]>
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> |
29c91e99 | 44 | #include <linux/jhash.h> |
42f8570f | 45 | #include <linux/hashtable.h> |
76af4d93 | 46 | #include <linux/rculist.h> |
e22bee78 | 47 | |
ea138446 | 48 | #include "workqueue_internal.h" |
1da177e4 | 49 | |
c8e55f36 | 50 | enum { |
24647570 TH |
51 | /* |
52 | * worker_pool flags | |
bc2ae0f5 | 53 | * |
24647570 | 54 | * A bound pool is either associated or disassociated with its CPU. |
bc2ae0f5 TH |
55 | * While associated (!DISASSOCIATED), all workers are bound to the |
56 | * CPU and none has %WORKER_UNBOUND set and concurrency management | |
57 | * is in effect. | |
58 | * | |
59 | * While DISASSOCIATED, the cpu may be offline and all workers have | |
60 | * %WORKER_UNBOUND set and concurrency management disabled, and may | |
24647570 | 61 | * be executing on any CPU. The pool behaves as an unbound one. |
bc2ae0f5 TH |
62 | * |
63 | * Note that DISASSOCIATED can be flipped only while holding | |
24647570 TH |
64 | * assoc_mutex to avoid changing binding state while |
65 | * create_worker() is in progress. | |
bc2ae0f5 | 66 | */ |
11ebea50 | 67 | POOL_MANAGE_WORKERS = 1 << 0, /* need to manage 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 | |
29c91e99 | 84 | UNBOUND_POOL_HASH_ORDER = 6, /* hashed by pool->attrs */ |
c8e55f36 | 85 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ |
db7bccf4 | 86 | |
e22bee78 TH |
87 | MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */ |
88 | IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */ | |
89 | ||
3233cdbd TH |
90 | MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2, |
91 | /* call for help after 10ms | |
92 | (min two ticks) */ | |
e22bee78 TH |
93 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ |
94 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
e22bee78 TH |
95 | |
96 | /* | |
97 | * Rescue workers are used only on emergencies and shared by | |
98 | * all cpus. Give -20. | |
99 | */ | |
100 | RESCUER_NICE_LEVEL = -20, | |
3270476a | 101 | HIGHPRI_NICE_LEVEL = -20, |
c8e55f36 | 102 | }; |
1da177e4 LT |
103 | |
104 | /* | |
4690c4ab TH |
105 | * Structure fields follow one of the following exclusion rules. |
106 | * | |
e41e704b TH |
107 | * I: Modifiable by initialization/destruction paths and read-only for |
108 | * everyone else. | |
4690c4ab | 109 | * |
e22bee78 TH |
110 | * P: Preemption protected. Disabling preemption is enough and should |
111 | * only be modified and accessed from the local cpu. | |
112 | * | |
d565ed63 | 113 | * L: pool->lock protected. Access with pool->lock held. |
4690c4ab | 114 | * |
d565ed63 TH |
115 | * X: During normal operation, modification requires pool->lock and should |
116 | * be done only from local cpu. Either disabling preemption on local | |
117 | * cpu or grabbing pool->lock is enough for read access. If | |
118 | * POOL_DISASSOCIATED is set, it's identical to L. | |
e22bee78 | 119 | * |
73f53c4a TH |
120 | * F: wq->flush_mutex protected. |
121 | * | |
4690c4ab | 122 | * W: workqueue_lock protected. |
76af4d93 TH |
123 | * |
124 | * R: workqueue_lock protected for writes. Sched-RCU protected for reads. | |
1da177e4 | 125 | */ |
1da177e4 | 126 | |
2eaebdb3 | 127 | /* struct worker is defined in workqueue_internal.h */ |
c34056a3 | 128 | |
bd7bdd43 | 129 | struct worker_pool { |
d565ed63 | 130 | spinlock_t lock; /* the pool lock */ |
d84ff051 | 131 | int cpu; /* I: the associated cpu */ |
9daf9e67 | 132 | int id; /* I: pool ID */ |
11ebea50 | 133 | unsigned int flags; /* X: flags */ |
bd7bdd43 TH |
134 | |
135 | struct list_head worklist; /* L: list of pending works */ | |
136 | int nr_workers; /* L: total number of workers */ | |
ea1abd61 LJ |
137 | |
138 | /* nr_idle includes the ones off idle_list for rebinding */ | |
bd7bdd43 TH |
139 | int nr_idle; /* L: currently idle ones */ |
140 | ||
141 | struct list_head idle_list; /* X: list of idle workers */ | |
142 | struct timer_list idle_timer; /* L: worker idle timeout */ | |
143 | struct timer_list mayday_timer; /* L: SOS timer for workers */ | |
144 | ||
c9e7cf27 TH |
145 | /* workers are chained either in busy_hash or idle_list */ |
146 | DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER); | |
147 | /* L: hash of busy workers */ | |
148 | ||
34a06bd6 | 149 | struct mutex manager_arb; /* manager arbitration */ |
24647570 | 150 | struct mutex assoc_mutex; /* protect POOL_DISASSOCIATED */ |
bd7bdd43 | 151 | struct ida worker_ida; /* L: for worker IDs */ |
e19e397a | 152 | |
7a4e344c | 153 | struct workqueue_attrs *attrs; /* I: worker attributes */ |
29c91e99 TH |
154 | struct hlist_node hash_node; /* R: unbound_pool_hash node */ |
155 | int refcnt; /* refcnt for unbound pools */ | |
7a4e344c | 156 | |
e19e397a TH |
157 | /* |
158 | * The current concurrency level. As it's likely to be accessed | |
159 | * from other CPUs during try_to_wake_up(), put it in a separate | |
160 | * cacheline. | |
161 | */ | |
162 | atomic_t nr_running ____cacheline_aligned_in_smp; | |
29c91e99 TH |
163 | |
164 | /* | |
165 | * Destruction of pool is sched-RCU protected to allow dereferences | |
166 | * from get_work_pool(). | |
167 | */ | |
168 | struct rcu_head rcu; | |
8b03ae3c TH |
169 | } ____cacheline_aligned_in_smp; |
170 | ||
1da177e4 | 171 | /* |
112202d9 TH |
172 | * The per-pool workqueue. While queued, the lower WORK_STRUCT_FLAG_BITS |
173 | * of work_struct->data are used for flags and the remaining high bits | |
174 | * point to the pwq; thus, pwqs need to be aligned at two's power of the | |
175 | * number of flag bits. | |
1da177e4 | 176 | */ |
112202d9 | 177 | struct pool_workqueue { |
bd7bdd43 | 178 | struct worker_pool *pool; /* I: the associated pool */ |
4690c4ab | 179 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
180 | int work_color; /* L: current color */ |
181 | int flush_color; /* L: flushing color */ | |
182 | int nr_in_flight[WORK_NR_COLORS]; | |
183 | /* L: nr of in_flight works */ | |
1e19ffc6 | 184 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 185 | int max_active; /* L: max active works */ |
1e19ffc6 | 186 | struct list_head delayed_works; /* L: delayed works */ |
76af4d93 | 187 | struct list_head pwqs_node; /* R: node on wq->pwqs */ |
493a1724 | 188 | struct list_head mayday_node; /* W: node on wq->maydays */ |
e904e6c2 | 189 | } __aligned(1 << WORK_STRUCT_FLAG_BITS); |
1da177e4 | 190 | |
73f53c4a TH |
191 | /* |
192 | * Structure used to wait for workqueue flush. | |
193 | */ | |
194 | struct wq_flusher { | |
195 | struct list_head list; /* F: list of flushers */ | |
196 | int flush_color; /* F: flush color waiting for */ | |
197 | struct completion done; /* flush completion */ | |
198 | }; | |
199 | ||
1da177e4 LT |
200 | /* |
201 | * The externally visible workqueue abstraction is an array of | |
202 | * per-CPU workqueues: | |
203 | */ | |
204 | struct workqueue_struct { | |
9c5a2ba7 | 205 | unsigned int flags; /* W: WQ_* flags */ |
420c0ddb | 206 | struct pool_workqueue __percpu *cpu_pwqs; /* I: per-cpu pwq's */ |
76af4d93 | 207 | struct list_head pwqs; /* R: all pwqs of this wq */ |
4690c4ab | 208 | struct list_head list; /* W: list of all workqueues */ |
73f53c4a TH |
209 | |
210 | struct mutex flush_mutex; /* protects wq flushing */ | |
211 | int work_color; /* F: current work color */ | |
212 | int flush_color; /* F: current flush color */ | |
112202d9 | 213 | atomic_t nr_pwqs_to_flush; /* flush in progress */ |
73f53c4a TH |
214 | struct wq_flusher *first_flusher; /* F: first flusher */ |
215 | struct list_head flusher_queue; /* F: flush waiters */ | |
216 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
217 | ||
493a1724 | 218 | struct list_head maydays; /* W: pwqs requesting rescue */ |
e22bee78 TH |
219 | struct worker *rescuer; /* I: rescue worker */ |
220 | ||
9c5a2ba7 | 221 | int nr_drainers; /* W: drain in progress */ |
112202d9 | 222 | int saved_max_active; /* W: saved pwq max_active */ |
4e6045f1 | 223 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 224 | struct lockdep_map lockdep_map; |
4e6045f1 | 225 | #endif |
b196be89 | 226 | char name[]; /* I: workqueue name */ |
1da177e4 LT |
227 | }; |
228 | ||
e904e6c2 TH |
229 | static struct kmem_cache *pwq_cache; |
230 | ||
29c91e99 TH |
231 | /* hash of all unbound pools keyed by pool->attrs */ |
232 | static DEFINE_HASHTABLE(unbound_pool_hash, UNBOUND_POOL_HASH_ORDER); | |
233 | ||
234 | static struct workqueue_attrs *unbound_std_wq_attrs[NR_STD_WORKER_POOLS]; | |
235 | ||
d320c038 | 236 | struct workqueue_struct *system_wq __read_mostly; |
d320c038 | 237 | EXPORT_SYMBOL_GPL(system_wq); |
044c782c | 238 | struct workqueue_struct *system_highpri_wq __read_mostly; |
1aabe902 | 239 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
044c782c | 240 | struct workqueue_struct *system_long_wq __read_mostly; |
d320c038 | 241 | EXPORT_SYMBOL_GPL(system_long_wq); |
044c782c | 242 | struct workqueue_struct *system_unbound_wq __read_mostly; |
f3421797 | 243 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
044c782c | 244 | struct workqueue_struct *system_freezable_wq __read_mostly; |
24d51add | 245 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
d320c038 | 246 | |
97bd2347 TH |
247 | #define CREATE_TRACE_POINTS |
248 | #include <trace/events/workqueue.h> | |
249 | ||
76af4d93 TH |
250 | #define assert_rcu_or_wq_lock() \ |
251 | rcu_lockdep_assert(rcu_read_lock_sched_held() || \ | |
252 | lockdep_is_held(&workqueue_lock), \ | |
253 | "sched RCU or workqueue lock should be held") | |
254 | ||
38db41d9 | 255 | #define for_each_std_worker_pool(pool, cpu) \ |
7a62c2c8 TH |
256 | for ((pool) = &per_cpu(cpu_std_worker_pools, cpu)[0]; \ |
257 | (pool) < &per_cpu(cpu_std_worker_pools, cpu)[NR_STD_WORKER_POOLS]; \ | |
258 | (pool)++) | |
4ce62e9e | 259 | |
b67bfe0d SL |
260 | #define for_each_busy_worker(worker, i, pool) \ |
261 | hash_for_each(pool->busy_hash, i, worker, hentry) | |
db7bccf4 | 262 | |
17116969 TH |
263 | /** |
264 | * for_each_pool - iterate through all worker_pools in the system | |
265 | * @pool: iteration cursor | |
266 | * @id: integer used for iteration | |
fa1b54e6 TH |
267 | * |
268 | * This must be called either with workqueue_lock held or sched RCU read | |
269 | * locked. If the pool needs to be used beyond the locking in effect, the | |
270 | * caller is responsible for guaranteeing that the pool stays online. | |
271 | * | |
272 | * The if/else clause exists only for the lockdep assertion and can be | |
273 | * ignored. | |
17116969 TH |
274 | */ |
275 | #define for_each_pool(pool, id) \ | |
fa1b54e6 TH |
276 | idr_for_each_entry(&worker_pool_idr, pool, id) \ |
277 | if (({ assert_rcu_or_wq_lock(); false; })) { } \ | |
278 | else | |
17116969 | 279 | |
49e3cf44 TH |
280 | /** |
281 | * for_each_pwq - iterate through all pool_workqueues of the specified workqueue | |
282 | * @pwq: iteration cursor | |
283 | * @wq: the target workqueue | |
76af4d93 TH |
284 | * |
285 | * This must be called either with workqueue_lock held or sched RCU read | |
286 | * locked. If the pwq needs to be used beyond the locking in effect, the | |
287 | * caller is responsible for guaranteeing that the pwq stays online. | |
288 | * | |
289 | * The if/else clause exists only for the lockdep assertion and can be | |
290 | * ignored. | |
49e3cf44 TH |
291 | */ |
292 | #define for_each_pwq(pwq, wq) \ | |
76af4d93 TH |
293 | list_for_each_entry_rcu((pwq), &(wq)->pwqs, pwqs_node) \ |
294 | if (({ assert_rcu_or_wq_lock(); false; })) { } \ | |
295 | else | |
f3421797 | 296 | |
dc186ad7 TG |
297 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
298 | ||
299 | static struct debug_obj_descr work_debug_descr; | |
300 | ||
99777288 SG |
301 | static void *work_debug_hint(void *addr) |
302 | { | |
303 | return ((struct work_struct *) addr)->func; | |
304 | } | |
305 | ||
dc186ad7 TG |
306 | /* |
307 | * fixup_init is called when: | |
308 | * - an active object is initialized | |
309 | */ | |
310 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
311 | { | |
312 | struct work_struct *work = addr; | |
313 | ||
314 | switch (state) { | |
315 | case ODEBUG_STATE_ACTIVE: | |
316 | cancel_work_sync(work); | |
317 | debug_object_init(work, &work_debug_descr); | |
318 | return 1; | |
319 | default: | |
320 | return 0; | |
321 | } | |
322 | } | |
323 | ||
324 | /* | |
325 | * fixup_activate is called when: | |
326 | * - an active object is activated | |
327 | * - an unknown object is activated (might be a statically initialized object) | |
328 | */ | |
329 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
330 | { | |
331 | struct work_struct *work = addr; | |
332 | ||
333 | switch (state) { | |
334 | ||
335 | case ODEBUG_STATE_NOTAVAILABLE: | |
336 | /* | |
337 | * This is not really a fixup. The work struct was | |
338 | * statically initialized. We just make sure that it | |
339 | * is tracked in the object tracker. | |
340 | */ | |
22df02bb | 341 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
342 | debug_object_init(work, &work_debug_descr); |
343 | debug_object_activate(work, &work_debug_descr); | |
344 | return 0; | |
345 | } | |
346 | WARN_ON_ONCE(1); | |
347 | return 0; | |
348 | ||
349 | case ODEBUG_STATE_ACTIVE: | |
350 | WARN_ON(1); | |
351 | ||
352 | default: | |
353 | return 0; | |
354 | } | |
355 | } | |
356 | ||
357 | /* | |
358 | * fixup_free is called when: | |
359 | * - an active object is freed | |
360 | */ | |
361 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
362 | { | |
363 | struct work_struct *work = addr; | |
364 | ||
365 | switch (state) { | |
366 | case ODEBUG_STATE_ACTIVE: | |
367 | cancel_work_sync(work); | |
368 | debug_object_free(work, &work_debug_descr); | |
369 | return 1; | |
370 | default: | |
371 | return 0; | |
372 | } | |
373 | } | |
374 | ||
375 | static struct debug_obj_descr work_debug_descr = { | |
376 | .name = "work_struct", | |
99777288 | 377 | .debug_hint = work_debug_hint, |
dc186ad7 TG |
378 | .fixup_init = work_fixup_init, |
379 | .fixup_activate = work_fixup_activate, | |
380 | .fixup_free = work_fixup_free, | |
381 | }; | |
382 | ||
383 | static inline void debug_work_activate(struct work_struct *work) | |
384 | { | |
385 | debug_object_activate(work, &work_debug_descr); | |
386 | } | |
387 | ||
388 | static inline void debug_work_deactivate(struct work_struct *work) | |
389 | { | |
390 | debug_object_deactivate(work, &work_debug_descr); | |
391 | } | |
392 | ||
393 | void __init_work(struct work_struct *work, int onstack) | |
394 | { | |
395 | if (onstack) | |
396 | debug_object_init_on_stack(work, &work_debug_descr); | |
397 | else | |
398 | debug_object_init(work, &work_debug_descr); | |
399 | } | |
400 | EXPORT_SYMBOL_GPL(__init_work); | |
401 | ||
402 | void destroy_work_on_stack(struct work_struct *work) | |
403 | { | |
404 | debug_object_free(work, &work_debug_descr); | |
405 | } | |
406 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
407 | ||
408 | #else | |
409 | static inline void debug_work_activate(struct work_struct *work) { } | |
410 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
411 | #endif | |
412 | ||
95402b38 GS |
413 | /* Serializes the accesses to the list of workqueues. */ |
414 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 415 | static LIST_HEAD(workqueues); |
a0a1a5fd | 416 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 417 | |
e22bee78 | 418 | /* |
e19e397a TH |
419 | * The CPU and unbound standard worker pools. The unbound ones have |
420 | * POOL_DISASSOCIATED set, and their workers have WORKER_UNBOUND set. | |
f3421797 | 421 | */ |
e19e397a TH |
422 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS], |
423 | cpu_std_worker_pools); | |
f3421797 | 424 | |
fa1b54e6 TH |
425 | /* |
426 | * idr of all pools. Modifications are protected by workqueue_lock. Read | |
427 | * accesses are protected by sched-RCU protected. | |
428 | */ | |
9daf9e67 TH |
429 | static DEFINE_IDR(worker_pool_idr); |
430 | ||
c34056a3 | 431 | static int worker_thread(void *__worker); |
1da177e4 | 432 | |
9daf9e67 TH |
433 | /* allocate ID and assign it to @pool */ |
434 | static int worker_pool_assign_id(struct worker_pool *pool) | |
435 | { | |
436 | int ret; | |
437 | ||
fa1b54e6 TH |
438 | do { |
439 | if (!idr_pre_get(&worker_pool_idr, GFP_KERNEL)) | |
440 | return -ENOMEM; | |
9daf9e67 | 441 | |
fa1b54e6 TH |
442 | spin_lock_irq(&workqueue_lock); |
443 | ret = idr_get_new(&worker_pool_idr, pool, &pool->id); | |
444 | spin_unlock_irq(&workqueue_lock); | |
445 | } while (ret == -EAGAIN); | |
9daf9e67 | 446 | |
fa1b54e6 | 447 | return ret; |
7c3eed5c TH |
448 | } |
449 | ||
76af4d93 TH |
450 | /** |
451 | * first_pwq - return the first pool_workqueue of the specified workqueue | |
452 | * @wq: the target workqueue | |
453 | * | |
454 | * This must be called either with workqueue_lock held or sched RCU read | |
455 | * locked. If the pwq needs to be used beyond the locking in effect, the | |
456 | * caller is responsible for guaranteeing that the pwq stays online. | |
457 | */ | |
7fb98ea7 | 458 | static struct pool_workqueue *first_pwq(struct workqueue_struct *wq) |
b1f4ec17 | 459 | { |
76af4d93 TH |
460 | assert_rcu_or_wq_lock(); |
461 | return list_first_or_null_rcu(&wq->pwqs, struct pool_workqueue, | |
462 | pwqs_node); | |
b1f4ec17 ON |
463 | } |
464 | ||
73f53c4a TH |
465 | static unsigned int work_color_to_flags(int color) |
466 | { | |
467 | return color << WORK_STRUCT_COLOR_SHIFT; | |
468 | } | |
469 | ||
470 | static int get_work_color(struct work_struct *work) | |
471 | { | |
472 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
473 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
474 | } | |
475 | ||
476 | static int work_next_color(int color) | |
477 | { | |
478 | return (color + 1) % WORK_NR_COLORS; | |
479 | } | |
1da177e4 | 480 | |
14441960 | 481 | /* |
112202d9 TH |
482 | * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data |
483 | * contain the pointer to the queued pwq. Once execution starts, the flag | |
7c3eed5c | 484 | * is cleared and the high bits contain OFFQ flags and pool ID. |
7a22ad75 | 485 | * |
112202d9 TH |
486 | * set_work_pwq(), set_work_pool_and_clear_pending(), mark_work_canceling() |
487 | * and clear_work_data() can be used to set the pwq, pool or clear | |
bbb68dfa TH |
488 | * work->data. These functions should only be called while the work is |
489 | * owned - ie. while the PENDING bit is set. | |
7a22ad75 | 490 | * |
112202d9 | 491 | * get_work_pool() and get_work_pwq() can be used to obtain the pool or pwq |
7c3eed5c | 492 | * corresponding to a work. Pool is available once the work has been |
112202d9 | 493 | * queued anywhere after initialization until it is sync canceled. pwq is |
7c3eed5c | 494 | * available only while the work item is queued. |
7a22ad75 | 495 | * |
bbb68dfa TH |
496 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
497 | * canceled. While being canceled, a work item may have its PENDING set | |
498 | * but stay off timer and worklist for arbitrarily long and nobody should | |
499 | * try to steal the PENDING bit. | |
14441960 | 500 | */ |
7a22ad75 TH |
501 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
502 | unsigned long flags) | |
365970a1 | 503 | { |
6183c009 | 504 | WARN_ON_ONCE(!work_pending(work)); |
7a22ad75 TH |
505 | atomic_long_set(&work->data, data | flags | work_static(work)); |
506 | } | |
365970a1 | 507 | |
112202d9 | 508 | static void set_work_pwq(struct work_struct *work, struct pool_workqueue *pwq, |
7a22ad75 TH |
509 | unsigned long extra_flags) |
510 | { | |
112202d9 TH |
511 | set_work_data(work, (unsigned long)pwq, |
512 | WORK_STRUCT_PENDING | WORK_STRUCT_PWQ | extra_flags); | |
365970a1 DH |
513 | } |
514 | ||
4468a00f LJ |
515 | static void set_work_pool_and_keep_pending(struct work_struct *work, |
516 | int pool_id) | |
517 | { | |
518 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, | |
519 | WORK_STRUCT_PENDING); | |
520 | } | |
521 | ||
7c3eed5c TH |
522 | static void set_work_pool_and_clear_pending(struct work_struct *work, |
523 | int pool_id) | |
7a22ad75 | 524 | { |
23657bb1 TH |
525 | /* |
526 | * The following wmb is paired with the implied mb in | |
527 | * test_and_set_bit(PENDING) and ensures all updates to @work made | |
528 | * here are visible to and precede any updates by the next PENDING | |
529 | * owner. | |
530 | */ | |
531 | smp_wmb(); | |
7c3eed5c | 532 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0); |
7a22ad75 | 533 | } |
f756d5e2 | 534 | |
7a22ad75 | 535 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 536 | { |
7c3eed5c TH |
537 | smp_wmb(); /* see set_work_pool_and_clear_pending() */ |
538 | set_work_data(work, WORK_STRUCT_NO_POOL, 0); | |
1da177e4 LT |
539 | } |
540 | ||
112202d9 | 541 | static struct pool_workqueue *get_work_pwq(struct work_struct *work) |
b1f4ec17 | 542 | { |
e120153d | 543 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 544 | |
112202d9 | 545 | if (data & WORK_STRUCT_PWQ) |
e120153d TH |
546 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); |
547 | else | |
548 | return NULL; | |
4d707b9f ON |
549 | } |
550 | ||
7c3eed5c TH |
551 | /** |
552 | * get_work_pool - return the worker_pool a given work was associated with | |
553 | * @work: the work item of interest | |
554 | * | |
555 | * Return the worker_pool @work was last associated with. %NULL if none. | |
fa1b54e6 TH |
556 | * |
557 | * Pools are created and destroyed under workqueue_lock, and allows read | |
558 | * access under sched-RCU read lock. As such, this function should be | |
559 | * called under workqueue_lock or with preemption disabled. | |
560 | * | |
561 | * All fields of the returned pool are accessible as long as the above | |
562 | * mentioned locking is in effect. If the returned pool needs to be used | |
563 | * beyond the critical section, the caller is responsible for ensuring the | |
564 | * returned pool is and stays online. | |
7c3eed5c TH |
565 | */ |
566 | static struct worker_pool *get_work_pool(struct work_struct *work) | |
365970a1 | 567 | { |
e120153d | 568 | unsigned long data = atomic_long_read(&work->data); |
7c3eed5c | 569 | int pool_id; |
7a22ad75 | 570 | |
fa1b54e6 TH |
571 | assert_rcu_or_wq_lock(); |
572 | ||
112202d9 TH |
573 | if (data & WORK_STRUCT_PWQ) |
574 | return ((struct pool_workqueue *) | |
7c3eed5c | 575 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool; |
7a22ad75 | 576 | |
7c3eed5c TH |
577 | pool_id = data >> WORK_OFFQ_POOL_SHIFT; |
578 | if (pool_id == WORK_OFFQ_POOL_NONE) | |
7a22ad75 TH |
579 | return NULL; |
580 | ||
fa1b54e6 | 581 | return idr_find(&worker_pool_idr, pool_id); |
7c3eed5c TH |
582 | } |
583 | ||
584 | /** | |
585 | * get_work_pool_id - return the worker pool ID a given work is associated with | |
586 | * @work: the work item of interest | |
587 | * | |
588 | * Return the worker_pool ID @work was last associated with. | |
589 | * %WORK_OFFQ_POOL_NONE if none. | |
590 | */ | |
591 | static int get_work_pool_id(struct work_struct *work) | |
592 | { | |
54d5b7d0 LJ |
593 | unsigned long data = atomic_long_read(&work->data); |
594 | ||
112202d9 TH |
595 | if (data & WORK_STRUCT_PWQ) |
596 | return ((struct pool_workqueue *) | |
54d5b7d0 | 597 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->id; |
7c3eed5c | 598 | |
54d5b7d0 | 599 | return data >> WORK_OFFQ_POOL_SHIFT; |
7c3eed5c TH |
600 | } |
601 | ||
bbb68dfa TH |
602 | static void mark_work_canceling(struct work_struct *work) |
603 | { | |
7c3eed5c | 604 | unsigned long pool_id = get_work_pool_id(work); |
bbb68dfa | 605 | |
7c3eed5c TH |
606 | pool_id <<= WORK_OFFQ_POOL_SHIFT; |
607 | set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING); | |
bbb68dfa TH |
608 | } |
609 | ||
610 | static bool work_is_canceling(struct work_struct *work) | |
611 | { | |
612 | unsigned long data = atomic_long_read(&work->data); | |
613 | ||
112202d9 | 614 | return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING); |
bbb68dfa TH |
615 | } |
616 | ||
e22bee78 | 617 | /* |
3270476a TH |
618 | * Policy functions. These define the policies on how the global worker |
619 | * pools are managed. Unless noted otherwise, these functions assume that | |
d565ed63 | 620 | * they're being called with pool->lock held. |
e22bee78 TH |
621 | */ |
622 | ||
63d95a91 | 623 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 624 | { |
e19e397a | 625 | return !atomic_read(&pool->nr_running); |
a848e3b6 ON |
626 | } |
627 | ||
4594bf15 | 628 | /* |
e22bee78 TH |
629 | * Need to wake up a worker? Called from anything but currently |
630 | * running workers. | |
974271c4 TH |
631 | * |
632 | * Note that, because unbound workers never contribute to nr_running, this | |
706026c2 | 633 | * function will always return %true for unbound pools as long as the |
974271c4 | 634 | * worklist isn't empty. |
4594bf15 | 635 | */ |
63d95a91 | 636 | static bool need_more_worker(struct worker_pool *pool) |
365970a1 | 637 | { |
63d95a91 | 638 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
e22bee78 | 639 | } |
4594bf15 | 640 | |
e22bee78 | 641 | /* Can I start working? Called from busy but !running workers. */ |
63d95a91 | 642 | static bool may_start_working(struct worker_pool *pool) |
e22bee78 | 643 | { |
63d95a91 | 644 | return pool->nr_idle; |
e22bee78 TH |
645 | } |
646 | ||
647 | /* Do I need to keep working? Called from currently running workers. */ | |
63d95a91 | 648 | static bool keep_working(struct worker_pool *pool) |
e22bee78 | 649 | { |
e19e397a TH |
650 | return !list_empty(&pool->worklist) && |
651 | atomic_read(&pool->nr_running) <= 1; | |
e22bee78 TH |
652 | } |
653 | ||
654 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 655 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 656 | { |
63d95a91 | 657 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 658 | } |
365970a1 | 659 | |
e22bee78 | 660 | /* Do I need to be the manager? */ |
63d95a91 | 661 | static bool need_to_manage_workers(struct worker_pool *pool) |
e22bee78 | 662 | { |
63d95a91 | 663 | return need_to_create_worker(pool) || |
11ebea50 | 664 | (pool->flags & POOL_MANAGE_WORKERS); |
e22bee78 TH |
665 | } |
666 | ||
667 | /* Do we have too many workers and should some go away? */ | |
63d95a91 | 668 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 669 | { |
34a06bd6 | 670 | bool managing = mutex_is_locked(&pool->manager_arb); |
63d95a91 TH |
671 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
672 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 | 673 | |
ea1abd61 LJ |
674 | /* |
675 | * nr_idle and idle_list may disagree if idle rebinding is in | |
676 | * progress. Never return %true if idle_list is empty. | |
677 | */ | |
678 | if (list_empty(&pool->idle_list)) | |
679 | return false; | |
680 | ||
e22bee78 | 681 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
365970a1 DH |
682 | } |
683 | ||
4d707b9f | 684 | /* |
e22bee78 TH |
685 | * Wake up functions. |
686 | */ | |
687 | ||
7e11629d | 688 | /* Return the first worker. Safe with preemption disabled */ |
63d95a91 | 689 | static struct worker *first_worker(struct worker_pool *pool) |
7e11629d | 690 | { |
63d95a91 | 691 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
692 | return NULL; |
693 | ||
63d95a91 | 694 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
695 | } |
696 | ||
697 | /** | |
698 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 699 | * @pool: worker pool to wake worker from |
7e11629d | 700 | * |
63d95a91 | 701 | * Wake up the first idle worker of @pool. |
7e11629d TH |
702 | * |
703 | * CONTEXT: | |
d565ed63 | 704 | * spin_lock_irq(pool->lock). |
7e11629d | 705 | */ |
63d95a91 | 706 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 707 | { |
63d95a91 | 708 | struct worker *worker = first_worker(pool); |
7e11629d TH |
709 | |
710 | if (likely(worker)) | |
711 | wake_up_process(worker->task); | |
712 | } | |
713 | ||
d302f017 | 714 | /** |
e22bee78 TH |
715 | * wq_worker_waking_up - a worker is waking up |
716 | * @task: task waking up | |
717 | * @cpu: CPU @task is waking up to | |
718 | * | |
719 | * This function is called during try_to_wake_up() when a worker is | |
720 | * being awoken. | |
721 | * | |
722 | * CONTEXT: | |
723 | * spin_lock_irq(rq->lock) | |
724 | */ | |
d84ff051 | 725 | void wq_worker_waking_up(struct task_struct *task, int cpu) |
e22bee78 TH |
726 | { |
727 | struct worker *worker = kthread_data(task); | |
728 | ||
36576000 | 729 | if (!(worker->flags & WORKER_NOT_RUNNING)) { |
ec22ca5e | 730 | WARN_ON_ONCE(worker->pool->cpu != cpu); |
e19e397a | 731 | atomic_inc(&worker->pool->nr_running); |
36576000 | 732 | } |
e22bee78 TH |
733 | } |
734 | ||
735 | /** | |
736 | * wq_worker_sleeping - a worker is going to sleep | |
737 | * @task: task going to sleep | |
738 | * @cpu: CPU in question, must be the current CPU number | |
739 | * | |
740 | * This function is called during schedule() when a busy worker is | |
741 | * going to sleep. Worker on the same cpu can be woken up by | |
742 | * returning pointer to its task. | |
743 | * | |
744 | * CONTEXT: | |
745 | * spin_lock_irq(rq->lock) | |
746 | * | |
747 | * RETURNS: | |
748 | * Worker task on @cpu to wake up, %NULL if none. | |
749 | */ | |
d84ff051 | 750 | struct task_struct *wq_worker_sleeping(struct task_struct *task, int cpu) |
e22bee78 TH |
751 | { |
752 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
111c225a | 753 | struct worker_pool *pool; |
e22bee78 | 754 | |
111c225a TH |
755 | /* |
756 | * Rescuers, which may not have all the fields set up like normal | |
757 | * workers, also reach here, let's not access anything before | |
758 | * checking NOT_RUNNING. | |
759 | */ | |
2d64672e | 760 | if (worker->flags & WORKER_NOT_RUNNING) |
e22bee78 TH |
761 | return NULL; |
762 | ||
111c225a | 763 | pool = worker->pool; |
111c225a | 764 | |
e22bee78 | 765 | /* this can only happen on the local cpu */ |
6183c009 TH |
766 | if (WARN_ON_ONCE(cpu != raw_smp_processor_id())) |
767 | return NULL; | |
e22bee78 TH |
768 | |
769 | /* | |
770 | * The counterpart of the following dec_and_test, implied mb, | |
771 | * worklist not empty test sequence is in insert_work(). | |
772 | * Please read comment there. | |
773 | * | |
628c78e7 TH |
774 | * NOT_RUNNING is clear. This means that we're bound to and |
775 | * running on the local cpu w/ rq lock held and preemption | |
776 | * disabled, which in turn means that none else could be | |
d565ed63 | 777 | * manipulating idle_list, so dereferencing idle_list without pool |
628c78e7 | 778 | * lock is safe. |
e22bee78 | 779 | */ |
e19e397a TH |
780 | if (atomic_dec_and_test(&pool->nr_running) && |
781 | !list_empty(&pool->worklist)) | |
63d95a91 | 782 | to_wakeup = first_worker(pool); |
e22bee78 TH |
783 | return to_wakeup ? to_wakeup->task : NULL; |
784 | } | |
785 | ||
786 | /** | |
787 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 788 | * @worker: self |
d302f017 TH |
789 | * @flags: flags to set |
790 | * @wakeup: wakeup an idle worker if necessary | |
791 | * | |
e22bee78 TH |
792 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
793 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
794 | * woken up. | |
d302f017 | 795 | * |
cb444766 | 796 | * CONTEXT: |
d565ed63 | 797 | * spin_lock_irq(pool->lock) |
d302f017 TH |
798 | */ |
799 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
800 | bool wakeup) | |
801 | { | |
bd7bdd43 | 802 | struct worker_pool *pool = worker->pool; |
e22bee78 | 803 | |
cb444766 TH |
804 | WARN_ON_ONCE(worker->task != current); |
805 | ||
e22bee78 TH |
806 | /* |
807 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
808 | * wake up an idle worker as necessary if requested by | |
809 | * @wakeup. | |
810 | */ | |
811 | if ((flags & WORKER_NOT_RUNNING) && | |
812 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
e22bee78 | 813 | if (wakeup) { |
e19e397a | 814 | if (atomic_dec_and_test(&pool->nr_running) && |
bd7bdd43 | 815 | !list_empty(&pool->worklist)) |
63d95a91 | 816 | wake_up_worker(pool); |
e22bee78 | 817 | } else |
e19e397a | 818 | atomic_dec(&pool->nr_running); |
e22bee78 TH |
819 | } |
820 | ||
d302f017 TH |
821 | worker->flags |= flags; |
822 | } | |
823 | ||
824 | /** | |
e22bee78 | 825 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 826 | * @worker: self |
d302f017 TH |
827 | * @flags: flags to clear |
828 | * | |
e22bee78 | 829 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 830 | * |
cb444766 | 831 | * CONTEXT: |
d565ed63 | 832 | * spin_lock_irq(pool->lock) |
d302f017 TH |
833 | */ |
834 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
835 | { | |
63d95a91 | 836 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
837 | unsigned int oflags = worker->flags; |
838 | ||
cb444766 TH |
839 | WARN_ON_ONCE(worker->task != current); |
840 | ||
d302f017 | 841 | worker->flags &= ~flags; |
e22bee78 | 842 | |
42c025f3 TH |
843 | /* |
844 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
845 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
846 | * of multiple flags, not a single flag. | |
847 | */ | |
e22bee78 TH |
848 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
849 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
e19e397a | 850 | atomic_inc(&pool->nr_running); |
d302f017 TH |
851 | } |
852 | ||
8cca0eea TH |
853 | /** |
854 | * find_worker_executing_work - find worker which is executing a work | |
c9e7cf27 | 855 | * @pool: pool of interest |
8cca0eea TH |
856 | * @work: work to find worker for |
857 | * | |
c9e7cf27 TH |
858 | * Find a worker which is executing @work on @pool by searching |
859 | * @pool->busy_hash which is keyed by the address of @work. For a worker | |
a2c1c57b TH |
860 | * to match, its current execution should match the address of @work and |
861 | * its work function. This is to avoid unwanted dependency between | |
862 | * unrelated work executions through a work item being recycled while still | |
863 | * being executed. | |
864 | * | |
865 | * This is a bit tricky. A work item may be freed once its execution | |
866 | * starts and nothing prevents the freed area from being recycled for | |
867 | * another work item. If the same work item address ends up being reused | |
868 | * before the original execution finishes, workqueue will identify the | |
869 | * recycled work item as currently executing and make it wait until the | |
870 | * current execution finishes, introducing an unwanted dependency. | |
871 | * | |
872 | * This function checks the work item address, work function and workqueue | |
873 | * to avoid false positives. Note that this isn't complete as one may | |
874 | * construct a work function which can introduce dependency onto itself | |
875 | * through a recycled work item. Well, if somebody wants to shoot oneself | |
876 | * in the foot that badly, there's only so much we can do, and if such | |
877 | * deadlock actually occurs, it should be easy to locate the culprit work | |
878 | * function. | |
8cca0eea TH |
879 | * |
880 | * CONTEXT: | |
d565ed63 | 881 | * spin_lock_irq(pool->lock). |
8cca0eea TH |
882 | * |
883 | * RETURNS: | |
884 | * Pointer to worker which is executing @work if found, NULL | |
885 | * otherwise. | |
4d707b9f | 886 | */ |
c9e7cf27 | 887 | static struct worker *find_worker_executing_work(struct worker_pool *pool, |
8cca0eea | 888 | struct work_struct *work) |
4d707b9f | 889 | { |
42f8570f | 890 | struct worker *worker; |
42f8570f | 891 | |
b67bfe0d | 892 | hash_for_each_possible(pool->busy_hash, worker, hentry, |
a2c1c57b TH |
893 | (unsigned long)work) |
894 | if (worker->current_work == work && | |
895 | worker->current_func == work->func) | |
42f8570f SL |
896 | return worker; |
897 | ||
898 | return NULL; | |
4d707b9f ON |
899 | } |
900 | ||
bf4ede01 TH |
901 | /** |
902 | * move_linked_works - move linked works to a list | |
903 | * @work: start of series of works to be scheduled | |
904 | * @head: target list to append @work to | |
905 | * @nextp: out paramter for nested worklist walking | |
906 | * | |
907 | * Schedule linked works starting from @work to @head. Work series to | |
908 | * be scheduled starts at @work and includes any consecutive work with | |
909 | * WORK_STRUCT_LINKED set in its predecessor. | |
910 | * | |
911 | * If @nextp is not NULL, it's updated to point to the next work of | |
912 | * the last scheduled work. This allows move_linked_works() to be | |
913 | * nested inside outer list_for_each_entry_safe(). | |
914 | * | |
915 | * CONTEXT: | |
d565ed63 | 916 | * spin_lock_irq(pool->lock). |
bf4ede01 TH |
917 | */ |
918 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
919 | struct work_struct **nextp) | |
920 | { | |
921 | struct work_struct *n; | |
922 | ||
923 | /* | |
924 | * Linked worklist will always end before the end of the list, | |
925 | * use NULL for list head. | |
926 | */ | |
927 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
928 | list_move_tail(&work->entry, head); | |
929 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
930 | break; | |
931 | } | |
932 | ||
933 | /* | |
934 | * If we're already inside safe list traversal and have moved | |
935 | * multiple works to the scheduled queue, the next position | |
936 | * needs to be updated. | |
937 | */ | |
938 | if (nextp) | |
939 | *nextp = n; | |
940 | } | |
941 | ||
112202d9 | 942 | static void pwq_activate_delayed_work(struct work_struct *work) |
bf4ede01 | 943 | { |
112202d9 | 944 | struct pool_workqueue *pwq = get_work_pwq(work); |
bf4ede01 TH |
945 | |
946 | trace_workqueue_activate_work(work); | |
112202d9 | 947 | move_linked_works(work, &pwq->pool->worklist, NULL); |
bf4ede01 | 948 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); |
112202d9 | 949 | pwq->nr_active++; |
bf4ede01 TH |
950 | } |
951 | ||
112202d9 | 952 | static void pwq_activate_first_delayed(struct pool_workqueue *pwq) |
3aa62497 | 953 | { |
112202d9 | 954 | struct work_struct *work = list_first_entry(&pwq->delayed_works, |
3aa62497 LJ |
955 | struct work_struct, entry); |
956 | ||
112202d9 | 957 | pwq_activate_delayed_work(work); |
3aa62497 LJ |
958 | } |
959 | ||
bf4ede01 | 960 | /** |
112202d9 TH |
961 | * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight |
962 | * @pwq: pwq of interest | |
bf4ede01 | 963 | * @color: color of work which left the queue |
bf4ede01 TH |
964 | * |
965 | * A work either has completed or is removed from pending queue, | |
112202d9 | 966 | * decrement nr_in_flight of its pwq and handle workqueue flushing. |
bf4ede01 TH |
967 | * |
968 | * CONTEXT: | |
d565ed63 | 969 | * spin_lock_irq(pool->lock). |
bf4ede01 | 970 | */ |
112202d9 | 971 | static void pwq_dec_nr_in_flight(struct pool_workqueue *pwq, int color) |
bf4ede01 TH |
972 | { |
973 | /* ignore uncolored works */ | |
974 | if (color == WORK_NO_COLOR) | |
975 | return; | |
976 | ||
112202d9 | 977 | pwq->nr_in_flight[color]--; |
bf4ede01 | 978 | |
112202d9 TH |
979 | pwq->nr_active--; |
980 | if (!list_empty(&pwq->delayed_works)) { | |
b3f9f405 | 981 | /* one down, submit a delayed one */ |
112202d9 TH |
982 | if (pwq->nr_active < pwq->max_active) |
983 | pwq_activate_first_delayed(pwq); | |
bf4ede01 TH |
984 | } |
985 | ||
986 | /* is flush in progress and are we at the flushing tip? */ | |
112202d9 | 987 | if (likely(pwq->flush_color != color)) |
bf4ede01 TH |
988 | return; |
989 | ||
990 | /* are there still in-flight works? */ | |
112202d9 | 991 | if (pwq->nr_in_flight[color]) |
bf4ede01 TH |
992 | return; |
993 | ||
112202d9 TH |
994 | /* this pwq is done, clear flush_color */ |
995 | pwq->flush_color = -1; | |
bf4ede01 TH |
996 | |
997 | /* | |
112202d9 | 998 | * If this was the last pwq, wake up the first flusher. It |
bf4ede01 TH |
999 | * will handle the rest. |
1000 | */ | |
112202d9 TH |
1001 | if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush)) |
1002 | complete(&pwq->wq->first_flusher->done); | |
bf4ede01 TH |
1003 | } |
1004 | ||
36e227d2 | 1005 | /** |
bbb68dfa | 1006 | * try_to_grab_pending - steal work item from worklist and disable irq |
36e227d2 TH |
1007 | * @work: work item to steal |
1008 | * @is_dwork: @work is a delayed_work | |
bbb68dfa | 1009 | * @flags: place to store irq state |
36e227d2 TH |
1010 | * |
1011 | * Try to grab PENDING bit of @work. This function can handle @work in any | |
1012 | * stable state - idle, on timer or on worklist. Return values are | |
1013 | * | |
1014 | * 1 if @work was pending and we successfully stole PENDING | |
1015 | * 0 if @work was idle and we claimed PENDING | |
1016 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry | |
bbb68dfa TH |
1017 | * -ENOENT if someone else is canceling @work, this state may persist |
1018 | * for arbitrarily long | |
36e227d2 | 1019 | * |
bbb68dfa | 1020 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
e0aecdd8 TH |
1021 | * interrupted while holding PENDING and @work off queue, irq must be |
1022 | * disabled on entry. This, combined with delayed_work->timer being | |
1023 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. | |
bbb68dfa TH |
1024 | * |
1025 | * On successful return, >= 0, irq is disabled and the caller is | |
1026 | * responsible for releasing it using local_irq_restore(*@flags). | |
1027 | * | |
e0aecdd8 | 1028 | * This function is safe to call from any context including IRQ handler. |
bf4ede01 | 1029 | */ |
bbb68dfa TH |
1030 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
1031 | unsigned long *flags) | |
bf4ede01 | 1032 | { |
d565ed63 | 1033 | struct worker_pool *pool; |
112202d9 | 1034 | struct pool_workqueue *pwq; |
bf4ede01 | 1035 | |
bbb68dfa TH |
1036 | local_irq_save(*flags); |
1037 | ||
36e227d2 TH |
1038 | /* try to steal the timer if it exists */ |
1039 | if (is_dwork) { | |
1040 | struct delayed_work *dwork = to_delayed_work(work); | |
1041 | ||
e0aecdd8 TH |
1042 | /* |
1043 | * dwork->timer is irqsafe. If del_timer() fails, it's | |
1044 | * guaranteed that the timer is not queued anywhere and not | |
1045 | * running on the local CPU. | |
1046 | */ | |
36e227d2 TH |
1047 | if (likely(del_timer(&dwork->timer))) |
1048 | return 1; | |
1049 | } | |
1050 | ||
1051 | /* try to claim PENDING the normal way */ | |
bf4ede01 TH |
1052 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1053 | return 0; | |
1054 | ||
1055 | /* | |
1056 | * The queueing is in progress, or it is already queued. Try to | |
1057 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1058 | */ | |
d565ed63 TH |
1059 | pool = get_work_pool(work); |
1060 | if (!pool) | |
bbb68dfa | 1061 | goto fail; |
bf4ede01 | 1062 | |
d565ed63 | 1063 | spin_lock(&pool->lock); |
0b3dae68 | 1064 | /* |
112202d9 TH |
1065 | * work->data is guaranteed to point to pwq only while the work |
1066 | * item is queued on pwq->wq, and both updating work->data to point | |
1067 | * to pwq on queueing and to pool on dequeueing are done under | |
1068 | * pwq->pool->lock. This in turn guarantees that, if work->data | |
1069 | * points to pwq which is associated with a locked pool, the work | |
0b3dae68 LJ |
1070 | * item is currently queued on that pool. |
1071 | */ | |
112202d9 TH |
1072 | pwq = get_work_pwq(work); |
1073 | if (pwq && pwq->pool == pool) { | |
16062836 TH |
1074 | debug_work_deactivate(work); |
1075 | ||
1076 | /* | |
1077 | * A delayed work item cannot be grabbed directly because | |
1078 | * it might have linked NO_COLOR work items which, if left | |
112202d9 | 1079 | * on the delayed_list, will confuse pwq->nr_active |
16062836 TH |
1080 | * management later on and cause stall. Make sure the work |
1081 | * item is activated before grabbing. | |
1082 | */ | |
1083 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) | |
112202d9 | 1084 | pwq_activate_delayed_work(work); |
16062836 TH |
1085 | |
1086 | list_del_init(&work->entry); | |
112202d9 | 1087 | pwq_dec_nr_in_flight(get_work_pwq(work), get_work_color(work)); |
16062836 | 1088 | |
112202d9 | 1089 | /* work->data points to pwq iff queued, point to pool */ |
16062836 TH |
1090 | set_work_pool_and_keep_pending(work, pool->id); |
1091 | ||
1092 | spin_unlock(&pool->lock); | |
1093 | return 1; | |
bf4ede01 | 1094 | } |
d565ed63 | 1095 | spin_unlock(&pool->lock); |
bbb68dfa TH |
1096 | fail: |
1097 | local_irq_restore(*flags); | |
1098 | if (work_is_canceling(work)) | |
1099 | return -ENOENT; | |
1100 | cpu_relax(); | |
36e227d2 | 1101 | return -EAGAIN; |
bf4ede01 TH |
1102 | } |
1103 | ||
4690c4ab | 1104 | /** |
706026c2 | 1105 | * insert_work - insert a work into a pool |
112202d9 | 1106 | * @pwq: pwq @work belongs to |
4690c4ab TH |
1107 | * @work: work to insert |
1108 | * @head: insertion point | |
1109 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
1110 | * | |
112202d9 | 1111 | * Insert @work which belongs to @pwq after @head. @extra_flags is or'd to |
706026c2 | 1112 | * work_struct flags. |
4690c4ab TH |
1113 | * |
1114 | * CONTEXT: | |
d565ed63 | 1115 | * spin_lock_irq(pool->lock). |
4690c4ab | 1116 | */ |
112202d9 TH |
1117 | static void insert_work(struct pool_workqueue *pwq, struct work_struct *work, |
1118 | struct list_head *head, unsigned int extra_flags) | |
b89deed3 | 1119 | { |
112202d9 | 1120 | struct worker_pool *pool = pwq->pool; |
e22bee78 | 1121 | |
4690c4ab | 1122 | /* we own @work, set data and link */ |
112202d9 | 1123 | set_work_pwq(work, pwq, extra_flags); |
1a4d9b0a | 1124 | list_add_tail(&work->entry, head); |
e22bee78 TH |
1125 | |
1126 | /* | |
1127 | * Ensure either worker_sched_deactivated() sees the above | |
1128 | * list_add_tail() or we see zero nr_running to avoid workers | |
1129 | * lying around lazily while there are works to be processed. | |
1130 | */ | |
1131 | smp_mb(); | |
1132 | ||
63d95a91 TH |
1133 | if (__need_more_worker(pool)) |
1134 | wake_up_worker(pool); | |
b89deed3 ON |
1135 | } |
1136 | ||
c8efcc25 TH |
1137 | /* |
1138 | * Test whether @work is being queued from another work executing on the | |
8d03ecfe | 1139 | * same workqueue. |
c8efcc25 TH |
1140 | */ |
1141 | static bool is_chained_work(struct workqueue_struct *wq) | |
1142 | { | |
8d03ecfe TH |
1143 | struct worker *worker; |
1144 | ||
1145 | worker = current_wq_worker(); | |
1146 | /* | |
1147 | * Return %true iff I'm a worker execuing a work item on @wq. If | |
1148 | * I'm @worker, it's safe to dereference it without locking. | |
1149 | */ | |
112202d9 | 1150 | return worker && worker->current_pwq->wq == wq; |
c8efcc25 TH |
1151 | } |
1152 | ||
d84ff051 | 1153 | static void __queue_work(int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
1154 | struct work_struct *work) |
1155 | { | |
112202d9 | 1156 | struct pool_workqueue *pwq; |
1e19ffc6 | 1157 | struct list_head *worklist; |
8a2e8e5d | 1158 | unsigned int work_flags; |
b75cac93 | 1159 | unsigned int req_cpu = cpu; |
8930caba TH |
1160 | |
1161 | /* | |
1162 | * While a work item is PENDING && off queue, a task trying to | |
1163 | * steal the PENDING will busy-loop waiting for it to either get | |
1164 | * queued or lose PENDING. Grabbing PENDING and queueing should | |
1165 | * happen with IRQ disabled. | |
1166 | */ | |
1167 | WARN_ON_ONCE(!irqs_disabled()); | |
1da177e4 | 1168 | |
dc186ad7 | 1169 | debug_work_activate(work); |
1e19ffc6 | 1170 | |
c8efcc25 | 1171 | /* if dying, only works from the same workqueue are allowed */ |
9c5a2ba7 | 1172 | if (unlikely(wq->flags & WQ_DRAINING) && |
c8efcc25 | 1173 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b TH |
1174 | return; |
1175 | ||
112202d9 | 1176 | /* determine the pwq to use */ |
c7fc77f7 | 1177 | if (!(wq->flags & WQ_UNBOUND)) { |
c9e7cf27 | 1178 | struct worker_pool *last_pool; |
18aa9eff | 1179 | |
57469821 | 1180 | if (cpu == WORK_CPU_UNBOUND) |
c7fc77f7 TH |
1181 | cpu = raw_smp_processor_id(); |
1182 | ||
18aa9eff | 1183 | /* |
dbf2576e TH |
1184 | * It's multi cpu. If @work was previously on a different |
1185 | * cpu, it might still be running there, in which case the | |
1186 | * work needs to be queued on that cpu to guarantee | |
1187 | * non-reentrancy. | |
18aa9eff | 1188 | */ |
7fb98ea7 | 1189 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); |
c9e7cf27 | 1190 | last_pool = get_work_pool(work); |
dbf2576e | 1191 | |
112202d9 | 1192 | if (last_pool && last_pool != pwq->pool) { |
18aa9eff TH |
1193 | struct worker *worker; |
1194 | ||
d565ed63 | 1195 | spin_lock(&last_pool->lock); |
18aa9eff | 1196 | |
c9e7cf27 | 1197 | worker = find_worker_executing_work(last_pool, work); |
18aa9eff | 1198 | |
112202d9 | 1199 | if (worker && worker->current_pwq->wq == wq) { |
7fb98ea7 | 1200 | pwq = per_cpu_ptr(wq->cpu_pwqs, last_pool->cpu); |
8594fade | 1201 | } else { |
18aa9eff | 1202 | /* meh... not running there, queue here */ |
d565ed63 | 1203 | spin_unlock(&last_pool->lock); |
112202d9 | 1204 | spin_lock(&pwq->pool->lock); |
18aa9eff | 1205 | } |
8930caba | 1206 | } else { |
112202d9 | 1207 | spin_lock(&pwq->pool->lock); |
8930caba | 1208 | } |
f3421797 | 1209 | } else { |
7fb98ea7 | 1210 | pwq = first_pwq(wq); |
112202d9 | 1211 | spin_lock(&pwq->pool->lock); |
502ca9d8 TH |
1212 | } |
1213 | ||
112202d9 TH |
1214 | /* pwq determined, queue */ |
1215 | trace_workqueue_queue_work(req_cpu, pwq, work); | |
502ca9d8 | 1216 | |
f5b2552b | 1217 | if (WARN_ON(!list_empty(&work->entry))) { |
112202d9 | 1218 | spin_unlock(&pwq->pool->lock); |
f5b2552b DC |
1219 | return; |
1220 | } | |
1e19ffc6 | 1221 | |
112202d9 TH |
1222 | pwq->nr_in_flight[pwq->work_color]++; |
1223 | work_flags = work_color_to_flags(pwq->work_color); | |
1e19ffc6 | 1224 | |
112202d9 | 1225 | if (likely(pwq->nr_active < pwq->max_active)) { |
cdadf009 | 1226 | trace_workqueue_activate_work(work); |
112202d9 TH |
1227 | pwq->nr_active++; |
1228 | worklist = &pwq->pool->worklist; | |
8a2e8e5d TH |
1229 | } else { |
1230 | work_flags |= WORK_STRUCT_DELAYED; | |
112202d9 | 1231 | worklist = &pwq->delayed_works; |
8a2e8e5d | 1232 | } |
1e19ffc6 | 1233 | |
112202d9 | 1234 | insert_work(pwq, work, worklist, work_flags); |
1e19ffc6 | 1235 | |
112202d9 | 1236 | spin_unlock(&pwq->pool->lock); |
1da177e4 LT |
1237 | } |
1238 | ||
0fcb78c2 | 1239 | /** |
c1a220e7 ZR |
1240 | * queue_work_on - queue work on specific cpu |
1241 | * @cpu: CPU number to execute work on | |
0fcb78c2 REB |
1242 | * @wq: workqueue to use |
1243 | * @work: work to queue | |
1244 | * | |
d4283e93 | 1245 | * Returns %false if @work was already on a queue, %true otherwise. |
1da177e4 | 1246 | * |
c1a220e7 ZR |
1247 | * We queue the work to a specific CPU, the caller must ensure it |
1248 | * can't go away. | |
1da177e4 | 1249 | */ |
d4283e93 TH |
1250 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
1251 | struct work_struct *work) | |
1da177e4 | 1252 | { |
d4283e93 | 1253 | bool ret = false; |
8930caba | 1254 | unsigned long flags; |
ef1ca236 | 1255 | |
8930caba | 1256 | local_irq_save(flags); |
c1a220e7 | 1257 | |
22df02bb | 1258 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1259 | __queue_work(cpu, wq, work); |
d4283e93 | 1260 | ret = true; |
c1a220e7 | 1261 | } |
ef1ca236 | 1262 | |
8930caba | 1263 | local_irq_restore(flags); |
1da177e4 LT |
1264 | return ret; |
1265 | } | |
c1a220e7 | 1266 | EXPORT_SYMBOL_GPL(queue_work_on); |
1da177e4 | 1267 | |
c1a220e7 | 1268 | /** |
0a13c00e | 1269 | * queue_work - queue work on a workqueue |
c1a220e7 ZR |
1270 | * @wq: workqueue to use |
1271 | * @work: work to queue | |
1272 | * | |
d4283e93 | 1273 | * Returns %false if @work was already on a queue, %true otherwise. |
c1a220e7 | 1274 | * |
0a13c00e TH |
1275 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
1276 | * it can be processed by another CPU. | |
c1a220e7 | 1277 | */ |
d4283e93 | 1278 | bool queue_work(struct workqueue_struct *wq, struct work_struct *work) |
c1a220e7 | 1279 | { |
57469821 | 1280 | return queue_work_on(WORK_CPU_UNBOUND, wq, work); |
c1a220e7 | 1281 | } |
0a13c00e | 1282 | EXPORT_SYMBOL_GPL(queue_work); |
c1a220e7 | 1283 | |
d8e794df | 1284 | void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1285 | { |
52bad64d | 1286 | struct delayed_work *dwork = (struct delayed_work *)__data; |
1da177e4 | 1287 | |
e0aecdd8 | 1288 | /* should have been called from irqsafe timer with irq already off */ |
60c057bc | 1289 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
1da177e4 | 1290 | } |
1438ade5 | 1291 | EXPORT_SYMBOL(delayed_work_timer_fn); |
1da177e4 | 1292 | |
7beb2edf TH |
1293 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
1294 | struct delayed_work *dwork, unsigned long delay) | |
1da177e4 | 1295 | { |
7beb2edf TH |
1296 | struct timer_list *timer = &dwork->timer; |
1297 | struct work_struct *work = &dwork->work; | |
7beb2edf TH |
1298 | |
1299 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || | |
1300 | timer->data != (unsigned long)dwork); | |
fc4b514f TH |
1301 | WARN_ON_ONCE(timer_pending(timer)); |
1302 | WARN_ON_ONCE(!list_empty(&work->entry)); | |
7beb2edf | 1303 | |
8852aac2 TH |
1304 | /* |
1305 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1306 | * both optimization and correctness. The earliest @timer can | |
1307 | * expire is on the closest next tick and delayed_work users depend | |
1308 | * on that there's no such delay when @delay is 0. | |
1309 | */ | |
1310 | if (!delay) { | |
1311 | __queue_work(cpu, wq, &dwork->work); | |
1312 | return; | |
1313 | } | |
1314 | ||
7beb2edf | 1315 | timer_stats_timer_set_start_info(&dwork->timer); |
1da177e4 | 1316 | |
60c057bc | 1317 | dwork->wq = wq; |
1265057f | 1318 | dwork->cpu = cpu; |
7beb2edf TH |
1319 | timer->expires = jiffies + delay; |
1320 | ||
1321 | if (unlikely(cpu != WORK_CPU_UNBOUND)) | |
1322 | add_timer_on(timer, cpu); | |
1323 | else | |
1324 | add_timer(timer); | |
1da177e4 LT |
1325 | } |
1326 | ||
0fcb78c2 REB |
1327 | /** |
1328 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1329 | * @cpu: CPU number to execute work on | |
1330 | * @wq: workqueue to use | |
af9997e4 | 1331 | * @dwork: work to queue |
0fcb78c2 REB |
1332 | * @delay: number of jiffies to wait before queueing |
1333 | * | |
715f1300 TH |
1334 | * Returns %false if @work was already on a queue, %true otherwise. If |
1335 | * @delay is zero and @dwork is idle, it will be scheduled for immediate | |
1336 | * execution. | |
0fcb78c2 | 1337 | */ |
d4283e93 TH |
1338 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
1339 | struct delayed_work *dwork, unsigned long delay) | |
7a6bc1cd | 1340 | { |
52bad64d | 1341 | struct work_struct *work = &dwork->work; |
d4283e93 | 1342 | bool ret = false; |
8930caba | 1343 | unsigned long flags; |
7a6bc1cd | 1344 | |
8930caba TH |
1345 | /* read the comment in __queue_work() */ |
1346 | local_irq_save(flags); | |
7a6bc1cd | 1347 | |
22df02bb | 1348 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7beb2edf | 1349 | __queue_delayed_work(cpu, wq, dwork, delay); |
d4283e93 | 1350 | ret = true; |
7a6bc1cd | 1351 | } |
8a3e77cc | 1352 | |
8930caba | 1353 | local_irq_restore(flags); |
7a6bc1cd VP |
1354 | return ret; |
1355 | } | |
ae90dd5d | 1356 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
c7fc77f7 | 1357 | |
0a13c00e TH |
1358 | /** |
1359 | * queue_delayed_work - queue work on a workqueue after delay | |
1360 | * @wq: workqueue to use | |
1361 | * @dwork: delayable work to queue | |
1362 | * @delay: number of jiffies to wait before queueing | |
1363 | * | |
715f1300 | 1364 | * Equivalent to queue_delayed_work_on() but tries to use the local CPU. |
0a13c00e | 1365 | */ |
d4283e93 | 1366 | bool queue_delayed_work(struct workqueue_struct *wq, |
0a13c00e TH |
1367 | struct delayed_work *dwork, unsigned long delay) |
1368 | { | |
57469821 | 1369 | return queue_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); |
0a13c00e TH |
1370 | } |
1371 | EXPORT_SYMBOL_GPL(queue_delayed_work); | |
c7fc77f7 | 1372 | |
8376fe22 TH |
1373 | /** |
1374 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU | |
1375 | * @cpu: CPU number to execute work on | |
1376 | * @wq: workqueue to use | |
1377 | * @dwork: work to queue | |
1378 | * @delay: number of jiffies to wait before queueing | |
1379 | * | |
1380 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, | |
1381 | * modify @dwork's timer so that it expires after @delay. If @delay is | |
1382 | * zero, @work is guaranteed to be scheduled immediately regardless of its | |
1383 | * current state. | |
1384 | * | |
1385 | * Returns %false if @dwork was idle and queued, %true if @dwork was | |
1386 | * pending and its timer was modified. | |
1387 | * | |
e0aecdd8 | 1388 | * This function is safe to call from any context including IRQ handler. |
8376fe22 TH |
1389 | * See try_to_grab_pending() for details. |
1390 | */ | |
1391 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
1392 | struct delayed_work *dwork, unsigned long delay) | |
1393 | { | |
1394 | unsigned long flags; | |
1395 | int ret; | |
c7fc77f7 | 1396 | |
8376fe22 TH |
1397 | do { |
1398 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
1399 | } while (unlikely(ret == -EAGAIN)); | |
63bc0362 | 1400 | |
8376fe22 TH |
1401 | if (likely(ret >= 0)) { |
1402 | __queue_delayed_work(cpu, wq, dwork, delay); | |
1403 | local_irq_restore(flags); | |
7a6bc1cd | 1404 | } |
8376fe22 TH |
1405 | |
1406 | /* -ENOENT from try_to_grab_pending() becomes %true */ | |
7a6bc1cd VP |
1407 | return ret; |
1408 | } | |
8376fe22 TH |
1409 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
1410 | ||
1411 | /** | |
1412 | * mod_delayed_work - modify delay of or queue a delayed work | |
1413 | * @wq: workqueue to use | |
1414 | * @dwork: work to queue | |
1415 | * @delay: number of jiffies to wait before queueing | |
1416 | * | |
1417 | * mod_delayed_work_on() on local CPU. | |
1418 | */ | |
1419 | bool mod_delayed_work(struct workqueue_struct *wq, struct delayed_work *dwork, | |
1420 | unsigned long delay) | |
1421 | { | |
1422 | return mod_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); | |
1423 | } | |
1424 | EXPORT_SYMBOL_GPL(mod_delayed_work); | |
1da177e4 | 1425 | |
c8e55f36 TH |
1426 | /** |
1427 | * worker_enter_idle - enter idle state | |
1428 | * @worker: worker which is entering idle state | |
1429 | * | |
1430 | * @worker is entering idle state. Update stats and idle timer if | |
1431 | * necessary. | |
1432 | * | |
1433 | * LOCKING: | |
d565ed63 | 1434 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1435 | */ |
1436 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1437 | { |
bd7bdd43 | 1438 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1439 | |
6183c009 TH |
1440 | if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) || |
1441 | WARN_ON_ONCE(!list_empty(&worker->entry) && | |
1442 | (worker->hentry.next || worker->hentry.pprev))) | |
1443 | return; | |
c8e55f36 | 1444 | |
cb444766 TH |
1445 | /* can't use worker_set_flags(), also called from start_worker() */ |
1446 | worker->flags |= WORKER_IDLE; | |
bd7bdd43 | 1447 | pool->nr_idle++; |
e22bee78 | 1448 | worker->last_active = jiffies; |
c8e55f36 TH |
1449 | |
1450 | /* idle_list is LIFO */ | |
bd7bdd43 | 1451 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1452 | |
628c78e7 TH |
1453 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
1454 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); | |
cb444766 | 1455 | |
544ecf31 | 1456 | /* |
706026c2 | 1457 | * Sanity check nr_running. Because wq_unbind_fn() releases |
d565ed63 | 1458 | * pool->lock between setting %WORKER_UNBOUND and zapping |
628c78e7 TH |
1459 | * nr_running, the warning may trigger spuriously. Check iff |
1460 | * unbind is not in progress. | |
544ecf31 | 1461 | */ |
24647570 | 1462 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
bd7bdd43 | 1463 | pool->nr_workers == pool->nr_idle && |
e19e397a | 1464 | atomic_read(&pool->nr_running)); |
c8e55f36 TH |
1465 | } |
1466 | ||
1467 | /** | |
1468 | * worker_leave_idle - leave idle state | |
1469 | * @worker: worker which is leaving idle state | |
1470 | * | |
1471 | * @worker is leaving idle state. Update stats. | |
1472 | * | |
1473 | * LOCKING: | |
d565ed63 | 1474 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1475 | */ |
1476 | static void worker_leave_idle(struct worker *worker) | |
1477 | { | |
bd7bdd43 | 1478 | struct worker_pool *pool = worker->pool; |
c8e55f36 | 1479 | |
6183c009 TH |
1480 | if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE))) |
1481 | return; | |
d302f017 | 1482 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1483 | pool->nr_idle--; |
c8e55f36 TH |
1484 | list_del_init(&worker->entry); |
1485 | } | |
1486 | ||
e22bee78 | 1487 | /** |
f36dc67b LJ |
1488 | * worker_maybe_bind_and_lock - try to bind %current to worker_pool and lock it |
1489 | * @pool: target worker_pool | |
1490 | * | |
1491 | * Bind %current to the cpu of @pool if it is associated and lock @pool. | |
e22bee78 TH |
1492 | * |
1493 | * Works which are scheduled while the cpu is online must at least be | |
1494 | * scheduled to a worker which is bound to the cpu so that if they are | |
1495 | * flushed from cpu callbacks while cpu is going down, they are | |
1496 | * guaranteed to execute on the cpu. | |
1497 | * | |
f5faa077 | 1498 | * This function is to be used by unbound workers and rescuers to bind |
e22bee78 TH |
1499 | * themselves to the target cpu and may race with cpu going down or |
1500 | * coming online. kthread_bind() can't be used because it may put the | |
1501 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
706026c2 | 1502 | * verbatim as it's best effort and blocking and pool may be |
e22bee78 TH |
1503 | * [dis]associated in the meantime. |
1504 | * | |
706026c2 | 1505 | * This function tries set_cpus_allowed() and locks pool and verifies the |
24647570 | 1506 | * binding against %POOL_DISASSOCIATED which is set during |
f2d5a0ee TH |
1507 | * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker |
1508 | * enters idle state or fetches works without dropping lock, it can | |
1509 | * guarantee the scheduling requirement described in the first paragraph. | |
e22bee78 TH |
1510 | * |
1511 | * CONTEXT: | |
d565ed63 | 1512 | * Might sleep. Called without any lock but returns with pool->lock |
e22bee78 TH |
1513 | * held. |
1514 | * | |
1515 | * RETURNS: | |
706026c2 | 1516 | * %true if the associated pool is online (@worker is successfully |
e22bee78 TH |
1517 | * bound), %false if offline. |
1518 | */ | |
f36dc67b | 1519 | static bool worker_maybe_bind_and_lock(struct worker_pool *pool) |
d565ed63 | 1520 | __acquires(&pool->lock) |
e22bee78 | 1521 | { |
e22bee78 | 1522 | while (true) { |
4e6045f1 | 1523 | /* |
e22bee78 TH |
1524 | * The following call may fail, succeed or succeed |
1525 | * without actually migrating the task to the cpu if | |
1526 | * it races with cpu hotunplug operation. Verify | |
24647570 | 1527 | * against POOL_DISASSOCIATED. |
4e6045f1 | 1528 | */ |
24647570 | 1529 | if (!(pool->flags & POOL_DISASSOCIATED)) |
7a4e344c | 1530 | set_cpus_allowed_ptr(current, pool->attrs->cpumask); |
e22bee78 | 1531 | |
d565ed63 | 1532 | spin_lock_irq(&pool->lock); |
24647570 | 1533 | if (pool->flags & POOL_DISASSOCIATED) |
e22bee78 | 1534 | return false; |
f5faa077 | 1535 | if (task_cpu(current) == pool->cpu && |
7a4e344c | 1536 | cpumask_equal(¤t->cpus_allowed, pool->attrs->cpumask)) |
e22bee78 | 1537 | return true; |
d565ed63 | 1538 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1539 | |
5035b20f TH |
1540 | /* |
1541 | * We've raced with CPU hot[un]plug. Give it a breather | |
1542 | * and retry migration. cond_resched() is required here; | |
1543 | * otherwise, we might deadlock against cpu_stop trying to | |
1544 | * bring down the CPU on non-preemptive kernel. | |
1545 | */ | |
e22bee78 | 1546 | cpu_relax(); |
5035b20f | 1547 | cond_resched(); |
e22bee78 TH |
1548 | } |
1549 | } | |
1550 | ||
25511a47 | 1551 | /* |
ea1abd61 | 1552 | * Rebind an idle @worker to its CPU. worker_thread() will test |
5f7dabfd | 1553 | * list_empty(@worker->entry) before leaving idle and call this function. |
25511a47 TH |
1554 | */ |
1555 | static void idle_worker_rebind(struct worker *worker) | |
1556 | { | |
5f7dabfd | 1557 | /* CPU may go down again inbetween, clear UNBOUND only on success */ |
f36dc67b | 1558 | if (worker_maybe_bind_and_lock(worker->pool)) |
5f7dabfd | 1559 | worker_clr_flags(worker, WORKER_UNBOUND); |
25511a47 | 1560 | |
ea1abd61 LJ |
1561 | /* rebind complete, become available again */ |
1562 | list_add(&worker->entry, &worker->pool->idle_list); | |
d565ed63 | 1563 | spin_unlock_irq(&worker->pool->lock); |
25511a47 TH |
1564 | } |
1565 | ||
e22bee78 | 1566 | /* |
25511a47 | 1567 | * Function for @worker->rebind.work used to rebind unbound busy workers to |
403c821d TH |
1568 | * the associated cpu which is coming back online. This is scheduled by |
1569 | * cpu up but can race with other cpu hotplug operations and may be | |
1570 | * executed twice without intervening cpu down. | |
e22bee78 | 1571 | */ |
25511a47 | 1572 | static void busy_worker_rebind_fn(struct work_struct *work) |
e22bee78 TH |
1573 | { |
1574 | struct worker *worker = container_of(work, struct worker, rebind_work); | |
e22bee78 | 1575 | |
f36dc67b | 1576 | if (worker_maybe_bind_and_lock(worker->pool)) |
eab6d828 | 1577 | worker_clr_flags(worker, WORKER_UNBOUND); |
e22bee78 | 1578 | |
d565ed63 | 1579 | spin_unlock_irq(&worker->pool->lock); |
e22bee78 TH |
1580 | } |
1581 | ||
25511a47 | 1582 | /** |
94cf58bb TH |
1583 | * rebind_workers - rebind all workers of a pool to the associated CPU |
1584 | * @pool: pool of interest | |
25511a47 | 1585 | * |
94cf58bb | 1586 | * @pool->cpu is coming online. Rebind all workers to the CPU. Rebinding |
25511a47 TH |
1587 | * is different for idle and busy ones. |
1588 | * | |
ea1abd61 LJ |
1589 | * Idle ones will be removed from the idle_list and woken up. They will |
1590 | * add themselves back after completing rebind. This ensures that the | |
1591 | * idle_list doesn't contain any unbound workers when re-bound busy workers | |
1592 | * try to perform local wake-ups for concurrency management. | |
25511a47 | 1593 | * |
ea1abd61 LJ |
1594 | * Busy workers can rebind after they finish their current work items. |
1595 | * Queueing the rebind work item at the head of the scheduled list is | |
1596 | * enough. Note that nr_running will be properly bumped as busy workers | |
1597 | * rebind. | |
25511a47 | 1598 | * |
ea1abd61 LJ |
1599 | * On return, all non-manager workers are scheduled for rebind - see |
1600 | * manage_workers() for the manager special case. Any idle worker | |
1601 | * including the manager will not appear on @idle_list until rebind is | |
1602 | * complete, making local wake-ups safe. | |
25511a47 | 1603 | */ |
94cf58bb | 1604 | static void rebind_workers(struct worker_pool *pool) |
25511a47 | 1605 | { |
ea1abd61 | 1606 | struct worker *worker, *n; |
25511a47 TH |
1607 | int i; |
1608 | ||
94cf58bb TH |
1609 | lockdep_assert_held(&pool->assoc_mutex); |
1610 | lockdep_assert_held(&pool->lock); | |
25511a47 | 1611 | |
5f7dabfd | 1612 | /* dequeue and kick idle ones */ |
94cf58bb TH |
1613 | list_for_each_entry_safe(worker, n, &pool->idle_list, entry) { |
1614 | /* | |
1615 | * idle workers should be off @pool->idle_list until rebind | |
1616 | * is complete to avoid receiving premature local wake-ups. | |
1617 | */ | |
1618 | list_del_init(&worker->entry); | |
25511a47 | 1619 | |
94cf58bb TH |
1620 | /* |
1621 | * worker_thread() will see the above dequeuing and call | |
1622 | * idle_worker_rebind(). | |
1623 | */ | |
1624 | wake_up_process(worker->task); | |
1625 | } | |
25511a47 | 1626 | |
94cf58bb | 1627 | /* rebind busy workers */ |
b67bfe0d | 1628 | for_each_busy_worker(worker, i, pool) { |
94cf58bb TH |
1629 | struct work_struct *rebind_work = &worker->rebind_work; |
1630 | struct workqueue_struct *wq; | |
25511a47 | 1631 | |
94cf58bb TH |
1632 | if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, |
1633 | work_data_bits(rebind_work))) | |
1634 | continue; | |
25511a47 | 1635 | |
94cf58bb | 1636 | debug_work_activate(rebind_work); |
90beca5d | 1637 | |
94cf58bb TH |
1638 | /* |
1639 | * wq doesn't really matter but let's keep @worker->pool | |
112202d9 | 1640 | * and @pwq->pool consistent for sanity. |
94cf58bb | 1641 | */ |
7a4e344c | 1642 | if (worker->pool->attrs->nice < 0) |
94cf58bb TH |
1643 | wq = system_highpri_wq; |
1644 | else | |
1645 | wq = system_wq; | |
1646 | ||
7fb98ea7 | 1647 | insert_work(per_cpu_ptr(wq->cpu_pwqs, pool->cpu), rebind_work, |
94cf58bb TH |
1648 | worker->scheduled.next, |
1649 | work_color_to_flags(WORK_NO_COLOR)); | |
ec58815a | 1650 | } |
25511a47 TH |
1651 | } |
1652 | ||
c34056a3 TH |
1653 | static struct worker *alloc_worker(void) |
1654 | { | |
1655 | struct worker *worker; | |
1656 | ||
1657 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1658 | if (worker) { |
1659 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1660 | INIT_LIST_HEAD(&worker->scheduled); |
25511a47 | 1661 | INIT_WORK(&worker->rebind_work, busy_worker_rebind_fn); |
e22bee78 TH |
1662 | /* on creation a worker is in !idle && prep state */ |
1663 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1664 | } |
c34056a3 TH |
1665 | return worker; |
1666 | } | |
1667 | ||
1668 | /** | |
1669 | * create_worker - create a new workqueue worker | |
63d95a91 | 1670 | * @pool: pool the new worker will belong to |
c34056a3 | 1671 | * |
63d95a91 | 1672 | * Create a new worker which is bound to @pool. The returned worker |
c34056a3 TH |
1673 | * can be started by calling start_worker() or destroyed using |
1674 | * destroy_worker(). | |
1675 | * | |
1676 | * CONTEXT: | |
1677 | * Might sleep. Does GFP_KERNEL allocations. | |
1678 | * | |
1679 | * RETURNS: | |
1680 | * Pointer to the newly created worker. | |
1681 | */ | |
bc2ae0f5 | 1682 | static struct worker *create_worker(struct worker_pool *pool) |
c34056a3 | 1683 | { |
7a4e344c | 1684 | const char *pri = pool->attrs->nice < 0 ? "H" : ""; |
c34056a3 | 1685 | struct worker *worker = NULL; |
f3421797 | 1686 | int id = -1; |
c34056a3 | 1687 | |
d565ed63 | 1688 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1689 | while (ida_get_new(&pool->worker_ida, &id)) { |
d565ed63 | 1690 | spin_unlock_irq(&pool->lock); |
bd7bdd43 | 1691 | if (!ida_pre_get(&pool->worker_ida, GFP_KERNEL)) |
c34056a3 | 1692 | goto fail; |
d565ed63 | 1693 | spin_lock_irq(&pool->lock); |
c34056a3 | 1694 | } |
d565ed63 | 1695 | spin_unlock_irq(&pool->lock); |
c34056a3 TH |
1696 | |
1697 | worker = alloc_worker(); | |
1698 | if (!worker) | |
1699 | goto fail; | |
1700 | ||
bd7bdd43 | 1701 | worker->pool = pool; |
c34056a3 TH |
1702 | worker->id = id; |
1703 | ||
29c91e99 | 1704 | if (pool->cpu >= 0) |
94dcf29a | 1705 | worker->task = kthread_create_on_node(worker_thread, |
ec22ca5e | 1706 | worker, cpu_to_node(pool->cpu), |
d84ff051 | 1707 | "kworker/%d:%d%s", pool->cpu, id, pri); |
f3421797 TH |
1708 | else |
1709 | worker->task = kthread_create(worker_thread, worker, | |
3270476a | 1710 | "kworker/u:%d%s", id, pri); |
c34056a3 TH |
1711 | if (IS_ERR(worker->task)) |
1712 | goto fail; | |
1713 | ||
7a4e344c TH |
1714 | set_user_nice(worker->task, pool->attrs->nice); |
1715 | set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask); | |
3270476a | 1716 | |
db7bccf4 | 1717 | /* |
7a4e344c TH |
1718 | * %PF_THREAD_BOUND is used to prevent userland from meddling with |
1719 | * cpumask of workqueue workers. This is an abuse. We need | |
1720 | * %PF_NO_SETAFFINITY. | |
db7bccf4 | 1721 | */ |
7a4e344c TH |
1722 | worker->task->flags |= PF_THREAD_BOUND; |
1723 | ||
1724 | /* | |
1725 | * The caller is responsible for ensuring %POOL_DISASSOCIATED | |
1726 | * remains stable across this function. See the comments above the | |
1727 | * flag definition for details. | |
1728 | */ | |
1729 | if (pool->flags & POOL_DISASSOCIATED) | |
bc2ae0f5 | 1730 | worker->flags |= WORKER_UNBOUND; |
c34056a3 TH |
1731 | |
1732 | return worker; | |
1733 | fail: | |
1734 | if (id >= 0) { | |
d565ed63 | 1735 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1736 | ida_remove(&pool->worker_ida, id); |
d565ed63 | 1737 | spin_unlock_irq(&pool->lock); |
c34056a3 TH |
1738 | } |
1739 | kfree(worker); | |
1740 | return NULL; | |
1741 | } | |
1742 | ||
1743 | /** | |
1744 | * start_worker - start a newly created worker | |
1745 | * @worker: worker to start | |
1746 | * | |
706026c2 | 1747 | * Make the pool aware of @worker and start it. |
c34056a3 TH |
1748 | * |
1749 | * CONTEXT: | |
d565ed63 | 1750 | * spin_lock_irq(pool->lock). |
c34056a3 TH |
1751 | */ |
1752 | static void start_worker(struct worker *worker) | |
1753 | { | |
cb444766 | 1754 | worker->flags |= WORKER_STARTED; |
bd7bdd43 | 1755 | worker->pool->nr_workers++; |
c8e55f36 | 1756 | worker_enter_idle(worker); |
c34056a3 TH |
1757 | wake_up_process(worker->task); |
1758 | } | |
1759 | ||
1760 | /** | |
1761 | * destroy_worker - destroy a workqueue worker | |
1762 | * @worker: worker to be destroyed | |
1763 | * | |
706026c2 | 1764 | * Destroy @worker and adjust @pool stats accordingly. |
c8e55f36 TH |
1765 | * |
1766 | * CONTEXT: | |
d565ed63 | 1767 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
c34056a3 TH |
1768 | */ |
1769 | static void destroy_worker(struct worker *worker) | |
1770 | { | |
bd7bdd43 | 1771 | struct worker_pool *pool = worker->pool; |
c34056a3 TH |
1772 | int id = worker->id; |
1773 | ||
1774 | /* sanity check frenzy */ | |
6183c009 TH |
1775 | if (WARN_ON(worker->current_work) || |
1776 | WARN_ON(!list_empty(&worker->scheduled))) | |
1777 | return; | |
c34056a3 | 1778 | |
c8e55f36 | 1779 | if (worker->flags & WORKER_STARTED) |
bd7bdd43 | 1780 | pool->nr_workers--; |
c8e55f36 | 1781 | if (worker->flags & WORKER_IDLE) |
bd7bdd43 | 1782 | pool->nr_idle--; |
c8e55f36 TH |
1783 | |
1784 | list_del_init(&worker->entry); | |
cb444766 | 1785 | worker->flags |= WORKER_DIE; |
c8e55f36 | 1786 | |
d565ed63 | 1787 | spin_unlock_irq(&pool->lock); |
c8e55f36 | 1788 | |
c34056a3 TH |
1789 | kthread_stop(worker->task); |
1790 | kfree(worker); | |
1791 | ||
d565ed63 | 1792 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1793 | ida_remove(&pool->worker_ida, id); |
c34056a3 TH |
1794 | } |
1795 | ||
63d95a91 | 1796 | static void idle_worker_timeout(unsigned long __pool) |
e22bee78 | 1797 | { |
63d95a91 | 1798 | struct worker_pool *pool = (void *)__pool; |
e22bee78 | 1799 | |
d565ed63 | 1800 | spin_lock_irq(&pool->lock); |
e22bee78 | 1801 | |
63d95a91 | 1802 | if (too_many_workers(pool)) { |
e22bee78 TH |
1803 | struct worker *worker; |
1804 | unsigned long expires; | |
1805 | ||
1806 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1807 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1808 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1809 | ||
1810 | if (time_before(jiffies, expires)) | |
63d95a91 | 1811 | mod_timer(&pool->idle_timer, expires); |
e22bee78 TH |
1812 | else { |
1813 | /* it's been idle for too long, wake up manager */ | |
11ebea50 | 1814 | pool->flags |= POOL_MANAGE_WORKERS; |
63d95a91 | 1815 | wake_up_worker(pool); |
d5abe669 | 1816 | } |
e22bee78 TH |
1817 | } |
1818 | ||
d565ed63 | 1819 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1820 | } |
d5abe669 | 1821 | |
493a1724 | 1822 | static void send_mayday(struct work_struct *work) |
e22bee78 | 1823 | { |
112202d9 TH |
1824 | struct pool_workqueue *pwq = get_work_pwq(work); |
1825 | struct workqueue_struct *wq = pwq->wq; | |
493a1724 TH |
1826 | |
1827 | lockdep_assert_held(&workqueue_lock); | |
e22bee78 TH |
1828 | |
1829 | if (!(wq->flags & WQ_RESCUER)) | |
493a1724 | 1830 | return; |
e22bee78 TH |
1831 | |
1832 | /* mayday mayday mayday */ | |
493a1724 TH |
1833 | if (list_empty(&pwq->mayday_node)) { |
1834 | list_add_tail(&pwq->mayday_node, &wq->maydays); | |
e22bee78 | 1835 | wake_up_process(wq->rescuer->task); |
493a1724 | 1836 | } |
e22bee78 TH |
1837 | } |
1838 | ||
706026c2 | 1839 | static void pool_mayday_timeout(unsigned long __pool) |
e22bee78 | 1840 | { |
63d95a91 | 1841 | struct worker_pool *pool = (void *)__pool; |
e22bee78 TH |
1842 | struct work_struct *work; |
1843 | ||
493a1724 TH |
1844 | spin_lock_irq(&workqueue_lock); /* for wq->maydays */ |
1845 | spin_lock(&pool->lock); | |
e22bee78 | 1846 | |
63d95a91 | 1847 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
1848 | /* |
1849 | * We've been trying to create a new worker but | |
1850 | * haven't been successful. We might be hitting an | |
1851 | * allocation deadlock. Send distress signals to | |
1852 | * rescuers. | |
1853 | */ | |
63d95a91 | 1854 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 1855 | send_mayday(work); |
1da177e4 | 1856 | } |
e22bee78 | 1857 | |
493a1724 TH |
1858 | spin_unlock(&pool->lock); |
1859 | spin_unlock_irq(&workqueue_lock); | |
e22bee78 | 1860 | |
63d95a91 | 1861 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
1862 | } |
1863 | ||
e22bee78 TH |
1864 | /** |
1865 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 1866 | * @pool: pool to create a new worker for |
e22bee78 | 1867 | * |
63d95a91 | 1868 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
1869 | * have at least one idle worker on return from this function. If |
1870 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 1871 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
1872 | * possible allocation deadlock. |
1873 | * | |
1874 | * On return, need_to_create_worker() is guaranteed to be false and | |
1875 | * may_start_working() true. | |
1876 | * | |
1877 | * LOCKING: | |
d565ed63 | 1878 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1879 | * multiple times. Does GFP_KERNEL allocations. Called only from |
1880 | * manager. | |
1881 | * | |
1882 | * RETURNS: | |
d565ed63 | 1883 | * false if no action was taken and pool->lock stayed locked, true |
e22bee78 TH |
1884 | * otherwise. |
1885 | */ | |
63d95a91 | 1886 | static bool maybe_create_worker(struct worker_pool *pool) |
d565ed63 TH |
1887 | __releases(&pool->lock) |
1888 | __acquires(&pool->lock) | |
1da177e4 | 1889 | { |
63d95a91 | 1890 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1891 | return false; |
1892 | restart: | |
d565ed63 | 1893 | spin_unlock_irq(&pool->lock); |
9f9c2364 | 1894 | |
e22bee78 | 1895 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 1896 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
1897 | |
1898 | while (true) { | |
1899 | struct worker *worker; | |
1900 | ||
bc2ae0f5 | 1901 | worker = create_worker(pool); |
e22bee78 | 1902 | if (worker) { |
63d95a91 | 1903 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1904 | spin_lock_irq(&pool->lock); |
e22bee78 | 1905 | start_worker(worker); |
6183c009 TH |
1906 | if (WARN_ON_ONCE(need_to_create_worker(pool))) |
1907 | goto restart; | |
e22bee78 TH |
1908 | return true; |
1909 | } | |
1910 | ||
63d95a91 | 1911 | if (!need_to_create_worker(pool)) |
e22bee78 | 1912 | break; |
1da177e4 | 1913 | |
e22bee78 TH |
1914 | __set_current_state(TASK_INTERRUPTIBLE); |
1915 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1916 | |
63d95a91 | 1917 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1918 | break; |
1919 | } | |
1920 | ||
63d95a91 | 1921 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1922 | spin_lock_irq(&pool->lock); |
63d95a91 | 1923 | if (need_to_create_worker(pool)) |
e22bee78 TH |
1924 | goto restart; |
1925 | return true; | |
1926 | } | |
1927 | ||
1928 | /** | |
1929 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
63d95a91 | 1930 | * @pool: pool to destroy workers for |
e22bee78 | 1931 | * |
63d95a91 | 1932 | * Destroy @pool workers which have been idle for longer than |
e22bee78 TH |
1933 | * IDLE_WORKER_TIMEOUT. |
1934 | * | |
1935 | * LOCKING: | |
d565ed63 | 1936 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1937 | * multiple times. Called only from manager. |
1938 | * | |
1939 | * RETURNS: | |
d565ed63 | 1940 | * false if no action was taken and pool->lock stayed locked, true |
e22bee78 TH |
1941 | * otherwise. |
1942 | */ | |
63d95a91 | 1943 | static bool maybe_destroy_workers(struct worker_pool *pool) |
e22bee78 TH |
1944 | { |
1945 | bool ret = false; | |
1da177e4 | 1946 | |
63d95a91 | 1947 | while (too_many_workers(pool)) { |
e22bee78 TH |
1948 | struct worker *worker; |
1949 | unsigned long expires; | |
3af24433 | 1950 | |
63d95a91 | 1951 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 | 1952 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
85f4186a | 1953 | |
e22bee78 | 1954 | if (time_before(jiffies, expires)) { |
63d95a91 | 1955 | mod_timer(&pool->idle_timer, expires); |
3af24433 | 1956 | break; |
e22bee78 | 1957 | } |
1da177e4 | 1958 | |
e22bee78 TH |
1959 | destroy_worker(worker); |
1960 | ret = true; | |
1da177e4 | 1961 | } |
1e19ffc6 | 1962 | |
e22bee78 | 1963 | return ret; |
1e19ffc6 TH |
1964 | } |
1965 | ||
73f53c4a | 1966 | /** |
e22bee78 TH |
1967 | * manage_workers - manage worker pool |
1968 | * @worker: self | |
73f53c4a | 1969 | * |
706026c2 | 1970 | * Assume the manager role and manage the worker pool @worker belongs |
e22bee78 | 1971 | * to. At any given time, there can be only zero or one manager per |
706026c2 | 1972 | * pool. The exclusion is handled automatically by this function. |
e22bee78 TH |
1973 | * |
1974 | * The caller can safely start processing works on false return. On | |
1975 | * true return, it's guaranteed that need_to_create_worker() is false | |
1976 | * and may_start_working() is true. | |
73f53c4a TH |
1977 | * |
1978 | * CONTEXT: | |
d565ed63 | 1979 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1980 | * multiple times. Does GFP_KERNEL allocations. |
1981 | * | |
1982 | * RETURNS: | |
d565ed63 TH |
1983 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
1984 | * multiple times. Does GFP_KERNEL allocations. | |
73f53c4a | 1985 | */ |
e22bee78 | 1986 | static bool manage_workers(struct worker *worker) |
73f53c4a | 1987 | { |
63d95a91 | 1988 | struct worker_pool *pool = worker->pool; |
e22bee78 | 1989 | bool ret = false; |
73f53c4a | 1990 | |
34a06bd6 | 1991 | if (!mutex_trylock(&pool->manager_arb)) |
e22bee78 | 1992 | return ret; |
1e19ffc6 | 1993 | |
ee378aa4 LJ |
1994 | /* |
1995 | * To simplify both worker management and CPU hotplug, hold off | |
1996 | * management while hotplug is in progress. CPU hotplug path can't | |
34a06bd6 TH |
1997 | * grab @pool->manager_arb to achieve this because that can lead to |
1998 | * idle worker depletion (all become busy thinking someone else is | |
1999 | * managing) which in turn can result in deadlock under extreme | |
2000 | * circumstances. Use @pool->assoc_mutex to synchronize manager | |
2001 | * against CPU hotplug. | |
ee378aa4 | 2002 | * |
b2eb83d1 | 2003 | * assoc_mutex would always be free unless CPU hotplug is in |
d565ed63 | 2004 | * progress. trylock first without dropping @pool->lock. |
ee378aa4 | 2005 | */ |
b2eb83d1 | 2006 | if (unlikely(!mutex_trylock(&pool->assoc_mutex))) { |
d565ed63 | 2007 | spin_unlock_irq(&pool->lock); |
b2eb83d1 | 2008 | mutex_lock(&pool->assoc_mutex); |
ee378aa4 LJ |
2009 | /* |
2010 | * CPU hotplug could have happened while we were waiting | |
b2eb83d1 | 2011 | * for assoc_mutex. Hotplug itself can't handle us |
ee378aa4 | 2012 | * because manager isn't either on idle or busy list, and |
706026c2 | 2013 | * @pool's state and ours could have deviated. |
ee378aa4 | 2014 | * |
b2eb83d1 | 2015 | * As hotplug is now excluded via assoc_mutex, we can |
ee378aa4 | 2016 | * simply try to bind. It will succeed or fail depending |
706026c2 | 2017 | * on @pool's current state. Try it and adjust |
ee378aa4 LJ |
2018 | * %WORKER_UNBOUND accordingly. |
2019 | */ | |
f36dc67b | 2020 | if (worker_maybe_bind_and_lock(pool)) |
ee378aa4 LJ |
2021 | worker->flags &= ~WORKER_UNBOUND; |
2022 | else | |
2023 | worker->flags |= WORKER_UNBOUND; | |
73f53c4a | 2024 | |
ee378aa4 LJ |
2025 | ret = true; |
2026 | } | |
73f53c4a | 2027 | |
11ebea50 | 2028 | pool->flags &= ~POOL_MANAGE_WORKERS; |
73f53c4a TH |
2029 | |
2030 | /* | |
e22bee78 TH |
2031 | * Destroy and then create so that may_start_working() is true |
2032 | * on return. | |
73f53c4a | 2033 | */ |
63d95a91 TH |
2034 | ret |= maybe_destroy_workers(pool); |
2035 | ret |= maybe_create_worker(pool); | |
e22bee78 | 2036 | |
b2eb83d1 | 2037 | mutex_unlock(&pool->assoc_mutex); |
34a06bd6 | 2038 | mutex_unlock(&pool->manager_arb); |
e22bee78 | 2039 | return ret; |
73f53c4a TH |
2040 | } |
2041 | ||
a62428c0 TH |
2042 | /** |
2043 | * process_one_work - process single work | |
c34056a3 | 2044 | * @worker: self |
a62428c0 TH |
2045 | * @work: work to process |
2046 | * | |
2047 | * Process @work. This function contains all the logics necessary to | |
2048 | * process a single work including synchronization against and | |
2049 | * interaction with other workers on the same cpu, queueing and | |
2050 | * flushing. As long as context requirement is met, any worker can | |
2051 | * call this function to process a work. | |
2052 | * | |
2053 | * CONTEXT: | |
d565ed63 | 2054 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
a62428c0 | 2055 | */ |
c34056a3 | 2056 | static void process_one_work(struct worker *worker, struct work_struct *work) |
d565ed63 TH |
2057 | __releases(&pool->lock) |
2058 | __acquires(&pool->lock) | |
a62428c0 | 2059 | { |
112202d9 | 2060 | struct pool_workqueue *pwq = get_work_pwq(work); |
bd7bdd43 | 2061 | struct worker_pool *pool = worker->pool; |
112202d9 | 2062 | bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE; |
73f53c4a | 2063 | int work_color; |
7e11629d | 2064 | struct worker *collision; |
a62428c0 TH |
2065 | #ifdef CONFIG_LOCKDEP |
2066 | /* | |
2067 | * It is permissible to free the struct work_struct from | |
2068 | * inside the function that is called from it, this we need to | |
2069 | * take into account for lockdep too. To avoid bogus "held | |
2070 | * lock freed" warnings as well as problems when looking into | |
2071 | * work->lockdep_map, make a copy and use that here. | |
2072 | */ | |
4d82a1de PZ |
2073 | struct lockdep_map lockdep_map; |
2074 | ||
2075 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 2076 | #endif |
6fec10a1 TH |
2077 | /* |
2078 | * Ensure we're on the correct CPU. DISASSOCIATED test is | |
2079 | * necessary to avoid spurious warnings from rescuers servicing the | |
24647570 | 2080 | * unbound or a disassociated pool. |
6fec10a1 | 2081 | */ |
5f7dabfd | 2082 | WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) && |
24647570 | 2083 | !(pool->flags & POOL_DISASSOCIATED) && |
ec22ca5e | 2084 | raw_smp_processor_id() != pool->cpu); |
25511a47 | 2085 | |
7e11629d TH |
2086 | /* |
2087 | * A single work shouldn't be executed concurrently by | |
2088 | * multiple workers on a single cpu. Check whether anyone is | |
2089 | * already processing the work. If so, defer the work to the | |
2090 | * currently executing one. | |
2091 | */ | |
c9e7cf27 | 2092 | collision = find_worker_executing_work(pool, work); |
7e11629d TH |
2093 | if (unlikely(collision)) { |
2094 | move_linked_works(work, &collision->scheduled, NULL); | |
2095 | return; | |
2096 | } | |
2097 | ||
8930caba | 2098 | /* claim and dequeue */ |
a62428c0 | 2099 | debug_work_deactivate(work); |
c9e7cf27 | 2100 | hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work); |
c34056a3 | 2101 | worker->current_work = work; |
a2c1c57b | 2102 | worker->current_func = work->func; |
112202d9 | 2103 | worker->current_pwq = pwq; |
73f53c4a | 2104 | work_color = get_work_color(work); |
7a22ad75 | 2105 | |
a62428c0 TH |
2106 | list_del_init(&work->entry); |
2107 | ||
fb0e7beb TH |
2108 | /* |
2109 | * CPU intensive works don't participate in concurrency | |
2110 | * management. They're the scheduler's responsibility. | |
2111 | */ | |
2112 | if (unlikely(cpu_intensive)) | |
2113 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
2114 | ||
974271c4 | 2115 | /* |
d565ed63 | 2116 | * Unbound pool isn't concurrency managed and work items should be |
974271c4 TH |
2117 | * executed ASAP. Wake up another worker if necessary. |
2118 | */ | |
63d95a91 TH |
2119 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
2120 | wake_up_worker(pool); | |
974271c4 | 2121 | |
8930caba | 2122 | /* |
7c3eed5c | 2123 | * Record the last pool and clear PENDING which should be the last |
d565ed63 | 2124 | * update to @work. Also, do this inside @pool->lock so that |
23657bb1 TH |
2125 | * PENDING and queued state changes happen together while IRQ is |
2126 | * disabled. | |
8930caba | 2127 | */ |
7c3eed5c | 2128 | set_work_pool_and_clear_pending(work, pool->id); |
a62428c0 | 2129 | |
d565ed63 | 2130 | spin_unlock_irq(&pool->lock); |
a62428c0 | 2131 | |
112202d9 | 2132 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
a62428c0 | 2133 | lock_map_acquire(&lockdep_map); |
e36c886a | 2134 | trace_workqueue_execute_start(work); |
a2c1c57b | 2135 | worker->current_func(work); |
e36c886a AV |
2136 | /* |
2137 | * While we must be careful to not use "work" after this, the trace | |
2138 | * point will only record its address. | |
2139 | */ | |
2140 | trace_workqueue_execute_end(work); | |
a62428c0 | 2141 | lock_map_release(&lockdep_map); |
112202d9 | 2142 | lock_map_release(&pwq->wq->lockdep_map); |
a62428c0 TH |
2143 | |
2144 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
044c782c VI |
2145 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
2146 | " last function: %pf\n", | |
a2c1c57b TH |
2147 | current->comm, preempt_count(), task_pid_nr(current), |
2148 | worker->current_func); | |
a62428c0 TH |
2149 | debug_show_held_locks(current); |
2150 | dump_stack(); | |
2151 | } | |
2152 | ||
d565ed63 | 2153 | spin_lock_irq(&pool->lock); |
a62428c0 | 2154 | |
fb0e7beb TH |
2155 | /* clear cpu intensive status */ |
2156 | if (unlikely(cpu_intensive)) | |
2157 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
2158 | ||
a62428c0 | 2159 | /* we're done with it, release */ |
42f8570f | 2160 | hash_del(&worker->hentry); |
c34056a3 | 2161 | worker->current_work = NULL; |
a2c1c57b | 2162 | worker->current_func = NULL; |
112202d9 TH |
2163 | worker->current_pwq = NULL; |
2164 | pwq_dec_nr_in_flight(pwq, work_color); | |
a62428c0 TH |
2165 | } |
2166 | ||
affee4b2 TH |
2167 | /** |
2168 | * process_scheduled_works - process scheduled works | |
2169 | * @worker: self | |
2170 | * | |
2171 | * Process all scheduled works. Please note that the scheduled list | |
2172 | * may change while processing a work, so this function repeatedly | |
2173 | * fetches a work from the top and executes it. | |
2174 | * | |
2175 | * CONTEXT: | |
d565ed63 | 2176 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
affee4b2 TH |
2177 | * multiple times. |
2178 | */ | |
2179 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 2180 | { |
affee4b2 TH |
2181 | while (!list_empty(&worker->scheduled)) { |
2182 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 2183 | struct work_struct, entry); |
c34056a3 | 2184 | process_one_work(worker, work); |
1da177e4 | 2185 | } |
1da177e4 LT |
2186 | } |
2187 | ||
4690c4ab TH |
2188 | /** |
2189 | * worker_thread - the worker thread function | |
c34056a3 | 2190 | * @__worker: self |
4690c4ab | 2191 | * |
706026c2 TH |
2192 | * The worker thread function. There are NR_CPU_WORKER_POOLS dynamic pools |
2193 | * of these per each cpu. These workers process all works regardless of | |
e22bee78 TH |
2194 | * their specific target workqueue. The only exception is works which |
2195 | * belong to workqueues with a rescuer which will be explained in | |
2196 | * rescuer_thread(). | |
4690c4ab | 2197 | */ |
c34056a3 | 2198 | static int worker_thread(void *__worker) |
1da177e4 | 2199 | { |
c34056a3 | 2200 | struct worker *worker = __worker; |
bd7bdd43 | 2201 | struct worker_pool *pool = worker->pool; |
1da177e4 | 2202 | |
e22bee78 TH |
2203 | /* tell the scheduler that this is a workqueue worker */ |
2204 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 2205 | woke_up: |
d565ed63 | 2206 | spin_lock_irq(&pool->lock); |
1da177e4 | 2207 | |
5f7dabfd LJ |
2208 | /* we are off idle list if destruction or rebind is requested */ |
2209 | if (unlikely(list_empty(&worker->entry))) { | |
d565ed63 | 2210 | spin_unlock_irq(&pool->lock); |
25511a47 | 2211 | |
5f7dabfd | 2212 | /* if DIE is set, destruction is requested */ |
25511a47 TH |
2213 | if (worker->flags & WORKER_DIE) { |
2214 | worker->task->flags &= ~PF_WQ_WORKER; | |
2215 | return 0; | |
2216 | } | |
2217 | ||
5f7dabfd | 2218 | /* otherwise, rebind */ |
25511a47 TH |
2219 | idle_worker_rebind(worker); |
2220 | goto woke_up; | |
c8e55f36 | 2221 | } |
affee4b2 | 2222 | |
c8e55f36 | 2223 | worker_leave_idle(worker); |
db7bccf4 | 2224 | recheck: |
e22bee78 | 2225 | /* no more worker necessary? */ |
63d95a91 | 2226 | if (!need_more_worker(pool)) |
e22bee78 TH |
2227 | goto sleep; |
2228 | ||
2229 | /* do we need to manage? */ | |
63d95a91 | 2230 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
2231 | goto recheck; |
2232 | ||
c8e55f36 TH |
2233 | /* |
2234 | * ->scheduled list can only be filled while a worker is | |
2235 | * preparing to process a work or actually processing it. | |
2236 | * Make sure nobody diddled with it while I was sleeping. | |
2237 | */ | |
6183c009 | 2238 | WARN_ON_ONCE(!list_empty(&worker->scheduled)); |
c8e55f36 | 2239 | |
e22bee78 TH |
2240 | /* |
2241 | * When control reaches this point, we're guaranteed to have | |
2242 | * at least one idle worker or that someone else has already | |
2243 | * assumed the manager role. | |
2244 | */ | |
2245 | worker_clr_flags(worker, WORKER_PREP); | |
2246 | ||
2247 | do { | |
c8e55f36 | 2248 | struct work_struct *work = |
bd7bdd43 | 2249 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2250 | struct work_struct, entry); |
2251 | ||
2252 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
2253 | /* optimization path, not strictly necessary */ | |
2254 | process_one_work(worker, work); | |
2255 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2256 | process_scheduled_works(worker); |
c8e55f36 TH |
2257 | } else { |
2258 | move_linked_works(work, &worker->scheduled, NULL); | |
2259 | process_scheduled_works(worker); | |
affee4b2 | 2260 | } |
63d95a91 | 2261 | } while (keep_working(pool)); |
e22bee78 TH |
2262 | |
2263 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 2264 | sleep: |
63d95a91 | 2265 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
e22bee78 | 2266 | goto recheck; |
d313dd85 | 2267 | |
c8e55f36 | 2268 | /* |
d565ed63 TH |
2269 | * pool->lock is held and there's no work to process and no need to |
2270 | * manage, sleep. Workers are woken up only while holding | |
2271 | * pool->lock or from local cpu, so setting the current state | |
2272 | * before releasing pool->lock is enough to prevent losing any | |
2273 | * event. | |
c8e55f36 TH |
2274 | */ |
2275 | worker_enter_idle(worker); | |
2276 | __set_current_state(TASK_INTERRUPTIBLE); | |
d565ed63 | 2277 | spin_unlock_irq(&pool->lock); |
c8e55f36 TH |
2278 | schedule(); |
2279 | goto woke_up; | |
1da177e4 LT |
2280 | } |
2281 | ||
e22bee78 TH |
2282 | /** |
2283 | * rescuer_thread - the rescuer thread function | |
111c225a | 2284 | * @__rescuer: self |
e22bee78 TH |
2285 | * |
2286 | * Workqueue rescuer thread function. There's one rescuer for each | |
2287 | * workqueue which has WQ_RESCUER set. | |
2288 | * | |
706026c2 | 2289 | * Regular work processing on a pool may block trying to create a new |
e22bee78 TH |
2290 | * worker which uses GFP_KERNEL allocation which has slight chance of |
2291 | * developing into deadlock if some works currently on the same queue | |
2292 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2293 | * the problem rescuer solves. | |
2294 | * | |
706026c2 TH |
2295 | * When such condition is possible, the pool summons rescuers of all |
2296 | * workqueues which have works queued on the pool and let them process | |
e22bee78 TH |
2297 | * those works so that forward progress can be guaranteed. |
2298 | * | |
2299 | * This should happen rarely. | |
2300 | */ | |
111c225a | 2301 | static int rescuer_thread(void *__rescuer) |
e22bee78 | 2302 | { |
111c225a TH |
2303 | struct worker *rescuer = __rescuer; |
2304 | struct workqueue_struct *wq = rescuer->rescue_wq; | |
e22bee78 | 2305 | struct list_head *scheduled = &rescuer->scheduled; |
e22bee78 TH |
2306 | |
2307 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
111c225a TH |
2308 | |
2309 | /* | |
2310 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it | |
2311 | * doesn't participate in concurrency management. | |
2312 | */ | |
2313 | rescuer->task->flags |= PF_WQ_WORKER; | |
e22bee78 TH |
2314 | repeat: |
2315 | set_current_state(TASK_INTERRUPTIBLE); | |
2316 | ||
412d32e6 MG |
2317 | if (kthread_should_stop()) { |
2318 | __set_current_state(TASK_RUNNING); | |
111c225a | 2319 | rescuer->task->flags &= ~PF_WQ_WORKER; |
e22bee78 | 2320 | return 0; |
412d32e6 | 2321 | } |
e22bee78 | 2322 | |
493a1724 TH |
2323 | /* see whether any pwq is asking for help */ |
2324 | spin_lock_irq(&workqueue_lock); | |
2325 | ||
2326 | while (!list_empty(&wq->maydays)) { | |
2327 | struct pool_workqueue *pwq = list_first_entry(&wq->maydays, | |
2328 | struct pool_workqueue, mayday_node); | |
112202d9 | 2329 | struct worker_pool *pool = pwq->pool; |
e22bee78 TH |
2330 | struct work_struct *work, *n; |
2331 | ||
2332 | __set_current_state(TASK_RUNNING); | |
493a1724 TH |
2333 | list_del_init(&pwq->mayday_node); |
2334 | ||
2335 | spin_unlock_irq(&workqueue_lock); | |
e22bee78 TH |
2336 | |
2337 | /* migrate to the target cpu if possible */ | |
f36dc67b | 2338 | worker_maybe_bind_and_lock(pool); |
b3104104 | 2339 | rescuer->pool = pool; |
e22bee78 TH |
2340 | |
2341 | /* | |
2342 | * Slurp in all works issued via this workqueue and | |
2343 | * process'em. | |
2344 | */ | |
6183c009 | 2345 | WARN_ON_ONCE(!list_empty(&rescuer->scheduled)); |
bd7bdd43 | 2346 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
112202d9 | 2347 | if (get_work_pwq(work) == pwq) |
e22bee78 TH |
2348 | move_linked_works(work, scheduled, &n); |
2349 | ||
2350 | process_scheduled_works(rescuer); | |
7576958a TH |
2351 | |
2352 | /* | |
d565ed63 | 2353 | * Leave this pool. If keep_working() is %true, notify a |
7576958a TH |
2354 | * regular worker; otherwise, we end up with 0 concurrency |
2355 | * and stalling the execution. | |
2356 | */ | |
63d95a91 TH |
2357 | if (keep_working(pool)) |
2358 | wake_up_worker(pool); | |
7576958a | 2359 | |
b3104104 | 2360 | rescuer->pool = NULL; |
493a1724 TH |
2361 | spin_unlock(&pool->lock); |
2362 | spin_lock(&workqueue_lock); | |
e22bee78 TH |
2363 | } |
2364 | ||
493a1724 TH |
2365 | spin_unlock_irq(&workqueue_lock); |
2366 | ||
111c225a TH |
2367 | /* rescuers should never participate in concurrency management */ |
2368 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); | |
e22bee78 TH |
2369 | schedule(); |
2370 | goto repeat; | |
1da177e4 LT |
2371 | } |
2372 | ||
fc2e4d70 ON |
2373 | struct wq_barrier { |
2374 | struct work_struct work; | |
2375 | struct completion done; | |
2376 | }; | |
2377 | ||
2378 | static void wq_barrier_func(struct work_struct *work) | |
2379 | { | |
2380 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2381 | complete(&barr->done); | |
2382 | } | |
2383 | ||
4690c4ab TH |
2384 | /** |
2385 | * insert_wq_barrier - insert a barrier work | |
112202d9 | 2386 | * @pwq: pwq to insert barrier into |
4690c4ab | 2387 | * @barr: wq_barrier to insert |
affee4b2 TH |
2388 | * @target: target work to attach @barr to |
2389 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2390 | * |
affee4b2 TH |
2391 | * @barr is linked to @target such that @barr is completed only after |
2392 | * @target finishes execution. Please note that the ordering | |
2393 | * guarantee is observed only with respect to @target and on the local | |
2394 | * cpu. | |
2395 | * | |
2396 | * Currently, a queued barrier can't be canceled. This is because | |
2397 | * try_to_grab_pending() can't determine whether the work to be | |
2398 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2399 | * flag of the previous work while there must be a valid next work | |
2400 | * after a work with LINKED flag set. | |
2401 | * | |
2402 | * Note that when @worker is non-NULL, @target may be modified | |
112202d9 | 2403 | * underneath us, so we can't reliably determine pwq from @target. |
4690c4ab TH |
2404 | * |
2405 | * CONTEXT: | |
d565ed63 | 2406 | * spin_lock_irq(pool->lock). |
4690c4ab | 2407 | */ |
112202d9 | 2408 | static void insert_wq_barrier(struct pool_workqueue *pwq, |
affee4b2 TH |
2409 | struct wq_barrier *barr, |
2410 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2411 | { |
affee4b2 TH |
2412 | struct list_head *head; |
2413 | unsigned int linked = 0; | |
2414 | ||
dc186ad7 | 2415 | /* |
d565ed63 | 2416 | * debugobject calls are safe here even with pool->lock locked |
dc186ad7 TG |
2417 | * as we know for sure that this will not trigger any of the |
2418 | * checks and call back into the fixup functions where we | |
2419 | * might deadlock. | |
2420 | */ | |
ca1cab37 | 2421 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2422 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2423 | init_completion(&barr->done); |
83c22520 | 2424 | |
affee4b2 TH |
2425 | /* |
2426 | * If @target is currently being executed, schedule the | |
2427 | * barrier to the worker; otherwise, put it after @target. | |
2428 | */ | |
2429 | if (worker) | |
2430 | head = worker->scheduled.next; | |
2431 | else { | |
2432 | unsigned long *bits = work_data_bits(target); | |
2433 | ||
2434 | head = target->entry.next; | |
2435 | /* there can already be other linked works, inherit and set */ | |
2436 | linked = *bits & WORK_STRUCT_LINKED; | |
2437 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2438 | } | |
2439 | ||
dc186ad7 | 2440 | debug_work_activate(&barr->work); |
112202d9 | 2441 | insert_work(pwq, &barr->work, head, |
affee4b2 | 2442 | work_color_to_flags(WORK_NO_COLOR) | linked); |
fc2e4d70 ON |
2443 | } |
2444 | ||
73f53c4a | 2445 | /** |
112202d9 | 2446 | * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing |
73f53c4a TH |
2447 | * @wq: workqueue being flushed |
2448 | * @flush_color: new flush color, < 0 for no-op | |
2449 | * @work_color: new work color, < 0 for no-op | |
2450 | * | |
112202d9 | 2451 | * Prepare pwqs for workqueue flushing. |
73f53c4a | 2452 | * |
112202d9 TH |
2453 | * If @flush_color is non-negative, flush_color on all pwqs should be |
2454 | * -1. If no pwq has in-flight commands at the specified color, all | |
2455 | * pwq->flush_color's stay at -1 and %false is returned. If any pwq | |
2456 | * has in flight commands, its pwq->flush_color is set to | |
2457 | * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq | |
73f53c4a TH |
2458 | * wakeup logic is armed and %true is returned. |
2459 | * | |
2460 | * The caller should have initialized @wq->first_flusher prior to | |
2461 | * calling this function with non-negative @flush_color. If | |
2462 | * @flush_color is negative, no flush color update is done and %false | |
2463 | * is returned. | |
2464 | * | |
112202d9 | 2465 | * If @work_color is non-negative, all pwqs should have the same |
73f53c4a TH |
2466 | * work_color which is previous to @work_color and all will be |
2467 | * advanced to @work_color. | |
2468 | * | |
2469 | * CONTEXT: | |
2470 | * mutex_lock(wq->flush_mutex). | |
2471 | * | |
2472 | * RETURNS: | |
2473 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2474 | * otherwise. | |
2475 | */ | |
112202d9 | 2476 | static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq, |
73f53c4a | 2477 | int flush_color, int work_color) |
1da177e4 | 2478 | { |
73f53c4a | 2479 | bool wait = false; |
49e3cf44 | 2480 | struct pool_workqueue *pwq; |
1da177e4 | 2481 | |
73f53c4a | 2482 | if (flush_color >= 0) { |
6183c009 | 2483 | WARN_ON_ONCE(atomic_read(&wq->nr_pwqs_to_flush)); |
112202d9 | 2484 | atomic_set(&wq->nr_pwqs_to_flush, 1); |
1da177e4 | 2485 | } |
2355b70f | 2486 | |
76af4d93 TH |
2487 | local_irq_disable(); |
2488 | ||
49e3cf44 | 2489 | for_each_pwq(pwq, wq) { |
112202d9 | 2490 | struct worker_pool *pool = pwq->pool; |
fc2e4d70 | 2491 | |
76af4d93 | 2492 | spin_lock(&pool->lock); |
83c22520 | 2493 | |
73f53c4a | 2494 | if (flush_color >= 0) { |
6183c009 | 2495 | WARN_ON_ONCE(pwq->flush_color != -1); |
fc2e4d70 | 2496 | |
112202d9 TH |
2497 | if (pwq->nr_in_flight[flush_color]) { |
2498 | pwq->flush_color = flush_color; | |
2499 | atomic_inc(&wq->nr_pwqs_to_flush); | |
73f53c4a TH |
2500 | wait = true; |
2501 | } | |
2502 | } | |
1da177e4 | 2503 | |
73f53c4a | 2504 | if (work_color >= 0) { |
6183c009 | 2505 | WARN_ON_ONCE(work_color != work_next_color(pwq->work_color)); |
112202d9 | 2506 | pwq->work_color = work_color; |
73f53c4a | 2507 | } |
1da177e4 | 2508 | |
76af4d93 | 2509 | spin_unlock(&pool->lock); |
1da177e4 | 2510 | } |
2355b70f | 2511 | |
76af4d93 TH |
2512 | local_irq_enable(); |
2513 | ||
112202d9 | 2514 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_pwqs_to_flush)) |
73f53c4a | 2515 | complete(&wq->first_flusher->done); |
14441960 | 2516 | |
73f53c4a | 2517 | return wait; |
1da177e4 LT |
2518 | } |
2519 | ||
0fcb78c2 | 2520 | /** |
1da177e4 | 2521 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2522 | * @wq: workqueue to flush |
1da177e4 LT |
2523 | * |
2524 | * Forces execution of the workqueue and blocks until its completion. | |
2525 | * This is typically used in driver shutdown handlers. | |
2526 | * | |
fc2e4d70 ON |
2527 | * We sleep until all works which were queued on entry have been handled, |
2528 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 2529 | */ |
7ad5b3a5 | 2530 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2531 | { |
73f53c4a TH |
2532 | struct wq_flusher this_flusher = { |
2533 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2534 | .flush_color = -1, | |
2535 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2536 | }; | |
2537 | int next_color; | |
1da177e4 | 2538 | |
3295f0ef IM |
2539 | lock_map_acquire(&wq->lockdep_map); |
2540 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
2541 | |
2542 | mutex_lock(&wq->flush_mutex); | |
2543 | ||
2544 | /* | |
2545 | * Start-to-wait phase | |
2546 | */ | |
2547 | next_color = work_next_color(wq->work_color); | |
2548 | ||
2549 | if (next_color != wq->flush_color) { | |
2550 | /* | |
2551 | * Color space is not full. The current work_color | |
2552 | * becomes our flush_color and work_color is advanced | |
2553 | * by one. | |
2554 | */ | |
6183c009 | 2555 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow)); |
73f53c4a TH |
2556 | this_flusher.flush_color = wq->work_color; |
2557 | wq->work_color = next_color; | |
2558 | ||
2559 | if (!wq->first_flusher) { | |
2560 | /* no flush in progress, become the first flusher */ | |
6183c009 | 2561 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); |
73f53c4a TH |
2562 | |
2563 | wq->first_flusher = &this_flusher; | |
2564 | ||
112202d9 | 2565 | if (!flush_workqueue_prep_pwqs(wq, wq->flush_color, |
73f53c4a TH |
2566 | wq->work_color)) { |
2567 | /* nothing to flush, done */ | |
2568 | wq->flush_color = next_color; | |
2569 | wq->first_flusher = NULL; | |
2570 | goto out_unlock; | |
2571 | } | |
2572 | } else { | |
2573 | /* wait in queue */ | |
6183c009 | 2574 | WARN_ON_ONCE(wq->flush_color == this_flusher.flush_color); |
73f53c4a | 2575 | list_add_tail(&this_flusher.list, &wq->flusher_queue); |
112202d9 | 2576 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2577 | } |
2578 | } else { | |
2579 | /* | |
2580 | * Oops, color space is full, wait on overflow queue. | |
2581 | * The next flush completion will assign us | |
2582 | * flush_color and transfer to flusher_queue. | |
2583 | */ | |
2584 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2585 | } | |
2586 | ||
2587 | mutex_unlock(&wq->flush_mutex); | |
2588 | ||
2589 | wait_for_completion(&this_flusher.done); | |
2590 | ||
2591 | /* | |
2592 | * Wake-up-and-cascade phase | |
2593 | * | |
2594 | * First flushers are responsible for cascading flushes and | |
2595 | * handling overflow. Non-first flushers can simply return. | |
2596 | */ | |
2597 | if (wq->first_flusher != &this_flusher) | |
2598 | return; | |
2599 | ||
2600 | mutex_lock(&wq->flush_mutex); | |
2601 | ||
4ce48b37 TH |
2602 | /* we might have raced, check again with mutex held */ |
2603 | if (wq->first_flusher != &this_flusher) | |
2604 | goto out_unlock; | |
2605 | ||
73f53c4a TH |
2606 | wq->first_flusher = NULL; |
2607 | ||
6183c009 TH |
2608 | WARN_ON_ONCE(!list_empty(&this_flusher.list)); |
2609 | WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color); | |
73f53c4a TH |
2610 | |
2611 | while (true) { | |
2612 | struct wq_flusher *next, *tmp; | |
2613 | ||
2614 | /* complete all the flushers sharing the current flush color */ | |
2615 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2616 | if (next->flush_color != wq->flush_color) | |
2617 | break; | |
2618 | list_del_init(&next->list); | |
2619 | complete(&next->done); | |
2620 | } | |
2621 | ||
6183c009 TH |
2622 | WARN_ON_ONCE(!list_empty(&wq->flusher_overflow) && |
2623 | wq->flush_color != work_next_color(wq->work_color)); | |
73f53c4a TH |
2624 | |
2625 | /* this flush_color is finished, advance by one */ | |
2626 | wq->flush_color = work_next_color(wq->flush_color); | |
2627 | ||
2628 | /* one color has been freed, handle overflow queue */ | |
2629 | if (!list_empty(&wq->flusher_overflow)) { | |
2630 | /* | |
2631 | * Assign the same color to all overflowed | |
2632 | * flushers, advance work_color and append to | |
2633 | * flusher_queue. This is the start-to-wait | |
2634 | * phase for these overflowed flushers. | |
2635 | */ | |
2636 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2637 | tmp->flush_color = wq->work_color; | |
2638 | ||
2639 | wq->work_color = work_next_color(wq->work_color); | |
2640 | ||
2641 | list_splice_tail_init(&wq->flusher_overflow, | |
2642 | &wq->flusher_queue); | |
112202d9 | 2643 | flush_workqueue_prep_pwqs(wq, -1, wq->work_color); |
73f53c4a TH |
2644 | } |
2645 | ||
2646 | if (list_empty(&wq->flusher_queue)) { | |
6183c009 | 2647 | WARN_ON_ONCE(wq->flush_color != wq->work_color); |
73f53c4a TH |
2648 | break; |
2649 | } | |
2650 | ||
2651 | /* | |
2652 | * Need to flush more colors. Make the next flusher | |
112202d9 | 2653 | * the new first flusher and arm pwqs. |
73f53c4a | 2654 | */ |
6183c009 TH |
2655 | WARN_ON_ONCE(wq->flush_color == wq->work_color); |
2656 | WARN_ON_ONCE(wq->flush_color != next->flush_color); | |
73f53c4a TH |
2657 | |
2658 | list_del_init(&next->list); | |
2659 | wq->first_flusher = next; | |
2660 | ||
112202d9 | 2661 | if (flush_workqueue_prep_pwqs(wq, wq->flush_color, -1)) |
73f53c4a TH |
2662 | break; |
2663 | ||
2664 | /* | |
2665 | * Meh... this color is already done, clear first | |
2666 | * flusher and repeat cascading. | |
2667 | */ | |
2668 | wq->first_flusher = NULL; | |
2669 | } | |
2670 | ||
2671 | out_unlock: | |
2672 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 2673 | } |
ae90dd5d | 2674 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2675 | |
9c5a2ba7 TH |
2676 | /** |
2677 | * drain_workqueue - drain a workqueue | |
2678 | * @wq: workqueue to drain | |
2679 | * | |
2680 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2681 | * only chain queueing is allowed. IOW, only currently pending or running | |
2682 | * work items on @wq can queue further work items on it. @wq is flushed | |
2683 | * repeatedly until it becomes empty. The number of flushing is detemined | |
2684 | * by the depth of chaining and should be relatively short. Whine if it | |
2685 | * takes too long. | |
2686 | */ | |
2687 | void drain_workqueue(struct workqueue_struct *wq) | |
2688 | { | |
2689 | unsigned int flush_cnt = 0; | |
49e3cf44 | 2690 | struct pool_workqueue *pwq; |
9c5a2ba7 TH |
2691 | |
2692 | /* | |
2693 | * __queue_work() needs to test whether there are drainers, is much | |
2694 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
2695 | * Use WQ_DRAINING so that queue doesn't have to check nr_drainers. | |
2696 | */ | |
e98d5b16 | 2697 | spin_lock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2698 | if (!wq->nr_drainers++) |
2699 | wq->flags |= WQ_DRAINING; | |
e98d5b16 | 2700 | spin_unlock_irq(&workqueue_lock); |
9c5a2ba7 TH |
2701 | reflush: |
2702 | flush_workqueue(wq); | |
2703 | ||
76af4d93 TH |
2704 | local_irq_disable(); |
2705 | ||
49e3cf44 | 2706 | for_each_pwq(pwq, wq) { |
fa2563e4 | 2707 | bool drained; |
9c5a2ba7 | 2708 | |
76af4d93 | 2709 | spin_lock(&pwq->pool->lock); |
112202d9 | 2710 | drained = !pwq->nr_active && list_empty(&pwq->delayed_works); |
76af4d93 | 2711 | spin_unlock(&pwq->pool->lock); |
fa2563e4 TT |
2712 | |
2713 | if (drained) | |
9c5a2ba7 TH |
2714 | continue; |
2715 | ||
2716 | if (++flush_cnt == 10 || | |
2717 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
044c782c VI |
2718 | pr_warn("workqueue %s: flush on destruction isn't complete after %u tries\n", |
2719 | wq->name, flush_cnt); | |
76af4d93 TH |
2720 | |
2721 | local_irq_enable(); | |
9c5a2ba7 TH |
2722 | goto reflush; |
2723 | } | |
2724 | ||
76af4d93 | 2725 | spin_lock(&workqueue_lock); |
9c5a2ba7 TH |
2726 | if (!--wq->nr_drainers) |
2727 | wq->flags &= ~WQ_DRAINING; | |
76af4d93 TH |
2728 | spin_unlock(&workqueue_lock); |
2729 | ||
2730 | local_irq_enable(); | |
9c5a2ba7 TH |
2731 | } |
2732 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2733 | ||
606a5020 | 2734 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) |
db700897 | 2735 | { |
affee4b2 | 2736 | struct worker *worker = NULL; |
c9e7cf27 | 2737 | struct worker_pool *pool; |
112202d9 | 2738 | struct pool_workqueue *pwq; |
db700897 ON |
2739 | |
2740 | might_sleep(); | |
fa1b54e6 TH |
2741 | |
2742 | local_irq_disable(); | |
c9e7cf27 | 2743 | pool = get_work_pool(work); |
fa1b54e6 TH |
2744 | if (!pool) { |
2745 | local_irq_enable(); | |
baf59022 | 2746 | return false; |
fa1b54e6 | 2747 | } |
db700897 | 2748 | |
fa1b54e6 | 2749 | spin_lock(&pool->lock); |
0b3dae68 | 2750 | /* see the comment in try_to_grab_pending() with the same code */ |
112202d9 TH |
2751 | pwq = get_work_pwq(work); |
2752 | if (pwq) { | |
2753 | if (unlikely(pwq->pool != pool)) | |
4690c4ab | 2754 | goto already_gone; |
606a5020 | 2755 | } else { |
c9e7cf27 | 2756 | worker = find_worker_executing_work(pool, work); |
affee4b2 | 2757 | if (!worker) |
4690c4ab | 2758 | goto already_gone; |
112202d9 | 2759 | pwq = worker->current_pwq; |
606a5020 | 2760 | } |
db700897 | 2761 | |
112202d9 | 2762 | insert_wq_barrier(pwq, barr, work, worker); |
d565ed63 | 2763 | spin_unlock_irq(&pool->lock); |
7a22ad75 | 2764 | |
e159489b TH |
2765 | /* |
2766 | * If @max_active is 1 or rescuer is in use, flushing another work | |
2767 | * item on the same workqueue may lead to deadlock. Make sure the | |
2768 | * flusher is not running on the same workqueue by verifying write | |
2769 | * access. | |
2770 | */ | |
112202d9 TH |
2771 | if (pwq->wq->saved_max_active == 1 || pwq->wq->flags & WQ_RESCUER) |
2772 | lock_map_acquire(&pwq->wq->lockdep_map); | |
e159489b | 2773 | else |
112202d9 TH |
2774 | lock_map_acquire_read(&pwq->wq->lockdep_map); |
2775 | lock_map_release(&pwq->wq->lockdep_map); | |
e159489b | 2776 | |
401a8d04 | 2777 | return true; |
4690c4ab | 2778 | already_gone: |
d565ed63 | 2779 | spin_unlock_irq(&pool->lock); |
401a8d04 | 2780 | return false; |
db700897 | 2781 | } |
baf59022 TH |
2782 | |
2783 | /** | |
2784 | * flush_work - wait for a work to finish executing the last queueing instance | |
2785 | * @work: the work to flush | |
2786 | * | |
606a5020 TH |
2787 | * Wait until @work has finished execution. @work is guaranteed to be idle |
2788 | * on return if it hasn't been requeued since flush started. | |
baf59022 TH |
2789 | * |
2790 | * RETURNS: | |
2791 | * %true if flush_work() waited for the work to finish execution, | |
2792 | * %false if it was already idle. | |
2793 | */ | |
2794 | bool flush_work(struct work_struct *work) | |
2795 | { | |
2796 | struct wq_barrier barr; | |
2797 | ||
0976dfc1 SB |
2798 | lock_map_acquire(&work->lockdep_map); |
2799 | lock_map_release(&work->lockdep_map); | |
2800 | ||
606a5020 | 2801 | if (start_flush_work(work, &barr)) { |
401a8d04 TH |
2802 | wait_for_completion(&barr.done); |
2803 | destroy_work_on_stack(&barr.work); | |
2804 | return true; | |
606a5020 | 2805 | } else { |
401a8d04 | 2806 | return false; |
6e84d644 | 2807 | } |
6e84d644 | 2808 | } |
606a5020 | 2809 | EXPORT_SYMBOL_GPL(flush_work); |
6e84d644 | 2810 | |
36e227d2 | 2811 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
1f1f642e | 2812 | { |
bbb68dfa | 2813 | unsigned long flags; |
1f1f642e ON |
2814 | int ret; |
2815 | ||
2816 | do { | |
bbb68dfa TH |
2817 | ret = try_to_grab_pending(work, is_dwork, &flags); |
2818 | /* | |
2819 | * If someone else is canceling, wait for the same event it | |
2820 | * would be waiting for before retrying. | |
2821 | */ | |
2822 | if (unlikely(ret == -ENOENT)) | |
606a5020 | 2823 | flush_work(work); |
1f1f642e ON |
2824 | } while (unlikely(ret < 0)); |
2825 | ||
bbb68dfa TH |
2826 | /* tell other tasks trying to grab @work to back off */ |
2827 | mark_work_canceling(work); | |
2828 | local_irq_restore(flags); | |
2829 | ||
606a5020 | 2830 | flush_work(work); |
7a22ad75 | 2831 | clear_work_data(work); |
1f1f642e ON |
2832 | return ret; |
2833 | } | |
2834 | ||
6e84d644 | 2835 | /** |
401a8d04 TH |
2836 | * cancel_work_sync - cancel a work and wait for it to finish |
2837 | * @work: the work to cancel | |
6e84d644 | 2838 | * |
401a8d04 TH |
2839 | * Cancel @work and wait for its execution to finish. This function |
2840 | * can be used even if the work re-queues itself or migrates to | |
2841 | * another workqueue. On return from this function, @work is | |
2842 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2843 | * |
401a8d04 TH |
2844 | * cancel_work_sync(&delayed_work->work) must not be used for |
2845 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2846 | * |
401a8d04 | 2847 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2848 | * queued can't be destroyed before this function returns. |
401a8d04 TH |
2849 | * |
2850 | * RETURNS: | |
2851 | * %true if @work was pending, %false otherwise. | |
6e84d644 | 2852 | */ |
401a8d04 | 2853 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2854 | { |
36e227d2 | 2855 | return __cancel_work_timer(work, false); |
b89deed3 | 2856 | } |
28e53bdd | 2857 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2858 | |
6e84d644 | 2859 | /** |
401a8d04 TH |
2860 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2861 | * @dwork: the delayed work to flush | |
6e84d644 | 2862 | * |
401a8d04 TH |
2863 | * Delayed timer is cancelled and the pending work is queued for |
2864 | * immediate execution. Like flush_work(), this function only | |
2865 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2866 | * |
401a8d04 TH |
2867 | * RETURNS: |
2868 | * %true if flush_work() waited for the work to finish execution, | |
2869 | * %false if it was already idle. | |
6e84d644 | 2870 | */ |
401a8d04 TH |
2871 | bool flush_delayed_work(struct delayed_work *dwork) |
2872 | { | |
8930caba | 2873 | local_irq_disable(); |
401a8d04 | 2874 | if (del_timer_sync(&dwork->timer)) |
60c057bc | 2875 | __queue_work(dwork->cpu, dwork->wq, &dwork->work); |
8930caba | 2876 | local_irq_enable(); |
401a8d04 TH |
2877 | return flush_work(&dwork->work); |
2878 | } | |
2879 | EXPORT_SYMBOL(flush_delayed_work); | |
2880 | ||
09383498 | 2881 | /** |
57b30ae7 TH |
2882 | * cancel_delayed_work - cancel a delayed work |
2883 | * @dwork: delayed_work to cancel | |
09383498 | 2884 | * |
57b30ae7 TH |
2885 | * Kill off a pending delayed_work. Returns %true if @dwork was pending |
2886 | * and canceled; %false if wasn't pending. Note that the work callback | |
2887 | * function may still be running on return, unless it returns %true and the | |
2888 | * work doesn't re-arm itself. Explicitly flush or use | |
2889 | * cancel_delayed_work_sync() to wait on it. | |
09383498 | 2890 | * |
57b30ae7 | 2891 | * This function is safe to call from any context including IRQ handler. |
09383498 | 2892 | */ |
57b30ae7 | 2893 | bool cancel_delayed_work(struct delayed_work *dwork) |
09383498 | 2894 | { |
57b30ae7 TH |
2895 | unsigned long flags; |
2896 | int ret; | |
2897 | ||
2898 | do { | |
2899 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
2900 | } while (unlikely(ret == -EAGAIN)); | |
2901 | ||
2902 | if (unlikely(ret < 0)) | |
2903 | return false; | |
2904 | ||
7c3eed5c TH |
2905 | set_work_pool_and_clear_pending(&dwork->work, |
2906 | get_work_pool_id(&dwork->work)); | |
57b30ae7 | 2907 | local_irq_restore(flags); |
c0158ca6 | 2908 | return ret; |
09383498 | 2909 | } |
57b30ae7 | 2910 | EXPORT_SYMBOL(cancel_delayed_work); |
09383498 | 2911 | |
401a8d04 TH |
2912 | /** |
2913 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
2914 | * @dwork: the delayed work cancel | |
2915 | * | |
2916 | * This is cancel_work_sync() for delayed works. | |
2917 | * | |
2918 | * RETURNS: | |
2919 | * %true if @dwork was pending, %false otherwise. | |
2920 | */ | |
2921 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 2922 | { |
36e227d2 | 2923 | return __cancel_work_timer(&dwork->work, true); |
6e84d644 | 2924 | } |
f5a421a4 | 2925 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 2926 | |
0fcb78c2 | 2927 | /** |
c1a220e7 ZR |
2928 | * schedule_work_on - put work task on a specific cpu |
2929 | * @cpu: cpu to put the work task on | |
2930 | * @work: job to be done | |
2931 | * | |
2932 | * This puts a job on a specific cpu | |
2933 | */ | |
d4283e93 | 2934 | bool schedule_work_on(int cpu, struct work_struct *work) |
c1a220e7 | 2935 | { |
d320c038 | 2936 | return queue_work_on(cpu, system_wq, work); |
c1a220e7 ZR |
2937 | } |
2938 | EXPORT_SYMBOL(schedule_work_on); | |
2939 | ||
0fcb78c2 | 2940 | /** |
0fcb78c2 REB |
2941 | * schedule_work - put work task in global workqueue |
2942 | * @work: job to be done | |
0fcb78c2 | 2943 | * |
d4283e93 TH |
2944 | * Returns %false if @work was already on the kernel-global workqueue and |
2945 | * %true otherwise. | |
5b0f437d BVA |
2946 | * |
2947 | * This puts a job in the kernel-global workqueue if it was not already | |
2948 | * queued and leaves it in the same position on the kernel-global | |
2949 | * workqueue otherwise. | |
0fcb78c2 | 2950 | */ |
d4283e93 | 2951 | bool schedule_work(struct work_struct *work) |
1da177e4 | 2952 | { |
d320c038 | 2953 | return queue_work(system_wq, work); |
1da177e4 | 2954 | } |
ae90dd5d | 2955 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 2956 | |
0fcb78c2 REB |
2957 | /** |
2958 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
2959 | * @cpu: cpu to use | |
52bad64d | 2960 | * @dwork: job to be done |
0fcb78c2 REB |
2961 | * @delay: number of jiffies to wait |
2962 | * | |
2963 | * After waiting for a given time this puts a job in the kernel-global | |
2964 | * workqueue on the specified CPU. | |
2965 | */ | |
d4283e93 TH |
2966 | bool schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
2967 | unsigned long delay) | |
1da177e4 | 2968 | { |
d320c038 | 2969 | return queue_delayed_work_on(cpu, system_wq, dwork, delay); |
1da177e4 | 2970 | } |
ae90dd5d | 2971 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 2972 | |
0fcb78c2 REB |
2973 | /** |
2974 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
2975 | * @dwork: job to be done |
2976 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
2977 | * |
2978 | * After waiting for a given time this puts a job in the kernel-global | |
2979 | * workqueue. | |
2980 | */ | |
d4283e93 | 2981 | bool schedule_delayed_work(struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 2982 | { |
d320c038 | 2983 | return queue_delayed_work(system_wq, dwork, delay); |
1da177e4 | 2984 | } |
ae90dd5d | 2985 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 2986 | |
b6136773 | 2987 | /** |
31ddd871 | 2988 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 2989 | * @func: the function to call |
b6136773 | 2990 | * |
31ddd871 TH |
2991 | * schedule_on_each_cpu() executes @func on each online CPU using the |
2992 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 2993 | * schedule_on_each_cpu() is very slow. |
31ddd871 TH |
2994 | * |
2995 | * RETURNS: | |
2996 | * 0 on success, -errno on failure. | |
b6136773 | 2997 | */ |
65f27f38 | 2998 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
2999 | { |
3000 | int cpu; | |
38f51568 | 3001 | struct work_struct __percpu *works; |
15316ba8 | 3002 | |
b6136773 AM |
3003 | works = alloc_percpu(struct work_struct); |
3004 | if (!works) | |
15316ba8 | 3005 | return -ENOMEM; |
b6136773 | 3006 | |
93981800 TH |
3007 | get_online_cpus(); |
3008 | ||
15316ba8 | 3009 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
3010 | struct work_struct *work = per_cpu_ptr(works, cpu); |
3011 | ||
3012 | INIT_WORK(work, func); | |
b71ab8c2 | 3013 | schedule_work_on(cpu, work); |
65a64464 | 3014 | } |
93981800 TH |
3015 | |
3016 | for_each_online_cpu(cpu) | |
3017 | flush_work(per_cpu_ptr(works, cpu)); | |
3018 | ||
95402b38 | 3019 | put_online_cpus(); |
b6136773 | 3020 | free_percpu(works); |
15316ba8 CL |
3021 | return 0; |
3022 | } | |
3023 | ||
eef6a7d5 AS |
3024 | /** |
3025 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
3026 | * | |
3027 | * Forces execution of the kernel-global workqueue and blocks until its | |
3028 | * completion. | |
3029 | * | |
3030 | * Think twice before calling this function! It's very easy to get into | |
3031 | * trouble if you don't take great care. Either of the following situations | |
3032 | * will lead to deadlock: | |
3033 | * | |
3034 | * One of the work items currently on the workqueue needs to acquire | |
3035 | * a lock held by your code or its caller. | |
3036 | * | |
3037 | * Your code is running in the context of a work routine. | |
3038 | * | |
3039 | * They will be detected by lockdep when they occur, but the first might not | |
3040 | * occur very often. It depends on what work items are on the workqueue and | |
3041 | * what locks they need, which you have no control over. | |
3042 | * | |
3043 | * In most situations flushing the entire workqueue is overkill; you merely | |
3044 | * need to know that a particular work item isn't queued and isn't running. | |
3045 | * In such cases you should use cancel_delayed_work_sync() or | |
3046 | * cancel_work_sync() instead. | |
3047 | */ | |
1da177e4 LT |
3048 | void flush_scheduled_work(void) |
3049 | { | |
d320c038 | 3050 | flush_workqueue(system_wq); |
1da177e4 | 3051 | } |
ae90dd5d | 3052 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 3053 | |
1fa44eca JB |
3054 | /** |
3055 | * execute_in_process_context - reliably execute the routine with user context | |
3056 | * @fn: the function to execute | |
1fa44eca JB |
3057 | * @ew: guaranteed storage for the execute work structure (must |
3058 | * be available when the work executes) | |
3059 | * | |
3060 | * Executes the function immediately if process context is available, | |
3061 | * otherwise schedules the function for delayed execution. | |
3062 | * | |
3063 | * Returns: 0 - function was executed | |
3064 | * 1 - function was scheduled for execution | |
3065 | */ | |
65f27f38 | 3066 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
3067 | { |
3068 | if (!in_interrupt()) { | |
65f27f38 | 3069 | fn(&ew->work); |
1fa44eca JB |
3070 | return 0; |
3071 | } | |
3072 | ||
65f27f38 | 3073 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
3074 | schedule_work(&ew->work); |
3075 | ||
3076 | return 1; | |
3077 | } | |
3078 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
3079 | ||
1da177e4 LT |
3080 | int keventd_up(void) |
3081 | { | |
d320c038 | 3082 | return system_wq != NULL; |
1da177e4 LT |
3083 | } |
3084 | ||
7a4e344c TH |
3085 | /** |
3086 | * free_workqueue_attrs - free a workqueue_attrs | |
3087 | * @attrs: workqueue_attrs to free | |
3088 | * | |
3089 | * Undo alloc_workqueue_attrs(). | |
3090 | */ | |
3091 | void free_workqueue_attrs(struct workqueue_attrs *attrs) | |
3092 | { | |
3093 | if (attrs) { | |
3094 | free_cpumask_var(attrs->cpumask); | |
3095 | kfree(attrs); | |
3096 | } | |
3097 | } | |
3098 | ||
3099 | /** | |
3100 | * alloc_workqueue_attrs - allocate a workqueue_attrs | |
3101 | * @gfp_mask: allocation mask to use | |
3102 | * | |
3103 | * Allocate a new workqueue_attrs, initialize with default settings and | |
3104 | * return it. Returns NULL on failure. | |
3105 | */ | |
3106 | struct workqueue_attrs *alloc_workqueue_attrs(gfp_t gfp_mask) | |
3107 | { | |
3108 | struct workqueue_attrs *attrs; | |
3109 | ||
3110 | attrs = kzalloc(sizeof(*attrs), gfp_mask); | |
3111 | if (!attrs) | |
3112 | goto fail; | |
3113 | if (!alloc_cpumask_var(&attrs->cpumask, gfp_mask)) | |
3114 | goto fail; | |
3115 | ||
3116 | cpumask_setall(attrs->cpumask); | |
3117 | return attrs; | |
3118 | fail: | |
3119 | free_workqueue_attrs(attrs); | |
3120 | return NULL; | |
3121 | } | |
3122 | ||
29c91e99 TH |
3123 | static void copy_workqueue_attrs(struct workqueue_attrs *to, |
3124 | const struct workqueue_attrs *from) | |
3125 | { | |
3126 | to->nice = from->nice; | |
3127 | cpumask_copy(to->cpumask, from->cpumask); | |
3128 | } | |
3129 | ||
3130 | /* | |
3131 | * Hacky implementation of jhash of bitmaps which only considers the | |
3132 | * specified number of bits. We probably want a proper implementation in | |
3133 | * include/linux/jhash.h. | |
3134 | */ | |
3135 | static u32 jhash_bitmap(const unsigned long *bitmap, int bits, u32 hash) | |
3136 | { | |
3137 | int nr_longs = bits / BITS_PER_LONG; | |
3138 | int nr_leftover = bits % BITS_PER_LONG; | |
3139 | unsigned long leftover = 0; | |
3140 | ||
3141 | if (nr_longs) | |
3142 | hash = jhash(bitmap, nr_longs * sizeof(long), hash); | |
3143 | if (nr_leftover) { | |
3144 | bitmap_copy(&leftover, bitmap + nr_longs, nr_leftover); | |
3145 | hash = jhash(&leftover, sizeof(long), hash); | |
3146 | } | |
3147 | return hash; | |
3148 | } | |
3149 | ||
3150 | /* hash value of the content of @attr */ | |
3151 | static u32 wqattrs_hash(const struct workqueue_attrs *attrs) | |
3152 | { | |
3153 | u32 hash = 0; | |
3154 | ||
3155 | hash = jhash_1word(attrs->nice, hash); | |
3156 | hash = jhash_bitmap(cpumask_bits(attrs->cpumask), nr_cpu_ids, hash); | |
3157 | return hash; | |
3158 | } | |
3159 | ||
3160 | /* content equality test */ | |
3161 | static bool wqattrs_equal(const struct workqueue_attrs *a, | |
3162 | const struct workqueue_attrs *b) | |
3163 | { | |
3164 | if (a->nice != b->nice) | |
3165 | return false; | |
3166 | if (!cpumask_equal(a->cpumask, b->cpumask)) | |
3167 | return false; | |
3168 | return true; | |
3169 | } | |
3170 | ||
7a4e344c TH |
3171 | /** |
3172 | * init_worker_pool - initialize a newly zalloc'd worker_pool | |
3173 | * @pool: worker_pool to initialize | |
3174 | * | |
3175 | * Initiailize a newly zalloc'd @pool. It also allocates @pool->attrs. | |
29c91e99 TH |
3176 | * Returns 0 on success, -errno on failure. Even on failure, all fields |
3177 | * inside @pool proper are initialized and put_unbound_pool() can be called | |
3178 | * on @pool safely to release it. | |
7a4e344c TH |
3179 | */ |
3180 | static int init_worker_pool(struct worker_pool *pool) | |
4e1a1f9a TH |
3181 | { |
3182 | spin_lock_init(&pool->lock); | |
29c91e99 TH |
3183 | pool->id = -1; |
3184 | pool->cpu = -1; | |
4e1a1f9a TH |
3185 | pool->flags |= POOL_DISASSOCIATED; |
3186 | INIT_LIST_HEAD(&pool->worklist); | |
3187 | INIT_LIST_HEAD(&pool->idle_list); | |
3188 | hash_init(pool->busy_hash); | |
3189 | ||
3190 | init_timer_deferrable(&pool->idle_timer); | |
3191 | pool->idle_timer.function = idle_worker_timeout; | |
3192 | pool->idle_timer.data = (unsigned long)pool; | |
3193 | ||
3194 | setup_timer(&pool->mayday_timer, pool_mayday_timeout, | |
3195 | (unsigned long)pool); | |
3196 | ||
3197 | mutex_init(&pool->manager_arb); | |
3198 | mutex_init(&pool->assoc_mutex); | |
3199 | ida_init(&pool->worker_ida); | |
7a4e344c | 3200 | |
29c91e99 TH |
3201 | INIT_HLIST_NODE(&pool->hash_node); |
3202 | pool->refcnt = 1; | |
3203 | ||
3204 | /* shouldn't fail above this point */ | |
7a4e344c TH |
3205 | pool->attrs = alloc_workqueue_attrs(GFP_KERNEL); |
3206 | if (!pool->attrs) | |
3207 | return -ENOMEM; | |
3208 | return 0; | |
4e1a1f9a TH |
3209 | } |
3210 | ||
29c91e99 TH |
3211 | static void rcu_free_pool(struct rcu_head *rcu) |
3212 | { | |
3213 | struct worker_pool *pool = container_of(rcu, struct worker_pool, rcu); | |
3214 | ||
3215 | ida_destroy(&pool->worker_ida); | |
3216 | free_workqueue_attrs(pool->attrs); | |
3217 | kfree(pool); | |
3218 | } | |
3219 | ||
3220 | /** | |
3221 | * put_unbound_pool - put a worker_pool | |
3222 | * @pool: worker_pool to put | |
3223 | * | |
3224 | * Put @pool. If its refcnt reaches zero, it gets destroyed in sched-RCU | |
3225 | * safe manner. | |
3226 | */ | |
3227 | static void put_unbound_pool(struct worker_pool *pool) | |
3228 | { | |
3229 | struct worker *worker; | |
3230 | ||
3231 | spin_lock_irq(&workqueue_lock); | |
3232 | if (--pool->refcnt) { | |
3233 | spin_unlock_irq(&workqueue_lock); | |
3234 | return; | |
3235 | } | |
3236 | ||
3237 | /* sanity checks */ | |
3238 | if (WARN_ON(!(pool->flags & POOL_DISASSOCIATED)) || | |
3239 | WARN_ON(!list_empty(&pool->worklist))) { | |
3240 | spin_unlock_irq(&workqueue_lock); | |
3241 | return; | |
3242 | } | |
3243 | ||
3244 | /* release id and unhash */ | |
3245 | if (pool->id >= 0) | |
3246 | idr_remove(&worker_pool_idr, pool->id); | |
3247 | hash_del(&pool->hash_node); | |
3248 | ||
3249 | spin_unlock_irq(&workqueue_lock); | |
3250 | ||
3251 | /* lock out manager and destroy all workers */ | |
3252 | mutex_lock(&pool->manager_arb); | |
3253 | spin_lock_irq(&pool->lock); | |
3254 | ||
3255 | while ((worker = first_worker(pool))) | |
3256 | destroy_worker(worker); | |
3257 | WARN_ON(pool->nr_workers || pool->nr_idle); | |
3258 | ||
3259 | spin_unlock_irq(&pool->lock); | |
3260 | mutex_unlock(&pool->manager_arb); | |
3261 | ||
3262 | /* shut down the timers */ | |
3263 | del_timer_sync(&pool->idle_timer); | |
3264 | del_timer_sync(&pool->mayday_timer); | |
3265 | ||
3266 | /* sched-RCU protected to allow dereferences from get_work_pool() */ | |
3267 | call_rcu_sched(&pool->rcu, rcu_free_pool); | |
3268 | } | |
3269 | ||
3270 | /** | |
3271 | * get_unbound_pool - get a worker_pool with the specified attributes | |
3272 | * @attrs: the attributes of the worker_pool to get | |
3273 | * | |
3274 | * Obtain a worker_pool which has the same attributes as @attrs, bump the | |
3275 | * reference count and return it. If there already is a matching | |
3276 | * worker_pool, it will be used; otherwise, this function attempts to | |
3277 | * create a new one. On failure, returns NULL. | |
3278 | */ | |
3279 | static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs) | |
3280 | { | |
3281 | static DEFINE_MUTEX(create_mutex); | |
3282 | u32 hash = wqattrs_hash(attrs); | |
3283 | struct worker_pool *pool; | |
3284 | struct worker *worker; | |
3285 | ||
3286 | mutex_lock(&create_mutex); | |
3287 | ||
3288 | /* do we already have a matching pool? */ | |
3289 | spin_lock_irq(&workqueue_lock); | |
3290 | hash_for_each_possible(unbound_pool_hash, pool, hash_node, hash) { | |
3291 | if (wqattrs_equal(pool->attrs, attrs)) { | |
3292 | pool->refcnt++; | |
3293 | goto out_unlock; | |
3294 | } | |
3295 | } | |
3296 | spin_unlock_irq(&workqueue_lock); | |
3297 | ||
3298 | /* nope, create a new one */ | |
3299 | pool = kzalloc(sizeof(*pool), GFP_KERNEL); | |
3300 | if (!pool || init_worker_pool(pool) < 0) | |
3301 | goto fail; | |
3302 | ||
3303 | copy_workqueue_attrs(pool->attrs, attrs); | |
3304 | ||
3305 | if (worker_pool_assign_id(pool) < 0) | |
3306 | goto fail; | |
3307 | ||
3308 | /* create and start the initial worker */ | |
3309 | worker = create_worker(pool); | |
3310 | if (!worker) | |
3311 | goto fail; | |
3312 | ||
3313 | spin_lock_irq(&pool->lock); | |
3314 | start_worker(worker); | |
3315 | spin_unlock_irq(&pool->lock); | |
3316 | ||
3317 | /* install */ | |
3318 | spin_lock_irq(&workqueue_lock); | |
3319 | hash_add(unbound_pool_hash, &pool->hash_node, hash); | |
3320 | out_unlock: | |
3321 | spin_unlock_irq(&workqueue_lock); | |
3322 | mutex_unlock(&create_mutex); | |
3323 | return pool; | |
3324 | fail: | |
3325 | mutex_unlock(&create_mutex); | |
3326 | if (pool) | |
3327 | put_unbound_pool(pool); | |
3328 | return NULL; | |
3329 | } | |
3330 | ||
30cdf249 | 3331 | static int alloc_and_link_pwqs(struct workqueue_struct *wq) |
0f900049 | 3332 | { |
49e3cf44 | 3333 | bool highpri = wq->flags & WQ_HIGHPRI; |
30cdf249 TH |
3334 | int cpu; |
3335 | ||
3336 | if (!(wq->flags & WQ_UNBOUND)) { | |
420c0ddb TH |
3337 | wq->cpu_pwqs = alloc_percpu(struct pool_workqueue); |
3338 | if (!wq->cpu_pwqs) | |
30cdf249 TH |
3339 | return -ENOMEM; |
3340 | ||
3341 | for_each_possible_cpu(cpu) { | |
7fb98ea7 TH |
3342 | struct pool_workqueue *pwq = |
3343 | per_cpu_ptr(wq->cpu_pwqs, cpu); | |
7a62c2c8 TH |
3344 | struct worker_pool *cpu_pools = |
3345 | per_cpu(cpu_std_worker_pools, cpu); | |
f3421797 | 3346 | |
7a62c2c8 | 3347 | pwq->pool = &cpu_pools[highpri]; |
76af4d93 | 3348 | list_add_tail_rcu(&pwq->pwqs_node, &wq->pwqs); |
30cdf249 TH |
3349 | } |
3350 | } else { | |
3351 | struct pool_workqueue *pwq; | |
3352 | ||
3353 | pwq = kmem_cache_zalloc(pwq_cache, GFP_KERNEL); | |
3354 | if (!pwq) | |
3355 | return -ENOMEM; | |
3356 | ||
29c91e99 TH |
3357 | pwq->pool = get_unbound_pool(unbound_std_wq_attrs[highpri]); |
3358 | if (!pwq->pool) { | |
3359 | kmem_cache_free(pwq_cache, pwq); | |
3360 | return -ENOMEM; | |
3361 | } | |
3362 | ||
76af4d93 | 3363 | list_add_tail_rcu(&pwq->pwqs_node, &wq->pwqs); |
30cdf249 TH |
3364 | } |
3365 | ||
3366 | return 0; | |
0f900049 TH |
3367 | } |
3368 | ||
112202d9 | 3369 | static void free_pwqs(struct workqueue_struct *wq) |
0f900049 | 3370 | { |
e06ffa1e | 3371 | if (!(wq->flags & WQ_UNBOUND)) |
420c0ddb TH |
3372 | free_percpu(wq->cpu_pwqs); |
3373 | else if (!list_empty(&wq->pwqs)) | |
3374 | kmem_cache_free(pwq_cache, list_first_entry(&wq->pwqs, | |
3375 | struct pool_workqueue, pwqs_node)); | |
0f900049 TH |
3376 | } |
3377 | ||
f3421797 TH |
3378 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
3379 | const char *name) | |
b71ab8c2 | 3380 | { |
f3421797 TH |
3381 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
3382 | ||
3383 | if (max_active < 1 || max_active > lim) | |
044c782c VI |
3384 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
3385 | max_active, name, 1, lim); | |
b71ab8c2 | 3386 | |
f3421797 | 3387 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
3388 | } |
3389 | ||
b196be89 | 3390 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
d320c038 TH |
3391 | unsigned int flags, |
3392 | int max_active, | |
3393 | struct lock_class_key *key, | |
b196be89 | 3394 | const char *lock_name, ...) |
1da177e4 | 3395 | { |
b196be89 | 3396 | va_list args, args1; |
1da177e4 | 3397 | struct workqueue_struct *wq; |
49e3cf44 | 3398 | struct pool_workqueue *pwq; |
b196be89 TH |
3399 | size_t namelen; |
3400 | ||
3401 | /* determine namelen, allocate wq and format name */ | |
3402 | va_start(args, lock_name); | |
3403 | va_copy(args1, args); | |
3404 | namelen = vsnprintf(NULL, 0, fmt, args) + 1; | |
3405 | ||
3406 | wq = kzalloc(sizeof(*wq) + namelen, GFP_KERNEL); | |
3407 | if (!wq) | |
3408 | goto err; | |
3409 | ||
3410 | vsnprintf(wq->name, namelen, fmt, args1); | |
3411 | va_end(args); | |
3412 | va_end(args1); | |
1da177e4 | 3413 | |
6370a6ad TH |
3414 | /* |
3415 | * Workqueues which may be used during memory reclaim should | |
3416 | * have a rescuer to guarantee forward progress. | |
3417 | */ | |
3418 | if (flags & WQ_MEM_RECLAIM) | |
3419 | flags |= WQ_RESCUER; | |
3420 | ||
d320c038 | 3421 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 3422 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 3423 | |
b196be89 | 3424 | /* init wq */ |
97e37d7b | 3425 | wq->flags = flags; |
a0a1a5fd | 3426 | wq->saved_max_active = max_active; |
73f53c4a | 3427 | mutex_init(&wq->flush_mutex); |
112202d9 | 3428 | atomic_set(&wq->nr_pwqs_to_flush, 0); |
30cdf249 | 3429 | INIT_LIST_HEAD(&wq->pwqs); |
73f53c4a TH |
3430 | INIT_LIST_HEAD(&wq->flusher_queue); |
3431 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
493a1724 | 3432 | INIT_LIST_HEAD(&wq->maydays); |
502ca9d8 | 3433 | |
eb13ba87 | 3434 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 3435 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 3436 | |
30cdf249 | 3437 | if (alloc_and_link_pwqs(wq) < 0) |
bdbc5dd7 TH |
3438 | goto err; |
3439 | ||
76af4d93 | 3440 | local_irq_disable(); |
49e3cf44 | 3441 | for_each_pwq(pwq, wq) { |
112202d9 | 3442 | BUG_ON((unsigned long)pwq & WORK_STRUCT_FLAG_MASK); |
112202d9 TH |
3443 | pwq->wq = wq; |
3444 | pwq->flush_color = -1; | |
3445 | pwq->max_active = max_active; | |
3446 | INIT_LIST_HEAD(&pwq->delayed_works); | |
493a1724 | 3447 | INIT_LIST_HEAD(&pwq->mayday_node); |
e22bee78 | 3448 | } |
76af4d93 | 3449 | local_irq_enable(); |
1537663f | 3450 | |
e22bee78 TH |
3451 | if (flags & WQ_RESCUER) { |
3452 | struct worker *rescuer; | |
3453 | ||
e22bee78 TH |
3454 | wq->rescuer = rescuer = alloc_worker(); |
3455 | if (!rescuer) | |
3456 | goto err; | |
3457 | ||
111c225a TH |
3458 | rescuer->rescue_wq = wq; |
3459 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", | |
b196be89 | 3460 | wq->name); |
e22bee78 TH |
3461 | if (IS_ERR(rescuer->task)) |
3462 | goto err; | |
3463 | ||
e22bee78 TH |
3464 | rescuer->task->flags |= PF_THREAD_BOUND; |
3465 | wake_up_process(rescuer->task); | |
3af24433 ON |
3466 | } |
3467 | ||
a0a1a5fd TH |
3468 | /* |
3469 | * workqueue_lock protects global freeze state and workqueues | |
3470 | * list. Grab it, set max_active accordingly and add the new | |
3471 | * workqueue to workqueues list. | |
3472 | */ | |
e98d5b16 | 3473 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3474 | |
58a69cb4 | 3475 | if (workqueue_freezing && wq->flags & WQ_FREEZABLE) |
49e3cf44 TH |
3476 | for_each_pwq(pwq, wq) |
3477 | pwq->max_active = 0; | |
a0a1a5fd | 3478 | |
1537663f | 3479 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 3480 | |
e98d5b16 | 3481 | spin_unlock_irq(&workqueue_lock); |
1537663f | 3482 | |
3af24433 | 3483 | return wq; |
4690c4ab TH |
3484 | err: |
3485 | if (wq) { | |
112202d9 | 3486 | free_pwqs(wq); |
e22bee78 | 3487 | kfree(wq->rescuer); |
4690c4ab TH |
3488 | kfree(wq); |
3489 | } | |
3490 | return NULL; | |
3af24433 | 3491 | } |
d320c038 | 3492 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 3493 | |
3af24433 ON |
3494 | /** |
3495 | * destroy_workqueue - safely terminate a workqueue | |
3496 | * @wq: target workqueue | |
3497 | * | |
3498 | * Safely destroy a workqueue. All work currently pending will be done first. | |
3499 | */ | |
3500 | void destroy_workqueue(struct workqueue_struct *wq) | |
3501 | { | |
49e3cf44 | 3502 | struct pool_workqueue *pwq; |
3af24433 | 3503 | |
9c5a2ba7 TH |
3504 | /* drain it before proceeding with destruction */ |
3505 | drain_workqueue(wq); | |
c8efcc25 | 3506 | |
76af4d93 TH |
3507 | spin_lock_irq(&workqueue_lock); |
3508 | ||
6183c009 | 3509 | /* sanity checks */ |
49e3cf44 | 3510 | for_each_pwq(pwq, wq) { |
6183c009 TH |
3511 | int i; |
3512 | ||
76af4d93 TH |
3513 | for (i = 0; i < WORK_NR_COLORS; i++) { |
3514 | if (WARN_ON(pwq->nr_in_flight[i])) { | |
3515 | spin_unlock_irq(&workqueue_lock); | |
6183c009 | 3516 | return; |
76af4d93 TH |
3517 | } |
3518 | } | |
3519 | ||
6183c009 | 3520 | if (WARN_ON(pwq->nr_active) || |
76af4d93 TH |
3521 | WARN_ON(!list_empty(&pwq->delayed_works))) { |
3522 | spin_unlock_irq(&workqueue_lock); | |
6183c009 | 3523 | return; |
76af4d93 | 3524 | } |
6183c009 TH |
3525 | } |
3526 | ||
a0a1a5fd TH |
3527 | /* |
3528 | * wq list is used to freeze wq, remove from list after | |
3529 | * flushing is complete in case freeze races us. | |
3530 | */ | |
b1f4ec17 | 3531 | list_del(&wq->list); |
76af4d93 | 3532 | |
e98d5b16 | 3533 | spin_unlock_irq(&workqueue_lock); |
3af24433 | 3534 | |
e22bee78 TH |
3535 | if (wq->flags & WQ_RESCUER) { |
3536 | kthread_stop(wq->rescuer->task); | |
8d9df9f0 | 3537 | kfree(wq->rescuer); |
e22bee78 TH |
3538 | } |
3539 | ||
29c91e99 TH |
3540 | /* |
3541 | * We're the sole accessor of @wq at this point. Directly access | |
3542 | * the first pwq and put its pool. | |
3543 | */ | |
3544 | if (wq->flags & WQ_UNBOUND) { | |
3545 | pwq = list_first_entry(&wq->pwqs, struct pool_workqueue, | |
3546 | pwqs_node); | |
3547 | put_unbound_pool(pwq->pool); | |
3548 | } | |
112202d9 | 3549 | free_pwqs(wq); |
3af24433 ON |
3550 | kfree(wq); |
3551 | } | |
3552 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
3553 | ||
9f4bd4cd | 3554 | /** |
112202d9 TH |
3555 | * pwq_set_max_active - adjust max_active of a pwq |
3556 | * @pwq: target pool_workqueue | |
9f4bd4cd LJ |
3557 | * @max_active: new max_active value. |
3558 | * | |
112202d9 | 3559 | * Set @pwq->max_active to @max_active and activate delayed works if |
9f4bd4cd LJ |
3560 | * increased. |
3561 | * | |
3562 | * CONTEXT: | |
d565ed63 | 3563 | * spin_lock_irq(pool->lock). |
9f4bd4cd | 3564 | */ |
112202d9 | 3565 | static void pwq_set_max_active(struct pool_workqueue *pwq, int max_active) |
9f4bd4cd | 3566 | { |
112202d9 | 3567 | pwq->max_active = max_active; |
9f4bd4cd | 3568 | |
112202d9 TH |
3569 | while (!list_empty(&pwq->delayed_works) && |
3570 | pwq->nr_active < pwq->max_active) | |
3571 | pwq_activate_first_delayed(pwq); | |
9f4bd4cd LJ |
3572 | } |
3573 | ||
dcd989cb TH |
3574 | /** |
3575 | * workqueue_set_max_active - adjust max_active of a workqueue | |
3576 | * @wq: target workqueue | |
3577 | * @max_active: new max_active value. | |
3578 | * | |
3579 | * Set max_active of @wq to @max_active. | |
3580 | * | |
3581 | * CONTEXT: | |
3582 | * Don't call from IRQ context. | |
3583 | */ | |
3584 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
3585 | { | |
49e3cf44 | 3586 | struct pool_workqueue *pwq; |
dcd989cb | 3587 | |
f3421797 | 3588 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb | 3589 | |
e98d5b16 | 3590 | spin_lock_irq(&workqueue_lock); |
dcd989cb TH |
3591 | |
3592 | wq->saved_max_active = max_active; | |
3593 | ||
49e3cf44 | 3594 | for_each_pwq(pwq, wq) { |
112202d9 | 3595 | struct worker_pool *pool = pwq->pool; |
dcd989cb | 3596 | |
e98d5b16 | 3597 | spin_lock(&pool->lock); |
dcd989cb | 3598 | |
58a69cb4 | 3599 | if (!(wq->flags & WQ_FREEZABLE) || |
35b6bb63 | 3600 | !(pool->flags & POOL_FREEZING)) |
112202d9 | 3601 | pwq_set_max_active(pwq, max_active); |
9bfb1839 | 3602 | |
e98d5b16 | 3603 | spin_unlock(&pool->lock); |
65a64464 | 3604 | } |
93981800 | 3605 | |
e98d5b16 | 3606 | spin_unlock_irq(&workqueue_lock); |
15316ba8 | 3607 | } |
dcd989cb | 3608 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 3609 | |
eef6a7d5 | 3610 | /** |
dcd989cb TH |
3611 | * workqueue_congested - test whether a workqueue is congested |
3612 | * @cpu: CPU in question | |
3613 | * @wq: target workqueue | |
eef6a7d5 | 3614 | * |
dcd989cb TH |
3615 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
3616 | * no synchronization around this function and the test result is | |
3617 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 3618 | * |
dcd989cb TH |
3619 | * RETURNS: |
3620 | * %true if congested, %false otherwise. | |
eef6a7d5 | 3621 | */ |
d84ff051 | 3622 | bool workqueue_congested(int cpu, struct workqueue_struct *wq) |
1da177e4 | 3623 | { |
7fb98ea7 | 3624 | struct pool_workqueue *pwq; |
76af4d93 TH |
3625 | bool ret; |
3626 | ||
3627 | preempt_disable(); | |
7fb98ea7 TH |
3628 | |
3629 | if (!(wq->flags & WQ_UNBOUND)) | |
3630 | pwq = per_cpu_ptr(wq->cpu_pwqs, cpu); | |
3631 | else | |
3632 | pwq = first_pwq(wq); | |
dcd989cb | 3633 | |
76af4d93 TH |
3634 | ret = !list_empty(&pwq->delayed_works); |
3635 | preempt_enable(); | |
3636 | ||
3637 | return ret; | |
1da177e4 | 3638 | } |
dcd989cb | 3639 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 3640 | |
dcd989cb TH |
3641 | /** |
3642 | * work_busy - test whether a work is currently pending or running | |
3643 | * @work: the work to be tested | |
3644 | * | |
3645 | * Test whether @work is currently pending or running. There is no | |
3646 | * synchronization around this function and the test result is | |
3647 | * unreliable and only useful as advisory hints or for debugging. | |
dcd989cb TH |
3648 | * |
3649 | * RETURNS: | |
3650 | * OR'd bitmask of WORK_BUSY_* bits. | |
3651 | */ | |
3652 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 3653 | { |
fa1b54e6 | 3654 | struct worker_pool *pool; |
dcd989cb TH |
3655 | unsigned long flags; |
3656 | unsigned int ret = 0; | |
1da177e4 | 3657 | |
dcd989cb TH |
3658 | if (work_pending(work)) |
3659 | ret |= WORK_BUSY_PENDING; | |
1da177e4 | 3660 | |
fa1b54e6 TH |
3661 | local_irq_save(flags); |
3662 | pool = get_work_pool(work); | |
038366c5 | 3663 | if (pool) { |
fa1b54e6 | 3664 | spin_lock(&pool->lock); |
038366c5 LJ |
3665 | if (find_worker_executing_work(pool, work)) |
3666 | ret |= WORK_BUSY_RUNNING; | |
fa1b54e6 | 3667 | spin_unlock(&pool->lock); |
038366c5 | 3668 | } |
fa1b54e6 | 3669 | local_irq_restore(flags); |
1da177e4 | 3670 | |
dcd989cb | 3671 | return ret; |
1da177e4 | 3672 | } |
dcd989cb | 3673 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 3674 | |
db7bccf4 TH |
3675 | /* |
3676 | * CPU hotplug. | |
3677 | * | |
e22bee78 | 3678 | * There are two challenges in supporting CPU hotplug. Firstly, there |
112202d9 | 3679 | * are a lot of assumptions on strong associations among work, pwq and |
706026c2 | 3680 | * pool which make migrating pending and scheduled works very |
e22bee78 | 3681 | * difficult to implement without impacting hot paths. Secondly, |
94cf58bb | 3682 | * worker pools serve mix of short, long and very long running works making |
e22bee78 TH |
3683 | * blocked draining impractical. |
3684 | * | |
24647570 | 3685 | * This is solved by allowing the pools to be disassociated from the CPU |
628c78e7 TH |
3686 | * running as an unbound one and allowing it to be reattached later if the |
3687 | * cpu comes back online. | |
db7bccf4 | 3688 | */ |
1da177e4 | 3689 | |
706026c2 | 3690 | static void wq_unbind_fn(struct work_struct *work) |
3af24433 | 3691 | { |
38db41d9 | 3692 | int cpu = smp_processor_id(); |
4ce62e9e | 3693 | struct worker_pool *pool; |
db7bccf4 | 3694 | struct worker *worker; |
db7bccf4 | 3695 | int i; |
3af24433 | 3696 | |
38db41d9 | 3697 | for_each_std_worker_pool(pool, cpu) { |
6183c009 | 3698 | WARN_ON_ONCE(cpu != smp_processor_id()); |
db7bccf4 | 3699 | |
94cf58bb TH |
3700 | mutex_lock(&pool->assoc_mutex); |
3701 | spin_lock_irq(&pool->lock); | |
3af24433 | 3702 | |
94cf58bb TH |
3703 | /* |
3704 | * We've claimed all manager positions. Make all workers | |
3705 | * unbound and set DISASSOCIATED. Before this, all workers | |
3706 | * except for the ones which are still executing works from | |
3707 | * before the last CPU down must be on the cpu. After | |
3708 | * this, they may become diasporas. | |
3709 | */ | |
4ce62e9e | 3710 | list_for_each_entry(worker, &pool->idle_list, entry) |
403c821d | 3711 | worker->flags |= WORKER_UNBOUND; |
3af24433 | 3712 | |
b67bfe0d | 3713 | for_each_busy_worker(worker, i, pool) |
c9e7cf27 | 3714 | worker->flags |= WORKER_UNBOUND; |
06ba38a9 | 3715 | |
24647570 | 3716 | pool->flags |= POOL_DISASSOCIATED; |
f2d5a0ee | 3717 | |
94cf58bb TH |
3718 | spin_unlock_irq(&pool->lock); |
3719 | mutex_unlock(&pool->assoc_mutex); | |
3720 | } | |
628c78e7 | 3721 | |
e22bee78 | 3722 | /* |
403c821d | 3723 | * Call schedule() so that we cross rq->lock and thus can guarantee |
628c78e7 TH |
3724 | * sched callbacks see the %WORKER_UNBOUND flag. This is necessary |
3725 | * as scheduler callbacks may be invoked from other cpus. | |
e22bee78 | 3726 | */ |
e22bee78 | 3727 | schedule(); |
06ba38a9 | 3728 | |
e22bee78 | 3729 | /* |
628c78e7 TH |
3730 | * Sched callbacks are disabled now. Zap nr_running. After this, |
3731 | * nr_running stays zero and need_more_worker() and keep_working() | |
38db41d9 TH |
3732 | * are always true as long as the worklist is not empty. Pools on |
3733 | * @cpu now behave as unbound (in terms of concurrency management) | |
3734 | * pools which are served by workers tied to the CPU. | |
628c78e7 TH |
3735 | * |
3736 | * On return from this function, the current worker would trigger | |
3737 | * unbound chain execution of pending work items if other workers | |
3738 | * didn't already. | |
e22bee78 | 3739 | */ |
38db41d9 | 3740 | for_each_std_worker_pool(pool, cpu) |
e19e397a | 3741 | atomic_set(&pool->nr_running, 0); |
3af24433 | 3742 | } |
3af24433 | 3743 | |
8db25e78 TH |
3744 | /* |
3745 | * Workqueues should be brought up before normal priority CPU notifiers. | |
3746 | * This will be registered high priority CPU notifier. | |
3747 | */ | |
9fdf9b73 | 3748 | static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb, |
8db25e78 TH |
3749 | unsigned long action, |
3750 | void *hcpu) | |
3af24433 | 3751 | { |
d84ff051 | 3752 | int cpu = (unsigned long)hcpu; |
4ce62e9e | 3753 | struct worker_pool *pool; |
3ce63377 | 3754 | |
8db25e78 | 3755 | switch (action & ~CPU_TASKS_FROZEN) { |
3af24433 | 3756 | case CPU_UP_PREPARE: |
38db41d9 | 3757 | for_each_std_worker_pool(pool, cpu) { |
3ce63377 TH |
3758 | struct worker *worker; |
3759 | ||
3760 | if (pool->nr_workers) | |
3761 | continue; | |
3762 | ||
3763 | worker = create_worker(pool); | |
3764 | if (!worker) | |
3765 | return NOTIFY_BAD; | |
3766 | ||
d565ed63 | 3767 | spin_lock_irq(&pool->lock); |
3ce63377 | 3768 | start_worker(worker); |
d565ed63 | 3769 | spin_unlock_irq(&pool->lock); |
3af24433 | 3770 | } |
8db25e78 | 3771 | break; |
3af24433 | 3772 | |
db7bccf4 TH |
3773 | case CPU_DOWN_FAILED: |
3774 | case CPU_ONLINE: | |
38db41d9 | 3775 | for_each_std_worker_pool(pool, cpu) { |
94cf58bb TH |
3776 | mutex_lock(&pool->assoc_mutex); |
3777 | spin_lock_irq(&pool->lock); | |
3778 | ||
24647570 | 3779 | pool->flags &= ~POOL_DISASSOCIATED; |
94cf58bb TH |
3780 | rebind_workers(pool); |
3781 | ||
3782 | spin_unlock_irq(&pool->lock); | |
3783 | mutex_unlock(&pool->assoc_mutex); | |
3784 | } | |
db7bccf4 | 3785 | break; |
00dfcaf7 | 3786 | } |
65758202 TH |
3787 | return NOTIFY_OK; |
3788 | } | |
3789 | ||
3790 | /* | |
3791 | * Workqueues should be brought down after normal priority CPU notifiers. | |
3792 | * This will be registered as low priority CPU notifier. | |
3793 | */ | |
9fdf9b73 | 3794 | static int __cpuinit workqueue_cpu_down_callback(struct notifier_block *nfb, |
65758202 TH |
3795 | unsigned long action, |
3796 | void *hcpu) | |
3797 | { | |
d84ff051 | 3798 | int cpu = (unsigned long)hcpu; |
8db25e78 TH |
3799 | struct work_struct unbind_work; |
3800 | ||
65758202 TH |
3801 | switch (action & ~CPU_TASKS_FROZEN) { |
3802 | case CPU_DOWN_PREPARE: | |
8db25e78 | 3803 | /* unbinding should happen on the local CPU */ |
706026c2 | 3804 | INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn); |
7635d2fd | 3805 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
8db25e78 TH |
3806 | flush_work(&unbind_work); |
3807 | break; | |
65758202 TH |
3808 | } |
3809 | return NOTIFY_OK; | |
3810 | } | |
3811 | ||
2d3854a3 | 3812 | #ifdef CONFIG_SMP |
8ccad40d | 3813 | |
2d3854a3 | 3814 | struct work_for_cpu { |
ed48ece2 | 3815 | struct work_struct work; |
2d3854a3 RR |
3816 | long (*fn)(void *); |
3817 | void *arg; | |
3818 | long ret; | |
3819 | }; | |
3820 | ||
ed48ece2 | 3821 | static void work_for_cpu_fn(struct work_struct *work) |
2d3854a3 | 3822 | { |
ed48ece2 TH |
3823 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
3824 | ||
2d3854a3 RR |
3825 | wfc->ret = wfc->fn(wfc->arg); |
3826 | } | |
3827 | ||
3828 | /** | |
3829 | * work_on_cpu - run a function in user context on a particular cpu | |
3830 | * @cpu: the cpu to run on | |
3831 | * @fn: the function to run | |
3832 | * @arg: the function arg | |
3833 | * | |
31ad9081 RR |
3834 | * This will return the value @fn returns. |
3835 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 3836 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 | 3837 | */ |
d84ff051 | 3838 | long work_on_cpu(int cpu, long (*fn)(void *), void *arg) |
2d3854a3 | 3839 | { |
ed48ece2 | 3840 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; |
6b44003e | 3841 | |
ed48ece2 TH |
3842 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
3843 | schedule_work_on(cpu, &wfc.work); | |
3844 | flush_work(&wfc.work); | |
2d3854a3 RR |
3845 | return wfc.ret; |
3846 | } | |
3847 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
3848 | #endif /* CONFIG_SMP */ | |
3849 | ||
a0a1a5fd TH |
3850 | #ifdef CONFIG_FREEZER |
3851 | ||
3852 | /** | |
3853 | * freeze_workqueues_begin - begin freezing workqueues | |
3854 | * | |
58a69cb4 TH |
3855 | * Start freezing workqueues. After this function returns, all freezable |
3856 | * workqueues will queue new works to their frozen_works list instead of | |
706026c2 | 3857 | * pool->worklist. |
a0a1a5fd TH |
3858 | * |
3859 | * CONTEXT: | |
d565ed63 | 3860 | * Grabs and releases workqueue_lock and pool->lock's. |
a0a1a5fd TH |
3861 | */ |
3862 | void freeze_workqueues_begin(void) | |
3863 | { | |
17116969 | 3864 | struct worker_pool *pool; |
24b8a847 TH |
3865 | struct workqueue_struct *wq; |
3866 | struct pool_workqueue *pwq; | |
17116969 | 3867 | int id; |
a0a1a5fd | 3868 | |
e98d5b16 | 3869 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3870 | |
6183c009 | 3871 | WARN_ON_ONCE(workqueue_freezing); |
a0a1a5fd TH |
3872 | workqueue_freezing = true; |
3873 | ||
24b8a847 | 3874 | /* set FREEZING */ |
17116969 | 3875 | for_each_pool(pool, id) { |
17116969 | 3876 | spin_lock(&pool->lock); |
17116969 TH |
3877 | WARN_ON_ONCE(pool->flags & POOL_FREEZING); |
3878 | pool->flags |= POOL_FREEZING; | |
24b8a847 TH |
3879 | spin_unlock(&pool->lock); |
3880 | } | |
a0a1a5fd | 3881 | |
24b8a847 TH |
3882 | /* suppress further executions by setting max_active to zero */ |
3883 | list_for_each_entry(wq, &workqueues, list) { | |
3884 | if (!(wq->flags & WQ_FREEZABLE)) | |
3885 | continue; | |
8b03ae3c | 3886 | |
24b8a847 TH |
3887 | for_each_pwq(pwq, wq) { |
3888 | spin_lock(&pwq->pool->lock); | |
3889 | pwq->max_active = 0; | |
3890 | spin_unlock(&pwq->pool->lock); | |
a1056305 | 3891 | } |
a0a1a5fd TH |
3892 | } |
3893 | ||
e98d5b16 | 3894 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3895 | } |
3896 | ||
3897 | /** | |
58a69cb4 | 3898 | * freeze_workqueues_busy - are freezable workqueues still busy? |
a0a1a5fd TH |
3899 | * |
3900 | * Check whether freezing is complete. This function must be called | |
3901 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
3902 | * | |
3903 | * CONTEXT: | |
3904 | * Grabs and releases workqueue_lock. | |
3905 | * | |
3906 | * RETURNS: | |
58a69cb4 TH |
3907 | * %true if some freezable workqueues are still busy. %false if freezing |
3908 | * is complete. | |
a0a1a5fd TH |
3909 | */ |
3910 | bool freeze_workqueues_busy(void) | |
3911 | { | |
a0a1a5fd | 3912 | bool busy = false; |
24b8a847 TH |
3913 | struct workqueue_struct *wq; |
3914 | struct pool_workqueue *pwq; | |
a0a1a5fd | 3915 | |
e98d5b16 | 3916 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd | 3917 | |
6183c009 | 3918 | WARN_ON_ONCE(!workqueue_freezing); |
a0a1a5fd | 3919 | |
24b8a847 TH |
3920 | list_for_each_entry(wq, &workqueues, list) { |
3921 | if (!(wq->flags & WQ_FREEZABLE)) | |
3922 | continue; | |
a0a1a5fd TH |
3923 | /* |
3924 | * nr_active is monotonically decreasing. It's safe | |
3925 | * to peek without lock. | |
3926 | */ | |
24b8a847 | 3927 | for_each_pwq(pwq, wq) { |
6183c009 | 3928 | WARN_ON_ONCE(pwq->nr_active < 0); |
112202d9 | 3929 | if (pwq->nr_active) { |
a0a1a5fd TH |
3930 | busy = true; |
3931 | goto out_unlock; | |
3932 | } | |
3933 | } | |
3934 | } | |
3935 | out_unlock: | |
e98d5b16 | 3936 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3937 | return busy; |
3938 | } | |
3939 | ||
3940 | /** | |
3941 | * thaw_workqueues - thaw workqueues | |
3942 | * | |
3943 | * Thaw workqueues. Normal queueing is restored and all collected | |
706026c2 | 3944 | * frozen works are transferred to their respective pool worklists. |
a0a1a5fd TH |
3945 | * |
3946 | * CONTEXT: | |
d565ed63 | 3947 | * Grabs and releases workqueue_lock and pool->lock's. |
a0a1a5fd TH |
3948 | */ |
3949 | void thaw_workqueues(void) | |
3950 | { | |
24b8a847 TH |
3951 | struct workqueue_struct *wq; |
3952 | struct pool_workqueue *pwq; | |
3953 | struct worker_pool *pool; | |
3954 | int id; | |
a0a1a5fd | 3955 | |
e98d5b16 | 3956 | spin_lock_irq(&workqueue_lock); |
a0a1a5fd TH |
3957 | |
3958 | if (!workqueue_freezing) | |
3959 | goto out_unlock; | |
3960 | ||
24b8a847 TH |
3961 | /* clear FREEZING */ |
3962 | for_each_pool(pool, id) { | |
3963 | spin_lock(&pool->lock); | |
3964 | WARN_ON_ONCE(!(pool->flags & POOL_FREEZING)); | |
3965 | pool->flags &= ~POOL_FREEZING; | |
3966 | spin_unlock(&pool->lock); | |
3967 | } | |
8b03ae3c | 3968 | |
24b8a847 TH |
3969 | /* restore max_active and repopulate worklist */ |
3970 | list_for_each_entry(wq, &workqueues, list) { | |
3971 | if (!(wq->flags & WQ_FREEZABLE)) | |
3972 | continue; | |
a1056305 | 3973 | |
24b8a847 TH |
3974 | for_each_pwq(pwq, wq) { |
3975 | spin_lock(&pwq->pool->lock); | |
3976 | pwq_set_max_active(pwq, wq->saved_max_active); | |
3977 | spin_unlock(&pwq->pool->lock); | |
d565ed63 | 3978 | } |
a0a1a5fd TH |
3979 | } |
3980 | ||
24b8a847 TH |
3981 | /* kick workers */ |
3982 | for_each_pool(pool, id) { | |
3983 | spin_lock(&pool->lock); | |
3984 | wake_up_worker(pool); | |
3985 | spin_unlock(&pool->lock); | |
3986 | } | |
3987 | ||
a0a1a5fd TH |
3988 | workqueue_freezing = false; |
3989 | out_unlock: | |
e98d5b16 | 3990 | spin_unlock_irq(&workqueue_lock); |
a0a1a5fd TH |
3991 | } |
3992 | #endif /* CONFIG_FREEZER */ | |
3993 | ||
6ee0578b | 3994 | static int __init init_workqueues(void) |
1da177e4 | 3995 | { |
7a4e344c TH |
3996 | int std_nice[NR_STD_WORKER_POOLS] = { 0, HIGHPRI_NICE_LEVEL }; |
3997 | int i, cpu; | |
c34056a3 | 3998 | |
7c3eed5c TH |
3999 | /* make sure we have enough bits for OFFQ pool ID */ |
4000 | BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_POOL_SHIFT)) < | |
6be19588 | 4001 | WORK_CPU_END * NR_STD_WORKER_POOLS); |
b5490077 | 4002 | |
e904e6c2 TH |
4003 | WARN_ON(__alignof__(struct pool_workqueue) < __alignof__(long long)); |
4004 | ||
4005 | pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC); | |
4006 | ||
65758202 | 4007 | cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); |
a5b4e57d | 4008 | hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); |
8b03ae3c | 4009 | |
706026c2 | 4010 | /* initialize CPU pools */ |
29c91e99 | 4011 | for_each_possible_cpu(cpu) { |
4ce62e9e | 4012 | struct worker_pool *pool; |
8b03ae3c | 4013 | |
7a4e344c | 4014 | i = 0; |
38db41d9 | 4015 | for_each_std_worker_pool(pool, cpu) { |
7a4e344c | 4016 | BUG_ON(init_worker_pool(pool)); |
ec22ca5e | 4017 | pool->cpu = cpu; |
29c91e99 | 4018 | cpumask_copy(pool->attrs->cpumask, cpumask_of(cpu)); |
7a4e344c TH |
4019 | pool->attrs->nice = std_nice[i++]; |
4020 | ||
9daf9e67 TH |
4021 | /* alloc pool ID */ |
4022 | BUG_ON(worker_pool_assign_id(pool)); | |
4ce62e9e | 4023 | } |
8b03ae3c TH |
4024 | } |
4025 | ||
e22bee78 | 4026 | /* create the initial worker */ |
29c91e99 | 4027 | for_each_online_cpu(cpu) { |
4ce62e9e | 4028 | struct worker_pool *pool; |
e22bee78 | 4029 | |
38db41d9 | 4030 | for_each_std_worker_pool(pool, cpu) { |
4ce62e9e TH |
4031 | struct worker *worker; |
4032 | ||
29c91e99 | 4033 | pool->flags &= ~POOL_DISASSOCIATED; |
24647570 | 4034 | |
bc2ae0f5 | 4035 | worker = create_worker(pool); |
4ce62e9e | 4036 | BUG_ON(!worker); |
d565ed63 | 4037 | spin_lock_irq(&pool->lock); |
4ce62e9e | 4038 | start_worker(worker); |
d565ed63 | 4039 | spin_unlock_irq(&pool->lock); |
4ce62e9e | 4040 | } |
e22bee78 TH |
4041 | } |
4042 | ||
29c91e99 TH |
4043 | /* create default unbound wq attrs */ |
4044 | for (i = 0; i < NR_STD_WORKER_POOLS; i++) { | |
4045 | struct workqueue_attrs *attrs; | |
4046 | ||
4047 | BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL))); | |
4048 | ||
4049 | attrs->nice = std_nice[i]; | |
4050 | cpumask_setall(attrs->cpumask); | |
4051 | ||
4052 | unbound_std_wq_attrs[i] = attrs; | |
4053 | } | |
4054 | ||
d320c038 | 4055 | system_wq = alloc_workqueue("events", 0, 0); |
1aabe902 | 4056 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
d320c038 | 4057 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
f3421797 TH |
4058 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
4059 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
4060 | system_freezable_wq = alloc_workqueue("events_freezable", |
4061 | WQ_FREEZABLE, 0); | |
1aabe902 | 4062 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
ae930e0f | 4063 | !system_unbound_wq || !system_freezable_wq); |
6ee0578b | 4064 | return 0; |
1da177e4 | 4065 | } |
6ee0578b | 4066 | early_initcall(init_workqueues); |