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> |
e22bee78 TH |
44 | |
45 | #include "workqueue_sched.h" | |
1da177e4 | 46 | |
c8e55f36 | 47 | enum { |
db7bccf4 | 48 | /* global_cwq flags */ |
e22bee78 TH |
49 | GCWQ_MANAGE_WORKERS = 1 << 0, /* need to manage workers */ |
50 | GCWQ_MANAGING_WORKERS = 1 << 1, /* managing workers */ | |
51 | GCWQ_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ | |
db7bccf4 | 52 | GCWQ_FREEZING = 1 << 3, /* freeze in progress */ |
649027d7 | 53 | GCWQ_HIGHPRI_PENDING = 1 << 4, /* highpri works on queue */ |
db7bccf4 | 54 | |
c8e55f36 TH |
55 | /* worker flags */ |
56 | WORKER_STARTED = 1 << 0, /* started */ | |
57 | WORKER_DIE = 1 << 1, /* die die die */ | |
58 | WORKER_IDLE = 1 << 2, /* is idle */ | |
e22bee78 | 59 | WORKER_PREP = 1 << 3, /* preparing to run works */ |
db7bccf4 | 60 | WORKER_ROGUE = 1 << 4, /* not bound to any cpu */ |
e22bee78 | 61 | WORKER_REBIND = 1 << 5, /* mom is home, come back */ |
fb0e7beb | 62 | WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ |
f3421797 | 63 | WORKER_UNBOUND = 1 << 7, /* worker is unbound */ |
e22bee78 | 64 | |
fb0e7beb | 65 | WORKER_NOT_RUNNING = WORKER_PREP | WORKER_ROGUE | WORKER_REBIND | |
f3421797 | 66 | WORKER_CPU_INTENSIVE | WORKER_UNBOUND, |
db7bccf4 TH |
67 | |
68 | /* gcwq->trustee_state */ | |
69 | TRUSTEE_START = 0, /* start */ | |
70 | TRUSTEE_IN_CHARGE = 1, /* trustee in charge of gcwq */ | |
71 | TRUSTEE_BUTCHER = 2, /* butcher workers */ | |
72 | TRUSTEE_RELEASE = 3, /* release workers */ | |
73 | TRUSTEE_DONE = 4, /* trustee is done */ | |
c8e55f36 TH |
74 | |
75 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ | |
76 | BUSY_WORKER_HASH_SIZE = 1 << BUSY_WORKER_HASH_ORDER, | |
77 | BUSY_WORKER_HASH_MASK = BUSY_WORKER_HASH_SIZE - 1, | |
db7bccf4 | 78 | |
e22bee78 TH |
79 | MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */ |
80 | IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */ | |
81 | ||
3233cdbd TH |
82 | MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2, |
83 | /* call for help after 10ms | |
84 | (min two ticks) */ | |
e22bee78 TH |
85 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ |
86 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
db7bccf4 | 87 | TRUSTEE_COOLDOWN = HZ / 10, /* for trustee draining */ |
e22bee78 TH |
88 | |
89 | /* | |
90 | * Rescue workers are used only on emergencies and shared by | |
91 | * all cpus. Give -20. | |
92 | */ | |
93 | RESCUER_NICE_LEVEL = -20, | |
c8e55f36 | 94 | }; |
1da177e4 LT |
95 | |
96 | /* | |
4690c4ab TH |
97 | * Structure fields follow one of the following exclusion rules. |
98 | * | |
e41e704b TH |
99 | * I: Modifiable by initialization/destruction paths and read-only for |
100 | * everyone else. | |
4690c4ab | 101 | * |
e22bee78 TH |
102 | * P: Preemption protected. Disabling preemption is enough and should |
103 | * only be modified and accessed from the local cpu. | |
104 | * | |
8b03ae3c | 105 | * L: gcwq->lock protected. Access with gcwq->lock held. |
4690c4ab | 106 | * |
e22bee78 TH |
107 | * X: During normal operation, modification requires gcwq->lock and |
108 | * should be done only from local cpu. Either disabling preemption | |
109 | * on local cpu or grabbing gcwq->lock is enough for read access. | |
f3421797 | 110 | * If GCWQ_DISASSOCIATED is set, it's identical to L. |
e22bee78 | 111 | * |
73f53c4a TH |
112 | * F: wq->flush_mutex protected. |
113 | * | |
4690c4ab | 114 | * W: workqueue_lock protected. |
1da177e4 | 115 | */ |
1da177e4 | 116 | |
8b03ae3c | 117 | struct global_cwq; |
bd7bdd43 | 118 | struct worker_pool; |
1da177e4 | 119 | |
e22bee78 TH |
120 | /* |
121 | * The poor guys doing the actual heavy lifting. All on-duty workers | |
122 | * are either serving the manager role, on idle list or on busy hash. | |
123 | */ | |
c34056a3 | 124 | struct worker { |
c8e55f36 TH |
125 | /* on idle list while idle, on busy hash table while busy */ |
126 | union { | |
127 | struct list_head entry; /* L: while idle */ | |
128 | struct hlist_node hentry; /* L: while busy */ | |
129 | }; | |
1da177e4 | 130 | |
c34056a3 | 131 | struct work_struct *current_work; /* L: work being processed */ |
8cca0eea | 132 | struct cpu_workqueue_struct *current_cwq; /* L: current_work's cwq */ |
affee4b2 | 133 | struct list_head scheduled; /* L: scheduled works */ |
c34056a3 | 134 | struct task_struct *task; /* I: worker task */ |
bd7bdd43 | 135 | struct worker_pool *pool; /* I: the associated pool */ |
e22bee78 TH |
136 | /* 64 bytes boundary on 64bit, 32 on 32bit */ |
137 | unsigned long last_active; /* L: last active timestamp */ | |
138 | unsigned int flags; /* X: flags */ | |
c34056a3 | 139 | int id; /* I: worker id */ |
e22bee78 | 140 | struct work_struct rebind_work; /* L: rebind worker to cpu */ |
c34056a3 TH |
141 | }; |
142 | ||
bd7bdd43 TH |
143 | struct worker_pool { |
144 | struct global_cwq *gcwq; /* I: the owning gcwq */ | |
145 | ||
146 | struct list_head worklist; /* L: list of pending works */ | |
147 | int nr_workers; /* L: total number of workers */ | |
148 | int nr_idle; /* L: currently idle ones */ | |
149 | ||
150 | struct list_head idle_list; /* X: list of idle workers */ | |
151 | struct timer_list idle_timer; /* L: worker idle timeout */ | |
152 | struct timer_list mayday_timer; /* L: SOS timer for workers */ | |
153 | ||
154 | struct ida worker_ida; /* L: for worker IDs */ | |
155 | struct worker *first_idle; /* L: first idle worker */ | |
156 | }; | |
157 | ||
8b03ae3c | 158 | /* |
e22bee78 TH |
159 | * Global per-cpu workqueue. There's one and only one for each cpu |
160 | * and all works are queued and processed here regardless of their | |
161 | * target workqueues. | |
8b03ae3c TH |
162 | */ |
163 | struct global_cwq { | |
164 | spinlock_t lock; /* the gcwq lock */ | |
165 | unsigned int cpu; /* I: the associated cpu */ | |
db7bccf4 | 166 | unsigned int flags; /* L: GCWQ_* flags */ |
c8e55f36 | 167 | |
bd7bdd43 | 168 | /* workers are chained either in busy_hash or pool idle_list */ |
c8e55f36 TH |
169 | struct hlist_head busy_hash[BUSY_WORKER_HASH_SIZE]; |
170 | /* L: hash of busy workers */ | |
171 | ||
bd7bdd43 | 172 | struct worker_pool pool; /* the worker pools */ |
db7bccf4 TH |
173 | |
174 | struct task_struct *trustee; /* L: for gcwq shutdown */ | |
175 | unsigned int trustee_state; /* L: trustee state */ | |
176 | wait_queue_head_t trustee_wait; /* trustee wait */ | |
8b03ae3c TH |
177 | } ____cacheline_aligned_in_smp; |
178 | ||
1da177e4 | 179 | /* |
502ca9d8 | 180 | * The per-CPU workqueue. The lower WORK_STRUCT_FLAG_BITS of |
0f900049 TH |
181 | * work_struct->data are used for flags and thus cwqs need to be |
182 | * aligned at two's power of the number of flag bits. | |
1da177e4 LT |
183 | */ |
184 | struct cpu_workqueue_struct { | |
bd7bdd43 | 185 | struct worker_pool *pool; /* I: the associated pool */ |
4690c4ab | 186 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
187 | int work_color; /* L: current color */ |
188 | int flush_color; /* L: flushing color */ | |
189 | int nr_in_flight[WORK_NR_COLORS]; | |
190 | /* L: nr of in_flight works */ | |
1e19ffc6 | 191 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 192 | int max_active; /* L: max active works */ |
1e19ffc6 | 193 | struct list_head delayed_works; /* L: delayed works */ |
0f900049 | 194 | }; |
1da177e4 | 195 | |
73f53c4a TH |
196 | /* |
197 | * Structure used to wait for workqueue flush. | |
198 | */ | |
199 | struct wq_flusher { | |
200 | struct list_head list; /* F: list of flushers */ | |
201 | int flush_color; /* F: flush color waiting for */ | |
202 | struct completion done; /* flush completion */ | |
203 | }; | |
204 | ||
f2e005aa TH |
205 | /* |
206 | * All cpumasks are assumed to be always set on UP and thus can't be | |
207 | * used to determine whether there's something to be done. | |
208 | */ | |
209 | #ifdef CONFIG_SMP | |
210 | typedef cpumask_var_t mayday_mask_t; | |
211 | #define mayday_test_and_set_cpu(cpu, mask) \ | |
212 | cpumask_test_and_set_cpu((cpu), (mask)) | |
213 | #define mayday_clear_cpu(cpu, mask) cpumask_clear_cpu((cpu), (mask)) | |
214 | #define for_each_mayday_cpu(cpu, mask) for_each_cpu((cpu), (mask)) | |
9c37547a | 215 | #define alloc_mayday_mask(maskp, gfp) zalloc_cpumask_var((maskp), (gfp)) |
f2e005aa TH |
216 | #define free_mayday_mask(mask) free_cpumask_var((mask)) |
217 | #else | |
218 | typedef unsigned long mayday_mask_t; | |
219 | #define mayday_test_and_set_cpu(cpu, mask) test_and_set_bit(0, &(mask)) | |
220 | #define mayday_clear_cpu(cpu, mask) clear_bit(0, &(mask)) | |
221 | #define for_each_mayday_cpu(cpu, mask) if ((cpu) = 0, (mask)) | |
222 | #define alloc_mayday_mask(maskp, gfp) true | |
223 | #define free_mayday_mask(mask) do { } while (0) | |
224 | #endif | |
1da177e4 LT |
225 | |
226 | /* | |
227 | * The externally visible workqueue abstraction is an array of | |
228 | * per-CPU workqueues: | |
229 | */ | |
230 | struct workqueue_struct { | |
9c5a2ba7 | 231 | unsigned int flags; /* W: WQ_* flags */ |
bdbc5dd7 TH |
232 | union { |
233 | struct cpu_workqueue_struct __percpu *pcpu; | |
234 | struct cpu_workqueue_struct *single; | |
235 | unsigned long v; | |
236 | } cpu_wq; /* I: cwq's */ | |
4690c4ab | 237 | struct list_head list; /* W: list of all workqueues */ |
73f53c4a TH |
238 | |
239 | struct mutex flush_mutex; /* protects wq flushing */ | |
240 | int work_color; /* F: current work color */ | |
241 | int flush_color; /* F: current flush color */ | |
242 | atomic_t nr_cwqs_to_flush; /* flush in progress */ | |
243 | struct wq_flusher *first_flusher; /* F: first flusher */ | |
244 | struct list_head flusher_queue; /* F: flush waiters */ | |
245 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
246 | ||
f2e005aa | 247 | mayday_mask_t mayday_mask; /* cpus requesting rescue */ |
e22bee78 TH |
248 | struct worker *rescuer; /* I: rescue worker */ |
249 | ||
9c5a2ba7 | 250 | int nr_drainers; /* W: drain in progress */ |
dcd989cb | 251 | int saved_max_active; /* W: saved cwq max_active */ |
4e6045f1 | 252 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 253 | struct lockdep_map lockdep_map; |
4e6045f1 | 254 | #endif |
b196be89 | 255 | char name[]; /* I: workqueue name */ |
1da177e4 LT |
256 | }; |
257 | ||
d320c038 TH |
258 | struct workqueue_struct *system_wq __read_mostly; |
259 | struct workqueue_struct *system_long_wq __read_mostly; | |
260 | struct workqueue_struct *system_nrt_wq __read_mostly; | |
f3421797 | 261 | struct workqueue_struct *system_unbound_wq __read_mostly; |
24d51add | 262 | struct workqueue_struct *system_freezable_wq __read_mostly; |
62d3c543 | 263 | struct workqueue_struct *system_nrt_freezable_wq __read_mostly; |
d320c038 TH |
264 | EXPORT_SYMBOL_GPL(system_wq); |
265 | EXPORT_SYMBOL_GPL(system_long_wq); | |
266 | EXPORT_SYMBOL_GPL(system_nrt_wq); | |
f3421797 | 267 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
24d51add | 268 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
62d3c543 | 269 | EXPORT_SYMBOL_GPL(system_nrt_freezable_wq); |
d320c038 | 270 | |
97bd2347 TH |
271 | #define CREATE_TRACE_POINTS |
272 | #include <trace/events/workqueue.h> | |
273 | ||
db7bccf4 TH |
274 | #define for_each_busy_worker(worker, i, pos, gcwq) \ |
275 | for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++) \ | |
276 | hlist_for_each_entry(worker, pos, &gcwq->busy_hash[i], hentry) | |
277 | ||
f3421797 TH |
278 | static inline int __next_gcwq_cpu(int cpu, const struct cpumask *mask, |
279 | unsigned int sw) | |
280 | { | |
281 | if (cpu < nr_cpu_ids) { | |
282 | if (sw & 1) { | |
283 | cpu = cpumask_next(cpu, mask); | |
284 | if (cpu < nr_cpu_ids) | |
285 | return cpu; | |
286 | } | |
287 | if (sw & 2) | |
288 | return WORK_CPU_UNBOUND; | |
289 | } | |
290 | return WORK_CPU_NONE; | |
291 | } | |
292 | ||
293 | static inline int __next_wq_cpu(int cpu, const struct cpumask *mask, | |
294 | struct workqueue_struct *wq) | |
295 | { | |
296 | return __next_gcwq_cpu(cpu, mask, !(wq->flags & WQ_UNBOUND) ? 1 : 2); | |
297 | } | |
298 | ||
09884951 TH |
299 | /* |
300 | * CPU iterators | |
301 | * | |
302 | * An extra gcwq is defined for an invalid cpu number | |
303 | * (WORK_CPU_UNBOUND) to host workqueues which are not bound to any | |
304 | * specific CPU. The following iterators are similar to | |
305 | * for_each_*_cpu() iterators but also considers the unbound gcwq. | |
306 | * | |
307 | * for_each_gcwq_cpu() : possible CPUs + WORK_CPU_UNBOUND | |
308 | * for_each_online_gcwq_cpu() : online CPUs + WORK_CPU_UNBOUND | |
309 | * for_each_cwq_cpu() : possible CPUs for bound workqueues, | |
310 | * WORK_CPU_UNBOUND for unbound workqueues | |
311 | */ | |
f3421797 TH |
312 | #define for_each_gcwq_cpu(cpu) \ |
313 | for ((cpu) = __next_gcwq_cpu(-1, cpu_possible_mask, 3); \ | |
314 | (cpu) < WORK_CPU_NONE; \ | |
315 | (cpu) = __next_gcwq_cpu((cpu), cpu_possible_mask, 3)) | |
316 | ||
317 | #define for_each_online_gcwq_cpu(cpu) \ | |
318 | for ((cpu) = __next_gcwq_cpu(-1, cpu_online_mask, 3); \ | |
319 | (cpu) < WORK_CPU_NONE; \ | |
320 | (cpu) = __next_gcwq_cpu((cpu), cpu_online_mask, 3)) | |
321 | ||
322 | #define for_each_cwq_cpu(cpu, wq) \ | |
323 | for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, (wq)); \ | |
324 | (cpu) < WORK_CPU_NONE; \ | |
325 | (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, (wq))) | |
326 | ||
dc186ad7 TG |
327 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
328 | ||
329 | static struct debug_obj_descr work_debug_descr; | |
330 | ||
99777288 SG |
331 | static void *work_debug_hint(void *addr) |
332 | { | |
333 | return ((struct work_struct *) addr)->func; | |
334 | } | |
335 | ||
dc186ad7 TG |
336 | /* |
337 | * fixup_init is called when: | |
338 | * - an active object is initialized | |
339 | */ | |
340 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
341 | { | |
342 | struct work_struct *work = addr; | |
343 | ||
344 | switch (state) { | |
345 | case ODEBUG_STATE_ACTIVE: | |
346 | cancel_work_sync(work); | |
347 | debug_object_init(work, &work_debug_descr); | |
348 | return 1; | |
349 | default: | |
350 | return 0; | |
351 | } | |
352 | } | |
353 | ||
354 | /* | |
355 | * fixup_activate is called when: | |
356 | * - an active object is activated | |
357 | * - an unknown object is activated (might be a statically initialized object) | |
358 | */ | |
359 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
360 | { | |
361 | struct work_struct *work = addr; | |
362 | ||
363 | switch (state) { | |
364 | ||
365 | case ODEBUG_STATE_NOTAVAILABLE: | |
366 | /* | |
367 | * This is not really a fixup. The work struct was | |
368 | * statically initialized. We just make sure that it | |
369 | * is tracked in the object tracker. | |
370 | */ | |
22df02bb | 371 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
372 | debug_object_init(work, &work_debug_descr); |
373 | debug_object_activate(work, &work_debug_descr); | |
374 | return 0; | |
375 | } | |
376 | WARN_ON_ONCE(1); | |
377 | return 0; | |
378 | ||
379 | case ODEBUG_STATE_ACTIVE: | |
380 | WARN_ON(1); | |
381 | ||
382 | default: | |
383 | return 0; | |
384 | } | |
385 | } | |
386 | ||
387 | /* | |
388 | * fixup_free is called when: | |
389 | * - an active object is freed | |
390 | */ | |
391 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
392 | { | |
393 | struct work_struct *work = addr; | |
394 | ||
395 | switch (state) { | |
396 | case ODEBUG_STATE_ACTIVE: | |
397 | cancel_work_sync(work); | |
398 | debug_object_free(work, &work_debug_descr); | |
399 | return 1; | |
400 | default: | |
401 | return 0; | |
402 | } | |
403 | } | |
404 | ||
405 | static struct debug_obj_descr work_debug_descr = { | |
406 | .name = "work_struct", | |
99777288 | 407 | .debug_hint = work_debug_hint, |
dc186ad7 TG |
408 | .fixup_init = work_fixup_init, |
409 | .fixup_activate = work_fixup_activate, | |
410 | .fixup_free = work_fixup_free, | |
411 | }; | |
412 | ||
413 | static inline void debug_work_activate(struct work_struct *work) | |
414 | { | |
415 | debug_object_activate(work, &work_debug_descr); | |
416 | } | |
417 | ||
418 | static inline void debug_work_deactivate(struct work_struct *work) | |
419 | { | |
420 | debug_object_deactivate(work, &work_debug_descr); | |
421 | } | |
422 | ||
423 | void __init_work(struct work_struct *work, int onstack) | |
424 | { | |
425 | if (onstack) | |
426 | debug_object_init_on_stack(work, &work_debug_descr); | |
427 | else | |
428 | debug_object_init(work, &work_debug_descr); | |
429 | } | |
430 | EXPORT_SYMBOL_GPL(__init_work); | |
431 | ||
432 | void destroy_work_on_stack(struct work_struct *work) | |
433 | { | |
434 | debug_object_free(work, &work_debug_descr); | |
435 | } | |
436 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
437 | ||
438 | #else | |
439 | static inline void debug_work_activate(struct work_struct *work) { } | |
440 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
441 | #endif | |
442 | ||
95402b38 GS |
443 | /* Serializes the accesses to the list of workqueues. */ |
444 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 445 | static LIST_HEAD(workqueues); |
a0a1a5fd | 446 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 447 | |
e22bee78 TH |
448 | /* |
449 | * The almighty global cpu workqueues. nr_running is the only field | |
450 | * which is expected to be used frequently by other cpus via | |
451 | * try_to_wake_up(). Put it in a separate cacheline. | |
452 | */ | |
8b03ae3c | 453 | static DEFINE_PER_CPU(struct global_cwq, global_cwq); |
e22bee78 | 454 | static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, gcwq_nr_running); |
8b03ae3c | 455 | |
f3421797 TH |
456 | /* |
457 | * Global cpu workqueue and nr_running counter for unbound gcwq. The | |
458 | * gcwq is always online, has GCWQ_DISASSOCIATED set, and all its | |
459 | * workers have WORKER_UNBOUND set. | |
460 | */ | |
461 | static struct global_cwq unbound_global_cwq; | |
462 | static atomic_t unbound_gcwq_nr_running = ATOMIC_INIT(0); /* always 0 */ | |
463 | ||
c34056a3 | 464 | static int worker_thread(void *__worker); |
1da177e4 | 465 | |
8b03ae3c TH |
466 | static struct global_cwq *get_gcwq(unsigned int cpu) |
467 | { | |
f3421797 TH |
468 | if (cpu != WORK_CPU_UNBOUND) |
469 | return &per_cpu(global_cwq, cpu); | |
470 | else | |
471 | return &unbound_global_cwq; | |
8b03ae3c TH |
472 | } |
473 | ||
63d95a91 | 474 | static atomic_t *get_pool_nr_running(struct worker_pool *pool) |
e22bee78 | 475 | { |
63d95a91 TH |
476 | int cpu = pool->gcwq->cpu; |
477 | ||
f3421797 TH |
478 | if (cpu != WORK_CPU_UNBOUND) |
479 | return &per_cpu(gcwq_nr_running, cpu); | |
480 | else | |
481 | return &unbound_gcwq_nr_running; | |
e22bee78 TH |
482 | } |
483 | ||
1537663f TH |
484 | static struct cpu_workqueue_struct *get_cwq(unsigned int cpu, |
485 | struct workqueue_struct *wq) | |
b1f4ec17 | 486 | { |
f3421797 | 487 | if (!(wq->flags & WQ_UNBOUND)) { |
e06ffa1e | 488 | if (likely(cpu < nr_cpu_ids)) |
f3421797 | 489 | return per_cpu_ptr(wq->cpu_wq.pcpu, cpu); |
f3421797 TH |
490 | } else if (likely(cpu == WORK_CPU_UNBOUND)) |
491 | return wq->cpu_wq.single; | |
492 | return NULL; | |
b1f4ec17 ON |
493 | } |
494 | ||
73f53c4a TH |
495 | static unsigned int work_color_to_flags(int color) |
496 | { | |
497 | return color << WORK_STRUCT_COLOR_SHIFT; | |
498 | } | |
499 | ||
500 | static int get_work_color(struct work_struct *work) | |
501 | { | |
502 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
503 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
504 | } | |
505 | ||
506 | static int work_next_color(int color) | |
507 | { | |
508 | return (color + 1) % WORK_NR_COLORS; | |
509 | } | |
1da177e4 | 510 | |
14441960 | 511 | /* |
e120153d TH |
512 | * A work's data points to the cwq with WORK_STRUCT_CWQ set while the |
513 | * work is on queue. Once execution starts, WORK_STRUCT_CWQ is | |
514 | * cleared and the work data contains the cpu number it was last on. | |
7a22ad75 TH |
515 | * |
516 | * set_work_{cwq|cpu}() and clear_work_data() can be used to set the | |
517 | * cwq, cpu or clear work->data. These functions should only be | |
518 | * called while the work is owned - ie. while the PENDING bit is set. | |
519 | * | |
520 | * get_work_[g]cwq() can be used to obtain the gcwq or cwq | |
521 | * corresponding to a work. gcwq is available once the work has been | |
522 | * queued anywhere after initialization. cwq is available only from | |
523 | * queueing until execution starts. | |
14441960 | 524 | */ |
7a22ad75 TH |
525 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
526 | unsigned long flags) | |
365970a1 | 527 | { |
4594bf15 | 528 | BUG_ON(!work_pending(work)); |
7a22ad75 TH |
529 | atomic_long_set(&work->data, data | flags | work_static(work)); |
530 | } | |
365970a1 | 531 | |
7a22ad75 TH |
532 | static void set_work_cwq(struct work_struct *work, |
533 | struct cpu_workqueue_struct *cwq, | |
534 | unsigned long extra_flags) | |
535 | { | |
536 | set_work_data(work, (unsigned long)cwq, | |
e120153d | 537 | WORK_STRUCT_PENDING | WORK_STRUCT_CWQ | extra_flags); |
365970a1 DH |
538 | } |
539 | ||
7a22ad75 TH |
540 | static void set_work_cpu(struct work_struct *work, unsigned int cpu) |
541 | { | |
542 | set_work_data(work, cpu << WORK_STRUCT_FLAG_BITS, WORK_STRUCT_PENDING); | |
543 | } | |
f756d5e2 | 544 | |
7a22ad75 | 545 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 546 | { |
7a22ad75 | 547 | set_work_data(work, WORK_STRUCT_NO_CPU, 0); |
1da177e4 LT |
548 | } |
549 | ||
7a22ad75 | 550 | static struct cpu_workqueue_struct *get_work_cwq(struct work_struct *work) |
b1f4ec17 | 551 | { |
e120153d | 552 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 553 | |
e120153d TH |
554 | if (data & WORK_STRUCT_CWQ) |
555 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); | |
556 | else | |
557 | return NULL; | |
4d707b9f ON |
558 | } |
559 | ||
7a22ad75 | 560 | static struct global_cwq *get_work_gcwq(struct work_struct *work) |
365970a1 | 561 | { |
e120153d | 562 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 TH |
563 | unsigned int cpu; |
564 | ||
e120153d TH |
565 | if (data & WORK_STRUCT_CWQ) |
566 | return ((struct cpu_workqueue_struct *) | |
bd7bdd43 | 567 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool->gcwq; |
7a22ad75 TH |
568 | |
569 | cpu = data >> WORK_STRUCT_FLAG_BITS; | |
bdbc5dd7 | 570 | if (cpu == WORK_CPU_NONE) |
7a22ad75 TH |
571 | return NULL; |
572 | ||
f3421797 | 573 | BUG_ON(cpu >= nr_cpu_ids && cpu != WORK_CPU_UNBOUND); |
7a22ad75 | 574 | return get_gcwq(cpu); |
b1f4ec17 ON |
575 | } |
576 | ||
e22bee78 TH |
577 | /* |
578 | * Policy functions. These define the policies on how the global | |
579 | * worker pool is managed. Unless noted otherwise, these functions | |
580 | * assume that they're being called with gcwq->lock held. | |
581 | */ | |
582 | ||
63d95a91 | 583 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 584 | { |
63d95a91 TH |
585 | return !atomic_read(get_pool_nr_running(pool)) || |
586 | pool->gcwq->flags & GCWQ_HIGHPRI_PENDING; | |
a848e3b6 ON |
587 | } |
588 | ||
4594bf15 | 589 | /* |
e22bee78 TH |
590 | * Need to wake up a worker? Called from anything but currently |
591 | * running workers. | |
974271c4 TH |
592 | * |
593 | * Note that, because unbound workers never contribute to nr_running, this | |
594 | * function will always return %true for unbound gcwq as long as the | |
595 | * worklist isn't empty. | |
4594bf15 | 596 | */ |
63d95a91 | 597 | static bool need_more_worker(struct worker_pool *pool) |
365970a1 | 598 | { |
63d95a91 | 599 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
e22bee78 | 600 | } |
4594bf15 | 601 | |
e22bee78 | 602 | /* Can I start working? Called from busy but !running workers. */ |
63d95a91 | 603 | static bool may_start_working(struct worker_pool *pool) |
e22bee78 | 604 | { |
63d95a91 | 605 | return pool->nr_idle; |
e22bee78 TH |
606 | } |
607 | ||
608 | /* Do I need to keep working? Called from currently running workers. */ | |
63d95a91 | 609 | static bool keep_working(struct worker_pool *pool) |
e22bee78 | 610 | { |
63d95a91 | 611 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 | 612 | |
63d95a91 | 613 | return !list_empty(&pool->worklist) && |
30310045 | 614 | (atomic_read(nr_running) <= 1 || |
63d95a91 | 615 | pool->gcwq->flags & GCWQ_HIGHPRI_PENDING); |
e22bee78 TH |
616 | } |
617 | ||
618 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 619 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 620 | { |
63d95a91 | 621 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 622 | } |
365970a1 | 623 | |
e22bee78 | 624 | /* Do I need to be the manager? */ |
63d95a91 | 625 | static bool need_to_manage_workers(struct worker_pool *pool) |
e22bee78 | 626 | { |
63d95a91 TH |
627 | return need_to_create_worker(pool) || |
628 | pool->gcwq->flags & GCWQ_MANAGE_WORKERS; | |
e22bee78 TH |
629 | } |
630 | ||
631 | /* Do we have too many workers and should some go away? */ | |
63d95a91 | 632 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 633 | { |
63d95a91 TH |
634 | bool managing = pool->gcwq->flags & GCWQ_MANAGING_WORKERS; |
635 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ | |
636 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 TH |
637 | |
638 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; | |
365970a1 DH |
639 | } |
640 | ||
4d707b9f | 641 | /* |
e22bee78 TH |
642 | * Wake up functions. |
643 | */ | |
644 | ||
7e11629d | 645 | /* Return the first worker. Safe with preemption disabled */ |
63d95a91 | 646 | static struct worker *first_worker(struct worker_pool *pool) |
7e11629d | 647 | { |
63d95a91 | 648 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
649 | return NULL; |
650 | ||
63d95a91 | 651 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
652 | } |
653 | ||
654 | /** | |
655 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 656 | * @pool: worker pool to wake worker from |
7e11629d | 657 | * |
63d95a91 | 658 | * Wake up the first idle worker of @pool. |
7e11629d TH |
659 | * |
660 | * CONTEXT: | |
661 | * spin_lock_irq(gcwq->lock). | |
662 | */ | |
63d95a91 | 663 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 664 | { |
63d95a91 | 665 | struct worker *worker = first_worker(pool); |
7e11629d TH |
666 | |
667 | if (likely(worker)) | |
668 | wake_up_process(worker->task); | |
669 | } | |
670 | ||
d302f017 | 671 | /** |
e22bee78 TH |
672 | * wq_worker_waking_up - a worker is waking up |
673 | * @task: task waking up | |
674 | * @cpu: CPU @task is waking up to | |
675 | * | |
676 | * This function is called during try_to_wake_up() when a worker is | |
677 | * being awoken. | |
678 | * | |
679 | * CONTEXT: | |
680 | * spin_lock_irq(rq->lock) | |
681 | */ | |
682 | void wq_worker_waking_up(struct task_struct *task, unsigned int cpu) | |
683 | { | |
684 | struct worker *worker = kthread_data(task); | |
685 | ||
2d64672e | 686 | if (!(worker->flags & WORKER_NOT_RUNNING)) |
63d95a91 | 687 | atomic_inc(get_pool_nr_running(worker->pool)); |
e22bee78 TH |
688 | } |
689 | ||
690 | /** | |
691 | * wq_worker_sleeping - a worker is going to sleep | |
692 | * @task: task going to sleep | |
693 | * @cpu: CPU in question, must be the current CPU number | |
694 | * | |
695 | * This function is called during schedule() when a busy worker is | |
696 | * going to sleep. Worker on the same cpu can be woken up by | |
697 | * returning pointer to its task. | |
698 | * | |
699 | * CONTEXT: | |
700 | * spin_lock_irq(rq->lock) | |
701 | * | |
702 | * RETURNS: | |
703 | * Worker task on @cpu to wake up, %NULL if none. | |
704 | */ | |
705 | struct task_struct *wq_worker_sleeping(struct task_struct *task, | |
706 | unsigned int cpu) | |
707 | { | |
708 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
bd7bdd43 | 709 | struct worker_pool *pool = worker->pool; |
63d95a91 | 710 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 | 711 | |
2d64672e | 712 | if (worker->flags & WORKER_NOT_RUNNING) |
e22bee78 TH |
713 | return NULL; |
714 | ||
715 | /* this can only happen on the local cpu */ | |
716 | BUG_ON(cpu != raw_smp_processor_id()); | |
717 | ||
718 | /* | |
719 | * The counterpart of the following dec_and_test, implied mb, | |
720 | * worklist not empty test sequence is in insert_work(). | |
721 | * Please read comment there. | |
722 | * | |
723 | * NOT_RUNNING is clear. This means that trustee is not in | |
724 | * charge and we're running on the local cpu w/ rq lock held | |
725 | * and preemption disabled, which in turn means that none else | |
726 | * could be manipulating idle_list, so dereferencing idle_list | |
727 | * without gcwq lock is safe. | |
728 | */ | |
bd7bdd43 | 729 | if (atomic_dec_and_test(nr_running) && !list_empty(&pool->worklist)) |
63d95a91 | 730 | to_wakeup = first_worker(pool); |
e22bee78 TH |
731 | return to_wakeup ? to_wakeup->task : NULL; |
732 | } | |
733 | ||
734 | /** | |
735 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 736 | * @worker: self |
d302f017 TH |
737 | * @flags: flags to set |
738 | * @wakeup: wakeup an idle worker if necessary | |
739 | * | |
e22bee78 TH |
740 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
741 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
742 | * woken up. | |
d302f017 | 743 | * |
cb444766 TH |
744 | * CONTEXT: |
745 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
746 | */ |
747 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
748 | bool wakeup) | |
749 | { | |
bd7bdd43 | 750 | struct worker_pool *pool = worker->pool; |
e22bee78 | 751 | |
cb444766 TH |
752 | WARN_ON_ONCE(worker->task != current); |
753 | ||
e22bee78 TH |
754 | /* |
755 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
756 | * wake up an idle worker as necessary if requested by | |
757 | * @wakeup. | |
758 | */ | |
759 | if ((flags & WORKER_NOT_RUNNING) && | |
760 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
63d95a91 | 761 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 TH |
762 | |
763 | if (wakeup) { | |
764 | if (atomic_dec_and_test(nr_running) && | |
bd7bdd43 | 765 | !list_empty(&pool->worklist)) |
63d95a91 | 766 | wake_up_worker(pool); |
e22bee78 TH |
767 | } else |
768 | atomic_dec(nr_running); | |
769 | } | |
770 | ||
d302f017 TH |
771 | worker->flags |= flags; |
772 | } | |
773 | ||
774 | /** | |
e22bee78 | 775 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 776 | * @worker: self |
d302f017 TH |
777 | * @flags: flags to clear |
778 | * | |
e22bee78 | 779 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 780 | * |
cb444766 TH |
781 | * CONTEXT: |
782 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
783 | */ |
784 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
785 | { | |
63d95a91 | 786 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
787 | unsigned int oflags = worker->flags; |
788 | ||
cb444766 TH |
789 | WARN_ON_ONCE(worker->task != current); |
790 | ||
d302f017 | 791 | worker->flags &= ~flags; |
e22bee78 | 792 | |
42c025f3 TH |
793 | /* |
794 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
795 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
796 | * of multiple flags, not a single flag. | |
797 | */ | |
e22bee78 TH |
798 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
799 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
63d95a91 | 800 | atomic_inc(get_pool_nr_running(pool)); |
d302f017 TH |
801 | } |
802 | ||
c8e55f36 TH |
803 | /** |
804 | * busy_worker_head - return the busy hash head for a work | |
805 | * @gcwq: gcwq of interest | |
806 | * @work: work to be hashed | |
807 | * | |
808 | * Return hash head of @gcwq for @work. | |
809 | * | |
810 | * CONTEXT: | |
811 | * spin_lock_irq(gcwq->lock). | |
812 | * | |
813 | * RETURNS: | |
814 | * Pointer to the hash head. | |
815 | */ | |
816 | static struct hlist_head *busy_worker_head(struct global_cwq *gcwq, | |
817 | struct work_struct *work) | |
818 | { | |
819 | const int base_shift = ilog2(sizeof(struct work_struct)); | |
820 | unsigned long v = (unsigned long)work; | |
821 | ||
822 | /* simple shift and fold hash, do we need something better? */ | |
823 | v >>= base_shift; | |
824 | v += v >> BUSY_WORKER_HASH_ORDER; | |
825 | v &= BUSY_WORKER_HASH_MASK; | |
826 | ||
827 | return &gcwq->busy_hash[v]; | |
828 | } | |
829 | ||
8cca0eea TH |
830 | /** |
831 | * __find_worker_executing_work - find worker which is executing a work | |
832 | * @gcwq: gcwq of interest | |
833 | * @bwh: hash head as returned by busy_worker_head() | |
834 | * @work: work to find worker for | |
835 | * | |
836 | * Find a worker which is executing @work on @gcwq. @bwh should be | |
837 | * the hash head obtained by calling busy_worker_head() with the same | |
838 | * work. | |
839 | * | |
840 | * CONTEXT: | |
841 | * spin_lock_irq(gcwq->lock). | |
842 | * | |
843 | * RETURNS: | |
844 | * Pointer to worker which is executing @work if found, NULL | |
845 | * otherwise. | |
846 | */ | |
847 | static struct worker *__find_worker_executing_work(struct global_cwq *gcwq, | |
848 | struct hlist_head *bwh, | |
849 | struct work_struct *work) | |
850 | { | |
851 | struct worker *worker; | |
852 | struct hlist_node *tmp; | |
853 | ||
854 | hlist_for_each_entry(worker, tmp, bwh, hentry) | |
855 | if (worker->current_work == work) | |
856 | return worker; | |
857 | return NULL; | |
858 | } | |
859 | ||
860 | /** | |
861 | * find_worker_executing_work - find worker which is executing a work | |
862 | * @gcwq: gcwq of interest | |
863 | * @work: work to find worker for | |
864 | * | |
865 | * Find a worker which is executing @work on @gcwq. This function is | |
866 | * identical to __find_worker_executing_work() except that this | |
867 | * function calculates @bwh itself. | |
868 | * | |
869 | * CONTEXT: | |
870 | * spin_lock_irq(gcwq->lock). | |
871 | * | |
872 | * RETURNS: | |
873 | * Pointer to worker which is executing @work if found, NULL | |
874 | * otherwise. | |
4d707b9f | 875 | */ |
8cca0eea TH |
876 | static struct worker *find_worker_executing_work(struct global_cwq *gcwq, |
877 | struct work_struct *work) | |
4d707b9f | 878 | { |
8cca0eea TH |
879 | return __find_worker_executing_work(gcwq, busy_worker_head(gcwq, work), |
880 | work); | |
4d707b9f ON |
881 | } |
882 | ||
649027d7 | 883 | /** |
63d95a91 TH |
884 | * pool_determine_ins_pos - find insertion position |
885 | * @pool: pool of interest | |
649027d7 TH |
886 | * @cwq: cwq a work is being queued for |
887 | * | |
63d95a91 | 888 | * A work for @cwq is about to be queued on @pool, determine insertion |
649027d7 TH |
889 | * position for the work. If @cwq is for HIGHPRI wq, the work is |
890 | * queued at the head of the queue but in FIFO order with respect to | |
891 | * other HIGHPRI works; otherwise, at the end of the queue. This | |
63d95a91 | 892 | * function also sets GCWQ_HIGHPRI_PENDING flag to hint @pool that |
649027d7 TH |
893 | * there are HIGHPRI works pending. |
894 | * | |
895 | * CONTEXT: | |
896 | * spin_lock_irq(gcwq->lock). | |
897 | * | |
898 | * RETURNS: | |
899 | * Pointer to inserstion position. | |
900 | */ | |
63d95a91 | 901 | static inline struct list_head *pool_determine_ins_pos(struct worker_pool *pool, |
649027d7 | 902 | struct cpu_workqueue_struct *cwq) |
365970a1 | 903 | { |
649027d7 TH |
904 | struct work_struct *twork; |
905 | ||
906 | if (likely(!(cwq->wq->flags & WQ_HIGHPRI))) | |
63d95a91 | 907 | return &pool->worklist; |
649027d7 | 908 | |
63d95a91 | 909 | list_for_each_entry(twork, &pool->worklist, entry) { |
649027d7 TH |
910 | struct cpu_workqueue_struct *tcwq = get_work_cwq(twork); |
911 | ||
912 | if (!(tcwq->wq->flags & WQ_HIGHPRI)) | |
913 | break; | |
914 | } | |
915 | ||
63d95a91 | 916 | pool->gcwq->flags |= GCWQ_HIGHPRI_PENDING; |
649027d7 | 917 | return &twork->entry; |
365970a1 DH |
918 | } |
919 | ||
4690c4ab | 920 | /** |
7e11629d | 921 | * insert_work - insert a work into gcwq |
4690c4ab TH |
922 | * @cwq: cwq @work belongs to |
923 | * @work: work to insert | |
924 | * @head: insertion point | |
925 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
926 | * | |
7e11629d TH |
927 | * Insert @work which belongs to @cwq into @gcwq after @head. |
928 | * @extra_flags is or'd to work_struct flags. | |
4690c4ab TH |
929 | * |
930 | * CONTEXT: | |
8b03ae3c | 931 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 932 | */ |
b89deed3 | 933 | static void insert_work(struct cpu_workqueue_struct *cwq, |
4690c4ab TH |
934 | struct work_struct *work, struct list_head *head, |
935 | unsigned int extra_flags) | |
b89deed3 | 936 | { |
63d95a91 | 937 | struct worker_pool *pool = cwq->pool; |
e22bee78 | 938 | |
4690c4ab | 939 | /* we own @work, set data and link */ |
7a22ad75 | 940 | set_work_cwq(work, cwq, extra_flags); |
e1d8aa9f | 941 | |
6e84d644 ON |
942 | /* |
943 | * Ensure that we get the right work->data if we see the | |
944 | * result of list_add() below, see try_to_grab_pending(). | |
945 | */ | |
946 | smp_wmb(); | |
4690c4ab | 947 | |
1a4d9b0a | 948 | list_add_tail(&work->entry, head); |
e22bee78 TH |
949 | |
950 | /* | |
951 | * Ensure either worker_sched_deactivated() sees the above | |
952 | * list_add_tail() or we see zero nr_running to avoid workers | |
953 | * lying around lazily while there are works to be processed. | |
954 | */ | |
955 | smp_mb(); | |
956 | ||
63d95a91 TH |
957 | if (__need_more_worker(pool)) |
958 | wake_up_worker(pool); | |
b89deed3 ON |
959 | } |
960 | ||
c8efcc25 TH |
961 | /* |
962 | * Test whether @work is being queued from another work executing on the | |
963 | * same workqueue. This is rather expensive and should only be used from | |
964 | * cold paths. | |
965 | */ | |
966 | static bool is_chained_work(struct workqueue_struct *wq) | |
967 | { | |
968 | unsigned long flags; | |
969 | unsigned int cpu; | |
970 | ||
971 | for_each_gcwq_cpu(cpu) { | |
972 | struct global_cwq *gcwq = get_gcwq(cpu); | |
973 | struct worker *worker; | |
974 | struct hlist_node *pos; | |
975 | int i; | |
976 | ||
977 | spin_lock_irqsave(&gcwq->lock, flags); | |
978 | for_each_busy_worker(worker, i, pos, gcwq) { | |
979 | if (worker->task != current) | |
980 | continue; | |
981 | spin_unlock_irqrestore(&gcwq->lock, flags); | |
982 | /* | |
983 | * I'm @worker, no locking necessary. See if @work | |
984 | * is headed to the same workqueue. | |
985 | */ | |
986 | return worker->current_cwq->wq == wq; | |
987 | } | |
988 | spin_unlock_irqrestore(&gcwq->lock, flags); | |
989 | } | |
990 | return false; | |
991 | } | |
992 | ||
4690c4ab | 993 | static void __queue_work(unsigned int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
994 | struct work_struct *work) |
995 | { | |
502ca9d8 TH |
996 | struct global_cwq *gcwq; |
997 | struct cpu_workqueue_struct *cwq; | |
1e19ffc6 | 998 | struct list_head *worklist; |
8a2e8e5d | 999 | unsigned int work_flags; |
1da177e4 LT |
1000 | unsigned long flags; |
1001 | ||
dc186ad7 | 1002 | debug_work_activate(work); |
1e19ffc6 | 1003 | |
c8efcc25 | 1004 | /* if dying, only works from the same workqueue are allowed */ |
9c5a2ba7 | 1005 | if (unlikely(wq->flags & WQ_DRAINING) && |
c8efcc25 | 1006 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b TH |
1007 | return; |
1008 | ||
c7fc77f7 TH |
1009 | /* determine gcwq to use */ |
1010 | if (!(wq->flags & WQ_UNBOUND)) { | |
18aa9eff TH |
1011 | struct global_cwq *last_gcwq; |
1012 | ||
c7fc77f7 TH |
1013 | if (unlikely(cpu == WORK_CPU_UNBOUND)) |
1014 | cpu = raw_smp_processor_id(); | |
1015 | ||
18aa9eff TH |
1016 | /* |
1017 | * It's multi cpu. If @wq is non-reentrant and @work | |
1018 | * was previously on a different cpu, it might still | |
1019 | * be running there, in which case the work needs to | |
1020 | * be queued on that cpu to guarantee non-reentrance. | |
1021 | */ | |
502ca9d8 | 1022 | gcwq = get_gcwq(cpu); |
18aa9eff TH |
1023 | if (wq->flags & WQ_NON_REENTRANT && |
1024 | (last_gcwq = get_work_gcwq(work)) && last_gcwq != gcwq) { | |
1025 | struct worker *worker; | |
1026 | ||
1027 | spin_lock_irqsave(&last_gcwq->lock, flags); | |
1028 | ||
1029 | worker = find_worker_executing_work(last_gcwq, work); | |
1030 | ||
1031 | if (worker && worker->current_cwq->wq == wq) | |
1032 | gcwq = last_gcwq; | |
1033 | else { | |
1034 | /* meh... not running there, queue here */ | |
1035 | spin_unlock_irqrestore(&last_gcwq->lock, flags); | |
1036 | spin_lock_irqsave(&gcwq->lock, flags); | |
1037 | } | |
1038 | } else | |
1039 | spin_lock_irqsave(&gcwq->lock, flags); | |
f3421797 TH |
1040 | } else { |
1041 | gcwq = get_gcwq(WORK_CPU_UNBOUND); | |
1042 | spin_lock_irqsave(&gcwq->lock, flags); | |
502ca9d8 TH |
1043 | } |
1044 | ||
1045 | /* gcwq determined, get cwq and queue */ | |
1046 | cwq = get_cwq(gcwq->cpu, wq); | |
cdadf009 | 1047 | trace_workqueue_queue_work(cpu, cwq, work); |
502ca9d8 | 1048 | |
f5b2552b DC |
1049 | if (WARN_ON(!list_empty(&work->entry))) { |
1050 | spin_unlock_irqrestore(&gcwq->lock, flags); | |
1051 | return; | |
1052 | } | |
1e19ffc6 | 1053 | |
73f53c4a | 1054 | cwq->nr_in_flight[cwq->work_color]++; |
8a2e8e5d | 1055 | work_flags = work_color_to_flags(cwq->work_color); |
1e19ffc6 TH |
1056 | |
1057 | if (likely(cwq->nr_active < cwq->max_active)) { | |
cdadf009 | 1058 | trace_workqueue_activate_work(work); |
1e19ffc6 | 1059 | cwq->nr_active++; |
63d95a91 | 1060 | worklist = pool_determine_ins_pos(cwq->pool, cwq); |
8a2e8e5d TH |
1061 | } else { |
1062 | work_flags |= WORK_STRUCT_DELAYED; | |
1e19ffc6 | 1063 | worklist = &cwq->delayed_works; |
8a2e8e5d | 1064 | } |
1e19ffc6 | 1065 | |
8a2e8e5d | 1066 | insert_work(cwq, work, worklist, work_flags); |
1e19ffc6 | 1067 | |
8b03ae3c | 1068 | spin_unlock_irqrestore(&gcwq->lock, flags); |
1da177e4 LT |
1069 | } |
1070 | ||
0fcb78c2 REB |
1071 | /** |
1072 | * queue_work - queue work on a workqueue | |
1073 | * @wq: workqueue to use | |
1074 | * @work: work to queue | |
1075 | * | |
057647fc | 1076 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
1da177e4 | 1077 | * |
00dfcaf7 ON |
1078 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
1079 | * it can be processed by another CPU. | |
1da177e4 | 1080 | */ |
7ad5b3a5 | 1081 | int queue_work(struct workqueue_struct *wq, struct work_struct *work) |
1da177e4 | 1082 | { |
ef1ca236 ON |
1083 | int ret; |
1084 | ||
1085 | ret = queue_work_on(get_cpu(), wq, work); | |
1086 | put_cpu(); | |
1087 | ||
1da177e4 LT |
1088 | return ret; |
1089 | } | |
ae90dd5d | 1090 | EXPORT_SYMBOL_GPL(queue_work); |
1da177e4 | 1091 | |
c1a220e7 ZR |
1092 | /** |
1093 | * queue_work_on - queue work on specific cpu | |
1094 | * @cpu: CPU number to execute work on | |
1095 | * @wq: workqueue to use | |
1096 | * @work: work to queue | |
1097 | * | |
1098 | * Returns 0 if @work was already on a queue, non-zero otherwise. | |
1099 | * | |
1100 | * We queue the work to a specific CPU, the caller must ensure it | |
1101 | * can't go away. | |
1102 | */ | |
1103 | int | |
1104 | queue_work_on(int cpu, struct workqueue_struct *wq, struct work_struct *work) | |
1105 | { | |
1106 | int ret = 0; | |
1107 | ||
22df02bb | 1108 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1109 | __queue_work(cpu, wq, work); |
c1a220e7 ZR |
1110 | ret = 1; |
1111 | } | |
1112 | return ret; | |
1113 | } | |
1114 | EXPORT_SYMBOL_GPL(queue_work_on); | |
1115 | ||
6d141c3f | 1116 | static void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1117 | { |
52bad64d | 1118 | struct delayed_work *dwork = (struct delayed_work *)__data; |
7a22ad75 | 1119 | struct cpu_workqueue_struct *cwq = get_work_cwq(&dwork->work); |
1da177e4 | 1120 | |
4690c4ab | 1121 | __queue_work(smp_processor_id(), cwq->wq, &dwork->work); |
1da177e4 LT |
1122 | } |
1123 | ||
0fcb78c2 REB |
1124 | /** |
1125 | * queue_delayed_work - queue work on a workqueue after delay | |
1126 | * @wq: workqueue to use | |
af9997e4 | 1127 | * @dwork: delayable work to queue |
0fcb78c2 REB |
1128 | * @delay: number of jiffies to wait before queueing |
1129 | * | |
057647fc | 1130 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 1131 | */ |
7ad5b3a5 | 1132 | int queue_delayed_work(struct workqueue_struct *wq, |
52bad64d | 1133 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 1134 | { |
52bad64d | 1135 | if (delay == 0) |
63bc0362 | 1136 | return queue_work(wq, &dwork->work); |
1da177e4 | 1137 | |
63bc0362 | 1138 | return queue_delayed_work_on(-1, wq, dwork, delay); |
1da177e4 | 1139 | } |
ae90dd5d | 1140 | EXPORT_SYMBOL_GPL(queue_delayed_work); |
1da177e4 | 1141 | |
0fcb78c2 REB |
1142 | /** |
1143 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1144 | * @cpu: CPU number to execute work on | |
1145 | * @wq: workqueue to use | |
af9997e4 | 1146 | * @dwork: work to queue |
0fcb78c2 REB |
1147 | * @delay: number of jiffies to wait before queueing |
1148 | * | |
057647fc | 1149 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 1150 | */ |
7a6bc1cd | 1151 | int queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
52bad64d | 1152 | struct delayed_work *dwork, unsigned long delay) |
7a6bc1cd VP |
1153 | { |
1154 | int ret = 0; | |
52bad64d DH |
1155 | struct timer_list *timer = &dwork->timer; |
1156 | struct work_struct *work = &dwork->work; | |
7a6bc1cd | 1157 | |
22df02bb | 1158 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
c7fc77f7 | 1159 | unsigned int lcpu; |
7a22ad75 | 1160 | |
7a6bc1cd VP |
1161 | BUG_ON(timer_pending(timer)); |
1162 | BUG_ON(!list_empty(&work->entry)); | |
1163 | ||
8a3e77cc AL |
1164 | timer_stats_timer_set_start_info(&dwork->timer); |
1165 | ||
7a22ad75 TH |
1166 | /* |
1167 | * This stores cwq for the moment, for the timer_fn. | |
1168 | * Note that the work's gcwq is preserved to allow | |
1169 | * reentrance detection for delayed works. | |
1170 | */ | |
c7fc77f7 TH |
1171 | if (!(wq->flags & WQ_UNBOUND)) { |
1172 | struct global_cwq *gcwq = get_work_gcwq(work); | |
1173 | ||
1174 | if (gcwq && gcwq->cpu != WORK_CPU_UNBOUND) | |
1175 | lcpu = gcwq->cpu; | |
1176 | else | |
1177 | lcpu = raw_smp_processor_id(); | |
1178 | } else | |
1179 | lcpu = WORK_CPU_UNBOUND; | |
1180 | ||
7a22ad75 | 1181 | set_work_cwq(work, get_cwq(lcpu, wq), 0); |
c7fc77f7 | 1182 | |
7a6bc1cd | 1183 | timer->expires = jiffies + delay; |
52bad64d | 1184 | timer->data = (unsigned long)dwork; |
7a6bc1cd | 1185 | timer->function = delayed_work_timer_fn; |
63bc0362 ON |
1186 | |
1187 | if (unlikely(cpu >= 0)) | |
1188 | add_timer_on(timer, cpu); | |
1189 | else | |
1190 | add_timer(timer); | |
7a6bc1cd VP |
1191 | ret = 1; |
1192 | } | |
1193 | return ret; | |
1194 | } | |
ae90dd5d | 1195 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
1da177e4 | 1196 | |
c8e55f36 TH |
1197 | /** |
1198 | * worker_enter_idle - enter idle state | |
1199 | * @worker: worker which is entering idle state | |
1200 | * | |
1201 | * @worker is entering idle state. Update stats and idle timer if | |
1202 | * necessary. | |
1203 | * | |
1204 | * LOCKING: | |
1205 | * spin_lock_irq(gcwq->lock). | |
1206 | */ | |
1207 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1208 | { |
bd7bdd43 TH |
1209 | struct worker_pool *pool = worker->pool; |
1210 | struct global_cwq *gcwq = pool->gcwq; | |
c8e55f36 TH |
1211 | |
1212 | BUG_ON(worker->flags & WORKER_IDLE); | |
1213 | BUG_ON(!list_empty(&worker->entry) && | |
1214 | (worker->hentry.next || worker->hentry.pprev)); | |
1215 | ||
cb444766 TH |
1216 | /* can't use worker_set_flags(), also called from start_worker() */ |
1217 | worker->flags |= WORKER_IDLE; | |
bd7bdd43 | 1218 | pool->nr_idle++; |
e22bee78 | 1219 | worker->last_active = jiffies; |
c8e55f36 TH |
1220 | |
1221 | /* idle_list is LIFO */ | |
bd7bdd43 | 1222 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1223 | |
e22bee78 | 1224 | if (likely(!(worker->flags & WORKER_ROGUE))) { |
63d95a91 | 1225 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
bd7bdd43 | 1226 | mod_timer(&pool->idle_timer, |
e22bee78 TH |
1227 | jiffies + IDLE_WORKER_TIMEOUT); |
1228 | } else | |
db7bccf4 | 1229 | wake_up_all(&gcwq->trustee_wait); |
cb444766 | 1230 | |
544ecf31 TH |
1231 | /* |
1232 | * Sanity check nr_running. Because trustee releases gcwq->lock | |
1233 | * between setting %WORKER_ROGUE and zapping nr_running, the | |
1234 | * warning may trigger spuriously. Check iff trustee is idle. | |
1235 | */ | |
1236 | WARN_ON_ONCE(gcwq->trustee_state == TRUSTEE_DONE && | |
bd7bdd43 | 1237 | pool->nr_workers == pool->nr_idle && |
63d95a91 | 1238 | atomic_read(get_pool_nr_running(pool))); |
c8e55f36 TH |
1239 | } |
1240 | ||
1241 | /** | |
1242 | * worker_leave_idle - leave idle state | |
1243 | * @worker: worker which is leaving idle state | |
1244 | * | |
1245 | * @worker is leaving idle state. Update stats. | |
1246 | * | |
1247 | * LOCKING: | |
1248 | * spin_lock_irq(gcwq->lock). | |
1249 | */ | |
1250 | static void worker_leave_idle(struct worker *worker) | |
1251 | { | |
bd7bdd43 | 1252 | struct worker_pool *pool = worker->pool; |
c8e55f36 TH |
1253 | |
1254 | BUG_ON(!(worker->flags & WORKER_IDLE)); | |
d302f017 | 1255 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1256 | pool->nr_idle--; |
c8e55f36 TH |
1257 | list_del_init(&worker->entry); |
1258 | } | |
1259 | ||
e22bee78 TH |
1260 | /** |
1261 | * worker_maybe_bind_and_lock - bind worker to its cpu if possible and lock gcwq | |
1262 | * @worker: self | |
1263 | * | |
1264 | * Works which are scheduled while the cpu is online must at least be | |
1265 | * scheduled to a worker which is bound to the cpu so that if they are | |
1266 | * flushed from cpu callbacks while cpu is going down, they are | |
1267 | * guaranteed to execute on the cpu. | |
1268 | * | |
1269 | * This function is to be used by rogue workers and rescuers to bind | |
1270 | * themselves to the target cpu and may race with cpu going down or | |
1271 | * coming online. kthread_bind() can't be used because it may put the | |
1272 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
1273 | * verbatim as it's best effort and blocking and gcwq may be | |
1274 | * [dis]associated in the meantime. | |
1275 | * | |
1276 | * This function tries set_cpus_allowed() and locks gcwq and verifies | |
1277 | * the binding against GCWQ_DISASSOCIATED which is set during | |
1278 | * CPU_DYING and cleared during CPU_ONLINE, so if the worker enters | |
1279 | * idle state or fetches works without dropping lock, it can guarantee | |
1280 | * the scheduling requirement described in the first paragraph. | |
1281 | * | |
1282 | * CONTEXT: | |
1283 | * Might sleep. Called without any lock but returns with gcwq->lock | |
1284 | * held. | |
1285 | * | |
1286 | * RETURNS: | |
1287 | * %true if the associated gcwq is online (@worker is successfully | |
1288 | * bound), %false if offline. | |
1289 | */ | |
1290 | static bool worker_maybe_bind_and_lock(struct worker *worker) | |
972fa1c5 | 1291 | __acquires(&gcwq->lock) |
e22bee78 | 1292 | { |
bd7bdd43 | 1293 | struct global_cwq *gcwq = worker->pool->gcwq; |
e22bee78 TH |
1294 | struct task_struct *task = worker->task; |
1295 | ||
1296 | while (true) { | |
4e6045f1 | 1297 | /* |
e22bee78 TH |
1298 | * The following call may fail, succeed or succeed |
1299 | * without actually migrating the task to the cpu if | |
1300 | * it races with cpu hotunplug operation. Verify | |
1301 | * against GCWQ_DISASSOCIATED. | |
4e6045f1 | 1302 | */ |
f3421797 TH |
1303 | if (!(gcwq->flags & GCWQ_DISASSOCIATED)) |
1304 | set_cpus_allowed_ptr(task, get_cpu_mask(gcwq->cpu)); | |
e22bee78 TH |
1305 | |
1306 | spin_lock_irq(&gcwq->lock); | |
1307 | if (gcwq->flags & GCWQ_DISASSOCIATED) | |
1308 | return false; | |
1309 | if (task_cpu(task) == gcwq->cpu && | |
1310 | cpumask_equal(¤t->cpus_allowed, | |
1311 | get_cpu_mask(gcwq->cpu))) | |
1312 | return true; | |
1313 | spin_unlock_irq(&gcwq->lock); | |
1314 | ||
5035b20f TH |
1315 | /* |
1316 | * We've raced with CPU hot[un]plug. Give it a breather | |
1317 | * and retry migration. cond_resched() is required here; | |
1318 | * otherwise, we might deadlock against cpu_stop trying to | |
1319 | * bring down the CPU on non-preemptive kernel. | |
1320 | */ | |
e22bee78 | 1321 | cpu_relax(); |
5035b20f | 1322 | cond_resched(); |
e22bee78 TH |
1323 | } |
1324 | } | |
1325 | ||
1326 | /* | |
1327 | * Function for worker->rebind_work used to rebind rogue busy workers | |
1328 | * to the associated cpu which is coming back online. This is | |
1329 | * scheduled by cpu up but can race with other cpu hotplug operations | |
1330 | * and may be executed twice without intervening cpu down. | |
1331 | */ | |
1332 | static void worker_rebind_fn(struct work_struct *work) | |
1333 | { | |
1334 | struct worker *worker = container_of(work, struct worker, rebind_work); | |
bd7bdd43 | 1335 | struct global_cwq *gcwq = worker->pool->gcwq; |
e22bee78 TH |
1336 | |
1337 | if (worker_maybe_bind_and_lock(worker)) | |
1338 | worker_clr_flags(worker, WORKER_REBIND); | |
1339 | ||
1340 | spin_unlock_irq(&gcwq->lock); | |
1341 | } | |
1342 | ||
c34056a3 TH |
1343 | static struct worker *alloc_worker(void) |
1344 | { | |
1345 | struct worker *worker; | |
1346 | ||
1347 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1348 | if (worker) { |
1349 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1350 | INIT_LIST_HEAD(&worker->scheduled); |
e22bee78 TH |
1351 | INIT_WORK(&worker->rebind_work, worker_rebind_fn); |
1352 | /* on creation a worker is in !idle && prep state */ | |
1353 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1354 | } |
c34056a3 TH |
1355 | return worker; |
1356 | } | |
1357 | ||
1358 | /** | |
1359 | * create_worker - create a new workqueue worker | |
63d95a91 | 1360 | * @pool: pool the new worker will belong to |
c34056a3 TH |
1361 | * @bind: whether to set affinity to @cpu or not |
1362 | * | |
63d95a91 | 1363 | * Create a new worker which is bound to @pool. The returned worker |
c34056a3 TH |
1364 | * can be started by calling start_worker() or destroyed using |
1365 | * destroy_worker(). | |
1366 | * | |
1367 | * CONTEXT: | |
1368 | * Might sleep. Does GFP_KERNEL allocations. | |
1369 | * | |
1370 | * RETURNS: | |
1371 | * Pointer to the newly created worker. | |
1372 | */ | |
63d95a91 | 1373 | static struct worker *create_worker(struct worker_pool *pool, bool bind) |
c34056a3 | 1374 | { |
63d95a91 | 1375 | struct global_cwq *gcwq = pool->gcwq; |
f3421797 | 1376 | bool on_unbound_cpu = gcwq->cpu == WORK_CPU_UNBOUND; |
c34056a3 | 1377 | struct worker *worker = NULL; |
f3421797 | 1378 | int id = -1; |
c34056a3 | 1379 | |
8b03ae3c | 1380 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1381 | while (ida_get_new(&pool->worker_ida, &id)) { |
8b03ae3c | 1382 | spin_unlock_irq(&gcwq->lock); |
bd7bdd43 | 1383 | if (!ida_pre_get(&pool->worker_ida, GFP_KERNEL)) |
c34056a3 | 1384 | goto fail; |
8b03ae3c | 1385 | spin_lock_irq(&gcwq->lock); |
c34056a3 | 1386 | } |
8b03ae3c | 1387 | spin_unlock_irq(&gcwq->lock); |
c34056a3 TH |
1388 | |
1389 | worker = alloc_worker(); | |
1390 | if (!worker) | |
1391 | goto fail; | |
1392 | ||
bd7bdd43 | 1393 | worker->pool = pool; |
c34056a3 TH |
1394 | worker->id = id; |
1395 | ||
f3421797 | 1396 | if (!on_unbound_cpu) |
94dcf29a ED |
1397 | worker->task = kthread_create_on_node(worker_thread, |
1398 | worker, | |
1399 | cpu_to_node(gcwq->cpu), | |
1400 | "kworker/%u:%d", gcwq->cpu, id); | |
f3421797 TH |
1401 | else |
1402 | worker->task = kthread_create(worker_thread, worker, | |
1403 | "kworker/u:%d", id); | |
c34056a3 TH |
1404 | if (IS_ERR(worker->task)) |
1405 | goto fail; | |
1406 | ||
db7bccf4 TH |
1407 | /* |
1408 | * A rogue worker will become a regular one if CPU comes | |
1409 | * online later on. Make sure every worker has | |
1410 | * PF_THREAD_BOUND set. | |
1411 | */ | |
f3421797 | 1412 | if (bind && !on_unbound_cpu) |
8b03ae3c | 1413 | kthread_bind(worker->task, gcwq->cpu); |
f3421797 | 1414 | else { |
db7bccf4 | 1415 | worker->task->flags |= PF_THREAD_BOUND; |
f3421797 TH |
1416 | if (on_unbound_cpu) |
1417 | worker->flags |= WORKER_UNBOUND; | |
1418 | } | |
c34056a3 TH |
1419 | |
1420 | return worker; | |
1421 | fail: | |
1422 | if (id >= 0) { | |
8b03ae3c | 1423 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1424 | ida_remove(&pool->worker_ida, id); |
8b03ae3c | 1425 | spin_unlock_irq(&gcwq->lock); |
c34056a3 TH |
1426 | } |
1427 | kfree(worker); | |
1428 | return NULL; | |
1429 | } | |
1430 | ||
1431 | /** | |
1432 | * start_worker - start a newly created worker | |
1433 | * @worker: worker to start | |
1434 | * | |
c8e55f36 | 1435 | * Make the gcwq aware of @worker and start it. |
c34056a3 TH |
1436 | * |
1437 | * CONTEXT: | |
8b03ae3c | 1438 | * spin_lock_irq(gcwq->lock). |
c34056a3 TH |
1439 | */ |
1440 | static void start_worker(struct worker *worker) | |
1441 | { | |
cb444766 | 1442 | worker->flags |= WORKER_STARTED; |
bd7bdd43 | 1443 | worker->pool->nr_workers++; |
c8e55f36 | 1444 | worker_enter_idle(worker); |
c34056a3 TH |
1445 | wake_up_process(worker->task); |
1446 | } | |
1447 | ||
1448 | /** | |
1449 | * destroy_worker - destroy a workqueue worker | |
1450 | * @worker: worker to be destroyed | |
1451 | * | |
c8e55f36 TH |
1452 | * Destroy @worker and adjust @gcwq stats accordingly. |
1453 | * | |
1454 | * CONTEXT: | |
1455 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. | |
c34056a3 TH |
1456 | */ |
1457 | static void destroy_worker(struct worker *worker) | |
1458 | { | |
bd7bdd43 TH |
1459 | struct worker_pool *pool = worker->pool; |
1460 | struct global_cwq *gcwq = pool->gcwq; | |
c34056a3 TH |
1461 | int id = worker->id; |
1462 | ||
1463 | /* sanity check frenzy */ | |
1464 | BUG_ON(worker->current_work); | |
affee4b2 | 1465 | BUG_ON(!list_empty(&worker->scheduled)); |
c34056a3 | 1466 | |
c8e55f36 | 1467 | if (worker->flags & WORKER_STARTED) |
bd7bdd43 | 1468 | pool->nr_workers--; |
c8e55f36 | 1469 | if (worker->flags & WORKER_IDLE) |
bd7bdd43 | 1470 | pool->nr_idle--; |
c8e55f36 TH |
1471 | |
1472 | list_del_init(&worker->entry); | |
cb444766 | 1473 | worker->flags |= WORKER_DIE; |
c8e55f36 TH |
1474 | |
1475 | spin_unlock_irq(&gcwq->lock); | |
1476 | ||
c34056a3 TH |
1477 | kthread_stop(worker->task); |
1478 | kfree(worker); | |
1479 | ||
8b03ae3c | 1480 | spin_lock_irq(&gcwq->lock); |
bd7bdd43 | 1481 | ida_remove(&pool->worker_ida, id); |
c34056a3 TH |
1482 | } |
1483 | ||
63d95a91 | 1484 | static void idle_worker_timeout(unsigned long __pool) |
e22bee78 | 1485 | { |
63d95a91 TH |
1486 | struct worker_pool *pool = (void *)__pool; |
1487 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
1488 | |
1489 | spin_lock_irq(&gcwq->lock); | |
1490 | ||
63d95a91 | 1491 | if (too_many_workers(pool)) { |
e22bee78 TH |
1492 | struct worker *worker; |
1493 | unsigned long expires; | |
1494 | ||
1495 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1496 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1497 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1498 | ||
1499 | if (time_before(jiffies, expires)) | |
63d95a91 | 1500 | mod_timer(&pool->idle_timer, expires); |
e22bee78 TH |
1501 | else { |
1502 | /* it's been idle for too long, wake up manager */ | |
1503 | gcwq->flags |= GCWQ_MANAGE_WORKERS; | |
63d95a91 | 1504 | wake_up_worker(pool); |
d5abe669 | 1505 | } |
e22bee78 TH |
1506 | } |
1507 | ||
1508 | spin_unlock_irq(&gcwq->lock); | |
1509 | } | |
d5abe669 | 1510 | |
e22bee78 TH |
1511 | static bool send_mayday(struct work_struct *work) |
1512 | { | |
1513 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); | |
1514 | struct workqueue_struct *wq = cwq->wq; | |
f3421797 | 1515 | unsigned int cpu; |
e22bee78 TH |
1516 | |
1517 | if (!(wq->flags & WQ_RESCUER)) | |
1518 | return false; | |
1519 | ||
1520 | /* mayday mayday mayday */ | |
bd7bdd43 | 1521 | cpu = cwq->pool->gcwq->cpu; |
f3421797 TH |
1522 | /* WORK_CPU_UNBOUND can't be set in cpumask, use cpu 0 instead */ |
1523 | if (cpu == WORK_CPU_UNBOUND) | |
1524 | cpu = 0; | |
f2e005aa | 1525 | if (!mayday_test_and_set_cpu(cpu, wq->mayday_mask)) |
e22bee78 TH |
1526 | wake_up_process(wq->rescuer->task); |
1527 | return true; | |
1528 | } | |
1529 | ||
63d95a91 | 1530 | static void gcwq_mayday_timeout(unsigned long __pool) |
e22bee78 | 1531 | { |
63d95a91 TH |
1532 | struct worker_pool *pool = (void *)__pool; |
1533 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
1534 | struct work_struct *work; |
1535 | ||
1536 | spin_lock_irq(&gcwq->lock); | |
1537 | ||
63d95a91 | 1538 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
1539 | /* |
1540 | * We've been trying to create a new worker but | |
1541 | * haven't been successful. We might be hitting an | |
1542 | * allocation deadlock. Send distress signals to | |
1543 | * rescuers. | |
1544 | */ | |
63d95a91 | 1545 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 1546 | send_mayday(work); |
1da177e4 | 1547 | } |
e22bee78 TH |
1548 | |
1549 | spin_unlock_irq(&gcwq->lock); | |
1550 | ||
63d95a91 | 1551 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
1552 | } |
1553 | ||
e22bee78 TH |
1554 | /** |
1555 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 1556 | * @pool: pool to create a new worker for |
e22bee78 | 1557 | * |
63d95a91 | 1558 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
1559 | * have at least one idle worker on return from this function. If |
1560 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 1561 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
1562 | * possible allocation deadlock. |
1563 | * | |
1564 | * On return, need_to_create_worker() is guaranteed to be false and | |
1565 | * may_start_working() true. | |
1566 | * | |
1567 | * LOCKING: | |
1568 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1569 | * multiple times. Does GFP_KERNEL allocations. Called only from | |
1570 | * manager. | |
1571 | * | |
1572 | * RETURNS: | |
1573 | * false if no action was taken and gcwq->lock stayed locked, true | |
1574 | * otherwise. | |
1575 | */ | |
63d95a91 | 1576 | static bool maybe_create_worker(struct worker_pool *pool) |
06bd6ebf NK |
1577 | __releases(&gcwq->lock) |
1578 | __acquires(&gcwq->lock) | |
1da177e4 | 1579 | { |
63d95a91 TH |
1580 | struct global_cwq *gcwq = pool->gcwq; |
1581 | ||
1582 | if (!need_to_create_worker(pool)) | |
e22bee78 TH |
1583 | return false; |
1584 | restart: | |
9f9c2364 TH |
1585 | spin_unlock_irq(&gcwq->lock); |
1586 | ||
e22bee78 | 1587 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 1588 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
1589 | |
1590 | while (true) { | |
1591 | struct worker *worker; | |
1592 | ||
63d95a91 | 1593 | worker = create_worker(pool, true); |
e22bee78 | 1594 | if (worker) { |
63d95a91 | 1595 | del_timer_sync(&pool->mayday_timer); |
e22bee78 TH |
1596 | spin_lock_irq(&gcwq->lock); |
1597 | start_worker(worker); | |
63d95a91 | 1598 | BUG_ON(need_to_create_worker(pool)); |
e22bee78 TH |
1599 | return true; |
1600 | } | |
1601 | ||
63d95a91 | 1602 | if (!need_to_create_worker(pool)) |
e22bee78 | 1603 | break; |
1da177e4 | 1604 | |
e22bee78 TH |
1605 | __set_current_state(TASK_INTERRUPTIBLE); |
1606 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1607 | |
63d95a91 | 1608 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1609 | break; |
1610 | } | |
1611 | ||
63d95a91 | 1612 | del_timer_sync(&pool->mayday_timer); |
e22bee78 | 1613 | spin_lock_irq(&gcwq->lock); |
63d95a91 | 1614 | if (need_to_create_worker(pool)) |
e22bee78 TH |
1615 | goto restart; |
1616 | return true; | |
1617 | } | |
1618 | ||
1619 | /** | |
1620 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
63d95a91 | 1621 | * @pool: pool to destroy workers for |
e22bee78 | 1622 | * |
63d95a91 | 1623 | * Destroy @pool workers which have been idle for longer than |
e22bee78 TH |
1624 | * IDLE_WORKER_TIMEOUT. |
1625 | * | |
1626 | * LOCKING: | |
1627 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1628 | * multiple times. Called only from manager. | |
1629 | * | |
1630 | * RETURNS: | |
1631 | * false if no action was taken and gcwq->lock stayed locked, true | |
1632 | * otherwise. | |
1633 | */ | |
63d95a91 | 1634 | static bool maybe_destroy_workers(struct worker_pool *pool) |
e22bee78 TH |
1635 | { |
1636 | bool ret = false; | |
1da177e4 | 1637 | |
63d95a91 | 1638 | while (too_many_workers(pool)) { |
e22bee78 TH |
1639 | struct worker *worker; |
1640 | unsigned long expires; | |
3af24433 | 1641 | |
63d95a91 | 1642 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 | 1643 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
85f4186a | 1644 | |
e22bee78 | 1645 | if (time_before(jiffies, expires)) { |
63d95a91 | 1646 | mod_timer(&pool->idle_timer, expires); |
3af24433 | 1647 | break; |
e22bee78 | 1648 | } |
1da177e4 | 1649 | |
e22bee78 TH |
1650 | destroy_worker(worker); |
1651 | ret = true; | |
1da177e4 | 1652 | } |
3af24433 | 1653 | |
e22bee78 TH |
1654 | return ret; |
1655 | } | |
1656 | ||
1657 | /** | |
1658 | * manage_workers - manage worker pool | |
1659 | * @worker: self | |
1660 | * | |
1661 | * Assume the manager role and manage gcwq worker pool @worker belongs | |
1662 | * to. At any given time, there can be only zero or one manager per | |
1663 | * gcwq. The exclusion is handled automatically by this function. | |
1664 | * | |
1665 | * The caller can safely start processing works on false return. On | |
1666 | * true return, it's guaranteed that need_to_create_worker() is false | |
1667 | * and may_start_working() is true. | |
1668 | * | |
1669 | * CONTEXT: | |
1670 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1671 | * multiple times. Does GFP_KERNEL allocations. | |
1672 | * | |
1673 | * RETURNS: | |
1674 | * false if no action was taken and gcwq->lock stayed locked, true if | |
1675 | * some action was taken. | |
1676 | */ | |
1677 | static bool manage_workers(struct worker *worker) | |
1678 | { | |
63d95a91 TH |
1679 | struct worker_pool *pool = worker->pool; |
1680 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
1681 | bool ret = false; |
1682 | ||
1683 | if (gcwq->flags & GCWQ_MANAGING_WORKERS) | |
1684 | return ret; | |
1685 | ||
1686 | gcwq->flags &= ~GCWQ_MANAGE_WORKERS; | |
1687 | gcwq->flags |= GCWQ_MANAGING_WORKERS; | |
1688 | ||
1689 | /* | |
1690 | * Destroy and then create so that may_start_working() is true | |
1691 | * on return. | |
1692 | */ | |
63d95a91 TH |
1693 | ret |= maybe_destroy_workers(pool); |
1694 | ret |= maybe_create_worker(pool); | |
e22bee78 TH |
1695 | |
1696 | gcwq->flags &= ~GCWQ_MANAGING_WORKERS; | |
1697 | ||
1698 | /* | |
1699 | * The trustee might be waiting to take over the manager | |
1700 | * position, tell it we're done. | |
1701 | */ | |
1702 | if (unlikely(gcwq->trustee)) | |
1703 | wake_up_all(&gcwq->trustee_wait); | |
1704 | ||
1705 | return ret; | |
1706 | } | |
1707 | ||
affee4b2 TH |
1708 | /** |
1709 | * move_linked_works - move linked works to a list | |
1710 | * @work: start of series of works to be scheduled | |
1711 | * @head: target list to append @work to | |
1712 | * @nextp: out paramter for nested worklist walking | |
1713 | * | |
1714 | * Schedule linked works starting from @work to @head. Work series to | |
1715 | * be scheduled starts at @work and includes any consecutive work with | |
1716 | * WORK_STRUCT_LINKED set in its predecessor. | |
1717 | * | |
1718 | * If @nextp is not NULL, it's updated to point to the next work of | |
1719 | * the last scheduled work. This allows move_linked_works() to be | |
1720 | * nested inside outer list_for_each_entry_safe(). | |
1721 | * | |
1722 | * CONTEXT: | |
8b03ae3c | 1723 | * spin_lock_irq(gcwq->lock). |
affee4b2 TH |
1724 | */ |
1725 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
1726 | struct work_struct **nextp) | |
1727 | { | |
1728 | struct work_struct *n; | |
1729 | ||
1730 | /* | |
1731 | * Linked worklist will always end before the end of the list, | |
1732 | * use NULL for list head. | |
1733 | */ | |
1734 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
1735 | list_move_tail(&work->entry, head); | |
1736 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
1737 | break; | |
1738 | } | |
1739 | ||
1740 | /* | |
1741 | * If we're already inside safe list traversal and have moved | |
1742 | * multiple works to the scheduled queue, the next position | |
1743 | * needs to be updated. | |
1744 | */ | |
1745 | if (nextp) | |
1746 | *nextp = n; | |
1747 | } | |
1748 | ||
1e19ffc6 TH |
1749 | static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq) |
1750 | { | |
1751 | struct work_struct *work = list_first_entry(&cwq->delayed_works, | |
1752 | struct work_struct, entry); | |
63d95a91 | 1753 | struct list_head *pos = pool_determine_ins_pos(cwq->pool, cwq); |
1e19ffc6 | 1754 | |
cdadf009 | 1755 | trace_workqueue_activate_work(work); |
649027d7 | 1756 | move_linked_works(work, pos, NULL); |
8a2e8e5d | 1757 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); |
1e19ffc6 TH |
1758 | cwq->nr_active++; |
1759 | } | |
1760 | ||
73f53c4a TH |
1761 | /** |
1762 | * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight | |
1763 | * @cwq: cwq of interest | |
1764 | * @color: color of work which left the queue | |
8a2e8e5d | 1765 | * @delayed: for a delayed work |
73f53c4a TH |
1766 | * |
1767 | * A work either has completed or is removed from pending queue, | |
1768 | * decrement nr_in_flight of its cwq and handle workqueue flushing. | |
1769 | * | |
1770 | * CONTEXT: | |
8b03ae3c | 1771 | * spin_lock_irq(gcwq->lock). |
73f53c4a | 1772 | */ |
8a2e8e5d TH |
1773 | static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color, |
1774 | bool delayed) | |
73f53c4a TH |
1775 | { |
1776 | /* ignore uncolored works */ | |
1777 | if (color == WORK_NO_COLOR) | |
1778 | return; | |
1779 | ||
1780 | cwq->nr_in_flight[color]--; | |
1e19ffc6 | 1781 | |
8a2e8e5d TH |
1782 | if (!delayed) { |
1783 | cwq->nr_active--; | |
1784 | if (!list_empty(&cwq->delayed_works)) { | |
1785 | /* one down, submit a delayed one */ | |
1786 | if (cwq->nr_active < cwq->max_active) | |
1787 | cwq_activate_first_delayed(cwq); | |
1788 | } | |
502ca9d8 | 1789 | } |
73f53c4a TH |
1790 | |
1791 | /* is flush in progress and are we at the flushing tip? */ | |
1792 | if (likely(cwq->flush_color != color)) | |
1793 | return; | |
1794 | ||
1795 | /* are there still in-flight works? */ | |
1796 | if (cwq->nr_in_flight[color]) | |
1797 | return; | |
1798 | ||
1799 | /* this cwq is done, clear flush_color */ | |
1800 | cwq->flush_color = -1; | |
1801 | ||
1802 | /* | |
1803 | * If this was the last cwq, wake up the first flusher. It | |
1804 | * will handle the rest. | |
1805 | */ | |
1806 | if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush)) | |
1807 | complete(&cwq->wq->first_flusher->done); | |
1808 | } | |
1809 | ||
a62428c0 TH |
1810 | /** |
1811 | * process_one_work - process single work | |
c34056a3 | 1812 | * @worker: self |
a62428c0 TH |
1813 | * @work: work to process |
1814 | * | |
1815 | * Process @work. This function contains all the logics necessary to | |
1816 | * process a single work including synchronization against and | |
1817 | * interaction with other workers on the same cpu, queueing and | |
1818 | * flushing. As long as context requirement is met, any worker can | |
1819 | * call this function to process a work. | |
1820 | * | |
1821 | * CONTEXT: | |
8b03ae3c | 1822 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. |
a62428c0 | 1823 | */ |
c34056a3 | 1824 | static void process_one_work(struct worker *worker, struct work_struct *work) |
06bd6ebf NK |
1825 | __releases(&gcwq->lock) |
1826 | __acquires(&gcwq->lock) | |
a62428c0 | 1827 | { |
7e11629d | 1828 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); |
bd7bdd43 TH |
1829 | struct worker_pool *pool = worker->pool; |
1830 | struct global_cwq *gcwq = pool->gcwq; | |
c8e55f36 | 1831 | struct hlist_head *bwh = busy_worker_head(gcwq, work); |
fb0e7beb | 1832 | bool cpu_intensive = cwq->wq->flags & WQ_CPU_INTENSIVE; |
a62428c0 | 1833 | work_func_t f = work->func; |
73f53c4a | 1834 | int work_color; |
7e11629d | 1835 | struct worker *collision; |
a62428c0 TH |
1836 | #ifdef CONFIG_LOCKDEP |
1837 | /* | |
1838 | * It is permissible to free the struct work_struct from | |
1839 | * inside the function that is called from it, this we need to | |
1840 | * take into account for lockdep too. To avoid bogus "held | |
1841 | * lock freed" warnings as well as problems when looking into | |
1842 | * work->lockdep_map, make a copy and use that here. | |
1843 | */ | |
4d82a1de PZ |
1844 | struct lockdep_map lockdep_map; |
1845 | ||
1846 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 1847 | #endif |
7e11629d TH |
1848 | /* |
1849 | * A single work shouldn't be executed concurrently by | |
1850 | * multiple workers on a single cpu. Check whether anyone is | |
1851 | * already processing the work. If so, defer the work to the | |
1852 | * currently executing one. | |
1853 | */ | |
1854 | collision = __find_worker_executing_work(gcwq, bwh, work); | |
1855 | if (unlikely(collision)) { | |
1856 | move_linked_works(work, &collision->scheduled, NULL); | |
1857 | return; | |
1858 | } | |
1859 | ||
a62428c0 | 1860 | /* claim and process */ |
a62428c0 | 1861 | debug_work_deactivate(work); |
c8e55f36 | 1862 | hlist_add_head(&worker->hentry, bwh); |
c34056a3 | 1863 | worker->current_work = work; |
8cca0eea | 1864 | worker->current_cwq = cwq; |
73f53c4a | 1865 | work_color = get_work_color(work); |
7a22ad75 | 1866 | |
7a22ad75 TH |
1867 | /* record the current cpu number in the work data and dequeue */ |
1868 | set_work_cpu(work, gcwq->cpu); | |
a62428c0 TH |
1869 | list_del_init(&work->entry); |
1870 | ||
649027d7 TH |
1871 | /* |
1872 | * If HIGHPRI_PENDING, check the next work, and, if HIGHPRI, | |
1873 | * wake up another worker; otherwise, clear HIGHPRI_PENDING. | |
1874 | */ | |
1875 | if (unlikely(gcwq->flags & GCWQ_HIGHPRI_PENDING)) { | |
bd7bdd43 TH |
1876 | struct work_struct *nwork = list_first_entry(&pool->worklist, |
1877 | struct work_struct, entry); | |
649027d7 | 1878 | |
bd7bdd43 | 1879 | if (!list_empty(&pool->worklist) && |
649027d7 | 1880 | get_work_cwq(nwork)->wq->flags & WQ_HIGHPRI) |
63d95a91 | 1881 | wake_up_worker(pool); |
649027d7 TH |
1882 | else |
1883 | gcwq->flags &= ~GCWQ_HIGHPRI_PENDING; | |
1884 | } | |
1885 | ||
fb0e7beb TH |
1886 | /* |
1887 | * CPU intensive works don't participate in concurrency | |
1888 | * management. They're the scheduler's responsibility. | |
1889 | */ | |
1890 | if (unlikely(cpu_intensive)) | |
1891 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
1892 | ||
974271c4 TH |
1893 | /* |
1894 | * Unbound gcwq isn't concurrency managed and work items should be | |
1895 | * executed ASAP. Wake up another worker if necessary. | |
1896 | */ | |
63d95a91 TH |
1897 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
1898 | wake_up_worker(pool); | |
974271c4 | 1899 | |
8b03ae3c | 1900 | spin_unlock_irq(&gcwq->lock); |
a62428c0 | 1901 | |
a62428c0 | 1902 | work_clear_pending(work); |
e159489b | 1903 | lock_map_acquire_read(&cwq->wq->lockdep_map); |
a62428c0 | 1904 | lock_map_acquire(&lockdep_map); |
e36c886a | 1905 | trace_workqueue_execute_start(work); |
a62428c0 | 1906 | f(work); |
e36c886a AV |
1907 | /* |
1908 | * While we must be careful to not use "work" after this, the trace | |
1909 | * point will only record its address. | |
1910 | */ | |
1911 | trace_workqueue_execute_end(work); | |
a62428c0 TH |
1912 | lock_map_release(&lockdep_map); |
1913 | lock_map_release(&cwq->wq->lockdep_map); | |
1914 | ||
1915 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
1916 | printk(KERN_ERR "BUG: workqueue leaked lock or atomic: " | |
1917 | "%s/0x%08x/%d\n", | |
1918 | current->comm, preempt_count(), task_pid_nr(current)); | |
1919 | printk(KERN_ERR " last function: "); | |
1920 | print_symbol("%s\n", (unsigned long)f); | |
1921 | debug_show_held_locks(current); | |
1922 | dump_stack(); | |
1923 | } | |
1924 | ||
8b03ae3c | 1925 | spin_lock_irq(&gcwq->lock); |
a62428c0 | 1926 | |
fb0e7beb TH |
1927 | /* clear cpu intensive status */ |
1928 | if (unlikely(cpu_intensive)) | |
1929 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
1930 | ||
a62428c0 | 1931 | /* we're done with it, release */ |
c8e55f36 | 1932 | hlist_del_init(&worker->hentry); |
c34056a3 | 1933 | worker->current_work = NULL; |
8cca0eea | 1934 | worker->current_cwq = NULL; |
8a2e8e5d | 1935 | cwq_dec_nr_in_flight(cwq, work_color, false); |
a62428c0 TH |
1936 | } |
1937 | ||
affee4b2 TH |
1938 | /** |
1939 | * process_scheduled_works - process scheduled works | |
1940 | * @worker: self | |
1941 | * | |
1942 | * Process all scheduled works. Please note that the scheduled list | |
1943 | * may change while processing a work, so this function repeatedly | |
1944 | * fetches a work from the top and executes it. | |
1945 | * | |
1946 | * CONTEXT: | |
8b03ae3c | 1947 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed |
affee4b2 TH |
1948 | * multiple times. |
1949 | */ | |
1950 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 1951 | { |
affee4b2 TH |
1952 | while (!list_empty(&worker->scheduled)) { |
1953 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 1954 | struct work_struct, entry); |
c34056a3 | 1955 | process_one_work(worker, work); |
1da177e4 | 1956 | } |
1da177e4 LT |
1957 | } |
1958 | ||
4690c4ab TH |
1959 | /** |
1960 | * worker_thread - the worker thread function | |
c34056a3 | 1961 | * @__worker: self |
4690c4ab | 1962 | * |
e22bee78 TH |
1963 | * The gcwq worker thread function. There's a single dynamic pool of |
1964 | * these per each cpu. These workers process all works regardless of | |
1965 | * their specific target workqueue. The only exception is works which | |
1966 | * belong to workqueues with a rescuer which will be explained in | |
1967 | * rescuer_thread(). | |
4690c4ab | 1968 | */ |
c34056a3 | 1969 | static int worker_thread(void *__worker) |
1da177e4 | 1970 | { |
c34056a3 | 1971 | struct worker *worker = __worker; |
bd7bdd43 TH |
1972 | struct worker_pool *pool = worker->pool; |
1973 | struct global_cwq *gcwq = pool->gcwq; | |
1da177e4 | 1974 | |
e22bee78 TH |
1975 | /* tell the scheduler that this is a workqueue worker */ |
1976 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 1977 | woke_up: |
c8e55f36 | 1978 | spin_lock_irq(&gcwq->lock); |
1da177e4 | 1979 | |
c8e55f36 TH |
1980 | /* DIE can be set only while we're idle, checking here is enough */ |
1981 | if (worker->flags & WORKER_DIE) { | |
1982 | spin_unlock_irq(&gcwq->lock); | |
e22bee78 | 1983 | worker->task->flags &= ~PF_WQ_WORKER; |
c8e55f36 TH |
1984 | return 0; |
1985 | } | |
affee4b2 | 1986 | |
c8e55f36 | 1987 | worker_leave_idle(worker); |
db7bccf4 | 1988 | recheck: |
e22bee78 | 1989 | /* no more worker necessary? */ |
63d95a91 | 1990 | if (!need_more_worker(pool)) |
e22bee78 TH |
1991 | goto sleep; |
1992 | ||
1993 | /* do we need to manage? */ | |
63d95a91 | 1994 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
1995 | goto recheck; |
1996 | ||
c8e55f36 TH |
1997 | /* |
1998 | * ->scheduled list can only be filled while a worker is | |
1999 | * preparing to process a work or actually processing it. | |
2000 | * Make sure nobody diddled with it while I was sleeping. | |
2001 | */ | |
2002 | BUG_ON(!list_empty(&worker->scheduled)); | |
2003 | ||
e22bee78 TH |
2004 | /* |
2005 | * When control reaches this point, we're guaranteed to have | |
2006 | * at least one idle worker or that someone else has already | |
2007 | * assumed the manager role. | |
2008 | */ | |
2009 | worker_clr_flags(worker, WORKER_PREP); | |
2010 | ||
2011 | do { | |
c8e55f36 | 2012 | struct work_struct *work = |
bd7bdd43 | 2013 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2014 | struct work_struct, entry); |
2015 | ||
2016 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
2017 | /* optimization path, not strictly necessary */ | |
2018 | process_one_work(worker, work); | |
2019 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2020 | process_scheduled_works(worker); |
c8e55f36 TH |
2021 | } else { |
2022 | move_linked_works(work, &worker->scheduled, NULL); | |
2023 | process_scheduled_works(worker); | |
affee4b2 | 2024 | } |
63d95a91 | 2025 | } while (keep_working(pool)); |
e22bee78 TH |
2026 | |
2027 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 2028 | sleep: |
63d95a91 | 2029 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
e22bee78 | 2030 | goto recheck; |
d313dd85 | 2031 | |
c8e55f36 | 2032 | /* |
e22bee78 TH |
2033 | * gcwq->lock is held and there's no work to process and no |
2034 | * need to manage, sleep. Workers are woken up only while | |
2035 | * holding gcwq->lock or from local cpu, so setting the | |
2036 | * current state before releasing gcwq->lock is enough to | |
2037 | * prevent losing any event. | |
c8e55f36 TH |
2038 | */ |
2039 | worker_enter_idle(worker); | |
2040 | __set_current_state(TASK_INTERRUPTIBLE); | |
2041 | spin_unlock_irq(&gcwq->lock); | |
2042 | schedule(); | |
2043 | goto woke_up; | |
1da177e4 LT |
2044 | } |
2045 | ||
e22bee78 TH |
2046 | /** |
2047 | * rescuer_thread - the rescuer thread function | |
2048 | * @__wq: the associated workqueue | |
2049 | * | |
2050 | * Workqueue rescuer thread function. There's one rescuer for each | |
2051 | * workqueue which has WQ_RESCUER set. | |
2052 | * | |
2053 | * Regular work processing on a gcwq may block trying to create a new | |
2054 | * worker which uses GFP_KERNEL allocation which has slight chance of | |
2055 | * developing into deadlock if some works currently on the same queue | |
2056 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2057 | * the problem rescuer solves. | |
2058 | * | |
2059 | * When such condition is possible, the gcwq summons rescuers of all | |
2060 | * workqueues which have works queued on the gcwq and let them process | |
2061 | * those works so that forward progress can be guaranteed. | |
2062 | * | |
2063 | * This should happen rarely. | |
2064 | */ | |
2065 | static int rescuer_thread(void *__wq) | |
2066 | { | |
2067 | struct workqueue_struct *wq = __wq; | |
2068 | struct worker *rescuer = wq->rescuer; | |
2069 | struct list_head *scheduled = &rescuer->scheduled; | |
f3421797 | 2070 | bool is_unbound = wq->flags & WQ_UNBOUND; |
e22bee78 TH |
2071 | unsigned int cpu; |
2072 | ||
2073 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
2074 | repeat: | |
2075 | set_current_state(TASK_INTERRUPTIBLE); | |
2076 | ||
2077 | if (kthread_should_stop()) | |
2078 | return 0; | |
2079 | ||
f3421797 TH |
2080 | /* |
2081 | * See whether any cpu is asking for help. Unbounded | |
2082 | * workqueues use cpu 0 in mayday_mask for CPU_UNBOUND. | |
2083 | */ | |
f2e005aa | 2084 | for_each_mayday_cpu(cpu, wq->mayday_mask) { |
f3421797 TH |
2085 | unsigned int tcpu = is_unbound ? WORK_CPU_UNBOUND : cpu; |
2086 | struct cpu_workqueue_struct *cwq = get_cwq(tcpu, wq); | |
bd7bdd43 TH |
2087 | struct worker_pool *pool = cwq->pool; |
2088 | struct global_cwq *gcwq = pool->gcwq; | |
e22bee78 TH |
2089 | struct work_struct *work, *n; |
2090 | ||
2091 | __set_current_state(TASK_RUNNING); | |
f2e005aa | 2092 | mayday_clear_cpu(cpu, wq->mayday_mask); |
e22bee78 TH |
2093 | |
2094 | /* migrate to the target cpu if possible */ | |
bd7bdd43 | 2095 | rescuer->pool = pool; |
e22bee78 TH |
2096 | worker_maybe_bind_and_lock(rescuer); |
2097 | ||
2098 | /* | |
2099 | * Slurp in all works issued via this workqueue and | |
2100 | * process'em. | |
2101 | */ | |
2102 | BUG_ON(!list_empty(&rescuer->scheduled)); | |
bd7bdd43 | 2103 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
e22bee78 TH |
2104 | if (get_work_cwq(work) == cwq) |
2105 | move_linked_works(work, scheduled, &n); | |
2106 | ||
2107 | process_scheduled_works(rescuer); | |
7576958a TH |
2108 | |
2109 | /* | |
2110 | * Leave this gcwq. If keep_working() is %true, notify a | |
2111 | * regular worker; otherwise, we end up with 0 concurrency | |
2112 | * and stalling the execution. | |
2113 | */ | |
63d95a91 TH |
2114 | if (keep_working(pool)) |
2115 | wake_up_worker(pool); | |
7576958a | 2116 | |
e22bee78 TH |
2117 | spin_unlock_irq(&gcwq->lock); |
2118 | } | |
2119 | ||
2120 | schedule(); | |
2121 | goto repeat; | |
1da177e4 LT |
2122 | } |
2123 | ||
fc2e4d70 ON |
2124 | struct wq_barrier { |
2125 | struct work_struct work; | |
2126 | struct completion done; | |
2127 | }; | |
2128 | ||
2129 | static void wq_barrier_func(struct work_struct *work) | |
2130 | { | |
2131 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2132 | complete(&barr->done); | |
2133 | } | |
2134 | ||
4690c4ab TH |
2135 | /** |
2136 | * insert_wq_barrier - insert a barrier work | |
2137 | * @cwq: cwq to insert barrier into | |
2138 | * @barr: wq_barrier to insert | |
affee4b2 TH |
2139 | * @target: target work to attach @barr to |
2140 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2141 | * |
affee4b2 TH |
2142 | * @barr is linked to @target such that @barr is completed only after |
2143 | * @target finishes execution. Please note that the ordering | |
2144 | * guarantee is observed only with respect to @target and on the local | |
2145 | * cpu. | |
2146 | * | |
2147 | * Currently, a queued barrier can't be canceled. This is because | |
2148 | * try_to_grab_pending() can't determine whether the work to be | |
2149 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2150 | * flag of the previous work while there must be a valid next work | |
2151 | * after a work with LINKED flag set. | |
2152 | * | |
2153 | * Note that when @worker is non-NULL, @target may be modified | |
2154 | * underneath us, so we can't reliably determine cwq from @target. | |
4690c4ab TH |
2155 | * |
2156 | * CONTEXT: | |
8b03ae3c | 2157 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 2158 | */ |
83c22520 | 2159 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, |
affee4b2 TH |
2160 | struct wq_barrier *barr, |
2161 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2162 | { |
affee4b2 TH |
2163 | struct list_head *head; |
2164 | unsigned int linked = 0; | |
2165 | ||
dc186ad7 | 2166 | /* |
8b03ae3c | 2167 | * debugobject calls are safe here even with gcwq->lock locked |
dc186ad7 TG |
2168 | * as we know for sure that this will not trigger any of the |
2169 | * checks and call back into the fixup functions where we | |
2170 | * might deadlock. | |
2171 | */ | |
ca1cab37 | 2172 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2173 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2174 | init_completion(&barr->done); |
83c22520 | 2175 | |
affee4b2 TH |
2176 | /* |
2177 | * If @target is currently being executed, schedule the | |
2178 | * barrier to the worker; otherwise, put it after @target. | |
2179 | */ | |
2180 | if (worker) | |
2181 | head = worker->scheduled.next; | |
2182 | else { | |
2183 | unsigned long *bits = work_data_bits(target); | |
2184 | ||
2185 | head = target->entry.next; | |
2186 | /* there can already be other linked works, inherit and set */ | |
2187 | linked = *bits & WORK_STRUCT_LINKED; | |
2188 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2189 | } | |
2190 | ||
dc186ad7 | 2191 | debug_work_activate(&barr->work); |
affee4b2 TH |
2192 | insert_work(cwq, &barr->work, head, |
2193 | work_color_to_flags(WORK_NO_COLOR) | linked); | |
fc2e4d70 ON |
2194 | } |
2195 | ||
73f53c4a TH |
2196 | /** |
2197 | * flush_workqueue_prep_cwqs - prepare cwqs for workqueue flushing | |
2198 | * @wq: workqueue being flushed | |
2199 | * @flush_color: new flush color, < 0 for no-op | |
2200 | * @work_color: new work color, < 0 for no-op | |
2201 | * | |
2202 | * Prepare cwqs for workqueue flushing. | |
2203 | * | |
2204 | * If @flush_color is non-negative, flush_color on all cwqs should be | |
2205 | * -1. If no cwq has in-flight commands at the specified color, all | |
2206 | * cwq->flush_color's stay at -1 and %false is returned. If any cwq | |
2207 | * has in flight commands, its cwq->flush_color is set to | |
2208 | * @flush_color, @wq->nr_cwqs_to_flush is updated accordingly, cwq | |
2209 | * wakeup logic is armed and %true is returned. | |
2210 | * | |
2211 | * The caller should have initialized @wq->first_flusher prior to | |
2212 | * calling this function with non-negative @flush_color. If | |
2213 | * @flush_color is negative, no flush color update is done and %false | |
2214 | * is returned. | |
2215 | * | |
2216 | * If @work_color is non-negative, all cwqs should have the same | |
2217 | * work_color which is previous to @work_color and all will be | |
2218 | * advanced to @work_color. | |
2219 | * | |
2220 | * CONTEXT: | |
2221 | * mutex_lock(wq->flush_mutex). | |
2222 | * | |
2223 | * RETURNS: | |
2224 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2225 | * otherwise. | |
2226 | */ | |
2227 | static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq, | |
2228 | int flush_color, int work_color) | |
1da177e4 | 2229 | { |
73f53c4a TH |
2230 | bool wait = false; |
2231 | unsigned int cpu; | |
1da177e4 | 2232 | |
73f53c4a TH |
2233 | if (flush_color >= 0) { |
2234 | BUG_ON(atomic_read(&wq->nr_cwqs_to_flush)); | |
2235 | atomic_set(&wq->nr_cwqs_to_flush, 1); | |
1da177e4 | 2236 | } |
2355b70f | 2237 | |
f3421797 | 2238 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a | 2239 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
bd7bdd43 | 2240 | struct global_cwq *gcwq = cwq->pool->gcwq; |
fc2e4d70 | 2241 | |
8b03ae3c | 2242 | spin_lock_irq(&gcwq->lock); |
83c22520 | 2243 | |
73f53c4a TH |
2244 | if (flush_color >= 0) { |
2245 | BUG_ON(cwq->flush_color != -1); | |
fc2e4d70 | 2246 | |
73f53c4a TH |
2247 | if (cwq->nr_in_flight[flush_color]) { |
2248 | cwq->flush_color = flush_color; | |
2249 | atomic_inc(&wq->nr_cwqs_to_flush); | |
2250 | wait = true; | |
2251 | } | |
2252 | } | |
1da177e4 | 2253 | |
73f53c4a TH |
2254 | if (work_color >= 0) { |
2255 | BUG_ON(work_color != work_next_color(cwq->work_color)); | |
2256 | cwq->work_color = work_color; | |
2257 | } | |
1da177e4 | 2258 | |
8b03ae3c | 2259 | spin_unlock_irq(&gcwq->lock); |
1da177e4 | 2260 | } |
2355b70f | 2261 | |
73f53c4a TH |
2262 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_cwqs_to_flush)) |
2263 | complete(&wq->first_flusher->done); | |
14441960 | 2264 | |
73f53c4a | 2265 | return wait; |
1da177e4 LT |
2266 | } |
2267 | ||
0fcb78c2 | 2268 | /** |
1da177e4 | 2269 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2270 | * @wq: workqueue to flush |
1da177e4 LT |
2271 | * |
2272 | * Forces execution of the workqueue and blocks until its completion. | |
2273 | * This is typically used in driver shutdown handlers. | |
2274 | * | |
fc2e4d70 ON |
2275 | * We sleep until all works which were queued on entry have been handled, |
2276 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 2277 | */ |
7ad5b3a5 | 2278 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2279 | { |
73f53c4a TH |
2280 | struct wq_flusher this_flusher = { |
2281 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2282 | .flush_color = -1, | |
2283 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2284 | }; | |
2285 | int next_color; | |
1da177e4 | 2286 | |
3295f0ef IM |
2287 | lock_map_acquire(&wq->lockdep_map); |
2288 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
2289 | |
2290 | mutex_lock(&wq->flush_mutex); | |
2291 | ||
2292 | /* | |
2293 | * Start-to-wait phase | |
2294 | */ | |
2295 | next_color = work_next_color(wq->work_color); | |
2296 | ||
2297 | if (next_color != wq->flush_color) { | |
2298 | /* | |
2299 | * Color space is not full. The current work_color | |
2300 | * becomes our flush_color and work_color is advanced | |
2301 | * by one. | |
2302 | */ | |
2303 | BUG_ON(!list_empty(&wq->flusher_overflow)); | |
2304 | this_flusher.flush_color = wq->work_color; | |
2305 | wq->work_color = next_color; | |
2306 | ||
2307 | if (!wq->first_flusher) { | |
2308 | /* no flush in progress, become the first flusher */ | |
2309 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2310 | ||
2311 | wq->first_flusher = &this_flusher; | |
2312 | ||
2313 | if (!flush_workqueue_prep_cwqs(wq, wq->flush_color, | |
2314 | wq->work_color)) { | |
2315 | /* nothing to flush, done */ | |
2316 | wq->flush_color = next_color; | |
2317 | wq->first_flusher = NULL; | |
2318 | goto out_unlock; | |
2319 | } | |
2320 | } else { | |
2321 | /* wait in queue */ | |
2322 | BUG_ON(wq->flush_color == this_flusher.flush_color); | |
2323 | list_add_tail(&this_flusher.list, &wq->flusher_queue); | |
2324 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2325 | } | |
2326 | } else { | |
2327 | /* | |
2328 | * Oops, color space is full, wait on overflow queue. | |
2329 | * The next flush completion will assign us | |
2330 | * flush_color and transfer to flusher_queue. | |
2331 | */ | |
2332 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2333 | } | |
2334 | ||
2335 | mutex_unlock(&wq->flush_mutex); | |
2336 | ||
2337 | wait_for_completion(&this_flusher.done); | |
2338 | ||
2339 | /* | |
2340 | * Wake-up-and-cascade phase | |
2341 | * | |
2342 | * First flushers are responsible for cascading flushes and | |
2343 | * handling overflow. Non-first flushers can simply return. | |
2344 | */ | |
2345 | if (wq->first_flusher != &this_flusher) | |
2346 | return; | |
2347 | ||
2348 | mutex_lock(&wq->flush_mutex); | |
2349 | ||
4ce48b37 TH |
2350 | /* we might have raced, check again with mutex held */ |
2351 | if (wq->first_flusher != &this_flusher) | |
2352 | goto out_unlock; | |
2353 | ||
73f53c4a TH |
2354 | wq->first_flusher = NULL; |
2355 | ||
2356 | BUG_ON(!list_empty(&this_flusher.list)); | |
2357 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2358 | ||
2359 | while (true) { | |
2360 | struct wq_flusher *next, *tmp; | |
2361 | ||
2362 | /* complete all the flushers sharing the current flush color */ | |
2363 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2364 | if (next->flush_color != wq->flush_color) | |
2365 | break; | |
2366 | list_del_init(&next->list); | |
2367 | complete(&next->done); | |
2368 | } | |
2369 | ||
2370 | BUG_ON(!list_empty(&wq->flusher_overflow) && | |
2371 | wq->flush_color != work_next_color(wq->work_color)); | |
2372 | ||
2373 | /* this flush_color is finished, advance by one */ | |
2374 | wq->flush_color = work_next_color(wq->flush_color); | |
2375 | ||
2376 | /* one color has been freed, handle overflow queue */ | |
2377 | if (!list_empty(&wq->flusher_overflow)) { | |
2378 | /* | |
2379 | * Assign the same color to all overflowed | |
2380 | * flushers, advance work_color and append to | |
2381 | * flusher_queue. This is the start-to-wait | |
2382 | * phase for these overflowed flushers. | |
2383 | */ | |
2384 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2385 | tmp->flush_color = wq->work_color; | |
2386 | ||
2387 | wq->work_color = work_next_color(wq->work_color); | |
2388 | ||
2389 | list_splice_tail_init(&wq->flusher_overflow, | |
2390 | &wq->flusher_queue); | |
2391 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2392 | } | |
2393 | ||
2394 | if (list_empty(&wq->flusher_queue)) { | |
2395 | BUG_ON(wq->flush_color != wq->work_color); | |
2396 | break; | |
2397 | } | |
2398 | ||
2399 | /* | |
2400 | * Need to flush more colors. Make the next flusher | |
2401 | * the new first flusher and arm cwqs. | |
2402 | */ | |
2403 | BUG_ON(wq->flush_color == wq->work_color); | |
2404 | BUG_ON(wq->flush_color != next->flush_color); | |
2405 | ||
2406 | list_del_init(&next->list); | |
2407 | wq->first_flusher = next; | |
2408 | ||
2409 | if (flush_workqueue_prep_cwqs(wq, wq->flush_color, -1)) | |
2410 | break; | |
2411 | ||
2412 | /* | |
2413 | * Meh... this color is already done, clear first | |
2414 | * flusher and repeat cascading. | |
2415 | */ | |
2416 | wq->first_flusher = NULL; | |
2417 | } | |
2418 | ||
2419 | out_unlock: | |
2420 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 2421 | } |
ae90dd5d | 2422 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2423 | |
9c5a2ba7 TH |
2424 | /** |
2425 | * drain_workqueue - drain a workqueue | |
2426 | * @wq: workqueue to drain | |
2427 | * | |
2428 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2429 | * only chain queueing is allowed. IOW, only currently pending or running | |
2430 | * work items on @wq can queue further work items on it. @wq is flushed | |
2431 | * repeatedly until it becomes empty. The number of flushing is detemined | |
2432 | * by the depth of chaining and should be relatively short. Whine if it | |
2433 | * takes too long. | |
2434 | */ | |
2435 | void drain_workqueue(struct workqueue_struct *wq) | |
2436 | { | |
2437 | unsigned int flush_cnt = 0; | |
2438 | unsigned int cpu; | |
2439 | ||
2440 | /* | |
2441 | * __queue_work() needs to test whether there are drainers, is much | |
2442 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
2443 | * Use WQ_DRAINING so that queue doesn't have to check nr_drainers. | |
2444 | */ | |
2445 | spin_lock(&workqueue_lock); | |
2446 | if (!wq->nr_drainers++) | |
2447 | wq->flags |= WQ_DRAINING; | |
2448 | spin_unlock(&workqueue_lock); | |
2449 | reflush: | |
2450 | flush_workqueue(wq); | |
2451 | ||
2452 | for_each_cwq_cpu(cpu, wq) { | |
2453 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
fa2563e4 | 2454 | bool drained; |
9c5a2ba7 | 2455 | |
bd7bdd43 | 2456 | spin_lock_irq(&cwq->pool->gcwq->lock); |
fa2563e4 | 2457 | drained = !cwq->nr_active && list_empty(&cwq->delayed_works); |
bd7bdd43 | 2458 | spin_unlock_irq(&cwq->pool->gcwq->lock); |
fa2563e4 TT |
2459 | |
2460 | if (drained) | |
9c5a2ba7 TH |
2461 | continue; |
2462 | ||
2463 | if (++flush_cnt == 10 || | |
2464 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
2465 | pr_warning("workqueue %s: flush on destruction isn't complete after %u tries\n", | |
2466 | wq->name, flush_cnt); | |
2467 | goto reflush; | |
2468 | } | |
2469 | ||
2470 | spin_lock(&workqueue_lock); | |
2471 | if (!--wq->nr_drainers) | |
2472 | wq->flags &= ~WQ_DRAINING; | |
2473 | spin_unlock(&workqueue_lock); | |
2474 | } | |
2475 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2476 | ||
baf59022 TH |
2477 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr, |
2478 | bool wait_executing) | |
db700897 | 2479 | { |
affee4b2 | 2480 | struct worker *worker = NULL; |
8b03ae3c | 2481 | struct global_cwq *gcwq; |
db700897 | 2482 | struct cpu_workqueue_struct *cwq; |
db700897 ON |
2483 | |
2484 | might_sleep(); | |
7a22ad75 TH |
2485 | gcwq = get_work_gcwq(work); |
2486 | if (!gcwq) | |
baf59022 | 2487 | return false; |
db700897 | 2488 | |
8b03ae3c | 2489 | spin_lock_irq(&gcwq->lock); |
db700897 ON |
2490 | if (!list_empty(&work->entry)) { |
2491 | /* | |
2492 | * See the comment near try_to_grab_pending()->smp_rmb(). | |
7a22ad75 TH |
2493 | * If it was re-queued to a different gcwq under us, we |
2494 | * are not going to wait. | |
db700897 ON |
2495 | */ |
2496 | smp_rmb(); | |
7a22ad75 | 2497 | cwq = get_work_cwq(work); |
bd7bdd43 | 2498 | if (unlikely(!cwq || gcwq != cwq->pool->gcwq)) |
4690c4ab | 2499 | goto already_gone; |
baf59022 | 2500 | } else if (wait_executing) { |
7a22ad75 | 2501 | worker = find_worker_executing_work(gcwq, work); |
affee4b2 | 2502 | if (!worker) |
4690c4ab | 2503 | goto already_gone; |
7a22ad75 | 2504 | cwq = worker->current_cwq; |
baf59022 TH |
2505 | } else |
2506 | goto already_gone; | |
db700897 | 2507 | |
baf59022 | 2508 | insert_wq_barrier(cwq, barr, work, worker); |
8b03ae3c | 2509 | spin_unlock_irq(&gcwq->lock); |
7a22ad75 | 2510 | |
e159489b TH |
2511 | /* |
2512 | * If @max_active is 1 or rescuer is in use, flushing another work | |
2513 | * item on the same workqueue may lead to deadlock. Make sure the | |
2514 | * flusher is not running on the same workqueue by verifying write | |
2515 | * access. | |
2516 | */ | |
2517 | if (cwq->wq->saved_max_active == 1 || cwq->wq->flags & WQ_RESCUER) | |
2518 | lock_map_acquire(&cwq->wq->lockdep_map); | |
2519 | else | |
2520 | lock_map_acquire_read(&cwq->wq->lockdep_map); | |
7a22ad75 | 2521 | lock_map_release(&cwq->wq->lockdep_map); |
e159489b | 2522 | |
401a8d04 | 2523 | return true; |
4690c4ab | 2524 | already_gone: |
8b03ae3c | 2525 | spin_unlock_irq(&gcwq->lock); |
401a8d04 | 2526 | return false; |
db700897 | 2527 | } |
baf59022 TH |
2528 | |
2529 | /** | |
2530 | * flush_work - wait for a work to finish executing the last queueing instance | |
2531 | * @work: the work to flush | |
2532 | * | |
2533 | * Wait until @work has finished execution. This function considers | |
2534 | * only the last queueing instance of @work. If @work has been | |
2535 | * enqueued across different CPUs on a non-reentrant workqueue or on | |
2536 | * multiple workqueues, @work might still be executing on return on | |
2537 | * some of the CPUs from earlier queueing. | |
2538 | * | |
2539 | * If @work was queued only on a non-reentrant, ordered or unbound | |
2540 | * workqueue, @work is guaranteed to be idle on return if it hasn't | |
2541 | * been requeued since flush started. | |
2542 | * | |
2543 | * RETURNS: | |
2544 | * %true if flush_work() waited for the work to finish execution, | |
2545 | * %false if it was already idle. | |
2546 | */ | |
2547 | bool flush_work(struct work_struct *work) | |
2548 | { | |
2549 | struct wq_barrier barr; | |
2550 | ||
0976dfc1 SB |
2551 | lock_map_acquire(&work->lockdep_map); |
2552 | lock_map_release(&work->lockdep_map); | |
2553 | ||
baf59022 TH |
2554 | if (start_flush_work(work, &barr, true)) { |
2555 | wait_for_completion(&barr.done); | |
2556 | destroy_work_on_stack(&barr.work); | |
2557 | return true; | |
2558 | } else | |
2559 | return false; | |
2560 | } | |
db700897 ON |
2561 | EXPORT_SYMBOL_GPL(flush_work); |
2562 | ||
401a8d04 TH |
2563 | static bool wait_on_cpu_work(struct global_cwq *gcwq, struct work_struct *work) |
2564 | { | |
2565 | struct wq_barrier barr; | |
2566 | struct worker *worker; | |
2567 | ||
2568 | spin_lock_irq(&gcwq->lock); | |
2569 | ||
2570 | worker = find_worker_executing_work(gcwq, work); | |
2571 | if (unlikely(worker)) | |
2572 | insert_wq_barrier(worker->current_cwq, &barr, work, worker); | |
2573 | ||
2574 | spin_unlock_irq(&gcwq->lock); | |
2575 | ||
2576 | if (unlikely(worker)) { | |
2577 | wait_for_completion(&barr.done); | |
2578 | destroy_work_on_stack(&barr.work); | |
2579 | return true; | |
2580 | } else | |
2581 | return false; | |
2582 | } | |
2583 | ||
2584 | static bool wait_on_work(struct work_struct *work) | |
2585 | { | |
2586 | bool ret = false; | |
2587 | int cpu; | |
2588 | ||
2589 | might_sleep(); | |
2590 | ||
2591 | lock_map_acquire(&work->lockdep_map); | |
2592 | lock_map_release(&work->lockdep_map); | |
2593 | ||
2594 | for_each_gcwq_cpu(cpu) | |
2595 | ret |= wait_on_cpu_work(get_gcwq(cpu), work); | |
2596 | return ret; | |
2597 | } | |
2598 | ||
09383498 TH |
2599 | /** |
2600 | * flush_work_sync - wait until a work has finished execution | |
2601 | * @work: the work to flush | |
2602 | * | |
2603 | * Wait until @work has finished execution. On return, it's | |
2604 | * guaranteed that all queueing instances of @work which happened | |
2605 | * before this function is called are finished. In other words, if | |
2606 | * @work hasn't been requeued since this function was called, @work is | |
2607 | * guaranteed to be idle on return. | |
2608 | * | |
2609 | * RETURNS: | |
2610 | * %true if flush_work_sync() waited for the work to finish execution, | |
2611 | * %false if it was already idle. | |
2612 | */ | |
2613 | bool flush_work_sync(struct work_struct *work) | |
2614 | { | |
2615 | struct wq_barrier barr; | |
2616 | bool pending, waited; | |
2617 | ||
2618 | /* we'll wait for executions separately, queue barr only if pending */ | |
2619 | pending = start_flush_work(work, &barr, false); | |
2620 | ||
2621 | /* wait for executions to finish */ | |
2622 | waited = wait_on_work(work); | |
2623 | ||
2624 | /* wait for the pending one */ | |
2625 | if (pending) { | |
2626 | wait_for_completion(&barr.done); | |
2627 | destroy_work_on_stack(&barr.work); | |
2628 | } | |
2629 | ||
2630 | return pending || waited; | |
2631 | } | |
2632 | EXPORT_SYMBOL_GPL(flush_work_sync); | |
2633 | ||
6e84d644 | 2634 | /* |
1f1f642e | 2635 | * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit, |
6e84d644 ON |
2636 | * so this work can't be re-armed in any way. |
2637 | */ | |
2638 | static int try_to_grab_pending(struct work_struct *work) | |
2639 | { | |
8b03ae3c | 2640 | struct global_cwq *gcwq; |
1f1f642e | 2641 | int ret = -1; |
6e84d644 | 2642 | |
22df02bb | 2643 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1f1f642e | 2644 | return 0; |
6e84d644 ON |
2645 | |
2646 | /* | |
2647 | * The queueing is in progress, or it is already queued. Try to | |
2648 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
2649 | */ | |
7a22ad75 TH |
2650 | gcwq = get_work_gcwq(work); |
2651 | if (!gcwq) | |
6e84d644 ON |
2652 | return ret; |
2653 | ||
8b03ae3c | 2654 | spin_lock_irq(&gcwq->lock); |
6e84d644 ON |
2655 | if (!list_empty(&work->entry)) { |
2656 | /* | |
7a22ad75 | 2657 | * This work is queued, but perhaps we locked the wrong gcwq. |
6e84d644 ON |
2658 | * In that case we must see the new value after rmb(), see |
2659 | * insert_work()->wmb(). | |
2660 | */ | |
2661 | smp_rmb(); | |
7a22ad75 | 2662 | if (gcwq == get_work_gcwq(work)) { |
dc186ad7 | 2663 | debug_work_deactivate(work); |
6e84d644 | 2664 | list_del_init(&work->entry); |
7a22ad75 | 2665 | cwq_dec_nr_in_flight(get_work_cwq(work), |
8a2e8e5d TH |
2666 | get_work_color(work), |
2667 | *work_data_bits(work) & WORK_STRUCT_DELAYED); | |
6e84d644 ON |
2668 | ret = 1; |
2669 | } | |
2670 | } | |
8b03ae3c | 2671 | spin_unlock_irq(&gcwq->lock); |
6e84d644 ON |
2672 | |
2673 | return ret; | |
2674 | } | |
2675 | ||
401a8d04 | 2676 | static bool __cancel_work_timer(struct work_struct *work, |
1f1f642e ON |
2677 | struct timer_list* timer) |
2678 | { | |
2679 | int ret; | |
2680 | ||
2681 | do { | |
2682 | ret = (timer && likely(del_timer(timer))); | |
2683 | if (!ret) | |
2684 | ret = try_to_grab_pending(work); | |
2685 | wait_on_work(work); | |
2686 | } while (unlikely(ret < 0)); | |
2687 | ||
7a22ad75 | 2688 | clear_work_data(work); |
1f1f642e ON |
2689 | return ret; |
2690 | } | |
2691 | ||
6e84d644 | 2692 | /** |
401a8d04 TH |
2693 | * cancel_work_sync - cancel a work and wait for it to finish |
2694 | * @work: the work to cancel | |
6e84d644 | 2695 | * |
401a8d04 TH |
2696 | * Cancel @work and wait for its execution to finish. This function |
2697 | * can be used even if the work re-queues itself or migrates to | |
2698 | * another workqueue. On return from this function, @work is | |
2699 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2700 | * |
401a8d04 TH |
2701 | * cancel_work_sync(&delayed_work->work) must not be used for |
2702 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2703 | * |
401a8d04 | 2704 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2705 | * queued can't be destroyed before this function returns. |
401a8d04 TH |
2706 | * |
2707 | * RETURNS: | |
2708 | * %true if @work was pending, %false otherwise. | |
6e84d644 | 2709 | */ |
401a8d04 | 2710 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2711 | { |
1f1f642e | 2712 | return __cancel_work_timer(work, NULL); |
b89deed3 | 2713 | } |
28e53bdd | 2714 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2715 | |
6e84d644 | 2716 | /** |
401a8d04 TH |
2717 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2718 | * @dwork: the delayed work to flush | |
6e84d644 | 2719 | * |
401a8d04 TH |
2720 | * Delayed timer is cancelled and the pending work is queued for |
2721 | * immediate execution. Like flush_work(), this function only | |
2722 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2723 | * |
401a8d04 TH |
2724 | * RETURNS: |
2725 | * %true if flush_work() waited for the work to finish execution, | |
2726 | * %false if it was already idle. | |
6e84d644 | 2727 | */ |
401a8d04 TH |
2728 | bool flush_delayed_work(struct delayed_work *dwork) |
2729 | { | |
2730 | if (del_timer_sync(&dwork->timer)) | |
2731 | __queue_work(raw_smp_processor_id(), | |
2732 | get_work_cwq(&dwork->work)->wq, &dwork->work); | |
2733 | return flush_work(&dwork->work); | |
2734 | } | |
2735 | EXPORT_SYMBOL(flush_delayed_work); | |
2736 | ||
09383498 TH |
2737 | /** |
2738 | * flush_delayed_work_sync - wait for a dwork to finish | |
2739 | * @dwork: the delayed work to flush | |
2740 | * | |
2741 | * Delayed timer is cancelled and the pending work is queued for | |
2742 | * execution immediately. Other than timer handling, its behavior | |
2743 | * is identical to flush_work_sync(). | |
2744 | * | |
2745 | * RETURNS: | |
2746 | * %true if flush_work_sync() waited for the work to finish execution, | |
2747 | * %false if it was already idle. | |
2748 | */ | |
2749 | bool flush_delayed_work_sync(struct delayed_work *dwork) | |
2750 | { | |
2751 | if (del_timer_sync(&dwork->timer)) | |
2752 | __queue_work(raw_smp_processor_id(), | |
2753 | get_work_cwq(&dwork->work)->wq, &dwork->work); | |
2754 | return flush_work_sync(&dwork->work); | |
2755 | } | |
2756 | EXPORT_SYMBOL(flush_delayed_work_sync); | |
2757 | ||
401a8d04 TH |
2758 | /** |
2759 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
2760 | * @dwork: the delayed work cancel | |
2761 | * | |
2762 | * This is cancel_work_sync() for delayed works. | |
2763 | * | |
2764 | * RETURNS: | |
2765 | * %true if @dwork was pending, %false otherwise. | |
2766 | */ | |
2767 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 2768 | { |
1f1f642e | 2769 | return __cancel_work_timer(&dwork->work, &dwork->timer); |
6e84d644 | 2770 | } |
f5a421a4 | 2771 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 2772 | |
0fcb78c2 REB |
2773 | /** |
2774 | * schedule_work - put work task in global workqueue | |
2775 | * @work: job to be done | |
2776 | * | |
5b0f437d BVA |
2777 | * Returns zero if @work was already on the kernel-global workqueue and |
2778 | * non-zero otherwise. | |
2779 | * | |
2780 | * This puts a job in the kernel-global workqueue if it was not already | |
2781 | * queued and leaves it in the same position on the kernel-global | |
2782 | * workqueue otherwise. | |
0fcb78c2 | 2783 | */ |
7ad5b3a5 | 2784 | int schedule_work(struct work_struct *work) |
1da177e4 | 2785 | { |
d320c038 | 2786 | return queue_work(system_wq, work); |
1da177e4 | 2787 | } |
ae90dd5d | 2788 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 2789 | |
c1a220e7 ZR |
2790 | /* |
2791 | * schedule_work_on - put work task on a specific cpu | |
2792 | * @cpu: cpu to put the work task on | |
2793 | * @work: job to be done | |
2794 | * | |
2795 | * This puts a job on a specific cpu | |
2796 | */ | |
2797 | int schedule_work_on(int cpu, struct work_struct *work) | |
2798 | { | |
d320c038 | 2799 | return queue_work_on(cpu, system_wq, work); |
c1a220e7 ZR |
2800 | } |
2801 | EXPORT_SYMBOL(schedule_work_on); | |
2802 | ||
0fcb78c2 REB |
2803 | /** |
2804 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
2805 | * @dwork: job to be done |
2806 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
2807 | * |
2808 | * After waiting for a given time this puts a job in the kernel-global | |
2809 | * workqueue. | |
2810 | */ | |
7ad5b3a5 | 2811 | int schedule_delayed_work(struct delayed_work *dwork, |
82f67cd9 | 2812 | unsigned long delay) |
1da177e4 | 2813 | { |
d320c038 | 2814 | return queue_delayed_work(system_wq, dwork, delay); |
1da177e4 | 2815 | } |
ae90dd5d | 2816 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 2817 | |
0fcb78c2 REB |
2818 | /** |
2819 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
2820 | * @cpu: cpu to use | |
52bad64d | 2821 | * @dwork: job to be done |
0fcb78c2 REB |
2822 | * @delay: number of jiffies to wait |
2823 | * | |
2824 | * After waiting for a given time this puts a job in the kernel-global | |
2825 | * workqueue on the specified CPU. | |
2826 | */ | |
1da177e4 | 2827 | int schedule_delayed_work_on(int cpu, |
52bad64d | 2828 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 2829 | { |
d320c038 | 2830 | return queue_delayed_work_on(cpu, system_wq, dwork, delay); |
1da177e4 | 2831 | } |
ae90dd5d | 2832 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 2833 | |
b6136773 | 2834 | /** |
31ddd871 | 2835 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 2836 | * @func: the function to call |
b6136773 | 2837 | * |
31ddd871 TH |
2838 | * schedule_on_each_cpu() executes @func on each online CPU using the |
2839 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 2840 | * schedule_on_each_cpu() is very slow. |
31ddd871 TH |
2841 | * |
2842 | * RETURNS: | |
2843 | * 0 on success, -errno on failure. | |
b6136773 | 2844 | */ |
65f27f38 | 2845 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
2846 | { |
2847 | int cpu; | |
38f51568 | 2848 | struct work_struct __percpu *works; |
15316ba8 | 2849 | |
b6136773 AM |
2850 | works = alloc_percpu(struct work_struct); |
2851 | if (!works) | |
15316ba8 | 2852 | return -ENOMEM; |
b6136773 | 2853 | |
93981800 TH |
2854 | get_online_cpus(); |
2855 | ||
15316ba8 | 2856 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
2857 | struct work_struct *work = per_cpu_ptr(works, cpu); |
2858 | ||
2859 | INIT_WORK(work, func); | |
b71ab8c2 | 2860 | schedule_work_on(cpu, work); |
65a64464 | 2861 | } |
93981800 TH |
2862 | |
2863 | for_each_online_cpu(cpu) | |
2864 | flush_work(per_cpu_ptr(works, cpu)); | |
2865 | ||
95402b38 | 2866 | put_online_cpus(); |
b6136773 | 2867 | free_percpu(works); |
15316ba8 CL |
2868 | return 0; |
2869 | } | |
2870 | ||
eef6a7d5 AS |
2871 | /** |
2872 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
2873 | * | |
2874 | * Forces execution of the kernel-global workqueue and blocks until its | |
2875 | * completion. | |
2876 | * | |
2877 | * Think twice before calling this function! It's very easy to get into | |
2878 | * trouble if you don't take great care. Either of the following situations | |
2879 | * will lead to deadlock: | |
2880 | * | |
2881 | * One of the work items currently on the workqueue needs to acquire | |
2882 | * a lock held by your code or its caller. | |
2883 | * | |
2884 | * Your code is running in the context of a work routine. | |
2885 | * | |
2886 | * They will be detected by lockdep when they occur, but the first might not | |
2887 | * occur very often. It depends on what work items are on the workqueue and | |
2888 | * what locks they need, which you have no control over. | |
2889 | * | |
2890 | * In most situations flushing the entire workqueue is overkill; you merely | |
2891 | * need to know that a particular work item isn't queued and isn't running. | |
2892 | * In such cases you should use cancel_delayed_work_sync() or | |
2893 | * cancel_work_sync() instead. | |
2894 | */ | |
1da177e4 LT |
2895 | void flush_scheduled_work(void) |
2896 | { | |
d320c038 | 2897 | flush_workqueue(system_wq); |
1da177e4 | 2898 | } |
ae90dd5d | 2899 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 2900 | |
1fa44eca JB |
2901 | /** |
2902 | * execute_in_process_context - reliably execute the routine with user context | |
2903 | * @fn: the function to execute | |
1fa44eca JB |
2904 | * @ew: guaranteed storage for the execute work structure (must |
2905 | * be available when the work executes) | |
2906 | * | |
2907 | * Executes the function immediately if process context is available, | |
2908 | * otherwise schedules the function for delayed execution. | |
2909 | * | |
2910 | * Returns: 0 - function was executed | |
2911 | * 1 - function was scheduled for execution | |
2912 | */ | |
65f27f38 | 2913 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
2914 | { |
2915 | if (!in_interrupt()) { | |
65f27f38 | 2916 | fn(&ew->work); |
1fa44eca JB |
2917 | return 0; |
2918 | } | |
2919 | ||
65f27f38 | 2920 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
2921 | schedule_work(&ew->work); |
2922 | ||
2923 | return 1; | |
2924 | } | |
2925 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
2926 | ||
1da177e4 LT |
2927 | int keventd_up(void) |
2928 | { | |
d320c038 | 2929 | return system_wq != NULL; |
1da177e4 LT |
2930 | } |
2931 | ||
bdbc5dd7 | 2932 | static int alloc_cwqs(struct workqueue_struct *wq) |
0f900049 | 2933 | { |
65a64464 | 2934 | /* |
0f900049 TH |
2935 | * cwqs are forced aligned according to WORK_STRUCT_FLAG_BITS. |
2936 | * Make sure that the alignment isn't lower than that of | |
2937 | * unsigned long long. | |
65a64464 | 2938 | */ |
0f900049 TH |
2939 | const size_t size = sizeof(struct cpu_workqueue_struct); |
2940 | const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS, | |
2941 | __alignof__(unsigned long long)); | |
65a64464 | 2942 | |
e06ffa1e | 2943 | if (!(wq->flags & WQ_UNBOUND)) |
f3421797 | 2944 | wq->cpu_wq.pcpu = __alloc_percpu(size, align); |
931ac77e | 2945 | else { |
f3421797 TH |
2946 | void *ptr; |
2947 | ||
2948 | /* | |
2949 | * Allocate enough room to align cwq and put an extra | |
2950 | * pointer at the end pointing back to the originally | |
2951 | * allocated pointer which will be used for free. | |
2952 | */ | |
2953 | ptr = kzalloc(size + align + sizeof(void *), GFP_KERNEL); | |
2954 | if (ptr) { | |
2955 | wq->cpu_wq.single = PTR_ALIGN(ptr, align); | |
2956 | *(void **)(wq->cpu_wq.single + 1) = ptr; | |
2957 | } | |
bdbc5dd7 | 2958 | } |
f3421797 | 2959 | |
0415b00d | 2960 | /* just in case, make sure it's actually aligned */ |
bdbc5dd7 TH |
2961 | BUG_ON(!IS_ALIGNED(wq->cpu_wq.v, align)); |
2962 | return wq->cpu_wq.v ? 0 : -ENOMEM; | |
0f900049 TH |
2963 | } |
2964 | ||
bdbc5dd7 | 2965 | static void free_cwqs(struct workqueue_struct *wq) |
0f900049 | 2966 | { |
e06ffa1e | 2967 | if (!(wq->flags & WQ_UNBOUND)) |
f3421797 TH |
2968 | free_percpu(wq->cpu_wq.pcpu); |
2969 | else if (wq->cpu_wq.single) { | |
2970 | /* the pointer to free is stored right after the cwq */ | |
bdbc5dd7 | 2971 | kfree(*(void **)(wq->cpu_wq.single + 1)); |
f3421797 | 2972 | } |
0f900049 TH |
2973 | } |
2974 | ||
f3421797 TH |
2975 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
2976 | const char *name) | |
b71ab8c2 | 2977 | { |
f3421797 TH |
2978 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
2979 | ||
2980 | if (max_active < 1 || max_active > lim) | |
b71ab8c2 TH |
2981 | printk(KERN_WARNING "workqueue: max_active %d requested for %s " |
2982 | "is out of range, clamping between %d and %d\n", | |
f3421797 | 2983 | max_active, name, 1, lim); |
b71ab8c2 | 2984 | |
f3421797 | 2985 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
2986 | } |
2987 | ||
b196be89 | 2988 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
d320c038 TH |
2989 | unsigned int flags, |
2990 | int max_active, | |
2991 | struct lock_class_key *key, | |
b196be89 | 2992 | const char *lock_name, ...) |
1da177e4 | 2993 | { |
b196be89 | 2994 | va_list args, args1; |
1da177e4 | 2995 | struct workqueue_struct *wq; |
c34056a3 | 2996 | unsigned int cpu; |
b196be89 TH |
2997 | size_t namelen; |
2998 | ||
2999 | /* determine namelen, allocate wq and format name */ | |
3000 | va_start(args, lock_name); | |
3001 | va_copy(args1, args); | |
3002 | namelen = vsnprintf(NULL, 0, fmt, args) + 1; | |
3003 | ||
3004 | wq = kzalloc(sizeof(*wq) + namelen, GFP_KERNEL); | |
3005 | if (!wq) | |
3006 | goto err; | |
3007 | ||
3008 | vsnprintf(wq->name, namelen, fmt, args1); | |
3009 | va_end(args); | |
3010 | va_end(args1); | |
1da177e4 | 3011 | |
6370a6ad TH |
3012 | /* |
3013 | * Workqueues which may be used during memory reclaim should | |
3014 | * have a rescuer to guarantee forward progress. | |
3015 | */ | |
3016 | if (flags & WQ_MEM_RECLAIM) | |
3017 | flags |= WQ_RESCUER; | |
3018 | ||
d320c038 | 3019 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 3020 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 3021 | |
b196be89 | 3022 | /* init wq */ |
97e37d7b | 3023 | wq->flags = flags; |
a0a1a5fd | 3024 | wq->saved_max_active = max_active; |
73f53c4a TH |
3025 | mutex_init(&wq->flush_mutex); |
3026 | atomic_set(&wq->nr_cwqs_to_flush, 0); | |
3027 | INIT_LIST_HEAD(&wq->flusher_queue); | |
3028 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
502ca9d8 | 3029 | |
eb13ba87 | 3030 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 3031 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 3032 | |
bdbc5dd7 TH |
3033 | if (alloc_cwqs(wq) < 0) |
3034 | goto err; | |
3035 | ||
f3421797 | 3036 | for_each_cwq_cpu(cpu, wq) { |
1537663f | 3037 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
8b03ae3c | 3038 | struct global_cwq *gcwq = get_gcwq(cpu); |
1537663f | 3039 | |
0f900049 | 3040 | BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK); |
bd7bdd43 | 3041 | cwq->pool = &gcwq->pool; |
c34056a3 | 3042 | cwq->wq = wq; |
73f53c4a | 3043 | cwq->flush_color = -1; |
1e19ffc6 | 3044 | cwq->max_active = max_active; |
1e19ffc6 | 3045 | INIT_LIST_HEAD(&cwq->delayed_works); |
e22bee78 | 3046 | } |
1537663f | 3047 | |
e22bee78 TH |
3048 | if (flags & WQ_RESCUER) { |
3049 | struct worker *rescuer; | |
3050 | ||
f2e005aa | 3051 | if (!alloc_mayday_mask(&wq->mayday_mask, GFP_KERNEL)) |
e22bee78 TH |
3052 | goto err; |
3053 | ||
3054 | wq->rescuer = rescuer = alloc_worker(); | |
3055 | if (!rescuer) | |
3056 | goto err; | |
3057 | ||
b196be89 TH |
3058 | rescuer->task = kthread_create(rescuer_thread, wq, "%s", |
3059 | wq->name); | |
e22bee78 TH |
3060 | if (IS_ERR(rescuer->task)) |
3061 | goto err; | |
3062 | ||
e22bee78 TH |
3063 | rescuer->task->flags |= PF_THREAD_BOUND; |
3064 | wake_up_process(rescuer->task); | |
3af24433 ON |
3065 | } |
3066 | ||
a0a1a5fd TH |
3067 | /* |
3068 | * workqueue_lock protects global freeze state and workqueues | |
3069 | * list. Grab it, set max_active accordingly and add the new | |
3070 | * workqueue to workqueues list. | |
3071 | */ | |
1537663f | 3072 | spin_lock(&workqueue_lock); |
a0a1a5fd | 3073 | |
58a69cb4 | 3074 | if (workqueue_freezing && wq->flags & WQ_FREEZABLE) |
f3421797 | 3075 | for_each_cwq_cpu(cpu, wq) |
a0a1a5fd TH |
3076 | get_cwq(cpu, wq)->max_active = 0; |
3077 | ||
1537663f | 3078 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 3079 | |
1537663f TH |
3080 | spin_unlock(&workqueue_lock); |
3081 | ||
3af24433 | 3082 | return wq; |
4690c4ab TH |
3083 | err: |
3084 | if (wq) { | |
bdbc5dd7 | 3085 | free_cwqs(wq); |
f2e005aa | 3086 | free_mayday_mask(wq->mayday_mask); |
e22bee78 | 3087 | kfree(wq->rescuer); |
4690c4ab TH |
3088 | kfree(wq); |
3089 | } | |
3090 | return NULL; | |
3af24433 | 3091 | } |
d320c038 | 3092 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 3093 | |
3af24433 ON |
3094 | /** |
3095 | * destroy_workqueue - safely terminate a workqueue | |
3096 | * @wq: target workqueue | |
3097 | * | |
3098 | * Safely destroy a workqueue. All work currently pending will be done first. | |
3099 | */ | |
3100 | void destroy_workqueue(struct workqueue_struct *wq) | |
3101 | { | |
c8e55f36 | 3102 | unsigned int cpu; |
3af24433 | 3103 | |
9c5a2ba7 TH |
3104 | /* drain it before proceeding with destruction */ |
3105 | drain_workqueue(wq); | |
c8efcc25 | 3106 | |
a0a1a5fd TH |
3107 | /* |
3108 | * wq list is used to freeze wq, remove from list after | |
3109 | * flushing is complete in case freeze races us. | |
3110 | */ | |
95402b38 | 3111 | spin_lock(&workqueue_lock); |
b1f4ec17 | 3112 | list_del(&wq->list); |
95402b38 | 3113 | spin_unlock(&workqueue_lock); |
3af24433 | 3114 | |
e22bee78 | 3115 | /* sanity check */ |
f3421797 | 3116 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a TH |
3117 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3118 | int i; | |
3119 | ||
73f53c4a TH |
3120 | for (i = 0; i < WORK_NR_COLORS; i++) |
3121 | BUG_ON(cwq->nr_in_flight[i]); | |
1e19ffc6 TH |
3122 | BUG_ON(cwq->nr_active); |
3123 | BUG_ON(!list_empty(&cwq->delayed_works)); | |
73f53c4a | 3124 | } |
9b41ea72 | 3125 | |
e22bee78 TH |
3126 | if (wq->flags & WQ_RESCUER) { |
3127 | kthread_stop(wq->rescuer->task); | |
f2e005aa | 3128 | free_mayday_mask(wq->mayday_mask); |
8d9df9f0 | 3129 | kfree(wq->rescuer); |
e22bee78 TH |
3130 | } |
3131 | ||
bdbc5dd7 | 3132 | free_cwqs(wq); |
3af24433 ON |
3133 | kfree(wq); |
3134 | } | |
3135 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
3136 | ||
dcd989cb TH |
3137 | /** |
3138 | * workqueue_set_max_active - adjust max_active of a workqueue | |
3139 | * @wq: target workqueue | |
3140 | * @max_active: new max_active value. | |
3141 | * | |
3142 | * Set max_active of @wq to @max_active. | |
3143 | * | |
3144 | * CONTEXT: | |
3145 | * Don't call from IRQ context. | |
3146 | */ | |
3147 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
3148 | { | |
3149 | unsigned int cpu; | |
3150 | ||
f3421797 | 3151 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb TH |
3152 | |
3153 | spin_lock(&workqueue_lock); | |
3154 | ||
3155 | wq->saved_max_active = max_active; | |
3156 | ||
f3421797 | 3157 | for_each_cwq_cpu(cpu, wq) { |
dcd989cb TH |
3158 | struct global_cwq *gcwq = get_gcwq(cpu); |
3159 | ||
3160 | spin_lock_irq(&gcwq->lock); | |
3161 | ||
58a69cb4 | 3162 | if (!(wq->flags & WQ_FREEZABLE) || |
dcd989cb TH |
3163 | !(gcwq->flags & GCWQ_FREEZING)) |
3164 | get_cwq(gcwq->cpu, wq)->max_active = max_active; | |
9bfb1839 | 3165 | |
dcd989cb | 3166 | spin_unlock_irq(&gcwq->lock); |
65a64464 | 3167 | } |
93981800 | 3168 | |
dcd989cb | 3169 | spin_unlock(&workqueue_lock); |
15316ba8 | 3170 | } |
dcd989cb | 3171 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 3172 | |
eef6a7d5 | 3173 | /** |
dcd989cb TH |
3174 | * workqueue_congested - test whether a workqueue is congested |
3175 | * @cpu: CPU in question | |
3176 | * @wq: target workqueue | |
eef6a7d5 | 3177 | * |
dcd989cb TH |
3178 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
3179 | * no synchronization around this function and the test result is | |
3180 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 3181 | * |
dcd989cb TH |
3182 | * RETURNS: |
3183 | * %true if congested, %false otherwise. | |
eef6a7d5 | 3184 | */ |
dcd989cb | 3185 | bool workqueue_congested(unsigned int cpu, struct workqueue_struct *wq) |
1da177e4 | 3186 | { |
dcd989cb TH |
3187 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3188 | ||
3189 | return !list_empty(&cwq->delayed_works); | |
1da177e4 | 3190 | } |
dcd989cb | 3191 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 3192 | |
1fa44eca | 3193 | /** |
dcd989cb TH |
3194 | * work_cpu - return the last known associated cpu for @work |
3195 | * @work: the work of interest | |
1fa44eca | 3196 | * |
dcd989cb | 3197 | * RETURNS: |
bdbc5dd7 | 3198 | * CPU number if @work was ever queued. WORK_CPU_NONE otherwise. |
1fa44eca | 3199 | */ |
dcd989cb | 3200 | unsigned int work_cpu(struct work_struct *work) |
1fa44eca | 3201 | { |
dcd989cb | 3202 | struct global_cwq *gcwq = get_work_gcwq(work); |
1fa44eca | 3203 | |
bdbc5dd7 | 3204 | return gcwq ? gcwq->cpu : WORK_CPU_NONE; |
1fa44eca | 3205 | } |
dcd989cb | 3206 | EXPORT_SYMBOL_GPL(work_cpu); |
1fa44eca | 3207 | |
dcd989cb TH |
3208 | /** |
3209 | * work_busy - test whether a work is currently pending or running | |
3210 | * @work: the work to be tested | |
3211 | * | |
3212 | * Test whether @work is currently pending or running. There is no | |
3213 | * synchronization around this function and the test result is | |
3214 | * unreliable and only useful as advisory hints or for debugging. | |
3215 | * Especially for reentrant wqs, the pending state might hide the | |
3216 | * running state. | |
3217 | * | |
3218 | * RETURNS: | |
3219 | * OR'd bitmask of WORK_BUSY_* bits. | |
3220 | */ | |
3221 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 3222 | { |
dcd989cb TH |
3223 | struct global_cwq *gcwq = get_work_gcwq(work); |
3224 | unsigned long flags; | |
3225 | unsigned int ret = 0; | |
1da177e4 | 3226 | |
dcd989cb TH |
3227 | if (!gcwq) |
3228 | return false; | |
1da177e4 | 3229 | |
dcd989cb | 3230 | spin_lock_irqsave(&gcwq->lock, flags); |
1da177e4 | 3231 | |
dcd989cb TH |
3232 | if (work_pending(work)) |
3233 | ret |= WORK_BUSY_PENDING; | |
3234 | if (find_worker_executing_work(gcwq, work)) | |
3235 | ret |= WORK_BUSY_RUNNING; | |
1da177e4 | 3236 | |
dcd989cb | 3237 | spin_unlock_irqrestore(&gcwq->lock, flags); |
1da177e4 | 3238 | |
dcd989cb | 3239 | return ret; |
1da177e4 | 3240 | } |
dcd989cb | 3241 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 3242 | |
db7bccf4 TH |
3243 | /* |
3244 | * CPU hotplug. | |
3245 | * | |
e22bee78 TH |
3246 | * There are two challenges in supporting CPU hotplug. Firstly, there |
3247 | * are a lot of assumptions on strong associations among work, cwq and | |
3248 | * gcwq which make migrating pending and scheduled works very | |
3249 | * difficult to implement without impacting hot paths. Secondly, | |
3250 | * gcwqs serve mix of short, long and very long running works making | |
3251 | * blocked draining impractical. | |
3252 | * | |
3253 | * This is solved by allowing a gcwq to be detached from CPU, running | |
3254 | * it with unbound (rogue) workers and allowing it to be reattached | |
3255 | * later if the cpu comes back online. A separate thread is created | |
3256 | * to govern a gcwq in such state and is called the trustee of the | |
3257 | * gcwq. | |
db7bccf4 TH |
3258 | * |
3259 | * Trustee states and their descriptions. | |
3260 | * | |
3261 | * START Command state used on startup. On CPU_DOWN_PREPARE, a | |
3262 | * new trustee is started with this state. | |
3263 | * | |
3264 | * IN_CHARGE Once started, trustee will enter this state after | |
e22bee78 TH |
3265 | * assuming the manager role and making all existing |
3266 | * workers rogue. DOWN_PREPARE waits for trustee to | |
3267 | * enter this state. After reaching IN_CHARGE, trustee | |
3268 | * tries to execute the pending worklist until it's empty | |
3269 | * and the state is set to BUTCHER, or the state is set | |
3270 | * to RELEASE. | |
db7bccf4 TH |
3271 | * |
3272 | * BUTCHER Command state which is set by the cpu callback after | |
3273 | * the cpu has went down. Once this state is set trustee | |
3274 | * knows that there will be no new works on the worklist | |
3275 | * and once the worklist is empty it can proceed to | |
3276 | * killing idle workers. | |
3277 | * | |
3278 | * RELEASE Command state which is set by the cpu callback if the | |
3279 | * cpu down has been canceled or it has come online | |
3280 | * again. After recognizing this state, trustee stops | |
e22bee78 TH |
3281 | * trying to drain or butcher and clears ROGUE, rebinds |
3282 | * all remaining workers back to the cpu and releases | |
3283 | * manager role. | |
db7bccf4 TH |
3284 | * |
3285 | * DONE Trustee will enter this state after BUTCHER or RELEASE | |
3286 | * is complete. | |
3287 | * | |
3288 | * trustee CPU draining | |
3289 | * took over down complete | |
3290 | * START -----------> IN_CHARGE -----------> BUTCHER -----------> DONE | |
3291 | * | | ^ | |
3292 | * | CPU is back online v return workers | | |
3293 | * ----------------> RELEASE -------------- | |
3294 | */ | |
1da177e4 | 3295 | |
db7bccf4 TH |
3296 | /** |
3297 | * trustee_wait_event_timeout - timed event wait for trustee | |
3298 | * @cond: condition to wait for | |
3299 | * @timeout: timeout in jiffies | |
3300 | * | |
3301 | * wait_event_timeout() for trustee to use. Handles locking and | |
3302 | * checks for RELEASE request. | |
3303 | * | |
3304 | * CONTEXT: | |
3305 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
3306 | * multiple times. To be used by trustee. | |
3307 | * | |
3308 | * RETURNS: | |
3309 | * Positive indicating left time if @cond is satisfied, 0 if timed | |
3310 | * out, -1 if canceled. | |
3311 | */ | |
3312 | #define trustee_wait_event_timeout(cond, timeout) ({ \ | |
3313 | long __ret = (timeout); \ | |
3314 | while (!((cond) || (gcwq->trustee_state == TRUSTEE_RELEASE)) && \ | |
3315 | __ret) { \ | |
3316 | spin_unlock_irq(&gcwq->lock); \ | |
3317 | __wait_event_timeout(gcwq->trustee_wait, (cond) || \ | |
3318 | (gcwq->trustee_state == TRUSTEE_RELEASE), \ | |
3319 | __ret); \ | |
3320 | spin_lock_irq(&gcwq->lock); \ | |
3321 | } \ | |
3322 | gcwq->trustee_state == TRUSTEE_RELEASE ? -1 : (__ret); \ | |
3323 | }) | |
3af24433 | 3324 | |
db7bccf4 TH |
3325 | /** |
3326 | * trustee_wait_event - event wait for trustee | |
3327 | * @cond: condition to wait for | |
3328 | * | |
3329 | * wait_event() for trustee to use. Automatically handles locking and | |
3330 | * checks for CANCEL request. | |
3331 | * | |
3332 | * CONTEXT: | |
3333 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
3334 | * multiple times. To be used by trustee. | |
3335 | * | |
3336 | * RETURNS: | |
3337 | * 0 if @cond is satisfied, -1 if canceled. | |
3338 | */ | |
3339 | #define trustee_wait_event(cond) ({ \ | |
3340 | long __ret1; \ | |
3341 | __ret1 = trustee_wait_event_timeout(cond, MAX_SCHEDULE_TIMEOUT);\ | |
3342 | __ret1 < 0 ? -1 : 0; \ | |
3343 | }) | |
1da177e4 | 3344 | |
db7bccf4 | 3345 | static int __cpuinit trustee_thread(void *__gcwq) |
3af24433 | 3346 | { |
db7bccf4 TH |
3347 | struct global_cwq *gcwq = __gcwq; |
3348 | struct worker *worker; | |
e22bee78 | 3349 | struct work_struct *work; |
db7bccf4 | 3350 | struct hlist_node *pos; |
e22bee78 | 3351 | long rc; |
db7bccf4 | 3352 | int i; |
3af24433 | 3353 | |
db7bccf4 TH |
3354 | BUG_ON(gcwq->cpu != smp_processor_id()); |
3355 | ||
3356 | spin_lock_irq(&gcwq->lock); | |
3af24433 | 3357 | /* |
e22bee78 TH |
3358 | * Claim the manager position and make all workers rogue. |
3359 | * Trustee must be bound to the target cpu and can't be | |
3360 | * cancelled. | |
3af24433 | 3361 | */ |
db7bccf4 | 3362 | BUG_ON(gcwq->cpu != smp_processor_id()); |
e22bee78 TH |
3363 | rc = trustee_wait_event(!(gcwq->flags & GCWQ_MANAGING_WORKERS)); |
3364 | BUG_ON(rc < 0); | |
3af24433 | 3365 | |
e22bee78 | 3366 | gcwq->flags |= GCWQ_MANAGING_WORKERS; |
e1d8aa9f | 3367 | |
bd7bdd43 | 3368 | list_for_each_entry(worker, &gcwq->pool.idle_list, entry) |
cb444766 | 3369 | worker->flags |= WORKER_ROGUE; |
3af24433 | 3370 | |
db7bccf4 | 3371 | for_each_busy_worker(worker, i, pos, gcwq) |
cb444766 | 3372 | worker->flags |= WORKER_ROGUE; |
06ba38a9 | 3373 | |
e22bee78 TH |
3374 | /* |
3375 | * Call schedule() so that we cross rq->lock and thus can | |
3376 | * guarantee sched callbacks see the rogue flag. This is | |
3377 | * necessary as scheduler callbacks may be invoked from other | |
3378 | * cpus. | |
3379 | */ | |
3380 | spin_unlock_irq(&gcwq->lock); | |
3381 | schedule(); | |
3382 | spin_lock_irq(&gcwq->lock); | |
06ba38a9 | 3383 | |
e22bee78 | 3384 | /* |
cb444766 TH |
3385 | * Sched callbacks are disabled now. Zap nr_running. After |
3386 | * this, nr_running stays zero and need_more_worker() and | |
3387 | * keep_working() are always true as long as the worklist is | |
3388 | * not empty. | |
e22bee78 | 3389 | */ |
63d95a91 | 3390 | atomic_set(get_pool_nr_running(&gcwq->pool), 0); |
1da177e4 | 3391 | |
e22bee78 | 3392 | spin_unlock_irq(&gcwq->lock); |
bd7bdd43 | 3393 | del_timer_sync(&gcwq->pool.idle_timer); |
e22bee78 | 3394 | spin_lock_irq(&gcwq->lock); |
3af24433 | 3395 | |
db7bccf4 TH |
3396 | /* |
3397 | * We're now in charge. Notify and proceed to drain. We need | |
3398 | * to keep the gcwq running during the whole CPU down | |
3399 | * procedure as other cpu hotunplug callbacks may need to | |
3400 | * flush currently running tasks. | |
3401 | */ | |
3402 | gcwq->trustee_state = TRUSTEE_IN_CHARGE; | |
3403 | wake_up_all(&gcwq->trustee_wait); | |
3af24433 | 3404 | |
db7bccf4 TH |
3405 | /* |
3406 | * The original cpu is in the process of dying and may go away | |
3407 | * anytime now. When that happens, we and all workers would | |
e22bee78 TH |
3408 | * be migrated to other cpus. Try draining any left work. We |
3409 | * want to get it over with ASAP - spam rescuers, wake up as | |
3410 | * many idlers as necessary and create new ones till the | |
3411 | * worklist is empty. Note that if the gcwq is frozen, there | |
58a69cb4 | 3412 | * may be frozen works in freezable cwqs. Don't declare |
e22bee78 | 3413 | * completion while frozen. |
db7bccf4 | 3414 | */ |
bd7bdd43 | 3415 | while (gcwq->pool.nr_workers != gcwq->pool.nr_idle || |
db7bccf4 TH |
3416 | gcwq->flags & GCWQ_FREEZING || |
3417 | gcwq->trustee_state == TRUSTEE_IN_CHARGE) { | |
e22bee78 TH |
3418 | int nr_works = 0; |
3419 | ||
bd7bdd43 | 3420 | list_for_each_entry(work, &gcwq->pool.worklist, entry) { |
e22bee78 TH |
3421 | send_mayday(work); |
3422 | nr_works++; | |
3423 | } | |
3af24433 | 3424 | |
bd7bdd43 | 3425 | list_for_each_entry(worker, &gcwq->pool.idle_list, entry) { |
e22bee78 TH |
3426 | if (!nr_works--) |
3427 | break; | |
3428 | wake_up_process(worker->task); | |
3429 | } | |
3430 | ||
63d95a91 | 3431 | if (need_to_create_worker(&gcwq->pool)) { |
e22bee78 | 3432 | spin_unlock_irq(&gcwq->lock); |
63d95a91 | 3433 | worker = create_worker(&gcwq->pool, false); |
e22bee78 TH |
3434 | spin_lock_irq(&gcwq->lock); |
3435 | if (worker) { | |
cb444766 | 3436 | worker->flags |= WORKER_ROGUE; |
e22bee78 TH |
3437 | start_worker(worker); |
3438 | } | |
1da177e4 | 3439 | } |
3af24433 | 3440 | |
db7bccf4 TH |
3441 | /* give a breather */ |
3442 | if (trustee_wait_event_timeout(false, TRUSTEE_COOLDOWN) < 0) | |
3443 | break; | |
3af24433 | 3444 | } |
1da177e4 | 3445 | |
14441960 | 3446 | /* |
e22bee78 TH |
3447 | * Either all works have been scheduled and cpu is down, or |
3448 | * cpu down has already been canceled. Wait for and butcher | |
3449 | * all workers till we're canceled. | |
14441960 | 3450 | */ |
e22bee78 | 3451 | do { |
bd7bdd43 TH |
3452 | rc = trustee_wait_event(!list_empty(&gcwq->pool.idle_list)); |
3453 | while (!list_empty(&gcwq->pool.idle_list)) | |
3454 | destroy_worker(list_first_entry(&gcwq->pool.idle_list, | |
e22bee78 | 3455 | struct worker, entry)); |
bd7bdd43 | 3456 | } while (gcwq->pool.nr_workers && rc >= 0); |
4e6045f1 | 3457 | |
14441960 | 3458 | /* |
e22bee78 TH |
3459 | * At this point, either draining has completed and no worker |
3460 | * is left, or cpu down has been canceled or the cpu is being | |
3461 | * brought back up. There shouldn't be any idle one left. | |
3462 | * Tell the remaining busy ones to rebind once it finishes the | |
3463 | * currently scheduled works by scheduling the rebind_work. | |
14441960 | 3464 | */ |
bd7bdd43 | 3465 | WARN_ON(!list_empty(&gcwq->pool.idle_list)); |
e22bee78 TH |
3466 | |
3467 | for_each_busy_worker(worker, i, pos, gcwq) { | |
3468 | struct work_struct *rebind_work = &worker->rebind_work; | |
3469 | ||
3470 | /* | |
3471 | * Rebind_work may race with future cpu hotplug | |
3472 | * operations. Use a separate flag to mark that | |
3473 | * rebinding is scheduled. | |
3474 | */ | |
cb444766 TH |
3475 | worker->flags |= WORKER_REBIND; |
3476 | worker->flags &= ~WORKER_ROGUE; | |
e22bee78 TH |
3477 | |
3478 | /* queue rebind_work, wq doesn't matter, use the default one */ | |
3479 | if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, | |
3480 | work_data_bits(rebind_work))) | |
3481 | continue; | |
3482 | ||
3483 | debug_work_activate(rebind_work); | |
d320c038 | 3484 | insert_work(get_cwq(gcwq->cpu, system_wq), rebind_work, |
e22bee78 TH |
3485 | worker->scheduled.next, |
3486 | work_color_to_flags(WORK_NO_COLOR)); | |
3487 | } | |
3488 | ||
3489 | /* relinquish manager role */ | |
3490 | gcwq->flags &= ~GCWQ_MANAGING_WORKERS; | |
3491 | ||
db7bccf4 TH |
3492 | /* notify completion */ |
3493 | gcwq->trustee = NULL; | |
3494 | gcwq->trustee_state = TRUSTEE_DONE; | |
3495 | wake_up_all(&gcwq->trustee_wait); | |
3496 | spin_unlock_irq(&gcwq->lock); | |
3497 | return 0; | |
3af24433 ON |
3498 | } |
3499 | ||
3500 | /** | |
db7bccf4 TH |
3501 | * wait_trustee_state - wait for trustee to enter the specified state |
3502 | * @gcwq: gcwq the trustee of interest belongs to | |
3503 | * @state: target state to wait for | |
3af24433 | 3504 | * |
db7bccf4 TH |
3505 | * Wait for the trustee to reach @state. DONE is already matched. |
3506 | * | |
3507 | * CONTEXT: | |
3508 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
3509 | * multiple times. To be used by cpu_callback. | |
3af24433 | 3510 | */ |
db7bccf4 | 3511 | static void __cpuinit wait_trustee_state(struct global_cwq *gcwq, int state) |
06bd6ebf NK |
3512 | __releases(&gcwq->lock) |
3513 | __acquires(&gcwq->lock) | |
3af24433 | 3514 | { |
db7bccf4 TH |
3515 | if (!(gcwq->trustee_state == state || |
3516 | gcwq->trustee_state == TRUSTEE_DONE)) { | |
3517 | spin_unlock_irq(&gcwq->lock); | |
3518 | __wait_event(gcwq->trustee_wait, | |
3519 | gcwq->trustee_state == state || | |
3520 | gcwq->trustee_state == TRUSTEE_DONE); | |
3521 | spin_lock_irq(&gcwq->lock); | |
3522 | } | |
3af24433 | 3523 | } |
3af24433 ON |
3524 | |
3525 | static int __devinit workqueue_cpu_callback(struct notifier_block *nfb, | |
3526 | unsigned long action, | |
3527 | void *hcpu) | |
3528 | { | |
3529 | unsigned int cpu = (unsigned long)hcpu; | |
db7bccf4 TH |
3530 | struct global_cwq *gcwq = get_gcwq(cpu); |
3531 | struct task_struct *new_trustee = NULL; | |
e22bee78 | 3532 | struct worker *uninitialized_var(new_worker); |
db7bccf4 | 3533 | unsigned long flags; |
3af24433 | 3534 | |
8bb78442 RW |
3535 | action &= ~CPU_TASKS_FROZEN; |
3536 | ||
3af24433 | 3537 | switch (action) { |
db7bccf4 TH |
3538 | case CPU_DOWN_PREPARE: |
3539 | new_trustee = kthread_create(trustee_thread, gcwq, | |
3540 | "workqueue_trustee/%d\n", cpu); | |
3541 | if (IS_ERR(new_trustee)) | |
3542 | return notifier_from_errno(PTR_ERR(new_trustee)); | |
3543 | kthread_bind(new_trustee, cpu); | |
e22bee78 | 3544 | /* fall through */ |
3af24433 | 3545 | case CPU_UP_PREPARE: |
bd7bdd43 | 3546 | BUG_ON(gcwq->pool.first_idle); |
63d95a91 | 3547 | new_worker = create_worker(&gcwq->pool, false); |
e22bee78 TH |
3548 | if (!new_worker) { |
3549 | if (new_trustee) | |
3550 | kthread_stop(new_trustee); | |
3551 | return NOTIFY_BAD; | |
3af24433 | 3552 | } |
1da177e4 LT |
3553 | } |
3554 | ||
db7bccf4 TH |
3555 | /* some are called w/ irq disabled, don't disturb irq status */ |
3556 | spin_lock_irqsave(&gcwq->lock, flags); | |
3af24433 | 3557 | |
00dfcaf7 | 3558 | switch (action) { |
db7bccf4 TH |
3559 | case CPU_DOWN_PREPARE: |
3560 | /* initialize trustee and tell it to acquire the gcwq */ | |
3561 | BUG_ON(gcwq->trustee || gcwq->trustee_state != TRUSTEE_DONE); | |
3562 | gcwq->trustee = new_trustee; | |
3563 | gcwq->trustee_state = TRUSTEE_START; | |
3564 | wake_up_process(gcwq->trustee); | |
3565 | wait_trustee_state(gcwq, TRUSTEE_IN_CHARGE); | |
e22bee78 TH |
3566 | /* fall through */ |
3567 | case CPU_UP_PREPARE: | |
bd7bdd43 TH |
3568 | BUG_ON(gcwq->pool.first_idle); |
3569 | gcwq->pool.first_idle = new_worker; | |
e22bee78 TH |
3570 | break; |
3571 | ||
3572 | case CPU_DYING: | |
3573 | /* | |
3574 | * Before this, the trustee and all workers except for | |
3575 | * the ones which are still executing works from | |
3576 | * before the last CPU down must be on the cpu. After | |
3577 | * this, they'll all be diasporas. | |
3578 | */ | |
3579 | gcwq->flags |= GCWQ_DISASSOCIATED; | |
db7bccf4 TH |
3580 | break; |
3581 | ||
3da1c84c | 3582 | case CPU_POST_DEAD: |
db7bccf4 | 3583 | gcwq->trustee_state = TRUSTEE_BUTCHER; |
e22bee78 TH |
3584 | /* fall through */ |
3585 | case CPU_UP_CANCELED: | |
bd7bdd43 TH |
3586 | destroy_worker(gcwq->pool.first_idle); |
3587 | gcwq->pool.first_idle = NULL; | |
db7bccf4 TH |
3588 | break; |
3589 | ||
3590 | case CPU_DOWN_FAILED: | |
3591 | case CPU_ONLINE: | |
e22bee78 | 3592 | gcwq->flags &= ~GCWQ_DISASSOCIATED; |
db7bccf4 TH |
3593 | if (gcwq->trustee_state != TRUSTEE_DONE) { |
3594 | gcwq->trustee_state = TRUSTEE_RELEASE; | |
3595 | wake_up_process(gcwq->trustee); | |
3596 | wait_trustee_state(gcwq, TRUSTEE_DONE); | |
3af24433 | 3597 | } |
db7bccf4 | 3598 | |
e22bee78 TH |
3599 | /* |
3600 | * Trustee is done and there might be no worker left. | |
3601 | * Put the first_idle in and request a real manager to | |
3602 | * take a look. | |
3603 | */ | |
3604 | spin_unlock_irq(&gcwq->lock); | |
bd7bdd43 | 3605 | kthread_bind(gcwq->pool.first_idle->task, cpu); |
e22bee78 TH |
3606 | spin_lock_irq(&gcwq->lock); |
3607 | gcwq->flags |= GCWQ_MANAGE_WORKERS; | |
bd7bdd43 TH |
3608 | start_worker(gcwq->pool.first_idle); |
3609 | gcwq->pool.first_idle = NULL; | |
db7bccf4 | 3610 | break; |
00dfcaf7 ON |
3611 | } |
3612 | ||
db7bccf4 TH |
3613 | spin_unlock_irqrestore(&gcwq->lock, flags); |
3614 | ||
1537663f | 3615 | return notifier_from_errno(0); |
1da177e4 | 3616 | } |
1da177e4 | 3617 | |
2d3854a3 | 3618 | #ifdef CONFIG_SMP |
8ccad40d | 3619 | |
2d3854a3 | 3620 | struct work_for_cpu { |
6b44003e | 3621 | struct completion completion; |
2d3854a3 RR |
3622 | long (*fn)(void *); |
3623 | void *arg; | |
3624 | long ret; | |
3625 | }; | |
3626 | ||
6b44003e | 3627 | static int do_work_for_cpu(void *_wfc) |
2d3854a3 | 3628 | { |
6b44003e | 3629 | struct work_for_cpu *wfc = _wfc; |
2d3854a3 | 3630 | wfc->ret = wfc->fn(wfc->arg); |
6b44003e AM |
3631 | complete(&wfc->completion); |
3632 | return 0; | |
2d3854a3 RR |
3633 | } |
3634 | ||
3635 | /** | |
3636 | * work_on_cpu - run a function in user context on a particular cpu | |
3637 | * @cpu: the cpu to run on | |
3638 | * @fn: the function to run | |
3639 | * @arg: the function arg | |
3640 | * | |
31ad9081 RR |
3641 | * This will return the value @fn returns. |
3642 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 3643 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 RR |
3644 | */ |
3645 | long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) | |
3646 | { | |
6b44003e AM |
3647 | struct task_struct *sub_thread; |
3648 | struct work_for_cpu wfc = { | |
3649 | .completion = COMPLETION_INITIALIZER_ONSTACK(wfc.completion), | |
3650 | .fn = fn, | |
3651 | .arg = arg, | |
3652 | }; | |
3653 | ||
3654 | sub_thread = kthread_create(do_work_for_cpu, &wfc, "work_for_cpu"); | |
3655 | if (IS_ERR(sub_thread)) | |
3656 | return PTR_ERR(sub_thread); | |
3657 | kthread_bind(sub_thread, cpu); | |
3658 | wake_up_process(sub_thread); | |
3659 | wait_for_completion(&wfc.completion); | |
2d3854a3 RR |
3660 | return wfc.ret; |
3661 | } | |
3662 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
3663 | #endif /* CONFIG_SMP */ | |
3664 | ||
a0a1a5fd TH |
3665 | #ifdef CONFIG_FREEZER |
3666 | ||
3667 | /** | |
3668 | * freeze_workqueues_begin - begin freezing workqueues | |
3669 | * | |
58a69cb4 TH |
3670 | * Start freezing workqueues. After this function returns, all freezable |
3671 | * workqueues will queue new works to their frozen_works list instead of | |
3672 | * gcwq->worklist. | |
a0a1a5fd TH |
3673 | * |
3674 | * CONTEXT: | |
8b03ae3c | 3675 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3676 | */ |
3677 | void freeze_workqueues_begin(void) | |
3678 | { | |
a0a1a5fd TH |
3679 | unsigned int cpu; |
3680 | ||
3681 | spin_lock(&workqueue_lock); | |
3682 | ||
3683 | BUG_ON(workqueue_freezing); | |
3684 | workqueue_freezing = true; | |
3685 | ||
f3421797 | 3686 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3687 | struct global_cwq *gcwq = get_gcwq(cpu); |
bdbc5dd7 | 3688 | struct workqueue_struct *wq; |
8b03ae3c TH |
3689 | |
3690 | spin_lock_irq(&gcwq->lock); | |
3691 | ||
db7bccf4 TH |
3692 | BUG_ON(gcwq->flags & GCWQ_FREEZING); |
3693 | gcwq->flags |= GCWQ_FREEZING; | |
3694 | ||
a0a1a5fd TH |
3695 | list_for_each_entry(wq, &workqueues, list) { |
3696 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3697 | ||
58a69cb4 | 3698 | if (cwq && wq->flags & WQ_FREEZABLE) |
a0a1a5fd | 3699 | cwq->max_active = 0; |
a0a1a5fd | 3700 | } |
8b03ae3c TH |
3701 | |
3702 | spin_unlock_irq(&gcwq->lock); | |
a0a1a5fd TH |
3703 | } |
3704 | ||
3705 | spin_unlock(&workqueue_lock); | |
3706 | } | |
3707 | ||
3708 | /** | |
58a69cb4 | 3709 | * freeze_workqueues_busy - are freezable workqueues still busy? |
a0a1a5fd TH |
3710 | * |
3711 | * Check whether freezing is complete. This function must be called | |
3712 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
3713 | * | |
3714 | * CONTEXT: | |
3715 | * Grabs and releases workqueue_lock. | |
3716 | * | |
3717 | * RETURNS: | |
58a69cb4 TH |
3718 | * %true if some freezable workqueues are still busy. %false if freezing |
3719 | * is complete. | |
a0a1a5fd TH |
3720 | */ |
3721 | bool freeze_workqueues_busy(void) | |
3722 | { | |
a0a1a5fd TH |
3723 | unsigned int cpu; |
3724 | bool busy = false; | |
3725 | ||
3726 | spin_lock(&workqueue_lock); | |
3727 | ||
3728 | BUG_ON(!workqueue_freezing); | |
3729 | ||
f3421797 | 3730 | for_each_gcwq_cpu(cpu) { |
bdbc5dd7 | 3731 | struct workqueue_struct *wq; |
a0a1a5fd TH |
3732 | /* |
3733 | * nr_active is monotonically decreasing. It's safe | |
3734 | * to peek without lock. | |
3735 | */ | |
3736 | list_for_each_entry(wq, &workqueues, list) { | |
3737 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3738 | ||
58a69cb4 | 3739 | if (!cwq || !(wq->flags & WQ_FREEZABLE)) |
a0a1a5fd TH |
3740 | continue; |
3741 | ||
3742 | BUG_ON(cwq->nr_active < 0); | |
3743 | if (cwq->nr_active) { | |
3744 | busy = true; | |
3745 | goto out_unlock; | |
3746 | } | |
3747 | } | |
3748 | } | |
3749 | out_unlock: | |
3750 | spin_unlock(&workqueue_lock); | |
3751 | return busy; | |
3752 | } | |
3753 | ||
3754 | /** | |
3755 | * thaw_workqueues - thaw workqueues | |
3756 | * | |
3757 | * Thaw workqueues. Normal queueing is restored and all collected | |
7e11629d | 3758 | * frozen works are transferred to their respective gcwq worklists. |
a0a1a5fd TH |
3759 | * |
3760 | * CONTEXT: | |
8b03ae3c | 3761 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3762 | */ |
3763 | void thaw_workqueues(void) | |
3764 | { | |
a0a1a5fd TH |
3765 | unsigned int cpu; |
3766 | ||
3767 | spin_lock(&workqueue_lock); | |
3768 | ||
3769 | if (!workqueue_freezing) | |
3770 | goto out_unlock; | |
3771 | ||
f3421797 | 3772 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3773 | struct global_cwq *gcwq = get_gcwq(cpu); |
bdbc5dd7 | 3774 | struct workqueue_struct *wq; |
8b03ae3c TH |
3775 | |
3776 | spin_lock_irq(&gcwq->lock); | |
3777 | ||
db7bccf4 TH |
3778 | BUG_ON(!(gcwq->flags & GCWQ_FREEZING)); |
3779 | gcwq->flags &= ~GCWQ_FREEZING; | |
3780 | ||
a0a1a5fd TH |
3781 | list_for_each_entry(wq, &workqueues, list) { |
3782 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3783 | ||
58a69cb4 | 3784 | if (!cwq || !(wq->flags & WQ_FREEZABLE)) |
a0a1a5fd TH |
3785 | continue; |
3786 | ||
a0a1a5fd TH |
3787 | /* restore max_active and repopulate worklist */ |
3788 | cwq->max_active = wq->saved_max_active; | |
3789 | ||
3790 | while (!list_empty(&cwq->delayed_works) && | |
3791 | cwq->nr_active < cwq->max_active) | |
3792 | cwq_activate_first_delayed(cwq); | |
a0a1a5fd | 3793 | } |
8b03ae3c | 3794 | |
63d95a91 | 3795 | wake_up_worker(&gcwq->pool); |
e22bee78 | 3796 | |
8b03ae3c | 3797 | spin_unlock_irq(&gcwq->lock); |
a0a1a5fd TH |
3798 | } |
3799 | ||
3800 | workqueue_freezing = false; | |
3801 | out_unlock: | |
3802 | spin_unlock(&workqueue_lock); | |
3803 | } | |
3804 | #endif /* CONFIG_FREEZER */ | |
3805 | ||
6ee0578b | 3806 | static int __init init_workqueues(void) |
1da177e4 | 3807 | { |
c34056a3 | 3808 | unsigned int cpu; |
c8e55f36 | 3809 | int i; |
c34056a3 | 3810 | |
f6500947 | 3811 | cpu_notifier(workqueue_cpu_callback, CPU_PRI_WORKQUEUE); |
8b03ae3c TH |
3812 | |
3813 | /* initialize gcwqs */ | |
f3421797 | 3814 | for_each_gcwq_cpu(cpu) { |
8b03ae3c TH |
3815 | struct global_cwq *gcwq = get_gcwq(cpu); |
3816 | ||
3817 | spin_lock_init(&gcwq->lock); | |
bd7bdd43 TH |
3818 | gcwq->pool.gcwq = gcwq; |
3819 | INIT_LIST_HEAD(&gcwq->pool.worklist); | |
8b03ae3c | 3820 | gcwq->cpu = cpu; |
477a3c33 | 3821 | gcwq->flags |= GCWQ_DISASSOCIATED; |
8b03ae3c | 3822 | |
bd7bdd43 | 3823 | INIT_LIST_HEAD(&gcwq->pool.idle_list); |
c8e55f36 TH |
3824 | for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++) |
3825 | INIT_HLIST_HEAD(&gcwq->busy_hash[i]); | |
3826 | ||
bd7bdd43 TH |
3827 | init_timer_deferrable(&gcwq->pool.idle_timer); |
3828 | gcwq->pool.idle_timer.function = idle_worker_timeout; | |
63d95a91 | 3829 | gcwq->pool.idle_timer.data = (unsigned long)&gcwq->pool; |
e7577c50 | 3830 | |
bd7bdd43 | 3831 | setup_timer(&gcwq->pool.mayday_timer, gcwq_mayday_timeout, |
63d95a91 | 3832 | (unsigned long)&gcwq->pool); |
e22bee78 | 3833 | |
bd7bdd43 | 3834 | ida_init(&gcwq->pool.worker_ida); |
db7bccf4 TH |
3835 | |
3836 | gcwq->trustee_state = TRUSTEE_DONE; | |
3837 | init_waitqueue_head(&gcwq->trustee_wait); | |
8b03ae3c TH |
3838 | } |
3839 | ||
e22bee78 | 3840 | /* create the initial worker */ |
f3421797 | 3841 | for_each_online_gcwq_cpu(cpu) { |
e22bee78 TH |
3842 | struct global_cwq *gcwq = get_gcwq(cpu); |
3843 | struct worker *worker; | |
3844 | ||
477a3c33 TH |
3845 | if (cpu != WORK_CPU_UNBOUND) |
3846 | gcwq->flags &= ~GCWQ_DISASSOCIATED; | |
63d95a91 | 3847 | worker = create_worker(&gcwq->pool, true); |
e22bee78 TH |
3848 | BUG_ON(!worker); |
3849 | spin_lock_irq(&gcwq->lock); | |
3850 | start_worker(worker); | |
3851 | spin_unlock_irq(&gcwq->lock); | |
3852 | } | |
3853 | ||
d320c038 TH |
3854 | system_wq = alloc_workqueue("events", 0, 0); |
3855 | system_long_wq = alloc_workqueue("events_long", 0, 0); | |
3856 | system_nrt_wq = alloc_workqueue("events_nrt", WQ_NON_REENTRANT, 0); | |
f3421797 TH |
3857 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
3858 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
3859 | system_freezable_wq = alloc_workqueue("events_freezable", |
3860 | WQ_FREEZABLE, 0); | |
62d3c543 AS |
3861 | system_nrt_freezable_wq = alloc_workqueue("events_nrt_freezable", |
3862 | WQ_NON_REENTRANT | WQ_FREEZABLE, 0); | |
e5cba24e | 3863 | BUG_ON(!system_wq || !system_long_wq || !system_nrt_wq || |
62d3c543 AS |
3864 | !system_unbound_wq || !system_freezable_wq || |
3865 | !system_nrt_freezable_wq); | |
6ee0578b | 3866 | return 0; |
1da177e4 | 3867 | } |
6ee0578b | 3868 | early_initcall(init_workqueues); |