1 // SPDX-License-Identifier: GPL-2.0-only
3 * kernel/workqueue.c - generic async execution with shared worker pool
5 * Copyright (C) 2002 Ingo Molnar
7 * Derived from the taskqueue/keventd code by:
8 * David Woodhouse <dwmw2@infradead.org>
10 * Kai Petzke <wpp@marie.physik.tu-berlin.de>
11 * Theodore Ts'o <tytso@mit.edu>
13 * Made to use alloc_percpu by Christoph Lameter.
15 * Copyright (C) 2010 SUSE Linux Products GmbH
16 * Copyright (C) 2010 Tejun Heo <tj@kernel.org>
18 * This is the generic async execution mechanism. Work items as are
19 * executed in process context. The worker pool is shared and
20 * automatically managed. There are two worker pools for each CPU (one for
21 * normal work items and the other for high priority ones) and some extra
22 * pools for workqueues which are not bound to any specific CPU - the
23 * number of these backing pools is dynamic.
25 * Please read Documentation/core-api/workqueue.rst for details.
28 #include <linux/export.h>
29 #include <linux/kernel.h>
30 #include <linux/sched.h>
31 #include <linux/init.h>
32 #include <linux/interrupt.h>
33 #include <linux/signal.h>
34 #include <linux/completion.h>
35 #include <linux/workqueue.h>
36 #include <linux/slab.h>
37 #include <linux/cpu.h>
38 #include <linux/notifier.h>
39 #include <linux/kthread.h>
40 #include <linux/hardirq.h>
41 #include <linux/mempolicy.h>
42 #include <linux/freezer.h>
43 #include <linux/debug_locks.h>
44 #include <linux/lockdep.h>
45 #include <linux/idr.h>
46 #include <linux/jhash.h>
47 #include <linux/hashtable.h>
48 #include <linux/rculist.h>
49 #include <linux/nodemask.h>
50 #include <linux/moduleparam.h>
51 #include <linux/uaccess.h>
52 #include <linux/sched/isolation.h>
53 #include <linux/sched/debug.h>
54 #include <linux/nmi.h>
55 #include <linux/kvm_para.h>
56 #include <linux/delay.h>
57 #include <linux/irq_work.h>
59 #include "workqueue_internal.h"
61 enum worker_pool_flags
{
65 * A bound pool is either associated or disassociated with its CPU.
66 * While associated (!DISASSOCIATED), all workers are bound to the
67 * CPU and none has %WORKER_UNBOUND set and concurrency management
70 * While DISASSOCIATED, the cpu may be offline and all workers have
71 * %WORKER_UNBOUND set and concurrency management disabled, and may
72 * be executing on any CPU. The pool behaves as an unbound one.
74 * Note that DISASSOCIATED should be flipped only while holding
75 * wq_pool_attach_mutex to avoid changing binding state while
76 * worker_attach_to_pool() is in progress.
78 * As there can only be one concurrent BH execution context per CPU, a
79 * BH pool is per-CPU and always DISASSOCIATED.
81 POOL_BH
= 1 << 0, /* is a BH pool */
82 POOL_MANAGER_ACTIVE
= 1 << 1, /* being managed */
83 POOL_DISASSOCIATED
= 1 << 2, /* cpu can't serve workers */
84 POOL_BH_DRAINING
= 1 << 3, /* draining after CPU offline */
89 WORKER_DIE
= 1 << 1, /* die die die */
90 WORKER_IDLE
= 1 << 2, /* is idle */
91 WORKER_PREP
= 1 << 3, /* preparing to run works */
92 WORKER_CPU_INTENSIVE
= 1 << 6, /* cpu intensive */
93 WORKER_UNBOUND
= 1 << 7, /* worker is unbound */
94 WORKER_REBOUND
= 1 << 8, /* worker was rebound */
96 WORKER_NOT_RUNNING
= WORKER_PREP
| WORKER_CPU_INTENSIVE
|
97 WORKER_UNBOUND
| WORKER_REBOUND
,
100 enum work_cancel_flags
{
101 WORK_CANCEL_DELAYED
= 1 << 0, /* canceling a delayed_work */
104 enum wq_internal_consts
{
105 NR_STD_WORKER_POOLS
= 2, /* # standard pools per cpu */
107 UNBOUND_POOL_HASH_ORDER
= 6, /* hashed by pool->attrs */
108 BUSY_WORKER_HASH_ORDER
= 6, /* 64 pointers */
110 MAX_IDLE_WORKERS_RATIO
= 4, /* 1/4 of busy can be idle */
111 IDLE_WORKER_TIMEOUT
= 300 * HZ
, /* keep idle ones for 5 mins */
113 MAYDAY_INITIAL_TIMEOUT
= HZ
/ 100 >= 2 ? HZ
/ 100 : 2,
114 /* call for help after 10ms
116 MAYDAY_INTERVAL
= HZ
/ 10, /* and then every 100ms */
117 CREATE_COOLDOWN
= HZ
, /* time to breath after fail */
120 * Rescue workers are used only on emergencies and shared by
121 * all cpus. Give MIN_NICE.
123 RESCUER_NICE_LEVEL
= MIN_NICE
,
124 HIGHPRI_NICE_LEVEL
= MIN_NICE
,
130 * We don't want to trap softirq for too long. See MAX_SOFTIRQ_TIME and
131 * MAX_SOFTIRQ_RESTART in kernel/softirq.c. These are macros because
132 * msecs_to_jiffies() can't be an initializer.
134 #define BH_WORKER_JIFFIES msecs_to_jiffies(2)
135 #define BH_WORKER_RESTARTS 10
138 * Structure fields follow one of the following exclusion rules.
140 * I: Modifiable by initialization/destruction paths and read-only for
143 * P: Preemption protected. Disabling preemption is enough and should
144 * only be modified and accessed from the local cpu.
146 * L: pool->lock protected. Access with pool->lock held.
148 * LN: pool->lock and wq_node_nr_active->lock protected for writes. Either for
151 * K: Only modified by worker while holding pool->lock. Can be safely read by
152 * self, while holding pool->lock or from IRQ context if %current is the
155 * S: Only modified by worker self.
157 * A: wq_pool_attach_mutex protected.
159 * PL: wq_pool_mutex protected.
161 * PR: wq_pool_mutex protected for writes. RCU protected for reads.
163 * PW: wq_pool_mutex and wq->mutex protected for writes. Either for reads.
165 * PWR: wq_pool_mutex and wq->mutex protected for writes. Either or
168 * WQ: wq->mutex protected.
170 * WR: wq->mutex protected for writes. RCU protected for reads.
172 * WO: wq->mutex protected for writes. Updated with WRITE_ONCE() and can be read
173 * with READ_ONCE() without locking.
175 * MD: wq_mayday_lock protected.
177 * WD: Used internally by the watchdog.
180 /* struct worker is defined in workqueue_internal.h */
183 raw_spinlock_t lock
; /* the pool lock */
184 int cpu
; /* I: the associated cpu */
185 int node
; /* I: the associated node ID */
186 int id
; /* I: pool ID */
187 unsigned int flags
; /* L: flags */
189 unsigned long watchdog_ts
; /* L: watchdog timestamp */
190 bool cpu_stall
; /* WD: stalled cpu bound pool */
193 * The counter is incremented in a process context on the associated CPU
194 * w/ preemption disabled, and decremented or reset in the same context
195 * but w/ pool->lock held. The readers grab pool->lock and are
196 * guaranteed to see if the counter reached zero.
200 struct list_head worklist
; /* L: list of pending works */
202 int nr_workers
; /* L: total number of workers */
203 int nr_idle
; /* L: currently idle workers */
205 struct list_head idle_list
; /* L: list of idle workers */
206 struct timer_list idle_timer
; /* L: worker idle timeout */
207 struct work_struct idle_cull_work
; /* L: worker idle cleanup */
209 struct timer_list mayday_timer
; /* L: SOS timer for workers */
211 /* a workers is either on busy_hash or idle_list, or the manager */
212 DECLARE_HASHTABLE(busy_hash
, BUSY_WORKER_HASH_ORDER
);
213 /* L: hash of busy workers */
215 struct worker
*manager
; /* L: purely informational */
216 struct list_head workers
; /* A: attached workers */
217 struct list_head dying_workers
; /* A: workers about to die */
218 struct completion
*detach_completion
; /* all workers detached */
220 struct ida worker_ida
; /* worker IDs for task name */
222 struct workqueue_attrs
*attrs
; /* I: worker attributes */
223 struct hlist_node hash_node
; /* PL: unbound_pool_hash node */
224 int refcnt
; /* PL: refcnt for unbound pools */
227 * Destruction of pool is RCU protected to allow dereferences
228 * from get_work_pool().
234 * Per-pool_workqueue statistics. These can be monitored using
235 * tools/workqueue/wq_monitor.py.
237 enum pool_workqueue_stats
{
238 PWQ_STAT_STARTED
, /* work items started execution */
239 PWQ_STAT_COMPLETED
, /* work items completed execution */
240 PWQ_STAT_CPU_TIME
, /* total CPU time consumed */
241 PWQ_STAT_CPU_INTENSIVE
, /* wq_cpu_intensive_thresh_us violations */
242 PWQ_STAT_CM_WAKEUP
, /* concurrency-management worker wakeups */
243 PWQ_STAT_REPATRIATED
, /* unbound workers brought back into scope */
244 PWQ_STAT_MAYDAY
, /* maydays to rescuer */
245 PWQ_STAT_RESCUED
, /* linked work items executed by rescuer */
251 * The per-pool workqueue. While queued, bits below WORK_PWQ_SHIFT
252 * of work_struct->data are used for flags and the remaining high bits
253 * point to the pwq; thus, pwqs need to be aligned at two's power of the
254 * number of flag bits.
256 struct pool_workqueue
{
257 struct worker_pool
*pool
; /* I: the associated pool */
258 struct workqueue_struct
*wq
; /* I: the owning workqueue */
259 int work_color
; /* L: current color */
260 int flush_color
; /* L: flushing color */
261 int refcnt
; /* L: reference count */
262 int nr_in_flight
[WORK_NR_COLORS
];
263 /* L: nr of in_flight works */
264 bool plugged
; /* L: execution suspended */
267 * nr_active management and WORK_STRUCT_INACTIVE:
269 * When pwq->nr_active >= max_active, new work item is queued to
270 * pwq->inactive_works instead of pool->worklist and marked with
271 * WORK_STRUCT_INACTIVE.
273 * All work items marked with WORK_STRUCT_INACTIVE do not participate in
274 * nr_active and all work items in pwq->inactive_works are marked with
275 * WORK_STRUCT_INACTIVE. But not all WORK_STRUCT_INACTIVE work items are
276 * in pwq->inactive_works. Some of them are ready to run in
277 * pool->worklist or worker->scheduled. Those work itmes are only struct
278 * wq_barrier which is used for flush_work() and should not participate
279 * in nr_active. For non-barrier work item, it is marked with
280 * WORK_STRUCT_INACTIVE iff it is in pwq->inactive_works.
282 int nr_active
; /* L: nr of active works */
283 struct list_head inactive_works
; /* L: inactive works */
284 struct list_head pending_node
; /* LN: node on wq_node_nr_active->pending_pwqs */
285 struct list_head pwqs_node
; /* WR: node on wq->pwqs */
286 struct list_head mayday_node
; /* MD: node on wq->maydays */
288 u64 stats
[PWQ_NR_STATS
];
291 * Release of unbound pwq is punted to a kthread_worker. See put_pwq()
292 * and pwq_release_workfn() for details. pool_workqueue itself is also
293 * RCU protected so that the first pwq can be determined without
294 * grabbing wq->mutex.
296 struct kthread_work release_work
;
298 } __aligned(1 << WORK_STRUCT_PWQ_SHIFT
);
301 * Structure used to wait for workqueue flush.
304 struct list_head list
; /* WQ: list of flushers */
305 int flush_color
; /* WQ: flush color waiting for */
306 struct completion done
; /* flush completion */
312 * Unlike in a per-cpu workqueue where max_active limits its concurrency level
313 * on each CPU, in an unbound workqueue, max_active applies to the whole system.
314 * As sharing a single nr_active across multiple sockets can be very expensive,
315 * the counting and enforcement is per NUMA node.
317 * The following struct is used to enforce per-node max_active. When a pwq wants
318 * to start executing a work item, it should increment ->nr using
319 * tryinc_node_nr_active(). If acquisition fails due to ->nr already being over
320 * ->max, the pwq is queued on ->pending_pwqs. As in-flight work items finish
321 * and decrement ->nr, node_activate_pending_pwq() activates the pending pwqs in
324 struct wq_node_nr_active
{
325 int max
; /* per-node max_active */
326 atomic_t nr
; /* per-node nr_active */
327 raw_spinlock_t lock
; /* nests inside pool locks */
328 struct list_head pending_pwqs
; /* LN: pwqs with inactive works */
332 * The externally visible workqueue. It relays the issued work items to
333 * the appropriate worker_pool through its pool_workqueues.
335 struct workqueue_struct
{
336 struct list_head pwqs
; /* WR: all pwqs of this wq */
337 struct list_head list
; /* PR: list of all workqueues */
339 struct mutex mutex
; /* protects this wq */
340 int work_color
; /* WQ: current work color */
341 int flush_color
; /* WQ: current flush color */
342 atomic_t nr_pwqs_to_flush
; /* flush in progress */
343 struct wq_flusher
*first_flusher
; /* WQ: first flusher */
344 struct list_head flusher_queue
; /* WQ: flush waiters */
345 struct list_head flusher_overflow
; /* WQ: flush overflow list */
347 struct list_head maydays
; /* MD: pwqs requesting rescue */
348 struct worker
*rescuer
; /* MD: rescue worker */
350 int nr_drainers
; /* WQ: drain in progress */
352 /* See alloc_workqueue() function comment for info on min/max_active */
353 int max_active
; /* WO: max active works */
354 int min_active
; /* WO: min active works */
355 int saved_max_active
; /* WQ: saved max_active */
356 int saved_min_active
; /* WQ: saved min_active */
358 struct workqueue_attrs
*unbound_attrs
; /* PW: only for unbound wqs */
359 struct pool_workqueue __rcu
*dfl_pwq
; /* PW: only for unbound wqs */
362 struct wq_device
*wq_dev
; /* I: for sysfs interface */
364 #ifdef CONFIG_LOCKDEP
366 struct lock_class_key key
;
367 struct lockdep_map lockdep_map
;
369 char name
[WQ_NAME_LEN
]; /* I: workqueue name */
372 * Destruction of workqueue_struct is RCU protected to allow walking
373 * the workqueues list without grabbing wq_pool_mutex.
374 * This is used to dump all workqueues from sysrq.
378 /* hot fields used during command issue, aligned to cacheline */
379 unsigned int flags ____cacheline_aligned
; /* WQ: WQ_* flags */
380 struct pool_workqueue __percpu __rcu
**cpu_pwq
; /* I: per-cpu pwqs */
381 struct wq_node_nr_active
*node_nr_active
[]; /* I: per-node nr_active */
385 * Each pod type describes how CPUs should be grouped for unbound workqueues.
386 * See the comment above workqueue_attrs->affn_scope.
389 int nr_pods
; /* number of pods */
390 cpumask_var_t
*pod_cpus
; /* pod -> cpus */
391 int *pod_node
; /* pod -> node */
392 int *cpu_pod
; /* cpu -> pod */
395 static const char *wq_affn_names
[WQ_AFFN_NR_TYPES
] = {
396 [WQ_AFFN_DFL
] = "default",
397 [WQ_AFFN_CPU
] = "cpu",
398 [WQ_AFFN_SMT
] = "smt",
399 [WQ_AFFN_CACHE
] = "cache",
400 [WQ_AFFN_NUMA
] = "numa",
401 [WQ_AFFN_SYSTEM
] = "system",
405 * Per-cpu work items which run for longer than the following threshold are
406 * automatically considered CPU intensive and excluded from concurrency
407 * management to prevent them from noticeably delaying other per-cpu work items.
408 * ULONG_MAX indicates that the user hasn't overridden it with a boot parameter.
409 * The actual value is initialized in wq_cpu_intensive_thresh_init().
411 static unsigned long wq_cpu_intensive_thresh_us
= ULONG_MAX
;
412 module_param_named(cpu_intensive_thresh_us
, wq_cpu_intensive_thresh_us
, ulong
, 0644);
413 #ifdef CONFIG_WQ_CPU_INTENSIVE_REPORT
414 static unsigned int wq_cpu_intensive_warning_thresh
= 4;
415 module_param_named(cpu_intensive_warning_thresh
, wq_cpu_intensive_warning_thresh
, uint
, 0644);
418 /* see the comment above the definition of WQ_POWER_EFFICIENT */
419 static bool wq_power_efficient
= IS_ENABLED(CONFIG_WQ_POWER_EFFICIENT_DEFAULT
);
420 module_param_named(power_efficient
, wq_power_efficient
, bool, 0444);
422 static bool wq_online
; /* can kworkers be created yet? */
423 static bool wq_topo_initialized __read_mostly
= false;
425 static struct kmem_cache
*pwq_cache
;
427 static struct wq_pod_type wq_pod_types
[WQ_AFFN_NR_TYPES
];
428 static enum wq_affn_scope wq_affn_dfl
= WQ_AFFN_CACHE
;
430 /* buf for wq_update_unbound_pod_attrs(), protected by CPU hotplug exclusion */
431 static struct workqueue_attrs
*wq_update_pod_attrs_buf
;
433 static DEFINE_MUTEX(wq_pool_mutex
); /* protects pools and workqueues list */
434 static DEFINE_MUTEX(wq_pool_attach_mutex
); /* protects worker attach/detach */
435 static DEFINE_RAW_SPINLOCK(wq_mayday_lock
); /* protects wq->maydays list */
436 /* wait for manager to go away */
437 static struct rcuwait manager_wait
= __RCUWAIT_INITIALIZER(manager_wait
);
439 static LIST_HEAD(workqueues
); /* PR: list of all workqueues */
440 static bool workqueue_freezing
; /* PL: have wqs started freezing? */
442 /* PL&A: allowable cpus for unbound wqs and work items */
443 static cpumask_var_t wq_unbound_cpumask
;
445 /* PL: user requested unbound cpumask via sysfs */
446 static cpumask_var_t wq_requested_unbound_cpumask
;
448 /* PL: isolated cpumask to be excluded from unbound cpumask */
449 static cpumask_var_t wq_isolated_cpumask
;
451 /* for further constrain wq_unbound_cpumask by cmdline parameter*/
452 static struct cpumask wq_cmdline_cpumask __initdata
;
454 /* CPU where unbound work was last round robin scheduled from this CPU */
455 static DEFINE_PER_CPU(int, wq_rr_cpu_last
);
458 * Local execution of unbound work items is no longer guaranteed. The
459 * following always forces round-robin CPU selection on unbound work items
460 * to uncover usages which depend on it.
462 #ifdef CONFIG_DEBUG_WQ_FORCE_RR_CPU
463 static bool wq_debug_force_rr_cpu
= true;
465 static bool wq_debug_force_rr_cpu
= false;
467 module_param_named(debug_force_rr_cpu
, wq_debug_force_rr_cpu
, bool, 0644);
469 /* to raise softirq for the BH worker pools on other CPUs */
470 static DEFINE_PER_CPU_SHARED_ALIGNED(struct irq_work
[NR_STD_WORKER_POOLS
],
473 /* the BH worker pools */
474 static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool
[NR_STD_WORKER_POOLS
],
477 /* the per-cpu worker pools */
478 static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool
[NR_STD_WORKER_POOLS
],
481 static DEFINE_IDR(worker_pool_idr
); /* PR: idr of all pools */
483 /* PL: hash of all unbound pools keyed by pool->attrs */
484 static DEFINE_HASHTABLE(unbound_pool_hash
, UNBOUND_POOL_HASH_ORDER
);
486 /* I: attributes used when instantiating standard unbound pools on demand */
487 static struct workqueue_attrs
*unbound_std_wq_attrs
[NR_STD_WORKER_POOLS
];
489 /* I: attributes used when instantiating ordered pools on demand */
490 static struct workqueue_attrs
*ordered_wq_attrs
[NR_STD_WORKER_POOLS
];
493 * Used to synchronize multiple cancel_sync attempts on the same work item. See
494 * work_grab_pending() and __cancel_work_sync().
496 static DECLARE_WAIT_QUEUE_HEAD(wq_cancel_waitq
);
499 * I: kthread_worker to release pwq's. pwq release needs to be bounced to a
500 * process context while holding a pool lock. Bounce to a dedicated kthread
501 * worker to avoid A-A deadlocks.
503 static struct kthread_worker
*pwq_release_worker __ro_after_init
;
505 struct workqueue_struct
*system_wq __ro_after_init
;
506 EXPORT_SYMBOL(system_wq
);
507 struct workqueue_struct
*system_highpri_wq __ro_after_init
;
508 EXPORT_SYMBOL_GPL(system_highpri_wq
);
509 struct workqueue_struct
*system_long_wq __ro_after_init
;
510 EXPORT_SYMBOL_GPL(system_long_wq
);
511 struct workqueue_struct
*system_unbound_wq __ro_after_init
;
512 EXPORT_SYMBOL_GPL(system_unbound_wq
);
513 struct workqueue_struct
*system_freezable_wq __ro_after_init
;
514 EXPORT_SYMBOL_GPL(system_freezable_wq
);
515 struct workqueue_struct
*system_power_efficient_wq __ro_after_init
;
516 EXPORT_SYMBOL_GPL(system_power_efficient_wq
);
517 struct workqueue_struct
*system_freezable_power_efficient_wq __ro_after_init
;
518 EXPORT_SYMBOL_GPL(system_freezable_power_efficient_wq
);
519 struct workqueue_struct
*system_bh_wq
;
520 EXPORT_SYMBOL_GPL(system_bh_wq
);
521 struct workqueue_struct
*system_bh_highpri_wq
;
522 EXPORT_SYMBOL_GPL(system_bh_highpri_wq
);
524 static int worker_thread(void *__worker
);
525 static void workqueue_sysfs_unregister(struct workqueue_struct
*wq
);
526 static void show_pwq(struct pool_workqueue
*pwq
);
527 static void show_one_worker_pool(struct worker_pool
*pool
);
529 #define CREATE_TRACE_POINTS
530 #include <trace/events/workqueue.h>
532 #define assert_rcu_or_pool_mutex() \
533 RCU_LOCKDEP_WARN(!rcu_read_lock_any_held() && \
534 !lockdep_is_held(&wq_pool_mutex), \
535 "RCU or wq_pool_mutex should be held")
537 #define assert_rcu_or_wq_mutex_or_pool_mutex(wq) \
538 RCU_LOCKDEP_WARN(!rcu_read_lock_any_held() && \
539 !lockdep_is_held(&wq->mutex) && \
540 !lockdep_is_held(&wq_pool_mutex), \
541 "RCU, wq->mutex or wq_pool_mutex should be held")
543 #define for_each_bh_worker_pool(pool, cpu) \
544 for ((pool) = &per_cpu(bh_worker_pools, cpu)[0]; \
545 (pool) < &per_cpu(bh_worker_pools, cpu)[NR_STD_WORKER_POOLS]; \
548 #define for_each_cpu_worker_pool(pool, cpu) \
549 for ((pool) = &per_cpu(cpu_worker_pools, cpu)[0]; \
550 (pool) < &per_cpu(cpu_worker_pools, cpu)[NR_STD_WORKER_POOLS]; \
554 * for_each_pool - iterate through all worker_pools in the system
555 * @pool: iteration cursor
556 * @pi: integer used for iteration
558 * This must be called either with wq_pool_mutex held or RCU read
559 * locked. If the pool needs to be used beyond the locking in effect, the
560 * caller is responsible for guaranteeing that the pool stays online.
562 * The if/else clause exists only for the lockdep assertion and can be
565 #define for_each_pool(pool, pi) \
566 idr_for_each_entry(&worker_pool_idr, pool, pi) \
567 if (({ assert_rcu_or_pool_mutex(); false; })) { } \
571 * for_each_pool_worker - iterate through all workers of a worker_pool
572 * @worker: iteration cursor
573 * @pool: worker_pool to iterate workers of
575 * This must be called with wq_pool_attach_mutex.
577 * The if/else clause exists only for the lockdep assertion and can be
580 #define for_each_pool_worker(worker, pool) \
581 list_for_each_entry((worker), &(pool)->workers, node) \
582 if (({ lockdep_assert_held(&wq_pool_attach_mutex); false; })) { } \
586 * for_each_pwq - iterate through all pool_workqueues of the specified workqueue
587 * @pwq: iteration cursor
588 * @wq: the target workqueue
590 * This must be called either with wq->mutex held or RCU read locked.
591 * If the pwq needs to be used beyond the locking in effect, the caller is
592 * responsible for guaranteeing that the pwq stays online.
594 * The if/else clause exists only for the lockdep assertion and can be
597 #define for_each_pwq(pwq, wq) \
598 list_for_each_entry_rcu((pwq), &(wq)->pwqs, pwqs_node, \
599 lockdep_is_held(&(wq->mutex)))
601 #ifdef CONFIG_DEBUG_OBJECTS_WORK
603 static const struct debug_obj_descr work_debug_descr
;
605 static void *work_debug_hint(void *addr
)
607 return ((struct work_struct
*) addr
)->func
;
610 static bool work_is_static_object(void *addr
)
612 struct work_struct
*work
= addr
;
614 return test_bit(WORK_STRUCT_STATIC_BIT
, work_data_bits(work
));
618 * fixup_init is called when:
619 * - an active object is initialized
621 static bool work_fixup_init(void *addr
, enum debug_obj_state state
)
623 struct work_struct
*work
= addr
;
626 case ODEBUG_STATE_ACTIVE
:
627 cancel_work_sync(work
);
628 debug_object_init(work
, &work_debug_descr
);
636 * fixup_free is called when:
637 * - an active object is freed
639 static bool work_fixup_free(void *addr
, enum debug_obj_state state
)
641 struct work_struct
*work
= addr
;
644 case ODEBUG_STATE_ACTIVE
:
645 cancel_work_sync(work
);
646 debug_object_free(work
, &work_debug_descr
);
653 static const struct debug_obj_descr work_debug_descr
= {
654 .name
= "work_struct",
655 .debug_hint
= work_debug_hint
,
656 .is_static_object
= work_is_static_object
,
657 .fixup_init
= work_fixup_init
,
658 .fixup_free
= work_fixup_free
,
661 static inline void debug_work_activate(struct work_struct
*work
)
663 debug_object_activate(work
, &work_debug_descr
);
666 static inline void debug_work_deactivate(struct work_struct
*work
)
668 debug_object_deactivate(work
, &work_debug_descr
);
671 void __init_work(struct work_struct
*work
, int onstack
)
674 debug_object_init_on_stack(work
, &work_debug_descr
);
676 debug_object_init(work
, &work_debug_descr
);
678 EXPORT_SYMBOL_GPL(__init_work
);
680 void destroy_work_on_stack(struct work_struct
*work
)
682 debug_object_free(work
, &work_debug_descr
);
684 EXPORT_SYMBOL_GPL(destroy_work_on_stack
);
686 void destroy_delayed_work_on_stack(struct delayed_work
*work
)
688 destroy_timer_on_stack(&work
->timer
);
689 debug_object_free(&work
->work
, &work_debug_descr
);
691 EXPORT_SYMBOL_GPL(destroy_delayed_work_on_stack
);
694 static inline void debug_work_activate(struct work_struct
*work
) { }
695 static inline void debug_work_deactivate(struct work_struct
*work
) { }
699 * worker_pool_assign_id - allocate ID and assign it to @pool
700 * @pool: the pool pointer of interest
702 * Returns 0 if ID in [0, WORK_OFFQ_POOL_NONE) is allocated and assigned
703 * successfully, -errno on failure.
705 static int worker_pool_assign_id(struct worker_pool
*pool
)
709 lockdep_assert_held(&wq_pool_mutex
);
711 ret
= idr_alloc(&worker_pool_idr
, pool
, 0, WORK_OFFQ_POOL_NONE
,
720 static struct pool_workqueue __rcu
**
721 unbound_pwq_slot(struct workqueue_struct
*wq
, int cpu
)
724 return per_cpu_ptr(wq
->cpu_pwq
, cpu
);
729 /* @cpu < 0 for dfl_pwq */
730 static struct pool_workqueue
*unbound_pwq(struct workqueue_struct
*wq
, int cpu
)
732 return rcu_dereference_check(*unbound_pwq_slot(wq
, cpu
),
733 lockdep_is_held(&wq_pool_mutex
) ||
734 lockdep_is_held(&wq
->mutex
));
738 * unbound_effective_cpumask - effective cpumask of an unbound workqueue
739 * @wq: workqueue of interest
741 * @wq->unbound_attrs->cpumask contains the cpumask requested by the user which
742 * is masked with wq_unbound_cpumask to determine the effective cpumask. The
743 * default pwq is always mapped to the pool with the current effective cpumask.
745 static struct cpumask
*unbound_effective_cpumask(struct workqueue_struct
*wq
)
747 return unbound_pwq(wq
, -1)->pool
->attrs
->__pod_cpumask
;
750 static unsigned int work_color_to_flags(int color
)
752 return color
<< WORK_STRUCT_COLOR_SHIFT
;
755 static int get_work_color(unsigned long work_data
)
757 return (work_data
>> WORK_STRUCT_COLOR_SHIFT
) &
758 ((1 << WORK_STRUCT_COLOR_BITS
) - 1);
761 static int work_next_color(int color
)
763 return (color
+ 1) % WORK_NR_COLORS
;
767 * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data
768 * contain the pointer to the queued pwq. Once execution starts, the flag
769 * is cleared and the high bits contain OFFQ flags and pool ID.
771 * set_work_pwq(), set_work_pool_and_clear_pending() and mark_work_canceling()
772 * can be used to set the pwq, pool or clear work->data. These functions should
773 * only be called while the work is owned - ie. while the PENDING bit is set.
775 * get_work_pool() and get_work_pwq() can be used to obtain the pool or pwq
776 * corresponding to a work. Pool is available once the work has been
777 * queued anywhere after initialization until it is sync canceled. pwq is
778 * available only while the work item is queued.
780 * %WORK_OFFQ_CANCELING is used to mark a work item which is being
781 * canceled. While being canceled, a work item may have its PENDING set
782 * but stay off timer and worklist for arbitrarily long and nobody should
783 * try to steal the PENDING bit.
785 static inline void set_work_data(struct work_struct
*work
, unsigned long data
)
787 WARN_ON_ONCE(!work_pending(work
));
788 atomic_long_set(&work
->data
, data
| work_static(work
));
791 static void set_work_pwq(struct work_struct
*work
, struct pool_workqueue
*pwq
,
794 set_work_data(work
, (unsigned long)pwq
| WORK_STRUCT_PENDING
|
795 WORK_STRUCT_PWQ
| flags
);
798 static void set_work_pool_and_keep_pending(struct work_struct
*work
,
799 int pool_id
, unsigned long flags
)
801 set_work_data(work
, ((unsigned long)pool_id
<< WORK_OFFQ_POOL_SHIFT
) |
802 WORK_STRUCT_PENDING
| flags
);
805 static void set_work_pool_and_clear_pending(struct work_struct
*work
,
806 int pool_id
, unsigned long flags
)
809 * The following wmb is paired with the implied mb in
810 * test_and_set_bit(PENDING) and ensures all updates to @work made
811 * here are visible to and precede any updates by the next PENDING
815 set_work_data(work
, ((unsigned long)pool_id
<< WORK_OFFQ_POOL_SHIFT
) |
818 * The following mb guarantees that previous clear of a PENDING bit
819 * will not be reordered with any speculative LOADS or STORES from
820 * work->current_func, which is executed afterwards. This possible
821 * reordering can lead to a missed execution on attempt to queue
822 * the same @work. E.g. consider this case:
825 * ---------------------------- --------------------------------
827 * 1 STORE event_indicated
828 * 2 queue_work_on() {
829 * 3 test_and_set_bit(PENDING)
830 * 4 } set_..._and_clear_pending() {
831 * 5 set_work_data() # clear bit
833 * 7 work->current_func() {
834 * 8 LOAD event_indicated
837 * Without an explicit full barrier speculative LOAD on line 8 can
838 * be executed before CPU#0 does STORE on line 1. If that happens,
839 * CPU#0 observes the PENDING bit is still set and new execution of
840 * a @work is not queued in a hope, that CPU#1 will eventually
841 * finish the queued @work. Meanwhile CPU#1 does not see
842 * event_indicated is set, because speculative LOAD was executed
843 * before actual STORE.
848 static inline struct pool_workqueue
*work_struct_pwq(unsigned long data
)
850 return (struct pool_workqueue
*)(data
& WORK_STRUCT_PWQ_MASK
);
853 static struct pool_workqueue
*get_work_pwq(struct work_struct
*work
)
855 unsigned long data
= atomic_long_read(&work
->data
);
857 if (data
& WORK_STRUCT_PWQ
)
858 return work_struct_pwq(data
);
864 * get_work_pool - return the worker_pool a given work was associated with
865 * @work: the work item of interest
867 * Pools are created and destroyed under wq_pool_mutex, and allows read
868 * access under RCU read lock. As such, this function should be
869 * called under wq_pool_mutex or inside of a rcu_read_lock() region.
871 * All fields of the returned pool are accessible as long as the above
872 * mentioned locking is in effect. If the returned pool needs to be used
873 * beyond the critical section, the caller is responsible for ensuring the
874 * returned pool is and stays online.
876 * Return: The worker_pool @work was last associated with. %NULL if none.
878 static struct worker_pool
*get_work_pool(struct work_struct
*work
)
880 unsigned long data
= atomic_long_read(&work
->data
);
883 assert_rcu_or_pool_mutex();
885 if (data
& WORK_STRUCT_PWQ
)
886 return work_struct_pwq(data
)->pool
;
888 pool_id
= data
>> WORK_OFFQ_POOL_SHIFT
;
889 if (pool_id
== WORK_OFFQ_POOL_NONE
)
892 return idr_find(&worker_pool_idr
, pool_id
);
896 * get_work_pool_id - return the worker pool ID a given work is associated with
897 * @work: the work item of interest
899 * Return: The worker_pool ID @work was last associated with.
900 * %WORK_OFFQ_POOL_NONE if none.
902 static int get_work_pool_id(struct work_struct
*work
)
904 unsigned long data
= atomic_long_read(&work
->data
);
906 if (data
& WORK_STRUCT_PWQ
)
907 return work_struct_pwq(data
)->pool
->id
;
909 return data
>> WORK_OFFQ_POOL_SHIFT
;
912 static void mark_work_canceling(struct work_struct
*work
)
914 unsigned long pool_id
= get_work_pool_id(work
);
916 pool_id
<<= WORK_OFFQ_POOL_SHIFT
;
917 set_work_data(work
, pool_id
| WORK_STRUCT_PENDING
| WORK_OFFQ_CANCELING
);
920 static bool work_is_canceling(struct work_struct
*work
)
922 unsigned long data
= atomic_long_read(&work
->data
);
924 return !(data
& WORK_STRUCT_PWQ
) && (data
& WORK_OFFQ_CANCELING
);
928 * Policy functions. These define the policies on how the global worker
929 * pools are managed. Unless noted otherwise, these functions assume that
930 * they're being called with pool->lock held.
934 * Need to wake up a worker? Called from anything but currently
937 * Note that, because unbound workers never contribute to nr_running, this
938 * function will always return %true for unbound pools as long as the
939 * worklist isn't empty.
941 static bool need_more_worker(struct worker_pool
*pool
)
943 return !list_empty(&pool
->worklist
) && !pool
->nr_running
;
946 /* Can I start working? Called from busy but !running workers. */
947 static bool may_start_working(struct worker_pool
*pool
)
949 return pool
->nr_idle
;
952 /* Do I need to keep working? Called from currently running workers. */
953 static bool keep_working(struct worker_pool
*pool
)
955 return !list_empty(&pool
->worklist
) && (pool
->nr_running
<= 1);
958 /* Do we need a new worker? Called from manager. */
959 static bool need_to_create_worker(struct worker_pool
*pool
)
961 return need_more_worker(pool
) && !may_start_working(pool
);
964 /* Do we have too many workers and should some go away? */
965 static bool too_many_workers(struct worker_pool
*pool
)
967 bool managing
= pool
->flags
& POOL_MANAGER_ACTIVE
;
968 int nr_idle
= pool
->nr_idle
+ managing
; /* manager is considered idle */
969 int nr_busy
= pool
->nr_workers
- nr_idle
;
971 return nr_idle
> 2 && (nr_idle
- 2) * MAX_IDLE_WORKERS_RATIO
>= nr_busy
;
975 * worker_set_flags - set worker flags and adjust nr_running accordingly
977 * @flags: flags to set
979 * Set @flags in @worker->flags and adjust nr_running accordingly.
981 static inline void worker_set_flags(struct worker
*worker
, unsigned int flags
)
983 struct worker_pool
*pool
= worker
->pool
;
985 lockdep_assert_held(&pool
->lock
);
987 /* If transitioning into NOT_RUNNING, adjust nr_running. */
988 if ((flags
& WORKER_NOT_RUNNING
) &&
989 !(worker
->flags
& WORKER_NOT_RUNNING
)) {
993 worker
->flags
|= flags
;
997 * worker_clr_flags - clear worker flags and adjust nr_running accordingly
999 * @flags: flags to clear
1001 * Clear @flags in @worker->flags and adjust nr_running accordingly.
1003 static inline void worker_clr_flags(struct worker
*worker
, unsigned int flags
)
1005 struct worker_pool
*pool
= worker
->pool
;
1006 unsigned int oflags
= worker
->flags
;
1008 lockdep_assert_held(&pool
->lock
);
1010 worker
->flags
&= ~flags
;
1013 * If transitioning out of NOT_RUNNING, increment nr_running. Note
1014 * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask
1015 * of multiple flags, not a single flag.
1017 if ((flags
& WORKER_NOT_RUNNING
) && (oflags
& WORKER_NOT_RUNNING
))
1018 if (!(worker
->flags
& WORKER_NOT_RUNNING
))
1022 /* Return the first idle worker. Called with pool->lock held. */
1023 static struct worker
*first_idle_worker(struct worker_pool
*pool
)
1025 if (unlikely(list_empty(&pool
->idle_list
)))
1028 return list_first_entry(&pool
->idle_list
, struct worker
, entry
);
1032 * worker_enter_idle - enter idle state
1033 * @worker: worker which is entering idle state
1035 * @worker is entering idle state. Update stats and idle timer if
1039 * raw_spin_lock_irq(pool->lock).
1041 static void worker_enter_idle(struct worker
*worker
)
1043 struct worker_pool
*pool
= worker
->pool
;
1045 if (WARN_ON_ONCE(worker
->flags
& WORKER_IDLE
) ||
1046 WARN_ON_ONCE(!list_empty(&worker
->entry
) &&
1047 (worker
->hentry
.next
|| worker
->hentry
.pprev
)))
1050 /* can't use worker_set_flags(), also called from create_worker() */
1051 worker
->flags
|= WORKER_IDLE
;
1053 worker
->last_active
= jiffies
;
1055 /* idle_list is LIFO */
1056 list_add(&worker
->entry
, &pool
->idle_list
);
1058 if (too_many_workers(pool
) && !timer_pending(&pool
->idle_timer
))
1059 mod_timer(&pool
->idle_timer
, jiffies
+ IDLE_WORKER_TIMEOUT
);
1061 /* Sanity check nr_running. */
1062 WARN_ON_ONCE(pool
->nr_workers
== pool
->nr_idle
&& pool
->nr_running
);
1066 * worker_leave_idle - leave idle state
1067 * @worker: worker which is leaving idle state
1069 * @worker is leaving idle state. Update stats.
1072 * raw_spin_lock_irq(pool->lock).
1074 static void worker_leave_idle(struct worker
*worker
)
1076 struct worker_pool
*pool
= worker
->pool
;
1078 if (WARN_ON_ONCE(!(worker
->flags
& WORKER_IDLE
)))
1080 worker_clr_flags(worker
, WORKER_IDLE
);
1082 list_del_init(&worker
->entry
);
1086 * find_worker_executing_work - find worker which is executing a work
1087 * @pool: pool of interest
1088 * @work: work to find worker for
1090 * Find a worker which is executing @work on @pool by searching
1091 * @pool->busy_hash which is keyed by the address of @work. For a worker
1092 * to match, its current execution should match the address of @work and
1093 * its work function. This is to avoid unwanted dependency between
1094 * unrelated work executions through a work item being recycled while still
1097 * This is a bit tricky. A work item may be freed once its execution
1098 * starts and nothing prevents the freed area from being recycled for
1099 * another work item. If the same work item address ends up being reused
1100 * before the original execution finishes, workqueue will identify the
1101 * recycled work item as currently executing and make it wait until the
1102 * current execution finishes, introducing an unwanted dependency.
1104 * This function checks the work item address and work function to avoid
1105 * false positives. Note that this isn't complete as one may construct a
1106 * work function which can introduce dependency onto itself through a
1107 * recycled work item. Well, if somebody wants to shoot oneself in the
1108 * foot that badly, there's only so much we can do, and if such deadlock
1109 * actually occurs, it should be easy to locate the culprit work function.
1112 * raw_spin_lock_irq(pool->lock).
1115 * Pointer to worker which is executing @work if found, %NULL
1118 static struct worker
*find_worker_executing_work(struct worker_pool
*pool
,
1119 struct work_struct
*work
)
1121 struct worker
*worker
;
1123 hash_for_each_possible(pool
->busy_hash
, worker
, hentry
,
1124 (unsigned long)work
)
1125 if (worker
->current_work
== work
&&
1126 worker
->current_func
== work
->func
)
1133 * move_linked_works - move linked works to a list
1134 * @work: start of series of works to be scheduled
1135 * @head: target list to append @work to
1136 * @nextp: out parameter for nested worklist walking
1138 * Schedule linked works starting from @work to @head. Work series to be
1139 * scheduled starts at @work and includes any consecutive work with
1140 * WORK_STRUCT_LINKED set in its predecessor. See assign_work() for details on
1144 * raw_spin_lock_irq(pool->lock).
1146 static void move_linked_works(struct work_struct
*work
, struct list_head
*head
,
1147 struct work_struct
**nextp
)
1149 struct work_struct
*n
;
1152 * Linked worklist will always end before the end of the list,
1153 * use NULL for list head.
1155 list_for_each_entry_safe_from(work
, n
, NULL
, entry
) {
1156 list_move_tail(&work
->entry
, head
);
1157 if (!(*work_data_bits(work
) & WORK_STRUCT_LINKED
))
1162 * If we're already inside safe list traversal and have moved
1163 * multiple works to the scheduled queue, the next position
1164 * needs to be updated.
1171 * assign_work - assign a work item and its linked work items to a worker
1172 * @work: work to assign
1173 * @worker: worker to assign to
1174 * @nextp: out parameter for nested worklist walking
1176 * Assign @work and its linked work items to @worker. If @work is already being
1177 * executed by another worker in the same pool, it'll be punted there.
1179 * If @nextp is not NULL, it's updated to point to the next work of the last
1180 * scheduled work. This allows assign_work() to be nested inside
1181 * list_for_each_entry_safe().
1183 * Returns %true if @work was successfully assigned to @worker. %false if @work
1184 * was punted to another worker already executing it.
1186 static bool assign_work(struct work_struct
*work
, struct worker
*worker
,
1187 struct work_struct
**nextp
)
1189 struct worker_pool
*pool
= worker
->pool
;
1190 struct worker
*collision
;
1192 lockdep_assert_held(&pool
->lock
);
1195 * A single work shouldn't be executed concurrently by multiple workers.
1196 * __queue_work() ensures that @work doesn't jump to a different pool
1197 * while still running in the previous pool. Here, we should ensure that
1198 * @work is not executed concurrently by multiple workers from the same
1199 * pool. Check whether anyone is already processing the work. If so,
1200 * defer the work to the currently executing one.
1202 collision
= find_worker_executing_work(pool
, work
);
1203 if (unlikely(collision
)) {
1204 move_linked_works(work
, &collision
->scheduled
, nextp
);
1208 move_linked_works(work
, &worker
->scheduled
, nextp
);
1212 static struct irq_work
*bh_pool_irq_work(struct worker_pool
*pool
)
1214 int high
= pool
->attrs
->nice
== HIGHPRI_NICE_LEVEL
? 1 : 0;
1216 return &per_cpu(bh_pool_irq_works
, pool
->cpu
)[high
];
1219 static void kick_bh_pool(struct worker_pool
*pool
)
1222 /* see drain_dead_softirq_workfn() for BH_DRAINING */
1223 if (unlikely(pool
->cpu
!= smp_processor_id() &&
1224 !(pool
->flags
& POOL_BH_DRAINING
))) {
1225 irq_work_queue_on(bh_pool_irq_work(pool
), pool
->cpu
);
1229 if (pool
->attrs
->nice
== HIGHPRI_NICE_LEVEL
)
1230 raise_softirq_irqoff(HI_SOFTIRQ
);
1232 raise_softirq_irqoff(TASKLET_SOFTIRQ
);
1236 * kick_pool - wake up an idle worker if necessary
1237 * @pool: pool to kick
1239 * @pool may have pending work items. Wake up worker if necessary. Returns
1240 * whether a worker was woken up.
1242 static bool kick_pool(struct worker_pool
*pool
)
1244 struct worker
*worker
= first_idle_worker(pool
);
1245 struct task_struct
*p
;
1247 lockdep_assert_held(&pool
->lock
);
1249 if (!need_more_worker(pool
) || !worker
)
1252 if (pool
->flags
& POOL_BH
) {
1261 * Idle @worker is about to execute @work and waking up provides an
1262 * opportunity to migrate @worker at a lower cost by setting the task's
1263 * wake_cpu field. Let's see if we want to move @worker to improve
1264 * execution locality.
1266 * We're waking the worker that went idle the latest and there's some
1267 * chance that @worker is marked idle but hasn't gone off CPU yet. If
1268 * so, setting the wake_cpu won't do anything. As this is a best-effort
1269 * optimization and the race window is narrow, let's leave as-is for
1270 * now. If this becomes pronounced, we can skip over workers which are
1271 * still on cpu when picking an idle worker.
1273 * If @pool has non-strict affinity, @worker might have ended up outside
1274 * its affinity scope. Repatriate.
1276 if (!pool
->attrs
->affn_strict
&&
1277 !cpumask_test_cpu(p
->wake_cpu
, pool
->attrs
->__pod_cpumask
)) {
1278 struct work_struct
*work
= list_first_entry(&pool
->worklist
,
1279 struct work_struct
, entry
);
1280 p
->wake_cpu
= cpumask_any_distribute(pool
->attrs
->__pod_cpumask
);
1281 get_work_pwq(work
)->stats
[PWQ_STAT_REPATRIATED
]++;
1288 #ifdef CONFIG_WQ_CPU_INTENSIVE_REPORT
1291 * Concurrency-managed per-cpu work items that hog CPU for longer than
1292 * wq_cpu_intensive_thresh_us trigger the automatic CPU_INTENSIVE mechanism,
1293 * which prevents them from stalling other concurrency-managed work items. If a
1294 * work function keeps triggering this mechanism, it's likely that the work item
1295 * should be using an unbound workqueue instead.
1297 * wq_cpu_intensive_report() tracks work functions which trigger such conditions
1298 * and report them so that they can be examined and converted to use unbound
1299 * workqueues as appropriate. To avoid flooding the console, each violating work
1300 * function is tracked and reported with exponential backoff.
1302 #define WCI_MAX_ENTS 128
1307 struct hlist_node hash_node
;
1310 static struct wci_ent wci_ents
[WCI_MAX_ENTS
];
1311 static int wci_nr_ents
;
1312 static DEFINE_RAW_SPINLOCK(wci_lock
);
1313 static DEFINE_HASHTABLE(wci_hash
, ilog2(WCI_MAX_ENTS
));
1315 static struct wci_ent
*wci_find_ent(work_func_t func
)
1317 struct wci_ent
*ent
;
1319 hash_for_each_possible_rcu(wci_hash
, ent
, hash_node
,
1320 (unsigned long)func
) {
1321 if (ent
->func
== func
)
1327 static void wq_cpu_intensive_report(work_func_t func
)
1329 struct wci_ent
*ent
;
1332 ent
= wci_find_ent(func
);
1337 * Start reporting from the warning_thresh and back off
1340 cnt
= atomic64_inc_return_relaxed(&ent
->cnt
);
1341 if (wq_cpu_intensive_warning_thresh
&&
1342 cnt
>= wq_cpu_intensive_warning_thresh
&&
1343 is_power_of_2(cnt
+ 1 - wq_cpu_intensive_warning_thresh
))
1344 printk_deferred(KERN_WARNING
"workqueue: %ps hogged CPU for >%luus %llu times, consider switching to WQ_UNBOUND\n",
1345 ent
->func
, wq_cpu_intensive_thresh_us
,
1346 atomic64_read(&ent
->cnt
));
1351 * @func is a new violation. Allocate a new entry for it. If wcn_ents[]
1352 * is exhausted, something went really wrong and we probably made enough
1355 if (wci_nr_ents
>= WCI_MAX_ENTS
)
1358 raw_spin_lock(&wci_lock
);
1360 if (wci_nr_ents
>= WCI_MAX_ENTS
) {
1361 raw_spin_unlock(&wci_lock
);
1365 if (wci_find_ent(func
)) {
1366 raw_spin_unlock(&wci_lock
);
1370 ent
= &wci_ents
[wci_nr_ents
++];
1372 atomic64_set(&ent
->cnt
, 0);
1373 hash_add_rcu(wci_hash
, &ent
->hash_node
, (unsigned long)func
);
1375 raw_spin_unlock(&wci_lock
);
1380 #else /* CONFIG_WQ_CPU_INTENSIVE_REPORT */
1381 static void wq_cpu_intensive_report(work_func_t func
) {}
1382 #endif /* CONFIG_WQ_CPU_INTENSIVE_REPORT */
1385 * wq_worker_running - a worker is running again
1386 * @task: task waking up
1388 * This function is called when a worker returns from schedule()
1390 void wq_worker_running(struct task_struct
*task
)
1392 struct worker
*worker
= kthread_data(task
);
1394 if (!READ_ONCE(worker
->sleeping
))
1398 * If preempted by unbind_workers() between the WORKER_NOT_RUNNING check
1399 * and the nr_running increment below, we may ruin the nr_running reset
1400 * and leave with an unexpected pool->nr_running == 1 on the newly unbound
1401 * pool. Protect against such race.
1404 if (!(worker
->flags
& WORKER_NOT_RUNNING
))
1405 worker
->pool
->nr_running
++;
1409 * CPU intensive auto-detection cares about how long a work item hogged
1410 * CPU without sleeping. Reset the starting timestamp on wakeup.
1412 worker
->current_at
= worker
->task
->se
.sum_exec_runtime
;
1414 WRITE_ONCE(worker
->sleeping
, 0);
1418 * wq_worker_sleeping - a worker is going to sleep
1419 * @task: task going to sleep
1421 * This function is called from schedule() when a busy worker is
1424 void wq_worker_sleeping(struct task_struct
*task
)
1426 struct worker
*worker
= kthread_data(task
);
1427 struct worker_pool
*pool
;
1430 * Rescuers, which may not have all the fields set up like normal
1431 * workers, also reach here, let's not access anything before
1432 * checking NOT_RUNNING.
1434 if (worker
->flags
& WORKER_NOT_RUNNING
)
1437 pool
= worker
->pool
;
1439 /* Return if preempted before wq_worker_running() was reached */
1440 if (READ_ONCE(worker
->sleeping
))
1443 WRITE_ONCE(worker
->sleeping
, 1);
1444 raw_spin_lock_irq(&pool
->lock
);
1447 * Recheck in case unbind_workers() preempted us. We don't
1448 * want to decrement nr_running after the worker is unbound
1449 * and nr_running has been reset.
1451 if (worker
->flags
& WORKER_NOT_RUNNING
) {
1452 raw_spin_unlock_irq(&pool
->lock
);
1457 if (kick_pool(pool
))
1458 worker
->current_pwq
->stats
[PWQ_STAT_CM_WAKEUP
]++;
1460 raw_spin_unlock_irq(&pool
->lock
);
1464 * wq_worker_tick - a scheduler tick occurred while a kworker is running
1465 * @task: task currently running
1467 * Called from scheduler_tick(). We're in the IRQ context and the current
1468 * worker's fields which follow the 'K' locking rule can be accessed safely.
1470 void wq_worker_tick(struct task_struct
*task
)
1472 struct worker
*worker
= kthread_data(task
);
1473 struct pool_workqueue
*pwq
= worker
->current_pwq
;
1474 struct worker_pool
*pool
= worker
->pool
;
1479 pwq
->stats
[PWQ_STAT_CPU_TIME
] += TICK_USEC
;
1481 if (!wq_cpu_intensive_thresh_us
)
1485 * If the current worker is concurrency managed and hogged the CPU for
1486 * longer than wq_cpu_intensive_thresh_us, it's automatically marked
1487 * CPU_INTENSIVE to avoid stalling other concurrency-managed work items.
1489 * Set @worker->sleeping means that @worker is in the process of
1490 * switching out voluntarily and won't be contributing to
1491 * @pool->nr_running until it wakes up. As wq_worker_sleeping() also
1492 * decrements ->nr_running, setting CPU_INTENSIVE here can lead to
1493 * double decrements. The task is releasing the CPU anyway. Let's skip.
1494 * We probably want to make this prettier in the future.
1496 if ((worker
->flags
& WORKER_NOT_RUNNING
) || READ_ONCE(worker
->sleeping
) ||
1497 worker
->task
->se
.sum_exec_runtime
- worker
->current_at
<
1498 wq_cpu_intensive_thresh_us
* NSEC_PER_USEC
)
1501 raw_spin_lock(&pool
->lock
);
1503 worker_set_flags(worker
, WORKER_CPU_INTENSIVE
);
1504 wq_cpu_intensive_report(worker
->current_func
);
1505 pwq
->stats
[PWQ_STAT_CPU_INTENSIVE
]++;
1507 if (kick_pool(pool
))
1508 pwq
->stats
[PWQ_STAT_CM_WAKEUP
]++;
1510 raw_spin_unlock(&pool
->lock
);
1514 * wq_worker_last_func - retrieve worker's last work function
1515 * @task: Task to retrieve last work function of.
1517 * Determine the last function a worker executed. This is called from
1518 * the scheduler to get a worker's last known identity.
1521 * raw_spin_lock_irq(rq->lock)
1523 * This function is called during schedule() when a kworker is going
1524 * to sleep. It's used by psi to identify aggregation workers during
1525 * dequeuing, to allow periodic aggregation to shut-off when that
1526 * worker is the last task in the system or cgroup to go to sleep.
1528 * As this function doesn't involve any workqueue-related locking, it
1529 * only returns stable values when called from inside the scheduler's
1530 * queuing and dequeuing paths, when @task, which must be a kworker,
1531 * is guaranteed to not be processing any works.
1534 * The last work function %current executed as a worker, NULL if it
1535 * hasn't executed any work yet.
1537 work_func_t
wq_worker_last_func(struct task_struct
*task
)
1539 struct worker
*worker
= kthread_data(task
);
1541 return worker
->last_func
;
1545 * wq_node_nr_active - Determine wq_node_nr_active to use
1546 * @wq: workqueue of interest
1547 * @node: NUMA node, can be %NUMA_NO_NODE
1549 * Determine wq_node_nr_active to use for @wq on @node. Returns:
1551 * - %NULL for per-cpu workqueues as they don't need to use shared nr_active.
1553 * - node_nr_active[nr_node_ids] if @node is %NUMA_NO_NODE.
1555 * - Otherwise, node_nr_active[@node].
1557 static struct wq_node_nr_active
*wq_node_nr_active(struct workqueue_struct
*wq
,
1560 if (!(wq
->flags
& WQ_UNBOUND
))
1563 if (node
== NUMA_NO_NODE
)
1566 return wq
->node_nr_active
[node
];
1570 * wq_update_node_max_active - Update per-node max_actives to use
1571 * @wq: workqueue to update
1572 * @off_cpu: CPU that's going down, -1 if a CPU is not going down
1574 * Update @wq->node_nr_active[]->max. @wq must be unbound. max_active is
1575 * distributed among nodes according to the proportions of numbers of online
1576 * cpus. The result is always between @wq->min_active and max_active.
1578 static void wq_update_node_max_active(struct workqueue_struct
*wq
, int off_cpu
)
1580 struct cpumask
*effective
= unbound_effective_cpumask(wq
);
1581 int min_active
= READ_ONCE(wq
->min_active
);
1582 int max_active
= READ_ONCE(wq
->max_active
);
1583 int total_cpus
, node
;
1585 lockdep_assert_held(&wq
->mutex
);
1587 if (!wq_topo_initialized
)
1590 if (off_cpu
>= 0 && !cpumask_test_cpu(off_cpu
, effective
))
1593 total_cpus
= cpumask_weight_and(effective
, cpu_online_mask
);
1597 for_each_node(node
) {
1600 node_cpus
= cpumask_weight_and(effective
, cpumask_of_node(node
));
1601 if (off_cpu
>= 0 && cpu_to_node(off_cpu
) == node
)
1604 wq_node_nr_active(wq
, node
)->max
=
1605 clamp(DIV_ROUND_UP(max_active
* node_cpus
, total_cpus
),
1606 min_active
, max_active
);
1609 wq_node_nr_active(wq
, NUMA_NO_NODE
)->max
= min_active
;
1613 * get_pwq - get an extra reference on the specified pool_workqueue
1614 * @pwq: pool_workqueue to get
1616 * Obtain an extra reference on @pwq. The caller should guarantee that
1617 * @pwq has positive refcnt and be holding the matching pool->lock.
1619 static void get_pwq(struct pool_workqueue
*pwq
)
1621 lockdep_assert_held(&pwq
->pool
->lock
);
1622 WARN_ON_ONCE(pwq
->refcnt
<= 0);
1627 * put_pwq - put a pool_workqueue reference
1628 * @pwq: pool_workqueue to put
1630 * Drop a reference of @pwq. If its refcnt reaches zero, schedule its
1631 * destruction. The caller should be holding the matching pool->lock.
1633 static void put_pwq(struct pool_workqueue
*pwq
)
1635 lockdep_assert_held(&pwq
->pool
->lock
);
1636 if (likely(--pwq
->refcnt
))
1639 * @pwq can't be released under pool->lock, bounce to a dedicated
1640 * kthread_worker to avoid A-A deadlocks.
1642 kthread_queue_work(pwq_release_worker
, &pwq
->release_work
);
1646 * put_pwq_unlocked - put_pwq() with surrounding pool lock/unlock
1647 * @pwq: pool_workqueue to put (can be %NULL)
1649 * put_pwq() with locking. This function also allows %NULL @pwq.
1651 static void put_pwq_unlocked(struct pool_workqueue
*pwq
)
1655 * As both pwqs and pools are RCU protected, the
1656 * following lock operations are safe.
1658 raw_spin_lock_irq(&pwq
->pool
->lock
);
1660 raw_spin_unlock_irq(&pwq
->pool
->lock
);
1664 static bool pwq_is_empty(struct pool_workqueue
*pwq
)
1666 return !pwq
->nr_active
&& list_empty(&pwq
->inactive_works
);
1669 static void __pwq_activate_work(struct pool_workqueue
*pwq
,
1670 struct work_struct
*work
)
1672 unsigned long *wdb
= work_data_bits(work
);
1674 WARN_ON_ONCE(!(*wdb
& WORK_STRUCT_INACTIVE
));
1675 trace_workqueue_activate_work(work
);
1676 if (list_empty(&pwq
->pool
->worklist
))
1677 pwq
->pool
->watchdog_ts
= jiffies
;
1678 move_linked_works(work
, &pwq
->pool
->worklist
, NULL
);
1679 __clear_bit(WORK_STRUCT_INACTIVE_BIT
, wdb
);
1683 * pwq_activate_work - Activate a work item if inactive
1684 * @pwq: pool_workqueue @work belongs to
1685 * @work: work item to activate
1687 * Returns %true if activated. %false if already active.
1689 static bool pwq_activate_work(struct pool_workqueue
*pwq
,
1690 struct work_struct
*work
)
1692 struct worker_pool
*pool
= pwq
->pool
;
1693 struct wq_node_nr_active
*nna
;
1695 lockdep_assert_held(&pool
->lock
);
1697 if (!(*work_data_bits(work
) & WORK_STRUCT_INACTIVE
))
1700 nna
= wq_node_nr_active(pwq
->wq
, pool
->node
);
1702 atomic_inc(&nna
->nr
);
1705 __pwq_activate_work(pwq
, work
);
1709 static bool tryinc_node_nr_active(struct wq_node_nr_active
*nna
)
1711 int max
= READ_ONCE(nna
->max
);
1716 old
= atomic_read(&nna
->nr
);
1719 tmp
= atomic_cmpxchg_relaxed(&nna
->nr
, old
, old
+ 1);
1726 * pwq_tryinc_nr_active - Try to increment nr_active for a pwq
1727 * @pwq: pool_workqueue of interest
1728 * @fill: max_active may have increased, try to increase concurrency level
1730 * Try to increment nr_active for @pwq. Returns %true if an nr_active count is
1731 * successfully obtained. %false otherwise.
1733 static bool pwq_tryinc_nr_active(struct pool_workqueue
*pwq
, bool fill
)
1735 struct workqueue_struct
*wq
= pwq
->wq
;
1736 struct worker_pool
*pool
= pwq
->pool
;
1737 struct wq_node_nr_active
*nna
= wq_node_nr_active(wq
, pool
->node
);
1738 bool obtained
= false;
1740 lockdep_assert_held(&pool
->lock
);
1743 /* BH or per-cpu workqueue, pwq->nr_active is sufficient */
1744 obtained
= pwq
->nr_active
< READ_ONCE(wq
->max_active
);
1748 if (unlikely(pwq
->plugged
))
1752 * Unbound workqueue uses per-node shared nr_active $nna. If @pwq is
1753 * already waiting on $nna, pwq_dec_nr_active() will maintain the
1754 * concurrency level. Don't jump the line.
1756 * We need to ignore the pending test after max_active has increased as
1757 * pwq_dec_nr_active() can only maintain the concurrency level but not
1758 * increase it. This is indicated by @fill.
1760 if (!list_empty(&pwq
->pending_node
) && likely(!fill
))
1763 obtained
= tryinc_node_nr_active(nna
);
1768 * Lockless acquisition failed. Lock, add ourself to $nna->pending_pwqs
1769 * and try again. The smp_mb() is paired with the implied memory barrier
1770 * of atomic_dec_return() in pwq_dec_nr_active() to ensure that either
1771 * we see the decremented $nna->nr or they see non-empty
1772 * $nna->pending_pwqs.
1774 raw_spin_lock(&nna
->lock
);
1776 if (list_empty(&pwq
->pending_node
))
1777 list_add_tail(&pwq
->pending_node
, &nna
->pending_pwqs
);
1778 else if (likely(!fill
))
1783 obtained
= tryinc_node_nr_active(nna
);
1786 * If @fill, @pwq might have already been pending. Being spuriously
1787 * pending in cold paths doesn't affect anything. Let's leave it be.
1789 if (obtained
&& likely(!fill
))
1790 list_del_init(&pwq
->pending_node
);
1793 raw_spin_unlock(&nna
->lock
);
1801 * pwq_activate_first_inactive - Activate the first inactive work item on a pwq
1802 * @pwq: pool_workqueue of interest
1803 * @fill: max_active may have increased, try to increase concurrency level
1805 * Activate the first inactive work item of @pwq if available and allowed by
1808 * Returns %true if an inactive work item has been activated. %false if no
1809 * inactive work item is found or max_active limit is reached.
1811 static bool pwq_activate_first_inactive(struct pool_workqueue
*pwq
, bool fill
)
1813 struct work_struct
*work
=
1814 list_first_entry_or_null(&pwq
->inactive_works
,
1815 struct work_struct
, entry
);
1817 if (work
&& pwq_tryinc_nr_active(pwq
, fill
)) {
1818 __pwq_activate_work(pwq
, work
);
1826 * unplug_oldest_pwq - unplug the oldest pool_workqueue
1827 * @wq: workqueue_struct where its oldest pwq is to be unplugged
1829 * This function should only be called for ordered workqueues where only the
1830 * oldest pwq is unplugged, the others are plugged to suspend execution to
1831 * ensure proper work item ordering::
1833 * dfl_pwq --------------+ [P] - plugged
1836 * pwqs -> A -> B [P] -> C [P] (newest)
1842 * When the oldest pwq is drained and removed, this function should be called
1843 * to unplug the next oldest one to start its work item execution. Note that
1844 * pwq's are linked into wq->pwqs with the oldest first, so the first one in
1845 * the list is the oldest.
1847 static void unplug_oldest_pwq(struct workqueue_struct
*wq
)
1849 struct pool_workqueue
*pwq
;
1851 lockdep_assert_held(&wq
->mutex
);
1853 /* Caller should make sure that pwqs isn't empty before calling */
1854 pwq
= list_first_entry_or_null(&wq
->pwqs
, struct pool_workqueue
,
1856 raw_spin_lock_irq(&pwq
->pool
->lock
);
1858 pwq
->plugged
= false;
1859 if (pwq_activate_first_inactive(pwq
, true))
1860 kick_pool(pwq
->pool
);
1862 raw_spin_unlock_irq(&pwq
->pool
->lock
);
1866 * node_activate_pending_pwq - Activate a pending pwq on a wq_node_nr_active
1867 * @nna: wq_node_nr_active to activate a pending pwq for
1868 * @caller_pool: worker_pool the caller is locking
1870 * Activate a pwq in @nna->pending_pwqs. Called with @caller_pool locked.
1871 * @caller_pool may be unlocked and relocked to lock other worker_pools.
1873 static void node_activate_pending_pwq(struct wq_node_nr_active
*nna
,
1874 struct worker_pool
*caller_pool
)
1876 struct worker_pool
*locked_pool
= caller_pool
;
1877 struct pool_workqueue
*pwq
;
1878 struct work_struct
*work
;
1880 lockdep_assert_held(&caller_pool
->lock
);
1882 raw_spin_lock(&nna
->lock
);
1884 pwq
= list_first_entry_or_null(&nna
->pending_pwqs
,
1885 struct pool_workqueue
, pending_node
);
1890 * If @pwq is for a different pool than @locked_pool, we need to lock
1891 * @pwq->pool->lock. Let's trylock first. If unsuccessful, do the unlock
1892 * / lock dance. For that, we also need to release @nna->lock as it's
1893 * nested inside pool locks.
1895 if (pwq
->pool
!= locked_pool
) {
1896 raw_spin_unlock(&locked_pool
->lock
);
1897 locked_pool
= pwq
->pool
;
1898 if (!raw_spin_trylock(&locked_pool
->lock
)) {
1899 raw_spin_unlock(&nna
->lock
);
1900 raw_spin_lock(&locked_pool
->lock
);
1901 raw_spin_lock(&nna
->lock
);
1907 * $pwq may not have any inactive work items due to e.g. cancellations.
1908 * Drop it from pending_pwqs and see if there's another one.
1910 work
= list_first_entry_or_null(&pwq
->inactive_works
,
1911 struct work_struct
, entry
);
1913 list_del_init(&pwq
->pending_node
);
1918 * Acquire an nr_active count and activate the inactive work item. If
1919 * $pwq still has inactive work items, rotate it to the end of the
1920 * pending_pwqs so that we round-robin through them. This means that
1921 * inactive work items are not activated in queueing order which is fine
1922 * given that there has never been any ordering across different pwqs.
1924 if (likely(tryinc_node_nr_active(nna
))) {
1926 __pwq_activate_work(pwq
, work
);
1928 if (list_empty(&pwq
->inactive_works
))
1929 list_del_init(&pwq
->pending_node
);
1931 list_move_tail(&pwq
->pending_node
, &nna
->pending_pwqs
);
1933 /* if activating a foreign pool, make sure it's running */
1934 if (pwq
->pool
!= caller_pool
)
1935 kick_pool(pwq
->pool
);
1939 raw_spin_unlock(&nna
->lock
);
1940 if (locked_pool
!= caller_pool
) {
1941 raw_spin_unlock(&locked_pool
->lock
);
1942 raw_spin_lock(&caller_pool
->lock
);
1947 * pwq_dec_nr_active - Retire an active count
1948 * @pwq: pool_workqueue of interest
1950 * Decrement @pwq's nr_active and try to activate the first inactive work item.
1951 * For unbound workqueues, this function may temporarily drop @pwq->pool->lock.
1953 static void pwq_dec_nr_active(struct pool_workqueue
*pwq
)
1955 struct worker_pool
*pool
= pwq
->pool
;
1956 struct wq_node_nr_active
*nna
= wq_node_nr_active(pwq
->wq
, pool
->node
);
1958 lockdep_assert_held(&pool
->lock
);
1961 * @pwq->nr_active should be decremented for both percpu and unbound
1967 * For a percpu workqueue, it's simple. Just need to kick the first
1968 * inactive work item on @pwq itself.
1971 pwq_activate_first_inactive(pwq
, false);
1976 * If @pwq is for an unbound workqueue, it's more complicated because
1977 * multiple pwqs and pools may be sharing the nr_active count. When a
1978 * pwq needs to wait for an nr_active count, it puts itself on
1979 * $nna->pending_pwqs. The following atomic_dec_return()'s implied
1980 * memory barrier is paired with smp_mb() in pwq_tryinc_nr_active() to
1981 * guarantee that either we see non-empty pending_pwqs or they see
1982 * decremented $nna->nr.
1984 * $nna->max may change as CPUs come online/offline and @pwq->wq's
1985 * max_active gets updated. However, it is guaranteed to be equal to or
1986 * larger than @pwq->wq->min_active which is above zero unless freezing.
1987 * This maintains the forward progress guarantee.
1989 if (atomic_dec_return(&nna
->nr
) >= READ_ONCE(nna
->max
))
1992 if (!list_empty(&nna
->pending_pwqs
))
1993 node_activate_pending_pwq(nna
, pool
);
1997 * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight
1998 * @pwq: pwq of interest
1999 * @work_data: work_data of work which left the queue
2001 * A work either has completed or is removed from pending queue,
2002 * decrement nr_in_flight of its pwq and handle workqueue flushing.
2005 * For unbound workqueues, this function may temporarily drop @pwq->pool->lock
2006 * and thus should be called after all other state updates for the in-flight
2007 * work item is complete.
2010 * raw_spin_lock_irq(pool->lock).
2012 static void pwq_dec_nr_in_flight(struct pool_workqueue
*pwq
, unsigned long work_data
)
2014 int color
= get_work_color(work_data
);
2016 if (!(work_data
& WORK_STRUCT_INACTIVE
))
2017 pwq_dec_nr_active(pwq
);
2019 pwq
->nr_in_flight
[color
]--;
2021 /* is flush in progress and are we at the flushing tip? */
2022 if (likely(pwq
->flush_color
!= color
))
2025 /* are there still in-flight works? */
2026 if (pwq
->nr_in_flight
[color
])
2029 /* this pwq is done, clear flush_color */
2030 pwq
->flush_color
= -1;
2033 * If this was the last pwq, wake up the first flusher. It
2034 * will handle the rest.
2036 if (atomic_dec_and_test(&pwq
->wq
->nr_pwqs_to_flush
))
2037 complete(&pwq
->wq
->first_flusher
->done
);
2043 * try_to_grab_pending - steal work item from worklist and disable irq
2044 * @work: work item to steal
2045 * @cflags: %WORK_CANCEL_ flags
2046 * @irq_flags: place to store irq state
2048 * Try to grab PENDING bit of @work. This function can handle @work in any
2049 * stable state - idle, on timer or on worklist.
2053 * ======== ================================================================
2054 * 1 if @work was pending and we successfully stole PENDING
2055 * 0 if @work was idle and we claimed PENDING
2056 * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry
2057 * -ENOENT if someone else is canceling @work, this state may persist
2058 * for arbitrarily long
2059 * ======== ================================================================
2062 * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting
2063 * interrupted while holding PENDING and @work off queue, irq must be
2064 * disabled on entry. This, combined with delayed_work->timer being
2065 * irqsafe, ensures that we return -EAGAIN for finite short period of time.
2067 * On successful return, >= 0, irq is disabled and the caller is
2068 * responsible for releasing it using local_irq_restore(*@irq_flags).
2070 * This function is safe to call from any context including IRQ handler.
2072 static int try_to_grab_pending(struct work_struct
*work
, u32 cflags
,
2073 unsigned long *irq_flags
)
2075 struct worker_pool
*pool
;
2076 struct pool_workqueue
*pwq
;
2078 local_irq_save(*irq_flags
);
2080 /* try to steal the timer if it exists */
2081 if (cflags
& WORK_CANCEL_DELAYED
) {
2082 struct delayed_work
*dwork
= to_delayed_work(work
);
2085 * dwork->timer is irqsafe. If del_timer() fails, it's
2086 * guaranteed that the timer is not queued anywhere and not
2087 * running on the local CPU.
2089 if (likely(del_timer(&dwork
->timer
)))
2093 /* try to claim PENDING the normal way */
2094 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT
, work_data_bits(work
)))
2099 * The queueing is in progress, or it is already queued. Try to
2100 * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
2102 pool
= get_work_pool(work
);
2106 raw_spin_lock(&pool
->lock
);
2108 * work->data is guaranteed to point to pwq only while the work
2109 * item is queued on pwq->wq, and both updating work->data to point
2110 * to pwq on queueing and to pool on dequeueing are done under
2111 * pwq->pool->lock. This in turn guarantees that, if work->data
2112 * points to pwq which is associated with a locked pool, the work
2113 * item is currently queued on that pool.
2115 pwq
= get_work_pwq(work
);
2116 if (pwq
&& pwq
->pool
== pool
) {
2117 unsigned long work_data
;
2119 debug_work_deactivate(work
);
2122 * A cancelable inactive work item must be in the
2123 * pwq->inactive_works since a queued barrier can't be
2124 * canceled (see the comments in insert_wq_barrier()).
2126 * An inactive work item cannot be grabbed directly because
2127 * it might have linked barrier work items which, if left
2128 * on the inactive_works list, will confuse pwq->nr_active
2129 * management later on and cause stall. Make sure the work
2130 * item is activated before grabbing.
2132 pwq_activate_work(pwq
, work
);
2134 list_del_init(&work
->entry
);
2137 * work->data points to pwq iff queued. Let's point to pool. As
2138 * this destroys work->data needed by the next step, stash it.
2140 work_data
= *work_data_bits(work
);
2141 set_work_pool_and_keep_pending(work
, pool
->id
, 0);
2143 /* must be the last step, see the function comment */
2144 pwq_dec_nr_in_flight(pwq
, work_data
);
2146 raw_spin_unlock(&pool
->lock
);
2150 raw_spin_unlock(&pool
->lock
);
2153 local_irq_restore(*irq_flags
);
2154 if (work_is_canceling(work
))
2161 wait_queue_entry_t wait
;
2162 struct work_struct
*work
;
2165 static int cwt_wakefn(wait_queue_entry_t
*wait
, unsigned mode
, int sync
, void *key
)
2167 struct cwt_wait
*cwait
= container_of(wait
, struct cwt_wait
, wait
);
2169 if (cwait
->work
!= key
)
2171 return autoremove_wake_function(wait
, mode
, sync
, key
);
2175 * work_grab_pending - steal work item from worklist and disable irq
2176 * @work: work item to steal
2177 * @cflags: %WORK_CANCEL_ flags
2178 * @irq_flags: place to store IRQ state
2180 * Grab PENDING bit of @work. @work can be in any stable state - idle, on timer
2183 * Must be called in process context. IRQ is disabled on return with IRQ state
2184 * stored in *@irq_flags. The caller is responsible for re-enabling it using
2185 * local_irq_restore().
2187 * Returns %true if @work was pending. %false if idle.
2189 static bool work_grab_pending(struct work_struct
*work
, u32 cflags
,
2190 unsigned long *irq_flags
)
2192 struct cwt_wait cwait
;
2197 ret
= try_to_grab_pending(work
, cflags
, irq_flags
);
2198 if (likely(ret
>= 0))
2204 * Someone is already canceling. Wait for it to finish. flush_work()
2205 * doesn't work for PREEMPT_NONE because we may get woken up between
2206 * @work's completion and the other canceling task resuming and clearing
2207 * CANCELING - flush_work() will return false immediately as @work is no
2208 * longer busy, try_to_grab_pending() will return -ENOENT as @work is
2209 * still being canceled and the other canceling task won't be able to
2210 * clear CANCELING as we're hogging the CPU.
2212 * Let's wait for completion using a waitqueue. As this may lead to the
2213 * thundering herd problem, use a custom wake function which matches
2214 * @work along with exclusive wait and wakeup.
2216 init_wait(&cwait
.wait
);
2217 cwait
.wait
.func
= cwt_wakefn
;
2220 prepare_to_wait_exclusive(&wq_cancel_waitq
, &cwait
.wait
,
2221 TASK_UNINTERRUPTIBLE
);
2222 if (work_is_canceling(work
))
2224 finish_wait(&wq_cancel_waitq
, &cwait
.wait
);
2230 * insert_work - insert a work into a pool
2231 * @pwq: pwq @work belongs to
2232 * @work: work to insert
2233 * @head: insertion point
2234 * @extra_flags: extra WORK_STRUCT_* flags to set
2236 * Insert @work which belongs to @pwq after @head. @extra_flags is or'd to
2237 * work_struct flags.
2240 * raw_spin_lock_irq(pool->lock).
2242 static void insert_work(struct pool_workqueue
*pwq
, struct work_struct
*work
,
2243 struct list_head
*head
, unsigned int extra_flags
)
2245 debug_work_activate(work
);
2247 /* record the work call stack in order to print it in KASAN reports */
2248 kasan_record_aux_stack_noalloc(work
);
2250 /* we own @work, set data and link */
2251 set_work_pwq(work
, pwq
, extra_flags
);
2252 list_add_tail(&work
->entry
, head
);
2257 * Test whether @work is being queued from another work executing on the
2260 static bool is_chained_work(struct workqueue_struct
*wq
)
2262 struct worker
*worker
;
2264 worker
= current_wq_worker();
2266 * Return %true iff I'm a worker executing a work item on @wq. If
2267 * I'm @worker, it's safe to dereference it without locking.
2269 return worker
&& worker
->current_pwq
->wq
== wq
;
2273 * When queueing an unbound work item to a wq, prefer local CPU if allowed
2274 * by wq_unbound_cpumask. Otherwise, round robin among the allowed ones to
2275 * avoid perturbing sensitive tasks.
2277 static int wq_select_unbound_cpu(int cpu
)
2281 if (likely(!wq_debug_force_rr_cpu
)) {
2282 if (cpumask_test_cpu(cpu
, wq_unbound_cpumask
))
2285 pr_warn_once("workqueue: round-robin CPU selection forced, expect performance impact\n");
2288 new_cpu
= __this_cpu_read(wq_rr_cpu_last
);
2289 new_cpu
= cpumask_next_and(new_cpu
, wq_unbound_cpumask
, cpu_online_mask
);
2290 if (unlikely(new_cpu
>= nr_cpu_ids
)) {
2291 new_cpu
= cpumask_first_and(wq_unbound_cpumask
, cpu_online_mask
);
2292 if (unlikely(new_cpu
>= nr_cpu_ids
))
2295 __this_cpu_write(wq_rr_cpu_last
, new_cpu
);
2300 static void __queue_work(int cpu
, struct workqueue_struct
*wq
,
2301 struct work_struct
*work
)
2303 struct pool_workqueue
*pwq
;
2304 struct worker_pool
*last_pool
, *pool
;
2305 unsigned int work_flags
;
2306 unsigned int req_cpu
= cpu
;
2309 * While a work item is PENDING && off queue, a task trying to
2310 * steal the PENDING will busy-loop waiting for it to either get
2311 * queued or lose PENDING. Grabbing PENDING and queueing should
2312 * happen with IRQ disabled.
2314 lockdep_assert_irqs_disabled();
2317 * For a draining wq, only works from the same workqueue are
2318 * allowed. The __WQ_DESTROYING helps to spot the issue that
2319 * queues a new work item to a wq after destroy_workqueue(wq).
2321 if (unlikely(wq
->flags
& (__WQ_DESTROYING
| __WQ_DRAINING
) &&
2322 WARN_ON_ONCE(!is_chained_work(wq
))))
2326 /* pwq which will be used unless @work is executing elsewhere */
2327 if (req_cpu
== WORK_CPU_UNBOUND
) {
2328 if (wq
->flags
& WQ_UNBOUND
)
2329 cpu
= wq_select_unbound_cpu(raw_smp_processor_id());
2331 cpu
= raw_smp_processor_id();
2334 pwq
= rcu_dereference(*per_cpu_ptr(wq
->cpu_pwq
, cpu
));
2338 * If @work was previously on a different pool, it might still be
2339 * running there, in which case the work needs to be queued on that
2340 * pool to guarantee non-reentrancy.
2342 last_pool
= get_work_pool(work
);
2343 if (last_pool
&& last_pool
!= pool
) {
2344 struct worker
*worker
;
2346 raw_spin_lock(&last_pool
->lock
);
2348 worker
= find_worker_executing_work(last_pool
, work
);
2350 if (worker
&& worker
->current_pwq
->wq
== wq
) {
2351 pwq
= worker
->current_pwq
;
2353 WARN_ON_ONCE(pool
!= last_pool
);
2355 /* meh... not running there, queue here */
2356 raw_spin_unlock(&last_pool
->lock
);
2357 raw_spin_lock(&pool
->lock
);
2360 raw_spin_lock(&pool
->lock
);
2364 * pwq is determined and locked. For unbound pools, we could have raced
2365 * with pwq release and it could already be dead. If its refcnt is zero,
2366 * repeat pwq selection. Note that unbound pwqs never die without
2367 * another pwq replacing it in cpu_pwq or while work items are executing
2368 * on it, so the retrying is guaranteed to make forward-progress.
2370 if (unlikely(!pwq
->refcnt
)) {
2371 if (wq
->flags
& WQ_UNBOUND
) {
2372 raw_spin_unlock(&pool
->lock
);
2377 WARN_ONCE(true, "workqueue: per-cpu pwq for %s on cpu%d has 0 refcnt",
2381 /* pwq determined, queue */
2382 trace_workqueue_queue_work(req_cpu
, pwq
, work
);
2384 if (WARN_ON(!list_empty(&work
->entry
)))
2387 pwq
->nr_in_flight
[pwq
->work_color
]++;
2388 work_flags
= work_color_to_flags(pwq
->work_color
);
2391 * Limit the number of concurrently active work items to max_active.
2392 * @work must also queue behind existing inactive work items to maintain
2393 * ordering when max_active changes. See wq_adjust_max_active().
2395 if (list_empty(&pwq
->inactive_works
) && pwq_tryinc_nr_active(pwq
, false)) {
2396 if (list_empty(&pool
->worklist
))
2397 pool
->watchdog_ts
= jiffies
;
2399 trace_workqueue_activate_work(work
);
2400 insert_work(pwq
, work
, &pool
->worklist
, work_flags
);
2403 work_flags
|= WORK_STRUCT_INACTIVE
;
2404 insert_work(pwq
, work
, &pwq
->inactive_works
, work_flags
);
2408 raw_spin_unlock(&pool
->lock
);
2413 * queue_work_on - queue work on specific cpu
2414 * @cpu: CPU number to execute work on
2415 * @wq: workqueue to use
2416 * @work: work to queue
2418 * We queue the work to a specific CPU, the caller must ensure it
2419 * can't go away. Callers that fail to ensure that the specified
2420 * CPU cannot go away will execute on a randomly chosen CPU.
2421 * But note well that callers specifying a CPU that never has been
2422 * online will get a splat.
2424 * Return: %false if @work was already on a queue, %true otherwise.
2426 bool queue_work_on(int cpu
, struct workqueue_struct
*wq
,
2427 struct work_struct
*work
)
2430 unsigned long irq_flags
;
2432 local_irq_save(irq_flags
);
2434 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT
, work_data_bits(work
))) {
2435 __queue_work(cpu
, wq
, work
);
2439 local_irq_restore(irq_flags
);
2442 EXPORT_SYMBOL(queue_work_on
);
2445 * select_numa_node_cpu - Select a CPU based on NUMA node
2446 * @node: NUMA node ID that we want to select a CPU from
2448 * This function will attempt to find a "random" cpu available on a given
2449 * node. If there are no CPUs available on the given node it will return
2450 * WORK_CPU_UNBOUND indicating that we should just schedule to any
2451 * available CPU if we need to schedule this work.
2453 static int select_numa_node_cpu(int node
)
2457 /* Delay binding to CPU if node is not valid or online */
2458 if (node
< 0 || node
>= MAX_NUMNODES
|| !node_online(node
))
2459 return WORK_CPU_UNBOUND
;
2461 /* Use local node/cpu if we are already there */
2462 cpu
= raw_smp_processor_id();
2463 if (node
== cpu_to_node(cpu
))
2466 /* Use "random" otherwise know as "first" online CPU of node */
2467 cpu
= cpumask_any_and(cpumask_of_node(node
), cpu_online_mask
);
2469 /* If CPU is valid return that, otherwise just defer */
2470 return cpu
< nr_cpu_ids
? cpu
: WORK_CPU_UNBOUND
;
2474 * queue_work_node - queue work on a "random" cpu for a given NUMA node
2475 * @node: NUMA node that we are targeting the work for
2476 * @wq: workqueue to use
2477 * @work: work to queue
2479 * We queue the work to a "random" CPU within a given NUMA node. The basic
2480 * idea here is to provide a way to somehow associate work with a given
2483 * This function will only make a best effort attempt at getting this onto
2484 * the right NUMA node. If no node is requested or the requested node is
2485 * offline then we just fall back to standard queue_work behavior.
2487 * Currently the "random" CPU ends up being the first available CPU in the
2488 * intersection of cpu_online_mask and the cpumask of the node, unless we
2489 * are running on the node. In that case we just use the current CPU.
2491 * Return: %false if @work was already on a queue, %true otherwise.
2493 bool queue_work_node(int node
, struct workqueue_struct
*wq
,
2494 struct work_struct
*work
)
2496 unsigned long irq_flags
;
2500 * This current implementation is specific to unbound workqueues.
2501 * Specifically we only return the first available CPU for a given
2502 * node instead of cycling through individual CPUs within the node.
2504 * If this is used with a per-cpu workqueue then the logic in
2505 * workqueue_select_cpu_near would need to be updated to allow for
2506 * some round robin type logic.
2508 WARN_ON_ONCE(!(wq
->flags
& WQ_UNBOUND
));
2510 local_irq_save(irq_flags
);
2512 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT
, work_data_bits(work
))) {
2513 int cpu
= select_numa_node_cpu(node
);
2515 __queue_work(cpu
, wq
, work
);
2519 local_irq_restore(irq_flags
);
2522 EXPORT_SYMBOL_GPL(queue_work_node
);
2524 void delayed_work_timer_fn(struct timer_list
*t
)
2526 struct delayed_work
*dwork
= from_timer(dwork
, t
, timer
);
2528 /* should have been called from irqsafe timer with irq already off */
2529 __queue_work(dwork
->cpu
, dwork
->wq
, &dwork
->work
);
2531 EXPORT_SYMBOL(delayed_work_timer_fn
);
2533 static void __queue_delayed_work(int cpu
, struct workqueue_struct
*wq
,
2534 struct delayed_work
*dwork
, unsigned long delay
)
2536 struct timer_list
*timer
= &dwork
->timer
;
2537 struct work_struct
*work
= &dwork
->work
;
2540 WARN_ON_ONCE(timer
->function
!= delayed_work_timer_fn
);
2541 WARN_ON_ONCE(timer_pending(timer
));
2542 WARN_ON_ONCE(!list_empty(&work
->entry
));
2545 * If @delay is 0, queue @dwork->work immediately. This is for
2546 * both optimization and correctness. The earliest @timer can
2547 * expire is on the closest next tick and delayed_work users depend
2548 * on that there's no such delay when @delay is 0.
2551 __queue_work(cpu
, wq
, &dwork
->work
);
2557 timer
->expires
= jiffies
+ delay
;
2559 if (housekeeping_enabled(HK_TYPE_TIMER
)) {
2560 /* If the current cpu is a housekeeping cpu, use it. */
2561 cpu
= smp_processor_id();
2562 if (!housekeeping_test_cpu(cpu
, HK_TYPE_TIMER
))
2563 cpu
= housekeeping_any_cpu(HK_TYPE_TIMER
);
2564 add_timer_on(timer
, cpu
);
2566 if (likely(cpu
== WORK_CPU_UNBOUND
))
2567 add_timer_global(timer
);
2569 add_timer_on(timer
, cpu
);
2574 * queue_delayed_work_on - queue work on specific CPU after delay
2575 * @cpu: CPU number to execute work on
2576 * @wq: workqueue to use
2577 * @dwork: work to queue
2578 * @delay: number of jiffies to wait before queueing
2580 * Return: %false if @work was already on a queue, %true otherwise. If
2581 * @delay is zero and @dwork is idle, it will be scheduled for immediate
2584 bool queue_delayed_work_on(int cpu
, struct workqueue_struct
*wq
,
2585 struct delayed_work
*dwork
, unsigned long delay
)
2587 struct work_struct
*work
= &dwork
->work
;
2589 unsigned long irq_flags
;
2591 /* read the comment in __queue_work() */
2592 local_irq_save(irq_flags
);
2594 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT
, work_data_bits(work
))) {
2595 __queue_delayed_work(cpu
, wq
, dwork
, delay
);
2599 local_irq_restore(irq_flags
);
2602 EXPORT_SYMBOL(queue_delayed_work_on
);
2605 * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU
2606 * @cpu: CPU number to execute work on
2607 * @wq: workqueue to use
2608 * @dwork: work to queue
2609 * @delay: number of jiffies to wait before queueing
2611 * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise,
2612 * modify @dwork's timer so that it expires after @delay. If @delay is
2613 * zero, @work is guaranteed to be scheduled immediately regardless of its
2616 * Return: %false if @dwork was idle and queued, %true if @dwork was
2617 * pending and its timer was modified.
2619 * This function is safe to call from any context including IRQ handler.
2620 * See try_to_grab_pending() for details.
2622 bool mod_delayed_work_on(int cpu
, struct workqueue_struct
*wq
,
2623 struct delayed_work
*dwork
, unsigned long delay
)
2625 unsigned long irq_flags
;
2629 ret
= try_to_grab_pending(&dwork
->work
, WORK_CANCEL_DELAYED
,
2631 } while (unlikely(ret
== -EAGAIN
));
2633 if (likely(ret
>= 0)) {
2634 __queue_delayed_work(cpu
, wq
, dwork
, delay
);
2635 local_irq_restore(irq_flags
);
2638 /* -ENOENT from try_to_grab_pending() becomes %true */
2641 EXPORT_SYMBOL_GPL(mod_delayed_work_on
);
2643 static void rcu_work_rcufn(struct rcu_head
*rcu
)
2645 struct rcu_work
*rwork
= container_of(rcu
, struct rcu_work
, rcu
);
2647 /* read the comment in __queue_work() */
2648 local_irq_disable();
2649 __queue_work(WORK_CPU_UNBOUND
, rwork
->wq
, &rwork
->work
);
2654 * queue_rcu_work - queue work after a RCU grace period
2655 * @wq: workqueue to use
2656 * @rwork: work to queue
2658 * Return: %false if @rwork was already pending, %true otherwise. Note
2659 * that a full RCU grace period is guaranteed only after a %true return.
2660 * While @rwork is guaranteed to be executed after a %false return, the
2661 * execution may happen before a full RCU grace period has passed.
2663 bool queue_rcu_work(struct workqueue_struct
*wq
, struct rcu_work
*rwork
)
2665 struct work_struct
*work
= &rwork
->work
;
2667 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT
, work_data_bits(work
))) {
2669 call_rcu_hurry(&rwork
->rcu
, rcu_work_rcufn
);
2675 EXPORT_SYMBOL(queue_rcu_work
);
2677 static struct worker
*alloc_worker(int node
)
2679 struct worker
*worker
;
2681 worker
= kzalloc_node(sizeof(*worker
), GFP_KERNEL
, node
);
2683 INIT_LIST_HEAD(&worker
->entry
);
2684 INIT_LIST_HEAD(&worker
->scheduled
);
2685 INIT_LIST_HEAD(&worker
->node
);
2686 /* on creation a worker is in !idle && prep state */
2687 worker
->flags
= WORKER_PREP
;
2692 static cpumask_t
*pool_allowed_cpus(struct worker_pool
*pool
)
2694 if (pool
->cpu
< 0 && pool
->attrs
->affn_strict
)
2695 return pool
->attrs
->__pod_cpumask
;
2697 return pool
->attrs
->cpumask
;
2701 * worker_attach_to_pool() - attach a worker to a pool
2702 * @worker: worker to be attached
2703 * @pool: the target pool
2705 * Attach @worker to @pool. Once attached, the %WORKER_UNBOUND flag and
2706 * cpu-binding of @worker are kept coordinated with the pool across
2709 static void worker_attach_to_pool(struct worker
*worker
,
2710 struct worker_pool
*pool
)
2712 mutex_lock(&wq_pool_attach_mutex
);
2715 * The wq_pool_attach_mutex ensures %POOL_DISASSOCIATED remains stable
2716 * across this function. See the comments above the flag definition for
2717 * details. BH workers are, while per-CPU, always DISASSOCIATED.
2719 if (pool
->flags
& POOL_DISASSOCIATED
) {
2720 worker
->flags
|= WORKER_UNBOUND
;
2722 WARN_ON_ONCE(pool
->flags
& POOL_BH
);
2723 kthread_set_per_cpu(worker
->task
, pool
->cpu
);
2726 if (worker
->rescue_wq
)
2727 set_cpus_allowed_ptr(worker
->task
, pool_allowed_cpus(pool
));
2729 list_add_tail(&worker
->node
, &pool
->workers
);
2730 worker
->pool
= pool
;
2732 mutex_unlock(&wq_pool_attach_mutex
);
2736 * worker_detach_from_pool() - detach a worker from its pool
2737 * @worker: worker which is attached to its pool
2739 * Undo the attaching which had been done in worker_attach_to_pool(). The
2740 * caller worker shouldn't access to the pool after detached except it has
2741 * other reference to the pool.
2743 static void worker_detach_from_pool(struct worker
*worker
)
2745 struct worker_pool
*pool
= worker
->pool
;
2746 struct completion
*detach_completion
= NULL
;
2748 /* there is one permanent BH worker per CPU which should never detach */
2749 WARN_ON_ONCE(pool
->flags
& POOL_BH
);
2751 mutex_lock(&wq_pool_attach_mutex
);
2753 kthread_set_per_cpu(worker
->task
, -1);
2754 list_del(&worker
->node
);
2755 worker
->pool
= NULL
;
2757 if (list_empty(&pool
->workers
) && list_empty(&pool
->dying_workers
))
2758 detach_completion
= pool
->detach_completion
;
2759 mutex_unlock(&wq_pool_attach_mutex
);
2761 /* clear leftover flags without pool->lock after it is detached */
2762 worker
->flags
&= ~(WORKER_UNBOUND
| WORKER_REBOUND
);
2764 if (detach_completion
)
2765 complete(detach_completion
);
2769 * create_worker - create a new workqueue worker
2770 * @pool: pool the new worker will belong to
2772 * Create and start a new worker which is attached to @pool.
2775 * Might sleep. Does GFP_KERNEL allocations.
2778 * Pointer to the newly created worker.
2780 static struct worker
*create_worker(struct worker_pool
*pool
)
2782 struct worker
*worker
;
2786 /* ID is needed to determine kthread name */
2787 id
= ida_alloc(&pool
->worker_ida
, GFP_KERNEL
);
2789 pr_err_once("workqueue: Failed to allocate a worker ID: %pe\n",
2794 worker
= alloc_worker(pool
->node
);
2796 pr_err_once("workqueue: Failed to allocate a worker\n");
2802 if (!(pool
->flags
& POOL_BH
)) {
2804 snprintf(id_buf
, sizeof(id_buf
), "%d:%d%s", pool
->cpu
, id
,
2805 pool
->attrs
->nice
< 0 ? "H" : "");
2807 snprintf(id_buf
, sizeof(id_buf
), "u%d:%d", pool
->id
, id
);
2809 worker
->task
= kthread_create_on_node(worker_thread
, worker
,
2810 pool
->node
, "kworker/%s", id_buf
);
2811 if (IS_ERR(worker
->task
)) {
2812 if (PTR_ERR(worker
->task
) == -EINTR
) {
2813 pr_err("workqueue: Interrupted when creating a worker thread \"kworker/%s\"\n",
2816 pr_err_once("workqueue: Failed to create a worker thread: %pe",
2822 set_user_nice(worker
->task
, pool
->attrs
->nice
);
2823 kthread_bind_mask(worker
->task
, pool_allowed_cpus(pool
));
2826 /* successful, attach the worker to the pool */
2827 worker_attach_to_pool(worker
, pool
);
2829 /* start the newly created worker */
2830 raw_spin_lock_irq(&pool
->lock
);
2832 worker
->pool
->nr_workers
++;
2833 worker_enter_idle(worker
);
2836 * @worker is waiting on a completion in kthread() and will trigger hung
2837 * check if not woken up soon. As kick_pool() is noop if @pool is empty,
2838 * wake it up explicitly.
2841 wake_up_process(worker
->task
);
2843 raw_spin_unlock_irq(&pool
->lock
);
2848 ida_free(&pool
->worker_ida
, id
);
2853 static void unbind_worker(struct worker
*worker
)
2855 lockdep_assert_held(&wq_pool_attach_mutex
);
2857 kthread_set_per_cpu(worker
->task
, -1);
2858 if (cpumask_intersects(wq_unbound_cpumask
, cpu_active_mask
))
2859 WARN_ON_ONCE(set_cpus_allowed_ptr(worker
->task
, wq_unbound_cpumask
) < 0);
2861 WARN_ON_ONCE(set_cpus_allowed_ptr(worker
->task
, cpu_possible_mask
) < 0);
2864 static void wake_dying_workers(struct list_head
*cull_list
)
2866 struct worker
*worker
, *tmp
;
2868 list_for_each_entry_safe(worker
, tmp
, cull_list
, entry
) {
2869 list_del_init(&worker
->entry
);
2870 unbind_worker(worker
);
2872 * If the worker was somehow already running, then it had to be
2873 * in pool->idle_list when set_worker_dying() happened or we
2874 * wouldn't have gotten here.
2876 * Thus, the worker must either have observed the WORKER_DIE
2877 * flag, or have set its state to TASK_IDLE. Either way, the
2878 * below will be observed by the worker and is safe to do
2879 * outside of pool->lock.
2881 wake_up_process(worker
->task
);
2886 * set_worker_dying - Tag a worker for destruction
2887 * @worker: worker to be destroyed
2888 * @list: transfer worker away from its pool->idle_list and into list
2890 * Tag @worker for destruction and adjust @pool stats accordingly. The worker
2894 * raw_spin_lock_irq(pool->lock).
2896 static void set_worker_dying(struct worker
*worker
, struct list_head
*list
)
2898 struct worker_pool
*pool
= worker
->pool
;
2900 lockdep_assert_held(&pool
->lock
);
2901 lockdep_assert_held(&wq_pool_attach_mutex
);
2903 /* sanity check frenzy */
2904 if (WARN_ON(worker
->current_work
) ||
2905 WARN_ON(!list_empty(&worker
->scheduled
)) ||
2906 WARN_ON(!(worker
->flags
& WORKER_IDLE
)))
2912 worker
->flags
|= WORKER_DIE
;
2914 list_move(&worker
->entry
, list
);
2915 list_move(&worker
->node
, &pool
->dying_workers
);
2919 * idle_worker_timeout - check if some idle workers can now be deleted.
2920 * @t: The pool's idle_timer that just expired
2922 * The timer is armed in worker_enter_idle(). Note that it isn't disarmed in
2923 * worker_leave_idle(), as a worker flicking between idle and active while its
2924 * pool is at the too_many_workers() tipping point would cause too much timer
2925 * housekeeping overhead. Since IDLE_WORKER_TIMEOUT is long enough, we just let
2926 * it expire and re-evaluate things from there.
2928 static void idle_worker_timeout(struct timer_list
*t
)
2930 struct worker_pool
*pool
= from_timer(pool
, t
, idle_timer
);
2931 bool do_cull
= false;
2933 if (work_pending(&pool
->idle_cull_work
))
2936 raw_spin_lock_irq(&pool
->lock
);
2938 if (too_many_workers(pool
)) {
2939 struct worker
*worker
;
2940 unsigned long expires
;
2942 /* idle_list is kept in LIFO order, check the last one */
2943 worker
= list_entry(pool
->idle_list
.prev
, struct worker
, entry
);
2944 expires
= worker
->last_active
+ IDLE_WORKER_TIMEOUT
;
2945 do_cull
= !time_before(jiffies
, expires
);
2948 mod_timer(&pool
->idle_timer
, expires
);
2950 raw_spin_unlock_irq(&pool
->lock
);
2953 queue_work(system_unbound_wq
, &pool
->idle_cull_work
);
2957 * idle_cull_fn - cull workers that have been idle for too long.
2958 * @work: the pool's work for handling these idle workers
2960 * This goes through a pool's idle workers and gets rid of those that have been
2961 * idle for at least IDLE_WORKER_TIMEOUT seconds.
2963 * We don't want to disturb isolated CPUs because of a pcpu kworker being
2964 * culled, so this also resets worker affinity. This requires a sleepable
2965 * context, hence the split between timer callback and work item.
2967 static void idle_cull_fn(struct work_struct
*work
)
2969 struct worker_pool
*pool
= container_of(work
, struct worker_pool
, idle_cull_work
);
2970 LIST_HEAD(cull_list
);
2973 * Grabbing wq_pool_attach_mutex here ensures an already-running worker
2974 * cannot proceed beyong worker_detach_from_pool() in its self-destruct
2975 * path. This is required as a previously-preempted worker could run after
2976 * set_worker_dying() has happened but before wake_dying_workers() did.
2978 mutex_lock(&wq_pool_attach_mutex
);
2979 raw_spin_lock_irq(&pool
->lock
);
2981 while (too_many_workers(pool
)) {
2982 struct worker
*worker
;
2983 unsigned long expires
;
2985 worker
= list_entry(pool
->idle_list
.prev
, struct worker
, entry
);
2986 expires
= worker
->last_active
+ IDLE_WORKER_TIMEOUT
;
2988 if (time_before(jiffies
, expires
)) {
2989 mod_timer(&pool
->idle_timer
, expires
);
2993 set_worker_dying(worker
, &cull_list
);
2996 raw_spin_unlock_irq(&pool
->lock
);
2997 wake_dying_workers(&cull_list
);
2998 mutex_unlock(&wq_pool_attach_mutex
);
3001 static void send_mayday(struct work_struct
*work
)
3003 struct pool_workqueue
*pwq
= get_work_pwq(work
);
3004 struct workqueue_struct
*wq
= pwq
->wq
;
3006 lockdep_assert_held(&wq_mayday_lock
);
3011 /* mayday mayday mayday */
3012 if (list_empty(&pwq
->mayday_node
)) {
3014 * If @pwq is for an unbound wq, its base ref may be put at
3015 * any time due to an attribute change. Pin @pwq until the
3016 * rescuer is done with it.
3019 list_add_tail(&pwq
->mayday_node
, &wq
->maydays
);
3020 wake_up_process(wq
->rescuer
->task
);
3021 pwq
->stats
[PWQ_STAT_MAYDAY
]++;
3025 static void pool_mayday_timeout(struct timer_list
*t
)
3027 struct worker_pool
*pool
= from_timer(pool
, t
, mayday_timer
);
3028 struct work_struct
*work
;
3030 raw_spin_lock_irq(&pool
->lock
);
3031 raw_spin_lock(&wq_mayday_lock
); /* for wq->maydays */
3033 if (need_to_create_worker(pool
)) {
3035 * We've been trying to create a new worker but
3036 * haven't been successful. We might be hitting an
3037 * allocation deadlock. Send distress signals to
3040 list_for_each_entry(work
, &pool
->worklist
, entry
)
3044 raw_spin_unlock(&wq_mayday_lock
);
3045 raw_spin_unlock_irq(&pool
->lock
);
3047 mod_timer(&pool
->mayday_timer
, jiffies
+ MAYDAY_INTERVAL
);
3051 * maybe_create_worker - create a new worker if necessary
3052 * @pool: pool to create a new worker for
3054 * Create a new worker for @pool if necessary. @pool is guaranteed to
3055 * have at least one idle worker on return from this function. If
3056 * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is
3057 * sent to all rescuers with works scheduled on @pool to resolve
3058 * possible allocation deadlock.
3060 * On return, need_to_create_worker() is guaranteed to be %false and
3061 * may_start_working() %true.
3064 * raw_spin_lock_irq(pool->lock) which may be released and regrabbed
3065 * multiple times. Does GFP_KERNEL allocations. Called only from
3068 static void maybe_create_worker(struct worker_pool
*pool
)
3069 __releases(&pool
->lock
)
3070 __acquires(&pool
->lock
)
3073 raw_spin_unlock_irq(&pool
->lock
);
3075 /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */
3076 mod_timer(&pool
->mayday_timer
, jiffies
+ MAYDAY_INITIAL_TIMEOUT
);
3079 if (create_worker(pool
) || !need_to_create_worker(pool
))
3082 schedule_timeout_interruptible(CREATE_COOLDOWN
);
3084 if (!need_to_create_worker(pool
))
3088 del_timer_sync(&pool
->mayday_timer
);
3089 raw_spin_lock_irq(&pool
->lock
);
3091 * This is necessary even after a new worker was just successfully
3092 * created as @pool->lock was dropped and the new worker might have
3093 * already become busy.
3095 if (need_to_create_worker(pool
))
3100 * manage_workers - manage worker pool
3103 * Assume the manager role and manage the worker pool @worker belongs
3104 * to. At any given time, there can be only zero or one manager per
3105 * pool. The exclusion is handled automatically by this function.
3107 * The caller can safely start processing works on false return. On
3108 * true return, it's guaranteed that need_to_create_worker() is false
3109 * and may_start_working() is true.
3112 * raw_spin_lock_irq(pool->lock) which may be released and regrabbed
3113 * multiple times. Does GFP_KERNEL allocations.
3116 * %false if the pool doesn't need management and the caller can safely
3117 * start processing works, %true if management function was performed and
3118 * the conditions that the caller verified before calling the function may
3119 * no longer be true.
3121 static bool manage_workers(struct worker
*worker
)
3123 struct worker_pool
*pool
= worker
->pool
;
3125 if (pool
->flags
& POOL_MANAGER_ACTIVE
)
3128 pool
->flags
|= POOL_MANAGER_ACTIVE
;
3129 pool
->manager
= worker
;
3131 maybe_create_worker(pool
);
3133 pool
->manager
= NULL
;
3134 pool
->flags
&= ~POOL_MANAGER_ACTIVE
;
3135 rcuwait_wake_up(&manager_wait
);
3140 * process_one_work - process single work
3142 * @work: work to process
3144 * Process @work. This function contains all the logics necessary to
3145 * process a single work including synchronization against and
3146 * interaction with other workers on the same cpu, queueing and
3147 * flushing. As long as context requirement is met, any worker can
3148 * call this function to process a work.
3151 * raw_spin_lock_irq(pool->lock) which is released and regrabbed.
3153 static void process_one_work(struct worker
*worker
, struct work_struct
*work
)
3154 __releases(&pool
->lock
)
3155 __acquires(&pool
->lock
)
3157 struct pool_workqueue
*pwq
= get_work_pwq(work
);
3158 struct worker_pool
*pool
= worker
->pool
;
3159 unsigned long work_data
;
3160 int lockdep_start_depth
, rcu_start_depth
;
3161 bool bh_draining
= pool
->flags
& POOL_BH_DRAINING
;
3162 #ifdef CONFIG_LOCKDEP
3164 * It is permissible to free the struct work_struct from
3165 * inside the function that is called from it, this we need to
3166 * take into account for lockdep too. To avoid bogus "held
3167 * lock freed" warnings as well as problems when looking into
3168 * work->lockdep_map, make a copy and use that here.
3170 struct lockdep_map lockdep_map
;
3172 lockdep_copy_map(&lockdep_map
, &work
->lockdep_map
);
3174 /* ensure we're on the correct CPU */
3175 WARN_ON_ONCE(!(pool
->flags
& POOL_DISASSOCIATED
) &&
3176 raw_smp_processor_id() != pool
->cpu
);
3178 /* claim and dequeue */
3179 debug_work_deactivate(work
);
3180 hash_add(pool
->busy_hash
, &worker
->hentry
, (unsigned long)work
);
3181 worker
->current_work
= work
;
3182 worker
->current_func
= work
->func
;
3183 worker
->current_pwq
= pwq
;
3185 worker
->current_at
= worker
->task
->se
.sum_exec_runtime
;
3186 work_data
= *work_data_bits(work
);
3187 worker
->current_color
= get_work_color(work_data
);
3190 * Record wq name for cmdline and debug reporting, may get
3191 * overridden through set_worker_desc().
3193 strscpy(worker
->desc
, pwq
->wq
->name
, WORKER_DESC_LEN
);
3195 list_del_init(&work
->entry
);
3198 * CPU intensive works don't participate in concurrency management.
3199 * They're the scheduler's responsibility. This takes @worker out
3200 * of concurrency management and the next code block will chain
3201 * execution of the pending work items.
3203 if (unlikely(pwq
->wq
->flags
& WQ_CPU_INTENSIVE
))
3204 worker_set_flags(worker
, WORKER_CPU_INTENSIVE
);
3207 * Kick @pool if necessary. It's always noop for per-cpu worker pools
3208 * since nr_running would always be >= 1 at this point. This is used to
3209 * chain execution of the pending work items for WORKER_NOT_RUNNING
3210 * workers such as the UNBOUND and CPU_INTENSIVE ones.
3215 * Record the last pool and clear PENDING which should be the last
3216 * update to @work. Also, do this inside @pool->lock so that
3217 * PENDING and queued state changes happen together while IRQ is
3220 set_work_pool_and_clear_pending(work
, pool
->id
, 0);
3222 pwq
->stats
[PWQ_STAT_STARTED
]++;
3223 raw_spin_unlock_irq(&pool
->lock
);
3225 rcu_start_depth
= rcu_preempt_depth();
3226 lockdep_start_depth
= lockdep_depth(current
);
3227 /* see drain_dead_softirq_workfn() */
3229 lock_map_acquire(&pwq
->wq
->lockdep_map
);
3230 lock_map_acquire(&lockdep_map
);
3232 * Strictly speaking we should mark the invariant state without holding
3233 * any locks, that is, before these two lock_map_acquire()'s.
3235 * However, that would result in:
3242 * Which would create W1->C->W1 dependencies, even though there is no
3243 * actual deadlock possible. There are two solutions, using a
3244 * read-recursive acquire on the work(queue) 'locks', but this will then
3245 * hit the lockdep limitation on recursive locks, or simply discard
3248 * AFAICT there is no possible deadlock scenario between the
3249 * flush_work() and complete() primitives (except for single-threaded
3250 * workqueues), so hiding them isn't a problem.
3252 lockdep_invariant_state(true);
3253 trace_workqueue_execute_start(work
);
3254 worker
->current_func(work
);
3256 * While we must be careful to not use "work" after this, the trace
3257 * point will only record its address.
3259 trace_workqueue_execute_end(work
, worker
->current_func
);
3260 pwq
->stats
[PWQ_STAT_COMPLETED
]++;
3261 lock_map_release(&lockdep_map
);
3263 lock_map_release(&pwq
->wq
->lockdep_map
);
3265 if (unlikely((worker
->task
&& in_atomic()) ||
3266 lockdep_depth(current
) != lockdep_start_depth
||
3267 rcu_preempt_depth() != rcu_start_depth
)) {
3268 pr_err("BUG: workqueue leaked atomic, lock or RCU: %s[%d]\n"
3269 " preempt=0x%08x lock=%d->%d RCU=%d->%d workfn=%ps\n",
3270 current
->comm
, task_pid_nr(current
), preempt_count(),
3271 lockdep_start_depth
, lockdep_depth(current
),
3272 rcu_start_depth
, rcu_preempt_depth(),
3273 worker
->current_func
);
3274 debug_show_held_locks(current
);
3279 * The following prevents a kworker from hogging CPU on !PREEMPTION
3280 * kernels, where a requeueing work item waiting for something to
3281 * happen could deadlock with stop_machine as such work item could
3282 * indefinitely requeue itself while all other CPUs are trapped in
3283 * stop_machine. At the same time, report a quiescent RCU state so
3284 * the same condition doesn't freeze RCU.
3289 raw_spin_lock_irq(&pool
->lock
);
3292 * In addition to %WQ_CPU_INTENSIVE, @worker may also have been marked
3293 * CPU intensive by wq_worker_tick() if @work hogged CPU longer than
3294 * wq_cpu_intensive_thresh_us. Clear it.
3296 worker_clr_flags(worker
, WORKER_CPU_INTENSIVE
);
3298 /* tag the worker for identification in schedule() */
3299 worker
->last_func
= worker
->current_func
;
3301 /* we're done with it, release */
3302 hash_del(&worker
->hentry
);
3303 worker
->current_work
= NULL
;
3304 worker
->current_func
= NULL
;
3305 worker
->current_pwq
= NULL
;
3306 worker
->current_color
= INT_MAX
;
3308 /* must be the last step, see the function comment */
3309 pwq_dec_nr_in_flight(pwq
, work_data
);
3313 * process_scheduled_works - process scheduled works
3316 * Process all scheduled works. Please note that the scheduled list
3317 * may change while processing a work, so this function repeatedly
3318 * fetches a work from the top and executes it.
3321 * raw_spin_lock_irq(pool->lock) which may be released and regrabbed
3324 static void process_scheduled_works(struct worker
*worker
)
3326 struct work_struct
*work
;
3329 while ((work
= list_first_entry_or_null(&worker
->scheduled
,
3330 struct work_struct
, entry
))) {
3332 worker
->pool
->watchdog_ts
= jiffies
;
3335 process_one_work(worker
, work
);
3339 static void set_pf_worker(bool val
)
3341 mutex_lock(&wq_pool_attach_mutex
);
3343 current
->flags
|= PF_WQ_WORKER
;
3345 current
->flags
&= ~PF_WQ_WORKER
;
3346 mutex_unlock(&wq_pool_attach_mutex
);
3350 * worker_thread - the worker thread function
3353 * The worker thread function. All workers belong to a worker_pool -
3354 * either a per-cpu one or dynamic unbound one. These workers process all
3355 * work items regardless of their specific target workqueue. The only
3356 * exception is work items which belong to workqueues with a rescuer which
3357 * will be explained in rescuer_thread().
3361 static int worker_thread(void *__worker
)
3363 struct worker
*worker
= __worker
;
3364 struct worker_pool
*pool
= worker
->pool
;
3366 /* tell the scheduler that this is a workqueue worker */
3367 set_pf_worker(true);
3369 raw_spin_lock_irq(&pool
->lock
);
3371 /* am I supposed to die? */
3372 if (unlikely(worker
->flags
& WORKER_DIE
)) {
3373 raw_spin_unlock_irq(&pool
->lock
);
3374 set_pf_worker(false);
3376 set_task_comm(worker
->task
, "kworker/dying");
3377 ida_free(&pool
->worker_ida
, worker
->id
);
3378 worker_detach_from_pool(worker
);
3379 WARN_ON_ONCE(!list_empty(&worker
->entry
));
3384 worker_leave_idle(worker
);
3386 /* no more worker necessary? */
3387 if (!need_more_worker(pool
))
3390 /* do we need to manage? */
3391 if (unlikely(!may_start_working(pool
)) && manage_workers(worker
))
3395 * ->scheduled list can only be filled while a worker is
3396 * preparing to process a work or actually processing it.
3397 * Make sure nobody diddled with it while I was sleeping.
3399 WARN_ON_ONCE(!list_empty(&worker
->scheduled
));
3402 * Finish PREP stage. We're guaranteed to have at least one idle
3403 * worker or that someone else has already assumed the manager
3404 * role. This is where @worker starts participating in concurrency
3405 * management if applicable and concurrency management is restored
3406 * after being rebound. See rebind_workers() for details.
3408 worker_clr_flags(worker
, WORKER_PREP
| WORKER_REBOUND
);
3411 struct work_struct
*work
=
3412 list_first_entry(&pool
->worklist
,
3413 struct work_struct
, entry
);
3415 if (assign_work(work
, worker
, NULL
))
3416 process_scheduled_works(worker
);
3417 } while (keep_working(pool
));
3419 worker_set_flags(worker
, WORKER_PREP
);
3422 * pool->lock is held and there's no work to process and no need to
3423 * manage, sleep. Workers are woken up only while holding
3424 * pool->lock or from local cpu, so setting the current state
3425 * before releasing pool->lock is enough to prevent losing any
3428 worker_enter_idle(worker
);
3429 __set_current_state(TASK_IDLE
);
3430 raw_spin_unlock_irq(&pool
->lock
);
3436 * rescuer_thread - the rescuer thread function
3439 * Workqueue rescuer thread function. There's one rescuer for each
3440 * workqueue which has WQ_MEM_RECLAIM set.
3442 * Regular work processing on a pool may block trying to create a new
3443 * worker which uses GFP_KERNEL allocation which has slight chance of
3444 * developing into deadlock if some works currently on the same queue
3445 * need to be processed to satisfy the GFP_KERNEL allocation. This is
3446 * the problem rescuer solves.
3448 * When such condition is possible, the pool summons rescuers of all
3449 * workqueues which have works queued on the pool and let them process
3450 * those works so that forward progress can be guaranteed.
3452 * This should happen rarely.
3456 static int rescuer_thread(void *__rescuer
)
3458 struct worker
*rescuer
= __rescuer
;
3459 struct workqueue_struct
*wq
= rescuer
->rescue_wq
;
3462 set_user_nice(current
, RESCUER_NICE_LEVEL
);
3465 * Mark rescuer as worker too. As WORKER_PREP is never cleared, it
3466 * doesn't participate in concurrency management.
3468 set_pf_worker(true);
3470 set_current_state(TASK_IDLE
);
3473 * By the time the rescuer is requested to stop, the workqueue
3474 * shouldn't have any work pending, but @wq->maydays may still have
3475 * pwq(s) queued. This can happen by non-rescuer workers consuming
3476 * all the work items before the rescuer got to them. Go through
3477 * @wq->maydays processing before acting on should_stop so that the
3478 * list is always empty on exit.
3480 should_stop
= kthread_should_stop();
3482 /* see whether any pwq is asking for help */
3483 raw_spin_lock_irq(&wq_mayday_lock
);
3485 while (!list_empty(&wq
->maydays
)) {
3486 struct pool_workqueue
*pwq
= list_first_entry(&wq
->maydays
,
3487 struct pool_workqueue
, mayday_node
);
3488 struct worker_pool
*pool
= pwq
->pool
;
3489 struct work_struct
*work
, *n
;
3491 __set_current_state(TASK_RUNNING
);
3492 list_del_init(&pwq
->mayday_node
);
3494 raw_spin_unlock_irq(&wq_mayday_lock
);
3496 worker_attach_to_pool(rescuer
, pool
);
3498 raw_spin_lock_irq(&pool
->lock
);
3501 * Slurp in all works issued via this workqueue and
3504 WARN_ON_ONCE(!list_empty(&rescuer
->scheduled
));
3505 list_for_each_entry_safe(work
, n
, &pool
->worklist
, entry
) {
3506 if (get_work_pwq(work
) == pwq
&&
3507 assign_work(work
, rescuer
, &n
))
3508 pwq
->stats
[PWQ_STAT_RESCUED
]++;
3511 if (!list_empty(&rescuer
->scheduled
)) {
3512 process_scheduled_works(rescuer
);
3515 * The above execution of rescued work items could
3516 * have created more to rescue through
3517 * pwq_activate_first_inactive() or chained
3518 * queueing. Let's put @pwq back on mayday list so
3519 * that such back-to-back work items, which may be
3520 * being used to relieve memory pressure, don't
3521 * incur MAYDAY_INTERVAL delay inbetween.
3523 if (pwq
->nr_active
&& need_to_create_worker(pool
)) {
3524 raw_spin_lock(&wq_mayday_lock
);
3526 * Queue iff we aren't racing destruction
3527 * and somebody else hasn't queued it already.
3529 if (wq
->rescuer
&& list_empty(&pwq
->mayday_node
)) {
3531 list_add_tail(&pwq
->mayday_node
, &wq
->maydays
);
3533 raw_spin_unlock(&wq_mayday_lock
);
3538 * Put the reference grabbed by send_mayday(). @pool won't
3539 * go away while we're still attached to it.
3544 * Leave this pool. Notify regular workers; otherwise, we end up
3545 * with 0 concurrency and stalling the execution.
3549 raw_spin_unlock_irq(&pool
->lock
);
3551 worker_detach_from_pool(rescuer
);
3553 raw_spin_lock_irq(&wq_mayday_lock
);
3556 raw_spin_unlock_irq(&wq_mayday_lock
);
3559 __set_current_state(TASK_RUNNING
);
3560 set_pf_worker(false);
3564 /* rescuers should never participate in concurrency management */
3565 WARN_ON_ONCE(!(rescuer
->flags
& WORKER_NOT_RUNNING
));
3570 static void bh_worker(struct worker
*worker
)
3572 struct worker_pool
*pool
= worker
->pool
;
3573 int nr_restarts
= BH_WORKER_RESTARTS
;
3574 unsigned long end
= jiffies
+ BH_WORKER_JIFFIES
;
3576 raw_spin_lock_irq(&pool
->lock
);
3577 worker_leave_idle(worker
);
3580 * This function follows the structure of worker_thread(). See there for
3581 * explanations on each step.
3583 if (!need_more_worker(pool
))
3586 WARN_ON_ONCE(!list_empty(&worker
->scheduled
));
3587 worker_clr_flags(worker
, WORKER_PREP
| WORKER_REBOUND
);
3590 struct work_struct
*work
=
3591 list_first_entry(&pool
->worklist
,
3592 struct work_struct
, entry
);
3594 if (assign_work(work
, worker
, NULL
))
3595 process_scheduled_works(worker
);
3596 } while (keep_working(pool
) &&
3597 --nr_restarts
&& time_before(jiffies
, end
));
3599 worker_set_flags(worker
, WORKER_PREP
);
3601 worker_enter_idle(worker
);
3603 raw_spin_unlock_irq(&pool
->lock
);
3607 * TODO: Convert all tasklet users to workqueue and use softirq directly.
3609 * This is currently called from tasklet[_hi]action() and thus is also called
3610 * whenever there are tasklets to run. Let's do an early exit if there's nothing
3611 * queued. Once conversion from tasklet is complete, the need_more_worker() test
3614 * After full conversion, we'll add worker->softirq_action, directly use the
3615 * softirq action and obtain the worker pointer from the softirq_action pointer.
3617 void workqueue_softirq_action(bool highpri
)
3619 struct worker_pool
*pool
=
3620 &per_cpu(bh_worker_pools
, smp_processor_id())[highpri
];
3621 if (need_more_worker(pool
))
3622 bh_worker(list_first_entry(&pool
->workers
, struct worker
, node
));
3625 struct wq_drain_dead_softirq_work
{
3626 struct work_struct work
;
3627 struct worker_pool
*pool
;
3628 struct completion done
;
3631 static void drain_dead_softirq_workfn(struct work_struct
*work
)
3633 struct wq_drain_dead_softirq_work
*dead_work
=
3634 container_of(work
, struct wq_drain_dead_softirq_work
, work
);
3635 struct worker_pool
*pool
= dead_work
->pool
;
3639 * @pool's CPU is dead and we want to execute its still pending work
3640 * items from this BH work item which is running on a different CPU. As
3641 * its CPU is dead, @pool can't be kicked and, as work execution path
3642 * will be nested, a lockdep annotation needs to be suppressed. Mark
3643 * @pool with %POOL_BH_DRAINING for the special treatments.
3645 raw_spin_lock_irq(&pool
->lock
);
3646 pool
->flags
|= POOL_BH_DRAINING
;
3647 raw_spin_unlock_irq(&pool
->lock
);
3649 bh_worker(list_first_entry(&pool
->workers
, struct worker
, node
));
3651 raw_spin_lock_irq(&pool
->lock
);
3652 pool
->flags
&= ~POOL_BH_DRAINING
;
3653 repeat
= need_more_worker(pool
);
3654 raw_spin_unlock_irq(&pool
->lock
);
3657 * bh_worker() might hit consecutive execution limit and bail. If there
3658 * still are pending work items, reschedule self and return so that we
3659 * don't hog this CPU's BH.
3662 if (pool
->attrs
->nice
== HIGHPRI_NICE_LEVEL
)
3663 queue_work(system_bh_highpri_wq
, work
);
3665 queue_work(system_bh_wq
, work
);
3667 complete(&dead_work
->done
);
3672 * @cpu is dead. Drain the remaining BH work items on the current CPU. It's
3673 * possible to allocate dead_work per CPU and avoid flushing. However, then we
3674 * have to worry about draining overlapping with CPU coming back online or
3675 * nesting (one CPU's dead_work queued on another CPU which is also dead and so
3676 * on). Let's keep it simple and drain them synchronously. These are BH work
3677 * items which shouldn't be requeued on the same pool. Shouldn't take long.
3679 void workqueue_softirq_dead(unsigned int cpu
)
3683 for (i
= 0; i
< NR_STD_WORKER_POOLS
; i
++) {
3684 struct worker_pool
*pool
= &per_cpu(bh_worker_pools
, cpu
)[i
];
3685 struct wq_drain_dead_softirq_work dead_work
;
3687 if (!need_more_worker(pool
))
3690 INIT_WORK(&dead_work
.work
, drain_dead_softirq_workfn
);
3691 dead_work
.pool
= pool
;
3692 init_completion(&dead_work
.done
);
3694 if (pool
->attrs
->nice
== HIGHPRI_NICE_LEVEL
)
3695 queue_work(system_bh_highpri_wq
, &dead_work
.work
);
3697 queue_work(system_bh_wq
, &dead_work
.work
);
3699 wait_for_completion(&dead_work
.done
);
3704 * check_flush_dependency - check for flush dependency sanity
3705 * @target_wq: workqueue being flushed
3706 * @target_work: work item being flushed (NULL for workqueue flushes)
3708 * %current is trying to flush the whole @target_wq or @target_work on it.
3709 * If @target_wq doesn't have %WQ_MEM_RECLAIM, verify that %current is not
3710 * reclaiming memory or running on a workqueue which doesn't have
3711 * %WQ_MEM_RECLAIM as that can break forward-progress guarantee leading to
3714 static void check_flush_dependency(struct workqueue_struct
*target_wq
,
3715 struct work_struct
*target_work
)
3717 work_func_t target_func
= target_work
? target_work
->func
: NULL
;
3718 struct worker
*worker
;
3720 if (target_wq
->flags
& WQ_MEM_RECLAIM
)
3723 worker
= current_wq_worker();
3725 WARN_ONCE(current
->flags
& PF_MEMALLOC
,
3726 "workqueue: PF_MEMALLOC task %d(%s) is flushing !WQ_MEM_RECLAIM %s:%ps",
3727 current
->pid
, current
->comm
, target_wq
->name
, target_func
);
3728 WARN_ONCE(worker
&& ((worker
->current_pwq
->wq
->flags
&
3729 (WQ_MEM_RECLAIM
| __WQ_LEGACY
)) == WQ_MEM_RECLAIM
),
3730 "workqueue: WQ_MEM_RECLAIM %s:%ps is flushing !WQ_MEM_RECLAIM %s:%ps",
3731 worker
->current_pwq
->wq
->name
, worker
->current_func
,
3732 target_wq
->name
, target_func
);
3736 struct work_struct work
;
3737 struct completion done
;
3738 struct task_struct
*task
; /* purely informational */
3741 static void wq_barrier_func(struct work_struct
*work
)
3743 struct wq_barrier
*barr
= container_of(work
, struct wq_barrier
, work
);
3744 complete(&barr
->done
);
3748 * insert_wq_barrier - insert a barrier work
3749 * @pwq: pwq to insert barrier into
3750 * @barr: wq_barrier to insert
3751 * @target: target work to attach @barr to
3752 * @worker: worker currently executing @target, NULL if @target is not executing
3754 * @barr is linked to @target such that @barr is completed only after
3755 * @target finishes execution. Please note that the ordering
3756 * guarantee is observed only with respect to @target and on the local
3759 * Currently, a queued barrier can't be canceled. This is because
3760 * try_to_grab_pending() can't determine whether the work to be
3761 * grabbed is at the head of the queue and thus can't clear LINKED
3762 * flag of the previous work while there must be a valid next work
3763 * after a work with LINKED flag set.
3765 * Note that when @worker is non-NULL, @target may be modified
3766 * underneath us, so we can't reliably determine pwq from @target.
3769 * raw_spin_lock_irq(pool->lock).
3771 static void insert_wq_barrier(struct pool_workqueue
*pwq
,
3772 struct wq_barrier
*barr
,
3773 struct work_struct
*target
, struct worker
*worker
)
3775 static __maybe_unused
struct lock_class_key bh_key
, thr_key
;
3776 unsigned int work_flags
= 0;
3777 unsigned int work_color
;
3778 struct list_head
*head
;
3781 * debugobject calls are safe here even with pool->lock locked
3782 * as we know for sure that this will not trigger any of the
3783 * checks and call back into the fixup functions where we
3786 * BH and threaded workqueues need separate lockdep keys to avoid
3787 * spuriously triggering "inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W}
3790 INIT_WORK_ONSTACK_KEY(&barr
->work
, wq_barrier_func
,
3791 (pwq
->wq
->flags
& WQ_BH
) ? &bh_key
: &thr_key
);
3792 __set_bit(WORK_STRUCT_PENDING_BIT
, work_data_bits(&barr
->work
));
3794 init_completion_map(&barr
->done
, &target
->lockdep_map
);
3796 barr
->task
= current
;
3798 /* The barrier work item does not participate in nr_active. */
3799 work_flags
|= WORK_STRUCT_INACTIVE
;
3802 * If @target is currently being executed, schedule the
3803 * barrier to the worker; otherwise, put it after @target.
3806 head
= worker
->scheduled
.next
;
3807 work_color
= worker
->current_color
;
3809 unsigned long *bits
= work_data_bits(target
);
3811 head
= target
->entry
.next
;
3812 /* there can already be other linked works, inherit and set */
3813 work_flags
|= *bits
& WORK_STRUCT_LINKED
;
3814 work_color
= get_work_color(*bits
);
3815 __set_bit(WORK_STRUCT_LINKED_BIT
, bits
);
3818 pwq
->nr_in_flight
[work_color
]++;
3819 work_flags
|= work_color_to_flags(work_color
);
3821 insert_work(pwq
, &barr
->work
, head
, work_flags
);
3825 * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing
3826 * @wq: workqueue being flushed
3827 * @flush_color: new flush color, < 0 for no-op
3828 * @work_color: new work color, < 0 for no-op
3830 * Prepare pwqs for workqueue flushing.
3832 * If @flush_color is non-negative, flush_color on all pwqs should be
3833 * -1. If no pwq has in-flight commands at the specified color, all
3834 * pwq->flush_color's stay at -1 and %false is returned. If any pwq
3835 * has in flight commands, its pwq->flush_color is set to
3836 * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq
3837 * wakeup logic is armed and %true is returned.
3839 * The caller should have initialized @wq->first_flusher prior to
3840 * calling this function with non-negative @flush_color. If
3841 * @flush_color is negative, no flush color update is done and %false
3844 * If @work_color is non-negative, all pwqs should have the same
3845 * work_color which is previous to @work_color and all will be
3846 * advanced to @work_color.
3849 * mutex_lock(wq->mutex).
3852 * %true if @flush_color >= 0 and there's something to flush. %false
3855 static bool flush_workqueue_prep_pwqs(struct workqueue_struct
*wq
,
3856 int flush_color
, int work_color
)
3859 struct pool_workqueue
*pwq
;
3861 if (flush_color
>= 0) {
3862 WARN_ON_ONCE(atomic_read(&wq
->nr_pwqs_to_flush
));
3863 atomic_set(&wq
->nr_pwqs_to_flush
, 1);
3866 for_each_pwq(pwq
, wq
) {
3867 struct worker_pool
*pool
= pwq
->pool
;
3869 raw_spin_lock_irq(&pool
->lock
);
3871 if (flush_color
>= 0) {
3872 WARN_ON_ONCE(pwq
->flush_color
!= -1);
3874 if (pwq
->nr_in_flight
[flush_color
]) {
3875 pwq
->flush_color
= flush_color
;
3876 atomic_inc(&wq
->nr_pwqs_to_flush
);
3881 if (work_color
>= 0) {
3882 WARN_ON_ONCE(work_color
!= work_next_color(pwq
->work_color
));
3883 pwq
->work_color
= work_color
;
3886 raw_spin_unlock_irq(&pool
->lock
);
3889 if (flush_color
>= 0 && atomic_dec_and_test(&wq
->nr_pwqs_to_flush
))
3890 complete(&wq
->first_flusher
->done
);
3895 static void touch_wq_lockdep_map(struct workqueue_struct
*wq
)
3897 #ifdef CONFIG_LOCKDEP
3898 if (wq
->flags
& WQ_BH
)
3901 lock_map_acquire(&wq
->lockdep_map
);
3902 lock_map_release(&wq
->lockdep_map
);
3904 if (wq
->flags
& WQ_BH
)
3909 static void touch_work_lockdep_map(struct work_struct
*work
,
3910 struct workqueue_struct
*wq
)
3912 #ifdef CONFIG_LOCKDEP
3913 if (wq
->flags
& WQ_BH
)
3916 lock_map_acquire(&work
->lockdep_map
);
3917 lock_map_release(&work
->lockdep_map
);
3919 if (wq
->flags
& WQ_BH
)
3925 * __flush_workqueue - ensure that any scheduled work has run to completion.
3926 * @wq: workqueue to flush
3928 * This function sleeps until all work items which were queued on entry
3929 * have finished execution, but it is not livelocked by new incoming ones.
3931 void __flush_workqueue(struct workqueue_struct
*wq
)
3933 struct wq_flusher this_flusher
= {
3934 .list
= LIST_HEAD_INIT(this_flusher
.list
),
3936 .done
= COMPLETION_INITIALIZER_ONSTACK_MAP(this_flusher
.done
, wq
->lockdep_map
),
3940 if (WARN_ON(!wq_online
))
3943 touch_wq_lockdep_map(wq
);
3945 mutex_lock(&wq
->mutex
);
3948 * Start-to-wait phase
3950 next_color
= work_next_color(wq
->work_color
);
3952 if (next_color
!= wq
->flush_color
) {
3954 * Color space is not full. The current work_color
3955 * becomes our flush_color and work_color is advanced
3958 WARN_ON_ONCE(!list_empty(&wq
->flusher_overflow
));
3959 this_flusher
.flush_color
= wq
->work_color
;
3960 wq
->work_color
= next_color
;
3962 if (!wq
->first_flusher
) {
3963 /* no flush in progress, become the first flusher */
3964 WARN_ON_ONCE(wq
->flush_color
!= this_flusher
.flush_color
);
3966 wq
->first_flusher
= &this_flusher
;
3968 if (!flush_workqueue_prep_pwqs(wq
, wq
->flush_color
,
3970 /* nothing to flush, done */
3971 wq
->flush_color
= next_color
;
3972 wq
->first_flusher
= NULL
;
3977 WARN_ON_ONCE(wq
->flush_color
== this_flusher
.flush_color
);
3978 list_add_tail(&this_flusher
.list
, &wq
->flusher_queue
);
3979 flush_workqueue_prep_pwqs(wq
, -1, wq
->work_color
);
3983 * Oops, color space is full, wait on overflow queue.
3984 * The next flush completion will assign us
3985 * flush_color and transfer to flusher_queue.
3987 list_add_tail(&this_flusher
.list
, &wq
->flusher_overflow
);
3990 check_flush_dependency(wq
, NULL
);
3992 mutex_unlock(&wq
->mutex
);
3994 wait_for_completion(&this_flusher
.done
);
3997 * Wake-up-and-cascade phase
3999 * First flushers are responsible for cascading flushes and
4000 * handling overflow. Non-first flushers can simply return.
4002 if (READ_ONCE(wq
->first_flusher
) != &this_flusher
)
4005 mutex_lock(&wq
->mutex
);
4007 /* we might have raced, check again with mutex held */
4008 if (wq
->first_flusher
!= &this_flusher
)
4011 WRITE_ONCE(wq
->first_flusher
, NULL
);
4013 WARN_ON_ONCE(!list_empty(&this_flusher
.list
));
4014 WARN_ON_ONCE(wq
->flush_color
!= this_flusher
.flush_color
);
4017 struct wq_flusher
*next
, *tmp
;
4019 /* complete all the flushers sharing the current flush color */
4020 list_for_each_entry_safe(next
, tmp
, &wq
->flusher_queue
, list
) {
4021 if (next
->flush_color
!= wq
->flush_color
)
4023 list_del_init(&next
->list
);
4024 complete(&next
->done
);
4027 WARN_ON_ONCE(!list_empty(&wq
->flusher_overflow
) &&
4028 wq
->flush_color
!= work_next_color(wq
->work_color
));
4030 /* this flush_color is finished, advance by one */
4031 wq
->flush_color
= work_next_color(wq
->flush_color
);
4033 /* one color has been freed, handle overflow queue */
4034 if (!list_empty(&wq
->flusher_overflow
)) {
4036 * Assign the same color to all overflowed
4037 * flushers, advance work_color and append to
4038 * flusher_queue. This is the start-to-wait
4039 * phase for these overflowed flushers.
4041 list_for_each_entry(tmp
, &wq
->flusher_overflow
, list
)
4042 tmp
->flush_color
= wq
->work_color
;
4044 wq
->work_color
= work_next_color(wq
->work_color
);
4046 list_splice_tail_init(&wq
->flusher_overflow
,
4047 &wq
->flusher_queue
);
4048 flush_workqueue_prep_pwqs(wq
, -1, wq
->work_color
);
4051 if (list_empty(&wq
->flusher_queue
)) {
4052 WARN_ON_ONCE(wq
->flush_color
!= wq
->work_color
);
4057 * Need to flush more colors. Make the next flusher
4058 * the new first flusher and arm pwqs.
4060 WARN_ON_ONCE(wq
->flush_color
== wq
->work_color
);
4061 WARN_ON_ONCE(wq
->flush_color
!= next
->flush_color
);
4063 list_del_init(&next
->list
);
4064 wq
->first_flusher
= next
;
4066 if (flush_workqueue_prep_pwqs(wq
, wq
->flush_color
, -1))
4070 * Meh... this color is already done, clear first
4071 * flusher and repeat cascading.
4073 wq
->first_flusher
= NULL
;
4077 mutex_unlock(&wq
->mutex
);
4079 EXPORT_SYMBOL(__flush_workqueue
);
4082 * drain_workqueue - drain a workqueue
4083 * @wq: workqueue to drain
4085 * Wait until the workqueue becomes empty. While draining is in progress,
4086 * only chain queueing is allowed. IOW, only currently pending or running
4087 * work items on @wq can queue further work items on it. @wq is flushed
4088 * repeatedly until it becomes empty. The number of flushing is determined
4089 * by the depth of chaining and should be relatively short. Whine if it
4092 void drain_workqueue(struct workqueue_struct
*wq
)
4094 unsigned int flush_cnt
= 0;
4095 struct pool_workqueue
*pwq
;
4098 * __queue_work() needs to test whether there are drainers, is much
4099 * hotter than drain_workqueue() and already looks at @wq->flags.
4100 * Use __WQ_DRAINING so that queue doesn't have to check nr_drainers.
4102 mutex_lock(&wq
->mutex
);
4103 if (!wq
->nr_drainers
++)
4104 wq
->flags
|= __WQ_DRAINING
;
4105 mutex_unlock(&wq
->mutex
);
4107 __flush_workqueue(wq
);
4109 mutex_lock(&wq
->mutex
);
4111 for_each_pwq(pwq
, wq
) {
4114 raw_spin_lock_irq(&pwq
->pool
->lock
);
4115 drained
= pwq_is_empty(pwq
);
4116 raw_spin_unlock_irq(&pwq
->pool
->lock
);
4121 if (++flush_cnt
== 10 ||
4122 (flush_cnt
% 100 == 0 && flush_cnt
<= 1000))
4123 pr_warn("workqueue %s: %s() isn't complete after %u tries\n",
4124 wq
->name
, __func__
, flush_cnt
);
4126 mutex_unlock(&wq
->mutex
);
4130 if (!--wq
->nr_drainers
)
4131 wq
->flags
&= ~__WQ_DRAINING
;
4132 mutex_unlock(&wq
->mutex
);
4134 EXPORT_SYMBOL_GPL(drain_workqueue
);
4136 static bool start_flush_work(struct work_struct
*work
, struct wq_barrier
*barr
,
4139 struct worker
*worker
= NULL
;
4140 struct worker_pool
*pool
;
4141 struct pool_workqueue
*pwq
;
4142 struct workqueue_struct
*wq
;
4147 pool
= get_work_pool(work
);
4153 raw_spin_lock_irq(&pool
->lock
);
4154 /* see the comment in try_to_grab_pending() with the same code */
4155 pwq
= get_work_pwq(work
);
4157 if (unlikely(pwq
->pool
!= pool
))
4160 worker
= find_worker_executing_work(pool
, work
);
4163 pwq
= worker
->current_pwq
;
4167 check_flush_dependency(wq
, work
);
4169 insert_wq_barrier(pwq
, barr
, work
, worker
);
4170 raw_spin_unlock_irq(&pool
->lock
);
4172 touch_work_lockdep_map(work
, wq
);
4175 * Force a lock recursion deadlock when using flush_work() inside a
4176 * single-threaded or rescuer equipped workqueue.
4178 * For single threaded workqueues the deadlock happens when the work
4179 * is after the work issuing the flush_work(). For rescuer equipped
4180 * workqueues the deadlock happens when the rescuer stalls, blocking
4183 if (!from_cancel
&& (wq
->saved_max_active
== 1 || wq
->rescuer
))
4184 touch_wq_lockdep_map(wq
);
4189 raw_spin_unlock_irq(&pool
->lock
);
4194 static bool __flush_work(struct work_struct
*work
, bool from_cancel
)
4196 struct wq_barrier barr
;
4198 if (WARN_ON(!wq_online
))
4201 if (WARN_ON(!work
->func
))
4204 if (start_flush_work(work
, &barr
, from_cancel
)) {
4205 wait_for_completion(&barr
.done
);
4206 destroy_work_on_stack(&barr
.work
);
4214 * flush_work - wait for a work to finish executing the last queueing instance
4215 * @work: the work to flush
4217 * Wait until @work has finished execution. @work is guaranteed to be idle
4218 * on return if it hasn't been requeued since flush started.
4221 * %true if flush_work() waited for the work to finish execution,
4222 * %false if it was already idle.
4224 bool flush_work(struct work_struct
*work
)
4226 return __flush_work(work
, false);
4228 EXPORT_SYMBOL_GPL(flush_work
);
4231 * flush_delayed_work - wait for a dwork to finish executing the last queueing
4232 * @dwork: the delayed work to flush
4234 * Delayed timer is cancelled and the pending work is queued for
4235 * immediate execution. Like flush_work(), this function only
4236 * considers the last queueing instance of @dwork.
4239 * %true if flush_work() waited for the work to finish execution,
4240 * %false if it was already idle.
4242 bool flush_delayed_work(struct delayed_work
*dwork
)
4244 local_irq_disable();
4245 if (del_timer_sync(&dwork
->timer
))
4246 __queue_work(dwork
->cpu
, dwork
->wq
, &dwork
->work
);
4248 return flush_work(&dwork
->work
);
4250 EXPORT_SYMBOL(flush_delayed_work
);
4253 * flush_rcu_work - wait for a rwork to finish executing the last queueing
4254 * @rwork: the rcu work to flush
4257 * %true if flush_rcu_work() waited for the work to finish execution,
4258 * %false if it was already idle.
4260 bool flush_rcu_work(struct rcu_work
*rwork
)
4262 if (test_bit(WORK_STRUCT_PENDING_BIT
, work_data_bits(&rwork
->work
))) {
4264 flush_work(&rwork
->work
);
4267 return flush_work(&rwork
->work
);
4270 EXPORT_SYMBOL(flush_rcu_work
);
4272 static bool __cancel_work(struct work_struct
*work
, u32 cflags
)
4274 unsigned long irq_flags
;
4278 ret
= try_to_grab_pending(work
, cflags
, &irq_flags
);
4279 } while (unlikely(ret
== -EAGAIN
));
4281 if (unlikely(ret
< 0))
4284 set_work_pool_and_clear_pending(work
, get_work_pool_id(work
), 0);
4285 local_irq_restore(irq_flags
);
4289 static bool __cancel_work_sync(struct work_struct
*work
, u32 cflags
)
4291 unsigned long irq_flags
;
4294 /* claim @work and tell other tasks trying to grab @work to back off */
4295 ret
= work_grab_pending(work
, cflags
, &irq_flags
);
4296 mark_work_canceling(work
);
4297 local_irq_restore(irq_flags
);
4300 * Skip __flush_work() during early boot when we know that @work isn't
4301 * executing. This allows canceling during early boot.
4304 __flush_work(work
, true);
4307 * smp_mb() at the end of set_work_pool_and_clear_pending() is paired
4308 * with prepare_to_wait() above so that either waitqueue_active() is
4309 * visible here or !work_is_canceling() is visible there.
4311 set_work_pool_and_clear_pending(work
, WORK_OFFQ_POOL_NONE
, 0);
4313 if (waitqueue_active(&wq_cancel_waitq
))
4314 __wake_up(&wq_cancel_waitq
, TASK_NORMAL
, 1, work
);
4320 * See cancel_delayed_work()
4322 bool cancel_work(struct work_struct
*work
)
4324 return __cancel_work(work
, 0);
4326 EXPORT_SYMBOL(cancel_work
);
4329 * cancel_work_sync - cancel a work and wait for it to finish
4330 * @work: the work to cancel
4332 * Cancel @work and wait for its execution to finish. This function
4333 * can be used even if the work re-queues itself or migrates to
4334 * another workqueue. On return from this function, @work is
4335 * guaranteed to be not pending or executing on any CPU.
4337 * cancel_work_sync(&delayed_work->work) must not be used for
4338 * delayed_work's. Use cancel_delayed_work_sync() instead.
4340 * The caller must ensure that the workqueue on which @work was last
4341 * queued can't be destroyed before this function returns.
4344 * %true if @work was pending, %false otherwise.
4346 bool cancel_work_sync(struct work_struct
*work
)
4348 return __cancel_work_sync(work
, 0);
4350 EXPORT_SYMBOL_GPL(cancel_work_sync
);
4353 * cancel_delayed_work - cancel a delayed work
4354 * @dwork: delayed_work to cancel
4356 * Kill off a pending delayed_work.
4358 * Return: %true if @dwork was pending and canceled; %false if it wasn't
4362 * The work callback function may still be running on return, unless
4363 * it returns %true and the work doesn't re-arm itself. Explicitly flush or
4364 * use cancel_delayed_work_sync() to wait on it.
4366 * This function is safe to call from any context including IRQ handler.
4368 bool cancel_delayed_work(struct delayed_work
*dwork
)
4370 return __cancel_work(&dwork
->work
, WORK_CANCEL_DELAYED
);
4372 EXPORT_SYMBOL(cancel_delayed_work
);
4375 * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish
4376 * @dwork: the delayed work cancel
4378 * This is cancel_work_sync() for delayed works.
4381 * %true if @dwork was pending, %false otherwise.
4383 bool cancel_delayed_work_sync(struct delayed_work
*dwork
)
4385 return __cancel_work_sync(&dwork
->work
, WORK_CANCEL_DELAYED
);
4387 EXPORT_SYMBOL(cancel_delayed_work_sync
);
4390 * schedule_on_each_cpu - execute a function synchronously on each online CPU
4391 * @func: the function to call
4393 * schedule_on_each_cpu() executes @func on each online CPU using the
4394 * system workqueue and blocks until all CPUs have completed.
4395 * schedule_on_each_cpu() is very slow.
4398 * 0 on success, -errno on failure.
4400 int schedule_on_each_cpu(work_func_t func
)
4403 struct work_struct __percpu
*works
;
4405 works
= alloc_percpu(struct work_struct
);
4411 for_each_online_cpu(cpu
) {
4412 struct work_struct
*work
= per_cpu_ptr(works
, cpu
);
4414 INIT_WORK(work
, func
);
4415 schedule_work_on(cpu
, work
);
4418 for_each_online_cpu(cpu
)
4419 flush_work(per_cpu_ptr(works
, cpu
));
4427 * execute_in_process_context - reliably execute the routine with user context
4428 * @fn: the function to execute
4429 * @ew: guaranteed storage for the execute work structure (must
4430 * be available when the work executes)
4432 * Executes the function immediately if process context is available,
4433 * otherwise schedules the function for delayed execution.
4435 * Return: 0 - function was executed
4436 * 1 - function was scheduled for execution
4438 int execute_in_process_context(work_func_t fn
, struct execute_work
*ew
)
4440 if (!in_interrupt()) {
4445 INIT_WORK(&ew
->work
, fn
);
4446 schedule_work(&ew
->work
);
4450 EXPORT_SYMBOL_GPL(execute_in_process_context
);
4453 * free_workqueue_attrs - free a workqueue_attrs
4454 * @attrs: workqueue_attrs to free
4456 * Undo alloc_workqueue_attrs().
4458 void free_workqueue_attrs(struct workqueue_attrs
*attrs
)
4461 free_cpumask_var(attrs
->cpumask
);
4462 free_cpumask_var(attrs
->__pod_cpumask
);
4468 * alloc_workqueue_attrs - allocate a workqueue_attrs
4470 * Allocate a new workqueue_attrs, initialize with default settings and
4473 * Return: The allocated new workqueue_attr on success. %NULL on failure.
4475 struct workqueue_attrs
*alloc_workqueue_attrs(void)
4477 struct workqueue_attrs
*attrs
;
4479 attrs
= kzalloc(sizeof(*attrs
), GFP_KERNEL
);
4482 if (!alloc_cpumask_var(&attrs
->cpumask
, GFP_KERNEL
))
4484 if (!alloc_cpumask_var(&attrs
->__pod_cpumask
, GFP_KERNEL
))
4487 cpumask_copy(attrs
->cpumask
, cpu_possible_mask
);
4488 attrs
->affn_scope
= WQ_AFFN_DFL
;
4491 free_workqueue_attrs(attrs
);
4495 static void copy_workqueue_attrs(struct workqueue_attrs
*to
,
4496 const struct workqueue_attrs
*from
)
4498 to
->nice
= from
->nice
;
4499 cpumask_copy(to
->cpumask
, from
->cpumask
);
4500 cpumask_copy(to
->__pod_cpumask
, from
->__pod_cpumask
);
4501 to
->affn_strict
= from
->affn_strict
;
4504 * Unlike hash and equality test, copying shouldn't ignore wq-only
4505 * fields as copying is used for both pool and wq attrs. Instead,
4506 * get_unbound_pool() explicitly clears the fields.
4508 to
->affn_scope
= from
->affn_scope
;
4509 to
->ordered
= from
->ordered
;
4513 * Some attrs fields are workqueue-only. Clear them for worker_pool's. See the
4514 * comments in 'struct workqueue_attrs' definition.
4516 static void wqattrs_clear_for_pool(struct workqueue_attrs
*attrs
)
4518 attrs
->affn_scope
= WQ_AFFN_NR_TYPES
;
4519 attrs
->ordered
= false;
4522 /* hash value of the content of @attr */
4523 static u32
wqattrs_hash(const struct workqueue_attrs
*attrs
)
4527 hash
= jhash_1word(attrs
->nice
, hash
);
4528 hash
= jhash(cpumask_bits(attrs
->cpumask
),
4529 BITS_TO_LONGS(nr_cpumask_bits
) * sizeof(long), hash
);
4530 hash
= jhash(cpumask_bits(attrs
->__pod_cpumask
),
4531 BITS_TO_LONGS(nr_cpumask_bits
) * sizeof(long), hash
);
4532 hash
= jhash_1word(attrs
->affn_strict
, hash
);
4536 /* content equality test */
4537 static bool wqattrs_equal(const struct workqueue_attrs
*a
,
4538 const struct workqueue_attrs
*b
)
4540 if (a
->nice
!= b
->nice
)
4542 if (!cpumask_equal(a
->cpumask
, b
->cpumask
))
4544 if (!cpumask_equal(a
->__pod_cpumask
, b
->__pod_cpumask
))
4546 if (a
->affn_strict
!= b
->affn_strict
)
4551 /* Update @attrs with actually available CPUs */
4552 static void wqattrs_actualize_cpumask(struct workqueue_attrs
*attrs
,
4553 const cpumask_t
*unbound_cpumask
)
4556 * Calculate the effective CPU mask of @attrs given @unbound_cpumask. If
4557 * @attrs->cpumask doesn't overlap with @unbound_cpumask, we fallback to
4560 cpumask_and(attrs
->cpumask
, attrs
->cpumask
, unbound_cpumask
);
4561 if (unlikely(cpumask_empty(attrs
->cpumask
)))
4562 cpumask_copy(attrs
->cpumask
, unbound_cpumask
);
4565 /* find wq_pod_type to use for @attrs */
4566 static const struct wq_pod_type
*
4567 wqattrs_pod_type(const struct workqueue_attrs
*attrs
)
4569 enum wq_affn_scope scope
;
4570 struct wq_pod_type
*pt
;
4572 /* to synchronize access to wq_affn_dfl */
4573 lockdep_assert_held(&wq_pool_mutex
);
4575 if (attrs
->affn_scope
== WQ_AFFN_DFL
)
4576 scope
= wq_affn_dfl
;
4578 scope
= attrs
->affn_scope
;
4580 pt
= &wq_pod_types
[scope
];
4582 if (!WARN_ON_ONCE(attrs
->affn_scope
== WQ_AFFN_NR_TYPES
) &&
4583 likely(pt
->nr_pods
))
4587 * Before workqueue_init_topology(), only SYSTEM is available which is
4588 * initialized in workqueue_init_early().
4590 pt
= &wq_pod_types
[WQ_AFFN_SYSTEM
];
4591 BUG_ON(!pt
->nr_pods
);
4596 * init_worker_pool - initialize a newly zalloc'd worker_pool
4597 * @pool: worker_pool to initialize
4599 * Initialize a newly zalloc'd @pool. It also allocates @pool->attrs.
4601 * Return: 0 on success, -errno on failure. Even on failure, all fields
4602 * inside @pool proper are initialized and put_unbound_pool() can be called
4603 * on @pool safely to release it.
4605 static int init_worker_pool(struct worker_pool
*pool
)
4607 raw_spin_lock_init(&pool
->lock
);
4610 pool
->node
= NUMA_NO_NODE
;
4611 pool
->flags
|= POOL_DISASSOCIATED
;
4612 pool
->watchdog_ts
= jiffies
;
4613 INIT_LIST_HEAD(&pool
->worklist
);
4614 INIT_LIST_HEAD(&pool
->idle_list
);
4615 hash_init(pool
->busy_hash
);
4617 timer_setup(&pool
->idle_timer
, idle_worker_timeout
, TIMER_DEFERRABLE
);
4618 INIT_WORK(&pool
->idle_cull_work
, idle_cull_fn
);
4620 timer_setup(&pool
->mayday_timer
, pool_mayday_timeout
, 0);
4622 INIT_LIST_HEAD(&pool
->workers
);
4623 INIT_LIST_HEAD(&pool
->dying_workers
);
4625 ida_init(&pool
->worker_ida
);
4626 INIT_HLIST_NODE(&pool
->hash_node
);
4629 /* shouldn't fail above this point */
4630 pool
->attrs
= alloc_workqueue_attrs();
4634 wqattrs_clear_for_pool(pool
->attrs
);
4639 #ifdef CONFIG_LOCKDEP
4640 static void wq_init_lockdep(struct workqueue_struct
*wq
)
4644 lockdep_register_key(&wq
->key
);
4645 lock_name
= kasprintf(GFP_KERNEL
, "%s%s", "(wq_completion)", wq
->name
);
4647 lock_name
= wq
->name
;
4649 wq
->lock_name
= lock_name
;
4650 lockdep_init_map(&wq
->lockdep_map
, lock_name
, &wq
->key
, 0);
4653 static void wq_unregister_lockdep(struct workqueue_struct
*wq
)
4655 lockdep_unregister_key(&wq
->key
);
4658 static void wq_free_lockdep(struct workqueue_struct
*wq
)
4660 if (wq
->lock_name
!= wq
->name
)
4661 kfree(wq
->lock_name
);
4664 static void wq_init_lockdep(struct workqueue_struct
*wq
)
4668 static void wq_unregister_lockdep(struct workqueue_struct
*wq
)
4672 static void wq_free_lockdep(struct workqueue_struct
*wq
)
4677 static void free_node_nr_active(struct wq_node_nr_active
**nna_ar
)
4681 for_each_node(node
) {
4682 kfree(nna_ar
[node
]);
4683 nna_ar
[node
] = NULL
;
4686 kfree(nna_ar
[nr_node_ids
]);
4687 nna_ar
[nr_node_ids
] = NULL
;
4690 static void init_node_nr_active(struct wq_node_nr_active
*nna
)
4692 nna
->max
= WQ_DFL_MIN_ACTIVE
;
4693 atomic_set(&nna
->nr
, 0);
4694 raw_spin_lock_init(&nna
->lock
);
4695 INIT_LIST_HEAD(&nna
->pending_pwqs
);
4699 * Each node's nr_active counter will be accessed mostly from its own node and
4700 * should be allocated in the node.
4702 static int alloc_node_nr_active(struct wq_node_nr_active
**nna_ar
)
4704 struct wq_node_nr_active
*nna
;
4707 for_each_node(node
) {
4708 nna
= kzalloc_node(sizeof(*nna
), GFP_KERNEL
, node
);
4711 init_node_nr_active(nna
);
4715 /* [nr_node_ids] is used as the fallback */
4716 nna
= kzalloc_node(sizeof(*nna
), GFP_KERNEL
, NUMA_NO_NODE
);
4719 init_node_nr_active(nna
);
4720 nna_ar
[nr_node_ids
] = nna
;
4725 free_node_nr_active(nna_ar
);
4729 static void rcu_free_wq(struct rcu_head
*rcu
)
4731 struct workqueue_struct
*wq
=
4732 container_of(rcu
, struct workqueue_struct
, rcu
);
4734 if (wq
->flags
& WQ_UNBOUND
)
4735 free_node_nr_active(wq
->node_nr_active
);
4737 wq_free_lockdep(wq
);
4738 free_percpu(wq
->cpu_pwq
);
4739 free_workqueue_attrs(wq
->unbound_attrs
);
4743 static void rcu_free_pool(struct rcu_head
*rcu
)
4745 struct worker_pool
*pool
= container_of(rcu
, struct worker_pool
, rcu
);
4747 ida_destroy(&pool
->worker_ida
);
4748 free_workqueue_attrs(pool
->attrs
);
4753 * put_unbound_pool - put a worker_pool
4754 * @pool: worker_pool to put
4756 * Put @pool. If its refcnt reaches zero, it gets destroyed in RCU
4757 * safe manner. get_unbound_pool() calls this function on its failure path
4758 * and this function should be able to release pools which went through,
4759 * successfully or not, init_worker_pool().
4761 * Should be called with wq_pool_mutex held.
4763 static void put_unbound_pool(struct worker_pool
*pool
)
4765 DECLARE_COMPLETION_ONSTACK(detach_completion
);
4766 struct worker
*worker
;
4767 LIST_HEAD(cull_list
);
4769 lockdep_assert_held(&wq_pool_mutex
);
4775 if (WARN_ON(!(pool
->cpu
< 0)) ||
4776 WARN_ON(!list_empty(&pool
->worklist
)))
4779 /* release id and unhash */
4781 idr_remove(&worker_pool_idr
, pool
->id
);
4782 hash_del(&pool
->hash_node
);
4785 * Become the manager and destroy all workers. This prevents
4786 * @pool's workers from blocking on attach_mutex. We're the last
4787 * manager and @pool gets freed with the flag set.
4789 * Having a concurrent manager is quite unlikely to happen as we can
4790 * only get here with
4791 * pwq->refcnt == pool->refcnt == 0
4792 * which implies no work queued to the pool, which implies no worker can
4793 * become the manager. However a worker could have taken the role of
4794 * manager before the refcnts dropped to 0, since maybe_create_worker()
4798 rcuwait_wait_event(&manager_wait
,
4799 !(pool
->flags
& POOL_MANAGER_ACTIVE
),
4800 TASK_UNINTERRUPTIBLE
);
4802 mutex_lock(&wq_pool_attach_mutex
);
4803 raw_spin_lock_irq(&pool
->lock
);
4804 if (!(pool
->flags
& POOL_MANAGER_ACTIVE
)) {
4805 pool
->flags
|= POOL_MANAGER_ACTIVE
;
4808 raw_spin_unlock_irq(&pool
->lock
);
4809 mutex_unlock(&wq_pool_attach_mutex
);
4812 while ((worker
= first_idle_worker(pool
)))
4813 set_worker_dying(worker
, &cull_list
);
4814 WARN_ON(pool
->nr_workers
|| pool
->nr_idle
);
4815 raw_spin_unlock_irq(&pool
->lock
);
4817 wake_dying_workers(&cull_list
);
4819 if (!list_empty(&pool
->workers
) || !list_empty(&pool
->dying_workers
))
4820 pool
->detach_completion
= &detach_completion
;
4821 mutex_unlock(&wq_pool_attach_mutex
);
4823 if (pool
->detach_completion
)
4824 wait_for_completion(pool
->detach_completion
);
4826 /* shut down the timers */
4827 del_timer_sync(&pool
->idle_timer
);
4828 cancel_work_sync(&pool
->idle_cull_work
);
4829 del_timer_sync(&pool
->mayday_timer
);
4831 /* RCU protected to allow dereferences from get_work_pool() */
4832 call_rcu(&pool
->rcu
, rcu_free_pool
);
4836 * get_unbound_pool - get a worker_pool with the specified attributes
4837 * @attrs: the attributes of the worker_pool to get
4839 * Obtain a worker_pool which has the same attributes as @attrs, bump the
4840 * reference count and return it. If there already is a matching
4841 * worker_pool, it will be used; otherwise, this function attempts to
4844 * Should be called with wq_pool_mutex held.
4846 * Return: On success, a worker_pool with the same attributes as @attrs.
4847 * On failure, %NULL.
4849 static struct worker_pool
*get_unbound_pool(const struct workqueue_attrs
*attrs
)
4851 struct wq_pod_type
*pt
= &wq_pod_types
[WQ_AFFN_NUMA
];
4852 u32 hash
= wqattrs_hash(attrs
);
4853 struct worker_pool
*pool
;
4854 int pod
, node
= NUMA_NO_NODE
;
4856 lockdep_assert_held(&wq_pool_mutex
);
4858 /* do we already have a matching pool? */
4859 hash_for_each_possible(unbound_pool_hash
, pool
, hash_node
, hash
) {
4860 if (wqattrs_equal(pool
->attrs
, attrs
)) {
4866 /* If __pod_cpumask is contained inside a NUMA pod, that's our node */
4867 for (pod
= 0; pod
< pt
->nr_pods
; pod
++) {
4868 if (cpumask_subset(attrs
->__pod_cpumask
, pt
->pod_cpus
[pod
])) {
4869 node
= pt
->pod_node
[pod
];
4874 /* nope, create a new one */
4875 pool
= kzalloc_node(sizeof(*pool
), GFP_KERNEL
, node
);
4876 if (!pool
|| init_worker_pool(pool
) < 0)
4880 copy_workqueue_attrs(pool
->attrs
, attrs
);
4881 wqattrs_clear_for_pool(pool
->attrs
);
4883 if (worker_pool_assign_id(pool
) < 0)
4886 /* create and start the initial worker */
4887 if (wq_online
&& !create_worker(pool
))
4891 hash_add(unbound_pool_hash
, &pool
->hash_node
, hash
);
4896 put_unbound_pool(pool
);
4900 static void rcu_free_pwq(struct rcu_head
*rcu
)
4902 kmem_cache_free(pwq_cache
,
4903 container_of(rcu
, struct pool_workqueue
, rcu
));
4907 * Scheduled on pwq_release_worker by put_pwq() when an unbound pwq hits zero
4908 * refcnt and needs to be destroyed.
4910 static void pwq_release_workfn(struct kthread_work
*work
)
4912 struct pool_workqueue
*pwq
= container_of(work
, struct pool_workqueue
,
4914 struct workqueue_struct
*wq
= pwq
->wq
;
4915 struct worker_pool
*pool
= pwq
->pool
;
4916 bool is_last
= false;
4919 * When @pwq is not linked, it doesn't hold any reference to the
4920 * @wq, and @wq is invalid to access.
4922 if (!list_empty(&pwq
->pwqs_node
)) {
4923 mutex_lock(&wq
->mutex
);
4924 list_del_rcu(&pwq
->pwqs_node
);
4925 is_last
= list_empty(&wq
->pwqs
);
4928 * For ordered workqueue with a plugged dfl_pwq, restart it now.
4930 if (!is_last
&& (wq
->flags
& __WQ_ORDERED
))
4931 unplug_oldest_pwq(wq
);
4933 mutex_unlock(&wq
->mutex
);
4936 if (wq
->flags
& WQ_UNBOUND
) {
4937 mutex_lock(&wq_pool_mutex
);
4938 put_unbound_pool(pool
);
4939 mutex_unlock(&wq_pool_mutex
);
4942 if (!list_empty(&pwq
->pending_node
)) {
4943 struct wq_node_nr_active
*nna
=
4944 wq_node_nr_active(pwq
->wq
, pwq
->pool
->node
);
4946 raw_spin_lock_irq(&nna
->lock
);
4947 list_del_init(&pwq
->pending_node
);
4948 raw_spin_unlock_irq(&nna
->lock
);
4951 call_rcu(&pwq
->rcu
, rcu_free_pwq
);
4954 * If we're the last pwq going away, @wq is already dead and no one
4955 * is gonna access it anymore. Schedule RCU free.
4958 wq_unregister_lockdep(wq
);
4959 call_rcu(&wq
->rcu
, rcu_free_wq
);
4963 /* initialize newly allocated @pwq which is associated with @wq and @pool */
4964 static void init_pwq(struct pool_workqueue
*pwq
, struct workqueue_struct
*wq
,
4965 struct worker_pool
*pool
)
4967 BUG_ON((unsigned long)pwq
& ~WORK_STRUCT_PWQ_MASK
);
4969 memset(pwq
, 0, sizeof(*pwq
));
4973 pwq
->flush_color
= -1;
4975 INIT_LIST_HEAD(&pwq
->inactive_works
);
4976 INIT_LIST_HEAD(&pwq
->pending_node
);
4977 INIT_LIST_HEAD(&pwq
->pwqs_node
);
4978 INIT_LIST_HEAD(&pwq
->mayday_node
);
4979 kthread_init_work(&pwq
->release_work
, pwq_release_workfn
);
4982 /* sync @pwq with the current state of its associated wq and link it */
4983 static void link_pwq(struct pool_workqueue
*pwq
)
4985 struct workqueue_struct
*wq
= pwq
->wq
;
4987 lockdep_assert_held(&wq
->mutex
);
4989 /* may be called multiple times, ignore if already linked */
4990 if (!list_empty(&pwq
->pwqs_node
))
4993 /* set the matching work_color */
4994 pwq
->work_color
= wq
->work_color
;
4997 list_add_tail_rcu(&pwq
->pwqs_node
, &wq
->pwqs
);
5000 /* obtain a pool matching @attr and create a pwq associating the pool and @wq */
5001 static struct pool_workqueue
*alloc_unbound_pwq(struct workqueue_struct
*wq
,
5002 const struct workqueue_attrs
*attrs
)
5004 struct worker_pool
*pool
;
5005 struct pool_workqueue
*pwq
;
5007 lockdep_assert_held(&wq_pool_mutex
);
5009 pool
= get_unbound_pool(attrs
);
5013 pwq
= kmem_cache_alloc_node(pwq_cache
, GFP_KERNEL
, pool
->node
);
5015 put_unbound_pool(pool
);
5019 init_pwq(pwq
, wq
, pool
);
5024 * wq_calc_pod_cpumask - calculate a wq_attrs' cpumask for a pod
5025 * @attrs: the wq_attrs of the default pwq of the target workqueue
5026 * @cpu: the target CPU
5027 * @cpu_going_down: if >= 0, the CPU to consider as offline
5029 * Calculate the cpumask a workqueue with @attrs should use on @pod. If
5030 * @cpu_going_down is >= 0, that cpu is considered offline during calculation.
5031 * The result is stored in @attrs->__pod_cpumask.
5033 * If pod affinity is not enabled, @attrs->cpumask is always used. If enabled
5034 * and @pod has online CPUs requested by @attrs, the returned cpumask is the
5035 * intersection of the possible CPUs of @pod and @attrs->cpumask.
5037 * The caller is responsible for ensuring that the cpumask of @pod stays stable.
5039 static void wq_calc_pod_cpumask(struct workqueue_attrs
*attrs
, int cpu
,
5042 const struct wq_pod_type
*pt
= wqattrs_pod_type(attrs
);
5043 int pod
= pt
->cpu_pod
[cpu
];
5045 /* does @pod have any online CPUs @attrs wants? */
5046 cpumask_and(attrs
->__pod_cpumask
, pt
->pod_cpus
[pod
], attrs
->cpumask
);
5047 cpumask_and(attrs
->__pod_cpumask
, attrs
->__pod_cpumask
, cpu_online_mask
);
5048 if (cpu_going_down
>= 0)
5049 cpumask_clear_cpu(cpu_going_down
, attrs
->__pod_cpumask
);
5051 if (cpumask_empty(attrs
->__pod_cpumask
)) {
5052 cpumask_copy(attrs
->__pod_cpumask
, attrs
->cpumask
);
5056 /* yeap, return possible CPUs in @pod that @attrs wants */
5057 cpumask_and(attrs
->__pod_cpumask
, attrs
->cpumask
, pt
->pod_cpus
[pod
]);
5059 if (cpumask_empty(attrs
->__pod_cpumask
))
5060 pr_warn_once("WARNING: workqueue cpumask: online intersect > "
5061 "possible intersect\n");
5064 /* install @pwq into @wq and return the old pwq, @cpu < 0 for dfl_pwq */
5065 static struct pool_workqueue
*install_unbound_pwq(struct workqueue_struct
*wq
,
5066 int cpu
, struct pool_workqueue
*pwq
)
5068 struct pool_workqueue __rcu
**slot
= unbound_pwq_slot(wq
, cpu
);
5069 struct pool_workqueue
*old_pwq
;
5071 lockdep_assert_held(&wq_pool_mutex
);
5072 lockdep_assert_held(&wq
->mutex
);
5074 /* link_pwq() can handle duplicate calls */
5077 old_pwq
= rcu_access_pointer(*slot
);
5078 rcu_assign_pointer(*slot
, pwq
);
5082 /* context to store the prepared attrs & pwqs before applying */
5083 struct apply_wqattrs_ctx
{
5084 struct workqueue_struct
*wq
; /* target workqueue */
5085 struct workqueue_attrs
*attrs
; /* attrs to apply */
5086 struct list_head list
; /* queued for batching commit */
5087 struct pool_workqueue
*dfl_pwq
;
5088 struct pool_workqueue
*pwq_tbl
[];
5091 /* free the resources after success or abort */
5092 static void apply_wqattrs_cleanup(struct apply_wqattrs_ctx
*ctx
)
5097 for_each_possible_cpu(cpu
)
5098 put_pwq_unlocked(ctx
->pwq_tbl
[cpu
]);
5099 put_pwq_unlocked(ctx
->dfl_pwq
);
5101 free_workqueue_attrs(ctx
->attrs
);
5107 /* allocate the attrs and pwqs for later installation */
5108 static struct apply_wqattrs_ctx
*
5109 apply_wqattrs_prepare(struct workqueue_struct
*wq
,
5110 const struct workqueue_attrs
*attrs
,
5111 const cpumask_var_t unbound_cpumask
)
5113 struct apply_wqattrs_ctx
*ctx
;
5114 struct workqueue_attrs
*new_attrs
;
5117 lockdep_assert_held(&wq_pool_mutex
);
5119 if (WARN_ON(attrs
->affn_scope
< 0 ||
5120 attrs
->affn_scope
>= WQ_AFFN_NR_TYPES
))
5121 return ERR_PTR(-EINVAL
);
5123 ctx
= kzalloc(struct_size(ctx
, pwq_tbl
, nr_cpu_ids
), GFP_KERNEL
);
5125 new_attrs
= alloc_workqueue_attrs();
5126 if (!ctx
|| !new_attrs
)
5130 * If something goes wrong during CPU up/down, we'll fall back to
5131 * the default pwq covering whole @attrs->cpumask. Always create
5132 * it even if we don't use it immediately.
5134 copy_workqueue_attrs(new_attrs
, attrs
);
5135 wqattrs_actualize_cpumask(new_attrs
, unbound_cpumask
);
5136 cpumask_copy(new_attrs
->__pod_cpumask
, new_attrs
->cpumask
);
5137 ctx
->dfl_pwq
= alloc_unbound_pwq(wq
, new_attrs
);
5141 for_each_possible_cpu(cpu
) {
5142 if (new_attrs
->ordered
) {
5143 ctx
->dfl_pwq
->refcnt
++;
5144 ctx
->pwq_tbl
[cpu
] = ctx
->dfl_pwq
;
5146 wq_calc_pod_cpumask(new_attrs
, cpu
, -1);
5147 ctx
->pwq_tbl
[cpu
] = alloc_unbound_pwq(wq
, new_attrs
);
5148 if (!ctx
->pwq_tbl
[cpu
])
5153 /* save the user configured attrs and sanitize it. */
5154 copy_workqueue_attrs(new_attrs
, attrs
);
5155 cpumask_and(new_attrs
->cpumask
, new_attrs
->cpumask
, cpu_possible_mask
);
5156 cpumask_copy(new_attrs
->__pod_cpumask
, new_attrs
->cpumask
);
5157 ctx
->attrs
= new_attrs
;
5160 * For initialized ordered workqueues, there should only be one pwq
5161 * (dfl_pwq). Set the plugged flag of ctx->dfl_pwq to suspend execution
5162 * of newly queued work items until execution of older work items in
5163 * the old pwq's have completed.
5165 if ((wq
->flags
& __WQ_ORDERED
) && !list_empty(&wq
->pwqs
))
5166 ctx
->dfl_pwq
->plugged
= true;
5172 free_workqueue_attrs(new_attrs
);
5173 apply_wqattrs_cleanup(ctx
);
5174 return ERR_PTR(-ENOMEM
);
5177 /* set attrs and install prepared pwqs, @ctx points to old pwqs on return */
5178 static void apply_wqattrs_commit(struct apply_wqattrs_ctx
*ctx
)
5182 /* all pwqs have been created successfully, let's install'em */
5183 mutex_lock(&ctx
->wq
->mutex
);
5185 copy_workqueue_attrs(ctx
->wq
->unbound_attrs
, ctx
->attrs
);
5187 /* save the previous pwqs and install the new ones */
5188 for_each_possible_cpu(cpu
)
5189 ctx
->pwq_tbl
[cpu
] = install_unbound_pwq(ctx
->wq
, cpu
,
5191 ctx
->dfl_pwq
= install_unbound_pwq(ctx
->wq
, -1, ctx
->dfl_pwq
);
5193 /* update node_nr_active->max */
5194 wq_update_node_max_active(ctx
->wq
, -1);
5196 /* rescuer needs to respect wq cpumask changes */
5197 if (ctx
->wq
->rescuer
)
5198 set_cpus_allowed_ptr(ctx
->wq
->rescuer
->task
,
5199 unbound_effective_cpumask(ctx
->wq
));
5201 mutex_unlock(&ctx
->wq
->mutex
);
5204 static int apply_workqueue_attrs_locked(struct workqueue_struct
*wq
,
5205 const struct workqueue_attrs
*attrs
)
5207 struct apply_wqattrs_ctx
*ctx
;
5209 /* only unbound workqueues can change attributes */
5210 if (WARN_ON(!(wq
->flags
& WQ_UNBOUND
)))
5213 ctx
= apply_wqattrs_prepare(wq
, attrs
, wq_unbound_cpumask
);
5215 return PTR_ERR(ctx
);
5217 /* the ctx has been prepared successfully, let's commit it */
5218 apply_wqattrs_commit(ctx
);
5219 apply_wqattrs_cleanup(ctx
);
5225 * apply_workqueue_attrs - apply new workqueue_attrs to an unbound workqueue
5226 * @wq: the target workqueue
5227 * @attrs: the workqueue_attrs to apply, allocated with alloc_workqueue_attrs()
5229 * Apply @attrs to an unbound workqueue @wq. Unless disabled, this function maps
5230 * a separate pwq to each CPU pod with possibles CPUs in @attrs->cpumask so that
5231 * work items are affine to the pod it was issued on. Older pwqs are released as
5232 * in-flight work items finish. Note that a work item which repeatedly requeues
5233 * itself back-to-back will stay on its current pwq.
5235 * Performs GFP_KERNEL allocations.
5237 * Assumes caller has CPU hotplug read exclusion, i.e. cpus_read_lock().
5239 * Return: 0 on success and -errno on failure.
5241 int apply_workqueue_attrs(struct workqueue_struct
*wq
,
5242 const struct workqueue_attrs
*attrs
)
5246 lockdep_assert_cpus_held();
5248 mutex_lock(&wq_pool_mutex
);
5249 ret
= apply_workqueue_attrs_locked(wq
, attrs
);
5250 mutex_unlock(&wq_pool_mutex
);
5256 * wq_update_pod - update pod affinity of a wq for CPU hot[un]plug
5257 * @wq: the target workqueue
5258 * @cpu: the CPU to update pool association for
5259 * @hotplug_cpu: the CPU coming up or going down
5260 * @online: whether @cpu is coming up or going down
5262 * This function is to be called from %CPU_DOWN_PREPARE, %CPU_ONLINE and
5263 * %CPU_DOWN_FAILED. @cpu is being hot[un]plugged, update pod affinity of
5267 * If pod affinity can't be adjusted due to memory allocation failure, it falls
5268 * back to @wq->dfl_pwq which may not be optimal but is always correct.
5270 * Note that when the last allowed CPU of a pod goes offline for a workqueue
5271 * with a cpumask spanning multiple pods, the workers which were already
5272 * executing the work items for the workqueue will lose their CPU affinity and
5273 * may execute on any CPU. This is similar to how per-cpu workqueues behave on
5274 * CPU_DOWN. If a workqueue user wants strict affinity, it's the user's
5275 * responsibility to flush the work item from CPU_DOWN_PREPARE.
5277 static void wq_update_pod(struct workqueue_struct
*wq
, int cpu
,
5278 int hotplug_cpu
, bool online
)
5280 int off_cpu
= online
? -1 : hotplug_cpu
;
5281 struct pool_workqueue
*old_pwq
= NULL
, *pwq
;
5282 struct workqueue_attrs
*target_attrs
;
5284 lockdep_assert_held(&wq_pool_mutex
);
5286 if (!(wq
->flags
& WQ_UNBOUND
) || wq
->unbound_attrs
->ordered
)
5290 * We don't wanna alloc/free wq_attrs for each wq for each CPU.
5291 * Let's use a preallocated one. The following buf is protected by
5292 * CPU hotplug exclusion.
5294 target_attrs
= wq_update_pod_attrs_buf
;
5296 copy_workqueue_attrs(target_attrs
, wq
->unbound_attrs
);
5297 wqattrs_actualize_cpumask(target_attrs
, wq_unbound_cpumask
);
5299 /* nothing to do if the target cpumask matches the current pwq */
5300 wq_calc_pod_cpumask(target_attrs
, cpu
, off_cpu
);
5301 if (wqattrs_equal(target_attrs
, unbound_pwq(wq
, cpu
)->pool
->attrs
))
5304 /* create a new pwq */
5305 pwq
= alloc_unbound_pwq(wq
, target_attrs
);
5307 pr_warn("workqueue: allocation failed while updating CPU pod affinity of \"%s\"\n",
5312 /* Install the new pwq. */
5313 mutex_lock(&wq
->mutex
);
5314 old_pwq
= install_unbound_pwq(wq
, cpu
, pwq
);
5318 mutex_lock(&wq
->mutex
);
5319 pwq
= unbound_pwq(wq
, -1);
5320 raw_spin_lock_irq(&pwq
->pool
->lock
);
5322 raw_spin_unlock_irq(&pwq
->pool
->lock
);
5323 old_pwq
= install_unbound_pwq(wq
, cpu
, pwq
);
5325 mutex_unlock(&wq
->mutex
);
5326 put_pwq_unlocked(old_pwq
);
5329 static int alloc_and_link_pwqs(struct workqueue_struct
*wq
)
5331 bool highpri
= wq
->flags
& WQ_HIGHPRI
;
5334 wq
->cpu_pwq
= alloc_percpu(struct pool_workqueue
*);
5338 if (!(wq
->flags
& WQ_UNBOUND
)) {
5339 for_each_possible_cpu(cpu
) {
5340 struct pool_workqueue
**pwq_p
;
5341 struct worker_pool __percpu
*pools
;
5342 struct worker_pool
*pool
;
5344 if (wq
->flags
& WQ_BH
)
5345 pools
= bh_worker_pools
;
5347 pools
= cpu_worker_pools
;
5349 pool
= &(per_cpu_ptr(pools
, cpu
)[highpri
]);
5350 pwq_p
= per_cpu_ptr(wq
->cpu_pwq
, cpu
);
5352 *pwq_p
= kmem_cache_alloc_node(pwq_cache
, GFP_KERNEL
,
5357 init_pwq(*pwq_p
, wq
, pool
);
5359 mutex_lock(&wq
->mutex
);
5361 mutex_unlock(&wq
->mutex
);
5367 if (wq
->flags
& __WQ_ORDERED
) {
5368 struct pool_workqueue
*dfl_pwq
;
5370 ret
= apply_workqueue_attrs(wq
, ordered_wq_attrs
[highpri
]);
5371 /* there should only be single pwq for ordering guarantee */
5372 dfl_pwq
= rcu_access_pointer(wq
->dfl_pwq
);
5373 WARN(!ret
&& (wq
->pwqs
.next
!= &dfl_pwq
->pwqs_node
||
5374 wq
->pwqs
.prev
!= &dfl_pwq
->pwqs_node
),
5375 "ordering guarantee broken for workqueue %s\n", wq
->name
);
5377 ret
= apply_workqueue_attrs(wq
, unbound_std_wq_attrs
[highpri
]);
5381 /* for unbound pwq, flush the pwq_release_worker ensures that the
5382 * pwq_release_workfn() completes before calling kfree(wq).
5385 kthread_flush_worker(pwq_release_worker
);
5391 for_each_possible_cpu(cpu
) {
5392 struct pool_workqueue
*pwq
= *per_cpu_ptr(wq
->cpu_pwq
, cpu
);
5395 kmem_cache_free(pwq_cache
, pwq
);
5397 free_percpu(wq
->cpu_pwq
);
5403 static int wq_clamp_max_active(int max_active
, unsigned int flags
,
5406 if (max_active
< 1 || max_active
> WQ_MAX_ACTIVE
)
5407 pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n",
5408 max_active
, name
, 1, WQ_MAX_ACTIVE
);
5410 return clamp_val(max_active
, 1, WQ_MAX_ACTIVE
);
5414 * Workqueues which may be used during memory reclaim should have a rescuer
5415 * to guarantee forward progress.
5417 static int init_rescuer(struct workqueue_struct
*wq
)
5419 struct worker
*rescuer
;
5422 if (!(wq
->flags
& WQ_MEM_RECLAIM
))
5425 rescuer
= alloc_worker(NUMA_NO_NODE
);
5427 pr_err("workqueue: Failed to allocate a rescuer for wq \"%s\"\n",
5432 rescuer
->rescue_wq
= wq
;
5433 rescuer
->task
= kthread_create(rescuer_thread
, rescuer
, "kworker/R-%s", wq
->name
);
5434 if (IS_ERR(rescuer
->task
)) {
5435 ret
= PTR_ERR(rescuer
->task
);
5436 pr_err("workqueue: Failed to create a rescuer kthread for wq \"%s\": %pe",
5437 wq
->name
, ERR_PTR(ret
));
5442 wq
->rescuer
= rescuer
;
5443 if (wq
->flags
& WQ_UNBOUND
)
5444 kthread_bind_mask(rescuer
->task
, wq_unbound_cpumask
);
5446 kthread_bind_mask(rescuer
->task
, cpu_possible_mask
);
5447 wake_up_process(rescuer
->task
);
5453 * wq_adjust_max_active - update a wq's max_active to the current setting
5454 * @wq: target workqueue
5456 * If @wq isn't freezing, set @wq->max_active to the saved_max_active and
5457 * activate inactive work items accordingly. If @wq is freezing, clear
5458 * @wq->max_active to zero.
5460 static void wq_adjust_max_active(struct workqueue_struct
*wq
)
5463 int new_max
, new_min
;
5465 lockdep_assert_held(&wq
->mutex
);
5467 if ((wq
->flags
& WQ_FREEZABLE
) && workqueue_freezing
) {
5471 new_max
= wq
->saved_max_active
;
5472 new_min
= wq
->saved_min_active
;
5475 if (wq
->max_active
== new_max
&& wq
->min_active
== new_min
)
5479 * Update @wq->max/min_active and then kick inactive work items if more
5480 * active work items are allowed. This doesn't break work item ordering
5481 * because new work items are always queued behind existing inactive
5482 * work items if there are any.
5484 WRITE_ONCE(wq
->max_active
, new_max
);
5485 WRITE_ONCE(wq
->min_active
, new_min
);
5487 if (wq
->flags
& WQ_UNBOUND
)
5488 wq_update_node_max_active(wq
, -1);
5494 * Round-robin through pwq's activating the first inactive work item
5495 * until max_active is filled.
5498 struct pool_workqueue
*pwq
;
5501 for_each_pwq(pwq
, wq
) {
5502 unsigned long irq_flags
;
5504 /* can be called during early boot w/ irq disabled */
5505 raw_spin_lock_irqsave(&pwq
->pool
->lock
, irq_flags
);
5506 if (pwq_activate_first_inactive(pwq
, true)) {
5508 kick_pool(pwq
->pool
);
5510 raw_spin_unlock_irqrestore(&pwq
->pool
->lock
, irq_flags
);
5512 } while (activated
);
5516 struct workqueue_struct
*alloc_workqueue(const char *fmt
,
5518 int max_active
, ...)
5521 struct workqueue_struct
*wq
;
5525 if (flags
& WQ_BH
) {
5526 if (WARN_ON_ONCE(flags
& ~__WQ_BH_ALLOWS
))
5528 if (WARN_ON_ONCE(max_active
))
5532 /* see the comment above the definition of WQ_POWER_EFFICIENT */
5533 if ((flags
& WQ_POWER_EFFICIENT
) && wq_power_efficient
)
5534 flags
|= WQ_UNBOUND
;
5536 /* allocate wq and format name */
5537 if (flags
& WQ_UNBOUND
)
5538 wq_size
= struct_size(wq
, node_nr_active
, nr_node_ids
+ 1);
5540 wq_size
= sizeof(*wq
);
5542 wq
= kzalloc(wq_size
, GFP_KERNEL
);
5546 if (flags
& WQ_UNBOUND
) {
5547 wq
->unbound_attrs
= alloc_workqueue_attrs();
5548 if (!wq
->unbound_attrs
)
5552 va_start(args
, max_active
);
5553 name_len
= vsnprintf(wq
->name
, sizeof(wq
->name
), fmt
, args
);
5556 if (name_len
>= WQ_NAME_LEN
)
5557 pr_warn_once("workqueue: name exceeds WQ_NAME_LEN. Truncating to: %s\n",
5560 if (flags
& WQ_BH
) {
5562 * BH workqueues always share a single execution context per CPU
5563 * and don't impose any max_active limit.
5565 max_active
= INT_MAX
;
5567 max_active
= max_active
?: WQ_DFL_ACTIVE
;
5568 max_active
= wq_clamp_max_active(max_active
, flags
, wq
->name
);
5573 wq
->max_active
= max_active
;
5574 wq
->min_active
= min(max_active
, WQ_DFL_MIN_ACTIVE
);
5575 wq
->saved_max_active
= wq
->max_active
;
5576 wq
->saved_min_active
= wq
->min_active
;
5577 mutex_init(&wq
->mutex
);
5578 atomic_set(&wq
->nr_pwqs_to_flush
, 0);
5579 INIT_LIST_HEAD(&wq
->pwqs
);
5580 INIT_LIST_HEAD(&wq
->flusher_queue
);
5581 INIT_LIST_HEAD(&wq
->flusher_overflow
);
5582 INIT_LIST_HEAD(&wq
->maydays
);
5584 wq_init_lockdep(wq
);
5585 INIT_LIST_HEAD(&wq
->list
);
5587 if (flags
& WQ_UNBOUND
) {
5588 if (alloc_node_nr_active(wq
->node_nr_active
) < 0)
5589 goto err_unreg_lockdep
;
5592 if (alloc_and_link_pwqs(wq
) < 0)
5593 goto err_free_node_nr_active
;
5595 if (wq_online
&& init_rescuer(wq
) < 0)
5598 if ((wq
->flags
& WQ_SYSFS
) && workqueue_sysfs_register(wq
))
5602 * wq_pool_mutex protects global freeze state and workqueues list.
5603 * Grab it, adjust max_active and add the new @wq to workqueues
5606 mutex_lock(&wq_pool_mutex
);
5608 mutex_lock(&wq
->mutex
);
5609 wq_adjust_max_active(wq
);
5610 mutex_unlock(&wq
->mutex
);
5612 list_add_tail_rcu(&wq
->list
, &workqueues
);
5614 mutex_unlock(&wq_pool_mutex
);
5618 err_free_node_nr_active
:
5619 if (wq
->flags
& WQ_UNBOUND
)
5620 free_node_nr_active(wq
->node_nr_active
);
5622 wq_unregister_lockdep(wq
);
5623 wq_free_lockdep(wq
);
5625 free_workqueue_attrs(wq
->unbound_attrs
);
5629 destroy_workqueue(wq
);
5632 EXPORT_SYMBOL_GPL(alloc_workqueue
);
5634 static bool pwq_busy(struct pool_workqueue
*pwq
)
5638 for (i
= 0; i
< WORK_NR_COLORS
; i
++)
5639 if (pwq
->nr_in_flight
[i
])
5642 if ((pwq
!= rcu_access_pointer(pwq
->wq
->dfl_pwq
)) && (pwq
->refcnt
> 1))
5644 if (!pwq_is_empty(pwq
))
5651 * destroy_workqueue - safely terminate a workqueue
5652 * @wq: target workqueue
5654 * Safely destroy a workqueue. All work currently pending will be done first.
5656 void destroy_workqueue(struct workqueue_struct
*wq
)
5658 struct pool_workqueue
*pwq
;
5662 * Remove it from sysfs first so that sanity check failure doesn't
5663 * lead to sysfs name conflicts.
5665 workqueue_sysfs_unregister(wq
);
5667 /* mark the workqueue destruction is in progress */
5668 mutex_lock(&wq
->mutex
);
5669 wq
->flags
|= __WQ_DESTROYING
;
5670 mutex_unlock(&wq
->mutex
);
5672 /* drain it before proceeding with destruction */
5673 drain_workqueue(wq
);
5675 /* kill rescuer, if sanity checks fail, leave it w/o rescuer */
5677 struct worker
*rescuer
= wq
->rescuer
;
5679 /* this prevents new queueing */
5680 raw_spin_lock_irq(&wq_mayday_lock
);
5682 raw_spin_unlock_irq(&wq_mayday_lock
);
5684 /* rescuer will empty maydays list before exiting */
5685 kthread_stop(rescuer
->task
);
5690 * Sanity checks - grab all the locks so that we wait for all
5691 * in-flight operations which may do put_pwq().
5693 mutex_lock(&wq_pool_mutex
);
5694 mutex_lock(&wq
->mutex
);
5695 for_each_pwq(pwq
, wq
) {
5696 raw_spin_lock_irq(&pwq
->pool
->lock
);
5697 if (WARN_ON(pwq_busy(pwq
))) {
5698 pr_warn("%s: %s has the following busy pwq\n",
5699 __func__
, wq
->name
);
5701 raw_spin_unlock_irq(&pwq
->pool
->lock
);
5702 mutex_unlock(&wq
->mutex
);
5703 mutex_unlock(&wq_pool_mutex
);
5704 show_one_workqueue(wq
);
5707 raw_spin_unlock_irq(&pwq
->pool
->lock
);
5709 mutex_unlock(&wq
->mutex
);
5712 * wq list is used to freeze wq, remove from list after
5713 * flushing is complete in case freeze races us.
5715 list_del_rcu(&wq
->list
);
5716 mutex_unlock(&wq_pool_mutex
);
5719 * We're the sole accessor of @wq. Directly access cpu_pwq and dfl_pwq
5720 * to put the base refs. @wq will be auto-destroyed from the last
5721 * pwq_put. RCU read lock prevents @wq from going away from under us.
5725 for_each_possible_cpu(cpu
) {
5726 put_pwq_unlocked(unbound_pwq(wq
, cpu
));
5727 RCU_INIT_POINTER(*unbound_pwq_slot(wq
, cpu
), NULL
);
5730 put_pwq_unlocked(unbound_pwq(wq
, -1));
5731 RCU_INIT_POINTER(*unbound_pwq_slot(wq
, -1), NULL
);
5735 EXPORT_SYMBOL_GPL(destroy_workqueue
);
5738 * workqueue_set_max_active - adjust max_active of a workqueue
5739 * @wq: target workqueue
5740 * @max_active: new max_active value.
5742 * Set max_active of @wq to @max_active. See the alloc_workqueue() function
5746 * Don't call from IRQ context.
5748 void workqueue_set_max_active(struct workqueue_struct
*wq
, int max_active
)
5750 /* max_active doesn't mean anything for BH workqueues */
5751 if (WARN_ON(wq
->flags
& WQ_BH
))
5753 /* disallow meddling with max_active for ordered workqueues */
5754 if (WARN_ON(wq
->flags
& __WQ_ORDERED
))
5757 max_active
= wq_clamp_max_active(max_active
, wq
->flags
, wq
->name
);
5759 mutex_lock(&wq
->mutex
);
5761 wq
->saved_max_active
= max_active
;
5762 if (wq
->flags
& WQ_UNBOUND
)
5763 wq
->saved_min_active
= min(wq
->saved_min_active
, max_active
);
5765 wq_adjust_max_active(wq
);
5767 mutex_unlock(&wq
->mutex
);
5769 EXPORT_SYMBOL_GPL(workqueue_set_max_active
);
5772 * workqueue_set_min_active - adjust min_active of an unbound workqueue
5773 * @wq: target unbound workqueue
5774 * @min_active: new min_active value
5776 * Set min_active of an unbound workqueue. Unlike other types of workqueues, an
5777 * unbound workqueue is not guaranteed to be able to process max_active
5778 * interdependent work items. Instead, an unbound workqueue is guaranteed to be
5779 * able to process min_active number of interdependent work items which is
5780 * %WQ_DFL_MIN_ACTIVE by default.
5782 * Use this function to adjust the min_active value between 0 and the current
5785 void workqueue_set_min_active(struct workqueue_struct
*wq
, int min_active
)
5787 /* min_active is only meaningful for non-ordered unbound workqueues */
5788 if (WARN_ON((wq
->flags
& (WQ_BH
| WQ_UNBOUND
| __WQ_ORDERED
)) !=
5792 mutex_lock(&wq
->mutex
);
5793 wq
->saved_min_active
= clamp(min_active
, 0, wq
->saved_max_active
);
5794 wq_adjust_max_active(wq
);
5795 mutex_unlock(&wq
->mutex
);
5799 * current_work - retrieve %current task's work struct
5801 * Determine if %current task is a workqueue worker and what it's working on.
5802 * Useful to find out the context that the %current task is running in.
5804 * Return: work struct if %current task is a workqueue worker, %NULL otherwise.
5806 struct work_struct
*current_work(void)
5808 struct worker
*worker
= current_wq_worker();
5810 return worker
? worker
->current_work
: NULL
;
5812 EXPORT_SYMBOL(current_work
);
5815 * current_is_workqueue_rescuer - is %current workqueue rescuer?
5817 * Determine whether %current is a workqueue rescuer. Can be used from
5818 * work functions to determine whether it's being run off the rescuer task.
5820 * Return: %true if %current is a workqueue rescuer. %false otherwise.
5822 bool current_is_workqueue_rescuer(void)
5824 struct worker
*worker
= current_wq_worker();
5826 return worker
&& worker
->rescue_wq
;
5830 * workqueue_congested - test whether a workqueue is congested
5831 * @cpu: CPU in question
5832 * @wq: target workqueue
5834 * Test whether @wq's cpu workqueue for @cpu is congested. There is
5835 * no synchronization around this function and the test result is
5836 * unreliable and only useful as advisory hints or for debugging.
5838 * If @cpu is WORK_CPU_UNBOUND, the test is performed on the local CPU.
5840 * With the exception of ordered workqueues, all workqueues have per-cpu
5841 * pool_workqueues, each with its own congested state. A workqueue being
5842 * congested on one CPU doesn't mean that the workqueue is contested on any
5846 * %true if congested, %false otherwise.
5848 bool workqueue_congested(int cpu
, struct workqueue_struct
*wq
)
5850 struct pool_workqueue
*pwq
;
5856 if (cpu
== WORK_CPU_UNBOUND
)
5857 cpu
= smp_processor_id();
5859 pwq
= *per_cpu_ptr(wq
->cpu_pwq
, cpu
);
5860 ret
= !list_empty(&pwq
->inactive_works
);
5867 EXPORT_SYMBOL_GPL(workqueue_congested
);
5870 * work_busy - test whether a work is currently pending or running
5871 * @work: the work to be tested
5873 * Test whether @work is currently pending or running. There is no
5874 * synchronization around this function and the test result is
5875 * unreliable and only useful as advisory hints or for debugging.
5878 * OR'd bitmask of WORK_BUSY_* bits.
5880 unsigned int work_busy(struct work_struct
*work
)
5882 struct worker_pool
*pool
;
5883 unsigned long irq_flags
;
5884 unsigned int ret
= 0;
5886 if (work_pending(work
))
5887 ret
|= WORK_BUSY_PENDING
;
5890 pool
= get_work_pool(work
);
5892 raw_spin_lock_irqsave(&pool
->lock
, irq_flags
);
5893 if (find_worker_executing_work(pool
, work
))
5894 ret
|= WORK_BUSY_RUNNING
;
5895 raw_spin_unlock_irqrestore(&pool
->lock
, irq_flags
);
5901 EXPORT_SYMBOL_GPL(work_busy
);
5904 * set_worker_desc - set description for the current work item
5905 * @fmt: printf-style format string
5906 * @...: arguments for the format string
5908 * This function can be called by a running work function to describe what
5909 * the work item is about. If the worker task gets dumped, this
5910 * information will be printed out together to help debugging. The
5911 * description can be at most WORKER_DESC_LEN including the trailing '\0'.
5913 void set_worker_desc(const char *fmt
, ...)
5915 struct worker
*worker
= current_wq_worker();
5919 va_start(args
, fmt
);
5920 vsnprintf(worker
->desc
, sizeof(worker
->desc
), fmt
, args
);
5924 EXPORT_SYMBOL_GPL(set_worker_desc
);
5927 * print_worker_info - print out worker information and description
5928 * @log_lvl: the log level to use when printing
5929 * @task: target task
5931 * If @task is a worker and currently executing a work item, print out the
5932 * name of the workqueue being serviced and worker description set with
5933 * set_worker_desc() by the currently executing work item.
5935 * This function can be safely called on any task as long as the
5936 * task_struct itself is accessible. While safe, this function isn't
5937 * synchronized and may print out mixups or garbages of limited length.
5939 void print_worker_info(const char *log_lvl
, struct task_struct
*task
)
5941 work_func_t
*fn
= NULL
;
5942 char name
[WQ_NAME_LEN
] = { };
5943 char desc
[WORKER_DESC_LEN
] = { };
5944 struct pool_workqueue
*pwq
= NULL
;
5945 struct workqueue_struct
*wq
= NULL
;
5946 struct worker
*worker
;
5948 if (!(task
->flags
& PF_WQ_WORKER
))
5952 * This function is called without any synchronization and @task
5953 * could be in any state. Be careful with dereferences.
5955 worker
= kthread_probe_data(task
);
5958 * Carefully copy the associated workqueue's workfn, name and desc.
5959 * Keep the original last '\0' in case the original is garbage.
5961 copy_from_kernel_nofault(&fn
, &worker
->current_func
, sizeof(fn
));
5962 copy_from_kernel_nofault(&pwq
, &worker
->current_pwq
, sizeof(pwq
));
5963 copy_from_kernel_nofault(&wq
, &pwq
->wq
, sizeof(wq
));
5964 copy_from_kernel_nofault(name
, wq
->name
, sizeof(name
) - 1);
5965 copy_from_kernel_nofault(desc
, worker
->desc
, sizeof(desc
) - 1);
5967 if (fn
|| name
[0] || desc
[0]) {
5968 printk("%sWorkqueue: %s %ps", log_lvl
, name
, fn
);
5969 if (strcmp(name
, desc
))
5970 pr_cont(" (%s)", desc
);
5975 static void pr_cont_pool_info(struct worker_pool
*pool
)
5977 pr_cont(" cpus=%*pbl", nr_cpumask_bits
, pool
->attrs
->cpumask
);
5978 if (pool
->node
!= NUMA_NO_NODE
)
5979 pr_cont(" node=%d", pool
->node
);
5980 pr_cont(" flags=0x%x", pool
->flags
);
5981 if (pool
->flags
& POOL_BH
)
5983 pool
->attrs
->nice
== HIGHPRI_NICE_LEVEL
? "-hi" : "");
5985 pr_cont(" nice=%d", pool
->attrs
->nice
);
5988 static void pr_cont_worker_id(struct worker
*worker
)
5990 struct worker_pool
*pool
= worker
->pool
;
5992 if (pool
->flags
& WQ_BH
)
5994 pool
->attrs
->nice
== HIGHPRI_NICE_LEVEL
? "-hi" : "");
5996 pr_cont("%d%s", task_pid_nr(worker
->task
),
5997 worker
->rescue_wq
? "(RESCUER)" : "");
6000 struct pr_cont_work_struct
{
6006 static void pr_cont_work_flush(bool comma
, work_func_t func
, struct pr_cont_work_struct
*pcwsp
)
6010 if (func
== pcwsp
->func
) {
6014 if (pcwsp
->ctr
== 1)
6015 pr_cont("%s %ps", pcwsp
->comma
? "," : "", pcwsp
->func
);
6017 pr_cont("%s %ld*%ps", pcwsp
->comma
? "," : "", pcwsp
->ctr
, pcwsp
->func
);
6020 if ((long)func
== -1L)
6022 pcwsp
->comma
= comma
;
6027 static void pr_cont_work(bool comma
, struct work_struct
*work
, struct pr_cont_work_struct
*pcwsp
)
6029 if (work
->func
== wq_barrier_func
) {
6030 struct wq_barrier
*barr
;
6032 barr
= container_of(work
, struct wq_barrier
, work
);
6034 pr_cont_work_flush(comma
, (work_func_t
)-1, pcwsp
);
6035 pr_cont("%s BAR(%d)", comma
? "," : "",
6036 task_pid_nr(barr
->task
));
6039 pr_cont_work_flush(comma
, (work_func_t
)-1, pcwsp
);
6040 pr_cont_work_flush(comma
, work
->func
, pcwsp
);
6044 static void show_pwq(struct pool_workqueue
*pwq
)
6046 struct pr_cont_work_struct pcws
= { .ctr
= 0, };
6047 struct worker_pool
*pool
= pwq
->pool
;
6048 struct work_struct
*work
;
6049 struct worker
*worker
;
6050 bool has_in_flight
= false, has_pending
= false;
6053 pr_info(" pwq %d:", pool
->id
);
6054 pr_cont_pool_info(pool
);
6056 pr_cont(" active=%d refcnt=%d%s\n",
6057 pwq
->nr_active
, pwq
->refcnt
,
6058 !list_empty(&pwq
->mayday_node
) ? " MAYDAY" : "");
6060 hash_for_each(pool
->busy_hash
, bkt
, worker
, hentry
) {
6061 if (worker
->current_pwq
== pwq
) {
6062 has_in_flight
= true;
6066 if (has_in_flight
) {
6069 pr_info(" in-flight:");
6070 hash_for_each(pool
->busy_hash
, bkt
, worker
, hentry
) {
6071 if (worker
->current_pwq
!= pwq
)
6074 pr_cont(" %s", comma
? "," : "");
6075 pr_cont_worker_id(worker
);
6076 pr_cont(":%ps", worker
->current_func
);
6077 list_for_each_entry(work
, &worker
->scheduled
, entry
)
6078 pr_cont_work(false, work
, &pcws
);
6079 pr_cont_work_flush(comma
, (work_func_t
)-1L, &pcws
);
6085 list_for_each_entry(work
, &pool
->worklist
, entry
) {
6086 if (get_work_pwq(work
) == pwq
) {
6094 pr_info(" pending:");
6095 list_for_each_entry(work
, &pool
->worklist
, entry
) {
6096 if (get_work_pwq(work
) != pwq
)
6099 pr_cont_work(comma
, work
, &pcws
);
6100 comma
= !(*work_data_bits(work
) & WORK_STRUCT_LINKED
);
6102 pr_cont_work_flush(comma
, (work_func_t
)-1L, &pcws
);
6106 if (!list_empty(&pwq
->inactive_works
)) {
6109 pr_info(" inactive:");
6110 list_for_each_entry(work
, &pwq
->inactive_works
, entry
) {
6111 pr_cont_work(comma
, work
, &pcws
);
6112 comma
= !(*work_data_bits(work
) & WORK_STRUCT_LINKED
);
6114 pr_cont_work_flush(comma
, (work_func_t
)-1L, &pcws
);
6120 * show_one_workqueue - dump state of specified workqueue
6121 * @wq: workqueue whose state will be printed
6123 void show_one_workqueue(struct workqueue_struct
*wq
)
6125 struct pool_workqueue
*pwq
;
6127 unsigned long irq_flags
;
6129 for_each_pwq(pwq
, wq
) {
6130 if (!pwq_is_empty(pwq
)) {
6135 if (idle
) /* Nothing to print for idle workqueue */
6138 pr_info("workqueue %s: flags=0x%x\n", wq
->name
, wq
->flags
);
6140 for_each_pwq(pwq
, wq
) {
6141 raw_spin_lock_irqsave(&pwq
->pool
->lock
, irq_flags
);
6142 if (!pwq_is_empty(pwq
)) {
6144 * Defer printing to avoid deadlocks in console
6145 * drivers that queue work while holding locks
6146 * also taken in their write paths.
6148 printk_deferred_enter();
6150 printk_deferred_exit();
6152 raw_spin_unlock_irqrestore(&pwq
->pool
->lock
, irq_flags
);
6154 * We could be printing a lot from atomic context, e.g.
6155 * sysrq-t -> show_all_workqueues(). Avoid triggering
6158 touch_nmi_watchdog();
6164 * show_one_worker_pool - dump state of specified worker pool
6165 * @pool: worker pool whose state will be printed
6167 static void show_one_worker_pool(struct worker_pool
*pool
)
6169 struct worker
*worker
;
6171 unsigned long irq_flags
;
6172 unsigned long hung
= 0;
6174 raw_spin_lock_irqsave(&pool
->lock
, irq_flags
);
6175 if (pool
->nr_workers
== pool
->nr_idle
)
6178 /* How long the first pending work is waiting for a worker. */
6179 if (!list_empty(&pool
->worklist
))
6180 hung
= jiffies_to_msecs(jiffies
- pool
->watchdog_ts
) / 1000;
6183 * Defer printing to avoid deadlocks in console drivers that
6184 * queue work while holding locks also taken in their write
6187 printk_deferred_enter();
6188 pr_info("pool %d:", pool
->id
);
6189 pr_cont_pool_info(pool
);
6190 pr_cont(" hung=%lus workers=%d", hung
, pool
->nr_workers
);
6192 pr_cont(" manager: %d",
6193 task_pid_nr(pool
->manager
->task
));
6194 list_for_each_entry(worker
, &pool
->idle_list
, entry
) {
6195 pr_cont(" %s", first
? "idle: " : "");
6196 pr_cont_worker_id(worker
);
6200 printk_deferred_exit();
6202 raw_spin_unlock_irqrestore(&pool
->lock
, irq_flags
);
6204 * We could be printing a lot from atomic context, e.g.
6205 * sysrq-t -> show_all_workqueues(). Avoid triggering
6208 touch_nmi_watchdog();
6213 * show_all_workqueues - dump workqueue state
6215 * Called from a sysrq handler and prints out all busy workqueues and pools.
6217 void show_all_workqueues(void)
6219 struct workqueue_struct
*wq
;
6220 struct worker_pool
*pool
;
6225 pr_info("Showing busy workqueues and worker pools:\n");
6227 list_for_each_entry_rcu(wq
, &workqueues
, list
)
6228 show_one_workqueue(wq
);
6230 for_each_pool(pool
, pi
)
6231 show_one_worker_pool(pool
);
6237 * show_freezable_workqueues - dump freezable workqueue state
6239 * Called from try_to_freeze_tasks() and prints out all freezable workqueues
6242 void show_freezable_workqueues(void)
6244 struct workqueue_struct
*wq
;
6248 pr_info("Showing freezable workqueues that are still busy:\n");
6250 list_for_each_entry_rcu(wq
, &workqueues
, list
) {
6251 if (!(wq
->flags
& WQ_FREEZABLE
))
6253 show_one_workqueue(wq
);
6259 /* used to show worker information through /proc/PID/{comm,stat,status} */
6260 void wq_worker_comm(char *buf
, size_t size
, struct task_struct
*task
)
6264 /* always show the actual comm */
6265 off
= strscpy(buf
, task
->comm
, size
);
6269 /* stabilize PF_WQ_WORKER and worker pool association */
6270 mutex_lock(&wq_pool_attach_mutex
);
6272 if (task
->flags
& PF_WQ_WORKER
) {
6273 struct worker
*worker
= kthread_data(task
);
6274 struct worker_pool
*pool
= worker
->pool
;
6277 raw_spin_lock_irq(&pool
->lock
);
6279 * ->desc tracks information (wq name or
6280 * set_worker_desc()) for the latest execution. If
6281 * current, prepend '+', otherwise '-'.
6283 if (worker
->desc
[0] != '\0') {
6284 if (worker
->current_work
)
6285 scnprintf(buf
+ off
, size
- off
, "+%s",
6288 scnprintf(buf
+ off
, size
- off
, "-%s",
6291 raw_spin_unlock_irq(&pool
->lock
);
6295 mutex_unlock(&wq_pool_attach_mutex
);
6303 * There are two challenges in supporting CPU hotplug. Firstly, there
6304 * are a lot of assumptions on strong associations among work, pwq and
6305 * pool which make migrating pending and scheduled works very
6306 * difficult to implement without impacting hot paths. Secondly,
6307 * worker pools serve mix of short, long and very long running works making
6308 * blocked draining impractical.
6310 * This is solved by allowing the pools to be disassociated from the CPU
6311 * running as an unbound one and allowing it to be reattached later if the
6312 * cpu comes back online.
6315 static void unbind_workers(int cpu
)
6317 struct worker_pool
*pool
;
6318 struct worker
*worker
;
6320 for_each_cpu_worker_pool(pool
, cpu
) {
6321 mutex_lock(&wq_pool_attach_mutex
);
6322 raw_spin_lock_irq(&pool
->lock
);
6325 * We've blocked all attach/detach operations. Make all workers
6326 * unbound and set DISASSOCIATED. Before this, all workers
6327 * must be on the cpu. After this, they may become diasporas.
6328 * And the preemption disabled section in their sched callbacks
6329 * are guaranteed to see WORKER_UNBOUND since the code here
6330 * is on the same cpu.
6332 for_each_pool_worker(worker
, pool
)
6333 worker
->flags
|= WORKER_UNBOUND
;
6335 pool
->flags
|= POOL_DISASSOCIATED
;
6338 * The handling of nr_running in sched callbacks are disabled
6339 * now. Zap nr_running. After this, nr_running stays zero and
6340 * need_more_worker() and keep_working() are always true as
6341 * long as the worklist is not empty. This pool now behaves as
6342 * an unbound (in terms of concurrency management) pool which
6343 * are served by workers tied to the pool.
6345 pool
->nr_running
= 0;
6348 * With concurrency management just turned off, a busy
6349 * worker blocking could lead to lengthy stalls. Kick off
6350 * unbound chain execution of currently pending work items.
6354 raw_spin_unlock_irq(&pool
->lock
);
6356 for_each_pool_worker(worker
, pool
)
6357 unbind_worker(worker
);
6359 mutex_unlock(&wq_pool_attach_mutex
);
6364 * rebind_workers - rebind all workers of a pool to the associated CPU
6365 * @pool: pool of interest
6367 * @pool->cpu is coming online. Rebind all workers to the CPU.
6369 static void rebind_workers(struct worker_pool
*pool
)
6371 struct worker
*worker
;
6373 lockdep_assert_held(&wq_pool_attach_mutex
);
6376 * Restore CPU affinity of all workers. As all idle workers should
6377 * be on the run-queue of the associated CPU before any local
6378 * wake-ups for concurrency management happen, restore CPU affinity
6379 * of all workers first and then clear UNBOUND. As we're called
6380 * from CPU_ONLINE, the following shouldn't fail.
6382 for_each_pool_worker(worker
, pool
) {
6383 kthread_set_per_cpu(worker
->task
, pool
->cpu
);
6384 WARN_ON_ONCE(set_cpus_allowed_ptr(worker
->task
,
6385 pool_allowed_cpus(pool
)) < 0);
6388 raw_spin_lock_irq(&pool
->lock
);
6390 pool
->flags
&= ~POOL_DISASSOCIATED
;
6392 for_each_pool_worker(worker
, pool
) {
6393 unsigned int worker_flags
= worker
->flags
;
6396 * We want to clear UNBOUND but can't directly call
6397 * worker_clr_flags() or adjust nr_running. Atomically
6398 * replace UNBOUND with another NOT_RUNNING flag REBOUND.
6399 * @worker will clear REBOUND using worker_clr_flags() when
6400 * it initiates the next execution cycle thus restoring
6401 * concurrency management. Note that when or whether
6402 * @worker clears REBOUND doesn't affect correctness.
6404 * WRITE_ONCE() is necessary because @worker->flags may be
6405 * tested without holding any lock in
6406 * wq_worker_running(). Without it, NOT_RUNNING test may
6407 * fail incorrectly leading to premature concurrency
6408 * management operations.
6410 WARN_ON_ONCE(!(worker_flags
& WORKER_UNBOUND
));
6411 worker_flags
|= WORKER_REBOUND
;
6412 worker_flags
&= ~WORKER_UNBOUND
;
6413 WRITE_ONCE(worker
->flags
, worker_flags
);
6416 raw_spin_unlock_irq(&pool
->lock
);
6420 * restore_unbound_workers_cpumask - restore cpumask of unbound workers
6421 * @pool: unbound pool of interest
6422 * @cpu: the CPU which is coming up
6424 * An unbound pool may end up with a cpumask which doesn't have any online
6425 * CPUs. When a worker of such pool get scheduled, the scheduler resets
6426 * its cpus_allowed. If @cpu is in @pool's cpumask which didn't have any
6427 * online CPU before, cpus_allowed of all its workers should be restored.
6429 static void restore_unbound_workers_cpumask(struct worker_pool
*pool
, int cpu
)
6431 static cpumask_t cpumask
;
6432 struct worker
*worker
;
6434 lockdep_assert_held(&wq_pool_attach_mutex
);
6436 /* is @cpu allowed for @pool? */
6437 if (!cpumask_test_cpu(cpu
, pool
->attrs
->cpumask
))
6440 cpumask_and(&cpumask
, pool
->attrs
->cpumask
, cpu_online_mask
);
6442 /* as we're called from CPU_ONLINE, the following shouldn't fail */
6443 for_each_pool_worker(worker
, pool
)
6444 WARN_ON_ONCE(set_cpus_allowed_ptr(worker
->task
, &cpumask
) < 0);
6447 int workqueue_prepare_cpu(unsigned int cpu
)
6449 struct worker_pool
*pool
;
6451 for_each_cpu_worker_pool(pool
, cpu
) {
6452 if (pool
->nr_workers
)
6454 if (!create_worker(pool
))
6460 int workqueue_online_cpu(unsigned int cpu
)
6462 struct worker_pool
*pool
;
6463 struct workqueue_struct
*wq
;
6466 mutex_lock(&wq_pool_mutex
);
6468 for_each_pool(pool
, pi
) {
6469 /* BH pools aren't affected by hotplug */
6470 if (pool
->flags
& POOL_BH
)
6473 mutex_lock(&wq_pool_attach_mutex
);
6474 if (pool
->cpu
== cpu
)
6475 rebind_workers(pool
);
6476 else if (pool
->cpu
< 0)
6477 restore_unbound_workers_cpumask(pool
, cpu
);
6478 mutex_unlock(&wq_pool_attach_mutex
);
6481 /* update pod affinity of unbound workqueues */
6482 list_for_each_entry(wq
, &workqueues
, list
) {
6483 struct workqueue_attrs
*attrs
= wq
->unbound_attrs
;
6486 const struct wq_pod_type
*pt
= wqattrs_pod_type(attrs
);
6489 for_each_cpu(tcpu
, pt
->pod_cpus
[pt
->cpu_pod
[cpu
]])
6490 wq_update_pod(wq
, tcpu
, cpu
, true);
6492 mutex_lock(&wq
->mutex
);
6493 wq_update_node_max_active(wq
, -1);
6494 mutex_unlock(&wq
->mutex
);
6498 mutex_unlock(&wq_pool_mutex
);
6502 int workqueue_offline_cpu(unsigned int cpu
)
6504 struct workqueue_struct
*wq
;
6506 /* unbinding per-cpu workers should happen on the local CPU */
6507 if (WARN_ON(cpu
!= smp_processor_id()))
6510 unbind_workers(cpu
);
6512 /* update pod affinity of unbound workqueues */
6513 mutex_lock(&wq_pool_mutex
);
6514 list_for_each_entry(wq
, &workqueues
, list
) {
6515 struct workqueue_attrs
*attrs
= wq
->unbound_attrs
;
6518 const struct wq_pod_type
*pt
= wqattrs_pod_type(attrs
);
6521 for_each_cpu(tcpu
, pt
->pod_cpus
[pt
->cpu_pod
[cpu
]])
6522 wq_update_pod(wq
, tcpu
, cpu
, false);
6524 mutex_lock(&wq
->mutex
);
6525 wq_update_node_max_active(wq
, cpu
);
6526 mutex_unlock(&wq
->mutex
);
6529 mutex_unlock(&wq_pool_mutex
);
6534 struct work_for_cpu
{
6535 struct work_struct work
;
6541 static void work_for_cpu_fn(struct work_struct
*work
)
6543 struct work_for_cpu
*wfc
= container_of(work
, struct work_for_cpu
, work
);
6545 wfc
->ret
= wfc
->fn(wfc
->arg
);
6549 * work_on_cpu_key - run a function in thread context on a particular cpu
6550 * @cpu: the cpu to run on
6551 * @fn: the function to run
6552 * @arg: the function arg
6553 * @key: The lock class key for lock debugging purposes
6555 * It is up to the caller to ensure that the cpu doesn't go offline.
6556 * The caller must not hold any locks which would prevent @fn from completing.
6558 * Return: The value @fn returns.
6560 long work_on_cpu_key(int cpu
, long (*fn
)(void *),
6561 void *arg
, struct lock_class_key
*key
)
6563 struct work_for_cpu wfc
= { .fn
= fn
, .arg
= arg
};
6565 INIT_WORK_ONSTACK_KEY(&wfc
.work
, work_for_cpu_fn
, key
);
6566 schedule_work_on(cpu
, &wfc
.work
);
6567 flush_work(&wfc
.work
);
6568 destroy_work_on_stack(&wfc
.work
);
6571 EXPORT_SYMBOL_GPL(work_on_cpu_key
);
6574 * work_on_cpu_safe_key - run a function in thread context on a particular cpu
6575 * @cpu: the cpu to run on
6576 * @fn: the function to run
6577 * @arg: the function argument
6578 * @key: The lock class key for lock debugging purposes
6580 * Disables CPU hotplug and calls work_on_cpu(). The caller must not hold
6581 * any locks which would prevent @fn from completing.
6583 * Return: The value @fn returns.
6585 long work_on_cpu_safe_key(int cpu
, long (*fn
)(void *),
6586 void *arg
, struct lock_class_key
*key
)
6591 if (cpu_online(cpu
))
6592 ret
= work_on_cpu_key(cpu
, fn
, arg
, key
);
6596 EXPORT_SYMBOL_GPL(work_on_cpu_safe_key
);
6597 #endif /* CONFIG_SMP */
6599 #ifdef CONFIG_FREEZER
6602 * freeze_workqueues_begin - begin freezing workqueues
6604 * Start freezing workqueues. After this function returns, all freezable
6605 * workqueues will queue new works to their inactive_works list instead of
6609 * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's.
6611 void freeze_workqueues_begin(void)
6613 struct workqueue_struct
*wq
;
6615 mutex_lock(&wq_pool_mutex
);
6617 WARN_ON_ONCE(workqueue_freezing
);
6618 workqueue_freezing
= true;
6620 list_for_each_entry(wq
, &workqueues
, list
) {
6621 mutex_lock(&wq
->mutex
);
6622 wq_adjust_max_active(wq
);
6623 mutex_unlock(&wq
->mutex
);
6626 mutex_unlock(&wq_pool_mutex
);
6630 * freeze_workqueues_busy - are freezable workqueues still busy?
6632 * Check whether freezing is complete. This function must be called
6633 * between freeze_workqueues_begin() and thaw_workqueues().
6636 * Grabs and releases wq_pool_mutex.
6639 * %true if some freezable workqueues are still busy. %false if freezing
6642 bool freeze_workqueues_busy(void)
6645 struct workqueue_struct
*wq
;
6646 struct pool_workqueue
*pwq
;
6648 mutex_lock(&wq_pool_mutex
);
6650 WARN_ON_ONCE(!workqueue_freezing
);
6652 list_for_each_entry(wq
, &workqueues
, list
) {
6653 if (!(wq
->flags
& WQ_FREEZABLE
))
6656 * nr_active is monotonically decreasing. It's safe
6657 * to peek without lock.
6660 for_each_pwq(pwq
, wq
) {
6661 WARN_ON_ONCE(pwq
->nr_active
< 0);
6662 if (pwq
->nr_active
) {
6671 mutex_unlock(&wq_pool_mutex
);
6676 * thaw_workqueues - thaw workqueues
6678 * Thaw workqueues. Normal queueing is restored and all collected
6679 * frozen works are transferred to their respective pool worklists.
6682 * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's.
6684 void thaw_workqueues(void)
6686 struct workqueue_struct
*wq
;
6688 mutex_lock(&wq_pool_mutex
);
6690 if (!workqueue_freezing
)
6693 workqueue_freezing
= false;
6695 /* restore max_active and repopulate worklist */
6696 list_for_each_entry(wq
, &workqueues
, list
) {
6697 mutex_lock(&wq
->mutex
);
6698 wq_adjust_max_active(wq
);
6699 mutex_unlock(&wq
->mutex
);
6703 mutex_unlock(&wq_pool_mutex
);
6705 #endif /* CONFIG_FREEZER */
6707 static int workqueue_apply_unbound_cpumask(const cpumask_var_t unbound_cpumask
)
6711 struct workqueue_struct
*wq
;
6712 struct apply_wqattrs_ctx
*ctx
, *n
;
6714 lockdep_assert_held(&wq_pool_mutex
);
6716 list_for_each_entry(wq
, &workqueues
, list
) {
6717 if (!(wq
->flags
& WQ_UNBOUND
) || (wq
->flags
& __WQ_DESTROYING
))
6720 ctx
= apply_wqattrs_prepare(wq
, wq
->unbound_attrs
, unbound_cpumask
);
6726 list_add_tail(&ctx
->list
, &ctxs
);
6729 list_for_each_entry_safe(ctx
, n
, &ctxs
, list
) {
6731 apply_wqattrs_commit(ctx
);
6732 apply_wqattrs_cleanup(ctx
);
6736 mutex_lock(&wq_pool_attach_mutex
);
6737 cpumask_copy(wq_unbound_cpumask
, unbound_cpumask
);
6738 mutex_unlock(&wq_pool_attach_mutex
);
6744 * workqueue_unbound_exclude_cpumask - Exclude given CPUs from unbound cpumask
6745 * @exclude_cpumask: the cpumask to be excluded from wq_unbound_cpumask
6747 * This function can be called from cpuset code to provide a set of isolated
6748 * CPUs that should be excluded from wq_unbound_cpumask. The caller must hold
6749 * either cpus_read_lock or cpus_write_lock.
6751 int workqueue_unbound_exclude_cpumask(cpumask_var_t exclude_cpumask
)
6753 cpumask_var_t cpumask
;
6756 if (!zalloc_cpumask_var(&cpumask
, GFP_KERNEL
))
6759 lockdep_assert_cpus_held();
6760 mutex_lock(&wq_pool_mutex
);
6762 /* Save the current isolated cpumask & export it via sysfs */
6763 cpumask_copy(wq_isolated_cpumask
, exclude_cpumask
);
6766 * If the operation fails, it will fall back to
6767 * wq_requested_unbound_cpumask which is initially set to
6768 * (HK_TYPE_WQ ∩ HK_TYPE_DOMAIN) house keeping mask and rewritten
6769 * by any subsequent write to workqueue/cpumask sysfs file.
6771 if (!cpumask_andnot(cpumask
, wq_requested_unbound_cpumask
, exclude_cpumask
))
6772 cpumask_copy(cpumask
, wq_requested_unbound_cpumask
);
6773 if (!cpumask_equal(cpumask
, wq_unbound_cpumask
))
6774 ret
= workqueue_apply_unbound_cpumask(cpumask
);
6776 mutex_unlock(&wq_pool_mutex
);
6777 free_cpumask_var(cpumask
);
6781 static int parse_affn_scope(const char *val
)
6785 for (i
= 0; i
< ARRAY_SIZE(wq_affn_names
); i
++) {
6786 if (!strncasecmp(val
, wq_affn_names
[i
], strlen(wq_affn_names
[i
])))
6792 static int wq_affn_dfl_set(const char *val
, const struct kernel_param
*kp
)
6794 struct workqueue_struct
*wq
;
6797 affn
= parse_affn_scope(val
);
6800 if (affn
== WQ_AFFN_DFL
)
6804 mutex_lock(&wq_pool_mutex
);
6808 list_for_each_entry(wq
, &workqueues
, list
) {
6809 for_each_online_cpu(cpu
) {
6810 wq_update_pod(wq
, cpu
, cpu
, true);
6814 mutex_unlock(&wq_pool_mutex
);
6820 static int wq_affn_dfl_get(char *buffer
, const struct kernel_param
*kp
)
6822 return scnprintf(buffer
, PAGE_SIZE
, "%s\n", wq_affn_names
[wq_affn_dfl
]);
6825 static const struct kernel_param_ops wq_affn_dfl_ops
= {
6826 .set
= wq_affn_dfl_set
,
6827 .get
= wq_affn_dfl_get
,
6830 module_param_cb(default_affinity_scope
, &wq_affn_dfl_ops
, NULL
, 0644);
6834 * Workqueues with WQ_SYSFS flag set is visible to userland via
6835 * /sys/bus/workqueue/devices/WQ_NAME. All visible workqueues have the
6836 * following attributes.
6838 * per_cpu RO bool : whether the workqueue is per-cpu or unbound
6839 * max_active RW int : maximum number of in-flight work items
6841 * Unbound workqueues have the following extra attributes.
6843 * nice RW int : nice value of the workers
6844 * cpumask RW mask : bitmask of allowed CPUs for the workers
6845 * affinity_scope RW str : worker CPU affinity scope (cache, numa, none)
6846 * affinity_strict RW bool : worker CPU affinity is strict
6849 struct workqueue_struct
*wq
;
6853 static struct workqueue_struct
*dev_to_wq(struct device
*dev
)
6855 struct wq_device
*wq_dev
= container_of(dev
, struct wq_device
, dev
);
6860 static ssize_t
per_cpu_show(struct device
*dev
, struct device_attribute
*attr
,
6863 struct workqueue_struct
*wq
= dev_to_wq(dev
);
6865 return scnprintf(buf
, PAGE_SIZE
, "%d\n", (bool)!(wq
->flags
& WQ_UNBOUND
));
6867 static DEVICE_ATTR_RO(per_cpu
);
6869 static ssize_t
max_active_show(struct device
*dev
,
6870 struct device_attribute
*attr
, char *buf
)
6872 struct workqueue_struct
*wq
= dev_to_wq(dev
);
6874 return scnprintf(buf
, PAGE_SIZE
, "%d\n", wq
->saved_max_active
);
6877 static ssize_t
max_active_store(struct device
*dev
,
6878 struct device_attribute
*attr
, const char *buf
,
6881 struct workqueue_struct
*wq
= dev_to_wq(dev
);
6884 if (sscanf(buf
, "%d", &val
) != 1 || val
<= 0)
6887 workqueue_set_max_active(wq
, val
);
6890 static DEVICE_ATTR_RW(max_active
);
6892 static struct attribute
*wq_sysfs_attrs
[] = {
6893 &dev_attr_per_cpu
.attr
,
6894 &dev_attr_max_active
.attr
,
6897 ATTRIBUTE_GROUPS(wq_sysfs
);
6899 static void apply_wqattrs_lock(void)
6901 /* CPUs should stay stable across pwq creations and installations */
6903 mutex_lock(&wq_pool_mutex
);
6906 static void apply_wqattrs_unlock(void)
6908 mutex_unlock(&wq_pool_mutex
);
6912 static ssize_t
wq_nice_show(struct device
*dev
, struct device_attribute
*attr
,
6915 struct workqueue_struct
*wq
= dev_to_wq(dev
);
6918 mutex_lock(&wq
->mutex
);
6919 written
= scnprintf(buf
, PAGE_SIZE
, "%d\n", wq
->unbound_attrs
->nice
);
6920 mutex_unlock(&wq
->mutex
);
6925 /* prepare workqueue_attrs for sysfs store operations */
6926 static struct workqueue_attrs
*wq_sysfs_prep_attrs(struct workqueue_struct
*wq
)
6928 struct workqueue_attrs
*attrs
;
6930 lockdep_assert_held(&wq_pool_mutex
);
6932 attrs
= alloc_workqueue_attrs();
6936 copy_workqueue_attrs(attrs
, wq
->unbound_attrs
);
6940 static ssize_t
wq_nice_store(struct device
*dev
, struct device_attribute
*attr
,
6941 const char *buf
, size_t count
)
6943 struct workqueue_struct
*wq
= dev_to_wq(dev
);
6944 struct workqueue_attrs
*attrs
;
6947 apply_wqattrs_lock();
6949 attrs
= wq_sysfs_prep_attrs(wq
);
6953 if (sscanf(buf
, "%d", &attrs
->nice
) == 1 &&
6954 attrs
->nice
>= MIN_NICE
&& attrs
->nice
<= MAX_NICE
)
6955 ret
= apply_workqueue_attrs_locked(wq
, attrs
);
6960 apply_wqattrs_unlock();
6961 free_workqueue_attrs(attrs
);
6962 return ret
?: count
;
6965 static ssize_t
wq_cpumask_show(struct device
*dev
,
6966 struct device_attribute
*attr
, char *buf
)
6968 struct workqueue_struct
*wq
= dev_to_wq(dev
);
6971 mutex_lock(&wq
->mutex
);
6972 written
= scnprintf(buf
, PAGE_SIZE
, "%*pb\n",
6973 cpumask_pr_args(wq
->unbound_attrs
->cpumask
));
6974 mutex_unlock(&wq
->mutex
);
6978 static ssize_t
wq_cpumask_store(struct device
*dev
,
6979 struct device_attribute
*attr
,
6980 const char *buf
, size_t count
)
6982 struct workqueue_struct
*wq
= dev_to_wq(dev
);
6983 struct workqueue_attrs
*attrs
;
6986 apply_wqattrs_lock();
6988 attrs
= wq_sysfs_prep_attrs(wq
);
6992 ret
= cpumask_parse(buf
, attrs
->cpumask
);
6994 ret
= apply_workqueue_attrs_locked(wq
, attrs
);
6997 apply_wqattrs_unlock();
6998 free_workqueue_attrs(attrs
);
6999 return ret
?: count
;
7002 static ssize_t
wq_affn_scope_show(struct device
*dev
,
7003 struct device_attribute
*attr
, char *buf
)
7005 struct workqueue_struct
*wq
= dev_to_wq(dev
);
7008 mutex_lock(&wq
->mutex
);
7009 if (wq
->unbound_attrs
->affn_scope
== WQ_AFFN_DFL
)
7010 written
= scnprintf(buf
, PAGE_SIZE
, "%s (%s)\n",
7011 wq_affn_names
[WQ_AFFN_DFL
],
7012 wq_affn_names
[wq_affn_dfl
]);
7014 written
= scnprintf(buf
, PAGE_SIZE
, "%s\n",
7015 wq_affn_names
[wq
->unbound_attrs
->affn_scope
]);
7016 mutex_unlock(&wq
->mutex
);
7021 static ssize_t
wq_affn_scope_store(struct device
*dev
,
7022 struct device_attribute
*attr
,
7023 const char *buf
, size_t count
)
7025 struct workqueue_struct
*wq
= dev_to_wq(dev
);
7026 struct workqueue_attrs
*attrs
;
7027 int affn
, ret
= -ENOMEM
;
7029 affn
= parse_affn_scope(buf
);
7033 apply_wqattrs_lock();
7034 attrs
= wq_sysfs_prep_attrs(wq
);
7036 attrs
->affn_scope
= affn
;
7037 ret
= apply_workqueue_attrs_locked(wq
, attrs
);
7039 apply_wqattrs_unlock();
7040 free_workqueue_attrs(attrs
);
7041 return ret
?: count
;
7044 static ssize_t
wq_affinity_strict_show(struct device
*dev
,
7045 struct device_attribute
*attr
, char *buf
)
7047 struct workqueue_struct
*wq
= dev_to_wq(dev
);
7049 return scnprintf(buf
, PAGE_SIZE
, "%d\n",
7050 wq
->unbound_attrs
->affn_strict
);
7053 static ssize_t
wq_affinity_strict_store(struct device
*dev
,
7054 struct device_attribute
*attr
,
7055 const char *buf
, size_t count
)
7057 struct workqueue_struct
*wq
= dev_to_wq(dev
);
7058 struct workqueue_attrs
*attrs
;
7059 int v
, ret
= -ENOMEM
;
7061 if (sscanf(buf
, "%d", &v
) != 1)
7064 apply_wqattrs_lock();
7065 attrs
= wq_sysfs_prep_attrs(wq
);
7067 attrs
->affn_strict
= (bool)v
;
7068 ret
= apply_workqueue_attrs_locked(wq
, attrs
);
7070 apply_wqattrs_unlock();
7071 free_workqueue_attrs(attrs
);
7072 return ret
?: count
;
7075 static struct device_attribute wq_sysfs_unbound_attrs
[] = {
7076 __ATTR(nice
, 0644, wq_nice_show
, wq_nice_store
),
7077 __ATTR(cpumask
, 0644, wq_cpumask_show
, wq_cpumask_store
),
7078 __ATTR(affinity_scope
, 0644, wq_affn_scope_show
, wq_affn_scope_store
),
7079 __ATTR(affinity_strict
, 0644, wq_affinity_strict_show
, wq_affinity_strict_store
),
7083 static const struct bus_type wq_subsys
= {
7084 .name
= "workqueue",
7085 .dev_groups
= wq_sysfs_groups
,
7089 * workqueue_set_unbound_cpumask - Set the low-level unbound cpumask
7090 * @cpumask: the cpumask to set
7092 * The low-level workqueues cpumask is a global cpumask that limits
7093 * the affinity of all unbound workqueues. This function check the @cpumask
7094 * and apply it to all unbound workqueues and updates all pwqs of them.
7096 * Return: 0 - Success
7097 * -EINVAL - Invalid @cpumask
7098 * -ENOMEM - Failed to allocate memory for attrs or pwqs.
7100 static int workqueue_set_unbound_cpumask(cpumask_var_t cpumask
)
7105 * Not excluding isolated cpus on purpose.
7106 * If the user wishes to include them, we allow that.
7108 cpumask_and(cpumask
, cpumask
, cpu_possible_mask
);
7109 if (!cpumask_empty(cpumask
)) {
7110 apply_wqattrs_lock();
7111 cpumask_copy(wq_requested_unbound_cpumask
, cpumask
);
7112 if (cpumask_equal(cpumask
, wq_unbound_cpumask
)) {
7117 ret
= workqueue_apply_unbound_cpumask(cpumask
);
7120 apply_wqattrs_unlock();
7126 static ssize_t
__wq_cpumask_show(struct device
*dev
,
7127 struct device_attribute
*attr
, char *buf
, cpumask_var_t mask
)
7131 mutex_lock(&wq_pool_mutex
);
7132 written
= scnprintf(buf
, PAGE_SIZE
, "%*pb\n", cpumask_pr_args(mask
));
7133 mutex_unlock(&wq_pool_mutex
);
7138 static ssize_t
wq_unbound_cpumask_show(struct device
*dev
,
7139 struct device_attribute
*attr
, char *buf
)
7141 return __wq_cpumask_show(dev
, attr
, buf
, wq_unbound_cpumask
);
7144 static ssize_t
wq_requested_cpumask_show(struct device
*dev
,
7145 struct device_attribute
*attr
, char *buf
)
7147 return __wq_cpumask_show(dev
, attr
, buf
, wq_requested_unbound_cpumask
);
7150 static ssize_t
wq_isolated_cpumask_show(struct device
*dev
,
7151 struct device_attribute
*attr
, char *buf
)
7153 return __wq_cpumask_show(dev
, attr
, buf
, wq_isolated_cpumask
);
7156 static ssize_t
wq_unbound_cpumask_store(struct device
*dev
,
7157 struct device_attribute
*attr
, const char *buf
, size_t count
)
7159 cpumask_var_t cpumask
;
7162 if (!zalloc_cpumask_var(&cpumask
, GFP_KERNEL
))
7165 ret
= cpumask_parse(buf
, cpumask
);
7167 ret
= workqueue_set_unbound_cpumask(cpumask
);
7169 free_cpumask_var(cpumask
);
7170 return ret
? ret
: count
;
7173 static struct device_attribute wq_sysfs_cpumask_attrs
[] = {
7174 __ATTR(cpumask
, 0644, wq_unbound_cpumask_show
,
7175 wq_unbound_cpumask_store
),
7176 __ATTR(cpumask_requested
, 0444, wq_requested_cpumask_show
, NULL
),
7177 __ATTR(cpumask_isolated
, 0444, wq_isolated_cpumask_show
, NULL
),
7181 static int __init
wq_sysfs_init(void)
7183 struct device
*dev_root
;
7186 err
= subsys_virtual_register(&wq_subsys
, NULL
);
7190 dev_root
= bus_get_dev_root(&wq_subsys
);
7192 struct device_attribute
*attr
;
7194 for (attr
= wq_sysfs_cpumask_attrs
; attr
->attr
.name
; attr
++) {
7195 err
= device_create_file(dev_root
, attr
);
7199 put_device(dev_root
);
7203 core_initcall(wq_sysfs_init
);
7205 static void wq_device_release(struct device
*dev
)
7207 struct wq_device
*wq_dev
= container_of(dev
, struct wq_device
, dev
);
7213 * workqueue_sysfs_register - make a workqueue visible in sysfs
7214 * @wq: the workqueue to register
7216 * Expose @wq in sysfs under /sys/bus/workqueue/devices.
7217 * alloc_workqueue*() automatically calls this function if WQ_SYSFS is set
7218 * which is the preferred method.
7220 * Workqueue user should use this function directly iff it wants to apply
7221 * workqueue_attrs before making the workqueue visible in sysfs; otherwise,
7222 * apply_workqueue_attrs() may race against userland updating the
7225 * Return: 0 on success, -errno on failure.
7227 int workqueue_sysfs_register(struct workqueue_struct
*wq
)
7229 struct wq_device
*wq_dev
;
7233 * Adjusting max_active breaks ordering guarantee. Disallow exposing
7234 * ordered workqueues.
7236 if (WARN_ON(wq
->flags
& __WQ_ORDERED
))
7239 wq
->wq_dev
= wq_dev
= kzalloc(sizeof(*wq_dev
), GFP_KERNEL
);
7244 wq_dev
->dev
.bus
= &wq_subsys
;
7245 wq_dev
->dev
.release
= wq_device_release
;
7246 dev_set_name(&wq_dev
->dev
, "%s", wq
->name
);
7249 * unbound_attrs are created separately. Suppress uevent until
7250 * everything is ready.
7252 dev_set_uevent_suppress(&wq_dev
->dev
, true);
7254 ret
= device_register(&wq_dev
->dev
);
7256 put_device(&wq_dev
->dev
);
7261 if (wq
->flags
& WQ_UNBOUND
) {
7262 struct device_attribute
*attr
;
7264 for (attr
= wq_sysfs_unbound_attrs
; attr
->attr
.name
; attr
++) {
7265 ret
= device_create_file(&wq_dev
->dev
, attr
);
7267 device_unregister(&wq_dev
->dev
);
7274 dev_set_uevent_suppress(&wq_dev
->dev
, false);
7275 kobject_uevent(&wq_dev
->dev
.kobj
, KOBJ_ADD
);
7280 * workqueue_sysfs_unregister - undo workqueue_sysfs_register()
7281 * @wq: the workqueue to unregister
7283 * If @wq is registered to sysfs by workqueue_sysfs_register(), unregister.
7285 static void workqueue_sysfs_unregister(struct workqueue_struct
*wq
)
7287 struct wq_device
*wq_dev
= wq
->wq_dev
;
7293 device_unregister(&wq_dev
->dev
);
7295 #else /* CONFIG_SYSFS */
7296 static void workqueue_sysfs_unregister(struct workqueue_struct
*wq
) { }
7297 #endif /* CONFIG_SYSFS */
7300 * Workqueue watchdog.
7302 * Stall may be caused by various bugs - missing WQ_MEM_RECLAIM, illegal
7303 * flush dependency, a concurrency managed work item which stays RUNNING
7304 * indefinitely. Workqueue stalls can be very difficult to debug as the
7305 * usual warning mechanisms don't trigger and internal workqueue state is
7308 * Workqueue watchdog monitors all worker pools periodically and dumps
7309 * state if some pools failed to make forward progress for a while where
7310 * forward progress is defined as the first item on ->worklist changing.
7312 * This mechanism is controlled through the kernel parameter
7313 * "workqueue.watchdog_thresh" which can be updated at runtime through the
7314 * corresponding sysfs parameter file.
7316 #ifdef CONFIG_WQ_WATCHDOG
7318 static unsigned long wq_watchdog_thresh
= 30;
7319 static struct timer_list wq_watchdog_timer
;
7321 static unsigned long wq_watchdog_touched
= INITIAL_JIFFIES
;
7322 static DEFINE_PER_CPU(unsigned long, wq_watchdog_touched_cpu
) = INITIAL_JIFFIES
;
7325 * Show workers that might prevent the processing of pending work items.
7326 * The only candidates are CPU-bound workers in the running state.
7327 * Pending work items should be handled by another idle worker
7328 * in all other situations.
7330 static void show_cpu_pool_hog(struct worker_pool
*pool
)
7332 struct worker
*worker
;
7333 unsigned long irq_flags
;
7336 raw_spin_lock_irqsave(&pool
->lock
, irq_flags
);
7338 hash_for_each(pool
->busy_hash
, bkt
, worker
, hentry
) {
7339 if (task_is_running(worker
->task
)) {
7341 * Defer printing to avoid deadlocks in console
7342 * drivers that queue work while holding locks
7343 * also taken in their write paths.
7345 printk_deferred_enter();
7347 pr_info("pool %d:\n", pool
->id
);
7348 sched_show_task(worker
->task
);
7350 printk_deferred_exit();
7354 raw_spin_unlock_irqrestore(&pool
->lock
, irq_flags
);
7357 static void show_cpu_pools_hogs(void)
7359 struct worker_pool
*pool
;
7362 pr_info("Showing backtraces of running workers in stalled CPU-bound worker pools:\n");
7366 for_each_pool(pool
, pi
) {
7367 if (pool
->cpu_stall
)
7368 show_cpu_pool_hog(pool
);
7375 static void wq_watchdog_reset_touched(void)
7379 wq_watchdog_touched
= jiffies
;
7380 for_each_possible_cpu(cpu
)
7381 per_cpu(wq_watchdog_touched_cpu
, cpu
) = jiffies
;
7384 static void wq_watchdog_timer_fn(struct timer_list
*unused
)
7386 unsigned long thresh
= READ_ONCE(wq_watchdog_thresh
) * HZ
;
7387 bool lockup_detected
= false;
7388 bool cpu_pool_stall
= false;
7389 unsigned long now
= jiffies
;
7390 struct worker_pool
*pool
;
7398 for_each_pool(pool
, pi
) {
7399 unsigned long pool_ts
, touched
, ts
;
7401 pool
->cpu_stall
= false;
7402 if (list_empty(&pool
->worklist
))
7406 * If a virtual machine is stopped by the host it can look to
7407 * the watchdog like a stall.
7409 kvm_check_and_clear_guest_paused();
7411 /* get the latest of pool and touched timestamps */
7413 touched
= READ_ONCE(per_cpu(wq_watchdog_touched_cpu
, pool
->cpu
));
7415 touched
= READ_ONCE(wq_watchdog_touched
);
7416 pool_ts
= READ_ONCE(pool
->watchdog_ts
);
7418 if (time_after(pool_ts
, touched
))
7424 if (time_after(now
, ts
+ thresh
)) {
7425 lockup_detected
= true;
7426 if (pool
->cpu
>= 0 && !(pool
->flags
& POOL_BH
)) {
7427 pool
->cpu_stall
= true;
7428 cpu_pool_stall
= true;
7430 pr_emerg("BUG: workqueue lockup - pool");
7431 pr_cont_pool_info(pool
);
7432 pr_cont(" stuck for %us!\n",
7433 jiffies_to_msecs(now
- pool_ts
) / 1000);
7441 if (lockup_detected
)
7442 show_all_workqueues();
7445 show_cpu_pools_hogs();
7447 wq_watchdog_reset_touched();
7448 mod_timer(&wq_watchdog_timer
, jiffies
+ thresh
);
7451 notrace
void wq_watchdog_touch(int cpu
)
7454 per_cpu(wq_watchdog_touched_cpu
, cpu
) = jiffies
;
7456 wq_watchdog_touched
= jiffies
;
7459 static void wq_watchdog_set_thresh(unsigned long thresh
)
7461 wq_watchdog_thresh
= 0;
7462 del_timer_sync(&wq_watchdog_timer
);
7465 wq_watchdog_thresh
= thresh
;
7466 wq_watchdog_reset_touched();
7467 mod_timer(&wq_watchdog_timer
, jiffies
+ thresh
* HZ
);
7471 static int wq_watchdog_param_set_thresh(const char *val
,
7472 const struct kernel_param
*kp
)
7474 unsigned long thresh
;
7477 ret
= kstrtoul(val
, 0, &thresh
);
7482 wq_watchdog_set_thresh(thresh
);
7484 wq_watchdog_thresh
= thresh
;
7489 static const struct kernel_param_ops wq_watchdog_thresh_ops
= {
7490 .set
= wq_watchdog_param_set_thresh
,
7491 .get
= param_get_ulong
,
7494 module_param_cb(watchdog_thresh
, &wq_watchdog_thresh_ops
, &wq_watchdog_thresh
,
7497 static void wq_watchdog_init(void)
7499 timer_setup(&wq_watchdog_timer
, wq_watchdog_timer_fn
, TIMER_DEFERRABLE
);
7500 wq_watchdog_set_thresh(wq_watchdog_thresh
);
7503 #else /* CONFIG_WQ_WATCHDOG */
7505 static inline void wq_watchdog_init(void) { }
7507 #endif /* CONFIG_WQ_WATCHDOG */
7509 static void bh_pool_kick_normal(struct irq_work
*irq_work
)
7511 raise_softirq_irqoff(TASKLET_SOFTIRQ
);
7514 static void bh_pool_kick_highpri(struct irq_work
*irq_work
)
7516 raise_softirq_irqoff(HI_SOFTIRQ
);
7519 static void __init
restrict_unbound_cpumask(const char *name
, const struct cpumask
*mask
)
7521 if (!cpumask_intersects(wq_unbound_cpumask
, mask
)) {
7522 pr_warn("workqueue: Restricting unbound_cpumask (%*pb) with %s (%*pb) leaves no CPU, ignoring\n",
7523 cpumask_pr_args(wq_unbound_cpumask
), name
, cpumask_pr_args(mask
));
7527 cpumask_and(wq_unbound_cpumask
, wq_unbound_cpumask
, mask
);
7530 static void __init
init_cpu_worker_pool(struct worker_pool
*pool
, int cpu
, int nice
)
7532 BUG_ON(init_worker_pool(pool
));
7534 cpumask_copy(pool
->attrs
->cpumask
, cpumask_of(cpu
));
7535 cpumask_copy(pool
->attrs
->__pod_cpumask
, cpumask_of(cpu
));
7536 pool
->attrs
->nice
= nice
;
7537 pool
->attrs
->affn_strict
= true;
7538 pool
->node
= cpu_to_node(cpu
);
7541 mutex_lock(&wq_pool_mutex
);
7542 BUG_ON(worker_pool_assign_id(pool
));
7543 mutex_unlock(&wq_pool_mutex
);
7547 * workqueue_init_early - early init for workqueue subsystem
7549 * This is the first step of three-staged workqueue subsystem initialization and
7550 * invoked as soon as the bare basics - memory allocation, cpumasks and idr are
7551 * up. It sets up all the data structures and system workqueues and allows early
7552 * boot code to create workqueues and queue/cancel work items. Actual work item
7553 * execution starts only after kthreads can be created and scheduled right
7554 * before early initcalls.
7556 void __init
workqueue_init_early(void)
7558 struct wq_pod_type
*pt
= &wq_pod_types
[WQ_AFFN_SYSTEM
];
7559 int std_nice
[NR_STD_WORKER_POOLS
] = { 0, HIGHPRI_NICE_LEVEL
};
7560 void (*irq_work_fns
[2])(struct irq_work
*) = { bh_pool_kick_normal
,
7561 bh_pool_kick_highpri
};
7564 BUILD_BUG_ON(__alignof__(struct pool_workqueue
) < __alignof__(long long));
7566 BUG_ON(!alloc_cpumask_var(&wq_unbound_cpumask
, GFP_KERNEL
));
7567 BUG_ON(!alloc_cpumask_var(&wq_requested_unbound_cpumask
, GFP_KERNEL
));
7568 BUG_ON(!zalloc_cpumask_var(&wq_isolated_cpumask
, GFP_KERNEL
));
7570 cpumask_copy(wq_unbound_cpumask
, cpu_possible_mask
);
7571 restrict_unbound_cpumask("HK_TYPE_WQ", housekeeping_cpumask(HK_TYPE_WQ
));
7572 restrict_unbound_cpumask("HK_TYPE_DOMAIN", housekeeping_cpumask(HK_TYPE_DOMAIN
));
7573 if (!cpumask_empty(&wq_cmdline_cpumask
))
7574 restrict_unbound_cpumask("workqueue.unbound_cpus", &wq_cmdline_cpumask
);
7576 cpumask_copy(wq_requested_unbound_cpumask
, wq_unbound_cpumask
);
7578 pwq_cache
= KMEM_CACHE(pool_workqueue
, SLAB_PANIC
);
7580 wq_update_pod_attrs_buf
= alloc_workqueue_attrs();
7581 BUG_ON(!wq_update_pod_attrs_buf
);
7584 * If nohz_full is enabled, set power efficient workqueue as unbound.
7585 * This allows workqueue items to be moved to HK CPUs.
7587 if (housekeeping_enabled(HK_TYPE_TICK
))
7588 wq_power_efficient
= true;
7590 /* initialize WQ_AFFN_SYSTEM pods */
7591 pt
->pod_cpus
= kcalloc(1, sizeof(pt
->pod_cpus
[0]), GFP_KERNEL
);
7592 pt
->pod_node
= kcalloc(1, sizeof(pt
->pod_node
[0]), GFP_KERNEL
);
7593 pt
->cpu_pod
= kcalloc(nr_cpu_ids
, sizeof(pt
->cpu_pod
[0]), GFP_KERNEL
);
7594 BUG_ON(!pt
->pod_cpus
|| !pt
->pod_node
|| !pt
->cpu_pod
);
7596 BUG_ON(!zalloc_cpumask_var_node(&pt
->pod_cpus
[0], GFP_KERNEL
, NUMA_NO_NODE
));
7599 cpumask_copy(pt
->pod_cpus
[0], cpu_possible_mask
);
7600 pt
->pod_node
[0] = NUMA_NO_NODE
;
7603 /* initialize BH and CPU pools */
7604 for_each_possible_cpu(cpu
) {
7605 struct worker_pool
*pool
;
7608 for_each_bh_worker_pool(pool
, cpu
) {
7609 init_cpu_worker_pool(pool
, cpu
, std_nice
[i
]);
7610 pool
->flags
|= POOL_BH
;
7611 init_irq_work(bh_pool_irq_work(pool
), irq_work_fns
[i
]);
7616 for_each_cpu_worker_pool(pool
, cpu
)
7617 init_cpu_worker_pool(pool
, cpu
, std_nice
[i
++]);
7620 /* create default unbound and ordered wq attrs */
7621 for (i
= 0; i
< NR_STD_WORKER_POOLS
; i
++) {
7622 struct workqueue_attrs
*attrs
;
7624 BUG_ON(!(attrs
= alloc_workqueue_attrs()));
7625 attrs
->nice
= std_nice
[i
];
7626 unbound_std_wq_attrs
[i
] = attrs
;
7629 * An ordered wq should have only one pwq as ordering is
7630 * guaranteed by max_active which is enforced by pwqs.
7632 BUG_ON(!(attrs
= alloc_workqueue_attrs()));
7633 attrs
->nice
= std_nice
[i
];
7634 attrs
->ordered
= true;
7635 ordered_wq_attrs
[i
] = attrs
;
7638 system_wq
= alloc_workqueue("events", 0, 0);
7639 system_highpri_wq
= alloc_workqueue("events_highpri", WQ_HIGHPRI
, 0);
7640 system_long_wq
= alloc_workqueue("events_long", 0, 0);
7641 system_unbound_wq
= alloc_workqueue("events_unbound", WQ_UNBOUND
,
7643 system_freezable_wq
= alloc_workqueue("events_freezable",
7645 system_power_efficient_wq
= alloc_workqueue("events_power_efficient",
7646 WQ_POWER_EFFICIENT
, 0);
7647 system_freezable_power_efficient_wq
= alloc_workqueue("events_freezable_pwr_efficient",
7648 WQ_FREEZABLE
| WQ_POWER_EFFICIENT
,
7650 system_bh_wq
= alloc_workqueue("events_bh", WQ_BH
, 0);
7651 system_bh_highpri_wq
= alloc_workqueue("events_bh_highpri",
7652 WQ_BH
| WQ_HIGHPRI
, 0);
7653 BUG_ON(!system_wq
|| !system_highpri_wq
|| !system_long_wq
||
7654 !system_unbound_wq
|| !system_freezable_wq
||
7655 !system_power_efficient_wq
||
7656 !system_freezable_power_efficient_wq
||
7657 !system_bh_wq
|| !system_bh_highpri_wq
);
7660 static void __init
wq_cpu_intensive_thresh_init(void)
7662 unsigned long thresh
;
7665 pwq_release_worker
= kthread_create_worker(0, "pool_workqueue_release");
7666 BUG_ON(IS_ERR(pwq_release_worker
));
7668 /* if the user set it to a specific value, keep it */
7669 if (wq_cpu_intensive_thresh_us
!= ULONG_MAX
)
7673 * The default of 10ms is derived from the fact that most modern (as of
7674 * 2023) processors can do a lot in 10ms and that it's just below what
7675 * most consider human-perceivable. However, the kernel also runs on a
7676 * lot slower CPUs including microcontrollers where the threshold is way
7679 * Let's scale up the threshold upto 1 second if BogoMips is below 4000.
7680 * This is by no means accurate but it doesn't have to be. The mechanism
7681 * is still useful even when the threshold is fully scaled up. Also, as
7682 * the reports would usually be applicable to everyone, some machines
7683 * operating on longer thresholds won't significantly diminish their
7686 thresh
= 10 * USEC_PER_MSEC
;
7688 /* see init/calibrate.c for lpj -> BogoMIPS calculation */
7689 bogo
= max_t(unsigned long, loops_per_jiffy
/ 500000 * HZ
, 1);
7691 thresh
= min_t(unsigned long, thresh
* 4000 / bogo
, USEC_PER_SEC
);
7693 pr_debug("wq_cpu_intensive_thresh: lpj=%lu BogoMIPS=%lu thresh_us=%lu\n",
7694 loops_per_jiffy
, bogo
, thresh
);
7696 wq_cpu_intensive_thresh_us
= thresh
;
7700 * workqueue_init - bring workqueue subsystem fully online
7702 * This is the second step of three-staged workqueue subsystem initialization
7703 * and invoked as soon as kthreads can be created and scheduled. Workqueues have
7704 * been created and work items queued on them, but there are no kworkers
7705 * executing the work items yet. Populate the worker pools with the initial
7706 * workers and enable future kworker creations.
7708 void __init
workqueue_init(void)
7710 struct workqueue_struct
*wq
;
7711 struct worker_pool
*pool
;
7714 wq_cpu_intensive_thresh_init();
7716 mutex_lock(&wq_pool_mutex
);
7719 * Per-cpu pools created earlier could be missing node hint. Fix them
7720 * up. Also, create a rescuer for workqueues that requested it.
7722 for_each_possible_cpu(cpu
) {
7723 for_each_bh_worker_pool(pool
, cpu
)
7724 pool
->node
= cpu_to_node(cpu
);
7725 for_each_cpu_worker_pool(pool
, cpu
)
7726 pool
->node
= cpu_to_node(cpu
);
7729 list_for_each_entry(wq
, &workqueues
, list
) {
7730 WARN(init_rescuer(wq
),
7731 "workqueue: failed to create early rescuer for %s",
7735 mutex_unlock(&wq_pool_mutex
);
7738 * Create the initial workers. A BH pool has one pseudo worker that
7739 * represents the shared BH execution context and thus doesn't get
7740 * affected by hotplug events. Create the BH pseudo workers for all
7741 * possible CPUs here.
7743 for_each_possible_cpu(cpu
)
7744 for_each_bh_worker_pool(pool
, cpu
)
7745 BUG_ON(!create_worker(pool
));
7747 for_each_online_cpu(cpu
) {
7748 for_each_cpu_worker_pool(pool
, cpu
) {
7749 pool
->flags
&= ~POOL_DISASSOCIATED
;
7750 BUG_ON(!create_worker(pool
));
7754 hash_for_each(unbound_pool_hash
, bkt
, pool
, hash_node
)
7755 BUG_ON(!create_worker(pool
));
7762 * Initialize @pt by first initializing @pt->cpu_pod[] with pod IDs according to
7763 * @cpu_shares_pod(). Each subset of CPUs that share a pod is assigned a unique
7764 * and consecutive pod ID. The rest of @pt is initialized accordingly.
7766 static void __init
init_pod_type(struct wq_pod_type
*pt
,
7767 bool (*cpus_share_pod
)(int, int))
7769 int cur
, pre
, cpu
, pod
;
7773 /* init @pt->cpu_pod[] according to @cpus_share_pod() */
7774 pt
->cpu_pod
= kcalloc(nr_cpu_ids
, sizeof(pt
->cpu_pod
[0]), GFP_KERNEL
);
7775 BUG_ON(!pt
->cpu_pod
);
7777 for_each_possible_cpu(cur
) {
7778 for_each_possible_cpu(pre
) {
7780 pt
->cpu_pod
[cur
] = pt
->nr_pods
++;
7783 if (cpus_share_pod(cur
, pre
)) {
7784 pt
->cpu_pod
[cur
] = pt
->cpu_pod
[pre
];
7790 /* init the rest to match @pt->cpu_pod[] */
7791 pt
->pod_cpus
= kcalloc(pt
->nr_pods
, sizeof(pt
->pod_cpus
[0]), GFP_KERNEL
);
7792 pt
->pod_node
= kcalloc(pt
->nr_pods
, sizeof(pt
->pod_node
[0]), GFP_KERNEL
);
7793 BUG_ON(!pt
->pod_cpus
|| !pt
->pod_node
);
7795 for (pod
= 0; pod
< pt
->nr_pods
; pod
++)
7796 BUG_ON(!zalloc_cpumask_var(&pt
->pod_cpus
[pod
], GFP_KERNEL
));
7798 for_each_possible_cpu(cpu
) {
7799 cpumask_set_cpu(cpu
, pt
->pod_cpus
[pt
->cpu_pod
[cpu
]]);
7800 pt
->pod_node
[pt
->cpu_pod
[cpu
]] = cpu_to_node(cpu
);
7804 static bool __init
cpus_dont_share(int cpu0
, int cpu1
)
7809 static bool __init
cpus_share_smt(int cpu0
, int cpu1
)
7811 #ifdef CONFIG_SCHED_SMT
7812 return cpumask_test_cpu(cpu0
, cpu_smt_mask(cpu1
));
7818 static bool __init
cpus_share_numa(int cpu0
, int cpu1
)
7820 return cpu_to_node(cpu0
) == cpu_to_node(cpu1
);
7824 * workqueue_init_topology - initialize CPU pods for unbound workqueues
7826 * This is the third step of three-staged workqueue subsystem initialization and
7827 * invoked after SMP and topology information are fully initialized. It
7828 * initializes the unbound CPU pods accordingly.
7830 void __init
workqueue_init_topology(void)
7832 struct workqueue_struct
*wq
;
7835 init_pod_type(&wq_pod_types
[WQ_AFFN_CPU
], cpus_dont_share
);
7836 init_pod_type(&wq_pod_types
[WQ_AFFN_SMT
], cpus_share_smt
);
7837 init_pod_type(&wq_pod_types
[WQ_AFFN_CACHE
], cpus_share_cache
);
7838 init_pod_type(&wq_pod_types
[WQ_AFFN_NUMA
], cpus_share_numa
);
7840 wq_topo_initialized
= true;
7842 mutex_lock(&wq_pool_mutex
);
7845 * Workqueues allocated earlier would have all CPUs sharing the default
7846 * worker pool. Explicitly call wq_update_pod() on all workqueue and CPU
7847 * combinations to apply per-pod sharing.
7849 list_for_each_entry(wq
, &workqueues
, list
) {
7850 for_each_online_cpu(cpu
)
7851 wq_update_pod(wq
, cpu
, cpu
, true);
7852 if (wq
->flags
& WQ_UNBOUND
) {
7853 mutex_lock(&wq
->mutex
);
7854 wq_update_node_max_active(wq
, -1);
7855 mutex_unlock(&wq
->mutex
);
7859 mutex_unlock(&wq_pool_mutex
);
7862 void __warn_flushing_systemwide_wq(void)
7864 pr_warn("WARNING: Flushing system-wide workqueues will be prohibited in near future.\n");
7867 EXPORT_SYMBOL(__warn_flushing_systemwide_wq
);
7869 static int __init
workqueue_unbound_cpus_setup(char *str
)
7871 if (cpulist_parse(str
, &wq_cmdline_cpumask
) < 0) {
7872 cpumask_clear(&wq_cmdline_cpumask
);
7873 pr_warn("workqueue.unbound_cpus: incorrect CPU range, using default\n");
7878 __setup("workqueue.unbound_cpus=", workqueue_unbound_cpus_setup
);