linux - kernel/workqueue.c

7878 lines · 220.4 KiB

1	// SPDX-License-Identifier: GPL-2.0-only
2	/*
3	* kernel/workqueue.c - generic async execution with shared worker pool
4	*
5	* Copyright (C) 2002 Ingo Molnar
6	*
7	* Derived from the taskqueue/keventd code by:
8	* David Woodhouse <[email protected]>
9	* Andrew Morton
10	* Kai Petzke <[email protected]>
11	* Theodore Ts'o <[email protected]>
12	*
13	* Made to use alloc_percpu by Christoph Lameter.
14	*
15	* Copyright (C) 2010 SUSE Linux Products GmbH
16	* Copyright (C) 2010 Tejun Heo <[email protected]>
17	*
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.
24	*
25	* Please read Documentation/core-api/workqueue.rst for details.
26	*/
27
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>
58
59	#include "workqueue_internal.h"
60
61	enum worker_pool_flags {
62	/*
63	* worker_pool flags
64	*
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
68	* is in effect.
69	*
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.
73	*
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.
77	*
78	* As there can only be one concurrent BH execution context per CPU, a
79	* BH pool is per-CPU and always DISASSOCIATED.
80	*/
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 */
85	};
86
87	enum worker_flags {
88	/* worker flags */
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 */
95
96	WORKER_NOT_RUNNING = WORKER_PREP \| WORKER_CPU_INTENSIVE \|
97	WORKER_UNBOUND \| WORKER_REBOUND,
98	};
99
100	enum work_cancel_flags {
101	WORK_CANCEL_DELAYED = 1 << 0, /* canceling a delayed_work */
102	};
103
104	enum wq_internal_consts {
105	NR_STD_WORKER_POOLS = 2, /* # standard pools per cpu */
106
107	UNBOUND_POOL_HASH_ORDER = 6, /* hashed by pool->attrs */
108	BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */
109
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 */
112
113	MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2,
114	/* call for help after 10ms
115	(min two ticks) */
116	MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */
117	CREATE_COOLDOWN = HZ, /* time to breath after fail */
118
119	/*
120	* Rescue workers are used only on emergencies and shared by
121	* all cpus. Give MIN_NICE.
122	*/
123	RESCUER_NICE_LEVEL = MIN_NICE,
124	HIGHPRI_NICE_LEVEL = MIN_NICE,
125
126	WQ_NAME_LEN = 32,
127	};
128
129	/*
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.
133	*/
134	#define BH_WORKER_JIFFIES msecs_to_jiffies(2)
135	#define BH_WORKER_RESTARTS 10
136
137	/*
138	* Structure fields follow one of the following exclusion rules.
139	*
140	* I: Modifiable by initialization/destruction paths and read-only for
141	* everyone else.
142	*
143	* P: Preemption protected. Disabling preemption is enough and should
144	* only be modified and accessed from the local cpu.
145	*
146	* L: pool->lock protected. Access with pool->lock held.
147	*
148	* LN: pool->lock and wq_node_nr_active->lock protected for writes. Either for
149	* reads.
150	*
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
153	* kworker.
154	*
155	* S: Only modified by worker self.
156	*
157	* A: wq_pool_attach_mutex protected.
158	*
159	* PL: wq_pool_mutex protected.
160	*
161	* PR: wq_pool_mutex protected for writes. RCU protected for reads.
162	*
163	* PW: wq_pool_mutex and wq->mutex protected for writes. Either for reads.
164	*
165	* PWR: wq_pool_mutex and wq->mutex protected for writes. Either or
166	* RCU for reads.
167	*
168	* WQ: wq->mutex protected.
169	*
170	* WR: wq->mutex protected for writes. RCU protected for reads.
171	*
172	* WO: wq->mutex protected for writes. Updated with WRITE_ONCE() and can be read
173	* with READ_ONCE() without locking.
174	*
175	* MD: wq_mayday_lock protected.
176	*
177	* WD: Used internally by the watchdog.
178	*/
179
180	/* struct worker is defined in workqueue_internal.h */
181
182	struct worker_pool {
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 */
188
189	unsigned long watchdog_ts; /* L: watchdog timestamp */
190	bool cpu_stall; /* WD: stalled cpu bound pool */
191
192	/*
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.
197	*/
198	int nr_running;
199
200	struct list_head worklist; /* L: list of pending works */
201
202	int nr_workers; /* L: total number of workers */
203	int nr_idle; /* L: currently idle workers */
204
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 */
208
209	struct timer_list mayday_timer; /* L: SOS timer for workers */
210
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 */
214
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 */
219
220	struct ida worker_ida; /* worker IDs for task name */
221
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 */
225
226	/*
227	* Destruction of pool is RCU protected to allow dereferences
228	* from get_work_pool().
229	*/
230	struct rcu_head rcu;
231	};
232
233	/*
234	* Per-pool_workqueue statistics. These can be monitored using
235	* tools/workqueue/wq_monitor.py.
236	*/
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 */
246
247	PWQ_NR_STATS,
248	};
249
250	/*
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.
255	*/
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 */
265
266	/*
267	* nr_active management and WORK_STRUCT_INACTIVE:
268	*
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.
272	*
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.
281	*/
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 */
287
288	u64 stats[PWQ_NR_STATS];
289
290	/*
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.
295	*/
296	struct kthread_work release_work;
297	struct rcu_head rcu;
298	} __aligned(1 << WORK_STRUCT_PWQ_SHIFT);
299
300	/*
301	* Structure used to wait for workqueue flush.
302	*/
303	struct wq_flusher {
304	struct list_head list; /* WQ: list of flushers */
305	int flush_color; /* WQ: flush color waiting for */
306	struct completion done; /* flush completion */
307	};
308
309	struct wq_device;
310
311	/*
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.
316	*
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
322	* round-robin order.
323	*/
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 */
329	};
330
331	/*
332	* The externally visible workqueue. It relays the issued work items to
333	* the appropriate worker_pool through its pool_workqueues.
334	*/
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 */
338
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 */
346
347	struct list_head maydays; /* MD: pwqs requesting rescue */
348	struct worker rescuer; / MD: rescue worker */
349
350	int nr_drainers; /* WQ: drain in progress */
351
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 */
357
358	struct workqueue_attrs unbound_attrs; / PW: only for unbound wqs */
359	struct pool_workqueue __rcu dfl_pwq; / PW: only for unbound wqs */
360
361	#ifdef CONFIG_SYSFS
362	struct wq_device wq_dev; / I: for sysfs interface */
363	#endif
364	#ifdef CONFIG_LOCKDEP
365	char *lock_name;
366	struct lock_class_key key;
367	struct lockdep_map lockdep_map;
368	#endif
369	char name[WQ_NAME_LEN]; /* I: workqueue name */
370
371	/*
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.
375	*/
376	struct rcu_head rcu;
377
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 */
382	};
383
384	/*
385	* Each pod type describes how CPUs should be grouped for unbound workqueues.
386	* See the comment above workqueue_attrs->affn_scope.
387	*/
388	struct wq_pod_type {
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 */
393	};
394
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",
402	};
403
404	/*
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().
410	*/
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);
416	#endif
417
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);
421
422	static bool wq_online; /* can kworkers be created yet? */
423	static bool wq_topo_initialized __read_mostly = false;
424
425	static struct kmem_cache *pwq_cache;
426
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;
429
430	/* buf for wq_update_unbound_pod_attrs(), protected by CPU hotplug exclusion */
431	static struct workqueue_attrs *wq_update_pod_attrs_buf;
432
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);
438
439	static LIST_HEAD(workqueues); /* PR: list of all workqueues */
440	static bool workqueue_freezing; /* PL: have wqs started freezing? */
441
442	/* PL&A: allowable cpus for unbound wqs and work items */
443	static cpumask_var_t wq_unbound_cpumask;
444
445	/* PL: user requested unbound cpumask via sysfs */
446	static cpumask_var_t wq_requested_unbound_cpumask;
447
448	/* PL: isolated cpumask to be excluded from unbound cpumask */
449	static cpumask_var_t wq_isolated_cpumask;
450
451	/* for further constrain wq_unbound_cpumask by cmdline parameter*/
452	static struct cpumask wq_cmdline_cpumask __initdata;
453
454	/* CPU where unbound work was last round robin scheduled from this CPU */
455	static DEFINE_PER_CPU(int, wq_rr_cpu_last);
456
457	/*
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.
461	*/
462	#ifdef CONFIG_DEBUG_WQ_FORCE_RR_CPU
463	static bool wq_debug_force_rr_cpu = true;
464	#else
465	static bool wq_debug_force_rr_cpu = false;
466	#endif
467	module_param_named(debug_force_rr_cpu, wq_debug_force_rr_cpu, bool, 0644);
468
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],
471	bh_pool_irq_works);
472
473	/* the BH worker pools */
474	static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS],
475	bh_worker_pools);
476
477	/* the per-cpu worker pools */
478	static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS],
479	cpu_worker_pools);
480
481	static DEFINE_IDR(worker_pool_idr); /* PR: idr of all pools */
482
483	/* PL: hash of all unbound pools keyed by pool->attrs */
484	static DEFINE_HASHTABLE(unbound_pool_hash, UNBOUND_POOL_HASH_ORDER);
485
486	/* I: attributes used when instantiating standard unbound pools on demand */
487	static struct workqueue_attrs *unbound_std_wq_attrs[NR_STD_WORKER_POOLS];
488
489	/* I: attributes used when instantiating ordered pools on demand */
490	static struct workqueue_attrs *ordered_wq_attrs[NR_STD_WORKER_POOLS];
491
492	/*
493	* Used to synchronize multiple cancel_sync attempts on the same work item. See
494	* work_grab_pending() and __cancel_work_sync().
495	*/
496	static DECLARE_WAIT_QUEUE_HEAD(wq_cancel_waitq);
497
498	/*
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.
502	*/
503	static struct kthread_worker *pwq_release_worker __ro_after_init;
504
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);
523
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);
528
529	#define CREATE_TRACE_POINTS
530	#include <trace/events/workqueue.h>
531
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")
536
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")
542
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]; \
546	(pool)++)
547
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]; \
551	(pool)++)
552
553	/**
554	* for_each_pool - iterate through all worker_pools in the system
555	* @pool: iteration cursor
556	* @pi: integer used for iteration
557	*
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.
561	*
562	* The if/else clause exists only for the lockdep assertion and can be
563	* ignored.
564	*/
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; })) { } \
568	else
569
570	/**
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
574	*
575	* This must be called with wq_pool_attach_mutex.
576	*
577	* The if/else clause exists only for the lockdep assertion and can be
578	* ignored.
579	*/
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; })) { } \
583	else
584
585	/**
586	* for_each_pwq - iterate through all pool_workqueues of the specified workqueue
587	* @pwq: iteration cursor
588	* @wq: the target workqueue
589	*
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.
593	*
594	* The if/else clause exists only for the lockdep assertion and can be
595	* ignored.
596	*/
597	#define for_each_pwq(pwq, wq) \
598	list_for_each_entry_rcu((pwq), &(wq)->pwqs, pwqs_node, \
599	lockdep_is_held(&(wq->mutex)))
600
601	#ifdef CONFIG_DEBUG_OBJECTS_WORK
602
603	static const struct debug_obj_descr work_debug_descr;
604
605	static void work_debug_hint(void addr)
606	{
607	return ((struct work_struct *) addr)->func;
608	}
609
610	static bool work_is_static_object(void *addr)
611	{
612	struct work_struct *work = addr;
613
614	return test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work));
615	}
616
617	/*
618	* fixup_init is called when:
619	* - an active object is initialized
620	*/
621	static bool work_fixup_init(void *addr, enum debug_obj_state state)
622	{
623	struct work_struct *work = addr;
624
625	switch (state) {
626	case ODEBUG_STATE_ACTIVE:
627	cancel_work_sync(work);
628	debug_object_init(work, &work_debug_descr);
629	return true;
630	default:
631	return false;
632	}
633	}
634
635	/*
636	* fixup_free is called when:
637	* - an active object is freed
638	*/
639	static bool work_fixup_free(void *addr, enum debug_obj_state state)
640	{
641	struct work_struct *work = addr;
642
643	switch (state) {
644	case ODEBUG_STATE_ACTIVE:
645	cancel_work_sync(work);
646	debug_object_free(work, &work_debug_descr);
647	return true;
648	default:
649	return false;
650	}
651	}
652
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,
659	};
660
661	static inline void debug_work_activate(struct work_struct *work)
662	{
663	debug_object_activate(work, &work_debug_descr);
664	}
665
666	static inline void debug_work_deactivate(struct work_struct *work)
667	{
668	debug_object_deactivate(work, &work_debug_descr);
669	}
670
671	void __init_work(struct work_struct *work, int onstack)
672	{
673	if (onstack)
674	debug_object_init_on_stack(work, &work_debug_descr);
675	else
676	debug_object_init(work, &work_debug_descr);
677	}
678	EXPORT_SYMBOL_GPL(__init_work);
679
680	void destroy_work_on_stack(struct work_struct *work)
681	{
682	debug_object_free(work, &work_debug_descr);
683	}
684	EXPORT_SYMBOL_GPL(destroy_work_on_stack);
685
686	void destroy_delayed_work_on_stack(struct delayed_work *work)
687	{
688	destroy_timer_on_stack(&work->timer);
689	debug_object_free(&work->work, &work_debug_descr);
690	}
691	EXPORT_SYMBOL_GPL(destroy_delayed_work_on_stack);
692
693	#else
694	static inline void debug_work_activate(struct work_struct *work) { }
695	static inline void debug_work_deactivate(struct work_struct *work) { }
696	#endif
697
698	/**
699	* worker_pool_assign_id - allocate ID and assign it to @pool
700	* @pool: the pool pointer of interest
701	*
702	* Returns 0 if ID in [0, WORK_OFFQ_POOL_NONE) is allocated and assigned
703	* successfully, -errno on failure.
704	*/
705	static int worker_pool_assign_id(struct worker_pool *pool)
706	{
707	int ret;
708
709	lockdep_assert_held(&wq_pool_mutex);
710
711	ret = idr_alloc(&worker_pool_idr, pool, 0, WORK_OFFQ_POOL_NONE,
712	GFP_KERNEL);
713	if (ret >= 0) {
714	pool->id = ret;
715	return 0;
716	}
717	return ret;
718	}
719
720	static struct pool_workqueue __rcu **
721	unbound_pwq_slot(struct workqueue_struct *wq, int cpu)
722	{
723	if (cpu >= 0)
724	return per_cpu_ptr(wq->cpu_pwq, cpu);
725	else
726	return &wq->dfl_pwq;
727	}
728
729	/* @cpu < 0 for dfl_pwq */
730	static struct pool_workqueue unbound_pwq(struct workqueue_struct wq, int cpu)
731	{
732	return rcu_dereference_check(*unbound_pwq_slot(wq, cpu),
733	lockdep_is_held(&wq_pool_mutex) \|\|
734	lockdep_is_held(&wq->mutex));
735	}
736
737	/**
738	* unbound_effective_cpumask - effective cpumask of an unbound workqueue
739	* @wq: workqueue of interest
740	*
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.
744	*/
745	static struct cpumask unbound_effective_cpumask(struct workqueue_struct wq)
746	{
747	return unbound_pwq(wq, -1)->pool->attrs->__pod_cpumask;
748	}
749
750	static unsigned int work_color_to_flags(int color)
751	{
752	return color << WORK_STRUCT_COLOR_SHIFT;
753	}
754
755	static int get_work_color(unsigned long work_data)
756	{
757	return (work_data >> WORK_STRUCT_COLOR_SHIFT) &
758	((1 << WORK_STRUCT_COLOR_BITS) - 1);
759	}
760
761	static int work_next_color(int color)
762	{
763	return (color + 1) % WORK_NR_COLORS;
764	}
765
766	/*
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.
770	*
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.
774	*
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.
779	*
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.
784	*/
785	static inline void set_work_data(struct work_struct *work, unsigned long data)
786	{
787	WARN_ON_ONCE(!work_pending(work));
788	atomic_long_set(&work->data, data \| work_static(work));
789	}
790
791	static void set_work_pwq(struct work_struct work, struct pool_workqueue pwq,
792	unsigned long flags)
793	{
794	set_work_data(work, (unsigned long)pwq \| WORK_STRUCT_PENDING \|
795	WORK_STRUCT_PWQ \| flags);
796	}
797
798	static void set_work_pool_and_keep_pending(struct work_struct *work,
799	int pool_id, unsigned long flags)
800	{
801	set_work_data(work, ((unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT) \|
802	WORK_STRUCT_PENDING \| flags);
803	}
804
805	static void set_work_pool_and_clear_pending(struct work_struct *work,
806	int pool_id, unsigned long flags)
807	{
808	/*
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
812	* owner.
813	*/
814	smp_wmb();
815	set_work_data(work, ((unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT) \|
816	flags);
817	/*
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:
823	*
824	* CPU#0 CPU#1
825	* ---------------------------- --------------------------------
826	*
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
832	* 6 smp_mb()
833	* 7 work->current_func() {
834	* 8 LOAD event_indicated
835	* }
836	*
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.
844	*/
845	smp_mb();
846	}
847
848	static inline struct pool_workqueue *work_struct_pwq(unsigned long data)
849	{
850	return (struct pool_workqueue *)(data & WORK_STRUCT_PWQ_MASK);
851	}
852
853	static struct pool_workqueue get_work_pwq(struct work_struct work)
854	{
855	unsigned long data = atomic_long_read(&work->data);
856
857	if (data & WORK_STRUCT_PWQ)
858	return work_struct_pwq(data);
859	else
860	return NULL;
861	}
862
863	/**
864	* get_work_pool - return the worker_pool a given work was associated with
865	* @work: the work item of interest
866	*
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.
870	*
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.
875	*
876	* Return: The worker_pool @work was last associated with. %NULL if none.
877	*/
878	static struct worker_pool get_work_pool(struct work_struct work)
879	{
880	unsigned long data = atomic_long_read(&work->data);
881	int pool_id;
882
883	assert_rcu_or_pool_mutex();
884
885	if (data & WORK_STRUCT_PWQ)
886	return work_struct_pwq(data)->pool;
887
888	pool_id = data >> WORK_OFFQ_POOL_SHIFT;
889	if (pool_id == WORK_OFFQ_POOL_NONE)
890	return NULL;
891
892	return idr_find(&worker_pool_idr, pool_id);
893	}
894
895	/**
896	* get_work_pool_id - return the worker pool ID a given work is associated with
897	* @work: the work item of interest
898	*
899	* Return: The worker_pool ID @work was last associated with.
900	* %WORK_OFFQ_POOL_NONE if none.
901	*/
902	static int get_work_pool_id(struct work_struct *work)
903	{
904	unsigned long data = atomic_long_read(&work->data);
905
906	if (data & WORK_STRUCT_PWQ)
907	return work_struct_pwq(data)->pool->id;
908
909	return data >> WORK_OFFQ_POOL_SHIFT;
910	}
911
912	static void mark_work_canceling(struct work_struct *work)
913	{
914	unsigned long pool_id = get_work_pool_id(work);
915
916	pool_id <<= WORK_OFFQ_POOL_SHIFT;
917	set_work_data(work, pool_id \| WORK_STRUCT_PENDING \| WORK_OFFQ_CANCELING);
918	}
919
920	static bool work_is_canceling(struct work_struct *work)
921	{
922	unsigned long data = atomic_long_read(&work->data);
923
924	return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING);
925	}
926
927	/*
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.
931	*/
932
933	/*
934	* Need to wake up a worker? Called from anything but currently
935	* running workers.
936	*
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.
940	*/
941	static bool need_more_worker(struct worker_pool *pool)
942	{
943	return !list_empty(&pool->worklist) && !pool->nr_running;
944	}
945
946	/* Can I start working? Called from busy but !running workers. */
947	static bool may_start_working(struct worker_pool *pool)
948	{
949	return pool->nr_idle;
950	}
951
952	/* Do I need to keep working? Called from currently running workers. */
953	static bool keep_working(struct worker_pool *pool)
954	{
955	return !list_empty(&pool->worklist) && (pool->nr_running <= 1);
956	}
957
958	/* Do we need a new worker? Called from manager. */
959	static bool need_to_create_worker(struct worker_pool *pool)
960	{
961	return need_more_worker(pool) && !may_start_working(pool);
962	}
963
964	/* Do we have too many workers and should some go away? */
965	static bool too_many_workers(struct worker_pool *pool)
966	{
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;
970
971	return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy;
972	}
973
974	/**
975	* worker_set_flags - set worker flags and adjust nr_running accordingly
976	* @worker: self
977	* @flags: flags to set
978	*
979	* Set @flags in @worker->flags and adjust nr_running accordingly.
980	*/
981	static inline void worker_set_flags(struct worker *worker, unsigned int flags)
982	{
983	struct worker_pool *pool = worker->pool;
984
985	lockdep_assert_held(&pool->lock);
986
987	/* If transitioning into NOT_RUNNING, adjust nr_running. */
988	if ((flags & WORKER_NOT_RUNNING) &&
989	!(worker->flags & WORKER_NOT_RUNNING)) {
990	pool->nr_running--;
991	}
992
993	worker->flags \|= flags;
994	}
995
996	/**
997	* worker_clr_flags - clear worker flags and adjust nr_running accordingly
998	* @worker: self
999	* @flags: flags to clear
1000	*
1001	* Clear @flags in @worker->flags and adjust nr_running accordingly.
1002	*/
1003	static inline void worker_clr_flags(struct worker *worker, unsigned int flags)
1004	{
1005	struct worker_pool *pool = worker->pool;
1006	unsigned int oflags = worker->flags;
1007
1008	lockdep_assert_held(&pool->lock);
1009
1010	worker->flags &= ~flags;
1011
1012	/*
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.
1016	*/
1017	if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING))
1018	if (!(worker->flags & WORKER_NOT_RUNNING))
1019	pool->nr_running++;
1020	}
1021
1022	/* Return the first idle worker. Called with pool->lock held. */
1023	static struct worker first_idle_worker(struct worker_pool pool)
1024	{
1025	if (unlikely(list_empty(&pool->idle_list)))
1026	return NULL;
1027
1028	return list_first_entry(&pool->idle_list, struct worker, entry);
1029	}
1030
1031	/**
1032	* worker_enter_idle - enter idle state
1033	* @worker: worker which is entering idle state
1034	*
1035	* @worker is entering idle state. Update stats and idle timer if
1036	* necessary.
1037	*
1038	* LOCKING:
1039	* raw_spin_lock_irq(pool->lock).
1040	*/
1041	static void worker_enter_idle(struct worker *worker)
1042	{
1043	struct worker_pool *pool = worker->pool;
1044
1045	if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) \|\|
1046	WARN_ON_ONCE(!list_empty(&worker->entry) &&
1047	(worker->hentry.next \|\| worker->hentry.pprev)))
1048	return;
1049
1050	/* can't use worker_set_flags(), also called from create_worker() */
1051	worker->flags \|= WORKER_IDLE;
1052	pool->nr_idle++;
1053	worker->last_active = jiffies;
1054
1055	/* idle_list is LIFO */
1056	list_add(&worker->entry, &pool->idle_list);
1057
1058	if (too_many_workers(pool) && !timer_pending(&pool->idle_timer))
1059	mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT);
1060
1061	/* Sanity check nr_running. */
1062	WARN_ON_ONCE(pool->nr_workers == pool->nr_idle && pool->nr_running);
1063	}
1064
1065	/**
1066	* worker_leave_idle - leave idle state
1067	* @worker: worker which is leaving idle state
1068	*
1069	* @worker is leaving idle state. Update stats.
1070	*
1071	* LOCKING:
1072	* raw_spin_lock_irq(pool->lock).
1073	*/
1074	static void worker_leave_idle(struct worker *worker)
1075	{
1076	struct worker_pool *pool = worker->pool;
1077
1078	if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE)))
1079	return;
1080	worker_clr_flags(worker, WORKER_IDLE);
1081	pool->nr_idle--;
1082	list_del_init(&worker->entry);
1083	}
1084
1085	/**
1086	* find_worker_executing_work - find worker which is executing a work
1087	* @pool: pool of interest
1088	* @work: work to find worker for
1089	*
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
1095	* being executed.
1096	*
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.
1103	*
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.
1110	*
1111	* CONTEXT:
1112	* raw_spin_lock_irq(pool->lock).
1113	*
1114	* Return:
1115	* Pointer to worker which is executing @work if found, %NULL
1116	* otherwise.
1117	*/
1118	static struct worker find_worker_executing_work(struct worker_pool pool,
1119	struct work_struct *work)
1120	{
1121	struct worker *worker;
1122
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)
1127	return worker;
1128
1129	return NULL;
1130	}
1131
1132	/**
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
1137	*
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
1141	* @nextp.
1142	*
1143	* CONTEXT:
1144	* raw_spin_lock_irq(pool->lock).
1145	*/
1146	static void move_linked_works(struct work_struct work, struct list_head head,
1147	struct work_struct **nextp)
1148	{
1149	struct work_struct *n;
1150
1151	/*
1152	* Linked worklist will always end before the end of the list,
1153	* use NULL for list head.
1154	*/
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))
1158	break;
1159	}
1160
1161	/*
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.
1165	*/
1166	if (nextp)
1167	*nextp = n;
1168	}
1169
1170	/**
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
1175	*
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.
1178	*
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().
1182	*
1183	* Returns %true if @work was successfully assigned to @worker. %false if @work
1184	* was punted to another worker already executing it.
1185	*/
1186	static bool assign_work(struct work_struct work, struct worker worker,
1187	struct work_struct **nextp)
1188	{
1189	struct worker_pool *pool = worker->pool;
1190	struct worker *collision;
1191
1192	lockdep_assert_held(&pool->lock);
1193
1194	/*
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.
1201	*/
1202	collision = find_worker_executing_work(pool, work);
1203	if (unlikely(collision)) {
1204	move_linked_works(work, &collision->scheduled, nextp);
1205	return false;
1206	}
1207
1208	move_linked_works(work, &worker->scheduled, nextp);
1209	return true;
1210	}
1211
1212	static struct irq_work bh_pool_irq_work(struct worker_pool pool)
1213	{
1214	int high = pool->attrs->nice == HIGHPRI_NICE_LEVEL ? 1 : 0;
1215
1216	return &per_cpu(bh_pool_irq_works, pool->cpu)[high];
1217	}
1218
1219	static void kick_bh_pool(struct worker_pool *pool)
1220	{
1221	#ifdef CONFIG_SMP
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);
1226	return;
1227	}
1228	#endif
1229	if (pool->attrs->nice == HIGHPRI_NICE_LEVEL)
1230	raise_softirq_irqoff(HI_SOFTIRQ);
1231	else
1232	raise_softirq_irqoff(TASKLET_SOFTIRQ);
1233	}
1234
1235	/**
1236	* kick_pool - wake up an idle worker if necessary
1237	* @pool: pool to kick
1238	*
1239	* @pool may have pending work items. Wake up worker if necessary. Returns
1240	* whether a worker was woken up.
1241	*/
1242	static bool kick_pool(struct worker_pool *pool)
1243	{
1244	struct worker *worker = first_idle_worker(pool);
1245	struct task_struct *p;
1246
1247	lockdep_assert_held(&pool->lock);
1248
1249	if (!need_more_worker(pool) \|\| !worker)
1250	return false;
1251
1252	if (pool->flags & POOL_BH) {
1253	kick_bh_pool(pool);
1254	return true;
1255	}
1256
1257	p = worker->task;
1258
1259	#ifdef CONFIG_SMP
1260	/*
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.
1265	*
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.
1272	*
1273	* If @pool has non-strict affinity, @worker might have ended up outside
1274	* its affinity scope. Repatriate.
1275	*/
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]++;
1282	}
1283	#endif
1284	wake_up_process(p);
1285	return true;
1286	}
1287
1288	#ifdef CONFIG_WQ_CPU_INTENSIVE_REPORT
1289
1290	/*
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.
1296	*
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.
1301	*/
1302	#define WCI_MAX_ENTS 128
1303
1304	struct wci_ent {
1305	work_func_t func;
1306	atomic64_t cnt;
1307	struct hlist_node hash_node;
1308	};
1309
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));
1314
1315	static struct wci_ent *wci_find_ent(work_func_t func)
1316	{
1317	struct wci_ent *ent;
1318
1319	hash_for_each_possible_rcu(wci_hash, ent, hash_node,
1320	(unsigned long)func) {
1321	if (ent->func == func)
1322	return ent;
1323	}
1324	return NULL;
1325	}
1326
1327	static void wq_cpu_intensive_report(work_func_t func)
1328	{
1329	struct wci_ent *ent;
1330
1331	restart:
1332	ent = wci_find_ent(func);
1333	if (ent) {
1334	u64 cnt;
1335
1336	/*
1337	* Start reporting from the warning_thresh and back off
1338	* exponentially.
1339	*/
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));
1347	return;
1348	}
1349
1350	/*
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
1353	* noise already.
1354	*/
1355	if (wci_nr_ents >= WCI_MAX_ENTS)
1356	return;
1357
1358	raw_spin_lock(&wci_lock);
1359
1360	if (wci_nr_ents >= WCI_MAX_ENTS) {
1361	raw_spin_unlock(&wci_lock);
1362	return;
1363	}
1364
1365	if (wci_find_ent(func)) {
1366	raw_spin_unlock(&wci_lock);
1367	goto restart;
1368	}
1369
1370	ent = &wci_ents[wci_nr_ents++];
1371	ent->func = func;
1372	atomic64_set(&ent->cnt, 0);
1373	hash_add_rcu(wci_hash, &ent->hash_node, (unsigned long)func);
1374
1375	raw_spin_unlock(&wci_lock);
1376
1377	goto restart;
1378	}
1379
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 */
1383
1384	/**
1385	* wq_worker_running - a worker is running again
1386	* @task: task waking up
1387	*
1388	* This function is called when a worker returns from schedule()
1389	*/
1390	void wq_worker_running(struct task_struct *task)
1391	{
1392	struct worker *worker = kthread_data(task);
1393
1394	if (!READ_ONCE(worker->sleeping))
1395	return;
1396
1397	/*
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.
1402	*/
1403	preempt_disable();
1404	if (!(worker->flags & WORKER_NOT_RUNNING))
1405	worker->pool->nr_running++;
1406	preempt_enable();
1407
1408	/*
1409	* CPU intensive auto-detection cares about how long a work item hogged
1410	* CPU without sleeping. Reset the starting timestamp on wakeup.
1411	*/
1412	worker->current_at = worker->task->se.sum_exec_runtime;
1413
1414	WRITE_ONCE(worker->sleeping, 0);
1415	}
1416
1417	/**
1418	* wq_worker_sleeping - a worker is going to sleep
1419	* @task: task going to sleep
1420	*
1421	* This function is called from schedule() when a busy worker is
1422	* going to sleep.
1423	*/
1424	void wq_worker_sleeping(struct task_struct *task)
1425	{
1426	struct worker *worker = kthread_data(task);
1427	struct worker_pool *pool;
1428
1429	/*
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.
1433	*/
1434	if (worker->flags & WORKER_NOT_RUNNING)
1435	return;
1436
1437	pool = worker->pool;
1438
1439	/* Return if preempted before wq_worker_running() was reached */
1440	if (READ_ONCE(worker->sleeping))
1441	return;
1442
1443	WRITE_ONCE(worker->sleeping, 1);
1444	raw_spin_lock_irq(&pool->lock);
1445
1446	/*
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.
1450	*/
1451	if (worker->flags & WORKER_NOT_RUNNING) {
1452	raw_spin_unlock_irq(&pool->lock);
1453	return;
1454	}
1455
1456	pool->nr_running--;
1457	if (kick_pool(pool))
1458	worker->current_pwq->stats[PWQ_STAT_CM_WAKEUP]++;
1459
1460	raw_spin_unlock_irq(&pool->lock);
1461	}
1462
1463	/**
1464	* wq_worker_tick - a scheduler tick occurred while a kworker is running
1465	* @task: task currently running
1466	*
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.
1469	*/
1470	void wq_worker_tick(struct task_struct *task)
1471	{
1472	struct worker *worker = kthread_data(task);
1473	struct pool_workqueue *pwq = worker->current_pwq;
1474	struct worker_pool *pool = worker->pool;
1475
1476	if (!pwq)
1477	return;
1478
1479	pwq->stats[PWQ_STAT_CPU_TIME] += TICK_USEC;
1480
1481	if (!wq_cpu_intensive_thresh_us)
1482	return;
1483
1484	/*
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.
1488	*
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.
1495	*/
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)
1499	return;
1500
1501	raw_spin_lock(&pool->lock);
1502
1503	worker_set_flags(worker, WORKER_CPU_INTENSIVE);
1504	wq_cpu_intensive_report(worker->current_func);
1505	pwq->stats[PWQ_STAT_CPU_INTENSIVE]++;
1506
1507	if (kick_pool(pool))
1508	pwq->stats[PWQ_STAT_CM_WAKEUP]++;
1509
1510	raw_spin_unlock(&pool->lock);
1511	}
1512
1513	/**
1514	* wq_worker_last_func - retrieve worker's last work function
1515	* @task: Task to retrieve last work function of.
1516	*
1517	* Determine the last function a worker executed. This is called from
1518	* the scheduler to get a worker's last known identity.
1519	*
1520	* CONTEXT:
1521	* raw_spin_lock_irq(rq->lock)
1522	*
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.
1527	*
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.
1532	*
1533	* Return:
1534	* The last work function %current executed as a worker, NULL if it
1535	* hasn't executed any work yet.
1536	*/
1537	work_func_t wq_worker_last_func(struct task_struct *task)
1538	{
1539	struct worker *worker = kthread_data(task);
1540
1541	return worker->last_func;
1542	}
1543
1544	/**
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
1548	*
1549	* Determine wq_node_nr_active to use for @wq on @node. Returns:
1550	*
1551	* - %NULL for per-cpu workqueues as they don't need to use shared nr_active.
1552	*
1553	* - node_nr_active[nr_node_ids] if @node is %NUMA_NO_NODE.
1554	*
1555	* - Otherwise, node_nr_active[@node].
1556	*/
1557	static struct wq_node_nr_active wq_node_nr_active(struct workqueue_struct wq,
1558	int node)
1559	{
1560	if (!(wq->flags & WQ_UNBOUND))
1561	return NULL;
1562
1563	if (node == NUMA_NO_NODE)
1564	node = nr_node_ids;
1565
1566	return wq->node_nr_active[node];
1567	}
1568
1569	/**
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
1573	*
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.
1577	*/
1578	static void wq_update_node_max_active(struct workqueue_struct *wq, int off_cpu)
1579	{
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;
1584
1585	lockdep_assert_held(&wq->mutex);
1586
1587	if (!wq_topo_initialized)
1588	return;
1589
1590	if (off_cpu >= 0 && !cpumask_test_cpu(off_cpu, effective))
1591	off_cpu = -1;
1592
1593	total_cpus = cpumask_weight_and(effective, cpu_online_mask);
1594	if (off_cpu >= 0)
1595	total_cpus--;
1596
1597	for_each_node(node) {
1598	int node_cpus;
1599
1600	node_cpus = cpumask_weight_and(effective, cpumask_of_node(node));
1601	if (off_cpu >= 0 && cpu_to_node(off_cpu) == node)
1602	node_cpus--;
1603
1604	wq_node_nr_active(wq, node)->max =
1605	clamp(DIV_ROUND_UP(max_active * node_cpus, total_cpus),
1606	min_active, max_active);
1607	}
1608
1609	wq_node_nr_active(wq, NUMA_NO_NODE)->max = min_active;
1610	}
1611
1612	/**
1613	* get_pwq - get an extra reference on the specified pool_workqueue
1614	* @pwq: pool_workqueue to get
1615	*
1616	* Obtain an extra reference on @pwq. The caller should guarantee that
1617	* @pwq has positive refcnt and be holding the matching pool->lock.
1618	*/
1619	static void get_pwq(struct pool_workqueue *pwq)
1620	{
1621	lockdep_assert_held(&pwq->pool->lock);
1622	WARN_ON_ONCE(pwq->refcnt <= 0);
1623	pwq->refcnt++;
1624	}
1625
1626	/**
1627	* put_pwq - put a pool_workqueue reference
1628	* @pwq: pool_workqueue to put
1629	*
1630	* Drop a reference of @pwq. If its refcnt reaches zero, schedule its
1631	* destruction. The caller should be holding the matching pool->lock.
1632	*/
1633	static void put_pwq(struct pool_workqueue *pwq)
1634	{
1635	lockdep_assert_held(&pwq->pool->lock);
1636	if (likely(--pwq->refcnt))
1637	return;
1638	/*
1639	* @pwq can't be released under pool->lock, bounce to a dedicated
1640	* kthread_worker to avoid A-A deadlocks.
1641	*/
1642	kthread_queue_work(pwq_release_worker, &pwq->release_work);
1643	}
1644
1645	/**
1646	* put_pwq_unlocked - put_pwq() with surrounding pool lock/unlock
1647	* @pwq: pool_workqueue to put (can be %NULL)
1648	*
1649	* put_pwq() with locking. This function also allows %NULL @pwq.
1650	*/
1651	static void put_pwq_unlocked(struct pool_workqueue *pwq)
1652	{
1653	if (pwq) {
1654	/*
1655	* As both pwqs and pools are RCU protected, the
1656	* following lock operations are safe.
1657	*/
1658	raw_spin_lock_irq(&pwq->pool->lock);
1659	put_pwq(pwq);
1660	raw_spin_unlock_irq(&pwq->pool->lock);
1661	}
1662	}
1663
1664	static bool pwq_is_empty(struct pool_workqueue *pwq)
1665	{
1666	return !pwq->nr_active && list_empty(&pwq->inactive_works);
1667	}
1668
1669	static void __pwq_activate_work(struct pool_workqueue *pwq,
1670	struct work_struct *work)
1671	{
1672	unsigned long *wdb = work_data_bits(work);
1673
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);
1680	}
1681
1682	/**
1683	* pwq_activate_work - Activate a work item if inactive
1684	* @pwq: pool_workqueue @work belongs to
1685	* @work: work item to activate
1686	*
1687	* Returns %true if activated. %false if already active.
1688	*/
1689	static bool pwq_activate_work(struct pool_workqueue *pwq,
1690	struct work_struct *work)
1691	{
1692	struct worker_pool *pool = pwq->pool;
1693	struct wq_node_nr_active *nna;
1694
1695	lockdep_assert_held(&pool->lock);
1696
1697	if (!(*work_data_bits(work) & WORK_STRUCT_INACTIVE))
1698	return false;
1699
1700	nna = wq_node_nr_active(pwq->wq, pool->node);
1701	if (nna)
1702	atomic_inc(&nna->nr);
1703
1704	pwq->nr_active++;
1705	__pwq_activate_work(pwq, work);
1706	return true;
1707	}
1708
1709	static bool tryinc_node_nr_active(struct wq_node_nr_active *nna)
1710	{
1711	int max = READ_ONCE(nna->max);
1712
1713	while (true) {
1714	int old, tmp;
1715
1716	old = atomic_read(&nna->nr);
1717	if (old >= max)
1718	return false;
1719	tmp = atomic_cmpxchg_relaxed(&nna->nr, old, old + 1);
1720	if (tmp == old)
1721	return true;
1722	}
1723	}
1724
1725	/**
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
1729	*
1730	* Try to increment nr_active for @pwq. Returns %true if an nr_active count is
1731	* successfully obtained. %false otherwise.
1732	*/
1733	static bool pwq_tryinc_nr_active(struct pool_workqueue *pwq, bool fill)
1734	{
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;
1739
1740	lockdep_assert_held(&pool->lock);
1741
1742	if (!nna) {
1743	/* BH or per-cpu workqueue, pwq->nr_active is sufficient */
1744	obtained = pwq->nr_active < READ_ONCE(wq->max_active);
1745	goto out;
1746	}
1747
1748	if (unlikely(pwq->plugged))
1749	return false;
1750
1751	/*
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.
1755	*
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.
1759	*/
1760	if (!list_empty(&pwq->pending_node) && likely(!fill))
1761	goto out;
1762
1763	obtained = tryinc_node_nr_active(nna);
1764	if (obtained)
1765	goto out;
1766
1767	/*
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.
1773	*/
1774	raw_spin_lock(&nna->lock);
1775
1776	if (list_empty(&pwq->pending_node))
1777	list_add_tail(&pwq->pending_node, &nna->pending_pwqs);
1778	else if (likely(!fill))
1779	goto out_unlock;
1780
1781	smp_mb();
1782
1783	obtained = tryinc_node_nr_active(nna);
1784
1785	/*
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.
1788	*/
1789	if (obtained && likely(!fill))
1790	list_del_init(&pwq->pending_node);
1791
1792	out_unlock:
1793	raw_spin_unlock(&nna->lock);
1794	out:
1795	if (obtained)
1796	pwq->nr_active++;
1797	return obtained;
1798	}
1799
1800	/**
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
1804	*
1805	* Activate the first inactive work item of @pwq if available and allowed by
1806	* max_active limit.
1807	*
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.
1810	*/
1811	static bool pwq_activate_first_inactive(struct pool_workqueue *pwq, bool fill)
1812	{
1813	struct work_struct *work =
1814	list_first_entry_or_null(&pwq->inactive_works,
1815	struct work_struct, entry);
1816
1817	if (work && pwq_tryinc_nr_active(pwq, fill)) {
1818	__pwq_activate_work(pwq, work);
1819	return true;
1820	} else {
1821	return false;
1822	}
1823	}
1824
1825	/**
1826	* unplug_oldest_pwq - unplug the oldest pool_workqueue
1827	* @wq: workqueue_struct where its oldest pwq is to be unplugged
1828	*
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::
1832	*
1833	* dfl_pwq --------------+ [P] - plugged
1834	* \|
1835	* v
1836	* pwqs -> A -> B [P] -> C [P] (newest)
1837	* \| \| \|
1838	* 1 3 5
1839	* \| \| \|
1840	* 2 4 6
1841	*
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.
1846	*/
1847	static void unplug_oldest_pwq(struct workqueue_struct *wq)
1848	{
1849	struct pool_workqueue *pwq;
1850
1851	lockdep_assert_held(&wq->mutex);
1852
1853	/* Caller should make sure that pwqs isn't empty before calling */
1854	pwq = list_first_entry_or_null(&wq->pwqs, struct pool_workqueue,
1855	pwqs_node);
1856	raw_spin_lock_irq(&pwq->pool->lock);
1857	if (pwq->plugged) {
1858	pwq->plugged = false;
1859	if (pwq_activate_first_inactive(pwq, true))
1860	kick_pool(pwq->pool);
1861	}
1862	raw_spin_unlock_irq(&pwq->pool->lock);
1863	}
1864
1865	/**
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
1869	*
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.
1872	*/
1873	static void node_activate_pending_pwq(struct wq_node_nr_active *nna,
1874	struct worker_pool *caller_pool)
1875	{
1876	struct worker_pool *locked_pool = caller_pool;
1877	struct pool_workqueue *pwq;
1878	struct work_struct *work;
1879
1880	lockdep_assert_held(&caller_pool->lock);
1881
1882	raw_spin_lock(&nna->lock);
1883	retry:
1884	pwq = list_first_entry_or_null(&nna->pending_pwqs,
1885	struct pool_workqueue, pending_node);
1886	if (!pwq)
1887	goto out_unlock;
1888
1889	/*
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.
1894	*/
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);
1902	goto retry;
1903	}
1904	}
1905
1906	/*
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.
1909	*/
1910	work = list_first_entry_or_null(&pwq->inactive_works,
1911	struct work_struct, entry);
1912	if (!work) {
1913	list_del_init(&pwq->pending_node);
1914	goto retry;
1915	}
1916
1917	/*
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.
1923	*/
1924	if (likely(tryinc_node_nr_active(nna))) {
1925	pwq->nr_active++;
1926	__pwq_activate_work(pwq, work);
1927
1928	if (list_empty(&pwq->inactive_works))
1929	list_del_init(&pwq->pending_node);
1930	else
1931	list_move_tail(&pwq->pending_node, &nna->pending_pwqs);
1932
1933	/* if activating a foreign pool, make sure it's running */
1934	if (pwq->pool != caller_pool)
1935	kick_pool(pwq->pool);
1936	}
1937
1938	out_unlock:
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);
1943	}
1944	}
1945
1946	/**
1947	* pwq_dec_nr_active - Retire an active count
1948	* @pwq: pool_workqueue of interest
1949	*
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.
1952	*/
1953	static void pwq_dec_nr_active(struct pool_workqueue *pwq)
1954	{
1955	struct worker_pool *pool = pwq->pool;
1956	struct wq_node_nr_active *nna = wq_node_nr_active(pwq->wq, pool->node);
1957
1958	lockdep_assert_held(&pool->lock);
1959
1960	/*
1961	* @pwq->nr_active should be decremented for both percpu and unbound
1962	* workqueues.
1963	*/
1964	pwq->nr_active--;
1965
1966	/*
1967	* For a percpu workqueue, it's simple. Just need to kick the first
1968	* inactive work item on @pwq itself.
1969	*/
1970	if (!nna) {
1971	pwq_activate_first_inactive(pwq, false);
1972	return;
1973	}
1974
1975	/*
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.
1983	*
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.
1988	*/
1989	if (atomic_dec_return(&nna->nr) >= READ_ONCE(nna->max))
1990	return;
1991
1992	if (!list_empty(&nna->pending_pwqs))
1993	node_activate_pending_pwq(nna, pool);
1994	}
1995
1996	/**
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
2000	*
2001	* A work either has completed or is removed from pending queue,
2002	* decrement nr_in_flight of its pwq and handle workqueue flushing.
2003	*
2004	* NOTE:
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.
2008	*
2009	* CONTEXT:
2010	* raw_spin_lock_irq(pool->lock).
2011	*/
2012	static void pwq_dec_nr_in_flight(struct pool_workqueue *pwq, unsigned long work_data)
2013	{
2014	int color = get_work_color(work_data);
2015
2016	if (!(work_data & WORK_STRUCT_INACTIVE))
2017	pwq_dec_nr_active(pwq);
2018
2019	pwq->nr_in_flight[color]--;
2020
2021	/* is flush in progress and are we at the flushing tip? */
2022	if (likely(pwq->flush_color != color))
2023	goto out_put;
2024
2025	/* are there still in-flight works? */
2026	if (pwq->nr_in_flight[color])
2027	goto out_put;
2028
2029	/* this pwq is done, clear flush_color */
2030	pwq->flush_color = -1;
2031
2032	/*
2033	* If this was the last pwq, wake up the first flusher. It
2034	* will handle the rest.
2035	*/
2036	if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush))
2037	complete(&pwq->wq->first_flusher->done);
2038	out_put:
2039	put_pwq(pwq);
2040	}
2041
2042	/**
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
2047	*
2048	* Try to grab PENDING bit of @work. This function can handle @work in any
2049	* stable state - idle, on timer or on worklist.
2050	*
2051	* Return:
2052	*
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	* ======== ================================================================
2060	*
2061	* Note:
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.
2066	*
2067	* On successful return, >= 0, irq is disabled and the caller is
2068	* responsible for releasing it using local_irq_restore(*@irq_flags).
2069	*
2070	* This function is safe to call from any context including IRQ handler.
2071	*/
2072	static int try_to_grab_pending(struct work_struct *work, u32 cflags,
2073	unsigned long *irq_flags)
2074	{
2075	struct worker_pool *pool;
2076	struct pool_workqueue *pwq;
2077
2078	local_irq_save(*irq_flags);
2079
2080	/* try to steal the timer if it exists */
2081	if (cflags & WORK_CANCEL_DELAYED) {
2082	struct delayed_work *dwork = to_delayed_work(work);
2083
2084	/*
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.
2088	*/
2089	if (likely(del_timer(&dwork->timer)))
2090	return 1;
2091	}
2092
2093	/* try to claim PENDING the normal way */
2094	if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work)))
2095	return 0;
2096
2097	rcu_read_lock();
2098	/*
2099	* The queueing is in progress, or it is already queued. Try to
2100	* steal it from ->worklist without clearing WORK_STRUCT_PENDING.
2101	*/
2102	pool = get_work_pool(work);
2103	if (!pool)
2104	goto fail;
2105
2106	raw_spin_lock(&pool->lock);
2107	/*
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.
2114	*/
2115	pwq = get_work_pwq(work);
2116	if (pwq && pwq->pool == pool) {
2117	unsigned long work_data;
2118
2119	debug_work_deactivate(work);
2120
2121	/*
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()).
2125	*
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.
2131	*/
2132	pwq_activate_work(pwq, work);
2133
2134	list_del_init(&work->entry);
2135
2136	/*
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.
2139	*/
2140	work_data = *work_data_bits(work);
2141	set_work_pool_and_keep_pending(work, pool->id, 0);
2142
2143	/* must be the last step, see the function comment */
2144	pwq_dec_nr_in_flight(pwq, work_data);
2145
2146	raw_spin_unlock(&pool->lock);
2147	rcu_read_unlock();
2148	return 1;
2149	}
2150	raw_spin_unlock(&pool->lock);
2151	fail:
2152	rcu_read_unlock();
2153	local_irq_restore(*irq_flags);
2154	if (work_is_canceling(work))
2155	return -ENOENT;
2156	cpu_relax();
2157	return -EAGAIN;
2158	}
2159
2160	struct cwt_wait {
2161	wait_queue_entry_t wait;
2162	struct work_struct *work;
2163	};
2164
2165	static int cwt_wakefn(wait_queue_entry_t wait, unsigned mode, int sync, void key)
2166	{
2167	struct cwt_wait *cwait = container_of(wait, struct cwt_wait, wait);
2168
2169	if (cwait->work != key)
2170	return 0;
2171	return autoremove_wake_function(wait, mode, sync, key);
2172	}
2173
2174	/**
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
2179	*
2180	* Grab PENDING bit of @work. @work can be in any stable state - idle, on timer
2181	* or on worklist.
2182	*
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().
2186	*
2187	* Returns %true if @work was pending. %false if idle.
2188	*/
2189	static bool work_grab_pending(struct work_struct *work, u32 cflags,
2190	unsigned long *irq_flags)
2191	{
2192	struct cwt_wait cwait;
2193	int ret;
2194
2195	might_sleep();
2196	repeat:
2197	ret = try_to_grab_pending(work, cflags, irq_flags);
2198	if (likely(ret >= 0))
2199	return ret;
2200	if (ret != -ENOENT)
2201	goto repeat;
2202
2203	/*
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.
2211	*
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.
2215	*/
2216	init_wait(&cwait.wait);
2217	cwait.wait.func = cwt_wakefn;
2218	cwait.work = work;
2219
2220	prepare_to_wait_exclusive(&wq_cancel_waitq, &cwait.wait,
2221	TASK_UNINTERRUPTIBLE);
2222	if (work_is_canceling(work))
2223	schedule();
2224	finish_wait(&wq_cancel_waitq, &cwait.wait);
2225
2226	goto repeat;
2227	}
2228
2229	/**
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
2235	*
2236	* Insert @work which belongs to @pwq after @head. @extra_flags is or'd to
2237	* work_struct flags.
2238	*
2239	* CONTEXT:
2240	* raw_spin_lock_irq(pool->lock).
2241	*/
2242	static void insert_work(struct pool_workqueue pwq, struct work_struct work,
2243	struct list_head *head, unsigned int extra_flags)
2244	{
2245	debug_work_activate(work);
2246
2247	/* record the work call stack in order to print it in KASAN reports */
2248	kasan_record_aux_stack_noalloc(work);
2249
2250	/* we own @work, set data and link */
2251	set_work_pwq(work, pwq, extra_flags);
2252	list_add_tail(&work->entry, head);
2253	get_pwq(pwq);
2254	}
2255
2256	/*
2257	* Test whether @work is being queued from another work executing on the
2258	* same workqueue.
2259	*/
2260	static bool is_chained_work(struct workqueue_struct *wq)
2261	{
2262	struct worker *worker;
2263
2264	worker = current_wq_worker();
2265	/*
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.
2268	*/
2269	return worker && worker->current_pwq->wq == wq;
2270	}
2271
2272	/*
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.
2276	*/
2277	static int wq_select_unbound_cpu(int cpu)
2278	{
2279	int new_cpu;
2280
2281	if (likely(!wq_debug_force_rr_cpu)) {
2282	if (cpumask_test_cpu(cpu, wq_unbound_cpumask))
2283	return cpu;
2284	} else {
2285	pr_warn_once("workqueue: round-robin CPU selection forced, expect performance impact\n");
2286	}
2287
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))
2293	return cpu;
2294	}
2295	__this_cpu_write(wq_rr_cpu_last, new_cpu);
2296
2297	return new_cpu;
2298	}
2299
2300	static void __queue_work(int cpu, struct workqueue_struct *wq,
2301	struct work_struct *work)
2302	{
2303	struct pool_workqueue *pwq;
2304	struct worker_pool last_pool, pool;
2305	unsigned int work_flags;
2306	unsigned int req_cpu = cpu;
2307
2308	/*
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.
2313	*/
2314	lockdep_assert_irqs_disabled();
2315
2316	/*
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).
2320	*/
2321	if (unlikely(wq->flags & (__WQ_DESTROYING \| __WQ_DRAINING) &&
2322	WARN_ON_ONCE(!is_chained_work(wq))))
2323	return;
2324	rcu_read_lock();
2325	retry:
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());
2330	else
2331	cpu = raw_smp_processor_id();
2332	}
2333
2334	pwq = rcu_dereference(*per_cpu_ptr(wq->cpu_pwq, cpu));
2335	pool = pwq->pool;
2336
2337	/*
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.
2341	*/
2342	last_pool = get_work_pool(work);
2343	if (last_pool && last_pool != pool) {
2344	struct worker *worker;
2345
2346	raw_spin_lock(&last_pool->lock);
2347
2348	worker = find_worker_executing_work(last_pool, work);
2349
2350	if (worker && worker->current_pwq->wq == wq) {
2351	pwq = worker->current_pwq;
2352	pool = pwq->pool;
2353	WARN_ON_ONCE(pool != last_pool);
2354	} else {
2355	/* meh... not running there, queue here */
2356	raw_spin_unlock(&last_pool->lock);
2357	raw_spin_lock(&pool->lock);
2358	}
2359	} else {
2360	raw_spin_lock(&pool->lock);
2361	}
2362
2363	/*
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.
2369	*/
2370	if (unlikely(!pwq->refcnt)) {
2371	if (wq->flags & WQ_UNBOUND) {
2372	raw_spin_unlock(&pool->lock);
2373	cpu_relax();
2374	goto retry;
2375	}
2376	/* oops */
2377	WARN_ONCE(true, "workqueue: per-cpu pwq for %s on cpu%d has 0 refcnt",
2378	wq->name, cpu);
2379	}
2380
2381	/* pwq determined, queue */
2382	trace_workqueue_queue_work(req_cpu, pwq, work);
2383
2384	if (WARN_ON(!list_empty(&work->entry)))
2385	goto out;
2386
2387	pwq->nr_in_flight[pwq->work_color]++;
2388	work_flags = work_color_to_flags(pwq->work_color);
2389
2390	/*
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().
2394	*/
2395	if (list_empty(&pwq->inactive_works) && pwq_tryinc_nr_active(pwq, false)) {
2396	if (list_empty(&pool->worklist))
2397	pool->watchdog_ts = jiffies;
2398
2399	trace_workqueue_activate_work(work);
2400	insert_work(pwq, work, &pool->worklist, work_flags);
2401	kick_pool(pool);
2402	} else {
2403	work_flags \|= WORK_STRUCT_INACTIVE;
2404	insert_work(pwq, work, &pwq->inactive_works, work_flags);
2405	}
2406
2407	out:
2408	raw_spin_unlock(&pool->lock);
2409	rcu_read_unlock();
2410	}
2411
2412	/**
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
2417	*
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.
2423	*
2424	* Return: %false if @work was already on a queue, %true otherwise.
2425	*/
2426	bool queue_work_on(int cpu, struct workqueue_struct *wq,
2427	struct work_struct *work)
2428	{
2429	bool ret = false;
2430	unsigned long irq_flags;
2431
2432	local_irq_save(irq_flags);
2433
2434	if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
2435	__queue_work(cpu, wq, work);
2436	ret = true;
2437	}
2438
2439	local_irq_restore(irq_flags);
2440	return ret;
2441	}
2442	EXPORT_SYMBOL(queue_work_on);
2443
2444	/**
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
2447	*
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.
2452	*/
2453	static int select_numa_node_cpu(int node)
2454	{
2455	int cpu;
2456
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;
2460
2461	/* Use local node/cpu if we are already there */
2462	cpu = raw_smp_processor_id();
2463	if (node == cpu_to_node(cpu))
2464	return cpu;
2465
2466	/* Use "random" otherwise know as "first" online CPU of node */
2467	cpu = cpumask_any_and(cpumask_of_node(node), cpu_online_mask);
2468
2469	/* If CPU is valid return that, otherwise just defer */
2470	return cpu < nr_cpu_ids ? cpu : WORK_CPU_UNBOUND;
2471	}
2472
2473	/**
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
2478	*
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
2481	* NUMA node.
2482	*
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.
2486	*
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.
2490	*
2491	* Return: %false if @work was already on a queue, %true otherwise.
2492	*/
2493	bool queue_work_node(int node, struct workqueue_struct *wq,
2494	struct work_struct *work)
2495	{
2496	unsigned long irq_flags;
2497	bool ret = false;
2498
2499	/*
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.
2503	*
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.
2507	*/
2508	WARN_ON_ONCE(!(wq->flags & WQ_UNBOUND));
2509
2510	local_irq_save(irq_flags);
2511
2512	if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
2513	int cpu = select_numa_node_cpu(node);
2514
2515	__queue_work(cpu, wq, work);
2516	ret = true;
2517	}
2518
2519	local_irq_restore(irq_flags);
2520	return ret;
2521	}
2522	EXPORT_SYMBOL_GPL(queue_work_node);
2523
2524	void delayed_work_timer_fn(struct timer_list *t)
2525	{
2526	struct delayed_work *dwork = from_timer(dwork, t, timer);
2527
2528	/* should have been called from irqsafe timer with irq already off */
2529	__queue_work(dwork->cpu, dwork->wq, &dwork->work);
2530	}
2531	EXPORT_SYMBOL(delayed_work_timer_fn);
2532
2533	static void __queue_delayed_work(int cpu, struct workqueue_struct *wq,
2534	struct delayed_work *dwork, unsigned long delay)
2535	{
2536	struct timer_list *timer = &dwork->timer;
2537	struct work_struct *work = &dwork->work;
2538
2539	WARN_ON_ONCE(!wq);
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));
2543
2544	/*
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.
2549	*/
2550	if (!delay) {
2551	__queue_work(cpu, wq, &dwork->work);
2552	return;
2553	}
2554
2555	dwork->wq = wq;
2556	dwork->cpu = cpu;
2557	timer->expires = jiffies + delay;
2558
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);
2565	} else {
2566	if (likely(cpu == WORK_CPU_UNBOUND))
2567	add_timer_global(timer);
2568	else
2569	add_timer_on(timer, cpu);
2570	}
2571	}
2572
2573	/**
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
2579	*
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
2582	* execution.
2583	*/
2584	bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
2585	struct delayed_work *dwork, unsigned long delay)
2586	{
2587	struct work_struct *work = &dwork->work;
2588	bool ret = false;
2589	unsigned long irq_flags;
2590
2591	/* read the comment in __queue_work() */
2592	local_irq_save(irq_flags);
2593
2594	if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
2595	__queue_delayed_work(cpu, wq, dwork, delay);
2596	ret = true;
2597	}
2598
2599	local_irq_restore(irq_flags);
2600	return ret;
2601	}
2602	EXPORT_SYMBOL(queue_delayed_work_on);
2603
2604	/**
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
2610	*
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
2614	* current state.
2615	*
2616	* Return: %false if @dwork was idle and queued, %true if @dwork was
2617	* pending and its timer was modified.
2618	*
2619	* This function is safe to call from any context including IRQ handler.
2620	* See try_to_grab_pending() for details.
2621	*/
2622	bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq,
2623	struct delayed_work *dwork, unsigned long delay)
2624	{
2625	unsigned long irq_flags;
2626	int ret;
2627
2628	do {
2629	ret = try_to_grab_pending(&dwork->work, WORK_CANCEL_DELAYED,
2630	&irq_flags);
2631	} while (unlikely(ret == -EAGAIN));
2632
2633	if (likely(ret >= 0)) {
2634	__queue_delayed_work(cpu, wq, dwork, delay);
2635	local_irq_restore(irq_flags);
2636	}
2637
2638	/* -ENOENT from try_to_grab_pending() becomes %true */
2639	return ret;
2640	}
2641	EXPORT_SYMBOL_GPL(mod_delayed_work_on);
2642
2643	static void rcu_work_rcufn(struct rcu_head *rcu)
2644	{
2645	struct rcu_work *rwork = container_of(rcu, struct rcu_work, rcu);
2646
2647	/* read the comment in __queue_work() */
2648	local_irq_disable();
2649	__queue_work(WORK_CPU_UNBOUND, rwork->wq, &rwork->work);
2650	local_irq_enable();
2651	}
2652
2653	/**
2654	* queue_rcu_work - queue work after a RCU grace period
2655	* @wq: workqueue to use
2656	* @rwork: work to queue
2657	*
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.
2662	*/
2663	bool queue_rcu_work(struct workqueue_struct wq, struct rcu_work rwork)
2664	{
2665	struct work_struct *work = &rwork->work;
2666
2667	if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
2668	rwork->wq = wq;
2669	call_rcu_hurry(&rwork->rcu, rcu_work_rcufn);
2670	return true;
2671	}
2672
2673	return false;
2674	}
2675	EXPORT_SYMBOL(queue_rcu_work);
2676
2677	static struct worker *alloc_worker(int node)
2678	{
2679	struct worker *worker;
2680
2681	worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node);
2682	if (worker) {
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;
2688	}
2689	return worker;
2690	}
2691
2692	static cpumask_t pool_allowed_cpus(struct worker_pool pool)
2693	{
2694	if (pool->cpu < 0 && pool->attrs->affn_strict)
2695	return pool->attrs->__pod_cpumask;
2696	else
2697	return pool->attrs->cpumask;
2698	}
2699
2700	/**
2701	* worker_attach_to_pool() - attach a worker to a pool
2702	* @worker: worker to be attached
2703	* @pool: the target pool
2704	*
2705	* Attach @worker to @pool. Once attached, the %WORKER_UNBOUND flag and
2706	* cpu-binding of @worker are kept coordinated with the pool across
2707	* cpu-[un]hotplugs.
2708	*/
2709	static void worker_attach_to_pool(struct worker *worker,
2710	struct worker_pool *pool)
2711	{
2712	mutex_lock(&wq_pool_attach_mutex);
2713
2714	/*
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.
2718	*/
2719	if (pool->flags & POOL_DISASSOCIATED) {
2720	worker->flags \|= WORKER_UNBOUND;
2721	} else {
2722	WARN_ON_ONCE(pool->flags & POOL_BH);
2723	kthread_set_per_cpu(worker->task, pool->cpu);
2724	}
2725
2726	if (worker->rescue_wq)
2727	set_cpus_allowed_ptr(worker->task, pool_allowed_cpus(pool));
2728
2729	list_add_tail(&worker->node, &pool->workers);
2730	worker->pool = pool;
2731
2732	mutex_unlock(&wq_pool_attach_mutex);
2733	}
2734
2735	/**
2736	* worker_detach_from_pool() - detach a worker from its pool
2737	* @worker: worker which is attached to its pool
2738	*
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.
2742	*/
2743	static void worker_detach_from_pool(struct worker *worker)
2744	{
2745	struct worker_pool *pool = worker->pool;
2746	struct completion *detach_completion = NULL;
2747
2748	/* there is one permanent BH worker per CPU which should never detach */
2749	WARN_ON_ONCE(pool->flags & POOL_BH);
2750
2751	mutex_lock(&wq_pool_attach_mutex);
2752
2753	kthread_set_per_cpu(worker->task, -1);
2754	list_del(&worker->node);
2755	worker->pool = NULL;
2756
2757	if (list_empty(&pool->workers) && list_empty(&pool->dying_workers))
2758	detach_completion = pool->detach_completion;
2759	mutex_unlock(&wq_pool_attach_mutex);
2760
2761	/* clear leftover flags without pool->lock after it is detached */
2762	worker->flags &= ~(WORKER_UNBOUND \| WORKER_REBOUND);
2763
2764	if (detach_completion)
2765	complete(detach_completion);
2766	}
2767
2768	/**
2769	* create_worker - create a new workqueue worker
2770	* @pool: pool the new worker will belong to
2771	*
2772	* Create and start a new worker which is attached to @pool.
2773	*
2774	* CONTEXT:
2775	* Might sleep. Does GFP_KERNEL allocations.
2776	*
2777	* Return:
2778	* Pointer to the newly created worker.
2779	*/
2780	static struct worker create_worker(struct worker_pool pool)
2781	{
2782	struct worker *worker;
2783	int id;
2784	char id_buf[23];
2785
2786	/* ID is needed to determine kthread name */
2787	id = ida_alloc(&pool->worker_ida, GFP_KERNEL);
2788	if (id < 0) {
2789	pr_err_once("workqueue: Failed to allocate a worker ID: %pe\n",
2790	ERR_PTR(id));
2791	return NULL;
2792	}
2793
2794	worker = alloc_worker(pool->node);
2795	if (!worker) {
2796	pr_err_once("workqueue: Failed to allocate a worker\n");
2797	goto fail;
2798	}
2799
2800	worker->id = id;
2801
2802	if (!(pool->flags & POOL_BH)) {
2803	if (pool->cpu >= 0)
2804	snprintf(id_buf, sizeof(id_buf), "%d:%d%s", pool->cpu, id,
2805	pool->attrs->nice < 0 ? "H" : "");
2806	else
2807	snprintf(id_buf, sizeof(id_buf), "u%d:%d", pool->id, id);
2808
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",
2814	id_buf);
2815	} else {
2816	pr_err_once("workqueue: Failed to create a worker thread: %pe",
2817	worker->task);
2818	}
2819	goto fail;
2820	}
2821
2822	set_user_nice(worker->task, pool->attrs->nice);
2823	kthread_bind_mask(worker->task, pool_allowed_cpus(pool));
2824	}
2825
2826	/* successful, attach the worker to the pool */
2827	worker_attach_to_pool(worker, pool);
2828
2829	/* start the newly created worker */
2830	raw_spin_lock_irq(&pool->lock);
2831
2832	worker->pool->nr_workers++;
2833	worker_enter_idle(worker);
2834
2835	/*
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.
2839	*/
2840	if (worker->task)
2841	wake_up_process(worker->task);
2842
2843	raw_spin_unlock_irq(&pool->lock);
2844
2845	return worker;
2846
2847	fail:
2848	ida_free(&pool->worker_ida, id);
2849	kfree(worker);
2850	return NULL;
2851	}
2852
2853	static void unbind_worker(struct worker *worker)
2854	{
2855	lockdep_assert_held(&wq_pool_attach_mutex);
2856
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);
2860	else
2861	WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, cpu_possible_mask) < 0);
2862	}
2863
2864	static void wake_dying_workers(struct list_head *cull_list)
2865	{
2866	struct worker worker, tmp;
2867
2868	list_for_each_entry_safe(worker, tmp, cull_list, entry) {
2869	list_del_init(&worker->entry);
2870	unbind_worker(worker);
2871	/*
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.
2875	*
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.
2880	*/
2881	wake_up_process(worker->task);
2882	}
2883	}
2884
2885	/**
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
2889	*
2890	* Tag @worker for destruction and adjust @pool stats accordingly. The worker
2891	* should be idle.
2892	*
2893	* CONTEXT:
2894	* raw_spin_lock_irq(pool->lock).
2895	*/
2896	static void set_worker_dying(struct worker worker, struct list_head list)
2897	{
2898	struct worker_pool *pool = worker->pool;
2899
2900	lockdep_assert_held(&pool->lock);
2901	lockdep_assert_held(&wq_pool_attach_mutex);
2902
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)))
2907	return;
2908
2909	pool->nr_workers--;
2910	pool->nr_idle--;
2911
2912	worker->flags \|= WORKER_DIE;
2913
2914	list_move(&worker->entry, list);
2915	list_move(&worker->node, &pool->dying_workers);
2916	}
2917
2918	/**
2919	* idle_worker_timeout - check if some idle workers can now be deleted.
2920	* @t: The pool's idle_timer that just expired
2921	*
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.
2927	*/
2928	static void idle_worker_timeout(struct timer_list *t)
2929	{
2930	struct worker_pool *pool = from_timer(pool, t, idle_timer);
2931	bool do_cull = false;
2932
2933	if (work_pending(&pool->idle_cull_work))
2934	return;
2935
2936	raw_spin_lock_irq(&pool->lock);
2937
2938	if (too_many_workers(pool)) {
2939	struct worker *worker;
2940	unsigned long expires;
2941
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);
2946
2947	if (!do_cull)
2948	mod_timer(&pool->idle_timer, expires);
2949	}
2950	raw_spin_unlock_irq(&pool->lock);
2951
2952	if (do_cull)
2953	queue_work(system_unbound_wq, &pool->idle_cull_work);
2954	}
2955
2956	/**
2957	* idle_cull_fn - cull workers that have been idle for too long.
2958	* @work: the pool's work for handling these idle workers
2959	*
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.
2962	*
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.
2966	*/
2967	static void idle_cull_fn(struct work_struct *work)
2968	{
2969	struct worker_pool *pool = container_of(work, struct worker_pool, idle_cull_work);
2970	LIST_HEAD(cull_list);
2971
2972	/*
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.
2977	*/
2978	mutex_lock(&wq_pool_attach_mutex);
2979	raw_spin_lock_irq(&pool->lock);
2980
2981	while (too_many_workers(pool)) {
2982	struct worker *worker;
2983	unsigned long expires;
2984
2985	worker = list_entry(pool->idle_list.prev, struct worker, entry);
2986	expires = worker->last_active + IDLE_WORKER_TIMEOUT;
2987
2988	if (time_before(jiffies, expires)) {
2989	mod_timer(&pool->idle_timer, expires);
2990	break;
2991	}
2992
2993	set_worker_dying(worker, &cull_list);
2994	}
2995
2996	raw_spin_unlock_irq(&pool->lock);
2997	wake_dying_workers(&cull_list);
2998	mutex_unlock(&wq_pool_attach_mutex);
2999	}
3000
3001	static void send_mayday(struct work_struct *work)
3002	{
3003	struct pool_workqueue *pwq = get_work_pwq(work);
3004	struct workqueue_struct *wq = pwq->wq;
3005
3006	lockdep_assert_held(&wq_mayday_lock);
3007
3008	if (!wq->rescuer)
3009	return;
3010
3011	/* mayday mayday mayday */
3012	if (list_empty(&pwq->mayday_node)) {
3013	/*
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.
3017	*/
3018	get_pwq(pwq);
3019	list_add_tail(&pwq->mayday_node, &wq->maydays);
3020	wake_up_process(wq->rescuer->task);
3021	pwq->stats[PWQ_STAT_MAYDAY]++;
3022	}
3023	}
3024
3025	static void pool_mayday_timeout(struct timer_list *t)
3026	{
3027	struct worker_pool *pool = from_timer(pool, t, mayday_timer);
3028	struct work_struct *work;
3029
3030	raw_spin_lock_irq(&pool->lock);
3031	raw_spin_lock(&wq_mayday_lock); /* for wq->maydays */
3032
3033	if (need_to_create_worker(pool)) {
3034	/*
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
3038	* rescuers.
3039	*/
3040	list_for_each_entry(work, &pool->worklist, entry)
3041	send_mayday(work);
3042	}
3043
3044	raw_spin_unlock(&wq_mayday_lock);
3045	raw_spin_unlock_irq(&pool->lock);
3046
3047	mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL);
3048	}
3049
3050	/**
3051	* maybe_create_worker - create a new worker if necessary
3052	* @pool: pool to create a new worker for
3053	*
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.
3059	*
3060	* On return, need_to_create_worker() is guaranteed to be %false and
3061	* may_start_working() %true.
3062	*
3063	* LOCKING:
3064	* raw_spin_lock_irq(pool->lock) which may be released and regrabbed
3065	* multiple times. Does GFP_KERNEL allocations. Called only from
3066	* manager.
3067	*/
3068	static void maybe_create_worker(struct worker_pool *pool)
3069	__releases(&pool->lock)
3070	__acquires(&pool->lock)
3071	{
3072	restart:
3073	raw_spin_unlock_irq(&pool->lock);
3074
3075	/* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */
3076	mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT);
3077
3078	while (true) {
3079	if (create_worker(pool) \|\| !need_to_create_worker(pool))
3080	break;
3081
3082	schedule_timeout_interruptible(CREATE_COOLDOWN);
3083
3084	if (!need_to_create_worker(pool))
3085	break;
3086	}
3087
3088	del_timer_sync(&pool->mayday_timer);
3089	raw_spin_lock_irq(&pool->lock);
3090	/*
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.
3094	*/
3095	if (need_to_create_worker(pool))
3096	goto restart;
3097	}
3098
3099	/**
3100	* manage_workers - manage worker pool
3101	* @worker: self
3102	*
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.
3106	*
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.
3110	*
3111	* CONTEXT:
3112	* raw_spin_lock_irq(pool->lock) which may be released and regrabbed
3113	* multiple times. Does GFP_KERNEL allocations.
3114	*
3115	* Return:
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.
3120	*/
3121	static bool manage_workers(struct worker *worker)
3122	{
3123	struct worker_pool *pool = worker->pool;
3124
3125	if (pool->flags & POOL_MANAGER_ACTIVE)
3126	return false;
3127
3128	pool->flags \|= POOL_MANAGER_ACTIVE;
3129	pool->manager = worker;
3130
3131	maybe_create_worker(pool);
3132
3133	pool->manager = NULL;
3134	pool->flags &= ~POOL_MANAGER_ACTIVE;
3135	rcuwait_wake_up(&manager_wait);
3136	return true;
3137	}
3138
3139	/**
3140	* process_one_work - process single work
3141	* @worker: self
3142	* @work: work to process
3143	*
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.
3149	*
3150	* CONTEXT:
3151	* raw_spin_lock_irq(pool->lock) which is released and regrabbed.
3152	*/
3153	static void process_one_work(struct worker worker, struct work_struct work)
3154	__releases(&pool->lock)
3155	__acquires(&pool->lock)
3156	{
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
3163	/*
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.
3169	*/
3170	struct lockdep_map lockdep_map;
3171
3172	lockdep_copy_map(&lockdep_map, &work->lockdep_map);
3173	#endif
3174	/* ensure we're on the correct CPU */
3175	WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) &&
3176	raw_smp_processor_id() != pool->cpu);
3177
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;
3184	if (worker->task)
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);
3188
3189	/*
3190	* Record wq name for cmdline and debug reporting, may get
3191	* overridden through set_worker_desc().
3192	*/
3193	strscpy(worker->desc, pwq->wq->name, WORKER_DESC_LEN);
3194
3195	list_del_init(&work->entry);
3196
3197	/*
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.
3202	*/
3203	if (unlikely(pwq->wq->flags & WQ_CPU_INTENSIVE))
3204	worker_set_flags(worker, WORKER_CPU_INTENSIVE);
3205
3206	/*
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.
3211	*/
3212	kick_pool(pool);
3213
3214	/*
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
3218	* disabled.
3219	*/
3220	set_work_pool_and_clear_pending(work, pool->id, 0);
3221
3222	pwq->stats[PWQ_STAT_STARTED]++;
3223	raw_spin_unlock_irq(&pool->lock);
3224
3225	rcu_start_depth = rcu_preempt_depth();
3226	lockdep_start_depth = lockdep_depth(current);
3227	/* see drain_dead_softirq_workfn() */
3228	if (!bh_draining)
3229	lock_map_acquire(&pwq->wq->lockdep_map);
3230	lock_map_acquire(&lockdep_map);
3231	/*
3232	* Strictly speaking we should mark the invariant state without holding
3233	* any locks, that is, before these two lock_map_acquire()'s.
3234	*
3235	* However, that would result in:
3236	*
3237	* A(W1)
3238	* WFC(C)
3239	* A(W1)
3240	* C(C)
3241	*
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
3246	* these locks.
3247	*
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.
3251	*/
3252	lockdep_invariant_state(true);
3253	trace_workqueue_execute_start(work);
3254	worker->current_func(work);
3255	/*
3256	* While we must be careful to not use "work" after this, the trace
3257	* point will only record its address.
3258	*/
3259	trace_workqueue_execute_end(work, worker->current_func);
3260	pwq->stats[PWQ_STAT_COMPLETED]++;
3261	lock_map_release(&lockdep_map);
3262	if (!bh_draining)
3263	lock_map_release(&pwq->wq->lockdep_map);
3264
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);
3275	dump_stack();
3276	}
3277
3278	/*
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.
3285	*/
3286	if (worker->task)
3287	cond_resched();
3288
3289	raw_spin_lock_irq(&pool->lock);
3290
3291	/*
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.
3295	*/
3296	worker_clr_flags(worker, WORKER_CPU_INTENSIVE);
3297
3298	/* tag the worker for identification in schedule() */
3299	worker->last_func = worker->current_func;
3300
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;
3307
3308	/* must be the last step, see the function comment */
3309	pwq_dec_nr_in_flight(pwq, work_data);
3310	}
3311
3312	/**
3313	* process_scheduled_works - process scheduled works
3314	* @worker: self
3315	*
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.
3319	*
3320	* CONTEXT:
3321	* raw_spin_lock_irq(pool->lock) which may be released and regrabbed
3322	* multiple times.
3323	*/
3324	static void process_scheduled_works(struct worker *worker)
3325	{
3326	struct work_struct *work;
3327	bool first = true;
3328
3329	while ((work = list_first_entry_or_null(&worker->scheduled,
3330	struct work_struct, entry))) {
3331	if (first) {
3332	worker->pool->watchdog_ts = jiffies;
3333	first = false;
3334	}
3335	process_one_work(worker, work);
3336	}
3337	}
3338
3339	static void set_pf_worker(bool val)
3340	{
3341	mutex_lock(&wq_pool_attach_mutex);
3342	if (val)
3343	current->flags \|= PF_WQ_WORKER;
3344	else
3345	current->flags &= ~PF_WQ_WORKER;
3346	mutex_unlock(&wq_pool_attach_mutex);
3347	}
3348
3349	/**
3350	* worker_thread - the worker thread function
3351	* @__worker: self
3352	*
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().
3358	*
3359	* Return: 0
3360	*/
3361	static int worker_thread(void *__worker)
3362	{
3363	struct worker *worker = __worker;
3364	struct worker_pool *pool = worker->pool;
3365
3366	/* tell the scheduler that this is a workqueue worker */
3367	set_pf_worker(true);
3368	woke_up:
3369	raw_spin_lock_irq(&pool->lock);
3370
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);
3375
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));
3380	kfree(worker);
3381	return 0;
3382	}
3383
3384	worker_leave_idle(worker);
3385	recheck:
3386	/* no more worker necessary? */
3387	if (!need_more_worker(pool))
3388	goto sleep;
3389
3390	/* do we need to manage? */
3391	if (unlikely(!may_start_working(pool)) && manage_workers(worker))
3392	goto recheck;
3393
3394	/*
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.
3398	*/
3399	WARN_ON_ONCE(!list_empty(&worker->scheduled));
3400
3401	/*
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.
3407	*/
3408	worker_clr_flags(worker, WORKER_PREP \| WORKER_REBOUND);
3409
3410	do {
3411	struct work_struct *work =
3412	list_first_entry(&pool->worklist,
3413	struct work_struct, entry);
3414
3415	if (assign_work(work, worker, NULL))
3416	process_scheduled_works(worker);
3417	} while (keep_working(pool));
3418
3419	worker_set_flags(worker, WORKER_PREP);
3420	sleep:
3421	/*
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
3426	* event.
3427	*/
3428	worker_enter_idle(worker);
3429	__set_current_state(TASK_IDLE);
3430	raw_spin_unlock_irq(&pool->lock);
3431	schedule();
3432	goto woke_up;
3433	}
3434
3435	/**
3436	* rescuer_thread - the rescuer thread function
3437	* @__rescuer: self
3438	*
3439	* Workqueue rescuer thread function. There's one rescuer for each
3440	* workqueue which has WQ_MEM_RECLAIM set.
3441	*
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.
3447	*
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.
3451	*
3452	* This should happen rarely.
3453	*
3454	* Return: 0
3455	*/
3456	static int rescuer_thread(void *__rescuer)
3457	{
3458	struct worker *rescuer = __rescuer;
3459	struct workqueue_struct *wq = rescuer->rescue_wq;
3460	bool should_stop;
3461
3462	set_user_nice(current, RESCUER_NICE_LEVEL);
3463
3464	/*
3465	* Mark rescuer as worker too. As WORKER_PREP is never cleared, it
3466	* doesn't participate in concurrency management.
3467	*/
3468	set_pf_worker(true);
3469	repeat:
3470	set_current_state(TASK_IDLE);
3471
3472	/*
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.
3479	*/
3480	should_stop = kthread_should_stop();
3481
3482	/* see whether any pwq is asking for help */
3483	raw_spin_lock_irq(&wq_mayday_lock);
3484
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;
3490
3491	__set_current_state(TASK_RUNNING);
3492	list_del_init(&pwq->mayday_node);
3493
3494	raw_spin_unlock_irq(&wq_mayday_lock);
3495
3496	worker_attach_to_pool(rescuer, pool);
3497
3498	raw_spin_lock_irq(&pool->lock);
3499
3500	/*
3501	* Slurp in all works issued via this workqueue and
3502	* process'em.
3503	*/
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]++;
3509	}
3510
3511	if (!list_empty(&rescuer->scheduled)) {
3512	process_scheduled_works(rescuer);
3513
3514	/*
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.
3522	*/
3523	if (pwq->nr_active && need_to_create_worker(pool)) {
3524	raw_spin_lock(&wq_mayday_lock);
3525	/*
3526	* Queue iff we aren't racing destruction
3527	* and somebody else hasn't queued it already.
3528	*/
3529	if (wq->rescuer && list_empty(&pwq->mayday_node)) {
3530	get_pwq(pwq);
3531	list_add_tail(&pwq->mayday_node, &wq->maydays);
3532	}
3533	raw_spin_unlock(&wq_mayday_lock);
3534	}
3535	}
3536
3537	/*
3538	* Put the reference grabbed by send_mayday(). @pool won't
3539	* go away while we're still attached to it.
3540	*/
3541	put_pwq(pwq);
3542
3543	/*
3544	* Leave this pool. Notify regular workers; otherwise, we end up
3545	* with 0 concurrency and stalling the execution.
3546	*/
3547	kick_pool(pool);
3548
3549	raw_spin_unlock_irq(&pool->lock);
3550
3551	worker_detach_from_pool(rescuer);
3552
3553	raw_spin_lock_irq(&wq_mayday_lock);
3554	}
3555
3556	raw_spin_unlock_irq(&wq_mayday_lock);
3557
3558	if (should_stop) {
3559	__set_current_state(TASK_RUNNING);
3560	set_pf_worker(false);
3561	return 0;
3562	}
3563
3564	/* rescuers should never participate in concurrency management */
3565	WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING));
3566	schedule();
3567	goto repeat;
3568	}
3569
3570	static void bh_worker(struct worker *worker)
3571	{
3572	struct worker_pool *pool = worker->pool;
3573	int nr_restarts = BH_WORKER_RESTARTS;
3574	unsigned long end = jiffies + BH_WORKER_JIFFIES;
3575
3576	raw_spin_lock_irq(&pool->lock);
3577	worker_leave_idle(worker);
3578
3579	/*
3580	* This function follows the structure of worker_thread(). See there for
3581	* explanations on each step.
3582	*/
3583	if (!need_more_worker(pool))
3584	goto done;
3585
3586	WARN_ON_ONCE(!list_empty(&worker->scheduled));
3587	worker_clr_flags(worker, WORKER_PREP \| WORKER_REBOUND);
3588
3589	do {
3590	struct work_struct *work =
3591	list_first_entry(&pool->worklist,
3592	struct work_struct, entry);
3593
3594	if (assign_work(work, worker, NULL))
3595	process_scheduled_works(worker);
3596	} while (keep_working(pool) &&
3597	--nr_restarts && time_before(jiffies, end));
3598
3599	worker_set_flags(worker, WORKER_PREP);
3600	done:
3601	worker_enter_idle(worker);
3602	kick_pool(pool);
3603	raw_spin_unlock_irq(&pool->lock);
3604	}
3605
3606	/*
3607	* TODO: Convert all tasklet users to workqueue and use softirq directly.
3608	*
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
3612	* can be dropped.
3613	*
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.
3616	*/
3617	void workqueue_softirq_action(bool highpri)
3618	{
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));
3623	}
3624
3625	struct wq_drain_dead_softirq_work {
3626	struct work_struct work;
3627	struct worker_pool *pool;
3628	struct completion done;
3629	};
3630
3631	static void drain_dead_softirq_workfn(struct work_struct *work)
3632	{
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;
3636	bool repeat;
3637
3638	/*
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.
3644	*/
3645	raw_spin_lock_irq(&pool->lock);
3646	pool->flags \|= POOL_BH_DRAINING;
3647	raw_spin_unlock_irq(&pool->lock);
3648
3649	bh_worker(list_first_entry(&pool->workers, struct worker, node));
3650
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);
3655
3656	/*
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.
3660	*/
3661	if (repeat) {
3662	if (pool->attrs->nice == HIGHPRI_NICE_LEVEL)
3663	queue_work(system_bh_highpri_wq, work);
3664	else
3665	queue_work(system_bh_wq, work);
3666	} else {
3667	complete(&dead_work->done);
3668	}
3669	}
3670
3671	/*
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.
3678	*/
3679	void workqueue_softirq_dead(unsigned int cpu)
3680	{
3681	int i;
3682
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;
3686
3687	if (!need_more_worker(pool))
3688	continue;
3689
3690	INIT_WORK(&dead_work.work, drain_dead_softirq_workfn);
3691	dead_work.pool = pool;
3692	init_completion(&dead_work.done);
3693
3694	if (pool->attrs->nice == HIGHPRI_NICE_LEVEL)
3695	queue_work(system_bh_highpri_wq, &dead_work.work);
3696	else
3697	queue_work(system_bh_wq, &dead_work.work);
3698
3699	wait_for_completion(&dead_work.done);
3700	}
3701	}
3702
3703	/**
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)
3707	*
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
3712	* a deadlock.
3713	*/
3714	static void check_flush_dependency(struct workqueue_struct *target_wq,
3715	struct work_struct *target_work)
3716	{
3717	work_func_t target_func = target_work ? target_work->func : NULL;
3718	struct worker *worker;
3719
3720	if (target_wq->flags & WQ_MEM_RECLAIM)
3721	return;
3722
3723	worker = current_wq_worker();
3724
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);
3733	}
3734
3735	struct wq_barrier {
3736	struct work_struct work;
3737	struct completion done;
3738	struct task_struct task; / purely informational */
3739	};
3740
3741	static void wq_barrier_func(struct work_struct *work)
3742	{
3743	struct wq_barrier *barr = container_of(work, struct wq_barrier, work);
3744	complete(&barr->done);
3745	}
3746
3747	/**
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
3753	*
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
3757	* cpu.
3758	*
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.
3764	*
3765	* Note that when @worker is non-NULL, @target may be modified
3766	* underneath us, so we can't reliably determine pwq from @target.
3767	*
3768	* CONTEXT:
3769	* raw_spin_lock_irq(pool->lock).
3770	*/
3771	static void insert_wq_barrier(struct pool_workqueue *pwq,
3772	struct wq_barrier *barr,
3773	struct work_struct target, struct worker worker)
3774	{
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;
3779
3780	/*
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
3784	* might deadlock.
3785	*
3786	* BH and threaded workqueues need separate lockdep keys to avoid
3787	* spuriously triggering "inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W}
3788	* usage".
3789	*/
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));
3793
3794	init_completion_map(&barr->done, &target->lockdep_map);
3795
3796	barr->task = current;
3797
3798	/* The barrier work item does not participate in nr_active. */
3799	work_flags \|= WORK_STRUCT_INACTIVE;
3800
3801	/*
3802	* If @target is currently being executed, schedule the
3803	* barrier to the worker; otherwise, put it after @target.
3804	*/
3805	if (worker) {
3806	head = worker->scheduled.next;
3807	work_color = worker->current_color;
3808	} else {
3809	unsigned long *bits = work_data_bits(target);
3810
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);
3816	}
3817
3818	pwq->nr_in_flight[work_color]++;
3819	work_flags \|= work_color_to_flags(work_color);
3820
3821	insert_work(pwq, &barr->work, head, work_flags);
3822	}
3823
3824	/**
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
3829	*
3830	* Prepare pwqs for workqueue flushing.
3831	*
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.
3838	*
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
3842	* is returned.
3843	*
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.
3847	*
3848	* CONTEXT:
3849	* mutex_lock(wq->mutex).
3850	*
3851	* Return:
3852	* %true if @flush_color >= 0 and there's something to flush. %false
3853	* otherwise.
3854	*/
3855	static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq,
3856	int flush_color, int work_color)
3857	{
3858	bool wait = false;
3859	struct pool_workqueue *pwq;
3860
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);
3864	}
3865
3866	for_each_pwq(pwq, wq) {
3867	struct worker_pool *pool = pwq->pool;
3868
3869	raw_spin_lock_irq(&pool->lock);
3870
3871	if (flush_color >= 0) {
3872	WARN_ON_ONCE(pwq->flush_color != -1);
3873
3874	if (pwq->nr_in_flight[flush_color]) {
3875	pwq->flush_color = flush_color;
3876	atomic_inc(&wq->nr_pwqs_to_flush);
3877	wait = true;
3878	}
3879	}
3880
3881	if (work_color >= 0) {
3882	WARN_ON_ONCE(work_color != work_next_color(pwq->work_color));
3883	pwq->work_color = work_color;
3884	}
3885
3886	raw_spin_unlock_irq(&pool->lock);
3887	}
3888
3889	if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_pwqs_to_flush))
3890	complete(&wq->first_flusher->done);
3891
3892	return wait;
3893	}
3894
3895	static void touch_wq_lockdep_map(struct workqueue_struct *wq)
3896	{
3897	#ifdef CONFIG_LOCKDEP
3898	if (wq->flags & WQ_BH)
3899	local_bh_disable();
3900
3901	lock_map_acquire(&wq->lockdep_map);
3902	lock_map_release(&wq->lockdep_map);
3903
3904	if (wq->flags & WQ_BH)
3905	local_bh_enable();
3906	#endif
3907	}
3908
3909	static void touch_work_lockdep_map(struct work_struct *work,
3910	struct workqueue_struct *wq)
3911	{
3912	#ifdef CONFIG_LOCKDEP
3913	if (wq->flags & WQ_BH)
3914	local_bh_disable();
3915
3916	lock_map_acquire(&work->lockdep_map);
3917	lock_map_release(&work->lockdep_map);
3918
3919	if (wq->flags & WQ_BH)
3920	local_bh_enable();
3921	#endif
3922	}
3923
3924	/**
3925	* __flush_workqueue - ensure that any scheduled work has run to completion.
3926	* @wq: workqueue to flush
3927	*
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.
3930	*/
3931	void __flush_workqueue(struct workqueue_struct *wq)
3932	{
3933	struct wq_flusher this_flusher = {
3934	.list = LIST_HEAD_INIT(this_flusher.list),
3935	.flush_color = -1,
3936	.done = COMPLETION_INITIALIZER_ONSTACK_MAP(this_flusher.done, wq->lockdep_map),
3937	};
3938	int next_color;
3939
3940	if (WARN_ON(!wq_online))
3941	return;
3942
3943	touch_wq_lockdep_map(wq);
3944
3945	mutex_lock(&wq->mutex);
3946
3947	/*
3948	* Start-to-wait phase
3949	*/
3950	next_color = work_next_color(wq->work_color);
3951
3952	if (next_color != wq->flush_color) {
3953	/*
3954	* Color space is not full. The current work_color
3955	* becomes our flush_color and work_color is advanced
3956	* by one.
3957	*/
3958	WARN_ON_ONCE(!list_empty(&wq->flusher_overflow));
3959	this_flusher.flush_color = wq->work_color;
3960	wq->work_color = next_color;
3961
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);
3965
3966	wq->first_flusher = &this_flusher;
3967
3968	if (!flush_workqueue_prep_pwqs(wq, wq->flush_color,
3969	wq->work_color)) {
3970	/* nothing to flush, done */
3971	wq->flush_color = next_color;
3972	wq->first_flusher = NULL;
3973	goto out_unlock;
3974	}
3975	} else {
3976	/* wait in queue */
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);
3980	}
3981	} else {
3982	/*
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.
3986	*/
3987	list_add_tail(&this_flusher.list, &wq->flusher_overflow);
3988	}
3989
3990	check_flush_dependency(wq, NULL);
3991
3992	mutex_unlock(&wq->mutex);
3993
3994	wait_for_completion(&this_flusher.done);
3995
3996	/*
3997	* Wake-up-and-cascade phase
3998	*
3999	* First flushers are responsible for cascading flushes and
4000	* handling overflow. Non-first flushers can simply return.
4001	*/
4002	if (READ_ONCE(wq->first_flusher) != &this_flusher)
4003	return;
4004
4005	mutex_lock(&wq->mutex);
4006
4007	/* we might have raced, check again with mutex held */
4008	if (wq->first_flusher != &this_flusher)
4009	goto out_unlock;
4010
4011	WRITE_ONCE(wq->first_flusher, NULL);
4012
4013	WARN_ON_ONCE(!list_empty(&this_flusher.list));
4014	WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color);
4015
4016	while (true) {
4017	struct wq_flusher next, tmp;
4018
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)
4022	break;
4023	list_del_init(&next->list);
4024	complete(&next->done);
4025	}
4026
4027	WARN_ON_ONCE(!list_empty(&wq->flusher_overflow) &&
4028	wq->flush_color != work_next_color(wq->work_color));
4029
4030	/* this flush_color is finished, advance by one */
4031	wq->flush_color = work_next_color(wq->flush_color);
4032
4033	/* one color has been freed, handle overflow queue */
4034	if (!list_empty(&wq->flusher_overflow)) {
4035	/*
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.
4040	*/
4041	list_for_each_entry(tmp, &wq->flusher_overflow, list)
4042	tmp->flush_color = wq->work_color;
4043
4044	wq->work_color = work_next_color(wq->work_color);
4045
4046	list_splice_tail_init(&wq->flusher_overflow,
4047	&wq->flusher_queue);
4048	flush_workqueue_prep_pwqs(wq, -1, wq->work_color);
4049	}
4050
4051	if (list_empty(&wq->flusher_queue)) {
4052	WARN_ON_ONCE(wq->flush_color != wq->work_color);
4053	break;
4054	}
4055
4056	/*
4057	* Need to flush more colors. Make the next flusher
4058	* the new first flusher and arm pwqs.
4059	*/
4060	WARN_ON_ONCE(wq->flush_color == wq->work_color);
4061	WARN_ON_ONCE(wq->flush_color != next->flush_color);
4062
4063	list_del_init(&next->list);
4064	wq->first_flusher = next;
4065
4066	if (flush_workqueue_prep_pwqs(wq, wq->flush_color, -1))
4067	break;
4068
4069	/*
4070	* Meh... this color is already done, clear first
4071	* flusher and repeat cascading.
4072	*/
4073	wq->first_flusher = NULL;
4074	}
4075
4076	out_unlock:
4077	mutex_unlock(&wq->mutex);
4078	}
4079	EXPORT_SYMBOL(__flush_workqueue);
4080
4081	/**
4082	* drain_workqueue - drain a workqueue
4083	* @wq: workqueue to drain
4084	*
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
4090	* takes too long.
4091	*/
4092	void drain_workqueue(struct workqueue_struct *wq)
4093	{
4094	unsigned int flush_cnt = 0;
4095	struct pool_workqueue *pwq;
4096
4097	/*
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.
4101	*/
4102	mutex_lock(&wq->mutex);
4103	if (!wq->nr_drainers++)
4104	wq->flags \|= __WQ_DRAINING;
4105	mutex_unlock(&wq->mutex);
4106	reflush:
4107	__flush_workqueue(wq);
4108
4109	mutex_lock(&wq->mutex);
4110
4111	for_each_pwq(pwq, wq) {
4112	bool drained;
4113
4114	raw_spin_lock_irq(&pwq->pool->lock);
4115	drained = pwq_is_empty(pwq);
4116	raw_spin_unlock_irq(&pwq->pool->lock);
4117
4118	if (drained)
4119	continue;
4120
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);
4125
4126	mutex_unlock(&wq->mutex);
4127	goto reflush;
4128	}
4129
4130	if (!--wq->nr_drainers)
4131	wq->flags &= ~__WQ_DRAINING;
4132	mutex_unlock(&wq->mutex);
4133	}
4134	EXPORT_SYMBOL_GPL(drain_workqueue);
4135
4136	static bool start_flush_work(struct work_struct work, struct wq_barrier barr,
4137	bool from_cancel)
4138	{
4139	struct worker *worker = NULL;
4140	struct worker_pool *pool;
4141	struct pool_workqueue *pwq;
4142	struct workqueue_struct *wq;
4143
4144	might_sleep();
4145
4146	rcu_read_lock();
4147	pool = get_work_pool(work);
4148	if (!pool) {
4149	rcu_read_unlock();
4150	return false;
4151	}
4152
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);
4156	if (pwq) {
4157	if (unlikely(pwq->pool != pool))
4158	goto already_gone;
4159	} else {
4160	worker = find_worker_executing_work(pool, work);
4161	if (!worker)
4162	goto already_gone;
4163	pwq = worker->current_pwq;
4164	}
4165
4166	wq = pwq->wq;
4167	check_flush_dependency(wq, work);
4168
4169	insert_wq_barrier(pwq, barr, work, worker);
4170	raw_spin_unlock_irq(&pool->lock);
4171
4172	touch_work_lockdep_map(work, wq);
4173
4174	/*
4175	* Force a lock recursion deadlock when using flush_work() inside a
4176	* single-threaded or rescuer equipped workqueue.
4177	*
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
4181	* forward progress.
4182	*/
4183	if (!from_cancel && (wq->saved_max_active == 1 \|\| wq->rescuer))
4184	touch_wq_lockdep_map(wq);
4185
4186	rcu_read_unlock();
4187	return true;
4188	already_gone:
4189	raw_spin_unlock_irq(&pool->lock);
4190	rcu_read_unlock();
4191	return false;
4192	}
4193
4194	static bool __flush_work(struct work_struct *work, bool from_cancel)
4195	{
4196	struct wq_barrier barr;
4197
4198	if (WARN_ON(!wq_online))
4199	return false;
4200
4201	if (WARN_ON(!work->func))
4202	return false;
4203
4204	if (start_flush_work(work, &barr, from_cancel)) {
4205	wait_for_completion(&barr.done);
4206	destroy_work_on_stack(&barr.work);
4207	return true;
4208	} else {
4209	return false;
4210	}
4211	}
4212
4213	/**
4214	* flush_work - wait for a work to finish executing the last queueing instance
4215	* @work: the work to flush
4216	*
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.
4219	*
4220	* Return:
4221	* %true if flush_work() waited for the work to finish execution,
4222	* %false if it was already idle.
4223	*/
4224	bool flush_work(struct work_struct *work)
4225	{
4226	return __flush_work(work, false);
4227	}
4228	EXPORT_SYMBOL_GPL(flush_work);
4229
4230	/**
4231	* flush_delayed_work - wait for a dwork to finish executing the last queueing
4232	* @dwork: the delayed work to flush
4233	*
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.
4237	*
4238	* Return:
4239	* %true if flush_work() waited for the work to finish execution,
4240	* %false if it was already idle.
4241	*/
4242	bool flush_delayed_work(struct delayed_work *dwork)
4243	{
4244	local_irq_disable();
4245	if (del_timer_sync(&dwork->timer))
4246	__queue_work(dwork->cpu, dwork->wq, &dwork->work);
4247	local_irq_enable();
4248	return flush_work(&dwork->work);
4249	}
4250	EXPORT_SYMBOL(flush_delayed_work);
4251
4252	/**
4253	* flush_rcu_work - wait for a rwork to finish executing the last queueing
4254	* @rwork: the rcu work to flush
4255	*
4256	* Return:
4257	* %true if flush_rcu_work() waited for the work to finish execution,
4258	* %false if it was already idle.
4259	*/
4260	bool flush_rcu_work(struct rcu_work *rwork)
4261	{
4262	if (test_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&rwork->work))) {
4263	rcu_barrier();
4264	flush_work(&rwork->work);
4265	return true;
4266	} else {
4267	return flush_work(&rwork->work);
4268	}
4269	}
4270	EXPORT_SYMBOL(flush_rcu_work);
4271
4272	static bool __cancel_work(struct work_struct *work, u32 cflags)
4273	{
4274	unsigned long irq_flags;
4275	int ret;
4276
4277	do {
4278	ret = try_to_grab_pending(work, cflags, &irq_flags);
4279	} while (unlikely(ret == -EAGAIN));
4280
4281	if (unlikely(ret < 0))
4282	return false;
4283
4284	set_work_pool_and_clear_pending(work, get_work_pool_id(work), 0);
4285	local_irq_restore(irq_flags);
4286	return ret;
4287	}
4288
4289	static bool __cancel_work_sync(struct work_struct *work, u32 cflags)
4290	{
4291	unsigned long irq_flags;
4292	bool ret;
4293
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);
4298
4299	/*
4300	* Skip __flush_work() during early boot when we know that @work isn't
4301	* executing. This allows canceling during early boot.
4302	*/
4303	if (wq_online)
4304	__flush_work(work, true);
4305
4306	/*
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.
4310	*/
4311	set_work_pool_and_clear_pending(work, WORK_OFFQ_POOL_NONE, 0);
4312
4313	if (waitqueue_active(&wq_cancel_waitq))
4314	__wake_up(&wq_cancel_waitq, TASK_NORMAL, 1, work);
4315
4316	return ret;
4317	}
4318
4319	/*
4320	* See cancel_delayed_work()
4321	*/
4322	bool cancel_work(struct work_struct *work)
4323	{
4324	return __cancel_work(work, 0);
4325	}
4326	EXPORT_SYMBOL(cancel_work);
4327
4328	/**
4329	* cancel_work_sync - cancel a work and wait for it to finish
4330	* @work: the work to cancel
4331	*
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.
4336	*
4337	* cancel_work_sync(&delayed_work->work) must not be used for
4338	* delayed_work's. Use cancel_delayed_work_sync() instead.
4339	*
4340	* The caller must ensure that the workqueue on which @work was last
4341	* queued can't be destroyed before this function returns.
4342	*
4343	* Return:
4344	* %true if @work was pending, %false otherwise.
4345	*/
4346	bool cancel_work_sync(struct work_struct *work)
4347	{
4348	return __cancel_work_sync(work, 0);
4349	}
4350	EXPORT_SYMBOL_GPL(cancel_work_sync);
4351
4352	/**
4353	* cancel_delayed_work - cancel a delayed work
4354	* @dwork: delayed_work to cancel
4355	*
4356	* Kill off a pending delayed_work.
4357	*
4358	* Return: %true if @dwork was pending and canceled; %false if it wasn't
4359	* pending.
4360	*
4361	* Note:
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.
4365	*
4366	* This function is safe to call from any context including IRQ handler.
4367	*/
4368	bool cancel_delayed_work(struct delayed_work *dwork)
4369	{
4370	return __cancel_work(&dwork->work, WORK_CANCEL_DELAYED);
4371	}
4372	EXPORT_SYMBOL(cancel_delayed_work);
4373
4374	/**
4375	* cancel_delayed_work_sync - cancel a delayed work and wait for it to finish
4376	* @dwork: the delayed work cancel
4377	*
4378	* This is cancel_work_sync() for delayed works.
4379	*
4380	* Return:
4381	* %true if @dwork was pending, %false otherwise.
4382	*/
4383	bool cancel_delayed_work_sync(struct delayed_work *dwork)
4384	{
4385	return __cancel_work_sync(&dwork->work, WORK_CANCEL_DELAYED);
4386	}
4387	EXPORT_SYMBOL(cancel_delayed_work_sync);
4388
4389	/**
4390	* schedule_on_each_cpu - execute a function synchronously on each online CPU
4391	* @func: the function to call
4392	*
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.
4396	*
4397	* Return:
4398	* 0 on success, -errno on failure.
4399	*/
4400	int schedule_on_each_cpu(work_func_t func)
4401	{
4402	int cpu;
4403	struct work_struct __percpu *works;
4404
4405	works = alloc_percpu(struct work_struct);
4406	if (!works)
4407	return -ENOMEM;
4408
4409	cpus_read_lock();
4410
4411	for_each_online_cpu(cpu) {
4412	struct work_struct *work = per_cpu_ptr(works, cpu);
4413
4414	INIT_WORK(work, func);
4415	schedule_work_on(cpu, work);
4416	}
4417
4418	for_each_online_cpu(cpu)
4419	flush_work(per_cpu_ptr(works, cpu));
4420
4421	cpus_read_unlock();
4422	free_percpu(works);
4423	return 0;
4424	}
4425
4426	/**
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)
4431	*
4432	* Executes the function immediately if process context is available,
4433	* otherwise schedules the function for delayed execution.
4434	*
4435	* Return: 0 - function was executed
4436	* 1 - function was scheduled for execution
4437	*/
4438	int execute_in_process_context(work_func_t fn, struct execute_work *ew)
4439	{
4440	if (!in_interrupt()) {
4441	fn(&ew->work);
4442	return 0;
4443	}
4444
4445	INIT_WORK(&ew->work, fn);
4446	schedule_work(&ew->work);
4447
4448	return 1;
4449	}
4450	EXPORT_SYMBOL_GPL(execute_in_process_context);
4451
4452	/**
4453	* free_workqueue_attrs - free a workqueue_attrs
4454	* @attrs: workqueue_attrs to free
4455	*
4456	* Undo alloc_workqueue_attrs().
4457	*/
4458	void free_workqueue_attrs(struct workqueue_attrs *attrs)
4459	{
4460	if (attrs) {
4461	free_cpumask_var(attrs->cpumask);
4462	free_cpumask_var(attrs->__pod_cpumask);
4463	kfree(attrs);
4464	}
4465	}
4466
4467	/**
4468	* alloc_workqueue_attrs - allocate a workqueue_attrs
4469	*
4470	* Allocate a new workqueue_attrs, initialize with default settings and
4471	* return it.
4472	*
4473	* Return: The allocated new workqueue_attr on success. %NULL on failure.
4474	*/
4475	struct workqueue_attrs *alloc_workqueue_attrs(void)
4476	{
4477	struct workqueue_attrs *attrs;
4478
4479	attrs = kzalloc(sizeof(*attrs), GFP_KERNEL);
4480	if (!attrs)
4481	goto fail;
4482	if (!alloc_cpumask_var(&attrs->cpumask, GFP_KERNEL))
4483	goto fail;
4484	if (!alloc_cpumask_var(&attrs->__pod_cpumask, GFP_KERNEL))
4485	goto fail;
4486
4487	cpumask_copy(attrs->cpumask, cpu_possible_mask);
4488	attrs->affn_scope = WQ_AFFN_DFL;
4489	return attrs;
4490	fail:
4491	free_workqueue_attrs(attrs);
4492	return NULL;
4493	}
4494
4495	static void copy_workqueue_attrs(struct workqueue_attrs *to,
4496	const struct workqueue_attrs *from)
4497	{
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;
4502
4503	/*
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.
4507	*/
4508	to->affn_scope = from->affn_scope;
4509	to->ordered = from->ordered;
4510	}
4511
4512	/*
4513	* Some attrs fields are workqueue-only. Clear them for worker_pool's. See the
4514	* comments in 'struct workqueue_attrs' definition.
4515	*/
4516	static void wqattrs_clear_for_pool(struct workqueue_attrs *attrs)
4517	{
4518	attrs->affn_scope = WQ_AFFN_NR_TYPES;
4519	attrs->ordered = false;
4520	}
4521
4522	/* hash value of the content of @attr */
4523	static u32 wqattrs_hash(const struct workqueue_attrs *attrs)
4524	{
4525	u32 hash = 0;
4526
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);
4533	return hash;
4534	}
4535
4536	/* content equality test */
4537	static bool wqattrs_equal(const struct workqueue_attrs *a,
4538	const struct workqueue_attrs *b)
4539	{
4540	if (a->nice != b->nice)
4541	return false;
4542	if (!cpumask_equal(a->cpumask, b->cpumask))
4543	return false;
4544	if (!cpumask_equal(a->__pod_cpumask, b->__pod_cpumask))
4545	return false;
4546	if (a->affn_strict != b->affn_strict)
4547	return false;
4548	return true;
4549	}
4550
4551	/* Update @attrs with actually available CPUs */
4552	static void wqattrs_actualize_cpumask(struct workqueue_attrs *attrs,
4553	const cpumask_t *unbound_cpumask)
4554	{
4555	/*
4556	* Calculate the effective CPU mask of @attrs given @unbound_cpumask. If
4557	* @attrs->cpumask doesn't overlap with @unbound_cpumask, we fallback to
4558	* @unbound_cpumask.
4559	*/
4560	cpumask_and(attrs->cpumask, attrs->cpumask, unbound_cpumask);
4561	if (unlikely(cpumask_empty(attrs->cpumask)))
4562	cpumask_copy(attrs->cpumask, unbound_cpumask);
4563	}
4564
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)
4568	{
4569	enum wq_affn_scope scope;
4570	struct wq_pod_type *pt;
4571
4572	/* to synchronize access to wq_affn_dfl */
4573	lockdep_assert_held(&wq_pool_mutex);
4574
4575	if (attrs->affn_scope == WQ_AFFN_DFL)
4576	scope = wq_affn_dfl;
4577	else
4578	scope = attrs->affn_scope;
4579
4580	pt = &wq_pod_types[scope];
4581
4582	if (!WARN_ON_ONCE(attrs->affn_scope == WQ_AFFN_NR_TYPES) &&
4583	likely(pt->nr_pods))
4584	return pt;
4585
4586	/*
4587	* Before workqueue_init_topology(), only SYSTEM is available which is
4588	* initialized in workqueue_init_early().
4589	*/
4590	pt = &wq_pod_types[WQ_AFFN_SYSTEM];
4591	BUG_ON(!pt->nr_pods);
4592	return pt;
4593	}
4594
4595	/**
4596	* init_worker_pool - initialize a newly zalloc'd worker_pool
4597	* @pool: worker_pool to initialize
4598	*
4599	* Initialize a newly zalloc'd @pool. It also allocates @pool->attrs.
4600	*
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.
4604	*/
4605	static int init_worker_pool(struct worker_pool *pool)
4606	{
4607	raw_spin_lock_init(&pool->lock);
4608	pool->id = -1;
4609	pool->cpu = -1;
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);
4616
4617	timer_setup(&pool->idle_timer, idle_worker_timeout, TIMER_DEFERRABLE);
4618	INIT_WORK(&pool->idle_cull_work, idle_cull_fn);
4619
4620	timer_setup(&pool->mayday_timer, pool_mayday_timeout, 0);
4621
4622	INIT_LIST_HEAD(&pool->workers);
4623	INIT_LIST_HEAD(&pool->dying_workers);
4624
4625	ida_init(&pool->worker_ida);
4626	INIT_HLIST_NODE(&pool->hash_node);
4627	pool->refcnt = 1;
4628
4629	/* shouldn't fail above this point */
4630	pool->attrs = alloc_workqueue_attrs();
4631	if (!pool->attrs)
4632	return -ENOMEM;
4633
4634	wqattrs_clear_for_pool(pool->attrs);
4635
4636	return 0;
4637	}
4638
4639	#ifdef CONFIG_LOCKDEP
4640	static void wq_init_lockdep(struct workqueue_struct *wq)
4641	{
4642	char *lock_name;
4643
4644	lockdep_register_key(&wq->key);
4645	lock_name = kasprintf(GFP_KERNEL, "%s%s", "(wq_completion)", wq->name);
4646	if (!lock_name)
4647	lock_name = wq->name;
4648
4649	wq->lock_name = lock_name;
4650	lockdep_init_map(&wq->lockdep_map, lock_name, &wq->key, 0);
4651	}
4652
4653	static void wq_unregister_lockdep(struct workqueue_struct *wq)
4654	{
4655	lockdep_unregister_key(&wq->key);
4656	}
4657
4658	static void wq_free_lockdep(struct workqueue_struct *wq)
4659	{
4660	if (wq->lock_name != wq->name)
4661	kfree(wq->lock_name);
4662	}
4663	#else
4664	static void wq_init_lockdep(struct workqueue_struct *wq)
4665	{
4666	}
4667
4668	static void wq_unregister_lockdep(struct workqueue_struct *wq)
4669	{
4670	}
4671
4672	static void wq_free_lockdep(struct workqueue_struct *wq)
4673	{
4674	}
4675	#endif
4676
4677	static void free_node_nr_active(struct wq_node_nr_active **nna_ar)
4678	{
4679	int node;
4680
4681	for_each_node(node) {
4682	kfree(nna_ar[node]);
4683	nna_ar[node] = NULL;
4684	}
4685
4686	kfree(nna_ar[nr_node_ids]);
4687	nna_ar[nr_node_ids] = NULL;
4688	}
4689
4690	static void init_node_nr_active(struct wq_node_nr_active *nna)
4691	{
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);
4696	}
4697
4698	/*
4699	* Each node's nr_active counter will be accessed mostly from its own node and
4700	* should be allocated in the node.
4701	*/
4702	static int alloc_node_nr_active(struct wq_node_nr_active **nna_ar)
4703	{
4704	struct wq_node_nr_active *nna;
4705	int node;
4706
4707	for_each_node(node) {
4708	nna = kzalloc_node(sizeof(*nna), GFP_KERNEL, node);
4709	if (!nna)
4710	goto err_free;
4711	init_node_nr_active(nna);
4712	nna_ar[node] = nna;
4713	}
4714
4715	/* [nr_node_ids] is used as the fallback */
4716	nna = kzalloc_node(sizeof(*nna), GFP_KERNEL, NUMA_NO_NODE);
4717	if (!nna)
4718	goto err_free;
4719	init_node_nr_active(nna);
4720	nna_ar[nr_node_ids] = nna;
4721
4722	return 0;
4723
4724	err_free:
4725	free_node_nr_active(nna_ar);
4726	return -ENOMEM;
4727	}
4728
4729	static void rcu_free_wq(struct rcu_head *rcu)
4730	{
4731	struct workqueue_struct *wq =
4732	container_of(rcu, struct workqueue_struct, rcu);
4733
4734	if (wq->flags & WQ_UNBOUND)
4735	free_node_nr_active(wq->node_nr_active);
4736
4737	wq_free_lockdep(wq);
4738	free_percpu(wq->cpu_pwq);
4739	free_workqueue_attrs(wq->unbound_attrs);
4740	kfree(wq);
4741	}
4742
4743	static void rcu_free_pool(struct rcu_head *rcu)
4744	{
4745	struct worker_pool *pool = container_of(rcu, struct worker_pool, rcu);
4746
4747	ida_destroy(&pool->worker_ida);
4748	free_workqueue_attrs(pool->attrs);
4749	kfree(pool);
4750	}
4751
4752	/**
4753	* put_unbound_pool - put a worker_pool
4754	* @pool: worker_pool to put
4755	*
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().
4760	*
4761	* Should be called with wq_pool_mutex held.
4762	*/
4763	static void put_unbound_pool(struct worker_pool *pool)
4764	{
4765	DECLARE_COMPLETION_ONSTACK(detach_completion);
4766	struct worker *worker;
4767	LIST_HEAD(cull_list);
4768
4769	lockdep_assert_held(&wq_pool_mutex);
4770
4771	if (--pool->refcnt)
4772	return;
4773
4774	/* sanity checks */
4775	if (WARN_ON(!(pool->cpu < 0)) \|\|
4776	WARN_ON(!list_empty(&pool->worklist)))
4777	return;
4778
4779	/* release id and unhash */
4780	if (pool->id >= 0)
4781	idr_remove(&worker_pool_idr, pool->id);
4782	hash_del(&pool->hash_node);
4783
4784	/*
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.
4788	*
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()
4795	* drops pool->lock
4796	*/
4797	while (true) {
4798	rcuwait_wait_event(&manager_wait,
4799	!(pool->flags & POOL_MANAGER_ACTIVE),
4800	TASK_UNINTERRUPTIBLE);
4801
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;
4806	break;
4807	}
4808	raw_spin_unlock_irq(&pool->lock);
4809	mutex_unlock(&wq_pool_attach_mutex);
4810	}
4811
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);
4816
4817	wake_dying_workers(&cull_list);
4818
4819	if (!list_empty(&pool->workers) \|\| !list_empty(&pool->dying_workers))
4820	pool->detach_completion = &detach_completion;
4821	mutex_unlock(&wq_pool_attach_mutex);
4822
4823	if (pool->detach_completion)
4824	wait_for_completion(pool->detach_completion);
4825
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);
4830
4831	/* RCU protected to allow dereferences from get_work_pool() */
4832	call_rcu(&pool->rcu, rcu_free_pool);
4833	}
4834
4835	/**
4836	* get_unbound_pool - get a worker_pool with the specified attributes
4837	* @attrs: the attributes of the worker_pool to get
4838	*
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
4842	* create a new one.
4843	*
4844	* Should be called with wq_pool_mutex held.
4845	*
4846	* Return: On success, a worker_pool with the same attributes as @attrs.
4847	* On failure, %NULL.
4848	*/
4849	static struct worker_pool get_unbound_pool(const struct workqueue_attrs attrs)
4850	{
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;
4855
4856	lockdep_assert_held(&wq_pool_mutex);
4857
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)) {
4861	pool->refcnt++;
4862	return pool;
4863	}
4864	}
4865
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];
4870	break;
4871	}
4872	}
4873
4874	/* nope, create a new one */
4875	pool = kzalloc_node(sizeof(*pool), GFP_KERNEL, node);
4876	if (!pool \|\| init_worker_pool(pool) < 0)
4877	goto fail;
4878
4879	pool->node = node;
4880	copy_workqueue_attrs(pool->attrs, attrs);
4881	wqattrs_clear_for_pool(pool->attrs);
4882
4883	if (worker_pool_assign_id(pool) < 0)
4884	goto fail;
4885
4886	/* create and start the initial worker */
4887	if (wq_online && !create_worker(pool))
4888	goto fail;
4889
4890	/* install */
4891	hash_add(unbound_pool_hash, &pool->hash_node, hash);
4892
4893	return pool;
4894	fail:
4895	if (pool)
4896	put_unbound_pool(pool);
4897	return NULL;
4898	}
4899
4900	static void rcu_free_pwq(struct rcu_head *rcu)
4901	{
4902	kmem_cache_free(pwq_cache,
4903	container_of(rcu, struct pool_workqueue, rcu));
4904	}
4905
4906	/*
4907	* Scheduled on pwq_release_worker by put_pwq() when an unbound pwq hits zero
4908	* refcnt and needs to be destroyed.
4909	*/
4910	static void pwq_release_workfn(struct kthread_work *work)
4911	{
4912	struct pool_workqueue *pwq = container_of(work, struct pool_workqueue,
4913	release_work);
4914	struct workqueue_struct *wq = pwq->wq;
4915	struct worker_pool *pool = pwq->pool;
4916	bool is_last = false;
4917
4918	/*
4919	* When @pwq is not linked, it doesn't hold any reference to the
4920	* @wq, and @wq is invalid to access.
4921	*/
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);
4926
4927	/*
4928	* For ordered workqueue with a plugged dfl_pwq, restart it now.
4929	*/
4930	if (!is_last && (wq->flags & __WQ_ORDERED))
4931	unplug_oldest_pwq(wq);
4932
4933	mutex_unlock(&wq->mutex);
4934	}
4935
4936	if (wq->flags & WQ_UNBOUND) {
4937	mutex_lock(&wq_pool_mutex);
4938	put_unbound_pool(pool);
4939	mutex_unlock(&wq_pool_mutex);
4940	}
4941
4942	if (!list_empty(&pwq->pending_node)) {
4943	struct wq_node_nr_active *nna =
4944	wq_node_nr_active(pwq->wq, pwq->pool->node);
4945
4946	raw_spin_lock_irq(&nna->lock);
4947	list_del_init(&pwq->pending_node);
4948	raw_spin_unlock_irq(&nna->lock);
4949	}
4950
4951	call_rcu(&pwq->rcu, rcu_free_pwq);
4952
4953	/*
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.
4956	*/
4957	if (is_last) {
4958	wq_unregister_lockdep(wq);
4959	call_rcu(&wq->rcu, rcu_free_wq);
4960	}
4961	}
4962
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)
4966	{
4967	BUG_ON((unsigned long)pwq & ~WORK_STRUCT_PWQ_MASK);
4968
4969	memset(pwq, 0, sizeof(*pwq));
4970
4971	pwq->pool = pool;
4972	pwq->wq = wq;
4973	pwq->flush_color = -1;
4974	pwq->refcnt = 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);
4980	}
4981
4982	/* sync @pwq with the current state of its associated wq and link it */
4983	static void link_pwq(struct pool_workqueue *pwq)
4984	{
4985	struct workqueue_struct *wq = pwq->wq;
4986
4987	lockdep_assert_held(&wq->mutex);
4988
4989	/* may be called multiple times, ignore if already linked */
4990	if (!list_empty(&pwq->pwqs_node))
4991	return;
4992
4993	/* set the matching work_color */
4994	pwq->work_color = wq->work_color;
4995
4996	/* link in @pwq */
4997	list_add_tail_rcu(&pwq->pwqs_node, &wq->pwqs);
4998	}
4999
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)
5003	{
5004	struct worker_pool *pool;
5005	struct pool_workqueue *pwq;
5006
5007	lockdep_assert_held(&wq_pool_mutex);
5008
5009	pool = get_unbound_pool(attrs);
5010	if (!pool)
5011	return NULL;
5012
5013	pwq = kmem_cache_alloc_node(pwq_cache, GFP_KERNEL, pool->node);
5014	if (!pwq) {
5015	put_unbound_pool(pool);
5016	return NULL;
5017	}
5018
5019	init_pwq(pwq, wq, pool);
5020	return pwq;
5021	}
5022
5023	/**
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
5028	*
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.
5032	*
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.
5036	*
5037	* The caller is responsible for ensuring that the cpumask of @pod stays stable.
5038	*/
5039	static void wq_calc_pod_cpumask(struct workqueue_attrs *attrs, int cpu,
5040	int cpu_going_down)
5041	{
5042	const struct wq_pod_type *pt = wqattrs_pod_type(attrs);
5043	int pod = pt->cpu_pod[cpu];
5044
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);
5050
5051	if (cpumask_empty(attrs->__pod_cpumask)) {
5052	cpumask_copy(attrs->__pod_cpumask, attrs->cpumask);
5053	return;
5054	}
5055
5056	/* yeap, return possible CPUs in @pod that @attrs wants */
5057	cpumask_and(attrs->__pod_cpumask, attrs->cpumask, pt->pod_cpus[pod]);
5058
5059	if (cpumask_empty(attrs->__pod_cpumask))
5060	pr_warn_once("WARNING: workqueue cpumask: online intersect > "
5061	"possible intersect\n");
5062	}
5063
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)
5067	{
5068	struct pool_workqueue __rcu **slot = unbound_pwq_slot(wq, cpu);
5069	struct pool_workqueue *old_pwq;
5070
5071	lockdep_assert_held(&wq_pool_mutex);
5072	lockdep_assert_held(&wq->mutex);
5073
5074	/* link_pwq() can handle duplicate calls */
5075	link_pwq(pwq);
5076
5077	old_pwq = rcu_access_pointer(*slot);
5078	rcu_assign_pointer(*slot, pwq);
5079	return old_pwq;
5080	}
5081
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[];
5089	};
5090
5091	/* free the resources after success or abort */
5092	static void apply_wqattrs_cleanup(struct apply_wqattrs_ctx *ctx)
5093	{
5094	if (ctx) {
5095	int cpu;
5096
5097	for_each_possible_cpu(cpu)
5098	put_pwq_unlocked(ctx->pwq_tbl[cpu]);
5099	put_pwq_unlocked(ctx->dfl_pwq);
5100
5101	free_workqueue_attrs(ctx->attrs);
5102
5103	kfree(ctx);
5104	}
5105	}
5106
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)
5112	{
5113	struct apply_wqattrs_ctx *ctx;
5114	struct workqueue_attrs *new_attrs;
5115	int cpu;
5116
5117	lockdep_assert_held(&wq_pool_mutex);
5118
5119	if (WARN_ON(attrs->affn_scope < 0 \|\|
5120	attrs->affn_scope >= WQ_AFFN_NR_TYPES))
5121	return ERR_PTR(-EINVAL);
5122
5123	ctx = kzalloc(struct_size(ctx, pwq_tbl, nr_cpu_ids), GFP_KERNEL);
5124
5125	new_attrs = alloc_workqueue_attrs();
5126	if (!ctx \|\| !new_attrs)
5127	goto out_free;
5128
5129	/*
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.
5133	*/
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);
5138	if (!ctx->dfl_pwq)
5139	goto out_free;
5140
5141	for_each_possible_cpu(cpu) {
5142	if (new_attrs->ordered) {
5143	ctx->dfl_pwq->refcnt++;
5144	ctx->pwq_tbl[cpu] = ctx->dfl_pwq;
5145	} else {
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])
5149	goto out_free;
5150	}
5151	}
5152
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;
5158
5159	/*
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.
5164	*/
5165	if ((wq->flags & __WQ_ORDERED) && !list_empty(&wq->pwqs))
5166	ctx->dfl_pwq->plugged = true;
5167
5168	ctx->wq = wq;
5169	return ctx;
5170
5171	out_free:
5172	free_workqueue_attrs(new_attrs);
5173	apply_wqattrs_cleanup(ctx);
5174	return ERR_PTR(-ENOMEM);
5175	}
5176
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)
5179	{
5180	int cpu;
5181
5182	/* all pwqs have been created successfully, let's install'em */
5183	mutex_lock(&ctx->wq->mutex);
5184
5185	copy_workqueue_attrs(ctx->wq->unbound_attrs, ctx->attrs);
5186
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,
5190	ctx->pwq_tbl[cpu]);
5191	ctx->dfl_pwq = install_unbound_pwq(ctx->wq, -1, ctx->dfl_pwq);
5192
5193	/* update node_nr_active->max */
5194	wq_update_node_max_active(ctx->wq, -1);
5195
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));
5200
5201	mutex_unlock(&ctx->wq->mutex);
5202	}
5203
5204	static int apply_workqueue_attrs_locked(struct workqueue_struct *wq,
5205	const struct workqueue_attrs *attrs)
5206	{
5207	struct apply_wqattrs_ctx *ctx;
5208
5209	/* only unbound workqueues can change attributes */
5210	if (WARN_ON(!(wq->flags & WQ_UNBOUND)))
5211	return -EINVAL;
5212
5213	ctx = apply_wqattrs_prepare(wq, attrs, wq_unbound_cpumask);
5214	if (IS_ERR(ctx))
5215	return PTR_ERR(ctx);
5216
5217	/* the ctx has been prepared successfully, let's commit it */
5218	apply_wqattrs_commit(ctx);
5219	apply_wqattrs_cleanup(ctx);
5220
5221	return 0;
5222	}
5223
5224	/**
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()
5228	*
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.
5234	*
5235	* Performs GFP_KERNEL allocations.
5236	*
5237	* Assumes caller has CPU hotplug read exclusion, i.e. cpus_read_lock().
5238	*
5239	* Return: 0 on success and -errno on failure.
5240	*/
5241	int apply_workqueue_attrs(struct workqueue_struct *wq,
5242	const struct workqueue_attrs *attrs)
5243	{
5244	int ret;
5245
5246	lockdep_assert_cpus_held();
5247
5248	mutex_lock(&wq_pool_mutex);
5249	ret = apply_workqueue_attrs_locked(wq, attrs);
5250	mutex_unlock(&wq_pool_mutex);
5251
5252	return ret;
5253	}
5254
5255	/**
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
5261	*
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
5264	* @wq accordingly.
5265	*
5266	*
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.
5269	*
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.
5276	*/
5277	static void wq_update_pod(struct workqueue_struct *wq, int cpu,
5278	int hotplug_cpu, bool online)
5279	{
5280	int off_cpu = online ? -1 : hotplug_cpu;
5281	struct pool_workqueue old_pwq = NULL, pwq;
5282	struct workqueue_attrs *target_attrs;
5283
5284	lockdep_assert_held(&wq_pool_mutex);
5285
5286	if (!(wq->flags & WQ_UNBOUND) \|\| wq->unbound_attrs->ordered)
5287	return;
5288
5289	/*
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.
5293	*/
5294	target_attrs = wq_update_pod_attrs_buf;
5295
5296	copy_workqueue_attrs(target_attrs, wq->unbound_attrs);
5297	wqattrs_actualize_cpumask(target_attrs, wq_unbound_cpumask);
5298
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))
5302	return;
5303
5304	/* create a new pwq */
5305	pwq = alloc_unbound_pwq(wq, target_attrs);
5306	if (!pwq) {
5307	pr_warn("workqueue: allocation failed while updating CPU pod affinity of \"%s\"\n",
5308	wq->name);
5309	goto use_dfl_pwq;
5310	}
5311
5312	/* Install the new pwq. */
5313	mutex_lock(&wq->mutex);
5314	old_pwq = install_unbound_pwq(wq, cpu, pwq);
5315	goto out_unlock;
5316
5317	use_dfl_pwq:
5318	mutex_lock(&wq->mutex);
5319	pwq = unbound_pwq(wq, -1);
5320	raw_spin_lock_irq(&pwq->pool->lock);
5321	get_pwq(pwq);
5322	raw_spin_unlock_irq(&pwq->pool->lock);
5323	old_pwq = install_unbound_pwq(wq, cpu, pwq);
5324	out_unlock:
5325	mutex_unlock(&wq->mutex);
5326	put_pwq_unlocked(old_pwq);
5327	}
5328
5329	static int alloc_and_link_pwqs(struct workqueue_struct *wq)
5330	{
5331	bool highpri = wq->flags & WQ_HIGHPRI;
5332	int cpu, ret;
5333
5334	wq->cpu_pwq = alloc_percpu(struct pool_workqueue *);
5335	if (!wq->cpu_pwq)
5336	goto enomem;
5337
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;
5343
5344	if (wq->flags & WQ_BH)
5345	pools = bh_worker_pools;
5346	else
5347	pools = cpu_worker_pools;
5348
5349	pool = &(per_cpu_ptr(pools, cpu)[highpri]);
5350	pwq_p = per_cpu_ptr(wq->cpu_pwq, cpu);
5351
5352	*pwq_p = kmem_cache_alloc_node(pwq_cache, GFP_KERNEL,
5353	pool->node);
5354	if (!*pwq_p)
5355	goto enomem;
5356
5357	init_pwq(*pwq_p, wq, pool);
5358
5359	mutex_lock(&wq->mutex);
5360	link_pwq(*pwq_p);
5361	mutex_unlock(&wq->mutex);
5362	}
5363	return 0;
5364	}
5365
5366	cpus_read_lock();
5367	if (wq->flags & __WQ_ORDERED) {
5368	struct pool_workqueue *dfl_pwq;
5369
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);
5376	} else {
5377	ret = apply_workqueue_attrs(wq, unbound_std_wq_attrs[highpri]);
5378	}
5379	cpus_read_unlock();
5380
5381	/* for unbound pwq, flush the pwq_release_worker ensures that the
5382	* pwq_release_workfn() completes before calling kfree(wq).
5383	*/
5384	if (ret)
5385	kthread_flush_worker(pwq_release_worker);
5386
5387	return ret;
5388
5389	enomem:
5390	if (wq->cpu_pwq) {
5391	for_each_possible_cpu(cpu) {
5392	struct pool_workqueue pwq = per_cpu_ptr(wq->cpu_pwq, cpu);
5393
5394	if (pwq)
5395	kmem_cache_free(pwq_cache, pwq);
5396	}
5397	free_percpu(wq->cpu_pwq);
5398	wq->cpu_pwq = NULL;
5399	}
5400	return -ENOMEM;
5401	}
5402
5403	static int wq_clamp_max_active(int max_active, unsigned int flags,
5404	const char *name)
5405	{
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);
5409
5410	return clamp_val(max_active, 1, WQ_MAX_ACTIVE);
5411	}
5412
5413	/*
5414	* Workqueues which may be used during memory reclaim should have a rescuer
5415	* to guarantee forward progress.
5416	*/
5417	static int init_rescuer(struct workqueue_struct *wq)
5418	{
5419	struct worker *rescuer;
5420	int ret;
5421
5422	if (!(wq->flags & WQ_MEM_RECLAIM))
5423	return 0;
5424
5425	rescuer = alloc_worker(NUMA_NO_NODE);
5426	if (!rescuer) {
5427	pr_err("workqueue: Failed to allocate a rescuer for wq \"%s\"\n",
5428	wq->name);
5429	return -ENOMEM;
5430	}
5431
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));
5438	kfree(rescuer);
5439	return ret;
5440	}
5441
5442	wq->rescuer = rescuer;
5443	if (wq->flags & WQ_UNBOUND)
5444	kthread_bind_mask(rescuer->task, wq_unbound_cpumask);
5445	else
5446	kthread_bind_mask(rescuer->task, cpu_possible_mask);
5447	wake_up_process(rescuer->task);
5448
5449	return 0;
5450	}
5451
5452	/**
5453	* wq_adjust_max_active - update a wq's max_active to the current setting
5454	* @wq: target workqueue
5455	*
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.
5459	*/
5460	static void wq_adjust_max_active(struct workqueue_struct *wq)
5461	{
5462	bool activated;
5463	int new_max, new_min;
5464
5465	lockdep_assert_held(&wq->mutex);
5466
5467	if ((wq->flags & WQ_FREEZABLE) && workqueue_freezing) {
5468	new_max = 0;
5469	new_min = 0;
5470	} else {
5471	new_max = wq->saved_max_active;
5472	new_min = wq->saved_min_active;
5473	}
5474
5475	if (wq->max_active == new_max && wq->min_active == new_min)
5476	return;
5477
5478	/*
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.
5483	*/
5484	WRITE_ONCE(wq->max_active, new_max);
5485	WRITE_ONCE(wq->min_active, new_min);
5486
5487	if (wq->flags & WQ_UNBOUND)
5488	wq_update_node_max_active(wq, -1);
5489
5490	if (new_max == 0)
5491	return;
5492
5493	/*
5494	* Round-robin through pwq's activating the first inactive work item
5495	* until max_active is filled.
5496	*/
5497	do {
5498	struct pool_workqueue *pwq;
5499
5500	activated = false;
5501	for_each_pwq(pwq, wq) {
5502	unsigned long irq_flags;
5503
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)) {
5507	activated = true;
5508	kick_pool(pwq->pool);
5509	}
5510	raw_spin_unlock_irqrestore(&pwq->pool->lock, irq_flags);
5511	}
5512	} while (activated);
5513	}
5514
5515	__printf(1, 4)
5516	struct workqueue_struct alloc_workqueue(const char fmt,
5517	unsigned int flags,
5518	int max_active, ...)
5519	{
5520	va_list args;
5521	struct workqueue_struct *wq;
5522	size_t wq_size;
5523	int name_len;
5524
5525	if (flags & WQ_BH) {
5526	if (WARN_ON_ONCE(flags & ~__WQ_BH_ALLOWS))
5527	return NULL;
5528	if (WARN_ON_ONCE(max_active))
5529	return NULL;
5530	}
5531
5532	/* see the comment above the definition of WQ_POWER_EFFICIENT */
5533	if ((flags & WQ_POWER_EFFICIENT) && wq_power_efficient)
5534	flags \|= WQ_UNBOUND;
5535
5536	/* allocate wq and format name */
5537	if (flags & WQ_UNBOUND)
5538	wq_size = struct_size(wq, node_nr_active, nr_node_ids + 1);
5539	else
5540	wq_size = sizeof(*wq);
5541
5542	wq = kzalloc(wq_size, GFP_KERNEL);
5543	if (!wq)
5544	return NULL;
5545
5546	if (flags & WQ_UNBOUND) {
5547	wq->unbound_attrs = alloc_workqueue_attrs();
5548	if (!wq->unbound_attrs)
5549	goto err_free_wq;
5550	}
5551
5552	va_start(args, max_active);
5553	name_len = vsnprintf(wq->name, sizeof(wq->name), fmt, args);
5554	va_end(args);
5555
5556	if (name_len >= WQ_NAME_LEN)
5557	pr_warn_once("workqueue: name exceeds WQ_NAME_LEN. Truncating to: %s\n",
5558	wq->name);
5559
5560	if (flags & WQ_BH) {
5561	/*
5562	* BH workqueues always share a single execution context per CPU
5563	* and don't impose any max_active limit.
5564	*/
5565	max_active = INT_MAX;
5566	} else {
5567	max_active = max_active ?: WQ_DFL_ACTIVE;
5568	max_active = wq_clamp_max_active(max_active, flags, wq->name);
5569	}
5570
5571	/* init wq */
5572	wq->flags = flags;
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);
5583
5584	wq_init_lockdep(wq);
5585	INIT_LIST_HEAD(&wq->list);
5586
5587	if (flags & WQ_UNBOUND) {
5588	if (alloc_node_nr_active(wq->node_nr_active) < 0)
5589	goto err_unreg_lockdep;
5590	}
5591
5592	if (alloc_and_link_pwqs(wq) < 0)
5593	goto err_free_node_nr_active;
5594
5595	if (wq_online && init_rescuer(wq) < 0)
5596	goto err_destroy;
5597
5598	if ((wq->flags & WQ_SYSFS) && workqueue_sysfs_register(wq))
5599	goto err_destroy;
5600
5601	/*
5602	* wq_pool_mutex protects global freeze state and workqueues list.
5603	* Grab it, adjust max_active and add the new @wq to workqueues
5604	* list.
5605	*/
5606	mutex_lock(&wq_pool_mutex);
5607
5608	mutex_lock(&wq->mutex);
5609	wq_adjust_max_active(wq);
5610	mutex_unlock(&wq->mutex);
5611
5612	list_add_tail_rcu(&wq->list, &workqueues);
5613
5614	mutex_unlock(&wq_pool_mutex);
5615
5616	return wq;
5617
5618	err_free_node_nr_active:
5619	if (wq->flags & WQ_UNBOUND)
5620	free_node_nr_active(wq->node_nr_active);
5621	err_unreg_lockdep:
5622	wq_unregister_lockdep(wq);
5623	wq_free_lockdep(wq);
5624	err_free_wq:
5625	free_workqueue_attrs(wq->unbound_attrs);
5626	kfree(wq);
5627	return NULL;
5628	err_destroy:
5629	destroy_workqueue(wq);
5630	return NULL;
5631	}
5632	EXPORT_SYMBOL_GPL(alloc_workqueue);
5633
5634	static bool pwq_busy(struct pool_workqueue *pwq)
5635	{
5636	int i;
5637
5638	for (i = 0; i < WORK_NR_COLORS; i++)
5639	if (pwq->nr_in_flight[i])
5640	return true;
5641
5642	if ((pwq != rcu_access_pointer(pwq->wq->dfl_pwq)) && (pwq->refcnt > 1))
5643	return true;
5644	if (!pwq_is_empty(pwq))
5645	return true;
5646
5647	return false;
5648	}
5649
5650	/**
5651	* destroy_workqueue - safely terminate a workqueue
5652	* @wq: target workqueue
5653	*
5654	* Safely destroy a workqueue. All work currently pending will be done first.
5655	*/
5656	void destroy_workqueue(struct workqueue_struct *wq)
5657	{
5658	struct pool_workqueue *pwq;
5659	int cpu;
5660
5661	/*
5662	* Remove it from sysfs first so that sanity check failure doesn't
5663	* lead to sysfs name conflicts.
5664	*/
5665	workqueue_sysfs_unregister(wq);
5666
5667	/* mark the workqueue destruction is in progress */
5668	mutex_lock(&wq->mutex);
5669	wq->flags \|= __WQ_DESTROYING;
5670	mutex_unlock(&wq->mutex);
5671
5672	/* drain it before proceeding with destruction */
5673	drain_workqueue(wq);
5674
5675	/* kill rescuer, if sanity checks fail, leave it w/o rescuer */
5676	if (wq->rescuer) {
5677	struct worker *rescuer = wq->rescuer;
5678
5679	/* this prevents new queueing */
5680	raw_spin_lock_irq(&wq_mayday_lock);
5681	wq->rescuer = NULL;
5682	raw_spin_unlock_irq(&wq_mayday_lock);
5683
5684	/* rescuer will empty maydays list before exiting */
5685	kthread_stop(rescuer->task);
5686	kfree(rescuer);
5687	}
5688
5689	/*
5690	* Sanity checks - grab all the locks so that we wait for all
5691	* in-flight operations which may do put_pwq().
5692	*/
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);
5700	show_pwq(pwq);
5701	raw_spin_unlock_irq(&pwq->pool->lock);
5702	mutex_unlock(&wq->mutex);
5703	mutex_unlock(&wq_pool_mutex);
5704	show_one_workqueue(wq);
5705	return;
5706	}
5707	raw_spin_unlock_irq(&pwq->pool->lock);
5708	}
5709	mutex_unlock(&wq->mutex);
5710
5711	/*
5712	* wq list is used to freeze wq, remove from list after
5713	* flushing is complete in case freeze races us.
5714	*/
5715	list_del_rcu(&wq->list);
5716	mutex_unlock(&wq_pool_mutex);
5717
5718	/*
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.
5722	*/
5723	rcu_read_lock();
5724
5725	for_each_possible_cpu(cpu) {
5726	put_pwq_unlocked(unbound_pwq(wq, cpu));
5727	RCU_INIT_POINTER(*unbound_pwq_slot(wq, cpu), NULL);
5728	}
5729
5730	put_pwq_unlocked(unbound_pwq(wq, -1));
5731	RCU_INIT_POINTER(*unbound_pwq_slot(wq, -1), NULL);
5732
5733	rcu_read_unlock();
5734	}
5735	EXPORT_SYMBOL_GPL(destroy_workqueue);
5736
5737	/**
5738	* workqueue_set_max_active - adjust max_active of a workqueue
5739	* @wq: target workqueue
5740	* @max_active: new max_active value.
5741	*
5742	* Set max_active of @wq to @max_active. See the alloc_workqueue() function
5743	* comment.
5744	*
5745	* CONTEXT:
5746	* Don't call from IRQ context.
5747	*/
5748	void workqueue_set_max_active(struct workqueue_struct *wq, int max_active)
5749	{
5750	/* max_active doesn't mean anything for BH workqueues */
5751	if (WARN_ON(wq->flags & WQ_BH))
5752	return;
5753	/* disallow meddling with max_active for ordered workqueues */
5754	if (WARN_ON(wq->flags & __WQ_ORDERED))
5755	return;
5756
5757	max_active = wq_clamp_max_active(max_active, wq->flags, wq->name);
5758
5759	mutex_lock(&wq->mutex);
5760
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);
5764
5765	wq_adjust_max_active(wq);
5766
5767	mutex_unlock(&wq->mutex);
5768	}
5769	EXPORT_SYMBOL_GPL(workqueue_set_max_active);
5770
5771	/**
5772	* workqueue_set_min_active - adjust min_active of an unbound workqueue
5773	* @wq: target unbound workqueue
5774	* @min_active: new min_active value
5775	*
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.
5781	*
5782	* Use this function to adjust the min_active value between 0 and the current
5783	* max_active.
5784	*/
5785	void workqueue_set_min_active(struct workqueue_struct *wq, int min_active)
5786	{
5787	/* min_active is only meaningful for non-ordered unbound workqueues */
5788	if (WARN_ON((wq->flags & (WQ_BH \| WQ_UNBOUND \| __WQ_ORDERED)) !=
5789	WQ_UNBOUND))
5790	return;
5791
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);
5796	}
5797
5798	/**
5799	* current_work - retrieve %current task's work struct
5800	*
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.
5803	*
5804	* Return: work struct if %current task is a workqueue worker, %NULL otherwise.
5805	*/
5806	struct work_struct *current_work(void)
5807	{
5808	struct worker *worker = current_wq_worker();
5809
5810	return worker ? worker->current_work : NULL;
5811	}
5812	EXPORT_SYMBOL(current_work);
5813
5814	/**
5815	* current_is_workqueue_rescuer - is %current workqueue rescuer?
5816	*
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.
5819	*
5820	* Return: %true if %current is a workqueue rescuer. %false otherwise.
5821	*/
5822	bool current_is_workqueue_rescuer(void)
5823	{
5824	struct worker *worker = current_wq_worker();
5825
5826	return worker && worker->rescue_wq;
5827	}
5828
5829	/**
5830	* workqueue_congested - test whether a workqueue is congested
5831	* @cpu: CPU in question
5832	* @wq: target workqueue
5833	*
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.
5837	*
5838	* If @cpu is WORK_CPU_UNBOUND, the test is performed on the local CPU.
5839	*
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
5843	* other CPUs.
5844	*
5845	* Return:
5846	* %true if congested, %false otherwise.
5847	*/
5848	bool workqueue_congested(int cpu, struct workqueue_struct *wq)
5849	{
5850	struct pool_workqueue *pwq;
5851	bool ret;
5852
5853	rcu_read_lock();
5854	preempt_disable();
5855
5856	if (cpu == WORK_CPU_UNBOUND)
5857	cpu = smp_processor_id();
5858
5859	pwq = *per_cpu_ptr(wq->cpu_pwq, cpu);
5860	ret = !list_empty(&pwq->inactive_works);
5861
5862	preempt_enable();
5863	rcu_read_unlock();
5864
5865	return ret;
5866	}
5867	EXPORT_SYMBOL_GPL(workqueue_congested);
5868
5869	/**
5870	* work_busy - test whether a work is currently pending or running
5871	* @work: the work to be tested
5872	*
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.
5876	*
5877	* Return:
5878	* OR'd bitmask of WORK_BUSY_* bits.
5879	*/
5880	unsigned int work_busy(struct work_struct *work)
5881	{
5882	struct worker_pool *pool;
5883	unsigned long irq_flags;
5884	unsigned int ret = 0;
5885
5886	if (work_pending(work))
5887	ret \|= WORK_BUSY_PENDING;
5888
5889	rcu_read_lock();
5890	pool = get_work_pool(work);
5891	if (pool) {
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);
5896	}
5897	rcu_read_unlock();
5898
5899	return ret;
5900	}
5901	EXPORT_SYMBOL_GPL(work_busy);
5902
5903	/**
5904	* set_worker_desc - set description for the current work item
5905	* @fmt: printf-style format string
5906	* @...: arguments for the format string
5907	*
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'.
5912	*/
5913	void set_worker_desc(const char *fmt, ...)
5914	{
5915	struct worker *worker = current_wq_worker();
5916	va_list args;
5917
5918	if (worker) {
5919	va_start(args, fmt);
5920	vsnprintf(worker->desc, sizeof(worker->desc), fmt, args);
5921	va_end(args);
5922	}
5923	}
5924	EXPORT_SYMBOL_GPL(set_worker_desc);
5925
5926	/**
5927	* print_worker_info - print out worker information and description
5928	* @log_lvl: the log level to use when printing
5929	* @task: target task
5930	*
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.
5934	*
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.
5938	*/
5939	void print_worker_info(const char log_lvl, struct task_struct task)
5940	{
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;
5947
5948	if (!(task->flags & PF_WQ_WORKER))
5949	return;
5950
5951	/*
5952	* This function is called without any synchronization and @task
5953	* could be in any state. Be careful with dereferences.
5954	*/
5955	worker = kthread_probe_data(task);
5956
5957	/*
5958	* Carefully copy the associated workqueue's workfn, name and desc.
5959	* Keep the original last '\0' in case the original is garbage.
5960	*/
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);
5966
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);
5971	pr_cont("\n");
5972	}
5973	}
5974
5975	static void pr_cont_pool_info(struct worker_pool *pool)
5976	{
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)
5982	pr_cont(" bh%s",
5983	pool->attrs->nice == HIGHPRI_NICE_LEVEL ? "-hi" : "");
5984	else
5985	pr_cont(" nice=%d", pool->attrs->nice);
5986	}
5987
5988	static void pr_cont_worker_id(struct worker *worker)
5989	{
5990	struct worker_pool *pool = worker->pool;
5991
5992	if (pool->flags & WQ_BH)
5993	pr_cont("bh%s",
5994	pool->attrs->nice == HIGHPRI_NICE_LEVEL ? "-hi" : "");
5995	else
5996	pr_cont("%d%s", task_pid_nr(worker->task),
5997	worker->rescue_wq ? "(RESCUER)" : "");
5998	}
5999
6000	struct pr_cont_work_struct {
6001	bool comma;
6002	work_func_t func;
6003	long ctr;
6004	};
6005
6006	static void pr_cont_work_flush(bool comma, work_func_t func, struct pr_cont_work_struct *pcwsp)
6007	{
6008	if (!pcwsp->ctr)
6009	goto out_record;
6010	if (func == pcwsp->func) {
6011	pcwsp->ctr++;
6012	return;
6013	}
6014	if (pcwsp->ctr == 1)
6015	pr_cont("%s %ps", pcwsp->comma ? "," : "", pcwsp->func);
6016	else
6017	pr_cont("%s %ld*%ps", pcwsp->comma ? "," : "", pcwsp->ctr, pcwsp->func);
6018	pcwsp->ctr = 0;
6019	out_record:
6020	if ((long)func == -1L)
6021	return;
6022	pcwsp->comma = comma;
6023	pcwsp->func = func;
6024	pcwsp->ctr = 1;
6025	}
6026
6027	static void pr_cont_work(bool comma, struct work_struct work, struct pr_cont_work_struct pcwsp)
6028	{
6029	if (work->func == wq_barrier_func) {
6030	struct wq_barrier *barr;
6031
6032	barr = container_of(work, struct wq_barrier, work);
6033
6034	pr_cont_work_flush(comma, (work_func_t)-1, pcwsp);
6035	pr_cont("%s BAR(%d)", comma ? "," : "",
6036	task_pid_nr(barr->task));
6037	} else {
6038	if (!comma)
6039	pr_cont_work_flush(comma, (work_func_t)-1, pcwsp);
6040	pr_cont_work_flush(comma, work->func, pcwsp);
6041	}
6042	}
6043
6044	static void show_pwq(struct pool_workqueue *pwq)
6045	{
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;
6051	int bkt;
6052
6053	pr_info(" pwq %d:", pool->id);
6054	pr_cont_pool_info(pool);
6055
6056	pr_cont(" active=%d refcnt=%d%s\n",
6057	pwq->nr_active, pwq->refcnt,
6058	!list_empty(&pwq->mayday_node) ? " MAYDAY" : "");
6059
6060	hash_for_each(pool->busy_hash, bkt, worker, hentry) {
6061	if (worker->current_pwq == pwq) {
6062	has_in_flight = true;
6063	break;
6064	}
6065	}
6066	if (has_in_flight) {
6067	bool comma = false;
6068
6069	pr_info(" in-flight:");
6070	hash_for_each(pool->busy_hash, bkt, worker, hentry) {
6071	if (worker->current_pwq != pwq)
6072	continue;
6073
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);
6080	comma = true;
6081	}
6082	pr_cont("\n");
6083	}
6084
6085	list_for_each_entry(work, &pool->worklist, entry) {
6086	if (get_work_pwq(work) == pwq) {
6087	has_pending = true;
6088	break;
6089	}
6090	}
6091	if (has_pending) {
6092	bool comma = false;
6093
6094	pr_info(" pending:");
6095	list_for_each_entry(work, &pool->worklist, entry) {
6096	if (get_work_pwq(work) != pwq)
6097	continue;
6098
6099	pr_cont_work(comma, work, &pcws);
6100	comma = !(*work_data_bits(work) & WORK_STRUCT_LINKED);
6101	}
6102	pr_cont_work_flush(comma, (work_func_t)-1L, &pcws);
6103	pr_cont("\n");
6104	}
6105
6106	if (!list_empty(&pwq->inactive_works)) {
6107	bool comma = false;
6108
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);
6113	}
6114	pr_cont_work_flush(comma, (work_func_t)-1L, &pcws);
6115	pr_cont("\n");
6116	}
6117	}
6118
6119	/**
6120	* show_one_workqueue - dump state of specified workqueue
6121	* @wq: workqueue whose state will be printed
6122	*/
6123	void show_one_workqueue(struct workqueue_struct *wq)
6124	{
6125	struct pool_workqueue *pwq;
6126	bool idle = true;
6127	unsigned long irq_flags;
6128
6129	for_each_pwq(pwq, wq) {
6130	if (!pwq_is_empty(pwq)) {
6131	idle = false;
6132	break;
6133	}
6134	}
6135	if (idle) /* Nothing to print for idle workqueue */
6136	return;
6137
6138	pr_info("workqueue %s: flags=0x%x\n", wq->name, wq->flags);
6139
6140	for_each_pwq(pwq, wq) {
6141	raw_spin_lock_irqsave(&pwq->pool->lock, irq_flags);
6142	if (!pwq_is_empty(pwq)) {
6143	/*
6144	* Defer printing to avoid deadlocks in console
6145	* drivers that queue work while holding locks
6146	* also taken in their write paths.
6147	*/
6148	printk_deferred_enter();
6149	show_pwq(pwq);
6150	printk_deferred_exit();
6151	}
6152	raw_spin_unlock_irqrestore(&pwq->pool->lock, irq_flags);
6153	/*
6154	* We could be printing a lot from atomic context, e.g.
6155	* sysrq-t -> show_all_workqueues(). Avoid triggering
6156	* hard lockup.
6157	*/
6158	touch_nmi_watchdog();
6159	}
6160
6161	}
6162
6163	/**
6164	* show_one_worker_pool - dump state of specified worker pool
6165	* @pool: worker pool whose state will be printed
6166	*/
6167	static void show_one_worker_pool(struct worker_pool *pool)
6168	{
6169	struct worker *worker;
6170	bool first = true;
6171	unsigned long irq_flags;
6172	unsigned long hung = 0;
6173
6174	raw_spin_lock_irqsave(&pool->lock, irq_flags);
6175	if (pool->nr_workers == pool->nr_idle)
6176	goto next_pool;
6177
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;
6181
6182	/*
6183	* Defer printing to avoid deadlocks in console drivers that
6184	* queue work while holding locks also taken in their write
6185	* paths.
6186	*/
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);
6191	if (pool->manager)
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);
6197	first = false;
6198	}
6199	pr_cont("\n");
6200	printk_deferred_exit();
6201	next_pool:
6202	raw_spin_unlock_irqrestore(&pool->lock, irq_flags);
6203	/*
6204	* We could be printing a lot from atomic context, e.g.
6205	* sysrq-t -> show_all_workqueues(). Avoid triggering
6206	* hard lockup.
6207	*/
6208	touch_nmi_watchdog();
6209
6210	}
6211
6212	/**
6213	* show_all_workqueues - dump workqueue state
6214	*
6215	* Called from a sysrq handler and prints out all busy workqueues and pools.
6216	*/
6217	void show_all_workqueues(void)
6218	{
6219	struct workqueue_struct *wq;
6220	struct worker_pool *pool;
6221	int pi;
6222
6223	rcu_read_lock();
6224
6225	pr_info("Showing busy workqueues and worker pools:\n");
6226
6227	list_for_each_entry_rcu(wq, &workqueues, list)
6228	show_one_workqueue(wq);
6229
6230	for_each_pool(pool, pi)
6231	show_one_worker_pool(pool);
6232
6233	rcu_read_unlock();
6234	}
6235
6236	/**
6237	* show_freezable_workqueues - dump freezable workqueue state
6238	*
6239	* Called from try_to_freeze_tasks() and prints out all freezable workqueues
6240	* still busy.
6241	*/
6242	void show_freezable_workqueues(void)
6243	{
6244	struct workqueue_struct *wq;
6245
6246	rcu_read_lock();
6247
6248	pr_info("Showing freezable workqueues that are still busy:\n");
6249
6250	list_for_each_entry_rcu(wq, &workqueues, list) {
6251	if (!(wq->flags & WQ_FREEZABLE))
6252	continue;
6253	show_one_workqueue(wq);
6254	}
6255
6256	rcu_read_unlock();
6257	}
6258
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)
6261	{
6262	int off;
6263
6264	/* always show the actual comm */
6265	off = strscpy(buf, task->comm, size);
6266	if (off < 0)
6267	return;
6268
6269	/* stabilize PF_WQ_WORKER and worker pool association */
6270	mutex_lock(&wq_pool_attach_mutex);
6271
6272	if (task->flags & PF_WQ_WORKER) {
6273	struct worker *worker = kthread_data(task);
6274	struct worker_pool *pool = worker->pool;
6275
6276	if (pool) {
6277	raw_spin_lock_irq(&pool->lock);
6278	/*
6279	* ->desc tracks information (wq name or
6280	* set_worker_desc()) for the latest execution. If
6281	* current, prepend '+', otherwise '-'.
6282	*/
6283	if (worker->desc[0] != '\0') {
6284	if (worker->current_work)
6285	scnprintf(buf + off, size - off, "+%s",
6286	worker->desc);
6287	else
6288	scnprintf(buf + off, size - off, "-%s",
6289	worker->desc);
6290	}
6291	raw_spin_unlock_irq(&pool->lock);
6292	}
6293	}
6294
6295	mutex_unlock(&wq_pool_attach_mutex);
6296	}
6297
6298	#ifdef CONFIG_SMP
6299
6300	/*
6301	* CPU hotplug.
6302	*
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.
6309	*
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.
6313	*/
6314
6315	static void unbind_workers(int cpu)
6316	{
6317	struct worker_pool *pool;
6318	struct worker *worker;
6319
6320	for_each_cpu_worker_pool(pool, cpu) {
6321	mutex_lock(&wq_pool_attach_mutex);
6322	raw_spin_lock_irq(&pool->lock);
6323
6324	/*
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.
6331	*/
6332	for_each_pool_worker(worker, pool)
6333	worker->flags \|= WORKER_UNBOUND;
6334
6335	pool->flags \|= POOL_DISASSOCIATED;
6336
6337	/*
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.
6344	*/
6345	pool->nr_running = 0;
6346
6347	/*
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.
6351	*/
6352	kick_pool(pool);
6353
6354	raw_spin_unlock_irq(&pool->lock);
6355
6356	for_each_pool_worker(worker, pool)
6357	unbind_worker(worker);
6358
6359	mutex_unlock(&wq_pool_attach_mutex);
6360	}
6361	}
6362
6363	/**
6364	* rebind_workers - rebind all workers of a pool to the associated CPU
6365	* @pool: pool of interest
6366	*
6367	* @pool->cpu is coming online. Rebind all workers to the CPU.
6368	*/
6369	static void rebind_workers(struct worker_pool *pool)
6370	{
6371	struct worker *worker;
6372
6373	lockdep_assert_held(&wq_pool_attach_mutex);
6374
6375	/*
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.
6381	*/
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);
6386	}
6387
6388	raw_spin_lock_irq(&pool->lock);
6389
6390	pool->flags &= ~POOL_DISASSOCIATED;
6391
6392	for_each_pool_worker(worker, pool) {
6393	unsigned int worker_flags = worker->flags;
6394
6395	/*
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.
6403	*
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.
6409	*/
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);
6414	}
6415
6416	raw_spin_unlock_irq(&pool->lock);
6417	}
6418
6419	/**
6420	* restore_unbound_workers_cpumask - restore cpumask of unbound workers
6421	* @pool: unbound pool of interest
6422	* @cpu: the CPU which is coming up
6423	*
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.
6428	*/
6429	static void restore_unbound_workers_cpumask(struct worker_pool *pool, int cpu)
6430	{
6431	static cpumask_t cpumask;
6432	struct worker *worker;
6433
6434	lockdep_assert_held(&wq_pool_attach_mutex);
6435
6436	/* is @cpu allowed for @pool? */
6437	if (!cpumask_test_cpu(cpu, pool->attrs->cpumask))
6438	return;
6439
6440	cpumask_and(&cpumask, pool->attrs->cpumask, cpu_online_mask);
6441
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);
6445	}
6446
6447	int workqueue_prepare_cpu(unsigned int cpu)
6448	{
6449	struct worker_pool *pool;
6450
6451	for_each_cpu_worker_pool(pool, cpu) {
6452	if (pool->nr_workers)
6453	continue;
6454	if (!create_worker(pool))
6455	return -ENOMEM;
6456	}
6457	return 0;
6458	}
6459
6460	int workqueue_online_cpu(unsigned int cpu)
6461	{
6462	struct worker_pool *pool;
6463	struct workqueue_struct *wq;
6464	int pi;
6465
6466	mutex_lock(&wq_pool_mutex);
6467
6468	for_each_pool(pool, pi) {
6469	/* BH pools aren't affected by hotplug */
6470	if (pool->flags & POOL_BH)
6471	continue;
6472
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);
6479	}
6480
6481	/* update pod affinity of unbound workqueues */
6482	list_for_each_entry(wq, &workqueues, list) {
6483	struct workqueue_attrs *attrs = wq->unbound_attrs;
6484
6485	if (attrs) {
6486	const struct wq_pod_type *pt = wqattrs_pod_type(attrs);
6487	int tcpu;
6488
6489	for_each_cpu(tcpu, pt->pod_cpus[pt->cpu_pod[cpu]])
6490	wq_update_pod(wq, tcpu, cpu, true);
6491
6492	mutex_lock(&wq->mutex);
6493	wq_update_node_max_active(wq, -1);
6494	mutex_unlock(&wq->mutex);
6495	}
6496	}
6497
6498	mutex_unlock(&wq_pool_mutex);
6499	return 0;
6500	}
6501
6502	int workqueue_offline_cpu(unsigned int cpu)
6503	{
6504	struct workqueue_struct *wq;
6505
6506	/* unbinding per-cpu workers should happen on the local CPU */
6507	if (WARN_ON(cpu != smp_processor_id()))
6508	return -1;
6509
6510	unbind_workers(cpu);
6511
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;
6516
6517	if (attrs) {
6518	const struct wq_pod_type *pt = wqattrs_pod_type(attrs);
6519	int tcpu;
6520
6521	for_each_cpu(tcpu, pt->pod_cpus[pt->cpu_pod[cpu]])
6522	wq_update_pod(wq, tcpu, cpu, false);
6523
6524	mutex_lock(&wq->mutex);
6525	wq_update_node_max_active(wq, cpu);
6526	mutex_unlock(&wq->mutex);
6527	}
6528	}
6529	mutex_unlock(&wq_pool_mutex);
6530
6531	return 0;
6532	}
6533
6534	struct work_for_cpu {
6535	struct work_struct work;
6536	long (fn)(void );
6537	void *arg;
6538	long ret;
6539	};
6540
6541	static void work_for_cpu_fn(struct work_struct *work)
6542	{
6543	struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work);
6544
6545	wfc->ret = wfc->fn(wfc->arg);
6546	}
6547
6548	/**
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
6554	*
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.
6557	*
6558	* Return: The value @fn returns.
6559	*/
6560	long work_on_cpu_key(int cpu, long (fn)(void ),
6561	void arg, struct lock_class_key key)
6562	{
6563	struct work_for_cpu wfc = { .fn = fn, .arg = arg };
6564
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);
6569	return wfc.ret;
6570	}
6571	EXPORT_SYMBOL_GPL(work_on_cpu_key);
6572
6573	/**
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
6579	*
6580	* Disables CPU hotplug and calls work_on_cpu(). The caller must not hold
6581	* any locks which would prevent @fn from completing.
6582	*
6583	* Return: The value @fn returns.
6584	*/
6585	long work_on_cpu_safe_key(int cpu, long (fn)(void ),
6586	void arg, struct lock_class_key key)
6587	{
6588	long ret = -ENODEV;
6589
6590	cpus_read_lock();
6591	if (cpu_online(cpu))
6592	ret = work_on_cpu_key(cpu, fn, arg, key);
6593	cpus_read_unlock();
6594	return ret;
6595	}
6596	EXPORT_SYMBOL_GPL(work_on_cpu_safe_key);
6597	#endif /* CONFIG_SMP */
6598
6599	#ifdef CONFIG_FREEZER
6600
6601	/**
6602	* freeze_workqueues_begin - begin freezing workqueues
6603	*
6604	* Start freezing workqueues. After this function returns, all freezable
6605	* workqueues will queue new works to their inactive_works list instead of
6606	* pool->worklist.
6607	*
6608	* CONTEXT:
6609	* Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's.
6610	*/
6611	void freeze_workqueues_begin(void)
6612	{
6613	struct workqueue_struct *wq;
6614
6615	mutex_lock(&wq_pool_mutex);
6616
6617	WARN_ON_ONCE(workqueue_freezing);
6618	workqueue_freezing = true;
6619
6620	list_for_each_entry(wq, &workqueues, list) {
6621	mutex_lock(&wq->mutex);
6622	wq_adjust_max_active(wq);
6623	mutex_unlock(&wq->mutex);
6624	}
6625
6626	mutex_unlock(&wq_pool_mutex);
6627	}
6628
6629	/**
6630	* freeze_workqueues_busy - are freezable workqueues still busy?
6631	*
6632	* Check whether freezing is complete. This function must be called
6633	* between freeze_workqueues_begin() and thaw_workqueues().
6634	*
6635	* CONTEXT:
6636	* Grabs and releases wq_pool_mutex.
6637	*
6638	* Return:
6639	* %true if some freezable workqueues are still busy. %false if freezing
6640	* is complete.
6641	*/
6642	bool freeze_workqueues_busy(void)
6643	{
6644	bool busy = false;
6645	struct workqueue_struct *wq;
6646	struct pool_workqueue *pwq;
6647
6648	mutex_lock(&wq_pool_mutex);
6649
6650	WARN_ON_ONCE(!workqueue_freezing);
6651
6652	list_for_each_entry(wq, &workqueues, list) {
6653	if (!(wq->flags & WQ_FREEZABLE))
6654	continue;
6655	/*
6656	* nr_active is monotonically decreasing. It's safe
6657	* to peek without lock.
6658	*/
6659	rcu_read_lock();
6660	for_each_pwq(pwq, wq) {
6661	WARN_ON_ONCE(pwq->nr_active < 0);
6662	if (pwq->nr_active) {
6663	busy = true;
6664	rcu_read_unlock();
6665	goto out_unlock;
6666	}
6667	}
6668	rcu_read_unlock();
6669	}
6670	out_unlock:
6671	mutex_unlock(&wq_pool_mutex);
6672	return busy;
6673	}
6674
6675	/**
6676	* thaw_workqueues - thaw workqueues
6677	*
6678	* Thaw workqueues. Normal queueing is restored and all collected
6679	* frozen works are transferred to their respective pool worklists.
6680	*
6681	* CONTEXT:
6682	* Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's.
6683	*/
6684	void thaw_workqueues(void)
6685	{
6686	struct workqueue_struct *wq;
6687
6688	mutex_lock(&wq_pool_mutex);
6689
6690	if (!workqueue_freezing)
6691	goto out_unlock;
6692
6693	workqueue_freezing = false;
6694
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);
6700	}
6701
6702	out_unlock:
6703	mutex_unlock(&wq_pool_mutex);
6704	}
6705	#endif /* CONFIG_FREEZER */
6706
6707	static int workqueue_apply_unbound_cpumask(const cpumask_var_t unbound_cpumask)
6708	{
6709	LIST_HEAD(ctxs);
6710	int ret = 0;
6711	struct workqueue_struct *wq;
6712	struct apply_wqattrs_ctx ctx, n;
6713
6714	lockdep_assert_held(&wq_pool_mutex);
6715
6716	list_for_each_entry(wq, &workqueues, list) {
6717	if (!(wq->flags & WQ_UNBOUND) \|\| (wq->flags & __WQ_DESTROYING))
6718	continue;
6719
6720	ctx = apply_wqattrs_prepare(wq, wq->unbound_attrs, unbound_cpumask);
6721	if (IS_ERR(ctx)) {
6722	ret = PTR_ERR(ctx);
6723	break;
6724	}
6725
6726	list_add_tail(&ctx->list, &ctxs);
6727	}
6728
6729	list_for_each_entry_safe(ctx, n, &ctxs, list) {
6730	if (!ret)
6731	apply_wqattrs_commit(ctx);
6732	apply_wqattrs_cleanup(ctx);
6733	}
6734
6735	if (!ret) {
6736	mutex_lock(&wq_pool_attach_mutex);
6737	cpumask_copy(wq_unbound_cpumask, unbound_cpumask);
6738	mutex_unlock(&wq_pool_attach_mutex);
6739	}
6740	return ret;
6741	}
6742
6743	/**
6744	* workqueue_unbound_exclude_cpumask - Exclude given CPUs from unbound cpumask
6745	* @exclude_cpumask: the cpumask to be excluded from wq_unbound_cpumask
6746	*
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.
6750	*/
6751	int workqueue_unbound_exclude_cpumask(cpumask_var_t exclude_cpumask)
6752	{
6753	cpumask_var_t cpumask;
6754	int ret = 0;
6755
6756	if (!zalloc_cpumask_var(&cpumask, GFP_KERNEL))
6757	return -ENOMEM;
6758
6759	lockdep_assert_cpus_held();
6760	mutex_lock(&wq_pool_mutex);
6761
6762	/* Save the current isolated cpumask & export it via sysfs */
6763	cpumask_copy(wq_isolated_cpumask, exclude_cpumask);
6764
6765	/*
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.
6770	*/
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);
6775
6776	mutex_unlock(&wq_pool_mutex);
6777	free_cpumask_var(cpumask);
6778	return ret;
6779	}
6780
6781	static int parse_affn_scope(const char *val)
6782	{
6783	int i;
6784
6785	for (i = 0; i < ARRAY_SIZE(wq_affn_names); i++) {
6786	if (!strncasecmp(val, wq_affn_names[i], strlen(wq_affn_names[i])))
6787	return i;
6788	}
6789	return -EINVAL;
6790	}
6791
6792	static int wq_affn_dfl_set(const char val, const struct kernel_param kp)
6793	{
6794	struct workqueue_struct *wq;
6795	int affn, cpu;
6796
6797	affn = parse_affn_scope(val);
6798	if (affn < 0)
6799	return affn;
6800	if (affn == WQ_AFFN_DFL)
6801	return -EINVAL;
6802
6803	cpus_read_lock();
6804	mutex_lock(&wq_pool_mutex);
6805
6806	wq_affn_dfl = affn;
6807
6808	list_for_each_entry(wq, &workqueues, list) {
6809	for_each_online_cpu(cpu) {
6810	wq_update_pod(wq, cpu, cpu, true);
6811	}
6812	}
6813
6814	mutex_unlock(&wq_pool_mutex);
6815	cpus_read_unlock();
6816
6817	return 0;
6818	}
6819
6820	static int wq_affn_dfl_get(char buffer, const struct kernel_param kp)
6821	{
6822	return scnprintf(buffer, PAGE_SIZE, "%s\n", wq_affn_names[wq_affn_dfl]);
6823	}
6824
6825	static const struct kernel_param_ops wq_affn_dfl_ops = {
6826	.set = wq_affn_dfl_set,
6827	.get = wq_affn_dfl_get,
6828	};
6829
6830	module_param_cb(default_affinity_scope, &wq_affn_dfl_ops, NULL, 0644);
6831
6832	#ifdef CONFIG_SYSFS
6833	/*
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.
6837	*
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
6840	*
6841	* Unbound workqueues have the following extra attributes.
6842	*
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
6847	*/
6848	struct wq_device {
6849	struct workqueue_struct *wq;
6850	struct device dev;
6851	};
6852
6853	static struct workqueue_struct dev_to_wq(struct device dev)
6854	{
6855	struct wq_device *wq_dev = container_of(dev, struct wq_device, dev);
6856
6857	return wq_dev->wq;
6858	}
6859
6860	static ssize_t per_cpu_show(struct device dev, struct device_attribute attr,
6861	char *buf)
6862	{
6863	struct workqueue_struct *wq = dev_to_wq(dev);
6864
6865	return scnprintf(buf, PAGE_SIZE, "%d\n", (bool)!(wq->flags & WQ_UNBOUND));
6866	}
6867	static DEVICE_ATTR_RO(per_cpu);
6868
6869	static ssize_t max_active_show(struct device *dev,
6870	struct device_attribute attr, char buf)
6871	{
6872	struct workqueue_struct *wq = dev_to_wq(dev);
6873
6874	return scnprintf(buf, PAGE_SIZE, "%d\n", wq->saved_max_active);
6875	}
6876
6877	static ssize_t max_active_store(struct device *dev,
6878	struct device_attribute attr, const char buf,
6879	size_t count)
6880	{
6881	struct workqueue_struct *wq = dev_to_wq(dev);
6882	int val;
6883
6884	if (sscanf(buf, "%d", &val) != 1 \|\| val <= 0)
6885	return -EINVAL;
6886
6887	workqueue_set_max_active(wq, val);
6888	return count;
6889	}
6890	static DEVICE_ATTR_RW(max_active);
6891
6892	static struct attribute *wq_sysfs_attrs[] = {
6893	&dev_attr_per_cpu.attr,
6894	&dev_attr_max_active.attr,
6895	NULL,
6896	};
6897	ATTRIBUTE_GROUPS(wq_sysfs);
6898
6899	static void apply_wqattrs_lock(void)
6900	{
6901	/* CPUs should stay stable across pwq creations and installations */
6902	cpus_read_lock();
6903	mutex_lock(&wq_pool_mutex);
6904	}
6905
6906	static void apply_wqattrs_unlock(void)
6907	{
6908	mutex_unlock(&wq_pool_mutex);
6909	cpus_read_unlock();
6910	}
6911
6912	static ssize_t wq_nice_show(struct device dev, struct device_attribute attr,
6913	char *buf)
6914	{
6915	struct workqueue_struct *wq = dev_to_wq(dev);
6916	int written;
6917
6918	mutex_lock(&wq->mutex);
6919	written = scnprintf(buf, PAGE_SIZE, "%d\n", wq->unbound_attrs->nice);
6920	mutex_unlock(&wq->mutex);
6921
6922	return written;
6923	}
6924
6925	/* prepare workqueue_attrs for sysfs store operations */
6926	static struct workqueue_attrs wq_sysfs_prep_attrs(struct workqueue_struct wq)
6927	{
6928	struct workqueue_attrs *attrs;
6929
6930	lockdep_assert_held(&wq_pool_mutex);
6931
6932	attrs = alloc_workqueue_attrs();
6933	if (!attrs)
6934	return NULL;
6935
6936	copy_workqueue_attrs(attrs, wq->unbound_attrs);
6937	return attrs;
6938	}
6939
6940	static ssize_t wq_nice_store(struct device dev, struct device_attribute attr,
6941	const char *buf, size_t count)
6942	{
6943	struct workqueue_struct *wq = dev_to_wq(dev);
6944	struct workqueue_attrs *attrs;
6945	int ret = -ENOMEM;
6946
6947	apply_wqattrs_lock();
6948
6949	attrs = wq_sysfs_prep_attrs(wq);
6950	if (!attrs)
6951	goto out_unlock;
6952
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);
6956	else
6957	ret = -EINVAL;
6958
6959	out_unlock:
6960	apply_wqattrs_unlock();
6961	free_workqueue_attrs(attrs);
6962	return ret ?: count;
6963	}
6964
6965	static ssize_t wq_cpumask_show(struct device *dev,
6966	struct device_attribute attr, char buf)
6967	{
6968	struct workqueue_struct *wq = dev_to_wq(dev);
6969	int written;
6970
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);
6975	return written;
6976	}
6977
6978	static ssize_t wq_cpumask_store(struct device *dev,
6979	struct device_attribute *attr,
6980	const char *buf, size_t count)
6981	{
6982	struct workqueue_struct *wq = dev_to_wq(dev);
6983	struct workqueue_attrs *attrs;
6984	int ret = -ENOMEM;
6985
6986	apply_wqattrs_lock();
6987
6988	attrs = wq_sysfs_prep_attrs(wq);
6989	if (!attrs)
6990	goto out_unlock;
6991
6992	ret = cpumask_parse(buf, attrs->cpumask);
6993	if (!ret)
6994	ret = apply_workqueue_attrs_locked(wq, attrs);
6995
6996	out_unlock:
6997	apply_wqattrs_unlock();
6998	free_workqueue_attrs(attrs);
6999	return ret ?: count;
7000	}
7001
7002	static ssize_t wq_affn_scope_show(struct device *dev,
7003	struct device_attribute attr, char buf)
7004	{
7005	struct workqueue_struct *wq = dev_to_wq(dev);
7006	int written;
7007
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]);
7013	else
7014	written = scnprintf(buf, PAGE_SIZE, "%s\n",
7015	wq_affn_names[wq->unbound_attrs->affn_scope]);
7016	mutex_unlock(&wq->mutex);
7017
7018	return written;
7019	}
7020
7021	static ssize_t wq_affn_scope_store(struct device *dev,
7022	struct device_attribute *attr,
7023	const char *buf, size_t count)
7024	{
7025	struct workqueue_struct *wq = dev_to_wq(dev);
7026	struct workqueue_attrs *attrs;
7027	int affn, ret = -ENOMEM;
7028
7029	affn = parse_affn_scope(buf);
7030	if (affn < 0)
7031	return affn;
7032
7033	apply_wqattrs_lock();
7034	attrs = wq_sysfs_prep_attrs(wq);
7035	if (attrs) {
7036	attrs->affn_scope = affn;
7037	ret = apply_workqueue_attrs_locked(wq, attrs);
7038	}
7039	apply_wqattrs_unlock();
7040	free_workqueue_attrs(attrs);
7041	return ret ?: count;
7042	}
7043
7044	static ssize_t wq_affinity_strict_show(struct device *dev,
7045	struct device_attribute attr, char buf)
7046	{
7047	struct workqueue_struct *wq = dev_to_wq(dev);
7048
7049	return scnprintf(buf, PAGE_SIZE, "%d\n",
7050	wq->unbound_attrs->affn_strict);
7051	}
7052
7053	static ssize_t wq_affinity_strict_store(struct device *dev,
7054	struct device_attribute *attr,
7055	const char *buf, size_t count)
7056	{
7057	struct workqueue_struct *wq = dev_to_wq(dev);
7058	struct workqueue_attrs *attrs;
7059	int v, ret = -ENOMEM;
7060
7061	if (sscanf(buf, "%d", &v) != 1)
7062	return -EINVAL;
7063
7064	apply_wqattrs_lock();
7065	attrs = wq_sysfs_prep_attrs(wq);
7066	if (attrs) {
7067	attrs->affn_strict = (bool)v;
7068	ret = apply_workqueue_attrs_locked(wq, attrs);
7069	}
7070	apply_wqattrs_unlock();
7071	free_workqueue_attrs(attrs);
7072	return ret ?: count;
7073	}
7074
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),
7080	__ATTR_NULL,
7081	};
7082
7083	static const struct bus_type wq_subsys = {
7084	.name = "workqueue",
7085	.dev_groups = wq_sysfs_groups,
7086	};
7087
7088	/**
7089	* workqueue_set_unbound_cpumask - Set the low-level unbound cpumask
7090	* @cpumask: the cpumask to set
7091	*
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.
7095	*
7096	* Return: 0 - Success
7097	* -EINVAL - Invalid @cpumask
7098	* -ENOMEM - Failed to allocate memory for attrs or pwqs.
7099	*/
7100	static int workqueue_set_unbound_cpumask(cpumask_var_t cpumask)
7101	{
7102	int ret = -EINVAL;
7103
7104	/*
7105	* Not excluding isolated cpus on purpose.
7106	* If the user wishes to include them, we allow that.
7107	*/
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)) {
7113	ret = 0;
7114	goto out_unlock;
7115	}
7116
7117	ret = workqueue_apply_unbound_cpumask(cpumask);
7118
7119	out_unlock:
7120	apply_wqattrs_unlock();
7121	}
7122
7123	return ret;
7124	}
7125
7126	static ssize_t __wq_cpumask_show(struct device *dev,
7127	struct device_attribute attr, char buf, cpumask_var_t mask)
7128	{
7129	int written;
7130
7131	mutex_lock(&wq_pool_mutex);
7132	written = scnprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
7133	mutex_unlock(&wq_pool_mutex);
7134
7135	return written;
7136	}
7137
7138	static ssize_t wq_unbound_cpumask_show(struct device *dev,
7139	struct device_attribute attr, char buf)
7140	{
7141	return __wq_cpumask_show(dev, attr, buf, wq_unbound_cpumask);
7142	}
7143
7144	static ssize_t wq_requested_cpumask_show(struct device *dev,
7145	struct device_attribute attr, char buf)
7146	{
7147	return __wq_cpumask_show(dev, attr, buf, wq_requested_unbound_cpumask);
7148	}
7149
7150	static ssize_t wq_isolated_cpumask_show(struct device *dev,
7151	struct device_attribute attr, char buf)
7152	{
7153	return __wq_cpumask_show(dev, attr, buf, wq_isolated_cpumask);
7154	}
7155
7156	static ssize_t wq_unbound_cpumask_store(struct device *dev,
7157	struct device_attribute attr, const char buf, size_t count)
7158	{
7159	cpumask_var_t cpumask;
7160	int ret;
7161
7162	if (!zalloc_cpumask_var(&cpumask, GFP_KERNEL))
7163	return -ENOMEM;
7164
7165	ret = cpumask_parse(buf, cpumask);
7166	if (!ret)
7167	ret = workqueue_set_unbound_cpumask(cpumask);
7168
7169	free_cpumask_var(cpumask);
7170	return ret ? ret : count;
7171	}
7172
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),
7178	__ATTR_NULL,
7179	};
7180
7181	static int __init wq_sysfs_init(void)
7182	{
7183	struct device *dev_root;
7184	int err;
7185
7186	err = subsys_virtual_register(&wq_subsys, NULL);
7187	if (err)
7188	return err;
7189
7190	dev_root = bus_get_dev_root(&wq_subsys);
7191	if (dev_root) {
7192	struct device_attribute *attr;
7193
7194	for (attr = wq_sysfs_cpumask_attrs; attr->attr.name; attr++) {
7195	err = device_create_file(dev_root, attr);
7196	if (err)
7197	break;
7198	}
7199	put_device(dev_root);
7200	}
7201	return err;
7202	}
7203	core_initcall(wq_sysfs_init);
7204
7205	static void wq_device_release(struct device *dev)
7206	{
7207	struct wq_device *wq_dev = container_of(dev, struct wq_device, dev);
7208
7209	kfree(wq_dev);
7210	}
7211
7212	/**
7213	* workqueue_sysfs_register - make a workqueue visible in sysfs
7214	* @wq: the workqueue to register
7215	*
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.
7219	*
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
7223	* attributes.
7224	*
7225	* Return: 0 on success, -errno on failure.
7226	*/
7227	int workqueue_sysfs_register(struct workqueue_struct *wq)
7228	{
7229	struct wq_device *wq_dev;
7230	int ret;
7231
7232	/*
7233	* Adjusting max_active breaks ordering guarantee. Disallow exposing
7234	* ordered workqueues.
7235	*/
7236	if (WARN_ON(wq->flags & __WQ_ORDERED))
7237	return -EINVAL;
7238
7239	wq->wq_dev = wq_dev = kzalloc(sizeof(*wq_dev), GFP_KERNEL);
7240	if (!wq_dev)
7241	return -ENOMEM;
7242
7243	wq_dev->wq = wq;
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);
7247
7248	/*
7249	* unbound_attrs are created separately. Suppress uevent until
7250	* everything is ready.
7251	*/
7252	dev_set_uevent_suppress(&wq_dev->dev, true);
7253
7254	ret = device_register(&wq_dev->dev);
7255	if (ret) {
7256	put_device(&wq_dev->dev);
7257	wq->wq_dev = NULL;
7258	return ret;
7259	}
7260
7261	if (wq->flags & WQ_UNBOUND) {
7262	struct device_attribute *attr;
7263
7264	for (attr = wq_sysfs_unbound_attrs; attr->attr.name; attr++) {
7265	ret = device_create_file(&wq_dev->dev, attr);
7266	if (ret) {
7267	device_unregister(&wq_dev->dev);
7268	wq->wq_dev = NULL;
7269	return ret;
7270	}
7271	}
7272	}
7273
7274	dev_set_uevent_suppress(&wq_dev->dev, false);
7275	kobject_uevent(&wq_dev->dev.kobj, KOBJ_ADD);
7276	return 0;
7277	}
7278
7279	/**
7280	* workqueue_sysfs_unregister - undo workqueue_sysfs_register()
7281	* @wq: the workqueue to unregister
7282	*
7283	* If @wq is registered to sysfs by workqueue_sysfs_register(), unregister.
7284	*/
7285	static void workqueue_sysfs_unregister(struct workqueue_struct *wq)
7286	{
7287	struct wq_device *wq_dev = wq->wq_dev;
7288
7289	if (!wq->wq_dev)
7290	return;
7291
7292	wq->wq_dev = NULL;
7293	device_unregister(&wq_dev->dev);
7294	}
7295	#else /* CONFIG_SYSFS */
7296	static void workqueue_sysfs_unregister(struct workqueue_struct *wq) { }
7297	#endif /* CONFIG_SYSFS */
7298
7299	/*
7300	* Workqueue watchdog.
7301	*
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
7306	* largely opaque.
7307	*
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.
7311	*
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.
7315	*/
7316	#ifdef CONFIG_WQ_WATCHDOG
7317
7318	static unsigned long wq_watchdog_thresh = 30;
7319	static struct timer_list wq_watchdog_timer;
7320
7321	static unsigned long wq_watchdog_touched = INITIAL_JIFFIES;
7322	static DEFINE_PER_CPU(unsigned long, wq_watchdog_touched_cpu) = INITIAL_JIFFIES;
7323
7324	/*
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.
7329	*/
7330	static void show_cpu_pool_hog(struct worker_pool *pool)
7331	{
7332	struct worker *worker;
7333	unsigned long irq_flags;
7334	int bkt;
7335
7336	raw_spin_lock_irqsave(&pool->lock, irq_flags);
7337
7338	hash_for_each(pool->busy_hash, bkt, worker, hentry) {
7339	if (task_is_running(worker->task)) {
7340	/*
7341	* Defer printing to avoid deadlocks in console
7342	* drivers that queue work while holding locks
7343	* also taken in their write paths.
7344	*/
7345	printk_deferred_enter();
7346
7347	pr_info("pool %d:\n", pool->id);
7348	sched_show_task(worker->task);
7349
7350	printk_deferred_exit();
7351	}
7352	}
7353
7354	raw_spin_unlock_irqrestore(&pool->lock, irq_flags);
7355	}
7356
7357	static void show_cpu_pools_hogs(void)
7358	{
7359	struct worker_pool *pool;
7360	int pi;
7361
7362	pr_info("Showing backtraces of running workers in stalled CPU-bound worker pools:\n");
7363
7364	rcu_read_lock();
7365
7366	for_each_pool(pool, pi) {
7367	if (pool->cpu_stall)
7368	show_cpu_pool_hog(pool);
7369
7370	}
7371
7372	rcu_read_unlock();
7373	}
7374
7375	static void wq_watchdog_reset_touched(void)
7376	{
7377	int cpu;
7378
7379	wq_watchdog_touched = jiffies;
7380	for_each_possible_cpu(cpu)
7381	per_cpu(wq_watchdog_touched_cpu, cpu) = jiffies;
7382	}
7383
7384	static void wq_watchdog_timer_fn(struct timer_list *unused)
7385	{
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;
7391	int pi;
7392
7393	if (!thresh)
7394	return;
7395
7396	rcu_read_lock();
7397
7398	for_each_pool(pool, pi) {
7399	unsigned long pool_ts, touched, ts;
7400
7401	pool->cpu_stall = false;
7402	if (list_empty(&pool->worklist))
7403	continue;
7404
7405	/*
7406	* If a virtual machine is stopped by the host it can look to
7407	* the watchdog like a stall.
7408	*/
7409	kvm_check_and_clear_guest_paused();
7410
7411	/* get the latest of pool and touched timestamps */
7412	if (pool->cpu >= 0)
7413	touched = READ_ONCE(per_cpu(wq_watchdog_touched_cpu, pool->cpu));
7414	else
7415	touched = READ_ONCE(wq_watchdog_touched);
7416	pool_ts = READ_ONCE(pool->watchdog_ts);
7417
7418	if (time_after(pool_ts, touched))
7419	ts = pool_ts;
7420	else
7421	ts = touched;
7422
7423	/* did we stall? */
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;
7429	}
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);
7434	}
7435
7436
7437	}
7438
7439	rcu_read_unlock();
7440
7441	if (lockup_detected)
7442	show_all_workqueues();
7443
7444	if (cpu_pool_stall)
7445	show_cpu_pools_hogs();
7446
7447	wq_watchdog_reset_touched();
7448	mod_timer(&wq_watchdog_timer, jiffies + thresh);
7449	}
7450
7451	notrace void wq_watchdog_touch(int cpu)
7452	{
7453	if (cpu >= 0)
7454	per_cpu(wq_watchdog_touched_cpu, cpu) = jiffies;
7455
7456	wq_watchdog_touched = jiffies;
7457	}
7458
7459	static void wq_watchdog_set_thresh(unsigned long thresh)
7460	{
7461	wq_watchdog_thresh = 0;
7462	del_timer_sync(&wq_watchdog_timer);
7463
7464	if (thresh) {
7465	wq_watchdog_thresh = thresh;
7466	wq_watchdog_reset_touched();
7467	mod_timer(&wq_watchdog_timer, jiffies + thresh * HZ);
7468	}
7469	}
7470
7471	static int wq_watchdog_param_set_thresh(const char *val,
7472	const struct kernel_param *kp)
7473	{
7474	unsigned long thresh;
7475	int ret;
7476
7477	ret = kstrtoul(val, 0, &thresh);
7478	if (ret)
7479	return ret;
7480
7481	if (system_wq)
7482	wq_watchdog_set_thresh(thresh);
7483	else
7484	wq_watchdog_thresh = thresh;
7485
7486	return 0;
7487	}
7488
7489	static const struct kernel_param_ops wq_watchdog_thresh_ops = {
7490	.set = wq_watchdog_param_set_thresh,
7491	.get = param_get_ulong,
7492	};
7493
7494	module_param_cb(watchdog_thresh, &wq_watchdog_thresh_ops, &wq_watchdog_thresh,
7495	0644);
7496
7497	static void wq_watchdog_init(void)
7498	{
7499	timer_setup(&wq_watchdog_timer, wq_watchdog_timer_fn, TIMER_DEFERRABLE);
7500	wq_watchdog_set_thresh(wq_watchdog_thresh);
7501	}
7502
7503	#else /* CONFIG_WQ_WATCHDOG */
7504
7505	static inline void wq_watchdog_init(void) { }
7506
7507	#endif /* CONFIG_WQ_WATCHDOG */
7508
7509	static void bh_pool_kick_normal(struct irq_work *irq_work)
7510	{
7511	raise_softirq_irqoff(TASKLET_SOFTIRQ);
7512	}
7513
7514	static void bh_pool_kick_highpri(struct irq_work *irq_work)
7515	{
7516	raise_softirq_irqoff(HI_SOFTIRQ);
7517	}
7518
7519	static void __init restrict_unbound_cpumask(const char name, const struct cpumask mask)
7520	{
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));
7524	return;
7525	}
7526
7527	cpumask_and(wq_unbound_cpumask, wq_unbound_cpumask, mask);
7528	}
7529
7530	static void __init init_cpu_worker_pool(struct worker_pool *pool, int cpu, int nice)
7531	{
7532	BUG_ON(init_worker_pool(pool));
7533	pool->cpu = cpu;
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);
7539
7540	/* alloc pool ID */
7541	mutex_lock(&wq_pool_mutex);
7542	BUG_ON(worker_pool_assign_id(pool));
7543	mutex_unlock(&wq_pool_mutex);
7544	}
7545
7546	/**
7547	* workqueue_init_early - early init for workqueue subsystem
7548	*
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.
7555	*/
7556	void __init workqueue_init_early(void)
7557	{
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 };
7562	int i, cpu;
7563
7564	BUILD_BUG_ON(__alignof__(struct pool_workqueue) < __alignof__(long long));
7565
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));
7569
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);
7575
7576	cpumask_copy(wq_requested_unbound_cpumask, wq_unbound_cpumask);
7577
7578	pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC);
7579
7580	wq_update_pod_attrs_buf = alloc_workqueue_attrs();
7581	BUG_ON(!wq_update_pod_attrs_buf);
7582
7583	/*
7584	* If nohz_full is enabled, set power efficient workqueue as unbound.
7585	* This allows workqueue items to be moved to HK CPUs.
7586	*/
7587	if (housekeeping_enabled(HK_TYPE_TICK))
7588	wq_power_efficient = true;
7589
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);
7595
7596	BUG_ON(!zalloc_cpumask_var_node(&pt->pod_cpus[0], GFP_KERNEL, NUMA_NO_NODE));
7597
7598	pt->nr_pods = 1;
7599	cpumask_copy(pt->pod_cpus[0], cpu_possible_mask);
7600	pt->pod_node[0] = NUMA_NO_NODE;
7601	pt->cpu_pod[0] = 0;
7602
7603	/* initialize BH and CPU pools */
7604	for_each_possible_cpu(cpu) {
7605	struct worker_pool *pool;
7606
7607	i = 0;
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]);
7612	i++;
7613	}
7614
7615	i = 0;
7616	for_each_cpu_worker_pool(pool, cpu)
7617	init_cpu_worker_pool(pool, cpu, std_nice[i++]);
7618	}
7619
7620	/* create default unbound and ordered wq attrs */
7621	for (i = 0; i < NR_STD_WORKER_POOLS; i++) {
7622	struct workqueue_attrs *attrs;
7623
7624	BUG_ON(!(attrs = alloc_workqueue_attrs()));
7625	attrs->nice = std_nice[i];
7626	unbound_std_wq_attrs[i] = attrs;
7627
7628	/*
7629	* An ordered wq should have only one pwq as ordering is
7630	* guaranteed by max_active which is enforced by pwqs.
7631	*/
7632	BUG_ON(!(attrs = alloc_workqueue_attrs()));
7633	attrs->nice = std_nice[i];
7634	attrs->ordered = true;
7635	ordered_wq_attrs[i] = attrs;
7636	}
7637
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,
7642	WQ_MAX_ACTIVE);
7643	system_freezable_wq = alloc_workqueue("events_freezable",
7644	WQ_FREEZABLE, 0);
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,
7649	0);
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);
7658	}
7659
7660	static void __init wq_cpu_intensive_thresh_init(void)
7661	{
7662	unsigned long thresh;
7663	unsigned long bogo;
7664
7665	pwq_release_worker = kthread_create_worker(0, "pool_workqueue_release");
7666	BUG_ON(IS_ERR(pwq_release_worker));
7667
7668	/* if the user set it to a specific value, keep it */
7669	if (wq_cpu_intensive_thresh_us != ULONG_MAX)
7670	return;
7671
7672	/*
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
7677	* too low.
7678	*
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
7684	* usefulness.
7685	*/
7686	thresh = 10 * USEC_PER_MSEC;
7687
7688	/* see init/calibrate.c for lpj -> BogoMIPS calculation */
7689	bogo = max_t(unsigned long, loops_per_jiffy / 500000 * HZ, 1);
7690	if (bogo < 4000)
7691	thresh = min_t(unsigned long, thresh * 4000 / bogo, USEC_PER_SEC);
7692
7693	pr_debug("wq_cpu_intensive_thresh: lpj=%lu BogoMIPS=%lu thresh_us=%lu\n",
7694	loops_per_jiffy, bogo, thresh);
7695
7696	wq_cpu_intensive_thresh_us = thresh;
7697	}
7698
7699	/**
7700	* workqueue_init - bring workqueue subsystem fully online
7701	*
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.
7707	*/
7708	void __init workqueue_init(void)
7709	{
7710	struct workqueue_struct *wq;
7711	struct worker_pool *pool;
7712	int cpu, bkt;
7713
7714	wq_cpu_intensive_thresh_init();
7715
7716	mutex_lock(&wq_pool_mutex);
7717
7718	/*
7719	* Per-cpu pools created earlier could be missing node hint. Fix them
7720	* up. Also, create a rescuer for workqueues that requested it.
7721	*/
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);
7727	}
7728
7729	list_for_each_entry(wq, &workqueues, list) {
7730	WARN(init_rescuer(wq),
7731	"workqueue: failed to create early rescuer for %s",
7732	wq->name);
7733	}
7734
7735	mutex_unlock(&wq_pool_mutex);
7736
7737	/*
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.
7742	*/
7743	for_each_possible_cpu(cpu)
7744	for_each_bh_worker_pool(pool, cpu)
7745	BUG_ON(!create_worker(pool));
7746
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));
7751	}
7752	}
7753
7754	hash_for_each(unbound_pool_hash, bkt, pool, hash_node)
7755	BUG_ON(!create_worker(pool));
7756
7757	wq_online = true;
7758	wq_watchdog_init();
7759	}
7760
7761	/*
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.
7765	*/
7766	static void __init init_pod_type(struct wq_pod_type *pt,
7767	bool (*cpus_share_pod)(int, int))
7768	{
7769	int cur, pre, cpu, pod;
7770
7771	pt->nr_pods = 0;
7772
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);
7776
7777	for_each_possible_cpu(cur) {
7778	for_each_possible_cpu(pre) {
7779	if (pre >= cur) {
7780	pt->cpu_pod[cur] = pt->nr_pods++;
7781	break;
7782	}
7783	if (cpus_share_pod(cur, pre)) {
7784	pt->cpu_pod[cur] = pt->cpu_pod[pre];
7785	break;
7786	}
7787	}
7788	}
7789
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);
7794
7795	for (pod = 0; pod < pt->nr_pods; pod++)
7796	BUG_ON(!zalloc_cpumask_var(&pt->pod_cpus[pod], GFP_KERNEL));
7797
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);
7801	}
7802	}
7803
7804	static bool __init cpus_dont_share(int cpu0, int cpu1)
7805	{
7806	return false;
7807	}
7808
7809	static bool __init cpus_share_smt(int cpu0, int cpu1)
7810	{
7811	#ifdef CONFIG_SCHED_SMT
7812	return cpumask_test_cpu(cpu0, cpu_smt_mask(cpu1));
7813	#else
7814	return false;
7815	#endif
7816	}
7817
7818	static bool __init cpus_share_numa(int cpu0, int cpu1)
7819	{
7820	return cpu_to_node(cpu0) == cpu_to_node(cpu1);
7821	}
7822
7823	/**
7824	* workqueue_init_topology - initialize CPU pods for unbound workqueues
7825	*
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.
7829	*/
7830	void __init workqueue_init_topology(void)
7831	{
7832	struct workqueue_struct *wq;
7833	int cpu;
7834
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);
7839
7840	wq_topo_initialized = true;
7841
7842	mutex_lock(&wq_pool_mutex);
7843
7844	/*
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.
7848	*/
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);
7856	}
7857	}
7858
7859	mutex_unlock(&wq_pool_mutex);
7860	}
7861
7862	void __warn_flushing_systemwide_wq(void)
7863	{
7864	pr_warn("WARNING: Flushing system-wide workqueues will be prohibited in near future.\n");
7865	dump_stack();
7866	}
7867	EXPORT_SYMBOL(__warn_flushing_systemwide_wq);
7868
7869	static int __init workqueue_unbound_cpus_setup(char *str)
7870	{
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");
7874	}
7875
7876	return 1;
7877	}
7878	__setup("workqueue.unbound_cpus=", workqueue_unbound_cpus_setup);