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> |
42f8570f | 44 | #include <linux/hashtable.h> |
e22bee78 | 45 | |
ea138446 | 46 | #include "workqueue_internal.h" |
1da177e4 | 47 | |
c8e55f36 | 48 | enum { |
bc2ae0f5 TH |
49 | /* |
50 | * global_cwq flags | |
24647570 TH |
51 | */ |
52 | GCWQ_FREEZING = 1 << 1, /* freeze in progress */ | |
53 | ||
54 | /* | |
55 | * worker_pool flags | |
bc2ae0f5 | 56 | * |
24647570 | 57 | * A bound pool is either associated or disassociated with its CPU. |
bc2ae0f5 TH |
58 | * While associated (!DISASSOCIATED), all workers are bound to the |
59 | * CPU and none has %WORKER_UNBOUND set and concurrency management | |
60 | * is in effect. | |
61 | * | |
62 | * While DISASSOCIATED, the cpu may be offline and all workers have | |
63 | * %WORKER_UNBOUND set and concurrency management disabled, and may | |
24647570 | 64 | * be executing on any CPU. The pool behaves as an unbound one. |
bc2ae0f5 TH |
65 | * |
66 | * Note that DISASSOCIATED can be flipped only while holding | |
24647570 TH |
67 | * assoc_mutex to avoid changing binding state while |
68 | * create_worker() is in progress. | |
bc2ae0f5 | 69 | */ |
11ebea50 | 70 | POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */ |
552a37e9 | 71 | POOL_MANAGING_WORKERS = 1 << 1, /* managing workers */ |
24647570 | 72 | POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ |
db7bccf4 | 73 | |
c8e55f36 TH |
74 | /* worker flags */ |
75 | WORKER_STARTED = 1 << 0, /* started */ | |
76 | WORKER_DIE = 1 << 1, /* die die die */ | |
77 | WORKER_IDLE = 1 << 2, /* is idle */ | |
e22bee78 | 78 | WORKER_PREP = 1 << 3, /* preparing to run works */ |
fb0e7beb | 79 | WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ |
f3421797 | 80 | WORKER_UNBOUND = 1 << 7, /* worker is unbound */ |
e22bee78 | 81 | |
5f7dabfd | 82 | WORKER_NOT_RUNNING = WORKER_PREP | WORKER_UNBOUND | |
403c821d | 83 | WORKER_CPU_INTENSIVE, |
db7bccf4 | 84 | |
e34cdddb | 85 | NR_STD_WORKER_POOLS = 2, /* # standard pools per cpu */ |
4ce62e9e | 86 | |
c8e55f36 | 87 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ |
db7bccf4 | 88 | |
e22bee78 TH |
89 | MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */ |
90 | IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */ | |
91 | ||
3233cdbd TH |
92 | MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2, |
93 | /* call for help after 10ms | |
94 | (min two ticks) */ | |
e22bee78 TH |
95 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ |
96 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
e22bee78 TH |
97 | |
98 | /* | |
99 | * Rescue workers are used only on emergencies and shared by | |
100 | * all cpus. Give -20. | |
101 | */ | |
102 | RESCUER_NICE_LEVEL = -20, | |
3270476a | 103 | HIGHPRI_NICE_LEVEL = -20, |
c8e55f36 | 104 | }; |
1da177e4 LT |
105 | |
106 | /* | |
4690c4ab TH |
107 | * Structure fields follow one of the following exclusion rules. |
108 | * | |
e41e704b TH |
109 | * I: Modifiable by initialization/destruction paths and read-only for |
110 | * everyone else. | |
4690c4ab | 111 | * |
e22bee78 TH |
112 | * P: Preemption protected. Disabling preemption is enough and should |
113 | * only be modified and accessed from the local cpu. | |
114 | * | |
8b03ae3c | 115 | * L: gcwq->lock protected. Access with gcwq->lock held. |
4690c4ab | 116 | * |
e22bee78 TH |
117 | * X: During normal operation, modification requires gcwq->lock and |
118 | * should be done only from local cpu. Either disabling preemption | |
119 | * on local cpu or grabbing gcwq->lock is enough for read access. | |
24647570 | 120 | * If POOL_DISASSOCIATED is set, it's identical to L. |
e22bee78 | 121 | * |
73f53c4a TH |
122 | * F: wq->flush_mutex protected. |
123 | * | |
4690c4ab | 124 | * W: workqueue_lock protected. |
1da177e4 | 125 | */ |
1da177e4 | 126 | |
2eaebdb3 | 127 | /* struct worker is defined in workqueue_internal.h */ |
c34056a3 | 128 | |
bd7bdd43 TH |
129 | struct worker_pool { |
130 | struct global_cwq *gcwq; /* I: the owning gcwq */ | |
11ebea50 | 131 | unsigned int flags; /* X: flags */ |
bd7bdd43 TH |
132 | |
133 | struct list_head worklist; /* L: list of pending works */ | |
134 | int nr_workers; /* L: total number of workers */ | |
ea1abd61 LJ |
135 | |
136 | /* nr_idle includes the ones off idle_list for rebinding */ | |
bd7bdd43 TH |
137 | int nr_idle; /* L: currently idle ones */ |
138 | ||
139 | struct list_head idle_list; /* X: list of idle workers */ | |
140 | struct timer_list idle_timer; /* L: worker idle timeout */ | |
141 | struct timer_list mayday_timer; /* L: SOS timer for workers */ | |
142 | ||
24647570 | 143 | struct mutex assoc_mutex; /* protect POOL_DISASSOCIATED */ |
bd7bdd43 | 144 | struct ida worker_ida; /* L: for worker IDs */ |
bd7bdd43 TH |
145 | }; |
146 | ||
8b03ae3c | 147 | /* |
e22bee78 TH |
148 | * Global per-cpu workqueue. There's one and only one for each cpu |
149 | * and all works are queued and processed here regardless of their | |
150 | * target workqueues. | |
8b03ae3c TH |
151 | */ |
152 | struct global_cwq { | |
153 | spinlock_t lock; /* the gcwq lock */ | |
154 | unsigned int cpu; /* I: the associated cpu */ | |
db7bccf4 | 155 | unsigned int flags; /* L: GCWQ_* flags */ |
c8e55f36 | 156 | |
bd7bdd43 | 157 | /* workers are chained either in busy_hash or pool idle_list */ |
42f8570f | 158 | DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER); |
c8e55f36 TH |
159 | /* L: hash of busy workers */ |
160 | ||
e34cdddb | 161 | struct worker_pool pools[NR_STD_WORKER_POOLS]; |
330dad5b | 162 | /* normal and highpri pools */ |
8b03ae3c TH |
163 | } ____cacheline_aligned_in_smp; |
164 | ||
1da177e4 | 165 | /* |
502ca9d8 | 166 | * The per-CPU workqueue. The lower WORK_STRUCT_FLAG_BITS of |
0f900049 TH |
167 | * work_struct->data are used for flags and thus cwqs need to be |
168 | * aligned at two's power of the number of flag bits. | |
1da177e4 LT |
169 | */ |
170 | struct cpu_workqueue_struct { | |
bd7bdd43 | 171 | struct worker_pool *pool; /* I: the associated pool */ |
4690c4ab | 172 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
173 | int work_color; /* L: current color */ |
174 | int flush_color; /* L: flushing color */ | |
175 | int nr_in_flight[WORK_NR_COLORS]; | |
176 | /* L: nr of in_flight works */ | |
1e19ffc6 | 177 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 178 | int max_active; /* L: max active works */ |
1e19ffc6 | 179 | struct list_head delayed_works; /* L: delayed works */ |
0f900049 | 180 | }; |
1da177e4 | 181 | |
73f53c4a TH |
182 | /* |
183 | * Structure used to wait for workqueue flush. | |
184 | */ | |
185 | struct wq_flusher { | |
186 | struct list_head list; /* F: list of flushers */ | |
187 | int flush_color; /* F: flush color waiting for */ | |
188 | struct completion done; /* flush completion */ | |
189 | }; | |
190 | ||
f2e005aa TH |
191 | /* |
192 | * All cpumasks are assumed to be always set on UP and thus can't be | |
193 | * used to determine whether there's something to be done. | |
194 | */ | |
195 | #ifdef CONFIG_SMP | |
196 | typedef cpumask_var_t mayday_mask_t; | |
197 | #define mayday_test_and_set_cpu(cpu, mask) \ | |
198 | cpumask_test_and_set_cpu((cpu), (mask)) | |
199 | #define mayday_clear_cpu(cpu, mask) cpumask_clear_cpu((cpu), (mask)) | |
200 | #define for_each_mayday_cpu(cpu, mask) for_each_cpu((cpu), (mask)) | |
9c37547a | 201 | #define alloc_mayday_mask(maskp, gfp) zalloc_cpumask_var((maskp), (gfp)) |
f2e005aa TH |
202 | #define free_mayday_mask(mask) free_cpumask_var((mask)) |
203 | #else | |
204 | typedef unsigned long mayday_mask_t; | |
205 | #define mayday_test_and_set_cpu(cpu, mask) test_and_set_bit(0, &(mask)) | |
206 | #define mayday_clear_cpu(cpu, mask) clear_bit(0, &(mask)) | |
207 | #define for_each_mayday_cpu(cpu, mask) if ((cpu) = 0, (mask)) | |
208 | #define alloc_mayday_mask(maskp, gfp) true | |
209 | #define free_mayday_mask(mask) do { } while (0) | |
210 | #endif | |
1da177e4 LT |
211 | |
212 | /* | |
213 | * The externally visible workqueue abstraction is an array of | |
214 | * per-CPU workqueues: | |
215 | */ | |
216 | struct workqueue_struct { | |
9c5a2ba7 | 217 | unsigned int flags; /* W: WQ_* flags */ |
bdbc5dd7 TH |
218 | union { |
219 | struct cpu_workqueue_struct __percpu *pcpu; | |
220 | struct cpu_workqueue_struct *single; | |
221 | unsigned long v; | |
222 | } cpu_wq; /* I: cwq's */ | |
4690c4ab | 223 | struct list_head list; /* W: list of all workqueues */ |
73f53c4a TH |
224 | |
225 | struct mutex flush_mutex; /* protects wq flushing */ | |
226 | int work_color; /* F: current work color */ | |
227 | int flush_color; /* F: current flush color */ | |
228 | atomic_t nr_cwqs_to_flush; /* flush in progress */ | |
229 | struct wq_flusher *first_flusher; /* F: first flusher */ | |
230 | struct list_head flusher_queue; /* F: flush waiters */ | |
231 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
232 | ||
f2e005aa | 233 | mayday_mask_t mayday_mask; /* cpus requesting rescue */ |
e22bee78 TH |
234 | struct worker *rescuer; /* I: rescue worker */ |
235 | ||
9c5a2ba7 | 236 | int nr_drainers; /* W: drain in progress */ |
dcd989cb | 237 | int saved_max_active; /* W: saved cwq max_active */ |
4e6045f1 | 238 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 239 | struct lockdep_map lockdep_map; |
4e6045f1 | 240 | #endif |
b196be89 | 241 | char name[]; /* I: workqueue name */ |
1da177e4 LT |
242 | }; |
243 | ||
d320c038 | 244 | struct workqueue_struct *system_wq __read_mostly; |
d320c038 | 245 | EXPORT_SYMBOL_GPL(system_wq); |
044c782c | 246 | struct workqueue_struct *system_highpri_wq __read_mostly; |
1aabe902 | 247 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
044c782c | 248 | struct workqueue_struct *system_long_wq __read_mostly; |
d320c038 | 249 | EXPORT_SYMBOL_GPL(system_long_wq); |
044c782c | 250 | struct workqueue_struct *system_unbound_wq __read_mostly; |
f3421797 | 251 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
044c782c | 252 | struct workqueue_struct *system_freezable_wq __read_mostly; |
24d51add | 253 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
d320c038 | 254 | |
97bd2347 TH |
255 | #define CREATE_TRACE_POINTS |
256 | #include <trace/events/workqueue.h> | |
257 | ||
4ce62e9e | 258 | #define for_each_worker_pool(pool, gcwq) \ |
3270476a | 259 | for ((pool) = &(gcwq)->pools[0]; \ |
e34cdddb | 260 | (pool) < &(gcwq)->pools[NR_STD_WORKER_POOLS]; (pool)++) |
4ce62e9e | 261 | |
db7bccf4 | 262 | #define for_each_busy_worker(worker, i, pos, gcwq) \ |
42f8570f | 263 | hash_for_each(gcwq->busy_hash, i, pos, worker, hentry) |
db7bccf4 | 264 | |
f3421797 TH |
265 | static inline int __next_gcwq_cpu(int cpu, const struct cpumask *mask, |
266 | unsigned int sw) | |
267 | { | |
268 | if (cpu < nr_cpu_ids) { | |
269 | if (sw & 1) { | |
270 | cpu = cpumask_next(cpu, mask); | |
271 | if (cpu < nr_cpu_ids) | |
272 | return cpu; | |
273 | } | |
274 | if (sw & 2) | |
275 | return WORK_CPU_UNBOUND; | |
276 | } | |
277 | return WORK_CPU_NONE; | |
278 | } | |
279 | ||
280 | static inline int __next_wq_cpu(int cpu, const struct cpumask *mask, | |
281 | struct workqueue_struct *wq) | |
282 | { | |
283 | return __next_gcwq_cpu(cpu, mask, !(wq->flags & WQ_UNBOUND) ? 1 : 2); | |
284 | } | |
285 | ||
09884951 TH |
286 | /* |
287 | * CPU iterators | |
288 | * | |
289 | * An extra gcwq is defined for an invalid cpu number | |
290 | * (WORK_CPU_UNBOUND) to host workqueues which are not bound to any | |
291 | * specific CPU. The following iterators are similar to | |
292 | * for_each_*_cpu() iterators but also considers the unbound gcwq. | |
293 | * | |
294 | * for_each_gcwq_cpu() : possible CPUs + WORK_CPU_UNBOUND | |
295 | * for_each_online_gcwq_cpu() : online CPUs + WORK_CPU_UNBOUND | |
296 | * for_each_cwq_cpu() : possible CPUs for bound workqueues, | |
297 | * WORK_CPU_UNBOUND for unbound workqueues | |
298 | */ | |
f3421797 TH |
299 | #define for_each_gcwq_cpu(cpu) \ |
300 | for ((cpu) = __next_gcwq_cpu(-1, cpu_possible_mask, 3); \ | |
301 | (cpu) < WORK_CPU_NONE; \ | |
302 | (cpu) = __next_gcwq_cpu((cpu), cpu_possible_mask, 3)) | |
303 | ||
304 | #define for_each_online_gcwq_cpu(cpu) \ | |
305 | for ((cpu) = __next_gcwq_cpu(-1, cpu_online_mask, 3); \ | |
306 | (cpu) < WORK_CPU_NONE; \ | |
307 | (cpu) = __next_gcwq_cpu((cpu), cpu_online_mask, 3)) | |
308 | ||
309 | #define for_each_cwq_cpu(cpu, wq) \ | |
310 | for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, (wq)); \ | |
311 | (cpu) < WORK_CPU_NONE; \ | |
312 | (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, (wq))) | |
313 | ||
dc186ad7 TG |
314 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
315 | ||
316 | static struct debug_obj_descr work_debug_descr; | |
317 | ||
99777288 SG |
318 | static void *work_debug_hint(void *addr) |
319 | { | |
320 | return ((struct work_struct *) addr)->func; | |
321 | } | |
322 | ||
dc186ad7 TG |
323 | /* |
324 | * fixup_init is called when: | |
325 | * - an active object is initialized | |
326 | */ | |
327 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
328 | { | |
329 | struct work_struct *work = addr; | |
330 | ||
331 | switch (state) { | |
332 | case ODEBUG_STATE_ACTIVE: | |
333 | cancel_work_sync(work); | |
334 | debug_object_init(work, &work_debug_descr); | |
335 | return 1; | |
336 | default: | |
337 | return 0; | |
338 | } | |
339 | } | |
340 | ||
341 | /* | |
342 | * fixup_activate is called when: | |
343 | * - an active object is activated | |
344 | * - an unknown object is activated (might be a statically initialized object) | |
345 | */ | |
346 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
347 | { | |
348 | struct work_struct *work = addr; | |
349 | ||
350 | switch (state) { | |
351 | ||
352 | case ODEBUG_STATE_NOTAVAILABLE: | |
353 | /* | |
354 | * This is not really a fixup. The work struct was | |
355 | * statically initialized. We just make sure that it | |
356 | * is tracked in the object tracker. | |
357 | */ | |
22df02bb | 358 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
359 | debug_object_init(work, &work_debug_descr); |
360 | debug_object_activate(work, &work_debug_descr); | |
361 | return 0; | |
362 | } | |
363 | WARN_ON_ONCE(1); | |
364 | return 0; | |
365 | ||
366 | case ODEBUG_STATE_ACTIVE: | |
367 | WARN_ON(1); | |
368 | ||
369 | default: | |
370 | return 0; | |
371 | } | |
372 | } | |
373 | ||
374 | /* | |
375 | * fixup_free is called when: | |
376 | * - an active object is freed | |
377 | */ | |
378 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
379 | { | |
380 | struct work_struct *work = addr; | |
381 | ||
382 | switch (state) { | |
383 | case ODEBUG_STATE_ACTIVE: | |
384 | cancel_work_sync(work); | |
385 | debug_object_free(work, &work_debug_descr); | |
386 | return 1; | |
387 | default: | |
388 | return 0; | |
389 | } | |
390 | } | |
391 | ||
392 | static struct debug_obj_descr work_debug_descr = { | |
393 | .name = "work_struct", | |
99777288 | 394 | .debug_hint = work_debug_hint, |
dc186ad7 TG |
395 | .fixup_init = work_fixup_init, |
396 | .fixup_activate = work_fixup_activate, | |
397 | .fixup_free = work_fixup_free, | |
398 | }; | |
399 | ||
400 | static inline void debug_work_activate(struct work_struct *work) | |
401 | { | |
402 | debug_object_activate(work, &work_debug_descr); | |
403 | } | |
404 | ||
405 | static inline void debug_work_deactivate(struct work_struct *work) | |
406 | { | |
407 | debug_object_deactivate(work, &work_debug_descr); | |
408 | } | |
409 | ||
410 | void __init_work(struct work_struct *work, int onstack) | |
411 | { | |
412 | if (onstack) | |
413 | debug_object_init_on_stack(work, &work_debug_descr); | |
414 | else | |
415 | debug_object_init(work, &work_debug_descr); | |
416 | } | |
417 | EXPORT_SYMBOL_GPL(__init_work); | |
418 | ||
419 | void destroy_work_on_stack(struct work_struct *work) | |
420 | { | |
421 | debug_object_free(work, &work_debug_descr); | |
422 | } | |
423 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
424 | ||
425 | #else | |
426 | static inline void debug_work_activate(struct work_struct *work) { } | |
427 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
428 | #endif | |
429 | ||
95402b38 GS |
430 | /* Serializes the accesses to the list of workqueues. */ |
431 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 432 | static LIST_HEAD(workqueues); |
a0a1a5fd | 433 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 434 | |
e22bee78 TH |
435 | /* |
436 | * The almighty global cpu workqueues. nr_running is the only field | |
437 | * which is expected to be used frequently by other cpus via | |
438 | * try_to_wake_up(). Put it in a separate cacheline. | |
439 | */ | |
8b03ae3c | 440 | static DEFINE_PER_CPU(struct global_cwq, global_cwq); |
e34cdddb | 441 | static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_STD_WORKER_POOLS]); |
8b03ae3c | 442 | |
f3421797 | 443 | /* |
24647570 TH |
444 | * Global cpu workqueue and nr_running counter for unbound gcwq. The pools |
445 | * for online CPUs have POOL_DISASSOCIATED set, and all their workers have | |
446 | * WORKER_UNBOUND set. | |
f3421797 TH |
447 | */ |
448 | static struct global_cwq unbound_global_cwq; | |
e34cdddb TH |
449 | static atomic_t unbound_pool_nr_running[NR_STD_WORKER_POOLS] = { |
450 | [0 ... NR_STD_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */ | |
4ce62e9e | 451 | }; |
f3421797 | 452 | |
c34056a3 | 453 | static int worker_thread(void *__worker); |
e2905b29 | 454 | static unsigned int work_cpu(struct work_struct *work); |
1da177e4 | 455 | |
e34cdddb | 456 | static int std_worker_pool_pri(struct worker_pool *pool) |
3270476a TH |
457 | { |
458 | return pool - pool->gcwq->pools; | |
459 | } | |
460 | ||
8b03ae3c TH |
461 | static struct global_cwq *get_gcwq(unsigned int cpu) |
462 | { | |
f3421797 TH |
463 | if (cpu != WORK_CPU_UNBOUND) |
464 | return &per_cpu(global_cwq, cpu); | |
465 | else | |
466 | return &unbound_global_cwq; | |
8b03ae3c TH |
467 | } |
468 | ||
63d95a91 | 469 | static atomic_t *get_pool_nr_running(struct worker_pool *pool) |
e22bee78 | 470 | { |
63d95a91 | 471 | int cpu = pool->gcwq->cpu; |
e34cdddb | 472 | int idx = std_worker_pool_pri(pool); |
63d95a91 | 473 | |
f3421797 | 474 | if (cpu != WORK_CPU_UNBOUND) |
4ce62e9e | 475 | return &per_cpu(pool_nr_running, cpu)[idx]; |
f3421797 | 476 | else |
4ce62e9e | 477 | return &unbound_pool_nr_running[idx]; |
e22bee78 TH |
478 | } |
479 | ||
1537663f TH |
480 | static struct cpu_workqueue_struct *get_cwq(unsigned int cpu, |
481 | struct workqueue_struct *wq) | |
b1f4ec17 | 482 | { |
f3421797 | 483 | if (!(wq->flags & WQ_UNBOUND)) { |
e06ffa1e | 484 | if (likely(cpu < nr_cpu_ids)) |
f3421797 | 485 | return per_cpu_ptr(wq->cpu_wq.pcpu, cpu); |
f3421797 TH |
486 | } else if (likely(cpu == WORK_CPU_UNBOUND)) |
487 | return wq->cpu_wq.single; | |
488 | return NULL; | |
b1f4ec17 ON |
489 | } |
490 | ||
73f53c4a TH |
491 | static unsigned int work_color_to_flags(int color) |
492 | { | |
493 | return color << WORK_STRUCT_COLOR_SHIFT; | |
494 | } | |
495 | ||
496 | static int get_work_color(struct work_struct *work) | |
497 | { | |
498 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
499 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
500 | } | |
501 | ||
502 | static int work_next_color(int color) | |
503 | { | |
504 | return (color + 1) % WORK_NR_COLORS; | |
505 | } | |
1da177e4 | 506 | |
14441960 | 507 | /* |
b5490077 TH |
508 | * While queued, %WORK_STRUCT_CWQ is set and non flag bits of a work's data |
509 | * contain the pointer to the queued cwq. Once execution starts, the flag | |
510 | * is cleared and the high bits contain OFFQ flags and CPU number. | |
7a22ad75 | 511 | * |
bbb68dfa TH |
512 | * set_work_cwq(), set_work_cpu_and_clear_pending(), mark_work_canceling() |
513 | * and clear_work_data() can be used to set the cwq, cpu or clear | |
514 | * work->data. These functions should only be called while the work is | |
515 | * owned - ie. while the PENDING bit is set. | |
7a22ad75 | 516 | * |
bbb68dfa TH |
517 | * get_work_[g]cwq() can be used to obtain the gcwq or cwq corresponding to |
518 | * a work. gcwq is available once the work has been queued anywhere after | |
519 | * initialization until it is sync canceled. cwq is available only while | |
520 | * the work item is queued. | |
7a22ad75 | 521 | * |
bbb68dfa TH |
522 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
523 | * canceled. While being canceled, a work item may have its PENDING set | |
524 | * but stay off timer and worklist for arbitrarily long and nobody should | |
525 | * try to steal the PENDING bit. | |
14441960 | 526 | */ |
7a22ad75 TH |
527 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
528 | unsigned long flags) | |
365970a1 | 529 | { |
4594bf15 | 530 | BUG_ON(!work_pending(work)); |
7a22ad75 TH |
531 | atomic_long_set(&work->data, data | flags | work_static(work)); |
532 | } | |
365970a1 | 533 | |
7a22ad75 TH |
534 | static void set_work_cwq(struct work_struct *work, |
535 | struct cpu_workqueue_struct *cwq, | |
536 | unsigned long extra_flags) | |
537 | { | |
538 | set_work_data(work, (unsigned long)cwq, | |
e120153d | 539 | WORK_STRUCT_PENDING | WORK_STRUCT_CWQ | extra_flags); |
365970a1 DH |
540 | } |
541 | ||
8930caba TH |
542 | static void set_work_cpu_and_clear_pending(struct work_struct *work, |
543 | unsigned int cpu) | |
7a22ad75 | 544 | { |
23657bb1 TH |
545 | /* |
546 | * The following wmb is paired with the implied mb in | |
547 | * test_and_set_bit(PENDING) and ensures all updates to @work made | |
548 | * here are visible to and precede any updates by the next PENDING | |
549 | * owner. | |
550 | */ | |
551 | smp_wmb(); | |
b5490077 | 552 | set_work_data(work, (unsigned long)cpu << WORK_OFFQ_CPU_SHIFT, 0); |
7a22ad75 | 553 | } |
f756d5e2 | 554 | |
7a22ad75 | 555 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 556 | { |
23657bb1 | 557 | smp_wmb(); /* see set_work_cpu_and_clear_pending() */ |
7a22ad75 | 558 | set_work_data(work, WORK_STRUCT_NO_CPU, 0); |
1da177e4 LT |
559 | } |
560 | ||
7a22ad75 | 561 | static struct cpu_workqueue_struct *get_work_cwq(struct work_struct *work) |
b1f4ec17 | 562 | { |
e120153d | 563 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 564 | |
e120153d TH |
565 | if (data & WORK_STRUCT_CWQ) |
566 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); | |
567 | else | |
568 | return NULL; | |
4d707b9f ON |
569 | } |
570 | ||
7a22ad75 | 571 | static struct global_cwq *get_work_gcwq(struct work_struct *work) |
365970a1 | 572 | { |
e120153d | 573 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 TH |
574 | unsigned int cpu; |
575 | ||
e120153d TH |
576 | if (data & WORK_STRUCT_CWQ) |
577 | return ((struct cpu_workqueue_struct *) | |
bd7bdd43 | 578 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->gcwq; |
7a22ad75 | 579 | |
b5490077 | 580 | cpu = data >> WORK_OFFQ_CPU_SHIFT; |
bdbc5dd7 | 581 | if (cpu == WORK_CPU_NONE) |
7a22ad75 TH |
582 | return NULL; |
583 | ||
f3421797 | 584 | BUG_ON(cpu >= nr_cpu_ids && cpu != WORK_CPU_UNBOUND); |
7a22ad75 | 585 | return get_gcwq(cpu); |
b1f4ec17 ON |
586 | } |
587 | ||
bbb68dfa TH |
588 | static void mark_work_canceling(struct work_struct *work) |
589 | { | |
590 | struct global_cwq *gcwq = get_work_gcwq(work); | |
591 | unsigned long cpu = gcwq ? gcwq->cpu : WORK_CPU_NONE; | |
592 | ||
593 | set_work_data(work, (cpu << WORK_OFFQ_CPU_SHIFT) | WORK_OFFQ_CANCELING, | |
594 | WORK_STRUCT_PENDING); | |
595 | } | |
596 | ||
597 | static bool work_is_canceling(struct work_struct *work) | |
598 | { | |
599 | unsigned long data = atomic_long_read(&work->data); | |
600 | ||
601 | return !(data & WORK_STRUCT_CWQ) && (data & WORK_OFFQ_CANCELING); | |
602 | } | |
603 | ||
e22bee78 | 604 | /* |
3270476a TH |
605 | * Policy functions. These define the policies on how the global worker |
606 | * pools are managed. Unless noted otherwise, these functions assume that | |
607 | * they're being called with gcwq->lock held. | |
e22bee78 TH |
608 | */ |
609 | ||
63d95a91 | 610 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 611 | { |
3270476a | 612 | return !atomic_read(get_pool_nr_running(pool)); |
a848e3b6 ON |
613 | } |
614 | ||
4594bf15 | 615 | /* |
e22bee78 TH |
616 | * Need to wake up a worker? Called from anything but currently |
617 | * running workers. | |
974271c4 TH |
618 | * |
619 | * Note that, because unbound workers never contribute to nr_running, this | |
620 | * function will always return %true for unbound gcwq as long as the | |
621 | * worklist isn't empty. | |
4594bf15 | 622 | */ |
63d95a91 | 623 | static bool need_more_worker(struct worker_pool *pool) |
365970a1 | 624 | { |
63d95a91 | 625 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
e22bee78 | 626 | } |
4594bf15 | 627 | |
e22bee78 | 628 | /* Can I start working? Called from busy but !running workers. */ |
63d95a91 | 629 | static bool may_start_working(struct worker_pool *pool) |
e22bee78 | 630 | { |
63d95a91 | 631 | return pool->nr_idle; |
e22bee78 TH |
632 | } |
633 | ||
634 | /* Do I need to keep working? Called from currently running workers. */ | |
63d95a91 | 635 | static bool keep_working(struct worker_pool *pool) |
e22bee78 | 636 | { |
63d95a91 | 637 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 | 638 | |
3270476a | 639 | return !list_empty(&pool->worklist) && atomic_read(nr_running) <= 1; |
e22bee78 TH |
640 | } |
641 | ||
642 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 643 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 644 | { |
63d95a91 | 645 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 646 | } |
365970a1 | 647 | |
e22bee78 | 648 | /* Do I need to be the manager? */ |
63d95a91 | 649 | static bool need_to_manage_workers(struct worker_pool *pool) |
e22bee78 | 650 | { |
63d95a91 | 651 | return need_to_create_worker(pool) || |
11ebea50 | 652 | (pool->flags & POOL_MANAGE_WORKERS); |
e22bee78 TH |
653 | } |
654 | ||
655 | /* Do we have too many workers and should some go away? */ | |
63d95a91 | 656 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 657 | { |
552a37e9 | 658 | bool managing = pool->flags & POOL_MANAGING_WORKERS; |
63d95a91 TH |
659 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
660 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 | 661 | |
ea1abd61 LJ |
662 | /* |
663 | * nr_idle and idle_list may disagree if idle rebinding is in | |
664 | * progress. Never return %true if idle_list is empty. | |
665 | */ | |
666 | if (list_empty(&pool->idle_list)) | |
667 | return false; | |
668 | ||
e22bee78 | 669 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
365970a1 DH |
670 | } |
671 | ||
4d707b9f | 672 | /* |
e22bee78 TH |
673 | * Wake up functions. |
674 | */ | |
675 | ||
7e11629d | 676 | /* Return the first worker. Safe with preemption disabled */ |
63d95a91 | 677 | static struct worker *first_worker(struct worker_pool *pool) |
7e11629d | 678 | { |
63d95a91 | 679 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
680 | return NULL; |
681 | ||
63d95a91 | 682 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
683 | } |
684 | ||
685 | /** | |
686 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 687 | * @pool: worker pool to wake worker from |
7e11629d | 688 | * |
63d95a91 | 689 | * Wake up the first idle worker of @pool. |
7e11629d TH |
690 | * |
691 | * CONTEXT: | |
692 | * spin_lock_irq(gcwq->lock). | |
693 | */ | |
63d95a91 | 694 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 695 | { |
63d95a91 | 696 | struct worker *worker = first_worker(pool); |
7e11629d TH |
697 | |
698 | if (likely(worker)) | |
699 | wake_up_process(worker->task); | |
700 | } | |
701 | ||
d302f017 | 702 | /** |
e22bee78 TH |
703 | * wq_worker_waking_up - a worker is waking up |
704 | * @task: task waking up | |
705 | * @cpu: CPU @task is waking up to | |
706 | * | |
707 | * This function is called during try_to_wake_up() when a worker is | |
708 | * being awoken. | |
709 | * | |
710 | * CONTEXT: | |
711 | * spin_lock_irq(rq->lock) | |
712 | */ | |
713 | void wq_worker_waking_up(struct task_struct *task, unsigned int cpu) | |
714 | { | |
715 | struct worker *worker = kthread_data(task); | |
716 | ||
36576000 JK |
717 | if (!(worker->flags & WORKER_NOT_RUNNING)) { |
718 | WARN_ON_ONCE(worker->pool->gcwq->cpu != cpu); | |
63d95a91 | 719 | atomic_inc(get_pool_nr_running(worker->pool)); |
36576000 | 720 | } |
e22bee78 TH |
721 | } |
722 | ||
723 | /** | |
724 | * wq_worker_sleeping - a worker is going to sleep | |
725 | * @task: task going to sleep | |
726 | * @cpu: CPU in question, must be the current CPU number | |
727 | * | |
728 | * This function is called during schedule() when a busy worker is | |
729 | * going to sleep. Worker on the same cpu can be woken up by | |
730 | * returning pointer to its task. | |
731 | * | |
732 | * CONTEXT: | |
733 | * spin_lock_irq(rq->lock) | |
734 | * | |
735 | * RETURNS: | |
736 | * Worker task on @cpu to wake up, %NULL if none. | |
737 | */ | |
738 | struct task_struct *wq_worker_sleeping(struct task_struct *task, | |
739 | unsigned int cpu) | |
740 | { | |
741 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
111c225a TH |
742 | struct worker_pool *pool; |
743 | atomic_t *nr_running; | |
e22bee78 | 744 | |
111c225a TH |
745 | /* |
746 | * Rescuers, which may not have all the fields set up like normal | |
747 | * workers, also reach here, let's not access anything before | |
748 | * checking NOT_RUNNING. | |
749 | */ | |
2d64672e | 750 | if (worker->flags & WORKER_NOT_RUNNING) |
e22bee78 TH |
751 | return NULL; |
752 | ||
111c225a TH |
753 | pool = worker->pool; |
754 | nr_running = get_pool_nr_running(pool); | |
755 | ||
e22bee78 TH |
756 | /* this can only happen on the local cpu */ |
757 | BUG_ON(cpu != raw_smp_processor_id()); | |
758 | ||
759 | /* | |
760 | * The counterpart of the following dec_and_test, implied mb, | |
761 | * worklist not empty test sequence is in insert_work(). | |
762 | * Please read comment there. | |
763 | * | |
628c78e7 TH |
764 | * NOT_RUNNING is clear. This means that we're bound to and |
765 | * running on the local cpu w/ rq lock held and preemption | |
766 | * disabled, which in turn means that none else could be | |
767 | * manipulating idle_list, so dereferencing idle_list without gcwq | |
768 | * lock is safe. | |
e22bee78 | 769 | */ |
bd7bdd43 | 770 | if (atomic_dec_and_test(nr_running) && !list_empty(&pool->worklist)) |
63d95a91 | 771 | to_wakeup = first_worker(pool); |
e22bee78 TH |
772 | return to_wakeup ? to_wakeup->task : NULL; |
773 | } | |
774 | ||
775 | /** | |
776 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 777 | * @worker: self |
d302f017 TH |
778 | * @flags: flags to set |
779 | * @wakeup: wakeup an idle worker if necessary | |
780 | * | |
e22bee78 TH |
781 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
782 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
783 | * woken up. | |
d302f017 | 784 | * |
cb444766 TH |
785 | * CONTEXT: |
786 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
787 | */ |
788 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
789 | bool wakeup) | |
790 | { | |
bd7bdd43 | 791 | struct worker_pool *pool = worker->pool; |
e22bee78 | 792 | |
cb444766 TH |
793 | WARN_ON_ONCE(worker->task != current); |
794 | ||
e22bee78 TH |
795 | /* |
796 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
797 | * wake up an idle worker as necessary if requested by | |
798 | * @wakeup. | |
799 | */ | |
800 | if ((flags & WORKER_NOT_RUNNING) && | |
801 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
63d95a91 | 802 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 TH |
803 | |
804 | if (wakeup) { | |
805 | if (atomic_dec_and_test(nr_running) && | |
bd7bdd43 | 806 | !list_empty(&pool->worklist)) |
63d95a91 | 807 | wake_up_worker(pool); |
e22bee78 TH |
808 | } else |
809 | atomic_dec(nr_running); | |
810 | } | |
811 | ||
d302f017 TH |
812 | worker->flags |= flags; |
813 | } | |
814 | ||
815 | /** | |
e22bee78 | 816 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 817 | * @worker: self |
d302f017 TH |
818 | * @flags: flags to clear |
819 | * | |
e22bee78 | 820 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 821 | * |
cb444766 TH |
822 | * CONTEXT: |
823 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
824 | */ |
825 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
826 | { | |
63d95a91 | 827 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
828 | unsigned int oflags = worker->flags; |
829 | ||
cb444766 TH |
830 | WARN_ON_ONCE(worker->task != current); |
831 | ||
d302f017 | 832 | worker->flags &= ~flags; |
e22bee78 | 833 | |
42c025f3 TH |
834 | /* |
835 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
836 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
837 | * of multiple flags, not a single flag. | |
838 | */ | |
e22bee78 TH |
839 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
840 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
63d95a91 | 841 | atomic_inc(get_pool_nr_running(pool)); |
d302f017 TH |
842 | } |
843 | ||
8cca0eea TH |
844 | /** |
845 | * find_worker_executing_work - find worker which is executing a work | |
846 | * @gcwq: gcwq of interest | |
847 | * @work: work to find worker for | |
848 | * | |
a2c1c57b TH |
849 | * Find a worker which is executing @work on @gcwq by searching |
850 | * @gcwq->busy_hash which is keyed by the address of @work. For a worker | |
851 | * to match, its current execution should match the address of @work and | |
852 | * its work function. This is to avoid unwanted dependency between | |
853 | * unrelated work executions through a work item being recycled while still | |
854 | * being executed. | |
855 | * | |
856 | * This is a bit tricky. A work item may be freed once its execution | |
857 | * starts and nothing prevents the freed area from being recycled for | |
858 | * another work item. If the same work item address ends up being reused | |
859 | * before the original execution finishes, workqueue will identify the | |
860 | * recycled work item as currently executing and make it wait until the | |
861 | * current execution finishes, introducing an unwanted dependency. | |
862 | * | |
863 | * This function checks the work item address, work function and workqueue | |
864 | * to avoid false positives. Note that this isn't complete as one may | |
865 | * construct a work function which can introduce dependency onto itself | |
866 | * through a recycled work item. Well, if somebody wants to shoot oneself | |
867 | * in the foot that badly, there's only so much we can do, and if such | |
868 | * deadlock actually occurs, it should be easy to locate the culprit work | |
869 | * function. | |
8cca0eea TH |
870 | * |
871 | * CONTEXT: | |
872 | * spin_lock_irq(gcwq->lock). | |
873 | * | |
874 | * RETURNS: | |
875 | * Pointer to worker which is executing @work if found, NULL | |
876 | * otherwise. | |
4d707b9f | 877 | */ |
8cca0eea TH |
878 | static struct worker *find_worker_executing_work(struct global_cwq *gcwq, |
879 | struct work_struct *work) | |
4d707b9f | 880 | { |
42f8570f SL |
881 | struct worker *worker; |
882 | struct hlist_node *tmp; | |
883 | ||
a2c1c57b TH |
884 | hash_for_each_possible(gcwq->busy_hash, worker, tmp, hentry, |
885 | (unsigned long)work) | |
886 | if (worker->current_work == work && | |
887 | worker->current_func == work->func) | |
42f8570f SL |
888 | return worker; |
889 | ||
890 | return NULL; | |
4d707b9f ON |
891 | } |
892 | ||
bf4ede01 TH |
893 | /** |
894 | * move_linked_works - move linked works to a list | |
895 | * @work: start of series of works to be scheduled | |
896 | * @head: target list to append @work to | |
897 | * @nextp: out paramter for nested worklist walking | |
898 | * | |
899 | * Schedule linked works starting from @work to @head. Work series to | |
900 | * be scheduled starts at @work and includes any consecutive work with | |
901 | * WORK_STRUCT_LINKED set in its predecessor. | |
902 | * | |
903 | * If @nextp is not NULL, it's updated to point to the next work of | |
904 | * the last scheduled work. This allows move_linked_works() to be | |
905 | * nested inside outer list_for_each_entry_safe(). | |
906 | * | |
907 | * CONTEXT: | |
908 | * spin_lock_irq(gcwq->lock). | |
909 | */ | |
910 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
911 | struct work_struct **nextp) | |
912 | { | |
913 | struct work_struct *n; | |
914 | ||
915 | /* | |
916 | * Linked worklist will always end before the end of the list, | |
917 | * use NULL for list head. | |
918 | */ | |
919 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
920 | list_move_tail(&work->entry, head); | |
921 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
922 | break; | |
923 | } | |
924 | ||
925 | /* | |
926 | * If we're already inside safe list traversal and have moved | |
927 | * multiple works to the scheduled queue, the next position | |
928 | * needs to be updated. | |
929 | */ | |
930 | if (nextp) | |
931 | *nextp = n; | |
932 | } | |
933 | ||
3aa62497 | 934 | static void cwq_activate_delayed_work(struct work_struct *work) |
bf4ede01 | 935 | { |
3aa62497 | 936 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); |
bf4ede01 TH |
937 | |
938 | trace_workqueue_activate_work(work); | |
939 | move_linked_works(work, &cwq->pool->worklist, NULL); | |
940 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); | |
941 | cwq->nr_active++; | |
942 | } | |
943 | ||
3aa62497 LJ |
944 | static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq) |
945 | { | |
946 | struct work_struct *work = list_first_entry(&cwq->delayed_works, | |
947 | struct work_struct, entry); | |
948 | ||
949 | cwq_activate_delayed_work(work); | |
950 | } | |
951 | ||
bf4ede01 TH |
952 | /** |
953 | * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight | |
954 | * @cwq: cwq of interest | |
955 | * @color: color of work which left the queue | |
bf4ede01 TH |
956 | * |
957 | * A work either has completed or is removed from pending queue, | |
958 | * decrement nr_in_flight of its cwq and handle workqueue flushing. | |
959 | * | |
960 | * CONTEXT: | |
961 | * spin_lock_irq(gcwq->lock). | |
962 | */ | |
b3f9f405 | 963 | static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color) |
bf4ede01 TH |
964 | { |
965 | /* ignore uncolored works */ | |
966 | if (color == WORK_NO_COLOR) | |
967 | return; | |
968 | ||
969 | cwq->nr_in_flight[color]--; | |
970 | ||
b3f9f405 LJ |
971 | cwq->nr_active--; |
972 | if (!list_empty(&cwq->delayed_works)) { | |
973 | /* one down, submit a delayed one */ | |
974 | if (cwq->nr_active < cwq->max_active) | |
975 | cwq_activate_first_delayed(cwq); | |
bf4ede01 TH |
976 | } |
977 | ||
978 | /* is flush in progress and are we at the flushing tip? */ | |
979 | if (likely(cwq->flush_color != color)) | |
980 | return; | |
981 | ||
982 | /* are there still in-flight works? */ | |
983 | if (cwq->nr_in_flight[color]) | |
984 | return; | |
985 | ||
986 | /* this cwq is done, clear flush_color */ | |
987 | cwq->flush_color = -1; | |
988 | ||
989 | /* | |
990 | * If this was the last cwq, wake up the first flusher. It | |
991 | * will handle the rest. | |
992 | */ | |
993 | if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush)) | |
994 | complete(&cwq->wq->first_flusher->done); | |
995 | } | |
996 | ||
36e227d2 | 997 | /** |
bbb68dfa | 998 | * try_to_grab_pending - steal work item from worklist and disable irq |
36e227d2 TH |
999 | * @work: work item to steal |
1000 | * @is_dwork: @work is a delayed_work | |
bbb68dfa | 1001 | * @flags: place to store irq state |
36e227d2 TH |
1002 | * |
1003 | * Try to grab PENDING bit of @work. This function can handle @work in any | |
1004 | * stable state - idle, on timer or on worklist. Return values are | |
1005 | * | |
1006 | * 1 if @work was pending and we successfully stole PENDING | |
1007 | * 0 if @work was idle and we claimed PENDING | |
1008 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry | |
bbb68dfa TH |
1009 | * -ENOENT if someone else is canceling @work, this state may persist |
1010 | * for arbitrarily long | |
36e227d2 | 1011 | * |
bbb68dfa | 1012 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
e0aecdd8 TH |
1013 | * interrupted while holding PENDING and @work off queue, irq must be |
1014 | * disabled on entry. This, combined with delayed_work->timer being | |
1015 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. | |
bbb68dfa TH |
1016 | * |
1017 | * On successful return, >= 0, irq is disabled and the caller is | |
1018 | * responsible for releasing it using local_irq_restore(*@flags). | |
1019 | * | |
e0aecdd8 | 1020 | * This function is safe to call from any context including IRQ handler. |
bf4ede01 | 1021 | */ |
bbb68dfa TH |
1022 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
1023 | unsigned long *flags) | |
bf4ede01 TH |
1024 | { |
1025 | struct global_cwq *gcwq; | |
bf4ede01 | 1026 | |
bbb68dfa TH |
1027 | local_irq_save(*flags); |
1028 | ||
36e227d2 TH |
1029 | /* try to steal the timer if it exists */ |
1030 | if (is_dwork) { | |
1031 | struct delayed_work *dwork = to_delayed_work(work); | |
1032 | ||
e0aecdd8 TH |
1033 | /* |
1034 | * dwork->timer is irqsafe. If del_timer() fails, it's | |
1035 | * guaranteed that the timer is not queued anywhere and not | |
1036 | * running on the local CPU. | |
1037 | */ | |
36e227d2 TH |
1038 | if (likely(del_timer(&dwork->timer))) |
1039 | return 1; | |
1040 | } | |
1041 | ||
1042 | /* try to claim PENDING the normal way */ | |
bf4ede01 TH |
1043 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1044 | return 0; | |
1045 | ||
1046 | /* | |
1047 | * The queueing is in progress, or it is already queued. Try to | |
1048 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1049 | */ | |
1050 | gcwq = get_work_gcwq(work); | |
1051 | if (!gcwq) | |
bbb68dfa | 1052 | goto fail; |
bf4ede01 | 1053 | |
bbb68dfa | 1054 | spin_lock(&gcwq->lock); |
bf4ede01 TH |
1055 | if (!list_empty(&work->entry)) { |
1056 | /* | |
1057 | * This work is queued, but perhaps we locked the wrong gcwq. | |
1058 | * In that case we must see the new value after rmb(), see | |
1059 | * insert_work()->wmb(). | |
1060 | */ | |
1061 | smp_rmb(); | |
1062 | if (gcwq == get_work_gcwq(work)) { | |
1063 | debug_work_deactivate(work); | |
3aa62497 LJ |
1064 | |
1065 | /* | |
1066 | * A delayed work item cannot be grabbed directly | |
1067 | * because it might have linked NO_COLOR work items | |
1068 | * which, if left on the delayed_list, will confuse | |
1069 | * cwq->nr_active management later on and cause | |
1070 | * stall. Make sure the work item is activated | |
1071 | * before grabbing. | |
1072 | */ | |
1073 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) | |
1074 | cwq_activate_delayed_work(work); | |
1075 | ||
bf4ede01 TH |
1076 | list_del_init(&work->entry); |
1077 | cwq_dec_nr_in_flight(get_work_cwq(work), | |
b3f9f405 | 1078 | get_work_color(work)); |
36e227d2 | 1079 | |
bbb68dfa | 1080 | spin_unlock(&gcwq->lock); |
36e227d2 | 1081 | return 1; |
bf4ede01 TH |
1082 | } |
1083 | } | |
bbb68dfa TH |
1084 | spin_unlock(&gcwq->lock); |
1085 | fail: | |
1086 | local_irq_restore(*flags); | |
1087 | if (work_is_canceling(work)) | |
1088 | return -ENOENT; | |
1089 | cpu_relax(); | |
36e227d2 | 1090 | return -EAGAIN; |
bf4ede01 TH |
1091 | } |
1092 | ||
4690c4ab | 1093 | /** |
7e11629d | 1094 | * insert_work - insert a work into gcwq |
4690c4ab TH |
1095 | * @cwq: cwq @work belongs to |
1096 | * @work: work to insert | |
1097 | * @head: insertion point | |
1098 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
1099 | * | |
7e11629d TH |
1100 | * Insert @work which belongs to @cwq into @gcwq after @head. |
1101 | * @extra_flags is or'd to work_struct flags. | |
4690c4ab TH |
1102 | * |
1103 | * CONTEXT: | |
8b03ae3c | 1104 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 1105 | */ |
b89deed3 | 1106 | static void insert_work(struct cpu_workqueue_struct *cwq, |
4690c4ab TH |
1107 | struct work_struct *work, struct list_head *head, |
1108 | unsigned int extra_flags) | |
b89deed3 | 1109 | { |
63d95a91 | 1110 | struct worker_pool *pool = cwq->pool; |
e22bee78 | 1111 | |
4690c4ab | 1112 | /* we own @work, set data and link */ |
7a22ad75 | 1113 | set_work_cwq(work, cwq, extra_flags); |
e1d8aa9f | 1114 | |
6e84d644 ON |
1115 | /* |
1116 | * Ensure that we get the right work->data if we see the | |
1117 | * result of list_add() below, see try_to_grab_pending(). | |
1118 | */ | |
1119 | smp_wmb(); | |
4690c4ab | 1120 | |
1a4d9b0a | 1121 | list_add_tail(&work->entry, head); |
e22bee78 TH |
1122 | |
1123 | /* | |
1124 | * Ensure either worker_sched_deactivated() sees the above | |
1125 | * list_add_tail() or we see zero nr_running to avoid workers | |
1126 | * lying around lazily while there are works to be processed. | |
1127 | */ | |
1128 | smp_mb(); | |
1129 | ||
63d95a91 TH |
1130 | if (__need_more_worker(pool)) |
1131 | wake_up_worker(pool); | |
b89deed3 ON |
1132 | } |
1133 | ||
c8efcc25 TH |
1134 | /* |
1135 | * Test whether @work is being queued from another work executing on the | |
1136 | * same workqueue. This is rather expensive and should only be used from | |
1137 | * cold paths. | |
1138 | */ | |
1139 | static bool is_chained_work(struct workqueue_struct *wq) | |
1140 | { | |
1141 | unsigned long flags; | |
1142 | unsigned int cpu; | |
1143 | ||
1144 | for_each_gcwq_cpu(cpu) { | |
1145 | struct global_cwq *gcwq = get_gcwq(cpu); | |
1146 | struct worker *worker; | |
1147 | struct hlist_node *pos; | |
1148 | int i; | |
1149 | ||
1150 | spin_lock_irqsave(&gcwq->lock, flags); | |
1151 | for_each_busy_worker(worker, i, pos, gcwq) { | |
1152 | if (worker->task != current) | |
1153 | continue; | |
1154 | spin_unlock_irqrestore(&gcwq->lock, flags); | |
1155 | /* | |
1156 | * I'm @worker, no locking necessary. See if @work | |
1157 | * is headed to the same workqueue. | |
1158 | */ | |
1159 | return worker->current_cwq->wq == wq; | |
1160 | } | |
1161 | spin_unlock_irqrestore(&gcwq->lock, flags); | |
1162 | } | |
1163 | return false; | |
1164 | } | |
1165 | ||
4690c4ab | 1166 | static void __queue_work(unsigned int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
1167 | struct work_struct *work) |
1168 | { | |
502ca9d8 TH |
1169 | struct global_cwq *gcwq; |
1170 | struct cpu_workqueue_struct *cwq; | |
1e19ffc6 | 1171 | struct list_head *worklist; |
8a2e8e5d | 1172 | unsigned int work_flags; |
b75cac93 | 1173 | unsigned int req_cpu = cpu; |
8930caba TH |
1174 | |
1175 | /* | |
1176 | * While a work item is PENDING && off queue, a task trying to | |
1177 | * steal the PENDING will busy-loop waiting for it to either get | |
1178 | * queued or lose PENDING. Grabbing PENDING and queueing should | |
1179 | * happen with IRQ disabled. | |
1180 | */ | |
1181 | WARN_ON_ONCE(!irqs_disabled()); | |
1da177e4 | 1182 | |
dc186ad7 | 1183 | debug_work_activate(work); |
1e19ffc6 | 1184 | |
c8efcc25 | 1185 | /* if dying, only works from the same workqueue are allowed */ |
9c5a2ba7 | 1186 | if (unlikely(wq->flags & WQ_DRAINING) && |
c8efcc25 | 1187 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b TH |
1188 | return; |
1189 | ||
c7fc77f7 TH |
1190 | /* determine gcwq to use */ |
1191 | if (!(wq->flags & WQ_UNBOUND)) { | |
18aa9eff TH |
1192 | struct global_cwq *last_gcwq; |
1193 | ||
57469821 | 1194 | if (cpu == WORK_CPU_UNBOUND) |
c7fc77f7 TH |
1195 | cpu = raw_smp_processor_id(); |
1196 | ||
18aa9eff | 1197 | /* |
dbf2576e TH |
1198 | * It's multi cpu. If @work was previously on a different |
1199 | * cpu, it might still be running there, in which case the | |
1200 | * work needs to be queued on that cpu to guarantee | |
1201 | * non-reentrancy. | |
18aa9eff | 1202 | */ |
502ca9d8 | 1203 | gcwq = get_gcwq(cpu); |
dbf2576e TH |
1204 | last_gcwq = get_work_gcwq(work); |
1205 | ||
1206 | if (last_gcwq && last_gcwq != gcwq) { | |
18aa9eff TH |
1207 | struct worker *worker; |
1208 | ||
8930caba | 1209 | spin_lock(&last_gcwq->lock); |
18aa9eff TH |
1210 | |
1211 | worker = find_worker_executing_work(last_gcwq, work); | |
1212 | ||
1213 | if (worker && worker->current_cwq->wq == wq) | |
1214 | gcwq = last_gcwq; | |
1215 | else { | |
1216 | /* meh... not running there, queue here */ | |
8930caba TH |
1217 | spin_unlock(&last_gcwq->lock); |
1218 | spin_lock(&gcwq->lock); | |
18aa9eff | 1219 | } |
8930caba TH |
1220 | } else { |
1221 | spin_lock(&gcwq->lock); | |
1222 | } | |
f3421797 TH |
1223 | } else { |
1224 | gcwq = get_gcwq(WORK_CPU_UNBOUND); | |
8930caba | 1225 | spin_lock(&gcwq->lock); |
502ca9d8 TH |
1226 | } |
1227 | ||
1228 | /* gcwq determined, get cwq and queue */ | |
1229 | cwq = get_cwq(gcwq->cpu, wq); | |
b75cac93 | 1230 | trace_workqueue_queue_work(req_cpu, cwq, work); |
502ca9d8 | 1231 | |
f5b2552b | 1232 | if (WARN_ON(!list_empty(&work->entry))) { |
8930caba | 1233 | spin_unlock(&gcwq->lock); |
f5b2552b DC |
1234 | return; |
1235 | } | |
1e19ffc6 | 1236 | |
73f53c4a | 1237 | cwq->nr_in_flight[cwq->work_color]++; |
8a2e8e5d | 1238 | work_flags = work_color_to_flags(cwq->work_color); |
1e19ffc6 TH |
1239 | |
1240 | if (likely(cwq->nr_active < cwq->max_active)) { | |
cdadf009 | 1241 | trace_workqueue_activate_work(work); |
1e19ffc6 | 1242 | cwq->nr_active++; |
3270476a | 1243 | worklist = &cwq->pool->worklist; |
8a2e8e5d TH |
1244 | } else { |
1245 | work_flags |= WORK_STRUCT_DELAYED; | |
1e19ffc6 | 1246 | worklist = &cwq->delayed_works; |
8a2e8e5d | 1247 | } |
1e19ffc6 | 1248 | |
8a2e8e5d | 1249 | insert_work(cwq, work, worklist, work_flags); |
1e19ffc6 | 1250 | |
8930caba | 1251 | spin_unlock(&gcwq->lock); |
1da177e4 LT |
1252 | } |
1253 | ||
0fcb78c2 | 1254 | /** |
c1a220e7 ZR |
1255 | * queue_work_on - queue work on specific cpu |
1256 | * @cpu: CPU number to execute work on | |
0fcb78c2 REB |
1257 | * @wq: workqueue to use |
1258 | * @work: work to queue | |
1259 | * | |
d4283e93 | 1260 | * Returns %false if @work was already on a queue, %true otherwise. |
1da177e4 | 1261 | * |
c1a220e7 ZR |
1262 | * We queue the work to a specific CPU, the caller must ensure it |
1263 | * can't go away. | |
1da177e4 | 1264 | */ |
d4283e93 TH |
1265 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
1266 | struct work_struct *work) | |
1da177e4 | 1267 | { |
d4283e93 | 1268 | bool ret = false; |
8930caba | 1269 | unsigned long flags; |
ef1ca236 | 1270 | |
8930caba | 1271 | local_irq_save(flags); |
c1a220e7 | 1272 | |
22df02bb | 1273 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1274 | __queue_work(cpu, wq, work); |
d4283e93 | 1275 | ret = true; |
c1a220e7 | 1276 | } |
ef1ca236 | 1277 | |
8930caba | 1278 | local_irq_restore(flags); |
1da177e4 LT |
1279 | return ret; |
1280 | } | |
c1a220e7 | 1281 | EXPORT_SYMBOL_GPL(queue_work_on); |
1da177e4 | 1282 | |
c1a220e7 | 1283 | /** |
0a13c00e | 1284 | * queue_work - queue work on a workqueue |
c1a220e7 ZR |
1285 | * @wq: workqueue to use |
1286 | * @work: work to queue | |
1287 | * | |
d4283e93 | 1288 | * Returns %false if @work was already on a queue, %true otherwise. |
c1a220e7 | 1289 | * |
0a13c00e TH |
1290 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
1291 | * it can be processed by another CPU. | |
c1a220e7 | 1292 | */ |
d4283e93 | 1293 | bool queue_work(struct workqueue_struct *wq, struct work_struct *work) |
c1a220e7 | 1294 | { |
57469821 | 1295 | return queue_work_on(WORK_CPU_UNBOUND, wq, work); |
c1a220e7 | 1296 | } |
0a13c00e | 1297 | EXPORT_SYMBOL_GPL(queue_work); |
c1a220e7 | 1298 | |
d8e794df | 1299 | void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1300 | { |
52bad64d | 1301 | struct delayed_work *dwork = (struct delayed_work *)__data; |
7a22ad75 | 1302 | struct cpu_workqueue_struct *cwq = get_work_cwq(&dwork->work); |
1da177e4 | 1303 | |
e0aecdd8 | 1304 | /* should have been called from irqsafe timer with irq already off */ |
1265057f | 1305 | __queue_work(dwork->cpu, cwq->wq, &dwork->work); |
1da177e4 | 1306 | } |
d8e794df | 1307 | EXPORT_SYMBOL_GPL(delayed_work_timer_fn); |
1da177e4 | 1308 | |
7beb2edf TH |
1309 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
1310 | struct delayed_work *dwork, unsigned long delay) | |
1da177e4 | 1311 | { |
7beb2edf TH |
1312 | struct timer_list *timer = &dwork->timer; |
1313 | struct work_struct *work = &dwork->work; | |
1314 | unsigned int lcpu; | |
1315 | ||
1316 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || | |
1317 | timer->data != (unsigned long)dwork); | |
fc4b514f TH |
1318 | WARN_ON_ONCE(timer_pending(timer)); |
1319 | WARN_ON_ONCE(!list_empty(&work->entry)); | |
7beb2edf | 1320 | |
8852aac2 TH |
1321 | /* |
1322 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1323 | * both optimization and correctness. The earliest @timer can | |
1324 | * expire is on the closest next tick and delayed_work users depend | |
1325 | * on that there's no such delay when @delay is 0. | |
1326 | */ | |
1327 | if (!delay) { | |
1328 | __queue_work(cpu, wq, &dwork->work); | |
1329 | return; | |
1330 | } | |
1331 | ||
7beb2edf | 1332 | timer_stats_timer_set_start_info(&dwork->timer); |
1da177e4 | 1333 | |
7beb2edf TH |
1334 | /* |
1335 | * This stores cwq for the moment, for the timer_fn. Note that the | |
1336 | * work's gcwq is preserved to allow reentrance detection for | |
1337 | * delayed works. | |
1338 | */ | |
1339 | if (!(wq->flags & WQ_UNBOUND)) { | |
1340 | struct global_cwq *gcwq = get_work_gcwq(work); | |
1341 | ||
e42986de JK |
1342 | /* |
1343 | * If we cannot get the last gcwq from @work directly, | |
1344 | * select the last CPU such that it avoids unnecessarily | |
1345 | * triggering non-reentrancy check in __queue_work(). | |
1346 | */ | |
1347 | lcpu = cpu; | |
1348 | if (gcwq) | |
7beb2edf | 1349 | lcpu = gcwq->cpu; |
e42986de | 1350 | if (lcpu == WORK_CPU_UNBOUND) |
7beb2edf TH |
1351 | lcpu = raw_smp_processor_id(); |
1352 | } else { | |
1353 | lcpu = WORK_CPU_UNBOUND; | |
1354 | } | |
1355 | ||
1356 | set_work_cwq(work, get_cwq(lcpu, wq), 0); | |
1357 | ||
1265057f | 1358 | dwork->cpu = cpu; |
7beb2edf TH |
1359 | timer->expires = jiffies + delay; |
1360 | ||
1361 | if (unlikely(cpu != WORK_CPU_UNBOUND)) | |
1362 | add_timer_on(timer, cpu); | |
1363 | else | |
1364 | add_timer(timer); | |
1da177e4 LT |
1365 | } |
1366 | ||
0fcb78c2 REB |
1367 | /** |
1368 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1369 | * @cpu: CPU number to execute work on | |
1370 | * @wq: workqueue to use | |
af9997e4 | 1371 | * @dwork: work to queue |
0fcb78c2 REB |
1372 | * @delay: number of jiffies to wait before queueing |
1373 | * | |
715f1300 TH |
1374 | * Returns %false if @work was already on a queue, %true otherwise. If |
1375 | * @delay is zero and @dwork is idle, it will be scheduled for immediate | |
1376 | * execution. | |
0fcb78c2 | 1377 | */ |
d4283e93 TH |
1378 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
1379 | struct delayed_work *dwork, unsigned long delay) | |
7a6bc1cd | 1380 | { |
52bad64d | 1381 | struct work_struct *work = &dwork->work; |
d4283e93 | 1382 | bool ret = false; |
8930caba | 1383 | unsigned long flags; |
7a6bc1cd | 1384 | |
8930caba TH |
1385 | /* read the comment in __queue_work() */ |
1386 | local_irq_save(flags); | |
7a6bc1cd | 1387 | |
22df02bb | 1388 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7beb2edf | 1389 | __queue_delayed_work(cpu, wq, dwork, delay); |
d4283e93 | 1390 | ret = true; |
7a6bc1cd | 1391 | } |
8a3e77cc | 1392 | |
8930caba | 1393 | local_irq_restore(flags); |
7a6bc1cd VP |
1394 | return ret; |
1395 | } | |
ae90dd5d | 1396 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
c7fc77f7 | 1397 | |
0a13c00e TH |
1398 | /** |
1399 | * queue_delayed_work - queue work on a workqueue after delay | |
1400 | * @wq: workqueue to use | |
1401 | * @dwork: delayable work to queue | |
1402 | * @delay: number of jiffies to wait before queueing | |
1403 | * | |
715f1300 | 1404 | * Equivalent to queue_delayed_work_on() but tries to use the local CPU. |
0a13c00e | 1405 | */ |
d4283e93 | 1406 | bool queue_delayed_work(struct workqueue_struct *wq, |
0a13c00e TH |
1407 | struct delayed_work *dwork, unsigned long delay) |
1408 | { | |
57469821 | 1409 | return queue_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); |
0a13c00e TH |
1410 | } |
1411 | EXPORT_SYMBOL_GPL(queue_delayed_work); | |
c7fc77f7 | 1412 | |
8376fe22 TH |
1413 | /** |
1414 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU | |
1415 | * @cpu: CPU number to execute work on | |
1416 | * @wq: workqueue to use | |
1417 | * @dwork: work to queue | |
1418 | * @delay: number of jiffies to wait before queueing | |
1419 | * | |
1420 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, | |
1421 | * modify @dwork's timer so that it expires after @delay. If @delay is | |
1422 | * zero, @work is guaranteed to be scheduled immediately regardless of its | |
1423 | * current state. | |
1424 | * | |
1425 | * Returns %false if @dwork was idle and queued, %true if @dwork was | |
1426 | * pending and its timer was modified. | |
1427 | * | |
e0aecdd8 | 1428 | * This function is safe to call from any context including IRQ handler. |
8376fe22 TH |
1429 | * See try_to_grab_pending() for details. |
1430 | */ | |
1431 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
1432 | struct delayed_work *dwork, unsigned long delay) | |
1433 | { | |
1434 | unsigned long flags; | |
1435 | int ret; | |
c7fc77f7 | 1436 | |
8376fe22 TH |
1437 | do { |
1438 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
1439 | } while (unlikely(ret == -EAGAIN)); | |
63bc0362 | 1440 | |
8376fe22 TH |
1441 | if (likely(ret >= 0)) { |
1442 | __queue_delayed_work(cpu, wq, dwork, delay); | |
1443 | local_irq_restore(flags); | |
7a6bc1cd | 1444 | } |
8376fe22 TH |
1445 | |
1446 | /* -ENOENT from try_to_grab_pending() becomes %true */ | |
7a6bc1cd VP |
1447 | return ret; |
1448 | } | |
8376fe22 TH |
1449 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
1450 | ||
1451 | /** | |
1452 | * mod_delayed_work - modify delay of or queue a delayed work | |
1453 | * @wq: workqueue to use | |
1454 | * @dwork: work to queue | |
1455 | * @delay: number of jiffies to wait before queueing | |
1456 | * | |
1457 | * mod_delayed_work_on() on local CPU. | |
1458 | */ | |
1459 | bool mod_delayed_work(struct workqueue_struct *wq, struct delayed_work *dwork, | |
1460 | unsigned long delay) | |
1461 | { | |
1462 | return mod_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); | |
1463 | } | |
1464 | EXPORT_SYMBOL_GPL(mod_delayed_work); | |
1da177e4 | 1465 | |
c8e55f36 TH |
1466 | /** |
1467 | * worker_enter_idle - enter idle state | |
1468 | * @worker: worker which is entering idle state | |
1469 | * | |
1470 | * @worker is entering idle state. Update stats and idle timer if | |
1471 | * necessary. | |
1472 | * | |
1473 | * LOCKING: | |
1474 | * spin_lock_irq(gcwq->lock). | |
1475 | */ | |
1476 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1477 | { |
bd7bdd43 | 1478 | struct worker_pool *pool = worker->pool; |
c8e55f36 TH |
1479 | |
1480 | BUG_ON(worker->flags & WORKER_IDLE); | |
1481 | BUG_ON(!list_empty(&worker->entry) && | |
1482 | (worker->hentry.next || worker->hentry.pprev)); | |
1483 | ||
cb444766 TH |
1484 | /* can't use worker_set_flags(), also called from start_worker() */ |
1485 | worker->flags |= WORKER_IDLE; | |
bd7bdd43 | 1486 | pool->nr_idle++; |
e22bee78 | 1487 | worker->last_active = jiffies; |
c8e55f36 TH |
1488 | |
1489 | /* idle_list is LIFO */ | |
bd7bdd43 | 1490 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1491 | |
628c78e7 TH |
1492 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
1493 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); | |
cb444766 | 1494 | |
544ecf31 | 1495 | /* |
628c78e7 TH |
1496 | * Sanity check nr_running. Because gcwq_unbind_fn() releases |
1497 | * gcwq->lock between setting %WORKER_UNBOUND and zapping | |
1498 | * nr_running, the warning may trigger spuriously. Check iff | |
1499 | * unbind is not in progress. | |
544ecf31 | 1500 | */ |
24647570 | 1501 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
bd7bdd43 | 1502 | pool->nr_workers == pool->nr_idle && |
63d95a91 | 1503 | atomic_read(get_pool_nr_running(pool))); |
c8e55f36 TH |
1504 | } |
1505 | ||
1506 | /** | |
1507 | * worker_leave_idle - leave idle state | |
1508 | * @worker: worker which is leaving idle state | |
1509 | * | |
1510 | * @worker is leaving idle state. Update stats. | |
1511 | * | |
1512 | * LOCKING: | |
1513 | * spin_lock_irq(gcwq->lock). | |
1514 | */ | |
1515 | static void worker_leave_idle(struct worker *worker) | |
1516 | { | |
bd7bdd43 | 1517 | struct worker_pool *pool = worker->pool; |
c8e55f36 TH |
1518 | |
1519 | BUG_ON(!(worker->flags & WORKER_IDLE)); | |
d302f017 | 1520 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1521 | pool->nr_idle--; |
c8e55f36 TH |
1522 | list_del_init(&worker->entry); |
1523 | } | |
1524 | ||
e22bee78 TH |
1525 | /** |
1526 | * worker_maybe_bind_and_lock - bind worker to its cpu if possible and lock gcwq | |
1527 | * @worker: self | |
1528 | * | |
1529 | * Works which are scheduled while the cpu is online must at least be | |
1530 | * scheduled to a worker which is bound to the cpu so that if they are | |
1531 | * flushed from cpu callbacks while cpu is going down, they are | |
1532 | * guaranteed to execute on the cpu. | |
1533 | * | |
1534 | * This function is to be used by rogue workers and rescuers to bind | |
1535 | * themselves to the target cpu and may race with cpu going down or | |
1536 | * coming online. kthread_bind() can't be used because it may put the | |
1537 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
1538 | * verbatim as it's best effort and blocking and gcwq may be | |
1539 | * [dis]associated in the meantime. | |
1540 | * | |
f2d5a0ee | 1541 | * This function tries set_cpus_allowed() and locks gcwq and verifies the |
24647570 | 1542 | * binding against %POOL_DISASSOCIATED which is set during |
f2d5a0ee TH |
1543 | * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker |
1544 | * enters idle state or fetches works without dropping lock, it can | |
1545 | * guarantee the scheduling requirement described in the first paragraph. | |
e22bee78 TH |
1546 | * |
1547 | * CONTEXT: | |
1548 | * Might sleep. Called without any lock but returns with gcwq->lock | |
1549 | * held. | |
1550 | * | |
1551 | * RETURNS: | |
1552 | * %true if the associated gcwq is online (@worker is successfully | |
1553 | * bound), %false if offline. | |
1554 | */ | |
1555 | static bool worker_maybe_bind_and_lock(struct worker *worker) | |
972fa1c5 | 1556 | __acquires(&gcwq->lock) |
e22bee78 | 1557 | { |
24647570 TH |
1558 | struct worker_pool *pool = worker->pool; |
1559 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
1560 | struct task_struct *task = worker->task; |
1561 | ||
1562 | while (true) { | |
4e6045f1 | 1563 | /* |
e22bee78 TH |
1564 | * The following call may fail, succeed or succeed |
1565 | * without actually migrating the task to the cpu if | |
1566 | * it races with cpu hotunplug operation. Verify | |
24647570 | 1567 | * against POOL_DISASSOCIATED. |
4e6045f1 | 1568 | */ |
24647570 | 1569 | if (!(pool->flags & POOL_DISASSOCIATED)) |
f3421797 | 1570 | set_cpus_allowed_ptr(task, get_cpu_mask(gcwq->cpu)); |
e22bee78 TH |
1571 | |
1572 | spin_lock_irq(&gcwq->lock); | |
24647570 | 1573 | if (pool->flags & POOL_DISASSOCIATED) |
e22bee78 TH |
1574 | return false; |
1575 | if (task_cpu(task) == gcwq->cpu && | |
1576 | cpumask_equal(¤t->cpus_allowed, | |
1577 | get_cpu_mask(gcwq->cpu))) | |
1578 | return true; | |
1579 | spin_unlock_irq(&gcwq->lock); | |
1580 | ||
5035b20f TH |
1581 | /* |
1582 | * We've raced with CPU hot[un]plug. Give it a breather | |
1583 | * and retry migration. cond_resched() is required here; | |
1584 | * otherwise, we might deadlock against cpu_stop trying to | |
1585 | * bring down the CPU on non-preemptive kernel. | |
1586 | */ | |
e22bee78 | 1587 | cpu_relax(); |
5035b20f | 1588 | cond_resched(); |
e22bee78 TH |
1589 | } |
1590 | } | |
1591 | ||
25511a47 | 1592 | /* |
ea1abd61 | 1593 | * Rebind an idle @worker to its CPU. worker_thread() will test |
5f7dabfd | 1594 | * list_empty(@worker->entry) before leaving idle and call this function. |
25511a47 TH |
1595 | */ |
1596 | static void idle_worker_rebind(struct worker *worker) | |
1597 | { | |
1598 | struct global_cwq *gcwq = worker->pool->gcwq; | |
1599 | ||
5f7dabfd LJ |
1600 | /* CPU may go down again inbetween, clear UNBOUND only on success */ |
1601 | if (worker_maybe_bind_and_lock(worker)) | |
1602 | worker_clr_flags(worker, WORKER_UNBOUND); | |
25511a47 | 1603 | |
ea1abd61 LJ |
1604 | /* rebind complete, become available again */ |
1605 | list_add(&worker->entry, &worker->pool->idle_list); | |
1606 | spin_unlock_irq(&gcwq->lock); | |
25511a47 TH |
1607 | } |
1608 | ||
e22bee78 | 1609 | /* |
25511a47 | 1610 | * Function for @worker->rebind.work used to rebind unbound busy workers to |
403c821d TH |
1611 | * the associated cpu which is coming back online. This is scheduled by |
1612 | * cpu up but can race with other cpu hotplug operations and may be | |
1613 | * executed twice without intervening cpu down. | |
e22bee78 | 1614 | */ |
25511a47 | 1615 | static void busy_worker_rebind_fn(struct work_struct *work) |
e22bee78 TH |
1616 | { |
1617 | struct worker *worker = container_of(work, struct worker, rebind_work); | |
bd7bdd43 | 1618 | struct global_cwq *gcwq = worker->pool->gcwq; |
e22bee78 | 1619 | |
eab6d828 LJ |
1620 | if (worker_maybe_bind_and_lock(worker)) |
1621 | worker_clr_flags(worker, WORKER_UNBOUND); | |
e22bee78 TH |
1622 | |
1623 | spin_unlock_irq(&gcwq->lock); | |
1624 | } | |
1625 | ||
25511a47 TH |
1626 | /** |
1627 | * rebind_workers - rebind all workers of a gcwq to the associated CPU | |
1628 | * @gcwq: gcwq of interest | |
1629 | * | |
1630 | * @gcwq->cpu is coming online. Rebind all workers to the CPU. Rebinding | |
1631 | * is different for idle and busy ones. | |
1632 | * | |
ea1abd61 LJ |
1633 | * Idle ones will be removed from the idle_list and woken up. They will |
1634 | * add themselves back after completing rebind. This ensures that the | |
1635 | * idle_list doesn't contain any unbound workers when re-bound busy workers | |
1636 | * try to perform local wake-ups for concurrency management. | |
25511a47 | 1637 | * |
ea1abd61 LJ |
1638 | * Busy workers can rebind after they finish their current work items. |
1639 | * Queueing the rebind work item at the head of the scheduled list is | |
1640 | * enough. Note that nr_running will be properly bumped as busy workers | |
1641 | * rebind. | |
25511a47 | 1642 | * |
ea1abd61 LJ |
1643 | * On return, all non-manager workers are scheduled for rebind - see |
1644 | * manage_workers() for the manager special case. Any idle worker | |
1645 | * including the manager will not appear on @idle_list until rebind is | |
1646 | * complete, making local wake-ups safe. | |
25511a47 TH |
1647 | */ |
1648 | static void rebind_workers(struct global_cwq *gcwq) | |
25511a47 | 1649 | { |
25511a47 | 1650 | struct worker_pool *pool; |
ea1abd61 | 1651 | struct worker *worker, *n; |
25511a47 TH |
1652 | struct hlist_node *pos; |
1653 | int i; | |
1654 | ||
1655 | lockdep_assert_held(&gcwq->lock); | |
1656 | ||
1657 | for_each_worker_pool(pool, gcwq) | |
b2eb83d1 | 1658 | lockdep_assert_held(&pool->assoc_mutex); |
25511a47 | 1659 | |
5f7dabfd | 1660 | /* dequeue and kick idle ones */ |
25511a47 | 1661 | for_each_worker_pool(pool, gcwq) { |
ea1abd61 | 1662 | list_for_each_entry_safe(worker, n, &pool->idle_list, entry) { |
ea1abd61 LJ |
1663 | /* |
1664 | * idle workers should be off @pool->idle_list | |
1665 | * until rebind is complete to avoid receiving | |
1666 | * premature local wake-ups. | |
1667 | */ | |
1668 | list_del_init(&worker->entry); | |
25511a47 | 1669 | |
5f7dabfd LJ |
1670 | /* |
1671 | * worker_thread() will see the above dequeuing | |
1672 | * and call idle_worker_rebind(). | |
1673 | */ | |
25511a47 TH |
1674 | wake_up_process(worker->task); |
1675 | } | |
1676 | } | |
1677 | ||
ea1abd61 | 1678 | /* rebind busy workers */ |
25511a47 TH |
1679 | for_each_busy_worker(worker, i, pos, gcwq) { |
1680 | struct work_struct *rebind_work = &worker->rebind_work; | |
e2b6a6d5 | 1681 | struct workqueue_struct *wq; |
25511a47 TH |
1682 | |
1683 | if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, | |
1684 | work_data_bits(rebind_work))) | |
1685 | continue; | |
1686 | ||
25511a47 | 1687 | debug_work_activate(rebind_work); |
90beca5d | 1688 | |
e2b6a6d5 JK |
1689 | /* |
1690 | * wq doesn't really matter but let's keep @worker->pool | |
1691 | * and @cwq->pool consistent for sanity. | |
1692 | */ | |
e34cdddb | 1693 | if (std_worker_pool_pri(worker->pool)) |
e2b6a6d5 JK |
1694 | wq = system_highpri_wq; |
1695 | else | |
1696 | wq = system_wq; | |
ec58815a | 1697 | |
e2b6a6d5 JK |
1698 | insert_work(get_cwq(gcwq->cpu, wq), rebind_work, |
1699 | worker->scheduled.next, | |
1700 | work_color_to_flags(WORK_NO_COLOR)); | |
ec58815a | 1701 | } |
25511a47 TH |
1702 | } |
1703 | ||
c34056a3 TH |
1704 | static struct worker *alloc_worker(void) |
1705 | { | |
1706 | struct worker *worker; | |
1707 | ||
1708 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1709 | if (worker) { |
1710 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1711 | INIT_LIST_HEAD(&worker->scheduled); |
25511a47 | 1712 | INIT_WORK(&worker->rebind_work, busy_worker_rebind_fn); |
e22bee78 TH |
1713 | /* on creation a worker is in !idle && prep state */ |
1714 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1715 | } |
c34056a3 TH |
1716 | return worker; |
1717 | } | |
1718 | ||
1719 | /** | |
1720 | * create_worker - create a new workqueue worker | |
63d95a91 | 1721 | * @pool: pool the new worker will belong to |
c34056a3 | 1722 | * |
63d95a91 | 1723 | * Create a new worker which is bound to @pool. The returned worker |
c34056a3 TH |
1724 | * can be started by calling start_worker() or destroyed using |
1725 | * destroy_worker(). | |
1726 | * | |
1727 | * CONTEXT: | |
1728 | * Might sleep. Does GFP_KERNEL allocations. | |
1729 | * | |
1730 | * RETURNS: | |
1731 | * Pointer to the newly created worker. | |
1732 | */ | |
bc2ae0f5 | 1733 | static struct worker *create_worker(struct worker_pool *pool) |
c34056a3 | 1734 | { |
63d95a91 | 1735 | struct global_cwq *gcwq = pool->gcwq; |
e34cdddb | 1736 | const char *pri = std_worker_pool_pri(pool) ? "H" : ""; |
c34056a3 | 1737 | struct worker *worker = NULL; |
f3421797 | 1738 | int id = -1; |
c34056a3 | 1739 | |
8b03ae3c | 1740 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1741 | while (ida_get_new(&pool->worker_ida, &id)) { |
8b03ae3c | 1742 | spin_unlock_irq(&gcwq->lock); |
bd7bdd43 | 1743 | if (!ida_pre_get(&pool->worker_ida, GFP_KERNEL)) |
c34056a3 | 1744 | goto fail; |
8b03ae3c | 1745 | spin_lock_irq(&gcwq->lock); |
c34056a3 | 1746 | } |
8b03ae3c | 1747 | spin_unlock_irq(&gcwq->lock); |
c34056a3 TH |
1748 | |
1749 | worker = alloc_worker(); | |
1750 | if (!worker) | |
1751 | goto fail; | |
1752 | ||
bd7bdd43 | 1753 | worker->pool = pool; |
c34056a3 TH |
1754 | worker->id = id; |
1755 | ||
bc2ae0f5 | 1756 | if (gcwq->cpu != WORK_CPU_UNBOUND) |
94dcf29a | 1757 | worker->task = kthread_create_on_node(worker_thread, |
3270476a TH |
1758 | worker, cpu_to_node(gcwq->cpu), |
1759 | "kworker/%u:%d%s", gcwq->cpu, id, pri); | |
f3421797 TH |
1760 | else |
1761 | worker->task = kthread_create(worker_thread, worker, | |
3270476a | 1762 | "kworker/u:%d%s", id, pri); |
c34056a3 TH |
1763 | if (IS_ERR(worker->task)) |
1764 | goto fail; | |
1765 | ||
e34cdddb | 1766 | if (std_worker_pool_pri(pool)) |
3270476a TH |
1767 | set_user_nice(worker->task, HIGHPRI_NICE_LEVEL); |
1768 | ||
db7bccf4 | 1769 | /* |
bc2ae0f5 | 1770 | * Determine CPU binding of the new worker depending on |
24647570 | 1771 | * %POOL_DISASSOCIATED. The caller is responsible for ensuring the |
bc2ae0f5 TH |
1772 | * flag remains stable across this function. See the comments |
1773 | * above the flag definition for details. | |
1774 | * | |
1775 | * As an unbound worker may later become a regular one if CPU comes | |
1776 | * online, make sure every worker has %PF_THREAD_BOUND set. | |
db7bccf4 | 1777 | */ |
24647570 | 1778 | if (!(pool->flags & POOL_DISASSOCIATED)) { |
8b03ae3c | 1779 | kthread_bind(worker->task, gcwq->cpu); |
bc2ae0f5 | 1780 | } else { |
db7bccf4 | 1781 | worker->task->flags |= PF_THREAD_BOUND; |
bc2ae0f5 | 1782 | worker->flags |= WORKER_UNBOUND; |
f3421797 | 1783 | } |
c34056a3 TH |
1784 | |
1785 | return worker; | |
1786 | fail: | |
1787 | if (id >= 0) { | |
8b03ae3c | 1788 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1789 | ida_remove(&pool->worker_ida, id); |
8b03ae3c | 1790 | spin_unlock_irq(&gcwq->lock); |
c34056a3 TH |
1791 | } |
1792 | kfree(worker); | |
1793 | return NULL; | |
1794 | } | |
1795 | ||
1796 | /** | |
1797 | * start_worker - start a newly created worker | |
1798 | * @worker: worker to start | |
1799 | * | |
c8e55f36 | 1800 | * Make the gcwq aware of @worker and start it. |
c34056a3 TH |
1801 | * |
1802 | * CONTEXT: | |
8b03ae3c | 1803 | * spin_lock_irq(gcwq->lock). |
c34056a3 TH |
1804 | */ |
1805 | static void start_worker(struct worker *worker) | |
1806 | { | |
cb444766 | 1807 | worker->flags |= WORKER_STARTED; |
bd7bdd43 | 1808 | worker->pool->nr_workers++; |
c8e55f36 | 1809 | worker_enter_idle(worker); |
c34056a3 TH |
1810 | wake_up_process(worker->task); |
1811 | } | |
1812 | ||
1813 | /** | |
1814 | * destroy_worker - destroy a workqueue worker | |
1815 | * @worker: worker to be destroyed | |
1816 | * | |
c8e55f36 TH |
1817 | * Destroy @worker and adjust @gcwq stats accordingly. |
1818 | * | |
1819 | * CONTEXT: | |
1820 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. | |
c34056a3 TH |
1821 | */ |
1822 | static void destroy_worker(struct worker *worker) | |
1823 | { | |
bd7bdd43 TH |
1824 | struct worker_pool *pool = worker->pool; |
1825 | struct global_cwq *gcwq = pool->gcwq; | |
c34056a3 TH |
1826 | int id = worker->id; |
1827 | ||
1828 | /* sanity check frenzy */ | |
1829 | BUG_ON(worker->current_work); | |
affee4b2 | 1830 | BUG_ON(!list_empty(&worker->scheduled)); |
c34056a3 | 1831 | |
c8e55f36 | 1832 | if (worker->flags & WORKER_STARTED) |
bd7bdd43 | 1833 | pool->nr_workers--; |
c8e55f36 | 1834 | if (worker->flags & WORKER_IDLE) |
bd7bdd43 | 1835 | pool->nr_idle--; |
c8e55f36 TH |
1836 | |
1837 | list_del_init(&worker->entry); | |
cb444766 | 1838 | worker->flags |= WORKER_DIE; |
c8e55f36 TH |
1839 | |
1840 | spin_unlock_irq(&gcwq->lock); | |
1841 | ||
c34056a3 TH |
1842 | kthread_stop(worker->task); |
1843 | kfree(worker); | |
1844 | ||
8b03ae3c | 1845 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1846 | ida_remove(&pool->worker_ida, id); |
c34056a3 TH |
1847 | } |
1848 | ||
63d95a91 | 1849 | static void idle_worker_timeout(unsigned long __pool) |
e22bee78 | 1850 | { |
63d95a91 TH |
1851 | struct worker_pool *pool = (void *)__pool; |
1852 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
1853 | |
1854 | spin_lock_irq(&gcwq->lock); | |
1855 | ||
63d95a91 | 1856 | if (too_many_workers(pool)) { |
e22bee78 TH |
1857 | struct worker *worker; |
1858 | unsigned long expires; | |
1859 | ||
1860 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1861 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1862 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1863 | ||
1864 | if (time_before(jiffies, expires)) | |
63d95a91 | 1865 | mod_timer(&pool->idle_timer, expires); |
e22bee78 TH |
1866 | else { |
1867 | /* it's been idle for too long, wake up manager */ | |
11ebea50 | 1868 | pool->flags |= POOL_MANAGE_WORKERS; |
63d95a91 | 1869 | wake_up_worker(pool); |
d5abe669 | 1870 | } |
e22bee78 TH |
1871 | } |
1872 | ||
1873 | spin_unlock_irq(&gcwq->lock); | |
1874 | } | |
d5abe669 | 1875 | |
e22bee78 TH |
1876 | static bool send_mayday(struct work_struct *work) |
1877 | { | |
1878 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); | |
1879 | struct workqueue_struct *wq = cwq->wq; | |
f3421797 | 1880 | unsigned int cpu; |
e22bee78 TH |
1881 | |
1882 | if (!(wq->flags & WQ_RESCUER)) | |
1883 | return false; | |
1884 | ||
1885 | /* mayday mayday mayday */ | |
bd7bdd43 | 1886 | cpu = cwq->pool->gcwq->cpu; |
f3421797 TH |
1887 | /* WORK_CPU_UNBOUND can't be set in cpumask, use cpu 0 instead */ |
1888 | if (cpu == WORK_CPU_UNBOUND) | |
1889 | cpu = 0; | |
f2e005aa | 1890 | if (!mayday_test_and_set_cpu(cpu, wq->mayday_mask)) |
e22bee78 TH |
1891 | wake_up_process(wq->rescuer->task); |
1892 | return true; | |
1893 | } | |
1894 | ||
63d95a91 | 1895 | static void gcwq_mayday_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 | struct work_struct *work; |
1900 | ||
1901 | spin_lock_irq(&gcwq->lock); | |
1902 | ||
63d95a91 | 1903 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
1904 | /* |
1905 | * We've been trying to create a new worker but | |
1906 | * haven't been successful. We might be hitting an | |
1907 | * allocation deadlock. Send distress signals to | |
1908 | * rescuers. | |
1909 | */ | |
63d95a91 | 1910 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 1911 | send_mayday(work); |
1da177e4 | 1912 | } |
e22bee78 TH |
1913 | |
1914 | spin_unlock_irq(&gcwq->lock); | |
1915 | ||
63d95a91 | 1916 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
1917 | } |
1918 | ||
e22bee78 TH |
1919 | /** |
1920 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 1921 | * @pool: pool to create a new worker for |
e22bee78 | 1922 | * |
63d95a91 | 1923 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
1924 | * have at least one idle worker on return from this function. If |
1925 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 1926 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
1927 | * possible allocation deadlock. |
1928 | * | |
1929 | * On return, need_to_create_worker() is guaranteed to be false and | |
1930 | * may_start_working() true. | |
1931 | * | |
1932 | * LOCKING: | |
1933 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1934 | * multiple times. Does GFP_KERNEL allocations. Called only from | |
1935 | * manager. | |
1936 | * | |
1937 | * RETURNS: | |
1938 | * false if no action was taken and gcwq->lock stayed locked, true | |
1939 | * otherwise. | |
1940 | */ | |
63d95a91 | 1941 | static bool maybe_create_worker(struct worker_pool *pool) |
06bd6ebf NK |
1942 | __releases(&gcwq->lock) |
1943 | __acquires(&gcwq->lock) | |
1da177e4 | 1944 | { |
63d95a91 TH |
1945 | struct global_cwq *gcwq = pool->gcwq; |
1946 | ||
1947 | if (!need_to_create_worker(pool)) | |
e22bee78 TH |
1948 | return false; |
1949 | restart: | |
9f9c2364 TH |
1950 | spin_unlock_irq(&gcwq->lock); |
1951 | ||
e22bee78 | 1952 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 1953 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
1954 | |
1955 | while (true) { | |
1956 | struct worker *worker; | |
1957 | ||
bc2ae0f5 | 1958 | worker = create_worker(pool); |
e22bee78 | 1959 | if (worker) { |
63d95a91 | 1960 | del_timer_sync(&pool->mayday_timer); |
e22bee78 TH |
1961 | spin_lock_irq(&gcwq->lock); |
1962 | start_worker(worker); | |
63d95a91 | 1963 | BUG_ON(need_to_create_worker(pool)); |
e22bee78 TH |
1964 | return true; |
1965 | } | |
1966 | ||
63d95a91 | 1967 | if (!need_to_create_worker(pool)) |
e22bee78 | 1968 | break; |
1da177e4 | 1969 | |
e22bee78 TH |
1970 | __set_current_state(TASK_INTERRUPTIBLE); |
1971 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1972 | |
63d95a91 | 1973 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1974 | break; |
1975 | } | |
1976 | ||
63d95a91 | 1977 | del_timer_sync(&pool->mayday_timer); |
e22bee78 | 1978 | spin_lock_irq(&gcwq->lock); |
63d95a91 | 1979 | if (need_to_create_worker(pool)) |
e22bee78 TH |
1980 | goto restart; |
1981 | return true; | |
1982 | } | |
1983 | ||
1984 | /** | |
1985 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
63d95a91 | 1986 | * @pool: pool to destroy workers for |
e22bee78 | 1987 | * |
63d95a91 | 1988 | * Destroy @pool workers which have been idle for longer than |
e22bee78 TH |
1989 | * IDLE_WORKER_TIMEOUT. |
1990 | * | |
1991 | * LOCKING: | |
1992 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1993 | * multiple times. Called only from manager. | |
1994 | * | |
1995 | * RETURNS: | |
1996 | * false if no action was taken and gcwq->lock stayed locked, true | |
1997 | * otherwise. | |
1998 | */ | |
63d95a91 | 1999 | static bool maybe_destroy_workers(struct worker_pool *pool) |
e22bee78 TH |
2000 | { |
2001 | bool ret = false; | |
1da177e4 | 2002 | |
63d95a91 | 2003 | while (too_many_workers(pool)) { |
e22bee78 TH |
2004 | struct worker *worker; |
2005 | unsigned long expires; | |
3af24433 | 2006 | |
63d95a91 | 2007 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 | 2008 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
85f4186a | 2009 | |
e22bee78 | 2010 | if (time_before(jiffies, expires)) { |
63d95a91 | 2011 | mod_timer(&pool->idle_timer, expires); |
3af24433 | 2012 | break; |
e22bee78 | 2013 | } |
1da177e4 | 2014 | |
e22bee78 TH |
2015 | destroy_worker(worker); |
2016 | ret = true; | |
1da177e4 | 2017 | } |
1e19ffc6 | 2018 | |
e22bee78 | 2019 | return ret; |
1e19ffc6 TH |
2020 | } |
2021 | ||
73f53c4a | 2022 | /** |
e22bee78 TH |
2023 | * manage_workers - manage worker pool |
2024 | * @worker: self | |
73f53c4a | 2025 | * |
e22bee78 TH |
2026 | * Assume the manager role and manage gcwq worker pool @worker belongs |
2027 | * to. At any given time, there can be only zero or one manager per | |
2028 | * gcwq. The exclusion is handled automatically by this function. | |
2029 | * | |
2030 | * The caller can safely start processing works on false return. On | |
2031 | * true return, it's guaranteed that need_to_create_worker() is false | |
2032 | * and may_start_working() is true. | |
73f53c4a TH |
2033 | * |
2034 | * CONTEXT: | |
e22bee78 TH |
2035 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed |
2036 | * multiple times. Does GFP_KERNEL allocations. | |
2037 | * | |
2038 | * RETURNS: | |
2039 | * false if no action was taken and gcwq->lock stayed locked, true if | |
2040 | * some action was taken. | |
73f53c4a | 2041 | */ |
e22bee78 | 2042 | static bool manage_workers(struct worker *worker) |
73f53c4a | 2043 | { |
63d95a91 | 2044 | struct worker_pool *pool = worker->pool; |
e22bee78 | 2045 | bool ret = false; |
73f53c4a | 2046 | |
ee378aa4 | 2047 | if (pool->flags & POOL_MANAGING_WORKERS) |
e22bee78 | 2048 | return ret; |
1e19ffc6 | 2049 | |
552a37e9 | 2050 | pool->flags |= POOL_MANAGING_WORKERS; |
73f53c4a | 2051 | |
ee378aa4 LJ |
2052 | /* |
2053 | * To simplify both worker management and CPU hotplug, hold off | |
2054 | * management while hotplug is in progress. CPU hotplug path can't | |
2055 | * grab %POOL_MANAGING_WORKERS to achieve this because that can | |
2056 | * lead to idle worker depletion (all become busy thinking someone | |
2057 | * else is managing) which in turn can result in deadlock under | |
b2eb83d1 | 2058 | * extreme circumstances. Use @pool->assoc_mutex to synchronize |
ee378aa4 LJ |
2059 | * manager against CPU hotplug. |
2060 | * | |
b2eb83d1 | 2061 | * assoc_mutex would always be free unless CPU hotplug is in |
ee378aa4 LJ |
2062 | * progress. trylock first without dropping @gcwq->lock. |
2063 | */ | |
b2eb83d1 | 2064 | if (unlikely(!mutex_trylock(&pool->assoc_mutex))) { |
ee378aa4 | 2065 | spin_unlock_irq(&pool->gcwq->lock); |
b2eb83d1 | 2066 | mutex_lock(&pool->assoc_mutex); |
ee378aa4 LJ |
2067 | /* |
2068 | * CPU hotplug could have happened while we were waiting | |
b2eb83d1 | 2069 | * for assoc_mutex. Hotplug itself can't handle us |
ee378aa4 LJ |
2070 | * because manager isn't either on idle or busy list, and |
2071 | * @gcwq's state and ours could have deviated. | |
2072 | * | |
b2eb83d1 | 2073 | * As hotplug is now excluded via assoc_mutex, we can |
ee378aa4 LJ |
2074 | * simply try to bind. It will succeed or fail depending |
2075 | * on @gcwq's current state. Try it and adjust | |
2076 | * %WORKER_UNBOUND accordingly. | |
2077 | */ | |
2078 | if (worker_maybe_bind_and_lock(worker)) | |
2079 | worker->flags &= ~WORKER_UNBOUND; | |
2080 | else | |
2081 | worker->flags |= WORKER_UNBOUND; | |
73f53c4a | 2082 | |
ee378aa4 LJ |
2083 | ret = true; |
2084 | } | |
73f53c4a | 2085 | |
11ebea50 | 2086 | pool->flags &= ~POOL_MANAGE_WORKERS; |
73f53c4a TH |
2087 | |
2088 | /* | |
e22bee78 TH |
2089 | * Destroy and then create so that may_start_working() is true |
2090 | * on return. | |
73f53c4a | 2091 | */ |
63d95a91 TH |
2092 | ret |= maybe_destroy_workers(pool); |
2093 | ret |= maybe_create_worker(pool); | |
e22bee78 | 2094 | |
552a37e9 | 2095 | pool->flags &= ~POOL_MANAGING_WORKERS; |
b2eb83d1 | 2096 | mutex_unlock(&pool->assoc_mutex); |
e22bee78 | 2097 | return ret; |
73f53c4a TH |
2098 | } |
2099 | ||
a62428c0 TH |
2100 | /** |
2101 | * process_one_work - process single work | |
c34056a3 | 2102 | * @worker: self |
a62428c0 TH |
2103 | * @work: work to process |
2104 | * | |
2105 | * Process @work. This function contains all the logics necessary to | |
2106 | * process a single work including synchronization against and | |
2107 | * interaction with other workers on the same cpu, queueing and | |
2108 | * flushing. As long as context requirement is met, any worker can | |
2109 | * call this function to process a work. | |
2110 | * | |
2111 | * CONTEXT: | |
8b03ae3c | 2112 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. |
a62428c0 | 2113 | */ |
c34056a3 | 2114 | static void process_one_work(struct worker *worker, struct work_struct *work) |
06bd6ebf NK |
2115 | __releases(&gcwq->lock) |
2116 | __acquires(&gcwq->lock) | |
a62428c0 | 2117 | { |
7e11629d | 2118 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); |
bd7bdd43 TH |
2119 | struct worker_pool *pool = worker->pool; |
2120 | struct global_cwq *gcwq = pool->gcwq; | |
fb0e7beb | 2121 | bool cpu_intensive = cwq->wq->flags & WQ_CPU_INTENSIVE; |
73f53c4a | 2122 | int work_color; |
7e11629d | 2123 | struct worker *collision; |
a62428c0 TH |
2124 | #ifdef CONFIG_LOCKDEP |
2125 | /* | |
2126 | * It is permissible to free the struct work_struct from | |
2127 | * inside the function that is called from it, this we need to | |
2128 | * take into account for lockdep too. To avoid bogus "held | |
2129 | * lock freed" warnings as well as problems when looking into | |
2130 | * work->lockdep_map, make a copy and use that here. | |
2131 | */ | |
4d82a1de PZ |
2132 | struct lockdep_map lockdep_map; |
2133 | ||
2134 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 2135 | #endif |
6fec10a1 TH |
2136 | /* |
2137 | * Ensure we're on the correct CPU. DISASSOCIATED test is | |
2138 | * necessary to avoid spurious warnings from rescuers servicing the | |
24647570 | 2139 | * unbound or a disassociated pool. |
6fec10a1 | 2140 | */ |
5f7dabfd | 2141 | WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) && |
24647570 | 2142 | !(pool->flags & POOL_DISASSOCIATED) && |
25511a47 TH |
2143 | raw_smp_processor_id() != gcwq->cpu); |
2144 | ||
7e11629d TH |
2145 | /* |
2146 | * A single work shouldn't be executed concurrently by | |
2147 | * multiple workers on a single cpu. Check whether anyone is | |
2148 | * already processing the work. If so, defer the work to the | |
2149 | * currently executing one. | |
2150 | */ | |
42f8570f | 2151 | collision = find_worker_executing_work(gcwq, work); |
7e11629d TH |
2152 | if (unlikely(collision)) { |
2153 | move_linked_works(work, &collision->scheduled, NULL); | |
2154 | return; | |
2155 | } | |
2156 | ||
8930caba | 2157 | /* claim and dequeue */ |
a62428c0 | 2158 | debug_work_deactivate(work); |
023f27d3 | 2159 | hash_add(gcwq->busy_hash, &worker->hentry, (unsigned long)work); |
c34056a3 | 2160 | worker->current_work = work; |
a2c1c57b | 2161 | worker->current_func = work->func; |
8cca0eea | 2162 | worker->current_cwq = cwq; |
73f53c4a | 2163 | work_color = get_work_color(work); |
7a22ad75 | 2164 | |
a62428c0 TH |
2165 | list_del_init(&work->entry); |
2166 | ||
fb0e7beb TH |
2167 | /* |
2168 | * CPU intensive works don't participate in concurrency | |
2169 | * management. They're the scheduler's responsibility. | |
2170 | */ | |
2171 | if (unlikely(cpu_intensive)) | |
2172 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
2173 | ||
974271c4 TH |
2174 | /* |
2175 | * Unbound gcwq isn't concurrency managed and work items should be | |
2176 | * executed ASAP. Wake up another worker if necessary. | |
2177 | */ | |
63d95a91 TH |
2178 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
2179 | wake_up_worker(pool); | |
974271c4 | 2180 | |
8930caba | 2181 | /* |
23657bb1 TH |
2182 | * Record the last CPU and clear PENDING which should be the last |
2183 | * update to @work. Also, do this inside @gcwq->lock so that | |
2184 | * PENDING and queued state changes happen together while IRQ is | |
2185 | * disabled. | |
8930caba | 2186 | */ |
8930caba | 2187 | set_work_cpu_and_clear_pending(work, gcwq->cpu); |
a62428c0 | 2188 | |
8b03ae3c | 2189 | spin_unlock_irq(&gcwq->lock); |
a62428c0 | 2190 | |
e159489b | 2191 | lock_map_acquire_read(&cwq->wq->lockdep_map); |
a62428c0 | 2192 | lock_map_acquire(&lockdep_map); |
e36c886a | 2193 | trace_workqueue_execute_start(work); |
a2c1c57b | 2194 | worker->current_func(work); |
e36c886a AV |
2195 | /* |
2196 | * While we must be careful to not use "work" after this, the trace | |
2197 | * point will only record its address. | |
2198 | */ | |
2199 | trace_workqueue_execute_end(work); | |
a62428c0 TH |
2200 | lock_map_release(&lockdep_map); |
2201 | lock_map_release(&cwq->wq->lockdep_map); | |
2202 | ||
2203 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
044c782c VI |
2204 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
2205 | " last function: %pf\n", | |
a2c1c57b TH |
2206 | current->comm, preempt_count(), task_pid_nr(current), |
2207 | worker->current_func); | |
a62428c0 TH |
2208 | debug_show_held_locks(current); |
2209 | dump_stack(); | |
2210 | } | |
2211 | ||
8b03ae3c | 2212 | spin_lock_irq(&gcwq->lock); |
a62428c0 | 2213 | |
fb0e7beb TH |
2214 | /* clear cpu intensive status */ |
2215 | if (unlikely(cpu_intensive)) | |
2216 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
2217 | ||
a62428c0 | 2218 | /* we're done with it, release */ |
42f8570f | 2219 | hash_del(&worker->hentry); |
c34056a3 | 2220 | worker->current_work = NULL; |
a2c1c57b | 2221 | worker->current_func = NULL; |
8cca0eea | 2222 | worker->current_cwq = NULL; |
b3f9f405 | 2223 | cwq_dec_nr_in_flight(cwq, work_color); |
a62428c0 TH |
2224 | } |
2225 | ||
affee4b2 TH |
2226 | /** |
2227 | * process_scheduled_works - process scheduled works | |
2228 | * @worker: self | |
2229 | * | |
2230 | * Process all scheduled works. Please note that the scheduled list | |
2231 | * may change while processing a work, so this function repeatedly | |
2232 | * fetches a work from the top and executes it. | |
2233 | * | |
2234 | * CONTEXT: | |
8b03ae3c | 2235 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed |
affee4b2 TH |
2236 | * multiple times. |
2237 | */ | |
2238 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 2239 | { |
affee4b2 TH |
2240 | while (!list_empty(&worker->scheduled)) { |
2241 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 2242 | struct work_struct, entry); |
c34056a3 | 2243 | process_one_work(worker, work); |
1da177e4 | 2244 | } |
1da177e4 LT |
2245 | } |
2246 | ||
4690c4ab TH |
2247 | /** |
2248 | * worker_thread - the worker thread function | |
c34056a3 | 2249 | * @__worker: self |
4690c4ab | 2250 | * |
e22bee78 TH |
2251 | * The gcwq worker thread function. There's a single dynamic pool of |
2252 | * these per each cpu. These workers process all works regardless of | |
2253 | * their specific target workqueue. The only exception is works which | |
2254 | * belong to workqueues with a rescuer which will be explained in | |
2255 | * rescuer_thread(). | |
4690c4ab | 2256 | */ |
c34056a3 | 2257 | static int worker_thread(void *__worker) |
1da177e4 | 2258 | { |
c34056a3 | 2259 | struct worker *worker = __worker; |
bd7bdd43 TH |
2260 | struct worker_pool *pool = worker->pool; |
2261 | struct global_cwq *gcwq = pool->gcwq; | |
1da177e4 | 2262 | |
e22bee78 TH |
2263 | /* tell the scheduler that this is a workqueue worker */ |
2264 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 2265 | woke_up: |
c8e55f36 | 2266 | spin_lock_irq(&gcwq->lock); |
1da177e4 | 2267 | |
5f7dabfd LJ |
2268 | /* we are off idle list if destruction or rebind is requested */ |
2269 | if (unlikely(list_empty(&worker->entry))) { | |
c8e55f36 | 2270 | spin_unlock_irq(&gcwq->lock); |
25511a47 | 2271 | |
5f7dabfd | 2272 | /* if DIE is set, destruction is requested */ |
25511a47 TH |
2273 | if (worker->flags & WORKER_DIE) { |
2274 | worker->task->flags &= ~PF_WQ_WORKER; | |
2275 | return 0; | |
2276 | } | |
2277 | ||
5f7dabfd | 2278 | /* otherwise, rebind */ |
25511a47 TH |
2279 | idle_worker_rebind(worker); |
2280 | goto woke_up; | |
c8e55f36 | 2281 | } |
affee4b2 | 2282 | |
c8e55f36 | 2283 | worker_leave_idle(worker); |
db7bccf4 | 2284 | recheck: |
e22bee78 | 2285 | /* no more worker necessary? */ |
63d95a91 | 2286 | if (!need_more_worker(pool)) |
e22bee78 TH |
2287 | goto sleep; |
2288 | ||
2289 | /* do we need to manage? */ | |
63d95a91 | 2290 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
2291 | goto recheck; |
2292 | ||
c8e55f36 TH |
2293 | /* |
2294 | * ->scheduled list can only be filled while a worker is | |
2295 | * preparing to process a work or actually processing it. | |
2296 | * Make sure nobody diddled with it while I was sleeping. | |
2297 | */ | |
2298 | BUG_ON(!list_empty(&worker->scheduled)); | |
2299 | ||
e22bee78 TH |
2300 | /* |
2301 | * When control reaches this point, we're guaranteed to have | |
2302 | * at least one idle worker or that someone else has already | |
2303 | * assumed the manager role. | |
2304 | */ | |
2305 | worker_clr_flags(worker, WORKER_PREP); | |
2306 | ||
2307 | do { | |
c8e55f36 | 2308 | struct work_struct *work = |
bd7bdd43 | 2309 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2310 | struct work_struct, entry); |
2311 | ||
2312 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
2313 | /* optimization path, not strictly necessary */ | |
2314 | process_one_work(worker, work); | |
2315 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2316 | process_scheduled_works(worker); |
c8e55f36 TH |
2317 | } else { |
2318 | move_linked_works(work, &worker->scheduled, NULL); | |
2319 | process_scheduled_works(worker); | |
affee4b2 | 2320 | } |
63d95a91 | 2321 | } while (keep_working(pool)); |
e22bee78 TH |
2322 | |
2323 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 2324 | sleep: |
63d95a91 | 2325 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
e22bee78 | 2326 | goto recheck; |
d313dd85 | 2327 | |
c8e55f36 | 2328 | /* |
e22bee78 TH |
2329 | * gcwq->lock is held and there's no work to process and no |
2330 | * need to manage, sleep. Workers are woken up only while | |
2331 | * holding gcwq->lock or from local cpu, so setting the | |
2332 | * current state before releasing gcwq->lock is enough to | |
2333 | * prevent losing any event. | |
c8e55f36 TH |
2334 | */ |
2335 | worker_enter_idle(worker); | |
2336 | __set_current_state(TASK_INTERRUPTIBLE); | |
2337 | spin_unlock_irq(&gcwq->lock); | |
2338 | schedule(); | |
2339 | goto woke_up; | |
1da177e4 LT |
2340 | } |
2341 | ||
e22bee78 TH |
2342 | /** |
2343 | * rescuer_thread - the rescuer thread function | |
111c225a | 2344 | * @__rescuer: self |
e22bee78 TH |
2345 | * |
2346 | * Workqueue rescuer thread function. There's one rescuer for each | |
2347 | * workqueue which has WQ_RESCUER set. | |
2348 | * | |
2349 | * Regular work processing on a gcwq may block trying to create a new | |
2350 | * worker which uses GFP_KERNEL allocation which has slight chance of | |
2351 | * developing into deadlock if some works currently on the same queue | |
2352 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2353 | * the problem rescuer solves. | |
2354 | * | |
2355 | * When such condition is possible, the gcwq summons rescuers of all | |
2356 | * workqueues which have works queued on the gcwq and let them process | |
2357 | * those works so that forward progress can be guaranteed. | |
2358 | * | |
2359 | * This should happen rarely. | |
2360 | */ | |
111c225a | 2361 | static int rescuer_thread(void *__rescuer) |
e22bee78 | 2362 | { |
111c225a TH |
2363 | struct worker *rescuer = __rescuer; |
2364 | struct workqueue_struct *wq = rescuer->rescue_wq; | |
e22bee78 | 2365 | struct list_head *scheduled = &rescuer->scheduled; |
f3421797 | 2366 | bool is_unbound = wq->flags & WQ_UNBOUND; |
e22bee78 TH |
2367 | unsigned int cpu; |
2368 | ||
2369 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
111c225a TH |
2370 | |
2371 | /* | |
2372 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it | |
2373 | * doesn't participate in concurrency management. | |
2374 | */ | |
2375 | rescuer->task->flags |= PF_WQ_WORKER; | |
e22bee78 TH |
2376 | repeat: |
2377 | set_current_state(TASK_INTERRUPTIBLE); | |
2378 | ||
412d32e6 MG |
2379 | if (kthread_should_stop()) { |
2380 | __set_current_state(TASK_RUNNING); | |
111c225a | 2381 | rescuer->task->flags &= ~PF_WQ_WORKER; |
e22bee78 | 2382 | return 0; |
412d32e6 | 2383 | } |
e22bee78 | 2384 | |
f3421797 TH |
2385 | /* |
2386 | * See whether any cpu is asking for help. Unbounded | |
2387 | * workqueues use cpu 0 in mayday_mask for CPU_UNBOUND. | |
2388 | */ | |
f2e005aa | 2389 | for_each_mayday_cpu(cpu, wq->mayday_mask) { |
f3421797 TH |
2390 | unsigned int tcpu = is_unbound ? WORK_CPU_UNBOUND : cpu; |
2391 | struct cpu_workqueue_struct *cwq = get_cwq(tcpu, wq); | |
bd7bdd43 TH |
2392 | struct worker_pool *pool = cwq->pool; |
2393 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
2394 | struct work_struct *work, *n; |
2395 | ||
2396 | __set_current_state(TASK_RUNNING); | |
f2e005aa | 2397 | mayday_clear_cpu(cpu, wq->mayday_mask); |
e22bee78 TH |
2398 | |
2399 | /* migrate to the target cpu if possible */ | |
bd7bdd43 | 2400 | rescuer->pool = pool; |
e22bee78 TH |
2401 | worker_maybe_bind_and_lock(rescuer); |
2402 | ||
2403 | /* | |
2404 | * Slurp in all works issued via this workqueue and | |
2405 | * process'em. | |
2406 | */ | |
2407 | BUG_ON(!list_empty(&rescuer->scheduled)); | |
bd7bdd43 | 2408 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
e22bee78 TH |
2409 | if (get_work_cwq(work) == cwq) |
2410 | move_linked_works(work, scheduled, &n); | |
2411 | ||
2412 | process_scheduled_works(rescuer); | |
7576958a TH |
2413 | |
2414 | /* | |
2415 | * Leave this gcwq. If keep_working() is %true, notify a | |
2416 | * regular worker; otherwise, we end up with 0 concurrency | |
2417 | * and stalling the execution. | |
2418 | */ | |
63d95a91 TH |
2419 | if (keep_working(pool)) |
2420 | wake_up_worker(pool); | |
7576958a | 2421 | |
e22bee78 TH |
2422 | spin_unlock_irq(&gcwq->lock); |
2423 | } | |
2424 | ||
111c225a TH |
2425 | /* rescuers should never participate in concurrency management */ |
2426 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); | |
e22bee78 TH |
2427 | schedule(); |
2428 | goto repeat; | |
1da177e4 LT |
2429 | } |
2430 | ||
fc2e4d70 ON |
2431 | struct wq_barrier { |
2432 | struct work_struct work; | |
2433 | struct completion done; | |
2434 | }; | |
2435 | ||
2436 | static void wq_barrier_func(struct work_struct *work) | |
2437 | { | |
2438 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2439 | complete(&barr->done); | |
2440 | } | |
2441 | ||
4690c4ab TH |
2442 | /** |
2443 | * insert_wq_barrier - insert a barrier work | |
2444 | * @cwq: cwq to insert barrier into | |
2445 | * @barr: wq_barrier to insert | |
affee4b2 TH |
2446 | * @target: target work to attach @barr to |
2447 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2448 | * |
affee4b2 TH |
2449 | * @barr is linked to @target such that @barr is completed only after |
2450 | * @target finishes execution. Please note that the ordering | |
2451 | * guarantee is observed only with respect to @target and on the local | |
2452 | * cpu. | |
2453 | * | |
2454 | * Currently, a queued barrier can't be canceled. This is because | |
2455 | * try_to_grab_pending() can't determine whether the work to be | |
2456 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2457 | * flag of the previous work while there must be a valid next work | |
2458 | * after a work with LINKED flag set. | |
2459 | * | |
2460 | * Note that when @worker is non-NULL, @target may be modified | |
2461 | * underneath us, so we can't reliably determine cwq from @target. | |
4690c4ab TH |
2462 | * |
2463 | * CONTEXT: | |
8b03ae3c | 2464 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 2465 | */ |
83c22520 | 2466 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, |
affee4b2 TH |
2467 | struct wq_barrier *barr, |
2468 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2469 | { |
affee4b2 TH |
2470 | struct list_head *head; |
2471 | unsigned int linked = 0; | |
2472 | ||
dc186ad7 | 2473 | /* |
8b03ae3c | 2474 | * debugobject calls are safe here even with gcwq->lock locked |
dc186ad7 TG |
2475 | * as we know for sure that this will not trigger any of the |
2476 | * checks and call back into the fixup functions where we | |
2477 | * might deadlock. | |
2478 | */ | |
ca1cab37 | 2479 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2480 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2481 | init_completion(&barr->done); |
83c22520 | 2482 | |
affee4b2 TH |
2483 | /* |
2484 | * If @target is currently being executed, schedule the | |
2485 | * barrier to the worker; otherwise, put it after @target. | |
2486 | */ | |
2487 | if (worker) | |
2488 | head = worker->scheduled.next; | |
2489 | else { | |
2490 | unsigned long *bits = work_data_bits(target); | |
2491 | ||
2492 | head = target->entry.next; | |
2493 | /* there can already be other linked works, inherit and set */ | |
2494 | linked = *bits & WORK_STRUCT_LINKED; | |
2495 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2496 | } | |
2497 | ||
dc186ad7 | 2498 | debug_work_activate(&barr->work); |
affee4b2 TH |
2499 | insert_work(cwq, &barr->work, head, |
2500 | work_color_to_flags(WORK_NO_COLOR) | linked); | |
fc2e4d70 ON |
2501 | } |
2502 | ||
73f53c4a TH |
2503 | /** |
2504 | * flush_workqueue_prep_cwqs - prepare cwqs for workqueue flushing | |
2505 | * @wq: workqueue being flushed | |
2506 | * @flush_color: new flush color, < 0 for no-op | |
2507 | * @work_color: new work color, < 0 for no-op | |
2508 | * | |
2509 | * Prepare cwqs for workqueue flushing. | |
2510 | * | |
2511 | * If @flush_color is non-negative, flush_color on all cwqs should be | |
2512 | * -1. If no cwq has in-flight commands at the specified color, all | |
2513 | * cwq->flush_color's stay at -1 and %false is returned. If any cwq | |
2514 | * has in flight commands, its cwq->flush_color is set to | |
2515 | * @flush_color, @wq->nr_cwqs_to_flush is updated accordingly, cwq | |
2516 | * wakeup logic is armed and %true is returned. | |
2517 | * | |
2518 | * The caller should have initialized @wq->first_flusher prior to | |
2519 | * calling this function with non-negative @flush_color. If | |
2520 | * @flush_color is negative, no flush color update is done and %false | |
2521 | * is returned. | |
2522 | * | |
2523 | * If @work_color is non-negative, all cwqs should have the same | |
2524 | * work_color which is previous to @work_color and all will be | |
2525 | * advanced to @work_color. | |
2526 | * | |
2527 | * CONTEXT: | |
2528 | * mutex_lock(wq->flush_mutex). | |
2529 | * | |
2530 | * RETURNS: | |
2531 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2532 | * otherwise. | |
2533 | */ | |
2534 | static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq, | |
2535 | int flush_color, int work_color) | |
1da177e4 | 2536 | { |
73f53c4a TH |
2537 | bool wait = false; |
2538 | unsigned int cpu; | |
1da177e4 | 2539 | |
73f53c4a TH |
2540 | if (flush_color >= 0) { |
2541 | BUG_ON(atomic_read(&wq->nr_cwqs_to_flush)); | |
2542 | atomic_set(&wq->nr_cwqs_to_flush, 1); | |
1da177e4 | 2543 | } |
2355b70f | 2544 | |
f3421797 | 2545 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a | 2546 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
bd7bdd43 | 2547 | struct global_cwq *gcwq = cwq->pool->gcwq; |
fc2e4d70 | 2548 | |
8b03ae3c | 2549 | spin_lock_irq(&gcwq->lock); |
83c22520 | 2550 | |
73f53c4a TH |
2551 | if (flush_color >= 0) { |
2552 | BUG_ON(cwq->flush_color != -1); | |
fc2e4d70 | 2553 | |
73f53c4a TH |
2554 | if (cwq->nr_in_flight[flush_color]) { |
2555 | cwq->flush_color = flush_color; | |
2556 | atomic_inc(&wq->nr_cwqs_to_flush); | |
2557 | wait = true; | |
2558 | } | |
2559 | } | |
1da177e4 | 2560 | |
73f53c4a TH |
2561 | if (work_color >= 0) { |
2562 | BUG_ON(work_color != work_next_color(cwq->work_color)); | |
2563 | cwq->work_color = work_color; | |
2564 | } | |
1da177e4 | 2565 | |
8b03ae3c | 2566 | spin_unlock_irq(&gcwq->lock); |
1da177e4 | 2567 | } |
2355b70f | 2568 | |
73f53c4a TH |
2569 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_cwqs_to_flush)) |
2570 | complete(&wq->first_flusher->done); | |
14441960 | 2571 | |
73f53c4a | 2572 | return wait; |
1da177e4 LT |
2573 | } |
2574 | ||
0fcb78c2 | 2575 | /** |
1da177e4 | 2576 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2577 | * @wq: workqueue to flush |
1da177e4 LT |
2578 | * |
2579 | * Forces execution of the workqueue and blocks until its completion. | |
2580 | * This is typically used in driver shutdown handlers. | |
2581 | * | |
fc2e4d70 ON |
2582 | * We sleep until all works which were queued on entry have been handled, |
2583 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 2584 | */ |
7ad5b3a5 | 2585 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2586 | { |
73f53c4a TH |
2587 | struct wq_flusher this_flusher = { |
2588 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2589 | .flush_color = -1, | |
2590 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2591 | }; | |
2592 | int next_color; | |
1da177e4 | 2593 | |
3295f0ef IM |
2594 | lock_map_acquire(&wq->lockdep_map); |
2595 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
2596 | |
2597 | mutex_lock(&wq->flush_mutex); | |
2598 | ||
2599 | /* | |
2600 | * Start-to-wait phase | |
2601 | */ | |
2602 | next_color = work_next_color(wq->work_color); | |
2603 | ||
2604 | if (next_color != wq->flush_color) { | |
2605 | /* | |
2606 | * Color space is not full. The current work_color | |
2607 | * becomes our flush_color and work_color is advanced | |
2608 | * by one. | |
2609 | */ | |
2610 | BUG_ON(!list_empty(&wq->flusher_overflow)); | |
2611 | this_flusher.flush_color = wq->work_color; | |
2612 | wq->work_color = next_color; | |
2613 | ||
2614 | if (!wq->first_flusher) { | |
2615 | /* no flush in progress, become the first flusher */ | |
2616 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2617 | ||
2618 | wq->first_flusher = &this_flusher; | |
2619 | ||
2620 | if (!flush_workqueue_prep_cwqs(wq, wq->flush_color, | |
2621 | wq->work_color)) { | |
2622 | /* nothing to flush, done */ | |
2623 | wq->flush_color = next_color; | |
2624 | wq->first_flusher = NULL; | |
2625 | goto out_unlock; | |
2626 | } | |
2627 | } else { | |
2628 | /* wait in queue */ | |
2629 | BUG_ON(wq->flush_color == this_flusher.flush_color); | |
2630 | list_add_tail(&this_flusher.list, &wq->flusher_queue); | |
2631 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2632 | } | |
2633 | } else { | |
2634 | /* | |
2635 | * Oops, color space is full, wait on overflow queue. | |
2636 | * The next flush completion will assign us | |
2637 | * flush_color and transfer to flusher_queue. | |
2638 | */ | |
2639 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2640 | } | |
2641 | ||
2642 | mutex_unlock(&wq->flush_mutex); | |
2643 | ||
2644 | wait_for_completion(&this_flusher.done); | |
2645 | ||
2646 | /* | |
2647 | * Wake-up-and-cascade phase | |
2648 | * | |
2649 | * First flushers are responsible for cascading flushes and | |
2650 | * handling overflow. Non-first flushers can simply return. | |
2651 | */ | |
2652 | if (wq->first_flusher != &this_flusher) | |
2653 | return; | |
2654 | ||
2655 | mutex_lock(&wq->flush_mutex); | |
2656 | ||
4ce48b37 TH |
2657 | /* we might have raced, check again with mutex held */ |
2658 | if (wq->first_flusher != &this_flusher) | |
2659 | goto out_unlock; | |
2660 | ||
73f53c4a TH |
2661 | wq->first_flusher = NULL; |
2662 | ||
2663 | BUG_ON(!list_empty(&this_flusher.list)); | |
2664 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2665 | ||
2666 | while (true) { | |
2667 | struct wq_flusher *next, *tmp; | |
2668 | ||
2669 | /* complete all the flushers sharing the current flush color */ | |
2670 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2671 | if (next->flush_color != wq->flush_color) | |
2672 | break; | |
2673 | list_del_init(&next->list); | |
2674 | complete(&next->done); | |
2675 | } | |
2676 | ||
2677 | BUG_ON(!list_empty(&wq->flusher_overflow) && | |
2678 | wq->flush_color != work_next_color(wq->work_color)); | |
2679 | ||
2680 | /* this flush_color is finished, advance by one */ | |
2681 | wq->flush_color = work_next_color(wq->flush_color); | |
2682 | ||
2683 | /* one color has been freed, handle overflow queue */ | |
2684 | if (!list_empty(&wq->flusher_overflow)) { | |
2685 | /* | |
2686 | * Assign the same color to all overflowed | |
2687 | * flushers, advance work_color and append to | |
2688 | * flusher_queue. This is the start-to-wait | |
2689 | * phase for these overflowed flushers. | |
2690 | */ | |
2691 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2692 | tmp->flush_color = wq->work_color; | |
2693 | ||
2694 | wq->work_color = work_next_color(wq->work_color); | |
2695 | ||
2696 | list_splice_tail_init(&wq->flusher_overflow, | |
2697 | &wq->flusher_queue); | |
2698 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2699 | } | |
2700 | ||
2701 | if (list_empty(&wq->flusher_queue)) { | |
2702 | BUG_ON(wq->flush_color != wq->work_color); | |
2703 | break; | |
2704 | } | |
2705 | ||
2706 | /* | |
2707 | * Need to flush more colors. Make the next flusher | |
2708 | * the new first flusher and arm cwqs. | |
2709 | */ | |
2710 | BUG_ON(wq->flush_color == wq->work_color); | |
2711 | BUG_ON(wq->flush_color != next->flush_color); | |
2712 | ||
2713 | list_del_init(&next->list); | |
2714 | wq->first_flusher = next; | |
2715 | ||
2716 | if (flush_workqueue_prep_cwqs(wq, wq->flush_color, -1)) | |
2717 | break; | |
2718 | ||
2719 | /* | |
2720 | * Meh... this color is already done, clear first | |
2721 | * flusher and repeat cascading. | |
2722 | */ | |
2723 | wq->first_flusher = NULL; | |
2724 | } | |
2725 | ||
2726 | out_unlock: | |
2727 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 2728 | } |
ae90dd5d | 2729 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2730 | |
9c5a2ba7 TH |
2731 | /** |
2732 | * drain_workqueue - drain a workqueue | |
2733 | * @wq: workqueue to drain | |
2734 | * | |
2735 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2736 | * only chain queueing is allowed. IOW, only currently pending or running | |
2737 | * work items on @wq can queue further work items on it. @wq is flushed | |
2738 | * repeatedly until it becomes empty. The number of flushing is detemined | |
2739 | * by the depth of chaining and should be relatively short. Whine if it | |
2740 | * takes too long. | |
2741 | */ | |
2742 | void drain_workqueue(struct workqueue_struct *wq) | |
2743 | { | |
2744 | unsigned int flush_cnt = 0; | |
2745 | unsigned int cpu; | |
2746 | ||
2747 | /* | |
2748 | * __queue_work() needs to test whether there are drainers, is much | |
2749 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
2750 | * Use WQ_DRAINING so that queue doesn't have to check nr_drainers. | |
2751 | */ | |
2752 | spin_lock(&workqueue_lock); | |
2753 | if (!wq->nr_drainers++) | |
2754 | wq->flags |= WQ_DRAINING; | |
2755 | spin_unlock(&workqueue_lock); | |
2756 | reflush: | |
2757 | flush_workqueue(wq); | |
2758 | ||
2759 | for_each_cwq_cpu(cpu, wq) { | |
2760 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
fa2563e4 | 2761 | bool drained; |
9c5a2ba7 | 2762 | |
bd7bdd43 | 2763 | spin_lock_irq(&cwq->pool->gcwq->lock); |
fa2563e4 | 2764 | drained = !cwq->nr_active && list_empty(&cwq->delayed_works); |
bd7bdd43 | 2765 | spin_unlock_irq(&cwq->pool->gcwq->lock); |
fa2563e4 TT |
2766 | |
2767 | if (drained) | |
9c5a2ba7 TH |
2768 | continue; |
2769 | ||
2770 | if (++flush_cnt == 10 || | |
2771 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
044c782c VI |
2772 | pr_warn("workqueue %s: flush on destruction isn't complete after %u tries\n", |
2773 | wq->name, flush_cnt); | |
9c5a2ba7 TH |
2774 | goto reflush; |
2775 | } | |
2776 | ||
2777 | spin_lock(&workqueue_lock); | |
2778 | if (!--wq->nr_drainers) | |
2779 | wq->flags &= ~WQ_DRAINING; | |
2780 | spin_unlock(&workqueue_lock); | |
2781 | } | |
2782 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2783 | ||
606a5020 | 2784 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) |
db700897 | 2785 | { |
affee4b2 | 2786 | struct worker *worker = NULL; |
8b03ae3c | 2787 | struct global_cwq *gcwq; |
db700897 | 2788 | struct cpu_workqueue_struct *cwq; |
db700897 ON |
2789 | |
2790 | might_sleep(); | |
7a22ad75 TH |
2791 | gcwq = get_work_gcwq(work); |
2792 | if (!gcwq) | |
baf59022 | 2793 | return false; |
db700897 | 2794 | |
8b03ae3c | 2795 | spin_lock_irq(&gcwq->lock); |
db700897 ON |
2796 | if (!list_empty(&work->entry)) { |
2797 | /* | |
2798 | * See the comment near try_to_grab_pending()->smp_rmb(). | |
7a22ad75 TH |
2799 | * If it was re-queued to a different gcwq under us, we |
2800 | * are not going to wait. | |
db700897 ON |
2801 | */ |
2802 | smp_rmb(); | |
7a22ad75 | 2803 | cwq = get_work_cwq(work); |
bd7bdd43 | 2804 | if (unlikely(!cwq || gcwq != cwq->pool->gcwq)) |
4690c4ab | 2805 | goto already_gone; |
606a5020 | 2806 | } else { |
7a22ad75 | 2807 | worker = find_worker_executing_work(gcwq, work); |
affee4b2 | 2808 | if (!worker) |
4690c4ab | 2809 | goto already_gone; |
7a22ad75 | 2810 | cwq = worker->current_cwq; |
606a5020 | 2811 | } |
db700897 | 2812 | |
baf59022 | 2813 | insert_wq_barrier(cwq, barr, work, worker); |
8b03ae3c | 2814 | spin_unlock_irq(&gcwq->lock); |
7a22ad75 | 2815 | |
e159489b TH |
2816 | /* |
2817 | * If @max_active is 1 or rescuer is in use, flushing another work | |
2818 | * item on the same workqueue may lead to deadlock. Make sure the | |
2819 | * flusher is not running on the same workqueue by verifying write | |
2820 | * access. | |
2821 | */ | |
2822 | if (cwq->wq->saved_max_active == 1 || cwq->wq->flags & WQ_RESCUER) | |
2823 | lock_map_acquire(&cwq->wq->lockdep_map); | |
2824 | else | |
2825 | lock_map_acquire_read(&cwq->wq->lockdep_map); | |
7a22ad75 | 2826 | lock_map_release(&cwq->wq->lockdep_map); |
e159489b | 2827 | |
401a8d04 | 2828 | return true; |
4690c4ab | 2829 | already_gone: |
8b03ae3c | 2830 | spin_unlock_irq(&gcwq->lock); |
401a8d04 | 2831 | return false; |
db700897 | 2832 | } |
baf59022 TH |
2833 | |
2834 | /** | |
2835 | * flush_work - wait for a work to finish executing the last queueing instance | |
2836 | * @work: the work to flush | |
2837 | * | |
606a5020 TH |
2838 | * Wait until @work has finished execution. @work is guaranteed to be idle |
2839 | * on return if it hasn't been requeued since flush started. | |
baf59022 TH |
2840 | * |
2841 | * RETURNS: | |
2842 | * %true if flush_work() waited for the work to finish execution, | |
2843 | * %false if it was already idle. | |
2844 | */ | |
2845 | bool flush_work(struct work_struct *work) | |
2846 | { | |
2847 | struct wq_barrier barr; | |
2848 | ||
0976dfc1 SB |
2849 | lock_map_acquire(&work->lockdep_map); |
2850 | lock_map_release(&work->lockdep_map); | |
2851 | ||
606a5020 | 2852 | if (start_flush_work(work, &barr)) { |
401a8d04 TH |
2853 | wait_for_completion(&barr.done); |
2854 | destroy_work_on_stack(&barr.work); | |
2855 | return true; | |
606a5020 | 2856 | } else { |
401a8d04 | 2857 | return false; |
6e84d644 | 2858 | } |
6e84d644 | 2859 | } |
606a5020 | 2860 | EXPORT_SYMBOL_GPL(flush_work); |
6e84d644 | 2861 | |
36e227d2 | 2862 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
1f1f642e | 2863 | { |
bbb68dfa | 2864 | unsigned long flags; |
1f1f642e ON |
2865 | int ret; |
2866 | ||
2867 | do { | |
bbb68dfa TH |
2868 | ret = try_to_grab_pending(work, is_dwork, &flags); |
2869 | /* | |
2870 | * If someone else is canceling, wait for the same event it | |
2871 | * would be waiting for before retrying. | |
2872 | */ | |
2873 | if (unlikely(ret == -ENOENT)) | |
606a5020 | 2874 | flush_work(work); |
1f1f642e ON |
2875 | } while (unlikely(ret < 0)); |
2876 | ||
bbb68dfa TH |
2877 | /* tell other tasks trying to grab @work to back off */ |
2878 | mark_work_canceling(work); | |
2879 | local_irq_restore(flags); | |
2880 | ||
606a5020 | 2881 | flush_work(work); |
7a22ad75 | 2882 | clear_work_data(work); |
1f1f642e ON |
2883 | return ret; |
2884 | } | |
2885 | ||
6e84d644 | 2886 | /** |
401a8d04 TH |
2887 | * cancel_work_sync - cancel a work and wait for it to finish |
2888 | * @work: the work to cancel | |
6e84d644 | 2889 | * |
401a8d04 TH |
2890 | * Cancel @work and wait for its execution to finish. This function |
2891 | * can be used even if the work re-queues itself or migrates to | |
2892 | * another workqueue. On return from this function, @work is | |
2893 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2894 | * |
401a8d04 TH |
2895 | * cancel_work_sync(&delayed_work->work) must not be used for |
2896 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2897 | * |
401a8d04 | 2898 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2899 | * queued can't be destroyed before this function returns. |
401a8d04 TH |
2900 | * |
2901 | * RETURNS: | |
2902 | * %true if @work was pending, %false otherwise. | |
6e84d644 | 2903 | */ |
401a8d04 | 2904 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2905 | { |
36e227d2 | 2906 | return __cancel_work_timer(work, false); |
b89deed3 | 2907 | } |
28e53bdd | 2908 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2909 | |
6e84d644 | 2910 | /** |
401a8d04 TH |
2911 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2912 | * @dwork: the delayed work to flush | |
6e84d644 | 2913 | * |
401a8d04 TH |
2914 | * Delayed timer is cancelled and the pending work is queued for |
2915 | * immediate execution. Like flush_work(), this function only | |
2916 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2917 | * |
401a8d04 TH |
2918 | * RETURNS: |
2919 | * %true if flush_work() waited for the work to finish execution, | |
2920 | * %false if it was already idle. | |
6e84d644 | 2921 | */ |
401a8d04 TH |
2922 | bool flush_delayed_work(struct delayed_work *dwork) |
2923 | { | |
8930caba | 2924 | local_irq_disable(); |
401a8d04 | 2925 | if (del_timer_sync(&dwork->timer)) |
1265057f | 2926 | __queue_work(dwork->cpu, |
401a8d04 | 2927 | get_work_cwq(&dwork->work)->wq, &dwork->work); |
8930caba | 2928 | local_irq_enable(); |
401a8d04 TH |
2929 | return flush_work(&dwork->work); |
2930 | } | |
2931 | EXPORT_SYMBOL(flush_delayed_work); | |
2932 | ||
09383498 | 2933 | /** |
57b30ae7 TH |
2934 | * cancel_delayed_work - cancel a delayed work |
2935 | * @dwork: delayed_work to cancel | |
09383498 | 2936 | * |
57b30ae7 TH |
2937 | * Kill off a pending delayed_work. Returns %true if @dwork was pending |
2938 | * and canceled; %false if wasn't pending. Note that the work callback | |
2939 | * function may still be running on return, unless it returns %true and the | |
2940 | * work doesn't re-arm itself. Explicitly flush or use | |
2941 | * cancel_delayed_work_sync() to wait on it. | |
09383498 | 2942 | * |
57b30ae7 | 2943 | * This function is safe to call from any context including IRQ handler. |
09383498 | 2944 | */ |
57b30ae7 | 2945 | bool cancel_delayed_work(struct delayed_work *dwork) |
09383498 | 2946 | { |
57b30ae7 TH |
2947 | unsigned long flags; |
2948 | int ret; | |
2949 | ||
2950 | do { | |
2951 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
2952 | } while (unlikely(ret == -EAGAIN)); | |
2953 | ||
2954 | if (unlikely(ret < 0)) | |
2955 | return false; | |
2956 | ||
2957 | set_work_cpu_and_clear_pending(&dwork->work, work_cpu(&dwork->work)); | |
2958 | local_irq_restore(flags); | |
c0158ca6 | 2959 | return ret; |
09383498 | 2960 | } |
57b30ae7 | 2961 | EXPORT_SYMBOL(cancel_delayed_work); |
09383498 | 2962 | |
401a8d04 TH |
2963 | /** |
2964 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
2965 | * @dwork: the delayed work cancel | |
2966 | * | |
2967 | * This is cancel_work_sync() for delayed works. | |
2968 | * | |
2969 | * RETURNS: | |
2970 | * %true if @dwork was pending, %false otherwise. | |
2971 | */ | |
2972 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 2973 | { |
36e227d2 | 2974 | return __cancel_work_timer(&dwork->work, true); |
6e84d644 | 2975 | } |
f5a421a4 | 2976 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 2977 | |
0fcb78c2 | 2978 | /** |
c1a220e7 ZR |
2979 | * schedule_work_on - put work task on a specific cpu |
2980 | * @cpu: cpu to put the work task on | |
2981 | * @work: job to be done | |
2982 | * | |
2983 | * This puts a job on a specific cpu | |
2984 | */ | |
d4283e93 | 2985 | bool schedule_work_on(int cpu, struct work_struct *work) |
c1a220e7 | 2986 | { |
d320c038 | 2987 | return queue_work_on(cpu, system_wq, work); |
c1a220e7 ZR |
2988 | } |
2989 | EXPORT_SYMBOL(schedule_work_on); | |
2990 | ||
0fcb78c2 | 2991 | /** |
0fcb78c2 REB |
2992 | * schedule_work - put work task in global workqueue |
2993 | * @work: job to be done | |
0fcb78c2 | 2994 | * |
d4283e93 TH |
2995 | * Returns %false if @work was already on the kernel-global workqueue and |
2996 | * %true otherwise. | |
5b0f437d BVA |
2997 | * |
2998 | * This puts a job in the kernel-global workqueue if it was not already | |
2999 | * queued and leaves it in the same position on the kernel-global | |
3000 | * workqueue otherwise. | |
0fcb78c2 | 3001 | */ |
d4283e93 | 3002 | bool schedule_work(struct work_struct *work) |
1da177e4 | 3003 | { |
d320c038 | 3004 | return queue_work(system_wq, work); |
1da177e4 | 3005 | } |
ae90dd5d | 3006 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 3007 | |
0fcb78c2 REB |
3008 | /** |
3009 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
3010 | * @cpu: cpu to use | |
52bad64d | 3011 | * @dwork: job to be done |
0fcb78c2 REB |
3012 | * @delay: number of jiffies to wait |
3013 | * | |
3014 | * After waiting for a given time this puts a job in the kernel-global | |
3015 | * workqueue on the specified CPU. | |
3016 | */ | |
d4283e93 TH |
3017 | bool schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
3018 | unsigned long delay) | |
1da177e4 | 3019 | { |
d320c038 | 3020 | return queue_delayed_work_on(cpu, system_wq, dwork, delay); |
1da177e4 | 3021 | } |
ae90dd5d | 3022 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 3023 | |
0fcb78c2 REB |
3024 | /** |
3025 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
3026 | * @dwork: job to be done |
3027 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
3028 | * |
3029 | * After waiting for a given time this puts a job in the kernel-global | |
3030 | * workqueue. | |
3031 | */ | |
d4283e93 | 3032 | bool schedule_delayed_work(struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 3033 | { |
d320c038 | 3034 | return queue_delayed_work(system_wq, dwork, delay); |
1da177e4 | 3035 | } |
ae90dd5d | 3036 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 3037 | |
b6136773 | 3038 | /** |
31ddd871 | 3039 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 3040 | * @func: the function to call |
b6136773 | 3041 | * |
31ddd871 TH |
3042 | * schedule_on_each_cpu() executes @func on each online CPU using the |
3043 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 3044 | * schedule_on_each_cpu() is very slow. |
31ddd871 TH |
3045 | * |
3046 | * RETURNS: | |
3047 | * 0 on success, -errno on failure. | |
b6136773 | 3048 | */ |
65f27f38 | 3049 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
3050 | { |
3051 | int cpu; | |
38f51568 | 3052 | struct work_struct __percpu *works; |
15316ba8 | 3053 | |
b6136773 AM |
3054 | works = alloc_percpu(struct work_struct); |
3055 | if (!works) | |
15316ba8 | 3056 | return -ENOMEM; |
b6136773 | 3057 | |
93981800 TH |
3058 | get_online_cpus(); |
3059 | ||
15316ba8 | 3060 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
3061 | struct work_struct *work = per_cpu_ptr(works, cpu); |
3062 | ||
3063 | INIT_WORK(work, func); | |
b71ab8c2 | 3064 | schedule_work_on(cpu, work); |
65a64464 | 3065 | } |
93981800 TH |
3066 | |
3067 | for_each_online_cpu(cpu) | |
3068 | flush_work(per_cpu_ptr(works, cpu)); | |
3069 | ||
95402b38 | 3070 | put_online_cpus(); |
b6136773 | 3071 | free_percpu(works); |
15316ba8 CL |
3072 | return 0; |
3073 | } | |
3074 | ||
eef6a7d5 AS |
3075 | /** |
3076 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
3077 | * | |
3078 | * Forces execution of the kernel-global workqueue and blocks until its | |
3079 | * completion. | |
3080 | * | |
3081 | * Think twice before calling this function! It's very easy to get into | |
3082 | * trouble if you don't take great care. Either of the following situations | |
3083 | * will lead to deadlock: | |
3084 | * | |
3085 | * One of the work items currently on the workqueue needs to acquire | |
3086 | * a lock held by your code or its caller. | |
3087 | * | |
3088 | * Your code is running in the context of a work routine. | |
3089 | * | |
3090 | * They will be detected by lockdep when they occur, but the first might not | |
3091 | * occur very often. It depends on what work items are on the workqueue and | |
3092 | * what locks they need, which you have no control over. | |
3093 | * | |
3094 | * In most situations flushing the entire workqueue is overkill; you merely | |
3095 | * need to know that a particular work item isn't queued and isn't running. | |
3096 | * In such cases you should use cancel_delayed_work_sync() or | |
3097 | * cancel_work_sync() instead. | |
3098 | */ | |
1da177e4 LT |
3099 | void flush_scheduled_work(void) |
3100 | { | |
d320c038 | 3101 | flush_workqueue(system_wq); |
1da177e4 | 3102 | } |
ae90dd5d | 3103 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 3104 | |
1fa44eca JB |
3105 | /** |
3106 | * execute_in_process_context - reliably execute the routine with user context | |
3107 | * @fn: the function to execute | |
1fa44eca JB |
3108 | * @ew: guaranteed storage for the execute work structure (must |
3109 | * be available when the work executes) | |
3110 | * | |
3111 | * Executes the function immediately if process context is available, | |
3112 | * otherwise schedules the function for delayed execution. | |
3113 | * | |
3114 | * Returns: 0 - function was executed | |
3115 | * 1 - function was scheduled for execution | |
3116 | */ | |
65f27f38 | 3117 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
3118 | { |
3119 | if (!in_interrupt()) { | |
65f27f38 | 3120 | fn(&ew->work); |
1fa44eca JB |
3121 | return 0; |
3122 | } | |
3123 | ||
65f27f38 | 3124 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
3125 | schedule_work(&ew->work); |
3126 | ||
3127 | return 1; | |
3128 | } | |
3129 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
3130 | ||
1da177e4 LT |
3131 | int keventd_up(void) |
3132 | { | |
d320c038 | 3133 | return system_wq != NULL; |
1da177e4 LT |
3134 | } |
3135 | ||
bdbc5dd7 | 3136 | static int alloc_cwqs(struct workqueue_struct *wq) |
0f900049 | 3137 | { |
65a64464 | 3138 | /* |
0f900049 TH |
3139 | * cwqs are forced aligned according to WORK_STRUCT_FLAG_BITS. |
3140 | * Make sure that the alignment isn't lower than that of | |
3141 | * unsigned long long. | |
65a64464 | 3142 | */ |
0f900049 TH |
3143 | const size_t size = sizeof(struct cpu_workqueue_struct); |
3144 | const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS, | |
3145 | __alignof__(unsigned long long)); | |
65a64464 | 3146 | |
e06ffa1e | 3147 | if (!(wq->flags & WQ_UNBOUND)) |
f3421797 | 3148 | wq->cpu_wq.pcpu = __alloc_percpu(size, align); |
931ac77e | 3149 | else { |
f3421797 TH |
3150 | void *ptr; |
3151 | ||
3152 | /* | |
3153 | * Allocate enough room to align cwq and put an extra | |
3154 | * pointer at the end pointing back to the originally | |
3155 | * allocated pointer which will be used for free. | |
3156 | */ | |
3157 | ptr = kzalloc(size + align + sizeof(void *), GFP_KERNEL); | |
3158 | if (ptr) { | |
3159 | wq->cpu_wq.single = PTR_ALIGN(ptr, align); | |
3160 | *(void **)(wq->cpu_wq.single + 1) = ptr; | |
3161 | } | |
bdbc5dd7 | 3162 | } |
f3421797 | 3163 | |
0415b00d | 3164 | /* just in case, make sure it's actually aligned */ |
bdbc5dd7 TH |
3165 | BUG_ON(!IS_ALIGNED(wq->cpu_wq.v, align)); |
3166 | return wq->cpu_wq.v ? 0 : -ENOMEM; | |
0f900049 TH |
3167 | } |
3168 | ||
bdbc5dd7 | 3169 | static void free_cwqs(struct workqueue_struct *wq) |
0f900049 | 3170 | { |
e06ffa1e | 3171 | if (!(wq->flags & WQ_UNBOUND)) |
f3421797 TH |
3172 | free_percpu(wq->cpu_wq.pcpu); |
3173 | else if (wq->cpu_wq.single) { | |
3174 | /* the pointer to free is stored right after the cwq */ | |
bdbc5dd7 | 3175 | kfree(*(void **)(wq->cpu_wq.single + 1)); |
f3421797 | 3176 | } |
0f900049 TH |
3177 | } |
3178 | ||
f3421797 TH |
3179 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
3180 | const char *name) | |
b71ab8c2 | 3181 | { |
f3421797 TH |
3182 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
3183 | ||
3184 | if (max_active < 1 || max_active > lim) | |
044c782c VI |
3185 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
3186 | max_active, name, 1, lim); | |
b71ab8c2 | 3187 | |
f3421797 | 3188 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
3189 | } |
3190 | ||
b196be89 | 3191 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
d320c038 TH |
3192 | unsigned int flags, |
3193 | int max_active, | |
3194 | struct lock_class_key *key, | |
b196be89 | 3195 | const char *lock_name, ...) |
1da177e4 | 3196 | { |
b196be89 | 3197 | va_list args, args1; |
1da177e4 | 3198 | struct workqueue_struct *wq; |
c34056a3 | 3199 | unsigned int cpu; |
b196be89 TH |
3200 | size_t namelen; |
3201 | ||
3202 | /* determine namelen, allocate wq and format name */ | |
3203 | va_start(args, lock_name); | |
3204 | va_copy(args1, args); | |
3205 | namelen = vsnprintf(NULL, 0, fmt, args) + 1; | |
3206 | ||
3207 | wq = kzalloc(sizeof(*wq) + namelen, GFP_KERNEL); | |
3208 | if (!wq) | |
3209 | goto err; | |
3210 | ||
3211 | vsnprintf(wq->name, namelen, fmt, args1); | |
3212 | va_end(args); | |
3213 | va_end(args1); | |
1da177e4 | 3214 | |
6370a6ad TH |
3215 | /* |
3216 | * Workqueues which may be used during memory reclaim should | |
3217 | * have a rescuer to guarantee forward progress. | |
3218 | */ | |
3219 | if (flags & WQ_MEM_RECLAIM) | |
3220 | flags |= WQ_RESCUER; | |
3221 | ||
d320c038 | 3222 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 3223 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 3224 | |
b196be89 | 3225 | /* init wq */ |
97e37d7b | 3226 | wq->flags = flags; |
a0a1a5fd | 3227 | wq->saved_max_active = max_active; |
73f53c4a TH |
3228 | mutex_init(&wq->flush_mutex); |
3229 | atomic_set(&wq->nr_cwqs_to_flush, 0); | |
3230 | INIT_LIST_HEAD(&wq->flusher_queue); | |
3231 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
502ca9d8 | 3232 | |
eb13ba87 | 3233 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 3234 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 3235 | |
bdbc5dd7 TH |
3236 | if (alloc_cwqs(wq) < 0) |
3237 | goto err; | |
3238 | ||
f3421797 | 3239 | for_each_cwq_cpu(cpu, wq) { |
1537663f | 3240 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
8b03ae3c | 3241 | struct global_cwq *gcwq = get_gcwq(cpu); |
3270476a | 3242 | int pool_idx = (bool)(flags & WQ_HIGHPRI); |
1537663f | 3243 | |
0f900049 | 3244 | BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK); |
3270476a | 3245 | cwq->pool = &gcwq->pools[pool_idx]; |
c34056a3 | 3246 | cwq->wq = wq; |
73f53c4a | 3247 | cwq->flush_color = -1; |
1e19ffc6 | 3248 | cwq->max_active = max_active; |
1e19ffc6 | 3249 | INIT_LIST_HEAD(&cwq->delayed_works); |
e22bee78 | 3250 | } |
1537663f | 3251 | |
e22bee78 TH |
3252 | if (flags & WQ_RESCUER) { |
3253 | struct worker *rescuer; | |
3254 | ||
f2e005aa | 3255 | if (!alloc_mayday_mask(&wq->mayday_mask, GFP_KERNEL)) |
e22bee78 TH |
3256 | goto err; |
3257 | ||
3258 | wq->rescuer = rescuer = alloc_worker(); | |
3259 | if (!rescuer) | |
3260 | goto err; | |
3261 | ||
111c225a TH |
3262 | rescuer->rescue_wq = wq; |
3263 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", | |
b196be89 | 3264 | wq->name); |
e22bee78 TH |
3265 | if (IS_ERR(rescuer->task)) |
3266 | goto err; | |
3267 | ||
e22bee78 TH |
3268 | rescuer->task->flags |= PF_THREAD_BOUND; |
3269 | wake_up_process(rescuer->task); | |
3af24433 ON |
3270 | } |
3271 | ||
a0a1a5fd TH |
3272 | /* |
3273 | * workqueue_lock protects global freeze state and workqueues | |
3274 | * list. Grab it, set max_active accordingly and add the new | |
3275 | * workqueue to workqueues list. | |
3276 | */ | |
1537663f | 3277 | spin_lock(&workqueue_lock); |
a0a1a5fd | 3278 | |
58a69cb4 | 3279 | if (workqueue_freezing && wq->flags & WQ_FREEZABLE) |
f3421797 | 3280 | for_each_cwq_cpu(cpu, wq) |
a0a1a5fd TH |
3281 | get_cwq(cpu, wq)->max_active = 0; |
3282 | ||
1537663f | 3283 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 3284 | |
1537663f TH |
3285 | spin_unlock(&workqueue_lock); |
3286 | ||
3af24433 | 3287 | return wq; |
4690c4ab TH |
3288 | err: |
3289 | if (wq) { | |
bdbc5dd7 | 3290 | free_cwqs(wq); |
f2e005aa | 3291 | free_mayday_mask(wq->mayday_mask); |
e22bee78 | 3292 | kfree(wq->rescuer); |
4690c4ab TH |
3293 | kfree(wq); |
3294 | } | |
3295 | return NULL; | |
3af24433 | 3296 | } |
d320c038 | 3297 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 3298 | |
3af24433 ON |
3299 | /** |
3300 | * destroy_workqueue - safely terminate a workqueue | |
3301 | * @wq: target workqueue | |
3302 | * | |
3303 | * Safely destroy a workqueue. All work currently pending will be done first. | |
3304 | */ | |
3305 | void destroy_workqueue(struct workqueue_struct *wq) | |
3306 | { | |
c8e55f36 | 3307 | unsigned int cpu; |
3af24433 | 3308 | |
9c5a2ba7 TH |
3309 | /* drain it before proceeding with destruction */ |
3310 | drain_workqueue(wq); | |
c8efcc25 | 3311 | |
a0a1a5fd TH |
3312 | /* |
3313 | * wq list is used to freeze wq, remove from list after | |
3314 | * flushing is complete in case freeze races us. | |
3315 | */ | |
95402b38 | 3316 | spin_lock(&workqueue_lock); |
b1f4ec17 | 3317 | list_del(&wq->list); |
95402b38 | 3318 | spin_unlock(&workqueue_lock); |
3af24433 | 3319 | |
e22bee78 | 3320 | /* sanity check */ |
f3421797 | 3321 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a TH |
3322 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3323 | int i; | |
3324 | ||
73f53c4a TH |
3325 | for (i = 0; i < WORK_NR_COLORS; i++) |
3326 | BUG_ON(cwq->nr_in_flight[i]); | |
1e19ffc6 TH |
3327 | BUG_ON(cwq->nr_active); |
3328 | BUG_ON(!list_empty(&cwq->delayed_works)); | |
73f53c4a | 3329 | } |
9b41ea72 | 3330 | |
e22bee78 TH |
3331 | if (wq->flags & WQ_RESCUER) { |
3332 | kthread_stop(wq->rescuer->task); | |
f2e005aa | 3333 | free_mayday_mask(wq->mayday_mask); |
8d9df9f0 | 3334 | kfree(wq->rescuer); |
e22bee78 TH |
3335 | } |
3336 | ||
bdbc5dd7 | 3337 | free_cwqs(wq); |
3af24433 ON |
3338 | kfree(wq); |
3339 | } | |
3340 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
3341 | ||
9f4bd4cd LJ |
3342 | /** |
3343 | * cwq_set_max_active - adjust max_active of a cwq | |
3344 | * @cwq: target cpu_workqueue_struct | |
3345 | * @max_active: new max_active value. | |
3346 | * | |
3347 | * Set @cwq->max_active to @max_active and activate delayed works if | |
3348 | * increased. | |
3349 | * | |
3350 | * CONTEXT: | |
3351 | * spin_lock_irq(gcwq->lock). | |
3352 | */ | |
3353 | static void cwq_set_max_active(struct cpu_workqueue_struct *cwq, int max_active) | |
3354 | { | |
3355 | cwq->max_active = max_active; | |
3356 | ||
3357 | while (!list_empty(&cwq->delayed_works) && | |
3358 | cwq->nr_active < cwq->max_active) | |
3359 | cwq_activate_first_delayed(cwq); | |
3360 | } | |
3361 | ||
dcd989cb TH |
3362 | /** |
3363 | * workqueue_set_max_active - adjust max_active of a workqueue | |
3364 | * @wq: target workqueue | |
3365 | * @max_active: new max_active value. | |
3366 | * | |
3367 | * Set max_active of @wq to @max_active. | |
3368 | * | |
3369 | * CONTEXT: | |
3370 | * Don't call from IRQ context. | |
3371 | */ | |
3372 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
3373 | { | |
3374 | unsigned int cpu; | |
3375 | ||
f3421797 | 3376 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb TH |
3377 | |
3378 | spin_lock(&workqueue_lock); | |
3379 | ||
3380 | wq->saved_max_active = max_active; | |
3381 | ||
f3421797 | 3382 | for_each_cwq_cpu(cpu, wq) { |
dcd989cb TH |
3383 | struct global_cwq *gcwq = get_gcwq(cpu); |
3384 | ||
3385 | spin_lock_irq(&gcwq->lock); | |
3386 | ||
58a69cb4 | 3387 | if (!(wq->flags & WQ_FREEZABLE) || |
dcd989cb | 3388 | !(gcwq->flags & GCWQ_FREEZING)) |
70369b11 | 3389 | cwq_set_max_active(get_cwq(gcwq->cpu, wq), max_active); |
9bfb1839 | 3390 | |
dcd989cb | 3391 | spin_unlock_irq(&gcwq->lock); |
65a64464 | 3392 | } |
93981800 | 3393 | |
dcd989cb | 3394 | spin_unlock(&workqueue_lock); |
15316ba8 | 3395 | } |
dcd989cb | 3396 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 3397 | |
eef6a7d5 | 3398 | /** |
dcd989cb TH |
3399 | * workqueue_congested - test whether a workqueue is congested |
3400 | * @cpu: CPU in question | |
3401 | * @wq: target workqueue | |
eef6a7d5 | 3402 | * |
dcd989cb TH |
3403 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
3404 | * no synchronization around this function and the test result is | |
3405 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 3406 | * |
dcd989cb TH |
3407 | * RETURNS: |
3408 | * %true if congested, %false otherwise. | |
eef6a7d5 | 3409 | */ |
dcd989cb | 3410 | bool workqueue_congested(unsigned int cpu, struct workqueue_struct *wq) |
1da177e4 | 3411 | { |
dcd989cb TH |
3412 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3413 | ||
3414 | return !list_empty(&cwq->delayed_works); | |
1da177e4 | 3415 | } |
dcd989cb | 3416 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 3417 | |
1fa44eca | 3418 | /** |
dcd989cb TH |
3419 | * work_cpu - return the last known associated cpu for @work |
3420 | * @work: the work of interest | |
1fa44eca | 3421 | * |
dcd989cb | 3422 | * RETURNS: |
bdbc5dd7 | 3423 | * CPU number if @work was ever queued. WORK_CPU_NONE otherwise. |
1fa44eca | 3424 | */ |
e2905b29 | 3425 | static unsigned int work_cpu(struct work_struct *work) |
1fa44eca | 3426 | { |
dcd989cb | 3427 | struct global_cwq *gcwq = get_work_gcwq(work); |
1fa44eca | 3428 | |
bdbc5dd7 | 3429 | return gcwq ? gcwq->cpu : WORK_CPU_NONE; |
1fa44eca | 3430 | } |
1fa44eca | 3431 | |
dcd989cb TH |
3432 | /** |
3433 | * work_busy - test whether a work is currently pending or running | |
3434 | * @work: the work to be tested | |
3435 | * | |
3436 | * Test whether @work is currently pending or running. There is no | |
3437 | * synchronization around this function and the test result is | |
3438 | * unreliable and only useful as advisory hints or for debugging. | |
3439 | * Especially for reentrant wqs, the pending state might hide the | |
3440 | * running state. | |
3441 | * | |
3442 | * RETURNS: | |
3443 | * OR'd bitmask of WORK_BUSY_* bits. | |
3444 | */ | |
3445 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 3446 | { |
dcd989cb TH |
3447 | struct global_cwq *gcwq = get_work_gcwq(work); |
3448 | unsigned long flags; | |
3449 | unsigned int ret = 0; | |
1da177e4 | 3450 | |
dcd989cb | 3451 | if (!gcwq) |
999767be | 3452 | return 0; |
1da177e4 | 3453 | |
dcd989cb | 3454 | spin_lock_irqsave(&gcwq->lock, flags); |
1da177e4 | 3455 | |
dcd989cb TH |
3456 | if (work_pending(work)) |
3457 | ret |= WORK_BUSY_PENDING; | |
3458 | if (find_worker_executing_work(gcwq, work)) | |
3459 | ret |= WORK_BUSY_RUNNING; | |
1da177e4 | 3460 | |
dcd989cb | 3461 | spin_unlock_irqrestore(&gcwq->lock, flags); |
1da177e4 | 3462 | |
dcd989cb | 3463 | return ret; |
1da177e4 | 3464 | } |
dcd989cb | 3465 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 3466 | |
db7bccf4 TH |
3467 | /* |
3468 | * CPU hotplug. | |
3469 | * | |
e22bee78 TH |
3470 | * There are two challenges in supporting CPU hotplug. Firstly, there |
3471 | * are a lot of assumptions on strong associations among work, cwq and | |
3472 | * gcwq which make migrating pending and scheduled works very | |
3473 | * difficult to implement without impacting hot paths. Secondly, | |
3474 | * gcwqs serve mix of short, long and very long running works making | |
3475 | * blocked draining impractical. | |
3476 | * | |
24647570 | 3477 | * This is solved by allowing the pools to be disassociated from the CPU |
628c78e7 TH |
3478 | * running as an unbound one and allowing it to be reattached later if the |
3479 | * cpu comes back online. | |
db7bccf4 | 3480 | */ |
1da177e4 | 3481 | |
60373152 | 3482 | /* claim manager positions of all pools */ |
b2eb83d1 | 3483 | static void gcwq_claim_assoc_and_lock(struct global_cwq *gcwq) |
60373152 TH |
3484 | { |
3485 | struct worker_pool *pool; | |
3486 | ||
3487 | for_each_worker_pool(pool, gcwq) | |
b2eb83d1 | 3488 | mutex_lock_nested(&pool->assoc_mutex, pool - gcwq->pools); |
8db25e78 | 3489 | spin_lock_irq(&gcwq->lock); |
60373152 TH |
3490 | } |
3491 | ||
3492 | /* release manager positions */ | |
b2eb83d1 | 3493 | static void gcwq_release_assoc_and_unlock(struct global_cwq *gcwq) |
60373152 TH |
3494 | { |
3495 | struct worker_pool *pool; | |
3496 | ||
8db25e78 | 3497 | spin_unlock_irq(&gcwq->lock); |
60373152 | 3498 | for_each_worker_pool(pool, gcwq) |
b2eb83d1 | 3499 | mutex_unlock(&pool->assoc_mutex); |
60373152 TH |
3500 | } |
3501 | ||
628c78e7 | 3502 | static void gcwq_unbind_fn(struct work_struct *work) |
3af24433 | 3503 | { |
628c78e7 | 3504 | struct global_cwq *gcwq = get_gcwq(smp_processor_id()); |
4ce62e9e | 3505 | struct worker_pool *pool; |
db7bccf4 TH |
3506 | struct worker *worker; |
3507 | struct hlist_node *pos; | |
3508 | int i; | |
3af24433 | 3509 | |
db7bccf4 TH |
3510 | BUG_ON(gcwq->cpu != smp_processor_id()); |
3511 | ||
b2eb83d1 | 3512 | gcwq_claim_assoc_and_lock(gcwq); |
3af24433 | 3513 | |
f2d5a0ee TH |
3514 | /* |
3515 | * We've claimed all manager positions. Make all workers unbound | |
3516 | * and set DISASSOCIATED. Before this, all workers except for the | |
3517 | * ones which are still executing works from before the last CPU | |
3518 | * down must be on the cpu. After this, they may become diasporas. | |
3519 | */ | |
60373152 | 3520 | for_each_worker_pool(pool, gcwq) |
4ce62e9e | 3521 | list_for_each_entry(worker, &pool->idle_list, entry) |
403c821d | 3522 | worker->flags |= WORKER_UNBOUND; |
3af24433 | 3523 | |
db7bccf4 | 3524 | for_each_busy_worker(worker, i, pos, gcwq) |
403c821d | 3525 | worker->flags |= WORKER_UNBOUND; |
06ba38a9 | 3526 | |
24647570 TH |
3527 | for_each_worker_pool(pool, gcwq) |
3528 | pool->flags |= POOL_DISASSOCIATED; | |
f2d5a0ee | 3529 | |
b2eb83d1 | 3530 | gcwq_release_assoc_and_unlock(gcwq); |
628c78e7 | 3531 | |
e22bee78 | 3532 | /* |
403c821d | 3533 | * Call schedule() so that we cross rq->lock and thus can guarantee |
628c78e7 TH |
3534 | * sched callbacks see the %WORKER_UNBOUND flag. This is necessary |
3535 | * as scheduler callbacks may be invoked from other cpus. | |
e22bee78 | 3536 | */ |
e22bee78 | 3537 | schedule(); |
06ba38a9 | 3538 | |
e22bee78 | 3539 | /* |
628c78e7 TH |
3540 | * Sched callbacks are disabled now. Zap nr_running. After this, |
3541 | * nr_running stays zero and need_more_worker() and keep_working() | |
3542 | * are always true as long as the worklist is not empty. @gcwq now | |
3543 | * behaves as unbound (in terms of concurrency management) gcwq | |
3544 | * which is served by workers tied to the CPU. | |
3545 | * | |
3546 | * On return from this function, the current worker would trigger | |
3547 | * unbound chain execution of pending work items if other workers | |
3548 | * didn't already. | |
e22bee78 | 3549 | */ |
4ce62e9e TH |
3550 | for_each_worker_pool(pool, gcwq) |
3551 | atomic_set(get_pool_nr_running(pool), 0); | |
3af24433 | 3552 | } |
3af24433 | 3553 | |
8db25e78 TH |
3554 | /* |
3555 | * Workqueues should be brought up before normal priority CPU notifiers. | |
3556 | * This will be registered high priority CPU notifier. | |
3557 | */ | |
9fdf9b73 | 3558 | static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb, |
8db25e78 TH |
3559 | unsigned long action, |
3560 | void *hcpu) | |
3af24433 ON |
3561 | { |
3562 | unsigned int cpu = (unsigned long)hcpu; | |
db7bccf4 | 3563 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3564 | struct worker_pool *pool; |
3ce63377 | 3565 | |
8db25e78 | 3566 | switch (action & ~CPU_TASKS_FROZEN) { |
3af24433 | 3567 | case CPU_UP_PREPARE: |
4ce62e9e | 3568 | for_each_worker_pool(pool, gcwq) { |
3ce63377 TH |
3569 | struct worker *worker; |
3570 | ||
3571 | if (pool->nr_workers) | |
3572 | continue; | |
3573 | ||
3574 | worker = create_worker(pool); | |
3575 | if (!worker) | |
3576 | return NOTIFY_BAD; | |
3577 | ||
3578 | spin_lock_irq(&gcwq->lock); | |
3579 | start_worker(worker); | |
3580 | spin_unlock_irq(&gcwq->lock); | |
3af24433 | 3581 | } |
8db25e78 | 3582 | break; |
3af24433 | 3583 | |
db7bccf4 TH |
3584 | case CPU_DOWN_FAILED: |
3585 | case CPU_ONLINE: | |
b2eb83d1 | 3586 | gcwq_claim_assoc_and_lock(gcwq); |
24647570 TH |
3587 | for_each_worker_pool(pool, gcwq) |
3588 | pool->flags &= ~POOL_DISASSOCIATED; | |
25511a47 | 3589 | rebind_workers(gcwq); |
b2eb83d1 | 3590 | gcwq_release_assoc_and_unlock(gcwq); |
db7bccf4 | 3591 | break; |
00dfcaf7 | 3592 | } |
65758202 TH |
3593 | return NOTIFY_OK; |
3594 | } | |
3595 | ||
3596 | /* | |
3597 | * Workqueues should be brought down after normal priority CPU notifiers. | |
3598 | * This will be registered as low priority CPU notifier. | |
3599 | */ | |
9fdf9b73 | 3600 | static int __cpuinit workqueue_cpu_down_callback(struct notifier_block *nfb, |
65758202 TH |
3601 | unsigned long action, |
3602 | void *hcpu) | |
3603 | { | |
8db25e78 TH |
3604 | unsigned int cpu = (unsigned long)hcpu; |
3605 | struct work_struct unbind_work; | |
3606 | ||
65758202 TH |
3607 | switch (action & ~CPU_TASKS_FROZEN) { |
3608 | case CPU_DOWN_PREPARE: | |
8db25e78 TH |
3609 | /* unbinding should happen on the local CPU */ |
3610 | INIT_WORK_ONSTACK(&unbind_work, gcwq_unbind_fn); | |
7635d2fd | 3611 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
8db25e78 TH |
3612 | flush_work(&unbind_work); |
3613 | break; | |
65758202 TH |
3614 | } |
3615 | return NOTIFY_OK; | |
3616 | } | |
3617 | ||
2d3854a3 | 3618 | #ifdef CONFIG_SMP |
8ccad40d | 3619 | |
2d3854a3 | 3620 | struct work_for_cpu { |
ed48ece2 | 3621 | struct work_struct work; |
2d3854a3 RR |
3622 | long (*fn)(void *); |
3623 | void *arg; | |
3624 | long ret; | |
3625 | }; | |
3626 | ||
ed48ece2 | 3627 | static void work_for_cpu_fn(struct work_struct *work) |
2d3854a3 | 3628 | { |
ed48ece2 TH |
3629 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
3630 | ||
2d3854a3 RR |
3631 | wfc->ret = wfc->fn(wfc->arg); |
3632 | } | |
3633 | ||
3634 | /** | |
3635 | * work_on_cpu - run a function in user context on a particular cpu | |
3636 | * @cpu: the cpu to run on | |
3637 | * @fn: the function to run | |
3638 | * @arg: the function arg | |
3639 | * | |
31ad9081 RR |
3640 | * This will return the value @fn returns. |
3641 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 3642 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 RR |
3643 | */ |
3644 | long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) | |
3645 | { | |
ed48ece2 | 3646 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; |
6b44003e | 3647 | |
ed48ece2 TH |
3648 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
3649 | schedule_work_on(cpu, &wfc.work); | |
3650 | flush_work(&wfc.work); | |
2d3854a3 RR |
3651 | return wfc.ret; |
3652 | } | |
3653 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
3654 | #endif /* CONFIG_SMP */ | |
3655 | ||
a0a1a5fd TH |
3656 | #ifdef CONFIG_FREEZER |
3657 | ||
3658 | /** | |
3659 | * freeze_workqueues_begin - begin freezing workqueues | |
3660 | * | |
58a69cb4 TH |
3661 | * Start freezing workqueues. After this function returns, all freezable |
3662 | * workqueues will queue new works to their frozen_works list instead of | |
3663 | * gcwq->worklist. | |
a0a1a5fd TH |
3664 | * |
3665 | * CONTEXT: | |
8b03ae3c | 3666 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3667 | */ |
3668 | void freeze_workqueues_begin(void) | |
3669 | { | |
a0a1a5fd TH |
3670 | unsigned int cpu; |
3671 | ||
3672 | spin_lock(&workqueue_lock); | |
3673 | ||
3674 | BUG_ON(workqueue_freezing); | |
3675 | workqueue_freezing = true; | |
3676 | ||
f3421797 | 3677 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3678 | struct global_cwq *gcwq = get_gcwq(cpu); |
bdbc5dd7 | 3679 | struct workqueue_struct *wq; |
8b03ae3c TH |
3680 | |
3681 | spin_lock_irq(&gcwq->lock); | |
3682 | ||
db7bccf4 TH |
3683 | BUG_ON(gcwq->flags & GCWQ_FREEZING); |
3684 | gcwq->flags |= GCWQ_FREEZING; | |
3685 | ||
a0a1a5fd TH |
3686 | list_for_each_entry(wq, &workqueues, list) { |
3687 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3688 | ||
58a69cb4 | 3689 | if (cwq && wq->flags & WQ_FREEZABLE) |
a0a1a5fd | 3690 | cwq->max_active = 0; |
a0a1a5fd | 3691 | } |
8b03ae3c TH |
3692 | |
3693 | spin_unlock_irq(&gcwq->lock); | |
a0a1a5fd TH |
3694 | } |
3695 | ||
3696 | spin_unlock(&workqueue_lock); | |
3697 | } | |
3698 | ||
3699 | /** | |
58a69cb4 | 3700 | * freeze_workqueues_busy - are freezable workqueues still busy? |
a0a1a5fd TH |
3701 | * |
3702 | * Check whether freezing is complete. This function must be called | |
3703 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
3704 | * | |
3705 | * CONTEXT: | |
3706 | * Grabs and releases workqueue_lock. | |
3707 | * | |
3708 | * RETURNS: | |
58a69cb4 TH |
3709 | * %true if some freezable workqueues are still busy. %false if freezing |
3710 | * is complete. | |
a0a1a5fd TH |
3711 | */ |
3712 | bool freeze_workqueues_busy(void) | |
3713 | { | |
a0a1a5fd TH |
3714 | unsigned int cpu; |
3715 | bool busy = false; | |
3716 | ||
3717 | spin_lock(&workqueue_lock); | |
3718 | ||
3719 | BUG_ON(!workqueue_freezing); | |
3720 | ||
f3421797 | 3721 | for_each_gcwq_cpu(cpu) { |
bdbc5dd7 | 3722 | struct workqueue_struct *wq; |
a0a1a5fd TH |
3723 | /* |
3724 | * nr_active is monotonically decreasing. It's safe | |
3725 | * to peek without lock. | |
3726 | */ | |
3727 | list_for_each_entry(wq, &workqueues, list) { | |
3728 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3729 | ||
58a69cb4 | 3730 | if (!cwq || !(wq->flags & WQ_FREEZABLE)) |
a0a1a5fd TH |
3731 | continue; |
3732 | ||
3733 | BUG_ON(cwq->nr_active < 0); | |
3734 | if (cwq->nr_active) { | |
3735 | busy = true; | |
3736 | goto out_unlock; | |
3737 | } | |
3738 | } | |
3739 | } | |
3740 | out_unlock: | |
3741 | spin_unlock(&workqueue_lock); | |
3742 | return busy; | |
3743 | } | |
3744 | ||
3745 | /** | |
3746 | * thaw_workqueues - thaw workqueues | |
3747 | * | |
3748 | * Thaw workqueues. Normal queueing is restored and all collected | |
7e11629d | 3749 | * frozen works are transferred to their respective gcwq worklists. |
a0a1a5fd TH |
3750 | * |
3751 | * CONTEXT: | |
8b03ae3c | 3752 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3753 | */ |
3754 | void thaw_workqueues(void) | |
3755 | { | |
a0a1a5fd TH |
3756 | unsigned int cpu; |
3757 | ||
3758 | spin_lock(&workqueue_lock); | |
3759 | ||
3760 | if (!workqueue_freezing) | |
3761 | goto out_unlock; | |
3762 | ||
f3421797 | 3763 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3764 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3765 | struct worker_pool *pool; |
bdbc5dd7 | 3766 | struct workqueue_struct *wq; |
8b03ae3c TH |
3767 | |
3768 | spin_lock_irq(&gcwq->lock); | |
3769 | ||
db7bccf4 TH |
3770 | BUG_ON(!(gcwq->flags & GCWQ_FREEZING)); |
3771 | gcwq->flags &= ~GCWQ_FREEZING; | |
3772 | ||
a0a1a5fd TH |
3773 | list_for_each_entry(wq, &workqueues, list) { |
3774 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3775 | ||
58a69cb4 | 3776 | if (!cwq || !(wq->flags & WQ_FREEZABLE)) |
a0a1a5fd TH |
3777 | continue; |
3778 | ||
a0a1a5fd | 3779 | /* restore max_active and repopulate worklist */ |
9f4bd4cd | 3780 | cwq_set_max_active(cwq, wq->saved_max_active); |
a0a1a5fd | 3781 | } |
8b03ae3c | 3782 | |
4ce62e9e TH |
3783 | for_each_worker_pool(pool, gcwq) |
3784 | wake_up_worker(pool); | |
e22bee78 | 3785 | |
8b03ae3c | 3786 | spin_unlock_irq(&gcwq->lock); |
a0a1a5fd TH |
3787 | } |
3788 | ||
3789 | workqueue_freezing = false; | |
3790 | out_unlock: | |
3791 | spin_unlock(&workqueue_lock); | |
3792 | } | |
3793 | #endif /* CONFIG_FREEZER */ | |
3794 | ||
6ee0578b | 3795 | static int __init init_workqueues(void) |
1da177e4 | 3796 | { |
c34056a3 TH |
3797 | unsigned int cpu; |
3798 | ||
b5490077 TH |
3799 | /* make sure we have enough bits for OFFQ CPU number */ |
3800 | BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_CPU_SHIFT)) < | |
3801 | WORK_CPU_LAST); | |
3802 | ||
65758202 | 3803 | cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); |
a5b4e57d | 3804 | hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); |
8b03ae3c TH |
3805 | |
3806 | /* initialize gcwqs */ | |
f3421797 | 3807 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3808 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3809 | struct worker_pool *pool; |
8b03ae3c TH |
3810 | |
3811 | spin_lock_init(&gcwq->lock); | |
3812 | gcwq->cpu = cpu; | |
3813 | ||
42f8570f | 3814 | hash_init(gcwq->busy_hash); |
c8e55f36 | 3815 | |
4ce62e9e TH |
3816 | for_each_worker_pool(pool, gcwq) { |
3817 | pool->gcwq = gcwq; | |
24647570 | 3818 | pool->flags |= POOL_DISASSOCIATED; |
4ce62e9e TH |
3819 | INIT_LIST_HEAD(&pool->worklist); |
3820 | INIT_LIST_HEAD(&pool->idle_list); | |
e7577c50 | 3821 | |
4ce62e9e TH |
3822 | init_timer_deferrable(&pool->idle_timer); |
3823 | pool->idle_timer.function = idle_worker_timeout; | |
3824 | pool->idle_timer.data = (unsigned long)pool; | |
e22bee78 | 3825 | |
4ce62e9e TH |
3826 | setup_timer(&pool->mayday_timer, gcwq_mayday_timeout, |
3827 | (unsigned long)pool); | |
3828 | ||
b2eb83d1 | 3829 | mutex_init(&pool->assoc_mutex); |
4ce62e9e TH |
3830 | ida_init(&pool->worker_ida); |
3831 | } | |
8b03ae3c TH |
3832 | } |
3833 | ||
e22bee78 | 3834 | /* create the initial worker */ |
f3421797 | 3835 | for_each_online_gcwq_cpu(cpu) { |
e22bee78 | 3836 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3837 | struct worker_pool *pool; |
e22bee78 | 3838 | |
4ce62e9e TH |
3839 | for_each_worker_pool(pool, gcwq) { |
3840 | struct worker *worker; | |
3841 | ||
24647570 TH |
3842 | if (cpu != WORK_CPU_UNBOUND) |
3843 | pool->flags &= ~POOL_DISASSOCIATED; | |
3844 | ||
bc2ae0f5 | 3845 | worker = create_worker(pool); |
4ce62e9e TH |
3846 | BUG_ON(!worker); |
3847 | spin_lock_irq(&gcwq->lock); | |
3848 | start_worker(worker); | |
3849 | spin_unlock_irq(&gcwq->lock); | |
3850 | } | |
e22bee78 TH |
3851 | } |
3852 | ||
d320c038 | 3853 | system_wq = alloc_workqueue("events", 0, 0); |
1aabe902 | 3854 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
d320c038 | 3855 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
f3421797 TH |
3856 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
3857 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
3858 | system_freezable_wq = alloc_workqueue("events_freezable", |
3859 | WQ_FREEZABLE, 0); | |
1aabe902 | 3860 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
ae930e0f | 3861 | !system_unbound_wq || !system_freezable_wq); |
6ee0578b | 3862 | return 0; |
1da177e4 | 3863 | } |
6ee0578b | 3864 | early_initcall(init_workqueues); |