]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
c54fce6e | 2 | * kernel/workqueue.c - generic async execution with shared worker pool |
1da177e4 | 3 | * |
c54fce6e | 4 | * Copyright (C) 2002 Ingo Molnar |
1da177e4 | 5 | * |
c54fce6e TH |
6 | * Derived from the taskqueue/keventd code by: |
7 | * David Woodhouse <dwmw2@infradead.org> | |
8 | * Andrew Morton | |
9 | * Kai Petzke <wpp@marie.physik.tu-berlin.de> | |
10 | * Theodore Ts'o <tytso@mit.edu> | |
1da177e4 | 11 | * |
c54fce6e | 12 | * Made to use alloc_percpu by Christoph Lameter. |
1da177e4 | 13 | * |
c54fce6e TH |
14 | * Copyright (C) 2010 SUSE Linux Products GmbH |
15 | * Copyright (C) 2010 Tejun Heo <tj@kernel.org> | |
89ada679 | 16 | * |
c54fce6e TH |
17 | * This is the generic async execution mechanism. Work items as are |
18 | * executed in process context. The worker pool is shared and | |
19 | * automatically managed. There is one worker pool for each CPU and | |
20 | * one extra for works which are better served by workers which are | |
21 | * not bound to any specific CPU. | |
22 | * | |
23 | * Please read Documentation/workqueue.txt for details. | |
1da177e4 LT |
24 | */ |
25 | ||
9984de1a | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/kernel.h> |
28 | #include <linux/sched.h> | |
29 | #include <linux/init.h> | |
30 | #include <linux/signal.h> | |
31 | #include <linux/completion.h> | |
32 | #include <linux/workqueue.h> | |
33 | #include <linux/slab.h> | |
34 | #include <linux/cpu.h> | |
35 | #include <linux/notifier.h> | |
36 | #include <linux/kthread.h> | |
1fa44eca | 37 | #include <linux/hardirq.h> |
46934023 | 38 | #include <linux/mempolicy.h> |
341a5958 | 39 | #include <linux/freezer.h> |
d5abe669 PZ |
40 | #include <linux/kallsyms.h> |
41 | #include <linux/debug_locks.h> | |
4e6045f1 | 42 | #include <linux/lockdep.h> |
c34056a3 | 43 | #include <linux/idr.h> |
42f8570f | 44 | #include <linux/hashtable.h> |
e22bee78 | 45 | |
ea138446 | 46 | #include "workqueue_internal.h" |
1da177e4 | 47 | |
c8e55f36 | 48 | enum { |
24647570 TH |
49 | /* |
50 | * worker_pool flags | |
bc2ae0f5 | 51 | * |
24647570 | 52 | * A bound pool is either associated or disassociated with its CPU. |
bc2ae0f5 TH |
53 | * While associated (!DISASSOCIATED), all workers are bound to the |
54 | * CPU and none has %WORKER_UNBOUND set and concurrency management | |
55 | * is in effect. | |
56 | * | |
57 | * While DISASSOCIATED, the cpu may be offline and all workers have | |
58 | * %WORKER_UNBOUND set and concurrency management disabled, and may | |
24647570 | 59 | * be executing on any CPU. The pool behaves as an unbound one. |
bc2ae0f5 TH |
60 | * |
61 | * Note that DISASSOCIATED can be flipped only while holding | |
24647570 TH |
62 | * assoc_mutex to avoid changing binding state while |
63 | * create_worker() is in progress. | |
bc2ae0f5 | 64 | */ |
11ebea50 | 65 | POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */ |
552a37e9 | 66 | POOL_MANAGING_WORKERS = 1 << 1, /* managing workers */ |
24647570 | 67 | POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ |
35b6bb63 | 68 | POOL_FREEZING = 1 << 3, /* freeze in progress */ |
db7bccf4 | 69 | |
c8e55f36 TH |
70 | /* worker flags */ |
71 | WORKER_STARTED = 1 << 0, /* started */ | |
72 | WORKER_DIE = 1 << 1, /* die die die */ | |
73 | WORKER_IDLE = 1 << 2, /* is idle */ | |
e22bee78 | 74 | WORKER_PREP = 1 << 3, /* preparing to run works */ |
fb0e7beb | 75 | WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ |
f3421797 | 76 | WORKER_UNBOUND = 1 << 7, /* worker is unbound */ |
e22bee78 | 77 | |
5f7dabfd | 78 | WORKER_NOT_RUNNING = WORKER_PREP | WORKER_UNBOUND | |
403c821d | 79 | WORKER_CPU_INTENSIVE, |
db7bccf4 | 80 | |
e34cdddb | 81 | NR_STD_WORKER_POOLS = 2, /* # standard pools per cpu */ |
4ce62e9e | 82 | |
c8e55f36 | 83 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ |
db7bccf4 | 84 | |
e22bee78 TH |
85 | MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */ |
86 | IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */ | |
87 | ||
3233cdbd TH |
88 | MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2, |
89 | /* call for help after 10ms | |
90 | (min two ticks) */ | |
e22bee78 TH |
91 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ |
92 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
e22bee78 TH |
93 | |
94 | /* | |
95 | * Rescue workers are used only on emergencies and shared by | |
96 | * all cpus. Give -20. | |
97 | */ | |
98 | RESCUER_NICE_LEVEL = -20, | |
3270476a | 99 | HIGHPRI_NICE_LEVEL = -20, |
c8e55f36 | 100 | }; |
1da177e4 LT |
101 | |
102 | /* | |
4690c4ab TH |
103 | * Structure fields follow one of the following exclusion rules. |
104 | * | |
e41e704b TH |
105 | * I: Modifiable by initialization/destruction paths and read-only for |
106 | * everyone else. | |
4690c4ab | 107 | * |
e22bee78 TH |
108 | * P: Preemption protected. Disabling preemption is enough and should |
109 | * only be modified and accessed from the local cpu. | |
110 | * | |
8b03ae3c | 111 | * L: gcwq->lock protected. Access with gcwq->lock held. |
4690c4ab | 112 | * |
e22bee78 TH |
113 | * X: During normal operation, modification requires gcwq->lock and |
114 | * should be done only from local cpu. Either disabling preemption | |
115 | * on local cpu or grabbing gcwq->lock is enough for read access. | |
24647570 | 116 | * If POOL_DISASSOCIATED is set, it's identical to L. |
e22bee78 | 117 | * |
73f53c4a TH |
118 | * F: wq->flush_mutex protected. |
119 | * | |
4690c4ab | 120 | * W: workqueue_lock protected. |
1da177e4 | 121 | */ |
1da177e4 | 122 | |
2eaebdb3 | 123 | /* struct worker is defined in workqueue_internal.h */ |
c34056a3 | 124 | |
bd7bdd43 TH |
125 | struct worker_pool { |
126 | struct global_cwq *gcwq; /* I: the owning gcwq */ | |
9daf9e67 | 127 | int id; /* I: pool ID */ |
11ebea50 | 128 | unsigned int flags; /* X: flags */ |
bd7bdd43 TH |
129 | |
130 | struct list_head worklist; /* L: list of pending works */ | |
131 | int nr_workers; /* L: total number of workers */ | |
ea1abd61 LJ |
132 | |
133 | /* nr_idle includes the ones off idle_list for rebinding */ | |
bd7bdd43 TH |
134 | int nr_idle; /* L: currently idle ones */ |
135 | ||
136 | struct list_head idle_list; /* X: list of idle workers */ | |
137 | struct timer_list idle_timer; /* L: worker idle timeout */ | |
138 | struct timer_list mayday_timer; /* L: SOS timer for workers */ | |
139 | ||
24647570 | 140 | struct mutex assoc_mutex; /* protect POOL_DISASSOCIATED */ |
bd7bdd43 | 141 | struct ida worker_ida; /* L: for worker IDs */ |
bd7bdd43 TH |
142 | }; |
143 | ||
8b03ae3c | 144 | /* |
e22bee78 TH |
145 | * Global per-cpu workqueue. There's one and only one for each cpu |
146 | * and all works are queued and processed here regardless of their | |
147 | * target workqueues. | |
8b03ae3c TH |
148 | */ |
149 | struct global_cwq { | |
150 | spinlock_t lock; /* the gcwq lock */ | |
151 | unsigned int cpu; /* I: the associated cpu */ | |
c8e55f36 | 152 | |
bd7bdd43 | 153 | /* workers are chained either in busy_hash or pool idle_list */ |
42f8570f | 154 | DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER); |
c8e55f36 TH |
155 | /* L: hash of busy workers */ |
156 | ||
e34cdddb | 157 | struct worker_pool pools[NR_STD_WORKER_POOLS]; |
330dad5b | 158 | /* normal and highpri pools */ |
8b03ae3c TH |
159 | } ____cacheline_aligned_in_smp; |
160 | ||
1da177e4 | 161 | /* |
502ca9d8 | 162 | * The per-CPU workqueue. The lower WORK_STRUCT_FLAG_BITS of |
0f900049 TH |
163 | * work_struct->data are used for flags and thus cwqs need to be |
164 | * aligned at two's power of the number of flag bits. | |
1da177e4 LT |
165 | */ |
166 | struct cpu_workqueue_struct { | |
bd7bdd43 | 167 | struct worker_pool *pool; /* I: the associated pool */ |
4690c4ab | 168 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
169 | int work_color; /* L: current color */ |
170 | int flush_color; /* L: flushing color */ | |
171 | int nr_in_flight[WORK_NR_COLORS]; | |
172 | /* L: nr of in_flight works */ | |
1e19ffc6 | 173 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 174 | int max_active; /* L: max active works */ |
1e19ffc6 | 175 | struct list_head delayed_works; /* L: delayed works */ |
0f900049 | 176 | }; |
1da177e4 | 177 | |
73f53c4a TH |
178 | /* |
179 | * Structure used to wait for workqueue flush. | |
180 | */ | |
181 | struct wq_flusher { | |
182 | struct list_head list; /* F: list of flushers */ | |
183 | int flush_color; /* F: flush color waiting for */ | |
184 | struct completion done; /* flush completion */ | |
185 | }; | |
186 | ||
f2e005aa TH |
187 | /* |
188 | * All cpumasks are assumed to be always set on UP and thus can't be | |
189 | * used to determine whether there's something to be done. | |
190 | */ | |
191 | #ifdef CONFIG_SMP | |
192 | typedef cpumask_var_t mayday_mask_t; | |
193 | #define mayday_test_and_set_cpu(cpu, mask) \ | |
194 | cpumask_test_and_set_cpu((cpu), (mask)) | |
195 | #define mayday_clear_cpu(cpu, mask) cpumask_clear_cpu((cpu), (mask)) | |
196 | #define for_each_mayday_cpu(cpu, mask) for_each_cpu((cpu), (mask)) | |
9c37547a | 197 | #define alloc_mayday_mask(maskp, gfp) zalloc_cpumask_var((maskp), (gfp)) |
f2e005aa TH |
198 | #define free_mayday_mask(mask) free_cpumask_var((mask)) |
199 | #else | |
200 | typedef unsigned long mayday_mask_t; | |
201 | #define mayday_test_and_set_cpu(cpu, mask) test_and_set_bit(0, &(mask)) | |
202 | #define mayday_clear_cpu(cpu, mask) clear_bit(0, &(mask)) | |
203 | #define for_each_mayday_cpu(cpu, mask) if ((cpu) = 0, (mask)) | |
204 | #define alloc_mayday_mask(maskp, gfp) true | |
205 | #define free_mayday_mask(mask) do { } while (0) | |
206 | #endif | |
1da177e4 LT |
207 | |
208 | /* | |
209 | * The externally visible workqueue abstraction is an array of | |
210 | * per-CPU workqueues: | |
211 | */ | |
212 | struct workqueue_struct { | |
9c5a2ba7 | 213 | unsigned int flags; /* W: WQ_* flags */ |
bdbc5dd7 TH |
214 | union { |
215 | struct cpu_workqueue_struct __percpu *pcpu; | |
216 | struct cpu_workqueue_struct *single; | |
217 | unsigned long v; | |
218 | } cpu_wq; /* I: cwq's */ | |
4690c4ab | 219 | struct list_head list; /* W: list of all workqueues */ |
73f53c4a TH |
220 | |
221 | struct mutex flush_mutex; /* protects wq flushing */ | |
222 | int work_color; /* F: current work color */ | |
223 | int flush_color; /* F: current flush color */ | |
224 | atomic_t nr_cwqs_to_flush; /* flush in progress */ | |
225 | struct wq_flusher *first_flusher; /* F: first flusher */ | |
226 | struct list_head flusher_queue; /* F: flush waiters */ | |
227 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
228 | ||
f2e005aa | 229 | mayday_mask_t mayday_mask; /* cpus requesting rescue */ |
e22bee78 TH |
230 | struct worker *rescuer; /* I: rescue worker */ |
231 | ||
9c5a2ba7 | 232 | int nr_drainers; /* W: drain in progress */ |
dcd989cb | 233 | int saved_max_active; /* W: saved cwq max_active */ |
4e6045f1 | 234 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 235 | struct lockdep_map lockdep_map; |
4e6045f1 | 236 | #endif |
b196be89 | 237 | char name[]; /* I: workqueue name */ |
1da177e4 LT |
238 | }; |
239 | ||
d320c038 | 240 | struct workqueue_struct *system_wq __read_mostly; |
d320c038 | 241 | EXPORT_SYMBOL_GPL(system_wq); |
044c782c | 242 | struct workqueue_struct *system_highpri_wq __read_mostly; |
1aabe902 | 243 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
044c782c | 244 | struct workqueue_struct *system_long_wq __read_mostly; |
d320c038 | 245 | EXPORT_SYMBOL_GPL(system_long_wq); |
044c782c | 246 | struct workqueue_struct *system_unbound_wq __read_mostly; |
f3421797 | 247 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
044c782c | 248 | struct workqueue_struct *system_freezable_wq __read_mostly; |
24d51add | 249 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
d320c038 | 250 | |
97bd2347 TH |
251 | #define CREATE_TRACE_POINTS |
252 | #include <trace/events/workqueue.h> | |
253 | ||
4ce62e9e | 254 | #define for_each_worker_pool(pool, gcwq) \ |
3270476a | 255 | for ((pool) = &(gcwq)->pools[0]; \ |
e34cdddb | 256 | (pool) < &(gcwq)->pools[NR_STD_WORKER_POOLS]; (pool)++) |
4ce62e9e | 257 | |
db7bccf4 | 258 | #define for_each_busy_worker(worker, i, pos, gcwq) \ |
42f8570f | 259 | hash_for_each(gcwq->busy_hash, i, pos, worker, hentry) |
db7bccf4 | 260 | |
f3421797 TH |
261 | static inline int __next_gcwq_cpu(int cpu, const struct cpumask *mask, |
262 | unsigned int sw) | |
263 | { | |
264 | if (cpu < nr_cpu_ids) { | |
265 | if (sw & 1) { | |
266 | cpu = cpumask_next(cpu, mask); | |
267 | if (cpu < nr_cpu_ids) | |
268 | return cpu; | |
269 | } | |
270 | if (sw & 2) | |
271 | return WORK_CPU_UNBOUND; | |
272 | } | |
273 | return WORK_CPU_NONE; | |
274 | } | |
275 | ||
276 | static inline int __next_wq_cpu(int cpu, const struct cpumask *mask, | |
277 | struct workqueue_struct *wq) | |
278 | { | |
279 | return __next_gcwq_cpu(cpu, mask, !(wq->flags & WQ_UNBOUND) ? 1 : 2); | |
280 | } | |
281 | ||
09884951 TH |
282 | /* |
283 | * CPU iterators | |
284 | * | |
285 | * An extra gcwq is defined for an invalid cpu number | |
286 | * (WORK_CPU_UNBOUND) to host workqueues which are not bound to any | |
287 | * specific CPU. The following iterators are similar to | |
288 | * for_each_*_cpu() iterators but also considers the unbound gcwq. | |
289 | * | |
290 | * for_each_gcwq_cpu() : possible CPUs + WORK_CPU_UNBOUND | |
291 | * for_each_online_gcwq_cpu() : online CPUs + WORK_CPU_UNBOUND | |
292 | * for_each_cwq_cpu() : possible CPUs for bound workqueues, | |
293 | * WORK_CPU_UNBOUND for unbound workqueues | |
294 | */ | |
f3421797 TH |
295 | #define for_each_gcwq_cpu(cpu) \ |
296 | for ((cpu) = __next_gcwq_cpu(-1, cpu_possible_mask, 3); \ | |
297 | (cpu) < WORK_CPU_NONE; \ | |
298 | (cpu) = __next_gcwq_cpu((cpu), cpu_possible_mask, 3)) | |
299 | ||
300 | #define for_each_online_gcwq_cpu(cpu) \ | |
301 | for ((cpu) = __next_gcwq_cpu(-1, cpu_online_mask, 3); \ | |
302 | (cpu) < WORK_CPU_NONE; \ | |
303 | (cpu) = __next_gcwq_cpu((cpu), cpu_online_mask, 3)) | |
304 | ||
305 | #define for_each_cwq_cpu(cpu, wq) \ | |
306 | for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, (wq)); \ | |
307 | (cpu) < WORK_CPU_NONE; \ | |
308 | (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, (wq))) | |
309 | ||
dc186ad7 TG |
310 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
311 | ||
312 | static struct debug_obj_descr work_debug_descr; | |
313 | ||
99777288 SG |
314 | static void *work_debug_hint(void *addr) |
315 | { | |
316 | return ((struct work_struct *) addr)->func; | |
317 | } | |
318 | ||
dc186ad7 TG |
319 | /* |
320 | * fixup_init is called when: | |
321 | * - an active object is initialized | |
322 | */ | |
323 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
324 | { | |
325 | struct work_struct *work = addr; | |
326 | ||
327 | switch (state) { | |
328 | case ODEBUG_STATE_ACTIVE: | |
329 | cancel_work_sync(work); | |
330 | debug_object_init(work, &work_debug_descr); | |
331 | return 1; | |
332 | default: | |
333 | return 0; | |
334 | } | |
335 | } | |
336 | ||
337 | /* | |
338 | * fixup_activate is called when: | |
339 | * - an active object is activated | |
340 | * - an unknown object is activated (might be a statically initialized object) | |
341 | */ | |
342 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
343 | { | |
344 | struct work_struct *work = addr; | |
345 | ||
346 | switch (state) { | |
347 | ||
348 | case ODEBUG_STATE_NOTAVAILABLE: | |
349 | /* | |
350 | * This is not really a fixup. The work struct was | |
351 | * statically initialized. We just make sure that it | |
352 | * is tracked in the object tracker. | |
353 | */ | |
22df02bb | 354 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
355 | debug_object_init(work, &work_debug_descr); |
356 | debug_object_activate(work, &work_debug_descr); | |
357 | return 0; | |
358 | } | |
359 | WARN_ON_ONCE(1); | |
360 | return 0; | |
361 | ||
362 | case ODEBUG_STATE_ACTIVE: | |
363 | WARN_ON(1); | |
364 | ||
365 | default: | |
366 | return 0; | |
367 | } | |
368 | } | |
369 | ||
370 | /* | |
371 | * fixup_free is called when: | |
372 | * - an active object is freed | |
373 | */ | |
374 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
375 | { | |
376 | struct work_struct *work = addr; | |
377 | ||
378 | switch (state) { | |
379 | case ODEBUG_STATE_ACTIVE: | |
380 | cancel_work_sync(work); | |
381 | debug_object_free(work, &work_debug_descr); | |
382 | return 1; | |
383 | default: | |
384 | return 0; | |
385 | } | |
386 | } | |
387 | ||
388 | static struct debug_obj_descr work_debug_descr = { | |
389 | .name = "work_struct", | |
99777288 | 390 | .debug_hint = work_debug_hint, |
dc186ad7 TG |
391 | .fixup_init = work_fixup_init, |
392 | .fixup_activate = work_fixup_activate, | |
393 | .fixup_free = work_fixup_free, | |
394 | }; | |
395 | ||
396 | static inline void debug_work_activate(struct work_struct *work) | |
397 | { | |
398 | debug_object_activate(work, &work_debug_descr); | |
399 | } | |
400 | ||
401 | static inline void debug_work_deactivate(struct work_struct *work) | |
402 | { | |
403 | debug_object_deactivate(work, &work_debug_descr); | |
404 | } | |
405 | ||
406 | void __init_work(struct work_struct *work, int onstack) | |
407 | { | |
408 | if (onstack) | |
409 | debug_object_init_on_stack(work, &work_debug_descr); | |
410 | else | |
411 | debug_object_init(work, &work_debug_descr); | |
412 | } | |
413 | EXPORT_SYMBOL_GPL(__init_work); | |
414 | ||
415 | void destroy_work_on_stack(struct work_struct *work) | |
416 | { | |
417 | debug_object_free(work, &work_debug_descr); | |
418 | } | |
419 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
420 | ||
421 | #else | |
422 | static inline void debug_work_activate(struct work_struct *work) { } | |
423 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
424 | #endif | |
425 | ||
95402b38 GS |
426 | /* Serializes the accesses to the list of workqueues. */ |
427 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 428 | static LIST_HEAD(workqueues); |
a0a1a5fd | 429 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 430 | |
e22bee78 TH |
431 | /* |
432 | * The almighty global cpu workqueues. nr_running is the only field | |
433 | * which is expected to be used frequently by other cpus via | |
434 | * try_to_wake_up(). Put it in a separate cacheline. | |
435 | */ | |
8b03ae3c | 436 | static DEFINE_PER_CPU(struct global_cwq, global_cwq); |
e34cdddb | 437 | static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, pool_nr_running[NR_STD_WORKER_POOLS]); |
8b03ae3c | 438 | |
f3421797 | 439 | /* |
24647570 TH |
440 | * Global cpu workqueue and nr_running counter for unbound gcwq. The pools |
441 | * for online CPUs have POOL_DISASSOCIATED set, and all their workers have | |
442 | * WORKER_UNBOUND set. | |
f3421797 TH |
443 | */ |
444 | static struct global_cwq unbound_global_cwq; | |
e34cdddb TH |
445 | static atomic_t unbound_pool_nr_running[NR_STD_WORKER_POOLS] = { |
446 | [0 ... NR_STD_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */ | |
4ce62e9e | 447 | }; |
f3421797 | 448 | |
9daf9e67 TH |
449 | /* idr of all pools */ |
450 | static DEFINE_MUTEX(worker_pool_idr_mutex); | |
451 | static DEFINE_IDR(worker_pool_idr); | |
452 | ||
c34056a3 | 453 | static int worker_thread(void *__worker); |
e2905b29 | 454 | static unsigned int work_cpu(struct work_struct *work); |
1da177e4 | 455 | |
e34cdddb | 456 | static int std_worker_pool_pri(struct worker_pool *pool) |
3270476a TH |
457 | { |
458 | return pool - pool->gcwq->pools; | |
459 | } | |
460 | ||
8b03ae3c TH |
461 | static struct global_cwq *get_gcwq(unsigned int cpu) |
462 | { | |
f3421797 TH |
463 | if (cpu != WORK_CPU_UNBOUND) |
464 | return &per_cpu(global_cwq, cpu); | |
465 | else | |
466 | return &unbound_global_cwq; | |
8b03ae3c TH |
467 | } |
468 | ||
9daf9e67 TH |
469 | /* allocate ID and assign it to @pool */ |
470 | static int worker_pool_assign_id(struct worker_pool *pool) | |
471 | { | |
472 | int ret; | |
473 | ||
474 | mutex_lock(&worker_pool_idr_mutex); | |
475 | idr_pre_get(&worker_pool_idr, GFP_KERNEL); | |
476 | ret = idr_get_new(&worker_pool_idr, pool, &pool->id); | |
477 | mutex_unlock(&worker_pool_idr_mutex); | |
478 | ||
479 | return ret; | |
480 | } | |
481 | ||
63d95a91 | 482 | static atomic_t *get_pool_nr_running(struct worker_pool *pool) |
e22bee78 | 483 | { |
63d95a91 | 484 | int cpu = pool->gcwq->cpu; |
e34cdddb | 485 | int idx = std_worker_pool_pri(pool); |
63d95a91 | 486 | |
f3421797 | 487 | if (cpu != WORK_CPU_UNBOUND) |
4ce62e9e | 488 | return &per_cpu(pool_nr_running, cpu)[idx]; |
f3421797 | 489 | else |
4ce62e9e | 490 | return &unbound_pool_nr_running[idx]; |
e22bee78 TH |
491 | } |
492 | ||
1537663f TH |
493 | static struct cpu_workqueue_struct *get_cwq(unsigned int cpu, |
494 | struct workqueue_struct *wq) | |
b1f4ec17 | 495 | { |
f3421797 | 496 | if (!(wq->flags & WQ_UNBOUND)) { |
e06ffa1e | 497 | if (likely(cpu < nr_cpu_ids)) |
f3421797 | 498 | return per_cpu_ptr(wq->cpu_wq.pcpu, cpu); |
f3421797 TH |
499 | } else if (likely(cpu == WORK_CPU_UNBOUND)) |
500 | return wq->cpu_wq.single; | |
501 | return NULL; | |
b1f4ec17 ON |
502 | } |
503 | ||
73f53c4a TH |
504 | static unsigned int work_color_to_flags(int color) |
505 | { | |
506 | return color << WORK_STRUCT_COLOR_SHIFT; | |
507 | } | |
508 | ||
509 | static int get_work_color(struct work_struct *work) | |
510 | { | |
511 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
512 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
513 | } | |
514 | ||
515 | static int work_next_color(int color) | |
516 | { | |
517 | return (color + 1) % WORK_NR_COLORS; | |
518 | } | |
1da177e4 | 519 | |
14441960 | 520 | /* |
b5490077 TH |
521 | * While queued, %WORK_STRUCT_CWQ is set and non flag bits of a work's data |
522 | * contain the pointer to the queued cwq. Once execution starts, the flag | |
523 | * is cleared and the high bits contain OFFQ flags and CPU number. | |
7a22ad75 | 524 | * |
bbb68dfa TH |
525 | * set_work_cwq(), set_work_cpu_and_clear_pending(), mark_work_canceling() |
526 | * and clear_work_data() can be used to set the cwq, cpu or clear | |
527 | * work->data. These functions should only be called while the work is | |
528 | * owned - ie. while the PENDING bit is set. | |
7a22ad75 | 529 | * |
bbb68dfa TH |
530 | * get_work_[g]cwq() can be used to obtain the gcwq or cwq corresponding to |
531 | * a work. gcwq is available once the work has been queued anywhere after | |
532 | * initialization until it is sync canceled. cwq is available only while | |
533 | * the work item is queued. | |
7a22ad75 | 534 | * |
bbb68dfa TH |
535 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
536 | * canceled. While being canceled, a work item may have its PENDING set | |
537 | * but stay off timer and worklist for arbitrarily long and nobody should | |
538 | * try to steal the PENDING bit. | |
14441960 | 539 | */ |
7a22ad75 TH |
540 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
541 | unsigned long flags) | |
365970a1 | 542 | { |
4594bf15 | 543 | BUG_ON(!work_pending(work)); |
7a22ad75 TH |
544 | atomic_long_set(&work->data, data | flags | work_static(work)); |
545 | } | |
365970a1 | 546 | |
7a22ad75 TH |
547 | static void set_work_cwq(struct work_struct *work, |
548 | struct cpu_workqueue_struct *cwq, | |
549 | unsigned long extra_flags) | |
550 | { | |
551 | set_work_data(work, (unsigned long)cwq, | |
e120153d | 552 | WORK_STRUCT_PENDING | WORK_STRUCT_CWQ | extra_flags); |
365970a1 DH |
553 | } |
554 | ||
8930caba TH |
555 | static void set_work_cpu_and_clear_pending(struct work_struct *work, |
556 | unsigned int cpu) | |
7a22ad75 | 557 | { |
23657bb1 TH |
558 | /* |
559 | * The following wmb is paired with the implied mb in | |
560 | * test_and_set_bit(PENDING) and ensures all updates to @work made | |
561 | * here are visible to and precede any updates by the next PENDING | |
562 | * owner. | |
563 | */ | |
564 | smp_wmb(); | |
b5490077 | 565 | set_work_data(work, (unsigned long)cpu << WORK_OFFQ_CPU_SHIFT, 0); |
7a22ad75 | 566 | } |
f756d5e2 | 567 | |
7a22ad75 | 568 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 569 | { |
23657bb1 | 570 | smp_wmb(); /* see set_work_cpu_and_clear_pending() */ |
7a22ad75 | 571 | set_work_data(work, WORK_STRUCT_NO_CPU, 0); |
1da177e4 LT |
572 | } |
573 | ||
7a22ad75 | 574 | static struct cpu_workqueue_struct *get_work_cwq(struct work_struct *work) |
b1f4ec17 | 575 | { |
e120153d | 576 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 577 | |
e120153d TH |
578 | if (data & WORK_STRUCT_CWQ) |
579 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); | |
580 | else | |
581 | return NULL; | |
4d707b9f ON |
582 | } |
583 | ||
7a22ad75 | 584 | static struct global_cwq *get_work_gcwq(struct work_struct *work) |
365970a1 | 585 | { |
e120153d | 586 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 TH |
587 | unsigned int cpu; |
588 | ||
e120153d TH |
589 | if (data & WORK_STRUCT_CWQ) |
590 | return ((struct cpu_workqueue_struct *) | |
bd7bdd43 | 591 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->gcwq; |
7a22ad75 | 592 | |
b5490077 | 593 | cpu = data >> WORK_OFFQ_CPU_SHIFT; |
715b06b8 | 594 | if (cpu == WORK_OFFQ_CPU_NONE) |
7a22ad75 TH |
595 | return NULL; |
596 | ||
f3421797 | 597 | BUG_ON(cpu >= nr_cpu_ids && cpu != WORK_CPU_UNBOUND); |
7a22ad75 | 598 | return get_gcwq(cpu); |
b1f4ec17 ON |
599 | } |
600 | ||
bbb68dfa TH |
601 | static void mark_work_canceling(struct work_struct *work) |
602 | { | |
603 | struct global_cwq *gcwq = get_work_gcwq(work); | |
715b06b8 | 604 | unsigned long cpu = gcwq ? gcwq->cpu : WORK_OFFQ_CPU_NONE; |
bbb68dfa TH |
605 | |
606 | set_work_data(work, (cpu << WORK_OFFQ_CPU_SHIFT) | WORK_OFFQ_CANCELING, | |
607 | WORK_STRUCT_PENDING); | |
608 | } | |
609 | ||
610 | static bool work_is_canceling(struct work_struct *work) | |
611 | { | |
612 | unsigned long data = atomic_long_read(&work->data); | |
613 | ||
614 | return !(data & WORK_STRUCT_CWQ) && (data & WORK_OFFQ_CANCELING); | |
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 | |
620 | * they're being called with gcwq->lock held. | |
e22bee78 TH |
621 | */ |
622 | ||
63d95a91 | 623 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 624 | { |
3270476a | 625 | return !atomic_read(get_pool_nr_running(pool)); |
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 | |
633 | * function will always return %true for unbound gcwq as long as the | |
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 | { |
63d95a91 | 650 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 | 651 | |
3270476a | 652 | return !list_empty(&pool->worklist) && atomic_read(nr_running) <= 1; |
e22bee78 TH |
653 | } |
654 | ||
655 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 656 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 657 | { |
63d95a91 | 658 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 659 | } |
365970a1 | 660 | |
e22bee78 | 661 | /* Do I need to be the manager? */ |
63d95a91 | 662 | static bool need_to_manage_workers(struct worker_pool *pool) |
e22bee78 | 663 | { |
63d95a91 | 664 | return need_to_create_worker(pool) || |
11ebea50 | 665 | (pool->flags & POOL_MANAGE_WORKERS); |
e22bee78 TH |
666 | } |
667 | ||
668 | /* Do we have too many workers and should some go away? */ | |
63d95a91 | 669 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 670 | { |
552a37e9 | 671 | bool managing = pool->flags & POOL_MANAGING_WORKERS; |
63d95a91 TH |
672 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
673 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 | 674 | |
ea1abd61 LJ |
675 | /* |
676 | * nr_idle and idle_list may disagree if idle rebinding is in | |
677 | * progress. Never return %true if idle_list is empty. | |
678 | */ | |
679 | if (list_empty(&pool->idle_list)) | |
680 | return false; | |
681 | ||
e22bee78 | 682 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
365970a1 DH |
683 | } |
684 | ||
4d707b9f | 685 | /* |
e22bee78 TH |
686 | * Wake up functions. |
687 | */ | |
688 | ||
7e11629d | 689 | /* Return the first worker. Safe with preemption disabled */ |
63d95a91 | 690 | static struct worker *first_worker(struct worker_pool *pool) |
7e11629d | 691 | { |
63d95a91 | 692 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
693 | return NULL; |
694 | ||
63d95a91 | 695 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
696 | } |
697 | ||
698 | /** | |
699 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 700 | * @pool: worker pool to wake worker from |
7e11629d | 701 | * |
63d95a91 | 702 | * Wake up the first idle worker of @pool. |
7e11629d TH |
703 | * |
704 | * CONTEXT: | |
705 | * spin_lock_irq(gcwq->lock). | |
706 | */ | |
63d95a91 | 707 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 708 | { |
63d95a91 | 709 | struct worker *worker = first_worker(pool); |
7e11629d TH |
710 | |
711 | if (likely(worker)) | |
712 | wake_up_process(worker->task); | |
713 | } | |
714 | ||
d302f017 | 715 | /** |
e22bee78 TH |
716 | * wq_worker_waking_up - a worker is waking up |
717 | * @task: task waking up | |
718 | * @cpu: CPU @task is waking up to | |
719 | * | |
720 | * This function is called during try_to_wake_up() when a worker is | |
721 | * being awoken. | |
722 | * | |
723 | * CONTEXT: | |
724 | * spin_lock_irq(rq->lock) | |
725 | */ | |
726 | void wq_worker_waking_up(struct task_struct *task, unsigned int cpu) | |
727 | { | |
728 | struct worker *worker = kthread_data(task); | |
729 | ||
36576000 JK |
730 | if (!(worker->flags & WORKER_NOT_RUNNING)) { |
731 | WARN_ON_ONCE(worker->pool->gcwq->cpu != cpu); | |
63d95a91 | 732 | atomic_inc(get_pool_nr_running(worker->pool)); |
36576000 | 733 | } |
e22bee78 TH |
734 | } |
735 | ||
736 | /** | |
737 | * wq_worker_sleeping - a worker is going to sleep | |
738 | * @task: task going to sleep | |
739 | * @cpu: CPU in question, must be the current CPU number | |
740 | * | |
741 | * This function is called during schedule() when a busy worker is | |
742 | * going to sleep. Worker on the same cpu can be woken up by | |
743 | * returning pointer to its task. | |
744 | * | |
745 | * CONTEXT: | |
746 | * spin_lock_irq(rq->lock) | |
747 | * | |
748 | * RETURNS: | |
749 | * Worker task on @cpu to wake up, %NULL if none. | |
750 | */ | |
751 | struct task_struct *wq_worker_sleeping(struct task_struct *task, | |
752 | unsigned int cpu) | |
753 | { | |
754 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
111c225a TH |
755 | struct worker_pool *pool; |
756 | atomic_t *nr_running; | |
e22bee78 | 757 | |
111c225a TH |
758 | /* |
759 | * Rescuers, which may not have all the fields set up like normal | |
760 | * workers, also reach here, let's not access anything before | |
761 | * checking NOT_RUNNING. | |
762 | */ | |
2d64672e | 763 | if (worker->flags & WORKER_NOT_RUNNING) |
e22bee78 TH |
764 | return NULL; |
765 | ||
111c225a TH |
766 | pool = worker->pool; |
767 | nr_running = get_pool_nr_running(pool); | |
768 | ||
e22bee78 TH |
769 | /* this can only happen on the local cpu */ |
770 | BUG_ON(cpu != raw_smp_processor_id()); | |
771 | ||
772 | /* | |
773 | * The counterpart of the following dec_and_test, implied mb, | |
774 | * worklist not empty test sequence is in insert_work(). | |
775 | * Please read comment there. | |
776 | * | |
628c78e7 TH |
777 | * NOT_RUNNING is clear. This means that we're bound to and |
778 | * running on the local cpu w/ rq lock held and preemption | |
779 | * disabled, which in turn means that none else could be | |
780 | * manipulating idle_list, so dereferencing idle_list without gcwq | |
781 | * lock is safe. | |
e22bee78 | 782 | */ |
bd7bdd43 | 783 | if (atomic_dec_and_test(nr_running) && !list_empty(&pool->worklist)) |
63d95a91 | 784 | to_wakeup = first_worker(pool); |
e22bee78 TH |
785 | return to_wakeup ? to_wakeup->task : NULL; |
786 | } | |
787 | ||
788 | /** | |
789 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 790 | * @worker: self |
d302f017 TH |
791 | * @flags: flags to set |
792 | * @wakeup: wakeup an idle worker if necessary | |
793 | * | |
e22bee78 TH |
794 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
795 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
796 | * woken up. | |
d302f017 | 797 | * |
cb444766 TH |
798 | * CONTEXT: |
799 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
800 | */ |
801 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
802 | bool wakeup) | |
803 | { | |
bd7bdd43 | 804 | struct worker_pool *pool = worker->pool; |
e22bee78 | 805 | |
cb444766 TH |
806 | WARN_ON_ONCE(worker->task != current); |
807 | ||
e22bee78 TH |
808 | /* |
809 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
810 | * wake up an idle worker as necessary if requested by | |
811 | * @wakeup. | |
812 | */ | |
813 | if ((flags & WORKER_NOT_RUNNING) && | |
814 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
63d95a91 | 815 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 TH |
816 | |
817 | if (wakeup) { | |
818 | if (atomic_dec_and_test(nr_running) && | |
bd7bdd43 | 819 | !list_empty(&pool->worklist)) |
63d95a91 | 820 | wake_up_worker(pool); |
e22bee78 TH |
821 | } else |
822 | atomic_dec(nr_running); | |
823 | } | |
824 | ||
d302f017 TH |
825 | worker->flags |= flags; |
826 | } | |
827 | ||
828 | /** | |
e22bee78 | 829 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 830 | * @worker: self |
d302f017 TH |
831 | * @flags: flags to clear |
832 | * | |
e22bee78 | 833 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 834 | * |
cb444766 TH |
835 | * CONTEXT: |
836 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
837 | */ |
838 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
839 | { | |
63d95a91 | 840 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
841 | unsigned int oflags = worker->flags; |
842 | ||
cb444766 TH |
843 | WARN_ON_ONCE(worker->task != current); |
844 | ||
d302f017 | 845 | worker->flags &= ~flags; |
e22bee78 | 846 | |
42c025f3 TH |
847 | /* |
848 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
849 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
850 | * of multiple flags, not a single flag. | |
851 | */ | |
e22bee78 TH |
852 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
853 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
63d95a91 | 854 | atomic_inc(get_pool_nr_running(pool)); |
d302f017 TH |
855 | } |
856 | ||
8cca0eea TH |
857 | /** |
858 | * find_worker_executing_work - find worker which is executing a work | |
859 | * @gcwq: gcwq of interest | |
860 | * @work: work to find worker for | |
861 | * | |
a2c1c57b TH |
862 | * Find a worker which is executing @work on @gcwq by searching |
863 | * @gcwq->busy_hash which is keyed by the address of @work. For a worker | |
864 | * to match, its current execution should match the address of @work and | |
865 | * its work function. This is to avoid unwanted dependency between | |
866 | * unrelated work executions through a work item being recycled while still | |
867 | * being executed. | |
868 | * | |
869 | * This is a bit tricky. A work item may be freed once its execution | |
870 | * starts and nothing prevents the freed area from being recycled for | |
871 | * another work item. If the same work item address ends up being reused | |
872 | * before the original execution finishes, workqueue will identify the | |
873 | * recycled work item as currently executing and make it wait until the | |
874 | * current execution finishes, introducing an unwanted dependency. | |
875 | * | |
876 | * This function checks the work item address, work function and workqueue | |
877 | * to avoid false positives. Note that this isn't complete as one may | |
878 | * construct a work function which can introduce dependency onto itself | |
879 | * through a recycled work item. Well, if somebody wants to shoot oneself | |
880 | * in the foot that badly, there's only so much we can do, and if such | |
881 | * deadlock actually occurs, it should be easy to locate the culprit work | |
882 | * function. | |
8cca0eea TH |
883 | * |
884 | * CONTEXT: | |
885 | * spin_lock_irq(gcwq->lock). | |
886 | * | |
887 | * RETURNS: | |
888 | * Pointer to worker which is executing @work if found, NULL | |
889 | * otherwise. | |
4d707b9f | 890 | */ |
8cca0eea TH |
891 | static struct worker *find_worker_executing_work(struct global_cwq *gcwq, |
892 | struct work_struct *work) | |
4d707b9f | 893 | { |
42f8570f SL |
894 | struct worker *worker; |
895 | struct hlist_node *tmp; | |
896 | ||
a2c1c57b TH |
897 | hash_for_each_possible(gcwq->busy_hash, worker, tmp, hentry, |
898 | (unsigned long)work) | |
899 | if (worker->current_work == work && | |
900 | worker->current_func == work->func) | |
42f8570f SL |
901 | return worker; |
902 | ||
903 | return NULL; | |
4d707b9f ON |
904 | } |
905 | ||
bf4ede01 TH |
906 | /** |
907 | * move_linked_works - move linked works to a list | |
908 | * @work: start of series of works to be scheduled | |
909 | * @head: target list to append @work to | |
910 | * @nextp: out paramter for nested worklist walking | |
911 | * | |
912 | * Schedule linked works starting from @work to @head. Work series to | |
913 | * be scheduled starts at @work and includes any consecutive work with | |
914 | * WORK_STRUCT_LINKED set in its predecessor. | |
915 | * | |
916 | * If @nextp is not NULL, it's updated to point to the next work of | |
917 | * the last scheduled work. This allows move_linked_works() to be | |
918 | * nested inside outer list_for_each_entry_safe(). | |
919 | * | |
920 | * CONTEXT: | |
921 | * spin_lock_irq(gcwq->lock). | |
922 | */ | |
923 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
924 | struct work_struct **nextp) | |
925 | { | |
926 | struct work_struct *n; | |
927 | ||
928 | /* | |
929 | * Linked worklist will always end before the end of the list, | |
930 | * use NULL for list head. | |
931 | */ | |
932 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
933 | list_move_tail(&work->entry, head); | |
934 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
935 | break; | |
936 | } | |
937 | ||
938 | /* | |
939 | * If we're already inside safe list traversal and have moved | |
940 | * multiple works to the scheduled queue, the next position | |
941 | * needs to be updated. | |
942 | */ | |
943 | if (nextp) | |
944 | *nextp = n; | |
945 | } | |
946 | ||
3aa62497 | 947 | static void cwq_activate_delayed_work(struct work_struct *work) |
bf4ede01 | 948 | { |
3aa62497 | 949 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); |
bf4ede01 TH |
950 | |
951 | trace_workqueue_activate_work(work); | |
952 | move_linked_works(work, &cwq->pool->worklist, NULL); | |
953 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); | |
954 | cwq->nr_active++; | |
955 | } | |
956 | ||
3aa62497 LJ |
957 | static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq) |
958 | { | |
959 | struct work_struct *work = list_first_entry(&cwq->delayed_works, | |
960 | struct work_struct, entry); | |
961 | ||
962 | cwq_activate_delayed_work(work); | |
963 | } | |
964 | ||
bf4ede01 TH |
965 | /** |
966 | * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight | |
967 | * @cwq: cwq of interest | |
968 | * @color: color of work which left the queue | |
bf4ede01 TH |
969 | * |
970 | * A work either has completed or is removed from pending queue, | |
971 | * decrement nr_in_flight of its cwq and handle workqueue flushing. | |
972 | * | |
973 | * CONTEXT: | |
974 | * spin_lock_irq(gcwq->lock). | |
975 | */ | |
b3f9f405 | 976 | static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color) |
bf4ede01 TH |
977 | { |
978 | /* ignore uncolored works */ | |
979 | if (color == WORK_NO_COLOR) | |
980 | return; | |
981 | ||
982 | cwq->nr_in_flight[color]--; | |
983 | ||
b3f9f405 LJ |
984 | cwq->nr_active--; |
985 | if (!list_empty(&cwq->delayed_works)) { | |
986 | /* one down, submit a delayed one */ | |
987 | if (cwq->nr_active < cwq->max_active) | |
988 | cwq_activate_first_delayed(cwq); | |
bf4ede01 TH |
989 | } |
990 | ||
991 | /* is flush in progress and are we at the flushing tip? */ | |
992 | if (likely(cwq->flush_color != color)) | |
993 | return; | |
994 | ||
995 | /* are there still in-flight works? */ | |
996 | if (cwq->nr_in_flight[color]) | |
997 | return; | |
998 | ||
999 | /* this cwq is done, clear flush_color */ | |
1000 | cwq->flush_color = -1; | |
1001 | ||
1002 | /* | |
1003 | * If this was the last cwq, wake up the first flusher. It | |
1004 | * will handle the rest. | |
1005 | */ | |
1006 | if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush)) | |
1007 | complete(&cwq->wq->first_flusher->done); | |
1008 | } | |
1009 | ||
36e227d2 | 1010 | /** |
bbb68dfa | 1011 | * try_to_grab_pending - steal work item from worklist and disable irq |
36e227d2 TH |
1012 | * @work: work item to steal |
1013 | * @is_dwork: @work is a delayed_work | |
bbb68dfa | 1014 | * @flags: place to store irq state |
36e227d2 TH |
1015 | * |
1016 | * Try to grab PENDING bit of @work. This function can handle @work in any | |
1017 | * stable state - idle, on timer or on worklist. Return values are | |
1018 | * | |
1019 | * 1 if @work was pending and we successfully stole PENDING | |
1020 | * 0 if @work was idle and we claimed PENDING | |
1021 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry | |
bbb68dfa TH |
1022 | * -ENOENT if someone else is canceling @work, this state may persist |
1023 | * for arbitrarily long | |
36e227d2 | 1024 | * |
bbb68dfa | 1025 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
e0aecdd8 TH |
1026 | * interrupted while holding PENDING and @work off queue, irq must be |
1027 | * disabled on entry. This, combined with delayed_work->timer being | |
1028 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. | |
bbb68dfa TH |
1029 | * |
1030 | * On successful return, >= 0, irq is disabled and the caller is | |
1031 | * responsible for releasing it using local_irq_restore(*@flags). | |
1032 | * | |
e0aecdd8 | 1033 | * This function is safe to call from any context including IRQ handler. |
bf4ede01 | 1034 | */ |
bbb68dfa TH |
1035 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
1036 | unsigned long *flags) | |
bf4ede01 TH |
1037 | { |
1038 | struct global_cwq *gcwq; | |
bf4ede01 | 1039 | |
bbb68dfa TH |
1040 | local_irq_save(*flags); |
1041 | ||
36e227d2 TH |
1042 | /* try to steal the timer if it exists */ |
1043 | if (is_dwork) { | |
1044 | struct delayed_work *dwork = to_delayed_work(work); | |
1045 | ||
e0aecdd8 TH |
1046 | /* |
1047 | * dwork->timer is irqsafe. If del_timer() fails, it's | |
1048 | * guaranteed that the timer is not queued anywhere and not | |
1049 | * running on the local CPU. | |
1050 | */ | |
36e227d2 TH |
1051 | if (likely(del_timer(&dwork->timer))) |
1052 | return 1; | |
1053 | } | |
1054 | ||
1055 | /* try to claim PENDING the normal way */ | |
bf4ede01 TH |
1056 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1057 | return 0; | |
1058 | ||
1059 | /* | |
1060 | * The queueing is in progress, or it is already queued. Try to | |
1061 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1062 | */ | |
1063 | gcwq = get_work_gcwq(work); | |
1064 | if (!gcwq) | |
bbb68dfa | 1065 | goto fail; |
bf4ede01 | 1066 | |
bbb68dfa | 1067 | spin_lock(&gcwq->lock); |
bf4ede01 TH |
1068 | if (!list_empty(&work->entry)) { |
1069 | /* | |
1070 | * This work is queued, but perhaps we locked the wrong gcwq. | |
1071 | * In that case we must see the new value after rmb(), see | |
1072 | * insert_work()->wmb(). | |
1073 | */ | |
1074 | smp_rmb(); | |
1075 | if (gcwq == get_work_gcwq(work)) { | |
1076 | debug_work_deactivate(work); | |
3aa62497 LJ |
1077 | |
1078 | /* | |
1079 | * A delayed work item cannot be grabbed directly | |
1080 | * because it might have linked NO_COLOR work items | |
1081 | * which, if left on the delayed_list, will confuse | |
1082 | * cwq->nr_active management later on and cause | |
1083 | * stall. Make sure the work item is activated | |
1084 | * before grabbing. | |
1085 | */ | |
1086 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) | |
1087 | cwq_activate_delayed_work(work); | |
1088 | ||
bf4ede01 TH |
1089 | list_del_init(&work->entry); |
1090 | cwq_dec_nr_in_flight(get_work_cwq(work), | |
b3f9f405 | 1091 | get_work_color(work)); |
36e227d2 | 1092 | |
bbb68dfa | 1093 | spin_unlock(&gcwq->lock); |
36e227d2 | 1094 | return 1; |
bf4ede01 TH |
1095 | } |
1096 | } | |
bbb68dfa TH |
1097 | spin_unlock(&gcwq->lock); |
1098 | fail: | |
1099 | local_irq_restore(*flags); | |
1100 | if (work_is_canceling(work)) | |
1101 | return -ENOENT; | |
1102 | cpu_relax(); | |
36e227d2 | 1103 | return -EAGAIN; |
bf4ede01 TH |
1104 | } |
1105 | ||
4690c4ab | 1106 | /** |
7e11629d | 1107 | * insert_work - insert a work into gcwq |
4690c4ab TH |
1108 | * @cwq: cwq @work belongs to |
1109 | * @work: work to insert | |
1110 | * @head: insertion point | |
1111 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
1112 | * | |
7e11629d TH |
1113 | * Insert @work which belongs to @cwq into @gcwq after @head. |
1114 | * @extra_flags is or'd to work_struct flags. | |
4690c4ab TH |
1115 | * |
1116 | * CONTEXT: | |
8b03ae3c | 1117 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 1118 | */ |
b89deed3 | 1119 | static void insert_work(struct cpu_workqueue_struct *cwq, |
4690c4ab TH |
1120 | struct work_struct *work, struct list_head *head, |
1121 | unsigned int extra_flags) | |
b89deed3 | 1122 | { |
63d95a91 | 1123 | struct worker_pool *pool = cwq->pool; |
e22bee78 | 1124 | |
4690c4ab | 1125 | /* we own @work, set data and link */ |
7a22ad75 | 1126 | set_work_cwq(work, cwq, extra_flags); |
e1d8aa9f | 1127 | |
6e84d644 ON |
1128 | /* |
1129 | * Ensure that we get the right work->data if we see the | |
1130 | * result of list_add() below, see try_to_grab_pending(). | |
1131 | */ | |
1132 | smp_wmb(); | |
4690c4ab | 1133 | |
1a4d9b0a | 1134 | list_add_tail(&work->entry, head); |
e22bee78 TH |
1135 | |
1136 | /* | |
1137 | * Ensure either worker_sched_deactivated() sees the above | |
1138 | * list_add_tail() or we see zero nr_running to avoid workers | |
1139 | * lying around lazily while there are works to be processed. | |
1140 | */ | |
1141 | smp_mb(); | |
1142 | ||
63d95a91 TH |
1143 | if (__need_more_worker(pool)) |
1144 | wake_up_worker(pool); | |
b89deed3 ON |
1145 | } |
1146 | ||
c8efcc25 TH |
1147 | /* |
1148 | * Test whether @work is being queued from another work executing on the | |
1149 | * same workqueue. This is rather expensive and should only be used from | |
1150 | * cold paths. | |
1151 | */ | |
1152 | static bool is_chained_work(struct workqueue_struct *wq) | |
1153 | { | |
1154 | unsigned long flags; | |
1155 | unsigned int cpu; | |
1156 | ||
1157 | for_each_gcwq_cpu(cpu) { | |
1158 | struct global_cwq *gcwq = get_gcwq(cpu); | |
1159 | struct worker *worker; | |
1160 | struct hlist_node *pos; | |
1161 | int i; | |
1162 | ||
1163 | spin_lock_irqsave(&gcwq->lock, flags); | |
1164 | for_each_busy_worker(worker, i, pos, gcwq) { | |
1165 | if (worker->task != current) | |
1166 | continue; | |
1167 | spin_unlock_irqrestore(&gcwq->lock, flags); | |
1168 | /* | |
1169 | * I'm @worker, no locking necessary. See if @work | |
1170 | * is headed to the same workqueue. | |
1171 | */ | |
1172 | return worker->current_cwq->wq == wq; | |
1173 | } | |
1174 | spin_unlock_irqrestore(&gcwq->lock, flags); | |
1175 | } | |
1176 | return false; | |
1177 | } | |
1178 | ||
4690c4ab | 1179 | static void __queue_work(unsigned int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
1180 | struct work_struct *work) |
1181 | { | |
502ca9d8 TH |
1182 | struct global_cwq *gcwq; |
1183 | struct cpu_workqueue_struct *cwq; | |
1e19ffc6 | 1184 | struct list_head *worklist; |
8a2e8e5d | 1185 | unsigned int work_flags; |
b75cac93 | 1186 | unsigned int req_cpu = cpu; |
8930caba TH |
1187 | |
1188 | /* | |
1189 | * While a work item is PENDING && off queue, a task trying to | |
1190 | * steal the PENDING will busy-loop waiting for it to either get | |
1191 | * queued or lose PENDING. Grabbing PENDING and queueing should | |
1192 | * happen with IRQ disabled. | |
1193 | */ | |
1194 | WARN_ON_ONCE(!irqs_disabled()); | |
1da177e4 | 1195 | |
dc186ad7 | 1196 | debug_work_activate(work); |
1e19ffc6 | 1197 | |
c8efcc25 | 1198 | /* if dying, only works from the same workqueue are allowed */ |
9c5a2ba7 | 1199 | if (unlikely(wq->flags & WQ_DRAINING) && |
c8efcc25 | 1200 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b TH |
1201 | return; |
1202 | ||
c7fc77f7 TH |
1203 | /* determine gcwq to use */ |
1204 | if (!(wq->flags & WQ_UNBOUND)) { | |
18aa9eff TH |
1205 | struct global_cwq *last_gcwq; |
1206 | ||
57469821 | 1207 | if (cpu == WORK_CPU_UNBOUND) |
c7fc77f7 TH |
1208 | cpu = raw_smp_processor_id(); |
1209 | ||
18aa9eff | 1210 | /* |
dbf2576e TH |
1211 | * It's multi cpu. If @work was previously on a different |
1212 | * cpu, it might still be running there, in which case the | |
1213 | * work needs to be queued on that cpu to guarantee | |
1214 | * non-reentrancy. | |
18aa9eff | 1215 | */ |
502ca9d8 | 1216 | gcwq = get_gcwq(cpu); |
dbf2576e TH |
1217 | last_gcwq = get_work_gcwq(work); |
1218 | ||
1219 | if (last_gcwq && last_gcwq != gcwq) { | |
18aa9eff TH |
1220 | struct worker *worker; |
1221 | ||
8930caba | 1222 | spin_lock(&last_gcwq->lock); |
18aa9eff TH |
1223 | |
1224 | worker = find_worker_executing_work(last_gcwq, work); | |
1225 | ||
1226 | if (worker && worker->current_cwq->wq == wq) | |
1227 | gcwq = last_gcwq; | |
1228 | else { | |
1229 | /* meh... not running there, queue here */ | |
8930caba TH |
1230 | spin_unlock(&last_gcwq->lock); |
1231 | spin_lock(&gcwq->lock); | |
18aa9eff | 1232 | } |
8930caba TH |
1233 | } else { |
1234 | spin_lock(&gcwq->lock); | |
1235 | } | |
f3421797 TH |
1236 | } else { |
1237 | gcwq = get_gcwq(WORK_CPU_UNBOUND); | |
8930caba | 1238 | spin_lock(&gcwq->lock); |
502ca9d8 TH |
1239 | } |
1240 | ||
1241 | /* gcwq determined, get cwq and queue */ | |
1242 | cwq = get_cwq(gcwq->cpu, wq); | |
b75cac93 | 1243 | trace_workqueue_queue_work(req_cpu, cwq, work); |
502ca9d8 | 1244 | |
f5b2552b | 1245 | if (WARN_ON(!list_empty(&work->entry))) { |
8930caba | 1246 | spin_unlock(&gcwq->lock); |
f5b2552b DC |
1247 | return; |
1248 | } | |
1e19ffc6 | 1249 | |
73f53c4a | 1250 | cwq->nr_in_flight[cwq->work_color]++; |
8a2e8e5d | 1251 | work_flags = work_color_to_flags(cwq->work_color); |
1e19ffc6 TH |
1252 | |
1253 | if (likely(cwq->nr_active < cwq->max_active)) { | |
cdadf009 | 1254 | trace_workqueue_activate_work(work); |
1e19ffc6 | 1255 | cwq->nr_active++; |
3270476a | 1256 | worklist = &cwq->pool->worklist; |
8a2e8e5d TH |
1257 | } else { |
1258 | work_flags |= WORK_STRUCT_DELAYED; | |
1e19ffc6 | 1259 | worklist = &cwq->delayed_works; |
8a2e8e5d | 1260 | } |
1e19ffc6 | 1261 | |
8a2e8e5d | 1262 | insert_work(cwq, work, worklist, work_flags); |
1e19ffc6 | 1263 | |
8930caba | 1264 | spin_unlock(&gcwq->lock); |
1da177e4 LT |
1265 | } |
1266 | ||
0fcb78c2 | 1267 | /** |
c1a220e7 ZR |
1268 | * queue_work_on - queue work on specific cpu |
1269 | * @cpu: CPU number to execute work on | |
0fcb78c2 REB |
1270 | * @wq: workqueue to use |
1271 | * @work: work to queue | |
1272 | * | |
d4283e93 | 1273 | * Returns %false if @work was already on a queue, %true otherwise. |
1da177e4 | 1274 | * |
c1a220e7 ZR |
1275 | * We queue the work to a specific CPU, the caller must ensure it |
1276 | * can't go away. | |
1da177e4 | 1277 | */ |
d4283e93 TH |
1278 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
1279 | struct work_struct *work) | |
1da177e4 | 1280 | { |
d4283e93 | 1281 | bool ret = false; |
8930caba | 1282 | unsigned long flags; |
ef1ca236 | 1283 | |
8930caba | 1284 | local_irq_save(flags); |
c1a220e7 | 1285 | |
22df02bb | 1286 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1287 | __queue_work(cpu, wq, work); |
d4283e93 | 1288 | ret = true; |
c1a220e7 | 1289 | } |
ef1ca236 | 1290 | |
8930caba | 1291 | local_irq_restore(flags); |
1da177e4 LT |
1292 | return ret; |
1293 | } | |
c1a220e7 | 1294 | EXPORT_SYMBOL_GPL(queue_work_on); |
1da177e4 | 1295 | |
c1a220e7 | 1296 | /** |
0a13c00e | 1297 | * queue_work - queue work on a workqueue |
c1a220e7 ZR |
1298 | * @wq: workqueue to use |
1299 | * @work: work to queue | |
1300 | * | |
d4283e93 | 1301 | * Returns %false if @work was already on a queue, %true otherwise. |
c1a220e7 | 1302 | * |
0a13c00e TH |
1303 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
1304 | * it can be processed by another CPU. | |
c1a220e7 | 1305 | */ |
d4283e93 | 1306 | bool queue_work(struct workqueue_struct *wq, struct work_struct *work) |
c1a220e7 | 1307 | { |
57469821 | 1308 | return queue_work_on(WORK_CPU_UNBOUND, wq, work); |
c1a220e7 | 1309 | } |
0a13c00e | 1310 | EXPORT_SYMBOL_GPL(queue_work); |
c1a220e7 | 1311 | |
d8e794df | 1312 | void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1313 | { |
52bad64d | 1314 | struct delayed_work *dwork = (struct delayed_work *)__data; |
7a22ad75 | 1315 | struct cpu_workqueue_struct *cwq = get_work_cwq(&dwork->work); |
1da177e4 | 1316 | |
e0aecdd8 | 1317 | /* should have been called from irqsafe timer with irq already off */ |
1265057f | 1318 | __queue_work(dwork->cpu, cwq->wq, &dwork->work); |
1da177e4 | 1319 | } |
d8e794df | 1320 | EXPORT_SYMBOL_GPL(delayed_work_timer_fn); |
1da177e4 | 1321 | |
7beb2edf TH |
1322 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
1323 | struct delayed_work *dwork, unsigned long delay) | |
1da177e4 | 1324 | { |
7beb2edf TH |
1325 | struct timer_list *timer = &dwork->timer; |
1326 | struct work_struct *work = &dwork->work; | |
1327 | unsigned int lcpu; | |
1328 | ||
1329 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || | |
1330 | timer->data != (unsigned long)dwork); | |
fc4b514f TH |
1331 | WARN_ON_ONCE(timer_pending(timer)); |
1332 | WARN_ON_ONCE(!list_empty(&work->entry)); | |
7beb2edf | 1333 | |
8852aac2 TH |
1334 | /* |
1335 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1336 | * both optimization and correctness. The earliest @timer can | |
1337 | * expire is on the closest next tick and delayed_work users depend | |
1338 | * on that there's no such delay when @delay is 0. | |
1339 | */ | |
1340 | if (!delay) { | |
1341 | __queue_work(cpu, wq, &dwork->work); | |
1342 | return; | |
1343 | } | |
1344 | ||
7beb2edf | 1345 | timer_stats_timer_set_start_info(&dwork->timer); |
1da177e4 | 1346 | |
7beb2edf TH |
1347 | /* |
1348 | * This stores cwq for the moment, for the timer_fn. Note that the | |
1349 | * work's gcwq is preserved to allow reentrance detection for | |
1350 | * delayed works. | |
1351 | */ | |
1352 | if (!(wq->flags & WQ_UNBOUND)) { | |
1353 | struct global_cwq *gcwq = get_work_gcwq(work); | |
1354 | ||
e42986de JK |
1355 | /* |
1356 | * If we cannot get the last gcwq from @work directly, | |
1357 | * select the last CPU such that it avoids unnecessarily | |
1358 | * triggering non-reentrancy check in __queue_work(). | |
1359 | */ | |
1360 | lcpu = cpu; | |
1361 | if (gcwq) | |
7beb2edf | 1362 | lcpu = gcwq->cpu; |
e42986de | 1363 | if (lcpu == WORK_CPU_UNBOUND) |
7beb2edf TH |
1364 | lcpu = raw_smp_processor_id(); |
1365 | } else { | |
1366 | lcpu = WORK_CPU_UNBOUND; | |
1367 | } | |
1368 | ||
1369 | set_work_cwq(work, get_cwq(lcpu, wq), 0); | |
1370 | ||
1265057f | 1371 | dwork->cpu = cpu; |
7beb2edf TH |
1372 | timer->expires = jiffies + delay; |
1373 | ||
1374 | if (unlikely(cpu != WORK_CPU_UNBOUND)) | |
1375 | add_timer_on(timer, cpu); | |
1376 | else | |
1377 | add_timer(timer); | |
1da177e4 LT |
1378 | } |
1379 | ||
0fcb78c2 REB |
1380 | /** |
1381 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1382 | * @cpu: CPU number to execute work on | |
1383 | * @wq: workqueue to use | |
af9997e4 | 1384 | * @dwork: work to queue |
0fcb78c2 REB |
1385 | * @delay: number of jiffies to wait before queueing |
1386 | * | |
715f1300 TH |
1387 | * Returns %false if @work was already on a queue, %true otherwise. If |
1388 | * @delay is zero and @dwork is idle, it will be scheduled for immediate | |
1389 | * execution. | |
0fcb78c2 | 1390 | */ |
d4283e93 TH |
1391 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
1392 | struct delayed_work *dwork, unsigned long delay) | |
7a6bc1cd | 1393 | { |
52bad64d | 1394 | struct work_struct *work = &dwork->work; |
d4283e93 | 1395 | bool ret = false; |
8930caba | 1396 | unsigned long flags; |
7a6bc1cd | 1397 | |
8930caba TH |
1398 | /* read the comment in __queue_work() */ |
1399 | local_irq_save(flags); | |
7a6bc1cd | 1400 | |
22df02bb | 1401 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7beb2edf | 1402 | __queue_delayed_work(cpu, wq, dwork, delay); |
d4283e93 | 1403 | ret = true; |
7a6bc1cd | 1404 | } |
8a3e77cc | 1405 | |
8930caba | 1406 | local_irq_restore(flags); |
7a6bc1cd VP |
1407 | return ret; |
1408 | } | |
ae90dd5d | 1409 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
c7fc77f7 | 1410 | |
0a13c00e TH |
1411 | /** |
1412 | * queue_delayed_work - queue work on a workqueue after delay | |
1413 | * @wq: workqueue to use | |
1414 | * @dwork: delayable work to queue | |
1415 | * @delay: number of jiffies to wait before queueing | |
1416 | * | |
715f1300 | 1417 | * Equivalent to queue_delayed_work_on() but tries to use the local CPU. |
0a13c00e | 1418 | */ |
d4283e93 | 1419 | bool queue_delayed_work(struct workqueue_struct *wq, |
0a13c00e TH |
1420 | struct delayed_work *dwork, unsigned long delay) |
1421 | { | |
57469821 | 1422 | return queue_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); |
0a13c00e TH |
1423 | } |
1424 | EXPORT_SYMBOL_GPL(queue_delayed_work); | |
c7fc77f7 | 1425 | |
8376fe22 TH |
1426 | /** |
1427 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU | |
1428 | * @cpu: CPU number to execute work on | |
1429 | * @wq: workqueue to use | |
1430 | * @dwork: work to queue | |
1431 | * @delay: number of jiffies to wait before queueing | |
1432 | * | |
1433 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, | |
1434 | * modify @dwork's timer so that it expires after @delay. If @delay is | |
1435 | * zero, @work is guaranteed to be scheduled immediately regardless of its | |
1436 | * current state. | |
1437 | * | |
1438 | * Returns %false if @dwork was idle and queued, %true if @dwork was | |
1439 | * pending and its timer was modified. | |
1440 | * | |
e0aecdd8 | 1441 | * This function is safe to call from any context including IRQ handler. |
8376fe22 TH |
1442 | * See try_to_grab_pending() for details. |
1443 | */ | |
1444 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
1445 | struct delayed_work *dwork, unsigned long delay) | |
1446 | { | |
1447 | unsigned long flags; | |
1448 | int ret; | |
c7fc77f7 | 1449 | |
8376fe22 TH |
1450 | do { |
1451 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
1452 | } while (unlikely(ret == -EAGAIN)); | |
63bc0362 | 1453 | |
8376fe22 TH |
1454 | if (likely(ret >= 0)) { |
1455 | __queue_delayed_work(cpu, wq, dwork, delay); | |
1456 | local_irq_restore(flags); | |
7a6bc1cd | 1457 | } |
8376fe22 TH |
1458 | |
1459 | /* -ENOENT from try_to_grab_pending() becomes %true */ | |
7a6bc1cd VP |
1460 | return ret; |
1461 | } | |
8376fe22 TH |
1462 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
1463 | ||
1464 | /** | |
1465 | * mod_delayed_work - modify delay of or queue a delayed work | |
1466 | * @wq: workqueue to use | |
1467 | * @dwork: work to queue | |
1468 | * @delay: number of jiffies to wait before queueing | |
1469 | * | |
1470 | * mod_delayed_work_on() on local CPU. | |
1471 | */ | |
1472 | bool mod_delayed_work(struct workqueue_struct *wq, struct delayed_work *dwork, | |
1473 | unsigned long delay) | |
1474 | { | |
1475 | return mod_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); | |
1476 | } | |
1477 | EXPORT_SYMBOL_GPL(mod_delayed_work); | |
1da177e4 | 1478 | |
c8e55f36 TH |
1479 | /** |
1480 | * worker_enter_idle - enter idle state | |
1481 | * @worker: worker which is entering idle state | |
1482 | * | |
1483 | * @worker is entering idle state. Update stats and idle timer if | |
1484 | * necessary. | |
1485 | * | |
1486 | * LOCKING: | |
1487 | * spin_lock_irq(gcwq->lock). | |
1488 | */ | |
1489 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1490 | { |
bd7bdd43 | 1491 | struct worker_pool *pool = worker->pool; |
c8e55f36 TH |
1492 | |
1493 | BUG_ON(worker->flags & WORKER_IDLE); | |
1494 | BUG_ON(!list_empty(&worker->entry) && | |
1495 | (worker->hentry.next || worker->hentry.pprev)); | |
1496 | ||
cb444766 TH |
1497 | /* can't use worker_set_flags(), also called from start_worker() */ |
1498 | worker->flags |= WORKER_IDLE; | |
bd7bdd43 | 1499 | pool->nr_idle++; |
e22bee78 | 1500 | worker->last_active = jiffies; |
c8e55f36 TH |
1501 | |
1502 | /* idle_list is LIFO */ | |
bd7bdd43 | 1503 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1504 | |
628c78e7 TH |
1505 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
1506 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); | |
cb444766 | 1507 | |
544ecf31 | 1508 | /* |
628c78e7 TH |
1509 | * Sanity check nr_running. Because gcwq_unbind_fn() releases |
1510 | * gcwq->lock between setting %WORKER_UNBOUND and zapping | |
1511 | * nr_running, the warning may trigger spuriously. Check iff | |
1512 | * unbind is not in progress. | |
544ecf31 | 1513 | */ |
24647570 | 1514 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
bd7bdd43 | 1515 | pool->nr_workers == pool->nr_idle && |
63d95a91 | 1516 | atomic_read(get_pool_nr_running(pool))); |
c8e55f36 TH |
1517 | } |
1518 | ||
1519 | /** | |
1520 | * worker_leave_idle - leave idle state | |
1521 | * @worker: worker which is leaving idle state | |
1522 | * | |
1523 | * @worker is leaving idle state. Update stats. | |
1524 | * | |
1525 | * LOCKING: | |
1526 | * spin_lock_irq(gcwq->lock). | |
1527 | */ | |
1528 | static void worker_leave_idle(struct worker *worker) | |
1529 | { | |
bd7bdd43 | 1530 | struct worker_pool *pool = worker->pool; |
c8e55f36 TH |
1531 | |
1532 | BUG_ON(!(worker->flags & WORKER_IDLE)); | |
d302f017 | 1533 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1534 | pool->nr_idle--; |
c8e55f36 TH |
1535 | list_del_init(&worker->entry); |
1536 | } | |
1537 | ||
e22bee78 TH |
1538 | /** |
1539 | * worker_maybe_bind_and_lock - bind worker to its cpu if possible and lock gcwq | |
1540 | * @worker: self | |
1541 | * | |
1542 | * Works which are scheduled while the cpu is online must at least be | |
1543 | * scheduled to a worker which is bound to the cpu so that if they are | |
1544 | * flushed from cpu callbacks while cpu is going down, they are | |
1545 | * guaranteed to execute on the cpu. | |
1546 | * | |
1547 | * This function is to be used by rogue workers and rescuers to bind | |
1548 | * themselves to the target cpu and may race with cpu going down or | |
1549 | * coming online. kthread_bind() can't be used because it may put the | |
1550 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
1551 | * verbatim as it's best effort and blocking and gcwq may be | |
1552 | * [dis]associated in the meantime. | |
1553 | * | |
f2d5a0ee | 1554 | * This function tries set_cpus_allowed() and locks gcwq and verifies the |
24647570 | 1555 | * binding against %POOL_DISASSOCIATED which is set during |
f2d5a0ee TH |
1556 | * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker |
1557 | * enters idle state or fetches works without dropping lock, it can | |
1558 | * guarantee the scheduling requirement described in the first paragraph. | |
e22bee78 TH |
1559 | * |
1560 | * CONTEXT: | |
1561 | * Might sleep. Called without any lock but returns with gcwq->lock | |
1562 | * held. | |
1563 | * | |
1564 | * RETURNS: | |
1565 | * %true if the associated gcwq is online (@worker is successfully | |
1566 | * bound), %false if offline. | |
1567 | */ | |
1568 | static bool worker_maybe_bind_and_lock(struct worker *worker) | |
972fa1c5 | 1569 | __acquires(&gcwq->lock) |
e22bee78 | 1570 | { |
24647570 TH |
1571 | struct worker_pool *pool = worker->pool; |
1572 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
1573 | struct task_struct *task = worker->task; |
1574 | ||
1575 | while (true) { | |
4e6045f1 | 1576 | /* |
e22bee78 TH |
1577 | * The following call may fail, succeed or succeed |
1578 | * without actually migrating the task to the cpu if | |
1579 | * it races with cpu hotunplug operation. Verify | |
24647570 | 1580 | * against POOL_DISASSOCIATED. |
4e6045f1 | 1581 | */ |
24647570 | 1582 | if (!(pool->flags & POOL_DISASSOCIATED)) |
f3421797 | 1583 | set_cpus_allowed_ptr(task, get_cpu_mask(gcwq->cpu)); |
e22bee78 TH |
1584 | |
1585 | spin_lock_irq(&gcwq->lock); | |
24647570 | 1586 | if (pool->flags & POOL_DISASSOCIATED) |
e22bee78 TH |
1587 | return false; |
1588 | if (task_cpu(task) == gcwq->cpu && | |
1589 | cpumask_equal(¤t->cpus_allowed, | |
1590 | get_cpu_mask(gcwq->cpu))) | |
1591 | return true; | |
1592 | spin_unlock_irq(&gcwq->lock); | |
1593 | ||
5035b20f TH |
1594 | /* |
1595 | * We've raced with CPU hot[un]plug. Give it a breather | |
1596 | * and retry migration. cond_resched() is required here; | |
1597 | * otherwise, we might deadlock against cpu_stop trying to | |
1598 | * bring down the CPU on non-preemptive kernel. | |
1599 | */ | |
e22bee78 | 1600 | cpu_relax(); |
5035b20f | 1601 | cond_resched(); |
e22bee78 TH |
1602 | } |
1603 | } | |
1604 | ||
25511a47 | 1605 | /* |
ea1abd61 | 1606 | * Rebind an idle @worker to its CPU. worker_thread() will test |
5f7dabfd | 1607 | * list_empty(@worker->entry) before leaving idle and call this function. |
25511a47 TH |
1608 | */ |
1609 | static void idle_worker_rebind(struct worker *worker) | |
1610 | { | |
1611 | struct global_cwq *gcwq = worker->pool->gcwq; | |
1612 | ||
5f7dabfd LJ |
1613 | /* CPU may go down again inbetween, clear UNBOUND only on success */ |
1614 | if (worker_maybe_bind_and_lock(worker)) | |
1615 | worker_clr_flags(worker, WORKER_UNBOUND); | |
25511a47 | 1616 | |
ea1abd61 LJ |
1617 | /* rebind complete, become available again */ |
1618 | list_add(&worker->entry, &worker->pool->idle_list); | |
1619 | spin_unlock_irq(&gcwq->lock); | |
25511a47 TH |
1620 | } |
1621 | ||
e22bee78 | 1622 | /* |
25511a47 | 1623 | * Function for @worker->rebind.work used to rebind unbound busy workers to |
403c821d TH |
1624 | * the associated cpu which is coming back online. This is scheduled by |
1625 | * cpu up but can race with other cpu hotplug operations and may be | |
1626 | * executed twice without intervening cpu down. | |
e22bee78 | 1627 | */ |
25511a47 | 1628 | static void busy_worker_rebind_fn(struct work_struct *work) |
e22bee78 TH |
1629 | { |
1630 | struct worker *worker = container_of(work, struct worker, rebind_work); | |
bd7bdd43 | 1631 | struct global_cwq *gcwq = worker->pool->gcwq; |
e22bee78 | 1632 | |
eab6d828 LJ |
1633 | if (worker_maybe_bind_and_lock(worker)) |
1634 | worker_clr_flags(worker, WORKER_UNBOUND); | |
e22bee78 TH |
1635 | |
1636 | spin_unlock_irq(&gcwq->lock); | |
1637 | } | |
1638 | ||
25511a47 TH |
1639 | /** |
1640 | * rebind_workers - rebind all workers of a gcwq to the associated CPU | |
1641 | * @gcwq: gcwq of interest | |
1642 | * | |
1643 | * @gcwq->cpu is coming online. Rebind all workers to the CPU. Rebinding | |
1644 | * is different for idle and busy ones. | |
1645 | * | |
ea1abd61 LJ |
1646 | * Idle ones will be removed from the idle_list and woken up. They will |
1647 | * add themselves back after completing rebind. This ensures that the | |
1648 | * idle_list doesn't contain any unbound workers when re-bound busy workers | |
1649 | * try to perform local wake-ups for concurrency management. | |
25511a47 | 1650 | * |
ea1abd61 LJ |
1651 | * Busy workers can rebind after they finish their current work items. |
1652 | * Queueing the rebind work item at the head of the scheduled list is | |
1653 | * enough. Note that nr_running will be properly bumped as busy workers | |
1654 | * rebind. | |
25511a47 | 1655 | * |
ea1abd61 LJ |
1656 | * On return, all non-manager workers are scheduled for rebind - see |
1657 | * manage_workers() for the manager special case. Any idle worker | |
1658 | * including the manager will not appear on @idle_list until rebind is | |
1659 | * complete, making local wake-ups safe. | |
25511a47 TH |
1660 | */ |
1661 | static void rebind_workers(struct global_cwq *gcwq) | |
25511a47 | 1662 | { |
25511a47 | 1663 | struct worker_pool *pool; |
ea1abd61 | 1664 | struct worker *worker, *n; |
25511a47 TH |
1665 | struct hlist_node *pos; |
1666 | int i; | |
1667 | ||
1668 | lockdep_assert_held(&gcwq->lock); | |
1669 | ||
1670 | for_each_worker_pool(pool, gcwq) | |
b2eb83d1 | 1671 | lockdep_assert_held(&pool->assoc_mutex); |
25511a47 | 1672 | |
5f7dabfd | 1673 | /* dequeue and kick idle ones */ |
25511a47 | 1674 | for_each_worker_pool(pool, gcwq) { |
ea1abd61 | 1675 | list_for_each_entry_safe(worker, n, &pool->idle_list, entry) { |
ea1abd61 LJ |
1676 | /* |
1677 | * idle workers should be off @pool->idle_list | |
1678 | * until rebind is complete to avoid receiving | |
1679 | * premature local wake-ups. | |
1680 | */ | |
1681 | list_del_init(&worker->entry); | |
25511a47 | 1682 | |
5f7dabfd LJ |
1683 | /* |
1684 | * worker_thread() will see the above dequeuing | |
1685 | * and call idle_worker_rebind(). | |
1686 | */ | |
25511a47 TH |
1687 | wake_up_process(worker->task); |
1688 | } | |
1689 | } | |
1690 | ||
ea1abd61 | 1691 | /* rebind busy workers */ |
25511a47 TH |
1692 | for_each_busy_worker(worker, i, pos, gcwq) { |
1693 | struct work_struct *rebind_work = &worker->rebind_work; | |
e2b6a6d5 | 1694 | struct workqueue_struct *wq; |
25511a47 TH |
1695 | |
1696 | if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, | |
1697 | work_data_bits(rebind_work))) | |
1698 | continue; | |
1699 | ||
25511a47 | 1700 | debug_work_activate(rebind_work); |
90beca5d | 1701 | |
e2b6a6d5 JK |
1702 | /* |
1703 | * wq doesn't really matter but let's keep @worker->pool | |
1704 | * and @cwq->pool consistent for sanity. | |
1705 | */ | |
e34cdddb | 1706 | if (std_worker_pool_pri(worker->pool)) |
e2b6a6d5 JK |
1707 | wq = system_highpri_wq; |
1708 | else | |
1709 | wq = system_wq; | |
ec58815a | 1710 | |
e2b6a6d5 JK |
1711 | insert_work(get_cwq(gcwq->cpu, wq), rebind_work, |
1712 | worker->scheduled.next, | |
1713 | work_color_to_flags(WORK_NO_COLOR)); | |
ec58815a | 1714 | } |
25511a47 TH |
1715 | } |
1716 | ||
c34056a3 TH |
1717 | static struct worker *alloc_worker(void) |
1718 | { | |
1719 | struct worker *worker; | |
1720 | ||
1721 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1722 | if (worker) { |
1723 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1724 | INIT_LIST_HEAD(&worker->scheduled); |
25511a47 | 1725 | INIT_WORK(&worker->rebind_work, busy_worker_rebind_fn); |
e22bee78 TH |
1726 | /* on creation a worker is in !idle && prep state */ |
1727 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1728 | } |
c34056a3 TH |
1729 | return worker; |
1730 | } | |
1731 | ||
1732 | /** | |
1733 | * create_worker - create a new workqueue worker | |
63d95a91 | 1734 | * @pool: pool the new worker will belong to |
c34056a3 | 1735 | * |
63d95a91 | 1736 | * Create a new worker which is bound to @pool. The returned worker |
c34056a3 TH |
1737 | * can be started by calling start_worker() or destroyed using |
1738 | * destroy_worker(). | |
1739 | * | |
1740 | * CONTEXT: | |
1741 | * Might sleep. Does GFP_KERNEL allocations. | |
1742 | * | |
1743 | * RETURNS: | |
1744 | * Pointer to the newly created worker. | |
1745 | */ | |
bc2ae0f5 | 1746 | static struct worker *create_worker(struct worker_pool *pool) |
c34056a3 | 1747 | { |
63d95a91 | 1748 | struct global_cwq *gcwq = pool->gcwq; |
e34cdddb | 1749 | const char *pri = std_worker_pool_pri(pool) ? "H" : ""; |
c34056a3 | 1750 | struct worker *worker = NULL; |
f3421797 | 1751 | int id = -1; |
c34056a3 | 1752 | |
8b03ae3c | 1753 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1754 | while (ida_get_new(&pool->worker_ida, &id)) { |
8b03ae3c | 1755 | spin_unlock_irq(&gcwq->lock); |
bd7bdd43 | 1756 | if (!ida_pre_get(&pool->worker_ida, GFP_KERNEL)) |
c34056a3 | 1757 | goto fail; |
8b03ae3c | 1758 | spin_lock_irq(&gcwq->lock); |
c34056a3 | 1759 | } |
8b03ae3c | 1760 | spin_unlock_irq(&gcwq->lock); |
c34056a3 TH |
1761 | |
1762 | worker = alloc_worker(); | |
1763 | if (!worker) | |
1764 | goto fail; | |
1765 | ||
bd7bdd43 | 1766 | worker->pool = pool; |
c34056a3 TH |
1767 | worker->id = id; |
1768 | ||
bc2ae0f5 | 1769 | if (gcwq->cpu != WORK_CPU_UNBOUND) |
94dcf29a | 1770 | worker->task = kthread_create_on_node(worker_thread, |
3270476a TH |
1771 | worker, cpu_to_node(gcwq->cpu), |
1772 | "kworker/%u:%d%s", gcwq->cpu, id, pri); | |
f3421797 TH |
1773 | else |
1774 | worker->task = kthread_create(worker_thread, worker, | |
3270476a | 1775 | "kworker/u:%d%s", id, pri); |
c34056a3 TH |
1776 | if (IS_ERR(worker->task)) |
1777 | goto fail; | |
1778 | ||
e34cdddb | 1779 | if (std_worker_pool_pri(pool)) |
3270476a TH |
1780 | set_user_nice(worker->task, HIGHPRI_NICE_LEVEL); |
1781 | ||
db7bccf4 | 1782 | /* |
bc2ae0f5 | 1783 | * Determine CPU binding of the new worker depending on |
24647570 | 1784 | * %POOL_DISASSOCIATED. The caller is responsible for ensuring the |
bc2ae0f5 TH |
1785 | * flag remains stable across this function. See the comments |
1786 | * above the flag definition for details. | |
1787 | * | |
1788 | * As an unbound worker may later become a regular one if CPU comes | |
1789 | * online, make sure every worker has %PF_THREAD_BOUND set. | |
db7bccf4 | 1790 | */ |
24647570 | 1791 | if (!(pool->flags & POOL_DISASSOCIATED)) { |
8b03ae3c | 1792 | kthread_bind(worker->task, gcwq->cpu); |
bc2ae0f5 | 1793 | } else { |
db7bccf4 | 1794 | worker->task->flags |= PF_THREAD_BOUND; |
bc2ae0f5 | 1795 | worker->flags |= WORKER_UNBOUND; |
f3421797 | 1796 | } |
c34056a3 TH |
1797 | |
1798 | return worker; | |
1799 | fail: | |
1800 | if (id >= 0) { | |
8b03ae3c | 1801 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1802 | ida_remove(&pool->worker_ida, id); |
8b03ae3c | 1803 | spin_unlock_irq(&gcwq->lock); |
c34056a3 TH |
1804 | } |
1805 | kfree(worker); | |
1806 | return NULL; | |
1807 | } | |
1808 | ||
1809 | /** | |
1810 | * start_worker - start a newly created worker | |
1811 | * @worker: worker to start | |
1812 | * | |
c8e55f36 | 1813 | * Make the gcwq aware of @worker and start it. |
c34056a3 TH |
1814 | * |
1815 | * CONTEXT: | |
8b03ae3c | 1816 | * spin_lock_irq(gcwq->lock). |
c34056a3 TH |
1817 | */ |
1818 | static void start_worker(struct worker *worker) | |
1819 | { | |
cb444766 | 1820 | worker->flags |= WORKER_STARTED; |
bd7bdd43 | 1821 | worker->pool->nr_workers++; |
c8e55f36 | 1822 | worker_enter_idle(worker); |
c34056a3 TH |
1823 | wake_up_process(worker->task); |
1824 | } | |
1825 | ||
1826 | /** | |
1827 | * destroy_worker - destroy a workqueue worker | |
1828 | * @worker: worker to be destroyed | |
1829 | * | |
c8e55f36 TH |
1830 | * Destroy @worker and adjust @gcwq stats accordingly. |
1831 | * | |
1832 | * CONTEXT: | |
1833 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. | |
c34056a3 TH |
1834 | */ |
1835 | static void destroy_worker(struct worker *worker) | |
1836 | { | |
bd7bdd43 TH |
1837 | struct worker_pool *pool = worker->pool; |
1838 | struct global_cwq *gcwq = pool->gcwq; | |
c34056a3 TH |
1839 | int id = worker->id; |
1840 | ||
1841 | /* sanity check frenzy */ | |
1842 | BUG_ON(worker->current_work); | |
affee4b2 | 1843 | BUG_ON(!list_empty(&worker->scheduled)); |
c34056a3 | 1844 | |
c8e55f36 | 1845 | if (worker->flags & WORKER_STARTED) |
bd7bdd43 | 1846 | pool->nr_workers--; |
c8e55f36 | 1847 | if (worker->flags & WORKER_IDLE) |
bd7bdd43 | 1848 | pool->nr_idle--; |
c8e55f36 TH |
1849 | |
1850 | list_del_init(&worker->entry); | |
cb444766 | 1851 | worker->flags |= WORKER_DIE; |
c8e55f36 TH |
1852 | |
1853 | spin_unlock_irq(&gcwq->lock); | |
1854 | ||
c34056a3 TH |
1855 | kthread_stop(worker->task); |
1856 | kfree(worker); | |
1857 | ||
8b03ae3c | 1858 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1859 | ida_remove(&pool->worker_ida, id); |
c34056a3 TH |
1860 | } |
1861 | ||
63d95a91 | 1862 | static void idle_worker_timeout(unsigned long __pool) |
e22bee78 | 1863 | { |
63d95a91 TH |
1864 | struct worker_pool *pool = (void *)__pool; |
1865 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
1866 | |
1867 | spin_lock_irq(&gcwq->lock); | |
1868 | ||
63d95a91 | 1869 | if (too_many_workers(pool)) { |
e22bee78 TH |
1870 | struct worker *worker; |
1871 | unsigned long expires; | |
1872 | ||
1873 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1874 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1875 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1876 | ||
1877 | if (time_before(jiffies, expires)) | |
63d95a91 | 1878 | mod_timer(&pool->idle_timer, expires); |
e22bee78 TH |
1879 | else { |
1880 | /* it's been idle for too long, wake up manager */ | |
11ebea50 | 1881 | pool->flags |= POOL_MANAGE_WORKERS; |
63d95a91 | 1882 | wake_up_worker(pool); |
d5abe669 | 1883 | } |
e22bee78 TH |
1884 | } |
1885 | ||
1886 | spin_unlock_irq(&gcwq->lock); | |
1887 | } | |
d5abe669 | 1888 | |
e22bee78 TH |
1889 | static bool send_mayday(struct work_struct *work) |
1890 | { | |
1891 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); | |
1892 | struct workqueue_struct *wq = cwq->wq; | |
f3421797 | 1893 | unsigned int cpu; |
e22bee78 TH |
1894 | |
1895 | if (!(wq->flags & WQ_RESCUER)) | |
1896 | return false; | |
1897 | ||
1898 | /* mayday mayday mayday */ | |
bd7bdd43 | 1899 | cpu = cwq->pool->gcwq->cpu; |
f3421797 TH |
1900 | /* WORK_CPU_UNBOUND can't be set in cpumask, use cpu 0 instead */ |
1901 | if (cpu == WORK_CPU_UNBOUND) | |
1902 | cpu = 0; | |
f2e005aa | 1903 | if (!mayday_test_and_set_cpu(cpu, wq->mayday_mask)) |
e22bee78 TH |
1904 | wake_up_process(wq->rescuer->task); |
1905 | return true; | |
1906 | } | |
1907 | ||
63d95a91 | 1908 | static void gcwq_mayday_timeout(unsigned long __pool) |
e22bee78 | 1909 | { |
63d95a91 TH |
1910 | struct worker_pool *pool = (void *)__pool; |
1911 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
1912 | struct work_struct *work; |
1913 | ||
1914 | spin_lock_irq(&gcwq->lock); | |
1915 | ||
63d95a91 | 1916 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
1917 | /* |
1918 | * We've been trying to create a new worker but | |
1919 | * haven't been successful. We might be hitting an | |
1920 | * allocation deadlock. Send distress signals to | |
1921 | * rescuers. | |
1922 | */ | |
63d95a91 | 1923 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 1924 | send_mayday(work); |
1da177e4 | 1925 | } |
e22bee78 TH |
1926 | |
1927 | spin_unlock_irq(&gcwq->lock); | |
1928 | ||
63d95a91 | 1929 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
1930 | } |
1931 | ||
e22bee78 TH |
1932 | /** |
1933 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 1934 | * @pool: pool to create a new worker for |
e22bee78 | 1935 | * |
63d95a91 | 1936 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
1937 | * have at least one idle worker on return from this function. If |
1938 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 1939 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
1940 | * possible allocation deadlock. |
1941 | * | |
1942 | * On return, need_to_create_worker() is guaranteed to be false and | |
1943 | * may_start_working() true. | |
1944 | * | |
1945 | * LOCKING: | |
1946 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1947 | * multiple times. Does GFP_KERNEL allocations. Called only from | |
1948 | * manager. | |
1949 | * | |
1950 | * RETURNS: | |
1951 | * false if no action was taken and gcwq->lock stayed locked, true | |
1952 | * otherwise. | |
1953 | */ | |
63d95a91 | 1954 | static bool maybe_create_worker(struct worker_pool *pool) |
06bd6ebf NK |
1955 | __releases(&gcwq->lock) |
1956 | __acquires(&gcwq->lock) | |
1da177e4 | 1957 | { |
63d95a91 TH |
1958 | struct global_cwq *gcwq = pool->gcwq; |
1959 | ||
1960 | if (!need_to_create_worker(pool)) | |
e22bee78 TH |
1961 | return false; |
1962 | restart: | |
9f9c2364 TH |
1963 | spin_unlock_irq(&gcwq->lock); |
1964 | ||
e22bee78 | 1965 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 1966 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
1967 | |
1968 | while (true) { | |
1969 | struct worker *worker; | |
1970 | ||
bc2ae0f5 | 1971 | worker = create_worker(pool); |
e22bee78 | 1972 | if (worker) { |
63d95a91 | 1973 | del_timer_sync(&pool->mayday_timer); |
e22bee78 TH |
1974 | spin_lock_irq(&gcwq->lock); |
1975 | start_worker(worker); | |
63d95a91 | 1976 | BUG_ON(need_to_create_worker(pool)); |
e22bee78 TH |
1977 | return true; |
1978 | } | |
1979 | ||
63d95a91 | 1980 | if (!need_to_create_worker(pool)) |
e22bee78 | 1981 | break; |
1da177e4 | 1982 | |
e22bee78 TH |
1983 | __set_current_state(TASK_INTERRUPTIBLE); |
1984 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1985 | |
63d95a91 | 1986 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1987 | break; |
1988 | } | |
1989 | ||
63d95a91 | 1990 | del_timer_sync(&pool->mayday_timer); |
e22bee78 | 1991 | spin_lock_irq(&gcwq->lock); |
63d95a91 | 1992 | if (need_to_create_worker(pool)) |
e22bee78 TH |
1993 | goto restart; |
1994 | return true; | |
1995 | } | |
1996 | ||
1997 | /** | |
1998 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
63d95a91 | 1999 | * @pool: pool to destroy workers for |
e22bee78 | 2000 | * |
63d95a91 | 2001 | * Destroy @pool workers which have been idle for longer than |
e22bee78 TH |
2002 | * IDLE_WORKER_TIMEOUT. |
2003 | * | |
2004 | * LOCKING: | |
2005 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
2006 | * multiple times. Called only from manager. | |
2007 | * | |
2008 | * RETURNS: | |
2009 | * false if no action was taken and gcwq->lock stayed locked, true | |
2010 | * otherwise. | |
2011 | */ | |
63d95a91 | 2012 | static bool maybe_destroy_workers(struct worker_pool *pool) |
e22bee78 TH |
2013 | { |
2014 | bool ret = false; | |
1da177e4 | 2015 | |
63d95a91 | 2016 | while (too_many_workers(pool)) { |
e22bee78 TH |
2017 | struct worker *worker; |
2018 | unsigned long expires; | |
3af24433 | 2019 | |
63d95a91 | 2020 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 | 2021 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
85f4186a | 2022 | |
e22bee78 | 2023 | if (time_before(jiffies, expires)) { |
63d95a91 | 2024 | mod_timer(&pool->idle_timer, expires); |
3af24433 | 2025 | break; |
e22bee78 | 2026 | } |
1da177e4 | 2027 | |
e22bee78 TH |
2028 | destroy_worker(worker); |
2029 | ret = true; | |
1da177e4 | 2030 | } |
1e19ffc6 | 2031 | |
e22bee78 | 2032 | return ret; |
1e19ffc6 TH |
2033 | } |
2034 | ||
73f53c4a | 2035 | /** |
e22bee78 TH |
2036 | * manage_workers - manage worker pool |
2037 | * @worker: self | |
73f53c4a | 2038 | * |
e22bee78 TH |
2039 | * Assume the manager role and manage gcwq worker pool @worker belongs |
2040 | * to. At any given time, there can be only zero or one manager per | |
2041 | * gcwq. The exclusion is handled automatically by this function. | |
2042 | * | |
2043 | * The caller can safely start processing works on false return. On | |
2044 | * true return, it's guaranteed that need_to_create_worker() is false | |
2045 | * and may_start_working() is true. | |
73f53c4a TH |
2046 | * |
2047 | * CONTEXT: | |
e22bee78 TH |
2048 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed |
2049 | * multiple times. Does GFP_KERNEL allocations. | |
2050 | * | |
2051 | * RETURNS: | |
2052 | * false if no action was taken and gcwq->lock stayed locked, true if | |
2053 | * some action was taken. | |
73f53c4a | 2054 | */ |
e22bee78 | 2055 | static bool manage_workers(struct worker *worker) |
73f53c4a | 2056 | { |
63d95a91 | 2057 | struct worker_pool *pool = worker->pool; |
e22bee78 | 2058 | bool ret = false; |
73f53c4a | 2059 | |
ee378aa4 | 2060 | if (pool->flags & POOL_MANAGING_WORKERS) |
e22bee78 | 2061 | return ret; |
1e19ffc6 | 2062 | |
552a37e9 | 2063 | pool->flags |= POOL_MANAGING_WORKERS; |
73f53c4a | 2064 | |
ee378aa4 LJ |
2065 | /* |
2066 | * To simplify both worker management and CPU hotplug, hold off | |
2067 | * management while hotplug is in progress. CPU hotplug path can't | |
2068 | * grab %POOL_MANAGING_WORKERS to achieve this because that can | |
2069 | * lead to idle worker depletion (all become busy thinking someone | |
2070 | * else is managing) which in turn can result in deadlock under | |
b2eb83d1 | 2071 | * extreme circumstances. Use @pool->assoc_mutex to synchronize |
ee378aa4 LJ |
2072 | * manager against CPU hotplug. |
2073 | * | |
b2eb83d1 | 2074 | * assoc_mutex would always be free unless CPU hotplug is in |
ee378aa4 LJ |
2075 | * progress. trylock first without dropping @gcwq->lock. |
2076 | */ | |
b2eb83d1 | 2077 | if (unlikely(!mutex_trylock(&pool->assoc_mutex))) { |
ee378aa4 | 2078 | spin_unlock_irq(&pool->gcwq->lock); |
b2eb83d1 | 2079 | mutex_lock(&pool->assoc_mutex); |
ee378aa4 LJ |
2080 | /* |
2081 | * CPU hotplug could have happened while we were waiting | |
b2eb83d1 | 2082 | * for assoc_mutex. Hotplug itself can't handle us |
ee378aa4 LJ |
2083 | * because manager isn't either on idle or busy list, and |
2084 | * @gcwq's state and ours could have deviated. | |
2085 | * | |
b2eb83d1 | 2086 | * As hotplug is now excluded via assoc_mutex, we can |
ee378aa4 LJ |
2087 | * simply try to bind. It will succeed or fail depending |
2088 | * on @gcwq's current state. Try it and adjust | |
2089 | * %WORKER_UNBOUND accordingly. | |
2090 | */ | |
2091 | if (worker_maybe_bind_and_lock(worker)) | |
2092 | worker->flags &= ~WORKER_UNBOUND; | |
2093 | else | |
2094 | worker->flags |= WORKER_UNBOUND; | |
73f53c4a | 2095 | |
ee378aa4 LJ |
2096 | ret = true; |
2097 | } | |
73f53c4a | 2098 | |
11ebea50 | 2099 | pool->flags &= ~POOL_MANAGE_WORKERS; |
73f53c4a TH |
2100 | |
2101 | /* | |
e22bee78 TH |
2102 | * Destroy and then create so that may_start_working() is true |
2103 | * on return. | |
73f53c4a | 2104 | */ |
63d95a91 TH |
2105 | ret |= maybe_destroy_workers(pool); |
2106 | ret |= maybe_create_worker(pool); | |
e22bee78 | 2107 | |
552a37e9 | 2108 | pool->flags &= ~POOL_MANAGING_WORKERS; |
b2eb83d1 | 2109 | mutex_unlock(&pool->assoc_mutex); |
e22bee78 | 2110 | return ret; |
73f53c4a TH |
2111 | } |
2112 | ||
a62428c0 TH |
2113 | /** |
2114 | * process_one_work - process single work | |
c34056a3 | 2115 | * @worker: self |
a62428c0 TH |
2116 | * @work: work to process |
2117 | * | |
2118 | * Process @work. This function contains all the logics necessary to | |
2119 | * process a single work including synchronization against and | |
2120 | * interaction with other workers on the same cpu, queueing and | |
2121 | * flushing. As long as context requirement is met, any worker can | |
2122 | * call this function to process a work. | |
2123 | * | |
2124 | * CONTEXT: | |
8b03ae3c | 2125 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. |
a62428c0 | 2126 | */ |
c34056a3 | 2127 | static void process_one_work(struct worker *worker, struct work_struct *work) |
06bd6ebf NK |
2128 | __releases(&gcwq->lock) |
2129 | __acquires(&gcwq->lock) | |
a62428c0 | 2130 | { |
7e11629d | 2131 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); |
bd7bdd43 TH |
2132 | struct worker_pool *pool = worker->pool; |
2133 | struct global_cwq *gcwq = pool->gcwq; | |
fb0e7beb | 2134 | bool cpu_intensive = cwq->wq->flags & WQ_CPU_INTENSIVE; |
73f53c4a | 2135 | int work_color; |
7e11629d | 2136 | struct worker *collision; |
a62428c0 TH |
2137 | #ifdef CONFIG_LOCKDEP |
2138 | /* | |
2139 | * It is permissible to free the struct work_struct from | |
2140 | * inside the function that is called from it, this we need to | |
2141 | * take into account for lockdep too. To avoid bogus "held | |
2142 | * lock freed" warnings as well as problems when looking into | |
2143 | * work->lockdep_map, make a copy and use that here. | |
2144 | */ | |
4d82a1de PZ |
2145 | struct lockdep_map lockdep_map; |
2146 | ||
2147 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 2148 | #endif |
6fec10a1 TH |
2149 | /* |
2150 | * Ensure we're on the correct CPU. DISASSOCIATED test is | |
2151 | * necessary to avoid spurious warnings from rescuers servicing the | |
24647570 | 2152 | * unbound or a disassociated pool. |
6fec10a1 | 2153 | */ |
5f7dabfd | 2154 | WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) && |
24647570 | 2155 | !(pool->flags & POOL_DISASSOCIATED) && |
25511a47 TH |
2156 | raw_smp_processor_id() != gcwq->cpu); |
2157 | ||
7e11629d TH |
2158 | /* |
2159 | * A single work shouldn't be executed concurrently by | |
2160 | * multiple workers on a single cpu. Check whether anyone is | |
2161 | * already processing the work. If so, defer the work to the | |
2162 | * currently executing one. | |
2163 | */ | |
42f8570f | 2164 | collision = find_worker_executing_work(gcwq, work); |
7e11629d TH |
2165 | if (unlikely(collision)) { |
2166 | move_linked_works(work, &collision->scheduled, NULL); | |
2167 | return; | |
2168 | } | |
2169 | ||
8930caba | 2170 | /* claim and dequeue */ |
a62428c0 | 2171 | debug_work_deactivate(work); |
023f27d3 | 2172 | hash_add(gcwq->busy_hash, &worker->hentry, (unsigned long)work); |
c34056a3 | 2173 | worker->current_work = work; |
a2c1c57b | 2174 | worker->current_func = work->func; |
8cca0eea | 2175 | worker->current_cwq = cwq; |
73f53c4a | 2176 | work_color = get_work_color(work); |
7a22ad75 | 2177 | |
a62428c0 TH |
2178 | list_del_init(&work->entry); |
2179 | ||
fb0e7beb TH |
2180 | /* |
2181 | * CPU intensive works don't participate in concurrency | |
2182 | * management. They're the scheduler's responsibility. | |
2183 | */ | |
2184 | if (unlikely(cpu_intensive)) | |
2185 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
2186 | ||
974271c4 TH |
2187 | /* |
2188 | * Unbound gcwq isn't concurrency managed and work items should be | |
2189 | * executed ASAP. Wake up another worker if necessary. | |
2190 | */ | |
63d95a91 TH |
2191 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
2192 | wake_up_worker(pool); | |
974271c4 | 2193 | |
8930caba | 2194 | /* |
23657bb1 TH |
2195 | * Record the last CPU and clear PENDING which should be the last |
2196 | * update to @work. Also, do this inside @gcwq->lock so that | |
2197 | * PENDING and queued state changes happen together while IRQ is | |
2198 | * disabled. | |
8930caba | 2199 | */ |
8930caba | 2200 | set_work_cpu_and_clear_pending(work, gcwq->cpu); |
a62428c0 | 2201 | |
8b03ae3c | 2202 | spin_unlock_irq(&gcwq->lock); |
a62428c0 | 2203 | |
e159489b | 2204 | lock_map_acquire_read(&cwq->wq->lockdep_map); |
a62428c0 | 2205 | lock_map_acquire(&lockdep_map); |
e36c886a | 2206 | trace_workqueue_execute_start(work); |
a2c1c57b | 2207 | worker->current_func(work); |
e36c886a AV |
2208 | /* |
2209 | * While we must be careful to not use "work" after this, the trace | |
2210 | * point will only record its address. | |
2211 | */ | |
2212 | trace_workqueue_execute_end(work); | |
a62428c0 TH |
2213 | lock_map_release(&lockdep_map); |
2214 | lock_map_release(&cwq->wq->lockdep_map); | |
2215 | ||
2216 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
044c782c VI |
2217 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
2218 | " last function: %pf\n", | |
a2c1c57b TH |
2219 | current->comm, preempt_count(), task_pid_nr(current), |
2220 | worker->current_func); | |
a62428c0 TH |
2221 | debug_show_held_locks(current); |
2222 | dump_stack(); | |
2223 | } | |
2224 | ||
8b03ae3c | 2225 | spin_lock_irq(&gcwq->lock); |
a62428c0 | 2226 | |
fb0e7beb TH |
2227 | /* clear cpu intensive status */ |
2228 | if (unlikely(cpu_intensive)) | |
2229 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
2230 | ||
a62428c0 | 2231 | /* we're done with it, release */ |
42f8570f | 2232 | hash_del(&worker->hentry); |
c34056a3 | 2233 | worker->current_work = NULL; |
a2c1c57b | 2234 | worker->current_func = NULL; |
8cca0eea | 2235 | worker->current_cwq = NULL; |
b3f9f405 | 2236 | cwq_dec_nr_in_flight(cwq, work_color); |
a62428c0 TH |
2237 | } |
2238 | ||
affee4b2 TH |
2239 | /** |
2240 | * process_scheduled_works - process scheduled works | |
2241 | * @worker: self | |
2242 | * | |
2243 | * Process all scheduled works. Please note that the scheduled list | |
2244 | * may change while processing a work, so this function repeatedly | |
2245 | * fetches a work from the top and executes it. | |
2246 | * | |
2247 | * CONTEXT: | |
8b03ae3c | 2248 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed |
affee4b2 TH |
2249 | * multiple times. |
2250 | */ | |
2251 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 2252 | { |
affee4b2 TH |
2253 | while (!list_empty(&worker->scheduled)) { |
2254 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 2255 | struct work_struct, entry); |
c34056a3 | 2256 | process_one_work(worker, work); |
1da177e4 | 2257 | } |
1da177e4 LT |
2258 | } |
2259 | ||
4690c4ab TH |
2260 | /** |
2261 | * worker_thread - the worker thread function | |
c34056a3 | 2262 | * @__worker: self |
4690c4ab | 2263 | * |
e22bee78 TH |
2264 | * The gcwq worker thread function. There's a single dynamic pool of |
2265 | * these per each cpu. These workers process all works regardless of | |
2266 | * their specific target workqueue. The only exception is works which | |
2267 | * belong to workqueues with a rescuer which will be explained in | |
2268 | * rescuer_thread(). | |
4690c4ab | 2269 | */ |
c34056a3 | 2270 | static int worker_thread(void *__worker) |
1da177e4 | 2271 | { |
c34056a3 | 2272 | struct worker *worker = __worker; |
bd7bdd43 TH |
2273 | struct worker_pool *pool = worker->pool; |
2274 | struct global_cwq *gcwq = pool->gcwq; | |
1da177e4 | 2275 | |
e22bee78 TH |
2276 | /* tell the scheduler that this is a workqueue worker */ |
2277 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 2278 | woke_up: |
c8e55f36 | 2279 | spin_lock_irq(&gcwq->lock); |
1da177e4 | 2280 | |
5f7dabfd LJ |
2281 | /* we are off idle list if destruction or rebind is requested */ |
2282 | if (unlikely(list_empty(&worker->entry))) { | |
c8e55f36 | 2283 | spin_unlock_irq(&gcwq->lock); |
25511a47 | 2284 | |
5f7dabfd | 2285 | /* if DIE is set, destruction is requested */ |
25511a47 TH |
2286 | if (worker->flags & WORKER_DIE) { |
2287 | worker->task->flags &= ~PF_WQ_WORKER; | |
2288 | return 0; | |
2289 | } | |
2290 | ||
5f7dabfd | 2291 | /* otherwise, rebind */ |
25511a47 TH |
2292 | idle_worker_rebind(worker); |
2293 | goto woke_up; | |
c8e55f36 | 2294 | } |
affee4b2 | 2295 | |
c8e55f36 | 2296 | worker_leave_idle(worker); |
db7bccf4 | 2297 | recheck: |
e22bee78 | 2298 | /* no more worker necessary? */ |
63d95a91 | 2299 | if (!need_more_worker(pool)) |
e22bee78 TH |
2300 | goto sleep; |
2301 | ||
2302 | /* do we need to manage? */ | |
63d95a91 | 2303 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
2304 | goto recheck; |
2305 | ||
c8e55f36 TH |
2306 | /* |
2307 | * ->scheduled list can only be filled while a worker is | |
2308 | * preparing to process a work or actually processing it. | |
2309 | * Make sure nobody diddled with it while I was sleeping. | |
2310 | */ | |
2311 | BUG_ON(!list_empty(&worker->scheduled)); | |
2312 | ||
e22bee78 TH |
2313 | /* |
2314 | * When control reaches this point, we're guaranteed to have | |
2315 | * at least one idle worker or that someone else has already | |
2316 | * assumed the manager role. | |
2317 | */ | |
2318 | worker_clr_flags(worker, WORKER_PREP); | |
2319 | ||
2320 | do { | |
c8e55f36 | 2321 | struct work_struct *work = |
bd7bdd43 | 2322 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2323 | struct work_struct, entry); |
2324 | ||
2325 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
2326 | /* optimization path, not strictly necessary */ | |
2327 | process_one_work(worker, work); | |
2328 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2329 | process_scheduled_works(worker); |
c8e55f36 TH |
2330 | } else { |
2331 | move_linked_works(work, &worker->scheduled, NULL); | |
2332 | process_scheduled_works(worker); | |
affee4b2 | 2333 | } |
63d95a91 | 2334 | } while (keep_working(pool)); |
e22bee78 TH |
2335 | |
2336 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 2337 | sleep: |
63d95a91 | 2338 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
e22bee78 | 2339 | goto recheck; |
d313dd85 | 2340 | |
c8e55f36 | 2341 | /* |
e22bee78 TH |
2342 | * gcwq->lock is held and there's no work to process and no |
2343 | * need to manage, sleep. Workers are woken up only while | |
2344 | * holding gcwq->lock or from local cpu, so setting the | |
2345 | * current state before releasing gcwq->lock is enough to | |
2346 | * prevent losing any event. | |
c8e55f36 TH |
2347 | */ |
2348 | worker_enter_idle(worker); | |
2349 | __set_current_state(TASK_INTERRUPTIBLE); | |
2350 | spin_unlock_irq(&gcwq->lock); | |
2351 | schedule(); | |
2352 | goto woke_up; | |
1da177e4 LT |
2353 | } |
2354 | ||
e22bee78 TH |
2355 | /** |
2356 | * rescuer_thread - the rescuer thread function | |
111c225a | 2357 | * @__rescuer: self |
e22bee78 TH |
2358 | * |
2359 | * Workqueue rescuer thread function. There's one rescuer for each | |
2360 | * workqueue which has WQ_RESCUER set. | |
2361 | * | |
2362 | * Regular work processing on a gcwq may block trying to create a new | |
2363 | * worker which uses GFP_KERNEL allocation which has slight chance of | |
2364 | * developing into deadlock if some works currently on the same queue | |
2365 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2366 | * the problem rescuer solves. | |
2367 | * | |
2368 | * When such condition is possible, the gcwq summons rescuers of all | |
2369 | * workqueues which have works queued on the gcwq and let them process | |
2370 | * those works so that forward progress can be guaranteed. | |
2371 | * | |
2372 | * This should happen rarely. | |
2373 | */ | |
111c225a | 2374 | static int rescuer_thread(void *__rescuer) |
e22bee78 | 2375 | { |
111c225a TH |
2376 | struct worker *rescuer = __rescuer; |
2377 | struct workqueue_struct *wq = rescuer->rescue_wq; | |
e22bee78 | 2378 | struct list_head *scheduled = &rescuer->scheduled; |
f3421797 | 2379 | bool is_unbound = wq->flags & WQ_UNBOUND; |
e22bee78 TH |
2380 | unsigned int cpu; |
2381 | ||
2382 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
111c225a TH |
2383 | |
2384 | /* | |
2385 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it | |
2386 | * doesn't participate in concurrency management. | |
2387 | */ | |
2388 | rescuer->task->flags |= PF_WQ_WORKER; | |
e22bee78 TH |
2389 | repeat: |
2390 | set_current_state(TASK_INTERRUPTIBLE); | |
2391 | ||
412d32e6 MG |
2392 | if (kthread_should_stop()) { |
2393 | __set_current_state(TASK_RUNNING); | |
111c225a | 2394 | rescuer->task->flags &= ~PF_WQ_WORKER; |
e22bee78 | 2395 | return 0; |
412d32e6 | 2396 | } |
e22bee78 | 2397 | |
f3421797 TH |
2398 | /* |
2399 | * See whether any cpu is asking for help. Unbounded | |
2400 | * workqueues use cpu 0 in mayday_mask for CPU_UNBOUND. | |
2401 | */ | |
f2e005aa | 2402 | for_each_mayday_cpu(cpu, wq->mayday_mask) { |
f3421797 TH |
2403 | unsigned int tcpu = is_unbound ? WORK_CPU_UNBOUND : cpu; |
2404 | struct cpu_workqueue_struct *cwq = get_cwq(tcpu, wq); | |
bd7bdd43 TH |
2405 | struct worker_pool *pool = cwq->pool; |
2406 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
2407 | struct work_struct *work, *n; |
2408 | ||
2409 | __set_current_state(TASK_RUNNING); | |
f2e005aa | 2410 | mayday_clear_cpu(cpu, wq->mayday_mask); |
e22bee78 TH |
2411 | |
2412 | /* migrate to the target cpu if possible */ | |
bd7bdd43 | 2413 | rescuer->pool = pool; |
e22bee78 TH |
2414 | worker_maybe_bind_and_lock(rescuer); |
2415 | ||
2416 | /* | |
2417 | * Slurp in all works issued via this workqueue and | |
2418 | * process'em. | |
2419 | */ | |
2420 | BUG_ON(!list_empty(&rescuer->scheduled)); | |
bd7bdd43 | 2421 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
e22bee78 TH |
2422 | if (get_work_cwq(work) == cwq) |
2423 | move_linked_works(work, scheduled, &n); | |
2424 | ||
2425 | process_scheduled_works(rescuer); | |
7576958a TH |
2426 | |
2427 | /* | |
2428 | * Leave this gcwq. If keep_working() is %true, notify a | |
2429 | * regular worker; otherwise, we end up with 0 concurrency | |
2430 | * and stalling the execution. | |
2431 | */ | |
63d95a91 TH |
2432 | if (keep_working(pool)) |
2433 | wake_up_worker(pool); | |
7576958a | 2434 | |
e22bee78 TH |
2435 | spin_unlock_irq(&gcwq->lock); |
2436 | } | |
2437 | ||
111c225a TH |
2438 | /* rescuers should never participate in concurrency management */ |
2439 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); | |
e22bee78 TH |
2440 | schedule(); |
2441 | goto repeat; | |
1da177e4 LT |
2442 | } |
2443 | ||
fc2e4d70 ON |
2444 | struct wq_barrier { |
2445 | struct work_struct work; | |
2446 | struct completion done; | |
2447 | }; | |
2448 | ||
2449 | static void wq_barrier_func(struct work_struct *work) | |
2450 | { | |
2451 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2452 | complete(&barr->done); | |
2453 | } | |
2454 | ||
4690c4ab TH |
2455 | /** |
2456 | * insert_wq_barrier - insert a barrier work | |
2457 | * @cwq: cwq to insert barrier into | |
2458 | * @barr: wq_barrier to insert | |
affee4b2 TH |
2459 | * @target: target work to attach @barr to |
2460 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2461 | * |
affee4b2 TH |
2462 | * @barr is linked to @target such that @barr is completed only after |
2463 | * @target finishes execution. Please note that the ordering | |
2464 | * guarantee is observed only with respect to @target and on the local | |
2465 | * cpu. | |
2466 | * | |
2467 | * Currently, a queued barrier can't be canceled. This is because | |
2468 | * try_to_grab_pending() can't determine whether the work to be | |
2469 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2470 | * flag of the previous work while there must be a valid next work | |
2471 | * after a work with LINKED flag set. | |
2472 | * | |
2473 | * Note that when @worker is non-NULL, @target may be modified | |
2474 | * underneath us, so we can't reliably determine cwq from @target. | |
4690c4ab TH |
2475 | * |
2476 | * CONTEXT: | |
8b03ae3c | 2477 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 2478 | */ |
83c22520 | 2479 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, |
affee4b2 TH |
2480 | struct wq_barrier *barr, |
2481 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2482 | { |
affee4b2 TH |
2483 | struct list_head *head; |
2484 | unsigned int linked = 0; | |
2485 | ||
dc186ad7 | 2486 | /* |
8b03ae3c | 2487 | * debugobject calls are safe here even with gcwq->lock locked |
dc186ad7 TG |
2488 | * as we know for sure that this will not trigger any of the |
2489 | * checks and call back into the fixup functions where we | |
2490 | * might deadlock. | |
2491 | */ | |
ca1cab37 | 2492 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2493 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2494 | init_completion(&barr->done); |
83c22520 | 2495 | |
affee4b2 TH |
2496 | /* |
2497 | * If @target is currently being executed, schedule the | |
2498 | * barrier to the worker; otherwise, put it after @target. | |
2499 | */ | |
2500 | if (worker) | |
2501 | head = worker->scheduled.next; | |
2502 | else { | |
2503 | unsigned long *bits = work_data_bits(target); | |
2504 | ||
2505 | head = target->entry.next; | |
2506 | /* there can already be other linked works, inherit and set */ | |
2507 | linked = *bits & WORK_STRUCT_LINKED; | |
2508 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2509 | } | |
2510 | ||
dc186ad7 | 2511 | debug_work_activate(&barr->work); |
affee4b2 TH |
2512 | insert_work(cwq, &barr->work, head, |
2513 | work_color_to_flags(WORK_NO_COLOR) | linked); | |
fc2e4d70 ON |
2514 | } |
2515 | ||
73f53c4a TH |
2516 | /** |
2517 | * flush_workqueue_prep_cwqs - prepare cwqs for workqueue flushing | |
2518 | * @wq: workqueue being flushed | |
2519 | * @flush_color: new flush color, < 0 for no-op | |
2520 | * @work_color: new work color, < 0 for no-op | |
2521 | * | |
2522 | * Prepare cwqs for workqueue flushing. | |
2523 | * | |
2524 | * If @flush_color is non-negative, flush_color on all cwqs should be | |
2525 | * -1. If no cwq has in-flight commands at the specified color, all | |
2526 | * cwq->flush_color's stay at -1 and %false is returned. If any cwq | |
2527 | * has in flight commands, its cwq->flush_color is set to | |
2528 | * @flush_color, @wq->nr_cwqs_to_flush is updated accordingly, cwq | |
2529 | * wakeup logic is armed and %true is returned. | |
2530 | * | |
2531 | * The caller should have initialized @wq->first_flusher prior to | |
2532 | * calling this function with non-negative @flush_color. If | |
2533 | * @flush_color is negative, no flush color update is done and %false | |
2534 | * is returned. | |
2535 | * | |
2536 | * If @work_color is non-negative, all cwqs should have the same | |
2537 | * work_color which is previous to @work_color and all will be | |
2538 | * advanced to @work_color. | |
2539 | * | |
2540 | * CONTEXT: | |
2541 | * mutex_lock(wq->flush_mutex). | |
2542 | * | |
2543 | * RETURNS: | |
2544 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2545 | * otherwise. | |
2546 | */ | |
2547 | static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq, | |
2548 | int flush_color, int work_color) | |
1da177e4 | 2549 | { |
73f53c4a TH |
2550 | bool wait = false; |
2551 | unsigned int cpu; | |
1da177e4 | 2552 | |
73f53c4a TH |
2553 | if (flush_color >= 0) { |
2554 | BUG_ON(atomic_read(&wq->nr_cwqs_to_flush)); | |
2555 | atomic_set(&wq->nr_cwqs_to_flush, 1); | |
1da177e4 | 2556 | } |
2355b70f | 2557 | |
f3421797 | 2558 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a | 2559 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
bd7bdd43 | 2560 | struct global_cwq *gcwq = cwq->pool->gcwq; |
fc2e4d70 | 2561 | |
8b03ae3c | 2562 | spin_lock_irq(&gcwq->lock); |
83c22520 | 2563 | |
73f53c4a TH |
2564 | if (flush_color >= 0) { |
2565 | BUG_ON(cwq->flush_color != -1); | |
fc2e4d70 | 2566 | |
73f53c4a TH |
2567 | if (cwq->nr_in_flight[flush_color]) { |
2568 | cwq->flush_color = flush_color; | |
2569 | atomic_inc(&wq->nr_cwqs_to_flush); | |
2570 | wait = true; | |
2571 | } | |
2572 | } | |
1da177e4 | 2573 | |
73f53c4a TH |
2574 | if (work_color >= 0) { |
2575 | BUG_ON(work_color != work_next_color(cwq->work_color)); | |
2576 | cwq->work_color = work_color; | |
2577 | } | |
1da177e4 | 2578 | |
8b03ae3c | 2579 | spin_unlock_irq(&gcwq->lock); |
1da177e4 | 2580 | } |
2355b70f | 2581 | |
73f53c4a TH |
2582 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_cwqs_to_flush)) |
2583 | complete(&wq->first_flusher->done); | |
14441960 | 2584 | |
73f53c4a | 2585 | return wait; |
1da177e4 LT |
2586 | } |
2587 | ||
0fcb78c2 | 2588 | /** |
1da177e4 | 2589 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2590 | * @wq: workqueue to flush |
1da177e4 LT |
2591 | * |
2592 | * Forces execution of the workqueue and blocks until its completion. | |
2593 | * This is typically used in driver shutdown handlers. | |
2594 | * | |
fc2e4d70 ON |
2595 | * We sleep until all works which were queued on entry have been handled, |
2596 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 2597 | */ |
7ad5b3a5 | 2598 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2599 | { |
73f53c4a TH |
2600 | struct wq_flusher this_flusher = { |
2601 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2602 | .flush_color = -1, | |
2603 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2604 | }; | |
2605 | int next_color; | |
1da177e4 | 2606 | |
3295f0ef IM |
2607 | lock_map_acquire(&wq->lockdep_map); |
2608 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
2609 | |
2610 | mutex_lock(&wq->flush_mutex); | |
2611 | ||
2612 | /* | |
2613 | * Start-to-wait phase | |
2614 | */ | |
2615 | next_color = work_next_color(wq->work_color); | |
2616 | ||
2617 | if (next_color != wq->flush_color) { | |
2618 | /* | |
2619 | * Color space is not full. The current work_color | |
2620 | * becomes our flush_color and work_color is advanced | |
2621 | * by one. | |
2622 | */ | |
2623 | BUG_ON(!list_empty(&wq->flusher_overflow)); | |
2624 | this_flusher.flush_color = wq->work_color; | |
2625 | wq->work_color = next_color; | |
2626 | ||
2627 | if (!wq->first_flusher) { | |
2628 | /* no flush in progress, become the first flusher */ | |
2629 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2630 | ||
2631 | wq->first_flusher = &this_flusher; | |
2632 | ||
2633 | if (!flush_workqueue_prep_cwqs(wq, wq->flush_color, | |
2634 | wq->work_color)) { | |
2635 | /* nothing to flush, done */ | |
2636 | wq->flush_color = next_color; | |
2637 | wq->first_flusher = NULL; | |
2638 | goto out_unlock; | |
2639 | } | |
2640 | } else { | |
2641 | /* wait in queue */ | |
2642 | BUG_ON(wq->flush_color == this_flusher.flush_color); | |
2643 | list_add_tail(&this_flusher.list, &wq->flusher_queue); | |
2644 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2645 | } | |
2646 | } else { | |
2647 | /* | |
2648 | * Oops, color space is full, wait on overflow queue. | |
2649 | * The next flush completion will assign us | |
2650 | * flush_color and transfer to flusher_queue. | |
2651 | */ | |
2652 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2653 | } | |
2654 | ||
2655 | mutex_unlock(&wq->flush_mutex); | |
2656 | ||
2657 | wait_for_completion(&this_flusher.done); | |
2658 | ||
2659 | /* | |
2660 | * Wake-up-and-cascade phase | |
2661 | * | |
2662 | * First flushers are responsible for cascading flushes and | |
2663 | * handling overflow. Non-first flushers can simply return. | |
2664 | */ | |
2665 | if (wq->first_flusher != &this_flusher) | |
2666 | return; | |
2667 | ||
2668 | mutex_lock(&wq->flush_mutex); | |
2669 | ||
4ce48b37 TH |
2670 | /* we might have raced, check again with mutex held */ |
2671 | if (wq->first_flusher != &this_flusher) | |
2672 | goto out_unlock; | |
2673 | ||
73f53c4a TH |
2674 | wq->first_flusher = NULL; |
2675 | ||
2676 | BUG_ON(!list_empty(&this_flusher.list)); | |
2677 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2678 | ||
2679 | while (true) { | |
2680 | struct wq_flusher *next, *tmp; | |
2681 | ||
2682 | /* complete all the flushers sharing the current flush color */ | |
2683 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2684 | if (next->flush_color != wq->flush_color) | |
2685 | break; | |
2686 | list_del_init(&next->list); | |
2687 | complete(&next->done); | |
2688 | } | |
2689 | ||
2690 | BUG_ON(!list_empty(&wq->flusher_overflow) && | |
2691 | wq->flush_color != work_next_color(wq->work_color)); | |
2692 | ||
2693 | /* this flush_color is finished, advance by one */ | |
2694 | wq->flush_color = work_next_color(wq->flush_color); | |
2695 | ||
2696 | /* one color has been freed, handle overflow queue */ | |
2697 | if (!list_empty(&wq->flusher_overflow)) { | |
2698 | /* | |
2699 | * Assign the same color to all overflowed | |
2700 | * flushers, advance work_color and append to | |
2701 | * flusher_queue. This is the start-to-wait | |
2702 | * phase for these overflowed flushers. | |
2703 | */ | |
2704 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2705 | tmp->flush_color = wq->work_color; | |
2706 | ||
2707 | wq->work_color = work_next_color(wq->work_color); | |
2708 | ||
2709 | list_splice_tail_init(&wq->flusher_overflow, | |
2710 | &wq->flusher_queue); | |
2711 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2712 | } | |
2713 | ||
2714 | if (list_empty(&wq->flusher_queue)) { | |
2715 | BUG_ON(wq->flush_color != wq->work_color); | |
2716 | break; | |
2717 | } | |
2718 | ||
2719 | /* | |
2720 | * Need to flush more colors. Make the next flusher | |
2721 | * the new first flusher and arm cwqs. | |
2722 | */ | |
2723 | BUG_ON(wq->flush_color == wq->work_color); | |
2724 | BUG_ON(wq->flush_color != next->flush_color); | |
2725 | ||
2726 | list_del_init(&next->list); | |
2727 | wq->first_flusher = next; | |
2728 | ||
2729 | if (flush_workqueue_prep_cwqs(wq, wq->flush_color, -1)) | |
2730 | break; | |
2731 | ||
2732 | /* | |
2733 | * Meh... this color is already done, clear first | |
2734 | * flusher and repeat cascading. | |
2735 | */ | |
2736 | wq->first_flusher = NULL; | |
2737 | } | |
2738 | ||
2739 | out_unlock: | |
2740 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 2741 | } |
ae90dd5d | 2742 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2743 | |
9c5a2ba7 TH |
2744 | /** |
2745 | * drain_workqueue - drain a workqueue | |
2746 | * @wq: workqueue to drain | |
2747 | * | |
2748 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2749 | * only chain queueing is allowed. IOW, only currently pending or running | |
2750 | * work items on @wq can queue further work items on it. @wq is flushed | |
2751 | * repeatedly until it becomes empty. The number of flushing is detemined | |
2752 | * by the depth of chaining and should be relatively short. Whine if it | |
2753 | * takes too long. | |
2754 | */ | |
2755 | void drain_workqueue(struct workqueue_struct *wq) | |
2756 | { | |
2757 | unsigned int flush_cnt = 0; | |
2758 | unsigned int cpu; | |
2759 | ||
2760 | /* | |
2761 | * __queue_work() needs to test whether there are drainers, is much | |
2762 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
2763 | * Use WQ_DRAINING so that queue doesn't have to check nr_drainers. | |
2764 | */ | |
2765 | spin_lock(&workqueue_lock); | |
2766 | if (!wq->nr_drainers++) | |
2767 | wq->flags |= WQ_DRAINING; | |
2768 | spin_unlock(&workqueue_lock); | |
2769 | reflush: | |
2770 | flush_workqueue(wq); | |
2771 | ||
2772 | for_each_cwq_cpu(cpu, wq) { | |
2773 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
fa2563e4 | 2774 | bool drained; |
9c5a2ba7 | 2775 | |
bd7bdd43 | 2776 | spin_lock_irq(&cwq->pool->gcwq->lock); |
fa2563e4 | 2777 | drained = !cwq->nr_active && list_empty(&cwq->delayed_works); |
bd7bdd43 | 2778 | spin_unlock_irq(&cwq->pool->gcwq->lock); |
fa2563e4 TT |
2779 | |
2780 | if (drained) | |
9c5a2ba7 TH |
2781 | continue; |
2782 | ||
2783 | if (++flush_cnt == 10 || | |
2784 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
044c782c VI |
2785 | pr_warn("workqueue %s: flush on destruction isn't complete after %u tries\n", |
2786 | wq->name, flush_cnt); | |
9c5a2ba7 TH |
2787 | goto reflush; |
2788 | } | |
2789 | ||
2790 | spin_lock(&workqueue_lock); | |
2791 | if (!--wq->nr_drainers) | |
2792 | wq->flags &= ~WQ_DRAINING; | |
2793 | spin_unlock(&workqueue_lock); | |
2794 | } | |
2795 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2796 | ||
606a5020 | 2797 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) |
db700897 | 2798 | { |
affee4b2 | 2799 | struct worker *worker = NULL; |
8b03ae3c | 2800 | struct global_cwq *gcwq; |
db700897 | 2801 | struct cpu_workqueue_struct *cwq; |
db700897 ON |
2802 | |
2803 | might_sleep(); | |
7a22ad75 TH |
2804 | gcwq = get_work_gcwq(work); |
2805 | if (!gcwq) | |
baf59022 | 2806 | return false; |
db700897 | 2807 | |
8b03ae3c | 2808 | spin_lock_irq(&gcwq->lock); |
db700897 ON |
2809 | if (!list_empty(&work->entry)) { |
2810 | /* | |
2811 | * See the comment near try_to_grab_pending()->smp_rmb(). | |
7a22ad75 TH |
2812 | * If it was re-queued to a different gcwq under us, we |
2813 | * are not going to wait. | |
db700897 ON |
2814 | */ |
2815 | smp_rmb(); | |
7a22ad75 | 2816 | cwq = get_work_cwq(work); |
bd7bdd43 | 2817 | if (unlikely(!cwq || gcwq != cwq->pool->gcwq)) |
4690c4ab | 2818 | goto already_gone; |
606a5020 | 2819 | } else { |
7a22ad75 | 2820 | worker = find_worker_executing_work(gcwq, work); |
affee4b2 | 2821 | if (!worker) |
4690c4ab | 2822 | goto already_gone; |
7a22ad75 | 2823 | cwq = worker->current_cwq; |
606a5020 | 2824 | } |
db700897 | 2825 | |
baf59022 | 2826 | insert_wq_barrier(cwq, barr, work, worker); |
8b03ae3c | 2827 | spin_unlock_irq(&gcwq->lock); |
7a22ad75 | 2828 | |
e159489b TH |
2829 | /* |
2830 | * If @max_active is 1 or rescuer is in use, flushing another work | |
2831 | * item on the same workqueue may lead to deadlock. Make sure the | |
2832 | * flusher is not running on the same workqueue by verifying write | |
2833 | * access. | |
2834 | */ | |
2835 | if (cwq->wq->saved_max_active == 1 || cwq->wq->flags & WQ_RESCUER) | |
2836 | lock_map_acquire(&cwq->wq->lockdep_map); | |
2837 | else | |
2838 | lock_map_acquire_read(&cwq->wq->lockdep_map); | |
7a22ad75 | 2839 | lock_map_release(&cwq->wq->lockdep_map); |
e159489b | 2840 | |
401a8d04 | 2841 | return true; |
4690c4ab | 2842 | already_gone: |
8b03ae3c | 2843 | spin_unlock_irq(&gcwq->lock); |
401a8d04 | 2844 | return false; |
db700897 | 2845 | } |
baf59022 TH |
2846 | |
2847 | /** | |
2848 | * flush_work - wait for a work to finish executing the last queueing instance | |
2849 | * @work: the work to flush | |
2850 | * | |
606a5020 TH |
2851 | * Wait until @work has finished execution. @work is guaranteed to be idle |
2852 | * on return if it hasn't been requeued since flush started. | |
baf59022 TH |
2853 | * |
2854 | * RETURNS: | |
2855 | * %true if flush_work() waited for the work to finish execution, | |
2856 | * %false if it was already idle. | |
2857 | */ | |
2858 | bool flush_work(struct work_struct *work) | |
2859 | { | |
2860 | struct wq_barrier barr; | |
2861 | ||
0976dfc1 SB |
2862 | lock_map_acquire(&work->lockdep_map); |
2863 | lock_map_release(&work->lockdep_map); | |
2864 | ||
606a5020 | 2865 | if (start_flush_work(work, &barr)) { |
401a8d04 TH |
2866 | wait_for_completion(&barr.done); |
2867 | destroy_work_on_stack(&barr.work); | |
2868 | return true; | |
606a5020 | 2869 | } else { |
401a8d04 | 2870 | return false; |
6e84d644 | 2871 | } |
6e84d644 | 2872 | } |
606a5020 | 2873 | EXPORT_SYMBOL_GPL(flush_work); |
6e84d644 | 2874 | |
36e227d2 | 2875 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
1f1f642e | 2876 | { |
bbb68dfa | 2877 | unsigned long flags; |
1f1f642e ON |
2878 | int ret; |
2879 | ||
2880 | do { | |
bbb68dfa TH |
2881 | ret = try_to_grab_pending(work, is_dwork, &flags); |
2882 | /* | |
2883 | * If someone else is canceling, wait for the same event it | |
2884 | * would be waiting for before retrying. | |
2885 | */ | |
2886 | if (unlikely(ret == -ENOENT)) | |
606a5020 | 2887 | flush_work(work); |
1f1f642e ON |
2888 | } while (unlikely(ret < 0)); |
2889 | ||
bbb68dfa TH |
2890 | /* tell other tasks trying to grab @work to back off */ |
2891 | mark_work_canceling(work); | |
2892 | local_irq_restore(flags); | |
2893 | ||
606a5020 | 2894 | flush_work(work); |
7a22ad75 | 2895 | clear_work_data(work); |
1f1f642e ON |
2896 | return ret; |
2897 | } | |
2898 | ||
6e84d644 | 2899 | /** |
401a8d04 TH |
2900 | * cancel_work_sync - cancel a work and wait for it to finish |
2901 | * @work: the work to cancel | |
6e84d644 | 2902 | * |
401a8d04 TH |
2903 | * Cancel @work and wait for its execution to finish. This function |
2904 | * can be used even if the work re-queues itself or migrates to | |
2905 | * another workqueue. On return from this function, @work is | |
2906 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2907 | * |
401a8d04 TH |
2908 | * cancel_work_sync(&delayed_work->work) must not be used for |
2909 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2910 | * |
401a8d04 | 2911 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2912 | * queued can't be destroyed before this function returns. |
401a8d04 TH |
2913 | * |
2914 | * RETURNS: | |
2915 | * %true if @work was pending, %false otherwise. | |
6e84d644 | 2916 | */ |
401a8d04 | 2917 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2918 | { |
36e227d2 | 2919 | return __cancel_work_timer(work, false); |
b89deed3 | 2920 | } |
28e53bdd | 2921 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2922 | |
6e84d644 | 2923 | /** |
401a8d04 TH |
2924 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2925 | * @dwork: the delayed work to flush | |
6e84d644 | 2926 | * |
401a8d04 TH |
2927 | * Delayed timer is cancelled and the pending work is queued for |
2928 | * immediate execution. Like flush_work(), this function only | |
2929 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2930 | * |
401a8d04 TH |
2931 | * RETURNS: |
2932 | * %true if flush_work() waited for the work to finish execution, | |
2933 | * %false if it was already idle. | |
6e84d644 | 2934 | */ |
401a8d04 TH |
2935 | bool flush_delayed_work(struct delayed_work *dwork) |
2936 | { | |
8930caba | 2937 | local_irq_disable(); |
401a8d04 | 2938 | if (del_timer_sync(&dwork->timer)) |
1265057f | 2939 | __queue_work(dwork->cpu, |
401a8d04 | 2940 | get_work_cwq(&dwork->work)->wq, &dwork->work); |
8930caba | 2941 | local_irq_enable(); |
401a8d04 TH |
2942 | return flush_work(&dwork->work); |
2943 | } | |
2944 | EXPORT_SYMBOL(flush_delayed_work); | |
2945 | ||
09383498 | 2946 | /** |
57b30ae7 TH |
2947 | * cancel_delayed_work - cancel a delayed work |
2948 | * @dwork: delayed_work to cancel | |
09383498 | 2949 | * |
57b30ae7 TH |
2950 | * Kill off a pending delayed_work. Returns %true if @dwork was pending |
2951 | * and canceled; %false if wasn't pending. Note that the work callback | |
2952 | * function may still be running on return, unless it returns %true and the | |
2953 | * work doesn't re-arm itself. Explicitly flush or use | |
2954 | * cancel_delayed_work_sync() to wait on it. | |
09383498 | 2955 | * |
57b30ae7 | 2956 | * This function is safe to call from any context including IRQ handler. |
09383498 | 2957 | */ |
57b30ae7 | 2958 | bool cancel_delayed_work(struct delayed_work *dwork) |
09383498 | 2959 | { |
57b30ae7 TH |
2960 | unsigned long flags; |
2961 | int ret; | |
2962 | ||
2963 | do { | |
2964 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
2965 | } while (unlikely(ret == -EAGAIN)); | |
2966 | ||
2967 | if (unlikely(ret < 0)) | |
2968 | return false; | |
2969 | ||
2970 | set_work_cpu_and_clear_pending(&dwork->work, work_cpu(&dwork->work)); | |
2971 | local_irq_restore(flags); | |
c0158ca6 | 2972 | return ret; |
09383498 | 2973 | } |
57b30ae7 | 2974 | EXPORT_SYMBOL(cancel_delayed_work); |
09383498 | 2975 | |
401a8d04 TH |
2976 | /** |
2977 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
2978 | * @dwork: the delayed work cancel | |
2979 | * | |
2980 | * This is cancel_work_sync() for delayed works. | |
2981 | * | |
2982 | * RETURNS: | |
2983 | * %true if @dwork was pending, %false otherwise. | |
2984 | */ | |
2985 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 2986 | { |
36e227d2 | 2987 | return __cancel_work_timer(&dwork->work, true); |
6e84d644 | 2988 | } |
f5a421a4 | 2989 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 2990 | |
0fcb78c2 | 2991 | /** |
c1a220e7 ZR |
2992 | * schedule_work_on - put work task on a specific cpu |
2993 | * @cpu: cpu to put the work task on | |
2994 | * @work: job to be done | |
2995 | * | |
2996 | * This puts a job on a specific cpu | |
2997 | */ | |
d4283e93 | 2998 | bool schedule_work_on(int cpu, struct work_struct *work) |
c1a220e7 | 2999 | { |
d320c038 | 3000 | return queue_work_on(cpu, system_wq, work); |
c1a220e7 ZR |
3001 | } |
3002 | EXPORT_SYMBOL(schedule_work_on); | |
3003 | ||
0fcb78c2 | 3004 | /** |
0fcb78c2 REB |
3005 | * schedule_work - put work task in global workqueue |
3006 | * @work: job to be done | |
0fcb78c2 | 3007 | * |
d4283e93 TH |
3008 | * Returns %false if @work was already on the kernel-global workqueue and |
3009 | * %true otherwise. | |
5b0f437d BVA |
3010 | * |
3011 | * This puts a job in the kernel-global workqueue if it was not already | |
3012 | * queued and leaves it in the same position on the kernel-global | |
3013 | * workqueue otherwise. | |
0fcb78c2 | 3014 | */ |
d4283e93 | 3015 | bool schedule_work(struct work_struct *work) |
1da177e4 | 3016 | { |
d320c038 | 3017 | return queue_work(system_wq, work); |
1da177e4 | 3018 | } |
ae90dd5d | 3019 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 3020 | |
0fcb78c2 REB |
3021 | /** |
3022 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
3023 | * @cpu: cpu to use | |
52bad64d | 3024 | * @dwork: job to be done |
0fcb78c2 REB |
3025 | * @delay: number of jiffies to wait |
3026 | * | |
3027 | * After waiting for a given time this puts a job in the kernel-global | |
3028 | * workqueue on the specified CPU. | |
3029 | */ | |
d4283e93 TH |
3030 | bool schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
3031 | unsigned long delay) | |
1da177e4 | 3032 | { |
d320c038 | 3033 | return queue_delayed_work_on(cpu, system_wq, dwork, delay); |
1da177e4 | 3034 | } |
ae90dd5d | 3035 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 3036 | |
0fcb78c2 REB |
3037 | /** |
3038 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
3039 | * @dwork: job to be done |
3040 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
3041 | * |
3042 | * After waiting for a given time this puts a job in the kernel-global | |
3043 | * workqueue. | |
3044 | */ | |
d4283e93 | 3045 | bool schedule_delayed_work(struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 3046 | { |
d320c038 | 3047 | return queue_delayed_work(system_wq, dwork, delay); |
1da177e4 | 3048 | } |
ae90dd5d | 3049 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 3050 | |
b6136773 | 3051 | /** |
31ddd871 | 3052 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 3053 | * @func: the function to call |
b6136773 | 3054 | * |
31ddd871 TH |
3055 | * schedule_on_each_cpu() executes @func on each online CPU using the |
3056 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 3057 | * schedule_on_each_cpu() is very slow. |
31ddd871 TH |
3058 | * |
3059 | * RETURNS: | |
3060 | * 0 on success, -errno on failure. | |
b6136773 | 3061 | */ |
65f27f38 | 3062 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
3063 | { |
3064 | int cpu; | |
38f51568 | 3065 | struct work_struct __percpu *works; |
15316ba8 | 3066 | |
b6136773 AM |
3067 | works = alloc_percpu(struct work_struct); |
3068 | if (!works) | |
15316ba8 | 3069 | return -ENOMEM; |
b6136773 | 3070 | |
93981800 TH |
3071 | get_online_cpus(); |
3072 | ||
15316ba8 | 3073 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
3074 | struct work_struct *work = per_cpu_ptr(works, cpu); |
3075 | ||
3076 | INIT_WORK(work, func); | |
b71ab8c2 | 3077 | schedule_work_on(cpu, work); |
65a64464 | 3078 | } |
93981800 TH |
3079 | |
3080 | for_each_online_cpu(cpu) | |
3081 | flush_work(per_cpu_ptr(works, cpu)); | |
3082 | ||
95402b38 | 3083 | put_online_cpus(); |
b6136773 | 3084 | free_percpu(works); |
15316ba8 CL |
3085 | return 0; |
3086 | } | |
3087 | ||
eef6a7d5 AS |
3088 | /** |
3089 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
3090 | * | |
3091 | * Forces execution of the kernel-global workqueue and blocks until its | |
3092 | * completion. | |
3093 | * | |
3094 | * Think twice before calling this function! It's very easy to get into | |
3095 | * trouble if you don't take great care. Either of the following situations | |
3096 | * will lead to deadlock: | |
3097 | * | |
3098 | * One of the work items currently on the workqueue needs to acquire | |
3099 | * a lock held by your code or its caller. | |
3100 | * | |
3101 | * Your code is running in the context of a work routine. | |
3102 | * | |
3103 | * They will be detected by lockdep when they occur, but the first might not | |
3104 | * occur very often. It depends on what work items are on the workqueue and | |
3105 | * what locks they need, which you have no control over. | |
3106 | * | |
3107 | * In most situations flushing the entire workqueue is overkill; you merely | |
3108 | * need to know that a particular work item isn't queued and isn't running. | |
3109 | * In such cases you should use cancel_delayed_work_sync() or | |
3110 | * cancel_work_sync() instead. | |
3111 | */ | |
1da177e4 LT |
3112 | void flush_scheduled_work(void) |
3113 | { | |
d320c038 | 3114 | flush_workqueue(system_wq); |
1da177e4 | 3115 | } |
ae90dd5d | 3116 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 3117 | |
1fa44eca JB |
3118 | /** |
3119 | * execute_in_process_context - reliably execute the routine with user context | |
3120 | * @fn: the function to execute | |
1fa44eca JB |
3121 | * @ew: guaranteed storage for the execute work structure (must |
3122 | * be available when the work executes) | |
3123 | * | |
3124 | * Executes the function immediately if process context is available, | |
3125 | * otherwise schedules the function for delayed execution. | |
3126 | * | |
3127 | * Returns: 0 - function was executed | |
3128 | * 1 - function was scheduled for execution | |
3129 | */ | |
65f27f38 | 3130 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
3131 | { |
3132 | if (!in_interrupt()) { | |
65f27f38 | 3133 | fn(&ew->work); |
1fa44eca JB |
3134 | return 0; |
3135 | } | |
3136 | ||
65f27f38 | 3137 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
3138 | schedule_work(&ew->work); |
3139 | ||
3140 | return 1; | |
3141 | } | |
3142 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
3143 | ||
1da177e4 LT |
3144 | int keventd_up(void) |
3145 | { | |
d320c038 | 3146 | return system_wq != NULL; |
1da177e4 LT |
3147 | } |
3148 | ||
bdbc5dd7 | 3149 | static int alloc_cwqs(struct workqueue_struct *wq) |
0f900049 | 3150 | { |
65a64464 | 3151 | /* |
0f900049 TH |
3152 | * cwqs are forced aligned according to WORK_STRUCT_FLAG_BITS. |
3153 | * Make sure that the alignment isn't lower than that of | |
3154 | * unsigned long long. | |
65a64464 | 3155 | */ |
0f900049 TH |
3156 | const size_t size = sizeof(struct cpu_workqueue_struct); |
3157 | const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS, | |
3158 | __alignof__(unsigned long long)); | |
65a64464 | 3159 | |
e06ffa1e | 3160 | if (!(wq->flags & WQ_UNBOUND)) |
f3421797 | 3161 | wq->cpu_wq.pcpu = __alloc_percpu(size, align); |
931ac77e | 3162 | else { |
f3421797 TH |
3163 | void *ptr; |
3164 | ||
3165 | /* | |
3166 | * Allocate enough room to align cwq and put an extra | |
3167 | * pointer at the end pointing back to the originally | |
3168 | * allocated pointer which will be used for free. | |
3169 | */ | |
3170 | ptr = kzalloc(size + align + sizeof(void *), GFP_KERNEL); | |
3171 | if (ptr) { | |
3172 | wq->cpu_wq.single = PTR_ALIGN(ptr, align); | |
3173 | *(void **)(wq->cpu_wq.single + 1) = ptr; | |
3174 | } | |
bdbc5dd7 | 3175 | } |
f3421797 | 3176 | |
0415b00d | 3177 | /* just in case, make sure it's actually aligned */ |
bdbc5dd7 TH |
3178 | BUG_ON(!IS_ALIGNED(wq->cpu_wq.v, align)); |
3179 | return wq->cpu_wq.v ? 0 : -ENOMEM; | |
0f900049 TH |
3180 | } |
3181 | ||
bdbc5dd7 | 3182 | static void free_cwqs(struct workqueue_struct *wq) |
0f900049 | 3183 | { |
e06ffa1e | 3184 | if (!(wq->flags & WQ_UNBOUND)) |
f3421797 TH |
3185 | free_percpu(wq->cpu_wq.pcpu); |
3186 | else if (wq->cpu_wq.single) { | |
3187 | /* the pointer to free is stored right after the cwq */ | |
bdbc5dd7 | 3188 | kfree(*(void **)(wq->cpu_wq.single + 1)); |
f3421797 | 3189 | } |
0f900049 TH |
3190 | } |
3191 | ||
f3421797 TH |
3192 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
3193 | const char *name) | |
b71ab8c2 | 3194 | { |
f3421797 TH |
3195 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
3196 | ||
3197 | if (max_active < 1 || max_active > lim) | |
044c782c VI |
3198 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
3199 | max_active, name, 1, lim); | |
b71ab8c2 | 3200 | |
f3421797 | 3201 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
3202 | } |
3203 | ||
b196be89 | 3204 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
d320c038 TH |
3205 | unsigned int flags, |
3206 | int max_active, | |
3207 | struct lock_class_key *key, | |
b196be89 | 3208 | const char *lock_name, ...) |
1da177e4 | 3209 | { |
b196be89 | 3210 | va_list args, args1; |
1da177e4 | 3211 | struct workqueue_struct *wq; |
c34056a3 | 3212 | unsigned int cpu; |
b196be89 TH |
3213 | size_t namelen; |
3214 | ||
3215 | /* determine namelen, allocate wq and format name */ | |
3216 | va_start(args, lock_name); | |
3217 | va_copy(args1, args); | |
3218 | namelen = vsnprintf(NULL, 0, fmt, args) + 1; | |
3219 | ||
3220 | wq = kzalloc(sizeof(*wq) + namelen, GFP_KERNEL); | |
3221 | if (!wq) | |
3222 | goto err; | |
3223 | ||
3224 | vsnprintf(wq->name, namelen, fmt, args1); | |
3225 | va_end(args); | |
3226 | va_end(args1); | |
1da177e4 | 3227 | |
6370a6ad TH |
3228 | /* |
3229 | * Workqueues which may be used during memory reclaim should | |
3230 | * have a rescuer to guarantee forward progress. | |
3231 | */ | |
3232 | if (flags & WQ_MEM_RECLAIM) | |
3233 | flags |= WQ_RESCUER; | |
3234 | ||
d320c038 | 3235 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 3236 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 3237 | |
b196be89 | 3238 | /* init wq */ |
97e37d7b | 3239 | wq->flags = flags; |
a0a1a5fd | 3240 | wq->saved_max_active = max_active; |
73f53c4a TH |
3241 | mutex_init(&wq->flush_mutex); |
3242 | atomic_set(&wq->nr_cwqs_to_flush, 0); | |
3243 | INIT_LIST_HEAD(&wq->flusher_queue); | |
3244 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
502ca9d8 | 3245 | |
eb13ba87 | 3246 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 3247 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 3248 | |
bdbc5dd7 TH |
3249 | if (alloc_cwqs(wq) < 0) |
3250 | goto err; | |
3251 | ||
f3421797 | 3252 | for_each_cwq_cpu(cpu, wq) { |
1537663f | 3253 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
8b03ae3c | 3254 | struct global_cwq *gcwq = get_gcwq(cpu); |
3270476a | 3255 | int pool_idx = (bool)(flags & WQ_HIGHPRI); |
1537663f | 3256 | |
0f900049 | 3257 | BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK); |
3270476a | 3258 | cwq->pool = &gcwq->pools[pool_idx]; |
c34056a3 | 3259 | cwq->wq = wq; |
73f53c4a | 3260 | cwq->flush_color = -1; |
1e19ffc6 | 3261 | cwq->max_active = max_active; |
1e19ffc6 | 3262 | INIT_LIST_HEAD(&cwq->delayed_works); |
e22bee78 | 3263 | } |
1537663f | 3264 | |
e22bee78 TH |
3265 | if (flags & WQ_RESCUER) { |
3266 | struct worker *rescuer; | |
3267 | ||
f2e005aa | 3268 | if (!alloc_mayday_mask(&wq->mayday_mask, GFP_KERNEL)) |
e22bee78 TH |
3269 | goto err; |
3270 | ||
3271 | wq->rescuer = rescuer = alloc_worker(); | |
3272 | if (!rescuer) | |
3273 | goto err; | |
3274 | ||
111c225a TH |
3275 | rescuer->rescue_wq = wq; |
3276 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", | |
b196be89 | 3277 | wq->name); |
e22bee78 TH |
3278 | if (IS_ERR(rescuer->task)) |
3279 | goto err; | |
3280 | ||
e22bee78 TH |
3281 | rescuer->task->flags |= PF_THREAD_BOUND; |
3282 | wake_up_process(rescuer->task); | |
3af24433 ON |
3283 | } |
3284 | ||
a0a1a5fd TH |
3285 | /* |
3286 | * workqueue_lock protects global freeze state and workqueues | |
3287 | * list. Grab it, set max_active accordingly and add the new | |
3288 | * workqueue to workqueues list. | |
3289 | */ | |
1537663f | 3290 | spin_lock(&workqueue_lock); |
a0a1a5fd | 3291 | |
58a69cb4 | 3292 | if (workqueue_freezing && wq->flags & WQ_FREEZABLE) |
f3421797 | 3293 | for_each_cwq_cpu(cpu, wq) |
a0a1a5fd TH |
3294 | get_cwq(cpu, wq)->max_active = 0; |
3295 | ||
1537663f | 3296 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 3297 | |
1537663f TH |
3298 | spin_unlock(&workqueue_lock); |
3299 | ||
3af24433 | 3300 | return wq; |
4690c4ab TH |
3301 | err: |
3302 | if (wq) { | |
bdbc5dd7 | 3303 | free_cwqs(wq); |
f2e005aa | 3304 | free_mayday_mask(wq->mayday_mask); |
e22bee78 | 3305 | kfree(wq->rescuer); |
4690c4ab TH |
3306 | kfree(wq); |
3307 | } | |
3308 | return NULL; | |
3af24433 | 3309 | } |
d320c038 | 3310 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 3311 | |
3af24433 ON |
3312 | /** |
3313 | * destroy_workqueue - safely terminate a workqueue | |
3314 | * @wq: target workqueue | |
3315 | * | |
3316 | * Safely destroy a workqueue. All work currently pending will be done first. | |
3317 | */ | |
3318 | void destroy_workqueue(struct workqueue_struct *wq) | |
3319 | { | |
c8e55f36 | 3320 | unsigned int cpu; |
3af24433 | 3321 | |
9c5a2ba7 TH |
3322 | /* drain it before proceeding with destruction */ |
3323 | drain_workqueue(wq); | |
c8efcc25 | 3324 | |
a0a1a5fd TH |
3325 | /* |
3326 | * wq list is used to freeze wq, remove from list after | |
3327 | * flushing is complete in case freeze races us. | |
3328 | */ | |
95402b38 | 3329 | spin_lock(&workqueue_lock); |
b1f4ec17 | 3330 | list_del(&wq->list); |
95402b38 | 3331 | spin_unlock(&workqueue_lock); |
3af24433 | 3332 | |
e22bee78 | 3333 | /* sanity check */ |
f3421797 | 3334 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a TH |
3335 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3336 | int i; | |
3337 | ||
73f53c4a TH |
3338 | for (i = 0; i < WORK_NR_COLORS; i++) |
3339 | BUG_ON(cwq->nr_in_flight[i]); | |
1e19ffc6 TH |
3340 | BUG_ON(cwq->nr_active); |
3341 | BUG_ON(!list_empty(&cwq->delayed_works)); | |
73f53c4a | 3342 | } |
9b41ea72 | 3343 | |
e22bee78 TH |
3344 | if (wq->flags & WQ_RESCUER) { |
3345 | kthread_stop(wq->rescuer->task); | |
f2e005aa | 3346 | free_mayday_mask(wq->mayday_mask); |
8d9df9f0 | 3347 | kfree(wq->rescuer); |
e22bee78 TH |
3348 | } |
3349 | ||
bdbc5dd7 | 3350 | free_cwqs(wq); |
3af24433 ON |
3351 | kfree(wq); |
3352 | } | |
3353 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
3354 | ||
9f4bd4cd LJ |
3355 | /** |
3356 | * cwq_set_max_active - adjust max_active of a cwq | |
3357 | * @cwq: target cpu_workqueue_struct | |
3358 | * @max_active: new max_active value. | |
3359 | * | |
3360 | * Set @cwq->max_active to @max_active and activate delayed works if | |
3361 | * increased. | |
3362 | * | |
3363 | * CONTEXT: | |
3364 | * spin_lock_irq(gcwq->lock). | |
3365 | */ | |
3366 | static void cwq_set_max_active(struct cpu_workqueue_struct *cwq, int max_active) | |
3367 | { | |
3368 | cwq->max_active = max_active; | |
3369 | ||
3370 | while (!list_empty(&cwq->delayed_works) && | |
3371 | cwq->nr_active < cwq->max_active) | |
3372 | cwq_activate_first_delayed(cwq); | |
3373 | } | |
3374 | ||
dcd989cb TH |
3375 | /** |
3376 | * workqueue_set_max_active - adjust max_active of a workqueue | |
3377 | * @wq: target workqueue | |
3378 | * @max_active: new max_active value. | |
3379 | * | |
3380 | * Set max_active of @wq to @max_active. | |
3381 | * | |
3382 | * CONTEXT: | |
3383 | * Don't call from IRQ context. | |
3384 | */ | |
3385 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
3386 | { | |
3387 | unsigned int cpu; | |
3388 | ||
f3421797 | 3389 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb TH |
3390 | |
3391 | spin_lock(&workqueue_lock); | |
3392 | ||
3393 | wq->saved_max_active = max_active; | |
3394 | ||
f3421797 | 3395 | for_each_cwq_cpu(cpu, wq) { |
35b6bb63 TH |
3396 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3397 | struct worker_pool *pool = cwq->pool; | |
3398 | struct global_cwq *gcwq = pool->gcwq; | |
dcd989cb TH |
3399 | |
3400 | spin_lock_irq(&gcwq->lock); | |
3401 | ||
58a69cb4 | 3402 | if (!(wq->flags & WQ_FREEZABLE) || |
35b6bb63 TH |
3403 | !(pool->flags & POOL_FREEZING)) |
3404 | cwq_set_max_active(cwq, max_active); | |
9bfb1839 | 3405 | |
dcd989cb | 3406 | spin_unlock_irq(&gcwq->lock); |
65a64464 | 3407 | } |
93981800 | 3408 | |
dcd989cb | 3409 | spin_unlock(&workqueue_lock); |
15316ba8 | 3410 | } |
dcd989cb | 3411 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 3412 | |
eef6a7d5 | 3413 | /** |
dcd989cb TH |
3414 | * workqueue_congested - test whether a workqueue is congested |
3415 | * @cpu: CPU in question | |
3416 | * @wq: target workqueue | |
eef6a7d5 | 3417 | * |
dcd989cb TH |
3418 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
3419 | * no synchronization around this function and the test result is | |
3420 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 3421 | * |
dcd989cb TH |
3422 | * RETURNS: |
3423 | * %true if congested, %false otherwise. | |
eef6a7d5 | 3424 | */ |
dcd989cb | 3425 | bool workqueue_congested(unsigned int cpu, struct workqueue_struct *wq) |
1da177e4 | 3426 | { |
dcd989cb TH |
3427 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3428 | ||
3429 | return !list_empty(&cwq->delayed_works); | |
1da177e4 | 3430 | } |
dcd989cb | 3431 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 3432 | |
1fa44eca | 3433 | /** |
dcd989cb TH |
3434 | * work_cpu - return the last known associated cpu for @work |
3435 | * @work: the work of interest | |
1fa44eca | 3436 | * |
dcd989cb | 3437 | * RETURNS: |
bdbc5dd7 | 3438 | * CPU number if @work was ever queued. WORK_CPU_NONE otherwise. |
1fa44eca | 3439 | */ |
e2905b29 | 3440 | static unsigned int work_cpu(struct work_struct *work) |
1fa44eca | 3441 | { |
dcd989cb | 3442 | struct global_cwq *gcwq = get_work_gcwq(work); |
1fa44eca | 3443 | |
bdbc5dd7 | 3444 | return gcwq ? gcwq->cpu : WORK_CPU_NONE; |
1fa44eca | 3445 | } |
1fa44eca | 3446 | |
dcd989cb TH |
3447 | /** |
3448 | * work_busy - test whether a work is currently pending or running | |
3449 | * @work: the work to be tested | |
3450 | * | |
3451 | * Test whether @work is currently pending or running. There is no | |
3452 | * synchronization around this function and the test result is | |
3453 | * unreliable and only useful as advisory hints or for debugging. | |
3454 | * Especially for reentrant wqs, the pending state might hide the | |
3455 | * running state. | |
3456 | * | |
3457 | * RETURNS: | |
3458 | * OR'd bitmask of WORK_BUSY_* bits. | |
3459 | */ | |
3460 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 3461 | { |
dcd989cb TH |
3462 | struct global_cwq *gcwq = get_work_gcwq(work); |
3463 | unsigned long flags; | |
3464 | unsigned int ret = 0; | |
1da177e4 | 3465 | |
dcd989cb | 3466 | if (!gcwq) |
999767be | 3467 | return 0; |
1da177e4 | 3468 | |
dcd989cb | 3469 | spin_lock_irqsave(&gcwq->lock, flags); |
1da177e4 | 3470 | |
dcd989cb TH |
3471 | if (work_pending(work)) |
3472 | ret |= WORK_BUSY_PENDING; | |
3473 | if (find_worker_executing_work(gcwq, work)) | |
3474 | ret |= WORK_BUSY_RUNNING; | |
1da177e4 | 3475 | |
dcd989cb | 3476 | spin_unlock_irqrestore(&gcwq->lock, flags); |
1da177e4 | 3477 | |
dcd989cb | 3478 | return ret; |
1da177e4 | 3479 | } |
dcd989cb | 3480 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 3481 | |
db7bccf4 TH |
3482 | /* |
3483 | * CPU hotplug. | |
3484 | * | |
e22bee78 TH |
3485 | * There are two challenges in supporting CPU hotplug. Firstly, there |
3486 | * are a lot of assumptions on strong associations among work, cwq and | |
3487 | * gcwq which make migrating pending and scheduled works very | |
3488 | * difficult to implement without impacting hot paths. Secondly, | |
3489 | * gcwqs serve mix of short, long and very long running works making | |
3490 | * blocked draining impractical. | |
3491 | * | |
24647570 | 3492 | * This is solved by allowing the pools to be disassociated from the CPU |
628c78e7 TH |
3493 | * running as an unbound one and allowing it to be reattached later if the |
3494 | * cpu comes back online. | |
db7bccf4 | 3495 | */ |
1da177e4 | 3496 | |
60373152 | 3497 | /* claim manager positions of all pools */ |
b2eb83d1 | 3498 | static void gcwq_claim_assoc_and_lock(struct global_cwq *gcwq) |
60373152 TH |
3499 | { |
3500 | struct worker_pool *pool; | |
3501 | ||
3502 | for_each_worker_pool(pool, gcwq) | |
b2eb83d1 | 3503 | mutex_lock_nested(&pool->assoc_mutex, pool - gcwq->pools); |
8db25e78 | 3504 | spin_lock_irq(&gcwq->lock); |
60373152 TH |
3505 | } |
3506 | ||
3507 | /* release manager positions */ | |
b2eb83d1 | 3508 | static void gcwq_release_assoc_and_unlock(struct global_cwq *gcwq) |
60373152 TH |
3509 | { |
3510 | struct worker_pool *pool; | |
3511 | ||
8db25e78 | 3512 | spin_unlock_irq(&gcwq->lock); |
60373152 | 3513 | for_each_worker_pool(pool, gcwq) |
b2eb83d1 | 3514 | mutex_unlock(&pool->assoc_mutex); |
60373152 TH |
3515 | } |
3516 | ||
628c78e7 | 3517 | static void gcwq_unbind_fn(struct work_struct *work) |
3af24433 | 3518 | { |
628c78e7 | 3519 | struct global_cwq *gcwq = get_gcwq(smp_processor_id()); |
4ce62e9e | 3520 | struct worker_pool *pool; |
db7bccf4 TH |
3521 | struct worker *worker; |
3522 | struct hlist_node *pos; | |
3523 | int i; | |
3af24433 | 3524 | |
db7bccf4 TH |
3525 | BUG_ON(gcwq->cpu != smp_processor_id()); |
3526 | ||
b2eb83d1 | 3527 | gcwq_claim_assoc_and_lock(gcwq); |
3af24433 | 3528 | |
f2d5a0ee TH |
3529 | /* |
3530 | * We've claimed all manager positions. Make all workers unbound | |
3531 | * and set DISASSOCIATED. Before this, all workers except for the | |
3532 | * ones which are still executing works from before the last CPU | |
3533 | * down must be on the cpu. After this, they may become diasporas. | |
3534 | */ | |
60373152 | 3535 | for_each_worker_pool(pool, gcwq) |
4ce62e9e | 3536 | list_for_each_entry(worker, &pool->idle_list, entry) |
403c821d | 3537 | worker->flags |= WORKER_UNBOUND; |
3af24433 | 3538 | |
db7bccf4 | 3539 | for_each_busy_worker(worker, i, pos, gcwq) |
403c821d | 3540 | worker->flags |= WORKER_UNBOUND; |
06ba38a9 | 3541 | |
24647570 TH |
3542 | for_each_worker_pool(pool, gcwq) |
3543 | pool->flags |= POOL_DISASSOCIATED; | |
f2d5a0ee | 3544 | |
b2eb83d1 | 3545 | gcwq_release_assoc_and_unlock(gcwq); |
628c78e7 | 3546 | |
e22bee78 | 3547 | /* |
403c821d | 3548 | * Call schedule() so that we cross rq->lock and thus can guarantee |
628c78e7 TH |
3549 | * sched callbacks see the %WORKER_UNBOUND flag. This is necessary |
3550 | * as scheduler callbacks may be invoked from other cpus. | |
e22bee78 | 3551 | */ |
e22bee78 | 3552 | schedule(); |
06ba38a9 | 3553 | |
e22bee78 | 3554 | /* |
628c78e7 TH |
3555 | * Sched callbacks are disabled now. Zap nr_running. After this, |
3556 | * nr_running stays zero and need_more_worker() and keep_working() | |
3557 | * are always true as long as the worklist is not empty. @gcwq now | |
3558 | * behaves as unbound (in terms of concurrency management) gcwq | |
3559 | * which is served by workers tied to the CPU. | |
3560 | * | |
3561 | * On return from this function, the current worker would trigger | |
3562 | * unbound chain execution of pending work items if other workers | |
3563 | * didn't already. | |
e22bee78 | 3564 | */ |
4ce62e9e TH |
3565 | for_each_worker_pool(pool, gcwq) |
3566 | atomic_set(get_pool_nr_running(pool), 0); | |
3af24433 | 3567 | } |
3af24433 | 3568 | |
8db25e78 TH |
3569 | /* |
3570 | * Workqueues should be brought up before normal priority CPU notifiers. | |
3571 | * This will be registered high priority CPU notifier. | |
3572 | */ | |
9fdf9b73 | 3573 | static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb, |
8db25e78 TH |
3574 | unsigned long action, |
3575 | void *hcpu) | |
3af24433 ON |
3576 | { |
3577 | unsigned int cpu = (unsigned long)hcpu; | |
db7bccf4 | 3578 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3579 | struct worker_pool *pool; |
3ce63377 | 3580 | |
8db25e78 | 3581 | switch (action & ~CPU_TASKS_FROZEN) { |
3af24433 | 3582 | case CPU_UP_PREPARE: |
4ce62e9e | 3583 | for_each_worker_pool(pool, gcwq) { |
3ce63377 TH |
3584 | struct worker *worker; |
3585 | ||
3586 | if (pool->nr_workers) | |
3587 | continue; | |
3588 | ||
3589 | worker = create_worker(pool); | |
3590 | if (!worker) | |
3591 | return NOTIFY_BAD; | |
3592 | ||
3593 | spin_lock_irq(&gcwq->lock); | |
3594 | start_worker(worker); | |
3595 | spin_unlock_irq(&gcwq->lock); | |
3af24433 | 3596 | } |
8db25e78 | 3597 | break; |
3af24433 | 3598 | |
db7bccf4 TH |
3599 | case CPU_DOWN_FAILED: |
3600 | case CPU_ONLINE: | |
b2eb83d1 | 3601 | gcwq_claim_assoc_and_lock(gcwq); |
24647570 TH |
3602 | for_each_worker_pool(pool, gcwq) |
3603 | pool->flags &= ~POOL_DISASSOCIATED; | |
25511a47 | 3604 | rebind_workers(gcwq); |
b2eb83d1 | 3605 | gcwq_release_assoc_and_unlock(gcwq); |
db7bccf4 | 3606 | break; |
00dfcaf7 | 3607 | } |
65758202 TH |
3608 | return NOTIFY_OK; |
3609 | } | |
3610 | ||
3611 | /* | |
3612 | * Workqueues should be brought down after normal priority CPU notifiers. | |
3613 | * This will be registered as low priority CPU notifier. | |
3614 | */ | |
9fdf9b73 | 3615 | static int __cpuinit workqueue_cpu_down_callback(struct notifier_block *nfb, |
65758202 TH |
3616 | unsigned long action, |
3617 | void *hcpu) | |
3618 | { | |
8db25e78 TH |
3619 | unsigned int cpu = (unsigned long)hcpu; |
3620 | struct work_struct unbind_work; | |
3621 | ||
65758202 TH |
3622 | switch (action & ~CPU_TASKS_FROZEN) { |
3623 | case CPU_DOWN_PREPARE: | |
8db25e78 TH |
3624 | /* unbinding should happen on the local CPU */ |
3625 | INIT_WORK_ONSTACK(&unbind_work, gcwq_unbind_fn); | |
7635d2fd | 3626 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
8db25e78 TH |
3627 | flush_work(&unbind_work); |
3628 | break; | |
65758202 TH |
3629 | } |
3630 | return NOTIFY_OK; | |
3631 | } | |
3632 | ||
2d3854a3 | 3633 | #ifdef CONFIG_SMP |
8ccad40d | 3634 | |
2d3854a3 | 3635 | struct work_for_cpu { |
ed48ece2 | 3636 | struct work_struct work; |
2d3854a3 RR |
3637 | long (*fn)(void *); |
3638 | void *arg; | |
3639 | long ret; | |
3640 | }; | |
3641 | ||
ed48ece2 | 3642 | static void work_for_cpu_fn(struct work_struct *work) |
2d3854a3 | 3643 | { |
ed48ece2 TH |
3644 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
3645 | ||
2d3854a3 RR |
3646 | wfc->ret = wfc->fn(wfc->arg); |
3647 | } | |
3648 | ||
3649 | /** | |
3650 | * work_on_cpu - run a function in user context on a particular cpu | |
3651 | * @cpu: the cpu to run on | |
3652 | * @fn: the function to run | |
3653 | * @arg: the function arg | |
3654 | * | |
31ad9081 RR |
3655 | * This will return the value @fn returns. |
3656 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 3657 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 RR |
3658 | */ |
3659 | long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) | |
3660 | { | |
ed48ece2 | 3661 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; |
6b44003e | 3662 | |
ed48ece2 TH |
3663 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
3664 | schedule_work_on(cpu, &wfc.work); | |
3665 | flush_work(&wfc.work); | |
2d3854a3 RR |
3666 | return wfc.ret; |
3667 | } | |
3668 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
3669 | #endif /* CONFIG_SMP */ | |
3670 | ||
a0a1a5fd TH |
3671 | #ifdef CONFIG_FREEZER |
3672 | ||
3673 | /** | |
3674 | * freeze_workqueues_begin - begin freezing workqueues | |
3675 | * | |
58a69cb4 TH |
3676 | * Start freezing workqueues. After this function returns, all freezable |
3677 | * workqueues will queue new works to their frozen_works list instead of | |
3678 | * gcwq->worklist. | |
a0a1a5fd TH |
3679 | * |
3680 | * CONTEXT: | |
8b03ae3c | 3681 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3682 | */ |
3683 | void freeze_workqueues_begin(void) | |
3684 | { | |
a0a1a5fd TH |
3685 | unsigned int cpu; |
3686 | ||
3687 | spin_lock(&workqueue_lock); | |
3688 | ||
3689 | BUG_ON(workqueue_freezing); | |
3690 | workqueue_freezing = true; | |
3691 | ||
f3421797 | 3692 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3693 | struct global_cwq *gcwq = get_gcwq(cpu); |
35b6bb63 | 3694 | struct worker_pool *pool; |
bdbc5dd7 | 3695 | struct workqueue_struct *wq; |
8b03ae3c TH |
3696 | |
3697 | spin_lock_irq(&gcwq->lock); | |
3698 | ||
35b6bb63 TH |
3699 | for_each_worker_pool(pool, gcwq) { |
3700 | WARN_ON_ONCE(pool->flags & POOL_FREEZING); | |
3701 | pool->flags |= POOL_FREEZING; | |
3702 | } | |
db7bccf4 | 3703 | |
a0a1a5fd TH |
3704 | list_for_each_entry(wq, &workqueues, list) { |
3705 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3706 | ||
58a69cb4 | 3707 | if (cwq && wq->flags & WQ_FREEZABLE) |
a0a1a5fd | 3708 | cwq->max_active = 0; |
a0a1a5fd | 3709 | } |
8b03ae3c TH |
3710 | |
3711 | spin_unlock_irq(&gcwq->lock); | |
a0a1a5fd TH |
3712 | } |
3713 | ||
3714 | spin_unlock(&workqueue_lock); | |
3715 | } | |
3716 | ||
3717 | /** | |
58a69cb4 | 3718 | * freeze_workqueues_busy - are freezable workqueues still busy? |
a0a1a5fd TH |
3719 | * |
3720 | * Check whether freezing is complete. This function must be called | |
3721 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
3722 | * | |
3723 | * CONTEXT: | |
3724 | * Grabs and releases workqueue_lock. | |
3725 | * | |
3726 | * RETURNS: | |
58a69cb4 TH |
3727 | * %true if some freezable workqueues are still busy. %false if freezing |
3728 | * is complete. | |
a0a1a5fd TH |
3729 | */ |
3730 | bool freeze_workqueues_busy(void) | |
3731 | { | |
a0a1a5fd TH |
3732 | unsigned int cpu; |
3733 | bool busy = false; | |
3734 | ||
3735 | spin_lock(&workqueue_lock); | |
3736 | ||
3737 | BUG_ON(!workqueue_freezing); | |
3738 | ||
f3421797 | 3739 | for_each_gcwq_cpu(cpu) { |
bdbc5dd7 | 3740 | struct workqueue_struct *wq; |
a0a1a5fd TH |
3741 | /* |
3742 | * nr_active is monotonically decreasing. It's safe | |
3743 | * to peek without lock. | |
3744 | */ | |
3745 | list_for_each_entry(wq, &workqueues, list) { | |
3746 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3747 | ||
58a69cb4 | 3748 | if (!cwq || !(wq->flags & WQ_FREEZABLE)) |
a0a1a5fd TH |
3749 | continue; |
3750 | ||
3751 | BUG_ON(cwq->nr_active < 0); | |
3752 | if (cwq->nr_active) { | |
3753 | busy = true; | |
3754 | goto out_unlock; | |
3755 | } | |
3756 | } | |
3757 | } | |
3758 | out_unlock: | |
3759 | spin_unlock(&workqueue_lock); | |
3760 | return busy; | |
3761 | } | |
3762 | ||
3763 | /** | |
3764 | * thaw_workqueues - thaw workqueues | |
3765 | * | |
3766 | * Thaw workqueues. Normal queueing is restored and all collected | |
7e11629d | 3767 | * frozen works are transferred to their respective gcwq worklists. |
a0a1a5fd TH |
3768 | * |
3769 | * CONTEXT: | |
8b03ae3c | 3770 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3771 | */ |
3772 | void thaw_workqueues(void) | |
3773 | { | |
a0a1a5fd TH |
3774 | unsigned int cpu; |
3775 | ||
3776 | spin_lock(&workqueue_lock); | |
3777 | ||
3778 | if (!workqueue_freezing) | |
3779 | goto out_unlock; | |
3780 | ||
f3421797 | 3781 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3782 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3783 | struct worker_pool *pool; |
bdbc5dd7 | 3784 | struct workqueue_struct *wq; |
8b03ae3c TH |
3785 | |
3786 | spin_lock_irq(&gcwq->lock); | |
3787 | ||
35b6bb63 TH |
3788 | for_each_worker_pool(pool, gcwq) { |
3789 | WARN_ON_ONCE(!(pool->flags & POOL_FREEZING)); | |
3790 | pool->flags &= ~POOL_FREEZING; | |
3791 | } | |
db7bccf4 | 3792 | |
a0a1a5fd TH |
3793 | list_for_each_entry(wq, &workqueues, list) { |
3794 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3795 | ||
58a69cb4 | 3796 | if (!cwq || !(wq->flags & WQ_FREEZABLE)) |
a0a1a5fd TH |
3797 | continue; |
3798 | ||
a0a1a5fd | 3799 | /* restore max_active and repopulate worklist */ |
9f4bd4cd | 3800 | cwq_set_max_active(cwq, wq->saved_max_active); |
a0a1a5fd | 3801 | } |
8b03ae3c | 3802 | |
4ce62e9e TH |
3803 | for_each_worker_pool(pool, gcwq) |
3804 | wake_up_worker(pool); | |
e22bee78 | 3805 | |
8b03ae3c | 3806 | spin_unlock_irq(&gcwq->lock); |
a0a1a5fd TH |
3807 | } |
3808 | ||
3809 | workqueue_freezing = false; | |
3810 | out_unlock: | |
3811 | spin_unlock(&workqueue_lock); | |
3812 | } | |
3813 | #endif /* CONFIG_FREEZER */ | |
3814 | ||
6ee0578b | 3815 | static int __init init_workqueues(void) |
1da177e4 | 3816 | { |
c34056a3 TH |
3817 | unsigned int cpu; |
3818 | ||
b5490077 TH |
3819 | /* make sure we have enough bits for OFFQ CPU number */ |
3820 | BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_CPU_SHIFT)) < | |
3821 | WORK_CPU_LAST); | |
3822 | ||
65758202 | 3823 | cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); |
a5b4e57d | 3824 | hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); |
8b03ae3c TH |
3825 | |
3826 | /* initialize gcwqs */ | |
f3421797 | 3827 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3828 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3829 | struct worker_pool *pool; |
8b03ae3c TH |
3830 | |
3831 | spin_lock_init(&gcwq->lock); | |
3832 | gcwq->cpu = cpu; | |
3833 | ||
42f8570f | 3834 | hash_init(gcwq->busy_hash); |
c8e55f36 | 3835 | |
4ce62e9e TH |
3836 | for_each_worker_pool(pool, gcwq) { |
3837 | pool->gcwq = gcwq; | |
24647570 | 3838 | pool->flags |= POOL_DISASSOCIATED; |
4ce62e9e TH |
3839 | INIT_LIST_HEAD(&pool->worklist); |
3840 | INIT_LIST_HEAD(&pool->idle_list); | |
e7577c50 | 3841 | |
4ce62e9e TH |
3842 | init_timer_deferrable(&pool->idle_timer); |
3843 | pool->idle_timer.function = idle_worker_timeout; | |
3844 | pool->idle_timer.data = (unsigned long)pool; | |
e22bee78 | 3845 | |
4ce62e9e TH |
3846 | setup_timer(&pool->mayday_timer, gcwq_mayday_timeout, |
3847 | (unsigned long)pool); | |
3848 | ||
b2eb83d1 | 3849 | mutex_init(&pool->assoc_mutex); |
4ce62e9e | 3850 | ida_init(&pool->worker_ida); |
9daf9e67 TH |
3851 | |
3852 | /* alloc pool ID */ | |
3853 | BUG_ON(worker_pool_assign_id(pool)); | |
4ce62e9e | 3854 | } |
8b03ae3c TH |
3855 | } |
3856 | ||
e22bee78 | 3857 | /* create the initial worker */ |
f3421797 | 3858 | for_each_online_gcwq_cpu(cpu) { |
e22bee78 | 3859 | struct global_cwq *gcwq = get_gcwq(cpu); |
4ce62e9e | 3860 | struct worker_pool *pool; |
e22bee78 | 3861 | |
4ce62e9e TH |
3862 | for_each_worker_pool(pool, gcwq) { |
3863 | struct worker *worker; | |
3864 | ||
24647570 TH |
3865 | if (cpu != WORK_CPU_UNBOUND) |
3866 | pool->flags &= ~POOL_DISASSOCIATED; | |
3867 | ||
bc2ae0f5 | 3868 | worker = create_worker(pool); |
4ce62e9e TH |
3869 | BUG_ON(!worker); |
3870 | spin_lock_irq(&gcwq->lock); | |
3871 | start_worker(worker); | |
3872 | spin_unlock_irq(&gcwq->lock); | |
3873 | } | |
e22bee78 TH |
3874 | } |
3875 | ||
d320c038 | 3876 | system_wq = alloc_workqueue("events", 0, 0); |
1aabe902 | 3877 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
d320c038 | 3878 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
f3421797 TH |
3879 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
3880 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
3881 | system_freezable_wq = alloc_workqueue("events_freezable", |
3882 | WQ_FREEZABLE, 0); | |
1aabe902 | 3883 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
ae930e0f | 3884 | !system_unbound_wq || !system_freezable_wq); |
6ee0578b | 3885 | return 0; |
1da177e4 | 3886 | } |
6ee0578b | 3887 | early_initcall(init_workqueues); |