]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/kernel/workqueue.c | |
3 | * | |
4 | * Generic mechanism for defining kernel helper threads for running | |
5 | * arbitrary tasks in process context. | |
6 | * | |
7 | * Started by Ingo Molnar, Copyright (C) 2002 | |
8 | * | |
9 | * Derived from the taskqueue/keventd code by: | |
10 | * | |
11 | * David Woodhouse <dwmw2@infradead.org> | |
e1f8e874 | 12 | * Andrew Morton |
1da177e4 LT |
13 | * Kai Petzke <wpp@marie.physik.tu-berlin.de> |
14 | * Theodore Ts'o <tytso@mit.edu> | |
89ada679 | 15 | * |
cde53535 | 16 | * Made to use alloc_percpu by Christoph Lameter. |
1da177e4 LT |
17 | */ |
18 | ||
19 | #include <linux/module.h> | |
20 | #include <linux/kernel.h> | |
21 | #include <linux/sched.h> | |
22 | #include <linux/init.h> | |
23 | #include <linux/signal.h> | |
24 | #include <linux/completion.h> | |
25 | #include <linux/workqueue.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/cpu.h> | |
28 | #include <linux/notifier.h> | |
29 | #include <linux/kthread.h> | |
1fa44eca | 30 | #include <linux/hardirq.h> |
46934023 | 31 | #include <linux/mempolicy.h> |
341a5958 | 32 | #include <linux/freezer.h> |
d5abe669 PZ |
33 | #include <linux/kallsyms.h> |
34 | #include <linux/debug_locks.h> | |
4e6045f1 | 35 | #include <linux/lockdep.h> |
c34056a3 | 36 | #include <linux/idr.h> |
e22bee78 TH |
37 | |
38 | #include "workqueue_sched.h" | |
1da177e4 | 39 | |
c8e55f36 | 40 | enum { |
db7bccf4 | 41 | /* global_cwq flags */ |
e22bee78 TH |
42 | GCWQ_MANAGE_WORKERS = 1 << 0, /* need to manage workers */ |
43 | GCWQ_MANAGING_WORKERS = 1 << 1, /* managing workers */ | |
44 | GCWQ_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */ | |
db7bccf4 | 45 | GCWQ_FREEZING = 1 << 3, /* freeze in progress */ |
649027d7 | 46 | GCWQ_HIGHPRI_PENDING = 1 << 4, /* highpri works on queue */ |
db7bccf4 | 47 | |
c8e55f36 TH |
48 | /* worker flags */ |
49 | WORKER_STARTED = 1 << 0, /* started */ | |
50 | WORKER_DIE = 1 << 1, /* die die die */ | |
51 | WORKER_IDLE = 1 << 2, /* is idle */ | |
e22bee78 | 52 | WORKER_PREP = 1 << 3, /* preparing to run works */ |
db7bccf4 | 53 | WORKER_ROGUE = 1 << 4, /* not bound to any cpu */ |
e22bee78 | 54 | WORKER_REBIND = 1 << 5, /* mom is home, come back */ |
fb0e7beb | 55 | WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ |
f3421797 | 56 | WORKER_UNBOUND = 1 << 7, /* worker is unbound */ |
e22bee78 | 57 | |
fb0e7beb | 58 | WORKER_NOT_RUNNING = WORKER_PREP | WORKER_ROGUE | WORKER_REBIND | |
f3421797 | 59 | WORKER_CPU_INTENSIVE | WORKER_UNBOUND, |
db7bccf4 TH |
60 | |
61 | /* gcwq->trustee_state */ | |
62 | TRUSTEE_START = 0, /* start */ | |
63 | TRUSTEE_IN_CHARGE = 1, /* trustee in charge of gcwq */ | |
64 | TRUSTEE_BUTCHER = 2, /* butcher workers */ | |
65 | TRUSTEE_RELEASE = 3, /* release workers */ | |
66 | TRUSTEE_DONE = 4, /* trustee is done */ | |
c8e55f36 TH |
67 | |
68 | BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */ | |
69 | BUSY_WORKER_HASH_SIZE = 1 << BUSY_WORKER_HASH_ORDER, | |
70 | BUSY_WORKER_HASH_MASK = BUSY_WORKER_HASH_SIZE - 1, | |
db7bccf4 | 71 | |
e22bee78 TH |
72 | MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */ |
73 | IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */ | |
74 | ||
75 | MAYDAY_INITIAL_TIMEOUT = HZ / 100, /* call for help after 10ms */ | |
76 | MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */ | |
77 | CREATE_COOLDOWN = HZ, /* time to breath after fail */ | |
db7bccf4 | 78 | TRUSTEE_COOLDOWN = HZ / 10, /* for trustee draining */ |
e22bee78 TH |
79 | |
80 | /* | |
81 | * Rescue workers are used only on emergencies and shared by | |
82 | * all cpus. Give -20. | |
83 | */ | |
84 | RESCUER_NICE_LEVEL = -20, | |
c8e55f36 TH |
85 | }; |
86 | ||
4690c4ab TH |
87 | /* |
88 | * Structure fields follow one of the following exclusion rules. | |
89 | * | |
90 | * I: Set during initialization and read-only afterwards. | |
91 | * | |
e22bee78 TH |
92 | * P: Preemption protected. Disabling preemption is enough and should |
93 | * only be modified and accessed from the local cpu. | |
94 | * | |
8b03ae3c | 95 | * L: gcwq->lock protected. Access with gcwq->lock held. |
4690c4ab | 96 | * |
e22bee78 TH |
97 | * X: During normal operation, modification requires gcwq->lock and |
98 | * should be done only from local cpu. Either disabling preemption | |
99 | * on local cpu or grabbing gcwq->lock is enough for read access. | |
f3421797 | 100 | * If GCWQ_DISASSOCIATED is set, it's identical to L. |
e22bee78 | 101 | * |
73f53c4a TH |
102 | * F: wq->flush_mutex protected. |
103 | * | |
4690c4ab TH |
104 | * W: workqueue_lock protected. |
105 | */ | |
106 | ||
8b03ae3c | 107 | struct global_cwq; |
c34056a3 | 108 | |
e22bee78 TH |
109 | /* |
110 | * The poor guys doing the actual heavy lifting. All on-duty workers | |
111 | * are either serving the manager role, on idle list or on busy hash. | |
112 | */ | |
c34056a3 | 113 | struct worker { |
c8e55f36 TH |
114 | /* on idle list while idle, on busy hash table while busy */ |
115 | union { | |
116 | struct list_head entry; /* L: while idle */ | |
117 | struct hlist_node hentry; /* L: while busy */ | |
118 | }; | |
119 | ||
c34056a3 | 120 | struct work_struct *current_work; /* L: work being processed */ |
8cca0eea | 121 | struct cpu_workqueue_struct *current_cwq; /* L: current_work's cwq */ |
affee4b2 | 122 | struct list_head scheduled; /* L: scheduled works */ |
c34056a3 | 123 | struct task_struct *task; /* I: worker task */ |
8b03ae3c | 124 | struct global_cwq *gcwq; /* I: the associated gcwq */ |
e22bee78 TH |
125 | /* 64 bytes boundary on 64bit, 32 on 32bit */ |
126 | unsigned long last_active; /* L: last active timestamp */ | |
127 | unsigned int flags; /* X: flags */ | |
c34056a3 | 128 | int id; /* I: worker id */ |
e22bee78 | 129 | struct work_struct rebind_work; /* L: rebind worker to cpu */ |
c34056a3 TH |
130 | }; |
131 | ||
8b03ae3c | 132 | /* |
e22bee78 TH |
133 | * Global per-cpu workqueue. There's one and only one for each cpu |
134 | * and all works are queued and processed here regardless of their | |
135 | * target workqueues. | |
8b03ae3c TH |
136 | */ |
137 | struct global_cwq { | |
138 | spinlock_t lock; /* the gcwq lock */ | |
7e11629d | 139 | struct list_head worklist; /* L: list of pending works */ |
8b03ae3c | 140 | unsigned int cpu; /* I: the associated cpu */ |
db7bccf4 | 141 | unsigned int flags; /* L: GCWQ_* flags */ |
c8e55f36 TH |
142 | |
143 | int nr_workers; /* L: total number of workers */ | |
144 | int nr_idle; /* L: currently idle ones */ | |
145 | ||
146 | /* workers are chained either in the idle_list or busy_hash */ | |
e22bee78 | 147 | struct list_head idle_list; /* X: list of idle workers */ |
c8e55f36 TH |
148 | struct hlist_head busy_hash[BUSY_WORKER_HASH_SIZE]; |
149 | /* L: hash of busy workers */ | |
150 | ||
e22bee78 TH |
151 | struct timer_list idle_timer; /* L: worker idle timeout */ |
152 | struct timer_list mayday_timer; /* L: SOS timer for dworkers */ | |
153 | ||
8b03ae3c | 154 | struct ida worker_ida; /* L: for worker IDs */ |
db7bccf4 TH |
155 | |
156 | struct task_struct *trustee; /* L: for gcwq shutdown */ | |
157 | unsigned int trustee_state; /* L: trustee state */ | |
158 | wait_queue_head_t trustee_wait; /* trustee wait */ | |
e22bee78 | 159 | struct worker *first_idle; /* L: first idle worker */ |
8b03ae3c TH |
160 | } ____cacheline_aligned_in_smp; |
161 | ||
1da177e4 | 162 | /* |
502ca9d8 | 163 | * The per-CPU workqueue. The lower WORK_STRUCT_FLAG_BITS of |
0f900049 TH |
164 | * work_struct->data are used for flags and thus cwqs need to be |
165 | * aligned at two's power of the number of flag bits. | |
1da177e4 LT |
166 | */ |
167 | struct cpu_workqueue_struct { | |
8b03ae3c | 168 | struct global_cwq *gcwq; /* I: the associated gcwq */ |
4690c4ab | 169 | struct workqueue_struct *wq; /* I: the owning workqueue */ |
73f53c4a TH |
170 | int work_color; /* L: current color */ |
171 | int flush_color; /* L: flushing color */ | |
172 | int nr_in_flight[WORK_NR_COLORS]; | |
173 | /* L: nr of in_flight works */ | |
1e19ffc6 | 174 | int nr_active; /* L: nr of active works */ |
a0a1a5fd | 175 | int max_active; /* L: max active works */ |
1e19ffc6 | 176 | struct list_head delayed_works; /* L: delayed works */ |
0f900049 | 177 | }; |
1da177e4 | 178 | |
73f53c4a TH |
179 | /* |
180 | * Structure used to wait for workqueue flush. | |
181 | */ | |
182 | struct wq_flusher { | |
183 | struct list_head list; /* F: list of flushers */ | |
184 | int flush_color; /* F: flush color waiting for */ | |
185 | struct completion done; /* flush completion */ | |
186 | }; | |
187 | ||
f2e005aa TH |
188 | /* |
189 | * All cpumasks are assumed to be always set on UP and thus can't be | |
190 | * used to determine whether there's something to be done. | |
191 | */ | |
192 | #ifdef CONFIG_SMP | |
193 | typedef cpumask_var_t mayday_mask_t; | |
194 | #define mayday_test_and_set_cpu(cpu, mask) \ | |
195 | cpumask_test_and_set_cpu((cpu), (mask)) | |
196 | #define mayday_clear_cpu(cpu, mask) cpumask_clear_cpu((cpu), (mask)) | |
197 | #define for_each_mayday_cpu(cpu, mask) for_each_cpu((cpu), (mask)) | |
198 | #define alloc_mayday_mask(maskp, gfp) alloc_cpumask_var((maskp), (gfp)) | |
199 | #define free_mayday_mask(mask) free_cpumask_var((mask)) | |
200 | #else | |
201 | typedef unsigned long mayday_mask_t; | |
202 | #define mayday_test_and_set_cpu(cpu, mask) test_and_set_bit(0, &(mask)) | |
203 | #define mayday_clear_cpu(cpu, mask) clear_bit(0, &(mask)) | |
204 | #define for_each_mayday_cpu(cpu, mask) if ((cpu) = 0, (mask)) | |
205 | #define alloc_mayday_mask(maskp, gfp) true | |
206 | #define free_mayday_mask(mask) do { } while (0) | |
207 | #endif | |
208 | ||
1da177e4 LT |
209 | /* |
210 | * The externally visible workqueue abstraction is an array of | |
211 | * per-CPU workqueues: | |
212 | */ | |
213 | struct workqueue_struct { | |
97e37d7b | 214 | unsigned int flags; /* I: WQ_* flags */ |
bdbc5dd7 TH |
215 | union { |
216 | struct cpu_workqueue_struct __percpu *pcpu; | |
217 | struct cpu_workqueue_struct *single; | |
218 | unsigned long v; | |
219 | } cpu_wq; /* I: cwq's */ | |
4690c4ab | 220 | struct list_head list; /* W: list of all workqueues */ |
73f53c4a TH |
221 | |
222 | struct mutex flush_mutex; /* protects wq flushing */ | |
223 | int work_color; /* F: current work color */ | |
224 | int flush_color; /* F: current flush color */ | |
225 | atomic_t nr_cwqs_to_flush; /* flush in progress */ | |
226 | struct wq_flusher *first_flusher; /* F: first flusher */ | |
227 | struct list_head flusher_queue; /* F: flush waiters */ | |
228 | struct list_head flusher_overflow; /* F: flush overflow list */ | |
229 | ||
f2e005aa | 230 | mayday_mask_t mayday_mask; /* cpus requesting rescue */ |
e22bee78 TH |
231 | struct worker *rescuer; /* I: rescue worker */ |
232 | ||
dcd989cb | 233 | int saved_max_active; /* W: saved cwq max_active */ |
4690c4ab | 234 | const char *name; /* I: workqueue name */ |
4e6045f1 | 235 | #ifdef CONFIG_LOCKDEP |
4690c4ab | 236 | struct lockdep_map lockdep_map; |
4e6045f1 | 237 | #endif |
1da177e4 LT |
238 | }; |
239 | ||
d320c038 TH |
240 | struct workqueue_struct *system_wq __read_mostly; |
241 | struct workqueue_struct *system_long_wq __read_mostly; | |
242 | struct workqueue_struct *system_nrt_wq __read_mostly; | |
f3421797 | 243 | struct workqueue_struct *system_unbound_wq __read_mostly; |
d320c038 TH |
244 | EXPORT_SYMBOL_GPL(system_wq); |
245 | EXPORT_SYMBOL_GPL(system_long_wq); | |
246 | EXPORT_SYMBOL_GPL(system_nrt_wq); | |
f3421797 | 247 | EXPORT_SYMBOL_GPL(system_unbound_wq); |
d320c038 | 248 | |
db7bccf4 TH |
249 | #define for_each_busy_worker(worker, i, pos, gcwq) \ |
250 | for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++) \ | |
251 | hlist_for_each_entry(worker, pos, &gcwq->busy_hash[i], hentry) | |
252 | ||
f3421797 TH |
253 | static inline int __next_gcwq_cpu(int cpu, const struct cpumask *mask, |
254 | unsigned int sw) | |
255 | { | |
256 | if (cpu < nr_cpu_ids) { | |
257 | if (sw & 1) { | |
258 | cpu = cpumask_next(cpu, mask); | |
259 | if (cpu < nr_cpu_ids) | |
260 | return cpu; | |
261 | } | |
262 | if (sw & 2) | |
263 | return WORK_CPU_UNBOUND; | |
264 | } | |
265 | return WORK_CPU_NONE; | |
266 | } | |
267 | ||
268 | static inline int __next_wq_cpu(int cpu, const struct cpumask *mask, | |
269 | struct workqueue_struct *wq) | |
270 | { | |
271 | return __next_gcwq_cpu(cpu, mask, !(wq->flags & WQ_UNBOUND) ? 1 : 2); | |
272 | } | |
273 | ||
09884951 TH |
274 | /* |
275 | * CPU iterators | |
276 | * | |
277 | * An extra gcwq is defined for an invalid cpu number | |
278 | * (WORK_CPU_UNBOUND) to host workqueues which are not bound to any | |
279 | * specific CPU. The following iterators are similar to | |
280 | * for_each_*_cpu() iterators but also considers the unbound gcwq. | |
281 | * | |
282 | * for_each_gcwq_cpu() : possible CPUs + WORK_CPU_UNBOUND | |
283 | * for_each_online_gcwq_cpu() : online CPUs + WORK_CPU_UNBOUND | |
284 | * for_each_cwq_cpu() : possible CPUs for bound workqueues, | |
285 | * WORK_CPU_UNBOUND for unbound workqueues | |
286 | */ | |
f3421797 TH |
287 | #define for_each_gcwq_cpu(cpu) \ |
288 | for ((cpu) = __next_gcwq_cpu(-1, cpu_possible_mask, 3); \ | |
289 | (cpu) < WORK_CPU_NONE; \ | |
290 | (cpu) = __next_gcwq_cpu((cpu), cpu_possible_mask, 3)) | |
291 | ||
292 | #define for_each_online_gcwq_cpu(cpu) \ | |
293 | for ((cpu) = __next_gcwq_cpu(-1, cpu_online_mask, 3); \ | |
294 | (cpu) < WORK_CPU_NONE; \ | |
295 | (cpu) = __next_gcwq_cpu((cpu), cpu_online_mask, 3)) | |
296 | ||
297 | #define for_each_cwq_cpu(cpu, wq) \ | |
298 | for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, (wq)); \ | |
299 | (cpu) < WORK_CPU_NONE; \ | |
300 | (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, (wq))) | |
301 | ||
dc186ad7 TG |
302 | #ifdef CONFIG_DEBUG_OBJECTS_WORK |
303 | ||
304 | static struct debug_obj_descr work_debug_descr; | |
305 | ||
306 | /* | |
307 | * fixup_init is called when: | |
308 | * - an active object is initialized | |
309 | */ | |
310 | static int work_fixup_init(void *addr, enum debug_obj_state state) | |
311 | { | |
312 | struct work_struct *work = addr; | |
313 | ||
314 | switch (state) { | |
315 | case ODEBUG_STATE_ACTIVE: | |
316 | cancel_work_sync(work); | |
317 | debug_object_init(work, &work_debug_descr); | |
318 | return 1; | |
319 | default: | |
320 | return 0; | |
321 | } | |
322 | } | |
323 | ||
324 | /* | |
325 | * fixup_activate is called when: | |
326 | * - an active object is activated | |
327 | * - an unknown object is activated (might be a statically initialized object) | |
328 | */ | |
329 | static int work_fixup_activate(void *addr, enum debug_obj_state state) | |
330 | { | |
331 | struct work_struct *work = addr; | |
332 | ||
333 | switch (state) { | |
334 | ||
335 | case ODEBUG_STATE_NOTAVAILABLE: | |
336 | /* | |
337 | * This is not really a fixup. The work struct was | |
338 | * statically initialized. We just make sure that it | |
339 | * is tracked in the object tracker. | |
340 | */ | |
22df02bb | 341 | if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) { |
dc186ad7 TG |
342 | debug_object_init(work, &work_debug_descr); |
343 | debug_object_activate(work, &work_debug_descr); | |
344 | return 0; | |
345 | } | |
346 | WARN_ON_ONCE(1); | |
347 | return 0; | |
348 | ||
349 | case ODEBUG_STATE_ACTIVE: | |
350 | WARN_ON(1); | |
351 | ||
352 | default: | |
353 | return 0; | |
354 | } | |
355 | } | |
356 | ||
357 | /* | |
358 | * fixup_free is called when: | |
359 | * - an active object is freed | |
360 | */ | |
361 | static int work_fixup_free(void *addr, enum debug_obj_state state) | |
362 | { | |
363 | struct work_struct *work = addr; | |
364 | ||
365 | switch (state) { | |
366 | case ODEBUG_STATE_ACTIVE: | |
367 | cancel_work_sync(work); | |
368 | debug_object_free(work, &work_debug_descr); | |
369 | return 1; | |
370 | default: | |
371 | return 0; | |
372 | } | |
373 | } | |
374 | ||
375 | static struct debug_obj_descr work_debug_descr = { | |
376 | .name = "work_struct", | |
377 | .fixup_init = work_fixup_init, | |
378 | .fixup_activate = work_fixup_activate, | |
379 | .fixup_free = work_fixup_free, | |
380 | }; | |
381 | ||
382 | static inline void debug_work_activate(struct work_struct *work) | |
383 | { | |
384 | debug_object_activate(work, &work_debug_descr); | |
385 | } | |
386 | ||
387 | static inline void debug_work_deactivate(struct work_struct *work) | |
388 | { | |
389 | debug_object_deactivate(work, &work_debug_descr); | |
390 | } | |
391 | ||
392 | void __init_work(struct work_struct *work, int onstack) | |
393 | { | |
394 | if (onstack) | |
395 | debug_object_init_on_stack(work, &work_debug_descr); | |
396 | else | |
397 | debug_object_init(work, &work_debug_descr); | |
398 | } | |
399 | EXPORT_SYMBOL_GPL(__init_work); | |
400 | ||
401 | void destroy_work_on_stack(struct work_struct *work) | |
402 | { | |
403 | debug_object_free(work, &work_debug_descr); | |
404 | } | |
405 | EXPORT_SYMBOL_GPL(destroy_work_on_stack); | |
406 | ||
407 | #else | |
408 | static inline void debug_work_activate(struct work_struct *work) { } | |
409 | static inline void debug_work_deactivate(struct work_struct *work) { } | |
410 | #endif | |
411 | ||
95402b38 GS |
412 | /* Serializes the accesses to the list of workqueues. */ |
413 | static DEFINE_SPINLOCK(workqueue_lock); | |
1da177e4 | 414 | static LIST_HEAD(workqueues); |
a0a1a5fd | 415 | static bool workqueue_freezing; /* W: have wqs started freezing? */ |
c34056a3 | 416 | |
e22bee78 TH |
417 | /* |
418 | * The almighty global cpu workqueues. nr_running is the only field | |
419 | * which is expected to be used frequently by other cpus via | |
420 | * try_to_wake_up(). Put it in a separate cacheline. | |
421 | */ | |
8b03ae3c | 422 | static DEFINE_PER_CPU(struct global_cwq, global_cwq); |
e22bee78 | 423 | static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t, gcwq_nr_running); |
8b03ae3c | 424 | |
f3421797 TH |
425 | /* |
426 | * Global cpu workqueue and nr_running counter for unbound gcwq. The | |
427 | * gcwq is always online, has GCWQ_DISASSOCIATED set, and all its | |
428 | * workers have WORKER_UNBOUND set. | |
429 | */ | |
430 | static struct global_cwq unbound_global_cwq; | |
431 | static atomic_t unbound_gcwq_nr_running = ATOMIC_INIT(0); /* always 0 */ | |
432 | ||
c34056a3 | 433 | static int worker_thread(void *__worker); |
1da177e4 | 434 | |
8b03ae3c TH |
435 | static struct global_cwq *get_gcwq(unsigned int cpu) |
436 | { | |
f3421797 TH |
437 | if (cpu != WORK_CPU_UNBOUND) |
438 | return &per_cpu(global_cwq, cpu); | |
439 | else | |
440 | return &unbound_global_cwq; | |
8b03ae3c TH |
441 | } |
442 | ||
e22bee78 TH |
443 | static atomic_t *get_gcwq_nr_running(unsigned int cpu) |
444 | { | |
f3421797 TH |
445 | if (cpu != WORK_CPU_UNBOUND) |
446 | return &per_cpu(gcwq_nr_running, cpu); | |
447 | else | |
448 | return &unbound_gcwq_nr_running; | |
e22bee78 TH |
449 | } |
450 | ||
1537663f TH |
451 | static struct cpu_workqueue_struct *get_cwq(unsigned int cpu, |
452 | struct workqueue_struct *wq) | |
b1f4ec17 | 453 | { |
f3421797 TH |
454 | if (!(wq->flags & WQ_UNBOUND)) { |
455 | if (likely(cpu < nr_cpu_ids)) { | |
456 | #ifdef CONFIG_SMP | |
457 | return per_cpu_ptr(wq->cpu_wq.pcpu, cpu); | |
bdbc5dd7 | 458 | #else |
f3421797 | 459 | return wq->cpu_wq.single; |
bdbc5dd7 | 460 | #endif |
f3421797 TH |
461 | } |
462 | } else if (likely(cpu == WORK_CPU_UNBOUND)) | |
463 | return wq->cpu_wq.single; | |
464 | return NULL; | |
b1f4ec17 ON |
465 | } |
466 | ||
73f53c4a TH |
467 | static unsigned int work_color_to_flags(int color) |
468 | { | |
469 | return color << WORK_STRUCT_COLOR_SHIFT; | |
470 | } | |
471 | ||
472 | static int get_work_color(struct work_struct *work) | |
473 | { | |
474 | return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) & | |
475 | ((1 << WORK_STRUCT_COLOR_BITS) - 1); | |
476 | } | |
477 | ||
478 | static int work_next_color(int color) | |
479 | { | |
480 | return (color + 1) % WORK_NR_COLORS; | |
481 | } | |
482 | ||
4594bf15 | 483 | /* |
e120153d TH |
484 | * A work's data points to the cwq with WORK_STRUCT_CWQ set while the |
485 | * work is on queue. Once execution starts, WORK_STRUCT_CWQ is | |
486 | * cleared and the work data contains the cpu number it was last on. | |
7a22ad75 TH |
487 | * |
488 | * set_work_{cwq|cpu}() and clear_work_data() can be used to set the | |
489 | * cwq, cpu or clear work->data. These functions should only be | |
490 | * called while the work is owned - ie. while the PENDING bit is set. | |
491 | * | |
492 | * get_work_[g]cwq() can be used to obtain the gcwq or cwq | |
493 | * corresponding to a work. gcwq is available once the work has been | |
494 | * queued anywhere after initialization. cwq is available only from | |
495 | * queueing until execution starts. | |
4594bf15 | 496 | */ |
7a22ad75 TH |
497 | static inline void set_work_data(struct work_struct *work, unsigned long data, |
498 | unsigned long flags) | |
365970a1 | 499 | { |
4594bf15 | 500 | BUG_ON(!work_pending(work)); |
7a22ad75 TH |
501 | atomic_long_set(&work->data, data | flags | work_static(work)); |
502 | } | |
365970a1 | 503 | |
7a22ad75 TH |
504 | static void set_work_cwq(struct work_struct *work, |
505 | struct cpu_workqueue_struct *cwq, | |
506 | unsigned long extra_flags) | |
507 | { | |
508 | set_work_data(work, (unsigned long)cwq, | |
e120153d | 509 | WORK_STRUCT_PENDING | WORK_STRUCT_CWQ | extra_flags); |
365970a1 DH |
510 | } |
511 | ||
7a22ad75 TH |
512 | static void set_work_cpu(struct work_struct *work, unsigned int cpu) |
513 | { | |
514 | set_work_data(work, cpu << WORK_STRUCT_FLAG_BITS, WORK_STRUCT_PENDING); | |
515 | } | |
516 | ||
517 | static void clear_work_data(struct work_struct *work) | |
518 | { | |
519 | set_work_data(work, WORK_STRUCT_NO_CPU, 0); | |
520 | } | |
521 | ||
7a22ad75 | 522 | static struct cpu_workqueue_struct *get_work_cwq(struct work_struct *work) |
4d707b9f | 523 | { |
e120153d | 524 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 | 525 | |
e120153d TH |
526 | if (data & WORK_STRUCT_CWQ) |
527 | return (void *)(data & WORK_STRUCT_WQ_DATA_MASK); | |
528 | else | |
529 | return NULL; | |
4d707b9f ON |
530 | } |
531 | ||
7a22ad75 | 532 | static struct global_cwq *get_work_gcwq(struct work_struct *work) |
365970a1 | 533 | { |
e120153d | 534 | unsigned long data = atomic_long_read(&work->data); |
7a22ad75 TH |
535 | unsigned int cpu; |
536 | ||
e120153d TH |
537 | if (data & WORK_STRUCT_CWQ) |
538 | return ((struct cpu_workqueue_struct *) | |
539 | (data & WORK_STRUCT_WQ_DATA_MASK))->gcwq; | |
7a22ad75 TH |
540 | |
541 | cpu = data >> WORK_STRUCT_FLAG_BITS; | |
bdbc5dd7 | 542 | if (cpu == WORK_CPU_NONE) |
7a22ad75 TH |
543 | return NULL; |
544 | ||
f3421797 | 545 | BUG_ON(cpu >= nr_cpu_ids && cpu != WORK_CPU_UNBOUND); |
7a22ad75 | 546 | return get_gcwq(cpu); |
365970a1 DH |
547 | } |
548 | ||
e22bee78 TH |
549 | /* |
550 | * Policy functions. These define the policies on how the global | |
551 | * worker pool is managed. Unless noted otherwise, these functions | |
552 | * assume that they're being called with gcwq->lock held. | |
553 | */ | |
554 | ||
649027d7 TH |
555 | static bool __need_more_worker(struct global_cwq *gcwq) |
556 | { | |
557 | return !atomic_read(get_gcwq_nr_running(gcwq->cpu)) || | |
558 | gcwq->flags & GCWQ_HIGHPRI_PENDING; | |
559 | } | |
560 | ||
e22bee78 TH |
561 | /* |
562 | * Need to wake up a worker? Called from anything but currently | |
563 | * running workers. | |
564 | */ | |
565 | static bool need_more_worker(struct global_cwq *gcwq) | |
566 | { | |
649027d7 | 567 | return !list_empty(&gcwq->worklist) && __need_more_worker(gcwq); |
e22bee78 TH |
568 | } |
569 | ||
570 | /* Can I start working? Called from busy but !running workers. */ | |
571 | static bool may_start_working(struct global_cwq *gcwq) | |
572 | { | |
573 | return gcwq->nr_idle; | |
574 | } | |
575 | ||
576 | /* Do I need to keep working? Called from currently running workers. */ | |
577 | static bool keep_working(struct global_cwq *gcwq) | |
578 | { | |
579 | atomic_t *nr_running = get_gcwq_nr_running(gcwq->cpu); | |
580 | ||
581 | return !list_empty(&gcwq->worklist) && atomic_read(nr_running) <= 1; | |
582 | } | |
583 | ||
584 | /* Do we need a new worker? Called from manager. */ | |
585 | static bool need_to_create_worker(struct global_cwq *gcwq) | |
586 | { | |
587 | return need_more_worker(gcwq) && !may_start_working(gcwq); | |
588 | } | |
589 | ||
590 | /* Do I need to be the manager? */ | |
591 | static bool need_to_manage_workers(struct global_cwq *gcwq) | |
592 | { | |
593 | return need_to_create_worker(gcwq) || gcwq->flags & GCWQ_MANAGE_WORKERS; | |
594 | } | |
595 | ||
596 | /* Do we have too many workers and should some go away? */ | |
597 | static bool too_many_workers(struct global_cwq *gcwq) | |
598 | { | |
599 | bool managing = gcwq->flags & GCWQ_MANAGING_WORKERS; | |
600 | int nr_idle = gcwq->nr_idle + managing; /* manager is considered idle */ | |
601 | int nr_busy = gcwq->nr_workers - nr_idle; | |
602 | ||
603 | return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; | |
604 | } | |
605 | ||
606 | /* | |
607 | * Wake up functions. | |
608 | */ | |
609 | ||
7e11629d TH |
610 | /* Return the first worker. Safe with preemption disabled */ |
611 | static struct worker *first_worker(struct global_cwq *gcwq) | |
612 | { | |
613 | if (unlikely(list_empty(&gcwq->idle_list))) | |
614 | return NULL; | |
615 | ||
616 | return list_first_entry(&gcwq->idle_list, struct worker, entry); | |
617 | } | |
618 | ||
619 | /** | |
620 | * wake_up_worker - wake up an idle worker | |
621 | * @gcwq: gcwq to wake worker for | |
622 | * | |
623 | * Wake up the first idle worker of @gcwq. | |
624 | * | |
625 | * CONTEXT: | |
626 | * spin_lock_irq(gcwq->lock). | |
627 | */ | |
628 | static void wake_up_worker(struct global_cwq *gcwq) | |
629 | { | |
630 | struct worker *worker = first_worker(gcwq); | |
631 | ||
632 | if (likely(worker)) | |
633 | wake_up_process(worker->task); | |
634 | } | |
635 | ||
d302f017 | 636 | /** |
e22bee78 TH |
637 | * wq_worker_waking_up - a worker is waking up |
638 | * @task: task waking up | |
639 | * @cpu: CPU @task is waking up to | |
640 | * | |
641 | * This function is called during try_to_wake_up() when a worker is | |
642 | * being awoken. | |
643 | * | |
644 | * CONTEXT: | |
645 | * spin_lock_irq(rq->lock) | |
646 | */ | |
647 | void wq_worker_waking_up(struct task_struct *task, unsigned int cpu) | |
648 | { | |
649 | struct worker *worker = kthread_data(task); | |
650 | ||
651 | if (likely(!(worker->flags & WORKER_NOT_RUNNING))) | |
652 | atomic_inc(get_gcwq_nr_running(cpu)); | |
653 | } | |
654 | ||
655 | /** | |
656 | * wq_worker_sleeping - a worker is going to sleep | |
657 | * @task: task going to sleep | |
658 | * @cpu: CPU in question, must be the current CPU number | |
659 | * | |
660 | * This function is called during schedule() when a busy worker is | |
661 | * going to sleep. Worker on the same cpu can be woken up by | |
662 | * returning pointer to its task. | |
663 | * | |
664 | * CONTEXT: | |
665 | * spin_lock_irq(rq->lock) | |
666 | * | |
667 | * RETURNS: | |
668 | * Worker task on @cpu to wake up, %NULL if none. | |
669 | */ | |
670 | struct task_struct *wq_worker_sleeping(struct task_struct *task, | |
671 | unsigned int cpu) | |
672 | { | |
673 | struct worker *worker = kthread_data(task), *to_wakeup = NULL; | |
674 | struct global_cwq *gcwq = get_gcwq(cpu); | |
675 | atomic_t *nr_running = get_gcwq_nr_running(cpu); | |
676 | ||
677 | if (unlikely(worker->flags & WORKER_NOT_RUNNING)) | |
678 | return NULL; | |
679 | ||
680 | /* this can only happen on the local cpu */ | |
681 | BUG_ON(cpu != raw_smp_processor_id()); | |
682 | ||
683 | /* | |
684 | * The counterpart of the following dec_and_test, implied mb, | |
685 | * worklist not empty test sequence is in insert_work(). | |
686 | * Please read comment there. | |
687 | * | |
688 | * NOT_RUNNING is clear. This means that trustee is not in | |
689 | * charge and we're running on the local cpu w/ rq lock held | |
690 | * and preemption disabled, which in turn means that none else | |
691 | * could be manipulating idle_list, so dereferencing idle_list | |
692 | * without gcwq lock is safe. | |
693 | */ | |
694 | if (atomic_dec_and_test(nr_running) && !list_empty(&gcwq->worklist)) | |
695 | to_wakeup = first_worker(gcwq); | |
696 | return to_wakeup ? to_wakeup->task : NULL; | |
697 | } | |
698 | ||
699 | /** | |
700 | * worker_set_flags - set worker flags and adjust nr_running accordingly | |
cb444766 | 701 | * @worker: self |
d302f017 TH |
702 | * @flags: flags to set |
703 | * @wakeup: wakeup an idle worker if necessary | |
704 | * | |
e22bee78 TH |
705 | * Set @flags in @worker->flags and adjust nr_running accordingly. If |
706 | * nr_running becomes zero and @wakeup is %true, an idle worker is | |
707 | * woken up. | |
d302f017 | 708 | * |
cb444766 TH |
709 | * CONTEXT: |
710 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
711 | */ |
712 | static inline void worker_set_flags(struct worker *worker, unsigned int flags, | |
713 | bool wakeup) | |
714 | { | |
e22bee78 TH |
715 | struct global_cwq *gcwq = worker->gcwq; |
716 | ||
cb444766 TH |
717 | WARN_ON_ONCE(worker->task != current); |
718 | ||
e22bee78 TH |
719 | /* |
720 | * If transitioning into NOT_RUNNING, adjust nr_running and | |
721 | * wake up an idle worker as necessary if requested by | |
722 | * @wakeup. | |
723 | */ | |
724 | if ((flags & WORKER_NOT_RUNNING) && | |
725 | !(worker->flags & WORKER_NOT_RUNNING)) { | |
726 | atomic_t *nr_running = get_gcwq_nr_running(gcwq->cpu); | |
727 | ||
728 | if (wakeup) { | |
729 | if (atomic_dec_and_test(nr_running) && | |
730 | !list_empty(&gcwq->worklist)) | |
731 | wake_up_worker(gcwq); | |
732 | } else | |
733 | atomic_dec(nr_running); | |
734 | } | |
735 | ||
d302f017 TH |
736 | worker->flags |= flags; |
737 | } | |
738 | ||
739 | /** | |
e22bee78 | 740 | * worker_clr_flags - clear worker flags and adjust nr_running accordingly |
cb444766 | 741 | * @worker: self |
d302f017 TH |
742 | * @flags: flags to clear |
743 | * | |
e22bee78 | 744 | * Clear @flags in @worker->flags and adjust nr_running accordingly. |
d302f017 | 745 | * |
cb444766 TH |
746 | * CONTEXT: |
747 | * spin_lock_irq(gcwq->lock) | |
d302f017 TH |
748 | */ |
749 | static inline void worker_clr_flags(struct worker *worker, unsigned int flags) | |
750 | { | |
e22bee78 TH |
751 | struct global_cwq *gcwq = worker->gcwq; |
752 | unsigned int oflags = worker->flags; | |
753 | ||
cb444766 TH |
754 | WARN_ON_ONCE(worker->task != current); |
755 | ||
d302f017 | 756 | worker->flags &= ~flags; |
e22bee78 TH |
757 | |
758 | /* if transitioning out of NOT_RUNNING, increment nr_running */ | |
759 | if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING)) | |
760 | if (!(worker->flags & WORKER_NOT_RUNNING)) | |
761 | atomic_inc(get_gcwq_nr_running(gcwq->cpu)); | |
d302f017 TH |
762 | } |
763 | ||
c8e55f36 TH |
764 | /** |
765 | * busy_worker_head - return the busy hash head for a work | |
766 | * @gcwq: gcwq of interest | |
767 | * @work: work to be hashed | |
768 | * | |
769 | * Return hash head of @gcwq for @work. | |
770 | * | |
771 | * CONTEXT: | |
772 | * spin_lock_irq(gcwq->lock). | |
773 | * | |
774 | * RETURNS: | |
775 | * Pointer to the hash head. | |
776 | */ | |
777 | static struct hlist_head *busy_worker_head(struct global_cwq *gcwq, | |
778 | struct work_struct *work) | |
779 | { | |
780 | const int base_shift = ilog2(sizeof(struct work_struct)); | |
781 | unsigned long v = (unsigned long)work; | |
782 | ||
783 | /* simple shift and fold hash, do we need something better? */ | |
784 | v >>= base_shift; | |
785 | v += v >> BUSY_WORKER_HASH_ORDER; | |
786 | v &= BUSY_WORKER_HASH_MASK; | |
787 | ||
788 | return &gcwq->busy_hash[v]; | |
789 | } | |
790 | ||
8cca0eea TH |
791 | /** |
792 | * __find_worker_executing_work - find worker which is executing a work | |
793 | * @gcwq: gcwq of interest | |
794 | * @bwh: hash head as returned by busy_worker_head() | |
795 | * @work: work to find worker for | |
796 | * | |
797 | * Find a worker which is executing @work on @gcwq. @bwh should be | |
798 | * the hash head obtained by calling busy_worker_head() with the same | |
799 | * work. | |
800 | * | |
801 | * CONTEXT: | |
802 | * spin_lock_irq(gcwq->lock). | |
803 | * | |
804 | * RETURNS: | |
805 | * Pointer to worker which is executing @work if found, NULL | |
806 | * otherwise. | |
807 | */ | |
808 | static struct worker *__find_worker_executing_work(struct global_cwq *gcwq, | |
809 | struct hlist_head *bwh, | |
810 | struct work_struct *work) | |
811 | { | |
812 | struct worker *worker; | |
813 | struct hlist_node *tmp; | |
814 | ||
815 | hlist_for_each_entry(worker, tmp, bwh, hentry) | |
816 | if (worker->current_work == work) | |
817 | return worker; | |
818 | return NULL; | |
819 | } | |
820 | ||
821 | /** | |
822 | * find_worker_executing_work - find worker which is executing a work | |
823 | * @gcwq: gcwq of interest | |
824 | * @work: work to find worker for | |
825 | * | |
826 | * Find a worker which is executing @work on @gcwq. This function is | |
827 | * identical to __find_worker_executing_work() except that this | |
828 | * function calculates @bwh itself. | |
829 | * | |
830 | * CONTEXT: | |
831 | * spin_lock_irq(gcwq->lock). | |
832 | * | |
833 | * RETURNS: | |
834 | * Pointer to worker which is executing @work if found, NULL | |
835 | * otherwise. | |
836 | */ | |
837 | static struct worker *find_worker_executing_work(struct global_cwq *gcwq, | |
838 | struct work_struct *work) | |
839 | { | |
840 | return __find_worker_executing_work(gcwq, busy_worker_head(gcwq, work), | |
841 | work); | |
842 | } | |
843 | ||
649027d7 TH |
844 | /** |
845 | * gcwq_determine_ins_pos - find insertion position | |
846 | * @gcwq: gcwq of interest | |
847 | * @cwq: cwq a work is being queued for | |
848 | * | |
849 | * A work for @cwq is about to be queued on @gcwq, determine insertion | |
850 | * position for the work. If @cwq is for HIGHPRI wq, the work is | |
851 | * queued at the head of the queue but in FIFO order with respect to | |
852 | * other HIGHPRI works; otherwise, at the end of the queue. This | |
853 | * function also sets GCWQ_HIGHPRI_PENDING flag to hint @gcwq that | |
854 | * there are HIGHPRI works pending. | |
855 | * | |
856 | * CONTEXT: | |
857 | * spin_lock_irq(gcwq->lock). | |
858 | * | |
859 | * RETURNS: | |
860 | * Pointer to inserstion position. | |
861 | */ | |
862 | static inline struct list_head *gcwq_determine_ins_pos(struct global_cwq *gcwq, | |
863 | struct cpu_workqueue_struct *cwq) | |
864 | { | |
865 | struct work_struct *twork; | |
866 | ||
867 | if (likely(!(cwq->wq->flags & WQ_HIGHPRI))) | |
868 | return &gcwq->worklist; | |
869 | ||
870 | list_for_each_entry(twork, &gcwq->worklist, entry) { | |
871 | struct cpu_workqueue_struct *tcwq = get_work_cwq(twork); | |
872 | ||
873 | if (!(tcwq->wq->flags & WQ_HIGHPRI)) | |
874 | break; | |
875 | } | |
876 | ||
877 | gcwq->flags |= GCWQ_HIGHPRI_PENDING; | |
878 | return &twork->entry; | |
879 | } | |
880 | ||
4690c4ab | 881 | /** |
7e11629d | 882 | * insert_work - insert a work into gcwq |
4690c4ab TH |
883 | * @cwq: cwq @work belongs to |
884 | * @work: work to insert | |
885 | * @head: insertion point | |
886 | * @extra_flags: extra WORK_STRUCT_* flags to set | |
887 | * | |
7e11629d TH |
888 | * Insert @work which belongs to @cwq into @gcwq after @head. |
889 | * @extra_flags is or'd to work_struct flags. | |
4690c4ab TH |
890 | * |
891 | * CONTEXT: | |
8b03ae3c | 892 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 893 | */ |
b89deed3 | 894 | static void insert_work(struct cpu_workqueue_struct *cwq, |
4690c4ab TH |
895 | struct work_struct *work, struct list_head *head, |
896 | unsigned int extra_flags) | |
b89deed3 | 897 | { |
e22bee78 TH |
898 | struct global_cwq *gcwq = cwq->gcwq; |
899 | ||
4690c4ab | 900 | /* we own @work, set data and link */ |
7a22ad75 | 901 | set_work_cwq(work, cwq, extra_flags); |
4690c4ab | 902 | |
6e84d644 ON |
903 | /* |
904 | * Ensure that we get the right work->data if we see the | |
905 | * result of list_add() below, see try_to_grab_pending(). | |
906 | */ | |
907 | smp_wmb(); | |
4690c4ab | 908 | |
1a4d9b0a | 909 | list_add_tail(&work->entry, head); |
e22bee78 TH |
910 | |
911 | /* | |
912 | * Ensure either worker_sched_deactivated() sees the above | |
913 | * list_add_tail() or we see zero nr_running to avoid workers | |
914 | * lying around lazily while there are works to be processed. | |
915 | */ | |
916 | smp_mb(); | |
917 | ||
649027d7 | 918 | if (__need_more_worker(gcwq)) |
e22bee78 | 919 | wake_up_worker(gcwq); |
b89deed3 ON |
920 | } |
921 | ||
4690c4ab | 922 | static void __queue_work(unsigned int cpu, struct workqueue_struct *wq, |
1da177e4 LT |
923 | struct work_struct *work) |
924 | { | |
502ca9d8 TH |
925 | struct global_cwq *gcwq; |
926 | struct cpu_workqueue_struct *cwq; | |
1e19ffc6 | 927 | struct list_head *worklist; |
1da177e4 LT |
928 | unsigned long flags; |
929 | ||
dc186ad7 | 930 | debug_work_activate(work); |
1e19ffc6 | 931 | |
c7fc77f7 TH |
932 | /* determine gcwq to use */ |
933 | if (!(wq->flags & WQ_UNBOUND)) { | |
18aa9eff TH |
934 | struct global_cwq *last_gcwq; |
935 | ||
c7fc77f7 TH |
936 | if (unlikely(cpu == WORK_CPU_UNBOUND)) |
937 | cpu = raw_smp_processor_id(); | |
938 | ||
18aa9eff TH |
939 | /* |
940 | * It's multi cpu. If @wq is non-reentrant and @work | |
941 | * was previously on a different cpu, it might still | |
942 | * be running there, in which case the work needs to | |
943 | * be queued on that cpu to guarantee non-reentrance. | |
944 | */ | |
502ca9d8 | 945 | gcwq = get_gcwq(cpu); |
18aa9eff TH |
946 | if (wq->flags & WQ_NON_REENTRANT && |
947 | (last_gcwq = get_work_gcwq(work)) && last_gcwq != gcwq) { | |
948 | struct worker *worker; | |
949 | ||
950 | spin_lock_irqsave(&last_gcwq->lock, flags); | |
951 | ||
952 | worker = find_worker_executing_work(last_gcwq, work); | |
953 | ||
954 | if (worker && worker->current_cwq->wq == wq) | |
955 | gcwq = last_gcwq; | |
956 | else { | |
957 | /* meh... not running there, queue here */ | |
958 | spin_unlock_irqrestore(&last_gcwq->lock, flags); | |
959 | spin_lock_irqsave(&gcwq->lock, flags); | |
960 | } | |
961 | } else | |
962 | spin_lock_irqsave(&gcwq->lock, flags); | |
f3421797 TH |
963 | } else { |
964 | gcwq = get_gcwq(WORK_CPU_UNBOUND); | |
965 | spin_lock_irqsave(&gcwq->lock, flags); | |
502ca9d8 TH |
966 | } |
967 | ||
968 | /* gcwq determined, get cwq and queue */ | |
969 | cwq = get_cwq(gcwq->cpu, wq); | |
970 | ||
4690c4ab | 971 | BUG_ON(!list_empty(&work->entry)); |
1e19ffc6 | 972 | |
73f53c4a | 973 | cwq->nr_in_flight[cwq->work_color]++; |
1e19ffc6 TH |
974 | |
975 | if (likely(cwq->nr_active < cwq->max_active)) { | |
976 | cwq->nr_active++; | |
649027d7 | 977 | worklist = gcwq_determine_ins_pos(gcwq, cwq); |
1e19ffc6 TH |
978 | } else |
979 | worklist = &cwq->delayed_works; | |
980 | ||
981 | insert_work(cwq, work, worklist, work_color_to_flags(cwq->work_color)); | |
982 | ||
8b03ae3c | 983 | spin_unlock_irqrestore(&gcwq->lock, flags); |
1da177e4 LT |
984 | } |
985 | ||
0fcb78c2 REB |
986 | /** |
987 | * queue_work - queue work on a workqueue | |
988 | * @wq: workqueue to use | |
989 | * @work: work to queue | |
990 | * | |
057647fc | 991 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
1da177e4 | 992 | * |
00dfcaf7 ON |
993 | * We queue the work to the CPU on which it was submitted, but if the CPU dies |
994 | * it can be processed by another CPU. | |
1da177e4 | 995 | */ |
7ad5b3a5 | 996 | int queue_work(struct workqueue_struct *wq, struct work_struct *work) |
1da177e4 | 997 | { |
ef1ca236 ON |
998 | int ret; |
999 | ||
1000 | ret = queue_work_on(get_cpu(), wq, work); | |
1001 | put_cpu(); | |
1002 | ||
1da177e4 LT |
1003 | return ret; |
1004 | } | |
ae90dd5d | 1005 | EXPORT_SYMBOL_GPL(queue_work); |
1da177e4 | 1006 | |
c1a220e7 ZR |
1007 | /** |
1008 | * queue_work_on - queue work on specific cpu | |
1009 | * @cpu: CPU number to execute work on | |
1010 | * @wq: workqueue to use | |
1011 | * @work: work to queue | |
1012 | * | |
1013 | * Returns 0 if @work was already on a queue, non-zero otherwise. | |
1014 | * | |
1015 | * We queue the work to a specific CPU, the caller must ensure it | |
1016 | * can't go away. | |
1017 | */ | |
1018 | int | |
1019 | queue_work_on(int cpu, struct workqueue_struct *wq, struct work_struct *work) | |
1020 | { | |
1021 | int ret = 0; | |
1022 | ||
22df02bb | 1023 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
4690c4ab | 1024 | __queue_work(cpu, wq, work); |
c1a220e7 ZR |
1025 | ret = 1; |
1026 | } | |
1027 | return ret; | |
1028 | } | |
1029 | EXPORT_SYMBOL_GPL(queue_work_on); | |
1030 | ||
6d141c3f | 1031 | static void delayed_work_timer_fn(unsigned long __data) |
1da177e4 | 1032 | { |
52bad64d | 1033 | struct delayed_work *dwork = (struct delayed_work *)__data; |
7a22ad75 | 1034 | struct cpu_workqueue_struct *cwq = get_work_cwq(&dwork->work); |
1da177e4 | 1035 | |
4690c4ab | 1036 | __queue_work(smp_processor_id(), cwq->wq, &dwork->work); |
1da177e4 LT |
1037 | } |
1038 | ||
0fcb78c2 REB |
1039 | /** |
1040 | * queue_delayed_work - queue work on a workqueue after delay | |
1041 | * @wq: workqueue to use | |
af9997e4 | 1042 | * @dwork: delayable work to queue |
0fcb78c2 REB |
1043 | * @delay: number of jiffies to wait before queueing |
1044 | * | |
057647fc | 1045 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 1046 | */ |
7ad5b3a5 | 1047 | int queue_delayed_work(struct workqueue_struct *wq, |
52bad64d | 1048 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 1049 | { |
52bad64d | 1050 | if (delay == 0) |
63bc0362 | 1051 | return queue_work(wq, &dwork->work); |
1da177e4 | 1052 | |
63bc0362 | 1053 | return queue_delayed_work_on(-1, wq, dwork, delay); |
1da177e4 | 1054 | } |
ae90dd5d | 1055 | EXPORT_SYMBOL_GPL(queue_delayed_work); |
1da177e4 | 1056 | |
0fcb78c2 REB |
1057 | /** |
1058 | * queue_delayed_work_on - queue work on specific CPU after delay | |
1059 | * @cpu: CPU number to execute work on | |
1060 | * @wq: workqueue to use | |
af9997e4 | 1061 | * @dwork: work to queue |
0fcb78c2 REB |
1062 | * @delay: number of jiffies to wait before queueing |
1063 | * | |
057647fc | 1064 | * Returns 0 if @work was already on a queue, non-zero otherwise. |
0fcb78c2 | 1065 | */ |
7a6bc1cd | 1066 | int queue_delayed_work_on(int cpu, struct workqueue_struct *wq, |
52bad64d | 1067 | struct delayed_work *dwork, unsigned long delay) |
7a6bc1cd VP |
1068 | { |
1069 | int ret = 0; | |
52bad64d DH |
1070 | struct timer_list *timer = &dwork->timer; |
1071 | struct work_struct *work = &dwork->work; | |
7a6bc1cd | 1072 | |
22df02bb | 1073 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { |
c7fc77f7 | 1074 | unsigned int lcpu; |
7a22ad75 | 1075 | |
7a6bc1cd VP |
1076 | BUG_ON(timer_pending(timer)); |
1077 | BUG_ON(!list_empty(&work->entry)); | |
1078 | ||
8a3e77cc | 1079 | timer_stats_timer_set_start_info(&dwork->timer); |
c7fc77f7 | 1080 | |
7a22ad75 TH |
1081 | /* |
1082 | * This stores cwq for the moment, for the timer_fn. | |
1083 | * Note that the work's gcwq is preserved to allow | |
1084 | * reentrance detection for delayed works. | |
1085 | */ | |
c7fc77f7 TH |
1086 | if (!(wq->flags & WQ_UNBOUND)) { |
1087 | struct global_cwq *gcwq = get_work_gcwq(work); | |
1088 | ||
1089 | if (gcwq && gcwq->cpu != WORK_CPU_UNBOUND) | |
1090 | lcpu = gcwq->cpu; | |
1091 | else | |
1092 | lcpu = raw_smp_processor_id(); | |
1093 | } else | |
1094 | lcpu = WORK_CPU_UNBOUND; | |
1095 | ||
7a22ad75 | 1096 | set_work_cwq(work, get_cwq(lcpu, wq), 0); |
c7fc77f7 | 1097 | |
7a6bc1cd | 1098 | timer->expires = jiffies + delay; |
52bad64d | 1099 | timer->data = (unsigned long)dwork; |
7a6bc1cd | 1100 | timer->function = delayed_work_timer_fn; |
63bc0362 ON |
1101 | |
1102 | if (unlikely(cpu >= 0)) | |
1103 | add_timer_on(timer, cpu); | |
1104 | else | |
1105 | add_timer(timer); | |
7a6bc1cd VP |
1106 | ret = 1; |
1107 | } | |
1108 | return ret; | |
1109 | } | |
ae90dd5d | 1110 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); |
1da177e4 | 1111 | |
c8e55f36 TH |
1112 | /** |
1113 | * worker_enter_idle - enter idle state | |
1114 | * @worker: worker which is entering idle state | |
1115 | * | |
1116 | * @worker is entering idle state. Update stats and idle timer if | |
1117 | * necessary. | |
1118 | * | |
1119 | * LOCKING: | |
1120 | * spin_lock_irq(gcwq->lock). | |
1121 | */ | |
1122 | static void worker_enter_idle(struct worker *worker) | |
1123 | { | |
1124 | struct global_cwq *gcwq = worker->gcwq; | |
1125 | ||
1126 | BUG_ON(worker->flags & WORKER_IDLE); | |
1127 | BUG_ON(!list_empty(&worker->entry) && | |
1128 | (worker->hentry.next || worker->hentry.pprev)); | |
1129 | ||
cb444766 TH |
1130 | /* can't use worker_set_flags(), also called from start_worker() */ |
1131 | worker->flags |= WORKER_IDLE; | |
c8e55f36 | 1132 | gcwq->nr_idle++; |
e22bee78 | 1133 | worker->last_active = jiffies; |
c8e55f36 TH |
1134 | |
1135 | /* idle_list is LIFO */ | |
1136 | list_add(&worker->entry, &gcwq->idle_list); | |
db7bccf4 | 1137 | |
e22bee78 TH |
1138 | if (likely(!(worker->flags & WORKER_ROGUE))) { |
1139 | if (too_many_workers(gcwq) && !timer_pending(&gcwq->idle_timer)) | |
1140 | mod_timer(&gcwq->idle_timer, | |
1141 | jiffies + IDLE_WORKER_TIMEOUT); | |
1142 | } else | |
db7bccf4 | 1143 | wake_up_all(&gcwq->trustee_wait); |
cb444766 TH |
1144 | |
1145 | /* sanity check nr_running */ | |
1146 | WARN_ON_ONCE(gcwq->nr_workers == gcwq->nr_idle && | |
1147 | atomic_read(get_gcwq_nr_running(gcwq->cpu))); | |
c8e55f36 TH |
1148 | } |
1149 | ||
1150 | /** | |
1151 | * worker_leave_idle - leave idle state | |
1152 | * @worker: worker which is leaving idle state | |
1153 | * | |
1154 | * @worker is leaving idle state. Update stats. | |
1155 | * | |
1156 | * LOCKING: | |
1157 | * spin_lock_irq(gcwq->lock). | |
1158 | */ | |
1159 | static void worker_leave_idle(struct worker *worker) | |
1160 | { | |
1161 | struct global_cwq *gcwq = worker->gcwq; | |
1162 | ||
1163 | BUG_ON(!(worker->flags & WORKER_IDLE)); | |
d302f017 | 1164 | worker_clr_flags(worker, WORKER_IDLE); |
c8e55f36 TH |
1165 | gcwq->nr_idle--; |
1166 | list_del_init(&worker->entry); | |
1167 | } | |
1168 | ||
e22bee78 TH |
1169 | /** |
1170 | * worker_maybe_bind_and_lock - bind worker to its cpu if possible and lock gcwq | |
1171 | * @worker: self | |
1172 | * | |
1173 | * Works which are scheduled while the cpu is online must at least be | |
1174 | * scheduled to a worker which is bound to the cpu so that if they are | |
1175 | * flushed from cpu callbacks while cpu is going down, they are | |
1176 | * guaranteed to execute on the cpu. | |
1177 | * | |
1178 | * This function is to be used by rogue workers and rescuers to bind | |
1179 | * themselves to the target cpu and may race with cpu going down or | |
1180 | * coming online. kthread_bind() can't be used because it may put the | |
1181 | * worker to already dead cpu and set_cpus_allowed_ptr() can't be used | |
1182 | * verbatim as it's best effort and blocking and gcwq may be | |
1183 | * [dis]associated in the meantime. | |
1184 | * | |
1185 | * This function tries set_cpus_allowed() and locks gcwq and verifies | |
1186 | * the binding against GCWQ_DISASSOCIATED which is set during | |
1187 | * CPU_DYING and cleared during CPU_ONLINE, so if the worker enters | |
1188 | * idle state or fetches works without dropping lock, it can guarantee | |
1189 | * the scheduling requirement described in the first paragraph. | |
1190 | * | |
1191 | * CONTEXT: | |
1192 | * Might sleep. Called without any lock but returns with gcwq->lock | |
1193 | * held. | |
1194 | * | |
1195 | * RETURNS: | |
1196 | * %true if the associated gcwq is online (@worker is successfully | |
1197 | * bound), %false if offline. | |
1198 | */ | |
1199 | static bool worker_maybe_bind_and_lock(struct worker *worker) | |
1200 | { | |
1201 | struct global_cwq *gcwq = worker->gcwq; | |
1202 | struct task_struct *task = worker->task; | |
1203 | ||
1204 | while (true) { | |
1205 | /* | |
1206 | * The following call may fail, succeed or succeed | |
1207 | * without actually migrating the task to the cpu if | |
1208 | * it races with cpu hotunplug operation. Verify | |
1209 | * against GCWQ_DISASSOCIATED. | |
1210 | */ | |
f3421797 TH |
1211 | if (!(gcwq->flags & GCWQ_DISASSOCIATED)) |
1212 | set_cpus_allowed_ptr(task, get_cpu_mask(gcwq->cpu)); | |
e22bee78 TH |
1213 | |
1214 | spin_lock_irq(&gcwq->lock); | |
1215 | if (gcwq->flags & GCWQ_DISASSOCIATED) | |
1216 | return false; | |
1217 | if (task_cpu(task) == gcwq->cpu && | |
1218 | cpumask_equal(¤t->cpus_allowed, | |
1219 | get_cpu_mask(gcwq->cpu))) | |
1220 | return true; | |
1221 | spin_unlock_irq(&gcwq->lock); | |
1222 | ||
1223 | /* CPU has come up inbetween, retry migration */ | |
1224 | cpu_relax(); | |
1225 | } | |
1226 | } | |
1227 | ||
1228 | /* | |
1229 | * Function for worker->rebind_work used to rebind rogue busy workers | |
1230 | * to the associated cpu which is coming back online. This is | |
1231 | * scheduled by cpu up but can race with other cpu hotplug operations | |
1232 | * and may be executed twice without intervening cpu down. | |
1233 | */ | |
1234 | static void worker_rebind_fn(struct work_struct *work) | |
1235 | { | |
1236 | struct worker *worker = container_of(work, struct worker, rebind_work); | |
1237 | struct global_cwq *gcwq = worker->gcwq; | |
1238 | ||
1239 | if (worker_maybe_bind_and_lock(worker)) | |
1240 | worker_clr_flags(worker, WORKER_REBIND); | |
1241 | ||
1242 | spin_unlock_irq(&gcwq->lock); | |
1243 | } | |
1244 | ||
c34056a3 TH |
1245 | static struct worker *alloc_worker(void) |
1246 | { | |
1247 | struct worker *worker; | |
1248 | ||
1249 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
c8e55f36 TH |
1250 | if (worker) { |
1251 | INIT_LIST_HEAD(&worker->entry); | |
affee4b2 | 1252 | INIT_LIST_HEAD(&worker->scheduled); |
e22bee78 TH |
1253 | INIT_WORK(&worker->rebind_work, worker_rebind_fn); |
1254 | /* on creation a worker is in !idle && prep state */ | |
1255 | worker->flags = WORKER_PREP; | |
c8e55f36 | 1256 | } |
c34056a3 TH |
1257 | return worker; |
1258 | } | |
1259 | ||
1260 | /** | |
1261 | * create_worker - create a new workqueue worker | |
7e11629d | 1262 | * @gcwq: gcwq the new worker will belong to |
c34056a3 TH |
1263 | * @bind: whether to set affinity to @cpu or not |
1264 | * | |
7e11629d | 1265 | * Create a new worker which is bound to @gcwq. The returned worker |
c34056a3 TH |
1266 | * can be started by calling start_worker() or destroyed using |
1267 | * destroy_worker(). | |
1268 | * | |
1269 | * CONTEXT: | |
1270 | * Might sleep. Does GFP_KERNEL allocations. | |
1271 | * | |
1272 | * RETURNS: | |
1273 | * Pointer to the newly created worker. | |
1274 | */ | |
7e11629d | 1275 | static struct worker *create_worker(struct global_cwq *gcwq, bool bind) |
c34056a3 | 1276 | { |
f3421797 | 1277 | bool on_unbound_cpu = gcwq->cpu == WORK_CPU_UNBOUND; |
c34056a3 | 1278 | struct worker *worker = NULL; |
f3421797 | 1279 | int id = -1; |
c34056a3 | 1280 | |
8b03ae3c TH |
1281 | spin_lock_irq(&gcwq->lock); |
1282 | while (ida_get_new(&gcwq->worker_ida, &id)) { | |
1283 | spin_unlock_irq(&gcwq->lock); | |
1284 | if (!ida_pre_get(&gcwq->worker_ida, GFP_KERNEL)) | |
c34056a3 | 1285 | goto fail; |
8b03ae3c | 1286 | spin_lock_irq(&gcwq->lock); |
c34056a3 | 1287 | } |
8b03ae3c | 1288 | spin_unlock_irq(&gcwq->lock); |
c34056a3 TH |
1289 | |
1290 | worker = alloc_worker(); | |
1291 | if (!worker) | |
1292 | goto fail; | |
1293 | ||
8b03ae3c | 1294 | worker->gcwq = gcwq; |
c34056a3 TH |
1295 | worker->id = id; |
1296 | ||
f3421797 TH |
1297 | if (!on_unbound_cpu) |
1298 | worker->task = kthread_create(worker_thread, worker, | |
1299 | "kworker/%u:%d", gcwq->cpu, id); | |
1300 | else | |
1301 | worker->task = kthread_create(worker_thread, worker, | |
1302 | "kworker/u:%d", id); | |
c34056a3 TH |
1303 | if (IS_ERR(worker->task)) |
1304 | goto fail; | |
1305 | ||
db7bccf4 TH |
1306 | /* |
1307 | * A rogue worker will become a regular one if CPU comes | |
1308 | * online later on. Make sure every worker has | |
1309 | * PF_THREAD_BOUND set. | |
1310 | */ | |
f3421797 | 1311 | if (bind && !on_unbound_cpu) |
8b03ae3c | 1312 | kthread_bind(worker->task, gcwq->cpu); |
f3421797 | 1313 | else { |
db7bccf4 | 1314 | worker->task->flags |= PF_THREAD_BOUND; |
f3421797 TH |
1315 | if (on_unbound_cpu) |
1316 | worker->flags |= WORKER_UNBOUND; | |
1317 | } | |
c34056a3 TH |
1318 | |
1319 | return worker; | |
1320 | fail: | |
1321 | if (id >= 0) { | |
8b03ae3c TH |
1322 | spin_lock_irq(&gcwq->lock); |
1323 | ida_remove(&gcwq->worker_ida, id); | |
1324 | spin_unlock_irq(&gcwq->lock); | |
c34056a3 TH |
1325 | } |
1326 | kfree(worker); | |
1327 | return NULL; | |
1328 | } | |
1329 | ||
1330 | /** | |
1331 | * start_worker - start a newly created worker | |
1332 | * @worker: worker to start | |
1333 | * | |
c8e55f36 | 1334 | * Make the gcwq aware of @worker and start it. |
c34056a3 TH |
1335 | * |
1336 | * CONTEXT: | |
8b03ae3c | 1337 | * spin_lock_irq(gcwq->lock). |
c34056a3 TH |
1338 | */ |
1339 | static void start_worker(struct worker *worker) | |
1340 | { | |
cb444766 | 1341 | worker->flags |= WORKER_STARTED; |
c8e55f36 TH |
1342 | worker->gcwq->nr_workers++; |
1343 | worker_enter_idle(worker); | |
c34056a3 TH |
1344 | wake_up_process(worker->task); |
1345 | } | |
1346 | ||
1347 | /** | |
1348 | * destroy_worker - destroy a workqueue worker | |
1349 | * @worker: worker to be destroyed | |
1350 | * | |
c8e55f36 TH |
1351 | * Destroy @worker and adjust @gcwq stats accordingly. |
1352 | * | |
1353 | * CONTEXT: | |
1354 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. | |
c34056a3 TH |
1355 | */ |
1356 | static void destroy_worker(struct worker *worker) | |
1357 | { | |
8b03ae3c | 1358 | struct global_cwq *gcwq = worker->gcwq; |
c34056a3 TH |
1359 | int id = worker->id; |
1360 | ||
1361 | /* sanity check frenzy */ | |
1362 | BUG_ON(worker->current_work); | |
affee4b2 | 1363 | BUG_ON(!list_empty(&worker->scheduled)); |
c34056a3 | 1364 | |
c8e55f36 TH |
1365 | if (worker->flags & WORKER_STARTED) |
1366 | gcwq->nr_workers--; | |
1367 | if (worker->flags & WORKER_IDLE) | |
1368 | gcwq->nr_idle--; | |
1369 | ||
1370 | list_del_init(&worker->entry); | |
cb444766 | 1371 | worker->flags |= WORKER_DIE; |
c8e55f36 TH |
1372 | |
1373 | spin_unlock_irq(&gcwq->lock); | |
1374 | ||
c34056a3 TH |
1375 | kthread_stop(worker->task); |
1376 | kfree(worker); | |
1377 | ||
8b03ae3c TH |
1378 | spin_lock_irq(&gcwq->lock); |
1379 | ida_remove(&gcwq->worker_ida, id); | |
c34056a3 TH |
1380 | } |
1381 | ||
e22bee78 TH |
1382 | static void idle_worker_timeout(unsigned long __gcwq) |
1383 | { | |
1384 | struct global_cwq *gcwq = (void *)__gcwq; | |
1385 | ||
1386 | spin_lock_irq(&gcwq->lock); | |
1387 | ||
1388 | if (too_many_workers(gcwq)) { | |
1389 | struct worker *worker; | |
1390 | unsigned long expires; | |
1391 | ||
1392 | /* idle_list is kept in LIFO order, check the last one */ | |
1393 | worker = list_entry(gcwq->idle_list.prev, struct worker, entry); | |
1394 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; | |
1395 | ||
1396 | if (time_before(jiffies, expires)) | |
1397 | mod_timer(&gcwq->idle_timer, expires); | |
1398 | else { | |
1399 | /* it's been idle for too long, wake up manager */ | |
1400 | gcwq->flags |= GCWQ_MANAGE_WORKERS; | |
1401 | wake_up_worker(gcwq); | |
1402 | } | |
1403 | } | |
1404 | ||
1405 | spin_unlock_irq(&gcwq->lock); | |
1406 | } | |
1407 | ||
1408 | static bool send_mayday(struct work_struct *work) | |
1409 | { | |
1410 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); | |
1411 | struct workqueue_struct *wq = cwq->wq; | |
f3421797 | 1412 | unsigned int cpu; |
e22bee78 TH |
1413 | |
1414 | if (!(wq->flags & WQ_RESCUER)) | |
1415 | return false; | |
1416 | ||
1417 | /* mayday mayday mayday */ | |
f3421797 TH |
1418 | cpu = cwq->gcwq->cpu; |
1419 | /* WORK_CPU_UNBOUND can't be set in cpumask, use cpu 0 instead */ | |
1420 | if (cpu == WORK_CPU_UNBOUND) | |
1421 | cpu = 0; | |
f2e005aa | 1422 | if (!mayday_test_and_set_cpu(cpu, wq->mayday_mask)) |
e22bee78 TH |
1423 | wake_up_process(wq->rescuer->task); |
1424 | return true; | |
1425 | } | |
1426 | ||
1427 | static void gcwq_mayday_timeout(unsigned long __gcwq) | |
1428 | { | |
1429 | struct global_cwq *gcwq = (void *)__gcwq; | |
1430 | struct work_struct *work; | |
1431 | ||
1432 | spin_lock_irq(&gcwq->lock); | |
1433 | ||
1434 | if (need_to_create_worker(gcwq)) { | |
1435 | /* | |
1436 | * We've been trying to create a new worker but | |
1437 | * haven't been successful. We might be hitting an | |
1438 | * allocation deadlock. Send distress signals to | |
1439 | * rescuers. | |
1440 | */ | |
1441 | list_for_each_entry(work, &gcwq->worklist, entry) | |
1442 | send_mayday(work); | |
1443 | } | |
1444 | ||
1445 | spin_unlock_irq(&gcwq->lock); | |
1446 | ||
1447 | mod_timer(&gcwq->mayday_timer, jiffies + MAYDAY_INTERVAL); | |
1448 | } | |
1449 | ||
1450 | /** | |
1451 | * maybe_create_worker - create a new worker if necessary | |
1452 | * @gcwq: gcwq to create a new worker for | |
1453 | * | |
1454 | * Create a new worker for @gcwq if necessary. @gcwq is guaranteed to | |
1455 | * have at least one idle worker on return from this function. If | |
1456 | * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is | |
1457 | * sent to all rescuers with works scheduled on @gcwq to resolve | |
1458 | * possible allocation deadlock. | |
1459 | * | |
1460 | * On return, need_to_create_worker() is guaranteed to be false and | |
1461 | * may_start_working() true. | |
1462 | * | |
1463 | * LOCKING: | |
1464 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1465 | * multiple times. Does GFP_KERNEL allocations. Called only from | |
1466 | * manager. | |
1467 | * | |
1468 | * RETURNS: | |
1469 | * false if no action was taken and gcwq->lock stayed locked, true | |
1470 | * otherwise. | |
1471 | */ | |
1472 | static bool maybe_create_worker(struct global_cwq *gcwq) | |
1473 | { | |
1474 | if (!need_to_create_worker(gcwq)) | |
1475 | return false; | |
1476 | restart: | |
9f9c2364 TH |
1477 | spin_unlock_irq(&gcwq->lock); |
1478 | ||
e22bee78 TH |
1479 | /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */ |
1480 | mod_timer(&gcwq->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT); | |
1481 | ||
1482 | while (true) { | |
1483 | struct worker *worker; | |
1484 | ||
e22bee78 TH |
1485 | worker = create_worker(gcwq, true); |
1486 | if (worker) { | |
1487 | del_timer_sync(&gcwq->mayday_timer); | |
1488 | spin_lock_irq(&gcwq->lock); | |
1489 | start_worker(worker); | |
1490 | BUG_ON(need_to_create_worker(gcwq)); | |
1491 | return true; | |
1492 | } | |
1493 | ||
1494 | if (!need_to_create_worker(gcwq)) | |
1495 | break; | |
1496 | ||
e22bee78 TH |
1497 | __set_current_state(TASK_INTERRUPTIBLE); |
1498 | schedule_timeout(CREATE_COOLDOWN); | |
9f9c2364 | 1499 | |
e22bee78 TH |
1500 | if (!need_to_create_worker(gcwq)) |
1501 | break; | |
1502 | } | |
1503 | ||
e22bee78 TH |
1504 | del_timer_sync(&gcwq->mayday_timer); |
1505 | spin_lock_irq(&gcwq->lock); | |
1506 | if (need_to_create_worker(gcwq)) | |
1507 | goto restart; | |
1508 | return true; | |
1509 | } | |
1510 | ||
1511 | /** | |
1512 | * maybe_destroy_worker - destroy workers which have been idle for a while | |
1513 | * @gcwq: gcwq to destroy workers for | |
1514 | * | |
1515 | * Destroy @gcwq workers which have been idle for longer than | |
1516 | * IDLE_WORKER_TIMEOUT. | |
1517 | * | |
1518 | * LOCKING: | |
1519 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1520 | * multiple times. Called only from manager. | |
1521 | * | |
1522 | * RETURNS: | |
1523 | * false if no action was taken and gcwq->lock stayed locked, true | |
1524 | * otherwise. | |
1525 | */ | |
1526 | static bool maybe_destroy_workers(struct global_cwq *gcwq) | |
1527 | { | |
1528 | bool ret = false; | |
1529 | ||
1530 | while (too_many_workers(gcwq)) { | |
1531 | struct worker *worker; | |
1532 | unsigned long expires; | |
1533 | ||
1534 | worker = list_entry(gcwq->idle_list.prev, struct worker, entry); | |
1535 | expires = worker->last_active + IDLE_WORKER_TIMEOUT; | |
1536 | ||
1537 | if (time_before(jiffies, expires)) { | |
1538 | mod_timer(&gcwq->idle_timer, expires); | |
1539 | break; | |
1540 | } | |
1541 | ||
1542 | destroy_worker(worker); | |
1543 | ret = true; | |
1544 | } | |
1545 | ||
1546 | return ret; | |
1547 | } | |
1548 | ||
1549 | /** | |
1550 | * manage_workers - manage worker pool | |
1551 | * @worker: self | |
1552 | * | |
1553 | * Assume the manager role and manage gcwq worker pool @worker belongs | |
1554 | * to. At any given time, there can be only zero or one manager per | |
1555 | * gcwq. The exclusion is handled automatically by this function. | |
1556 | * | |
1557 | * The caller can safely start processing works on false return. On | |
1558 | * true return, it's guaranteed that need_to_create_worker() is false | |
1559 | * and may_start_working() is true. | |
1560 | * | |
1561 | * CONTEXT: | |
1562 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
1563 | * multiple times. Does GFP_KERNEL allocations. | |
1564 | * | |
1565 | * RETURNS: | |
1566 | * false if no action was taken and gcwq->lock stayed locked, true if | |
1567 | * some action was taken. | |
1568 | */ | |
1569 | static bool manage_workers(struct worker *worker) | |
1570 | { | |
1571 | struct global_cwq *gcwq = worker->gcwq; | |
1572 | bool ret = false; | |
1573 | ||
1574 | if (gcwq->flags & GCWQ_MANAGING_WORKERS) | |
1575 | return ret; | |
1576 | ||
1577 | gcwq->flags &= ~GCWQ_MANAGE_WORKERS; | |
1578 | gcwq->flags |= GCWQ_MANAGING_WORKERS; | |
1579 | ||
1580 | /* | |
1581 | * Destroy and then create so that may_start_working() is true | |
1582 | * on return. | |
1583 | */ | |
1584 | ret |= maybe_destroy_workers(gcwq); | |
1585 | ret |= maybe_create_worker(gcwq); | |
1586 | ||
1587 | gcwq->flags &= ~GCWQ_MANAGING_WORKERS; | |
1588 | ||
1589 | /* | |
1590 | * The trustee might be waiting to take over the manager | |
1591 | * position, tell it we're done. | |
1592 | */ | |
1593 | if (unlikely(gcwq->trustee)) | |
1594 | wake_up_all(&gcwq->trustee_wait); | |
1595 | ||
1596 | return ret; | |
1597 | } | |
1598 | ||
affee4b2 TH |
1599 | /** |
1600 | * move_linked_works - move linked works to a list | |
1601 | * @work: start of series of works to be scheduled | |
1602 | * @head: target list to append @work to | |
1603 | * @nextp: out paramter for nested worklist walking | |
1604 | * | |
1605 | * Schedule linked works starting from @work to @head. Work series to | |
1606 | * be scheduled starts at @work and includes any consecutive work with | |
1607 | * WORK_STRUCT_LINKED set in its predecessor. | |
1608 | * | |
1609 | * If @nextp is not NULL, it's updated to point to the next work of | |
1610 | * the last scheduled work. This allows move_linked_works() to be | |
1611 | * nested inside outer list_for_each_entry_safe(). | |
1612 | * | |
1613 | * CONTEXT: | |
8b03ae3c | 1614 | * spin_lock_irq(gcwq->lock). |
affee4b2 TH |
1615 | */ |
1616 | static void move_linked_works(struct work_struct *work, struct list_head *head, | |
1617 | struct work_struct **nextp) | |
1618 | { | |
1619 | struct work_struct *n; | |
1620 | ||
1621 | /* | |
1622 | * Linked worklist will always end before the end of the list, | |
1623 | * use NULL for list head. | |
1624 | */ | |
1625 | list_for_each_entry_safe_from(work, n, NULL, entry) { | |
1626 | list_move_tail(&work->entry, head); | |
1627 | if (!(*work_data_bits(work) & WORK_STRUCT_LINKED)) | |
1628 | break; | |
1629 | } | |
1630 | ||
1631 | /* | |
1632 | * If we're already inside safe list traversal and have moved | |
1633 | * multiple works to the scheduled queue, the next position | |
1634 | * needs to be updated. | |
1635 | */ | |
1636 | if (nextp) | |
1637 | *nextp = n; | |
1638 | } | |
1639 | ||
1e19ffc6 TH |
1640 | static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq) |
1641 | { | |
1642 | struct work_struct *work = list_first_entry(&cwq->delayed_works, | |
1643 | struct work_struct, entry); | |
649027d7 | 1644 | struct list_head *pos = gcwq_determine_ins_pos(cwq->gcwq, cwq); |
1e19ffc6 | 1645 | |
649027d7 | 1646 | move_linked_works(work, pos, NULL); |
1e19ffc6 TH |
1647 | cwq->nr_active++; |
1648 | } | |
1649 | ||
73f53c4a TH |
1650 | /** |
1651 | * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight | |
1652 | * @cwq: cwq of interest | |
1653 | * @color: color of work which left the queue | |
1654 | * | |
1655 | * A work either has completed or is removed from pending queue, | |
1656 | * decrement nr_in_flight of its cwq and handle workqueue flushing. | |
1657 | * | |
1658 | * CONTEXT: | |
8b03ae3c | 1659 | * spin_lock_irq(gcwq->lock). |
73f53c4a TH |
1660 | */ |
1661 | static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color) | |
1662 | { | |
1663 | /* ignore uncolored works */ | |
1664 | if (color == WORK_NO_COLOR) | |
1665 | return; | |
1666 | ||
1667 | cwq->nr_in_flight[color]--; | |
1e19ffc6 TH |
1668 | cwq->nr_active--; |
1669 | ||
502ca9d8 TH |
1670 | if (!list_empty(&cwq->delayed_works)) { |
1671 | /* one down, submit a delayed one */ | |
1672 | if (cwq->nr_active < cwq->max_active) | |
1673 | cwq_activate_first_delayed(cwq); | |
502ca9d8 | 1674 | } |
73f53c4a TH |
1675 | |
1676 | /* is flush in progress and are we at the flushing tip? */ | |
1677 | if (likely(cwq->flush_color != color)) | |
1678 | return; | |
1679 | ||
1680 | /* are there still in-flight works? */ | |
1681 | if (cwq->nr_in_flight[color]) | |
1682 | return; | |
1683 | ||
1684 | /* this cwq is done, clear flush_color */ | |
1685 | cwq->flush_color = -1; | |
1686 | ||
1687 | /* | |
1688 | * If this was the last cwq, wake up the first flusher. It | |
1689 | * will handle the rest. | |
1690 | */ | |
1691 | if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush)) | |
1692 | complete(&cwq->wq->first_flusher->done); | |
1693 | } | |
1694 | ||
a62428c0 TH |
1695 | /** |
1696 | * process_one_work - process single work | |
c34056a3 | 1697 | * @worker: self |
a62428c0 TH |
1698 | * @work: work to process |
1699 | * | |
1700 | * Process @work. This function contains all the logics necessary to | |
1701 | * process a single work including synchronization against and | |
1702 | * interaction with other workers on the same cpu, queueing and | |
1703 | * flushing. As long as context requirement is met, any worker can | |
1704 | * call this function to process a work. | |
1705 | * | |
1706 | * CONTEXT: | |
8b03ae3c | 1707 | * spin_lock_irq(gcwq->lock) which is released and regrabbed. |
a62428c0 | 1708 | */ |
c34056a3 | 1709 | static void process_one_work(struct worker *worker, struct work_struct *work) |
a62428c0 | 1710 | { |
7e11629d | 1711 | struct cpu_workqueue_struct *cwq = get_work_cwq(work); |
8b03ae3c | 1712 | struct global_cwq *gcwq = cwq->gcwq; |
c8e55f36 | 1713 | struct hlist_head *bwh = busy_worker_head(gcwq, work); |
fb0e7beb | 1714 | bool cpu_intensive = cwq->wq->flags & WQ_CPU_INTENSIVE; |
a62428c0 | 1715 | work_func_t f = work->func; |
73f53c4a | 1716 | int work_color; |
7e11629d | 1717 | struct worker *collision; |
a62428c0 TH |
1718 | #ifdef CONFIG_LOCKDEP |
1719 | /* | |
1720 | * It is permissible to free the struct work_struct from | |
1721 | * inside the function that is called from it, this we need to | |
1722 | * take into account for lockdep too. To avoid bogus "held | |
1723 | * lock freed" warnings as well as problems when looking into | |
1724 | * work->lockdep_map, make a copy and use that here. | |
1725 | */ | |
1726 | struct lockdep_map lockdep_map = work->lockdep_map; | |
1727 | #endif | |
7e11629d TH |
1728 | /* |
1729 | * A single work shouldn't be executed concurrently by | |
1730 | * multiple workers on a single cpu. Check whether anyone is | |
1731 | * already processing the work. If so, defer the work to the | |
1732 | * currently executing one. | |
1733 | */ | |
1734 | collision = __find_worker_executing_work(gcwq, bwh, work); | |
1735 | if (unlikely(collision)) { | |
1736 | move_linked_works(work, &collision->scheduled, NULL); | |
1737 | return; | |
1738 | } | |
1739 | ||
a62428c0 | 1740 | /* claim and process */ |
a62428c0 | 1741 | debug_work_deactivate(work); |
c8e55f36 | 1742 | hlist_add_head(&worker->hentry, bwh); |
c34056a3 | 1743 | worker->current_work = work; |
8cca0eea | 1744 | worker->current_cwq = cwq; |
73f53c4a | 1745 | work_color = get_work_color(work); |
7a22ad75 | 1746 | |
7a22ad75 TH |
1747 | /* record the current cpu number in the work data and dequeue */ |
1748 | set_work_cpu(work, gcwq->cpu); | |
a62428c0 TH |
1749 | list_del_init(&work->entry); |
1750 | ||
649027d7 TH |
1751 | /* |
1752 | * If HIGHPRI_PENDING, check the next work, and, if HIGHPRI, | |
1753 | * wake up another worker; otherwise, clear HIGHPRI_PENDING. | |
1754 | */ | |
1755 | if (unlikely(gcwq->flags & GCWQ_HIGHPRI_PENDING)) { | |
1756 | struct work_struct *nwork = list_first_entry(&gcwq->worklist, | |
1757 | struct work_struct, entry); | |
1758 | ||
1759 | if (!list_empty(&gcwq->worklist) && | |
1760 | get_work_cwq(nwork)->wq->flags & WQ_HIGHPRI) | |
1761 | wake_up_worker(gcwq); | |
1762 | else | |
1763 | gcwq->flags &= ~GCWQ_HIGHPRI_PENDING; | |
1764 | } | |
1765 | ||
fb0e7beb TH |
1766 | /* |
1767 | * CPU intensive works don't participate in concurrency | |
1768 | * management. They're the scheduler's responsibility. | |
1769 | */ | |
1770 | if (unlikely(cpu_intensive)) | |
1771 | worker_set_flags(worker, WORKER_CPU_INTENSIVE, true); | |
1772 | ||
8b03ae3c | 1773 | spin_unlock_irq(&gcwq->lock); |
a62428c0 | 1774 | |
a62428c0 TH |
1775 | work_clear_pending(work); |
1776 | lock_map_acquire(&cwq->wq->lockdep_map); | |
1777 | lock_map_acquire(&lockdep_map); | |
1778 | f(work); | |
1779 | lock_map_release(&lockdep_map); | |
1780 | lock_map_release(&cwq->wq->lockdep_map); | |
1781 | ||
1782 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
1783 | printk(KERN_ERR "BUG: workqueue leaked lock or atomic: " | |
1784 | "%s/0x%08x/%d\n", | |
1785 | current->comm, preempt_count(), task_pid_nr(current)); | |
1786 | printk(KERN_ERR " last function: "); | |
1787 | print_symbol("%s\n", (unsigned long)f); | |
1788 | debug_show_held_locks(current); | |
1789 | dump_stack(); | |
1790 | } | |
1791 | ||
8b03ae3c | 1792 | spin_lock_irq(&gcwq->lock); |
a62428c0 | 1793 | |
fb0e7beb TH |
1794 | /* clear cpu intensive status */ |
1795 | if (unlikely(cpu_intensive)) | |
1796 | worker_clr_flags(worker, WORKER_CPU_INTENSIVE); | |
1797 | ||
a62428c0 | 1798 | /* we're done with it, release */ |
c8e55f36 | 1799 | hlist_del_init(&worker->hentry); |
c34056a3 | 1800 | worker->current_work = NULL; |
8cca0eea | 1801 | worker->current_cwq = NULL; |
73f53c4a | 1802 | cwq_dec_nr_in_flight(cwq, work_color); |
a62428c0 TH |
1803 | } |
1804 | ||
affee4b2 TH |
1805 | /** |
1806 | * process_scheduled_works - process scheduled works | |
1807 | * @worker: self | |
1808 | * | |
1809 | * Process all scheduled works. Please note that the scheduled list | |
1810 | * may change while processing a work, so this function repeatedly | |
1811 | * fetches a work from the top and executes it. | |
1812 | * | |
1813 | * CONTEXT: | |
8b03ae3c | 1814 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed |
affee4b2 TH |
1815 | * multiple times. |
1816 | */ | |
1817 | static void process_scheduled_works(struct worker *worker) | |
1da177e4 | 1818 | { |
affee4b2 TH |
1819 | while (!list_empty(&worker->scheduled)) { |
1820 | struct work_struct *work = list_first_entry(&worker->scheduled, | |
1da177e4 | 1821 | struct work_struct, entry); |
c34056a3 | 1822 | process_one_work(worker, work); |
1da177e4 | 1823 | } |
1da177e4 LT |
1824 | } |
1825 | ||
4690c4ab TH |
1826 | /** |
1827 | * worker_thread - the worker thread function | |
c34056a3 | 1828 | * @__worker: self |
4690c4ab | 1829 | * |
e22bee78 TH |
1830 | * The gcwq worker thread function. There's a single dynamic pool of |
1831 | * these per each cpu. These workers process all works regardless of | |
1832 | * their specific target workqueue. The only exception is works which | |
1833 | * belong to workqueues with a rescuer which will be explained in | |
1834 | * rescuer_thread(). | |
4690c4ab | 1835 | */ |
c34056a3 | 1836 | static int worker_thread(void *__worker) |
1da177e4 | 1837 | { |
c34056a3 | 1838 | struct worker *worker = __worker; |
8b03ae3c | 1839 | struct global_cwq *gcwq = worker->gcwq; |
1da177e4 | 1840 | |
e22bee78 TH |
1841 | /* tell the scheduler that this is a workqueue worker */ |
1842 | worker->task->flags |= PF_WQ_WORKER; | |
c8e55f36 | 1843 | woke_up: |
c8e55f36 | 1844 | spin_lock_irq(&gcwq->lock); |
1da177e4 | 1845 | |
c8e55f36 TH |
1846 | /* DIE can be set only while we're idle, checking here is enough */ |
1847 | if (worker->flags & WORKER_DIE) { | |
1848 | spin_unlock_irq(&gcwq->lock); | |
e22bee78 | 1849 | worker->task->flags &= ~PF_WQ_WORKER; |
c8e55f36 TH |
1850 | return 0; |
1851 | } | |
affee4b2 | 1852 | |
c8e55f36 | 1853 | worker_leave_idle(worker); |
db7bccf4 | 1854 | recheck: |
e22bee78 TH |
1855 | /* no more worker necessary? */ |
1856 | if (!need_more_worker(gcwq)) | |
1857 | goto sleep; | |
1858 | ||
1859 | /* do we need to manage? */ | |
1860 | if (unlikely(!may_start_working(gcwq)) && manage_workers(worker)) | |
1861 | goto recheck; | |
1862 | ||
c8e55f36 TH |
1863 | /* |
1864 | * ->scheduled list can only be filled while a worker is | |
1865 | * preparing to process a work or actually processing it. | |
1866 | * Make sure nobody diddled with it while I was sleeping. | |
1867 | */ | |
1868 | BUG_ON(!list_empty(&worker->scheduled)); | |
1869 | ||
e22bee78 TH |
1870 | /* |
1871 | * When control reaches this point, we're guaranteed to have | |
1872 | * at least one idle worker or that someone else has already | |
1873 | * assumed the manager role. | |
1874 | */ | |
1875 | worker_clr_flags(worker, WORKER_PREP); | |
1876 | ||
1877 | do { | |
c8e55f36 | 1878 | struct work_struct *work = |
7e11629d | 1879 | list_first_entry(&gcwq->worklist, |
c8e55f36 TH |
1880 | struct work_struct, entry); |
1881 | ||
1882 | if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) { | |
1883 | /* optimization path, not strictly necessary */ | |
1884 | process_one_work(worker, work); | |
1885 | if (unlikely(!list_empty(&worker->scheduled))) | |
affee4b2 | 1886 | process_scheduled_works(worker); |
c8e55f36 TH |
1887 | } else { |
1888 | move_linked_works(work, &worker->scheduled, NULL); | |
1889 | process_scheduled_works(worker); | |
affee4b2 | 1890 | } |
e22bee78 TH |
1891 | } while (keep_working(gcwq)); |
1892 | ||
1893 | worker_set_flags(worker, WORKER_PREP, false); | |
d313dd85 | 1894 | sleep: |
e22bee78 TH |
1895 | if (unlikely(need_to_manage_workers(gcwq)) && manage_workers(worker)) |
1896 | goto recheck; | |
d313dd85 | 1897 | |
c8e55f36 | 1898 | /* |
e22bee78 TH |
1899 | * gcwq->lock is held and there's no work to process and no |
1900 | * need to manage, sleep. Workers are woken up only while | |
1901 | * holding gcwq->lock or from local cpu, so setting the | |
1902 | * current state before releasing gcwq->lock is enough to | |
1903 | * prevent losing any event. | |
c8e55f36 TH |
1904 | */ |
1905 | worker_enter_idle(worker); | |
1906 | __set_current_state(TASK_INTERRUPTIBLE); | |
1907 | spin_unlock_irq(&gcwq->lock); | |
1908 | schedule(); | |
1909 | goto woke_up; | |
1da177e4 LT |
1910 | } |
1911 | ||
e22bee78 TH |
1912 | /** |
1913 | * rescuer_thread - the rescuer thread function | |
1914 | * @__wq: the associated workqueue | |
1915 | * | |
1916 | * Workqueue rescuer thread function. There's one rescuer for each | |
1917 | * workqueue which has WQ_RESCUER set. | |
1918 | * | |
1919 | * Regular work processing on a gcwq may block trying to create a new | |
1920 | * worker which uses GFP_KERNEL allocation which has slight chance of | |
1921 | * developing into deadlock if some works currently on the same queue | |
1922 | * need to be processed to satisfy the GFP_KERNEL allocation. This is | |
1923 | * the problem rescuer solves. | |
1924 | * | |
1925 | * When such condition is possible, the gcwq summons rescuers of all | |
1926 | * workqueues which have works queued on the gcwq and let them process | |
1927 | * those works so that forward progress can be guaranteed. | |
1928 | * | |
1929 | * This should happen rarely. | |
1930 | */ | |
1931 | static int rescuer_thread(void *__wq) | |
1932 | { | |
1933 | struct workqueue_struct *wq = __wq; | |
1934 | struct worker *rescuer = wq->rescuer; | |
1935 | struct list_head *scheduled = &rescuer->scheduled; | |
f3421797 | 1936 | bool is_unbound = wq->flags & WQ_UNBOUND; |
e22bee78 TH |
1937 | unsigned int cpu; |
1938 | ||
1939 | set_user_nice(current, RESCUER_NICE_LEVEL); | |
1940 | repeat: | |
1941 | set_current_state(TASK_INTERRUPTIBLE); | |
1942 | ||
1943 | if (kthread_should_stop()) | |
1944 | return 0; | |
1945 | ||
f3421797 TH |
1946 | /* |
1947 | * See whether any cpu is asking for help. Unbounded | |
1948 | * workqueues use cpu 0 in mayday_mask for CPU_UNBOUND. | |
1949 | */ | |
f2e005aa | 1950 | for_each_mayday_cpu(cpu, wq->mayday_mask) { |
f3421797 TH |
1951 | unsigned int tcpu = is_unbound ? WORK_CPU_UNBOUND : cpu; |
1952 | struct cpu_workqueue_struct *cwq = get_cwq(tcpu, wq); | |
e22bee78 TH |
1953 | struct global_cwq *gcwq = cwq->gcwq; |
1954 | struct work_struct *work, *n; | |
1955 | ||
1956 | __set_current_state(TASK_RUNNING); | |
f2e005aa | 1957 | mayday_clear_cpu(cpu, wq->mayday_mask); |
e22bee78 TH |
1958 | |
1959 | /* migrate to the target cpu if possible */ | |
1960 | rescuer->gcwq = gcwq; | |
1961 | worker_maybe_bind_and_lock(rescuer); | |
1962 | ||
1963 | /* | |
1964 | * Slurp in all works issued via this workqueue and | |
1965 | * process'em. | |
1966 | */ | |
1967 | BUG_ON(!list_empty(&rescuer->scheduled)); | |
1968 | list_for_each_entry_safe(work, n, &gcwq->worklist, entry) | |
1969 | if (get_work_cwq(work) == cwq) | |
1970 | move_linked_works(work, scheduled, &n); | |
1971 | ||
1972 | process_scheduled_works(rescuer); | |
1973 | spin_unlock_irq(&gcwq->lock); | |
1974 | } | |
1975 | ||
1976 | schedule(); | |
1977 | goto repeat; | |
1978 | } | |
1979 | ||
fc2e4d70 ON |
1980 | struct wq_barrier { |
1981 | struct work_struct work; | |
1982 | struct completion done; | |
1983 | }; | |
1984 | ||
1985 | static void wq_barrier_func(struct work_struct *work) | |
1986 | { | |
1987 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
1988 | complete(&barr->done); | |
1989 | } | |
1990 | ||
4690c4ab TH |
1991 | /** |
1992 | * insert_wq_barrier - insert a barrier work | |
1993 | * @cwq: cwq to insert barrier into | |
1994 | * @barr: wq_barrier to insert | |
affee4b2 TH |
1995 | * @target: target work to attach @barr to |
1996 | * @worker: worker currently executing @target, NULL if @target is not executing | |
4690c4ab | 1997 | * |
affee4b2 TH |
1998 | * @barr is linked to @target such that @barr is completed only after |
1999 | * @target finishes execution. Please note that the ordering | |
2000 | * guarantee is observed only with respect to @target and on the local | |
2001 | * cpu. | |
2002 | * | |
2003 | * Currently, a queued barrier can't be canceled. This is because | |
2004 | * try_to_grab_pending() can't determine whether the work to be | |
2005 | * grabbed is at the head of the queue and thus can't clear LINKED | |
2006 | * flag of the previous work while there must be a valid next work | |
2007 | * after a work with LINKED flag set. | |
2008 | * | |
2009 | * Note that when @worker is non-NULL, @target may be modified | |
2010 | * underneath us, so we can't reliably determine cwq from @target. | |
4690c4ab TH |
2011 | * |
2012 | * CONTEXT: | |
8b03ae3c | 2013 | * spin_lock_irq(gcwq->lock). |
4690c4ab | 2014 | */ |
83c22520 | 2015 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, |
affee4b2 TH |
2016 | struct wq_barrier *barr, |
2017 | struct work_struct *target, struct worker *worker) | |
fc2e4d70 | 2018 | { |
affee4b2 TH |
2019 | struct list_head *head; |
2020 | unsigned int linked = 0; | |
2021 | ||
dc186ad7 | 2022 | /* |
8b03ae3c | 2023 | * debugobject calls are safe here even with gcwq->lock locked |
dc186ad7 TG |
2024 | * as we know for sure that this will not trigger any of the |
2025 | * checks and call back into the fixup functions where we | |
2026 | * might deadlock. | |
2027 | */ | |
2028 | INIT_WORK_ON_STACK(&barr->work, wq_barrier_func); | |
22df02bb | 2029 | __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work)); |
fc2e4d70 | 2030 | init_completion(&barr->done); |
83c22520 | 2031 | |
affee4b2 TH |
2032 | /* |
2033 | * If @target is currently being executed, schedule the | |
2034 | * barrier to the worker; otherwise, put it after @target. | |
2035 | */ | |
2036 | if (worker) | |
2037 | head = worker->scheduled.next; | |
2038 | else { | |
2039 | unsigned long *bits = work_data_bits(target); | |
2040 | ||
2041 | head = target->entry.next; | |
2042 | /* there can already be other linked works, inherit and set */ | |
2043 | linked = *bits & WORK_STRUCT_LINKED; | |
2044 | __set_bit(WORK_STRUCT_LINKED_BIT, bits); | |
2045 | } | |
2046 | ||
dc186ad7 | 2047 | debug_work_activate(&barr->work); |
affee4b2 TH |
2048 | insert_work(cwq, &barr->work, head, |
2049 | work_color_to_flags(WORK_NO_COLOR) | linked); | |
fc2e4d70 ON |
2050 | } |
2051 | ||
73f53c4a TH |
2052 | /** |
2053 | * flush_workqueue_prep_cwqs - prepare cwqs for workqueue flushing | |
2054 | * @wq: workqueue being flushed | |
2055 | * @flush_color: new flush color, < 0 for no-op | |
2056 | * @work_color: new work color, < 0 for no-op | |
2057 | * | |
2058 | * Prepare cwqs for workqueue flushing. | |
2059 | * | |
2060 | * If @flush_color is non-negative, flush_color on all cwqs should be | |
2061 | * -1. If no cwq has in-flight commands at the specified color, all | |
2062 | * cwq->flush_color's stay at -1 and %false is returned. If any cwq | |
2063 | * has in flight commands, its cwq->flush_color is set to | |
2064 | * @flush_color, @wq->nr_cwqs_to_flush is updated accordingly, cwq | |
2065 | * wakeup logic is armed and %true is returned. | |
2066 | * | |
2067 | * The caller should have initialized @wq->first_flusher prior to | |
2068 | * calling this function with non-negative @flush_color. If | |
2069 | * @flush_color is negative, no flush color update is done and %false | |
2070 | * is returned. | |
2071 | * | |
2072 | * If @work_color is non-negative, all cwqs should have the same | |
2073 | * work_color which is previous to @work_color and all will be | |
2074 | * advanced to @work_color. | |
2075 | * | |
2076 | * CONTEXT: | |
2077 | * mutex_lock(wq->flush_mutex). | |
2078 | * | |
2079 | * RETURNS: | |
2080 | * %true if @flush_color >= 0 and there's something to flush. %false | |
2081 | * otherwise. | |
2082 | */ | |
2083 | static bool flush_workqueue_prep_cwqs(struct workqueue_struct *wq, | |
2084 | int flush_color, int work_color) | |
1da177e4 | 2085 | { |
73f53c4a TH |
2086 | bool wait = false; |
2087 | unsigned int cpu; | |
1da177e4 | 2088 | |
73f53c4a TH |
2089 | if (flush_color >= 0) { |
2090 | BUG_ON(atomic_read(&wq->nr_cwqs_to_flush)); | |
2091 | atomic_set(&wq->nr_cwqs_to_flush, 1); | |
1da177e4 | 2092 | } |
2355b70f | 2093 | |
f3421797 | 2094 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a | 2095 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
8b03ae3c | 2096 | struct global_cwq *gcwq = cwq->gcwq; |
73f53c4a | 2097 | |
8b03ae3c | 2098 | spin_lock_irq(&gcwq->lock); |
73f53c4a TH |
2099 | |
2100 | if (flush_color >= 0) { | |
2101 | BUG_ON(cwq->flush_color != -1); | |
2102 | ||
2103 | if (cwq->nr_in_flight[flush_color]) { | |
2104 | cwq->flush_color = flush_color; | |
2105 | atomic_inc(&wq->nr_cwqs_to_flush); | |
2106 | wait = true; | |
2107 | } | |
2108 | } | |
2109 | ||
2110 | if (work_color >= 0) { | |
2111 | BUG_ON(work_color != work_next_color(cwq->work_color)); | |
2112 | cwq->work_color = work_color; | |
2113 | } | |
2114 | ||
8b03ae3c | 2115 | spin_unlock_irq(&gcwq->lock); |
dc186ad7 | 2116 | } |
14441960 | 2117 | |
73f53c4a TH |
2118 | if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_cwqs_to_flush)) |
2119 | complete(&wq->first_flusher->done); | |
2120 | ||
2121 | return wait; | |
1da177e4 LT |
2122 | } |
2123 | ||
0fcb78c2 | 2124 | /** |
1da177e4 | 2125 | * flush_workqueue - ensure that any scheduled work has run to completion. |
0fcb78c2 | 2126 | * @wq: workqueue to flush |
1da177e4 LT |
2127 | * |
2128 | * Forces execution of the workqueue and blocks until its completion. | |
2129 | * This is typically used in driver shutdown handlers. | |
2130 | * | |
fc2e4d70 ON |
2131 | * We sleep until all works which were queued on entry have been handled, |
2132 | * but we are not livelocked by new incoming ones. | |
1da177e4 | 2133 | */ |
7ad5b3a5 | 2134 | void flush_workqueue(struct workqueue_struct *wq) |
1da177e4 | 2135 | { |
73f53c4a TH |
2136 | struct wq_flusher this_flusher = { |
2137 | .list = LIST_HEAD_INIT(this_flusher.list), | |
2138 | .flush_color = -1, | |
2139 | .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done), | |
2140 | }; | |
2141 | int next_color; | |
1da177e4 | 2142 | |
3295f0ef IM |
2143 | lock_map_acquire(&wq->lockdep_map); |
2144 | lock_map_release(&wq->lockdep_map); | |
73f53c4a TH |
2145 | |
2146 | mutex_lock(&wq->flush_mutex); | |
2147 | ||
2148 | /* | |
2149 | * Start-to-wait phase | |
2150 | */ | |
2151 | next_color = work_next_color(wq->work_color); | |
2152 | ||
2153 | if (next_color != wq->flush_color) { | |
2154 | /* | |
2155 | * Color space is not full. The current work_color | |
2156 | * becomes our flush_color and work_color is advanced | |
2157 | * by one. | |
2158 | */ | |
2159 | BUG_ON(!list_empty(&wq->flusher_overflow)); | |
2160 | this_flusher.flush_color = wq->work_color; | |
2161 | wq->work_color = next_color; | |
2162 | ||
2163 | if (!wq->first_flusher) { | |
2164 | /* no flush in progress, become the first flusher */ | |
2165 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2166 | ||
2167 | wq->first_flusher = &this_flusher; | |
2168 | ||
2169 | if (!flush_workqueue_prep_cwqs(wq, wq->flush_color, | |
2170 | wq->work_color)) { | |
2171 | /* nothing to flush, done */ | |
2172 | wq->flush_color = next_color; | |
2173 | wq->first_flusher = NULL; | |
2174 | goto out_unlock; | |
2175 | } | |
2176 | } else { | |
2177 | /* wait in queue */ | |
2178 | BUG_ON(wq->flush_color == this_flusher.flush_color); | |
2179 | list_add_tail(&this_flusher.list, &wq->flusher_queue); | |
2180 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2181 | } | |
2182 | } else { | |
2183 | /* | |
2184 | * Oops, color space is full, wait on overflow queue. | |
2185 | * The next flush completion will assign us | |
2186 | * flush_color and transfer to flusher_queue. | |
2187 | */ | |
2188 | list_add_tail(&this_flusher.list, &wq->flusher_overflow); | |
2189 | } | |
2190 | ||
2191 | mutex_unlock(&wq->flush_mutex); | |
2192 | ||
2193 | wait_for_completion(&this_flusher.done); | |
2194 | ||
2195 | /* | |
2196 | * Wake-up-and-cascade phase | |
2197 | * | |
2198 | * First flushers are responsible for cascading flushes and | |
2199 | * handling overflow. Non-first flushers can simply return. | |
2200 | */ | |
2201 | if (wq->first_flusher != &this_flusher) | |
2202 | return; | |
2203 | ||
2204 | mutex_lock(&wq->flush_mutex); | |
2205 | ||
4ce48b37 TH |
2206 | /* we might have raced, check again with mutex held */ |
2207 | if (wq->first_flusher != &this_flusher) | |
2208 | goto out_unlock; | |
2209 | ||
73f53c4a TH |
2210 | wq->first_flusher = NULL; |
2211 | ||
2212 | BUG_ON(!list_empty(&this_flusher.list)); | |
2213 | BUG_ON(wq->flush_color != this_flusher.flush_color); | |
2214 | ||
2215 | while (true) { | |
2216 | struct wq_flusher *next, *tmp; | |
2217 | ||
2218 | /* complete all the flushers sharing the current flush color */ | |
2219 | list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) { | |
2220 | if (next->flush_color != wq->flush_color) | |
2221 | break; | |
2222 | list_del_init(&next->list); | |
2223 | complete(&next->done); | |
2224 | } | |
2225 | ||
2226 | BUG_ON(!list_empty(&wq->flusher_overflow) && | |
2227 | wq->flush_color != work_next_color(wq->work_color)); | |
2228 | ||
2229 | /* this flush_color is finished, advance by one */ | |
2230 | wq->flush_color = work_next_color(wq->flush_color); | |
2231 | ||
2232 | /* one color has been freed, handle overflow queue */ | |
2233 | if (!list_empty(&wq->flusher_overflow)) { | |
2234 | /* | |
2235 | * Assign the same color to all overflowed | |
2236 | * flushers, advance work_color and append to | |
2237 | * flusher_queue. This is the start-to-wait | |
2238 | * phase for these overflowed flushers. | |
2239 | */ | |
2240 | list_for_each_entry(tmp, &wq->flusher_overflow, list) | |
2241 | tmp->flush_color = wq->work_color; | |
2242 | ||
2243 | wq->work_color = work_next_color(wq->work_color); | |
2244 | ||
2245 | list_splice_tail_init(&wq->flusher_overflow, | |
2246 | &wq->flusher_queue); | |
2247 | flush_workqueue_prep_cwqs(wq, -1, wq->work_color); | |
2248 | } | |
2249 | ||
2250 | if (list_empty(&wq->flusher_queue)) { | |
2251 | BUG_ON(wq->flush_color != wq->work_color); | |
2252 | break; | |
2253 | } | |
2254 | ||
2255 | /* | |
2256 | * Need to flush more colors. Make the next flusher | |
2257 | * the new first flusher and arm cwqs. | |
2258 | */ | |
2259 | BUG_ON(wq->flush_color == wq->work_color); | |
2260 | BUG_ON(wq->flush_color != next->flush_color); | |
2261 | ||
2262 | list_del_init(&next->list); | |
2263 | wq->first_flusher = next; | |
2264 | ||
2265 | if (flush_workqueue_prep_cwqs(wq, wq->flush_color, -1)) | |
2266 | break; | |
2267 | ||
2268 | /* | |
2269 | * Meh... this color is already done, clear first | |
2270 | * flusher and repeat cascading. | |
2271 | */ | |
2272 | wq->first_flusher = NULL; | |
2273 | } | |
2274 | ||
2275 | out_unlock: | |
2276 | mutex_unlock(&wq->flush_mutex); | |
1da177e4 | 2277 | } |
ae90dd5d | 2278 | EXPORT_SYMBOL_GPL(flush_workqueue); |
1da177e4 | 2279 | |
db700897 ON |
2280 | /** |
2281 | * flush_work - block until a work_struct's callback has terminated | |
2282 | * @work: the work which is to be flushed | |
2283 | * | |
a67da70d ON |
2284 | * Returns false if @work has already terminated. |
2285 | * | |
db700897 ON |
2286 | * It is expected that, prior to calling flush_work(), the caller has |
2287 | * arranged for the work to not be requeued, otherwise it doesn't make | |
2288 | * sense to use this function. | |
2289 | */ | |
2290 | int flush_work(struct work_struct *work) | |
2291 | { | |
affee4b2 | 2292 | struct worker *worker = NULL; |
8b03ae3c | 2293 | struct global_cwq *gcwq; |
7a22ad75 | 2294 | struct cpu_workqueue_struct *cwq; |
db700897 ON |
2295 | struct wq_barrier barr; |
2296 | ||
2297 | might_sleep(); | |
7a22ad75 TH |
2298 | gcwq = get_work_gcwq(work); |
2299 | if (!gcwq) | |
db700897 | 2300 | return 0; |
a67da70d | 2301 | |
8b03ae3c | 2302 | spin_lock_irq(&gcwq->lock); |
db700897 ON |
2303 | if (!list_empty(&work->entry)) { |
2304 | /* | |
2305 | * See the comment near try_to_grab_pending()->smp_rmb(). | |
7a22ad75 TH |
2306 | * If it was re-queued to a different gcwq under us, we |
2307 | * are not going to wait. | |
db700897 ON |
2308 | */ |
2309 | smp_rmb(); | |
7a22ad75 TH |
2310 | cwq = get_work_cwq(work); |
2311 | if (unlikely(!cwq || gcwq != cwq->gcwq)) | |
4690c4ab | 2312 | goto already_gone; |
db700897 | 2313 | } else { |
7a22ad75 | 2314 | worker = find_worker_executing_work(gcwq, work); |
affee4b2 | 2315 | if (!worker) |
4690c4ab | 2316 | goto already_gone; |
7a22ad75 | 2317 | cwq = worker->current_cwq; |
db700897 | 2318 | } |
db700897 | 2319 | |
affee4b2 | 2320 | insert_wq_barrier(cwq, &barr, work, worker); |
8b03ae3c | 2321 | spin_unlock_irq(&gcwq->lock); |
7a22ad75 TH |
2322 | |
2323 | lock_map_acquire(&cwq->wq->lockdep_map); | |
2324 | lock_map_release(&cwq->wq->lockdep_map); | |
2325 | ||
db700897 | 2326 | wait_for_completion(&barr.done); |
dc186ad7 | 2327 | destroy_work_on_stack(&barr.work); |
db700897 | 2328 | return 1; |
4690c4ab | 2329 | already_gone: |
8b03ae3c | 2330 | spin_unlock_irq(&gcwq->lock); |
4690c4ab | 2331 | return 0; |
db700897 ON |
2332 | } |
2333 | EXPORT_SYMBOL_GPL(flush_work); | |
2334 | ||
6e84d644 | 2335 | /* |
1f1f642e | 2336 | * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit, |
6e84d644 ON |
2337 | * so this work can't be re-armed in any way. |
2338 | */ | |
2339 | static int try_to_grab_pending(struct work_struct *work) | |
2340 | { | |
8b03ae3c | 2341 | struct global_cwq *gcwq; |
1f1f642e | 2342 | int ret = -1; |
6e84d644 | 2343 | |
22df02bb | 2344 | if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) |
1f1f642e | 2345 | return 0; |
6e84d644 ON |
2346 | |
2347 | /* | |
2348 | * The queueing is in progress, or it is already queued. Try to | |
2349 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
2350 | */ | |
7a22ad75 TH |
2351 | gcwq = get_work_gcwq(work); |
2352 | if (!gcwq) | |
6e84d644 ON |
2353 | return ret; |
2354 | ||
8b03ae3c | 2355 | spin_lock_irq(&gcwq->lock); |
6e84d644 ON |
2356 | if (!list_empty(&work->entry)) { |
2357 | /* | |
7a22ad75 | 2358 | * This work is queued, but perhaps we locked the wrong gcwq. |
6e84d644 ON |
2359 | * In that case we must see the new value after rmb(), see |
2360 | * insert_work()->wmb(). | |
2361 | */ | |
2362 | smp_rmb(); | |
7a22ad75 | 2363 | if (gcwq == get_work_gcwq(work)) { |
dc186ad7 | 2364 | debug_work_deactivate(work); |
6e84d644 | 2365 | list_del_init(&work->entry); |
7a22ad75 TH |
2366 | cwq_dec_nr_in_flight(get_work_cwq(work), |
2367 | get_work_color(work)); | |
6e84d644 ON |
2368 | ret = 1; |
2369 | } | |
2370 | } | |
8b03ae3c | 2371 | spin_unlock_irq(&gcwq->lock); |
6e84d644 ON |
2372 | |
2373 | return ret; | |
2374 | } | |
2375 | ||
7a22ad75 | 2376 | static void wait_on_cpu_work(struct global_cwq *gcwq, struct work_struct *work) |
b89deed3 ON |
2377 | { |
2378 | struct wq_barrier barr; | |
affee4b2 | 2379 | struct worker *worker; |
b89deed3 | 2380 | |
8b03ae3c | 2381 | spin_lock_irq(&gcwq->lock); |
affee4b2 | 2382 | |
7a22ad75 TH |
2383 | worker = find_worker_executing_work(gcwq, work); |
2384 | if (unlikely(worker)) | |
2385 | insert_wq_barrier(worker->current_cwq, &barr, work, worker); | |
affee4b2 | 2386 | |
8b03ae3c | 2387 | spin_unlock_irq(&gcwq->lock); |
b89deed3 | 2388 | |
affee4b2 | 2389 | if (unlikely(worker)) { |
b89deed3 | 2390 | wait_for_completion(&barr.done); |
dc186ad7 TG |
2391 | destroy_work_on_stack(&barr.work); |
2392 | } | |
b89deed3 ON |
2393 | } |
2394 | ||
6e84d644 | 2395 | static void wait_on_work(struct work_struct *work) |
b89deed3 | 2396 | { |
b1f4ec17 | 2397 | int cpu; |
b89deed3 | 2398 | |
f293ea92 ON |
2399 | might_sleep(); |
2400 | ||
3295f0ef IM |
2401 | lock_map_acquire(&work->lockdep_map); |
2402 | lock_map_release(&work->lockdep_map); | |
4e6045f1 | 2403 | |
f3421797 | 2404 | for_each_gcwq_cpu(cpu) |
7a22ad75 | 2405 | wait_on_cpu_work(get_gcwq(cpu), work); |
6e84d644 ON |
2406 | } |
2407 | ||
1f1f642e ON |
2408 | static int __cancel_work_timer(struct work_struct *work, |
2409 | struct timer_list* timer) | |
2410 | { | |
2411 | int ret; | |
2412 | ||
2413 | do { | |
2414 | ret = (timer && likely(del_timer(timer))); | |
2415 | if (!ret) | |
2416 | ret = try_to_grab_pending(work); | |
2417 | wait_on_work(work); | |
2418 | } while (unlikely(ret < 0)); | |
2419 | ||
7a22ad75 | 2420 | clear_work_data(work); |
1f1f642e ON |
2421 | return ret; |
2422 | } | |
2423 | ||
6e84d644 ON |
2424 | /** |
2425 | * cancel_work_sync - block until a work_struct's callback has terminated | |
2426 | * @work: the work which is to be flushed | |
2427 | * | |
1f1f642e ON |
2428 | * Returns true if @work was pending. |
2429 | * | |
6e84d644 ON |
2430 | * cancel_work_sync() will cancel the work if it is queued. If the work's |
2431 | * callback appears to be running, cancel_work_sync() will block until it | |
2432 | * has completed. | |
2433 | * | |
2434 | * It is possible to use this function if the work re-queues itself. It can | |
2435 | * cancel the work even if it migrates to another workqueue, however in that | |
2436 | * case it only guarantees that work->func() has completed on the last queued | |
2437 | * workqueue. | |
2438 | * | |
2439 | * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not | |
2440 | * pending, otherwise it goes into a busy-wait loop until the timer expires. | |
2441 | * | |
2442 | * The caller must ensure that workqueue_struct on which this work was last | |
2443 | * queued can't be destroyed before this function returns. | |
2444 | */ | |
1f1f642e | 2445 | int cancel_work_sync(struct work_struct *work) |
6e84d644 | 2446 | { |
1f1f642e | 2447 | return __cancel_work_timer(work, NULL); |
b89deed3 | 2448 | } |
28e53bdd | 2449 | EXPORT_SYMBOL_GPL(cancel_work_sync); |
b89deed3 | 2450 | |
6e84d644 | 2451 | /** |
f5a421a4 | 2452 | * cancel_delayed_work_sync - reliably kill off a delayed work. |
6e84d644 ON |
2453 | * @dwork: the delayed work struct |
2454 | * | |
1f1f642e ON |
2455 | * Returns true if @dwork was pending. |
2456 | * | |
6e84d644 ON |
2457 | * It is possible to use this function if @dwork rearms itself via queue_work() |
2458 | * or queue_delayed_work(). See also the comment for cancel_work_sync(). | |
2459 | */ | |
1f1f642e | 2460 | int cancel_delayed_work_sync(struct delayed_work *dwork) |
6e84d644 | 2461 | { |
1f1f642e | 2462 | return __cancel_work_timer(&dwork->work, &dwork->timer); |
6e84d644 | 2463 | } |
f5a421a4 | 2464 | EXPORT_SYMBOL(cancel_delayed_work_sync); |
1da177e4 | 2465 | |
0fcb78c2 REB |
2466 | /** |
2467 | * schedule_work - put work task in global workqueue | |
2468 | * @work: job to be done | |
2469 | * | |
5b0f437d BVA |
2470 | * Returns zero if @work was already on the kernel-global workqueue and |
2471 | * non-zero otherwise. | |
2472 | * | |
2473 | * This puts a job in the kernel-global workqueue if it was not already | |
2474 | * queued and leaves it in the same position on the kernel-global | |
2475 | * workqueue otherwise. | |
0fcb78c2 | 2476 | */ |
7ad5b3a5 | 2477 | int schedule_work(struct work_struct *work) |
1da177e4 | 2478 | { |
d320c038 | 2479 | return queue_work(system_wq, work); |
1da177e4 | 2480 | } |
ae90dd5d | 2481 | EXPORT_SYMBOL(schedule_work); |
1da177e4 | 2482 | |
c1a220e7 ZR |
2483 | /* |
2484 | * schedule_work_on - put work task on a specific cpu | |
2485 | * @cpu: cpu to put the work task on | |
2486 | * @work: job to be done | |
2487 | * | |
2488 | * This puts a job on a specific cpu | |
2489 | */ | |
2490 | int schedule_work_on(int cpu, struct work_struct *work) | |
2491 | { | |
d320c038 | 2492 | return queue_work_on(cpu, system_wq, work); |
c1a220e7 ZR |
2493 | } |
2494 | EXPORT_SYMBOL(schedule_work_on); | |
2495 | ||
0fcb78c2 REB |
2496 | /** |
2497 | * schedule_delayed_work - put work task in global workqueue after delay | |
52bad64d DH |
2498 | * @dwork: job to be done |
2499 | * @delay: number of jiffies to wait or 0 for immediate execution | |
0fcb78c2 REB |
2500 | * |
2501 | * After waiting for a given time this puts a job in the kernel-global | |
2502 | * workqueue. | |
2503 | */ | |
7ad5b3a5 | 2504 | int schedule_delayed_work(struct delayed_work *dwork, |
82f67cd9 | 2505 | unsigned long delay) |
1da177e4 | 2506 | { |
d320c038 | 2507 | return queue_delayed_work(system_wq, dwork, delay); |
1da177e4 | 2508 | } |
ae90dd5d | 2509 | EXPORT_SYMBOL(schedule_delayed_work); |
1da177e4 | 2510 | |
8c53e463 LT |
2511 | /** |
2512 | * flush_delayed_work - block until a dwork_struct's callback has terminated | |
2513 | * @dwork: the delayed work which is to be flushed | |
2514 | * | |
2515 | * Any timeout is cancelled, and any pending work is run immediately. | |
2516 | */ | |
2517 | void flush_delayed_work(struct delayed_work *dwork) | |
2518 | { | |
2519 | if (del_timer_sync(&dwork->timer)) { | |
7a22ad75 | 2520 | __queue_work(get_cpu(), get_work_cwq(&dwork->work)->wq, |
4690c4ab | 2521 | &dwork->work); |
8c53e463 LT |
2522 | put_cpu(); |
2523 | } | |
2524 | flush_work(&dwork->work); | |
2525 | } | |
2526 | EXPORT_SYMBOL(flush_delayed_work); | |
2527 | ||
0fcb78c2 REB |
2528 | /** |
2529 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
2530 | * @cpu: cpu to use | |
52bad64d | 2531 | * @dwork: job to be done |
0fcb78c2 REB |
2532 | * @delay: number of jiffies to wait |
2533 | * | |
2534 | * After waiting for a given time this puts a job in the kernel-global | |
2535 | * workqueue on the specified CPU. | |
2536 | */ | |
1da177e4 | 2537 | int schedule_delayed_work_on(int cpu, |
52bad64d | 2538 | struct delayed_work *dwork, unsigned long delay) |
1da177e4 | 2539 | { |
d320c038 | 2540 | return queue_delayed_work_on(cpu, system_wq, dwork, delay); |
1da177e4 | 2541 | } |
ae90dd5d | 2542 | EXPORT_SYMBOL(schedule_delayed_work_on); |
1da177e4 | 2543 | |
b6136773 AM |
2544 | /** |
2545 | * schedule_on_each_cpu - call a function on each online CPU from keventd | |
2546 | * @func: the function to call | |
b6136773 AM |
2547 | * |
2548 | * Returns zero on success. | |
2549 | * Returns -ve errno on failure. | |
2550 | * | |
b6136773 AM |
2551 | * schedule_on_each_cpu() is very slow. |
2552 | */ | |
65f27f38 | 2553 | int schedule_on_each_cpu(work_func_t func) |
15316ba8 CL |
2554 | { |
2555 | int cpu; | |
b6136773 | 2556 | struct work_struct *works; |
15316ba8 | 2557 | |
b6136773 AM |
2558 | works = alloc_percpu(struct work_struct); |
2559 | if (!works) | |
15316ba8 | 2560 | return -ENOMEM; |
b6136773 | 2561 | |
93981800 TH |
2562 | get_online_cpus(); |
2563 | ||
15316ba8 | 2564 | for_each_online_cpu(cpu) { |
9bfb1839 IM |
2565 | struct work_struct *work = per_cpu_ptr(works, cpu); |
2566 | ||
2567 | INIT_WORK(work, func); | |
b71ab8c2 | 2568 | schedule_work_on(cpu, work); |
65a64464 | 2569 | } |
93981800 TH |
2570 | |
2571 | for_each_online_cpu(cpu) | |
2572 | flush_work(per_cpu_ptr(works, cpu)); | |
2573 | ||
95402b38 | 2574 | put_online_cpus(); |
b6136773 | 2575 | free_percpu(works); |
15316ba8 CL |
2576 | return 0; |
2577 | } | |
2578 | ||
eef6a7d5 AS |
2579 | /** |
2580 | * flush_scheduled_work - ensure that any scheduled work has run to completion. | |
2581 | * | |
2582 | * Forces execution of the kernel-global workqueue and blocks until its | |
2583 | * completion. | |
2584 | * | |
2585 | * Think twice before calling this function! It's very easy to get into | |
2586 | * trouble if you don't take great care. Either of the following situations | |
2587 | * will lead to deadlock: | |
2588 | * | |
2589 | * One of the work items currently on the workqueue needs to acquire | |
2590 | * a lock held by your code or its caller. | |
2591 | * | |
2592 | * Your code is running in the context of a work routine. | |
2593 | * | |
2594 | * They will be detected by lockdep when they occur, but the first might not | |
2595 | * occur very often. It depends on what work items are on the workqueue and | |
2596 | * what locks they need, which you have no control over. | |
2597 | * | |
2598 | * In most situations flushing the entire workqueue is overkill; you merely | |
2599 | * need to know that a particular work item isn't queued and isn't running. | |
2600 | * In such cases you should use cancel_delayed_work_sync() or | |
2601 | * cancel_work_sync() instead. | |
2602 | */ | |
1da177e4 LT |
2603 | void flush_scheduled_work(void) |
2604 | { | |
d320c038 | 2605 | flush_workqueue(system_wq); |
1da177e4 | 2606 | } |
ae90dd5d | 2607 | EXPORT_SYMBOL(flush_scheduled_work); |
1da177e4 | 2608 | |
1fa44eca JB |
2609 | /** |
2610 | * execute_in_process_context - reliably execute the routine with user context | |
2611 | * @fn: the function to execute | |
1fa44eca JB |
2612 | * @ew: guaranteed storage for the execute work structure (must |
2613 | * be available when the work executes) | |
2614 | * | |
2615 | * Executes the function immediately if process context is available, | |
2616 | * otherwise schedules the function for delayed execution. | |
2617 | * | |
2618 | * Returns: 0 - function was executed | |
2619 | * 1 - function was scheduled for execution | |
2620 | */ | |
65f27f38 | 2621 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) |
1fa44eca JB |
2622 | { |
2623 | if (!in_interrupt()) { | |
65f27f38 | 2624 | fn(&ew->work); |
1fa44eca JB |
2625 | return 0; |
2626 | } | |
2627 | ||
65f27f38 | 2628 | INIT_WORK(&ew->work, fn); |
1fa44eca JB |
2629 | schedule_work(&ew->work); |
2630 | ||
2631 | return 1; | |
2632 | } | |
2633 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
2634 | ||
1da177e4 LT |
2635 | int keventd_up(void) |
2636 | { | |
d320c038 | 2637 | return system_wq != NULL; |
1da177e4 LT |
2638 | } |
2639 | ||
bdbc5dd7 | 2640 | static int alloc_cwqs(struct workqueue_struct *wq) |
0f900049 TH |
2641 | { |
2642 | /* | |
2643 | * cwqs are forced aligned according to WORK_STRUCT_FLAG_BITS. | |
2644 | * Make sure that the alignment isn't lower than that of | |
2645 | * unsigned long long. | |
2646 | */ | |
2647 | const size_t size = sizeof(struct cpu_workqueue_struct); | |
2648 | const size_t align = max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS, | |
2649 | __alignof__(unsigned long long)); | |
931ac77e TH |
2650 | #ifdef CONFIG_SMP |
2651 | bool percpu = !(wq->flags & WQ_UNBOUND); | |
2652 | #else | |
2653 | bool percpu = false; | |
2654 | #endif | |
0f900049 | 2655 | |
931ac77e | 2656 | if (percpu) |
f3421797 | 2657 | wq->cpu_wq.pcpu = __alloc_percpu(size, align); |
931ac77e | 2658 | else { |
f3421797 TH |
2659 | void *ptr; |
2660 | ||
2661 | /* | |
2662 | * Allocate enough room to align cwq and put an extra | |
2663 | * pointer at the end pointing back to the originally | |
2664 | * allocated pointer which will be used for free. | |
2665 | */ | |
2666 | ptr = kzalloc(size + align + sizeof(void *), GFP_KERNEL); | |
2667 | if (ptr) { | |
2668 | wq->cpu_wq.single = PTR_ALIGN(ptr, align); | |
2669 | *(void **)(wq->cpu_wq.single + 1) = ptr; | |
2670 | } | |
bdbc5dd7 | 2671 | } |
f3421797 | 2672 | |
0f900049 | 2673 | /* just in case, make sure it's actually aligned */ |
bdbc5dd7 TH |
2674 | BUG_ON(!IS_ALIGNED(wq->cpu_wq.v, align)); |
2675 | return wq->cpu_wq.v ? 0 : -ENOMEM; | |
0f900049 TH |
2676 | } |
2677 | ||
bdbc5dd7 | 2678 | static void free_cwqs(struct workqueue_struct *wq) |
0f900049 | 2679 | { |
931ac77e TH |
2680 | #ifdef CONFIG_SMP |
2681 | bool percpu = !(wq->flags & WQ_UNBOUND); | |
2682 | #else | |
2683 | bool percpu = false; | |
2684 | #endif | |
2685 | ||
2686 | if (percpu) | |
f3421797 TH |
2687 | free_percpu(wq->cpu_wq.pcpu); |
2688 | else if (wq->cpu_wq.single) { | |
2689 | /* the pointer to free is stored right after the cwq */ | |
bdbc5dd7 | 2690 | kfree(*(void **)(wq->cpu_wq.single + 1)); |
f3421797 | 2691 | } |
0f900049 TH |
2692 | } |
2693 | ||
f3421797 TH |
2694 | static int wq_clamp_max_active(int max_active, unsigned int flags, |
2695 | const char *name) | |
b71ab8c2 | 2696 | { |
f3421797 TH |
2697 | int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; |
2698 | ||
2699 | if (max_active < 1 || max_active > lim) | |
b71ab8c2 TH |
2700 | printk(KERN_WARNING "workqueue: max_active %d requested for %s " |
2701 | "is out of range, clamping between %d and %d\n", | |
f3421797 | 2702 | max_active, name, 1, lim); |
b71ab8c2 | 2703 | |
f3421797 | 2704 | return clamp_val(max_active, 1, lim); |
b71ab8c2 TH |
2705 | } |
2706 | ||
d320c038 TH |
2707 | struct workqueue_struct *__alloc_workqueue_key(const char *name, |
2708 | unsigned int flags, | |
2709 | int max_active, | |
2710 | struct lock_class_key *key, | |
2711 | const char *lock_name) | |
1da177e4 | 2712 | { |
1da177e4 | 2713 | struct workqueue_struct *wq; |
c34056a3 | 2714 | unsigned int cpu; |
1da177e4 | 2715 | |
f3421797 TH |
2716 | /* |
2717 | * Unbound workqueues aren't concurrency managed and should be | |
2718 | * dispatched to workers immediately. | |
2719 | */ | |
2720 | if (flags & WQ_UNBOUND) | |
2721 | flags |= WQ_HIGHPRI; | |
2722 | ||
d320c038 | 2723 | max_active = max_active ?: WQ_DFL_ACTIVE; |
f3421797 | 2724 | max_active = wq_clamp_max_active(max_active, flags, name); |
1e19ffc6 | 2725 | |
3af24433 ON |
2726 | wq = kzalloc(sizeof(*wq), GFP_KERNEL); |
2727 | if (!wq) | |
4690c4ab | 2728 | goto err; |
3af24433 | 2729 | |
97e37d7b | 2730 | wq->flags = flags; |
a0a1a5fd | 2731 | wq->saved_max_active = max_active; |
73f53c4a TH |
2732 | mutex_init(&wq->flush_mutex); |
2733 | atomic_set(&wq->nr_cwqs_to_flush, 0); | |
2734 | INIT_LIST_HEAD(&wq->flusher_queue); | |
2735 | INIT_LIST_HEAD(&wq->flusher_overflow); | |
502ca9d8 | 2736 | |
3af24433 | 2737 | wq->name = name; |
eb13ba87 | 2738 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); |
cce1a165 | 2739 | INIT_LIST_HEAD(&wq->list); |
3af24433 | 2740 | |
bdbc5dd7 TH |
2741 | if (alloc_cwqs(wq) < 0) |
2742 | goto err; | |
2743 | ||
f3421797 | 2744 | for_each_cwq_cpu(cpu, wq) { |
1537663f | 2745 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
8b03ae3c | 2746 | struct global_cwq *gcwq = get_gcwq(cpu); |
1537663f | 2747 | |
0f900049 | 2748 | BUG_ON((unsigned long)cwq & WORK_STRUCT_FLAG_MASK); |
8b03ae3c | 2749 | cwq->gcwq = gcwq; |
c34056a3 | 2750 | cwq->wq = wq; |
73f53c4a | 2751 | cwq->flush_color = -1; |
1e19ffc6 | 2752 | cwq->max_active = max_active; |
1e19ffc6 | 2753 | INIT_LIST_HEAD(&cwq->delayed_works); |
e22bee78 | 2754 | } |
1537663f | 2755 | |
e22bee78 TH |
2756 | if (flags & WQ_RESCUER) { |
2757 | struct worker *rescuer; | |
2758 | ||
f2e005aa | 2759 | if (!alloc_mayday_mask(&wq->mayday_mask, GFP_KERNEL)) |
e22bee78 TH |
2760 | goto err; |
2761 | ||
2762 | wq->rescuer = rescuer = alloc_worker(); | |
2763 | if (!rescuer) | |
2764 | goto err; | |
2765 | ||
2766 | rescuer->task = kthread_create(rescuer_thread, wq, "%s", name); | |
2767 | if (IS_ERR(rescuer->task)) | |
2768 | goto err; | |
2769 | ||
2770 | wq->rescuer = rescuer; | |
2771 | rescuer->task->flags |= PF_THREAD_BOUND; | |
2772 | wake_up_process(rescuer->task); | |
3af24433 ON |
2773 | } |
2774 | ||
a0a1a5fd TH |
2775 | /* |
2776 | * workqueue_lock protects global freeze state and workqueues | |
2777 | * list. Grab it, set max_active accordingly and add the new | |
2778 | * workqueue to workqueues list. | |
2779 | */ | |
1537663f | 2780 | spin_lock(&workqueue_lock); |
a0a1a5fd TH |
2781 | |
2782 | if (workqueue_freezing && wq->flags & WQ_FREEZEABLE) | |
f3421797 | 2783 | for_each_cwq_cpu(cpu, wq) |
a0a1a5fd TH |
2784 | get_cwq(cpu, wq)->max_active = 0; |
2785 | ||
1537663f | 2786 | list_add(&wq->list, &workqueues); |
a0a1a5fd | 2787 | |
1537663f TH |
2788 | spin_unlock(&workqueue_lock); |
2789 | ||
3af24433 | 2790 | return wq; |
4690c4ab TH |
2791 | err: |
2792 | if (wq) { | |
bdbc5dd7 | 2793 | free_cwqs(wq); |
f2e005aa | 2794 | free_mayday_mask(wq->mayday_mask); |
e22bee78 | 2795 | kfree(wq->rescuer); |
4690c4ab TH |
2796 | kfree(wq); |
2797 | } | |
2798 | return NULL; | |
3af24433 | 2799 | } |
d320c038 | 2800 | EXPORT_SYMBOL_GPL(__alloc_workqueue_key); |
1da177e4 | 2801 | |
3af24433 ON |
2802 | /** |
2803 | * destroy_workqueue - safely terminate a workqueue | |
2804 | * @wq: target workqueue | |
2805 | * | |
2806 | * Safely destroy a workqueue. All work currently pending will be done first. | |
2807 | */ | |
2808 | void destroy_workqueue(struct workqueue_struct *wq) | |
2809 | { | |
c8e55f36 | 2810 | unsigned int cpu; |
3af24433 | 2811 | |
a0a1a5fd TH |
2812 | flush_workqueue(wq); |
2813 | ||
2814 | /* | |
2815 | * wq list is used to freeze wq, remove from list after | |
2816 | * flushing is complete in case freeze races us. | |
2817 | */ | |
95402b38 | 2818 | spin_lock(&workqueue_lock); |
b1f4ec17 | 2819 | list_del(&wq->list); |
95402b38 | 2820 | spin_unlock(&workqueue_lock); |
3af24433 | 2821 | |
e22bee78 | 2822 | /* sanity check */ |
f3421797 | 2823 | for_each_cwq_cpu(cpu, wq) { |
73f53c4a TH |
2824 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); |
2825 | int i; | |
2826 | ||
73f53c4a TH |
2827 | for (i = 0; i < WORK_NR_COLORS; i++) |
2828 | BUG_ON(cwq->nr_in_flight[i]); | |
1e19ffc6 TH |
2829 | BUG_ON(cwq->nr_active); |
2830 | BUG_ON(!list_empty(&cwq->delayed_works)); | |
73f53c4a | 2831 | } |
9b41ea72 | 2832 | |
e22bee78 TH |
2833 | if (wq->flags & WQ_RESCUER) { |
2834 | kthread_stop(wq->rescuer->task); | |
f2e005aa | 2835 | free_mayday_mask(wq->mayday_mask); |
e22bee78 TH |
2836 | } |
2837 | ||
bdbc5dd7 | 2838 | free_cwqs(wq); |
3af24433 ON |
2839 | kfree(wq); |
2840 | } | |
2841 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
2842 | ||
dcd989cb TH |
2843 | /** |
2844 | * workqueue_set_max_active - adjust max_active of a workqueue | |
2845 | * @wq: target workqueue | |
2846 | * @max_active: new max_active value. | |
2847 | * | |
2848 | * Set max_active of @wq to @max_active. | |
2849 | * | |
2850 | * CONTEXT: | |
2851 | * Don't call from IRQ context. | |
2852 | */ | |
2853 | void workqueue_set_max_active(struct workqueue_struct *wq, int max_active) | |
2854 | { | |
2855 | unsigned int cpu; | |
2856 | ||
f3421797 | 2857 | max_active = wq_clamp_max_active(max_active, wq->flags, wq->name); |
dcd989cb TH |
2858 | |
2859 | spin_lock(&workqueue_lock); | |
2860 | ||
2861 | wq->saved_max_active = max_active; | |
2862 | ||
f3421797 | 2863 | for_each_cwq_cpu(cpu, wq) { |
dcd989cb TH |
2864 | struct global_cwq *gcwq = get_gcwq(cpu); |
2865 | ||
2866 | spin_lock_irq(&gcwq->lock); | |
2867 | ||
2868 | if (!(wq->flags & WQ_FREEZEABLE) || | |
2869 | !(gcwq->flags & GCWQ_FREEZING)) | |
2870 | get_cwq(gcwq->cpu, wq)->max_active = max_active; | |
2871 | ||
2872 | spin_unlock_irq(&gcwq->lock); | |
2873 | } | |
2874 | ||
2875 | spin_unlock(&workqueue_lock); | |
2876 | } | |
2877 | EXPORT_SYMBOL_GPL(workqueue_set_max_active); | |
2878 | ||
2879 | /** | |
2880 | * workqueue_congested - test whether a workqueue is congested | |
2881 | * @cpu: CPU in question | |
2882 | * @wq: target workqueue | |
2883 | * | |
2884 | * Test whether @wq's cpu workqueue for @cpu is congested. There is | |
2885 | * no synchronization around this function and the test result is | |
2886 | * unreliable and only useful as advisory hints or for debugging. | |
2887 | * | |
2888 | * RETURNS: | |
2889 | * %true if congested, %false otherwise. | |
2890 | */ | |
2891 | bool workqueue_congested(unsigned int cpu, struct workqueue_struct *wq) | |
2892 | { | |
2893 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
2894 | ||
2895 | return !list_empty(&cwq->delayed_works); | |
2896 | } | |
2897 | EXPORT_SYMBOL_GPL(workqueue_congested); | |
2898 | ||
2899 | /** | |
2900 | * work_cpu - return the last known associated cpu for @work | |
2901 | * @work: the work of interest | |
2902 | * | |
2903 | * RETURNS: | |
bdbc5dd7 | 2904 | * CPU number if @work was ever queued. WORK_CPU_NONE otherwise. |
dcd989cb TH |
2905 | */ |
2906 | unsigned int work_cpu(struct work_struct *work) | |
2907 | { | |
2908 | struct global_cwq *gcwq = get_work_gcwq(work); | |
2909 | ||
bdbc5dd7 | 2910 | return gcwq ? gcwq->cpu : WORK_CPU_NONE; |
dcd989cb TH |
2911 | } |
2912 | EXPORT_SYMBOL_GPL(work_cpu); | |
2913 | ||
2914 | /** | |
2915 | * work_busy - test whether a work is currently pending or running | |
2916 | * @work: the work to be tested | |
2917 | * | |
2918 | * Test whether @work is currently pending or running. There is no | |
2919 | * synchronization around this function and the test result is | |
2920 | * unreliable and only useful as advisory hints or for debugging. | |
2921 | * Especially for reentrant wqs, the pending state might hide the | |
2922 | * running state. | |
2923 | * | |
2924 | * RETURNS: | |
2925 | * OR'd bitmask of WORK_BUSY_* bits. | |
2926 | */ | |
2927 | unsigned int work_busy(struct work_struct *work) | |
2928 | { | |
2929 | struct global_cwq *gcwq = get_work_gcwq(work); | |
2930 | unsigned long flags; | |
2931 | unsigned int ret = 0; | |
2932 | ||
2933 | if (!gcwq) | |
2934 | return false; | |
2935 | ||
2936 | spin_lock_irqsave(&gcwq->lock, flags); | |
2937 | ||
2938 | if (work_pending(work)) | |
2939 | ret |= WORK_BUSY_PENDING; | |
2940 | if (find_worker_executing_work(gcwq, work)) | |
2941 | ret |= WORK_BUSY_RUNNING; | |
2942 | ||
2943 | spin_unlock_irqrestore(&gcwq->lock, flags); | |
2944 | ||
2945 | return ret; | |
2946 | } | |
2947 | EXPORT_SYMBOL_GPL(work_busy); | |
2948 | ||
db7bccf4 TH |
2949 | /* |
2950 | * CPU hotplug. | |
2951 | * | |
e22bee78 TH |
2952 | * There are two challenges in supporting CPU hotplug. Firstly, there |
2953 | * are a lot of assumptions on strong associations among work, cwq and | |
2954 | * gcwq which make migrating pending and scheduled works very | |
2955 | * difficult to implement without impacting hot paths. Secondly, | |
2956 | * gcwqs serve mix of short, long and very long running works making | |
2957 | * blocked draining impractical. | |
2958 | * | |
2959 | * This is solved by allowing a gcwq to be detached from CPU, running | |
2960 | * it with unbound (rogue) workers and allowing it to be reattached | |
2961 | * later if the cpu comes back online. A separate thread is created | |
2962 | * to govern a gcwq in such state and is called the trustee of the | |
2963 | * gcwq. | |
db7bccf4 TH |
2964 | * |
2965 | * Trustee states and their descriptions. | |
2966 | * | |
2967 | * START Command state used on startup. On CPU_DOWN_PREPARE, a | |
2968 | * new trustee is started with this state. | |
2969 | * | |
2970 | * IN_CHARGE Once started, trustee will enter this state after | |
e22bee78 TH |
2971 | * assuming the manager role and making all existing |
2972 | * workers rogue. DOWN_PREPARE waits for trustee to | |
2973 | * enter this state. After reaching IN_CHARGE, trustee | |
2974 | * tries to execute the pending worklist until it's empty | |
2975 | * and the state is set to BUTCHER, or the state is set | |
2976 | * to RELEASE. | |
db7bccf4 TH |
2977 | * |
2978 | * BUTCHER Command state which is set by the cpu callback after | |
2979 | * the cpu has went down. Once this state is set trustee | |
2980 | * knows that there will be no new works on the worklist | |
2981 | * and once the worklist is empty it can proceed to | |
2982 | * killing idle workers. | |
2983 | * | |
2984 | * RELEASE Command state which is set by the cpu callback if the | |
2985 | * cpu down has been canceled or it has come online | |
2986 | * again. After recognizing this state, trustee stops | |
e22bee78 TH |
2987 | * trying to drain or butcher and clears ROGUE, rebinds |
2988 | * all remaining workers back to the cpu and releases | |
2989 | * manager role. | |
db7bccf4 TH |
2990 | * |
2991 | * DONE Trustee will enter this state after BUTCHER or RELEASE | |
2992 | * is complete. | |
2993 | * | |
2994 | * trustee CPU draining | |
2995 | * took over down complete | |
2996 | * START -----------> IN_CHARGE -----------> BUTCHER -----------> DONE | |
2997 | * | | ^ | |
2998 | * | CPU is back online v return workers | | |
2999 | * ----------------> RELEASE -------------- | |
3000 | */ | |
3001 | ||
3002 | /** | |
3003 | * trustee_wait_event_timeout - timed event wait for trustee | |
3004 | * @cond: condition to wait for | |
3005 | * @timeout: timeout in jiffies | |
3006 | * | |
3007 | * wait_event_timeout() for trustee to use. Handles locking and | |
3008 | * checks for RELEASE request. | |
3009 | * | |
3010 | * CONTEXT: | |
3011 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
3012 | * multiple times. To be used by trustee. | |
3013 | * | |
3014 | * RETURNS: | |
3015 | * Positive indicating left time if @cond is satisfied, 0 if timed | |
3016 | * out, -1 if canceled. | |
3017 | */ | |
3018 | #define trustee_wait_event_timeout(cond, timeout) ({ \ | |
3019 | long __ret = (timeout); \ | |
3020 | while (!((cond) || (gcwq->trustee_state == TRUSTEE_RELEASE)) && \ | |
3021 | __ret) { \ | |
3022 | spin_unlock_irq(&gcwq->lock); \ | |
3023 | __wait_event_timeout(gcwq->trustee_wait, (cond) || \ | |
3024 | (gcwq->trustee_state == TRUSTEE_RELEASE), \ | |
3025 | __ret); \ | |
3026 | spin_lock_irq(&gcwq->lock); \ | |
3027 | } \ | |
3028 | gcwq->trustee_state == TRUSTEE_RELEASE ? -1 : (__ret); \ | |
3029 | }) | |
3030 | ||
3031 | /** | |
3032 | * trustee_wait_event - event wait for trustee | |
3033 | * @cond: condition to wait for | |
3034 | * | |
3035 | * wait_event() for trustee to use. Automatically handles locking and | |
3036 | * checks for CANCEL request. | |
3037 | * | |
3038 | * CONTEXT: | |
3039 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
3040 | * multiple times. To be used by trustee. | |
3041 | * | |
3042 | * RETURNS: | |
3043 | * 0 if @cond is satisfied, -1 if canceled. | |
3044 | */ | |
3045 | #define trustee_wait_event(cond) ({ \ | |
3046 | long __ret1; \ | |
3047 | __ret1 = trustee_wait_event_timeout(cond, MAX_SCHEDULE_TIMEOUT);\ | |
3048 | __ret1 < 0 ? -1 : 0; \ | |
3049 | }) | |
3050 | ||
3051 | static int __cpuinit trustee_thread(void *__gcwq) | |
3052 | { | |
3053 | struct global_cwq *gcwq = __gcwq; | |
3054 | struct worker *worker; | |
e22bee78 | 3055 | struct work_struct *work; |
db7bccf4 | 3056 | struct hlist_node *pos; |
e22bee78 | 3057 | long rc; |
db7bccf4 TH |
3058 | int i; |
3059 | ||
3060 | BUG_ON(gcwq->cpu != smp_processor_id()); | |
3061 | ||
3062 | spin_lock_irq(&gcwq->lock); | |
3063 | /* | |
e22bee78 TH |
3064 | * Claim the manager position and make all workers rogue. |
3065 | * Trustee must be bound to the target cpu and can't be | |
3066 | * cancelled. | |
db7bccf4 TH |
3067 | */ |
3068 | BUG_ON(gcwq->cpu != smp_processor_id()); | |
e22bee78 TH |
3069 | rc = trustee_wait_event(!(gcwq->flags & GCWQ_MANAGING_WORKERS)); |
3070 | BUG_ON(rc < 0); | |
3071 | ||
3072 | gcwq->flags |= GCWQ_MANAGING_WORKERS; | |
db7bccf4 TH |
3073 | |
3074 | list_for_each_entry(worker, &gcwq->idle_list, entry) | |
cb444766 | 3075 | worker->flags |= WORKER_ROGUE; |
db7bccf4 TH |
3076 | |
3077 | for_each_busy_worker(worker, i, pos, gcwq) | |
cb444766 | 3078 | worker->flags |= WORKER_ROGUE; |
db7bccf4 | 3079 | |
e22bee78 TH |
3080 | /* |
3081 | * Call schedule() so that we cross rq->lock and thus can | |
3082 | * guarantee sched callbacks see the rogue flag. This is | |
3083 | * necessary as scheduler callbacks may be invoked from other | |
3084 | * cpus. | |
3085 | */ | |
3086 | spin_unlock_irq(&gcwq->lock); | |
3087 | schedule(); | |
3088 | spin_lock_irq(&gcwq->lock); | |
3089 | ||
3090 | /* | |
cb444766 TH |
3091 | * Sched callbacks are disabled now. Zap nr_running. After |
3092 | * this, nr_running stays zero and need_more_worker() and | |
3093 | * keep_working() are always true as long as the worklist is | |
3094 | * not empty. | |
e22bee78 | 3095 | */ |
cb444766 | 3096 | atomic_set(get_gcwq_nr_running(gcwq->cpu), 0); |
e22bee78 TH |
3097 | |
3098 | spin_unlock_irq(&gcwq->lock); | |
3099 | del_timer_sync(&gcwq->idle_timer); | |
3100 | spin_lock_irq(&gcwq->lock); | |
3101 | ||
db7bccf4 TH |
3102 | /* |
3103 | * We're now in charge. Notify and proceed to drain. We need | |
3104 | * to keep the gcwq running during the whole CPU down | |
3105 | * procedure as other cpu hotunplug callbacks may need to | |
3106 | * flush currently running tasks. | |
3107 | */ | |
3108 | gcwq->trustee_state = TRUSTEE_IN_CHARGE; | |
3109 | wake_up_all(&gcwq->trustee_wait); | |
3110 | ||
3111 | /* | |
3112 | * The original cpu is in the process of dying and may go away | |
3113 | * anytime now. When that happens, we and all workers would | |
e22bee78 TH |
3114 | * be migrated to other cpus. Try draining any left work. We |
3115 | * want to get it over with ASAP - spam rescuers, wake up as | |
3116 | * many idlers as necessary and create new ones till the | |
3117 | * worklist is empty. Note that if the gcwq is frozen, there | |
3118 | * may be frozen works in freezeable cwqs. Don't declare | |
3119 | * completion while frozen. | |
db7bccf4 TH |
3120 | */ |
3121 | while (gcwq->nr_workers != gcwq->nr_idle || | |
3122 | gcwq->flags & GCWQ_FREEZING || | |
3123 | gcwq->trustee_state == TRUSTEE_IN_CHARGE) { | |
e22bee78 TH |
3124 | int nr_works = 0; |
3125 | ||
3126 | list_for_each_entry(work, &gcwq->worklist, entry) { | |
3127 | send_mayday(work); | |
3128 | nr_works++; | |
3129 | } | |
3130 | ||
3131 | list_for_each_entry(worker, &gcwq->idle_list, entry) { | |
3132 | if (!nr_works--) | |
3133 | break; | |
3134 | wake_up_process(worker->task); | |
3135 | } | |
3136 | ||
3137 | if (need_to_create_worker(gcwq)) { | |
3138 | spin_unlock_irq(&gcwq->lock); | |
3139 | worker = create_worker(gcwq, false); | |
3140 | spin_lock_irq(&gcwq->lock); | |
3141 | if (worker) { | |
cb444766 | 3142 | worker->flags |= WORKER_ROGUE; |
e22bee78 TH |
3143 | start_worker(worker); |
3144 | } | |
3145 | } | |
3146 | ||
db7bccf4 TH |
3147 | /* give a breather */ |
3148 | if (trustee_wait_event_timeout(false, TRUSTEE_COOLDOWN) < 0) | |
3149 | break; | |
3150 | } | |
3151 | ||
e22bee78 TH |
3152 | /* |
3153 | * Either all works have been scheduled and cpu is down, or | |
3154 | * cpu down has already been canceled. Wait for and butcher | |
3155 | * all workers till we're canceled. | |
3156 | */ | |
3157 | do { | |
3158 | rc = trustee_wait_event(!list_empty(&gcwq->idle_list)); | |
3159 | while (!list_empty(&gcwq->idle_list)) | |
3160 | destroy_worker(list_first_entry(&gcwq->idle_list, | |
3161 | struct worker, entry)); | |
3162 | } while (gcwq->nr_workers && rc >= 0); | |
3163 | ||
3164 | /* | |
3165 | * At this point, either draining has completed and no worker | |
3166 | * is left, or cpu down has been canceled or the cpu is being | |
3167 | * brought back up. There shouldn't be any idle one left. | |
3168 | * Tell the remaining busy ones to rebind once it finishes the | |
3169 | * currently scheduled works by scheduling the rebind_work. | |
3170 | */ | |
3171 | WARN_ON(!list_empty(&gcwq->idle_list)); | |
3172 | ||
3173 | for_each_busy_worker(worker, i, pos, gcwq) { | |
3174 | struct work_struct *rebind_work = &worker->rebind_work; | |
3175 | ||
3176 | /* | |
3177 | * Rebind_work may race with future cpu hotplug | |
3178 | * operations. Use a separate flag to mark that | |
3179 | * rebinding is scheduled. | |
3180 | */ | |
cb444766 TH |
3181 | worker->flags |= WORKER_REBIND; |
3182 | worker->flags &= ~WORKER_ROGUE; | |
e22bee78 TH |
3183 | |
3184 | /* queue rebind_work, wq doesn't matter, use the default one */ | |
3185 | if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, | |
3186 | work_data_bits(rebind_work))) | |
3187 | continue; | |
3188 | ||
3189 | debug_work_activate(rebind_work); | |
d320c038 | 3190 | insert_work(get_cwq(gcwq->cpu, system_wq), rebind_work, |
e22bee78 TH |
3191 | worker->scheduled.next, |
3192 | work_color_to_flags(WORK_NO_COLOR)); | |
3193 | } | |
3194 | ||
3195 | /* relinquish manager role */ | |
3196 | gcwq->flags &= ~GCWQ_MANAGING_WORKERS; | |
3197 | ||
db7bccf4 TH |
3198 | /* notify completion */ |
3199 | gcwq->trustee = NULL; | |
3200 | gcwq->trustee_state = TRUSTEE_DONE; | |
3201 | wake_up_all(&gcwq->trustee_wait); | |
3202 | spin_unlock_irq(&gcwq->lock); | |
3203 | return 0; | |
3204 | } | |
3205 | ||
3206 | /** | |
3207 | * wait_trustee_state - wait for trustee to enter the specified state | |
3208 | * @gcwq: gcwq the trustee of interest belongs to | |
3209 | * @state: target state to wait for | |
3210 | * | |
3211 | * Wait for the trustee to reach @state. DONE is already matched. | |
3212 | * | |
3213 | * CONTEXT: | |
3214 | * spin_lock_irq(gcwq->lock) which may be released and regrabbed | |
3215 | * multiple times. To be used by cpu_callback. | |
3216 | */ | |
3217 | static void __cpuinit wait_trustee_state(struct global_cwq *gcwq, int state) | |
3218 | { | |
3219 | if (!(gcwq->trustee_state == state || | |
3220 | gcwq->trustee_state == TRUSTEE_DONE)) { | |
3221 | spin_unlock_irq(&gcwq->lock); | |
3222 | __wait_event(gcwq->trustee_wait, | |
3223 | gcwq->trustee_state == state || | |
3224 | gcwq->trustee_state == TRUSTEE_DONE); | |
3225 | spin_lock_irq(&gcwq->lock); | |
3226 | } | |
3227 | } | |
3228 | ||
3af24433 ON |
3229 | static int __devinit workqueue_cpu_callback(struct notifier_block *nfb, |
3230 | unsigned long action, | |
3231 | void *hcpu) | |
3232 | { | |
3233 | unsigned int cpu = (unsigned long)hcpu; | |
db7bccf4 TH |
3234 | struct global_cwq *gcwq = get_gcwq(cpu); |
3235 | struct task_struct *new_trustee = NULL; | |
e22bee78 | 3236 | struct worker *uninitialized_var(new_worker); |
db7bccf4 | 3237 | unsigned long flags; |
3af24433 | 3238 | |
8bb78442 RW |
3239 | action &= ~CPU_TASKS_FROZEN; |
3240 | ||
db7bccf4 TH |
3241 | switch (action) { |
3242 | case CPU_DOWN_PREPARE: | |
3243 | new_trustee = kthread_create(trustee_thread, gcwq, | |
3244 | "workqueue_trustee/%d\n", cpu); | |
3245 | if (IS_ERR(new_trustee)) | |
3246 | return notifier_from_errno(PTR_ERR(new_trustee)); | |
3247 | kthread_bind(new_trustee, cpu); | |
e22bee78 TH |
3248 | /* fall through */ |
3249 | case CPU_UP_PREPARE: | |
3250 | BUG_ON(gcwq->first_idle); | |
3251 | new_worker = create_worker(gcwq, false); | |
3252 | if (!new_worker) { | |
3253 | if (new_trustee) | |
3254 | kthread_stop(new_trustee); | |
3255 | return NOTIFY_BAD; | |
3256 | } | |
db7bccf4 | 3257 | } |
3af24433 | 3258 | |
db7bccf4 TH |
3259 | /* some are called w/ irq disabled, don't disturb irq status */ |
3260 | spin_lock_irqsave(&gcwq->lock, flags); | |
3af24433 | 3261 | |
db7bccf4 TH |
3262 | switch (action) { |
3263 | case CPU_DOWN_PREPARE: | |
3264 | /* initialize trustee and tell it to acquire the gcwq */ | |
3265 | BUG_ON(gcwq->trustee || gcwq->trustee_state != TRUSTEE_DONE); | |
3266 | gcwq->trustee = new_trustee; | |
3267 | gcwq->trustee_state = TRUSTEE_START; | |
3268 | wake_up_process(gcwq->trustee); | |
3269 | wait_trustee_state(gcwq, TRUSTEE_IN_CHARGE); | |
e22bee78 TH |
3270 | /* fall through */ |
3271 | case CPU_UP_PREPARE: | |
3272 | BUG_ON(gcwq->first_idle); | |
3273 | gcwq->first_idle = new_worker; | |
3274 | break; | |
3275 | ||
3276 | case CPU_DYING: | |
3277 | /* | |
3278 | * Before this, the trustee and all workers except for | |
3279 | * the ones which are still executing works from | |
3280 | * before the last CPU down must be on the cpu. After | |
3281 | * this, they'll all be diasporas. | |
3282 | */ | |
3283 | gcwq->flags |= GCWQ_DISASSOCIATED; | |
db7bccf4 TH |
3284 | break; |
3285 | ||
3286 | case CPU_POST_DEAD: | |
3287 | gcwq->trustee_state = TRUSTEE_BUTCHER; | |
e22bee78 TH |
3288 | /* fall through */ |
3289 | case CPU_UP_CANCELED: | |
3290 | destroy_worker(gcwq->first_idle); | |
3291 | gcwq->first_idle = NULL; | |
db7bccf4 TH |
3292 | break; |
3293 | ||
3294 | case CPU_DOWN_FAILED: | |
3295 | case CPU_ONLINE: | |
e22bee78 | 3296 | gcwq->flags &= ~GCWQ_DISASSOCIATED; |
db7bccf4 TH |
3297 | if (gcwq->trustee_state != TRUSTEE_DONE) { |
3298 | gcwq->trustee_state = TRUSTEE_RELEASE; | |
3299 | wake_up_process(gcwq->trustee); | |
3300 | wait_trustee_state(gcwq, TRUSTEE_DONE); | |
3af24433 | 3301 | } |
db7bccf4 | 3302 | |
e22bee78 TH |
3303 | /* |
3304 | * Trustee is done and there might be no worker left. | |
3305 | * Put the first_idle in and request a real manager to | |
3306 | * take a look. | |
3307 | */ | |
3308 | spin_unlock_irq(&gcwq->lock); | |
3309 | kthread_bind(gcwq->first_idle->task, cpu); | |
3310 | spin_lock_irq(&gcwq->lock); | |
3311 | gcwq->flags |= GCWQ_MANAGE_WORKERS; | |
3312 | start_worker(gcwq->first_idle); | |
3313 | gcwq->first_idle = NULL; | |
db7bccf4 | 3314 | break; |
1da177e4 LT |
3315 | } |
3316 | ||
db7bccf4 TH |
3317 | spin_unlock_irqrestore(&gcwq->lock, flags); |
3318 | ||
1537663f | 3319 | return notifier_from_errno(0); |
1da177e4 | 3320 | } |
1da177e4 | 3321 | |
2d3854a3 | 3322 | #ifdef CONFIG_SMP |
8ccad40d | 3323 | |
2d3854a3 | 3324 | struct work_for_cpu { |
6b44003e | 3325 | struct completion completion; |
2d3854a3 RR |
3326 | long (*fn)(void *); |
3327 | void *arg; | |
3328 | long ret; | |
3329 | }; | |
3330 | ||
6b44003e | 3331 | static int do_work_for_cpu(void *_wfc) |
2d3854a3 | 3332 | { |
6b44003e | 3333 | struct work_for_cpu *wfc = _wfc; |
2d3854a3 | 3334 | wfc->ret = wfc->fn(wfc->arg); |
6b44003e AM |
3335 | complete(&wfc->completion); |
3336 | return 0; | |
2d3854a3 RR |
3337 | } |
3338 | ||
3339 | /** | |
3340 | * work_on_cpu - run a function in user context on a particular cpu | |
3341 | * @cpu: the cpu to run on | |
3342 | * @fn: the function to run | |
3343 | * @arg: the function arg | |
3344 | * | |
31ad9081 RR |
3345 | * This will return the value @fn returns. |
3346 | * It is up to the caller to ensure that the cpu doesn't go offline. | |
6b44003e | 3347 | * The caller must not hold any locks which would prevent @fn from completing. |
2d3854a3 RR |
3348 | */ |
3349 | long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) | |
3350 | { | |
6b44003e AM |
3351 | struct task_struct *sub_thread; |
3352 | struct work_for_cpu wfc = { | |
3353 | .completion = COMPLETION_INITIALIZER_ONSTACK(wfc.completion), | |
3354 | .fn = fn, | |
3355 | .arg = arg, | |
3356 | }; | |
3357 | ||
3358 | sub_thread = kthread_create(do_work_for_cpu, &wfc, "work_for_cpu"); | |
3359 | if (IS_ERR(sub_thread)) | |
3360 | return PTR_ERR(sub_thread); | |
3361 | kthread_bind(sub_thread, cpu); | |
3362 | wake_up_process(sub_thread); | |
3363 | wait_for_completion(&wfc.completion); | |
2d3854a3 RR |
3364 | return wfc.ret; |
3365 | } | |
3366 | EXPORT_SYMBOL_GPL(work_on_cpu); | |
3367 | #endif /* CONFIG_SMP */ | |
3368 | ||
a0a1a5fd TH |
3369 | #ifdef CONFIG_FREEZER |
3370 | ||
3371 | /** | |
3372 | * freeze_workqueues_begin - begin freezing workqueues | |
3373 | * | |
3374 | * Start freezing workqueues. After this function returns, all | |
3375 | * freezeable workqueues will queue new works to their frozen_works | |
7e11629d | 3376 | * list instead of gcwq->worklist. |
a0a1a5fd TH |
3377 | * |
3378 | * CONTEXT: | |
8b03ae3c | 3379 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3380 | */ |
3381 | void freeze_workqueues_begin(void) | |
3382 | { | |
a0a1a5fd TH |
3383 | unsigned int cpu; |
3384 | ||
3385 | spin_lock(&workqueue_lock); | |
3386 | ||
3387 | BUG_ON(workqueue_freezing); | |
3388 | workqueue_freezing = true; | |
3389 | ||
f3421797 | 3390 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3391 | struct global_cwq *gcwq = get_gcwq(cpu); |
bdbc5dd7 | 3392 | struct workqueue_struct *wq; |
8b03ae3c TH |
3393 | |
3394 | spin_lock_irq(&gcwq->lock); | |
3395 | ||
db7bccf4 TH |
3396 | BUG_ON(gcwq->flags & GCWQ_FREEZING); |
3397 | gcwq->flags |= GCWQ_FREEZING; | |
3398 | ||
a0a1a5fd TH |
3399 | list_for_each_entry(wq, &workqueues, list) { |
3400 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3401 | ||
f3421797 | 3402 | if (cwq && wq->flags & WQ_FREEZEABLE) |
a0a1a5fd | 3403 | cwq->max_active = 0; |
a0a1a5fd | 3404 | } |
8b03ae3c TH |
3405 | |
3406 | spin_unlock_irq(&gcwq->lock); | |
a0a1a5fd TH |
3407 | } |
3408 | ||
3409 | spin_unlock(&workqueue_lock); | |
3410 | } | |
3411 | ||
3412 | /** | |
3413 | * freeze_workqueues_busy - are freezeable workqueues still busy? | |
3414 | * | |
3415 | * Check whether freezing is complete. This function must be called | |
3416 | * between freeze_workqueues_begin() and thaw_workqueues(). | |
3417 | * | |
3418 | * CONTEXT: | |
3419 | * Grabs and releases workqueue_lock. | |
3420 | * | |
3421 | * RETURNS: | |
3422 | * %true if some freezeable workqueues are still busy. %false if | |
3423 | * freezing is complete. | |
3424 | */ | |
3425 | bool freeze_workqueues_busy(void) | |
3426 | { | |
a0a1a5fd TH |
3427 | unsigned int cpu; |
3428 | bool busy = false; | |
3429 | ||
3430 | spin_lock(&workqueue_lock); | |
3431 | ||
3432 | BUG_ON(!workqueue_freezing); | |
3433 | ||
f3421797 | 3434 | for_each_gcwq_cpu(cpu) { |
bdbc5dd7 | 3435 | struct workqueue_struct *wq; |
a0a1a5fd TH |
3436 | /* |
3437 | * nr_active is monotonically decreasing. It's safe | |
3438 | * to peek without lock. | |
3439 | */ | |
3440 | list_for_each_entry(wq, &workqueues, list) { | |
3441 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3442 | ||
f3421797 | 3443 | if (!cwq || !(wq->flags & WQ_FREEZEABLE)) |
a0a1a5fd TH |
3444 | continue; |
3445 | ||
3446 | BUG_ON(cwq->nr_active < 0); | |
3447 | if (cwq->nr_active) { | |
3448 | busy = true; | |
3449 | goto out_unlock; | |
3450 | } | |
3451 | } | |
3452 | } | |
3453 | out_unlock: | |
3454 | spin_unlock(&workqueue_lock); | |
3455 | return busy; | |
3456 | } | |
3457 | ||
3458 | /** | |
3459 | * thaw_workqueues - thaw workqueues | |
3460 | * | |
3461 | * Thaw workqueues. Normal queueing is restored and all collected | |
7e11629d | 3462 | * frozen works are transferred to their respective gcwq worklists. |
a0a1a5fd TH |
3463 | * |
3464 | * CONTEXT: | |
8b03ae3c | 3465 | * Grabs and releases workqueue_lock and gcwq->lock's. |
a0a1a5fd TH |
3466 | */ |
3467 | void thaw_workqueues(void) | |
3468 | { | |
a0a1a5fd TH |
3469 | unsigned int cpu; |
3470 | ||
3471 | spin_lock(&workqueue_lock); | |
3472 | ||
3473 | if (!workqueue_freezing) | |
3474 | goto out_unlock; | |
3475 | ||
f3421797 | 3476 | for_each_gcwq_cpu(cpu) { |
8b03ae3c | 3477 | struct global_cwq *gcwq = get_gcwq(cpu); |
bdbc5dd7 | 3478 | struct workqueue_struct *wq; |
8b03ae3c TH |
3479 | |
3480 | spin_lock_irq(&gcwq->lock); | |
3481 | ||
db7bccf4 TH |
3482 | BUG_ON(!(gcwq->flags & GCWQ_FREEZING)); |
3483 | gcwq->flags &= ~GCWQ_FREEZING; | |
3484 | ||
a0a1a5fd TH |
3485 | list_for_each_entry(wq, &workqueues, list) { |
3486 | struct cpu_workqueue_struct *cwq = get_cwq(cpu, wq); | |
3487 | ||
f3421797 | 3488 | if (!cwq || !(wq->flags & WQ_FREEZEABLE)) |
a0a1a5fd TH |
3489 | continue; |
3490 | ||
a0a1a5fd TH |
3491 | /* restore max_active and repopulate worklist */ |
3492 | cwq->max_active = wq->saved_max_active; | |
3493 | ||
3494 | while (!list_empty(&cwq->delayed_works) && | |
3495 | cwq->nr_active < cwq->max_active) | |
3496 | cwq_activate_first_delayed(cwq); | |
a0a1a5fd | 3497 | } |
8b03ae3c | 3498 | |
e22bee78 TH |
3499 | wake_up_worker(gcwq); |
3500 | ||
8b03ae3c | 3501 | spin_unlock_irq(&gcwq->lock); |
a0a1a5fd TH |
3502 | } |
3503 | ||
3504 | workqueue_freezing = false; | |
3505 | out_unlock: | |
3506 | spin_unlock(&workqueue_lock); | |
3507 | } | |
3508 | #endif /* CONFIG_FREEZER */ | |
3509 | ||
c12920d1 | 3510 | void __init init_workqueues(void) |
1da177e4 | 3511 | { |
c34056a3 | 3512 | unsigned int cpu; |
c8e55f36 | 3513 | int i; |
c34056a3 | 3514 | |
db7bccf4 | 3515 | hotcpu_notifier(workqueue_cpu_callback, CPU_PRI_WORKQUEUE); |
8b03ae3c TH |
3516 | |
3517 | /* initialize gcwqs */ | |
f3421797 | 3518 | for_each_gcwq_cpu(cpu) { |
8b03ae3c TH |
3519 | struct global_cwq *gcwq = get_gcwq(cpu); |
3520 | ||
3521 | spin_lock_init(&gcwq->lock); | |
7e11629d | 3522 | INIT_LIST_HEAD(&gcwq->worklist); |
8b03ae3c | 3523 | gcwq->cpu = cpu; |
f3421797 TH |
3524 | if (cpu == WORK_CPU_UNBOUND) |
3525 | gcwq->flags |= GCWQ_DISASSOCIATED; | |
8b03ae3c | 3526 | |
c8e55f36 TH |
3527 | INIT_LIST_HEAD(&gcwq->idle_list); |
3528 | for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++) | |
3529 | INIT_HLIST_HEAD(&gcwq->busy_hash[i]); | |
3530 | ||
e22bee78 TH |
3531 | init_timer_deferrable(&gcwq->idle_timer); |
3532 | gcwq->idle_timer.function = idle_worker_timeout; | |
3533 | gcwq->idle_timer.data = (unsigned long)gcwq; | |
3534 | ||
3535 | setup_timer(&gcwq->mayday_timer, gcwq_mayday_timeout, | |
3536 | (unsigned long)gcwq); | |
3537 | ||
8b03ae3c | 3538 | ida_init(&gcwq->worker_ida); |
db7bccf4 TH |
3539 | |
3540 | gcwq->trustee_state = TRUSTEE_DONE; | |
3541 | init_waitqueue_head(&gcwq->trustee_wait); | |
8b03ae3c TH |
3542 | } |
3543 | ||
e22bee78 | 3544 | /* create the initial worker */ |
f3421797 | 3545 | for_each_online_gcwq_cpu(cpu) { |
e22bee78 TH |
3546 | struct global_cwq *gcwq = get_gcwq(cpu); |
3547 | struct worker *worker; | |
3548 | ||
3549 | worker = create_worker(gcwq, true); | |
3550 | BUG_ON(!worker); | |
3551 | spin_lock_irq(&gcwq->lock); | |
3552 | start_worker(worker); | |
3553 | spin_unlock_irq(&gcwq->lock); | |
3554 | } | |
3555 | ||
d320c038 TH |
3556 | system_wq = alloc_workqueue("events", 0, 0); |
3557 | system_long_wq = alloc_workqueue("events_long", 0, 0); | |
3558 | system_nrt_wq = alloc_workqueue("events_nrt", WQ_NON_REENTRANT, 0); | |
f3421797 TH |
3559 | system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, |
3560 | WQ_UNBOUND_MAX_ACTIVE); | |
d320c038 | 3561 | BUG_ON(!system_wq || !system_long_wq || !system_nrt_wq); |
1da177e4 | 3562 | } |