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 | ||
26 | #include <linux/module.h> | |
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 | ||
82 | MAYDAY_INITIAL_TIMEOUT = HZ / 100, /* call for help after 10ms */ | |
83 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ | |
84 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
db7bccf4 | 85 | TRUSTEE_COOLDOWN = HZ / 10, /* for trustee draining */ |
e22bee78 TH |
86 | |
87 | /* | |
88 | * Rescue workers are used only on emergencies and shared by | |
89 | * all cpus. Give -20. | |
90 | */ | |
91 | RESCUER_NICE_LEVEL = -20, | |
c8e55f36 | 92 | }; |
1da177e4 LT |
93 | |
94 | /* | |
4690c4ab TH |
95 | * Structure fields follow one of the following exclusion rules. |
96 | * | |
e41e704b TH |
97 | * I: Modifiable by initialization/destruction paths and read-only for |
98 | * everyone else. | |
4690c4ab | 99 | * |
e22bee78 TH |
100 | * P: Preemption protected. Disabling preemption is enough and should |
101 | * only be modified and accessed from the local cpu. | |
102 | * | |
8b03ae3c | 103 | * L: gcwq->lock protected. Access with gcwq->lock held. |
4690c4ab | 104 | * |
e22bee78 TH |
105 | * X: During normal operation, modification requires gcwq->lock and |
106 | * should be done only from local cpu. Either disabling preemption | |
107 | * on local cpu or grabbing gcwq->lock is enough for read access. | |
f3421797 | 108 | * If GCWQ_DISASSOCIATED is set, it's identical to L. |
e22bee78 | 109 | * |
73f53c4a TH |
110 | * F: wq->flush_mutex protected. |
111 | * | |
4690c4ab | 112 | * W: workqueue_lock protected. |
1da177e4 | 113 | */ |
1da177e4 | 114 | |
8b03ae3c | 115 | struct global_cwq; |
1da177e4 | 116 | |
e22bee78 TH |
117 | /* |
118 | * The poor guys doing the actual heavy lifting. All on-duty workers | |
119 | * are either serving the manager role, on idle list or on busy hash. | |
120 | */ | |
c34056a3 | 121 | struct worker { |
c8e55f36 TH |
122 | /* on idle list while idle, on busy hash table while busy */ |
123 | union { | |
124 | struct list_head entry; /* L: while idle */ | |
125 | struct hlist_node hentry; /* L: while busy */ | |
126 | }; | |
1da177e4 | 127 | |
c34056a3 | 128 | struct work_struct *current_work; /* L: work being processed */ |
8cca0eea | 129 | struct cpu_workqueue_struct *current_cwq; /* L: current_work's cwq */ |
affee4b2 | 130 | struct list_head scheduled; /* L: scheduled works */ |
c34056a3 | 131 | struct task_struct *task; /* I: worker task */ |
8b03ae3c | 132 | struct global_cwq *gcwq; /* I: the associated gcwq */ |
e22bee78 TH |
133 | /* 64 bytes boundary on 64bit, 32 on 32bit */ |
134 | unsigned long last_active; /* L: last active timestamp */ | |
135 | unsigned int flags; /* X: flags */ | |
c34056a3 | 136 | int id; /* I: worker id */ |
e22bee78 | 137 | struct work_struct rebind_work; /* L: rebind worker to cpu */ |
c34056a3 TH |
138 | }; |
139 | ||
8b03ae3c | 140 | /* |
e22bee78 TH |
141 | * Global per-cpu workqueue. There's one and only one for each cpu |
142 | * and all works are queued and processed here regardless of their | |
143 | * target workqueues. | |
8b03ae3c TH |
144 | */ |
145 | struct global_cwq { | |
146 | spinlock_t lock; /* the gcwq lock */ | |
7e11629d | 147 | struct list_head worklist; /* L: list of pending works */ |
8b03ae3c | 148 | unsigned int cpu; /* I: the associated cpu */ |
db7bccf4 | 149 | unsigned int flags; /* L: GCWQ_* flags */ |
c8e55f36 TH |
150 | |
151 | int nr_workers; /* L: total number of workers */ | |
152 | int nr_idle; /* L: currently idle ones */ | |
153 | ||
154 | /* workers are chained either in the idle_list or busy_hash */ | |
e22bee78 | 155 | struct list_head idle_list; /* X: list of idle workers */ |
c8e55f36 TH |
156 | struct hlist_head busy_hash[BUSY_WORKER_HASH_SIZE]; |
157 | /* L: hash of busy workers */ | |
158 | ||
e22bee78 TH |
159 | struct timer_list idle_timer; /* L: worker idle timeout */ |
160 | struct timer_list mayday_timer; /* L: SOS timer for dworkers */ | |
161 | ||
8b03ae3c | 162 | struct ida worker_ida; /* L: for worker IDs */ |
db7bccf4 TH |
163 | |
164 | struct task_struct *trustee; /* L: for gcwq shutdown */ | |
165 | unsigned int trustee_state; /* L: trustee state */ | |
166 | wait_queue_head_t trustee_wait; /* trustee wait */ | |
e22bee78 | 167 | struct worker *first_idle; /* L: first idle worker */ |
8b03ae3c TH |
168 | } ____cacheline_aligned_in_smp; |
169 | ||
1da177e4 | 170 | /* |
502ca9d8 | 171 | * The per-CPU workqueue. The lower WORK_STRUCT_FLAG_BITS of |
0f900049 TH |
172 | * work_struct->data are used for flags and thus cwqs need to be |
173 | * aligned at two's power of the number of flag bits. | |
1da177e4 LT |
174 | */ |
175 | struct cpu_workqueue_struct { | |
8b03ae3c | 176 | struct global_cwq *gcwq; /* I: the associated gcwq */ |
4690c4ab | 177 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
178 | int work_color; /* L: current color */ |
179 | int flush_color; /* L: flushing color */ | |
180 | int nr_in_flight[WORK_NR_COLORS]; | |
181 | /* L: nr of in_flight works */ | |
1e19ffc6 | 182 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 183 | int max_active; /* L: max active works */ |
1e19ffc6 | 184 | struct list_head delayed_works; /* L: delayed works */ |
0f900049 | 185 | }; |
1da177e4 | 186 | |
73f53c4a TH |
187 | /* |
188 | * Structure used to wait for workqueue flush. | |
189 | */ | |
190 | struct wq_flusher { | |
191 | struct list_head list; /* F: list of flushers */ | |
192 | int flush_color; /* F: flush color waiting for */ | |
193 | struct completion done; /* flush completion */ | |
194 | }; | |
195 | ||
f2e005aa TH |
196 | /* |
197 | * All cpumasks are assumed to be always set on UP and thus can't be | |
198 | * used to determine whether there's something to be done. | |
199 | */ | |
200 | #ifdef CONFIG_SMP | |
201 | typedef cpumask_var_t mayday_mask_t; | |
202 | #define mayday_test_and_set_cpu(cpu, mask) \ | |
203 | cpumask_test_and_set_cpu((cpu), (mask)) | |
204 | #define mayday_clear_cpu(cpu, mask) cpumask_clear_cpu((cpu), (mask)) | |
205 | #define for_each_mayday_cpu(cpu, mask) for_each_cpu((cpu), (mask)) | |
9c37547a | 206 | #define alloc_mayday_mask(maskp, gfp) zalloc_cpumask_var((maskp), (gfp)) |
f2e005aa TH |
207 | #define free_mayday_mask(mask) free_cpumask_var((mask)) |
208 | #else | |
209 | typedef unsigned long mayday_mask_t; | |
210 | #define mayday_test_and_set_cpu(cpu, mask) test_and_set_bit(0, &(mask)) | |
211 | #define mayday_clear_cpu(cpu, mask) clear_bit(0, &(mask)) | |
212 | #define for_each_mayday_cpu(cpu, mask) if ((cpu) = 0, (mask)) | |
213 | #define alloc_mayday_mask(maskp, gfp) true | |
214 | #define free_mayday_mask(mask) do { } while (0) | |
215 | #endif | |
1da177e4 LT |
216 | |
217 | /* | |
218 | * The externally visible workqueue abstraction is an array of | |
219 | * per-CPU workqueues: | |
220 | */ | |
221 | struct workqueue_struct { | |
97e37d7b | 222 | unsigned int flags; /* I: WQ_* flags */ |
bdbc5dd7 TH |
223 | union { |
224 | struct cpu_workqueue_struct __percpu *pcpu; | |
225 | struct cpu_workqueue_struct *single; | |
226 | unsigned long v; | |
227 | } cpu_wq; /* I: cwq's */ | |
4690c4ab | 228 | struct list_head list; /* W: list of all workqueues */ |
73f53c4a TH |
229 | |
230 | struct mutex flush_mutex; /* protects wq flushing */ | |
231 | int work_color; /* F: current work color */ | |
232 | int flush_color; /* F: current flush color */ | |
233 | atomic_t nr_cwqs_to_flush; /* flush in progress */ | |
234 | struct wq_flusher *first_flusher; /* F: first flusher */ | |
235 | struct list_head flusher_queue; /* F: flush waiters */ | |
236 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
237 | ||
f2e005aa | 238 | mayday_mask_t mayday_mask; /* cpus requesting rescue */ |
e22bee78 TH |
239 | struct worker *rescuer; /* I: rescue worker */ |
240 | ||
dcd989cb | 241 | int saved_max_active; /* W: saved cwq max_active */ |
4690c4ab | 242 | const char *name; /* I: workqueue name */ |
4e6045f1 | 243 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 244 | struct lockdep_map lockdep_map; |
4e6045f1 | 245 | #endif |
1da177e4 LT |
246 | }; |
247 | ||
d320c038 TH |
248 | struct workqueue_struct *system_wq __read_mostly; |
249 | struct workqueue_struct *system_long_wq __read_mostly; | |
250 | struct workqueue_struct *system_nrt_wq __read_mostly; | |
f3421797 | 251 | struct workqueue_struct *system_unbound_wq __read_mostly; |
d320c038 TH |
252 | EXPORT_SYMBOL_GPL(system_wq); |
253 | EXPORT_SYMBOL_GPL(system_long_wq); | |
254 | EXPORT_SYMBOL_GPL(system_nrt_wq); | |
f3421797 | 255 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
d320c038 | 256 | |
97bd2347 TH |
257 | #define CREATE_TRACE_POINTS |
258 | #include <trace/events/workqueue.h> | |
259 | ||
db7bccf4 TH |
260 | #define for_each_busy_worker(worker, i, pos, gcwq) \ |
261 | for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++) \ | |
262 | hlist_for_each_entry(worker, pos, &gcwq->busy_hash[i], hentry) | |
263 | ||
f3421797 TH |
264 | static inline int __next_gcwq_cpu(int cpu, const struct cpumask *mask, |
265 | unsigned int sw) | |
266 | { | |
267 | if (cpu < nr_cpu_ids) { | |
268 | if (sw & 1) { | |
269 | cpu = cpumask_next(cpu, mask); | |
270 | if (cpu < nr_cpu_ids) | |
271 | return cpu; | |
272 | } | |
273 | if (sw & 2) | |
274 | return WORK_CPU_UNBOUND; | |
275 | } | |
276 | return WORK_CPU_NONE; | |
277 | } | |
278 | ||
279 | static inline int __next_wq_cpu(int cpu, const struct cpumask *mask, | |
280 | struct workqueue_struct *wq) | |
281 | { | |
282 | return __next_gcwq_cpu(cpu, mask, !(wq->flags & WQ_UNBOUND) ? 1 : 2); | |
283 | } | |
284 | ||
09884951 TH |
285 | /* |
286 | * CPU iterators | |
287 | * | |
288 | * An extra gcwq is defined for an invalid cpu number | |
289 | * (WORK_CPU_UNBOUND) to host workqueues which are not bound to any | |
290 | * specific CPU. The following iterators are similar to | |
291 | * for_each_*_cpu() iterators but also considers the unbound gcwq. | |
292 | * | |
293 | * for_each_gcwq_cpu() : possible CPUs + WORK_CPU_UNBOUND | |
294 | * for_each_online_gcwq_cpu() : online CPUs + WORK_CPU_UNBOUND | |
295 | * for_each_cwq_cpu() : possible CPUs for bound workqueues, | |
296 | * WORK_CPU_UNBOUND for unbound workqueues | |
297 | */ | |
f3421797 TH |
298 | #define for_each_gcwq_cpu(cpu) \ |
299 | for ((cpu) = __next_gcwq_cpu(-1, cpu_possible_mask, 3); \ | |
300 | (cpu) < WORK_CPU_NONE; \ | |
301 | (cpu) = __next_gcwq_cpu((cpu), cpu_possible_mask, 3)) | |
302 | ||
303 | #define for_each_online_gcwq_cpu(cpu) \ | |
304 | for ((cpu) = __next_gcwq_cpu(-1, cpu_online_mask, 3); \ | |
305 | (cpu) < WORK_CPU_NONE; \ | |
306 | (cpu) = __next_gcwq_cpu((cpu), cpu_online_mask, 3)) | |
307 | ||
308 | #define for_each_cwq_cpu(cpu, wq) \ | |
309 | for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, (wq)); \ | |
310 | (cpu) < WORK_CPU_NONE; \ | |
311 | (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, (wq))) | |
312 | ||
a25909a4 PM |
313 | #ifdef CONFIG_LOCKDEP |
314 | /** | |
315 | * in_workqueue_context() - in context of specified workqueue? | |
316 | * @wq: the workqueue of interest | |
317 | * | |
318 | * Checks lockdep state to see if the current task is executing from | |
319 | * within a workqueue item. This function exists only if lockdep is | |
320 | * enabled. | |
321 | */ | |
322 | int in_workqueue_context(struct workqueue_struct *wq) | |
323 | { | |
324 | return lock_is_held(&wq->lockdep_map); | |
325 | } | |
326 | #endif | |
327 | ||
dc186ad7 TG |
328 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
329 | ||
330 | static struct debug_obj_descr work_debug_descr; | |
331 | ||
332 | /* | |
333 | * fixup_init is called when: | |
334 | * - an active object is initialized | |
335 | */ | |
336 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
337 | { | |
338 | struct work_struct *work = addr; | |
339 | ||
340 | switch (state) { | |
341 | case ODEBUG_STATE_ACTIVE: | |
342 | cancel_work_sync(work); | |
343 | debug_object_init(work, &work_debug_descr); | |
344 | return 1; | |
345 | default: | |
346 | return 0; | |
347 | } | |
348 | } | |
349 | ||
350 | /* | |
351 | * fixup_activate is called when: | |
352 | * - an active object is activated | |
353 | * - an unknown object is activated (might be a statically initialized object) | |
354 | */ | |
355 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
356 | { | |
357 | struct work_struct *work = addr; | |
358 | ||
359 | switch (state) { | |
360 | ||
361 | case ODEBUG_STATE_NOTAVAILABLE: | |
362 | /* | |
363 | * This is not really a fixup. The work struct was | |
364 | * statically initialized. We just make sure that it | |
365 | * is tracked in the object tracker. | |
366 | */ | |
22df02bb | 367 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
368 | debug_object_init(work, &work_debug_descr); |
369 | debug_object_activate(work, &work_debug_descr); | |
370 | return 0; | |
371 | } | |
372 | WARN_ON_ONCE(1); | |
373 | return 0; | |
374 | ||
375 | case ODEBUG_STATE_ACTIVE: | |
376 | WARN_ON(1); | |
377 | ||
378 | default: | |
379 | return 0; | |
380 | } | |
381 | } | |
382 | ||
383 | /* | |
384 | * fixup_free is called when: | |
385 | * - an active object is freed | |
386 | */ | |
387 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
388 | { | |
389 | struct work_struct *work = addr; | |
390 | ||
391 | switch (state) { | |
392 | case ODEBUG_STATE_ACTIVE: | |
393 | cancel_work_sync(work); | |
394 | debug_object_free(work, &work_debug_descr); | |
395 | return 1; | |
396 | default: | |
397 | return 0; | |
398 | } | |
399 | } | |
400 | ||
401 | static struct debug_obj_descr work_debug_descr = { | |
402 | .name = "work_struct", | |
403 | .fixup_init = work_fixup_init, | |
404 | .fixup_activate = work_fixup_activate, | |
405 | .fixup_free = work_fixup_free, | |
406 | }; | |
407 | ||
408 | static inline void debug_work_activate(struct work_struct *work) | |
409 | { | |
410 | debug_object_activate(work, &work_debug_descr); | |
411 | } | |
412 | ||
413 | static inline void debug_work_deactivate(struct work_struct *work) | |
414 | { | |
415 | debug_object_deactivate(work, &work_debug_descr); | |
416 | } | |
417 | ||
418 | void __init_work(struct work_struct *work, int onstack) | |
419 | { | |
420 | if (onstack) | |
421 | debug_object_init_on_stack(work, &work_debug_descr); | |
422 | else | |
423 | debug_object_init(work, &work_debug_descr); | |
424 | } | |
425 | EXPORT_SYMBOL_GPL(__init_work); | |
426 | ||
427 | void destroy_work_on_stack(struct work_struct *work) | |
428 | { | |
429 | debug_object_free(work, &work_debug_descr); | |
430 | } | |
431 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
432 | ||
433 | #else | |
434 | static inline void debug_work_activate(struct work_struct *work) { } | |
435 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
436 | #endif | |
437 | ||
95402b38 GS |
438 | /* Serializes the accesses to the list of workqueues. */ |
439 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 440 | static LIST_HEAD(workqueues); |
a0a1a5fd | 441 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 442 | |
e22bee78 TH |
443 | /* |
444 | * The almighty global cpu workqueues. nr_running is the only field | |
445 | * which is expected to be used frequently by other cpus via | |
446 | * try_to_wake_up(). Put it in a separate cacheline. | |
447 | */ | |
8b03ae3c | 448 | static DEFINE_PER_CPU(struct global_cwq, global_cwq); |
e22bee78 | 449 | static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, gcwq_nr_running); |
8b03ae3c | 450 | |
f3421797 TH |
451 | /* |
452 | * Global cpu workqueue and nr_running counter for unbound gcwq. The | |
453 | * gcwq is always online, has GCWQ_DISASSOCIATED set, and all its | |
454 | * workers have WORKER_UNBOUND set. | |
455 | */ | |
456 | static struct global_cwq unbound_global_cwq; | |
457 | static atomic_t unbound_gcwq_nr_running = ATOMIC_INIT(0); /* always 0 */ | |
458 | ||
c34056a3 | 459 | static int worker_thread(void *__worker); |
1da177e4 | 460 | |
8b03ae3c TH |
461 | static struct global_cwq *get_gcwq(unsigned int cpu) |
462 | { | |
f3421797 TH |
463 | if (cpu != WORK_CPU_UNBOUND) |
464 | return &per_cpu(global_cwq, cpu); | |
465 | else | |
466 | return &unbound_global_cwq; | |
8b03ae3c TH |
467 | } |
468 | ||
e22bee78 TH |
469 | static atomic_t *get_gcwq_nr_running(unsigned int cpu) |
470 | { | |
f3421797 TH |
471 | if (cpu != WORK_CPU_UNBOUND) |
472 | return &per_cpu(gcwq_nr_running, cpu); | |
473 | else | |
474 | return &unbound_gcwq_nr_running; | |
e22bee78 TH |
475 | } |
476 | ||
1537663f TH |
477 | static struct cpu_workqueue_struct *get_cwq(unsigned int cpu, |
478 | struct workqueue_struct *wq) | |
b1f4ec17 | 479 | { |
f3421797 TH |
480 | if (!(wq->flags & WQ_UNBOUND)) { |
481 | if (likely(cpu < nr_cpu_ids)) { | |
482 | #ifdef CONFIG_SMP | |
483 | return per_cpu_ptr(wq->cpu_wq.pcpu, cpu); | |
bdbc5dd7 | 484 | #else |
f3421797 | 485 | return wq->cpu_wq.single; |
bdbc5dd7 | 486 | #endif |
f3421797 TH |
487 | } |
488 | } else if (likely(cpu == WORK_CPU_UNBOUND)) | |
489 | return wq->cpu_wq.single; | |
490 | return NULL; | |
b1f4ec17 ON |
491 | } |
492 | ||
73f53c4a TH |
493 | static unsigned int work_color_to_flags(int color) |
494 | { | |
495 | return color << WORK_STRUCT_COLOR_SHIFT; | |
496 | } | |
497 | ||
498 | static int get_work_color(struct work_struct *work) | |
499 | { | |
500 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
501 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
502 | } | |
503 | ||
504 | static int work_next_color(int color) | |
505 | { | |
506 | return (color + 1) % WORK_NR_COLORS; | |
507 | } | |
1da177e4 | 508 | |
14441960 | 509 | /* |
e120153d TH |
510 | * A work's data points to the cwq with WORK_STRUCT_CWQ set while the |
511 | * work is on queue. Once execution starts, WORK_STRUCT_CWQ is | |
512 | * cleared and the work data contains the cpu number it was last on. | |
7a22ad75 TH |
513 | * |
514 | * set_work_{cwq|cpu}() and clear_work_data() can be used to set the | |
515 | * cwq, cpu or clear work->data. These functions should only be | |
516 | * called while the work is owned - ie. while the PENDING bit is set. | |
517 | * | |
518 | * get_work_[g]cwq() can be used to obtain the gcwq or cwq | |
519 | * corresponding to a work. gcwq is available once the work has been | |
520 | * queued anywhere after initialization. cwq is available only from | |
521 | * queueing until execution starts. | |
14441960 | 522 | */ |
7a22ad75 TH |
523 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
524 | unsigned long flags) | |
365970a1 | 525 | { |
4594bf15 | 526 | BUG_ON(!work_pending(work)); |
7a22ad75 TH |
527 | atomic_long_set(&work->data, data | flags | work_static(work)); |
528 | } | |
365970a1 | 529 | |
7a22ad75 TH |
530 | static void set_work_cwq(struct work_struct *work, |
531 | struct cpu_workqueue_struct *cwq, | |
532 | unsigned long extra_flags) | |
533 | { | |
534 | set_work_data(work, (unsigned long)cwq, | |
e120153d | 535 | WORK_STRUCT_PENDING | WORK_STRUCT_CWQ | extra_flags); |
365970a1 DH |
536 | } |
537 | ||
7a22ad75 TH |
538 | static void set_work_cpu(struct work_struct *work, unsigned int cpu) |
539 | { | |
540 | set_work_data(work, cpu << WORK_STRUCT_FLAG_BITS, WORK_STRUCT_PENDING); | |
541 | } | |
f756d5e2 | 542 | |
7a22ad75 | 543 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 544 | { |
7a22ad75 | 545 | set_work_data(work, WORK_STRUCT_NO_CPU, 0); |
1da177e4 LT |
546 | } |
547 | ||
7a22ad75 | 548 | static struct cpu_workqueue_struct *get_work_cwq(struct work_struct *work) |
b1f4ec17 | 549 | { |
e120153d | 550 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 551 | |
e120153d TH |
552 | if (data & WORK_STRUCT_CWQ) |
553 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); | |
554 | else | |
555 | return NULL; | |
4d707b9f ON |
556 | } |
557 | ||
7a22ad75 | 558 | static struct global_cwq *get_work_gcwq(struct work_struct *work) |
365970a1 | 559 | { |
e120153d | 560 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 TH |
561 | unsigned int cpu; |
562 | ||
e120153d TH |
563 | if (data & WORK_STRUCT_CWQ) |
564 | return ((struct cpu_workqueue_struct *) | |
565 | (data & WORK_STRUCT_WQ_DATA_MASK))->gcwq; | |
7a22ad75 TH |
566 | |
567 | cpu = data >> WORK_STRUCT_FLAG_BITS; | |
bdbc5dd7 | 568 | if (cpu == WORK_CPU_NONE) |
7a22ad75 TH |
569 | return NULL; |
570 | ||
f3421797 | 571 | BUG_ON(cpu >= nr_cpu_ids && cpu != WORK_CPU_UNBOUND); |
7a22ad75 | 572 | return get_gcwq(cpu); |
b1f4ec17 ON |
573 | } |
574 | ||
e22bee78 TH |
575 | /* |
576 | * Policy functions. These define the policies on how the global | |
577 | * worker pool is managed. Unless noted otherwise, these functions | |
578 | * assume that they're being called with gcwq->lock held. | |
579 | */ | |
580 | ||
649027d7 | 581 | static bool __need_more_worker(struct global_cwq *gcwq) |
a848e3b6 | 582 | { |
649027d7 TH |
583 | return !atomic_read(get_gcwq_nr_running(gcwq->cpu)) || |
584 | gcwq->flags & GCWQ_HIGHPRI_PENDING; | |
a848e3b6 ON |
585 | } |
586 | ||
4594bf15 | 587 | /* |
e22bee78 TH |
588 | * Need to wake up a worker? Called from anything but currently |
589 | * running workers. | |
4594bf15 | 590 | */ |
e22bee78 | 591 | static bool need_more_worker(struct global_cwq *gcwq) |
365970a1 | 592 | { |
649027d7 | 593 | return !list_empty(&gcwq->worklist) && __need_more_worker(gcwq); |
e22bee78 | 594 | } |
4594bf15 | 595 | |
e22bee78 TH |
596 | /* Can I start working? Called from busy but !running workers. */ |
597 | static bool may_start_working(struct global_cwq *gcwq) | |
598 | { | |
599 | return gcwq->nr_idle; | |
600 | } | |
601 | ||
602 | /* Do I need to keep working? Called from currently running workers. */ | |
603 | static bool keep_working(struct global_cwq *gcwq) | |
604 | { | |
605 | atomic_t *nr_running = get_gcwq_nr_running(gcwq->cpu); | |
606 | ||
30310045 TH |
607 | return !list_empty(&gcwq->worklist) && |
608 | (atomic_read(nr_running) <= 1 || | |
609 | gcwq->flags & GCWQ_HIGHPRI_PENDING); | |
e22bee78 TH |
610 | } |
611 | ||
612 | /* Do we need a new worker? Called from manager. */ | |
613 | static bool need_to_create_worker(struct global_cwq *gcwq) | |
614 | { | |
615 | return need_more_worker(gcwq) && !may_start_working(gcwq); | |
616 | } | |
365970a1 | 617 | |
e22bee78 TH |
618 | /* Do I need to be the manager? */ |
619 | static bool need_to_manage_workers(struct global_cwq *gcwq) | |
620 | { | |
621 | return need_to_create_worker(gcwq) || gcwq->flags & GCWQ_MANAGE_WORKERS; | |
622 | } | |
623 | ||
624 | /* Do we have too many workers and should some go away? */ | |
625 | static bool too_many_workers(struct global_cwq *gcwq) | |
626 | { | |
627 | bool managing = gcwq->flags & GCWQ_MANAGING_WORKERS; | |
628 | int nr_idle = gcwq->nr_idle + managing; /* manager is considered idle */ | |
629 | int nr_busy = gcwq->nr_workers - nr_idle; | |
630 | ||
631 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; | |
365970a1 DH |
632 | } |
633 | ||
4d707b9f | 634 | /* |
e22bee78 TH |
635 | * Wake up functions. |
636 | */ | |
637 | ||
7e11629d TH |
638 | /* Return the first worker. Safe with preemption disabled */ |
639 | static struct worker *first_worker(struct global_cwq *gcwq) | |
640 | { | |
641 | if (unlikely(list_empty(&gcwq->idle_list))) | |
642 | return NULL; | |
643 | ||
644 | return list_first_entry(&gcwq->idle_list, struct worker, entry); | |
645 | } | |
646 | ||
647 | /** | |
648 | * wake_up_worker - wake up an idle worker | |
649 | * @gcwq: gcwq to wake worker for | |
650 | * | |
651 | * Wake up the first idle worker of @gcwq. | |
652 | * | |
653 | * CONTEXT: | |
654 | * spin_lock_irq(gcwq->lock). | |
655 | */ | |
656 | static void wake_up_worker(struct global_cwq *gcwq) | |
657 | { | |
658 | struct worker *worker = first_worker(gcwq); | |
659 | ||
660 | if (likely(worker)) | |
661 | wake_up_process(worker->task); | |
662 | } | |
663 | ||
d302f017 | 664 | /** |
e22bee78 TH |
665 | * wq_worker_waking_up - a worker is waking up |
666 | * @task: task waking up | |
667 | * @cpu: CPU @task is waking up to | |
668 | * | |
669 | * This function is called during try_to_wake_up() when a worker is | |
670 | * being awoken. | |
671 | * | |
672 | * CONTEXT: | |
673 | * spin_lock_irq(rq->lock) | |
674 | */ | |
675 | void wq_worker_waking_up(struct task_struct *task, unsigned int cpu) | |
676 | { | |
677 | struct worker *worker = kthread_data(task); | |
678 | ||
679 | if (likely(!(worker->flags & WORKER_NOT_RUNNING))) | |
680 | atomic_inc(get_gcwq_nr_running(cpu)); | |
681 | } | |
682 | ||
683 | /** | |
684 | * wq_worker_sleeping - a worker is going to sleep | |
685 | * @task: task going to sleep | |
686 | * @cpu: CPU in question, must be the current CPU number | |
687 | * | |
688 | * This function is called during schedule() when a busy worker is | |
689 | * going to sleep. Worker on the same cpu can be woken up by | |
690 | * returning pointer to its task. | |
691 | * | |
692 | * CONTEXT: | |
693 | * spin_lock_irq(rq->lock) | |
694 | * | |
695 | * RETURNS: | |
696 | * Worker task on @cpu to wake up, %NULL if none. | |
697 | */ | |
698 | struct task_struct *wq_worker_sleeping(struct task_struct *task, | |
699 | unsigned int cpu) | |
700 | { | |
701 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
702 | struct global_cwq *gcwq = get_gcwq(cpu); | |
703 | atomic_t *nr_running = get_gcwq_nr_running(cpu); | |
704 | ||
705 | if (unlikely(worker->flags & WORKER_NOT_RUNNING)) | |
706 | return NULL; | |
707 | ||
708 | /* this can only happen on the local cpu */ | |
709 | BUG_ON(cpu != raw_smp_processor_id()); | |
710 | ||
711 | /* | |
712 | * The counterpart of the following dec_and_test, implied mb, | |
713 | * worklist not empty test sequence is in insert_work(). | |
714 | * Please read comment there. | |
715 | * | |
716 | * NOT_RUNNING is clear. This means that trustee is not in | |
717 | * charge and we're running on the local cpu w/ rq lock held | |
718 | * and preemption disabled, which in turn means that none else | |
719 | * could be manipulating idle_list, so dereferencing idle_list | |
720 | * without gcwq lock is safe. | |
721 | */ | |
722 | if (atomic_dec_and_test(nr_running) && !list_empty(&gcwq->worklist)) | |
723 | to_wakeup = first_worker(gcwq); | |
724 | return to_wakeup ? to_wakeup->task : NULL; | |
725 | } | |
726 | ||
727 | /** | |
728 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 729 | * @worker: self |
d302f017 TH |
730 | * @flags: flags to set |
731 | * @wakeup: wakeup an idle worker if necessary | |
732 | * | |
e22bee78 TH |
733 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
734 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
735 | * woken up. | |
d302f017 | 736 | * |
cb444766 TH |
737 | * CONTEXT: |
738 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
739 | */ |
740 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
741 | bool wakeup) | |
742 | { | |
e22bee78 TH |
743 | struct global_cwq *gcwq = worker->gcwq; |
744 | ||
cb444766 TH |
745 | WARN_ON_ONCE(worker->task != current); |
746 | ||
e22bee78 TH |
747 | /* |
748 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
749 | * wake up an idle worker as necessary if requested by | |
750 | * @wakeup. | |
751 | */ | |
752 | if ((flags & WORKER_NOT_RUNNING) && | |
753 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
754 | atomic_t *nr_running = get_gcwq_nr_running(gcwq->cpu); | |
755 | ||
756 | if (wakeup) { | |
757 | if (atomic_dec_and_test(nr_running) && | |
758 | !list_empty(&gcwq->worklist)) | |
759 | wake_up_worker(gcwq); | |
760 | } else | |
761 | atomic_dec(nr_running); | |
762 | } | |
763 | ||
d302f017 TH |
764 | worker->flags |= flags; |
765 | } | |
766 | ||
767 | /** | |
e22bee78 | 768 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 769 | * @worker: self |
d302f017 TH |
770 | * @flags: flags to clear |
771 | * | |
e22bee78 | 772 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 773 | * |
cb444766 TH |
774 | * CONTEXT: |
775 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
776 | */ |
777 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
778 | { | |
e22bee78 TH |
779 | struct global_cwq *gcwq = worker->gcwq; |
780 | unsigned int oflags = worker->flags; | |
781 | ||
cb444766 TH |
782 | WARN_ON_ONCE(worker->task != current); |
783 | ||
d302f017 | 784 | worker->flags &= ~flags; |
e22bee78 TH |
785 | |
786 | /* if transitioning out of NOT_RUNNING, increment nr_running */ | |
787 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) | |
788 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
789 | atomic_inc(get_gcwq_nr_running(gcwq->cpu)); | |
d302f017 TH |
790 | } |
791 | ||
c8e55f36 TH |
792 | /** |
793 | * busy_worker_head - return the busy hash head for a work | |
794 | * @gcwq: gcwq of interest | |
795 | * @work: work to be hashed | |
796 | * | |
797 | * Return hash head of @gcwq for @work. | |
798 | * | |
799 | * CONTEXT: | |
800 | * spin_lock_irq(gcwq->lock). | |
801 | * | |
802 | * RETURNS: | |
803 | * Pointer to the hash head. | |
804 | */ | |
805 | static struct hlist_head *busy_worker_head(struct global_cwq *gcwq, | |
806 | struct work_struct *work) | |
807 | { | |
808 | const int base_shift = ilog2(sizeof(struct work_struct)); | |
809 | unsigned long v = (unsigned long)work; | |
810 | ||
811 | /* simple shift and fold hash, do we need something better? */ | |
812 | v >>= base_shift; | |
813 | v += v >> BUSY_WORKER_HASH_ORDER; | |
814 | v &= BUSY_WORKER_HASH_MASK; | |
815 | ||
816 | return &gcwq->busy_hash[v]; | |
817 | } | |
818 | ||
8cca0eea TH |
819 | /** |
820 | * __find_worker_executing_work - find worker which is executing a work | |
821 | * @gcwq: gcwq of interest | |
822 | * @bwh: hash head as returned by busy_worker_head() | |
823 | * @work: work to find worker for | |
824 | * | |
825 | * Find a worker which is executing @work on @gcwq. @bwh should be | |
826 | * the hash head obtained by calling busy_worker_head() with the same | |
827 | * work. | |
828 | * | |
829 | * CONTEXT: | |
830 | * spin_lock_irq(gcwq->lock). | |
831 | * | |
832 | * RETURNS: | |
833 | * Pointer to worker which is executing @work if found, NULL | |
834 | * otherwise. | |
835 | */ | |
836 | static struct worker *__find_worker_executing_work(struct global_cwq *gcwq, | |
837 | struct hlist_head *bwh, | |
838 | struct work_struct *work) | |
839 | { | |
840 | struct worker *worker; | |
841 | struct hlist_node *tmp; | |
842 | ||
843 | hlist_for_each_entry(worker, tmp, bwh, hentry) | |
844 | if (worker->current_work == work) | |
845 | return worker; | |
846 | return NULL; | |
847 | } | |
848 | ||
849 | /** | |
850 | * find_worker_executing_work - find worker which is executing a work | |
851 | * @gcwq: gcwq of interest | |
852 | * @work: work to find worker for | |
853 | * | |
854 | * Find a worker which is executing @work on @gcwq. This function is | |
855 | * identical to __find_worker_executing_work() except that this | |
856 | * function calculates @bwh itself. | |
857 | * | |
858 | * CONTEXT: | |
859 | * spin_lock_irq(gcwq->lock). | |
860 | * | |
861 | * RETURNS: | |
862 | * Pointer to worker which is executing @work if found, NULL | |
863 | * otherwise. | |
4d707b9f | 864 | */ |
8cca0eea TH |
865 | static struct worker *find_worker_executing_work(struct global_cwq *gcwq, |
866 | struct work_struct *work) | |
4d707b9f | 867 | { |
8cca0eea TH |
868 | return __find_worker_executing_work(gcwq, busy_worker_head(gcwq, work), |
869 | work); | |
4d707b9f ON |
870 | } |
871 | ||
649027d7 TH |
872 | /** |
873 | * gcwq_determine_ins_pos - find insertion position | |
874 | * @gcwq: gcwq of interest | |
875 | * @cwq: cwq a work is being queued for | |
876 | * | |
877 | * A work for @cwq is about to be queued on @gcwq, determine insertion | |
878 | * position for the work. If @cwq is for HIGHPRI wq, the work is | |
879 | * queued at the head of the queue but in FIFO order with respect to | |
880 | * other HIGHPRI works; otherwise, at the end of the queue. This | |
881 | * function also sets GCWQ_HIGHPRI_PENDING flag to hint @gcwq that | |
882 | * there are HIGHPRI works pending. | |
883 | * | |
884 | * CONTEXT: | |
885 | * spin_lock_irq(gcwq->lock). | |
886 | * | |
887 | * RETURNS: | |
888 | * Pointer to inserstion position. | |
889 | */ | |
890 | static inline struct list_head *gcwq_determine_ins_pos(struct global_cwq *gcwq, | |
891 | struct cpu_workqueue_struct *cwq) | |
365970a1 | 892 | { |
649027d7 TH |
893 | struct work_struct *twork; |
894 | ||
895 | if (likely(!(cwq->wq->flags & WQ_HIGHPRI))) | |
896 | return &gcwq->worklist; | |
897 | ||
898 | list_for_each_entry(twork, &gcwq->worklist, entry) { | |
899 | struct cpu_workqueue_struct *tcwq = get_work_cwq(twork); | |
900 | ||
901 | if (!(tcwq->wq->flags & WQ_HIGHPRI)) | |
902 | break; | |
903 | } | |
904 | ||
905 | gcwq->flags |= GCWQ_HIGHPRI_PENDING; | |
906 | return &twork->entry; | |
365970a1 DH |
907 | } |
908 | ||
4690c4ab | 909 | /** |
7e11629d | 910 | * insert_work - insert a work into gcwq |
4690c4ab TH |
911 | * @cwq: cwq @work belongs to |
912 | * @work: work to insert | |
913 | * @head: insertion point | |
914 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
915 | * | |
7e11629d TH |
916 | * Insert @work which belongs to @cwq into @gcwq after @head. |
917 | * @extra_flags is or'd to work_struct flags. | |
4690c4ab TH |
918 | * |
919 | * CONTEXT: | |
8b03ae3c | 920 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 921 | */ |
b89deed3 | 922 | static void insert_work(struct cpu_workqueue_struct *cwq, |
4690c4ab TH |
923 | struct work_struct *work, struct list_head *head, |
924 | unsigned int extra_flags) | |
b89deed3 | 925 | { |
e22bee78 TH |
926 | struct global_cwq *gcwq = cwq->gcwq; |
927 | ||
4690c4ab | 928 | /* we own @work, set data and link */ |
7a22ad75 | 929 | set_work_cwq(work, cwq, extra_flags); |
e1d8aa9f | 930 | |
6e84d644 ON |
931 | /* |
932 | * Ensure that we get the right work->data if we see the | |
933 | * result of list_add() below, see try_to_grab_pending(). | |
934 | */ | |
935 | smp_wmb(); | |
4690c4ab | 936 | |
1a4d9b0a | 937 | list_add_tail(&work->entry, head); |
e22bee78 TH |
938 | |
939 | /* | |
940 | * Ensure either worker_sched_deactivated() sees the above | |
941 | * list_add_tail() or we see zero nr_running to avoid workers | |
942 | * lying around lazily while there are works to be processed. | |
943 | */ | |
944 | smp_mb(); | |
945 | ||
649027d7 | 946 | if (__need_more_worker(gcwq)) |
e22bee78 | 947 | wake_up_worker(gcwq); |
b89deed3 ON |
948 | } |
949 | ||
4690c4ab | 950 | static void __queue_work(unsigned int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
951 | struct work_struct *work) |
952 | { | |
502ca9d8 TH |
953 | struct global_cwq *gcwq; |
954 | struct cpu_workqueue_struct *cwq; | |
1e19ffc6 | 955 | struct list_head *worklist; |
8a2e8e5d | 956 | unsigned int work_flags; |
1da177e4 LT |
957 | unsigned long flags; |
958 | ||
dc186ad7 | 959 | debug_work_activate(work); |
1e19ffc6 | 960 | |
e41e704b TH |
961 | if (WARN_ON_ONCE(wq->flags & WQ_DYING)) |
962 | return; | |
963 | ||
c7fc77f7 TH |
964 | /* determine gcwq to use */ |
965 | if (!(wq->flags & WQ_UNBOUND)) { | |
18aa9eff TH |
966 | struct global_cwq *last_gcwq; |
967 | ||
c7fc77f7 TH |
968 | if (unlikely(cpu == WORK_CPU_UNBOUND)) |
969 | cpu = raw_smp_processor_id(); | |
970 | ||
18aa9eff TH |
971 | /* |
972 | * It's multi cpu. If @wq is non-reentrant and @work | |
973 | * was previously on a different cpu, it might still | |
974 | * be running there, in which case the work needs to | |
975 | * be queued on that cpu to guarantee non-reentrance. | |
976 | */ | |
502ca9d8 | 977 | gcwq = get_gcwq(cpu); |
18aa9eff TH |
978 | if (wq->flags & WQ_NON_REENTRANT && |
979 | (last_gcwq = get_work_gcwq(work)) && last_gcwq != gcwq) { | |
980 | struct worker *worker; | |
981 | ||
982 | spin_lock_irqsave(&last_gcwq->lock, flags); | |
983 | ||
984 | worker = find_worker_executing_work(last_gcwq, work); | |
985 | ||
986 | if (worker && worker->current_cwq->wq == wq) | |
987 | gcwq = last_gcwq; | |
988 | else { | |
989 | /* meh... not running there, queue here */ | |
990 | spin_unlock_irqrestore(&last_gcwq->lock, flags); | |
991 | spin_lock_irqsave(&gcwq->lock, flags); | |
992 | } | |
993 | } else | |
994 | spin_lock_irqsave(&gcwq->lock, flags); | |
f3421797 TH |
995 | } else { |
996 | gcwq = get_gcwq(WORK_CPU_UNBOUND); | |
997 | spin_lock_irqsave(&gcwq->lock, flags); | |
502ca9d8 TH |
998 | } |
999 | ||
1000 | /* gcwq determined, get cwq and queue */ | |
1001 | cwq = get_cwq(gcwq->cpu, wq); | |
cdadf009 | 1002 | trace_workqueue_queue_work(cpu, cwq, work); |
502ca9d8 | 1003 | |
4690c4ab | 1004 | BUG_ON(!list_empty(&work->entry)); |
1e19ffc6 | 1005 | |
73f53c4a | 1006 | cwq->nr_in_flight[cwq->work_color]++; |
8a2e8e5d | 1007 | work_flags = work_color_to_flags(cwq->work_color); |
1e19ffc6 TH |
1008 | |
1009 | if (likely(cwq->nr_active < cwq->max_active)) { | |
cdadf009 | 1010 | trace_workqueue_activate_work(work); |
1e19ffc6 | 1011 | cwq->nr_active++; |
649027d7 | 1012 | worklist = gcwq_determine_ins_pos(gcwq, cwq); |
8a2e8e5d TH |
1013 | } else { |
1014 | work_flags |= WORK_STRUCT_DELAYED; | |
1e19ffc6 | 1015 | worklist = &cwq->delayed_works; |
8a2e8e5d | 1016 | } |
1e19ffc6 | 1017 | |
8a2e8e5d | 1018 | insert_work(cwq, work, worklist, work_flags); |
1e19ffc6 | 1019 | |
8b03ae3c | 1020 | spin_unlock_irqrestore(&gcwq->lock, flags); |
1da177e4 LT |
1021 | } |
1022 | ||
0fcb78c2 REB |
1023 | /** |
1024 | * queue_work - queue work on a workqueue | |
1025 | * @wq: workqueue to use | |
1026 | * @work: work to queue | |
1027 | * | |
057647fc | 1028 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
1da177e4 | 1029 | * |
00dfcaf7 ON |
1030 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
1031 | * it can be processed by another CPU. | |
1da177e4 | 1032 | */ |
7ad5b3a5 | 1033 | int queue_work(struct workqueue_struct *wq, struct work_struct *work) |
1da177e4 | 1034 | { |
ef1ca236 ON |
1035 | int ret; |
1036 | ||
1037 | ret = queue_work_on(get_cpu(), wq, work); | |
1038 | put_cpu(); | |
1039 | ||
1da177e4 LT |
1040 | return ret; |
1041 | } | |
ae90dd5d | 1042 | EXPORT_SYMBOL_GPL(queue_work); |
1da177e4 | 1043 | |
c1a220e7 ZR |
1044 | /** |
1045 | * queue_work_on - queue work on specific cpu | |
1046 | * @cpu: CPU number to execute work on | |
1047 | * @wq: workqueue to use | |
1048 | * @work: work to queue | |
1049 | * | |
1050 | * Returns 0 if @work was already on a queue, non-zero otherwise. | |
1051 | * | |
1052 | * We queue the work to a specific CPU, the caller must ensure it | |
1053 | * can't go away. | |
1054 | */ | |
1055 | int | |
1056 | queue_work_on(int cpu, struct workqueue_struct *wq, struct work_struct *work) | |
1057 | { | |
1058 | int ret = 0; | |
1059 | ||
22df02bb | 1060 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1061 | __queue_work(cpu, wq, work); |
c1a220e7 ZR |
1062 | ret = 1; |
1063 | } | |
1064 | return ret; | |
1065 | } | |
1066 | EXPORT_SYMBOL_GPL(queue_work_on); | |
1067 | ||
6d141c3f | 1068 | static void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1069 | { |
52bad64d | 1070 | struct delayed_work *dwork = (struct delayed_work *)__data; |
7a22ad75 | 1071 | struct cpu_workqueue_struct *cwq = get_work_cwq(&dwork->work); |
1da177e4 | 1072 | |
4690c4ab | 1073 | __queue_work(smp_processor_id(), cwq->wq, &dwork->work); |
1da177e4 LT |
1074 | } |
1075 | ||
0fcb78c2 REB |
1076 | /** |
1077 | * queue_delayed_work - queue work on a workqueue after delay | |
1078 | * @wq: workqueue to use | |
af9997e4 | 1079 | * @dwork: delayable work to queue |
0fcb78c2 REB |
1080 | * @delay: number of jiffies to wait before queueing |
1081 | * | |
057647fc | 1082 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 1083 | */ |
7ad5b3a5 | 1084 | int queue_delayed_work(struct workqueue_struct *wq, |
52bad64d | 1085 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 1086 | { |
52bad64d | 1087 | if (delay == 0) |
63bc0362 | 1088 | return queue_work(wq, &dwork->work); |
1da177e4 | 1089 | |
63bc0362 | 1090 | return queue_delayed_work_on(-1, wq, dwork, delay); |
1da177e4 | 1091 | } |
ae90dd5d | 1092 | EXPORT_SYMBOL_GPL(queue_delayed_work); |
1da177e4 | 1093 | |
0fcb78c2 REB |
1094 | /** |
1095 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1096 | * @cpu: CPU number to execute work on | |
1097 | * @wq: workqueue to use | |
af9997e4 | 1098 | * @dwork: work to queue |
0fcb78c2 REB |
1099 | * @delay: number of jiffies to wait before queueing |
1100 | * | |
057647fc | 1101 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 1102 | */ |
7a6bc1cd | 1103 | int queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
52bad64d | 1104 | struct delayed_work *dwork, unsigned long delay) |
7a6bc1cd VP |
1105 | { |
1106 | int ret = 0; | |
52bad64d DH |
1107 | struct timer_list *timer = &dwork->timer; |
1108 | struct work_struct *work = &dwork->work; | |
7a6bc1cd | 1109 | |
22df02bb | 1110 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
c7fc77f7 | 1111 | unsigned int lcpu; |
7a22ad75 | 1112 | |
7a6bc1cd VP |
1113 | BUG_ON(timer_pending(timer)); |
1114 | BUG_ON(!list_empty(&work->entry)); | |
1115 | ||
8a3e77cc AL |
1116 | timer_stats_timer_set_start_info(&dwork->timer); |
1117 | ||
7a22ad75 TH |
1118 | /* |
1119 | * This stores cwq for the moment, for the timer_fn. | |
1120 | * Note that the work's gcwq is preserved to allow | |
1121 | * reentrance detection for delayed works. | |
1122 | */ | |
c7fc77f7 TH |
1123 | if (!(wq->flags & WQ_UNBOUND)) { |
1124 | struct global_cwq *gcwq = get_work_gcwq(work); | |
1125 | ||
1126 | if (gcwq && gcwq->cpu != WORK_CPU_UNBOUND) | |
1127 | lcpu = gcwq->cpu; | |
1128 | else | |
1129 | lcpu = raw_smp_processor_id(); | |
1130 | } else | |
1131 | lcpu = WORK_CPU_UNBOUND; | |
1132 | ||
7a22ad75 | 1133 | set_work_cwq(work, get_cwq(lcpu, wq), 0); |
c7fc77f7 | 1134 | |
7a6bc1cd | 1135 | timer->expires = jiffies + delay; |
52bad64d | 1136 | timer->data = (unsigned long)dwork; |
7a6bc1cd | 1137 | timer->function = delayed_work_timer_fn; |
63bc0362 ON |
1138 | |
1139 | if (unlikely(cpu >= 0)) | |
1140 | add_timer_on(timer, cpu); | |
1141 | else | |
1142 | add_timer(timer); | |
7a6bc1cd VP |
1143 | ret = 1; |
1144 | } | |
1145 | return ret; | |
1146 | } | |
ae90dd5d | 1147 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
1da177e4 | 1148 | |
c8e55f36 TH |
1149 | /** |
1150 | * worker_enter_idle - enter idle state | |
1151 | * @worker: worker which is entering idle state | |
1152 | * | |
1153 | * @worker is entering idle state. Update stats and idle timer if | |
1154 | * necessary. | |
1155 | * | |
1156 | * LOCKING: | |
1157 | * spin_lock_irq(gcwq->lock). | |
1158 | */ | |
1159 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1160 | { |
c8e55f36 TH |
1161 | struct global_cwq *gcwq = worker->gcwq; |
1162 | ||
1163 | BUG_ON(worker->flags & WORKER_IDLE); | |
1164 | BUG_ON(!list_empty(&worker->entry) && | |
1165 | (worker->hentry.next || worker->hentry.pprev)); | |
1166 | ||
cb444766 TH |
1167 | /* can't use worker_set_flags(), also called from start_worker() */ |
1168 | worker->flags |= WORKER_IDLE; | |
c8e55f36 | 1169 | gcwq->nr_idle++; |
e22bee78 | 1170 | worker->last_active = jiffies; |
c8e55f36 TH |
1171 | |
1172 | /* idle_list is LIFO */ | |
1173 | list_add(&worker->entry, &gcwq->idle_list); | |
db7bccf4 | 1174 | |
e22bee78 TH |
1175 | if (likely(!(worker->flags & WORKER_ROGUE))) { |
1176 | if (too_many_workers(gcwq) && !timer_pending(&gcwq->idle_timer)) | |
1177 | mod_timer(&gcwq->idle_timer, | |
1178 | jiffies + IDLE_WORKER_TIMEOUT); | |
1179 | } else | |
db7bccf4 | 1180 | wake_up_all(&gcwq->trustee_wait); |
cb444766 TH |
1181 | |
1182 | /* sanity check nr_running */ | |
1183 | WARN_ON_ONCE(gcwq->nr_workers == gcwq->nr_idle && | |
1184 | atomic_read(get_gcwq_nr_running(gcwq->cpu))); | |
c8e55f36 TH |
1185 | } |
1186 | ||
1187 | /** | |
1188 | * worker_leave_idle - leave idle state | |
1189 | * @worker: worker which is leaving idle state | |
1190 | * | |
1191 | * @worker is leaving idle state. Update stats. | |
1192 | * | |
1193 | * LOCKING: | |
1194 | * spin_lock_irq(gcwq->lock). | |
1195 | */ | |
1196 | static void worker_leave_idle(struct worker *worker) | |
1197 | { | |
1198 | struct global_cwq *gcwq = worker->gcwq; | |
1199 | ||
1200 | BUG_ON(!(worker->flags & WORKER_IDLE)); | |
d302f017 | 1201 | worker_clr_flags(worker, WORKER_IDLE); |
c8e55f36 TH |
1202 | gcwq->nr_idle--; |
1203 | list_del_init(&worker->entry); | |
1204 | } | |
1205 | ||
e22bee78 TH |
1206 | /** |
1207 | * worker_maybe_bind_and_lock - bind worker to its cpu if possible and lock gcwq | |
1208 | * @worker: self | |
1209 | * | |
1210 | * Works which are scheduled while the cpu is online must at least be | |
1211 | * scheduled to a worker which is bound to the cpu so that if they are | |
1212 | * flushed from cpu callbacks while cpu is going down, they are | |
1213 | * guaranteed to execute on the cpu. | |
1214 | * | |
1215 | * This function is to be used by rogue workers and rescuers to bind | |
1216 | * themselves to the target cpu and may race with cpu going down or | |
1217 | * coming online. kthread_bind() can't be used because it may put the | |
1218 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
1219 | * verbatim as it's best effort and blocking and gcwq may be | |
1220 | * [dis]associated in the meantime. | |
1221 | * | |
1222 | * This function tries set_cpus_allowed() and locks gcwq and verifies | |
1223 | * the binding against GCWQ_DISASSOCIATED which is set during | |
1224 | * CPU_DYING and cleared during CPU_ONLINE, so if the worker enters | |
1225 | * idle state or fetches works without dropping lock, it can guarantee | |
1226 | * the scheduling requirement described in the first paragraph. | |
1227 | * | |
1228 | * CONTEXT: | |
1229 | * Might sleep. Called without any lock but returns with gcwq->lock | |
1230 | * held. | |
1231 | * | |
1232 | * RETURNS: | |
1233 | * %true if the associated gcwq is online (@worker is successfully | |
1234 | * bound), %false if offline. | |
1235 | */ | |
1236 | static bool worker_maybe_bind_and_lock(struct worker *worker) | |
972fa1c5 | 1237 | __acquires(&gcwq->lock) |
e22bee78 TH |
1238 | { |
1239 | struct global_cwq *gcwq = worker->gcwq; | |
1240 | struct task_struct *task = worker->task; | |
1241 | ||
1242 | while (true) { | |
4e6045f1 | 1243 | /* |
e22bee78 TH |
1244 | * The following call may fail, succeed or succeed |
1245 | * without actually migrating the task to the cpu if | |
1246 | * it races with cpu hotunplug operation. Verify | |
1247 | * against GCWQ_DISASSOCIATED. | |
4e6045f1 | 1248 | */ |
f3421797 TH |
1249 | if (!(gcwq->flags & GCWQ_DISASSOCIATED)) |
1250 | set_cpus_allowed_ptr(task, get_cpu_mask(gcwq->cpu)); | |
e22bee78 TH |
1251 | |
1252 | spin_lock_irq(&gcwq->lock); | |
1253 | if (gcwq->flags & GCWQ_DISASSOCIATED) | |
1254 | return false; | |
1255 | if (task_cpu(task) == gcwq->cpu && | |
1256 | cpumask_equal(¤t->cpus_allowed, | |
1257 | get_cpu_mask(gcwq->cpu))) | |
1258 | return true; | |
1259 | spin_unlock_irq(&gcwq->lock); | |
1260 | ||
1261 | /* CPU has come up inbetween, retry migration */ | |
1262 | cpu_relax(); | |
1263 | } | |
1264 | } | |
1265 | ||
1266 | /* | |
1267 | * Function for worker->rebind_work used to rebind rogue busy workers | |
1268 | * to the associated cpu which is coming back online. This is | |
1269 | * scheduled by cpu up but can race with other cpu hotplug operations | |
1270 | * and may be executed twice without intervening cpu down. | |
1271 | */ | |
1272 | static void worker_rebind_fn(struct work_struct *work) | |
1273 | { | |
1274 | struct worker *worker = container_of(work, struct worker, rebind_work); | |
1275 | struct global_cwq *gcwq = worker->gcwq; | |
1276 | ||
1277 | if (worker_maybe_bind_and_lock(worker)) | |
1278 | worker_clr_flags(worker, WORKER_REBIND); | |
1279 | ||
1280 | spin_unlock_irq(&gcwq->lock); | |
1281 | } | |
1282 | ||
c34056a3 TH |
1283 | static struct worker *alloc_worker(void) |
1284 | { | |
1285 | struct worker *worker; | |
1286 | ||
1287 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1288 | if (worker) { |
1289 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1290 | INIT_LIST_HEAD(&worker->scheduled); |
e22bee78 TH |
1291 | INIT_WORK(&worker->rebind_work, worker_rebind_fn); |
1292 | /* on creation a worker is in !idle && prep state */ | |
1293 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1294 | } |
c34056a3 TH |
1295 | return worker; |
1296 | } | |
1297 | ||
1298 | /** | |
1299 | * create_worker - create a new workqueue worker | |
7e11629d | 1300 | * @gcwq: gcwq the new worker will belong to |
c34056a3 TH |
1301 | * @bind: whether to set affinity to @cpu or not |
1302 | * | |
7e11629d | 1303 | * Create a new worker which is bound to @gcwq. The returned worker |
c34056a3 TH |
1304 | * can be started by calling start_worker() or destroyed using |
1305 | * destroy_worker(). | |
1306 | * | |
1307 | * CONTEXT: | |
1308 | * Might sleep. Does GFP_KERNEL allocations. | |
1309 | * | |
1310 | * RETURNS: | |
1311 | * Pointer to the newly created worker. | |
1312 | */ | |
7e11629d | 1313 | static struct worker *create_worker(struct global_cwq *gcwq, bool bind) |
c34056a3 | 1314 | { |
f3421797 | 1315 | bool on_unbound_cpu = gcwq->cpu == WORK_CPU_UNBOUND; |
c34056a3 | 1316 | struct worker *worker = NULL; |
f3421797 | 1317 | int id = -1; |
c34056a3 | 1318 | |
8b03ae3c TH |
1319 | spin_lock_irq(&gcwq->lock); |
1320 | while (ida_get_new(&gcwq->worker_ida, &id)) { | |
1321 | spin_unlock_irq(&gcwq->lock); | |
1322 | if (!ida_pre_get(&gcwq->worker_ida, GFP_KERNEL)) | |
c34056a3 | 1323 | goto fail; |
8b03ae3c | 1324 | spin_lock_irq(&gcwq->lock); |
c34056a3 | 1325 | } |
8b03ae3c | 1326 | spin_unlock_irq(&gcwq->lock); |
c34056a3 TH |
1327 | |
1328 | worker = alloc_worker(); | |
1329 | if (!worker) | |
1330 | goto fail; | |
1331 | ||
8b03ae3c | 1332 | worker->gcwq = gcwq; |
c34056a3 TH |
1333 | worker->id = id; |
1334 | ||
f3421797 TH |
1335 | if (!on_unbound_cpu) |
1336 | worker->task = kthread_create(worker_thread, worker, | |
1337 | "kworker/%u:%d", gcwq->cpu, id); | |
1338 | else | |
1339 | worker->task = kthread_create(worker_thread, worker, | |
1340 | "kworker/u:%d", id); | |
c34056a3 TH |
1341 | if (IS_ERR(worker->task)) |
1342 | goto fail; | |
1343 | ||
db7bccf4 TH |
1344 | /* |
1345 | * A rogue worker will become a regular one if CPU comes | |
1346 | * online later on. Make sure every worker has | |
1347 | * PF_THREAD_BOUND set. | |
1348 | */ | |
f3421797 | 1349 | if (bind && !on_unbound_cpu) |
8b03ae3c | 1350 | kthread_bind(worker->task, gcwq->cpu); |
f3421797 | 1351 | else { |
db7bccf4 | 1352 | worker->task->flags |= PF_THREAD_BOUND; |
f3421797 TH |
1353 | if (on_unbound_cpu) |
1354 | worker->flags |= WORKER_UNBOUND; | |
1355 | } | |
c34056a3 TH |
1356 | |
1357 | return worker; | |
1358 | fail: | |
1359 | if (id >= 0) { | |
8b03ae3c TH |
1360 | spin_lock_irq(&gcwq->lock); |
1361 | ida_remove(&gcwq->worker_ida, id); | |
1362 | spin_unlock_irq(&gcwq->lock); | |
c34056a3 TH |
1363 | } |
1364 | kfree(worker); | |
1365 | return NULL; | |
1366 | } | |
1367 | ||
1368 | /** | |
1369 | * start_worker - start a newly created worker | |
1370 | * @worker: worker to start | |
1371 | * | |
c8e55f36 | 1372 | * Make the gcwq aware of @worker and start it. |
c34056a3 TH |
1373 | * |
1374 | * CONTEXT: | |
8b03ae3c | 1375 | * spin_lock_irq(gcwq->lock). |
c34056a3 TH |
1376 | */ |
1377 | static void start_worker(struct worker *worker) | |
1378 | { | |
cb444766 | 1379 | worker->flags |= WORKER_STARTED; |
c8e55f36 TH |
1380 | worker->gcwq->nr_workers++; |
1381 | worker_enter_idle(worker); | |
c34056a3 TH |
1382 | wake_up_process(worker->task); |
1383 | } | |
1384 | ||
1385 | /** | |
1386 | * destroy_worker - destroy a workqueue worker | |
1387 | * @worker: worker to be destroyed | |
1388 | * | |
c8e55f36 TH |
1389 | * Destroy @worker and adjust @gcwq stats accordingly. |
1390 | * | |
1391 | * CONTEXT: | |
1392 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. | |
c34056a3 TH |
1393 | */ |
1394 | static void destroy_worker(struct worker *worker) | |
1395 | { | |
8b03ae3c | 1396 | struct global_cwq *gcwq = worker->gcwq; |
c34056a3 TH |
1397 | int id = worker->id; |
1398 | ||
1399 | /* sanity check frenzy */ | |
1400 | BUG_ON(worker->current_work); | |
affee4b2 | 1401 | BUG_ON(!list_empty(&worker->scheduled)); |
c34056a3 | 1402 | |
c8e55f36 TH |
1403 | if (worker->flags & WORKER_STARTED) |
1404 | gcwq->nr_workers--; | |
1405 | if (worker->flags & WORKER_IDLE) | |
1406 | gcwq->nr_idle--; | |
1407 | ||
1408 | list_del_init(&worker->entry); | |
cb444766 | 1409 | worker->flags |= WORKER_DIE; |
c8e55f36 TH |
1410 | |
1411 | spin_unlock_irq(&gcwq->lock); | |
1412 | ||
c34056a3 TH |
1413 | kthread_stop(worker->task); |
1414 | kfree(worker); | |
1415 | ||
8b03ae3c TH |
1416 | spin_lock_irq(&gcwq->lock); |
1417 | ida_remove(&gcwq->worker_ida, id); | |
c34056a3 TH |
1418 | } |
1419 | ||
e22bee78 TH |
1420 | static void idle_worker_timeout(unsigned long __gcwq) |
1421 | { | |
1422 | struct global_cwq *gcwq = (void *)__gcwq; | |
1423 | ||
1424 | spin_lock_irq(&gcwq->lock); | |
1425 | ||
1426 | if (too_many_workers(gcwq)) { | |
1427 | struct worker *worker; | |
1428 | unsigned long expires; | |
1429 | ||
1430 | /* idle_list is kept in LIFO order, check the last one */ | |
1431 | worker = list_entry(gcwq->idle_list.prev, struct worker, entry); | |
1432 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; | |
1433 | ||
1434 | if (time_before(jiffies, expires)) | |
1435 | mod_timer(&gcwq->idle_timer, expires); | |
1436 | else { | |
1437 | /* it's been idle for too long, wake up manager */ | |
1438 | gcwq->flags |= GCWQ_MANAGE_WORKERS; | |
1439 | wake_up_worker(gcwq); | |
d5abe669 | 1440 | } |
e22bee78 TH |
1441 | } |
1442 | ||
1443 | spin_unlock_irq(&gcwq->lock); | |
1444 | } | |
d5abe669 | 1445 | |
e22bee78 TH |
1446 | static bool send_mayday(struct work_struct *work) |
1447 | { | |
1448 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); | |
1449 | struct workqueue_struct *wq = cwq->wq; | |
f3421797 | 1450 | unsigned int cpu; |
e22bee78 TH |
1451 | |
1452 | if (!(wq->flags & WQ_RESCUER)) | |
1453 | return false; | |
1454 | ||
1455 | /* mayday mayday mayday */ | |
f3421797 TH |
1456 | cpu = cwq->gcwq->cpu; |
1457 | /* WORK_CPU_UNBOUND can't be set in cpumask, use cpu 0 instead */ | |
1458 | if (cpu == WORK_CPU_UNBOUND) | |
1459 | cpu = 0; | |
f2e005aa | 1460 | if (!mayday_test_and_set_cpu(cpu, wq->mayday_mask)) |
e22bee78 TH |
1461 | wake_up_process(wq->rescuer->task); |
1462 | return true; | |
1463 | } | |
1464 | ||
1465 | static void gcwq_mayday_timeout(unsigned long __gcwq) | |
1466 | { | |
1467 | struct global_cwq *gcwq = (void *)__gcwq; | |
1468 | struct work_struct *work; | |
1469 | ||
1470 | spin_lock_irq(&gcwq->lock); | |
1471 | ||
1472 | if (need_to_create_worker(gcwq)) { | |
1473 | /* | |
1474 | * We've been trying to create a new worker but | |
1475 | * haven't been successful. We might be hitting an | |
1476 | * allocation deadlock. Send distress signals to | |
1477 | * rescuers. | |
1478 | */ | |
1479 | list_for_each_entry(work, &gcwq->worklist, entry) | |
1480 | send_mayday(work); | |
1da177e4 | 1481 | } |
e22bee78 TH |
1482 | |
1483 | spin_unlock_irq(&gcwq->lock); | |
1484 | ||
1485 | mod_timer(&gcwq->mayday_timer, jiffies + MAYDAY_INTERVAL); | |
1da177e4 LT |
1486 | } |
1487 | ||
e22bee78 TH |
1488 | /** |
1489 | * maybe_create_worker - create a new worker if necessary | |
1490 | * @gcwq: gcwq to create a new worker for | |
1491 | * | |
1492 | * Create a new worker for @gcwq if necessary. @gcwq is guaranteed to | |
1493 | * have at least one idle worker on return from this function. If | |
1494 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
1495 | * sent to all rescuers with works scheduled on @gcwq to resolve | |
1496 | * possible allocation deadlock. | |
1497 | * | |
1498 | * On return, need_to_create_worker() is guaranteed to be false and | |
1499 | * may_start_working() true. | |
1500 | * | |
1501 | * LOCKING: | |
1502 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1503 | * multiple times. Does GFP_KERNEL allocations. Called only from | |
1504 | * manager. | |
1505 | * | |
1506 | * RETURNS: | |
1507 | * false if no action was taken and gcwq->lock stayed locked, true | |
1508 | * otherwise. | |
1509 | */ | |
1510 | static bool maybe_create_worker(struct global_cwq *gcwq) | |
06bd6ebf NK |
1511 | __releases(&gcwq->lock) |
1512 | __acquires(&gcwq->lock) | |
1da177e4 | 1513 | { |
e22bee78 TH |
1514 | if (!need_to_create_worker(gcwq)) |
1515 | return false; | |
1516 | restart: | |
9f9c2364 TH |
1517 | spin_unlock_irq(&gcwq->lock); |
1518 | ||
e22bee78 TH |
1519 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
1520 | mod_timer(&gcwq->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); | |
1521 | ||
1522 | while (true) { | |
1523 | struct worker *worker; | |
1524 | ||
e22bee78 TH |
1525 | worker = create_worker(gcwq, true); |
1526 | if (worker) { | |
1527 | del_timer_sync(&gcwq->mayday_timer); | |
1528 | spin_lock_irq(&gcwq->lock); | |
1529 | start_worker(worker); | |
1530 | BUG_ON(need_to_create_worker(gcwq)); | |
1531 | return true; | |
1532 | } | |
1533 | ||
1534 | if (!need_to_create_worker(gcwq)) | |
1535 | break; | |
1da177e4 | 1536 | |
e22bee78 TH |
1537 | __set_current_state(TASK_INTERRUPTIBLE); |
1538 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1539 | |
e22bee78 TH |
1540 | if (!need_to_create_worker(gcwq)) |
1541 | break; | |
1542 | } | |
1543 | ||
e22bee78 TH |
1544 | del_timer_sync(&gcwq->mayday_timer); |
1545 | spin_lock_irq(&gcwq->lock); | |
1546 | if (need_to_create_worker(gcwq)) | |
1547 | goto restart; | |
1548 | return true; | |
1549 | } | |
1550 | ||
1551 | /** | |
1552 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
1553 | * @gcwq: gcwq to destroy workers for | |
1554 | * | |
1555 | * Destroy @gcwq workers which have been idle for longer than | |
1556 | * IDLE_WORKER_TIMEOUT. | |
1557 | * | |
1558 | * LOCKING: | |
1559 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1560 | * multiple times. Called only from manager. | |
1561 | * | |
1562 | * RETURNS: | |
1563 | * false if no action was taken and gcwq->lock stayed locked, true | |
1564 | * otherwise. | |
1565 | */ | |
1566 | static bool maybe_destroy_workers(struct global_cwq *gcwq) | |
1567 | { | |
1568 | bool ret = false; | |
1da177e4 | 1569 | |
e22bee78 TH |
1570 | while (too_many_workers(gcwq)) { |
1571 | struct worker *worker; | |
1572 | unsigned long expires; | |
3af24433 | 1573 | |
e22bee78 TH |
1574 | worker = list_entry(gcwq->idle_list.prev, struct worker, entry); |
1575 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; | |
85f4186a | 1576 | |
e22bee78 TH |
1577 | if (time_before(jiffies, expires)) { |
1578 | mod_timer(&gcwq->idle_timer, expires); | |
3af24433 | 1579 | break; |
e22bee78 | 1580 | } |
1da177e4 | 1581 | |
e22bee78 TH |
1582 | destroy_worker(worker); |
1583 | ret = true; | |
1da177e4 | 1584 | } |
3af24433 | 1585 | |
e22bee78 TH |
1586 | return ret; |
1587 | } | |
1588 | ||
1589 | /** | |
1590 | * manage_workers - manage worker pool | |
1591 | * @worker: self | |
1592 | * | |
1593 | * Assume the manager role and manage gcwq worker pool @worker belongs | |
1594 | * to. At any given time, there can be only zero or one manager per | |
1595 | * gcwq. The exclusion is handled automatically by this function. | |
1596 | * | |
1597 | * The caller can safely start processing works on false return. On | |
1598 | * true return, it's guaranteed that need_to_create_worker() is false | |
1599 | * and may_start_working() is true. | |
1600 | * | |
1601 | * CONTEXT: | |
1602 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1603 | * multiple times. Does GFP_KERNEL allocations. | |
1604 | * | |
1605 | * RETURNS: | |
1606 | * false if no action was taken and gcwq->lock stayed locked, true if | |
1607 | * some action was taken. | |
1608 | */ | |
1609 | static bool manage_workers(struct worker *worker) | |
1610 | { | |
1611 | struct global_cwq *gcwq = worker->gcwq; | |
1612 | bool ret = false; | |
1613 | ||
1614 | if (gcwq->flags & GCWQ_MANAGING_WORKERS) | |
1615 | return ret; | |
1616 | ||
1617 | gcwq->flags &= ~GCWQ_MANAGE_WORKERS; | |
1618 | gcwq->flags |= GCWQ_MANAGING_WORKERS; | |
1619 | ||
1620 | /* | |
1621 | * Destroy and then create so that may_start_working() is true | |
1622 | * on return. | |
1623 | */ | |
1624 | ret |= maybe_destroy_workers(gcwq); | |
1625 | ret |= maybe_create_worker(gcwq); | |
1626 | ||
1627 | gcwq->flags &= ~GCWQ_MANAGING_WORKERS; | |
1628 | ||
1629 | /* | |
1630 | * The trustee might be waiting to take over the manager | |
1631 | * position, tell it we're done. | |
1632 | */ | |
1633 | if (unlikely(gcwq->trustee)) | |
1634 | wake_up_all(&gcwq->trustee_wait); | |
1635 | ||
1636 | return ret; | |
1637 | } | |
1638 | ||
affee4b2 TH |
1639 | /** |
1640 | * move_linked_works - move linked works to a list | |
1641 | * @work: start of series of works to be scheduled | |
1642 | * @head: target list to append @work to | |
1643 | * @nextp: out paramter for nested worklist walking | |
1644 | * | |
1645 | * Schedule linked works starting from @work to @head. Work series to | |
1646 | * be scheduled starts at @work and includes any consecutive work with | |
1647 | * WORK_STRUCT_LINKED set in its predecessor. | |
1648 | * | |
1649 | * If @nextp is not NULL, it's updated to point to the next work of | |
1650 | * the last scheduled work. This allows move_linked_works() to be | |
1651 | * nested inside outer list_for_each_entry_safe(). | |
1652 | * | |
1653 | * CONTEXT: | |
8b03ae3c | 1654 | * spin_lock_irq(gcwq->lock). |
affee4b2 TH |
1655 | */ |
1656 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
1657 | struct work_struct **nextp) | |
1658 | { | |
1659 | struct work_struct *n; | |
1660 | ||
1661 | /* | |
1662 | * Linked worklist will always end before the end of the list, | |
1663 | * use NULL for list head. | |
1664 | */ | |
1665 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
1666 | list_move_tail(&work->entry, head); | |
1667 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
1668 | break; | |
1669 | } | |
1670 | ||
1671 | /* | |
1672 | * If we're already inside safe list traversal and have moved | |
1673 | * multiple works to the scheduled queue, the next position | |
1674 | * needs to be updated. | |
1675 | */ | |
1676 | if (nextp) | |
1677 | *nextp = n; | |
1678 | } | |
1679 | ||
1e19ffc6 TH |
1680 | static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq) |
1681 | { | |
1682 | struct work_struct *work = list_first_entry(&cwq->delayed_works, | |
1683 | struct work_struct, entry); | |
649027d7 | 1684 | struct list_head *pos = gcwq_determine_ins_pos(cwq->gcwq, cwq); |
1e19ffc6 | 1685 | |
cdadf009 | 1686 | trace_workqueue_activate_work(work); |
649027d7 | 1687 | move_linked_works(work, pos, NULL); |
8a2e8e5d | 1688 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); |
1e19ffc6 TH |
1689 | cwq->nr_active++; |
1690 | } | |
1691 | ||
73f53c4a TH |
1692 | /** |
1693 | * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight | |
1694 | * @cwq: cwq of interest | |
1695 | * @color: color of work which left the queue | |
8a2e8e5d | 1696 | * @delayed: for a delayed work |
73f53c4a TH |
1697 | * |
1698 | * A work either has completed or is removed from pending queue, | |
1699 | * decrement nr_in_flight of its cwq and handle workqueue flushing. | |
1700 | * | |
1701 | * CONTEXT: | |
8b03ae3c | 1702 | * spin_lock_irq(gcwq->lock). |
73f53c4a | 1703 | */ |
8a2e8e5d TH |
1704 | static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color, |
1705 | bool delayed) | |
73f53c4a TH |
1706 | { |
1707 | /* ignore uncolored works */ | |
1708 | if (color == WORK_NO_COLOR) | |
1709 | return; | |
1710 | ||
1711 | cwq->nr_in_flight[color]--; | |
1e19ffc6 | 1712 | |
8a2e8e5d TH |
1713 | if (!delayed) { |
1714 | cwq->nr_active--; | |
1715 | if (!list_empty(&cwq->delayed_works)) { | |
1716 | /* one down, submit a delayed one */ | |
1717 | if (cwq->nr_active < cwq->max_active) | |
1718 | cwq_activate_first_delayed(cwq); | |
1719 | } | |
502ca9d8 | 1720 | } |
73f53c4a TH |
1721 | |
1722 | /* is flush in progress and are we at the flushing tip? */ | |
1723 | if (likely(cwq->flush_color != color)) | |
1724 | return; | |
1725 | ||
1726 | /* are there still in-flight works? */ | |
1727 | if (cwq->nr_in_flight[color]) | |
1728 | return; | |
1729 | ||
1730 | /* this cwq is done, clear flush_color */ | |
1731 | cwq->flush_color = -1; | |
1732 | ||
1733 | /* | |
1734 | * If this was the last cwq, wake up the first flusher. It | |
1735 | * will handle the rest. | |
1736 | */ | |
1737 | if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush)) | |
1738 | complete(&cwq->wq->first_flusher->done); | |
1739 | } | |
1740 | ||
a62428c0 TH |
1741 | /** |
1742 | * process_one_work - process single work | |
c34056a3 | 1743 | * @worker: self |
a62428c0 TH |
1744 | * @work: work to process |
1745 | * | |
1746 | * Process @work. This function contains all the logics necessary to | |
1747 | * process a single work including synchronization against and | |
1748 | * interaction with other workers on the same cpu, queueing and | |
1749 | * flushing. As long as context requirement is met, any worker can | |
1750 | * call this function to process a work. | |
1751 | * | |
1752 | * CONTEXT: | |
8b03ae3c | 1753 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. |
a62428c0 | 1754 | */ |
c34056a3 | 1755 | static void process_one_work(struct worker *worker, struct work_struct *work) |
06bd6ebf NK |
1756 | __releases(&gcwq->lock) |
1757 | __acquires(&gcwq->lock) | |
a62428c0 | 1758 | { |
7e11629d | 1759 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); |
8b03ae3c | 1760 | struct global_cwq *gcwq = cwq->gcwq; |
c8e55f36 | 1761 | struct hlist_head *bwh = busy_worker_head(gcwq, work); |
fb0e7beb | 1762 | bool cpu_intensive = cwq->wq->flags & WQ_CPU_INTENSIVE; |
a62428c0 | 1763 | work_func_t f = work->func; |
73f53c4a | 1764 | int work_color; |
7e11629d | 1765 | struct worker *collision; |
a62428c0 TH |
1766 | #ifdef CONFIG_LOCKDEP |
1767 | /* | |
1768 | * It is permissible to free the struct work_struct from | |
1769 | * inside the function that is called from it, this we need to | |
1770 | * take into account for lockdep too. To avoid bogus "held | |
1771 | * lock freed" warnings as well as problems when looking into | |
1772 | * work->lockdep_map, make a copy and use that here. | |
1773 | */ | |
1774 | struct lockdep_map lockdep_map = work->lockdep_map; | |
1775 | #endif | |
7e11629d TH |
1776 | /* |
1777 | * A single work shouldn't be executed concurrently by | |
1778 | * multiple workers on a single cpu. Check whether anyone is | |
1779 | * already processing the work. If so, defer the work to the | |
1780 | * currently executing one. | |
1781 | */ | |
1782 | collision = __find_worker_executing_work(gcwq, bwh, work); | |
1783 | if (unlikely(collision)) { | |
1784 | move_linked_works(work, &collision->scheduled, NULL); | |
1785 | return; | |
1786 | } | |
1787 | ||
a62428c0 | 1788 | /* claim and process */ |
a62428c0 | 1789 | debug_work_deactivate(work); |
c8e55f36 | 1790 | hlist_add_head(&worker->hentry, bwh); |
c34056a3 | 1791 | worker->current_work = work; |
8cca0eea | 1792 | worker->current_cwq = cwq; |
73f53c4a | 1793 | work_color = get_work_color(work); |
7a22ad75 | 1794 | |
7a22ad75 TH |
1795 | /* record the current cpu number in the work data and dequeue */ |
1796 | set_work_cpu(work, gcwq->cpu); | |
a62428c0 TH |
1797 | list_del_init(&work->entry); |
1798 | ||
649027d7 TH |
1799 | /* |
1800 | * If HIGHPRI_PENDING, check the next work, and, if HIGHPRI, | |
1801 | * wake up another worker; otherwise, clear HIGHPRI_PENDING. | |
1802 | */ | |
1803 | if (unlikely(gcwq->flags & GCWQ_HIGHPRI_PENDING)) { | |
1804 | struct work_struct *nwork = list_first_entry(&gcwq->worklist, | |
1805 | struct work_struct, entry); | |
1806 | ||
1807 | if (!list_empty(&gcwq->worklist) && | |
1808 | get_work_cwq(nwork)->wq->flags & WQ_HIGHPRI) | |
1809 | wake_up_worker(gcwq); | |
1810 | else | |
1811 | gcwq->flags &= ~GCWQ_HIGHPRI_PENDING; | |
1812 | } | |
1813 | ||
fb0e7beb TH |
1814 | /* |
1815 | * CPU intensive works don't participate in concurrency | |
1816 | * management. They're the scheduler's responsibility. | |
1817 | */ | |
1818 | if (unlikely(cpu_intensive)) | |
1819 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
1820 | ||
8b03ae3c | 1821 | spin_unlock_irq(&gcwq->lock); |
a62428c0 | 1822 | |
a62428c0 TH |
1823 | work_clear_pending(work); |
1824 | lock_map_acquire(&cwq->wq->lockdep_map); | |
1825 | lock_map_acquire(&lockdep_map); | |
e36c886a | 1826 | trace_workqueue_execute_start(work); |
a62428c0 | 1827 | f(work); |
e36c886a AV |
1828 | /* |
1829 | * While we must be careful to not use "work" after this, the trace | |
1830 | * point will only record its address. | |
1831 | */ | |
1832 | trace_workqueue_execute_end(work); | |
a62428c0 TH |
1833 | lock_map_release(&lockdep_map); |
1834 | lock_map_release(&cwq->wq->lockdep_map); | |
1835 | ||
1836 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
1837 | printk(KERN_ERR "BUG: workqueue leaked lock or atomic: " | |
1838 | "%s/0x%08x/%d\n", | |
1839 | current->comm, preempt_count(), task_pid_nr(current)); | |
1840 | printk(KERN_ERR " last function: "); | |
1841 | print_symbol("%s\n", (unsigned long)f); | |
1842 | debug_show_held_locks(current); | |
1843 | dump_stack(); | |
1844 | } | |
1845 | ||
8b03ae3c | 1846 | spin_lock_irq(&gcwq->lock); |
a62428c0 | 1847 | |
fb0e7beb TH |
1848 | /* clear cpu intensive status */ |
1849 | if (unlikely(cpu_intensive)) | |
1850 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
1851 | ||
a62428c0 | 1852 | /* we're done with it, release */ |
c8e55f36 | 1853 | hlist_del_init(&worker->hentry); |
c34056a3 | 1854 | worker->current_work = NULL; |
8cca0eea | 1855 | worker->current_cwq = NULL; |
8a2e8e5d | 1856 | cwq_dec_nr_in_flight(cwq, work_color, false); |
a62428c0 TH |
1857 | } |
1858 | ||
affee4b2 TH |
1859 | /** |
1860 | * process_scheduled_works - process scheduled works | |
1861 | * @worker: self | |
1862 | * | |
1863 | * Process all scheduled works. Please note that the scheduled list | |
1864 | * may change while processing a work, so this function repeatedly | |
1865 | * fetches a work from the top and executes it. | |
1866 | * | |
1867 | * CONTEXT: | |
8b03ae3c | 1868 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed |
affee4b2 TH |
1869 | * multiple times. |
1870 | */ | |
1871 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 1872 | { |
affee4b2 TH |
1873 | while (!list_empty(&worker->scheduled)) { |
1874 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 1875 | struct work_struct, entry); |
c34056a3 | 1876 | process_one_work(worker, work); |
1da177e4 | 1877 | } |
1da177e4 LT |
1878 | } |
1879 | ||
4690c4ab TH |
1880 | /** |
1881 | * worker_thread - the worker thread function | |
c34056a3 | 1882 | * @__worker: self |
4690c4ab | 1883 | * |
e22bee78 TH |
1884 | * The gcwq worker thread function. There's a single dynamic pool of |
1885 | * these per each cpu. These workers process all works regardless of | |
1886 | * their specific target workqueue. The only exception is works which | |
1887 | * belong to workqueues with a rescuer which will be explained in | |
1888 | * rescuer_thread(). | |
4690c4ab | 1889 | */ |
c34056a3 | 1890 | static int worker_thread(void *__worker) |
1da177e4 | 1891 | { |
c34056a3 | 1892 | struct worker *worker = __worker; |
8b03ae3c | 1893 | struct global_cwq *gcwq = worker->gcwq; |
1da177e4 | 1894 | |
e22bee78 TH |
1895 | /* tell the scheduler that this is a workqueue worker */ |
1896 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 1897 | woke_up: |
c8e55f36 | 1898 | spin_lock_irq(&gcwq->lock); |
1da177e4 | 1899 | |
c8e55f36 TH |
1900 | /* DIE can be set only while we're idle, checking here is enough */ |
1901 | if (worker->flags & WORKER_DIE) { | |
1902 | spin_unlock_irq(&gcwq->lock); | |
e22bee78 | 1903 | worker->task->flags &= ~PF_WQ_WORKER; |
c8e55f36 TH |
1904 | return 0; |
1905 | } | |
affee4b2 | 1906 | |
c8e55f36 | 1907 | worker_leave_idle(worker); |
db7bccf4 | 1908 | recheck: |
e22bee78 TH |
1909 | /* no more worker necessary? */ |
1910 | if (!need_more_worker(gcwq)) | |
1911 | goto sleep; | |
1912 | ||
1913 | /* do we need to manage? */ | |
1914 | if (unlikely(!may_start_working(gcwq)) && manage_workers(worker)) | |
1915 | goto recheck; | |
1916 | ||
c8e55f36 TH |
1917 | /* |
1918 | * ->scheduled list can only be filled while a worker is | |
1919 | * preparing to process a work or actually processing it. | |
1920 | * Make sure nobody diddled with it while I was sleeping. | |
1921 | */ | |
1922 | BUG_ON(!list_empty(&worker->scheduled)); | |
1923 | ||
e22bee78 TH |
1924 | /* |
1925 | * When control reaches this point, we're guaranteed to have | |
1926 | * at least one idle worker or that someone else has already | |
1927 | * assumed the manager role. | |
1928 | */ | |
1929 | worker_clr_flags(worker, WORKER_PREP); | |
1930 | ||
1931 | do { | |
c8e55f36 | 1932 | struct work_struct *work = |
7e11629d | 1933 | list_first_entry(&gcwq->worklist, |
c8e55f36 TH |
1934 | struct work_struct, entry); |
1935 | ||
1936 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
1937 | /* optimization path, not strictly necessary */ | |
1938 | process_one_work(worker, work); | |
1939 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 1940 | process_scheduled_works(worker); |
c8e55f36 TH |
1941 | } else { |
1942 | move_linked_works(work, &worker->scheduled, NULL); | |
1943 | process_scheduled_works(worker); | |
affee4b2 | 1944 | } |
e22bee78 TH |
1945 | } while (keep_working(gcwq)); |
1946 | ||
1947 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 1948 | sleep: |
e22bee78 TH |
1949 | if (unlikely(need_to_manage_workers(gcwq)) && manage_workers(worker)) |
1950 | goto recheck; | |
d313dd85 | 1951 | |
c8e55f36 | 1952 | /* |
e22bee78 TH |
1953 | * gcwq->lock is held and there's no work to process and no |
1954 | * need to manage, sleep. Workers are woken up only while | |
1955 | * holding gcwq->lock or from local cpu, so setting the | |
1956 | * current state before releasing gcwq->lock is enough to | |
1957 | * prevent losing any event. | |
c8e55f36 TH |
1958 | */ |
1959 | worker_enter_idle(worker); | |
1960 | __set_current_state(TASK_INTERRUPTIBLE); | |
1961 | spin_unlock_irq(&gcwq->lock); | |
1962 | schedule(); | |
1963 | goto woke_up; | |
1da177e4 LT |
1964 | } |
1965 | ||
e22bee78 TH |
1966 | /** |
1967 | * rescuer_thread - the rescuer thread function | |
1968 | * @__wq: the associated workqueue | |
1969 | * | |
1970 | * Workqueue rescuer thread function. There's one rescuer for each | |
1971 | * workqueue which has WQ_RESCUER set. | |
1972 | * | |
1973 | * Regular work processing on a gcwq may block trying to create a new | |
1974 | * worker which uses GFP_KERNEL allocation which has slight chance of | |
1975 | * developing into deadlock if some works currently on the same queue | |
1976 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
1977 | * the problem rescuer solves. | |
1978 | * | |
1979 | * When such condition is possible, the gcwq summons rescuers of all | |
1980 | * workqueues which have works queued on the gcwq and let them process | |
1981 | * those works so that forward progress can be guaranteed. | |
1982 | * | |
1983 | * This should happen rarely. | |
1984 | */ | |
1985 | static int rescuer_thread(void *__wq) | |
1986 | { | |
1987 | struct workqueue_struct *wq = __wq; | |
1988 | struct worker *rescuer = wq->rescuer; | |
1989 | struct list_head *scheduled = &rescuer->scheduled; | |
f3421797 | 1990 | bool is_unbound = wq->flags & WQ_UNBOUND; |
e22bee78 TH |
1991 | unsigned int cpu; |
1992 | ||
1993 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
1994 | repeat: | |
1995 | set_current_state(TASK_INTERRUPTIBLE); | |
1996 | ||
1997 | if (kthread_should_stop()) | |
1998 | return 0; | |
1999 | ||
f3421797 TH |
2000 | /* |
2001 | * See whether any cpu is asking for help. Unbounded | |
2002 | * workqueues use cpu 0 in mayday_mask for CPU_UNBOUND. | |
2003 | */ | |
f2e005aa | 2004 | for_each_mayday_cpu(cpu, wq->mayday_mask) { |
f3421797 TH |
2005 | unsigned int tcpu = is_unbound ? WORK_CPU_UNBOUND : cpu; |
2006 | struct cpu_workqueue_struct *cwq = get_cwq(tcpu, wq); | |
e22bee78 TH |
2007 | struct global_cwq *gcwq = cwq->gcwq; |
2008 | struct work_struct *work, *n; | |
2009 | ||
2010 | __set_current_state(TASK_RUNNING); | |
f2e005aa | 2011 | mayday_clear_cpu(cpu, wq->mayday_mask); |
e22bee78 TH |
2012 | |
2013 | /* migrate to the target cpu if possible */ | |
2014 | rescuer->gcwq = gcwq; | |
2015 | worker_maybe_bind_and_lock(rescuer); | |
2016 | ||
2017 | /* | |
2018 | * Slurp in all works issued via this workqueue and | |
2019 | * process'em. | |
2020 | */ | |
2021 | BUG_ON(!list_empty(&rescuer->scheduled)); | |
2022 | list_for_each_entry_safe(work, n, &gcwq->worklist, entry) | |
2023 | if (get_work_cwq(work) == cwq) | |
2024 | move_linked_works(work, scheduled, &n); | |
2025 | ||
2026 | process_scheduled_works(rescuer); | |
2027 | spin_unlock_irq(&gcwq->lock); | |
2028 | } | |
2029 | ||
2030 | schedule(); | |
2031 | goto repeat; | |
1da177e4 LT |
2032 | } |
2033 | ||
fc2e4d70 ON |
2034 | struct wq_barrier { |
2035 | struct work_struct work; | |
2036 | struct completion done; | |
2037 | }; | |
2038 | ||
2039 | static void wq_barrier_func(struct work_struct *work) | |
2040 | { | |
2041 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2042 | complete(&barr->done); | |
2043 | } | |
2044 | ||
4690c4ab TH |
2045 | /** |
2046 | * insert_wq_barrier - insert a barrier work | |
2047 | * @cwq: cwq to insert barrier into | |
2048 | * @barr: wq_barrier to insert | |
affee4b2 TH |
2049 | * @target: target work to attach @barr to |
2050 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2051 | * |
affee4b2 TH |
2052 | * @barr is linked to @target such that @barr is completed only after |
2053 | * @target finishes execution. Please note that the ordering | |
2054 | * guarantee is observed only with respect to @target and on the local | |
2055 | * cpu. | |
2056 | * | |
2057 | * Currently, a queued barrier can't be canceled. This is because | |
2058 | * try_to_grab_pending() can't determine whether the work to be | |
2059 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2060 | * flag of the previous work while there must be a valid next work | |
2061 | * after a work with LINKED flag set. | |
2062 | * | |
2063 | * Note that when @worker is non-NULL, @target may be modified | |
2064 | * underneath us, so we can't reliably determine cwq from @target. | |
4690c4ab TH |
2065 | * |
2066 | * CONTEXT: | |
8b03ae3c | 2067 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 2068 | */ |
83c22520 | 2069 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, |
affee4b2 TH |
2070 | struct wq_barrier *barr, |
2071 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2072 | { |
affee4b2 TH |
2073 | struct list_head *head; |
2074 | unsigned int linked = 0; | |
2075 | ||
dc186ad7 | 2076 | /* |
8b03ae3c | 2077 | * debugobject calls are safe here even with gcwq->lock locked |
dc186ad7 TG |
2078 | * as we know for sure that this will not trigger any of the |
2079 | * checks and call back into the fixup functions where we | |
2080 | * might deadlock. | |
2081 | */ | |
2082 | INIT_WORK_ON_STACK(&barr->work, wq_barrier_func); | |
22df02bb | 2083 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2084 | init_completion(&barr->done); |
83c22520 | 2085 | |
affee4b2 TH |
2086 | /* |
2087 | * If @target is currently being executed, schedule the | |
2088 | * barrier to the worker; otherwise, put it after @target. | |
2089 | */ | |
2090 | if (worker) | |
2091 | head = worker->scheduled.next; | |
2092 | else { | |
2093 | unsigned long *bits = work_data_bits(target); | |
2094 | ||
2095 | head = target->entry.next; | |
2096 | /* there can already be other linked works, inherit and set */ | |
2097 | linked = *bits & WORK_STRUCT_LINKED; | |
2098 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2099 | } | |
2100 | ||
dc186ad7 | 2101 | debug_work_activate(&barr->work); |
affee4b2 TH |
2102 | insert_work(cwq, &barr->work, head, |
2103 | work_color_to_flags(WORK_NO_COLOR) | linked); | |
fc2e4d70 ON |
2104 | } |
2105 | ||
73f53c4a TH |
2106 | /** |
2107 | * flush_workqueue_prep_cwqs - prepare cwqs for workqueue flushing | |
2108 | * @wq: workqueue being flushed | |
2109 | * @flush_color: new flush color, < 0 for no-op | |
2110 | * @work_color: new work color, < 0 for no-op | |
2111 | * | |
2112 | * Prepare cwqs for workqueue flushing. | |
2113 | * | |
2114 | * If @flush_color is non-negative, flush_color on all cwqs should be | |
2115 | * -1. If no cwq has in-flight commands at the specified color, all | |
2116 | * cwq->flush_color's stay at -1 and %false is returned. If any cwq | |
2117 | * has in flight commands, its cwq->flush_color is set to | |
2118 | * @flush_color, @wq->nr_cwqs_to_flush is updated accordingly, cwq | |
2119 | * wakeup logic is armed and %true is returned. | |
2120 | * | |
2121 | * The caller should have initialized @wq->first_flusher prior to | |
2122 | * calling this function with non-negative @flush_color. If | |
2123 | * @flush_color is negative, no flush color update is done and %false | |
2124 | * is returned. | |
2125 | * | |
2126 | * If @work_color is non-negative, all cwqs should have the same | |
2127 | * work_color which is previous to @work_color and all will be | |
2128 | * advanced to @work_color. | |
2129 | * | |
2130 | * CONTEXT: | |
2131 | * mutex_lock(wq->flush_mutex). | |
2132 | * | |
2133 | * RETURNS: | |
2134 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2135 | * otherwise. | |
2136 | */ | |
2137 | static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq, | |
2138 | int flush_color, int work_color) | |
1da177e4 | 2139 | { |
73f53c4a TH |
2140 | bool wait = false; |
2141 | unsigned int cpu; | |
1da177e4 | 2142 | |
73f53c4a TH |
2143 | if (flush_color >= 0) { |
2144 | BUG_ON(atomic_read(&wq->nr_cwqs_to_flush)); | |
2145 | atomic_set(&wq->nr_cwqs_to_flush, 1); | |
1da177e4 | 2146 | } |
2355b70f | 2147 | |
f3421797 | 2148 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a | 2149 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
8b03ae3c | 2150 | struct global_cwq *gcwq = cwq->gcwq; |
fc2e4d70 | 2151 | |
8b03ae3c | 2152 | spin_lock_irq(&gcwq->lock); |
83c22520 | 2153 | |
73f53c4a TH |
2154 | if (flush_color >= 0) { |
2155 | BUG_ON(cwq->flush_color != -1); | |
fc2e4d70 | 2156 | |
73f53c4a TH |
2157 | if (cwq->nr_in_flight[flush_color]) { |
2158 | cwq->flush_color = flush_color; | |
2159 | atomic_inc(&wq->nr_cwqs_to_flush); | |
2160 | wait = true; | |
2161 | } | |
2162 | } | |
1da177e4 | 2163 | |
73f53c4a TH |
2164 | if (work_color >= 0) { |
2165 | BUG_ON(work_color != work_next_color(cwq->work_color)); | |
2166 | cwq->work_color = work_color; | |
2167 | } | |
1da177e4 | 2168 | |
8b03ae3c | 2169 | spin_unlock_irq(&gcwq->lock); |
1da177e4 | 2170 | } |
2355b70f | 2171 | |
73f53c4a TH |
2172 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_cwqs_to_flush)) |
2173 | complete(&wq->first_flusher->done); | |
14441960 | 2174 | |
73f53c4a | 2175 | return wait; |
1da177e4 LT |
2176 | } |
2177 | ||
0fcb78c2 | 2178 | /** |
1da177e4 | 2179 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2180 | * @wq: workqueue to flush |
1da177e4 LT |
2181 | * |
2182 | * Forces execution of the workqueue and blocks until its completion. | |
2183 | * This is typically used in driver shutdown handlers. | |
2184 | * | |
fc2e4d70 ON |
2185 | * We sleep until all works which were queued on entry have been handled, |
2186 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 2187 | */ |
7ad5b3a5 | 2188 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2189 | { |
73f53c4a TH |
2190 | struct wq_flusher this_flusher = { |
2191 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2192 | .flush_color = -1, | |
2193 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2194 | }; | |
2195 | int next_color; | |
1da177e4 | 2196 | |
3295f0ef IM |
2197 | lock_map_acquire(&wq->lockdep_map); |
2198 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
2199 | |
2200 | mutex_lock(&wq->flush_mutex); | |
2201 | ||
2202 | /* | |
2203 | * Start-to-wait phase | |
2204 | */ | |
2205 | next_color = work_next_color(wq->work_color); | |
2206 | ||
2207 | if (next_color != wq->flush_color) { | |
2208 | /* | |
2209 | * Color space is not full. The current work_color | |
2210 | * becomes our flush_color and work_color is advanced | |
2211 | * by one. | |
2212 | */ | |
2213 | BUG_ON(!list_empty(&wq->flusher_overflow)); | |
2214 | this_flusher.flush_color = wq->work_color; | |
2215 | wq->work_color = next_color; | |
2216 | ||
2217 | if (!wq->first_flusher) { | |
2218 | /* no flush in progress, become the first flusher */ | |
2219 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2220 | ||
2221 | wq->first_flusher = &this_flusher; | |
2222 | ||
2223 | if (!flush_workqueue_prep_cwqs(wq, wq->flush_color, | |
2224 | wq->work_color)) { | |
2225 | /* nothing to flush, done */ | |
2226 | wq->flush_color = next_color; | |
2227 | wq->first_flusher = NULL; | |
2228 | goto out_unlock; | |
2229 | } | |
2230 | } else { | |
2231 | /* wait in queue */ | |
2232 | BUG_ON(wq->flush_color == this_flusher.flush_color); | |
2233 | list_add_tail(&this_flusher.list, &wq->flusher_queue); | |
2234 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2235 | } | |
2236 | } else { | |
2237 | /* | |
2238 | * Oops, color space is full, wait on overflow queue. | |
2239 | * The next flush completion will assign us | |
2240 | * flush_color and transfer to flusher_queue. | |
2241 | */ | |
2242 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2243 | } | |
2244 | ||
2245 | mutex_unlock(&wq->flush_mutex); | |
2246 | ||
2247 | wait_for_completion(&this_flusher.done); | |
2248 | ||
2249 | /* | |
2250 | * Wake-up-and-cascade phase | |
2251 | * | |
2252 | * First flushers are responsible for cascading flushes and | |
2253 | * handling overflow. Non-first flushers can simply return. | |
2254 | */ | |
2255 | if (wq->first_flusher != &this_flusher) | |
2256 | return; | |
2257 | ||
2258 | mutex_lock(&wq->flush_mutex); | |
2259 | ||
4ce48b37 TH |
2260 | /* we might have raced, check again with mutex held */ |
2261 | if (wq->first_flusher != &this_flusher) | |
2262 | goto out_unlock; | |
2263 | ||
73f53c4a TH |
2264 | wq->first_flusher = NULL; |
2265 | ||
2266 | BUG_ON(!list_empty(&this_flusher.list)); | |
2267 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2268 | ||
2269 | while (true) { | |
2270 | struct wq_flusher *next, *tmp; | |
2271 | ||
2272 | /* complete all the flushers sharing the current flush color */ | |
2273 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2274 | if (next->flush_color != wq->flush_color) | |
2275 | break; | |
2276 | list_del_init(&next->list); | |
2277 | complete(&next->done); | |
2278 | } | |
2279 | ||
2280 | BUG_ON(!list_empty(&wq->flusher_overflow) && | |
2281 | wq->flush_color != work_next_color(wq->work_color)); | |
2282 | ||
2283 | /* this flush_color is finished, advance by one */ | |
2284 | wq->flush_color = work_next_color(wq->flush_color); | |
2285 | ||
2286 | /* one color has been freed, handle overflow queue */ | |
2287 | if (!list_empty(&wq->flusher_overflow)) { | |
2288 | /* | |
2289 | * Assign the same color to all overflowed | |
2290 | * flushers, advance work_color and append to | |
2291 | * flusher_queue. This is the start-to-wait | |
2292 | * phase for these overflowed flushers. | |
2293 | */ | |
2294 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2295 | tmp->flush_color = wq->work_color; | |
2296 | ||
2297 | wq->work_color = work_next_color(wq->work_color); | |
2298 | ||
2299 | list_splice_tail_init(&wq->flusher_overflow, | |
2300 | &wq->flusher_queue); | |
2301 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2302 | } | |
2303 | ||
2304 | if (list_empty(&wq->flusher_queue)) { | |
2305 | BUG_ON(wq->flush_color != wq->work_color); | |
2306 | break; | |
2307 | } | |
2308 | ||
2309 | /* | |
2310 | * Need to flush more colors. Make the next flusher | |
2311 | * the new first flusher and arm cwqs. | |
2312 | */ | |
2313 | BUG_ON(wq->flush_color == wq->work_color); | |
2314 | BUG_ON(wq->flush_color != next->flush_color); | |
2315 | ||
2316 | list_del_init(&next->list); | |
2317 | wq->first_flusher = next; | |
2318 | ||
2319 | if (flush_workqueue_prep_cwqs(wq, wq->flush_color, -1)) | |
2320 | break; | |
2321 | ||
2322 | /* | |
2323 | * Meh... this color is already done, clear first | |
2324 | * flusher and repeat cascading. | |
2325 | */ | |
2326 | wq->first_flusher = NULL; | |
2327 | } | |
2328 | ||
2329 | out_unlock: | |
2330 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 2331 | } |
ae90dd5d | 2332 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2333 | |
baf59022 TH |
2334 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr, |
2335 | bool wait_executing) | |
db700897 | 2336 | { |
affee4b2 | 2337 | struct worker *worker = NULL; |
8b03ae3c | 2338 | struct global_cwq *gcwq; |
db700897 | 2339 | struct cpu_workqueue_struct *cwq; |
db700897 ON |
2340 | |
2341 | might_sleep(); | |
7a22ad75 TH |
2342 | gcwq = get_work_gcwq(work); |
2343 | if (!gcwq) | |
baf59022 | 2344 | return false; |
db700897 | 2345 | |
8b03ae3c | 2346 | spin_lock_irq(&gcwq->lock); |
db700897 ON |
2347 | if (!list_empty(&work->entry)) { |
2348 | /* | |
2349 | * See the comment near try_to_grab_pending()->smp_rmb(). | |
7a22ad75 TH |
2350 | * If it was re-queued to a different gcwq under us, we |
2351 | * are not going to wait. | |
db700897 ON |
2352 | */ |
2353 | smp_rmb(); | |
7a22ad75 TH |
2354 | cwq = get_work_cwq(work); |
2355 | if (unlikely(!cwq || gcwq != cwq->gcwq)) | |
4690c4ab | 2356 | goto already_gone; |
baf59022 | 2357 | } else if (wait_executing) { |
7a22ad75 | 2358 | worker = find_worker_executing_work(gcwq, work); |
affee4b2 | 2359 | if (!worker) |
4690c4ab | 2360 | goto already_gone; |
7a22ad75 | 2361 | cwq = worker->current_cwq; |
baf59022 TH |
2362 | } else |
2363 | goto already_gone; | |
db700897 | 2364 | |
baf59022 | 2365 | insert_wq_barrier(cwq, barr, work, worker); |
8b03ae3c | 2366 | spin_unlock_irq(&gcwq->lock); |
7a22ad75 TH |
2367 | |
2368 | lock_map_acquire(&cwq->wq->lockdep_map); | |
2369 | lock_map_release(&cwq->wq->lockdep_map); | |
401a8d04 | 2370 | return true; |
4690c4ab | 2371 | already_gone: |
8b03ae3c | 2372 | spin_unlock_irq(&gcwq->lock); |
401a8d04 | 2373 | return false; |
db700897 | 2374 | } |
baf59022 TH |
2375 | |
2376 | /** | |
2377 | * flush_work - wait for a work to finish executing the last queueing instance | |
2378 | * @work: the work to flush | |
2379 | * | |
2380 | * Wait until @work has finished execution. This function considers | |
2381 | * only the last queueing instance of @work. If @work has been | |
2382 | * enqueued across different CPUs on a non-reentrant workqueue or on | |
2383 | * multiple workqueues, @work might still be executing on return on | |
2384 | * some of the CPUs from earlier queueing. | |
2385 | * | |
2386 | * If @work was queued only on a non-reentrant, ordered or unbound | |
2387 | * workqueue, @work is guaranteed to be idle on return if it hasn't | |
2388 | * been requeued since flush started. | |
2389 | * | |
2390 | * RETURNS: | |
2391 | * %true if flush_work() waited for the work to finish execution, | |
2392 | * %false if it was already idle. | |
2393 | */ | |
2394 | bool flush_work(struct work_struct *work) | |
2395 | { | |
2396 | struct wq_barrier barr; | |
2397 | ||
2398 | if (start_flush_work(work, &barr, true)) { | |
2399 | wait_for_completion(&barr.done); | |
2400 | destroy_work_on_stack(&barr.work); | |
2401 | return true; | |
2402 | } else | |
2403 | return false; | |
2404 | } | |
db700897 ON |
2405 | EXPORT_SYMBOL_GPL(flush_work); |
2406 | ||
401a8d04 TH |
2407 | static bool wait_on_cpu_work(struct global_cwq *gcwq, struct work_struct *work) |
2408 | { | |
2409 | struct wq_barrier barr; | |
2410 | struct worker *worker; | |
2411 | ||
2412 | spin_lock_irq(&gcwq->lock); | |
2413 | ||
2414 | worker = find_worker_executing_work(gcwq, work); | |
2415 | if (unlikely(worker)) | |
2416 | insert_wq_barrier(worker->current_cwq, &barr, work, worker); | |
2417 | ||
2418 | spin_unlock_irq(&gcwq->lock); | |
2419 | ||
2420 | if (unlikely(worker)) { | |
2421 | wait_for_completion(&barr.done); | |
2422 | destroy_work_on_stack(&barr.work); | |
2423 | return true; | |
2424 | } else | |
2425 | return false; | |
2426 | } | |
2427 | ||
2428 | static bool wait_on_work(struct work_struct *work) | |
2429 | { | |
2430 | bool ret = false; | |
2431 | int cpu; | |
2432 | ||
2433 | might_sleep(); | |
2434 | ||
2435 | lock_map_acquire(&work->lockdep_map); | |
2436 | lock_map_release(&work->lockdep_map); | |
2437 | ||
2438 | for_each_gcwq_cpu(cpu) | |
2439 | ret |= wait_on_cpu_work(get_gcwq(cpu), work); | |
2440 | return ret; | |
2441 | } | |
2442 | ||
09383498 TH |
2443 | /** |
2444 | * flush_work_sync - wait until a work has finished execution | |
2445 | * @work: the work to flush | |
2446 | * | |
2447 | * Wait until @work has finished execution. On return, it's | |
2448 | * guaranteed that all queueing instances of @work which happened | |
2449 | * before this function is called are finished. In other words, if | |
2450 | * @work hasn't been requeued since this function was called, @work is | |
2451 | * guaranteed to be idle on return. | |
2452 | * | |
2453 | * RETURNS: | |
2454 | * %true if flush_work_sync() waited for the work to finish execution, | |
2455 | * %false if it was already idle. | |
2456 | */ | |
2457 | bool flush_work_sync(struct work_struct *work) | |
2458 | { | |
2459 | struct wq_barrier barr; | |
2460 | bool pending, waited; | |
2461 | ||
2462 | /* we'll wait for executions separately, queue barr only if pending */ | |
2463 | pending = start_flush_work(work, &barr, false); | |
2464 | ||
2465 | /* wait for executions to finish */ | |
2466 | waited = wait_on_work(work); | |
2467 | ||
2468 | /* wait for the pending one */ | |
2469 | if (pending) { | |
2470 | wait_for_completion(&barr.done); | |
2471 | destroy_work_on_stack(&barr.work); | |
2472 | } | |
2473 | ||
2474 | return pending || waited; | |
2475 | } | |
2476 | EXPORT_SYMBOL_GPL(flush_work_sync); | |
2477 | ||
6e84d644 | 2478 | /* |
1f1f642e | 2479 | * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit, |
6e84d644 ON |
2480 | * so this work can't be re-armed in any way. |
2481 | */ | |
2482 | static int try_to_grab_pending(struct work_struct *work) | |
2483 | { | |
8b03ae3c | 2484 | struct global_cwq *gcwq; |
1f1f642e | 2485 | int ret = -1; |
6e84d644 | 2486 | |
22df02bb | 2487 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1f1f642e | 2488 | return 0; |
6e84d644 ON |
2489 | |
2490 | /* | |
2491 | * The queueing is in progress, or it is already queued. Try to | |
2492 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
2493 | */ | |
7a22ad75 TH |
2494 | gcwq = get_work_gcwq(work); |
2495 | if (!gcwq) | |
6e84d644 ON |
2496 | return ret; |
2497 | ||
8b03ae3c | 2498 | spin_lock_irq(&gcwq->lock); |
6e84d644 ON |
2499 | if (!list_empty(&work->entry)) { |
2500 | /* | |
7a22ad75 | 2501 | * This work is queued, but perhaps we locked the wrong gcwq. |
6e84d644 ON |
2502 | * In that case we must see the new value after rmb(), see |
2503 | * insert_work()->wmb(). | |
2504 | */ | |
2505 | smp_rmb(); | |
7a22ad75 | 2506 | if (gcwq == get_work_gcwq(work)) { |
dc186ad7 | 2507 | debug_work_deactivate(work); |
6e84d644 | 2508 | list_del_init(&work->entry); |
7a22ad75 | 2509 | cwq_dec_nr_in_flight(get_work_cwq(work), |
8a2e8e5d TH |
2510 | get_work_color(work), |
2511 | *work_data_bits(work) & WORK_STRUCT_DELAYED); | |
6e84d644 ON |
2512 | ret = 1; |
2513 | } | |
2514 | } | |
8b03ae3c | 2515 | spin_unlock_irq(&gcwq->lock); |
6e84d644 ON |
2516 | |
2517 | return ret; | |
2518 | } | |
2519 | ||
401a8d04 | 2520 | static bool __cancel_work_timer(struct work_struct *work, |
1f1f642e ON |
2521 | struct timer_list* timer) |
2522 | { | |
2523 | int ret; | |
2524 | ||
2525 | do { | |
2526 | ret = (timer && likely(del_timer(timer))); | |
2527 | if (!ret) | |
2528 | ret = try_to_grab_pending(work); | |
2529 | wait_on_work(work); | |
2530 | } while (unlikely(ret < 0)); | |
2531 | ||
7a22ad75 | 2532 | clear_work_data(work); |
1f1f642e ON |
2533 | return ret; |
2534 | } | |
2535 | ||
6e84d644 | 2536 | /** |
401a8d04 TH |
2537 | * cancel_work_sync - cancel a work and wait for it to finish |
2538 | * @work: the work to cancel | |
6e84d644 | 2539 | * |
401a8d04 TH |
2540 | * Cancel @work and wait for its execution to finish. This function |
2541 | * can be used even if the work re-queues itself or migrates to | |
2542 | * another workqueue. On return from this function, @work is | |
2543 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2544 | * |
401a8d04 TH |
2545 | * cancel_work_sync(&delayed_work->work) must not be used for |
2546 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2547 | * |
401a8d04 | 2548 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2549 | * queued can't be destroyed before this function returns. |
401a8d04 TH |
2550 | * |
2551 | * RETURNS: | |
2552 | * %true if @work was pending, %false otherwise. | |
6e84d644 | 2553 | */ |
401a8d04 | 2554 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2555 | { |
1f1f642e | 2556 | return __cancel_work_timer(work, NULL); |
b89deed3 | 2557 | } |
28e53bdd | 2558 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2559 | |
6e84d644 | 2560 | /** |
401a8d04 TH |
2561 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2562 | * @dwork: the delayed work to flush | |
6e84d644 | 2563 | * |
401a8d04 TH |
2564 | * Delayed timer is cancelled and the pending work is queued for |
2565 | * immediate execution. Like flush_work(), this function only | |
2566 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2567 | * |
401a8d04 TH |
2568 | * RETURNS: |
2569 | * %true if flush_work() waited for the work to finish execution, | |
2570 | * %false if it was already idle. | |
6e84d644 | 2571 | */ |
401a8d04 TH |
2572 | bool flush_delayed_work(struct delayed_work *dwork) |
2573 | { | |
2574 | if (del_timer_sync(&dwork->timer)) | |
2575 | __queue_work(raw_smp_processor_id(), | |
2576 | get_work_cwq(&dwork->work)->wq, &dwork->work); | |
2577 | return flush_work(&dwork->work); | |
2578 | } | |
2579 | EXPORT_SYMBOL(flush_delayed_work); | |
2580 | ||
09383498 TH |
2581 | /** |
2582 | * flush_delayed_work_sync - wait for a dwork to finish | |
2583 | * @dwork: the delayed work to flush | |
2584 | * | |
2585 | * Delayed timer is cancelled and the pending work is queued for | |
2586 | * execution immediately. Other than timer handling, its behavior | |
2587 | * is identical to flush_work_sync(). | |
2588 | * | |
2589 | * RETURNS: | |
2590 | * %true if flush_work_sync() waited for the work to finish execution, | |
2591 | * %false if it was already idle. | |
2592 | */ | |
2593 | bool flush_delayed_work_sync(struct delayed_work *dwork) | |
2594 | { | |
2595 | if (del_timer_sync(&dwork->timer)) | |
2596 | __queue_work(raw_smp_processor_id(), | |
2597 | get_work_cwq(&dwork->work)->wq, &dwork->work); | |
2598 | return flush_work_sync(&dwork->work); | |
2599 | } | |
2600 | EXPORT_SYMBOL(flush_delayed_work_sync); | |
2601 | ||
401a8d04 TH |
2602 | /** |
2603 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
2604 | * @dwork: the delayed work cancel | |
2605 | * | |
2606 | * This is cancel_work_sync() for delayed works. | |
2607 | * | |
2608 | * RETURNS: | |
2609 | * %true if @dwork was pending, %false otherwise. | |
2610 | */ | |
2611 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 2612 | { |
1f1f642e | 2613 | return __cancel_work_timer(&dwork->work, &dwork->timer); |
6e84d644 | 2614 | } |
f5a421a4 | 2615 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 2616 | |
0fcb78c2 REB |
2617 | /** |
2618 | * schedule_work - put work task in global workqueue | |
2619 | * @work: job to be done | |
2620 | * | |
5b0f437d BVA |
2621 | * Returns zero if @work was already on the kernel-global workqueue and |
2622 | * non-zero otherwise. | |
2623 | * | |
2624 | * This puts a job in the kernel-global workqueue if it was not already | |
2625 | * queued and leaves it in the same position on the kernel-global | |
2626 | * workqueue otherwise. | |
0fcb78c2 | 2627 | */ |
7ad5b3a5 | 2628 | int schedule_work(struct work_struct *work) |
1da177e4 | 2629 | { |
d320c038 | 2630 | return queue_work(system_wq, work); |
1da177e4 | 2631 | } |
ae90dd5d | 2632 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 2633 | |
c1a220e7 ZR |
2634 | /* |
2635 | * schedule_work_on - put work task on a specific cpu | |
2636 | * @cpu: cpu to put the work task on | |
2637 | * @work: job to be done | |
2638 | * | |
2639 | * This puts a job on a specific cpu | |
2640 | */ | |
2641 | int schedule_work_on(int cpu, struct work_struct *work) | |
2642 | { | |
d320c038 | 2643 | return queue_work_on(cpu, system_wq, work); |
c1a220e7 ZR |
2644 | } |
2645 | EXPORT_SYMBOL(schedule_work_on); | |
2646 | ||
0fcb78c2 REB |
2647 | /** |
2648 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
2649 | * @dwork: job to be done |
2650 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
2651 | * |
2652 | * After waiting for a given time this puts a job in the kernel-global | |
2653 | * workqueue. | |
2654 | */ | |
7ad5b3a5 | 2655 | int schedule_delayed_work(struct delayed_work *dwork, |
82f67cd9 | 2656 | unsigned long delay) |
1da177e4 | 2657 | { |
d320c038 | 2658 | return queue_delayed_work(system_wq, dwork, delay); |
1da177e4 | 2659 | } |
ae90dd5d | 2660 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 2661 | |
0fcb78c2 REB |
2662 | /** |
2663 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
2664 | * @cpu: cpu to use | |
52bad64d | 2665 | * @dwork: job to be done |
0fcb78c2 REB |
2666 | * @delay: number of jiffies to wait |
2667 | * | |
2668 | * After waiting for a given time this puts a job in the kernel-global | |
2669 | * workqueue on the specified CPU. | |
2670 | */ | |
1da177e4 | 2671 | int schedule_delayed_work_on(int cpu, |
52bad64d | 2672 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 2673 | { |
d320c038 | 2674 | return queue_delayed_work_on(cpu, system_wq, dwork, delay); |
1da177e4 | 2675 | } |
ae90dd5d | 2676 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 2677 | |
b6136773 AM |
2678 | /** |
2679 | * schedule_on_each_cpu - call a function on each online CPU from keventd | |
2680 | * @func: the function to call | |
b6136773 AM |
2681 | * |
2682 | * Returns zero on success. | |
2683 | * Returns -ve errno on failure. | |
2684 | * | |
b6136773 AM |
2685 | * schedule_on_each_cpu() is very slow. |
2686 | */ | |
65f27f38 | 2687 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
2688 | { |
2689 | int cpu; | |
38f51568 | 2690 | struct work_struct __percpu *works; |
15316ba8 | 2691 | |
b6136773 AM |
2692 | works = alloc_percpu(struct work_struct); |
2693 | if (!works) | |
15316ba8 | 2694 | return -ENOMEM; |
b6136773 | 2695 | |
93981800 TH |
2696 | get_online_cpus(); |
2697 | ||
15316ba8 | 2698 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
2699 | struct work_struct *work = per_cpu_ptr(works, cpu); |
2700 | ||
2701 | INIT_WORK(work, func); | |
b71ab8c2 | 2702 | schedule_work_on(cpu, work); |
65a64464 | 2703 | } |
93981800 TH |
2704 | |
2705 | for_each_online_cpu(cpu) | |
2706 | flush_work(per_cpu_ptr(works, cpu)); | |
2707 | ||
95402b38 | 2708 | put_online_cpus(); |
b6136773 | 2709 | free_percpu(works); |
15316ba8 CL |
2710 | return 0; |
2711 | } | |
2712 | ||
eef6a7d5 AS |
2713 | /** |
2714 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
2715 | * | |
2716 | * Forces execution of the kernel-global workqueue and blocks until its | |
2717 | * completion. | |
2718 | * | |
2719 | * Think twice before calling this function! It's very easy to get into | |
2720 | * trouble if you don't take great care. Either of the following situations | |
2721 | * will lead to deadlock: | |
2722 | * | |
2723 | * One of the work items currently on the workqueue needs to acquire | |
2724 | * a lock held by your code or its caller. | |
2725 | * | |
2726 | * Your code is running in the context of a work routine. | |
2727 | * | |
2728 | * They will be detected by lockdep when they occur, but the first might not | |
2729 | * occur very often. It depends on what work items are on the workqueue and | |
2730 | * what locks they need, which you have no control over. | |
2731 | * | |
2732 | * In most situations flushing the entire workqueue is overkill; you merely | |
2733 | * need to know that a particular work item isn't queued and isn't running. | |
2734 | * In such cases you should use cancel_delayed_work_sync() or | |
2735 | * cancel_work_sync() instead. | |
2736 | */ | |
1da177e4 LT |
2737 | void flush_scheduled_work(void) |
2738 | { | |
d320c038 | 2739 | flush_workqueue(system_wq); |
1da177e4 | 2740 | } |
ae90dd5d | 2741 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 2742 | |
1fa44eca JB |
2743 | /** |
2744 | * execute_in_process_context - reliably execute the routine with user context | |
2745 | * @fn: the function to execute | |
1fa44eca JB |
2746 | * @ew: guaranteed storage for the execute work structure (must |
2747 | * be available when the work executes) | |
2748 | * | |
2749 | * Executes the function immediately if process context is available, | |
2750 | * otherwise schedules the function for delayed execution. | |
2751 | * | |
2752 | * Returns: 0 - function was executed | |
2753 | * 1 - function was scheduled for execution | |
2754 | */ | |
65f27f38 | 2755 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
2756 | { |
2757 | if (!in_interrupt()) { | |
65f27f38 | 2758 | fn(&ew->work); |
1fa44eca JB |
2759 | return 0; |
2760 | } | |
2761 | ||
65f27f38 | 2762 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
2763 | schedule_work(&ew->work); |
2764 | ||
2765 | return 1; | |
2766 | } | |
2767 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
2768 | ||
1da177e4 LT |
2769 | int keventd_up(void) |
2770 | { | |
d320c038 | 2771 | return system_wq != NULL; |
1da177e4 LT |
2772 | } |
2773 | ||
bdbc5dd7 | 2774 | static int alloc_cwqs(struct workqueue_struct *wq) |
0f900049 | 2775 | { |
65a64464 | 2776 | /* |
0f900049 TH |
2777 | * cwqs are forced aligned according to WORK_STRUCT_FLAG_BITS. |
2778 | * Make sure that the alignment isn't lower than that of | |
2779 | * unsigned long long. | |
65a64464 | 2780 | */ |
0f900049 TH |
2781 | const size_t size = sizeof(struct cpu_workqueue_struct); |
2782 | const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS, | |
2783 | __alignof__(unsigned long long)); | |
931ac77e TH |
2784 | #ifdef CONFIG_SMP |
2785 | bool percpu = !(wq->flags & WQ_UNBOUND); | |
2786 | #else | |
2787 | bool percpu = false; | |
2788 | #endif | |
65a64464 | 2789 | |
931ac77e | 2790 | if (percpu) |
f3421797 | 2791 | wq->cpu_wq.pcpu = __alloc_percpu(size, align); |
931ac77e | 2792 | else { |
f3421797 TH |
2793 | void *ptr; |
2794 | ||
2795 | /* | |
2796 | * Allocate enough room to align cwq and put an extra | |
2797 | * pointer at the end pointing back to the originally | |
2798 | * allocated pointer which will be used for free. | |
2799 | */ | |
2800 | ptr = kzalloc(size + align + sizeof(void *), GFP_KERNEL); | |
2801 | if (ptr) { | |
2802 | wq->cpu_wq.single = PTR_ALIGN(ptr, align); | |
2803 | *(void **)(wq->cpu_wq.single + 1) = ptr; | |
2804 | } | |
bdbc5dd7 | 2805 | } |
f3421797 | 2806 | |
0f900049 | 2807 | /* just in case, make sure it's actually aligned */ |
bdbc5dd7 TH |
2808 | BUG_ON(!IS_ALIGNED(wq->cpu_wq.v, align)); |
2809 | return wq->cpu_wq.v ? 0 : -ENOMEM; | |
0f900049 TH |
2810 | } |
2811 | ||
bdbc5dd7 | 2812 | static void free_cwqs(struct workqueue_struct *wq) |
0f900049 | 2813 | { |
931ac77e TH |
2814 | #ifdef CONFIG_SMP |
2815 | bool percpu = !(wq->flags & WQ_UNBOUND); | |
2816 | #else | |
2817 | bool percpu = false; | |
2818 | #endif | |
2819 | ||
2820 | if (percpu) | |
f3421797 TH |
2821 | free_percpu(wq->cpu_wq.pcpu); |
2822 | else if (wq->cpu_wq.single) { | |
2823 | /* the pointer to free is stored right after the cwq */ | |
bdbc5dd7 | 2824 | kfree(*(void **)(wq->cpu_wq.single + 1)); |
f3421797 | 2825 | } |
0f900049 TH |
2826 | } |
2827 | ||
f3421797 TH |
2828 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
2829 | const char *name) | |
b71ab8c2 | 2830 | { |
f3421797 TH |
2831 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
2832 | ||
2833 | if (max_active < 1 || max_active > lim) | |
b71ab8c2 TH |
2834 | printk(KERN_WARNING "workqueue: max_active %d requested for %s " |
2835 | "is out of range, clamping between %d and %d\n", | |
f3421797 | 2836 | max_active, name, 1, lim); |
b71ab8c2 | 2837 | |
f3421797 | 2838 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
2839 | } |
2840 | ||
d320c038 TH |
2841 | struct workqueue_struct *__alloc_workqueue_key(const char *name, |
2842 | unsigned int flags, | |
2843 | int max_active, | |
2844 | struct lock_class_key *key, | |
2845 | const char *lock_name) | |
1da177e4 | 2846 | { |
1da177e4 | 2847 | struct workqueue_struct *wq; |
c34056a3 | 2848 | unsigned int cpu; |
1da177e4 | 2849 | |
f3421797 TH |
2850 | /* |
2851 | * Unbound workqueues aren't concurrency managed and should be | |
2852 | * dispatched to workers immediately. | |
2853 | */ | |
2854 | if (flags & WQ_UNBOUND) | |
2855 | flags |= WQ_HIGHPRI; | |
2856 | ||
d320c038 | 2857 | max_active = max_active ?: WQ_DFL_ACTIVE; |
f3421797 | 2858 | max_active = wq_clamp_max_active(max_active, flags, name); |
1e19ffc6 | 2859 | |
3af24433 ON |
2860 | wq = kzalloc(sizeof(*wq), GFP_KERNEL); |
2861 | if (!wq) | |
4690c4ab | 2862 | goto err; |
3af24433 | 2863 | |
97e37d7b | 2864 | wq->flags = flags; |
a0a1a5fd | 2865 | wq->saved_max_active = max_active; |
73f53c4a TH |
2866 | mutex_init(&wq->flush_mutex); |
2867 | atomic_set(&wq->nr_cwqs_to_flush, 0); | |
2868 | INIT_LIST_HEAD(&wq->flusher_queue); | |
2869 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
502ca9d8 | 2870 | |
3af24433 | 2871 | wq->name = name; |
eb13ba87 | 2872 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 2873 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 2874 | |
bdbc5dd7 TH |
2875 | if (alloc_cwqs(wq) < 0) |
2876 | goto err; | |
2877 | ||
f3421797 | 2878 | for_each_cwq_cpu(cpu, wq) { |
1537663f | 2879 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
8b03ae3c | 2880 | struct global_cwq *gcwq = get_gcwq(cpu); |
1537663f | 2881 | |
0f900049 | 2882 | BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK); |
8b03ae3c | 2883 | cwq->gcwq = gcwq; |
c34056a3 | 2884 | cwq->wq = wq; |
73f53c4a | 2885 | cwq->flush_color = -1; |
1e19ffc6 | 2886 | cwq->max_active = max_active; |
1e19ffc6 | 2887 | INIT_LIST_HEAD(&cwq->delayed_works); |
e22bee78 | 2888 | } |
1537663f | 2889 | |
e22bee78 TH |
2890 | if (flags & WQ_RESCUER) { |
2891 | struct worker *rescuer; | |
2892 | ||
f2e005aa | 2893 | if (!alloc_mayday_mask(&wq->mayday_mask, GFP_KERNEL)) |
e22bee78 TH |
2894 | goto err; |
2895 | ||
2896 | wq->rescuer = rescuer = alloc_worker(); | |
2897 | if (!rescuer) | |
2898 | goto err; | |
2899 | ||
2900 | rescuer->task = kthread_create(rescuer_thread, wq, "%s", name); | |
2901 | if (IS_ERR(rescuer->task)) | |
2902 | goto err; | |
2903 | ||
e22bee78 TH |
2904 | rescuer->task->flags |= PF_THREAD_BOUND; |
2905 | wake_up_process(rescuer->task); | |
3af24433 ON |
2906 | } |
2907 | ||
a0a1a5fd TH |
2908 | /* |
2909 | * workqueue_lock protects global freeze state and workqueues | |
2910 | * list. Grab it, set max_active accordingly and add the new | |
2911 | * workqueue to workqueues list. | |
2912 | */ | |
1537663f | 2913 | spin_lock(&workqueue_lock); |
a0a1a5fd TH |
2914 | |
2915 | if (workqueue_freezing && wq->flags & WQ_FREEZEABLE) | |
f3421797 | 2916 | for_each_cwq_cpu(cpu, wq) |
a0a1a5fd TH |
2917 | get_cwq(cpu, wq)->max_active = 0; |
2918 | ||
1537663f | 2919 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 2920 | |
1537663f TH |
2921 | spin_unlock(&workqueue_lock); |
2922 | ||
3af24433 | 2923 | return wq; |
4690c4ab TH |
2924 | err: |
2925 | if (wq) { | |
bdbc5dd7 | 2926 | free_cwqs(wq); |
f2e005aa | 2927 | free_mayday_mask(wq->mayday_mask); |
e22bee78 | 2928 | kfree(wq->rescuer); |
4690c4ab TH |
2929 | kfree(wq); |
2930 | } | |
2931 | return NULL; | |
3af24433 | 2932 | } |
d320c038 | 2933 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 2934 | |
3af24433 ON |
2935 | /** |
2936 | * destroy_workqueue - safely terminate a workqueue | |
2937 | * @wq: target workqueue | |
2938 | * | |
2939 | * Safely destroy a workqueue. All work currently pending will be done first. | |
2940 | */ | |
2941 | void destroy_workqueue(struct workqueue_struct *wq) | |
2942 | { | |
c8e55f36 | 2943 | unsigned int cpu; |
3af24433 | 2944 | |
e41e704b | 2945 | wq->flags |= WQ_DYING; |
a0a1a5fd TH |
2946 | flush_workqueue(wq); |
2947 | ||
2948 | /* | |
2949 | * wq list is used to freeze wq, remove from list after | |
2950 | * flushing is complete in case freeze races us. | |
2951 | */ | |
95402b38 | 2952 | spin_lock(&workqueue_lock); |
b1f4ec17 | 2953 | list_del(&wq->list); |
95402b38 | 2954 | spin_unlock(&workqueue_lock); |
3af24433 | 2955 | |
e22bee78 | 2956 | /* sanity check */ |
f3421797 | 2957 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a TH |
2958 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
2959 | int i; | |
2960 | ||
73f53c4a TH |
2961 | for (i = 0; i < WORK_NR_COLORS; i++) |
2962 | BUG_ON(cwq->nr_in_flight[i]); | |
1e19ffc6 TH |
2963 | BUG_ON(cwq->nr_active); |
2964 | BUG_ON(!list_empty(&cwq->delayed_works)); | |
73f53c4a | 2965 | } |
9b41ea72 | 2966 | |
e22bee78 TH |
2967 | if (wq->flags & WQ_RESCUER) { |
2968 | kthread_stop(wq->rescuer->task); | |
f2e005aa | 2969 | free_mayday_mask(wq->mayday_mask); |
8d9df9f0 | 2970 | kfree(wq->rescuer); |
e22bee78 TH |
2971 | } |
2972 | ||
bdbc5dd7 | 2973 | free_cwqs(wq); |
3af24433 ON |
2974 | kfree(wq); |
2975 | } | |
2976 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
2977 | ||
dcd989cb TH |
2978 | /** |
2979 | * workqueue_set_max_active - adjust max_active of a workqueue | |
2980 | * @wq: target workqueue | |
2981 | * @max_active: new max_active value. | |
2982 | * | |
2983 | * Set max_active of @wq to @max_active. | |
2984 | * | |
2985 | * CONTEXT: | |
2986 | * Don't call from IRQ context. | |
2987 | */ | |
2988 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
2989 | { | |
2990 | unsigned int cpu; | |
2991 | ||
f3421797 | 2992 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb TH |
2993 | |
2994 | spin_lock(&workqueue_lock); | |
2995 | ||
2996 | wq->saved_max_active = max_active; | |
2997 | ||
f3421797 | 2998 | for_each_cwq_cpu(cpu, wq) { |
dcd989cb TH |
2999 | struct global_cwq *gcwq = get_gcwq(cpu); |
3000 | ||
3001 | spin_lock_irq(&gcwq->lock); | |
3002 | ||
3003 | if (!(wq->flags & WQ_FREEZEABLE) || | |
3004 | !(gcwq->flags & GCWQ_FREEZING)) | |
3005 | get_cwq(gcwq->cpu, wq)->max_active = max_active; | |
9bfb1839 | 3006 | |
dcd989cb | 3007 | spin_unlock_irq(&gcwq->lock); |
65a64464 | 3008 | } |
93981800 | 3009 | |
dcd989cb | 3010 | spin_unlock(&workqueue_lock); |
15316ba8 | 3011 | } |
dcd989cb | 3012 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 3013 | |
eef6a7d5 | 3014 | /** |
dcd989cb TH |
3015 | * workqueue_congested - test whether a workqueue is congested |
3016 | * @cpu: CPU in question | |
3017 | * @wq: target workqueue | |
eef6a7d5 | 3018 | * |
dcd989cb TH |
3019 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
3020 | * no synchronization around this function and the test result is | |
3021 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 3022 | * |
dcd989cb TH |
3023 | * RETURNS: |
3024 | * %true if congested, %false otherwise. | |
eef6a7d5 | 3025 | */ |
dcd989cb | 3026 | bool workqueue_congested(unsigned int cpu, struct workqueue_struct *wq) |
1da177e4 | 3027 | { |
dcd989cb TH |
3028 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3029 | ||
3030 | return !list_empty(&cwq->delayed_works); | |
1da177e4 | 3031 | } |
dcd989cb | 3032 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 3033 | |
1fa44eca | 3034 | /** |
dcd989cb TH |
3035 | * work_cpu - return the last known associated cpu for @work |
3036 | * @work: the work of interest | |
1fa44eca | 3037 | * |
dcd989cb | 3038 | * RETURNS: |
bdbc5dd7 | 3039 | * CPU number if @work was ever queued. WORK_CPU_NONE otherwise. |
1fa44eca | 3040 | */ |
dcd989cb | 3041 | unsigned int work_cpu(struct work_struct *work) |
1fa44eca | 3042 | { |
dcd989cb | 3043 | struct global_cwq *gcwq = get_work_gcwq(work); |
1fa44eca | 3044 | |
bdbc5dd7 | 3045 | return gcwq ? gcwq->cpu : WORK_CPU_NONE; |
1fa44eca | 3046 | } |
dcd989cb | 3047 | EXPORT_SYMBOL_GPL(work_cpu); |
1fa44eca | 3048 | |
dcd989cb TH |
3049 | /** |
3050 | * work_busy - test whether a work is currently pending or running | |
3051 | * @work: the work to be tested | |
3052 | * | |
3053 | * Test whether @work is currently pending or running. There is no | |
3054 | * synchronization around this function and the test result is | |
3055 | * unreliable and only useful as advisory hints or for debugging. | |
3056 | * Especially for reentrant wqs, the pending state might hide the | |
3057 | * running state. | |
3058 | * | |
3059 | * RETURNS: | |
3060 | * OR'd bitmask of WORK_BUSY_* bits. | |
3061 | */ | |
3062 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 3063 | { |
dcd989cb TH |
3064 | struct global_cwq *gcwq = get_work_gcwq(work); |
3065 | unsigned long flags; | |
3066 | unsigned int ret = 0; | |
1da177e4 | 3067 | |
dcd989cb TH |
3068 | if (!gcwq) |
3069 | return false; | |
1da177e4 | 3070 | |
dcd989cb | 3071 | spin_lock_irqsave(&gcwq->lock, flags); |
1da177e4 | 3072 | |
dcd989cb TH |
3073 | if (work_pending(work)) |
3074 | ret |= WORK_BUSY_PENDING; | |
3075 | if (find_worker_executing_work(gcwq, work)) | |
3076 | ret |= WORK_BUSY_RUNNING; | |
1da177e4 | 3077 | |
dcd989cb | 3078 | spin_unlock_irqrestore(&gcwq->lock, flags); |
1da177e4 | 3079 | |
dcd989cb | 3080 | return ret; |
1da177e4 | 3081 | } |
dcd989cb | 3082 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 3083 | |
db7bccf4 TH |
3084 | /* |
3085 | * CPU hotplug. | |
3086 | * | |
e22bee78 TH |
3087 | * There are two challenges in supporting CPU hotplug. Firstly, there |
3088 | * are a lot of assumptions on strong associations among work, cwq and | |
3089 | * gcwq which make migrating pending and scheduled works very | |
3090 | * difficult to implement without impacting hot paths. Secondly, | |
3091 | * gcwqs serve mix of short, long and very long running works making | |
3092 | * blocked draining impractical. | |
3093 | * | |
3094 | * This is solved by allowing a gcwq to be detached from CPU, running | |
3095 | * it with unbound (rogue) workers and allowing it to be reattached | |
3096 | * later if the cpu comes back online. A separate thread is created | |
3097 | * to govern a gcwq in such state and is called the trustee of the | |
3098 | * gcwq. | |
db7bccf4 TH |
3099 | * |
3100 | * Trustee states and their descriptions. | |
3101 | * | |
3102 | * START Command state used on startup. On CPU_DOWN_PREPARE, a | |
3103 | * new trustee is started with this state. | |
3104 | * | |
3105 | * IN_CHARGE Once started, trustee will enter this state after | |
e22bee78 TH |
3106 | * assuming the manager role and making all existing |
3107 | * workers rogue. DOWN_PREPARE waits for trustee to | |
3108 | * enter this state. After reaching IN_CHARGE, trustee | |
3109 | * tries to execute the pending worklist until it's empty | |
3110 | * and the state is set to BUTCHER, or the state is set | |
3111 | * to RELEASE. | |
db7bccf4 TH |
3112 | * |
3113 | * BUTCHER Command state which is set by the cpu callback after | |
3114 | * the cpu has went down. Once this state is set trustee | |
3115 | * knows that there will be no new works on the worklist | |
3116 | * and once the worklist is empty it can proceed to | |
3117 | * killing idle workers. | |
3118 | * | |
3119 | * RELEASE Command state which is set by the cpu callback if the | |
3120 | * cpu down has been canceled or it has come online | |
3121 | * again. After recognizing this state, trustee stops | |
e22bee78 TH |
3122 | * trying to drain or butcher and clears ROGUE, rebinds |
3123 | * all remaining workers back to the cpu and releases | |
3124 | * manager role. | |
db7bccf4 TH |
3125 | * |
3126 | * DONE Trustee will enter this state after BUTCHER or RELEASE | |
3127 | * is complete. | |
3128 | * | |
3129 | * trustee CPU draining | |
3130 | * took over down complete | |
3131 | * START -----------> IN_CHARGE -----------> BUTCHER -----------> DONE | |
3132 | * | | ^ | |
3133 | * | CPU is back online v return workers | | |
3134 | * ----------------> RELEASE -------------- | |
3135 | */ | |
1da177e4 | 3136 | |
db7bccf4 TH |
3137 | /** |
3138 | * trustee_wait_event_timeout - timed event wait for trustee | |
3139 | * @cond: condition to wait for | |
3140 | * @timeout: timeout in jiffies | |
3141 | * | |
3142 | * wait_event_timeout() for trustee to use. Handles locking and | |
3143 | * checks for RELEASE request. | |
3144 | * | |
3145 | * CONTEXT: | |
3146 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
3147 | * multiple times. To be used by trustee. | |
3148 | * | |
3149 | * RETURNS: | |
3150 | * Positive indicating left time if @cond is satisfied, 0 if timed | |
3151 | * out, -1 if canceled. | |
3152 | */ | |
3153 | #define trustee_wait_event_timeout(cond, timeout) ({ \ | |
3154 | long __ret = (timeout); \ | |
3155 | while (!((cond) || (gcwq->trustee_state == TRUSTEE_RELEASE)) && \ | |
3156 | __ret) { \ | |
3157 | spin_unlock_irq(&gcwq->lock); \ | |
3158 | __wait_event_timeout(gcwq->trustee_wait, (cond) || \ | |
3159 | (gcwq->trustee_state == TRUSTEE_RELEASE), \ | |
3160 | __ret); \ | |
3161 | spin_lock_irq(&gcwq->lock); \ | |
3162 | } \ | |
3163 | gcwq->trustee_state == TRUSTEE_RELEASE ? -1 : (__ret); \ | |
3164 | }) | |
3af24433 | 3165 | |
db7bccf4 TH |
3166 | /** |
3167 | * trustee_wait_event - event wait for trustee | |
3168 | * @cond: condition to wait for | |
3169 | * | |
3170 | * wait_event() for trustee to use. Automatically handles locking and | |
3171 | * checks for CANCEL request. | |
3172 | * | |
3173 | * CONTEXT: | |
3174 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
3175 | * multiple times. To be used by trustee. | |
3176 | * | |
3177 | * RETURNS: | |
3178 | * 0 if @cond is satisfied, -1 if canceled. | |
3179 | */ | |
3180 | #define trustee_wait_event(cond) ({ \ | |
3181 | long __ret1; \ | |
3182 | __ret1 = trustee_wait_event_timeout(cond, MAX_SCHEDULE_TIMEOUT);\ | |
3183 | __ret1 < 0 ? -1 : 0; \ | |
3184 | }) | |
1da177e4 | 3185 | |
db7bccf4 | 3186 | static int __cpuinit trustee_thread(void *__gcwq) |
3af24433 | 3187 | { |
db7bccf4 TH |
3188 | struct global_cwq *gcwq = __gcwq; |
3189 | struct worker *worker; | |
e22bee78 | 3190 | struct work_struct *work; |
db7bccf4 | 3191 | struct hlist_node *pos; |
e22bee78 | 3192 | long rc; |
db7bccf4 | 3193 | int i; |
3af24433 | 3194 | |
db7bccf4 TH |
3195 | BUG_ON(gcwq->cpu != smp_processor_id()); |
3196 | ||
3197 | spin_lock_irq(&gcwq->lock); | |
3af24433 | 3198 | /* |
e22bee78 TH |
3199 | * Claim the manager position and make all workers rogue. |
3200 | * Trustee must be bound to the target cpu and can't be | |
3201 | * cancelled. | |
3af24433 | 3202 | */ |
db7bccf4 | 3203 | BUG_ON(gcwq->cpu != smp_processor_id()); |
e22bee78 TH |
3204 | rc = trustee_wait_event(!(gcwq->flags & GCWQ_MANAGING_WORKERS)); |
3205 | BUG_ON(rc < 0); | |
3af24433 | 3206 | |
e22bee78 | 3207 | gcwq->flags |= GCWQ_MANAGING_WORKERS; |
e1d8aa9f | 3208 | |
db7bccf4 | 3209 | list_for_each_entry(worker, &gcwq->idle_list, entry) |
cb444766 | 3210 | worker->flags |= WORKER_ROGUE; |
3af24433 | 3211 | |
db7bccf4 | 3212 | for_each_busy_worker(worker, i, pos, gcwq) |
cb444766 | 3213 | worker->flags |= WORKER_ROGUE; |
06ba38a9 | 3214 | |
e22bee78 TH |
3215 | /* |
3216 | * Call schedule() so that we cross rq->lock and thus can | |
3217 | * guarantee sched callbacks see the rogue flag. This is | |
3218 | * necessary as scheduler callbacks may be invoked from other | |
3219 | * cpus. | |
3220 | */ | |
3221 | spin_unlock_irq(&gcwq->lock); | |
3222 | schedule(); | |
3223 | spin_lock_irq(&gcwq->lock); | |
06ba38a9 | 3224 | |
e22bee78 | 3225 | /* |
cb444766 TH |
3226 | * Sched callbacks are disabled now. Zap nr_running. After |
3227 | * this, nr_running stays zero and need_more_worker() and | |
3228 | * keep_working() are always true as long as the worklist is | |
3229 | * not empty. | |
e22bee78 | 3230 | */ |
cb444766 | 3231 | atomic_set(get_gcwq_nr_running(gcwq->cpu), 0); |
1da177e4 | 3232 | |
e22bee78 TH |
3233 | spin_unlock_irq(&gcwq->lock); |
3234 | del_timer_sync(&gcwq->idle_timer); | |
3235 | spin_lock_irq(&gcwq->lock); | |
3af24433 | 3236 | |
db7bccf4 TH |
3237 | /* |
3238 | * We're now in charge. Notify and proceed to drain. We need | |
3239 | * to keep the gcwq running during the whole CPU down | |
3240 | * procedure as other cpu hotunplug callbacks may need to | |
3241 | * flush currently running tasks. | |
3242 | */ | |
3243 | gcwq->trustee_state = TRUSTEE_IN_CHARGE; | |
3244 | wake_up_all(&gcwq->trustee_wait); | |
3af24433 | 3245 | |
db7bccf4 TH |
3246 | /* |
3247 | * The original cpu is in the process of dying and may go away | |
3248 | * anytime now. When that happens, we and all workers would | |
e22bee78 TH |
3249 | * be migrated to other cpus. Try draining any left work. We |
3250 | * want to get it over with ASAP - spam rescuers, wake up as | |
3251 | * many idlers as necessary and create new ones till the | |
3252 | * worklist is empty. Note that if the gcwq is frozen, there | |
3253 | * may be frozen works in freezeable cwqs. Don't declare | |
3254 | * completion while frozen. | |
db7bccf4 TH |
3255 | */ |
3256 | while (gcwq->nr_workers != gcwq->nr_idle || | |
3257 | gcwq->flags & GCWQ_FREEZING || | |
3258 | gcwq->trustee_state == TRUSTEE_IN_CHARGE) { | |
e22bee78 TH |
3259 | int nr_works = 0; |
3260 | ||
3261 | list_for_each_entry(work, &gcwq->worklist, entry) { | |
3262 | send_mayday(work); | |
3263 | nr_works++; | |
3264 | } | |
3af24433 | 3265 | |
e22bee78 TH |
3266 | list_for_each_entry(worker, &gcwq->idle_list, entry) { |
3267 | if (!nr_works--) | |
3268 | break; | |
3269 | wake_up_process(worker->task); | |
3270 | } | |
3271 | ||
3272 | if (need_to_create_worker(gcwq)) { | |
3273 | spin_unlock_irq(&gcwq->lock); | |
3274 | worker = create_worker(gcwq, false); | |
3275 | spin_lock_irq(&gcwq->lock); | |
3276 | if (worker) { | |
cb444766 | 3277 | worker->flags |= WORKER_ROGUE; |
e22bee78 TH |
3278 | start_worker(worker); |
3279 | } | |
1da177e4 | 3280 | } |
3af24433 | 3281 | |
db7bccf4 TH |
3282 | /* give a breather */ |
3283 | if (trustee_wait_event_timeout(false, TRUSTEE_COOLDOWN) < 0) | |
3284 | break; | |
3af24433 | 3285 | } |
1da177e4 | 3286 | |
14441960 | 3287 | /* |
e22bee78 TH |
3288 | * Either all works have been scheduled and cpu is down, or |
3289 | * cpu down has already been canceled. Wait for and butcher | |
3290 | * all workers till we're canceled. | |
14441960 | 3291 | */ |
e22bee78 TH |
3292 | do { |
3293 | rc = trustee_wait_event(!list_empty(&gcwq->idle_list)); | |
3294 | while (!list_empty(&gcwq->idle_list)) | |
3295 | destroy_worker(list_first_entry(&gcwq->idle_list, | |
3296 | struct worker, entry)); | |
3297 | } while (gcwq->nr_workers && rc >= 0); | |
4e6045f1 | 3298 | |
14441960 | 3299 | /* |
e22bee78 TH |
3300 | * At this point, either draining has completed and no worker |
3301 | * is left, or cpu down has been canceled or the cpu is being | |
3302 | * brought back up. There shouldn't be any idle one left. | |
3303 | * Tell the remaining busy ones to rebind once it finishes the | |
3304 | * currently scheduled works by scheduling the rebind_work. | |
14441960 | 3305 | */ |
e22bee78 TH |
3306 | WARN_ON(!list_empty(&gcwq->idle_list)); |
3307 | ||
3308 | for_each_busy_worker(worker, i, pos, gcwq) { | |
3309 | struct work_struct *rebind_work = &worker->rebind_work; | |
3310 | ||
3311 | /* | |
3312 | * Rebind_work may race with future cpu hotplug | |
3313 | * operations. Use a separate flag to mark that | |
3314 | * rebinding is scheduled. | |
3315 | */ | |
cb444766 TH |
3316 | worker->flags |= WORKER_REBIND; |
3317 | worker->flags &= ~WORKER_ROGUE; | |
e22bee78 TH |
3318 | |
3319 | /* queue rebind_work, wq doesn't matter, use the default one */ | |
3320 | if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, | |
3321 | work_data_bits(rebind_work))) | |
3322 | continue; | |
3323 | ||
3324 | debug_work_activate(rebind_work); | |
d320c038 | 3325 | insert_work(get_cwq(gcwq->cpu, system_wq), rebind_work, |
e22bee78 TH |
3326 | worker->scheduled.next, |
3327 | work_color_to_flags(WORK_NO_COLOR)); | |
3328 | } | |
3329 | ||
3330 | /* relinquish manager role */ | |
3331 | gcwq->flags &= ~GCWQ_MANAGING_WORKERS; | |
3332 | ||
db7bccf4 TH |
3333 | /* notify completion */ |
3334 | gcwq->trustee = NULL; | |
3335 | gcwq->trustee_state = TRUSTEE_DONE; | |
3336 | wake_up_all(&gcwq->trustee_wait); | |
3337 | spin_unlock_irq(&gcwq->lock); | |
3338 | return 0; | |
3af24433 ON |
3339 | } |
3340 | ||
3341 | /** | |
db7bccf4 TH |
3342 | * wait_trustee_state - wait for trustee to enter the specified state |
3343 | * @gcwq: gcwq the trustee of interest belongs to | |
3344 | * @state: target state to wait for | |
3af24433 | 3345 | * |
db7bccf4 TH |
3346 | * Wait for the trustee to reach @state. DONE is already matched. |
3347 | * | |
3348 | * CONTEXT: | |
3349 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
3350 | * multiple times. To be used by cpu_callback. | |
3af24433 | 3351 | */ |
db7bccf4 | 3352 | static void __cpuinit wait_trustee_state(struct global_cwq *gcwq, int state) |
06bd6ebf NK |
3353 | __releases(&gcwq->lock) |
3354 | __acquires(&gcwq->lock) | |
3af24433 | 3355 | { |
db7bccf4 TH |
3356 | if (!(gcwq->trustee_state == state || |
3357 | gcwq->trustee_state == TRUSTEE_DONE)) { | |
3358 | spin_unlock_irq(&gcwq->lock); | |
3359 | __wait_event(gcwq->trustee_wait, | |
3360 | gcwq->trustee_state == state || | |
3361 | gcwq->trustee_state == TRUSTEE_DONE); | |
3362 | spin_lock_irq(&gcwq->lock); | |
3363 | } | |
3af24433 | 3364 | } |
3af24433 ON |
3365 | |
3366 | static int __devinit workqueue_cpu_callback(struct notifier_block *nfb, | |
3367 | unsigned long action, | |
3368 | void *hcpu) | |
3369 | { | |
3370 | unsigned int cpu = (unsigned long)hcpu; | |
db7bccf4 TH |
3371 | struct global_cwq *gcwq = get_gcwq(cpu); |
3372 | struct task_struct *new_trustee = NULL; | |
e22bee78 | 3373 | struct worker *uninitialized_var(new_worker); |
db7bccf4 | 3374 | unsigned long flags; |
3af24433 | 3375 | |
8bb78442 RW |
3376 | action &= ~CPU_TASKS_FROZEN; |
3377 | ||
3af24433 | 3378 | switch (action) { |
db7bccf4 TH |
3379 | case CPU_DOWN_PREPARE: |
3380 | new_trustee = kthread_create(trustee_thread, gcwq, | |
3381 | "workqueue_trustee/%d\n", cpu); | |
3382 | if (IS_ERR(new_trustee)) | |
3383 | return notifier_from_errno(PTR_ERR(new_trustee)); | |
3384 | kthread_bind(new_trustee, cpu); | |
e22bee78 | 3385 | /* fall through */ |
3af24433 | 3386 | case CPU_UP_PREPARE: |
e22bee78 TH |
3387 | BUG_ON(gcwq->first_idle); |
3388 | new_worker = create_worker(gcwq, false); | |
3389 | if (!new_worker) { | |
3390 | if (new_trustee) | |
3391 | kthread_stop(new_trustee); | |
3392 | return NOTIFY_BAD; | |
3af24433 | 3393 | } |
1da177e4 LT |
3394 | } |
3395 | ||
db7bccf4 TH |
3396 | /* some are called w/ irq disabled, don't disturb irq status */ |
3397 | spin_lock_irqsave(&gcwq->lock, flags); | |
3af24433 | 3398 | |
00dfcaf7 | 3399 | switch (action) { |
db7bccf4 TH |
3400 | case CPU_DOWN_PREPARE: |
3401 | /* initialize trustee and tell it to acquire the gcwq */ | |
3402 | BUG_ON(gcwq->trustee || gcwq->trustee_state != TRUSTEE_DONE); | |
3403 | gcwq->trustee = new_trustee; | |
3404 | gcwq->trustee_state = TRUSTEE_START; | |
3405 | wake_up_process(gcwq->trustee); | |
3406 | wait_trustee_state(gcwq, TRUSTEE_IN_CHARGE); | |
e22bee78 TH |
3407 | /* fall through */ |
3408 | case CPU_UP_PREPARE: | |
3409 | BUG_ON(gcwq->first_idle); | |
3410 | gcwq->first_idle = new_worker; | |
3411 | break; | |
3412 | ||
3413 | case CPU_DYING: | |
3414 | /* | |
3415 | * Before this, the trustee and all workers except for | |
3416 | * the ones which are still executing works from | |
3417 | * before the last CPU down must be on the cpu. After | |
3418 | * this, they'll all be diasporas. | |
3419 | */ | |
3420 | gcwq->flags |= GCWQ_DISASSOCIATED; | |
db7bccf4 TH |
3421 | break; |
3422 | ||
3da1c84c | 3423 | case CPU_POST_DEAD: |
db7bccf4 | 3424 | gcwq->trustee_state = TRUSTEE_BUTCHER; |
e22bee78 TH |
3425 | /* fall through */ |
3426 | case CPU_UP_CANCELED: | |
3427 | destroy_worker(gcwq->first_idle); | |
3428 | gcwq->first_idle = NULL; | |
db7bccf4 TH |
3429 | break; |
3430 | ||
3431 | case CPU_DOWN_FAILED: | |
3432 | case CPU_ONLINE: | |
e22bee78 | 3433 | gcwq->flags &= ~GCWQ_DISASSOCIATED; |
db7bccf4 TH |
3434 | if (gcwq->trustee_state != TRUSTEE_DONE) { |
3435 | gcwq->trustee_state = TRUSTEE_RELEASE; | |
3436 | wake_up_process(gcwq->trustee); | |
3437 | wait_trustee_state(gcwq, TRUSTEE_DONE); | |
3af24433 | 3438 | } |
db7bccf4 | 3439 | |
e22bee78 TH |
3440 | /* |
3441 | * Trustee is done and there might be no worker left. | |
3442 | * Put the first_idle in and request a real manager to | |
3443 | * take a look. | |
3444 | */ | |
3445 | spin_unlock_irq(&gcwq->lock); | |
3446 | kthread_bind(gcwq->first_idle->task, cpu); | |
3447 | spin_lock_irq(&gcwq->lock); | |
3448 | gcwq->flags |= GCWQ_MANAGE_WORKERS; | |
3449 | start_worker(gcwq->first_idle); | |
3450 | gcwq->first_idle = NULL; | |
db7bccf4 | 3451 | break; |
00dfcaf7 ON |
3452 | } |
3453 | ||
db7bccf4 TH |
3454 | spin_unlock_irqrestore(&gcwq->lock, flags); |
3455 | ||
1537663f | 3456 | return notifier_from_errno(0); |
1da177e4 | 3457 | } |
1da177e4 | 3458 | |
2d3854a3 | 3459 | #ifdef CONFIG_SMP |
8ccad40d | 3460 | |
2d3854a3 | 3461 | struct work_for_cpu { |
6b44003e | 3462 | struct completion completion; |
2d3854a3 RR |
3463 | long (*fn)(void *); |
3464 | void *arg; | |
3465 | long ret; | |
3466 | }; | |
3467 | ||
6b44003e | 3468 | static int do_work_for_cpu(void *_wfc) |
2d3854a3 | 3469 | { |
6b44003e | 3470 | struct work_for_cpu *wfc = _wfc; |
2d3854a3 | 3471 | wfc->ret = wfc->fn(wfc->arg); |
6b44003e AM |
3472 | complete(&wfc->completion); |
3473 | return 0; | |
2d3854a3 RR |
3474 | } |
3475 | ||
3476 | /** | |
3477 | * work_on_cpu - run a function in user context on a particular cpu | |
3478 | * @cpu: the cpu to run on | |
3479 | * @fn: the function to run | |
3480 | * @arg: the function arg | |
3481 | * | |
31ad9081 RR |
3482 | * This will return the value @fn returns. |
3483 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 3484 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 RR |
3485 | */ |
3486 | long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) | |
3487 | { | |
6b44003e AM |
3488 | struct task_struct *sub_thread; |
3489 | struct work_for_cpu wfc = { | |
3490 | .completion = COMPLETION_INITIALIZER_ONSTACK(wfc.completion), | |
3491 | .fn = fn, | |
3492 | .arg = arg, | |
3493 | }; | |
3494 | ||
3495 | sub_thread = kthread_create(do_work_for_cpu, &wfc, "work_for_cpu"); | |
3496 | if (IS_ERR(sub_thread)) | |
3497 | return PTR_ERR(sub_thread); | |
3498 | kthread_bind(sub_thread, cpu); | |
3499 | wake_up_process(sub_thread); | |
3500 | wait_for_completion(&wfc.completion); | |
2d3854a3 RR |
3501 | return wfc.ret; |
3502 | } | |
3503 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
3504 | #endif /* CONFIG_SMP */ | |
3505 | ||
a0a1a5fd TH |
3506 | #ifdef CONFIG_FREEZER |
3507 | ||
3508 | /** | |
3509 | * freeze_workqueues_begin - begin freezing workqueues | |
3510 | * | |
3511 | * Start freezing workqueues. After this function returns, all | |
3512 | * freezeable workqueues will queue new works to their frozen_works | |
7e11629d | 3513 | * list instead of gcwq->worklist. |
a0a1a5fd TH |
3514 | * |
3515 | * CONTEXT: | |
8b03ae3c | 3516 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3517 | */ |
3518 | void freeze_workqueues_begin(void) | |
3519 | { | |
a0a1a5fd TH |
3520 | unsigned int cpu; |
3521 | ||
3522 | spin_lock(&workqueue_lock); | |
3523 | ||
3524 | BUG_ON(workqueue_freezing); | |
3525 | workqueue_freezing = true; | |
3526 | ||
f3421797 | 3527 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3528 | struct global_cwq *gcwq = get_gcwq(cpu); |
bdbc5dd7 | 3529 | struct workqueue_struct *wq; |
8b03ae3c TH |
3530 | |
3531 | spin_lock_irq(&gcwq->lock); | |
3532 | ||
db7bccf4 TH |
3533 | BUG_ON(gcwq->flags & GCWQ_FREEZING); |
3534 | gcwq->flags |= GCWQ_FREEZING; | |
3535 | ||
a0a1a5fd TH |
3536 | list_for_each_entry(wq, &workqueues, list) { |
3537 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3538 | ||
f3421797 | 3539 | if (cwq && wq->flags & WQ_FREEZEABLE) |
a0a1a5fd | 3540 | cwq->max_active = 0; |
a0a1a5fd | 3541 | } |
8b03ae3c TH |
3542 | |
3543 | spin_unlock_irq(&gcwq->lock); | |
a0a1a5fd TH |
3544 | } |
3545 | ||
3546 | spin_unlock(&workqueue_lock); | |
3547 | } | |
3548 | ||
3549 | /** | |
3550 | * freeze_workqueues_busy - are freezeable workqueues still busy? | |
3551 | * | |
3552 | * Check whether freezing is complete. This function must be called | |
3553 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
3554 | * | |
3555 | * CONTEXT: | |
3556 | * Grabs and releases workqueue_lock. | |
3557 | * | |
3558 | * RETURNS: | |
3559 | * %true if some freezeable workqueues are still busy. %false if | |
3560 | * freezing is complete. | |
3561 | */ | |
3562 | bool freeze_workqueues_busy(void) | |
3563 | { | |
a0a1a5fd TH |
3564 | unsigned int cpu; |
3565 | bool busy = false; | |
3566 | ||
3567 | spin_lock(&workqueue_lock); | |
3568 | ||
3569 | BUG_ON(!workqueue_freezing); | |
3570 | ||
f3421797 | 3571 | for_each_gcwq_cpu(cpu) { |
bdbc5dd7 | 3572 | struct workqueue_struct *wq; |
a0a1a5fd TH |
3573 | /* |
3574 | * nr_active is monotonically decreasing. It's safe | |
3575 | * to peek without lock. | |
3576 | */ | |
3577 | list_for_each_entry(wq, &workqueues, list) { | |
3578 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3579 | ||
f3421797 | 3580 | if (!cwq || !(wq->flags & WQ_FREEZEABLE)) |
a0a1a5fd TH |
3581 | continue; |
3582 | ||
3583 | BUG_ON(cwq->nr_active < 0); | |
3584 | if (cwq->nr_active) { | |
3585 | busy = true; | |
3586 | goto out_unlock; | |
3587 | } | |
3588 | } | |
3589 | } | |
3590 | out_unlock: | |
3591 | spin_unlock(&workqueue_lock); | |
3592 | return busy; | |
3593 | } | |
3594 | ||
3595 | /** | |
3596 | * thaw_workqueues - thaw workqueues | |
3597 | * | |
3598 | * Thaw workqueues. Normal queueing is restored and all collected | |
7e11629d | 3599 | * frozen works are transferred to their respective gcwq worklists. |
a0a1a5fd TH |
3600 | * |
3601 | * CONTEXT: | |
8b03ae3c | 3602 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3603 | */ |
3604 | void thaw_workqueues(void) | |
3605 | { | |
a0a1a5fd TH |
3606 | unsigned int cpu; |
3607 | ||
3608 | spin_lock(&workqueue_lock); | |
3609 | ||
3610 | if (!workqueue_freezing) | |
3611 | goto out_unlock; | |
3612 | ||
f3421797 | 3613 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3614 | struct global_cwq *gcwq = get_gcwq(cpu); |
bdbc5dd7 | 3615 | struct workqueue_struct *wq; |
8b03ae3c TH |
3616 | |
3617 | spin_lock_irq(&gcwq->lock); | |
3618 | ||
db7bccf4 TH |
3619 | BUG_ON(!(gcwq->flags & GCWQ_FREEZING)); |
3620 | gcwq->flags &= ~GCWQ_FREEZING; | |
3621 | ||
a0a1a5fd TH |
3622 | list_for_each_entry(wq, &workqueues, list) { |
3623 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3624 | ||
f3421797 | 3625 | if (!cwq || !(wq->flags & WQ_FREEZEABLE)) |
a0a1a5fd TH |
3626 | continue; |
3627 | ||
a0a1a5fd TH |
3628 | /* restore max_active and repopulate worklist */ |
3629 | cwq->max_active = wq->saved_max_active; | |
3630 | ||
3631 | while (!list_empty(&cwq->delayed_works) && | |
3632 | cwq->nr_active < cwq->max_active) | |
3633 | cwq_activate_first_delayed(cwq); | |
a0a1a5fd | 3634 | } |
8b03ae3c | 3635 | |
e22bee78 TH |
3636 | wake_up_worker(gcwq); |
3637 | ||
8b03ae3c | 3638 | spin_unlock_irq(&gcwq->lock); |
a0a1a5fd TH |
3639 | } |
3640 | ||
3641 | workqueue_freezing = false; | |
3642 | out_unlock: | |
3643 | spin_unlock(&workqueue_lock); | |
3644 | } | |
3645 | #endif /* CONFIG_FREEZER */ | |
3646 | ||
6ee0578b | 3647 | static int __init init_workqueues(void) |
1da177e4 | 3648 | { |
c34056a3 | 3649 | unsigned int cpu; |
c8e55f36 | 3650 | int i; |
c34056a3 | 3651 | |
f6500947 | 3652 | cpu_notifier(workqueue_cpu_callback, CPU_PRI_WORKQUEUE); |
8b03ae3c TH |
3653 | |
3654 | /* initialize gcwqs */ | |
f3421797 | 3655 | for_each_gcwq_cpu(cpu) { |
8b03ae3c TH |
3656 | struct global_cwq *gcwq = get_gcwq(cpu); |
3657 | ||
3658 | spin_lock_init(&gcwq->lock); | |
7e11629d | 3659 | INIT_LIST_HEAD(&gcwq->worklist); |
8b03ae3c | 3660 | gcwq->cpu = cpu; |
477a3c33 | 3661 | gcwq->flags |= GCWQ_DISASSOCIATED; |
8b03ae3c | 3662 | |
c8e55f36 TH |
3663 | INIT_LIST_HEAD(&gcwq->idle_list); |
3664 | for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++) | |
3665 | INIT_HLIST_HEAD(&gcwq->busy_hash[i]); | |
3666 | ||
e22bee78 TH |
3667 | init_timer_deferrable(&gcwq->idle_timer); |
3668 | gcwq->idle_timer.function = idle_worker_timeout; | |
3669 | gcwq->idle_timer.data = (unsigned long)gcwq; | |
e7577c50 | 3670 | |
e22bee78 TH |
3671 | setup_timer(&gcwq->mayday_timer, gcwq_mayday_timeout, |
3672 | (unsigned long)gcwq); | |
3673 | ||
8b03ae3c | 3674 | ida_init(&gcwq->worker_ida); |
db7bccf4 TH |
3675 | |
3676 | gcwq->trustee_state = TRUSTEE_DONE; | |
3677 | init_waitqueue_head(&gcwq->trustee_wait); | |
8b03ae3c TH |
3678 | } |
3679 | ||
e22bee78 | 3680 | /* create the initial worker */ |
f3421797 | 3681 | for_each_online_gcwq_cpu(cpu) { |
e22bee78 TH |
3682 | struct global_cwq *gcwq = get_gcwq(cpu); |
3683 | struct worker *worker; | |
3684 | ||
477a3c33 TH |
3685 | if (cpu != WORK_CPU_UNBOUND) |
3686 | gcwq->flags &= ~GCWQ_DISASSOCIATED; | |
e22bee78 TH |
3687 | worker = create_worker(gcwq, true); |
3688 | BUG_ON(!worker); | |
3689 | spin_lock_irq(&gcwq->lock); | |
3690 | start_worker(worker); | |
3691 | spin_unlock_irq(&gcwq->lock); | |
3692 | } | |
3693 | ||
d320c038 TH |
3694 | system_wq = alloc_workqueue("events", 0, 0); |
3695 | system_long_wq = alloc_workqueue("events_long", 0, 0); | |
3696 | system_nrt_wq = alloc_workqueue("events_nrt", WQ_NON_REENTRANT, 0); | |
f3421797 TH |
3697 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
3698 | WQ_UNBOUND_MAX_ACTIVE); | |
d320c038 | 3699 | BUG_ON(!system_wq || !system_long_wq || !system_nrt_wq); |
6ee0578b | 3700 | return 0; |
1da177e4 | 3701 | } |
6ee0578b | 3702 | early_initcall(init_workqueues); |