2 * kernel/workqueue.c - generic async execution with shared worker pool
4 * Copyright (C) 2002 Ingo Molnar
6 * Derived from the taskqueue/keventd code by:
7 * David Woodhouse <dwmw2@infradead.org>
9 * Kai Petzke <wpp@marie.physik.tu-berlin.de>
10 * Theodore Ts'o <tytso@mit.edu>
12 * Made to use alloc_percpu by Christoph Lameter.
14 * Copyright (C) 2010 SUSE Linux Products GmbH
15 * Copyright (C) 2010 Tejun Heo <tj@kernel.org>
17 * This is the generic async execution mechanism. Work items as are
18 * executed in process context. The worker pool is shared and
19 * automatically managed. There is one worker pool for each CPU and
20 * one extra for works which are better served by workers which are
21 * not bound to any specific CPU.
23 * Please read Documentation/workqueue.txt for details.
26 #include <linux/export.h>
27 #include <linux/kernel.h>
28 #include <linux/sched.h>
29 #include <linux/init.h>
30 #include <linux/signal.h>
31 #include <linux/completion.h>
32 #include <linux/workqueue.h>
33 #include <linux/slab.h>
34 #include <linux/cpu.h>
35 #include <linux/notifier.h>
36 #include <linux/kthread.h>
37 #include <linux/hardirq.h>
38 #include <linux/mempolicy.h>
39 #include <linux/freezer.h>
40 #include <linux/kallsyms.h>
41 #include <linux/debug_locks.h>
42 #include <linux/lockdep.h>
43 #include <linux/idr.h>
45 #include "workqueue_sched.h"
51 * A bound gcwq is either associated or disassociated with its CPU.
52 * While associated (!DISASSOCIATED), all workers are bound to the
53 * CPU and none has %WORKER_UNBOUND set and concurrency management
56 * While DISASSOCIATED, the cpu may be offline and all workers have
57 * %WORKER_UNBOUND set and concurrency management disabled, and may
58 * be executing on any CPU. The gcwq behaves as an unbound one.
60 * Note that DISASSOCIATED can be flipped only while holding
61 * managership of all pools on the gcwq to avoid changing binding
62 * state while create_worker() is in progress.
64 GCWQ_DISASSOCIATED
= 1 << 0, /* cpu can't serve workers */
65 GCWQ_FREEZING
= 1 << 1, /* freeze in progress */
68 POOL_MANAGE_WORKERS
= 1 << 0, /* need to manage workers */
71 WORKER_STARTED
= 1 << 0, /* started */
72 WORKER_DIE
= 1 << 1, /* die die die */
73 WORKER_IDLE
= 1 << 2, /* is idle */
74 WORKER_PREP
= 1 << 3, /* preparing to run works */
75 WORKER_REBIND
= 1 << 5, /* mom is home, come back */
76 WORKER_CPU_INTENSIVE
= 1 << 6, /* cpu intensive */
77 WORKER_UNBOUND
= 1 << 7, /* worker is unbound */
79 WORKER_NOT_RUNNING
= WORKER_PREP
| WORKER_REBIND
| WORKER_UNBOUND
|
82 NR_WORKER_POOLS
= 2, /* # worker pools per gcwq */
84 BUSY_WORKER_HASH_ORDER
= 6, /* 64 pointers */
85 BUSY_WORKER_HASH_SIZE
= 1 << BUSY_WORKER_HASH_ORDER
,
86 BUSY_WORKER_HASH_MASK
= BUSY_WORKER_HASH_SIZE
- 1,
88 MAX_IDLE_WORKERS_RATIO
= 4, /* 1/4 of busy can be idle */
89 IDLE_WORKER_TIMEOUT
= 300 * HZ
, /* keep idle ones for 5 mins */
91 MAYDAY_INITIAL_TIMEOUT
= HZ
/ 100 >= 2 ? HZ
/ 100 : 2,
92 /* call for help after 10ms
94 MAYDAY_INTERVAL
= HZ
/ 10, /* and then every 100ms */
95 CREATE_COOLDOWN
= HZ
, /* time to breath after fail */
98 * Rescue workers are used only on emergencies and shared by
101 RESCUER_NICE_LEVEL
= -20,
102 HIGHPRI_NICE_LEVEL
= -20,
106 * Structure fields follow one of the following exclusion rules.
108 * I: Modifiable by initialization/destruction paths and read-only for
111 * P: Preemption protected. Disabling preemption is enough and should
112 * only be modified and accessed from the local cpu.
114 * L: gcwq->lock protected. Access with gcwq->lock held.
116 * X: During normal operation, modification requires gcwq->lock and
117 * should be done only from local cpu. Either disabling preemption
118 * on local cpu or grabbing gcwq->lock is enough for read access.
119 * If GCWQ_DISASSOCIATED is set, it's identical to L.
121 * F: wq->flush_mutex protected.
123 * W: workqueue_lock protected.
131 * The poor guys doing the actual heavy lifting. All on-duty workers
132 * are either serving the manager role, on idle list or on busy hash.
135 /* on idle list while idle, on busy hash table while busy */
137 struct list_head entry
; /* L: while idle */
138 struct hlist_node hentry
; /* L: while busy */
141 struct work_struct
*current_work
; /* L: work being processed */
142 struct cpu_workqueue_struct
*current_cwq
; /* L: current_work's cwq */
143 struct list_head scheduled
; /* L: scheduled works */
144 struct task_struct
*task
; /* I: worker task */
145 struct worker_pool
*pool
; /* I: the associated pool */
146 /* 64 bytes boundary on 64bit, 32 on 32bit */
147 unsigned long last_active
; /* L: last active timestamp */
148 unsigned int flags
; /* X: flags */
149 int id
; /* I: worker id */
151 /* for rebinding worker to CPU */
152 struct idle_rebind
*idle_rebind
; /* L: for idle worker */
153 struct work_struct rebind_work
; /* L: for busy worker */
157 struct global_cwq
*gcwq
; /* I: the owning gcwq */
158 unsigned int flags
; /* X: flags */
160 struct list_head worklist
; /* L: list of pending works */
161 int nr_workers
; /* L: total number of workers */
162 int nr_idle
; /* L: currently idle ones */
164 struct list_head idle_list
; /* X: list of idle workers */
165 struct timer_list idle_timer
; /* L: worker idle timeout */
166 struct timer_list mayday_timer
; /* L: SOS timer for workers */
168 struct mutex manager_mutex
; /* mutex manager should hold */
169 struct ida worker_ida
; /* L: for worker IDs */
173 * Global per-cpu workqueue. There's one and only one for each cpu
174 * and all works are queued and processed here regardless of their
178 spinlock_t lock
; /* the gcwq lock */
179 unsigned int cpu
; /* I: the associated cpu */
180 unsigned int flags
; /* L: GCWQ_* flags */
182 /* workers are chained either in busy_hash or pool idle_list */
183 struct hlist_head busy_hash
[BUSY_WORKER_HASH_SIZE
];
184 /* L: hash of busy workers */
186 struct worker_pool pools
[NR_WORKER_POOLS
];
187 /* normal and highpri pools */
189 wait_queue_head_t rebind_hold
; /* rebind hold wait */
190 } ____cacheline_aligned_in_smp
;
193 * The per-CPU workqueue. The lower WORK_STRUCT_FLAG_BITS of
194 * work_struct->data are used for flags and thus cwqs need to be
195 * aligned at two's power of the number of flag bits.
197 struct cpu_workqueue_struct
{
198 struct worker_pool
*pool
; /* I: the associated pool */
199 struct workqueue_struct
*wq
; /* I: the owning workqueue */
200 int work_color
; /* L: current color */
201 int flush_color
; /* L: flushing color */
202 int nr_in_flight
[WORK_NR_COLORS
];
203 /* L: nr of in_flight works */
204 int nr_active
; /* L: nr of active works */
205 int max_active
; /* L: max active works */
206 struct list_head delayed_works
; /* L: delayed works */
210 * Structure used to wait for workqueue flush.
213 struct list_head list
; /* F: list of flushers */
214 int flush_color
; /* F: flush color waiting for */
215 struct completion done
; /* flush completion */
219 * All cpumasks are assumed to be always set on UP and thus can't be
220 * used to determine whether there's something to be done.
223 typedef cpumask_var_t mayday_mask_t
;
224 #define mayday_test_and_set_cpu(cpu, mask) \
225 cpumask_test_and_set_cpu((cpu), (mask))
226 #define mayday_clear_cpu(cpu, mask) cpumask_clear_cpu((cpu), (mask))
227 #define for_each_mayday_cpu(cpu, mask) for_each_cpu((cpu), (mask))
228 #define alloc_mayday_mask(maskp, gfp) zalloc_cpumask_var((maskp), (gfp))
229 #define free_mayday_mask(mask) free_cpumask_var((mask))
231 typedef unsigned long mayday_mask_t
;
232 #define mayday_test_and_set_cpu(cpu, mask) test_and_set_bit(0, &(mask))
233 #define mayday_clear_cpu(cpu, mask) clear_bit(0, &(mask))
234 #define for_each_mayday_cpu(cpu, mask) if ((cpu) = 0, (mask))
235 #define alloc_mayday_mask(maskp, gfp) true
236 #define free_mayday_mask(mask) do { } while (0)
240 * The externally visible workqueue abstraction is an array of
241 * per-CPU workqueues:
243 struct workqueue_struct
{
244 unsigned int flags
; /* W: WQ_* flags */
246 struct cpu_workqueue_struct __percpu
*pcpu
;
247 struct cpu_workqueue_struct
*single
;
249 } cpu_wq
; /* I: cwq's */
250 struct list_head list
; /* W: list of all workqueues */
252 struct mutex flush_mutex
; /* protects wq flushing */
253 int work_color
; /* F: current work color */
254 int flush_color
; /* F: current flush color */
255 atomic_t nr_cwqs_to_flush
; /* flush in progress */
256 struct wq_flusher
*first_flusher
; /* F: first flusher */
257 struct list_head flusher_queue
; /* F: flush waiters */
258 struct list_head flusher_overflow
; /* F: flush overflow list */
260 mayday_mask_t mayday_mask
; /* cpus requesting rescue */
261 struct worker
*rescuer
; /* I: rescue worker */
263 int nr_drainers
; /* W: drain in progress */
264 int saved_max_active
; /* W: saved cwq max_active */
265 #ifdef CONFIG_LOCKDEP
266 struct lockdep_map lockdep_map
;
268 char name
[]; /* I: workqueue name */
271 struct workqueue_struct
*system_wq __read_mostly
;
272 EXPORT_SYMBOL_GPL(system_wq
);
273 struct workqueue_struct
*system_highpri_wq __read_mostly
;
274 EXPORT_SYMBOL_GPL(system_highpri_wq
);
275 struct workqueue_struct
*system_long_wq __read_mostly
;
276 EXPORT_SYMBOL_GPL(system_long_wq
);
277 struct workqueue_struct
*system_nrt_wq __read_mostly
;
278 EXPORT_SYMBOL_GPL(system_nrt_wq
);
279 struct workqueue_struct
*system_unbound_wq __read_mostly
;
280 EXPORT_SYMBOL_GPL(system_unbound_wq
);
281 struct workqueue_struct
*system_freezable_wq __read_mostly
;
282 EXPORT_SYMBOL_GPL(system_freezable_wq
);
283 struct workqueue_struct
*system_nrt_freezable_wq __read_mostly
;
284 EXPORT_SYMBOL_GPL(system_nrt_freezable_wq
);
286 #define CREATE_TRACE_POINTS
287 #include <trace/events/workqueue.h>
289 #define for_each_worker_pool(pool, gcwq) \
290 for ((pool) = &(gcwq)->pools[0]; \
291 (pool) < &(gcwq)->pools[NR_WORKER_POOLS]; (pool)++)
293 #define for_each_busy_worker(worker, i, pos, gcwq) \
294 for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++) \
295 hlist_for_each_entry(worker, pos, &gcwq->busy_hash[i], hentry)
297 static inline int __next_gcwq_cpu(int cpu
, const struct cpumask
*mask
,
300 if (cpu
< nr_cpu_ids
) {
302 cpu
= cpumask_next(cpu
, mask
);
303 if (cpu
< nr_cpu_ids
)
307 return WORK_CPU_UNBOUND
;
309 return WORK_CPU_NONE
;
312 static inline int __next_wq_cpu(int cpu
, const struct cpumask
*mask
,
313 struct workqueue_struct
*wq
)
315 return __next_gcwq_cpu(cpu
, mask
, !(wq
->flags
& WQ_UNBOUND
) ? 1 : 2);
321 * An extra gcwq is defined for an invalid cpu number
322 * (WORK_CPU_UNBOUND) to host workqueues which are not bound to any
323 * specific CPU. The following iterators are similar to
324 * for_each_*_cpu() iterators but also considers the unbound gcwq.
326 * for_each_gcwq_cpu() : possible CPUs + WORK_CPU_UNBOUND
327 * for_each_online_gcwq_cpu() : online CPUs + WORK_CPU_UNBOUND
328 * for_each_cwq_cpu() : possible CPUs for bound workqueues,
329 * WORK_CPU_UNBOUND for unbound workqueues
331 #define for_each_gcwq_cpu(cpu) \
332 for ((cpu) = __next_gcwq_cpu(-1, cpu_possible_mask, 3); \
333 (cpu) < WORK_CPU_NONE; \
334 (cpu) = __next_gcwq_cpu((cpu), cpu_possible_mask, 3))
336 #define for_each_online_gcwq_cpu(cpu) \
337 for ((cpu) = __next_gcwq_cpu(-1, cpu_online_mask, 3); \
338 (cpu) < WORK_CPU_NONE; \
339 (cpu) = __next_gcwq_cpu((cpu), cpu_online_mask, 3))
341 #define for_each_cwq_cpu(cpu, wq) \
342 for ((cpu) = __next_wq_cpu(-1, cpu_possible_mask, (wq)); \
343 (cpu) < WORK_CPU_NONE; \
344 (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, (wq)))
346 #ifdef CONFIG_DEBUG_OBJECTS_WORK
348 static struct debug_obj_descr work_debug_descr
;
350 static void *work_debug_hint(void *addr
)
352 return ((struct work_struct
*) addr
)->func
;
356 * fixup_init is called when:
357 * - an active object is initialized
359 static int work_fixup_init(void *addr
, enum debug_obj_state state
)
361 struct work_struct
*work
= addr
;
364 case ODEBUG_STATE_ACTIVE
:
365 cancel_work_sync(work
);
366 debug_object_init(work
, &work_debug_descr
);
374 * fixup_activate is called when:
375 * - an active object is activated
376 * - an unknown object is activated (might be a statically initialized object)
378 static int work_fixup_activate(void *addr
, enum debug_obj_state state
)
380 struct work_struct
*work
= addr
;
384 case ODEBUG_STATE_NOTAVAILABLE
:
386 * This is not really a fixup. The work struct was
387 * statically initialized. We just make sure that it
388 * is tracked in the object tracker.
390 if (test_bit(WORK_STRUCT_STATIC_BIT
, work_data_bits(work
))) {
391 debug_object_init(work
, &work_debug_descr
);
392 debug_object_activate(work
, &work_debug_descr
);
398 case ODEBUG_STATE_ACTIVE
:
407 * fixup_free is called when:
408 * - an active object is freed
410 static int work_fixup_free(void *addr
, enum debug_obj_state state
)
412 struct work_struct
*work
= addr
;
415 case ODEBUG_STATE_ACTIVE
:
416 cancel_work_sync(work
);
417 debug_object_free(work
, &work_debug_descr
);
424 static struct debug_obj_descr work_debug_descr
= {
425 .name
= "work_struct",
426 .debug_hint
= work_debug_hint
,
427 .fixup_init
= work_fixup_init
,
428 .fixup_activate
= work_fixup_activate
,
429 .fixup_free
= work_fixup_free
,
432 static inline void debug_work_activate(struct work_struct
*work
)
434 debug_object_activate(work
, &work_debug_descr
);
437 static inline void debug_work_deactivate(struct work_struct
*work
)
439 debug_object_deactivate(work
, &work_debug_descr
);
442 void __init_work(struct work_struct
*work
, int onstack
)
445 debug_object_init_on_stack(work
, &work_debug_descr
);
447 debug_object_init(work
, &work_debug_descr
);
449 EXPORT_SYMBOL_GPL(__init_work
);
451 void destroy_work_on_stack(struct work_struct
*work
)
453 debug_object_free(work
, &work_debug_descr
);
455 EXPORT_SYMBOL_GPL(destroy_work_on_stack
);
458 static inline void debug_work_activate(struct work_struct
*work
) { }
459 static inline void debug_work_deactivate(struct work_struct
*work
) { }
462 /* Serializes the accesses to the list of workqueues. */
463 static DEFINE_SPINLOCK(workqueue_lock
);
464 static LIST_HEAD(workqueues
);
465 static bool workqueue_freezing
; /* W: have wqs started freezing? */
468 * The almighty global cpu workqueues. nr_running is the only field
469 * which is expected to be used frequently by other cpus via
470 * try_to_wake_up(). Put it in a separate cacheline.
472 static DEFINE_PER_CPU(struct global_cwq
, global_cwq
);
473 static DEFINE_PER_CPU_SHARED_ALIGNED(atomic_t
, pool_nr_running
[NR_WORKER_POOLS
]);
476 * Global cpu workqueue and nr_running counter for unbound gcwq. The
477 * gcwq is always online, has GCWQ_DISASSOCIATED set, and all its
478 * workers have WORKER_UNBOUND set.
480 static struct global_cwq unbound_global_cwq
;
481 static atomic_t unbound_pool_nr_running
[NR_WORKER_POOLS
] = {
482 [0 ... NR_WORKER_POOLS
- 1] = ATOMIC_INIT(0), /* always 0 */
485 static int worker_thread(void *__worker
);
487 static int worker_pool_pri(struct worker_pool
*pool
)
489 return pool
- pool
->gcwq
->pools
;
492 static struct global_cwq
*get_gcwq(unsigned int cpu
)
494 if (cpu
!= WORK_CPU_UNBOUND
)
495 return &per_cpu(global_cwq
, cpu
);
497 return &unbound_global_cwq
;
500 static atomic_t
*get_pool_nr_running(struct worker_pool
*pool
)
502 int cpu
= pool
->gcwq
->cpu
;
503 int idx
= worker_pool_pri(pool
);
505 if (cpu
!= WORK_CPU_UNBOUND
)
506 return &per_cpu(pool_nr_running
, cpu
)[idx
];
508 return &unbound_pool_nr_running
[idx
];
511 static struct cpu_workqueue_struct
*get_cwq(unsigned int cpu
,
512 struct workqueue_struct
*wq
)
514 if (!(wq
->flags
& WQ_UNBOUND
)) {
515 if (likely(cpu
< nr_cpu_ids
))
516 return per_cpu_ptr(wq
->cpu_wq
.pcpu
, cpu
);
517 } else if (likely(cpu
== WORK_CPU_UNBOUND
))
518 return wq
->cpu_wq
.single
;
522 static unsigned int work_color_to_flags(int color
)
524 return color
<< WORK_STRUCT_COLOR_SHIFT
;
527 static int get_work_color(struct work_struct
*work
)
529 return (*work_data_bits(work
) >> WORK_STRUCT_COLOR_SHIFT
) &
530 ((1 << WORK_STRUCT_COLOR_BITS
) - 1);
533 static int work_next_color(int color
)
535 return (color
+ 1) % WORK_NR_COLORS
;
539 * While queued, %WORK_STRUCT_CWQ is set and non flag bits of a work's data
540 * contain the pointer to the queued cwq. Once execution starts, the flag
541 * is cleared and the high bits contain OFFQ flags and CPU number.
543 * set_work_cwq(), set_work_cpu_and_clear_pending(), mark_work_canceling()
544 * and clear_work_data() can be used to set the cwq, cpu or clear
545 * work->data. These functions should only be called while the work is
546 * owned - ie. while the PENDING bit is set.
548 * get_work_[g]cwq() can be used to obtain the gcwq or cwq corresponding to
549 * a work. gcwq is available once the work has been queued anywhere after
550 * initialization until it is sync canceled. cwq is available only while
551 * the work item is queued.
553 * %WORK_OFFQ_CANCELING is used to mark a work item which is being
554 * canceled. While being canceled, a work item may have its PENDING set
555 * but stay off timer and worklist for arbitrarily long and nobody should
556 * try to steal the PENDING bit.
558 static inline void set_work_data(struct work_struct
*work
, unsigned long data
,
561 BUG_ON(!work_pending(work
));
562 atomic_long_set(&work
->data
, data
| flags
| work_static(work
));
565 static void set_work_cwq(struct work_struct
*work
,
566 struct cpu_workqueue_struct
*cwq
,
567 unsigned long extra_flags
)
569 set_work_data(work
, (unsigned long)cwq
,
570 WORK_STRUCT_PENDING
| WORK_STRUCT_CWQ
| extra_flags
);
573 static void set_work_cpu_and_clear_pending(struct work_struct
*work
,
577 * The following wmb is paired with the implied mb in
578 * test_and_set_bit(PENDING) and ensures all updates to @work made
579 * here are visible to and precede any updates by the next PENDING
583 set_work_data(work
, (unsigned long)cpu
<< WORK_OFFQ_CPU_SHIFT
, 0);
586 static void clear_work_data(struct work_struct
*work
)
588 smp_wmb(); /* see set_work_cpu_and_clear_pending() */
589 set_work_data(work
, WORK_STRUCT_NO_CPU
, 0);
592 static struct cpu_workqueue_struct
*get_work_cwq(struct work_struct
*work
)
594 unsigned long data
= atomic_long_read(&work
->data
);
596 if (data
& WORK_STRUCT_CWQ
)
597 return (void *)(data
& WORK_STRUCT_WQ_DATA_MASK
);
602 static struct global_cwq
*get_work_gcwq(struct work_struct
*work
)
604 unsigned long data
= atomic_long_read(&work
->data
);
607 if (data
& WORK_STRUCT_CWQ
)
608 return ((struct cpu_workqueue_struct
*)
609 (data
& WORK_STRUCT_WQ_DATA_MASK
))->pool
->gcwq
;
611 cpu
= data
>> WORK_OFFQ_CPU_SHIFT
;
612 if (cpu
== WORK_CPU_NONE
)
615 BUG_ON(cpu
>= nr_cpu_ids
&& cpu
!= WORK_CPU_UNBOUND
);
616 return get_gcwq(cpu
);
619 static void mark_work_canceling(struct work_struct
*work
)
621 struct global_cwq
*gcwq
= get_work_gcwq(work
);
622 unsigned long cpu
= gcwq
? gcwq
->cpu
: WORK_CPU_NONE
;
624 set_work_data(work
, (cpu
<< WORK_OFFQ_CPU_SHIFT
) | WORK_OFFQ_CANCELING
,
625 WORK_STRUCT_PENDING
);
628 static bool work_is_canceling(struct work_struct
*work
)
630 unsigned long data
= atomic_long_read(&work
->data
);
632 return !(data
& WORK_STRUCT_CWQ
) && (data
& WORK_OFFQ_CANCELING
);
636 * Policy functions. These define the policies on how the global worker
637 * pools are managed. Unless noted otherwise, these functions assume that
638 * they're being called with gcwq->lock held.
641 static bool __need_more_worker(struct worker_pool
*pool
)
643 return !atomic_read(get_pool_nr_running(pool
));
647 * Need to wake up a worker? Called from anything but currently
650 * Note that, because unbound workers never contribute to nr_running, this
651 * function will always return %true for unbound gcwq as long as the
652 * worklist isn't empty.
654 static bool need_more_worker(struct worker_pool
*pool
)
656 return !list_empty(&pool
->worklist
) && __need_more_worker(pool
);
659 /* Can I start working? Called from busy but !running workers. */
660 static bool may_start_working(struct worker_pool
*pool
)
662 return pool
->nr_idle
;
665 /* Do I need to keep working? Called from currently running workers. */
666 static bool keep_working(struct worker_pool
*pool
)
668 atomic_t
*nr_running
= get_pool_nr_running(pool
);
670 return !list_empty(&pool
->worklist
) && atomic_read(nr_running
) <= 1;
673 /* Do we need a new worker? Called from manager. */
674 static bool need_to_create_worker(struct worker_pool
*pool
)
676 return need_more_worker(pool
) && !may_start_working(pool
);
679 /* Do I need to be the manager? */
680 static bool need_to_manage_workers(struct worker_pool
*pool
)
682 return need_to_create_worker(pool
) ||
683 (pool
->flags
& POOL_MANAGE_WORKERS
);
686 /* Do we have too many workers and should some go away? */
687 static bool too_many_workers(struct worker_pool
*pool
)
689 bool managing
= mutex_is_locked(&pool
->manager_mutex
);
690 int nr_idle
= pool
->nr_idle
+ managing
; /* manager is considered idle */
691 int nr_busy
= pool
->nr_workers
- nr_idle
;
693 return nr_idle
> 2 && (nr_idle
- 2) * MAX_IDLE_WORKERS_RATIO
>= nr_busy
;
700 /* Return the first worker. Safe with preemption disabled */
701 static struct worker
*first_worker(struct worker_pool
*pool
)
703 if (unlikely(list_empty(&pool
->idle_list
)))
706 return list_first_entry(&pool
->idle_list
, struct worker
, entry
);
710 * wake_up_worker - wake up an idle worker
711 * @pool: worker pool to wake worker from
713 * Wake up the first idle worker of @pool.
716 * spin_lock_irq(gcwq->lock).
718 static void wake_up_worker(struct worker_pool
*pool
)
720 struct worker
*worker
= first_worker(pool
);
723 wake_up_process(worker
->task
);
727 * wq_worker_waking_up - a worker is waking up
728 * @task: task waking up
729 * @cpu: CPU @task is waking up to
731 * This function is called during try_to_wake_up() when a worker is
735 * spin_lock_irq(rq->lock)
737 void wq_worker_waking_up(struct task_struct
*task
, unsigned int cpu
)
739 struct worker
*worker
= kthread_data(task
);
741 if (!(worker
->flags
& WORKER_NOT_RUNNING
))
742 atomic_inc(get_pool_nr_running(worker
->pool
));
746 * wq_worker_sleeping - a worker is going to sleep
747 * @task: task going to sleep
748 * @cpu: CPU in question, must be the current CPU number
750 * This function is called during schedule() when a busy worker is
751 * going to sleep. Worker on the same cpu can be woken up by
752 * returning pointer to its task.
755 * spin_lock_irq(rq->lock)
758 * Worker task on @cpu to wake up, %NULL if none.
760 struct task_struct
*wq_worker_sleeping(struct task_struct
*task
,
763 struct worker
*worker
= kthread_data(task
), *to_wakeup
= NULL
;
764 struct worker_pool
*pool
= worker
->pool
;
765 atomic_t
*nr_running
= get_pool_nr_running(pool
);
767 if (worker
->flags
& WORKER_NOT_RUNNING
)
770 /* this can only happen on the local cpu */
771 BUG_ON(cpu
!= raw_smp_processor_id());
774 * The counterpart of the following dec_and_test, implied mb,
775 * worklist not empty test sequence is in insert_work().
776 * Please read comment there.
778 * NOT_RUNNING is clear. This means that we're bound to and
779 * running on the local cpu w/ rq lock held and preemption
780 * disabled, which in turn means that none else could be
781 * manipulating idle_list, so dereferencing idle_list without gcwq
784 if (atomic_dec_and_test(nr_running
) && !list_empty(&pool
->worklist
))
785 to_wakeup
= first_worker(pool
);
786 return to_wakeup
? to_wakeup
->task
: NULL
;
790 * worker_set_flags - set worker flags and adjust nr_running accordingly
792 * @flags: flags to set
793 * @wakeup: wakeup an idle worker if necessary
795 * Set @flags in @worker->flags and adjust nr_running accordingly. If
796 * nr_running becomes zero and @wakeup is %true, an idle worker is
800 * spin_lock_irq(gcwq->lock)
802 static inline void worker_set_flags(struct worker
*worker
, unsigned int flags
,
805 struct worker_pool
*pool
= worker
->pool
;
807 WARN_ON_ONCE(worker
->task
!= current
);
810 * If transitioning into NOT_RUNNING, adjust nr_running and
811 * wake up an idle worker as necessary if requested by
814 if ((flags
& WORKER_NOT_RUNNING
) &&
815 !(worker
->flags
& WORKER_NOT_RUNNING
)) {
816 atomic_t
*nr_running
= get_pool_nr_running(pool
);
819 if (atomic_dec_and_test(nr_running
) &&
820 !list_empty(&pool
->worklist
))
821 wake_up_worker(pool
);
823 atomic_dec(nr_running
);
826 worker
->flags
|= flags
;
830 * worker_clr_flags - clear worker flags and adjust nr_running accordingly
832 * @flags: flags to clear
834 * Clear @flags in @worker->flags and adjust nr_running accordingly.
837 * spin_lock_irq(gcwq->lock)
839 static inline void worker_clr_flags(struct worker
*worker
, unsigned int flags
)
841 struct worker_pool
*pool
= worker
->pool
;
842 unsigned int oflags
= worker
->flags
;
844 WARN_ON_ONCE(worker
->task
!= current
);
846 worker
->flags
&= ~flags
;
849 * If transitioning out of NOT_RUNNING, increment nr_running. Note
850 * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask
851 * of multiple flags, not a single flag.
853 if ((flags
& WORKER_NOT_RUNNING
) && (oflags
& WORKER_NOT_RUNNING
))
854 if (!(worker
->flags
& WORKER_NOT_RUNNING
))
855 atomic_inc(get_pool_nr_running(pool
));
859 * busy_worker_head - return the busy hash head for a work
860 * @gcwq: gcwq of interest
861 * @work: work to be hashed
863 * Return hash head of @gcwq for @work.
866 * spin_lock_irq(gcwq->lock).
869 * Pointer to the hash head.
871 static struct hlist_head
*busy_worker_head(struct global_cwq
*gcwq
,
872 struct work_struct
*work
)
874 const int base_shift
= ilog2(sizeof(struct work_struct
));
875 unsigned long v
= (unsigned long)work
;
877 /* simple shift and fold hash, do we need something better? */
879 v
+= v
>> BUSY_WORKER_HASH_ORDER
;
880 v
&= BUSY_WORKER_HASH_MASK
;
882 return &gcwq
->busy_hash
[v
];
886 * __find_worker_executing_work - find worker which is executing a work
887 * @gcwq: gcwq of interest
888 * @bwh: hash head as returned by busy_worker_head()
889 * @work: work to find worker for
891 * Find a worker which is executing @work on @gcwq. @bwh should be
892 * the hash head obtained by calling busy_worker_head() with the same
896 * spin_lock_irq(gcwq->lock).
899 * Pointer to worker which is executing @work if found, NULL
902 static struct worker
*__find_worker_executing_work(struct global_cwq
*gcwq
,
903 struct hlist_head
*bwh
,
904 struct work_struct
*work
)
906 struct worker
*worker
;
907 struct hlist_node
*tmp
;
909 hlist_for_each_entry(worker
, tmp
, bwh
, hentry
)
910 if (worker
->current_work
== work
)
916 * find_worker_executing_work - find worker which is executing a work
917 * @gcwq: gcwq of interest
918 * @work: work to find worker for
920 * Find a worker which is executing @work on @gcwq. This function is
921 * identical to __find_worker_executing_work() except that this
922 * function calculates @bwh itself.
925 * spin_lock_irq(gcwq->lock).
928 * Pointer to worker which is executing @work if found, NULL
931 static struct worker
*find_worker_executing_work(struct global_cwq
*gcwq
,
932 struct work_struct
*work
)
934 return __find_worker_executing_work(gcwq
, busy_worker_head(gcwq
, work
),
939 * move_linked_works - move linked works to a list
940 * @work: start of series of works to be scheduled
941 * @head: target list to append @work to
942 * @nextp: out paramter for nested worklist walking
944 * Schedule linked works starting from @work to @head. Work series to
945 * be scheduled starts at @work and includes any consecutive work with
946 * WORK_STRUCT_LINKED set in its predecessor.
948 * If @nextp is not NULL, it's updated to point to the next work of
949 * the last scheduled work. This allows move_linked_works() to be
950 * nested inside outer list_for_each_entry_safe().
953 * spin_lock_irq(gcwq->lock).
955 static void move_linked_works(struct work_struct
*work
, struct list_head
*head
,
956 struct work_struct
**nextp
)
958 struct work_struct
*n
;
961 * Linked worklist will always end before the end of the list,
962 * use NULL for list head.
964 list_for_each_entry_safe_from(work
, n
, NULL
, entry
) {
965 list_move_tail(&work
->entry
, head
);
966 if (!(*work_data_bits(work
) & WORK_STRUCT_LINKED
))
971 * If we're already inside safe list traversal and have moved
972 * multiple works to the scheduled queue, the next position
973 * needs to be updated.
979 static void cwq_activate_first_delayed(struct cpu_workqueue_struct
*cwq
)
981 struct work_struct
*work
= list_first_entry(&cwq
->delayed_works
,
982 struct work_struct
, entry
);
984 trace_workqueue_activate_work(work
);
985 move_linked_works(work
, &cwq
->pool
->worklist
, NULL
);
986 __clear_bit(WORK_STRUCT_DELAYED_BIT
, work_data_bits(work
));
991 * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight
992 * @cwq: cwq of interest
993 * @color: color of work which left the queue
994 * @delayed: for a delayed work
996 * A work either has completed or is removed from pending queue,
997 * decrement nr_in_flight of its cwq and handle workqueue flushing.
1000 * spin_lock_irq(gcwq->lock).
1002 static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct
*cwq
, int color
,
1005 /* ignore uncolored works */
1006 if (color
== WORK_NO_COLOR
)
1009 cwq
->nr_in_flight
[color
]--;
1013 if (!list_empty(&cwq
->delayed_works
)) {
1014 /* one down, submit a delayed one */
1015 if (cwq
->nr_active
< cwq
->max_active
)
1016 cwq_activate_first_delayed(cwq
);
1020 /* is flush in progress and are we at the flushing tip? */
1021 if (likely(cwq
->flush_color
!= color
))
1024 /* are there still in-flight works? */
1025 if (cwq
->nr_in_flight
[color
])
1028 /* this cwq is done, clear flush_color */
1029 cwq
->flush_color
= -1;
1032 * If this was the last cwq, wake up the first flusher. It
1033 * will handle the rest.
1035 if (atomic_dec_and_test(&cwq
->wq
->nr_cwqs_to_flush
))
1036 complete(&cwq
->wq
->first_flusher
->done
);
1040 * try_to_grab_pending - steal work item from worklist and disable irq
1041 * @work: work item to steal
1042 * @is_dwork: @work is a delayed_work
1043 * @flags: place to store irq state
1045 * Try to grab PENDING bit of @work. This function can handle @work in any
1046 * stable state - idle, on timer or on worklist. Return values are
1048 * 1 if @work was pending and we successfully stole PENDING
1049 * 0 if @work was idle and we claimed PENDING
1050 * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry
1051 * -ENOENT if someone else is canceling @work, this state may persist
1052 * for arbitrarily long
1054 * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting
1055 * preempted while holding PENDING and @work off queue, preemption must be
1056 * disabled on entry. This ensures that we don't return -EAGAIN while
1057 * another task is preempted in this function.
1059 * On successful return, >= 0, irq is disabled and the caller is
1060 * responsible for releasing it using local_irq_restore(*@flags).
1062 * This function is safe to call from any context other than IRQ handler.
1063 * An IRQ handler may run on top of delayed_work_timer_fn() which can make
1064 * this function return -EAGAIN perpetually.
1066 static int try_to_grab_pending(struct work_struct
*work
, bool is_dwork
,
1067 unsigned long *flags
)
1069 struct global_cwq
*gcwq
;
1071 WARN_ON_ONCE(in_irq());
1073 local_irq_save(*flags
);
1075 /* try to steal the timer if it exists */
1077 struct delayed_work
*dwork
= to_delayed_work(work
);
1079 if (likely(del_timer(&dwork
->timer
)))
1083 /* try to claim PENDING the normal way */
1084 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT
, work_data_bits(work
)))
1088 * The queueing is in progress, or it is already queued. Try to
1089 * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
1091 gcwq
= get_work_gcwq(work
);
1095 spin_lock(&gcwq
->lock
);
1096 if (!list_empty(&work
->entry
)) {
1098 * This work is queued, but perhaps we locked the wrong gcwq.
1099 * In that case we must see the new value after rmb(), see
1100 * insert_work()->wmb().
1103 if (gcwq
== get_work_gcwq(work
)) {
1104 debug_work_deactivate(work
);
1105 list_del_init(&work
->entry
);
1106 cwq_dec_nr_in_flight(get_work_cwq(work
),
1107 get_work_color(work
),
1108 *work_data_bits(work
) & WORK_STRUCT_DELAYED
);
1110 spin_unlock(&gcwq
->lock
);
1114 spin_unlock(&gcwq
->lock
);
1116 local_irq_restore(*flags
);
1117 if (work_is_canceling(work
))
1124 * insert_work - insert a work into gcwq
1125 * @cwq: cwq @work belongs to
1126 * @work: work to insert
1127 * @head: insertion point
1128 * @extra_flags: extra WORK_STRUCT_* flags to set
1130 * Insert @work which belongs to @cwq into @gcwq after @head.
1131 * @extra_flags is or'd to work_struct flags.
1134 * spin_lock_irq(gcwq->lock).
1136 static void insert_work(struct cpu_workqueue_struct
*cwq
,
1137 struct work_struct
*work
, struct list_head
*head
,
1138 unsigned int extra_flags
)
1140 struct worker_pool
*pool
= cwq
->pool
;
1142 /* we own @work, set data and link */
1143 set_work_cwq(work
, cwq
, extra_flags
);
1146 * Ensure that we get the right work->data if we see the
1147 * result of list_add() below, see try_to_grab_pending().
1151 list_add_tail(&work
->entry
, head
);
1154 * Ensure either worker_sched_deactivated() sees the above
1155 * list_add_tail() or we see zero nr_running to avoid workers
1156 * lying around lazily while there are works to be processed.
1160 if (__need_more_worker(pool
))
1161 wake_up_worker(pool
);
1165 * Test whether @work is being queued from another work executing on the
1166 * same workqueue. This is rather expensive and should only be used from
1169 static bool is_chained_work(struct workqueue_struct
*wq
)
1171 unsigned long flags
;
1174 for_each_gcwq_cpu(cpu
) {
1175 struct global_cwq
*gcwq
= get_gcwq(cpu
);
1176 struct worker
*worker
;
1177 struct hlist_node
*pos
;
1180 spin_lock_irqsave(&gcwq
->lock
, flags
);
1181 for_each_busy_worker(worker
, i
, pos
, gcwq
) {
1182 if (worker
->task
!= current
)
1184 spin_unlock_irqrestore(&gcwq
->lock
, flags
);
1186 * I'm @worker, no locking necessary. See if @work
1187 * is headed to the same workqueue.
1189 return worker
->current_cwq
->wq
== wq
;
1191 spin_unlock_irqrestore(&gcwq
->lock
, flags
);
1196 static void __queue_work(unsigned int cpu
, struct workqueue_struct
*wq
,
1197 struct work_struct
*work
)
1199 struct global_cwq
*gcwq
;
1200 struct cpu_workqueue_struct
*cwq
;
1201 struct list_head
*worklist
;
1202 unsigned int work_flags
;
1203 unsigned int req_cpu
= cpu
;
1206 * While a work item is PENDING && off queue, a task trying to
1207 * steal the PENDING will busy-loop waiting for it to either get
1208 * queued or lose PENDING. Grabbing PENDING and queueing should
1209 * happen with IRQ disabled.
1211 WARN_ON_ONCE(!irqs_disabled());
1213 debug_work_activate(work
);
1215 /* if dying, only works from the same workqueue are allowed */
1216 if (unlikely(wq
->flags
& WQ_DRAINING
) &&
1217 WARN_ON_ONCE(!is_chained_work(wq
)))
1220 /* determine gcwq to use */
1221 if (!(wq
->flags
& WQ_UNBOUND
)) {
1222 struct global_cwq
*last_gcwq
;
1224 if (cpu
== WORK_CPU_UNBOUND
)
1225 cpu
= raw_smp_processor_id();
1228 * It's multi cpu. If @work was previously on a different
1229 * cpu, it might still be running there, in which case the
1230 * work needs to be queued on that cpu to guarantee
1233 gcwq
= get_gcwq(cpu
);
1234 last_gcwq
= get_work_gcwq(work
);
1236 if (last_gcwq
&& last_gcwq
!= gcwq
) {
1237 struct worker
*worker
;
1239 spin_lock(&last_gcwq
->lock
);
1241 worker
= find_worker_executing_work(last_gcwq
, work
);
1243 if (worker
&& worker
->current_cwq
->wq
== wq
)
1246 /* meh... not running there, queue here */
1247 spin_unlock(&last_gcwq
->lock
);
1248 spin_lock(&gcwq
->lock
);
1251 spin_lock(&gcwq
->lock
);
1254 gcwq
= get_gcwq(WORK_CPU_UNBOUND
);
1255 spin_lock(&gcwq
->lock
);
1258 /* gcwq determined, get cwq and queue */
1259 cwq
= get_cwq(gcwq
->cpu
, wq
);
1260 trace_workqueue_queue_work(req_cpu
, cwq
, work
);
1262 if (WARN_ON(!list_empty(&work
->entry
))) {
1263 spin_unlock(&gcwq
->lock
);
1267 cwq
->nr_in_flight
[cwq
->work_color
]++;
1268 work_flags
= work_color_to_flags(cwq
->work_color
);
1270 if (likely(cwq
->nr_active
< cwq
->max_active
)) {
1271 trace_workqueue_activate_work(work
);
1273 worklist
= &cwq
->pool
->worklist
;
1275 work_flags
|= WORK_STRUCT_DELAYED
;
1276 worklist
= &cwq
->delayed_works
;
1279 insert_work(cwq
, work
, worklist
, work_flags
);
1281 spin_unlock(&gcwq
->lock
);
1285 * queue_work_on - queue work on specific cpu
1286 * @cpu: CPU number to execute work on
1287 * @wq: workqueue to use
1288 * @work: work to queue
1290 * Returns %false if @work was already on a queue, %true otherwise.
1292 * We queue the work to a specific CPU, the caller must ensure it
1295 bool queue_work_on(int cpu
, struct workqueue_struct
*wq
,
1296 struct work_struct
*work
)
1299 unsigned long flags
;
1301 local_irq_save(flags
);
1303 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT
, work_data_bits(work
))) {
1304 __queue_work(cpu
, wq
, work
);
1308 local_irq_restore(flags
);
1311 EXPORT_SYMBOL_GPL(queue_work_on
);
1314 * queue_work - queue work on a workqueue
1315 * @wq: workqueue to use
1316 * @work: work to queue
1318 * Returns %false if @work was already on a queue, %true otherwise.
1320 * We queue the work to the CPU on which it was submitted, but if the CPU dies
1321 * it can be processed by another CPU.
1323 bool queue_work(struct workqueue_struct
*wq
, struct work_struct
*work
)
1325 return queue_work_on(WORK_CPU_UNBOUND
, wq
, work
);
1327 EXPORT_SYMBOL_GPL(queue_work
);
1329 void delayed_work_timer_fn(unsigned long __data
)
1331 struct delayed_work
*dwork
= (struct delayed_work
*)__data
;
1332 struct cpu_workqueue_struct
*cwq
= get_work_cwq(&dwork
->work
);
1334 local_irq_disable();
1335 __queue_work(dwork
->cpu
, cwq
->wq
, &dwork
->work
);
1338 EXPORT_SYMBOL_GPL(delayed_work_timer_fn
);
1340 static void __queue_delayed_work(int cpu
, struct workqueue_struct
*wq
,
1341 struct delayed_work
*dwork
, unsigned long delay
)
1343 struct timer_list
*timer
= &dwork
->timer
;
1344 struct work_struct
*work
= &dwork
->work
;
1347 WARN_ON_ONCE(timer
->function
!= delayed_work_timer_fn
||
1348 timer
->data
!= (unsigned long)dwork
);
1349 BUG_ON(timer_pending(timer
));
1350 BUG_ON(!list_empty(&work
->entry
));
1352 timer_stats_timer_set_start_info(&dwork
->timer
);
1355 * This stores cwq for the moment, for the timer_fn. Note that the
1356 * work's gcwq is preserved to allow reentrance detection for
1359 if (!(wq
->flags
& WQ_UNBOUND
)) {
1360 struct global_cwq
*gcwq
= get_work_gcwq(work
);
1363 * If we cannot get the last gcwq from @work directly,
1364 * select the last CPU such that it avoids unnecessarily
1365 * triggering non-reentrancy check in __queue_work().
1370 if (lcpu
== WORK_CPU_UNBOUND
)
1371 lcpu
= raw_smp_processor_id();
1373 lcpu
= WORK_CPU_UNBOUND
;
1376 set_work_cwq(work
, get_cwq(lcpu
, wq
), 0);
1379 timer
->expires
= jiffies
+ delay
;
1381 if (unlikely(cpu
!= WORK_CPU_UNBOUND
))
1382 add_timer_on(timer
, cpu
);
1388 * queue_delayed_work_on - queue work on specific CPU after delay
1389 * @cpu: CPU number to execute work on
1390 * @wq: workqueue to use
1391 * @dwork: work to queue
1392 * @delay: number of jiffies to wait before queueing
1394 * Returns %false if @work was already on a queue, %true otherwise. If
1395 * @delay is zero and @dwork is idle, it will be scheduled for immediate
1398 bool queue_delayed_work_on(int cpu
, struct workqueue_struct
*wq
,
1399 struct delayed_work
*dwork
, unsigned long delay
)
1401 struct work_struct
*work
= &dwork
->work
;
1403 unsigned long flags
;
1406 return queue_work_on(cpu
, wq
, &dwork
->work
);
1408 /* read the comment in __queue_work() */
1409 local_irq_save(flags
);
1411 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT
, work_data_bits(work
))) {
1412 __queue_delayed_work(cpu
, wq
, dwork
, delay
);
1416 local_irq_restore(flags
);
1419 EXPORT_SYMBOL_GPL(queue_delayed_work_on
);
1422 * queue_delayed_work - queue work on a workqueue after delay
1423 * @wq: workqueue to use
1424 * @dwork: delayable work to queue
1425 * @delay: number of jiffies to wait before queueing
1427 * Equivalent to queue_delayed_work_on() but tries to use the local CPU.
1429 bool queue_delayed_work(struct workqueue_struct
*wq
,
1430 struct delayed_work
*dwork
, unsigned long delay
)
1432 return queue_delayed_work_on(WORK_CPU_UNBOUND
, wq
, dwork
, delay
);
1434 EXPORT_SYMBOL_GPL(queue_delayed_work
);
1437 * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU
1438 * @cpu: CPU number to execute work on
1439 * @wq: workqueue to use
1440 * @dwork: work to queue
1441 * @delay: number of jiffies to wait before queueing
1443 * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise,
1444 * modify @dwork's timer so that it expires after @delay. If @delay is
1445 * zero, @work is guaranteed to be scheduled immediately regardless of its
1448 * Returns %false if @dwork was idle and queued, %true if @dwork was
1449 * pending and its timer was modified.
1451 * This function is safe to call from any context other than IRQ handler.
1452 * See try_to_grab_pending() for details.
1454 bool mod_delayed_work_on(int cpu
, struct workqueue_struct
*wq
,
1455 struct delayed_work
*dwork
, unsigned long delay
)
1457 unsigned long flags
;
1461 ret
= try_to_grab_pending(&dwork
->work
, true, &flags
);
1462 } while (unlikely(ret
== -EAGAIN
));
1464 if (likely(ret
>= 0)) {
1465 __queue_delayed_work(cpu
, wq
, dwork
, delay
);
1466 local_irq_restore(flags
);
1469 /* -ENOENT from try_to_grab_pending() becomes %true */
1472 EXPORT_SYMBOL_GPL(mod_delayed_work_on
);
1475 * mod_delayed_work - modify delay of or queue a delayed work
1476 * @wq: workqueue to use
1477 * @dwork: work to queue
1478 * @delay: number of jiffies to wait before queueing
1480 * mod_delayed_work_on() on local CPU.
1482 bool mod_delayed_work(struct workqueue_struct
*wq
, struct delayed_work
*dwork
,
1483 unsigned long delay
)
1485 return mod_delayed_work_on(WORK_CPU_UNBOUND
, wq
, dwork
, delay
);
1487 EXPORT_SYMBOL_GPL(mod_delayed_work
);
1490 * worker_enter_idle - enter idle state
1491 * @worker: worker which is entering idle state
1493 * @worker is entering idle state. Update stats and idle timer if
1497 * spin_lock_irq(gcwq->lock).
1499 static void worker_enter_idle(struct worker
*worker
)
1501 struct worker_pool
*pool
= worker
->pool
;
1502 struct global_cwq
*gcwq
= pool
->gcwq
;
1504 BUG_ON(worker
->flags
& WORKER_IDLE
);
1505 BUG_ON(!list_empty(&worker
->entry
) &&
1506 (worker
->hentry
.next
|| worker
->hentry
.pprev
));
1508 /* can't use worker_set_flags(), also called from start_worker() */
1509 worker
->flags
|= WORKER_IDLE
;
1511 worker
->last_active
= jiffies
;
1513 /* idle_list is LIFO */
1514 list_add(&worker
->entry
, &pool
->idle_list
);
1516 if (too_many_workers(pool
) && !timer_pending(&pool
->idle_timer
))
1517 mod_timer(&pool
->idle_timer
, jiffies
+ IDLE_WORKER_TIMEOUT
);
1520 * Sanity check nr_running. Because gcwq_unbind_fn() releases
1521 * gcwq->lock between setting %WORKER_UNBOUND and zapping
1522 * nr_running, the warning may trigger spuriously. Check iff
1523 * unbind is not in progress.
1525 WARN_ON_ONCE(!(gcwq
->flags
& GCWQ_DISASSOCIATED
) &&
1526 pool
->nr_workers
== pool
->nr_idle
&&
1527 atomic_read(get_pool_nr_running(pool
)));
1531 * worker_leave_idle - leave idle state
1532 * @worker: worker which is leaving idle state
1534 * @worker is leaving idle state. Update stats.
1537 * spin_lock_irq(gcwq->lock).
1539 static void worker_leave_idle(struct worker
*worker
)
1541 struct worker_pool
*pool
= worker
->pool
;
1543 BUG_ON(!(worker
->flags
& WORKER_IDLE
));
1544 worker_clr_flags(worker
, WORKER_IDLE
);
1546 list_del_init(&worker
->entry
);
1550 * worker_maybe_bind_and_lock - bind worker to its cpu if possible and lock gcwq
1553 * Works which are scheduled while the cpu is online must at least be
1554 * scheduled to a worker which is bound to the cpu so that if they are
1555 * flushed from cpu callbacks while cpu is going down, they are
1556 * guaranteed to execute on the cpu.
1558 * This function is to be used by rogue workers and rescuers to bind
1559 * themselves to the target cpu and may race with cpu going down or
1560 * coming online. kthread_bind() can't be used because it may put the
1561 * worker to already dead cpu and set_cpus_allowed_ptr() can't be used
1562 * verbatim as it's best effort and blocking and gcwq may be
1563 * [dis]associated in the meantime.
1565 * This function tries set_cpus_allowed() and locks gcwq and verifies the
1566 * binding against %GCWQ_DISASSOCIATED which is set during
1567 * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker
1568 * enters idle state or fetches works without dropping lock, it can
1569 * guarantee the scheduling requirement described in the first paragraph.
1572 * Might sleep. Called without any lock but returns with gcwq->lock
1576 * %true if the associated gcwq is online (@worker is successfully
1577 * bound), %false if offline.
1579 static bool worker_maybe_bind_and_lock(struct worker
*worker
)
1580 __acquires(&gcwq
->lock
)
1582 struct global_cwq
*gcwq
= worker
->pool
->gcwq
;
1583 struct task_struct
*task
= worker
->task
;
1587 * The following call may fail, succeed or succeed
1588 * without actually migrating the task to the cpu if
1589 * it races with cpu hotunplug operation. Verify
1590 * against GCWQ_DISASSOCIATED.
1592 if (!(gcwq
->flags
& GCWQ_DISASSOCIATED
))
1593 set_cpus_allowed_ptr(task
, get_cpu_mask(gcwq
->cpu
));
1595 spin_lock_irq(&gcwq
->lock
);
1596 if (gcwq
->flags
& GCWQ_DISASSOCIATED
)
1598 if (task_cpu(task
) == gcwq
->cpu
&&
1599 cpumask_equal(¤t
->cpus_allowed
,
1600 get_cpu_mask(gcwq
->cpu
)))
1602 spin_unlock_irq(&gcwq
->lock
);
1605 * We've raced with CPU hot[un]plug. Give it a breather
1606 * and retry migration. cond_resched() is required here;
1607 * otherwise, we might deadlock against cpu_stop trying to
1608 * bring down the CPU on non-preemptive kernel.
1615 struct idle_rebind
{
1616 int cnt
; /* # workers to be rebound */
1617 struct completion done
; /* all workers rebound */
1621 * Rebind an idle @worker to its CPU. During CPU onlining, this has to
1622 * happen synchronously for idle workers. worker_thread() will test
1623 * %WORKER_REBIND before leaving idle and call this function.
1625 static void idle_worker_rebind(struct worker
*worker
)
1627 struct global_cwq
*gcwq
= worker
->pool
->gcwq
;
1629 /* CPU must be online at this point */
1630 WARN_ON(!worker_maybe_bind_and_lock(worker
));
1631 if (!--worker
->idle_rebind
->cnt
)
1632 complete(&worker
->idle_rebind
->done
);
1633 spin_unlock_irq(&worker
->pool
->gcwq
->lock
);
1635 /* we did our part, wait for rebind_workers() to finish up */
1636 wait_event(gcwq
->rebind_hold
, !(worker
->flags
& WORKER_REBIND
));
1640 * Function for @worker->rebind.work used to rebind unbound busy workers to
1641 * the associated cpu which is coming back online. This is scheduled by
1642 * cpu up but can race with other cpu hotplug operations and may be
1643 * executed twice without intervening cpu down.
1645 static void busy_worker_rebind_fn(struct work_struct
*work
)
1647 struct worker
*worker
= container_of(work
, struct worker
, rebind_work
);
1648 struct global_cwq
*gcwq
= worker
->pool
->gcwq
;
1650 if (worker_maybe_bind_and_lock(worker
))
1651 worker_clr_flags(worker
, WORKER_REBIND
);
1653 spin_unlock_irq(&gcwq
->lock
);
1657 * rebind_workers - rebind all workers of a gcwq to the associated CPU
1658 * @gcwq: gcwq of interest
1660 * @gcwq->cpu is coming online. Rebind all workers to the CPU. Rebinding
1661 * is different for idle and busy ones.
1663 * The idle ones should be rebound synchronously and idle rebinding should
1664 * be complete before any worker starts executing work items with
1665 * concurrency management enabled; otherwise, scheduler may oops trying to
1666 * wake up non-local idle worker from wq_worker_sleeping().
1668 * This is achieved by repeatedly requesting rebinding until all idle
1669 * workers are known to have been rebound under @gcwq->lock and holding all
1670 * idle workers from becoming busy until idle rebinding is complete.
1672 * Once idle workers are rebound, busy workers can be rebound as they
1673 * finish executing their current work items. Queueing the rebind work at
1674 * the head of their scheduled lists is enough. Note that nr_running will
1675 * be properbly bumped as busy workers rebind.
1677 * On return, all workers are guaranteed to either be bound or have rebind
1678 * work item scheduled.
1680 static void rebind_workers(struct global_cwq
*gcwq
)
1681 __releases(&gcwq
->lock
) __acquires(&gcwq
->lock
)
1683 struct idle_rebind idle_rebind
;
1684 struct worker_pool
*pool
;
1685 struct worker
*worker
;
1686 struct hlist_node
*pos
;
1689 lockdep_assert_held(&gcwq
->lock
);
1691 for_each_worker_pool(pool
, gcwq
)
1692 lockdep_assert_held(&pool
->manager_mutex
);
1695 * Rebind idle workers. Interlocked both ways. We wait for
1696 * workers to rebind via @idle_rebind.done. Workers will wait for
1697 * us to finish up by watching %WORKER_REBIND.
1699 init_completion(&idle_rebind
.done
);
1701 idle_rebind
.cnt
= 1;
1702 INIT_COMPLETION(idle_rebind
.done
);
1704 /* set REBIND and kick idle ones, we'll wait for these later */
1705 for_each_worker_pool(pool
, gcwq
) {
1706 list_for_each_entry(worker
, &pool
->idle_list
, entry
) {
1707 if (worker
->flags
& WORKER_REBIND
)
1710 /* morph UNBOUND to REBIND */
1711 worker
->flags
&= ~WORKER_UNBOUND
;
1712 worker
->flags
|= WORKER_REBIND
;
1715 worker
->idle_rebind
= &idle_rebind
;
1717 /* worker_thread() will call idle_worker_rebind() */
1718 wake_up_process(worker
->task
);
1722 if (--idle_rebind
.cnt
) {
1723 spin_unlock_irq(&gcwq
->lock
);
1724 wait_for_completion(&idle_rebind
.done
);
1725 spin_lock_irq(&gcwq
->lock
);
1726 /* busy ones might have become idle while waiting, retry */
1731 * All idle workers are rebound and waiting for %WORKER_REBIND to
1732 * be cleared inside idle_worker_rebind(). Clear and release.
1733 * Clearing %WORKER_REBIND from this foreign context is safe
1734 * because these workers are still guaranteed to be idle.
1736 for_each_worker_pool(pool
, gcwq
)
1737 list_for_each_entry(worker
, &pool
->idle_list
, entry
)
1738 worker
->flags
&= ~WORKER_REBIND
;
1740 wake_up_all(&gcwq
->rebind_hold
);
1742 /* rebind busy workers */
1743 for_each_busy_worker(worker
, i
, pos
, gcwq
) {
1744 struct work_struct
*rebind_work
= &worker
->rebind_work
;
1745 struct workqueue_struct
*wq
;
1747 /* morph UNBOUND to REBIND */
1748 worker
->flags
&= ~WORKER_UNBOUND
;
1749 worker
->flags
|= WORKER_REBIND
;
1751 if (test_and_set_bit(WORK_STRUCT_PENDING_BIT
,
1752 work_data_bits(rebind_work
)))
1755 debug_work_activate(rebind_work
);
1758 * wq doesn't really matter but let's keep @worker->pool
1759 * and @cwq->pool consistent for sanity.
1761 if (worker_pool_pri(worker
->pool
))
1762 wq
= system_highpri_wq
;
1766 insert_work(get_cwq(gcwq
->cpu
, wq
), rebind_work
,
1767 worker
->scheduled
.next
,
1768 work_color_to_flags(WORK_NO_COLOR
));
1772 static struct worker
*alloc_worker(void)
1774 struct worker
*worker
;
1776 worker
= kzalloc(sizeof(*worker
), GFP_KERNEL
);
1778 INIT_LIST_HEAD(&worker
->entry
);
1779 INIT_LIST_HEAD(&worker
->scheduled
);
1780 INIT_WORK(&worker
->rebind_work
, busy_worker_rebind_fn
);
1781 /* on creation a worker is in !idle && prep state */
1782 worker
->flags
= WORKER_PREP
;
1788 * create_worker - create a new workqueue worker
1789 * @pool: pool the new worker will belong to
1791 * Create a new worker which is bound to @pool. The returned worker
1792 * can be started by calling start_worker() or destroyed using
1796 * Might sleep. Does GFP_KERNEL allocations.
1799 * Pointer to the newly created worker.
1801 static struct worker
*create_worker(struct worker_pool
*pool
)
1803 struct global_cwq
*gcwq
= pool
->gcwq
;
1804 const char *pri
= worker_pool_pri(pool
) ? "H" : "";
1805 struct worker
*worker
= NULL
;
1808 spin_lock_irq(&gcwq
->lock
);
1809 while (ida_get_new(&pool
->worker_ida
, &id
)) {
1810 spin_unlock_irq(&gcwq
->lock
);
1811 if (!ida_pre_get(&pool
->worker_ida
, GFP_KERNEL
))
1813 spin_lock_irq(&gcwq
->lock
);
1815 spin_unlock_irq(&gcwq
->lock
);
1817 worker
= alloc_worker();
1821 worker
->pool
= pool
;
1824 if (gcwq
->cpu
!= WORK_CPU_UNBOUND
)
1825 worker
->task
= kthread_create_on_node(worker_thread
,
1826 worker
, cpu_to_node(gcwq
->cpu
),
1827 "kworker/%u:%d%s", gcwq
->cpu
, id
, pri
);
1829 worker
->task
= kthread_create(worker_thread
, worker
,
1830 "kworker/u:%d%s", id
, pri
);
1831 if (IS_ERR(worker
->task
))
1834 if (worker_pool_pri(pool
))
1835 set_user_nice(worker
->task
, HIGHPRI_NICE_LEVEL
);
1838 * Determine CPU binding of the new worker depending on
1839 * %GCWQ_DISASSOCIATED. The caller is responsible for ensuring the
1840 * flag remains stable across this function. See the comments
1841 * above the flag definition for details.
1843 * As an unbound worker may later become a regular one if CPU comes
1844 * online, make sure every worker has %PF_THREAD_BOUND set.
1846 if (!(gcwq
->flags
& GCWQ_DISASSOCIATED
)) {
1847 kthread_bind(worker
->task
, gcwq
->cpu
);
1849 worker
->task
->flags
|= PF_THREAD_BOUND
;
1850 worker
->flags
|= WORKER_UNBOUND
;
1856 spin_lock_irq(&gcwq
->lock
);
1857 ida_remove(&pool
->worker_ida
, id
);
1858 spin_unlock_irq(&gcwq
->lock
);
1865 * start_worker - start a newly created worker
1866 * @worker: worker to start
1868 * Make the gcwq aware of @worker and start it.
1871 * spin_lock_irq(gcwq->lock).
1873 static void start_worker(struct worker
*worker
)
1875 worker
->flags
|= WORKER_STARTED
;
1876 worker
->pool
->nr_workers
++;
1877 worker_enter_idle(worker
);
1878 wake_up_process(worker
->task
);
1882 * destroy_worker - destroy a workqueue worker
1883 * @worker: worker to be destroyed
1885 * Destroy @worker and adjust @gcwq stats accordingly.
1888 * spin_lock_irq(gcwq->lock) which is released and regrabbed.
1890 static void destroy_worker(struct worker
*worker
)
1892 struct worker_pool
*pool
= worker
->pool
;
1893 struct global_cwq
*gcwq
= pool
->gcwq
;
1894 int id
= worker
->id
;
1896 /* sanity check frenzy */
1897 BUG_ON(worker
->current_work
);
1898 BUG_ON(!list_empty(&worker
->scheduled
));
1900 if (worker
->flags
& WORKER_STARTED
)
1902 if (worker
->flags
& WORKER_IDLE
)
1905 list_del_init(&worker
->entry
);
1906 worker
->flags
|= WORKER_DIE
;
1908 spin_unlock_irq(&gcwq
->lock
);
1910 kthread_stop(worker
->task
);
1913 spin_lock_irq(&gcwq
->lock
);
1914 ida_remove(&pool
->worker_ida
, id
);
1917 static void idle_worker_timeout(unsigned long __pool
)
1919 struct worker_pool
*pool
= (void *)__pool
;
1920 struct global_cwq
*gcwq
= pool
->gcwq
;
1922 spin_lock_irq(&gcwq
->lock
);
1924 if (too_many_workers(pool
)) {
1925 struct worker
*worker
;
1926 unsigned long expires
;
1928 /* idle_list is kept in LIFO order, check the last one */
1929 worker
= list_entry(pool
->idle_list
.prev
, struct worker
, entry
);
1930 expires
= worker
->last_active
+ IDLE_WORKER_TIMEOUT
;
1932 if (time_before(jiffies
, expires
))
1933 mod_timer(&pool
->idle_timer
, expires
);
1935 /* it's been idle for too long, wake up manager */
1936 pool
->flags
|= POOL_MANAGE_WORKERS
;
1937 wake_up_worker(pool
);
1941 spin_unlock_irq(&gcwq
->lock
);
1944 static bool send_mayday(struct work_struct
*work
)
1946 struct cpu_workqueue_struct
*cwq
= get_work_cwq(work
);
1947 struct workqueue_struct
*wq
= cwq
->wq
;
1950 if (!(wq
->flags
& WQ_RESCUER
))
1953 /* mayday mayday mayday */
1954 cpu
= cwq
->pool
->gcwq
->cpu
;
1955 /* WORK_CPU_UNBOUND can't be set in cpumask, use cpu 0 instead */
1956 if (cpu
== WORK_CPU_UNBOUND
)
1958 if (!mayday_test_and_set_cpu(cpu
, wq
->mayday_mask
))
1959 wake_up_process(wq
->rescuer
->task
);
1963 static void gcwq_mayday_timeout(unsigned long __pool
)
1965 struct worker_pool
*pool
= (void *)__pool
;
1966 struct global_cwq
*gcwq
= pool
->gcwq
;
1967 struct work_struct
*work
;
1969 spin_lock_irq(&gcwq
->lock
);
1971 if (need_to_create_worker(pool
)) {
1973 * We've been trying to create a new worker but
1974 * haven't been successful. We might be hitting an
1975 * allocation deadlock. Send distress signals to
1978 list_for_each_entry(work
, &pool
->worklist
, entry
)
1982 spin_unlock_irq(&gcwq
->lock
);
1984 mod_timer(&pool
->mayday_timer
, jiffies
+ MAYDAY_INTERVAL
);
1988 * maybe_create_worker - create a new worker if necessary
1989 * @pool: pool to create a new worker for
1991 * Create a new worker for @pool if necessary. @pool is guaranteed to
1992 * have at least one idle worker on return from this function. If
1993 * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is
1994 * sent to all rescuers with works scheduled on @pool to resolve
1995 * possible allocation deadlock.
1997 * On return, need_to_create_worker() is guaranteed to be false and
1998 * may_start_working() true.
2001 * spin_lock_irq(gcwq->lock) which may be released and regrabbed
2002 * multiple times. Does GFP_KERNEL allocations. Called only from
2006 * false if no action was taken and gcwq->lock stayed locked, true
2009 static bool maybe_create_worker(struct worker_pool
*pool
)
2010 __releases(&gcwq
->lock
)
2011 __acquires(&gcwq
->lock
)
2013 struct global_cwq
*gcwq
= pool
->gcwq
;
2015 if (!need_to_create_worker(pool
))
2018 spin_unlock_irq(&gcwq
->lock
);
2020 /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */
2021 mod_timer(&pool
->mayday_timer
, jiffies
+ MAYDAY_INITIAL_TIMEOUT
);
2024 struct worker
*worker
;
2026 worker
= create_worker(pool
);
2028 del_timer_sync(&pool
->mayday_timer
);
2029 spin_lock_irq(&gcwq
->lock
);
2030 start_worker(worker
);
2031 BUG_ON(need_to_create_worker(pool
));
2035 if (!need_to_create_worker(pool
))
2038 __set_current_state(TASK_INTERRUPTIBLE
);
2039 schedule_timeout(CREATE_COOLDOWN
);
2041 if (!need_to_create_worker(pool
))
2045 del_timer_sync(&pool
->mayday_timer
);
2046 spin_lock_irq(&gcwq
->lock
);
2047 if (need_to_create_worker(pool
))
2053 * maybe_destroy_worker - destroy workers which have been idle for a while
2054 * @pool: pool to destroy workers for
2056 * Destroy @pool workers which have been idle for longer than
2057 * IDLE_WORKER_TIMEOUT.
2060 * spin_lock_irq(gcwq->lock) which may be released and regrabbed
2061 * multiple times. Called only from manager.
2064 * false if no action was taken and gcwq->lock stayed locked, true
2067 static bool maybe_destroy_workers(struct worker_pool
*pool
)
2071 while (too_many_workers(pool
)) {
2072 struct worker
*worker
;
2073 unsigned long expires
;
2075 worker
= list_entry(pool
->idle_list
.prev
, struct worker
, entry
);
2076 expires
= worker
->last_active
+ IDLE_WORKER_TIMEOUT
;
2078 if (time_before(jiffies
, expires
)) {
2079 mod_timer(&pool
->idle_timer
, expires
);
2083 destroy_worker(worker
);
2091 * manage_workers - manage worker pool
2094 * Assume the manager role and manage gcwq worker pool @worker belongs
2095 * to. At any given time, there can be only zero or one manager per
2096 * gcwq. The exclusion is handled automatically by this function.
2098 * The caller can safely start processing works on false return. On
2099 * true return, it's guaranteed that need_to_create_worker() is false
2100 * and may_start_working() is true.
2103 * spin_lock_irq(gcwq->lock) which may be released and regrabbed
2104 * multiple times. Does GFP_KERNEL allocations.
2107 * false if no action was taken and gcwq->lock stayed locked, true if
2108 * some action was taken.
2110 static bool manage_workers(struct worker
*worker
)
2112 struct worker_pool
*pool
= worker
->pool
;
2115 if (!mutex_trylock(&pool
->manager_mutex
))
2118 pool
->flags
&= ~POOL_MANAGE_WORKERS
;
2121 * Destroy and then create so that may_start_working() is true
2124 ret
|= maybe_destroy_workers(pool
);
2125 ret
|= maybe_create_worker(pool
);
2127 mutex_unlock(&pool
->manager_mutex
);
2132 * process_one_work - process single work
2134 * @work: work to process
2136 * Process @work. This function contains all the logics necessary to
2137 * process a single work including synchronization against and
2138 * interaction with other workers on the same cpu, queueing and
2139 * flushing. As long as context requirement is met, any worker can
2140 * call this function to process a work.
2143 * spin_lock_irq(gcwq->lock) which is released and regrabbed.
2145 static void process_one_work(struct worker
*worker
, struct work_struct
*work
)
2146 __releases(&gcwq
->lock
)
2147 __acquires(&gcwq
->lock
)
2149 struct cpu_workqueue_struct
*cwq
= get_work_cwq(work
);
2150 struct worker_pool
*pool
= worker
->pool
;
2151 struct global_cwq
*gcwq
= pool
->gcwq
;
2152 struct hlist_head
*bwh
= busy_worker_head(gcwq
, work
);
2153 bool cpu_intensive
= cwq
->wq
->flags
& WQ_CPU_INTENSIVE
;
2154 work_func_t f
= work
->func
;
2156 struct worker
*collision
;
2157 #ifdef CONFIG_LOCKDEP
2159 * It is permissible to free the struct work_struct from
2160 * inside the function that is called from it, this we need to
2161 * take into account for lockdep too. To avoid bogus "held
2162 * lock freed" warnings as well as problems when looking into
2163 * work->lockdep_map, make a copy and use that here.
2165 struct lockdep_map lockdep_map
;
2167 lockdep_copy_map(&lockdep_map
, &work
->lockdep_map
);
2170 * Ensure we're on the correct CPU. DISASSOCIATED test is
2171 * necessary to avoid spurious warnings from rescuers servicing the
2172 * unbound or a disassociated gcwq.
2174 WARN_ON_ONCE(!(worker
->flags
& (WORKER_UNBOUND
| WORKER_REBIND
)) &&
2175 !(gcwq
->flags
& GCWQ_DISASSOCIATED
) &&
2176 raw_smp_processor_id() != gcwq
->cpu
);
2179 * A single work shouldn't be executed concurrently by
2180 * multiple workers on a single cpu. Check whether anyone is
2181 * already processing the work. If so, defer the work to the
2182 * currently executing one.
2184 collision
= __find_worker_executing_work(gcwq
, bwh
, work
);
2185 if (unlikely(collision
)) {
2186 move_linked_works(work
, &collision
->scheduled
, NULL
);
2190 /* claim and dequeue */
2191 debug_work_deactivate(work
);
2192 hlist_add_head(&worker
->hentry
, bwh
);
2193 worker
->current_work
= work
;
2194 worker
->current_cwq
= cwq
;
2195 work_color
= get_work_color(work
);
2197 list_del_init(&work
->entry
);
2200 * CPU intensive works don't participate in concurrency
2201 * management. They're the scheduler's responsibility.
2203 if (unlikely(cpu_intensive
))
2204 worker_set_flags(worker
, WORKER_CPU_INTENSIVE
, true);
2207 * Unbound gcwq isn't concurrency managed and work items should be
2208 * executed ASAP. Wake up another worker if necessary.
2210 if ((worker
->flags
& WORKER_UNBOUND
) && need_more_worker(pool
))
2211 wake_up_worker(pool
);
2214 * Record the last CPU and clear PENDING which should be the last
2215 * update to @work. Also, do this inside @gcwq->lock so that
2216 * PENDING and queued state changes happen together while IRQ is
2219 set_work_cpu_and_clear_pending(work
, gcwq
->cpu
);
2221 spin_unlock_irq(&gcwq
->lock
);
2223 lock_map_acquire_read(&cwq
->wq
->lockdep_map
);
2224 lock_map_acquire(&lockdep_map
);
2225 trace_workqueue_execute_start(work
);
2228 * While we must be careful to not use "work" after this, the trace
2229 * point will only record its address.
2231 trace_workqueue_execute_end(work
);
2232 lock_map_release(&lockdep_map
);
2233 lock_map_release(&cwq
->wq
->lockdep_map
);
2235 if (unlikely(in_atomic() || lockdep_depth(current
) > 0)) {
2236 pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n"
2237 " last function: %pf\n",
2238 current
->comm
, preempt_count(), task_pid_nr(current
), f
);
2239 debug_show_held_locks(current
);
2243 spin_lock_irq(&gcwq
->lock
);
2245 /* clear cpu intensive status */
2246 if (unlikely(cpu_intensive
))
2247 worker_clr_flags(worker
, WORKER_CPU_INTENSIVE
);
2249 /* we're done with it, release */
2250 hlist_del_init(&worker
->hentry
);
2251 worker
->current_work
= NULL
;
2252 worker
->current_cwq
= NULL
;
2253 cwq_dec_nr_in_flight(cwq
, work_color
, false);
2257 * process_scheduled_works - process scheduled works
2260 * Process all scheduled works. Please note that the scheduled list
2261 * may change while processing a work, so this function repeatedly
2262 * fetches a work from the top and executes it.
2265 * spin_lock_irq(gcwq->lock) which may be released and regrabbed
2268 static void process_scheduled_works(struct worker
*worker
)
2270 while (!list_empty(&worker
->scheduled
)) {
2271 struct work_struct
*work
= list_first_entry(&worker
->scheduled
,
2272 struct work_struct
, entry
);
2273 process_one_work(worker
, work
);
2278 * worker_thread - the worker thread function
2281 * The gcwq worker thread function. There's a single dynamic pool of
2282 * these per each cpu. These workers process all works regardless of
2283 * their specific target workqueue. The only exception is works which
2284 * belong to workqueues with a rescuer which will be explained in
2287 static int worker_thread(void *__worker
)
2289 struct worker
*worker
= __worker
;
2290 struct worker_pool
*pool
= worker
->pool
;
2291 struct global_cwq
*gcwq
= pool
->gcwq
;
2293 /* tell the scheduler that this is a workqueue worker */
2294 worker
->task
->flags
|= PF_WQ_WORKER
;
2296 spin_lock_irq(&gcwq
->lock
);
2299 * DIE can be set only while idle and REBIND set while busy has
2300 * @worker->rebind_work scheduled. Checking here is enough.
2302 if (unlikely(worker
->flags
& (WORKER_REBIND
| WORKER_DIE
))) {
2303 spin_unlock_irq(&gcwq
->lock
);
2305 if (worker
->flags
& WORKER_DIE
) {
2306 worker
->task
->flags
&= ~PF_WQ_WORKER
;
2310 idle_worker_rebind(worker
);
2314 worker_leave_idle(worker
);
2316 /* no more worker necessary? */
2317 if (!need_more_worker(pool
))
2320 /* do we need to manage? */
2321 if (unlikely(!may_start_working(pool
)) && manage_workers(worker
))
2325 * ->scheduled list can only be filled while a worker is
2326 * preparing to process a work or actually processing it.
2327 * Make sure nobody diddled with it while I was sleeping.
2329 BUG_ON(!list_empty(&worker
->scheduled
));
2332 * When control reaches this point, we're guaranteed to have
2333 * at least one idle worker or that someone else has already
2334 * assumed the manager role.
2336 worker_clr_flags(worker
, WORKER_PREP
);
2339 struct work_struct
*work
=
2340 list_first_entry(&pool
->worklist
,
2341 struct work_struct
, entry
);
2343 if (likely(!(*work_data_bits(work
) & WORK_STRUCT_LINKED
))) {
2344 /* optimization path, not strictly necessary */
2345 process_one_work(worker
, work
);
2346 if (unlikely(!list_empty(&worker
->scheduled
)))
2347 process_scheduled_works(worker
);
2349 move_linked_works(work
, &worker
->scheduled
, NULL
);
2350 process_scheduled_works(worker
);
2352 } while (keep_working(pool
));
2354 worker_set_flags(worker
, WORKER_PREP
, false);
2356 if (unlikely(need_to_manage_workers(pool
)) && manage_workers(worker
))
2360 * gcwq->lock is held and there's no work to process and no
2361 * need to manage, sleep. Workers are woken up only while
2362 * holding gcwq->lock or from local cpu, so setting the
2363 * current state before releasing gcwq->lock is enough to
2364 * prevent losing any event.
2366 worker_enter_idle(worker
);
2367 __set_current_state(TASK_INTERRUPTIBLE
);
2368 spin_unlock_irq(&gcwq
->lock
);
2374 * rescuer_thread - the rescuer thread function
2375 * @__wq: the associated workqueue
2377 * Workqueue rescuer thread function. There's one rescuer for each
2378 * workqueue which has WQ_RESCUER set.
2380 * Regular work processing on a gcwq may block trying to create a new
2381 * worker which uses GFP_KERNEL allocation which has slight chance of
2382 * developing into deadlock if some works currently on the same queue
2383 * need to be processed to satisfy the GFP_KERNEL allocation. This is
2384 * the problem rescuer solves.
2386 * When such condition is possible, the gcwq summons rescuers of all
2387 * workqueues which have works queued on the gcwq and let them process
2388 * those works so that forward progress can be guaranteed.
2390 * This should happen rarely.
2392 static int rescuer_thread(void *__wq
)
2394 struct workqueue_struct
*wq
= __wq
;
2395 struct worker
*rescuer
= wq
->rescuer
;
2396 struct list_head
*scheduled
= &rescuer
->scheduled
;
2397 bool is_unbound
= wq
->flags
& WQ_UNBOUND
;
2400 set_user_nice(current
, RESCUER_NICE_LEVEL
);
2402 set_current_state(TASK_INTERRUPTIBLE
);
2404 if (kthread_should_stop())
2408 * See whether any cpu is asking for help. Unbounded
2409 * workqueues use cpu 0 in mayday_mask for CPU_UNBOUND.
2411 for_each_mayday_cpu(cpu
, wq
->mayday_mask
) {
2412 unsigned int tcpu
= is_unbound
? WORK_CPU_UNBOUND
: cpu
;
2413 struct cpu_workqueue_struct
*cwq
= get_cwq(tcpu
, wq
);
2414 struct worker_pool
*pool
= cwq
->pool
;
2415 struct global_cwq
*gcwq
= pool
->gcwq
;
2416 struct work_struct
*work
, *n
;
2418 __set_current_state(TASK_RUNNING
);
2419 mayday_clear_cpu(cpu
, wq
->mayday_mask
);
2421 /* migrate to the target cpu if possible */
2422 rescuer
->pool
= pool
;
2423 worker_maybe_bind_and_lock(rescuer
);
2426 * Slurp in all works issued via this workqueue and
2429 BUG_ON(!list_empty(&rescuer
->scheduled
));
2430 list_for_each_entry_safe(work
, n
, &pool
->worklist
, entry
)
2431 if (get_work_cwq(work
) == cwq
)
2432 move_linked_works(work
, scheduled
, &n
);
2434 process_scheduled_works(rescuer
);
2437 * Leave this gcwq. If keep_working() is %true, notify a
2438 * regular worker; otherwise, we end up with 0 concurrency
2439 * and stalling the execution.
2441 if (keep_working(pool
))
2442 wake_up_worker(pool
);
2444 spin_unlock_irq(&gcwq
->lock
);
2452 struct work_struct work
;
2453 struct completion done
;
2456 static void wq_barrier_func(struct work_struct
*work
)
2458 struct wq_barrier
*barr
= container_of(work
, struct wq_barrier
, work
);
2459 complete(&barr
->done
);
2463 * insert_wq_barrier - insert a barrier work
2464 * @cwq: cwq to insert barrier into
2465 * @barr: wq_barrier to insert
2466 * @target: target work to attach @barr to
2467 * @worker: worker currently executing @target, NULL if @target is not executing
2469 * @barr is linked to @target such that @barr is completed only after
2470 * @target finishes execution. Please note that the ordering
2471 * guarantee is observed only with respect to @target and on the local
2474 * Currently, a queued barrier can't be canceled. This is because
2475 * try_to_grab_pending() can't determine whether the work to be
2476 * grabbed is at the head of the queue and thus can't clear LINKED
2477 * flag of the previous work while there must be a valid next work
2478 * after a work with LINKED flag set.
2480 * Note that when @worker is non-NULL, @target may be modified
2481 * underneath us, so we can't reliably determine cwq from @target.
2484 * spin_lock_irq(gcwq->lock).
2486 static void insert_wq_barrier(struct cpu_workqueue_struct
*cwq
,
2487 struct wq_barrier
*barr
,
2488 struct work_struct
*target
, struct worker
*worker
)
2490 struct list_head
*head
;
2491 unsigned int linked
= 0;
2494 * debugobject calls are safe here even with gcwq->lock locked
2495 * as we know for sure that this will not trigger any of the
2496 * checks and call back into the fixup functions where we
2499 INIT_WORK_ONSTACK(&barr
->work
, wq_barrier_func
);
2500 __set_bit(WORK_STRUCT_PENDING_BIT
, work_data_bits(&barr
->work
));
2501 init_completion(&barr
->done
);
2504 * If @target is currently being executed, schedule the
2505 * barrier to the worker; otherwise, put it after @target.
2508 head
= worker
->scheduled
.next
;
2510 unsigned long *bits
= work_data_bits(target
);
2512 head
= target
->entry
.next
;
2513 /* there can already be other linked works, inherit and set */
2514 linked
= *bits
& WORK_STRUCT_LINKED
;
2515 __set_bit(WORK_STRUCT_LINKED_BIT
, bits
);
2518 debug_work_activate(&barr
->work
);
2519 insert_work(cwq
, &barr
->work
, head
,
2520 work_color_to_flags(WORK_NO_COLOR
) | linked
);
2524 * flush_workqueue_prep_cwqs - prepare cwqs for workqueue flushing
2525 * @wq: workqueue being flushed
2526 * @flush_color: new flush color, < 0 for no-op
2527 * @work_color: new work color, < 0 for no-op
2529 * Prepare cwqs for workqueue flushing.
2531 * If @flush_color is non-negative, flush_color on all cwqs should be
2532 * -1. If no cwq has in-flight commands at the specified color, all
2533 * cwq->flush_color's stay at -1 and %false is returned. If any cwq
2534 * has in flight commands, its cwq->flush_color is set to
2535 * @flush_color, @wq->nr_cwqs_to_flush is updated accordingly, cwq
2536 * wakeup logic is armed and %true is returned.
2538 * The caller should have initialized @wq->first_flusher prior to
2539 * calling this function with non-negative @flush_color. If
2540 * @flush_color is negative, no flush color update is done and %false
2543 * If @work_color is non-negative, all cwqs should have the same
2544 * work_color which is previous to @work_color and all will be
2545 * advanced to @work_color.
2548 * mutex_lock(wq->flush_mutex).
2551 * %true if @flush_color >= 0 and there's something to flush. %false
2554 static bool flush_workqueue_prep_cwqs(struct workqueue_struct
*wq
,
2555 int flush_color
, int work_color
)
2560 if (flush_color
>= 0) {
2561 BUG_ON(atomic_read(&wq
->nr_cwqs_to_flush
));
2562 atomic_set(&wq
->nr_cwqs_to_flush
, 1);
2565 for_each_cwq_cpu(cpu
, wq
) {
2566 struct cpu_workqueue_struct
*cwq
= get_cwq(cpu
, wq
);
2567 struct global_cwq
*gcwq
= cwq
->pool
->gcwq
;
2569 spin_lock_irq(&gcwq
->lock
);
2571 if (flush_color
>= 0) {
2572 BUG_ON(cwq
->flush_color
!= -1);
2574 if (cwq
->nr_in_flight
[flush_color
]) {
2575 cwq
->flush_color
= flush_color
;
2576 atomic_inc(&wq
->nr_cwqs_to_flush
);
2581 if (work_color
>= 0) {
2582 BUG_ON(work_color
!= work_next_color(cwq
->work_color
));
2583 cwq
->work_color
= work_color
;
2586 spin_unlock_irq(&gcwq
->lock
);
2589 if (flush_color
>= 0 && atomic_dec_and_test(&wq
->nr_cwqs_to_flush
))
2590 complete(&wq
->first_flusher
->done
);
2596 * flush_workqueue - ensure that any scheduled work has run to completion.
2597 * @wq: workqueue to flush
2599 * Forces execution of the workqueue and blocks until its completion.
2600 * This is typically used in driver shutdown handlers.
2602 * We sleep until all works which were queued on entry have been handled,
2603 * but we are not livelocked by new incoming ones.
2605 void flush_workqueue(struct workqueue_struct
*wq
)
2607 struct wq_flusher this_flusher
= {
2608 .list
= LIST_HEAD_INIT(this_flusher
.list
),
2610 .done
= COMPLETION_INITIALIZER_ONSTACK(this_flusher
.done
),
2614 lock_map_acquire(&wq
->lockdep_map
);
2615 lock_map_release(&wq
->lockdep_map
);
2617 mutex_lock(&wq
->flush_mutex
);
2620 * Start-to-wait phase
2622 next_color
= work_next_color(wq
->work_color
);
2624 if (next_color
!= wq
->flush_color
) {
2626 * Color space is not full. The current work_color
2627 * becomes our flush_color and work_color is advanced
2630 BUG_ON(!list_empty(&wq
->flusher_overflow
));
2631 this_flusher
.flush_color
= wq
->work_color
;
2632 wq
->work_color
= next_color
;
2634 if (!wq
->first_flusher
) {
2635 /* no flush in progress, become the first flusher */
2636 BUG_ON(wq
->flush_color
!= this_flusher
.flush_color
);
2638 wq
->first_flusher
= &this_flusher
;
2640 if (!flush_workqueue_prep_cwqs(wq
, wq
->flush_color
,
2642 /* nothing to flush, done */
2643 wq
->flush_color
= next_color
;
2644 wq
->first_flusher
= NULL
;
2649 BUG_ON(wq
->flush_color
== this_flusher
.flush_color
);
2650 list_add_tail(&this_flusher
.list
, &wq
->flusher_queue
);
2651 flush_workqueue_prep_cwqs(wq
, -1, wq
->work_color
);
2655 * Oops, color space is full, wait on overflow queue.
2656 * The next flush completion will assign us
2657 * flush_color and transfer to flusher_queue.
2659 list_add_tail(&this_flusher
.list
, &wq
->flusher_overflow
);
2662 mutex_unlock(&wq
->flush_mutex
);
2664 wait_for_completion(&this_flusher
.done
);
2667 * Wake-up-and-cascade phase
2669 * First flushers are responsible for cascading flushes and
2670 * handling overflow. Non-first flushers can simply return.
2672 if (wq
->first_flusher
!= &this_flusher
)
2675 mutex_lock(&wq
->flush_mutex
);
2677 /* we might have raced, check again with mutex held */
2678 if (wq
->first_flusher
!= &this_flusher
)
2681 wq
->first_flusher
= NULL
;
2683 BUG_ON(!list_empty(&this_flusher
.list
));
2684 BUG_ON(wq
->flush_color
!= this_flusher
.flush_color
);
2687 struct wq_flusher
*next
, *tmp
;
2689 /* complete all the flushers sharing the current flush color */
2690 list_for_each_entry_safe(next
, tmp
, &wq
->flusher_queue
, list
) {
2691 if (next
->flush_color
!= wq
->flush_color
)
2693 list_del_init(&next
->list
);
2694 complete(&next
->done
);
2697 BUG_ON(!list_empty(&wq
->flusher_overflow
) &&
2698 wq
->flush_color
!= work_next_color(wq
->work_color
));
2700 /* this flush_color is finished, advance by one */
2701 wq
->flush_color
= work_next_color(wq
->flush_color
);
2703 /* one color has been freed, handle overflow queue */
2704 if (!list_empty(&wq
->flusher_overflow
)) {
2706 * Assign the same color to all overflowed
2707 * flushers, advance work_color and append to
2708 * flusher_queue. This is the start-to-wait
2709 * phase for these overflowed flushers.
2711 list_for_each_entry(tmp
, &wq
->flusher_overflow
, list
)
2712 tmp
->flush_color
= wq
->work_color
;
2714 wq
->work_color
= work_next_color(wq
->work_color
);
2716 list_splice_tail_init(&wq
->flusher_overflow
,
2717 &wq
->flusher_queue
);
2718 flush_workqueue_prep_cwqs(wq
, -1, wq
->work_color
);
2721 if (list_empty(&wq
->flusher_queue
)) {
2722 BUG_ON(wq
->flush_color
!= wq
->work_color
);
2727 * Need to flush more colors. Make the next flusher
2728 * the new first flusher and arm cwqs.
2730 BUG_ON(wq
->flush_color
== wq
->work_color
);
2731 BUG_ON(wq
->flush_color
!= next
->flush_color
);
2733 list_del_init(&next
->list
);
2734 wq
->first_flusher
= next
;
2736 if (flush_workqueue_prep_cwqs(wq
, wq
->flush_color
, -1))
2740 * Meh... this color is already done, clear first
2741 * flusher and repeat cascading.
2743 wq
->first_flusher
= NULL
;
2747 mutex_unlock(&wq
->flush_mutex
);
2749 EXPORT_SYMBOL_GPL(flush_workqueue
);
2752 * drain_workqueue - drain a workqueue
2753 * @wq: workqueue to drain
2755 * Wait until the workqueue becomes empty. While draining is in progress,
2756 * only chain queueing is allowed. IOW, only currently pending or running
2757 * work items on @wq can queue further work items on it. @wq is flushed
2758 * repeatedly until it becomes empty. The number of flushing is detemined
2759 * by the depth of chaining and should be relatively short. Whine if it
2762 void drain_workqueue(struct workqueue_struct
*wq
)
2764 unsigned int flush_cnt
= 0;
2768 * __queue_work() needs to test whether there are drainers, is much
2769 * hotter than drain_workqueue() and already looks at @wq->flags.
2770 * Use WQ_DRAINING so that queue doesn't have to check nr_drainers.
2772 spin_lock(&workqueue_lock
);
2773 if (!wq
->nr_drainers
++)
2774 wq
->flags
|= WQ_DRAINING
;
2775 spin_unlock(&workqueue_lock
);
2777 flush_workqueue(wq
);
2779 for_each_cwq_cpu(cpu
, wq
) {
2780 struct cpu_workqueue_struct
*cwq
= get_cwq(cpu
, wq
);
2783 spin_lock_irq(&cwq
->pool
->gcwq
->lock
);
2784 drained
= !cwq
->nr_active
&& list_empty(&cwq
->delayed_works
);
2785 spin_unlock_irq(&cwq
->pool
->gcwq
->lock
);
2790 if (++flush_cnt
== 10 ||
2791 (flush_cnt
% 100 == 0 && flush_cnt
<= 1000))
2792 pr_warn("workqueue %s: flush on destruction isn't complete after %u tries\n",
2793 wq
->name
, flush_cnt
);
2797 spin_lock(&workqueue_lock
);
2798 if (!--wq
->nr_drainers
)
2799 wq
->flags
&= ~WQ_DRAINING
;
2800 spin_unlock(&workqueue_lock
);
2802 EXPORT_SYMBOL_GPL(drain_workqueue
);
2804 static bool start_flush_work(struct work_struct
*work
, struct wq_barrier
*barr
)
2806 struct worker
*worker
= NULL
;
2807 struct global_cwq
*gcwq
;
2808 struct cpu_workqueue_struct
*cwq
;
2811 gcwq
= get_work_gcwq(work
);
2815 spin_lock_irq(&gcwq
->lock
);
2816 if (!list_empty(&work
->entry
)) {
2818 * See the comment near try_to_grab_pending()->smp_rmb().
2819 * If it was re-queued to a different gcwq under us, we
2820 * are not going to wait.
2823 cwq
= get_work_cwq(work
);
2824 if (unlikely(!cwq
|| gcwq
!= cwq
->pool
->gcwq
))
2827 worker
= find_worker_executing_work(gcwq
, work
);
2830 cwq
= worker
->current_cwq
;
2833 insert_wq_barrier(cwq
, barr
, work
, worker
);
2834 spin_unlock_irq(&gcwq
->lock
);
2837 * If @max_active is 1 or rescuer is in use, flushing another work
2838 * item on the same workqueue may lead to deadlock. Make sure the
2839 * flusher is not running on the same workqueue by verifying write
2842 if (cwq
->wq
->saved_max_active
== 1 || cwq
->wq
->flags
& WQ_RESCUER
)
2843 lock_map_acquire(&cwq
->wq
->lockdep_map
);
2845 lock_map_acquire_read(&cwq
->wq
->lockdep_map
);
2846 lock_map_release(&cwq
->wq
->lockdep_map
);
2850 spin_unlock_irq(&gcwq
->lock
);
2855 * flush_work - wait for a work to finish executing the last queueing instance
2856 * @work: the work to flush
2858 * Wait until @work has finished execution. @work is guaranteed to be idle
2859 * on return if it hasn't been requeued since flush started.
2862 * %true if flush_work() waited for the work to finish execution,
2863 * %false if it was already idle.
2865 bool flush_work(struct work_struct
*work
)
2867 struct wq_barrier barr
;
2869 lock_map_acquire(&work
->lockdep_map
);
2870 lock_map_release(&work
->lockdep_map
);
2872 if (start_flush_work(work
, &barr
)) {
2873 wait_for_completion(&barr
.done
);
2874 destroy_work_on_stack(&barr
.work
);
2880 EXPORT_SYMBOL_GPL(flush_work
);
2882 static bool __cancel_work_timer(struct work_struct
*work
, bool is_dwork
)
2884 unsigned long flags
;
2888 ret
= try_to_grab_pending(work
, is_dwork
, &flags
);
2890 * If someone else is canceling, wait for the same event it
2891 * would be waiting for before retrying.
2893 if (unlikely(ret
== -ENOENT
))
2895 } while (unlikely(ret
< 0));
2897 /* tell other tasks trying to grab @work to back off */
2898 mark_work_canceling(work
);
2899 local_irq_restore(flags
);
2902 clear_work_data(work
);
2907 * cancel_work_sync - cancel a work and wait for it to finish
2908 * @work: the work to cancel
2910 * Cancel @work and wait for its execution to finish. This function
2911 * can be used even if the work re-queues itself or migrates to
2912 * another workqueue. On return from this function, @work is
2913 * guaranteed to be not pending or executing on any CPU.
2915 * cancel_work_sync(&delayed_work->work) must not be used for
2916 * delayed_work's. Use cancel_delayed_work_sync() instead.
2918 * The caller must ensure that the workqueue on which @work was last
2919 * queued can't be destroyed before this function returns.
2922 * %true if @work was pending, %false otherwise.
2924 bool cancel_work_sync(struct work_struct
*work
)
2926 return __cancel_work_timer(work
, false);
2928 EXPORT_SYMBOL_GPL(cancel_work_sync
);
2931 * flush_delayed_work - wait for a dwork to finish executing the last queueing
2932 * @dwork: the delayed work to flush
2934 * Delayed timer is cancelled and the pending work is queued for
2935 * immediate execution. Like flush_work(), this function only
2936 * considers the last queueing instance of @dwork.
2939 * %true if flush_work() waited for the work to finish execution,
2940 * %false if it was already idle.
2942 bool flush_delayed_work(struct delayed_work
*dwork
)
2944 local_irq_disable();
2945 if (del_timer_sync(&dwork
->timer
))
2946 __queue_work(dwork
->cpu
,
2947 get_work_cwq(&dwork
->work
)->wq
, &dwork
->work
);
2949 return flush_work(&dwork
->work
);
2951 EXPORT_SYMBOL(flush_delayed_work
);
2954 * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish
2955 * @dwork: the delayed work cancel
2957 * This is cancel_work_sync() for delayed works.
2960 * %true if @dwork was pending, %false otherwise.
2962 bool cancel_delayed_work_sync(struct delayed_work
*dwork
)
2964 return __cancel_work_timer(&dwork
->work
, true);
2966 EXPORT_SYMBOL(cancel_delayed_work_sync
);
2969 * schedule_work_on - put work task on a specific cpu
2970 * @cpu: cpu to put the work task on
2971 * @work: job to be done
2973 * This puts a job on a specific cpu
2975 bool schedule_work_on(int cpu
, struct work_struct
*work
)
2977 return queue_work_on(cpu
, system_wq
, work
);
2979 EXPORT_SYMBOL(schedule_work_on
);
2982 * schedule_work - put work task in global workqueue
2983 * @work: job to be done
2985 * Returns %false if @work was already on the kernel-global workqueue and
2988 * This puts a job in the kernel-global workqueue if it was not already
2989 * queued and leaves it in the same position on the kernel-global
2990 * workqueue otherwise.
2992 bool schedule_work(struct work_struct
*work
)
2994 return queue_work(system_wq
, work
);
2996 EXPORT_SYMBOL(schedule_work
);
2999 * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
3001 * @dwork: job to be done
3002 * @delay: number of jiffies to wait
3004 * After waiting for a given time this puts a job in the kernel-global
3005 * workqueue on the specified CPU.
3007 bool schedule_delayed_work_on(int cpu
, struct delayed_work
*dwork
,
3008 unsigned long delay
)
3010 return queue_delayed_work_on(cpu
, system_wq
, dwork
, delay
);
3012 EXPORT_SYMBOL(schedule_delayed_work_on
);
3015 * schedule_delayed_work - put work task in global workqueue after delay
3016 * @dwork: job to be done
3017 * @delay: number of jiffies to wait or 0 for immediate execution
3019 * After waiting for a given time this puts a job in the kernel-global
3022 bool schedule_delayed_work(struct delayed_work
*dwork
, unsigned long delay
)
3024 return queue_delayed_work(system_wq
, dwork
, delay
);
3026 EXPORT_SYMBOL(schedule_delayed_work
);
3029 * schedule_on_each_cpu - execute a function synchronously on each online CPU
3030 * @func: the function to call
3032 * schedule_on_each_cpu() executes @func on each online CPU using the
3033 * system workqueue and blocks until all CPUs have completed.
3034 * schedule_on_each_cpu() is very slow.
3037 * 0 on success, -errno on failure.
3039 int schedule_on_each_cpu(work_func_t func
)
3042 struct work_struct __percpu
*works
;
3044 works
= alloc_percpu(struct work_struct
);
3050 for_each_online_cpu(cpu
) {
3051 struct work_struct
*work
= per_cpu_ptr(works
, cpu
);
3053 INIT_WORK(work
, func
);
3054 schedule_work_on(cpu
, work
);
3057 for_each_online_cpu(cpu
)
3058 flush_work(per_cpu_ptr(works
, cpu
));
3066 * flush_scheduled_work - ensure that any scheduled work has run to completion.
3068 * Forces execution of the kernel-global workqueue and blocks until its
3071 * Think twice before calling this function! It's very easy to get into
3072 * trouble if you don't take great care. Either of the following situations
3073 * will lead to deadlock:
3075 * One of the work items currently on the workqueue needs to acquire
3076 * a lock held by your code or its caller.
3078 * Your code is running in the context of a work routine.
3080 * They will be detected by lockdep when they occur, but the first might not
3081 * occur very often. It depends on what work items are on the workqueue and
3082 * what locks they need, which you have no control over.
3084 * In most situations flushing the entire workqueue is overkill; you merely
3085 * need to know that a particular work item isn't queued and isn't running.
3086 * In such cases you should use cancel_delayed_work_sync() or
3087 * cancel_work_sync() instead.
3089 void flush_scheduled_work(void)
3091 flush_workqueue(system_wq
);
3093 EXPORT_SYMBOL(flush_scheduled_work
);
3096 * execute_in_process_context - reliably execute the routine with user context
3097 * @fn: the function to execute
3098 * @ew: guaranteed storage for the execute work structure (must
3099 * be available when the work executes)
3101 * Executes the function immediately if process context is available,
3102 * otherwise schedules the function for delayed execution.
3104 * Returns: 0 - function was executed
3105 * 1 - function was scheduled for execution
3107 int execute_in_process_context(work_func_t fn
, struct execute_work
*ew
)
3109 if (!in_interrupt()) {
3114 INIT_WORK(&ew
->work
, fn
);
3115 schedule_work(&ew
->work
);
3119 EXPORT_SYMBOL_GPL(execute_in_process_context
);
3121 int keventd_up(void)
3123 return system_wq
!= NULL
;
3126 static int alloc_cwqs(struct workqueue_struct
*wq
)
3129 * cwqs are forced aligned according to WORK_STRUCT_FLAG_BITS.
3130 * Make sure that the alignment isn't lower than that of
3131 * unsigned long long.
3133 const size_t size
= sizeof(struct cpu_workqueue_struct
);
3134 const size_t align
= max_t(size_t, 1 << WORK_STRUCT_FLAG_BITS
,
3135 __alignof__(unsigned long long));
3137 if (!(wq
->flags
& WQ_UNBOUND
))
3138 wq
->cpu_wq
.pcpu
= __alloc_percpu(size
, align
);
3143 * Allocate enough room to align cwq and put an extra
3144 * pointer at the end pointing back to the originally
3145 * allocated pointer which will be used for free.
3147 ptr
= kzalloc(size
+ align
+ sizeof(void *), GFP_KERNEL
);
3149 wq
->cpu_wq
.single
= PTR_ALIGN(ptr
, align
);
3150 *(void **)(wq
->cpu_wq
.single
+ 1) = ptr
;
3154 /* just in case, make sure it's actually aligned */
3155 BUG_ON(!IS_ALIGNED(wq
->cpu_wq
.v
, align
));
3156 return wq
->cpu_wq
.v
? 0 : -ENOMEM
;
3159 static void free_cwqs(struct workqueue_struct
*wq
)
3161 if (!(wq
->flags
& WQ_UNBOUND
))
3162 free_percpu(wq
->cpu_wq
.pcpu
);
3163 else if (wq
->cpu_wq
.single
) {
3164 /* the pointer to free is stored right after the cwq */
3165 kfree(*(void **)(wq
->cpu_wq
.single
+ 1));
3169 static int wq_clamp_max_active(int max_active
, unsigned int flags
,
3172 int lim
= flags
& WQ_UNBOUND
? WQ_UNBOUND_MAX_ACTIVE
: WQ_MAX_ACTIVE
;
3174 if (max_active
< 1 || max_active
> lim
)
3175 pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n",
3176 max_active
, name
, 1, lim
);
3178 return clamp_val(max_active
, 1, lim
);
3181 struct workqueue_struct
*__alloc_workqueue_key(const char *fmt
,
3184 struct lock_class_key
*key
,
3185 const char *lock_name
, ...)
3187 va_list args
, args1
;
3188 struct workqueue_struct
*wq
;
3192 /* determine namelen, allocate wq and format name */
3193 va_start(args
, lock_name
);
3194 va_copy(args1
, args
);
3195 namelen
= vsnprintf(NULL
, 0, fmt
, args
) + 1;
3197 wq
= kzalloc(sizeof(*wq
) + namelen
, GFP_KERNEL
);
3201 vsnprintf(wq
->name
, namelen
, fmt
, args1
);
3206 * Workqueues which may be used during memory reclaim should
3207 * have a rescuer to guarantee forward progress.
3209 if (flags
& WQ_MEM_RECLAIM
)
3210 flags
|= WQ_RESCUER
;
3212 max_active
= max_active
?: WQ_DFL_ACTIVE
;
3213 max_active
= wq_clamp_max_active(max_active
, flags
, wq
->name
);
3217 wq
->saved_max_active
= max_active
;
3218 mutex_init(&wq
->flush_mutex
);
3219 atomic_set(&wq
->nr_cwqs_to_flush
, 0);
3220 INIT_LIST_HEAD(&wq
->flusher_queue
);
3221 INIT_LIST_HEAD(&wq
->flusher_overflow
);
3223 lockdep_init_map(&wq
->lockdep_map
, lock_name
, key
, 0);
3224 INIT_LIST_HEAD(&wq
->list
);
3226 if (alloc_cwqs(wq
) < 0)
3229 for_each_cwq_cpu(cpu
, wq
) {
3230 struct cpu_workqueue_struct
*cwq
= get_cwq(cpu
, wq
);
3231 struct global_cwq
*gcwq
= get_gcwq(cpu
);
3232 int pool_idx
= (bool)(flags
& WQ_HIGHPRI
);
3234 BUG_ON((unsigned long)cwq
& WORK_STRUCT_FLAG_MASK
);
3235 cwq
->pool
= &gcwq
->pools
[pool_idx
];
3237 cwq
->flush_color
= -1;
3238 cwq
->max_active
= max_active
;
3239 INIT_LIST_HEAD(&cwq
->delayed_works
);
3242 if (flags
& WQ_RESCUER
) {
3243 struct worker
*rescuer
;
3245 if (!alloc_mayday_mask(&wq
->mayday_mask
, GFP_KERNEL
))
3248 wq
->rescuer
= rescuer
= alloc_worker();
3252 rescuer
->task
= kthread_create(rescuer_thread
, wq
, "%s",
3254 if (IS_ERR(rescuer
->task
))
3257 rescuer
->task
->flags
|= PF_THREAD_BOUND
;
3258 wake_up_process(rescuer
->task
);
3262 * workqueue_lock protects global freeze state and workqueues
3263 * list. Grab it, set max_active accordingly and add the new
3264 * workqueue to workqueues list.
3266 spin_lock(&workqueue_lock
);
3268 if (workqueue_freezing
&& wq
->flags
& WQ_FREEZABLE
)
3269 for_each_cwq_cpu(cpu
, wq
)
3270 get_cwq(cpu
, wq
)->max_active
= 0;
3272 list_add(&wq
->list
, &workqueues
);
3274 spin_unlock(&workqueue_lock
);
3280 free_mayday_mask(wq
->mayday_mask
);
3286 EXPORT_SYMBOL_GPL(__alloc_workqueue_key
);
3289 * destroy_workqueue - safely terminate a workqueue
3290 * @wq: target workqueue
3292 * Safely destroy a workqueue. All work currently pending will be done first.
3294 void destroy_workqueue(struct workqueue_struct
*wq
)
3298 /* drain it before proceeding with destruction */
3299 drain_workqueue(wq
);
3302 * wq list is used to freeze wq, remove from list after
3303 * flushing is complete in case freeze races us.
3305 spin_lock(&workqueue_lock
);
3306 list_del(&wq
->list
);
3307 spin_unlock(&workqueue_lock
);
3310 for_each_cwq_cpu(cpu
, wq
) {
3311 struct cpu_workqueue_struct
*cwq
= get_cwq(cpu
, wq
);
3314 for (i
= 0; i
< WORK_NR_COLORS
; i
++)
3315 BUG_ON(cwq
->nr_in_flight
[i
]);
3316 BUG_ON(cwq
->nr_active
);
3317 BUG_ON(!list_empty(&cwq
->delayed_works
));
3320 if (wq
->flags
& WQ_RESCUER
) {
3321 kthread_stop(wq
->rescuer
->task
);
3322 free_mayday_mask(wq
->mayday_mask
);
3329 EXPORT_SYMBOL_GPL(destroy_workqueue
);
3332 * workqueue_set_max_active - adjust max_active of a workqueue
3333 * @wq: target workqueue
3334 * @max_active: new max_active value.
3336 * Set max_active of @wq to @max_active.
3339 * Don't call from IRQ context.
3341 void workqueue_set_max_active(struct workqueue_struct
*wq
, int max_active
)
3345 max_active
= wq_clamp_max_active(max_active
, wq
->flags
, wq
->name
);
3347 spin_lock(&workqueue_lock
);
3349 wq
->saved_max_active
= max_active
;
3351 for_each_cwq_cpu(cpu
, wq
) {
3352 struct global_cwq
*gcwq
= get_gcwq(cpu
);
3354 spin_lock_irq(&gcwq
->lock
);
3356 if (!(wq
->flags
& WQ_FREEZABLE
) ||
3357 !(gcwq
->flags
& GCWQ_FREEZING
))
3358 get_cwq(gcwq
->cpu
, wq
)->max_active
= max_active
;
3360 spin_unlock_irq(&gcwq
->lock
);
3363 spin_unlock(&workqueue_lock
);
3365 EXPORT_SYMBOL_GPL(workqueue_set_max_active
);
3368 * workqueue_congested - test whether a workqueue is congested
3369 * @cpu: CPU in question
3370 * @wq: target workqueue
3372 * Test whether @wq's cpu workqueue for @cpu is congested. There is
3373 * no synchronization around this function and the test result is
3374 * unreliable and only useful as advisory hints or for debugging.
3377 * %true if congested, %false otherwise.
3379 bool workqueue_congested(unsigned int cpu
, struct workqueue_struct
*wq
)
3381 struct cpu_workqueue_struct
*cwq
= get_cwq(cpu
, wq
);
3383 return !list_empty(&cwq
->delayed_works
);
3385 EXPORT_SYMBOL_GPL(workqueue_congested
);
3388 * work_cpu - return the last known associated cpu for @work
3389 * @work: the work of interest
3392 * CPU number if @work was ever queued. WORK_CPU_NONE otherwise.
3394 unsigned int work_cpu(struct work_struct
*work
)
3396 struct global_cwq
*gcwq
= get_work_gcwq(work
);
3398 return gcwq
? gcwq
->cpu
: WORK_CPU_NONE
;
3400 EXPORT_SYMBOL_GPL(work_cpu
);
3403 * work_busy - test whether a work is currently pending or running
3404 * @work: the work to be tested
3406 * Test whether @work is currently pending or running. There is no
3407 * synchronization around this function and the test result is
3408 * unreliable and only useful as advisory hints or for debugging.
3409 * Especially for reentrant wqs, the pending state might hide the
3413 * OR'd bitmask of WORK_BUSY_* bits.
3415 unsigned int work_busy(struct work_struct
*work
)
3417 struct global_cwq
*gcwq
= get_work_gcwq(work
);
3418 unsigned long flags
;
3419 unsigned int ret
= 0;
3424 spin_lock_irqsave(&gcwq
->lock
, flags
);
3426 if (work_pending(work
))
3427 ret
|= WORK_BUSY_PENDING
;
3428 if (find_worker_executing_work(gcwq
, work
))
3429 ret
|= WORK_BUSY_RUNNING
;
3431 spin_unlock_irqrestore(&gcwq
->lock
, flags
);
3435 EXPORT_SYMBOL_GPL(work_busy
);
3440 * There are two challenges in supporting CPU hotplug. Firstly, there
3441 * are a lot of assumptions on strong associations among work, cwq and
3442 * gcwq which make migrating pending and scheduled works very
3443 * difficult to implement without impacting hot paths. Secondly,
3444 * gcwqs serve mix of short, long and very long running works making
3445 * blocked draining impractical.
3447 * This is solved by allowing a gcwq to be disassociated from the CPU
3448 * running as an unbound one and allowing it to be reattached later if the
3449 * cpu comes back online.
3452 /* claim manager positions of all pools */
3453 static void gcwq_claim_management_and_lock(struct global_cwq
*gcwq
)
3455 struct worker_pool
*pool
;
3457 for_each_worker_pool(pool
, gcwq
)
3458 mutex_lock_nested(&pool
->manager_mutex
, pool
- gcwq
->pools
);
3459 spin_lock_irq(&gcwq
->lock
);
3462 /* release manager positions */
3463 static void gcwq_release_management_and_unlock(struct global_cwq
*gcwq
)
3465 struct worker_pool
*pool
;
3467 spin_unlock_irq(&gcwq
->lock
);
3468 for_each_worker_pool(pool
, gcwq
)
3469 mutex_unlock(&pool
->manager_mutex
);
3472 static void gcwq_unbind_fn(struct work_struct
*work
)
3474 struct global_cwq
*gcwq
= get_gcwq(smp_processor_id());
3475 struct worker_pool
*pool
;
3476 struct worker
*worker
;
3477 struct hlist_node
*pos
;
3480 BUG_ON(gcwq
->cpu
!= smp_processor_id());
3482 gcwq_claim_management_and_lock(gcwq
);
3485 * We've claimed all manager positions. Make all workers unbound
3486 * and set DISASSOCIATED. Before this, all workers except for the
3487 * ones which are still executing works from before the last CPU
3488 * down must be on the cpu. After this, they may become diasporas.
3490 for_each_worker_pool(pool
, gcwq
)
3491 list_for_each_entry(worker
, &pool
->idle_list
, entry
)
3492 worker
->flags
|= WORKER_UNBOUND
;
3494 for_each_busy_worker(worker
, i
, pos
, gcwq
)
3495 worker
->flags
|= WORKER_UNBOUND
;
3497 gcwq
->flags
|= GCWQ_DISASSOCIATED
;
3499 gcwq_release_management_and_unlock(gcwq
);
3502 * Call schedule() so that we cross rq->lock and thus can guarantee
3503 * sched callbacks see the %WORKER_UNBOUND flag. This is necessary
3504 * as scheduler callbacks may be invoked from other cpus.
3509 * Sched callbacks are disabled now. Zap nr_running. After this,
3510 * nr_running stays zero and need_more_worker() and keep_working()
3511 * are always true as long as the worklist is not empty. @gcwq now
3512 * behaves as unbound (in terms of concurrency management) gcwq
3513 * which is served by workers tied to the CPU.
3515 * On return from this function, the current worker would trigger
3516 * unbound chain execution of pending work items if other workers
3519 for_each_worker_pool(pool
, gcwq
)
3520 atomic_set(get_pool_nr_running(pool
), 0);
3524 * Workqueues should be brought up before normal priority CPU notifiers.
3525 * This will be registered high priority CPU notifier.
3527 static int __devinit
workqueue_cpu_up_callback(struct notifier_block
*nfb
,
3528 unsigned long action
,
3531 unsigned int cpu
= (unsigned long)hcpu
;
3532 struct global_cwq
*gcwq
= get_gcwq(cpu
);
3533 struct worker_pool
*pool
;
3535 switch (action
& ~CPU_TASKS_FROZEN
) {
3536 case CPU_UP_PREPARE
:
3537 for_each_worker_pool(pool
, gcwq
) {
3538 struct worker
*worker
;
3540 if (pool
->nr_workers
)
3543 worker
= create_worker(pool
);
3547 spin_lock_irq(&gcwq
->lock
);
3548 start_worker(worker
);
3549 spin_unlock_irq(&gcwq
->lock
);
3553 case CPU_DOWN_FAILED
:
3555 gcwq_claim_management_and_lock(gcwq
);
3556 gcwq
->flags
&= ~GCWQ_DISASSOCIATED
;
3557 rebind_workers(gcwq
);
3558 gcwq_release_management_and_unlock(gcwq
);
3565 * Workqueues should be brought down after normal priority CPU notifiers.
3566 * This will be registered as low priority CPU notifier.
3568 static int __devinit
workqueue_cpu_down_callback(struct notifier_block
*nfb
,
3569 unsigned long action
,
3572 unsigned int cpu
= (unsigned long)hcpu
;
3573 struct work_struct unbind_work
;
3575 switch (action
& ~CPU_TASKS_FROZEN
) {
3576 case CPU_DOWN_PREPARE
:
3577 /* unbinding should happen on the local CPU */
3578 INIT_WORK_ONSTACK(&unbind_work
, gcwq_unbind_fn
);
3579 queue_work_on(cpu
, system_highpri_wq
, &unbind_work
);
3580 flush_work(&unbind_work
);
3588 struct work_for_cpu
{
3589 struct completion completion
;
3595 static int do_work_for_cpu(void *_wfc
)
3597 struct work_for_cpu
*wfc
= _wfc
;
3598 wfc
->ret
= wfc
->fn(wfc
->arg
);
3599 complete(&wfc
->completion
);
3604 * work_on_cpu - run a function in user context on a particular cpu
3605 * @cpu: the cpu to run on
3606 * @fn: the function to run
3607 * @arg: the function arg
3609 * This will return the value @fn returns.
3610 * It is up to the caller to ensure that the cpu doesn't go offline.
3611 * The caller must not hold any locks which would prevent @fn from completing.
3613 long work_on_cpu(unsigned int cpu
, long (*fn
)(void *), void *arg
)
3615 struct task_struct
*sub_thread
;
3616 struct work_for_cpu wfc
= {
3617 .completion
= COMPLETION_INITIALIZER_ONSTACK(wfc
.completion
),
3622 sub_thread
= kthread_create(do_work_for_cpu
, &wfc
, "work_for_cpu");
3623 if (IS_ERR(sub_thread
))
3624 return PTR_ERR(sub_thread
);
3625 kthread_bind(sub_thread
, cpu
);
3626 wake_up_process(sub_thread
);
3627 wait_for_completion(&wfc
.completion
);
3630 EXPORT_SYMBOL_GPL(work_on_cpu
);
3631 #endif /* CONFIG_SMP */
3633 #ifdef CONFIG_FREEZER
3636 * freeze_workqueues_begin - begin freezing workqueues
3638 * Start freezing workqueues. After this function returns, all freezable
3639 * workqueues will queue new works to their frozen_works list instead of
3643 * Grabs and releases workqueue_lock and gcwq->lock's.
3645 void freeze_workqueues_begin(void)
3649 spin_lock(&workqueue_lock
);
3651 BUG_ON(workqueue_freezing
);
3652 workqueue_freezing
= true;
3654 for_each_gcwq_cpu(cpu
) {
3655 struct global_cwq
*gcwq
= get_gcwq(cpu
);
3656 struct workqueue_struct
*wq
;
3658 spin_lock_irq(&gcwq
->lock
);
3660 BUG_ON(gcwq
->flags
& GCWQ_FREEZING
);
3661 gcwq
->flags
|= GCWQ_FREEZING
;
3663 list_for_each_entry(wq
, &workqueues
, list
) {
3664 struct cpu_workqueue_struct
*cwq
= get_cwq(cpu
, wq
);
3666 if (cwq
&& wq
->flags
& WQ_FREEZABLE
)
3667 cwq
->max_active
= 0;
3670 spin_unlock_irq(&gcwq
->lock
);
3673 spin_unlock(&workqueue_lock
);
3677 * freeze_workqueues_busy - are freezable workqueues still busy?
3679 * Check whether freezing is complete. This function must be called
3680 * between freeze_workqueues_begin() and thaw_workqueues().
3683 * Grabs and releases workqueue_lock.
3686 * %true if some freezable workqueues are still busy. %false if freezing
3689 bool freeze_workqueues_busy(void)
3694 spin_lock(&workqueue_lock
);
3696 BUG_ON(!workqueue_freezing
);
3698 for_each_gcwq_cpu(cpu
) {
3699 struct workqueue_struct
*wq
;
3701 * nr_active is monotonically decreasing. It's safe
3702 * to peek without lock.
3704 list_for_each_entry(wq
, &workqueues
, list
) {
3705 struct cpu_workqueue_struct
*cwq
= get_cwq(cpu
, wq
);
3707 if (!cwq
|| !(wq
->flags
& WQ_FREEZABLE
))
3710 BUG_ON(cwq
->nr_active
< 0);
3711 if (cwq
->nr_active
) {
3718 spin_unlock(&workqueue_lock
);
3723 * thaw_workqueues - thaw workqueues
3725 * Thaw workqueues. Normal queueing is restored and all collected
3726 * frozen works are transferred to their respective gcwq worklists.
3729 * Grabs and releases workqueue_lock and gcwq->lock's.
3731 void thaw_workqueues(void)
3735 spin_lock(&workqueue_lock
);
3737 if (!workqueue_freezing
)
3740 for_each_gcwq_cpu(cpu
) {
3741 struct global_cwq
*gcwq
= get_gcwq(cpu
);
3742 struct worker_pool
*pool
;
3743 struct workqueue_struct
*wq
;
3745 spin_lock_irq(&gcwq
->lock
);
3747 BUG_ON(!(gcwq
->flags
& GCWQ_FREEZING
));
3748 gcwq
->flags
&= ~GCWQ_FREEZING
;
3750 list_for_each_entry(wq
, &workqueues
, list
) {
3751 struct cpu_workqueue_struct
*cwq
= get_cwq(cpu
, wq
);
3753 if (!cwq
|| !(wq
->flags
& WQ_FREEZABLE
))
3756 /* restore max_active and repopulate worklist */
3757 cwq
->max_active
= wq
->saved_max_active
;
3759 while (!list_empty(&cwq
->delayed_works
) &&
3760 cwq
->nr_active
< cwq
->max_active
)
3761 cwq_activate_first_delayed(cwq
);
3764 for_each_worker_pool(pool
, gcwq
)
3765 wake_up_worker(pool
);
3767 spin_unlock_irq(&gcwq
->lock
);
3770 workqueue_freezing
= false;
3772 spin_unlock(&workqueue_lock
);
3774 #endif /* CONFIG_FREEZER */
3776 static int __init
init_workqueues(void)
3781 /* make sure we have enough bits for OFFQ CPU number */
3782 BUILD_BUG_ON((1LU << (BITS_PER_LONG
- WORK_OFFQ_CPU_SHIFT
)) <
3785 cpu_notifier(workqueue_cpu_up_callback
, CPU_PRI_WORKQUEUE_UP
);
3786 cpu_notifier(workqueue_cpu_down_callback
, CPU_PRI_WORKQUEUE_DOWN
);
3788 /* initialize gcwqs */
3789 for_each_gcwq_cpu(cpu
) {
3790 struct global_cwq
*gcwq
= get_gcwq(cpu
);
3791 struct worker_pool
*pool
;
3793 spin_lock_init(&gcwq
->lock
);
3795 gcwq
->flags
|= GCWQ_DISASSOCIATED
;
3797 for (i
= 0; i
< BUSY_WORKER_HASH_SIZE
; i
++)
3798 INIT_HLIST_HEAD(&gcwq
->busy_hash
[i
]);
3800 for_each_worker_pool(pool
, gcwq
) {
3802 INIT_LIST_HEAD(&pool
->worklist
);
3803 INIT_LIST_HEAD(&pool
->idle_list
);
3805 init_timer_deferrable(&pool
->idle_timer
);
3806 pool
->idle_timer
.function
= idle_worker_timeout
;
3807 pool
->idle_timer
.data
= (unsigned long)pool
;
3809 setup_timer(&pool
->mayday_timer
, gcwq_mayday_timeout
,
3810 (unsigned long)pool
);
3812 mutex_init(&pool
->manager_mutex
);
3813 ida_init(&pool
->worker_ida
);
3816 init_waitqueue_head(&gcwq
->rebind_hold
);
3819 /* create the initial worker */
3820 for_each_online_gcwq_cpu(cpu
) {
3821 struct global_cwq
*gcwq
= get_gcwq(cpu
);
3822 struct worker_pool
*pool
;
3824 if (cpu
!= WORK_CPU_UNBOUND
)
3825 gcwq
->flags
&= ~GCWQ_DISASSOCIATED
;
3827 for_each_worker_pool(pool
, gcwq
) {
3828 struct worker
*worker
;
3830 worker
= create_worker(pool
);
3832 spin_lock_irq(&gcwq
->lock
);
3833 start_worker(worker
);
3834 spin_unlock_irq(&gcwq
->lock
);
3838 system_wq
= alloc_workqueue("events", 0, 0);
3839 system_highpri_wq
= alloc_workqueue("events_highpri", WQ_HIGHPRI
, 0);
3840 system_long_wq
= alloc_workqueue("events_long", 0, 0);
3841 system_nrt_wq
= alloc_workqueue("events_nrt", WQ_NON_REENTRANT
, 0);
3842 system_unbound_wq
= alloc_workqueue("events_unbound", WQ_UNBOUND
,
3843 WQ_UNBOUND_MAX_ACTIVE
);
3844 system_freezable_wq
= alloc_workqueue("events_freezable",
3846 system_nrt_freezable_wq
= alloc_workqueue("events_nrt_freezable",
3847 WQ_NON_REENTRANT
| WQ_FREEZABLE
, 0);
3848 BUG_ON(!system_wq
|| !system_highpri_wq
|| !system_long_wq
||
3849 !system_nrt_wq
|| !system_unbound_wq
|| !system_freezable_wq
||
3850 !system_nrt_freezable_wq
);
3853 early_initcall(init_workqueues
);