]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
c54fce6e | 2 | * kernel/workqueue.c - generic async execution with shared worker pool |
1da177e4 | 3 | * |
c54fce6e | 4 | * Copyright (C) 2002 Ingo Molnar |
1da177e4 | 5 | * |
c54fce6e TH |
6 | * Derived from the taskqueue/keventd code by: |
7 | * David Woodhouse <dwmw2@infradead.org> | |
8 | * Andrew Morton | |
9 | * Kai Petzke <wpp@marie.physik.tu-berlin.de> | |
10 | * Theodore Ts'o <tytso@mit.edu> | |
1da177e4 | 11 | * |
c54fce6e | 12 | * Made to use alloc_percpu by Christoph Lameter. |
1da177e4 | 13 | * |
c54fce6e TH |
14 | * Copyright (C) 2010 SUSE Linux Products GmbH |
15 | * Copyright (C) 2010 Tejun Heo <tj@kernel.org> | |
89ada679 | 16 | * |
c54fce6e TH |
17 | * This is the generic async execution mechanism. Work items as are |
18 | * executed in process context. The worker pool is shared and | |
19 | * automatically managed. There is one worker pool for each CPU and | |
20 | * one extra for works which are better served by workers which are | |
21 | * not bound to any specific CPU. | |
22 | * | |
23 | * Please read Documentation/workqueue.txt for details. | |
1da177e4 LT |
24 | */ |
25 | ||
9984de1a | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/kernel.h> |
28 | #include <linux/sched.h> | |
29 | #include <linux/init.h> | |
30 | #include <linux/signal.h> | |
31 | #include <linux/completion.h> | |
32 | #include <linux/workqueue.h> | |
33 | #include <linux/slab.h> | |
34 | #include <linux/cpu.h> | |
35 | #include <linux/notifier.h> | |
36 | #include <linux/kthread.h> | |
1fa44eca | 37 | #include <linux/hardirq.h> |
46934023 | 38 | #include <linux/mempolicy.h> |
341a5958 | 39 | #include <linux/freezer.h> |
d5abe669 PZ |
40 | #include <linux/kallsyms.h> |
41 | #include <linux/debug_locks.h> | |
4e6045f1 | 42 | #include <linux/lockdep.h> |
c34056a3 | 43 | #include <linux/idr.h> |
42f8570f | 44 | #include <linux/hashtable.h> |
e22bee78 | 45 | |
ea138446 | 46 | #include "workqueue_internal.h" |
1da177e4 | 47 | |
c8e55f36 | 48 | enum { |
24647570 TH |
49 | /* |
50 | * worker_pool flags | |
bc2ae0f5 | 51 | * |
24647570 | 52 | * A bound pool is either associated or disassociated with its CPU. |
bc2ae0f5 TH |
53 | * While associated (!DISASSOCIATED), all workers are bound to the |
54 | * CPU and none has %WORKER_UNBOUND set and concurrency management | |
55 | * is in effect. | |
56 | * | |
57 | * While DISASSOCIATED, the cpu may be offline and all workers have | |
58 | * %WORKER_UNBOUND set and concurrency management disabled, and may | |
24647570 | 59 | * be executing on any CPU. The pool behaves as an unbound one. |
bc2ae0f5 TH |
60 | * |
61 | * Note that DISASSOCIATED can be flipped only while holding | |
24647570 TH |
62 | * assoc_mutex to avoid changing binding state while |
63 | * create_worker() is in progress. | |
bc2ae0f5 | 64 | */ |
11ebea50 | 65 | POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */ |
552a37e9 | 66 | POOL_MANAGING_WORKERS = 1 << 1, /* managing workers */ |
24647570 | 67 | POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ |
35b6bb63 | 68 | POOL_FREEZING = 1 << 3, /* freeze in progress */ |
db7bccf4 | 69 | |
c8e55f36 TH |
70 | /* worker flags */ |
71 | WORKER_STARTED = 1 << 0, /* started */ | |
72 | WORKER_DIE = 1 << 1, /* die die die */ | |
73 | WORKER_IDLE = 1 << 2, /* is idle */ | |
e22bee78 | 74 | WORKER_PREP = 1 << 3, /* preparing to run works */ |
fb0e7beb | 75 | WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ |
f3421797 | 76 | WORKER_UNBOUND = 1 << 7, /* worker is unbound */ |
e22bee78 | 77 | |
5f7dabfd | 78 | WORKER_NOT_RUNNING = WORKER_PREP | WORKER_UNBOUND | |
403c821d | 79 | WORKER_CPU_INTENSIVE, |
db7bccf4 | 80 | |
e34cdddb | 81 | NR_STD_WORKER_POOLS = 2, /* # standard pools per cpu */ |
4ce62e9e | 82 | |
c8e55f36 | 83 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ |
db7bccf4 | 84 | |
e22bee78 TH |
85 | MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */ |
86 | IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */ | |
87 | ||
3233cdbd TH |
88 | MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2, |
89 | /* call for help after 10ms | |
90 | (min two ticks) */ | |
e22bee78 TH |
91 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ |
92 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
e22bee78 TH |
93 | |
94 | /* | |
95 | * Rescue workers are used only on emergencies and shared by | |
96 | * all cpus. Give -20. | |
97 | */ | |
98 | RESCUER_NICE_LEVEL = -20, | |
3270476a | 99 | HIGHPRI_NICE_LEVEL = -20, |
c8e55f36 | 100 | }; |
1da177e4 LT |
101 | |
102 | /* | |
4690c4ab TH |
103 | * Structure fields follow one of the following exclusion rules. |
104 | * | |
e41e704b TH |
105 | * I: Modifiable by initialization/destruction paths and read-only for |
106 | * everyone else. | |
4690c4ab | 107 | * |
e22bee78 TH |
108 | * P: Preemption protected. Disabling preemption is enough and should |
109 | * only be modified and accessed from the local cpu. | |
110 | * | |
d565ed63 | 111 | * L: pool->lock protected. Access with pool->lock held. |
4690c4ab | 112 | * |
d565ed63 TH |
113 | * X: During normal operation, modification requires pool->lock and should |
114 | * be done only from local cpu. Either disabling preemption on local | |
115 | * cpu or grabbing pool->lock is enough for read access. If | |
116 | * POOL_DISASSOCIATED is set, it's identical to L. | |
e22bee78 | 117 | * |
73f53c4a TH |
118 | * F: wq->flush_mutex protected. |
119 | * | |
4690c4ab | 120 | * W: workqueue_lock protected. |
1da177e4 | 121 | */ |
1da177e4 | 122 | |
2eaebdb3 | 123 | /* struct worker is defined in workqueue_internal.h */ |
c34056a3 | 124 | |
bd7bdd43 | 125 | struct worker_pool { |
d565ed63 | 126 | spinlock_t lock; /* the pool lock */ |
ec22ca5e | 127 | unsigned int cpu; /* I: the associated cpu */ |
9daf9e67 | 128 | int id; /* I: pool ID */ |
11ebea50 | 129 | unsigned int flags; /* X: flags */ |
bd7bdd43 TH |
130 | |
131 | struct list_head worklist; /* L: list of pending works */ | |
132 | int nr_workers; /* L: total number of workers */ | |
ea1abd61 LJ |
133 | |
134 | /* nr_idle includes the ones off idle_list for rebinding */ | |
bd7bdd43 TH |
135 | int nr_idle; /* L: currently idle ones */ |
136 | ||
137 | struct list_head idle_list; /* X: list of idle workers */ | |
138 | struct timer_list idle_timer; /* L: worker idle timeout */ | |
139 | struct timer_list mayday_timer; /* L: SOS timer for workers */ | |
140 | ||
c9e7cf27 TH |
141 | /* workers are chained either in busy_hash or idle_list */ |
142 | DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER); | |
143 | /* L: hash of busy workers */ | |
144 | ||
24647570 | 145 | struct mutex assoc_mutex; /* protect POOL_DISASSOCIATED */ |
bd7bdd43 | 146 | struct ida worker_ida; /* L: for worker IDs */ |
8b03ae3c TH |
147 | } ____cacheline_aligned_in_smp; |
148 | ||
1da177e4 | 149 | /* |
502ca9d8 | 150 | * The per-CPU workqueue. The lower WORK_STRUCT_FLAG_BITS of |
0f900049 TH |
151 | * work_struct->data are used for flags and thus cwqs need to be |
152 | * aligned at two's power of the number of flag bits. | |
1da177e4 LT |
153 | */ |
154 | struct cpu_workqueue_struct { | |
bd7bdd43 | 155 | struct worker_pool *pool; /* I: the associated pool */ |
4690c4ab | 156 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
157 | int work_color; /* L: current color */ |
158 | int flush_color; /* L: flushing color */ | |
159 | int nr_in_flight[WORK_NR_COLORS]; | |
160 | /* L: nr of in_flight works */ | |
1e19ffc6 | 161 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 162 | int max_active; /* L: max active works */ |
1e19ffc6 | 163 | struct list_head delayed_works; /* L: delayed works */ |
0f900049 | 164 | }; |
1da177e4 | 165 | |
73f53c4a TH |
166 | /* |
167 | * Structure used to wait for workqueue flush. | |
168 | */ | |
169 | struct wq_flusher { | |
170 | struct list_head list; /* F: list of flushers */ | |
171 | int flush_color; /* F: flush color waiting for */ | |
172 | struct completion done; /* flush completion */ | |
173 | }; | |
174 | ||
f2e005aa TH |
175 | /* |
176 | * All cpumasks are assumed to be always set on UP and thus can't be | |
177 | * used to determine whether there's something to be done. | |
178 | */ | |
179 | #ifdef CONFIG_SMP | |
180 | typedef cpumask_var_t mayday_mask_t; | |
181 | #define mayday_test_and_set_cpu(cpu, mask) \ | |
182 | cpumask_test_and_set_cpu((cpu), (mask)) | |
183 | #define mayday_clear_cpu(cpu, mask) cpumask_clear_cpu((cpu), (mask)) | |
184 | #define for_each_mayday_cpu(cpu, mask) for_each_cpu((cpu), (mask)) | |
9c37547a | 185 | #define alloc_mayday_mask(maskp, gfp) zalloc_cpumask_var((maskp), (gfp)) |
f2e005aa TH |
186 | #define free_mayday_mask(mask) free_cpumask_var((mask)) |
187 | #else | |
188 | typedef unsigned long mayday_mask_t; | |
189 | #define mayday_test_and_set_cpu(cpu, mask) test_and_set_bit(0, &(mask)) | |
190 | #define mayday_clear_cpu(cpu, mask) clear_bit(0, &(mask)) | |
191 | #define for_each_mayday_cpu(cpu, mask) if ((cpu) = 0, (mask)) | |
192 | #define alloc_mayday_mask(maskp, gfp) true | |
193 | #define free_mayday_mask(mask) do { } while (0) | |
194 | #endif | |
1da177e4 LT |
195 | |
196 | /* | |
197 | * The externally visible workqueue abstraction is an array of | |
198 | * per-CPU workqueues: | |
199 | */ | |
200 | struct workqueue_struct { | |
9c5a2ba7 | 201 | unsigned int flags; /* W: WQ_* flags */ |
bdbc5dd7 TH |
202 | union { |
203 | struct cpu_workqueue_struct __percpu *pcpu; | |
204 | struct cpu_workqueue_struct *single; | |
205 | unsigned long v; | |
206 | } cpu_wq; /* I: cwq's */ | |
4690c4ab | 207 | struct list_head list; /* W: list of all workqueues */ |
73f53c4a TH |
208 | |
209 | struct mutex flush_mutex; /* protects wq flushing */ | |
210 | int work_color; /* F: current work color */ | |
211 | int flush_color; /* F: current flush color */ | |
212 | atomic_t nr_cwqs_to_flush; /* flush in progress */ | |
213 | struct wq_flusher *first_flusher; /* F: first flusher */ | |
214 | struct list_head flusher_queue; /* F: flush waiters */ | |
215 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
216 | ||
f2e005aa | 217 | mayday_mask_t mayday_mask; /* cpus requesting rescue */ |
e22bee78 TH |
218 | struct worker *rescuer; /* I: rescue worker */ |
219 | ||
9c5a2ba7 | 220 | int nr_drainers; /* W: drain in progress */ |
dcd989cb | 221 | int saved_max_active; /* W: saved cwq max_active */ |
4e6045f1 | 222 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 223 | struct lockdep_map lockdep_map; |
4e6045f1 | 224 | #endif |
b196be89 | 225 | char name[]; /* I: workqueue name */ |
1da177e4 LT |
226 | }; |
227 | ||
d320c038 | 228 | struct workqueue_struct *system_wq __read_mostly; |
d320c038 | 229 | EXPORT_SYMBOL_GPL(system_wq); |
044c782c | 230 | struct workqueue_struct *system_highpri_wq __read_mostly; |
1aabe902 | 231 | EXPORT_SYMBOL_GPL(system_highpri_wq); |
044c782c | 232 | struct workqueue_struct *system_long_wq __read_mostly; |
d320c038 | 233 | EXPORT_SYMBOL_GPL(system_long_wq); |
044c782c | 234 | struct workqueue_struct *system_unbound_wq __read_mostly; |
f3421797 | 235 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
044c782c | 236 | struct workqueue_struct *system_freezable_wq __read_mostly; |
24d51add | 237 | EXPORT_SYMBOL_GPL(system_freezable_wq); |
d320c038 | 238 | |
97bd2347 TH |
239 | #define CREATE_TRACE_POINTS |
240 | #include <trace/events/workqueue.h> | |
241 | ||
38db41d9 | 242 | #define for_each_std_worker_pool(pool, cpu) \ |
a60dc39c TH |
243 | for ((pool) = &std_worker_pools(cpu)[0]; \ |
244 | (pool) < &std_worker_pools(cpu)[NR_STD_WORKER_POOLS]; (pool)++) | |
4ce62e9e | 245 | |
c9e7cf27 TH |
246 | #define for_each_busy_worker(worker, i, pos, pool) \ |
247 | hash_for_each(pool->busy_hash, i, pos, worker, hentry) | |
db7bccf4 | 248 | |
706026c2 TH |
249 | static inline int __next_wq_cpu(int cpu, const struct cpumask *mask, |
250 | unsigned int sw) | |
f3421797 TH |
251 | { |
252 | if (cpu < nr_cpu_ids) { | |
253 | if (sw & 1) { | |
254 | cpu = cpumask_next(cpu, mask); | |
255 | if (cpu < nr_cpu_ids) | |
256 | return cpu; | |
257 | } | |
258 | if (sw & 2) | |
259 | return WORK_CPU_UNBOUND; | |
260 | } | |
261 | return WORK_CPU_NONE; | |
262 | } | |
263 | ||
706026c2 TH |
264 | static inline int __next_cwq_cpu(int cpu, const struct cpumask *mask, |
265 | struct workqueue_struct *wq) | |
f3421797 | 266 | { |
706026c2 | 267 | return __next_wq_cpu(cpu, mask, !(wq->flags & WQ_UNBOUND) ? 1 : 2); |
f3421797 TH |
268 | } |
269 | ||
09884951 TH |
270 | /* |
271 | * CPU iterators | |
272 | * | |
706026c2 | 273 | * An extra cpu number is defined using an invalid cpu number |
09884951 | 274 | * (WORK_CPU_UNBOUND) to host workqueues which are not bound to any |
706026c2 TH |
275 | * specific CPU. The following iterators are similar to for_each_*_cpu() |
276 | * iterators but also considers the unbound CPU. | |
09884951 | 277 | * |
706026c2 TH |
278 | * for_each_wq_cpu() : possible CPUs + WORK_CPU_UNBOUND |
279 | * for_each_online_wq_cpu() : online CPUs + WORK_CPU_UNBOUND | |
09884951 TH |
280 | * for_each_cwq_cpu() : possible CPUs for bound workqueues, |
281 | * WORK_CPU_UNBOUND for unbound workqueues | |
282 | */ | |
706026c2 TH |
283 | #define for_each_wq_cpu(cpu) \ |
284 | for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, 3); \ | |
f3421797 | 285 | (cpu) < WORK_CPU_NONE; \ |
706026c2 | 286 | (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, 3)) |
f3421797 | 287 | |
706026c2 TH |
288 | #define for_each_online_wq_cpu(cpu) \ |
289 | for ((cpu) = __next_wq_cpu(-1, cpu_online_mask, 3); \ | |
f3421797 | 290 | (cpu) < WORK_CPU_NONE; \ |
706026c2 | 291 | (cpu) = __next_wq_cpu((cpu), cpu_online_mask, 3)) |
f3421797 TH |
292 | |
293 | #define for_each_cwq_cpu(cpu, wq) \ | |
706026c2 | 294 | for ((cpu) = __next_cwq_cpu(-1, cpu_possible_mask, (wq)); \ |
f3421797 | 295 | (cpu) < WORK_CPU_NONE; \ |
706026c2 | 296 | (cpu) = __next_cwq_cpu((cpu), cpu_possible_mask, (wq))) |
f3421797 | 297 | |
dc186ad7 TG |
298 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
299 | ||
300 | static struct debug_obj_descr work_debug_descr; | |
301 | ||
99777288 SG |
302 | static void *work_debug_hint(void *addr) |
303 | { | |
304 | return ((struct work_struct *) addr)->func; | |
305 | } | |
306 | ||
dc186ad7 TG |
307 | /* |
308 | * fixup_init is called when: | |
309 | * - an active object is initialized | |
310 | */ | |
311 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
312 | { | |
313 | struct work_struct *work = addr; | |
314 | ||
315 | switch (state) { | |
316 | case ODEBUG_STATE_ACTIVE: | |
317 | cancel_work_sync(work); | |
318 | debug_object_init(work, &work_debug_descr); | |
319 | return 1; | |
320 | default: | |
321 | return 0; | |
322 | } | |
323 | } | |
324 | ||
325 | /* | |
326 | * fixup_activate is called when: | |
327 | * - an active object is activated | |
328 | * - an unknown object is activated (might be a statically initialized object) | |
329 | */ | |
330 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
331 | { | |
332 | struct work_struct *work = addr; | |
333 | ||
334 | switch (state) { | |
335 | ||
336 | case ODEBUG_STATE_NOTAVAILABLE: | |
337 | /* | |
338 | * This is not really a fixup. The work struct was | |
339 | * statically initialized. We just make sure that it | |
340 | * is tracked in the object tracker. | |
341 | */ | |
22df02bb | 342 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
343 | debug_object_init(work, &work_debug_descr); |
344 | debug_object_activate(work, &work_debug_descr); | |
345 | return 0; | |
346 | } | |
347 | WARN_ON_ONCE(1); | |
348 | return 0; | |
349 | ||
350 | case ODEBUG_STATE_ACTIVE: | |
351 | WARN_ON(1); | |
352 | ||
353 | default: | |
354 | return 0; | |
355 | } | |
356 | } | |
357 | ||
358 | /* | |
359 | * fixup_free is called when: | |
360 | * - an active object is freed | |
361 | */ | |
362 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
363 | { | |
364 | struct work_struct *work = addr; | |
365 | ||
366 | switch (state) { | |
367 | case ODEBUG_STATE_ACTIVE: | |
368 | cancel_work_sync(work); | |
369 | debug_object_free(work, &work_debug_descr); | |
370 | return 1; | |
371 | default: | |
372 | return 0; | |
373 | } | |
374 | } | |
375 | ||
376 | static struct debug_obj_descr work_debug_descr = { | |
377 | .name = "work_struct", | |
99777288 | 378 | .debug_hint = work_debug_hint, |
dc186ad7 TG |
379 | .fixup_init = work_fixup_init, |
380 | .fixup_activate = work_fixup_activate, | |
381 | .fixup_free = work_fixup_free, | |
382 | }; | |
383 | ||
384 | static inline void debug_work_activate(struct work_struct *work) | |
385 | { | |
386 | debug_object_activate(work, &work_debug_descr); | |
387 | } | |
388 | ||
389 | static inline void debug_work_deactivate(struct work_struct *work) | |
390 | { | |
391 | debug_object_deactivate(work, &work_debug_descr); | |
392 | } | |
393 | ||
394 | void __init_work(struct work_struct *work, int onstack) | |
395 | { | |
396 | if (onstack) | |
397 | debug_object_init_on_stack(work, &work_debug_descr); | |
398 | else | |
399 | debug_object_init(work, &work_debug_descr); | |
400 | } | |
401 | EXPORT_SYMBOL_GPL(__init_work); | |
402 | ||
403 | void destroy_work_on_stack(struct work_struct *work) | |
404 | { | |
405 | debug_object_free(work, &work_debug_descr); | |
406 | } | |
407 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
408 | ||
409 | #else | |
410 | static inline void debug_work_activate(struct work_struct *work) { } | |
411 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
412 | #endif | |
413 | ||
95402b38 GS |
414 | /* Serializes the accesses to the list of workqueues. */ |
415 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 416 | static LIST_HEAD(workqueues); |
a0a1a5fd | 417 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 418 | |
e22bee78 | 419 | /* |
a60dc39c TH |
420 | * The CPU standard worker pools. nr_running is the only field which is |
421 | * expected to be used frequently by other cpus via try_to_wake_up(). Put | |
422 | * it in a separate cacheline. | |
e22bee78 | 423 | */ |
a60dc39c TH |
424 | static DEFINE_PER_CPU(struct worker_pool [NR_STD_WORKER_POOLS], |
425 | cpu_std_worker_pools); | |
e6e380ed TH |
426 | static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t [NR_STD_WORKER_POOLS], |
427 | cpu_std_pool_nr_running); | |
8b03ae3c | 428 | |
f3421797 | 429 | /* |
a60dc39c TH |
430 | * Standard worker pools and nr_running counter for unbound CPU. The pools |
431 | * have POOL_DISASSOCIATED set, and all workers have WORKER_UNBOUND set. | |
f3421797 | 432 | */ |
a60dc39c | 433 | static struct worker_pool unbound_std_worker_pools[NR_STD_WORKER_POOLS]; |
e6e380ed | 434 | static atomic_t unbound_std_pool_nr_running[NR_STD_WORKER_POOLS] = { |
e34cdddb | 435 | [0 ... NR_STD_WORKER_POOLS - 1] = ATOMIC_INIT(0), /* always 0 */ |
4ce62e9e | 436 | }; |
f3421797 | 437 | |
9daf9e67 TH |
438 | /* idr of all pools */ |
439 | static DEFINE_MUTEX(worker_pool_idr_mutex); | |
440 | static DEFINE_IDR(worker_pool_idr); | |
441 | ||
c34056a3 | 442 | static int worker_thread(void *__worker); |
1da177e4 | 443 | |
a60dc39c | 444 | static struct worker_pool *std_worker_pools(int cpu) |
8b03ae3c | 445 | { |
f3421797 | 446 | if (cpu != WORK_CPU_UNBOUND) |
a60dc39c | 447 | return per_cpu(cpu_std_worker_pools, cpu); |
f3421797 | 448 | else |
a60dc39c | 449 | return unbound_std_worker_pools; |
8b03ae3c TH |
450 | } |
451 | ||
4e8f0a60 TH |
452 | static int std_worker_pool_pri(struct worker_pool *pool) |
453 | { | |
a60dc39c | 454 | return pool - std_worker_pools(pool->cpu); |
4e8f0a60 TH |
455 | } |
456 | ||
9daf9e67 TH |
457 | /* allocate ID and assign it to @pool */ |
458 | static int worker_pool_assign_id(struct worker_pool *pool) | |
459 | { | |
460 | int ret; | |
461 | ||
462 | mutex_lock(&worker_pool_idr_mutex); | |
463 | idr_pre_get(&worker_pool_idr, GFP_KERNEL); | |
464 | ret = idr_get_new(&worker_pool_idr, pool, &pool->id); | |
465 | mutex_unlock(&worker_pool_idr_mutex); | |
466 | ||
467 | return ret; | |
468 | } | |
469 | ||
7c3eed5c TH |
470 | /* |
471 | * Lookup worker_pool by id. The idr currently is built during boot and | |
472 | * never modified. Don't worry about locking for now. | |
473 | */ | |
474 | static struct worker_pool *worker_pool_by_id(int pool_id) | |
475 | { | |
476 | return idr_find(&worker_pool_idr, pool_id); | |
477 | } | |
478 | ||
d565ed63 TH |
479 | static struct worker_pool *get_std_worker_pool(int cpu, bool highpri) |
480 | { | |
a60dc39c | 481 | struct worker_pool *pools = std_worker_pools(cpu); |
d565ed63 | 482 | |
a60dc39c | 483 | return &pools[highpri]; |
d565ed63 TH |
484 | } |
485 | ||
63d95a91 | 486 | static atomic_t *get_pool_nr_running(struct worker_pool *pool) |
e22bee78 | 487 | { |
ec22ca5e | 488 | int cpu = pool->cpu; |
e34cdddb | 489 | int idx = std_worker_pool_pri(pool); |
63d95a91 | 490 | |
f3421797 | 491 | if (cpu != WORK_CPU_UNBOUND) |
e6e380ed | 492 | return &per_cpu(cpu_std_pool_nr_running, cpu)[idx]; |
f3421797 | 493 | else |
e6e380ed | 494 | return &unbound_std_pool_nr_running[idx]; |
e22bee78 TH |
495 | } |
496 | ||
1537663f TH |
497 | static struct cpu_workqueue_struct *get_cwq(unsigned int cpu, |
498 | struct workqueue_struct *wq) | |
b1f4ec17 | 499 | { |
f3421797 | 500 | if (!(wq->flags & WQ_UNBOUND)) { |
e06ffa1e | 501 | if (likely(cpu < nr_cpu_ids)) |
f3421797 | 502 | return per_cpu_ptr(wq->cpu_wq.pcpu, cpu); |
f3421797 TH |
503 | } else if (likely(cpu == WORK_CPU_UNBOUND)) |
504 | return wq->cpu_wq.single; | |
505 | return NULL; | |
b1f4ec17 ON |
506 | } |
507 | ||
73f53c4a TH |
508 | static unsigned int work_color_to_flags(int color) |
509 | { | |
510 | return color << WORK_STRUCT_COLOR_SHIFT; | |
511 | } | |
512 | ||
513 | static int get_work_color(struct work_struct *work) | |
514 | { | |
515 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
516 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
517 | } | |
518 | ||
519 | static int work_next_color(int color) | |
520 | { | |
521 | return (color + 1) % WORK_NR_COLORS; | |
522 | } | |
1da177e4 | 523 | |
14441960 | 524 | /* |
b5490077 TH |
525 | * While queued, %WORK_STRUCT_CWQ is set and non flag bits of a work's data |
526 | * contain the pointer to the queued cwq. Once execution starts, the flag | |
7c3eed5c | 527 | * is cleared and the high bits contain OFFQ flags and pool ID. |
7a22ad75 | 528 | * |
7c3eed5c TH |
529 | * set_work_cwq(), set_work_pool_and_clear_pending(), mark_work_canceling() |
530 | * and clear_work_data() can be used to set the cwq, pool or clear | |
bbb68dfa TH |
531 | * work->data. These functions should only be called while the work is |
532 | * owned - ie. while the PENDING bit is set. | |
7a22ad75 | 533 | * |
7c3eed5c TH |
534 | * get_work_pool() and get_work_cwq() can be used to obtain the pool or cwq |
535 | * corresponding to a work. Pool is available once the work has been | |
536 | * queued anywhere after initialization until it is sync canceled. cwq is | |
537 | * available only while the work item is queued. | |
7a22ad75 | 538 | * |
bbb68dfa TH |
539 | * %WORK_OFFQ_CANCELING is used to mark a work item which is being |
540 | * canceled. While being canceled, a work item may have its PENDING set | |
541 | * but stay off timer and worklist for arbitrarily long and nobody should | |
542 | * try to steal the PENDING bit. | |
14441960 | 543 | */ |
7a22ad75 TH |
544 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
545 | unsigned long flags) | |
365970a1 | 546 | { |
4594bf15 | 547 | BUG_ON(!work_pending(work)); |
7a22ad75 TH |
548 | atomic_long_set(&work->data, data | flags | work_static(work)); |
549 | } | |
365970a1 | 550 | |
7a22ad75 TH |
551 | static void set_work_cwq(struct work_struct *work, |
552 | struct cpu_workqueue_struct *cwq, | |
553 | unsigned long extra_flags) | |
554 | { | |
555 | set_work_data(work, (unsigned long)cwq, | |
e120153d | 556 | WORK_STRUCT_PENDING | WORK_STRUCT_CWQ | extra_flags); |
365970a1 DH |
557 | } |
558 | ||
7c3eed5c TH |
559 | static void set_work_pool_and_clear_pending(struct work_struct *work, |
560 | int pool_id) | |
7a22ad75 | 561 | { |
23657bb1 TH |
562 | /* |
563 | * The following wmb is paired with the implied mb in | |
564 | * test_and_set_bit(PENDING) and ensures all updates to @work made | |
565 | * here are visible to and precede any updates by the next PENDING | |
566 | * owner. | |
567 | */ | |
568 | smp_wmb(); | |
7c3eed5c | 569 | set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0); |
7a22ad75 | 570 | } |
f756d5e2 | 571 | |
7a22ad75 | 572 | static void clear_work_data(struct work_struct *work) |
1da177e4 | 573 | { |
7c3eed5c TH |
574 | smp_wmb(); /* see set_work_pool_and_clear_pending() */ |
575 | set_work_data(work, WORK_STRUCT_NO_POOL, 0); | |
1da177e4 LT |
576 | } |
577 | ||
7a22ad75 | 578 | static struct cpu_workqueue_struct *get_work_cwq(struct work_struct *work) |
b1f4ec17 | 579 | { |
e120153d | 580 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 581 | |
e120153d TH |
582 | if (data & WORK_STRUCT_CWQ) |
583 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); | |
584 | else | |
585 | return NULL; | |
4d707b9f ON |
586 | } |
587 | ||
7c3eed5c TH |
588 | /** |
589 | * get_work_pool - return the worker_pool a given work was associated with | |
590 | * @work: the work item of interest | |
591 | * | |
592 | * Return the worker_pool @work was last associated with. %NULL if none. | |
593 | */ | |
594 | static struct worker_pool *get_work_pool(struct work_struct *work) | |
365970a1 | 595 | { |
e120153d | 596 | unsigned long data = atomic_long_read(&work->data); |
7c3eed5c TH |
597 | struct worker_pool *pool; |
598 | int pool_id; | |
7a22ad75 | 599 | |
e120153d TH |
600 | if (data & WORK_STRUCT_CWQ) |
601 | return ((struct cpu_workqueue_struct *) | |
7c3eed5c | 602 | (data & WORK_STRUCT_WQ_DATA_MASK))->pool; |
7a22ad75 | 603 | |
7c3eed5c TH |
604 | pool_id = data >> WORK_OFFQ_POOL_SHIFT; |
605 | if (pool_id == WORK_OFFQ_POOL_NONE) | |
7a22ad75 TH |
606 | return NULL; |
607 | ||
7c3eed5c TH |
608 | pool = worker_pool_by_id(pool_id); |
609 | WARN_ON_ONCE(!pool); | |
610 | return pool; | |
611 | } | |
612 | ||
613 | /** | |
614 | * get_work_pool_id - return the worker pool ID a given work is associated with | |
615 | * @work: the work item of interest | |
616 | * | |
617 | * Return the worker_pool ID @work was last associated with. | |
618 | * %WORK_OFFQ_POOL_NONE if none. | |
619 | */ | |
620 | static int get_work_pool_id(struct work_struct *work) | |
621 | { | |
622 | struct worker_pool *pool = get_work_pool(work); | |
623 | ||
624 | return pool ? pool->id : WORK_OFFQ_POOL_NONE; | |
625 | } | |
626 | ||
bbb68dfa TH |
627 | static void mark_work_canceling(struct work_struct *work) |
628 | { | |
7c3eed5c | 629 | unsigned long pool_id = get_work_pool_id(work); |
bbb68dfa | 630 | |
7c3eed5c TH |
631 | pool_id <<= WORK_OFFQ_POOL_SHIFT; |
632 | set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING); | |
bbb68dfa TH |
633 | } |
634 | ||
635 | static bool work_is_canceling(struct work_struct *work) | |
636 | { | |
637 | unsigned long data = atomic_long_read(&work->data); | |
638 | ||
639 | return !(data & WORK_STRUCT_CWQ) && (data & WORK_OFFQ_CANCELING); | |
640 | } | |
641 | ||
e22bee78 | 642 | /* |
3270476a TH |
643 | * Policy functions. These define the policies on how the global worker |
644 | * pools are managed. Unless noted otherwise, these functions assume that | |
d565ed63 | 645 | * they're being called with pool->lock held. |
e22bee78 TH |
646 | */ |
647 | ||
63d95a91 | 648 | static bool __need_more_worker(struct worker_pool *pool) |
a848e3b6 | 649 | { |
3270476a | 650 | return !atomic_read(get_pool_nr_running(pool)); |
a848e3b6 ON |
651 | } |
652 | ||
4594bf15 | 653 | /* |
e22bee78 TH |
654 | * Need to wake up a worker? Called from anything but currently |
655 | * running workers. | |
974271c4 TH |
656 | * |
657 | * Note that, because unbound workers never contribute to nr_running, this | |
706026c2 | 658 | * function will always return %true for unbound pools as long as the |
974271c4 | 659 | * worklist isn't empty. |
4594bf15 | 660 | */ |
63d95a91 | 661 | static bool need_more_worker(struct worker_pool *pool) |
365970a1 | 662 | { |
63d95a91 | 663 | return !list_empty(&pool->worklist) && __need_more_worker(pool); |
e22bee78 | 664 | } |
4594bf15 | 665 | |
e22bee78 | 666 | /* Can I start working? Called from busy but !running workers. */ |
63d95a91 | 667 | static bool may_start_working(struct worker_pool *pool) |
e22bee78 | 668 | { |
63d95a91 | 669 | return pool->nr_idle; |
e22bee78 TH |
670 | } |
671 | ||
672 | /* Do I need to keep working? Called from currently running workers. */ | |
63d95a91 | 673 | static bool keep_working(struct worker_pool *pool) |
e22bee78 | 674 | { |
63d95a91 | 675 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 | 676 | |
3270476a | 677 | return !list_empty(&pool->worklist) && atomic_read(nr_running) <= 1; |
e22bee78 TH |
678 | } |
679 | ||
680 | /* Do we need a new worker? Called from manager. */ | |
63d95a91 | 681 | static bool need_to_create_worker(struct worker_pool *pool) |
e22bee78 | 682 | { |
63d95a91 | 683 | return need_more_worker(pool) && !may_start_working(pool); |
e22bee78 | 684 | } |
365970a1 | 685 | |
e22bee78 | 686 | /* Do I need to be the manager? */ |
63d95a91 | 687 | static bool need_to_manage_workers(struct worker_pool *pool) |
e22bee78 | 688 | { |
63d95a91 | 689 | return need_to_create_worker(pool) || |
11ebea50 | 690 | (pool->flags & POOL_MANAGE_WORKERS); |
e22bee78 TH |
691 | } |
692 | ||
693 | /* Do we have too many workers and should some go away? */ | |
63d95a91 | 694 | static bool too_many_workers(struct worker_pool *pool) |
e22bee78 | 695 | { |
552a37e9 | 696 | bool managing = pool->flags & POOL_MANAGING_WORKERS; |
63d95a91 TH |
697 | int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ |
698 | int nr_busy = pool->nr_workers - nr_idle; | |
e22bee78 | 699 | |
ea1abd61 LJ |
700 | /* |
701 | * nr_idle and idle_list may disagree if idle rebinding is in | |
702 | * progress. Never return %true if idle_list is empty. | |
703 | */ | |
704 | if (list_empty(&pool->idle_list)) | |
705 | return false; | |
706 | ||
e22bee78 | 707 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; |
365970a1 DH |
708 | } |
709 | ||
4d707b9f | 710 | /* |
e22bee78 TH |
711 | * Wake up functions. |
712 | */ | |
713 | ||
7e11629d | 714 | /* Return the first worker. Safe with preemption disabled */ |
63d95a91 | 715 | static struct worker *first_worker(struct worker_pool *pool) |
7e11629d | 716 | { |
63d95a91 | 717 | if (unlikely(list_empty(&pool->idle_list))) |
7e11629d TH |
718 | return NULL; |
719 | ||
63d95a91 | 720 | return list_first_entry(&pool->idle_list, struct worker, entry); |
7e11629d TH |
721 | } |
722 | ||
723 | /** | |
724 | * wake_up_worker - wake up an idle worker | |
63d95a91 | 725 | * @pool: worker pool to wake worker from |
7e11629d | 726 | * |
63d95a91 | 727 | * Wake up the first idle worker of @pool. |
7e11629d TH |
728 | * |
729 | * CONTEXT: | |
d565ed63 | 730 | * spin_lock_irq(pool->lock). |
7e11629d | 731 | */ |
63d95a91 | 732 | static void wake_up_worker(struct worker_pool *pool) |
7e11629d | 733 | { |
63d95a91 | 734 | struct worker *worker = first_worker(pool); |
7e11629d TH |
735 | |
736 | if (likely(worker)) | |
737 | wake_up_process(worker->task); | |
738 | } | |
739 | ||
d302f017 | 740 | /** |
e22bee78 TH |
741 | * wq_worker_waking_up - a worker is waking up |
742 | * @task: task waking up | |
743 | * @cpu: CPU @task is waking up to | |
744 | * | |
745 | * This function is called during try_to_wake_up() when a worker is | |
746 | * being awoken. | |
747 | * | |
748 | * CONTEXT: | |
749 | * spin_lock_irq(rq->lock) | |
750 | */ | |
751 | void wq_worker_waking_up(struct task_struct *task, unsigned int cpu) | |
752 | { | |
753 | struct worker *worker = kthread_data(task); | |
754 | ||
36576000 | 755 | if (!(worker->flags & WORKER_NOT_RUNNING)) { |
ec22ca5e | 756 | WARN_ON_ONCE(worker->pool->cpu != cpu); |
63d95a91 | 757 | atomic_inc(get_pool_nr_running(worker->pool)); |
36576000 | 758 | } |
e22bee78 TH |
759 | } |
760 | ||
761 | /** | |
762 | * wq_worker_sleeping - a worker is going to sleep | |
763 | * @task: task going to sleep | |
764 | * @cpu: CPU in question, must be the current CPU number | |
765 | * | |
766 | * This function is called during schedule() when a busy worker is | |
767 | * going to sleep. Worker on the same cpu can be woken up by | |
768 | * returning pointer to its task. | |
769 | * | |
770 | * CONTEXT: | |
771 | * spin_lock_irq(rq->lock) | |
772 | * | |
773 | * RETURNS: | |
774 | * Worker task on @cpu to wake up, %NULL if none. | |
775 | */ | |
776 | struct task_struct *wq_worker_sleeping(struct task_struct *task, | |
777 | unsigned int cpu) | |
778 | { | |
779 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
111c225a TH |
780 | struct worker_pool *pool; |
781 | atomic_t *nr_running; | |
e22bee78 | 782 | |
111c225a TH |
783 | /* |
784 | * Rescuers, which may not have all the fields set up like normal | |
785 | * workers, also reach here, let's not access anything before | |
786 | * checking NOT_RUNNING. | |
787 | */ | |
2d64672e | 788 | if (worker->flags & WORKER_NOT_RUNNING) |
e22bee78 TH |
789 | return NULL; |
790 | ||
111c225a TH |
791 | pool = worker->pool; |
792 | nr_running = get_pool_nr_running(pool); | |
793 | ||
e22bee78 TH |
794 | /* this can only happen on the local cpu */ |
795 | BUG_ON(cpu != raw_smp_processor_id()); | |
796 | ||
797 | /* | |
798 | * The counterpart of the following dec_and_test, implied mb, | |
799 | * worklist not empty test sequence is in insert_work(). | |
800 | * Please read comment there. | |
801 | * | |
628c78e7 TH |
802 | * NOT_RUNNING is clear. This means that we're bound to and |
803 | * running on the local cpu w/ rq lock held and preemption | |
804 | * disabled, which in turn means that none else could be | |
d565ed63 | 805 | * manipulating idle_list, so dereferencing idle_list without pool |
628c78e7 | 806 | * lock is safe. |
e22bee78 | 807 | */ |
bd7bdd43 | 808 | if (atomic_dec_and_test(nr_running) && !list_empty(&pool->worklist)) |
63d95a91 | 809 | to_wakeup = first_worker(pool); |
e22bee78 TH |
810 | return to_wakeup ? to_wakeup->task : NULL; |
811 | } | |
812 | ||
813 | /** | |
814 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 815 | * @worker: self |
d302f017 TH |
816 | * @flags: flags to set |
817 | * @wakeup: wakeup an idle worker if necessary | |
818 | * | |
e22bee78 TH |
819 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
820 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
821 | * woken up. | |
d302f017 | 822 | * |
cb444766 | 823 | * CONTEXT: |
d565ed63 | 824 | * spin_lock_irq(pool->lock) |
d302f017 TH |
825 | */ |
826 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
827 | bool wakeup) | |
828 | { | |
bd7bdd43 | 829 | struct worker_pool *pool = worker->pool; |
e22bee78 | 830 | |
cb444766 TH |
831 | WARN_ON_ONCE(worker->task != current); |
832 | ||
e22bee78 TH |
833 | /* |
834 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
835 | * wake up an idle worker as necessary if requested by | |
836 | * @wakeup. | |
837 | */ | |
838 | if ((flags & WORKER_NOT_RUNNING) && | |
839 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
63d95a91 | 840 | atomic_t *nr_running = get_pool_nr_running(pool); |
e22bee78 TH |
841 | |
842 | if (wakeup) { | |
843 | if (atomic_dec_and_test(nr_running) && | |
bd7bdd43 | 844 | !list_empty(&pool->worklist)) |
63d95a91 | 845 | wake_up_worker(pool); |
e22bee78 TH |
846 | } else |
847 | atomic_dec(nr_running); | |
848 | } | |
849 | ||
d302f017 TH |
850 | worker->flags |= flags; |
851 | } | |
852 | ||
853 | /** | |
e22bee78 | 854 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 855 | * @worker: self |
d302f017 TH |
856 | * @flags: flags to clear |
857 | * | |
e22bee78 | 858 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 859 | * |
cb444766 | 860 | * CONTEXT: |
d565ed63 | 861 | * spin_lock_irq(pool->lock) |
d302f017 TH |
862 | */ |
863 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
864 | { | |
63d95a91 | 865 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
866 | unsigned int oflags = worker->flags; |
867 | ||
cb444766 TH |
868 | WARN_ON_ONCE(worker->task != current); |
869 | ||
d302f017 | 870 | worker->flags &= ~flags; |
e22bee78 | 871 | |
42c025f3 TH |
872 | /* |
873 | * If transitioning out of NOT_RUNNING, increment nr_running. Note | |
874 | * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask | |
875 | * of multiple flags, not a single flag. | |
876 | */ | |
e22bee78 TH |
877 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) |
878 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
63d95a91 | 879 | atomic_inc(get_pool_nr_running(pool)); |
d302f017 TH |
880 | } |
881 | ||
8cca0eea TH |
882 | /** |
883 | * find_worker_executing_work - find worker which is executing a work | |
c9e7cf27 | 884 | * @pool: pool of interest |
8cca0eea TH |
885 | * @work: work to find worker for |
886 | * | |
c9e7cf27 TH |
887 | * Find a worker which is executing @work on @pool by searching |
888 | * @pool->busy_hash which is keyed by the address of @work. For a worker | |
a2c1c57b TH |
889 | * to match, its current execution should match the address of @work and |
890 | * its work function. This is to avoid unwanted dependency between | |
891 | * unrelated work executions through a work item being recycled while still | |
892 | * being executed. | |
893 | * | |
894 | * This is a bit tricky. A work item may be freed once its execution | |
895 | * starts and nothing prevents the freed area from being recycled for | |
896 | * another work item. If the same work item address ends up being reused | |
897 | * before the original execution finishes, workqueue will identify the | |
898 | * recycled work item as currently executing and make it wait until the | |
899 | * current execution finishes, introducing an unwanted dependency. | |
900 | * | |
901 | * This function checks the work item address, work function and workqueue | |
902 | * to avoid false positives. Note that this isn't complete as one may | |
903 | * construct a work function which can introduce dependency onto itself | |
904 | * through a recycled work item. Well, if somebody wants to shoot oneself | |
905 | * in the foot that badly, there's only so much we can do, and if such | |
906 | * deadlock actually occurs, it should be easy to locate the culprit work | |
907 | * function. | |
8cca0eea TH |
908 | * |
909 | * CONTEXT: | |
d565ed63 | 910 | * spin_lock_irq(pool->lock). |
8cca0eea TH |
911 | * |
912 | * RETURNS: | |
913 | * Pointer to worker which is executing @work if found, NULL | |
914 | * otherwise. | |
4d707b9f | 915 | */ |
c9e7cf27 | 916 | static struct worker *find_worker_executing_work(struct worker_pool *pool, |
8cca0eea | 917 | struct work_struct *work) |
4d707b9f | 918 | { |
42f8570f SL |
919 | struct worker *worker; |
920 | struct hlist_node *tmp; | |
921 | ||
c9e7cf27 | 922 | hash_for_each_possible(pool->busy_hash, worker, tmp, hentry, |
a2c1c57b TH |
923 | (unsigned long)work) |
924 | if (worker->current_work == work && | |
925 | worker->current_func == work->func) | |
42f8570f SL |
926 | return worker; |
927 | ||
928 | return NULL; | |
4d707b9f ON |
929 | } |
930 | ||
bf4ede01 TH |
931 | /** |
932 | * move_linked_works - move linked works to a list | |
933 | * @work: start of series of works to be scheduled | |
934 | * @head: target list to append @work to | |
935 | * @nextp: out paramter for nested worklist walking | |
936 | * | |
937 | * Schedule linked works starting from @work to @head. Work series to | |
938 | * be scheduled starts at @work and includes any consecutive work with | |
939 | * WORK_STRUCT_LINKED set in its predecessor. | |
940 | * | |
941 | * If @nextp is not NULL, it's updated to point to the next work of | |
942 | * the last scheduled work. This allows move_linked_works() to be | |
943 | * nested inside outer list_for_each_entry_safe(). | |
944 | * | |
945 | * CONTEXT: | |
d565ed63 | 946 | * spin_lock_irq(pool->lock). |
bf4ede01 TH |
947 | */ |
948 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
949 | struct work_struct **nextp) | |
950 | { | |
951 | struct work_struct *n; | |
952 | ||
953 | /* | |
954 | * Linked worklist will always end before the end of the list, | |
955 | * use NULL for list head. | |
956 | */ | |
957 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
958 | list_move_tail(&work->entry, head); | |
959 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
960 | break; | |
961 | } | |
962 | ||
963 | /* | |
964 | * If we're already inside safe list traversal and have moved | |
965 | * multiple works to the scheduled queue, the next position | |
966 | * needs to be updated. | |
967 | */ | |
968 | if (nextp) | |
969 | *nextp = n; | |
970 | } | |
971 | ||
3aa62497 | 972 | static void cwq_activate_delayed_work(struct work_struct *work) |
bf4ede01 | 973 | { |
3aa62497 | 974 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); |
bf4ede01 TH |
975 | |
976 | trace_workqueue_activate_work(work); | |
977 | move_linked_works(work, &cwq->pool->worklist, NULL); | |
978 | __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); | |
979 | cwq->nr_active++; | |
980 | } | |
981 | ||
3aa62497 LJ |
982 | static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq) |
983 | { | |
984 | struct work_struct *work = list_first_entry(&cwq->delayed_works, | |
985 | struct work_struct, entry); | |
986 | ||
987 | cwq_activate_delayed_work(work); | |
988 | } | |
989 | ||
bf4ede01 TH |
990 | /** |
991 | * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight | |
992 | * @cwq: cwq of interest | |
993 | * @color: color of work which left the queue | |
bf4ede01 TH |
994 | * |
995 | * A work either has completed or is removed from pending queue, | |
996 | * decrement nr_in_flight of its cwq and handle workqueue flushing. | |
997 | * | |
998 | * CONTEXT: | |
d565ed63 | 999 | * spin_lock_irq(pool->lock). |
bf4ede01 | 1000 | */ |
b3f9f405 | 1001 | static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color) |
bf4ede01 TH |
1002 | { |
1003 | /* ignore uncolored works */ | |
1004 | if (color == WORK_NO_COLOR) | |
1005 | return; | |
1006 | ||
1007 | cwq->nr_in_flight[color]--; | |
1008 | ||
b3f9f405 LJ |
1009 | cwq->nr_active--; |
1010 | if (!list_empty(&cwq->delayed_works)) { | |
1011 | /* one down, submit a delayed one */ | |
1012 | if (cwq->nr_active < cwq->max_active) | |
1013 | cwq_activate_first_delayed(cwq); | |
bf4ede01 TH |
1014 | } |
1015 | ||
1016 | /* is flush in progress and are we at the flushing tip? */ | |
1017 | if (likely(cwq->flush_color != color)) | |
1018 | return; | |
1019 | ||
1020 | /* are there still in-flight works? */ | |
1021 | if (cwq->nr_in_flight[color]) | |
1022 | return; | |
1023 | ||
1024 | /* this cwq is done, clear flush_color */ | |
1025 | cwq->flush_color = -1; | |
1026 | ||
1027 | /* | |
1028 | * If this was the last cwq, wake up the first flusher. It | |
1029 | * will handle the rest. | |
1030 | */ | |
1031 | if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush)) | |
1032 | complete(&cwq->wq->first_flusher->done); | |
1033 | } | |
1034 | ||
36e227d2 | 1035 | /** |
bbb68dfa | 1036 | * try_to_grab_pending - steal work item from worklist and disable irq |
36e227d2 TH |
1037 | * @work: work item to steal |
1038 | * @is_dwork: @work is a delayed_work | |
bbb68dfa | 1039 | * @flags: place to store irq state |
36e227d2 TH |
1040 | * |
1041 | * Try to grab PENDING bit of @work. This function can handle @work in any | |
1042 | * stable state - idle, on timer or on worklist. Return values are | |
1043 | * | |
1044 | * 1 if @work was pending and we successfully stole PENDING | |
1045 | * 0 if @work was idle and we claimed PENDING | |
1046 | * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry | |
bbb68dfa TH |
1047 | * -ENOENT if someone else is canceling @work, this state may persist |
1048 | * for arbitrarily long | |
36e227d2 | 1049 | * |
bbb68dfa | 1050 | * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting |
e0aecdd8 TH |
1051 | * interrupted while holding PENDING and @work off queue, irq must be |
1052 | * disabled on entry. This, combined with delayed_work->timer being | |
1053 | * irqsafe, ensures that we return -EAGAIN for finite short period of time. | |
bbb68dfa TH |
1054 | * |
1055 | * On successful return, >= 0, irq is disabled and the caller is | |
1056 | * responsible for releasing it using local_irq_restore(*@flags). | |
1057 | * | |
e0aecdd8 | 1058 | * This function is safe to call from any context including IRQ handler. |
bf4ede01 | 1059 | */ |
bbb68dfa TH |
1060 | static int try_to_grab_pending(struct work_struct *work, bool is_dwork, |
1061 | unsigned long *flags) | |
bf4ede01 | 1062 | { |
d565ed63 | 1063 | struct worker_pool *pool; |
bf4ede01 | 1064 | |
bbb68dfa TH |
1065 | local_irq_save(*flags); |
1066 | ||
36e227d2 TH |
1067 | /* try to steal the timer if it exists */ |
1068 | if (is_dwork) { | |
1069 | struct delayed_work *dwork = to_delayed_work(work); | |
1070 | ||
e0aecdd8 TH |
1071 | /* |
1072 | * dwork->timer is irqsafe. If del_timer() fails, it's | |
1073 | * guaranteed that the timer is not queued anywhere and not | |
1074 | * running on the local CPU. | |
1075 | */ | |
36e227d2 TH |
1076 | if (likely(del_timer(&dwork->timer))) |
1077 | return 1; | |
1078 | } | |
1079 | ||
1080 | /* try to claim PENDING the normal way */ | |
bf4ede01 TH |
1081 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1082 | return 0; | |
1083 | ||
1084 | /* | |
1085 | * The queueing is in progress, or it is already queued. Try to | |
1086 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
1087 | */ | |
d565ed63 TH |
1088 | pool = get_work_pool(work); |
1089 | if (!pool) | |
bbb68dfa | 1090 | goto fail; |
bf4ede01 | 1091 | |
d565ed63 | 1092 | spin_lock(&pool->lock); |
bf4ede01 TH |
1093 | if (!list_empty(&work->entry)) { |
1094 | /* | |
d565ed63 TH |
1095 | * This work is queued, but perhaps we locked the wrong |
1096 | * pool. In that case we must see the new value after | |
1097 | * rmb(), see insert_work()->wmb(). | |
bf4ede01 TH |
1098 | */ |
1099 | smp_rmb(); | |
d565ed63 | 1100 | if (pool == get_work_pool(work)) { |
bf4ede01 | 1101 | debug_work_deactivate(work); |
3aa62497 LJ |
1102 | |
1103 | /* | |
1104 | * A delayed work item cannot be grabbed directly | |
1105 | * because it might have linked NO_COLOR work items | |
1106 | * which, if left on the delayed_list, will confuse | |
1107 | * cwq->nr_active management later on and cause | |
1108 | * stall. Make sure the work item is activated | |
1109 | * before grabbing. | |
1110 | */ | |
1111 | if (*work_data_bits(work) & WORK_STRUCT_DELAYED) | |
1112 | cwq_activate_delayed_work(work); | |
1113 | ||
bf4ede01 TH |
1114 | list_del_init(&work->entry); |
1115 | cwq_dec_nr_in_flight(get_work_cwq(work), | |
b3f9f405 | 1116 | get_work_color(work)); |
36e227d2 | 1117 | |
d565ed63 | 1118 | spin_unlock(&pool->lock); |
36e227d2 | 1119 | return 1; |
bf4ede01 TH |
1120 | } |
1121 | } | |
d565ed63 | 1122 | spin_unlock(&pool->lock); |
bbb68dfa TH |
1123 | fail: |
1124 | local_irq_restore(*flags); | |
1125 | if (work_is_canceling(work)) | |
1126 | return -ENOENT; | |
1127 | cpu_relax(); | |
36e227d2 | 1128 | return -EAGAIN; |
bf4ede01 TH |
1129 | } |
1130 | ||
4690c4ab | 1131 | /** |
706026c2 | 1132 | * insert_work - insert a work into a pool |
4690c4ab TH |
1133 | * @cwq: cwq @work belongs to |
1134 | * @work: work to insert | |
1135 | * @head: insertion point | |
1136 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
1137 | * | |
706026c2 TH |
1138 | * Insert @work which belongs to @cwq after @head. @extra_flags is or'd to |
1139 | * work_struct flags. | |
4690c4ab TH |
1140 | * |
1141 | * CONTEXT: | |
d565ed63 | 1142 | * spin_lock_irq(pool->lock). |
4690c4ab | 1143 | */ |
b89deed3 | 1144 | static void insert_work(struct cpu_workqueue_struct *cwq, |
4690c4ab TH |
1145 | struct work_struct *work, struct list_head *head, |
1146 | unsigned int extra_flags) | |
b89deed3 | 1147 | { |
63d95a91 | 1148 | struct worker_pool *pool = cwq->pool; |
e22bee78 | 1149 | |
4690c4ab | 1150 | /* we own @work, set data and link */ |
7a22ad75 | 1151 | set_work_cwq(work, cwq, extra_flags); |
e1d8aa9f | 1152 | |
6e84d644 ON |
1153 | /* |
1154 | * Ensure that we get the right work->data if we see the | |
1155 | * result of list_add() below, see try_to_grab_pending(). | |
1156 | */ | |
1157 | smp_wmb(); | |
4690c4ab | 1158 | |
1a4d9b0a | 1159 | list_add_tail(&work->entry, head); |
e22bee78 TH |
1160 | |
1161 | /* | |
1162 | * Ensure either worker_sched_deactivated() sees the above | |
1163 | * list_add_tail() or we see zero nr_running to avoid workers | |
1164 | * lying around lazily while there are works to be processed. | |
1165 | */ | |
1166 | smp_mb(); | |
1167 | ||
63d95a91 TH |
1168 | if (__need_more_worker(pool)) |
1169 | wake_up_worker(pool); | |
b89deed3 ON |
1170 | } |
1171 | ||
c8efcc25 TH |
1172 | /* |
1173 | * Test whether @work is being queued from another work executing on the | |
1174 | * same workqueue. This is rather expensive and should only be used from | |
1175 | * cold paths. | |
1176 | */ | |
1177 | static bool is_chained_work(struct workqueue_struct *wq) | |
1178 | { | |
1179 | unsigned long flags; | |
1180 | unsigned int cpu; | |
1181 | ||
706026c2 | 1182 | for_each_wq_cpu(cpu) { |
c9e7cf27 TH |
1183 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
1184 | struct worker_pool *pool = cwq->pool; | |
c8efcc25 TH |
1185 | struct worker *worker; |
1186 | struct hlist_node *pos; | |
1187 | int i; | |
1188 | ||
d565ed63 | 1189 | spin_lock_irqsave(&pool->lock, flags); |
c9e7cf27 | 1190 | for_each_busy_worker(worker, i, pos, pool) { |
c8efcc25 TH |
1191 | if (worker->task != current) |
1192 | continue; | |
d565ed63 | 1193 | spin_unlock_irqrestore(&pool->lock, flags); |
c8efcc25 TH |
1194 | /* |
1195 | * I'm @worker, no locking necessary. See if @work | |
1196 | * is headed to the same workqueue. | |
1197 | */ | |
1198 | return worker->current_cwq->wq == wq; | |
1199 | } | |
d565ed63 | 1200 | spin_unlock_irqrestore(&pool->lock, flags); |
c8efcc25 TH |
1201 | } |
1202 | return false; | |
1203 | } | |
1204 | ||
4690c4ab | 1205 | static void __queue_work(unsigned int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
1206 | struct work_struct *work) |
1207 | { | |
d565ed63 TH |
1208 | bool highpri = wq->flags & WQ_HIGHPRI; |
1209 | struct worker_pool *pool; | |
502ca9d8 | 1210 | struct cpu_workqueue_struct *cwq; |
1e19ffc6 | 1211 | struct list_head *worklist; |
8a2e8e5d | 1212 | unsigned int work_flags; |
b75cac93 | 1213 | unsigned int req_cpu = cpu; |
8930caba TH |
1214 | |
1215 | /* | |
1216 | * While a work item is PENDING && off queue, a task trying to | |
1217 | * steal the PENDING will busy-loop waiting for it to either get | |
1218 | * queued or lose PENDING. Grabbing PENDING and queueing should | |
1219 | * happen with IRQ disabled. | |
1220 | */ | |
1221 | WARN_ON_ONCE(!irqs_disabled()); | |
1da177e4 | 1222 | |
dc186ad7 | 1223 | debug_work_activate(work); |
1e19ffc6 | 1224 | |
c8efcc25 | 1225 | /* if dying, only works from the same workqueue are allowed */ |
9c5a2ba7 | 1226 | if (unlikely(wq->flags & WQ_DRAINING) && |
c8efcc25 | 1227 | WARN_ON_ONCE(!is_chained_work(wq))) |
e41e704b TH |
1228 | return; |
1229 | ||
d565ed63 | 1230 | /* determine pool to use */ |
c7fc77f7 | 1231 | if (!(wq->flags & WQ_UNBOUND)) { |
c9e7cf27 | 1232 | struct worker_pool *last_pool; |
18aa9eff | 1233 | |
57469821 | 1234 | if (cpu == WORK_CPU_UNBOUND) |
c7fc77f7 TH |
1235 | cpu = raw_smp_processor_id(); |
1236 | ||
18aa9eff | 1237 | /* |
dbf2576e TH |
1238 | * It's multi cpu. If @work was previously on a different |
1239 | * cpu, it might still be running there, in which case the | |
1240 | * work needs to be queued on that cpu to guarantee | |
1241 | * non-reentrancy. | |
18aa9eff | 1242 | */ |
d565ed63 | 1243 | pool = get_std_worker_pool(cpu, highpri); |
c9e7cf27 | 1244 | last_pool = get_work_pool(work); |
dbf2576e | 1245 | |
d565ed63 | 1246 | if (last_pool && last_pool != pool) { |
18aa9eff TH |
1247 | struct worker *worker; |
1248 | ||
d565ed63 | 1249 | spin_lock(&last_pool->lock); |
18aa9eff | 1250 | |
c9e7cf27 | 1251 | worker = find_worker_executing_work(last_pool, work); |
18aa9eff TH |
1252 | |
1253 | if (worker && worker->current_cwq->wq == wq) | |
d565ed63 | 1254 | pool = last_pool; |
18aa9eff TH |
1255 | else { |
1256 | /* meh... not running there, queue here */ | |
d565ed63 TH |
1257 | spin_unlock(&last_pool->lock); |
1258 | spin_lock(&pool->lock); | |
18aa9eff | 1259 | } |
8930caba | 1260 | } else { |
d565ed63 | 1261 | spin_lock(&pool->lock); |
8930caba | 1262 | } |
f3421797 | 1263 | } else { |
d565ed63 TH |
1264 | pool = get_std_worker_pool(WORK_CPU_UNBOUND, highpri); |
1265 | spin_lock(&pool->lock); | |
502ca9d8 TH |
1266 | } |
1267 | ||
d565ed63 TH |
1268 | /* pool determined, get cwq and queue */ |
1269 | cwq = get_cwq(pool->cpu, wq); | |
b75cac93 | 1270 | trace_workqueue_queue_work(req_cpu, cwq, work); |
502ca9d8 | 1271 | |
f5b2552b | 1272 | if (WARN_ON(!list_empty(&work->entry))) { |
d565ed63 | 1273 | spin_unlock(&pool->lock); |
f5b2552b DC |
1274 | return; |
1275 | } | |
1e19ffc6 | 1276 | |
73f53c4a | 1277 | cwq->nr_in_flight[cwq->work_color]++; |
8a2e8e5d | 1278 | work_flags = work_color_to_flags(cwq->work_color); |
1e19ffc6 TH |
1279 | |
1280 | if (likely(cwq->nr_active < cwq->max_active)) { | |
cdadf009 | 1281 | trace_workqueue_activate_work(work); |
1e19ffc6 | 1282 | cwq->nr_active++; |
3270476a | 1283 | worklist = &cwq->pool->worklist; |
8a2e8e5d TH |
1284 | } else { |
1285 | work_flags |= WORK_STRUCT_DELAYED; | |
1e19ffc6 | 1286 | worklist = &cwq->delayed_works; |
8a2e8e5d | 1287 | } |
1e19ffc6 | 1288 | |
8a2e8e5d | 1289 | insert_work(cwq, work, worklist, work_flags); |
1e19ffc6 | 1290 | |
d565ed63 | 1291 | spin_unlock(&pool->lock); |
1da177e4 LT |
1292 | } |
1293 | ||
0fcb78c2 | 1294 | /** |
c1a220e7 ZR |
1295 | * queue_work_on - queue work on specific cpu |
1296 | * @cpu: CPU number to execute work on | |
0fcb78c2 REB |
1297 | * @wq: workqueue to use |
1298 | * @work: work to queue | |
1299 | * | |
d4283e93 | 1300 | * Returns %false if @work was already on a queue, %true otherwise. |
1da177e4 | 1301 | * |
c1a220e7 ZR |
1302 | * We queue the work to a specific CPU, the caller must ensure it |
1303 | * can't go away. | |
1da177e4 | 1304 | */ |
d4283e93 TH |
1305 | bool queue_work_on(int cpu, struct workqueue_struct *wq, |
1306 | struct work_struct *work) | |
1da177e4 | 1307 | { |
d4283e93 | 1308 | bool ret = false; |
8930caba | 1309 | unsigned long flags; |
ef1ca236 | 1310 | |
8930caba | 1311 | local_irq_save(flags); |
c1a220e7 | 1312 | |
22df02bb | 1313 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1314 | __queue_work(cpu, wq, work); |
d4283e93 | 1315 | ret = true; |
c1a220e7 | 1316 | } |
ef1ca236 | 1317 | |
8930caba | 1318 | local_irq_restore(flags); |
1da177e4 LT |
1319 | return ret; |
1320 | } | |
c1a220e7 | 1321 | EXPORT_SYMBOL_GPL(queue_work_on); |
1da177e4 | 1322 | |
c1a220e7 | 1323 | /** |
0a13c00e | 1324 | * queue_work - queue work on a workqueue |
c1a220e7 ZR |
1325 | * @wq: workqueue to use |
1326 | * @work: work to queue | |
1327 | * | |
d4283e93 | 1328 | * Returns %false if @work was already on a queue, %true otherwise. |
c1a220e7 | 1329 | * |
0a13c00e TH |
1330 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
1331 | * it can be processed by another CPU. | |
c1a220e7 | 1332 | */ |
d4283e93 | 1333 | bool queue_work(struct workqueue_struct *wq, struct work_struct *work) |
c1a220e7 | 1334 | { |
57469821 | 1335 | return queue_work_on(WORK_CPU_UNBOUND, wq, work); |
c1a220e7 | 1336 | } |
0a13c00e | 1337 | EXPORT_SYMBOL_GPL(queue_work); |
c1a220e7 | 1338 | |
d8e794df | 1339 | void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1340 | { |
52bad64d | 1341 | struct delayed_work *dwork = (struct delayed_work *)__data; |
7a22ad75 | 1342 | struct cpu_workqueue_struct *cwq = get_work_cwq(&dwork->work); |
1da177e4 | 1343 | |
e0aecdd8 | 1344 | /* should have been called from irqsafe timer with irq already off */ |
1265057f | 1345 | __queue_work(dwork->cpu, cwq->wq, &dwork->work); |
1da177e4 | 1346 | } |
d8e794df | 1347 | EXPORT_SYMBOL_GPL(delayed_work_timer_fn); |
1da177e4 | 1348 | |
7beb2edf TH |
1349 | static void __queue_delayed_work(int cpu, struct workqueue_struct *wq, |
1350 | struct delayed_work *dwork, unsigned long delay) | |
1da177e4 | 1351 | { |
7beb2edf TH |
1352 | struct timer_list *timer = &dwork->timer; |
1353 | struct work_struct *work = &dwork->work; | |
1354 | unsigned int lcpu; | |
1355 | ||
1356 | WARN_ON_ONCE(timer->function != delayed_work_timer_fn || | |
1357 | timer->data != (unsigned long)dwork); | |
fc4b514f TH |
1358 | WARN_ON_ONCE(timer_pending(timer)); |
1359 | WARN_ON_ONCE(!list_empty(&work->entry)); | |
7beb2edf | 1360 | |
8852aac2 TH |
1361 | /* |
1362 | * If @delay is 0, queue @dwork->work immediately. This is for | |
1363 | * both optimization and correctness. The earliest @timer can | |
1364 | * expire is on the closest next tick and delayed_work users depend | |
1365 | * on that there's no such delay when @delay is 0. | |
1366 | */ | |
1367 | if (!delay) { | |
1368 | __queue_work(cpu, wq, &dwork->work); | |
1369 | return; | |
1370 | } | |
1371 | ||
7beb2edf | 1372 | timer_stats_timer_set_start_info(&dwork->timer); |
1da177e4 | 1373 | |
7beb2edf TH |
1374 | /* |
1375 | * This stores cwq for the moment, for the timer_fn. Note that the | |
ec22ca5e | 1376 | * work's pool is preserved to allow reentrance detection for |
7beb2edf TH |
1377 | * delayed works. |
1378 | */ | |
1379 | if (!(wq->flags & WQ_UNBOUND)) { | |
ec22ca5e | 1380 | struct worker_pool *pool = get_work_pool(work); |
7beb2edf | 1381 | |
e42986de | 1382 | /* |
ec22ca5e | 1383 | * If we cannot get the last pool from @work directly, |
e42986de JK |
1384 | * select the last CPU such that it avoids unnecessarily |
1385 | * triggering non-reentrancy check in __queue_work(). | |
1386 | */ | |
1387 | lcpu = cpu; | |
ec22ca5e TH |
1388 | if (pool) |
1389 | lcpu = pool->cpu; | |
e42986de | 1390 | if (lcpu == WORK_CPU_UNBOUND) |
7beb2edf TH |
1391 | lcpu = raw_smp_processor_id(); |
1392 | } else { | |
1393 | lcpu = WORK_CPU_UNBOUND; | |
1394 | } | |
1395 | ||
1396 | set_work_cwq(work, get_cwq(lcpu, wq), 0); | |
1397 | ||
1265057f | 1398 | dwork->cpu = cpu; |
7beb2edf TH |
1399 | timer->expires = jiffies + delay; |
1400 | ||
1401 | if (unlikely(cpu != WORK_CPU_UNBOUND)) | |
1402 | add_timer_on(timer, cpu); | |
1403 | else | |
1404 | add_timer(timer); | |
1da177e4 LT |
1405 | } |
1406 | ||
0fcb78c2 REB |
1407 | /** |
1408 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1409 | * @cpu: CPU number to execute work on | |
1410 | * @wq: workqueue to use | |
af9997e4 | 1411 | * @dwork: work to queue |
0fcb78c2 REB |
1412 | * @delay: number of jiffies to wait before queueing |
1413 | * | |
715f1300 TH |
1414 | * Returns %false if @work was already on a queue, %true otherwise. If |
1415 | * @delay is zero and @dwork is idle, it will be scheduled for immediate | |
1416 | * execution. | |
0fcb78c2 | 1417 | */ |
d4283e93 TH |
1418 | bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
1419 | struct delayed_work *dwork, unsigned long delay) | |
7a6bc1cd | 1420 | { |
52bad64d | 1421 | struct work_struct *work = &dwork->work; |
d4283e93 | 1422 | bool ret = false; |
8930caba | 1423 | unsigned long flags; |
7a6bc1cd | 1424 | |
8930caba TH |
1425 | /* read the comment in __queue_work() */ |
1426 | local_irq_save(flags); | |
7a6bc1cd | 1427 | |
22df02bb | 1428 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
7beb2edf | 1429 | __queue_delayed_work(cpu, wq, dwork, delay); |
d4283e93 | 1430 | ret = true; |
7a6bc1cd | 1431 | } |
8a3e77cc | 1432 | |
8930caba | 1433 | local_irq_restore(flags); |
7a6bc1cd VP |
1434 | return ret; |
1435 | } | |
ae90dd5d | 1436 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
c7fc77f7 | 1437 | |
0a13c00e TH |
1438 | /** |
1439 | * queue_delayed_work - queue work on a workqueue after delay | |
1440 | * @wq: workqueue to use | |
1441 | * @dwork: delayable work to queue | |
1442 | * @delay: number of jiffies to wait before queueing | |
1443 | * | |
715f1300 | 1444 | * Equivalent to queue_delayed_work_on() but tries to use the local CPU. |
0a13c00e | 1445 | */ |
d4283e93 | 1446 | bool queue_delayed_work(struct workqueue_struct *wq, |
0a13c00e TH |
1447 | struct delayed_work *dwork, unsigned long delay) |
1448 | { | |
57469821 | 1449 | return queue_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); |
0a13c00e TH |
1450 | } |
1451 | EXPORT_SYMBOL_GPL(queue_delayed_work); | |
c7fc77f7 | 1452 | |
8376fe22 TH |
1453 | /** |
1454 | * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU | |
1455 | * @cpu: CPU number to execute work on | |
1456 | * @wq: workqueue to use | |
1457 | * @dwork: work to queue | |
1458 | * @delay: number of jiffies to wait before queueing | |
1459 | * | |
1460 | * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise, | |
1461 | * modify @dwork's timer so that it expires after @delay. If @delay is | |
1462 | * zero, @work is guaranteed to be scheduled immediately regardless of its | |
1463 | * current state. | |
1464 | * | |
1465 | * Returns %false if @dwork was idle and queued, %true if @dwork was | |
1466 | * pending and its timer was modified. | |
1467 | * | |
e0aecdd8 | 1468 | * This function is safe to call from any context including IRQ handler. |
8376fe22 TH |
1469 | * See try_to_grab_pending() for details. |
1470 | */ | |
1471 | bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
1472 | struct delayed_work *dwork, unsigned long delay) | |
1473 | { | |
1474 | unsigned long flags; | |
1475 | int ret; | |
c7fc77f7 | 1476 | |
8376fe22 TH |
1477 | do { |
1478 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
1479 | } while (unlikely(ret == -EAGAIN)); | |
63bc0362 | 1480 | |
8376fe22 TH |
1481 | if (likely(ret >= 0)) { |
1482 | __queue_delayed_work(cpu, wq, dwork, delay); | |
1483 | local_irq_restore(flags); | |
7a6bc1cd | 1484 | } |
8376fe22 TH |
1485 | |
1486 | /* -ENOENT from try_to_grab_pending() becomes %true */ | |
7a6bc1cd VP |
1487 | return ret; |
1488 | } | |
8376fe22 TH |
1489 | EXPORT_SYMBOL_GPL(mod_delayed_work_on); |
1490 | ||
1491 | /** | |
1492 | * mod_delayed_work - modify delay of or queue a delayed work | |
1493 | * @wq: workqueue to use | |
1494 | * @dwork: work to queue | |
1495 | * @delay: number of jiffies to wait before queueing | |
1496 | * | |
1497 | * mod_delayed_work_on() on local CPU. | |
1498 | */ | |
1499 | bool mod_delayed_work(struct workqueue_struct *wq, struct delayed_work *dwork, | |
1500 | unsigned long delay) | |
1501 | { | |
1502 | return mod_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay); | |
1503 | } | |
1504 | EXPORT_SYMBOL_GPL(mod_delayed_work); | |
1da177e4 | 1505 | |
c8e55f36 TH |
1506 | /** |
1507 | * worker_enter_idle - enter idle state | |
1508 | * @worker: worker which is entering idle state | |
1509 | * | |
1510 | * @worker is entering idle state. Update stats and idle timer if | |
1511 | * necessary. | |
1512 | * | |
1513 | * LOCKING: | |
d565ed63 | 1514 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1515 | */ |
1516 | static void worker_enter_idle(struct worker *worker) | |
1da177e4 | 1517 | { |
bd7bdd43 | 1518 | struct worker_pool *pool = worker->pool; |
c8e55f36 TH |
1519 | |
1520 | BUG_ON(worker->flags & WORKER_IDLE); | |
1521 | BUG_ON(!list_empty(&worker->entry) && | |
1522 | (worker->hentry.next || worker->hentry.pprev)); | |
1523 | ||
cb444766 TH |
1524 | /* can't use worker_set_flags(), also called from start_worker() */ |
1525 | worker->flags |= WORKER_IDLE; | |
bd7bdd43 | 1526 | pool->nr_idle++; |
e22bee78 | 1527 | worker->last_active = jiffies; |
c8e55f36 TH |
1528 | |
1529 | /* idle_list is LIFO */ | |
bd7bdd43 | 1530 | list_add(&worker->entry, &pool->idle_list); |
db7bccf4 | 1531 | |
628c78e7 TH |
1532 | if (too_many_workers(pool) && !timer_pending(&pool->idle_timer)) |
1533 | mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT); | |
cb444766 | 1534 | |
544ecf31 | 1535 | /* |
706026c2 | 1536 | * Sanity check nr_running. Because wq_unbind_fn() releases |
d565ed63 | 1537 | * pool->lock between setting %WORKER_UNBOUND and zapping |
628c78e7 TH |
1538 | * nr_running, the warning may trigger spuriously. Check iff |
1539 | * unbind is not in progress. | |
544ecf31 | 1540 | */ |
24647570 | 1541 | WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) && |
bd7bdd43 | 1542 | pool->nr_workers == pool->nr_idle && |
63d95a91 | 1543 | atomic_read(get_pool_nr_running(pool))); |
c8e55f36 TH |
1544 | } |
1545 | ||
1546 | /** | |
1547 | * worker_leave_idle - leave idle state | |
1548 | * @worker: worker which is leaving idle state | |
1549 | * | |
1550 | * @worker is leaving idle state. Update stats. | |
1551 | * | |
1552 | * LOCKING: | |
d565ed63 | 1553 | * spin_lock_irq(pool->lock). |
c8e55f36 TH |
1554 | */ |
1555 | static void worker_leave_idle(struct worker *worker) | |
1556 | { | |
bd7bdd43 | 1557 | struct worker_pool *pool = worker->pool; |
c8e55f36 TH |
1558 | |
1559 | BUG_ON(!(worker->flags & WORKER_IDLE)); | |
d302f017 | 1560 | worker_clr_flags(worker, WORKER_IDLE); |
bd7bdd43 | 1561 | pool->nr_idle--; |
c8e55f36 TH |
1562 | list_del_init(&worker->entry); |
1563 | } | |
1564 | ||
e22bee78 | 1565 | /** |
706026c2 | 1566 | * worker_maybe_bind_and_lock - bind worker to its cpu if possible and lock pool |
e22bee78 TH |
1567 | * @worker: self |
1568 | * | |
1569 | * Works which are scheduled while the cpu is online must at least be | |
1570 | * scheduled to a worker which is bound to the cpu so that if they are | |
1571 | * flushed from cpu callbacks while cpu is going down, they are | |
1572 | * guaranteed to execute on the cpu. | |
1573 | * | |
1574 | * This function is to be used by rogue workers and rescuers to bind | |
1575 | * themselves to the target cpu and may race with cpu going down or | |
1576 | * coming online. kthread_bind() can't be used because it may put the | |
1577 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
706026c2 | 1578 | * verbatim as it's best effort and blocking and pool may be |
e22bee78 TH |
1579 | * [dis]associated in the meantime. |
1580 | * | |
706026c2 | 1581 | * This function tries set_cpus_allowed() and locks pool and verifies the |
24647570 | 1582 | * binding against %POOL_DISASSOCIATED which is set during |
f2d5a0ee TH |
1583 | * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker |
1584 | * enters idle state or fetches works without dropping lock, it can | |
1585 | * guarantee the scheduling requirement described in the first paragraph. | |
e22bee78 TH |
1586 | * |
1587 | * CONTEXT: | |
d565ed63 | 1588 | * Might sleep. Called without any lock but returns with pool->lock |
e22bee78 TH |
1589 | * held. |
1590 | * | |
1591 | * RETURNS: | |
706026c2 | 1592 | * %true if the associated pool is online (@worker is successfully |
e22bee78 TH |
1593 | * bound), %false if offline. |
1594 | */ | |
1595 | static bool worker_maybe_bind_and_lock(struct worker *worker) | |
d565ed63 | 1596 | __acquires(&pool->lock) |
e22bee78 | 1597 | { |
24647570 | 1598 | struct worker_pool *pool = worker->pool; |
e22bee78 TH |
1599 | struct task_struct *task = worker->task; |
1600 | ||
1601 | while (true) { | |
4e6045f1 | 1602 | /* |
e22bee78 TH |
1603 | * The following call may fail, succeed or succeed |
1604 | * without actually migrating the task to the cpu if | |
1605 | * it races with cpu hotunplug operation. Verify | |
24647570 | 1606 | * against POOL_DISASSOCIATED. |
4e6045f1 | 1607 | */ |
24647570 | 1608 | if (!(pool->flags & POOL_DISASSOCIATED)) |
ec22ca5e | 1609 | set_cpus_allowed_ptr(task, get_cpu_mask(pool->cpu)); |
e22bee78 | 1610 | |
d565ed63 | 1611 | spin_lock_irq(&pool->lock); |
24647570 | 1612 | if (pool->flags & POOL_DISASSOCIATED) |
e22bee78 | 1613 | return false; |
ec22ca5e | 1614 | if (task_cpu(task) == pool->cpu && |
e22bee78 | 1615 | cpumask_equal(¤t->cpus_allowed, |
ec22ca5e | 1616 | get_cpu_mask(pool->cpu))) |
e22bee78 | 1617 | return true; |
d565ed63 | 1618 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1619 | |
5035b20f TH |
1620 | /* |
1621 | * We've raced with CPU hot[un]plug. Give it a breather | |
1622 | * and retry migration. cond_resched() is required here; | |
1623 | * otherwise, we might deadlock against cpu_stop trying to | |
1624 | * bring down the CPU on non-preemptive kernel. | |
1625 | */ | |
e22bee78 | 1626 | cpu_relax(); |
5035b20f | 1627 | cond_resched(); |
e22bee78 TH |
1628 | } |
1629 | } | |
1630 | ||
25511a47 | 1631 | /* |
ea1abd61 | 1632 | * Rebind an idle @worker to its CPU. worker_thread() will test |
5f7dabfd | 1633 | * list_empty(@worker->entry) before leaving idle and call this function. |
25511a47 TH |
1634 | */ |
1635 | static void idle_worker_rebind(struct worker *worker) | |
1636 | { | |
5f7dabfd LJ |
1637 | /* CPU may go down again inbetween, clear UNBOUND only on success */ |
1638 | if (worker_maybe_bind_and_lock(worker)) | |
1639 | worker_clr_flags(worker, WORKER_UNBOUND); | |
25511a47 | 1640 | |
ea1abd61 LJ |
1641 | /* rebind complete, become available again */ |
1642 | list_add(&worker->entry, &worker->pool->idle_list); | |
d565ed63 | 1643 | spin_unlock_irq(&worker->pool->lock); |
25511a47 TH |
1644 | } |
1645 | ||
e22bee78 | 1646 | /* |
25511a47 | 1647 | * Function for @worker->rebind.work used to rebind unbound busy workers to |
403c821d TH |
1648 | * the associated cpu which is coming back online. This is scheduled by |
1649 | * cpu up but can race with other cpu hotplug operations and may be | |
1650 | * executed twice without intervening cpu down. | |
e22bee78 | 1651 | */ |
25511a47 | 1652 | static void busy_worker_rebind_fn(struct work_struct *work) |
e22bee78 TH |
1653 | { |
1654 | struct worker *worker = container_of(work, struct worker, rebind_work); | |
e22bee78 | 1655 | |
eab6d828 LJ |
1656 | if (worker_maybe_bind_and_lock(worker)) |
1657 | worker_clr_flags(worker, WORKER_UNBOUND); | |
e22bee78 | 1658 | |
d565ed63 | 1659 | spin_unlock_irq(&worker->pool->lock); |
e22bee78 TH |
1660 | } |
1661 | ||
25511a47 | 1662 | /** |
94cf58bb TH |
1663 | * rebind_workers - rebind all workers of a pool to the associated CPU |
1664 | * @pool: pool of interest | |
25511a47 | 1665 | * |
94cf58bb | 1666 | * @pool->cpu is coming online. Rebind all workers to the CPU. Rebinding |
25511a47 TH |
1667 | * is different for idle and busy ones. |
1668 | * | |
ea1abd61 LJ |
1669 | * Idle ones will be removed from the idle_list and woken up. They will |
1670 | * add themselves back after completing rebind. This ensures that the | |
1671 | * idle_list doesn't contain any unbound workers when re-bound busy workers | |
1672 | * try to perform local wake-ups for concurrency management. | |
25511a47 | 1673 | * |
ea1abd61 LJ |
1674 | * Busy workers can rebind after they finish their current work items. |
1675 | * Queueing the rebind work item at the head of the scheduled list is | |
1676 | * enough. Note that nr_running will be properly bumped as busy workers | |
1677 | * rebind. | |
25511a47 | 1678 | * |
ea1abd61 LJ |
1679 | * On return, all non-manager workers are scheduled for rebind - see |
1680 | * manage_workers() for the manager special case. Any idle worker | |
1681 | * including the manager will not appear on @idle_list until rebind is | |
1682 | * complete, making local wake-ups safe. | |
25511a47 | 1683 | */ |
94cf58bb | 1684 | static void rebind_workers(struct worker_pool *pool) |
25511a47 | 1685 | { |
ea1abd61 | 1686 | struct worker *worker, *n; |
25511a47 TH |
1687 | struct hlist_node *pos; |
1688 | int i; | |
1689 | ||
94cf58bb TH |
1690 | lockdep_assert_held(&pool->assoc_mutex); |
1691 | lockdep_assert_held(&pool->lock); | |
25511a47 | 1692 | |
5f7dabfd | 1693 | /* dequeue and kick idle ones */ |
94cf58bb TH |
1694 | list_for_each_entry_safe(worker, n, &pool->idle_list, entry) { |
1695 | /* | |
1696 | * idle workers should be off @pool->idle_list until rebind | |
1697 | * is complete to avoid receiving premature local wake-ups. | |
1698 | */ | |
1699 | list_del_init(&worker->entry); | |
25511a47 | 1700 | |
94cf58bb TH |
1701 | /* |
1702 | * worker_thread() will see the above dequeuing and call | |
1703 | * idle_worker_rebind(). | |
1704 | */ | |
1705 | wake_up_process(worker->task); | |
1706 | } | |
25511a47 | 1707 | |
94cf58bb TH |
1708 | /* rebind busy workers */ |
1709 | for_each_busy_worker(worker, i, pos, pool) { | |
1710 | struct work_struct *rebind_work = &worker->rebind_work; | |
1711 | struct workqueue_struct *wq; | |
25511a47 | 1712 | |
94cf58bb TH |
1713 | if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, |
1714 | work_data_bits(rebind_work))) | |
1715 | continue; | |
25511a47 | 1716 | |
94cf58bb | 1717 | debug_work_activate(rebind_work); |
90beca5d | 1718 | |
94cf58bb TH |
1719 | /* |
1720 | * wq doesn't really matter but let's keep @worker->pool | |
1721 | * and @cwq->pool consistent for sanity. | |
1722 | */ | |
1723 | if (std_worker_pool_pri(worker->pool)) | |
1724 | wq = system_highpri_wq; | |
1725 | else | |
1726 | wq = system_wq; | |
1727 | ||
1728 | insert_work(get_cwq(pool->cpu, wq), rebind_work, | |
1729 | worker->scheduled.next, | |
1730 | work_color_to_flags(WORK_NO_COLOR)); | |
ec58815a | 1731 | } |
25511a47 TH |
1732 | } |
1733 | ||
c34056a3 TH |
1734 | static struct worker *alloc_worker(void) |
1735 | { | |
1736 | struct worker *worker; | |
1737 | ||
1738 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1739 | if (worker) { |
1740 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1741 | INIT_LIST_HEAD(&worker->scheduled); |
25511a47 | 1742 | INIT_WORK(&worker->rebind_work, busy_worker_rebind_fn); |
e22bee78 TH |
1743 | /* on creation a worker is in !idle && prep state */ |
1744 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1745 | } |
c34056a3 TH |
1746 | return worker; |
1747 | } | |
1748 | ||
1749 | /** | |
1750 | * create_worker - create a new workqueue worker | |
63d95a91 | 1751 | * @pool: pool the new worker will belong to |
c34056a3 | 1752 | * |
63d95a91 | 1753 | * Create a new worker which is bound to @pool. The returned worker |
c34056a3 TH |
1754 | * can be started by calling start_worker() or destroyed using |
1755 | * destroy_worker(). | |
1756 | * | |
1757 | * CONTEXT: | |
1758 | * Might sleep. Does GFP_KERNEL allocations. | |
1759 | * | |
1760 | * RETURNS: | |
1761 | * Pointer to the newly created worker. | |
1762 | */ | |
bc2ae0f5 | 1763 | static struct worker *create_worker(struct worker_pool *pool) |
c34056a3 | 1764 | { |
e34cdddb | 1765 | const char *pri = std_worker_pool_pri(pool) ? "H" : ""; |
c34056a3 | 1766 | struct worker *worker = NULL; |
f3421797 | 1767 | int id = -1; |
c34056a3 | 1768 | |
d565ed63 | 1769 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1770 | while (ida_get_new(&pool->worker_ida, &id)) { |
d565ed63 | 1771 | spin_unlock_irq(&pool->lock); |
bd7bdd43 | 1772 | if (!ida_pre_get(&pool->worker_ida, GFP_KERNEL)) |
c34056a3 | 1773 | goto fail; |
d565ed63 | 1774 | spin_lock_irq(&pool->lock); |
c34056a3 | 1775 | } |
d565ed63 | 1776 | spin_unlock_irq(&pool->lock); |
c34056a3 TH |
1777 | |
1778 | worker = alloc_worker(); | |
1779 | if (!worker) | |
1780 | goto fail; | |
1781 | ||
bd7bdd43 | 1782 | worker->pool = pool; |
c34056a3 TH |
1783 | worker->id = id; |
1784 | ||
ec22ca5e | 1785 | if (pool->cpu != WORK_CPU_UNBOUND) |
94dcf29a | 1786 | worker->task = kthread_create_on_node(worker_thread, |
ec22ca5e TH |
1787 | worker, cpu_to_node(pool->cpu), |
1788 | "kworker/%u:%d%s", pool->cpu, id, pri); | |
f3421797 TH |
1789 | else |
1790 | worker->task = kthread_create(worker_thread, worker, | |
3270476a | 1791 | "kworker/u:%d%s", id, pri); |
c34056a3 TH |
1792 | if (IS_ERR(worker->task)) |
1793 | goto fail; | |
1794 | ||
e34cdddb | 1795 | if (std_worker_pool_pri(pool)) |
3270476a TH |
1796 | set_user_nice(worker->task, HIGHPRI_NICE_LEVEL); |
1797 | ||
db7bccf4 | 1798 | /* |
bc2ae0f5 | 1799 | * Determine CPU binding of the new worker depending on |
24647570 | 1800 | * %POOL_DISASSOCIATED. The caller is responsible for ensuring the |
bc2ae0f5 TH |
1801 | * flag remains stable across this function. See the comments |
1802 | * above the flag definition for details. | |
1803 | * | |
1804 | * As an unbound worker may later become a regular one if CPU comes | |
1805 | * online, make sure every worker has %PF_THREAD_BOUND set. | |
db7bccf4 | 1806 | */ |
24647570 | 1807 | if (!(pool->flags & POOL_DISASSOCIATED)) { |
ec22ca5e | 1808 | kthread_bind(worker->task, pool->cpu); |
bc2ae0f5 | 1809 | } else { |
db7bccf4 | 1810 | worker->task->flags |= PF_THREAD_BOUND; |
bc2ae0f5 | 1811 | worker->flags |= WORKER_UNBOUND; |
f3421797 | 1812 | } |
c34056a3 TH |
1813 | |
1814 | return worker; | |
1815 | fail: | |
1816 | if (id >= 0) { | |
d565ed63 | 1817 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1818 | ida_remove(&pool->worker_ida, id); |
d565ed63 | 1819 | spin_unlock_irq(&pool->lock); |
c34056a3 TH |
1820 | } |
1821 | kfree(worker); | |
1822 | return NULL; | |
1823 | } | |
1824 | ||
1825 | /** | |
1826 | * start_worker - start a newly created worker | |
1827 | * @worker: worker to start | |
1828 | * | |
706026c2 | 1829 | * Make the pool aware of @worker and start it. |
c34056a3 TH |
1830 | * |
1831 | * CONTEXT: | |
d565ed63 | 1832 | * spin_lock_irq(pool->lock). |
c34056a3 TH |
1833 | */ |
1834 | static void start_worker(struct worker *worker) | |
1835 | { | |
cb444766 | 1836 | worker->flags |= WORKER_STARTED; |
bd7bdd43 | 1837 | worker->pool->nr_workers++; |
c8e55f36 | 1838 | worker_enter_idle(worker); |
c34056a3 TH |
1839 | wake_up_process(worker->task); |
1840 | } | |
1841 | ||
1842 | /** | |
1843 | * destroy_worker - destroy a workqueue worker | |
1844 | * @worker: worker to be destroyed | |
1845 | * | |
706026c2 | 1846 | * Destroy @worker and adjust @pool stats accordingly. |
c8e55f36 TH |
1847 | * |
1848 | * CONTEXT: | |
d565ed63 | 1849 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
c34056a3 TH |
1850 | */ |
1851 | static void destroy_worker(struct worker *worker) | |
1852 | { | |
bd7bdd43 | 1853 | struct worker_pool *pool = worker->pool; |
c34056a3 TH |
1854 | int id = worker->id; |
1855 | ||
1856 | /* sanity check frenzy */ | |
1857 | BUG_ON(worker->current_work); | |
affee4b2 | 1858 | BUG_ON(!list_empty(&worker->scheduled)); |
c34056a3 | 1859 | |
c8e55f36 | 1860 | if (worker->flags & WORKER_STARTED) |
bd7bdd43 | 1861 | pool->nr_workers--; |
c8e55f36 | 1862 | if (worker->flags & WORKER_IDLE) |
bd7bdd43 | 1863 | pool->nr_idle--; |
c8e55f36 TH |
1864 | |
1865 | list_del_init(&worker->entry); | |
cb444766 | 1866 | worker->flags |= WORKER_DIE; |
c8e55f36 | 1867 | |
d565ed63 | 1868 | spin_unlock_irq(&pool->lock); |
c8e55f36 | 1869 | |
c34056a3 TH |
1870 | kthread_stop(worker->task); |
1871 | kfree(worker); | |
1872 | ||
d565ed63 | 1873 | spin_lock_irq(&pool->lock); |
bd7bdd43 | 1874 | ida_remove(&pool->worker_ida, id); |
c34056a3 TH |
1875 | } |
1876 | ||
63d95a91 | 1877 | static void idle_worker_timeout(unsigned long __pool) |
e22bee78 | 1878 | { |
63d95a91 | 1879 | struct worker_pool *pool = (void *)__pool; |
e22bee78 | 1880 | |
d565ed63 | 1881 | spin_lock_irq(&pool->lock); |
e22bee78 | 1882 | |
63d95a91 | 1883 | if (too_many_workers(pool)) { |
e22bee78 TH |
1884 | struct worker *worker; |
1885 | unsigned long expires; | |
1886 | ||
1887 | /* idle_list is kept in LIFO order, check the last one */ | |
63d95a91 | 1888 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 TH |
1889 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
1890 | ||
1891 | if (time_before(jiffies, expires)) | |
63d95a91 | 1892 | mod_timer(&pool->idle_timer, expires); |
e22bee78 TH |
1893 | else { |
1894 | /* it's been idle for too long, wake up manager */ | |
11ebea50 | 1895 | pool->flags |= POOL_MANAGE_WORKERS; |
63d95a91 | 1896 | wake_up_worker(pool); |
d5abe669 | 1897 | } |
e22bee78 TH |
1898 | } |
1899 | ||
d565ed63 | 1900 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1901 | } |
d5abe669 | 1902 | |
e22bee78 TH |
1903 | static bool send_mayday(struct work_struct *work) |
1904 | { | |
1905 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); | |
1906 | struct workqueue_struct *wq = cwq->wq; | |
f3421797 | 1907 | unsigned int cpu; |
e22bee78 TH |
1908 | |
1909 | if (!(wq->flags & WQ_RESCUER)) | |
1910 | return false; | |
1911 | ||
1912 | /* mayday mayday mayday */ | |
ec22ca5e | 1913 | cpu = cwq->pool->cpu; |
f3421797 TH |
1914 | /* WORK_CPU_UNBOUND can't be set in cpumask, use cpu 0 instead */ |
1915 | if (cpu == WORK_CPU_UNBOUND) | |
1916 | cpu = 0; | |
f2e005aa | 1917 | if (!mayday_test_and_set_cpu(cpu, wq->mayday_mask)) |
e22bee78 TH |
1918 | wake_up_process(wq->rescuer->task); |
1919 | return true; | |
1920 | } | |
1921 | ||
706026c2 | 1922 | static void pool_mayday_timeout(unsigned long __pool) |
e22bee78 | 1923 | { |
63d95a91 | 1924 | struct worker_pool *pool = (void *)__pool; |
e22bee78 TH |
1925 | struct work_struct *work; |
1926 | ||
d565ed63 | 1927 | spin_lock_irq(&pool->lock); |
e22bee78 | 1928 | |
63d95a91 | 1929 | if (need_to_create_worker(pool)) { |
e22bee78 TH |
1930 | /* |
1931 | * We've been trying to create a new worker but | |
1932 | * haven't been successful. We might be hitting an | |
1933 | * allocation deadlock. Send distress signals to | |
1934 | * rescuers. | |
1935 | */ | |
63d95a91 | 1936 | list_for_each_entry(work, &pool->worklist, entry) |
e22bee78 | 1937 | send_mayday(work); |
1da177e4 | 1938 | } |
e22bee78 | 1939 | |
d565ed63 | 1940 | spin_unlock_irq(&pool->lock); |
e22bee78 | 1941 | |
63d95a91 | 1942 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL); |
1da177e4 LT |
1943 | } |
1944 | ||
e22bee78 TH |
1945 | /** |
1946 | * maybe_create_worker - create a new worker if necessary | |
63d95a91 | 1947 | * @pool: pool to create a new worker for |
e22bee78 | 1948 | * |
63d95a91 | 1949 | * Create a new worker for @pool if necessary. @pool is guaranteed to |
e22bee78 TH |
1950 | * have at least one idle worker on return from this function. If |
1951 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
63d95a91 | 1952 | * sent to all rescuers with works scheduled on @pool to resolve |
e22bee78 TH |
1953 | * possible allocation deadlock. |
1954 | * | |
1955 | * On return, need_to_create_worker() is guaranteed to be false and | |
1956 | * may_start_working() true. | |
1957 | * | |
1958 | * LOCKING: | |
d565ed63 | 1959 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
1960 | * multiple times. Does GFP_KERNEL allocations. Called only from |
1961 | * manager. | |
1962 | * | |
1963 | * RETURNS: | |
d565ed63 | 1964 | * false if no action was taken and pool->lock stayed locked, true |
e22bee78 TH |
1965 | * otherwise. |
1966 | */ | |
63d95a91 | 1967 | static bool maybe_create_worker(struct worker_pool *pool) |
d565ed63 TH |
1968 | __releases(&pool->lock) |
1969 | __acquires(&pool->lock) | |
1da177e4 | 1970 | { |
63d95a91 | 1971 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1972 | return false; |
1973 | restart: | |
d565ed63 | 1974 | spin_unlock_irq(&pool->lock); |
9f9c2364 | 1975 | |
e22bee78 | 1976 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
63d95a91 | 1977 | mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); |
e22bee78 TH |
1978 | |
1979 | while (true) { | |
1980 | struct worker *worker; | |
1981 | ||
bc2ae0f5 | 1982 | worker = create_worker(pool); |
e22bee78 | 1983 | if (worker) { |
63d95a91 | 1984 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 1985 | spin_lock_irq(&pool->lock); |
e22bee78 | 1986 | start_worker(worker); |
63d95a91 | 1987 | BUG_ON(need_to_create_worker(pool)); |
e22bee78 TH |
1988 | return true; |
1989 | } | |
1990 | ||
63d95a91 | 1991 | if (!need_to_create_worker(pool)) |
e22bee78 | 1992 | break; |
1da177e4 | 1993 | |
e22bee78 TH |
1994 | __set_current_state(TASK_INTERRUPTIBLE); |
1995 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1996 | |
63d95a91 | 1997 | if (!need_to_create_worker(pool)) |
e22bee78 TH |
1998 | break; |
1999 | } | |
2000 | ||
63d95a91 | 2001 | del_timer_sync(&pool->mayday_timer); |
d565ed63 | 2002 | spin_lock_irq(&pool->lock); |
63d95a91 | 2003 | if (need_to_create_worker(pool)) |
e22bee78 TH |
2004 | goto restart; |
2005 | return true; | |
2006 | } | |
2007 | ||
2008 | /** | |
2009 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
63d95a91 | 2010 | * @pool: pool to destroy workers for |
e22bee78 | 2011 | * |
63d95a91 | 2012 | * Destroy @pool workers which have been idle for longer than |
e22bee78 TH |
2013 | * IDLE_WORKER_TIMEOUT. |
2014 | * | |
2015 | * LOCKING: | |
d565ed63 | 2016 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
2017 | * multiple times. Called only from manager. |
2018 | * | |
2019 | * RETURNS: | |
d565ed63 | 2020 | * false if no action was taken and pool->lock stayed locked, true |
e22bee78 TH |
2021 | * otherwise. |
2022 | */ | |
63d95a91 | 2023 | static bool maybe_destroy_workers(struct worker_pool *pool) |
e22bee78 TH |
2024 | { |
2025 | bool ret = false; | |
1da177e4 | 2026 | |
63d95a91 | 2027 | while (too_many_workers(pool)) { |
e22bee78 TH |
2028 | struct worker *worker; |
2029 | unsigned long expires; | |
3af24433 | 2030 | |
63d95a91 | 2031 | worker = list_entry(pool->idle_list.prev, struct worker, entry); |
e22bee78 | 2032 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; |
85f4186a | 2033 | |
e22bee78 | 2034 | if (time_before(jiffies, expires)) { |
63d95a91 | 2035 | mod_timer(&pool->idle_timer, expires); |
3af24433 | 2036 | break; |
e22bee78 | 2037 | } |
1da177e4 | 2038 | |
e22bee78 TH |
2039 | destroy_worker(worker); |
2040 | ret = true; | |
1da177e4 | 2041 | } |
1e19ffc6 | 2042 | |
e22bee78 | 2043 | return ret; |
1e19ffc6 TH |
2044 | } |
2045 | ||
73f53c4a | 2046 | /** |
e22bee78 TH |
2047 | * manage_workers - manage worker pool |
2048 | * @worker: self | |
73f53c4a | 2049 | * |
706026c2 | 2050 | * Assume the manager role and manage the worker pool @worker belongs |
e22bee78 | 2051 | * to. At any given time, there can be only zero or one manager per |
706026c2 | 2052 | * pool. The exclusion is handled automatically by this function. |
e22bee78 TH |
2053 | * |
2054 | * The caller can safely start processing works on false return. On | |
2055 | * true return, it's guaranteed that need_to_create_worker() is false | |
2056 | * and may_start_working() is true. | |
73f53c4a TH |
2057 | * |
2058 | * CONTEXT: | |
d565ed63 | 2059 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
e22bee78 TH |
2060 | * multiple times. Does GFP_KERNEL allocations. |
2061 | * | |
2062 | * RETURNS: | |
d565ed63 TH |
2063 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
2064 | * multiple times. Does GFP_KERNEL allocations. | |
73f53c4a | 2065 | */ |
e22bee78 | 2066 | static bool manage_workers(struct worker *worker) |
73f53c4a | 2067 | { |
63d95a91 | 2068 | struct worker_pool *pool = worker->pool; |
e22bee78 | 2069 | bool ret = false; |
73f53c4a | 2070 | |
ee378aa4 | 2071 | if (pool->flags & POOL_MANAGING_WORKERS) |
e22bee78 | 2072 | return ret; |
1e19ffc6 | 2073 | |
552a37e9 | 2074 | pool->flags |= POOL_MANAGING_WORKERS; |
73f53c4a | 2075 | |
ee378aa4 LJ |
2076 | /* |
2077 | * To simplify both worker management and CPU hotplug, hold off | |
2078 | * management while hotplug is in progress. CPU hotplug path can't | |
2079 | * grab %POOL_MANAGING_WORKERS to achieve this because that can | |
2080 | * lead to idle worker depletion (all become busy thinking someone | |
2081 | * else is managing) which in turn can result in deadlock under | |
b2eb83d1 | 2082 | * extreme circumstances. Use @pool->assoc_mutex to synchronize |
ee378aa4 LJ |
2083 | * manager against CPU hotplug. |
2084 | * | |
b2eb83d1 | 2085 | * assoc_mutex would always be free unless CPU hotplug is in |
d565ed63 | 2086 | * progress. trylock first without dropping @pool->lock. |
ee378aa4 | 2087 | */ |
b2eb83d1 | 2088 | if (unlikely(!mutex_trylock(&pool->assoc_mutex))) { |
d565ed63 | 2089 | spin_unlock_irq(&pool->lock); |
b2eb83d1 | 2090 | mutex_lock(&pool->assoc_mutex); |
ee378aa4 LJ |
2091 | /* |
2092 | * CPU hotplug could have happened while we were waiting | |
b2eb83d1 | 2093 | * for assoc_mutex. Hotplug itself can't handle us |
ee378aa4 | 2094 | * because manager isn't either on idle or busy list, and |
706026c2 | 2095 | * @pool's state and ours could have deviated. |
ee378aa4 | 2096 | * |
b2eb83d1 | 2097 | * As hotplug is now excluded via assoc_mutex, we can |
ee378aa4 | 2098 | * simply try to bind. It will succeed or fail depending |
706026c2 | 2099 | * on @pool's current state. Try it and adjust |
ee378aa4 LJ |
2100 | * %WORKER_UNBOUND accordingly. |
2101 | */ | |
2102 | if (worker_maybe_bind_and_lock(worker)) | |
2103 | worker->flags &= ~WORKER_UNBOUND; | |
2104 | else | |
2105 | worker->flags |= WORKER_UNBOUND; | |
73f53c4a | 2106 | |
ee378aa4 LJ |
2107 | ret = true; |
2108 | } | |
73f53c4a | 2109 | |
11ebea50 | 2110 | pool->flags &= ~POOL_MANAGE_WORKERS; |
73f53c4a TH |
2111 | |
2112 | /* | |
e22bee78 TH |
2113 | * Destroy and then create so that may_start_working() is true |
2114 | * on return. | |
73f53c4a | 2115 | */ |
63d95a91 TH |
2116 | ret |= maybe_destroy_workers(pool); |
2117 | ret |= maybe_create_worker(pool); | |
e22bee78 | 2118 | |
552a37e9 | 2119 | pool->flags &= ~POOL_MANAGING_WORKERS; |
b2eb83d1 | 2120 | mutex_unlock(&pool->assoc_mutex); |
e22bee78 | 2121 | return ret; |
73f53c4a TH |
2122 | } |
2123 | ||
a62428c0 TH |
2124 | /** |
2125 | * process_one_work - process single work | |
c34056a3 | 2126 | * @worker: self |
a62428c0 TH |
2127 | * @work: work to process |
2128 | * | |
2129 | * Process @work. This function contains all the logics necessary to | |
2130 | * process a single work including synchronization against and | |
2131 | * interaction with other workers on the same cpu, queueing and | |
2132 | * flushing. As long as context requirement is met, any worker can | |
2133 | * call this function to process a work. | |
2134 | * | |
2135 | * CONTEXT: | |
d565ed63 | 2136 | * spin_lock_irq(pool->lock) which is released and regrabbed. |
a62428c0 | 2137 | */ |
c34056a3 | 2138 | static void process_one_work(struct worker *worker, struct work_struct *work) |
d565ed63 TH |
2139 | __releases(&pool->lock) |
2140 | __acquires(&pool->lock) | |
a62428c0 | 2141 | { |
7e11629d | 2142 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); |
bd7bdd43 | 2143 | struct worker_pool *pool = worker->pool; |
fb0e7beb | 2144 | bool cpu_intensive = cwq->wq->flags & WQ_CPU_INTENSIVE; |
73f53c4a | 2145 | int work_color; |
7e11629d | 2146 | struct worker *collision; |
a62428c0 TH |
2147 | #ifdef CONFIG_LOCKDEP |
2148 | /* | |
2149 | * It is permissible to free the struct work_struct from | |
2150 | * inside the function that is called from it, this we need to | |
2151 | * take into account for lockdep too. To avoid bogus "held | |
2152 | * lock freed" warnings as well as problems when looking into | |
2153 | * work->lockdep_map, make a copy and use that here. | |
2154 | */ | |
4d82a1de PZ |
2155 | struct lockdep_map lockdep_map; |
2156 | ||
2157 | lockdep_copy_map(&lockdep_map, &work->lockdep_map); | |
a62428c0 | 2158 | #endif |
6fec10a1 TH |
2159 | /* |
2160 | * Ensure we're on the correct CPU. DISASSOCIATED test is | |
2161 | * necessary to avoid spurious warnings from rescuers servicing the | |
24647570 | 2162 | * unbound or a disassociated pool. |
6fec10a1 | 2163 | */ |
5f7dabfd | 2164 | WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) && |
24647570 | 2165 | !(pool->flags & POOL_DISASSOCIATED) && |
ec22ca5e | 2166 | raw_smp_processor_id() != pool->cpu); |
25511a47 | 2167 | |
7e11629d TH |
2168 | /* |
2169 | * A single work shouldn't be executed concurrently by | |
2170 | * multiple workers on a single cpu. Check whether anyone is | |
2171 | * already processing the work. If so, defer the work to the | |
2172 | * currently executing one. | |
2173 | */ | |
c9e7cf27 | 2174 | collision = find_worker_executing_work(pool, work); |
7e11629d TH |
2175 | if (unlikely(collision)) { |
2176 | move_linked_works(work, &collision->scheduled, NULL); | |
2177 | return; | |
2178 | } | |
2179 | ||
8930caba | 2180 | /* claim and dequeue */ |
a62428c0 | 2181 | debug_work_deactivate(work); |
c9e7cf27 | 2182 | hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work); |
c34056a3 | 2183 | worker->current_work = work; |
a2c1c57b | 2184 | worker->current_func = work->func; |
8cca0eea | 2185 | worker->current_cwq = cwq; |
73f53c4a | 2186 | work_color = get_work_color(work); |
7a22ad75 | 2187 | |
a62428c0 TH |
2188 | list_del_init(&work->entry); |
2189 | ||
fb0e7beb TH |
2190 | /* |
2191 | * CPU intensive works don't participate in concurrency | |
2192 | * management. They're the scheduler's responsibility. | |
2193 | */ | |
2194 | if (unlikely(cpu_intensive)) | |
2195 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
2196 | ||
974271c4 | 2197 | /* |
d565ed63 | 2198 | * Unbound pool isn't concurrency managed and work items should be |
974271c4 TH |
2199 | * executed ASAP. Wake up another worker if necessary. |
2200 | */ | |
63d95a91 TH |
2201 | if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) |
2202 | wake_up_worker(pool); | |
974271c4 | 2203 | |
8930caba | 2204 | /* |
7c3eed5c | 2205 | * Record the last pool and clear PENDING which should be the last |
d565ed63 | 2206 | * update to @work. Also, do this inside @pool->lock so that |
23657bb1 TH |
2207 | * PENDING and queued state changes happen together while IRQ is |
2208 | * disabled. | |
8930caba | 2209 | */ |
7c3eed5c | 2210 | set_work_pool_and_clear_pending(work, pool->id); |
a62428c0 | 2211 | |
d565ed63 | 2212 | spin_unlock_irq(&pool->lock); |
a62428c0 | 2213 | |
e159489b | 2214 | lock_map_acquire_read(&cwq->wq->lockdep_map); |
a62428c0 | 2215 | lock_map_acquire(&lockdep_map); |
e36c886a | 2216 | trace_workqueue_execute_start(work); |
a2c1c57b | 2217 | worker->current_func(work); |
e36c886a AV |
2218 | /* |
2219 | * While we must be careful to not use "work" after this, the trace | |
2220 | * point will only record its address. | |
2221 | */ | |
2222 | trace_workqueue_execute_end(work); | |
a62428c0 TH |
2223 | lock_map_release(&lockdep_map); |
2224 | lock_map_release(&cwq->wq->lockdep_map); | |
2225 | ||
2226 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
044c782c VI |
2227 | pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n" |
2228 | " last function: %pf\n", | |
a2c1c57b TH |
2229 | current->comm, preempt_count(), task_pid_nr(current), |
2230 | worker->current_func); | |
a62428c0 TH |
2231 | debug_show_held_locks(current); |
2232 | dump_stack(); | |
2233 | } | |
2234 | ||
d565ed63 | 2235 | spin_lock_irq(&pool->lock); |
a62428c0 | 2236 | |
fb0e7beb TH |
2237 | /* clear cpu intensive status */ |
2238 | if (unlikely(cpu_intensive)) | |
2239 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
2240 | ||
a62428c0 | 2241 | /* we're done with it, release */ |
42f8570f | 2242 | hash_del(&worker->hentry); |
c34056a3 | 2243 | worker->current_work = NULL; |
a2c1c57b | 2244 | worker->current_func = NULL; |
8cca0eea | 2245 | worker->current_cwq = NULL; |
b3f9f405 | 2246 | cwq_dec_nr_in_flight(cwq, work_color); |
a62428c0 TH |
2247 | } |
2248 | ||
affee4b2 TH |
2249 | /** |
2250 | * process_scheduled_works - process scheduled works | |
2251 | * @worker: self | |
2252 | * | |
2253 | * Process all scheduled works. Please note that the scheduled list | |
2254 | * may change while processing a work, so this function repeatedly | |
2255 | * fetches a work from the top and executes it. | |
2256 | * | |
2257 | * CONTEXT: | |
d565ed63 | 2258 | * spin_lock_irq(pool->lock) which may be released and regrabbed |
affee4b2 TH |
2259 | * multiple times. |
2260 | */ | |
2261 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 2262 | { |
affee4b2 TH |
2263 | while (!list_empty(&worker->scheduled)) { |
2264 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 2265 | struct work_struct, entry); |
c34056a3 | 2266 | process_one_work(worker, work); |
1da177e4 | 2267 | } |
1da177e4 LT |
2268 | } |
2269 | ||
4690c4ab TH |
2270 | /** |
2271 | * worker_thread - the worker thread function | |
c34056a3 | 2272 | * @__worker: self |
4690c4ab | 2273 | * |
706026c2 TH |
2274 | * The worker thread function. There are NR_CPU_WORKER_POOLS dynamic pools |
2275 | * of these per each cpu. These workers process all works regardless of | |
e22bee78 TH |
2276 | * their specific target workqueue. The only exception is works which |
2277 | * belong to workqueues with a rescuer which will be explained in | |
2278 | * rescuer_thread(). | |
4690c4ab | 2279 | */ |
c34056a3 | 2280 | static int worker_thread(void *__worker) |
1da177e4 | 2281 | { |
c34056a3 | 2282 | struct worker *worker = __worker; |
bd7bdd43 | 2283 | struct worker_pool *pool = worker->pool; |
1da177e4 | 2284 | |
e22bee78 TH |
2285 | /* tell the scheduler that this is a workqueue worker */ |
2286 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 2287 | woke_up: |
d565ed63 | 2288 | spin_lock_irq(&pool->lock); |
1da177e4 | 2289 | |
5f7dabfd LJ |
2290 | /* we are off idle list if destruction or rebind is requested */ |
2291 | if (unlikely(list_empty(&worker->entry))) { | |
d565ed63 | 2292 | spin_unlock_irq(&pool->lock); |
25511a47 | 2293 | |
5f7dabfd | 2294 | /* if DIE is set, destruction is requested */ |
25511a47 TH |
2295 | if (worker->flags & WORKER_DIE) { |
2296 | worker->task->flags &= ~PF_WQ_WORKER; | |
2297 | return 0; | |
2298 | } | |
2299 | ||
5f7dabfd | 2300 | /* otherwise, rebind */ |
25511a47 TH |
2301 | idle_worker_rebind(worker); |
2302 | goto woke_up; | |
c8e55f36 | 2303 | } |
affee4b2 | 2304 | |
c8e55f36 | 2305 | worker_leave_idle(worker); |
db7bccf4 | 2306 | recheck: |
e22bee78 | 2307 | /* no more worker necessary? */ |
63d95a91 | 2308 | if (!need_more_worker(pool)) |
e22bee78 TH |
2309 | goto sleep; |
2310 | ||
2311 | /* do we need to manage? */ | |
63d95a91 | 2312 | if (unlikely(!may_start_working(pool)) && manage_workers(worker)) |
e22bee78 TH |
2313 | goto recheck; |
2314 | ||
c8e55f36 TH |
2315 | /* |
2316 | * ->scheduled list can only be filled while a worker is | |
2317 | * preparing to process a work or actually processing it. | |
2318 | * Make sure nobody diddled with it while I was sleeping. | |
2319 | */ | |
2320 | BUG_ON(!list_empty(&worker->scheduled)); | |
2321 | ||
e22bee78 TH |
2322 | /* |
2323 | * When control reaches this point, we're guaranteed to have | |
2324 | * at least one idle worker or that someone else has already | |
2325 | * assumed the manager role. | |
2326 | */ | |
2327 | worker_clr_flags(worker, WORKER_PREP); | |
2328 | ||
2329 | do { | |
c8e55f36 | 2330 | struct work_struct *work = |
bd7bdd43 | 2331 | list_first_entry(&pool->worklist, |
c8e55f36 TH |
2332 | struct work_struct, entry); |
2333 | ||
2334 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
2335 | /* optimization path, not strictly necessary */ | |
2336 | process_one_work(worker, work); | |
2337 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 2338 | process_scheduled_works(worker); |
c8e55f36 TH |
2339 | } else { |
2340 | move_linked_works(work, &worker->scheduled, NULL); | |
2341 | process_scheduled_works(worker); | |
affee4b2 | 2342 | } |
63d95a91 | 2343 | } while (keep_working(pool)); |
e22bee78 TH |
2344 | |
2345 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 2346 | sleep: |
63d95a91 | 2347 | if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker)) |
e22bee78 | 2348 | goto recheck; |
d313dd85 | 2349 | |
c8e55f36 | 2350 | /* |
d565ed63 TH |
2351 | * pool->lock is held and there's no work to process and no need to |
2352 | * manage, sleep. Workers are woken up only while holding | |
2353 | * pool->lock or from local cpu, so setting the current state | |
2354 | * before releasing pool->lock is enough to prevent losing any | |
2355 | * event. | |
c8e55f36 TH |
2356 | */ |
2357 | worker_enter_idle(worker); | |
2358 | __set_current_state(TASK_INTERRUPTIBLE); | |
d565ed63 | 2359 | spin_unlock_irq(&pool->lock); |
c8e55f36 TH |
2360 | schedule(); |
2361 | goto woke_up; | |
1da177e4 LT |
2362 | } |
2363 | ||
e22bee78 TH |
2364 | /** |
2365 | * rescuer_thread - the rescuer thread function | |
111c225a | 2366 | * @__rescuer: self |
e22bee78 TH |
2367 | * |
2368 | * Workqueue rescuer thread function. There's one rescuer for each | |
2369 | * workqueue which has WQ_RESCUER set. | |
2370 | * | |
706026c2 | 2371 | * Regular work processing on a pool may block trying to create a new |
e22bee78 TH |
2372 | * worker which uses GFP_KERNEL allocation which has slight chance of |
2373 | * developing into deadlock if some works currently on the same queue | |
2374 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
2375 | * the problem rescuer solves. | |
2376 | * | |
706026c2 TH |
2377 | * When such condition is possible, the pool summons rescuers of all |
2378 | * workqueues which have works queued on the pool and let them process | |
e22bee78 TH |
2379 | * those works so that forward progress can be guaranteed. |
2380 | * | |
2381 | * This should happen rarely. | |
2382 | */ | |
111c225a | 2383 | static int rescuer_thread(void *__rescuer) |
e22bee78 | 2384 | { |
111c225a TH |
2385 | struct worker *rescuer = __rescuer; |
2386 | struct workqueue_struct *wq = rescuer->rescue_wq; | |
e22bee78 | 2387 | struct list_head *scheduled = &rescuer->scheduled; |
f3421797 | 2388 | bool is_unbound = wq->flags & WQ_UNBOUND; |
e22bee78 TH |
2389 | unsigned int cpu; |
2390 | ||
2391 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
111c225a TH |
2392 | |
2393 | /* | |
2394 | * Mark rescuer as worker too. As WORKER_PREP is never cleared, it | |
2395 | * doesn't participate in concurrency management. | |
2396 | */ | |
2397 | rescuer->task->flags |= PF_WQ_WORKER; | |
e22bee78 TH |
2398 | repeat: |
2399 | set_current_state(TASK_INTERRUPTIBLE); | |
2400 | ||
412d32e6 MG |
2401 | if (kthread_should_stop()) { |
2402 | __set_current_state(TASK_RUNNING); | |
111c225a | 2403 | rescuer->task->flags &= ~PF_WQ_WORKER; |
e22bee78 | 2404 | return 0; |
412d32e6 | 2405 | } |
e22bee78 | 2406 | |
f3421797 TH |
2407 | /* |
2408 | * See whether any cpu is asking for help. Unbounded | |
2409 | * workqueues use cpu 0 in mayday_mask for CPU_UNBOUND. | |
2410 | */ | |
f2e005aa | 2411 | for_each_mayday_cpu(cpu, wq->mayday_mask) { |
f3421797 TH |
2412 | unsigned int tcpu = is_unbound ? WORK_CPU_UNBOUND : cpu; |
2413 | struct cpu_workqueue_struct *cwq = get_cwq(tcpu, wq); | |
bd7bdd43 | 2414 | struct worker_pool *pool = cwq->pool; |
e22bee78 TH |
2415 | struct work_struct *work, *n; |
2416 | ||
2417 | __set_current_state(TASK_RUNNING); | |
f2e005aa | 2418 | mayday_clear_cpu(cpu, wq->mayday_mask); |
e22bee78 TH |
2419 | |
2420 | /* migrate to the target cpu if possible */ | |
bd7bdd43 | 2421 | rescuer->pool = pool; |
e22bee78 TH |
2422 | worker_maybe_bind_and_lock(rescuer); |
2423 | ||
2424 | /* | |
2425 | * Slurp in all works issued via this workqueue and | |
2426 | * process'em. | |
2427 | */ | |
2428 | BUG_ON(!list_empty(&rescuer->scheduled)); | |
bd7bdd43 | 2429 | list_for_each_entry_safe(work, n, &pool->worklist, entry) |
e22bee78 TH |
2430 | if (get_work_cwq(work) == cwq) |
2431 | move_linked_works(work, scheduled, &n); | |
2432 | ||
2433 | process_scheduled_works(rescuer); | |
7576958a TH |
2434 | |
2435 | /* | |
d565ed63 | 2436 | * Leave this pool. If keep_working() is %true, notify a |
7576958a TH |
2437 | * regular worker; otherwise, we end up with 0 concurrency |
2438 | * and stalling the execution. | |
2439 | */ | |
63d95a91 TH |
2440 | if (keep_working(pool)) |
2441 | wake_up_worker(pool); | |
7576958a | 2442 | |
d565ed63 | 2443 | spin_unlock_irq(&pool->lock); |
e22bee78 TH |
2444 | } |
2445 | ||
111c225a TH |
2446 | /* rescuers should never participate in concurrency management */ |
2447 | WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING)); | |
e22bee78 TH |
2448 | schedule(); |
2449 | goto repeat; | |
1da177e4 LT |
2450 | } |
2451 | ||
fc2e4d70 ON |
2452 | struct wq_barrier { |
2453 | struct work_struct work; | |
2454 | struct completion done; | |
2455 | }; | |
2456 | ||
2457 | static void wq_barrier_func(struct work_struct *work) | |
2458 | { | |
2459 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
2460 | complete(&barr->done); | |
2461 | } | |
2462 | ||
4690c4ab TH |
2463 | /** |
2464 | * insert_wq_barrier - insert a barrier work | |
2465 | * @cwq: cwq to insert barrier into | |
2466 | * @barr: wq_barrier to insert | |
affee4b2 TH |
2467 | * @target: target work to attach @barr to |
2468 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 2469 | * |
affee4b2 TH |
2470 | * @barr is linked to @target such that @barr is completed only after |
2471 | * @target finishes execution. Please note that the ordering | |
2472 | * guarantee is observed only with respect to @target and on the local | |
2473 | * cpu. | |
2474 | * | |
2475 | * Currently, a queued barrier can't be canceled. This is because | |
2476 | * try_to_grab_pending() can't determine whether the work to be | |
2477 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2478 | * flag of the previous work while there must be a valid next work | |
2479 | * after a work with LINKED flag set. | |
2480 | * | |
2481 | * Note that when @worker is non-NULL, @target may be modified | |
2482 | * underneath us, so we can't reliably determine cwq from @target. | |
4690c4ab TH |
2483 | * |
2484 | * CONTEXT: | |
d565ed63 | 2485 | * spin_lock_irq(pool->lock). |
4690c4ab | 2486 | */ |
83c22520 | 2487 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, |
affee4b2 TH |
2488 | struct wq_barrier *barr, |
2489 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2490 | { |
affee4b2 TH |
2491 | struct list_head *head; |
2492 | unsigned int linked = 0; | |
2493 | ||
dc186ad7 | 2494 | /* |
d565ed63 | 2495 | * debugobject calls are safe here even with pool->lock locked |
dc186ad7 TG |
2496 | * as we know for sure that this will not trigger any of the |
2497 | * checks and call back into the fixup functions where we | |
2498 | * might deadlock. | |
2499 | */ | |
ca1cab37 | 2500 | INIT_WORK_ONSTACK(&barr->work, wq_barrier_func); |
22df02bb | 2501 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2502 | init_completion(&barr->done); |
83c22520 | 2503 | |
affee4b2 TH |
2504 | /* |
2505 | * If @target is currently being executed, schedule the | |
2506 | * barrier to the worker; otherwise, put it after @target. | |
2507 | */ | |
2508 | if (worker) | |
2509 | head = worker->scheduled.next; | |
2510 | else { | |
2511 | unsigned long *bits = work_data_bits(target); | |
2512 | ||
2513 | head = target->entry.next; | |
2514 | /* there can already be other linked works, inherit and set */ | |
2515 | linked = *bits & WORK_STRUCT_LINKED; | |
2516 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2517 | } | |
2518 | ||
dc186ad7 | 2519 | debug_work_activate(&barr->work); |
affee4b2 TH |
2520 | insert_work(cwq, &barr->work, head, |
2521 | work_color_to_flags(WORK_NO_COLOR) | linked); | |
fc2e4d70 ON |
2522 | } |
2523 | ||
73f53c4a TH |
2524 | /** |
2525 | * flush_workqueue_prep_cwqs - prepare cwqs for workqueue flushing | |
2526 | * @wq: workqueue being flushed | |
2527 | * @flush_color: new flush color, < 0 for no-op | |
2528 | * @work_color: new work color, < 0 for no-op | |
2529 | * | |
2530 | * Prepare cwqs for workqueue flushing. | |
2531 | * | |
2532 | * If @flush_color is non-negative, flush_color on all cwqs should be | |
2533 | * -1. If no cwq has in-flight commands at the specified color, all | |
2534 | * cwq->flush_color's stay at -1 and %false is returned. If any cwq | |
2535 | * has in flight commands, its cwq->flush_color is set to | |
2536 | * @flush_color, @wq->nr_cwqs_to_flush is updated accordingly, cwq | |
2537 | * wakeup logic is armed and %true is returned. | |
2538 | * | |
2539 | * The caller should have initialized @wq->first_flusher prior to | |
2540 | * calling this function with non-negative @flush_color. If | |
2541 | * @flush_color is negative, no flush color update is done and %false | |
2542 | * is returned. | |
2543 | * | |
2544 | * If @work_color is non-negative, all cwqs should have the same | |
2545 | * work_color which is previous to @work_color and all will be | |
2546 | * advanced to @work_color. | |
2547 | * | |
2548 | * CONTEXT: | |
2549 | * mutex_lock(wq->flush_mutex). | |
2550 | * | |
2551 | * RETURNS: | |
2552 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2553 | * otherwise. | |
2554 | */ | |
2555 | static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq, | |
2556 | int flush_color, int work_color) | |
1da177e4 | 2557 | { |
73f53c4a TH |
2558 | bool wait = false; |
2559 | unsigned int cpu; | |
1da177e4 | 2560 | |
73f53c4a TH |
2561 | if (flush_color >= 0) { |
2562 | BUG_ON(atomic_read(&wq->nr_cwqs_to_flush)); | |
2563 | atomic_set(&wq->nr_cwqs_to_flush, 1); | |
1da177e4 | 2564 | } |
2355b70f | 2565 | |
f3421797 | 2566 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a | 2567 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
d565ed63 | 2568 | struct worker_pool *pool = cwq->pool; |
fc2e4d70 | 2569 | |
d565ed63 | 2570 | spin_lock_irq(&pool->lock); |
83c22520 | 2571 | |
73f53c4a TH |
2572 | if (flush_color >= 0) { |
2573 | BUG_ON(cwq->flush_color != -1); | |
fc2e4d70 | 2574 | |
73f53c4a TH |
2575 | if (cwq->nr_in_flight[flush_color]) { |
2576 | cwq->flush_color = flush_color; | |
2577 | atomic_inc(&wq->nr_cwqs_to_flush); | |
2578 | wait = true; | |
2579 | } | |
2580 | } | |
1da177e4 | 2581 | |
73f53c4a TH |
2582 | if (work_color >= 0) { |
2583 | BUG_ON(work_color != work_next_color(cwq->work_color)); | |
2584 | cwq->work_color = work_color; | |
2585 | } | |
1da177e4 | 2586 | |
d565ed63 | 2587 | spin_unlock_irq(&pool->lock); |
1da177e4 | 2588 | } |
2355b70f | 2589 | |
73f53c4a TH |
2590 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_cwqs_to_flush)) |
2591 | complete(&wq->first_flusher->done); | |
14441960 | 2592 | |
73f53c4a | 2593 | return wait; |
1da177e4 LT |
2594 | } |
2595 | ||
0fcb78c2 | 2596 | /** |
1da177e4 | 2597 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2598 | * @wq: workqueue to flush |
1da177e4 LT |
2599 | * |
2600 | * Forces execution of the workqueue and blocks until its completion. | |
2601 | * This is typically used in driver shutdown handlers. | |
2602 | * | |
fc2e4d70 ON |
2603 | * We sleep until all works which were queued on entry have been handled, |
2604 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 2605 | */ |
7ad5b3a5 | 2606 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2607 | { |
73f53c4a TH |
2608 | struct wq_flusher this_flusher = { |
2609 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2610 | .flush_color = -1, | |
2611 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2612 | }; | |
2613 | int next_color; | |
1da177e4 | 2614 | |
3295f0ef IM |
2615 | lock_map_acquire(&wq->lockdep_map); |
2616 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
2617 | |
2618 | mutex_lock(&wq->flush_mutex); | |
2619 | ||
2620 | /* | |
2621 | * Start-to-wait phase | |
2622 | */ | |
2623 | next_color = work_next_color(wq->work_color); | |
2624 | ||
2625 | if (next_color != wq->flush_color) { | |
2626 | /* | |
2627 | * Color space is not full. The current work_color | |
2628 | * becomes our flush_color and work_color is advanced | |
2629 | * by one. | |
2630 | */ | |
2631 | BUG_ON(!list_empty(&wq->flusher_overflow)); | |
2632 | this_flusher.flush_color = wq->work_color; | |
2633 | wq->work_color = next_color; | |
2634 | ||
2635 | if (!wq->first_flusher) { | |
2636 | /* no flush in progress, become the first flusher */ | |
2637 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2638 | ||
2639 | wq->first_flusher = &this_flusher; | |
2640 | ||
2641 | if (!flush_workqueue_prep_cwqs(wq, wq->flush_color, | |
2642 | wq->work_color)) { | |
2643 | /* nothing to flush, done */ | |
2644 | wq->flush_color = next_color; | |
2645 | wq->first_flusher = NULL; | |
2646 | goto out_unlock; | |
2647 | } | |
2648 | } else { | |
2649 | /* wait in queue */ | |
2650 | BUG_ON(wq->flush_color == this_flusher.flush_color); | |
2651 | list_add_tail(&this_flusher.list, &wq->flusher_queue); | |
2652 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2653 | } | |
2654 | } else { | |
2655 | /* | |
2656 | * Oops, color space is full, wait on overflow queue. | |
2657 | * The next flush completion will assign us | |
2658 | * flush_color and transfer to flusher_queue. | |
2659 | */ | |
2660 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2661 | } | |
2662 | ||
2663 | mutex_unlock(&wq->flush_mutex); | |
2664 | ||
2665 | wait_for_completion(&this_flusher.done); | |
2666 | ||
2667 | /* | |
2668 | * Wake-up-and-cascade phase | |
2669 | * | |
2670 | * First flushers are responsible for cascading flushes and | |
2671 | * handling overflow. Non-first flushers can simply return. | |
2672 | */ | |
2673 | if (wq->first_flusher != &this_flusher) | |
2674 | return; | |
2675 | ||
2676 | mutex_lock(&wq->flush_mutex); | |
2677 | ||
4ce48b37 TH |
2678 | /* we might have raced, check again with mutex held */ |
2679 | if (wq->first_flusher != &this_flusher) | |
2680 | goto out_unlock; | |
2681 | ||
73f53c4a TH |
2682 | wq->first_flusher = NULL; |
2683 | ||
2684 | BUG_ON(!list_empty(&this_flusher.list)); | |
2685 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2686 | ||
2687 | while (true) { | |
2688 | struct wq_flusher *next, *tmp; | |
2689 | ||
2690 | /* complete all the flushers sharing the current flush color */ | |
2691 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2692 | if (next->flush_color != wq->flush_color) | |
2693 | break; | |
2694 | list_del_init(&next->list); | |
2695 | complete(&next->done); | |
2696 | } | |
2697 | ||
2698 | BUG_ON(!list_empty(&wq->flusher_overflow) && | |
2699 | wq->flush_color != work_next_color(wq->work_color)); | |
2700 | ||
2701 | /* this flush_color is finished, advance by one */ | |
2702 | wq->flush_color = work_next_color(wq->flush_color); | |
2703 | ||
2704 | /* one color has been freed, handle overflow queue */ | |
2705 | if (!list_empty(&wq->flusher_overflow)) { | |
2706 | /* | |
2707 | * Assign the same color to all overflowed | |
2708 | * flushers, advance work_color and append to | |
2709 | * flusher_queue. This is the start-to-wait | |
2710 | * phase for these overflowed flushers. | |
2711 | */ | |
2712 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2713 | tmp->flush_color = wq->work_color; | |
2714 | ||
2715 | wq->work_color = work_next_color(wq->work_color); | |
2716 | ||
2717 | list_splice_tail_init(&wq->flusher_overflow, | |
2718 | &wq->flusher_queue); | |
2719 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2720 | } | |
2721 | ||
2722 | if (list_empty(&wq->flusher_queue)) { | |
2723 | BUG_ON(wq->flush_color != wq->work_color); | |
2724 | break; | |
2725 | } | |
2726 | ||
2727 | /* | |
2728 | * Need to flush more colors. Make the next flusher | |
2729 | * the new first flusher and arm cwqs. | |
2730 | */ | |
2731 | BUG_ON(wq->flush_color == wq->work_color); | |
2732 | BUG_ON(wq->flush_color != next->flush_color); | |
2733 | ||
2734 | list_del_init(&next->list); | |
2735 | wq->first_flusher = next; | |
2736 | ||
2737 | if (flush_workqueue_prep_cwqs(wq, wq->flush_color, -1)) | |
2738 | break; | |
2739 | ||
2740 | /* | |
2741 | * Meh... this color is already done, clear first | |
2742 | * flusher and repeat cascading. | |
2743 | */ | |
2744 | wq->first_flusher = NULL; | |
2745 | } | |
2746 | ||
2747 | out_unlock: | |
2748 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 2749 | } |
ae90dd5d | 2750 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2751 | |
9c5a2ba7 TH |
2752 | /** |
2753 | * drain_workqueue - drain a workqueue | |
2754 | * @wq: workqueue to drain | |
2755 | * | |
2756 | * Wait until the workqueue becomes empty. While draining is in progress, | |
2757 | * only chain queueing is allowed. IOW, only currently pending or running | |
2758 | * work items on @wq can queue further work items on it. @wq is flushed | |
2759 | * repeatedly until it becomes empty. The number of flushing is detemined | |
2760 | * by the depth of chaining and should be relatively short. Whine if it | |
2761 | * takes too long. | |
2762 | */ | |
2763 | void drain_workqueue(struct workqueue_struct *wq) | |
2764 | { | |
2765 | unsigned int flush_cnt = 0; | |
2766 | unsigned int cpu; | |
2767 | ||
2768 | /* | |
2769 | * __queue_work() needs to test whether there are drainers, is much | |
2770 | * hotter than drain_workqueue() and already looks at @wq->flags. | |
2771 | * Use WQ_DRAINING so that queue doesn't have to check nr_drainers. | |
2772 | */ | |
2773 | spin_lock(&workqueue_lock); | |
2774 | if (!wq->nr_drainers++) | |
2775 | wq->flags |= WQ_DRAINING; | |
2776 | spin_unlock(&workqueue_lock); | |
2777 | reflush: | |
2778 | flush_workqueue(wq); | |
2779 | ||
2780 | for_each_cwq_cpu(cpu, wq) { | |
2781 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
fa2563e4 | 2782 | bool drained; |
9c5a2ba7 | 2783 | |
d565ed63 | 2784 | spin_lock_irq(&cwq->pool->lock); |
fa2563e4 | 2785 | drained = !cwq->nr_active && list_empty(&cwq->delayed_works); |
d565ed63 | 2786 | spin_unlock_irq(&cwq->pool->lock); |
fa2563e4 TT |
2787 | |
2788 | if (drained) | |
9c5a2ba7 TH |
2789 | continue; |
2790 | ||
2791 | if (++flush_cnt == 10 || | |
2792 | (flush_cnt % 100 == 0 && flush_cnt <= 1000)) | |
044c782c VI |
2793 | pr_warn("workqueue %s: flush on destruction isn't complete after %u tries\n", |
2794 | wq->name, flush_cnt); | |
9c5a2ba7 TH |
2795 | goto reflush; |
2796 | } | |
2797 | ||
2798 | spin_lock(&workqueue_lock); | |
2799 | if (!--wq->nr_drainers) | |
2800 | wq->flags &= ~WQ_DRAINING; | |
2801 | spin_unlock(&workqueue_lock); | |
2802 | } | |
2803 | EXPORT_SYMBOL_GPL(drain_workqueue); | |
2804 | ||
606a5020 | 2805 | static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr) |
db700897 | 2806 | { |
affee4b2 | 2807 | struct worker *worker = NULL; |
c9e7cf27 | 2808 | struct worker_pool *pool; |
db700897 | 2809 | struct cpu_workqueue_struct *cwq; |
db700897 ON |
2810 | |
2811 | might_sleep(); | |
c9e7cf27 TH |
2812 | pool = get_work_pool(work); |
2813 | if (!pool) | |
baf59022 | 2814 | return false; |
db700897 | 2815 | |
d565ed63 | 2816 | spin_lock_irq(&pool->lock); |
db700897 ON |
2817 | if (!list_empty(&work->entry)) { |
2818 | /* | |
2819 | * See the comment near try_to_grab_pending()->smp_rmb(). | |
d565ed63 | 2820 | * If it was re-queued to a different pool under us, we |
7a22ad75 | 2821 | * are not going to wait. |
db700897 ON |
2822 | */ |
2823 | smp_rmb(); | |
7a22ad75 | 2824 | cwq = get_work_cwq(work); |
d565ed63 | 2825 | if (unlikely(!cwq || pool != cwq->pool)) |
4690c4ab | 2826 | goto already_gone; |
606a5020 | 2827 | } else { |
c9e7cf27 | 2828 | worker = find_worker_executing_work(pool, work); |
affee4b2 | 2829 | if (!worker) |
4690c4ab | 2830 | goto already_gone; |
7a22ad75 | 2831 | cwq = worker->current_cwq; |
606a5020 | 2832 | } |
db700897 | 2833 | |
baf59022 | 2834 | insert_wq_barrier(cwq, barr, work, worker); |
d565ed63 | 2835 | spin_unlock_irq(&pool->lock); |
7a22ad75 | 2836 | |
e159489b TH |
2837 | /* |
2838 | * If @max_active is 1 or rescuer is in use, flushing another work | |
2839 | * item on the same workqueue may lead to deadlock. Make sure the | |
2840 | * flusher is not running on the same workqueue by verifying write | |
2841 | * access. | |
2842 | */ | |
2843 | if (cwq->wq->saved_max_active == 1 || cwq->wq->flags & WQ_RESCUER) | |
2844 | lock_map_acquire(&cwq->wq->lockdep_map); | |
2845 | else | |
2846 | lock_map_acquire_read(&cwq->wq->lockdep_map); | |
7a22ad75 | 2847 | lock_map_release(&cwq->wq->lockdep_map); |
e159489b | 2848 | |
401a8d04 | 2849 | return true; |
4690c4ab | 2850 | already_gone: |
d565ed63 | 2851 | spin_unlock_irq(&pool->lock); |
401a8d04 | 2852 | return false; |
db700897 | 2853 | } |
baf59022 TH |
2854 | |
2855 | /** | |
2856 | * flush_work - wait for a work to finish executing the last queueing instance | |
2857 | * @work: the work to flush | |
2858 | * | |
606a5020 TH |
2859 | * Wait until @work has finished execution. @work is guaranteed to be idle |
2860 | * on return if it hasn't been requeued since flush started. | |
baf59022 TH |
2861 | * |
2862 | * RETURNS: | |
2863 | * %true if flush_work() waited for the work to finish execution, | |
2864 | * %false if it was already idle. | |
2865 | */ | |
2866 | bool flush_work(struct work_struct *work) | |
2867 | { | |
2868 | struct wq_barrier barr; | |
2869 | ||
0976dfc1 SB |
2870 | lock_map_acquire(&work->lockdep_map); |
2871 | lock_map_release(&work->lockdep_map); | |
2872 | ||
606a5020 | 2873 | if (start_flush_work(work, &barr)) { |
401a8d04 TH |
2874 | wait_for_completion(&barr.done); |
2875 | destroy_work_on_stack(&barr.work); | |
2876 | return true; | |
606a5020 | 2877 | } else { |
401a8d04 | 2878 | return false; |
6e84d644 | 2879 | } |
6e84d644 | 2880 | } |
606a5020 | 2881 | EXPORT_SYMBOL_GPL(flush_work); |
6e84d644 | 2882 | |
36e227d2 | 2883 | static bool __cancel_work_timer(struct work_struct *work, bool is_dwork) |
1f1f642e | 2884 | { |
bbb68dfa | 2885 | unsigned long flags; |
1f1f642e ON |
2886 | int ret; |
2887 | ||
2888 | do { | |
bbb68dfa TH |
2889 | ret = try_to_grab_pending(work, is_dwork, &flags); |
2890 | /* | |
2891 | * If someone else is canceling, wait for the same event it | |
2892 | * would be waiting for before retrying. | |
2893 | */ | |
2894 | if (unlikely(ret == -ENOENT)) | |
606a5020 | 2895 | flush_work(work); |
1f1f642e ON |
2896 | } while (unlikely(ret < 0)); |
2897 | ||
bbb68dfa TH |
2898 | /* tell other tasks trying to grab @work to back off */ |
2899 | mark_work_canceling(work); | |
2900 | local_irq_restore(flags); | |
2901 | ||
606a5020 | 2902 | flush_work(work); |
7a22ad75 | 2903 | clear_work_data(work); |
1f1f642e ON |
2904 | return ret; |
2905 | } | |
2906 | ||
6e84d644 | 2907 | /** |
401a8d04 TH |
2908 | * cancel_work_sync - cancel a work and wait for it to finish |
2909 | * @work: the work to cancel | |
6e84d644 | 2910 | * |
401a8d04 TH |
2911 | * Cancel @work and wait for its execution to finish. This function |
2912 | * can be used even if the work re-queues itself or migrates to | |
2913 | * another workqueue. On return from this function, @work is | |
2914 | * guaranteed to be not pending or executing on any CPU. | |
1f1f642e | 2915 | * |
401a8d04 TH |
2916 | * cancel_work_sync(&delayed_work->work) must not be used for |
2917 | * delayed_work's. Use cancel_delayed_work_sync() instead. | |
6e84d644 | 2918 | * |
401a8d04 | 2919 | * The caller must ensure that the workqueue on which @work was last |
6e84d644 | 2920 | * queued can't be destroyed before this function returns. |
401a8d04 TH |
2921 | * |
2922 | * RETURNS: | |
2923 | * %true if @work was pending, %false otherwise. | |
6e84d644 | 2924 | */ |
401a8d04 | 2925 | bool cancel_work_sync(struct work_struct *work) |
6e84d644 | 2926 | { |
36e227d2 | 2927 | return __cancel_work_timer(work, false); |
b89deed3 | 2928 | } |
28e53bdd | 2929 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2930 | |
6e84d644 | 2931 | /** |
401a8d04 TH |
2932 | * flush_delayed_work - wait for a dwork to finish executing the last queueing |
2933 | * @dwork: the delayed work to flush | |
6e84d644 | 2934 | * |
401a8d04 TH |
2935 | * Delayed timer is cancelled and the pending work is queued for |
2936 | * immediate execution. Like flush_work(), this function only | |
2937 | * considers the last queueing instance of @dwork. | |
1f1f642e | 2938 | * |
401a8d04 TH |
2939 | * RETURNS: |
2940 | * %true if flush_work() waited for the work to finish execution, | |
2941 | * %false if it was already idle. | |
6e84d644 | 2942 | */ |
401a8d04 TH |
2943 | bool flush_delayed_work(struct delayed_work *dwork) |
2944 | { | |
8930caba | 2945 | local_irq_disable(); |
401a8d04 | 2946 | if (del_timer_sync(&dwork->timer)) |
1265057f | 2947 | __queue_work(dwork->cpu, |
401a8d04 | 2948 | get_work_cwq(&dwork->work)->wq, &dwork->work); |
8930caba | 2949 | local_irq_enable(); |
401a8d04 TH |
2950 | return flush_work(&dwork->work); |
2951 | } | |
2952 | EXPORT_SYMBOL(flush_delayed_work); | |
2953 | ||
09383498 | 2954 | /** |
57b30ae7 TH |
2955 | * cancel_delayed_work - cancel a delayed work |
2956 | * @dwork: delayed_work to cancel | |
09383498 | 2957 | * |
57b30ae7 TH |
2958 | * Kill off a pending delayed_work. Returns %true if @dwork was pending |
2959 | * and canceled; %false if wasn't pending. Note that the work callback | |
2960 | * function may still be running on return, unless it returns %true and the | |
2961 | * work doesn't re-arm itself. Explicitly flush or use | |
2962 | * cancel_delayed_work_sync() to wait on it. | |
09383498 | 2963 | * |
57b30ae7 | 2964 | * This function is safe to call from any context including IRQ handler. |
09383498 | 2965 | */ |
57b30ae7 | 2966 | bool cancel_delayed_work(struct delayed_work *dwork) |
09383498 | 2967 | { |
57b30ae7 TH |
2968 | unsigned long flags; |
2969 | int ret; | |
2970 | ||
2971 | do { | |
2972 | ret = try_to_grab_pending(&dwork->work, true, &flags); | |
2973 | } while (unlikely(ret == -EAGAIN)); | |
2974 | ||
2975 | if (unlikely(ret < 0)) | |
2976 | return false; | |
2977 | ||
7c3eed5c TH |
2978 | set_work_pool_and_clear_pending(&dwork->work, |
2979 | get_work_pool_id(&dwork->work)); | |
57b30ae7 | 2980 | local_irq_restore(flags); |
c0158ca6 | 2981 | return ret; |
09383498 | 2982 | } |
57b30ae7 | 2983 | EXPORT_SYMBOL(cancel_delayed_work); |
09383498 | 2984 | |
401a8d04 TH |
2985 | /** |
2986 | * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish | |
2987 | * @dwork: the delayed work cancel | |
2988 | * | |
2989 | * This is cancel_work_sync() for delayed works. | |
2990 | * | |
2991 | * RETURNS: | |
2992 | * %true if @dwork was pending, %false otherwise. | |
2993 | */ | |
2994 | bool cancel_delayed_work_sync(struct delayed_work *dwork) | |
6e84d644 | 2995 | { |
36e227d2 | 2996 | return __cancel_work_timer(&dwork->work, true); |
6e84d644 | 2997 | } |
f5a421a4 | 2998 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 2999 | |
0fcb78c2 | 3000 | /** |
c1a220e7 ZR |
3001 | * schedule_work_on - put work task on a specific cpu |
3002 | * @cpu: cpu to put the work task on | |
3003 | * @work: job to be done | |
3004 | * | |
3005 | * This puts a job on a specific cpu | |
3006 | */ | |
d4283e93 | 3007 | bool schedule_work_on(int cpu, struct work_struct *work) |
c1a220e7 | 3008 | { |
d320c038 | 3009 | return queue_work_on(cpu, system_wq, work); |
c1a220e7 ZR |
3010 | } |
3011 | EXPORT_SYMBOL(schedule_work_on); | |
3012 | ||
0fcb78c2 | 3013 | /** |
0fcb78c2 REB |
3014 | * schedule_work - put work task in global workqueue |
3015 | * @work: job to be done | |
0fcb78c2 | 3016 | * |
d4283e93 TH |
3017 | * Returns %false if @work was already on the kernel-global workqueue and |
3018 | * %true otherwise. | |
5b0f437d BVA |
3019 | * |
3020 | * This puts a job in the kernel-global workqueue if it was not already | |
3021 | * queued and leaves it in the same position on the kernel-global | |
3022 | * workqueue otherwise. | |
0fcb78c2 | 3023 | */ |
d4283e93 | 3024 | bool schedule_work(struct work_struct *work) |
1da177e4 | 3025 | { |
d320c038 | 3026 | return queue_work(system_wq, work); |
1da177e4 | 3027 | } |
ae90dd5d | 3028 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 3029 | |
0fcb78c2 REB |
3030 | /** |
3031 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
3032 | * @cpu: cpu to use | |
52bad64d | 3033 | * @dwork: job to be done |
0fcb78c2 REB |
3034 | * @delay: number of jiffies to wait |
3035 | * | |
3036 | * After waiting for a given time this puts a job in the kernel-global | |
3037 | * workqueue on the specified CPU. | |
3038 | */ | |
d4283e93 TH |
3039 | bool schedule_delayed_work_on(int cpu, struct delayed_work *dwork, |
3040 | unsigned long delay) | |
1da177e4 | 3041 | { |
d320c038 | 3042 | return queue_delayed_work_on(cpu, system_wq, dwork, delay); |
1da177e4 | 3043 | } |
ae90dd5d | 3044 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 3045 | |
0fcb78c2 REB |
3046 | /** |
3047 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
3048 | * @dwork: job to be done |
3049 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
3050 | * |
3051 | * After waiting for a given time this puts a job in the kernel-global | |
3052 | * workqueue. | |
3053 | */ | |
d4283e93 | 3054 | bool schedule_delayed_work(struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 3055 | { |
d320c038 | 3056 | return queue_delayed_work(system_wq, dwork, delay); |
1da177e4 | 3057 | } |
ae90dd5d | 3058 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 3059 | |
b6136773 | 3060 | /** |
31ddd871 | 3061 | * schedule_on_each_cpu - execute a function synchronously on each online CPU |
b6136773 | 3062 | * @func: the function to call |
b6136773 | 3063 | * |
31ddd871 TH |
3064 | * schedule_on_each_cpu() executes @func on each online CPU using the |
3065 | * system workqueue and blocks until all CPUs have completed. | |
b6136773 | 3066 | * schedule_on_each_cpu() is very slow. |
31ddd871 TH |
3067 | * |
3068 | * RETURNS: | |
3069 | * 0 on success, -errno on failure. | |
b6136773 | 3070 | */ |
65f27f38 | 3071 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
3072 | { |
3073 | int cpu; | |
38f51568 | 3074 | struct work_struct __percpu *works; |
15316ba8 | 3075 | |
b6136773 AM |
3076 | works = alloc_percpu(struct work_struct); |
3077 | if (!works) | |
15316ba8 | 3078 | return -ENOMEM; |
b6136773 | 3079 | |
93981800 TH |
3080 | get_online_cpus(); |
3081 | ||
15316ba8 | 3082 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
3083 | struct work_struct *work = per_cpu_ptr(works, cpu); |
3084 | ||
3085 | INIT_WORK(work, func); | |
b71ab8c2 | 3086 | schedule_work_on(cpu, work); |
65a64464 | 3087 | } |
93981800 TH |
3088 | |
3089 | for_each_online_cpu(cpu) | |
3090 | flush_work(per_cpu_ptr(works, cpu)); | |
3091 | ||
95402b38 | 3092 | put_online_cpus(); |
b6136773 | 3093 | free_percpu(works); |
15316ba8 CL |
3094 | return 0; |
3095 | } | |
3096 | ||
eef6a7d5 AS |
3097 | /** |
3098 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
3099 | * | |
3100 | * Forces execution of the kernel-global workqueue and blocks until its | |
3101 | * completion. | |
3102 | * | |
3103 | * Think twice before calling this function! It's very easy to get into | |
3104 | * trouble if you don't take great care. Either of the following situations | |
3105 | * will lead to deadlock: | |
3106 | * | |
3107 | * One of the work items currently on the workqueue needs to acquire | |
3108 | * a lock held by your code or its caller. | |
3109 | * | |
3110 | * Your code is running in the context of a work routine. | |
3111 | * | |
3112 | * They will be detected by lockdep when they occur, but the first might not | |
3113 | * occur very often. It depends on what work items are on the workqueue and | |
3114 | * what locks they need, which you have no control over. | |
3115 | * | |
3116 | * In most situations flushing the entire workqueue is overkill; you merely | |
3117 | * need to know that a particular work item isn't queued and isn't running. | |
3118 | * In such cases you should use cancel_delayed_work_sync() or | |
3119 | * cancel_work_sync() instead. | |
3120 | */ | |
1da177e4 LT |
3121 | void flush_scheduled_work(void) |
3122 | { | |
d320c038 | 3123 | flush_workqueue(system_wq); |
1da177e4 | 3124 | } |
ae90dd5d | 3125 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 3126 | |
1fa44eca JB |
3127 | /** |
3128 | * execute_in_process_context - reliably execute the routine with user context | |
3129 | * @fn: the function to execute | |
1fa44eca JB |
3130 | * @ew: guaranteed storage for the execute work structure (must |
3131 | * be available when the work executes) | |
3132 | * | |
3133 | * Executes the function immediately if process context is available, | |
3134 | * otherwise schedules the function for delayed execution. | |
3135 | * | |
3136 | * Returns: 0 - function was executed | |
3137 | * 1 - function was scheduled for execution | |
3138 | */ | |
65f27f38 | 3139 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
3140 | { |
3141 | if (!in_interrupt()) { | |
65f27f38 | 3142 | fn(&ew->work); |
1fa44eca JB |
3143 | return 0; |
3144 | } | |
3145 | ||
65f27f38 | 3146 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
3147 | schedule_work(&ew->work); |
3148 | ||
3149 | return 1; | |
3150 | } | |
3151 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
3152 | ||
1da177e4 LT |
3153 | int keventd_up(void) |
3154 | { | |
d320c038 | 3155 | return system_wq != NULL; |
1da177e4 LT |
3156 | } |
3157 | ||
bdbc5dd7 | 3158 | static int alloc_cwqs(struct workqueue_struct *wq) |
0f900049 | 3159 | { |
65a64464 | 3160 | /* |
0f900049 TH |
3161 | * cwqs are forced aligned according to WORK_STRUCT_FLAG_BITS. |
3162 | * Make sure that the alignment isn't lower than that of | |
3163 | * unsigned long long. | |
65a64464 | 3164 | */ |
0f900049 TH |
3165 | const size_t size = sizeof(struct cpu_workqueue_struct); |
3166 | const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS, | |
3167 | __alignof__(unsigned long long)); | |
65a64464 | 3168 | |
e06ffa1e | 3169 | if (!(wq->flags & WQ_UNBOUND)) |
f3421797 | 3170 | wq->cpu_wq.pcpu = __alloc_percpu(size, align); |
931ac77e | 3171 | else { |
f3421797 TH |
3172 | void *ptr; |
3173 | ||
3174 | /* | |
3175 | * Allocate enough room to align cwq and put an extra | |
3176 | * pointer at the end pointing back to the originally | |
3177 | * allocated pointer which will be used for free. | |
3178 | */ | |
3179 | ptr = kzalloc(size + align + sizeof(void *), GFP_KERNEL); | |
3180 | if (ptr) { | |
3181 | wq->cpu_wq.single = PTR_ALIGN(ptr, align); | |
3182 | *(void **)(wq->cpu_wq.single + 1) = ptr; | |
3183 | } | |
bdbc5dd7 | 3184 | } |
f3421797 | 3185 | |
0415b00d | 3186 | /* just in case, make sure it's actually aligned */ |
bdbc5dd7 TH |
3187 | BUG_ON(!IS_ALIGNED(wq->cpu_wq.v, align)); |
3188 | return wq->cpu_wq.v ? 0 : -ENOMEM; | |
0f900049 TH |
3189 | } |
3190 | ||
bdbc5dd7 | 3191 | static void free_cwqs(struct workqueue_struct *wq) |
0f900049 | 3192 | { |
e06ffa1e | 3193 | if (!(wq->flags & WQ_UNBOUND)) |
f3421797 TH |
3194 | free_percpu(wq->cpu_wq.pcpu); |
3195 | else if (wq->cpu_wq.single) { | |
3196 | /* the pointer to free is stored right after the cwq */ | |
bdbc5dd7 | 3197 | kfree(*(void **)(wq->cpu_wq.single + 1)); |
f3421797 | 3198 | } |
0f900049 TH |
3199 | } |
3200 | ||
f3421797 TH |
3201 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
3202 | const char *name) | |
b71ab8c2 | 3203 | { |
f3421797 TH |
3204 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
3205 | ||
3206 | if (max_active < 1 || max_active > lim) | |
044c782c VI |
3207 | pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n", |
3208 | max_active, name, 1, lim); | |
b71ab8c2 | 3209 | |
f3421797 | 3210 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
3211 | } |
3212 | ||
b196be89 | 3213 | struct workqueue_struct *__alloc_workqueue_key(const char *fmt, |
d320c038 TH |
3214 | unsigned int flags, |
3215 | int max_active, | |
3216 | struct lock_class_key *key, | |
b196be89 | 3217 | const char *lock_name, ...) |
1da177e4 | 3218 | { |
b196be89 | 3219 | va_list args, args1; |
1da177e4 | 3220 | struct workqueue_struct *wq; |
c34056a3 | 3221 | unsigned int cpu; |
b196be89 TH |
3222 | size_t namelen; |
3223 | ||
3224 | /* determine namelen, allocate wq and format name */ | |
3225 | va_start(args, lock_name); | |
3226 | va_copy(args1, args); | |
3227 | namelen = vsnprintf(NULL, 0, fmt, args) + 1; | |
3228 | ||
3229 | wq = kzalloc(sizeof(*wq) + namelen, GFP_KERNEL); | |
3230 | if (!wq) | |
3231 | goto err; | |
3232 | ||
3233 | vsnprintf(wq->name, namelen, fmt, args1); | |
3234 | va_end(args); | |
3235 | va_end(args1); | |
1da177e4 | 3236 | |
6370a6ad TH |
3237 | /* |
3238 | * Workqueues which may be used during memory reclaim should | |
3239 | * have a rescuer to guarantee forward progress. | |
3240 | */ | |
3241 | if (flags & WQ_MEM_RECLAIM) | |
3242 | flags |= WQ_RESCUER; | |
3243 | ||
d320c038 | 3244 | max_active = max_active ?: WQ_DFL_ACTIVE; |
b196be89 | 3245 | max_active = wq_clamp_max_active(max_active, flags, wq->name); |
3af24433 | 3246 | |
b196be89 | 3247 | /* init wq */ |
97e37d7b | 3248 | wq->flags = flags; |
a0a1a5fd | 3249 | wq->saved_max_active = max_active; |
73f53c4a TH |
3250 | mutex_init(&wq->flush_mutex); |
3251 | atomic_set(&wq->nr_cwqs_to_flush, 0); | |
3252 | INIT_LIST_HEAD(&wq->flusher_queue); | |
3253 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
502ca9d8 | 3254 | |
eb13ba87 | 3255 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 3256 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 3257 | |
bdbc5dd7 TH |
3258 | if (alloc_cwqs(wq) < 0) |
3259 | goto err; | |
3260 | ||
f3421797 | 3261 | for_each_cwq_cpu(cpu, wq) { |
1537663f TH |
3262 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3263 | ||
0f900049 | 3264 | BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK); |
a60dc39c | 3265 | cwq->pool = get_std_worker_pool(cpu, flags & WQ_HIGHPRI); |
c34056a3 | 3266 | cwq->wq = wq; |
73f53c4a | 3267 | cwq->flush_color = -1; |
1e19ffc6 | 3268 | cwq->max_active = max_active; |
1e19ffc6 | 3269 | INIT_LIST_HEAD(&cwq->delayed_works); |
e22bee78 | 3270 | } |
1537663f | 3271 | |
e22bee78 TH |
3272 | if (flags & WQ_RESCUER) { |
3273 | struct worker *rescuer; | |
3274 | ||
f2e005aa | 3275 | if (!alloc_mayday_mask(&wq->mayday_mask, GFP_KERNEL)) |
e22bee78 TH |
3276 | goto err; |
3277 | ||
3278 | wq->rescuer = rescuer = alloc_worker(); | |
3279 | if (!rescuer) | |
3280 | goto err; | |
3281 | ||
111c225a TH |
3282 | rescuer->rescue_wq = wq; |
3283 | rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", | |
b196be89 | 3284 | wq->name); |
e22bee78 TH |
3285 | if (IS_ERR(rescuer->task)) |
3286 | goto err; | |
3287 | ||
e22bee78 TH |
3288 | rescuer->task->flags |= PF_THREAD_BOUND; |
3289 | wake_up_process(rescuer->task); | |
3af24433 ON |
3290 | } |
3291 | ||
a0a1a5fd TH |
3292 | /* |
3293 | * workqueue_lock protects global freeze state and workqueues | |
3294 | * list. Grab it, set max_active accordingly and add the new | |
3295 | * workqueue to workqueues list. | |
3296 | */ | |
1537663f | 3297 | spin_lock(&workqueue_lock); |
a0a1a5fd | 3298 | |
58a69cb4 | 3299 | if (workqueue_freezing && wq->flags & WQ_FREEZABLE) |
f3421797 | 3300 | for_each_cwq_cpu(cpu, wq) |
a0a1a5fd TH |
3301 | get_cwq(cpu, wq)->max_active = 0; |
3302 | ||
1537663f | 3303 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 3304 | |
1537663f TH |
3305 | spin_unlock(&workqueue_lock); |
3306 | ||
3af24433 | 3307 | return wq; |
4690c4ab TH |
3308 | err: |
3309 | if (wq) { | |
bdbc5dd7 | 3310 | free_cwqs(wq); |
f2e005aa | 3311 | free_mayday_mask(wq->mayday_mask); |
e22bee78 | 3312 | kfree(wq->rescuer); |
4690c4ab TH |
3313 | kfree(wq); |
3314 | } | |
3315 | return NULL; | |
3af24433 | 3316 | } |
d320c038 | 3317 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 3318 | |
3af24433 ON |
3319 | /** |
3320 | * destroy_workqueue - safely terminate a workqueue | |
3321 | * @wq: target workqueue | |
3322 | * | |
3323 | * Safely destroy a workqueue. All work currently pending will be done first. | |
3324 | */ | |
3325 | void destroy_workqueue(struct workqueue_struct *wq) | |
3326 | { | |
c8e55f36 | 3327 | unsigned int cpu; |
3af24433 | 3328 | |
9c5a2ba7 TH |
3329 | /* drain it before proceeding with destruction */ |
3330 | drain_workqueue(wq); | |
c8efcc25 | 3331 | |
a0a1a5fd TH |
3332 | /* |
3333 | * wq list is used to freeze wq, remove from list after | |
3334 | * flushing is complete in case freeze races us. | |
3335 | */ | |
95402b38 | 3336 | spin_lock(&workqueue_lock); |
b1f4ec17 | 3337 | list_del(&wq->list); |
95402b38 | 3338 | spin_unlock(&workqueue_lock); |
3af24433 | 3339 | |
e22bee78 | 3340 | /* sanity check */ |
f3421797 | 3341 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a TH |
3342 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3343 | int i; | |
3344 | ||
73f53c4a TH |
3345 | for (i = 0; i < WORK_NR_COLORS; i++) |
3346 | BUG_ON(cwq->nr_in_flight[i]); | |
1e19ffc6 TH |
3347 | BUG_ON(cwq->nr_active); |
3348 | BUG_ON(!list_empty(&cwq->delayed_works)); | |
73f53c4a | 3349 | } |
9b41ea72 | 3350 | |
e22bee78 TH |
3351 | if (wq->flags & WQ_RESCUER) { |
3352 | kthread_stop(wq->rescuer->task); | |
f2e005aa | 3353 | free_mayday_mask(wq->mayday_mask); |
8d9df9f0 | 3354 | kfree(wq->rescuer); |
e22bee78 TH |
3355 | } |
3356 | ||
bdbc5dd7 | 3357 | free_cwqs(wq); |
3af24433 ON |
3358 | kfree(wq); |
3359 | } | |
3360 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
3361 | ||
9f4bd4cd LJ |
3362 | /** |
3363 | * cwq_set_max_active - adjust max_active of a cwq | |
3364 | * @cwq: target cpu_workqueue_struct | |
3365 | * @max_active: new max_active value. | |
3366 | * | |
3367 | * Set @cwq->max_active to @max_active and activate delayed works if | |
3368 | * increased. | |
3369 | * | |
3370 | * CONTEXT: | |
d565ed63 | 3371 | * spin_lock_irq(pool->lock). |
9f4bd4cd LJ |
3372 | */ |
3373 | static void cwq_set_max_active(struct cpu_workqueue_struct *cwq, int max_active) | |
3374 | { | |
3375 | cwq->max_active = max_active; | |
3376 | ||
3377 | while (!list_empty(&cwq->delayed_works) && | |
3378 | cwq->nr_active < cwq->max_active) | |
3379 | cwq_activate_first_delayed(cwq); | |
3380 | } | |
3381 | ||
dcd989cb TH |
3382 | /** |
3383 | * workqueue_set_max_active - adjust max_active of a workqueue | |
3384 | * @wq: target workqueue | |
3385 | * @max_active: new max_active value. | |
3386 | * | |
3387 | * Set max_active of @wq to @max_active. | |
3388 | * | |
3389 | * CONTEXT: | |
3390 | * Don't call from IRQ context. | |
3391 | */ | |
3392 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
3393 | { | |
3394 | unsigned int cpu; | |
3395 | ||
f3421797 | 3396 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb TH |
3397 | |
3398 | spin_lock(&workqueue_lock); | |
3399 | ||
3400 | wq->saved_max_active = max_active; | |
3401 | ||
f3421797 | 3402 | for_each_cwq_cpu(cpu, wq) { |
35b6bb63 TH |
3403 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3404 | struct worker_pool *pool = cwq->pool; | |
dcd989cb | 3405 | |
d565ed63 | 3406 | spin_lock_irq(&pool->lock); |
dcd989cb | 3407 | |
58a69cb4 | 3408 | if (!(wq->flags & WQ_FREEZABLE) || |
35b6bb63 TH |
3409 | !(pool->flags & POOL_FREEZING)) |
3410 | cwq_set_max_active(cwq, max_active); | |
9bfb1839 | 3411 | |
d565ed63 | 3412 | spin_unlock_irq(&pool->lock); |
65a64464 | 3413 | } |
93981800 | 3414 | |
dcd989cb | 3415 | spin_unlock(&workqueue_lock); |
15316ba8 | 3416 | } |
dcd989cb | 3417 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); |
15316ba8 | 3418 | |
eef6a7d5 | 3419 | /** |
dcd989cb TH |
3420 | * workqueue_congested - test whether a workqueue is congested |
3421 | * @cpu: CPU in question | |
3422 | * @wq: target workqueue | |
eef6a7d5 | 3423 | * |
dcd989cb TH |
3424 | * Test whether @wq's cpu workqueue for @cpu is congested. There is |
3425 | * no synchronization around this function and the test result is | |
3426 | * unreliable and only useful as advisory hints or for debugging. | |
eef6a7d5 | 3427 | * |
dcd989cb TH |
3428 | * RETURNS: |
3429 | * %true if congested, %false otherwise. | |
eef6a7d5 | 3430 | */ |
dcd989cb | 3431 | bool workqueue_congested(unsigned int cpu, struct workqueue_struct *wq) |
1da177e4 | 3432 | { |
dcd989cb TH |
3433 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
3434 | ||
3435 | return !list_empty(&cwq->delayed_works); | |
1da177e4 | 3436 | } |
dcd989cb | 3437 | EXPORT_SYMBOL_GPL(workqueue_congested); |
1da177e4 | 3438 | |
dcd989cb TH |
3439 | /** |
3440 | * work_busy - test whether a work is currently pending or running | |
3441 | * @work: the work to be tested | |
3442 | * | |
3443 | * Test whether @work is currently pending or running. There is no | |
3444 | * synchronization around this function and the test result is | |
3445 | * unreliable and only useful as advisory hints or for debugging. | |
3446 | * Especially for reentrant wqs, the pending state might hide the | |
3447 | * running state. | |
3448 | * | |
3449 | * RETURNS: | |
3450 | * OR'd bitmask of WORK_BUSY_* bits. | |
3451 | */ | |
3452 | unsigned int work_busy(struct work_struct *work) | |
1da177e4 | 3453 | { |
c9e7cf27 | 3454 | struct worker_pool *pool = get_work_pool(work); |
dcd989cb TH |
3455 | unsigned long flags; |
3456 | unsigned int ret = 0; | |
1da177e4 | 3457 | |
c9e7cf27 | 3458 | if (!pool) |
999767be | 3459 | return 0; |
1da177e4 | 3460 | |
d565ed63 | 3461 | spin_lock_irqsave(&pool->lock, flags); |
1da177e4 | 3462 | |
dcd989cb TH |
3463 | if (work_pending(work)) |
3464 | ret |= WORK_BUSY_PENDING; | |
c9e7cf27 | 3465 | if (find_worker_executing_work(pool, work)) |
dcd989cb | 3466 | ret |= WORK_BUSY_RUNNING; |
1da177e4 | 3467 | |
d565ed63 | 3468 | spin_unlock_irqrestore(&pool->lock, flags); |
1da177e4 | 3469 | |
dcd989cb | 3470 | return ret; |
1da177e4 | 3471 | } |
dcd989cb | 3472 | EXPORT_SYMBOL_GPL(work_busy); |
1da177e4 | 3473 | |
db7bccf4 TH |
3474 | /* |
3475 | * CPU hotplug. | |
3476 | * | |
e22bee78 TH |
3477 | * There are two challenges in supporting CPU hotplug. Firstly, there |
3478 | * are a lot of assumptions on strong associations among work, cwq and | |
706026c2 | 3479 | * pool which make migrating pending and scheduled works very |
e22bee78 | 3480 | * difficult to implement without impacting hot paths. Secondly, |
94cf58bb | 3481 | * worker pools serve mix of short, long and very long running works making |
e22bee78 TH |
3482 | * blocked draining impractical. |
3483 | * | |
24647570 | 3484 | * This is solved by allowing the pools to be disassociated from the CPU |
628c78e7 TH |
3485 | * running as an unbound one and allowing it to be reattached later if the |
3486 | * cpu comes back online. | |
db7bccf4 | 3487 | */ |
1da177e4 | 3488 | |
706026c2 | 3489 | static void wq_unbind_fn(struct work_struct *work) |
3af24433 | 3490 | { |
38db41d9 | 3491 | int cpu = smp_processor_id(); |
4ce62e9e | 3492 | struct worker_pool *pool; |
db7bccf4 TH |
3493 | struct worker *worker; |
3494 | struct hlist_node *pos; | |
3495 | int i; | |
3af24433 | 3496 | |
38db41d9 TH |
3497 | for_each_std_worker_pool(pool, cpu) { |
3498 | BUG_ON(cpu != smp_processor_id()); | |
db7bccf4 | 3499 | |
94cf58bb TH |
3500 | mutex_lock(&pool->assoc_mutex); |
3501 | spin_lock_irq(&pool->lock); | |
3af24433 | 3502 | |
94cf58bb TH |
3503 | /* |
3504 | * We've claimed all manager positions. Make all workers | |
3505 | * unbound and set DISASSOCIATED. Before this, all workers | |
3506 | * except for the ones which are still executing works from | |
3507 | * before the last CPU down must be on the cpu. After | |
3508 | * this, they may become diasporas. | |
3509 | */ | |
4ce62e9e | 3510 | list_for_each_entry(worker, &pool->idle_list, entry) |
403c821d | 3511 | worker->flags |= WORKER_UNBOUND; |
3af24433 | 3512 | |
c9e7cf27 TH |
3513 | for_each_busy_worker(worker, i, pos, pool) |
3514 | worker->flags |= WORKER_UNBOUND; | |
06ba38a9 | 3515 | |
24647570 | 3516 | pool->flags |= POOL_DISASSOCIATED; |
f2d5a0ee | 3517 | |
94cf58bb TH |
3518 | spin_unlock_irq(&pool->lock); |
3519 | mutex_unlock(&pool->assoc_mutex); | |
3520 | } | |
628c78e7 | 3521 | |
e22bee78 | 3522 | /* |
403c821d | 3523 | * Call schedule() so that we cross rq->lock and thus can guarantee |
628c78e7 TH |
3524 | * sched callbacks see the %WORKER_UNBOUND flag. This is necessary |
3525 | * as scheduler callbacks may be invoked from other cpus. | |
e22bee78 | 3526 | */ |
e22bee78 | 3527 | schedule(); |
06ba38a9 | 3528 | |
e22bee78 | 3529 | /* |
628c78e7 TH |
3530 | * Sched callbacks are disabled now. Zap nr_running. After this, |
3531 | * nr_running stays zero and need_more_worker() and keep_working() | |
38db41d9 TH |
3532 | * are always true as long as the worklist is not empty. Pools on |
3533 | * @cpu now behave as unbound (in terms of concurrency management) | |
3534 | * pools which are served by workers tied to the CPU. | |
628c78e7 TH |
3535 | * |
3536 | * On return from this function, the current worker would trigger | |
3537 | * unbound chain execution of pending work items if other workers | |
3538 | * didn't already. | |
e22bee78 | 3539 | */ |
38db41d9 | 3540 | for_each_std_worker_pool(pool, cpu) |
4ce62e9e | 3541 | atomic_set(get_pool_nr_running(pool), 0); |
3af24433 | 3542 | } |
3af24433 | 3543 | |
8db25e78 TH |
3544 | /* |
3545 | * Workqueues should be brought up before normal priority CPU notifiers. | |
3546 | * This will be registered high priority CPU notifier. | |
3547 | */ | |
9fdf9b73 | 3548 | static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb, |
8db25e78 TH |
3549 | unsigned long action, |
3550 | void *hcpu) | |
3af24433 ON |
3551 | { |
3552 | unsigned int cpu = (unsigned long)hcpu; | |
4ce62e9e | 3553 | struct worker_pool *pool; |
3ce63377 | 3554 | |
8db25e78 | 3555 | switch (action & ~CPU_TASKS_FROZEN) { |
3af24433 | 3556 | case CPU_UP_PREPARE: |
38db41d9 | 3557 | for_each_std_worker_pool(pool, cpu) { |
3ce63377 TH |
3558 | struct worker *worker; |
3559 | ||
3560 | if (pool->nr_workers) | |
3561 | continue; | |
3562 | ||
3563 | worker = create_worker(pool); | |
3564 | if (!worker) | |
3565 | return NOTIFY_BAD; | |
3566 | ||
d565ed63 | 3567 | spin_lock_irq(&pool->lock); |
3ce63377 | 3568 | start_worker(worker); |
d565ed63 | 3569 | spin_unlock_irq(&pool->lock); |
3af24433 | 3570 | } |
8db25e78 | 3571 | break; |
3af24433 | 3572 | |
db7bccf4 TH |
3573 | case CPU_DOWN_FAILED: |
3574 | case CPU_ONLINE: | |
38db41d9 | 3575 | for_each_std_worker_pool(pool, cpu) { |
94cf58bb TH |
3576 | mutex_lock(&pool->assoc_mutex); |
3577 | spin_lock_irq(&pool->lock); | |
3578 | ||
24647570 | 3579 | pool->flags &= ~POOL_DISASSOCIATED; |
94cf58bb TH |
3580 | rebind_workers(pool); |
3581 | ||
3582 | spin_unlock_irq(&pool->lock); | |
3583 | mutex_unlock(&pool->assoc_mutex); | |
3584 | } | |
db7bccf4 | 3585 | break; |
00dfcaf7 | 3586 | } |
65758202 TH |
3587 | return NOTIFY_OK; |
3588 | } | |
3589 | ||
3590 | /* | |
3591 | * Workqueues should be brought down after normal priority CPU notifiers. | |
3592 | * This will be registered as low priority CPU notifier. | |
3593 | */ | |
9fdf9b73 | 3594 | static int __cpuinit workqueue_cpu_down_callback(struct notifier_block *nfb, |
65758202 TH |
3595 | unsigned long action, |
3596 | void *hcpu) | |
3597 | { | |
8db25e78 TH |
3598 | unsigned int cpu = (unsigned long)hcpu; |
3599 | struct work_struct unbind_work; | |
3600 | ||
65758202 TH |
3601 | switch (action & ~CPU_TASKS_FROZEN) { |
3602 | case CPU_DOWN_PREPARE: | |
8db25e78 | 3603 | /* unbinding should happen on the local CPU */ |
706026c2 | 3604 | INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn); |
7635d2fd | 3605 | queue_work_on(cpu, system_highpri_wq, &unbind_work); |
8db25e78 TH |
3606 | flush_work(&unbind_work); |
3607 | break; | |
65758202 TH |
3608 | } |
3609 | return NOTIFY_OK; | |
3610 | } | |
3611 | ||
2d3854a3 | 3612 | #ifdef CONFIG_SMP |
8ccad40d | 3613 | |
2d3854a3 | 3614 | struct work_for_cpu { |
ed48ece2 | 3615 | struct work_struct work; |
2d3854a3 RR |
3616 | long (*fn)(void *); |
3617 | void *arg; | |
3618 | long ret; | |
3619 | }; | |
3620 | ||
ed48ece2 | 3621 | static void work_for_cpu_fn(struct work_struct *work) |
2d3854a3 | 3622 | { |
ed48ece2 TH |
3623 | struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work); |
3624 | ||
2d3854a3 RR |
3625 | wfc->ret = wfc->fn(wfc->arg); |
3626 | } | |
3627 | ||
3628 | /** | |
3629 | * work_on_cpu - run a function in user context on a particular cpu | |
3630 | * @cpu: the cpu to run on | |
3631 | * @fn: the function to run | |
3632 | * @arg: the function arg | |
3633 | * | |
31ad9081 RR |
3634 | * This will return the value @fn returns. |
3635 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 3636 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 RR |
3637 | */ |
3638 | long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) | |
3639 | { | |
ed48ece2 | 3640 | struct work_for_cpu wfc = { .fn = fn, .arg = arg }; |
6b44003e | 3641 | |
ed48ece2 TH |
3642 | INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn); |
3643 | schedule_work_on(cpu, &wfc.work); | |
3644 | flush_work(&wfc.work); | |
2d3854a3 RR |
3645 | return wfc.ret; |
3646 | } | |
3647 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
3648 | #endif /* CONFIG_SMP */ | |
3649 | ||
a0a1a5fd TH |
3650 | #ifdef CONFIG_FREEZER |
3651 | ||
3652 | /** | |
3653 | * freeze_workqueues_begin - begin freezing workqueues | |
3654 | * | |
58a69cb4 TH |
3655 | * Start freezing workqueues. After this function returns, all freezable |
3656 | * workqueues will queue new works to their frozen_works list instead of | |
706026c2 | 3657 | * pool->worklist. |
a0a1a5fd TH |
3658 | * |
3659 | * CONTEXT: | |
d565ed63 | 3660 | * Grabs and releases workqueue_lock and pool->lock's. |
a0a1a5fd TH |
3661 | */ |
3662 | void freeze_workqueues_begin(void) | |
3663 | { | |
a0a1a5fd TH |
3664 | unsigned int cpu; |
3665 | ||
3666 | spin_lock(&workqueue_lock); | |
3667 | ||
3668 | BUG_ON(workqueue_freezing); | |
3669 | workqueue_freezing = true; | |
3670 | ||
706026c2 | 3671 | for_each_wq_cpu(cpu) { |
35b6bb63 | 3672 | struct worker_pool *pool; |
bdbc5dd7 | 3673 | struct workqueue_struct *wq; |
8b03ae3c | 3674 | |
38db41d9 | 3675 | for_each_std_worker_pool(pool, cpu) { |
a1056305 | 3676 | spin_lock_irq(&pool->lock); |
d565ed63 | 3677 | |
35b6bb63 TH |
3678 | WARN_ON_ONCE(pool->flags & POOL_FREEZING); |
3679 | pool->flags |= POOL_FREEZING; | |
db7bccf4 | 3680 | |
a1056305 TH |
3681 | list_for_each_entry(wq, &workqueues, list) { |
3682 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
a0a1a5fd | 3683 | |
a1056305 TH |
3684 | if (cwq && cwq->pool == pool && |
3685 | (wq->flags & WQ_FREEZABLE)) | |
3686 | cwq->max_active = 0; | |
3687 | } | |
8b03ae3c | 3688 | |
a1056305 TH |
3689 | spin_unlock_irq(&pool->lock); |
3690 | } | |
a0a1a5fd TH |
3691 | } |
3692 | ||
3693 | spin_unlock(&workqueue_lock); | |
3694 | } | |
3695 | ||
3696 | /** | |
58a69cb4 | 3697 | * freeze_workqueues_busy - are freezable workqueues still busy? |
a0a1a5fd TH |
3698 | * |
3699 | * Check whether freezing is complete. This function must be called | |
3700 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
3701 | * | |
3702 | * CONTEXT: | |
3703 | * Grabs and releases workqueue_lock. | |
3704 | * | |
3705 | * RETURNS: | |
58a69cb4 TH |
3706 | * %true if some freezable workqueues are still busy. %false if freezing |
3707 | * is complete. | |
a0a1a5fd TH |
3708 | */ |
3709 | bool freeze_workqueues_busy(void) | |
3710 | { | |
a0a1a5fd TH |
3711 | unsigned int cpu; |
3712 | bool busy = false; | |
3713 | ||
3714 | spin_lock(&workqueue_lock); | |
3715 | ||
3716 | BUG_ON(!workqueue_freezing); | |
3717 | ||
706026c2 | 3718 | for_each_wq_cpu(cpu) { |
bdbc5dd7 | 3719 | struct workqueue_struct *wq; |
a0a1a5fd TH |
3720 | /* |
3721 | * nr_active is monotonically decreasing. It's safe | |
3722 | * to peek without lock. | |
3723 | */ | |
3724 | list_for_each_entry(wq, &workqueues, list) { | |
3725 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3726 | ||
58a69cb4 | 3727 | if (!cwq || !(wq->flags & WQ_FREEZABLE)) |
a0a1a5fd TH |
3728 | continue; |
3729 | ||
3730 | BUG_ON(cwq->nr_active < 0); | |
3731 | if (cwq->nr_active) { | |
3732 | busy = true; | |
3733 | goto out_unlock; | |
3734 | } | |
3735 | } | |
3736 | } | |
3737 | out_unlock: | |
3738 | spin_unlock(&workqueue_lock); | |
3739 | return busy; | |
3740 | } | |
3741 | ||
3742 | /** | |
3743 | * thaw_workqueues - thaw workqueues | |
3744 | * | |
3745 | * Thaw workqueues. Normal queueing is restored and all collected | |
706026c2 | 3746 | * frozen works are transferred to their respective pool worklists. |
a0a1a5fd TH |
3747 | * |
3748 | * CONTEXT: | |
d565ed63 | 3749 | * Grabs and releases workqueue_lock and pool->lock's. |
a0a1a5fd TH |
3750 | */ |
3751 | void thaw_workqueues(void) | |
3752 | { | |
a0a1a5fd TH |
3753 | unsigned int cpu; |
3754 | ||
3755 | spin_lock(&workqueue_lock); | |
3756 | ||
3757 | if (!workqueue_freezing) | |
3758 | goto out_unlock; | |
3759 | ||
706026c2 | 3760 | for_each_wq_cpu(cpu) { |
4ce62e9e | 3761 | struct worker_pool *pool; |
bdbc5dd7 | 3762 | struct workqueue_struct *wq; |
8b03ae3c | 3763 | |
38db41d9 | 3764 | for_each_std_worker_pool(pool, cpu) { |
a1056305 | 3765 | spin_lock_irq(&pool->lock); |
d565ed63 | 3766 | |
35b6bb63 TH |
3767 | WARN_ON_ONCE(!(pool->flags & POOL_FREEZING)); |
3768 | pool->flags &= ~POOL_FREEZING; | |
db7bccf4 | 3769 | |
a1056305 TH |
3770 | list_for_each_entry(wq, &workqueues, list) { |
3771 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
a0a1a5fd | 3772 | |
a1056305 TH |
3773 | if (!cwq || cwq->pool != pool || |
3774 | !(wq->flags & WQ_FREEZABLE)) | |
3775 | continue; | |
a0a1a5fd | 3776 | |
a1056305 TH |
3777 | /* restore max_active and repopulate worklist */ |
3778 | cwq_set_max_active(cwq, wq->saved_max_active); | |
3779 | } | |
8b03ae3c | 3780 | |
4ce62e9e | 3781 | wake_up_worker(pool); |
a1056305 TH |
3782 | |
3783 | spin_unlock_irq(&pool->lock); | |
d565ed63 | 3784 | } |
a0a1a5fd TH |
3785 | } |
3786 | ||
3787 | workqueue_freezing = false; | |
3788 | out_unlock: | |
3789 | spin_unlock(&workqueue_lock); | |
3790 | } | |
3791 | #endif /* CONFIG_FREEZER */ | |
3792 | ||
6ee0578b | 3793 | static int __init init_workqueues(void) |
1da177e4 | 3794 | { |
c34056a3 TH |
3795 | unsigned int cpu; |
3796 | ||
7c3eed5c TH |
3797 | /* make sure we have enough bits for OFFQ pool ID */ |
3798 | BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_POOL_SHIFT)) < | |
3799 | WORK_CPU_LAST * NR_STD_WORKER_POOLS); | |
b5490077 | 3800 | |
65758202 | 3801 | cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); |
a5b4e57d | 3802 | hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); |
8b03ae3c | 3803 | |
706026c2 TH |
3804 | /* initialize CPU pools */ |
3805 | for_each_wq_cpu(cpu) { | |
4ce62e9e | 3806 | struct worker_pool *pool; |
8b03ae3c | 3807 | |
38db41d9 | 3808 | for_each_std_worker_pool(pool, cpu) { |
d565ed63 | 3809 | spin_lock_init(&pool->lock); |
ec22ca5e | 3810 | pool->cpu = cpu; |
24647570 | 3811 | pool->flags |= POOL_DISASSOCIATED; |
4ce62e9e TH |
3812 | INIT_LIST_HEAD(&pool->worklist); |
3813 | INIT_LIST_HEAD(&pool->idle_list); | |
c9e7cf27 | 3814 | hash_init(pool->busy_hash); |
e7577c50 | 3815 | |
4ce62e9e TH |
3816 | init_timer_deferrable(&pool->idle_timer); |
3817 | pool->idle_timer.function = idle_worker_timeout; | |
3818 | pool->idle_timer.data = (unsigned long)pool; | |
e22bee78 | 3819 | |
706026c2 | 3820 | setup_timer(&pool->mayday_timer, pool_mayday_timeout, |
4ce62e9e TH |
3821 | (unsigned long)pool); |
3822 | ||
b2eb83d1 | 3823 | mutex_init(&pool->assoc_mutex); |
4ce62e9e | 3824 | ida_init(&pool->worker_ida); |
9daf9e67 TH |
3825 | |
3826 | /* alloc pool ID */ | |
3827 | BUG_ON(worker_pool_assign_id(pool)); | |
4ce62e9e | 3828 | } |
8b03ae3c TH |
3829 | } |
3830 | ||
e22bee78 | 3831 | /* create the initial worker */ |
706026c2 | 3832 | for_each_online_wq_cpu(cpu) { |
4ce62e9e | 3833 | struct worker_pool *pool; |
e22bee78 | 3834 | |
38db41d9 | 3835 | for_each_std_worker_pool(pool, cpu) { |
4ce62e9e TH |
3836 | struct worker *worker; |
3837 | ||
24647570 TH |
3838 | if (cpu != WORK_CPU_UNBOUND) |
3839 | pool->flags &= ~POOL_DISASSOCIATED; | |
3840 | ||
bc2ae0f5 | 3841 | worker = create_worker(pool); |
4ce62e9e | 3842 | BUG_ON(!worker); |
d565ed63 | 3843 | spin_lock_irq(&pool->lock); |
4ce62e9e | 3844 | start_worker(worker); |
d565ed63 | 3845 | spin_unlock_irq(&pool->lock); |
4ce62e9e | 3846 | } |
e22bee78 TH |
3847 | } |
3848 | ||
d320c038 | 3849 | system_wq = alloc_workqueue("events", 0, 0); |
1aabe902 | 3850 | system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0); |
d320c038 | 3851 | system_long_wq = alloc_workqueue("events_long", 0, 0); |
f3421797 TH |
3852 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
3853 | WQ_UNBOUND_MAX_ACTIVE); | |
24d51add TH |
3854 | system_freezable_wq = alloc_workqueue("events_freezable", |
3855 | WQ_FREEZABLE, 0); | |
1aabe902 | 3856 | BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq || |
ae930e0f | 3857 | !system_unbound_wq || !system_freezable_wq); |
6ee0578b | 3858 | return 0; |
1da177e4 | 3859 | } |
6ee0578b | 3860 | early_initcall(init_workqueues); |