
1477 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
1478 | struct delayed_work *dwork, unsigned long delay) | |
1479 | { | |
1480 | unsigned long flags; | |
1481 | int ret; | |
c7fc77f7 | 1482 | |
8376fe22 TH |
1483 | do { |
1484 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
1485 | } while (unlikely(ret == -EAGAIN)); | |
63bc0362 | 1486 | |
8376fe22 TH |
1487 | if (likely(ret >= 0)) { |
1488 | __queue_delayed_work(cpu, wq, dwork, delay); | |
1489 | local_irq_restore(flags); | |
7a6bc1cd | 1490 | } |
8376fe22 TH |
1491 | |
1492 | /* -ENOENT from try_to_grab_pending() becomes %true */ | |
7a6bc1cd VP |
1493 | return ret; |
1494 | } | |
8376fe22 TH |
1495 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
1496 | ||
1497 | /** | |
1498 | * mod_delayed_work - modify delay of or queue a delayed work | |
1499 | * @wq: workqueue to use | |
1500 | * @dwork: work to queue | |
1501 | * @delay: number of jiffies to wait before queueing | |
1502 | * | |
1503 | * mod_delayed_work_on() on local CPU. | |
1504 | */ | |
1505 | bool mod_delayed_work(struct workqueue_struct *wq, struct delayed_work *dwork, | |
1506 | unsigned long delay) | |
1507 | { | |
1508 | return mod_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); | |
1509 | } | |
1510 | EXPORT_SYMBOL_GPL(mod_delayed_work); | |
1da177e4 | 1511 | |
c8e55f36 TH |
1512 | /** |
1513 | * worker_enter_idle - enter idle state | |
1514 | * @worker: worker which is entering idle state | |
1515 | * | |
1516 | * @worker is entering idle state. Update stats and idle timer if | |
1517 | * necessary. | |
1518 | * | |
1519 | * LOCKING: | |
1520 | * spin_lock_irq(gcwq->lock). | |
1521 | */ | |
1522 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1523 | { |
bd7bdd43 TH |
1524 | struct worker_pool *pool = worker->pool; |
1525 | struct global_cwq *gcwq = pool->gcwq; | |
c8e55f36 TH |
1526 | |
1527 | BUG_ON(worker->flags & WORKER_IDLE); | |
1528 | BUG_ON(!list_empty(&worker->entry) && | |
1529 | (worker->hentry.next || worker->hentry.pprev)); | |
1530 | ||
cb444766 TH |
1531 | /* can't use worker_set_flags(), also called from start_worker() */ |
1532 | worker->flags |= WORKER_IDLE; | |
bd7bdd43 | 1533 | pool->nr_idle++; |
e22bee78 | 1534 | worker->last_active = jiffies; |
c8e55f36 TH |
1535 | |
1536 | /* idle_list is LIFO */ | |
bd7bdd43 | 1537 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1538 | |
628c78e7 TH |
1539 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
1540 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); | |
cb444766 | 1541 | |
544ecf31 | 1542 | /* |
628c78e7 TH |
1543 | * Sanity check nr_running. Because gcwq_unbind_fn() releases |
1544 | * gcwq->lock between setting %WORKER_UNBOUND and zapping | |
1545 | * nr_running, the warning may trigger spuriously. Check iff | |
1546 | * unbind is not in progress. | |
544ecf31 | 1547 | */ |
628c78e7 | 1548 | WARN_ON_ONCE(!(gcwq->flags & GCWQ_DISASSOCIATED) && |
bd7bdd43 | 1549 | pool->nr_workers == pool->nr_idle && |
63d95a91 | 1550 | atomic_read(get_pool_nr_running(pool))); |
c8e55f36 TH |
1551 | } |
1552 | ||
1553 | /** | |
1554 | * worker_leave_idle - leave idle state | |
1555 | * @worker: worker which is leaving idle state | |
1556 | * | |
1557 | * @worker is leaving idle state. Update stats. | |
1558 | * | |
1559 | * LOCKING: | |
1560 | * spin_lock_irq(gcwq->lock). | |
1561 | */ | |
1562 | static void worker_leave_idle(struct worker *worker) | |
1563 | { | |
bd7bdd43 | 1564 | struct worker_pool *pool = worker->pool; |
c8e55f36 TH |
1565 | |
1566 | BUG_ON(!(worker->flags & WORKER_IDLE)); | |
d302f017 | 1567 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1568 | pool->nr_idle--; |
c8e55f36 TH |
1569 | list_del_init(&worker->entry); |
1570 | } | |
1571 | ||
e22bee78 TH |
1572 | /** |
1573 | * worker_maybe_bind_and_lock - bind worker to its cpu if possible and lock gcwq | |
1574 | * @worker: self | |
1575 | * | |
1576 | * Works which are scheduled while the cpu is online must at least be | |
1577 | * scheduled to a worker which is bound to the cpu so that if they are | |
1578 | * flushed from cpu callbacks while cpu is going down, they are | |
1579 | * guaranteed to execute on the cpu. | |
1580 | * | |
1581 | * This function is to be used by rogue workers and rescuers to bind | |
1582 | * themselves to the target cpu and may race with cpu going down or | |
1583 | * coming online. kthread_bind() can't be used because it may put the | |
1584 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
1585 | * verbatim as it's best effort and blocking and gcwq may be | |
1586 | * [dis]associated in the meantime. | |
1587 | * | |
f2d5a0ee TH |
1588 | * This function tries set_cpus_allowed() and locks gcwq and verifies the |
1589 | * binding against %GCWQ_DISASSOCIATED which is set during | |
1590 | * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker | |
1591 | * enters idle state or fetches works without dropping lock, it can | |
1592 | * guarantee the scheduling requirement described in the first paragraph. | |
e22bee78 TH |
1593 | * |
1594 | * CONTEXT: | |
1595 | * Might sleep. Called without any lock but returns with gcwq->lock | |
1596 | * held. | |
1597 | * | |
1598 | * RETURNS: | |
1599 | * %true if the associated gcwq is online (@worker is successfully | |
1600 | * bound), %false if offline. | |
1601 | */ | |
1602 | static bool worker_maybe_bind_and_lock(struct worker *worker) | |
972fa1c5 | 1603 | __acquires(&gcwq->lock) |
e22bee78 | 1604 | { |
bd7bdd43 | 1605 | struct global_cwq *gcwq = worker->pool->gcwq; |
e22bee78 TH |
1606 | struct task_struct *task = worker->task; |
1607 | ||
1608 | while (true) { | |
4e6045f1 | 1609 | /* |
e22bee78 TH |
1610 | * The following call may fail, succeed or succeed |
1611 | * without actually migrating the task to the cpu if | |
1612 | * it races with cpu hotunplug operation. Verify | |
1613 | * against GCWQ_DISASSOCIATED. | |
4e6045f1 | 1614 | */ |
f3421797 TH |
1615 | if (!(gcwq->flags & GCWQ_DISASSOCIATED)) |
1616 | set_cpus_allowed_ptr(task, get_cpu_mask(gcwq->cpu)); | |
e22bee78 TH |
1617 | |
1618 | spin_lock_irq(&gcwq->lock); | |
1619 | if (gcwq->flags & GCWQ_DISASSOCIATED) | |
1620 | return false; | |
1621 | if (task_cpu(task) == gcwq->cpu && | |
1622 | cpumask_equal(¤t->cpus_allowed, | |
1623 | get_cpu_mask(gcwq->cpu))) | |
1624 | return true; | |
1625 | spin_unlock_irq(&gcwq->lock); | |
1626 | ||
5035b20f TH |
1627 | /* |
1628 | * We've raced with CPU hot[un]plug. Give it a breather | |
1629 | * and retry migration. cond_resched() is required here; | |
1630 | * otherwise, we might deadlock against cpu_stop trying to | |
1631 | * bring down the CPU on non-preemptive kernel. | |
1632 | */ | |
e22bee78 | 1633 | cpu_relax(); |
5035b20f | 1634 | cond_resched(); |
e22bee78 TH |
1635 | } |
1636 | } | |
1637 | ||
25511a47 | 1638 | /* |
ea1abd61 | 1639 | * Rebind an idle @worker to its CPU. worker_thread() will test |
5f7dabfd | 1640 | * list_empty(@worker->entry) before leaving idle and call this function. |
25511a47 TH |
1641 | */ |
1642 | static void idle_worker_rebind(struct worker *worker) | |
1643 | { | |
1644 | struct global_cwq *gcwq = worker->pool->gcwq; | |
1645 | ||
5f7dabfd LJ |
1646 | /* CPU may go down again inbetween, clear UNBOUND only on success */ |
1647 | if (worker_maybe_bind_and_lock(worker)) | |
1648 | worker_clr_flags(worker, WORKER_UNBOUND); | |
25511a47 | 1649 | |
ea1abd61 LJ |
1650 | /* rebind complete, become available again */ |
1651 | list_add(&worker->entry, &worker->pool->idle_list); | |
1652 | spin_unlock_irq(&gcwq->lock); | |
25511a47 TH |
1653 | } |
1654 | ||
e22bee78 | 1655 | /* |
25511a47 | 1656 | * Function for @worker->rebind.work used to rebind unbound busy workers to |
403c821d TH |
1657 | * the associated cpu which is coming back online. This is scheduled by |
1658 | * cpu up but can race with other cpu hotplug operations and may be | |
1659 | * executed twice without intervening cpu down. | |
e22bee78 | 1660 | */ |
25511a47 | 1661 | static void busy_worker_rebind_fn(struct work_struct *work) |
e22bee78 TH |
1662 | { |
1663 | struct worker *worker = container_of(work, struct worker, rebind_work); | |
bd7bdd43 | 1664 | struct global_cwq *gcwq = worker->pool->gcwq; |
e22bee78 | 1665 | |
eab6d828 LJ |
1666 | if (worker_maybe_bind_and_lock(worker)) |
1667 | worker_clr_flags(worker, WORKER_UNBOUND); | |
e22bee78 TH |
1668 | |
1669 | spin_unlock_irq(&gcwq->lock); | |
1670 | } | |
1671 | ||
25511a47 TH |
1672 | /** |
1673 | * rebind_workers - rebind all workers of a gcwq to the associated CPU | |
1674 | * @gcwq: gcwq of interest | |
1675 | * | |
1676 | * @gcwq->cpu is coming online. Rebind all workers to the CPU. Rebinding | |
1677 | * is different for idle and busy ones. | |
1678 | * | |
ea1abd61 LJ |
1679 | * Idle ones will be removed from the idle_list and woken up. They will |
1680 | * add themselves back after completing rebind. This ensures that the | |
1681 | * idle_list doesn't contain any unbound workers when re-bound busy workers | |
1682 | * try to perform local wake-ups for concurrency management. | |
25511a47 | 1683 | * |
ea1abd61 LJ |
1684 | * Busy workers can rebind after they finish their current work items. |
1685 | * Queueing the rebind work item at the head of the scheduled list is | |
1686 | * enough. Note that nr_running will be properly bumped as busy workers | |
1687 | * rebind. | |
25511a47 | 1688 | * |
ea1abd61 LJ |
1689 | * On return, all non-manager workers are scheduled for rebind - see |
1690 | * manage_workers() for the manager special case. Any idle worker | |
1691 | * including the manager will not appear on @idle_list until rebind is | |
1692 | * complete, making local wake-ups safe. | |
25511a47 TH |
1693 | */ |
1694 | static void rebind_workers(struct global_cwq *gcwq) | |
25511a47 | 1695 | { |
25511a47 | 1696 | struct worker_pool *pool; |
ea1abd61 | 1697 | struct worker *worker, *n; |
25511a47 TH |
1698 | struct hlist_node *pos; |
1699 | int i; | |
1700 | ||
1701 | lockdep_assert_held(&gcwq->lock); | |
1702 | ||
1703 | for_each_worker_pool(pool, gcwq) | |
b2eb83d1 | 1704 | lockdep_assert_held(&pool->assoc_mutex); |
25511a47 | 1705 | |
5f7dabfd | 1706 | /* dequeue and kick idle ones */ |
25511a47 | 1707 | for_each_worker_pool(pool, gcwq) { |
ea1abd61 | 1708 | list_for_each_entry_safe(worker, n, &pool->idle_list, entry) { |
ea1abd61 LJ |
1709 | /* |
1710 | * idle workers should be off @pool->idle_list | |
1711 | * until rebind is complete to avoid receiving | |
1712 | * premature local wake-ups. | |
1713 | */ | |
1714 | list_del_init(&worker->entry); | |
25511a47 | 1715 | |
5f7dabfd LJ |
1716 | /* |
1717 | * worker_thread() will see the above dequeuing | |
1718 | * and call idle_worker_rebind(). | |
1719 | */ | |
25511a47 TH |
1720 | wake_up_process(worker->task); |
1721 | } | |
1722 | } | |
1723 | ||
ea1abd61 | 1724 | /* rebind busy workers */ |
25511a47 TH |
1725 | for_each_busy_worker(worker, i, pos, gcwq) { |
1726 | struct work_struct *rebind_work = &worker->rebind_work; | |
e2b6a6d5 | 1727 | struct workqueue_struct *wq; |
25511a47 TH |
1728 | |
1729 | if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, | |
1730 | work_data_bits(rebind_work))) | |
1731 | continue; | |
1732 | ||
25511a47 | 1733 | debug_work_activate(rebind_work); |
90beca5d | 1734 | |
e2b6a6d5 JK |
1735 | /* |
1736 | * wq doesn't really matter but let's keep @worker->pool | |
1737 | * and @cwq->pool consistent for sanity. | |
1738 | */ | |
1739 | if (worker_pool_pri(worker->pool)) | |
1740 | wq = system_highpri_wq; | |
1741 | else | |
1742 | wq = system_wq; | |
ec58815a | 1743 | |
e2b6a6d5 JK |
1744 | insert_work(get_cwq(gcwq->cpu, wq), rebind_work, |
1745 | worker->scheduled.next, | |
1746 | work_color_to_flags(WORK_NO_COLOR)); | |
ec58815a | 1747 | } |
25511a47 TH |
1748 | } |
1749 | ||
c34056a3 TH |
1750 | static struct worker *alloc_worker(void) |
1751 | { | |
1752 | struct worker *worker; | |
1753 | ||
1754 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1755 | if (worker) { |
1756 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1757 | INIT_LIST_HEAD(&worker->scheduled); |
25511a47 | 1758 | INIT_WORK(&worker->rebind_work, busy_worker_rebind_fn); |
e22bee78 TH |
1759 | /* on creation a worker is in !idle && prep state */ |
1760 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1761 | } |
c34056a3 TH |
1762 | return worker; |
1763 | } | |
1764 | ||
1765 | /** | |
1766 | * create_worker - create a new workqueue worker | |
63d95a91 | 1767 | * @pool: pool the new worker will belong to |
c34056a3 | 1768 | * |
63d95a91 | 1769 | * Create a new worker which is bound to @pool. The returned worker |
c34056a3 TH |
1770 | * can be started by calling start_worker() or destroyed using |
1771 | * destroy_worker(). | |
1772 | * | |
1773 | * CONTEXT: | |
1774 | * Might sleep. Does GFP_KERNEL allocations. | |
1775 | * | |
1776 | * RETURNS: | |
1777 | * Pointer to the newly created worker. | |
1778 | */ | |
bc2ae0f5 | 1779 | static struct worker *create_worker(struct worker_pool *pool) |
c34056a3 | 1780 | { |
63d95a91 | 1781 | struct global_cwq *gcwq = pool->gcwq; |
3270476a | 1782 | const char *pri = worker_pool_pri(pool) ? "H" : ""; |
c34056a3 | 1783 | struct worker *worker = NULL; |
f3421797 | 1784 | int id = -1; |
c34056a3 | 1785 | |
8b03ae3c | 1786 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1787 | while (ida_get_new(&pool->worker_ida, &id)) { |
8b03ae3c | 1788 | spin_unlock_irq(&gcwq->lock); |
bd7bdd43 | 1789 | if (!ida_pre_get(&pool->worker_ida, GFP_KERNEL)) |
c34056a3 | 1790 | goto fail; |
8b03ae3c | 1791 | spin_lock_irq(&gcwq->lock); |
c34056a3 | 1792 | } |
8b03ae3c | 1793 | spin_unlock_irq(&gcwq->lock); |
c34056a3 TH |
1794 | |
1795 | worker = alloc_worker(); | |
1796 | if (!worker) | |
1797 | goto fail; | |
1798 | ||
bd7bdd43 | 1799 | worker->pool = pool; |
c34056a3 TH |
1800 | worker->id = id; |
1801 | ||
bc2ae0f5 | 1802 | if (gcwq->cpu != WORK_CPU_UNBOUND) |
94dcf29a | 1803 | worker->task = kthread_create_on_node(worker_thread, |
3270476a TH |
1804 | worker, cpu_to_node(gcwq->cpu), |
1805 | "kworker/%u:%d%s", gcwq->cpu, id, pri); | |
f3421797 TH |
1806 | else |
1807 | worker->task = kthread_create(worker_thread, worker, | |
3270476a | 1808 | "kworker/u:%d%s", id, pri); |
c34056a3 TH |
1809 | if (IS_ERR(worker->task)) |
1810 | goto fail; | |
1811 | ||
3270476a TH |
1812 | if (worker_pool_pri(pool)) |
1813 | set_user_nice(worker->task, HIGHPRI_NICE_LEVEL); | |
1814 | ||
db7bccf4 | 1815 | /* |
bc2ae0f5 TH |
1816 | * Determine CPU binding of the new worker depending on |
1817 | * %GCWQ_DISASSOCIATED. The caller is responsible for ensuring the | |
1818 | * flag remains stable across this function. See the comments | |
1819 | * above the flag definition for details. | |
1820 | * | |
1821 | * As an unbound worker may later become a regular one if CPU comes | |
1822 | * online, make sure every worker has %PF_THREAD_BOUND set. | |
db7bccf4 | 1823 | */ |
bc2ae0f5 | 1824 | if (!(gcwq->flags & GCWQ_DISASSOCIATED)) { |
8b03ae3c | 1825 | kthread_bind(worker->task, gcwq->cpu); |
bc2ae0f5 | 1826 | } else { |
db7bccf4 | 1827 | worker->task->flags |= PF_THREAD_BOUND; |
bc2ae0f5 | 1828 | worker->flags |= WORKER_UNBOUND; |
f3421797 | 1829 | } |
c34056a3 TH |
1830 | |
1831 | return worker; | |
1832 | fail: | |
1833 | if (id >= 0) { | |
8b03ae3c | 1834 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1835 | ida_remove(&pool->worker_ida, id); |
8b03ae3c | 1836 | spin_unlock_irq(&gcwq->lock); |
c34056a3 TH |
1837 | } |
1838 | kfree(worker); | |
1839 | return NULL; | |
1840 | } | |
1841 | ||
1842 | /** | |
1843 | * start_worker - start a newly created worker | |
1844 | * @worker: worker to start | |
1845 | * | |
c8e55f36 | 1846 | * Make the gcwq aware of @worker and start it. |
c34056a3 TH |
1847 | * |
1848 | * CONTEXT: | |
8b03ae3c | 1849 | * spin_lock_irq(gcwq->lock). |
c34056a3 TH |
1850 | */ |
1851 | static void start_worker(struct worker *worker) | |
1852 | { | |
cb444766 | 1853 | worker->flags |= WORKER_STARTED; |
bd7bdd43 | 1854 | worker->pool->nr_workers++; |
c8e55f36 | 1855 | worker_enter_idle(worker); |
c34056a3 TH |
1856 | wake_up_process(worker->task); |
1857 | } | |
1858 | ||
1859 | /** | |
1860 | * destroy_worker - destroy a workqueue worker | |
1861 | * @worker: worker to be destroyed | |
1862 | * | |
c8e55f36 TH |
1863 | * Destroy @worker and adjust @gcwq stats accordingly. |
1864 | * | |
1865 | * CONTEXT: | |
1866 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. | |
c34056a3 TH |
1867 | */ |
1868 | static void destroy_worker(struct worker *worker) | |
1869 | { | |
bd7bdd43 TH |
1870 | struct worker_pool *pool = worker->pool; |
1871 | struct global_cwq *gcwq = pool->gcwq; | |
c34056a3 TH |
1872 | int id = worker->id; |
1873 | ||
1874 | /* sanity check frenzy */ | |
1875 | BUG_ON(worker->current_work); | |
affee4b2 | 1876 | BUG_ON(!list_empty(&worker->scheduled)); |
c34056a3 | 1877 | |
c8e55f36 | 1878 | if (worker->flags & WORKER_STARTED) |
bd7bdd43 | 1879 | pool->nr_workers--; |
c8e55f36 | 1880 | if (worker->flags & WORKER_IDLE) |
bd7bdd43 | 1881 | pool->nr_idle--; |
c8e55f36 TH |
1882 | |
1883 | list_del_init(&worker->entry); | |
cb444766 | 1884 | worker->flags |= WORKER_DIE; |
c8e55f36 TH |
1885 | |
1886 | spin_unlock_irq(&gcwq->lock); | |
1887 | ||
c34056a3 TH |
1888 | kthread_stop(worker->task); |
1889 | kfree(worker); | |
1890 | ||
8b03ae3c | 1891 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1892 | ida_remove(&pool->worker_ida, id); |
c34056a3 TH |
1893 | } |
1894 | ||
63d95a91 | 1895 | static void idle_worker_timeout(unsigned long __pool) |
e22bee78 | 1896 | { |
63d95a91 TH |
1897 | struct worker_pool *pool = (void *)__pool; |
1898 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
1899 | |
1900 | spin_lock_irq(&gcwq->lock); | |
1901 | ||
63d95a91 | 1902 | if (too_many_workers(pool)) { |
e22bee78 TH |
1903 | struct worker *worker; |
1904 | unsigned long expires; | |
1905 | ||
1906 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1907 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1908 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1909 | ||
1910 | if (time_before(jiffies, expires)) | |
63d95a91 | 1911 | mod_timer(&pool->idle_timer, expires); |
e22bee78 TH |
1912 | else { |
1913 | /* it's been idle for too long, wake up manager */ | |
11ebea50 | 1914 | pool->flags |= POOL_MANAGE_WORKERS; |
63d95a91 | 1915 | wake_up_worker(pool); |
d5abe669 | 1916 | } |
e22bee78 TH |
1917 | } |
1918 | ||
1919 | spin_unlock_irq(&gcwq->lock); | |
1920 | } | |
d5abe669 | 1921 | |
e22bee78 TH |
1922 | static bool send_mayday(struct work_struct *work) |
1923 | { | |
1924 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); | |
1925 | struct workqueue_struct *wq = cwq->wq; | |
f3421797 | 1926 | unsigned int cpu; |
e22bee78 TH |
1927 | |
1928 | if (!(wq->flags & WQ_RESCUER)) | |
1929 | return false; | |
1930 | ||
1931 | /* mayday mayday mayday */ | |
bd7bdd43 | 1932 | cpu = cwq->pool->gcwq->cpu; |
f3421797 TH |
1933 | /* WORK_CPU_UNBOUND can't be set in cpumask, use cpu 0 instead */ |
1934 | if (cpu == WORK_CPU_UNBOUND) | |
1935 | cpu = 0; | |
f2e005aa | 1936 | if (!mayday_test_and_set_cpu(cpu, wq->mayday_mask)) |
e22bee78 TH |
1937 | wake_up_process(wq->rescuer->task); |
1938 | return true; | |
1939 | } | |
1940 | ||
63d95a91 | 1941 | static void gcwq_mayday_timeout(unsigned long __pool) |
e22bee78 | 1942 | { |
63d95a91 TH |
1943 | struct worker_pool *pool = (void *)__pool; |
1944 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
1945 | struct work_struct *work; |
1946 | ||
1947 | spin_lock_irq(&gcwq->lock); | |
1948 | ||
63d95a91 | 1949 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
1950 | /* |
1951 | * We've been trying to create a new worker but | |
1952 | * haven't been successful. We might be hitting an | |
1953 | * allocation deadlock. Send distress signals to | |
1954 | * rescuers. | |
1955 | */ | |
63d95a91 | 1956 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 1957 | send_mayday(work); |
1da177e4 | 1958 | } |
e22bee78 TH |
1959 | |
1960 | spin_unlock_irq(&gcwq->lock); | |
1961 | ||
63d95a91 | 1962 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
1963 | } |
1964 | ||
e22bee78 TH |
1965 | /** |
1966 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 1967 | * @pool: pool to create a new worker for |
e22bee78 | 1968 | * |
63d95a91 | 1969 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
1970 | * have at least one idle worker on return from this function. If |
1971 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 1972 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
1973 | * possible allocation deadlock. |
1974 | * | |
1975 | * On return, need_to_create_worker() is guaranteed to be false and | |
1976 | * may_start_working() true. | |
1977 | * | |
1978 | * LOCKING: | |
1979 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1980 | * multiple times. Does GFP_KERNEL allocations. Called only from | |
1981 | * manager. | |
1982 | * | |
1983 | * RETURNS: | |
1984 | * false if no action was taken and gcwq->lock stayed locked, true | |
1985 | * otherwise. | |
1986 | */ | |
63d95a91 | 1987 | static bool maybe_create_worker(struct worker_pool *pool) |
06bd6ebf NK |
1988 | __releases(&gcwq->lock) |
1989 | __acquires(&gcwq->lock) | |
1da177e4 | 1990 | { |
63d95a91 TH |
1991 | struct global_cwq *gcwq = pool->gcwq; |
1992 | ||
1993 | if (!need_to_create_worker(pool)) | |
e22bee78 TH |
1994 | return false; |
1995 | restart: | |
9f9c2364 TH |
1996 | spin_unlock_irq(&gcwq->lock); |
1997 | ||
e22bee78 | 1998 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 1999 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
2000 | |
2001 | while (true) { | |
2002 | struct worker *worker; | |
2003 | ||
bc2ae0f5 | 2004 | worker = create_worker(pool); |
e22bee78 | 2005 | if (worker) { |
63d95a91 | 2006 | del_timer_sync(&pool->mayday_timer); |
e22bee78 TH |
2007 | spin_lock_irq(&gcwq->lock); |
2008 | start_worker(worker); | |
63d95a91 | 2009 | BUG_ON(need_to_create_worker(pool)); |
e22bee78 TH |
2010 | return true; |
2011 | } | |
2012 | ||
63d95a91 | 2013 | if (!need_to_create_worker(pool)) |
e22bee78 | 2014 | break; |
1da177e4 | 2015 | |
e22bee78 TH |
2016 | __set_current_state(TASK_INTERRUPTIBLE); |
2017 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 2018 | |
63d95a91 | 2019 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
2020 | break; |
2021 | } | |
2022 | ||
63d95a91 | 2023 | del_timer_sync(&pool->mayday_timer); |
e22bee78 | 2024 | spin_lock_irq(&gcwq->lock); |
63d95a91 | 2025 | if (need_to_create_worker(pool)) |
e22bee78 TH |
2026 | goto restart; |
2027 | return true; | |
2028 | } | |
2029 | ||
2030 | /** | |
2031 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
63d95a91 | 2032 | * @pool: pool to destroy workers for |
e22bee78 | 2033 | * |
63d95a91 | 2034 | * Destroy @pool workers which have been idle for longer than |
e22bee78 TH |
2035 | * IDLE_WORKER_TIMEOUT. |
2036 | * | |
2037 | * LOCKING: | |
2038 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
2039 | * multiple times. Called only from manager. | |
2040 | * | |
2041 | * RETURNS: | |
2042 | * false if no action was taken and gcwq->lock stayed locked, true | |
2043 | * otherwise. | |
2044 | */ | |
63d95a91 | 2045 | static bool maybe_destroy_workers(struct worker_pool *pool) |
e22bee78 TH |
2046 | { |
2047 | bool ret = false; | |
1da177e4 | 2048 | |
63d95a91 | 2049 | while (too_many_workers(pool)) { |
e22bee78 TH |
2050 | struct worker *worker; |
2051 | unsigned long expires; | |
3af24433 | 2052 | |
63d95a91 | 2053 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 | 2054 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
85f4186a | 2055 | |
e22bee78 | 2056 | if (time_before(jiffies, expires)) { |
63d95a91 | 2057 | mod_timer(&pool->idle_timer, expires); |
3af24433 | 2058 | break; |
e22bee78 | 2059 | } |
1da177e4 | 2060 | |
e22bee78 TH |
2061 | destroy_worker(worker); |
2062 | ret = true; | |
1da177e4 | 2063 | } |
1e19ffc6 | 2064 | |
e22bee78 | 2065 | return ret; |
1e19ffc6 TH |
2066 | } |
2067 | ||
73f53c4a | 2068 | /** |
e22bee78 TH |
2069 | * manage_workers - manage worker pool |
2070 | * @worker: self | |
73f53c4a | 2071 | * |
e22bee78 TH |
2072 | * Assume the manager role and manage gcwq worker pool @worker belongs |
2073 | * to. At any given time, there can be only zero or one manager per | |
2074 | * gcwq. The exclusion is handled automatically by this function. | |
2075 | * | |
2076 | * The caller can safely start processing works on false return. On | |
2077 | * true return, it's guaranteed that need_to_create_worker() is false | |
2078 | * and may_start_working() is true. | |
73f53c4a TH |
2079 | * |
2080 | * CONTEXT: | |
e22bee78 TH |
2081 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed |
2082 | * multiple times. Does GFP_KERNEL allocations. | |
2083 | * | |
2084 | * RETURNS: | |
2085 | * false if no action was taken and gcwq->lock stayed locked, true if | |
2086 | * some action was taken. | |
73f53c4a | 2087 | */ |
e22bee78 | 2088 | static bool manage_workers(struct worker *worker) |
73f53c4a | 2089 | { |
63d95a91 | 2090 | struct worker_pool *pool = worker->pool; |
e22bee78 | 2091 | bool ret = false; |
73f53c4a | 2092 | |
ee378aa4 | 2093 | if (pool->flags & POOL_MANAGING_WORKERS) |
e22bee78 | 2094 | return ret; |
1e19ffc6 | 2095 | |
552a37e9 | 2096 | pool->flags |= POOL_MANAGING_WORKERS; |
73f53c4a | 2097 | |
ee378aa4 LJ |
2098 | /* |
2099 | * To simplify both worker management and CPU hotplug, hold off | |
2100 | * management while hotplug is in progress. CPU hotplug path can't | |
2101 | * grab %POOL_MANAGING_WORKERS to achieve this because that can | |
2102 | * lead to idle worker depletion (all become busy thinking someone | |
2103 | * else is managing) which in turn can result in deadlock under | |
b2eb83d1 | 2104 | * extreme circumstances. Use @pool->assoc_mutex to synchronize |
ee378aa4 LJ |
2105 | * manager against CPU hotplug. |
2106 | * | |
b2eb83d1 | 2107 | * assoc_mutex would always be free unless CPU hotplug is in |
ee378aa4 LJ |
2108 | * progress. trylock first without dropping @gcwq->lock. |
2109 | */ | |
b2eb83d1 | 2110 | if (unlikely(!mutex_trylock(&pool->assoc_mutex))) { |
ee378aa4 | 2111 | spin_unlock_irq(&pool->gcwq->lock); |
b2eb83d1 | 2112 | mutex_lock(&pool->assoc_mutex); |
ee378aa4 LJ |
2113 | /* |
2114 | * CPU hotplug could have happened while we were waiting | |
b2eb83d1 | 2115 | * for assoc_mutex. Hotplug itself can't handle us |
ee378aa4 LJ |
2116 | * because manager isn't either on idle or busy list, and |
2117 | * @gcwq's state and ours could have deviated. | |
2118 | * | |
b2eb83d1 | 2119 | * As hotplug is now excluded via assoc_mutex, we can |
ee378aa4 LJ |
2120 | * simply try to bind. It will succeed or fail depending |
2121 | * on @gcwq's current state. Try it and adjust | |
2122 | * %WORKER_UNBOUND accordingly. | |
2123 | */ | |
2124 | if (worker_maybe_bind_and_lock(worker)) | |
2125 | worker->flags &= ~WORKER_UNBOUND; | |
2126 | else | |
2127 | worker->flags |= WORKER_UNBOUND; | |
73f53c4a | 2128 | |
ee378aa4 LJ |
2129 | ret = true; |
2130 | } | |
73f53c4a | 2131 | |
11ebea50 | 2132 | pool->flags &= ~POOL_MANAGE_WORKERS; |
73f53c4a TH |
2133 | |
2134 | /* | |
e22bee78 TH |
2135 | * Destroy and then create so that may_start_working() is true |
2136 | * on return. | |
73f53c4a | 2137 | */ |
63d95a91 TH |
2138 | ret |= maybe_destroy_workers(pool); |
2139 | ret |= maybe_create_worker(pool); | |
e22bee78 | 2140 | |
552a37e9 | 2141 | pool->flags &= ~POOL_MANAGING_WORKERS; |
b2eb83d1 | 2142 | mutex_unlock(&pool->assoc_mutex); |
e22bee78 | 2143 | return ret; |
73f53c4a TH |
2144 | } |
2145 | ||
a62428c0 TH |
2146 | /** |
2147 | * process_one_work - process single work | |
c34056a3 | 2148 | * @worker: self |
a62428c0 TH |
2149 | * @work: work to process |
2150 | * | |
2151 | * Process @work. This function contains all the logics necessary to | |
2152 | * process a single work including synchronization against and | |
2153 | * interaction with other workers on the same cpu, queueing and | |
2154 | * flushing. As long as context requirement is met, any worker can | |
2155 | * call this function to process a work. | |
2156 | * | |
2157 | * CONTEXT: | |
8b03ae3c | 2158 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. |
a62428c0 | 2159 | */ |
c34056a3 | 2160 | static void process_one_work(struct worker *worker, struct work_struct *work) |
06bd6ebf NK |
2161 | __releases(&gcwq->lock) |
2162 | __acquires(&gcwq->lock) | |
a62428c0 | 2163 | { |
7e11629d | 2164 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); |
bd7bdd43 TH |
2165 | struct worker_pool *pool = worker->pool; |
2166 | struct global_cwq *gcwq = pool->gcwq; | |
c8e55f36 | 2167 | struct hlist_head *bwh = busy_worker_head(gcwq, work); |
fb0e7beb | 2168 | bool cpu_intensive = cwq->wq->flags & WQ_CPU_INTENSIVE; |
a62428c0 | 2169 | work_func_t f = work->func; |
73f53c4a | 2170 | int work_color; |
7e11629d | 2171 | struct worker *collision; |
a62428c0 TH |
2172 | #ifdef CONFIG_LOCKDEP |
2173 | /* | |
2174 | * It is permissible to free the struct work_struct from | |
2175 | * inside the function that is called from it, this we need to | |
2176 | * take into account for lockdep too. To avoid bogus "held | |
2177 | * lock freed" warnings as well as problems when looking into | |
2178 | * work->lockdep_map, make a copy and use that here. | |
2179 | */ | |
4d82a1de PZ |
2180 | struct lockdep_map lockdep_map; |
2181 | ||
2182 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 2183 | #endif |
6fec10a1 TH |
2184 | /* |
2185 | * Ensure we're on the correct CPU. DISASSOCIATED test is | |
2186 | * necessary to avoid spurious warnings from rescuers servicing the | |
2187 | * unbound or a disassociated gcwq. | |
2188 | */ | |
5f7dabfd | 2189 | WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) && |
6fec10a1 | 2190 | !(gcwq->flags & GCWQ_DISASSOCIATED) && |
25511a47 TH |
2191 | raw_smp_processor_id() != gcwq->cpu); |
2192 | ||
7e11629d TH |
2193 | /* |
2194 | * A single work shouldn't be executed concurrently by | |
2195 | * multiple workers on a single cpu. Check whether anyone is | |
2196 | * already processing the work. If so, defer the work to the | |
2197 | * currently executing one. | |
2198 | */ | |
2199 | collision = __find_worker_executing_work(gcwq, bwh, work); | |
2200 | if (unlikely(collision)) { | |
2201 | move_linked_works(work, &collision->scheduled, NULL); | |
2202 | return; | |
2203 | } | |
2204 | ||
8930caba | 2205 | /* claim and dequeue */ |
a62428c0 | 2206 | debug_work_deactivate(work); |
c8e55f36 | 2207 | hlist_add_head(&worker->hentry, bwh); |
c34056a3 | 2208 | worker->current_work = work; |
8cca0eea | 2209 | worker->current_cwq = cwq; |
73f53c4a | 2210 | work_color = get_work_color(work); |
7a22ad75 | 2211 | |
a62428c0 TH |
2212 | list_del_init(&work->entry); |
2213 | ||
fb0e7beb TH |
2214 | /* |
2215 | * CPU intensive works don't participate in concurrency | |
2216 | * management. They're the scheduler's responsibility. | |
2217 | */ | |
2218 | if (unlikely(cpu_intensive)) | |
2219 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
2220 | ||
974271c4 TH |
2221 | /* |
2222 | * Unbound gcwq isn't concurrency managed and work items should be | |
2223 | * executed ASAP. Wake up another worker if necessary. | |
2224 | */ | |
63d95a91 TH |
2225 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
2226 | wake_up_worker(pool); | |
974271c4 | 2227 | |
8930caba | 2228 | /* |
23657bb1 TH |
2229 | * Record the last CPU and clear PENDING which should be the last |
2230 | * update to @work. Also, do this inside @gcwq->lock so that | |
2231 | * PENDING and queued state changes happen together while IRQ is | |
2232 | * disabled. | |
8930caba | 2233 | */ |
8930caba | 2234 | set_work_cpu_and_clear_pending(work, gcwq->cpu); |
a62428c0 | 2235 | |
8b03ae3c | 2236 | spin_unlock_irq(&gcwq->lock); |
a62428c0 | 2237 | |
e159489b | 2238 | lock_map_acquire_read(&cwq->wq->lockdep_map); |
a62428c0 | 2239 | lock_map_acquire(&lockdep_map); |
e36c886a | 2240 | trace_workqueue_execute_start(work); |
a62428c0 | 2241 | f(work); |
e36c886a AV |
2242 | /* |
2243 | * While we must be careful to not use "work" after this, the trace | |
2244 | * point will only record its address. | |
2245 | */ | |
2246 | trace_workqueue_execute_end(work); | |
a62428c0 TH |
2247 | lock_map_release(&lockdep_map); |
2248 | lock_map_release(&cwq->wq->lockdep_map); | |
2249 | ||
2250 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
044c782c VI |
2251 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
2252 | " last function: %pf\n", | |
2253 | current->comm, preempt_count(), task_pid_nr(current), f); | |
a62428c0 TH |
2254 | debug_show_held_locks(current); |
2255 | dump_stack(); | |
2256 | } | |
2257 | ||
8b03ae3c | 2258 | spin_lock_irq(&gcwq->lock); |
a62428c0 | 2259 | |
fb0e7beb TH |
2260 | /* clear cpu intensive status */ |
2261 | if (unlikely(cpu_intensive)) | |
2262 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
2263 | ||
a62428c0 | 2264 | /* we're done with it, release */ |
c8e55f36 | 2265 | hlist_del_init(&worker->hentry); |
c34056a3 | 2266 | worker->current_work = NULL; |
8cca0eea | 2267 | worker->current_cwq = NULL; |
b3f9f405 | 2268 | cwq_dec_nr_in_flight(cwq, work_color); |
a62428c0 TH |
2269 | } |
2270 | ||
affee4b2 TH |
2271 | /** |
2272 | * process_scheduled_works - process scheduled works | |
2273 | * @worker: self | |
2274 | * | |
2275 | * Process all scheduled works. Please note that the scheduled list | |
2276 | * may change while processing a work, so this function repeatedly | |
2277 | * fetches a work from the top and executes it. | |
2278 | * | |
2279 | * CONTEXT: | |
8b03ae3c | 2280 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed |
affee4b2 TH |
2281 | * multiple times. |
2282 | */ | |
2283 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 2284 | { |
affee4b2 TH |
2285 | while (!list_empty(&worker->scheduled)) { |
2286 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 2287 | struct work_struct, entry); |
c34056a3 | 2288 | process_one_work(worker, work); |
1da177e4 | 2289 | } |
1da177e4 LT |
2290 | } |
2291 | ||
4690c4ab TH |
2292 | /** |
2293 | * worker_thread - the worker thread function | |
c34056a3 | 2294 | * @__worker: self |
4690c4ab | 2295 | * |
e22bee78 TH |
2296 | * The gcwq worker thread function. There's a single dynamic pool of |
2297 | * these per each cpu. These workers process all works regardless of | |
2298 | * their specific target workqueue. The only exception is works which | |
2299 | * belong to workqueues with a rescuer which will be explained in | |
2300 | * rescuer_thread(). | |
4690c4ab | 2301 | */ |
c34056a3 | 2302 | static int worker_thread(void *__worker) |
1da177e4 | 2303 | { |
c34056a3 | 2304 | struct worker *worker = __worker; |
bd7bdd43 TH |
2305 | struct worker_pool *pool = worker->pool; |
2306 | struct global_cwq *gcwq = pool->gcwq; | |
1da177e4 | 2307 | |
e22bee78 TH |
2308 | /* tell the scheduler that this is a workqueue worker */ |
2309 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 2310 | woke_up: |
c8e55f36 | 2311 | spin_lock_irq(&gcwq->lock); |
1da177e4 | 2312 | |
5f7dabfd LJ |
2313 | /* we are off idle list if destruction or rebind is requested */ |
2314 | if (unlikely(list_empty(&worker->entry))) { | |
c8e55f36 | 2315 | spin_unlock_irq(&gcwq->lock); |
25511a47 | 2316 | |
5f7dabfd | 2317 | /* if DIE is set, destruction is requested */ |
25511a47 TH |
2318 | if (worker->flags & WORKER_DIE) { |
2319 | worker->task->flags &= ~PF_WQ_WORKER; | |
2320 | return 0; | |
2321 | } | |
2322 | ||
5f7dabfd | 2323 | /* otherwise, rebind */ |
25511a47 TH |
2324 | idle_worker_rebind(worker); |
2325 | goto woke_up; | |
c8e55f36 | 2326 | } |
affee4b2 | 2327 | |
c8e55f36 | 2328 | worker_leave_idle(worker); |
db7bccf4 | 2329 | recheck: |
e22bee78 | 2330 | /* no more worker necessary? */ |
63d95a91 | 2331 | if (!need_more_worker(pool)) |
e22bee78 TH |
2332 | goto sleep; |
2333 | ||
2334 | /* do we need to manage? */ | |
63d95a91 | 2335 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
2336 | goto recheck; |
2337 | ||
c8e55f36 TH |
2338 | /* |
2339 | * ->scheduled list can only be filled while a worker is | |
2340 | * preparing to process a work or actually processing it. | |
2341 | * Make sure nobody diddled with it while I was sleeping. | |
2342 | */ | |
2343 | BUG_ON(!list_empty(&worker->scheduled)); | |
2344 | ||
e22bee78 TH |
2345 | /* |
2346 | * When control reaches this point, we're guaranteed to have | |
2347 | * at least one idle worker or that someone else has already | |
2348 | * assumed the manager role. | |
2349 | */ | |
2350 | worker_clr_flags(worker, WORKER_PREP); | |
2351 | ||
2352 | do { | |
c8e55f36 | 2353 | struct work_struct *work = |
bd7bdd43 | 2354 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2355 | struct work_struct, entry); |
2356 | ||
2357 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
2358 | /* optimization path, not strictly necessary */ | |
2359 | process_one_work(worker, work); | |
2360 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2361 | process_scheduled_works(worker); |
c8e55f36 TH |
2362 | } else { |
2363 | move_linked_works(work, &worker->scheduled, NULL); | |
2364 | process_scheduled_works(worker); | |
affee4b2 | 2365 | } |
63d95a91 | 2366 | } while (keep_working(pool)); |
e22bee78 TH |
2367 | |
2368 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 2369 | sleep: |
63d95a91 | 2370 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
e22bee78 | 2371 | goto recheck; |
d313dd85 | 2372 | |
c8e55f36 | 2373 | /* |
e22bee78 TH |
2374 | * gcwq->lock is held and there's no work to process and no |
2375 | * need to manage, sleep. Workers are woken up only while | |
2376 | * holding gcwq->lock or from local cpu, so setting the | |
2377 | * current state before releasing gcwq->lock is enough to | |
2378 | * prevent losing any event. | |
c8e55f36 TH |
2379 | */ |
2380 | worker_enter_idle(worker); | |
2381 | __set_current_state(TASK_INTERRUPTIBLE); | |
2382 | spin_unlock_irq(&gcwq->lock); | |
2383 | schedule(); | |
2384 | goto woke_up; | |
1da177e4 LT |
2385 | } |
2386 | ||
e22bee78 TH |
2387 | /** |
2388 | * rescuer_thread - the rescuer thread function | |
2389 | * @__wq: the associated workqueue | |
2390 | * | |
2391 | * Workqueue rescuer thread function. There's one rescuer for each | |
2392 | * workqueue which has WQ_RESCUER set. | |
2393 | * | |
2394 | * Regular work processing on a gcwq may block trying to create a new | |
2395 | * worker which uses GFP_KERNEL allocation which has slight chance of | |
2396 | * developing into deadlock if some works currently on the same queue | |
2397 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2398 | * the problem rescuer solves. | |
2399 | * | |
2400 | * When such condition is possible, the gcwq summons rescuers of all | |
2401 | * workqueues which have works queued on the gcwq and let them process | |
2402 | * those works so that forward progress can be guaranteed. | |
2403 | * | |
2404 | * This should happen rarely. | |
2405 | */ | |
2406 | static int rescuer_thread(void *__wq) | |
2407 | { | |
2408 | struct workqueue_struct *wq = __wq; | |
2409 | struct worker *rescuer = wq->rescuer; | |
2410 | struct list_head *scheduled = &rescuer->scheduled; | |
f3421797 | 2411 | bool is_unbound = wq->flags & WQ_UNBOUND; |
e22bee78 TH |
2412 | unsigned int cpu; |
2413 | ||
2414 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
2415 | repeat: | |
2416 | set_current_state(TASK_INTERRUPTIBLE); | |
2417 | ||
412d32e6 MG |
2418 | if (kthread_should_stop()) { |
2419 | __set_current_state(TASK_RUNNING); | |
e22bee78 | 2420 | return 0; |
412d32e6 | 2421 | } |
e22bee78 | 2422 | |
f3421797 TH |
2423 | /* |
2424 | * See whether any cpu is asking for help. Unbounded | |
2425 | * workqueues use cpu 0 in mayday_mask for CPU_UNBOUND. | |
2426 | */ | |
f2e005aa | 2427 | for_each_mayday_cpu(cpu, wq->mayday_mask) { |
f3421797 TH |
2428 | unsigned int tcpu = is_unbound ? WORK_CPU_UNBOUND : cpu; |
2429 | struct cpu_workqueue_struct *cwq = get_cwq(tcpu, wq); | |
bd7bdd43 TH |
2430 | struct worker_pool *pool = cwq->pool; |
2431 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
2432 | struct work_struct *work, *n; |
2433 | ||
2434 | __set_current_state(TASK_RUNNING); | |
f2e005aa | 2435 | mayday_clear_cpu(cpu, wq->mayday_mask); |
e22bee78 TH |
2436 | |
2437 | /* migrate to the target cpu if possible */ | |
bd7bdd43 | 2438 | rescuer->pool = pool; |
e22bee78 TH |
2439 | worker_maybe_bind_and_lock(rescuer); |
2440 | ||
2441 | /* | |
2442 | * Slurp in all works issued via this workqueue and | |
2443 | * process'em. | |
2444 | */ | |
2445 | BUG_ON(!list_empty(&rescuer->scheduled)); | |
bd7bdd43 | 2446 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
e22bee78 TH |
2447 | if (get_work_cwq(work) == cwq) |
2448 | move_linked_works(work, scheduled, &n); | |
2449 | ||
2450 | process_scheduled_works(rescuer); | |
7576958a TH |
2451 | |
2452 | /* | |
2453 | * Leave this gcwq. If keep_working() is %true, notify a | |
2454 | * regular worker; otherwise, we end up with 0 concurrency | |
2455 | * and stalling the execution. | |
2456 | */ | |
63d95a91 TH |
2457 | if (keep_working(pool)) |
2458 | wake_up_worker(pool); | |
7576958a | 2459 | |
e22bee78 TH |
2460 | spin_unlock_irq(&gcwq->lock); |
2461 | } | |
2462 | ||
2463 | schedule(); | |
2464 | goto repeat; | |
1da177e4 LT |
2465 | } |
2466 | ||
fc2e4d70 ON |
2467 | struct wq_barrier { |
2468 | struct work_struct work; | |
2469 | struct completion done; | |
2470 | }; | |
2471 | ||
2472 | static void wq_barrier_func(struct work_struct *work) | |
2473 | { | |
2474 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2475 | complete(&barr->done); | |
2476 | } | |
2477 | ||
4690c4ab TH |
2478 | /** |
2479 | * insert_wq_barrier - insert a barrier work | |
2480 | * @cwq: cwq to insert barrier into | |
2481 | * @barr: wq_barrier to insert | |
affee4b2 TH |
2482 | * @target: target work to attach @barr to |
2483 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2484 | * |
affee4b2 TH |
2485 | * @barr is linked to @target such that @barr is completed only after |
2486 | * @target finishes execution. Please note that the ordering | |
2487 | * guarantee is observed only with respect to @target and on the local | |
2488 | * cpu. | |
2489 | * | |
2490 | * Currently, a queued barrier can't be canceled. This is because | |
2491 | * try_to_grab_pending() can't determine whether the work to be | |
2492 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2493 | * flag of the previous work while there must be a valid next work | |
2494 | * after a work with LINKED flag set. | |
2495 | * | |
2496 | * Note that when @worker is non-NULL, @target may be modified | |
2497 | * underneath us, so we can't reliably determine cwq from @target. | |
4690c4ab TH |
2498 | * |
2499 | * CONTEXT: | |
8b03ae3c | 2500 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 2501 | */ |
83c22520 | 2502 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, |
affee4b2 TH |
2503 | struct wq_barrier *barr, |
2504 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2505 | { |
affee4b2 TH |
2506 | struct list_head *head; |
2507 | unsigned int linked = 0; | |
2508 | ||
dc186ad7 | 2509 | /* |
8b03ae3c | 2510 | * debugobject calls are safe here even with gcwq->lock locked |
dc186ad7 TG |
2511 | * as we know for sure that this will not trigger any of the |
2512 | * checks and call back into the fixup functions where we | |
2513 | * might deadlock. | |
2514 | */ | |
ca1cab37 | 2515 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2516 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2517 | init_completion(&barr->done); |
83c22520 | 2518 | |
affee4b2 TH |
2519 | /* |
2520 | * If @target is currently being executed, schedule the | |
2521 | * barrier to the worker; otherwise, put it after @target. | |
2522 | */ | |
2523 | if (worker) | |
2524 | head = worker->scheduled.next; | |
2525 | else { | |
2526 | unsigned long *bits = work_data_bits(target); | |
2527 | ||
2528 | head = target->entry.next; | |
2529 | /* there can already be other linked works, inherit and set */ | |
2530 | linked = *bits & WORK_STRUCT_LINKED; | |
2531 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2532 | } | |
2533 | ||
dc186ad7 | 2534 | debug_work_activate(&barr->work); |
affee4b2 TH |
2535 | insert_work(cwq, &barr->work, head, |
2536 | work_color_to_flags(WORK_NO_COLOR) | linked); | |
fc2e4d70 ON |
2537 | } |
2538 | ||
73f53c4a TH |
2539 | /** |
2540 | * flush_workqueue_prep_cwqs - prepare cwqs for workqueue flushing | |
2541 | * @wq: workqueue being flushed | |
2542 | * @flush_color: new flush color, < 0 for no-op | |
2543 | * @work_color: new work color, < 0 for no-op | |
2544 | * | |
2545 | * Prepare cwqs for workqueue flushing. | |
2546 | * | |
2547 | * If @flush_color is non-negative, flush_color on all cwqs should be | |
2548 | * -1. If no cwq has in-flight commands at the specified color, all | |
2549 | * cwq->flush_color's stay at -1 and %false is returned. If any cwq | |
2550 | * has in flight commands, its cwq->flush_color is set to | |
2551 | * @flush_color, @wq->nr_cwqs_to_flush is updated accordingly, cwq | |
2552 | * wakeup logic is armed and %true is returned. | |
2553 | * | |
2554 | * The caller should have initialized @wq->first_flusher prior to | |
2555 | * calling this function with non-negative @flush_color. If | |
2556 | * @flush_color is negative, no flush color update is done and %false | |
2557 | * is returned. | |
2558 | * | |
2559 | * If @work_color is non-negative, all cwqs should have the same | |
2560 | * work_color which is previous to @work_color and all will be | |
2561 | * advanced to @work_color. | |
2562 | * | |
2563 | * CONTEXT: | |
2564 | * mutex_lock(wq->flush_mutex). | |
2565 | * | |
2566 | * RETURNS: | |
2567 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2568 | * otherwise. | |
2569 | */ | |
2570 | static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq, | |
2571 | int flush_color, int work_color) | |
1da177e4 | 2572 | { |
73f53c4a TH |
2573 | bool wait = false; |
2574 | unsigned int cpu; | |
1da177e4 | 2575 | |
73f53c4a TH |
2576 | if (flush_color >= 0) { |
2577 | BUG_ON(atomic_read(&wq->nr_cwqs_to_flush)); | |
2578 | atomic_set(&wq->nr_cwqs_to_flush, 1); | |
1da177e4 | 2579 | } |
2355b70f | 2580 | |
f3421797 | 2581 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a | 2582 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
bd7bdd43 | 2583 | struct global_cwq *gcwq = cwq->pool->gcwq; |
fc2e4d70 | 2584 | |
8b03ae3c | 2585 | spin_lock_irq(&gcwq->lock); |
83c22520 | 2586 | |
73f53c4a TH |
2587 | if (flush_color >= 0) { |
2588 | BUG_ON(cwq->flush_color != -1); | |
fc2e4d70 | 2589 | |
73f53c4a TH |
2590 | if (cwq->nr_in_flight[flush_color]) { |
2591 | cwq->flush_color = flush_color; | |
2592 | atomic_inc(&wq->nr_cwqs_to_flush); | |
2593 | wait = true; | |
2594 | } | |
2595 | } | |
1da177e4 | 2596 | |
73f53c4a TH |
2597 | if (work_color >= 0) { |
2598 | BUG_ON(work_color != work_next_color(cwq->work_color)); | |
2599 | cwq->work_color = work_color; | |
2600 | } | |
1da177e4 | 2601 | |
8b03ae3c | 2602 | spin_unlock_irq(&gcwq->lock); |
1da177e4 | 2603 | } |
2355b70f | 2604 | |
73f53c4a TH |
2605 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_cwqs_to_flush)) |
2606 | complete(&wq->first_flusher->done); | |
14441960 | 2607 | |
73f53c4a | 2608 | return wait; |
1da177e4 LT |
2609 | } |
2610 | ||
0fcb78c2 | 2611 | /** |
1da177e4 | 2612 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2613 | * @wq: workqueue to flush |
1da177e4 LT |
2614 | * |
2615 | * Forces execution of the workqueue and blocks until its completion. | |
2616 | * This is typically used in driver shutdown handlers. | |
2617 | * | |
fc2e4d70 ON |
2618 | * We sleep until all works which were queued on entry have been handled, |
2619 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 2620 | */ |
7ad5b3a5 | 2621 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2622 | { |
73f53c4a TH |
2623 | struct wq_flusher this_flusher = { |
2624 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2625 | .flush_color = -1, | |
2626 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2627 | }; | |
2628 | int next_color; | |
1da177e4 | 2629 | |
3295f0ef IM |
2630 | lock_map_acquire(&wq->lockdep_map); |
2631 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
2632 | |
2633 | mutex_lock(&wq->flush_mutex); | |
2634 | ||
2635 | /* | |
2636 | * Start-to-wait phase | |
2637 | */ | |
2638 | next_color = work_next_color(wq->work_color); | |
2639 | ||
2640 | if (next_color != wq->flush_color) { | |
2641 | /* | |
2642 | * Color space is not full. The current work_color | |
2643 | * becomes our flush_color and work_color is advanced | |
2644 | * by one. | |
2645 | */ | |
2646 | BUG_ON(!list_empty(&wq->flusher_overflow)); | |
2647 | this_flusher.flush_color = wq->work_color; | |
2648 | wq->work_color = next_color; | |
2649 | ||
2650 | if (!wq->first_flusher) { | |
2651 | /* no flush in progress, become the first flusher */ | |
2652 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2653 | ||
2654 | wq->first_flusher = &this_flusher; | |
2655 | ||
2656 | if (!flush_workqueue_prep_cwqs(wq, wq->flush_color, | |
2657 | wq->work_color)) { | |
2658 | /* nothing to flush, done */ | |
2659 | wq->flush_color = next_color; | |
2660 | wq->first_flusher = NULL; | |
2661 | goto out_unlock; | |
2662 | } | |
2663 | } else { | |
2664 | /* wait in queue */ | |
2665 | BUG_ON(wq->flush_color == this_flusher.flush_color); | |
2666 | list_add_tail(&this_flusher.list, &wq->flusher_queue); | |
2667 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2668 | } | |
2669 | } else { | |
2670 | /* | |
2671 | * Oops, color space is full, wait on overflow queue. | |
2672 | * The next flush completion will assign us | |
2673 | * flush_color and transfer to flusher_queue. | |
2674 | */ | |
2675 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2676 | } | |
2677 | ||
2678 | mutex_unlock(&wq->flush_mutex); | |
2679 | ||
2680 | wait_for_completion(&this_flusher.done); | |
2681 | ||
2682 | /* | |
2683 | * Wake-up-and-cascade phase | |
2684 | * | |
2685 | * First flushers are responsible for cascading flushes and | |
2686 | * handling overflow. Non-first flushers can simply return. | |
2687 | */ | |
2688 | if (wq->first_flusher != &this_flusher) | |
2689 | return; | |
2690 | ||
2691 | mutex_lock(&wq->flush_mutex); | |
2692 | ||
4ce48b37 TH |
2693 | /* we might have raced, check again with mutex held */ |
2694 | if (wq->first_flusher != &this_flusher) | |
2695 | goto out_unlock; | |
2696 | ||
73f53c4a TH |
2697 | wq->first_flusher = NULL; |
2698 | ||
2699 | BUG_ON(!list_empty(&this_flusher.list)); | |
2700 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2701 | ||
2702 | while (true) { | |
2703 | struct wq_flusher *next, *tmp; | |
2704 | ||
2705 | /* complete all the flushers sharing the current flush color */ | |
2706 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2707 | if (next->flush_color != wq->flush_color) | |
2708 | break; | |
2709 | list_del_init(&next->list); | |
2710 | complete(&next->done); | |
2711 | } | |
2712 | ||
2713 | BUG_ON(!list_empty(&wq->flusher_overflow) && | |
2714 | wq->flush_color != work_next_color(wq->work_color)); | |
2715 | ||
2716 | /* this flush_color is finished, advance by one */ | |
2717 | wq->flush_color = work_next_color(wq->flush_color); | |
2718 | ||
2719 | /* one color has been freed, handle overflow queue */ | |
2720 | if (!list_empty(&wq->flusher_overflow)) { | |
2721 | /* | |
2722 | * Assign the same color to all overflowed | |
2723 | * flushers, advance work_color and append to | |
2724 | * flusher_queue. This is the start-to-wait | |
2725 | * phase for these overflowed flushers. | |
2726 | */ | |
2727 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2728 | tmp->flush_color = wq->work_color; | |
2729 | ||
2730 | wq->work_color = work_next_color(wq->work_color); | |
2731 | ||
2732 | list_splice_tail_init(&wq->flusher_overflow, | |
2733 | &wq->flusher_queue); | |
2734 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2735 | } | |
2736 | ||
2737 | if (list_empty(&wq->flusher_queue)) { | |
2738 | BUG_ON(wq->flush_color != wq->work_color); | |
2739 | break; | |
2740 | } | |
2741 | ||
2742 | /* | |
2743 | * Need to flush more colors. Make the next flusher | |
2744 | * the new first flusher and arm cwqs. | |
2745 | */ | |
2746 | BUG_ON(wq->flush_color == wq->work_color); | |
2747 | BUG_ON(wq->flush_color != next->flush_color); | |
2748 | ||
2749 | list_del_init(&next->list); | |
2750 | wq->first_flusher = next; | |
2751 | ||
2752 | if (flush_workqueue_prep_cwqs(wq, wq->flush_color, -1)) | |
2753 | break; | |
2754 | ||
2755 | /* | |
2756 | * Meh... this color is already done, clear first | |
2757 | * flusher and repeat cascading. | |
2758 | */ | |
2759 | wq->first_flusher = NULL; | |
2760 | } | |
2761 | ||
2762 | out_unlock: | |
2763 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 2764 | } |
ae90dd5d | 2765 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2766 | |
9c5a2ba7 TH |
2767 | /** |
2768 | * drain_workqueue - drain a workqueue | |
2769 | * @wq: workqueue to drain | |
2770 | * | |
2771 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2772 | * only chain queueing is allowed. IOW, only currently pending or running | |
2773 | * work items on @wq can queue further work items on it. @wq is flushed | |
2774 | * repeatedly until it becomes empty. The number of flushing is detemined | |
2775 | * by the depth of chaining and should be relatively short. Whine if it | |
2776 | * takes too long. | |
2777 | */ | |
2778 | void drain_workqueue(struct workqueue_struct *wq) | |
2779 | { | |
2780 | unsigned int flush_cnt = 0; | |
2781 | unsigned int cpu; | |
2782 | ||
2783 | /* | |
2784 | * __queue_work() needs to test whether there are drainers, is much | |
2785 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
2786 | * Use WQ_DRAINING so that queue doesn't have to check nr_drainers. | |
2787 | */ | |
2788 | spin_lock(&workqueue_lock); | |
2789 | if (!wq->nr_drainers++) | |
2790 | wq->flags |= WQ_DRAINING; | |
2791 | spin_unlock(&workqueue_lock); | |
2792 | reflush: | |
2793 | flush_workqueue(wq); | |
2794 | ||
2795 | for_each_cwq_cpu(cpu, wq) { | |
2796 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
fa2563e4 | 2797 | bool drained; |
9c5a2ba7 | 2798 | |
bd7bdd43 | 2799 | spin_lock_irq(&cwq->pool->gcwq->lock); |
fa2563e4 | 2800 | drained = !cwq->nr_active && list_empty(&cwq->delayed_works); |
bd7bdd43 | 2801 | spin_unlock_irq(&cwq->pool->gcwq->lock); |
fa2563e4 TT |
2802 | |
2803 | if (drained) | |
9c5a2ba7 TH |
2804 | continue; |
2805 | ||
2806 | if (++flush_cnt == 10 || | |
2807 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
044c782c VI |
2808 | pr_warn("workqueue %s: flush on destruction isn't complete after %u tries\n", |
2809 | wq->name, flush_cnt); | |
9c5a2ba7 TH |
2810 | goto reflush; |
2811 | } | |
2812 | ||
2813 | spin_lock(&workqueue_lock); | |
2814 | if (!--wq->nr_drainers) | |
2815 | wq->flags &= ~WQ_DRAINING; | |
2816 | spin_unlock(&workqueue_lock); | |
2817 | } | |
2818 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2819 | ||
606a5020 | 2820 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) |
db700897 | 2821 | { |
affee4b2 | 2822 | struct worker *worker = NULL; |
8b03ae3c | 2823 | struct global_cwq *gcwq; |
db700897 | 2824 | struct cpu_workqueue_struct *cwq; |
db700897 ON |
2825 | |
2826 | might_sleep(); | |
7a22ad75 TH |
2827 | gcwq = get_work_gcwq(work); |
2828 | if (!gcwq) | |
baf59022 | 2829 | return false; |
db700897 | 2830 | |
8b03ae3c | 2831 | spin_lock_irq(&gcwq->lock); |
db700897 ON |
2832 | if (!list_empty(&work->entry)) { |
2833 | /* | |
2834 | * See the comment near try_to_grab_pending()->smp_rmb(). | |
7a22ad75 TH |
2835 | * If it was re-queued to a different gcwq under us, we |
2836 | * are not going to wait. | |
db700897 ON |
2837 | */ |
2838 | smp_rmb(); | |
7a22ad75 | 2839 | cwq = get_work_cwq(work); |
bd7bdd43 | 2840 | if (unlikely(!cwq || gcwq != cwq->pool->gcwq)) |
4690c4ab | 2841 | goto already_gone; |
606a5020 | 2842 | } else { |
7a22ad75 | 2843 | worker = find_worker_executing_work(gcwq, work); |
affee4b2 | 2844 | if (!worker) |
4690c4ab | 2845 | goto already_gone; |
7a22ad75 | 2846 | cwq = worker->current_cwq; |
606a5020 | 2847 | } |
db700897 | 2848 | |
baf59022 | 2849 | insert_wq_barrier(cwq, barr, work, worker); |
8b03ae3c | 2850 | spin_unlock_irq(&gcwq->lock); |
7a22ad75 | 2851 | |
e159489b TH |
2852 | /* |
2853 | * If @max_active is 1 or rescuer is in use, flushing another work | |
2854 | * item on the same workqueue may lead to deadlock. Make sure the | |
2855 | * flusher is not running on the same workqueue by verifying write | |
2856 | * access. | |
2857 | */ | |
2858 | if (cwq->wq->saved_max_active == 1 || cwq->wq->flags & WQ_RESCUER) | |
2859 | lock_map_acquire(&cwq->wq->lockdep_map); | |
2860 | else | |
2861 | lock_map_acquire_read(&cwq->wq->lockdep_map); | |
7a22ad75 | 2862 | lock_map_release(&cwq->wq->lockdep_map); |
e159489b | 2863 | |
401a8d04 | 2864 | return true; |
4690c4ab | 2865 | already_gone: |
8b03ae3c | 2866 | spin_unlock_irq(&gcwq->lock); |
401a8d04 | 2867 | return false; |
db700897 | 2868 | } |
baf59022 TH |
2869 | |
2870 | /** | |
2871 | * flush_work - wait for a work to finish executing the last queueing instance | |
2872 | * @work: the work to flush | |
2873 | * | |
606a5020 TH |
2874 | * Wait until @work has finished execution. @work is guaranteed to be idle |
2875 | * on return if it hasn't been requeued since flush started. | |
baf59022 TH |
2876 | * |
2877 | * RETURNS: | |
2878 | * %true if flush_work() waited for the work to finish execution, | |
2879 | * %false if it was already idle. | |
2880 | */ | |
2881 | bool flush_work(struct work_struct *work) | |
2882 | { | |
2883 | struct wq_barrier barr; | |
2884 | ||
0976dfc1 SB |
2885 | lock_map_acquire(&work->lockdep_map); |
2886 | lock_map_release(&work->lockdep_map); | |
2887 | ||
606a5020 | 2888 | if (start_flush_work(work, &barr)) { |
401a8d04 TH |
2889 | wait_for_completion(&barr.done); |
2890 | destroy_work_on_stack(&barr.work); | |
2891 | return true; | |
606a5020 | 2892 | } else { |
401a8d04 | 2893 | return false; |
6e84d644 | 2894 | } |
6e84d644 | 2895 | } |
606a5020 | 2896 | EXPORT_SYMBOL_GPL(flush_work); |
6e84d644 | 2897 | |
36e227d2 | 2898 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
1f1f642e | 2899 | { |
bbb68dfa | 2900 | unsigned long flags; |
1f1f642e ON |
2901 | int ret; |
2902 | ||
2903 | do { | |
bbb68dfa TH |
2904 | ret = try_to_grab_pending(work, is_dwork, &flags); |
2905 | /* | |
2906 | * If someone else is canceling, wait for the same event it | |
2907 | * would be waiting for before retrying. | |
2908 | */ | |
2909 | if (unlikely(ret == -ENOENT)) | |
606a5020 | 2910 | flush_work(work); |
1f1f642e ON |
2911 | } while (unlikely(ret < 0)); |
2912 | ||
bbb68dfa TH |
2913 | /* tell other tasks trying to grab @work to back off */ |
2914 | mark_work_canceling(work); | |
2915 | local_irq_restore(flags); | |
2916 | ||
606a5020 | 2917 | flush_work(work); |
7a22ad75 | 2918 | clear_work_data(work); |
1f1f642e ON |
2919 | return ret; |
2920 | } | |
2921 | ||
6e84d644 | 2922 | /** |
401a8d04 TH |
2923 | * cancel_work_sync - cancel a work and wait for it to finish |
2924 | * @work: the work to cancel | |
6e84d644 | 2925 | * |
401a8d04 TH |
2926 | * Cancel @work and wait for its execution to finish. This function |
2927 | * can be used even if the work re-queues itself or migrates to | |
2928 | * another workqueue. On return from this function, @work is | |
2929 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2930 | * |
401a8d04 TH |
2931 | * cancel_work_sync(&delayed_work->work) must not be used for |
2932 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2933 | * |
401a8d04 | 2934 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2935 | * queued can't be destroyed before this function returns. |
401a8d04 TH |
2936 | * |
2937 | * RETURNS: | |
2938 | * %true if @work was pending, %false otherwise. | |
6e84d644 | 2939 | */ |
401a8d04 | 2940 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2941 | { |
36e227d2 | 2942 | return __cancel_work_timer(work, false); |
b89deed3 | 2943 | } |
28e53bdd | 2944 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2945 | |
6e84d644 | 2946 | /** |
401a8d04 TH |
2947 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2948 | * @dwork: the delayed work to flush | |
6e84d644 | 2949 | * |
401a8d04 TH |
2950 | * Delayed timer is cancelled and the pending work is queued for |
2951 | * immediate execution. Like flush_work(), this function only | |
2952 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2953 | * |
401a8d04 TH |
2954 | * RETURNS: |
2955 | * %true if flush_work() waited for the work to finish execution, | |
2956 | * %false if it was already idle. | |
6e84d644 | 2957 | */ |
401a8d04 TH |
2958 | bool flush_delayed_work(struct delayed_work *dwork) |
2959 | { | |
8930caba | 2960 | local_irq_disable(); |
401a8d04 | 2961 | if (del_timer_sync(&dwork->timer)) |
1265057f | 2962 | __queue_work(dwork->cpu, |
401a8d04 | 2963 | get_work_cwq(&dwork->work)->wq, &dwork->work); |
8930caba | 2964 | local_irq_enable(); |
401a8d04 TH |
2965 | return flush_work(&dwork->work); |
2966 | } | |
2967 | EXPORT_SYMBOL(flush_delayed_work); | |
2968 | ||
09383498 | 2969 | /** |
57b30ae7 TH |
2970 | * cancel_delayed_work - cancel a delayed work |
2971 | * @dwork: delayed_work to cancel | |
09383498 | 2972 | * |
57b30ae7 TH |
2973 | * Kill off a pending delayed_work. Returns %true if @dwork was pending |
2974 | * and canceled; %false if wasn't pending. Note that the work callback | |
2975 | * function may still be running on return, unless it returns %true and the | |
2976 | * work doesn't re-arm itself. Explicitly flush or use | |
2977 | * cancel_delayed_work_sync() to wait on it. | |
09383498 | 2978 | * |
57b30ae7 | 2979 | * This function is safe to call from any context including IRQ handler. |
09383498 | 2980 | */ |
57b30ae7 | 2981 | bool cancel_delayed_work(struct delayed_work *dwork) |
09383498 | 2982 | { |
57b30ae7 TH |
2983 | unsigned long flags; |
2984 | int ret; | |
2985 | ||
2986 | do { | |
2987 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
2988 | } while (unlikely(ret == -EAGAIN)); | |
2989 | ||
2990 | if (unlikely(ret < 0)) | |
2991 | return false; | |
2992 | ||
2993 | set_work_cpu_and_clear_pending(&dwork->work, work_cpu(&dwork->work)); | |
2994 | local_irq_restore(flags); | |
c0158ca6 | 2995 | return ret; |
09383498 | 2996 | } |
57b30ae7 | 2997 | EXPORT_SYMBOL(cancel_delayed_work); |
09383498 | 2998 | |
401a8d04 TH |
2999 | /** |
3000 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
3001 | * @dwork: the delayed work cancel | |
3002 | * | |
3003 | * This is cancel_work_sync() for delayed works. | |
3004 | * | |
3005 | * RETURNS: | |
3006 | * %true if @dwork was pending, %false otherwise. | |
3007 | */ | |
3008 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 3009 | { |
36e227d2 | 3010 | return __cancel_work_timer(&dwork->work, true); |
6e84d644 | 3011 | } |
f5a421a4 | 3012 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 3013 | |
0fcb78c2 | 3014 | /** |
c1a220e7 ZR |
3015 | * schedule_work_on - put work task on a specific cpu |
3016 | * @cpu: cpu to put the work task on | |
3017 | * @work: job to be done | |
3018 | * | |
3019 | * This puts a job on a specific cpu | |
3020 | */ | |
d4283e93 | 3021 | bool schedule_work_on(int cpu, struct work_struct *work) |
c1a220e7 | 3022 | { |
d320c038 | 3023 | return queue_work_on(cpu, system_wq, work); |
c1a220e7 ZR |
3024 | } |
3025 | EXPORT_SYMBOL(schedule_work_on); | |
3026 | ||
0fcb78c2 | 3027 | /** |
0fcb78c2 REB |
3028 | * schedule_work - put work task in global workqueue |
3029 | * @work: job to be done | |
0fcb78c2 | 3030 | * |
d4283e93 TH |
3031 | * Returns %false if @work was already on the kernel-global workqueue and |
3032 | * %true otherwise. | |
5b0f437d BVA |
3033 | * |
3034 | * This puts a job in the kernel-global workqueue if it was not already | |
3035 | * queued and leaves it in the same position on the kernel-global | |
3036 | * workqueue otherwise. | |
0fcb78c2 | 3037 | */ |
d4283e93 | 3038 | bool schedule_work(struct work_struct *work) |
1da177e4 | 3039 | { |
d320c038 | 3040 | return queue_work(system_wq, work); |
1da177e4 | 3041 | } |
ae90dd5d | 3042 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 3043 | |
0fcb78c2 REB |
3044 | /** |
3045 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
3046 | * @cpu: cpu to use | |
52bad64d | 3047 | * @dwork: job to be done |
0fcb78c2 REB |
3048 | * @delay: number of jiffies to wait |
3049 | * | |
3050 | * After waiting for a given time this puts a job in the kernel-global | |
3051 | * workqueue on the specified CPU. | |
3052 | */ | |
d4283e93 TH |
3053 | bool schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
3054 | unsigned long delay) | |
1da177e4 | 3055 | { |
d320c038 | 3056 | return queue_delayed_work_on(cpu, system_wq, dwork, delay); |
1da177e4 | 3057 | } |
ae90dd5d | 3058 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 3059 | |
0fcb78c2 REB |
3060 | /** |
3061 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
3062 | * @dwork: job to be done |
3063 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
3064 | * |
3065 | * After waiting for a given time this puts a job in the kernel-global | |
3066 | * workqueue. | |
3067 | */ | |
d4283e93 | 3068 | bool schedule_delayed_work(struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 3069 | { |
d320c038 | 3070 | return queue_delayed_work(system_wq, dwork, delay); |
1da177e4 | 3071 | } |
ae90dd5d | 3072 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 3073 | |
b6136773 | 3074 | /** |
31ddd871 | 3075 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 3076 | * @func: the function to call |
b6136773 | 3077 | * |
31ddd871 TH |
3078 | * schedule_on_each_cpu() executes @func on each online CPU using the |
3079 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 3080 | * schedule_on_each_cpu() is very slow. |
31ddd871 TH |
3081 | * |
3082 | * RETURNS: | |
3083 | * 0 on success, -errno on failure. | |
b6136773 | 3084 | */ |
65f27f38 | 3085 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
3086 | { |
3087 | int cpu; | |
38f51568 | 3088 | struct work_struct __percpu *works; |
15316ba8 | 3089 | |
b6136773 AM |
3090 | works = alloc_percpu(struct work_struct); |
3091 | if (!works) | |
15316ba8 | 3092 | return -ENOMEM; |
b6136773 | 3093 | |
93981800 TH |
3094 | get_online_cpus(); |
3095 | ||
15316ba8 | 3096 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
3097 | struct work_struct *work = per_cpu_ptr(works, cpu); |
3098 | ||
3099 | INIT_WORK(work, func); | |
b71ab8c2 | 3100 | schedule_work_on(cpu, work); |
65a64464 | 3101 | } |
93981800 TH |
3102 | |
3103 | for_each_online_cpu(cpu) | |
3104 | flush_work(per_cpu_ptr(works, cpu)); | |
3105 | ||
95402b38 | 3106 | put_online_cpus(); |
b6136773 | 3107 | free_percpu(works); |
15316ba8 CL |
3108 | return 0; |
3109 | } | |
3110 | ||
eef6a7d5 AS |
3111 | /** |
3112 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
3113 | * | |
3114 | * Forces execution of the kernel-global workqueue and blocks until its | |
3115 | * completion. | |
3116 | * | |
3117 | * Think twice before calling this function! It's very easy to get into | |
3118 | * trouble if you don't take great care. Either of the following situations | |
3119 | * will lead to deadlock: | |
3120 | * | |
3121 | * One of the work items currently on the workqueue needs to acquire | |
3122 | * a lock held by your code or its caller. | |
3123 | * | |
3124 | * Your code is running in the context of a work routine. | |
3125 | * | |
3126 | * They will be detected by lockdep when they occur, but the first might not | |
3127 | * occur very often. It depends on what work items are on the workqueue and | |
3128 | * what locks they need, which you have no control over. | |
3129 | * | |
3130 | * In most situations flushing the entire workqueue is overkill; you merely | |
3131 | * need to know that a particular work item isn't queued and isn't running. | |
3132 | * In such cases you should use cancel_delayed_work_sync() or | |
3133 | * cancel_work_sync() instead. | |
3134 | */ | |
1da177e4 LT |
3135 | void flush_scheduled_work(void) |
3136 | { | |
d320c038 | 3137 | flush_workqueue(system_wq); |
1da177e4 | 3138 | } |
ae90dd5d | 3139 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 3140 | |
1fa44eca JB |
3141 | /** |
3142 | * execute_in_process_context - reliably execute the routine with user context | |
3143 | * @fn: the function to execute | |
1fa44eca JB |
3144 | * @ew: guaranteed storage for the execute work structure (must |
3145 | * be available when the work executes) | |
3146 | * | |
3147 | * Executes the function immediately if process context is available, | |
3148 | * otherwise schedules the function for delayed execution. | |
3149 | * | |
3150 | * Returns: 0 - function was executed | |
3151 | * 1 - function was scheduled for execution | |
3152 | */ | |
65f27f38 | 3153 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
3154 | { |
3155 | if (!in_interrupt()) { | |
65f27f38 | 3156 | fn(&ew->work); |
1fa44eca JB |
3157 | return 0; |
3158 | } | |
3159 | ||
65f27f38 | 3160 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
3161 | schedule_work(&ew->work); |
3162 | ||
3163 | return 1; | |
3164 | } | |
3165 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
3166 | ||
1da177e4 LT |
3167 | int keventd_up(void) |
3168 | { | |
d320c038 | 3169 | return system_wq != NULL; |
1da177e4 LT |
3170 | } |
3171 | ||
bdbc5dd7 | 3172 | static int alloc_cwqs(struct workqueue_struct *wq) |
0f900049 | 3173 | { |
65a64464 | 3174 | /* |
0f900049 TH |
3175 | * cwqs are forced aligned according to WORK_STRUCT_FLAG_BITS. |
3176 | * Make sure that the alignment isn't lower than that of | |
3177 | * unsigned long long. | |
65a64464 | 3178 | */ |
0f900049 TH |
3179 | const size_t size = sizeof(struct cpu_workqueue_struct); |
3180 | const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS, | |
3181 | __alignof__(unsigned long long)); | |
65a64464 | 3182 | |
e06ffa1e | 3183 | if (!(wq->flags & WQ_UNBOUND)) |
f3421797 | 3184 | wq->cpu_wq.pcpu = __alloc_percpu(size, align); |
931ac77e | 3185 | else { |
f3421797 TH |
3186 | void *ptr; |
3187 | ||
3188 | /* | |
3189 | * Allocate enough room to align cwq and put an extra | |
3190 | * pointer at the end pointing back to the originally | |
3191 | * allocated pointer which will be used for free. | |
3192 | */ | |
3193 | ptr = kzalloc(size + align + sizeof(void *), GFP_KERNEL); | |
3194 | if (ptr) { | |
3195 | wq->cpu_wq.single = PTR_ALIGN(ptr, align); | |
3196 | *(void **)(wq->cpu_wq.single + 1) = ptr; | |
3197 | } | |
bdbc5dd7 | 3198 | } |
f3421797 | 3199 | |
0415b00d | 3200 | /* just in case, make sure it's actually aligned */ |
bdbc5dd7 TH |
3201 | BUG_ON(!IS_ALIGNED(wq->cpu_wq.v, align)); |
3202 | return wq->cpu_wq.v ? 0 : -ENOMEM; | |
0f900049 TH |
3203 | } |
3204 | ||
bdbc5dd7 | 3205 | static void free_cwqs(struct workqueue_struct *wq) |
0f900049 | 3206 | { |
e06ffa1e | 3207 | if (!(wq->flags & WQ_UNBOUND)) |
f3421797 TH |
3208 | free_percpu(wq->cpu_wq.pcpu); |
3209 | else if (wq->cpu_wq.single) { | |
3210 | /* the pointer to free is stored right after the cwq */ | |
bdbc5dd7 | 3211 | kfree(*(void **)(wq->cpu_wq.single + 1)); |
f3421797 | 3212 | } |
0f900049 TH |
3213 | } |
3214 | ||
f3421797 TH |
3215 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
3216 | const char *name) | |
b71ab8c2 | 3217 | { |
f3421797 TH |
3218 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
3219 | ||
3220 | if (max_active < 1 || max_active > lim) | |
044c782c VI |
3221 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
3222 | max_active, name, 1, lim); | |
b71ab8c2 | 3223 | |
f3421797 | 3224 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
3225 | } |
3226 | ||
b196be89 | 3227 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
d320c038 TH |
3228 | unsigned int flags, |
3229 | int max_active, | |
3230 | struct lock_class_key *key, | |
b196be89 | 3231 | const char *lock_name, ...) |
1da177e4 | 3232 | { |
b196be89 | 3233 | va_list args, args1; |
1da177e4 | 3234 | struct workqueue_struct *wq; |
c34056a3 | 3235 | unsigned int cpu; |
b196be89 TH |
3236 | size_t namelen; |
3237 | ||
3238 | /* determine namelen, allocate wq and format name */ | |
3239 | va_start(args, lock_name); | |
3240 | va_copy(args1, args); | |
3241 | namelen = vsnprintf(NULL, 0, fmt, args) + 1; | |
3242 | ||
3243 | wq = kzalloc(sizeof(*wq) + namelen, GFP_KERNEL); | |
3244 | if (!wq) | |
3245 | goto err; | |
3246 | ||
3247 | vsnprintf(wq->name, namelen, fmt, args1); | |
3248 | va_end(args); | |
3249 | va_end(args1); | |
1da177e4 | 3250 | |
6370a6ad TH |
3251 | /* |
3252 | * Workqueues which may be used during memory reclaim should | |
3253 | * have a rescuer to guarantee forward progress. | |
3254 | */ | |
3255 | if (flags & WQ_MEM_RECLAIM) | |
3256 | flags |= WQ_RESCUER; | |
3257 | ||
d320c038 | 3258 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 3259 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 3260 | |
b196be89 | 3261 | /* init wq */ |
97e37d7b | 3262 | wq->flags = flags; |
a0a1a5fd | 3263 | wq->saved_max_active = max_active; |
73f53c4a TH |
3264 | mutex_init(&wq->flush_mutex); |
3265 | atomic_set(&wq->nr_cwqs_to_flush, 0); | |
3266 | INIT_LIST_HEAD(&wq->flusher_queue); | |
3267 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
502ca9d8 | 3268 | |
eb13ba87 | 3269 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 3270 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 3271 | |
bdbc5dd7 TH |
3272 | if (alloc_cwqs(wq) < 0) |
3273 | goto err; | |
3274 | ||
f3421797 | 3275 | for_each_cwq_cpu(cpu, wq) { |
1537663f | 3276 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
8b03ae3c | 3277 | struct global_cwq *gcwq = get_gcwq(cpu); |
3270476a | 3278 | int pool_idx = (bool)(flags & WQ_HIGHPRI); |
1537663f | 3279 | |
0f900049 | 3280 | BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK); |
3270476a | 3281 | cwq->pool = &gcwq->pools[pool_idx]; |
c34056a3 | 3282 | cwq->wq = wq; |
73f53c4a | 3283 | cwq->flush_color = -1; |
1e19ffc6 | 3284 | cwq->max_active = max_active; |
1e19ffc6 | 3285 | INIT_LIST_HEAD(&cwq->delayed_works); |
e22bee78 | 3286 | } |
1537663f | 3287 | |
e22bee78 TH |
3288 | if (flags & WQ_RESCUER) { |
3289 | struct worker *rescuer; | |
3290 | ||
f2e005aa | 3291 | if (!alloc_mayday_mask(&wq->mayday_mask, GFP_KERNEL)) |
e22bee78 TH |
3292 | goto err; |
3293 | ||
3294 | wq->rescuer = rescuer = alloc_worker(); | |
3295 | if (!rescuer) | |
3296 | goto err; | |
3297 | ||
b196be89 TH |
3298 | rescuer->task = kthread_create(rescuer_thread, wq, "%s", |
3299 | wq->name); | |
e22bee78 TH |
3300 | if (IS_ERR(rescuer->task)) |
3301 | goto err; | |
3302 | ||
e22bee78 TH |
3303 | rescuer->task->flags |= PF_THREAD_BOUND; |
3304 | wake_up_process(rescuer->task); | |
3af24433 ON |
3305 | } |
3306 | ||
a0a1a5fd TH |
3307 | /* |
3308 | * workqueue_lock protects global freeze state and workqueues | |
3309 | * list. Grab it, set max_active accordingly and add the new | |
3310 | * workqueue to workqueues list. | |
3311 | */ | |
1537663f | 3312 | spin_lock(&workqueue_lock); |
a0a1a5fd | 3313 | |
58a69cb4 | 3314 | if (workqueue_freezing && wq->flags & WQ_FREEZABLE) |
f3421797 | 3315 | for_each_cwq_cpu(cpu, wq) |
a0a1a5fd TH |
3316 | get_cwq(cpu, wq)->max_active = 0; |
3317 | ||
1537663f | 3318 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 3319 | |
1537663f TH |
3320 | spin_unlock(&workqueue_lock); |
3321 | ||
3af24433 | 3322 | return wq; |
4690c4ab TH |
3323 | err: |
3324 | if (wq) { | |
bdbc5dd7 | 3325 | free_cwqs(wq); |
f2e005aa | 3326 | free_mayday_mask(wq->mayday_mask); |
e22bee78 | 3327 | kfree(wq->rescuer); |
4690c4ab TH |
3328 | kfree(wq); |
3329 | } | |
3330 | return NULL; | |
3af24433 | 3331 | } |
d320c038 | 3332 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 3333 | |
3af24433 ON |
3334 | /** |
3335 | * destroy_workqueue - safely terminate a workqueue | |
3336 | * @wq: target workqueue | |
3337 | * | |
3338 | * Safely destroy a workqueue. All work currently pending will be done first. | |
3339 | */ | |
3340 | void destroy_workqueue(struct workqueue_struct *wq) | |
3341 | { | |
c8e55f36 | 3342 | unsigned int cpu; |
3af24433 | 3343 | |
9c5a2ba7 TH |
3344 | /* drain it before proceeding with destruction */ |
3345 | drain_workqueue(wq); | |
c8efcc25 | 3346 | |
a0a1a5fd TH |
3347 | /* |
3348 | * wq list is used to freeze wq, remove from list after | |
3349 | * flushing is complete in case freeze races us. | |
3350 | */ | |
95402b38 | 3351 | spin_lock(&workqueue_lock); |
b1f4ec17 | 3352 | list_del(&wq->list); |
95402b38 | 3353 | spin_unlock(&workqueue_lock); |
3af24433 | 3354 | |
e22bee78 | 3355 | /* sanity check */ |
f3421797 | 3356 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a TH |
3357 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3358 | int i; | |
3359 | ||
73f53c4a TH |
3360 | for (i = 0; i < WORK_NR_COLORS; i++) |
3361 | BUG_ON(cwq->nr_in_flight[i]); | |
1e19ffc6 TH |
3362 | BUG_ON(cwq->nr_active); |
3363 | BUG_ON(!list_empty(&cwq->delayed_works)); | |
73f53c4a | 3364 | } |
9b41ea72 | 3365 | |
e22bee78 TH |
3366 | if (wq->flags & WQ_RESCUER) { |
3367 | kthread_stop(wq->rescuer->task); | |
f2e005aa | 3368 | free_mayday_mask(wq->mayday_mask); |
8d9df9f0 | 3369 | kfree(wq->rescuer); |
e22bee78 TH |
3370 | } |
3371 | ||
bdbc5dd7 | 3372 | free_cwqs(wq); |
3af24433 ON |
3373 | kfree(wq); |
3374 | } | |
3375 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
3376 | ||
9f4bd4cd LJ |
3377 | /** |
3378 | * cwq_set_max_active - adjust max_active of a cwq | |
3379 | * @cwq: target cpu_workqueue_struct | |
3380 | * @max_active: new max_active value. | |
3381 | * | |
3382 | * Set @cwq->max_active to @max_active and activate delayed works if | |
3383 | * increased. | |
3384 | * | |
3385 | * CONTEXT: | |
3386 | * spin_lock_irq(gcwq->lock). | |
3387 | */ | |
3388 | static void cwq_set_max_active(struct cpu_workqueue_struct *cwq, int max_active) | |
3389 | { | |
3390 | cwq->max_active = max_active; | |
3391 | ||
3392 | while (!list_empty(&cwq->delayed_works) && | |
3393 | cwq->nr_active < cwq->max_active) | |
3394 | cwq_activate_first_delayed(cwq); | |
3395 | } | |
3396 | ||
dcd989cb TH |
3397 | /** |
3398 | * workqueue_set_max_active - adjust max_active of a workqueue | |
3399 | * @wq: target workqueue | |
3400 | * @max_active: new max_active value. | |
3401 | * | |
3402 | * Set max_active of @wq to @max_active. | |
3403 | * | |
3404 | * CONTEXT: | |
3405 | * Don't call from IRQ context. | |
3406 | */ | |
3407 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
3408 | { | |
3409 | unsigned int cpu; | |
3410 | ||
f3421797 | 3411 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb TH |
3412 | |
3413 | spin_lock(&workqueue_lock); | |
3414 | ||
3415 | wq->saved_max_active = max_active; | |
3416 | ||
f3421797 | 3417 | for_each_cwq_cpu(cpu, wq) { |
dcd989cb TH |
3418 | struct global_cwq *gcwq = get_gcwq(cpu); |
3419 | ||
3420 | spin_lock_irq(&gcwq->lock); | |
3421 | ||
58a69cb4 | 3422 | if (!(wq->flags & WQ_FREEZABLE) || |
dcd989cb | 3423 | !(gcwq->flags & GCWQ_FREEZING)) |
70369b11 | 3424 | cwq_set_max_active(get_cwq(gcwq->cpu, wq), max_active); |
9bfb1839 | 3425 | |
dcd989cb | 3426 | spin_unlock_irq(&gcwq->lock); |
65a64464 | 3427 | } |
93981800 | 3428 | |
dcd989cb | 3429 | spin_unlock(&workqueue_lock); |
15316ba8 | 3430 | } |
dcd989cb | 3431 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 3432 | |
eef6a7d5 | 3433 | /** |
dcd989cb TH |
3434 | * workqueue_congested - test whether a workqueue is congested |
3435 | * @cpu: CPU in question | |
3436 | * @wq: target workqueue | |
eef6a7d5 | 3437 | * |
dcd989cb TH |
3438 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
3439 | * no synchronization around this function and the test result is | |
3440 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 3441 | * |
dcd989cb TH |
3442 | * RETURNS: |
3443 | * %true if congested, %false otherwise. | |
eef6a7d5 | 3444 | */ |
dcd989cb | 3445 | bool workqueue_congested(unsigned int cpu, struct workqueue_struct *wq) |
1da177e4 | 3446 | { |
dcd989cb TH |
3447 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3448 | ||
3449 | return !list_empty(&cwq->delayed_works); | |
1da177e4 | 3450 | } |
dcd989cb | 3451 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 3452 | |
1fa44eca | 3453 | /** |
dcd989cb TH |
3454 | * work_cpu - return the last known associated cpu for @work |
3455 | * @work: the work of interest | |
1fa44eca | 3456 | * |
dcd989cb | 3457 | * RETURNS: |
bdbc5dd7 | 3458 | * CPU number if @work was ever queued. WORK_CPU_NONE otherwise. |
1fa44eca | 3459 | */ |
dcd989cb | 3460 | unsigned int work_cpu(struct work_struct *work) |
1fa44eca | 3461 | { |
dcd989cb | 3462 | struct global_cwq *gcwq = get_work_gcwq(work); |
1fa44eca | 3463 | |
bdbc5dd7 | 3464 | return gcwq ? gcwq->cpu : WORK_CPU_NONE; |
1fa44eca | 3465 | } |
dcd989cb | 3466 | EXPORT_SYMBOL_GPL(work_cpu); |
1fa44eca | 3467 | |
dcd989cb TH |
3468 | /** |
3469 | * work_busy - test whether a work is currently pending or running | |
3470 | * @work: the work to be tested | |
3471 | * | |
3472 | * Test whether @work is currently pending or running. There is no | |
3473 | * synchronization around this function and the test result is | |
3474 | * unreliable and only useful as advisory hints or for debugging. | |
3475 | * Especially for reentrant wqs, the pending state might hide the | |
3476 | * running state. | |
3477 | * | |
3478 | * RETURNS: | |
3479 | * OR'd bitmask of WORK_BUSY_* bits. | |
3480 | */ | |
3481 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 3482 | { |
dcd989cb TH |
3483 | struct global_cwq *gcwq = get_work_gcwq(work); |
3484 | unsigned long flags; | |
3485 | unsigned int ret = 0; | |
1da177e4 | 3486 | |
dcd989cb TH |
3487 | if (!gcwq) |
3488 | return false; | |
1da177e4 | 3489 | |
dcd989cb | 3490 | spin_lock_irqsave(&gcwq->lock, flags); |
1da177e4 | 3491 | |
dcd989cb TH |
3492 | if (work_pending(work)) |
3493 | ret |= WORK_BUSY_PENDING; | |
3494 | if (find_worker_executing_work(gcwq, work)) | |
3495 | ret |= WORK_BUSY_RUNNING; | |
1da177e4 | 3496 | |
dcd989cb | 3497 | spin_unlock_irqrestore(&gcwq->lock, flags); |
1da177e4 | 3498 | |
dcd989cb | 3499 | return ret; |
1da177e4 | 3500 | } |
dcd989cb | 3501 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 3502 | |
db7bccf4 TH |
3503 | /* |
3504 | * CPU hotplug. | |
3505 | * | |
e22bee78 TH |
3506 | * There are two challenges in supporting CPU hotplug. Firstly, there |
3507 | * are a lot of assumptions on strong associations among work, cwq and | |
3508 | * gcwq which make migrating pending and scheduled works very | |
3509 | * difficult to implement without impacting hot paths. Secondly, | |
3510 | * gcwqs serve mix of short, long and very long running works making | |
3511 | * blocked draining impractical. | |
3512 | * | |
628c78e7 TH |
3513 | * This is solved by allowing a gcwq to be disassociated from the CPU |
3514 | * running as an unbound one and allowing it to be reattached later if the | |
3515 | * cpu comes back online. | |
db7bccf4 | 3516 | */ |
1da177e4 | 3517 | |
60373152 | 3518 | /* claim manager positions of all pools */ |
b2eb83d1 | 3519 | static void gcwq_claim_assoc_and_lock(struct global_cwq *gcwq) |
60373152 TH |
3520 | { |
3521 | struct worker_pool *pool; | |
3522 | ||
3523 | for_each_worker_pool(pool, gcwq) | |
b2eb83d1 | 3524 | mutex_lock_nested(&pool->assoc_mutex, pool - gcwq->pools); |
8db25e78 | 3525 | spin_lock_irq(&gcwq->lock); |
60373152 TH |
3526 | } |
3527 | ||
3528 | /* release manager positions */ | |
b2eb83d1 | 3529 | static void gcwq_release_assoc_and_unlock(struct global_cwq *gcwq) |
60373152 TH |
3530 | { |
3531 | struct worker_pool *pool; | |
3532 | ||
8db25e78 | 3533 | spin_unlock_irq(&gcwq->lock); |
60373152 | 3534 | for_each_worker_pool(pool, gcwq) |
b2eb83d1 | 3535 | mutex_unlock(&pool->assoc_mutex); |
60373152 TH |
3536 | } |
3537 | ||
628c78e7 | 3538 | static void gcwq_unbind_fn(struct work_struct *work) |
3af24433 | 3539 | { |
628c78e7 | 3540 | struct global_cwq *gcwq = get_gcwq(smp_processor_id()); |
4ce62e9e | 3541 | struct worker_pool *pool; |
db7bccf4 TH |
3542 | struct worker *worker; |
3543 | struct hlist_node *pos; | |
3544 | int i; | |
3af24433 | 3545 | |
db7bccf4 TH |
3546 | BUG_ON(gcwq->cpu != smp_processor_id()); |
3547 | ||
b2eb83d1 | 3548 | gcwq_claim_assoc_and_lock(gcwq); |
3af24433 | 3549 | |
f2d5a0ee TH |
3550 | /* |
3551 | * We've claimed all manager positions. Make all workers unbound | |
3552 | * and set DISASSOCIATED. Before this, all workers except for the | |
3553 | * ones which are still executing works from before the last CPU | |
3554 | * down must be on the cpu. After this, they may become diasporas. | |
3555 | */ | |
60373152 | 3556 | for_each_worker_pool(pool, gcwq) |
4ce62e9e | 3557 | list_for_each_entry(worker, &pool->idle_list, entry) |
403c821d | 3558 | worker->flags |= WORKER_UNBOUND; |
3af24433 | 3559 | |
db7bccf4 | 3560 | for_each_busy_worker(worker, i, pos, gcwq) |
403c821d | 3561 | worker->flags |= WORKER_UNBOUND; |
06ba38a9 | 3562 | |
f2d5a0ee TH |
3563 | gcwq->flags |= GCWQ_DISASSOCIATED; |
3564 | ||
b2eb83d1 | 3565 | gcwq_release_assoc_and_unlock(gcwq); |
628c78e7 | 3566 | |
e22bee78 | 3567 | /* |
403c821d | 3568 | * Call schedule() so that we cross rq->lock and thus can guarantee |
628c78e7 TH |
3569 | * sched callbacks see the %WORKER_UNBOUND flag. This is necessary |
3570 | * as scheduler callbacks may be invoked from other cpus. | |
e22bee78 | 3571 | */ |
e22bee78 | 3572 | schedule(); |
06ba38a9 | 3573 | |
e22bee78 | 3574 | /* |
628c78e7 TH |
3575 | * Sched callbacks are disabled now. Zap nr_running. After this, |
3576 | * nr_running stays zero and need_more_worker() and keep_working() | |
3577 | * are always true as long as the worklist is not empty. @gcwq now | |
3578 | * behaves as unbound (in terms of concurrency management) gcwq | |
3579 | * which is served by workers tied to the CPU. | |
3580 | * | |
3581 | * On return from this function, the current worker would trigger | |
3582 | * unbound chain execution of pending work items if other workers | |
3583 | * didn't already. | |
e22bee78 | 3584 | */ |
4ce62e9e TH |
3585 | for_each_worker_pool(pool, gcwq) |
3586 | atomic_set(get_pool_nr_running(pool), 0); | |
3af24433 | 3587 | } |
3af24433 | 3588 | |
8db25e78 TH |
3589 | /* |
3590 | * Workqueues should be brought up before normal priority CPU notifiers. | |
3591 | * This will be registered high priority CPU notifier. | |
3592 | */ | |
9fdf9b73 | 3593 | static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb, |
8db25e78 TH |
3594 | unsigned long action, |
3595 | void *hcpu) | |
3af24433 ON |
3596 | { |
3597 | unsigned int cpu = (unsigned long)hcpu; | |
db7bccf4 | 3598 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3599 | struct worker_pool *pool; |
3ce63377 | 3600 | |
8db25e78 | 3601 | switch (action & ~CPU_TASKS_FROZEN) { |
3af24433 | 3602 | case CPU_UP_PREPARE: |
4ce62e9e | 3603 | for_each_worker_pool(pool, gcwq) { |
3ce63377 TH |
3604 | struct worker *worker; |
3605 | ||
3606 | if (pool->nr_workers) | |
3607 | continue; | |
3608 | ||
3609 | worker = create_worker(pool); | |
3610 | if (!worker) | |
3611 | return NOTIFY_BAD; | |
3612 | ||
3613 | spin_lock_irq(&gcwq->lock); | |
3614 | start_worker(worker); | |
3615 | spin_unlock_irq(&gcwq->lock); | |
3af24433 | 3616 | } |
8db25e78 | 3617 | break; |
3af24433 | 3618 | |
db7bccf4 TH |
3619 | case CPU_DOWN_FAILED: |
3620 | case CPU_ONLINE: | |
b2eb83d1 | 3621 | gcwq_claim_assoc_and_lock(gcwq); |
bc2ae0f5 | 3622 | gcwq->flags &= ~GCWQ_DISASSOCIATED; |
25511a47 | 3623 | rebind_workers(gcwq); |
b2eb83d1 | 3624 | gcwq_release_assoc_and_unlock(gcwq); |
db7bccf4 | 3625 | break; |
00dfcaf7 | 3626 | } |
65758202 TH |
3627 | return NOTIFY_OK; |
3628 | } | |
3629 | ||
3630 | /* | |
3631 | * Workqueues should be brought down after normal priority CPU notifiers. | |
3632 | * This will be registered as low priority CPU notifier. | |
3633 | */ | |
9fdf9b73 | 3634 | static int __cpuinit workqueue_cpu_down_callback(struct notifier_block *nfb, |
65758202 TH |
3635 | unsigned long action, |
3636 | void *hcpu) | |
3637 | { | |
8db25e78 TH |
3638 | unsigned int cpu = (unsigned long)hcpu; |
3639 | struct work_struct unbind_work; | |
3640 | ||
65758202 TH |
3641 | switch (action & ~CPU_TASKS_FROZEN) { |
3642 | case CPU_DOWN_PREPARE: | |
8db25e78 TH |
3643 | /* unbinding should happen on the local CPU */ |
3644 | INIT_WORK_ONSTACK(&unbind_work, gcwq_unbind_fn); | |
7635d2fd | 3645 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
8db25e78 TH |
3646 | flush_work(&unbind_work); |
3647 | break; | |
65758202 TH |
3648 | } |
3649 | return NOTIFY_OK; | |
3650 | } | |
3651 | ||
2d3854a3 | 3652 | #ifdef CONFIG_SMP |
8ccad40d | 3653 | |
2d3854a3 | 3654 | struct work_for_cpu { |
ed48ece2 | 3655 | struct work_struct work; |
2d3854a3 RR |
3656 | long (*fn)(void *); |
3657 | void *arg; | |
3658 | long ret; | |
3659 | }; | |
3660 | ||
ed48ece2 | 3661 | static void work_for_cpu_fn(struct work_struct *work) |
2d3854a3 | 3662 | { |
ed48ece2 TH |
3663 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
3664 | ||
2d3854a3 RR |
3665 | wfc->ret = wfc->fn(wfc->arg); |
3666 | } | |
3667 | ||
3668 | /** | |
3669 | * work_on_cpu - run a function in user context on a particular cpu | |
3670 | * @cpu: the cpu to run on | |
3671 | * @fn: the function to run | |
3672 | * @arg: the function arg | |
3673 | * | |
31ad9081 RR |
3674 | * This will return the value @fn returns. |
3675 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 3676 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 RR |
3677 | */ |
3678 | long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) | |
3679 | { | |
ed48ece2 | 3680 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; |
6b44003e | 3681 | |
ed48ece2 TH |
3682 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
3683 | schedule_work_on(cpu, &wfc.work); | |
3684 | flush_work(&wfc.work); | |
2d3854a3 RR |
3685 | return wfc.ret; |
3686 | } | |
3687 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
3688 | #endif /* CONFIG_SMP */ | |
3689 | ||
a0a1a5fd TH |
3690 | #ifdef CONFIG_FREEZER |
3691 | ||
3692 | /** | |
3693 | * freeze_workqueues_begin - begin freezing workqueues | |
3694 | * | |
58a69cb4 TH |
3695 | * Start freezing workqueues. After this function returns, all freezable |
3696 | * workqueues will queue new works to their frozen_works list instead of | |
3697 | * gcwq->worklist. | |
a0a1a5fd TH |
3698 | * |
3699 | * CONTEXT: | |
8b03ae3c | 3700 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3701 | */ |
3702 | void freeze_workqueues_begin(void) | |
3703 | { | |
a0a1a5fd TH |
3704 | unsigned int cpu; |
3705 | ||
3706 | spin_lock(&workqueue_lock); | |
3707 | ||
3708 | BUG_ON(workqueue_freezing); | |
3709 | workqueue_freezing = true; | |
3710 | ||
f3421797 | 3711 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3712 | struct global_cwq *gcwq = get_gcwq(cpu); |
bdbc5dd7 | 3713 | struct workqueue_struct *wq; |
8b03ae3c TH |
3714 | |
3715 | spin_lock_irq(&gcwq->lock); | |
3716 | ||
db7bccf4 TH |
3717 | BUG_ON(gcwq->flags & GCWQ_FREEZING); |
3718 | gcwq->flags |= GCWQ_FREEZING; | |
3719 | ||
a0a1a5fd TH |
3720 | list_for_each_entry(wq, &workqueues, list) { |
3721 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3722 | ||
58a69cb4 | 3723 | if (cwq && wq->flags & WQ_FREEZABLE) |
a0a1a5fd | 3724 | cwq->max_active = 0; |
a0a1a5fd | 3725 | } |
8b03ae3c TH |
3726 | |
3727 | spin_unlock_irq(&gcwq->lock); | |
a0a1a5fd TH |
3728 | } |
3729 | ||
3730 | spin_unlock(&workqueue_lock); | |
3731 | } | |
3732 | ||
3733 | /** | |
58a69cb4 | 3734 | * freeze_workqueues_busy - are freezable workqueues still busy? |
a0a1a5fd TH |
3735 | * |
3736 | * Check whether freezing is complete. This function must be called | |
3737 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
3738 | * | |
3739 | * CONTEXT: | |
3740 | * Grabs and releases workqueue_lock. | |
3741 | * | |
3742 | * RETURNS: | |
58a69cb4 TH |
3743 | * %true if some freezable workqueues are still busy. %false if freezing |
3744 | * is complete. | |
a0a1a5fd TH |
3745 | */ |
3746 | bool freeze_workqueues_busy(void) | |
3747 | { | |
a0a1a5fd TH |
3748 | unsigned int cpu; |
3749 | bool busy = false; | |
3750 | ||
3751 | spin_lock(&workqueue_lock); | |
3752 | ||
3753 | BUG_ON(!workqueue_freezing); | |
3754 | ||
f3421797 | 3755 | for_each_gcwq_cpu(cpu) { |
bdbc5dd7 | 3756 | struct workqueue_struct *wq; |
a0a1a5fd TH |
3757 | /* |
3758 | * nr_active is monotonically decreasing. It's safe | |
3759 | * to peek without lock. | |
3760 | */ | |
3761 | list_for_each_entry(wq, &workqueues, list) { | |
3762 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3763 | ||
58a69cb4 | 3764 | if (!cwq || !(wq->flags & WQ_FREEZABLE)) |
a0a1a5fd TH |
3765 | continue; |
3766 | ||
3767 | BUG_ON(cwq->nr_active < 0); | |
3768 | if (cwq->nr_active) { | |
3769 | busy = true; | |
3770 | goto out_unlock; | |
3771 | } | |
3772 | } | |
3773 | } | |
3774 | out_unlock: | |
3775 | spin_unlock(&workqueue_lock); | |
3776 | return busy; | |
3777 | } | |
3778 | ||
3779 | /** | |
3780 | * thaw_workqueues - thaw workqueues | |
3781 | * | |
3782 | * Thaw workqueues. Normal queueing is restored and all collected | |
7e11629d | 3783 | * frozen works are transferred to their respective gcwq worklists. |
a0a1a5fd TH |
3784 | * |
3785 | * CONTEXT: | |
8b03ae3c | 3786 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3787 | */ |
3788 | void thaw_workqueues(void) | |
3789 | { | |
a0a1a5fd TH |
3790 | unsigned int cpu; |
3791 | ||
3792 | spin_lock(&workqueue_lock); | |
3793 | ||
3794 | if (!workqueue_freezing) | |
3795 | goto out_unlock; | |
3796 | ||
f3421797 | 3797 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3798 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3799 | struct worker_pool *pool; |
bdbc5dd7 | 3800 | struct workqueue_struct *wq; |
8b03ae3c TH |
3801 | |
3802 | spin_lock_irq(&gcwq->lock); | |
3803 | ||
db7bccf4 TH |
3804 | BUG_ON(!(gcwq->flags & GCWQ_FREEZING)); |
3805 | gcwq->flags &= ~GCWQ_FREEZING; | |
3806 | ||
a0a1a5fd TH |
3807 | list_for_each_entry(wq, &workqueues, list) { |
3808 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3809 | ||
58a69cb4 | 3810 | if (!cwq || !(wq->flags & WQ_FREEZABLE)) |
a0a1a5fd TH |
3811 | continue; |
3812 | ||
a0a1a5fd | 3813 | /* restore max_active and repopulate worklist */ |
9f4bd4cd | 3814 | cwq_set_max_active(cwq, wq->saved_max_active); |
a0a1a5fd | 3815 | } |
8b03ae3c | 3816 | |
4ce62e9e TH |
3817 | for_each_worker_pool(pool, gcwq) |
3818 | wake_up_worker(pool); | |
e22bee78 | 3819 | |
8b03ae3c | 3820 | spin_unlock_irq(&gcwq->lock); |
a0a1a5fd TH |
3821 | } |
3822 | ||
3823 | workqueue_freezing = false; | |
3824 | out_unlock: | |
3825 | spin_unlock(&workqueue_lock); | |
3826 | } | |
3827 | #endif /* CONFIG_FREEZER */ | |
3828 | ||
6ee0578b | 3829 | static int __init init_workqueues(void) |
1da177e4 | 3830 | { |
c34056a3 | 3831 | unsigned int cpu; |
c8e55f36 | 3832 | int i; |
c34056a3 | 3833 | |
b5490077 TH |
3834 | /* make sure we have enough bits for OFFQ CPU number */ |
3835 | BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_CPU_SHIFT)) < | |
3836 | WORK_CPU_LAST); | |
3837 | ||
65758202 | 3838 | cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); |
a5b4e57d | 3839 | hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); |
8b03ae3c TH |
3840 | |
3841 | /* initialize gcwqs */ | |
f3421797 | 3842 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3843 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3844 | struct worker_pool *pool; |
8b03ae3c TH |
3845 | |
3846 | spin_lock_init(&gcwq->lock); | |
3847 | gcwq->cpu = cpu; | |
477a3c33 | 3848 | gcwq->flags |= GCWQ_DISASSOCIATED; |
8b03ae3c | 3849 | |
c8e55f36 TH |
3850 | for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++) |
3851 | INIT_HLIST_HEAD(&gcwq->busy_hash[i]); | |
3852 | ||
4ce62e9e TH |
3853 | for_each_worker_pool(pool, gcwq) { |
3854 | pool->gcwq = gcwq; | |
3855 | INIT_LIST_HEAD(&pool->worklist); | |
3856 | INIT_LIST_HEAD(&pool->idle_list); | |
e7577c50 | 3857 | |
4ce62e9e TH |
3858 | init_timer_deferrable(&pool->idle_timer); |
3859 | pool->idle_timer.function = idle_worker_timeout; | |
3860 | pool->idle_timer.data = (unsigned long)pool; | |
e22bee78 | 3861 | |
4ce62e9e TH |
3862 | setup_timer(&pool->mayday_timer, gcwq_mayday_timeout, |
3863 | (unsigned long)pool); | |
3864 | ||
b2eb83d1 | 3865 | mutex_init(&pool->assoc_mutex); |
4ce62e9e TH |
3866 | ida_init(&pool->worker_ida); |
3867 | } | |
8b03ae3c TH |
3868 | } |
3869 | ||
e22bee78 | 3870 | /* create the initial worker */ |
f3421797 | 3871 | for_each_online_gcwq_cpu(cpu) { |
e22bee78 | 3872 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3873 | struct worker_pool *pool; |
e22bee78 | 3874 | |
477a3c33 TH |
3875 | if (cpu != WORK_CPU_UNBOUND) |
3876 | gcwq->flags &= ~GCWQ_DISASSOCIATED; | |
4ce62e9e TH |
3877 | |
3878 | for_each_worker_pool(pool, gcwq) { | |
3879 | struct worker *worker; | |
3880 | ||
bc2ae0f5 | 3881 | worker = create_worker(pool); |
4ce62e9e TH |
3882 | BUG_ON(!worker); |
3883 | spin_lock_irq(&gcwq->lock); | |
3884 | start_worker(worker); | |
3885 | spin_unlock_irq(&gcwq->lock); | |
3886 | } | |
e22bee78 TH |
3887 | } |
3888 | ||
d320c038 | 3889 | system_wq = alloc_workqueue("events", 0, 0); |
1aabe902 | 3890 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
d320c038 | 3891 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
f3421797 TH |
3892 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
3893 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
3894 | system_freezable_wq = alloc_workqueue("events_freezable", |
3895 | WQ_FREEZABLE, 0); | |
1aabe902 | 3896 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
ae930e0f | 3897 | !system_unbound_wq || !system_freezable_wq); |
6ee0578b | 3898 | return 0; |
1da177e4 | 3899 | } |
6ee0578b | 3900 | early_initcall(init_workqueues); |