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