]>
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 */ |
d84ff051 | 127 | 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 */ |
30cdf249 | 172 | struct list_head pwqs_node; /* I: node on wq->pwqs */ |
493a1724 | 173 | struct list_head mayday_node; /* W: node on wq->maydays */ |
e904e6c2 | 174 | } __aligned(1 << WORK_STRUCT_FLAG_BITS); |
1da177e4 | 175 | |
73f53c4a TH |
176 | /* |
177 | * Structure used to wait for workqueue flush. | |
178 | */ | |
179 | struct wq_flusher { | |
180 | struct list_head list; /* F: list of flushers */ | |
181 | int flush_color; /* F: flush color waiting for */ | |
182 | struct completion done; /* flush completion */ | |
183 | }; | |
184 | ||
1da177e4 LT |
185 | /* |
186 | * The externally visible workqueue abstraction is an array of | |
187 | * per-CPU workqueues: | |
188 | */ | |
189 | struct workqueue_struct { | |
9c5a2ba7 | 190 | unsigned int flags; /* W: WQ_* flags */ |
420c0ddb | 191 | struct pool_workqueue __percpu *cpu_pwqs; /* I: per-cpu pwq's */ |
30cdf249 | 192 | struct list_head pwqs; /* I: all pwqs of this wq */ |
4690c4ab | 193 | struct list_head list; /* W: list of all workqueues */ |
73f53c4a TH |
194 | |
195 | struct mutex flush_mutex; /* protects wq flushing */ | |
196 | int work_color; /* F: current work color */ | |
197 | int flush_color; /* F: current flush color */ | |
112202d9 | 198 | atomic_t nr_pwqs_to_flush; /* flush in progress */ |
73f53c4a TH |
199 | struct wq_flusher *first_flusher; /* F: first flusher */ |
200 | struct list_head flusher_queue; /* F: flush waiters */ | |
201 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
202 | ||
493a1724 | 203 | struct list_head maydays; /* W: pwqs requesting rescue */ |
e22bee78 TH |
204 | struct worker *rescuer; /* I: rescue worker */ |
205 | ||
9c5a2ba7 | 206 | int nr_drainers; /* W: drain in progress */ |
112202d9 | 207 | int saved_max_active; /* W: saved pwq max_active */ |
4e6045f1 | 208 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 209 | struct lockdep_map lockdep_map; |
4e6045f1 | 210 | #endif |
b196be89 | 211 | char name[]; /* I: workqueue name */ |
1da177e4 LT |
212 | }; |
213 | ||
e904e6c2 TH |
214 | static struct kmem_cache *pwq_cache; |
215 | ||
d320c038 | 216 | struct workqueue_struct *system_wq __read_mostly; |
d320c038 | 217 | EXPORT_SYMBOL_GPL(system_wq); |
044c782c | 218 | struct workqueue_struct *system_highpri_wq __read_mostly; |
1aabe902 | 219 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
044c782c | 220 | struct workqueue_struct *system_long_wq __read_mostly; |
d320c038 | 221 | EXPORT_SYMBOL_GPL(system_long_wq); |
044c782c | 222 | struct workqueue_struct *system_unbound_wq __read_mostly; |
f3421797 | 223 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
044c782c | 224 | struct workqueue_struct *system_freezable_wq __read_mostly; |
24d51add | 225 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
d320c038 | 226 | |
97bd2347 TH |
227 | #define CREATE_TRACE_POINTS |
228 | #include <trace/events/workqueue.h> | |
229 | ||
38db41d9 | 230 | #define for_each_std_worker_pool(pool, cpu) \ |
a60dc39c TH |
231 | for ((pool) = &std_worker_pools(cpu)[0]; \ |
232 | (pool) < &std_worker_pools(cpu)[NR_STD_WORKER_POOLS]; (pool)++) | |
4ce62e9e | 233 | |
b67bfe0d SL |
234 | #define for_each_busy_worker(worker, i, pool) \ |
235 | hash_for_each(pool->busy_hash, i, worker, hentry) | |
db7bccf4 | 236 | |
706026c2 TH |
237 | static inline int __next_wq_cpu(int cpu, const struct cpumask *mask, |
238 | unsigned int sw) | |
f3421797 TH |
239 | { |
240 | if (cpu < nr_cpu_ids) { | |
241 | if (sw & 1) { | |
242 | cpu = cpumask_next(cpu, mask); | |
243 | if (cpu < nr_cpu_ids) | |
244 | return cpu; | |
245 | } | |
246 | if (sw & 2) | |
247 | return WORK_CPU_UNBOUND; | |
248 | } | |
6be19588 | 249 | return WORK_CPU_END; |
f3421797 TH |
250 | } |
251 | ||
09884951 TH |
252 | /* |
253 | * CPU iterators | |
254 | * | |
706026c2 | 255 | * An extra cpu number is defined using an invalid cpu number |
09884951 | 256 | * (WORK_CPU_UNBOUND) to host workqueues which are not bound to any |
706026c2 TH |
257 | * specific CPU. The following iterators are similar to for_each_*_cpu() |
258 | * iterators but also considers the unbound CPU. | |
09884951 | 259 | * |
706026c2 TH |
260 | * for_each_wq_cpu() : possible CPUs + WORK_CPU_UNBOUND |
261 | * for_each_online_wq_cpu() : online CPUs + WORK_CPU_UNBOUND | |
09884951 | 262 | */ |
706026c2 TH |
263 | #define for_each_wq_cpu(cpu) \ |
264 | for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, 3); \ | |
6be19588 | 265 | (cpu) < WORK_CPU_END; \ |
706026c2 | 266 | (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, 3)) |
f3421797 | 267 | |
706026c2 TH |
268 | #define for_each_online_wq_cpu(cpu) \ |
269 | for ((cpu) = __next_wq_cpu(-1, cpu_online_mask, 3); \ | |
6be19588 | 270 | (cpu) < WORK_CPU_END; \ |
706026c2 | 271 | (cpu) = __next_wq_cpu((cpu), cpu_online_mask, 3)) |
f3421797 | 272 | |
17116969 TH |
273 | /** |
274 | * for_each_pool - iterate through all worker_pools in the system | |
275 | * @pool: iteration cursor | |
276 | * @id: integer used for iteration | |
277 | */ | |
278 | #define for_each_pool(pool, id) \ | |
279 | idr_for_each_entry(&worker_pool_idr, pool, id) | |
280 | ||
49e3cf44 TH |
281 | /** |
282 | * for_each_pwq - iterate through all pool_workqueues of the specified workqueue | |
283 | * @pwq: iteration cursor | |
284 | * @wq: the target workqueue | |
285 | */ | |
286 | #define for_each_pwq(pwq, wq) \ | |
287 | list_for_each_entry((pwq), &(wq)->pwqs, pwqs_node) | |
f3421797 | 288 | |
dc186ad7 TG |
289 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
290 | ||
291 | static struct debug_obj_descr work_debug_descr; | |
292 | ||
99777288 SG |
293 | static void *work_debug_hint(void *addr) |
294 | { | |
295 | return ((struct work_struct *) addr)->func; | |
296 | } | |
297 | ||
dc186ad7 TG |
298 | /* |
299 | * fixup_init is called when: | |
300 | * - an active object is initialized | |
301 | */ | |
302 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
303 | { | |
304 | struct work_struct *work = addr; | |
305 | ||
306 | switch (state) { | |
307 | case ODEBUG_STATE_ACTIVE: | |
308 | cancel_work_sync(work); | |
309 | debug_object_init(work, &work_debug_descr); | |
310 | return 1; | |
311 | default: | |
312 | return 0; | |
313 | } | |
314 | } | |
315 | ||
316 | /* | |
317 | * fixup_activate is called when: | |
318 | * - an active object is activated | |
319 | * - an unknown object is activated (might be a statically initialized object) | |
320 | */ | |
321 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
322 | { | |
323 | struct work_struct *work = addr; | |
324 | ||
325 | switch (state) { | |
326 | ||
327 | case ODEBUG_STATE_NOTAVAILABLE: | |
328 | /* | |
329 | * This is not really a fixup. The work struct was | |
330 | * statically initialized. We just make sure that it | |
331 | * is tracked in the object tracker. | |
332 | */ | |
22df02bb | 333 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
334 | debug_object_init(work, &work_debug_descr); |
335 | debug_object_activate(work, &work_debug_descr); | |
336 | return 0; | |
337 | } | |
338 | WARN_ON_ONCE(1); | |
339 | return 0; | |
340 | ||
341 | case ODEBUG_STATE_ACTIVE: | |
342 | WARN_ON(1); | |
343 | ||
344 | default: | |
345 | return 0; | |
346 | } | |
347 | } | |
348 | ||
349 | /* | |
350 | * fixup_free is called when: | |
351 | * - an active object is freed | |
352 | */ | |
353 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
354 | { | |
355 | struct work_struct *work = addr; | |
356 | ||
357 | switch (state) { | |
358 | case ODEBUG_STATE_ACTIVE: | |
359 | cancel_work_sync(work); | |
360 | debug_object_free(work, &work_debug_descr); | |
361 | return 1; | |
362 | default: | |
363 | return 0; | |
364 | } | |
365 | } | |
366 | ||
367 | static struct debug_obj_descr work_debug_descr = { | |
368 | .name = "work_struct", | |
99777288 | 369 | .debug_hint = work_debug_hint, |
dc186ad7 TG |
370 | .fixup_init = work_fixup_init, |
371 | .fixup_activate = work_fixup_activate, | |
372 | .fixup_free = work_fixup_free, | |
373 | }; | |
374 | ||
375 | static inline void debug_work_activate(struct work_struct *work) | |
376 | { | |
377 | debug_object_activate(work, &work_debug_descr); | |
378 | } | |
379 | ||
380 | static inline void debug_work_deactivate(struct work_struct *work) | |
381 | { | |
382 | debug_object_deactivate(work, &work_debug_descr); | |
383 | } | |
384 | ||
385 | void __init_work(struct work_struct *work, int onstack) | |
386 | { | |
387 | if (onstack) | |
388 | debug_object_init_on_stack(work, &work_debug_descr); | |
389 | else | |
390 | debug_object_init(work, &work_debug_descr); | |
391 | } | |
392 | EXPORT_SYMBOL_GPL(__init_work); | |
393 | ||
394 | void destroy_work_on_stack(struct work_struct *work) | |
395 | { | |
396 | debug_object_free(work, &work_debug_descr); | |
397 | } | |
398 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
399 | ||
400 | #else | |
401 | static inline void debug_work_activate(struct work_struct *work) { } | |
402 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
403 | #endif | |
404 | ||
95402b38 GS |
405 | /* Serializes the accesses to the list of workqueues. */ |
406 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 407 | static LIST_HEAD(workqueues); |
a0a1a5fd | 408 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 409 | |
e22bee78 | 410 | /* |
e19e397a TH |
411 | * The CPU and unbound standard worker pools. The unbound ones have |
412 | * POOL_DISASSOCIATED set, and their workers have WORKER_UNBOUND set. | |
f3421797 | 413 | */ |
e19e397a TH |
414 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS], |
415 | cpu_std_worker_pools); | |
a60dc39c | 416 | static struct worker_pool unbound_std_worker_pools[NR_STD_WORKER_POOLS]; |
f3421797 | 417 | |
9daf9e67 TH |
418 | /* idr of all pools */ |
419 | static DEFINE_MUTEX(worker_pool_idr_mutex); | |
420 | static DEFINE_IDR(worker_pool_idr); | |
421 | ||
c34056a3 | 422 | static int worker_thread(void *__worker); |
1da177e4 | 423 | |
a60dc39c | 424 | static struct worker_pool *std_worker_pools(int cpu) |
8b03ae3c | 425 | { |
f3421797 | 426 | if (cpu != WORK_CPU_UNBOUND) |
a60dc39c | 427 | return per_cpu(cpu_std_worker_pools, cpu); |
f3421797 | 428 | else |
a60dc39c | 429 | return unbound_std_worker_pools; |
8b03ae3c TH |
430 | } |
431 | ||
4e8f0a60 TH |
432 | static int std_worker_pool_pri(struct worker_pool *pool) |
433 | { | |
a60dc39c | 434 | return pool - std_worker_pools(pool->cpu); |
4e8f0a60 TH |
435 | } |
436 | ||
9daf9e67 TH |
437 | /* allocate ID and assign it to @pool */ |
438 | static int worker_pool_assign_id(struct worker_pool *pool) | |
439 | { | |
440 | int ret; | |
441 | ||
442 | mutex_lock(&worker_pool_idr_mutex); | |
443 | idr_pre_get(&worker_pool_idr, GFP_KERNEL); | |
444 | ret = idr_get_new(&worker_pool_idr, pool, &pool->id); | |
445 | mutex_unlock(&worker_pool_idr_mutex); | |
446 | ||
447 | return ret; | |
448 | } | |
449 | ||
7c3eed5c TH |
450 | /* |
451 | * Lookup worker_pool by id. The idr currently is built during boot and | |
452 | * never modified. Don't worry about locking for now. | |
453 | */ | |
454 | static struct worker_pool *worker_pool_by_id(int pool_id) | |
455 | { | |
456 | return idr_find(&worker_pool_idr, pool_id); | |
457 | } | |
458 | ||
d565ed63 TH |
459 | static struct worker_pool *get_std_worker_pool(int cpu, bool highpri) |
460 | { | |
a60dc39c | 461 | struct worker_pool *pools = std_worker_pools(cpu); |
d565ed63 | 462 | |
a60dc39c | 463 | return &pools[highpri]; |
d565ed63 TH |
464 | } |
465 | ||
d84ff051 | 466 | static struct pool_workqueue *get_pwq(int cpu, struct workqueue_struct *wq) |
b1f4ec17 | 467 | { |
f3421797 | 468 | if (!(wq->flags & WQ_UNBOUND)) { |
e06ffa1e | 469 | if (likely(cpu < nr_cpu_ids)) |
420c0ddb TH |
470 | return per_cpu_ptr(wq->cpu_pwqs, cpu); |
471 | } else if (likely(cpu == WORK_CPU_UNBOUND)) { | |
472 | return list_first_entry(&wq->pwqs, struct pool_workqueue, | |
473 | pwqs_node); | |
474 | } | |
f3421797 | 475 | return NULL; |
b1f4ec17 ON |
476 | } |
477 | ||
73f53c4a TH |
478 | static unsigned int work_color_to_flags(int color) |
479 | { | |
480 | return color << WORK_STRUCT_COLOR_SHIFT; | |
481 | } | |
482 | ||
483 | static int get_work_color(struct work_struct *work) | |
484 | { | |
485 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
486 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
487 | } | |
488 | ||
489 | static int work_next_color(int color) | |
490 | { | |
491 | return (color + 1) % WORK_NR_COLORS; | |
492 | } | |
1da177e4 | 493 | |
14441960 | 494 | /* |
112202d9 TH |
495 | * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data |
496 | * contain the pointer to the queued pwq. Once execution starts, the flag | |
7c3eed5c | 497 | * is cleared and the high bits contain OFFQ flags and pool ID. |
7a22ad75 | 498 | * |
112202d9 TH |
499 | * set_work_pwq(), set_work_pool_and_clear_pending(), mark_work_canceling() |
500 | * and clear_work_data() can be used to set the pwq, pool or clear | |
bbb68dfa TH |
501 | * work->data. These functions should only be called while the work is |
502 | * owned - ie. while the PENDING bit is set. | |
7a22ad75 | 503 | * |
112202d9 | 504 | * get_work_pool() and get_work_pwq() can be used to obtain the pool or pwq |
7c3eed5c | 505 | * corresponding to a work. Pool is available once the work has been |
112202d9 | 506 | * queued anywhere after initialization until it is sync canceled. pwq is |
7c3eed5c | 507 | * available only while the work item is queued. |
7a22ad75 | 508 | * |
bbb68dfa TH |
509 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
510 | * canceled. While being canceled, a work item may have its PENDING set | |
511 | * but stay off timer and worklist for arbitrarily long and nobody should | |
512 | * try to steal the PENDING bit. | |
14441960 | 513 | */ |
7a22ad75 TH |
514 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
515 | unsigned long flags) | |
365970a1 | 516 | { |
6183c009 | 517 | WARN_ON_ONCE(!work_pending(work)); |
7a22ad75 TH |
518 | atomic_long_set(&work->data, data | flags | work_static(work)); |
519 | } | |
365970a1 | 520 | |
112202d9 | 521 | static void set_work_pwq(struct work_struct *work, struct pool_workqueue *pwq, |
7a22ad75 TH |
522 | unsigned long extra_flags) |
523 | { | |
112202d9 TH |
524 | set_work_data(work, (unsigned long)pwq, |
525 | WORK_STRUCT_PENDING | WORK_STRUCT_PWQ | extra_flags); | |
365970a1 DH |
526 | } |
527 | ||
4468a00f LJ |
528 | static void set_work_pool_and_keep_pending(struct work_struct *work, |
529 | int pool_id) | |
530 | { | |
531 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, | |
532 | WORK_STRUCT_PENDING); | |
533 | } | |
534 | ||
7c3eed5c TH |
535 | static void set_work_pool_and_clear_pending(struct work_struct *work, |
536 | int pool_id) | |
7a22ad75 | 537 | { |
23657bb1 TH |
538 | /* |
539 | * The following wmb is paired with the implied mb in | |
540 | * test_and_set_bit(PENDING) and ensures all updates to @work made | |
541 | * here are visible to and precede any updates by the next PENDING | |
542 | * owner. | |
543 | */ | |
544 | smp_wmb(); | |
7c3eed5c | 545 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0); |
7a22ad75 | 546 | } |
f756d5e2 | 547 | |
7a22ad75 | 548 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 549 | { |
7c3eed5c TH |
550 | smp_wmb(); /* see set_work_pool_and_clear_pending() */ |
551 | set_work_data(work, WORK_STRUCT_NO_POOL, 0); | |
1da177e4 LT |
552 | } |
553 | ||
112202d9 | 554 | static struct pool_workqueue *get_work_pwq(struct work_struct *work) |
b1f4ec17 | 555 | { |
e120153d | 556 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 557 | |
112202d9 | 558 | if (data & WORK_STRUCT_PWQ) |
e120153d TH |
559 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); |
560 | else | |
561 | return NULL; | |
4d707b9f ON |
562 | } |
563 | ||
7c3eed5c TH |
564 | /** |
565 | * get_work_pool - return the worker_pool a given work was associated with | |
566 | * @work: the work item of interest | |
567 | * | |
568 | * Return the worker_pool @work was last associated with. %NULL if none. | |
569 | */ | |
570 | static struct worker_pool *get_work_pool(struct work_struct *work) | |
365970a1 | 571 | { |
e120153d | 572 | unsigned long data = atomic_long_read(&work->data); |
7c3eed5c TH |
573 | struct worker_pool *pool; |
574 | int pool_id; | |
7a22ad75 | 575 | |
112202d9 TH |
576 | if (data & WORK_STRUCT_PWQ) |
577 | return ((struct pool_workqueue *) | |
7c3eed5c | 578 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool; |
7a22ad75 | 579 | |
7c3eed5c TH |
580 | pool_id = data >> WORK_OFFQ_POOL_SHIFT; |
581 | if (pool_id == WORK_OFFQ_POOL_NONE) | |
7a22ad75 TH |
582 | return NULL; |
583 | ||
7c3eed5c TH |
584 | pool = worker_pool_by_id(pool_id); |
585 | WARN_ON_ONCE(!pool); | |
586 | return pool; | |
587 | } | |
588 | ||
589 | /** | |
590 | * get_work_pool_id - return the worker pool ID a given work is associated with | |
591 | * @work: the work item of interest | |
592 | * | |
593 | * Return the worker_pool ID @work was last associated with. | |
594 | * %WORK_OFFQ_POOL_NONE if none. | |
595 | */ | |
596 | static int get_work_pool_id(struct work_struct *work) | |
597 | { | |
54d5b7d0 LJ |
598 | unsigned long data = atomic_long_read(&work->data); |
599 | ||
112202d9 TH |
600 | if (data & WORK_STRUCT_PWQ) |
601 | return ((struct pool_workqueue *) | |
54d5b7d0 | 602 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->id; |
7c3eed5c | 603 | |
54d5b7d0 | 604 | return data >> WORK_OFFQ_POOL_SHIFT; |
7c3eed5c TH |
605 | } |
606 | ||
bbb68dfa TH |
607 | static void mark_work_canceling(struct work_struct *work) |
608 | { | |
7c3eed5c | 609 | unsigned long pool_id = get_work_pool_id(work); |
bbb68dfa | 610 | |
7c3eed5c TH |
611 | pool_id <<= WORK_OFFQ_POOL_SHIFT; |
612 | set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING); | |
bbb68dfa TH |
613 | } |
614 | ||
615 | static bool work_is_canceling(struct work_struct *work) | |
616 | { | |
617 | unsigned long data = atomic_long_read(&work->data); | |
618 | ||
112202d9 | 619 | return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING); |
bbb68dfa TH |
620 | } |
621 | ||
e22bee78 | 622 | /* |
3270476a TH |
623 | * Policy functions. These define the policies on how the global worker |
624 | * pools are managed. Unless noted otherwise, these functions assume that | |
d565ed63 | 625 | * they're being called with pool->lock held. |
e22bee78 TH |
626 | */ |
627 | ||
63d95a91 | 628 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 629 | { |
e19e397a | 630 | return !atomic_read(&pool->nr_running); |
a848e3b6 ON |
631 | } |
632 | ||
4594bf15 | 633 | /* |
e22bee78 TH |
634 | * Need to wake up a worker? Called from anything but currently |
635 | * running workers. | |
974271c4 TH |
636 | * |
637 | * Note that, because unbound workers never contribute to nr_running, this | |
706026c2 | 638 | * function will always return %true for unbound pools as long as the |
974271c4 | 639 | * worklist isn't empty. |
4594bf15 | 640 | */ |
63d95a91 | 641 | static bool need_more_worker(struct worker_pool *pool) |
365970a1 | 642 | { |
63d95a91 | 643 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
e22bee78 | 644 | } |
4594bf15 | 645 | |
e22bee78 | 646 | /* Can I start working? Called from busy but !running workers. */ |
63d95a91 | 647 | static bool may_start_working(struct worker_pool *pool) |
e22bee78 | 648 | { |
63d95a91 | 649 | return pool->nr_idle; |
e22bee78 TH |
650 | } |
651 | ||
652 | /* Do I need to keep working? Called from currently running workers. */ | |
63d95a91 | 653 | static bool keep_working(struct worker_pool *pool) |
e22bee78 | 654 | { |
e19e397a TH |
655 | return !list_empty(&pool->worklist) && |
656 | atomic_read(&pool->nr_running) <= 1; | |
e22bee78 TH |
657 | } |
658 | ||
659 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 660 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 661 | { |
63d95a91 | 662 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 663 | } |
365970a1 | 664 | |
e22bee78 | 665 | /* Do I need to be the manager? */ |
63d95a91 | 666 | static bool need_to_manage_workers(struct worker_pool *pool) |
e22bee78 | 667 | { |
63d95a91 | 668 | return need_to_create_worker(pool) || |
11ebea50 | 669 | (pool->flags & POOL_MANAGE_WORKERS); |
e22bee78 TH |
670 | } |
671 | ||
672 | /* Do we have too many workers and should some go away? */ | |
63d95a91 | 673 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 674 | { |
552a37e9 | 675 | bool managing = pool->flags & POOL_MANAGING_WORKERS; |
63d95a91 TH |
676 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
677 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 | 678 | |
ea1abd61 LJ |
679 | /* |
680 | * nr_idle and idle_list may disagree if idle rebinding is in | |
681 | * progress. Never return %true if idle_list is empty. | |
682 | */ | |
683 | if (list_empty(&pool->idle_list)) | |
684 | return false; | |
685 | ||
e22bee78 | 686 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
365970a1 DH |
687 | } |
688 | ||
4d707b9f | 689 | /* |
e22bee78 TH |
690 | * Wake up functions. |
691 | */ | |
692 | ||
7e11629d | 693 | /* Return the first worker. Safe with preemption disabled */ |
63d95a91 | 694 | static struct worker *first_worker(struct worker_pool *pool) |
7e11629d | 695 | { |
63d95a91 | 696 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
697 | return NULL; |
698 | ||
63d95a91 | 699 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
700 | } |
701 | ||
702 | /** | |
703 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 704 | * @pool: worker pool to wake worker from |
7e11629d | 705 | * |
63d95a91 | 706 | * Wake up the first idle worker of @pool. |
7e11629d TH |
707 | * |
708 | * CONTEXT: | |
d565ed63 | 709 | * spin_lock_irq(pool->lock). |
7e11629d | 710 | */ |
63d95a91 | 711 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 712 | { |
63d95a91 | 713 | struct worker *worker = first_worker(pool); |
7e11629d TH |
714 | |
715 | if (likely(worker)) | |
716 | wake_up_process(worker->task); | |
717 | } | |
718 | ||
d302f017 | 719 | /** |
e22bee78 TH |
720 | * wq_worker_waking_up - a worker is waking up |
721 | * @task: task waking up | |
722 | * @cpu: CPU @task is waking up to | |
723 | * | |
724 | * This function is called during try_to_wake_up() when a worker is | |
725 | * being awoken. | |
726 | * | |
727 | * CONTEXT: | |
728 | * spin_lock_irq(rq->lock) | |
729 | */ | |
d84ff051 | 730 | void wq_worker_waking_up(struct task_struct *task, int cpu) |
e22bee78 TH |
731 | { |
732 | struct worker *worker = kthread_data(task); | |
733 | ||
36576000 | 734 | if (!(worker->flags & WORKER_NOT_RUNNING)) { |
ec22ca5e | 735 | WARN_ON_ONCE(worker->pool->cpu != cpu); |
e19e397a | 736 | atomic_inc(&worker->pool->nr_running); |
36576000 | 737 | } |
e22bee78 TH |
738 | } |
739 | ||
740 | /** | |
741 | * wq_worker_sleeping - a worker is going to sleep | |
742 | * @task: task going to sleep | |
743 | * @cpu: CPU in question, must be the current CPU number | |
744 | * | |
745 | * This function is called during schedule() when a busy worker is | |
746 | * going to sleep. Worker on the same cpu can be woken up by | |
747 | * returning pointer to its task. | |
748 | * | |
749 | * CONTEXT: | |
750 | * spin_lock_irq(rq->lock) | |
751 | * | |
752 | * RETURNS: | |
753 | * Worker task on @cpu to wake up, %NULL if none. | |
754 | */ | |
d84ff051 | 755 | struct task_struct *wq_worker_sleeping(struct task_struct *task, int cpu) |
e22bee78 TH |
756 | { |
757 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
111c225a | 758 | struct worker_pool *pool; |
e22bee78 | 759 | |
111c225a TH |
760 | /* |
761 | * Rescuers, which may not have all the fields set up like normal | |
762 | * workers, also reach here, let's not access anything before | |
763 | * checking NOT_RUNNING. | |
764 | */ | |
2d64672e | 765 | if (worker->flags & WORKER_NOT_RUNNING) |
e22bee78 TH |
766 | return NULL; |
767 | ||
111c225a | 768 | pool = worker->pool; |
111c225a | 769 | |
e22bee78 | 770 | /* this can only happen on the local cpu */ |
6183c009 TH |
771 | if (WARN_ON_ONCE(cpu != raw_smp_processor_id())) |
772 | return NULL; | |
e22bee78 TH |
773 | |
774 | /* | |
775 | * The counterpart of the following dec_and_test, implied mb, | |
776 | * worklist not empty test sequence is in insert_work(). | |
777 | * Please read comment there. | |
778 | * | |
628c78e7 TH |
779 | * NOT_RUNNING is clear. This means that we're bound to and |
780 | * running on the local cpu w/ rq lock held and preemption | |
781 | * disabled, which in turn means that none else could be | |
d565ed63 | 782 | * manipulating idle_list, so dereferencing idle_list without pool |
628c78e7 | 783 | * lock is safe. |
e22bee78 | 784 | */ |
e19e397a TH |
785 | if (atomic_dec_and_test(&pool->nr_running) && |
786 | !list_empty(&pool->worklist)) | |
63d95a91 | 787 | to_wakeup = first_worker(pool); |
e22bee78 TH |
788 | return to_wakeup ? to_wakeup->task : NULL; |
789 | } | |
790 | ||
791 | /** | |
792 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 793 | * @worker: self |
d302f017 TH |
794 | * @flags: flags to set |
795 | * @wakeup: wakeup an idle worker if necessary | |
796 | * | |
e22bee78 TH |
797 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
798 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
799 | * woken up. | |
d302f017 | 800 | * |
cb444766 | 801 | * CONTEXT: |
d565ed63 | 802 | * spin_lock_irq(pool->lock) |
d302f017 TH |
803 | */ |
804 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
805 | bool wakeup) | |
806 | { | |
bd7bdd43 | 807 | struct worker_pool *pool = worker->pool; |
e22bee78 | 808 | |
cb444766 TH |
809 | WARN_ON_ONCE(worker->task != current); |
810 | ||
e22bee78 TH |
811 | /* |
812 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
813 | * wake up an idle worker as necessary if requested by | |
814 | * @wakeup. | |
815 | */ | |
816 | if ((flags & WORKER_NOT_RUNNING) && | |
817 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
e22bee78 | 818 | if (wakeup) { |
e19e397a | 819 | if (atomic_dec_and_test(&pool->nr_running) && |
bd7bdd43 | 820 | !list_empty(&pool->worklist)) |
63d95a91 | 821 | wake_up_worker(pool); |
e22bee78 | 822 | } else |
e19e397a | 823 | atomic_dec(&pool->nr_running); |
e22bee78 TH |
824 | } |
825 | ||
d302f017 TH |
826 | worker->flags |= flags; |
827 | } | |
828 | ||
829 | /** | |
e22bee78 | 830 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 831 | * @worker: self |
d302f017 TH |
832 | * @flags: flags to clear |
833 | * | |
e22bee78 | 834 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 835 | * |
cb444766 | 836 | * CONTEXT: |
d565ed63 | 837 | * spin_lock_irq(pool->lock) |
d302f017 TH |
838 | */ |
839 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
840 | { | |
63d95a91 | 841 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
842 | unsigned int oflags = worker->flags; |
843 | ||
cb444766 TH |
844 | WARN_ON_ONCE(worker->task != current); |
845 | ||
d302f017 | 846 | worker->flags &= ~flags; |
e22bee78 | 847 | |
42c025f3 TH |
848 | /* |
849 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
850 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
851 | * of multiple flags, not a single flag. | |
852 | */ | |
e22bee78 TH |
853 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
854 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
e19e397a | 855 | atomic_inc(&pool->nr_running); |
d302f017 TH |
856 | } |
857 | ||
8cca0eea TH |
858 | /** |
859 | * find_worker_executing_work - find worker which is executing a work | |
c9e7cf27 | 860 | * @pool: pool of interest |
8cca0eea TH |
861 | * @work: work to find worker for |
862 | * | |
c9e7cf27 TH |
863 | * Find a worker which is executing @work on @pool by searching |
864 | * @pool->busy_hash which is keyed by the address of @work. For a worker | |
a2c1c57b TH |
865 | * to match, its current execution should match the address of @work and |
866 | * its work function. This is to avoid unwanted dependency between | |
867 | * unrelated work executions through a work item being recycled while still | |
868 | * being executed. | |
869 | * | |
870 | * This is a bit tricky. A work item may be freed once its execution | |
871 | * starts and nothing prevents the freed area from being recycled for | |
872 | * another work item. If the same work item address ends up being reused | |
873 | * before the original execution finishes, workqueue will identify the | |
874 | * recycled work item as currently executing and make it wait until the | |
875 | * current execution finishes, introducing an unwanted dependency. | |
876 | * | |
877 | * This function checks the work item address, work function and workqueue | |
878 | * to avoid false positives. Note that this isn't complete as one may | |
879 | * construct a work function which can introduce dependency onto itself | |
880 | * through a recycled work item. Well, if somebody wants to shoot oneself | |
881 | * in the foot that badly, there's only so much we can do, and if such | |
882 | * deadlock actually occurs, it should be easy to locate the culprit work | |
883 | * function. | |
8cca0eea TH |
884 | * |
885 | * CONTEXT: | |
d565ed63 | 886 | * spin_lock_irq(pool->lock). |
8cca0eea TH |
887 | * |
888 | * RETURNS: | |
889 | * Pointer to worker which is executing @work if found, NULL | |
890 | * otherwise. | |
4d707b9f | 891 | */ |
c9e7cf27 | 892 | static struct worker *find_worker_executing_work(struct worker_pool *pool, |
8cca0eea | 893 | struct work_struct *work) |
4d707b9f | 894 | { |
42f8570f | 895 | struct worker *worker; |
42f8570f | 896 | |
b67bfe0d | 897 | hash_for_each_possible(pool->busy_hash, worker, hentry, |
a2c1c57b TH |
898 | (unsigned long)work) |
899 | if (worker->current_work == work && | |
900 | worker->current_func == work->func) | |
42f8570f SL |
901 | return worker; |
902 | ||
903 | return NULL; | |
4d707b9f ON |
904 | } |
905 | ||
bf4ede01 TH |
906 | /** |
907 | * move_linked_works - move linked works to a list | |
908 | * @work: start of series of works to be scheduled | |
909 | * @head: target list to append @work to | |
910 | * @nextp: out paramter for nested worklist walking | |
911 | * | |
912 | * Schedule linked works starting from @work to @head. Work series to | |
913 | * be scheduled starts at @work and includes any consecutive work with | |
914 | * WORK_STRUCT_LINKED set in its predecessor. | |
915 | * | |
916 | * If @nextp is not NULL, it's updated to point to the next work of | |
917 | * the last scheduled work. This allows move_linked_works() to be | |
918 | * nested inside outer list_for_each_entry_safe(). | |
919 | * | |
920 | * CONTEXT: | |
d565ed63 | 921 | * spin_lock_irq(pool->lock). |
bf4ede01 TH |
922 | */ |
923 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
924 | struct work_struct **nextp) | |
925 | { | |
926 | struct work_struct *n; | |
927 | ||
928 | /* | |
929 | * Linked worklist will always end before the end of the list, | |
930 | * use NULL for list head. | |
931 | */ | |
932 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
933 | list_move_tail(&work->entry, head); | |
934 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
935 | break; | |
936 | } | |
937 | ||
938 | /* | |
939 | * If we're already inside safe list traversal and have moved | |
940 | * multiple works to the scheduled queue, the next position | |
941 | * needs to be updated. | |
942 | */ | |
943 | if (nextp) | |
944 | *nextp = n; | |
945 | } | |
946 | ||
112202d9 | 947 | static void pwq_activate_delayed_work(struct work_struct *work) |
bf4ede01 | 948 | { |
112202d9 | 949 | struct pool_workqueue *pwq = get_work_pwq(work); |
bf4ede01 TH |
950 | |
951 | trace_workqueue_activate_work(work); | |
112202d9 | 952 | move_linked_works(work, &pwq->pool->worklist, NULL); |
bf4ede01 | 953 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); |
112202d9 | 954 | pwq->nr_active++; |
bf4ede01 TH |
955 | } |
956 | ||
112202d9 | 957 | static void pwq_activate_first_delayed(struct pool_workqueue *pwq) |
3aa62497 | 958 | { |
112202d9 | 959 | struct work_struct *work = list_first_entry(&pwq->delayed_works, |
3aa62497 LJ |
960 | struct work_struct, entry); |
961 | ||
112202d9 | 962 | pwq_activate_delayed_work(work); |
3aa62497 LJ |
963 | } |
964 | ||
bf4ede01 | 965 | /** |
112202d9 TH |
966 | * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight |
967 | * @pwq: pwq of interest | |
bf4ede01 | 968 | * @color: color of work which left the queue |
bf4ede01 TH |
969 | * |
970 | * A work either has completed or is removed from pending queue, | |
112202d9 | 971 | * decrement nr_in_flight of its pwq and handle workqueue flushing. |
bf4ede01 TH |
972 | * |
973 | * CONTEXT: | |
d565ed63 | 974 | * spin_lock_irq(pool->lock). |
bf4ede01 | 975 | */ |
112202d9 | 976 | static void pwq_dec_nr_in_flight(struct pool_workqueue *pwq, int color) |
bf4ede01 TH |
977 | { |
978 | /* ignore uncolored works */ | |
979 | if (color == WORK_NO_COLOR) | |
980 | return; | |
981 | ||
112202d9 | 982 | pwq->nr_in_flight[color]--; |
bf4ede01 | 983 | |
112202d9 TH |
984 | pwq->nr_active--; |
985 | if (!list_empty(&pwq->delayed_works)) { | |
b3f9f405 | 986 | /* one down, submit a delayed one */ |
112202d9 TH |
987 | if (pwq->nr_active < pwq->max_active) |
988 | pwq_activate_first_delayed(pwq); | |
bf4ede01 TH |
989 | } |
990 | ||
991 | /* is flush in progress and are we at the flushing tip? */ | |
112202d9 | 992 | if (likely(pwq->flush_color != color)) |
bf4ede01 TH |
993 | return; |
994 | ||
995 | /* are there still in-flight works? */ | |
112202d9 | 996 | if (pwq->nr_in_flight[color]) |
bf4ede01 TH |
997 | return; |
998 | ||
112202d9 TH |
999 | /* this pwq is done, clear flush_color */ |
1000 | pwq->flush_color = -1; | |
bf4ede01 TH |
1001 | |
1002 | /* | |
112202d9 | 1003 | * If this was the last pwq, wake up the first flusher. It |
bf4ede01 TH |
1004 | * will handle the rest. |
1005 | */ | |
112202d9 TH |
1006 | if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush)) |
1007 | complete(&pwq->wq->first_flusher->done); | |
bf4ede01 TH |
1008 | } |
1009 | ||
36e227d2 | 1010 | /** |
bbb68dfa | 1011 | * try_to_grab_pending - steal work item from worklist and disable irq |
36e227d2 TH |
1012 | * @work: work item to steal |
1013 | * @is_dwork: @work is a delayed_work | |
bbb68dfa | 1014 | * @flags: place to store irq state |
36e227d2 TH |
1015 | * |
1016 | * Try to grab PENDING bit of @work. This function can handle @work in any | |
1017 | * stable state - idle, on timer or on worklist. Return values are | |
1018 | * | |
1019 | * 1 if @work was pending and we successfully stole PENDING | |
1020 | * 0 if @work was idle and we claimed PENDING | |
1021 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry | |
bbb68dfa TH |
1022 | * -ENOENT if someone else is canceling @work, this state may persist |
1023 | * for arbitrarily long | |
36e227d2 | 1024 | * |
bbb68dfa | 1025 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
e0aecdd8 TH |
1026 | * interrupted while holding PENDING and @work off queue, irq must be |
1027 | * disabled on entry. This, combined with delayed_work->timer being | |
1028 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. | |
bbb68dfa TH |
1029 | * |
1030 | * On successful return, >= 0, irq is disabled and the caller is | |
1031 | * responsible for releasing it using local_irq_restore(*@flags). | |
1032 | * | |
e0aecdd8 | 1033 | * This function is safe to call from any context including IRQ handler. |
bf4ede01 | 1034 | */ |
bbb68dfa TH |
1035 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
1036 | unsigned long *flags) | |
bf4ede01 | 1037 | { |
d565ed63 | 1038 | struct worker_pool *pool; |
112202d9 | 1039 | struct pool_workqueue *pwq; |
bf4ede01 | 1040 | |
bbb68dfa TH |
1041 | local_irq_save(*flags); |
1042 | ||
36e227d2 TH |
1043 | /* try to steal the timer if it exists */ |
1044 | if (is_dwork) { | |
1045 | struct delayed_work *dwork = to_delayed_work(work); | |
1046 | ||
e0aecdd8 TH |
1047 | /* |
1048 | * dwork->timer is irqsafe. If del_timer() fails, it's | |
1049 | * guaranteed that the timer is not queued anywhere and not | |
1050 | * running on the local CPU. | |
1051 | */ | |
36e227d2 TH |
1052 | if (likely(del_timer(&dwork->timer))) |
1053 | return 1; | |
1054 | } | |
1055 | ||
1056 | /* try to claim PENDING the normal way */ | |
bf4ede01 TH |
1057 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1058 | return 0; | |
1059 | ||
1060 | /* | |
1061 | * The queueing is in progress, or it is already queued. Try to | |
1062 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1063 | */ | |
d565ed63 TH |
1064 | pool = get_work_pool(work); |
1065 | if (!pool) | |
bbb68dfa | 1066 | goto fail; |
bf4ede01 | 1067 | |
d565ed63 | 1068 | spin_lock(&pool->lock); |
0b3dae68 | 1069 | /* |
112202d9 TH |
1070 | * work->data is guaranteed to point to pwq only while the work |
1071 | * item is queued on pwq->wq, and both updating work->data to point | |
1072 | * to pwq on queueing and to pool on dequeueing are done under | |
1073 | * pwq->pool->lock. This in turn guarantees that, if work->data | |
1074 | * points to pwq which is associated with a locked pool, the work | |
0b3dae68 LJ |
1075 | * item is currently queued on that pool. |
1076 | */ | |
112202d9 TH |
1077 | pwq = get_work_pwq(work); |
1078 | if (pwq && pwq->pool == pool) { | |
16062836 TH |
1079 | debug_work_deactivate(work); |
1080 | ||
1081 | /* | |
1082 | * A delayed work item cannot be grabbed directly because | |
1083 | * it might have linked NO_COLOR work items which, if left | |
112202d9 | 1084 | * on the delayed_list, will confuse pwq->nr_active |
16062836 TH |
1085 | * management later on and cause stall. Make sure the work |
1086 | * item is activated before grabbing. | |
1087 | */ | |
1088 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) | |
112202d9 | 1089 | pwq_activate_delayed_work(work); |
16062836 TH |
1090 | |
1091 | list_del_init(&work->entry); | |
112202d9 | 1092 | pwq_dec_nr_in_flight(get_work_pwq(work), get_work_color(work)); |
16062836 | 1093 | |
112202d9 | 1094 | /* work->data points to pwq iff queued, point to pool */ |
16062836 TH |
1095 | set_work_pool_and_keep_pending(work, pool->id); |
1096 | ||
1097 | spin_unlock(&pool->lock); | |
1098 | return 1; | |
bf4ede01 | 1099 | } |
d565ed63 | 1100 | spin_unlock(&pool->lock); |
bbb68dfa TH |
1101 | fail: |
1102 | local_irq_restore(*flags); | |
1103 | if (work_is_canceling(work)) | |
1104 | return -ENOENT; | |
1105 | cpu_relax(); | |
36e227d2 | 1106 | return -EAGAIN; |
bf4ede01 TH |
1107 | } |
1108 | ||
4690c4ab | 1109 | /** |
706026c2 | 1110 | * insert_work - insert a work into a pool |
112202d9 | 1111 | * @pwq: pwq @work belongs to |
4690c4ab TH |
1112 | * @work: work to insert |
1113 | * @head: insertion point | |
1114 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
1115 | * | |
112202d9 | 1116 | * Insert @work which belongs to @pwq after @head. @extra_flags is or'd to |
706026c2 | 1117 | * work_struct flags. |
4690c4ab TH |
1118 | * |
1119 | * CONTEXT: | |
d565ed63 | 1120 | * spin_lock_irq(pool->lock). |
4690c4ab | 1121 | */ |
112202d9 TH |
1122 | static void insert_work(struct pool_workqueue *pwq, struct work_struct *work, |
1123 | struct list_head *head, unsigned int extra_flags) | |
b89deed3 | 1124 | { |
112202d9 | 1125 | struct worker_pool *pool = pwq->pool; |
e22bee78 | 1126 | |
4690c4ab | 1127 | /* we own @work, set data and link */ |
112202d9 | 1128 | set_work_pwq(work, pwq, extra_flags); |
1a4d9b0a | 1129 | list_add_tail(&work->entry, head); |
e22bee78 TH |
1130 | |
1131 | /* | |
1132 | * Ensure either worker_sched_deactivated() sees the above | |
1133 | * list_add_tail() or we see zero nr_running to avoid workers | |
1134 | * lying around lazily while there are works to be processed. | |
1135 | */ | |
1136 | smp_mb(); | |
1137 | ||
63d95a91 TH |
1138 | if (__need_more_worker(pool)) |
1139 | wake_up_worker(pool); | |
b89deed3 ON |
1140 | } |
1141 | ||
c8efcc25 TH |
1142 | /* |
1143 | * Test whether @work is being queued from another work executing on the | |
8d03ecfe | 1144 | * same workqueue. |
c8efcc25 TH |
1145 | */ |
1146 | static bool is_chained_work(struct workqueue_struct *wq) | |
1147 | { | |
8d03ecfe TH |
1148 | struct worker *worker; |
1149 | ||
1150 | worker = current_wq_worker(); | |
1151 | /* | |
1152 | * Return %true iff I'm a worker execuing a work item on @wq. If | |
1153 | * I'm @worker, it's safe to dereference it without locking. | |
1154 | */ | |
112202d9 | 1155 | return worker && worker->current_pwq->wq == wq; |
c8efcc25 TH |
1156 | } |
1157 | ||
d84ff051 | 1158 | static void __queue_work(int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
1159 | struct work_struct *work) |
1160 | { | |
112202d9 | 1161 | struct pool_workqueue *pwq; |
1e19ffc6 | 1162 | struct list_head *worklist; |
8a2e8e5d | 1163 | unsigned int work_flags; |
b75cac93 | 1164 | unsigned int req_cpu = cpu; |
8930caba TH |
1165 | |
1166 | /* | |
1167 | * While a work item is PENDING && off queue, a task trying to | |
1168 | * steal the PENDING will busy-loop waiting for it to either get | |
1169 | * queued or lose PENDING. Grabbing PENDING and queueing should | |
1170 | * happen with IRQ disabled. | |
1171 | */ | |
1172 | WARN_ON_ONCE(!irqs_disabled()); | |
1da177e4 | 1173 | |
dc186ad7 | 1174 | debug_work_activate(work); |
1e19ffc6 | 1175 | |
c8efcc25 | 1176 | /* if dying, only works from the same workqueue are allowed */ |
9c5a2ba7 | 1177 | if (unlikely(wq->flags & WQ_DRAINING) && |
c8efcc25 | 1178 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b TH |
1179 | return; |
1180 | ||
112202d9 | 1181 | /* determine the pwq to use */ |
c7fc77f7 | 1182 | if (!(wq->flags & WQ_UNBOUND)) { |
c9e7cf27 | 1183 | struct worker_pool *last_pool; |
18aa9eff | 1184 | |
57469821 | 1185 | if (cpu == WORK_CPU_UNBOUND) |
c7fc77f7 TH |
1186 | cpu = raw_smp_processor_id(); |
1187 | ||
18aa9eff | 1188 | /* |
dbf2576e TH |
1189 | * It's multi cpu. If @work was previously on a different |
1190 | * cpu, it might still be running there, in which case the | |
1191 | * work needs to be queued on that cpu to guarantee | |
1192 | * non-reentrancy. | |
18aa9eff | 1193 | */ |
112202d9 | 1194 | pwq = get_pwq(cpu, wq); |
c9e7cf27 | 1195 | last_pool = get_work_pool(work); |
dbf2576e | 1196 | |
112202d9 | 1197 | if (last_pool && last_pool != pwq->pool) { |
18aa9eff TH |
1198 | struct worker *worker; |
1199 | ||
d565ed63 | 1200 | spin_lock(&last_pool->lock); |
18aa9eff | 1201 | |
c9e7cf27 | 1202 | worker = find_worker_executing_work(last_pool, work); |
18aa9eff | 1203 | |
112202d9 TH |
1204 | if (worker && worker->current_pwq->wq == wq) { |
1205 | pwq = get_pwq(last_pool->cpu, wq); | |
8594fade | 1206 | } else { |
18aa9eff | 1207 | /* meh... not running there, queue here */ |
d565ed63 | 1208 | spin_unlock(&last_pool->lock); |
112202d9 | 1209 | spin_lock(&pwq->pool->lock); |
18aa9eff | 1210 | } |
8930caba | 1211 | } else { |
112202d9 | 1212 | spin_lock(&pwq->pool->lock); |
8930caba | 1213 | } |
f3421797 | 1214 | } else { |
112202d9 TH |
1215 | pwq = get_pwq(WORK_CPU_UNBOUND, wq); |
1216 | spin_lock(&pwq->pool->lock); | |
502ca9d8 TH |
1217 | } |
1218 | ||
112202d9 TH |
1219 | /* pwq determined, queue */ |
1220 | trace_workqueue_queue_work(req_cpu, pwq, work); | |
502ca9d8 | 1221 | |
f5b2552b | 1222 | if (WARN_ON(!list_empty(&work->entry))) { |
112202d9 | 1223 | spin_unlock(&pwq->pool->lock); |
f5b2552b DC |
1224 | return; |
1225 | } | |
1e19ffc6 | 1226 | |
112202d9 TH |
1227 | pwq->nr_in_flight[pwq->work_color]++; |
1228 | work_flags = work_color_to_flags(pwq->work_color); | |
1e19ffc6 | 1229 | |
112202d9 | 1230 | if (likely(pwq->nr_active < pwq->max_active)) { |
cdadf009 | 1231 | trace_workqueue_activate_work(work); |
112202d9 TH |
1232 | pwq->nr_active++; |
1233 | worklist = &pwq->pool->worklist; | |
8a2e8e5d TH |
1234 | } else { |
1235 | work_flags |= WORK_STRUCT_DELAYED; | |
112202d9 | 1236 | worklist = &pwq->delayed_works; |
8a2e8e5d | 1237 | } |
1e19ffc6 | 1238 | |
112202d9 | 1239 | insert_work(pwq, work, worklist, work_flags); |
1e19ffc6 | 1240 | |
112202d9 | 1241 | spin_unlock(&pwq->pool->lock); |
1da177e4 LT |
1242 | } |
1243 | ||
0fcb78c2 | 1244 | /** |
c1a220e7 ZR |
1245 | * queue_work_on - queue work on specific cpu |
1246 | * @cpu: CPU number to execute work on | |
0fcb78c2 REB |
1247 | * @wq: workqueue to use |
1248 | * @work: work to queue | |
1249 | * | |
d4283e93 | 1250 | * Returns %false if @work was already on a queue, %true otherwise. |
1da177e4 | 1251 | * |
c1a220e7 ZR |
1252 | * We queue the work to a specific CPU, the caller must ensure it |
1253 | * can't go away. | |
1da177e4 | 1254 | */ |
d4283e93 TH |
1255 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
1256 | struct work_struct *work) | |
1da177e4 | 1257 | { |
d4283e93 | 1258 | bool ret = false; |
8930caba | 1259 | unsigned long flags; |
ef1ca236 | 1260 | |
8930caba | 1261 | local_irq_save(flags); |
c1a220e7 | 1262 | |
22df02bb | 1263 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1264 | __queue_work(cpu, wq, work); |
d4283e93 | 1265 | ret = true; |
c1a220e7 | 1266 | } |
ef1ca236 | 1267 | |
8930caba | 1268 | local_irq_restore(flags); |
1da177e4 LT |
1269 | return ret; |
1270 | } | |
c1a220e7 | 1271 | EXPORT_SYMBOL_GPL(queue_work_on); |
1da177e4 | 1272 | |
c1a220e7 | 1273 | /** |
0a13c00e | 1274 | * queue_work - queue work on a workqueue |
c1a220e7 ZR |
1275 | * @wq: workqueue to use |
1276 | * @work: work to queue | |
1277 | * | |
d4283e93 | 1278 | * Returns %false if @work was already on a queue, %true otherwise. |
c1a220e7 | 1279 | * |
0a13c00e TH |
1280 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
1281 | * it can be processed by another CPU. | |
c1a220e7 | 1282 | */ |
d4283e93 | 1283 | bool queue_work(struct workqueue_struct *wq, struct work_struct *work) |
c1a220e7 | 1284 | { |
57469821 | 1285 | return queue_work_on(WORK_CPU_UNBOUND, wq, work); |
c1a220e7 | 1286 | } |
0a13c00e | 1287 | EXPORT_SYMBOL_GPL(queue_work); |
c1a220e7 | 1288 | |
d8e794df | 1289 | void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1290 | { |
52bad64d | 1291 | struct delayed_work *dwork = (struct delayed_work *)__data; |
1da177e4 | 1292 | |
e0aecdd8 | 1293 | /* should have been called from irqsafe timer with irq already off */ |
60c057bc | 1294 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
1da177e4 | 1295 | } |
1438ade5 | 1296 | EXPORT_SYMBOL(delayed_work_timer_fn); |
1da177e4 | 1297 | |
7beb2edf TH |
1298 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
1299 | struct delayed_work *dwork, unsigned long delay) | |
1da177e4 | 1300 | { |
7beb2edf TH |
1301 | struct timer_list *timer = &dwork->timer; |
1302 | struct work_struct *work = &dwork->work; | |
7beb2edf TH |
1303 | |
1304 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || | |
1305 | timer->data != (unsigned long)dwork); | |
fc4b514f TH |
1306 | WARN_ON_ONCE(timer_pending(timer)); |
1307 | WARN_ON_ONCE(!list_empty(&work->entry)); | |
7beb2edf | 1308 | |
8852aac2 TH |
1309 | /* |
1310 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1311 | * both optimization and correctness. The earliest @timer can | |
1312 | * expire is on the closest next tick and delayed_work users depend | |
1313 | * on that there's no such delay when @delay is 0. | |
1314 | */ | |
1315 | if (!delay) { | |
1316 | __queue_work(cpu, wq, &dwork->work); | |
1317 | return; | |
1318 | } | |
1319 | ||
7beb2edf | 1320 | timer_stats_timer_set_start_info(&dwork->timer); |
1da177e4 | 1321 | |
60c057bc | 1322 | dwork->wq = wq; |
1265057f | 1323 | dwork->cpu = cpu; |
7beb2edf TH |
1324 | timer->expires = jiffies + delay; |
1325 | ||
1326 | if (unlikely(cpu != WORK_CPU_UNBOUND)) | |
1327 | add_timer_on(timer, cpu); | |
1328 | else | |
1329 | add_timer(timer); | |
1da177e4 LT |
1330 | } |
1331 | ||
0fcb78c2 REB |
1332 | /** |
1333 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1334 | * @cpu: CPU number to execute work on | |
1335 | * @wq: workqueue to use | |
af9997e4 | 1336 | * @dwork: work to queue |
0fcb78c2 REB |
1337 | * @delay: number of jiffies to wait before queueing |
1338 | * | |
715f1300 TH |
1339 | * Returns %false if @work was already on a queue, %true otherwise. If |
1340 | * @delay is zero and @dwork is idle, it will be scheduled for immediate | |
1341 | * execution. | |
0fcb78c2 | 1342 | */ |
d4283e93 TH |
1343 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
1344 | struct delayed_work *dwork, unsigned long delay) | |
7a6bc1cd | 1345 | { |
52bad64d | 1346 | struct work_struct *work = &dwork->work; |
d4283e93 | 1347 | bool ret = false; |
8930caba | 1348 | unsigned long flags; |
7a6bc1cd | 1349 | |
8930caba TH |
1350 | /* read the comment in __queue_work() */ |
1351 | local_irq_save(flags); | |
7a6bc1cd | 1352 | |
22df02bb | 1353 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7beb2edf | 1354 | __queue_delayed_work(cpu, wq, dwork, delay); |
d4283e93 | 1355 | ret = true; |
7a6bc1cd | 1356 | } |
8a3e77cc | 1357 | |
8930caba | 1358 | local_irq_restore(flags); |
7a6bc1cd VP |
1359 | return ret; |
1360 | } | |
ae90dd5d | 1361 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
c7fc77f7 | 1362 | |
0a13c00e TH |
1363 | /** |
1364 | * queue_delayed_work - queue work on a workqueue after delay | |
1365 | * @wq: workqueue to use | |
1366 | * @dwork: delayable work to queue | |
1367 | * @delay: number of jiffies to wait before queueing | |
1368 | * | |
715f1300 | 1369 | * Equivalent to queue_delayed_work_on() but tries to use the local CPU. |
0a13c00e | 1370 | */ |
d4283e93 | 1371 | bool queue_delayed_work(struct workqueue_struct *wq, |
0a13c00e TH |
1372 | struct delayed_work *dwork, unsigned long delay) |
1373 | { | |
57469821 | 1374 | return queue_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); |
0a13c00e TH |
1375 | } |
1376 | EXPORT_SYMBOL_GPL(queue_delayed_work); | |
c7fc77f7 | 1377 | |
8376fe22 TH |
1378 | /** |
1379 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU | |
1380 | * @cpu: CPU number to execute work on | |
1381 | * @wq: workqueue to use | |
1382 | * @dwork: work to queue | |
1383 | * @delay: number of jiffies to wait before queueing | |
1384 | * | |
1385 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, | |
1386 | * modify @dwork's timer so that it expires after @delay. If @delay is | |
1387 | * zero, @work is guaranteed to be scheduled immediately regardless of its | |
1388 | * current state. | |
1389 | * | |
1390 | * Returns %false if @dwork was idle and queued, %true if @dwork was | |
1391 | * pending and its timer was modified. | |
1392 | * | |
e0aecdd8 | 1393 | * This function is safe to call from any context including IRQ handler. |
8376fe22 TH |
1394 | * See try_to_grab_pending() for details. |
1395 | */ | |
1396 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
1397 | struct delayed_work *dwork, unsigned long delay) | |
1398 | { | |
1399 | unsigned long flags; | |
1400 | int ret; | |
c7fc77f7 | 1401 | |
8376fe22 TH |
1402 | do { |
1403 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
1404 | } while (unlikely(ret == -EAGAIN)); | |
63bc0362 | 1405 | |
8376fe22 TH |
1406 | if (likely(ret >= 0)) { |
1407 | __queue_delayed_work(cpu, wq, dwork, delay); | |
1408 | local_irq_restore(flags); | |
7a6bc1cd | 1409 | } |
8376fe22 TH |
1410 | |
1411 | /* -ENOENT from try_to_grab_pending() becomes %true */ | |
7a6bc1cd VP |
1412 | return ret; |
1413 | } | |
8376fe22 TH |
1414 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
1415 | ||
1416 | /** | |
1417 | * mod_delayed_work - modify delay of or queue a delayed work | |
1418 | * @wq: workqueue to use | |
1419 | * @dwork: work to queue | |
1420 | * @delay: number of jiffies to wait before queueing | |
1421 | * | |
1422 | * mod_delayed_work_on() on local CPU. | |
1423 | */ | |
1424 | bool mod_delayed_work(struct workqueue_struct *wq, struct delayed_work *dwork, | |
1425 | unsigned long delay) | |
1426 | { | |
1427 | return mod_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); | |
1428 | } | |
1429 | EXPORT_SYMBOL_GPL(mod_delayed_work); | |
1da177e4 | 1430 | |
c8e55f36 TH |
1431 | /** |
1432 | * worker_enter_idle - enter idle state | |
1433 | * @worker: worker which is entering idle state | |
1434 | * | |
1435 | * @worker is entering idle state. Update stats and idle timer if | |
1436 | * necessary. | |
1437 | * | |
1438 | * LOCKING: | |
d565ed63 | 1439 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1440 | */ |
1441 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1442 | { |
bd7bdd43 | 1443 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1444 | |
6183c009 TH |
1445 | if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) || |
1446 | WARN_ON_ONCE(!list_empty(&worker->entry) && | |
1447 | (worker->hentry.next || worker->hentry.pprev))) | |
1448 | return; | |
c8e55f36 | 1449 | |
cb444766 TH |
1450 | /* can't use worker_set_flags(), also called from start_worker() */ |
1451 | worker->flags |= WORKER_IDLE; | |
bd7bdd43 | 1452 | pool->nr_idle++; |
e22bee78 | 1453 | worker->last_active = jiffies; |
c8e55f36 TH |
1454 | |
1455 | /* idle_list is LIFO */ | |
bd7bdd43 | 1456 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1457 | |
628c78e7 TH |
1458 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
1459 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); | |
cb444766 | 1460 | |
544ecf31 | 1461 | /* |
706026c2 | 1462 | * Sanity check nr_running. Because wq_unbind_fn() releases |
d565ed63 | 1463 | * pool->lock between setting %WORKER_UNBOUND and zapping |
628c78e7 TH |
1464 | * nr_running, the warning may trigger spuriously. Check iff |
1465 | * unbind is not in progress. | |
544ecf31 | 1466 | */ |
24647570 | 1467 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
bd7bdd43 | 1468 | pool->nr_workers == pool->nr_idle && |
e19e397a | 1469 | atomic_read(&pool->nr_running)); |
c8e55f36 TH |
1470 | } |
1471 | ||
1472 | /** | |
1473 | * worker_leave_idle - leave idle state | |
1474 | * @worker: worker which is leaving idle state | |
1475 | * | |
1476 | * @worker is leaving idle state. Update stats. | |
1477 | * | |
1478 | * LOCKING: | |
d565ed63 | 1479 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1480 | */ |
1481 | static void worker_leave_idle(struct worker *worker) | |
1482 | { | |
bd7bdd43 | 1483 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1484 | |
6183c009 TH |
1485 | if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE))) |
1486 | return; | |
d302f017 | 1487 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1488 | pool->nr_idle--; |
c8e55f36 TH |
1489 | list_del_init(&worker->entry); |
1490 | } | |
1491 | ||
e22bee78 | 1492 | /** |
f36dc67b LJ |
1493 | * worker_maybe_bind_and_lock - try to bind %current to worker_pool and lock it |
1494 | * @pool: target worker_pool | |
1495 | * | |
1496 | * Bind %current to the cpu of @pool if it is associated and lock @pool. | |
e22bee78 TH |
1497 | * |
1498 | * Works which are scheduled while the cpu is online must at least be | |
1499 | * scheduled to a worker which is bound to the cpu so that if they are | |
1500 | * flushed from cpu callbacks while cpu is going down, they are | |
1501 | * guaranteed to execute on the cpu. | |
1502 | * | |
f5faa077 | 1503 | * This function is to be used by unbound workers and rescuers to bind |
e22bee78 TH |
1504 | * themselves to the target cpu and may race with cpu going down or |
1505 | * coming online. kthread_bind() can't be used because it may put the | |
1506 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
706026c2 | 1507 | * verbatim as it's best effort and blocking and pool may be |
e22bee78 TH |
1508 | * [dis]associated in the meantime. |
1509 | * | |
706026c2 | 1510 | * This function tries set_cpus_allowed() and locks pool and verifies the |
24647570 | 1511 | * binding against %POOL_DISASSOCIATED which is set during |
f2d5a0ee TH |
1512 | * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker |
1513 | * enters idle state or fetches works without dropping lock, it can | |
1514 | * guarantee the scheduling requirement described in the first paragraph. | |
e22bee78 TH |
1515 | * |
1516 | * CONTEXT: | |
d565ed63 | 1517 | * Might sleep. Called without any lock but returns with pool->lock |
e22bee78 TH |
1518 | * held. |
1519 | * | |
1520 | * RETURNS: | |
706026c2 | 1521 | * %true if the associated pool is online (@worker is successfully |
e22bee78 TH |
1522 | * bound), %false if offline. |
1523 | */ | |
f36dc67b | 1524 | static bool worker_maybe_bind_and_lock(struct worker_pool *pool) |
d565ed63 | 1525 | __acquires(&pool->lock) |
e22bee78 | 1526 | { |
e22bee78 | 1527 | while (true) { |
4e6045f1 | 1528 | /* |
e22bee78 TH |
1529 | * The following call may fail, succeed or succeed |
1530 | * without actually migrating the task to the cpu if | |
1531 | * it races with cpu hotunplug operation. Verify | |
24647570 | 1532 | * against POOL_DISASSOCIATED. |
4e6045f1 | 1533 | */ |
24647570 | 1534 | if (!(pool->flags & POOL_DISASSOCIATED)) |
f5faa077 | 1535 | set_cpus_allowed_ptr(current, get_cpu_mask(pool->cpu)); |
e22bee78 | 1536 | |
d565ed63 | 1537 | spin_lock_irq(&pool->lock); |
24647570 | 1538 | if (pool->flags & POOL_DISASSOCIATED) |
e22bee78 | 1539 | return false; |
f5faa077 | 1540 | if (task_cpu(current) == pool->cpu && |
e22bee78 | 1541 | cpumask_equal(¤t->cpus_allowed, |
ec22ca5e | 1542 | get_cpu_mask(pool->cpu))) |
e22bee78 | 1543 | return true; |
d565ed63 | 1544 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1545 | |
5035b20f TH |
1546 | /* |
1547 | * We've raced with CPU hot[un]plug. Give it a breather | |
1548 | * and retry migration. cond_resched() is required here; | |
1549 | * otherwise, we might deadlock against cpu_stop trying to | |
1550 | * bring down the CPU on non-preemptive kernel. | |
1551 | */ | |
e22bee78 | 1552 | cpu_relax(); |
5035b20f | 1553 | cond_resched(); |
e22bee78 TH |
1554 | } |
1555 | } | |
1556 | ||
25511a47 | 1557 | /* |
ea1abd61 | 1558 | * Rebind an idle @worker to its CPU. worker_thread() will test |
5f7dabfd | 1559 | * list_empty(@worker->entry) before leaving idle and call this function. |
25511a47 TH |
1560 | */ |
1561 | static void idle_worker_rebind(struct worker *worker) | |
1562 | { | |
5f7dabfd | 1563 | /* CPU may go down again inbetween, clear UNBOUND only on success */ |
f36dc67b | 1564 | if (worker_maybe_bind_and_lock(worker->pool)) |
5f7dabfd | 1565 | worker_clr_flags(worker, WORKER_UNBOUND); |
25511a47 | 1566 | |
ea1abd61 LJ |
1567 | /* rebind complete, become available again */ |
1568 | list_add(&worker->entry, &worker->pool->idle_list); | |
d565ed63 | 1569 | spin_unlock_irq(&worker->pool->lock); |
25511a47 TH |
1570 | } |
1571 | ||
e22bee78 | 1572 | /* |
25511a47 | 1573 | * Function for @worker->rebind.work used to rebind unbound busy workers to |
403c821d TH |
1574 | * the associated cpu which is coming back online. This is scheduled by |
1575 | * cpu up but can race with other cpu hotplug operations and may be | |
1576 | * executed twice without intervening cpu down. | |
e22bee78 | 1577 | */ |
25511a47 | 1578 | static void busy_worker_rebind_fn(struct work_struct *work) |
e22bee78 TH |
1579 | { |
1580 | struct worker *worker = container_of(work, struct worker, rebind_work); | |
e22bee78 | 1581 | |
f36dc67b | 1582 | if (worker_maybe_bind_and_lock(worker->pool)) |
eab6d828 | 1583 | worker_clr_flags(worker, WORKER_UNBOUND); |
e22bee78 | 1584 | |
d565ed63 | 1585 | spin_unlock_irq(&worker->pool->lock); |
e22bee78 TH |
1586 | } |
1587 | ||
25511a47 | 1588 | /** |
94cf58bb TH |
1589 | * rebind_workers - rebind all workers of a pool to the associated CPU |
1590 | * @pool: pool of interest | |
25511a47 | 1591 | * |
94cf58bb | 1592 | * @pool->cpu is coming online. Rebind all workers to the CPU. Rebinding |
25511a47 TH |
1593 | * is different for idle and busy ones. |
1594 | * | |
ea1abd61 LJ |
1595 | * Idle ones will be removed from the idle_list and woken up. They will |
1596 | * add themselves back after completing rebind. This ensures that the | |
1597 | * idle_list doesn't contain any unbound workers when re-bound busy workers | |
1598 | * try to perform local wake-ups for concurrency management. | |
25511a47 | 1599 | * |
ea1abd61 LJ |
1600 | * Busy workers can rebind after they finish their current work items. |
1601 | * Queueing the rebind work item at the head of the scheduled list is | |
1602 | * enough. Note that nr_running will be properly bumped as busy workers | |
1603 | * rebind. | |
25511a47 | 1604 | * |
ea1abd61 LJ |
1605 | * On return, all non-manager workers are scheduled for rebind - see |
1606 | * manage_workers() for the manager special case. Any idle worker | |
1607 | * including the manager will not appear on @idle_list until rebind is | |
1608 | * complete, making local wake-ups safe. | |
25511a47 | 1609 | */ |
94cf58bb | 1610 | static void rebind_workers(struct worker_pool *pool) |
25511a47 | 1611 | { |
ea1abd61 | 1612 | struct worker *worker, *n; |
25511a47 TH |
1613 | int i; |
1614 | ||
94cf58bb TH |
1615 | lockdep_assert_held(&pool->assoc_mutex); |
1616 | lockdep_assert_held(&pool->lock); | |
25511a47 | 1617 | |
5f7dabfd | 1618 | /* dequeue and kick idle ones */ |
94cf58bb TH |
1619 | list_for_each_entry_safe(worker, n, &pool->idle_list, entry) { |
1620 | /* | |
1621 | * idle workers should be off @pool->idle_list until rebind | |
1622 | * is complete to avoid receiving premature local wake-ups. | |
1623 | */ | |
1624 | list_del_init(&worker->entry); | |
25511a47 | 1625 | |
94cf58bb TH |
1626 | /* |
1627 | * worker_thread() will see the above dequeuing and call | |
1628 | * idle_worker_rebind(). | |
1629 | */ | |
1630 | wake_up_process(worker->task); | |
1631 | } | |
25511a47 | 1632 | |
94cf58bb | 1633 | /* rebind busy workers */ |
b67bfe0d | 1634 | for_each_busy_worker(worker, i, pool) { |
94cf58bb TH |
1635 | struct work_struct *rebind_work = &worker->rebind_work; |
1636 | struct workqueue_struct *wq; | |
25511a47 | 1637 | |
94cf58bb TH |
1638 | if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, |
1639 | work_data_bits(rebind_work))) | |
1640 | continue; | |
25511a47 | 1641 | |
94cf58bb | 1642 | debug_work_activate(rebind_work); |
90beca5d | 1643 | |
94cf58bb TH |
1644 | /* |
1645 | * wq doesn't really matter but let's keep @worker->pool | |
112202d9 | 1646 | * and @pwq->pool consistent for sanity. |
94cf58bb TH |
1647 | */ |
1648 | if (std_worker_pool_pri(worker->pool)) | |
1649 | wq = system_highpri_wq; | |
1650 | else | |
1651 | wq = system_wq; | |
1652 | ||
112202d9 | 1653 | insert_work(get_pwq(pool->cpu, wq), rebind_work, |
94cf58bb TH |
1654 | worker->scheduled.next, |
1655 | work_color_to_flags(WORK_NO_COLOR)); | |
ec58815a | 1656 | } |
25511a47 TH |
1657 | } |
1658 | ||
c34056a3 TH |
1659 | static struct worker *alloc_worker(void) |
1660 | { | |
1661 | struct worker *worker; | |
1662 | ||
1663 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1664 | if (worker) { |
1665 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1666 | INIT_LIST_HEAD(&worker->scheduled); |
25511a47 | 1667 | INIT_WORK(&worker->rebind_work, busy_worker_rebind_fn); |
e22bee78 TH |
1668 | /* on creation a worker is in !idle && prep state */ |
1669 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1670 | } |
c34056a3 TH |
1671 | return worker; |
1672 | } | |
1673 | ||
1674 | /** | |
1675 | * create_worker - create a new workqueue worker | |
63d95a91 | 1676 | * @pool: pool the new worker will belong to |
c34056a3 | 1677 | * |
63d95a91 | 1678 | * Create a new worker which is bound to @pool. The returned worker |
c34056a3 TH |
1679 | * can be started by calling start_worker() or destroyed using |
1680 | * destroy_worker(). | |
1681 | * | |
1682 | * CONTEXT: | |
1683 | * Might sleep. Does GFP_KERNEL allocations. | |
1684 | * | |
1685 | * RETURNS: | |
1686 | * Pointer to the newly created worker. | |
1687 | */ | |
bc2ae0f5 | 1688 | static struct worker *create_worker(struct worker_pool *pool) |
c34056a3 | 1689 | { |
e34cdddb | 1690 | const char *pri = std_worker_pool_pri(pool) ? "H" : ""; |
c34056a3 | 1691 | struct worker *worker = NULL; |
f3421797 | 1692 | int id = -1; |
c34056a3 | 1693 | |
d565ed63 | 1694 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1695 | while (ida_get_new(&pool->worker_ida, &id)) { |
d565ed63 | 1696 | spin_unlock_irq(&pool->lock); |
bd7bdd43 | 1697 | if (!ida_pre_get(&pool->worker_ida, GFP_KERNEL)) |
c34056a3 | 1698 | goto fail; |
d565ed63 | 1699 | spin_lock_irq(&pool->lock); |
c34056a3 | 1700 | } |
d565ed63 | 1701 | spin_unlock_irq(&pool->lock); |
c34056a3 TH |
1702 | |
1703 | worker = alloc_worker(); | |
1704 | if (!worker) | |
1705 | goto fail; | |
1706 | ||
bd7bdd43 | 1707 | worker->pool = pool; |
c34056a3 TH |
1708 | worker->id = id; |
1709 | ||
ec22ca5e | 1710 | if (pool->cpu != WORK_CPU_UNBOUND) |
94dcf29a | 1711 | worker->task = kthread_create_on_node(worker_thread, |
ec22ca5e | 1712 | worker, cpu_to_node(pool->cpu), |
d84ff051 | 1713 | "kworker/%d:%d%s", pool->cpu, id, pri); |
f3421797 TH |
1714 | else |
1715 | worker->task = kthread_create(worker_thread, worker, | |
3270476a | 1716 | "kworker/u:%d%s", id, pri); |
c34056a3 TH |
1717 | if (IS_ERR(worker->task)) |
1718 | goto fail; | |
1719 | ||
e34cdddb | 1720 | if (std_worker_pool_pri(pool)) |
3270476a TH |
1721 | set_user_nice(worker->task, HIGHPRI_NICE_LEVEL); |
1722 | ||
db7bccf4 | 1723 | /* |
bc2ae0f5 | 1724 | * Determine CPU binding of the new worker depending on |
24647570 | 1725 | * %POOL_DISASSOCIATED. The caller is responsible for ensuring the |
bc2ae0f5 TH |
1726 | * flag remains stable across this function. See the comments |
1727 | * above the flag definition for details. | |
1728 | * | |
1729 | * As an unbound worker may later become a regular one if CPU comes | |
1730 | * online, make sure every worker has %PF_THREAD_BOUND set. | |
db7bccf4 | 1731 | */ |
24647570 | 1732 | if (!(pool->flags & POOL_DISASSOCIATED)) { |
ec22ca5e | 1733 | kthread_bind(worker->task, pool->cpu); |
bc2ae0f5 | 1734 | } else { |
db7bccf4 | 1735 | worker->task->flags |= PF_THREAD_BOUND; |
bc2ae0f5 | 1736 | worker->flags |= WORKER_UNBOUND; |
f3421797 | 1737 | } |
c34056a3 TH |
1738 | |
1739 | return worker; | |
1740 | fail: | |
1741 | if (id >= 0) { | |
d565ed63 | 1742 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1743 | ida_remove(&pool->worker_ida, id); |
d565ed63 | 1744 | spin_unlock_irq(&pool->lock); |
c34056a3 TH |
1745 | } |
1746 | kfree(worker); | |
1747 | return NULL; | |
1748 | } | |
1749 | ||
1750 | /** | |
1751 | * start_worker - start a newly created worker | |
1752 | * @worker: worker to start | |
1753 | * | |
706026c2 | 1754 | * Make the pool aware of @worker and start it. |
c34056a3 TH |
1755 | * |
1756 | * CONTEXT: | |
d565ed63 | 1757 | * spin_lock_irq(pool->lock). |
c34056a3 TH |
1758 | */ |
1759 | static void start_worker(struct worker *worker) | |
1760 | { | |
cb444766 | 1761 | worker->flags |= WORKER_STARTED; |
bd7bdd43 | 1762 | worker->pool->nr_workers++; |
c8e55f36 | 1763 | worker_enter_idle(worker); |
c34056a3 TH |
1764 | wake_up_process(worker->task); |
1765 | } | |
1766 | ||
1767 | /** | |
1768 | * destroy_worker - destroy a workqueue worker | |
1769 | * @worker: worker to be destroyed | |
1770 | * | |
706026c2 | 1771 | * Destroy @worker and adjust @pool stats accordingly. |
c8e55f36 TH |
1772 | * |
1773 | * CONTEXT: | |
d565ed63 | 1774 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
c34056a3 TH |
1775 | */ |
1776 | static void destroy_worker(struct worker *worker) | |
1777 | { | |
bd7bdd43 | 1778 | struct worker_pool *pool = worker->pool; |
c34056a3 TH |
1779 | int id = worker->id; |
1780 | ||
1781 | /* sanity check frenzy */ | |
6183c009 TH |
1782 | if (WARN_ON(worker->current_work) || |
1783 | WARN_ON(!list_empty(&worker->scheduled))) | |
1784 | return; | |
c34056a3 | 1785 | |
c8e55f36 | 1786 | if (worker->flags & WORKER_STARTED) |
bd7bdd43 | 1787 | pool->nr_workers--; |
c8e55f36 | 1788 | if (worker->flags & WORKER_IDLE) |
bd7bdd43 | 1789 | pool->nr_idle--; |
c8e55f36 TH |
1790 | |
1791 | list_del_init(&worker->entry); | |
cb444766 | 1792 | worker->flags |= WORKER_DIE; |
c8e55f36 | 1793 | |
d565ed63 | 1794 | spin_unlock_irq(&pool->lock); |
c8e55f36 | 1795 | |
c34056a3 TH |
1796 | kthread_stop(worker->task); |
1797 | kfree(worker); | |
1798 | ||
d565ed63 | 1799 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1800 | ida_remove(&pool->worker_ida, id); |
c34056a3 TH |
1801 | } |
1802 | ||
63d95a91 | 1803 | static void idle_worker_timeout(unsigned long __pool) |
e22bee78 | 1804 | { |
63d95a91 | 1805 | struct worker_pool *pool = (void *)__pool; |
e22bee78 | 1806 | |
d565ed63 | 1807 | spin_lock_irq(&pool->lock); |
e22bee78 | 1808 | |
63d95a91 | 1809 | if (too_many_workers(pool)) { |
e22bee78 TH |
1810 | struct worker *worker; |
1811 | unsigned long expires; | |
1812 | ||
1813 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1814 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1815 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1816 | ||
1817 | if (time_before(jiffies, expires)) | |
63d95a91 | 1818 | mod_timer(&pool->idle_timer, expires); |
e22bee78 TH |
1819 | else { |
1820 | /* it's been idle for too long, wake up manager */ | |
11ebea50 | 1821 | pool->flags |= POOL_MANAGE_WORKERS; |
63d95a91 | 1822 | wake_up_worker(pool); |
d5abe669 | 1823 | } |
e22bee78 TH |
1824 | } |
1825 | ||
d565ed63 | 1826 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1827 | } |
d5abe669 | 1828 | |
493a1724 | 1829 | static void send_mayday(struct work_struct *work) |
e22bee78 | 1830 | { |
112202d9 TH |
1831 | struct pool_workqueue *pwq = get_work_pwq(work); |
1832 | struct workqueue_struct *wq = pwq->wq; | |
493a1724 TH |
1833 | |
1834 | lockdep_assert_held(&workqueue_lock); | |
e22bee78 TH |
1835 | |
1836 | if (!(wq->flags & WQ_RESCUER)) | |
493a1724 | 1837 | return; |
e22bee78 TH |
1838 | |
1839 | /* mayday mayday mayday */ | |
493a1724 TH |
1840 | if (list_empty(&pwq->mayday_node)) { |
1841 | list_add_tail(&pwq->mayday_node, &wq->maydays); | |
e22bee78 | 1842 | wake_up_process(wq->rescuer->task); |
493a1724 | 1843 | } |
e22bee78 TH |
1844 | } |
1845 | ||
706026c2 | 1846 | static void pool_mayday_timeout(unsigned long __pool) |
e22bee78 | 1847 | { |
63d95a91 | 1848 | struct worker_pool *pool = (void *)__pool; |
e22bee78 TH |
1849 | struct work_struct *work; |
1850 | ||
493a1724 TH |
1851 | spin_lock_irq(&workqueue_lock); /* for wq->maydays */ |
1852 | spin_lock(&pool->lock); | |
e22bee78 | 1853 | |
63d95a91 | 1854 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
1855 | /* |
1856 | * We've been trying to create a new worker but | |
1857 | * haven't been successful. We might be hitting an | |
1858 | * allocation deadlock. Send distress signals to | |
1859 | * rescuers. | |
1860 | */ | |
63d95a91 | 1861 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 1862 | send_mayday(work); |
1da177e4 | 1863 | } |
e22bee78 | 1864 | |
493a1724 TH |
1865 | spin_unlock(&pool->lock); |
1866 | spin_unlock_irq(&workqueue_lock); | |
e22bee78 | 1867 | |
63d95a91 | 1868 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
1869 | } |
1870 | ||
e22bee78 TH |
1871 | /** |
1872 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 1873 | * @pool: pool to create a new worker for |
e22bee78 | 1874 | * |
63d95a91 | 1875 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
1876 | * have at least one idle worker on return from this function. If |
1877 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 1878 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
1879 | * possible allocation deadlock. |
1880 | * | |
1881 | * On return, need_to_create_worker() is guaranteed to be false and | |
1882 | * may_start_working() true. | |
1883 | * | |
1884 | * LOCKING: | |
d565ed63 | 1885 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1886 | * multiple times. Does GFP_KERNEL allocations. Called only from |
1887 | * manager. | |
1888 | * | |
1889 | * RETURNS: | |
d565ed63 | 1890 | * false if no action was taken and pool->lock stayed locked, true |
e22bee78 TH |
1891 | * otherwise. |
1892 | */ | |
63d95a91 | 1893 | static bool maybe_create_worker(struct worker_pool *pool) |
d565ed63 TH |
1894 | __releases(&pool->lock) |
1895 | __acquires(&pool->lock) | |
1da177e4 | 1896 | { |
63d95a91 | 1897 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1898 | return false; |
1899 | restart: | |
d565ed63 | 1900 | spin_unlock_irq(&pool->lock); |
9f9c2364 | 1901 | |
e22bee78 | 1902 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 1903 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
1904 | |
1905 | while (true) { | |
1906 | struct worker *worker; | |
1907 | ||
bc2ae0f5 | 1908 | worker = create_worker(pool); |
e22bee78 | 1909 | if (worker) { |
63d95a91 | 1910 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1911 | spin_lock_irq(&pool->lock); |
e22bee78 | 1912 | start_worker(worker); |
6183c009 TH |
1913 | if (WARN_ON_ONCE(need_to_create_worker(pool))) |
1914 | goto restart; | |
e22bee78 TH |
1915 | return true; |
1916 | } | |
1917 | ||
63d95a91 | 1918 | if (!need_to_create_worker(pool)) |
e22bee78 | 1919 | break; |
1da177e4 | 1920 | |
e22bee78 TH |
1921 | __set_current_state(TASK_INTERRUPTIBLE); |
1922 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1923 | |
63d95a91 | 1924 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1925 | break; |
1926 | } | |
1927 | ||
63d95a91 | 1928 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1929 | spin_lock_irq(&pool->lock); |
63d95a91 | 1930 | if (need_to_create_worker(pool)) |
e22bee78 TH |
1931 | goto restart; |
1932 | return true; | |
1933 | } | |
1934 | ||
1935 | /** | |
1936 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
63d95a91 | 1937 | * @pool: pool to destroy workers for |
e22bee78 | 1938 | * |
63d95a91 | 1939 | * Destroy @pool workers which have been idle for longer than |
e22bee78 TH |
1940 | * IDLE_WORKER_TIMEOUT. |
1941 | * | |
1942 | * LOCKING: | |
d565ed63 | 1943 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1944 | * multiple times. Called only from manager. |
1945 | * | |
1946 | * RETURNS: | |
d565ed63 | 1947 | * false if no action was taken and pool->lock stayed locked, true |
e22bee78 TH |
1948 | * otherwise. |
1949 | */ | |
63d95a91 | 1950 | static bool maybe_destroy_workers(struct worker_pool *pool) |
e22bee78 TH |
1951 | { |
1952 | bool ret = false; | |
1da177e4 | 1953 | |
63d95a91 | 1954 | while (too_many_workers(pool)) { |
e22bee78 TH |
1955 | struct worker *worker; |
1956 | unsigned long expires; | |
3af24433 | 1957 | |
63d95a91 | 1958 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 | 1959 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
85f4186a | 1960 | |
e22bee78 | 1961 | if (time_before(jiffies, expires)) { |
63d95a91 | 1962 | mod_timer(&pool->idle_timer, expires); |
3af24433 | 1963 | break; |
e22bee78 | 1964 | } |
1da177e4 | 1965 | |
e22bee78 TH |
1966 | destroy_worker(worker); |
1967 | ret = true; | |
1da177e4 | 1968 | } |
1e19ffc6 | 1969 | |
e22bee78 | 1970 | return ret; |
1e19ffc6 TH |
1971 | } |
1972 | ||
73f53c4a | 1973 | /** |
e22bee78 TH |
1974 | * manage_workers - manage worker pool |
1975 | * @worker: self | |
73f53c4a | 1976 | * |
706026c2 | 1977 | * Assume the manager role and manage the worker pool @worker belongs |
e22bee78 | 1978 | * to. At any given time, there can be only zero or one manager per |
706026c2 | 1979 | * pool. The exclusion is handled automatically by this function. |
e22bee78 TH |
1980 | * |
1981 | * The caller can safely start processing works on false return. On | |
1982 | * true return, it's guaranteed that need_to_create_worker() is false | |
1983 | * and may_start_working() is true. | |
73f53c4a TH |
1984 | * |
1985 | * CONTEXT: | |
d565ed63 | 1986 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1987 | * multiple times. Does GFP_KERNEL allocations. |
1988 | * | |
1989 | * RETURNS: | |
d565ed63 TH |
1990 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
1991 | * multiple times. Does GFP_KERNEL allocations. | |
73f53c4a | 1992 | */ |
e22bee78 | 1993 | static bool manage_workers(struct worker *worker) |
73f53c4a | 1994 | { |
63d95a91 | 1995 | struct worker_pool *pool = worker->pool; |
e22bee78 | 1996 | bool ret = false; |
73f53c4a | 1997 | |
ee378aa4 | 1998 | if (pool->flags & POOL_MANAGING_WORKERS) |
e22bee78 | 1999 | return ret; |
1e19ffc6 | 2000 | |
552a37e9 | 2001 | pool->flags |= POOL_MANAGING_WORKERS; |
73f53c4a | 2002 | |
ee378aa4 LJ |
2003 | /* |
2004 | * To simplify both worker management and CPU hotplug, hold off | |
2005 | * management while hotplug is in progress. CPU hotplug path can't | |
2006 | * grab %POOL_MANAGING_WORKERS to achieve this because that can | |
2007 | * lead to idle worker depletion (all become busy thinking someone | |
2008 | * else is managing) which in turn can result in deadlock under | |
b2eb83d1 | 2009 | * extreme circumstances. Use @pool->assoc_mutex to synchronize |
ee378aa4 LJ |
2010 | * manager against CPU hotplug. |
2011 | * | |
b2eb83d1 | 2012 | * assoc_mutex would always be free unless CPU hotplug is in |
d565ed63 | 2013 | * progress. trylock first without dropping @pool->lock. |
ee378aa4 | 2014 | */ |
b2eb83d1 | 2015 | if (unlikely(!mutex_trylock(&pool->assoc_mutex))) { |
d565ed63 | 2016 | spin_unlock_irq(&pool->lock); |
b2eb83d1 | 2017 | mutex_lock(&pool->assoc_mutex); |
ee378aa4 LJ |
2018 | /* |
2019 | * CPU hotplug could have happened while we were waiting | |
b2eb83d1 | 2020 | * for assoc_mutex. Hotplug itself can't handle us |
ee378aa4 | 2021 | * because manager isn't either on idle or busy list, and |
706026c2 | 2022 | * @pool's state and ours could have deviated. |
ee378aa4 | 2023 | * |
b2eb83d1 | 2024 | * As hotplug is now excluded via assoc_mutex, we can |
ee378aa4 | 2025 | * simply try to bind. It will succeed or fail depending |
706026c2 | 2026 | * on @pool's current state. Try it and adjust |
ee378aa4 LJ |
2027 | * %WORKER_UNBOUND accordingly. |
2028 | */ | |
f36dc67b | 2029 | if (worker_maybe_bind_and_lock(pool)) |
ee378aa4 LJ |
2030 | worker->flags &= ~WORKER_UNBOUND; |
2031 | else | |
2032 | worker->flags |= WORKER_UNBOUND; | |
73f53c4a | 2033 | |
ee378aa4 LJ |
2034 | ret = true; |
2035 | } | |
73f53c4a | 2036 | |
11ebea50 | 2037 | pool->flags &= ~POOL_MANAGE_WORKERS; |
73f53c4a TH |
2038 | |
2039 | /* | |
e22bee78 TH |
2040 | * Destroy and then create so that may_start_working() is true |
2041 | * on return. | |
73f53c4a | 2042 | */ |
63d95a91 TH |
2043 | ret |= maybe_destroy_workers(pool); |
2044 | ret |= maybe_create_worker(pool); | |
e22bee78 | 2045 | |
552a37e9 | 2046 | pool->flags &= ~POOL_MANAGING_WORKERS; |
b2eb83d1 | 2047 | mutex_unlock(&pool->assoc_mutex); |
e22bee78 | 2048 | return ret; |
73f53c4a TH |
2049 | } |
2050 | ||
a62428c0 TH |
2051 | /** |
2052 | * process_one_work - process single work | |
c34056a3 | 2053 | * @worker: self |
a62428c0 TH |
2054 | * @work: work to process |
2055 | * | |
2056 | * Process @work. This function contains all the logics necessary to | |
2057 | * process a single work including synchronization against and | |
2058 | * interaction with other workers on the same cpu, queueing and | |
2059 | * flushing. As long as context requirement is met, any worker can | |
2060 | * call this function to process a work. | |
2061 | * | |
2062 | * CONTEXT: | |
d565ed63 | 2063 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
a62428c0 | 2064 | */ |
c34056a3 | 2065 | static void process_one_work(struct worker *worker, struct work_struct *work) |
d565ed63 TH |
2066 | __releases(&pool->lock) |
2067 | __acquires(&pool->lock) | |
a62428c0 | 2068 | { |
112202d9 | 2069 | struct pool_workqueue *pwq = get_work_pwq(work); |
bd7bdd43 | 2070 | struct worker_pool *pool = worker->pool; |
112202d9 | 2071 | bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE; |
73f53c4a | 2072 | int work_color; |
7e11629d | 2073 | struct worker *collision; |
a62428c0 TH |
2074 | #ifdef CONFIG_LOCKDEP |
2075 | /* | |
2076 | * It is permissible to free the struct work_struct from | |
2077 | * inside the function that is called from it, this we need to | |
2078 | * take into account for lockdep too. To avoid bogus "held | |
2079 | * lock freed" warnings as well as problems when looking into | |
2080 | * work->lockdep_map, make a copy and use that here. | |
2081 | */ | |
4d82a1de PZ |
2082 | struct lockdep_map lockdep_map; |
2083 | ||
2084 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 2085 | #endif |
6fec10a1 TH |
2086 | /* |
2087 | * Ensure we're on the correct CPU. DISASSOCIATED test is | |
2088 | * necessary to avoid spurious warnings from rescuers servicing the | |
24647570 | 2089 | * unbound or a disassociated pool. |
6fec10a1 | 2090 | */ |
5f7dabfd | 2091 | WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) && |
24647570 | 2092 | !(pool->flags & POOL_DISASSOCIATED) && |
ec22ca5e | 2093 | raw_smp_processor_id() != pool->cpu); |
25511a47 | 2094 | |
7e11629d TH |
2095 | /* |
2096 | * A single work shouldn't be executed concurrently by | |
2097 | * multiple workers on a single cpu. Check whether anyone is | |
2098 | * already processing the work. If so, defer the work to the | |
2099 | * currently executing one. | |
2100 | */ | |
c9e7cf27 | 2101 | collision = find_worker_executing_work(pool, work); |
7e11629d TH |
2102 | if (unlikely(collision)) { |
2103 | move_linked_works(work, &collision->scheduled, NULL); | |
2104 | return; | |
2105 | } | |
2106 | ||
8930caba | 2107 | /* claim and dequeue */ |
a62428c0 | 2108 | debug_work_deactivate(work); |
c9e7cf27 | 2109 | hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work); |
c34056a3 | 2110 | worker->current_work = work; |
a2c1c57b | 2111 | worker->current_func = work->func; |
112202d9 | 2112 | worker->current_pwq = pwq; |
73f53c4a | 2113 | work_color = get_work_color(work); |
7a22ad75 | 2114 | |
a62428c0 TH |
2115 | list_del_init(&work->entry); |
2116 | ||
fb0e7beb TH |
2117 | /* |
2118 | * CPU intensive works don't participate in concurrency | |
2119 | * management. They're the scheduler's responsibility. | |
2120 | */ | |
2121 | if (unlikely(cpu_intensive)) | |
2122 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
2123 | ||
974271c4 | 2124 | /* |
d565ed63 | 2125 | * Unbound pool isn't concurrency managed and work items should be |
974271c4 TH |
2126 | * executed ASAP. Wake up another worker if necessary. |
2127 | */ | |
63d95a91 TH |
2128 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
2129 | wake_up_worker(pool); | |
974271c4 | 2130 | |
8930caba | 2131 | /* |
7c3eed5c | 2132 | * Record the last pool and clear PENDING which should be the last |
d565ed63 | 2133 | * update to @work. Also, do this inside @pool->lock so that |
23657bb1 TH |
2134 | * PENDING and queued state changes happen together while IRQ is |
2135 | * disabled. | |
8930caba | 2136 | */ |
7c3eed5c | 2137 | set_work_pool_and_clear_pending(work, pool->id); |
a62428c0 | 2138 | |
d565ed63 | 2139 | spin_unlock_irq(&pool->lock); |
a62428c0 | 2140 | |
112202d9 | 2141 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
a62428c0 | 2142 | lock_map_acquire(&lockdep_map); |
e36c886a | 2143 | trace_workqueue_execute_start(work); |
a2c1c57b | 2144 | worker->current_func(work); |
e36c886a AV |
2145 | /* |
2146 | * While we must be careful to not use "work" after this, the trace | |
2147 | * point will only record its address. | |
2148 | */ | |
2149 | trace_workqueue_execute_end(work); | |
a62428c0 | 2150 | lock_map_release(&lockdep_map); |
112202d9 | 2151 | lock_map_release(&pwq->wq->lockdep_map); |
a62428c0 TH |
2152 | |
2153 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
044c782c VI |
2154 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
2155 | " last function: %pf\n", | |
a2c1c57b TH |
2156 | current->comm, preempt_count(), task_pid_nr(current), |
2157 | worker->current_func); | |
a62428c0 TH |
2158 | debug_show_held_locks(current); |
2159 | dump_stack(); | |
2160 | } | |
2161 | ||
d565ed63 | 2162 | spin_lock_irq(&pool->lock); |
a62428c0 | 2163 | |
fb0e7beb TH |
2164 | /* clear cpu intensive status */ |
2165 | if (unlikely(cpu_intensive)) | |
2166 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
2167 | ||
a62428c0 | 2168 | /* we're done with it, release */ |
42f8570f | 2169 | hash_del(&worker->hentry); |
c34056a3 | 2170 | worker->current_work = NULL; |
a2c1c57b | 2171 | worker->current_func = NULL; |
112202d9 TH |
2172 | worker->current_pwq = NULL; |
2173 | pwq_dec_nr_in_flight(pwq, work_color); | |
a62428c0 TH |
2174 | } |
2175 | ||
affee4b2 TH |
2176 | /** |
2177 | * process_scheduled_works - process scheduled works | |
2178 | * @worker: self | |
2179 | * | |
2180 | * Process all scheduled works. Please note that the scheduled list | |
2181 | * may change while processing a work, so this function repeatedly | |
2182 | * fetches a work from the top and executes it. | |
2183 | * | |
2184 | * CONTEXT: | |
d565ed63 | 2185 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
affee4b2 TH |
2186 | * multiple times. |
2187 | */ | |
2188 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 2189 | { |
affee4b2 TH |
2190 | while (!list_empty(&worker->scheduled)) { |
2191 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 2192 | struct work_struct, entry); |
c34056a3 | 2193 | process_one_work(worker, work); |
1da177e4 | 2194 | } |
1da177e4 LT |
2195 | } |
2196 | ||
4690c4ab TH |
2197 | /** |
2198 | * worker_thread - the worker thread function | |
c34056a3 | 2199 | * @__worker: self |
4690c4ab | 2200 | * |
706026c2 TH |
2201 | * The worker thread function. There are NR_CPU_WORKER_POOLS dynamic pools |
2202 | * of these per each cpu. These workers process all works regardless of | |
e22bee78 TH |
2203 | * their specific target workqueue. The only exception is works which |
2204 | * belong to workqueues with a rescuer which will be explained in | |
2205 | * rescuer_thread(). | |
4690c4ab | 2206 | */ |
c34056a3 | 2207 | static int worker_thread(void *__worker) |
1da177e4 | 2208 | { |
c34056a3 | 2209 | struct worker *worker = __worker; |
bd7bdd43 | 2210 | struct worker_pool *pool = worker->pool; |
1da177e4 | 2211 | |
e22bee78 TH |
2212 | /* tell the scheduler that this is a workqueue worker */ |
2213 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 2214 | woke_up: |
d565ed63 | 2215 | spin_lock_irq(&pool->lock); |
1da177e4 | 2216 | |
5f7dabfd LJ |
2217 | /* we are off idle list if destruction or rebind is requested */ |
2218 | if (unlikely(list_empty(&worker->entry))) { | |
d565ed63 | 2219 | spin_unlock_irq(&pool->lock); |
25511a47 | 2220 | |
5f7dabfd | 2221 | /* if DIE is set, destruction is requested */ |
25511a47 TH |
2222 | if (worker->flags & WORKER_DIE) { |
2223 | worker->task->flags &= ~PF_WQ_WORKER; | |
2224 | return 0; | |
2225 | } | |
2226 | ||
5f7dabfd | 2227 | /* otherwise, rebind */ |
25511a47 TH |
2228 | idle_worker_rebind(worker); |
2229 | goto woke_up; | |
c8e55f36 | 2230 | } |
affee4b2 | 2231 | |
c8e55f36 | 2232 | worker_leave_idle(worker); |
db7bccf4 | 2233 | recheck: |
e22bee78 | 2234 | /* no more worker necessary? */ |
63d95a91 | 2235 | if (!need_more_worker(pool)) |
e22bee78 TH |
2236 | goto sleep; |
2237 | ||
2238 | /* do we need to manage? */ | |
63d95a91 | 2239 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
2240 | goto recheck; |
2241 | ||
c8e55f36 TH |
2242 | /* |
2243 | * ->scheduled list can only be filled while a worker is | |
2244 | * preparing to process a work or actually processing it. | |
2245 | * Make sure nobody diddled with it while I was sleeping. | |
2246 | */ | |
6183c009 | 2247 | WARN_ON_ONCE(!list_empty(&worker->scheduled)); |
c8e55f36 | 2248 | |
e22bee78 TH |
2249 | /* |
2250 | * When control reaches this point, we're guaranteed to have | |
2251 | * at least one idle worker or that someone else has already | |
2252 | * assumed the manager role. | |
2253 | */ | |
2254 | worker_clr_flags(worker, WORKER_PREP); | |
2255 | ||
2256 | do { | |
c8e55f36 | 2257 | struct work_struct *work = |
bd7bdd43 | 2258 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2259 | struct work_struct, entry); |
2260 | ||
2261 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
2262 | /* optimization path, not strictly necessary */ | |
2263 | process_one_work(worker, work); | |
2264 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2265 | process_scheduled_works(worker); |
c8e55f36 TH |
2266 | } else { |
2267 | move_linked_works(work, &worker->scheduled, NULL); | |
2268 | process_scheduled_works(worker); | |
affee4b2 | 2269 | } |
63d95a91 | 2270 | } while (keep_working(pool)); |
e22bee78 TH |
2271 | |
2272 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 2273 | sleep: |
63d95a91 | 2274 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
e22bee78 | 2275 | goto recheck; |
d313dd85 | 2276 | |
c8e55f36 | 2277 | /* |
d565ed63 TH |
2278 | * pool->lock is held and there's no work to process and no need to |
2279 | * manage, sleep. Workers are woken up only while holding | |
2280 | * pool->lock or from local cpu, so setting the current state | |
2281 | * before releasing pool->lock is enough to prevent losing any | |
2282 | * event. | |
c8e55f36 TH |
2283 | */ |
2284 | worker_enter_idle(worker); | |
2285 | __set_current_state(TASK_INTERRUPTIBLE); | |
d565ed63 | 2286 | spin_unlock_irq(&pool->lock); |
c8e55f36 TH |
2287 | schedule(); |
2288 | goto woke_up; | |
1da177e4 LT |
2289 | } |
2290 | ||
e22bee78 TH |
2291 | /** |
2292 | * rescuer_thread - the rescuer thread function | |
111c225a | 2293 | * @__rescuer: self |
e22bee78 TH |
2294 | * |
2295 | * Workqueue rescuer thread function. There's one rescuer for each | |
2296 | * workqueue which has WQ_RESCUER set. | |
2297 | * | |
706026c2 | 2298 | * Regular work processing on a pool may block trying to create a new |
e22bee78 TH |
2299 | * worker which uses GFP_KERNEL allocation which has slight chance of |
2300 | * developing into deadlock if some works currently on the same queue | |
2301 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2302 | * the problem rescuer solves. | |
2303 | * | |
706026c2 TH |
2304 | * When such condition is possible, the pool summons rescuers of all |
2305 | * workqueues which have works queued on the pool and let them process | |
e22bee78 TH |
2306 | * those works so that forward progress can be guaranteed. |
2307 | * | |
2308 | * This should happen rarely. | |
2309 | */ | |
111c225a | 2310 | static int rescuer_thread(void *__rescuer) |
e22bee78 | 2311 | { |
111c225a TH |
2312 | struct worker *rescuer = __rescuer; |
2313 | struct workqueue_struct *wq = rescuer->rescue_wq; | |
e22bee78 | 2314 | struct list_head *scheduled = &rescuer->scheduled; |
e22bee78 TH |
2315 | |
2316 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
111c225a TH |
2317 | |
2318 | /* | |
2319 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it | |
2320 | * doesn't participate in concurrency management. | |
2321 | */ | |
2322 | rescuer->task->flags |= PF_WQ_WORKER; | |
e22bee78 TH |
2323 | repeat: |
2324 | set_current_state(TASK_INTERRUPTIBLE); | |
2325 | ||
412d32e6 MG |
2326 | if (kthread_should_stop()) { |
2327 | __set_current_state(TASK_RUNNING); | |
111c225a | 2328 | rescuer->task->flags &= ~PF_WQ_WORKER; |
e22bee78 | 2329 | return 0; |
412d32e6 | 2330 | } |
e22bee78 | 2331 | |
493a1724 TH |
2332 | /* see whether any pwq is asking for help */ |
2333 | spin_lock_irq(&workqueue_lock); | |
2334 | ||
2335 | while (!list_empty(&wq->maydays)) { | |
2336 | struct pool_workqueue *pwq = list_first_entry(&wq->maydays, | |
2337 | struct pool_workqueue, mayday_node); | |
112202d9 | 2338 | struct worker_pool *pool = pwq->pool; |
e22bee78 TH |
2339 | struct work_struct *work, *n; |
2340 | ||
2341 | __set_current_state(TASK_RUNNING); | |
493a1724 TH |
2342 | list_del_init(&pwq->mayday_node); |
2343 | ||
2344 | spin_unlock_irq(&workqueue_lock); | |
e22bee78 TH |
2345 | |
2346 | /* migrate to the target cpu if possible */ | |
f36dc67b | 2347 | worker_maybe_bind_and_lock(pool); |
b3104104 | 2348 | rescuer->pool = pool; |
e22bee78 TH |
2349 | |
2350 | /* | |
2351 | * Slurp in all works issued via this workqueue and | |
2352 | * process'em. | |
2353 | */ | |
6183c009 | 2354 | WARN_ON_ONCE(!list_empty(&rescuer->scheduled)); |
bd7bdd43 | 2355 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
112202d9 | 2356 | if (get_work_pwq(work) == pwq) |
e22bee78 TH |
2357 | move_linked_works(work, scheduled, &n); |
2358 | ||
2359 | process_scheduled_works(rescuer); | |
7576958a TH |
2360 | |
2361 | /* | |
d565ed63 | 2362 | * Leave this pool. If keep_working() is %true, notify a |
7576958a TH |
2363 | * regular worker; otherwise, we end up with 0 concurrency |
2364 | * and stalling the execution. | |
2365 | */ | |
63d95a91 TH |
2366 | if (keep_working(pool)) |
2367 | wake_up_worker(pool); | |
7576958a | 2368 | |
b3104104 | 2369 | rescuer->pool = NULL; |
493a1724 TH |
2370 | spin_unlock(&pool->lock); |
2371 | spin_lock(&workqueue_lock); | |
e22bee78 TH |
2372 | } |
2373 | ||
493a1724 TH |
2374 | spin_unlock_irq(&workqueue_lock); |
2375 | ||
111c225a TH |
2376 | /* rescuers should never participate in concurrency management */ |
2377 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); | |
e22bee78 TH |
2378 | schedule(); |
2379 | goto repeat; | |
1da177e4 LT |
2380 | } |
2381 | ||
fc2e4d70 ON |
2382 | struct wq_barrier { |
2383 | struct work_struct work; | |
2384 | struct completion done; | |
2385 | }; | |
2386 | ||
2387 | static void wq_barrier_func(struct work_struct *work) | |
2388 | { | |
2389 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2390 | complete(&barr->done); | |
2391 | } | |
2392 | ||
4690c4ab TH |
2393 | /** |
2394 | * insert_wq_barrier - insert a barrier work | |
112202d9 | 2395 | * @pwq: pwq to insert barrier into |
4690c4ab | 2396 | * @barr: wq_barrier to insert |
affee4b2 TH |
2397 | * @target: target work to attach @barr to |
2398 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2399 | * |
affee4b2 TH |
2400 | * @barr is linked to @target such that @barr is completed only after |
2401 | * @target finishes execution. Please note that the ordering | |
2402 | * guarantee is observed only with respect to @target and on the local | |
2403 | * cpu. | |
2404 | * | |
2405 | * Currently, a queued barrier can't be canceled. This is because | |
2406 | * try_to_grab_pending() can't determine whether the work to be | |
2407 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2408 | * flag of the previous work while there must be a valid next work | |
2409 | * after a work with LINKED flag set. | |
2410 | * | |
2411 | * Note that when @worker is non-NULL, @target may be modified | |
112202d9 | 2412 | * underneath us, so we can't reliably determine pwq from @target. |
4690c4ab TH |
2413 | * |
2414 | * CONTEXT: | |
d565ed63 | 2415 | * spin_lock_irq(pool->lock). |
4690c4ab | 2416 | */ |
112202d9 | 2417 | static void insert_wq_barrier(struct pool_workqueue *pwq, |
affee4b2 TH |
2418 | struct wq_barrier *barr, |
2419 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2420 | { |
affee4b2 TH |
2421 | struct list_head *head; |
2422 | unsigned int linked = 0; | |
2423 | ||
dc186ad7 | 2424 | /* |
d565ed63 | 2425 | * debugobject calls are safe here even with pool->lock locked |
dc186ad7 TG |
2426 | * as we know for sure that this will not trigger any of the |
2427 | * checks and call back into the fixup functions where we | |
2428 | * might deadlock. | |
2429 | */ | |
ca1cab37 | 2430 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2431 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2432 | init_completion(&barr->done); |
83c22520 | 2433 | |
affee4b2 TH |
2434 | /* |
2435 | * If @target is currently being executed, schedule the | |
2436 | * barrier to the worker; otherwise, put it after @target. | |
2437 | */ | |
2438 | if (worker) | |
2439 | head = worker->scheduled.next; | |
2440 | else { | |
2441 | unsigned long *bits = work_data_bits(target); | |
2442 | ||
2443 | head = target->entry.next; | |
2444 | /* there can already be other linked works, inherit and set */ | |
2445 | linked = *bits & WORK_STRUCT_LINKED; | |
2446 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2447 | } | |
2448 | ||
dc186ad7 | 2449 | debug_work_activate(&barr->work); |
112202d9 | 2450 | insert_work(pwq, &barr->work, head, |
affee4b2 | 2451 | work_color_to_flags(WORK_NO_COLOR) | linked); |
fc2e4d70 ON |
2452 | } |
2453 | ||
73f53c4a | 2454 | /** |
112202d9 | 2455 | * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing |
73f53c4a TH |
2456 | * @wq: workqueue being flushed |
2457 | * @flush_color: new flush color, < 0 for no-op | |
2458 | * @work_color: new work color, < 0 for no-op | |
2459 | * | |
112202d9 | 2460 | * Prepare pwqs for workqueue flushing. |
73f53c4a | 2461 | * |
112202d9 TH |
2462 | * If @flush_color is non-negative, flush_color on all pwqs should be |
2463 | * -1. If no pwq has in-flight commands at the specified color, all | |
2464 | * pwq->flush_color's stay at -1 and %false is returned. If any pwq | |
2465 | * has in flight commands, its pwq->flush_color is set to | |
2466 | * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq | |
73f53c4a TH |
2467 | * wakeup logic is armed and %true is returned. |
2468 | * | |
2469 | * The caller should have initialized @wq->first_flusher prior to | |
2470 | * calling this function with non-negative @flush_color. If | |
2471 | * @flush_color is negative, no flush color update is done and %false | |
2472 | * is returned. | |
2473 | * | |
112202d9 | 2474 | * If @work_color is non-negative, all pwqs should have the same |
73f53c4a TH |
2475 | * work_color which is previous to @work_color and all will be |
2476 | * advanced to @work_color. | |
2477 | * | |
2478 | * CONTEXT: | |
2479 | * mutex_lock(wq->flush_mutex). | |
2480 | * | |
2481 | * RETURNS: | |
2482 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2483 | * otherwise. | |
2484 | */ | |
112202d9 | 2485 | static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq, |
73f53c4a | 2486 | int flush_color, int work_color) |
1da177e4 | 2487 | { |
73f53c4a | 2488 | bool wait = false; |
49e3cf44 | 2489 | struct pool_workqueue *pwq; |
1da177e4 | 2490 | |
73f53c4a | 2491 | if (flush_color >= 0) { |
6183c009 | 2492 | WARN_ON_ONCE(atomic_read(&wq->nr_pwqs_to_flush)); |
112202d9 | 2493 | atomic_set(&wq->nr_pwqs_to_flush, 1); |
1da177e4 | 2494 | } |
2355b70f | 2495 | |
49e3cf44 | 2496 | for_each_pwq(pwq, wq) { |
112202d9 | 2497 | struct worker_pool *pool = pwq->pool; |
fc2e4d70 | 2498 | |
d565ed63 | 2499 | spin_lock_irq(&pool->lock); |
83c22520 | 2500 | |
73f53c4a | 2501 | if (flush_color >= 0) { |
6183c009 | 2502 | WARN_ON_ONCE(pwq->flush_color != -1); |
fc2e4d70 | 2503 | |
112202d9 TH |
2504 | if (pwq->nr_in_flight[flush_color]) { |
2505 | pwq->flush_color = flush_color; | |
2506 | atomic_inc(&wq->nr_pwqs_to_flush); | |
73f53c4a TH |
2507 | wait = true; |
2508 | } | |
2509 | } | |
1da177e4 | 2510 | |
73f53c4a | 2511 | if (work_color >= 0) { |
6183c009 | 2512 | WARN_ON_ONCE(work_color != work_next_color(pwq->work_color)); |
112202d9 | 2513 | pwq->work_color = work_color; |
73f53c4a | 2514 | } |
1da177e4 | 2515 | |
d565ed63 | 2516 | spin_unlock_irq(&pool->lock); |
1da177e4 | 2517 | } |
2355b70f | 2518 | |
112202d9 | 2519 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_pwqs_to_flush)) |
73f53c4a | 2520 | complete(&wq->first_flusher->done); |
14441960 | 2521 | |
73f53c4a | 2522 | return wait; |
1da177e4 LT |
2523 | } |
2524 | ||
0fcb78c2 | 2525 | /** |
1da177e4 | 2526 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2527 | * @wq: workqueue to flush |
1da177e4 LT |
2528 | * |
2529 | * Forces execution of the workqueue and blocks until its completion. | |
2530 | * This is typically used in driver shutdown handlers. | |
2531 | * | |
fc2e4d70 ON |
2532 | * We sleep until all works which were queued on entry have been handled, |
2533 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 2534 | */ |
7ad5b3a5 | 2535 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2536 | { |
73f53c4a TH |
2537 | struct wq_flusher this_flusher = { |
2538 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2539 | .flush_color = -1, | |
2540 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2541 | }; | |
2542 | int next_color; | |
1da177e4 | 2543 | |
3295f0ef IM |
2544 | lock_map_acquire(&wq->lockdep_map); |
2545 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
2546 | |
2547 | mutex_lock(&wq->flush_mutex); | |
2548 | ||
2549 | /* | |
2550 | * Start-to-wait phase | |
2551 | */ | |
2552 | next_color = work_next_color(wq->work_color); | |
2553 | ||
2554 | if (next_color != wq->flush_color) { | |
2555 | /* | |
2556 | * Color space is not full. The current work_color | |
2557 | * becomes our flush_color and work_color is advanced | |
2558 | * by one. | |
2559 | */ | |
6183c009 | 2560 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow)); |
73f53c4a TH |
2561 | this_flusher.flush_color = wq->work_color; |
2562 | wq->work_color = next_color; | |
2563 | ||
2564 | if (!wq->first_flusher) { | |
2565 | /* no flush in progress, become the first flusher */ | |
6183c009 | 2566 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); |
73f53c4a TH |
2567 | |
2568 | wq->first_flusher = &this_flusher; | |
2569 | ||
112202d9 | 2570 | if (!flush_workqueue_prep_pwqs(wq, wq->flush_color, |
73f53c4a TH |
2571 | wq->work_color)) { |
2572 | /* nothing to flush, done */ | |
2573 | wq->flush_color = next_color; | |
2574 | wq->first_flusher = NULL; | |
2575 | goto out_unlock; | |
2576 | } | |
2577 | } else { | |
2578 | /* wait in queue */ | |
6183c009 | 2579 | WARN_ON_ONCE(wq->flush_color == this_flusher.flush_color); |
73f53c4a | 2580 | list_add_tail(&this_flusher.list, &wq->flusher_queue); |
112202d9 | 2581 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2582 | } |
2583 | } else { | |
2584 | /* | |
2585 | * Oops, color space is full, wait on overflow queue. | |
2586 | * The next flush completion will assign us | |
2587 | * flush_color and transfer to flusher_queue. | |
2588 | */ | |
2589 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2590 | } | |
2591 | ||
2592 | mutex_unlock(&wq->flush_mutex); | |
2593 | ||
2594 | wait_for_completion(&this_flusher.done); | |
2595 | ||
2596 | /* | |
2597 | * Wake-up-and-cascade phase | |
2598 | * | |
2599 | * First flushers are responsible for cascading flushes and | |
2600 | * handling overflow. Non-first flushers can simply return. | |
2601 | */ | |
2602 | if (wq->first_flusher != &this_flusher) | |
2603 | return; | |
2604 | ||
2605 | mutex_lock(&wq->flush_mutex); | |
2606 | ||
4ce48b37 TH |
2607 | /* we might have raced, check again with mutex held */ |
2608 | if (wq->first_flusher != &this_flusher) | |
2609 | goto out_unlock; | |
2610 | ||
73f53c4a TH |
2611 | wq->first_flusher = NULL; |
2612 | ||
6183c009 TH |
2613 | WARN_ON_ONCE(!list_empty(&this_flusher.list)); |
2614 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); | |
73f53c4a TH |
2615 | |
2616 | while (true) { | |
2617 | struct wq_flusher *next, *tmp; | |
2618 | ||
2619 | /* complete all the flushers sharing the current flush color */ | |
2620 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2621 | if (next->flush_color != wq->flush_color) | |
2622 | break; | |
2623 | list_del_init(&next->list); | |
2624 | complete(&next->done); | |
2625 | } | |
2626 | ||
6183c009 TH |
2627 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow) && |
2628 | wq->flush_color != work_next_color(wq->work_color)); | |
73f53c4a TH |
2629 | |
2630 | /* this flush_color is finished, advance by one */ | |
2631 | wq->flush_color = work_next_color(wq->flush_color); | |
2632 | ||
2633 | /* one color has been freed, handle overflow queue */ | |
2634 | if (!list_empty(&wq->flusher_overflow)) { | |
2635 | /* | |
2636 | * Assign the same color to all overflowed | |
2637 | * flushers, advance work_color and append to | |
2638 | * flusher_queue. This is the start-to-wait | |
2639 | * phase for these overflowed flushers. | |
2640 | */ | |
2641 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2642 | tmp->flush_color = wq->work_color; | |
2643 | ||
2644 | wq->work_color = work_next_color(wq->work_color); | |
2645 | ||
2646 | list_splice_tail_init(&wq->flusher_overflow, | |
2647 | &wq->flusher_queue); | |
112202d9 | 2648 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2649 | } |
2650 | ||
2651 | if (list_empty(&wq->flusher_queue)) { | |
6183c009 | 2652 | WARN_ON_ONCE(wq->flush_color != wq->work_color); |
73f53c4a TH |
2653 | break; |
2654 | } | |
2655 | ||
2656 | /* | |
2657 | * Need to flush more colors. Make the next flusher | |
112202d9 | 2658 | * the new first flusher and arm pwqs. |
73f53c4a | 2659 | */ |
6183c009 TH |
2660 | WARN_ON_ONCE(wq->flush_color == wq->work_color); |
2661 | WARN_ON_ONCE(wq->flush_color != next->flush_color); | |
73f53c4a TH |
2662 | |
2663 | list_del_init(&next->list); | |
2664 | wq->first_flusher = next; | |
2665 | ||
112202d9 | 2666 | if (flush_workqueue_prep_pwqs(wq, wq->flush_color, -1)) |
73f53c4a TH |
2667 | break; |
2668 | ||
2669 | /* | |
2670 | * Meh... this color is already done, clear first | |
2671 | * flusher and repeat cascading. | |
2672 | */ | |
2673 | wq->first_flusher = NULL; | |
2674 | } | |
2675 | ||
2676 | out_unlock: | |
2677 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 2678 | } |
ae90dd5d | 2679 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2680 | |
9c5a2ba7 TH |
2681 | /** |
2682 | * drain_workqueue - drain a workqueue | |
2683 | * @wq: workqueue to drain | |
2684 | * | |
2685 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2686 | * only chain queueing is allowed. IOW, only currently pending or running | |
2687 | * work items on @wq can queue further work items on it. @wq is flushed | |
2688 | * repeatedly until it becomes empty. The number of flushing is detemined | |
2689 | * by the depth of chaining and should be relatively short. Whine if it | |
2690 | * takes too long. | |
2691 | */ | |
2692 | void drain_workqueue(struct workqueue_struct *wq) | |
2693 | { | |
2694 | unsigned int flush_cnt = 0; | |
49e3cf44 | 2695 | struct pool_workqueue *pwq; |
9c5a2ba7 TH |
2696 | |
2697 | /* | |
2698 | * __queue_work() needs to test whether there are drainers, is much | |
2699 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
2700 | * Use WQ_DRAINING so that queue doesn't have to check nr_drainers. | |
2701 | */ | |
e98d5b16 | 2702 | spin_lock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2703 | if (!wq->nr_drainers++) |
2704 | wq->flags |= WQ_DRAINING; | |
e98d5b16 | 2705 | spin_unlock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2706 | reflush: |
2707 | flush_workqueue(wq); | |
2708 | ||
49e3cf44 | 2709 | for_each_pwq(pwq, wq) { |
fa2563e4 | 2710 | bool drained; |
9c5a2ba7 | 2711 | |
112202d9 TH |
2712 | spin_lock_irq(&pwq->pool->lock); |
2713 | drained = !pwq->nr_active && list_empty(&pwq->delayed_works); | |
2714 | spin_unlock_irq(&pwq->pool->lock); | |
fa2563e4 TT |
2715 | |
2716 | if (drained) | |
9c5a2ba7 TH |
2717 | continue; |
2718 | ||
2719 | if (++flush_cnt == 10 || | |
2720 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
044c782c VI |
2721 | pr_warn("workqueue %s: flush on destruction isn't complete after %u tries\n", |
2722 | wq->name, flush_cnt); | |
9c5a2ba7 TH |
2723 | goto reflush; |
2724 | } | |
2725 | ||
e98d5b16 | 2726 | spin_lock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2727 | if (!--wq->nr_drainers) |
2728 | wq->flags &= ~WQ_DRAINING; | |
e98d5b16 | 2729 | spin_unlock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2730 | } |
2731 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2732 | ||
606a5020 | 2733 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) |
db700897 | 2734 | { |
affee4b2 | 2735 | struct worker *worker = NULL; |
c9e7cf27 | 2736 | struct worker_pool *pool; |
112202d9 | 2737 | struct pool_workqueue *pwq; |
db700897 ON |
2738 | |
2739 | might_sleep(); | |
c9e7cf27 TH |
2740 | pool = get_work_pool(work); |
2741 | if (!pool) | |
baf59022 | 2742 | return false; |
db700897 | 2743 | |
d565ed63 | 2744 | spin_lock_irq(&pool->lock); |
0b3dae68 | 2745 | /* see the comment in try_to_grab_pending() with the same code */ |
112202d9 TH |
2746 | pwq = get_work_pwq(work); |
2747 | if (pwq) { | |
2748 | if (unlikely(pwq->pool != pool)) | |
4690c4ab | 2749 | goto already_gone; |
606a5020 | 2750 | } else { |
c9e7cf27 | 2751 | worker = find_worker_executing_work(pool, work); |
affee4b2 | 2752 | if (!worker) |
4690c4ab | 2753 | goto already_gone; |
112202d9 | 2754 | pwq = worker->current_pwq; |
606a5020 | 2755 | } |
db700897 | 2756 | |
112202d9 | 2757 | insert_wq_barrier(pwq, barr, work, worker); |
d565ed63 | 2758 | spin_unlock_irq(&pool->lock); |
7a22ad75 | 2759 | |
e159489b TH |
2760 | /* |
2761 | * If @max_active is 1 or rescuer is in use, flushing another work | |
2762 | * item on the same workqueue may lead to deadlock. Make sure the | |
2763 | * flusher is not running on the same workqueue by verifying write | |
2764 | * access. | |
2765 | */ | |
112202d9 TH |
2766 | if (pwq->wq->saved_max_active == 1 || pwq->wq->flags & WQ_RESCUER) |
2767 | lock_map_acquire(&pwq->wq->lockdep_map); | |
e159489b | 2768 | else |
112202d9 TH |
2769 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
2770 | lock_map_release(&pwq->wq->lockdep_map); | |
e159489b | 2771 | |
401a8d04 | 2772 | return true; |
4690c4ab | 2773 | already_gone: |
d565ed63 | 2774 | spin_unlock_irq(&pool->lock); |
401a8d04 | 2775 | return false; |
db700897 | 2776 | } |
baf59022 TH |
2777 | |
2778 | /** | |
2779 | * flush_work - wait for a work to finish executing the last queueing instance | |
2780 | * @work: the work to flush | |
2781 | * | |
606a5020 TH |
2782 | * Wait until @work has finished execution. @work is guaranteed to be idle |
2783 | * on return if it hasn't been requeued since flush started. | |
baf59022 TH |
2784 | * |
2785 | * RETURNS: | |
2786 | * %true if flush_work() waited for the work to finish execution, | |
2787 | * %false if it was already idle. | |
2788 | */ | |
2789 | bool flush_work(struct work_struct *work) | |
2790 | { | |
2791 | struct wq_barrier barr; | |
2792 | ||
0976dfc1 SB |
2793 | lock_map_acquire(&work->lockdep_map); |
2794 | lock_map_release(&work->lockdep_map); | |
2795 | ||
606a5020 | 2796 | if (start_flush_work(work, &barr)) { |
401a8d04 TH |
2797 | wait_for_completion(&barr.done); |
2798 | destroy_work_on_stack(&barr.work); | |
2799 | return true; | |
606a5020 | 2800 | } else { |
401a8d04 | 2801 | return false; |
6e84d644 | 2802 | } |
6e84d644 | 2803 | } |
606a5020 | 2804 | EXPORT_SYMBOL_GPL(flush_work); |
6e84d644 | 2805 | |
36e227d2 | 2806 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
1f1f642e | 2807 | { |
bbb68dfa | 2808 | unsigned long flags; |
1f1f642e ON |
2809 | int ret; |
2810 | ||
2811 | do { | |
bbb68dfa TH |
2812 | ret = try_to_grab_pending(work, is_dwork, &flags); |
2813 | /* | |
2814 | * If someone else is canceling, wait for the same event it | |
2815 | * would be waiting for before retrying. | |
2816 | */ | |
2817 | if (unlikely(ret == -ENOENT)) | |
606a5020 | 2818 | flush_work(work); |
1f1f642e ON |
2819 | } while (unlikely(ret < 0)); |
2820 | ||
bbb68dfa TH |
2821 | /* tell other tasks trying to grab @work to back off */ |
2822 | mark_work_canceling(work); | |
2823 | local_irq_restore(flags); | |
2824 | ||
606a5020 | 2825 | flush_work(work); |
7a22ad75 | 2826 | clear_work_data(work); |
1f1f642e ON |
2827 | return ret; |
2828 | } | |
2829 | ||
6e84d644 | 2830 | /** |
401a8d04 TH |
2831 | * cancel_work_sync - cancel a work and wait for it to finish |
2832 | * @work: the work to cancel | |
6e84d644 | 2833 | * |
401a8d04 TH |
2834 | * Cancel @work and wait for its execution to finish. This function |
2835 | * can be used even if the work re-queues itself or migrates to | |
2836 | * another workqueue. On return from this function, @work is | |
2837 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2838 | * |
401a8d04 TH |
2839 | * cancel_work_sync(&delayed_work->work) must not be used for |
2840 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2841 | * |
401a8d04 | 2842 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2843 | * queued can't be destroyed before this function returns. |
401a8d04 TH |
2844 | * |
2845 | * RETURNS: | |
2846 | * %true if @work was pending, %false otherwise. | |
6e84d644 | 2847 | */ |
401a8d04 | 2848 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2849 | { |
36e227d2 | 2850 | return __cancel_work_timer(work, false); |
b89deed3 | 2851 | } |
28e53bdd | 2852 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2853 | |
6e84d644 | 2854 | /** |
401a8d04 TH |
2855 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2856 | * @dwork: the delayed work to flush | |
6e84d644 | 2857 | * |
401a8d04 TH |
2858 | * Delayed timer is cancelled and the pending work is queued for |
2859 | * immediate execution. Like flush_work(), this function only | |
2860 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2861 | * |
401a8d04 TH |
2862 | * RETURNS: |
2863 | * %true if flush_work() waited for the work to finish execution, | |
2864 | * %false if it was already idle. | |
6e84d644 | 2865 | */ |
401a8d04 TH |
2866 | bool flush_delayed_work(struct delayed_work *dwork) |
2867 | { | |
8930caba | 2868 | local_irq_disable(); |
401a8d04 | 2869 | if (del_timer_sync(&dwork->timer)) |
60c057bc | 2870 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
8930caba | 2871 | local_irq_enable(); |
401a8d04 TH |
2872 | return flush_work(&dwork->work); |
2873 | } | |
2874 | EXPORT_SYMBOL(flush_delayed_work); | |
2875 | ||
09383498 | 2876 | /** |
57b30ae7 TH |
2877 | * cancel_delayed_work - cancel a delayed work |
2878 | * @dwork: delayed_work to cancel | |
09383498 | 2879 | * |
57b30ae7 TH |
2880 | * Kill off a pending delayed_work. Returns %true if @dwork was pending |
2881 | * and canceled; %false if wasn't pending. Note that the work callback | |
2882 | * function may still be running on return, unless it returns %true and the | |
2883 | * work doesn't re-arm itself. Explicitly flush or use | |
2884 | * cancel_delayed_work_sync() to wait on it. | |
09383498 | 2885 | * |
57b30ae7 | 2886 | * This function is safe to call from any context including IRQ handler. |
09383498 | 2887 | */ |
57b30ae7 | 2888 | bool cancel_delayed_work(struct delayed_work *dwork) |
09383498 | 2889 | { |
57b30ae7 TH |
2890 | unsigned long flags; |
2891 | int ret; | |
2892 | ||
2893 | do { | |
2894 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
2895 | } while (unlikely(ret == -EAGAIN)); | |
2896 | ||
2897 | if (unlikely(ret < 0)) | |
2898 | return false; | |
2899 | ||
7c3eed5c TH |
2900 | set_work_pool_and_clear_pending(&dwork->work, |
2901 | get_work_pool_id(&dwork->work)); | |
57b30ae7 | 2902 | local_irq_restore(flags); |
c0158ca6 | 2903 | return ret; |
09383498 | 2904 | } |
57b30ae7 | 2905 | EXPORT_SYMBOL(cancel_delayed_work); |
09383498 | 2906 | |
401a8d04 TH |
2907 | /** |
2908 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
2909 | * @dwork: the delayed work cancel | |
2910 | * | |
2911 | * This is cancel_work_sync() for delayed works. | |
2912 | * | |
2913 | * RETURNS: | |
2914 | * %true if @dwork was pending, %false otherwise. | |
2915 | */ | |
2916 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 2917 | { |
36e227d2 | 2918 | return __cancel_work_timer(&dwork->work, true); |
6e84d644 | 2919 | } |
f5a421a4 | 2920 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 2921 | |
0fcb78c2 | 2922 | /** |
c1a220e7 ZR |
2923 | * schedule_work_on - put work task on a specific cpu |
2924 | * @cpu: cpu to put the work task on | |
2925 | * @work: job to be done | |
2926 | * | |
2927 | * This puts a job on a specific cpu | |
2928 | */ | |
d4283e93 | 2929 | bool schedule_work_on(int cpu, struct work_struct *work) |
c1a220e7 | 2930 | { |
d320c038 | 2931 | return queue_work_on(cpu, system_wq, work); |
c1a220e7 ZR |
2932 | } |
2933 | EXPORT_SYMBOL(schedule_work_on); | |
2934 | ||
0fcb78c2 | 2935 | /** |
0fcb78c2 REB |
2936 | * schedule_work - put work task in global workqueue |
2937 | * @work: job to be done | |
0fcb78c2 | 2938 | * |
d4283e93 TH |
2939 | * Returns %false if @work was already on the kernel-global workqueue and |
2940 | * %true otherwise. | |
5b0f437d BVA |
2941 | * |
2942 | * This puts a job in the kernel-global workqueue if it was not already | |
2943 | * queued and leaves it in the same position on the kernel-global | |
2944 | * workqueue otherwise. | |
0fcb78c2 | 2945 | */ |
d4283e93 | 2946 | bool schedule_work(struct work_struct *work) |
1da177e4 | 2947 | { |
d320c038 | 2948 | return queue_work(system_wq, work); |
1da177e4 | 2949 | } |
ae90dd5d | 2950 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 2951 | |
0fcb78c2 REB |
2952 | /** |
2953 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
2954 | * @cpu: cpu to use | |
52bad64d | 2955 | * @dwork: job to be done |
0fcb78c2 REB |
2956 | * @delay: number of jiffies to wait |
2957 | * | |
2958 | * After waiting for a given time this puts a job in the kernel-global | |
2959 | * workqueue on the specified CPU. | |
2960 | */ | |
d4283e93 TH |
2961 | bool schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
2962 | unsigned long delay) | |
1da177e4 | 2963 | { |
d320c038 | 2964 | return queue_delayed_work_on(cpu, system_wq, dwork, delay); |
1da177e4 | 2965 | } |
ae90dd5d | 2966 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 2967 | |
0fcb78c2 REB |
2968 | /** |
2969 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
2970 | * @dwork: job to be done |
2971 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
2972 | * |
2973 | * After waiting for a given time this puts a job in the kernel-global | |
2974 | * workqueue. | |
2975 | */ | |
d4283e93 | 2976 | bool schedule_delayed_work(struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 2977 | { |
d320c038 | 2978 | return queue_delayed_work(system_wq, dwork, delay); |
1da177e4 | 2979 | } |
ae90dd5d | 2980 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 2981 | |
b6136773 | 2982 | /** |
31ddd871 | 2983 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 2984 | * @func: the function to call |
b6136773 | 2985 | * |
31ddd871 TH |
2986 | * schedule_on_each_cpu() executes @func on each online CPU using the |
2987 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 2988 | * schedule_on_each_cpu() is very slow. |
31ddd871 TH |
2989 | * |
2990 | * RETURNS: | |
2991 | * 0 on success, -errno on failure. | |
b6136773 | 2992 | */ |
65f27f38 | 2993 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
2994 | { |
2995 | int cpu; | |
38f51568 | 2996 | struct work_struct __percpu *works; |
15316ba8 | 2997 | |
b6136773 AM |
2998 | works = alloc_percpu(struct work_struct); |
2999 | if (!works) | |
15316ba8 | 3000 | return -ENOMEM; |
b6136773 | 3001 | |
93981800 TH |
3002 | get_online_cpus(); |
3003 | ||
15316ba8 | 3004 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
3005 | struct work_struct *work = per_cpu_ptr(works, cpu); |
3006 | ||
3007 | INIT_WORK(work, func); | |
b71ab8c2 | 3008 | schedule_work_on(cpu, work); |
65a64464 | 3009 | } |
93981800 TH |
3010 | |
3011 | for_each_online_cpu(cpu) | |
3012 | flush_work(per_cpu_ptr(works, cpu)); | |
3013 | ||
95402b38 | 3014 | put_online_cpus(); |
b6136773 | 3015 | free_percpu(works); |
15316ba8 CL |
3016 | return 0; |
3017 | } | |
3018 | ||
eef6a7d5 AS |
3019 | /** |
3020 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
3021 | * | |
3022 | * Forces execution of the kernel-global workqueue and blocks until its | |
3023 | * completion. | |
3024 | * | |
3025 | * Think twice before calling this function! It's very easy to get into | |
3026 | * trouble if you don't take great care. Either of the following situations | |
3027 | * will lead to deadlock: | |
3028 | * | |
3029 | * One of the work items currently on the workqueue needs to acquire | |
3030 | * a lock held by your code or its caller. | |
3031 | * | |
3032 | * Your code is running in the context of a work routine. | |
3033 | * | |
3034 | * They will be detected by lockdep when they occur, but the first might not | |
3035 | * occur very often. It depends on what work items are on the workqueue and | |
3036 | * what locks they need, which you have no control over. | |
3037 | * | |
3038 | * In most situations flushing the entire workqueue is overkill; you merely | |
3039 | * need to know that a particular work item isn't queued and isn't running. | |
3040 | * In such cases you should use cancel_delayed_work_sync() or | |
3041 | * cancel_work_sync() instead. | |
3042 | */ | |
1da177e4 LT |
3043 | void flush_scheduled_work(void) |
3044 | { | |
d320c038 | 3045 | flush_workqueue(system_wq); |
1da177e4 | 3046 | } |
ae90dd5d | 3047 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 3048 | |
1fa44eca JB |
3049 | /** |
3050 | * execute_in_process_context - reliably execute the routine with user context | |
3051 | * @fn: the function to execute | |
1fa44eca JB |
3052 | * @ew: guaranteed storage for the execute work structure (must |
3053 | * be available when the work executes) | |
3054 | * | |
3055 | * Executes the function immediately if process context is available, | |
3056 | * otherwise schedules the function for delayed execution. | |
3057 | * | |
3058 | * Returns: 0 - function was executed | |
3059 | * 1 - function was scheduled for execution | |
3060 | */ | |
65f27f38 | 3061 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
3062 | { |
3063 | if (!in_interrupt()) { | |
65f27f38 | 3064 | fn(&ew->work); |
1fa44eca JB |
3065 | return 0; |
3066 | } | |
3067 | ||
65f27f38 | 3068 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
3069 | schedule_work(&ew->work); |
3070 | ||
3071 | return 1; | |
3072 | } | |
3073 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
3074 | ||
1da177e4 LT |
3075 | int keventd_up(void) |
3076 | { | |
d320c038 | 3077 | return system_wq != NULL; |
1da177e4 LT |
3078 | } |
3079 | ||
30cdf249 | 3080 | static int alloc_and_link_pwqs(struct workqueue_struct *wq) |
0f900049 | 3081 | { |
49e3cf44 | 3082 | bool highpri = wq->flags & WQ_HIGHPRI; |
30cdf249 TH |
3083 | int cpu; |
3084 | ||
3085 | if (!(wq->flags & WQ_UNBOUND)) { | |
420c0ddb TH |
3086 | wq->cpu_pwqs = alloc_percpu(struct pool_workqueue); |
3087 | if (!wq->cpu_pwqs) | |
30cdf249 TH |
3088 | return -ENOMEM; |
3089 | ||
3090 | for_each_possible_cpu(cpu) { | |
3091 | struct pool_workqueue *pwq = get_pwq(cpu, wq); | |
f3421797 | 3092 | |
49e3cf44 | 3093 | pwq->pool = get_std_worker_pool(cpu, highpri); |
30cdf249 TH |
3094 | list_add_tail(&pwq->pwqs_node, &wq->pwqs); |
3095 | } | |
3096 | } else { | |
3097 | struct pool_workqueue *pwq; | |
3098 | ||
3099 | pwq = kmem_cache_zalloc(pwq_cache, GFP_KERNEL); | |
3100 | if (!pwq) | |
3101 | return -ENOMEM; | |
3102 | ||
49e3cf44 | 3103 | pwq->pool = get_std_worker_pool(WORK_CPU_UNBOUND, highpri); |
30cdf249 TH |
3104 | list_add_tail(&pwq->pwqs_node, &wq->pwqs); |
3105 | } | |
3106 | ||
3107 | return 0; | |
0f900049 TH |
3108 | } |
3109 | ||
112202d9 | 3110 | static void free_pwqs(struct workqueue_struct *wq) |
0f900049 | 3111 | { |
e06ffa1e | 3112 | if (!(wq->flags & WQ_UNBOUND)) |
420c0ddb TH |
3113 | free_percpu(wq->cpu_pwqs); |
3114 | else if (!list_empty(&wq->pwqs)) | |
3115 | kmem_cache_free(pwq_cache, list_first_entry(&wq->pwqs, | |
3116 | struct pool_workqueue, pwqs_node)); | |
0f900049 TH |
3117 | } |
3118 | ||
f3421797 TH |
3119 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
3120 | const char *name) | |
b71ab8c2 | 3121 | { |
f3421797 TH |
3122 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
3123 | ||
3124 | if (max_active < 1 || max_active > lim) | |
044c782c VI |
3125 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
3126 | max_active, name, 1, lim); | |
b71ab8c2 | 3127 | |
f3421797 | 3128 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
3129 | } |
3130 | ||
b196be89 | 3131 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
d320c038 TH |
3132 | unsigned int flags, |
3133 | int max_active, | |
3134 | struct lock_class_key *key, | |
b196be89 | 3135 | const char *lock_name, ...) |
1da177e4 | 3136 | { |
b196be89 | 3137 | va_list args, args1; |
1da177e4 | 3138 | struct workqueue_struct *wq; |
49e3cf44 | 3139 | struct pool_workqueue *pwq; |
b196be89 TH |
3140 | size_t namelen; |
3141 | ||
3142 | /* determine namelen, allocate wq and format name */ | |
3143 | va_start(args, lock_name); | |
3144 | va_copy(args1, args); | |
3145 | namelen = vsnprintf(NULL, 0, fmt, args) + 1; | |
3146 | ||
3147 | wq = kzalloc(sizeof(*wq) + namelen, GFP_KERNEL); | |
3148 | if (!wq) | |
3149 | goto err; | |
3150 | ||
3151 | vsnprintf(wq->name, namelen, fmt, args1); | |
3152 | va_end(args); | |
3153 | va_end(args1); | |
1da177e4 | 3154 | |
6370a6ad TH |
3155 | /* |
3156 | * Workqueues which may be used during memory reclaim should | |
3157 | * have a rescuer to guarantee forward progress. | |
3158 | */ | |
3159 | if (flags & WQ_MEM_RECLAIM) | |
3160 | flags |= WQ_RESCUER; | |
3161 | ||
d320c038 | 3162 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 3163 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 3164 | |
b196be89 | 3165 | /* init wq */ |
97e37d7b | 3166 | wq->flags = flags; |
a0a1a5fd | 3167 | wq->saved_max_active = max_active; |
73f53c4a | 3168 | mutex_init(&wq->flush_mutex); |
112202d9 | 3169 | atomic_set(&wq->nr_pwqs_to_flush, 0); |
30cdf249 | 3170 | INIT_LIST_HEAD(&wq->pwqs); |
73f53c4a TH |
3171 | INIT_LIST_HEAD(&wq->flusher_queue); |
3172 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
493a1724 | 3173 | INIT_LIST_HEAD(&wq->maydays); |
502ca9d8 | 3174 | |
eb13ba87 | 3175 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 3176 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 3177 | |
30cdf249 | 3178 | if (alloc_and_link_pwqs(wq) < 0) |
bdbc5dd7 TH |
3179 | goto err; |
3180 | ||
49e3cf44 | 3181 | for_each_pwq(pwq, wq) { |
112202d9 | 3182 | BUG_ON((unsigned long)pwq & WORK_STRUCT_FLAG_MASK); |
112202d9 TH |
3183 | pwq->wq = wq; |
3184 | pwq->flush_color = -1; | |
3185 | pwq->max_active = max_active; | |
3186 | INIT_LIST_HEAD(&pwq->delayed_works); | |
493a1724 | 3187 | INIT_LIST_HEAD(&pwq->mayday_node); |
e22bee78 | 3188 | } |
1537663f | 3189 | |
e22bee78 TH |
3190 | if (flags & WQ_RESCUER) { |
3191 | struct worker *rescuer; | |
3192 | ||
e22bee78 TH |
3193 | wq->rescuer = rescuer = alloc_worker(); |
3194 | if (!rescuer) | |
3195 | goto err; | |
3196 | ||
111c225a TH |
3197 | rescuer->rescue_wq = wq; |
3198 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", | |
b196be89 | 3199 | wq->name); |
e22bee78 TH |
3200 | if (IS_ERR(rescuer->task)) |
3201 | goto err; | |
3202 | ||
e22bee78 TH |
3203 | rescuer->task->flags |= PF_THREAD_BOUND; |
3204 | wake_up_process(rescuer->task); | |
3af24433 ON |
3205 | } |
3206 | ||
a0a1a5fd TH |
3207 | /* |
3208 | * workqueue_lock protects global freeze state and workqueues | |
3209 | * list. Grab it, set max_active accordingly and add the new | |
3210 | * workqueue to workqueues list. | |
3211 | */ | |
e98d5b16 | 3212 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3213 | |
58a69cb4 | 3214 | if (workqueue_freezing && wq->flags & WQ_FREEZABLE) |
49e3cf44 TH |
3215 | for_each_pwq(pwq, wq) |
3216 | pwq->max_active = 0; | |
a0a1a5fd | 3217 | |
1537663f | 3218 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 3219 | |
e98d5b16 | 3220 | spin_unlock_irq(&workqueue_lock); |
1537663f | 3221 | |
3af24433 | 3222 | return wq; |
4690c4ab TH |
3223 | err: |
3224 | if (wq) { | |
112202d9 | 3225 | free_pwqs(wq); |
e22bee78 | 3226 | kfree(wq->rescuer); |
4690c4ab TH |
3227 | kfree(wq); |
3228 | } | |
3229 | return NULL; | |
3af24433 | 3230 | } |
d320c038 | 3231 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 3232 | |
3af24433 ON |
3233 | /** |
3234 | * destroy_workqueue - safely terminate a workqueue | |
3235 | * @wq: target workqueue | |
3236 | * | |
3237 | * Safely destroy a workqueue. All work currently pending will be done first. | |
3238 | */ | |
3239 | void destroy_workqueue(struct workqueue_struct *wq) | |
3240 | { | |
49e3cf44 | 3241 | struct pool_workqueue *pwq; |
3af24433 | 3242 | |
9c5a2ba7 TH |
3243 | /* drain it before proceeding with destruction */ |
3244 | drain_workqueue(wq); | |
c8efcc25 | 3245 | |
6183c009 | 3246 | /* sanity checks */ |
49e3cf44 | 3247 | for_each_pwq(pwq, wq) { |
6183c009 TH |
3248 | int i; |
3249 | ||
3250 | for (i = 0; i < WORK_NR_COLORS; i++) | |
3251 | if (WARN_ON(pwq->nr_in_flight[i])) | |
3252 | return; | |
3253 | if (WARN_ON(pwq->nr_active) || | |
3254 | WARN_ON(!list_empty(&pwq->delayed_works))) | |
3255 | return; | |
3256 | } | |
3257 | ||
a0a1a5fd TH |
3258 | /* |
3259 | * wq list is used to freeze wq, remove from list after | |
3260 | * flushing is complete in case freeze races us. | |
3261 | */ | |
e98d5b16 | 3262 | spin_lock_irq(&workqueue_lock); |
b1f4ec17 | 3263 | list_del(&wq->list); |
e98d5b16 | 3264 | spin_unlock_irq(&workqueue_lock); |
3af24433 | 3265 | |
e22bee78 TH |
3266 | if (wq->flags & WQ_RESCUER) { |
3267 | kthread_stop(wq->rescuer->task); | |
8d9df9f0 | 3268 | kfree(wq->rescuer); |
e22bee78 TH |
3269 | } |
3270 | ||
112202d9 | 3271 | free_pwqs(wq); |
3af24433 ON |
3272 | kfree(wq); |
3273 | } | |
3274 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
3275 | ||
9f4bd4cd | 3276 | /** |
112202d9 TH |
3277 | * pwq_set_max_active - adjust max_active of a pwq |
3278 | * @pwq: target pool_workqueue | |
9f4bd4cd LJ |
3279 | * @max_active: new max_active value. |
3280 | * | |
112202d9 | 3281 | * Set @pwq->max_active to @max_active and activate delayed works if |
9f4bd4cd LJ |
3282 | * increased. |
3283 | * | |
3284 | * CONTEXT: | |
d565ed63 | 3285 | * spin_lock_irq(pool->lock). |
9f4bd4cd | 3286 | */ |
112202d9 | 3287 | static void pwq_set_max_active(struct pool_workqueue *pwq, int max_active) |
9f4bd4cd | 3288 | { |
112202d9 | 3289 | pwq->max_active = max_active; |
9f4bd4cd | 3290 | |
112202d9 TH |
3291 | while (!list_empty(&pwq->delayed_works) && |
3292 | pwq->nr_active < pwq->max_active) | |
3293 | pwq_activate_first_delayed(pwq); | |
9f4bd4cd LJ |
3294 | } |
3295 | ||
dcd989cb TH |
3296 | /** |
3297 | * workqueue_set_max_active - adjust max_active of a workqueue | |
3298 | * @wq: target workqueue | |
3299 | * @max_active: new max_active value. | |
3300 | * | |
3301 | * Set max_active of @wq to @max_active. | |
3302 | * | |
3303 | * CONTEXT: | |
3304 | * Don't call from IRQ context. | |
3305 | */ | |
3306 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
3307 | { | |
49e3cf44 | 3308 | struct pool_workqueue *pwq; |
dcd989cb | 3309 | |
f3421797 | 3310 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb | 3311 | |
e98d5b16 | 3312 | spin_lock_irq(&workqueue_lock); |
dcd989cb TH |
3313 | |
3314 | wq->saved_max_active = max_active; | |
3315 | ||
49e3cf44 | 3316 | for_each_pwq(pwq, wq) { |
112202d9 | 3317 | struct worker_pool *pool = pwq->pool; |
dcd989cb | 3318 | |
e98d5b16 | 3319 | spin_lock(&pool->lock); |
dcd989cb | 3320 | |
58a69cb4 | 3321 | if (!(wq->flags & WQ_FREEZABLE) || |
35b6bb63 | 3322 | !(pool->flags & POOL_FREEZING)) |
112202d9 | 3323 | pwq_set_max_active(pwq, max_active); |
9bfb1839 | 3324 | |
e98d5b16 | 3325 | spin_unlock(&pool->lock); |
65a64464 | 3326 | } |
93981800 | 3327 | |
e98d5b16 | 3328 | spin_unlock_irq(&workqueue_lock); |
15316ba8 | 3329 | } |
dcd989cb | 3330 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 3331 | |
eef6a7d5 | 3332 | /** |
dcd989cb TH |
3333 | * workqueue_congested - test whether a workqueue is congested |
3334 | * @cpu: CPU in question | |
3335 | * @wq: target workqueue | |
eef6a7d5 | 3336 | * |
dcd989cb TH |
3337 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
3338 | * no synchronization around this function and the test result is | |
3339 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 3340 | * |
dcd989cb TH |
3341 | * RETURNS: |
3342 | * %true if congested, %false otherwise. | |
eef6a7d5 | 3343 | */ |
d84ff051 | 3344 | bool workqueue_congested(int cpu, struct workqueue_struct *wq) |
1da177e4 | 3345 | { |
112202d9 | 3346 | struct pool_workqueue *pwq = get_pwq(cpu, wq); |
dcd989cb | 3347 | |
112202d9 | 3348 | return !list_empty(&pwq->delayed_works); |
1da177e4 | 3349 | } |
dcd989cb | 3350 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 3351 | |
dcd989cb TH |
3352 | /** |
3353 | * work_busy - test whether a work is currently pending or running | |
3354 | * @work: the work to be tested | |
3355 | * | |
3356 | * Test whether @work is currently pending or running. There is no | |
3357 | * synchronization around this function and the test result is | |
3358 | * unreliable and only useful as advisory hints or for debugging. | |
dcd989cb TH |
3359 | * |
3360 | * RETURNS: | |
3361 | * OR'd bitmask of WORK_BUSY_* bits. | |
3362 | */ | |
3363 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 3364 | { |
c9e7cf27 | 3365 | struct worker_pool *pool = get_work_pool(work); |
dcd989cb TH |
3366 | unsigned long flags; |
3367 | unsigned int ret = 0; | |
1da177e4 | 3368 | |
dcd989cb TH |
3369 | if (work_pending(work)) |
3370 | ret |= WORK_BUSY_PENDING; | |
1da177e4 | 3371 | |
038366c5 LJ |
3372 | if (pool) { |
3373 | spin_lock_irqsave(&pool->lock, flags); | |
3374 | if (find_worker_executing_work(pool, work)) | |
3375 | ret |= WORK_BUSY_RUNNING; | |
3376 | spin_unlock_irqrestore(&pool->lock, flags); | |
3377 | } | |
1da177e4 | 3378 | |
dcd989cb | 3379 | return ret; |
1da177e4 | 3380 | } |
dcd989cb | 3381 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 3382 | |
db7bccf4 TH |
3383 | /* |
3384 | * CPU hotplug. | |
3385 | * | |
e22bee78 | 3386 | * There are two challenges in supporting CPU hotplug. Firstly, there |
112202d9 | 3387 | * are a lot of assumptions on strong associations among work, pwq and |
706026c2 | 3388 | * pool which make migrating pending and scheduled works very |
e22bee78 | 3389 | * difficult to implement without impacting hot paths. Secondly, |
94cf58bb | 3390 | * worker pools serve mix of short, long and very long running works making |
e22bee78 TH |
3391 | * blocked draining impractical. |
3392 | * | |
24647570 | 3393 | * This is solved by allowing the pools to be disassociated from the CPU |
628c78e7 TH |
3394 | * running as an unbound one and allowing it to be reattached later if the |
3395 | * cpu comes back online. | |
db7bccf4 | 3396 | */ |
1da177e4 | 3397 | |
706026c2 | 3398 | static void wq_unbind_fn(struct work_struct *work) |
3af24433 | 3399 | { |
38db41d9 | 3400 | int cpu = smp_processor_id(); |
4ce62e9e | 3401 | struct worker_pool *pool; |
db7bccf4 | 3402 | struct worker *worker; |
db7bccf4 | 3403 | int i; |
3af24433 | 3404 | |
38db41d9 | 3405 | for_each_std_worker_pool(pool, cpu) { |
6183c009 | 3406 | WARN_ON_ONCE(cpu != smp_processor_id()); |
db7bccf4 | 3407 | |
94cf58bb TH |
3408 | mutex_lock(&pool->assoc_mutex); |
3409 | spin_lock_irq(&pool->lock); | |
3af24433 | 3410 | |
94cf58bb TH |
3411 | /* |
3412 | * We've claimed all manager positions. Make all workers | |
3413 | * unbound and set DISASSOCIATED. Before this, all workers | |
3414 | * except for the ones which are still executing works from | |
3415 | * before the last CPU down must be on the cpu. After | |
3416 | * this, they may become diasporas. | |
3417 | */ | |
4ce62e9e | 3418 | list_for_each_entry(worker, &pool->idle_list, entry) |
403c821d | 3419 | worker->flags |= WORKER_UNBOUND; |
3af24433 | 3420 | |
b67bfe0d | 3421 | for_each_busy_worker(worker, i, pool) |
c9e7cf27 | 3422 | worker->flags |= WORKER_UNBOUND; |
06ba38a9 | 3423 | |
24647570 | 3424 | pool->flags |= POOL_DISASSOCIATED; |
f2d5a0ee | 3425 | |
94cf58bb TH |
3426 | spin_unlock_irq(&pool->lock); |
3427 | mutex_unlock(&pool->assoc_mutex); | |
3428 | } | |
628c78e7 | 3429 | |
e22bee78 | 3430 | /* |
403c821d | 3431 | * Call schedule() so that we cross rq->lock and thus can guarantee |
628c78e7 TH |
3432 | * sched callbacks see the %WORKER_UNBOUND flag. This is necessary |
3433 | * as scheduler callbacks may be invoked from other cpus. | |
e22bee78 | 3434 | */ |
e22bee78 | 3435 | schedule(); |
06ba38a9 | 3436 | |
e22bee78 | 3437 | /* |
628c78e7 TH |
3438 | * Sched callbacks are disabled now. Zap nr_running. After this, |
3439 | * nr_running stays zero and need_more_worker() and keep_working() | |
38db41d9 TH |
3440 | * are always true as long as the worklist is not empty. Pools on |
3441 | * @cpu now behave as unbound (in terms of concurrency management) | |
3442 | * pools which are served by workers tied to the CPU. | |
628c78e7 TH |
3443 | * |
3444 | * On return from this function, the current worker would trigger | |
3445 | * unbound chain execution of pending work items if other workers | |
3446 | * didn't already. | |
e22bee78 | 3447 | */ |
38db41d9 | 3448 | for_each_std_worker_pool(pool, cpu) |
e19e397a | 3449 | atomic_set(&pool->nr_running, 0); |
3af24433 | 3450 | } |
3af24433 | 3451 | |
8db25e78 TH |
3452 | /* |
3453 | * Workqueues should be brought up before normal priority CPU notifiers. | |
3454 | * This will be registered high priority CPU notifier. | |
3455 | */ | |
9fdf9b73 | 3456 | static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb, |
8db25e78 TH |
3457 | unsigned long action, |
3458 | void *hcpu) | |
3af24433 | 3459 | { |
d84ff051 | 3460 | int cpu = (unsigned long)hcpu; |
4ce62e9e | 3461 | struct worker_pool *pool; |
3ce63377 | 3462 | |
8db25e78 | 3463 | switch (action & ~CPU_TASKS_FROZEN) { |
3af24433 | 3464 | case CPU_UP_PREPARE: |
38db41d9 | 3465 | for_each_std_worker_pool(pool, cpu) { |
3ce63377 TH |
3466 | struct worker *worker; |
3467 | ||
3468 | if (pool->nr_workers) | |
3469 | continue; | |
3470 | ||
3471 | worker = create_worker(pool); | |
3472 | if (!worker) | |
3473 | return NOTIFY_BAD; | |
3474 | ||
d565ed63 | 3475 | spin_lock_irq(&pool->lock); |
3ce63377 | 3476 | start_worker(worker); |
d565ed63 | 3477 | spin_unlock_irq(&pool->lock); |
3af24433 | 3478 | } |
8db25e78 | 3479 | break; |
3af24433 | 3480 | |
db7bccf4 TH |
3481 | case CPU_DOWN_FAILED: |
3482 | case CPU_ONLINE: | |
38db41d9 | 3483 | for_each_std_worker_pool(pool, cpu) { |
94cf58bb TH |
3484 | mutex_lock(&pool->assoc_mutex); |
3485 | spin_lock_irq(&pool->lock); | |
3486 | ||
24647570 | 3487 | pool->flags &= ~POOL_DISASSOCIATED; |
94cf58bb TH |
3488 | rebind_workers(pool); |
3489 | ||
3490 | spin_unlock_irq(&pool->lock); | |
3491 | mutex_unlock(&pool->assoc_mutex); | |
3492 | } | |
db7bccf4 | 3493 | break; |
00dfcaf7 | 3494 | } |
65758202 TH |
3495 | return NOTIFY_OK; |
3496 | } | |
3497 | ||
3498 | /* | |
3499 | * Workqueues should be brought down after normal priority CPU notifiers. | |
3500 | * This will be registered as low priority CPU notifier. | |
3501 | */ | |
9fdf9b73 | 3502 | static int __cpuinit workqueue_cpu_down_callback(struct notifier_block *nfb, |
65758202 TH |
3503 | unsigned long action, |
3504 | void *hcpu) | |
3505 | { | |
d84ff051 | 3506 | int cpu = (unsigned long)hcpu; |
8db25e78 TH |
3507 | struct work_struct unbind_work; |
3508 | ||
65758202 TH |
3509 | switch (action & ~CPU_TASKS_FROZEN) { |
3510 | case CPU_DOWN_PREPARE: | |
8db25e78 | 3511 | /* unbinding should happen on the local CPU */ |
706026c2 | 3512 | INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn); |
7635d2fd | 3513 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
8db25e78 TH |
3514 | flush_work(&unbind_work); |
3515 | break; | |
65758202 TH |
3516 | } |
3517 | return NOTIFY_OK; | |
3518 | } | |
3519 | ||
2d3854a3 | 3520 | #ifdef CONFIG_SMP |
8ccad40d | 3521 | |
2d3854a3 | 3522 | struct work_for_cpu { |
ed48ece2 | 3523 | struct work_struct work; |
2d3854a3 RR |
3524 | long (*fn)(void *); |
3525 | void *arg; | |
3526 | long ret; | |
3527 | }; | |
3528 | ||
ed48ece2 | 3529 | static void work_for_cpu_fn(struct work_struct *work) |
2d3854a3 | 3530 | { |
ed48ece2 TH |
3531 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
3532 | ||
2d3854a3 RR |
3533 | wfc->ret = wfc->fn(wfc->arg); |
3534 | } | |
3535 | ||
3536 | /** | |
3537 | * work_on_cpu - run a function in user context on a particular cpu | |
3538 | * @cpu: the cpu to run on | |
3539 | * @fn: the function to run | |
3540 | * @arg: the function arg | |
3541 | * | |
31ad9081 RR |
3542 | * This will return the value @fn returns. |
3543 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 3544 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 | 3545 | */ |
d84ff051 | 3546 | long work_on_cpu(int cpu, long (*fn)(void *), void *arg) |
2d3854a3 | 3547 | { |
ed48ece2 | 3548 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; |
6b44003e | 3549 | |
ed48ece2 TH |
3550 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
3551 | schedule_work_on(cpu, &wfc.work); | |
3552 | flush_work(&wfc.work); | |
2d3854a3 RR |
3553 | return wfc.ret; |
3554 | } | |
3555 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
3556 | #endif /* CONFIG_SMP */ | |
3557 | ||
a0a1a5fd TH |
3558 | #ifdef CONFIG_FREEZER |
3559 | ||
3560 | /** | |
3561 | * freeze_workqueues_begin - begin freezing workqueues | |
3562 | * | |
58a69cb4 TH |
3563 | * Start freezing workqueues. After this function returns, all freezable |
3564 | * workqueues will queue new works to their frozen_works list instead of | |
706026c2 | 3565 | * pool->worklist. |
a0a1a5fd TH |
3566 | * |
3567 | * CONTEXT: | |
d565ed63 | 3568 | * Grabs and releases workqueue_lock and pool->lock's. |
a0a1a5fd TH |
3569 | */ |
3570 | void freeze_workqueues_begin(void) | |
3571 | { | |
17116969 | 3572 | struct worker_pool *pool; |
24b8a847 TH |
3573 | struct workqueue_struct *wq; |
3574 | struct pool_workqueue *pwq; | |
17116969 | 3575 | int id; |
a0a1a5fd | 3576 | |
e98d5b16 | 3577 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3578 | |
6183c009 | 3579 | WARN_ON_ONCE(workqueue_freezing); |
a0a1a5fd TH |
3580 | workqueue_freezing = true; |
3581 | ||
24b8a847 | 3582 | /* set FREEZING */ |
17116969 | 3583 | for_each_pool(pool, id) { |
17116969 | 3584 | spin_lock(&pool->lock); |
17116969 TH |
3585 | WARN_ON_ONCE(pool->flags & POOL_FREEZING); |
3586 | pool->flags |= POOL_FREEZING; | |
24b8a847 TH |
3587 | spin_unlock(&pool->lock); |
3588 | } | |
a0a1a5fd | 3589 | |
24b8a847 TH |
3590 | /* suppress further executions by setting max_active to zero */ |
3591 | list_for_each_entry(wq, &workqueues, list) { | |
3592 | if (!(wq->flags & WQ_FREEZABLE)) | |
3593 | continue; | |
8b03ae3c | 3594 | |
24b8a847 TH |
3595 | for_each_pwq(pwq, wq) { |
3596 | spin_lock(&pwq->pool->lock); | |
3597 | pwq->max_active = 0; | |
3598 | spin_unlock(&pwq->pool->lock); | |
a1056305 | 3599 | } |
a0a1a5fd TH |
3600 | } |
3601 | ||
e98d5b16 | 3602 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3603 | } |
3604 | ||
3605 | /** | |
58a69cb4 | 3606 | * freeze_workqueues_busy - are freezable workqueues still busy? |
a0a1a5fd TH |
3607 | * |
3608 | * Check whether freezing is complete. This function must be called | |
3609 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
3610 | * | |
3611 | * CONTEXT: | |
3612 | * Grabs and releases workqueue_lock. | |
3613 | * | |
3614 | * RETURNS: | |
58a69cb4 TH |
3615 | * %true if some freezable workqueues are still busy. %false if freezing |
3616 | * is complete. | |
a0a1a5fd TH |
3617 | */ |
3618 | bool freeze_workqueues_busy(void) | |
3619 | { | |
a0a1a5fd | 3620 | bool busy = false; |
24b8a847 TH |
3621 | struct workqueue_struct *wq; |
3622 | struct pool_workqueue *pwq; | |
a0a1a5fd | 3623 | |
e98d5b16 | 3624 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3625 | |
6183c009 | 3626 | WARN_ON_ONCE(!workqueue_freezing); |
a0a1a5fd | 3627 | |
24b8a847 TH |
3628 | list_for_each_entry(wq, &workqueues, list) { |
3629 | if (!(wq->flags & WQ_FREEZABLE)) | |
3630 | continue; | |
a0a1a5fd TH |
3631 | /* |
3632 | * nr_active is monotonically decreasing. It's safe | |
3633 | * to peek without lock. | |
3634 | */ | |
24b8a847 | 3635 | for_each_pwq(pwq, wq) { |
6183c009 | 3636 | WARN_ON_ONCE(pwq->nr_active < 0); |
112202d9 | 3637 | if (pwq->nr_active) { |
a0a1a5fd TH |
3638 | busy = true; |
3639 | goto out_unlock; | |
3640 | } | |
3641 | } | |
3642 | } | |
3643 | out_unlock: | |
e98d5b16 | 3644 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3645 | return busy; |
3646 | } | |
3647 | ||
3648 | /** | |
3649 | * thaw_workqueues - thaw workqueues | |
3650 | * | |
3651 | * Thaw workqueues. Normal queueing is restored and all collected | |
706026c2 | 3652 | * frozen works are transferred to their respective pool worklists. |
a0a1a5fd TH |
3653 | * |
3654 | * CONTEXT: | |
d565ed63 | 3655 | * Grabs and releases workqueue_lock and pool->lock's. |
a0a1a5fd TH |
3656 | */ |
3657 | void thaw_workqueues(void) | |
3658 | { | |
24b8a847 TH |
3659 | struct workqueue_struct *wq; |
3660 | struct pool_workqueue *pwq; | |
3661 | struct worker_pool *pool; | |
3662 | int id; | |
a0a1a5fd | 3663 | |
e98d5b16 | 3664 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd TH |
3665 | |
3666 | if (!workqueue_freezing) | |
3667 | goto out_unlock; | |
3668 | ||
24b8a847 TH |
3669 | /* clear FREEZING */ |
3670 | for_each_pool(pool, id) { | |
3671 | spin_lock(&pool->lock); | |
3672 | WARN_ON_ONCE(!(pool->flags & POOL_FREEZING)); | |
3673 | pool->flags &= ~POOL_FREEZING; | |
3674 | spin_unlock(&pool->lock); | |
3675 | } | |
8b03ae3c | 3676 | |
24b8a847 TH |
3677 | /* restore max_active and repopulate worklist */ |
3678 | list_for_each_entry(wq, &workqueues, list) { | |
3679 | if (!(wq->flags & WQ_FREEZABLE)) | |
3680 | continue; | |
a1056305 | 3681 | |
24b8a847 TH |
3682 | for_each_pwq(pwq, wq) { |
3683 | spin_lock(&pwq->pool->lock); | |
3684 | pwq_set_max_active(pwq, wq->saved_max_active); | |
3685 | spin_unlock(&pwq->pool->lock); | |
d565ed63 | 3686 | } |
a0a1a5fd TH |
3687 | } |
3688 | ||
24b8a847 TH |
3689 | /* kick workers */ |
3690 | for_each_pool(pool, id) { | |
3691 | spin_lock(&pool->lock); | |
3692 | wake_up_worker(pool); | |
3693 | spin_unlock(&pool->lock); | |
3694 | } | |
3695 | ||
a0a1a5fd TH |
3696 | workqueue_freezing = false; |
3697 | out_unlock: | |
e98d5b16 | 3698 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3699 | } |
3700 | #endif /* CONFIG_FREEZER */ | |
3701 | ||
6ee0578b | 3702 | static int __init init_workqueues(void) |
1da177e4 | 3703 | { |
d84ff051 | 3704 | int cpu; |
c34056a3 | 3705 | |
7c3eed5c TH |
3706 | /* make sure we have enough bits for OFFQ pool ID */ |
3707 | BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_POOL_SHIFT)) < | |
6be19588 | 3708 | WORK_CPU_END * NR_STD_WORKER_POOLS); |
b5490077 | 3709 | |
e904e6c2 TH |
3710 | WARN_ON(__alignof__(struct pool_workqueue) < __alignof__(long long)); |
3711 | ||
3712 | pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC); | |
3713 | ||
65758202 | 3714 | cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); |
a5b4e57d | 3715 | hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); |
8b03ae3c | 3716 | |
706026c2 TH |
3717 | /* initialize CPU pools */ |
3718 | for_each_wq_cpu(cpu) { | |
4ce62e9e | 3719 | struct worker_pool *pool; |
8b03ae3c | 3720 | |
38db41d9 | 3721 | for_each_std_worker_pool(pool, cpu) { |
d565ed63 | 3722 | spin_lock_init(&pool->lock); |
ec22ca5e | 3723 | pool->cpu = cpu; |
24647570 | 3724 | pool->flags |= POOL_DISASSOCIATED; |
4ce62e9e TH |
3725 | INIT_LIST_HEAD(&pool->worklist); |
3726 | INIT_LIST_HEAD(&pool->idle_list); | |
c9e7cf27 | 3727 | hash_init(pool->busy_hash); |
e7577c50 | 3728 | |
4ce62e9e TH |
3729 | init_timer_deferrable(&pool->idle_timer); |
3730 | pool->idle_timer.function = idle_worker_timeout; | |
3731 | pool->idle_timer.data = (unsigned long)pool; | |
e22bee78 | 3732 | |
706026c2 | 3733 | setup_timer(&pool->mayday_timer, pool_mayday_timeout, |
4ce62e9e TH |
3734 | (unsigned long)pool); |
3735 | ||
b2eb83d1 | 3736 | mutex_init(&pool->assoc_mutex); |
4ce62e9e | 3737 | ida_init(&pool->worker_ida); |
9daf9e67 TH |
3738 | |
3739 | /* alloc pool ID */ | |
3740 | BUG_ON(worker_pool_assign_id(pool)); | |
4ce62e9e | 3741 | } |
8b03ae3c TH |
3742 | } |
3743 | ||
e22bee78 | 3744 | /* create the initial worker */ |
706026c2 | 3745 | for_each_online_wq_cpu(cpu) { |
4ce62e9e | 3746 | struct worker_pool *pool; |
e22bee78 | 3747 | |
38db41d9 | 3748 | for_each_std_worker_pool(pool, cpu) { |
4ce62e9e TH |
3749 | struct worker *worker; |
3750 | ||
24647570 TH |
3751 | if (cpu != WORK_CPU_UNBOUND) |
3752 | pool->flags &= ~POOL_DISASSOCIATED; | |
3753 | ||
bc2ae0f5 | 3754 | worker = create_worker(pool); |
4ce62e9e | 3755 | BUG_ON(!worker); |
d565ed63 | 3756 | spin_lock_irq(&pool->lock); |
4ce62e9e | 3757 | start_worker(worker); |
d565ed63 | 3758 | spin_unlock_irq(&pool->lock); |
4ce62e9e | 3759 | } |
e22bee78 TH |
3760 | } |
3761 | ||
d320c038 | 3762 | system_wq = alloc_workqueue("events", 0, 0); |
1aabe902 | 3763 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
d320c038 | 3764 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
f3421797 TH |
3765 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
3766 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
3767 | system_freezable_wq = alloc_workqueue("events_freezable", |
3768 | WQ_FREEZABLE, 0); | |
1aabe902 | 3769 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
ae930e0f | 3770 | !system_unbound_wq || !system_freezable_wq); |
6ee0578b | 3771 | return 0; |
1da177e4 | 3772 | } |
6ee0578b | 3773 | early_initcall(init_workqueues); |