]>
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> |
76af4d93 | 45 | #include <linux/rculist.h> |
e22bee78 | 46 | |
ea138446 | 47 | #include "workqueue_internal.h" |
1da177e4 | 48 | |
c8e55f36 | 49 | enum { |
24647570 TH |
50 | /* |
51 | * worker_pool flags | |
bc2ae0f5 | 52 | * |
24647570 | 53 | * A bound pool is either associated or disassociated with its CPU. |
bc2ae0f5 TH |
54 | * While associated (!DISASSOCIATED), all workers are bound to the |
55 | * CPU and none has %WORKER_UNBOUND set and concurrency management | |
56 | * is in effect. | |
57 | * | |
58 | * While DISASSOCIATED, the cpu may be offline and all workers have | |
59 | * %WORKER_UNBOUND set and concurrency management disabled, and may | |
24647570 | 60 | * be executing on any CPU. The pool behaves as an unbound one. |
bc2ae0f5 TH |
61 | * |
62 | * Note that DISASSOCIATED can be flipped only while holding | |
24647570 TH |
63 | * assoc_mutex to avoid changing binding state while |
64 | * create_worker() is in progress. | |
bc2ae0f5 | 65 | */ |
11ebea50 | 66 | POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */ |
552a37e9 | 67 | POOL_MANAGING_WORKERS = 1 << 1, /* managing workers */ |
24647570 | 68 | POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ |
35b6bb63 | 69 | POOL_FREEZING = 1 << 3, /* freeze in progress */ |
db7bccf4 | 70 | |
c8e55f36 TH |
71 | /* worker flags */ |
72 | WORKER_STARTED = 1 << 0, /* started */ | |
73 | WORKER_DIE = 1 << 1, /* die die die */ | |
74 | WORKER_IDLE = 1 << 2, /* is idle */ | |
e22bee78 | 75 | WORKER_PREP = 1 << 3, /* preparing to run works */ |
fb0e7beb | 76 | WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ |
f3421797 | 77 | WORKER_UNBOUND = 1 << 7, /* worker is unbound */ |
e22bee78 | 78 | |
5f7dabfd | 79 | WORKER_NOT_RUNNING = WORKER_PREP | WORKER_UNBOUND | |
403c821d | 80 | WORKER_CPU_INTENSIVE, |
db7bccf4 | 81 | |
e34cdddb | 82 | NR_STD_WORKER_POOLS = 2, /* # standard pools per cpu */ |
4ce62e9e | 83 | |
c8e55f36 | 84 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ |
db7bccf4 | 85 | |
e22bee78 TH |
86 | MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */ |
87 | IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */ | |
88 | ||
3233cdbd TH |
89 | MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2, |
90 | /* call for help after 10ms | |
91 | (min two ticks) */ | |
e22bee78 TH |
92 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ |
93 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
e22bee78 TH |
94 | |
95 | /* | |
96 | * Rescue workers are used only on emergencies and shared by | |
97 | * all cpus. Give -20. | |
98 | */ | |
99 | RESCUER_NICE_LEVEL = -20, | |
3270476a | 100 | HIGHPRI_NICE_LEVEL = -20, |
c8e55f36 | 101 | }; |
1da177e4 LT |
102 | |
103 | /* | |
4690c4ab TH |
104 | * Structure fields follow one of the following exclusion rules. |
105 | * | |
e41e704b TH |
106 | * I: Modifiable by initialization/destruction paths and read-only for |
107 | * everyone else. | |
4690c4ab | 108 | * |
e22bee78 TH |
109 | * P: Preemption protected. Disabling preemption is enough and should |
110 | * only be modified and accessed from the local cpu. | |
111 | * | |
d565ed63 | 112 | * L: pool->lock protected. Access with pool->lock held. |
4690c4ab | 113 | * |
d565ed63 TH |
114 | * X: During normal operation, modification requires pool->lock and should |
115 | * be done only from local cpu. Either disabling preemption on local | |
116 | * cpu or grabbing pool->lock is enough for read access. If | |
117 | * POOL_DISASSOCIATED is set, it's identical to L. | |
e22bee78 | 118 | * |
73f53c4a TH |
119 | * F: wq->flush_mutex protected. |
120 | * | |
4690c4ab | 121 | * W: workqueue_lock protected. |
76af4d93 TH |
122 | * |
123 | * R: workqueue_lock protected for writes. Sched-RCU protected for reads. | |
1da177e4 | 124 | */ |
1da177e4 | 125 | |
2eaebdb3 | 126 | /* struct worker is defined in workqueue_internal.h */ |
c34056a3 | 127 | |
bd7bdd43 | 128 | struct worker_pool { |
d565ed63 | 129 | spinlock_t lock; /* the pool lock */ |
d84ff051 | 130 | int cpu; /* I: the associated cpu */ |
9daf9e67 | 131 | int id; /* I: pool ID */ |
11ebea50 | 132 | unsigned int flags; /* X: flags */ |
bd7bdd43 TH |
133 | |
134 | struct list_head worklist; /* L: list of pending works */ | |
135 | int nr_workers; /* L: total number of workers */ | |
ea1abd61 LJ |
136 | |
137 | /* nr_idle includes the ones off idle_list for rebinding */ | |
bd7bdd43 TH |
138 | int nr_idle; /* L: currently idle ones */ |
139 | ||
140 | struct list_head idle_list; /* X: list of idle workers */ | |
141 | struct timer_list idle_timer; /* L: worker idle timeout */ | |
142 | struct timer_list mayday_timer; /* L: SOS timer for workers */ | |
143 | ||
c9e7cf27 TH |
144 | /* workers are chained either in busy_hash or idle_list */ |
145 | DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER); | |
146 | /* L: hash of busy workers */ | |
147 | ||
24647570 | 148 | struct mutex assoc_mutex; /* protect POOL_DISASSOCIATED */ |
bd7bdd43 | 149 | struct ida worker_ida; /* L: for worker IDs */ |
e19e397a TH |
150 | |
151 | /* | |
152 | * The current concurrency level. As it's likely to be accessed | |
153 | * from other CPUs during try_to_wake_up(), put it in a separate | |
154 | * cacheline. | |
155 | */ | |
156 | atomic_t nr_running ____cacheline_aligned_in_smp; | |
8b03ae3c TH |
157 | } ____cacheline_aligned_in_smp; |
158 | ||
1da177e4 | 159 | /* |
112202d9 TH |
160 | * The per-pool workqueue. While queued, the lower WORK_STRUCT_FLAG_BITS |
161 | * of work_struct->data are used for flags and the remaining high bits | |
162 | * point to the pwq; thus, pwqs need to be aligned at two's power of the | |
163 | * number of flag bits. | |
1da177e4 | 164 | */ |
112202d9 | 165 | struct pool_workqueue { |
bd7bdd43 | 166 | struct worker_pool *pool; /* I: the associated pool */ |
4690c4ab | 167 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
168 | int work_color; /* L: current color */ |
169 | int flush_color; /* L: flushing color */ | |
170 | int nr_in_flight[WORK_NR_COLORS]; | |
171 | /* L: nr of in_flight works */ | |
1e19ffc6 | 172 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 173 | int max_active; /* L: max active works */ |
1e19ffc6 | 174 | struct list_head delayed_works; /* L: delayed works */ |
76af4d93 | 175 | struct list_head pwqs_node; /* R: node on wq->pwqs */ |
493a1724 | 176 | struct list_head mayday_node; /* W: node on wq->maydays */ |
e904e6c2 | 177 | } __aligned(1 << WORK_STRUCT_FLAG_BITS); |
1da177e4 | 178 | |
73f53c4a TH |
179 | /* |
180 | * Structure used to wait for workqueue flush. | |
181 | */ | |
182 | struct wq_flusher { | |
183 | struct list_head list; /* F: list of flushers */ | |
184 | int flush_color; /* F: flush color waiting for */ | |
185 | struct completion done; /* flush completion */ | |
186 | }; | |
187 | ||
1da177e4 LT |
188 | /* |
189 | * The externally visible workqueue abstraction is an array of | |
190 | * per-CPU workqueues: | |
191 | */ | |
192 | struct workqueue_struct { | |
9c5a2ba7 | 193 | unsigned int flags; /* W: WQ_* flags */ |
420c0ddb | 194 | struct pool_workqueue __percpu *cpu_pwqs; /* I: per-cpu pwq's */ |
76af4d93 | 195 | struct list_head pwqs; /* R: all pwqs of this wq */ |
4690c4ab | 196 | struct list_head list; /* W: list of all workqueues */ |
73f53c4a TH |
197 | |
198 | struct mutex flush_mutex; /* protects wq flushing */ | |
199 | int work_color; /* F: current work color */ | |
200 | int flush_color; /* F: current flush color */ | |
112202d9 | 201 | atomic_t nr_pwqs_to_flush; /* flush in progress */ |
73f53c4a TH |
202 | struct wq_flusher *first_flusher; /* F: first flusher */ |
203 | struct list_head flusher_queue; /* F: flush waiters */ | |
204 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
205 | ||
493a1724 | 206 | struct list_head maydays; /* W: pwqs requesting rescue */ |
e22bee78 TH |
207 | struct worker *rescuer; /* I: rescue worker */ |
208 | ||
9c5a2ba7 | 209 | int nr_drainers; /* W: drain in progress */ |
112202d9 | 210 | int saved_max_active; /* W: saved pwq max_active */ |
4e6045f1 | 211 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 212 | struct lockdep_map lockdep_map; |
4e6045f1 | 213 | #endif |
b196be89 | 214 | char name[]; /* I: workqueue name */ |
1da177e4 LT |
215 | }; |
216 | ||
e904e6c2 TH |
217 | static struct kmem_cache *pwq_cache; |
218 | ||
d320c038 | 219 | struct workqueue_struct *system_wq __read_mostly; |
d320c038 | 220 | EXPORT_SYMBOL_GPL(system_wq); |
044c782c | 221 | struct workqueue_struct *system_highpri_wq __read_mostly; |
1aabe902 | 222 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
044c782c | 223 | struct workqueue_struct *system_long_wq __read_mostly; |
d320c038 | 224 | EXPORT_SYMBOL_GPL(system_long_wq); |
044c782c | 225 | struct workqueue_struct *system_unbound_wq __read_mostly; |
f3421797 | 226 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
044c782c | 227 | struct workqueue_struct *system_freezable_wq __read_mostly; |
24d51add | 228 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
d320c038 | 229 | |
97bd2347 TH |
230 | #define CREATE_TRACE_POINTS |
231 | #include <trace/events/workqueue.h> | |
232 | ||
76af4d93 TH |
233 | #define assert_rcu_or_wq_lock() \ |
234 | rcu_lockdep_assert(rcu_read_lock_sched_held() || \ | |
235 | lockdep_is_held(&workqueue_lock), \ | |
236 | "sched RCU or workqueue lock should be held") | |
237 | ||
38db41d9 | 238 | #define for_each_std_worker_pool(pool, cpu) \ |
a60dc39c TH |
239 | for ((pool) = &std_worker_pools(cpu)[0]; \ |
240 | (pool) < &std_worker_pools(cpu)[NR_STD_WORKER_POOLS]; (pool)++) | |
4ce62e9e | 241 | |
b67bfe0d SL |
242 | #define for_each_busy_worker(worker, i, pool) \ |
243 | hash_for_each(pool->busy_hash, i, worker, hentry) | |
db7bccf4 | 244 | |
706026c2 TH |
245 | static inline int __next_wq_cpu(int cpu, const struct cpumask *mask, |
246 | unsigned int sw) | |
f3421797 TH |
247 | { |
248 | if (cpu < nr_cpu_ids) { | |
249 | if (sw & 1) { | |
250 | cpu = cpumask_next(cpu, mask); | |
251 | if (cpu < nr_cpu_ids) | |
252 | return cpu; | |
253 | } | |
254 | if (sw & 2) | |
255 | return WORK_CPU_UNBOUND; | |
256 | } | |
6be19588 | 257 | return WORK_CPU_END; |
f3421797 TH |
258 | } |
259 | ||
09884951 TH |
260 | /* |
261 | * CPU iterators | |
262 | * | |
706026c2 | 263 | * An extra cpu number is defined using an invalid cpu number |
09884951 | 264 | * (WORK_CPU_UNBOUND) to host workqueues which are not bound to any |
706026c2 TH |
265 | * specific CPU. The following iterators are similar to for_each_*_cpu() |
266 | * iterators but also considers the unbound CPU. | |
09884951 | 267 | * |
706026c2 TH |
268 | * for_each_wq_cpu() : possible CPUs + WORK_CPU_UNBOUND |
269 | * for_each_online_wq_cpu() : online CPUs + WORK_CPU_UNBOUND | |
09884951 | 270 | */ |
706026c2 TH |
271 | #define for_each_wq_cpu(cpu) \ |
272 | for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, 3); \ | |
6be19588 | 273 | (cpu) < WORK_CPU_END; \ |
706026c2 | 274 | (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, 3)) |
f3421797 | 275 | |
706026c2 TH |
276 | #define for_each_online_wq_cpu(cpu) \ |
277 | for ((cpu) = __next_wq_cpu(-1, cpu_online_mask, 3); \ | |
6be19588 | 278 | (cpu) < WORK_CPU_END; \ |
706026c2 | 279 | (cpu) = __next_wq_cpu((cpu), cpu_online_mask, 3)) |
f3421797 | 280 | |
17116969 TH |
281 | /** |
282 | * for_each_pool - iterate through all worker_pools in the system | |
283 | * @pool: iteration cursor | |
284 | * @id: integer used for iteration | |
fa1b54e6 TH |
285 | * |
286 | * This must be called either with workqueue_lock held or sched RCU read | |
287 | * locked. If the pool needs to be used beyond the locking in effect, the | |
288 | * caller is responsible for guaranteeing that the pool stays online. | |
289 | * | |
290 | * The if/else clause exists only for the lockdep assertion and can be | |
291 | * ignored. | |
17116969 TH |
292 | */ |
293 | #define for_each_pool(pool, id) \ | |
fa1b54e6 TH |
294 | idr_for_each_entry(&worker_pool_idr, pool, id) \ |
295 | if (({ assert_rcu_or_wq_lock(); false; })) { } \ | |
296 | else | |
17116969 | 297 | |
49e3cf44 TH |
298 | /** |
299 | * for_each_pwq - iterate through all pool_workqueues of the specified workqueue | |
300 | * @pwq: iteration cursor | |
301 | * @wq: the target workqueue | |
76af4d93 TH |
302 | * |
303 | * This must be called either with workqueue_lock held or sched RCU read | |
304 | * locked. If the pwq needs to be used beyond the locking in effect, the | |
305 | * caller is responsible for guaranteeing that the pwq stays online. | |
306 | * | |
307 | * The if/else clause exists only for the lockdep assertion and can be | |
308 | * ignored. | |
49e3cf44 TH |
309 | */ |
310 | #define for_each_pwq(pwq, wq) \ | |
76af4d93 TH |
311 | list_for_each_entry_rcu((pwq), &(wq)->pwqs, pwqs_node) \ |
312 | if (({ assert_rcu_or_wq_lock(); false; })) { } \ | |
313 | else | |
f3421797 | 314 | |
dc186ad7 TG |
315 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
316 | ||
317 | static struct debug_obj_descr work_debug_descr; | |
318 | ||
99777288 SG |
319 | static void *work_debug_hint(void *addr) |
320 | { | |
321 | return ((struct work_struct *) addr)->func; | |
322 | } | |
323 | ||
dc186ad7 TG |
324 | /* |
325 | * fixup_init is called when: | |
326 | * - an active object is initialized | |
327 | */ | |
328 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
329 | { | |
330 | struct work_struct *work = addr; | |
331 | ||
332 | switch (state) { | |
333 | case ODEBUG_STATE_ACTIVE: | |
334 | cancel_work_sync(work); | |
335 | debug_object_init(work, &work_debug_descr); | |
336 | return 1; | |
337 | default: | |
338 | return 0; | |
339 | } | |
340 | } | |
341 | ||
342 | /* | |
343 | * fixup_activate is called when: | |
344 | * - an active object is activated | |
345 | * - an unknown object is activated (might be a statically initialized object) | |
346 | */ | |
347 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
348 | { | |
349 | struct work_struct *work = addr; | |
350 | ||
351 | switch (state) { | |
352 | ||
353 | case ODEBUG_STATE_NOTAVAILABLE: | |
354 | /* | |
355 | * This is not really a fixup. The work struct was | |
356 | * statically initialized. We just make sure that it | |
357 | * is tracked in the object tracker. | |
358 | */ | |
22df02bb | 359 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
360 | debug_object_init(work, &work_debug_descr); |
361 | debug_object_activate(work, &work_debug_descr); | |
362 | return 0; | |
363 | } | |
364 | WARN_ON_ONCE(1); | |
365 | return 0; | |
366 | ||
367 | case ODEBUG_STATE_ACTIVE: | |
368 | WARN_ON(1); | |
369 | ||
370 | default: | |
371 | return 0; | |
372 | } | |
373 | } | |
374 | ||
375 | /* | |
376 | * fixup_free is called when: | |
377 | * - an active object is freed | |
378 | */ | |
379 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
380 | { | |
381 | struct work_struct *work = addr; | |
382 | ||
383 | switch (state) { | |
384 | case ODEBUG_STATE_ACTIVE: | |
385 | cancel_work_sync(work); | |
386 | debug_object_free(work, &work_debug_descr); | |
387 | return 1; | |
388 | default: | |
389 | return 0; | |
390 | } | |
391 | } | |
392 | ||
393 | static struct debug_obj_descr work_debug_descr = { | |
394 | .name = "work_struct", | |
99777288 | 395 | .debug_hint = work_debug_hint, |
dc186ad7 TG |
396 | .fixup_init = work_fixup_init, |
397 | .fixup_activate = work_fixup_activate, | |
398 | .fixup_free = work_fixup_free, | |
399 | }; | |
400 | ||
401 | static inline void debug_work_activate(struct work_struct *work) | |
402 | { | |
403 | debug_object_activate(work, &work_debug_descr); | |
404 | } | |
405 | ||
406 | static inline void debug_work_deactivate(struct work_struct *work) | |
407 | { | |
408 | debug_object_deactivate(work, &work_debug_descr); | |
409 | } | |
410 | ||
411 | void __init_work(struct work_struct *work, int onstack) | |
412 | { | |
413 | if (onstack) | |
414 | debug_object_init_on_stack(work, &work_debug_descr); | |
415 | else | |
416 | debug_object_init(work, &work_debug_descr); | |
417 | } | |
418 | EXPORT_SYMBOL_GPL(__init_work); | |
419 | ||
420 | void destroy_work_on_stack(struct work_struct *work) | |
421 | { | |
422 | debug_object_free(work, &work_debug_descr); | |
423 | } | |
424 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
425 | ||
426 | #else | |
427 | static inline void debug_work_activate(struct work_struct *work) { } | |
428 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
429 | #endif | |
430 | ||
95402b38 GS |
431 | /* Serializes the accesses to the list of workqueues. */ |
432 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 433 | static LIST_HEAD(workqueues); |
a0a1a5fd | 434 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 435 | |
e22bee78 | 436 | /* |
e19e397a TH |
437 | * The CPU and unbound standard worker pools. The unbound ones have |
438 | * POOL_DISASSOCIATED set, and their workers have WORKER_UNBOUND set. | |
f3421797 | 439 | */ |
e19e397a TH |
440 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS], |
441 | cpu_std_worker_pools); | |
a60dc39c | 442 | static struct worker_pool unbound_std_worker_pools[NR_STD_WORKER_POOLS]; |
f3421797 | 443 | |
fa1b54e6 TH |
444 | /* |
445 | * idr of all pools. Modifications are protected by workqueue_lock. Read | |
446 | * accesses are protected by sched-RCU protected. | |
447 | */ | |
9daf9e67 TH |
448 | static DEFINE_IDR(worker_pool_idr); |
449 | ||
c34056a3 | 450 | static int worker_thread(void *__worker); |
1da177e4 | 451 | |
a60dc39c | 452 | static struct worker_pool *std_worker_pools(int cpu) |
8b03ae3c | 453 | { |
f3421797 | 454 | if (cpu != WORK_CPU_UNBOUND) |
a60dc39c | 455 | return per_cpu(cpu_std_worker_pools, cpu); |
f3421797 | 456 | else |
a60dc39c | 457 | return unbound_std_worker_pools; |
8b03ae3c TH |
458 | } |
459 | ||
4e8f0a60 TH |
460 | static int std_worker_pool_pri(struct worker_pool *pool) |
461 | { | |
a60dc39c | 462 | return pool - std_worker_pools(pool->cpu); |
4e8f0a60 TH |
463 | } |
464 | ||
9daf9e67 TH |
465 | /* allocate ID and assign it to @pool */ |
466 | static int worker_pool_assign_id(struct worker_pool *pool) | |
467 | { | |
468 | int ret; | |
469 | ||
fa1b54e6 TH |
470 | do { |
471 | if (!idr_pre_get(&worker_pool_idr, GFP_KERNEL)) | |
472 | return -ENOMEM; | |
9daf9e67 | 473 | |
fa1b54e6 TH |
474 | spin_lock_irq(&workqueue_lock); |
475 | ret = idr_get_new(&worker_pool_idr, pool, &pool->id); | |
476 | spin_unlock_irq(&workqueue_lock); | |
477 | } while (ret == -EAGAIN); | |
9daf9e67 | 478 | |
fa1b54e6 | 479 | return ret; |
7c3eed5c TH |
480 | } |
481 | ||
d565ed63 TH |
482 | static struct worker_pool *get_std_worker_pool(int cpu, bool highpri) |
483 | { | |
a60dc39c | 484 | struct worker_pool *pools = std_worker_pools(cpu); |
d565ed63 | 485 | |
a60dc39c | 486 | return &pools[highpri]; |
d565ed63 TH |
487 | } |
488 | ||
76af4d93 TH |
489 | /** |
490 | * first_pwq - return the first pool_workqueue of the specified workqueue | |
491 | * @wq: the target workqueue | |
492 | * | |
493 | * This must be called either with workqueue_lock held or sched RCU read | |
494 | * locked. If the pwq needs to be used beyond the locking in effect, the | |
495 | * caller is responsible for guaranteeing that the pwq stays online. | |
496 | */ | |
7fb98ea7 | 497 | static struct pool_workqueue *first_pwq(struct workqueue_struct *wq) |
b1f4ec17 | 498 | { |
76af4d93 TH |
499 | assert_rcu_or_wq_lock(); |
500 | return list_first_or_null_rcu(&wq->pwqs, struct pool_workqueue, | |
501 | pwqs_node); | |
b1f4ec17 ON |
502 | } |
503 | ||
73f53c4a TH |
504 | static unsigned int work_color_to_flags(int color) |
505 | { | |
506 | return color << WORK_STRUCT_COLOR_SHIFT; | |
507 | } | |
508 | ||
509 | static int get_work_color(struct work_struct *work) | |
510 | { | |
511 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
512 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
513 | } | |
514 | ||
515 | static int work_next_color(int color) | |
516 | { | |
517 | return (color + 1) % WORK_NR_COLORS; | |
518 | } | |
1da177e4 | 519 | |
14441960 | 520 | /* |
112202d9 TH |
521 | * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data |
522 | * contain the pointer to the queued pwq. Once execution starts, the flag | |
7c3eed5c | 523 | * is cleared and the high bits contain OFFQ flags and pool ID. |
7a22ad75 | 524 | * |
112202d9 TH |
525 | * set_work_pwq(), set_work_pool_and_clear_pending(), mark_work_canceling() |
526 | * and clear_work_data() can be used to set the pwq, pool or clear | |
bbb68dfa TH |
527 | * work->data. These functions should only be called while the work is |
528 | * owned - ie. while the PENDING bit is set. | |
7a22ad75 | 529 | * |
112202d9 | 530 | * get_work_pool() and get_work_pwq() can be used to obtain the pool or pwq |
7c3eed5c | 531 | * corresponding to a work. Pool is available once the work has been |
112202d9 | 532 | * queued anywhere after initialization until it is sync canceled. pwq is |
7c3eed5c | 533 | * available only while the work item is queued. |
7a22ad75 | 534 | * |
bbb68dfa TH |
535 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
536 | * canceled. While being canceled, a work item may have its PENDING set | |
537 | * but stay off timer and worklist for arbitrarily long and nobody should | |
538 | * try to steal the PENDING bit. | |
14441960 | 539 | */ |
7a22ad75 TH |
540 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
541 | unsigned long flags) | |
365970a1 | 542 | { |
6183c009 | 543 | WARN_ON_ONCE(!work_pending(work)); |
7a22ad75 TH |
544 | atomic_long_set(&work->data, data | flags | work_static(work)); |
545 | } | |
365970a1 | 546 | |
112202d9 | 547 | static void set_work_pwq(struct work_struct *work, struct pool_workqueue *pwq, |
7a22ad75 TH |
548 | unsigned long extra_flags) |
549 | { | |
112202d9 TH |
550 | set_work_data(work, (unsigned long)pwq, |
551 | WORK_STRUCT_PENDING | WORK_STRUCT_PWQ | extra_flags); | |
365970a1 DH |
552 | } |
553 | ||
4468a00f LJ |
554 | static void set_work_pool_and_keep_pending(struct work_struct *work, |
555 | int pool_id) | |
556 | { | |
557 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, | |
558 | WORK_STRUCT_PENDING); | |
559 | } | |
560 | ||
7c3eed5c TH |
561 | static void set_work_pool_and_clear_pending(struct work_struct *work, |
562 | int pool_id) | |
7a22ad75 | 563 | { |
23657bb1 TH |
564 | /* |
565 | * The following wmb is paired with the implied mb in | |
566 | * test_and_set_bit(PENDING) and ensures all updates to @work made | |
567 | * here are visible to and precede any updates by the next PENDING | |
568 | * owner. | |
569 | */ | |
570 | smp_wmb(); | |
7c3eed5c | 571 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0); |
7a22ad75 | 572 | } |
f756d5e2 | 573 | |
7a22ad75 | 574 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 575 | { |
7c3eed5c TH |
576 | smp_wmb(); /* see set_work_pool_and_clear_pending() */ |
577 | set_work_data(work, WORK_STRUCT_NO_POOL, 0); | |
1da177e4 LT |
578 | } |
579 | ||
112202d9 | 580 | static struct pool_workqueue *get_work_pwq(struct work_struct *work) |
b1f4ec17 | 581 | { |
e120153d | 582 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 583 | |
112202d9 | 584 | if (data & WORK_STRUCT_PWQ) |
e120153d TH |
585 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); |
586 | else | |
587 | return NULL; | |
4d707b9f ON |
588 | } |
589 | ||
7c3eed5c TH |
590 | /** |
591 | * get_work_pool - return the worker_pool a given work was associated with | |
592 | * @work: the work item of interest | |
593 | * | |
594 | * Return the worker_pool @work was last associated with. %NULL if none. | |
fa1b54e6 TH |
595 | * |
596 | * Pools are created and destroyed under workqueue_lock, and allows read | |
597 | * access under sched-RCU read lock. As such, this function should be | |
598 | * called under workqueue_lock or with preemption disabled. | |
599 | * | |
600 | * All fields of the returned pool are accessible as long as the above | |
601 | * mentioned locking is in effect. If the returned pool needs to be used | |
602 | * beyond the critical section, the caller is responsible for ensuring the | |
603 | * returned pool is and stays online. | |
7c3eed5c TH |
604 | */ |
605 | static struct worker_pool *get_work_pool(struct work_struct *work) | |
365970a1 | 606 | { |
e120153d | 607 | unsigned long data = atomic_long_read(&work->data); |
7c3eed5c | 608 | int pool_id; |
7a22ad75 | 609 | |
fa1b54e6 TH |
610 | assert_rcu_or_wq_lock(); |
611 | ||
112202d9 TH |
612 | if (data & WORK_STRUCT_PWQ) |
613 | return ((struct pool_workqueue *) | |
7c3eed5c | 614 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool; |
7a22ad75 | 615 | |
7c3eed5c TH |
616 | pool_id = data >> WORK_OFFQ_POOL_SHIFT; |
617 | if (pool_id == WORK_OFFQ_POOL_NONE) | |
7a22ad75 TH |
618 | return NULL; |
619 | ||
fa1b54e6 | 620 | return idr_find(&worker_pool_idr, pool_id); |
7c3eed5c TH |
621 | } |
622 | ||
623 | /** | |
624 | * get_work_pool_id - return the worker pool ID a given work is associated with | |
625 | * @work: the work item of interest | |
626 | * | |
627 | * Return the worker_pool ID @work was last associated with. | |
628 | * %WORK_OFFQ_POOL_NONE if none. | |
629 | */ | |
630 | static int get_work_pool_id(struct work_struct *work) | |
631 | { | |
54d5b7d0 LJ |
632 | unsigned long data = atomic_long_read(&work->data); |
633 | ||
112202d9 TH |
634 | if (data & WORK_STRUCT_PWQ) |
635 | return ((struct pool_workqueue *) | |
54d5b7d0 | 636 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->id; |
7c3eed5c | 637 | |
54d5b7d0 | 638 | return data >> WORK_OFFQ_POOL_SHIFT; |
7c3eed5c TH |
639 | } |
640 | ||
bbb68dfa TH |
641 | static void mark_work_canceling(struct work_struct *work) |
642 | { | |
7c3eed5c | 643 | unsigned long pool_id = get_work_pool_id(work); |
bbb68dfa | 644 | |
7c3eed5c TH |
645 | pool_id <<= WORK_OFFQ_POOL_SHIFT; |
646 | set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING); | |
bbb68dfa TH |
647 | } |
648 | ||
649 | static bool work_is_canceling(struct work_struct *work) | |
650 | { | |
651 | unsigned long data = atomic_long_read(&work->data); | |
652 | ||
112202d9 | 653 | return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING); |
bbb68dfa TH |
654 | } |
655 | ||
e22bee78 | 656 | /* |
3270476a TH |
657 | * Policy functions. These define the policies on how the global worker |
658 | * pools are managed. Unless noted otherwise, these functions assume that | |
d565ed63 | 659 | * they're being called with pool->lock held. |
e22bee78 TH |
660 | */ |
661 | ||
63d95a91 | 662 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 663 | { |
e19e397a | 664 | return !atomic_read(&pool->nr_running); |
a848e3b6 ON |
665 | } |
666 | ||
4594bf15 | 667 | /* |
e22bee78 TH |
668 | * Need to wake up a worker? Called from anything but currently |
669 | * running workers. | |
974271c4 TH |
670 | * |
671 | * Note that, because unbound workers never contribute to nr_running, this | |
706026c2 | 672 | * function will always return %true for unbound pools as long as the |
974271c4 | 673 | * worklist isn't empty. |
4594bf15 | 674 | */ |
63d95a91 | 675 | static bool need_more_worker(struct worker_pool *pool) |
365970a1 | 676 | { |
63d95a91 | 677 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
e22bee78 | 678 | } |
4594bf15 | 679 | |
e22bee78 | 680 | /* Can I start working? Called from busy but !running workers. */ |
63d95a91 | 681 | static bool may_start_working(struct worker_pool *pool) |
e22bee78 | 682 | { |
63d95a91 | 683 | return pool->nr_idle; |
e22bee78 TH |
684 | } |
685 | ||
686 | /* Do I need to keep working? Called from currently running workers. */ | |
63d95a91 | 687 | static bool keep_working(struct worker_pool *pool) |
e22bee78 | 688 | { |
e19e397a TH |
689 | return !list_empty(&pool->worklist) && |
690 | atomic_read(&pool->nr_running) <= 1; | |
e22bee78 TH |
691 | } |
692 | ||
693 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 694 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 695 | { |
63d95a91 | 696 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 697 | } |
365970a1 | 698 | |
e22bee78 | 699 | /* Do I need to be the manager? */ |
63d95a91 | 700 | static bool need_to_manage_workers(struct worker_pool *pool) |
e22bee78 | 701 | { |
63d95a91 | 702 | return need_to_create_worker(pool) || |
11ebea50 | 703 | (pool->flags & POOL_MANAGE_WORKERS); |
e22bee78 TH |
704 | } |
705 | ||
706 | /* Do we have too many workers and should some go away? */ | |
63d95a91 | 707 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 708 | { |
552a37e9 | 709 | bool managing = pool->flags & POOL_MANAGING_WORKERS; |
63d95a91 TH |
710 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
711 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 | 712 | |
ea1abd61 LJ |
713 | /* |
714 | * nr_idle and idle_list may disagree if idle rebinding is in | |
715 | * progress. Never return %true if idle_list is empty. | |
716 | */ | |
717 | if (list_empty(&pool->idle_list)) | |
718 | return false; | |
719 | ||
e22bee78 | 720 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
365970a1 DH |
721 | } |
722 | ||
4d707b9f | 723 | /* |
e22bee78 TH |
724 | * Wake up functions. |
725 | */ | |
726 | ||
7e11629d | 727 | /* Return the first worker. Safe with preemption disabled */ |
63d95a91 | 728 | static struct worker *first_worker(struct worker_pool *pool) |
7e11629d | 729 | { |
63d95a91 | 730 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
731 | return NULL; |
732 | ||
63d95a91 | 733 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
734 | } |
735 | ||
736 | /** | |
737 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 738 | * @pool: worker pool to wake worker from |
7e11629d | 739 | * |
63d95a91 | 740 | * Wake up the first idle worker of @pool. |
7e11629d TH |
741 | * |
742 | * CONTEXT: | |
d565ed63 | 743 | * spin_lock_irq(pool->lock). |
7e11629d | 744 | */ |
63d95a91 | 745 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 746 | { |
63d95a91 | 747 | struct worker *worker = first_worker(pool); |
7e11629d TH |
748 | |
749 | if (likely(worker)) | |
750 | wake_up_process(worker->task); | |
751 | } | |
752 | ||
d302f017 | 753 | /** |
e22bee78 TH |
754 | * wq_worker_waking_up - a worker is waking up |
755 | * @task: task waking up | |
756 | * @cpu: CPU @task is waking up to | |
757 | * | |
758 | * This function is called during try_to_wake_up() when a worker is | |
759 | * being awoken. | |
760 | * | |
761 | * CONTEXT: | |
762 | * spin_lock_irq(rq->lock) | |
763 | */ | |
d84ff051 | 764 | void wq_worker_waking_up(struct task_struct *task, int cpu) |
e22bee78 TH |
765 | { |
766 | struct worker *worker = kthread_data(task); | |
767 | ||
36576000 | 768 | if (!(worker->flags & WORKER_NOT_RUNNING)) { |
ec22ca5e | 769 | WARN_ON_ONCE(worker->pool->cpu != cpu); |
e19e397a | 770 | atomic_inc(&worker->pool->nr_running); |
36576000 | 771 | } |
e22bee78 TH |
772 | } |
773 | ||
774 | /** | |
775 | * wq_worker_sleeping - a worker is going to sleep | |
776 | * @task: task going to sleep | |
777 | * @cpu: CPU in question, must be the current CPU number | |
778 | * | |
779 | * This function is called during schedule() when a busy worker is | |
780 | * going to sleep. Worker on the same cpu can be woken up by | |
781 | * returning pointer to its task. | |
782 | * | |
783 | * CONTEXT: | |
784 | * spin_lock_irq(rq->lock) | |
785 | * | |
786 | * RETURNS: | |
787 | * Worker task on @cpu to wake up, %NULL if none. | |
788 | */ | |
d84ff051 | 789 | struct task_struct *wq_worker_sleeping(struct task_struct *task, int cpu) |
e22bee78 TH |
790 | { |
791 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
111c225a | 792 | struct worker_pool *pool; |
e22bee78 | 793 | |
111c225a TH |
794 | /* |
795 | * Rescuers, which may not have all the fields set up like normal | |
796 | * workers, also reach here, let's not access anything before | |
797 | * checking NOT_RUNNING. | |
798 | */ | |
2d64672e | 799 | if (worker->flags & WORKER_NOT_RUNNING) |
e22bee78 TH |
800 | return NULL; |
801 | ||
111c225a | 802 | pool = worker->pool; |
111c225a | 803 | |
e22bee78 | 804 | /* this can only happen on the local cpu */ |
6183c009 TH |
805 | if (WARN_ON_ONCE(cpu != raw_smp_processor_id())) |
806 | return NULL; | |
e22bee78 TH |
807 | |
808 | /* | |
809 | * The counterpart of the following dec_and_test, implied mb, | |
810 | * worklist not empty test sequence is in insert_work(). | |
811 | * Please read comment there. | |
812 | * | |
628c78e7 TH |
813 | * NOT_RUNNING is clear. This means that we're bound to and |
814 | * running on the local cpu w/ rq lock held and preemption | |
815 | * disabled, which in turn means that none else could be | |
d565ed63 | 816 | * manipulating idle_list, so dereferencing idle_list without pool |
628c78e7 | 817 | * lock is safe. |
e22bee78 | 818 | */ |
e19e397a TH |
819 | if (atomic_dec_and_test(&pool->nr_running) && |
820 | !list_empty(&pool->worklist)) | |
63d95a91 | 821 | to_wakeup = first_worker(pool); |
e22bee78 TH |
822 | return to_wakeup ? to_wakeup->task : NULL; |
823 | } | |
824 | ||
825 | /** | |
826 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 827 | * @worker: self |
d302f017 TH |
828 | * @flags: flags to set |
829 | * @wakeup: wakeup an idle worker if necessary | |
830 | * | |
e22bee78 TH |
831 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
832 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
833 | * woken up. | |
d302f017 | 834 | * |
cb444766 | 835 | * CONTEXT: |
d565ed63 | 836 | * spin_lock_irq(pool->lock) |
d302f017 TH |
837 | */ |
838 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
839 | bool wakeup) | |
840 | { | |
bd7bdd43 | 841 | struct worker_pool *pool = worker->pool; |
e22bee78 | 842 | |
cb444766 TH |
843 | WARN_ON_ONCE(worker->task != current); |
844 | ||
e22bee78 TH |
845 | /* |
846 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
847 | * wake up an idle worker as necessary if requested by | |
848 | * @wakeup. | |
849 | */ | |
850 | if ((flags & WORKER_NOT_RUNNING) && | |
851 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
e22bee78 | 852 | if (wakeup) { |
e19e397a | 853 | if (atomic_dec_and_test(&pool->nr_running) && |
bd7bdd43 | 854 | !list_empty(&pool->worklist)) |
63d95a91 | 855 | wake_up_worker(pool); |
e22bee78 | 856 | } else |
e19e397a | 857 | atomic_dec(&pool->nr_running); |
e22bee78 TH |
858 | } |
859 | ||
d302f017 TH |
860 | worker->flags |= flags; |
861 | } | |
862 | ||
863 | /** | |
e22bee78 | 864 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 865 | * @worker: self |
d302f017 TH |
866 | * @flags: flags to clear |
867 | * | |
e22bee78 | 868 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 869 | * |
cb444766 | 870 | * CONTEXT: |
d565ed63 | 871 | * spin_lock_irq(pool->lock) |
d302f017 TH |
872 | */ |
873 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
874 | { | |
63d95a91 | 875 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
876 | unsigned int oflags = worker->flags; |
877 | ||
cb444766 TH |
878 | WARN_ON_ONCE(worker->task != current); |
879 | ||
d302f017 | 880 | worker->flags &= ~flags; |
e22bee78 | 881 | |
42c025f3 TH |
882 | /* |
883 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
884 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
885 | * of multiple flags, not a single flag. | |
886 | */ | |
e22bee78 TH |
887 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
888 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
e19e397a | 889 | atomic_inc(&pool->nr_running); |
d302f017 TH |
890 | } |
891 | ||
8cca0eea TH |
892 | /** |
893 | * find_worker_executing_work - find worker which is executing a work | |
c9e7cf27 | 894 | * @pool: pool of interest |
8cca0eea TH |
895 | * @work: work to find worker for |
896 | * | |
c9e7cf27 TH |
897 | * Find a worker which is executing @work on @pool by searching |
898 | * @pool->busy_hash which is keyed by the address of @work. For a worker | |
a2c1c57b TH |
899 | * to match, its current execution should match the address of @work and |
900 | * its work function. This is to avoid unwanted dependency between | |
901 | * unrelated work executions through a work item being recycled while still | |
902 | * being executed. | |
903 | * | |
904 | * This is a bit tricky. A work item may be freed once its execution | |
905 | * starts and nothing prevents the freed area from being recycled for | |
906 | * another work item. If the same work item address ends up being reused | |
907 | * before the original execution finishes, workqueue will identify the | |
908 | * recycled work item as currently executing and make it wait until the | |
909 | * current execution finishes, introducing an unwanted dependency. | |
910 | * | |
911 | * This function checks the work item address, work function and workqueue | |
912 | * to avoid false positives. Note that this isn't complete as one may | |
913 | * construct a work function which can introduce dependency onto itself | |
914 | * through a recycled work item. Well, if somebody wants to shoot oneself | |
915 | * in the foot that badly, there's only so much we can do, and if such | |
916 | * deadlock actually occurs, it should be easy to locate the culprit work | |
917 | * function. | |
8cca0eea TH |
918 | * |
919 | * CONTEXT: | |
d565ed63 | 920 | * spin_lock_irq(pool->lock). |
8cca0eea TH |
921 | * |
922 | * RETURNS: | |
923 | * Pointer to worker which is executing @work if found, NULL | |
924 | * otherwise. | |
4d707b9f | 925 | */ |
c9e7cf27 | 926 | static struct worker *find_worker_executing_work(struct worker_pool *pool, |
8cca0eea | 927 | struct work_struct *work) |
4d707b9f | 928 | { |
42f8570f | 929 | struct worker *worker; |
42f8570f | 930 | |
b67bfe0d | 931 | hash_for_each_possible(pool->busy_hash, worker, hentry, |
a2c1c57b TH |
932 | (unsigned long)work) |
933 | if (worker->current_work == work && | |
934 | worker->current_func == work->func) | |
42f8570f SL |
935 | return worker; |
936 | ||
937 | return NULL; | |
4d707b9f ON |
938 | } |
939 | ||
bf4ede01 TH |
940 | /** |
941 | * move_linked_works - move linked works to a list | |
942 | * @work: start of series of works to be scheduled | |
943 | * @head: target list to append @work to | |
944 | * @nextp: out paramter for nested worklist walking | |
945 | * | |
946 | * Schedule linked works starting from @work to @head. Work series to | |
947 | * be scheduled starts at @work and includes any consecutive work with | |
948 | * WORK_STRUCT_LINKED set in its predecessor. | |
949 | * | |
950 | * If @nextp is not NULL, it's updated to point to the next work of | |
951 | * the last scheduled work. This allows move_linked_works() to be | |
952 | * nested inside outer list_for_each_entry_safe(). | |
953 | * | |
954 | * CONTEXT: | |
d565ed63 | 955 | * spin_lock_irq(pool->lock). |
bf4ede01 TH |
956 | */ |
957 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
958 | struct work_struct **nextp) | |
959 | { | |
960 | struct work_struct *n; | |
961 | ||
962 | /* | |
963 | * Linked worklist will always end before the end of the list, | |
964 | * use NULL for list head. | |
965 | */ | |
966 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
967 | list_move_tail(&work->entry, head); | |
968 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
969 | break; | |
970 | } | |
971 | ||
972 | /* | |
973 | * If we're already inside safe list traversal and have moved | |
974 | * multiple works to the scheduled queue, the next position | |
975 | * needs to be updated. | |
976 | */ | |
977 | if (nextp) | |
978 | *nextp = n; | |
979 | } | |
980 | ||
112202d9 | 981 | static void pwq_activate_delayed_work(struct work_struct *work) |
bf4ede01 | 982 | { |
112202d9 | 983 | struct pool_workqueue *pwq = get_work_pwq(work); |
bf4ede01 TH |
984 | |
985 | trace_workqueue_activate_work(work); | |
112202d9 | 986 | move_linked_works(work, &pwq->pool->worklist, NULL); |
bf4ede01 | 987 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); |
112202d9 | 988 | pwq->nr_active++; |
bf4ede01 TH |
989 | } |
990 | ||
112202d9 | 991 | static void pwq_activate_first_delayed(struct pool_workqueue *pwq) |
3aa62497 | 992 | { |
112202d9 | 993 | struct work_struct *work = list_first_entry(&pwq->delayed_works, |
3aa62497 LJ |
994 | struct work_struct, entry); |
995 | ||
112202d9 | 996 | pwq_activate_delayed_work(work); |
3aa62497 LJ |
997 | } |
998 | ||
bf4ede01 | 999 | /** |
112202d9 TH |
1000 | * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight |
1001 | * @pwq: pwq of interest | |
bf4ede01 | 1002 | * @color: color of work which left the queue |
bf4ede01 TH |
1003 | * |
1004 | * A work either has completed or is removed from pending queue, | |
112202d9 | 1005 | * decrement nr_in_flight of its pwq and handle workqueue flushing. |
bf4ede01 TH |
1006 | * |
1007 | * CONTEXT: | |
d565ed63 | 1008 | * spin_lock_irq(pool->lock). |
bf4ede01 | 1009 | */ |
112202d9 | 1010 | static void pwq_dec_nr_in_flight(struct pool_workqueue *pwq, int color) |
bf4ede01 TH |
1011 | { |
1012 | /* ignore uncolored works */ | |
1013 | if (color == WORK_NO_COLOR) | |
1014 | return; | |
1015 | ||
112202d9 | 1016 | pwq->nr_in_flight[color]--; |
bf4ede01 | 1017 | |
112202d9 TH |
1018 | pwq->nr_active--; |
1019 | if (!list_empty(&pwq->delayed_works)) { | |
b3f9f405 | 1020 | /* one down, submit a delayed one */ |
112202d9 TH |
1021 | if (pwq->nr_active < pwq->max_active) |
1022 | pwq_activate_first_delayed(pwq); | |
bf4ede01 TH |
1023 | } |
1024 | ||
1025 | /* is flush in progress and are we at the flushing tip? */ | |
112202d9 | 1026 | if (likely(pwq->flush_color != color)) |
bf4ede01 TH |
1027 | return; |
1028 | ||
1029 | /* are there still in-flight works? */ | |
112202d9 | 1030 | if (pwq->nr_in_flight[color]) |
bf4ede01 TH |
1031 | return; |
1032 | ||
112202d9 TH |
1033 | /* this pwq is done, clear flush_color */ |
1034 | pwq->flush_color = -1; | |
bf4ede01 TH |
1035 | |
1036 | /* | |
112202d9 | 1037 | * If this was the last pwq, wake up the first flusher. It |
bf4ede01 TH |
1038 | * will handle the rest. |
1039 | */ | |
112202d9 TH |
1040 | if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush)) |
1041 | complete(&pwq->wq->first_flusher->done); | |
bf4ede01 TH |
1042 | } |
1043 | ||
36e227d2 | 1044 | /** |
bbb68dfa | 1045 | * try_to_grab_pending - steal work item from worklist and disable irq |
36e227d2 TH |
1046 | * @work: work item to steal |
1047 | * @is_dwork: @work is a delayed_work | |
bbb68dfa | 1048 | * @flags: place to store irq state |
36e227d2 TH |
1049 | * |
1050 | * Try to grab PENDING bit of @work. This function can handle @work in any | |
1051 | * stable state - idle, on timer or on worklist. Return values are | |
1052 | * | |
1053 | * 1 if @work was pending and we successfully stole PENDING | |
1054 | * 0 if @work was idle and we claimed PENDING | |
1055 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry | |
bbb68dfa TH |
1056 | * -ENOENT if someone else is canceling @work, this state may persist |
1057 | * for arbitrarily long | |
36e227d2 | 1058 | * |
bbb68dfa | 1059 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
e0aecdd8 TH |
1060 | * interrupted while holding PENDING and @work off queue, irq must be |
1061 | * disabled on entry. This, combined with delayed_work->timer being | |
1062 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. | |
bbb68dfa TH |
1063 | * |
1064 | * On successful return, >= 0, irq is disabled and the caller is | |
1065 | * responsible for releasing it using local_irq_restore(*@flags). | |
1066 | * | |
e0aecdd8 | 1067 | * This function is safe to call from any context including IRQ handler. |
bf4ede01 | 1068 | */ |
bbb68dfa TH |
1069 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
1070 | unsigned long *flags) | |
bf4ede01 | 1071 | { |
d565ed63 | 1072 | struct worker_pool *pool; |
112202d9 | 1073 | struct pool_workqueue *pwq; |
bf4ede01 | 1074 | |
bbb68dfa TH |
1075 | local_irq_save(*flags); |
1076 | ||
36e227d2 TH |
1077 | /* try to steal the timer if it exists */ |
1078 | if (is_dwork) { | |
1079 | struct delayed_work *dwork = to_delayed_work(work); | |
1080 | ||
e0aecdd8 TH |
1081 | /* |
1082 | * dwork->timer is irqsafe. If del_timer() fails, it's | |
1083 | * guaranteed that the timer is not queued anywhere and not | |
1084 | * running on the local CPU. | |
1085 | */ | |
36e227d2 TH |
1086 | if (likely(del_timer(&dwork->timer))) |
1087 | return 1; | |
1088 | } | |
1089 | ||
1090 | /* try to claim PENDING the normal way */ | |
bf4ede01 TH |
1091 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1092 | return 0; | |
1093 | ||
1094 | /* | |
1095 | * The queueing is in progress, or it is already queued. Try to | |
1096 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1097 | */ | |
d565ed63 TH |
1098 | pool = get_work_pool(work); |
1099 | if (!pool) | |
bbb68dfa | 1100 | goto fail; |
bf4ede01 | 1101 | |
d565ed63 | 1102 | spin_lock(&pool->lock); |
0b3dae68 | 1103 | /* |
112202d9 TH |
1104 | * work->data is guaranteed to point to pwq only while the work |
1105 | * item is queued on pwq->wq, and both updating work->data to point | |
1106 | * to pwq on queueing and to pool on dequeueing are done under | |
1107 | * pwq->pool->lock. This in turn guarantees that, if work->data | |
1108 | * points to pwq which is associated with a locked pool, the work | |
0b3dae68 LJ |
1109 | * item is currently queued on that pool. |
1110 | */ | |
112202d9 TH |
1111 | pwq = get_work_pwq(work); |
1112 | if (pwq && pwq->pool == pool) { | |
16062836 TH |
1113 | debug_work_deactivate(work); |
1114 | ||
1115 | /* | |
1116 | * A delayed work item cannot be grabbed directly because | |
1117 | * it might have linked NO_COLOR work items which, if left | |
112202d9 | 1118 | * on the delayed_list, will confuse pwq->nr_active |
16062836 TH |
1119 | * management later on and cause stall. Make sure the work |
1120 | * item is activated before grabbing. | |
1121 | */ | |
1122 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) | |
112202d9 | 1123 | pwq_activate_delayed_work(work); |
16062836 TH |
1124 | |
1125 | list_del_init(&work->entry); | |
112202d9 | 1126 | pwq_dec_nr_in_flight(get_work_pwq(work), get_work_color(work)); |
16062836 | 1127 | |
112202d9 | 1128 | /* work->data points to pwq iff queued, point to pool */ |
16062836 TH |
1129 | set_work_pool_and_keep_pending(work, pool->id); |
1130 | ||
1131 | spin_unlock(&pool->lock); | |
1132 | return 1; | |
bf4ede01 | 1133 | } |
d565ed63 | 1134 | spin_unlock(&pool->lock); |
bbb68dfa TH |
1135 | fail: |
1136 | local_irq_restore(*flags); | |
1137 | if (work_is_canceling(work)) | |
1138 | return -ENOENT; | |
1139 | cpu_relax(); | |
36e227d2 | 1140 | return -EAGAIN; |
bf4ede01 TH |
1141 | } |
1142 | ||
4690c4ab | 1143 | /** |
706026c2 | 1144 | * insert_work - insert a work into a pool |
112202d9 | 1145 | * @pwq: pwq @work belongs to |
4690c4ab TH |
1146 | * @work: work to insert |
1147 | * @head: insertion point | |
1148 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
1149 | * | |
112202d9 | 1150 | * Insert @work which belongs to @pwq after @head. @extra_flags is or'd to |
706026c2 | 1151 | * work_struct flags. |
4690c4ab TH |
1152 | * |
1153 | * CONTEXT: | |
d565ed63 | 1154 | * spin_lock_irq(pool->lock). |
4690c4ab | 1155 | */ |
112202d9 TH |
1156 | static void insert_work(struct pool_workqueue *pwq, struct work_struct *work, |
1157 | struct list_head *head, unsigned int extra_flags) | |
b89deed3 | 1158 | { |
112202d9 | 1159 | struct worker_pool *pool = pwq->pool; |
e22bee78 | 1160 | |
4690c4ab | 1161 | /* we own @work, set data and link */ |
112202d9 | 1162 | set_work_pwq(work, pwq, extra_flags); |
1a4d9b0a | 1163 | list_add_tail(&work->entry, head); |
e22bee78 TH |
1164 | |
1165 | /* | |
1166 | * Ensure either worker_sched_deactivated() sees the above | |
1167 | * list_add_tail() or we see zero nr_running to avoid workers | |
1168 | * lying around lazily while there are works to be processed. | |
1169 | */ | |
1170 | smp_mb(); | |
1171 | ||
63d95a91 TH |
1172 | if (__need_more_worker(pool)) |
1173 | wake_up_worker(pool); | |
b89deed3 ON |
1174 | } |
1175 | ||
c8efcc25 TH |
1176 | /* |
1177 | * Test whether @work is being queued from another work executing on the | |
8d03ecfe | 1178 | * same workqueue. |
c8efcc25 TH |
1179 | */ |
1180 | static bool is_chained_work(struct workqueue_struct *wq) | |
1181 | { | |
8d03ecfe TH |
1182 | struct worker *worker; |
1183 | ||
1184 | worker = current_wq_worker(); | |
1185 | /* | |
1186 | * Return %true iff I'm a worker execuing a work item on @wq. If | |
1187 | * I'm @worker, it's safe to dereference it without locking. | |
1188 | */ | |
112202d9 | 1189 | return worker && worker->current_pwq->wq == wq; |
c8efcc25 TH |
1190 | } |
1191 | ||
d84ff051 | 1192 | static void __queue_work(int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
1193 | struct work_struct *work) |
1194 | { | |
112202d9 | 1195 | struct pool_workqueue *pwq; |
1e19ffc6 | 1196 | struct list_head *worklist; |
8a2e8e5d | 1197 | unsigned int work_flags; |
b75cac93 | 1198 | unsigned int req_cpu = cpu; |
8930caba TH |
1199 | |
1200 | /* | |
1201 | * While a work item is PENDING && off queue, a task trying to | |
1202 | * steal the PENDING will busy-loop waiting for it to either get | |
1203 | * queued or lose PENDING. Grabbing PENDING and queueing should | |
1204 | * happen with IRQ disabled. | |
1205 | */ | |
1206 | WARN_ON_ONCE(!irqs_disabled()); | |
1da177e4 | 1207 | |
dc186ad7 | 1208 | debug_work_activate(work); |
1e19ffc6 | 1209 | |
c8efcc25 | 1210 | /* if dying, only works from the same workqueue are allowed */ |
9c5a2ba7 | 1211 | if (unlikely(wq->flags & WQ_DRAINING) && |
c8efcc25 | 1212 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b TH |
1213 | return; |
1214 | ||
112202d9 | 1215 | /* determine the pwq to use */ |
c7fc77f7 | 1216 | if (!(wq->flags & WQ_UNBOUND)) { |
c9e7cf27 | 1217 | struct worker_pool *last_pool; |
18aa9eff | 1218 | |
57469821 | 1219 | if (cpu == WORK_CPU_UNBOUND) |
c7fc77f7 TH |
1220 | cpu = raw_smp_processor_id(); |
1221 | ||
18aa9eff | 1222 | /* |
dbf2576e TH |
1223 | * It's multi cpu. If @work was previously on a different |
1224 | * cpu, it might still be running there, in which case the | |
1225 | * work needs to be queued on that cpu to guarantee | |
1226 | * non-reentrancy. | |
18aa9eff | 1227 | */ |
7fb98ea7 | 1228 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); |
c9e7cf27 | 1229 | last_pool = get_work_pool(work); |
dbf2576e | 1230 | |
112202d9 | 1231 | if (last_pool && last_pool != pwq->pool) { |
18aa9eff TH |
1232 | struct worker *worker; |
1233 | ||
d565ed63 | 1234 | spin_lock(&last_pool->lock); |
18aa9eff | 1235 | |
c9e7cf27 | 1236 | worker = find_worker_executing_work(last_pool, work); |
18aa9eff | 1237 | |
112202d9 | 1238 | if (worker && worker->current_pwq->wq == wq) { |
7fb98ea7 | 1239 | pwq = per_cpu_ptr(wq->cpu_pwqs, last_pool->cpu); |
8594fade | 1240 | } else { |
18aa9eff | 1241 | /* meh... not running there, queue here */ |
d565ed63 | 1242 | spin_unlock(&last_pool->lock); |
112202d9 | 1243 | spin_lock(&pwq->pool->lock); |
18aa9eff | 1244 | } |
8930caba | 1245 | } else { |
112202d9 | 1246 | spin_lock(&pwq->pool->lock); |
8930caba | 1247 | } |
f3421797 | 1248 | } else { |
7fb98ea7 | 1249 | pwq = first_pwq(wq); |
112202d9 | 1250 | spin_lock(&pwq->pool->lock); |
502ca9d8 TH |
1251 | } |
1252 | ||
112202d9 TH |
1253 | /* pwq determined, queue */ |
1254 | trace_workqueue_queue_work(req_cpu, pwq, work); | |
502ca9d8 | 1255 | |
f5b2552b | 1256 | if (WARN_ON(!list_empty(&work->entry))) { |
112202d9 | 1257 | spin_unlock(&pwq->pool->lock); |
f5b2552b DC |
1258 | return; |
1259 | } | |
1e19ffc6 | 1260 | |
112202d9 TH |
1261 | pwq->nr_in_flight[pwq->work_color]++; |
1262 | work_flags = work_color_to_flags(pwq->work_color); | |
1e19ffc6 | 1263 | |
112202d9 | 1264 | if (likely(pwq->nr_active < pwq->max_active)) { |
cdadf009 | 1265 | trace_workqueue_activate_work(work); |
112202d9 TH |
1266 | pwq->nr_active++; |
1267 | worklist = &pwq->pool->worklist; | |
8a2e8e5d TH |
1268 | } else { |
1269 | work_flags |= WORK_STRUCT_DELAYED; | |
112202d9 | 1270 | worklist = &pwq->delayed_works; |
8a2e8e5d | 1271 | } |
1e19ffc6 | 1272 | |
112202d9 | 1273 | insert_work(pwq, work, worklist, work_flags); |
1e19ffc6 | 1274 | |
112202d9 | 1275 | spin_unlock(&pwq->pool->lock); |
1da177e4 LT |
1276 | } |
1277 | ||
0fcb78c2 | 1278 | /** |
c1a220e7 ZR |
1279 | * queue_work_on - queue work on specific cpu |
1280 | * @cpu: CPU number to execute work on | |
0fcb78c2 REB |
1281 | * @wq: workqueue to use |
1282 | * @work: work to queue | |
1283 | * | |
d4283e93 | 1284 | * Returns %false if @work was already on a queue, %true otherwise. |
1da177e4 | 1285 | * |
c1a220e7 ZR |
1286 | * We queue the work to a specific CPU, the caller must ensure it |
1287 | * can't go away. | |
1da177e4 | 1288 | */ |
d4283e93 TH |
1289 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
1290 | struct work_struct *work) | |
1da177e4 | 1291 | { |
d4283e93 | 1292 | bool ret = false; |
8930caba | 1293 | unsigned long flags; |
ef1ca236 | 1294 | |
8930caba | 1295 | local_irq_save(flags); |
c1a220e7 | 1296 | |
22df02bb | 1297 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1298 | __queue_work(cpu, wq, work); |
d4283e93 | 1299 | ret = true; |
c1a220e7 | 1300 | } |
ef1ca236 | 1301 | |
8930caba | 1302 | local_irq_restore(flags); |
1da177e4 LT |
1303 | return ret; |
1304 | } | |
c1a220e7 | 1305 | EXPORT_SYMBOL_GPL(queue_work_on); |
1da177e4 | 1306 | |
c1a220e7 | 1307 | /** |
0a13c00e | 1308 | * queue_work - queue work on a workqueue |
c1a220e7 ZR |
1309 | * @wq: workqueue to use |
1310 | * @work: work to queue | |
1311 | * | |
d4283e93 | 1312 | * Returns %false if @work was already on a queue, %true otherwise. |
c1a220e7 | 1313 | * |
0a13c00e TH |
1314 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
1315 | * it can be processed by another CPU. | |
c1a220e7 | 1316 | */ |
d4283e93 | 1317 | bool queue_work(struct workqueue_struct *wq, struct work_struct *work) |
c1a220e7 | 1318 | { |
57469821 | 1319 | return queue_work_on(WORK_CPU_UNBOUND, wq, work); |
c1a220e7 | 1320 | } |
0a13c00e | 1321 | EXPORT_SYMBOL_GPL(queue_work); |
c1a220e7 | 1322 | |
d8e794df | 1323 | void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1324 | { |
52bad64d | 1325 | struct delayed_work *dwork = (struct delayed_work *)__data; |
1da177e4 | 1326 | |
e0aecdd8 | 1327 | /* should have been called from irqsafe timer with irq already off */ |
60c057bc | 1328 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
1da177e4 | 1329 | } |
1438ade5 | 1330 | EXPORT_SYMBOL(delayed_work_timer_fn); |
1da177e4 | 1331 | |
7beb2edf TH |
1332 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
1333 | struct delayed_work *dwork, unsigned long delay) | |
1da177e4 | 1334 | { |
7beb2edf TH |
1335 | struct timer_list *timer = &dwork->timer; |
1336 | struct work_struct *work = &dwork->work; | |
7beb2edf TH |
1337 | |
1338 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || | |
1339 | timer->data != (unsigned long)dwork); | |
fc4b514f TH |
1340 | WARN_ON_ONCE(timer_pending(timer)); |
1341 | WARN_ON_ONCE(!list_empty(&work->entry)); | |
7beb2edf | 1342 | |
8852aac2 TH |
1343 | /* |
1344 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1345 | * both optimization and correctness. The earliest @timer can | |
1346 | * expire is on the closest next tick and delayed_work users depend | |
1347 | * on that there's no such delay when @delay is 0. | |
1348 | */ | |
1349 | if (!delay) { | |
1350 | __queue_work(cpu, wq, &dwork->work); | |
1351 | return; | |
1352 | } | |
1353 | ||
7beb2edf | 1354 | timer_stats_timer_set_start_info(&dwork->timer); |
1da177e4 | 1355 | |
60c057bc | 1356 | dwork->wq = wq; |
1265057f | 1357 | dwork->cpu = cpu; |
7beb2edf TH |
1358 | timer->expires = jiffies + delay; |
1359 | ||
1360 | if (unlikely(cpu != WORK_CPU_UNBOUND)) | |
1361 | add_timer_on(timer, cpu); | |
1362 | else | |
1363 | add_timer(timer); | |
1da177e4 LT |
1364 | } |
1365 | ||
0fcb78c2 REB |
1366 | /** |
1367 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1368 | * @cpu: CPU number to execute work on | |
1369 | * @wq: workqueue to use | |
af9997e4 | 1370 | * @dwork: work to queue |
0fcb78c2 REB |
1371 | * @delay: number of jiffies to wait before queueing |
1372 | * | |
715f1300 TH |
1373 | * Returns %false if @work was already on a queue, %true otherwise. If |
1374 | * @delay is zero and @dwork is idle, it will be scheduled for immediate | |
1375 | * execution. | |
0fcb78c2 | 1376 | */ |
d4283e93 TH |
1377 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
1378 | struct delayed_work *dwork, unsigned long delay) | |
7a6bc1cd | 1379 | { |
52bad64d | 1380 | struct work_struct *work = &dwork->work; |
d4283e93 | 1381 | bool ret = false; |
8930caba | 1382 | unsigned long flags; |
7a6bc1cd | 1383 | |
8930caba TH |
1384 | /* read the comment in __queue_work() */ |
1385 | local_irq_save(flags); | |
7a6bc1cd | 1386 | |
22df02bb | 1387 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7beb2edf | 1388 | __queue_delayed_work(cpu, wq, dwork, delay); |
d4283e93 | 1389 | ret = true; |
7a6bc1cd | 1390 | } |
8a3e77cc | 1391 | |
8930caba | 1392 | local_irq_restore(flags); |
7a6bc1cd VP |
1393 | return ret; |
1394 | } | |
ae90dd5d | 1395 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
c7fc77f7 | 1396 | |
0a13c00e TH |
1397 | /** |
1398 | * queue_delayed_work - queue work on a workqueue after delay | |
1399 | * @wq: workqueue to use | |
1400 | * @dwork: delayable work to queue | |
1401 | * @delay: number of jiffies to wait before queueing | |
1402 | * | |
715f1300 | 1403 | * Equivalent to queue_delayed_work_on() but tries to use the local CPU. |
0a13c00e | 1404 | */ |
d4283e93 | 1405 | bool queue_delayed_work(struct workqueue_struct *wq, |
0a13c00e TH |
1406 | struct delayed_work *dwork, unsigned long delay) |
1407 | { | |
57469821 | 1408 | return queue_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); |
0a13c00e TH |
1409 | } |
1410 | EXPORT_SYMBOL_GPL(queue_delayed_work); | |
c7fc77f7 | 1411 | |
8376fe22 TH |
1412 | /** |
1413 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU | |
1414 | * @cpu: CPU number to execute work on | |
1415 | * @wq: workqueue to use | |
1416 | * @dwork: work to queue | |
1417 | * @delay: number of jiffies to wait before queueing | |
1418 | * | |
1419 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, | |
1420 | * modify @dwork's timer so that it expires after @delay. If @delay is | |
1421 | * zero, @work is guaranteed to be scheduled immediately regardless of its | |
1422 | * current state. | |
1423 | * | |
1424 | * Returns %false if @dwork was idle and queued, %true if @dwork was | |
1425 | * pending and its timer was modified. | |
1426 | * | |
e0aecdd8 | 1427 | * This function is safe to call from any context including IRQ handler. |
8376fe22 TH |
1428 | * See try_to_grab_pending() for details. |
1429 | */ | |
1430 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
1431 | struct delayed_work *dwork, unsigned long delay) | |
1432 | { | |
1433 | unsigned long flags; | |
1434 | int ret; | |
c7fc77f7 | 1435 | |
8376fe22 TH |
1436 | do { |
1437 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
1438 | } while (unlikely(ret == -EAGAIN)); | |
63bc0362 | 1439 | |
8376fe22 TH |
1440 | if (likely(ret >= 0)) { |
1441 | __queue_delayed_work(cpu, wq, dwork, delay); | |
1442 | local_irq_restore(flags); | |
7a6bc1cd | 1443 | } |
8376fe22 TH |
1444 | |
1445 | /* -ENOENT from try_to_grab_pending() becomes %true */ | |
7a6bc1cd VP |
1446 | return ret; |
1447 | } | |
8376fe22 TH |
1448 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
1449 | ||
1450 | /** | |
1451 | * mod_delayed_work - modify delay of or queue a delayed work | |
1452 | * @wq: workqueue to use | |
1453 | * @dwork: work to queue | |
1454 | * @delay: number of jiffies to wait before queueing | |
1455 | * | |
1456 | * mod_delayed_work_on() on local CPU. | |
1457 | */ | |
1458 | bool mod_delayed_work(struct workqueue_struct *wq, struct delayed_work *dwork, | |
1459 | unsigned long delay) | |
1460 | { | |
1461 | return mod_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); | |
1462 | } | |
1463 | EXPORT_SYMBOL_GPL(mod_delayed_work); | |
1da177e4 | 1464 | |
c8e55f36 TH |
1465 | /** |
1466 | * worker_enter_idle - enter idle state | |
1467 | * @worker: worker which is entering idle state | |
1468 | * | |
1469 | * @worker is entering idle state. Update stats and idle timer if | |
1470 | * necessary. | |
1471 | * | |
1472 | * LOCKING: | |
d565ed63 | 1473 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1474 | */ |
1475 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1476 | { |
bd7bdd43 | 1477 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1478 | |
6183c009 TH |
1479 | if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) || |
1480 | WARN_ON_ONCE(!list_empty(&worker->entry) && | |
1481 | (worker->hentry.next || worker->hentry.pprev))) | |
1482 | return; | |
c8e55f36 | 1483 | |
cb444766 TH |
1484 | /* can't use worker_set_flags(), also called from start_worker() */ |
1485 | worker->flags |= WORKER_IDLE; | |
bd7bdd43 | 1486 | pool->nr_idle++; |
e22bee78 | 1487 | worker->last_active = jiffies; |
c8e55f36 TH |
1488 | |
1489 | /* idle_list is LIFO */ | |
bd7bdd43 | 1490 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1491 | |
628c78e7 TH |
1492 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
1493 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); | |
cb444766 | 1494 | |
544ecf31 | 1495 | /* |
706026c2 | 1496 | * Sanity check nr_running. Because wq_unbind_fn() releases |
d565ed63 | 1497 | * pool->lock between setting %WORKER_UNBOUND and zapping |
628c78e7 TH |
1498 | * nr_running, the warning may trigger spuriously. Check iff |
1499 | * unbind is not in progress. | |
544ecf31 | 1500 | */ |
24647570 | 1501 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
bd7bdd43 | 1502 | pool->nr_workers == pool->nr_idle && |
e19e397a | 1503 | atomic_read(&pool->nr_running)); |
c8e55f36 TH |
1504 | } |
1505 | ||
1506 | /** | |
1507 | * worker_leave_idle - leave idle state | |
1508 | * @worker: worker which is leaving idle state | |
1509 | * | |
1510 | * @worker is leaving idle state. Update stats. | |
1511 | * | |
1512 | * LOCKING: | |
d565ed63 | 1513 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1514 | */ |
1515 | static void worker_leave_idle(struct worker *worker) | |
1516 | { | |
bd7bdd43 | 1517 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1518 | |
6183c009 TH |
1519 | if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE))) |
1520 | return; | |
d302f017 | 1521 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1522 | pool->nr_idle--; |
c8e55f36 TH |
1523 | list_del_init(&worker->entry); |
1524 | } | |
1525 | ||
e22bee78 | 1526 | /** |
f36dc67b LJ |
1527 | * worker_maybe_bind_and_lock - try to bind %current to worker_pool and lock it |
1528 | * @pool: target worker_pool | |
1529 | * | |
1530 | * Bind %current to the cpu of @pool if it is associated and lock @pool. | |
e22bee78 TH |
1531 | * |
1532 | * Works which are scheduled while the cpu is online must at least be | |
1533 | * scheduled to a worker which is bound to the cpu so that if they are | |
1534 | * flushed from cpu callbacks while cpu is going down, they are | |
1535 | * guaranteed to execute on the cpu. | |
1536 | * | |
f5faa077 | 1537 | * This function is to be used by unbound workers and rescuers to bind |
e22bee78 TH |
1538 | * themselves to the target cpu and may race with cpu going down or |
1539 | * coming online. kthread_bind() can't be used because it may put the | |
1540 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
706026c2 | 1541 | * verbatim as it's best effort and blocking and pool may be |
e22bee78 TH |
1542 | * [dis]associated in the meantime. |
1543 | * | |
706026c2 | 1544 | * This function tries set_cpus_allowed() and locks pool and verifies the |
24647570 | 1545 | * binding against %POOL_DISASSOCIATED which is set during |
f2d5a0ee TH |
1546 | * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker |
1547 | * enters idle state or fetches works without dropping lock, it can | |
1548 | * guarantee the scheduling requirement described in the first paragraph. | |
e22bee78 TH |
1549 | * |
1550 | * CONTEXT: | |
d565ed63 | 1551 | * Might sleep. Called without any lock but returns with pool->lock |
e22bee78 TH |
1552 | * held. |
1553 | * | |
1554 | * RETURNS: | |
706026c2 | 1555 | * %true if the associated pool is online (@worker is successfully |
e22bee78 TH |
1556 | * bound), %false if offline. |
1557 | */ | |
f36dc67b | 1558 | static bool worker_maybe_bind_and_lock(struct worker_pool *pool) |
d565ed63 | 1559 | __acquires(&pool->lock) |
e22bee78 | 1560 | { |
e22bee78 | 1561 | while (true) { |
4e6045f1 | 1562 | /* |
e22bee78 TH |
1563 | * The following call may fail, succeed or succeed |
1564 | * without actually migrating the task to the cpu if | |
1565 | * it races with cpu hotunplug operation. Verify | |
24647570 | 1566 | * against POOL_DISASSOCIATED. |
4e6045f1 | 1567 | */ |
24647570 | 1568 | if (!(pool->flags & POOL_DISASSOCIATED)) |
f5faa077 | 1569 | set_cpus_allowed_ptr(current, get_cpu_mask(pool->cpu)); |
e22bee78 | 1570 | |
d565ed63 | 1571 | spin_lock_irq(&pool->lock); |
24647570 | 1572 | if (pool->flags & POOL_DISASSOCIATED) |
e22bee78 | 1573 | return false; |
f5faa077 | 1574 | if (task_cpu(current) == pool->cpu && |
e22bee78 | 1575 | cpumask_equal(¤t->cpus_allowed, |
ec22ca5e | 1576 | get_cpu_mask(pool->cpu))) |
e22bee78 | 1577 | return true; |
d565ed63 | 1578 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1579 | |
5035b20f TH |
1580 | /* |
1581 | * We've raced with CPU hot[un]plug. Give it a breather | |
1582 | * and retry migration. cond_resched() is required here; | |
1583 | * otherwise, we might deadlock against cpu_stop trying to | |
1584 | * bring down the CPU on non-preemptive kernel. | |
1585 | */ | |
e22bee78 | 1586 | cpu_relax(); |
5035b20f | 1587 | cond_resched(); |
e22bee78 TH |
1588 | } |
1589 | } | |
1590 | ||
25511a47 | 1591 | /* |
ea1abd61 | 1592 | * Rebind an idle @worker to its CPU. worker_thread() will test |
5f7dabfd | 1593 | * list_empty(@worker->entry) before leaving idle and call this function. |
25511a47 TH |
1594 | */ |
1595 | static void idle_worker_rebind(struct worker *worker) | |
1596 | { | |
5f7dabfd | 1597 | /* CPU may go down again inbetween, clear UNBOUND only on success */ |
f36dc67b | 1598 | if (worker_maybe_bind_and_lock(worker->pool)) |
5f7dabfd | 1599 | worker_clr_flags(worker, WORKER_UNBOUND); |
25511a47 | 1600 | |
ea1abd61 LJ |
1601 | /* rebind complete, become available again */ |
1602 | list_add(&worker->entry, &worker->pool->idle_list); | |
d565ed63 | 1603 | spin_unlock_irq(&worker->pool->lock); |
25511a47 TH |
1604 | } |
1605 | ||
e22bee78 | 1606 | /* |
25511a47 | 1607 | * Function for @worker->rebind.work used to rebind unbound busy workers to |
403c821d TH |
1608 | * the associated cpu which is coming back online. This is scheduled by |
1609 | * cpu up but can race with other cpu hotplug operations and may be | |
1610 | * executed twice without intervening cpu down. | |
e22bee78 | 1611 | */ |
25511a47 | 1612 | static void busy_worker_rebind_fn(struct work_struct *work) |
e22bee78 TH |
1613 | { |
1614 | struct worker *worker = container_of(work, struct worker, rebind_work); | |
e22bee78 | 1615 | |
f36dc67b | 1616 | if (worker_maybe_bind_and_lock(worker->pool)) |
eab6d828 | 1617 | worker_clr_flags(worker, WORKER_UNBOUND); |
e22bee78 | 1618 | |
d565ed63 | 1619 | spin_unlock_irq(&worker->pool->lock); |
e22bee78 TH |
1620 | } |
1621 | ||
25511a47 | 1622 | /** |
94cf58bb TH |
1623 | * rebind_workers - rebind all workers of a pool to the associated CPU |
1624 | * @pool: pool of interest | |
25511a47 | 1625 | * |
94cf58bb | 1626 | * @pool->cpu is coming online. Rebind all workers to the CPU. Rebinding |
25511a47 TH |
1627 | * is different for idle and busy ones. |
1628 | * | |
ea1abd61 LJ |
1629 | * Idle ones will be removed from the idle_list and woken up. They will |
1630 | * add themselves back after completing rebind. This ensures that the | |
1631 | * idle_list doesn't contain any unbound workers when re-bound busy workers | |
1632 | * try to perform local wake-ups for concurrency management. | |
25511a47 | 1633 | * |
ea1abd61 LJ |
1634 | * Busy workers can rebind after they finish their current work items. |
1635 | * Queueing the rebind work item at the head of the scheduled list is | |
1636 | * enough. Note that nr_running will be properly bumped as busy workers | |
1637 | * rebind. | |
25511a47 | 1638 | * |
ea1abd61 LJ |
1639 | * On return, all non-manager workers are scheduled for rebind - see |
1640 | * manage_workers() for the manager special case. Any idle worker | |
1641 | * including the manager will not appear on @idle_list until rebind is | |
1642 | * complete, making local wake-ups safe. | |
25511a47 | 1643 | */ |
94cf58bb | 1644 | static void rebind_workers(struct worker_pool *pool) |
25511a47 | 1645 | { |
ea1abd61 | 1646 | struct worker *worker, *n; |
25511a47 TH |
1647 | int i; |
1648 | ||
94cf58bb TH |
1649 | lockdep_assert_held(&pool->assoc_mutex); |
1650 | lockdep_assert_held(&pool->lock); | |
25511a47 | 1651 | |
5f7dabfd | 1652 | /* dequeue and kick idle ones */ |
94cf58bb TH |
1653 | list_for_each_entry_safe(worker, n, &pool->idle_list, entry) { |
1654 | /* | |
1655 | * idle workers should be off @pool->idle_list until rebind | |
1656 | * is complete to avoid receiving premature local wake-ups. | |
1657 | */ | |
1658 | list_del_init(&worker->entry); | |
25511a47 | 1659 | |
94cf58bb TH |
1660 | /* |
1661 | * worker_thread() will see the above dequeuing and call | |
1662 | * idle_worker_rebind(). | |
1663 | */ | |
1664 | wake_up_process(worker->task); | |
1665 | } | |
25511a47 | 1666 | |
94cf58bb | 1667 | /* rebind busy workers */ |
b67bfe0d | 1668 | for_each_busy_worker(worker, i, pool) { |
94cf58bb TH |
1669 | struct work_struct *rebind_work = &worker->rebind_work; |
1670 | struct workqueue_struct *wq; | |
25511a47 | 1671 | |
94cf58bb TH |
1672 | if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, |
1673 | work_data_bits(rebind_work))) | |
1674 | continue; | |
25511a47 | 1675 | |
94cf58bb | 1676 | debug_work_activate(rebind_work); |
90beca5d | 1677 | |
94cf58bb TH |
1678 | /* |
1679 | * wq doesn't really matter but let's keep @worker->pool | |
112202d9 | 1680 | * and @pwq->pool consistent for sanity. |
94cf58bb TH |
1681 | */ |
1682 | if (std_worker_pool_pri(worker->pool)) | |
1683 | wq = system_highpri_wq; | |
1684 | else | |
1685 | wq = system_wq; | |
1686 | ||
7fb98ea7 | 1687 | insert_work(per_cpu_ptr(wq->cpu_pwqs, pool->cpu), rebind_work, |
94cf58bb TH |
1688 | worker->scheduled.next, |
1689 | work_color_to_flags(WORK_NO_COLOR)); | |
ec58815a | 1690 | } |
25511a47 TH |
1691 | } |
1692 | ||
c34056a3 TH |
1693 | static struct worker *alloc_worker(void) |
1694 | { | |
1695 | struct worker *worker; | |
1696 | ||
1697 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1698 | if (worker) { |
1699 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1700 | INIT_LIST_HEAD(&worker->scheduled); |
25511a47 | 1701 | INIT_WORK(&worker->rebind_work, busy_worker_rebind_fn); |
e22bee78 TH |
1702 | /* on creation a worker is in !idle && prep state */ |
1703 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1704 | } |
c34056a3 TH |
1705 | return worker; |
1706 | } | |
1707 | ||
1708 | /** | |
1709 | * create_worker - create a new workqueue worker | |
63d95a91 | 1710 | * @pool: pool the new worker will belong to |
c34056a3 | 1711 | * |
63d95a91 | 1712 | * Create a new worker which is bound to @pool. The returned worker |
c34056a3 TH |
1713 | * can be started by calling start_worker() or destroyed using |
1714 | * destroy_worker(). | |
1715 | * | |
1716 | * CONTEXT: | |
1717 | * Might sleep. Does GFP_KERNEL allocations. | |
1718 | * | |
1719 | * RETURNS: | |
1720 | * Pointer to the newly created worker. | |
1721 | */ | |
bc2ae0f5 | 1722 | static struct worker *create_worker(struct worker_pool *pool) |
c34056a3 | 1723 | { |
e34cdddb | 1724 | const char *pri = std_worker_pool_pri(pool) ? "H" : ""; |
c34056a3 | 1725 | struct worker *worker = NULL; |
f3421797 | 1726 | int id = -1; |
c34056a3 | 1727 | |
d565ed63 | 1728 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1729 | while (ida_get_new(&pool->worker_ida, &id)) { |
d565ed63 | 1730 | spin_unlock_irq(&pool->lock); |
bd7bdd43 | 1731 | if (!ida_pre_get(&pool->worker_ida, GFP_KERNEL)) |
c34056a3 | 1732 | goto fail; |
d565ed63 | 1733 | spin_lock_irq(&pool->lock); |
c34056a3 | 1734 | } |
d565ed63 | 1735 | spin_unlock_irq(&pool->lock); |
c34056a3 TH |
1736 | |
1737 | worker = alloc_worker(); | |
1738 | if (!worker) | |
1739 | goto fail; | |
1740 | ||
bd7bdd43 | 1741 | worker->pool = pool; |
c34056a3 TH |
1742 | worker->id = id; |
1743 | ||
ec22ca5e | 1744 | if (pool->cpu != WORK_CPU_UNBOUND) |
94dcf29a | 1745 | worker->task = kthread_create_on_node(worker_thread, |
ec22ca5e | 1746 | worker, cpu_to_node(pool->cpu), |
d84ff051 | 1747 | "kworker/%d:%d%s", pool->cpu, id, pri); |
f3421797 TH |
1748 | else |
1749 | worker->task = kthread_create(worker_thread, worker, | |
3270476a | 1750 | "kworker/u:%d%s", id, pri); |
c34056a3 TH |
1751 | if (IS_ERR(worker->task)) |
1752 | goto fail; | |
1753 | ||
e34cdddb | 1754 | if (std_worker_pool_pri(pool)) |
3270476a TH |
1755 | set_user_nice(worker->task, HIGHPRI_NICE_LEVEL); |
1756 | ||
db7bccf4 | 1757 | /* |
bc2ae0f5 | 1758 | * Determine CPU binding of the new worker depending on |
24647570 | 1759 | * %POOL_DISASSOCIATED. The caller is responsible for ensuring the |
bc2ae0f5 TH |
1760 | * flag remains stable across this function. See the comments |
1761 | * above the flag definition for details. | |
1762 | * | |
1763 | * As an unbound worker may later become a regular one if CPU comes | |
1764 | * online, make sure every worker has %PF_THREAD_BOUND set. | |
db7bccf4 | 1765 | */ |
24647570 | 1766 | if (!(pool->flags & POOL_DISASSOCIATED)) { |
ec22ca5e | 1767 | kthread_bind(worker->task, pool->cpu); |
bc2ae0f5 | 1768 | } else { |
db7bccf4 | 1769 | worker->task->flags |= PF_THREAD_BOUND; |
bc2ae0f5 | 1770 | worker->flags |= WORKER_UNBOUND; |
f3421797 | 1771 | } |
c34056a3 TH |
1772 | |
1773 | return worker; | |
1774 | fail: | |
1775 | if (id >= 0) { | |
d565ed63 | 1776 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1777 | ida_remove(&pool->worker_ida, id); |
d565ed63 | 1778 | spin_unlock_irq(&pool->lock); |
c34056a3 TH |
1779 | } |
1780 | kfree(worker); | |
1781 | return NULL; | |
1782 | } | |
1783 | ||
1784 | /** | |
1785 | * start_worker - start a newly created worker | |
1786 | * @worker: worker to start | |
1787 | * | |
706026c2 | 1788 | * Make the pool aware of @worker and start it. |
c34056a3 TH |
1789 | * |
1790 | * CONTEXT: | |
d565ed63 | 1791 | * spin_lock_irq(pool->lock). |
c34056a3 TH |
1792 | */ |
1793 | static void start_worker(struct worker *worker) | |
1794 | { | |
cb444766 | 1795 | worker->flags |= WORKER_STARTED; |
bd7bdd43 | 1796 | worker->pool->nr_workers++; |
c8e55f36 | 1797 | worker_enter_idle(worker); |
c34056a3 TH |
1798 | wake_up_process(worker->task); |
1799 | } | |
1800 | ||
1801 | /** | |
1802 | * destroy_worker - destroy a workqueue worker | |
1803 | * @worker: worker to be destroyed | |
1804 | * | |
706026c2 | 1805 | * Destroy @worker and adjust @pool stats accordingly. |
c8e55f36 TH |
1806 | * |
1807 | * CONTEXT: | |
d565ed63 | 1808 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
c34056a3 TH |
1809 | */ |
1810 | static void destroy_worker(struct worker *worker) | |
1811 | { | |
bd7bdd43 | 1812 | struct worker_pool *pool = worker->pool; |
c34056a3 TH |
1813 | int id = worker->id; |
1814 | ||
1815 | /* sanity check frenzy */ | |
6183c009 TH |
1816 | if (WARN_ON(worker->current_work) || |
1817 | WARN_ON(!list_empty(&worker->scheduled))) | |
1818 | return; | |
c34056a3 | 1819 | |
c8e55f36 | 1820 | if (worker->flags & WORKER_STARTED) |
bd7bdd43 | 1821 | pool->nr_workers--; |
c8e55f36 | 1822 | if (worker->flags & WORKER_IDLE) |
bd7bdd43 | 1823 | pool->nr_idle--; |
c8e55f36 TH |
1824 | |
1825 | list_del_init(&worker->entry); | |
cb444766 | 1826 | worker->flags |= WORKER_DIE; |
c8e55f36 | 1827 | |
d565ed63 | 1828 | spin_unlock_irq(&pool->lock); |
c8e55f36 | 1829 | |
c34056a3 TH |
1830 | kthread_stop(worker->task); |
1831 | kfree(worker); | |
1832 | ||
d565ed63 | 1833 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1834 | ida_remove(&pool->worker_ida, id); |
c34056a3 TH |
1835 | } |
1836 | ||
63d95a91 | 1837 | static void idle_worker_timeout(unsigned long __pool) |
e22bee78 | 1838 | { |
63d95a91 | 1839 | struct worker_pool *pool = (void *)__pool; |
e22bee78 | 1840 | |
d565ed63 | 1841 | spin_lock_irq(&pool->lock); |
e22bee78 | 1842 | |
63d95a91 | 1843 | if (too_many_workers(pool)) { |
e22bee78 TH |
1844 | struct worker *worker; |
1845 | unsigned long expires; | |
1846 | ||
1847 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1848 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1849 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1850 | ||
1851 | if (time_before(jiffies, expires)) | |
63d95a91 | 1852 | mod_timer(&pool->idle_timer, expires); |
e22bee78 TH |
1853 | else { |
1854 | /* it's been idle for too long, wake up manager */ | |
11ebea50 | 1855 | pool->flags |= POOL_MANAGE_WORKERS; |
63d95a91 | 1856 | wake_up_worker(pool); |
d5abe669 | 1857 | } |
e22bee78 TH |
1858 | } |
1859 | ||
d565ed63 | 1860 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1861 | } |
d5abe669 | 1862 | |
493a1724 | 1863 | static void send_mayday(struct work_struct *work) |
e22bee78 | 1864 | { |
112202d9 TH |
1865 | struct pool_workqueue *pwq = get_work_pwq(work); |
1866 | struct workqueue_struct *wq = pwq->wq; | |
493a1724 TH |
1867 | |
1868 | lockdep_assert_held(&workqueue_lock); | |
e22bee78 TH |
1869 | |
1870 | if (!(wq->flags & WQ_RESCUER)) | |
493a1724 | 1871 | return; |
e22bee78 TH |
1872 | |
1873 | /* mayday mayday mayday */ | |
493a1724 TH |
1874 | if (list_empty(&pwq->mayday_node)) { |
1875 | list_add_tail(&pwq->mayday_node, &wq->maydays); | |
e22bee78 | 1876 | wake_up_process(wq->rescuer->task); |
493a1724 | 1877 | } |
e22bee78 TH |
1878 | } |
1879 | ||
706026c2 | 1880 | static void pool_mayday_timeout(unsigned long __pool) |
e22bee78 | 1881 | { |
63d95a91 | 1882 | struct worker_pool *pool = (void *)__pool; |
e22bee78 TH |
1883 | struct work_struct *work; |
1884 | ||
493a1724 TH |
1885 | spin_lock_irq(&workqueue_lock); /* for wq->maydays */ |
1886 | spin_lock(&pool->lock); | |
e22bee78 | 1887 | |
63d95a91 | 1888 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
1889 | /* |
1890 | * We've been trying to create a new worker but | |
1891 | * haven't been successful. We might be hitting an | |
1892 | * allocation deadlock. Send distress signals to | |
1893 | * rescuers. | |
1894 | */ | |
63d95a91 | 1895 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 1896 | send_mayday(work); |
1da177e4 | 1897 | } |
e22bee78 | 1898 | |
493a1724 TH |
1899 | spin_unlock(&pool->lock); |
1900 | spin_unlock_irq(&workqueue_lock); | |
e22bee78 | 1901 | |
63d95a91 | 1902 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
1903 | } |
1904 | ||
e22bee78 TH |
1905 | /** |
1906 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 1907 | * @pool: pool to create a new worker for |
e22bee78 | 1908 | * |
63d95a91 | 1909 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
1910 | * have at least one idle worker on return from this function. If |
1911 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 1912 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
1913 | * possible allocation deadlock. |
1914 | * | |
1915 | * On return, need_to_create_worker() is guaranteed to be false and | |
1916 | * may_start_working() true. | |
1917 | * | |
1918 | * LOCKING: | |
d565ed63 | 1919 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1920 | * multiple times. Does GFP_KERNEL allocations. Called only from |
1921 | * manager. | |
1922 | * | |
1923 | * RETURNS: | |
d565ed63 | 1924 | * false if no action was taken and pool->lock stayed locked, true |
e22bee78 TH |
1925 | * otherwise. |
1926 | */ | |
63d95a91 | 1927 | static bool maybe_create_worker(struct worker_pool *pool) |
d565ed63 TH |
1928 | __releases(&pool->lock) |
1929 | __acquires(&pool->lock) | |
1da177e4 | 1930 | { |
63d95a91 | 1931 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1932 | return false; |
1933 | restart: | |
d565ed63 | 1934 | spin_unlock_irq(&pool->lock); |
9f9c2364 | 1935 | |
e22bee78 | 1936 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 1937 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
1938 | |
1939 | while (true) { | |
1940 | struct worker *worker; | |
1941 | ||
bc2ae0f5 | 1942 | worker = create_worker(pool); |
e22bee78 | 1943 | if (worker) { |
63d95a91 | 1944 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1945 | spin_lock_irq(&pool->lock); |
e22bee78 | 1946 | start_worker(worker); |
6183c009 TH |
1947 | if (WARN_ON_ONCE(need_to_create_worker(pool))) |
1948 | goto restart; | |
e22bee78 TH |
1949 | return true; |
1950 | } | |
1951 | ||
63d95a91 | 1952 | if (!need_to_create_worker(pool)) |
e22bee78 | 1953 | break; |
1da177e4 | 1954 | |
e22bee78 TH |
1955 | __set_current_state(TASK_INTERRUPTIBLE); |
1956 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1957 | |
63d95a91 | 1958 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1959 | break; |
1960 | } | |
1961 | ||
63d95a91 | 1962 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1963 | spin_lock_irq(&pool->lock); |
63d95a91 | 1964 | if (need_to_create_worker(pool)) |
e22bee78 TH |
1965 | goto restart; |
1966 | return true; | |
1967 | } | |
1968 | ||
1969 | /** | |
1970 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
63d95a91 | 1971 | * @pool: pool to destroy workers for |
e22bee78 | 1972 | * |
63d95a91 | 1973 | * Destroy @pool workers which have been idle for longer than |
e22bee78 TH |
1974 | * IDLE_WORKER_TIMEOUT. |
1975 | * | |
1976 | * LOCKING: | |
d565ed63 | 1977 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1978 | * multiple times. Called only from manager. |
1979 | * | |
1980 | * RETURNS: | |
d565ed63 | 1981 | * false if no action was taken and pool->lock stayed locked, true |
e22bee78 TH |
1982 | * otherwise. |
1983 | */ | |
63d95a91 | 1984 | static bool maybe_destroy_workers(struct worker_pool *pool) |
e22bee78 TH |
1985 | { |
1986 | bool ret = false; | |
1da177e4 | 1987 | |
63d95a91 | 1988 | while (too_many_workers(pool)) { |
e22bee78 TH |
1989 | struct worker *worker; |
1990 | unsigned long expires; | |
3af24433 | 1991 | |
63d95a91 | 1992 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 | 1993 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
85f4186a | 1994 | |
e22bee78 | 1995 | if (time_before(jiffies, expires)) { |
63d95a91 | 1996 | mod_timer(&pool->idle_timer, expires); |
3af24433 | 1997 | break; |
e22bee78 | 1998 | } |
1da177e4 | 1999 | |
e22bee78 TH |
2000 | destroy_worker(worker); |
2001 | ret = true; | |
1da177e4 | 2002 | } |
1e19ffc6 | 2003 | |
e22bee78 | 2004 | return ret; |
1e19ffc6 TH |
2005 | } |
2006 | ||
73f53c4a | 2007 | /** |
e22bee78 TH |
2008 | * manage_workers - manage worker pool |
2009 | * @worker: self | |
73f53c4a | 2010 | * |
706026c2 | 2011 | * Assume the manager role and manage the worker pool @worker belongs |
e22bee78 | 2012 | * to. At any given time, there can be only zero or one manager per |
706026c2 | 2013 | * pool. The exclusion is handled automatically by this function. |
e22bee78 TH |
2014 | * |
2015 | * The caller can safely start processing works on false return. On | |
2016 | * true return, it's guaranteed that need_to_create_worker() is false | |
2017 | * and may_start_working() is true. | |
73f53c4a TH |
2018 | * |
2019 | * CONTEXT: | |
d565ed63 | 2020 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
2021 | * multiple times. Does GFP_KERNEL allocations. |
2022 | * | |
2023 | * RETURNS: | |
d565ed63 TH |
2024 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
2025 | * multiple times. Does GFP_KERNEL allocations. | |
73f53c4a | 2026 | */ |
e22bee78 | 2027 | static bool manage_workers(struct worker *worker) |
73f53c4a | 2028 | { |
63d95a91 | 2029 | struct worker_pool *pool = worker->pool; |
e22bee78 | 2030 | bool ret = false; |
73f53c4a | 2031 | |
ee378aa4 | 2032 | if (pool->flags & POOL_MANAGING_WORKERS) |
e22bee78 | 2033 | return ret; |
1e19ffc6 | 2034 | |
552a37e9 | 2035 | pool->flags |= POOL_MANAGING_WORKERS; |
73f53c4a | 2036 | |
ee378aa4 LJ |
2037 | /* |
2038 | * To simplify both worker management and CPU hotplug, hold off | |
2039 | * management while hotplug is in progress. CPU hotplug path can't | |
2040 | * grab %POOL_MANAGING_WORKERS to achieve this because that can | |
2041 | * lead to idle worker depletion (all become busy thinking someone | |
2042 | * else is managing) which in turn can result in deadlock under | |
b2eb83d1 | 2043 | * extreme circumstances. Use @pool->assoc_mutex to synchronize |
ee378aa4 LJ |
2044 | * manager against CPU hotplug. |
2045 | * | |
b2eb83d1 | 2046 | * assoc_mutex would always be free unless CPU hotplug is in |
d565ed63 | 2047 | * progress. trylock first without dropping @pool->lock. |
ee378aa4 | 2048 | */ |
b2eb83d1 | 2049 | if (unlikely(!mutex_trylock(&pool->assoc_mutex))) { |
d565ed63 | 2050 | spin_unlock_irq(&pool->lock); |
b2eb83d1 | 2051 | mutex_lock(&pool->assoc_mutex); |
ee378aa4 LJ |
2052 | /* |
2053 | * CPU hotplug could have happened while we were waiting | |
b2eb83d1 | 2054 | * for assoc_mutex. Hotplug itself can't handle us |
ee378aa4 | 2055 | * because manager isn't either on idle or busy list, and |
706026c2 | 2056 | * @pool's state and ours could have deviated. |
ee378aa4 | 2057 | * |
b2eb83d1 | 2058 | * As hotplug is now excluded via assoc_mutex, we can |
ee378aa4 | 2059 | * simply try to bind. It will succeed or fail depending |
706026c2 | 2060 | * on @pool's current state. Try it and adjust |
ee378aa4 LJ |
2061 | * %WORKER_UNBOUND accordingly. |
2062 | */ | |
f36dc67b | 2063 | if (worker_maybe_bind_and_lock(pool)) |
ee378aa4 LJ |
2064 | worker->flags &= ~WORKER_UNBOUND; |
2065 | else | |
2066 | worker->flags |= WORKER_UNBOUND; | |
73f53c4a | 2067 | |
ee378aa4 LJ |
2068 | ret = true; |
2069 | } | |
73f53c4a | 2070 | |
11ebea50 | 2071 | pool->flags &= ~POOL_MANAGE_WORKERS; |
73f53c4a TH |
2072 | |
2073 | /* | |
e22bee78 TH |
2074 | * Destroy and then create so that may_start_working() is true |
2075 | * on return. | |
73f53c4a | 2076 | */ |
63d95a91 TH |
2077 | ret |= maybe_destroy_workers(pool); |
2078 | ret |= maybe_create_worker(pool); | |
e22bee78 | 2079 | |
552a37e9 | 2080 | pool->flags &= ~POOL_MANAGING_WORKERS; |
b2eb83d1 | 2081 | mutex_unlock(&pool->assoc_mutex); |
e22bee78 | 2082 | return ret; |
73f53c4a TH |
2083 | } |
2084 | ||
a62428c0 TH |
2085 | /** |
2086 | * process_one_work - process single work | |
c34056a3 | 2087 | * @worker: self |
a62428c0 TH |
2088 | * @work: work to process |
2089 | * | |
2090 | * Process @work. This function contains all the logics necessary to | |
2091 | * process a single work including synchronization against and | |
2092 | * interaction with other workers on the same cpu, queueing and | |
2093 | * flushing. As long as context requirement is met, any worker can | |
2094 | * call this function to process a work. | |
2095 | * | |
2096 | * CONTEXT: | |
d565ed63 | 2097 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
a62428c0 | 2098 | */ |
c34056a3 | 2099 | static void process_one_work(struct worker *worker, struct work_struct *work) |
d565ed63 TH |
2100 | __releases(&pool->lock) |
2101 | __acquires(&pool->lock) | |
a62428c0 | 2102 | { |
112202d9 | 2103 | struct pool_workqueue *pwq = get_work_pwq(work); |
bd7bdd43 | 2104 | struct worker_pool *pool = worker->pool; |
112202d9 | 2105 | bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE; |
73f53c4a | 2106 | int work_color; |
7e11629d | 2107 | struct worker *collision; |
a62428c0 TH |
2108 | #ifdef CONFIG_LOCKDEP |
2109 | /* | |
2110 | * It is permissible to free the struct work_struct from | |
2111 | * inside the function that is called from it, this we need to | |
2112 | * take into account for lockdep too. To avoid bogus "held | |
2113 | * lock freed" warnings as well as problems when looking into | |
2114 | * work->lockdep_map, make a copy and use that here. | |
2115 | */ | |
4d82a1de PZ |
2116 | struct lockdep_map lockdep_map; |
2117 | ||
2118 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 2119 | #endif |
6fec10a1 TH |
2120 | /* |
2121 | * Ensure we're on the correct CPU. DISASSOCIATED test is | |
2122 | * necessary to avoid spurious warnings from rescuers servicing the | |
24647570 | 2123 | * unbound or a disassociated pool. |
6fec10a1 | 2124 | */ |
5f7dabfd | 2125 | WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) && |
24647570 | 2126 | !(pool->flags & POOL_DISASSOCIATED) && |
ec22ca5e | 2127 | raw_smp_processor_id() != pool->cpu); |
25511a47 | 2128 | |
7e11629d TH |
2129 | /* |
2130 | * A single work shouldn't be executed concurrently by | |
2131 | * multiple workers on a single cpu. Check whether anyone is | |
2132 | * already processing the work. If so, defer the work to the | |
2133 | * currently executing one. | |
2134 | */ | |
c9e7cf27 | 2135 | collision = find_worker_executing_work(pool, work); |
7e11629d TH |
2136 | if (unlikely(collision)) { |
2137 | move_linked_works(work, &collision->scheduled, NULL); | |
2138 | return; | |
2139 | } | |
2140 | ||
8930caba | 2141 | /* claim and dequeue */ |
a62428c0 | 2142 | debug_work_deactivate(work); |
c9e7cf27 | 2143 | hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work); |
c34056a3 | 2144 | worker->current_work = work; |
a2c1c57b | 2145 | worker->current_func = work->func; |
112202d9 | 2146 | worker->current_pwq = pwq; |
73f53c4a | 2147 | work_color = get_work_color(work); |
7a22ad75 | 2148 | |
a62428c0 TH |
2149 | list_del_init(&work->entry); |
2150 | ||
fb0e7beb TH |
2151 | /* |
2152 | * CPU intensive works don't participate in concurrency | |
2153 | * management. They're the scheduler's responsibility. | |
2154 | */ | |
2155 | if (unlikely(cpu_intensive)) | |
2156 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
2157 | ||
974271c4 | 2158 | /* |
d565ed63 | 2159 | * Unbound pool isn't concurrency managed and work items should be |
974271c4 TH |
2160 | * executed ASAP. Wake up another worker if necessary. |
2161 | */ | |
63d95a91 TH |
2162 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
2163 | wake_up_worker(pool); | |
974271c4 | 2164 | |
8930caba | 2165 | /* |
7c3eed5c | 2166 | * Record the last pool and clear PENDING which should be the last |
d565ed63 | 2167 | * update to @work. Also, do this inside @pool->lock so that |
23657bb1 TH |
2168 | * PENDING and queued state changes happen together while IRQ is |
2169 | * disabled. | |
8930caba | 2170 | */ |
7c3eed5c | 2171 | set_work_pool_and_clear_pending(work, pool->id); |
a62428c0 | 2172 | |
d565ed63 | 2173 | spin_unlock_irq(&pool->lock); |
a62428c0 | 2174 | |
112202d9 | 2175 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
a62428c0 | 2176 | lock_map_acquire(&lockdep_map); |
e36c886a | 2177 | trace_workqueue_execute_start(work); |
a2c1c57b | 2178 | worker->current_func(work); |
e36c886a AV |
2179 | /* |
2180 | * While we must be careful to not use "work" after this, the trace | |
2181 | * point will only record its address. | |
2182 | */ | |
2183 | trace_workqueue_execute_end(work); | |
a62428c0 | 2184 | lock_map_release(&lockdep_map); |
112202d9 | 2185 | lock_map_release(&pwq->wq->lockdep_map); |
a62428c0 TH |
2186 | |
2187 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
044c782c VI |
2188 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
2189 | " last function: %pf\n", | |
a2c1c57b TH |
2190 | current->comm, preempt_count(), task_pid_nr(current), |
2191 | worker->current_func); | |
a62428c0 TH |
2192 | debug_show_held_locks(current); |
2193 | dump_stack(); | |
2194 | } | |
2195 | ||
d565ed63 | 2196 | spin_lock_irq(&pool->lock); |
a62428c0 | 2197 | |
fb0e7beb TH |
2198 | /* clear cpu intensive status */ |
2199 | if (unlikely(cpu_intensive)) | |
2200 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
2201 | ||
a62428c0 | 2202 | /* we're done with it, release */ |
42f8570f | 2203 | hash_del(&worker->hentry); |
c34056a3 | 2204 | worker->current_work = NULL; |
a2c1c57b | 2205 | worker->current_func = NULL; |
112202d9 TH |
2206 | worker->current_pwq = NULL; |
2207 | pwq_dec_nr_in_flight(pwq, work_color); | |
a62428c0 TH |
2208 | } |
2209 | ||
affee4b2 TH |
2210 | /** |
2211 | * process_scheduled_works - process scheduled works | |
2212 | * @worker: self | |
2213 | * | |
2214 | * Process all scheduled works. Please note that the scheduled list | |
2215 | * may change while processing a work, so this function repeatedly | |
2216 | * fetches a work from the top and executes it. | |
2217 | * | |
2218 | * CONTEXT: | |
d565ed63 | 2219 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
affee4b2 TH |
2220 | * multiple times. |
2221 | */ | |
2222 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 2223 | { |
affee4b2 TH |
2224 | while (!list_empty(&worker->scheduled)) { |
2225 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 2226 | struct work_struct, entry); |
c34056a3 | 2227 | process_one_work(worker, work); |
1da177e4 | 2228 | } |
1da177e4 LT |
2229 | } |
2230 | ||
4690c4ab TH |
2231 | /** |
2232 | * worker_thread - the worker thread function | |
c34056a3 | 2233 | * @__worker: self |
4690c4ab | 2234 | * |
706026c2 TH |
2235 | * The worker thread function. There are NR_CPU_WORKER_POOLS dynamic pools |
2236 | * of these per each cpu. These workers process all works regardless of | |
e22bee78 TH |
2237 | * their specific target workqueue. The only exception is works which |
2238 | * belong to workqueues with a rescuer which will be explained in | |
2239 | * rescuer_thread(). | |
4690c4ab | 2240 | */ |
c34056a3 | 2241 | static int worker_thread(void *__worker) |
1da177e4 | 2242 | { |
c34056a3 | 2243 | struct worker *worker = __worker; |
bd7bdd43 | 2244 | struct worker_pool *pool = worker->pool; |
1da177e4 | 2245 | |
e22bee78 TH |
2246 | /* tell the scheduler that this is a workqueue worker */ |
2247 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 2248 | woke_up: |
d565ed63 | 2249 | spin_lock_irq(&pool->lock); |
1da177e4 | 2250 | |
5f7dabfd LJ |
2251 | /* we are off idle list if destruction or rebind is requested */ |
2252 | if (unlikely(list_empty(&worker->entry))) { | |
d565ed63 | 2253 | spin_unlock_irq(&pool->lock); |
25511a47 | 2254 | |
5f7dabfd | 2255 | /* if DIE is set, destruction is requested */ |
25511a47 TH |
2256 | if (worker->flags & WORKER_DIE) { |
2257 | worker->task->flags &= ~PF_WQ_WORKER; | |
2258 | return 0; | |
2259 | } | |
2260 | ||
5f7dabfd | 2261 | /* otherwise, rebind */ |
25511a47 TH |
2262 | idle_worker_rebind(worker); |
2263 | goto woke_up; | |
c8e55f36 | 2264 | } |
affee4b2 | 2265 | |
c8e55f36 | 2266 | worker_leave_idle(worker); |
db7bccf4 | 2267 | recheck: |
e22bee78 | 2268 | /* no more worker necessary? */ |
63d95a91 | 2269 | if (!need_more_worker(pool)) |
e22bee78 TH |
2270 | goto sleep; |
2271 | ||
2272 | /* do we need to manage? */ | |
63d95a91 | 2273 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
2274 | goto recheck; |
2275 | ||
c8e55f36 TH |
2276 | /* |
2277 | * ->scheduled list can only be filled while a worker is | |
2278 | * preparing to process a work or actually processing it. | |
2279 | * Make sure nobody diddled with it while I was sleeping. | |
2280 | */ | |
6183c009 | 2281 | WARN_ON_ONCE(!list_empty(&worker->scheduled)); |
c8e55f36 | 2282 | |
e22bee78 TH |
2283 | /* |
2284 | * When control reaches this point, we're guaranteed to have | |
2285 | * at least one idle worker or that someone else has already | |
2286 | * assumed the manager role. | |
2287 | */ | |
2288 | worker_clr_flags(worker, WORKER_PREP); | |
2289 | ||
2290 | do { | |
c8e55f36 | 2291 | struct work_struct *work = |
bd7bdd43 | 2292 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2293 | struct work_struct, entry); |
2294 | ||
2295 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
2296 | /* optimization path, not strictly necessary */ | |
2297 | process_one_work(worker, work); | |
2298 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2299 | process_scheduled_works(worker); |
c8e55f36 TH |
2300 | } else { |
2301 | move_linked_works(work, &worker->scheduled, NULL); | |
2302 | process_scheduled_works(worker); | |
affee4b2 | 2303 | } |
63d95a91 | 2304 | } while (keep_working(pool)); |
e22bee78 TH |
2305 | |
2306 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 2307 | sleep: |
63d95a91 | 2308 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
e22bee78 | 2309 | goto recheck; |
d313dd85 | 2310 | |
c8e55f36 | 2311 | /* |
d565ed63 TH |
2312 | * pool->lock is held and there's no work to process and no need to |
2313 | * manage, sleep. Workers are woken up only while holding | |
2314 | * pool->lock or from local cpu, so setting the current state | |
2315 | * before releasing pool->lock is enough to prevent losing any | |
2316 | * event. | |
c8e55f36 TH |
2317 | */ |
2318 | worker_enter_idle(worker); | |
2319 | __set_current_state(TASK_INTERRUPTIBLE); | |
d565ed63 | 2320 | spin_unlock_irq(&pool->lock); |
c8e55f36 TH |
2321 | schedule(); |
2322 | goto woke_up; | |
1da177e4 LT |
2323 | } |
2324 | ||
e22bee78 TH |
2325 | /** |
2326 | * rescuer_thread - the rescuer thread function | |
111c225a | 2327 | * @__rescuer: self |
e22bee78 TH |
2328 | * |
2329 | * Workqueue rescuer thread function. There's one rescuer for each | |
2330 | * workqueue which has WQ_RESCUER set. | |
2331 | * | |
706026c2 | 2332 | * Regular work processing on a pool may block trying to create a new |
e22bee78 TH |
2333 | * worker which uses GFP_KERNEL allocation which has slight chance of |
2334 | * developing into deadlock if some works currently on the same queue | |
2335 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2336 | * the problem rescuer solves. | |
2337 | * | |
706026c2 TH |
2338 | * When such condition is possible, the pool summons rescuers of all |
2339 | * workqueues which have works queued on the pool and let them process | |
e22bee78 TH |
2340 | * those works so that forward progress can be guaranteed. |
2341 | * | |
2342 | * This should happen rarely. | |
2343 | */ | |
111c225a | 2344 | static int rescuer_thread(void *__rescuer) |
e22bee78 | 2345 | { |
111c225a TH |
2346 | struct worker *rescuer = __rescuer; |
2347 | struct workqueue_struct *wq = rescuer->rescue_wq; | |
e22bee78 | 2348 | struct list_head *scheduled = &rescuer->scheduled; |
e22bee78 TH |
2349 | |
2350 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
111c225a TH |
2351 | |
2352 | /* | |
2353 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it | |
2354 | * doesn't participate in concurrency management. | |
2355 | */ | |
2356 | rescuer->task->flags |= PF_WQ_WORKER; | |
e22bee78 TH |
2357 | repeat: |
2358 | set_current_state(TASK_INTERRUPTIBLE); | |
2359 | ||
412d32e6 MG |
2360 | if (kthread_should_stop()) { |
2361 | __set_current_state(TASK_RUNNING); | |
111c225a | 2362 | rescuer->task->flags &= ~PF_WQ_WORKER; |
e22bee78 | 2363 | return 0; |
412d32e6 | 2364 | } |
e22bee78 | 2365 | |
493a1724 TH |
2366 | /* see whether any pwq is asking for help */ |
2367 | spin_lock_irq(&workqueue_lock); | |
2368 | ||
2369 | while (!list_empty(&wq->maydays)) { | |
2370 | struct pool_workqueue *pwq = list_first_entry(&wq->maydays, | |
2371 | struct pool_workqueue, mayday_node); | |
112202d9 | 2372 | struct worker_pool *pool = pwq->pool; |
e22bee78 TH |
2373 | struct work_struct *work, *n; |
2374 | ||
2375 | __set_current_state(TASK_RUNNING); | |
493a1724 TH |
2376 | list_del_init(&pwq->mayday_node); |
2377 | ||
2378 | spin_unlock_irq(&workqueue_lock); | |
e22bee78 TH |
2379 | |
2380 | /* migrate to the target cpu if possible */ | |
f36dc67b | 2381 | worker_maybe_bind_and_lock(pool); |
b3104104 | 2382 | rescuer->pool = pool; |
e22bee78 TH |
2383 | |
2384 | /* | |
2385 | * Slurp in all works issued via this workqueue and | |
2386 | * process'em. | |
2387 | */ | |
6183c009 | 2388 | WARN_ON_ONCE(!list_empty(&rescuer->scheduled)); |
bd7bdd43 | 2389 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
112202d9 | 2390 | if (get_work_pwq(work) == pwq) |
e22bee78 TH |
2391 | move_linked_works(work, scheduled, &n); |
2392 | ||
2393 | process_scheduled_works(rescuer); | |
7576958a TH |
2394 | |
2395 | /* | |
d565ed63 | 2396 | * Leave this pool. If keep_working() is %true, notify a |
7576958a TH |
2397 | * regular worker; otherwise, we end up with 0 concurrency |
2398 | * and stalling the execution. | |
2399 | */ | |
63d95a91 TH |
2400 | if (keep_working(pool)) |
2401 | wake_up_worker(pool); | |
7576958a | 2402 | |
b3104104 | 2403 | rescuer->pool = NULL; |
493a1724 TH |
2404 | spin_unlock(&pool->lock); |
2405 | spin_lock(&workqueue_lock); | |
e22bee78 TH |
2406 | } |
2407 | ||
493a1724 TH |
2408 | spin_unlock_irq(&workqueue_lock); |
2409 | ||
111c225a TH |
2410 | /* rescuers should never participate in concurrency management */ |
2411 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); | |
e22bee78 TH |
2412 | schedule(); |
2413 | goto repeat; | |
1da177e4 LT |
2414 | } |
2415 | ||
fc2e4d70 ON |
2416 | struct wq_barrier { |
2417 | struct work_struct work; | |
2418 | struct completion done; | |
2419 | }; | |
2420 | ||
2421 | static void wq_barrier_func(struct work_struct *work) | |
2422 | { | |
2423 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2424 | complete(&barr->done); | |
2425 | } | |
2426 | ||
4690c4ab TH |
2427 | /** |
2428 | * insert_wq_barrier - insert a barrier work | |
112202d9 | 2429 | * @pwq: pwq to insert barrier into |
4690c4ab | 2430 | * @barr: wq_barrier to insert |
affee4b2 TH |
2431 | * @target: target work to attach @barr to |
2432 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2433 | * |
affee4b2 TH |
2434 | * @barr is linked to @target such that @barr is completed only after |
2435 | * @target finishes execution. Please note that the ordering | |
2436 | * guarantee is observed only with respect to @target and on the local | |
2437 | * cpu. | |
2438 | * | |
2439 | * Currently, a queued barrier can't be canceled. This is because | |
2440 | * try_to_grab_pending() can't determine whether the work to be | |
2441 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2442 | * flag of the previous work while there must be a valid next work | |
2443 | * after a work with LINKED flag set. | |
2444 | * | |
2445 | * Note that when @worker is non-NULL, @target may be modified | |
112202d9 | 2446 | * underneath us, so we can't reliably determine pwq from @target. |
4690c4ab TH |
2447 | * |
2448 | * CONTEXT: | |
d565ed63 | 2449 | * spin_lock_irq(pool->lock). |
4690c4ab | 2450 | */ |
112202d9 | 2451 | static void insert_wq_barrier(struct pool_workqueue *pwq, |
affee4b2 TH |
2452 | struct wq_barrier *barr, |
2453 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2454 | { |
affee4b2 TH |
2455 | struct list_head *head; |
2456 | unsigned int linked = 0; | |
2457 | ||
dc186ad7 | 2458 | /* |
d565ed63 | 2459 | * debugobject calls are safe here even with pool->lock locked |
dc186ad7 TG |
2460 | * as we know for sure that this will not trigger any of the |
2461 | * checks and call back into the fixup functions where we | |
2462 | * might deadlock. | |
2463 | */ | |
ca1cab37 | 2464 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2465 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2466 | init_completion(&barr->done); |
83c22520 | 2467 | |
affee4b2 TH |
2468 | /* |
2469 | * If @target is currently being executed, schedule the | |
2470 | * barrier to the worker; otherwise, put it after @target. | |
2471 | */ | |
2472 | if (worker) | |
2473 | head = worker->scheduled.next; | |
2474 | else { | |
2475 | unsigned long *bits = work_data_bits(target); | |
2476 | ||
2477 | head = target->entry.next; | |
2478 | /* there can already be other linked works, inherit and set */ | |
2479 | linked = *bits & WORK_STRUCT_LINKED; | |
2480 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2481 | } | |
2482 | ||
dc186ad7 | 2483 | debug_work_activate(&barr->work); |
112202d9 | 2484 | insert_work(pwq, &barr->work, head, |
affee4b2 | 2485 | work_color_to_flags(WORK_NO_COLOR) | linked); |
fc2e4d70 ON |
2486 | } |
2487 | ||
73f53c4a | 2488 | /** |
112202d9 | 2489 | * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing |
73f53c4a TH |
2490 | * @wq: workqueue being flushed |
2491 | * @flush_color: new flush color, < 0 for no-op | |
2492 | * @work_color: new work color, < 0 for no-op | |
2493 | * | |
112202d9 | 2494 | * Prepare pwqs for workqueue flushing. |
73f53c4a | 2495 | * |
112202d9 TH |
2496 | * If @flush_color is non-negative, flush_color on all pwqs should be |
2497 | * -1. If no pwq has in-flight commands at the specified color, all | |
2498 | * pwq->flush_color's stay at -1 and %false is returned. If any pwq | |
2499 | * has in flight commands, its pwq->flush_color is set to | |
2500 | * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq | |
73f53c4a TH |
2501 | * wakeup logic is armed and %true is returned. |
2502 | * | |
2503 | * The caller should have initialized @wq->first_flusher prior to | |
2504 | * calling this function with non-negative @flush_color. If | |
2505 | * @flush_color is negative, no flush color update is done and %false | |
2506 | * is returned. | |
2507 | * | |
112202d9 | 2508 | * If @work_color is non-negative, all pwqs should have the same |
73f53c4a TH |
2509 | * work_color which is previous to @work_color and all will be |
2510 | * advanced to @work_color. | |
2511 | * | |
2512 | * CONTEXT: | |
2513 | * mutex_lock(wq->flush_mutex). | |
2514 | * | |
2515 | * RETURNS: | |
2516 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2517 | * otherwise. | |
2518 | */ | |
112202d9 | 2519 | static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq, |
73f53c4a | 2520 | int flush_color, int work_color) |
1da177e4 | 2521 | { |
73f53c4a | 2522 | bool wait = false; |
49e3cf44 | 2523 | struct pool_workqueue *pwq; |
1da177e4 | 2524 | |
73f53c4a | 2525 | if (flush_color >= 0) { |
6183c009 | 2526 | WARN_ON_ONCE(atomic_read(&wq->nr_pwqs_to_flush)); |
112202d9 | 2527 | atomic_set(&wq->nr_pwqs_to_flush, 1); |
1da177e4 | 2528 | } |
2355b70f | 2529 | |
76af4d93 TH |
2530 | local_irq_disable(); |
2531 | ||
49e3cf44 | 2532 | for_each_pwq(pwq, wq) { |
112202d9 | 2533 | struct worker_pool *pool = pwq->pool; |
fc2e4d70 | 2534 | |
76af4d93 | 2535 | spin_lock(&pool->lock); |
83c22520 | 2536 | |
73f53c4a | 2537 | if (flush_color >= 0) { |
6183c009 | 2538 | WARN_ON_ONCE(pwq->flush_color != -1); |
fc2e4d70 | 2539 | |
112202d9 TH |
2540 | if (pwq->nr_in_flight[flush_color]) { |
2541 | pwq->flush_color = flush_color; | |
2542 | atomic_inc(&wq->nr_pwqs_to_flush); | |
73f53c4a TH |
2543 | wait = true; |
2544 | } | |
2545 | } | |
1da177e4 | 2546 | |
73f53c4a | 2547 | if (work_color >= 0) { |
6183c009 | 2548 | WARN_ON_ONCE(work_color != work_next_color(pwq->work_color)); |
112202d9 | 2549 | pwq->work_color = work_color; |
73f53c4a | 2550 | } |
1da177e4 | 2551 | |
76af4d93 | 2552 | spin_unlock(&pool->lock); |
1da177e4 | 2553 | } |
2355b70f | 2554 | |
76af4d93 TH |
2555 | local_irq_enable(); |
2556 | ||
112202d9 | 2557 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_pwqs_to_flush)) |
73f53c4a | 2558 | complete(&wq->first_flusher->done); |
14441960 | 2559 | |
73f53c4a | 2560 | return wait; |
1da177e4 LT |
2561 | } |
2562 | ||
0fcb78c2 | 2563 | /** |
1da177e4 | 2564 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2565 | * @wq: workqueue to flush |
1da177e4 LT |
2566 | * |
2567 | * Forces execution of the workqueue and blocks until its completion. | |
2568 | * This is typically used in driver shutdown handlers. | |
2569 | * | |
fc2e4d70 ON |
2570 | * We sleep until all works which were queued on entry have been handled, |
2571 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 2572 | */ |
7ad5b3a5 | 2573 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2574 | { |
73f53c4a TH |
2575 | struct wq_flusher this_flusher = { |
2576 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2577 | .flush_color = -1, | |
2578 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2579 | }; | |
2580 | int next_color; | |
1da177e4 | 2581 | |
3295f0ef IM |
2582 | lock_map_acquire(&wq->lockdep_map); |
2583 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
2584 | |
2585 | mutex_lock(&wq->flush_mutex); | |
2586 | ||
2587 | /* | |
2588 | * Start-to-wait phase | |
2589 | */ | |
2590 | next_color = work_next_color(wq->work_color); | |
2591 | ||
2592 | if (next_color != wq->flush_color) { | |
2593 | /* | |
2594 | * Color space is not full. The current work_color | |
2595 | * becomes our flush_color and work_color is advanced | |
2596 | * by one. | |
2597 | */ | |
6183c009 | 2598 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow)); |
73f53c4a TH |
2599 | this_flusher.flush_color = wq->work_color; |
2600 | wq->work_color = next_color; | |
2601 | ||
2602 | if (!wq->first_flusher) { | |
2603 | /* no flush in progress, become the first flusher */ | |
6183c009 | 2604 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); |
73f53c4a TH |
2605 | |
2606 | wq->first_flusher = &this_flusher; | |
2607 | ||
112202d9 | 2608 | if (!flush_workqueue_prep_pwqs(wq, wq->flush_color, |
73f53c4a TH |
2609 | wq->work_color)) { |
2610 | /* nothing to flush, done */ | |
2611 | wq->flush_color = next_color; | |
2612 | wq->first_flusher = NULL; | |
2613 | goto out_unlock; | |
2614 | } | |
2615 | } else { | |
2616 | /* wait in queue */ | |
6183c009 | 2617 | WARN_ON_ONCE(wq->flush_color == this_flusher.flush_color); |
73f53c4a | 2618 | list_add_tail(&this_flusher.list, &wq->flusher_queue); |
112202d9 | 2619 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2620 | } |
2621 | } else { | |
2622 | /* | |
2623 | * Oops, color space is full, wait on overflow queue. | |
2624 | * The next flush completion will assign us | |
2625 | * flush_color and transfer to flusher_queue. | |
2626 | */ | |
2627 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2628 | } | |
2629 | ||
2630 | mutex_unlock(&wq->flush_mutex); | |
2631 | ||
2632 | wait_for_completion(&this_flusher.done); | |
2633 | ||
2634 | /* | |
2635 | * Wake-up-and-cascade phase | |
2636 | * | |
2637 | * First flushers are responsible for cascading flushes and | |
2638 | * handling overflow. Non-first flushers can simply return. | |
2639 | */ | |
2640 | if (wq->first_flusher != &this_flusher) | |
2641 | return; | |
2642 | ||
2643 | mutex_lock(&wq->flush_mutex); | |
2644 | ||
4ce48b37 TH |
2645 | /* we might have raced, check again with mutex held */ |
2646 | if (wq->first_flusher != &this_flusher) | |
2647 | goto out_unlock; | |
2648 | ||
73f53c4a TH |
2649 | wq->first_flusher = NULL; |
2650 | ||
6183c009 TH |
2651 | WARN_ON_ONCE(!list_empty(&this_flusher.list)); |
2652 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); | |
73f53c4a TH |
2653 | |
2654 | while (true) { | |
2655 | struct wq_flusher *next, *tmp; | |
2656 | ||
2657 | /* complete all the flushers sharing the current flush color */ | |
2658 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2659 | if (next->flush_color != wq->flush_color) | |
2660 | break; | |
2661 | list_del_init(&next->list); | |
2662 | complete(&next->done); | |
2663 | } | |
2664 | ||
6183c009 TH |
2665 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow) && |
2666 | wq->flush_color != work_next_color(wq->work_color)); | |
73f53c4a TH |
2667 | |
2668 | /* this flush_color is finished, advance by one */ | |
2669 | wq->flush_color = work_next_color(wq->flush_color); | |
2670 | ||
2671 | /* one color has been freed, handle overflow queue */ | |
2672 | if (!list_empty(&wq->flusher_overflow)) { | |
2673 | /* | |
2674 | * Assign the same color to all overflowed | |
2675 | * flushers, advance work_color and append to | |
2676 | * flusher_queue. This is the start-to-wait | |
2677 | * phase for these overflowed flushers. | |
2678 | */ | |
2679 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2680 | tmp->flush_color = wq->work_color; | |
2681 | ||
2682 | wq->work_color = work_next_color(wq->work_color); | |
2683 | ||
2684 | list_splice_tail_init(&wq->flusher_overflow, | |
2685 | &wq->flusher_queue); | |
112202d9 | 2686 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2687 | } |
2688 | ||
2689 | if (list_empty(&wq->flusher_queue)) { | |
6183c009 | 2690 | WARN_ON_ONCE(wq->flush_color != wq->work_color); |
73f53c4a TH |
2691 | break; |
2692 | } | |
2693 | ||
2694 | /* | |
2695 | * Need to flush more colors. Make the next flusher | |
112202d9 | 2696 | * the new first flusher and arm pwqs. |
73f53c4a | 2697 | */ |
6183c009 TH |
2698 | WARN_ON_ONCE(wq->flush_color == wq->work_color); |
2699 | WARN_ON_ONCE(wq->flush_color != next->flush_color); | |
73f53c4a TH |
2700 | |
2701 | list_del_init(&next->list); | |
2702 | wq->first_flusher = next; | |
2703 | ||
112202d9 | 2704 | if (flush_workqueue_prep_pwqs(wq, wq->flush_color, -1)) |
73f53c4a TH |
2705 | break; |
2706 | ||
2707 | /* | |
2708 | * Meh... this color is already done, clear first | |
2709 | * flusher and repeat cascading. | |
2710 | */ | |
2711 | wq->first_flusher = NULL; | |
2712 | } | |
2713 | ||
2714 | out_unlock: | |
2715 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 2716 | } |
ae90dd5d | 2717 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2718 | |
9c5a2ba7 TH |
2719 | /** |
2720 | * drain_workqueue - drain a workqueue | |
2721 | * @wq: workqueue to drain | |
2722 | * | |
2723 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2724 | * only chain queueing is allowed. IOW, only currently pending or running | |
2725 | * work items on @wq can queue further work items on it. @wq is flushed | |
2726 | * repeatedly until it becomes empty. The number of flushing is detemined | |
2727 | * by the depth of chaining and should be relatively short. Whine if it | |
2728 | * takes too long. | |
2729 | */ | |
2730 | void drain_workqueue(struct workqueue_struct *wq) | |
2731 | { | |
2732 | unsigned int flush_cnt = 0; | |
49e3cf44 | 2733 | struct pool_workqueue *pwq; |
9c5a2ba7 TH |
2734 | |
2735 | /* | |
2736 | * __queue_work() needs to test whether there are drainers, is much | |
2737 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
2738 | * Use WQ_DRAINING so that queue doesn't have to check nr_drainers. | |
2739 | */ | |
e98d5b16 | 2740 | spin_lock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2741 | if (!wq->nr_drainers++) |
2742 | wq->flags |= WQ_DRAINING; | |
e98d5b16 | 2743 | spin_unlock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2744 | reflush: |
2745 | flush_workqueue(wq); | |
2746 | ||
76af4d93 TH |
2747 | local_irq_disable(); |
2748 | ||
49e3cf44 | 2749 | for_each_pwq(pwq, wq) { |
fa2563e4 | 2750 | bool drained; |
9c5a2ba7 | 2751 | |
76af4d93 | 2752 | spin_lock(&pwq->pool->lock); |
112202d9 | 2753 | drained = !pwq->nr_active && list_empty(&pwq->delayed_works); |
76af4d93 | 2754 | spin_unlock(&pwq->pool->lock); |
fa2563e4 TT |
2755 | |
2756 | if (drained) | |
9c5a2ba7 TH |
2757 | continue; |
2758 | ||
2759 | if (++flush_cnt == 10 || | |
2760 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
044c782c VI |
2761 | pr_warn("workqueue %s: flush on destruction isn't complete after %u tries\n", |
2762 | wq->name, flush_cnt); | |
76af4d93 TH |
2763 | |
2764 | local_irq_enable(); | |
9c5a2ba7 TH |
2765 | goto reflush; |
2766 | } | |
2767 | ||
76af4d93 | 2768 | spin_lock(&workqueue_lock); |
9c5a2ba7 TH |
2769 | if (!--wq->nr_drainers) |
2770 | wq->flags &= ~WQ_DRAINING; | |
76af4d93 TH |
2771 | spin_unlock(&workqueue_lock); |
2772 | ||
2773 | local_irq_enable(); | |
9c5a2ba7 TH |
2774 | } |
2775 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2776 | ||
606a5020 | 2777 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) |
db700897 | 2778 | { |
affee4b2 | 2779 | struct worker *worker = NULL; |
c9e7cf27 | 2780 | struct worker_pool *pool; |
112202d9 | 2781 | struct pool_workqueue *pwq; |
db700897 ON |
2782 | |
2783 | might_sleep(); | |
fa1b54e6 TH |
2784 | |
2785 | local_irq_disable(); | |
c9e7cf27 | 2786 | pool = get_work_pool(work); |
fa1b54e6 TH |
2787 | if (!pool) { |
2788 | local_irq_enable(); | |
baf59022 | 2789 | return false; |
fa1b54e6 | 2790 | } |
db700897 | 2791 | |
fa1b54e6 | 2792 | spin_lock(&pool->lock); |
0b3dae68 | 2793 | /* see the comment in try_to_grab_pending() with the same code */ |
112202d9 TH |
2794 | pwq = get_work_pwq(work); |
2795 | if (pwq) { | |
2796 | if (unlikely(pwq->pool != pool)) | |
4690c4ab | 2797 | goto already_gone; |
606a5020 | 2798 | } else { |
c9e7cf27 | 2799 | worker = find_worker_executing_work(pool, work); |
affee4b2 | 2800 | if (!worker) |
4690c4ab | 2801 | goto already_gone; |
112202d9 | 2802 | pwq = worker->current_pwq; |
606a5020 | 2803 | } |
db700897 | 2804 | |
112202d9 | 2805 | insert_wq_barrier(pwq, barr, work, worker); |
d565ed63 | 2806 | spin_unlock_irq(&pool->lock); |
7a22ad75 | 2807 | |
e159489b TH |
2808 | /* |
2809 | * If @max_active is 1 or rescuer is in use, flushing another work | |
2810 | * item on the same workqueue may lead to deadlock. Make sure the | |
2811 | * flusher is not running on the same workqueue by verifying write | |
2812 | * access. | |
2813 | */ | |
112202d9 TH |
2814 | if (pwq->wq->saved_max_active == 1 || pwq->wq->flags & WQ_RESCUER) |
2815 | lock_map_acquire(&pwq->wq->lockdep_map); | |
e159489b | 2816 | else |
112202d9 TH |
2817 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
2818 | lock_map_release(&pwq->wq->lockdep_map); | |
e159489b | 2819 | |
401a8d04 | 2820 | return true; |
4690c4ab | 2821 | already_gone: |
d565ed63 | 2822 | spin_unlock_irq(&pool->lock); |
401a8d04 | 2823 | return false; |
db700897 | 2824 | } |
baf59022 TH |
2825 | |
2826 | /** | |
2827 | * flush_work - wait for a work to finish executing the last queueing instance | |
2828 | * @work: the work to flush | |
2829 | * | |
606a5020 TH |
2830 | * Wait until @work has finished execution. @work is guaranteed to be idle |
2831 | * on return if it hasn't been requeued since flush started. | |
baf59022 TH |
2832 | * |
2833 | * RETURNS: | |
2834 | * %true if flush_work() waited for the work to finish execution, | |
2835 | * %false if it was already idle. | |
2836 | */ | |
2837 | bool flush_work(struct work_struct *work) | |
2838 | { | |
2839 | struct wq_barrier barr; | |
2840 | ||
0976dfc1 SB |
2841 | lock_map_acquire(&work->lockdep_map); |
2842 | lock_map_release(&work->lockdep_map); | |
2843 | ||
606a5020 | 2844 | if (start_flush_work(work, &barr)) { |
401a8d04 TH |
2845 | wait_for_completion(&barr.done); |
2846 | destroy_work_on_stack(&barr.work); | |
2847 | return true; | |
606a5020 | 2848 | } else { |
401a8d04 | 2849 | return false; |
6e84d644 | 2850 | } |
6e84d644 | 2851 | } |
606a5020 | 2852 | EXPORT_SYMBOL_GPL(flush_work); |
6e84d644 | 2853 | |
36e227d2 | 2854 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
1f1f642e | 2855 | { |
bbb68dfa | 2856 | unsigned long flags; |
1f1f642e ON |
2857 | int ret; |
2858 | ||
2859 | do { | |
bbb68dfa TH |
2860 | ret = try_to_grab_pending(work, is_dwork, &flags); |
2861 | /* | |
2862 | * If someone else is canceling, wait for the same event it | |
2863 | * would be waiting for before retrying. | |
2864 | */ | |
2865 | if (unlikely(ret == -ENOENT)) | |
606a5020 | 2866 | flush_work(work); |
1f1f642e ON |
2867 | } while (unlikely(ret < 0)); |
2868 | ||
bbb68dfa TH |
2869 | /* tell other tasks trying to grab @work to back off */ |
2870 | mark_work_canceling(work); | |
2871 | local_irq_restore(flags); | |
2872 | ||
606a5020 | 2873 | flush_work(work); |
7a22ad75 | 2874 | clear_work_data(work); |
1f1f642e ON |
2875 | return ret; |
2876 | } | |
2877 | ||
6e84d644 | 2878 | /** |
401a8d04 TH |
2879 | * cancel_work_sync - cancel a work and wait for it to finish |
2880 | * @work: the work to cancel | |
6e84d644 | 2881 | * |
401a8d04 TH |
2882 | * Cancel @work and wait for its execution to finish. This function |
2883 | * can be used even if the work re-queues itself or migrates to | |
2884 | * another workqueue. On return from this function, @work is | |
2885 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2886 | * |
401a8d04 TH |
2887 | * cancel_work_sync(&delayed_work->work) must not be used for |
2888 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2889 | * |
401a8d04 | 2890 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2891 | * queued can't be destroyed before this function returns. |
401a8d04 TH |
2892 | * |
2893 | * RETURNS: | |
2894 | * %true if @work was pending, %false otherwise. | |
6e84d644 | 2895 | */ |
401a8d04 | 2896 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2897 | { |
36e227d2 | 2898 | return __cancel_work_timer(work, false); |
b89deed3 | 2899 | } |
28e53bdd | 2900 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2901 | |
6e84d644 | 2902 | /** |
401a8d04 TH |
2903 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2904 | * @dwork: the delayed work to flush | |
6e84d644 | 2905 | * |
401a8d04 TH |
2906 | * Delayed timer is cancelled and the pending work is queued for |
2907 | * immediate execution. Like flush_work(), this function only | |
2908 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2909 | * |
401a8d04 TH |
2910 | * RETURNS: |
2911 | * %true if flush_work() waited for the work to finish execution, | |
2912 | * %false if it was already idle. | |
6e84d644 | 2913 | */ |
401a8d04 TH |
2914 | bool flush_delayed_work(struct delayed_work *dwork) |
2915 | { | |
8930caba | 2916 | local_irq_disable(); |
401a8d04 | 2917 | if (del_timer_sync(&dwork->timer)) |
60c057bc | 2918 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
8930caba | 2919 | local_irq_enable(); |
401a8d04 TH |
2920 | return flush_work(&dwork->work); |
2921 | } | |
2922 | EXPORT_SYMBOL(flush_delayed_work); | |
2923 | ||
09383498 | 2924 | /** |
57b30ae7 TH |
2925 | * cancel_delayed_work - cancel a delayed work |
2926 | * @dwork: delayed_work to cancel | |
09383498 | 2927 | * |
57b30ae7 TH |
2928 | * Kill off a pending delayed_work. Returns %true if @dwork was pending |
2929 | * and canceled; %false if wasn't pending. Note that the work callback | |
2930 | * function may still be running on return, unless it returns %true and the | |
2931 | * work doesn't re-arm itself. Explicitly flush or use | |
2932 | * cancel_delayed_work_sync() to wait on it. | |
09383498 | 2933 | * |
57b30ae7 | 2934 | * This function is safe to call from any context including IRQ handler. |
09383498 | 2935 | */ |
57b30ae7 | 2936 | bool cancel_delayed_work(struct delayed_work *dwork) |
09383498 | 2937 | { |
57b30ae7 TH |
2938 | unsigned long flags; |
2939 | int ret; | |
2940 | ||
2941 | do { | |
2942 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
2943 | } while (unlikely(ret == -EAGAIN)); | |
2944 | ||
2945 | if (unlikely(ret < 0)) | |
2946 | return false; | |
2947 | ||
7c3eed5c TH |
2948 | set_work_pool_and_clear_pending(&dwork->work, |
2949 | get_work_pool_id(&dwork->work)); | |
57b30ae7 | 2950 | local_irq_restore(flags); |
c0158ca6 | 2951 | return ret; |
09383498 | 2952 | } |
57b30ae7 | 2953 | EXPORT_SYMBOL(cancel_delayed_work); |
09383498 | 2954 | |
401a8d04 TH |
2955 | /** |
2956 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
2957 | * @dwork: the delayed work cancel | |
2958 | * | |
2959 | * This is cancel_work_sync() for delayed works. | |
2960 | * | |
2961 | * RETURNS: | |
2962 | * %true if @dwork was pending, %false otherwise. | |
2963 | */ | |
2964 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 2965 | { |
36e227d2 | 2966 | return __cancel_work_timer(&dwork->work, true); |
6e84d644 | 2967 | } |
f5a421a4 | 2968 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 2969 | |
0fcb78c2 | 2970 | /** |
c1a220e7 ZR |
2971 | * schedule_work_on - put work task on a specific cpu |
2972 | * @cpu: cpu to put the work task on | |
2973 | * @work: job to be done | |
2974 | * | |
2975 | * This puts a job on a specific cpu | |
2976 | */ | |
d4283e93 | 2977 | bool schedule_work_on(int cpu, struct work_struct *work) |
c1a220e7 | 2978 | { |
d320c038 | 2979 | return queue_work_on(cpu, system_wq, work); |
c1a220e7 ZR |
2980 | } |
2981 | EXPORT_SYMBOL(schedule_work_on); | |
2982 | ||
0fcb78c2 | 2983 | /** |
0fcb78c2 REB |
2984 | * schedule_work - put work task in global workqueue |
2985 | * @work: job to be done | |
0fcb78c2 | 2986 | * |
d4283e93 TH |
2987 | * Returns %false if @work was already on the kernel-global workqueue and |
2988 | * %true otherwise. | |
5b0f437d BVA |
2989 | * |
2990 | * This puts a job in the kernel-global workqueue if it was not already | |
2991 | * queued and leaves it in the same position on the kernel-global | |
2992 | * workqueue otherwise. | |
0fcb78c2 | 2993 | */ |
d4283e93 | 2994 | bool schedule_work(struct work_struct *work) |
1da177e4 | 2995 | { |
d320c038 | 2996 | return queue_work(system_wq, work); |
1da177e4 | 2997 | } |
ae90dd5d | 2998 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 2999 | |
0fcb78c2 REB |
3000 | /** |
3001 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
3002 | * @cpu: cpu to use | |
52bad64d | 3003 | * @dwork: job to be done |
0fcb78c2 REB |
3004 | * @delay: number of jiffies to wait |
3005 | * | |
3006 | * After waiting for a given time this puts a job in the kernel-global | |
3007 | * workqueue on the specified CPU. | |
3008 | */ | |
d4283e93 TH |
3009 | bool schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
3010 | unsigned long delay) | |
1da177e4 | 3011 | { |
d320c038 | 3012 | return queue_delayed_work_on(cpu, system_wq, dwork, delay); |
1da177e4 | 3013 | } |
ae90dd5d | 3014 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 3015 | |
0fcb78c2 REB |
3016 | /** |
3017 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
3018 | * @dwork: job to be done |
3019 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
3020 | * |
3021 | * After waiting for a given time this puts a job in the kernel-global | |
3022 | * workqueue. | |
3023 | */ | |
d4283e93 | 3024 | bool schedule_delayed_work(struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 3025 | { |
d320c038 | 3026 | return queue_delayed_work(system_wq, dwork, delay); |
1da177e4 | 3027 | } |
ae90dd5d | 3028 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 3029 | |
b6136773 | 3030 | /** |
31ddd871 | 3031 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 3032 | * @func: the function to call |
b6136773 | 3033 | * |
31ddd871 TH |
3034 | * schedule_on_each_cpu() executes @func on each online CPU using the |
3035 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 3036 | * schedule_on_each_cpu() is very slow. |
31ddd871 TH |
3037 | * |
3038 | * RETURNS: | |
3039 | * 0 on success, -errno on failure. | |
b6136773 | 3040 | */ |
65f27f38 | 3041 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
3042 | { |
3043 | int cpu; | |
38f51568 | 3044 | struct work_struct __percpu *works; |
15316ba8 | 3045 | |
b6136773 AM |
3046 | works = alloc_percpu(struct work_struct); |
3047 | if (!works) | |
15316ba8 | 3048 | return -ENOMEM; |
b6136773 | 3049 | |
93981800 TH |
3050 | get_online_cpus(); |
3051 | ||
15316ba8 | 3052 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
3053 | struct work_struct *work = per_cpu_ptr(works, cpu); |
3054 | ||
3055 | INIT_WORK(work, func); | |
b71ab8c2 | 3056 | schedule_work_on(cpu, work); |
65a64464 | 3057 | } |
93981800 TH |
3058 | |
3059 | for_each_online_cpu(cpu) | |
3060 | flush_work(per_cpu_ptr(works, cpu)); | |
3061 | ||
95402b38 | 3062 | put_online_cpus(); |
b6136773 | 3063 | free_percpu(works); |
15316ba8 CL |
3064 | return 0; |
3065 | } | |
3066 | ||
eef6a7d5 AS |
3067 | /** |
3068 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
3069 | * | |
3070 | * Forces execution of the kernel-global workqueue and blocks until its | |
3071 | * completion. | |
3072 | * | |
3073 | * Think twice before calling this function! It's very easy to get into | |
3074 | * trouble if you don't take great care. Either of the following situations | |
3075 | * will lead to deadlock: | |
3076 | * | |
3077 | * One of the work items currently on the workqueue needs to acquire | |
3078 | * a lock held by your code or its caller. | |
3079 | * | |
3080 | * Your code is running in the context of a work routine. | |
3081 | * | |
3082 | * They will be detected by lockdep when they occur, but the first might not | |
3083 | * occur very often. It depends on what work items are on the workqueue and | |
3084 | * what locks they need, which you have no control over. | |
3085 | * | |
3086 | * In most situations flushing the entire workqueue is overkill; you merely | |
3087 | * need to know that a particular work item isn't queued and isn't running. | |
3088 | * In such cases you should use cancel_delayed_work_sync() or | |
3089 | * cancel_work_sync() instead. | |
3090 | */ | |
1da177e4 LT |
3091 | void flush_scheduled_work(void) |
3092 | { | |
d320c038 | 3093 | flush_workqueue(system_wq); |
1da177e4 | 3094 | } |
ae90dd5d | 3095 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 3096 | |
1fa44eca JB |
3097 | /** |
3098 | * execute_in_process_context - reliably execute the routine with user context | |
3099 | * @fn: the function to execute | |
1fa44eca JB |
3100 | * @ew: guaranteed storage for the execute work structure (must |
3101 | * be available when the work executes) | |
3102 | * | |
3103 | * Executes the function immediately if process context is available, | |
3104 | * otherwise schedules the function for delayed execution. | |
3105 | * | |
3106 | * Returns: 0 - function was executed | |
3107 | * 1 - function was scheduled for execution | |
3108 | */ | |
65f27f38 | 3109 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
3110 | { |
3111 | if (!in_interrupt()) { | |
65f27f38 | 3112 | fn(&ew->work); |
1fa44eca JB |
3113 | return 0; |
3114 | } | |
3115 | ||
65f27f38 | 3116 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
3117 | schedule_work(&ew->work); |
3118 | ||
3119 | return 1; | |
3120 | } | |
3121 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
3122 | ||
1da177e4 LT |
3123 | int keventd_up(void) |
3124 | { | |
d320c038 | 3125 | return system_wq != NULL; |
1da177e4 LT |
3126 | } |
3127 | ||
30cdf249 | 3128 | static int alloc_and_link_pwqs(struct workqueue_struct *wq) |
0f900049 | 3129 | { |
49e3cf44 | 3130 | bool highpri = wq->flags & WQ_HIGHPRI; |
30cdf249 TH |
3131 | int cpu; |
3132 | ||
3133 | if (!(wq->flags & WQ_UNBOUND)) { | |
420c0ddb TH |
3134 | wq->cpu_pwqs = alloc_percpu(struct pool_workqueue); |
3135 | if (!wq->cpu_pwqs) | |
30cdf249 TH |
3136 | return -ENOMEM; |
3137 | ||
3138 | for_each_possible_cpu(cpu) { | |
7fb98ea7 TH |
3139 | struct pool_workqueue *pwq = |
3140 | per_cpu_ptr(wq->cpu_pwqs, cpu); | |
f3421797 | 3141 | |
49e3cf44 | 3142 | pwq->pool = get_std_worker_pool(cpu, highpri); |
76af4d93 | 3143 | list_add_tail_rcu(&pwq->pwqs_node, &wq->pwqs); |
30cdf249 TH |
3144 | } |
3145 | } else { | |
3146 | struct pool_workqueue *pwq; | |
3147 | ||
3148 | pwq = kmem_cache_zalloc(pwq_cache, GFP_KERNEL); | |
3149 | if (!pwq) | |
3150 | return -ENOMEM; | |
3151 | ||
49e3cf44 | 3152 | pwq->pool = get_std_worker_pool(WORK_CPU_UNBOUND, highpri); |
76af4d93 | 3153 | list_add_tail_rcu(&pwq->pwqs_node, &wq->pwqs); |
30cdf249 TH |
3154 | } |
3155 | ||
3156 | return 0; | |
0f900049 TH |
3157 | } |
3158 | ||
112202d9 | 3159 | static void free_pwqs(struct workqueue_struct *wq) |
0f900049 | 3160 | { |
e06ffa1e | 3161 | if (!(wq->flags & WQ_UNBOUND)) |
420c0ddb TH |
3162 | free_percpu(wq->cpu_pwqs); |
3163 | else if (!list_empty(&wq->pwqs)) | |
3164 | kmem_cache_free(pwq_cache, list_first_entry(&wq->pwqs, | |
3165 | struct pool_workqueue, pwqs_node)); | |
0f900049 TH |
3166 | } |
3167 | ||
f3421797 TH |
3168 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
3169 | const char *name) | |
b71ab8c2 | 3170 | { |
f3421797 TH |
3171 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
3172 | ||
3173 | if (max_active < 1 || max_active > lim) | |
044c782c VI |
3174 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
3175 | max_active, name, 1, lim); | |
b71ab8c2 | 3176 | |
f3421797 | 3177 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
3178 | } |
3179 | ||
b196be89 | 3180 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
d320c038 TH |
3181 | unsigned int flags, |
3182 | int max_active, | |
3183 | struct lock_class_key *key, | |
b196be89 | 3184 | const char *lock_name, ...) |
1da177e4 | 3185 | { |
b196be89 | 3186 | va_list args, args1; |
1da177e4 | 3187 | struct workqueue_struct *wq; |
49e3cf44 | 3188 | struct pool_workqueue *pwq; |
b196be89 TH |
3189 | size_t namelen; |
3190 | ||
3191 | /* determine namelen, allocate wq and format name */ | |
3192 | va_start(args, lock_name); | |
3193 | va_copy(args1, args); | |
3194 | namelen = vsnprintf(NULL, 0, fmt, args) + 1; | |
3195 | ||
3196 | wq = kzalloc(sizeof(*wq) + namelen, GFP_KERNEL); | |
3197 | if (!wq) | |
3198 | goto err; | |
3199 | ||
3200 | vsnprintf(wq->name, namelen, fmt, args1); | |
3201 | va_end(args); | |
3202 | va_end(args1); | |
1da177e4 | 3203 | |
6370a6ad TH |
3204 | /* |
3205 | * Workqueues which may be used during memory reclaim should | |
3206 | * have a rescuer to guarantee forward progress. | |
3207 | */ | |
3208 | if (flags & WQ_MEM_RECLAIM) | |
3209 | flags |= WQ_RESCUER; | |
3210 | ||
d320c038 | 3211 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 3212 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 3213 | |
b196be89 | 3214 | /* init wq */ |
97e37d7b | 3215 | wq->flags = flags; |
a0a1a5fd | 3216 | wq->saved_max_active = max_active; |
73f53c4a | 3217 | mutex_init(&wq->flush_mutex); |
112202d9 | 3218 | atomic_set(&wq->nr_pwqs_to_flush, 0); |
30cdf249 | 3219 | INIT_LIST_HEAD(&wq->pwqs); |
73f53c4a TH |
3220 | INIT_LIST_HEAD(&wq->flusher_queue); |
3221 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
493a1724 | 3222 | INIT_LIST_HEAD(&wq->maydays); |
502ca9d8 | 3223 | |
eb13ba87 | 3224 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 3225 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 3226 | |
30cdf249 | 3227 | if (alloc_and_link_pwqs(wq) < 0) |
bdbc5dd7 TH |
3228 | goto err; |
3229 | ||
76af4d93 | 3230 | local_irq_disable(); |
49e3cf44 | 3231 | for_each_pwq(pwq, wq) { |
112202d9 | 3232 | BUG_ON((unsigned long)pwq & WORK_STRUCT_FLAG_MASK); |
112202d9 TH |
3233 | pwq->wq = wq; |
3234 | pwq->flush_color = -1; | |
3235 | pwq->max_active = max_active; | |
3236 | INIT_LIST_HEAD(&pwq->delayed_works); | |
493a1724 | 3237 | INIT_LIST_HEAD(&pwq->mayday_node); |
e22bee78 | 3238 | } |
76af4d93 | 3239 | local_irq_enable(); |
1537663f | 3240 | |
e22bee78 TH |
3241 | if (flags & WQ_RESCUER) { |
3242 | struct worker *rescuer; | |
3243 | ||
e22bee78 TH |
3244 | wq->rescuer = rescuer = alloc_worker(); |
3245 | if (!rescuer) | |
3246 | goto err; | |
3247 | ||
111c225a TH |
3248 | rescuer->rescue_wq = wq; |
3249 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", | |
b196be89 | 3250 | wq->name); |
e22bee78 TH |
3251 | if (IS_ERR(rescuer->task)) |
3252 | goto err; | |
3253 | ||
e22bee78 TH |
3254 | rescuer->task->flags |= PF_THREAD_BOUND; |
3255 | wake_up_process(rescuer->task); | |
3af24433 ON |
3256 | } |
3257 | ||
a0a1a5fd TH |
3258 | /* |
3259 | * workqueue_lock protects global freeze state and workqueues | |
3260 | * list. Grab it, set max_active accordingly and add the new | |
3261 | * workqueue to workqueues list. | |
3262 | */ | |
e98d5b16 | 3263 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3264 | |
58a69cb4 | 3265 | if (workqueue_freezing && wq->flags & WQ_FREEZABLE) |
49e3cf44 TH |
3266 | for_each_pwq(pwq, wq) |
3267 | pwq->max_active = 0; | |
a0a1a5fd | 3268 | |
1537663f | 3269 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 3270 | |
e98d5b16 | 3271 | spin_unlock_irq(&workqueue_lock); |
1537663f | 3272 | |
3af24433 | 3273 | return wq; |
4690c4ab TH |
3274 | err: |
3275 | if (wq) { | |
112202d9 | 3276 | free_pwqs(wq); |
e22bee78 | 3277 | kfree(wq->rescuer); |
4690c4ab TH |
3278 | kfree(wq); |
3279 | } | |
3280 | return NULL; | |
3af24433 | 3281 | } |
d320c038 | 3282 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 3283 | |
3af24433 ON |
3284 | /** |
3285 | * destroy_workqueue - safely terminate a workqueue | |
3286 | * @wq: target workqueue | |
3287 | * | |
3288 | * Safely destroy a workqueue. All work currently pending will be done first. | |
3289 | */ | |
3290 | void destroy_workqueue(struct workqueue_struct *wq) | |
3291 | { | |
49e3cf44 | 3292 | struct pool_workqueue *pwq; |
3af24433 | 3293 | |
9c5a2ba7 TH |
3294 | /* drain it before proceeding with destruction */ |
3295 | drain_workqueue(wq); | |
c8efcc25 | 3296 | |
76af4d93 TH |
3297 | spin_lock_irq(&workqueue_lock); |
3298 | ||
6183c009 | 3299 | /* sanity checks */ |
49e3cf44 | 3300 | for_each_pwq(pwq, wq) { |
6183c009 TH |
3301 | int i; |
3302 | ||
76af4d93 TH |
3303 | for (i = 0; i < WORK_NR_COLORS; i++) { |
3304 | if (WARN_ON(pwq->nr_in_flight[i])) { | |
3305 | spin_unlock_irq(&workqueue_lock); | |
6183c009 | 3306 | return; |
76af4d93 TH |
3307 | } |
3308 | } | |
3309 | ||
6183c009 | 3310 | if (WARN_ON(pwq->nr_active) || |
76af4d93 TH |
3311 | WARN_ON(!list_empty(&pwq->delayed_works))) { |
3312 | spin_unlock_irq(&workqueue_lock); | |
6183c009 | 3313 | return; |
76af4d93 | 3314 | } |
6183c009 TH |
3315 | } |
3316 | ||
a0a1a5fd TH |
3317 | /* |
3318 | * wq list is used to freeze wq, remove from list after | |
3319 | * flushing is complete in case freeze races us. | |
3320 | */ | |
b1f4ec17 | 3321 | list_del(&wq->list); |
76af4d93 | 3322 | |
e98d5b16 | 3323 | spin_unlock_irq(&workqueue_lock); |
3af24433 | 3324 | |
e22bee78 TH |
3325 | if (wq->flags & WQ_RESCUER) { |
3326 | kthread_stop(wq->rescuer->task); | |
8d9df9f0 | 3327 | kfree(wq->rescuer); |
e22bee78 TH |
3328 | } |
3329 | ||
112202d9 | 3330 | free_pwqs(wq); |
3af24433 ON |
3331 | kfree(wq); |
3332 | } | |
3333 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
3334 | ||
9f4bd4cd | 3335 | /** |
112202d9 TH |
3336 | * pwq_set_max_active - adjust max_active of a pwq |
3337 | * @pwq: target pool_workqueue | |
9f4bd4cd LJ |
3338 | * @max_active: new max_active value. |
3339 | * | |
112202d9 | 3340 | * Set @pwq->max_active to @max_active and activate delayed works if |
9f4bd4cd LJ |
3341 | * increased. |
3342 | * | |
3343 | * CONTEXT: | |
d565ed63 | 3344 | * spin_lock_irq(pool->lock). |
9f4bd4cd | 3345 | */ |
112202d9 | 3346 | static void pwq_set_max_active(struct pool_workqueue *pwq, int max_active) |
9f4bd4cd | 3347 | { |
112202d9 | 3348 | pwq->max_active = max_active; |
9f4bd4cd | 3349 | |
112202d9 TH |
3350 | while (!list_empty(&pwq->delayed_works) && |
3351 | pwq->nr_active < pwq->max_active) | |
3352 | pwq_activate_first_delayed(pwq); | |
9f4bd4cd LJ |
3353 | } |
3354 | ||
dcd989cb TH |
3355 | /** |
3356 | * workqueue_set_max_active - adjust max_active of a workqueue | |
3357 | * @wq: target workqueue | |
3358 | * @max_active: new max_active value. | |
3359 | * | |
3360 | * Set max_active of @wq to @max_active. | |
3361 | * | |
3362 | * CONTEXT: | |
3363 | * Don't call from IRQ context. | |
3364 | */ | |
3365 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
3366 | { | |
49e3cf44 | 3367 | struct pool_workqueue *pwq; |
dcd989cb | 3368 | |
f3421797 | 3369 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb | 3370 | |
e98d5b16 | 3371 | spin_lock_irq(&workqueue_lock); |
dcd989cb TH |
3372 | |
3373 | wq->saved_max_active = max_active; | |
3374 | ||
49e3cf44 | 3375 | for_each_pwq(pwq, wq) { |
112202d9 | 3376 | struct worker_pool *pool = pwq->pool; |
dcd989cb | 3377 | |
e98d5b16 | 3378 | spin_lock(&pool->lock); |
dcd989cb | 3379 | |
58a69cb4 | 3380 | if (!(wq->flags & WQ_FREEZABLE) || |
35b6bb63 | 3381 | !(pool->flags & POOL_FREEZING)) |
112202d9 | 3382 | pwq_set_max_active(pwq, max_active); |
9bfb1839 | 3383 | |
e98d5b16 | 3384 | spin_unlock(&pool->lock); |
65a64464 | 3385 | } |
93981800 | 3386 | |
e98d5b16 | 3387 | spin_unlock_irq(&workqueue_lock); |
15316ba8 | 3388 | } |
dcd989cb | 3389 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 3390 | |
eef6a7d5 | 3391 | /** |
dcd989cb TH |
3392 | * workqueue_congested - test whether a workqueue is congested |
3393 | * @cpu: CPU in question | |
3394 | * @wq: target workqueue | |
eef6a7d5 | 3395 | * |
dcd989cb TH |
3396 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
3397 | * no synchronization around this function and the test result is | |
3398 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 3399 | * |
dcd989cb TH |
3400 | * RETURNS: |
3401 | * %true if congested, %false otherwise. | |
eef6a7d5 | 3402 | */ |
d84ff051 | 3403 | bool workqueue_congested(int cpu, struct workqueue_struct *wq) |
1da177e4 | 3404 | { |
7fb98ea7 | 3405 | struct pool_workqueue *pwq; |
76af4d93 TH |
3406 | bool ret; |
3407 | ||
3408 | preempt_disable(); | |
7fb98ea7 TH |
3409 | |
3410 | if (!(wq->flags & WQ_UNBOUND)) | |
3411 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); | |
3412 | else | |
3413 | pwq = first_pwq(wq); | |
dcd989cb | 3414 | |
76af4d93 TH |
3415 | ret = !list_empty(&pwq->delayed_works); |
3416 | preempt_enable(); | |
3417 | ||
3418 | return ret; | |
1da177e4 | 3419 | } |
dcd989cb | 3420 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 3421 | |
dcd989cb TH |
3422 | /** |
3423 | * work_busy - test whether a work is currently pending or running | |
3424 | * @work: the work to be tested | |
3425 | * | |
3426 | * Test whether @work is currently pending or running. There is no | |
3427 | * synchronization around this function and the test result is | |
3428 | * unreliable and only useful as advisory hints or for debugging. | |
dcd989cb TH |
3429 | * |
3430 | * RETURNS: | |
3431 | * OR'd bitmask of WORK_BUSY_* bits. | |
3432 | */ | |
3433 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 3434 | { |
fa1b54e6 | 3435 | struct worker_pool *pool; |
dcd989cb TH |
3436 | unsigned long flags; |
3437 | unsigned int ret = 0; | |
1da177e4 | 3438 | |
dcd989cb TH |
3439 | if (work_pending(work)) |
3440 | ret |= WORK_BUSY_PENDING; | |
1da177e4 | 3441 | |
fa1b54e6 TH |
3442 | local_irq_save(flags); |
3443 | pool = get_work_pool(work); | |
038366c5 | 3444 | if (pool) { |
fa1b54e6 | 3445 | spin_lock(&pool->lock); |
038366c5 LJ |
3446 | if (find_worker_executing_work(pool, work)) |
3447 | ret |= WORK_BUSY_RUNNING; | |
fa1b54e6 | 3448 | spin_unlock(&pool->lock); |
038366c5 | 3449 | } |
fa1b54e6 | 3450 | local_irq_restore(flags); |
1da177e4 | 3451 | |
dcd989cb | 3452 | return ret; |
1da177e4 | 3453 | } |
dcd989cb | 3454 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 3455 | |
db7bccf4 TH |
3456 | /* |
3457 | * CPU hotplug. | |
3458 | * | |
e22bee78 | 3459 | * There are two challenges in supporting CPU hotplug. Firstly, there |
112202d9 | 3460 | * are a lot of assumptions on strong associations among work, pwq and |
706026c2 | 3461 | * pool which make migrating pending and scheduled works very |
e22bee78 | 3462 | * difficult to implement without impacting hot paths. Secondly, |
94cf58bb | 3463 | * worker pools serve mix of short, long and very long running works making |
e22bee78 TH |
3464 | * blocked draining impractical. |
3465 | * | |
24647570 | 3466 | * This is solved by allowing the pools to be disassociated from the CPU |
628c78e7 TH |
3467 | * running as an unbound one and allowing it to be reattached later if the |
3468 | * cpu comes back online. | |
db7bccf4 | 3469 | */ |
1da177e4 | 3470 | |
706026c2 | 3471 | static void wq_unbind_fn(struct work_struct *work) |
3af24433 | 3472 | { |
38db41d9 | 3473 | int cpu = smp_processor_id(); |
4ce62e9e | 3474 | struct worker_pool *pool; |
db7bccf4 | 3475 | struct worker *worker; |
db7bccf4 | 3476 | int i; |
3af24433 | 3477 | |
38db41d9 | 3478 | for_each_std_worker_pool(pool, cpu) { |
6183c009 | 3479 | WARN_ON_ONCE(cpu != smp_processor_id()); |
db7bccf4 | 3480 | |
94cf58bb TH |
3481 | mutex_lock(&pool->assoc_mutex); |
3482 | spin_lock_irq(&pool->lock); | |
3af24433 | 3483 | |
94cf58bb TH |
3484 | /* |
3485 | * We've claimed all manager positions. Make all workers | |
3486 | * unbound and set DISASSOCIATED. Before this, all workers | |
3487 | * except for the ones which are still executing works from | |
3488 | * before the last CPU down must be on the cpu. After | |
3489 | * this, they may become diasporas. | |
3490 | */ | |
4ce62e9e | 3491 | list_for_each_entry(worker, &pool->idle_list, entry) |
403c821d | 3492 | worker->flags |= WORKER_UNBOUND; |
3af24433 | 3493 | |
b67bfe0d | 3494 | for_each_busy_worker(worker, i, pool) |
c9e7cf27 | 3495 | worker->flags |= WORKER_UNBOUND; |
06ba38a9 | 3496 | |
24647570 | 3497 | pool->flags |= POOL_DISASSOCIATED; |
f2d5a0ee | 3498 | |
94cf58bb TH |
3499 | spin_unlock_irq(&pool->lock); |
3500 | mutex_unlock(&pool->assoc_mutex); | |
3501 | } | |
628c78e7 | 3502 | |
e22bee78 | 3503 | /* |
403c821d | 3504 | * Call schedule() so that we cross rq->lock and thus can guarantee |
628c78e7 TH |
3505 | * sched callbacks see the %WORKER_UNBOUND flag. This is necessary |
3506 | * as scheduler callbacks may be invoked from other cpus. | |
e22bee78 | 3507 | */ |
e22bee78 | 3508 | schedule(); |
06ba38a9 | 3509 | |
e22bee78 | 3510 | /* |
628c78e7 TH |
3511 | * Sched callbacks are disabled now. Zap nr_running. After this, |
3512 | * nr_running stays zero and need_more_worker() and keep_working() | |
38db41d9 TH |
3513 | * are always true as long as the worklist is not empty. Pools on |
3514 | * @cpu now behave as unbound (in terms of concurrency management) | |
3515 | * pools which are served by workers tied to the CPU. | |
628c78e7 TH |
3516 | * |
3517 | * On return from this function, the current worker would trigger | |
3518 | * unbound chain execution of pending work items if other workers | |
3519 | * didn't already. | |
e22bee78 | 3520 | */ |
38db41d9 | 3521 | for_each_std_worker_pool(pool, cpu) |
e19e397a | 3522 | atomic_set(&pool->nr_running, 0); |
3af24433 | 3523 | } |
3af24433 | 3524 | |
8db25e78 TH |
3525 | /* |
3526 | * Workqueues should be brought up before normal priority CPU notifiers. | |
3527 | * This will be registered high priority CPU notifier. | |
3528 | */ | |
9fdf9b73 | 3529 | static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb, |
8db25e78 TH |
3530 | unsigned long action, |
3531 | void *hcpu) | |
3af24433 | 3532 | { |
d84ff051 | 3533 | int cpu = (unsigned long)hcpu; |
4ce62e9e | 3534 | struct worker_pool *pool; |
3ce63377 | 3535 | |
8db25e78 | 3536 | switch (action & ~CPU_TASKS_FROZEN) { |
3af24433 | 3537 | case CPU_UP_PREPARE: |
38db41d9 | 3538 | for_each_std_worker_pool(pool, cpu) { |
3ce63377 TH |
3539 | struct worker *worker; |
3540 | ||
3541 | if (pool->nr_workers) | |
3542 | continue; | |
3543 | ||
3544 | worker = create_worker(pool); | |
3545 | if (!worker) | |
3546 | return NOTIFY_BAD; | |
3547 | ||
d565ed63 | 3548 | spin_lock_irq(&pool->lock); |
3ce63377 | 3549 | start_worker(worker); |
d565ed63 | 3550 | spin_unlock_irq(&pool->lock); |
3af24433 | 3551 | } |
8db25e78 | 3552 | break; |
3af24433 | 3553 | |
db7bccf4 TH |
3554 | case CPU_DOWN_FAILED: |
3555 | case CPU_ONLINE: | |
38db41d9 | 3556 | for_each_std_worker_pool(pool, cpu) { |
94cf58bb TH |
3557 | mutex_lock(&pool->assoc_mutex); |
3558 | spin_lock_irq(&pool->lock); | |
3559 | ||
24647570 | 3560 | pool->flags &= ~POOL_DISASSOCIATED; |
94cf58bb TH |
3561 | rebind_workers(pool); |
3562 | ||
3563 | spin_unlock_irq(&pool->lock); | |
3564 | mutex_unlock(&pool->assoc_mutex); | |
3565 | } | |
db7bccf4 | 3566 | break; |
00dfcaf7 | 3567 | } |
65758202 TH |
3568 | return NOTIFY_OK; |
3569 | } | |
3570 | ||
3571 | /* | |
3572 | * Workqueues should be brought down after normal priority CPU notifiers. | |
3573 | * This will be registered as low priority CPU notifier. | |
3574 | */ | |
9fdf9b73 | 3575 | static int __cpuinit workqueue_cpu_down_callback(struct notifier_block *nfb, |
65758202 TH |
3576 | unsigned long action, |
3577 | void *hcpu) | |
3578 | { | |
d84ff051 | 3579 | int cpu = (unsigned long)hcpu; |
8db25e78 TH |
3580 | struct work_struct unbind_work; |
3581 | ||
65758202 TH |
3582 | switch (action & ~CPU_TASKS_FROZEN) { |
3583 | case CPU_DOWN_PREPARE: | |
8db25e78 | 3584 | /* unbinding should happen on the local CPU */ |
706026c2 | 3585 | INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn); |
7635d2fd | 3586 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
8db25e78 TH |
3587 | flush_work(&unbind_work); |
3588 | break; | |
65758202 TH |
3589 | } |
3590 | return NOTIFY_OK; | |
3591 | } | |
3592 | ||
2d3854a3 | 3593 | #ifdef CONFIG_SMP |
8ccad40d | 3594 | |
2d3854a3 | 3595 | struct work_for_cpu { |
ed48ece2 | 3596 | struct work_struct work; |
2d3854a3 RR |
3597 | long (*fn)(void *); |
3598 | void *arg; | |
3599 | long ret; | |
3600 | }; | |
3601 | ||
ed48ece2 | 3602 | static void work_for_cpu_fn(struct work_struct *work) |
2d3854a3 | 3603 | { |
ed48ece2 TH |
3604 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
3605 | ||
2d3854a3 RR |
3606 | wfc->ret = wfc->fn(wfc->arg); |
3607 | } | |
3608 | ||
3609 | /** | |
3610 | * work_on_cpu - run a function in user context on a particular cpu | |
3611 | * @cpu: the cpu to run on | |
3612 | * @fn: the function to run | |
3613 | * @arg: the function arg | |
3614 | * | |
31ad9081 RR |
3615 | * This will return the value @fn returns. |
3616 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 3617 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 | 3618 | */ |
d84ff051 | 3619 | long work_on_cpu(int cpu, long (*fn)(void *), void *arg) |
2d3854a3 | 3620 | { |
ed48ece2 | 3621 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; |
6b44003e | 3622 | |
ed48ece2 TH |
3623 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
3624 | schedule_work_on(cpu, &wfc.work); | |
3625 | flush_work(&wfc.work); | |
2d3854a3 RR |
3626 | return wfc.ret; |
3627 | } | |
3628 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
3629 | #endif /* CONFIG_SMP */ | |
3630 | ||
a0a1a5fd TH |
3631 | #ifdef CONFIG_FREEZER |
3632 | ||
3633 | /** | |
3634 | * freeze_workqueues_begin - begin freezing workqueues | |
3635 | * | |
58a69cb4 TH |
3636 | * Start freezing workqueues. After this function returns, all freezable |
3637 | * workqueues will queue new works to their frozen_works list instead of | |
706026c2 | 3638 | * pool->worklist. |
a0a1a5fd TH |
3639 | * |
3640 | * CONTEXT: | |
d565ed63 | 3641 | * Grabs and releases workqueue_lock and pool->lock's. |
a0a1a5fd TH |
3642 | */ |
3643 | void freeze_workqueues_begin(void) | |
3644 | { | |
17116969 | 3645 | struct worker_pool *pool; |
24b8a847 TH |
3646 | struct workqueue_struct *wq; |
3647 | struct pool_workqueue *pwq; | |
17116969 | 3648 | int id; |
a0a1a5fd | 3649 | |
e98d5b16 | 3650 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3651 | |
6183c009 | 3652 | WARN_ON_ONCE(workqueue_freezing); |
a0a1a5fd TH |
3653 | workqueue_freezing = true; |
3654 | ||
24b8a847 | 3655 | /* set FREEZING */ |
17116969 | 3656 | for_each_pool(pool, id) { |
17116969 | 3657 | spin_lock(&pool->lock); |
17116969 TH |
3658 | WARN_ON_ONCE(pool->flags & POOL_FREEZING); |
3659 | pool->flags |= POOL_FREEZING; | |
24b8a847 TH |
3660 | spin_unlock(&pool->lock); |
3661 | } | |
a0a1a5fd | 3662 | |
24b8a847 TH |
3663 | /* suppress further executions by setting max_active to zero */ |
3664 | list_for_each_entry(wq, &workqueues, list) { | |
3665 | if (!(wq->flags & WQ_FREEZABLE)) | |
3666 | continue; | |
8b03ae3c | 3667 | |
24b8a847 TH |
3668 | for_each_pwq(pwq, wq) { |
3669 | spin_lock(&pwq->pool->lock); | |
3670 | pwq->max_active = 0; | |
3671 | spin_unlock(&pwq->pool->lock); | |
a1056305 | 3672 | } |
a0a1a5fd TH |
3673 | } |
3674 | ||
e98d5b16 | 3675 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3676 | } |
3677 | ||
3678 | /** | |
58a69cb4 | 3679 | * freeze_workqueues_busy - are freezable workqueues still busy? |
a0a1a5fd TH |
3680 | * |
3681 | * Check whether freezing is complete. This function must be called | |
3682 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
3683 | * | |
3684 | * CONTEXT: | |
3685 | * Grabs and releases workqueue_lock. | |
3686 | * | |
3687 | * RETURNS: | |
58a69cb4 TH |
3688 | * %true if some freezable workqueues are still busy. %false if freezing |
3689 | * is complete. | |
a0a1a5fd TH |
3690 | */ |
3691 | bool freeze_workqueues_busy(void) | |
3692 | { | |
a0a1a5fd | 3693 | bool busy = false; |
24b8a847 TH |
3694 | struct workqueue_struct *wq; |
3695 | struct pool_workqueue *pwq; | |
a0a1a5fd | 3696 | |
e98d5b16 | 3697 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3698 | |
6183c009 | 3699 | WARN_ON_ONCE(!workqueue_freezing); |
a0a1a5fd | 3700 | |
24b8a847 TH |
3701 | list_for_each_entry(wq, &workqueues, list) { |
3702 | if (!(wq->flags & WQ_FREEZABLE)) | |
3703 | continue; | |
a0a1a5fd TH |
3704 | /* |
3705 | * nr_active is monotonically decreasing. It's safe | |
3706 | * to peek without lock. | |
3707 | */ | |
24b8a847 | 3708 | for_each_pwq(pwq, wq) { |
6183c009 | 3709 | WARN_ON_ONCE(pwq->nr_active < 0); |
112202d9 | 3710 | if (pwq->nr_active) { |
a0a1a5fd TH |
3711 | busy = true; |
3712 | goto out_unlock; | |
3713 | } | |
3714 | } | |
3715 | } | |
3716 | out_unlock: | |
e98d5b16 | 3717 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3718 | return busy; |
3719 | } | |
3720 | ||
3721 | /** | |
3722 | * thaw_workqueues - thaw workqueues | |
3723 | * | |
3724 | * Thaw workqueues. Normal queueing is restored and all collected | |
706026c2 | 3725 | * frozen works are transferred to their respective pool worklists. |
a0a1a5fd TH |
3726 | * |
3727 | * CONTEXT: | |
d565ed63 | 3728 | * Grabs and releases workqueue_lock and pool->lock's. |
a0a1a5fd TH |
3729 | */ |
3730 | void thaw_workqueues(void) | |
3731 | { | |
24b8a847 TH |
3732 | struct workqueue_struct *wq; |
3733 | struct pool_workqueue *pwq; | |
3734 | struct worker_pool *pool; | |
3735 | int id; | |
a0a1a5fd | 3736 | |
e98d5b16 | 3737 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd TH |
3738 | |
3739 | if (!workqueue_freezing) | |
3740 | goto out_unlock; | |
3741 | ||
24b8a847 TH |
3742 | /* clear FREEZING */ |
3743 | for_each_pool(pool, id) { | |
3744 | spin_lock(&pool->lock); | |
3745 | WARN_ON_ONCE(!(pool->flags & POOL_FREEZING)); | |
3746 | pool->flags &= ~POOL_FREEZING; | |
3747 | spin_unlock(&pool->lock); | |
3748 | } | |
8b03ae3c | 3749 | |
24b8a847 TH |
3750 | /* restore max_active and repopulate worklist */ |
3751 | list_for_each_entry(wq, &workqueues, list) { | |
3752 | if (!(wq->flags & WQ_FREEZABLE)) | |
3753 | continue; | |
a1056305 | 3754 | |
24b8a847 TH |
3755 | for_each_pwq(pwq, wq) { |
3756 | spin_lock(&pwq->pool->lock); | |
3757 | pwq_set_max_active(pwq, wq->saved_max_active); | |
3758 | spin_unlock(&pwq->pool->lock); | |
d565ed63 | 3759 | } |
a0a1a5fd TH |
3760 | } |
3761 | ||
24b8a847 TH |
3762 | /* kick workers */ |
3763 | for_each_pool(pool, id) { | |
3764 | spin_lock(&pool->lock); | |
3765 | wake_up_worker(pool); | |
3766 | spin_unlock(&pool->lock); | |
3767 | } | |
3768 | ||
a0a1a5fd TH |
3769 | workqueue_freezing = false; |
3770 | out_unlock: | |
e98d5b16 | 3771 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3772 | } |
3773 | #endif /* CONFIG_FREEZER */ | |
3774 | ||
6ee0578b | 3775 | static int __init init_workqueues(void) |
1da177e4 | 3776 | { |
d84ff051 | 3777 | int cpu; |
c34056a3 | 3778 | |
7c3eed5c TH |
3779 | /* make sure we have enough bits for OFFQ pool ID */ |
3780 | BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_POOL_SHIFT)) < | |
6be19588 | 3781 | WORK_CPU_END * NR_STD_WORKER_POOLS); |
b5490077 | 3782 | |
e904e6c2 TH |
3783 | WARN_ON(__alignof__(struct pool_workqueue) < __alignof__(long long)); |
3784 | ||
3785 | pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC); | |
3786 | ||
65758202 | 3787 | cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); |
a5b4e57d | 3788 | hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); |
8b03ae3c | 3789 | |
706026c2 TH |
3790 | /* initialize CPU pools */ |
3791 | for_each_wq_cpu(cpu) { | |
4ce62e9e | 3792 | struct worker_pool *pool; |
8b03ae3c | 3793 | |
38db41d9 | 3794 | for_each_std_worker_pool(pool, cpu) { |
d565ed63 | 3795 | spin_lock_init(&pool->lock); |
ec22ca5e | 3796 | pool->cpu = cpu; |
24647570 | 3797 | pool->flags |= POOL_DISASSOCIATED; |
4ce62e9e TH |
3798 | INIT_LIST_HEAD(&pool->worklist); |
3799 | INIT_LIST_HEAD(&pool->idle_list); | |
c9e7cf27 | 3800 | hash_init(pool->busy_hash); |
e7577c50 | 3801 | |
4ce62e9e TH |
3802 | init_timer_deferrable(&pool->idle_timer); |
3803 | pool->idle_timer.function = idle_worker_timeout; | |
3804 | pool->idle_timer.data = (unsigned long)pool; | |
e22bee78 | 3805 | |
706026c2 | 3806 | setup_timer(&pool->mayday_timer, pool_mayday_timeout, |
4ce62e9e TH |
3807 | (unsigned long)pool); |
3808 | ||
b2eb83d1 | 3809 | mutex_init(&pool->assoc_mutex); |
4ce62e9e | 3810 | ida_init(&pool->worker_ida); |
9daf9e67 TH |
3811 | |
3812 | /* alloc pool ID */ | |
3813 | BUG_ON(worker_pool_assign_id(pool)); | |
4ce62e9e | 3814 | } |
8b03ae3c TH |
3815 | } |
3816 | ||
e22bee78 | 3817 | /* create the initial worker */ |
706026c2 | 3818 | for_each_online_wq_cpu(cpu) { |
4ce62e9e | 3819 | struct worker_pool *pool; |
e22bee78 | 3820 | |
38db41d9 | 3821 | for_each_std_worker_pool(pool, cpu) { |
4ce62e9e TH |
3822 | struct worker *worker; |
3823 | ||
24647570 TH |
3824 | if (cpu != WORK_CPU_UNBOUND) |
3825 | pool->flags &= ~POOL_DISASSOCIATED; | |
3826 | ||
bc2ae0f5 | 3827 | worker = create_worker(pool); |
4ce62e9e | 3828 | BUG_ON(!worker); |
d565ed63 | 3829 | spin_lock_irq(&pool->lock); |
4ce62e9e | 3830 | start_worker(worker); |
d565ed63 | 3831 | spin_unlock_irq(&pool->lock); |
4ce62e9e | 3832 | } |
e22bee78 TH |
3833 | } |
3834 | ||
d320c038 | 3835 | system_wq = alloc_workqueue("events", 0, 0); |
1aabe902 | 3836 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
d320c038 | 3837 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
f3421797 TH |
3838 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
3839 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
3840 | system_freezable_wq = alloc_workqueue("events_freezable", |
3841 | WQ_FREEZABLE, 0); | |
1aabe902 | 3842 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
ae930e0f | 3843 | !system_unbound_wq || !system_freezable_wq); |
6ee0578b | 3844 | return 0; |
1da177e4 | 3845 | } |
6ee0578b | 3846 | early_initcall(init_workqueues); |