Branch data Line data Source code
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 <dwmw2@infradead.org>
9 : : * Andrew Morton
10 : : * Kai Petzke <wpp@marie.physik.tu-berlin.de>
11 : : * Theodore Ts'o <tytso@mit.edu>
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 <tj@kernel.org>
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/signal.h>
33 : : #include <linux/completion.h>
34 : : #include <linux/workqueue.h>
35 : : #include <linux/slab.h>
36 : : #include <linux/cpu.h>
37 : : #include <linux/notifier.h>
38 : : #include <linux/kthread.h>
39 : : #include <linux/hardirq.h>
40 : : #include <linux/mempolicy.h>
41 : : #include <linux/freezer.h>
42 : : #include <linux/debug_locks.h>
43 : : #include <linux/lockdep.h>
44 : : #include <linux/idr.h>
45 : : #include <linux/jhash.h>
46 : : #include <linux/hashtable.h>
47 : : #include <linux/rculist.h>
48 : : #include <linux/nodemask.h>
49 : : #include <linux/moduleparam.h>
50 : : #include <linux/uaccess.h>
51 : : #include <linux/sched/isolation.h>
52 : : #include <linux/nmi.h>
53 : :
54 : : #include "workqueue_internal.h"
55 : :
56 : : enum {
57 : : /*
58 : : * worker_pool flags
59 : : *
60 : : * A bound pool is either associated or disassociated with its CPU.
61 : : * While associated (!DISASSOCIATED), all workers are bound to the
62 : : * CPU and none has %WORKER_UNBOUND set and concurrency management
63 : : * is in effect.
64 : : *
65 : : * While DISASSOCIATED, the cpu may be offline and all workers have
66 : : * %WORKER_UNBOUND set and concurrency management disabled, and may
67 : : * be executing on any CPU. The pool behaves as an unbound one.
68 : : *
69 : : * Note that DISASSOCIATED should be flipped only while holding
70 : : * wq_pool_attach_mutex to avoid changing binding state while
71 : : * worker_attach_to_pool() is in progress.
72 : : */
73 : : POOL_MANAGER_ACTIVE = 1 << 0, /* being managed */
74 : : POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */
75 : :
76 : : /* worker flags */
77 : : WORKER_DIE = 1 << 1, /* die die die */
78 : : WORKER_IDLE = 1 << 2, /* is idle */
79 : : WORKER_PREP = 1 << 3, /* preparing to run works */
80 : : WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */
81 : : WORKER_UNBOUND = 1 << 7, /* worker is unbound */
82 : : WORKER_REBOUND = 1 << 8, /* worker was rebound */
83 : :
84 : : WORKER_NOT_RUNNING = WORKER_PREP | WORKER_CPU_INTENSIVE |
85 : : WORKER_UNBOUND | WORKER_REBOUND,
86 : :
87 : : NR_STD_WORKER_POOLS = 2, /* # standard pools per cpu */
88 : :
89 : : UNBOUND_POOL_HASH_ORDER = 6, /* hashed by pool->attrs */
90 : : BUSY_WORKER_HASH_ORDER = 6, /* 64 pointers */
91 : :
92 : : MAX_IDLE_WORKERS_RATIO = 4, /* 1/4 of busy can be idle */
93 : : IDLE_WORKER_TIMEOUT = 300 * HZ, /* keep idle ones for 5 mins */
94 : :
95 : : MAYDAY_INITIAL_TIMEOUT = HZ / 100 >= 2 ? HZ / 100 : 2,
96 : : /* call for help after 10ms
97 : : (min two ticks) */
98 : : MAYDAY_INTERVAL = HZ / 10, /* and then every 100ms */
99 : : CREATE_COOLDOWN = HZ, /* time to breath after fail */
100 : :
101 : : /*
102 : : * Rescue workers are used only on emergencies and shared by
103 : : * all cpus. Give MIN_NICE.
104 : : */
105 : : RESCUER_NICE_LEVEL = MIN_NICE,
106 : : HIGHPRI_NICE_LEVEL = MIN_NICE,
107 : :
108 : : WQ_NAME_LEN = 24,
109 : : };
110 : :
111 : : /*
112 : : * Structure fields follow one of the following exclusion rules.
113 : : *
114 : : * I: Modifiable by initialization/destruction paths and read-only for
115 : : * everyone else.
116 : : *
117 : : * P: Preemption protected. Disabling preemption is enough and should
118 : : * only be modified and accessed from the local cpu.
119 : : *
120 : : * L: pool->lock protected. Access with pool->lock held.
121 : : *
122 : : * X: During normal operation, modification requires pool->lock and should
123 : : * be done only from local cpu. Either disabling preemption on local
124 : : * cpu or grabbing pool->lock is enough for read access. If
125 : : * POOL_DISASSOCIATED is set, it's identical to L.
126 : : *
127 : : * A: wq_pool_attach_mutex protected.
128 : : *
129 : : * PL: wq_pool_mutex protected.
130 : : *
131 : : * PR: wq_pool_mutex protected for writes. RCU protected for reads.
132 : : *
133 : : * PW: wq_pool_mutex and wq->mutex protected for writes. Either for reads.
134 : : *
135 : : * PWR: wq_pool_mutex and wq->mutex protected for writes. Either or
136 : : * RCU for reads.
137 : : *
138 : : * WQ: wq->mutex protected.
139 : : *
140 : : * WR: wq->mutex protected for writes. RCU protected for reads.
141 : : *
142 : : * MD: wq_mayday_lock protected.
143 : : */
144 : :
145 : : /* struct worker is defined in workqueue_internal.h */
146 : :
147 : : struct worker_pool {
148 : : spinlock_t lock; /* the pool lock */
149 : : int cpu; /* I: the associated cpu */
150 : : int node; /* I: the associated node ID */
151 : : int id; /* I: pool ID */
152 : : unsigned int flags; /* X: flags */
153 : :
154 : : unsigned long watchdog_ts; /* L: watchdog timestamp */
155 : :
156 : : struct list_head worklist; /* L: list of pending works */
157 : :
158 : : int nr_workers; /* L: total number of workers */
159 : : int nr_idle; /* L: currently idle workers */
160 : :
161 : : struct list_head idle_list; /* X: list of idle workers */
162 : : struct timer_list idle_timer; /* L: worker idle timeout */
163 : : struct timer_list mayday_timer; /* L: SOS timer for workers */
164 : :
165 : : /* a workers is either on busy_hash or idle_list, or the manager */
166 : : DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER);
167 : : /* L: hash of busy workers */
168 : :
169 : : struct worker *manager; /* L: purely informational */
170 : : struct list_head workers; /* A: attached workers */
171 : : struct completion *detach_completion; /* all workers detached */
172 : :
173 : : struct ida worker_ida; /* worker IDs for task name */
174 : :
175 : : struct workqueue_attrs *attrs; /* I: worker attributes */
176 : : struct hlist_node hash_node; /* PL: unbound_pool_hash node */
177 : : int refcnt; /* PL: refcnt for unbound pools */
178 : :
179 : : /*
180 : : * The current concurrency level. As it's likely to be accessed
181 : : * from other CPUs during try_to_wake_up(), put it in a separate
182 : : * cacheline.
183 : : */
184 : : atomic_t nr_running ____cacheline_aligned_in_smp;
185 : :
186 : : /*
187 : : * Destruction of pool is RCU protected to allow dereferences
188 : : * from get_work_pool().
189 : : */
190 : : struct rcu_head rcu;
191 : : } ____cacheline_aligned_in_smp;
192 : :
193 : : /*
194 : : * The per-pool workqueue. While queued, the lower WORK_STRUCT_FLAG_BITS
195 : : * of work_struct->data are used for flags and the remaining high bits
196 : : * point to the pwq; thus, pwqs need to be aligned at two's power of the
197 : : * number of flag bits.
198 : : */
199 : : struct pool_workqueue {
200 : : struct worker_pool *pool; /* I: the associated pool */
201 : : struct workqueue_struct *wq; /* I: the owning workqueue */
202 : : int work_color; /* L: current color */
203 : : int flush_color; /* L: flushing color */
204 : : int refcnt; /* L: reference count */
205 : : int nr_in_flight[WORK_NR_COLORS];
206 : : /* L: nr of in_flight works */
207 : : int nr_active; /* L: nr of active works */
208 : : int max_active; /* L: max active works */
209 : : struct list_head delayed_works; /* L: delayed works */
210 : : struct list_head pwqs_node; /* WR: node on wq->pwqs */
211 : : struct list_head mayday_node; /* MD: node on wq->maydays */
212 : :
213 : : /*
214 : : * Release of unbound pwq is punted to system_wq. See put_pwq()
215 : : * and pwq_unbound_release_workfn() for details. pool_workqueue
216 : : * itself is also RCU protected so that the first pwq can be
217 : : * determined without grabbing wq->mutex.
218 : : */
219 : : struct work_struct unbound_release_work;
220 : : struct rcu_head rcu;
221 : : } __aligned(1 << WORK_STRUCT_FLAG_BITS);
222 : :
223 : : /*
224 : : * Structure used to wait for workqueue flush.
225 : : */
226 : : struct wq_flusher {
227 : : struct list_head list; /* WQ: list of flushers */
228 : : int flush_color; /* WQ: flush color waiting for */
229 : : struct completion done; /* flush completion */
230 : : };
231 : :
232 : : struct wq_device;
233 : :
234 : : /*
235 : : * The externally visible workqueue. It relays the issued work items to
236 : : * the appropriate worker_pool through its pool_workqueues.
237 : : */
238 : : struct workqueue_struct {
239 : : struct list_head pwqs; /* WR: all pwqs of this wq */
240 : : struct list_head list; /* PR: list of all workqueues */
241 : :
242 : : struct mutex mutex; /* protects this wq */
243 : : int work_color; /* WQ: current work color */
244 : : int flush_color; /* WQ: current flush color */
245 : : atomic_t nr_pwqs_to_flush; /* flush in progress */
246 : : struct wq_flusher *first_flusher; /* WQ: first flusher */
247 : : struct list_head flusher_queue; /* WQ: flush waiters */
248 : : struct list_head flusher_overflow; /* WQ: flush overflow list */
249 : :
250 : : struct list_head maydays; /* MD: pwqs requesting rescue */
251 : : struct worker *rescuer; /* MD: rescue worker */
252 : :
253 : : int nr_drainers; /* WQ: drain in progress */
254 : : int saved_max_active; /* WQ: saved pwq max_active */
255 : :
256 : : struct workqueue_attrs *unbound_attrs; /* PW: only for unbound wqs */
257 : : struct pool_workqueue *dfl_pwq; /* PW: only for unbound wqs */
258 : :
259 : : #ifdef CONFIG_SYSFS
260 : : struct wq_device *wq_dev; /* I: for sysfs interface */
261 : : #endif
262 : : #ifdef CONFIG_LOCKDEP
263 : : char *lock_name;
264 : : struct lock_class_key key;
265 : : struct lockdep_map lockdep_map;
266 : : #endif
267 : : char name[WQ_NAME_LEN]; /* I: workqueue name */
268 : :
269 : : /*
270 : : * Destruction of workqueue_struct is RCU protected to allow walking
271 : : * the workqueues list without grabbing wq_pool_mutex.
272 : : * This is used to dump all workqueues from sysrq.
273 : : */
274 : : struct rcu_head rcu;
275 : :
276 : : /* hot fields used during command issue, aligned to cacheline */
277 : : unsigned int flags ____cacheline_aligned; /* WQ: WQ_* flags */
278 : : struct pool_workqueue __percpu *cpu_pwqs; /* I: per-cpu pwqs */
279 : : struct pool_workqueue __rcu *numa_pwq_tbl[]; /* PWR: unbound pwqs indexed by node */
280 : : };
281 : :
282 : : static struct kmem_cache *pwq_cache;
283 : :
284 : : static cpumask_var_t *wq_numa_possible_cpumask;
285 : : /* possible CPUs of each node */
286 : :
287 : : static bool wq_disable_numa;
288 : : module_param_named(disable_numa, wq_disable_numa, bool, 0444);
289 : :
290 : : /* see the comment above the definition of WQ_POWER_EFFICIENT */
291 : : static bool wq_power_efficient = IS_ENABLED(CONFIG_WQ_POWER_EFFICIENT_DEFAULT);
292 : : module_param_named(power_efficient, wq_power_efficient, bool, 0444);
293 : :
294 : : static bool wq_online; /* can kworkers be created yet? */
295 : :
296 : : static bool wq_numa_enabled; /* unbound NUMA affinity enabled */
297 : :
298 : : /* buf for wq_update_unbound_numa_attrs(), protected by CPU hotplug exclusion */
299 : : static struct workqueue_attrs *wq_update_unbound_numa_attrs_buf;
300 : :
301 : : static DEFINE_MUTEX(wq_pool_mutex); /* protects pools and workqueues list */
302 : : static DEFINE_MUTEX(wq_pool_attach_mutex); /* protects worker attach/detach */
303 : : static DEFINE_SPINLOCK(wq_mayday_lock); /* protects wq->maydays list */
304 : : static DECLARE_WAIT_QUEUE_HEAD(wq_manager_wait); /* wait for manager to go away */
305 : :
306 : : static LIST_HEAD(workqueues); /* PR: list of all workqueues */
307 : : static bool workqueue_freezing; /* PL: have wqs started freezing? */
308 : :
309 : : /* PL: allowable cpus for unbound wqs and work items */
310 : : static cpumask_var_t wq_unbound_cpumask;
311 : :
312 : : /* CPU where unbound work was last round robin scheduled from this CPU */
313 : : static DEFINE_PER_CPU(int, wq_rr_cpu_last);
314 : :
315 : : /*
316 : : * Local execution of unbound work items is no longer guaranteed. The
317 : : * following always forces round-robin CPU selection on unbound work items
318 : : * to uncover usages which depend on it.
319 : : */
320 : : #ifdef CONFIG_DEBUG_WQ_FORCE_RR_CPU
321 : : static bool wq_debug_force_rr_cpu = true;
322 : : #else
323 : : static bool wq_debug_force_rr_cpu = false;
324 : : #endif
325 : : module_param_named(debug_force_rr_cpu, wq_debug_force_rr_cpu, bool, 0644);
326 : :
327 : : /* the per-cpu worker pools */
328 : : static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS], cpu_worker_pools);
329 : :
330 : : static DEFINE_IDR(worker_pool_idr); /* PR: idr of all pools */
331 : :
332 : : /* PL: hash of all unbound pools keyed by pool->attrs */
333 : : static DEFINE_HASHTABLE(unbound_pool_hash, UNBOUND_POOL_HASH_ORDER);
334 : :
335 : : /* I: attributes used when instantiating standard unbound pools on demand */
336 : : static struct workqueue_attrs *unbound_std_wq_attrs[NR_STD_WORKER_POOLS];
337 : :
338 : : /* I: attributes used when instantiating ordered pools on demand */
339 : : static struct workqueue_attrs *ordered_wq_attrs[NR_STD_WORKER_POOLS];
340 : :
341 : : struct workqueue_struct *system_wq __read_mostly;
342 : : EXPORT_SYMBOL(system_wq);
343 : : struct workqueue_struct *system_highpri_wq __read_mostly;
344 : : EXPORT_SYMBOL_GPL(system_highpri_wq);
345 : : struct workqueue_struct *system_long_wq __read_mostly;
346 : : EXPORT_SYMBOL_GPL(system_long_wq);
347 : : struct workqueue_struct *system_unbound_wq __read_mostly;
348 : : EXPORT_SYMBOL_GPL(system_unbound_wq);
349 : : struct workqueue_struct *system_freezable_wq __read_mostly;
350 : : EXPORT_SYMBOL_GPL(system_freezable_wq);
351 : : struct workqueue_struct *system_power_efficient_wq __read_mostly;
352 : : EXPORT_SYMBOL_GPL(system_power_efficient_wq);
353 : : struct workqueue_struct *system_freezable_power_efficient_wq __read_mostly;
354 : : EXPORT_SYMBOL_GPL(system_freezable_power_efficient_wq);
355 : :
356 : : static int worker_thread(void *__worker);
357 : : static void workqueue_sysfs_unregister(struct workqueue_struct *wq);
358 : : static void show_pwq(struct pool_workqueue *pwq);
359 : :
360 : : #define CREATE_TRACE_POINTS
361 : : #include <trace/events/workqueue.h>
362 : :
363 : : #define assert_rcu_or_pool_mutex() \
364 : : RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
365 : : !lockdep_is_held(&wq_pool_mutex), \
366 : : "RCU or wq_pool_mutex should be held")
367 : :
368 : : #define assert_rcu_or_wq_mutex_or_pool_mutex(wq) \
369 : : RCU_LOCKDEP_WARN(!rcu_read_lock_held() && \
370 : : !lockdep_is_held(&wq->mutex) && \
371 : : !lockdep_is_held(&wq_pool_mutex), \
372 : : "RCU, wq->mutex or wq_pool_mutex should be held")
373 : :
374 : : #define for_each_cpu_worker_pool(pool, cpu) \
375 : : for ((pool) = &per_cpu(cpu_worker_pools, cpu)[0]; \
376 : : (pool) < &per_cpu(cpu_worker_pools, cpu)[NR_STD_WORKER_POOLS]; \
377 : : (pool)++)
378 : :
379 : : /**
380 : : * for_each_pool - iterate through all worker_pools in the system
381 : : * @pool: iteration cursor
382 : : * @pi: integer used for iteration
383 : : *
384 : : * This must be called either with wq_pool_mutex held or RCU read
385 : : * locked. If the pool needs to be used beyond the locking in effect, the
386 : : * caller is responsible for guaranteeing that the pool stays online.
387 : : *
388 : : * The if/else clause exists only for the lockdep assertion and can be
389 : : * ignored.
390 : : */
391 : : #define for_each_pool(pool, pi) \
392 : : idr_for_each_entry(&worker_pool_idr, pool, pi) \
393 : : if (({ assert_rcu_or_pool_mutex(); false; })) { } \
394 : : else
395 : :
396 : : /**
397 : : * for_each_pool_worker - iterate through all workers of a worker_pool
398 : : * @worker: iteration cursor
399 : : * @pool: worker_pool to iterate workers of
400 : : *
401 : : * This must be called with wq_pool_attach_mutex.
402 : : *
403 : : * The if/else clause exists only for the lockdep assertion and can be
404 : : * ignored.
405 : : */
406 : : #define for_each_pool_worker(worker, pool) \
407 : : list_for_each_entry((worker), &(pool)->workers, node) \
408 : : if (({ lockdep_assert_held(&wq_pool_attach_mutex); false; })) { } \
409 : : else
410 : :
411 : : /**
412 : : * for_each_pwq - iterate through all pool_workqueues of the specified workqueue
413 : : * @pwq: iteration cursor
414 : : * @wq: the target workqueue
415 : : *
416 : : * This must be called either with wq->mutex held or RCU read locked.
417 : : * If the pwq needs to be used beyond the locking in effect, the caller is
418 : : * responsible for guaranteeing that the pwq stays online.
419 : : *
420 : : * The if/else clause exists only for the lockdep assertion and can be
421 : : * ignored.
422 : : */
423 : : #define for_each_pwq(pwq, wq) \
424 : : list_for_each_entry_rcu((pwq), &(wq)->pwqs, pwqs_node, \
425 : : lockdep_is_held(&(wq->mutex)))
426 : :
427 : : #ifdef CONFIG_DEBUG_OBJECTS_WORK
428 : :
429 : : static struct debug_obj_descr work_debug_descr;
430 : :
431 : : static void *work_debug_hint(void *addr)
432 : : {
433 : : return ((struct work_struct *) addr)->func;
434 : : }
435 : :
436 : : static bool work_is_static_object(void *addr)
437 : : {
438 : : struct work_struct *work = addr;
439 : :
440 : : return test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work));
441 : : }
442 : :
443 : : /*
444 : : * fixup_init is called when:
445 : : * - an active object is initialized
446 : : */
447 : : static bool work_fixup_init(void *addr, enum debug_obj_state state)
448 : : {
449 : : struct work_struct *work = addr;
450 : :
451 : : switch (state) {
452 : : case ODEBUG_STATE_ACTIVE:
453 : : cancel_work_sync(work);
454 : : debug_object_init(work, &work_debug_descr);
455 : : return true;
456 : : default:
457 : : return false;
458 : : }
459 : : }
460 : :
461 : : /*
462 : : * fixup_free is called when:
463 : : * - an active object is freed
464 : : */
465 : : static bool work_fixup_free(void *addr, enum debug_obj_state state)
466 : : {
467 : : struct work_struct *work = addr;
468 : :
469 : : switch (state) {
470 : : case ODEBUG_STATE_ACTIVE:
471 : : cancel_work_sync(work);
472 : : debug_object_free(work, &work_debug_descr);
473 : : return true;
474 : : default:
475 : : return false;
476 : : }
477 : : }
478 : :
479 : : static struct debug_obj_descr work_debug_descr = {
480 : : .name = "work_struct",
481 : : .debug_hint = work_debug_hint,
482 : : .is_static_object = work_is_static_object,
483 : : .fixup_init = work_fixup_init,
484 : : .fixup_free = work_fixup_free,
485 : : };
486 : :
487 : : static inline void debug_work_activate(struct work_struct *work)
488 : : {
489 : : debug_object_activate(work, &work_debug_descr);
490 : : }
491 : :
492 : : static inline void debug_work_deactivate(struct work_struct *work)
493 : : {
494 : : debug_object_deactivate(work, &work_debug_descr);
495 : : }
496 : :
497 : : void __init_work(struct work_struct *work, int onstack)
498 : : {
499 : : if (onstack)
500 : : debug_object_init_on_stack(work, &work_debug_descr);
501 : : else
502 : : debug_object_init(work, &work_debug_descr);
503 : : }
504 : : EXPORT_SYMBOL_GPL(__init_work);
505 : :
506 : : void destroy_work_on_stack(struct work_struct *work)
507 : : {
508 : : debug_object_free(work, &work_debug_descr);
509 : : }
510 : : EXPORT_SYMBOL_GPL(destroy_work_on_stack);
511 : :
512 : : void destroy_delayed_work_on_stack(struct delayed_work *work)
513 : : {
514 : : destroy_timer_on_stack(&work->timer);
515 : : debug_object_free(&work->work, &work_debug_descr);
516 : : }
517 : : EXPORT_SYMBOL_GPL(destroy_delayed_work_on_stack);
518 : :
519 : : #else
520 : 9274 : static inline void debug_work_activate(struct work_struct *work) { }
521 : 9274 : static inline void debug_work_deactivate(struct work_struct *work) { }
522 : : #endif
523 : :
524 : : /**
525 : : * worker_pool_assign_id - allocate ID and assing it to @pool
526 : : * @pool: the pool pointer of interest
527 : : *
528 : : * Returns 0 if ID in [0, WORK_OFFQ_POOL_NONE) is allocated and assigned
529 : : * successfully, -errno on failure.
530 : : */
531 : 44 : static int worker_pool_assign_id(struct worker_pool *pool)
532 : : {
533 : 44 : int ret;
534 : :
535 : 44 : lockdep_assert_held(&wq_pool_mutex);
536 : :
537 : 44 : ret = idr_alloc(&worker_pool_idr, pool, 0, WORK_OFFQ_POOL_NONE,
538 : : GFP_KERNEL);
539 [ + - + - ]: 44 : if (ret >= 0) {
540 : 44 : pool->id = ret;
541 : 22 : return 0;
542 : : }
543 : : return ret;
544 : : }
545 : :
546 : : /**
547 : : * unbound_pwq_by_node - return the unbound pool_workqueue for the given node
548 : : * @wq: the target workqueue
549 : : * @node: the node ID
550 : : *
551 : : * This must be called with any of wq_pool_mutex, wq->mutex or RCU
552 : : * read locked.
553 : : * If the pwq needs to be used beyond the locking in effect, the caller is
554 : : * responsible for guaranteeing that the pwq stays online.
555 : : *
556 : : * Return: The unbound pool_workqueue for @node.
557 : : */
558 : 941 : static struct pool_workqueue *unbound_pwq_by_node(struct workqueue_struct *wq,
559 : : int node)
560 : : {
561 : 941 : assert_rcu_or_wq_mutex_or_pool_mutex(wq);
562 : :
563 : : /*
564 : : * XXX: @node can be NUMA_NO_NODE if CPU goes offline while a
565 : : * delayed item is pending. The plan is to keep CPU -> NODE
566 : : * mapping valid and stable across CPU on/offlines. Once that
567 : : * happens, this workaround can be removed.
568 : : */
569 [ - - - + ]: 941 : if (unlikely(node == NUMA_NO_NODE))
570 : 0 : return wq->dfl_pwq;
571 : :
572 : 941 : return rcu_dereference_raw(wq->numa_pwq_tbl[node]);
573 : : }
574 : :
575 : 9274 : static unsigned int work_color_to_flags(int color)
576 : : {
577 : 9274 : return color << WORK_STRUCT_COLOR_SHIFT;
578 : : }
579 : :
580 : 9274 : static int get_work_color(struct work_struct *work)
581 : : {
582 : 9274 : return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) &
583 : : ((1 << WORK_STRUCT_COLOR_BITS) - 1);
584 : : }
585 : :
586 : 3784 : static int work_next_color(int color)
587 : : {
588 : 3784 : return (color + 1) % WORK_NR_COLORS;
589 : : }
590 : :
591 : : /*
592 : : * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data
593 : : * contain the pointer to the queued pwq. Once execution starts, the flag
594 : : * is cleared and the high bits contain OFFQ flags and pool ID.
595 : : *
596 : : * set_work_pwq(), set_work_pool_and_clear_pending(), mark_work_canceling()
597 : : * and clear_work_data() can be used to set the pwq, pool or clear
598 : : * work->data. These functions should only be called while the work is
599 : : * owned - ie. while the PENDING bit is set.
600 : : *
601 : : * get_work_pool() and get_work_pwq() can be used to obtain the pool or pwq
602 : : * corresponding to a work. Pool is available once the work has been
603 : : * queued anywhere after initialization until it is sync canceled. pwq is
604 : : * available only while the work item is queued.
605 : : *
606 : : * %WORK_OFFQ_CANCELING is used to mark a work item which is being
607 : : * canceled. While being canceled, a work item may have its PENDING set
608 : : * but stay off timer and worklist for arbitrarily long and nobody should
609 : : * try to steal the PENDING bit.
610 : : */
611 : 22002 : static inline void set_work_data(struct work_struct *work, unsigned long data,
612 : : unsigned long flags)
613 : : {
614 [ - + ]: 22002 : WARN_ON_ONCE(!work_pending(work));
615 : 22002 : atomic_long_set(&work->data, data | flags | work_static(work));
616 : 22002 : }
617 : :
618 : 9274 : static void set_work_pwq(struct work_struct *work, struct pool_workqueue *pwq,
619 : : unsigned long extra_flags)
620 : : {
621 : 9274 : set_work_data(work, (unsigned long)pwq,
622 : : WORK_STRUCT_PENDING | WORK_STRUCT_PWQ | extra_flags);
623 : : }
624 : :
625 : 59 : static void set_work_pool_and_keep_pending(struct work_struct *work,
626 : : int pool_id)
627 : : {
628 : 59 : set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT,
629 : : WORK_STRUCT_PENDING);
630 : : }
631 : :
632 : 9259 : static void set_work_pool_and_clear_pending(struct work_struct *work,
633 : : int pool_id)
634 : : {
635 : : /*
636 : : * The following wmb is paired with the implied mb in
637 : : * test_and_set_bit(PENDING) and ensures all updates to @work made
638 : : * here are visible to and precede any updates by the next PENDING
639 : : * owner.
640 : : */
641 : 44 : smp_wmb();
642 : 9259 : set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0);
643 : : /*
644 : : * The following mb guarantees that previous clear of a PENDING bit
645 : : * will not be reordered with any speculative LOADS or STORES from
646 : : * work->current_func, which is executed afterwards. This possible
647 : : * reordering can lead to a missed execution on attempt to queue
648 : : * the same @work. E.g. consider this case:
649 : : *
650 : : * CPU#0 CPU#1
651 : : * ---------------------------- --------------------------------
652 : : *
653 : : * 1 STORE event_indicated
654 : : * 2 queue_work_on() {
655 : : * 3 test_and_set_bit(PENDING)
656 : : * 4 } set_..._and_clear_pending() {
657 : : * 5 set_work_data() # clear bit
658 : : * 6 smp_mb()
659 : : * 7 work->current_func() {
660 : : * 8 LOAD event_indicated
661 : : * }
662 : : *
663 : : * Without an explicit full barrier speculative LOAD on line 8 can
664 : : * be executed before CPU#0 does STORE on line 1. If that happens,
665 : : * CPU#0 observes the PENDING bit is still set and new execution of
666 : : * a @work is not queued in a hope, that CPU#1 will eventually
667 : : * finish the queued @work. Meanwhile CPU#1 does not see
668 : : * event_indicated is set, because speculative LOAD was executed
669 : : * before actual STORE.
670 : : */
671 : 9259 : smp_mb();
672 : : }
673 : :
674 : 1705 : static void clear_work_data(struct work_struct *work)
675 : : {
676 : 1705 : smp_wmb(); /* see set_work_pool_and_clear_pending() */
677 : 1705 : set_work_data(work, WORK_STRUCT_NO_POOL, 0);
678 : : }
679 : :
680 : 10013 : static struct pool_workqueue *get_work_pwq(struct work_struct *work)
681 : : {
682 : 10013 : unsigned long data = atomic_long_read(&work->data);
683 : :
684 [ - - - - : 10013 : if (data & WORK_STRUCT_PWQ)
+ + - - +
- - - + -
+ - ]
685 : 9525 : return (void *)(data & WORK_STRUCT_WQ_DATA_MASK);
686 : : else
687 : : return NULL;
688 : : }
689 : :
690 : : /**
691 : : * get_work_pool - return the worker_pool a given work was associated with
692 : : * @work: the work item of interest
693 : : *
694 : : * Pools are created and destroyed under wq_pool_mutex, and allows read
695 : : * access under RCU read lock. As such, this function should be
696 : : * called under wq_pool_mutex or inside of a rcu_read_lock() region.
697 : : *
698 : : * All fields of the returned pool are accessible as long as the above
699 : : * mentioned locking is in effect. If the returned pool needs to be used
700 : : * beyond the critical section, the caller is responsible for ensuring the
701 : : * returned pool is and stays online.
702 : : *
703 : : * Return: The worker_pool @work was last associated with. %NULL if none.
704 : : */
705 : 13046 : static struct worker_pool *get_work_pool(struct work_struct *work)
706 : : {
707 : 13046 : unsigned long data = atomic_long_read(&work->data);
708 : 13046 : int pool_id;
709 : :
710 : 13046 : assert_rcu_or_pool_mutex();
711 : :
712 [ + + ]: 13046 : if (data & WORK_STRUCT_PWQ)
713 : 84 : return ((struct pool_workqueue *)
714 : 84 : (data & WORK_STRUCT_WQ_DATA_MASK))->pool;
715 : :
716 : 12962 : pool_id = data >> WORK_OFFQ_POOL_SHIFT;
717 [ + + ]: 12962 : if (pool_id == WORK_OFFQ_POOL_NONE)
718 : : return NULL;
719 : :
720 : 6382 : return idr_find(&worker_pool_idr, pool_id);
721 : : }
722 : :
723 : : /**
724 : : * get_work_pool_id - return the worker pool ID a given work is associated with
725 : : * @work: the work item of interest
726 : : *
727 : : * Return: The worker_pool ID @work was last associated with.
728 : : * %WORK_OFFQ_POOL_NONE if none.
729 : : */
730 : 1749 : static int get_work_pool_id(struct work_struct *work)
731 : : {
732 : 1749 : unsigned long data = atomic_long_read(&work->data);
733 : :
734 [ - + ]: 1749 : if (data & WORK_STRUCT_PWQ)
735 : 0 : return ((struct pool_workqueue *)
736 : 0 : (data & WORK_STRUCT_WQ_DATA_MASK))->pool->id;
737 : :
738 : 1749 : return data >> WORK_OFFQ_POOL_SHIFT;
739 : : }
740 : :
741 : 1705 : static void mark_work_canceling(struct work_struct *work)
742 : : {
743 : 1705 : unsigned long pool_id = get_work_pool_id(work);
744 : :
745 : 1705 : pool_id <<= WORK_OFFQ_POOL_SHIFT;
746 : 1705 : set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING);
747 : 1705 : }
748 : :
749 : 0 : static bool work_is_canceling(struct work_struct *work)
750 : : {
751 : 0 : unsigned long data = atomic_long_read(&work->data);
752 : :
753 : 0 : return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING);
754 : : }
755 : :
756 : : /*
757 : : * Policy functions. These define the policies on how the global worker
758 : : * pools are managed. Unless noted otherwise, these functions assume that
759 : : * they're being called with pool->lock held.
760 : : */
761 : :
762 : 17829 : static bool __need_more_worker(struct worker_pool *pool)
763 : : {
764 : 8555 : return !atomic_read(&pool->nr_running);
765 : : }
766 : :
767 : : /*
768 : : * Need to wake up a worker? Called from anything but currently
769 : : * running workers.
770 : : *
771 : : * Note that, because unbound workers never contribute to nr_running, this
772 : : * function will always return %true for unbound pools as long as the
773 : : * worklist isn't empty.
774 : : */
775 : 17418 : static bool need_more_worker(struct worker_pool *pool)
776 : : {
777 [ - - - + : 8555 : return !list_empty(&pool->worklist) && __need_more_worker(pool);
+ + - + ]
778 : : }
779 : :
780 : : /* Can I start working? Called from busy but !running workers. */
781 : 7664 : static bool may_start_working(struct worker_pool *pool)
782 : : {
783 : 7664 : return pool->nr_idle;
784 : : }
785 : :
786 : : /* Do I need to keep working? Called from currently running workers. */
787 : 9177 : static bool keep_working(struct worker_pool *pool)
788 : : {
789 [ - + ]: 1678 : return !list_empty(&pool->worklist) &&
790 : 1678 : atomic_read(&pool->nr_running) <= 1;
791 : : }
792 : :
793 : : /* Do we need a new worker? Called from manager. */
794 : 88 : static bool need_to_create_worker(struct worker_pool *pool)
795 : : {
796 [ + + + + : 165 : return need_more_worker(pool) && !may_start_working(pool);
+ - ]
797 : : }
798 : :
799 : : /* Do we have too many workers and should some go away? */
800 : 8159 : static bool too_many_workers(struct worker_pool *pool)
801 : : {
802 : 8159 : bool managing = pool->flags & POOL_MANAGER_ACTIVE;
803 : 8159 : int nr_idle = pool->nr_idle + managing; /* manager is considered idle */
804 : 8159 : int nr_busy = pool->nr_workers - nr_idle;
805 : :
806 [ - - - + ]: 3886 : return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy;
807 : : }
808 : :
809 : : /*
810 : : * Wake up functions.
811 : : */
812 : :
813 : : /* Return the first idle worker. Safe with preemption disabled */
814 : 8878 : static struct worker *first_idle_worker(struct worker_pool *pool)
815 : : {
816 : 8878 : if (unlikely(list_empty(&pool->idle_list)))
817 : : return NULL;
818 : :
819 : 8618 : return list_first_entry(&pool->idle_list, struct worker, entry);
820 : : }
821 : :
822 : : /**
823 : : * wake_up_worker - wake up an idle worker
824 : : * @pool: worker pool to wake worker from
825 : : *
826 : : * Wake up the first idle worker of @pool.
827 : : *
828 : : * CONTEXT:
829 : : * spin_lock_irq(pool->lock).
830 : : */
831 : 8789 : static void wake_up_worker(struct worker_pool *pool)
832 : : {
833 : 8789 : struct worker *worker = first_idle_worker(pool);
834 : :
835 [ - - + + : 8789 : if (likely(worker))
- - + - +
+ ]
836 : 8618 : wake_up_process(worker->task);
837 : : }
838 : :
839 : : /**
840 : : * wq_worker_running - a worker is running again
841 : : * @task: task waking up
842 : : *
843 : : * This function is called when a worker returns from schedule()
844 : : */
845 : 9223 : void wq_worker_running(struct task_struct *task)
846 : : {
847 : 9223 : struct worker *worker = kthread_data(task);
848 : :
849 [ + + ]: 9223 : if (!worker->sleeping)
850 : : return;
851 [ + - ]: 134 : if (!(worker->flags & WORKER_NOT_RUNNING))
852 : 134 : atomic_inc(&worker->pool->nr_running);
853 : 134 : worker->sleeping = 0;
854 : : }
855 : :
856 : : /**
857 : : * wq_worker_sleeping - a worker is going to sleep
858 : : * @task: task going to sleep
859 : : *
860 : : * This function is called from schedule() when a busy worker is
861 : : * going to sleep.
862 : : */
863 : 9570 : void wq_worker_sleeping(struct task_struct *task)
864 : : {
865 : 9570 : struct worker *next, *worker = kthread_data(task);
866 : 9570 : struct worker_pool *pool;
867 : :
868 : : /*
869 : : * Rescuers, which may not have all the fields set up like normal
870 : : * workers, also reach here, let's not access anything before
871 : : * checking NOT_RUNNING.
872 : : */
873 [ + + ]: 9570 : if (worker->flags & WORKER_NOT_RUNNING)
874 : : return;
875 : :
876 : 134 : pool = worker->pool;
877 : :
878 [ - + + - ]: 134 : if (WARN_ON_ONCE(worker->sleeping))
879 : : return;
880 : :
881 : 134 : worker->sleeping = 1;
882 : 134 : spin_lock_irq(&pool->lock);
883 : :
884 : : /*
885 : : * The counterpart of the following dec_and_test, implied mb,
886 : : * worklist not empty test sequence is in insert_work().
887 : : * Please read comment there.
888 : : *
889 : : * NOT_RUNNING is clear. This means that we're bound to and
890 : : * running on the local cpu w/ rq lock held and preemption
891 : : * disabled, which in turn means that none else could be
892 : : * manipulating idle_list, so dereferencing idle_list without pool
893 : : * lock is safe.
894 : : */
895 [ + - + + ]: 134 : if (atomic_dec_and_test(&pool->nr_running) &&
896 [ + + ]: 134 : !list_empty(&pool->worklist)) {
897 [ + - ]: 89 : next = first_idle_worker(pool);
898 [ + - ]: 89 : if (next)
899 : 89 : wake_up_process(next->task);
900 : : }
901 : 134 : spin_unlock_irq(&pool->lock);
902 : : }
903 : :
904 : : /**
905 : : * wq_worker_last_func - retrieve worker's last work function
906 : : * @task: Task to retrieve last work function of.
907 : : *
908 : : * Determine the last function a worker executed. This is called from
909 : : * the scheduler to get a worker's last known identity.
910 : : *
911 : : * CONTEXT:
912 : : * spin_lock_irq(rq->lock)
913 : : *
914 : : * This function is called during schedule() when a kworker is going
915 : : * to sleep. It's used by psi to identify aggregation workers during
916 : : * dequeuing, to allow periodic aggregation to shut-off when that
917 : : * worker is the last task in the system or cgroup to go to sleep.
918 : : *
919 : : * As this function doesn't involve any workqueue-related locking, it
920 : : * only returns stable values when called from inside the scheduler's
921 : : * queuing and dequeuing paths, when @task, which must be a kworker,
922 : : * is guaranteed to not be processing any works.
923 : : *
924 : : * Return:
925 : : * The last work function %current executed as a worker, NULL if it
926 : : * hasn't executed any work yet.
927 : : */
928 : 0 : work_func_t wq_worker_last_func(struct task_struct *task)
929 : : {
930 : 0 : struct worker *worker = kthread_data(task);
931 : :
932 : 0 : return worker->last_func;
933 : : }
934 : :
935 : : /**
936 : : * worker_set_flags - set worker flags and adjust nr_running accordingly
937 : : * @worker: self
938 : : * @flags: flags to set
939 : : *
940 : : * Set @flags in @worker->flags and adjust nr_running accordingly.
941 : : *
942 : : * CONTEXT:
943 : : * spin_lock_irq(pool->lock)
944 : : */
945 : 7499 : static inline void worker_set_flags(struct worker *worker, unsigned int flags)
946 : : {
947 : 7499 : struct worker_pool *pool = worker->pool;
948 : :
949 [ - + ]: 7499 : WARN_ON_ONCE(worker->task != current);
950 : :
951 : : /* If transitioning into NOT_RUNNING, adjust nr_running. */
952 [ + - ]: 7499 : if ((flags & WORKER_NOT_RUNNING) &&
953 [ + + ]: 7499 : !(worker->flags & WORKER_NOT_RUNNING)) {
954 : 6570 : atomic_dec(&pool->nr_running);
955 : : }
956 : :
957 : 7499 : worker->flags |= flags;
958 : 7499 : }
959 : :
960 : : /**
961 : : * worker_clr_flags - clear worker flags and adjust nr_running accordingly
962 : : * @worker: self
963 : : * @flags: flags to clear
964 : : *
965 : : * Clear @flags in @worker->flags and adjust nr_running accordingly.
966 : : *
967 : : * CONTEXT:
968 : : * spin_lock_irq(pool->lock)
969 : : */
970 : 15526 : static inline void worker_clr_flags(struct worker *worker, unsigned int flags)
971 : : {
972 : 15526 : struct worker_pool *pool = worker->pool;
973 : 15526 : unsigned int oflags = worker->flags;
974 : :
975 [ - + ]: 15526 : WARN_ON_ONCE(worker->task != current);
976 : :
977 : 15526 : worker->flags &= ~flags;
978 : :
979 : : /*
980 : : * If transitioning out of NOT_RUNNING, increment nr_running. Note
981 : : * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask
982 : : * of multiple flags, not a single flag.
983 : : */
984 [ + + + - ]: 15526 : if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING))
985 [ + + ]: 7499 : if (!(worker->flags & WORKER_NOT_RUNNING))
986 : 6570 : atomic_inc(&pool->nr_running);
987 : 15526 : }
988 : :
989 : : /**
990 : : * find_worker_executing_work - find worker which is executing a work
991 : : * @pool: pool of interest
992 : : * @work: work to find worker for
993 : : *
994 : : * Find a worker which is executing @work on @pool by searching
995 : : * @pool->busy_hash which is keyed by the address of @work. For a worker
996 : : * to match, its current execution should match the address of @work and
997 : : * its work function. This is to avoid unwanted dependency between
998 : : * unrelated work executions through a work item being recycled while still
999 : : * being executed.
1000 : : *
1001 : : * This is a bit tricky. A work item may be freed once its execution
1002 : : * starts and nothing prevents the freed area from being recycled for
1003 : : * another work item. If the same work item address ends up being reused
1004 : : * before the original execution finishes, workqueue will identify the
1005 : : * recycled work item as currently executing and make it wait until the
1006 : : * current execution finishes, introducing an unwanted dependency.
1007 : : *
1008 : : * This function checks the work item address and work function to avoid
1009 : : * false positives. Note that this isn't complete as one may construct a
1010 : : * work function which can introduce dependency onto itself through a
1011 : : * recycled work item. Well, if somebody wants to shoot oneself in the
1012 : : * foot that badly, there's only so much we can do, and if such deadlock
1013 : : * actually occurs, it should be easy to locate the culprit work function.
1014 : : *
1015 : : * CONTEXT:
1016 : : * spin_lock_irq(pool->lock).
1017 : : *
1018 : : * Return:
1019 : : * Pointer to worker which is executing @work if found, %NULL
1020 : : * otherwise.
1021 : : */
1022 : 9619 : static struct worker *find_worker_executing_work(struct worker_pool *pool,
1023 : : struct work_struct *work)
1024 : : {
1025 : 9619 : struct worker *worker;
1026 : :
1027 [ - - - - : 9623 : hash_for_each_possible(pool->busy_hash, worker, hentry,
- - + + -
+ + + - -
- - ]
1028 : : (unsigned long)work)
1029 [ - - + - : 17 : if (worker->current_work == work &&
- + - - ]
1030 [ - - - + : 13 : worker->current_func == work->func)
- - - - ]
1031 : : return worker;
1032 : :
1033 : : return NULL;
1034 : : }
1035 : :
1036 : : /**
1037 : : * move_linked_works - move linked works to a list
1038 : : * @work: start of series of works to be scheduled
1039 : : * @head: target list to append @work to
1040 : : * @nextp: out parameter for nested worklist walking
1041 : : *
1042 : : * Schedule linked works starting from @work to @head. Work series to
1043 : : * be scheduled starts at @work and includes any consecutive work with
1044 : : * WORK_STRUCT_LINKED set in its predecessor.
1045 : : *
1046 : : * If @nextp is not NULL, it's updated to point to the next work of
1047 : : * the last scheduled work. This allows move_linked_works() to be
1048 : : * nested inside outer list_for_each_entry_safe().
1049 : : *
1050 : : * CONTEXT:
1051 : : * spin_lock_irq(pool->lock).
1052 : : */
1053 : 335 : static void move_linked_works(struct work_struct *work, struct list_head *head,
1054 : : struct work_struct **nextp)
1055 : : {
1056 : 335 : struct work_struct *n;
1057 : :
1058 : : /*
1059 : : * Linked worklist will always end before the end of the list,
1060 : : * use NULL for list head.
1061 : : */
1062 [ - - + - : 360 : list_for_each_entry_safe_from(work, n, NULL, entry) {
- - + - ]
1063 [ - - + + : 360 : list_move_tail(&work->entry, head);
- - - + ]
1064 [ - - + + : 360 : if (!(*work_data_bits(work) & WORK_STRUCT_LINKED))
- - - + ]
1065 : : break;
1066 : : }
1067 : :
1068 : : /*
1069 : : * If we're already inside safe list traversal and have moved
1070 : : * multiple works to the scheduled queue, the next position
1071 : : * needs to be updated.
1072 : : */
1073 : 310 : if (nextp)
1074 : : *nextp = n;
1075 : : }
1076 : :
1077 : : /**
1078 : : * get_pwq - get an extra reference on the specified pool_workqueue
1079 : : * @pwq: pool_workqueue to get
1080 : : *
1081 : : * Obtain an extra reference on @pwq. The caller should guarantee that
1082 : : * @pwq has positive refcnt and be holding the matching pool->lock.
1083 : : */
1084 : 9274 : static void get_pwq(struct pool_workqueue *pwq)
1085 : : {
1086 : 9274 : lockdep_assert_held(&pwq->pool->lock);
1087 : 0 : WARN_ON_ONCE(pwq->refcnt <= 0);
1088 : 9274 : pwq->refcnt++;
1089 : : }
1090 : :
1091 : : /**
1092 : : * put_pwq - put a pool_workqueue reference
1093 : : * @pwq: pool_workqueue to put
1094 : : *
1095 : : * Drop a reference of @pwq. If its refcnt reaches zero, schedule its
1096 : : * destruction. The caller should be holding the matching pool->lock.
1097 : : */
1098 : 9296 : static void put_pwq(struct pool_workqueue *pwq)
1099 : : {
1100 : 9296 : lockdep_assert_held(&pwq->pool->lock);
1101 [ + + ]: 9296 : if (likely(--pwq->refcnt))
1102 : : return;
1103 [ - + + - ]: 11 : if (WARN_ON_ONCE(!(pwq->wq->flags & WQ_UNBOUND)))
1104 : : return;
1105 : : /*
1106 : : * @pwq can't be released under pool->lock, bounce to
1107 : : * pwq_unbound_release_workfn(). This never recurses on the same
1108 : : * pool->lock as this path is taken only for unbound workqueues and
1109 : : * the release work item is scheduled on a per-cpu workqueue. To
1110 : : * avoid lockdep warning, unbound pool->locks are given lockdep
1111 : : * subclass of 1 in get_unbound_pool().
1112 : : */
1113 : 11 : schedule_work(&pwq->unbound_release_work);
1114 : : }
1115 : :
1116 : : /**
1117 : : * put_pwq_unlocked - put_pwq() with surrounding pool lock/unlock
1118 : : * @pwq: pool_workqueue to put (can be %NULL)
1119 : : *
1120 : : * put_pwq() with locking. This function also allows %NULL @pwq.
1121 : : */
1122 : 440 : static void put_pwq_unlocked(struct pool_workqueue *pwq)
1123 : : {
1124 [ + + ]: 440 : if (pwq) {
1125 : : /*
1126 : : * As both pwqs and pools are RCU protected, the
1127 : : * following lock operations are safe.
1128 : : */
1129 : 22 : spin_lock_irq(&pwq->pool->lock);
1130 : 22 : put_pwq(pwq);
1131 : 22 : spin_unlock_irq(&pwq->pool->lock);
1132 : : }
1133 : 440 : }
1134 : :
1135 : 310 : static void pwq_activate_delayed_work(struct work_struct *work)
1136 : : {
1137 : 310 : struct pool_workqueue *pwq = get_work_pwq(work);
1138 : :
1139 : 310 : trace_workqueue_activate_work(work);
1140 [ + + ]: 310 : if (list_empty(&pwq->pool->worklist))
1141 : 175 : pwq->pool->watchdog_ts = jiffies;
1142 : 310 : move_linked_works(work, &pwq->pool->worklist, NULL);
1143 : 310 : __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work));
1144 : 310 : pwq->nr_active++;
1145 : 310 : }
1146 : :
1147 : 310 : static void pwq_activate_first_delayed(struct pool_workqueue *pwq)
1148 : : {
1149 : 310 : struct work_struct *work = list_first_entry(&pwq->delayed_works,
1150 : : struct work_struct, entry);
1151 : :
1152 : 310 : pwq_activate_delayed_work(work);
1153 : 310 : }
1154 : :
1155 : : /**
1156 : : * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight
1157 : : * @pwq: pwq of interest
1158 : : * @color: color of work which left the queue
1159 : : *
1160 : : * A work either has completed or is removed from pending queue,
1161 : : * decrement nr_in_flight of its pwq and handle workqueue flushing.
1162 : : *
1163 : : * CONTEXT:
1164 : : * spin_lock_irq(pool->lock).
1165 : : */
1166 : 9274 : static void pwq_dec_nr_in_flight(struct pool_workqueue *pwq, int color)
1167 : : {
1168 : : /* uncolored work items don't participate in flushing or nr_active */
1169 [ + + ]: 9274 : if (color == WORK_NO_COLOR)
1170 : 38 : goto out_put;
1171 : :
1172 : 9236 : pwq->nr_in_flight[color]--;
1173 : :
1174 : 9236 : pwq->nr_active--;
1175 [ + + ]: 9236 : if (!list_empty(&pwq->delayed_works)) {
1176 : : /* one down, submit a delayed one */
1177 [ + - ]: 310 : if (pwq->nr_active < pwq->max_active)
1178 : 310 : pwq_activate_first_delayed(pwq);
1179 : : }
1180 : :
1181 : : /* is flush in progress and are we at the flushing tip? */
1182 [ + - ]: 9236 : if (likely(pwq->flush_color != color))
1183 : 9236 : goto out_put;
1184 : :
1185 : : /* are there still in-flight works? */
1186 [ # # ]: 0 : if (pwq->nr_in_flight[color])
1187 : 0 : goto out_put;
1188 : :
1189 : : /* this pwq is done, clear flush_color */
1190 : 0 : pwq->flush_color = -1;
1191 : :
1192 : : /*
1193 : : * If this was the last pwq, wake up the first flusher. It
1194 : : * will handle the rest.
1195 : : */
1196 [ # # ]: 0 : if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush))
1197 : 0 : complete(&pwq->wq->first_flusher->done);
1198 : 0 : out_put:
1199 : 9274 : put_pwq(pwq);
1200 : 9274 : }
1201 : :
1202 : : /**
1203 : : * try_to_grab_pending - steal work item from worklist and disable irq
1204 : : * @work: work item to steal
1205 : : * @is_dwork: @work is a delayed_work
1206 : : * @flags: place to store irq state
1207 : : *
1208 : : * Try to grab PENDING bit of @work. This function can handle @work in any
1209 : : * stable state - idle, on timer or on worklist.
1210 : : *
1211 : : * Return:
1212 : : * 1 if @work was pending and we successfully stole PENDING
1213 : : * 0 if @work was idle and we claimed PENDING
1214 : : * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry
1215 : : * -ENOENT if someone else is canceling @work, this state may persist
1216 : : * for arbitrarily long
1217 : : *
1218 : : * Note:
1219 : : * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting
1220 : : * interrupted while holding PENDING and @work off queue, irq must be
1221 : : * disabled on entry. This, combined with delayed_work->timer being
1222 : : * irqsafe, ensures that we return -EAGAIN for finite short period of time.
1223 : : *
1224 : : * On successful return, >= 0, irq is disabled and the caller is
1225 : : * responsible for releasing it using local_irq_restore(*@flags).
1226 : : *
1227 : : * This function is safe to call from any context including IRQ handler.
1228 : : */
1229 : 4759 : static int try_to_grab_pending(struct work_struct *work, bool is_dwork,
1230 : : unsigned long *flags)
1231 : : {
1232 : 4759 : struct worker_pool *pool;
1233 : 4759 : struct pool_workqueue *pwq;
1234 : :
1235 : 4759 : local_irq_save(*flags);
1236 : :
1237 : : /* try to steal the timer if it exists */
1238 [ + + ]: 4759 : if (is_dwork) {
1239 : 3450 : struct delayed_work *dwork = to_delayed_work(work);
1240 : :
1241 : : /*
1242 : : * dwork->timer is irqsafe. If del_timer() fails, it's
1243 : : * guaranteed that the timer is not queued anywhere and not
1244 : : * running on the local CPU.
1245 : : */
1246 [ + + ]: 3450 : if (likely(del_timer(&dwork->timer)))
1247 : : return 1;
1248 : : }
1249 : :
1250 : : /* try to claim PENDING the normal way */
1251 [ + + ]: 4661 : if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work)))
1252 : : return 0;
1253 : :
1254 : 59 : rcu_read_lock();
1255 : : /*
1256 : : * The queueing is in progress, or it is already queued. Try to
1257 : : * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
1258 : : */
1259 : 59 : pool = get_work_pool(work);
1260 [ - + ]: 59 : if (!pool)
1261 : 0 : goto fail;
1262 : :
1263 : 59 : spin_lock(&pool->lock);
1264 : : /*
1265 : : * work->data is guaranteed to point to pwq only while the work
1266 : : * item is queued on pwq->wq, and both updating work->data to point
1267 : : * to pwq on queueing and to pool on dequeueing are done under
1268 : : * pwq->pool->lock. This in turn guarantees that, if work->data
1269 : : * points to pwq which is associated with a locked pool, the work
1270 : : * item is currently queued on that pool.
1271 : : */
1272 : 59 : pwq = get_work_pwq(work);
1273 [ + - + - ]: 59 : if (pwq && pwq->pool == pool) {
1274 : 59 : debug_work_deactivate(work);
1275 : :
1276 : : /*
1277 : : * A delayed work item cannot be grabbed directly because
1278 : : * it might have linked NO_COLOR work items which, if left
1279 : : * on the delayed_list, will confuse pwq->nr_active
1280 : : * management later on and cause stall. Make sure the work
1281 : : * item is activated before grabbing.
1282 : : */
1283 [ - + ]: 59 : if (*work_data_bits(work) & WORK_STRUCT_DELAYED)
1284 : 0 : pwq_activate_delayed_work(work);
1285 : :
1286 : 59 : list_del_init(&work->entry);
1287 : 59 : pwq_dec_nr_in_flight(pwq, get_work_color(work));
1288 : :
1289 : : /* work->data points to pwq iff queued, point to pool */
1290 : 59 : set_work_pool_and_keep_pending(work, pool->id);
1291 : :
1292 : 59 : spin_unlock(&pool->lock);
1293 : 59 : rcu_read_unlock();
1294 : 59 : return 1;
1295 : : }
1296 : 0 : spin_unlock(&pool->lock);
1297 : 0 : fail:
1298 : 0 : rcu_read_unlock();
1299 : 0 : local_irq_restore(*flags);
1300 [ # # ]: 0 : if (work_is_canceling(work))
1301 : : return -ENOENT;
1302 : 0 : cpu_relax();
1303 : 0 : return -EAGAIN;
1304 : : }
1305 : :
1306 : : /**
1307 : : * insert_work - insert a work into a pool
1308 : : * @pwq: pwq @work belongs to
1309 : : * @work: work to insert
1310 : : * @head: insertion point
1311 : : * @extra_flags: extra WORK_STRUCT_* flags to set
1312 : : *
1313 : : * Insert @work which belongs to @pwq after @head. @extra_flags is or'd to
1314 : : * work_struct flags.
1315 : : *
1316 : : * CONTEXT:
1317 : : * spin_lock_irq(pool->lock).
1318 : : */
1319 : 9274 : static void insert_work(struct pool_workqueue *pwq, struct work_struct *work,
1320 : : struct list_head *head, unsigned int extra_flags)
1321 : : {
1322 : 9274 : struct worker_pool *pool = pwq->pool;
1323 : :
1324 : : /* we own @work, set data and link */
1325 : 9274 : set_work_pwq(work, pwq, extra_flags);
1326 [ - + ]: 9274 : list_add_tail(&work->entry, head);
1327 [ - + ]: 9274 : get_pwq(pwq);
1328 : :
1329 : : /*
1330 : : * Ensure either wq_worker_sleeping() sees the above
1331 : : * list_add_tail() or we see zero nr_running to avoid workers lying
1332 : : * around lazily while there are works to be processed.
1333 : : */
1334 : 9274 : smp_mb();
1335 : :
1336 [ + + ]: 9274 : if (__need_more_worker(pool))
1337 [ + + ]: 8236 : wake_up_worker(pool);
1338 : 9274 : }
1339 : :
1340 : : /*
1341 : : * Test whether @work is being queued from another work executing on the
1342 : : * same workqueue.
1343 : : */
1344 : 0 : static bool is_chained_work(struct workqueue_struct *wq)
1345 : : {
1346 : 0 : struct worker *worker;
1347 : :
1348 : 0 : worker = current_wq_worker();
1349 : : /*
1350 : : * Return %true iff I'm a worker executing a work item on @wq. If
1351 : : * I'm @worker, it's safe to dereference it without locking.
1352 : : */
1353 [ # # # # ]: 0 : return worker && worker->current_pwq->wq == wq;
1354 : : }
1355 : :
1356 : : /*
1357 : : * When queueing an unbound work item to a wq, prefer local CPU if allowed
1358 : : * by wq_unbound_cpumask. Otherwise, round robin among the allowed ones to
1359 : : * avoid perturbing sensitive tasks.
1360 : : */
1361 : 941 : static int wq_select_unbound_cpu(int cpu)
1362 : : {
1363 : 941 : static bool printed_dbg_warning;
1364 : 941 : int new_cpu;
1365 : :
1366 [ + - ]: 941 : if (likely(!wq_debug_force_rr_cpu)) {
1367 [ - + ]: 941 : if (cpumask_test_cpu(cpu, wq_unbound_cpumask))
1368 : : return cpu;
1369 [ # # ]: 0 : } else if (!printed_dbg_warning) {
1370 : 0 : pr_warn("workqueue: round-robin CPU selection forced, expect performance impact\n");
1371 : 0 : printed_dbg_warning = true;
1372 : : }
1373 : :
1374 [ # # ]: 0 : if (cpumask_empty(wq_unbound_cpumask))
1375 : : return cpu;
1376 : :
1377 : 0 : new_cpu = __this_cpu_read(wq_rr_cpu_last);
1378 : 0 : new_cpu = cpumask_next_and(new_cpu, wq_unbound_cpumask, cpu_online_mask);
1379 [ # # ]: 0 : if (unlikely(new_cpu >= nr_cpu_ids)) {
1380 : 0 : new_cpu = cpumask_first_and(wq_unbound_cpumask, cpu_online_mask);
1381 [ # # ]: 0 : if (unlikely(new_cpu >= nr_cpu_ids))
1382 : : return cpu;
1383 : : }
1384 : 0 : __this_cpu_write(wq_rr_cpu_last, new_cpu);
1385 : :
1386 : 0 : return new_cpu;
1387 : : }
1388 : :
1389 : 9236 : static void __queue_work(int cpu, struct workqueue_struct *wq,
1390 : : struct work_struct *work)
1391 : : {
1392 : 9236 : struct pool_workqueue *pwq;
1393 : 9236 : struct worker_pool *last_pool;
1394 : 9236 : struct list_head *worklist;
1395 : 9236 : unsigned int work_flags;
1396 : 9236 : unsigned int req_cpu = cpu;
1397 : :
1398 : : /*
1399 : : * While a work item is PENDING && off queue, a task trying to
1400 : : * steal the PENDING will busy-loop waiting for it to either get
1401 : : * queued or lose PENDING. Grabbing PENDING and queueing should
1402 : : * happen with IRQ disabled.
1403 : : */
1404 : 9236 : lockdep_assert_irqs_disabled();
1405 : :
1406 : 9236 : debug_work_activate(work);
1407 : :
1408 : : /* if draining, only works from the same workqueue are allowed */
1409 [ - + ]: 9236 : if (unlikely(wq->flags & __WQ_DRAINING) &&
1410 [ # # # # ]: 0 : WARN_ON_ONCE(!is_chained_work(wq)))
1411 : : return;
1412 : 9236 : rcu_read_lock();
1413 : 9236 : retry:
1414 : : /* pwq which will be used unless @work is executing elsewhere */
1415 [ + + ]: 9236 : if (wq->flags & WQ_UNBOUND) {
1416 [ + - ]: 941 : if (req_cpu == WORK_CPU_UNBOUND)
1417 : 941 : cpu = wq_select_unbound_cpu(raw_smp_processor_id());
1418 [ - + ]: 941 : pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu));
1419 : : } else {
1420 [ + + ]: 8295 : if (req_cpu == WORK_CPU_UNBOUND)
1421 : 7870 : cpu = raw_smp_processor_id();
1422 : 8295 : pwq = per_cpu_ptr(wq->cpu_pwqs, cpu);
1423 : : }
1424 : :
1425 : : /*
1426 : : * If @work was previously on a different pool, it might still be
1427 : : * running there, in which case the work needs to be queued on that
1428 : : * pool to guarantee non-reentrancy.
1429 : : */
1430 : 9236 : last_pool = get_work_pool(work);
1431 [ + + - + ]: 9236 : if (last_pool && last_pool != pwq->pool) {
1432 : 0 : struct worker *worker;
1433 : :
1434 : 0 : spin_lock(&last_pool->lock);
1435 : :
1436 [ # # ]: 0 : worker = find_worker_executing_work(last_pool, work);
1437 : :
1438 [ # # # # ]: 0 : if (worker && worker->current_pwq->wq == wq) {
1439 : : pwq = worker->current_pwq;
1440 : : } else {
1441 : : /* meh... not running there, queue here */
1442 : 0 : spin_unlock(&last_pool->lock);
1443 : 0 : spin_lock(&pwq->pool->lock);
1444 : : }
1445 : : } else {
1446 : 9236 : spin_lock(&pwq->pool->lock);
1447 : : }
1448 : :
1449 : : /*
1450 : : * pwq is determined and locked. For unbound pools, we could have
1451 : : * raced with pwq release and it could already be dead. If its
1452 : : * refcnt is zero, repeat pwq selection. Note that pwqs never die
1453 : : * without another pwq replacing it in the numa_pwq_tbl or while
1454 : : * work items are executing on it, so the retrying is guaranteed to
1455 : : * make forward-progress.
1456 : : */
1457 [ - + ]: 9236 : if (unlikely(!pwq->refcnt)) {
1458 [ # # ]: 0 : if (wq->flags & WQ_UNBOUND) {
1459 : 0 : spin_unlock(&pwq->pool->lock);
1460 : 0 : cpu_relax();
1461 : 0 : goto retry;
1462 : : }
1463 : : /* oops */
1464 [ # # ]: 0 : WARN_ONCE(true, "workqueue: per-cpu pwq for %s on cpu%d has 0 refcnt",
1465 : : wq->name, cpu);
1466 : : }
1467 : :
1468 : : /* pwq determined, queue */
1469 : 9236 : trace_workqueue_queue_work(req_cpu, pwq, work);
1470 : :
1471 [ - + - + ]: 9236 : if (WARN_ON(!list_empty(&work->entry)))
1472 : 0 : goto out;
1473 : :
1474 : 9236 : pwq->nr_in_flight[pwq->work_color]++;
1475 : 9236 : work_flags = work_color_to_flags(pwq->work_color);
1476 : :
1477 [ + + ]: 9236 : if (likely(pwq->nr_active < pwq->max_active)) {
1478 : 8926 : trace_workqueue_activate_work(work);
1479 : 8926 : pwq->nr_active++;
1480 : 8926 : worklist = &pwq->pool->worklist;
1481 [ + + ]: 8926 : if (list_empty(worklist))
1482 : 8159 : pwq->pool->watchdog_ts = jiffies;
1483 : : } else {
1484 : 310 : work_flags |= WORK_STRUCT_DELAYED;
1485 : 310 : worklist = &pwq->delayed_works;
1486 : : }
1487 : :
1488 : 9236 : insert_work(pwq, work, worklist, work_flags);
1489 : :
1490 : 9236 : out:
1491 : 9236 : spin_unlock(&pwq->pool->lock);
1492 : 9236 : rcu_read_unlock();
1493 : : }
1494 : :
1495 : : /**
1496 : : * queue_work_on - queue work on specific cpu
1497 : : * @cpu: CPU number to execute work on
1498 : : * @wq: workqueue to use
1499 : : * @work: work to queue
1500 : : *
1501 : : * We queue the work to a specific CPU, the caller must ensure it
1502 : : * can't go away.
1503 : : *
1504 : : * Return: %false if @work was already on a queue, %true otherwise.
1505 : : */
1506 : 4163 : bool queue_work_on(int cpu, struct workqueue_struct *wq,
1507 : : struct work_struct *work)
1508 : : {
1509 : 4163 : bool ret = false;
1510 : 4163 : unsigned long flags;
1511 : :
1512 : 4163 : local_irq_save(flags);
1513 : :
1514 [ + + ]: 4163 : if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
1515 : 3900 : __queue_work(cpu, wq, work);
1516 : 3900 : ret = true;
1517 : : }
1518 : :
1519 : 4163 : local_irq_restore(flags);
1520 : 4163 : return ret;
1521 : : }
1522 : : EXPORT_SYMBOL(queue_work_on);
1523 : :
1524 : : /**
1525 : : * workqueue_select_cpu_near - Select a CPU based on NUMA node
1526 : : * @node: NUMA node ID that we want to select a CPU from
1527 : : *
1528 : : * This function will attempt to find a "random" cpu available on a given
1529 : : * node. If there are no CPUs available on the given node it will return
1530 : : * WORK_CPU_UNBOUND indicating that we should just schedule to any
1531 : : * available CPU if we need to schedule this work.
1532 : : */
1533 : 55 : static int workqueue_select_cpu_near(int node)
1534 : : {
1535 : 55 : int cpu;
1536 : :
1537 : : /* No point in doing this if NUMA isn't enabled for workqueues */
1538 [ - + ]: 55 : if (!wq_numa_enabled)
1539 : : return WORK_CPU_UNBOUND;
1540 : :
1541 : : /* Delay binding to CPU if node is not valid or online */
1542 [ # # # # ]: 0 : if (node < 0 || node >= MAX_NUMNODES || !node_online(node))
1543 : 0 : return WORK_CPU_UNBOUND;
1544 : :
1545 : : /* Use local node/cpu if we are already there */
1546 : 0 : cpu = raw_smp_processor_id();
1547 [ # # ]: 0 : if (node == cpu_to_node(cpu))
1548 : : return cpu;
1549 : :
1550 : : /* Use "random" otherwise know as "first" online CPU of node */
1551 : 0 : cpu = cpumask_any_and(cpumask_of_node(node), cpu_online_mask);
1552 : :
1553 : : /* If CPU is valid return that, otherwise just defer */
1554 [ # # ]: 0 : return cpu < nr_cpu_ids ? cpu : WORK_CPU_UNBOUND;
1555 : : }
1556 : :
1557 : : /**
1558 : : * queue_work_node - queue work on a "random" cpu for a given NUMA node
1559 : : * @node: NUMA node that we are targeting the work for
1560 : : * @wq: workqueue to use
1561 : : * @work: work to queue
1562 : : *
1563 : : * We queue the work to a "random" CPU within a given NUMA node. The basic
1564 : : * idea here is to provide a way to somehow associate work with a given
1565 : : * NUMA node.
1566 : : *
1567 : : * This function will only make a best effort attempt at getting this onto
1568 : : * the right NUMA node. If no node is requested or the requested node is
1569 : : * offline then we just fall back to standard queue_work behavior.
1570 : : *
1571 : : * Currently the "random" CPU ends up being the first available CPU in the
1572 : : * intersection of cpu_online_mask and the cpumask of the node, unless we
1573 : : * are running on the node. In that case we just use the current CPU.
1574 : : *
1575 : : * Return: %false if @work was already on a queue, %true otherwise.
1576 : : */
1577 : 55 : bool queue_work_node(int node, struct workqueue_struct *wq,
1578 : : struct work_struct *work)
1579 : : {
1580 : 55 : unsigned long flags;
1581 : 55 : bool ret = false;
1582 : :
1583 : : /*
1584 : : * This current implementation is specific to unbound workqueues.
1585 : : * Specifically we only return the first available CPU for a given
1586 : : * node instead of cycling through individual CPUs within the node.
1587 : : *
1588 : : * If this is used with a per-cpu workqueue then the logic in
1589 : : * workqueue_select_cpu_near would need to be updated to allow for
1590 : : * some round robin type logic.
1591 : : */
1592 [ - + ]: 55 : WARN_ON_ONCE(!(wq->flags & WQ_UNBOUND));
1593 : :
1594 : 55 : local_irq_save(flags);
1595 : :
1596 [ + - ]: 55 : if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
1597 : 55 : int cpu = workqueue_select_cpu_near(node);
1598 : :
1599 : 55 : __queue_work(cpu, wq, work);
1600 : 55 : ret = true;
1601 : : }
1602 : :
1603 : 55 : local_irq_restore(flags);
1604 : 55 : return ret;
1605 : : }
1606 : : EXPORT_SYMBOL_GPL(queue_work_node);
1607 : :
1608 : 732 : void delayed_work_timer_fn(struct timer_list *t)
1609 : : {
1610 : 732 : struct delayed_work *dwork = from_timer(dwork, t, timer);
1611 : :
1612 : : /* should have been called from irqsafe timer with irq already off */
1613 : 732 : __queue_work(dwork->cpu, dwork->wq, &dwork->work);
1614 : 732 : }
1615 : : EXPORT_SYMBOL(delayed_work_timer_fn);
1616 : :
1617 : 4985 : static void __queue_delayed_work(int cpu, struct workqueue_struct *wq,
1618 : : struct delayed_work *dwork, unsigned long delay)
1619 : : {
1620 : 4985 : struct timer_list *timer = &dwork->timer;
1621 : 4985 : struct work_struct *work = &dwork->work;
1622 : :
1623 [ - + ]: 4985 : WARN_ON_ONCE(!wq);
1624 [ - + ]: 4985 : WARN_ON_ONCE(timer->function != delayed_work_timer_fn);
1625 [ - + ]: 4985 : WARN_ON_ONCE(timer_pending(timer));
1626 [ - + ]: 4985 : WARN_ON_ONCE(!list_empty(&work->entry));
1627 : :
1628 : : /*
1629 : : * If @delay is 0, queue @dwork->work immediately. This is for
1630 : : * both optimization and correctness. The earliest @timer can
1631 : : * expire is on the closest next tick and delayed_work users depend
1632 : : * on that there's no such delay when @delay is 0.
1633 : : */
1634 [ + + ]: 4985 : if (!delay) {
1635 : 3977 : __queue_work(cpu, wq, &dwork->work);
1636 : 3977 : return;
1637 : : }
1638 : :
1639 : 1008 : dwork->wq = wq;
1640 : 1008 : dwork->cpu = cpu;
1641 : 1008 : timer->expires = jiffies + delay;
1642 : :
1643 [ + + ]: 1008 : if (unlikely(cpu != WORK_CPU_UNBOUND))
1644 : 12 : add_timer_on(timer, cpu);
1645 : : else
1646 : 996 : add_timer(timer);
1647 : : }
1648 : :
1649 : : /**
1650 : : * queue_delayed_work_on - queue work on specific CPU after delay
1651 : : * @cpu: CPU number to execute work on
1652 : : * @wq: workqueue to use
1653 : : * @dwork: work to queue
1654 : : * @delay: number of jiffies to wait before queueing
1655 : : *
1656 : : * Return: %false if @work was already on a queue, %true otherwise. If
1657 : : * @delay is zero and @dwork is idle, it will be scheduled for immediate
1658 : : * execution.
1659 : : */
1660 : 2041 : bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
1661 : : struct delayed_work *dwork, unsigned long delay)
1662 : : {
1663 : 2041 : struct work_struct *work = &dwork->work;
1664 : 2041 : bool ret = false;
1665 : 2041 : unsigned long flags;
1666 : :
1667 : : /* read the comment in __queue_work() */
1668 : 2041 : local_irq_save(flags);
1669 : :
1670 [ + + ]: 2041 : if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
1671 : 1975 : __queue_delayed_work(cpu, wq, dwork, delay);
1672 : 1975 : ret = true;
1673 : : }
1674 : :
1675 : 2041 : local_irq_restore(flags);
1676 : 2041 : return ret;
1677 : : }
1678 : : EXPORT_SYMBOL(queue_delayed_work_on);
1679 : :
1680 : : /**
1681 : : * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU
1682 : : * @cpu: CPU number to execute work on
1683 : : * @wq: workqueue to use
1684 : : * @dwork: work to queue
1685 : : * @delay: number of jiffies to wait before queueing
1686 : : *
1687 : : * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise,
1688 : : * modify @dwork's timer so that it expires after @delay. If @delay is
1689 : : * zero, @work is guaranteed to be scheduled immediately regardless of its
1690 : : * current state.
1691 : : *
1692 : : * Return: %false if @dwork was idle and queued, %true if @dwork was
1693 : : * pending and its timer was modified.
1694 : : *
1695 : : * This function is safe to call from any context including IRQ handler.
1696 : : * See try_to_grab_pending() for details.
1697 : : */
1698 : 3010 : bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq,
1699 : : struct delayed_work *dwork, unsigned long delay)
1700 : : {
1701 : 3010 : unsigned long flags;
1702 : 3010 : int ret;
1703 : :
1704 : 3010 : do {
1705 : 3010 : ret = try_to_grab_pending(&dwork->work, true, &flags);
1706 [ - + ]: 3010 : } while (unlikely(ret == -EAGAIN));
1707 : :
1708 [ + - ]: 3010 : if (likely(ret >= 0)) {
1709 : 3010 : __queue_delayed_work(cpu, wq, dwork, delay);
1710 : 3010 : local_irq_restore(flags);
1711 : : }
1712 : :
1713 : : /* -ENOENT from try_to_grab_pending() becomes %true */
1714 : 3010 : return ret;
1715 : : }
1716 : : EXPORT_SYMBOL_GPL(mod_delayed_work_on);
1717 : :
1718 : 550 : static void rcu_work_rcufn(struct rcu_head *rcu)
1719 : : {
1720 : 550 : struct rcu_work *rwork = container_of(rcu, struct rcu_work, rcu);
1721 : :
1722 : : /* read the comment in __queue_work() */
1723 : 550 : local_irq_disable();
1724 : 550 : __queue_work(WORK_CPU_UNBOUND, rwork->wq, &rwork->work);
1725 : 550 : local_irq_enable();
1726 : 550 : }
1727 : :
1728 : : /**
1729 : : * queue_rcu_work - queue work after a RCU grace period
1730 : : * @wq: workqueue to use
1731 : : * @rwork: work to queue
1732 : : *
1733 : : * Return: %false if @rwork was already pending, %true otherwise. Note
1734 : : * that a full RCU grace period is guaranteed only after a %true return.
1735 : : * While @rwork is guaranteed to be executed after a %false return, the
1736 : : * execution may happen before a full RCU grace period has passed.
1737 : : */
1738 : 550 : bool queue_rcu_work(struct workqueue_struct *wq, struct rcu_work *rwork)
1739 : : {
1740 : 550 : struct work_struct *work = &rwork->work;
1741 : :
1742 [ + - ]: 550 : if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
1743 : 550 : rwork->wq = wq;
1744 : 550 : call_rcu(&rwork->rcu, rcu_work_rcufn);
1745 : 550 : return true;
1746 : : }
1747 : :
1748 : : return false;
1749 : : }
1750 : : EXPORT_SYMBOL(queue_rcu_work);
1751 : :
1752 : : /**
1753 : : * worker_enter_idle - enter idle state
1754 : : * @worker: worker which is entering idle state
1755 : : *
1756 : : * @worker is entering idle state. Update stats and idle timer if
1757 : : * necessary.
1758 : : *
1759 : : * LOCKING:
1760 : : * spin_lock_irq(pool->lock).
1761 : : */
1762 : 8159 : static void worker_enter_idle(struct worker *worker)
1763 : : {
1764 : 8159 : struct worker_pool *pool = worker->pool;
1765 : :
1766 [ - + + - ]: 8159 : if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) ||
1767 [ + + + - : 16318 : WARN_ON_ONCE(!list_empty(&worker->entry) &&
+ - - + +
- ]
1768 : : (worker->hentry.next || worker->hentry.pprev)))
1769 : : return;
1770 : :
1771 : : /* can't use worker_set_flags(), also called from create_worker() */
1772 : 8159 : worker->flags |= WORKER_IDLE;
1773 : 8159 : pool->nr_idle++;
1774 : 8159 : worker->last_active = jiffies;
1775 : :
1776 : : /* idle_list is LIFO */
1777 [ + + ]: 8159 : list_add(&worker->entry, &pool->idle_list);
1778 : :
1779 [ + + + + : 16318 : if (too_many_workers(pool) && !timer_pending(&pool->idle_timer))
+ + ]
1780 : 22 : mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT);
1781 : :
1782 : : /*
1783 : : * Sanity check nr_running. Because unbind_workers() releases
1784 : : * pool->lock between setting %WORKER_UNBOUND and zapping
1785 : : * nr_running, the warning may trigger spuriously. Check iff
1786 : : * unbind is not in progress.
1787 : : */
1788 [ + + + + : 14511 : WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) &&
+ - - + ]
1789 : : pool->nr_workers == pool->nr_idle &&
1790 : : atomic_read(&pool->nr_running));
1791 : : }
1792 : :
1793 : : /**
1794 : : * worker_leave_idle - leave idle state
1795 : : * @worker: worker which is leaving idle state
1796 : : *
1797 : : * @worker is leaving idle state. Update stats.
1798 : : *
1799 : : * LOCKING:
1800 : : * spin_lock_irq(pool->lock).
1801 : : */
1802 : 8027 : static void worker_leave_idle(struct worker *worker)
1803 : : {
1804 : 8027 : struct worker_pool *pool = worker->pool;
1805 : :
1806 [ - + + - ]: 8027 : if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE)))
1807 : : return;
1808 : 8027 : worker_clr_flags(worker, WORKER_IDLE);
1809 : 8027 : pool->nr_idle--;
1810 : 8027 : list_del_init(&worker->entry);
1811 : : }
1812 : :
1813 : 363 : static struct worker *alloc_worker(int node)
1814 : : {
1815 : 363 : struct worker *worker;
1816 : :
1817 : 363 : worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, node);
1818 [ + - ]: 363 : if (worker) {
1819 : 363 : INIT_LIST_HEAD(&worker->entry);
1820 : 363 : INIT_LIST_HEAD(&worker->scheduled);
1821 : 363 : INIT_LIST_HEAD(&worker->node);
1822 : : /* on creation a worker is in !idle && prep state */
1823 : 363 : worker->flags = WORKER_PREP;
1824 : : }
1825 : 363 : return worker;
1826 : : }
1827 : :
1828 : : /**
1829 : : * worker_attach_to_pool() - attach a worker to a pool
1830 : : * @worker: worker to be attached
1831 : : * @pool: the target pool
1832 : : *
1833 : : * Attach @worker to @pool. Once attached, the %WORKER_UNBOUND flag and
1834 : : * cpu-binding of @worker are kept coordinated with the pool across
1835 : : * cpu-[un]hotplugs.
1836 : : */
1837 : 132 : static void worker_attach_to_pool(struct worker *worker,
1838 : : struct worker_pool *pool)
1839 : : {
1840 : 132 : mutex_lock(&wq_pool_attach_mutex);
1841 : :
1842 : : /*
1843 : : * set_cpus_allowed_ptr() will fail if the cpumask doesn't have any
1844 : : * online CPUs. It'll be re-applied when any of the CPUs come up.
1845 : : */
1846 : 132 : set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask);
1847 : :
1848 : : /*
1849 : : * The wq_pool_attach_mutex ensures %POOL_DISASSOCIATED remains
1850 : : * stable across this function. See the comments above the flag
1851 : : * definition for details.
1852 : : */
1853 [ + + ]: 132 : if (pool->flags & POOL_DISASSOCIATED)
1854 : 66 : worker->flags |= WORKER_UNBOUND;
1855 : :
1856 : 132 : list_add_tail(&worker->node, &pool->workers);
1857 : 132 : worker->pool = pool;
1858 : :
1859 : 132 : mutex_unlock(&wq_pool_attach_mutex);
1860 : 132 : }
1861 : :
1862 : : /**
1863 : : * worker_detach_from_pool() - detach a worker from its pool
1864 : : * @worker: worker which is attached to its pool
1865 : : *
1866 : : * Undo the attaching which had been done in worker_attach_to_pool(). The
1867 : : * caller worker shouldn't access to the pool after detached except it has
1868 : : * other reference to the pool.
1869 : : */
1870 : 0 : static void worker_detach_from_pool(struct worker *worker)
1871 : : {
1872 : 0 : struct worker_pool *pool = worker->pool;
1873 : 0 : struct completion *detach_completion = NULL;
1874 : :
1875 : 0 : mutex_lock(&wq_pool_attach_mutex);
1876 : :
1877 [ # # ]: 0 : list_del(&worker->node);
1878 : 0 : worker->pool = NULL;
1879 : :
1880 [ # # ]: 0 : if (list_empty(&pool->workers))
1881 : 0 : detach_completion = pool->detach_completion;
1882 : 0 : mutex_unlock(&wq_pool_attach_mutex);
1883 : :
1884 : : /* clear leftover flags without pool->lock after it is detached */
1885 : 0 : worker->flags &= ~(WORKER_UNBOUND | WORKER_REBOUND);
1886 : :
1887 [ # # ]: 0 : if (detach_completion)
1888 : 0 : complete(detach_completion);
1889 : 0 : }
1890 : :
1891 : : /**
1892 : : * create_worker - create a new workqueue worker
1893 : : * @pool: pool the new worker will belong to
1894 : : *
1895 : : * Create and start a new worker which is attached to @pool.
1896 : : *
1897 : : * CONTEXT:
1898 : : * Might sleep. Does GFP_KERNEL allocations.
1899 : : *
1900 : : * Return:
1901 : : * Pointer to the newly created worker.
1902 : : */
1903 : 132 : static struct worker *create_worker(struct worker_pool *pool)
1904 : : {
1905 : 132 : struct worker *worker = NULL;
1906 : 132 : int id = -1;
1907 : 132 : char id_buf[16];
1908 : :
1909 : : /* ID is needed to determine kthread name */
1910 : 132 : id = ida_simple_get(&pool->worker_ida, 0, 0, GFP_KERNEL);
1911 [ - + ]: 132 : if (id < 0)
1912 : 0 : goto fail;
1913 : :
1914 : 132 : worker = alloc_worker(pool->node);
1915 [ - + ]: 132 : if (!worker)
1916 : 0 : goto fail;
1917 : :
1918 : 132 : worker->id = id;
1919 : :
1920 [ + + ]: 132 : if (pool->cpu >= 0)
1921 : 66 : snprintf(id_buf, sizeof(id_buf), "%d:%d%s", pool->cpu, id,
1922 [ + + ]: 66 : pool->attrs->nice < 0 ? "H" : "");
1923 : : else
1924 : 66 : snprintf(id_buf, sizeof(id_buf), "u%d:%d", pool->id, id);
1925 : :
1926 : 132 : worker->task = kthread_create_on_node(worker_thread, worker, pool->node,
1927 : : "kworker/%s", id_buf);
1928 [ - + ]: 132 : if (IS_ERR(worker->task))
1929 : 0 : goto fail;
1930 : :
1931 : 132 : set_user_nice(worker->task, pool->attrs->nice);
1932 : 132 : kthread_bind_mask(worker->task, pool->attrs->cpumask);
1933 : :
1934 : : /* successful, attach the worker to the pool */
1935 : 132 : worker_attach_to_pool(worker, pool);
1936 : :
1937 : : /* start the newly created worker */
1938 : 132 : spin_lock_irq(&pool->lock);
1939 : 132 : worker->pool->nr_workers++;
1940 : 132 : worker_enter_idle(worker);
1941 : 132 : wake_up_process(worker->task);
1942 : 132 : spin_unlock_irq(&pool->lock);
1943 : :
1944 : 132 : return worker;
1945 : :
1946 : 0 : fail:
1947 [ # # ]: 0 : if (id >= 0)
1948 : 0 : ida_simple_remove(&pool->worker_ida, id);
1949 : 0 : kfree(worker);
1950 : 0 : return NULL;
1951 : : }
1952 : :
1953 : : /**
1954 : : * destroy_worker - destroy a workqueue worker
1955 : : * @worker: worker to be destroyed
1956 : : *
1957 : : * Destroy @worker and adjust @pool stats accordingly. The worker should
1958 : : * be idle.
1959 : : *
1960 : : * CONTEXT:
1961 : : * spin_lock_irq(pool->lock).
1962 : : */
1963 : 0 : static void destroy_worker(struct worker *worker)
1964 : : {
1965 : 0 : struct worker_pool *pool = worker->pool;
1966 : :
1967 : 0 : lockdep_assert_held(&pool->lock);
1968 : :
1969 : : /* sanity check frenzy */
1970 [ # # # # ]: 0 : if (WARN_ON(worker->current_work) ||
1971 [ # # # # ]: 0 : WARN_ON(!list_empty(&worker->scheduled)) ||
1972 [ # # # # ]: 0 : WARN_ON(!(worker->flags & WORKER_IDLE)))
1973 : : return;
1974 : :
1975 : 0 : pool->nr_workers--;
1976 : 0 : pool->nr_idle--;
1977 : :
1978 : 0 : list_del_init(&worker->entry);
1979 : 0 : worker->flags |= WORKER_DIE;
1980 : 0 : wake_up_process(worker->task);
1981 : : }
1982 : :
1983 : 0 : static void idle_worker_timeout(struct timer_list *t)
1984 : : {
1985 : 0 : struct worker_pool *pool = from_timer(pool, t, idle_timer);
1986 : :
1987 : 0 : spin_lock_irq(&pool->lock);
1988 : :
1989 [ # # # # ]: 0 : while (too_many_workers(pool)) {
1990 : 0 : struct worker *worker;
1991 : 0 : unsigned long expires;
1992 : :
1993 : : /* idle_list is kept in LIFO order, check the last one */
1994 : 0 : worker = list_entry(pool->idle_list.prev, struct worker, entry);
1995 : 0 : expires = worker->last_active + IDLE_WORKER_TIMEOUT;
1996 : :
1997 [ # # ]: 0 : if (time_before(jiffies, expires)) {
1998 : 0 : mod_timer(&pool->idle_timer, expires);
1999 : 0 : break;
2000 : : }
2001 : :
2002 : 0 : destroy_worker(worker);
2003 : : }
2004 : :
2005 : 0 : spin_unlock_irq(&pool->lock);
2006 : 0 : }
2007 : :
2008 : 0 : static void send_mayday(struct work_struct *work)
2009 : : {
2010 : 0 : struct pool_workqueue *pwq = get_work_pwq(work);
2011 : 0 : struct workqueue_struct *wq = pwq->wq;
2012 : :
2013 : 0 : lockdep_assert_held(&wq_mayday_lock);
2014 : :
2015 [ # # ]: 0 : if (!wq->rescuer)
2016 : : return;
2017 : :
2018 : : /* mayday mayday mayday */
2019 [ # # ]: 0 : if (list_empty(&pwq->mayday_node)) {
2020 : : /*
2021 : : * If @pwq is for an unbound wq, its base ref may be put at
2022 : : * any time due to an attribute change. Pin @pwq until the
2023 : : * rescuer is done with it.
2024 : : */
2025 [ # # ]: 0 : get_pwq(pwq);
2026 : 0 : list_add_tail(&pwq->mayday_node, &wq->maydays);
2027 : 0 : wake_up_process(wq->rescuer->task);
2028 : : }
2029 : : }
2030 : :
2031 : 0 : static void pool_mayday_timeout(struct timer_list *t)
2032 : : {
2033 : 0 : struct worker_pool *pool = from_timer(pool, t, mayday_timer);
2034 : 0 : struct work_struct *work;
2035 : :
2036 : 0 : spin_lock_irq(&pool->lock);
2037 : 0 : spin_lock(&wq_mayday_lock); /* for wq->maydays */
2038 : :
2039 [ # # ]: 0 : if (need_to_create_worker(pool)) {
2040 : : /*
2041 : : * We've been trying to create a new worker but
2042 : : * haven't been successful. We might be hitting an
2043 : : * allocation deadlock. Send distress signals to
2044 : : * rescuers.
2045 : : */
2046 [ # # ]: 0 : list_for_each_entry(work, &pool->worklist, entry)
2047 : 0 : send_mayday(work);
2048 : : }
2049 : :
2050 : 0 : spin_unlock(&wq_mayday_lock);
2051 : 0 : spin_unlock_irq(&pool->lock);
2052 : :
2053 : 0 : mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL);
2054 : 0 : }
2055 : :
2056 : : /**
2057 : : * maybe_create_worker - create a new worker if necessary
2058 : : * @pool: pool to create a new worker for
2059 : : *
2060 : : * Create a new worker for @pool if necessary. @pool is guaranteed to
2061 : : * have at least one idle worker on return from this function. If
2062 : : * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is
2063 : : * sent to all rescuers with works scheduled on @pool to resolve
2064 : : * possible allocation deadlock.
2065 : : *
2066 : : * On return, need_to_create_worker() is guaranteed to be %false and
2067 : : * may_start_working() %true.
2068 : : *
2069 : : * LOCKING:
2070 : : * spin_lock_irq(pool->lock) which may be released and regrabbed
2071 : : * multiple times. Does GFP_KERNEL allocations. Called only from
2072 : : * manager.
2073 : : */
2074 : 88 : static void maybe_create_worker(struct worker_pool *pool)
2075 : : __releases(&pool->lock)
2076 : : __acquires(&pool->lock)
2077 : : {
2078 : 88 : restart:
2079 : 88 : spin_unlock_irq(&pool->lock);
2080 : :
2081 : : /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */
2082 : 88 : mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT);
2083 : :
2084 : 88 : while (true) {
2085 [ - + - - ]: 88 : if (create_worker(pool) || !need_to_create_worker(pool))
2086 : : break;
2087 : :
2088 : 0 : schedule_timeout_interruptible(CREATE_COOLDOWN);
2089 : :
2090 [ # # ]: 0 : if (!need_to_create_worker(pool))
2091 : : break;
2092 : : }
2093 : :
2094 : 88 : del_timer_sync(&pool->mayday_timer);
2095 : 88 : spin_lock_irq(&pool->lock);
2096 : : /*
2097 : : * This is necessary even after a new worker was just successfully
2098 : : * created as @pool->lock was dropped and the new worker might have
2099 : : * already become busy.
2100 : : */
2101 [ - + ]: 88 : if (need_to_create_worker(pool))
2102 : 0 : goto restart;
2103 : 88 : }
2104 : :
2105 : : /**
2106 : : * manage_workers - manage worker pool
2107 : : * @worker: self
2108 : : *
2109 : : * Assume the manager role and manage the worker pool @worker belongs
2110 : : * to. At any given time, there can be only zero or one manager per
2111 : : * pool. The exclusion is handled automatically by this function.
2112 : : *
2113 : : * The caller can safely start processing works on false return. On
2114 : : * true return, it's guaranteed that need_to_create_worker() is false
2115 : : * and may_start_working() is true.
2116 : : *
2117 : : * CONTEXT:
2118 : : * spin_lock_irq(pool->lock) which may be released and regrabbed
2119 : : * multiple times. Does GFP_KERNEL allocations.
2120 : : *
2121 : : * Return:
2122 : : * %false if the pool doesn't need management and the caller can safely
2123 : : * start processing works, %true if management function was performed and
2124 : : * the conditions that the caller verified before calling the function may
2125 : : * no longer be true.
2126 : : */
2127 : 88 : static bool manage_workers(struct worker *worker)
2128 : : {
2129 : 88 : struct worker_pool *pool = worker->pool;
2130 : :
2131 [ + - ]: 88 : if (pool->flags & POOL_MANAGER_ACTIVE)
2132 : : return false;
2133 : :
2134 : 88 : pool->flags |= POOL_MANAGER_ACTIVE;
2135 : 88 : pool->manager = worker;
2136 : :
2137 : 88 : maybe_create_worker(pool);
2138 : :
2139 : 88 : pool->manager = NULL;
2140 : 88 : pool->flags &= ~POOL_MANAGER_ACTIVE;
2141 : 88 : wake_up(&wq_manager_wait);
2142 : 88 : return true;
2143 : : }
2144 : :
2145 : : /**
2146 : : * process_one_work - process single work
2147 : : * @worker: self
2148 : : * @work: work to process
2149 : : *
2150 : : * Process @work. This function contains all the logics necessary to
2151 : : * process a single work including synchronization against and
2152 : : * interaction with other workers on the same cpu, queueing and
2153 : : * flushing. As long as context requirement is met, any worker can
2154 : : * call this function to process a work.
2155 : : *
2156 : : * CONTEXT:
2157 : : * spin_lock_irq(pool->lock) which is released and regrabbed.
2158 : : */
2159 : 9215 : static void process_one_work(struct worker *worker, struct work_struct *work)
2160 : : __releases(&pool->lock)
2161 : : __acquires(&pool->lock)
2162 : : {
2163 : 9215 : struct pool_workqueue *pwq = get_work_pwq(work);
2164 : 9215 : struct worker_pool *pool = worker->pool;
2165 : 9215 : bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE;
2166 : 9215 : int work_color;
2167 : 9215 : struct worker *collision;
2168 : : #ifdef CONFIG_LOCKDEP
2169 : : /*
2170 : : * It is permissible to free the struct work_struct from
2171 : : * inside the function that is called from it, this we need to
2172 : : * take into account for lockdep too. To avoid bogus "held
2173 : : * lock freed" warnings as well as problems when looking into
2174 : : * work->lockdep_map, make a copy and use that here.
2175 : : */
2176 : : struct lockdep_map lockdep_map;
2177 : :
2178 : : lockdep_copy_map(&lockdep_map, &work->lockdep_map);
2179 : : #endif
2180 : : /* ensure we're on the correct CPU */
2181 [ + + + - : 18430 : WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) &&
- + ]
2182 : : raw_smp_processor_id() != pool->cpu);
2183 : :
2184 : : /*
2185 : : * A single work shouldn't be executed concurrently by
2186 : : * multiple workers on a single cpu. Check whether anyone is
2187 : : * already processing the work. If so, defer the work to the
2188 : : * currently executing one.
2189 : : */
2190 [ + + ]: 9215 : collision = find_worker_executing_work(pool, work);
2191 [ - + ]: 9215 : if (unlikely(collision)) {
2192 : 0 : move_linked_works(work, &collision->scheduled, NULL);
2193 : 0 : return;
2194 : : }
2195 : :
2196 : : /* claim and dequeue */
2197 : 9215 : debug_work_deactivate(work);
2198 [ + + ]: 9215 : hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work);
2199 : 9215 : worker->current_work = work;
2200 : 9215 : worker->current_func = work->func;
2201 : 9215 : worker->current_pwq = pwq;
2202 : 9215 : work_color = get_work_color(work);
2203 : :
2204 : : /*
2205 : : * Record wq name for cmdline and debug reporting, may get
2206 : : * overridden through set_worker_desc().
2207 : : */
2208 : 9215 : strscpy(worker->desc, pwq->wq->name, WORKER_DESC_LEN);
2209 : :
2210 [ - + ]: 9215 : list_del_init(&work->entry);
2211 : :
2212 : : /*
2213 : : * CPU intensive works don't participate in concurrency management.
2214 : : * They're the scheduler's responsibility. This takes @worker out
2215 : : * of concurrency management and the next code block will chain
2216 : : * execution of the pending work items.
2217 : : */
2218 [ - + ]: 9215 : if (unlikely(cpu_intensive))
2219 : 0 : worker_set_flags(worker, WORKER_CPU_INTENSIVE);
2220 : :
2221 : : /*
2222 : : * Wake up another worker if necessary. The condition is always
2223 : : * false for normal per-cpu workers since nr_running would always
2224 : : * be >= 1 at this point. This is used to chain execution of the
2225 : : * pending work items for WORKER_NOT_RUNNING workers such as the
2226 : : * UNBOUND and CPU_INTENSIVE ones.
2227 : : */
2228 [ + + + + ]: 10106 : if (need_more_worker(pool))
2229 [ + - ]: 25 : wake_up_worker(pool);
2230 : :
2231 : : /*
2232 : : * Record the last pool and clear PENDING which should be the last
2233 : : * update to @work. Also, do this inside @pool->lock so that
2234 : : * PENDING and queued state changes happen together while IRQ is
2235 : : * disabled.
2236 : : */
2237 : 9215 : set_work_pool_and_clear_pending(work, pool->id);
2238 : :
2239 : 9215 : spin_unlock_irq(&pool->lock);
2240 : :
2241 : 9215 : lock_map_acquire(&pwq->wq->lockdep_map);
2242 : 9215 : lock_map_acquire(&lockdep_map);
2243 : : /*
2244 : : * Strictly speaking we should mark the invariant state without holding
2245 : : * any locks, that is, before these two lock_map_acquire()'s.
2246 : : *
2247 : : * However, that would result in:
2248 : : *
2249 : : * A(W1)
2250 : : * WFC(C)
2251 : : * A(W1)
2252 : : * C(C)
2253 : : *
2254 : : * Which would create W1->C->W1 dependencies, even though there is no
2255 : : * actual deadlock possible. There are two solutions, using a
2256 : : * read-recursive acquire on the work(queue) 'locks', but this will then
2257 : : * hit the lockdep limitation on recursive locks, or simply discard
2258 : : * these locks.
2259 : : *
2260 : : * AFAICT there is no possible deadlock scenario between the
2261 : : * flush_work() and complete() primitives (except for single-threaded
2262 : : * workqueues), so hiding them isn't a problem.
2263 : : */
2264 : 9215 : lockdep_invariant_state(true);
2265 : 9215 : trace_workqueue_execute_start(work);
2266 : 9215 : worker->current_func(work);
2267 : : /*
2268 : : * While we must be careful to not use "work" after this, the trace
2269 : : * point will only record its address.
2270 : : */
2271 : 9215 : trace_workqueue_execute_end(work, worker->current_func);
2272 : 9215 : lock_map_release(&lockdep_map);
2273 : 9215 : lock_map_release(&pwq->wq->lockdep_map);
2274 : :
2275 [ - + ]: 9215 : if (unlikely(in_atomic() || lockdep_depth(current) > 0)) {
2276 : 0 : pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n"
2277 : : " last function: %ps\n",
2278 : : current->comm, preempt_count(), task_pid_nr(current),
2279 : : worker->current_func);
2280 : 0 : debug_show_held_locks(current);
2281 : 0 : dump_stack();
2282 : : }
2283 : :
2284 : : /*
2285 : : * The following prevents a kworker from hogging CPU on !PREEMPTION
2286 : : * kernels, where a requeueing work item waiting for something to
2287 : : * happen could deadlock with stop_machine as such work item could
2288 : : * indefinitely requeue itself while all other CPUs are trapped in
2289 : : * stop_machine. At the same time, report a quiescent RCU state so
2290 : : * the same condition doesn't freeze RCU.
2291 : : */
2292 : 9215 : cond_resched();
2293 : :
2294 : 9215 : spin_lock_irq(&pool->lock);
2295 : :
2296 : : /* clear cpu intensive status */
2297 [ - + ]: 9215 : if (unlikely(cpu_intensive))
2298 : 0 : worker_clr_flags(worker, WORKER_CPU_INTENSIVE);
2299 : :
2300 : : /* tag the worker for identification in schedule() */
2301 : 9215 : worker->last_func = worker->current_func;
2302 : :
2303 : : /* we're done with it, release */
2304 [ + - ]: 9215 : hash_del(&worker->hentry);
2305 : 9215 : worker->current_work = NULL;
2306 : 9215 : worker->current_func = NULL;
2307 : 9215 : worker->current_pwq = NULL;
2308 : 9215 : pwq_dec_nr_in_flight(pwq, work_color);
2309 : : }
2310 : :
2311 : : /**
2312 : : * process_scheduled_works - process scheduled works
2313 : : * @worker: self
2314 : : *
2315 : : * Process all scheduled works. Please note that the scheduled list
2316 : : * may change while processing a work, so this function repeatedly
2317 : : * fetches a work from the top and executes it.
2318 : : *
2319 : : * CONTEXT:
2320 : : * spin_lock_irq(pool->lock) which may be released and regrabbed
2321 : : * multiple times.
2322 : : */
2323 : : static void process_scheduled_works(struct worker *worker)
2324 : : {
2325 [ - - + + : 101 : while (!list_empty(&worker->scheduled)) {
+ + ]
2326 : 63 : struct work_struct *work = list_first_entry(&worker->scheduled,
2327 : : struct work_struct, entry);
2328 : 63 : process_one_work(worker, work);
2329 : : }
2330 : : }
2331 : :
2332 : 374 : static void set_pf_worker(bool val)
2333 : : {
2334 : 374 : mutex_lock(&wq_pool_attach_mutex);
2335 [ + + ]: 374 : if (val)
2336 : 363 : current->flags |= PF_WQ_WORKER;
2337 : : else
2338 : 11 : current->flags &= ~PF_WQ_WORKER;
2339 : 374 : mutex_unlock(&wq_pool_attach_mutex);
2340 : 374 : }
2341 : :
2342 : : /**
2343 : : * worker_thread - the worker thread function
2344 : : * @__worker: self
2345 : : *
2346 : : * The worker thread function. All workers belong to a worker_pool -
2347 : : * either a per-cpu one or dynamic unbound one. These workers process all
2348 : : * work items regardless of their specific target workqueue. The only
2349 : : * exception is work items which belong to workqueues with a rescuer which
2350 : : * will be explained in rescuer_thread().
2351 : : *
2352 : : * Return: 0
2353 : : */
2354 : 132 : static int worker_thread(void *__worker)
2355 : : {
2356 : 132 : struct worker *worker = __worker;
2357 : 132 : struct worker_pool *pool = worker->pool;
2358 : :
2359 : : /* tell the scheduler that this is a workqueue worker */
2360 : 132 : set_pf_worker(true);
2361 : 8027 : woke_up:
2362 : 8027 : spin_lock_irq(&pool->lock);
2363 : :
2364 : : /* am I supposed to die? */
2365 [ - + ]: 8027 : if (unlikely(worker->flags & WORKER_DIE)) {
2366 : 0 : spin_unlock_irq(&pool->lock);
2367 [ # # ]: 0 : WARN_ON_ONCE(!list_empty(&worker->entry));
2368 : 0 : set_pf_worker(false);
2369 : :
2370 : 0 : set_task_comm(worker->task, "kworker/dying");
2371 : 0 : ida_simple_remove(&pool->worker_ida, worker->id);
2372 : 0 : worker_detach_from_pool(worker);
2373 : 0 : kfree(worker);
2374 : 0 : return 0;
2375 : : }
2376 : :
2377 : 8027 : worker_leave_idle(worker);
2378 : 8115 : recheck:
2379 : : /* no more worker necessary? */
2380 [ + + + + ]: 15702 : if (!need_more_worker(pool))
2381 : 528 : goto sleep;
2382 : :
2383 : : /* do we need to manage? */
2384 [ + + + - ]: 7587 : if (unlikely(!may_start_working(pool)) && manage_workers(worker))
2385 : 88 : goto recheck;
2386 : :
2387 : : /*
2388 : : * ->scheduled list can only be filled while a worker is
2389 : : * preparing to process a work or actually processing it.
2390 : : * Make sure nobody diddled with it while I was sleeping.
2391 : : */
2392 [ - + ]: 7499 : WARN_ON_ONCE(!list_empty(&worker->scheduled));
2393 : :
2394 : : /*
2395 : : * Finish PREP stage. We're guaranteed to have at least one idle
2396 : : * worker or that someone else has already assumed the manager
2397 : : * role. This is where @worker starts participating in concurrency
2398 : : * management if applicable and concurrency management is restored
2399 : : * after being rebound. See rebind_workers() for details.
2400 : : */
2401 : 7499 : worker_clr_flags(worker, WORKER_PREP | WORKER_REBOUND);
2402 : :
2403 : 9177 : do {
2404 : 9177 : struct work_struct *work =
2405 : 9177 : list_first_entry(&pool->worklist,
2406 : : struct work_struct, entry);
2407 : :
2408 : 9177 : pool->watchdog_ts = jiffies;
2409 : :
2410 [ + + ]: 9177 : if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) {
2411 : : /* optimization path, not strictly necessary */
2412 : 9152 : process_one_work(worker, work);
2413 [ + + ]: 9152 : if (unlikely(!list_empty(&worker->scheduled)))
2414 : : process_scheduled_works(worker);
2415 : : } else {
2416 : 25 : move_linked_works(work, &worker->scheduled, NULL);
2417 : : process_scheduled_works(worker);
2418 : : }
2419 [ + + + + ]: 10855 : } while (keep_working(pool));
2420 : :
2421 : 7499 : worker_set_flags(worker, WORKER_PREP);
2422 : 8027 : sleep:
2423 : : /*
2424 : : * pool->lock is held and there's no work to process and no need to
2425 : : * manage, sleep. Workers are woken up only while holding
2426 : : * pool->lock or from local cpu, so setting the current state
2427 : : * before releasing pool->lock is enough to prevent losing any
2428 : : * event.
2429 : : */
2430 : 8027 : worker_enter_idle(worker);
2431 : 8027 : __set_current_state(TASK_IDLE);
2432 : 8027 : spin_unlock_irq(&pool->lock);
2433 : 8027 : schedule();
2434 : 7895 : goto woke_up;
2435 : : }
2436 : :
2437 : : /**
2438 : : * rescuer_thread - the rescuer thread function
2439 : : * @__rescuer: self
2440 : : *
2441 : : * Workqueue rescuer thread function. There's one rescuer for each
2442 : : * workqueue which has WQ_MEM_RECLAIM set.
2443 : : *
2444 : : * Regular work processing on a pool may block trying to create a new
2445 : : * worker which uses GFP_KERNEL allocation which has slight chance of
2446 : : * developing into deadlock if some works currently on the same queue
2447 : : * need to be processed to satisfy the GFP_KERNEL allocation. This is
2448 : : * the problem rescuer solves.
2449 : : *
2450 : : * When such condition is possible, the pool summons rescuers of all
2451 : : * workqueues which have works queued on the pool and let them process
2452 : : * those works so that forward progress can be guaranteed.
2453 : : *
2454 : : * This should happen rarely.
2455 : : *
2456 : : * Return: 0
2457 : : */
2458 : 231 : static int rescuer_thread(void *__rescuer)
2459 : : {
2460 : 231 : struct worker *rescuer = __rescuer;
2461 : 231 : struct workqueue_struct *wq = rescuer->rescue_wq;
2462 : 231 : struct list_head *scheduled = &rescuer->scheduled;
2463 : 231 : bool should_stop;
2464 : :
2465 : 231 : set_user_nice(current, RESCUER_NICE_LEVEL);
2466 : :
2467 : : /*
2468 : : * Mark rescuer as worker too. As WORKER_PREP is never cleared, it
2469 : : * doesn't participate in concurrency management.
2470 : : */
2471 : 231 : set_pf_worker(true);
2472 : 242 : repeat:
2473 : 242 : set_current_state(TASK_IDLE);
2474 : :
2475 : : /*
2476 : : * By the time the rescuer is requested to stop, the workqueue
2477 : : * shouldn't have any work pending, but @wq->maydays may still have
2478 : : * pwq(s) queued. This can happen by non-rescuer workers consuming
2479 : : * all the work items before the rescuer got to them. Go through
2480 : : * @wq->maydays processing before acting on should_stop so that the
2481 : : * list is always empty on exit.
2482 : : */
2483 : 242 : should_stop = kthread_should_stop();
2484 : :
2485 : : /* see whether any pwq is asking for help */
2486 : 242 : spin_lock_irq(&wq_mayday_lock);
2487 : :
2488 [ - + ]: 242 : while (!list_empty(&wq->maydays)) {
2489 : 0 : struct pool_workqueue *pwq = list_first_entry(&wq->maydays,
2490 : : struct pool_workqueue, mayday_node);
2491 : 0 : struct worker_pool *pool = pwq->pool;
2492 : 0 : struct work_struct *work, *n;
2493 : 0 : bool first = true;
2494 : :
2495 : 0 : __set_current_state(TASK_RUNNING);
2496 : 0 : list_del_init(&pwq->mayday_node);
2497 : :
2498 : 0 : spin_unlock_irq(&wq_mayday_lock);
2499 : :
2500 : 0 : worker_attach_to_pool(rescuer, pool);
2501 : :
2502 : 0 : spin_lock_irq(&pool->lock);
2503 : :
2504 : : /*
2505 : : * Slurp in all works issued via this workqueue and
2506 : : * process'em.
2507 : : */
2508 [ # # ]: 0 : WARN_ON_ONCE(!list_empty(scheduled));
2509 [ # # ]: 0 : list_for_each_entry_safe(work, n, &pool->worklist, entry) {
2510 [ # # ]: 0 : if (get_work_pwq(work) == pwq) {
2511 [ # # ]: 0 : if (first)
2512 : 0 : pool->watchdog_ts = jiffies;
2513 : 0 : move_linked_works(work, scheduled, &n);
2514 : : }
2515 : 0 : first = false;
2516 : : }
2517 : :
2518 [ # # ]: 0 : if (!list_empty(scheduled)) {
2519 : : process_scheduled_works(rescuer);
2520 : :
2521 : : /*
2522 : : * The above execution of rescued work items could
2523 : : * have created more to rescue through
2524 : : * pwq_activate_first_delayed() or chained
2525 : : * queueing. Let's put @pwq back on mayday list so
2526 : : * that such back-to-back work items, which may be
2527 : : * being used to relieve memory pressure, don't
2528 : : * incur MAYDAY_INTERVAL delay inbetween.
2529 : : */
2530 [ # # ]: 0 : if (need_to_create_worker(pool)) {
2531 : 0 : spin_lock(&wq_mayday_lock);
2532 : : /*
2533 : : * Queue iff we aren't racing destruction
2534 : : * and somebody else hasn't queued it already.
2535 : : */
2536 [ # # # # ]: 0 : if (wq->rescuer && list_empty(&pwq->mayday_node)) {
2537 [ # # ]: 0 : get_pwq(pwq);
2538 : 0 : list_add_tail(&pwq->mayday_node, &wq->maydays);
2539 : : }
2540 : 0 : spin_unlock(&wq_mayday_lock);
2541 : : }
2542 : : }
2543 : :
2544 : : /*
2545 : : * Put the reference grabbed by send_mayday(). @pool won't
2546 : : * go away while we're still attached to it.
2547 : : */
2548 : 0 : put_pwq(pwq);
2549 : :
2550 : : /*
2551 : : * Leave this pool. If need_more_worker() is %true, notify a
2552 : : * regular worker; otherwise, we end up with 0 concurrency
2553 : : * and stalling the execution.
2554 : : */
2555 [ # # # # ]: 0 : if (need_more_worker(pool))
2556 [ # # ]: 0 : wake_up_worker(pool);
2557 : :
2558 : 0 : spin_unlock_irq(&pool->lock);
2559 : :
2560 : 0 : worker_detach_from_pool(rescuer);
2561 : :
2562 : 0 : spin_lock_irq(&wq_mayday_lock);
2563 : : }
2564 : :
2565 : 242 : spin_unlock_irq(&wq_mayday_lock);
2566 : :
2567 [ + + ]: 242 : if (should_stop) {
2568 : 11 : __set_current_state(TASK_RUNNING);
2569 : 11 : set_pf_worker(false);
2570 : 11 : return 0;
2571 : : }
2572 : :
2573 : : /* rescuers should never participate in concurrency management */
2574 [ - + ]: 231 : WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING));
2575 : 231 : schedule();
2576 : 11 : goto repeat;
2577 : : }
2578 : :
2579 : : /**
2580 : : * check_flush_dependency - check for flush dependency sanity
2581 : : * @target_wq: workqueue being flushed
2582 : : * @target_work: work item being flushed (NULL for workqueue flushes)
2583 : : *
2584 : : * %current is trying to flush the whole @target_wq or @target_work on it.
2585 : : * If @target_wq doesn't have %WQ_MEM_RECLAIM, verify that %current is not
2586 : : * reclaiming memory or running on a workqueue which doesn't have
2587 : : * %WQ_MEM_RECLAIM as that can break forward-progress guarantee leading to
2588 : : * a deadlock.
2589 : : */
2590 : 38 : static void check_flush_dependency(struct workqueue_struct *target_wq,
2591 : : struct work_struct *target_work)
2592 : : {
2593 [ + - ]: 38 : work_func_t target_func = target_work ? target_work->func : NULL;
2594 : 38 : struct worker *worker;
2595 : :
2596 [ + + ]: 38 : if (target_wq->flags & WQ_MEM_RECLAIM)
2597 : : return;
2598 : :
2599 : 34 : worker = current_wq_worker();
2600 : :
2601 [ - + - - ]: 34 : WARN_ONCE(current->flags & PF_MEMALLOC,
2602 : : "workqueue: PF_MEMALLOC task %d(%s) is flushing !WQ_MEM_RECLAIM %s:%ps",
2603 : : current->pid, current->comm, target_wq->name, target_func);
2604 [ - + - - : 68 : WARN_ONCE(worker && ((worker->current_pwq->wq->flags &
- + - - ]
2605 : : (WQ_MEM_RECLAIM | __WQ_LEGACY)) == WQ_MEM_RECLAIM),
2606 : : "workqueue: WQ_MEM_RECLAIM %s:%ps is flushing !WQ_MEM_RECLAIM %s:%ps",
2607 : : worker->current_pwq->wq->name, worker->current_func,
2608 : : target_wq->name, target_func);
2609 : : }
2610 : :
2611 : : struct wq_barrier {
2612 : : struct work_struct work;
2613 : : struct completion done;
2614 : : struct task_struct *task; /* purely informational */
2615 : : };
2616 : :
2617 : 38 : static void wq_barrier_func(struct work_struct *work)
2618 : : {
2619 : 38 : struct wq_barrier *barr = container_of(work, struct wq_barrier, work);
2620 : 38 : complete(&barr->done);
2621 : 38 : }
2622 : :
2623 : : /**
2624 : : * insert_wq_barrier - insert a barrier work
2625 : : * @pwq: pwq to insert barrier into
2626 : : * @barr: wq_barrier to insert
2627 : : * @target: target work to attach @barr to
2628 : : * @worker: worker currently executing @target, NULL if @target is not executing
2629 : : *
2630 : : * @barr is linked to @target such that @barr is completed only after
2631 : : * @target finishes execution. Please note that the ordering
2632 : : * guarantee is observed only with respect to @target and on the local
2633 : : * cpu.
2634 : : *
2635 : : * Currently, a queued barrier can't be canceled. This is because
2636 : : * try_to_grab_pending() can't determine whether the work to be
2637 : : * grabbed is at the head of the queue and thus can't clear LINKED
2638 : : * flag of the previous work while there must be a valid next work
2639 : : * after a work with LINKED flag set.
2640 : : *
2641 : : * Note that when @worker is non-NULL, @target may be modified
2642 : : * underneath us, so we can't reliably determine pwq from @target.
2643 : : *
2644 : : * CONTEXT:
2645 : : * spin_lock_irq(pool->lock).
2646 : : */
2647 : 38 : static void insert_wq_barrier(struct pool_workqueue *pwq,
2648 : : struct wq_barrier *barr,
2649 : : struct work_struct *target, struct worker *worker)
2650 : : {
2651 : 38 : struct list_head *head;
2652 : 38 : unsigned int linked = 0;
2653 : :
2654 : : /*
2655 : : * debugobject calls are safe here even with pool->lock locked
2656 : : * as we know for sure that this will not trigger any of the
2657 : : * checks and call back into the fixup functions where we
2658 : : * might deadlock.
2659 : : */
2660 : 38 : INIT_WORK_ONSTACK(&barr->work, wq_barrier_func);
2661 : 38 : __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work));
2662 : :
2663 : 38 : init_completion_map(&barr->done, &target->lockdep_map);
2664 : :
2665 [ + + ]: 38 : barr->task = current;
2666 : :
2667 : : /*
2668 : : * If @target is currently being executed, schedule the
2669 : : * barrier to the worker; otherwise, put it after @target.
2670 : : */
2671 [ + + ]: 38 : if (worker)
2672 : 13 : head = worker->scheduled.next;
2673 : : else {
2674 : 25 : unsigned long *bits = work_data_bits(target);
2675 : :
2676 : 25 : head = target->entry.next;
2677 : : /* there can already be other linked works, inherit and set */
2678 : 25 : linked = *bits & WORK_STRUCT_LINKED;
2679 : 25 : __set_bit(WORK_STRUCT_LINKED_BIT, bits);
2680 : : }
2681 : :
2682 : 38 : debug_work_activate(&barr->work);
2683 : 38 : insert_work(pwq, &barr->work, head,
2684 : : work_color_to_flags(WORK_NO_COLOR) | linked);
2685 : 38 : }
2686 : :
2687 : : /**
2688 : : * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing
2689 : : * @wq: workqueue being flushed
2690 : : * @flush_color: new flush color, < 0 for no-op
2691 : : * @work_color: new work color, < 0 for no-op
2692 : : *
2693 : : * Prepare pwqs for workqueue flushing.
2694 : : *
2695 : : * If @flush_color is non-negative, flush_color on all pwqs should be
2696 : : * -1. If no pwq has in-flight commands at the specified color, all
2697 : : * pwq->flush_color's stay at -1 and %false is returned. If any pwq
2698 : : * has in flight commands, its pwq->flush_color is set to
2699 : : * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq
2700 : : * wakeup logic is armed and %true is returned.
2701 : : *
2702 : : * The caller should have initialized @wq->first_flusher prior to
2703 : : * calling this function with non-negative @flush_color. If
2704 : : * @flush_color is negative, no flush color update is done and %false
2705 : : * is returned.
2706 : : *
2707 : : * If @work_color is non-negative, all pwqs should have the same
2708 : : * work_color which is previous to @work_color and all will be
2709 : : * advanced to @work_color.
2710 : : *
2711 : : * CONTEXT:
2712 : : * mutex_lock(wq->mutex).
2713 : : *
2714 : : * Return:
2715 : : * %true if @flush_color >= 0 and there's something to flush. %false
2716 : : * otherwise.
2717 : : */
2718 : 1892 : static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq,
2719 : : int flush_color, int work_color)
2720 : : {
2721 : 1892 : bool wait = false;
2722 : 1892 : struct pool_workqueue *pwq;
2723 : :
2724 [ + - ]: 1892 : if (flush_color >= 0) {
2725 [ - + ]: 1892 : WARN_ON_ONCE(atomic_read(&wq->nr_pwqs_to_flush));
2726 : 1892 : atomic_set(&wq->nr_pwqs_to_flush, 1);
2727 : : }
2728 : :
2729 [ + + ]: 3784 : for_each_pwq(pwq, wq) {
2730 : 1892 : struct worker_pool *pool = pwq->pool;
2731 : :
2732 : 1892 : spin_lock_irq(&pool->lock);
2733 : :
2734 [ + - ]: 1892 : if (flush_color >= 0) {
2735 [ - + ]: 1892 : WARN_ON_ONCE(pwq->flush_color != -1);
2736 : :
2737 [ - + ]: 1892 : if (pwq->nr_in_flight[flush_color]) {
2738 : 0 : pwq->flush_color = flush_color;
2739 : 0 : atomic_inc(&wq->nr_pwqs_to_flush);
2740 : 0 : wait = true;
2741 : : }
2742 : : }
2743 : :
2744 [ + - ]: 1892 : if (work_color >= 0) {
2745 [ - + ]: 1892 : WARN_ON_ONCE(work_color != work_next_color(pwq->work_color));
2746 : 1892 : pwq->work_color = work_color;
2747 : : }
2748 : :
2749 : 1892 : spin_unlock_irq(&pool->lock);
2750 : : }
2751 : :
2752 [ + - + - ]: 1892 : if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_pwqs_to_flush))
2753 : 1892 : complete(&wq->first_flusher->done);
2754 : :
2755 : 1892 : return wait;
2756 : : }
2757 : :
2758 : : /**
2759 : : * flush_workqueue - ensure that any scheduled work has run to completion.
2760 : : * @wq: workqueue to flush
2761 : : *
2762 : : * This function sleeps until all work items which were queued on entry
2763 : : * have finished execution, but it is not livelocked by new incoming ones.
2764 : : */
2765 : 1892 : void flush_workqueue(struct workqueue_struct *wq)
2766 : : {
2767 : 3784 : struct wq_flusher this_flusher = {
2768 : : .list = LIST_HEAD_INIT(this_flusher.list),
2769 : : .flush_color = -1,
2770 : 1892 : .done = COMPLETION_INITIALIZER_ONSTACK_MAP(this_flusher.done, wq->lockdep_map),
2771 : : };
2772 : 1892 : int next_color;
2773 : :
2774 [ - + + - ]: 1892 : if (WARN_ON(!wq_online))
2775 : 0 : return;
2776 : :
2777 : 1892 : lock_map_acquire(&wq->lockdep_map);
2778 : 1892 : lock_map_release(&wq->lockdep_map);
2779 : :
2780 : 1892 : mutex_lock(&wq->mutex);
2781 : :
2782 : : /*
2783 : : * Start-to-wait phase
2784 : : */
2785 : 1892 : next_color = work_next_color(wq->work_color);
2786 : :
2787 [ + - ]: 1892 : if (next_color != wq->flush_color) {
2788 : : /*
2789 : : * Color space is not full. The current work_color
2790 : : * becomes our flush_color and work_color is advanced
2791 : : * by one.
2792 : : */
2793 [ - + ]: 1892 : WARN_ON_ONCE(!list_empty(&wq->flusher_overflow));
2794 : 1892 : this_flusher.flush_color = wq->work_color;
2795 : 1892 : wq->work_color = next_color;
2796 : :
2797 [ + - ]: 1892 : if (!wq->first_flusher) {
2798 : : /* no flush in progress, become the first flusher */
2799 [ - + ]: 1892 : WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color);
2800 : :
2801 : 1892 : wq->first_flusher = &this_flusher;
2802 : :
2803 [ + - ]: 1892 : if (!flush_workqueue_prep_pwqs(wq, wq->flush_color,
2804 : : wq->work_color)) {
2805 : : /* nothing to flush, done */
2806 : 1892 : wq->flush_color = next_color;
2807 : 1892 : wq->first_flusher = NULL;
2808 : 1892 : goto out_unlock;
2809 : : }
2810 : : } else {
2811 : : /* wait in queue */
2812 [ # # ]: 0 : WARN_ON_ONCE(wq->flush_color == this_flusher.flush_color);
2813 : 0 : list_add_tail(&this_flusher.list, &wq->flusher_queue);
2814 : 0 : flush_workqueue_prep_pwqs(wq, -1, wq->work_color);
2815 : : }
2816 : : } else {
2817 : : /*
2818 : : * Oops, color space is full, wait on overflow queue.
2819 : : * The next flush completion will assign us
2820 : : * flush_color and transfer to flusher_queue.
2821 : : */
2822 : 0 : list_add_tail(&this_flusher.list, &wq->flusher_overflow);
2823 : : }
2824 : :
2825 : 0 : check_flush_dependency(wq, NULL);
2826 : :
2827 : 0 : mutex_unlock(&wq->mutex);
2828 : :
2829 : 0 : wait_for_completion(&this_flusher.done);
2830 : :
2831 : : /*
2832 : : * Wake-up-and-cascade phase
2833 : : *
2834 : : * First flushers are responsible for cascading flushes and
2835 : : * handling overflow. Non-first flushers can simply return.
2836 : : */
2837 [ # # ]: 0 : if (wq->first_flusher != &this_flusher)
2838 : : return;
2839 : :
2840 : 0 : mutex_lock(&wq->mutex);
2841 : :
2842 : : /* we might have raced, check again with mutex held */
2843 [ # # ]: 0 : if (wq->first_flusher != &this_flusher)
2844 : 0 : goto out_unlock;
2845 : :
2846 : 0 : wq->first_flusher = NULL;
2847 : :
2848 [ # # ]: 0 : WARN_ON_ONCE(!list_empty(&this_flusher.list));
2849 [ # # ]: 0 : WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color);
2850 : :
2851 : 0 : while (true) {
2852 : 0 : struct wq_flusher *next, *tmp;
2853 : :
2854 : : /* complete all the flushers sharing the current flush color */
2855 [ # # ]: 0 : list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) {
2856 [ # # ]: 0 : if (next->flush_color != wq->flush_color)
2857 : : break;
2858 : 0 : list_del_init(&next->list);
2859 : 0 : complete(&next->done);
2860 : : }
2861 : :
2862 [ # # # # : 0 : WARN_ON_ONCE(!list_empty(&wq->flusher_overflow) &&
# # ]
2863 : : wq->flush_color != work_next_color(wq->work_color));
2864 : :
2865 : : /* this flush_color is finished, advance by one */
2866 : 0 : wq->flush_color = work_next_color(wq->flush_color);
2867 : :
2868 : : /* one color has been freed, handle overflow queue */
2869 [ # # ]: 0 : if (!list_empty(&wq->flusher_overflow)) {
2870 : : /*
2871 : : * Assign the same color to all overflowed
2872 : : * flushers, advance work_color and append to
2873 : : * flusher_queue. This is the start-to-wait
2874 : : * phase for these overflowed flushers.
2875 : : */
2876 [ # # ]: 0 : list_for_each_entry(tmp, &wq->flusher_overflow, list)
2877 : 0 : tmp->flush_color = wq->work_color;
2878 : :
2879 : 0 : wq->work_color = work_next_color(wq->work_color);
2880 : :
2881 [ # # ]: 0 : list_splice_tail_init(&wq->flusher_overflow,
2882 : : &wq->flusher_queue);
2883 : 0 : flush_workqueue_prep_pwqs(wq, -1, wq->work_color);
2884 : : }
2885 : :
2886 [ # # ]: 0 : if (list_empty(&wq->flusher_queue)) {
2887 [ # # ]: 0 : WARN_ON_ONCE(wq->flush_color != wq->work_color);
2888 : : break;
2889 : : }
2890 : :
2891 : : /*
2892 : : * Need to flush more colors. Make the next flusher
2893 : : * the new first flusher and arm pwqs.
2894 : : */
2895 [ # # ]: 0 : WARN_ON_ONCE(wq->flush_color == wq->work_color);
2896 [ # # ]: 0 : WARN_ON_ONCE(wq->flush_color != next->flush_color);
2897 : :
2898 : 0 : list_del_init(&next->list);
2899 : 0 : wq->first_flusher = next;
2900 : :
2901 [ # # ]: 0 : if (flush_workqueue_prep_pwqs(wq, wq->flush_color, -1))
2902 : : break;
2903 : :
2904 : : /*
2905 : : * Meh... this color is already done, clear first
2906 : : * flusher and repeat cascading.
2907 : : */
2908 : 0 : wq->first_flusher = NULL;
2909 : : }
2910 : :
2911 : 0 : out_unlock:
2912 : 1892 : mutex_unlock(&wq->mutex);
2913 : : }
2914 : : EXPORT_SYMBOL(flush_workqueue);
2915 : :
2916 : : /**
2917 : : * drain_workqueue - drain a workqueue
2918 : : * @wq: workqueue to drain
2919 : : *
2920 : : * Wait until the workqueue becomes empty. While draining is in progress,
2921 : : * only chain queueing is allowed. IOW, only currently pending or running
2922 : : * work items on @wq can queue further work items on it. @wq is flushed
2923 : : * repeatedly until it becomes empty. The number of flushing is determined
2924 : : * by the depth of chaining and should be relatively short. Whine if it
2925 : : * takes too long.
2926 : : */
2927 : 11 : void drain_workqueue(struct workqueue_struct *wq)
2928 : : {
2929 : 11 : unsigned int flush_cnt = 0;
2930 : 11 : struct pool_workqueue *pwq;
2931 : :
2932 : : /*
2933 : : * __queue_work() needs to test whether there are drainers, is much
2934 : : * hotter than drain_workqueue() and already looks at @wq->flags.
2935 : : * Use __WQ_DRAINING so that queue doesn't have to check nr_drainers.
2936 : : */
2937 : 11 : mutex_lock(&wq->mutex);
2938 [ + - ]: 11 : if (!wq->nr_drainers++)
2939 : 11 : wq->flags |= __WQ_DRAINING;
2940 : 11 : mutex_unlock(&wq->mutex);
2941 : 11 : reflush:
2942 : 11 : flush_workqueue(wq);
2943 : :
2944 : 11 : mutex_lock(&wq->mutex);
2945 : :
2946 [ + + ]: 22 : for_each_pwq(pwq, wq) {
2947 : 11 : bool drained;
2948 : :
2949 : 11 : spin_lock_irq(&pwq->pool->lock);
2950 [ + - - + ]: 11 : drained = !pwq->nr_active && list_empty(&pwq->delayed_works);
2951 : 11 : spin_unlock_irq(&pwq->pool->lock);
2952 : :
2953 [ + - ]: 11 : if (drained)
2954 : 11 : continue;
2955 : :
2956 [ # # ]: 0 : if (++flush_cnt == 10 ||
2957 [ # # # # ]: 0 : (flush_cnt % 100 == 0 && flush_cnt <= 1000))
2958 : 0 : pr_warn("workqueue %s: drain_workqueue() isn't complete after %u tries\n",
2959 : : wq->name, flush_cnt);
2960 : :
2961 : 0 : mutex_unlock(&wq->mutex);
2962 : 0 : goto reflush;
2963 : : }
2964 : :
2965 [ + - ]: 11 : if (!--wq->nr_drainers)
2966 : 11 : wq->flags &= ~__WQ_DRAINING;
2967 : 11 : mutex_unlock(&wq->mutex);
2968 : 11 : }
2969 : : EXPORT_SYMBOL_GPL(drain_workqueue);
2970 : :
2971 : 3751 : static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr,
2972 : : bool from_cancel)
2973 : : {
2974 : 3751 : struct worker *worker = NULL;
2975 : 3751 : struct worker_pool *pool;
2976 : 3751 : struct pool_workqueue *pwq;
2977 : :
2978 : 3751 : might_sleep();
2979 : :
2980 : 3751 : rcu_read_lock();
2981 : 3751 : pool = get_work_pool(work);
2982 [ + + ]: 3751 : if (!pool) {
2983 : 3322 : rcu_read_unlock();
2984 : 3322 : return false;
2985 : : }
2986 : :
2987 : 429 : spin_lock_irq(&pool->lock);
2988 : : /* see the comment in try_to_grab_pending() with the same code */
2989 : 429 : pwq = get_work_pwq(work);
2990 [ + - ]: 25 : if (pwq) {
2991 [ - + ]: 25 : if (unlikely(pwq->pool != pool))
2992 : 0 : goto already_gone;
2993 : : } else {
2994 [ + + ]: 404 : worker = find_worker_executing_work(pool, work);
2995 [ + + ]: 404 : if (!worker)
2996 : 391 : goto already_gone;
2997 : 13 : pwq = worker->current_pwq;
2998 : : }
2999 : :
3000 : 38 : check_flush_dependency(pwq->wq, work);
3001 : :
3002 : 38 : insert_wq_barrier(pwq, barr, work, worker);
3003 : 38 : spin_unlock_irq(&pool->lock);
3004 : :
3005 : : /*
3006 : : * Force a lock recursion deadlock when using flush_work() inside a
3007 : : * single-threaded or rescuer equipped workqueue.
3008 : : *
3009 : : * For single threaded workqueues the deadlock happens when the work
3010 : : * is after the work issuing the flush_work(). For rescuer equipped
3011 : : * workqueues the deadlock happens when the rescuer stalls, blocking
3012 : : * forward progress.
3013 : : */
3014 : 38 : if (!from_cancel &&
3015 : : (pwq->wq->saved_max_active == 1 || pwq->wq->rescuer)) {
3016 : 38 : lock_map_acquire(&pwq->wq->lockdep_map);
3017 : 38 : lock_map_release(&pwq->wq->lockdep_map);
3018 : : }
3019 : 38 : rcu_read_unlock();
3020 : 38 : return true;
3021 : 391 : already_gone:
3022 : 391 : spin_unlock_irq(&pool->lock);
3023 : 391 : rcu_read_unlock();
3024 : 391 : return false;
3025 : : }
3026 : :
3027 : 3751 : static bool __flush_work(struct work_struct *work, bool from_cancel)
3028 : : {
3029 : 3751 : struct wq_barrier barr;
3030 : :
3031 [ - + + - ]: 3751 : if (WARN_ON(!wq_online))
3032 : : return false;
3033 : :
3034 [ - + + - ]: 3751 : if (WARN_ON(!work->func))
3035 : : return false;
3036 : :
3037 : 3751 : if (!from_cancel) {
3038 : 3751 : lock_map_acquire(&work->lockdep_map);
3039 : 3751 : lock_map_release(&work->lockdep_map);
3040 : : }
3041 : :
3042 [ + + ]: 3751 : if (start_flush_work(work, &barr, from_cancel)) {
3043 : 38 : wait_for_completion(&barr.done);
3044 : 38 : destroy_work_on_stack(&barr.work);
3045 : 38 : return true;
3046 : : } else {
3047 : : return false;
3048 : : }
3049 : : }
3050 : :
3051 : : /**
3052 : : * flush_work - wait for a work to finish executing the last queueing instance
3053 : : * @work: the work to flush
3054 : : *
3055 : : * Wait until @work has finished execution. @work is guaranteed to be idle
3056 : : * on return if it hasn't been requeued since flush started.
3057 : : *
3058 : : * Return:
3059 : : * %true if flush_work() waited for the work to finish execution,
3060 : : * %false if it was already idle.
3061 : : */
3062 : 2046 : bool flush_work(struct work_struct *work)
3063 : : {
3064 : 2024 : return __flush_work(work, false);
3065 : : }
3066 : : EXPORT_SYMBOL_GPL(flush_work);
3067 : :
3068 : : struct cwt_wait {
3069 : : wait_queue_entry_t wait;
3070 : : struct work_struct *work;
3071 : : };
3072 : :
3073 : 0 : static int cwt_wakefn(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
3074 : : {
3075 : 0 : struct cwt_wait *cwait = container_of(wait, struct cwt_wait, wait);
3076 : :
3077 [ # # ]: 0 : if (cwait->work != key)
3078 : : return 0;
3079 : 0 : return autoremove_wake_function(wait, mode, sync, key);
3080 : : }
3081 : :
3082 : 1705 : static bool __cancel_work_timer(struct work_struct *work, bool is_dwork)
3083 : : {
3084 : 1705 : static DECLARE_WAIT_QUEUE_HEAD(cancel_waitq);
3085 : 1705 : unsigned long flags;
3086 : 1705 : int ret;
3087 : :
3088 : 1705 : do {
3089 : 1705 : ret = try_to_grab_pending(work, is_dwork, &flags);
3090 : : /*
3091 : : * If someone else is already canceling, wait for it to
3092 : : * finish. flush_work() doesn't work for PREEMPT_NONE
3093 : : * because we may get scheduled between @work's completion
3094 : : * and the other canceling task resuming and clearing
3095 : : * CANCELING - flush_work() will return false immediately
3096 : : * as @work is no longer busy, try_to_grab_pending() will
3097 : : * return -ENOENT as @work is still being canceled and the
3098 : : * other canceling task won't be able to clear CANCELING as
3099 : : * we're hogging the CPU.
3100 : : *
3101 : : * Let's wait for completion using a waitqueue. As this
3102 : : * may lead to the thundering herd problem, use a custom
3103 : : * wake function which matches @work along with exclusive
3104 : : * wait and wakeup.
3105 : : */
3106 [ - + ]: 1705 : if (unlikely(ret == -ENOENT)) {
3107 : 0 : struct cwt_wait cwait;
3108 : :
3109 : 0 : init_wait(&cwait.wait);
3110 : 0 : cwait.wait.func = cwt_wakefn;
3111 : 0 : cwait.work = work;
3112 : :
3113 : 0 : prepare_to_wait_exclusive(&cancel_waitq, &cwait.wait,
3114 : : TASK_UNINTERRUPTIBLE);
3115 [ # # ]: 0 : if (work_is_canceling(work))
3116 : 0 : schedule();
3117 : 0 : finish_wait(&cancel_waitq, &cwait.wait);
3118 : : }
3119 [ - + ]: 1705 : } while (unlikely(ret < 0));
3120 : :
3121 : : /* tell other tasks trying to grab @work to back off */
3122 : 1705 : mark_work_canceling(work);
3123 : 1705 : local_irq_restore(flags);
3124 : :
3125 : : /*
3126 : : * This allows canceling during early boot. We know that @work
3127 : : * isn't executing.
3128 : : */
3129 [ + - ]: 1705 : if (wq_online)
3130 : 1705 : __flush_work(work, true);
3131 : :
3132 : 1705 : clear_work_data(work);
3133 : :
3134 : : /*
3135 : : * Paired with prepare_to_wait() above so that either
3136 : : * waitqueue_active() is visible here or !work_is_canceling() is
3137 : : * visible there.
3138 : : */
3139 : 1705 : smp_mb();
3140 [ - + ]: 1705 : if (waitqueue_active(&cancel_waitq))
3141 : 0 : __wake_up(&cancel_waitq, TASK_NORMAL, 1, work);
3142 : :
3143 : 1705 : return ret;
3144 : : }
3145 : :
3146 : : /**
3147 : : * cancel_work_sync - cancel a work and wait for it to finish
3148 : : * @work: the work to cancel
3149 : : *
3150 : : * Cancel @work and wait for its execution to finish. This function
3151 : : * can be used even if the work re-queues itself or migrates to
3152 : : * another workqueue. On return from this function, @work is
3153 : : * guaranteed to be not pending or executing on any CPU.
3154 : : *
3155 : : * cancel_work_sync(&delayed_work->work) must not be used for
3156 : : * delayed_work's. Use cancel_delayed_work_sync() instead.
3157 : : *
3158 : : * The caller must ensure that the workqueue on which @work was last
3159 : : * queued can't be destroyed before this function returns.
3160 : : *
3161 : : * Return:
3162 : : * %true if @work was pending, %false otherwise.
3163 : : */
3164 : 1309 : bool cancel_work_sync(struct work_struct *work)
3165 : : {
3166 : 1309 : return __cancel_work_timer(work, false);
3167 : : }
3168 : : EXPORT_SYMBOL_GPL(cancel_work_sync);
3169 : :
3170 : : /**
3171 : : * flush_delayed_work - wait for a dwork to finish executing the last queueing
3172 : : * @dwork: the delayed work to flush
3173 : : *
3174 : : * Delayed timer is cancelled and the pending work is queued for
3175 : : * immediate execution. Like flush_work(), this function only
3176 : : * considers the last queueing instance of @dwork.
3177 : : *
3178 : : * Return:
3179 : : * %true if flush_work() waited for the work to finish execution,
3180 : : * %false if it was already idle.
3181 : : */
3182 : 22 : bool flush_delayed_work(struct delayed_work *dwork)
3183 : : {
3184 : 22 : local_irq_disable();
3185 [ + - ]: 22 : if (del_timer_sync(&dwork->timer))
3186 : 22 : __queue_work(dwork->cpu, dwork->wq, &dwork->work);
3187 : 22 : local_irq_enable();
3188 : 22 : return flush_work(&dwork->work);
3189 : : }
3190 : : EXPORT_SYMBOL(flush_delayed_work);
3191 : :
3192 : : /**
3193 : : * flush_rcu_work - wait for a rwork to finish executing the last queueing
3194 : : * @rwork: the rcu work to flush
3195 : : *
3196 : : * Return:
3197 : : * %true if flush_rcu_work() waited for the work to finish execution,
3198 : : * %false if it was already idle.
3199 : : */
3200 : 0 : bool flush_rcu_work(struct rcu_work *rwork)
3201 : : {
3202 [ # # ]: 0 : if (test_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&rwork->work))) {
3203 : 0 : rcu_barrier();
3204 : 0 : flush_work(&rwork->work);
3205 : 0 : return true;
3206 : : } else {
3207 : 0 : return flush_work(&rwork->work);
3208 : : }
3209 : : }
3210 : : EXPORT_SYMBOL(flush_rcu_work);
3211 : :
3212 : 44 : static bool __cancel_work(struct work_struct *work, bool is_dwork)
3213 : : {
3214 : 44 : unsigned long flags;
3215 : 44 : int ret;
3216 : :
3217 : 44 : do {
3218 : 44 : ret = try_to_grab_pending(work, is_dwork, &flags);
3219 [ - + ]: 44 : } while (unlikely(ret == -EAGAIN));
3220 : :
3221 [ + - ]: 44 : if (unlikely(ret < 0))
3222 : : return false;
3223 : :
3224 : 44 : set_work_pool_and_clear_pending(work, get_work_pool_id(work));
3225 : 44 : local_irq_restore(flags);
3226 : 44 : return ret;
3227 : : }
3228 : :
3229 : : /**
3230 : : * cancel_delayed_work - cancel a delayed work
3231 : : * @dwork: delayed_work to cancel
3232 : : *
3233 : : * Kill off a pending delayed_work.
3234 : : *
3235 : : * Return: %true if @dwork was pending and canceled; %false if it wasn't
3236 : : * pending.
3237 : : *
3238 : : * Note:
3239 : : * The work callback function may still be running on return, unless
3240 : : * it returns %true and the work doesn't re-arm itself. Explicitly flush or
3241 : : * use cancel_delayed_work_sync() to wait on it.
3242 : : *
3243 : : * This function is safe to call from any context including IRQ handler.
3244 : : */
3245 : 44 : bool cancel_delayed_work(struct delayed_work *dwork)
3246 : : {
3247 : 44 : return __cancel_work(&dwork->work, true);
3248 : : }
3249 : : EXPORT_SYMBOL(cancel_delayed_work);
3250 : :
3251 : : /**
3252 : : * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish
3253 : : * @dwork: the delayed work cancel
3254 : : *
3255 : : * This is cancel_work_sync() for delayed works.
3256 : : *
3257 : : * Return:
3258 : : * %true if @dwork was pending, %false otherwise.
3259 : : */
3260 : 396 : bool cancel_delayed_work_sync(struct delayed_work *dwork)
3261 : : {
3262 : 396 : return __cancel_work_timer(&dwork->work, true);
3263 : : }
3264 : : EXPORT_SYMBOL(cancel_delayed_work_sync);
3265 : :
3266 : : /**
3267 : : * schedule_on_each_cpu - execute a function synchronously on each online CPU
3268 : : * @func: the function to call
3269 : : *
3270 : : * schedule_on_each_cpu() executes @func on each online CPU using the
3271 : : * system workqueue and blocks until all CPUs have completed.
3272 : : * schedule_on_each_cpu() is very slow.
3273 : : *
3274 : : * Return:
3275 : : * 0 on success, -errno on failure.
3276 : : */
3277 : 0 : int schedule_on_each_cpu(work_func_t func)
3278 : : {
3279 : 0 : int cpu;
3280 : 0 : struct work_struct __percpu *works;
3281 : :
3282 : 0 : works = alloc_percpu(struct work_struct);
3283 [ # # ]: 0 : if (!works)
3284 : : return -ENOMEM;
3285 : :
3286 : 0 : get_online_cpus();
3287 : :
3288 [ # # ]: 0 : for_each_online_cpu(cpu) {
3289 : 0 : struct work_struct *work = per_cpu_ptr(works, cpu);
3290 : :
3291 : 0 : INIT_WORK(work, func);
3292 : 0 : schedule_work_on(cpu, work);
3293 : : }
3294 : :
3295 [ # # ]: 0 : for_each_online_cpu(cpu)
3296 : 0 : flush_work(per_cpu_ptr(works, cpu));
3297 : :
3298 : 0 : put_online_cpus();
3299 : 0 : free_percpu(works);
3300 : 0 : return 0;
3301 : : }
3302 : :
3303 : : /**
3304 : : * execute_in_process_context - reliably execute the routine with user context
3305 : : * @fn: the function to execute
3306 : : * @ew: guaranteed storage for the execute work structure (must
3307 : : * be available when the work executes)
3308 : : *
3309 : : * Executes the function immediately if process context is available,
3310 : : * otherwise schedules the function for delayed execution.
3311 : : *
3312 : : * Return: 0 - function was executed
3313 : : * 1 - function was scheduled for execution
3314 : : */
3315 : 0 : int execute_in_process_context(work_func_t fn, struct execute_work *ew)
3316 : : {
3317 [ # # ]: 0 : if (!in_interrupt()) {
3318 : 0 : fn(&ew->work);
3319 : 0 : return 0;
3320 : : }
3321 : :
3322 : 0 : INIT_WORK(&ew->work, fn);
3323 : 0 : schedule_work(&ew->work);
3324 : :
3325 : 0 : return 1;
3326 : : }
3327 : : EXPORT_SYMBOL_GPL(execute_in_process_context);
3328 : :
3329 : : /**
3330 : : * free_workqueue_attrs - free a workqueue_attrs
3331 : : * @attrs: workqueue_attrs to free
3332 : : *
3333 : : * Undo alloc_workqueue_attrs().
3334 : : */
3335 : 429 : void free_workqueue_attrs(struct workqueue_attrs *attrs)
3336 : : {
3337 [ # # ]: 0 : if (attrs) {
3338 : 220 : free_cpumask_var(attrs->cpumask);
3339 : 220 : kfree(attrs);
3340 : : }
3341 : 0 : }
3342 : :
3343 : : /**
3344 : : * alloc_workqueue_attrs - allocate a workqueue_attrs
3345 : : *
3346 : : * Allocate a new workqueue_attrs, initialize with default settings and
3347 : : * return it.
3348 : : *
3349 : : * Return: The allocated new workqueue_attr on success. %NULL on failure.
3350 : : */
3351 : 715 : struct workqueue_attrs *alloc_workqueue_attrs(void)
3352 : : {
3353 : 715 : struct workqueue_attrs *attrs;
3354 : :
3355 : 715 : attrs = kzalloc(sizeof(*attrs), GFP_KERNEL);
3356 [ - + ]: 715 : if (!attrs)
3357 : 0 : goto fail;
3358 : 715 : if (!alloc_cpumask_var(&attrs->cpumask, GFP_KERNEL))
3359 : : goto fail;
3360 : :
3361 : 715 : cpumask_copy(attrs->cpumask, cpu_possible_mask);
3362 : 715 : return attrs;
3363 : : fail:
3364 : 0 : free_workqueue_attrs(attrs);
3365 : 0 : return NULL;
3366 : : }
3367 : :
3368 : 858 : static void copy_workqueue_attrs(struct workqueue_attrs *to,
3369 : : const struct workqueue_attrs *from)
3370 : : {
3371 : 858 : to->nice = from->nice;
3372 : 858 : cpumask_copy(to->cpumask, from->cpumask);
3373 : : /*
3374 : : * Unlike hash and equality test, this function doesn't ignore
3375 : : * ->no_numa as it is used for both pool and wq attrs. Instead,
3376 : : * get_unbound_pool() explicitly clears ->no_numa after copying.
3377 : : */
3378 : 858 : to->no_numa = from->no_numa;
3379 : : }
3380 : :
3381 : : /* hash value of the content of @attr */
3382 : 209 : static u32 wqattrs_hash(const struct workqueue_attrs *attrs)
3383 : : {
3384 : 209 : u32 hash = 0;
3385 : :
3386 : 209 : hash = jhash_1word(attrs->nice, hash);
3387 : 209 : hash = jhash(cpumask_bits(attrs->cpumask),
3388 : : BITS_TO_LONGS(nr_cpumask_bits) * sizeof(long), hash);
3389 : 209 : return hash;
3390 : : }
3391 : :
3392 : : /* content equality test */
3393 : 187 : static bool wqattrs_equal(const struct workqueue_attrs *a,
3394 : : const struct workqueue_attrs *b)
3395 : : {
3396 : 187 : if (a->nice != b->nice)
3397 : : return false;
3398 [ + - ]: 187 : if (!cpumask_equal(a->cpumask, b->cpumask))
3399 : : return false;
3400 : : return true;
3401 : : }
3402 : :
3403 : : /**
3404 : : * init_worker_pool - initialize a newly zalloc'd worker_pool
3405 : : * @pool: worker_pool to initialize
3406 : : *
3407 : : * Initialize a newly zalloc'd @pool. It also allocates @pool->attrs.
3408 : : *
3409 : : * Return: 0 on success, -errno on failure. Even on failure, all fields
3410 : : * inside @pool proper are initialized and put_unbound_pool() can be called
3411 : : * on @pool safely to release it.
3412 : : */
3413 : 44 : static int init_worker_pool(struct worker_pool *pool)
3414 : : {
3415 : 44 : spin_lock_init(&pool->lock);
3416 : 44 : pool->id = -1;
3417 : 44 : pool->cpu = -1;
3418 : 44 : pool->node = NUMA_NO_NODE;
3419 : 44 : pool->flags |= POOL_DISASSOCIATED;
3420 : 44 : pool->watchdog_ts = jiffies;
3421 : 44 : INIT_LIST_HEAD(&pool->worklist);
3422 : 44 : INIT_LIST_HEAD(&pool->idle_list);
3423 : 44 : hash_init(pool->busy_hash);
3424 : :
3425 : 44 : timer_setup(&pool->idle_timer, idle_worker_timeout, TIMER_DEFERRABLE);
3426 : :
3427 : 44 : timer_setup(&pool->mayday_timer, pool_mayday_timeout, 0);
3428 : :
3429 : 44 : INIT_LIST_HEAD(&pool->workers);
3430 : :
3431 : 44 : ida_init(&pool->worker_ida);
3432 : 44 : INIT_HLIST_NODE(&pool->hash_node);
3433 : 44 : pool->refcnt = 1;
3434 : :
3435 : : /* shouldn't fail above this point */
3436 : 44 : pool->attrs = alloc_workqueue_attrs();
3437 [ - + ]: 44 : if (!pool->attrs)
3438 : 0 : return -ENOMEM;
3439 : : return 0;
3440 : : }
3441 : :
3442 : : #ifdef CONFIG_LOCKDEP
3443 : : static void wq_init_lockdep(struct workqueue_struct *wq)
3444 : : {
3445 : : char *lock_name;
3446 : :
3447 : : lockdep_register_key(&wq->key);
3448 : : lock_name = kasprintf(GFP_KERNEL, "%s%s", "(wq_completion)", wq->name);
3449 : : if (!lock_name)
3450 : : lock_name = wq->name;
3451 : :
3452 : : wq->lock_name = lock_name;
3453 : : lockdep_init_map(&wq->lockdep_map, lock_name, &wq->key, 0);
3454 : : }
3455 : :
3456 : : static void wq_unregister_lockdep(struct workqueue_struct *wq)
3457 : : {
3458 : : lockdep_unregister_key(&wq->key);
3459 : : }
3460 : :
3461 : : static void wq_free_lockdep(struct workqueue_struct *wq)
3462 : : {
3463 : : if (wq->lock_name != wq->name)
3464 : : kfree(wq->lock_name);
3465 : : }
3466 : : #else
3467 : 484 : static void wq_init_lockdep(struct workqueue_struct *wq)
3468 : : {
3469 : 484 : }
3470 : :
3471 : 11 : static void wq_unregister_lockdep(struct workqueue_struct *wq)
3472 : : {
3473 : 11 : }
3474 : :
3475 : 11 : static void wq_free_lockdep(struct workqueue_struct *wq)
3476 : : {
3477 : 11 : }
3478 : : #endif
3479 : :
3480 : 11 : static void rcu_free_wq(struct rcu_head *rcu)
3481 : : {
3482 : 11 : struct workqueue_struct *wq =
3483 : 11 : container_of(rcu, struct workqueue_struct, rcu);
3484 : :
3485 : 11 : wq_free_lockdep(wq);
3486 : :
3487 [ - + ]: 11 : if (!(wq->flags & WQ_UNBOUND))
3488 : 0 : free_percpu(wq->cpu_pwqs);
3489 : : else
3490 [ + - ]: 11 : free_workqueue_attrs(wq->unbound_attrs);
3491 : :
3492 : 11 : kfree(wq->rescuer);
3493 : 11 : kfree(wq);
3494 : 11 : }
3495 : :
3496 : 0 : static void rcu_free_pool(struct rcu_head *rcu)
3497 : : {
3498 : 0 : struct worker_pool *pool = container_of(rcu, struct worker_pool, rcu);
3499 : :
3500 : 0 : ida_destroy(&pool->worker_ida);
3501 [ # # ]: 0 : free_workqueue_attrs(pool->attrs);
3502 : 0 : kfree(pool);
3503 : 0 : }
3504 : :
3505 : : /**
3506 : : * put_unbound_pool - put a worker_pool
3507 : : * @pool: worker_pool to put
3508 : : *
3509 : : * Put @pool. If its refcnt reaches zero, it gets destroyed in RCU
3510 : : * safe manner. get_unbound_pool() calls this function on its failure path
3511 : : * and this function should be able to release pools which went through,
3512 : : * successfully or not, init_worker_pool().
3513 : : *
3514 : : * Should be called with wq_pool_mutex held.
3515 : : */
3516 : 11 : static void put_unbound_pool(struct worker_pool *pool)
3517 : : {
3518 : 11 : DECLARE_COMPLETION_ONSTACK(detach_completion);
3519 : 11 : struct worker *worker;
3520 : :
3521 : 11 : lockdep_assert_held(&wq_pool_mutex);
3522 : :
3523 [ - + ]: 11 : if (--pool->refcnt)
3524 : 11 : return;
3525 : :
3526 : : /* sanity checks */
3527 [ # # # # ]: 0 : if (WARN_ON(!(pool->cpu < 0)) ||
3528 [ # # # # ]: 0 : WARN_ON(!list_empty(&pool->worklist)))
3529 : : return;
3530 : :
3531 : : /* release id and unhash */
3532 [ # # ]: 0 : if (pool->id >= 0)
3533 : 0 : idr_remove(&worker_pool_idr, pool->id);
3534 [ # # ]: 0 : hash_del(&pool->hash_node);
3535 : :
3536 : : /*
3537 : : * Become the manager and destroy all workers. This prevents
3538 : : * @pool's workers from blocking on attach_mutex. We're the last
3539 : : * manager and @pool gets freed with the flag set.
3540 : : */
3541 : 0 : spin_lock_irq(&pool->lock);
3542 [ # # # # ]: 0 : wait_event_lock_irq(wq_manager_wait,
3543 : : !(pool->flags & POOL_MANAGER_ACTIVE), pool->lock);
3544 : 0 : pool->flags |= POOL_MANAGER_ACTIVE;
3545 : :
3546 [ # # # # ]: 0 : while ((worker = first_idle_worker(pool)))
3547 : 0 : destroy_worker(worker);
3548 [ # # ]: 0 : WARN_ON(pool->nr_workers || pool->nr_idle);
3549 : 0 : spin_unlock_irq(&pool->lock);
3550 : :
3551 : 0 : mutex_lock(&wq_pool_attach_mutex);
3552 [ # # ]: 0 : if (!list_empty(&pool->workers))
3553 : 0 : pool->detach_completion = &detach_completion;
3554 : 0 : mutex_unlock(&wq_pool_attach_mutex);
3555 : :
3556 [ # # ]: 0 : if (pool->detach_completion)
3557 : 0 : wait_for_completion(pool->detach_completion);
3558 : :
3559 : : /* shut down the timers */
3560 : 0 : del_timer_sync(&pool->idle_timer);
3561 : 0 : del_timer_sync(&pool->mayday_timer);
3562 : :
3563 : : /* RCU protected to allow dereferences from get_work_pool() */
3564 : 0 : call_rcu(&pool->rcu, rcu_free_pool);
3565 : : }
3566 : :
3567 : : /**
3568 : : * get_unbound_pool - get a worker_pool with the specified attributes
3569 : : * @attrs: the attributes of the worker_pool to get
3570 : : *
3571 : : * Obtain a worker_pool which has the same attributes as @attrs, bump the
3572 : : * reference count and return it. If there already is a matching
3573 : : * worker_pool, it will be used; otherwise, this function attempts to
3574 : : * create a new one.
3575 : : *
3576 : : * Should be called with wq_pool_mutex held.
3577 : : *
3578 : : * Return: On success, a worker_pool with the same attributes as @attrs.
3579 : : * On failure, %NULL.
3580 : : */
3581 : 209 : static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs)
3582 : : {
3583 : 209 : u32 hash = wqattrs_hash(attrs);
3584 : 209 : struct worker_pool *pool;
3585 : 209 : int node;
3586 : 209 : int target_node = NUMA_NO_NODE;
3587 : :
3588 : 209 : lockdep_assert_held(&wq_pool_mutex);
3589 : :
3590 : : /* do we already have a matching pool? */
3591 [ + + - - : 418 : hash_for_each_possible(unbound_pool_hash, pool, hash_node, hash) {
+ + ]
3592 [ + - ]: 187 : if (wqattrs_equal(pool->attrs, attrs)) {
3593 : 187 : pool->refcnt++;
3594 : 187 : return pool;
3595 : : }
3596 : : }
3597 : :
3598 : : /* if cpumask is contained inside a NUMA node, we belong to that node */
3599 [ - + ]: 22 : if (wq_numa_enabled) {
3600 [ # # ]: 0 : for_each_node(node) {
3601 : 0 : if (cpumask_subset(attrs->cpumask,
3602 [ # # ]: 0 : wq_numa_possible_cpumask[node])) {
3603 : : target_node = node;
3604 : : break;
3605 : : }
3606 : : }
3607 : : }
3608 : :
3609 : : /* nope, create a new one */
3610 : 22 : pool = kzalloc_node(sizeof(*pool), GFP_KERNEL, target_node);
3611 [ + - - + ]: 22 : if (!pool || init_worker_pool(pool) < 0)
3612 : 0 : goto fail;
3613 : :
3614 : 22 : lockdep_set_subclass(&pool->lock, 1); /* see put_pwq() */
3615 : 22 : copy_workqueue_attrs(pool->attrs, attrs);
3616 : 22 : pool->node = target_node;
3617 : :
3618 : : /*
3619 : : * no_numa isn't a worker_pool attribute, always clear it. See
3620 : : * 'struct workqueue_attrs' comments for detail.
3621 : : */
3622 : 22 : pool->attrs->no_numa = false;
3623 : :
3624 : 22 : if (worker_pool_assign_id(pool) < 0)
3625 : 0 : goto fail;
3626 : :
3627 : : /* create and start the initial worker */
3628 [ + + - + ]: 22 : if (wq_online && !create_worker(pool))
3629 : 0 : goto fail;
3630 : :
3631 : : /* install */
3632 [ - + ]: 22 : hash_add(unbound_pool_hash, &pool->hash_node, hash);
3633 : :
3634 : 22 : return pool;
3635 : 0 : fail:
3636 [ # # ]: 0 : if (pool)
3637 : 0 : put_unbound_pool(pool);
3638 : : return NULL;
3639 : : }
3640 : :
3641 : 11 : static void rcu_free_pwq(struct rcu_head *rcu)
3642 : : {
3643 : 11 : kmem_cache_free(pwq_cache,
3644 : 11 : container_of(rcu, struct pool_workqueue, rcu));
3645 : 11 : }
3646 : :
3647 : : /*
3648 : : * Scheduled on system_wq by put_pwq() when an unbound pwq hits zero refcnt
3649 : : * and needs to be destroyed.
3650 : : */
3651 : 11 : static void pwq_unbound_release_workfn(struct work_struct *work)
3652 : : {
3653 : 11 : struct pool_workqueue *pwq = container_of(work, struct pool_workqueue,
3654 : : unbound_release_work);
3655 : 11 : struct workqueue_struct *wq = pwq->wq;
3656 : 11 : struct worker_pool *pool = pwq->pool;
3657 : 11 : bool is_last;
3658 : :
3659 [ - + + - ]: 11 : if (WARN_ON_ONCE(!(wq->flags & WQ_UNBOUND)))
3660 : : return;
3661 : :
3662 : 11 : mutex_lock(&wq->mutex);
3663 : 11 : list_del_rcu(&pwq->pwqs_node);
3664 : 11 : is_last = list_empty(&wq->pwqs);
3665 : 11 : mutex_unlock(&wq->mutex);
3666 : :
3667 : 11 : mutex_lock(&wq_pool_mutex);
3668 : 11 : put_unbound_pool(pool);
3669 : 11 : mutex_unlock(&wq_pool_mutex);
3670 : :
3671 : 11 : call_rcu(&pwq->rcu, rcu_free_pwq);
3672 : :
3673 : : /*
3674 : : * If we're the last pwq going away, @wq is already dead and no one
3675 : : * is gonna access it anymore. Schedule RCU free.
3676 : : */
3677 [ + - ]: 11 : if (is_last) {
3678 : 11 : wq_unregister_lockdep(wq);
3679 : 11 : call_rcu(&wq->rcu, rcu_free_wq);
3680 : : }
3681 : : }
3682 : :
3683 : : /**
3684 : : * pwq_adjust_max_active - update a pwq's max_active to the current setting
3685 : : * @pwq: target pool_workqueue
3686 : : *
3687 : : * If @pwq isn't freezing, set @pwq->max_active to the associated
3688 : : * workqueue's saved_max_active and activate delayed work items
3689 : : * accordingly. If @pwq is freezing, clear @pwq->max_active to zero.
3690 : : */
3691 : 968 : static void pwq_adjust_max_active(struct pool_workqueue *pwq)
3692 : : {
3693 : 968 : struct workqueue_struct *wq = pwq->wq;
3694 : 968 : bool freezable = wq->flags & WQ_FREEZABLE;
3695 : 968 : unsigned long flags;
3696 : :
3697 : : /* for @wq->saved_max_active */
3698 : 968 : lockdep_assert_held(&wq->mutex);
3699 : :
3700 : : /* fast exit for non-freezable wqs */
3701 [ + + + + ]: 968 : if (!freezable && pwq->max_active == wq->saved_max_active)
3702 : : return;
3703 : :
3704 : : /* this function can be called during early boot w/ irq disabled */
3705 : 528 : spin_lock_irqsave(&pwq->pool->lock, flags);
3706 : :
3707 : : /*
3708 : : * During [un]freezing, the caller is responsible for ensuring that
3709 : : * this function is called at least once after @workqueue_freezing
3710 : : * is updated and visible.
3711 : : */
3712 [ + + + - ]: 528 : if (!freezable || !workqueue_freezing) {
3713 : 528 : pwq->max_active = wq->saved_max_active;
3714 : :
3715 [ - + ]: 528 : while (!list_empty(&pwq->delayed_works) &&
3716 [ # # ]: 0 : pwq->nr_active < pwq->max_active)
3717 : 0 : pwq_activate_first_delayed(pwq);
3718 : :
3719 : : /*
3720 : : * Need to kick a worker after thawed or an unbound wq's
3721 : : * max_active is bumped. It's a slow path. Do it always.
3722 : : */
3723 [ + + ]: 528 : wake_up_worker(pwq->pool);
3724 : : } else {
3725 : 0 : pwq->max_active = 0;
3726 : : }
3727 : :
3728 : 528 : spin_unlock_irqrestore(&pwq->pool->lock, flags);
3729 : : }
3730 : :
3731 : : /* initialize newly alloced @pwq which is associated with @wq and @pool */
3732 : 484 : static void init_pwq(struct pool_workqueue *pwq, struct workqueue_struct *wq,
3733 : : struct worker_pool *pool)
3734 : : {
3735 [ - + ]: 484 : BUG_ON((unsigned long)pwq & WORK_STRUCT_FLAG_MASK);
3736 : :
3737 : 484 : memset(pwq, 0, sizeof(*pwq));
3738 : :
3739 : 484 : pwq->pool = pool;
3740 : 484 : pwq->wq = wq;
3741 : 484 : pwq->flush_color = -1;
3742 : 484 : pwq->refcnt = 1;
3743 : 484 : INIT_LIST_HEAD(&pwq->delayed_works);
3744 : 484 : INIT_LIST_HEAD(&pwq->pwqs_node);
3745 : 484 : INIT_LIST_HEAD(&pwq->mayday_node);
3746 : 484 : INIT_WORK(&pwq->unbound_release_work, pwq_unbound_release_workfn);
3747 : 484 : }
3748 : :
3749 : : /* sync @pwq with the current state of its associated wq and link it */
3750 : 693 : static void link_pwq(struct pool_workqueue *pwq)
3751 : : {
3752 : 693 : struct workqueue_struct *wq = pwq->wq;
3753 : :
3754 : 693 : lockdep_assert_held(&wq->mutex);
3755 : :
3756 : : /* may be called multiple times, ignore if already linked */
3757 [ + + ]: 693 : if (!list_empty(&pwq->pwqs_node))
3758 : : return;
3759 : :
3760 : : /* set the matching work_color */
3761 : 484 : pwq->work_color = wq->work_color;
3762 : :
3763 : : /* sync max_active to the current setting */
3764 : 484 : pwq_adjust_max_active(pwq);
3765 : :
3766 : : /* link in @pwq */
3767 : 484 : list_add_rcu(&pwq->pwqs_node, &wq->pwqs);
3768 : : }
3769 : :
3770 : : /* obtain a pool matching @attr and create a pwq associating the pool and @wq */
3771 : 209 : static struct pool_workqueue *alloc_unbound_pwq(struct workqueue_struct *wq,
3772 : : const struct workqueue_attrs *attrs)
3773 : : {
3774 : 209 : struct worker_pool *pool;
3775 : 209 : struct pool_workqueue *pwq;
3776 : :
3777 : 209 : lockdep_assert_held(&wq_pool_mutex);
3778 : :
3779 : 209 : pool = get_unbound_pool(attrs);
3780 [ + - ]: 209 : if (!pool)
3781 : : return NULL;
3782 : :
3783 : 209 : pwq = kmem_cache_alloc_node(pwq_cache, GFP_KERNEL, pool->node);
3784 [ - + ]: 209 : if (!pwq) {
3785 : 0 : put_unbound_pool(pool);
3786 : 0 : return NULL;
3787 : : }
3788 : :
3789 : 209 : init_pwq(pwq, wq, pool);
3790 : 209 : return pwq;
3791 : : }
3792 : :
3793 : : /**
3794 : : * wq_calc_node_cpumask - calculate a wq_attrs' cpumask for the specified node
3795 : : * @attrs: the wq_attrs of the default pwq of the target workqueue
3796 : : * @node: the target NUMA node
3797 : : * @cpu_going_down: if >= 0, the CPU to consider as offline
3798 : : * @cpumask: outarg, the resulting cpumask
3799 : : *
3800 : : * Calculate the cpumask a workqueue with @attrs should use on @node. If
3801 : : * @cpu_going_down is >= 0, that cpu is considered offline during
3802 : : * calculation. The result is stored in @cpumask.
3803 : : *
3804 : : * If NUMA affinity is not enabled, @attrs->cpumask is always used. If
3805 : : * enabled and @node has online CPUs requested by @attrs, the returned
3806 : : * cpumask is the intersection of the possible CPUs of @node and
3807 : : * @attrs->cpumask.
3808 : : *
3809 : : * The caller is responsible for ensuring that the cpumask of @node stays
3810 : : * stable.
3811 : : *
3812 : : * Return: %true if the resulting @cpumask is different from @attrs->cpumask,
3813 : : * %false if equal.
3814 : : */
3815 : 209 : static bool wq_calc_node_cpumask(const struct workqueue_attrs *attrs, int node,
3816 : : int cpu_going_down, cpumask_t *cpumask)
3817 : : {
3818 [ - + - - ]: 209 : if (!wq_numa_enabled || attrs->no_numa)
3819 : 209 : goto use_dfl;
3820 : :
3821 : : /* does @node have any online CPUs @attrs wants? */
3822 [ # # ]: 0 : cpumask_and(cpumask, cpumask_of_node(node), attrs->cpumask);
3823 [ # # ]: 0 : if (cpu_going_down >= 0)
3824 : 0 : cpumask_clear_cpu(cpu_going_down, cpumask);
3825 : :
3826 [ # # ]: 0 : if (cpumask_empty(cpumask))
3827 : 0 : goto use_dfl;
3828 : :
3829 : : /* yeap, return possible CPUs in @node that @attrs wants */
3830 [ # # ]: 0 : cpumask_and(cpumask, attrs->cpumask, wq_numa_possible_cpumask[node]);
3831 : :
3832 [ # # ]: 0 : if (cpumask_empty(cpumask)) {
3833 [ # # ]: 0 : pr_warn_once("WARNING: workqueue cpumask: online intersect > "
3834 : : "possible intersect\n");
3835 : 0 : return false;
3836 : : }
3837 : :
3838 : 0 : return !cpumask_equal(cpumask, attrs->cpumask);
3839 : :
3840 : 209 : use_dfl:
3841 : 209 : cpumask_copy(cpumask, attrs->cpumask);
3842 : 209 : return false;
3843 : : }
3844 : :
3845 : : /* install @pwq into @wq's numa_pwq_tbl[] for @node and return the old pwq */
3846 : 209 : static struct pool_workqueue *numa_pwq_tbl_install(struct workqueue_struct *wq,
3847 : : int node,
3848 : : struct pool_workqueue *pwq)
3849 : : {
3850 : 209 : struct pool_workqueue *old_pwq;
3851 : :
3852 : 209 : lockdep_assert_held(&wq_pool_mutex);
3853 : 209 : lockdep_assert_held(&wq->mutex);
3854 : :
3855 : : /* link_pwq() can handle duplicate calls */
3856 : 209 : link_pwq(pwq);
3857 : :
3858 : 209 : old_pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]);
3859 : 209 : rcu_assign_pointer(wq->numa_pwq_tbl[node], pwq);
3860 : 0 : return old_pwq;
3861 : : }
3862 : :
3863 : : /* context to store the prepared attrs & pwqs before applying */
3864 : : struct apply_wqattrs_ctx {
3865 : : struct workqueue_struct *wq; /* target workqueue */
3866 : : struct workqueue_attrs *attrs; /* attrs to apply */
3867 : : struct list_head list; /* queued for batching commit */
3868 : : struct pool_workqueue *dfl_pwq;
3869 : : struct pool_workqueue *pwq_tbl[];
3870 : : };
3871 : :
3872 : : /* free the resources after success or abort */
3873 : 209 : static void apply_wqattrs_cleanup(struct apply_wqattrs_ctx *ctx)
3874 : : {
3875 [ + - ]: 209 : if (ctx) {
3876 : 209 : int node;
3877 : :
3878 [ + + ]: 836 : for_each_node(node)
3879 : 209 : put_pwq_unlocked(ctx->pwq_tbl[node]);
3880 : 209 : put_pwq_unlocked(ctx->dfl_pwq);
3881 : :
3882 [ + - ]: 209 : free_workqueue_attrs(ctx->attrs);
3883 : :
3884 : 209 : kfree(ctx);
3885 : : }
3886 : 209 : }
3887 : :
3888 : : /* allocate the attrs and pwqs for later installation */
3889 : : static struct apply_wqattrs_ctx *
3890 : 209 : apply_wqattrs_prepare(struct workqueue_struct *wq,
3891 : : const struct workqueue_attrs *attrs)
3892 : : {
3893 : 209 : struct apply_wqattrs_ctx *ctx;
3894 : 209 : struct workqueue_attrs *new_attrs, *tmp_attrs;
3895 : 209 : int node;
3896 : :
3897 : 209 : lockdep_assert_held(&wq_pool_mutex);
3898 : :
3899 [ - + ]: 209 : ctx = kzalloc(struct_size(ctx, pwq_tbl, nr_node_ids), GFP_KERNEL);
3900 : :
3901 : 209 : new_attrs = alloc_workqueue_attrs();
3902 : 209 : tmp_attrs = alloc_workqueue_attrs();
3903 [ + - - + ]: 209 : if (!ctx || !new_attrs || !tmp_attrs)
3904 : 0 : goto out_free;
3905 : :
3906 : : /*
3907 : : * Calculate the attrs of the default pwq.
3908 : : * If the user configured cpumask doesn't overlap with the
3909 : : * wq_unbound_cpumask, we fallback to the wq_unbound_cpumask.
3910 : : */
3911 [ - + ]: 209 : copy_workqueue_attrs(new_attrs, attrs);
3912 [ - + ]: 209 : cpumask_and(new_attrs->cpumask, new_attrs->cpumask, wq_unbound_cpumask);
3913 [ - + ]: 209 : if (unlikely(cpumask_empty(new_attrs->cpumask)))
3914 : 0 : cpumask_copy(new_attrs->cpumask, wq_unbound_cpumask);
3915 : :
3916 : : /*
3917 : : * We may create multiple pwqs with differing cpumasks. Make a
3918 : : * copy of @new_attrs which will be modified and used to obtain
3919 : : * pools.
3920 : : */
3921 : 209 : copy_workqueue_attrs(tmp_attrs, new_attrs);
3922 : :
3923 : : /*
3924 : : * If something goes wrong during CPU up/down, we'll fall back to
3925 : : * the default pwq covering whole @attrs->cpumask. Always create
3926 : : * it even if we don't use it immediately.
3927 : : */
3928 : 209 : ctx->dfl_pwq = alloc_unbound_pwq(wq, new_attrs);
3929 [ - + ]: 209 : if (!ctx->dfl_pwq)
3930 : 0 : goto out_free;
3931 : :
3932 [ + + ]: 836 : for_each_node(node) {
3933 [ - + ]: 209 : if (wq_calc_node_cpumask(new_attrs, node, -1, tmp_attrs->cpumask)) {
3934 : 0 : ctx->pwq_tbl[node] = alloc_unbound_pwq(wq, tmp_attrs);
3935 [ # # ]: 0 : if (!ctx->pwq_tbl[node])
3936 : 0 : goto out_free;
3937 : : } else {
3938 : 209 : ctx->dfl_pwq->refcnt++;
3939 : 209 : ctx->pwq_tbl[node] = ctx->dfl_pwq;
3940 : : }
3941 : : }
3942 : :
3943 : : /* save the user configured attrs and sanitize it. */
3944 : 209 : copy_workqueue_attrs(new_attrs, attrs);
3945 : 209 : cpumask_and(new_attrs->cpumask, new_attrs->cpumask, cpu_possible_mask);
3946 : 209 : ctx->attrs = new_attrs;
3947 : :
3948 : 209 : ctx->wq = wq;
3949 : 209 : free_workqueue_attrs(tmp_attrs);
3950 : 209 : return ctx;
3951 : :
3952 : 0 : out_free:
3953 [ # # ]: 0 : free_workqueue_attrs(tmp_attrs);
3954 [ # # ]: 0 : free_workqueue_attrs(new_attrs);
3955 : 0 : apply_wqattrs_cleanup(ctx);
3956 : 0 : return NULL;
3957 : : }
3958 : :
3959 : : /* set attrs and install prepared pwqs, @ctx points to old pwqs on return */
3960 : 209 : static void apply_wqattrs_commit(struct apply_wqattrs_ctx *ctx)
3961 : : {
3962 : 209 : int node;
3963 : :
3964 : : /* all pwqs have been created successfully, let's install'em */
3965 : 209 : mutex_lock(&ctx->wq->mutex);
3966 : :
3967 : 209 : copy_workqueue_attrs(ctx->wq->unbound_attrs, ctx->attrs);
3968 : :
3969 : : /* save the previous pwq and install the new one */
3970 [ + + ]: 627 : for_each_node(node)
3971 : 418 : ctx->pwq_tbl[node] = numa_pwq_tbl_install(ctx->wq, node,
3972 : : ctx->pwq_tbl[node]);
3973 : :
3974 : : /* @dfl_pwq might not have been used, ensure it's linked */
3975 : 209 : link_pwq(ctx->dfl_pwq);
3976 : 209 : swap(ctx->wq->dfl_pwq, ctx->dfl_pwq);
3977 : :
3978 : 209 : mutex_unlock(&ctx->wq->mutex);
3979 : 209 : }
3980 : :
3981 : 0 : static void apply_wqattrs_lock(void)
3982 : : {
3983 : : /* CPUs should stay stable across pwq creations and installations */
3984 : 0 : get_online_cpus();
3985 : 0 : mutex_lock(&wq_pool_mutex);
3986 : : }
3987 : :
3988 : 0 : static void apply_wqattrs_unlock(void)
3989 : : {
3990 : 0 : mutex_unlock(&wq_pool_mutex);
3991 : 0 : put_online_cpus();
3992 : 0 : }
3993 : :
3994 : 209 : static int apply_workqueue_attrs_locked(struct workqueue_struct *wq,
3995 : : const struct workqueue_attrs *attrs)
3996 : : {
3997 : 209 : struct apply_wqattrs_ctx *ctx;
3998 : :
3999 : : /* only unbound workqueues can change attributes */
4000 [ - + + - ]: 209 : if (WARN_ON(!(wq->flags & WQ_UNBOUND)))
4001 : : return -EINVAL;
4002 : :
4003 : : /* creating multiple pwqs breaks ordering guarantee */
4004 [ - + ]: 209 : if (!list_empty(&wq->pwqs)) {
4005 [ # # # # ]: 0 : if (WARN_ON(wq->flags & __WQ_ORDERED_EXPLICIT))
4006 : : return -EINVAL;
4007 : :
4008 : 0 : wq->flags &= ~__WQ_ORDERED;
4009 : : }
4010 : :
4011 : 209 : ctx = apply_wqattrs_prepare(wq, attrs);
4012 [ + - ]: 209 : if (!ctx)
4013 : : return -ENOMEM;
4014 : :
4015 : : /* the ctx has been prepared successfully, let's commit it */
4016 : 209 : apply_wqattrs_commit(ctx);
4017 : 209 : apply_wqattrs_cleanup(ctx);
4018 : :
4019 : 209 : return 0;
4020 : : }
4021 : :
4022 : : /**
4023 : : * apply_workqueue_attrs - apply new workqueue_attrs to an unbound workqueue
4024 : : * @wq: the target workqueue
4025 : : * @attrs: the workqueue_attrs to apply, allocated with alloc_workqueue_attrs()
4026 : : *
4027 : : * Apply @attrs to an unbound workqueue @wq. Unless disabled, on NUMA
4028 : : * machines, this function maps a separate pwq to each NUMA node with
4029 : : * possibles CPUs in @attrs->cpumask so that work items are affine to the
4030 : : * NUMA node it was issued on. Older pwqs are released as in-flight work
4031 : : * items finish. Note that a work item which repeatedly requeues itself
4032 : : * back-to-back will stay on its current pwq.
4033 : : *
4034 : : * Performs GFP_KERNEL allocations.
4035 : : *
4036 : : * Assumes caller has CPU hotplug read exclusion, i.e. get_online_cpus().
4037 : : *
4038 : : * Return: 0 on success and -errno on failure.
4039 : : */
4040 : 209 : int apply_workqueue_attrs(struct workqueue_struct *wq,
4041 : : const struct workqueue_attrs *attrs)
4042 : : {
4043 : 209 : int ret;
4044 : :
4045 : 209 : lockdep_assert_cpus_held();
4046 : :
4047 : 209 : mutex_lock(&wq_pool_mutex);
4048 : 209 : ret = apply_workqueue_attrs_locked(wq, attrs);
4049 : 209 : mutex_unlock(&wq_pool_mutex);
4050 : :
4051 : 209 : return ret;
4052 : : }
4053 : :
4054 : : /**
4055 : : * wq_update_unbound_numa - update NUMA affinity of a wq for CPU hot[un]plug
4056 : : * @wq: the target workqueue
4057 : : * @cpu: the CPU coming up or going down
4058 : : * @online: whether @cpu is coming up or going down
4059 : : *
4060 : : * This function is to be called from %CPU_DOWN_PREPARE, %CPU_ONLINE and
4061 : : * %CPU_DOWN_FAILED. @cpu is being hot[un]plugged, update NUMA affinity of
4062 : : * @wq accordingly.
4063 : : *
4064 : : * If NUMA affinity can't be adjusted due to memory allocation failure, it
4065 : : * falls back to @wq->dfl_pwq which may not be optimal but is always
4066 : : * correct.
4067 : : *
4068 : : * Note that when the last allowed CPU of a NUMA node goes offline for a
4069 : : * workqueue with a cpumask spanning multiple nodes, the workers which were
4070 : : * already executing the work items for the workqueue will lose their CPU
4071 : : * affinity and may execute on any CPU. This is similar to how per-cpu
4072 : : * workqueues behave on CPU_DOWN. If a workqueue user wants strict
4073 : : * affinity, it's the user's responsibility to flush the work item from
4074 : : * CPU_DOWN_PREPARE.
4075 : : */
4076 : 99 : static void wq_update_unbound_numa(struct workqueue_struct *wq, int cpu,
4077 : : bool online)
4078 : : {
4079 [ - + ]: 99 : int node = cpu_to_node(cpu);
4080 [ - + ]: 99 : int cpu_off = online ? -1 : cpu;
4081 : 99 : struct pool_workqueue *old_pwq = NULL, *pwq;
4082 : 99 : struct workqueue_attrs *target_attrs;
4083 : 99 : cpumask_t *cpumask;
4084 : :
4085 : 99 : lockdep_assert_held(&wq_pool_mutex);
4086 : :
4087 [ - + - - ]: 99 : if (!wq_numa_enabled || !(wq->flags & WQ_UNBOUND) ||
4088 [ # # ]: 0 : wq->unbound_attrs->no_numa)
4089 : : return;
4090 : :
4091 : : /*
4092 : : * We don't wanna alloc/free wq_attrs for each wq for each CPU.
4093 : : * Let's use a preallocated one. The following buf is protected by
4094 : : * CPU hotplug exclusion.
4095 : : */
4096 : 0 : target_attrs = wq_update_unbound_numa_attrs_buf;
4097 : 0 : cpumask = target_attrs->cpumask;
4098 : :
4099 [ # # ]: 0 : copy_workqueue_attrs(target_attrs, wq->unbound_attrs);
4100 [ # # ]: 0 : pwq = unbound_pwq_by_node(wq, node);
4101 : :
4102 : : /*
4103 : : * Let's determine what needs to be done. If the target cpumask is
4104 : : * different from the default pwq's, we need to compare it to @pwq's
4105 : : * and create a new one if they don't match. If the target cpumask
4106 : : * equals the default pwq's, the default pwq should be used.
4107 : : */
4108 [ # # ]: 0 : if (wq_calc_node_cpumask(wq->dfl_pwq->pool->attrs, node, cpu_off, cpumask)) {
4109 [ # # ]: 0 : if (cpumask_equal(cpumask, pwq->pool->attrs->cpumask))
4110 : : return;
4111 : : } else {
4112 : 0 : goto use_dfl_pwq;
4113 : : }
4114 : :
4115 : : /* create a new pwq */
4116 : 0 : pwq = alloc_unbound_pwq(wq, target_attrs);
4117 [ # # ]: 0 : if (!pwq) {
4118 : 0 : pr_warn("workqueue: allocation failed while updating NUMA affinity of \"%s\"\n",
4119 : : wq->name);
4120 : 0 : goto use_dfl_pwq;
4121 : : }
4122 : :
4123 : : /* Install the new pwq. */
4124 : 0 : mutex_lock(&wq->mutex);
4125 : 0 : old_pwq = numa_pwq_tbl_install(wq, node, pwq);
4126 : 0 : goto out_unlock;
4127 : :
4128 : 0 : use_dfl_pwq:
4129 : 0 : mutex_lock(&wq->mutex);
4130 : 0 : spin_lock_irq(&wq->dfl_pwq->pool->lock);
4131 [ # # ]: 0 : get_pwq(wq->dfl_pwq);
4132 : 0 : spin_unlock_irq(&wq->dfl_pwq->pool->lock);
4133 : 0 : old_pwq = numa_pwq_tbl_install(wq, node, wq->dfl_pwq);
4134 : 0 : out_unlock:
4135 : 0 : mutex_unlock(&wq->mutex);
4136 : 0 : put_pwq_unlocked(old_pwq);
4137 : : }
4138 : :
4139 : 484 : static int alloc_and_link_pwqs(struct workqueue_struct *wq)
4140 : : {
4141 : 484 : bool highpri = wq->flags & WQ_HIGHPRI;
4142 : 484 : int cpu, ret;
4143 : :
4144 [ + + ]: 484 : if (!(wq->flags & WQ_UNBOUND)) {
4145 : 275 : wq->cpu_pwqs = alloc_percpu(struct pool_workqueue);
4146 [ + - ]: 275 : if (!wq->cpu_pwqs)
4147 : : return -ENOMEM;
4148 : :
4149 [ + + ]: 550 : for_each_possible_cpu(cpu) {
4150 : 275 : struct pool_workqueue *pwq =
4151 : 275 : per_cpu_ptr(wq->cpu_pwqs, cpu);
4152 : 275 : struct worker_pool *cpu_pools =
4153 : 275 : per_cpu(cpu_worker_pools, cpu);
4154 : :
4155 : 275 : init_pwq(pwq, wq, &cpu_pools[highpri]);
4156 : :
4157 : 275 : mutex_lock(&wq->mutex);
4158 : 275 : link_pwq(pwq);
4159 : 275 : mutex_unlock(&wq->mutex);
4160 : : }
4161 : : return 0;
4162 : : }
4163 : :
4164 : 209 : get_online_cpus();
4165 [ + + ]: 209 : if (wq->flags & __WQ_ORDERED) {
4166 : 165 : ret = apply_workqueue_attrs(wq, ordered_wq_attrs[highpri]);
4167 : : /* there should only be single pwq for ordering guarantee */
4168 [ + - + - : 330 : WARN(!ret && (wq->pwqs.next != &wq->dfl_pwq->pwqs_node ||
+ - - + ]
4169 : : wq->pwqs.prev != &wq->dfl_pwq->pwqs_node),
4170 : : "ordering guarantee broken for workqueue %s\n", wq->name);
4171 : : } else {
4172 : 44 : ret = apply_workqueue_attrs(wq, unbound_std_wq_attrs[highpri]);
4173 : : }
4174 : 209 : put_online_cpus();
4175 : :
4176 : 209 : return ret;
4177 : : }
4178 : :
4179 : 484 : static int wq_clamp_max_active(int max_active, unsigned int flags,
4180 : : const char *name)
4181 : : {
4182 [ + + ]: 484 : int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE;
4183 : :
4184 [ - + ]: 484 : if (max_active < 1 || max_active > lim)
4185 : 0 : pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n",
4186 : : max_active, name, 1, lim);
4187 : :
4188 : 484 : return clamp_val(max_active, 1, lim);
4189 : : }
4190 : :
4191 : : /*
4192 : : * Workqueues which may be used during memory reclaim should have a rescuer
4193 : : * to guarantee forward progress.
4194 : : */
4195 : 484 : static int init_rescuer(struct workqueue_struct *wq)
4196 : : {
4197 : 484 : struct worker *rescuer;
4198 : 484 : int ret;
4199 : :
4200 [ + + ]: 484 : if (!(wq->flags & WQ_MEM_RECLAIM))
4201 : : return 0;
4202 : :
4203 : 231 : rescuer = alloc_worker(NUMA_NO_NODE);
4204 [ + - ]: 231 : if (!rescuer)
4205 : : return -ENOMEM;
4206 : :
4207 : 231 : rescuer->rescue_wq = wq;
4208 : 231 : rescuer->task = kthread_create(rescuer_thread, rescuer, "%s", wq->name);
4209 [ - + ]: 231 : ret = PTR_ERR_OR_ZERO(rescuer->task);
4210 [ # # ]: 0 : if (ret) {
4211 : 0 : kfree(rescuer);
4212 : 0 : return ret;
4213 : : }
4214 : :
4215 : 231 : wq->rescuer = rescuer;
4216 : 231 : kthread_bind_mask(rescuer->task, cpu_possible_mask);
4217 : 231 : wake_up_process(rescuer->task);
4218 : :
4219 : 231 : return 0;
4220 : : }
4221 : :
4222 : : __printf(1, 4)
4223 : 484 : struct workqueue_struct *alloc_workqueue(const char *fmt,
4224 : : unsigned int flags,
4225 : : int max_active, ...)
4226 : : {
4227 : 484 : size_t tbl_size = 0;
4228 : 484 : va_list args;
4229 : 484 : struct workqueue_struct *wq;
4230 : 484 : struct pool_workqueue *pwq;
4231 : :
4232 : : /*
4233 : : * Unbound && max_active == 1 used to imply ordered, which is no
4234 : : * longer the case on NUMA machines due to per-node pools. While
4235 : : * alloc_ordered_workqueue() is the right way to create an ordered
4236 : : * workqueue, keep the previous behavior to avoid subtle breakages
4237 : : * on NUMA.
4238 : : */
4239 [ + + + + ]: 484 : if ((flags & WQ_UNBOUND) && max_active == 1)
4240 : 165 : flags |= __WQ_ORDERED;
4241 : :
4242 : : /* see the comment above the definition of WQ_POWER_EFFICIENT */
4243 [ + + - + ]: 484 : if ((flags & WQ_POWER_EFFICIENT) && wq_power_efficient)
4244 : 0 : flags |= WQ_UNBOUND;
4245 : :
4246 : : /* allocate wq and format name */
4247 [ + + ]: 484 : if (flags & WQ_UNBOUND)
4248 : 209 : tbl_size = nr_node_ids * sizeof(wq->numa_pwq_tbl[0]);
4249 : :
4250 : 484 : wq = kzalloc(sizeof(*wq) + tbl_size, GFP_KERNEL);
4251 [ + - ]: 484 : if (!wq)
4252 : : return NULL;
4253 : :
4254 [ + + ]: 484 : if (flags & WQ_UNBOUND) {
4255 : 209 : wq->unbound_attrs = alloc_workqueue_attrs();
4256 [ - + ]: 209 : if (!wq->unbound_attrs)
4257 : 0 : goto err_free_wq;
4258 : : }
4259 : :
4260 : 484 : va_start(args, max_active);
4261 : 484 : vsnprintf(wq->name, sizeof(wq->name), fmt, args);
4262 : 484 : va_end(args);
4263 : :
4264 [ + + ]: 484 : max_active = max_active ?: WQ_DFL_ACTIVE;
4265 : 484 : max_active = wq_clamp_max_active(max_active, flags, wq->name);
4266 : :
4267 : : /* init wq */
4268 : 484 : wq->flags = flags;
4269 : 484 : wq->saved_max_active = max_active;
4270 : 484 : mutex_init(&wq->mutex);
4271 : 484 : atomic_set(&wq->nr_pwqs_to_flush, 0);
4272 : 484 : INIT_LIST_HEAD(&wq->pwqs);
4273 : 484 : INIT_LIST_HEAD(&wq->flusher_queue);
4274 : 484 : INIT_LIST_HEAD(&wq->flusher_overflow);
4275 : 484 : INIT_LIST_HEAD(&wq->maydays);
4276 : :
4277 : 484 : wq_init_lockdep(wq);
4278 : 484 : INIT_LIST_HEAD(&wq->list);
4279 : :
4280 [ - + ]: 484 : if (alloc_and_link_pwqs(wq) < 0)
4281 : 0 : goto err_unreg_lockdep;
4282 : :
4283 [ + + - + ]: 484 : if (wq_online && init_rescuer(wq) < 0)
4284 : 0 : goto err_destroy;
4285 : :
4286 [ + + - + ]: 484 : if ((wq->flags & WQ_SYSFS) && workqueue_sysfs_register(wq))
4287 : 0 : goto err_destroy;
4288 : :
4289 : : /*
4290 : : * wq_pool_mutex protects global freeze state and workqueues list.
4291 : : * Grab it, adjust max_active and add the new @wq to workqueues
4292 : : * list.
4293 : : */
4294 : 484 : mutex_lock(&wq_pool_mutex);
4295 : :
4296 : 484 : mutex_lock(&wq->mutex);
4297 [ + + ]: 968 : for_each_pwq(pwq, wq)
4298 : 484 : pwq_adjust_max_active(pwq);
4299 : 484 : mutex_unlock(&wq->mutex);
4300 : :
4301 : 484 : list_add_tail_rcu(&wq->list, &workqueues);
4302 : :
4303 : 484 : mutex_unlock(&wq_pool_mutex);
4304 : :
4305 : 484 : return wq;
4306 : :
4307 : : err_unreg_lockdep:
4308 : 0 : wq_unregister_lockdep(wq);
4309 : : wq_free_lockdep(wq);
4310 : 0 : err_free_wq:
4311 [ # # ]: 0 : free_workqueue_attrs(wq->unbound_attrs);
4312 : 0 : kfree(wq);
4313 : 0 : return NULL;
4314 : 0 : err_destroy:
4315 : 0 : destroy_workqueue(wq);
4316 : 0 : return NULL;
4317 : : }
4318 : : EXPORT_SYMBOL_GPL(alloc_workqueue);
4319 : :
4320 : : static bool pwq_busy(struct pool_workqueue *pwq)
4321 : : {
4322 : : int i;
4323 : :
4324 [ + + ]: 176 : for (i = 0; i < WORK_NR_COLORS; i++)
4325 [ + - ]: 165 : if (pwq->nr_in_flight[i])
4326 : : return true;
4327 : :
4328 [ - + - - ]: 11 : if ((pwq != pwq->wq->dfl_pwq) && (pwq->refcnt > 1))
4329 : : return true;
4330 [ + - - + ]: 11 : if (pwq->nr_active || !list_empty(&pwq->delayed_works))
4331 : 0 : return true;
4332 : :
4333 : : return false;
4334 : : }
4335 : :
4336 : : /**
4337 : : * destroy_workqueue - safely terminate a workqueue
4338 : : * @wq: target workqueue
4339 : : *
4340 : : * Safely destroy a workqueue. All work currently pending will be done first.
4341 : : */
4342 : 11 : void destroy_workqueue(struct workqueue_struct *wq)
4343 : : {
4344 : 11 : struct pool_workqueue *pwq;
4345 : 11 : int node;
4346 : :
4347 : : /*
4348 : : * Remove it from sysfs first so that sanity check failure doesn't
4349 : : * lead to sysfs name conflicts.
4350 : : */
4351 : 11 : workqueue_sysfs_unregister(wq);
4352 : :
4353 : : /* drain it before proceeding with destruction */
4354 : 11 : drain_workqueue(wq);
4355 : :
4356 : : /* kill rescuer, if sanity checks fail, leave it w/o rescuer */
4357 [ + - ]: 11 : if (wq->rescuer) {
4358 : 11 : struct worker *rescuer = wq->rescuer;
4359 : :
4360 : : /* this prevents new queueing */
4361 : 11 : spin_lock_irq(&wq_mayday_lock);
4362 : 11 : wq->rescuer = NULL;
4363 : 11 : spin_unlock_irq(&wq_mayday_lock);
4364 : :
4365 : : /* rescuer will empty maydays list before exiting */
4366 : 11 : kthread_stop(rescuer->task);
4367 : 11 : kfree(rescuer);
4368 : : }
4369 : :
4370 : : /*
4371 : : * Sanity checks - grab all the locks so that we wait for all
4372 : : * in-flight operations which may do put_pwq().
4373 : : */
4374 : 11 : mutex_lock(&wq_pool_mutex);
4375 : 11 : mutex_lock(&wq->mutex);
4376 [ + + ]: 22 : for_each_pwq(pwq, wq) {
4377 : 11 : spin_lock_irq(&pwq->pool->lock);
4378 [ - + - + ]: 11 : if (WARN_ON(pwq_busy(pwq))) {
4379 : 0 : pr_warn("%s: %s has the following busy pwq\n",
4380 : : __func__, wq->name);
4381 : 0 : show_pwq(pwq);
4382 : 0 : spin_unlock_irq(&pwq->pool->lock);
4383 : 0 : mutex_unlock(&wq->mutex);
4384 : 0 : mutex_unlock(&wq_pool_mutex);
4385 : 0 : show_workqueue_state();
4386 : 0 : return;
4387 : : }
4388 : 11 : spin_unlock_irq(&pwq->pool->lock);
4389 : : }
4390 : 11 : mutex_unlock(&wq->mutex);
4391 : 11 : mutex_unlock(&wq_pool_mutex);
4392 : :
4393 : : /*
4394 : : * wq list is used to freeze wq, remove from list after
4395 : : * flushing is complete in case freeze races us.
4396 : : */
4397 : 11 : mutex_lock(&wq_pool_mutex);
4398 : 11 : list_del_rcu(&wq->list);
4399 : 11 : mutex_unlock(&wq_pool_mutex);
4400 : :
4401 [ - + ]: 11 : if (!(wq->flags & WQ_UNBOUND)) {
4402 : 0 : wq_unregister_lockdep(wq);
4403 : : /*
4404 : : * The base ref is never dropped on per-cpu pwqs. Directly
4405 : : * schedule RCU free.
4406 : : */
4407 : 0 : call_rcu(&wq->rcu, rcu_free_wq);
4408 : : } else {
4409 : : /*
4410 : : * We're the sole accessor of @wq at this point. Directly
4411 : : * access numa_pwq_tbl[] and dfl_pwq to put the base refs.
4412 : : * @wq will be freed when the last pwq is released.
4413 : : */
4414 [ + + ]: 44 : for_each_node(node) {
4415 : 11 : pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]);
4416 : 11 : RCU_INIT_POINTER(wq->numa_pwq_tbl[node], NULL);
4417 : 11 : put_pwq_unlocked(pwq);
4418 : : }
4419 : :
4420 : : /*
4421 : : * Put dfl_pwq. @wq may be freed any time after dfl_pwq is
4422 : : * put. Don't access it afterwards.
4423 : : */
4424 : 11 : pwq = wq->dfl_pwq;
4425 : 11 : wq->dfl_pwq = NULL;
4426 : 11 : put_pwq_unlocked(pwq);
4427 : : }
4428 : : }
4429 : : EXPORT_SYMBOL_GPL(destroy_workqueue);
4430 : :
4431 : : /**
4432 : : * workqueue_set_max_active - adjust max_active of a workqueue
4433 : : * @wq: target workqueue
4434 : : * @max_active: new max_active value.
4435 : : *
4436 : : * Set max_active of @wq to @max_active.
4437 : : *
4438 : : * CONTEXT:
4439 : : * Don't call from IRQ context.
4440 : : */
4441 : 0 : void workqueue_set_max_active(struct workqueue_struct *wq, int max_active)
4442 : : {
4443 : 0 : struct pool_workqueue *pwq;
4444 : :
4445 : : /* disallow meddling with max_active for ordered workqueues */
4446 [ # # # # ]: 0 : if (WARN_ON(wq->flags & __WQ_ORDERED_EXPLICIT))
4447 : : return;
4448 : :
4449 : 0 : max_active = wq_clamp_max_active(max_active, wq->flags, wq->name);
4450 : :
4451 : 0 : mutex_lock(&wq->mutex);
4452 : :
4453 : 0 : wq->flags &= ~__WQ_ORDERED;
4454 : 0 : wq->saved_max_active = max_active;
4455 : :
4456 [ # # ]: 0 : for_each_pwq(pwq, wq)
4457 : 0 : pwq_adjust_max_active(pwq);
4458 : :
4459 : 0 : mutex_unlock(&wq->mutex);
4460 : : }
4461 : : EXPORT_SYMBOL_GPL(workqueue_set_max_active);
4462 : :
4463 : : /**
4464 : : * current_work - retrieve %current task's work struct
4465 : : *
4466 : : * Determine if %current task is a workqueue worker and what it's working on.
4467 : : * Useful to find out the context that the %current task is running in.
4468 : : *
4469 : : * Return: work struct if %current task is a workqueue worker, %NULL otherwise.
4470 : : */
4471 : 0 : struct work_struct *current_work(void)
4472 : : {
4473 : 0 : struct worker *worker = current_wq_worker();
4474 : :
4475 [ # # ]: 0 : return worker ? worker->current_work : NULL;
4476 : : }
4477 : : EXPORT_SYMBOL(current_work);
4478 : :
4479 : : /**
4480 : : * current_is_workqueue_rescuer - is %current workqueue rescuer?
4481 : : *
4482 : : * Determine whether %current is a workqueue rescuer. Can be used from
4483 : : * work functions to determine whether it's being run off the rescuer task.
4484 : : *
4485 : : * Return: %true if %current is a workqueue rescuer. %false otherwise.
4486 : : */
4487 : 0 : bool current_is_workqueue_rescuer(void)
4488 : : {
4489 : 0 : struct worker *worker = current_wq_worker();
4490 : :
4491 [ # # # # ]: 0 : return worker && worker->rescue_wq;
4492 : : }
4493 : :
4494 : : /**
4495 : : * workqueue_congested - test whether a workqueue is congested
4496 : : * @cpu: CPU in question
4497 : : * @wq: target workqueue
4498 : : *
4499 : : * Test whether @wq's cpu workqueue for @cpu is congested. There is
4500 : : * no synchronization around this function and the test result is
4501 : : * unreliable and only useful as advisory hints or for debugging.
4502 : : *
4503 : : * If @cpu is WORK_CPU_UNBOUND, the test is performed on the local CPU.
4504 : : * Note that both per-cpu and unbound workqueues may be associated with
4505 : : * multiple pool_workqueues which have separate congested states. A
4506 : : * workqueue being congested on one CPU doesn't mean the workqueue is also
4507 : : * contested on other CPUs / NUMA nodes.
4508 : : *
4509 : : * Return:
4510 : : * %true if congested, %false otherwise.
4511 : : */
4512 : 0 : bool workqueue_congested(int cpu, struct workqueue_struct *wq)
4513 : : {
4514 : 0 : struct pool_workqueue *pwq;
4515 : 0 : bool ret;
4516 : :
4517 : 0 : rcu_read_lock();
4518 : 0 : preempt_disable();
4519 : :
4520 [ # # ]: 0 : if (cpu == WORK_CPU_UNBOUND)
4521 : 0 : cpu = smp_processor_id();
4522 : :
4523 [ # # ]: 0 : if (!(wq->flags & WQ_UNBOUND))
4524 : 0 : pwq = per_cpu_ptr(wq->cpu_pwqs, cpu);
4525 : : else
4526 [ # # ]: 0 : pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu));
4527 : :
4528 : 0 : ret = !list_empty(&pwq->delayed_works);
4529 : 0 : preempt_enable();
4530 : 0 : rcu_read_unlock();
4531 : :
4532 : 0 : return ret;
4533 : : }
4534 : : EXPORT_SYMBOL_GPL(workqueue_congested);
4535 : :
4536 : : /**
4537 : : * work_busy - test whether a work is currently pending or running
4538 : : * @work: the work to be tested
4539 : : *
4540 : : * Test whether @work is currently pending or running. There is no
4541 : : * synchronization around this function and the test result is
4542 : : * unreliable and only useful as advisory hints or for debugging.
4543 : : *
4544 : : * Return:
4545 : : * OR'd bitmask of WORK_BUSY_* bits.
4546 : : */
4547 : 0 : unsigned int work_busy(struct work_struct *work)
4548 : : {
4549 : 0 : struct worker_pool *pool;
4550 : 0 : unsigned long flags;
4551 : 0 : unsigned int ret = 0;
4552 : :
4553 [ # # ]: 0 : if (work_pending(work))
4554 : 0 : ret |= WORK_BUSY_PENDING;
4555 : :
4556 : 0 : rcu_read_lock();
4557 : 0 : pool = get_work_pool(work);
4558 [ # # ]: 0 : if (pool) {
4559 : 0 : spin_lock_irqsave(&pool->lock, flags);
4560 [ # # # # ]: 0 : if (find_worker_executing_work(pool, work))
4561 : 0 : ret |= WORK_BUSY_RUNNING;
4562 : 0 : spin_unlock_irqrestore(&pool->lock, flags);
4563 : : }
4564 : 0 : rcu_read_unlock();
4565 : :
4566 : 0 : return ret;
4567 : : }
4568 : : EXPORT_SYMBOL_GPL(work_busy);
4569 : :
4570 : : /**
4571 : : * set_worker_desc - set description for the current work item
4572 : : * @fmt: printf-style format string
4573 : : * @...: arguments for the format string
4574 : : *
4575 : : * This function can be called by a running work function to describe what
4576 : : * the work item is about. If the worker task gets dumped, this
4577 : : * information will be printed out together to help debugging. The
4578 : : * description can be at most WORKER_DESC_LEN including the trailing '\0'.
4579 : : */
4580 : 0 : void set_worker_desc(const char *fmt, ...)
4581 : : {
4582 : 0 : struct worker *worker = current_wq_worker();
4583 : 0 : va_list args;
4584 : :
4585 [ # # ]: 0 : if (worker) {
4586 : 0 : va_start(args, fmt);
4587 : 0 : vsnprintf(worker->desc, sizeof(worker->desc), fmt, args);
4588 : 0 : va_end(args);
4589 : : }
4590 : 0 : }
4591 : : EXPORT_SYMBOL_GPL(set_worker_desc);
4592 : :
4593 : : /**
4594 : : * print_worker_info - print out worker information and description
4595 : : * @log_lvl: the log level to use when printing
4596 : : * @task: target task
4597 : : *
4598 : : * If @task is a worker and currently executing a work item, print out the
4599 : : * name of the workqueue being serviced and worker description set with
4600 : : * set_worker_desc() by the currently executing work item.
4601 : : *
4602 : : * This function can be safely called on any task as long as the
4603 : : * task_struct itself is accessible. While safe, this function isn't
4604 : : * synchronized and may print out mixups or garbages of limited length.
4605 : : */
4606 : 0 : void print_worker_info(const char *log_lvl, struct task_struct *task)
4607 : : {
4608 : 0 : work_func_t *fn = NULL;
4609 : 0 : char name[WQ_NAME_LEN] = { };
4610 : 0 : char desc[WORKER_DESC_LEN] = { };
4611 : 0 : struct pool_workqueue *pwq = NULL;
4612 : 0 : struct workqueue_struct *wq = NULL;
4613 : 0 : struct worker *worker;
4614 : :
4615 [ # # ]: 0 : if (!(task->flags & PF_WQ_WORKER))
4616 : 0 : return;
4617 : :
4618 : : /*
4619 : : * This function is called without any synchronization and @task
4620 : : * could be in any state. Be careful with dereferences.
4621 : : */
4622 : 0 : worker = kthread_probe_data(task);
4623 : :
4624 : : /*
4625 : : * Carefully copy the associated workqueue's workfn, name and desc.
4626 : : * Keep the original last '\0' in case the original is garbage.
4627 : : */
4628 : 0 : probe_kernel_read(&fn, &worker->current_func, sizeof(fn));
4629 : 0 : probe_kernel_read(&pwq, &worker->current_pwq, sizeof(pwq));
4630 : 0 : probe_kernel_read(&wq, &pwq->wq, sizeof(wq));
4631 : 0 : probe_kernel_read(name, wq->name, sizeof(name) - 1);
4632 : 0 : probe_kernel_read(desc, worker->desc, sizeof(desc) - 1);
4633 : :
4634 [ # # # # : 0 : if (fn || name[0] || desc[0]) {
# # ]
4635 : 0 : printk("%sWorkqueue: %s %ps", log_lvl, name, fn);
4636 [ # # ]: 0 : if (strcmp(name, desc))
4637 : 0 : pr_cont(" (%s)", desc);
4638 : 0 : pr_cont("\n");
4639 : : }
4640 : : }
4641 : :
4642 : 0 : static void pr_cont_pool_info(struct worker_pool *pool)
4643 : : {
4644 : 0 : pr_cont(" cpus=%*pbl", nr_cpumask_bits, pool->attrs->cpumask);
4645 [ # # ]: 0 : if (pool->node != NUMA_NO_NODE)
4646 : 0 : pr_cont(" node=%d", pool->node);
4647 : 0 : pr_cont(" flags=0x%x nice=%d", pool->flags, pool->attrs->nice);
4648 : 0 : }
4649 : :
4650 : 0 : static void pr_cont_work(bool comma, struct work_struct *work)
4651 : : {
4652 [ # # ]: 0 : if (work->func == wq_barrier_func) {
4653 : 0 : struct wq_barrier *barr;
4654 : :
4655 : 0 : barr = container_of(work, struct wq_barrier, work);
4656 : :
4657 [ # # ]: 0 : pr_cont("%s BAR(%d)", comma ? "," : "",
4658 : : task_pid_nr(barr->task));
4659 : : } else {
4660 [ # # ]: 0 : pr_cont("%s %ps", comma ? "," : "", work->func);
4661 : : }
4662 : 0 : }
4663 : :
4664 : 0 : static void show_pwq(struct pool_workqueue *pwq)
4665 : : {
4666 : 0 : struct worker_pool *pool = pwq->pool;
4667 : 0 : struct work_struct *work;
4668 : 0 : struct worker *worker;
4669 : 0 : bool has_in_flight = false, has_pending = false;
4670 : 0 : int bkt;
4671 : :
4672 : 0 : pr_info(" pwq %d:", pool->id);
4673 : 0 : pr_cont_pool_info(pool);
4674 : :
4675 [ # # ]: 0 : pr_cont(" active=%d/%d refcnt=%d%s\n",
4676 : : pwq->nr_active, pwq->max_active, pwq->refcnt,
4677 : : !list_empty(&pwq->mayday_node) ? " MAYDAY" : "");
4678 : :
4679 [ # # # # : 0 : hash_for_each(pool->busy_hash, bkt, worker, hentry) {
# # # # ]
4680 [ # # ]: 0 : if (worker->current_pwq == pwq) {
4681 : : has_in_flight = true;
4682 : : break;
4683 : : }
4684 : : }
4685 [ # # ]: 0 : if (has_in_flight) {
4686 : 0 : bool comma = false;
4687 : :
4688 : 0 : pr_info(" in-flight:");
4689 [ # # # # : 0 : hash_for_each(pool->busy_hash, bkt, worker, hentry) {
# # # # ]
4690 [ # # ]: 0 : if (worker->current_pwq != pwq)
4691 : 0 : continue;
4692 : :
4693 [ # # # # ]: 0 : pr_cont("%s %d%s:%ps", comma ? "," : "",
4694 : : task_pid_nr(worker->task),
4695 : : worker->rescue_wq ? "(RESCUER)" : "",
4696 : : worker->current_func);
4697 [ # # ]: 0 : list_for_each_entry(work, &worker->scheduled, entry)
4698 : 0 : pr_cont_work(false, work);
4699 : : comma = true;
4700 : : }
4701 : 0 : pr_cont("\n");
4702 : : }
4703 : :
4704 [ # # ]: 0 : list_for_each_entry(work, &pool->worklist, entry) {
4705 [ # # ]: 0 : if (get_work_pwq(work) == pwq) {
4706 : : has_pending = true;
4707 : : break;
4708 : : }
4709 : : }
4710 [ # # ]: 0 : if (has_pending) {
4711 : 0 : bool comma = false;
4712 : :
4713 : 0 : pr_info(" pending:");
4714 [ # # ]: 0 : list_for_each_entry(work, &pool->worklist, entry) {
4715 [ # # ]: 0 : if (get_work_pwq(work) != pwq)
4716 : 0 : continue;
4717 : :
4718 : 0 : pr_cont_work(comma, work);
4719 : 0 : comma = !(*work_data_bits(work) & WORK_STRUCT_LINKED);
4720 : : }
4721 : 0 : pr_cont("\n");
4722 : : }
4723 : :
4724 [ # # ]: 0 : if (!list_empty(&pwq->delayed_works)) {
4725 : 0 : bool comma = false;
4726 : :
4727 : 0 : pr_info(" delayed:");
4728 [ # # ]: 0 : list_for_each_entry(work, &pwq->delayed_works, entry) {
4729 : 0 : pr_cont_work(comma, work);
4730 : 0 : comma = !(*work_data_bits(work) & WORK_STRUCT_LINKED);
4731 : : }
4732 : 0 : pr_cont("\n");
4733 : : }
4734 : 0 : }
4735 : :
4736 : : /**
4737 : : * show_workqueue_state - dump workqueue state
4738 : : *
4739 : : * Called from a sysrq handler or try_to_freeze_tasks() and prints out
4740 : : * all busy workqueues and pools.
4741 : : */
4742 : 0 : void show_workqueue_state(void)
4743 : : {
4744 : 0 : struct workqueue_struct *wq;
4745 : 0 : struct worker_pool *pool;
4746 : 0 : unsigned long flags;
4747 : 0 : int pi;
4748 : :
4749 : 0 : rcu_read_lock();
4750 : :
4751 : 0 : pr_info("Showing busy workqueues and worker pools:\n");
4752 : :
4753 [ # # ]: 0 : list_for_each_entry_rcu(wq, &workqueues, list) {
4754 : 0 : struct pool_workqueue *pwq;
4755 : 0 : bool idle = true;
4756 : :
4757 [ # # ]: 0 : for_each_pwq(pwq, wq) {
4758 [ # # # # ]: 0 : if (pwq->nr_active || !list_empty(&pwq->delayed_works)) {
4759 : 0 : idle = false;
4760 : 0 : break;
4761 : : }
4762 : : }
4763 : 0 : if (idle)
4764 : 0 : continue;
4765 : :
4766 : 0 : pr_info("workqueue %s: flags=0x%x\n", wq->name, wq->flags);
4767 : :
4768 [ # # ]: 0 : for_each_pwq(pwq, wq) {
4769 : 0 : spin_lock_irqsave(&pwq->pool->lock, flags);
4770 [ # # # # ]: 0 : if (pwq->nr_active || !list_empty(&pwq->delayed_works))
4771 : 0 : show_pwq(pwq);
4772 : 0 : spin_unlock_irqrestore(&pwq->pool->lock, flags);
4773 : : /*
4774 : : * We could be printing a lot from atomic context, e.g.
4775 : : * sysrq-t -> show_workqueue_state(). Avoid triggering
4776 : : * hard lockup.
4777 : : */
4778 : 0 : touch_nmi_watchdog();
4779 : : }
4780 : : }
4781 : :
4782 [ # # ]: 0 : for_each_pool(pool, pi) {
4783 : 0 : struct worker *worker;
4784 : 0 : bool first = true;
4785 : :
4786 : 0 : spin_lock_irqsave(&pool->lock, flags);
4787 [ # # ]: 0 : if (pool->nr_workers == pool->nr_idle)
4788 : 0 : goto next_pool;
4789 : :
4790 : 0 : pr_info("pool %d:", pool->id);
4791 : 0 : pr_cont_pool_info(pool);
4792 : 0 : pr_cont(" hung=%us workers=%d",
4793 : : jiffies_to_msecs(jiffies - pool->watchdog_ts) / 1000,
4794 : : pool->nr_workers);
4795 [ # # ]: 0 : if (pool->manager)
4796 : 0 : pr_cont(" manager: %d",
4797 : : task_pid_nr(pool->manager->task));
4798 [ # # ]: 0 : list_for_each_entry(worker, &pool->idle_list, entry) {
4799 [ # # ]: 0 : pr_cont(" %s%d", first ? "idle: " : "",
4800 : : task_pid_nr(worker->task));
4801 : 0 : first = false;
4802 : : }
4803 : 0 : pr_cont("\n");
4804 : 0 : next_pool:
4805 : 0 : spin_unlock_irqrestore(&pool->lock, flags);
4806 : : /*
4807 : : * We could be printing a lot from atomic context, e.g.
4808 : : * sysrq-t -> show_workqueue_state(). Avoid triggering
4809 : : * hard lockup.
4810 : : */
4811 : : touch_nmi_watchdog();
4812 : : }
4813 : :
4814 : 0 : rcu_read_unlock();
4815 : 0 : }
4816 : :
4817 : : /* used to show worker information through /proc/PID/{comm,stat,status} */
4818 : 330 : void wq_worker_comm(char *buf, size_t size, struct task_struct *task)
4819 : : {
4820 : 330 : int off;
4821 : :
4822 : : /* always show the actual comm */
4823 : 330 : off = strscpy(buf, task->comm, size);
4824 [ + - ]: 330 : if (off < 0)
4825 : : return;
4826 : :
4827 : : /* stabilize PF_WQ_WORKER and worker pool association */
4828 : 330 : mutex_lock(&wq_pool_attach_mutex);
4829 : :
4830 [ + - ]: 330 : if (task->flags & PF_WQ_WORKER) {
4831 : 330 : struct worker *worker = kthread_data(task);
4832 : 330 : struct worker_pool *pool = worker->pool;
4833 : :
4834 [ + + ]: 330 : if (pool) {
4835 : 132 : spin_lock_irq(&pool->lock);
4836 : : /*
4837 : : * ->desc tracks information (wq name or
4838 : : * set_worker_desc()) for the latest execution. If
4839 : : * current, prepend '+', otherwise '-'.
4840 : : */
4841 [ + + ]: 132 : if (worker->desc[0] != '\0') {
4842 [ - + ]: 88 : if (worker->current_work)
4843 : 0 : scnprintf(buf + off, size - off, "+%s",
4844 : 0 : worker->desc);
4845 : : else
4846 : 88 : scnprintf(buf + off, size - off, "-%s",
4847 : 88 : worker->desc);
4848 : : }
4849 : 132 : spin_unlock_irq(&pool->lock);
4850 : : }
4851 : : }
4852 : :
4853 : 330 : mutex_unlock(&wq_pool_attach_mutex);
4854 : : }
4855 : :
4856 : : #ifdef CONFIG_SMP
4857 : :
4858 : : /*
4859 : : * CPU hotplug.
4860 : : *
4861 : : * There are two challenges in supporting CPU hotplug. Firstly, there
4862 : : * are a lot of assumptions on strong associations among work, pwq and
4863 : : * pool which make migrating pending and scheduled works very
4864 : : * difficult to implement without impacting hot paths. Secondly,
4865 : : * worker pools serve mix of short, long and very long running works making
4866 : : * blocked draining impractical.
4867 : : *
4868 : : * This is solved by allowing the pools to be disassociated from the CPU
4869 : : * running as an unbound one and allowing it to be reattached later if the
4870 : : * cpu comes back online.
4871 : : */
4872 : :
4873 : 0 : static void unbind_workers(int cpu)
4874 : : {
4875 : 0 : struct worker_pool *pool;
4876 : 0 : struct worker *worker;
4877 : :
4878 [ # # ]: 0 : for_each_cpu_worker_pool(pool, cpu) {
4879 : 0 : mutex_lock(&wq_pool_attach_mutex);
4880 : 0 : spin_lock_irq(&pool->lock);
4881 : :
4882 : : /*
4883 : : * We've blocked all attach/detach operations. Make all workers
4884 : : * unbound and set DISASSOCIATED. Before this, all workers
4885 : : * except for the ones which are still executing works from
4886 : : * before the last CPU down must be on the cpu. After
4887 : : * this, they may become diasporas.
4888 : : */
4889 [ # # ]: 0 : for_each_pool_worker(worker, pool)
4890 : 0 : worker->flags |= WORKER_UNBOUND;
4891 : :
4892 : 0 : pool->flags |= POOL_DISASSOCIATED;
4893 : :
4894 : 0 : spin_unlock_irq(&pool->lock);
4895 : 0 : mutex_unlock(&wq_pool_attach_mutex);
4896 : :
4897 : : /*
4898 : : * Call schedule() so that we cross rq->lock and thus can
4899 : : * guarantee sched callbacks see the %WORKER_UNBOUND flag.
4900 : : * This is necessary as scheduler callbacks may be invoked
4901 : : * from other cpus.
4902 : : */
4903 : 0 : schedule();
4904 : :
4905 : : /*
4906 : : * Sched callbacks are disabled now. Zap nr_running.
4907 : : * After this, nr_running stays zero and need_more_worker()
4908 : : * and keep_working() are always true as long as the
4909 : : * worklist is not empty. This pool now behaves as an
4910 : : * unbound (in terms of concurrency management) pool which
4911 : : * are served by workers tied to the pool.
4912 : : */
4913 : 0 : atomic_set(&pool->nr_running, 0);
4914 : :
4915 : : /*
4916 : : * With concurrency management just turned off, a busy
4917 : : * worker blocking could lead to lengthy stalls. Kick off
4918 : : * unbound chain execution of currently pending work items.
4919 : : */
4920 : 0 : spin_lock_irq(&pool->lock);
4921 [ # # ]: 0 : wake_up_worker(pool);
4922 : 0 : spin_unlock_irq(&pool->lock);
4923 : : }
4924 : 0 : }
4925 : :
4926 : : /**
4927 : : * rebind_workers - rebind all workers of a pool to the associated CPU
4928 : : * @pool: pool of interest
4929 : : *
4930 : : * @pool->cpu is coming online. Rebind all workers to the CPU.
4931 : : */
4932 : 0 : static void rebind_workers(struct worker_pool *pool)
4933 : : {
4934 : 0 : struct worker *worker;
4935 : :
4936 : 0 : lockdep_assert_held(&wq_pool_attach_mutex);
4937 : :
4938 : : /*
4939 : : * Restore CPU affinity of all workers. As all idle workers should
4940 : : * be on the run-queue of the associated CPU before any local
4941 : : * wake-ups for concurrency management happen, restore CPU affinity
4942 : : * of all workers first and then clear UNBOUND. As we're called
4943 : : * from CPU_ONLINE, the following shouldn't fail.
4944 : : */
4945 [ # # ]: 0 : for_each_pool_worker(worker, pool)
4946 [ # # ]: 0 : WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task,
4947 : : pool->attrs->cpumask) < 0);
4948 : :
4949 : 0 : spin_lock_irq(&pool->lock);
4950 : :
4951 : 0 : pool->flags &= ~POOL_DISASSOCIATED;
4952 : :
4953 [ # # ]: 0 : for_each_pool_worker(worker, pool) {
4954 : 0 : unsigned int worker_flags = worker->flags;
4955 : :
4956 : : /*
4957 : : * A bound idle worker should actually be on the runqueue
4958 : : * of the associated CPU for local wake-ups targeting it to
4959 : : * work. Kick all idle workers so that they migrate to the
4960 : : * associated CPU. Doing this in the same loop as
4961 : : * replacing UNBOUND with REBOUND is safe as no worker will
4962 : : * be bound before @pool->lock is released.
4963 : : */
4964 [ # # ]: 0 : if (worker_flags & WORKER_IDLE)
4965 : 0 : wake_up_process(worker->task);
4966 : :
4967 : : /*
4968 : : * We want to clear UNBOUND but can't directly call
4969 : : * worker_clr_flags() or adjust nr_running. Atomically
4970 : : * replace UNBOUND with another NOT_RUNNING flag REBOUND.
4971 : : * @worker will clear REBOUND using worker_clr_flags() when
4972 : : * it initiates the next execution cycle thus restoring
4973 : : * concurrency management. Note that when or whether
4974 : : * @worker clears REBOUND doesn't affect correctness.
4975 : : *
4976 : : * WRITE_ONCE() is necessary because @worker->flags may be
4977 : : * tested without holding any lock in
4978 : : * wq_worker_running(). Without it, NOT_RUNNING test may
4979 : : * fail incorrectly leading to premature concurrency
4980 : : * management operations.
4981 : : */
4982 [ # # ]: 0 : WARN_ON_ONCE(!(worker_flags & WORKER_UNBOUND));
4983 : 0 : worker_flags |= WORKER_REBOUND;
4984 : 0 : worker_flags &= ~WORKER_UNBOUND;
4985 : 0 : WRITE_ONCE(worker->flags, worker_flags);
4986 : : }
4987 : :
4988 : 0 : spin_unlock_irq(&pool->lock);
4989 : 0 : }
4990 : :
4991 : : /**
4992 : : * restore_unbound_workers_cpumask - restore cpumask of unbound workers
4993 : : * @pool: unbound pool of interest
4994 : : * @cpu: the CPU which is coming up
4995 : : *
4996 : : * An unbound pool may end up with a cpumask which doesn't have any online
4997 : : * CPUs. When a worker of such pool get scheduled, the scheduler resets
4998 : : * its cpus_allowed. If @cpu is in @pool's cpumask which didn't have any
4999 : : * online CPU before, cpus_allowed of all its workers should be restored.
5000 : : */
5001 : 0 : static void restore_unbound_workers_cpumask(struct worker_pool *pool, int cpu)
5002 : : {
5003 : 0 : static cpumask_t cpumask;
5004 : 0 : struct worker *worker;
5005 : :
5006 : 0 : lockdep_assert_held(&wq_pool_attach_mutex);
5007 : :
5008 : : /* is @cpu allowed for @pool? */
5009 [ # # ]: 0 : if (!cpumask_test_cpu(cpu, pool->attrs->cpumask))
5010 : : return;
5011 : :
5012 : 0 : cpumask_and(&cpumask, pool->attrs->cpumask, cpu_online_mask);
5013 : :
5014 : : /* as we're called from CPU_ONLINE, the following shouldn't fail */
5015 [ # # ]: 0 : for_each_pool_worker(worker, pool)
5016 [ # # ]: 0 : WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task, &cpumask) < 0);
5017 : : }
5018 : :
5019 : 0 : int workqueue_prepare_cpu(unsigned int cpu)
5020 : : {
5021 : 0 : struct worker_pool *pool;
5022 : :
5023 [ # # ]: 0 : for_each_cpu_worker_pool(pool, cpu) {
5024 [ # # ]: 0 : if (pool->nr_workers)
5025 : 0 : continue;
5026 [ # # ]: 0 : if (!create_worker(pool))
5027 : : return -ENOMEM;
5028 : : }
5029 : : return 0;
5030 : : }
5031 : :
5032 : 0 : int workqueue_online_cpu(unsigned int cpu)
5033 : : {
5034 : 0 : struct worker_pool *pool;
5035 : 0 : struct workqueue_struct *wq;
5036 : 0 : int pi;
5037 : :
5038 : 0 : mutex_lock(&wq_pool_mutex);
5039 : :
5040 [ # # ]: 0 : for_each_pool(pool, pi) {
5041 : 0 : mutex_lock(&wq_pool_attach_mutex);
5042 : :
5043 [ # # ]: 0 : if (pool->cpu == cpu)
5044 : 0 : rebind_workers(pool);
5045 [ # # ]: 0 : else if (pool->cpu < 0)
5046 : 0 : restore_unbound_workers_cpumask(pool, cpu);
5047 : :
5048 : 0 : mutex_unlock(&wq_pool_attach_mutex);
5049 : : }
5050 : :
5051 : : /* update NUMA affinity of unbound workqueues */
5052 [ # # ]: 0 : list_for_each_entry(wq, &workqueues, list)
5053 : 0 : wq_update_unbound_numa(wq, cpu, true);
5054 : :
5055 : 0 : mutex_unlock(&wq_pool_mutex);
5056 : 0 : return 0;
5057 : : }
5058 : :
5059 : 0 : int workqueue_offline_cpu(unsigned int cpu)
5060 : : {
5061 : 0 : struct workqueue_struct *wq;
5062 : :
5063 : : /* unbinding per-cpu workers should happen on the local CPU */
5064 [ # # # # ]: 0 : if (WARN_ON(cpu != smp_processor_id()))
5065 : : return -1;
5066 : :
5067 : 0 : unbind_workers(cpu);
5068 : :
5069 : : /* update NUMA affinity of unbound workqueues */
5070 : 0 : mutex_lock(&wq_pool_mutex);
5071 [ # # ]: 0 : list_for_each_entry(wq, &workqueues, list)
5072 : 0 : wq_update_unbound_numa(wq, cpu, false);
5073 : 0 : mutex_unlock(&wq_pool_mutex);
5074 : :
5075 : 0 : return 0;
5076 : : }
5077 : :
5078 : : struct work_for_cpu {
5079 : : struct work_struct work;
5080 : : long (*fn)(void *);
5081 : : void *arg;
5082 : : long ret;
5083 : : };
5084 : :
5085 : 0 : static void work_for_cpu_fn(struct work_struct *work)
5086 : : {
5087 : 0 : struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work);
5088 : :
5089 : 0 : wfc->ret = wfc->fn(wfc->arg);
5090 : 0 : }
5091 : :
5092 : : /**
5093 : : * work_on_cpu - run a function in thread context on a particular cpu
5094 : : * @cpu: the cpu to run on
5095 : : * @fn: the function to run
5096 : : * @arg: the function arg
5097 : : *
5098 : : * It is up to the caller to ensure that the cpu doesn't go offline.
5099 : : * The caller must not hold any locks which would prevent @fn from completing.
5100 : : *
5101 : : * Return: The value @fn returns.
5102 : : */
5103 : 0 : long work_on_cpu(int cpu, long (*fn)(void *), void *arg)
5104 : : {
5105 : 0 : struct work_for_cpu wfc = { .fn = fn, .arg = arg };
5106 : :
5107 : 0 : INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn);
5108 : 0 : schedule_work_on(cpu, &wfc.work);
5109 : 0 : flush_work(&wfc.work);
5110 : 0 : destroy_work_on_stack(&wfc.work);
5111 : 0 : return wfc.ret;
5112 : : }
5113 : : EXPORT_SYMBOL_GPL(work_on_cpu);
5114 : :
5115 : : /**
5116 : : * work_on_cpu_safe - run a function in thread context on a particular cpu
5117 : : * @cpu: the cpu to run on
5118 : : * @fn: the function to run
5119 : : * @arg: the function argument
5120 : : *
5121 : : * Disables CPU hotplug and calls work_on_cpu(). The caller must not hold
5122 : : * any locks which would prevent @fn from completing.
5123 : : *
5124 : : * Return: The value @fn returns.
5125 : : */
5126 : 0 : long work_on_cpu_safe(int cpu, long (*fn)(void *), void *arg)
5127 : : {
5128 : 0 : long ret = -ENODEV;
5129 : :
5130 : 0 : get_online_cpus();
5131 [ # # ]: 0 : if (cpu_online(cpu))
5132 : 0 : ret = work_on_cpu(cpu, fn, arg);
5133 : 0 : put_online_cpus();
5134 : 0 : return ret;
5135 : : }
5136 : : EXPORT_SYMBOL_GPL(work_on_cpu_safe);
5137 : : #endif /* CONFIG_SMP */
5138 : :
5139 : : #ifdef CONFIG_FREEZER
5140 : :
5141 : : /**
5142 : : * freeze_workqueues_begin - begin freezing workqueues
5143 : : *
5144 : : * Start freezing workqueues. After this function returns, all freezable
5145 : : * workqueues will queue new works to their delayed_works list instead of
5146 : : * pool->worklist.
5147 : : *
5148 : : * CONTEXT:
5149 : : * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's.
5150 : : */
5151 : 0 : void freeze_workqueues_begin(void)
5152 : : {
5153 : 0 : struct workqueue_struct *wq;
5154 : 0 : struct pool_workqueue *pwq;
5155 : :
5156 : 0 : mutex_lock(&wq_pool_mutex);
5157 : :
5158 [ # # ]: 0 : WARN_ON_ONCE(workqueue_freezing);
5159 : 0 : workqueue_freezing = true;
5160 : :
5161 [ # # ]: 0 : list_for_each_entry(wq, &workqueues, list) {
5162 : 0 : mutex_lock(&wq->mutex);
5163 [ # # ]: 0 : for_each_pwq(pwq, wq)
5164 : 0 : pwq_adjust_max_active(pwq);
5165 : 0 : mutex_unlock(&wq->mutex);
5166 : : }
5167 : :
5168 : 0 : mutex_unlock(&wq_pool_mutex);
5169 : 0 : }
5170 : :
5171 : : /**
5172 : : * freeze_workqueues_busy - are freezable workqueues still busy?
5173 : : *
5174 : : * Check whether freezing is complete. This function must be called
5175 : : * between freeze_workqueues_begin() and thaw_workqueues().
5176 : : *
5177 : : * CONTEXT:
5178 : : * Grabs and releases wq_pool_mutex.
5179 : : *
5180 : : * Return:
5181 : : * %true if some freezable workqueues are still busy. %false if freezing
5182 : : * is complete.
5183 : : */
5184 : 0 : bool freeze_workqueues_busy(void)
5185 : : {
5186 : 0 : bool busy = false;
5187 : 0 : struct workqueue_struct *wq;
5188 : 0 : struct pool_workqueue *pwq;
5189 : :
5190 : 0 : mutex_lock(&wq_pool_mutex);
5191 : :
5192 [ # # ]: 0 : WARN_ON_ONCE(!workqueue_freezing);
5193 : :
5194 [ # # ]: 0 : list_for_each_entry(wq, &workqueues, list) {
5195 [ # # ]: 0 : if (!(wq->flags & WQ_FREEZABLE))
5196 : 0 : continue;
5197 : : /*
5198 : : * nr_active is monotonically decreasing. It's safe
5199 : : * to peek without lock.
5200 : : */
5201 : 0 : rcu_read_lock();
5202 [ # # ]: 0 : for_each_pwq(pwq, wq) {
5203 [ # # ]: 0 : WARN_ON_ONCE(pwq->nr_active < 0);
5204 [ # # ]: 0 : if (pwq->nr_active) {
5205 : 0 : busy = true;
5206 : 0 : rcu_read_unlock();
5207 : 0 : goto out_unlock;
5208 : : }
5209 : : }
5210 : 0 : rcu_read_unlock();
5211 : : }
5212 : 0 : out_unlock:
5213 : 0 : mutex_unlock(&wq_pool_mutex);
5214 : 0 : return busy;
5215 : : }
5216 : :
5217 : : /**
5218 : : * thaw_workqueues - thaw workqueues
5219 : : *
5220 : : * Thaw workqueues. Normal queueing is restored and all collected
5221 : : * frozen works are transferred to their respective pool worklists.
5222 : : *
5223 : : * CONTEXT:
5224 : : * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's.
5225 : : */
5226 : 0 : void thaw_workqueues(void)
5227 : : {
5228 : 0 : struct workqueue_struct *wq;
5229 : 0 : struct pool_workqueue *pwq;
5230 : :
5231 : 0 : mutex_lock(&wq_pool_mutex);
5232 : :
5233 [ # # ]: 0 : if (!workqueue_freezing)
5234 : 0 : goto out_unlock;
5235 : :
5236 : 0 : workqueue_freezing = false;
5237 : :
5238 : : /* restore max_active and repopulate worklist */
5239 [ # # ]: 0 : list_for_each_entry(wq, &workqueues, list) {
5240 : 0 : mutex_lock(&wq->mutex);
5241 [ # # ]: 0 : for_each_pwq(pwq, wq)
5242 : 0 : pwq_adjust_max_active(pwq);
5243 : 0 : mutex_unlock(&wq->mutex);
5244 : : }
5245 : :
5246 : 0 : out_unlock:
5247 : 0 : mutex_unlock(&wq_pool_mutex);
5248 : 0 : }
5249 : : #endif /* CONFIG_FREEZER */
5250 : :
5251 : 0 : static int workqueue_apply_unbound_cpumask(void)
5252 : : {
5253 : 0 : LIST_HEAD(ctxs);
5254 : 0 : int ret = 0;
5255 : 0 : struct workqueue_struct *wq;
5256 : 0 : struct apply_wqattrs_ctx *ctx, *n;
5257 : :
5258 : 0 : lockdep_assert_held(&wq_pool_mutex);
5259 : :
5260 [ # # ]: 0 : list_for_each_entry(wq, &workqueues, list) {
5261 [ # # ]: 0 : if (!(wq->flags & WQ_UNBOUND))
5262 : 0 : continue;
5263 : : /* creating multiple pwqs breaks ordering guarantee */
5264 [ # # ]: 0 : if (wq->flags & __WQ_ORDERED)
5265 : 0 : continue;
5266 : :
5267 : 0 : ctx = apply_wqattrs_prepare(wq, wq->unbound_attrs);
5268 [ # # ]: 0 : if (!ctx) {
5269 : : ret = -ENOMEM;
5270 : : break;
5271 : : }
5272 : :
5273 : 0 : list_add_tail(&ctx->list, &ctxs);
5274 : : }
5275 : :
5276 [ # # ]: 0 : list_for_each_entry_safe(ctx, n, &ctxs, list) {
5277 [ # # ]: 0 : if (!ret)
5278 : 0 : apply_wqattrs_commit(ctx);
5279 : 0 : apply_wqattrs_cleanup(ctx);
5280 : : }
5281 : :
5282 : 0 : return ret;
5283 : : }
5284 : :
5285 : : /**
5286 : : * workqueue_set_unbound_cpumask - Set the low-level unbound cpumask
5287 : : * @cpumask: the cpumask to set
5288 : : *
5289 : : * The low-level workqueues cpumask is a global cpumask that limits
5290 : : * the affinity of all unbound workqueues. This function check the @cpumask
5291 : : * and apply it to all unbound workqueues and updates all pwqs of them.
5292 : : *
5293 : : * Retun: 0 - Success
5294 : : * -EINVAL - Invalid @cpumask
5295 : : * -ENOMEM - Failed to allocate memory for attrs or pwqs.
5296 : : */
5297 : 0 : int workqueue_set_unbound_cpumask(cpumask_var_t cpumask)
5298 : : {
5299 : 0 : int ret = -EINVAL;
5300 : 0 : cpumask_var_t saved_cpumask;
5301 : :
5302 [ # # ]: 0 : if (!zalloc_cpumask_var(&saved_cpumask, GFP_KERNEL))
5303 : : return -ENOMEM;
5304 : :
5305 : : /*
5306 : : * Not excluding isolated cpus on purpose.
5307 : : * If the user wishes to include them, we allow that.
5308 : : */
5309 [ # # ]: 0 : cpumask_and(cpumask, cpumask, cpu_possible_mask);
5310 [ # # ]: 0 : if (!cpumask_empty(cpumask)) {
5311 : 0 : apply_wqattrs_lock();
5312 : :
5313 : : /* save the old wq_unbound_cpumask. */
5314 : 0 : cpumask_copy(saved_cpumask, wq_unbound_cpumask);
5315 : :
5316 : : /* update wq_unbound_cpumask at first and apply it to wqs. */
5317 : 0 : cpumask_copy(wq_unbound_cpumask, cpumask);
5318 : 0 : ret = workqueue_apply_unbound_cpumask();
5319 : :
5320 : : /* restore the wq_unbound_cpumask when failed. */
5321 [ # # ]: 0 : if (ret < 0)
5322 : 0 : cpumask_copy(wq_unbound_cpumask, saved_cpumask);
5323 : :
5324 : 0 : apply_wqattrs_unlock();
5325 : : }
5326 : :
5327 : 0 : free_cpumask_var(saved_cpumask);
5328 : 0 : return ret;
5329 : : }
5330 : :
5331 : : #ifdef CONFIG_SYSFS
5332 : : /*
5333 : : * Workqueues with WQ_SYSFS flag set is visible to userland via
5334 : : * /sys/bus/workqueue/devices/WQ_NAME. All visible workqueues have the
5335 : : * following attributes.
5336 : : *
5337 : : * per_cpu RO bool : whether the workqueue is per-cpu or unbound
5338 : : * max_active RW int : maximum number of in-flight work items
5339 : : *
5340 : : * Unbound workqueues have the following extra attributes.
5341 : : *
5342 : : * pool_ids RO int : the associated pool IDs for each node
5343 : : * nice RW int : nice value of the workers
5344 : : * cpumask RW mask : bitmask of allowed CPUs for the workers
5345 : : * numa RW bool : whether enable NUMA affinity
5346 : : */
5347 : : struct wq_device {
5348 : : struct workqueue_struct *wq;
5349 : : struct device dev;
5350 : : };
5351 : :
5352 : 0 : static struct workqueue_struct *dev_to_wq(struct device *dev)
5353 : : {
5354 : 0 : struct wq_device *wq_dev = container_of(dev, struct wq_device, dev);
5355 : :
5356 : 0 : return wq_dev->wq;
5357 : : }
5358 : :
5359 : 0 : static ssize_t per_cpu_show(struct device *dev, struct device_attribute *attr,
5360 : : char *buf)
5361 : : {
5362 : 0 : struct workqueue_struct *wq = dev_to_wq(dev);
5363 : :
5364 : 0 : return scnprintf(buf, PAGE_SIZE, "%d\n", (bool)!(wq->flags & WQ_UNBOUND));
5365 : : }
5366 : : static DEVICE_ATTR_RO(per_cpu);
5367 : :
5368 : 0 : static ssize_t max_active_show(struct device *dev,
5369 : : struct device_attribute *attr, char *buf)
5370 : : {
5371 : 0 : struct workqueue_struct *wq = dev_to_wq(dev);
5372 : :
5373 : 0 : return scnprintf(buf, PAGE_SIZE, "%d\n", wq->saved_max_active);
5374 : : }
5375 : :
5376 : 0 : static ssize_t max_active_store(struct device *dev,
5377 : : struct device_attribute *attr, const char *buf,
5378 : : size_t count)
5379 : : {
5380 : 0 : struct workqueue_struct *wq = dev_to_wq(dev);
5381 : 0 : int val;
5382 : :
5383 [ # # # # ]: 0 : if (sscanf(buf, "%d", &val) != 1 || val <= 0)
5384 : : return -EINVAL;
5385 : :
5386 : 0 : workqueue_set_max_active(wq, val);
5387 : 0 : return count;
5388 : : }
5389 : : static DEVICE_ATTR_RW(max_active);
5390 : :
5391 : : static struct attribute *wq_sysfs_attrs[] = {
5392 : : &dev_attr_per_cpu.attr,
5393 : : &dev_attr_max_active.attr,
5394 : : NULL,
5395 : : };
5396 : : ATTRIBUTE_GROUPS(wq_sysfs);
5397 : :
5398 : 0 : static ssize_t wq_pool_ids_show(struct device *dev,
5399 : : struct device_attribute *attr, char *buf)
5400 : : {
5401 : 0 : struct workqueue_struct *wq = dev_to_wq(dev);
5402 : 0 : const char *delim = "";
5403 : 0 : int node, written = 0;
5404 : :
5405 : 0 : get_online_cpus();
5406 : 0 : rcu_read_lock();
5407 [ # # ]: 0 : for_each_node(node) {
5408 [ # # ]: 0 : written += scnprintf(buf + written, PAGE_SIZE - written,
5409 : : "%s%d:%d", delim, node,
5410 : 0 : unbound_pwq_by_node(wq, node)->pool->id);
5411 : 0 : delim = " ";
5412 : : }
5413 : 0 : written += scnprintf(buf + written, PAGE_SIZE - written, "\n");
5414 : 0 : rcu_read_unlock();
5415 : 0 : put_online_cpus();
5416 : :
5417 : 0 : return written;
5418 : : }
5419 : :
5420 : 0 : static ssize_t wq_nice_show(struct device *dev, struct device_attribute *attr,
5421 : : char *buf)
5422 : : {
5423 : 0 : struct workqueue_struct *wq = dev_to_wq(dev);
5424 : 0 : int written;
5425 : :
5426 : 0 : mutex_lock(&wq->mutex);
5427 : 0 : written = scnprintf(buf, PAGE_SIZE, "%d\n", wq->unbound_attrs->nice);
5428 : 0 : mutex_unlock(&wq->mutex);
5429 : :
5430 : 0 : return written;
5431 : : }
5432 : :
5433 : : /* prepare workqueue_attrs for sysfs store operations */
5434 : 0 : static struct workqueue_attrs *wq_sysfs_prep_attrs(struct workqueue_struct *wq)
5435 : : {
5436 : 0 : struct workqueue_attrs *attrs;
5437 : :
5438 : 0 : lockdep_assert_held(&wq_pool_mutex);
5439 : :
5440 : 0 : attrs = alloc_workqueue_attrs();
5441 [ # # ]: 0 : if (!attrs)
5442 : : return NULL;
5443 : :
5444 : 0 : copy_workqueue_attrs(attrs, wq->unbound_attrs);
5445 : 0 : return attrs;
5446 : : }
5447 : :
5448 : 0 : static ssize_t wq_nice_store(struct device *dev, struct device_attribute *attr,
5449 : : const char *buf, size_t count)
5450 : : {
5451 : 0 : struct workqueue_struct *wq = dev_to_wq(dev);
5452 : 0 : struct workqueue_attrs *attrs;
5453 : 0 : int ret = -ENOMEM;
5454 : :
5455 : 0 : apply_wqattrs_lock();
5456 : :
5457 : 0 : attrs = wq_sysfs_prep_attrs(wq);
5458 [ # # ]: 0 : if (!attrs)
5459 : 0 : goto out_unlock;
5460 : :
5461 [ # # ]: 0 : if (sscanf(buf, "%d", &attrs->nice) == 1 &&
5462 [ # # # # ]: 0 : attrs->nice >= MIN_NICE && attrs->nice <= MAX_NICE)
5463 : 0 : ret = apply_workqueue_attrs_locked(wq, attrs);
5464 : : else
5465 : : ret = -EINVAL;
5466 : :
5467 : 0 : out_unlock:
5468 : 0 : apply_wqattrs_unlock();
5469 [ # # ]: 0 : free_workqueue_attrs(attrs);
5470 [ # # ]: 0 : return ret ?: count;
5471 : : }
5472 : :
5473 : 0 : static ssize_t wq_cpumask_show(struct device *dev,
5474 : : struct device_attribute *attr, char *buf)
5475 : : {
5476 : 0 : struct workqueue_struct *wq = dev_to_wq(dev);
5477 : 0 : int written;
5478 : :
5479 : 0 : mutex_lock(&wq->mutex);
5480 : 0 : written = scnprintf(buf, PAGE_SIZE, "%*pb\n",
5481 : 0 : cpumask_pr_args(wq->unbound_attrs->cpumask));
5482 : 0 : mutex_unlock(&wq->mutex);
5483 : 0 : return written;
5484 : : }
5485 : :
5486 : 0 : static ssize_t wq_cpumask_store(struct device *dev,
5487 : : struct device_attribute *attr,
5488 : : const char *buf, size_t count)
5489 : : {
5490 : 0 : struct workqueue_struct *wq = dev_to_wq(dev);
5491 : 0 : struct workqueue_attrs *attrs;
5492 : 0 : int ret = -ENOMEM;
5493 : :
5494 : 0 : apply_wqattrs_lock();
5495 : :
5496 : 0 : attrs = wq_sysfs_prep_attrs(wq);
5497 [ # # ]: 0 : if (!attrs)
5498 : 0 : goto out_unlock;
5499 : :
5500 : 0 : ret = cpumask_parse(buf, attrs->cpumask);
5501 [ # # ]: 0 : if (!ret)
5502 : 0 : ret = apply_workqueue_attrs_locked(wq, attrs);
5503 : :
5504 : 0 : out_unlock:
5505 : 0 : apply_wqattrs_unlock();
5506 [ # # ]: 0 : free_workqueue_attrs(attrs);
5507 [ # # ]: 0 : return ret ?: count;
5508 : : }
5509 : :
5510 : 0 : static ssize_t wq_numa_show(struct device *dev, struct device_attribute *attr,
5511 : : char *buf)
5512 : : {
5513 : 0 : struct workqueue_struct *wq = dev_to_wq(dev);
5514 : 0 : int written;
5515 : :
5516 : 0 : mutex_lock(&wq->mutex);
5517 : 0 : written = scnprintf(buf, PAGE_SIZE, "%d\n",
5518 : 0 : !wq->unbound_attrs->no_numa);
5519 : 0 : mutex_unlock(&wq->mutex);
5520 : :
5521 : 0 : return written;
5522 : : }
5523 : :
5524 : 0 : static ssize_t wq_numa_store(struct device *dev, struct device_attribute *attr,
5525 : : const char *buf, size_t count)
5526 : : {
5527 : 0 : struct workqueue_struct *wq = dev_to_wq(dev);
5528 : 0 : struct workqueue_attrs *attrs;
5529 : 0 : int v, ret = -ENOMEM;
5530 : :
5531 : 0 : apply_wqattrs_lock();
5532 : :
5533 : 0 : attrs = wq_sysfs_prep_attrs(wq);
5534 [ # # ]: 0 : if (!attrs)
5535 : 0 : goto out_unlock;
5536 : :
5537 : 0 : ret = -EINVAL;
5538 [ # # ]: 0 : if (sscanf(buf, "%d", &v) == 1) {
5539 : 0 : attrs->no_numa = !v;
5540 : 0 : ret = apply_workqueue_attrs_locked(wq, attrs);
5541 : : }
5542 : :
5543 : 0 : out_unlock:
5544 : 0 : apply_wqattrs_unlock();
5545 [ # # ]: 0 : free_workqueue_attrs(attrs);
5546 [ # # ]: 0 : return ret ?: count;
5547 : : }
5548 : :
5549 : : static struct device_attribute wq_sysfs_unbound_attrs[] = {
5550 : : __ATTR(pool_ids, 0444, wq_pool_ids_show, NULL),
5551 : : __ATTR(nice, 0644, wq_nice_show, wq_nice_store),
5552 : : __ATTR(cpumask, 0644, wq_cpumask_show, wq_cpumask_store),
5553 : : __ATTR(numa, 0644, wq_numa_show, wq_numa_store),
5554 : : __ATTR_NULL,
5555 : : };
5556 : :
5557 : : static struct bus_type wq_subsys = {
5558 : : .name = "workqueue",
5559 : : .dev_groups = wq_sysfs_groups,
5560 : : };
5561 : :
5562 : 0 : static ssize_t wq_unbound_cpumask_show(struct device *dev,
5563 : : struct device_attribute *attr, char *buf)
5564 : : {
5565 : 0 : int written;
5566 : :
5567 : 0 : mutex_lock(&wq_pool_mutex);
5568 : 0 : written = scnprintf(buf, PAGE_SIZE, "%*pb\n",
5569 : : cpumask_pr_args(wq_unbound_cpumask));
5570 : 0 : mutex_unlock(&wq_pool_mutex);
5571 : :
5572 : 0 : return written;
5573 : : }
5574 : :
5575 : 0 : static ssize_t wq_unbound_cpumask_store(struct device *dev,
5576 : : struct device_attribute *attr, const char *buf, size_t count)
5577 : : {
5578 : 0 : cpumask_var_t cpumask;
5579 : 0 : int ret;
5580 : :
5581 : 0 : if (!zalloc_cpumask_var(&cpumask, GFP_KERNEL))
5582 : : return -ENOMEM;
5583 : :
5584 : 0 : ret = cpumask_parse(buf, cpumask);
5585 [ # # ]: 0 : if (!ret)
5586 : 0 : ret = workqueue_set_unbound_cpumask(cpumask);
5587 : :
5588 [ # # ]: 0 : free_cpumask_var(cpumask);
5589 [ # # ]: 0 : return ret ? ret : count;
5590 : : }
5591 : :
5592 : : static struct device_attribute wq_sysfs_cpumask_attr =
5593 : : __ATTR(cpumask, 0644, wq_unbound_cpumask_show,
5594 : : wq_unbound_cpumask_store);
5595 : :
5596 : 11 : static int __init wq_sysfs_init(void)
5597 : : {
5598 : 11 : int err;
5599 : :
5600 : 11 : err = subsys_virtual_register(&wq_subsys, NULL);
5601 [ + - ]: 11 : if (err)
5602 : : return err;
5603 : :
5604 : 11 : return device_create_file(wq_subsys.dev_root, &wq_sysfs_cpumask_attr);
5605 : : }
5606 : : core_initcall(wq_sysfs_init);
5607 : :
5608 : 0 : static void wq_device_release(struct device *dev)
5609 : : {
5610 : 0 : struct wq_device *wq_dev = container_of(dev, struct wq_device, dev);
5611 : :
5612 : 0 : kfree(wq_dev);
5613 : 0 : }
5614 : :
5615 : : /**
5616 : : * workqueue_sysfs_register - make a workqueue visible in sysfs
5617 : : * @wq: the workqueue to register
5618 : : *
5619 : : * Expose @wq in sysfs under /sys/bus/workqueue/devices.
5620 : : * alloc_workqueue*() automatically calls this function if WQ_SYSFS is set
5621 : : * which is the preferred method.
5622 : : *
5623 : : * Workqueue user should use this function directly iff it wants to apply
5624 : : * workqueue_attrs before making the workqueue visible in sysfs; otherwise,
5625 : : * apply_workqueue_attrs() may race against userland updating the
5626 : : * attributes.
5627 : : *
5628 : : * Return: 0 on success, -errno on failure.
5629 : : */
5630 : 11 : int workqueue_sysfs_register(struct workqueue_struct *wq)
5631 : : {
5632 : 11 : struct wq_device *wq_dev;
5633 : 11 : int ret;
5634 : :
5635 : : /*
5636 : : * Adjusting max_active or creating new pwqs by applying
5637 : : * attributes breaks ordering guarantee. Disallow exposing ordered
5638 : : * workqueues.
5639 : : */
5640 [ - + + - ]: 11 : if (WARN_ON(wq->flags & __WQ_ORDERED_EXPLICIT))
5641 : : return -EINVAL;
5642 : :
5643 : 11 : wq->wq_dev = wq_dev = kzalloc(sizeof(*wq_dev), GFP_KERNEL);
5644 [ + - ]: 11 : if (!wq_dev)
5645 : : return -ENOMEM;
5646 : :
5647 : 11 : wq_dev->wq = wq;
5648 : 11 : wq_dev->dev.bus = &wq_subsys;
5649 : 11 : wq_dev->dev.release = wq_device_release;
5650 : 11 : dev_set_name(&wq_dev->dev, "%s", wq->name);
5651 : :
5652 : : /*
5653 : : * unbound_attrs are created separately. Suppress uevent until
5654 : : * everything is ready.
5655 : : */
5656 : 11 : dev_set_uevent_suppress(&wq_dev->dev, true);
5657 : :
5658 : 11 : ret = device_register(&wq_dev->dev);
5659 [ - + ]: 11 : if (ret) {
5660 : 0 : put_device(&wq_dev->dev);
5661 : 0 : wq->wq_dev = NULL;
5662 : 0 : return ret;
5663 : : }
5664 : :
5665 [ + - ]: 11 : if (wq->flags & WQ_UNBOUND) {
5666 : : struct device_attribute *attr;
5667 : :
5668 [ + + ]: 55 : for (attr = wq_sysfs_unbound_attrs; attr->attr.name; attr++) {
5669 : 44 : ret = device_create_file(&wq_dev->dev, attr);
5670 [ - + ]: 44 : if (ret) {
5671 : 0 : device_unregister(&wq_dev->dev);
5672 : 0 : wq->wq_dev = NULL;
5673 : 0 : return ret;
5674 : : }
5675 : : }
5676 : : }
5677 : :
5678 : 11 : dev_set_uevent_suppress(&wq_dev->dev, false);
5679 : 11 : kobject_uevent(&wq_dev->dev.kobj, KOBJ_ADD);
5680 : 11 : return 0;
5681 : : }
5682 : :
5683 : : /**
5684 : : * workqueue_sysfs_unregister - undo workqueue_sysfs_register()
5685 : : * @wq: the workqueue to unregister
5686 : : *
5687 : : * If @wq is registered to sysfs by workqueue_sysfs_register(), unregister.
5688 : : */
5689 : 11 : static void workqueue_sysfs_unregister(struct workqueue_struct *wq)
5690 : : {
5691 : 11 : struct wq_device *wq_dev = wq->wq_dev;
5692 : :
5693 [ - + ]: 11 : if (!wq->wq_dev)
5694 : : return;
5695 : :
5696 : 0 : wq->wq_dev = NULL;
5697 : 0 : device_unregister(&wq_dev->dev);
5698 : : }
5699 : : #else /* CONFIG_SYSFS */
5700 : : static void workqueue_sysfs_unregister(struct workqueue_struct *wq) { }
5701 : : #endif /* CONFIG_SYSFS */
5702 : :
5703 : : /*
5704 : : * Workqueue watchdog.
5705 : : *
5706 : : * Stall may be caused by various bugs - missing WQ_MEM_RECLAIM, illegal
5707 : : * flush dependency, a concurrency managed work item which stays RUNNING
5708 : : * indefinitely. Workqueue stalls can be very difficult to debug as the
5709 : : * usual warning mechanisms don't trigger and internal workqueue state is
5710 : : * largely opaque.
5711 : : *
5712 : : * Workqueue watchdog monitors all worker pools periodically and dumps
5713 : : * state if some pools failed to make forward progress for a while where
5714 : : * forward progress is defined as the first item on ->worklist changing.
5715 : : *
5716 : : * This mechanism is controlled through the kernel parameter
5717 : : * "workqueue.watchdog_thresh" which can be updated at runtime through the
5718 : : * corresponding sysfs parameter file.
5719 : : */
5720 : : #ifdef CONFIG_WQ_WATCHDOG
5721 : :
5722 : : static unsigned long wq_watchdog_thresh = 30;
5723 : : static struct timer_list wq_watchdog_timer;
5724 : :
5725 : : static unsigned long wq_watchdog_touched = INITIAL_JIFFIES;
5726 : : static DEFINE_PER_CPU(unsigned long, wq_watchdog_touched_cpu) = INITIAL_JIFFIES;
5727 : :
5728 : : static void wq_watchdog_reset_touched(void)
5729 : : {
5730 : : int cpu;
5731 : :
5732 : : wq_watchdog_touched = jiffies;
5733 : : for_each_possible_cpu(cpu)
5734 : : per_cpu(wq_watchdog_touched_cpu, cpu) = jiffies;
5735 : : }
5736 : :
5737 : : static void wq_watchdog_timer_fn(struct timer_list *unused)
5738 : : {
5739 : : unsigned long thresh = READ_ONCE(wq_watchdog_thresh) * HZ;
5740 : : bool lockup_detected = false;
5741 : : struct worker_pool *pool;
5742 : : int pi;
5743 : :
5744 : : if (!thresh)
5745 : : return;
5746 : :
5747 : : rcu_read_lock();
5748 : :
5749 : : for_each_pool(pool, pi) {
5750 : : unsigned long pool_ts, touched, ts;
5751 : :
5752 : : if (list_empty(&pool->worklist))
5753 : : continue;
5754 : :
5755 : : /* get the latest of pool and touched timestamps */
5756 : : pool_ts = READ_ONCE(pool->watchdog_ts);
5757 : : touched = READ_ONCE(wq_watchdog_touched);
5758 : :
5759 : : if (time_after(pool_ts, touched))
5760 : : ts = pool_ts;
5761 : : else
5762 : : ts = touched;
5763 : :
5764 : : if (pool->cpu >= 0) {
5765 : : unsigned long cpu_touched =
5766 : : READ_ONCE(per_cpu(wq_watchdog_touched_cpu,
5767 : : pool->cpu));
5768 : : if (time_after(cpu_touched, ts))
5769 : : ts = cpu_touched;
5770 : : }
5771 : :
5772 : : /* did we stall? */
5773 : : if (time_after(jiffies, ts + thresh)) {
5774 : : lockup_detected = true;
5775 : : pr_emerg("BUG: workqueue lockup - pool");
5776 : : pr_cont_pool_info(pool);
5777 : : pr_cont(" stuck for %us!\n",
5778 : : jiffies_to_msecs(jiffies - pool_ts) / 1000);
5779 : : }
5780 : : }
5781 : :
5782 : : rcu_read_unlock();
5783 : :
5784 : : if (lockup_detected)
5785 : : show_workqueue_state();
5786 : :
5787 : : wq_watchdog_reset_touched();
5788 : : mod_timer(&wq_watchdog_timer, jiffies + thresh);
5789 : : }
5790 : :
5791 : : notrace void wq_watchdog_touch(int cpu)
5792 : : {
5793 : : if (cpu >= 0)
5794 : : per_cpu(wq_watchdog_touched_cpu, cpu) = jiffies;
5795 : : else
5796 : : wq_watchdog_touched = jiffies;
5797 : : }
5798 : :
5799 : : static void wq_watchdog_set_thresh(unsigned long thresh)
5800 : : {
5801 : : wq_watchdog_thresh = 0;
5802 : : del_timer_sync(&wq_watchdog_timer);
5803 : :
5804 : : if (thresh) {
5805 : : wq_watchdog_thresh = thresh;
5806 : : wq_watchdog_reset_touched();
5807 : : mod_timer(&wq_watchdog_timer, jiffies + thresh * HZ);
5808 : : }
5809 : : }
5810 : :
5811 : : static int wq_watchdog_param_set_thresh(const char *val,
5812 : : const struct kernel_param *kp)
5813 : : {
5814 : : unsigned long thresh;
5815 : : int ret;
5816 : :
5817 : : ret = kstrtoul(val, 0, &thresh);
5818 : : if (ret)
5819 : : return ret;
5820 : :
5821 : : if (system_wq)
5822 : : wq_watchdog_set_thresh(thresh);
5823 : : else
5824 : : wq_watchdog_thresh = thresh;
5825 : :
5826 : : return 0;
5827 : : }
5828 : :
5829 : : static const struct kernel_param_ops wq_watchdog_thresh_ops = {
5830 : : .set = wq_watchdog_param_set_thresh,
5831 : : .get = param_get_ulong,
5832 : : };
5833 : :
5834 : : module_param_cb(watchdog_thresh, &wq_watchdog_thresh_ops, &wq_watchdog_thresh,
5835 : : 0644);
5836 : :
5837 : : static void wq_watchdog_init(void)
5838 : : {
5839 : : timer_setup(&wq_watchdog_timer, wq_watchdog_timer_fn, TIMER_DEFERRABLE);
5840 : : wq_watchdog_set_thresh(wq_watchdog_thresh);
5841 : : }
5842 : :
5843 : : #else /* CONFIG_WQ_WATCHDOG */
5844 : :
5845 : 11 : static inline void wq_watchdog_init(void) { }
5846 : :
5847 : : #endif /* CONFIG_WQ_WATCHDOG */
5848 : :
5849 : 11 : static void __init wq_numa_init(void)
5850 : : {
5851 : 11 : cpumask_var_t *tbl;
5852 : 11 : int node, cpu;
5853 : :
5854 [ - + ]: 11 : if (num_possible_nodes() <= 1)
5855 : : return;
5856 : :
5857 [ # # ]: 0 : if (wq_disable_numa) {
5858 : 0 : pr_info("workqueue: NUMA affinity support disabled\n");
5859 : 0 : return;
5860 : : }
5861 : :
5862 : 0 : wq_update_unbound_numa_attrs_buf = alloc_workqueue_attrs();
5863 [ # # ]: 0 : BUG_ON(!wq_update_unbound_numa_attrs_buf);
5864 : :
5865 : : /*
5866 : : * We want masks of possible CPUs of each node which isn't readily
5867 : : * available. Build one from cpu_to_node() which should have been
5868 : : * fully initialized by now.
5869 : : */
5870 : 0 : tbl = kcalloc(nr_node_ids, sizeof(tbl[0]), GFP_KERNEL);
5871 [ # # ]: 0 : BUG_ON(!tbl);
5872 : :
5873 [ # # ]: 0 : for_each_node(node)
5874 : 0 : BUG_ON(!zalloc_cpumask_var_node(&tbl[node], GFP_KERNEL,
5875 : : node_online(node) ? node : NUMA_NO_NODE));
5876 : :
5877 [ # # ]: 0 : for_each_possible_cpu(cpu) {
5878 [ # # ]: 0 : node = cpu_to_node(cpu);
5879 [ # # # # ]: 0 : if (WARN_ON(node == NUMA_NO_NODE)) {
5880 : 0 : pr_warn("workqueue: NUMA node mapping not available for cpu%d, disabling NUMA support\n", cpu);
5881 : : /* happens iff arch is bonkers, let's just proceed */
5882 : 0 : return;
5883 : : }
5884 : 0 : cpumask_set_cpu(cpu, tbl[node]);
5885 : : }
5886 : :
5887 : 0 : wq_numa_possible_cpumask = tbl;
5888 : 0 : wq_numa_enabled = true;
5889 : : }
5890 : :
5891 : : /**
5892 : : * workqueue_init_early - early init for workqueue subsystem
5893 : : *
5894 : : * This is the first half of two-staged workqueue subsystem initialization
5895 : : * and invoked as soon as the bare basics - memory allocation, cpumasks and
5896 : : * idr are up. It sets up all the data structures and system workqueues
5897 : : * and allows early boot code to create workqueues and queue/cancel work
5898 : : * items. Actual work item execution starts only after kthreads can be
5899 : : * created and scheduled right before early initcalls.
5900 : : */
5901 : 11 : int __init workqueue_init_early(void)
5902 : : {
5903 : 11 : int std_nice[NR_STD_WORKER_POOLS] = { 0, HIGHPRI_NICE_LEVEL };
5904 : 11 : int hk_flags = HK_FLAG_DOMAIN | HK_FLAG_WQ;
5905 : 11 : int i, cpu;
5906 : :
5907 : 11 : WARN_ON(__alignof__(struct pool_workqueue) < __alignof__(long long));
5908 : :
5909 : 11 : BUG_ON(!alloc_cpumask_var(&wq_unbound_cpumask, GFP_KERNEL));
5910 : 11 : cpumask_copy(wq_unbound_cpumask, housekeeping_cpumask(hk_flags));
5911 : :
5912 : 11 : pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC);
5913 : :
5914 : : /* initialize CPU pools */
5915 [ + + ]: 22 : for_each_possible_cpu(cpu) {
5916 : 11 : struct worker_pool *pool;
5917 : :
5918 : 11 : i = 0;
5919 [ + + ]: 33 : for_each_cpu_worker_pool(pool, cpu) {
5920 [ - + ]: 22 : BUG_ON(init_worker_pool(pool));
5921 : 22 : pool->cpu = cpu;
5922 : 22 : cpumask_copy(pool->attrs->cpumask, cpumask_of(cpu));
5923 : 22 : pool->attrs->nice = std_nice[i++];
5924 : 22 : pool->node = cpu_to_node(cpu);
5925 : :
5926 : : /* alloc pool ID */
5927 : 22 : mutex_lock(&wq_pool_mutex);
5928 [ - + ]: 44 : BUG_ON(worker_pool_assign_id(pool));
5929 : 22 : mutex_unlock(&wq_pool_mutex);
5930 : : }
5931 : : }
5932 : :
5933 : : /* create default unbound and ordered wq attrs */
5934 [ + + ]: 33 : for (i = 0; i < NR_STD_WORKER_POOLS; i++) {
5935 : 22 : struct workqueue_attrs *attrs;
5936 : :
5937 [ - + ]: 22 : BUG_ON(!(attrs = alloc_workqueue_attrs()));
5938 : 22 : attrs->nice = std_nice[i];
5939 : 22 : unbound_std_wq_attrs[i] = attrs;
5940 : :
5941 : : /*
5942 : : * An ordered wq should have only one pwq as ordering is
5943 : : * guaranteed by max_active which is enforced by pwqs.
5944 : : * Turn off NUMA so that dfl_pwq is used for all nodes.
5945 : : */
5946 [ - + ]: 22 : BUG_ON(!(attrs = alloc_workqueue_attrs()));
5947 : 22 : attrs->nice = std_nice[i];
5948 : 22 : attrs->no_numa = true;
5949 : 22 : ordered_wq_attrs[i] = attrs;
5950 : : }
5951 : :
5952 : 11 : system_wq = alloc_workqueue("events", 0, 0);
5953 : 11 : system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0);
5954 : 11 : system_long_wq = alloc_workqueue("events_long", 0, 0);
5955 : 22 : system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND,
5956 : 11 : WQ_UNBOUND_MAX_ACTIVE);
5957 : 11 : system_freezable_wq = alloc_workqueue("events_freezable",
5958 : : WQ_FREEZABLE, 0);
5959 : 11 : system_power_efficient_wq = alloc_workqueue("events_power_efficient",
5960 : : WQ_POWER_EFFICIENT, 0);
5961 : 11 : system_freezable_power_efficient_wq = alloc_workqueue("events_freezable_power_efficient",
5962 : : WQ_FREEZABLE | WQ_POWER_EFFICIENT,
5963 : : 0);
5964 [ + - + - : 11 : BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq ||
+ - + - +
- + - -
+ ]
5965 : : !system_unbound_wq || !system_freezable_wq ||
5966 : : !system_power_efficient_wq ||
5967 : : !system_freezable_power_efficient_wq);
5968 : :
5969 : 11 : return 0;
5970 : : }
5971 : :
5972 : : /**
5973 : : * workqueue_init - bring workqueue subsystem fully online
5974 : : *
5975 : : * This is the latter half of two-staged workqueue subsystem initialization
5976 : : * and invoked as soon as kthreads can be created and scheduled.
5977 : : * Workqueues have been created and work items queued on them, but there
5978 : : * are no kworkers executing the work items yet. Populate the worker pools
5979 : : * with the initial workers and enable future kworker creations.
5980 : : */
5981 : 11 : int __init workqueue_init(void)
5982 : : {
5983 : 11 : struct workqueue_struct *wq;
5984 : 11 : struct worker_pool *pool;
5985 : 11 : int cpu, bkt;
5986 : :
5987 : : /*
5988 : : * It'd be simpler to initialize NUMA in workqueue_init_early() but
5989 : : * CPU to node mapping may not be available that early on some
5990 : : * archs such as power and arm64. As per-cpu pools created
5991 : : * previously could be missing node hint and unbound pools NUMA
5992 : : * affinity, fix them up.
5993 : : *
5994 : : * Also, while iterating workqueues, create rescuers if requested.
5995 : : */
5996 : 11 : wq_numa_init();
5997 : :
5998 : 11 : mutex_lock(&wq_pool_mutex);
5999 : :
6000 [ + + ]: 33 : for_each_possible_cpu(cpu) {
6001 [ + + ]: 33 : for_each_cpu_worker_pool(pool, cpu) {
6002 : 22 : pool->node = cpu_to_node(cpu);
6003 : : }
6004 : : }
6005 : :
6006 [ + + ]: 110 : list_for_each_entry(wq, &workqueues, list) {
6007 : 99 : wq_update_unbound_numa(wq, smp_processor_id(), true);
6008 [ - + ]: 99 : WARN(init_rescuer(wq),
6009 : : "workqueue: failed to create early rescuer for %s",
6010 : : wq->name);
6011 : : }
6012 : :
6013 : 11 : mutex_unlock(&wq_pool_mutex);
6014 : :
6015 : : /* create the initial workers */
6016 [ + + ]: 33 : for_each_online_cpu(cpu) {
6017 [ + + ]: 33 : for_each_cpu_worker_pool(pool, cpu) {
6018 : 22 : pool->flags &= ~POOL_DISASSOCIATED;
6019 [ - + ]: 22 : BUG_ON(!create_worker(pool));
6020 : : }
6021 : : }
6022 : :
6023 [ + + + + : 726 : hash_for_each(unbound_pool_hash, bkt, pool, hash_node)
+ + ]
6024 [ - + - + ]: 11 : BUG_ON(!create_worker(pool));
6025 : :
6026 : 11 : wq_online = true;
6027 : 11 : wq_watchdog_init();
6028 : :
6029 : 11 : return 0;
6030 : : }
|