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 { |
24647570 TH |
49 | /* |
50 | * worker_pool flags | |
bc2ae0f5 | 51 | * |
24647570 | 52 | * A bound pool is either associated or disassociated with its CPU. |
bc2ae0f5 TH |
53 | * While associated (!DISASSOCIATED), all workers are bound to the |
54 | * CPU and none has %WORKER_UNBOUND set and concurrency management | |
55 | * is in effect. | |
56 | * | |
57 | * While DISASSOCIATED, the cpu may be offline and all workers have | |
58 | * %WORKER_UNBOUND set and concurrency management disabled, and may | |
24647570 | 59 | * be executing on any CPU. The pool behaves as an unbound one. |
bc2ae0f5 TH |
60 | * |
61 | * Note that DISASSOCIATED can be flipped only while holding | |
24647570 TH |
62 | * assoc_mutex to avoid changing binding state while |
63 | * create_worker() is in progress. | |
bc2ae0f5 | 64 | */ |
11ebea50 | 65 | POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */ |
552a37e9 | 66 | POOL_MANAGING_WORKERS = 1 << 1, /* managing workers */ |
24647570 | 67 | POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ |
35b6bb63 | 68 | POOL_FREEZING = 1 << 3, /* freeze in progress */ |
db7bccf4 | 69 | |
c8e55f36 TH |
70 | /* worker flags */ |
71 | WORKER_STARTED = 1 << 0, /* started */ | |
72 | WORKER_DIE = 1 << 1, /* die die die */ | |
73 | WORKER_IDLE = 1 << 2, /* is idle */ | |
e22bee78 | 74 | WORKER_PREP = 1 << 3, /* preparing to run works */ |
fb0e7beb | 75 | WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ |
f3421797 | 76 | WORKER_UNBOUND = 1 << 7, /* worker is unbound */ |
e22bee78 | 77 | |
5f7dabfd | 78 | WORKER_NOT_RUNNING = WORKER_PREP | WORKER_UNBOUND | |
403c821d | 79 | WORKER_CPU_INTENSIVE, |
db7bccf4 | 80 | |
e34cdddb | 81 | NR_STD_WORKER_POOLS = 2, /* # standard pools per cpu */ |
4ce62e9e | 82 | |
c8e55f36 | 83 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ |
db7bccf4 | 84 | |
e22bee78 TH |
85 | MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */ |
86 | IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */ | |
87 | ||
3233cdbd TH |
88 | MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2, |
89 | /* call for help after 10ms | |
90 | (min two ticks) */ | |
e22bee78 TH |
91 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ |
92 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
e22bee78 TH |
93 | |
94 | /* | |
95 | * Rescue workers are used only on emergencies and shared by | |
96 | * all cpus. Give -20. | |
97 | */ | |
98 | RESCUER_NICE_LEVEL = -20, | |
3270476a | 99 | HIGHPRI_NICE_LEVEL = -20, |
c8e55f36 | 100 | }; |
1da177e4 LT |
101 | |
102 | /* | |
4690c4ab TH |
103 | * Structure fields follow one of the following exclusion rules. |
104 | * | |
e41e704b TH |
105 | * I: Modifiable by initialization/destruction paths and read-only for |
106 | * everyone else. | |
4690c4ab | 107 | * |
e22bee78 TH |
108 | * P: Preemption protected. Disabling preemption is enough and should |
109 | * only be modified and accessed from the local cpu. | |
110 | * | |
d565ed63 | 111 | * L: pool->lock protected. Access with pool->lock held. |
4690c4ab | 112 | * |
d565ed63 TH |
113 | * X: During normal operation, modification requires pool->lock and should |
114 | * be done only from local cpu. Either disabling preemption on local | |
115 | * cpu or grabbing pool->lock is enough for read access. If | |
116 | * POOL_DISASSOCIATED is set, it's identical to L. | |
e22bee78 | 117 | * |
73f53c4a TH |
118 | * F: wq->flush_mutex protected. |
119 | * | |
4690c4ab | 120 | * W: workqueue_lock protected. |
1da177e4 | 121 | */ |
1da177e4 | 122 | |
2eaebdb3 | 123 | /* struct worker is defined in workqueue_internal.h */ |
c34056a3 | 124 | |
bd7bdd43 | 125 | struct worker_pool { |
d565ed63 | 126 | spinlock_t lock; /* the pool lock */ |
ec22ca5e | 127 | unsigned int cpu; /* I: the associated cpu */ |
9daf9e67 | 128 | int id; /* I: pool ID */ |
11ebea50 | 129 | unsigned int flags; /* X: flags */ |
bd7bdd43 TH |
130 | |
131 | struct list_head worklist; /* L: list of pending works */ | |
132 | int nr_workers; /* L: total number of workers */ | |
ea1abd61 LJ |
133 | |
134 | /* nr_idle includes the ones off idle_list for rebinding */ | |
bd7bdd43 TH |
135 | int nr_idle; /* L: currently idle ones */ |
136 | ||
137 | struct list_head idle_list; /* X: list of idle workers */ | |
138 | struct timer_list idle_timer; /* L: worker idle timeout */ | |
139 | struct timer_list mayday_timer; /* L: SOS timer for workers */ | |
140 | ||
c9e7cf27 TH |
141 | /* workers are chained either in busy_hash or idle_list */ |
142 | DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER); | |
143 | /* L: hash of busy workers */ | |
144 | ||
24647570 | 145 | struct mutex assoc_mutex; /* protect POOL_DISASSOCIATED */ |
bd7bdd43 | 146 | struct ida worker_ida; /* L: for worker IDs */ |
e19e397a TH |
147 | |
148 | /* | |
149 | * The current concurrency level. As it's likely to be accessed | |
150 | * from other CPUs during try_to_wake_up(), put it in a separate | |
151 | * cacheline. | |
152 | */ | |
153 | atomic_t nr_running ____cacheline_aligned_in_smp; | |
8b03ae3c TH |
154 | } ____cacheline_aligned_in_smp; |
155 | ||
1da177e4 | 156 | /* |
112202d9 TH |
157 | * The per-pool workqueue. While queued, the lower WORK_STRUCT_FLAG_BITS |
158 | * of work_struct->data are used for flags and the remaining high bits | |
159 | * point to the pwq; thus, pwqs need to be aligned at two's power of the | |
160 | * number of flag bits. | |
1da177e4 | 161 | */ |
112202d9 | 162 | struct pool_workqueue { |
bd7bdd43 | 163 | struct worker_pool *pool; /* I: the associated pool */ |
4690c4ab | 164 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
165 | int work_color; /* L: current color */ |
166 | int flush_color; /* L: flushing color */ | |
167 | int nr_in_flight[WORK_NR_COLORS]; | |
168 | /* L: nr of in_flight works */ | |
1e19ffc6 | 169 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 170 | int max_active; /* L: max active works */ |
1e19ffc6 | 171 | struct list_head delayed_works; /* L: delayed works */ |
0f900049 | 172 | }; |
1da177e4 | 173 | |
73f53c4a TH |
174 | /* |
175 | * Structure used to wait for workqueue flush. | |
176 | */ | |
177 | struct wq_flusher { | |
178 | struct list_head list; /* F: list of flushers */ | |
179 | int flush_color; /* F: flush color waiting for */ | |
180 | struct completion done; /* flush completion */ | |
181 | }; | |
182 | ||
f2e005aa TH |
183 | /* |
184 | * All cpumasks are assumed to be always set on UP and thus can't be | |
185 | * used to determine whether there's something to be done. | |
186 | */ | |
187 | #ifdef CONFIG_SMP | |
188 | typedef cpumask_var_t mayday_mask_t; | |
189 | #define mayday_test_and_set_cpu(cpu, mask) \ | |
190 | cpumask_test_and_set_cpu((cpu), (mask)) | |
191 | #define mayday_clear_cpu(cpu, mask) cpumask_clear_cpu((cpu), (mask)) | |
192 | #define for_each_mayday_cpu(cpu, mask) for_each_cpu((cpu), (mask)) | |
9c37547a | 193 | #define alloc_mayday_mask(maskp, gfp) zalloc_cpumask_var((maskp), (gfp)) |
f2e005aa TH |
194 | #define free_mayday_mask(mask) free_cpumask_var((mask)) |
195 | #else | |
196 | typedef unsigned long mayday_mask_t; | |
197 | #define mayday_test_and_set_cpu(cpu, mask) test_and_set_bit(0, &(mask)) | |
198 | #define mayday_clear_cpu(cpu, mask) clear_bit(0, &(mask)) | |
199 | #define for_each_mayday_cpu(cpu, mask) if ((cpu) = 0, (mask)) | |
200 | #define alloc_mayday_mask(maskp, gfp) true | |
201 | #define free_mayday_mask(mask) do { } while (0) | |
202 | #endif | |
1da177e4 LT |
203 | |
204 | /* | |
205 | * The externally visible workqueue abstraction is an array of | |
206 | * per-CPU workqueues: | |
207 | */ | |
208 | struct workqueue_struct { | |
9c5a2ba7 | 209 | unsigned int flags; /* W: WQ_* flags */ |
bdbc5dd7 | 210 | union { |
112202d9 TH |
211 | struct pool_workqueue __percpu *pcpu; |
212 | struct pool_workqueue *single; | |
bdbc5dd7 | 213 | unsigned long v; |
112202d9 | 214 | } pool_wq; /* I: pwq's */ |
4690c4ab | 215 | struct list_head list; /* W: list of all workqueues */ |
73f53c4a TH |
216 | |
217 | struct mutex flush_mutex; /* protects wq flushing */ | |
218 | int work_color; /* F: current work color */ | |
219 | int flush_color; /* F: current flush color */ | |
112202d9 | 220 | atomic_t nr_pwqs_to_flush; /* flush in progress */ |
73f53c4a TH |
221 | struct wq_flusher *first_flusher; /* F: first flusher */ |
222 | struct list_head flusher_queue; /* F: flush waiters */ | |
223 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
224 | ||
f2e005aa | 225 | mayday_mask_t mayday_mask; /* cpus requesting rescue */ |
e22bee78 TH |
226 | struct worker *rescuer; /* I: rescue worker */ |
227 | ||
9c5a2ba7 | 228 | int nr_drainers; /* W: drain in progress */ |
112202d9 | 229 | int saved_max_active; /* W: saved pwq max_active */ |
4e6045f1 | 230 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 231 | struct lockdep_map lockdep_map; |
4e6045f1 | 232 | #endif |
b196be89 | 233 | char name[]; /* I: workqueue name */ |
1da177e4 LT |
234 | }; |
235 | ||
d320c038 | 236 | struct workqueue_struct *system_wq __read_mostly; |
d320c038 | 237 | EXPORT_SYMBOL_GPL(system_wq); |
044c782c | 238 | struct workqueue_struct *system_highpri_wq __read_mostly; |
1aabe902 | 239 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
044c782c | 240 | struct workqueue_struct *system_long_wq __read_mostly; |
d320c038 | 241 | EXPORT_SYMBOL_GPL(system_long_wq); |
044c782c | 242 | struct workqueue_struct *system_unbound_wq __read_mostly; |
f3421797 | 243 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
044c782c | 244 | struct workqueue_struct *system_freezable_wq __read_mostly; |
24d51add | 245 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
d320c038 | 246 | |
97bd2347 TH |
247 | #define CREATE_TRACE_POINTS |
248 | #include <trace/events/workqueue.h> | |
249 | ||
38db41d9 | 250 | #define for_each_std_worker_pool(pool, cpu) \ |
a60dc39c TH |
251 | for ((pool) = &std_worker_pools(cpu)[0]; \ |
252 | (pool) < &std_worker_pools(cpu)[NR_STD_WORKER_POOLS]; (pool)++) | |
4ce62e9e | 253 | |
b67bfe0d SL |
254 | #define for_each_busy_worker(worker, i, pool) \ |
255 | hash_for_each(pool->busy_hash, i, worker, hentry) | |
db7bccf4 | 256 | |
706026c2 TH |
257 | static inline int __next_wq_cpu(int cpu, const struct cpumask *mask, |
258 | unsigned int sw) | |
f3421797 TH |
259 | { |
260 | if (cpu < nr_cpu_ids) { | |
261 | if (sw & 1) { | |
262 | cpu = cpumask_next(cpu, mask); | |
263 | if (cpu < nr_cpu_ids) | |
264 | return cpu; | |
265 | } | |
266 | if (sw & 2) | |
267 | return WORK_CPU_UNBOUND; | |
268 | } | |
6be19588 | 269 | return WORK_CPU_END; |
f3421797 TH |
270 | } |
271 | ||
112202d9 | 272 | static inline int __next_pwq_cpu(int cpu, const struct cpumask *mask, |
706026c2 | 273 | struct workqueue_struct *wq) |
f3421797 | 274 | { |
706026c2 | 275 | return __next_wq_cpu(cpu, mask, !(wq->flags & WQ_UNBOUND) ? 1 : 2); |
f3421797 TH |
276 | } |
277 | ||
09884951 TH |
278 | /* |
279 | * CPU iterators | |
280 | * | |
706026c2 | 281 | * An extra cpu number is defined using an invalid cpu number |
09884951 | 282 | * (WORK_CPU_UNBOUND) to host workqueues which are not bound to any |
706026c2 TH |
283 | * specific CPU. The following iterators are similar to for_each_*_cpu() |
284 | * iterators but also considers the unbound CPU. | |
09884951 | 285 | * |
706026c2 TH |
286 | * for_each_wq_cpu() : possible CPUs + WORK_CPU_UNBOUND |
287 | * for_each_online_wq_cpu() : online CPUs + WORK_CPU_UNBOUND | |
112202d9 | 288 | * for_each_pwq_cpu() : possible CPUs for bound workqueues, |
09884951 TH |
289 | * WORK_CPU_UNBOUND for unbound workqueues |
290 | */ | |
706026c2 TH |
291 | #define for_each_wq_cpu(cpu) \ |
292 | for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, 3); \ | |
6be19588 | 293 | (cpu) < WORK_CPU_END; \ |
706026c2 | 294 | (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, 3)) |
f3421797 | 295 | |
706026c2 TH |
296 | #define for_each_online_wq_cpu(cpu) \ |
297 | for ((cpu) = __next_wq_cpu(-1, cpu_online_mask, 3); \ | |
6be19588 | 298 | (cpu) < WORK_CPU_END; \ |
706026c2 | 299 | (cpu) = __next_wq_cpu((cpu), cpu_online_mask, 3)) |
f3421797 | 300 | |
112202d9 TH |
301 | #define for_each_pwq_cpu(cpu, wq) \ |
302 | for ((cpu) = __next_pwq_cpu(-1, cpu_possible_mask, (wq)); \ | |
6be19588 | 303 | (cpu) < WORK_CPU_END; \ |
112202d9 | 304 | (cpu) = __next_pwq_cpu((cpu), cpu_possible_mask, (wq))) |
f3421797 | 305 | |
dc186ad7 TG |
306 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
307 | ||
308 | static struct debug_obj_descr work_debug_descr; | |
309 | ||
99777288 SG |
310 | static void *work_debug_hint(void *addr) |
311 | { | |
312 | return ((struct work_struct *) addr)->func; | |
313 | } | |
314 | ||
dc186ad7 TG |
315 | /* |
316 | * fixup_init is called when: | |
317 | * - an active object is initialized | |
318 | */ | |
319 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
320 | { | |
321 | struct work_struct *work = addr; | |
322 | ||
323 | switch (state) { | |
324 | case ODEBUG_STATE_ACTIVE: | |
325 | cancel_work_sync(work); | |
326 | debug_object_init(work, &work_debug_descr); | |
327 | return 1; | |
328 | default: | |
329 | return 0; | |
330 | } | |
331 | } | |
332 | ||
333 | /* | |
334 | * fixup_activate is called when: | |
335 | * - an active object is activated | |
336 | * - an unknown object is activated (might be a statically initialized object) | |
337 | */ | |
338 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
339 | { | |
340 | struct work_struct *work = addr; | |
341 | ||
342 | switch (state) { | |
343 | ||
344 | case ODEBUG_STATE_NOTAVAILABLE: | |
345 | /* | |
346 | * This is not really a fixup. The work struct was | |
347 | * statically initialized. We just make sure that it | |
348 | * is tracked in the object tracker. | |
349 | */ | |
22df02bb | 350 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
351 | debug_object_init(work, &work_debug_descr); |
352 | debug_object_activate(work, &work_debug_descr); | |
353 | return 0; | |
354 | } | |
355 | WARN_ON_ONCE(1); | |
356 | return 0; | |
357 | ||
358 | case ODEBUG_STATE_ACTIVE: | |
359 | WARN_ON(1); | |
360 | ||
361 | default: | |
362 | return 0; | |
363 | } | |
364 | } | |
365 | ||
366 | /* | |
367 | * fixup_free is called when: | |
368 | * - an active object is freed | |
369 | */ | |
370 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
371 | { | |
372 | struct work_struct *work = addr; | |
373 | ||
374 | switch (state) { | |
375 | case ODEBUG_STATE_ACTIVE: | |
376 | cancel_work_sync(work); | |
377 | debug_object_free(work, &work_debug_descr); | |
378 | return 1; | |
379 | default: | |
380 | return 0; | |
381 | } | |
382 | } | |
383 | ||
384 | static struct debug_obj_descr work_debug_descr = { | |
385 | .name = "work_struct", | |
99777288 | 386 | .debug_hint = work_debug_hint, |
dc186ad7 TG |
387 | .fixup_init = work_fixup_init, |
388 | .fixup_activate = work_fixup_activate, | |
389 | .fixup_free = work_fixup_free, | |
390 | }; | |
391 | ||
392 | static inline void debug_work_activate(struct work_struct *work) | |
393 | { | |
394 | debug_object_activate(work, &work_debug_descr); | |
395 | } | |
396 | ||
397 | static inline void debug_work_deactivate(struct work_struct *work) | |
398 | { | |
399 | debug_object_deactivate(work, &work_debug_descr); | |
400 | } | |
401 | ||
402 | void __init_work(struct work_struct *work, int onstack) | |
403 | { | |
404 | if (onstack) | |
405 | debug_object_init_on_stack(work, &work_debug_descr); | |
406 | else | |
407 | debug_object_init(work, &work_debug_descr); | |
408 | } | |
409 | EXPORT_SYMBOL_GPL(__init_work); | |
410 | ||
411 | void destroy_work_on_stack(struct work_struct *work) | |
412 | { | |
413 | debug_object_free(work, &work_debug_descr); | |
414 | } | |
415 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
416 | ||
417 | #else | |
418 | static inline void debug_work_activate(struct work_struct *work) { } | |
419 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
420 | #endif | |
421 | ||
95402b38 GS |
422 | /* Serializes the accesses to the list of workqueues. */ |
423 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 424 | static LIST_HEAD(workqueues); |
a0a1a5fd | 425 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 426 | |
e22bee78 | 427 | /* |
e19e397a TH |
428 | * The CPU and unbound standard worker pools. The unbound ones have |
429 | * POOL_DISASSOCIATED set, and their workers have WORKER_UNBOUND set. | |
f3421797 | 430 | */ |
e19e397a TH |
431 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS], |
432 | cpu_std_worker_pools); | |
a60dc39c | 433 | static struct worker_pool unbound_std_worker_pools[NR_STD_WORKER_POOLS]; |
f3421797 | 434 | |
9daf9e67 TH |
435 | /* idr of all pools */ |
436 | static DEFINE_MUTEX(worker_pool_idr_mutex); | |
437 | static DEFINE_IDR(worker_pool_idr); | |
438 | ||
c34056a3 | 439 | static int worker_thread(void *__worker); |
1da177e4 | 440 | |
a60dc39c | 441 | static struct worker_pool *std_worker_pools(int cpu) |
8b03ae3c | 442 | { |
f3421797 | 443 | if (cpu != WORK_CPU_UNBOUND) |
a60dc39c | 444 | return per_cpu(cpu_std_worker_pools, cpu); |
f3421797 | 445 | else |
a60dc39c | 446 | return unbound_std_worker_pools; |
8b03ae3c TH |
447 | } |
448 | ||
4e8f0a60 TH |
449 | static int std_worker_pool_pri(struct worker_pool *pool) |
450 | { | |
a60dc39c | 451 | return pool - std_worker_pools(pool->cpu); |
4e8f0a60 TH |
452 | } |
453 | ||
9daf9e67 TH |
454 | /* allocate ID and assign it to @pool */ |
455 | static int worker_pool_assign_id(struct worker_pool *pool) | |
456 | { | |
457 | int ret; | |
458 | ||
459 | mutex_lock(&worker_pool_idr_mutex); | |
460 | idr_pre_get(&worker_pool_idr, GFP_KERNEL); | |
461 | ret = idr_get_new(&worker_pool_idr, pool, &pool->id); | |
462 | mutex_unlock(&worker_pool_idr_mutex); | |
463 | ||
464 | return ret; | |
465 | } | |
466 | ||
7c3eed5c TH |
467 | /* |
468 | * Lookup worker_pool by id. The idr currently is built during boot and | |
469 | * never modified. Don't worry about locking for now. | |
470 | */ | |
471 | static struct worker_pool *worker_pool_by_id(int pool_id) | |
472 | { | |
473 | return idr_find(&worker_pool_idr, pool_id); | |
474 | } | |
475 | ||
d565ed63 TH |
476 | static struct worker_pool *get_std_worker_pool(int cpu, bool highpri) |
477 | { | |
a60dc39c | 478 | struct worker_pool *pools = std_worker_pools(cpu); |
d565ed63 | 479 | |
a60dc39c | 480 | return &pools[highpri]; |
d565ed63 TH |
481 | } |
482 | ||
112202d9 TH |
483 | static struct pool_workqueue *get_pwq(unsigned int cpu, |
484 | struct workqueue_struct *wq) | |
b1f4ec17 | 485 | { |
f3421797 | 486 | if (!(wq->flags & WQ_UNBOUND)) { |
e06ffa1e | 487 | if (likely(cpu < nr_cpu_ids)) |
112202d9 | 488 | return per_cpu_ptr(wq->pool_wq.pcpu, cpu); |
f3421797 | 489 | } else if (likely(cpu == WORK_CPU_UNBOUND)) |
112202d9 | 490 | return wq->pool_wq.single; |
f3421797 | 491 | return NULL; |
b1f4ec17 ON |
492 | } |
493 | ||
73f53c4a TH |
494 | static unsigned int work_color_to_flags(int color) |
495 | { | |
496 | return color << WORK_STRUCT_COLOR_SHIFT; | |
497 | } | |
498 | ||
499 | static int get_work_color(struct work_struct *work) | |
500 | { | |
501 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
502 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
503 | } | |
504 | ||
505 | static int work_next_color(int color) | |
506 | { | |
507 | return (color + 1) % WORK_NR_COLORS; | |
508 | } | |
1da177e4 | 509 | |
14441960 | 510 | /* |
112202d9 TH |
511 | * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data |
512 | * contain the pointer to the queued pwq. Once execution starts, the flag | |
7c3eed5c | 513 | * is cleared and the high bits contain OFFQ flags and pool ID. |
7a22ad75 | 514 | * |
112202d9 TH |
515 | * set_work_pwq(), set_work_pool_and_clear_pending(), mark_work_canceling() |
516 | * and clear_work_data() can be used to set the pwq, pool or clear | |
bbb68dfa TH |
517 | * work->data. These functions should only be called while the work is |
518 | * owned - ie. while the PENDING bit is set. | |
7a22ad75 | 519 | * |
112202d9 | 520 | * get_work_pool() and get_work_pwq() can be used to obtain the pool or pwq |
7c3eed5c | 521 | * corresponding to a work. Pool is available once the work has been |
112202d9 | 522 | * queued anywhere after initialization until it is sync canceled. pwq is |
7c3eed5c | 523 | * available only while the work item is queued. |
7a22ad75 | 524 | * |
bbb68dfa TH |
525 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
526 | * canceled. While being canceled, a work item may have its PENDING set | |
527 | * but stay off timer and worklist for arbitrarily long and nobody should | |
528 | * try to steal the PENDING bit. | |
14441960 | 529 | */ |
7a22ad75 TH |
530 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
531 | unsigned long flags) | |
365970a1 | 532 | { |
6183c009 | 533 | WARN_ON_ONCE(!work_pending(work)); |
7a22ad75 TH |
534 | atomic_long_set(&work->data, data | flags | work_static(work)); |
535 | } | |
365970a1 | 536 | |
112202d9 | 537 | static void set_work_pwq(struct work_struct *work, struct pool_workqueue *pwq, |
7a22ad75 TH |
538 | unsigned long extra_flags) |
539 | { | |
112202d9 TH |
540 | set_work_data(work, (unsigned long)pwq, |
541 | WORK_STRUCT_PENDING | WORK_STRUCT_PWQ | extra_flags); | |
365970a1 DH |
542 | } |
543 | ||
4468a00f LJ |
544 | static void set_work_pool_and_keep_pending(struct work_struct *work, |
545 | int pool_id) | |
546 | { | |
547 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, | |
548 | WORK_STRUCT_PENDING); | |
549 | } | |
550 | ||
7c3eed5c TH |
551 | static void set_work_pool_and_clear_pending(struct work_struct *work, |
552 | int pool_id) | |
7a22ad75 | 553 | { |
23657bb1 TH |
554 | /* |
555 | * The following wmb is paired with the implied mb in | |
556 | * test_and_set_bit(PENDING) and ensures all updates to @work made | |
557 | * here are visible to and precede any updates by the next PENDING | |
558 | * owner. | |
559 | */ | |
560 | smp_wmb(); | |
7c3eed5c | 561 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0); |
7a22ad75 | 562 | } |
f756d5e2 | 563 | |
7a22ad75 | 564 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 565 | { |
7c3eed5c TH |
566 | smp_wmb(); /* see set_work_pool_and_clear_pending() */ |
567 | set_work_data(work, WORK_STRUCT_NO_POOL, 0); | |
1da177e4 LT |
568 | } |
569 | ||
112202d9 | 570 | static struct pool_workqueue *get_work_pwq(struct work_struct *work) |
b1f4ec17 | 571 | { |
e120153d | 572 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 573 | |
112202d9 | 574 | if (data & WORK_STRUCT_PWQ) |
e120153d TH |
575 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); |
576 | else | |
577 | return NULL; | |
4d707b9f ON |
578 | } |
579 | ||
7c3eed5c TH |
580 | /** |
581 | * get_work_pool - return the worker_pool a given work was associated with | |
582 | * @work: the work item of interest | |
583 | * | |
584 | * Return the worker_pool @work was last associated with. %NULL if none. | |
585 | */ | |
586 | static struct worker_pool *get_work_pool(struct work_struct *work) | |
365970a1 | 587 | { |
e120153d | 588 | unsigned long data = atomic_long_read(&work->data); |
7c3eed5c TH |
589 | struct worker_pool *pool; |
590 | int pool_id; | |
7a22ad75 | 591 | |
112202d9 TH |
592 | if (data & WORK_STRUCT_PWQ) |
593 | return ((struct pool_workqueue *) | |
7c3eed5c | 594 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool; |
7a22ad75 | 595 | |
7c3eed5c TH |
596 | pool_id = data >> WORK_OFFQ_POOL_SHIFT; |
597 | if (pool_id == WORK_OFFQ_POOL_NONE) | |
7a22ad75 TH |
598 | return NULL; |
599 | ||
7c3eed5c TH |
600 | pool = worker_pool_by_id(pool_id); |
601 | WARN_ON_ONCE(!pool); | |
602 | return pool; | |
603 | } | |
604 | ||
605 | /** | |
606 | * get_work_pool_id - return the worker pool ID a given work is associated with | |
607 | * @work: the work item of interest | |
608 | * | |
609 | * Return the worker_pool ID @work was last associated with. | |
610 | * %WORK_OFFQ_POOL_NONE if none. | |
611 | */ | |
612 | static int get_work_pool_id(struct work_struct *work) | |
613 | { | |
54d5b7d0 LJ |
614 | unsigned long data = atomic_long_read(&work->data); |
615 | ||
112202d9 TH |
616 | if (data & WORK_STRUCT_PWQ) |
617 | return ((struct pool_workqueue *) | |
54d5b7d0 | 618 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->id; |
7c3eed5c | 619 | |
54d5b7d0 | 620 | return data >> WORK_OFFQ_POOL_SHIFT; |
7c3eed5c TH |
621 | } |
622 | ||
bbb68dfa TH |
623 | static void mark_work_canceling(struct work_struct *work) |
624 | { | |
7c3eed5c | 625 | unsigned long pool_id = get_work_pool_id(work); |
bbb68dfa | 626 | |
7c3eed5c TH |
627 | pool_id <<= WORK_OFFQ_POOL_SHIFT; |
628 | set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING); | |
bbb68dfa TH |
629 | } |
630 | ||
631 | static bool work_is_canceling(struct work_struct *work) | |
632 | { | |
633 | unsigned long data = atomic_long_read(&work->data); | |
634 | ||
112202d9 | 635 | return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING); |
bbb68dfa TH |
636 | } |
637 | ||
e22bee78 | 638 | /* |
3270476a TH |
639 | * Policy functions. These define the policies on how the global worker |
640 | * pools are managed. Unless noted otherwise, these functions assume that | |
d565ed63 | 641 | * they're being called with pool->lock held. |
e22bee78 TH |
642 | */ |
643 | ||
63d95a91 | 644 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 645 | { |
e19e397a | 646 | return !atomic_read(&pool->nr_running); |
a848e3b6 ON |
647 | } |
648 | ||
4594bf15 | 649 | /* |
e22bee78 TH |
650 | * Need to wake up a worker? Called from anything but currently |
651 | * running workers. | |
974271c4 TH |
652 | * |
653 | * Note that, because unbound workers never contribute to nr_running, this | |
706026c2 | 654 | * function will always return %true for unbound pools as long as the |
974271c4 | 655 | * worklist isn't empty. |
4594bf15 | 656 | */ |
63d95a91 | 657 | static bool need_more_worker(struct worker_pool *pool) |
365970a1 | 658 | { |
63d95a91 | 659 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
e22bee78 | 660 | } |
4594bf15 | 661 | |
e22bee78 | 662 | /* Can I start working? Called from busy but !running workers. */ |
63d95a91 | 663 | static bool may_start_working(struct worker_pool *pool) |
e22bee78 | 664 | { |
63d95a91 | 665 | return pool->nr_idle; |
e22bee78 TH |
666 | } |
667 | ||
668 | /* Do I need to keep working? Called from currently running workers. */ | |
63d95a91 | 669 | static bool keep_working(struct worker_pool *pool) |
e22bee78 | 670 | { |
e19e397a TH |
671 | return !list_empty(&pool->worklist) && |
672 | atomic_read(&pool->nr_running) <= 1; | |
e22bee78 TH |
673 | } |
674 | ||
675 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 676 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 677 | { |
63d95a91 | 678 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 679 | } |
365970a1 | 680 | |
e22bee78 | 681 | /* Do I need to be the manager? */ |
63d95a91 | 682 | static bool need_to_manage_workers(struct worker_pool *pool) |
e22bee78 | 683 | { |
63d95a91 | 684 | return need_to_create_worker(pool) || |
11ebea50 | 685 | (pool->flags & POOL_MANAGE_WORKERS); |
e22bee78 TH |
686 | } |
687 | ||
688 | /* Do we have too many workers and should some go away? */ | |
63d95a91 | 689 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 690 | { |
552a37e9 | 691 | bool managing = pool->flags & POOL_MANAGING_WORKERS; |
63d95a91 TH |
692 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
693 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 | 694 | |
ea1abd61 LJ |
695 | /* |
696 | * nr_idle and idle_list may disagree if idle rebinding is in | |
697 | * progress. Never return %true if idle_list is empty. | |
698 | */ | |
699 | if (list_empty(&pool->idle_list)) | |
700 | return false; | |
701 | ||
e22bee78 | 702 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
365970a1 DH |
703 | } |
704 | ||
4d707b9f | 705 | /* |
e22bee78 TH |
706 | * Wake up functions. |
707 | */ | |
708 | ||
7e11629d | 709 | /* Return the first worker. Safe with preemption disabled */ |
63d95a91 | 710 | static struct worker *first_worker(struct worker_pool *pool) |
7e11629d | 711 | { |
63d95a91 | 712 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
713 | return NULL; |
714 | ||
63d95a91 | 715 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
716 | } |
717 | ||
718 | /** | |
719 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 720 | * @pool: worker pool to wake worker from |
7e11629d | 721 | * |
63d95a91 | 722 | * Wake up the first idle worker of @pool. |
7e11629d TH |
723 | * |
724 | * CONTEXT: | |
d565ed63 | 725 | * spin_lock_irq(pool->lock). |
7e11629d | 726 | */ |
63d95a91 | 727 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 728 | { |
63d95a91 | 729 | struct worker *worker = first_worker(pool); |
7e11629d TH |
730 | |
731 | if (likely(worker)) | |
732 | wake_up_process(worker->task); | |
733 | } | |
734 | ||
d302f017 | 735 | /** |
e22bee78 TH |
736 | * wq_worker_waking_up - a worker is waking up |
737 | * @task: task waking up | |
738 | * @cpu: CPU @task is waking up to | |
739 | * | |
740 | * This function is called during try_to_wake_up() when a worker is | |
741 | * being awoken. | |
742 | * | |
743 | * CONTEXT: | |
744 | * spin_lock_irq(rq->lock) | |
745 | */ | |
746 | void wq_worker_waking_up(struct task_struct *task, unsigned int cpu) | |
747 | { | |
748 | struct worker *worker = kthread_data(task); | |
749 | ||
36576000 | 750 | if (!(worker->flags & WORKER_NOT_RUNNING)) { |
ec22ca5e | 751 | WARN_ON_ONCE(worker->pool->cpu != cpu); |
e19e397a | 752 | atomic_inc(&worker->pool->nr_running); |
36576000 | 753 | } |
e22bee78 TH |
754 | } |
755 | ||
756 | /** | |
757 | * wq_worker_sleeping - a worker is going to sleep | |
758 | * @task: task going to sleep | |
759 | * @cpu: CPU in question, must be the current CPU number | |
760 | * | |
761 | * This function is called during schedule() when a busy worker is | |
762 | * going to sleep. Worker on the same cpu can be woken up by | |
763 | * returning pointer to its task. | |
764 | * | |
765 | * CONTEXT: | |
766 | * spin_lock_irq(rq->lock) | |
767 | * | |
768 | * RETURNS: | |
769 | * Worker task on @cpu to wake up, %NULL if none. | |
770 | */ | |
771 | struct task_struct *wq_worker_sleeping(struct task_struct *task, | |
772 | unsigned int cpu) | |
773 | { | |
774 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
111c225a | 775 | struct worker_pool *pool; |
e22bee78 | 776 | |
111c225a TH |
777 | /* |
778 | * Rescuers, which may not have all the fields set up like normal | |
779 | * workers, also reach here, let's not access anything before | |
780 | * checking NOT_RUNNING. | |
781 | */ | |
2d64672e | 782 | if (worker->flags & WORKER_NOT_RUNNING) |
e22bee78 TH |
783 | return NULL; |
784 | ||
111c225a | 785 | pool = worker->pool; |
111c225a | 786 | |
e22bee78 | 787 | /* this can only happen on the local cpu */ |
6183c009 TH |
788 | if (WARN_ON_ONCE(cpu != raw_smp_processor_id())) |
789 | return NULL; | |
e22bee78 TH |
790 | |
791 | /* | |
792 | * The counterpart of the following dec_and_test, implied mb, | |
793 | * worklist not empty test sequence is in insert_work(). | |
794 | * Please read comment there. | |
795 | * | |
628c78e7 TH |
796 | * NOT_RUNNING is clear. This means that we're bound to and |
797 | * running on the local cpu w/ rq lock held and preemption | |
798 | * disabled, which in turn means that none else could be | |
d565ed63 | 799 | * manipulating idle_list, so dereferencing idle_list without pool |
628c78e7 | 800 | * lock is safe. |
e22bee78 | 801 | */ |
e19e397a TH |
802 | if (atomic_dec_and_test(&pool->nr_running) && |
803 | !list_empty(&pool->worklist)) | |
63d95a91 | 804 | to_wakeup = first_worker(pool); |
e22bee78 TH |
805 | return to_wakeup ? to_wakeup->task : NULL; |
806 | } | |
807 | ||
808 | /** | |
809 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 810 | * @worker: self |
d302f017 TH |
811 | * @flags: flags to set |
812 | * @wakeup: wakeup an idle worker if necessary | |
813 | * | |
e22bee78 TH |
814 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
815 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
816 | * woken up. | |
d302f017 | 817 | * |
cb444766 | 818 | * CONTEXT: |
d565ed63 | 819 | * spin_lock_irq(pool->lock) |
d302f017 TH |
820 | */ |
821 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
822 | bool wakeup) | |
823 | { | |
bd7bdd43 | 824 | struct worker_pool *pool = worker->pool; |
e22bee78 | 825 | |
cb444766 TH |
826 | WARN_ON_ONCE(worker->task != current); |
827 | ||
e22bee78 TH |
828 | /* |
829 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
830 | * wake up an idle worker as necessary if requested by | |
831 | * @wakeup. | |
832 | */ | |
833 | if ((flags & WORKER_NOT_RUNNING) && | |
834 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
e22bee78 | 835 | if (wakeup) { |
e19e397a | 836 | if (atomic_dec_and_test(&pool->nr_running) && |
bd7bdd43 | 837 | !list_empty(&pool->worklist)) |
63d95a91 | 838 | wake_up_worker(pool); |
e22bee78 | 839 | } else |
e19e397a | 840 | atomic_dec(&pool->nr_running); |
e22bee78 TH |
841 | } |
842 | ||
d302f017 TH |
843 | worker->flags |= flags; |
844 | } | |
845 | ||
846 | /** | |
e22bee78 | 847 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 848 | * @worker: self |
d302f017 TH |
849 | * @flags: flags to clear |
850 | * | |
e22bee78 | 851 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 852 | * |
cb444766 | 853 | * CONTEXT: |
d565ed63 | 854 | * spin_lock_irq(pool->lock) |
d302f017 TH |
855 | */ |
856 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
857 | { | |
63d95a91 | 858 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
859 | unsigned int oflags = worker->flags; |
860 | ||
cb444766 TH |
861 | WARN_ON_ONCE(worker->task != current); |
862 | ||
d302f017 | 863 | worker->flags &= ~flags; |
e22bee78 | 864 | |
42c025f3 TH |
865 | /* |
866 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
867 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
868 | * of multiple flags, not a single flag. | |
869 | */ | |
e22bee78 TH |
870 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
871 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
e19e397a | 872 | atomic_inc(&pool->nr_running); |
d302f017 TH |
873 | } |
874 | ||
8cca0eea TH |
875 | /** |
876 | * find_worker_executing_work - find worker which is executing a work | |
c9e7cf27 | 877 | * @pool: pool of interest |
8cca0eea TH |
878 | * @work: work to find worker for |
879 | * | |
c9e7cf27 TH |
880 | * Find a worker which is executing @work on @pool by searching |
881 | * @pool->busy_hash which is keyed by the address of @work. For a worker | |
a2c1c57b TH |
882 | * to match, its current execution should match the address of @work and |
883 | * its work function. This is to avoid unwanted dependency between | |
884 | * unrelated work executions through a work item being recycled while still | |
885 | * being executed. | |
886 | * | |
887 | * This is a bit tricky. A work item may be freed once its execution | |
888 | * starts and nothing prevents the freed area from being recycled for | |
889 | * another work item. If the same work item address ends up being reused | |
890 | * before the original execution finishes, workqueue will identify the | |
891 | * recycled work item as currently executing and make it wait until the | |
892 | * current execution finishes, introducing an unwanted dependency. | |
893 | * | |
894 | * This function checks the work item address, work function and workqueue | |
895 | * to avoid false positives. Note that this isn't complete as one may | |
896 | * construct a work function which can introduce dependency onto itself | |
897 | * through a recycled work item. Well, if somebody wants to shoot oneself | |
898 | * in the foot that badly, there's only so much we can do, and if such | |
899 | * deadlock actually occurs, it should be easy to locate the culprit work | |
900 | * function. | |
8cca0eea TH |
901 | * |
902 | * CONTEXT: | |
d565ed63 | 903 | * spin_lock_irq(pool->lock). |
8cca0eea TH |
904 | * |
905 | * RETURNS: | |
906 | * Pointer to worker which is executing @work if found, NULL | |
907 | * otherwise. | |
4d707b9f | 908 | */ |
c9e7cf27 | 909 | static struct worker *find_worker_executing_work(struct worker_pool *pool, |
8cca0eea | 910 | struct work_struct *work) |
4d707b9f | 911 | { |
42f8570f | 912 | struct worker *worker; |
42f8570f | 913 | |
b67bfe0d | 914 | hash_for_each_possible(pool->busy_hash, worker, hentry, |
a2c1c57b TH |
915 | (unsigned long)work) |
916 | if (worker->current_work == work && | |
917 | worker->current_func == work->func) | |
42f8570f SL |
918 | return worker; |
919 | ||
920 | return NULL; | |
4d707b9f ON |
921 | } |
922 | ||
bf4ede01 TH |
923 | /** |
924 | * move_linked_works - move linked works to a list | |
925 | * @work: start of series of works to be scheduled | |
926 | * @head: target list to append @work to | |
927 | * @nextp: out paramter for nested worklist walking | |
928 | * | |
929 | * Schedule linked works starting from @work to @head. Work series to | |
930 | * be scheduled starts at @work and includes any consecutive work with | |
931 | * WORK_STRUCT_LINKED set in its predecessor. | |
932 | * | |
933 | * If @nextp is not NULL, it's updated to point to the next work of | |
934 | * the last scheduled work. This allows move_linked_works() to be | |
935 | * nested inside outer list_for_each_entry_safe(). | |
936 | * | |
937 | * CONTEXT: | |
d565ed63 | 938 | * spin_lock_irq(pool->lock). |
bf4ede01 TH |
939 | */ |
940 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
941 | struct work_struct **nextp) | |
942 | { | |
943 | struct work_struct *n; | |
944 | ||
945 | /* | |
946 | * Linked worklist will always end before the end of the list, | |
947 | * use NULL for list head. | |
948 | */ | |
949 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
950 | list_move_tail(&work->entry, head); | |
951 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
952 | break; | |
953 | } | |
954 | ||
955 | /* | |
956 | * If we're already inside safe list traversal and have moved | |
957 | * multiple works to the scheduled queue, the next position | |
958 | * needs to be updated. | |
959 | */ | |
960 | if (nextp) | |
961 | *nextp = n; | |
962 | } | |
963 | ||
112202d9 | 964 | static void pwq_activate_delayed_work(struct work_struct *work) |
bf4ede01 | 965 | { |
112202d9 | 966 | struct pool_workqueue *pwq = get_work_pwq(work); |
bf4ede01 TH |
967 | |
968 | trace_workqueue_activate_work(work); | |
112202d9 | 969 | move_linked_works(work, &pwq->pool->worklist, NULL); |
bf4ede01 | 970 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); |
112202d9 | 971 | pwq->nr_active++; |
bf4ede01 TH |
972 | } |
973 | ||
112202d9 | 974 | static void pwq_activate_first_delayed(struct pool_workqueue *pwq) |
3aa62497 | 975 | { |
112202d9 | 976 | struct work_struct *work = list_first_entry(&pwq->delayed_works, |
3aa62497 LJ |
977 | struct work_struct, entry); |
978 | ||
112202d9 | 979 | pwq_activate_delayed_work(work); |
3aa62497 LJ |
980 | } |
981 | ||
bf4ede01 | 982 | /** |
112202d9 TH |
983 | * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight |
984 | * @pwq: pwq of interest | |
bf4ede01 | 985 | * @color: color of work which left the queue |
bf4ede01 TH |
986 | * |
987 | * A work either has completed or is removed from pending queue, | |
112202d9 | 988 | * decrement nr_in_flight of its pwq and handle workqueue flushing. |
bf4ede01 TH |
989 | * |
990 | * CONTEXT: | |
d565ed63 | 991 | * spin_lock_irq(pool->lock). |
bf4ede01 | 992 | */ |
112202d9 | 993 | static void pwq_dec_nr_in_flight(struct pool_workqueue *pwq, int color) |
bf4ede01 TH |
994 | { |
995 | /* ignore uncolored works */ | |
996 | if (color == WORK_NO_COLOR) | |
997 | return; | |
998 | ||
112202d9 | 999 | pwq->nr_in_flight[color]--; |
bf4ede01 | 1000 | |
112202d9 TH |
1001 | pwq->nr_active--; |
1002 | if (!list_empty(&pwq->delayed_works)) { | |
b3f9f405 | 1003 | /* one down, submit a delayed one */ |
112202d9 TH |
1004 | if (pwq->nr_active < pwq->max_active) |
1005 | pwq_activate_first_delayed(pwq); | |
bf4ede01 TH |
1006 | } |
1007 | ||
1008 | /* is flush in progress and are we at the flushing tip? */ | |
112202d9 | 1009 | if (likely(pwq->flush_color != color)) |
bf4ede01 TH |
1010 | return; |
1011 | ||
1012 | /* are there still in-flight works? */ | |
112202d9 | 1013 | if (pwq->nr_in_flight[color]) |
bf4ede01 TH |
1014 | return; |
1015 | ||
112202d9 TH |
1016 | /* this pwq is done, clear flush_color */ |
1017 | pwq->flush_color = -1; | |
bf4ede01 TH |
1018 | |
1019 | /* | |
112202d9 | 1020 | * If this was the last pwq, wake up the first flusher. It |
bf4ede01 TH |
1021 | * will handle the rest. |
1022 | */ | |
112202d9 TH |
1023 | if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush)) |
1024 | complete(&pwq->wq->first_flusher->done); | |
bf4ede01 TH |
1025 | } |
1026 | ||
36e227d2 | 1027 | /** |
bbb68dfa | 1028 | * try_to_grab_pending - steal work item from worklist and disable irq |
36e227d2 TH |
1029 | * @work: work item to steal |
1030 | * @is_dwork: @work is a delayed_work | |
bbb68dfa | 1031 | * @flags: place to store irq state |
36e227d2 TH |
1032 | * |
1033 | * Try to grab PENDING bit of @work. This function can handle @work in any | |
1034 | * stable state - idle, on timer or on worklist. Return values are | |
1035 | * | |
1036 | * 1 if @work was pending and we successfully stole PENDING | |
1037 | * 0 if @work was idle and we claimed PENDING | |
1038 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry | |
bbb68dfa TH |
1039 | * -ENOENT if someone else is canceling @work, this state may persist |
1040 | * for arbitrarily long | |
36e227d2 | 1041 | * |
bbb68dfa | 1042 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
e0aecdd8 TH |
1043 | * interrupted while holding PENDING and @work off queue, irq must be |
1044 | * disabled on entry. This, combined with delayed_work->timer being | |
1045 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. | |
bbb68dfa TH |
1046 | * |
1047 | * On successful return, >= 0, irq is disabled and the caller is | |
1048 | * responsible for releasing it using local_irq_restore(*@flags). | |
1049 | * | |
e0aecdd8 | 1050 | * This function is safe to call from any context including IRQ handler. |
bf4ede01 | 1051 | */ |
bbb68dfa TH |
1052 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
1053 | unsigned long *flags) | |
bf4ede01 | 1054 | { |
d565ed63 | 1055 | struct worker_pool *pool; |
112202d9 | 1056 | struct pool_workqueue *pwq; |
bf4ede01 | 1057 | |
bbb68dfa TH |
1058 | local_irq_save(*flags); |
1059 | ||
36e227d2 TH |
1060 | /* try to steal the timer if it exists */ |
1061 | if (is_dwork) { | |
1062 | struct delayed_work *dwork = to_delayed_work(work); | |
1063 | ||
e0aecdd8 TH |
1064 | /* |
1065 | * dwork->timer is irqsafe. If del_timer() fails, it's | |
1066 | * guaranteed that the timer is not queued anywhere and not | |
1067 | * running on the local CPU. | |
1068 | */ | |
36e227d2 TH |
1069 | if (likely(del_timer(&dwork->timer))) |
1070 | return 1; | |
1071 | } | |
1072 | ||
1073 | /* try to claim PENDING the normal way */ | |
bf4ede01 TH |
1074 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1075 | return 0; | |
1076 | ||
1077 | /* | |
1078 | * The queueing is in progress, or it is already queued. Try to | |
1079 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1080 | */ | |
d565ed63 TH |
1081 | pool = get_work_pool(work); |
1082 | if (!pool) | |
bbb68dfa | 1083 | goto fail; |
bf4ede01 | 1084 | |
d565ed63 | 1085 | spin_lock(&pool->lock); |
0b3dae68 | 1086 | /* |
112202d9 TH |
1087 | * work->data is guaranteed to point to pwq only while the work |
1088 | * item is queued on pwq->wq, and both updating work->data to point | |
1089 | * to pwq on queueing and to pool on dequeueing are done under | |
1090 | * pwq->pool->lock. This in turn guarantees that, if work->data | |
1091 | * points to pwq which is associated with a locked pool, the work | |
0b3dae68 LJ |
1092 | * item is currently queued on that pool. |
1093 | */ | |
112202d9 TH |
1094 | pwq = get_work_pwq(work); |
1095 | if (pwq && pwq->pool == pool) { | |
16062836 TH |
1096 | debug_work_deactivate(work); |
1097 | ||
1098 | /* | |
1099 | * A delayed work item cannot be grabbed directly because | |
1100 | * it might have linked NO_COLOR work items which, if left | |
112202d9 | 1101 | * on the delayed_list, will confuse pwq->nr_active |
16062836 TH |
1102 | * management later on and cause stall. Make sure the work |
1103 | * item is activated before grabbing. | |
1104 | */ | |
1105 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) | |
112202d9 | 1106 | pwq_activate_delayed_work(work); |
16062836 TH |
1107 | |
1108 | list_del_init(&work->entry); | |
112202d9 | 1109 | pwq_dec_nr_in_flight(get_work_pwq(work), get_work_color(work)); |
16062836 | 1110 | |
112202d9 | 1111 | /* work->data points to pwq iff queued, point to pool */ |
16062836 TH |
1112 | set_work_pool_and_keep_pending(work, pool->id); |
1113 | ||
1114 | spin_unlock(&pool->lock); | |
1115 | return 1; | |
bf4ede01 | 1116 | } |
d565ed63 | 1117 | spin_unlock(&pool->lock); |
bbb68dfa TH |
1118 | fail: |
1119 | local_irq_restore(*flags); | |
1120 | if (work_is_canceling(work)) | |
1121 | return -ENOENT; | |
1122 | cpu_relax(); | |
36e227d2 | 1123 | return -EAGAIN; |
bf4ede01 TH |
1124 | } |
1125 | ||
4690c4ab | 1126 | /** |
706026c2 | 1127 | * insert_work - insert a work into a pool |
112202d9 | 1128 | * @pwq: pwq @work belongs to |
4690c4ab TH |
1129 | * @work: work to insert |
1130 | * @head: insertion point | |
1131 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
1132 | * | |
112202d9 | 1133 | * Insert @work which belongs to @pwq after @head. @extra_flags is or'd to |
706026c2 | 1134 | * work_struct flags. |
4690c4ab TH |
1135 | * |
1136 | * CONTEXT: | |
d565ed63 | 1137 | * spin_lock_irq(pool->lock). |
4690c4ab | 1138 | */ |
112202d9 TH |
1139 | static void insert_work(struct pool_workqueue *pwq, struct work_struct *work, |
1140 | struct list_head *head, unsigned int extra_flags) | |
b89deed3 | 1141 | { |
112202d9 | 1142 | struct worker_pool *pool = pwq->pool; |
e22bee78 | 1143 | |
4690c4ab | 1144 | /* we own @work, set data and link */ |
112202d9 | 1145 | set_work_pwq(work, pwq, extra_flags); |
1a4d9b0a | 1146 | list_add_tail(&work->entry, head); |
e22bee78 TH |
1147 | |
1148 | /* | |
1149 | * Ensure either worker_sched_deactivated() sees the above | |
1150 | * list_add_tail() or we see zero nr_running to avoid workers | |
1151 | * lying around lazily while there are works to be processed. | |
1152 | */ | |
1153 | smp_mb(); | |
1154 | ||
63d95a91 TH |
1155 | if (__need_more_worker(pool)) |
1156 | wake_up_worker(pool); | |
b89deed3 ON |
1157 | } |
1158 | ||
c8efcc25 TH |
1159 | /* |
1160 | * Test whether @work is being queued from another work executing on the | |
8d03ecfe | 1161 | * same workqueue. |
c8efcc25 TH |
1162 | */ |
1163 | static bool is_chained_work(struct workqueue_struct *wq) | |
1164 | { | |
8d03ecfe TH |
1165 | struct worker *worker; |
1166 | ||
1167 | worker = current_wq_worker(); | |
1168 | /* | |
1169 | * Return %true iff I'm a worker execuing a work item on @wq. If | |
1170 | * I'm @worker, it's safe to dereference it without locking. | |
1171 | */ | |
112202d9 | 1172 | return worker && worker->current_pwq->wq == wq; |
c8efcc25 TH |
1173 | } |
1174 | ||
4690c4ab | 1175 | static void __queue_work(unsigned int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
1176 | struct work_struct *work) |
1177 | { | |
112202d9 | 1178 | struct pool_workqueue *pwq; |
1e19ffc6 | 1179 | struct list_head *worklist; |
8a2e8e5d | 1180 | unsigned int work_flags; |
b75cac93 | 1181 | unsigned int req_cpu = cpu; |
8930caba TH |
1182 | |
1183 | /* | |
1184 | * While a work item is PENDING && off queue, a task trying to | |
1185 | * steal the PENDING will busy-loop waiting for it to either get | |
1186 | * queued or lose PENDING. Grabbing PENDING and queueing should | |
1187 | * happen with IRQ disabled. | |
1188 | */ | |
1189 | WARN_ON_ONCE(!irqs_disabled()); | |
1da177e4 | 1190 | |
dc186ad7 | 1191 | debug_work_activate(work); |
1e19ffc6 | 1192 | |
c8efcc25 | 1193 | /* if dying, only works from the same workqueue are allowed */ |
9c5a2ba7 | 1194 | if (unlikely(wq->flags & WQ_DRAINING) && |
c8efcc25 | 1195 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b TH |
1196 | return; |
1197 | ||
112202d9 | 1198 | /* determine the pwq to use */ |
c7fc77f7 | 1199 | if (!(wq->flags & WQ_UNBOUND)) { |
c9e7cf27 | 1200 | struct worker_pool *last_pool; |
18aa9eff | 1201 | |
57469821 | 1202 | if (cpu == WORK_CPU_UNBOUND) |
c7fc77f7 TH |
1203 | cpu = raw_smp_processor_id(); |
1204 | ||
18aa9eff | 1205 | /* |
dbf2576e TH |
1206 | * It's multi cpu. If @work was previously on a different |
1207 | * cpu, it might still be running there, in which case the | |
1208 | * work needs to be queued on that cpu to guarantee | |
1209 | * non-reentrancy. | |
18aa9eff | 1210 | */ |
112202d9 | 1211 | pwq = get_pwq(cpu, wq); |
c9e7cf27 | 1212 | last_pool = get_work_pool(work); |
dbf2576e | 1213 | |
112202d9 | 1214 | if (last_pool && last_pool != pwq->pool) { |
18aa9eff TH |
1215 | struct worker *worker; |
1216 | ||
d565ed63 | 1217 | spin_lock(&last_pool->lock); |
18aa9eff | 1218 | |
c9e7cf27 | 1219 | worker = find_worker_executing_work(last_pool, work); |
18aa9eff | 1220 | |
112202d9 TH |
1221 | if (worker && worker->current_pwq->wq == wq) { |
1222 | pwq = get_pwq(last_pool->cpu, wq); | |
8594fade | 1223 | } else { |
18aa9eff | 1224 | /* meh... not running there, queue here */ |
d565ed63 | 1225 | spin_unlock(&last_pool->lock); |
112202d9 | 1226 | spin_lock(&pwq->pool->lock); |
18aa9eff | 1227 | } |
8930caba | 1228 | } else { |
112202d9 | 1229 | spin_lock(&pwq->pool->lock); |
8930caba | 1230 | } |
f3421797 | 1231 | } else { |
112202d9 TH |
1232 | pwq = get_pwq(WORK_CPU_UNBOUND, wq); |
1233 | spin_lock(&pwq->pool->lock); | |
502ca9d8 TH |
1234 | } |
1235 | ||
112202d9 TH |
1236 | /* pwq determined, queue */ |
1237 | trace_workqueue_queue_work(req_cpu, pwq, work); | |
502ca9d8 | 1238 | |
f5b2552b | 1239 | if (WARN_ON(!list_empty(&work->entry))) { |
112202d9 | 1240 | spin_unlock(&pwq->pool->lock); |
f5b2552b DC |
1241 | return; |
1242 | } | |
1e19ffc6 | 1243 | |
112202d9 TH |
1244 | pwq->nr_in_flight[pwq->work_color]++; |
1245 | work_flags = work_color_to_flags(pwq->work_color); | |
1e19ffc6 | 1246 | |
112202d9 | 1247 | if (likely(pwq->nr_active < pwq->max_active)) { |
cdadf009 | 1248 | trace_workqueue_activate_work(work); |
112202d9 TH |
1249 | pwq->nr_active++; |
1250 | worklist = &pwq->pool->worklist; | |
8a2e8e5d TH |
1251 | } else { |
1252 | work_flags |= WORK_STRUCT_DELAYED; | |
112202d9 | 1253 | worklist = &pwq->delayed_works; |
8a2e8e5d | 1254 | } |
1e19ffc6 | 1255 | |
112202d9 | 1256 | insert_work(pwq, work, worklist, work_flags); |
1e19ffc6 | 1257 | |
112202d9 | 1258 | spin_unlock(&pwq->pool->lock); |
1da177e4 LT |
1259 | } |
1260 | ||
0fcb78c2 | 1261 | /** |
c1a220e7 ZR |
1262 | * queue_work_on - queue work on specific cpu |
1263 | * @cpu: CPU number to execute work on | |
0fcb78c2 REB |
1264 | * @wq: workqueue to use |
1265 | * @work: work to queue | |
1266 | * | |
d4283e93 | 1267 | * Returns %false if @work was already on a queue, %true otherwise. |
1da177e4 | 1268 | * |
c1a220e7 ZR |
1269 | * We queue the work to a specific CPU, the caller must ensure it |
1270 | * can't go away. | |
1da177e4 | 1271 | */ |
d4283e93 TH |
1272 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
1273 | struct work_struct *work) | |
1da177e4 | 1274 | { |
d4283e93 | 1275 | bool ret = false; |
8930caba | 1276 | unsigned long flags; |
ef1ca236 | 1277 | |
8930caba | 1278 | local_irq_save(flags); |
c1a220e7 | 1279 | |
22df02bb | 1280 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1281 | __queue_work(cpu, wq, work); |
d4283e93 | 1282 | ret = true; |
c1a220e7 | 1283 | } |
ef1ca236 | 1284 | |
8930caba | 1285 | local_irq_restore(flags); |
1da177e4 LT |
1286 | return ret; |
1287 | } | |
c1a220e7 | 1288 | EXPORT_SYMBOL_GPL(queue_work_on); |
1da177e4 | 1289 | |
c1a220e7 | 1290 | /** |
0a13c00e | 1291 | * queue_work - queue work on a workqueue |
c1a220e7 ZR |
1292 | * @wq: workqueue to use |
1293 | * @work: work to queue | |
1294 | * | |
d4283e93 | 1295 | * Returns %false if @work was already on a queue, %true otherwise. |
c1a220e7 | 1296 | * |
0a13c00e TH |
1297 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
1298 | * it can be processed by another CPU. | |
c1a220e7 | 1299 | */ |
d4283e93 | 1300 | bool queue_work(struct workqueue_struct *wq, struct work_struct *work) |
c1a220e7 | 1301 | { |
57469821 | 1302 | return queue_work_on(WORK_CPU_UNBOUND, wq, work); |
c1a220e7 | 1303 | } |
0a13c00e | 1304 | EXPORT_SYMBOL_GPL(queue_work); |
c1a220e7 | 1305 | |
d8e794df | 1306 | void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1307 | { |
52bad64d | 1308 | struct delayed_work *dwork = (struct delayed_work *)__data; |
1da177e4 | 1309 | |
e0aecdd8 | 1310 | /* should have been called from irqsafe timer with irq already off */ |
60c057bc | 1311 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
1da177e4 | 1312 | } |
1438ade5 | 1313 | EXPORT_SYMBOL(delayed_work_timer_fn); |
1da177e4 | 1314 | |
7beb2edf TH |
1315 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
1316 | struct delayed_work *dwork, unsigned long delay) | |
1da177e4 | 1317 | { |
7beb2edf TH |
1318 | struct timer_list *timer = &dwork->timer; |
1319 | struct work_struct *work = &dwork->work; | |
7beb2edf TH |
1320 | |
1321 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || | |
1322 | timer->data != (unsigned long)dwork); | |
fc4b514f TH |
1323 | WARN_ON_ONCE(timer_pending(timer)); |
1324 | WARN_ON_ONCE(!list_empty(&work->entry)); | |
7beb2edf | 1325 | |
8852aac2 TH |
1326 | /* |
1327 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1328 | * both optimization and correctness. The earliest @timer can | |
1329 | * expire is on the closest next tick and delayed_work users depend | |
1330 | * on that there's no such delay when @delay is 0. | |
1331 | */ | |
1332 | if (!delay) { | |
1333 | __queue_work(cpu, wq, &dwork->work); | |
1334 | return; | |
1335 | } | |
1336 | ||
7beb2edf | 1337 | timer_stats_timer_set_start_info(&dwork->timer); |
1da177e4 | 1338 | |
60c057bc | 1339 | dwork->wq = wq; |
1265057f | 1340 | dwork->cpu = cpu; |
7beb2edf TH |
1341 | timer->expires = jiffies + delay; |
1342 | ||
1343 | if (unlikely(cpu != WORK_CPU_UNBOUND)) | |
1344 | add_timer_on(timer, cpu); | |
1345 | else | |
1346 | add_timer(timer); | |
1da177e4 LT |
1347 | } |
1348 | ||
0fcb78c2 REB |
1349 | /** |
1350 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1351 | * @cpu: CPU number to execute work on | |
1352 | * @wq: workqueue to use | |
af9997e4 | 1353 | * @dwork: work to queue |
0fcb78c2 REB |
1354 | * @delay: number of jiffies to wait before queueing |
1355 | * | |
715f1300 TH |
1356 | * Returns %false if @work was already on a queue, %true otherwise. If |
1357 | * @delay is zero and @dwork is idle, it will be scheduled for immediate | |
1358 | * execution. | |
0fcb78c2 | 1359 | */ |
d4283e93 TH |
1360 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
1361 | struct delayed_work *dwork, unsigned long delay) | |
7a6bc1cd | 1362 | { |
52bad64d | 1363 | struct work_struct *work = &dwork->work; |
d4283e93 | 1364 | bool ret = false; |
8930caba | 1365 | unsigned long flags; |
7a6bc1cd | 1366 | |
8930caba TH |
1367 | /* read the comment in __queue_work() */ |
1368 | local_irq_save(flags); | |
7a6bc1cd | 1369 | |
22df02bb | 1370 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7beb2edf | 1371 | __queue_delayed_work(cpu, wq, dwork, delay); |
d4283e93 | 1372 | ret = true; |
7a6bc1cd | 1373 | } |
8a3e77cc | 1374 | |
8930caba | 1375 | local_irq_restore(flags); |
7a6bc1cd VP |
1376 | return ret; |
1377 | } | |
ae90dd5d | 1378 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
c7fc77f7 | 1379 | |
0a13c00e TH |
1380 | /** |
1381 | * queue_delayed_work - queue work on a workqueue after delay | |
1382 | * @wq: workqueue to use | |
1383 | * @dwork: delayable work to queue | |
1384 | * @delay: number of jiffies to wait before queueing | |
1385 | * | |
715f1300 | 1386 | * Equivalent to queue_delayed_work_on() but tries to use the local CPU. |
0a13c00e | 1387 | */ |
d4283e93 | 1388 | bool queue_delayed_work(struct workqueue_struct *wq, |
0a13c00e TH |
1389 | struct delayed_work *dwork, unsigned long delay) |
1390 | { | |
57469821 | 1391 | return queue_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); |
0a13c00e TH |
1392 | } |
1393 | EXPORT_SYMBOL_GPL(queue_delayed_work); | |
c7fc77f7 | 1394 | |
8376fe22 TH |
1395 | /** |
1396 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU | |
1397 | * @cpu: CPU number to execute work on | |
1398 | * @wq: workqueue to use | |
1399 | * @dwork: work to queue | |
1400 | * @delay: number of jiffies to wait before queueing | |
1401 | * | |
1402 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, | |
1403 | * modify @dwork's timer so that it expires after @delay. If @delay is | |
1404 | * zero, @work is guaranteed to be scheduled immediately regardless of its | |
1405 | * current state. | |
1406 | * | |
1407 | * Returns %false if @dwork was idle and queued, %true if @dwork was | |
1408 | * pending and its timer was modified. | |
1409 | * | |
e0aecdd8 | 1410 | * This function is safe to call from any context including IRQ handler. |
8376fe22 TH |
1411 | * See try_to_grab_pending() for details. |
1412 | */ | |
1413 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
1414 | struct delayed_work *dwork, unsigned long delay) | |
1415 | { | |
1416 | unsigned long flags; | |
1417 | int ret; | |
c7fc77f7 | 1418 | |
8376fe22 TH |
1419 | do { |
1420 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
1421 | } while (unlikely(ret == -EAGAIN)); | |
63bc0362 | 1422 | |
8376fe22 TH |
1423 | if (likely(ret >= 0)) { |
1424 | __queue_delayed_work(cpu, wq, dwork, delay); | |
1425 | local_irq_restore(flags); | |
7a6bc1cd | 1426 | } |
8376fe22 TH |
1427 | |
1428 | /* -ENOENT from try_to_grab_pending() becomes %true */ | |
7a6bc1cd VP |
1429 | return ret; |
1430 | } | |
8376fe22 TH |
1431 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
1432 | ||
1433 | /** | |
1434 | * mod_delayed_work - modify delay of or queue a delayed work | |
1435 | * @wq: workqueue to use | |
1436 | * @dwork: work to queue | |
1437 | * @delay: number of jiffies to wait before queueing | |
1438 | * | |
1439 | * mod_delayed_work_on() on local CPU. | |
1440 | */ | |
1441 | bool mod_delayed_work(struct workqueue_struct *wq, struct delayed_work *dwork, | |
1442 | unsigned long delay) | |
1443 | { | |
1444 | return mod_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); | |
1445 | } | |
1446 | EXPORT_SYMBOL_GPL(mod_delayed_work); | |
1da177e4 | 1447 | |
c8e55f36 TH |
1448 | /** |
1449 | * worker_enter_idle - enter idle state | |
1450 | * @worker: worker which is entering idle state | |
1451 | * | |
1452 | * @worker is entering idle state. Update stats and idle timer if | |
1453 | * necessary. | |
1454 | * | |
1455 | * LOCKING: | |
d565ed63 | 1456 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1457 | */ |
1458 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1459 | { |
bd7bdd43 | 1460 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1461 | |
6183c009 TH |
1462 | if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) || |
1463 | WARN_ON_ONCE(!list_empty(&worker->entry) && | |
1464 | (worker->hentry.next || worker->hentry.pprev))) | |
1465 | return; | |
c8e55f36 | 1466 | |
cb444766 TH |
1467 | /* can't use worker_set_flags(), also called from start_worker() */ |
1468 | worker->flags |= WORKER_IDLE; | |
bd7bdd43 | 1469 | pool->nr_idle++; |
e22bee78 | 1470 | worker->last_active = jiffies; |
c8e55f36 TH |
1471 | |
1472 | /* idle_list is LIFO */ | |
bd7bdd43 | 1473 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1474 | |
628c78e7 TH |
1475 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
1476 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); | |
cb444766 | 1477 | |
544ecf31 | 1478 | /* |
706026c2 | 1479 | * Sanity check nr_running. Because wq_unbind_fn() releases |
d565ed63 | 1480 | * pool->lock between setting %WORKER_UNBOUND and zapping |
628c78e7 TH |
1481 | * nr_running, the warning may trigger spuriously. Check iff |
1482 | * unbind is not in progress. | |
544ecf31 | 1483 | */ |
24647570 | 1484 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
bd7bdd43 | 1485 | pool->nr_workers == pool->nr_idle && |
e19e397a | 1486 | atomic_read(&pool->nr_running)); |
c8e55f36 TH |
1487 | } |
1488 | ||
1489 | /** | |
1490 | * worker_leave_idle - leave idle state | |
1491 | * @worker: worker which is leaving idle state | |
1492 | * | |
1493 | * @worker is leaving idle state. Update stats. | |
1494 | * | |
1495 | * LOCKING: | |
d565ed63 | 1496 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1497 | */ |
1498 | static void worker_leave_idle(struct worker *worker) | |
1499 | { | |
bd7bdd43 | 1500 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1501 | |
6183c009 TH |
1502 | if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE))) |
1503 | return; | |
d302f017 | 1504 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1505 | pool->nr_idle--; |
c8e55f36 TH |
1506 | list_del_init(&worker->entry); |
1507 | } | |
1508 | ||
e22bee78 | 1509 | /** |
f36dc67b LJ |
1510 | * worker_maybe_bind_and_lock - try to bind %current to worker_pool and lock it |
1511 | * @pool: target worker_pool | |
1512 | * | |
1513 | * Bind %current to the cpu of @pool if it is associated and lock @pool. | |
e22bee78 TH |
1514 | * |
1515 | * Works which are scheduled while the cpu is online must at least be | |
1516 | * scheduled to a worker which is bound to the cpu so that if they are | |
1517 | * flushed from cpu callbacks while cpu is going down, they are | |
1518 | * guaranteed to execute on the cpu. | |
1519 | * | |
f5faa077 | 1520 | * This function is to be used by unbound workers and rescuers to bind |
e22bee78 TH |
1521 | * themselves to the target cpu and may race with cpu going down or |
1522 | * coming online. kthread_bind() can't be used because it may put the | |
1523 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
706026c2 | 1524 | * verbatim as it's best effort and blocking and pool may be |
e22bee78 TH |
1525 | * [dis]associated in the meantime. |
1526 | * | |
706026c2 | 1527 | * This function tries set_cpus_allowed() and locks pool and verifies the |
24647570 | 1528 | * binding against %POOL_DISASSOCIATED which is set during |
f2d5a0ee TH |
1529 | * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker |
1530 | * enters idle state or fetches works without dropping lock, it can | |
1531 | * guarantee the scheduling requirement described in the first paragraph. | |
e22bee78 TH |
1532 | * |
1533 | * CONTEXT: | |
d565ed63 | 1534 | * Might sleep. Called without any lock but returns with pool->lock |
e22bee78 TH |
1535 | * held. |
1536 | * | |
1537 | * RETURNS: | |
706026c2 | 1538 | * %true if the associated pool is online (@worker is successfully |
e22bee78 TH |
1539 | * bound), %false if offline. |
1540 | */ | |
f36dc67b | 1541 | static bool worker_maybe_bind_and_lock(struct worker_pool *pool) |
d565ed63 | 1542 | __acquires(&pool->lock) |
e22bee78 | 1543 | { |
e22bee78 | 1544 | while (true) { |
4e6045f1 | 1545 | /* |
e22bee78 TH |
1546 | * The following call may fail, succeed or succeed |
1547 | * without actually migrating the task to the cpu if | |
1548 | * it races with cpu hotunplug operation. Verify | |
24647570 | 1549 | * against POOL_DISASSOCIATED. |
4e6045f1 | 1550 | */ |
24647570 | 1551 | if (!(pool->flags & POOL_DISASSOCIATED)) |
f5faa077 | 1552 | set_cpus_allowed_ptr(current, get_cpu_mask(pool->cpu)); |
e22bee78 | 1553 | |
d565ed63 | 1554 | spin_lock_irq(&pool->lock); |
24647570 | 1555 | if (pool->flags & POOL_DISASSOCIATED) |
e22bee78 | 1556 | return false; |
f5faa077 | 1557 | if (task_cpu(current) == pool->cpu && |
e22bee78 | 1558 | cpumask_equal(¤t->cpus_allowed, |
ec22ca5e | 1559 | get_cpu_mask(pool->cpu))) |
e22bee78 | 1560 | return true; |
d565ed63 | 1561 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1562 | |
5035b20f TH |
1563 | /* |
1564 | * We've raced with CPU hot[un]plug. Give it a breather | |
1565 | * and retry migration. cond_resched() is required here; | |
1566 | * otherwise, we might deadlock against cpu_stop trying to | |
1567 | * bring down the CPU on non-preemptive kernel. | |
1568 | */ | |
e22bee78 | 1569 | cpu_relax(); |
5035b20f | 1570 | cond_resched(); |
e22bee78 TH |
1571 | } |
1572 | } | |
1573 | ||
25511a47 | 1574 | /* |
ea1abd61 | 1575 | * Rebind an idle @worker to its CPU. worker_thread() will test |
5f7dabfd | 1576 | * list_empty(@worker->entry) before leaving idle and call this function. |
25511a47 TH |
1577 | */ |
1578 | static void idle_worker_rebind(struct worker *worker) | |
1579 | { | |
5f7dabfd | 1580 | /* CPU may go down again inbetween, clear UNBOUND only on success */ |
f36dc67b | 1581 | if (worker_maybe_bind_and_lock(worker->pool)) |
5f7dabfd | 1582 | worker_clr_flags(worker, WORKER_UNBOUND); |
25511a47 | 1583 | |
ea1abd61 LJ |
1584 | /* rebind complete, become available again */ |
1585 | list_add(&worker->entry, &worker->pool->idle_list); | |
d565ed63 | 1586 | spin_unlock_irq(&worker->pool->lock); |
25511a47 TH |
1587 | } |
1588 | ||
e22bee78 | 1589 | /* |
25511a47 | 1590 | * Function for @worker->rebind.work used to rebind unbound busy workers to |
403c821d TH |
1591 | * the associated cpu which is coming back online. This is scheduled by |
1592 | * cpu up but can race with other cpu hotplug operations and may be | |
1593 | * executed twice without intervening cpu down. | |
e22bee78 | 1594 | */ |
25511a47 | 1595 | static void busy_worker_rebind_fn(struct work_struct *work) |
e22bee78 TH |
1596 | { |
1597 | struct worker *worker = container_of(work, struct worker, rebind_work); | |
e22bee78 | 1598 | |
f36dc67b | 1599 | if (worker_maybe_bind_and_lock(worker->pool)) |
eab6d828 | 1600 | worker_clr_flags(worker, WORKER_UNBOUND); |
e22bee78 | 1601 | |
d565ed63 | 1602 | spin_unlock_irq(&worker->pool->lock); |
e22bee78 TH |
1603 | } |
1604 | ||
25511a47 | 1605 | /** |
94cf58bb TH |
1606 | * rebind_workers - rebind all workers of a pool to the associated CPU |
1607 | * @pool: pool of interest | |
25511a47 | 1608 | * |
94cf58bb | 1609 | * @pool->cpu is coming online. Rebind all workers to the CPU. Rebinding |
25511a47 TH |
1610 | * is different for idle and busy ones. |
1611 | * | |
ea1abd61 LJ |
1612 | * Idle ones will be removed from the idle_list and woken up. They will |
1613 | * add themselves back after completing rebind. This ensures that the | |
1614 | * idle_list doesn't contain any unbound workers when re-bound busy workers | |
1615 | * try to perform local wake-ups for concurrency management. | |
25511a47 | 1616 | * |
ea1abd61 LJ |
1617 | * Busy workers can rebind after they finish their current work items. |
1618 | * Queueing the rebind work item at the head of the scheduled list is | |
1619 | * enough. Note that nr_running will be properly bumped as busy workers | |
1620 | * rebind. | |
25511a47 | 1621 | * |
ea1abd61 LJ |
1622 | * On return, all non-manager workers are scheduled for rebind - see |
1623 | * manage_workers() for the manager special case. Any idle worker | |
1624 | * including the manager will not appear on @idle_list until rebind is | |
1625 | * complete, making local wake-ups safe. | |
25511a47 | 1626 | */ |
94cf58bb | 1627 | static void rebind_workers(struct worker_pool *pool) |
25511a47 | 1628 | { |
ea1abd61 | 1629 | struct worker *worker, *n; |
25511a47 TH |
1630 | int i; |
1631 | ||
94cf58bb TH |
1632 | lockdep_assert_held(&pool->assoc_mutex); |
1633 | lockdep_assert_held(&pool->lock); | |
25511a47 | 1634 | |
5f7dabfd | 1635 | /* dequeue and kick idle ones */ |
94cf58bb TH |
1636 | list_for_each_entry_safe(worker, n, &pool->idle_list, entry) { |
1637 | /* | |
1638 | * idle workers should be off @pool->idle_list until rebind | |
1639 | * is complete to avoid receiving premature local wake-ups. | |
1640 | */ | |
1641 | list_del_init(&worker->entry); | |
25511a47 | 1642 | |
94cf58bb TH |
1643 | /* |
1644 | * worker_thread() will see the above dequeuing and call | |
1645 | * idle_worker_rebind(). | |
1646 | */ | |
1647 | wake_up_process(worker->task); | |
1648 | } | |
25511a47 | 1649 | |
94cf58bb | 1650 | /* rebind busy workers */ |
b67bfe0d | 1651 | for_each_busy_worker(worker, i, pool) { |
94cf58bb TH |
1652 | struct work_struct *rebind_work = &worker->rebind_work; |
1653 | struct workqueue_struct *wq; | |
25511a47 | 1654 | |
94cf58bb TH |
1655 | if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, |
1656 | work_data_bits(rebind_work))) | |
1657 | continue; | |
25511a47 | 1658 | |
94cf58bb | 1659 | debug_work_activate(rebind_work); |
90beca5d | 1660 | |
94cf58bb TH |
1661 | /* |
1662 | * wq doesn't really matter but let's keep @worker->pool | |
112202d9 | 1663 | * and @pwq->pool consistent for sanity. |
94cf58bb TH |
1664 | */ |
1665 | if (std_worker_pool_pri(worker->pool)) | |
1666 | wq = system_highpri_wq; | |
1667 | else | |
1668 | wq = system_wq; | |
1669 | ||
112202d9 | 1670 | insert_work(get_pwq(pool->cpu, wq), rebind_work, |
94cf58bb TH |
1671 | worker->scheduled.next, |
1672 | work_color_to_flags(WORK_NO_COLOR)); | |
ec58815a | 1673 | } |
25511a47 TH |
1674 | } |
1675 | ||
c34056a3 TH |
1676 | static struct worker *alloc_worker(void) |
1677 | { | |
1678 | struct worker *worker; | |
1679 | ||
1680 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1681 | if (worker) { |
1682 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1683 | INIT_LIST_HEAD(&worker->scheduled); |
25511a47 | 1684 | INIT_WORK(&worker->rebind_work, busy_worker_rebind_fn); |
e22bee78 TH |
1685 | /* on creation a worker is in !idle && prep state */ |
1686 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1687 | } |
c34056a3 TH |
1688 | return worker; |
1689 | } | |
1690 | ||
1691 | /** | |
1692 | * create_worker - create a new workqueue worker | |
63d95a91 | 1693 | * @pool: pool the new worker will belong to |
c34056a3 | 1694 | * |
63d95a91 | 1695 | * Create a new worker which is bound to @pool. The returned worker |
c34056a3 TH |
1696 | * can be started by calling start_worker() or destroyed using |
1697 | * destroy_worker(). | |
1698 | * | |
1699 | * CONTEXT: | |
1700 | * Might sleep. Does GFP_KERNEL allocations. | |
1701 | * | |
1702 | * RETURNS: | |
1703 | * Pointer to the newly created worker. | |
1704 | */ | |
bc2ae0f5 | 1705 | static struct worker *create_worker(struct worker_pool *pool) |
c34056a3 | 1706 | { |
e34cdddb | 1707 | const char *pri = std_worker_pool_pri(pool) ? "H" : ""; |
c34056a3 | 1708 | struct worker *worker = NULL; |
f3421797 | 1709 | int id = -1; |
c34056a3 | 1710 | |
d565ed63 | 1711 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1712 | while (ida_get_new(&pool->worker_ida, &id)) { |
d565ed63 | 1713 | spin_unlock_irq(&pool->lock); |
bd7bdd43 | 1714 | if (!ida_pre_get(&pool->worker_ida, GFP_KERNEL)) |
c34056a3 | 1715 | goto fail; |
d565ed63 | 1716 | spin_lock_irq(&pool->lock); |
c34056a3 | 1717 | } |
d565ed63 | 1718 | spin_unlock_irq(&pool->lock); |
c34056a3 TH |
1719 | |
1720 | worker = alloc_worker(); | |
1721 | if (!worker) | |
1722 | goto fail; | |
1723 | ||
bd7bdd43 | 1724 | worker->pool = pool; |
c34056a3 TH |
1725 | worker->id = id; |
1726 | ||
ec22ca5e | 1727 | if (pool->cpu != WORK_CPU_UNBOUND) |
94dcf29a | 1728 | worker->task = kthread_create_on_node(worker_thread, |
ec22ca5e TH |
1729 | worker, cpu_to_node(pool->cpu), |
1730 | "kworker/%u:%d%s", pool->cpu, id, pri); | |
f3421797 TH |
1731 | else |
1732 | worker->task = kthread_create(worker_thread, worker, | |
3270476a | 1733 | "kworker/u:%d%s", id, pri); |
c34056a3 TH |
1734 | if (IS_ERR(worker->task)) |
1735 | goto fail; | |
1736 | ||
e34cdddb | 1737 | if (std_worker_pool_pri(pool)) |
3270476a TH |
1738 | set_user_nice(worker->task, HIGHPRI_NICE_LEVEL); |
1739 | ||
db7bccf4 | 1740 | /* |
bc2ae0f5 | 1741 | * Determine CPU binding of the new worker depending on |
24647570 | 1742 | * %POOL_DISASSOCIATED. The caller is responsible for ensuring the |
bc2ae0f5 TH |
1743 | * flag remains stable across this function. See the comments |
1744 | * above the flag definition for details. | |
1745 | * | |
1746 | * As an unbound worker may later become a regular one if CPU comes | |
1747 | * online, make sure every worker has %PF_THREAD_BOUND set. | |
db7bccf4 | 1748 | */ |
24647570 | 1749 | if (!(pool->flags & POOL_DISASSOCIATED)) { |
ec22ca5e | 1750 | kthread_bind(worker->task, pool->cpu); |
bc2ae0f5 | 1751 | } else { |
db7bccf4 | 1752 | worker->task->flags |= PF_THREAD_BOUND; |
bc2ae0f5 | 1753 | worker->flags |= WORKER_UNBOUND; |
f3421797 | 1754 | } |
c34056a3 TH |
1755 | |
1756 | return worker; | |
1757 | fail: | |
1758 | if (id >= 0) { | |
d565ed63 | 1759 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1760 | ida_remove(&pool->worker_ida, id); |
d565ed63 | 1761 | spin_unlock_irq(&pool->lock); |
c34056a3 TH |
1762 | } |
1763 | kfree(worker); | |
1764 | return NULL; | |
1765 | } | |
1766 | ||
1767 | /** | |
1768 | * start_worker - start a newly created worker | |
1769 | * @worker: worker to start | |
1770 | * | |
706026c2 | 1771 | * Make the pool aware of @worker and start it. |
c34056a3 TH |
1772 | * |
1773 | * CONTEXT: | |
d565ed63 | 1774 | * spin_lock_irq(pool->lock). |
c34056a3 TH |
1775 | */ |
1776 | static void start_worker(struct worker *worker) | |
1777 | { | |
cb444766 | 1778 | worker->flags |= WORKER_STARTED; |
bd7bdd43 | 1779 | worker->pool->nr_workers++; |
c8e55f36 | 1780 | worker_enter_idle(worker); |
c34056a3 TH |
1781 | wake_up_process(worker->task); |
1782 | } | |
1783 | ||
1784 | /** | |
1785 | * destroy_worker - destroy a workqueue worker | |
1786 | * @worker: worker to be destroyed | |
1787 | * | |
706026c2 | 1788 | * Destroy @worker and adjust @pool stats accordingly. |
c8e55f36 TH |
1789 | * |
1790 | * CONTEXT: | |
d565ed63 | 1791 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
c34056a3 TH |
1792 | */ |
1793 | static void destroy_worker(struct worker *worker) | |
1794 | { | |
bd7bdd43 | 1795 | struct worker_pool *pool = worker->pool; |
c34056a3 TH |
1796 | int id = worker->id; |
1797 | ||
1798 | /* sanity check frenzy */ | |
6183c009 TH |
1799 | if (WARN_ON(worker->current_work) || |
1800 | WARN_ON(!list_empty(&worker->scheduled))) | |
1801 | return; | |
c34056a3 | 1802 | |
c8e55f36 | 1803 | if (worker->flags & WORKER_STARTED) |
bd7bdd43 | 1804 | pool->nr_workers--; |
c8e55f36 | 1805 | if (worker->flags & WORKER_IDLE) |
bd7bdd43 | 1806 | pool->nr_idle--; |
c8e55f36 TH |
1807 | |
1808 | list_del_init(&worker->entry); | |
cb444766 | 1809 | worker->flags |= WORKER_DIE; |
c8e55f36 | 1810 | |
d565ed63 | 1811 | spin_unlock_irq(&pool->lock); |
c8e55f36 | 1812 | |
c34056a3 TH |
1813 | kthread_stop(worker->task); |
1814 | kfree(worker); | |
1815 | ||
d565ed63 | 1816 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1817 | ida_remove(&pool->worker_ida, id); |
c34056a3 TH |
1818 | } |
1819 | ||
63d95a91 | 1820 | static void idle_worker_timeout(unsigned long __pool) |
e22bee78 | 1821 | { |
63d95a91 | 1822 | struct worker_pool *pool = (void *)__pool; |
e22bee78 | 1823 | |
d565ed63 | 1824 | spin_lock_irq(&pool->lock); |
e22bee78 | 1825 | |
63d95a91 | 1826 | if (too_many_workers(pool)) { |
e22bee78 TH |
1827 | struct worker *worker; |
1828 | unsigned long expires; | |
1829 | ||
1830 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1831 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1832 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1833 | ||
1834 | if (time_before(jiffies, expires)) | |
63d95a91 | 1835 | mod_timer(&pool->idle_timer, expires); |
e22bee78 TH |
1836 | else { |
1837 | /* it's been idle for too long, wake up manager */ | |
11ebea50 | 1838 | pool->flags |= POOL_MANAGE_WORKERS; |
63d95a91 | 1839 | wake_up_worker(pool); |
d5abe669 | 1840 | } |
e22bee78 TH |
1841 | } |
1842 | ||
d565ed63 | 1843 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1844 | } |
d5abe669 | 1845 | |
e22bee78 TH |
1846 | static bool send_mayday(struct work_struct *work) |
1847 | { | |
112202d9 TH |
1848 | struct pool_workqueue *pwq = get_work_pwq(work); |
1849 | struct workqueue_struct *wq = pwq->wq; | |
f3421797 | 1850 | unsigned int cpu; |
e22bee78 TH |
1851 | |
1852 | if (!(wq->flags & WQ_RESCUER)) | |
1853 | return false; | |
1854 | ||
1855 | /* mayday mayday mayday */ | |
112202d9 | 1856 | cpu = pwq->pool->cpu; |
f3421797 TH |
1857 | /* WORK_CPU_UNBOUND can't be set in cpumask, use cpu 0 instead */ |
1858 | if (cpu == WORK_CPU_UNBOUND) | |
1859 | cpu = 0; | |
f2e005aa | 1860 | if (!mayday_test_and_set_cpu(cpu, wq->mayday_mask)) |
e22bee78 TH |
1861 | wake_up_process(wq->rescuer->task); |
1862 | return true; | |
1863 | } | |
1864 | ||
706026c2 | 1865 | static void pool_mayday_timeout(unsigned long __pool) |
e22bee78 | 1866 | { |
63d95a91 | 1867 | struct worker_pool *pool = (void *)__pool; |
e22bee78 TH |
1868 | struct work_struct *work; |
1869 | ||
d565ed63 | 1870 | spin_lock_irq(&pool->lock); |
e22bee78 | 1871 | |
63d95a91 | 1872 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
1873 | /* |
1874 | * We've been trying to create a new worker but | |
1875 | * haven't been successful. We might be hitting an | |
1876 | * allocation deadlock. Send distress signals to | |
1877 | * rescuers. | |
1878 | */ | |
63d95a91 | 1879 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 1880 | send_mayday(work); |
1da177e4 | 1881 | } |
e22bee78 | 1882 | |
d565ed63 | 1883 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1884 | |
63d95a91 | 1885 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
1886 | } |
1887 | ||
e22bee78 TH |
1888 | /** |
1889 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 1890 | * @pool: pool to create a new worker for |
e22bee78 | 1891 | * |
63d95a91 | 1892 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
1893 | * have at least one idle worker on return from this function. If |
1894 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 1895 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
1896 | * possible allocation deadlock. |
1897 | * | |
1898 | * On return, need_to_create_worker() is guaranteed to be false and | |
1899 | * may_start_working() true. | |
1900 | * | |
1901 | * LOCKING: | |
d565ed63 | 1902 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1903 | * multiple times. Does GFP_KERNEL allocations. Called only from |
1904 | * manager. | |
1905 | * | |
1906 | * RETURNS: | |
d565ed63 | 1907 | * false if no action was taken and pool->lock stayed locked, true |
e22bee78 TH |
1908 | * otherwise. |
1909 | */ | |
63d95a91 | 1910 | static bool maybe_create_worker(struct worker_pool *pool) |
d565ed63 TH |
1911 | __releases(&pool->lock) |
1912 | __acquires(&pool->lock) | |
1da177e4 | 1913 | { |
63d95a91 | 1914 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1915 | return false; |
1916 | restart: | |
d565ed63 | 1917 | spin_unlock_irq(&pool->lock); |
9f9c2364 | 1918 | |
e22bee78 | 1919 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 1920 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
1921 | |
1922 | while (true) { | |
1923 | struct worker *worker; | |
1924 | ||
bc2ae0f5 | 1925 | worker = create_worker(pool); |
e22bee78 | 1926 | if (worker) { |
63d95a91 | 1927 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1928 | spin_lock_irq(&pool->lock); |
e22bee78 | 1929 | start_worker(worker); |
6183c009 TH |
1930 | if (WARN_ON_ONCE(need_to_create_worker(pool))) |
1931 | goto restart; | |
e22bee78 TH |
1932 | return true; |
1933 | } | |
1934 | ||
63d95a91 | 1935 | if (!need_to_create_worker(pool)) |
e22bee78 | 1936 | break; |
1da177e4 | 1937 | |
e22bee78 TH |
1938 | __set_current_state(TASK_INTERRUPTIBLE); |
1939 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1940 | |
63d95a91 | 1941 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1942 | break; |
1943 | } | |
1944 | ||
63d95a91 | 1945 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1946 | spin_lock_irq(&pool->lock); |
63d95a91 | 1947 | if (need_to_create_worker(pool)) |
e22bee78 TH |
1948 | goto restart; |
1949 | return true; | |
1950 | } | |
1951 | ||
1952 | /** | |
1953 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
63d95a91 | 1954 | * @pool: pool to destroy workers for |
e22bee78 | 1955 | * |
63d95a91 | 1956 | * Destroy @pool workers which have been idle for longer than |
e22bee78 TH |
1957 | * IDLE_WORKER_TIMEOUT. |
1958 | * | |
1959 | * LOCKING: | |
d565ed63 | 1960 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1961 | * multiple times. Called only from manager. |
1962 | * | |
1963 | * RETURNS: | |
d565ed63 | 1964 | * false if no action was taken and pool->lock stayed locked, true |
e22bee78 TH |
1965 | * otherwise. |
1966 | */ | |
63d95a91 | 1967 | static bool maybe_destroy_workers(struct worker_pool *pool) |
e22bee78 TH |
1968 | { |
1969 | bool ret = false; | |
1da177e4 | 1970 | |
63d95a91 | 1971 | while (too_many_workers(pool)) { |
e22bee78 TH |
1972 | struct worker *worker; |
1973 | unsigned long expires; | |
3af24433 | 1974 | |
63d95a91 | 1975 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 | 1976 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
85f4186a | 1977 | |
e22bee78 | 1978 | if (time_before(jiffies, expires)) { |
63d95a91 | 1979 | mod_timer(&pool->idle_timer, expires); |
3af24433 | 1980 | break; |
e22bee78 | 1981 | } |
1da177e4 | 1982 | |
e22bee78 TH |
1983 | destroy_worker(worker); |
1984 | ret = true; | |
1da177e4 | 1985 | } |
1e19ffc6 | 1986 | |
e22bee78 | 1987 | return ret; |
1e19ffc6 TH |
1988 | } |
1989 | ||
73f53c4a | 1990 | /** |
e22bee78 TH |
1991 | * manage_workers - manage worker pool |
1992 | * @worker: self | |
73f53c4a | 1993 | * |
706026c2 | 1994 | * Assume the manager role and manage the worker pool @worker belongs |
e22bee78 | 1995 | * to. At any given time, there can be only zero or one manager per |
706026c2 | 1996 | * pool. The exclusion is handled automatically by this function. |
e22bee78 TH |
1997 | * |
1998 | * The caller can safely start processing works on false return. On | |
1999 | * true return, it's guaranteed that need_to_create_worker() is false | |
2000 | * and may_start_working() is true. | |
73f53c4a TH |
2001 | * |
2002 | * CONTEXT: | |
d565ed63 | 2003 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
2004 | * multiple times. Does GFP_KERNEL allocations. |
2005 | * | |
2006 | * RETURNS: | |
d565ed63 TH |
2007 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
2008 | * multiple times. Does GFP_KERNEL allocations. | |
73f53c4a | 2009 | */ |
e22bee78 | 2010 | static bool manage_workers(struct worker *worker) |
73f53c4a | 2011 | { |
63d95a91 | 2012 | struct worker_pool *pool = worker->pool; |
e22bee78 | 2013 | bool ret = false; |
73f53c4a | 2014 | |
ee378aa4 | 2015 | if (pool->flags & POOL_MANAGING_WORKERS) |
e22bee78 | 2016 | return ret; |
1e19ffc6 | 2017 | |
552a37e9 | 2018 | pool->flags |= POOL_MANAGING_WORKERS; |
73f53c4a | 2019 | |
ee378aa4 LJ |
2020 | /* |
2021 | * To simplify both worker management and CPU hotplug, hold off | |
2022 | * management while hotplug is in progress. CPU hotplug path can't | |
2023 | * grab %POOL_MANAGING_WORKERS to achieve this because that can | |
2024 | * lead to idle worker depletion (all become busy thinking someone | |
2025 | * else is managing) which in turn can result in deadlock under | |
b2eb83d1 | 2026 | * extreme circumstances. Use @pool->assoc_mutex to synchronize |
ee378aa4 LJ |
2027 | * manager against CPU hotplug. |
2028 | * | |
b2eb83d1 | 2029 | * assoc_mutex would always be free unless CPU hotplug is in |
d565ed63 | 2030 | * progress. trylock first without dropping @pool->lock. |
ee378aa4 | 2031 | */ |
b2eb83d1 | 2032 | if (unlikely(!mutex_trylock(&pool->assoc_mutex))) { |
d565ed63 | 2033 | spin_unlock_irq(&pool->lock); |
b2eb83d1 | 2034 | mutex_lock(&pool->assoc_mutex); |
ee378aa4 LJ |
2035 | /* |
2036 | * CPU hotplug could have happened while we were waiting | |
b2eb83d1 | 2037 | * for assoc_mutex. Hotplug itself can't handle us |
ee378aa4 | 2038 | * because manager isn't either on idle or busy list, and |
706026c2 | 2039 | * @pool's state and ours could have deviated. |
ee378aa4 | 2040 | * |
b2eb83d1 | 2041 | * As hotplug is now excluded via assoc_mutex, we can |
ee378aa4 | 2042 | * simply try to bind. It will succeed or fail depending |
706026c2 | 2043 | * on @pool's current state. Try it and adjust |
ee378aa4 LJ |
2044 | * %WORKER_UNBOUND accordingly. |
2045 | */ | |
f36dc67b | 2046 | if (worker_maybe_bind_and_lock(pool)) |
ee378aa4 LJ |
2047 | worker->flags &= ~WORKER_UNBOUND; |
2048 | else | |
2049 | worker->flags |= WORKER_UNBOUND; | |
73f53c4a | 2050 | |
ee378aa4 LJ |
2051 | ret = true; |
2052 | } | |
73f53c4a | 2053 | |
11ebea50 | 2054 | pool->flags &= ~POOL_MANAGE_WORKERS; |
73f53c4a TH |
2055 | |
2056 | /* | |
e22bee78 TH |
2057 | * Destroy and then create so that may_start_working() is true |
2058 | * on return. | |
73f53c4a | 2059 | */ |
63d95a91 TH |
2060 | ret |= maybe_destroy_workers(pool); |
2061 | ret |= maybe_create_worker(pool); | |
e22bee78 | 2062 | |
552a37e9 | 2063 | pool->flags &= ~POOL_MANAGING_WORKERS; |
b2eb83d1 | 2064 | mutex_unlock(&pool->assoc_mutex); |
e22bee78 | 2065 | return ret; |
73f53c4a TH |
2066 | } |
2067 | ||
a62428c0 TH |
2068 | /** |
2069 | * process_one_work - process single work | |
c34056a3 | 2070 | * @worker: self |
a62428c0 TH |
2071 | * @work: work to process |
2072 | * | |
2073 | * Process @work. This function contains all the logics necessary to | |
2074 | * process a single work including synchronization against and | |
2075 | * interaction with other workers on the same cpu, queueing and | |
2076 | * flushing. As long as context requirement is met, any worker can | |
2077 | * call this function to process a work. | |
2078 | * | |
2079 | * CONTEXT: | |
d565ed63 | 2080 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
a62428c0 | 2081 | */ |
c34056a3 | 2082 | static void process_one_work(struct worker *worker, struct work_struct *work) |
d565ed63 TH |
2083 | __releases(&pool->lock) |
2084 | __acquires(&pool->lock) | |
a62428c0 | 2085 | { |
112202d9 | 2086 | struct pool_workqueue *pwq = get_work_pwq(work); |
bd7bdd43 | 2087 | struct worker_pool *pool = worker->pool; |
112202d9 | 2088 | bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE; |
73f53c4a | 2089 | int work_color; |
7e11629d | 2090 | struct worker *collision; |
a62428c0 TH |
2091 | #ifdef CONFIG_LOCKDEP |
2092 | /* | |
2093 | * It is permissible to free the struct work_struct from | |
2094 | * inside the function that is called from it, this we need to | |
2095 | * take into account for lockdep too. To avoid bogus "held | |
2096 | * lock freed" warnings as well as problems when looking into | |
2097 | * work->lockdep_map, make a copy and use that here. | |
2098 | */ | |
4d82a1de PZ |
2099 | struct lockdep_map lockdep_map; |
2100 | ||
2101 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 2102 | #endif |
6fec10a1 TH |
2103 | /* |
2104 | * Ensure we're on the correct CPU. DISASSOCIATED test is | |
2105 | * necessary to avoid spurious warnings from rescuers servicing the | |
24647570 | 2106 | * unbound or a disassociated pool. |
6fec10a1 | 2107 | */ |
5f7dabfd | 2108 | WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) && |
24647570 | 2109 | !(pool->flags & POOL_DISASSOCIATED) && |
ec22ca5e | 2110 | raw_smp_processor_id() != pool->cpu); |
25511a47 | 2111 | |
7e11629d TH |
2112 | /* |
2113 | * A single work shouldn't be executed concurrently by | |
2114 | * multiple workers on a single cpu. Check whether anyone is | |
2115 | * already processing the work. If so, defer the work to the | |
2116 | * currently executing one. | |
2117 | */ | |
c9e7cf27 | 2118 | collision = find_worker_executing_work(pool, work); |
7e11629d TH |
2119 | if (unlikely(collision)) { |
2120 | move_linked_works(work, &collision->scheduled, NULL); | |
2121 | return; | |
2122 | } | |
2123 | ||
8930caba | 2124 | /* claim and dequeue */ |
a62428c0 | 2125 | debug_work_deactivate(work); |
c9e7cf27 | 2126 | hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work); |
c34056a3 | 2127 | worker->current_work = work; |
a2c1c57b | 2128 | worker->current_func = work->func; |
112202d9 | 2129 | worker->current_pwq = pwq; |
73f53c4a | 2130 | work_color = get_work_color(work); |
7a22ad75 | 2131 | |
a62428c0 TH |
2132 | list_del_init(&work->entry); |
2133 | ||
fb0e7beb TH |
2134 | /* |
2135 | * CPU intensive works don't participate in concurrency | |
2136 | * management. They're the scheduler's responsibility. | |
2137 | */ | |
2138 | if (unlikely(cpu_intensive)) | |
2139 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
2140 | ||
974271c4 | 2141 | /* |
d565ed63 | 2142 | * Unbound pool isn't concurrency managed and work items should be |
974271c4 TH |
2143 | * executed ASAP. Wake up another worker if necessary. |
2144 | */ | |
63d95a91 TH |
2145 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
2146 | wake_up_worker(pool); | |
974271c4 | 2147 | |
8930caba | 2148 | /* |
7c3eed5c | 2149 | * Record the last pool and clear PENDING which should be the last |
d565ed63 | 2150 | * update to @work. Also, do this inside @pool->lock so that |
23657bb1 TH |
2151 | * PENDING and queued state changes happen together while IRQ is |
2152 | * disabled. | |
8930caba | 2153 | */ |
7c3eed5c | 2154 | set_work_pool_and_clear_pending(work, pool->id); |
a62428c0 | 2155 | |
d565ed63 | 2156 | spin_unlock_irq(&pool->lock); |
a62428c0 | 2157 | |
112202d9 | 2158 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
a62428c0 | 2159 | lock_map_acquire(&lockdep_map); |
e36c886a | 2160 | trace_workqueue_execute_start(work); |
a2c1c57b | 2161 | worker->current_func(work); |
e36c886a AV |
2162 | /* |
2163 | * While we must be careful to not use "work" after this, the trace | |
2164 | * point will only record its address. | |
2165 | */ | |
2166 | trace_workqueue_execute_end(work); | |
a62428c0 | 2167 | lock_map_release(&lockdep_map); |
112202d9 | 2168 | lock_map_release(&pwq->wq->lockdep_map); |
a62428c0 TH |
2169 | |
2170 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
044c782c VI |
2171 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
2172 | " last function: %pf\n", | |
a2c1c57b TH |
2173 | current->comm, preempt_count(), task_pid_nr(current), |
2174 | worker->current_func); | |
a62428c0 TH |
2175 | debug_show_held_locks(current); |
2176 | dump_stack(); | |
2177 | } | |
2178 | ||
d565ed63 | 2179 | spin_lock_irq(&pool->lock); |
a62428c0 | 2180 | |
fb0e7beb TH |
2181 | /* clear cpu intensive status */ |
2182 | if (unlikely(cpu_intensive)) | |
2183 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
2184 | ||
a62428c0 | 2185 | /* we're done with it, release */ |
42f8570f | 2186 | hash_del(&worker->hentry); |
c34056a3 | 2187 | worker->current_work = NULL; |
a2c1c57b | 2188 | worker->current_func = NULL; |
112202d9 TH |
2189 | worker->current_pwq = NULL; |
2190 | pwq_dec_nr_in_flight(pwq, work_color); | |
a62428c0 TH |
2191 | } |
2192 | ||
affee4b2 TH |
2193 | /** |
2194 | * process_scheduled_works - process scheduled works | |
2195 | * @worker: self | |
2196 | * | |
2197 | * Process all scheduled works. Please note that the scheduled list | |
2198 | * may change while processing a work, so this function repeatedly | |
2199 | * fetches a work from the top and executes it. | |
2200 | * | |
2201 | * CONTEXT: | |
d565ed63 | 2202 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
affee4b2 TH |
2203 | * multiple times. |
2204 | */ | |
2205 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 2206 | { |
affee4b2 TH |
2207 | while (!list_empty(&worker->scheduled)) { |
2208 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 2209 | struct work_struct, entry); |
c34056a3 | 2210 | process_one_work(worker, work); |
1da177e4 | 2211 | } |
1da177e4 LT |
2212 | } |
2213 | ||
4690c4ab TH |
2214 | /** |
2215 | * worker_thread - the worker thread function | |
c34056a3 | 2216 | * @__worker: self |
4690c4ab | 2217 | * |
706026c2 TH |
2218 | * The worker thread function. There are NR_CPU_WORKER_POOLS dynamic pools |
2219 | * of these per each cpu. These workers process all works regardless of | |
e22bee78 TH |
2220 | * their specific target workqueue. The only exception is works which |
2221 | * belong to workqueues with a rescuer which will be explained in | |
2222 | * rescuer_thread(). | |
4690c4ab | 2223 | */ |
c34056a3 | 2224 | static int worker_thread(void *__worker) |
1da177e4 | 2225 | { |
c34056a3 | 2226 | struct worker *worker = __worker; |
bd7bdd43 | 2227 | struct worker_pool *pool = worker->pool; |
1da177e4 | 2228 | |
e22bee78 TH |
2229 | /* tell the scheduler that this is a workqueue worker */ |
2230 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 2231 | woke_up: |
d565ed63 | 2232 | spin_lock_irq(&pool->lock); |
1da177e4 | 2233 | |
5f7dabfd LJ |
2234 | /* we are off idle list if destruction or rebind is requested */ |
2235 | if (unlikely(list_empty(&worker->entry))) { | |
d565ed63 | 2236 | spin_unlock_irq(&pool->lock); |
25511a47 | 2237 | |
5f7dabfd | 2238 | /* if DIE is set, destruction is requested */ |
25511a47 TH |
2239 | if (worker->flags & WORKER_DIE) { |
2240 | worker->task->flags &= ~PF_WQ_WORKER; | |
2241 | return 0; | |
2242 | } | |
2243 | ||
5f7dabfd | 2244 | /* otherwise, rebind */ |
25511a47 TH |
2245 | idle_worker_rebind(worker); |
2246 | goto woke_up; | |
c8e55f36 | 2247 | } |
affee4b2 | 2248 | |
c8e55f36 | 2249 | worker_leave_idle(worker); |
db7bccf4 | 2250 | recheck: |
e22bee78 | 2251 | /* no more worker necessary? */ |
63d95a91 | 2252 | if (!need_more_worker(pool)) |
e22bee78 TH |
2253 | goto sleep; |
2254 | ||
2255 | /* do we need to manage? */ | |
63d95a91 | 2256 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
2257 | goto recheck; |
2258 | ||
c8e55f36 TH |
2259 | /* |
2260 | * ->scheduled list can only be filled while a worker is | |
2261 | * preparing to process a work or actually processing it. | |
2262 | * Make sure nobody diddled with it while I was sleeping. | |
2263 | */ | |
6183c009 | 2264 | WARN_ON_ONCE(!list_empty(&worker->scheduled)); |
c8e55f36 | 2265 | |
e22bee78 TH |
2266 | /* |
2267 | * When control reaches this point, we're guaranteed to have | |
2268 | * at least one idle worker or that someone else has already | |
2269 | * assumed the manager role. | |
2270 | */ | |
2271 | worker_clr_flags(worker, WORKER_PREP); | |
2272 | ||
2273 | do { | |
c8e55f36 | 2274 | struct work_struct *work = |
bd7bdd43 | 2275 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2276 | struct work_struct, entry); |
2277 | ||
2278 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
2279 | /* optimization path, not strictly necessary */ | |
2280 | process_one_work(worker, work); | |
2281 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2282 | process_scheduled_works(worker); |
c8e55f36 TH |
2283 | } else { |
2284 | move_linked_works(work, &worker->scheduled, NULL); | |
2285 | process_scheduled_works(worker); | |
affee4b2 | 2286 | } |
63d95a91 | 2287 | } while (keep_working(pool)); |
e22bee78 TH |
2288 | |
2289 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 2290 | sleep: |
63d95a91 | 2291 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
e22bee78 | 2292 | goto recheck; |
d313dd85 | 2293 | |
c8e55f36 | 2294 | /* |
d565ed63 TH |
2295 | * pool->lock is held and there's no work to process and no need to |
2296 | * manage, sleep. Workers are woken up only while holding | |
2297 | * pool->lock or from local cpu, so setting the current state | |
2298 | * before releasing pool->lock is enough to prevent losing any | |
2299 | * event. | |
c8e55f36 TH |
2300 | */ |
2301 | worker_enter_idle(worker); | |
2302 | __set_current_state(TASK_INTERRUPTIBLE); | |
d565ed63 | 2303 | spin_unlock_irq(&pool->lock); |
c8e55f36 TH |
2304 | schedule(); |
2305 | goto woke_up; | |
1da177e4 LT |
2306 | } |
2307 | ||
e22bee78 TH |
2308 | /** |
2309 | * rescuer_thread - the rescuer thread function | |
111c225a | 2310 | * @__rescuer: self |
e22bee78 TH |
2311 | * |
2312 | * Workqueue rescuer thread function. There's one rescuer for each | |
2313 | * workqueue which has WQ_RESCUER set. | |
2314 | * | |
706026c2 | 2315 | * Regular work processing on a pool may block trying to create a new |
e22bee78 TH |
2316 | * worker which uses GFP_KERNEL allocation which has slight chance of |
2317 | * developing into deadlock if some works currently on the same queue | |
2318 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2319 | * the problem rescuer solves. | |
2320 | * | |
706026c2 TH |
2321 | * When such condition is possible, the pool summons rescuers of all |
2322 | * workqueues which have works queued on the pool and let them process | |
e22bee78 TH |
2323 | * those works so that forward progress can be guaranteed. |
2324 | * | |
2325 | * This should happen rarely. | |
2326 | */ | |
111c225a | 2327 | static int rescuer_thread(void *__rescuer) |
e22bee78 | 2328 | { |
111c225a TH |
2329 | struct worker *rescuer = __rescuer; |
2330 | struct workqueue_struct *wq = rescuer->rescue_wq; | |
e22bee78 | 2331 | struct list_head *scheduled = &rescuer->scheduled; |
f3421797 | 2332 | bool is_unbound = wq->flags & WQ_UNBOUND; |
e22bee78 TH |
2333 | unsigned int cpu; |
2334 | ||
2335 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
111c225a TH |
2336 | |
2337 | /* | |
2338 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it | |
2339 | * doesn't participate in concurrency management. | |
2340 | */ | |
2341 | rescuer->task->flags |= PF_WQ_WORKER; | |
e22bee78 TH |
2342 | repeat: |
2343 | set_current_state(TASK_INTERRUPTIBLE); | |
2344 | ||
412d32e6 MG |
2345 | if (kthread_should_stop()) { |
2346 | __set_current_state(TASK_RUNNING); | |
111c225a | 2347 | rescuer->task->flags &= ~PF_WQ_WORKER; |
e22bee78 | 2348 | return 0; |
412d32e6 | 2349 | } |
e22bee78 | 2350 | |
f3421797 TH |
2351 | /* |
2352 | * See whether any cpu is asking for help. Unbounded | |
2353 | * workqueues use cpu 0 in mayday_mask for CPU_UNBOUND. | |
2354 | */ | |
f2e005aa | 2355 | for_each_mayday_cpu(cpu, wq->mayday_mask) { |
f3421797 | 2356 | unsigned int tcpu = is_unbound ? WORK_CPU_UNBOUND : cpu; |
112202d9 TH |
2357 | struct pool_workqueue *pwq = get_pwq(tcpu, wq); |
2358 | struct worker_pool *pool = pwq->pool; | |
e22bee78 TH |
2359 | struct work_struct *work, *n; |
2360 | ||
2361 | __set_current_state(TASK_RUNNING); | |
f2e005aa | 2362 | mayday_clear_cpu(cpu, wq->mayday_mask); |
e22bee78 TH |
2363 | |
2364 | /* migrate to the target cpu if possible */ | |
f36dc67b | 2365 | worker_maybe_bind_and_lock(pool); |
b3104104 | 2366 | rescuer->pool = pool; |
e22bee78 TH |
2367 | |
2368 | /* | |
2369 | * Slurp in all works issued via this workqueue and | |
2370 | * process'em. | |
2371 | */ | |
6183c009 | 2372 | WARN_ON_ONCE(!list_empty(&rescuer->scheduled)); |
bd7bdd43 | 2373 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
112202d9 | 2374 | if (get_work_pwq(work) == pwq) |
e22bee78 TH |
2375 | move_linked_works(work, scheduled, &n); |
2376 | ||
2377 | process_scheduled_works(rescuer); | |
7576958a TH |
2378 | |
2379 | /* | |
d565ed63 | 2380 | * Leave this pool. If keep_working() is %true, notify a |
7576958a TH |
2381 | * regular worker; otherwise, we end up with 0 concurrency |
2382 | * and stalling the execution. | |
2383 | */ | |
63d95a91 TH |
2384 | if (keep_working(pool)) |
2385 | wake_up_worker(pool); | |
7576958a | 2386 | |
b3104104 | 2387 | rescuer->pool = NULL; |
d565ed63 | 2388 | spin_unlock_irq(&pool->lock); |
e22bee78 TH |
2389 | } |
2390 | ||
111c225a TH |
2391 | /* rescuers should never participate in concurrency management */ |
2392 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); | |
e22bee78 TH |
2393 | schedule(); |
2394 | goto repeat; | |
1da177e4 LT |
2395 | } |
2396 | ||
fc2e4d70 ON |
2397 | struct wq_barrier { |
2398 | struct work_struct work; | |
2399 | struct completion done; | |
2400 | }; | |
2401 | ||
2402 | static void wq_barrier_func(struct work_struct *work) | |
2403 | { | |
2404 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2405 | complete(&barr->done); | |
2406 | } | |
2407 | ||
4690c4ab TH |
2408 | /** |
2409 | * insert_wq_barrier - insert a barrier work | |
112202d9 | 2410 | * @pwq: pwq to insert barrier into |
4690c4ab | 2411 | * @barr: wq_barrier to insert |
affee4b2 TH |
2412 | * @target: target work to attach @barr to |
2413 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2414 | * |
affee4b2 TH |
2415 | * @barr is linked to @target such that @barr is completed only after |
2416 | * @target finishes execution. Please note that the ordering | |
2417 | * guarantee is observed only with respect to @target and on the local | |
2418 | * cpu. | |
2419 | * | |
2420 | * Currently, a queued barrier can't be canceled. This is because | |
2421 | * try_to_grab_pending() can't determine whether the work to be | |
2422 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2423 | * flag of the previous work while there must be a valid next work | |
2424 | * after a work with LINKED flag set. | |
2425 | * | |
2426 | * Note that when @worker is non-NULL, @target may be modified | |
112202d9 | 2427 | * underneath us, so we can't reliably determine pwq from @target. |
4690c4ab TH |
2428 | * |
2429 | * CONTEXT: | |
d565ed63 | 2430 | * spin_lock_irq(pool->lock). |
4690c4ab | 2431 | */ |
112202d9 | 2432 | static void insert_wq_barrier(struct pool_workqueue *pwq, |
affee4b2 TH |
2433 | struct wq_barrier *barr, |
2434 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2435 | { |
affee4b2 TH |
2436 | struct list_head *head; |
2437 | unsigned int linked = 0; | |
2438 | ||
dc186ad7 | 2439 | /* |
d565ed63 | 2440 | * debugobject calls are safe here even with pool->lock locked |
dc186ad7 TG |
2441 | * as we know for sure that this will not trigger any of the |
2442 | * checks and call back into the fixup functions where we | |
2443 | * might deadlock. | |
2444 | */ | |
ca1cab37 | 2445 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2446 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2447 | init_completion(&barr->done); |
83c22520 | 2448 | |
affee4b2 TH |
2449 | /* |
2450 | * If @target is currently being executed, schedule the | |
2451 | * barrier to the worker; otherwise, put it after @target. | |
2452 | */ | |
2453 | if (worker) | |
2454 | head = worker->scheduled.next; | |
2455 | else { | |
2456 | unsigned long *bits = work_data_bits(target); | |
2457 | ||
2458 | head = target->entry.next; | |
2459 | /* there can already be other linked works, inherit and set */ | |
2460 | linked = *bits & WORK_STRUCT_LINKED; | |
2461 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2462 | } | |
2463 | ||
dc186ad7 | 2464 | debug_work_activate(&barr->work); |
112202d9 | 2465 | insert_work(pwq, &barr->work, head, |
affee4b2 | 2466 | work_color_to_flags(WORK_NO_COLOR) | linked); |
fc2e4d70 ON |
2467 | } |
2468 | ||
73f53c4a | 2469 | /** |
112202d9 | 2470 | * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing |
73f53c4a TH |
2471 | * @wq: workqueue being flushed |
2472 | * @flush_color: new flush color, < 0 for no-op | |
2473 | * @work_color: new work color, < 0 for no-op | |
2474 | * | |
112202d9 | 2475 | * Prepare pwqs for workqueue flushing. |
73f53c4a | 2476 | * |
112202d9 TH |
2477 | * If @flush_color is non-negative, flush_color on all pwqs should be |
2478 | * -1. If no pwq has in-flight commands at the specified color, all | |
2479 | * pwq->flush_color's stay at -1 and %false is returned. If any pwq | |
2480 | * has in flight commands, its pwq->flush_color is set to | |
2481 | * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq | |
73f53c4a TH |
2482 | * wakeup logic is armed and %true is returned. |
2483 | * | |
2484 | * The caller should have initialized @wq->first_flusher prior to | |
2485 | * calling this function with non-negative @flush_color. If | |
2486 | * @flush_color is negative, no flush color update is done and %false | |
2487 | * is returned. | |
2488 | * | |
112202d9 | 2489 | * If @work_color is non-negative, all pwqs should have the same |
73f53c4a TH |
2490 | * work_color which is previous to @work_color and all will be |
2491 | * advanced to @work_color. | |
2492 | * | |
2493 | * CONTEXT: | |
2494 | * mutex_lock(wq->flush_mutex). | |
2495 | * | |
2496 | * RETURNS: | |
2497 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2498 | * otherwise. | |
2499 | */ | |
112202d9 | 2500 | static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq, |
73f53c4a | 2501 | int flush_color, int work_color) |
1da177e4 | 2502 | { |
73f53c4a TH |
2503 | bool wait = false; |
2504 | unsigned int cpu; | |
1da177e4 | 2505 | |
73f53c4a | 2506 | if (flush_color >= 0) { |
6183c009 | 2507 | WARN_ON_ONCE(atomic_read(&wq->nr_pwqs_to_flush)); |
112202d9 | 2508 | atomic_set(&wq->nr_pwqs_to_flush, 1); |
1da177e4 | 2509 | } |
2355b70f | 2510 | |
112202d9 TH |
2511 | for_each_pwq_cpu(cpu, wq) { |
2512 | struct pool_workqueue *pwq = get_pwq(cpu, wq); | |
2513 | struct worker_pool *pool = pwq->pool; | |
fc2e4d70 | 2514 | |
d565ed63 | 2515 | spin_lock_irq(&pool->lock); |
83c22520 | 2516 | |
73f53c4a | 2517 | if (flush_color >= 0) { |
6183c009 | 2518 | WARN_ON_ONCE(pwq->flush_color != -1); |
fc2e4d70 | 2519 | |
112202d9 TH |
2520 | if (pwq->nr_in_flight[flush_color]) { |
2521 | pwq->flush_color = flush_color; | |
2522 | atomic_inc(&wq->nr_pwqs_to_flush); | |
73f53c4a TH |
2523 | wait = true; |
2524 | } | |
2525 | } | |
1da177e4 | 2526 | |
73f53c4a | 2527 | if (work_color >= 0) { |
6183c009 | 2528 | WARN_ON_ONCE(work_color != work_next_color(pwq->work_color)); |
112202d9 | 2529 | pwq->work_color = work_color; |
73f53c4a | 2530 | } |
1da177e4 | 2531 | |
d565ed63 | 2532 | spin_unlock_irq(&pool->lock); |
1da177e4 | 2533 | } |
2355b70f | 2534 | |
112202d9 | 2535 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_pwqs_to_flush)) |
73f53c4a | 2536 | complete(&wq->first_flusher->done); |
14441960 | 2537 | |
73f53c4a | 2538 | return wait; |
1da177e4 LT |
2539 | } |
2540 | ||
0fcb78c2 | 2541 | /** |
1da177e4 | 2542 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2543 | * @wq: workqueue to flush |
1da177e4 LT |
2544 | * |
2545 | * Forces execution of the workqueue and blocks until its completion. | |
2546 | * This is typically used in driver shutdown handlers. | |
2547 | * | |
fc2e4d70 ON |
2548 | * We sleep until all works which were queued on entry have been handled, |
2549 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 2550 | */ |
7ad5b3a5 | 2551 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2552 | { |
73f53c4a TH |
2553 | struct wq_flusher this_flusher = { |
2554 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2555 | .flush_color = -1, | |
2556 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2557 | }; | |
2558 | int next_color; | |
1da177e4 | 2559 | |
3295f0ef IM |
2560 | lock_map_acquire(&wq->lockdep_map); |
2561 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
2562 | |
2563 | mutex_lock(&wq->flush_mutex); | |
2564 | ||
2565 | /* | |
2566 | * Start-to-wait phase | |
2567 | */ | |
2568 | next_color = work_next_color(wq->work_color); | |
2569 | ||
2570 | if (next_color != wq->flush_color) { | |
2571 | /* | |
2572 | * Color space is not full. The current work_color | |
2573 | * becomes our flush_color and work_color is advanced | |
2574 | * by one. | |
2575 | */ | |
6183c009 | 2576 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow)); |
73f53c4a TH |
2577 | this_flusher.flush_color = wq->work_color; |
2578 | wq->work_color = next_color; | |
2579 | ||
2580 | if (!wq->first_flusher) { | |
2581 | /* no flush in progress, become the first flusher */ | |
6183c009 | 2582 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); |
73f53c4a TH |
2583 | |
2584 | wq->first_flusher = &this_flusher; | |
2585 | ||
112202d9 | 2586 | if (!flush_workqueue_prep_pwqs(wq, wq->flush_color, |
73f53c4a TH |
2587 | wq->work_color)) { |
2588 | /* nothing to flush, done */ | |
2589 | wq->flush_color = next_color; | |
2590 | wq->first_flusher = NULL; | |
2591 | goto out_unlock; | |
2592 | } | |
2593 | } else { | |
2594 | /* wait in queue */ | |
6183c009 | 2595 | WARN_ON_ONCE(wq->flush_color == this_flusher.flush_color); |
73f53c4a | 2596 | list_add_tail(&this_flusher.list, &wq->flusher_queue); |
112202d9 | 2597 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2598 | } |
2599 | } else { | |
2600 | /* | |
2601 | * Oops, color space is full, wait on overflow queue. | |
2602 | * The next flush completion will assign us | |
2603 | * flush_color and transfer to flusher_queue. | |
2604 | */ | |
2605 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2606 | } | |
2607 | ||
2608 | mutex_unlock(&wq->flush_mutex); | |
2609 | ||
2610 | wait_for_completion(&this_flusher.done); | |
2611 | ||
2612 | /* | |
2613 | * Wake-up-and-cascade phase | |
2614 | * | |
2615 | * First flushers are responsible for cascading flushes and | |
2616 | * handling overflow. Non-first flushers can simply return. | |
2617 | */ | |
2618 | if (wq->first_flusher != &this_flusher) | |
2619 | return; | |
2620 | ||
2621 | mutex_lock(&wq->flush_mutex); | |
2622 | ||
4ce48b37 TH |
2623 | /* we might have raced, check again with mutex held */ |
2624 | if (wq->first_flusher != &this_flusher) | |
2625 | goto out_unlock; | |
2626 | ||
73f53c4a TH |
2627 | wq->first_flusher = NULL; |
2628 | ||
6183c009 TH |
2629 | WARN_ON_ONCE(!list_empty(&this_flusher.list)); |
2630 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); | |
73f53c4a TH |
2631 | |
2632 | while (true) { | |
2633 | struct wq_flusher *next, *tmp; | |
2634 | ||
2635 | /* complete all the flushers sharing the current flush color */ | |
2636 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2637 | if (next->flush_color != wq->flush_color) | |
2638 | break; | |
2639 | list_del_init(&next->list); | |
2640 | complete(&next->done); | |
2641 | } | |
2642 | ||
6183c009 TH |
2643 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow) && |
2644 | wq->flush_color != work_next_color(wq->work_color)); | |
73f53c4a TH |
2645 | |
2646 | /* this flush_color is finished, advance by one */ | |
2647 | wq->flush_color = work_next_color(wq->flush_color); | |
2648 | ||
2649 | /* one color has been freed, handle overflow queue */ | |
2650 | if (!list_empty(&wq->flusher_overflow)) { | |
2651 | /* | |
2652 | * Assign the same color to all overflowed | |
2653 | * flushers, advance work_color and append to | |
2654 | * flusher_queue. This is the start-to-wait | |
2655 | * phase for these overflowed flushers. | |
2656 | */ | |
2657 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2658 | tmp->flush_color = wq->work_color; | |
2659 | ||
2660 | wq->work_color = work_next_color(wq->work_color); | |
2661 | ||
2662 | list_splice_tail_init(&wq->flusher_overflow, | |
2663 | &wq->flusher_queue); | |
112202d9 | 2664 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2665 | } |
2666 | ||
2667 | if (list_empty(&wq->flusher_queue)) { | |
6183c009 | 2668 | WARN_ON_ONCE(wq->flush_color != wq->work_color); |
73f53c4a TH |
2669 | break; |
2670 | } | |
2671 | ||
2672 | /* | |
2673 | * Need to flush more colors. Make the next flusher | |
112202d9 | 2674 | * the new first flusher and arm pwqs. |
73f53c4a | 2675 | */ |
6183c009 TH |
2676 | WARN_ON_ONCE(wq->flush_color == wq->work_color); |
2677 | WARN_ON_ONCE(wq->flush_color != next->flush_color); | |
73f53c4a TH |
2678 | |
2679 | list_del_init(&next->list); | |
2680 | wq->first_flusher = next; | |
2681 | ||
112202d9 | 2682 | if (flush_workqueue_prep_pwqs(wq, wq->flush_color, -1)) |
73f53c4a TH |
2683 | break; |
2684 | ||
2685 | /* | |
2686 | * Meh... this color is already done, clear first | |
2687 | * flusher and repeat cascading. | |
2688 | */ | |
2689 | wq->first_flusher = NULL; | |
2690 | } | |
2691 | ||
2692 | out_unlock: | |
2693 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 2694 | } |
ae90dd5d | 2695 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2696 | |
9c5a2ba7 TH |
2697 | /** |
2698 | * drain_workqueue - drain a workqueue | |
2699 | * @wq: workqueue to drain | |
2700 | * | |
2701 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2702 | * only chain queueing is allowed. IOW, only currently pending or running | |
2703 | * work items on @wq can queue further work items on it. @wq is flushed | |
2704 | * repeatedly until it becomes empty. The number of flushing is detemined | |
2705 | * by the depth of chaining and should be relatively short. Whine if it | |
2706 | * takes too long. | |
2707 | */ | |
2708 | void drain_workqueue(struct workqueue_struct *wq) | |
2709 | { | |
2710 | unsigned int flush_cnt = 0; | |
2711 | unsigned int cpu; | |
2712 | ||
2713 | /* | |
2714 | * __queue_work() needs to test whether there are drainers, is much | |
2715 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
2716 | * Use WQ_DRAINING so that queue doesn't have to check nr_drainers. | |
2717 | */ | |
e98d5b16 | 2718 | spin_lock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2719 | if (!wq->nr_drainers++) |
2720 | wq->flags |= WQ_DRAINING; | |
e98d5b16 | 2721 | spin_unlock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2722 | reflush: |
2723 | flush_workqueue(wq); | |
2724 | ||
112202d9 TH |
2725 | for_each_pwq_cpu(cpu, wq) { |
2726 | struct pool_workqueue *pwq = get_pwq(cpu, wq); | |
fa2563e4 | 2727 | bool drained; |
9c5a2ba7 | 2728 | |
112202d9 TH |
2729 | spin_lock_irq(&pwq->pool->lock); |
2730 | drained = !pwq->nr_active && list_empty(&pwq->delayed_works); | |
2731 | spin_unlock_irq(&pwq->pool->lock); | |
fa2563e4 TT |
2732 | |
2733 | if (drained) | |
9c5a2ba7 TH |
2734 | continue; |
2735 | ||
2736 | if (++flush_cnt == 10 || | |
2737 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
044c782c VI |
2738 | pr_warn("workqueue %s: flush on destruction isn't complete after %u tries\n", |
2739 | wq->name, flush_cnt); | |
9c5a2ba7 TH |
2740 | goto reflush; |
2741 | } | |
2742 | ||
e98d5b16 | 2743 | spin_lock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2744 | if (!--wq->nr_drainers) |
2745 | wq->flags &= ~WQ_DRAINING; | |
e98d5b16 | 2746 | spin_unlock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2747 | } |
2748 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2749 | ||
606a5020 | 2750 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) |
db700897 | 2751 | { |
affee4b2 | 2752 | struct worker *worker = NULL; |
c9e7cf27 | 2753 | struct worker_pool *pool; |
112202d9 | 2754 | struct pool_workqueue *pwq; |
db700897 ON |
2755 | |
2756 | might_sleep(); | |
c9e7cf27 TH |
2757 | pool = get_work_pool(work); |
2758 | if (!pool) | |
baf59022 | 2759 | return false; |
db700897 | 2760 | |
d565ed63 | 2761 | spin_lock_irq(&pool->lock); |
0b3dae68 | 2762 | /* see the comment in try_to_grab_pending() with the same code */ |
112202d9 TH |
2763 | pwq = get_work_pwq(work); |
2764 | if (pwq) { | |
2765 | if (unlikely(pwq->pool != pool)) | |
4690c4ab | 2766 | goto already_gone; |
606a5020 | 2767 | } else { |
c9e7cf27 | 2768 | worker = find_worker_executing_work(pool, work); |
affee4b2 | 2769 | if (!worker) |
4690c4ab | 2770 | goto already_gone; |
112202d9 | 2771 | pwq = worker->current_pwq; |
606a5020 | 2772 | } |
db700897 | 2773 | |
112202d9 | 2774 | insert_wq_barrier(pwq, barr, work, worker); |
d565ed63 | 2775 | spin_unlock_irq(&pool->lock); |
7a22ad75 | 2776 | |
e159489b TH |
2777 | /* |
2778 | * If @max_active is 1 or rescuer is in use, flushing another work | |
2779 | * item on the same workqueue may lead to deadlock. Make sure the | |
2780 | * flusher is not running on the same workqueue by verifying write | |
2781 | * access. | |
2782 | */ | |
112202d9 TH |
2783 | if (pwq->wq->saved_max_active == 1 || pwq->wq->flags & WQ_RESCUER) |
2784 | lock_map_acquire(&pwq->wq->lockdep_map); | |
e159489b | 2785 | else |
112202d9 TH |
2786 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
2787 | lock_map_release(&pwq->wq->lockdep_map); | |
e159489b | 2788 | |
401a8d04 | 2789 | return true; |
4690c4ab | 2790 | already_gone: |
d565ed63 | 2791 | spin_unlock_irq(&pool->lock); |
401a8d04 | 2792 | return false; |
db700897 | 2793 | } |
baf59022 TH |
2794 | |
2795 | /** | |
2796 | * flush_work - wait for a work to finish executing the last queueing instance | |
2797 | * @work: the work to flush | |
2798 | * | |
606a5020 TH |
2799 | * Wait until @work has finished execution. @work is guaranteed to be idle |
2800 | * on return if it hasn't been requeued since flush started. | |
baf59022 TH |
2801 | * |
2802 | * RETURNS: | |
2803 | * %true if flush_work() waited for the work to finish execution, | |
2804 | * %false if it was already idle. | |
2805 | */ | |
2806 | bool flush_work(struct work_struct *work) | |
2807 | { | |
2808 | struct wq_barrier barr; | |
2809 | ||
0976dfc1 SB |
2810 | lock_map_acquire(&work->lockdep_map); |
2811 | lock_map_release(&work->lockdep_map); | |
2812 | ||
606a5020 | 2813 | if (start_flush_work(work, &barr)) { |
401a8d04 TH |
2814 | wait_for_completion(&barr.done); |
2815 | destroy_work_on_stack(&barr.work); | |
2816 | return true; | |
606a5020 | 2817 | } else { |
401a8d04 | 2818 | return false; |
6e84d644 | 2819 | } |
6e84d644 | 2820 | } |
606a5020 | 2821 | EXPORT_SYMBOL_GPL(flush_work); |
6e84d644 | 2822 | |
36e227d2 | 2823 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
1f1f642e | 2824 | { |
bbb68dfa | 2825 | unsigned long flags; |
1f1f642e ON |
2826 | int ret; |
2827 | ||
2828 | do { | |
bbb68dfa TH |
2829 | ret = try_to_grab_pending(work, is_dwork, &flags); |
2830 | /* | |
2831 | * If someone else is canceling, wait for the same event it | |
2832 | * would be waiting for before retrying. | |
2833 | */ | |
2834 | if (unlikely(ret == -ENOENT)) | |
606a5020 | 2835 | flush_work(work); |
1f1f642e ON |
2836 | } while (unlikely(ret < 0)); |
2837 | ||
bbb68dfa TH |
2838 | /* tell other tasks trying to grab @work to back off */ |
2839 | mark_work_canceling(work); | |
2840 | local_irq_restore(flags); | |
2841 | ||
606a5020 | 2842 | flush_work(work); |
7a22ad75 | 2843 | clear_work_data(work); |
1f1f642e ON |
2844 | return ret; |
2845 | } | |
2846 | ||
6e84d644 | 2847 | /** |
401a8d04 TH |
2848 | * cancel_work_sync - cancel a work and wait for it to finish |
2849 | * @work: the work to cancel | |
6e84d644 | 2850 | * |
401a8d04 TH |
2851 | * Cancel @work and wait for its execution to finish. This function |
2852 | * can be used even if the work re-queues itself or migrates to | |
2853 | * another workqueue. On return from this function, @work is | |
2854 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2855 | * |
401a8d04 TH |
2856 | * cancel_work_sync(&delayed_work->work) must not be used for |
2857 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2858 | * |
401a8d04 | 2859 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2860 | * queued can't be destroyed before this function returns. |
401a8d04 TH |
2861 | * |
2862 | * RETURNS: | |
2863 | * %true if @work was pending, %false otherwise. | |
6e84d644 | 2864 | */ |
401a8d04 | 2865 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2866 | { |
36e227d2 | 2867 | return __cancel_work_timer(work, false); |
b89deed3 | 2868 | } |
28e53bdd | 2869 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2870 | |
6e84d644 | 2871 | /** |
401a8d04 TH |
2872 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2873 | * @dwork: the delayed work to flush | |
6e84d644 | 2874 | * |
401a8d04 TH |
2875 | * Delayed timer is cancelled and the pending work is queued for |
2876 | * immediate execution. Like flush_work(), this function only | |
2877 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2878 | * |
401a8d04 TH |
2879 | * RETURNS: |
2880 | * %true if flush_work() waited for the work to finish execution, | |
2881 | * %false if it was already idle. | |
6e84d644 | 2882 | */ |
401a8d04 TH |
2883 | bool flush_delayed_work(struct delayed_work *dwork) |
2884 | { | |
8930caba | 2885 | local_irq_disable(); |
401a8d04 | 2886 | if (del_timer_sync(&dwork->timer)) |
60c057bc | 2887 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
8930caba | 2888 | local_irq_enable(); |
401a8d04 TH |
2889 | return flush_work(&dwork->work); |
2890 | } | |
2891 | EXPORT_SYMBOL(flush_delayed_work); | |
2892 | ||
09383498 | 2893 | /** |
57b30ae7 TH |
2894 | * cancel_delayed_work - cancel a delayed work |
2895 | * @dwork: delayed_work to cancel | |
09383498 | 2896 | * |
57b30ae7 TH |
2897 | * Kill off a pending delayed_work. Returns %true if @dwork was pending |
2898 | * and canceled; %false if wasn't pending. Note that the work callback | |
2899 | * function may still be running on return, unless it returns %true and the | |
2900 | * work doesn't re-arm itself. Explicitly flush or use | |
2901 | * cancel_delayed_work_sync() to wait on it. | |
09383498 | 2902 | * |
57b30ae7 | 2903 | * This function is safe to call from any context including IRQ handler. |
09383498 | 2904 | */ |
57b30ae7 | 2905 | bool cancel_delayed_work(struct delayed_work *dwork) |
09383498 | 2906 | { |
57b30ae7 TH |
2907 | unsigned long flags; |
2908 | int ret; | |
2909 | ||
2910 | do { | |
2911 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
2912 | } while (unlikely(ret == -EAGAIN)); | |
2913 | ||
2914 | if (unlikely(ret < 0)) | |
2915 | return false; | |
2916 | ||
7c3eed5c TH |
2917 | set_work_pool_and_clear_pending(&dwork->work, |
2918 | get_work_pool_id(&dwork->work)); | |
57b30ae7 | 2919 | local_irq_restore(flags); |
c0158ca6 | 2920 | return ret; |
09383498 | 2921 | } |
57b30ae7 | 2922 | EXPORT_SYMBOL(cancel_delayed_work); |
09383498 | 2923 | |
401a8d04 TH |
2924 | /** |
2925 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
2926 | * @dwork: the delayed work cancel | |
2927 | * | |
2928 | * This is cancel_work_sync() for delayed works. | |
2929 | * | |
2930 | * RETURNS: | |
2931 | * %true if @dwork was pending, %false otherwise. | |
2932 | */ | |
2933 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 2934 | { |
36e227d2 | 2935 | return __cancel_work_timer(&dwork->work, true); |
6e84d644 | 2936 | } |
f5a421a4 | 2937 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 2938 | |
0fcb78c2 | 2939 | /** |
c1a220e7 ZR |
2940 | * schedule_work_on - put work task on a specific cpu |
2941 | * @cpu: cpu to put the work task on | |
2942 | * @work: job to be done | |
2943 | * | |
2944 | * This puts a job on a specific cpu | |
2945 | */ | |
d4283e93 | 2946 | bool schedule_work_on(int cpu, struct work_struct *work) |
c1a220e7 | 2947 | { |
d320c038 | 2948 | return queue_work_on(cpu, system_wq, work); |
c1a220e7 ZR |
2949 | } |
2950 | EXPORT_SYMBOL(schedule_work_on); | |
2951 | ||
0fcb78c2 | 2952 | /** |
0fcb78c2 REB |
2953 | * schedule_work - put work task in global workqueue |
2954 | * @work: job to be done | |
0fcb78c2 | 2955 | * |
d4283e93 TH |
2956 | * Returns %false if @work was already on the kernel-global workqueue and |
2957 | * %true otherwise. | |
5b0f437d BVA |
2958 | * |
2959 | * This puts a job in the kernel-global workqueue if it was not already | |
2960 | * queued and leaves it in the same position on the kernel-global | |
2961 | * workqueue otherwise. | |
0fcb78c2 | 2962 | */ |
d4283e93 | 2963 | bool schedule_work(struct work_struct *work) |
1da177e4 | 2964 | { |
d320c038 | 2965 | return queue_work(system_wq, work); |
1da177e4 | 2966 | } |
ae90dd5d | 2967 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 2968 | |
0fcb78c2 REB |
2969 | /** |
2970 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
2971 | * @cpu: cpu to use | |
52bad64d | 2972 | * @dwork: job to be done |
0fcb78c2 REB |
2973 | * @delay: number of jiffies to wait |
2974 | * | |
2975 | * After waiting for a given time this puts a job in the kernel-global | |
2976 | * workqueue on the specified CPU. | |
2977 | */ | |
d4283e93 TH |
2978 | bool schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
2979 | unsigned long delay) | |
1da177e4 | 2980 | { |
d320c038 | 2981 | return queue_delayed_work_on(cpu, system_wq, dwork, delay); |
1da177e4 | 2982 | } |
ae90dd5d | 2983 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 2984 | |
0fcb78c2 REB |
2985 | /** |
2986 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
2987 | * @dwork: job to be done |
2988 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
2989 | * |
2990 | * After waiting for a given time this puts a job in the kernel-global | |
2991 | * workqueue. | |
2992 | */ | |
d4283e93 | 2993 | bool schedule_delayed_work(struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 2994 | { |
d320c038 | 2995 | return queue_delayed_work(system_wq, dwork, delay); |
1da177e4 | 2996 | } |
ae90dd5d | 2997 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 2998 | |
b6136773 | 2999 | /** |
31ddd871 | 3000 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 3001 | * @func: the function to call |
b6136773 | 3002 | * |
31ddd871 TH |
3003 | * schedule_on_each_cpu() executes @func on each online CPU using the |
3004 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 3005 | * schedule_on_each_cpu() is very slow. |
31ddd871 TH |
3006 | * |
3007 | * RETURNS: | |
3008 | * 0 on success, -errno on failure. | |
b6136773 | 3009 | */ |
65f27f38 | 3010 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
3011 | { |
3012 | int cpu; | |
38f51568 | 3013 | struct work_struct __percpu *works; |
15316ba8 | 3014 | |
b6136773 AM |
3015 | works = alloc_percpu(struct work_struct); |
3016 | if (!works) | |
15316ba8 | 3017 | return -ENOMEM; |
b6136773 | 3018 | |
93981800 TH |
3019 | get_online_cpus(); |
3020 | ||
15316ba8 | 3021 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
3022 | struct work_struct *work = per_cpu_ptr(works, cpu); |
3023 | ||
3024 | INIT_WORK(work, func); | |
b71ab8c2 | 3025 | schedule_work_on(cpu, work); |
65a64464 | 3026 | } |
93981800 TH |
3027 | |
3028 | for_each_online_cpu(cpu) | |
3029 | flush_work(per_cpu_ptr(works, cpu)); | |
3030 | ||
95402b38 | 3031 | put_online_cpus(); |
b6136773 | 3032 | free_percpu(works); |
15316ba8 CL |
3033 | return 0; |
3034 | } | |
3035 | ||
eef6a7d5 AS |
3036 | /** |
3037 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
3038 | * | |
3039 | * Forces execution of the kernel-global workqueue and blocks until its | |
3040 | * completion. | |
3041 | * | |
3042 | * Think twice before calling this function! It's very easy to get into | |
3043 | * trouble if you don't take great care. Either of the following situations | |
3044 | * will lead to deadlock: | |
3045 | * | |
3046 | * One of the work items currently on the workqueue needs to acquire | |
3047 | * a lock held by your code or its caller. | |
3048 | * | |
3049 | * Your code is running in the context of a work routine. | |
3050 | * | |
3051 | * They will be detected by lockdep when they occur, but the first might not | |
3052 | * occur very often. It depends on what work items are on the workqueue and | |
3053 | * what locks they need, which you have no control over. | |
3054 | * | |
3055 | * In most situations flushing the entire workqueue is overkill; you merely | |
3056 | * need to know that a particular work item isn't queued and isn't running. | |
3057 | * In such cases you should use cancel_delayed_work_sync() or | |
3058 | * cancel_work_sync() instead. | |
3059 | */ | |
1da177e4 LT |
3060 | void flush_scheduled_work(void) |
3061 | { | |
d320c038 | 3062 | flush_workqueue(system_wq); |
1da177e4 | 3063 | } |
ae90dd5d | 3064 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 3065 | |
1fa44eca JB |
3066 | /** |
3067 | * execute_in_process_context - reliably execute the routine with user context | |
3068 | * @fn: the function to execute | |
1fa44eca JB |
3069 | * @ew: guaranteed storage for the execute work structure (must |
3070 | * be available when the work executes) | |
3071 | * | |
3072 | * Executes the function immediately if process context is available, | |
3073 | * otherwise schedules the function for delayed execution. | |
3074 | * | |
3075 | * Returns: 0 - function was executed | |
3076 | * 1 - function was scheduled for execution | |
3077 | */ | |
65f27f38 | 3078 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
3079 | { |
3080 | if (!in_interrupt()) { | |
65f27f38 | 3081 | fn(&ew->work); |
1fa44eca JB |
3082 | return 0; |
3083 | } | |
3084 | ||
65f27f38 | 3085 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
3086 | schedule_work(&ew->work); |
3087 | ||
3088 | return 1; | |
3089 | } | |
3090 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
3091 | ||
1da177e4 LT |
3092 | int keventd_up(void) |
3093 | { | |
d320c038 | 3094 | return system_wq != NULL; |
1da177e4 LT |
3095 | } |
3096 | ||
112202d9 | 3097 | static int alloc_pwqs(struct workqueue_struct *wq) |
0f900049 | 3098 | { |
65a64464 | 3099 | /* |
112202d9 | 3100 | * pwqs are forced aligned according to WORK_STRUCT_FLAG_BITS. |
0f900049 TH |
3101 | * Make sure that the alignment isn't lower than that of |
3102 | * unsigned long long. | |
65a64464 | 3103 | */ |
112202d9 | 3104 | const size_t size = sizeof(struct pool_workqueue); |
0f900049 TH |
3105 | const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS, |
3106 | __alignof__(unsigned long long)); | |
65a64464 | 3107 | |
e06ffa1e | 3108 | if (!(wq->flags & WQ_UNBOUND)) |
112202d9 | 3109 | wq->pool_wq.pcpu = __alloc_percpu(size, align); |
931ac77e | 3110 | else { |
f3421797 TH |
3111 | void *ptr; |
3112 | ||
3113 | /* | |
112202d9 | 3114 | * Allocate enough room to align pwq and put an extra |
f3421797 TH |
3115 | * pointer at the end pointing back to the originally |
3116 | * allocated pointer which will be used for free. | |
3117 | */ | |
3118 | ptr = kzalloc(size + align + sizeof(void *), GFP_KERNEL); | |
3119 | if (ptr) { | |
112202d9 TH |
3120 | wq->pool_wq.single = PTR_ALIGN(ptr, align); |
3121 | *(void **)(wq->pool_wq.single + 1) = ptr; | |
f3421797 | 3122 | } |
bdbc5dd7 | 3123 | } |
f3421797 | 3124 | |
0415b00d | 3125 | /* just in case, make sure it's actually aligned */ |
112202d9 TH |
3126 | BUG_ON(!IS_ALIGNED(wq->pool_wq.v, align)); |
3127 | return wq->pool_wq.v ? 0 : -ENOMEM; | |
0f900049 TH |
3128 | } |
3129 | ||
112202d9 | 3130 | static void free_pwqs(struct workqueue_struct *wq) |
0f900049 | 3131 | { |
e06ffa1e | 3132 | if (!(wq->flags & WQ_UNBOUND)) |
112202d9 TH |
3133 | free_percpu(wq->pool_wq.pcpu); |
3134 | else if (wq->pool_wq.single) { | |
3135 | /* the pointer to free is stored right after the pwq */ | |
3136 | kfree(*(void **)(wq->pool_wq.single + 1)); | |
f3421797 | 3137 | } |
0f900049 TH |
3138 | } |
3139 | ||
f3421797 TH |
3140 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
3141 | const char *name) | |
b71ab8c2 | 3142 | { |
f3421797 TH |
3143 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
3144 | ||
3145 | if (max_active < 1 || max_active > lim) | |
044c782c VI |
3146 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
3147 | max_active, name, 1, lim); | |
b71ab8c2 | 3148 | |
f3421797 | 3149 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
3150 | } |
3151 | ||
b196be89 | 3152 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
d320c038 TH |
3153 | unsigned int flags, |
3154 | int max_active, | |
3155 | struct lock_class_key *key, | |
b196be89 | 3156 | const char *lock_name, ...) |
1da177e4 | 3157 | { |
b196be89 | 3158 | va_list args, args1; |
1da177e4 | 3159 | struct workqueue_struct *wq; |
c34056a3 | 3160 | unsigned int cpu; |
b196be89 TH |
3161 | size_t namelen; |
3162 | ||
3163 | /* determine namelen, allocate wq and format name */ | |
3164 | va_start(args, lock_name); | |
3165 | va_copy(args1, args); | |
3166 | namelen = vsnprintf(NULL, 0, fmt, args) + 1; | |
3167 | ||
3168 | wq = kzalloc(sizeof(*wq) + namelen, GFP_KERNEL); | |
3169 | if (!wq) | |
3170 | goto err; | |
3171 | ||
3172 | vsnprintf(wq->name, namelen, fmt, args1); | |
3173 | va_end(args); | |
3174 | va_end(args1); | |
1da177e4 | 3175 | |
6370a6ad TH |
3176 | /* |
3177 | * Workqueues which may be used during memory reclaim should | |
3178 | * have a rescuer to guarantee forward progress. | |
3179 | */ | |
3180 | if (flags & WQ_MEM_RECLAIM) | |
3181 | flags |= WQ_RESCUER; | |
3182 | ||
d320c038 | 3183 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 3184 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 3185 | |
b196be89 | 3186 | /* init wq */ |
97e37d7b | 3187 | wq->flags = flags; |
a0a1a5fd | 3188 | wq->saved_max_active = max_active; |
73f53c4a | 3189 | mutex_init(&wq->flush_mutex); |
112202d9 | 3190 | atomic_set(&wq->nr_pwqs_to_flush, 0); |
73f53c4a TH |
3191 | INIT_LIST_HEAD(&wq->flusher_queue); |
3192 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
502ca9d8 | 3193 | |
eb13ba87 | 3194 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 3195 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 3196 | |
112202d9 | 3197 | if (alloc_pwqs(wq) < 0) |
bdbc5dd7 TH |
3198 | goto err; |
3199 | ||
112202d9 TH |
3200 | for_each_pwq_cpu(cpu, wq) { |
3201 | struct pool_workqueue *pwq = get_pwq(cpu, wq); | |
1537663f | 3202 | |
112202d9 TH |
3203 | BUG_ON((unsigned long)pwq & WORK_STRUCT_FLAG_MASK); |
3204 | pwq->pool = get_std_worker_pool(cpu, flags & WQ_HIGHPRI); | |
3205 | pwq->wq = wq; | |
3206 | pwq->flush_color = -1; | |
3207 | pwq->max_active = max_active; | |
3208 | INIT_LIST_HEAD(&pwq->delayed_works); | |
e22bee78 | 3209 | } |
1537663f | 3210 | |
e22bee78 TH |
3211 | if (flags & WQ_RESCUER) { |
3212 | struct worker *rescuer; | |
3213 | ||
f2e005aa | 3214 | if (!alloc_mayday_mask(&wq->mayday_mask, GFP_KERNEL)) |
e22bee78 TH |
3215 | goto err; |
3216 | ||
3217 | wq->rescuer = rescuer = alloc_worker(); | |
3218 | if (!rescuer) | |
3219 | goto err; | |
3220 | ||
111c225a TH |
3221 | rescuer->rescue_wq = wq; |
3222 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", | |
b196be89 | 3223 | wq->name); |
e22bee78 TH |
3224 | if (IS_ERR(rescuer->task)) |
3225 | goto err; | |
3226 | ||
e22bee78 TH |
3227 | rescuer->task->flags |= PF_THREAD_BOUND; |
3228 | wake_up_process(rescuer->task); | |
3af24433 ON |
3229 | } |
3230 | ||
a0a1a5fd TH |
3231 | /* |
3232 | * workqueue_lock protects global freeze state and workqueues | |
3233 | * list. Grab it, set max_active accordingly and add the new | |
3234 | * workqueue to workqueues list. | |
3235 | */ | |
e98d5b16 | 3236 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3237 | |
58a69cb4 | 3238 | if (workqueue_freezing && wq->flags & WQ_FREEZABLE) |
112202d9 TH |
3239 | for_each_pwq_cpu(cpu, wq) |
3240 | get_pwq(cpu, wq)->max_active = 0; | |
a0a1a5fd | 3241 | |
1537663f | 3242 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 3243 | |
e98d5b16 | 3244 | spin_unlock_irq(&workqueue_lock); |
1537663f | 3245 | |
3af24433 | 3246 | return wq; |
4690c4ab TH |
3247 | err: |
3248 | if (wq) { | |
112202d9 | 3249 | free_pwqs(wq); |
f2e005aa | 3250 | free_mayday_mask(wq->mayday_mask); |
e22bee78 | 3251 | kfree(wq->rescuer); |
4690c4ab TH |
3252 | kfree(wq); |
3253 | } | |
3254 | return NULL; | |
3af24433 | 3255 | } |
d320c038 | 3256 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 3257 | |
3af24433 ON |
3258 | /** |
3259 | * destroy_workqueue - safely terminate a workqueue | |
3260 | * @wq: target workqueue | |
3261 | * | |
3262 | * Safely destroy a workqueue. All work currently pending will be done first. | |
3263 | */ | |
3264 | void destroy_workqueue(struct workqueue_struct *wq) | |
3265 | { | |
c8e55f36 | 3266 | unsigned int cpu; |
3af24433 | 3267 | |
9c5a2ba7 TH |
3268 | /* drain it before proceeding with destruction */ |
3269 | drain_workqueue(wq); | |
c8efcc25 | 3270 | |
6183c009 TH |
3271 | /* sanity checks */ |
3272 | for_each_pwq_cpu(cpu, wq) { | |
3273 | struct pool_workqueue *pwq = get_pwq(cpu, wq); | |
3274 | int i; | |
3275 | ||
3276 | for (i = 0; i < WORK_NR_COLORS; i++) | |
3277 | if (WARN_ON(pwq->nr_in_flight[i])) | |
3278 | return; | |
3279 | if (WARN_ON(pwq->nr_active) || | |
3280 | WARN_ON(!list_empty(&pwq->delayed_works))) | |
3281 | return; | |
3282 | } | |
3283 | ||
a0a1a5fd TH |
3284 | /* |
3285 | * wq list is used to freeze wq, remove from list after | |
3286 | * flushing is complete in case freeze races us. | |
3287 | */ | |
e98d5b16 | 3288 | spin_lock_irq(&workqueue_lock); |
b1f4ec17 | 3289 | list_del(&wq->list); |
e98d5b16 | 3290 | spin_unlock_irq(&workqueue_lock); |
3af24433 | 3291 | |
e22bee78 TH |
3292 | if (wq->flags & WQ_RESCUER) { |
3293 | kthread_stop(wq->rescuer->task); | |
f2e005aa | 3294 | free_mayday_mask(wq->mayday_mask); |
8d9df9f0 | 3295 | kfree(wq->rescuer); |
e22bee78 TH |
3296 | } |
3297 | ||
112202d9 | 3298 | free_pwqs(wq); |
3af24433 ON |
3299 | kfree(wq); |
3300 | } | |
3301 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
3302 | ||
9f4bd4cd | 3303 | /** |
112202d9 TH |
3304 | * pwq_set_max_active - adjust max_active of a pwq |
3305 | * @pwq: target pool_workqueue | |
9f4bd4cd LJ |
3306 | * @max_active: new max_active value. |
3307 | * | |
112202d9 | 3308 | * Set @pwq->max_active to @max_active and activate delayed works if |
9f4bd4cd LJ |
3309 | * increased. |
3310 | * | |
3311 | * CONTEXT: | |
d565ed63 | 3312 | * spin_lock_irq(pool->lock). |
9f4bd4cd | 3313 | */ |
112202d9 | 3314 | static void pwq_set_max_active(struct pool_workqueue *pwq, int max_active) |
9f4bd4cd | 3315 | { |
112202d9 | 3316 | pwq->max_active = max_active; |
9f4bd4cd | 3317 | |
112202d9 TH |
3318 | while (!list_empty(&pwq->delayed_works) && |
3319 | pwq->nr_active < pwq->max_active) | |
3320 | pwq_activate_first_delayed(pwq); | |
9f4bd4cd LJ |
3321 | } |
3322 | ||
dcd989cb TH |
3323 | /** |
3324 | * workqueue_set_max_active - adjust max_active of a workqueue | |
3325 | * @wq: target workqueue | |
3326 | * @max_active: new max_active value. | |
3327 | * | |
3328 | * Set max_active of @wq to @max_active. | |
3329 | * | |
3330 | * CONTEXT: | |
3331 | * Don't call from IRQ context. | |
3332 | */ | |
3333 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
3334 | { | |
3335 | unsigned int cpu; | |
3336 | ||
f3421797 | 3337 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb | 3338 | |
e98d5b16 | 3339 | spin_lock_irq(&workqueue_lock); |
dcd989cb TH |
3340 | |
3341 | wq->saved_max_active = max_active; | |
3342 | ||
112202d9 TH |
3343 | for_each_pwq_cpu(cpu, wq) { |
3344 | struct pool_workqueue *pwq = get_pwq(cpu, wq); | |
3345 | struct worker_pool *pool = pwq->pool; | |
dcd989cb | 3346 | |
e98d5b16 | 3347 | spin_lock(&pool->lock); |
dcd989cb | 3348 | |
58a69cb4 | 3349 | if (!(wq->flags & WQ_FREEZABLE) || |
35b6bb63 | 3350 | !(pool->flags & POOL_FREEZING)) |
112202d9 | 3351 | pwq_set_max_active(pwq, max_active); |
9bfb1839 | 3352 | |
e98d5b16 | 3353 | spin_unlock(&pool->lock); |
65a64464 | 3354 | } |
93981800 | 3355 | |
e98d5b16 | 3356 | spin_unlock_irq(&workqueue_lock); |
15316ba8 | 3357 | } |
dcd989cb | 3358 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 3359 | |
eef6a7d5 | 3360 | /** |
dcd989cb TH |
3361 | * workqueue_congested - test whether a workqueue is congested |
3362 | * @cpu: CPU in question | |
3363 | * @wq: target workqueue | |
eef6a7d5 | 3364 | * |
dcd989cb TH |
3365 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
3366 | * no synchronization around this function and the test result is | |
3367 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 3368 | * |
dcd989cb TH |
3369 | * RETURNS: |
3370 | * %true if congested, %false otherwise. | |
eef6a7d5 | 3371 | */ |
dcd989cb | 3372 | bool workqueue_congested(unsigned int cpu, struct workqueue_struct *wq) |
1da177e4 | 3373 | { |
112202d9 | 3374 | struct pool_workqueue *pwq = get_pwq(cpu, wq); |
dcd989cb | 3375 | |
112202d9 | 3376 | return !list_empty(&pwq->delayed_works); |
1da177e4 | 3377 | } |
dcd989cb | 3378 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 3379 | |
dcd989cb TH |
3380 | /** |
3381 | * work_busy - test whether a work is currently pending or running | |
3382 | * @work: the work to be tested | |
3383 | * | |
3384 | * Test whether @work is currently pending or running. There is no | |
3385 | * synchronization around this function and the test result is | |
3386 | * unreliable and only useful as advisory hints or for debugging. | |
dcd989cb TH |
3387 | * |
3388 | * RETURNS: | |
3389 | * OR'd bitmask of WORK_BUSY_* bits. | |
3390 | */ | |
3391 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 3392 | { |
c9e7cf27 | 3393 | struct worker_pool *pool = get_work_pool(work); |
dcd989cb TH |
3394 | unsigned long flags; |
3395 | unsigned int ret = 0; | |
1da177e4 | 3396 | |
dcd989cb TH |
3397 | if (work_pending(work)) |
3398 | ret |= WORK_BUSY_PENDING; | |
1da177e4 | 3399 | |
038366c5 LJ |
3400 | if (pool) { |
3401 | spin_lock_irqsave(&pool->lock, flags); | |
3402 | if (find_worker_executing_work(pool, work)) | |
3403 | ret |= WORK_BUSY_RUNNING; | |
3404 | spin_unlock_irqrestore(&pool->lock, flags); | |
3405 | } | |
1da177e4 | 3406 | |
dcd989cb | 3407 | return ret; |
1da177e4 | 3408 | } |
dcd989cb | 3409 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 3410 | |
db7bccf4 TH |
3411 | /* |
3412 | * CPU hotplug. | |
3413 | * | |
e22bee78 | 3414 | * There are two challenges in supporting CPU hotplug. Firstly, there |
112202d9 | 3415 | * are a lot of assumptions on strong associations among work, pwq and |
706026c2 | 3416 | * pool which make migrating pending and scheduled works very |
e22bee78 | 3417 | * difficult to implement without impacting hot paths. Secondly, |
94cf58bb | 3418 | * worker pools serve mix of short, long and very long running works making |
e22bee78 TH |
3419 | * blocked draining impractical. |
3420 | * | |
24647570 | 3421 | * This is solved by allowing the pools to be disassociated from the CPU |
628c78e7 TH |
3422 | * running as an unbound one and allowing it to be reattached later if the |
3423 | * cpu comes back online. | |
db7bccf4 | 3424 | */ |
1da177e4 | 3425 | |
706026c2 | 3426 | static void wq_unbind_fn(struct work_struct *work) |
3af24433 | 3427 | { |
38db41d9 | 3428 | int cpu = smp_processor_id(); |
4ce62e9e | 3429 | struct worker_pool *pool; |
db7bccf4 | 3430 | struct worker *worker; |
db7bccf4 | 3431 | int i; |
3af24433 | 3432 | |
38db41d9 | 3433 | for_each_std_worker_pool(pool, cpu) { |
6183c009 | 3434 | WARN_ON_ONCE(cpu != smp_processor_id()); |
db7bccf4 | 3435 | |
94cf58bb TH |
3436 | mutex_lock(&pool->assoc_mutex); |
3437 | spin_lock_irq(&pool->lock); | |
3af24433 | 3438 | |
94cf58bb TH |
3439 | /* |
3440 | * We've claimed all manager positions. Make all workers | |
3441 | * unbound and set DISASSOCIATED. Before this, all workers | |
3442 | * except for the ones which are still executing works from | |
3443 | * before the last CPU down must be on the cpu. After | |
3444 | * this, they may become diasporas. | |
3445 | */ | |
4ce62e9e | 3446 | list_for_each_entry(worker, &pool->idle_list, entry) |
403c821d | 3447 | worker->flags |= WORKER_UNBOUND; |
3af24433 | 3448 | |
b67bfe0d | 3449 | for_each_busy_worker(worker, i, pool) |
c9e7cf27 | 3450 | worker->flags |= WORKER_UNBOUND; |
06ba38a9 | 3451 | |
24647570 | 3452 | pool->flags |= POOL_DISASSOCIATED; |
f2d5a0ee | 3453 | |
94cf58bb TH |
3454 | spin_unlock_irq(&pool->lock); |
3455 | mutex_unlock(&pool->assoc_mutex); | |
3456 | } | |
628c78e7 | 3457 | |
e22bee78 | 3458 | /* |
403c821d | 3459 | * Call schedule() so that we cross rq->lock and thus can guarantee |
628c78e7 TH |
3460 | * sched callbacks see the %WORKER_UNBOUND flag. This is necessary |
3461 | * as scheduler callbacks may be invoked from other cpus. | |
e22bee78 | 3462 | */ |
e22bee78 | 3463 | schedule(); |
06ba38a9 | 3464 | |
e22bee78 | 3465 | /* |
628c78e7 TH |
3466 | * Sched callbacks are disabled now. Zap nr_running. After this, |
3467 | * nr_running stays zero and need_more_worker() and keep_working() | |
38db41d9 TH |
3468 | * are always true as long as the worklist is not empty. Pools on |
3469 | * @cpu now behave as unbound (in terms of concurrency management) | |
3470 | * pools which are served by workers tied to the CPU. | |
628c78e7 TH |
3471 | * |
3472 | * On return from this function, the current worker would trigger | |
3473 | * unbound chain execution of pending work items if other workers | |
3474 | * didn't already. | |
e22bee78 | 3475 | */ |
38db41d9 | 3476 | for_each_std_worker_pool(pool, cpu) |
e19e397a | 3477 | atomic_set(&pool->nr_running, 0); |
3af24433 | 3478 | } |
3af24433 | 3479 | |
8db25e78 TH |
3480 | /* |
3481 | * Workqueues should be brought up before normal priority CPU notifiers. | |
3482 | * This will be registered high priority CPU notifier. | |
3483 | */ | |
9fdf9b73 | 3484 | static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb, |
8db25e78 TH |
3485 | unsigned long action, |
3486 | void *hcpu) | |
3af24433 ON |
3487 | { |
3488 | unsigned int cpu = (unsigned long)hcpu; | |
4ce62e9e | 3489 | struct worker_pool *pool; |
3ce63377 | 3490 | |
8db25e78 | 3491 | switch (action & ~CPU_TASKS_FROZEN) { |
3af24433 | 3492 | case CPU_UP_PREPARE: |
38db41d9 | 3493 | for_each_std_worker_pool(pool, cpu) { |
3ce63377 TH |
3494 | struct worker *worker; |
3495 | ||
3496 | if (pool->nr_workers) | |
3497 | continue; | |
3498 | ||
3499 | worker = create_worker(pool); | |
3500 | if (!worker) | |
3501 | return NOTIFY_BAD; | |
3502 | ||
d565ed63 | 3503 | spin_lock_irq(&pool->lock); |
3ce63377 | 3504 | start_worker(worker); |
d565ed63 | 3505 | spin_unlock_irq(&pool->lock); |
3af24433 | 3506 | } |
8db25e78 | 3507 | break; |
3af24433 | 3508 | |
db7bccf4 TH |
3509 | case CPU_DOWN_FAILED: |
3510 | case CPU_ONLINE: | |
38db41d9 | 3511 | for_each_std_worker_pool(pool, cpu) { |
94cf58bb TH |
3512 | mutex_lock(&pool->assoc_mutex); |
3513 | spin_lock_irq(&pool->lock); | |
3514 | ||
24647570 | 3515 | pool->flags &= ~POOL_DISASSOCIATED; |
94cf58bb TH |
3516 | rebind_workers(pool); |
3517 | ||
3518 | spin_unlock_irq(&pool->lock); | |
3519 | mutex_unlock(&pool->assoc_mutex); | |
3520 | } | |
db7bccf4 | 3521 | break; |
00dfcaf7 | 3522 | } |
65758202 TH |
3523 | return NOTIFY_OK; |
3524 | } | |
3525 | ||
3526 | /* | |
3527 | * Workqueues should be brought down after normal priority CPU notifiers. | |
3528 | * This will be registered as low priority CPU notifier. | |
3529 | */ | |
9fdf9b73 | 3530 | static int __cpuinit workqueue_cpu_down_callback(struct notifier_block *nfb, |
65758202 TH |
3531 | unsigned long action, |
3532 | void *hcpu) | |
3533 | { | |
8db25e78 TH |
3534 | unsigned int cpu = (unsigned long)hcpu; |
3535 | struct work_struct unbind_work; | |
3536 | ||
65758202 TH |
3537 | switch (action & ~CPU_TASKS_FROZEN) { |
3538 | case CPU_DOWN_PREPARE: | |
8db25e78 | 3539 | /* unbinding should happen on the local CPU */ |
706026c2 | 3540 | INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn); |
7635d2fd | 3541 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
8db25e78 TH |
3542 | flush_work(&unbind_work); |
3543 | break; | |
65758202 TH |
3544 | } |
3545 | return NOTIFY_OK; | |
3546 | } | |
3547 | ||
2d3854a3 | 3548 | #ifdef CONFIG_SMP |
8ccad40d | 3549 | |
2d3854a3 | 3550 | struct work_for_cpu { |
ed48ece2 | 3551 | struct work_struct work; |
2d3854a3 RR |
3552 | long (*fn)(void *); |
3553 | void *arg; | |
3554 | long ret; | |
3555 | }; | |
3556 | ||
ed48ece2 | 3557 | static void work_for_cpu_fn(struct work_struct *work) |
2d3854a3 | 3558 | { |
ed48ece2 TH |
3559 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
3560 | ||
2d3854a3 RR |
3561 | wfc->ret = wfc->fn(wfc->arg); |
3562 | } | |
3563 | ||
3564 | /** | |
3565 | * work_on_cpu - run a function in user context on a particular cpu | |
3566 | * @cpu: the cpu to run on | |
3567 | * @fn: the function to run | |
3568 | * @arg: the function arg | |
3569 | * | |
31ad9081 RR |
3570 | * This will return the value @fn returns. |
3571 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 3572 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 RR |
3573 | */ |
3574 | long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) | |
3575 | { | |
ed48ece2 | 3576 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; |
6b44003e | 3577 | |
ed48ece2 TH |
3578 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
3579 | schedule_work_on(cpu, &wfc.work); | |
3580 | flush_work(&wfc.work); | |
2d3854a3 RR |
3581 | return wfc.ret; |
3582 | } | |
3583 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
3584 | #endif /* CONFIG_SMP */ | |
3585 | ||
a0a1a5fd TH |
3586 | #ifdef CONFIG_FREEZER |
3587 | ||
3588 | /** | |
3589 | * freeze_workqueues_begin - begin freezing workqueues | |
3590 | * | |
58a69cb4 TH |
3591 | * Start freezing workqueues. After this function returns, all freezable |
3592 | * workqueues will queue new works to their frozen_works list instead of | |
706026c2 | 3593 | * pool->worklist. |
a0a1a5fd TH |
3594 | * |
3595 | * CONTEXT: | |
d565ed63 | 3596 | * Grabs and releases workqueue_lock and pool->lock's. |
a0a1a5fd TH |
3597 | */ |
3598 | void freeze_workqueues_begin(void) | |
3599 | { | |
a0a1a5fd TH |
3600 | unsigned int cpu; |
3601 | ||
e98d5b16 | 3602 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3603 | |
6183c009 | 3604 | WARN_ON_ONCE(workqueue_freezing); |
a0a1a5fd TH |
3605 | workqueue_freezing = true; |
3606 | ||
706026c2 | 3607 | for_each_wq_cpu(cpu) { |
35b6bb63 | 3608 | struct worker_pool *pool; |
bdbc5dd7 | 3609 | struct workqueue_struct *wq; |
8b03ae3c | 3610 | |
38db41d9 | 3611 | for_each_std_worker_pool(pool, cpu) { |
e98d5b16 | 3612 | spin_lock(&pool->lock); |
d565ed63 | 3613 | |
35b6bb63 TH |
3614 | WARN_ON_ONCE(pool->flags & POOL_FREEZING); |
3615 | pool->flags |= POOL_FREEZING; | |
db7bccf4 | 3616 | |
a1056305 | 3617 | list_for_each_entry(wq, &workqueues, list) { |
112202d9 | 3618 | struct pool_workqueue *pwq = get_pwq(cpu, wq); |
a0a1a5fd | 3619 | |
112202d9 | 3620 | if (pwq && pwq->pool == pool && |
a1056305 | 3621 | (wq->flags & WQ_FREEZABLE)) |
112202d9 | 3622 | pwq->max_active = 0; |
a1056305 | 3623 | } |
8b03ae3c | 3624 | |
e98d5b16 | 3625 | spin_unlock(&pool->lock); |
a1056305 | 3626 | } |
a0a1a5fd TH |
3627 | } |
3628 | ||
e98d5b16 | 3629 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3630 | } |
3631 | ||
3632 | /** | |
58a69cb4 | 3633 | * freeze_workqueues_busy - are freezable workqueues still busy? |
a0a1a5fd TH |
3634 | * |
3635 | * Check whether freezing is complete. This function must be called | |
3636 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
3637 | * | |
3638 | * CONTEXT: | |
3639 | * Grabs and releases workqueue_lock. | |
3640 | * | |
3641 | * RETURNS: | |
58a69cb4 TH |
3642 | * %true if some freezable workqueues are still busy. %false if freezing |
3643 | * is complete. | |
a0a1a5fd TH |
3644 | */ |
3645 | bool freeze_workqueues_busy(void) | |
3646 | { | |
a0a1a5fd TH |
3647 | unsigned int cpu; |
3648 | bool busy = false; | |
3649 | ||
e98d5b16 | 3650 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3651 | |
6183c009 | 3652 | WARN_ON_ONCE(!workqueue_freezing); |
a0a1a5fd | 3653 | |
706026c2 | 3654 | for_each_wq_cpu(cpu) { |
bdbc5dd7 | 3655 | struct workqueue_struct *wq; |
a0a1a5fd TH |
3656 | /* |
3657 | * nr_active is monotonically decreasing. It's safe | |
3658 | * to peek without lock. | |
3659 | */ | |
3660 | list_for_each_entry(wq, &workqueues, list) { | |
112202d9 | 3661 | struct pool_workqueue *pwq = get_pwq(cpu, wq); |
a0a1a5fd | 3662 | |
112202d9 | 3663 | if (!pwq || !(wq->flags & WQ_FREEZABLE)) |
a0a1a5fd TH |
3664 | continue; |
3665 | ||
6183c009 | 3666 | WARN_ON_ONCE(pwq->nr_active < 0); |
112202d9 | 3667 | if (pwq->nr_active) { |
a0a1a5fd TH |
3668 | busy = true; |
3669 | goto out_unlock; | |
3670 | } | |
3671 | } | |
3672 | } | |
3673 | out_unlock: | |
e98d5b16 | 3674 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3675 | return busy; |
3676 | } | |
3677 | ||
3678 | /** | |
3679 | * thaw_workqueues - thaw workqueues | |
3680 | * | |
3681 | * Thaw workqueues. Normal queueing is restored and all collected | |
706026c2 | 3682 | * frozen works are transferred to their respective pool worklists. |
a0a1a5fd TH |
3683 | * |
3684 | * CONTEXT: | |
d565ed63 | 3685 | * Grabs and releases workqueue_lock and pool->lock's. |
a0a1a5fd TH |
3686 | */ |
3687 | void thaw_workqueues(void) | |
3688 | { | |
a0a1a5fd TH |
3689 | unsigned int cpu; |
3690 | ||
e98d5b16 | 3691 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd TH |
3692 | |
3693 | if (!workqueue_freezing) | |
3694 | goto out_unlock; | |
3695 | ||
706026c2 | 3696 | for_each_wq_cpu(cpu) { |
4ce62e9e | 3697 | struct worker_pool *pool; |
bdbc5dd7 | 3698 | struct workqueue_struct *wq; |
8b03ae3c | 3699 | |
38db41d9 | 3700 | for_each_std_worker_pool(pool, cpu) { |
e98d5b16 | 3701 | spin_lock(&pool->lock); |
d565ed63 | 3702 | |
35b6bb63 TH |
3703 | WARN_ON_ONCE(!(pool->flags & POOL_FREEZING)); |
3704 | pool->flags &= ~POOL_FREEZING; | |
db7bccf4 | 3705 | |
a1056305 | 3706 | list_for_each_entry(wq, &workqueues, list) { |
112202d9 | 3707 | struct pool_workqueue *pwq = get_pwq(cpu, wq); |
a0a1a5fd | 3708 | |
112202d9 | 3709 | if (!pwq || pwq->pool != pool || |
a1056305 TH |
3710 | !(wq->flags & WQ_FREEZABLE)) |
3711 | continue; | |
a0a1a5fd | 3712 | |
a1056305 | 3713 | /* restore max_active and repopulate worklist */ |
112202d9 | 3714 | pwq_set_max_active(pwq, wq->saved_max_active); |
a1056305 | 3715 | } |
8b03ae3c | 3716 | |
4ce62e9e | 3717 | wake_up_worker(pool); |
a1056305 | 3718 | |
e98d5b16 | 3719 | spin_unlock(&pool->lock); |
d565ed63 | 3720 | } |
a0a1a5fd TH |
3721 | } |
3722 | ||
3723 | workqueue_freezing = false; | |
3724 | out_unlock: | |
e98d5b16 | 3725 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3726 | } |
3727 | #endif /* CONFIG_FREEZER */ | |
3728 | ||
6ee0578b | 3729 | static int __init init_workqueues(void) |
1da177e4 | 3730 | { |
c34056a3 TH |
3731 | unsigned int cpu; |
3732 | ||
7c3eed5c TH |
3733 | /* make sure we have enough bits for OFFQ pool ID */ |
3734 | BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_POOL_SHIFT)) < | |
6be19588 | 3735 | WORK_CPU_END * NR_STD_WORKER_POOLS); |
b5490077 | 3736 | |
65758202 | 3737 | cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); |
a5b4e57d | 3738 | hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); |
8b03ae3c | 3739 | |
706026c2 TH |
3740 | /* initialize CPU pools */ |
3741 | for_each_wq_cpu(cpu) { | |
4ce62e9e | 3742 | struct worker_pool *pool; |
8b03ae3c | 3743 | |
38db41d9 | 3744 | for_each_std_worker_pool(pool, cpu) { |
d565ed63 | 3745 | spin_lock_init(&pool->lock); |
ec22ca5e | 3746 | pool->cpu = cpu; |
24647570 | 3747 | pool->flags |= POOL_DISASSOCIATED; |
4ce62e9e TH |
3748 | INIT_LIST_HEAD(&pool->worklist); |
3749 | INIT_LIST_HEAD(&pool->idle_list); | |
c9e7cf27 | 3750 | hash_init(pool->busy_hash); |
e7577c50 | 3751 | |
4ce62e9e TH |
3752 | init_timer_deferrable(&pool->idle_timer); |
3753 | pool->idle_timer.function = idle_worker_timeout; | |
3754 | pool->idle_timer.data = (unsigned long)pool; | |
e22bee78 | 3755 | |
706026c2 | 3756 | setup_timer(&pool->mayday_timer, pool_mayday_timeout, |
4ce62e9e TH |
3757 | (unsigned long)pool); |
3758 | ||
b2eb83d1 | 3759 | mutex_init(&pool->assoc_mutex); |
4ce62e9e | 3760 | ida_init(&pool->worker_ida); |
9daf9e67 TH |
3761 | |
3762 | /* alloc pool ID */ | |
3763 | BUG_ON(worker_pool_assign_id(pool)); | |
4ce62e9e | 3764 | } |
8b03ae3c TH |
3765 | } |
3766 | ||
e22bee78 | 3767 | /* create the initial worker */ |
706026c2 | 3768 | for_each_online_wq_cpu(cpu) { |
4ce62e9e | 3769 | struct worker_pool *pool; |
e22bee78 | 3770 | |
38db41d9 | 3771 | for_each_std_worker_pool(pool, cpu) { |
4ce62e9e TH |
3772 | struct worker *worker; |
3773 | ||
24647570 TH |
3774 | if (cpu != WORK_CPU_UNBOUND) |
3775 | pool->flags &= ~POOL_DISASSOCIATED; | |
3776 | ||
bc2ae0f5 | 3777 | worker = create_worker(pool); |
4ce62e9e | 3778 | BUG_ON(!worker); |
d565ed63 | 3779 | spin_lock_irq(&pool->lock); |
4ce62e9e | 3780 | start_worker(worker); |
d565ed63 | 3781 | spin_unlock_irq(&pool->lock); |
4ce62e9e | 3782 | } |
e22bee78 TH |
3783 | } |
3784 | ||
d320c038 | 3785 | system_wq = alloc_workqueue("events", 0, 0); |
1aabe902 | 3786 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
d320c038 | 3787 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
f3421797 TH |
3788 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
3789 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
3790 | system_freezable_wq = alloc_workqueue("events_freezable", |
3791 | WQ_FREEZABLE, 0); | |
1aabe902 | 3792 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
ae930e0f | 3793 | !system_unbound_wq || !system_freezable_wq); |
6ee0578b | 3794 | return 0; |
1da177e4 | 3795 | } |
6ee0578b | 3796 | early_initcall(init_workqueues); |