Branch data Line data Source code
1 : : /*
2 : : * kernel/workqueue.c - generic async execution with shared worker pool
3 : : *
4 : : * Copyright (C) 2002 Ingo Molnar
5 : : *
6 : : * Derived from the taskqueue/keventd code by:
7 : : * David Woodhouse <dwmw2@infradead.org>
8 : : * Andrew Morton
9 : : * Kai Petzke <wpp@marie.physik.tu-berlin.de>
10 : : * Theodore Ts'o <tytso@mit.edu>
11 : : *
12 : : * Made to use alloc_percpu by Christoph Lameter.
13 : : *
14 : : * Copyright (C) 2010 SUSE Linux Products GmbH
15 : : * Copyright (C) 2010 Tejun Heo <tj@kernel.org>
16 : : *
17 : : * This is the generic async execution mechanism. Work items as are
18 : : * executed in process context. The worker pool is shared and
19 : : * automatically managed. There are two worker pools for each CPU (one for
20 : : * normal work items and the other for high priority ones) and some extra
21 : : * pools for workqueues which are not bound to any specific CPU - the
22 : : * number of these backing pools is dynamic.
23 : : *
24 : : * Please read Documentation/workqueue.txt for details.
25 : : */
26 : :
27 : : #include <linux/export.h>
28 : : #include <linux/kernel.h>
29 : : #include <linux/sched.h>
30 : : #include <linux/init.h>
31 : : #include <linux/signal.h>
32 : : #include <linux/completion.h>
33 : : #include <linux/workqueue.h>
34 : : #include <linux/slab.h>
35 : : #include <linux/cpu.h>
36 : : #include <linux/notifier.h>
37 : : #include <linux/kthread.h>
38 : : #include <linux/hardirq.h>
39 : : #include <linux/mempolicy.h>
40 : : #include <linux/freezer.h>
41 : : #include <linux/kallsyms.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 : :
52 : : #include "workqueue_internal.h"
53 : :
54 : : enum {
55 : : /*
56 : : * worker_pool flags
57 : : *
58 : : * A bound pool is either associated or disassociated with its CPU.
59 : : * While associated (!DISASSOCIATED), all workers are bound to the
60 : : * CPU and none has %WORKER_UNBOUND set and concurrency management
61 : : * is in effect.
62 : : *
63 : : * While DISASSOCIATED, the cpu may be offline and all workers have
64 : : * %WORKER_UNBOUND set and concurrency management disabled, and may
65 : : * be executing on any CPU. The pool behaves as an unbound one.
66 : : *
67 : : * Note that DISASSOCIATED should be flipped only while holding
68 : : * manager_mutex to avoid changing binding state while
69 : : * create_worker() is in progress.
70 : : */
71 : : POOL_MANAGE_WORKERS = 1 << 0, /* need to manage workers */
72 : : POOL_DISASSOCIATED = 1 << 2, /* cpu can't serve workers */
73 : : POOL_FREEZING = 1 << 3, /* freeze in progress */
74 : :
75 : : /* worker flags */
76 : : WORKER_STARTED = 1 << 0, /* started */
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 -20.
104 : : */
105 : : RESCUER_NICE_LEVEL = -20,
106 : : HIGHPRI_NICE_LEVEL = -20,
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 : : * MG: pool->manager_mutex and pool->lock protected. Writes require both
128 : : * locks. Reads can happen under either lock.
129 : : *
130 : : * PL: wq_pool_mutex protected.
131 : : *
132 : : * PR: wq_pool_mutex protected for writes. Sched-RCU protected for reads.
133 : : *
134 : : * WQ: wq->mutex protected.
135 : : *
136 : : * WR: wq->mutex protected for writes. Sched-RCU protected for reads.
137 : : *
138 : : * MD: wq_mayday_lock protected.
139 : : */
140 : :
141 : : /* struct worker is defined in workqueue_internal.h */
142 : :
143 : : struct worker_pool {
144 : : spinlock_t lock; /* the pool lock */
145 : : int cpu; /* I: the associated cpu */
146 : : int node; /* I: the associated node ID */
147 : : int id; /* I: pool ID */
148 : : unsigned int flags; /* X: flags */
149 : :
150 : : struct list_head worklist; /* L: list of pending works */
151 : : int nr_workers; /* L: total number of workers */
152 : :
153 : : /* nr_idle includes the ones off idle_list for rebinding */
154 : : int nr_idle; /* L: currently idle ones */
155 : :
156 : : struct list_head idle_list; /* X: list of idle workers */
157 : : struct timer_list idle_timer; /* L: worker idle timeout */
158 : : struct timer_list mayday_timer; /* L: SOS timer for workers */
159 : :
160 : : /* a workers is either on busy_hash or idle_list, or the manager */
161 : : DECLARE_HASHTABLE(busy_hash, BUSY_WORKER_HASH_ORDER);
162 : : /* L: hash of busy workers */
163 : :
164 : : /* see manage_workers() for details on the two manager mutexes */
165 : : struct mutex manager_arb; /* manager arbitration */
166 : : struct mutex manager_mutex; /* manager exclusion */
167 : : struct idr worker_idr; /* MG: worker IDs and iteration */
168 : :
169 : : struct workqueue_attrs *attrs; /* I: worker attributes */
170 : : struct hlist_node hash_node; /* PL: unbound_pool_hash node */
171 : : int refcnt; /* PL: refcnt for unbound pools */
172 : :
173 : : /*
174 : : * The current concurrency level. As it's likely to be accessed
175 : : * from other CPUs during try_to_wake_up(), put it in a separate
176 : : * cacheline.
177 : : */
178 : : atomic_t nr_running ____cacheline_aligned_in_smp;
179 : :
180 : : /*
181 : : * Destruction of pool is sched-RCU protected to allow dereferences
182 : : * from get_work_pool().
183 : : */
184 : : struct rcu_head rcu;
185 : : } ____cacheline_aligned_in_smp;
186 : :
187 : : /*
188 : : * The per-pool workqueue. While queued, the lower WORK_STRUCT_FLAG_BITS
189 : : * of work_struct->data are used for flags and the remaining high bits
190 : : * point to the pwq; thus, pwqs need to be aligned at two's power of the
191 : : * number of flag bits.
192 : : */
193 : : struct pool_workqueue {
194 : : struct worker_pool *pool; /* I: the associated pool */
195 : : struct workqueue_struct *wq; /* I: the owning workqueue */
196 : : int work_color; /* L: current color */
197 : : int flush_color; /* L: flushing color */
198 : : int refcnt; /* L: reference count */
199 : : int nr_in_flight[WORK_NR_COLORS];
200 : : /* L: nr of in_flight works */
201 : : int nr_active; /* L: nr of active works */
202 : : int max_active; /* L: max active works */
203 : : struct list_head delayed_works; /* L: delayed works */
204 : : struct list_head pwqs_node; /* WR: node on wq->pwqs */
205 : : struct list_head mayday_node; /* MD: node on wq->maydays */
206 : :
207 : : /*
208 : : * Release of unbound pwq is punted to system_wq. See put_pwq()
209 : : * and pwq_unbound_release_workfn() for details. pool_workqueue
210 : : * itself is also sched-RCU protected so that the first pwq can be
211 : : * determined without grabbing wq->mutex.
212 : : */
213 : : struct work_struct unbound_release_work;
214 : : struct rcu_head rcu;
215 : : } __aligned(1 << WORK_STRUCT_FLAG_BITS);
216 : :
217 : : /*
218 : : * Structure used to wait for workqueue flush.
219 : : */
220 : : struct wq_flusher {
221 : : struct list_head list; /* WQ: list of flushers */
222 : : int flush_color; /* WQ: flush color waiting for */
223 : : struct completion done; /* flush completion */
224 : : };
225 : :
226 : : struct wq_device;
227 : :
228 : : /*
229 : : * The externally visible workqueue. It relays the issued work items to
230 : : * the appropriate worker_pool through its pool_workqueues.
231 : : */
232 : : struct workqueue_struct {
233 : : struct list_head pwqs; /* WR: all pwqs of this wq */
234 : : struct list_head list; /* PL: list of all workqueues */
235 : :
236 : : struct mutex mutex; /* protects this wq */
237 : : int work_color; /* WQ: current work color */
238 : : int flush_color; /* WQ: current flush color */
239 : : atomic_t nr_pwqs_to_flush; /* flush in progress */
240 : : struct wq_flusher *first_flusher; /* WQ: first flusher */
241 : : struct list_head flusher_queue; /* WQ: flush waiters */
242 : : struct list_head flusher_overflow; /* WQ: flush overflow list */
243 : :
244 : : struct list_head maydays; /* MD: pwqs requesting rescue */
245 : : struct worker *rescuer; /* I: rescue worker */
246 : :
247 : : int nr_drainers; /* WQ: drain in progress */
248 : : int saved_max_active; /* WQ: saved pwq max_active */
249 : :
250 : : struct workqueue_attrs *unbound_attrs; /* WQ: only for unbound wqs */
251 : : struct pool_workqueue *dfl_pwq; /* WQ: only for unbound wqs */
252 : :
253 : : #ifdef CONFIG_SYSFS
254 : : struct wq_device *wq_dev; /* I: for sysfs interface */
255 : : #endif
256 : : #ifdef CONFIG_LOCKDEP
257 : : struct lockdep_map lockdep_map;
258 : : #endif
259 : : char name[WQ_NAME_LEN]; /* I: workqueue name */
260 : :
261 : : /* hot fields used during command issue, aligned to cacheline */
262 : : unsigned int flags ____cacheline_aligned; /* WQ: WQ_* flags */
263 : : struct pool_workqueue __percpu *cpu_pwqs; /* I: per-cpu pwqs */
264 : : struct pool_workqueue __rcu *numa_pwq_tbl[]; /* FR: unbound pwqs indexed by node */
265 : : };
266 : :
267 : : static struct kmem_cache *pwq_cache;
268 : :
269 : : static int wq_numa_tbl_len; /* highest possible NUMA node id + 1 */
270 : : static cpumask_var_t *wq_numa_possible_cpumask;
271 : : /* possible CPUs of each node */
272 : :
273 : : static bool wq_disable_numa;
274 : : module_param_named(disable_numa, wq_disable_numa, bool, 0444);
275 : :
276 : : /* see the comment above the definition of WQ_POWER_EFFICIENT */
277 : : #ifdef CONFIG_WQ_POWER_EFFICIENT_DEFAULT
278 : : static bool wq_power_efficient = true;
279 : : #else
280 : : static bool wq_power_efficient;
281 : : #endif
282 : :
283 : : module_param_named(power_efficient, wq_power_efficient, bool, 0444);
284 : :
285 : : static bool wq_numa_enabled; /* unbound NUMA affinity enabled */
286 : :
287 : : /* buf for wq_update_unbound_numa_attrs(), protected by CPU hotplug exclusion */
288 : : static struct workqueue_attrs *wq_update_unbound_numa_attrs_buf;
289 : :
290 : : static DEFINE_MUTEX(wq_pool_mutex); /* protects pools and workqueues list */
291 : : static DEFINE_SPINLOCK(wq_mayday_lock); /* protects wq->maydays list */
292 : :
293 : : static LIST_HEAD(workqueues); /* PL: list of all workqueues */
294 : : static bool workqueue_freezing; /* PL: have wqs started freezing? */
295 : :
296 : : /* the per-cpu worker pools */
297 : : static DEFINE_PER_CPU_SHARED_ALIGNED(struct worker_pool [NR_STD_WORKER_POOLS],
298 : : cpu_worker_pools);
299 : :
300 : : static DEFINE_IDR(worker_pool_idr); /* PR: idr of all pools */
301 : :
302 : : /* PL: hash of all unbound pools keyed by pool->attrs */
303 : : static DEFINE_HASHTABLE(unbound_pool_hash, UNBOUND_POOL_HASH_ORDER);
304 : :
305 : : /* I: attributes used when instantiating standard unbound pools on demand */
306 : : static struct workqueue_attrs *unbound_std_wq_attrs[NR_STD_WORKER_POOLS];
307 : :
308 : : /* I: attributes used when instantiating ordered pools on demand */
309 : : static struct workqueue_attrs *ordered_wq_attrs[NR_STD_WORKER_POOLS];
310 : :
311 : : struct workqueue_struct *system_wq __read_mostly;
312 : : EXPORT_SYMBOL(system_wq);
313 : : struct workqueue_struct *system_highpri_wq __read_mostly;
314 : : EXPORT_SYMBOL_GPL(system_highpri_wq);
315 : : struct workqueue_struct *system_long_wq __read_mostly;
316 : : EXPORT_SYMBOL_GPL(system_long_wq);
317 : : struct workqueue_struct *system_unbound_wq __read_mostly;
318 : : EXPORT_SYMBOL_GPL(system_unbound_wq);
319 : : struct workqueue_struct *system_freezable_wq __read_mostly;
320 : : EXPORT_SYMBOL_GPL(system_freezable_wq);
321 : : struct workqueue_struct *system_power_efficient_wq __read_mostly;
322 : : EXPORT_SYMBOL_GPL(system_power_efficient_wq);
323 : : struct workqueue_struct *system_freezable_power_efficient_wq __read_mostly;
324 : : EXPORT_SYMBOL_GPL(system_freezable_power_efficient_wq);
325 : :
326 : : static int worker_thread(void *__worker);
327 : : static void copy_workqueue_attrs(struct workqueue_attrs *to,
328 : : const struct workqueue_attrs *from);
329 : :
330 : : #define CREATE_TRACE_POINTS
331 : : #include <trace/events/workqueue.h>
332 : :
333 : : #define assert_rcu_or_pool_mutex() \
334 : : rcu_lockdep_assert(rcu_read_lock_sched_held() || \
335 : : lockdep_is_held(&wq_pool_mutex), \
336 : : "sched RCU or wq_pool_mutex should be held")
337 : :
338 : : #define assert_rcu_or_wq_mutex(wq) \
339 : : rcu_lockdep_assert(rcu_read_lock_sched_held() || \
340 : : lockdep_is_held(&wq->mutex), \
341 : : "sched RCU or wq->mutex should be held")
342 : :
343 : : #ifdef CONFIG_LOCKDEP
344 : : #define assert_manager_or_pool_lock(pool) \
345 : : WARN_ONCE(debug_locks && \
346 : : !lockdep_is_held(&(pool)->manager_mutex) && \
347 : : !lockdep_is_held(&(pool)->lock), \
348 : : "pool->manager_mutex or ->lock should be held")
349 : : #else
350 : : #define assert_manager_or_pool_lock(pool) do { } while (0)
351 : : #endif
352 : :
353 : : #define for_each_cpu_worker_pool(pool, cpu) \
354 : : for ((pool) = &per_cpu(cpu_worker_pools, cpu)[0]; \
355 : : (pool) < &per_cpu(cpu_worker_pools, cpu)[NR_STD_WORKER_POOLS]; \
356 : : (pool)++)
357 : :
358 : : /**
359 : : * for_each_pool - iterate through all worker_pools in the system
360 : : * @pool: iteration cursor
361 : : * @pi: integer used for iteration
362 : : *
363 : : * This must be called either with wq_pool_mutex held or sched RCU read
364 : : * locked. If the pool needs to be used beyond the locking in effect, the
365 : : * caller is responsible for guaranteeing that the pool stays online.
366 : : *
367 : : * The if/else clause exists only for the lockdep assertion and can be
368 : : * ignored.
369 : : */
370 : : #define for_each_pool(pool, pi) \
371 : : idr_for_each_entry(&worker_pool_idr, pool, pi) \
372 : : if (({ assert_rcu_or_pool_mutex(); false; })) { } \
373 : : else
374 : :
375 : : /**
376 : : * for_each_pool_worker - iterate through all workers of a worker_pool
377 : : * @worker: iteration cursor
378 : : * @wi: integer used for iteration
379 : : * @pool: worker_pool to iterate workers of
380 : : *
381 : : * This must be called with either @pool->manager_mutex or ->lock held.
382 : : *
383 : : * The if/else clause exists only for the lockdep assertion and can be
384 : : * ignored.
385 : : */
386 : : #define for_each_pool_worker(worker, wi, pool) \
387 : : idr_for_each_entry(&(pool)->worker_idr, (worker), (wi)) \
388 : : if (({ assert_manager_or_pool_lock((pool)); false; })) { } \
389 : : else
390 : :
391 : : /**
392 : : * for_each_pwq - iterate through all pool_workqueues of the specified workqueue
393 : : * @pwq: iteration cursor
394 : : * @wq: the target workqueue
395 : : *
396 : : * This must be called either with wq->mutex held or sched RCU read locked.
397 : : * If the pwq needs to be used beyond the locking in effect, the caller is
398 : : * responsible for guaranteeing that the pwq stays online.
399 : : *
400 : : * The if/else clause exists only for the lockdep assertion and can be
401 : : * ignored.
402 : : */
403 : : #define for_each_pwq(pwq, wq) \
404 : : list_for_each_entry_rcu((pwq), &(wq)->pwqs, pwqs_node) \
405 : : if (({ assert_rcu_or_wq_mutex(wq); false; })) { } \
406 : : else
407 : :
408 : : #ifdef CONFIG_DEBUG_OBJECTS_WORK
409 : :
410 : : static struct debug_obj_descr work_debug_descr;
411 : :
412 : : static void *work_debug_hint(void *addr)
413 : : {
414 : : return ((struct work_struct *) addr)->func;
415 : : }
416 : :
417 : : /*
418 : : * fixup_init is called when:
419 : : * - an active object is initialized
420 : : */
421 : : static int work_fixup_init(void *addr, enum debug_obj_state state)
422 : : {
423 : : struct work_struct *work = addr;
424 : :
425 : : switch (state) {
426 : : case ODEBUG_STATE_ACTIVE:
427 : : cancel_work_sync(work);
428 : : debug_object_init(work, &work_debug_descr);
429 : : return 1;
430 : : default:
431 : : return 0;
432 : : }
433 : : }
434 : :
435 : : /*
436 : : * fixup_activate is called when:
437 : : * - an active object is activated
438 : : * - an unknown object is activated (might be a statically initialized object)
439 : : */
440 : : static int work_fixup_activate(void *addr, enum debug_obj_state state)
441 : : {
442 : : struct work_struct *work = addr;
443 : :
444 : : switch (state) {
445 : :
446 : : case ODEBUG_STATE_NOTAVAILABLE:
447 : : /*
448 : : * This is not really a fixup. The work struct was
449 : : * statically initialized. We just make sure that it
450 : : * is tracked in the object tracker.
451 : : */
452 : : if (test_bit(WORK_STRUCT_STATIC_BIT, work_data_bits(work))) {
453 : : debug_object_init(work, &work_debug_descr);
454 : : debug_object_activate(work, &work_debug_descr);
455 : : return 0;
456 : : }
457 : : WARN_ON_ONCE(1);
458 : : return 0;
459 : :
460 : : case ODEBUG_STATE_ACTIVE:
461 : : WARN_ON(1);
462 : :
463 : : default:
464 : : return 0;
465 : : }
466 : : }
467 : :
468 : : /*
469 : : * fixup_free is called when:
470 : : * - an active object is freed
471 : : */
472 : : static int work_fixup_free(void *addr, enum debug_obj_state state)
473 : : {
474 : : struct work_struct *work = addr;
475 : :
476 : : switch (state) {
477 : : case ODEBUG_STATE_ACTIVE:
478 : : cancel_work_sync(work);
479 : : debug_object_free(work, &work_debug_descr);
480 : : return 1;
481 : : default:
482 : : return 0;
483 : : }
484 : : }
485 : :
486 : : static struct debug_obj_descr work_debug_descr = {
487 : : .name = "work_struct",
488 : : .debug_hint = work_debug_hint,
489 : : .fixup_init = work_fixup_init,
490 : : .fixup_activate = work_fixup_activate,
491 : : .fixup_free = work_fixup_free,
492 : : };
493 : :
494 : : static inline void debug_work_activate(struct work_struct *work)
495 : : {
496 : : debug_object_activate(work, &work_debug_descr);
497 : : }
498 : :
499 : : static inline void debug_work_deactivate(struct work_struct *work)
500 : : {
501 : : debug_object_deactivate(work, &work_debug_descr);
502 : : }
503 : :
504 : : void __init_work(struct work_struct *work, int onstack)
505 : : {
506 : : if (onstack)
507 : : debug_object_init_on_stack(work, &work_debug_descr);
508 : : else
509 : : debug_object_init(work, &work_debug_descr);
510 : : }
511 : : EXPORT_SYMBOL_GPL(__init_work);
512 : :
513 : : void destroy_work_on_stack(struct work_struct *work)
514 : : {
515 : : debug_object_free(work, &work_debug_descr);
516 : : }
517 : : EXPORT_SYMBOL_GPL(destroy_work_on_stack);
518 : :
519 : : #else
520 : : static inline void debug_work_activate(struct work_struct *work) { }
521 : : 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 : 0 : static int worker_pool_assign_id(struct worker_pool *pool)
532 : : {
533 : : int ret;
534 : :
535 : : lockdep_assert_held(&wq_pool_mutex);
536 : :
537 : 0 : ret = idr_alloc(&worker_pool_idr, pool, 0, WORK_OFFQ_POOL_NONE,
538 : : GFP_KERNEL);
539 [ # # ]: 0 : if (ret >= 0) {
540 : 0 : pool->id = ret;
541 : 0 : 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 either with pwq_lock held or sched RCU read locked.
552 : : * If the pwq needs to be used beyond the locking in effect, the caller is
553 : : * responsible for guaranteeing that the pwq stays online.
554 : : *
555 : : * Return: The unbound pool_workqueue for @node.
556 : : */
557 : : static struct pool_workqueue *unbound_pwq_by_node(struct workqueue_struct *wq,
558 : : int node)
559 : : {
560 : : assert_rcu_or_wq_mutex(wq);
561 : 160684 : return rcu_dereference_raw(wq->numa_pwq_tbl[node]);
562 : : }
563 : :
564 : : static unsigned int work_color_to_flags(int color)
565 : : {
566 : 1698564 : return color << WORK_STRUCT_COLOR_SHIFT;
567 : : }
568 : :
569 : : static int get_work_color(struct work_struct *work)
570 : : {
571 : 1696631 : return (*work_data_bits(work) >> WORK_STRUCT_COLOR_SHIFT) &
572 : : ((1 << WORK_STRUCT_COLOR_BITS) - 1);
573 : : }
574 : :
575 : : static int work_next_color(int color)
576 : : {
577 : 11944 : return (color + 1) % WORK_NR_COLORS;
578 : : }
579 : :
580 : : /*
581 : : * While queued, %WORK_STRUCT_PWQ is set and non flag bits of a work's data
582 : : * contain the pointer to the queued pwq. Once execution starts, the flag
583 : : * is cleared and the high bits contain OFFQ flags and pool ID.
584 : : *
585 : : * set_work_pwq(), set_work_pool_and_clear_pending(), mark_work_canceling()
586 : : * and clear_work_data() can be used to set the pwq, pool or clear
587 : : * work->data. These functions should only be called while the work is
588 : : * owned - ie. while the PENDING bit is set.
589 : : *
590 : : * get_work_pool() and get_work_pwq() can be used to obtain the pool or pwq
591 : : * corresponding to a work. Pool is available once the work has been
592 : : * queued anywhere after initialization until it is sync canceled. pwq is
593 : : * available only while the work item is queued.
594 : : *
595 : : * %WORK_OFFQ_CANCELING is used to mark a work item which is being
596 : : * canceled. While being canceled, a work item may have its PENDING set
597 : : * but stay off timer and worklist for arbitrarily long and nobody should
598 : : * try to steal the PENDING bit.
599 : : */
600 : : static inline void set_work_data(struct work_struct *work, unsigned long data,
601 : : unsigned long flags)
602 : : {
603 [ - + ][ # # ]: 3395012 : WARN_ON_ONCE(!work_pending(work));
[ # # ][ - + ]
[ # # ][ # # ]
[ - + ][ # # ]
[ # # ][ - + ]
[ # # ][ # # ]
[ - + ][ # # ]
[ # # ]
604 : 3393824 : atomic_long_set(&work->data, data | flags | work_static(work));
605 : : }
606 : :
607 : 0 : static void set_work_pwq(struct work_struct *work, struct pool_workqueue *pwq,
608 : : unsigned long extra_flags)
609 : : {
610 : 1697799 : set_work_data(work, (unsigned long)pwq,
611 : : WORK_STRUCT_PENDING | WORK_STRUCT_PWQ | extra_flags);
612 : 1697799 : }
613 : :
614 : 0 : static void set_work_pool_and_keep_pending(struct work_struct *work,
615 : : int pool_id)
616 : : {
617 : 7334 : set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT,
618 : : WORK_STRUCT_PENDING);
619 : 7334 : }
620 : :
621 : 0 : static void set_work_pool_and_clear_pending(struct work_struct *work,
622 : : int pool_id)
623 : : {
624 : : /*
625 : : * The following wmb is paired with the implied mb in
626 : : * test_and_set_bit(PENDING) and ensures all updates to @work made
627 : : * here are visible to and precede any updates by the next PENDING
628 : : * owner.
629 : : */
630 : 1691329 : smp_wmb();
631 : 1687503 : set_work_data(work, (unsigned long)pool_id << WORK_OFFQ_POOL_SHIFT, 0);
632 : 1687503 : }
633 : :
634 : 0 : static void clear_work_data(struct work_struct *work)
635 : : {
636 : 1188 : smp_wmb(); /* see set_work_pool_and_clear_pending() */
637 : : set_work_data(work, WORK_STRUCT_NO_POOL, 0);
638 : 1188 : }
639 : :
640 : : static struct pool_workqueue *get_work_pwq(struct work_struct *work)
641 : : {
642 : 1708527 : unsigned long data = atomic_long_read(&work->data);
643 : :
644 [ + + ][ # # ]: 1708527 : if (data & WORK_STRUCT_PWQ)
[ + + ]
[ # # + + ]
[ + - ][ # # ]
645 : 1701869 : return (void *)(data & WORK_STRUCT_WQ_DATA_MASK);
646 : : else
647 : : return NULL;
648 : : }
649 : :
650 : : /**
651 : : * get_work_pool - return the worker_pool a given work was associated with
652 : : * @work: the work item of interest
653 : : *
654 : : * Pools are created and destroyed under wq_pool_mutex, and allows read
655 : : * access under sched-RCU read lock. As such, this function should be
656 : : * called under wq_pool_mutex or with preemption disabled.
657 : : *
658 : : * All fields of the returned pool are accessible as long as the above
659 : : * mentioned locking is in effect. If the returned pool needs to be used
660 : : * beyond the critical section, the caller is responsible for ensuring the
661 : : * returned pool is and stays online.
662 : : *
663 : : * Return: The worker_pool @work was last associated with. %NULL if none.
664 : : */
665 : 0 : static struct worker_pool *get_work_pool(struct work_struct *work)
666 : : {
667 : 1708088 : unsigned long data = atomic_long_read(&work->data);
668 : : int pool_id;
669 : :
670 : : assert_rcu_or_pool_mutex();
671 : :
672 [ + + ]: 1708088 : if (data & WORK_STRUCT_PWQ)
673 : 7417 : return ((struct pool_workqueue *)
674 : 7417 : (data & WORK_STRUCT_WQ_DATA_MASK))->pool;
675 : :
676 : 1700671 : pool_id = data >> WORK_OFFQ_POOL_SHIFT;
677 [ + + ]: 1700671 : if (pool_id == WORK_OFFQ_POOL_NONE)
678 : : return NULL;
679 : :
680 : 1689454 : return idr_find(&worker_pool_idr, pool_id);
681 : : }
682 : :
683 : : /**
684 : : * get_work_pool_id - return the worker pool ID a given work is associated with
685 : : * @work: the work item of interest
686 : : *
687 : : * Return: The worker_pool ID @work was last associated with.
688 : : * %WORK_OFFQ_POOL_NONE if none.
689 : : */
690 : : static int get_work_pool_id(struct work_struct *work)
691 : : {
692 : 1188 : unsigned long data = atomic_long_read(&work->data);
693 : :
694 [ # # ][ - + ]: 1188 : if (data & WORK_STRUCT_PWQ)
695 : 0 : return ((struct pool_workqueue *)
696 : 0 : (data & WORK_STRUCT_WQ_DATA_MASK))->pool->id;
697 : :
698 : 1188 : return data >> WORK_OFFQ_POOL_SHIFT;
699 : : }
700 : :
701 : 0 : static void mark_work_canceling(struct work_struct *work)
702 : : {
703 : 1188 : unsigned long pool_id = get_work_pool_id(work);
704 : :
705 : 1188 : pool_id <<= WORK_OFFQ_POOL_SHIFT;
706 : : set_work_data(work, pool_id | WORK_OFFQ_CANCELING, WORK_STRUCT_PENDING);
707 : 1188 : }
708 : :
709 : : static bool work_is_canceling(struct work_struct *work)
710 : : {
711 : 76 : unsigned long data = atomic_long_read(&work->data);
712 : :
713 : 76 : return !(data & WORK_STRUCT_PWQ) && (data & WORK_OFFQ_CANCELING);
714 : : }
715 : :
716 : : /*
717 : : * Policy functions. These define the policies on how the global worker
718 : : * pools are managed. Unless noted otherwise, these functions assume that
719 : : * they're being called with pool->lock held.
720 : : */
721 : :
722 : : static bool __need_more_worker(struct worker_pool *pool)
723 : : {
724 : 2862883 : return !atomic_read(&pool->nr_running);
725 : : }
726 : :
727 : : /*
728 : : * Need to wake up a worker? Called from anything but currently
729 : : * running workers.
730 : : *
731 : : * Note that, because unbound workers never contribute to nr_running, this
732 : : * function will always return %true for unbound pools as long as the
733 : : * worklist isn't empty.
734 : : */
735 : : static bool need_more_worker(struct worker_pool *pool)
736 : : {
737 [ + + ][ + + ]: 2490892 : return !list_empty(&pool->worklist) && __need_more_worker(pool);
[ + + ][ - + ]
[ + + ][ - + ]
[ + + ][ + + ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ - + ][ # # ]
[ + + ][ + - ]
738 : : }
739 : :
740 : : /* Can I start working? Called from busy but !running workers. */
741 : : static bool may_start_working(struct worker_pool *pool)
742 : : {
743 : : return pool->nr_idle;
744 : : }
745 : :
746 : : /* Do I need to keep working? Called from currently running workers. */
747 : : static bool keep_working(struct worker_pool *pool)
748 : : {
749 [ # # ][ # # ]: 1688692 : return !list_empty(&pool->worklist) &&
[ + + ][ + + ]
750 : 528520 : atomic_read(&pool->nr_running) <= 1;
751 : : }
752 : :
753 : : /* Do we need a new worker? Called from manager. */
754 : 529 : static bool need_to_create_worker(struct worker_pool *pool)
755 : : {
756 [ + + ][ - + ]: 1167574 : return need_more_worker(pool) && !may_start_working(pool);
[ + + ][ + - ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ - + ][ # # ]
[ - + ][ # # ]
757 : : }
758 : :
759 : : /* Do I need to be the manager? */
760 : 0 : static bool need_to_manage_workers(struct worker_pool *pool)
761 : : {
762 [ # # ][ + + ]: 1166088 : return need_to_create_worker(pool) ||
763 : 1166088 : (pool->flags & POOL_MANAGE_WORKERS);
764 : : }
765 : :
766 : : /* Do we have too many workers and should some go away? */
767 : : static bool too_many_workers(struct worker_pool *pool)
768 : : {
769 : : bool managing = mutex_is_locked(&pool->manager_arb);
770 : 1168127 : int nr_idle = pool->nr_idle + managing; /* manager is considered idle */
771 : 1168127 : int nr_busy = pool->nr_workers - nr_idle;
772 : :
773 : : /*
774 : : * nr_idle and idle_list may disagree if idle rebinding is in
775 : : * progress. Never return %true if idle_list is empty.
776 : : */
777 [ + + + - ]: 1169117 : if (list_empty(&pool->idle_list))
[ + + ]
778 : : return false;
779 : :
780 [ + + ][ - + ]: 1167583 : return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy;
[ + + ][ - + ]
[ + + ][ + + ]
781 : : }
782 : :
783 : : /*
784 : : * Wake up functions.
785 : : */
786 : :
787 : : /* Return the first worker. Safe with preemption disabled */
788 : : static struct worker *first_worker(struct worker_pool *pool)
789 : : {
790 [ # # ][ + - ]: 1702588 : if (unlikely(list_empty(&pool->idle_list)))
[ + + ]
791 : : return NULL;
792 : :
793 : : return list_first_entry(&pool->idle_list, struct worker, entry);
794 : : }
795 : :
796 : : /**
797 : : * wake_up_worker - wake up an idle worker
798 : : * @pool: worker pool to wake worker from
799 : : *
800 : : * Wake up the first idle worker of @pool.
801 : : *
802 : : * CONTEXT:
803 : : * spin_lock_irq(pool->lock).
804 : : */
805 : 0 : static void wake_up_worker(struct worker_pool *pool)
806 : : {
807 : : struct worker *worker = first_worker(pool);
808 : :
809 [ + + ]: 1698835 : if (likely(worker))
810 : 1697929 : wake_up_process(worker->task);
811 : 0 : }
812 : :
813 : : /**
814 : : * wq_worker_waking_up - a worker is waking up
815 : : * @task: task waking up
816 : : * @cpu: CPU @task is waking up to
817 : : *
818 : : * This function is called during try_to_wake_up() when a worker is
819 : : * being awoken.
820 : : *
821 : : * CONTEXT:
822 : : * spin_lock_irq(rq->lock)
823 : : */
824 : 0 : void wq_worker_waking_up(struct task_struct *task, int cpu)
825 : : {
826 : 1244846 : struct worker *worker = kthread_data(task);
827 : :
828 [ + + ]: 1238875 : if (!(worker->flags & WORKER_NOT_RUNNING)) {
829 [ - + ][ # # ]: 73468 : WARN_ON_ONCE(worker->pool->cpu != cpu);
[ # # ]
830 : 73468 : atomic_inc(&worker->pool->nr_running);
831 : : }
832 : 1238875 : }
833 : :
834 : : /**
835 : : * wq_worker_sleeping - a worker is going to sleep
836 : : * @task: task going to sleep
837 : : * @cpu: CPU in question, must be the current CPU number
838 : : *
839 : : * This function is called during schedule() when a busy worker is
840 : : * going to sleep. Worker on the same cpu can be woken up by
841 : : * returning pointer to its task.
842 : : *
843 : : * CONTEXT:
844 : : * spin_lock_irq(rq->lock)
845 : : *
846 : : * Return:
847 : : * Worker task on @cpu to wake up, %NULL if none.
848 : : */
849 : 0 : struct task_struct *wq_worker_sleeping(struct task_struct *task, int cpu)
850 : : {
851 : 1244926 : struct worker *worker = kthread_data(task), *to_wakeup = NULL;
852 : : struct worker_pool *pool;
853 : :
854 : : /*
855 : : * Rescuers, which may not have all the fields set up like normal
856 : : * workers, also reach here, let's not access anything before
857 : : * checking NOT_RUNNING.
858 : : */
859 [ + + ]: 1244923 : if (worker->flags & WORKER_NOT_RUNNING)
860 : : return NULL;
861 : :
862 : 73468 : pool = worker->pool;
863 : :
864 : : /* this can only happen on the local cpu */
865 [ - + ][ # # ]: 73468 : if (WARN_ON_ONCE(cpu != raw_smp_processor_id()))
[ # # ][ + - ]
866 : : return NULL;
867 : :
868 : : /*
869 : : * The counterpart of the following dec_and_test, implied mb,
870 : : * worklist not empty test sequence is in insert_work().
871 : : * Please read comment there.
872 : : *
873 : : * NOT_RUNNING is clear. This means that we're bound to and
874 : : * running on the local cpu w/ rq lock held and preemption
875 : : * disabled, which in turn means that none else could be
876 : : * manipulating idle_list, so dereferencing idle_list without pool
877 : : * lock is safe.
878 : : */
879 [ + - ][ + + ]: 1318393 : if (atomic_dec_and_test(&pool->nr_running) &&
880 : 73467 : !list_empty(&pool->worklist))
881 : : to_wakeup = first_worker(pool);
882 [ + + ]: 73467 : return to_wakeup ? to_wakeup->task : NULL;
883 : : }
884 : :
885 : : /**
886 : : * worker_set_flags - set worker flags and adjust nr_running accordingly
887 : : * @worker: self
888 : : * @flags: flags to set
889 : : * @wakeup: wakeup an idle worker if necessary
890 : : *
891 : : * Set @flags in @worker->flags and adjust nr_running accordingly. If
892 : : * nr_running becomes zero and @wakeup is %true, an idle worker is
893 : : * woken up.
894 : : *
895 : : * CONTEXT:
896 : : * spin_lock_irq(pool->lock)
897 : : */
898 : : static inline void worker_set_flags(struct worker *worker, unsigned int flags,
899 : : bool wakeup)
900 : : {
901 : 1164237 : struct worker_pool *pool = worker->pool;
902 : :
903 [ - + ][ # # ]: 1164237 : WARN_ON_ONCE(worker->task != current);
[ - ][ # # ]
[ # # ][ # # ]
904 : :
905 : : /*
906 : : * If transitioning into NOT_RUNNING, adjust nr_running and
907 : : * wake up an idle worker as necessary if requested by
908 : : * @wakeup.
909 : : */
910 [ + + ][ # # ]: 1164226 : if ((flags & WORKER_NOT_RUNNING) &&
911 : 1164226 : !(worker->flags & WORKER_NOT_RUNNING)) {
912 : : if (wakeup) {
913 [ # # ][ # # ]: 0 : if (atomic_dec_and_test(&pool->nr_running) &&
914 : 0 : !list_empty(&pool->worklist))
915 : 0 : wake_up_worker(pool);
916 : : } else
917 : 1005694 : atomic_dec(&pool->nr_running);
918 : : }
919 : :
920 : 1164204 : worker->flags |= flags;
921 : : }
922 : :
923 : : /**
924 : : * worker_clr_flags - clear worker flags and adjust nr_running accordingly
925 : : * @worker: self
926 : : * @flags: flags to clear
927 : : *
928 : : * Clear @flags in @worker->flags and adjust nr_running accordingly.
929 : : *
930 : : * CONTEXT:
931 : : * spin_lock_irq(pool->lock)
932 : : */
933 : : static inline void worker_clr_flags(struct worker *worker, unsigned int flags)
934 : : {
935 : 1159674 : struct worker_pool *pool = worker->pool;
936 : 1159674 : unsigned int oflags = worker->flags;
937 : :
938 [ - + ][ # # ]: 2320590 : WARN_ON_ONCE(worker->task != current);
[ # # ][ # # ]
[ # # ][ - + ]
[ - + ][ # # ]
[ - + ]
939 : :
940 : 2324475 : worker->flags &= ~flags;
941 : :
942 : : /*
943 : : * If transitioning out of NOT_RUNNING, increment nr_running. Note
944 : : * that the nested NOT_RUNNING is not a noop. NOT_RUNNING is mask
945 : : * of multiple flags, not a single flag.
946 : : */
947 [ + ][ - + ]: 1159708 : if ((flags & WORKER_NOT_RUNNING) && (oflags & WORKER_NOT_RUNNING))
948 [ + + ][ # # ]: 1164220 : if (!(worker->flags & WORKER_NOT_RUNNING))
949 : 1000571 : atomic_inc(&pool->nr_running);
950 : : }
951 : :
952 : : /**
953 : : * find_worker_executing_work - find worker which is executing a work
954 : : * @pool: pool of interest
955 : : * @work: work to find worker for
956 : : *
957 : : * Find a worker which is executing @work on @pool by searching
958 : : * @pool->busy_hash which is keyed by the address of @work. For a worker
959 : : * to match, its current execution should match the address of @work and
960 : : * its work function. This is to avoid unwanted dependency between
961 : : * unrelated work executions through a work item being recycled while still
962 : : * being executed.
963 : : *
964 : : * This is a bit tricky. A work item may be freed once its execution
965 : : * starts and nothing prevents the freed area from being recycled for
966 : : * another work item. If the same work item address ends up being reused
967 : : * before the original execution finishes, workqueue will identify the
968 : : * recycled work item as currently executing and make it wait until the
969 : : * current execution finishes, introducing an unwanted dependency.
970 : : *
971 : : * This function checks the work item address and work function to avoid
972 : : * false positives. Note that this isn't complete as one may construct a
973 : : * work function which can introduce dependency onto itself through a
974 : : * recycled work item. Well, if somebody wants to shoot oneself in the
975 : : * foot that badly, there's only so much we can do, and if such deadlock
976 : : * actually occurs, it should be easy to locate the culprit work function.
977 : : *
978 : : * CONTEXT:
979 : : * spin_lock_irq(pool->lock).
980 : : *
981 : : * Return:
982 : : * Pointer to worker which is executing @work if found, %NULL
983 : : * otherwise.
984 : : */
985 : 0 : static struct worker *find_worker_executing_work(struct worker_pool *pool,
986 : : struct work_struct *work)
987 : : {
988 : : struct worker *worker;
989 : :
990 [ + + ][ # # ]: 3454838 : hash_for_each_possible(pool->busy_hash, worker, hentry,
[ + + ]
991 : : (unsigned long)work)
992 [ + - ][ + ]: 2407 : if (worker->current_work == work &&
993 : 2407 : worker->current_func == work->func)
994 : : return worker;
995 : :
996 : : return NULL;
997 : : }
998 : :
999 : : /**
1000 : : * move_linked_works - move linked works to a list
1001 : : * @work: start of series of works to be scheduled
1002 : : * @head: target list to append @work to
1003 : : * @nextp: out paramter for nested worklist walking
1004 : : *
1005 : : * Schedule linked works starting from @work to @head. Work series to
1006 : : * be scheduled starts at @work and includes any consecutive work with
1007 : : * WORK_STRUCT_LINKED set in its predecessor.
1008 : : *
1009 : : * If @nextp is not NULL, it's updated to point to the next work of
1010 : : * the last scheduled work. This allows move_linked_works() to be
1011 : : * nested inside outer list_for_each_entry_safe().
1012 : : *
1013 : : * CONTEXT:
1014 : : * spin_lock_irq(pool->lock).
1015 : : */
1016 : 0 : static void move_linked_works(struct work_struct *work, struct list_head *head,
1017 : : struct work_struct **nextp)
1018 : : {
1019 : : struct work_struct *n;
1020 : :
1021 : : /*
1022 : : * Linked worklist will always end before the end of the list,
1023 : : * use NULL for list head.
1024 : : */
1025 [ + - ]: 2374 : list_for_each_entry_safe_from(work, n, NULL, entry) {
1026 : : list_move_tail(&work->entry, head);
1027 [ + + ]: 2374 : if (!(*work_data_bits(work) & WORK_STRUCT_LINKED))
1028 : : break;
1029 : : }
1030 : :
1031 : : /*
1032 : : * If we're already inside safe list traversal and have moved
1033 : : * multiple works to the scheduled queue, the next position
1034 : : * needs to be updated.
1035 : : */
1036 [ - - ]: 2326 : if (nextp)
1037 : 0 : *nextp = n;
1038 : 0 : }
1039 : :
1040 : : /**
1041 : : * get_pwq - get an extra reference on the specified pool_workqueue
1042 : : * @pwq: pool_workqueue to get
1043 : : *
1044 : : * Obtain an extra reference on @pwq. The caller should guarantee that
1045 : : * @pwq has positive refcnt and be holding the matching pool->lock.
1046 : : */
1047 : 0 : static void get_pwq(struct pool_workqueue *pwq)
1048 : : {
1049 : : lockdep_assert_held(&pwq->pool->lock);
1050 [ - + ][ # # ]: 1697447 : WARN_ON_ONCE(pwq->refcnt <= 0);
[ # # ]
1051 : 1697447 : pwq->refcnt++;
1052 : 1697447 : }
1053 : :
1054 : : /**
1055 : : * put_pwq - put a pool_workqueue reference
1056 : : * @pwq: pool_workqueue to put
1057 : : *
1058 : : * Drop a reference of @pwq. If its refcnt reaches zero, schedule its
1059 : : * destruction. The caller should be holding the matching pool->lock.
1060 : : */
1061 : 0 : static void put_pwq(struct pool_workqueue *pwq)
1062 : : {
1063 : : lockdep_assert_held(&pwq->pool->lock);
1064 [ - + ]: 1698834 : if (likely(--pwq->refcnt))
1065 : : return;
1066 [ # # ][ # # ]: 0 : if (WARN_ON_ONCE(!(pwq->wq->flags & WQ_UNBOUND)))
[ # # ][ # # ]
1067 : : return;
1068 : : /*
1069 : : * @pwq can't be released under pool->lock, bounce to
1070 : : * pwq_unbound_release_workfn(). This never recurses on the same
1071 : : * pool->lock as this path is taken only for unbound workqueues and
1072 : : * the release work item is scheduled on a per-cpu workqueue. To
1073 : : * avoid lockdep warning, unbound pool->locks are given lockdep
1074 : : * subclass of 1 in get_unbound_pool().
1075 : : */
1076 : 0 : schedule_work(&pwq->unbound_release_work);
1077 : : }
1078 : :
1079 : : /**
1080 : : * put_pwq_unlocked - put_pwq() with surrounding pool lock/unlock
1081 : : * @pwq: pool_workqueue to put (can be %NULL)
1082 : : *
1083 : : * put_pwq() with locking. This function also allows %NULL @pwq.
1084 : : */
1085 : 0 : static void put_pwq_unlocked(struct pool_workqueue *pwq)
1086 : : {
1087 [ # # ]: 0 : if (pwq) {
1088 : : /*
1089 : : * As both pwqs and pools are sched-RCU protected, the
1090 : : * following lock operations are safe.
1091 : : */
1092 : 0 : spin_lock_irq(&pwq->pool->lock);
1093 : 0 : put_pwq(pwq);
1094 : 0 : spin_unlock_irq(&pwq->pool->lock);
1095 : : }
1096 : 0 : }
1097 : :
1098 : 0 : static void pwq_activate_delayed_work(struct work_struct *work)
1099 : : {
1100 : : struct pool_workqueue *pwq = get_work_pwq(work);
1101 : :
1102 : : trace_workqueue_activate_work(work);
1103 : 0 : move_linked_works(work, &pwq->pool->worklist, NULL);
1104 : : __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work));
1105 : 0 : pwq->nr_active++;
1106 : 0 : }
1107 : :
1108 : : static void pwq_activate_first_delayed(struct pool_workqueue *pwq)
1109 : : {
1110 : 0 : struct work_struct *work = list_first_entry(&pwq->delayed_works,
1111 : : struct work_struct, entry);
1112 : :
1113 : 0 : pwq_activate_delayed_work(work);
1114 : : }
1115 : :
1116 : : /**
1117 : : * pwq_dec_nr_in_flight - decrement pwq's nr_in_flight
1118 : : * @pwq: pwq of interest
1119 : : * @color: color of work which left the queue
1120 : : *
1121 : : * A work either has completed or is removed from pending queue,
1122 : : * decrement nr_in_flight of its pwq and handle workqueue flushing.
1123 : : *
1124 : : * CONTEXT:
1125 : : * spin_lock_irq(pool->lock).
1126 : : */
1127 : 0 : static void pwq_dec_nr_in_flight(struct pool_workqueue *pwq, int color)
1128 : : {
1129 : : /* uncolored work items don't participate in flushing or nr_active */
1130 [ + + ]: 1698851 : if (color == WORK_NO_COLOR)
1131 : : goto out_put;
1132 : :
1133 : 1698191 : pwq->nr_in_flight[color]--;
1134 : :
1135 : 1698191 : pwq->nr_active--;
1136 [ - + ]: 1698191 : if (!list_empty(&pwq->delayed_works)) {
1137 : : /* one down, submit a delayed one */
1138 [ # # ]: 0 : if (pwq->nr_active < pwq->max_active)
1139 : : pwq_activate_first_delayed(pwq);
1140 : : }
1141 : :
1142 : : /* is flush in progress and are we at the flushing tip? */
1143 [ - + ]: 1698801 : if (likely(pwq->flush_color != color))
1144 : : goto out_put;
1145 : :
1146 : : /* are there still in-flight works? */
1147 [ # # ]: 0 : if (pwq->nr_in_flight[color])
1148 : : goto out_put;
1149 : :
1150 : : /* this pwq is done, clear flush_color */
1151 : 0 : pwq->flush_color = -1;
1152 : :
1153 : : /*
1154 : : * If this was the last pwq, wake up the first flusher. It
1155 : : * will handle the rest.
1156 : : */
1157 [ # # ]: 0 : if (atomic_dec_and_test(&pwq->wq->nr_pwqs_to_flush))
1158 : 0 : complete(&pwq->wq->first_flusher->done);
1159 : : out_put:
1160 : 1699461 : put_pwq(pwq);
1161 : 1697624 : }
1162 : :
1163 : : /**
1164 : : * try_to_grab_pending - steal work item from worklist and disable irq
1165 : : * @work: work item to steal
1166 : : * @is_dwork: @work is a delayed_work
1167 : : * @flags: place to store irq state
1168 : : *
1169 : : * Try to grab PENDING bit of @work. This function can handle @work in any
1170 : : * stable state - idle, on timer or on worklist.
1171 : : *
1172 : : * Return:
1173 : : * 1 if @work was pending and we successfully stole PENDING
1174 : : * 0 if @work was idle and we claimed PENDING
1175 : : * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry
1176 : : * -ENOENT if someone else is canceling @work, this state may persist
1177 : : * for arbitrarily long
1178 : : *
1179 : : * Note:
1180 : : * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting
1181 : : * interrupted while holding PENDING and @work off queue, irq must be
1182 : : * disabled on entry. This, combined with delayed_work->timer being
1183 : : * irqsafe, ensures that we return -EAGAIN for finite short period of time.
1184 : : *
1185 : : * On successful return, >= 0, irq is disabled and the caller is
1186 : : * responsible for releasing it using local_irq_restore(*@flags).
1187 : : *
1188 : : * This function is safe to call from any context including IRQ handler.
1189 : : */
1190 : 0 : static int try_to_grab_pending(struct work_struct *work, bool is_dwork,
1191 : : unsigned long *flags)
1192 : : {
1193 : : struct worker_pool *pool;
1194 : : struct pool_workqueue *pwq;
1195 : :
1196 : 471149 : local_irq_save(*flags);
1197 : :
1198 : : /* try to steal the timer if it exists */
1199 [ + + ]: 471149 : if (is_dwork) {
1200 : : struct delayed_work *dwork = to_delayed_work(work);
1201 : :
1202 : : /*
1203 : : * dwork->timer is irqsafe. If del_timer() fails, it's
1204 : : * guaranteed that the timer is not queued anywhere and not
1205 : : * running on the local CPU.
1206 : : */
1207 [ + + ]: 470535 : if (likely(del_timer(&dwork->timer)))
1208 : : return 1;
1209 : : }
1210 : :
1211 : : /* try to claim PENDING the normal way */
1212 [ + + ]: 167995 : if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work)))
1213 : : return 0;
1214 : :
1215 : : /*
1216 : : * The queueing is in progress, or it is already queued. Try to
1217 : : * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
1218 : : */
1219 : 7410 : pool = get_work_pool(work);
1220 [ + - ]: 7410 : if (!pool)
1221 : : goto fail;
1222 : :
1223 : : spin_lock(&pool->lock);
1224 : : /*
1225 : : * work->data is guaranteed to point to pwq only while the work
1226 : : * item is queued on pwq->wq, and both updating work->data to point
1227 : : * to pwq on queueing and to pool on dequeueing are done under
1228 : : * pwq->pool->lock. This in turn guarantees that, if work->data
1229 : : * points to pwq which is associated with a locked pool, the work
1230 : : * item is currently queued on that pool.
1231 : : */
1232 : : pwq = get_work_pwq(work);
1233 [ + + ][ + - ]: 7410 : if (pwq && pwq->pool == pool) {
1234 : : debug_work_deactivate(work);
1235 : :
1236 : : /*
1237 : : * A delayed work item cannot be grabbed directly because
1238 : : * it might have linked NO_COLOR work items which, if left
1239 : : * on the delayed_list, will confuse pwq->nr_active
1240 : : * management later on and cause stall. Make sure the work
1241 : : * item is activated before grabbing.
1242 : : */
1243 [ - + ]: 7334 : if (*work_data_bits(work) & WORK_STRUCT_DELAYED)
1244 : 0 : pwq_activate_delayed_work(work);
1245 : :
1246 : 7334 : list_del_init(&work->entry);
1247 : 7334 : pwq_dec_nr_in_flight(get_work_pwq(work), get_work_color(work));
1248 : :
1249 : : /* work->data points to pwq iff queued, point to pool */
1250 : 7334 : set_work_pool_and_keep_pending(work, pool->id);
1251 : :
1252 : : spin_unlock(&pool->lock);
1253 : 7334 : return 1;
1254 : : }
1255 : : spin_unlock(&pool->lock);
1256 : : fail:
1257 [ - + ]: 76 : local_irq_restore(*flags);
1258 [ + - ]: 76 : if (work_is_canceling(work))
1259 : : return -ENOENT;
1260 : 76 : cpu_relax();
1261 : 76 : return -EAGAIN;
1262 : : }
1263 : :
1264 : : /**
1265 : : * insert_work - insert a work into a pool
1266 : : * @pwq: pwq @work belongs to
1267 : : * @work: work to insert
1268 : : * @head: insertion point
1269 : : * @extra_flags: extra WORK_STRUCT_* flags to set
1270 : : *
1271 : : * Insert @work which belongs to @pwq after @head. @extra_flags is or'd to
1272 : : * work_struct flags.
1273 : : *
1274 : : * CONTEXT:
1275 : : * spin_lock_irq(pool->lock).
1276 : : */
1277 : 0 : static void insert_work(struct pool_workqueue *pwq, struct work_struct *work,
1278 : : struct list_head *head, unsigned int extra_flags)
1279 : : {
1280 : 1697147 : struct worker_pool *pool = pwq->pool;
1281 : :
1282 : : /* we own @work, set data and link */
1283 : 1697147 : set_work_pwq(work, pwq, extra_flags);
1284 : 1698451 : list_add_tail(&work->entry, head);
1285 : 1698451 : get_pwq(pwq);
1286 : :
1287 : : /*
1288 : : * Ensure either wq_worker_sleeping() sees the above
1289 : : * list_add_tail() or we see zero nr_running to avoid workers lying
1290 : : * around lazily while there are works to be processed.
1291 : : */
1292 : 1697882 : smp_mb();
1293 : :
1294 [ + + ]: 1698177 : if (__need_more_worker(pool))
1295 : 1698156 : wake_up_worker(pool);
1296 : 385 : }
1297 : :
1298 : : /*
1299 : : * Test whether @work is being queued from another work executing on the
1300 : : * same workqueue.
1301 : : */
1302 : 0 : static bool is_chained_work(struct workqueue_struct *wq)
1303 : : {
1304 : : struct worker *worker;
1305 : :
1306 : : worker = current_wq_worker();
1307 : : /*
1308 : : * Return %true iff I'm a worker execuing a work item on @wq. If
1309 : : * I'm @worker, it's safe to dereference it without locking.
1310 : : */
1311 [ # # ][ # # ]: 0 : return worker && worker->current_pwq->wq == wq;
1312 : : }
1313 : :
1314 : 0 : static void __queue_work(int cpu, struct workqueue_struct *wq,
1315 : : struct work_struct *work)
1316 : : {
1317 : : struct pool_workqueue *pwq;
1318 : : struct worker_pool *last_pool;
1319 : : struct list_head *worklist;
1320 : : unsigned int work_flags;
1321 : 1698587 : unsigned int req_cpu = cpu;
1322 : :
1323 : : /*
1324 : : * While a work item is PENDING && off queue, a task trying to
1325 : : * steal the PENDING will busy-loop waiting for it to either get
1326 : : * queued or lose PENDING. Grabbing PENDING and queueing should
1327 : : * happen with IRQ disabled.
1328 : : */
1329 [ - + ][ # # ]: 1697792 : WARN_ON_ONCE(!irqs_disabled());
[ - + ]
1330 : :
1331 : : debug_work_activate(work);
1332 : :
1333 : : /* if draining, only works from the same workqueue are allowed */
1334 [ + - ][ # # ]: 1698389 : if (unlikely(wq->flags & __WQ_DRAINING) &&
1335 [ # # ][ # # ]: 0 : WARN_ON_ONCE(!is_chained_work(wq)))
[ # # ]
1336 : : return;
1337 : : retry:
1338 [ + + ]: 1698389 : if (req_cpu == WORK_CPU_UNBOUND)
1339 : 1538612 : cpu = raw_smp_processor_id();
1340 : :
1341 : : /* pwq which will be used unless @work is executing elsewhere */
1342 [ + + ]: 1698389 : if (!(wq->flags & WQ_UNBOUND))
1343 : 1537705 : pwq = per_cpu_ptr(wq->cpu_pwqs, cpu);
1344 : : else
1345 : : pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu));
1346 : :
1347 : : /*
1348 : : * If @work was previously on a different pool, it might still be
1349 : : * running there, in which case the work needs to be queued on that
1350 : : * pool to guarantee non-reentrancy.
1351 : : */
1352 : 1698389 : last_pool = get_work_pool(work);
1353 [ + + ][ + + ]: 1698448 : if (last_pool && last_pool != pwq->pool) {
1354 : : struct worker *worker;
1355 : :
1356 : : spin_lock(&last_pool->lock);
1357 : :
1358 : 33139 : worker = find_worker_executing_work(last_pool, work);
1359 : :
1360 [ + + ][ + ]: 33139 : if (worker && worker->current_pwq->wq == wq) {
1361 : : pwq = worker->current_pwq;
1362 : : } else {
1363 : : /* meh... not running there, queue here */
1364 : : spin_unlock(&last_pool->lock);
1365 : 33048 : spin_lock(&pwq->pool->lock);
1366 : : }
1367 : : } else {
1368 : 1665309 : spin_lock(&pwq->pool->lock);
1369 : : }
1370 : :
1371 : : /*
1372 : : * pwq is determined and locked. For unbound pools, we could have
1373 : : * raced with pwq release and it could already be dead. If its
1374 : : * refcnt is zero, repeat pwq selection. Note that pwqs never die
1375 : : * without another pwq replacing it in the numa_pwq_tbl or while
1376 : : * work items are executing on it, so the retrying is guaranteed to
1377 : : * make forward-progress.
1378 : : */
1379 [ - + ]: 3397151 : if (unlikely(!pwq->refcnt)) {
1380 [ # # ]: 0 : if (wq->flags & WQ_UNBOUND) {
1381 : 0 : spin_unlock(&pwq->pool->lock);
1382 : 0 : cpu_relax();
1383 : 0 : goto retry;
1384 : : }
1385 : : /* oops */
1386 [ # # ][ # # ]: 0 : WARN_ONCE(true, "workqueue: per-cpu pwq for %s on cpu%d has 0 refcnt",
1387 : : wq->name, cpu);
1388 : : }
1389 : :
1390 : : /* pwq determined, queue */
1391 : : trace_workqueue_queue_work(req_cpu, pwq, work);
1392 : :
1393 [ - + ][ - + ]: 1698564 : if (WARN_ON(!list_empty(&work->entry))) {
1394 : 0 : spin_unlock(&pwq->pool->lock);
1395 : : return;
1396 : : }
1397 : :
1398 : 1698564 : pwq->nr_in_flight[pwq->work_color]++;
1399 : : work_flags = work_color_to_flags(pwq->work_color);
1400 : :
1401 [ + - ]: 1698564 : if (likely(pwq->nr_active < pwq->max_active)) {
1402 : : trace_workqueue_activate_work(work);
1403 : 1698525 : pwq->nr_active++;
1404 : 1698525 : worklist = &pwq->pool->worklist;
1405 : : } else {
1406 : 0 : work_flags |= WORK_STRUCT_DELAYED;
1407 : 0 : worklist = &pwq->delayed_works;
1408 : : }
1409 : :
1410 : 1698525 : insert_work(pwq, work, worklist, work_flags);
1411 : :
1412 : 1698209 : spin_unlock(&pwq->pool->lock);
1413 : : }
1414 : :
1415 : : /**
1416 : : * queue_work_on - queue work on specific cpu
1417 : : * @cpu: CPU number to execute work on
1418 : : * @wq: workqueue to use
1419 : : * @work: work to queue
1420 : : *
1421 : : * We queue the work to a specific CPU, the caller must ensure it
1422 : : * can't go away.
1423 : : *
1424 : : * Return: %false if @work was already on a queue, %true otherwise.
1425 : : */
1426 : 0 : bool queue_work_on(int cpu, struct workqueue_struct *wq,
1427 : : struct work_struct *work)
1428 : : {
1429 : : bool ret = false;
1430 : : unsigned long flags;
1431 : :
1432 : : local_irq_save(flags);
1433 : :
1434 [ + + ]: 424466 : if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
1435 : 327905 : __queue_work(cpu, wq, work);
1436 : : ret = true;
1437 : : }
1438 : :
1439 [ + + ]: 424464 : local_irq_restore(flags);
1440 : 424465 : return ret;
1441 : : }
1442 : : EXPORT_SYMBOL(queue_work_on);
1443 : :
1444 : 0 : void delayed_work_timer_fn(unsigned long __data)
1445 : : {
1446 : 1362532 : struct delayed_work *dwork = (struct delayed_work *)__data;
1447 : :
1448 : : /* should have been called from irqsafe timer with irq already off */
1449 : 1362532 : __queue_work(dwork->cpu, dwork->wq, &dwork->work);
1450 : 1360165 : }
1451 : : EXPORT_SYMBOL(delayed_work_timer_fn);
1452 : :
1453 : 0 : static void __queue_delayed_work(int cpu, struct workqueue_struct *wq,
1454 : : struct delayed_work *dwork, unsigned long delay)
1455 : : {
1456 : 3347803 : struct timer_list *timer = &dwork->timer;
1457 : : struct work_struct *work = &dwork->work;
1458 : :
1459 [ + + ][ + ]: 1673750 : WARN_ON_ONCE(timer->function != delayed_work_timer_fn ||
[ - + ][ # # ]
[ - + ]
1460 : : timer->data != (unsigned long)dwork);
1461 [ - + ][ # # ]: 3347803 : WARN_ON_ONCE(timer_pending(timer));
[ # # ]
1462 [ - + ][ # # ]: 1674053 : WARN_ON_ONCE(!list_empty(&work->entry));
[ # # ]
1463 : :
1464 : : /*
1465 : : * If @delay is 0, queue @dwork->work immediately. This is for
1466 : : * both optimization and correctness. The earliest @timer can
1467 : : * expire is on the closest next tick and delayed_work users depend
1468 : : * on that there's no such delay when @delay is 0.
1469 : : */
1470 [ + + ]: 1674053 : if (!delay) {
1471 : 7939 : __queue_work(cpu, wq, &dwork->work);
1472 : 7939 : return;
1473 : : }
1474 : :
1475 : : timer_stats_timer_set_start_info(&dwork->timer);
1476 : :
1477 : 1665344 : dwork->wq = wq;
1478 : 1665344 : dwork->cpu = cpu;
1479 : 1665344 : timer->expires = jiffies + delay;
1480 : :
1481 [ + + ]: 1665344 : if (unlikely(cpu != WORK_CPU_UNBOUND))
1482 : 460115 : add_timer_on(timer, cpu);
1483 : : else
1484 : 1205229 : add_timer(timer);
1485 : : }
1486 : :
1487 : : /**
1488 : : * queue_delayed_work_on - queue work on specific CPU after delay
1489 : : * @cpu: CPU number to execute work on
1490 : : * @wq: workqueue to use
1491 : : * @dwork: work to queue
1492 : : * @delay: number of jiffies to wait before queueing
1493 : : *
1494 : : * Return: %false if @work was already on a queue, %true otherwise. If
1495 : : * @delay is zero and @dwork is idle, it will be scheduled for immediate
1496 : : * execution.
1497 : : */
1498 : 0 : bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
1499 : : struct delayed_work *dwork, unsigned long delay)
1500 : : {
1501 : : struct work_struct *work = &dwork->work;
1502 : : bool ret = false;
1503 : : unsigned long flags;
1504 : :
1505 : : /* read the comment in __queue_work() */
1506 : : local_irq_save(flags);
1507 : :
1508 [ + ]: 1203628 : if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
1509 : 1203328 : __queue_delayed_work(cpu, wq, dwork, delay);
1510 : : ret = true;
1511 : : }
1512 : :
1513 [ + + ]: 1204244 : local_irq_restore(flags);
1514 : 1204247 : return ret;
1515 : : }
1516 : : EXPORT_SYMBOL(queue_delayed_work_on);
1517 : :
1518 : : /**
1519 : : * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU
1520 : : * @cpu: CPU number to execute work on
1521 : : * @wq: workqueue to use
1522 : : * @dwork: work to queue
1523 : : * @delay: number of jiffies to wait before queueing
1524 : : *
1525 : : * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise,
1526 : : * modify @dwork's timer so that it expires after @delay. If @delay is
1527 : : * zero, @work is guaranteed to be scheduled immediately regardless of its
1528 : : * current state.
1529 : : *
1530 : : * Return: %false if @dwork was idle and queued, %true if @dwork was
1531 : : * pending and its timer was modified.
1532 : : *
1533 : : * This function is safe to call from any context including IRQ handler.
1534 : : * See try_to_grab_pending() for details.
1535 : : */
1536 : 469885 : bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq,
1537 : : struct delayed_work *dwork, unsigned long delay)
1538 : : {
1539 : : unsigned long flags;
1540 : : int ret;
1541 : :
1542 : : do {
1543 : 469961 : ret = try_to_grab_pending(&dwork->work, true, &flags);
1544 [ + + ]: 469961 : } while (unlikely(ret == -EAGAIN));
1545 : :
1546 [ + - ]: 469885 : if (likely(ret >= 0)) {
1547 : 469885 : __queue_delayed_work(cpu, wq, dwork, delay);
1548 [ + + ]: 469885 : local_irq_restore(flags);
1549 : : }
1550 : :
1551 : : /* -ENOENT from try_to_grab_pending() becomes %true */
1552 : 469885 : return ret;
1553 : : }
1554 : : EXPORT_SYMBOL_GPL(mod_delayed_work_on);
1555 : :
1556 : : /**
1557 : : * worker_enter_idle - enter idle state
1558 : : * @worker: worker which is entering idle state
1559 : : *
1560 : : * @worker is entering idle state. Update stats and idle timer if
1561 : : * necessary.
1562 : : *
1563 : : * LOCKING:
1564 : : * spin_lock_irq(pool->lock).
1565 : : */
1566 : 0 : static void worker_enter_idle(struct worker *worker)
1567 : : {
1568 : 1166126 : struct worker_pool *pool = worker->pool;
1569 : :
1570 [ - + ][ # # ]: 2332251 : if (WARN_ON_ONCE(worker->flags & WORKER_IDLE) ||
[ - + ][ + - ]
[ + - ]
1571 [ + + ][ + ]: 1166131 : WARN_ON_ONCE(!list_empty(&worker->entry) &&
[ + + ][ - + ]
[ # # ][ - ]
1572 : : (worker->hentry.next || worker->hentry.pprev)))
1573 : 1166119 : return;
1574 : :
1575 : : /* can't use worker_set_flags(), also called from start_worker() */
1576 : 1166125 : worker->flags |= WORKER_IDLE;
1577 : 1166125 : pool->nr_idle++;
1578 : 1166125 : worker->last_active = jiffies;
1579 : :
1580 : : /* idle_list is LIFO */
1581 : 1166125 : list_add(&worker->entry, &pool->idle_list);
1582 : :
1583 [ + ][ + + ]: 1166125 : if (too_many_workers(pool) && !timer_pending(&pool->idle_timer))
1584 : 491 : mod_timer(&pool->idle_timer, jiffies + IDLE_WORKER_TIMEOUT);
1585 : :
1586 : : /*
1587 : : * Sanity check nr_running. Because wq_unbind_fn() releases
1588 : : * pool->lock between setting %WORKER_UNBOUND and zapping
1589 : : * nr_running, the warning may trigger spuriously. Check iff
1590 : : * unbind is not in progress.
1591 : : */
1592 [ + + ][ + + ]: 1166119 : WARN_ON_ONCE(!(pool->flags & POOL_DISASSOCIATED) &&
[ + + ][ - + ]
[ # # ][ # # ]
1593 : : pool->nr_workers == pool->nr_idle &&
1594 : : atomic_read(&pool->nr_running));
1595 : : }
1596 : :
1597 : : /**
1598 : : * worker_leave_idle - leave idle state
1599 : : * @worker: worker which is leaving idle state
1600 : : *
1601 : : * @worker is leaving idle state. Update stats.
1602 : : *
1603 : : * LOCKING:
1604 : : * spin_lock_irq(pool->lock).
1605 : : */
1606 : 0 : static void worker_leave_idle(struct worker *worker)
1607 : : {
1608 : 1160437 : struct worker_pool *pool = worker->pool;
1609 : :
1610 [ - + ][ # # ]: 1160437 : if (WARN_ON_ONCE(!(worker->flags & WORKER_IDLE)))
[ - + ][ + - ]
1611 : 1164767 : return;
1612 : : worker_clr_flags(worker, WORKER_IDLE);
1613 : 1164767 : pool->nr_idle--;
1614 : 1164767 : list_del_init(&worker->entry);
1615 : : }
1616 : :
1617 : : /**
1618 : : * worker_maybe_bind_and_lock - try to bind %current to worker_pool and lock it
1619 : : * @pool: target worker_pool
1620 : : *
1621 : : * Bind %current to the cpu of @pool if it is associated and lock @pool.
1622 : : *
1623 : : * Works which are scheduled while the cpu is online must at least be
1624 : : * scheduled to a worker which is bound to the cpu so that if they are
1625 : : * flushed from cpu callbacks while cpu is going down, they are
1626 : : * guaranteed to execute on the cpu.
1627 : : *
1628 : : * This function is to be used by unbound workers and rescuers to bind
1629 : : * themselves to the target cpu and may race with cpu going down or
1630 : : * coming online. kthread_bind() can't be used because it may put the
1631 : : * worker to already dead cpu and set_cpus_allowed_ptr() can't be used
1632 : : * verbatim as it's best effort and blocking and pool may be
1633 : : * [dis]associated in the meantime.
1634 : : *
1635 : : * This function tries set_cpus_allowed() and locks pool and verifies the
1636 : : * binding against %POOL_DISASSOCIATED which is set during
1637 : : * %CPU_DOWN_PREPARE and cleared during %CPU_ONLINE, so if the worker
1638 : : * enters idle state or fetches works without dropping lock, it can
1639 : : * guarantee the scheduling requirement described in the first paragraph.
1640 : : *
1641 : : * CONTEXT:
1642 : : * Might sleep. Called without any lock but returns with pool->lock
1643 : : * held.
1644 : : *
1645 : : * Return:
1646 : : * %true if the associated pool is online (@worker is successfully
1647 : : * bound), %false if offline.
1648 : : */
1649 : 0 : static bool worker_maybe_bind_and_lock(struct worker_pool *pool)
1650 : : __acquires(&pool->lock)
1651 : : {
1652 : : while (true) {
1653 : : /*
1654 : : * The following call may fail, succeed or succeed
1655 : : * without actually migrating the task to the cpu if
1656 : : * it races with cpu hotunplug operation. Verify
1657 : : * against POOL_DISASSOCIATED.
1658 : : */
1659 [ # # ]: 0 : if (!(pool->flags & POOL_DISASSOCIATED))
1660 : 0 : set_cpus_allowed_ptr(current, pool->attrs->cpumask);
1661 : :
1662 : : spin_lock_irq(&pool->lock);
1663 [ # # ]: 0 : if (pool->flags & POOL_DISASSOCIATED)
1664 : : return false;
1665 [ # # ][ # # ]: 0 : if (task_cpu(current) == pool->cpu &&
1666 : 0 : cpumask_equal(¤t->cpus_allowed, pool->attrs->cpumask))
1667 : : return true;
1668 : : spin_unlock_irq(&pool->lock);
1669 : :
1670 : : /*
1671 : : * We've raced with CPU hot[un]plug. Give it a breather
1672 : : * and retry migration. cond_resched() is required here;
1673 : : * otherwise, we might deadlock against cpu_stop trying to
1674 : : * bring down the CPU on non-preemptive kernel.
1675 : : */
1676 : 0 : cpu_relax();
1677 : 0 : cond_resched();
1678 : 0 : }
1679 : : }
1680 : :
1681 : 0 : static struct worker *alloc_worker(void)
1682 : : {
1683 : : struct worker *worker;
1684 : :
1685 : : worker = kzalloc(sizeof(*worker), GFP_KERNEL);
1686 [ + - ]: 495 : if (worker) {
1687 : 495 : INIT_LIST_HEAD(&worker->entry);
1688 : 495 : INIT_LIST_HEAD(&worker->scheduled);
1689 : : /* on creation a worker is in !idle && prep state */
1690 : 495 : worker->flags = WORKER_PREP;
1691 : : }
1692 : 0 : return worker;
1693 : : }
1694 : :
1695 : : /**
1696 : : * create_worker - create a new workqueue worker
1697 : : * @pool: pool the new worker will belong to
1698 : : *
1699 : : * Create a new worker which is bound to @pool. The returned worker
1700 : : * can be started by calling start_worker() or destroyed using
1701 : : * destroy_worker().
1702 : : *
1703 : : * CONTEXT:
1704 : : * Might sleep. Does GFP_KERNEL allocations.
1705 : : *
1706 : : * Return:
1707 : : * Pointer to the newly created worker.
1708 : : */
1709 : 0 : static struct worker *create_worker(struct worker_pool *pool)
1710 : : {
1711 : : struct worker *worker = NULL;
1712 : : int id = -1;
1713 : : char id_buf[16];
1714 : :
1715 : : lockdep_assert_held(&pool->manager_mutex);
1716 : :
1717 : : /*
1718 : : * ID is needed to determine kthread name. Allocate ID first
1719 : : * without installing the pointer.
1720 : : */
1721 : 495 : idr_preload(GFP_KERNEL);
1722 : : spin_lock_irq(&pool->lock);
1723 : :
1724 : 495 : id = idr_alloc(&pool->worker_idr, NULL, 0, 0, GFP_NOWAIT);
1725 : :
1726 : : spin_unlock_irq(&pool->lock);
1727 : : idr_preload_end();
1728 [ + - ]: 495 : if (id < 0)
1729 : : goto fail;
1730 : :
1731 : 495 : worker = alloc_worker();
1732 [ + - ]: 495 : if (!worker)
1733 : : goto fail;
1734 : :
1735 : 495 : worker->pool = pool;
1736 : 495 : worker->id = id;
1737 : :
1738 [ + + ]: 495 : if (pool->cpu >= 0)
1739 [ + - ]: 469 : snprintf(id_buf, sizeof(id_buf), "%d:%d%s", pool->cpu, id,
1740 : 469 : pool->attrs->nice < 0 ? "H" : "");
1741 : : else
1742 : 26 : snprintf(id_buf, sizeof(id_buf), "u%d:%d", pool->id, id);
1743 : :
1744 : 495 : worker->task = kthread_create_on_node(worker_thread, worker, pool->node,
1745 : : "kworker/%s", id_buf);
1746 [ + - ]: 495 : if (IS_ERR(worker->task))
1747 : : goto fail;
1748 : :
1749 : 495 : set_user_nice(worker->task, pool->attrs->nice);
1750 : :
1751 : : /* prevent userland from meddling with cpumask of workqueue workers */
1752 : 495 : worker->task->flags |= PF_NO_SETAFFINITY;
1753 : :
1754 : : /*
1755 : : * set_cpus_allowed_ptr() will fail if the cpumask doesn't have any
1756 : : * online CPUs. It'll be re-applied when any of the CPUs come up.
1757 : : */
1758 : 495 : set_cpus_allowed_ptr(worker->task, pool->attrs->cpumask);
1759 : :
1760 : : /*
1761 : : * The caller is responsible for ensuring %POOL_DISASSOCIATED
1762 : : * remains stable across this function. See the comments above the
1763 : : * flag definition for details.
1764 : : */
1765 [ + + ]: 495 : if (pool->flags & POOL_DISASSOCIATED)
1766 : 26 : worker->flags |= WORKER_UNBOUND;
1767 : :
1768 : : /* successful, commit the pointer to idr */
1769 : : spin_lock_irq(&pool->lock);
1770 : 495 : idr_replace(&pool->worker_idr, worker, worker->id);
1771 : : spin_unlock_irq(&pool->lock);
1772 : :
1773 : 495 : return worker;
1774 : :
1775 : : fail:
1776 [ - ]: 0 : if (id >= 0) {
1777 : : spin_lock_irq(&pool->lock);
1778 : 0 : idr_remove(&pool->worker_idr, id);
1779 : : spin_unlock_irq(&pool->lock);
1780 : : }
1781 : 0 : kfree(worker);
1782 : 0 : return NULL;
1783 : : }
1784 : :
1785 : : /**
1786 : : * start_worker - start a newly created worker
1787 : : * @worker: worker to start
1788 : : *
1789 : : * Make the pool aware of @worker and start it.
1790 : : *
1791 : : * CONTEXT:
1792 : : * spin_lock_irq(pool->lock).
1793 : : */
1794 : 0 : static void start_worker(struct worker *worker)
1795 : : {
1796 : 495 : worker->flags |= WORKER_STARTED;
1797 : 495 : worker->pool->nr_workers++;
1798 : 495 : worker_enter_idle(worker);
1799 : 495 : wake_up_process(worker->task);
1800 : 495 : }
1801 : :
1802 : : /**
1803 : : * create_and_start_worker - create and start a worker for a pool
1804 : : * @pool: the target pool
1805 : : *
1806 : : * Grab the managership of @pool and create and start a new worker for it.
1807 : : *
1808 : : * Return: 0 on success. A negative error code otherwise.
1809 : : */
1810 : 0 : static int create_and_start_worker(struct worker_pool *pool)
1811 : : {
1812 : : struct worker *worker;
1813 : :
1814 : 0 : mutex_lock(&pool->manager_mutex);
1815 : :
1816 : 0 : worker = create_worker(pool);
1817 [ # # ]: 0 : if (worker) {
1818 : : spin_lock_irq(&pool->lock);
1819 : 0 : start_worker(worker);
1820 : : spin_unlock_irq(&pool->lock);
1821 : : }
1822 : :
1823 : 0 : mutex_unlock(&pool->manager_mutex);
1824 : :
1825 [ # # ]: 0 : return worker ? 0 : -ENOMEM;
1826 : : }
1827 : :
1828 : : /**
1829 : : * destroy_worker - destroy a workqueue worker
1830 : : * @worker: worker to be destroyed
1831 : : *
1832 : : * Destroy @worker and adjust @pool stats accordingly.
1833 : : *
1834 : : * CONTEXT:
1835 : : * spin_lock_irq(pool->lock) which is released and regrabbed.
1836 : : */
1837 : 0 : static void destroy_worker(struct worker *worker)
1838 : : {
1839 : 496 : struct worker_pool *pool = worker->pool;
1840 : :
1841 : : lockdep_assert_held(&pool->manager_mutex);
1842 : : lockdep_assert_held(&pool->lock);
1843 : :
1844 : : /* sanity check frenzy */
1845 [ - + ][ + - ]: 992 : if (WARN_ON(worker->current_work) ||
[ + - ]
1846 [ - + ]: 496 : WARN_ON(!list_empty(&worker->scheduled)))
1847 : 496 : return;
1848 : :
1849 [ + - ]: 496 : if (worker->flags & WORKER_STARTED)
1850 : 496 : pool->nr_workers--;
1851 [ + - ]: 496 : if (worker->flags & WORKER_IDLE)
1852 : 496 : pool->nr_idle--;
1853 : :
1854 : 0 : list_del_init(&worker->entry);
1855 : 0 : worker->flags |= WORKER_DIE;
1856 : :
1857 : 496 : idr_remove(&pool->worker_idr, worker->id);
1858 : :
1859 : : spin_unlock_irq(&pool->lock);
1860 : :
1861 : 496 : kthread_stop(worker->task);
1862 : 496 : kfree(worker);
1863 : :
1864 : : spin_lock_irq(&pool->lock);
1865 : : }
1866 : :
1867 : 0 : static void idle_worker_timeout(unsigned long __pool)
1868 : : {
1869 : 516 : struct worker_pool *pool = (void *)__pool;
1870 : :
1871 : : spin_lock_irq(&pool->lock);
1872 : :
1873 [ # # ]: 516 : if (too_many_workers(pool)) {
1874 : : struct worker *worker;
1875 : : unsigned long expires;
1876 : :
1877 : : /* idle_list is kept in LIFO order, check the last one */
1878 : 512 : worker = list_entry(pool->idle_list.prev, struct worker, entry);
1879 : 512 : expires = worker->last_active + IDLE_WORKER_TIMEOUT;
1880 : :
1881 [ + + ]: 512 : if (time_before(jiffies, expires))
1882 : 17 : mod_timer(&pool->idle_timer, expires);
1883 : : else {
1884 : : /* it's been idle for too long, wake up manager */
1885 : 495 : pool->flags |= POOL_MANAGE_WORKERS;
1886 : 495 : wake_up_worker(pool);
1887 : : }
1888 : : }
1889 : :
1890 : : spin_unlock_irq(&pool->lock);
1891 : 516 : }
1892 : :
1893 : 0 : static void send_mayday(struct work_struct *work)
1894 : : {
1895 : : struct pool_workqueue *pwq = get_work_pwq(work);
1896 : 0 : struct workqueue_struct *wq = pwq->wq;
1897 : :
1898 : : lockdep_assert_held(&wq_mayday_lock);
1899 : :
1900 [ # # ]: 0 : if (!wq->rescuer)
1901 : 0 : return;
1902 : :
1903 : : /* mayday mayday mayday */
1904 [ # # ]: 0 : if (list_empty(&pwq->mayday_node)) {
1905 : 0 : list_add_tail(&pwq->mayday_node, &wq->maydays);
1906 : 0 : wake_up_process(wq->rescuer->task);
1907 : : }
1908 : : }
1909 : :
1910 : 0 : static void pool_mayday_timeout(unsigned long __pool)
1911 : : {
1912 : 1 : struct worker_pool *pool = (void *)__pool;
1913 : : struct work_struct *work;
1914 : :
1915 : : spin_lock_irq(&wq_mayday_lock); /* for wq->maydays */
1916 : : spin_lock(&pool->lock);
1917 : :
1918 [ # # ]: 1 : if (need_to_create_worker(pool)) {
1919 : : /*
1920 : : * We've been trying to create a new worker but
1921 : : * haven't been successful. We might be hitting an
1922 : : * allocation deadlock. Send distress signals to
1923 : : * rescuers.
1924 : : */
1925 [ # # ]: 0 : list_for_each_entry(work, &pool->worklist, entry)
1926 : 0 : send_mayday(work);
1927 : : }
1928 : :
1929 : : spin_unlock(&pool->lock);
1930 : : spin_unlock_irq(&wq_mayday_lock);
1931 : :
1932 : 1 : mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INTERVAL);
1933 : 1 : }
1934 : :
1935 : : /**
1936 : : * maybe_create_worker - create a new worker if necessary
1937 : : * @pool: pool to create a new worker for
1938 : : *
1939 : : * Create a new worker for @pool if necessary. @pool is guaranteed to
1940 : : * have at least one idle worker on return from this function. If
1941 : : * creating a new worker takes longer than MAYDAY_INTERVAL, mayday is
1942 : : * sent to all rescuers with works scheduled on @pool to resolve
1943 : : * possible allocation deadlock.
1944 : : *
1945 : : * On return, need_to_create_worker() is guaranteed to be %false and
1946 : : * may_start_working() %true.
1947 : : *
1948 : : * LOCKING:
1949 : : * spin_lock_irq(pool->lock) which may be released and regrabbed
1950 : : * multiple times. Does GFP_KERNEL allocations. Called only from
1951 : : * manager.
1952 : : *
1953 : : * Return:
1954 : : * %false if no action was taken and pool->lock stayed locked, %true
1955 : : * otherwise.
1956 : : */
1957 : 0 : static bool maybe_create_worker(struct worker_pool *pool)
1958 : : __releases(&pool->lock)
1959 : : __acquires(&pool->lock)
1960 : : {
1961 [ + + ]: 990 : if (!need_to_create_worker(pool))
1962 : : return false;
1963 : : restart:
1964 : : spin_unlock_irq(&pool->lock);
1965 : :
1966 : : /* if we don't make progress in MAYDAY_INITIAL_TIMEOUT, call for help */
1967 : 495 : mod_timer(&pool->mayday_timer, jiffies + MAYDAY_INITIAL_TIMEOUT);
1968 : :
1969 : : while (true) {
1970 : : struct worker *worker;
1971 : :
1972 : 495 : worker = create_worker(pool);
1973 [ + - ]: 495 : if (worker) {
1974 : 495 : del_timer_sync(&pool->mayday_timer);
1975 : : spin_lock_irq(&pool->lock);
1976 : 495 : start_worker(worker);
1977 [ - + ][ # # ]: 495 : if (WARN_ON_ONCE(need_to_create_worker(pool)))
[ # # ][ - + ]
1978 : : goto restart;
1979 : : return true;
1980 : : }
1981 : :
1982 [ # # ]: 0 : if (!need_to_create_worker(pool))
1983 : : break;
1984 : :
1985 : 0 : __set_current_state(TASK_INTERRUPTIBLE);
1986 : 0 : schedule_timeout(CREATE_COOLDOWN);
1987 : :
1988 [ # # ]: 0 : if (!need_to_create_worker(pool))
1989 : : break;
1990 : : }
1991 : :
1992 : 0 : del_timer_sync(&pool->mayday_timer);
1993 : : spin_lock_irq(&pool->lock);
1994 [ # # ]: 0 : if (need_to_create_worker(pool))
1995 : : goto restart;
1996 : : return true;
1997 : : }
1998 : :
1999 : : /**
2000 : : * maybe_destroy_worker - destroy workers which have been idle for a while
2001 : : * @pool: pool to destroy workers for
2002 : : *
2003 : : * Destroy @pool workers which have been idle for longer than
2004 : : * IDLE_WORKER_TIMEOUT.
2005 : : *
2006 : : * LOCKING:
2007 : : * spin_lock_irq(pool->lock) which may be released and regrabbed
2008 : : * multiple times. Called only from manager.
2009 : : *
2010 : : * Return:
2011 : : * %false if no action was taken and pool->lock stayed locked, %true
2012 : : * otherwise.
2013 : : */
2014 : 0 : static bool maybe_destroy_workers(struct worker_pool *pool)
2015 : : {
2016 : : bool ret = false;
2017 : :
2018 [ + + ]: 2476 : while (too_many_workers(pool)) {
2019 : : struct worker *worker;
2020 : : unsigned long expires;
2021 : :
2022 : 509 : worker = list_entry(pool->idle_list.prev, struct worker, entry);
2023 : 509 : expires = worker->last_active + IDLE_WORKER_TIMEOUT;
2024 : :
2025 [ + + ]: 509 : if (time_before(jiffies, expires)) {
2026 : 13 : mod_timer(&pool->idle_timer, expires);
2027 : 13 : break;
2028 : : }
2029 : :
2030 : 496 : destroy_worker(worker);
2031 : : ret = true;
2032 : : }
2033 : :
2034 : 990 : return ret;
2035 : : }
2036 : :
2037 : : /**
2038 : : * manage_workers - manage worker pool
2039 : : * @worker: self
2040 : : *
2041 : : * Assume the manager role and manage the worker pool @worker belongs
2042 : : * to. At any given time, there can be only zero or one manager per
2043 : : * pool. The exclusion is handled automatically by this function.
2044 : : *
2045 : : * The caller can safely start processing works on false return. On
2046 : : * true return, it's guaranteed that need_to_create_worker() is false
2047 : : * and may_start_working() is true.
2048 : : *
2049 : : * CONTEXT:
2050 : : * spin_lock_irq(pool->lock) which may be released and regrabbed
2051 : : * multiple times. Does GFP_KERNEL allocations.
2052 : : *
2053 : : * Return:
2054 : : * %false if the pool don't need management and the caller can safely start
2055 : : * processing works, %true indicates that the function released pool->lock
2056 : : * and reacquired it to perform some management function and that the
2057 : : * conditions that the caller verified while holding the lock before
2058 : : * calling the function might no longer be true.
2059 : : */
2060 : 0 : static bool manage_workers(struct worker *worker)
2061 : : {
2062 : 1430 : struct worker_pool *pool = worker->pool;
2063 : : bool ret = false;
2064 : :
2065 : : /*
2066 : : * Managership is governed by two mutexes - manager_arb and
2067 : : * manager_mutex. manager_arb handles arbitration of manager role.
2068 : : * Anyone who successfully grabs manager_arb wins the arbitration
2069 : : * and becomes the manager. mutex_trylock() on pool->manager_arb
2070 : : * failure while holding pool->lock reliably indicates that someone
2071 : : * else is managing the pool and the worker which failed trylock
2072 : : * can proceed to executing work items. This means that anyone
2073 : : * grabbing manager_arb is responsible for actually performing
2074 : : * manager duties. If manager_arb is grabbed and released without
2075 : : * actual management, the pool may stall indefinitely.
2076 : : *
2077 : : * manager_mutex is used for exclusion of actual management
2078 : : * operations. The holder of manager_mutex can be sure that none
2079 : : * of management operations, including creation and destruction of
2080 : : * workers, won't take place until the mutex is released. Because
2081 : : * manager_mutex doesn't interfere with manager role arbitration,
2082 : : * it is guaranteed that the pool's management, while may be
2083 : : * delayed, won't be disturbed by someone else grabbing
2084 : : * manager_mutex.
2085 : : */
2086 [ + + ]: 1430 : if (!mutex_trylock(&pool->manager_arb))
2087 : : return ret;
2088 : :
2089 : : /*
2090 : : * With manager arbitration won, manager_mutex would be free in
2091 : : * most cases. trylock first without dropping @pool->lock.
2092 : : */
2093 [ - + ]: 990 : if (unlikely(!mutex_trylock(&pool->manager_mutex))) {
2094 : : spin_unlock_irq(&pool->lock);
2095 : 0 : mutex_lock(&pool->manager_mutex);
2096 : : spin_lock_irq(&pool->lock);
2097 : : ret = true;
2098 : : }
2099 : :
2100 : 990 : pool->flags &= ~POOL_MANAGE_WORKERS;
2101 : :
2102 : : /*
2103 : : * Destroy and then create so that may_start_working() is true
2104 : : * on return.
2105 : : */
2106 : 990 : ret |= maybe_destroy_workers(pool);
2107 : 990 : ret |= maybe_create_worker(pool);
2108 : :
2109 : 990 : mutex_unlock(&pool->manager_mutex);
2110 : 990 : mutex_unlock(&pool->manager_arb);
2111 : : return ret;
2112 : : }
2113 : :
2114 : : /**
2115 : : * process_one_work - process single work
2116 : : * @worker: self
2117 : : * @work: work to process
2118 : : *
2119 : : * Process @work. This function contains all the logics necessary to
2120 : : * process a single work including synchronization against and
2121 : : * interaction with other workers on the same cpu, queueing and
2122 : : * flushing. As long as context requirement is met, any worker can
2123 : : * call this function to process a work.
2124 : : *
2125 : : * CONTEXT:
2126 : : * spin_lock_irq(pool->lock) which is released and regrabbed.
2127 : : */
2128 : 0 : static void process_one_work(struct worker *worker, struct work_struct *work)
2129 : : __releases(&pool->lock)
2130 : : __acquires(&pool->lock)
2131 : : {
2132 : : struct pool_workqueue *pwq = get_work_pwq(work);
2133 : 1693105 : struct worker_pool *pool = worker->pool;
2134 : 1693105 : bool cpu_intensive = pwq->wq->flags & WQ_CPU_INTENSIVE;
2135 : : int work_color;
2136 : : struct worker *collision;
2137 : : #ifdef CONFIG_LOCKDEP
2138 : : /*
2139 : : * It is permissible to free the struct work_struct from
2140 : : * inside the function that is called from it, this we need to
2141 : : * take into account for lockdep too. To avoid bogus "held
2142 : : * lock freed" warnings as well as problems when looking into
2143 : : * work->lockdep_map, make a copy and use that here.
2144 : : */
2145 : : struct lockdep_map lockdep_map;
2146 : :
2147 : : lockdep_copy_map(&lockdep_map, &work->lockdep_map);
2148 : : #endif
2149 : : /*
2150 : : * Ensure we're on the correct CPU. DISASSOCIATED test is
2151 : : * necessary to avoid spurious warnings from rescuers servicing the
2152 : : * unbound or a disassociated pool.
2153 : : */
2154 [ + + ][ + + ]: 1693105 : WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) &&
[ + + ][ - + ]
[ # # ][ # # ]
2155 : : !(pool->flags & POOL_DISASSOCIATED) &&
2156 : : raw_smp_processor_id() != pool->cpu);
2157 : :
2158 : : /*
2159 : : * A single work shouldn't be executed concurrently by
2160 : : * multiple workers on a single cpu. Check whether anyone is
2161 : : * already processing the work. If so, defer the work to the
2162 : : * currently executing one.
2163 : : */
2164 : 1693105 : collision = find_worker_executing_work(pool, work);
2165 [ + + ]: 1691575 : if (unlikely(collision)) {
2166 : 2278 : move_linked_works(work, &collision->scheduled, NULL);
2167 : 2278 : return;
2168 : : }
2169 : :
2170 : : /* claim and dequeue */
2171 : : debug_work_deactivate(work);
2172 : 3378594 : hash_add(pool->busy_hash, &worker->hentry, (unsigned long)work);
2173 : 1689297 : worker->current_work = work;
2174 : 1689297 : worker->current_func = work->func;
2175 : 1689297 : worker->current_pwq = pwq;
2176 : : work_color = get_work_color(work);
2177 : :
2178 : 1689297 : list_del_init(&work->entry);
2179 : :
2180 : : /*
2181 : : * CPU intensive works don't participate in concurrency
2182 : : * management. They're the scheduler's responsibility.
2183 : : */
2184 [ - + ]: 1689297 : if (unlikely(cpu_intensive))
2185 : : worker_set_flags(worker, WORKER_CPU_INTENSIVE, true);
2186 : :
2187 : : /*
2188 : : * Unbound pool isn't concurrency managed and work items should be
2189 : : * executed ASAP. Wake up another worker if necessary.
2190 : : */
2191 [ + + ][ + + ]: 1846568 : if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool))
2192 : 20 : wake_up_worker(pool);
2193 : :
2194 : : /*
2195 : : * Record the last pool and clear PENDING which should be the last
2196 : : * update to @work. Also, do this inside @pool->lock so that
2197 : : * PENDING and queued state changes happen together while IRQ is
2198 : : * disabled.
2199 : : */
2200 : 1689297 : set_work_pool_and_clear_pending(work, pool->id);
2201 : :
2202 : : spin_unlock_irq(&pool->lock);
2203 : :
2204 : : lock_map_acquire_read(&pwq->wq->lockdep_map);
2205 : : lock_map_acquire(&lockdep_map);
2206 : : trace_workqueue_execute_start(work);
2207 : 1691265 : worker->current_func(work);
2208 : : /*
2209 : : * While we must be careful to not use "work" after this, the trace
2210 : : * point will only record its address.
2211 : : */
2212 : : trace_workqueue_execute_end(work);
2213 : : lock_map_release(&lockdep_map);
2214 : : lock_map_release(&pwq->wq->lockdep_map);
2215 : :
2216 [ - + ]: 1691533 : if (unlikely(in_atomic() || lockdep_depth(current) > 0)) {
2217 : 0 : pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n"
2218 : : " last function: %pf\n",
2219 : : current->comm, preempt_count(), task_pid_nr(current),
2220 : : worker->current_func);
2221 : : debug_show_held_locks(current);
2222 : 0 : dump_stack();
2223 : : }
2224 : :
2225 : : /*
2226 : : * The following prevents a kworker from hogging CPU on !PREEMPT
2227 : : * kernels, where a requeueing work item waiting for something to
2228 : : * happen could deadlock with stop_machine as such work item could
2229 : : * indefinitely requeue itself while all other CPUs are trapped in
2230 : : * stop_machine.
2231 : : */
2232 : 1691533 : cond_resched();
2233 : :
2234 : : spin_lock_irq(&pool->lock);
2235 : :
2236 : : /* clear cpu intensive status */
2237 [ - + ]: 1691495 : if (unlikely(cpu_intensive))
2238 : : worker_clr_flags(worker, WORKER_CPU_INTENSIVE);
2239 : :
2240 : : /* we're done with it, release */
2241 : : hash_del(&worker->hentry);
2242 : 1691529 : worker->current_work = NULL;
2243 : 1691529 : worker->current_func = NULL;
2244 : 1691529 : worker->current_pwq = NULL;
2245 : 1691529 : worker->desc_valid = false;
2246 : 1691529 : pwq_dec_nr_in_flight(pwq, work_color);
2247 : : }
2248 : :
2249 : : /**
2250 : : * process_scheduled_works - process scheduled works
2251 : : * @worker: self
2252 : : *
2253 : : * Process all scheduled works. Please note that the scheduled list
2254 : : * may change while processing a work, so this function repeatedly
2255 : : * fetches a work from the top and executes it.
2256 : : *
2257 : : * CONTEXT:
2258 : : * spin_lock_irq(pool->lock) which may be released and regrabbed
2259 : : * multiple times.
2260 : : */
2261 : : static void process_scheduled_works(struct worker *worker)
2262 : : {
2263 [ # # + ]: 0 : while (!list_empty(&worker->scheduled)) {
[ + + ]
2264 : 1022 : struct work_struct *work = list_first_entry(&worker->scheduled,
2265 : : struct work_struct, entry);
2266 : 1022 : process_one_work(worker, work);
2267 : : }
2268 : : }
2269 : :
2270 : : /**
2271 : : * worker_thread - the worker thread function
2272 : : * @__worker: self
2273 : : *
2274 : : * The worker thread function. All workers belong to a worker_pool -
2275 : : * either a per-cpu one or dynamic unbound one. These workers process all
2276 : : * work items regardless of their specific target workqueue. The only
2277 : : * exception is work items which belong to workqueues with a rescuer which
2278 : : * will be explained in rescuer_thread().
2279 : : *
2280 : : * Return: 0
2281 : : */
2282 : 0 : static int worker_thread(void *__worker)
2283 : : {
2284 : 1430 : struct worker *worker = __worker;
2285 : 1164674 : struct worker_pool *pool = worker->pool;
2286 : :
2287 : : /* tell the scheduler that this is a workqueue worker */
2288 : 495 : worker->task->flags |= PF_WQ_WORKER;
2289 : : woke_up:
2290 : : spin_lock_irq(&pool->lock);
2291 : :
2292 : : /* am I supposed to die? */
2293 [ + + ]: 1166045 : if (unlikely(worker->flags & WORKER_DIE)) {
2294 : : spin_unlock_irq(&pool->lock);
2295 [ - + ][ # # ]: 496 : WARN_ON_ONCE(!list_empty(&worker->entry));
[ # # ]
2296 : 496 : worker->task->flags &= ~PF_WQ_WORKER;
2297 : 496 : return 0;
2298 : : }
2299 : :
2300 : 1166538 : worker_leave_idle(worker);
2301 : : recheck:
2302 : : /* no more worker necessary? */
2303 [ + + ]: 1166047 : if (!need_more_worker(pool))
2304 : : goto sleep;
2305 : :
2306 : : /* do we need to manage? */
2307 [ + + ][ + + ]: 1164179 : if (unlikely(!may_start_working(pool)) && manage_workers(worker))
2308 : : goto recheck;
2309 : :
2310 : : /*
2311 : : * ->scheduled list can only be filled while a worker is
2312 : : * preparing to process a work or actually processing it.
2313 : : * Make sure nobody diddled with it while I was sleeping.
2314 : : */
2315 [ - + ][ # # ]: 1163684 : WARN_ON_ONCE(!list_empty(&worker->scheduled));
[ - ]
2316 : :
2317 : : /*
2318 : : * Finish PREP stage. We're guaranteed to have at least one idle
2319 : : * worker or that someone else has already assumed the manager
2320 : : * role. This is where @worker starts participating in concurrency
2321 : : * management if applicable and concurrency management is restored
2322 : : * after being rebound. See rebind_workers() for details.
2323 : : */
2324 : : worker_clr_flags(worker, WORKER_PREP | WORKER_REBOUND);
2325 : :
2326 : : do {
2327 : 1688184 : struct work_struct *work =
2328 : 1688184 : list_first_entry(&pool->worklist,
2329 : : struct work_struct, entry);
2330 : :
2331 [ + + ]: 1688184 : if (likely(!(*work_data_bits(work) & WORK_STRUCT_LINKED))) {
2332 : : /* optimization path, not strictly necessary */
2333 : 1688136 : process_one_work(worker, work);
2334 [ + + ]: 1692288 : if (unlikely(!list_empty(&worker->scheduled)))
2335 : : process_scheduled_works(worker);
2336 : : } else {
2337 : 48 : move_linked_works(work, &worker->scheduled, NULL);
2338 : : process_scheduled_works(worker);
2339 : : }
2340 [ + + ]: 1688692 : } while (keep_working(pool));
2341 : :
2342 : : worker_set_flags(worker, WORKER_PREP, false);
2343 : : sleep:
2344 [ + + ][ + + ]: 1166072 : if (unlikely(need_to_manage_workers(pool)) && manage_workers(worker))
2345 : : goto recheck;
2346 : :
2347 : : /*
2348 : : * pool->lock is held and there's no work to process and no need to
2349 : : * manage, sleep. Workers are woken up only while holding
2350 : : * pool->lock or from local cpu, so setting the current state
2351 : : * before releasing pool->lock is enough to prevent losing any
2352 : : * event.
2353 : : */
2354 : 1165561 : worker_enter_idle(worker);
2355 : 1165604 : __set_current_state(TASK_INTERRUPTIBLE);
2356 : : spin_unlock_irq(&pool->lock);
2357 : 1165636 : schedule();
2358 : 1165573 : goto woke_up;
2359 : : }
2360 : :
2361 : : /**
2362 : : * rescuer_thread - the rescuer thread function
2363 : : * @__rescuer: self
2364 : : *
2365 : : * Workqueue rescuer thread function. There's one rescuer for each
2366 : : * workqueue which has WQ_MEM_RECLAIM set.
2367 : : *
2368 : : * Regular work processing on a pool may block trying to create a new
2369 : : * worker which uses GFP_KERNEL allocation which has slight chance of
2370 : : * developing into deadlock if some works currently on the same queue
2371 : : * need to be processed to satisfy the GFP_KERNEL allocation. This is
2372 : : * the problem rescuer solves.
2373 : : *
2374 : : * When such condition is possible, the pool summons rescuers of all
2375 : : * workqueues which have works queued on the pool and let them process
2376 : : * those works so that forward progress can be guaranteed.
2377 : : *
2378 : : * This should happen rarely.
2379 : : *
2380 : : * Return: 0
2381 : : */
2382 : 0 : static int rescuer_thread(void *__rescuer)
2383 : : {
2384 : : struct worker *rescuer = __rescuer;
2385 : 0 : struct workqueue_struct *wq = rescuer->rescue_wq;
2386 : 0 : struct list_head *scheduled = &rescuer->scheduled;
2387 : :
2388 : 0 : set_user_nice(current, RESCUER_NICE_LEVEL);
2389 : :
2390 : : /*
2391 : : * Mark rescuer as worker too. As WORKER_PREP is never cleared, it
2392 : : * doesn't participate in concurrency management.
2393 : : */
2394 : 0 : rescuer->task->flags |= PF_WQ_WORKER;
2395 : : repeat:
2396 : 0 : set_current_state(TASK_INTERRUPTIBLE);
2397 : :
2398 [ # # ]: 0 : if (kthread_should_stop()) {
2399 : 0 : __set_current_state(TASK_RUNNING);
2400 : 0 : rescuer->task->flags &= ~PF_WQ_WORKER;
2401 : 0 : return 0;
2402 : : }
2403 : :
2404 : : /* see whether any pwq is asking for help */
2405 : : spin_lock_irq(&wq_mayday_lock);
2406 : :
2407 [ # # ]: 0 : while (!list_empty(&wq->maydays)) {
2408 : 0 : struct pool_workqueue *pwq = list_first_entry(&wq->maydays,
2409 : : struct pool_workqueue, mayday_node);
2410 : 0 : struct worker_pool *pool = pwq->pool;
2411 : : struct work_struct *work, *n;
2412 : :
2413 : 0 : __set_current_state(TASK_RUNNING);
2414 : 0 : list_del_init(&pwq->mayday_node);
2415 : :
2416 : : spin_unlock_irq(&wq_mayday_lock);
2417 : :
2418 : : /* migrate to the target cpu if possible */
2419 : 0 : worker_maybe_bind_and_lock(pool);
2420 : 0 : rescuer->pool = pool;
2421 : :
2422 : : /*
2423 : : * Slurp in all works issued via this workqueue and
2424 : : * process'em.
2425 : : */
2426 [ # # ][ # # ]: 0 : WARN_ON_ONCE(!list_empty(&rescuer->scheduled));
[ # # ]
2427 [ # # ]: 0 : list_for_each_entry_safe(work, n, &pool->worklist, entry)
2428 [ # # ]: 0 : if (get_work_pwq(work) == pwq)
2429 : 0 : move_linked_works(work, scheduled, &n);
2430 : :
2431 : : process_scheduled_works(rescuer);
2432 : :
2433 : : /*
2434 : : * Leave this pool. If keep_working() is %true, notify a
2435 : : * regular worker; otherwise, we end up with 0 concurrency
2436 : : * and stalling the execution.
2437 : : */
2438 [ # # ]: 0 : if (keep_working(pool))
2439 : 0 : wake_up_worker(pool);
2440 : :
2441 : 0 : rescuer->pool = NULL;
2442 : : spin_unlock(&pool->lock);
2443 : : spin_lock(&wq_mayday_lock);
2444 : : }
2445 : :
2446 : : spin_unlock_irq(&wq_mayday_lock);
2447 : :
2448 : : /* rescuers should never participate in concurrency management */
2449 [ # # ][ # # ]: 0 : WARN_ON_ONCE(!(rescuer->flags & WORKER_NOT_RUNNING));
[ # # ]
2450 : 0 : schedule();
2451 : 0 : goto repeat;
2452 : : }
2453 : :
2454 : : struct wq_barrier {
2455 : : struct work_struct work;
2456 : : struct completion done;
2457 : : };
2458 : :
2459 : 0 : static void wq_barrier_func(struct work_struct *work)
2460 : : {
2461 : : struct wq_barrier *barr = container_of(work, struct wq_barrier, work);
2462 : 56 : complete(&barr->done);
2463 : 56 : }
2464 : :
2465 : : /**
2466 : : * insert_wq_barrier - insert a barrier work
2467 : : * @pwq: pwq to insert barrier into
2468 : : * @barr: wq_barrier to insert
2469 : : * @target: target work to attach @barr to
2470 : : * @worker: worker currently executing @target, NULL if @target is not executing
2471 : : *
2472 : : * @barr is linked to @target such that @barr is completed only after
2473 : : * @target finishes execution. Please note that the ordering
2474 : : * guarantee is observed only with respect to @target and on the local
2475 : : * cpu.
2476 : : *
2477 : : * Currently, a queued barrier can't be canceled. This is because
2478 : : * try_to_grab_pending() can't determine whether the work to be
2479 : : * grabbed is at the head of the queue and thus can't clear LINKED
2480 : : * flag of the previous work while there must be a valid next work
2481 : : * after a work with LINKED flag set.
2482 : : *
2483 : : * Note that when @worker is non-NULL, @target may be modified
2484 : : * underneath us, so we can't reliably determine pwq from @target.
2485 : : *
2486 : : * CONTEXT:
2487 : : * spin_lock_irq(pool->lock).
2488 : : */
2489 : 0 : static void insert_wq_barrier(struct pool_workqueue *pwq,
2490 : : struct wq_barrier *barr,
2491 : : struct work_struct *target, struct worker *worker)
2492 : : {
2493 : : struct list_head *head;
2494 : : unsigned int linked = 0;
2495 : :
2496 : : /*
2497 : : * debugobject calls are safe here even with pool->lock locked
2498 : : * as we know for sure that this will not trigger any of the
2499 : : * checks and call back into the fixup functions where we
2500 : : * might deadlock.
2501 : : */
2502 : 112 : INIT_WORK_ONSTACK(&barr->work, wq_barrier_func);
2503 : : __set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(&barr->work));
2504 : : init_completion(&barr->done);
2505 : :
2506 : : /*
2507 : : * If @target is currently being executed, schedule the
2508 : : * barrier to the worker; otherwise, put it after @target.
2509 : : */
2510 [ + + ]: 112 : if (worker)
2511 : 8 : head = worker->scheduled.next;
2512 : : else {
2513 : : unsigned long *bits = work_data_bits(target);
2514 : :
2515 : 48 : head = target->entry.next;
2516 : : /* there can already be other linked works, inherit and set */
2517 : 48 : linked = *bits & WORK_STRUCT_LINKED;
2518 : : __set_bit(WORK_STRUCT_LINKED_BIT, bits);
2519 : : }
2520 : :
2521 : : debug_work_activate(&barr->work);
2522 : 56 : insert_work(pwq, &barr->work, head,
2523 : : work_color_to_flags(WORK_NO_COLOR) | linked);
2524 : 56 : }
2525 : :
2526 : : /**
2527 : : * flush_workqueue_prep_pwqs - prepare pwqs for workqueue flushing
2528 : : * @wq: workqueue being flushed
2529 : : * @flush_color: new flush color, < 0 for no-op
2530 : : * @work_color: new work color, < 0 for no-op
2531 : : *
2532 : : * Prepare pwqs for workqueue flushing.
2533 : : *
2534 : : * If @flush_color is non-negative, flush_color on all pwqs should be
2535 : : * -1. If no pwq has in-flight commands at the specified color, all
2536 : : * pwq->flush_color's stay at -1 and %false is returned. If any pwq
2537 : : * has in flight commands, its pwq->flush_color is set to
2538 : : * @flush_color, @wq->nr_pwqs_to_flush is updated accordingly, pwq
2539 : : * wakeup logic is armed and %true is returned.
2540 : : *
2541 : : * The caller should have initialized @wq->first_flusher prior to
2542 : : * calling this function with non-negative @flush_color. If
2543 : : * @flush_color is negative, no flush color update is done and %false
2544 : : * is returned.
2545 : : *
2546 : : * If @work_color is non-negative, all pwqs should have the same
2547 : : * work_color which is previous to @work_color and all will be
2548 : : * advanced to @work_color.
2549 : : *
2550 : : * CONTEXT:
2551 : : * mutex_lock(wq->mutex).
2552 : : *
2553 : : * Return:
2554 : : * %true if @flush_color >= 0 and there's something to flush. %false
2555 : : * otherwise.
2556 : : */
2557 : 0 : static bool flush_workqueue_prep_pwqs(struct workqueue_struct *wq,
2558 : : int flush_color, int work_color)
2559 : : {
2560 : : bool wait = false;
2561 : : struct pool_workqueue *pwq;
2562 : :
2563 [ + - ]: 5972 : if (flush_color >= 0) {
2564 [ - + ][ # # ]: 5972 : WARN_ON_ONCE(atomic_read(&wq->nr_pwqs_to_flush));
[ # # ]
2565 : 5972 : atomic_set(&wq->nr_pwqs_to_flush, 1);
2566 : : }
2567 : :
2568 [ + + ]: 11944 : for_each_pwq(pwq, wq) {
2569 : 5972 : struct worker_pool *pool = pwq->pool;
2570 : :
2571 : : spin_lock_irq(&pool->lock);
2572 : :
2573 [ + - ]: 5972 : if (flush_color >= 0) {
2574 [ - + ][ # # ]: 5972 : WARN_ON_ONCE(pwq->flush_color != -1);
[ # # ]
2575 : :
2576 [ - + ]: 5972 : if (pwq->nr_in_flight[flush_color]) {
2577 : 0 : pwq->flush_color = flush_color;
2578 : 0 : atomic_inc(&wq->nr_pwqs_to_flush);
2579 : : wait = true;
2580 : : }
2581 : : }
2582 : :
2583 [ + - ]: 5972 : if (work_color >= 0) {
2584 [ - + ][ # # ]: 5972 : WARN_ON_ONCE(work_color != work_next_color(pwq->work_color));
[ # # ]
2585 : 5972 : pwq->work_color = work_color;
2586 : : }
2587 : :
2588 : : spin_unlock_irq(&pool->lock);
2589 : : }
2590 : :
2591 [ + - + - ]: 11944 : if (flush_color >= 0 && atomic_dec_and_test(&wq->nr_pwqs_to_flush))
2592 : 5972 : complete(&wq->first_flusher->done);
2593 : :
2594 : 5972 : return wait;
2595 : : }
2596 : :
2597 : : /**
2598 : : * flush_workqueue - ensure that any scheduled work has run to completion.
2599 : : * @wq: workqueue to flush
2600 : : *
2601 : : * This function sleeps until all work items which were queued on entry
2602 : : * have finished execution, but it is not livelocked by new incoming ones.
2603 : : */
2604 : 0 : void flush_workqueue(struct workqueue_struct *wq)
2605 : : {
2606 : 11909 : struct wq_flusher this_flusher = {
2607 : : .list = LIST_HEAD_INIT(this_flusher.list),
2608 : : .flush_color = -1,
2609 : 5954 : .done = COMPLETION_INITIALIZER_ONSTACK(this_flusher.done),
2610 : : };
2611 : : int next_color;
2612 : :
2613 : : lock_map_acquire(&wq->lockdep_map);
2614 : : lock_map_release(&wq->lockdep_map);
2615 : :
2616 : 5954 : mutex_lock(&wq->mutex);
2617 : :
2618 : : /*
2619 : : * Start-to-wait phase
2620 : : */
2621 : 5972 : next_color = work_next_color(wq->work_color);
2622 : :
2623 [ + - ]: 5972 : if (next_color != wq->flush_color) {
2624 : : /*
2625 : : * Color space is not full. The current work_color
2626 : : * becomes our flush_color and work_color is advanced
2627 : : * by one.
2628 : : */
2629 [ - + ][ # # ]: 5972 : WARN_ON_ONCE(!list_empty(&wq->flusher_overflow));
[ # # ]
2630 : 5972 : this_flusher.flush_color = wq->work_color;
2631 : 5972 : wq->work_color = next_color;
2632 : :
2633 [ + - ]: 5972 : if (!wq->first_flusher) {
2634 : : /* no flush in progress, become the first flusher */
2635 [ - + ][ # # ]: 5972 : WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color);
[ # # ]
2636 : :
2637 : 5972 : wq->first_flusher = &this_flusher;
2638 : :
2639 [ + - ]: 5972 : if (!flush_workqueue_prep_pwqs(wq, wq->flush_color,
2640 : : wq->work_color)) {
2641 : : /* nothing to flush, done */
2642 : 5972 : wq->flush_color = next_color;
2643 : 5972 : wq->first_flusher = NULL;
2644 : 5972 : goto out_unlock;
2645 : : }
2646 : : } else {
2647 : : /* wait in queue */
2648 [ # # ][ # # ]: 0 : WARN_ON_ONCE(wq->flush_color == this_flusher.flush_color);
[ # # ]
2649 : 0 : list_add_tail(&this_flusher.list, &wq->flusher_queue);
2650 : 0 : flush_workqueue_prep_pwqs(wq, -1, wq->work_color);
2651 : : }
2652 : : } else {
2653 : : /*
2654 : : * Oops, color space is full, wait on overflow queue.
2655 : : * The next flush completion will assign us
2656 : : * flush_color and transfer to flusher_queue.
2657 : : */
2658 : 0 : list_add_tail(&this_flusher.list, &wq->flusher_overflow);
2659 : : }
2660 : :
2661 : 0 : mutex_unlock(&wq->mutex);
2662 : :
2663 : 0 : wait_for_completion(&this_flusher.done);
2664 : :
2665 : : /*
2666 : : * Wake-up-and-cascade phase
2667 : : *
2668 : : * First flushers are responsible for cascading flushes and
2669 : : * handling overflow. Non-first flushers can simply return.
2670 : : */
2671 [ # # ]: 0 : if (wq->first_flusher != &this_flusher)
2672 : 0 : return;
2673 : :
2674 : 0 : mutex_lock(&wq->mutex);
2675 : :
2676 : : /* we might have raced, check again with mutex held */
2677 [ # # ]: 0 : if (wq->first_flusher != &this_flusher)
2678 : : goto out_unlock;
2679 : :
2680 : 0 : wq->first_flusher = NULL;
2681 : :
2682 [ # # ][ # # ]: 0 : WARN_ON_ONCE(!list_empty(&this_flusher.list));
[ # # ]
2683 [ # # ][ # # ]: 0 : WARN_ON_ONCE(wq->flush_color != this_flusher.flush_color);
[ # # ]
2684 : :
2685 : : while (true) {
2686 : : struct wq_flusher *next, *tmp;
2687 : :
2688 : : /* complete all the flushers sharing the current flush color */
2689 [ # # ]: 0 : list_for_each_entry_safe(next, tmp, &wq->flusher_queue, list) {
2690 [ # # ]: 0 : if (next->flush_color != wq->flush_color)
2691 : : break;
2692 : : list_del_init(&next->list);
2693 : 0 : complete(&next->done);
2694 : : }
2695 : :
2696 [ # # ][ # # ]: 0 : WARN_ON_ONCE(!list_empty(&wq->flusher_overflow) &&
[ # # ][ # # ]
[ # # ]
2697 : : wq->flush_color != work_next_color(wq->work_color));
2698 : :
2699 : : /* this flush_color is finished, advance by one */
2700 : 0 : wq->flush_color = work_next_color(wq->flush_color);
2701 : :
2702 : : /* one color has been freed, handle overflow queue */
2703 [ # # ]: 0 : if (!list_empty(&wq->flusher_overflow)) {
2704 : : /*
2705 : : * Assign the same color to all overflowed
2706 : : * flushers, advance work_color and append to
2707 : : * flusher_queue. This is the start-to-wait
2708 : : * phase for these overflowed flushers.
2709 : : */
2710 [ # # ]: 0 : list_for_each_entry(tmp, &wq->flusher_overflow, list)
2711 : 0 : tmp->flush_color = wq->work_color;
2712 : :
2713 : 0 : wq->work_color = work_next_color(wq->work_color);
2714 : :
2715 : : list_splice_tail_init(&wq->flusher_overflow,
2716 : : &wq->flusher_queue);
2717 : 0 : flush_workqueue_prep_pwqs(wq, -1, wq->work_color);
2718 : : }
2719 : :
2720 [ # # ]: 0 : if (list_empty(&wq->flusher_queue)) {
2721 [ # # ][ # # ]: 0 : WARN_ON_ONCE(wq->flush_color != wq->work_color);
[ # # ]
2722 : : break;
2723 : : }
2724 : :
2725 : : /*
2726 : : * Need to flush more colors. Make the next flusher
2727 : : * the new first flusher and arm pwqs.
2728 : : */
2729 [ # # ][ # # ]: 0 : WARN_ON_ONCE(wq->flush_color == wq->work_color);
[ # # ]
2730 [ # # ][ # # ]: 0 : WARN_ON_ONCE(wq->flush_color != next->flush_color);
[ # # ]
2731 : :
2732 : : list_del_init(&next->list);
2733 : 0 : wq->first_flusher = next;
2734 : :
2735 [ # # ]: 0 : if (flush_workqueue_prep_pwqs(wq, wq->flush_color, -1))
2736 : : break;
2737 : :
2738 : : /*
2739 : : * Meh... this color is already done, clear first
2740 : : * flusher and repeat cascading.
2741 : : */
2742 : 0 : wq->first_flusher = NULL;
2743 : 0 : }
2744 : :
2745 : : out_unlock:
2746 : 5972 : mutex_unlock(&wq->mutex);
2747 : : }
2748 : : EXPORT_SYMBOL_GPL(flush_workqueue);
2749 : :
2750 : : /**
2751 : : * drain_workqueue - drain a workqueue
2752 : : * @wq: workqueue to drain
2753 : : *
2754 : : * Wait until the workqueue becomes empty. While draining is in progress,
2755 : : * only chain queueing is allowed. IOW, only currently pending or running
2756 : : * work items on @wq can queue further work items on it. @wq is flushed
2757 : : * repeatedly until it becomes empty. The number of flushing is detemined
2758 : : * by the depth of chaining and should be relatively short. Whine if it
2759 : : * takes too long.
2760 : : */
2761 : 0 : void drain_workqueue(struct workqueue_struct *wq)
2762 : : {
2763 : : unsigned int flush_cnt = 0;
2764 : : struct pool_workqueue *pwq;
2765 : :
2766 : : /*
2767 : : * __queue_work() needs to test whether there are drainers, is much
2768 : : * hotter than drain_workqueue() and already looks at @wq->flags.
2769 : : * Use __WQ_DRAINING so that queue doesn't have to check nr_drainers.
2770 : : */
2771 : 0 : mutex_lock(&wq->mutex);
2772 [ # # ]: 0 : if (!wq->nr_drainers++)
2773 : 0 : wq->flags |= __WQ_DRAINING;
2774 : 0 : mutex_unlock(&wq->mutex);
2775 : : reflush:
2776 : 0 : flush_workqueue(wq);
2777 : :
2778 : 0 : mutex_lock(&wq->mutex);
2779 : :
2780 [ # # ]: 0 : for_each_pwq(pwq, wq) {
2781 : : bool drained;
2782 : :
2783 : 0 : spin_lock_irq(&pwq->pool->lock);
2784 [ # # ][ # # ]: 0 : drained = !pwq->nr_active && list_empty(&pwq->delayed_works);
2785 : 0 : spin_unlock_irq(&pwq->pool->lock);
2786 : :
2787 [ # # ]: 0 : if (drained)
2788 : 0 : continue;
2789 : :
2790 [ # # ][ # # ]: 0 : if (++flush_cnt == 10 ||
2791 [ # # ]: 0 : (flush_cnt % 100 == 0 && flush_cnt <= 1000))
2792 : 0 : pr_warn("workqueue %s: drain_workqueue() isn't complete after %u tries\n",
2793 : : wq->name, flush_cnt);
2794 : :
2795 : 0 : mutex_unlock(&wq->mutex);
2796 : 0 : goto reflush;
2797 : : }
2798 : :
2799 [ # # ]: 0 : if (!--wq->nr_drainers)
2800 : 0 : wq->flags &= ~__WQ_DRAINING;
2801 : 0 : mutex_unlock(&wq->mutex);
2802 : 0 : }
2803 : : EXPORT_SYMBOL_GPL(drain_workqueue);
2804 : :
2805 : 0 : static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr)
2806 : : {
2807 : : struct worker *worker = NULL;
2808 : : struct worker_pool *pool;
2809 : : struct pool_workqueue *pwq;
2810 : :
2811 : : might_sleep();
2812 : :
2813 : : local_irq_disable();
2814 : 2336 : pool = get_work_pool(work);
2815 [ + + ]: 2336 : if (!pool) {
2816 : : local_irq_enable();
2817 : 1658 : return false;
2818 : : }
2819 : :
2820 : : spin_lock(&pool->lock);
2821 : : /* see the comment in try_to_grab_pending() with the same code */
2822 : : pwq = get_work_pwq(work);
2823 [ + + ]: 678 : if (pwq) {
2824 [ + - ]: 48 : if (unlikely(pwq->pool != pool))
2825 : : goto already_gone;
2826 : : } else {
2827 : 630 : worker = find_worker_executing_work(pool, work);
2828 [ + + ]: 630 : if (!worker)
2829 : : goto already_gone;
2830 : 8 : pwq = worker->current_pwq;
2831 : : }
2832 : :
2833 : 56 : insert_wq_barrier(pwq, barr, work, worker);
2834 : : spin_unlock_irq(&pool->lock);
2835 : :
2836 : : /*
2837 : : * If @max_active is 1 or rescuer is in use, flushing another work
2838 : : * item on the same workqueue may lead to deadlock. Make sure the
2839 : : * flusher is not running on the same workqueue by verifying write
2840 : : * access.
2841 : : */
2842 : : if (pwq->wq->saved_max_active == 1 || pwq->wq->rescuer)
2843 : : lock_map_acquire(&pwq->wq->lockdep_map);
2844 : : else
2845 : : lock_map_acquire_read(&pwq->wq->lockdep_map);
2846 : : lock_map_release(&pwq->wq->lockdep_map);
2847 : :
2848 : 56 : return true;
2849 : : already_gone:
2850 : : spin_unlock_irq(&pool->lock);
2851 : 622 : return false;
2852 : : }
2853 : :
2854 : : /**
2855 : : * flush_work - wait for a work to finish executing the last queueing instance
2856 : : * @work: the work to flush
2857 : : *
2858 : : * Wait until @work has finished execution. @work is guaranteed to be idle
2859 : : * on return if it hasn't been requeued since flush started.
2860 : : *
2861 : : * Return:
2862 : : * %true if flush_work() waited for the work to finish execution,
2863 : : * %false if it was already idle.
2864 : : */
2865 : 0 : bool flush_work(struct work_struct *work)
2866 : : {
2867 : : struct wq_barrier barr;
2868 : :
2869 : : lock_map_acquire(&work->lockdep_map);
2870 : : lock_map_release(&work->lockdep_map);
2871 : :
2872 [ + + ]: 2336 : if (start_flush_work(work, &barr)) {
2873 : 56 : wait_for_completion(&barr.done);
2874 : : destroy_work_on_stack(&barr.work);
2875 : 56 : return true;
2876 : : } else {
2877 : : return false;
2878 : : }
2879 : : }
2880 : : EXPORT_SYMBOL_GPL(flush_work);
2881 : :
2882 : 1188 : static bool __cancel_work_timer(struct work_struct *work, bool is_dwork)
2883 : : {
2884 : : unsigned long flags;
2885 : : int ret;
2886 : :
2887 : : do {
2888 : 1188 : ret = try_to_grab_pending(work, is_dwork, &flags);
2889 : : /*
2890 : : * If someone else is canceling, wait for the same event it
2891 : : * would be waiting for before retrying.
2892 : : */
2893 [ - + ]: 1188 : if (unlikely(ret == -ENOENT))
2894 : 0 : flush_work(work);
2895 [ - + ]: 1188 : } while (unlikely(ret < 0));
2896 : :
2897 : : /* tell other tasks trying to grab @work to back off */
2898 : 1188 : mark_work_canceling(work);
2899 [ - + ]: 1188 : local_irq_restore(flags);
2900 : :
2901 : 1188 : flush_work(work);
2902 : 1188 : clear_work_data(work);
2903 : 1188 : return ret;
2904 : : }
2905 : :
2906 : : /**
2907 : : * cancel_work_sync - cancel a work and wait for it to finish
2908 : : * @work: the work to cancel
2909 : : *
2910 : : * Cancel @work and wait for its execution to finish. This function
2911 : : * can be used even if the work re-queues itself or migrates to
2912 : : * another workqueue. On return from this function, @work is
2913 : : * guaranteed to be not pending or executing on any CPU.
2914 : : *
2915 : : * cancel_work_sync(&delayed_work->work) must not be used for
2916 : : * delayed_work's. Use cancel_delayed_work_sync() instead.
2917 : : *
2918 : : * The caller must ensure that the workqueue on which @work was last
2919 : : * queued can't be destroyed before this function returns.
2920 : : *
2921 : : * Return:
2922 : : * %true if @work was pending, %false otherwise.
2923 : : */
2924 : 0 : bool cancel_work_sync(struct work_struct *work)
2925 : : {
2926 : 612 : return __cancel_work_timer(work, false);
2927 : : }
2928 : : EXPORT_SYMBOL_GPL(cancel_work_sync);
2929 : :
2930 : : /**
2931 : : * flush_delayed_work - wait for a dwork to finish executing the last queueing
2932 : : * @dwork: the delayed work to flush
2933 : : *
2934 : : * Delayed timer is cancelled and the pending work is queued for
2935 : : * immediate execution. Like flush_work(), this function only
2936 : : * considers the last queueing instance of @dwork.
2937 : : *
2938 : : * Return:
2939 : : * %true if flush_work() waited for the work to finish execution,
2940 : : * %false if it was already idle.
2941 : : */
2942 : 0 : bool flush_delayed_work(struct delayed_work *dwork)
2943 : : {
2944 : : local_irq_disable();
2945 [ # # ]: 0 : if (del_timer_sync(&dwork->timer))
2946 : 0 : __queue_work(dwork->cpu, dwork->wq, &dwork->work);
2947 : : local_irq_enable();
2948 : 0 : return flush_work(&dwork->work);
2949 : : }
2950 : : EXPORT_SYMBOL(flush_delayed_work);
2951 : :
2952 : : /**
2953 : : * cancel_delayed_work - cancel a delayed work
2954 : : * @dwork: delayed_work to cancel
2955 : : *
2956 : : * Kill off a pending delayed_work.
2957 : : *
2958 : : * Return: %true if @dwork was pending and canceled; %false if it wasn't
2959 : : * pending.
2960 : : *
2961 : : * Note:
2962 : : * The work callback function may still be running on return, unless
2963 : : * it returns %true and the work doesn't re-arm itself. Explicitly flush or
2964 : : * use cancel_delayed_work_sync() to wait on it.
2965 : : *
2966 : : * This function is safe to call from any context including IRQ handler.
2967 : : */
2968 : 0 : bool cancel_delayed_work(struct delayed_work *dwork)
2969 : : {
2970 : : unsigned long flags;
2971 : : int ret;
2972 : :
2973 : : do {
2974 : 0 : ret = try_to_grab_pending(&dwork->work, true, &flags);
2975 [ # # ]: 0 : } while (unlikely(ret == -EAGAIN));
2976 : :
2977 [ # # ]: 0 : if (unlikely(ret < 0))
2978 : : return false;
2979 : :
2980 : 0 : set_work_pool_and_clear_pending(&dwork->work,
2981 : : get_work_pool_id(&dwork->work));
2982 [ # # ]: 0 : local_irq_restore(flags);
2983 : 0 : return ret;
2984 : : }
2985 : : EXPORT_SYMBOL(cancel_delayed_work);
2986 : :
2987 : : /**
2988 : : * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish
2989 : : * @dwork: the delayed work cancel
2990 : : *
2991 : : * This is cancel_work_sync() for delayed works.
2992 : : *
2993 : : * Return:
2994 : : * %true if @dwork was pending, %false otherwise.
2995 : : */
2996 : 0 : bool cancel_delayed_work_sync(struct delayed_work *dwork)
2997 : : {
2998 : 576 : return __cancel_work_timer(&dwork->work, true);
2999 : : }
3000 : : EXPORT_SYMBOL(cancel_delayed_work_sync);
3001 : :
3002 : : /**
3003 : : * schedule_on_each_cpu - execute a function synchronously on each online CPU
3004 : : * @func: the function to call
3005 : : *
3006 : : * schedule_on_each_cpu() executes @func on each online CPU using the
3007 : : * system workqueue and blocks until all CPUs have completed.
3008 : : * schedule_on_each_cpu() is very slow.
3009 : : *
3010 : : * Return:
3011 : : * 0 on success, -errno on failure.
3012 : : */
3013 : 0 : int schedule_on_each_cpu(work_func_t func)
3014 : : {
3015 : : int cpu;
3016 : : struct work_struct __percpu *works;
3017 : :
3018 : 0 : works = alloc_percpu(struct work_struct);
3019 [ # # ]: 0 : if (!works)
3020 : : return -ENOMEM;
3021 : :
3022 : 0 : get_online_cpus();
3023 : :
3024 [ # # ]: 0 : for_each_online_cpu(cpu) {
3025 : 0 : struct work_struct *work = per_cpu_ptr(works, cpu);
3026 : :
3027 : 0 : INIT_WORK(work, func);
3028 : : schedule_work_on(cpu, work);
3029 : : }
3030 : :
3031 [ # # ]: 0 : for_each_online_cpu(cpu)
3032 : 0 : flush_work(per_cpu_ptr(works, cpu));
3033 : :
3034 : 0 : put_online_cpus();
3035 : 0 : free_percpu(works);
3036 : 0 : return 0;
3037 : : }
3038 : :
3039 : : /**
3040 : : * flush_scheduled_work - ensure that any scheduled work has run to completion.
3041 : : *
3042 : : * Forces execution of the kernel-global workqueue and blocks until its
3043 : : * completion.
3044 : : *
3045 : : * Think twice before calling this function! It's very easy to get into
3046 : : * trouble if you don't take great care. Either of the following situations
3047 : : * will lead to deadlock:
3048 : : *
3049 : : * One of the work items currently on the workqueue needs to acquire
3050 : : * a lock held by your code or its caller.
3051 : : *
3052 : : * Your code is running in the context of a work routine.
3053 : : *
3054 : : * They will be detected by lockdep when they occur, but the first might not
3055 : : * occur very often. It depends on what work items are on the workqueue and
3056 : : * what locks they need, which you have no control over.
3057 : : *
3058 : : * In most situations flushing the entire workqueue is overkill; you merely
3059 : : * need to know that a particular work item isn't queued and isn't running.
3060 : : * In such cases you should use cancel_delayed_work_sync() or
3061 : : * cancel_work_sync() instead.
3062 : : */
3063 : 0 : void flush_scheduled_work(void)
3064 : : {
3065 : 0 : flush_workqueue(system_wq);
3066 : 0 : }
3067 : : EXPORT_SYMBOL(flush_scheduled_work);
3068 : :
3069 : : /**
3070 : : * execute_in_process_context - reliably execute the routine with user context
3071 : : * @fn: the function to execute
3072 : : * @ew: guaranteed storage for the execute work structure (must
3073 : : * be available when the work executes)
3074 : : *
3075 : : * Executes the function immediately if process context is available,
3076 : : * otherwise schedules the function for delayed execution.
3077 : : *
3078 : : * Return: 0 - function was executed
3079 : : * 1 - function was scheduled for execution
3080 : : */
3081 : 0 : int execute_in_process_context(work_func_t fn, struct execute_work *ew)
3082 : : {
3083 [ # # ]: 0 : if (!in_interrupt()) {
3084 : 0 : fn(&ew->work);
3085 : 0 : return 0;
3086 : : }
3087 : :
3088 : 0 : INIT_WORK(&ew->work, fn);
3089 : 0 : schedule_work(&ew->work);
3090 : :
3091 : 0 : return 1;
3092 : : }
3093 : : EXPORT_SYMBOL_GPL(execute_in_process_context);
3094 : :
3095 : : #ifdef CONFIG_SYSFS
3096 : : /*
3097 : : * Workqueues with WQ_SYSFS flag set is visible to userland via
3098 : : * /sys/bus/workqueue/devices/WQ_NAME. All visible workqueues have the
3099 : : * following attributes.
3100 : : *
3101 : : * per_cpu RO bool : whether the workqueue is per-cpu or unbound
3102 : : * max_active RW int : maximum number of in-flight work items
3103 : : *
3104 : : * Unbound workqueues have the following extra attributes.
3105 : : *
3106 : : * id RO int : the associated pool ID
3107 : : * nice RW int : nice value of the workers
3108 : : * cpumask RW mask : bitmask of allowed CPUs for the workers
3109 : : */
3110 : : struct wq_device {
3111 : : struct workqueue_struct *wq;
3112 : : struct device dev;
3113 : : };
3114 : :
3115 : : static struct workqueue_struct *dev_to_wq(struct device *dev)
3116 : : {
3117 : : struct wq_device *wq_dev = container_of(dev, struct wq_device, dev);
3118 : :
3119 : 0 : return wq_dev->wq;
3120 : : }
3121 : :
3122 : 0 : static ssize_t per_cpu_show(struct device *dev, struct device_attribute *attr,
3123 : : char *buf)
3124 : : {
3125 : : struct workqueue_struct *wq = dev_to_wq(dev);
3126 : :
3127 : 0 : return scnprintf(buf, PAGE_SIZE, "%d\n", (bool)!(wq->flags & WQ_UNBOUND));
3128 : : }
3129 : : static DEVICE_ATTR_RO(per_cpu);
3130 : :
3131 : 0 : static ssize_t max_active_show(struct device *dev,
3132 : : struct device_attribute *attr, char *buf)
3133 : : {
3134 : : struct workqueue_struct *wq = dev_to_wq(dev);
3135 : :
3136 : 0 : return scnprintf(buf, PAGE_SIZE, "%d\n", wq->saved_max_active);
3137 : : }
3138 : :
3139 : 0 : static ssize_t max_active_store(struct device *dev,
3140 : : struct device_attribute *attr, const char *buf,
3141 : : size_t count)
3142 : : {
3143 : : struct workqueue_struct *wq = dev_to_wq(dev);
3144 : : int val;
3145 : :
3146 [ # # ][ # # ]: 0 : if (sscanf(buf, "%d", &val) != 1 || val <= 0)
3147 : : return -EINVAL;
3148 : :
3149 : 0 : workqueue_set_max_active(wq, val);
3150 : 0 : return count;
3151 : : }
3152 : : static DEVICE_ATTR_RW(max_active);
3153 : :
3154 : : static struct attribute *wq_sysfs_attrs[] = {
3155 : : &dev_attr_per_cpu.attr,
3156 : : &dev_attr_max_active.attr,
3157 : : NULL,
3158 : : };
3159 : : ATTRIBUTE_GROUPS(wq_sysfs);
3160 : :
3161 : 0 : static ssize_t wq_pool_ids_show(struct device *dev,
3162 : : struct device_attribute *attr, char *buf)
3163 : : {
3164 : : struct workqueue_struct *wq = dev_to_wq(dev);
3165 : : const char *delim = "";
3166 : : int node, written = 0;
3167 : :
3168 : : rcu_read_lock_sched();
3169 [ # # ]: 0 : for_each_node(node) {
3170 : 0 : written += scnprintf(buf + written, PAGE_SIZE - written,
3171 : : "%s%d:%d", delim, node,
3172 : 0 : unbound_pwq_by_node(wq, node)->pool->id);
3173 : : delim = " ";
3174 : : }
3175 : 0 : written += scnprintf(buf + written, PAGE_SIZE - written, "\n");
3176 : : rcu_read_unlock_sched();
3177 : :
3178 : 0 : return written;
3179 : : }
3180 : :
3181 : 0 : static ssize_t wq_nice_show(struct device *dev, struct device_attribute *attr,
3182 : : char *buf)
3183 : : {
3184 : : struct workqueue_struct *wq = dev_to_wq(dev);
3185 : : int written;
3186 : :
3187 : 0 : mutex_lock(&wq->mutex);
3188 : 0 : written = scnprintf(buf, PAGE_SIZE, "%d\n", wq->unbound_attrs->nice);
3189 : 0 : mutex_unlock(&wq->mutex);
3190 : :
3191 : 0 : return written;
3192 : : }
3193 : :
3194 : : /* prepare workqueue_attrs for sysfs store operations */
3195 : 0 : static struct workqueue_attrs *wq_sysfs_prep_attrs(struct workqueue_struct *wq)
3196 : : {
3197 : : struct workqueue_attrs *attrs;
3198 : :
3199 : 0 : attrs = alloc_workqueue_attrs(GFP_KERNEL);
3200 [ # # ]: 0 : if (!attrs)
3201 : : return NULL;
3202 : :
3203 : 0 : mutex_lock(&wq->mutex);
3204 : 0 : copy_workqueue_attrs(attrs, wq->unbound_attrs);
3205 : 0 : mutex_unlock(&wq->mutex);
3206 : 0 : return attrs;
3207 : : }
3208 : :
3209 : 0 : static ssize_t wq_nice_store(struct device *dev, struct device_attribute *attr,
3210 : : const char *buf, size_t count)
3211 : : {
3212 : : struct workqueue_struct *wq = dev_to_wq(dev);
3213 : : struct workqueue_attrs *attrs;
3214 : : int ret;
3215 : :
3216 : 0 : attrs = wq_sysfs_prep_attrs(wq);
3217 [ # # ]: 0 : if (!attrs)
3218 : : return -ENOMEM;
3219 : :
3220 [ # # ][ # # ]: 0 : if (sscanf(buf, "%d", &attrs->nice) == 1 &&
3221 [ # # ]: 0 : attrs->nice >= -20 && attrs->nice <= 19)
3222 : 0 : ret = apply_workqueue_attrs(wq, attrs);
3223 : : else
3224 : : ret = -EINVAL;
3225 : :
3226 : : free_workqueue_attrs(attrs);
3227 [ # # ]: 0 : return ret ?: count;
3228 : : }
3229 : :
3230 : 0 : static ssize_t wq_cpumask_show(struct device *dev,
3231 : : struct device_attribute *attr, char *buf)
3232 : : {
3233 : : struct workqueue_struct *wq = dev_to_wq(dev);
3234 : : int written;
3235 : :
3236 : 0 : mutex_lock(&wq->mutex);
3237 : 0 : written = cpumask_scnprintf(buf, PAGE_SIZE, wq->unbound_attrs->cpumask);
3238 : 0 : mutex_unlock(&wq->mutex);
3239 : :
3240 : 0 : written += scnprintf(buf + written, PAGE_SIZE - written, "\n");
3241 : 0 : return written;
3242 : : }
3243 : :
3244 : 0 : static ssize_t wq_cpumask_store(struct device *dev,
3245 : : struct device_attribute *attr,
3246 : : const char *buf, size_t count)
3247 : : {
3248 : : struct workqueue_struct *wq = dev_to_wq(dev);
3249 : : struct workqueue_attrs *attrs;
3250 : : int ret;
3251 : :
3252 : 0 : attrs = wq_sysfs_prep_attrs(wq);
3253 [ # # ]: 0 : if (!attrs)
3254 : : return -ENOMEM;
3255 : :
3256 : : ret = cpumask_parse(buf, attrs->cpumask);
3257 [ # # ]: 0 : if (!ret)
3258 : 0 : ret = apply_workqueue_attrs(wq, attrs);
3259 : :
3260 : : free_workqueue_attrs(attrs);
3261 [ # # ]: 0 : return ret ?: count;
3262 : : }
3263 : :
3264 : 0 : static ssize_t wq_numa_show(struct device *dev, struct device_attribute *attr,
3265 : : char *buf)
3266 : : {
3267 : : struct workqueue_struct *wq = dev_to_wq(dev);
3268 : : int written;
3269 : :
3270 : 0 : mutex_lock(&wq->mutex);
3271 : 0 : written = scnprintf(buf, PAGE_SIZE, "%d\n",
3272 : 0 : !wq->unbound_attrs->no_numa);
3273 : 0 : mutex_unlock(&wq->mutex);
3274 : :
3275 : 0 : return written;
3276 : : }
3277 : :
3278 : 0 : static ssize_t wq_numa_store(struct device *dev, struct device_attribute *attr,
3279 : : const char *buf, size_t count)
3280 : : {
3281 : : struct workqueue_struct *wq = dev_to_wq(dev);
3282 : : struct workqueue_attrs *attrs;
3283 : : int v, ret;
3284 : :
3285 : 0 : attrs = wq_sysfs_prep_attrs(wq);
3286 [ # # ]: 0 : if (!attrs)
3287 : : return -ENOMEM;
3288 : :
3289 : : ret = -EINVAL;
3290 [ # # ]: 0 : if (sscanf(buf, "%d", &v) == 1) {
3291 : 0 : attrs->no_numa = !v;
3292 : 0 : ret = apply_workqueue_attrs(wq, attrs);
3293 : : }
3294 : :
3295 : : free_workqueue_attrs(attrs);
3296 [ # # ]: 0 : return ret ?: count;
3297 : : }
3298 : :
3299 : : static struct device_attribute wq_sysfs_unbound_attrs[] = {
3300 : : __ATTR(pool_ids, 0444, wq_pool_ids_show, NULL),
3301 : : __ATTR(nice, 0644, wq_nice_show, wq_nice_store),
3302 : : __ATTR(cpumask, 0644, wq_cpumask_show, wq_cpumask_store),
3303 : : __ATTR(numa, 0644, wq_numa_show, wq_numa_store),
3304 : : __ATTR_NULL,
3305 : : };
3306 : :
3307 : : static struct bus_type wq_subsys = {
3308 : : .name = "workqueue",
3309 : : .dev_groups = wq_sysfs_groups,
3310 : : };
3311 : :
3312 : 0 : static int __init wq_sysfs_init(void)
3313 : : {
3314 : 0 : return subsys_virtual_register(&wq_subsys, NULL);
3315 : : }
3316 : : core_initcall(wq_sysfs_init);
3317 : :
3318 : 0 : static void wq_device_release(struct device *dev)
3319 : : {
3320 : 0 : struct wq_device *wq_dev = container_of(dev, struct wq_device, dev);
3321 : :
3322 : 0 : kfree(wq_dev);
3323 : 0 : }
3324 : :
3325 : : /**
3326 : : * workqueue_sysfs_register - make a workqueue visible in sysfs
3327 : : * @wq: the workqueue to register
3328 : : *
3329 : : * Expose @wq in sysfs under /sys/bus/workqueue/devices.
3330 : : * alloc_workqueue*() automatically calls this function if WQ_SYSFS is set
3331 : : * which is the preferred method.
3332 : : *
3333 : : * Workqueue user should use this function directly iff it wants to apply
3334 : : * workqueue_attrs before making the workqueue visible in sysfs; otherwise,
3335 : : * apply_workqueue_attrs() may race against userland updating the
3336 : : * attributes.
3337 : : *
3338 : : * Return: 0 on success, -errno on failure.
3339 : : */
3340 : 0 : int workqueue_sysfs_register(struct workqueue_struct *wq)
3341 : : {
3342 : : struct wq_device *wq_dev;
3343 : : int ret;
3344 : :
3345 : : /*
3346 : : * Adjusting max_active or creating new pwqs by applyting
3347 : : * attributes breaks ordering guarantee. Disallow exposing ordered
3348 : : * workqueues.
3349 : : */
3350 [ # # ][ # # ]: 0 : if (WARN_ON(wq->flags & __WQ_ORDERED))
3351 : : return -EINVAL;
3352 : :
3353 : 0 : wq->wq_dev = wq_dev = kzalloc(sizeof(*wq_dev), GFP_KERNEL);
3354 [ # # ]: 0 : if (!wq_dev)
3355 : : return -ENOMEM;
3356 : :
3357 : 0 : wq_dev->wq = wq;
3358 : 0 : wq_dev->dev.bus = &wq_subsys;
3359 : 0 : wq_dev->dev.init_name = wq->name;
3360 : 0 : wq_dev->dev.release = wq_device_release;
3361 : :
3362 : : /*
3363 : : * unbound_attrs are created separately. Suppress uevent until
3364 : : * everything is ready.
3365 : : */
3366 : : dev_set_uevent_suppress(&wq_dev->dev, true);
3367 : :
3368 : 0 : ret = device_register(&wq_dev->dev);
3369 [ # # ]: 0 : if (ret) {
3370 : 0 : kfree(wq_dev);
3371 : 0 : wq->wq_dev = NULL;
3372 : 0 : return ret;
3373 : : }
3374 : :
3375 [ # # ]: 0 : if (wq->flags & WQ_UNBOUND) {
3376 : : struct device_attribute *attr;
3377 : :
3378 [ # # ]: 0 : for (attr = wq_sysfs_unbound_attrs; attr->attr.name; attr++) {
3379 : 0 : ret = device_create_file(&wq_dev->dev, attr);
3380 [ # # ]: 0 : if (ret) {
3381 : 0 : device_unregister(&wq_dev->dev);
3382 : 0 : wq->wq_dev = NULL;
3383 : 0 : return ret;
3384 : : }
3385 : : }
3386 : : }
3387 : :
3388 : 0 : kobject_uevent(&wq_dev->dev.kobj, KOBJ_ADD);
3389 : 0 : return 0;
3390 : : }
3391 : :
3392 : : /**
3393 : : * workqueue_sysfs_unregister - undo workqueue_sysfs_register()
3394 : : * @wq: the workqueue to unregister
3395 : : *
3396 : : * If @wq is registered to sysfs by workqueue_sysfs_register(), unregister.
3397 : : */
3398 : : static void workqueue_sysfs_unregister(struct workqueue_struct *wq)
3399 : : {
3400 : 0 : struct wq_device *wq_dev = wq->wq_dev;
3401 : :
3402 [ # # ]: 0 : if (!wq->wq_dev)
3403 : : return;
3404 : :
3405 : 0 : wq->wq_dev = NULL;
3406 : 0 : device_unregister(&wq_dev->dev);
3407 : : }
3408 : : #else /* CONFIG_SYSFS */
3409 : : static void workqueue_sysfs_unregister(struct workqueue_struct *wq) { }
3410 : : #endif /* CONFIG_SYSFS */
3411 : :
3412 : : /**
3413 : : * free_workqueue_attrs - free a workqueue_attrs
3414 : : * @attrs: workqueue_attrs to free
3415 : : *
3416 : : * Undo alloc_workqueue_attrs().
3417 : : */
3418 : 0 : void free_workqueue_attrs(struct workqueue_attrs *attrs)
3419 : : {
3420 [ # # ][ # # ]: 0 : if (attrs) {
[ # # ]
[ # # # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ]
3421 : : free_cpumask_var(attrs->cpumask);
3422 : 0 : kfree(attrs);
3423 : : }
3424 : 0 : }
3425 : :
3426 : : /**
3427 : : * alloc_workqueue_attrs - allocate a workqueue_attrs
3428 : : * @gfp_mask: allocation mask to use
3429 : : *
3430 : : * Allocate a new workqueue_attrs, initialize with default settings and
3431 : : * return it.
3432 : : *
3433 : : * Return: The allocated new workqueue_attr on success. %NULL on failure.
3434 : : */
3435 : 0 : struct workqueue_attrs *alloc_workqueue_attrs(gfp_t gfp_mask)
3436 : : {
3437 : : struct workqueue_attrs *attrs;
3438 : :
3439 : : attrs = kzalloc(sizeof(*attrs), gfp_mask);
3440 [ # # ]: 0 : if (!attrs)
3441 : : goto fail;
3442 : : if (!alloc_cpumask_var(&attrs->cpumask, gfp_mask))
3443 : : goto fail;
3444 : :
3445 : 0 : cpumask_copy(attrs->cpumask, cpu_possible_mask);
3446 : 0 : return attrs;
3447 : : fail:
3448 : : free_workqueue_attrs(attrs);
3449 : : return NULL;
3450 : : }
3451 : :
3452 : : static void copy_workqueue_attrs(struct workqueue_attrs *to,
3453 : : const struct workqueue_attrs *from)
3454 : : {
3455 : 0 : to->nice = from->nice;
3456 : : cpumask_copy(to->cpumask, from->cpumask);
3457 : : /*
3458 : : * Unlike hash and equality test, this function doesn't ignore
3459 : : * ->no_numa as it is used for both pool and wq attrs. Instead,
3460 : : * get_unbound_pool() explicitly clears ->no_numa after copying.
3461 : : */
3462 : 0 : to->no_numa = from->no_numa;
3463 : : }
3464 : :
3465 : : /* hash value of the content of @attr */
3466 : 0 : static u32 wqattrs_hash(const struct workqueue_attrs *attrs)
3467 : : {
3468 : : u32 hash = 0;
3469 : :
3470 : 0 : hash = jhash_1word(attrs->nice, hash);
3471 : 0 : hash = jhash(cpumask_bits(attrs->cpumask),
3472 : : BITS_TO_LONGS(nr_cpumask_bits) * sizeof(long), hash);
3473 : 0 : return hash;
3474 : : }
3475 : :
3476 : : /* content equality test */
3477 : : static bool wqattrs_equal(const struct workqueue_attrs *a,
3478 : : const struct workqueue_attrs *b)
3479 : : {
3480 [ # # ]: 0 : if (a->nice != b->nice)
3481 : : return false;
3482 [ # # ]: 0 : if (!cpumask_equal(a->cpumask, b->cpumask))
3483 : : return false;
3484 : : return true;
3485 : : }
3486 : :
3487 : : /**
3488 : : * init_worker_pool - initialize a newly zalloc'd worker_pool
3489 : : * @pool: worker_pool to initialize
3490 : : *
3491 : : * Initiailize a newly zalloc'd @pool. It also allocates @pool->attrs.
3492 : : *
3493 : : * Return: 0 on success, -errno on failure. Even on failure, all fields
3494 : : * inside @pool proper are initialized and put_unbound_pool() can be called
3495 : : * on @pool safely to release it.
3496 : : */
3497 : 0 : static int init_worker_pool(struct worker_pool *pool)
3498 : : {
3499 : 0 : spin_lock_init(&pool->lock);
3500 : 0 : pool->id = -1;
3501 : 0 : pool->cpu = -1;
3502 : 0 : pool->node = NUMA_NO_NODE;
3503 : 0 : pool->flags |= POOL_DISASSOCIATED;
3504 : 0 : INIT_LIST_HEAD(&pool->worklist);
3505 : 0 : INIT_LIST_HEAD(&pool->idle_list);
3506 : 0 : hash_init(pool->busy_hash);
3507 : :
3508 : 0 : init_timer_deferrable(&pool->idle_timer);
3509 : 0 : pool->idle_timer.function = idle_worker_timeout;
3510 : 0 : pool->idle_timer.data = (unsigned long)pool;
3511 : :
3512 : 0 : setup_timer(&pool->mayday_timer, pool_mayday_timeout,
3513 : : (unsigned long)pool);
3514 : :
3515 : 0 : mutex_init(&pool->manager_arb);
3516 : 0 : mutex_init(&pool->manager_mutex);
3517 : 0 : idr_init(&pool->worker_idr);
3518 : :
3519 : : INIT_HLIST_NODE(&pool->hash_node);
3520 : 0 : pool->refcnt = 1;
3521 : :
3522 : : /* shouldn't fail above this point */
3523 : 0 : pool->attrs = alloc_workqueue_attrs(GFP_KERNEL);
3524 [ # # ]: 0 : if (!pool->attrs)
3525 : : return -ENOMEM;
3526 : 0 : return 0;
3527 : : }
3528 : :
3529 : 0 : static void rcu_free_pool(struct rcu_head *rcu)
3530 : : {
3531 : 0 : struct worker_pool *pool = container_of(rcu, struct worker_pool, rcu);
3532 : :
3533 : 0 : idr_destroy(&pool->worker_idr);
3534 : 0 : free_workqueue_attrs(pool->attrs);
3535 : 0 : kfree(pool);
3536 : 0 : }
3537 : :
3538 : : /**
3539 : : * put_unbound_pool - put a worker_pool
3540 : : * @pool: worker_pool to put
3541 : : *
3542 : : * Put @pool. If its refcnt reaches zero, it gets destroyed in sched-RCU
3543 : : * safe manner. get_unbound_pool() calls this function on its failure path
3544 : : * and this function should be able to release pools which went through,
3545 : : * successfully or not, init_worker_pool().
3546 : : *
3547 : : * Should be called with wq_pool_mutex held.
3548 : : */
3549 : 0 : static void put_unbound_pool(struct worker_pool *pool)
3550 : : {
3551 : : struct worker *worker;
3552 : :
3553 : : lockdep_assert_held(&wq_pool_mutex);
3554 : :
3555 [ # # ]: 0 : if (--pool->refcnt)
3556 : : return;
3557 : :
3558 : : /* sanity checks */
3559 [ # # ][ # # ]: 0 : if (WARN_ON(!(pool->flags & POOL_DISASSOCIATED)) ||
[ # # ]
3560 [ # # ]: 0 : WARN_ON(!list_empty(&pool->worklist)))
3561 : : return;
3562 : :
3563 : : /* release id and unhash */
3564 [ # # ]: 0 : if (pool->id >= 0)
3565 : 0 : idr_remove(&worker_pool_idr, pool->id);
3566 : : hash_del(&pool->hash_node);
3567 : :
3568 : : /*
3569 : : * Become the manager and destroy all workers. Grabbing
3570 : : * manager_arb prevents @pool's workers from blocking on
3571 : : * manager_mutex.
3572 : : */
3573 : 0 : mutex_lock(&pool->manager_arb);
3574 : 0 : mutex_lock(&pool->manager_mutex);
3575 : : spin_lock_irq(&pool->lock);
3576 : :
3577 [ # # ]: 0 : while ((worker = first_worker(pool)))
3578 : 0 : destroy_worker(worker);
3579 [ # # ][ # # ]: 0 : WARN_ON(pool->nr_workers || pool->nr_idle);
[ # # ]
3580 : :
3581 : : spin_unlock_irq(&pool->lock);
3582 : 0 : mutex_unlock(&pool->manager_mutex);
3583 : 0 : mutex_unlock(&pool->manager_arb);
3584 : :
3585 : : /* shut down the timers */
3586 : 0 : del_timer_sync(&pool->idle_timer);
3587 : 0 : del_timer_sync(&pool->mayday_timer);
3588 : :
3589 : : /* sched-RCU protected to allow dereferences from get_work_pool() */
3590 : 0 : call_rcu_sched(&pool->rcu, rcu_free_pool);
3591 : : }
3592 : :
3593 : : /**
3594 : : * get_unbound_pool - get a worker_pool with the specified attributes
3595 : : * @attrs: the attributes of the worker_pool to get
3596 : : *
3597 : : * Obtain a worker_pool which has the same attributes as @attrs, bump the
3598 : : * reference count and return it. If there already is a matching
3599 : : * worker_pool, it will be used; otherwise, this function attempts to
3600 : : * create a new one.
3601 : : *
3602 : : * Should be called with wq_pool_mutex held.
3603 : : *
3604 : : * Return: On success, a worker_pool with the same attributes as @attrs.
3605 : : * On failure, %NULL.
3606 : : */
3607 : 0 : static struct worker_pool *get_unbound_pool(const struct workqueue_attrs *attrs)
3608 : : {
3609 : 0 : u32 hash = wqattrs_hash(attrs);
3610 : : struct worker_pool *pool;
3611 : : int node;
3612 : :
3613 : : lockdep_assert_held(&wq_pool_mutex);
3614 : :
3615 : : /* do we already have a matching pool? */
3616 [ # # ][ # # ]: 0 : hash_for_each_possible(unbound_pool_hash, pool, hash_node, hash) {
[ # # ]
3617 [ # # ]: 0 : if (wqattrs_equal(pool->attrs, attrs)) {
3618 : 0 : pool->refcnt++;
3619 : 0 : goto out_unlock;
3620 : : }
3621 : : }
3622 : :
3623 : : /* nope, create a new one */
3624 : : pool = kzalloc(sizeof(*pool), GFP_KERNEL);
3625 [ # # ][ # # ]: 0 : if (!pool || init_worker_pool(pool) < 0)
3626 : : goto fail;
3627 : :
3628 [ # # ]: 0 : if (workqueue_freezing)
3629 : 0 : pool->flags |= POOL_FREEZING;
3630 : :
3631 : : lockdep_set_subclass(&pool->lock, 1); /* see put_pwq() */
3632 : 0 : copy_workqueue_attrs(pool->attrs, attrs);
3633 : :
3634 : : /*
3635 : : * no_numa isn't a worker_pool attribute, always clear it. See
3636 : : * 'struct workqueue_attrs' comments for detail.
3637 : : */
3638 : 0 : pool->attrs->no_numa = false;
3639 : :
3640 : : /* if cpumask is contained inside a NUMA node, we belong to that node */
3641 [ # # ]: 0 : if (wq_numa_enabled) {
3642 [ # # ]: 0 : for_each_node(node) {
3643 [ # # ]: 0 : if (cpumask_subset(pool->attrs->cpumask,
3644 : 0 : wq_numa_possible_cpumask[node])) {
3645 : 0 : pool->node = node;
3646 : 0 : break;
3647 : : }
3648 : : }
3649 : : }
3650 : :
3651 [ # # ]: 0 : if (worker_pool_assign_id(pool) < 0)
3652 : : goto fail;
3653 : :
3654 : : /* create and start the initial worker */
3655 [ # # ]: 0 : if (create_and_start_worker(pool) < 0)
3656 : : goto fail;
3657 : :
3658 : : /* install */
3659 : 0 : hash_add(unbound_pool_hash, &pool->hash_node, hash);
3660 : : out_unlock:
3661 : 0 : return pool;
3662 : : fail:
3663 [ # # ]: 0 : if (pool)
3664 : 0 : put_unbound_pool(pool);
3665 : : return NULL;
3666 : : }
3667 : :
3668 : 0 : static void rcu_free_pwq(struct rcu_head *rcu)
3669 : : {
3670 : 0 : kmem_cache_free(pwq_cache,
3671 : 0 : container_of(rcu, struct pool_workqueue, rcu));
3672 : 0 : }
3673 : :
3674 : : /*
3675 : : * Scheduled on system_wq by put_pwq() when an unbound pwq hits zero refcnt
3676 : : * and needs to be destroyed.
3677 : : */
3678 : 0 : static void pwq_unbound_release_workfn(struct work_struct *work)
3679 : : {
3680 : : struct pool_workqueue *pwq = container_of(work, struct pool_workqueue,
3681 : : unbound_release_work);
3682 : 0 : struct workqueue_struct *wq = pwq->wq;
3683 : 0 : struct worker_pool *pool = pwq->pool;
3684 : : bool is_last;
3685 : :
3686 [ # # ][ # # ]: 0 : if (WARN_ON_ONCE(!(wq->flags & WQ_UNBOUND)))
[ # # ][ # # ]
3687 : 0 : return;
3688 : :
3689 : : /*
3690 : : * Unlink @pwq. Synchronization against wq->mutex isn't strictly
3691 : : * necessary on release but do it anyway. It's easier to verify
3692 : : * and consistent with the linking path.
3693 : : */
3694 : 0 : mutex_lock(&wq->mutex);
3695 : : list_del_rcu(&pwq->pwqs_node);
3696 : 0 : is_last = list_empty(&wq->pwqs);
3697 : 0 : mutex_unlock(&wq->mutex);
3698 : :
3699 : 0 : mutex_lock(&wq_pool_mutex);
3700 : 0 : put_unbound_pool(pool);
3701 : 0 : mutex_unlock(&wq_pool_mutex);
3702 : :
3703 : 0 : call_rcu_sched(&pwq->rcu, rcu_free_pwq);
3704 : :
3705 : : /*
3706 : : * If we're the last pwq going away, @wq is already dead and no one
3707 : : * is gonna access it anymore. Free it.
3708 : : */
3709 [ # # ]: 0 : if (is_last) {
3710 : 0 : free_workqueue_attrs(wq->unbound_attrs);
3711 : 0 : kfree(wq);
3712 : : }
3713 : : }
3714 : :
3715 : : /**
3716 : : * pwq_adjust_max_active - update a pwq's max_active to the current setting
3717 : : * @pwq: target pool_workqueue
3718 : : *
3719 : : * If @pwq isn't freezing, set @pwq->max_active to the associated
3720 : : * workqueue's saved_max_active and activate delayed work items
3721 : : * accordingly. If @pwq is freezing, clear @pwq->max_active to zero.
3722 : : */
3723 : 0 : static void pwq_adjust_max_active(struct pool_workqueue *pwq)
3724 : : {
3725 : 0 : struct workqueue_struct *wq = pwq->wq;
3726 : 0 : bool freezable = wq->flags & WQ_FREEZABLE;
3727 : :
3728 : : /* for @wq->saved_max_active */
3729 : : lockdep_assert_held(&wq->mutex);
3730 : :
3731 : : /* fast exit for non-freezable wqs */
3732 [ # # ][ # # ]: 0 : if (!freezable && pwq->max_active == wq->saved_max_active)
3733 : 0 : return;
3734 : :
3735 : 0 : spin_lock_irq(&pwq->pool->lock);
3736 : :
3737 [ # # ][ # # ]: 0 : if (!freezable || !(pwq->pool->flags & POOL_FREEZING)) {
3738 : 0 : pwq->max_active = wq->saved_max_active;
3739 : :
3740 [ # # ][ # # ]: 0 : while (!list_empty(&pwq->delayed_works) &&
3741 : 0 : pwq->nr_active < pwq->max_active)
3742 : : pwq_activate_first_delayed(pwq);
3743 : :
3744 : : /*
3745 : : * Need to kick a worker after thawed or an unbound wq's
3746 : : * max_active is bumped. It's a slow path. Do it always.
3747 : : */
3748 : 0 : wake_up_worker(pwq->pool);
3749 : : } else {
3750 : 0 : pwq->max_active = 0;
3751 : : }
3752 : :
3753 : 0 : spin_unlock_irq(&pwq->pool->lock);
3754 : : }
3755 : :
3756 : : /* initialize newly alloced @pwq which is associated with @wq and @pool */
3757 : 0 : static void init_pwq(struct pool_workqueue *pwq, struct workqueue_struct *wq,
3758 : : struct worker_pool *pool)
3759 : : {
3760 [ # # ]: 0 : BUG_ON((unsigned long)pwq & WORK_STRUCT_FLAG_MASK);
3761 : :
3762 : 0 : memset(pwq, 0, sizeof(*pwq));
3763 : :
3764 : 0 : pwq->pool = pool;
3765 : 0 : pwq->wq = wq;
3766 : 0 : pwq->flush_color = -1;
3767 : 0 : pwq->refcnt = 1;
3768 : 0 : INIT_LIST_HEAD(&pwq->delayed_works);
3769 : 0 : INIT_LIST_HEAD(&pwq->pwqs_node);
3770 : 0 : INIT_LIST_HEAD(&pwq->mayday_node);
3771 : 0 : INIT_WORK(&pwq->unbound_release_work, pwq_unbound_release_workfn);
3772 : 0 : }
3773 : :
3774 : : /* sync @pwq with the current state of its associated wq and link it */
3775 : 0 : static void link_pwq(struct pool_workqueue *pwq)
3776 : : {
3777 : 0 : struct workqueue_struct *wq = pwq->wq;
3778 : :
3779 : : lockdep_assert_held(&wq->mutex);
3780 : :
3781 : : /* may be called multiple times, ignore if already linked */
3782 [ # # ]: 0 : if (!list_empty(&pwq->pwqs_node))
3783 : 0 : return;
3784 : :
3785 : : /*
3786 : : * Set the matching work_color. This is synchronized with
3787 : : * wq->mutex to avoid confusing flush_workqueue().
3788 : : */
3789 : 0 : pwq->work_color = wq->work_color;
3790 : :
3791 : : /* sync max_active to the current setting */
3792 : 0 : pwq_adjust_max_active(pwq);
3793 : :
3794 : : /* link in @pwq */
3795 : 0 : list_add_rcu(&pwq->pwqs_node, &wq->pwqs);
3796 : : }
3797 : :
3798 : : /* obtain a pool matching @attr and create a pwq associating the pool and @wq */
3799 : 0 : static struct pool_workqueue *alloc_unbound_pwq(struct workqueue_struct *wq,
3800 : : const struct workqueue_attrs *attrs)
3801 : : {
3802 : : struct worker_pool *pool;
3803 : : struct pool_workqueue *pwq;
3804 : :
3805 : : lockdep_assert_held(&wq_pool_mutex);
3806 : :
3807 : 0 : pool = get_unbound_pool(attrs);
3808 [ # # ]: 0 : if (!pool)
3809 : : return NULL;
3810 : :
3811 : 0 : pwq = kmem_cache_alloc_node(pwq_cache, GFP_KERNEL, pool->node);
3812 [ # # ]: 0 : if (!pwq) {
3813 : 0 : put_unbound_pool(pool);
3814 : 0 : return NULL;
3815 : : }
3816 : :
3817 : 0 : init_pwq(pwq, wq, pool);
3818 : 0 : return pwq;
3819 : : }
3820 : :
3821 : : /* undo alloc_unbound_pwq(), used only in the error path */
3822 : 0 : static void free_unbound_pwq(struct pool_workqueue *pwq)
3823 : : {
3824 : : lockdep_assert_held(&wq_pool_mutex);
3825 : :
3826 [ # # ]: 0 : if (pwq) {
3827 : 0 : put_unbound_pool(pwq->pool);
3828 : 0 : kmem_cache_free(pwq_cache, pwq);
3829 : : }
3830 : 0 : }
3831 : :
3832 : : /**
3833 : : * wq_calc_node_mask - calculate a wq_attrs' cpumask for the specified node
3834 : : * @attrs: the wq_attrs of interest
3835 : : * @node: the target NUMA node
3836 : : * @cpu_going_down: if >= 0, the CPU to consider as offline
3837 : : * @cpumask: outarg, the resulting cpumask
3838 : : *
3839 : : * Calculate the cpumask a workqueue with @attrs should use on @node. If
3840 : : * @cpu_going_down is >= 0, that cpu is considered offline during
3841 : : * calculation. The result is stored in @cpumask.
3842 : : *
3843 : : * If NUMA affinity is not enabled, @attrs->cpumask is always used. If
3844 : : * enabled and @node has online CPUs requested by @attrs, the returned
3845 : : * cpumask is the intersection of the possible CPUs of @node and
3846 : : * @attrs->cpumask.
3847 : : *
3848 : : * The caller is responsible for ensuring that the cpumask of @node stays
3849 : : * stable.
3850 : : *
3851 : : * Return: %true if the resulting @cpumask is different from @attrs->cpumask,
3852 : : * %false if equal.
3853 : : */
3854 : 0 : static bool wq_calc_node_cpumask(const struct workqueue_attrs *attrs, int node,
3855 : : int cpu_going_down, cpumask_t *cpumask)
3856 : : {
3857 [ # # ][ # # ]: 0 : if (!wq_numa_enabled || attrs->no_numa)
3858 : : goto use_dfl;
3859 : :
3860 : : /* does @node have any online CPUs @attrs wants? */
3861 : 0 : cpumask_and(cpumask, cpumask_of_node(node), attrs->cpumask);
3862 [ # # ]: 0 : if (cpu_going_down >= 0)
3863 : : cpumask_clear_cpu(cpu_going_down, cpumask);
3864 : :
3865 [ # # ]: 0 : if (cpumask_empty(cpumask))
3866 : : goto use_dfl;
3867 : :
3868 : : /* yeap, return possible CPUs in @node that @attrs wants */
3869 : 0 : cpumask_and(cpumask, attrs->cpumask, wq_numa_possible_cpumask[node]);
3870 : 0 : return !cpumask_equal(cpumask, attrs->cpumask);
3871 : :
3872 : : use_dfl:
3873 : : cpumask_copy(cpumask, attrs->cpumask);
3874 : 0 : return false;
3875 : : }
3876 : :
3877 : : /* install @pwq into @wq's numa_pwq_tbl[] for @node and return the old pwq */
3878 : : static struct pool_workqueue *numa_pwq_tbl_install(struct workqueue_struct *wq,
3879 : : int node,
3880 : : struct pool_workqueue *pwq)
3881 : : {
3882 : : struct pool_workqueue *old_pwq;
3883 : :
3884 : : lockdep_assert_held(&wq->mutex);
3885 : :
3886 : : /* link_pwq() can handle duplicate calls */
3887 : 0 : link_pwq(pwq);
3888 : :
3889 : 0 : old_pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]);
3890 : 0 : rcu_assign_pointer(wq->numa_pwq_tbl[node], pwq);
3891 : : return old_pwq;
3892 : : }
3893 : :
3894 : : /**
3895 : : * apply_workqueue_attrs - apply new workqueue_attrs to an unbound workqueue
3896 : : * @wq: the target workqueue
3897 : : * @attrs: the workqueue_attrs to apply, allocated with alloc_workqueue_attrs()
3898 : : *
3899 : : * Apply @attrs to an unbound workqueue @wq. Unless disabled, on NUMA
3900 : : * machines, this function maps a separate pwq to each NUMA node with
3901 : : * possibles CPUs in @attrs->cpumask so that work items are affine to the
3902 : : * NUMA node it was issued on. Older pwqs are released as in-flight work
3903 : : * items finish. Note that a work item which repeatedly requeues itself
3904 : : * back-to-back will stay on its current pwq.
3905 : : *
3906 : : * Performs GFP_KERNEL allocations.
3907 : : *
3908 : : * Return: 0 on success and -errno on failure.
3909 : : */
3910 : 0 : int apply_workqueue_attrs(struct workqueue_struct *wq,
3911 : : const struct workqueue_attrs *attrs)
3912 : : {
3913 : : struct workqueue_attrs *new_attrs, *tmp_attrs;
3914 : : struct pool_workqueue **pwq_tbl, *dfl_pwq;
3915 : : int node, ret;
3916 : :
3917 : : /* only unbound workqueues can change attributes */
3918 [ # # ][ # # ]: 0 : if (WARN_ON(!(wq->flags & WQ_UNBOUND)))
3919 : : return -EINVAL;
3920 : :
3921 : : /* creating multiple pwqs breaks ordering guarantee */
3922 [ # # ][ # # ]: 0 : if (WARN_ON((wq->flags & __WQ_ORDERED) && !list_empty(&wq->pwqs)))
[ # # ][ # # ]
3923 : : return -EINVAL;
3924 : :
3925 : 0 : pwq_tbl = kzalloc(wq_numa_tbl_len * sizeof(pwq_tbl[0]), GFP_KERNEL);
3926 : 0 : new_attrs = alloc_workqueue_attrs(GFP_KERNEL);
3927 : 0 : tmp_attrs = alloc_workqueue_attrs(GFP_KERNEL);
3928 [ # # ][ # # ]: 0 : if (!pwq_tbl || !new_attrs || !tmp_attrs)
3929 : : goto enomem;
3930 : :
3931 : : /* make a copy of @attrs and sanitize it */
3932 : : copy_workqueue_attrs(new_attrs, attrs);
3933 : 0 : cpumask_and(new_attrs->cpumask, new_attrs->cpumask, cpu_possible_mask);
3934 : :
3935 : : /*
3936 : : * We may create multiple pwqs with differing cpumasks. Make a
3937 : : * copy of @new_attrs which will be modified and used to obtain
3938 : : * pools.
3939 : : */
3940 : : copy_workqueue_attrs(tmp_attrs, new_attrs);
3941 : :
3942 : : /*
3943 : : * CPUs should stay stable across pwq creations and installations.
3944 : : * Pin CPUs, determine the target cpumask for each node and create
3945 : : * pwqs accordingly.
3946 : : */
3947 : 0 : get_online_cpus();
3948 : :
3949 : 0 : mutex_lock(&wq_pool_mutex);
3950 : :
3951 : : /*
3952 : : * If something goes wrong during CPU up/down, we'll fall back to
3953 : : * the default pwq covering whole @attrs->cpumask. Always create
3954 : : * it even if we don't use it immediately.
3955 : : */
3956 : 0 : dfl_pwq = alloc_unbound_pwq(wq, new_attrs);
3957 [ # # ]: 0 : if (!dfl_pwq)
3958 : : goto enomem_pwq;
3959 : :
3960 [ # # ]: 0 : for_each_node(node) {
3961 [ # # ]: 0 : if (wq_calc_node_cpumask(attrs, node, -1, tmp_attrs->cpumask)) {
3962 : 0 : pwq_tbl[node] = alloc_unbound_pwq(wq, tmp_attrs);
3963 [ # # ]: 0 : if (!pwq_tbl[node])
3964 : : goto enomem_pwq;
3965 : : } else {
3966 : 0 : dfl_pwq->refcnt++;
3967 : 0 : pwq_tbl[node] = dfl_pwq;
3968 : : }
3969 : : }
3970 : :
3971 : 0 : mutex_unlock(&wq_pool_mutex);
3972 : :
3973 : : /* all pwqs have been created successfully, let's install'em */
3974 : 0 : mutex_lock(&wq->mutex);
3975 : :
3976 : 0 : copy_workqueue_attrs(wq->unbound_attrs, new_attrs);
3977 : :
3978 : : /* save the previous pwq and install the new one */
3979 [ # # ]: 0 : for_each_node(node)
3980 : 0 : pwq_tbl[node] = numa_pwq_tbl_install(wq, node, pwq_tbl[node]);
3981 : :
3982 : : /* @dfl_pwq might not have been used, ensure it's linked */
3983 : 0 : link_pwq(dfl_pwq);
3984 : 0 : swap(wq->dfl_pwq, dfl_pwq);
3985 : :
3986 : 0 : mutex_unlock(&wq->mutex);
3987 : :
3988 : : /* put the old pwqs */
3989 [ # # ]: 0 : for_each_node(node)
3990 : 0 : put_pwq_unlocked(pwq_tbl[node]);
3991 : 0 : put_pwq_unlocked(dfl_pwq);
3992 : :
3993 : 0 : put_online_cpus();
3994 : : ret = 0;
3995 : : /* fall through */
3996 : : out_free:
3997 : : free_workqueue_attrs(tmp_attrs);
3998 : : free_workqueue_attrs(new_attrs);
3999 : 0 : kfree(pwq_tbl);
4000 : 0 : return ret;
4001 : :
4002 : : enomem_pwq:
4003 : 0 : free_unbound_pwq(dfl_pwq);
4004 [ # # ]: 0 : for_each_node(node)
4005 [ # # ][ # # ]: 0 : if (pwq_tbl && pwq_tbl[node] != dfl_pwq)
4006 : 0 : free_unbound_pwq(pwq_tbl[node]);
4007 : 0 : mutex_unlock(&wq_pool_mutex);
4008 : 0 : put_online_cpus();
4009 : : enomem:
4010 : : ret = -ENOMEM;
4011 : : goto out_free;
4012 : : }
4013 : :
4014 : : /**
4015 : : * wq_update_unbound_numa - update NUMA affinity of a wq for CPU hot[un]plug
4016 : : * @wq: the target workqueue
4017 : : * @cpu: the CPU coming up or going down
4018 : : * @online: whether @cpu is coming up or going down
4019 : : *
4020 : : * This function is to be called from %CPU_DOWN_PREPARE, %CPU_ONLINE and
4021 : : * %CPU_DOWN_FAILED. @cpu is being hot[un]plugged, update NUMA affinity of
4022 : : * @wq accordingly.
4023 : : *
4024 : : * If NUMA affinity can't be adjusted due to memory allocation failure, it
4025 : : * falls back to @wq->dfl_pwq which may not be optimal but is always
4026 : : * correct.
4027 : : *
4028 : : * Note that when the last allowed CPU of a NUMA node goes offline for a
4029 : : * workqueue with a cpumask spanning multiple nodes, the workers which were
4030 : : * already executing the work items for the workqueue will lose their CPU
4031 : : * affinity and may execute on any CPU. This is similar to how per-cpu
4032 : : * workqueues behave on CPU_DOWN. If a workqueue user wants strict
4033 : : * affinity, it's the user's responsibility to flush the work item from
4034 : : * CPU_DOWN_PREPARE.
4035 : : */
4036 : 0 : static void wq_update_unbound_numa(struct workqueue_struct *wq, int cpu,
4037 : : bool online)
4038 : : {
4039 : : int node = cpu_to_node(cpu);
4040 [ # # ]: 0 : int cpu_off = online ? -1 : cpu;
4041 : : struct pool_workqueue *old_pwq = NULL, *pwq;
4042 : : struct workqueue_attrs *target_attrs;
4043 : : cpumask_t *cpumask;
4044 : :
4045 : : lockdep_assert_held(&wq_pool_mutex);
4046 : :
4047 [ # # ][ # # ]: 0 : if (!wq_numa_enabled || !(wq->flags & WQ_UNBOUND))
4048 : 0 : return;
4049 : :
4050 : : /*
4051 : : * We don't wanna alloc/free wq_attrs for each wq for each CPU.
4052 : : * Let's use a preallocated one. The following buf is protected by
4053 : : * CPU hotplug exclusion.
4054 : : */
4055 : 0 : target_attrs = wq_update_unbound_numa_attrs_buf;
4056 : 0 : cpumask = target_attrs->cpumask;
4057 : :
4058 : 0 : mutex_lock(&wq->mutex);
4059 [ # # ]: 0 : if (wq->unbound_attrs->no_numa)
4060 : : goto out_unlock;
4061 : :
4062 : : copy_workqueue_attrs(target_attrs, wq->unbound_attrs);
4063 : : pwq = unbound_pwq_by_node(wq, node);
4064 : :
4065 : : /*
4066 : : * Let's determine what needs to be done. If the target cpumask is
4067 : : * different from wq's, we need to compare it to @pwq's and create
4068 : : * a new one if they don't match. If the target cpumask equals
4069 : : * wq's, the default pwq should be used. If @pwq is already the
4070 : : * default one, nothing to do; otherwise, install the default one.
4071 : : */
4072 [ # # ]: 0 : if (wq_calc_node_cpumask(wq->unbound_attrs, node, cpu_off, cpumask)) {
4073 [ # # ]: 0 : if (cpumask_equal(cpumask, pwq->pool->attrs->cpumask))
4074 : : goto out_unlock;
4075 : : } else {
4076 [ # # ]: 0 : if (pwq == wq->dfl_pwq)
4077 : : goto out_unlock;
4078 : : else
4079 : : goto use_dfl_pwq;
4080 : : }
4081 : :
4082 : 0 : mutex_unlock(&wq->mutex);
4083 : :
4084 : : /* create a new pwq */
4085 : 0 : pwq = alloc_unbound_pwq(wq, target_attrs);
4086 [ # # ]: 0 : if (!pwq) {
4087 : 0 : pr_warning("workqueue: allocation failed while updating NUMA affinity of \"%s\"\n",
4088 : : wq->name);
4089 : 0 : goto out_unlock;
4090 : : }
4091 : :
4092 : : /*
4093 : : * Install the new pwq. As this function is called only from CPU
4094 : : * hotplug callbacks and applying a new attrs is wrapped with
4095 : : * get/put_online_cpus(), @wq->unbound_attrs couldn't have changed
4096 : : * inbetween.
4097 : : */
4098 : 0 : mutex_lock(&wq->mutex);
4099 : : old_pwq = numa_pwq_tbl_install(wq, node, pwq);
4100 : 0 : goto out_unlock;
4101 : :
4102 : : use_dfl_pwq:
4103 : 0 : spin_lock_irq(&wq->dfl_pwq->pool->lock);
4104 : 0 : get_pwq(wq->dfl_pwq);
4105 : 0 : spin_unlock_irq(&wq->dfl_pwq->pool->lock);
4106 : 0 : old_pwq = numa_pwq_tbl_install(wq, node, wq->dfl_pwq);
4107 : : out_unlock:
4108 : 0 : mutex_unlock(&wq->mutex);
4109 : 0 : put_pwq_unlocked(old_pwq);
4110 : : }
4111 : :
4112 : 0 : static int alloc_and_link_pwqs(struct workqueue_struct *wq)
4113 : : {
4114 : 0 : bool highpri = wq->flags & WQ_HIGHPRI;
4115 : : int cpu, ret;
4116 : :
4117 [ # # ]: 0 : if (!(wq->flags & WQ_UNBOUND)) {
4118 : 0 : wq->cpu_pwqs = alloc_percpu(struct pool_workqueue);
4119 [ # # ]: 0 : if (!wq->cpu_pwqs)
4120 : : return -ENOMEM;
4121 : :
4122 [ # # ]: 0 : for_each_possible_cpu(cpu) {
4123 : : struct pool_workqueue *pwq =
4124 : 0 : per_cpu_ptr(wq->cpu_pwqs, cpu);
4125 : : struct worker_pool *cpu_pools =
4126 : 0 : per_cpu(cpu_worker_pools, cpu);
4127 : :
4128 : 0 : init_pwq(pwq, wq, &cpu_pools[highpri]);
4129 : :
4130 : 0 : mutex_lock(&wq->mutex);
4131 : 0 : link_pwq(pwq);
4132 : 0 : mutex_unlock(&wq->mutex);
4133 : : }
4134 : : return 0;
4135 [ # # ]: 0 : } else if (wq->flags & __WQ_ORDERED) {
4136 : 0 : ret = apply_workqueue_attrs(wq, ordered_wq_attrs[highpri]);
4137 : : /* there should only be single pwq for ordering guarantee */
4138 [ # # ][ # # ]: 0 : WARN(!ret && (wq->pwqs.next != &wq->dfl_pwq->pwqs_node ||
[ # # ][ # # ]
4139 : : wq->pwqs.prev != &wq->dfl_pwq->pwqs_node),
4140 : : "ordering guarantee broken for workqueue %s\n", wq->name);
4141 : 0 : return ret;
4142 : : } else {
4143 : 0 : return apply_workqueue_attrs(wq, unbound_std_wq_attrs[highpri]);
4144 : : }
4145 : : }
4146 : :
4147 : 0 : static int wq_clamp_max_active(int max_active, unsigned int flags,
4148 : : const char *name)
4149 : : {
4150 [ # # ]: 0 : int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE;
4151 : :
4152 [ # # ]: 0 : if (max_active < 1 || max_active > lim)
4153 : 0 : pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n",
4154 : : max_active, name, 1, lim);
4155 : :
4156 : 0 : return clamp_val(max_active, 1, lim);
4157 : : }
4158 : :
4159 : 0 : struct workqueue_struct *__alloc_workqueue_key(const char *fmt,
4160 : : unsigned int flags,
4161 : : int max_active,
4162 : : struct lock_class_key *key,
4163 : : const char *lock_name, ...)
4164 : : {
4165 : : size_t tbl_size = 0;
4166 : : va_list args;
4167 : : struct workqueue_struct *wq;
4168 : : struct pool_workqueue *pwq;
4169 : :
4170 : : /* see the comment above the definition of WQ_POWER_EFFICIENT */
4171 [ # # ][ # # ]: 0 : if ((flags & WQ_POWER_EFFICIENT) && wq_power_efficient)
4172 : 0 : flags |= WQ_UNBOUND;
4173 : :
4174 : : /* allocate wq and format name */
4175 [ # # ]: 0 : if (flags & WQ_UNBOUND)
4176 : 0 : tbl_size = wq_numa_tbl_len * sizeof(wq->numa_pwq_tbl[0]);
4177 : :
4178 : 0 : wq = kzalloc(sizeof(*wq) + tbl_size, GFP_KERNEL);
4179 [ # # ]: 0 : if (!wq)
4180 : : return NULL;
4181 : :
4182 [ # # ]: 0 : if (flags & WQ_UNBOUND) {
4183 : 0 : wq->unbound_attrs = alloc_workqueue_attrs(GFP_KERNEL);
4184 [ # # ]: 0 : if (!wq->unbound_attrs)
4185 : : goto err_free_wq;
4186 : : }
4187 : :
4188 : 0 : va_start(args, lock_name);
4189 : 0 : vsnprintf(wq->name, sizeof(wq->name), fmt, args);
4190 : 0 : va_end(args);
4191 : :
4192 [ # # ]: 0 : max_active = max_active ?: WQ_DFL_ACTIVE;
4193 : 0 : max_active = wq_clamp_max_active(max_active, flags, wq->name);
4194 : :
4195 : : /* init wq */
4196 : 0 : wq->flags = flags;
4197 : 0 : wq->saved_max_active = max_active;
4198 : 0 : mutex_init(&wq->mutex);
4199 : 0 : atomic_set(&wq->nr_pwqs_to_flush, 0);
4200 : 0 : INIT_LIST_HEAD(&wq->pwqs);
4201 : 0 : INIT_LIST_HEAD(&wq->flusher_queue);
4202 : 0 : INIT_LIST_HEAD(&wq->flusher_overflow);
4203 : 0 : INIT_LIST_HEAD(&wq->maydays);
4204 : :
4205 : : lockdep_init_map(&wq->lockdep_map, lock_name, key, 0);
4206 : 0 : INIT_LIST_HEAD(&wq->list);
4207 : :
4208 [ # # ]: 0 : if (alloc_and_link_pwqs(wq) < 0)
4209 : : goto err_free_wq;
4210 : :
4211 : : /*
4212 : : * Workqueues which may be used during memory reclaim should
4213 : : * have a rescuer to guarantee forward progress.
4214 : : */
4215 [ # # ]: 0 : if (flags & WQ_MEM_RECLAIM) {
4216 : : struct worker *rescuer;
4217 : :
4218 : 0 : rescuer = alloc_worker();
4219 [ # # ]: 0 : if (!rescuer)
4220 : : goto err_destroy;
4221 : :
4222 : 0 : rescuer->rescue_wq = wq;
4223 : 0 : rescuer->task = kthread_create(rescuer_thread, rescuer, "%s",
4224 : : wq->name);
4225 [ # # ]: 0 : if (IS_ERR(rescuer->task)) {
4226 : 0 : kfree(rescuer);
4227 : 0 : goto err_destroy;
4228 : : }
4229 : :
4230 : 0 : wq->rescuer = rescuer;
4231 : 0 : rescuer->task->flags |= PF_NO_SETAFFINITY;
4232 : 0 : wake_up_process(rescuer->task);
4233 : : }
4234 : :
4235 [ # # ][ # # ]: 0 : if ((wq->flags & WQ_SYSFS) && workqueue_sysfs_register(wq))
4236 : : goto err_destroy;
4237 : :
4238 : : /*
4239 : : * wq_pool_mutex protects global freeze state and workqueues list.
4240 : : * Grab it, adjust max_active and add the new @wq to workqueues
4241 : : * list.
4242 : : */
4243 : 0 : mutex_lock(&wq_pool_mutex);
4244 : :
4245 : 0 : mutex_lock(&wq->mutex);
4246 [ # # ]: 0 : for_each_pwq(pwq, wq)
4247 : 0 : pwq_adjust_max_active(pwq);
4248 : 0 : mutex_unlock(&wq->mutex);
4249 : :
4250 : : list_add(&wq->list, &workqueues);
4251 : :
4252 : 0 : mutex_unlock(&wq_pool_mutex);
4253 : :
4254 : 0 : return wq;
4255 : :
4256 : : err_free_wq:
4257 : 0 : free_workqueue_attrs(wq->unbound_attrs);
4258 : 0 : kfree(wq);
4259 : 0 : return NULL;
4260 : : err_destroy:
4261 : 0 : destroy_workqueue(wq);
4262 : 0 : return NULL;
4263 : : }
4264 : : EXPORT_SYMBOL_GPL(__alloc_workqueue_key);
4265 : :
4266 : : /**
4267 : : * destroy_workqueue - safely terminate a workqueue
4268 : : * @wq: target workqueue
4269 : : *
4270 : : * Safely destroy a workqueue. All work currently pending will be done first.
4271 : : */
4272 : 0 : void destroy_workqueue(struct workqueue_struct *wq)
4273 : : {
4274 : : struct pool_workqueue *pwq;
4275 : : int node;
4276 : :
4277 : : /* drain it before proceeding with destruction */
4278 : 0 : drain_workqueue(wq);
4279 : :
4280 : : /* sanity checks */
4281 : 0 : mutex_lock(&wq->mutex);
4282 [ # # ]: 0 : for_each_pwq(pwq, wq) {
4283 : : int i;
4284 : :
4285 [ # # ]: 0 : for (i = 0; i < WORK_NR_COLORS; i++) {
4286 [ # # ][ # # ]: 0 : if (WARN_ON(pwq->nr_in_flight[i])) {
4287 : 0 : mutex_unlock(&wq->mutex);
4288 : 0 : return;
4289 : : }
4290 : : }
4291 : :
4292 [ # # ][ # # ]: 0 : if (WARN_ON((pwq != wq->dfl_pwq) && (pwq->refcnt > 1)) ||
[ # # ][ # # ]
[ # # ]
4293 [ # # ][ # # ]: 0 : WARN_ON(pwq->nr_active) ||
4294 [ # # ]: 0 : WARN_ON(!list_empty(&pwq->delayed_works))) {
4295 : 0 : mutex_unlock(&wq->mutex);
4296 : 0 : return;
4297 : : }
4298 : : }
4299 : 0 : mutex_unlock(&wq->mutex);
4300 : :
4301 : : /*
4302 : : * wq list is used to freeze wq, remove from list after
4303 : : * flushing is complete in case freeze races us.
4304 : : */
4305 : 0 : mutex_lock(&wq_pool_mutex);
4306 : 0 : list_del_init(&wq->list);
4307 : 0 : mutex_unlock(&wq_pool_mutex);
4308 : :
4309 : : workqueue_sysfs_unregister(wq);
4310 : :
4311 [ # # ]: 0 : if (wq->rescuer) {
4312 : 0 : kthread_stop(wq->rescuer->task);
4313 : 0 : kfree(wq->rescuer);
4314 : 0 : wq->rescuer = NULL;
4315 : : }
4316 : :
4317 [ # # ]: 0 : if (!(wq->flags & WQ_UNBOUND)) {
4318 : : /*
4319 : : * The base ref is never dropped on per-cpu pwqs. Directly
4320 : : * free the pwqs and wq.
4321 : : */
4322 : 0 : free_percpu(wq->cpu_pwqs);
4323 : 0 : kfree(wq);
4324 : : } else {
4325 : : /*
4326 : : * We're the sole accessor of @wq at this point. Directly
4327 : : * access numa_pwq_tbl[] and dfl_pwq to put the base refs.
4328 : : * @wq will be freed when the last pwq is released.
4329 : : */
4330 [ # # ]: 0 : for_each_node(node) {
4331 : 0 : pwq = rcu_access_pointer(wq->numa_pwq_tbl[node]);
4332 : 0 : RCU_INIT_POINTER(wq->numa_pwq_tbl[node], NULL);
4333 : 0 : put_pwq_unlocked(pwq);
4334 : : }
4335 : :
4336 : : /*
4337 : : * Put dfl_pwq. @wq may be freed any time after dfl_pwq is
4338 : : * put. Don't access it afterwards.
4339 : : */
4340 : 0 : pwq = wq->dfl_pwq;
4341 : 0 : wq->dfl_pwq = NULL;
4342 : 0 : put_pwq_unlocked(pwq);
4343 : : }
4344 : : }
4345 : : EXPORT_SYMBOL_GPL(destroy_workqueue);
4346 : :
4347 : : /**
4348 : : * workqueue_set_max_active - adjust max_active of a workqueue
4349 : : * @wq: target workqueue
4350 : : * @max_active: new max_active value.
4351 : : *
4352 : : * Set max_active of @wq to @max_active.
4353 : : *
4354 : : * CONTEXT:
4355 : : * Don't call from IRQ context.
4356 : : */
4357 : 0 : void workqueue_set_max_active(struct workqueue_struct *wq, int max_active)
4358 : : {
4359 : : struct pool_workqueue *pwq;
4360 : :
4361 : : /* disallow meddling with max_active for ordered workqueues */
4362 [ # # ][ # # ]: 0 : if (WARN_ON(wq->flags & __WQ_ORDERED))
4363 : 0 : return;
4364 : :
4365 : 0 : max_active = wq_clamp_max_active(max_active, wq->flags, wq->name);
4366 : :
4367 : 0 : mutex_lock(&wq->mutex);
4368 : :
4369 : 0 : wq->saved_max_active = max_active;
4370 : :
4371 [ # # ]: 0 : for_each_pwq(pwq, wq)
4372 : 0 : pwq_adjust_max_active(pwq);
4373 : :
4374 : 0 : mutex_unlock(&wq->mutex);
4375 : : }
4376 : : EXPORT_SYMBOL_GPL(workqueue_set_max_active);
4377 : :
4378 : : /**
4379 : : * current_is_workqueue_rescuer - is %current workqueue rescuer?
4380 : : *
4381 : : * Determine whether %current is a workqueue rescuer. Can be used from
4382 : : * work functions to determine whether it's being run off the rescuer task.
4383 : : *
4384 : : * Return: %true if %current is a workqueue rescuer. %false otherwise.
4385 : : */
4386 : 0 : bool current_is_workqueue_rescuer(void)
4387 : : {
4388 : : struct worker *worker = current_wq_worker();
4389 : :
4390 [ + - ][ + - ]: 9580 : return worker && worker->rescue_wq;
4391 : : }
4392 : :
4393 : : /**
4394 : : * workqueue_congested - test whether a workqueue is congested
4395 : : * @cpu: CPU in question
4396 : : * @wq: target workqueue
4397 : : *
4398 : : * Test whether @wq's cpu workqueue for @cpu is congested. There is
4399 : : * no synchronization around this function and the test result is
4400 : : * unreliable and only useful as advisory hints or for debugging.
4401 : : *
4402 : : * If @cpu is WORK_CPU_UNBOUND, the test is performed on the local CPU.
4403 : : * Note that both per-cpu and unbound workqueues may be associated with
4404 : : * multiple pool_workqueues which have separate congested states. A
4405 : : * workqueue being congested on one CPU doesn't mean the workqueue is also
4406 : : * contested on other CPUs / NUMA nodes.
4407 : : *
4408 : : * Return:
4409 : : * %true if congested, %false otherwise.
4410 : : */
4411 : 0 : bool workqueue_congested(int cpu, struct workqueue_struct *wq)
4412 : : {
4413 : : struct pool_workqueue *pwq;
4414 : : bool ret;
4415 : :
4416 : : rcu_read_lock_sched();
4417 : :
4418 [ # # ]: 0 : if (cpu == WORK_CPU_UNBOUND)
4419 : 0 : cpu = smp_processor_id();
4420 : :
4421 [ # # ]: 0 : if (!(wq->flags & WQ_UNBOUND))
4422 : 0 : pwq = per_cpu_ptr(wq->cpu_pwqs, cpu);
4423 : : else
4424 : : pwq = unbound_pwq_by_node(wq, cpu_to_node(cpu));
4425 : :
4426 : 0 : ret = !list_empty(&pwq->delayed_works);
4427 : : rcu_read_unlock_sched();
4428 : :
4429 : 0 : return ret;
4430 : : }
4431 : : EXPORT_SYMBOL_GPL(workqueue_congested);
4432 : :
4433 : : /**
4434 : : * work_busy - test whether a work is currently pending or running
4435 : : * @work: the work to be tested
4436 : : *
4437 : : * Test whether @work is currently pending or running. There is no
4438 : : * synchronization around this function and the test result is
4439 : : * unreliable and only useful as advisory hints or for debugging.
4440 : : *
4441 : : * Return:
4442 : : * OR'd bitmask of WORK_BUSY_* bits.
4443 : : */
4444 : 0 : unsigned int work_busy(struct work_struct *work)
4445 : : {
4446 : : struct worker_pool *pool;
4447 : : unsigned long flags;
4448 : : unsigned int ret = 0;
4449 : :
4450 [ # # ]: 0 : if (work_pending(work))
4451 : : ret |= WORK_BUSY_PENDING;
4452 : :
4453 : : local_irq_save(flags);
4454 : 0 : pool = get_work_pool(work);
4455 [ # # ]: 0 : if (pool) {
4456 : : spin_lock(&pool->lock);
4457 [ # # ]: 0 : if (find_worker_executing_work(pool, work))
4458 : 0 : ret |= WORK_BUSY_RUNNING;
4459 : : spin_unlock(&pool->lock);
4460 : : }
4461 [ # # ]: 0 : local_irq_restore(flags);
4462 : :
4463 : 0 : return ret;
4464 : : }
4465 : : EXPORT_SYMBOL_GPL(work_busy);
4466 : :
4467 : : /**
4468 : : * set_worker_desc - set description for the current work item
4469 : : * @fmt: printf-style format string
4470 : : * @...: arguments for the format string
4471 : : *
4472 : : * This function can be called by a running work function to describe what
4473 : : * the work item is about. If the worker task gets dumped, this
4474 : : * information will be printed out together to help debugging. The
4475 : : * description can be at most WORKER_DESC_LEN including the trailing '\0'.
4476 : : */
4477 : 0 : void set_worker_desc(const char *fmt, ...)
4478 : : {
4479 : : struct worker *worker = current_wq_worker();
4480 : : va_list args;
4481 : :
4482 [ + - ]: 9580 : if (worker) {
4483 : 4790 : va_start(args, fmt);
4484 : 4790 : vsnprintf(worker->desc, sizeof(worker->desc), fmt, args);
4485 : 4790 : va_end(args);
4486 : 4790 : worker->desc_valid = true;
4487 : : }
4488 : 4790 : }
4489 : :
4490 : : /**
4491 : : * print_worker_info - print out worker information and description
4492 : : * @log_lvl: the log level to use when printing
4493 : : * @task: target task
4494 : : *
4495 : : * If @task is a worker and currently executing a work item, print out the
4496 : : * name of the workqueue being serviced and worker description set with
4497 : : * set_worker_desc() by the currently executing work item.
4498 : : *
4499 : : * This function can be safely called on any task as long as the
4500 : : * task_struct itself is accessible. While safe, this function isn't
4501 : : * synchronized and may print out mixups or garbages of limited length.
4502 : : */
4503 : 0 : void print_worker_info(const char *log_lvl, struct task_struct *task)
4504 : : {
4505 : 3 : work_func_t *fn = NULL;
4506 : 3 : char name[WQ_NAME_LEN] = { };
4507 : 3 : char desc[WORKER_DESC_LEN] = { };
4508 : 3 : struct pool_workqueue *pwq = NULL;
4509 : 3 : struct workqueue_struct *wq = NULL;
4510 : 3 : bool desc_valid = false;
4511 : : struct worker *worker;
4512 : :
4513 [ - + ]: 3 : if (!(task->flags & PF_WQ_WORKER))
4514 : 3 : return;
4515 : :
4516 : : /*
4517 : : * This function is called without any synchronization and @task
4518 : : * could be in any state. Be careful with dereferences.
4519 : : */
4520 : 0 : worker = probe_kthread_data(task);
4521 : :
4522 : : /*
4523 : : * Carefully copy the associated workqueue's workfn and name. Keep
4524 : : * the original last '\0' in case the original contains garbage.
4525 : : */
4526 : 0 : probe_kernel_read(&fn, &worker->current_func, sizeof(fn));
4527 : 0 : probe_kernel_read(&pwq, &worker->current_pwq, sizeof(pwq));
4528 : 0 : probe_kernel_read(&wq, &pwq->wq, sizeof(wq));
4529 : 0 : probe_kernel_read(name, wq->name, sizeof(name) - 1);
4530 : :
4531 : : /* copy worker description */
4532 : 0 : probe_kernel_read(&desc_valid, &worker->desc_valid, sizeof(desc_valid));
4533 [ # # ]: 0 : if (desc_valid)
4534 : 0 : probe_kernel_read(desc, worker->desc, sizeof(desc) - 1);
4535 : :
4536 [ # # ][ # # ]: 0 : if (fn || name[0] || desc[0]) {
[ # # ]
4537 : 0 : printk("%sWorkqueue: %s %pf", log_lvl, name, fn);
4538 [ # # ]: 0 : if (desc[0])
4539 : 0 : pr_cont(" (%s)", desc);
4540 : 0 : pr_cont("\n");
4541 : : }
4542 : : }
4543 : :
4544 : : /*
4545 : : * CPU hotplug.
4546 : : *
4547 : : * There are two challenges in supporting CPU hotplug. Firstly, there
4548 : : * are a lot of assumptions on strong associations among work, pwq and
4549 : : * pool which make migrating pending and scheduled works very
4550 : : * difficult to implement without impacting hot paths. Secondly,
4551 : : * worker pools serve mix of short, long and very long running works making
4552 : : * blocked draining impractical.
4553 : : *
4554 : : * This is solved by allowing the pools to be disassociated from the CPU
4555 : : * running as an unbound one and allowing it to be reattached later if the
4556 : : * cpu comes back online.
4557 : : */
4558 : :
4559 : 0 : static void wq_unbind_fn(struct work_struct *work)
4560 : : {
4561 : 0 : int cpu = smp_processor_id();
4562 : : struct worker_pool *pool;
4563 : : struct worker *worker;
4564 : : int wi;
4565 : :
4566 [ # # ]: 0 : for_each_cpu_worker_pool(pool, cpu) {
4567 [ # # ][ # # ]: 0 : WARN_ON_ONCE(cpu != smp_processor_id());
[ # # ]
4568 : :
4569 : 0 : mutex_lock(&pool->manager_mutex);
4570 : : spin_lock_irq(&pool->lock);
4571 : :
4572 : : /*
4573 : : * We've blocked all manager operations. Make all workers
4574 : : * unbound and set DISASSOCIATED. Before this, all workers
4575 : : * except for the ones which are still executing works from
4576 : : * before the last CPU down must be on the cpu. After
4577 : : * this, they may become diasporas.
4578 : : */
4579 [ # # ]: 0 : for_each_pool_worker(worker, wi, pool)
4580 : 0 : worker->flags |= WORKER_UNBOUND;
4581 : :
4582 : 0 : pool->flags |= POOL_DISASSOCIATED;
4583 : :
4584 : : spin_unlock_irq(&pool->lock);
4585 : 0 : mutex_unlock(&pool->manager_mutex);
4586 : :
4587 : : /*
4588 : : * Call schedule() so that we cross rq->lock and thus can
4589 : : * guarantee sched callbacks see the %WORKER_UNBOUND flag.
4590 : : * This is necessary as scheduler callbacks may be invoked
4591 : : * from other cpus.
4592 : : */
4593 : 0 : schedule();
4594 : :
4595 : : /*
4596 : : * Sched callbacks are disabled now. Zap nr_running.
4597 : : * After this, nr_running stays zero and need_more_worker()
4598 : : * and keep_working() are always true as long as the
4599 : : * worklist is not empty. This pool now behaves as an
4600 : : * unbound (in terms of concurrency management) pool which
4601 : : * are served by workers tied to the pool.
4602 : : */
4603 : 0 : atomic_set(&pool->nr_running, 0);
4604 : :
4605 : : /*
4606 : : * With concurrency management just turned off, a busy
4607 : : * worker blocking could lead to lengthy stalls. Kick off
4608 : : * unbound chain execution of currently pending work items.
4609 : : */
4610 : : spin_lock_irq(&pool->lock);
4611 : 0 : wake_up_worker(pool);
4612 : : spin_unlock_irq(&pool->lock);
4613 : : }
4614 : 0 : }
4615 : :
4616 : : /**
4617 : : * rebind_workers - rebind all workers of a pool to the associated CPU
4618 : : * @pool: pool of interest
4619 : : *
4620 : : * @pool->cpu is coming online. Rebind all workers to the CPU.
4621 : : */
4622 : 0 : static void rebind_workers(struct worker_pool *pool)
4623 : : {
4624 : : struct worker *worker;
4625 : : int wi;
4626 : :
4627 : : lockdep_assert_held(&pool->manager_mutex);
4628 : :
4629 : : /*
4630 : : * Restore CPU affinity of all workers. As all idle workers should
4631 : : * be on the run-queue of the associated CPU before any local
4632 : : * wake-ups for concurrency management happen, restore CPU affinty
4633 : : * of all workers first and then clear UNBOUND. As we're called
4634 : : * from CPU_ONLINE, the following shouldn't fail.
4635 : : */
4636 [ # # ]: 0 : for_each_pool_worker(worker, wi, pool)
4637 [ # # ][ # # ]: 0 : WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task,
[ # # ]
4638 : : pool->attrs->cpumask) < 0);
4639 : :
4640 : : spin_lock_irq(&pool->lock);
4641 : :
4642 [ # # ]: 0 : for_each_pool_worker(worker, wi, pool) {
4643 : 0 : unsigned int worker_flags = worker->flags;
4644 : :
4645 : : /*
4646 : : * A bound idle worker should actually be on the runqueue
4647 : : * of the associated CPU for local wake-ups targeting it to
4648 : : * work. Kick all idle workers so that they migrate to the
4649 : : * associated CPU. Doing this in the same loop as
4650 : : * replacing UNBOUND with REBOUND is safe as no worker will
4651 : : * be bound before @pool->lock is released.
4652 : : */
4653 [ # # ]: 0 : if (worker_flags & WORKER_IDLE)
4654 : 0 : wake_up_process(worker->task);
4655 : :
4656 : : /*
4657 : : * We want to clear UNBOUND but can't directly call
4658 : : * worker_clr_flags() or adjust nr_running. Atomically
4659 : : * replace UNBOUND with another NOT_RUNNING flag REBOUND.
4660 : : * @worker will clear REBOUND using worker_clr_flags() when
4661 : : * it initiates the next execution cycle thus restoring
4662 : : * concurrency management. Note that when or whether
4663 : : * @worker clears REBOUND doesn't affect correctness.
4664 : : *
4665 : : * ACCESS_ONCE() is necessary because @worker->flags may be
4666 : : * tested without holding any lock in
4667 : : * wq_worker_waking_up(). Without it, NOT_RUNNING test may
4668 : : * fail incorrectly leading to premature concurrency
4669 : : * management operations.
4670 : : */
4671 [ # # ][ # # ]: 0 : WARN_ON_ONCE(!(worker_flags & WORKER_UNBOUND));
[ # # ]
4672 : : worker_flags |= WORKER_REBOUND;
4673 : 0 : worker_flags &= ~WORKER_UNBOUND;
4674 : 0 : ACCESS_ONCE(worker->flags) = worker_flags;
4675 : : }
4676 : :
4677 : : spin_unlock_irq(&pool->lock);
4678 : 0 : }
4679 : :
4680 : : /**
4681 : : * restore_unbound_workers_cpumask - restore cpumask of unbound workers
4682 : : * @pool: unbound pool of interest
4683 : : * @cpu: the CPU which is coming up
4684 : : *
4685 : : * An unbound pool may end up with a cpumask which doesn't have any online
4686 : : * CPUs. When a worker of such pool get scheduled, the scheduler resets
4687 : : * its cpus_allowed. If @cpu is in @pool's cpumask which didn't have any
4688 : : * online CPU before, cpus_allowed of all its workers should be restored.
4689 : : */
4690 : 0 : static void restore_unbound_workers_cpumask(struct worker_pool *pool, int cpu)
4691 : : {
4692 : : static cpumask_t cpumask;
4693 : : struct worker *worker;
4694 : : int wi;
4695 : :
4696 : : lockdep_assert_held(&pool->manager_mutex);
4697 : :
4698 : : /* is @cpu allowed for @pool? */
4699 [ # # ]: 0 : if (!cpumask_test_cpu(cpu, pool->attrs->cpumask))
4700 : 0 : return;
4701 : :
4702 : : /* is @cpu the only online CPU? */
4703 : 0 : cpumask_and(&cpumask, pool->attrs->cpumask, cpu_online_mask);
4704 [ # # ]: 0 : if (cpumask_weight(&cpumask) != 1)
4705 : : return;
4706 : :
4707 : : /* as we're called from CPU_ONLINE, the following shouldn't fail */
4708 [ # # ]: 0 : for_each_pool_worker(worker, wi, pool)
4709 [ # # ][ # # ]: 0 : WARN_ON_ONCE(set_cpus_allowed_ptr(worker->task,
[ # # ]
4710 : : pool->attrs->cpumask) < 0);
4711 : : }
4712 : :
4713 : : /*
4714 : : * Workqueues should be brought up before normal priority CPU notifiers.
4715 : : * This will be registered high priority CPU notifier.
4716 : : */
4717 : 0 : static int workqueue_cpu_up_callback(struct notifier_block *nfb,
4718 : : unsigned long action,
4719 : : void *hcpu)
4720 : : {
4721 : 0 : int cpu = (unsigned long)hcpu;
4722 : : struct worker_pool *pool;
4723 : : struct workqueue_struct *wq;
4724 : : int pi;
4725 : :
4726 [ # # # ]: 0 : switch (action & ~CPU_TASKS_FROZEN) {
4727 : : case CPU_UP_PREPARE:
4728 [ # # ]: 0 : for_each_cpu_worker_pool(pool, cpu) {
4729 [ # # ]: 0 : if (pool->nr_workers)
4730 : 0 : continue;
4731 [ # # ]: 0 : if (create_and_start_worker(pool) < 0)
4732 : : return NOTIFY_BAD;
4733 : : }
4734 : : break;
4735 : :
4736 : : case CPU_DOWN_FAILED:
4737 : : case CPU_ONLINE:
4738 : 0 : mutex_lock(&wq_pool_mutex);
4739 : :
4740 [ # # ]: 0 : for_each_pool(pool, pi) {
4741 : 0 : mutex_lock(&pool->manager_mutex);
4742 : :
4743 [ # # ]: 0 : if (pool->cpu == cpu) {
4744 : : spin_lock_irq(&pool->lock);
4745 : 0 : pool->flags &= ~POOL_DISASSOCIATED;
4746 : : spin_unlock_irq(&pool->lock);
4747 : :
4748 : 0 : rebind_workers(pool);
4749 [ # # ]: 0 : } else if (pool->cpu < 0) {
4750 : 0 : restore_unbound_workers_cpumask(pool, cpu);
4751 : : }
4752 : :
4753 : 0 : mutex_unlock(&pool->manager_mutex);
4754 : : }
4755 : :
4756 : : /* update NUMA affinity of unbound workqueues */
4757 [ # # ]: 0 : list_for_each_entry(wq, &workqueues, list)
4758 : 0 : wq_update_unbound_numa(wq, cpu, true);
4759 : :
4760 : 0 : mutex_unlock(&wq_pool_mutex);
4761 : 0 : break;
4762 : : }
4763 : : return NOTIFY_OK;
4764 : : }
4765 : :
4766 : : /*
4767 : : * Workqueues should be brought down after normal priority CPU notifiers.
4768 : : * This will be registered as low priority CPU notifier.
4769 : : */
4770 : 0 : static int workqueue_cpu_down_callback(struct notifier_block *nfb,
4771 : : unsigned long action,
4772 : : void *hcpu)
4773 : : {
4774 : 0 : int cpu = (unsigned long)hcpu;
4775 : : struct work_struct unbind_work;
4776 : : struct workqueue_struct *wq;
4777 : :
4778 [ # # ]: 0 : switch (action & ~CPU_TASKS_FROZEN) {
4779 : : case CPU_DOWN_PREPARE:
4780 : : /* unbinding per-cpu workers should happen on the local CPU */
4781 : 0 : INIT_WORK_ONSTACK(&unbind_work, wq_unbind_fn);
4782 : 0 : queue_work_on(cpu, system_highpri_wq, &unbind_work);
4783 : :
4784 : : /* update NUMA affinity of unbound workqueues */
4785 : 0 : mutex_lock(&wq_pool_mutex);
4786 [ # # ]: 0 : list_for_each_entry(wq, &workqueues, list)
4787 : 0 : wq_update_unbound_numa(wq, cpu, false);
4788 : 0 : mutex_unlock(&wq_pool_mutex);
4789 : :
4790 : : /* wait for per-cpu unbinding to finish */
4791 : 0 : flush_work(&unbind_work);
4792 : 0 : break;
4793 : : }
4794 : 0 : return NOTIFY_OK;
4795 : : }
4796 : :
4797 : : #ifdef CONFIG_SMP
4798 : :
4799 : : struct work_for_cpu {
4800 : : struct work_struct work;
4801 : : long (*fn)(void *);
4802 : : void *arg;
4803 : : long ret;
4804 : : };
4805 : :
4806 : 0 : static void work_for_cpu_fn(struct work_struct *work)
4807 : : {
4808 : : struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work);
4809 : :
4810 : 0 : wfc->ret = wfc->fn(wfc->arg);
4811 : 0 : }
4812 : :
4813 : : /**
4814 : : * work_on_cpu - run a function in user context on a particular cpu
4815 : : * @cpu: the cpu to run on
4816 : : * @fn: the function to run
4817 : : * @arg: the function arg
4818 : : *
4819 : : * It is up to the caller to ensure that the cpu doesn't go offline.
4820 : : * The caller must not hold any locks which would prevent @fn from completing.
4821 : : *
4822 : : * Return: The value @fn returns.
4823 : : */
4824 : 0 : long work_on_cpu(int cpu, long (*fn)(void *), void *arg)
4825 : : {
4826 : 0 : struct work_for_cpu wfc = { .fn = fn, .arg = arg };
4827 : :
4828 : 0 : INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn);
4829 : : schedule_work_on(cpu, &wfc.work);
4830 : 0 : flush_work(&wfc.work);
4831 : 0 : return wfc.ret;
4832 : : }
4833 : : EXPORT_SYMBOL_GPL(work_on_cpu);
4834 : : #endif /* CONFIG_SMP */
4835 : :
4836 : : #ifdef CONFIG_FREEZER
4837 : :
4838 : : /**
4839 : : * freeze_workqueues_begin - begin freezing workqueues
4840 : : *
4841 : : * Start freezing workqueues. After this function returns, all freezable
4842 : : * workqueues will queue new works to their delayed_works list instead of
4843 : : * pool->worklist.
4844 : : *
4845 : : * CONTEXT:
4846 : : * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's.
4847 : : */
4848 : 0 : void freeze_workqueues_begin(void)
4849 : : {
4850 : : struct worker_pool *pool;
4851 : : struct workqueue_struct *wq;
4852 : : struct pool_workqueue *pwq;
4853 : : int pi;
4854 : :
4855 : 0 : mutex_lock(&wq_pool_mutex);
4856 : :
4857 [ # # ][ # # ]: 0 : WARN_ON_ONCE(workqueue_freezing);
[ # # ]
4858 : 0 : workqueue_freezing = true;
4859 : :
4860 : : /* set FREEZING */
4861 [ # # ]: 0 : for_each_pool(pool, pi) {
4862 : : spin_lock_irq(&pool->lock);
4863 [ # # ][ # # ]: 0 : WARN_ON_ONCE(pool->flags & POOL_FREEZING);
[ # # ]
4864 : 0 : pool->flags |= POOL_FREEZING;
4865 : : spin_unlock_irq(&pool->lock);
4866 : : }
4867 : :
4868 [ # # ]: 0 : list_for_each_entry(wq, &workqueues, list) {
4869 : 0 : mutex_lock(&wq->mutex);
4870 [ # # ]: 0 : for_each_pwq(pwq, wq)
4871 : 0 : pwq_adjust_max_active(pwq);
4872 : 0 : mutex_unlock(&wq->mutex);
4873 : : }
4874 : :
4875 : 0 : mutex_unlock(&wq_pool_mutex);
4876 : 0 : }
4877 : :
4878 : : /**
4879 : : * freeze_workqueues_busy - are freezable workqueues still busy?
4880 : : *
4881 : : * Check whether freezing is complete. This function must be called
4882 : : * between freeze_workqueues_begin() and thaw_workqueues().
4883 : : *
4884 : : * CONTEXT:
4885 : : * Grabs and releases wq_pool_mutex.
4886 : : *
4887 : : * Return:
4888 : : * %true if some freezable workqueues are still busy. %false if freezing
4889 : : * is complete.
4890 : : */
4891 : 0 : bool freeze_workqueues_busy(void)
4892 : : {
4893 : : bool busy = false;
4894 : : struct workqueue_struct *wq;
4895 : : struct pool_workqueue *pwq;
4896 : :
4897 : 0 : mutex_lock(&wq_pool_mutex);
4898 : :
4899 [ # # ][ # # ]: 0 : WARN_ON_ONCE(!workqueue_freezing);
[ # # ]
4900 : :
4901 [ # # ]: 0 : list_for_each_entry(wq, &workqueues, list) {
4902 [ # # ]: 0 : if (!(wq->flags & WQ_FREEZABLE))
4903 : 0 : continue;
4904 : : /*
4905 : : * nr_active is monotonically decreasing. It's safe
4906 : : * to peek without lock.
4907 : : */
4908 : : rcu_read_lock_sched();
4909 [ # # ]: 0 : for_each_pwq(pwq, wq) {
4910 [ # # ][ # # ]: 0 : WARN_ON_ONCE(pwq->nr_active < 0);
[ # # ]
4911 [ # # ]: 0 : if (pwq->nr_active) {
4912 : : busy = true;
4913 : : rcu_read_unlock_sched();
4914 : : goto out_unlock;
4915 : : }
4916 : : }
4917 : : rcu_read_unlock_sched();
4918 : : }
4919 : : out_unlock:
4920 : 0 : mutex_unlock(&wq_pool_mutex);
4921 : 0 : return busy;
4922 : : }
4923 : :
4924 : : /**
4925 : : * thaw_workqueues - thaw workqueues
4926 : : *
4927 : : * Thaw workqueues. Normal queueing is restored and all collected
4928 : : * frozen works are transferred to their respective pool worklists.
4929 : : *
4930 : : * CONTEXT:
4931 : : * Grabs and releases wq_pool_mutex, wq->mutex and pool->lock's.
4932 : : */
4933 : 0 : void thaw_workqueues(void)
4934 : : {
4935 : : struct workqueue_struct *wq;
4936 : : struct pool_workqueue *pwq;
4937 : : struct worker_pool *pool;
4938 : : int pi;
4939 : :
4940 : 0 : mutex_lock(&wq_pool_mutex);
4941 : :
4942 [ # # ]: 0 : if (!workqueue_freezing)
4943 : : goto out_unlock;
4944 : :
4945 : : /* clear FREEZING */
4946 [ # # ]: 0 : for_each_pool(pool, pi) {
4947 : : spin_lock_irq(&pool->lock);
4948 [ # # ][ # # ]: 0 : WARN_ON_ONCE(!(pool->flags & POOL_FREEZING));
[ # # ]
4949 : 0 : pool->flags &= ~POOL_FREEZING;
4950 : : spin_unlock_irq(&pool->lock);
4951 : : }
4952 : :
4953 : : /* restore max_active and repopulate worklist */
4954 [ # # ]: 0 : list_for_each_entry(wq, &workqueues, list) {
4955 : 0 : mutex_lock(&wq->mutex);
4956 [ # # ]: 0 : for_each_pwq(pwq, wq)
4957 : 0 : pwq_adjust_max_active(pwq);
4958 : 0 : mutex_unlock(&wq->mutex);
4959 : : }
4960 : :
4961 : 0 : workqueue_freezing = false;
4962 : : out_unlock:
4963 : 0 : mutex_unlock(&wq_pool_mutex);
4964 : 0 : }
4965 : : #endif /* CONFIG_FREEZER */
4966 : :
4967 : 0 : static void __init wq_numa_init(void)
4968 : : {
4969 : : cpumask_var_t *tbl;
4970 : : int node, cpu;
4971 : :
4972 : : /* determine NUMA pwq table len - highest node id + 1 */
4973 [ # # ]: 0 : for_each_node(node)
4974 : 0 : wq_numa_tbl_len = max(wq_numa_tbl_len, node + 1);
4975 : :
4976 : : if (num_possible_nodes() <= 1)
4977 : : return;
4978 : :
4979 : : if (wq_disable_numa) {
4980 : : pr_info("workqueue: NUMA affinity support disabled\n");
4981 : : return;
4982 : : }
4983 : :
4984 : : wq_update_unbound_numa_attrs_buf = alloc_workqueue_attrs(GFP_KERNEL);
4985 : : BUG_ON(!wq_update_unbound_numa_attrs_buf);
4986 : :
4987 : : /*
4988 : : * We want masks of possible CPUs of each node which isn't readily
4989 : : * available. Build one from cpu_to_node() which should have been
4990 : : * fully initialized by now.
4991 : : */
4992 : : tbl = kzalloc(wq_numa_tbl_len * sizeof(tbl[0]), GFP_KERNEL);
4993 : : BUG_ON(!tbl);
4994 : :
4995 : : for_each_node(node)
4996 : : BUG_ON(!alloc_cpumask_var_node(&tbl[node], GFP_KERNEL,
4997 : : node_online(node) ? node : NUMA_NO_NODE));
4998 : :
4999 : : for_each_possible_cpu(cpu) {
5000 : : node = cpu_to_node(cpu);
5001 : : if (WARN_ON(node == NUMA_NO_NODE)) {
5002 : : pr_warn("workqueue: NUMA node mapping not available for cpu%d, disabling NUMA support\n", cpu);
5003 : : /* happens iff arch is bonkers, let's just proceed */
5004 : : return;
5005 : : }
5006 : : cpumask_set_cpu(cpu, tbl[node]);
5007 : : }
5008 : :
5009 : : wq_numa_possible_cpumask = tbl;
5010 : : wq_numa_enabled = true;
5011 : : }
5012 : :
5013 : 0 : static int __init init_workqueues(void)
5014 : : {
5015 : 0 : int std_nice[NR_STD_WORKER_POOLS] = { 0, HIGHPRI_NICE_LEVEL };
5016 : : int i, cpu;
5017 : :
5018 : : WARN_ON(__alignof__(struct pool_workqueue) < __alignof__(long long));
5019 : :
5020 : 0 : pwq_cache = KMEM_CACHE(pool_workqueue, SLAB_PANIC);
5021 : :
5022 : 0 : cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP);
5023 : 0 : hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN);
5024 : :
5025 : 0 : wq_numa_init();
5026 : :
5027 : : /* initialize CPU pools */
5028 [ # # ]: 0 : for_each_possible_cpu(cpu) {
5029 : : struct worker_pool *pool;
5030 : :
5031 : : i = 0;
5032 [ # # ]: 0 : for_each_cpu_worker_pool(pool, cpu) {
5033 [ # # ]: 0 : BUG_ON(init_worker_pool(pool));
5034 : 0 : pool->cpu = cpu;
5035 : 0 : cpumask_copy(pool->attrs->cpumask, cpumask_of(cpu));
5036 : 0 : pool->attrs->nice = std_nice[i++];
5037 : 0 : pool->node = cpu_to_node(cpu);
5038 : :
5039 : : /* alloc pool ID */
5040 : 0 : mutex_lock(&wq_pool_mutex);
5041 [ # # ]: 0 : BUG_ON(worker_pool_assign_id(pool));
5042 : 0 : mutex_unlock(&wq_pool_mutex);
5043 : : }
5044 : : }
5045 : :
5046 : : /* create the initial worker */
5047 [ # # ]: 0 : for_each_online_cpu(cpu) {
5048 : : struct worker_pool *pool;
5049 : :
5050 [ # # ]: 0 : for_each_cpu_worker_pool(pool, cpu) {
5051 : 0 : pool->flags &= ~POOL_DISASSOCIATED;
5052 [ # # ]: 0 : BUG_ON(create_and_start_worker(pool) < 0);
5053 : : }
5054 : : }
5055 : :
5056 : : /* create default unbound and ordered wq attrs */
5057 [ # # ]: 0 : for (i = 0; i < NR_STD_WORKER_POOLS; i++) {
5058 : : struct workqueue_attrs *attrs;
5059 : :
5060 [ # # ]: 0 : BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL)));
5061 : 0 : attrs->nice = std_nice[i];
5062 : 0 : unbound_std_wq_attrs[i] = attrs;
5063 : :
5064 : : /*
5065 : : * An ordered wq should have only one pwq as ordering is
5066 : : * guaranteed by max_active which is enforced by pwqs.
5067 : : * Turn off NUMA so that dfl_pwq is used for all nodes.
5068 : : */
5069 [ # # ]: 0 : BUG_ON(!(attrs = alloc_workqueue_attrs(GFP_KERNEL)));
5070 : 0 : attrs->nice = std_nice[i];
5071 : 0 : attrs->no_numa = true;
5072 : 0 : ordered_wq_attrs[i] = attrs;
5073 : : }
5074 : :
5075 : 0 : system_wq = alloc_workqueue("events", 0, 0);
5076 : 0 : system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0);
5077 : 0 : system_long_wq = alloc_workqueue("events_long", 0, 0);
5078 : 0 : system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND,
5079 : : WQ_UNBOUND_MAX_ACTIVE);
5080 : 0 : system_freezable_wq = alloc_workqueue("events_freezable",
5081 : : WQ_FREEZABLE, 0);
5082 : 0 : system_power_efficient_wq = alloc_workqueue("events_power_efficient",
5083 : : WQ_POWER_EFFICIENT, 0);
5084 : 0 : system_freezable_power_efficient_wq = alloc_workqueue("events_freezable_power_efficient",
5085 : : WQ_FREEZABLE | WQ_POWER_EFFICIENT,
5086 : : 0);
5087 [ # # ][ # # ]: 0 : BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq ||
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
[ # # ][ # # ]
5088 : : !system_unbound_wq || !system_freezable_wq ||
5089 : : !system_power_efficient_wq ||
5090 : : !system_freezable_power_efficient_wq);
5091 : 0 : return 0;
5092 : : }
5093 : : early_initcall(init_workqueues);
|